* [PATCH v3 00/20] Support VFIO cdev API in DPDK [not found] <cover.1763141462.git.anatoly.burakov@intel.com> @ 2025-11-18 16:29 ` Anatoly Burakov 2025-11-18 16:29 ` [PATCH v3 01/20] doc: add deprecation notice for VFIO API Anatoly Burakov ` (19 more replies) 2026-02-26 14:17 ` [PATCH v7 00/18] Support VFIO cdev API in DPDK Anatoly Burakov 1 sibling, 20 replies; 42+ messages in thread From: Anatoly Burakov @ 2025-11-18 16:29 UTC (permalink / raw) To: dev This patchset introduces a major refactor of the VFIO subsystem in DPDK to support character device (cdev) interface introduced in Linux kernel, as well as make the API more streamlined and useful. The goal is to simplify device management, improve compatibility, and clarify API responsibilities. The following sections outline the key issues addressed by this patchset and the corresponding changes introduced. 1. Only group mode is supported =============================== Since kernel version 4.14.327 (LTS), VFIO supports the new character device (cdev)-based way of working with VFIO devices (otherwise known as IOMMUFD). This is a device-centric mode and does away with all the complexity regarding groups and IOMMU types, delegating it all to the kernel, and exposes a much simpler interface to userspace. The old group interface is still around, and will need to be kept in DPDK both for compatibility reasons, as well as supporting special cases (FSLMC bus, NBL driver, etc.). To enable this, VFIO is heavily refactored, so that the code can support both modes while relying on (mostly) common infrastructure. Note that the existing `rte_vfio_device_setup/release` model is fundamentally incompatible with cdev mode, because for custom container cases, the expected flow is that the user binds the IOMMU group (and thus, implicitly, the device itself) to a specific container using `rte_vfio_container_group_bind`, whereas this step is not needed for cdev as the device fd is assigned to the container straight away. Therefore, what we do instead is introduce a new API for container device assignment which, semantically, will assign a device to specified container, so that when it is mapped using `rte_pci_map_device`, the appropriate container is selected. Under the hood though, we essentially transition to getting device fd straight away at assign stage, so that by the time the PCI bus attempts to map the device, it is already mapped and we just return an fd. Additionally, a new `rte_vfio_get_mode` API is added for those cases that need some introspection into VFIO's internals, with three new modes: group (old-style), no-iommu (old-style but without IOMMU), and cdev (the new mode). Although no-IOMMU is technically a variant of group mode, the distinction is largely irrelevant to the user, as all usages of noiommu checks in our codebase are for deciding whether to use IOVA or PA, not anything to do with managing groups. The current plan for kernel community is to *not* introduce no-IOMMU cdev implementation, which is why this will be kept for compatibility for these use cases. There were other users of VFIO which relied on group API but only for convenience purposes; no actual VFIO functionality depended on those API's. Therefore, group API's are removed and, where appropriate, replaced with the new API's. List of removed API's: * `rte_vfio_get_group_fd` * `rte_vfio_clear_group` * `rte_vfio_container_group_bind` * `rte_vfio_container_group_unbind` * `rte_vfio_noiommu_is_enabled` (replaced by new mode API) 2. The API responsibilities aren't clear and bleed into each other ================================================================== Some API's do multiple things at once. In particular: * `rte_vfio_get_device_info` will setup the device * `rte_vfio_setup_device` will get device info These API's have been adjusted to do one thing only. v3: - Make API removal cleaner - Fix `get_group_num` usages to align with new API - Fix issues with function exports - Fix issues with `setup_device` returning old-style values in some cases v2: - Make the entire API internal - More aggressive API pruning, complete removal of group API - Fixed a bug in group mode where device could not be used - Better documentation and deprecation notice patches - Moved doc patches to beginning of patchset Anatoly Burakov (20): doc: add deprecation notice for VFIO API doc: add deprecation notice for vDPA driver API uapi: update to v6.17 and add iommufd.h vfio: make all functions internal vfio: split get device info from setup vfio: add container device assignment API net/nbl: do not use VFIO group bind API net/ntnic: use container device assignment API vdpa/ifc: use container device assignment API vdpa/nfp: use container device assignment API vdpa/sfc: use container device assignment API vhost: remove group-related API from drivers vfio: remove group-based API vfio: cleanup and refactor bus/pci: use the new VFIO mode API bus/fslmc: use the new VFIO mode API net/hinic3: use the new VFIO mode API net/ntnic: use the new VFIO mode API vfio: remove no-IOMMU check API vfio: introduce cdev mode config/arm/meson.build | 1 + config/meson.build | 1 + doc/guides/prog_guide/vhost_lib.rst | 4 - doc/guides/rel_notes/deprecation.rst | 7 + drivers/bus/cdx/cdx_vfio.c | 25 +- drivers/bus/fslmc/fslmc_bus.c | 10 +- drivers/bus/fslmc/fslmc_vfio.c | 6 +- drivers/bus/pci/linux/pci.c | 2 +- drivers/bus/pci/linux/pci_vfio.c | 33 +- drivers/bus/platform/platform.c | 9 +- drivers/crypto/bcmfs/bcmfs_vfio.c | 14 +- drivers/net/hinic3/base/hinic3_hwdev.c | 2 +- drivers/net/nbl/nbl_common/nbl_userdev.c | 20 +- drivers/net/nbl/nbl_include/nbl_include.h | 1 + drivers/net/ntnic/ntnic_ethdev.c | 2 +- drivers/net/ntnic/ntnic_vfio.c | 30 +- drivers/vdpa/ifc/ifcvf_vdpa.c | 34 +- drivers/vdpa/mlx5/mlx5_vdpa.c | 1 - drivers/vdpa/nfp/nfp_vdpa.c | 37 +- drivers/vdpa/sfc/sfc_vdpa.c | 39 +- drivers/vdpa/sfc/sfc_vdpa.h | 2 - kernel/linux/uapi/linux/iommufd.h | 1292 +++++++++++ kernel/linux/uapi/linux/vduse.h | 2 +- kernel/linux/uapi/linux/vfio.h | 12 +- kernel/linux/uapi/version | 2 +- lib/eal/freebsd/eal.c | 98 +- lib/eal/include/rte_vfio.h | 387 ++-- lib/eal/linux/eal_vfio.c | 2433 ++++++++------------- lib/eal/linux/eal_vfio.h | 167 +- lib/eal/linux/eal_vfio_cdev.c | 389 ++++ lib/eal/linux/eal_vfio_group.c | 983 +++++++++ lib/eal/linux/eal_vfio_mp_sync.c | 80 +- lib/eal/linux/meson.build | 2 + lib/eal/windows/eal.c | 4 +- lib/vhost/vdpa_driver.h | 3 - 35 files changed, 4266 insertions(+), 1868 deletions(-) create mode 100644 kernel/linux/uapi/linux/iommufd.h create mode 100644 lib/eal/linux/eal_vfio_cdev.c create mode 100644 lib/eal/linux/eal_vfio_group.c -- 2.47.3 ^ permalink raw reply [flat|nested] 42+ messages in thread
* [PATCH v3 01/20] doc: add deprecation notice for VFIO API 2025-11-18 16:29 ` [PATCH v3 00/20] Support VFIO cdev API in DPDK Anatoly Burakov @ 2025-11-18 16:29 ` Anatoly Burakov 2025-11-18 16:29 ` [PATCH v3 02/20] doc: add deprecation notice for vDPA driver API Anatoly Burakov ` (18 subsequent siblings) 19 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2025-11-18 16:29 UTC (permalink / raw) To: dev The entirety of VFIO API will be either made internal and only available to drivers, or will be removed. Add upcoming changes to deprecation notices. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- doc/guides/rel_notes/deprecation.rst | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/doc/guides/rel_notes/deprecation.rst b/doc/guides/rel_notes/deprecation.rst index 1e60ac4e82..5e3df71048 100644 --- a/doc/guides/rel_notes/deprecation.rst +++ b/doc/guides/rel_notes/deprecation.rst @@ -158,3 +158,7 @@ Deprecation Notices without relying on flow API. Currently there is no alternative API providing the same functionality as with ``repr_matching_en`` set to 0. + +* eal: The entire VFIO API (``rte_vfio_*``) will be made internal only, and will only be available to EAL and drivers. + Group-based API (``rte_vfio_*_group_*``) will be removed and replaced with unified container device assignment API. + This change will be made in 26.11 release. -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* [PATCH v3 02/20] doc: add deprecation notice for vDPA driver API 2025-11-18 16:29 ` [PATCH v3 00/20] Support VFIO cdev API in DPDK Anatoly Burakov 2025-11-18 16:29 ` [PATCH v3 01/20] doc: add deprecation notice for VFIO API Anatoly Burakov @ 2025-11-18 16:29 ` Anatoly Burakov 2025-11-18 16:29 ` [PATCH v3 03/20] uapi: update to v6.17 and add iommufd.h Anatoly Burakov ` (17 subsequent siblings) 19 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2025-11-18 16:29 UTC (permalink / raw) To: dev Currently, vDPA driver API is partly built around VFIO group API, which will be removed in a future release and replaced with a device container assignment API. Make this know in deprecation notice. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- doc/guides/rel_notes/deprecation.rst | 3 +++ 1 file changed, 3 insertions(+) diff --git a/doc/guides/rel_notes/deprecation.rst b/doc/guides/rel_notes/deprecation.rst index 5e3df71048..e2c08387fe 100644 --- a/doc/guides/rel_notes/deprecation.rst +++ b/doc/guides/rel_notes/deprecation.rst @@ -162,3 +162,6 @@ Deprecation Notices * eal: The entire VFIO API (``rte_vfio_*``) will be made internal only, and will only be available to EAL and drivers. Group-based API (``rte_vfio_*_group_*``) will be removed and replaced with unified container device assignment API. This change will be made in 26.11 release. + +* vdpa: vDPA driver API will no longer offer ``get_vfio_group_fd`` as part of its internal API. + All drivers will be adjusted to use the new unified VFIO container device assignment API. -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* [PATCH v3 03/20] uapi: update to v6.17 and add iommufd.h 2025-11-18 16:29 ` [PATCH v3 00/20] Support VFIO cdev API in DPDK Anatoly Burakov 2025-11-18 16:29 ` [PATCH v3 01/20] doc: add deprecation notice for VFIO API Anatoly Burakov 2025-11-18 16:29 ` [PATCH v3 02/20] doc: add deprecation notice for vDPA driver API Anatoly Burakov @ 2025-11-18 16:29 ` Anatoly Burakov 2025-11-18 17:36 ` Stephen Hemminger 2025-11-18 16:29 ` [PATCH v3 04/20] vfio: make all functions internal Anatoly Burakov ` (16 subsequent siblings) 19 siblings, 1 reply; 42+ messages in thread From: Anatoly Burakov @ 2025-11-18 16:29 UTC (permalink / raw) To: dev, Maxime Coquelin In order to support VF tokens for cdev-based VFIO mode, kernel v6.17 is required. Update internal headers to version v6.17 and include iommufd.h. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- kernel/linux/uapi/linux/iommufd.h | 1292 +++++++++++++++++++++++++++++ kernel/linux/uapi/linux/vduse.h | 2 +- kernel/linux/uapi/linux/vfio.h | 12 +- kernel/linux/uapi/version | 2 +- 4 files changed, 1305 insertions(+), 3 deletions(-) create mode 100644 kernel/linux/uapi/linux/iommufd.h diff --git a/kernel/linux/uapi/linux/iommufd.h b/kernel/linux/uapi/linux/iommufd.h new file mode 100644 index 0000000000..2105a03955 --- /dev/null +++ b/kernel/linux/uapi/linux/iommufd.h @@ -0,0 +1,1292 @@ +/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */ +/* Copyright (c) 2021-2022, NVIDIA CORPORATION & AFFILIATES. + */ +#ifndef _IOMMUFD_H +#define _IOMMUFD_H + +#include <linux/ioctl.h> +#include <linux/types.h> + +#define IOMMUFD_TYPE (';') + +/** + * DOC: General ioctl format + * + * The ioctl interface follows a general format to allow for extensibility. Each + * ioctl is passed in a structure pointer as the argument providing the size of + * the structure in the first u32. The kernel checks that any structure space + * beyond what it understands is 0. This allows userspace to use the backward + * compatible portion while consistently using the newer, larger, structures. + * + * ioctls use a standard meaning for common errnos: + * + * - ENOTTY: The IOCTL number itself is not supported at all + * - E2BIG: The IOCTL number is supported, but the provided structure has + * non-zero in a part the kernel does not understand. + * - EOPNOTSUPP: The IOCTL number is supported, and the structure is + * understood, however a known field has a value the kernel does not + * understand or support. + * - EINVAL: Everything about the IOCTL was understood, but a field is not + * correct. + * - ENOENT: An ID or IOVA provided does not exist. + * - ENOMEM: Out of memory. + * - EOVERFLOW: Mathematics overflowed. + * + * As well as additional errnos, within specific ioctls. + */ +enum { + IOMMUFD_CMD_BASE = 0x80, + IOMMUFD_CMD_DESTROY = IOMMUFD_CMD_BASE, + IOMMUFD_CMD_IOAS_ALLOC = 0x81, + IOMMUFD_CMD_IOAS_ALLOW_IOVAS = 0x82, + IOMMUFD_CMD_IOAS_COPY = 0x83, + IOMMUFD_CMD_IOAS_IOVA_RANGES = 0x84, + IOMMUFD_CMD_IOAS_MAP = 0x85, + IOMMUFD_CMD_IOAS_UNMAP = 0x86, + IOMMUFD_CMD_OPTION = 0x87, + IOMMUFD_CMD_VFIO_IOAS = 0x88, + IOMMUFD_CMD_HWPT_ALLOC = 0x89, + IOMMUFD_CMD_GET_HW_INFO = 0x8a, + IOMMUFD_CMD_HWPT_SET_DIRTY_TRACKING = 0x8b, + IOMMUFD_CMD_HWPT_GET_DIRTY_BITMAP = 0x8c, + IOMMUFD_CMD_HWPT_INVALIDATE = 0x8d, + IOMMUFD_CMD_FAULT_QUEUE_ALLOC = 0x8e, + IOMMUFD_CMD_IOAS_MAP_FILE = 0x8f, + IOMMUFD_CMD_VIOMMU_ALLOC = 0x90, + IOMMUFD_CMD_VDEVICE_ALLOC = 0x91, + IOMMUFD_CMD_IOAS_CHANGE_PROCESS = 0x92, + IOMMUFD_CMD_VEVENTQ_ALLOC = 0x93, + IOMMUFD_CMD_HW_QUEUE_ALLOC = 0x94, +}; + +/** + * struct iommu_destroy - ioctl(IOMMU_DESTROY) + * @size: sizeof(struct iommu_destroy) + * @id: iommufd object ID to destroy. Can be any destroyable object type. + * + * Destroy any object held within iommufd. + */ +struct iommu_destroy { + __u32 size; + __u32 id; +}; +#define IOMMU_DESTROY _IO(IOMMUFD_TYPE, IOMMUFD_CMD_DESTROY) + +/** + * struct iommu_ioas_alloc - ioctl(IOMMU_IOAS_ALLOC) + * @size: sizeof(struct iommu_ioas_alloc) + * @flags: Must be 0 + * @out_ioas_id: Output IOAS ID for the allocated object + * + * Allocate an IO Address Space (IOAS) which holds an IO Virtual Address (IOVA) + * to memory mapping. + */ +struct iommu_ioas_alloc { + __u32 size; + __u32 flags; + __u32 out_ioas_id; +}; +#define IOMMU_IOAS_ALLOC _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_ALLOC) + +/** + * struct iommu_iova_range - ioctl(IOMMU_IOVA_RANGE) + * @start: First IOVA + * @last: Inclusive last IOVA + * + * An interval in IOVA space. + */ +struct iommu_iova_range { + __aligned_u64 start; + __aligned_u64 last; +}; + +/** + * struct iommu_ioas_iova_ranges - ioctl(IOMMU_IOAS_IOVA_RANGES) + * @size: sizeof(struct iommu_ioas_iova_ranges) + * @ioas_id: IOAS ID to read ranges from + * @num_iovas: Input/Output total number of ranges in the IOAS + * @__reserved: Must be 0 + * @allowed_iovas: Pointer to the output array of struct iommu_iova_range + * @out_iova_alignment: Minimum alignment required for mapping IOVA + * + * Query an IOAS for ranges of allowed IOVAs. Mapping IOVA outside these ranges + * is not allowed. num_iovas will be set to the total number of iovas and + * the allowed_iovas[] will be filled in as space permits. + * + * The allowed ranges are dependent on the HW path the DMA operation takes, and + * can change during the lifetime of the IOAS. A fresh empty IOAS will have a + * full range, and each attached device will narrow the ranges based on that + * device's HW restrictions. Detaching a device can widen the ranges. Userspace + * should query ranges after every attach/detach to know what IOVAs are valid + * for mapping. + * + * On input num_iovas is the length of the allowed_iovas array. On output it is + * the total number of iovas filled in. The ioctl will return -EMSGSIZE and set + * num_iovas to the required value if num_iovas is too small. In this case the + * caller should allocate a larger output array and re-issue the ioctl. + * + * out_iova_alignment returns the minimum IOVA alignment that can be given + * to IOMMU_IOAS_MAP/COPY. IOVA's must satisfy:: + * + * starting_iova % out_iova_alignment == 0 + * (starting_iova + length) % out_iova_alignment == 0 + * + * out_iova_alignment can be 1 indicating any IOVA is allowed. It cannot + * be higher than the system PAGE_SIZE. + */ +struct iommu_ioas_iova_ranges { + __u32 size; + __u32 ioas_id; + __u32 num_iovas; + __u32 __reserved; + __aligned_u64 allowed_iovas; + __aligned_u64 out_iova_alignment; +}; +#define IOMMU_IOAS_IOVA_RANGES _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_IOVA_RANGES) + +/** + * struct iommu_ioas_allow_iovas - ioctl(IOMMU_IOAS_ALLOW_IOVAS) + * @size: sizeof(struct iommu_ioas_allow_iovas) + * @ioas_id: IOAS ID to allow IOVAs from + * @num_iovas: Input/Output total number of ranges in the IOAS + * @__reserved: Must be 0 + * @allowed_iovas: Pointer to array of struct iommu_iova_range + * + * Ensure a range of IOVAs are always available for allocation. If this call + * succeeds then IOMMU_IOAS_IOVA_RANGES will never return a list of IOVA ranges + * that are narrower than the ranges provided here. This call will fail if + * IOMMU_IOAS_IOVA_RANGES is currently narrower than the given ranges. + * + * When an IOAS is first created the IOVA_RANGES will be maximally sized, and as + * devices are attached the IOVA will narrow based on the device restrictions. + * When an allowed range is specified any narrowing will be refused, ie device + * attachment can fail if the device requires limiting within the allowed range. + * + * Automatic IOVA allocation is also impacted by this call. MAP will only + * allocate within the allowed IOVAs if they are present. + * + * This call replaces the entire allowed list with the given list. + */ +struct iommu_ioas_allow_iovas { + __u32 size; + __u32 ioas_id; + __u32 num_iovas; + __u32 __reserved; + __aligned_u64 allowed_iovas; +}; +#define IOMMU_IOAS_ALLOW_IOVAS _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_ALLOW_IOVAS) + +/** + * enum iommufd_ioas_map_flags - Flags for map and copy + * @IOMMU_IOAS_MAP_FIXED_IOVA: If clear the kernel will compute an appropriate + * IOVA to place the mapping at + * @IOMMU_IOAS_MAP_WRITEABLE: DMA is allowed to write to this mapping + * @IOMMU_IOAS_MAP_READABLE: DMA is allowed to read from this mapping + */ +enum iommufd_ioas_map_flags { + IOMMU_IOAS_MAP_FIXED_IOVA = 1 << 0, + IOMMU_IOAS_MAP_WRITEABLE = 1 << 1, + IOMMU_IOAS_MAP_READABLE = 1 << 2, +}; + +/** + * struct iommu_ioas_map - ioctl(IOMMU_IOAS_MAP) + * @size: sizeof(struct iommu_ioas_map) + * @flags: Combination of enum iommufd_ioas_map_flags + * @ioas_id: IOAS ID to change the mapping of + * @__reserved: Must be 0 + * @user_va: Userspace pointer to start mapping from + * @length: Number of bytes to map + * @iova: IOVA the mapping was placed at. If IOMMU_IOAS_MAP_FIXED_IOVA is set + * then this must be provided as input. + * + * Set an IOVA mapping from a user pointer. If FIXED_IOVA is specified then the + * mapping will be established at iova, otherwise a suitable location based on + * the reserved and allowed lists will be automatically selected and returned in + * iova. + * + * If IOMMU_IOAS_MAP_FIXED_IOVA is specified then the iova range must currently + * be unused, existing IOVA cannot be replaced. + */ +struct iommu_ioas_map { + __u32 size; + __u32 flags; + __u32 ioas_id; + __u32 __reserved; + __aligned_u64 user_va; + __aligned_u64 length; + __aligned_u64 iova; +}; +#define IOMMU_IOAS_MAP _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_MAP) + +/** + * struct iommu_ioas_map_file - ioctl(IOMMU_IOAS_MAP_FILE) + * @size: sizeof(struct iommu_ioas_map_file) + * @flags: same as for iommu_ioas_map + * @ioas_id: same as for iommu_ioas_map + * @fd: the memfd to map + * @start: byte offset from start of file to map from + * @length: same as for iommu_ioas_map + * @iova: same as for iommu_ioas_map + * + * Set an IOVA mapping from a memfd file. All other arguments and semantics + * match those of IOMMU_IOAS_MAP. + */ +struct iommu_ioas_map_file { + __u32 size; + __u32 flags; + __u32 ioas_id; + __s32 fd; + __aligned_u64 start; + __aligned_u64 length; + __aligned_u64 iova; +}; +#define IOMMU_IOAS_MAP_FILE _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_MAP_FILE) + +/** + * struct iommu_ioas_copy - ioctl(IOMMU_IOAS_COPY) + * @size: sizeof(struct iommu_ioas_copy) + * @flags: Combination of enum iommufd_ioas_map_flags + * @dst_ioas_id: IOAS ID to change the mapping of + * @src_ioas_id: IOAS ID to copy from + * @length: Number of bytes to copy and map + * @dst_iova: IOVA the mapping was placed at. If IOMMU_IOAS_MAP_FIXED_IOVA is + * set then this must be provided as input. + * @src_iova: IOVA to start the copy + * + * Copy an already existing mapping from src_ioas_id and establish it in + * dst_ioas_id. The src iova/length must exactly match a range used with + * IOMMU_IOAS_MAP. + * + * This may be used to efficiently clone a subset of an IOAS to another, or as a + * kind of 'cache' to speed up mapping. Copy has an efficiency advantage over + * establishing equivalent new mappings, as internal resources are shared, and + * the kernel will pin the user memory only once. + */ +struct iommu_ioas_copy { + __u32 size; + __u32 flags; + __u32 dst_ioas_id; + __u32 src_ioas_id; + __aligned_u64 length; + __aligned_u64 dst_iova; + __aligned_u64 src_iova; +}; +#define IOMMU_IOAS_COPY _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_COPY) + +/** + * struct iommu_ioas_unmap - ioctl(IOMMU_IOAS_UNMAP) + * @size: sizeof(struct iommu_ioas_unmap) + * @ioas_id: IOAS ID to change the mapping of + * @iova: IOVA to start the unmapping at + * @length: Number of bytes to unmap, and return back the bytes unmapped + * + * Unmap an IOVA range. The iova/length must be a superset of a previously + * mapped range used with IOMMU_IOAS_MAP or IOMMU_IOAS_COPY. Splitting or + * truncating ranges is not allowed. The values 0 to U64_MAX will unmap + * everything. + */ +struct iommu_ioas_unmap { + __u32 size; + __u32 ioas_id; + __aligned_u64 iova; + __aligned_u64 length; +}; +#define IOMMU_IOAS_UNMAP _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_UNMAP) + +/** + * enum iommufd_option - ioctl(IOMMU_OPTION_RLIMIT_MODE) and + * ioctl(IOMMU_OPTION_HUGE_PAGES) + * @IOMMU_OPTION_RLIMIT_MODE: + * Change how RLIMIT_MEMLOCK accounting works. The caller must have privilege + * to invoke this. Value 0 (default) is user based accounting, 1 uses process + * based accounting. Global option, object_id must be 0 + * @IOMMU_OPTION_HUGE_PAGES: + * Value 1 (default) allows contiguous pages to be combined when generating + * iommu mappings. Value 0 disables combining, everything is mapped to + * PAGE_SIZE. This can be useful for benchmarking. This is a per-IOAS + * option, the object_id must be the IOAS ID. + */ +enum iommufd_option { + IOMMU_OPTION_RLIMIT_MODE = 0, + IOMMU_OPTION_HUGE_PAGES = 1, +}; + +/** + * enum iommufd_option_ops - ioctl(IOMMU_OPTION_OP_SET) and + * ioctl(IOMMU_OPTION_OP_GET) + * @IOMMU_OPTION_OP_SET: Set the option's value + * @IOMMU_OPTION_OP_GET: Get the option's value + */ +enum iommufd_option_ops { + IOMMU_OPTION_OP_SET = 0, + IOMMU_OPTION_OP_GET = 1, +}; + +/** + * struct iommu_option - iommu option multiplexer + * @size: sizeof(struct iommu_option) + * @option_id: One of enum iommufd_option + * @op: One of enum iommufd_option_ops + * @__reserved: Must be 0 + * @object_id: ID of the object if required + * @val64: Option value to set or value returned on get + * + * Change a simple option value. This multiplexor allows controlling options + * on objects. IOMMU_OPTION_OP_SET will load an option and IOMMU_OPTION_OP_GET + * will return the current value. + */ +struct iommu_option { + __u32 size; + __u32 option_id; + __u16 op; + __u16 __reserved; + __u32 object_id; + __aligned_u64 val64; +}; +#define IOMMU_OPTION _IO(IOMMUFD_TYPE, IOMMUFD_CMD_OPTION) + +/** + * enum iommufd_vfio_ioas_op - IOMMU_VFIO_IOAS_* ioctls + * @IOMMU_VFIO_IOAS_GET: Get the current compatibility IOAS + * @IOMMU_VFIO_IOAS_SET: Change the current compatibility IOAS + * @IOMMU_VFIO_IOAS_CLEAR: Disable VFIO compatibility + */ +enum iommufd_vfio_ioas_op { + IOMMU_VFIO_IOAS_GET = 0, + IOMMU_VFIO_IOAS_SET = 1, + IOMMU_VFIO_IOAS_CLEAR = 2, +}; + +/** + * struct iommu_vfio_ioas - ioctl(IOMMU_VFIO_IOAS) + * @size: sizeof(struct iommu_vfio_ioas) + * @ioas_id: For IOMMU_VFIO_IOAS_SET the input IOAS ID to set + * For IOMMU_VFIO_IOAS_GET will output the IOAS ID + * @op: One of enum iommufd_vfio_ioas_op + * @__reserved: Must be 0 + * + * The VFIO compatibility support uses a single ioas because VFIO APIs do not + * support the ID field. Set or Get the IOAS that VFIO compatibility will use. + * When VFIO_GROUP_SET_CONTAINER is used on an iommufd it will get the + * compatibility ioas, either by taking what is already set, or auto creating + * one. From then on VFIO will continue to use that ioas and is not effected by + * this ioctl. SET or CLEAR does not destroy any auto-created IOAS. + */ +struct iommu_vfio_ioas { + __u32 size; + __u32 ioas_id; + __u16 op; + __u16 __reserved; +}; +#define IOMMU_VFIO_IOAS _IO(IOMMUFD_TYPE, IOMMUFD_CMD_VFIO_IOAS) + +/** + * enum iommufd_hwpt_alloc_flags - Flags for HWPT allocation + * @IOMMU_HWPT_ALLOC_NEST_PARENT: If set, allocate a HWPT that can serve as + * the parent HWPT in a nesting configuration. + * @IOMMU_HWPT_ALLOC_DIRTY_TRACKING: Dirty tracking support for device IOMMU is + * enforced on device attachment + * @IOMMU_HWPT_FAULT_ID_VALID: The fault_id field of hwpt allocation data is + * valid. + * @IOMMU_HWPT_ALLOC_PASID: Requests a domain that can be used with PASID. The + * domain can be attached to any PASID on the device. + * Any domain attached to the non-PASID part of the + * device must also be flagged, otherwise attaching a + * PASID will blocked. + * For the user that wants to attach PASID, ioas is + * not recommended for both the non-PASID part + * and PASID part of the device. + * If IOMMU does not support PASID it will return + * error (-EOPNOTSUPP). + */ +enum iommufd_hwpt_alloc_flags { + IOMMU_HWPT_ALLOC_NEST_PARENT = 1 << 0, + IOMMU_HWPT_ALLOC_DIRTY_TRACKING = 1 << 1, + IOMMU_HWPT_FAULT_ID_VALID = 1 << 2, + IOMMU_HWPT_ALLOC_PASID = 1 << 3, +}; + +/** + * enum iommu_hwpt_vtd_s1_flags - Intel VT-d stage-1 page table + * entry attributes + * @IOMMU_VTD_S1_SRE: Supervisor request + * @IOMMU_VTD_S1_EAFE: Extended access enable + * @IOMMU_VTD_S1_WPE: Write protect enable + */ +enum iommu_hwpt_vtd_s1_flags { + IOMMU_VTD_S1_SRE = 1 << 0, + IOMMU_VTD_S1_EAFE = 1 << 1, + IOMMU_VTD_S1_WPE = 1 << 2, +}; + +/** + * struct iommu_hwpt_vtd_s1 - Intel VT-d stage-1 page table + * info (IOMMU_HWPT_DATA_VTD_S1) + * @flags: Combination of enum iommu_hwpt_vtd_s1_flags + * @pgtbl_addr: The base address of the stage-1 page table. + * @addr_width: The address width of the stage-1 page table + * @__reserved: Must be 0 + */ +struct iommu_hwpt_vtd_s1 { + __aligned_u64 flags; + __aligned_u64 pgtbl_addr; + __u32 addr_width; + __u32 __reserved; +}; + +/** + * struct iommu_hwpt_arm_smmuv3 - ARM SMMUv3 nested STE + * (IOMMU_HWPT_DATA_ARM_SMMUV3) + * + * @ste: The first two double words of the user space Stream Table Entry for + * the translation. Must be little-endian. + * Allowed fields: (Refer to "5.2 Stream Table Entry" in SMMUv3 HW Spec) + * - word-0: V, Cfg, S1Fmt, S1ContextPtr, S1CDMax + * - word-1: EATS, S1DSS, S1CIR, S1COR, S1CSH, S1STALLD + * + * -EIO will be returned if @ste is not legal or contains any non-allowed field. + * Cfg can be used to select a S1, Bypass or Abort configuration. A Bypass + * nested domain will translate the same as the nesting parent. The S1 will + * install a Context Descriptor Table pointing at userspace memory translated + * by the nesting parent. + */ +struct iommu_hwpt_arm_smmuv3 { + __aligned_le64 ste[2]; +}; + +/** + * enum iommu_hwpt_data_type - IOMMU HWPT Data Type + * @IOMMU_HWPT_DATA_NONE: no data + * @IOMMU_HWPT_DATA_VTD_S1: Intel VT-d stage-1 page table + * @IOMMU_HWPT_DATA_ARM_SMMUV3: ARM SMMUv3 Context Descriptor Table + */ +enum iommu_hwpt_data_type { + IOMMU_HWPT_DATA_NONE = 0, + IOMMU_HWPT_DATA_VTD_S1 = 1, + IOMMU_HWPT_DATA_ARM_SMMUV3 = 2, +}; + +/** + * struct iommu_hwpt_alloc - ioctl(IOMMU_HWPT_ALLOC) + * @size: sizeof(struct iommu_hwpt_alloc) + * @flags: Combination of enum iommufd_hwpt_alloc_flags + * @dev_id: The device to allocate this HWPT for + * @pt_id: The IOAS or HWPT or vIOMMU to connect this HWPT to + * @out_hwpt_id: The ID of the new HWPT + * @__reserved: Must be 0 + * @data_type: One of enum iommu_hwpt_data_type + * @data_len: Length of the type specific data + * @data_uptr: User pointer to the type specific data + * @fault_id: The ID of IOMMUFD_FAULT object. Valid only if flags field of + * IOMMU_HWPT_FAULT_ID_VALID is set. + * @__reserved2: Padding to 64-bit alignment. Must be 0. + * + * Explicitly allocate a hardware page table object. This is the same object + * type that is returned by iommufd_device_attach() and represents the + * underlying iommu driver's iommu_domain kernel object. + * + * A kernel-managed HWPT will be created with the mappings from the given + * IOAS via the @pt_id. The @data_type for this allocation must be set to + * IOMMU_HWPT_DATA_NONE. The HWPT can be allocated as a parent HWPT for a + * nesting configuration by passing IOMMU_HWPT_ALLOC_NEST_PARENT via @flags. + * + * A user-managed nested HWPT will be created from a given vIOMMU (wrapping a + * parent HWPT) or a parent HWPT via @pt_id, in which the parent HWPT must be + * allocated previously via the same ioctl from a given IOAS (@pt_id). In this + * case, the @data_type must be set to a pre-defined type corresponding to an + * I/O page table type supported by the underlying IOMMU hardware. The device + * via @dev_id and the vIOMMU via @pt_id must be associated to the same IOMMU + * instance. + * + * If the @data_type is set to IOMMU_HWPT_DATA_NONE, @data_len and + * @data_uptr should be zero. Otherwise, both @data_len and @data_uptr + * must be given. + */ +struct iommu_hwpt_alloc { + __u32 size; + __u32 flags; + __u32 dev_id; + __u32 pt_id; + __u32 out_hwpt_id; + __u32 __reserved; + __u32 data_type; + __u32 data_len; + __aligned_u64 data_uptr; + __u32 fault_id; + __u32 __reserved2; +}; +#define IOMMU_HWPT_ALLOC _IO(IOMMUFD_TYPE, IOMMUFD_CMD_HWPT_ALLOC) + +/** + * enum iommu_hw_info_vtd_flags - Flags for VT-d hw_info + * @IOMMU_HW_INFO_VTD_ERRATA_772415_SPR17: If set, disallow read-only mappings + * on a nested_parent domain. + * https://www.intel.com/content/www/us/en/content-details/772415/content-details.html + */ +enum iommu_hw_info_vtd_flags { + IOMMU_HW_INFO_VTD_ERRATA_772415_SPR17 = 1 << 0, +}; + +/** + * struct iommu_hw_info_vtd - Intel VT-d hardware information + * + * @flags: Combination of enum iommu_hw_info_vtd_flags + * @__reserved: Must be 0 + * + * @cap_reg: Value of Intel VT-d capability register defined in VT-d spec + * section 11.4.2 Capability Register. + * @ecap_reg: Value of Intel VT-d capability register defined in VT-d spec + * section 11.4.3 Extended Capability Register. + * + * User needs to understand the Intel VT-d specification to decode the + * register value. + */ +struct iommu_hw_info_vtd { + __u32 flags; + __u32 __reserved; + __aligned_u64 cap_reg; + __aligned_u64 ecap_reg; +}; + +/** + * struct iommu_hw_info_arm_smmuv3 - ARM SMMUv3 hardware information + * (IOMMU_HW_INFO_TYPE_ARM_SMMUV3) + * + * @flags: Must be set to 0 + * @__reserved: Must be 0 + * @idr: Implemented features for ARM SMMU Non-secure programming interface + * @iidr: Information about the implementation and implementer of ARM SMMU, + * and architecture version supported + * @aidr: ARM SMMU architecture version + * + * For the details of @idr, @iidr and @aidr, please refer to the chapters + * from 6.3.1 to 6.3.6 in the SMMUv3 Spec. + * + * This reports the raw HW capability, and not all bits are meaningful to be + * read by userspace. Only the following fields should be used: + * + * idr[0]: ST_LEVEL, TERM_MODEL, STALL_MODEL, TTENDIAN , CD2L, ASID16, TTF + * idr[1]: SIDSIZE, SSIDSIZE + * idr[3]: BBML, RIL + * idr[5]: VAX, GRAN64K, GRAN16K, GRAN4K + * + * - S1P should be assumed to be true if a NESTED HWPT can be created + * - VFIO/iommufd only support platforms with COHACC, it should be assumed to be + * true. + * - ATS is a per-device property. If the VMM describes any devices as ATS + * capable in ACPI/DT it should set the corresponding idr. + * + * This list may expand in future (eg E0PD, AIE, PBHA, D128, DS etc). It is + * important that VMMs do not read bits outside the list to allow for + * compatibility with future kernels. Several features in the SMMUv3 + * architecture are not currently supported by the kernel for nesting: HTTU, + * BTM, MPAM and others. + */ +struct iommu_hw_info_arm_smmuv3 { + __u32 flags; + __u32 __reserved; + __u32 idr[6]; + __u32 iidr; + __u32 aidr; +}; + +/** + * struct iommu_hw_info_tegra241_cmdqv - NVIDIA Tegra241 CMDQV Hardware + * Information (IOMMU_HW_INFO_TYPE_TEGRA241_CMDQV) + * + * @flags: Must be 0 + * @version: Version number for the CMDQ-V HW for PARAM bits[03:00] + * @log2vcmdqs: Log2 of the total number of VCMDQs for PARAM bits[07:04] + * @log2vsids: Log2 of the total number of SID replacements for PARAM bits[15:12] + * @__reserved: Must be 0 + * + * VMM can use these fields directly in its emulated global PARAM register. Note + * that only one Virtual Interface (VINTF) should be exposed to a VM, i.e. PARAM + * bits[11:08] should be set to 0 for log2 of the total number of VINTFs. + */ +struct iommu_hw_info_tegra241_cmdqv { + __u32 flags; + __u8 version; + __u8 log2vcmdqs; + __u8 log2vsids; + __u8 __reserved; +}; + +/** + * enum iommu_hw_info_type - IOMMU Hardware Info Types + * @IOMMU_HW_INFO_TYPE_NONE: Output by the drivers that do not report hardware + * info + * @IOMMU_HW_INFO_TYPE_DEFAULT: Input to request for a default type + * @IOMMU_HW_INFO_TYPE_INTEL_VTD: Intel VT-d iommu info type + * @IOMMU_HW_INFO_TYPE_ARM_SMMUV3: ARM SMMUv3 iommu info type + * @IOMMU_HW_INFO_TYPE_TEGRA241_CMDQV: NVIDIA Tegra241 CMDQV (extension for ARM + * SMMUv3) info type + */ +enum iommu_hw_info_type { + IOMMU_HW_INFO_TYPE_NONE = 0, + IOMMU_HW_INFO_TYPE_DEFAULT = 0, + IOMMU_HW_INFO_TYPE_INTEL_VTD = 1, + IOMMU_HW_INFO_TYPE_ARM_SMMUV3 = 2, + IOMMU_HW_INFO_TYPE_TEGRA241_CMDQV = 3, +}; + +/** + * enum iommufd_hw_capabilities + * @IOMMU_HW_CAP_DIRTY_TRACKING: IOMMU hardware support for dirty tracking + * If available, it means the following APIs + * are supported: + * + * IOMMU_HWPT_GET_DIRTY_BITMAP + * IOMMU_HWPT_SET_DIRTY_TRACKING + * + * @IOMMU_HW_CAP_PCI_PASID_EXEC: Execute Permission Supported, user ignores it + * when the struct + * iommu_hw_info::out_max_pasid_log2 is zero. + * @IOMMU_HW_CAP_PCI_PASID_PRIV: Privileged Mode Supported, user ignores it + * when the struct + * iommu_hw_info::out_max_pasid_log2 is zero. + */ +enum iommufd_hw_capabilities { + IOMMU_HW_CAP_DIRTY_TRACKING = 1 << 0, + IOMMU_HW_CAP_PCI_PASID_EXEC = 1 << 1, + IOMMU_HW_CAP_PCI_PASID_PRIV = 1 << 2, +}; + +/** + * enum iommufd_hw_info_flags - Flags for iommu_hw_info + * @IOMMU_HW_INFO_FLAG_INPUT_TYPE: If set, @in_data_type carries an input type + * for user space to request for a specific info + */ +enum iommufd_hw_info_flags { + IOMMU_HW_INFO_FLAG_INPUT_TYPE = 1 << 0, +}; + +/** + * struct iommu_hw_info - ioctl(IOMMU_GET_HW_INFO) + * @size: sizeof(struct iommu_hw_info) + * @flags: Must be 0 + * @dev_id: The device bound to the iommufd + * @data_len: Input the length of a user buffer in bytes. Output the length of + * data that kernel supports + * @data_uptr: User pointer to a user-space buffer used by the kernel to fill + * the iommu type specific hardware information data + * @in_data_type: This shares the same field with @out_data_type, making it be + * a bidirectional field. When IOMMU_HW_INFO_FLAG_INPUT_TYPE is + * set, an input type carried via this @in_data_type field will + * be valid, requesting for the info data to the given type. If + * IOMMU_HW_INFO_FLAG_INPUT_TYPE is unset, any input value will + * be seen as IOMMU_HW_INFO_TYPE_DEFAULT + * @out_data_type: Output the iommu hardware info type as defined in the enum + * iommu_hw_info_type. + * @out_capabilities: Output the generic iommu capability info type as defined + * in the enum iommu_hw_capabilities. + * @out_max_pasid_log2: Output the width of PASIDs. 0 means no PASID support. + * PCI devices turn to out_capabilities to check if the + * specific capabilities is supported or not. + * @__reserved: Must be 0 + * + * Query an iommu type specific hardware information data from an iommu behind + * a given device that has been bound to iommufd. This hardware info data will + * be used to sync capabilities between the virtual iommu and the physical + * iommu, e.g. a nested translation setup needs to check the hardware info, so + * a guest stage-1 page table can be compatible with the physical iommu. + * + * To capture an iommu type specific hardware information data, @data_uptr and + * its length @data_len must be provided. Trailing bytes will be zeroed if the + * user buffer is larger than the data that kernel has. Otherwise, kernel only + * fills the buffer using the given length in @data_len. If the ioctl succeeds, + * @data_len will be updated to the length that kernel actually supports, + * @out_data_type will be filled to decode the data filled in the buffer + * pointed by @data_uptr. Input @data_len == zero is allowed. + */ +struct iommu_hw_info { + __u32 size; + __u32 flags; + __u32 dev_id; + __u32 data_len; + __aligned_u64 data_uptr; + union { + __u32 in_data_type; + __u32 out_data_type; + }; + __u8 out_max_pasid_log2; + __u8 __reserved[3]; + __aligned_u64 out_capabilities; +}; +#define IOMMU_GET_HW_INFO _IO(IOMMUFD_TYPE, IOMMUFD_CMD_GET_HW_INFO) + +/* + * enum iommufd_hwpt_set_dirty_tracking_flags - Flags for steering dirty + * tracking + * @IOMMU_HWPT_DIRTY_TRACKING_ENABLE: Enable dirty tracking + */ +enum iommufd_hwpt_set_dirty_tracking_flags { + IOMMU_HWPT_DIRTY_TRACKING_ENABLE = 1, +}; + +/** + * struct iommu_hwpt_set_dirty_tracking - ioctl(IOMMU_HWPT_SET_DIRTY_TRACKING) + * @size: sizeof(struct iommu_hwpt_set_dirty_tracking) + * @flags: Combination of enum iommufd_hwpt_set_dirty_tracking_flags + * @hwpt_id: HW pagetable ID that represents the IOMMU domain + * @__reserved: Must be 0 + * + * Toggle dirty tracking on an HW pagetable. + */ +struct iommu_hwpt_set_dirty_tracking { + __u32 size; + __u32 flags; + __u32 hwpt_id; + __u32 __reserved; +}; +#define IOMMU_HWPT_SET_DIRTY_TRACKING _IO(IOMMUFD_TYPE, \ + IOMMUFD_CMD_HWPT_SET_DIRTY_TRACKING) + +/** + * enum iommufd_hwpt_get_dirty_bitmap_flags - Flags for getting dirty bits + * @IOMMU_HWPT_GET_DIRTY_BITMAP_NO_CLEAR: Just read the PTEs without clearing + * any dirty bits metadata. This flag + * can be passed in the expectation + * where the next operation is an unmap + * of the same IOVA range. + * + */ +enum iommufd_hwpt_get_dirty_bitmap_flags { + IOMMU_HWPT_GET_DIRTY_BITMAP_NO_CLEAR = 1, +}; + +/** + * struct iommu_hwpt_get_dirty_bitmap - ioctl(IOMMU_HWPT_GET_DIRTY_BITMAP) + * @size: sizeof(struct iommu_hwpt_get_dirty_bitmap) + * @hwpt_id: HW pagetable ID that represents the IOMMU domain + * @flags: Combination of enum iommufd_hwpt_get_dirty_bitmap_flags + * @__reserved: Must be 0 + * @iova: base IOVA of the bitmap first bit + * @length: IOVA range size + * @page_size: page size granularity of each bit in the bitmap + * @data: bitmap where to set the dirty bits. The bitmap bits each + * represent a page_size which you deviate from an arbitrary iova. + * + * Checking a given IOVA is dirty: + * + * data[(iova / page_size) / 64] & (1ULL << ((iova / page_size) % 64)) + * + * Walk the IOMMU pagetables for a given IOVA range to return a bitmap + * with the dirty IOVAs. In doing so it will also by default clear any + * dirty bit metadata set in the IOPTE. + */ +struct iommu_hwpt_get_dirty_bitmap { + __u32 size; + __u32 hwpt_id; + __u32 flags; + __u32 __reserved; + __aligned_u64 iova; + __aligned_u64 length; + __aligned_u64 page_size; + __aligned_u64 data; +}; +#define IOMMU_HWPT_GET_DIRTY_BITMAP _IO(IOMMUFD_TYPE, \ + IOMMUFD_CMD_HWPT_GET_DIRTY_BITMAP) + +/** + * enum iommu_hwpt_invalidate_data_type - IOMMU HWPT Cache Invalidation + * Data Type + * @IOMMU_HWPT_INVALIDATE_DATA_VTD_S1: Invalidation data for VTD_S1 + * @IOMMU_VIOMMU_INVALIDATE_DATA_ARM_SMMUV3: Invalidation data for ARM SMMUv3 + */ +enum iommu_hwpt_invalidate_data_type { + IOMMU_HWPT_INVALIDATE_DATA_VTD_S1 = 0, + IOMMU_VIOMMU_INVALIDATE_DATA_ARM_SMMUV3 = 1, +}; + +/** + * enum iommu_hwpt_vtd_s1_invalidate_flags - Flags for Intel VT-d + * stage-1 cache invalidation + * @IOMMU_VTD_INV_FLAGS_LEAF: Indicates whether the invalidation applies + * to all-levels page structure cache or just + * the leaf PTE cache. + */ +enum iommu_hwpt_vtd_s1_invalidate_flags { + IOMMU_VTD_INV_FLAGS_LEAF = 1 << 0, +}; + +/** + * struct iommu_hwpt_vtd_s1_invalidate - Intel VT-d cache invalidation + * (IOMMU_HWPT_INVALIDATE_DATA_VTD_S1) + * @addr: The start address of the range to be invalidated. It needs to + * be 4KB aligned. + * @npages: Number of contiguous 4K pages to be invalidated. + * @flags: Combination of enum iommu_hwpt_vtd_s1_invalidate_flags + * @__reserved: Must be 0 + * + * The Intel VT-d specific invalidation data for user-managed stage-1 cache + * invalidation in nested translation. Userspace uses this structure to + * tell the impacted cache scope after modifying the stage-1 page table. + * + * Invalidating all the caches related to the page table by setting @addr + * to be 0 and @npages to be U64_MAX. + * + * The device TLB will be invalidated automatically if ATS is enabled. + */ +struct iommu_hwpt_vtd_s1_invalidate { + __aligned_u64 addr; + __aligned_u64 npages; + __u32 flags; + __u32 __reserved; +}; + +/** + * struct iommu_viommu_arm_smmuv3_invalidate - ARM SMMUv3 cache invalidation + * (IOMMU_VIOMMU_INVALIDATE_DATA_ARM_SMMUV3) + * @cmd: 128-bit cache invalidation command that runs in SMMU CMDQ. + * Must be little-endian. + * + * Supported command list only when passing in a vIOMMU via @hwpt_id: + * CMDQ_OP_TLBI_NSNH_ALL + * CMDQ_OP_TLBI_NH_VA + * CMDQ_OP_TLBI_NH_VAA + * CMDQ_OP_TLBI_NH_ALL + * CMDQ_OP_TLBI_NH_ASID + * CMDQ_OP_ATC_INV + * CMDQ_OP_CFGI_CD + * CMDQ_OP_CFGI_CD_ALL + * + * -EIO will be returned if the command is not supported. + */ +struct iommu_viommu_arm_smmuv3_invalidate { + __aligned_le64 cmd[2]; +}; + +/** + * struct iommu_hwpt_invalidate - ioctl(IOMMU_HWPT_INVALIDATE) + * @size: sizeof(struct iommu_hwpt_invalidate) + * @hwpt_id: ID of a nested HWPT or a vIOMMU, for cache invalidation + * @data_uptr: User pointer to an array of driver-specific cache invalidation + * data. + * @data_type: One of enum iommu_hwpt_invalidate_data_type, defining the data + * type of all the entries in the invalidation request array. It + * should be a type supported by the hwpt pointed by @hwpt_id. + * @entry_len: Length (in bytes) of a request entry in the request array + * @entry_num: Input the number of cache invalidation requests in the array. + * Output the number of requests successfully handled by kernel. + * @__reserved: Must be 0. + * + * Invalidate iommu cache for user-managed page table or vIOMMU. Modifications + * on a user-managed page table should be followed by this operation, if a HWPT + * is passed in via @hwpt_id. Other caches, such as device cache or descriptor + * cache can be flushed if a vIOMMU is passed in via the @hwpt_id field. + * + * Each ioctl can support one or more cache invalidation requests in the array + * that has a total size of @entry_len * @entry_num. + * + * An empty invalidation request array by setting @entry_num==0 is allowed, and + * @entry_len and @data_uptr would be ignored in this case. This can be used to + * check if the given @data_type is supported or not by kernel. + */ +struct iommu_hwpt_invalidate { + __u32 size; + __u32 hwpt_id; + __aligned_u64 data_uptr; + __u32 data_type; + __u32 entry_len; + __u32 entry_num; + __u32 __reserved; +}; +#define IOMMU_HWPT_INVALIDATE _IO(IOMMUFD_TYPE, IOMMUFD_CMD_HWPT_INVALIDATE) + +/** + * enum iommu_hwpt_pgfault_flags - flags for struct iommu_hwpt_pgfault + * @IOMMU_PGFAULT_FLAGS_PASID_VALID: The pasid field of the fault data is + * valid. + * @IOMMU_PGFAULT_FLAGS_LAST_PAGE: It's the last fault of a fault group. + */ +enum iommu_hwpt_pgfault_flags { + IOMMU_PGFAULT_FLAGS_PASID_VALID = (1 << 0), + IOMMU_PGFAULT_FLAGS_LAST_PAGE = (1 << 1), +}; + +/** + * enum iommu_hwpt_pgfault_perm - perm bits for struct iommu_hwpt_pgfault + * @IOMMU_PGFAULT_PERM_READ: request for read permission + * @IOMMU_PGFAULT_PERM_WRITE: request for write permission + * @IOMMU_PGFAULT_PERM_EXEC: (PCIE 10.4.1) request with a PASID that has the + * Execute Requested bit set in PASID TLP Prefix. + * @IOMMU_PGFAULT_PERM_PRIV: (PCIE 10.4.1) request with a PASID that has the + * Privileged Mode Requested bit set in PASID TLP + * Prefix. + */ +enum iommu_hwpt_pgfault_perm { + IOMMU_PGFAULT_PERM_READ = (1 << 0), + IOMMU_PGFAULT_PERM_WRITE = (1 << 1), + IOMMU_PGFAULT_PERM_EXEC = (1 << 2), + IOMMU_PGFAULT_PERM_PRIV = (1 << 3), +}; + +/** + * struct iommu_hwpt_pgfault - iommu page fault data + * @flags: Combination of enum iommu_hwpt_pgfault_flags + * @dev_id: id of the originated device + * @pasid: Process Address Space ID + * @grpid: Page Request Group Index + * @perm: Combination of enum iommu_hwpt_pgfault_perm + * @__reserved: Must be 0. + * @addr: Fault address + * @length: a hint of how much data the requestor is expecting to fetch. For + * example, if the PRI initiator knows it is going to do a 10MB + * transfer, it could fill in 10MB and the OS could pre-fault in + * 10MB of IOVA. It's default to 0 if there's no such hint. + * @cookie: kernel-managed cookie identifying a group of fault messages. The + * cookie number encoded in the last page fault of the group should + * be echoed back in the response message. + */ +struct iommu_hwpt_pgfault { + __u32 flags; + __u32 dev_id; + __u32 pasid; + __u32 grpid; + __u32 perm; + __u32 __reserved; + __aligned_u64 addr; + __u32 length; + __u32 cookie; +}; + +/** + * enum iommufd_page_response_code - Return status of fault handlers + * @IOMMUFD_PAGE_RESP_SUCCESS: Fault has been handled and the page tables + * populated, retry the access. This is the + * "Success" defined in PCI 10.4.2.1. + * @IOMMUFD_PAGE_RESP_INVALID: Could not handle this fault, don't retry the + * access. This is the "Invalid Request" in PCI + * 10.4.2.1. + */ +enum iommufd_page_response_code { + IOMMUFD_PAGE_RESP_SUCCESS = 0, + IOMMUFD_PAGE_RESP_INVALID = 1, +}; + +/** + * struct iommu_hwpt_page_response - IOMMU page fault response + * @cookie: The kernel-managed cookie reported in the fault message. + * @code: One of response code in enum iommufd_page_response_code. + */ +struct iommu_hwpt_page_response { + __u32 cookie; + __u32 code; +}; + +/** + * struct iommu_fault_alloc - ioctl(IOMMU_FAULT_QUEUE_ALLOC) + * @size: sizeof(struct iommu_fault_alloc) + * @flags: Must be 0 + * @out_fault_id: The ID of the new FAULT + * @out_fault_fd: The fd of the new FAULT + * + * Explicitly allocate a fault handling object. + */ +struct iommu_fault_alloc { + __u32 size; + __u32 flags; + __u32 out_fault_id; + __u32 out_fault_fd; +}; +#define IOMMU_FAULT_QUEUE_ALLOC _IO(IOMMUFD_TYPE, IOMMUFD_CMD_FAULT_QUEUE_ALLOC) + +/** + * enum iommu_viommu_type - Virtual IOMMU Type + * @IOMMU_VIOMMU_TYPE_DEFAULT: Reserved for future use + * @IOMMU_VIOMMU_TYPE_ARM_SMMUV3: ARM SMMUv3 driver specific type + * @IOMMU_VIOMMU_TYPE_TEGRA241_CMDQV: NVIDIA Tegra241 CMDQV (extension for ARM + * SMMUv3) enabled ARM SMMUv3 type + */ +enum iommu_viommu_type { + IOMMU_VIOMMU_TYPE_DEFAULT = 0, + IOMMU_VIOMMU_TYPE_ARM_SMMUV3 = 1, + IOMMU_VIOMMU_TYPE_TEGRA241_CMDQV = 2, +}; + +/** + * struct iommu_viommu_tegra241_cmdqv - NVIDIA Tegra241 CMDQV Virtual Interface + * (IOMMU_VIOMMU_TYPE_TEGRA241_CMDQV) + * @out_vintf_mmap_offset: mmap offset argument for VINTF's page0 + * @out_vintf_mmap_length: mmap length argument for VINTF's page0 + * + * Both @out_vintf_mmap_offset and @out_vintf_mmap_length are reported by kernel + * for user space to mmap the VINTF page0 from the host physical address space + * to the guest physical address space so that a guest kernel can directly R/W + * access to the VINTF page0 in order to control its virtual command queues. + */ +struct iommu_viommu_tegra241_cmdqv { + __aligned_u64 out_vintf_mmap_offset; + __aligned_u64 out_vintf_mmap_length; +}; + +/** + * struct iommu_viommu_alloc - ioctl(IOMMU_VIOMMU_ALLOC) + * @size: sizeof(struct iommu_viommu_alloc) + * @flags: Must be 0 + * @type: Type of the virtual IOMMU. Must be defined in enum iommu_viommu_type + * @dev_id: The device's physical IOMMU will be used to back the virtual IOMMU + * @hwpt_id: ID of a nesting parent HWPT to associate to + * @out_viommu_id: Output virtual IOMMU ID for the allocated object + * @data_len: Length of the type specific data + * @__reserved: Must be 0 + * @data_uptr: User pointer to a driver-specific virtual IOMMU data + * + * Allocate a virtual IOMMU object, representing the underlying physical IOMMU's + * virtualization support that is a security-isolated slice of the real IOMMU HW + * that is unique to a specific VM. Operations global to the IOMMU are connected + * to the vIOMMU, such as: + * - Security namespace for guest owned ID, e.g. guest-controlled cache tags + * - Non-device-affiliated event reporting, e.g. invalidation queue errors + * - Access to a sharable nesting parent pagetable across physical IOMMUs + * - Virtualization of various platforms IDs, e.g. RIDs and others + * - Delivery of paravirtualized invalidation + * - Direct assigned invalidation queues + * - Direct assigned interrupts + */ +struct iommu_viommu_alloc { + __u32 size; + __u32 flags; + __u32 type; + __u32 dev_id; + __u32 hwpt_id; + __u32 out_viommu_id; + __u32 data_len; + __u32 __reserved; + __aligned_u64 data_uptr; +}; +#define IOMMU_VIOMMU_ALLOC _IO(IOMMUFD_TYPE, IOMMUFD_CMD_VIOMMU_ALLOC) + +/** + * struct iommu_vdevice_alloc - ioctl(IOMMU_VDEVICE_ALLOC) + * @size: sizeof(struct iommu_vdevice_alloc) + * @viommu_id: vIOMMU ID to associate with the virtual device + * @dev_id: The physical device to allocate a virtual instance on the vIOMMU + * @out_vdevice_id: Object handle for the vDevice. Pass to IOMMU_DESTORY + * @virt_id: Virtual device ID per vIOMMU, e.g. vSID of ARM SMMUv3, vDeviceID + * of AMD IOMMU, and vRID of Intel VT-d + * + * Allocate a virtual device instance (for a physical device) against a vIOMMU. + * This instance holds the device's information (related to its vIOMMU) in a VM. + * User should use IOMMU_DESTROY to destroy the virtual device before + * destroying the physical device (by closing vfio_cdev fd). Otherwise the + * virtual device would be forcibly destroyed on physical device destruction, + * its vdevice_id would be permanently leaked (unremovable & unreusable) until + * iommu fd closed. + */ +struct iommu_vdevice_alloc { + __u32 size; + __u32 viommu_id; + __u32 dev_id; + __u32 out_vdevice_id; + __aligned_u64 virt_id; +}; +#define IOMMU_VDEVICE_ALLOC _IO(IOMMUFD_TYPE, IOMMUFD_CMD_VDEVICE_ALLOC) + +/** + * struct iommu_ioas_change_process - ioctl(VFIO_IOAS_CHANGE_PROCESS) + * @size: sizeof(struct iommu_ioas_change_process) + * @__reserved: Must be 0 + * + * This transfers pinned memory counts for every memory map in every IOAS + * in the context to the current process. This only supports maps created + * with IOMMU_IOAS_MAP_FILE, and returns EINVAL if other maps are present. + * If the ioctl returns a failure status, then nothing is changed. + * + * This API is useful for transferring operation of a device from one process + * to another, such as during userland live update. + */ +struct iommu_ioas_change_process { + __u32 size; + __u32 __reserved; +}; + +#define IOMMU_IOAS_CHANGE_PROCESS \ + _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_CHANGE_PROCESS) + +/** + * enum iommu_veventq_flag - flag for struct iommufd_vevent_header + * @IOMMU_VEVENTQ_FLAG_LOST_EVENTS: vEVENTQ has lost vEVENTs + */ +enum iommu_veventq_flag { + IOMMU_VEVENTQ_FLAG_LOST_EVENTS = (1U << 0), +}; + +/** + * struct iommufd_vevent_header - Virtual Event Header for a vEVENTQ Status + * @flags: Combination of enum iommu_veventq_flag + * @sequence: The sequence index of a vEVENT in the vEVENTQ, with a range of + * [0, INT_MAX] where the following index of INT_MAX is 0 + * + * Each iommufd_vevent_header reports a sequence index of the following vEVENT: + * + * +----------------------+-------+----------------------+-------+---+-------+ + * | header0 {sequence=0} | data0 | header1 {sequence=1} | data1 |...| dataN | + * +----------------------+-------+----------------------+-------+---+-------+ + * + * And this sequence index is expected to be monotonic to the sequence index of + * the previous vEVENT. If two adjacent sequence indexes has a delta larger than + * 1, it means that delta - 1 number of vEVENTs has lost, e.g. two lost vEVENTs: + * + * +-----+----------------------+-------+----------------------+-------+-----+ + * | ... | header3 {sequence=3} | data3 | header6 {sequence=6} | data6 | ... | + * +-----+----------------------+-------+----------------------+-------+-----+ + * + * If a vEVENT lost at the tail of the vEVENTQ and there is no following vEVENT + * providing the next sequence index, an IOMMU_VEVENTQ_FLAG_LOST_EVENTS header + * would be added to the tail, and no data would follow this header: + * + * +--+----------------------+-------+-----------------------------------------+ + * |..| header3 {sequence=3} | data3 | header4 {flags=LOST_EVENTS, sequence=4} | + * +--+----------------------+-------+-----------------------------------------+ + */ +struct iommufd_vevent_header { + __u32 flags; + __u32 sequence; +}; + +/** + * enum iommu_veventq_type - Virtual Event Queue Type + * @IOMMU_VEVENTQ_TYPE_DEFAULT: Reserved for future use + * @IOMMU_VEVENTQ_TYPE_ARM_SMMUV3: ARM SMMUv3 Virtual Event Queue + * @IOMMU_VEVENTQ_TYPE_TEGRA241_CMDQV: NVIDIA Tegra241 CMDQV Extension IRQ + */ +enum iommu_veventq_type { + IOMMU_VEVENTQ_TYPE_DEFAULT = 0, + IOMMU_VEVENTQ_TYPE_ARM_SMMUV3 = 1, + IOMMU_VEVENTQ_TYPE_TEGRA241_CMDQV = 2, +}; + +/** + * struct iommu_vevent_arm_smmuv3 - ARM SMMUv3 Virtual Event + * (IOMMU_VEVENTQ_TYPE_ARM_SMMUV3) + * @evt: 256-bit ARM SMMUv3 Event record, little-endian. + * Reported event records: (Refer to "7.3 Event records" in SMMUv3 HW Spec) + * - 0x04 C_BAD_STE + * - 0x06 F_STREAM_DISABLED + * - 0x08 C_BAD_SUBSTREAMID + * - 0x0a C_BAD_CD + * - 0x10 F_TRANSLATION + * - 0x11 F_ADDR_SIZE + * - 0x12 F_ACCESS + * - 0x13 F_PERMISSION + * + * StreamID field reports a virtual device ID. To receive a virtual event for a + * device, a vDEVICE must be allocated via IOMMU_VDEVICE_ALLOC. + */ +struct iommu_vevent_arm_smmuv3 { + __aligned_le64 evt[4]; +}; + +/** + * struct iommu_vevent_tegra241_cmdqv - Tegra241 CMDQV IRQ + * (IOMMU_VEVENTQ_TYPE_TEGRA241_CMDQV) + * @lvcmdq_err_map: 128-bit logical vcmdq error map, little-endian. + * (Refer to register LVCMDQ_ERR_MAPs per VINTF ) + * + * The 128-bit register value from HW exclusively reflect the error bits for a + * Virtual Interface represented by a vIOMMU object. Read and report directly. + */ +struct iommu_vevent_tegra241_cmdqv { + __aligned_le64 lvcmdq_err_map[2]; +}; + +/** + * struct iommu_veventq_alloc - ioctl(IOMMU_VEVENTQ_ALLOC) + * @size: sizeof(struct iommu_veventq_alloc) + * @flags: Must be 0 + * @viommu_id: virtual IOMMU ID to associate the vEVENTQ with + * @type: Type of the vEVENTQ. Must be defined in enum iommu_veventq_type + * @veventq_depth: Maximum number of events in the vEVENTQ + * @out_veventq_id: The ID of the new vEVENTQ + * @out_veventq_fd: The fd of the new vEVENTQ. User space must close the + * successfully returned fd after using it + * @__reserved: Must be 0 + * + * Explicitly allocate a virtual event queue interface for a vIOMMU. A vIOMMU + * can have multiple FDs for different types, but is confined to one per @type. + * User space should open the @out_veventq_fd to read vEVENTs out of a vEVENTQ, + * if there are vEVENTs available. A vEVENTQ will lose events due to overflow, + * if the number of the vEVENTs hits @veventq_depth. + * + * Each vEVENT in a vEVENTQ encloses a struct iommufd_vevent_header followed by + * a type-specific data structure, in a normal case: + * + * +-+---------+-------+---------+-------+-----+---------+-------+-+ + * | | header0 | data0 | header1 | data1 | ... | headerN | dataN | | + * +-+---------+-------+---------+-------+-----+---------+-------+-+ + * + * unless a tailing IOMMU_VEVENTQ_FLAG_LOST_EVENTS header is logged (refer to + * struct iommufd_vevent_header). + */ +struct iommu_veventq_alloc { + __u32 size; + __u32 flags; + __u32 viommu_id; + __u32 type; + __u32 veventq_depth; + __u32 out_veventq_id; + __u32 out_veventq_fd; + __u32 __reserved; +}; +#define IOMMU_VEVENTQ_ALLOC _IO(IOMMUFD_TYPE, IOMMUFD_CMD_VEVENTQ_ALLOC) + +/** + * enum iommu_hw_queue_type - HW Queue Type + * @IOMMU_HW_QUEUE_TYPE_DEFAULT: Reserved for future use + * @IOMMU_HW_QUEUE_TYPE_TEGRA241_CMDQV: NVIDIA Tegra241 CMDQV (extension for ARM + * SMMUv3) Virtual Command Queue (VCMDQ) + */ +enum iommu_hw_queue_type { + IOMMU_HW_QUEUE_TYPE_DEFAULT = 0, + /* + * TEGRA241_CMDQV requirements (otherwise, allocation will fail) + * - alloc starts from the lowest @index=0 in ascending order + * - destroy starts from the last allocated @index in descending order + * - @base_addr must be aligned to @length in bytes and mapped in IOAS + * - @length must be a power of 2, with a minimum 32 bytes and a maximum + * 2 ^ idr[1].CMDQS * 16 bytes (use GET_HW_INFO call to read idr[1] + * from struct iommu_hw_info_arm_smmuv3) + * - suggest to back the queue memory with contiguous physical pages or + * a single huge page with alignment of the queue size, and limit the + * emulated vSMMU's IDR1.CMDQS to log2(huge page size / 16 bytes) + */ + IOMMU_HW_QUEUE_TYPE_TEGRA241_CMDQV = 1, +}; + +/** + * struct iommu_hw_queue_alloc - ioctl(IOMMU_HW_QUEUE_ALLOC) + * @size: sizeof(struct iommu_hw_queue_alloc) + * @flags: Must be 0 + * @viommu_id: Virtual IOMMU ID to associate the HW queue with + * @type: One of enum iommu_hw_queue_type + * @index: The logical index to the HW queue per virtual IOMMU for a multi-queue + * model + * @out_hw_queue_id: The ID of the new HW queue + * @nesting_parent_iova: Base address of the queue memory in the guest physical + * address space + * @length: Length of the queue memory + * + * Allocate a HW queue object for a vIOMMU-specific HW-accelerated queue, which + * allows HW to access a guest queue memory described using @nesting_parent_iova + * and @length. + * + * A vIOMMU can allocate multiple queues, but it must use a different @index per + * type to separate each allocation, e.g:: + * + * Type1 HW queue0, Type1 HW queue1, Type2 HW queue0, ... + */ +struct iommu_hw_queue_alloc { + __u32 size; + __u32 flags; + __u32 viommu_id; + __u32 type; + __u32 index; + __u32 out_hw_queue_id; + __aligned_u64 nesting_parent_iova; + __aligned_u64 length; +}; +#define IOMMU_HW_QUEUE_ALLOC _IO(IOMMUFD_TYPE, IOMMUFD_CMD_HW_QUEUE_ALLOC) +#endif diff --git a/kernel/linux/uapi/linux/vduse.h b/kernel/linux/uapi/linux/vduse.h index f46269af34..da6ac89af1 100644 --- a/kernel/linux/uapi/linux/vduse.h +++ b/kernel/linux/uapi/linux/vduse.h @@ -237,7 +237,7 @@ struct vduse_iova_umem { * struct vduse_iova_info - information of one IOVA region * @start: start of the IOVA region * @last: last of the IOVA region - * @capability: capability of the IOVA regsion + * @capability: capability of the IOVA region * @reserved: for future use, needs to be initialized to zero * * Structure used by VDUSE_IOTLB_GET_INFO ioctl to get information of diff --git a/kernel/linux/uapi/linux/vfio.h b/kernel/linux/uapi/linux/vfio.h index 79bf8c0cc5..4d96d1fc12 100644 --- a/kernel/linux/uapi/linux/vfio.h +++ b/kernel/linux/uapi/linux/vfio.h @@ -905,10 +905,12 @@ struct vfio_device_feature { * VFIO_DEVICE_BIND_IOMMUFD - _IOR(VFIO_TYPE, VFIO_BASE + 18, * struct vfio_device_bind_iommufd) * @argsz: User filled size of this data. - * @flags: Must be 0. + * @flags: Must be 0 or a bit flags of VFIO_DEVICE_BIND_* * @iommufd: iommufd to bind. * @out_devid: The device id generated by this bind. devid is a handle for * this device/iommufd bond and can be used in IOMMUFD commands. + * @token_uuid_ptr: Valid if VFIO_DEVICE_BIND_FLAG_TOKEN. Points to a 16 byte + * UUID in the same format as VFIO_DEVICE_FEATURE_PCI_VF_TOKEN. * * Bind a vfio_device to the specified iommufd. * @@ -917,13 +919,21 @@ struct vfio_device_feature { * * Unbind is automatically conducted when device fd is closed. * + * A token is sometimes required to open the device, unless this is known to be + * needed VFIO_DEVICE_BIND_FLAG_TOKEN should not be set and token_uuid_ptr is + * ignored. The only case today is a PF/VF relationship where the VF bind must + * be provided the same token as VFIO_DEVICE_FEATURE_PCI_VF_TOKEN provided to + * the PF. + * * Return: 0 on success, -errno on failure. */ struct vfio_device_bind_iommufd { __u32 argsz; __u32 flags; +#define VFIO_DEVICE_BIND_FLAG_TOKEN (1 << 0) __s32 iommufd; __u32 out_devid; + __aligned_u64 token_uuid_ptr; }; #define VFIO_DEVICE_BIND_IOMMUFD _IO(VFIO_TYPE, VFIO_BASE + 18) diff --git a/kernel/linux/uapi/version b/kernel/linux/uapi/version index 966a998301..d9e789dade 100644 --- a/kernel/linux/uapi/version +++ b/kernel/linux/uapi/version @@ -1 +1 @@ -v6.16 +v6.17 -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* Re: [PATCH v3 03/20] uapi: update to v6.17 and add iommufd.h 2025-11-18 16:29 ` [PATCH v3 03/20] uapi: update to v6.17 and add iommufd.h Anatoly Burakov @ 2025-11-18 17:36 ` Stephen Hemminger 0 siblings, 0 replies; 42+ messages in thread From: Stephen Hemminger @ 2025-11-18 17:36 UTC (permalink / raw) To: Anatoly Burakov; +Cc: dev, Maxime Coquelin On Tue, 18 Nov 2025 16:29:04 +0000 Anatoly Burakov <anatoly.burakov@intel.com> wrote: > In order to support VF tokens for cdev-based VFIO mode, kernel v6.17 is > required. Update internal headers to version v6.17 and include iommufd.h. > > Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Acked-by: Stephen Hemminger <stephen@networkplumber.org> ^ permalink raw reply [flat|nested] 42+ messages in thread
* [PATCH v3 04/20] vfio: make all functions internal 2025-11-18 16:29 ` [PATCH v3 00/20] Support VFIO cdev API in DPDK Anatoly Burakov ` (2 preceding siblings ...) 2025-11-18 16:29 ` [PATCH v3 03/20] uapi: update to v6.17 and add iommufd.h Anatoly Burakov @ 2025-11-18 16:29 ` Anatoly Burakov 2025-11-18 16:29 ` [PATCH v3 05/20] vfio: split get device info from setup Anatoly Burakov ` (15 subsequent siblings) 19 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2025-11-18 16:29 UTC (permalink / raw) To: dev, Bruce Richardson, Tyler Retzlaff, Dmitry Kozlyuk The VFIO API is an externally exported API because the original intent was to offer DMA mapping facilities to applications. However, practical usage of this API seems to be centered around drivers, so keeping this API exported to applications only creates needless API/ABI stability surface that has no added value. Make the entire VFIO API internal-only and visible only to drivers. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- lib/eal/freebsd/eal.c | 30 ++++++++++++------------ lib/eal/include/rte_vfio.h | 47 +++++++++++++++++++++++++++++++++----- lib/eal/linux/eal_vfio.c | 32 +++++++++++++------------- lib/eal/windows/eal.c | 4 ++-- 4 files changed, 74 insertions(+), 39 deletions(-) diff --git a/lib/eal/freebsd/eal.c b/lib/eal/freebsd/eal.c index 6215245ad5..3142ccd4bd 100644 --- a/lib/eal/freebsd/eal.c +++ b/lib/eal/freebsd/eal.c @@ -811,7 +811,7 @@ rte_eal_vfio_get_vf_token(__rte_unused rte_uuid_t vf_token) { } -RTE_EXPORT_SYMBOL(rte_vfio_setup_device) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_setup_device) int rte_vfio_setup_device(__rte_unused const char *sysfs_base, __rte_unused const char *dev_addr, __rte_unused int *vfio_dev_fd, @@ -821,7 +821,7 @@ int rte_vfio_setup_device(__rte_unused const char *sysfs_base, return -1; } -RTE_EXPORT_SYMBOL(rte_vfio_release_device) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_release_device) int rte_vfio_release_device(__rte_unused const char *sysfs_base, __rte_unused const char *dev_addr, __rte_unused int fd) @@ -830,33 +830,33 @@ int rte_vfio_release_device(__rte_unused const char *sysfs_base, return -1; } -RTE_EXPORT_SYMBOL(rte_vfio_enable) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_enable) int rte_vfio_enable(__rte_unused const char *modname) { rte_errno = ENOTSUP; return -1; } -RTE_EXPORT_SYMBOL(rte_vfio_is_enabled) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_is_enabled) int rte_vfio_is_enabled(__rte_unused const char *modname) { return 0; } -RTE_EXPORT_SYMBOL(rte_vfio_noiommu_is_enabled) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_noiommu_is_enabled) int rte_vfio_noiommu_is_enabled(void) { return 0; } -RTE_EXPORT_SYMBOL(rte_vfio_clear_group) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_clear_group) int rte_vfio_clear_group(__rte_unused int vfio_group_fd) { rte_errno = ENOTSUP; return -1; } -RTE_EXPORT_SYMBOL(rte_vfio_get_group_num) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_group_num) int rte_vfio_get_group_num(__rte_unused const char *sysfs_base, __rte_unused const char *dev_addr, @@ -866,7 +866,7 @@ rte_vfio_get_group_num(__rte_unused const char *sysfs_base, return -1; } -RTE_EXPORT_SYMBOL(rte_vfio_get_container_fd) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_container_fd) int rte_vfio_get_container_fd(void) { @@ -874,7 +874,7 @@ rte_vfio_get_container_fd(void) return -1; } -RTE_EXPORT_SYMBOL(rte_vfio_get_group_fd) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_group_fd) int rte_vfio_get_group_fd(__rte_unused int iommu_group_num) { @@ -882,7 +882,7 @@ rte_vfio_get_group_fd(__rte_unused int iommu_group_num) return -1; } -RTE_EXPORT_SYMBOL(rte_vfio_container_create) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_create) int rte_vfio_container_create(void) { @@ -890,7 +890,7 @@ rte_vfio_container_create(void) return -1; } -RTE_EXPORT_SYMBOL(rte_vfio_container_destroy) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_destroy) int rte_vfio_container_destroy(__rte_unused int container_fd) { @@ -898,7 +898,7 @@ rte_vfio_container_destroy(__rte_unused int container_fd) return -1; } -RTE_EXPORT_SYMBOL(rte_vfio_container_group_bind) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_group_bind) int rte_vfio_container_group_bind(__rte_unused int container_fd, __rte_unused int iommu_group_num) @@ -907,7 +907,7 @@ rte_vfio_container_group_bind(__rte_unused int container_fd, return -1; } -RTE_EXPORT_SYMBOL(rte_vfio_container_group_unbind) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_group_unbind) int rte_vfio_container_group_unbind(__rte_unused int container_fd, __rte_unused int iommu_group_num) @@ -916,7 +916,7 @@ rte_vfio_container_group_unbind(__rte_unused int container_fd, return -1; } -RTE_EXPORT_SYMBOL(rte_vfio_container_dma_map) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_dma_map) int rte_vfio_container_dma_map(__rte_unused int container_fd, __rte_unused uint64_t vaddr, @@ -927,7 +927,7 @@ rte_vfio_container_dma_map(__rte_unused int container_fd, return -1; } -RTE_EXPORT_SYMBOL(rte_vfio_container_dma_unmap) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_dma_unmap) int rte_vfio_container_dma_unmap(__rte_unused int container_fd, __rte_unused uint64_t vaddr, diff --git a/lib/eal/include/rte_vfio.h b/lib/eal/include/rte_vfio.h index d1e8bce56b..0ddeb08f94 100644 --- a/lib/eal/include/rte_vfio.h +++ b/lib/eal/include/rte_vfio.h @@ -7,7 +7,11 @@ /** * @file - * RTE VFIO. This library provides various VFIO related utility functions. + * @internal + * + * RTE VFIO internal API. + * + * This library provides VFIO related utility functions for use by drivers. */ #include <stdbool.h> @@ -36,6 +40,7 @@ struct vfio_device_info; #define RTE_VFIO_DEFAULT_CONTAINER_FD (-1) /** + * @internal * Setup vfio_cfg for the device identified by its address. * It discovers the configured I/O MMU groups or sets a new one for the device. * If a new groups is assigned, the DMA mapping is performed. @@ -60,10 +65,12 @@ struct vfio_device_info; * <0 on failure. * >1 if the device cannot be managed this way. */ +__rte_internal int rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, int *vfio_dev_fd, struct vfio_device_info *device_info); /** + * @internal * Release a device mapped to a VFIO-managed I/O MMU group. * * This function is only relevant to linux and will return @@ -82,9 +89,11 @@ int rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, * 0 on success. * <0 on failure. */ +__rte_internal int rte_vfio_release_device(const char *sysfs_base, const char *dev_addr, int fd); /** + * @internal * Enable a VFIO-related kmod. * * This function is only relevant to linux and will return @@ -97,9 +106,11 @@ int rte_vfio_release_device(const char *sysfs_base, const char *dev_addr, int fd * 0 on success. * <0 on failure. */ +__rte_internal int rte_vfio_enable(const char *modname); /** + * @internal * Check whether a VFIO-related kmod is enabled. * * This function is only relevant to Linux. @@ -111,9 +122,11 @@ int rte_vfio_enable(const char *modname); * 1 if true. * 0 otherwise. */ +__rte_internal int rte_vfio_is_enabled(const char *modname); /** + * @internal * Whether VFIO NOIOMMU mode is enabled. * * This function is only relevant to Linux. @@ -123,10 +136,12 @@ int rte_vfio_is_enabled(const char *modname); * 0 if false. * <0 for errors. */ +__rte_internal int rte_vfio_noiommu_is_enabled(void); /** - * Remove group fd from internal VFIO group fd array/ + * @internal + * Remove group fd from internal VFIO group fd array. * * This function is only relevant to linux and will return * an error on BSD. @@ -138,11 +153,13 @@ int rte_vfio_noiommu_is_enabled(void); * 0 on success. * <0 on failure. */ +__rte_internal int rte_vfio_clear_group(int vfio_group_fd); /** - * Parse IOMMU group number for a device + * @internal + * Parse IOMMU group number for a device. * * This function is only relevant to linux and will return * an error on BSD. @@ -161,12 +178,14 @@ rte_vfio_clear_group(int vfio_group_fd); * 0 for non-existent group or VFIO * <0 for errors */ +__rte_internal int rte_vfio_get_group_num(const char *sysfs_base, const char *dev_addr, int *iommu_group_num); /** - * Get device information + * @internal + * Get device information. * * This function is only relevant to Linux and will return an error on BSD. * @@ -186,12 +205,13 @@ rte_vfio_get_group_num(const char *sysfs_base, * 0 on success. * <0 on failure. */ -__rte_experimental +__rte_internal int rte_vfio_get_device_info(const char *sysfs_base, const char *dev_addr, int *vfio_dev_fd, struct vfio_device_info *device_info); /** + * @internal * Get the default VFIO container fd * * This function is only relevant to linux and will return @@ -201,11 +221,13 @@ rte_vfio_get_device_info(const char *sysfs_base, const char *dev_addr, * > 0 default container fd * < 0 if VFIO is not enabled or not supported */ +__rte_internal int rte_vfio_get_container_fd(void); /** - * Open VFIO group fd or get an existing one + * @internal + * Open VFIO group fd or get an existing one. * * This function is only relevant to linux and will return * an error on BSD. @@ -217,10 +239,12 @@ rte_vfio_get_container_fd(void); * > 0 group fd * < 0 for errors */ +__rte_internal int rte_vfio_get_group_fd(int iommu_group_num); /** + * @internal * Create a new container for device binding. * * @note Any newly allocated DPDK memory will not be mapped into these @@ -235,10 +259,12 @@ rte_vfio_get_group_fd(int iommu_group_num); * the container fd if successful * <0 if failed */ +__rte_internal int rte_vfio_container_create(void); /** + * @internal * Destroy the container, unbind all vfio groups within it. * * @param container_fd @@ -248,10 +274,12 @@ rte_vfio_container_create(void); * 0 if successful * <0 if failed */ +__rte_internal int rte_vfio_container_destroy(int container_fd); /** + * @internal * Bind a IOMMU group to a container. * * @param container_fd @@ -264,10 +292,12 @@ rte_vfio_container_destroy(int container_fd); * group fd if successful * <0 if failed */ +__rte_internal int rte_vfio_container_group_bind(int container_fd, int iommu_group_num); /** + * @internal * Unbind a IOMMU group from a container. * * @param container_fd @@ -280,10 +310,12 @@ rte_vfio_container_group_bind(int container_fd, int iommu_group_num); * 0 if successful * <0 if failed */ +__rte_internal int rte_vfio_container_group_unbind(int container_fd, int iommu_group_num); /** + * @internal * Perform DMA mapping for devices in a container. * * @param container_fd @@ -303,11 +335,13 @@ rte_vfio_container_group_unbind(int container_fd, int iommu_group_num); * 0 if successful * <0 if failed */ +__rte_internal int rte_vfio_container_dma_map(int container_fd, uint64_t vaddr, uint64_t iova, uint64_t len); /** + * @internal * Perform DMA unmapping for devices in a container. * * @param container_fd @@ -327,6 +361,7 @@ rte_vfio_container_dma_map(int container_fd, uint64_t vaddr, * 0 if successful * <0 if failed */ +__rte_internal int rte_vfio_container_dma_unmap(int container_fd, uint64_t vaddr, uint64_t iova, uint64_t len); diff --git a/lib/eal/linux/eal_vfio.c b/lib/eal/linux/eal_vfio.c index f1050ffa60..33fa04feaf 100644 --- a/lib/eal/linux/eal_vfio.c +++ b/lib/eal/linux/eal_vfio.c @@ -532,7 +532,7 @@ get_vfio_cfg_by_container_fd(int container_fd) return NULL; } -RTE_EXPORT_SYMBOL(rte_vfio_get_group_fd) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_group_fd) int rte_vfio_get_group_fd(int iommu_group_num) { @@ -731,7 +731,7 @@ vfio_sync_default_container(void) return -1; } -RTE_EXPORT_SYMBOL(rte_vfio_clear_group) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_clear_group) int rte_vfio_clear_group(int vfio_group_fd) { @@ -755,7 +755,7 @@ rte_vfio_clear_group(int vfio_group_fd) return 0; } -RTE_EXPORT_SYMBOL(rte_vfio_setup_device) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_setup_device) int rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, int *vfio_dev_fd, struct vfio_device_info *device_info) @@ -1009,7 +1009,7 @@ rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, return 0; } -RTE_EXPORT_SYMBOL(rte_vfio_release_device) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_release_device) int rte_vfio_release_device(const char *sysfs_base, const char *dev_addr, int vfio_dev_fd) @@ -1098,7 +1098,7 @@ rte_vfio_release_device(const char *sysfs_base, const char *dev_addr, return ret; } -RTE_EXPORT_SYMBOL(rte_vfio_enable) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_enable) int rte_vfio_enable(const char *modname) { @@ -1175,7 +1175,7 @@ rte_vfio_enable(const char *modname) return 0; } -RTE_EXPORT_SYMBOL(rte_vfio_is_enabled) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_is_enabled) int rte_vfio_is_enabled(const char *modname) { @@ -1215,7 +1215,7 @@ vfio_set_iommu_type(int vfio_container_fd) return NULL; } -RTE_EXPORT_EXPERIMENTAL_SYMBOL(rte_vfio_get_device_info, 24.03) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_device_info) int rte_vfio_get_device_info(const char *sysfs_base, const char *dev_addr, int *vfio_dev_fd, struct vfio_device_info *device_info) @@ -1349,7 +1349,7 @@ vfio_open_container_fd(bool mp_request) return -1; } -RTE_EXPORT_SYMBOL(rte_vfio_get_container_fd) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_container_fd) int rte_vfio_get_container_fd(void) { @@ -1363,7 +1363,7 @@ rte_vfio_get_container_fd(void) return default_vfio_cfg->vfio_container_fd; } -RTE_EXPORT_SYMBOL(rte_vfio_get_group_num) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_group_num) int rte_vfio_get_group_num(const char *sysfs_base, const char *dev_addr, int *iommu_group_num) @@ -2034,7 +2034,7 @@ container_dma_unmap(struct vfio_config *vfio_cfg, uint64_t vaddr, uint64_t iova, return ret; } -RTE_EXPORT_SYMBOL(rte_vfio_noiommu_is_enabled) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_noiommu_is_enabled) int rte_vfio_noiommu_is_enabled(void) { @@ -2067,7 +2067,7 @@ rte_vfio_noiommu_is_enabled(void) return c == 'Y'; } -RTE_EXPORT_SYMBOL(rte_vfio_container_create) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_create) int rte_vfio_container_create(void) { @@ -2094,7 +2094,7 @@ rte_vfio_container_create(void) return vfio_cfgs[i].vfio_container_fd; } -RTE_EXPORT_SYMBOL(rte_vfio_container_destroy) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_destroy) int rte_vfio_container_destroy(int container_fd) { @@ -2120,7 +2120,7 @@ rte_vfio_container_destroy(int container_fd) return 0; } -RTE_EXPORT_SYMBOL(rte_vfio_container_group_bind) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_group_bind) int rte_vfio_container_group_bind(int container_fd, int iommu_group_num) { @@ -2135,7 +2135,7 @@ rte_vfio_container_group_bind(int container_fd, int iommu_group_num) return vfio_get_group_fd(vfio_cfg, iommu_group_num); } -RTE_EXPORT_SYMBOL(rte_vfio_container_group_unbind) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_group_unbind) int rte_vfio_container_group_unbind(int container_fd, int iommu_group_num) { @@ -2176,7 +2176,7 @@ rte_vfio_container_group_unbind(int container_fd, int iommu_group_num) return 0; } -RTE_EXPORT_SYMBOL(rte_vfio_container_dma_map) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_dma_map) int rte_vfio_container_dma_map(int container_fd, uint64_t vaddr, uint64_t iova, uint64_t len) @@ -2197,7 +2197,7 @@ rte_vfio_container_dma_map(int container_fd, uint64_t vaddr, uint64_t iova, return container_dma_map(vfio_cfg, vaddr, iova, len); } -RTE_EXPORT_SYMBOL(rte_vfio_container_dma_unmap) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_dma_unmap) int rte_vfio_container_dma_unmap(int container_fd, uint64_t vaddr, uint64_t iova, uint64_t len) diff --git a/lib/eal/windows/eal.c b/lib/eal/windows/eal.c index 21fe7cb1d9..33854025da 100644 --- a/lib/eal/windows/eal.c +++ b/lib/eal/windows/eal.c @@ -445,7 +445,7 @@ eal_asprintf(char **buffer, const char *format, ...) return ret; } -RTE_EXPORT_SYMBOL(rte_vfio_container_dma_map) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_dma_map) int rte_vfio_container_dma_map(__rte_unused int container_fd, __rte_unused uint64_t vaddr, @@ -456,7 +456,7 @@ rte_vfio_container_dma_map(__rte_unused int container_fd, return -1; } -RTE_EXPORT_SYMBOL(rte_vfio_container_dma_unmap) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_dma_unmap) int rte_vfio_container_dma_unmap(__rte_unused int container_fd, __rte_unused uint64_t vaddr, -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* [PATCH v3 05/20] vfio: split get device info from setup 2025-11-18 16:29 ` [PATCH v3 00/20] Support VFIO cdev API in DPDK Anatoly Burakov ` (3 preceding siblings ...) 2025-11-18 16:29 ` [PATCH v3 04/20] vfio: make all functions internal Anatoly Burakov @ 2025-11-18 16:29 ` Anatoly Burakov 2025-11-18 16:29 ` [PATCH v3 06/20] vfio: add container device assignment API Anatoly Burakov ` (14 subsequent siblings) 19 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2025-11-18 16:29 UTC (permalink / raw) To: dev, Nipun Gupta, Nikhil Agarwal, Chenbo Xia, Tomasz Duszynski, Ajit Khaparde, Vikas Gupta, Bruce Richardson, Tyler Retzlaff Currently, setup gets device info as part of setup, while the separate get device info API also calls setup if the fd is zero. Untangle these two APIs and make each do one thing, and adjust all existing callers. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- drivers/bus/cdx/cdx_vfio.c | 12 ++++++++-- drivers/bus/pci/linux/pci_vfio.c | 18 ++++++++++---- drivers/bus/platform/platform.c | 9 ++++++- drivers/crypto/bcmfs/bcmfs_vfio.c | 8 ++++++- lib/eal/freebsd/eal.c | 3 +-- lib/eal/include/rte_vfio.h | 23 +++++++----------- lib/eal/linux/eal_vfio.c | 40 +++++++++---------------------- 7 files changed, 59 insertions(+), 54 deletions(-) diff --git a/drivers/bus/cdx/cdx_vfio.c b/drivers/bus/cdx/cdx_vfio.c index 11fe3265d2..9bae264409 100644 --- a/drivers/bus/cdx/cdx_vfio.c +++ b/drivers/bus/cdx/cdx_vfio.c @@ -401,10 +401,14 @@ cdx_vfio_map_resource_primary(struct rte_cdx_device *dev) return -1; ret = rte_vfio_setup_device(RTE_CDX_BUS_DEVICES_PATH, dev_name, - &vfio_dev_fd, &device_info); + &vfio_dev_fd); if (ret) return ret; + ret = rte_vfio_get_device_info(vfio_dev_fd, &device_info); + if (ret) + goto err_vfio_dev_fd; + /* allocate vfio_res and get region info */ vfio_res = rte_zmalloc("VFIO_RES", sizeof(*vfio_res), 0); if (vfio_res == NULL) { @@ -510,10 +514,14 @@ cdx_vfio_map_resource_secondary(struct rte_cdx_device *dev) } ret = rte_vfio_setup_device(RTE_CDX_BUS_DEVICES_PATH, dev_name, - &vfio_dev_fd, &device_info); + &vfio_dev_fd); if (ret) return ret; + ret = rte_vfio_get_device_info(vfio_dev_fd, &device_info); + if (ret) + goto err_vfio_dev_fd; + /* map MMIO regions */ maps = vfio_res->maps; diff --git a/drivers/bus/pci/linux/pci_vfio.c b/drivers/bus/pci/linux/pci_vfio.c index 8562fdcc6b..57ec9d7066 100644 --- a/drivers/bus/pci/linux/pci_vfio.c +++ b/drivers/bus/pci/linux/pci_vfio.c @@ -753,10 +753,14 @@ pci_vfio_map_resource_primary(struct rte_pci_device *dev) loc->domain, loc->bus, loc->devid, loc->function); ret = rte_vfio_setup_device(rte_pci_get_sysfs_path(), pci_addr, - &vfio_dev_fd, &device_info); + &vfio_dev_fd); if (ret) return ret; + ret = rte_vfio_get_device_info(vfio_dev_fd, &device_info); + if (ret) + goto err_vfio_dev_fd; + if (rte_intr_dev_fd_set(dev->intr_handle, vfio_dev_fd)) goto err_vfio_dev_fd; @@ -962,10 +966,14 @@ pci_vfio_map_resource_secondary(struct rte_pci_device *dev) } ret = rte_vfio_setup_device(rte_pci_get_sysfs_path(), pci_addr, - &vfio_dev_fd, &device_info); + &vfio_dev_fd); if (ret) return ret; + ret = rte_vfio_get_device_info(vfio_dev_fd, &device_info); + if (ret) + goto err_vfio_dev_fd; + ret = pci_vfio_fill_regions(dev, vfio_dev_fd, &device_info); if (ret) goto err_vfio_dev_fd; @@ -1194,12 +1202,14 @@ pci_vfio_ioport_map(struct rte_pci_device *dev, int bar, if (vfio_dev_fd < 0) { return -1; } else if (vfio_dev_fd == 0) { - if (rte_vfio_get_device_info(rte_pci_get_sysfs_path(), pci_addr, - &vfio_dev_fd, &device_info) != 0) + if (rte_vfio_setup_device(rte_pci_get_sysfs_path(), pci_addr, + &vfio_dev_fd) != 0) return -1; /* save vfio_dev_fd so it can be used during release */ if (rte_intr_dev_fd_set(dev->intr_handle, vfio_dev_fd) != 0) return -1; + if (rte_vfio_get_device_info(vfio_dev_fd, &device_info) != 0) + return -1; if (pci_vfio_fill_regions(dev, vfio_dev_fd, &device_info) != 0) return -1; diff --git a/drivers/bus/platform/platform.c b/drivers/bus/platform/platform.c index f6673cf181..9500b8aff5 100644 --- a/drivers/bus/platform/platform.c +++ b/drivers/bus/platform/platform.c @@ -328,12 +328,19 @@ device_setup(struct rte_platform_device *pdev) const char *name = pdev->name; int ret; - ret = rte_vfio_setup_device(PLATFORM_BUS_DEVICES_PATH, name, &pdev->dev_fd, &dev_info); + ret = rte_vfio_setup_device(PLATFORM_BUS_DEVICES_PATH, name, &pdev->dev_fd); if (ret) { PLATFORM_LOG_LINE(ERR, "failed to setup %s", name); return -ENODEV; } + ret = rte_vfio_get_device_info(pdev->dev_fd, &dev_info); + if (ret) { + PLATFORM_LOG_LINE(ERR, "failed to get device info for %s", name); + ret = -ENODEV; + goto out; + } + /* This is an extra check to confirm that platform device was initialized * by a kernel vfio-platform driver. On kernels that predate vfio-platform * driver this flag obviously does not exist. In such scenarios this diff --git a/drivers/crypto/bcmfs/bcmfs_vfio.c b/drivers/crypto/bcmfs/bcmfs_vfio.c index e7f7ed994c..d00aaf1bb7 100644 --- a/drivers/crypto/bcmfs/bcmfs_vfio.c +++ b/drivers/crypto/bcmfs/bcmfs_vfio.c @@ -25,12 +25,18 @@ vfio_map_dev_obj(const char *path, const char *dev_obj, struct vfio_device_info d_info = { .argsz = sizeof(d_info) }; struct vfio_region_info reg_info = { .argsz = sizeof(reg_info) }; - ret = rte_vfio_setup_device(path, dev_obj, dev_fd, &d_info); + ret = rte_vfio_setup_device(path, dev_obj, dev_fd); if (ret) { BCMFS_LOG(ERR, "VFIO Setting for device failed"); return ret; } + ret = rte_vfio_get_device_info(*dev_fd, &d_info); + if (ret) { + BCMFS_LOG(ERR, "VFIO Getting device info failed"); + goto map_failed; + } + /* getting device region info*/ ret = ioctl(*dev_fd, VFIO_DEVICE_GET_REGION_INFO, ®_info); if (ret < 0) { diff --git a/lib/eal/freebsd/eal.c b/lib/eal/freebsd/eal.c index 3142ccd4bd..dea32ae428 100644 --- a/lib/eal/freebsd/eal.c +++ b/lib/eal/freebsd/eal.c @@ -814,8 +814,7 @@ rte_eal_vfio_get_vf_token(__rte_unused rte_uuid_t vf_token) RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_setup_device) int rte_vfio_setup_device(__rte_unused const char *sysfs_base, __rte_unused const char *dev_addr, - __rte_unused int *vfio_dev_fd, - __rte_unused struct vfio_device_info *device_info) + __rte_unused int *vfio_dev_fd) { rte_errno = ENOTSUP; return -1; diff --git a/lib/eal/include/rte_vfio.h b/lib/eal/include/rte_vfio.h index 0ddeb08f94..fb666141f6 100644 --- a/lib/eal/include/rte_vfio.h +++ b/lib/eal/include/rte_vfio.h @@ -55,10 +55,7 @@ struct vfio_device_info; * device location. * * @param vfio_dev_fd - * VFIO fd. - * - * @param device_info - * Device information. + * Pointer to VFIO fd, will be set to the opened device fd on success. * * @return * 0 on success. @@ -67,7 +64,7 @@ struct vfio_device_info; */ __rte_internal int rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, - int *vfio_dev_fd, struct vfio_device_info *device_info); + int *vfio_dev_fd); /** * @internal @@ -187,19 +184,16 @@ rte_vfio_get_group_num(const char *sysfs_base, * @internal * Get device information. * + * This function retrieves VFIO device information from an already opened + * device. The device must be opened with rte_vfio_setup_device() first. + * * This function is only relevant to Linux and will return an error on BSD. * - * @param sysfs_base - * sysfs path prefix. - * - * @param dev_addr - * device location. - * * @param vfio_dev_fd - * VFIO fd. + * VFIO device fd (must be a valid, already opened fd). * * @param device_info - * Device information. + * Pointer to device information structure to be filled. * * @return * 0 on success. @@ -207,8 +201,7 @@ rte_vfio_get_group_num(const char *sysfs_base, */ __rte_internal int -rte_vfio_get_device_info(const char *sysfs_base, const char *dev_addr, - int *vfio_dev_fd, struct vfio_device_info *device_info); +rte_vfio_get_device_info(int vfio_dev_fd, struct vfio_device_info *device_info); /** * @internal diff --git a/lib/eal/linux/eal_vfio.c b/lib/eal/linux/eal_vfio.c index 33fa04feaf..47c973e49a 100644 --- a/lib/eal/linux/eal_vfio.c +++ b/lib/eal/linux/eal_vfio.c @@ -758,7 +758,7 @@ rte_vfio_clear_group(int vfio_group_fd) RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_setup_device) int rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, - int *vfio_dev_fd, struct vfio_device_info *device_info) + int *vfio_dev_fd) { struct vfio_group_status group_status = { .argsz = sizeof(group_status) @@ -975,7 +975,7 @@ rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, *vfio_dev_fd = ioctl(vfio_group_fd, VFIO_GROUP_GET_DEVICE_FD, dev); if (*vfio_dev_fd >= 0) - goto dev_get_info; + goto out; } /* get a file descriptor for the device */ @@ -992,18 +992,8 @@ rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, return -1; } - /* test and setup the device */ -dev_get_info: - ret = ioctl(*vfio_dev_fd, VFIO_DEVICE_GET_INFO, device_info); - if (ret) { - EAL_LOG(ERR, "%s cannot get device info, " - "error %i (%s)", dev_addr, errno, - strerror(errno)); - close(*vfio_dev_fd); - close(vfio_group_fd); - rte_vfio_clear_group(vfio_group_fd); - return -1; - } + /* device is now set up */ +out: vfio_group_device_get(vfio_group_fd); return 0; @@ -1217,26 +1207,18 @@ vfio_set_iommu_type(int vfio_container_fd) RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_device_info) int -rte_vfio_get_device_info(const char *sysfs_base, const char *dev_addr, - int *vfio_dev_fd, struct vfio_device_info *device_info) +rte_vfio_get_device_info(int vfio_dev_fd, struct vfio_device_info *device_info) { int ret; - if (device_info == NULL || *vfio_dev_fd < 0) + if (device_info == NULL || vfio_dev_fd < 0) return -1; - if (*vfio_dev_fd == 0) { - ret = rte_vfio_setup_device(sysfs_base, dev_addr, - vfio_dev_fd, device_info); - if (ret) - return -1; - } else { - ret = ioctl(*vfio_dev_fd, VFIO_DEVICE_GET_INFO, device_info); - if (ret) { - EAL_LOG(ERR, "%s cannot get device info, error %i (%s)", - dev_addr, errno, strerror(errno)); - return -1; - } + ret = ioctl(vfio_dev_fd, VFIO_DEVICE_GET_INFO, device_info); + if (ret) { + EAL_LOG(ERR, "Cannot get device info, error %i (%s)", + errno, strerror(errno)); + return -1; } return 0; -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* [PATCH v3 06/20] vfio: add container device assignment API 2025-11-18 16:29 ` [PATCH v3 00/20] Support VFIO cdev API in DPDK Anatoly Burakov ` (4 preceding siblings ...) 2025-11-18 16:29 ` [PATCH v3 05/20] vfio: split get device info from setup Anatoly Burakov @ 2025-11-18 16:29 ` Anatoly Burakov 2025-11-18 16:29 ` [PATCH v3 07/20] net/nbl: do not use VFIO group bind API Anatoly Burakov ` (13 subsequent siblings) 19 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2025-11-18 16:29 UTC (permalink / raw) To: dev, Bruce Richardson, Tyler Retzlaff Currently, VFIO has explicit group bind API's, but the way they're used is such that no one actually cares about VFIO groups, and the real goal of everyone using VFIO group bind API is to bind devices to particular VFIO container, such that when `rte_vfio_setup_device` call eventually comes, VFIO will pick up the correct container. To remove dependency on group API's, add a new "container assign device" API call that will do the same thing, but will not depend on managing VFIO group fd's. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- lib/eal/freebsd/eal.c | 10 ++++++++++ lib/eal/include/rte_vfio.h | 26 ++++++++++++++++++++++++++ lib/eal/linux/eal_vfio.c | 32 ++++++++++++++++++++++++++++++++ 3 files changed, 68 insertions(+) diff --git a/lib/eal/freebsd/eal.c b/lib/eal/freebsd/eal.c index dea32ae428..a7360db7a7 100644 --- a/lib/eal/freebsd/eal.c +++ b/lib/eal/freebsd/eal.c @@ -936,3 +936,13 @@ rte_vfio_container_dma_unmap(__rte_unused int container_fd, rte_errno = ENOTSUP; return -1; } + +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_assign_device) +int +rte_vfio_container_assign_device(__rte_unused int vfio_container_fd, + __rte_unused const char *sysfs_base, + __rte_unused const char *dev_addr) +{ + rte_errno = ENOTSUP; + return -1; +} diff --git a/lib/eal/include/rte_vfio.h b/lib/eal/include/rte_vfio.h index fb666141f6..e7e2ee950b 100644 --- a/lib/eal/include/rte_vfio.h +++ b/lib/eal/include/rte_vfio.h @@ -271,6 +271,32 @@ __rte_internal int rte_vfio_container_destroy(int container_fd); +/** + * @internal + * + * Assign a device to a VFIO container. + * + * Doing so will cause `rte_vfio_setup_device()` to set up the device with the VFIO container + * specified in this assign operation. + * + * This function is only relevant on Linux. + * + * @param vfio_container_fd + * VFIO container file descriptor. + * @param sysfs_base + * Sysfs path prefix. + * @param dev_addr + * Device identifier. + * + * @return + * 0 on success. + * <0 on failure, rte_errno is set. + */ +__rte_internal +int +rte_vfio_container_assign_device(int vfio_container_fd, const char *sysfs_base, + const char *dev_addr); + /** * @internal * Bind a IOMMU group to a container. diff --git a/lib/eal/linux/eal_vfio.c b/lib/eal/linux/eal_vfio.c index 47c973e49a..02fec64658 100644 --- a/lib/eal/linux/eal_vfio.c +++ b/lib/eal/linux/eal_vfio.c @@ -2102,6 +2102,38 @@ rte_vfio_container_destroy(int container_fd) return 0; } +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_assign_device) +int +rte_vfio_container_assign_device(int vfio_container_fd, const char *sysfs_base, + const char *dev_addr) +{ + int iommu_group_num; + int ret; + + ret = rte_vfio_get_group_num(sysfs_base, dev_addr, &iommu_group_num); + if (ret < 0) { + EAL_LOG(ERR, "Cannot get IOMMU group number for device %s", + dev_addr); + return -1; + } else if (ret == 0) { + EAL_LOG(ERR, + "Device %s is not assigned to any IOMMU group", + dev_addr); + return -1; + } + + ret = rte_vfio_container_group_bind(vfio_container_fd, + iommu_group_num); + if (ret < 0) { + EAL_LOG(ERR, + "Cannot bind IOMMU group %d for device %s", + iommu_group_num, dev_addr); + return -1; + } + + return 0; +} + RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_group_bind) int rte_vfio_container_group_bind(int container_fd, int iommu_group_num) -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* [PATCH v3 07/20] net/nbl: do not use VFIO group bind API 2025-11-18 16:29 ` [PATCH v3 00/20] Support VFIO cdev API in DPDK Anatoly Burakov ` (5 preceding siblings ...) 2025-11-18 16:29 ` [PATCH v3 06/20] vfio: add container device assignment API Anatoly Burakov @ 2025-11-18 16:29 ` Anatoly Burakov 2025-11-18 16:29 ` [PATCH v3 08/20] net/ntnic: use container device assignment API Anatoly Burakov ` (12 subsequent siblings) 19 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2025-11-18 16:29 UTC (permalink / raw) To: dev, Dimon Zhao, Kyo Liu, Leon Yu, Sam Chen The NBL driver currently uses group bind API, but it is using it only to get group fd and nothing else. In context of NBL driver, this is the only usage of VFIO API's in the driver, and it is not necessary to use it for what NBL driver is trying to accomplish. Use a direct `open()` call instead, and store the group fd in common structure. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- drivers/net/nbl/nbl_common/nbl_userdev.c | 18 +++++++++++------- drivers/net/nbl/nbl_include/nbl_include.h | 1 + 2 files changed, 12 insertions(+), 7 deletions(-) diff --git a/drivers/net/nbl/nbl_common/nbl_userdev.c b/drivers/net/nbl/nbl_common/nbl_userdev.c index 566f3a11ee..d641ecd9a8 100644 --- a/drivers/net/nbl/nbl_common/nbl_userdev.c +++ b/drivers/net/nbl/nbl_common/nbl_userdev.c @@ -387,6 +387,13 @@ nbl_userdev_mem_event_callback(enum rte_mem_event type, const void *addr, size_t } } +static int nbl_open_group_fd(int iommu_group_num) +{ + char path[PATH_MAX]; + snprintf(path, sizeof(path), RTE_VFIO_GROUP_FMT, iommu_group_num); + return open(path, O_RDWR); +} + static int nbl_mdev_map_device(struct nbl_adapter *adapter) { const struct rte_pci_device *pci_dev = adapter->pci_dev; @@ -424,11 +431,12 @@ static int nbl_mdev_map_device(struct nbl_adapter *adapter) } NBL_LOG(DEBUG, "nbl vfio container %d", container); - vfio_group_fd = rte_vfio_container_group_bind(container, common->iommu_group_num); + vfio_group_fd = nbl_open_group_fd(common->iommu_group_num); if (vfio_group_fd < 0) { NBL_LOG(ERR, "nbl vfio group bind failed, %d", vfio_group_fd); goto free_container; } + common->groupfd = vfio_group_fd; /* check if the group is viable */ ret = ioctl(vfio_group_fd, VFIO_GROUP_GET_STATUS, &group_status); @@ -535,7 +543,6 @@ static int nbl_mdev_map_device(struct nbl_adapter *adapter) } free_group: close(vfio_group_fd); - rte_vfio_clear_group(vfio_group_fd); free_container: if (container_create) rte_vfio_container_destroy(container); @@ -549,17 +556,14 @@ static int nbl_mdev_unmap_device(struct nbl_adapter *adapter) close(common->devfd); rte_mcfg_mem_read_lock(); - vfio_group_fd = rte_vfio_container_group_bind(nbl_default_container, - common->iommu_group_num); + vfio_group_fd = common->groupfd; NBL_LOG(DEBUG, "close vfio_group_fd %d", vfio_group_fd); ret = ioctl(vfio_group_fd, VFIO_GROUP_UNSET_CONTAINER, &nbl_default_container); if (ret) NBL_LOG(ERR, "unset container, error %i (%s) %d", errno, strerror(errno), ret); nbl_group_count--; - ret = rte_vfio_container_group_unbind(nbl_default_container, common->iommu_group_num); - if (ret) - NBL_LOG(ERR, "vfio container group unbind failed %d", ret); + close(vfio_group_fd); if (!nbl_group_count) { rte_mem_event_callback_unregister(NBL_USERDEV_EVENT_CLB_NAME, NULL); nbl_userdev_dma_free(); diff --git a/drivers/net/nbl/nbl_include/nbl_include.h b/drivers/net/nbl/nbl_include/nbl_include.h index eeae6a3301..ba99b9f8e7 100644 --- a/drivers/net/nbl/nbl_include/nbl_include.h +++ b/drivers/net/nbl/nbl_include/nbl_include.h @@ -132,6 +132,7 @@ struct nbl_common_info { u16 vsi_id; u16 instance_id; int devfd; + int groupfd; int eventfd; int ifindex; int iommu_group_num; -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* [PATCH v3 08/20] net/ntnic: use container device assignment API 2025-11-18 16:29 ` [PATCH v3 00/20] Support VFIO cdev API in DPDK Anatoly Burakov ` (6 preceding siblings ...) 2025-11-18 16:29 ` [PATCH v3 07/20] net/nbl: do not use VFIO group bind API Anatoly Burakov @ 2025-11-18 16:29 ` Anatoly Burakov 2025-11-18 16:29 ` [PATCH v3 09/20] vdpa/ifc: " Anatoly Burakov ` (11 subsequent siblings) 19 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2025-11-18 16:29 UTC (permalink / raw) To: dev, Christian Koue Muf, Serhii Iliushyk The NTNIC driver uses VFIO group bind/unbind functionality for container device assignment purposes. Use the new container device assignment API instead to provide clearer semantics. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- drivers/net/ntnic/ntnic_vfio.c | 30 +++++++++--------------------- 1 file changed, 9 insertions(+), 21 deletions(-) diff --git a/drivers/net/ntnic/ntnic_vfio.c b/drivers/net/ntnic/ntnic_vfio.c index 439468b3a2..c746b300b2 100644 --- a/drivers/net/ntnic/ntnic_vfio.c +++ b/drivers/net/ntnic/ntnic_vfio.c @@ -28,7 +28,6 @@ nt_vfio_vf_num(const struct rte_pci_device *pdev) /* Internal API */ struct vfio_dev { int container_fd; - int group_fd; int dev_fd; uint64_t iova_addr; }; @@ -50,7 +49,6 @@ nthw_vfio_setup(struct rte_pci_device *dev) { int ret; char devname[RTE_DEV_NAME_MAX_LEN] = { 0 }; - int iommu_group_num; int vf_num; struct vfio_dev *vfio; @@ -66,14 +64,9 @@ nthw_vfio_setup(struct rte_pci_device *dev) } vfio->dev_fd = -1; - vfio->group_fd = -1; vfio->iova_addr = START_VF_IOVA; rte_pci_device_name(&dev->addr, devname, RTE_DEV_NAME_MAX_LEN); - ret = rte_vfio_get_group_num(rte_pci_get_sysfs_path(), devname, &iommu_group_num); - if (ret <= 0) - return -1; - if (vf_num == 0) { /* use default container for pf0 */ vfio->container_fd = RTE_VFIO_DEFAULT_CONTAINER_FD; @@ -86,17 +79,14 @@ nthw_vfio_setup(struct rte_pci_device *dev) "VFIO device setup failed. VFIO container creation failed."); return -1; } - } + ret = rte_vfio_container_assign_device(vfio->container_fd, + rte_pci_get_sysfs_path(), devname); + if (ret < 0) { + NT_LOG(ERR, NTNIC, + "VFIO device setup failed. Assign device to container failed."); + goto err; + } - vfio->group_fd = rte_vfio_container_group_bind(vfio->container_fd, iommu_group_num); - - if (vfio->group_fd < 0) { - NT_LOG(ERR, NTNIC, - "VFIO device setup failed. VFIO container group bind failed."); - goto err; - } - - if (vf_num > 0) { if (rte_pci_map_device(dev)) { NT_LOG(ERR, NTNIC, "Map VFIO device failed. is the vfio-pci driver loaded?"); @@ -106,10 +96,8 @@ nthw_vfio_setup(struct rte_pci_device *dev) vfio->dev_fd = rte_intr_dev_fd_get(dev->intr_handle); - NT_LOG(DBG, NTNIC, - "%s: VFIO id=%d, dev_fd=%d, container_fd=%d, group_fd=%d, iommu_group_num=%d", - dev->name, vf_num, vfio->dev_fd, vfio->container_fd, vfio->group_fd, - iommu_group_num); + NT_LOG(DBG, NTNIC, "%s: VFIO id=%d, dev_fd=%d, container_fd=%d", + dev->name, vf_num, vfio->dev_fd, vfio->container_fd); return vf_num; -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* [PATCH v3 09/20] vdpa/ifc: use container device assignment API 2025-11-18 16:29 ` [PATCH v3 00/20] Support VFIO cdev API in DPDK Anatoly Burakov ` (7 preceding siblings ...) 2025-11-18 16:29 ` [PATCH v3 08/20] net/ntnic: use container device assignment API Anatoly Burakov @ 2025-11-18 16:29 ` Anatoly Burakov 2025-11-18 16:29 ` [PATCH v3 10/20] vdpa/nfp: " Anatoly Burakov ` (10 subsequent siblings) 19 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2025-11-18 16:29 UTC (permalink / raw) To: dev The IFC vDPA driver uses VFIO group bind/unbind functionality for container device assignment purposes. Use the new container device assignment API instead to provide clearer semantics. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- drivers/vdpa/ifc/ifcvf_vdpa.c | 15 +++------------ 1 file changed, 3 insertions(+), 12 deletions(-) diff --git a/drivers/vdpa/ifc/ifcvf_vdpa.c b/drivers/vdpa/ifc/ifcvf_vdpa.c index f319d455ba..6f1c050787 100644 --- a/drivers/vdpa/ifc/ifcvf_vdpa.c +++ b/drivers/vdpa/ifc/ifcvf_vdpa.c @@ -174,28 +174,19 @@ ifcvf_vfio_setup(struct ifcvf_internal *internal) { struct rte_pci_device *dev = internal->pdev; char devname[RTE_DEV_NAME_MAX_LEN] = {0}; - int iommu_group_num; - int i, ret; + int i; internal->vfio_dev_fd = -1; - internal->vfio_group_fd = -1; internal->vfio_container_fd = -1; rte_pci_device_name(&dev->addr, devname, RTE_DEV_NAME_MAX_LEN); - ret = rte_vfio_get_group_num(rte_pci_get_sysfs_path(), devname, - &iommu_group_num); - if (ret <= 0) { - DRV_LOG(ERR, "%s failed to get IOMMU group", devname); - return -1; - } internal->vfio_container_fd = rte_vfio_container_create(); if (internal->vfio_container_fd < 0) return -1; - internal->vfio_group_fd = rte_vfio_container_group_bind( - internal->vfio_container_fd, iommu_group_num); - if (internal->vfio_group_fd < 0) + if (rte_vfio_container_assign_device(internal->vfio_container_fd, + rte_pci_get_sysfs_path(), devname) < 0) goto err; if (rte_pci_map_device(dev)) -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* [PATCH v3 10/20] vdpa/nfp: use container device assignment API 2025-11-18 16:29 ` [PATCH v3 00/20] Support VFIO cdev API in DPDK Anatoly Burakov ` (8 preceding siblings ...) 2025-11-18 16:29 ` [PATCH v3 09/20] vdpa/ifc: " Anatoly Burakov @ 2025-11-18 16:29 ` Anatoly Burakov 2025-11-18 16:29 ` [PATCH v3 11/20] vdpa/sfc: " Anatoly Burakov ` (9 subsequent siblings) 19 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2025-11-18 16:29 UTC (permalink / raw) To: dev, Chaoyong He The NFP vDPA driver uses VFIO group bind/unbind functionality for container device assignment purposes. Use the new container device assignment API instead to provide clearer semantics. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- drivers/vdpa/nfp/nfp_vdpa.c | 17 +++++------------ 1 file changed, 5 insertions(+), 12 deletions(-) diff --git a/drivers/vdpa/nfp/nfp_vdpa.c b/drivers/vdpa/nfp/nfp_vdpa.c index f4fd5c92ec..4885fa5cbc 100644 --- a/drivers/vdpa/nfp/nfp_vdpa.c +++ b/drivers/vdpa/nfp/nfp_vdpa.c @@ -122,33 +122,26 @@ nfp_vdpa_vfio_setup(struct nfp_vdpa_dev *device) rte_pci_unmap_device(pci_dev); rte_pci_device_name(&pci_dev->addr, dev_name, RTE_DEV_NAME_MAX_LEN); - ret = rte_vfio_get_group_num(rte_pci_get_sysfs_path(), dev_name, - &device->iommu_group); - if (ret <= 0) - return -1; device->vfio_container_fd = rte_vfio_container_create(); if (device->vfio_container_fd < 0) return -1; - device->vfio_group_fd = rte_vfio_container_group_bind( - device->vfio_container_fd, device->iommu_group); - if (device->vfio_group_fd < 0) + ret = rte_vfio_container_assign_device(device->vfio_container_fd, + rte_pci_get_sysfs_path(), dev_name); + if (ret < 0) goto container_destroy; - DRV_VDPA_LOG(DEBUG, "The container_fd=%d, group_fd=%d.", - device->vfio_container_fd, device->vfio_group_fd); + DRV_VDPA_LOG(DEBUG, "container_fd=%d", device->vfio_container_fd); ret = rte_pci_map_device(pci_dev); if (ret != 0) - goto group_unbind; + goto container_destroy; device->vfio_dev_fd = rte_intr_dev_fd_get(pci_dev->intr_handle); return 0; -group_unbind: - rte_vfio_container_group_unbind(device->vfio_container_fd, device->iommu_group); container_destroy: rte_vfio_container_destroy(device->vfio_container_fd); -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* [PATCH v3 11/20] vdpa/sfc: use container device assignment API 2025-11-18 16:29 ` [PATCH v3 00/20] Support VFIO cdev API in DPDK Anatoly Burakov ` (9 preceding siblings ...) 2025-11-18 16:29 ` [PATCH v3 10/20] vdpa/nfp: " Anatoly Burakov @ 2025-11-18 16:29 ` Anatoly Burakov 2025-11-18 16:29 ` [PATCH v3 12/20] vhost: remove group-related API from drivers Anatoly Burakov ` (8 subsequent siblings) 19 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2025-11-18 16:29 UTC (permalink / raw) To: dev, Vijay Kumar Srivastava The SFC vDPA driver uses VFIO group bind/unbind functionality for container device assignment purposes. Use the new container device assignment API instead to provide clearer semantics. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- drivers/vdpa/sfc/sfc_vdpa.c | 39 +++++++------------------------------ drivers/vdpa/sfc/sfc_vdpa.h | 2 -- 2 files changed, 7 insertions(+), 34 deletions(-) diff --git a/drivers/vdpa/sfc/sfc_vdpa.c b/drivers/vdpa/sfc/sfc_vdpa.c index eda111954f..99b4ced3f4 100644 --- a/drivers/vdpa/sfc/sfc_vdpa.c +++ b/drivers/vdpa/sfc/sfc_vdpa.c @@ -80,22 +80,12 @@ sfc_vdpa_vfio_setup(struct sfc_vdpa_adapter *sva) goto fail_container_create; } - rc = rte_vfio_get_group_num(rte_pci_get_sysfs_path(), dev_name, - &sva->iommu_group_num); - if (rc <= 0) { - sfc_vdpa_err(sva, "failed to get IOMMU group for %s : %s", - dev_name, rte_strerror(-rc)); - goto fail_get_group_num; - } - - sva->vfio_group_fd = - rte_vfio_container_group_bind(sva->vfio_container_fd, - sva->iommu_group_num); - if (sva->vfio_group_fd < 0) { - sfc_vdpa_err(sva, - "failed to bind IOMMU group %d to container %d", - sva->iommu_group_num, sva->vfio_container_fd); - goto fail_group_bind; + rc = rte_vfio_container_assign_device(sva->vfio_container_fd, + rte_pci_get_sysfs_path(), dev_name); + if (rc < 0) { + sfc_vdpa_err(sva, "failed to assign device %s to container %d", + dev_name, sva->vfio_container_fd); + goto fail_device_assign; } if (rte_pci_map_device(dev) != 0) { @@ -109,15 +99,7 @@ sfc_vdpa_vfio_setup(struct sfc_vdpa_adapter *sva) return 0; fail_pci_map_device: - if (rte_vfio_container_group_unbind(sva->vfio_container_fd, - sva->iommu_group_num) != 0) { - sfc_vdpa_err(sva, - "failed to unbind IOMMU group %d from container %d", - sva->iommu_group_num, sva->vfio_container_fd); - } - -fail_group_bind: -fail_get_group_num: +fail_device_assign: if (rte_vfio_container_destroy(sva->vfio_container_fd) != 0) { sfc_vdpa_err(sva, "failed to destroy container %d", sva->vfio_container_fd); @@ -132,13 +114,6 @@ sfc_vdpa_vfio_teardown(struct sfc_vdpa_adapter *sva) { rte_pci_unmap_device(sva->pdev); - if (rte_vfio_container_group_unbind(sva->vfio_container_fd, - sva->iommu_group_num) != 0) { - sfc_vdpa_err(sva, - "failed to unbind IOMMU group %d from container %d", - sva->iommu_group_num, sva->vfio_container_fd); - } - if (rte_vfio_container_destroy(sva->vfio_container_fd) != 0) { sfc_vdpa_err(sva, "failed to destroy container %d", diff --git a/drivers/vdpa/sfc/sfc_vdpa.h b/drivers/vdpa/sfc/sfc_vdpa.h index 2b843e563d..99a81fd1b0 100644 --- a/drivers/vdpa/sfc/sfc_vdpa.h +++ b/drivers/vdpa/sfc/sfc_vdpa.h @@ -70,10 +70,8 @@ struct sfc_vdpa_adapter { sfc_vdpa_filter_t filters; - int vfio_group_fd; int vfio_dev_fd; int vfio_container_fd; - int iommu_group_num; struct sfc_vdpa_ops_data *ops_data; }; -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* [PATCH v3 12/20] vhost: remove group-related API from drivers 2025-11-18 16:29 ` [PATCH v3 00/20] Support VFIO cdev API in DPDK Anatoly Burakov ` (10 preceding siblings ...) 2025-11-18 16:29 ` [PATCH v3 11/20] vdpa/sfc: " Anatoly Burakov @ 2025-11-18 16:29 ` Anatoly Burakov 2025-11-18 16:29 ` [PATCH v3 13/20] vfio: remove group-based API Anatoly Burakov ` (7 subsequent siblings) 19 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2025-11-18 16:29 UTC (permalink / raw) To: dev, Maxime Coquelin, Chenbo Xia, Matan Azrad, Viacheslav Ovsiienko, Chaoyong He Some vDPA drivers have "get_vfio_group_fd" call in their internal driver API structure, but it is not used for anything beyond device assignment to containers which can now be achieved via other means, so remove this API and all its usages. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- doc/guides/prog_guide/vhost_lib.rst | 4 ---- drivers/vdpa/ifc/ifcvf_vdpa.c | 19 ------------------- drivers/vdpa/mlx5/mlx5_vdpa.c | 1 - drivers/vdpa/nfp/nfp_vdpa.c | 20 -------------------- lib/vhost/vdpa_driver.h | 3 --- 5 files changed, 47 deletions(-) diff --git a/doc/guides/prog_guide/vhost_lib.rst b/doc/guides/prog_guide/vhost_lib.rst index 0c2b4d020a..2f80cf4072 100644 --- a/doc/guides/prog_guide/vhost_lib.rst +++ b/doc/guides/prog_guide/vhost_lib.rst @@ -471,10 +471,6 @@ Finally, a set of device ops is defined for device specific operations: Called to allow the device to response to RARP sending. -* ``get_vfio_group_fd`` - - Called to get the VFIO group fd of the device. - * ``get_vfio_device_fd`` Called to get the VFIO device fd of the device. diff --git a/drivers/vdpa/ifc/ifcvf_vdpa.c b/drivers/vdpa/ifc/ifcvf_vdpa.c index 6f1c050787..63f4172da5 100644 --- a/drivers/vdpa/ifc/ifcvf_vdpa.c +++ b/drivers/vdpa/ifc/ifcvf_vdpa.c @@ -58,7 +58,6 @@ struct ifcvf_internal { struct ifcvf_hw hw; int configured; int vfio_container_fd; - int vfio_group_fd; int vfio_dev_fd; rte_thread_t tid; /* thread for notify relay */ rte_thread_t intr_tid; /* thread for config space change interrupt relay */ @@ -1204,22 +1203,6 @@ ifcvf_set_features(int vid) return 0; } -static int -ifcvf_get_vfio_group_fd(int vid) -{ - struct rte_vdpa_device *vdev; - struct internal_list *list; - - vdev = rte_vhost_get_vdpa_device(vid); - list = find_internal_resource_by_vdev(vdev); - if (list == NULL) { - DRV_LOG(ERR, "Invalid vDPA device: %p", vdev); - return -1; - } - - return list->internal->vfio_group_fd; -} - static int ifcvf_get_vfio_device_fd(int vid) { @@ -1465,7 +1448,6 @@ static struct rte_vdpa_dev_ops ifcvf_net_ops = { .set_vring_state = ifcvf_set_vring_state, .set_features = ifcvf_set_features, .migration_done = NULL, - .get_vfio_group_fd = ifcvf_get_vfio_group_fd, .get_vfio_device_fd = ifcvf_get_vfio_device_fd, .get_notify_area = ifcvf_get_notify_area, .get_dev_type = ifcvf_get_device_type, @@ -1596,7 +1578,6 @@ static struct rte_vdpa_dev_ops ifcvf_blk_ops = { .dev_close = ifcvf_dev_close, .set_vring_state = ifcvf_set_vring_state, .migration_done = NULL, - .get_vfio_group_fd = ifcvf_get_vfio_group_fd, .get_vfio_device_fd = ifcvf_get_vfio_device_fd, .get_notify_area = ifcvf_get_notify_area, .get_config = ifcvf_blk_get_config, diff --git a/drivers/vdpa/mlx5/mlx5_vdpa.c b/drivers/vdpa/mlx5/mlx5_vdpa.c index 11708e2005..1bb5de51b6 100644 --- a/drivers/vdpa/mlx5/mlx5_vdpa.c +++ b/drivers/vdpa/mlx5/mlx5_vdpa.c @@ -523,7 +523,6 @@ static struct rte_vdpa_dev_ops mlx5_vdpa_ops = { .set_vring_state = mlx5_vdpa_set_vring_state, .set_features = mlx5_vdpa_features_set, .migration_done = NULL, - .get_vfio_group_fd = NULL, .get_vfio_device_fd = mlx5_vdpa_get_device_fd, .get_notify_area = mlx5_vdpa_get_notify_area, .get_stats_names = mlx5_vdpa_get_stats_names, diff --git a/drivers/vdpa/nfp/nfp_vdpa.c b/drivers/vdpa/nfp/nfp_vdpa.c index 4885fa5cbc..efd7ee7d95 100644 --- a/drivers/vdpa/nfp/nfp_vdpa.c +++ b/drivers/vdpa/nfp/nfp_vdpa.c @@ -36,9 +36,7 @@ struct nfp_vdpa_dev { struct nfp_vdpa_hw hw; int vfio_container_fd; - int vfio_group_fd; int vfio_dev_fd; - int iommu_group; rte_thread_t tid; /**< Thread for notify relay */ int epoll_fd; @@ -152,7 +150,6 @@ static void nfp_vdpa_vfio_teardown(struct nfp_vdpa_dev *device) { rte_pci_unmap_device(device->pci_dev); - rte_vfio_container_group_unbind(device->vfio_container_fd, device->iommu_group); rte_vfio_container_destroy(device->vfio_container_fd); } @@ -1018,22 +1015,6 @@ nfp_vdpa_dev_close(int vid) return 0; } -static int -nfp_vdpa_get_vfio_group_fd(int vid) -{ - struct rte_vdpa_device *vdev; - struct nfp_vdpa_dev_node *node; - - vdev = rte_vhost_get_vdpa_device(vid); - node = nfp_vdpa_find_node_by_vdev(vdev); - if (node == NULL) { - DRV_VDPA_LOG(ERR, "Invalid vDPA device: %p.", vdev); - return -ENODEV; - } - - return node->device->vfio_group_fd; -} - static int nfp_vdpa_get_vfio_device_fd(int vid) { @@ -1185,7 +1166,6 @@ struct rte_vdpa_dev_ops nfp_vdpa_ops = { .dev_close = nfp_vdpa_dev_close, .set_vring_state = nfp_vdpa_set_vring_state, .set_features = nfp_vdpa_set_features, - .get_vfio_group_fd = nfp_vdpa_get_vfio_group_fd, .get_vfio_device_fd = nfp_vdpa_get_vfio_device_fd, .get_notify_area = nfp_vdpa_get_notify_area, }; diff --git a/lib/vhost/vdpa_driver.h b/lib/vhost/vdpa_driver.h index 42392a0d14..c7b9be09fb 100644 --- a/lib/vhost/vdpa_driver.h +++ b/lib/vhost/vdpa_driver.h @@ -50,9 +50,6 @@ struct rte_vdpa_dev_ops { /** Destination operations when migration done */ int (*migration_done)(int vid); - /** Get the vfio group fd */ - int (*get_vfio_group_fd)(int vid); - /** Get the vfio device fd */ int (*get_vfio_device_fd)(int vid); -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* [PATCH v3 13/20] vfio: remove group-based API 2025-11-18 16:29 ` [PATCH v3 00/20] Support VFIO cdev API in DPDK Anatoly Burakov ` (11 preceding siblings ...) 2025-11-18 16:29 ` [PATCH v3 12/20] vhost: remove group-related API from drivers Anatoly Burakov @ 2025-11-18 16:29 ` Anatoly Burakov 2025-11-18 16:29 ` [PATCH v3 14/20] vfio: cleanup and refactor Anatoly Burakov ` (6 subsequent siblings) 19 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2025-11-18 16:29 UTC (permalink / raw) To: dev, Bruce Richardson, Tyler Retzlaff All drivers have been adjusted to not use the VFIO group API directly and instead rely on container device assignment model, so the group API is no longer useful and can be removed. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- lib/eal/freebsd/eal.c | 33 --------------- lib/eal/include/rte_vfio.h | 72 -------------------------------- lib/eal/linux/eal_vfio.c | 47 ++++++++++----------- lib/eal/linux/eal_vfio.h | 2 + lib/eal/linux/eal_vfio_mp_sync.c | 2 +- 5 files changed, 26 insertions(+), 130 deletions(-) diff --git a/lib/eal/freebsd/eal.c b/lib/eal/freebsd/eal.c index a7360db7a7..f66ac01ba2 100644 --- a/lib/eal/freebsd/eal.c +++ b/lib/eal/freebsd/eal.c @@ -848,13 +848,6 @@ int rte_vfio_noiommu_is_enabled(void) return 0; } -RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_clear_group) -int rte_vfio_clear_group(__rte_unused int vfio_group_fd) -{ - rte_errno = ENOTSUP; - return -1; -} - RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_group_num) int rte_vfio_get_group_num(__rte_unused const char *sysfs_base, @@ -873,14 +866,6 @@ rte_vfio_get_container_fd(void) return -1; } -RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_group_fd) -int -rte_vfio_get_group_fd(__rte_unused int iommu_group_num) -{ - rte_errno = ENOTSUP; - return -1; -} - RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_create) int rte_vfio_container_create(void) @@ -897,24 +882,6 @@ rte_vfio_container_destroy(__rte_unused int container_fd) return -1; } -RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_group_bind) -int -rte_vfio_container_group_bind(__rte_unused int container_fd, - __rte_unused int iommu_group_num) -{ - rte_errno = ENOTSUP; - return -1; -} - -RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_group_unbind) -int -rte_vfio_container_group_unbind(__rte_unused int container_fd, - __rte_unused int iommu_group_num) -{ - rte_errno = ENOTSUP; - return -1; -} - RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_dma_map) int rte_vfio_container_dma_map(__rte_unused int container_fd, diff --git a/lib/eal/include/rte_vfio.h b/lib/eal/include/rte_vfio.h index e7e2ee950b..941b7d0541 100644 --- a/lib/eal/include/rte_vfio.h +++ b/lib/eal/include/rte_vfio.h @@ -136,24 +136,6 @@ int rte_vfio_is_enabled(const char *modname); __rte_internal int rte_vfio_noiommu_is_enabled(void); -/** - * @internal - * Remove group fd from internal VFIO group fd array. - * - * This function is only relevant to linux and will return - * an error on BSD. - * - * @param vfio_group_fd - * VFIO Group FD. - * - * @return - * 0 on success. - * <0 on failure. - */ -__rte_internal -int -rte_vfio_clear_group(int vfio_group_fd); - /** * @internal * Parse IOMMU group number for a device. @@ -218,24 +200,6 @@ __rte_internal int rte_vfio_get_container_fd(void); -/** - * @internal - * Open VFIO group fd or get an existing one. - * - * This function is only relevant to linux and will return - * an error on BSD. - * - * @param iommu_group_num - * iommu group number - * - * @return - * > 0 group fd - * < 0 for errors - */ -__rte_internal -int -rte_vfio_get_group_fd(int iommu_group_num); - /** * @internal * Create a new container for device binding. @@ -297,42 +261,6 @@ int rte_vfio_container_assign_device(int vfio_container_fd, const char *sysfs_base, const char *dev_addr); -/** - * @internal - * Bind a IOMMU group to a container. - * - * @param container_fd - * the container's fd - * - * @param iommu_group_num - * the iommu group number to bind to container - * - * @return - * group fd if successful - * <0 if failed - */ -__rte_internal -int -rte_vfio_container_group_bind(int container_fd, int iommu_group_num); - -/** - * @internal - * Unbind a IOMMU group from a container. - * - * @param container_fd - * the container fd of container - * - * @param iommu_group_num - * the iommu group number to delete from container - * - * @return - * 0 if successful - * <0 if failed - */ -__rte_internal -int -rte_vfio_container_group_unbind(int container_fd, int iommu_group_num); - /** * @internal * Perform DMA mapping for devices in a container. diff --git a/lib/eal/linux/eal_vfio.c b/lib/eal/linux/eal_vfio.c index 02fec64658..7893d334eb 100644 --- a/lib/eal/linux/eal_vfio.c +++ b/lib/eal/linux/eal_vfio.c @@ -67,6 +67,9 @@ static int vfio_noiommu_dma_mem_map(int, uint64_t, uint64_t, uint64_t, int); static int vfio_dma_mem_map(struct vfio_config *vfio_cfg, uint64_t vaddr, uint64_t iova, uint64_t len, int do_map); +static int vfio_container_group_bind(int container_fd, int iommu_group_num); +static int vfio_container_group_unbind(int container_fd, int iommu_group_num); + /* IOMMU types we support */ static const struct vfio_iommu_type iommu_types[] = { /* x86 IOMMU, otherwise known as type 1 */ @@ -532,9 +535,8 @@ get_vfio_cfg_by_container_fd(int container_fd) return NULL; } -RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_group_fd) int -rte_vfio_get_group_fd(int iommu_group_num) +vfio_get_group_fd_by_num(int iommu_group_num) { struct vfio_config *vfio_cfg; @@ -731,9 +733,8 @@ vfio_sync_default_container(void) return -1; } -RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_clear_group) -int -rte_vfio_clear_group(int vfio_group_fd) +static int +vfio_clear_group(int vfio_group_fd) { int i; struct vfio_config *vfio_cfg; @@ -787,7 +788,7 @@ rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, return -1; /* get the actual group fd */ - vfio_group_fd = rte_vfio_get_group_fd(iommu_group_num); + vfio_group_fd = vfio_get_group_fd_by_num(iommu_group_num); if (vfio_group_fd < 0 && vfio_group_fd != -ENOENT) return -1; @@ -813,14 +814,14 @@ rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, EAL_LOG(ERR, "%s cannot get VFIO group status, " "error %i (%s)", dev_addr, errno, strerror(errno)); close(vfio_group_fd); - rte_vfio_clear_group(vfio_group_fd); + vfio_clear_group(vfio_group_fd); return -1; } else if (!(group_status.flags & VFIO_GROUP_FLAGS_VIABLE)) { EAL_LOG(ERR, "%s VFIO group is not viable! " "Not all devices in IOMMU group bound to VFIO or unbound", dev_addr); close(vfio_group_fd); - rte_vfio_clear_group(vfio_group_fd); + vfio_clear_group(vfio_group_fd); return -1; } @@ -841,7 +842,7 @@ rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, "%s cannot add VFIO group to container, error " "%i (%s)", dev_addr, errno, strerror(errno)); close(vfio_group_fd); - rte_vfio_clear_group(vfio_group_fd); + vfio_clear_group(vfio_group_fd); return -1; } @@ -865,7 +866,7 @@ rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, "%s failed to select IOMMU type", dev_addr); close(vfio_group_fd); - rte_vfio_clear_group(vfio_group_fd); + vfio_clear_group(vfio_group_fd); return -1; } /* lock memory hotplug before mapping and release it @@ -882,7 +883,7 @@ rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, "%i (%s)", dev_addr, errno, strerror(errno)); close(vfio_group_fd); - rte_vfio_clear_group(vfio_group_fd); + vfio_clear_group(vfio_group_fd); rte_mcfg_mem_read_unlock(); return -1; } @@ -951,7 +952,7 @@ rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, if (ret < 0) { EAL_LOG(ERR, "Could not sync default VFIO container"); close(vfio_group_fd); - rte_vfio_clear_group(vfio_group_fd); + vfio_clear_group(vfio_group_fd); return -1; } /* we have successfully initialized VFIO, notify user */ @@ -988,7 +989,7 @@ rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, EAL_LOG(WARNING, "Getting a vfio_dev_fd for %s failed", dev_addr); close(vfio_group_fd); - rte_vfio_clear_group(vfio_group_fd); + vfio_clear_group(vfio_group_fd); return -1; } @@ -1026,9 +1027,9 @@ rte_vfio_release_device(const char *sysfs_base, const char *dev_addr, } /* get the actual group fd */ - vfio_group_fd = rte_vfio_get_group_fd(iommu_group_num); + vfio_group_fd = vfio_get_group_fd_by_num(iommu_group_num); if (vfio_group_fd < 0) { - EAL_LOG(INFO, "rte_vfio_get_group_fd failed for %s", + EAL_LOG(INFO, "vfio_get_group_fd_by_num failed for %s", dev_addr); ret = vfio_group_fd; goto out; @@ -1064,7 +1065,7 @@ rte_vfio_release_device(const char *sysfs_base, const char *dev_addr, goto out; } - if (rte_vfio_clear_group(vfio_group_fd) < 0) { + if (vfio_clear_group(vfio_group_fd) < 0) { EAL_LOG(INFO, "Error when clearing group for %s", dev_addr); ret = -1; @@ -2091,7 +2092,7 @@ rte_vfio_container_destroy(int container_fd) for (i = 0; i < RTE_DIM(vfio_cfg->vfio_groups); i++) if (vfio_cfg->vfio_groups[i].group_num != -1) - rte_vfio_container_group_unbind(container_fd, + vfio_container_group_unbind(container_fd, vfio_cfg->vfio_groups[i].group_num); close(container_fd); @@ -2122,7 +2123,7 @@ rte_vfio_container_assign_device(int vfio_container_fd, const char *sysfs_base, return -1; } - ret = rte_vfio_container_group_bind(vfio_container_fd, + ret = vfio_container_group_bind(vfio_container_fd, iommu_group_num); if (ret < 0) { EAL_LOG(ERR, @@ -2134,9 +2135,8 @@ rte_vfio_container_assign_device(int vfio_container_fd, const char *sysfs_base, return 0; } -RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_group_bind) -int -rte_vfio_container_group_bind(int container_fd, int iommu_group_num) +static int +vfio_container_group_bind(int container_fd, int iommu_group_num) { struct vfio_config *vfio_cfg; @@ -2149,9 +2149,8 @@ rte_vfio_container_group_bind(int container_fd, int iommu_group_num) return vfio_get_group_fd(vfio_cfg, iommu_group_num); } -RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_group_unbind) -int -rte_vfio_container_group_unbind(int container_fd, int iommu_group_num) +static int +vfio_container_group_unbind(int container_fd, int iommu_group_num) { struct vfio_group *cur_grp = NULL; struct vfio_config *vfio_cfg; diff --git a/lib/eal/linux/eal_vfio.h b/lib/eal/linux/eal_vfio.h index 89c4b5ba45..30389fb274 100644 --- a/lib/eal/linux/eal_vfio.h +++ b/lib/eal/linux/eal_vfio.h @@ -51,6 +51,8 @@ vfio_set_iommu_type(int vfio_container_fd); int vfio_get_iommu_type(void); +int vfio_get_group_fd_by_num(int iommu_group_num); + /* check if we have any supported extensions */ int vfio_has_supported_extensions(int vfio_container_fd); diff --git a/lib/eal/linux/eal_vfio_mp_sync.c b/lib/eal/linux/eal_vfio_mp_sync.c index 22136f2e8b..3eaeef2fc8 100644 --- a/lib/eal/linux/eal_vfio_mp_sync.c +++ b/lib/eal/linux/eal_vfio_mp_sync.c @@ -34,7 +34,7 @@ vfio_mp_primary(const struct rte_mp_msg *msg, const void *peer) case SOCKET_REQ_GROUP: r->req = SOCKET_REQ_GROUP; r->group_num = m->group_num; - fd = rte_vfio_get_group_fd(m->group_num); + fd = vfio_get_group_fd_by_num(m->group_num); if (fd < 0 && fd != -ENOENT) r->result = SOCKET_ERR; else if (fd == -ENOENT) -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* [PATCH v3 14/20] vfio: cleanup and refactor 2025-11-18 16:29 ` [PATCH v3 00/20] Support VFIO cdev API in DPDK Anatoly Burakov ` (12 preceding siblings ...) 2025-11-18 16:29 ` [PATCH v3 13/20] vfio: remove group-based API Anatoly Burakov @ 2025-11-18 16:29 ` Anatoly Burakov 2025-11-18 16:29 ` [PATCH v3 15/20] bus/pci: use the new VFIO mode API Anatoly Burakov ` (5 subsequent siblings) 19 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2025-11-18 16:29 UTC (permalink / raw) To: dev, Wathsala Vithanage, Bruce Richardson, Nipun Gupta, Nikhil Agarwal, Hemant Agrawal, Sachin Saxena, Chenbo Xia, Ajit Khaparde, Vikas Gupta, Dimon Zhao, Kyo Liu, Leon Yu, Sam Chen, Tyler Retzlaff Currently, VFIO code is a bit of an incoherent mess internally, with API's bleeding into each other, inconsistent returns, and a certain amount of spaghetti stemming from organic growth. Refactor VFIO code to achieve the following goals: - Make all error handling consistent, and provide/document rte_errno values returned from API's to indicate various conditions. - Introduce new "VFIO mode" concept. This new API will tell caller if VFIO is enabled, and whether it is using group API, and whether it is running in no-IOMMU mode. - Decouple rte_vfio_setup_device semantics from PCI bus return convention. Currently, when device is not managed by VFIO, rte_vfio_setup_device will return 1, which is bus speak for "skip this device", however VFIO has nothing to do with PCI bus and should not follow its API conventions. - Perform device setup in device assign, and make device setup use shared code path with device assign and explicitly assuming default container. This is technically not necessary for group mode as device set up is a two-step process in that mode, but coming cdev mode will have a single-step device setup, and it would be easier if the worked the same way under the hood. - Make VFIO internals more readable. Introduce a lot of infrastructure and more explicit validation, rather than over-reliance on sentinel values and implicit assumptions. This will also make it easier to integrate cdev mode down the line, as it will rely on most of this infrastructure. This will change behavior of the following functions: - `rte_vfio_setup_device` - when the device is not managed by VFIO, the function will now return -1 with `rte_errno` set to ENODEV - `rte_vfio_get_group_num` - when the device is not managed by VFIO, the function will now return -1 with `rte_errno` set to ENODEV - `rte_vfio_container_destroy` - the function will now release and close all group and device resources associated with the container being destroyed by this call All users of `rte_vfio_setup_device` and `rte_vfio_get_group_num` have been adjusted to account for API change. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- config/arm/meson.build | 1 + config/meson.build | 1 + drivers/bus/cdx/cdx_vfio.c | 17 +- drivers/bus/fslmc/fslmc_vfio.c | 4 +- drivers/bus/pci/linux/pci_vfio.c | 19 +- drivers/crypto/bcmfs/bcmfs_vfio.c | 6 +- drivers/net/nbl/nbl_common/nbl_userdev.c | 2 +- lib/eal/freebsd/eal.c | 16 + lib/eal/include/rte_vfio.h | 250 ++- lib/eal/linux/eal_vfio.c | 2285 ++++++++-------------- lib/eal/linux/eal_vfio.h | 142 +- lib/eal/linux/eal_vfio_group.c | 983 ++++++++++ lib/eal/linux/eal_vfio_mp_sync.c | 38 +- lib/eal/linux/meson.build | 1 + 14 files changed, 2129 insertions(+), 1636 deletions(-) create mode 100644 lib/eal/linux/eal_vfio_group.c diff --git a/config/arm/meson.build b/config/arm/meson.build index c0aa21b57d..cd9ccccefd 100644 --- a/config/arm/meson.build +++ b/config/arm/meson.build @@ -148,6 +148,7 @@ implementer_cavium = { 'description': 'Cavium', 'flags': [ ['RTE_MAX_VFIO_GROUPS', 128], + ['RTE_MAX_VFIO_DEVICES', 256], ['RTE_MAX_LCORE', 96], ['RTE_MAX_NUMA_NODES', 2] ], diff --git a/config/meson.build b/config/meson.build index 0cb074ab95..6752584a7b 100644 --- a/config/meson.build +++ b/config/meson.build @@ -375,6 +375,7 @@ dpdk_conf.set('RTE_ENABLE_TRACE_FP', get_option('enable_trace_fp')) dpdk_conf.set('RTE_PKTMBUF_HEADROOM', get_option('pkt_mbuf_headroom')) # values which have defaults which may be overridden dpdk_conf.set('RTE_MAX_VFIO_GROUPS', 64) +dpdk_conf.set('RTE_MAX_VFIO_DEVICES', 256) dpdk_conf.set('RTE_DRIVER_MEMPOOL_BUCKET_SIZE_KB', 64) dpdk_conf.set('RTE_LIBRTE_DPAA2_USE_PHYS_IOVA', true) if dpdk_conf.get('RTE_ARCH_64') diff --git a/drivers/bus/cdx/cdx_vfio.c b/drivers/bus/cdx/cdx_vfio.c index 9bae264409..873f0f3460 100644 --- a/drivers/bus/cdx/cdx_vfio.c +++ b/drivers/bus/cdx/cdx_vfio.c @@ -22,6 +22,7 @@ #include <eal_export.h> #include <rte_eal_paging.h> +#include <rte_errno.h> #include <rte_malloc.h> #include <rte_vfio.h> @@ -402,8 +403,12 @@ cdx_vfio_map_resource_primary(struct rte_cdx_device *dev) ret = rte_vfio_setup_device(RTE_CDX_BUS_DEVICES_PATH, dev_name, &vfio_dev_fd); - if (ret) + if (ret < 0) { + /* Device not managed by VFIO - skip */ + if (rte_errno == ENODEV) + ret = 1; return ret; + } ret = rte_vfio_get_device_info(vfio_dev_fd, &device_info); if (ret) @@ -513,11 +518,13 @@ cdx_vfio_map_resource_secondary(struct rte_cdx_device *dev) return -1; } - ret = rte_vfio_setup_device(RTE_CDX_BUS_DEVICES_PATH, dev_name, - &vfio_dev_fd); - if (ret) + ret = rte_vfio_setup_device(RTE_CDX_BUS_DEVICES_PATH, dev_name, &vfio_dev_fd); + if (ret < 0) { + /* Device not managed by VFIO - skip */ + if (rte_errno == ENODEV) + ret = 1; return ret; - + } ret = rte_vfio_get_device_info(vfio_dev_fd, &device_info); if (ret) goto err_vfio_dev_fd; diff --git a/drivers/bus/fslmc/fslmc_vfio.c b/drivers/bus/fslmc/fslmc_vfio.c index 550d4e0e8d..4b11b38322 100644 --- a/drivers/bus/fslmc/fslmc_vfio.c +++ b/drivers/bus/fslmc/fslmc_vfio.c @@ -364,9 +364,9 @@ fslmc_get_group_id(const char *group_name, /* get group number */ ret = rte_vfio_get_group_num(SYSFS_FSL_MC_DEVICES, group_name, groupid); - if (ret <= 0) { + if (ret < 0) { DPAA2_BUS_ERR("Find %s IOMMU group", group_name); - if (ret < 0) + if (rte_errno != ENODEV) return ret; return -EIO; diff --git a/drivers/bus/pci/linux/pci_vfio.c b/drivers/bus/pci/linux/pci_vfio.c index 57ec9d7066..03a61e7765 100644 --- a/drivers/bus/pci/linux/pci_vfio.c +++ b/drivers/bus/pci/linux/pci_vfio.c @@ -20,6 +20,7 @@ #include <rte_malloc.h> #include <rte_vfio.h> #include <rte_eal.h> +#include <rte_errno.h> #include <bus_driver.h> #include <rte_spinlock.h> #include <rte_tailq.h> @@ -752,10 +753,13 @@ pci_vfio_map_resource_primary(struct rte_pci_device *dev) snprintf(pci_addr, sizeof(pci_addr), PCI_PRI_FMT, loc->domain, loc->bus, loc->devid, loc->function); - ret = rte_vfio_setup_device(rte_pci_get_sysfs_path(), pci_addr, - &vfio_dev_fd); - if (ret) + ret = rte_vfio_setup_device(rte_pci_get_sysfs_path(), pci_addr, &vfio_dev_fd); + if (ret < 0) { + /* Device not managed by VFIO - skip */ + if (rte_errno == ENODEV) + ret = 1; return ret; + } ret = rte_vfio_get_device_info(vfio_dev_fd, &device_info); if (ret) @@ -965,10 +969,13 @@ pci_vfio_map_resource_secondary(struct rte_pci_device *dev) return -1; } - ret = rte_vfio_setup_device(rte_pci_get_sysfs_path(), pci_addr, - &vfio_dev_fd); - if (ret) + ret = rte_vfio_setup_device(rte_pci_get_sysfs_path(), pci_addr, &vfio_dev_fd); + if (ret < 0) { + /* Device not managed by VFIO - skip */ + if (rte_errno == ENODEV) + ret = 1; return ret; + } ret = rte_vfio_get_device_info(vfio_dev_fd, &device_info); if (ret) diff --git a/drivers/crypto/bcmfs/bcmfs_vfio.c b/drivers/crypto/bcmfs/bcmfs_vfio.c index d00aaf1bb7..92d8de4443 100644 --- a/drivers/crypto/bcmfs/bcmfs_vfio.c +++ b/drivers/crypto/bcmfs/bcmfs_vfio.c @@ -9,6 +9,7 @@ #include <sys/mman.h> #include <sys/ioctl.h> +#include <rte_errno.h> #include <rte_vfio.h> #include "bcmfs_device.h" @@ -26,7 +27,10 @@ vfio_map_dev_obj(const char *path, const char *dev_obj, struct vfio_region_info reg_info = { .argsz = sizeof(reg_info) }; ret = rte_vfio_setup_device(path, dev_obj, dev_fd); - if (ret) { + if (ret < 0) { + /* Device not managed by VFIO - skip */ + if (rte_errno == ENODEV) + ret = 1; BCMFS_LOG(ERR, "VFIO Setting for device failed"); return ret; } diff --git a/drivers/net/nbl/nbl_common/nbl_userdev.c b/drivers/net/nbl/nbl_common/nbl_userdev.c index d641ecd9a8..d71d4fc035 100644 --- a/drivers/net/nbl/nbl_common/nbl_userdev.c +++ b/drivers/net/nbl/nbl_common/nbl_userdev.c @@ -413,7 +413,7 @@ static int nbl_mdev_map_device(struct nbl_adapter *adapter) "%s/%s/", rte_pci_get_sysfs_path(), dev_name); ret = rte_vfio_get_group_num(pathname, dev_name, &common->iommu_group_num); - if (ret <= 0) { + if (ret < 0) { NBL_LOG(ERR, "nbl vfio group number failed"); return -1; } diff --git a/lib/eal/freebsd/eal.c b/lib/eal/freebsd/eal.c index f66ac01ba2..7b3ce875a8 100644 --- a/lib/eal/freebsd/eal.c +++ b/lib/eal/freebsd/eal.c @@ -913,3 +913,19 @@ rte_vfio_container_assign_device(__rte_unused int vfio_container_fd, rte_errno = ENOTSUP; return -1; } + +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_device_info) +int +rte_vfio_get_device_info(__rte_unused int vfio_dev_fd, + __rte_unused struct vfio_device_info *device_info) +{ + rte_errno = ENOTSUP; + return -1; +} + +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_mode) +enum rte_vfio_mode +rte_vfio_get_mode(void) +{ + return RTE_VFIO_MODE_NONE; +} diff --git a/lib/eal/include/rte_vfio.h b/lib/eal/include/rte_vfio.h index 941b7d0541..b6d7b4591f 100644 --- a/lib/eal/include/rte_vfio.h +++ b/lib/eal/include/rte_vfio.h @@ -18,6 +18,7 @@ #include <stdint.h> #include <rte_compat.h> +#include <rte_common.h> #ifdef __cplusplus extern "C" { @@ -29,8 +30,7 @@ extern "C" { #define RTE_VFIO_CONTAINER_PATH "/dev/vfio/vfio" #define RTE_VFIO_GROUP_FMT "/dev/vfio/%u" #define RTE_VFIO_NOIOMMU_GROUP_FMT "/dev/vfio/noiommu-%u" -#define RTE_VFIO_NOIOMMU_MODE \ - "/sys/module/vfio/parameters/enable_unsafe_noiommu_mode" +#define RTE_VFIO_NOIOMMU_MODE "/sys/module/vfio/parameters/enable_unsafe_noiommu_mode" #endif /* RTE_EXEC_ENV_LINUX */ @@ -41,26 +41,47 @@ struct vfio_device_info; /** * @internal - * Setup vfio_cfg for the device identified by its address. - * It discovers the configured I/O MMU groups or sets a new one for the device. - * If a new groups is assigned, the DMA mapping is performed. + * Enumeration of VFIO operational modes. * - * This function is only relevant to linux and will return - * an error on BSD. + * These modes define how VFIO devices are accessed and managed: + * + * - RTE_VFIO_MODE_NONE: VFIO is not enabled. + * - RTE_VFIO_MODE_GROUP: Legacy group mode. + * - RTE_VFIO_MODE_NOIOMMU: Unsafe no-IOMMU mode. + */ +enum rte_vfio_mode { + RTE_VFIO_MODE_NONE = 0, /**< VFIO not enabled */ + RTE_VFIO_MODE_GROUP, /**< Group mode */ + RTE_VFIO_MODE_NOIOMMU, /**< Group mode with no IOMMU protection */ +}; + +/** + * @internal + * Set up a device managed by VFIO driver. + * + * If the device was not previously assigned to a container using + * `rte_vfio_container_assign_device()`, default container will be used. + * + * This function is only relevant on Linux. * * @param sysfs_base - * sysfs path prefix. - * + * Sysfs path prefix. * @param dev_addr - * device location. - * + * Device identifier. * @param vfio_dev_fd - * Pointer to VFIO fd, will be set to the opened device fd on success. + * Pointer to where VFIO device file descriptor will be stored. * * @return * 0 on success. - * <0 on failure. - * >1 if the device cannot be managed this way. + * <0 on failure, rte_errno is set. + * + * Possible rte_errno values include: + * - ENODEV - Device not managed by VFIO. + * - ENOSPC - No space in VFIO container to track the device. + * - EINVAL - Invalid parameters. + * - EIO - Error during underlying VFIO operations. + * - ENXIO - VFIO support not initialized. + * - ENOTSUP - Unsupported VFIO mode. */ __rte_internal int rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, @@ -68,99 +89,127 @@ int rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, /** * @internal - * Release a device mapped to a VFIO-managed I/O MMU group. + * Release a device managed by VFIO driver. * - * This function is only relevant to linux and will return - * an error on BSD. + * This function is only relevant on Linux. + * + * @note As a result of this function, all internal resources used by the device will be released, + * so if the device was using a non-default container, it will need to be reassigned to the + * container before it can be used again. * * @param sysfs_base - * sysfs path prefix. - * + * Sysfs path prefix. * @param dev_addr - * device location. - * + * Device identifier. * @param fd - * VFIO fd. + * A previously set up VFIO file descriptor. * * @return * 0 on success. - * <0 on failure. + * <0 on failure, rte_errno is set. + * + * Possible rte_errno values include: + * - ENOENT - Device not found in any container. + * - EINVAL - Invalid parameters. + * - EIO - Error during underlying VFIO operations. + * - ENXIO - VFIO support not initialized. + * - ENOTSUP - Unsupported VFIO mode. */ __rte_internal int rte_vfio_release_device(const char *sysfs_base, const char *dev_addr, int fd); /** * @internal - * Enable a VFIO-related kmod. + * Enable VFIO subsystem and check if specified kernel module is loaded. * - * This function is only relevant to linux and will return - * an error on BSD. + * In case of success, `rte_vfio_get_mode()` can be used to retrieve the VFIO mode in use. + * + * This function is only relevant on Linux. * * @param modname - * kernel module name. + * Kernel module name. * * @return * 0 on success. - * <0 on failure. + * <0 on failure, rte_errno is set. + * + * Possible rte_errno values include: + * - EINVAL - Invalid parameters. + * - ENXIO - VFIO support not initialized. + * - ENOTSUP - Operation not supported. */ __rte_internal int rte_vfio_enable(const char *modname); /** * @internal - * Check whether a VFIO-related kmod is enabled. + * Check if VFIO subsystem is initialized and a specified kernel module is loaded. * - * This function is only relevant to Linux. + * This function is only relevant on Linux. * * @param modname - * kernel module name. + * Kernel module name. * * @return - * 1 if true. - * 0 otherwise. + * 1 if enabled. + * 0 if not enabled or not supported. */ __rte_internal int rte_vfio_is_enabled(const char *modname); /** * @internal - * Whether VFIO NOIOMMU mode is enabled. + * Get current VFIO mode. * - * This function is only relevant to Linux. + * This function is only relevant on Linux. * * @return - * 1 if true. - * 0 if false. - * <0 for errors. + * VFIO mode currently in use. */ __rte_internal -int rte_vfio_noiommu_is_enabled(void); +enum rte_vfio_mode +rte_vfio_get_mode(void); + +/** + * @internal + * Check if VFIO NOIOMMU mode is enabled. + * + * This function is only relevant on Linux in group mode. + * + * @return + * 1 if enabled. + * 0 if not enabled or not supported. + */ +__rte_internal +int +rte_vfio_noiommu_is_enabled(void); /** * @internal * Parse IOMMU group number for a device. * - * This function is only relevant to linux and will return - * an error on BSD. + * This function is only relevant on Linux in group mode. * * @param sysfs_base - * sysfs path prefix. - * + * Sysfs path prefix. * @param dev_addr - * device location. - * + * Device identifier. * @param iommu_group_num - * iommu group number + * Pointer to where IOMMU group number will be stored. * * @return - * >0 on success - * 0 for non-existent group or VFIO - * <0 for errors + * 0 on success. + * <0 on failure, rte_errno is set. + * + * Possible rte_errno values include: + * - ENODEV - Device not managed by VFIO. + * - EINVAL - Invalid parameters. + * - ENXIO - VFIO support not initialized. + * - ENOTSUP - Unsupported VFIO mode. */ __rte_internal int -rte_vfio_get_group_num(const char *sysfs_base, - const char *dev_addr, int *iommu_group_num); +rte_vfio_get_group_num(const char *sysfs_base, const char *dev_addr, int *iommu_group_num); /** * @internal @@ -179,7 +228,12 @@ rte_vfio_get_group_num(const char *sysfs_base, * * @return * 0 on success. - * <0 on failure. + * <0 on failure, rte_errno is set. + * + * Possible rte_errno values include: + * - EINVAL - Invalid parameters. + * - ENXIO - VFIO support not initialized. + * - ENOTSUP - Unsupported VFIO mode. */ __rte_internal int @@ -187,14 +241,17 @@ rte_vfio_get_device_info(int vfio_dev_fd, struct vfio_device_info *device_info); /** * @internal - * Get the default VFIO container fd + * Get the default VFIO container file descriptor. * - * This function is only relevant to linux and will return - * an error on BSD. + * This function is only relevant on Linux. * * @return - * > 0 default container fd - * < 0 if VFIO is not enabled or not supported + * Non-negative container file descriptor on success. + * <0 on failure, rte_errno is set. + * + * Possible rte_errno values include: + * - ENXIO - VFIO support not initialized. + * - ENOTSUP - Unsupported VFIO mode. */ __rte_internal int @@ -202,7 +259,9 @@ rte_vfio_get_container_fd(void); /** * @internal - * Create a new container for device binding. + * Create a new VFIO container for device assignment and DMA mapping. + * + * This function is only relevant on Linux. * * @note Any newly allocated DPDK memory will not be mapped into these * containers by default, user needs to manage DMA mappings for @@ -213,8 +272,14 @@ rte_vfio_get_container_fd(void); * devices between multiple processes is not supported. * * @return - * the container fd if successful - * <0 if failed + * Non-negative container file descriptor on success. + * <0 on failure, rte_errno is set. + * + * Possible rte_errno values include: + * - ENOSPC - Maximum number of containers reached. + * - EIO - Underlying VFIO operation failed. + * - ENXIO - VFIO support not initialized. + * - ENOTSUP - Unsupported VFIO mode. */ __rte_internal int @@ -222,14 +287,22 @@ rte_vfio_container_create(void); /** * @internal - * Destroy the container, unbind all vfio groups within it. + * Destroy a VFIO container and unmap all devices assigned to it. + * + * This function is only relevant on Linux. * * @param container_fd - * the container fd to destroy + * File descriptor of container to destroy. * * @return - * 0 if successful - * <0 if failed + * 0 on success. + * <0 on failure, rte_errno is set. + * + * Possible rte_errno values include: + * - ENODEV - Container not managed by VFIO. + * - EINVAL - Invalid container file descriptor. + * - ENXIO - VFIO support not initialized. + * - ENOTSUP - Unsupported VFIO mode. */ __rte_internal int @@ -255,32 +328,45 @@ rte_vfio_container_destroy(int container_fd); * @return * 0 on success. * <0 on failure, rte_errno is set. + * + * Possible rte_errno values include: + * - ENODEV - Device not managed by VFIO. + * - EEXIST - Device already assigned to the container. + * - ENOSPC - No space in VFIO container to assign device. + * - EINVAL - Invalid container file descriptor. + * - EIO - Error during underlying VFIO operations. + * - ENXIO - VFIO support not initialized. + * - ENOTSUP - Unsupported VFIO mode. */ __rte_internal int -rte_vfio_container_assign_device(int vfio_container_fd, const char *sysfs_base, - const char *dev_addr); +rte_vfio_container_assign_device(int vfio_container_fd, + const char *sysfs_base, const char *dev_addr); /** * @internal * Perform DMA mapping for devices in a container. * - * @param container_fd - * the specified container fd. Use RTE_VFIO_DEFAULT_CONTAINER_FD to - * use the default container. + * This function is only relevant on Linux. * + * @param container_fd + * Container file descriptor. Use RTE_VFIO_DEFAULT_CONTAINER_FD to use the default container. * @param vaddr * Starting virtual address of memory to be mapped. - * * @param iova * Starting IOVA address of memory to be mapped. - * * @param len * Length of memory segment being mapped. * * @return - * 0 if successful - * <0 if failed + * 0 on success. + * <0 on failure, rte_errno is set. + * + * Possible rte_errno values include: + * - EIO - DMA mapping operation failed. + * - EINVAL - Invalid parameters. + * - ENXIO - VFIO support not initialized. + * - ENOTSUP - Unsupported VFIO mode. */ __rte_internal int @@ -291,22 +377,26 @@ rte_vfio_container_dma_map(int container_fd, uint64_t vaddr, * @internal * Perform DMA unmapping for devices in a container. * - * @param container_fd - * the specified container fd. Use RTE_VFIO_DEFAULT_CONTAINER_FD to - * use the default container. + * This function is only relevant on Linux. * + * @param container_fd + * Container file descriptor. Use RTE_VFIO_DEFAULT_CONTAINER_FD to use the default container. * @param vaddr * Starting virtual address of memory to be unmapped. - * * @param iova * Starting IOVA address of memory to be unmapped. - * * @param len * Length of memory segment being unmapped. * * @return - * 0 if successful - * <0 if failed + * 0 on success. + * <0 on failure, rte_errno is set. + * + * Possible rte_errno values include: + * - EIO - DMA unmapping operation failed. + * - EINVAL - Invalid parameters. + * - ENXIO - VFIO support not initialized. + * - ENOTSUP - Unsupported VFIO mode. */ __rte_internal int diff --git a/lib/eal/linux/eal_vfio.c b/lib/eal/linux/eal_vfio.c index 7893d334eb..8545552308 100644 --- a/lib/eal/linux/eal_vfio.c +++ b/lib/eal/linux/eal_vfio.c @@ -9,6 +9,7 @@ #include <fcntl.h> #include <unistd.h> #include <sys/ioctl.h> +#include <sys/stat.h> #include <dirent.h> #include <rte_errno.h> @@ -24,80 +25,39 @@ #include "eal_private.h" #include "eal_internal_cfg.h" -#define VFIO_MEM_EVENT_CLB_NAME "vfio_mem_event_clb" - -/* hot plug/unplug of VFIO groups may cause all DMA maps to be dropped. we can - * recreate the mappings for DPDK segments, but we cannot do so for memory that - * was registered by the user themselves, so we need to store the user mappings - * somewhere, to recreate them later. +/* + * rte_errno convention: + * + * - EINVAL: invalid parameters + * - ENOTSUP: current mode does not support this operation + * - ENOXIO: VFIO not initialized + * - ENODEV: device not managed by VFIO + * - ENOSPC: no space in config + * - EEXIST: device already assigned + * - ENOENT: group or device not found + * - EIO: underlying VFIO operation failed */ -#define EAL_VFIO_MAX_USER_MEM_MAPS 256 -struct user_mem_map { - uint64_t addr; /**< start VA */ - uint64_t iova; /**< start IOVA */ - uint64_t len; /**< total length of the mapping */ - uint64_t chunk; /**< this mapping can be split in chunks of this size */ -}; -struct user_mem_maps { - rte_spinlock_recursive_t lock; - int n_maps; - struct user_mem_map maps[EAL_VFIO_MAX_USER_MEM_MAPS]; +/* functions can fail for multiple reasons, and errno is tedious */ +enum vfio_result { + VFIO_SUCCESS, + VFIO_ERROR, + VFIO_EXISTS, + VFIO_NOT_SUPPORTED, + VFIO_NOT_MANAGED, + VFIO_NOT_FOUND, + VFIO_NO_SPACE, }; -struct vfio_config { - int vfio_enabled; - int vfio_container_fd; - int vfio_active_groups; - const struct vfio_iommu_type *vfio_iommu_type; - struct vfio_group vfio_groups[RTE_MAX_VFIO_GROUPS]; - struct user_mem_maps mem_maps; +struct container containers[RTE_MAX_VFIO_CONTAINERS] = {0}; +struct vfio_config vfio_cfg = { + .mode = RTE_VFIO_MODE_NONE, + .default_cfg = &containers[0] }; -/* per-process VFIO config */ -static struct vfio_config vfio_cfgs[RTE_MAX_VFIO_CONTAINERS]; -static struct vfio_config *default_vfio_cfg = &vfio_cfgs[0]; - -static int vfio_type1_dma_map(int); -static int vfio_type1_dma_mem_map(int, uint64_t, uint64_t, uint64_t, int); -static int vfio_spapr_dma_map(int); -static int vfio_spapr_dma_mem_map(int, uint64_t, uint64_t, uint64_t, int); -static int vfio_noiommu_dma_map(int); -static int vfio_noiommu_dma_mem_map(int, uint64_t, uint64_t, uint64_t, int); -static int vfio_dma_mem_map(struct vfio_config *vfio_cfg, uint64_t vaddr, +static int vfio_dma_mem_map(struct container *cfg, uint64_t vaddr, uint64_t iova, uint64_t len, int do_map); -static int vfio_container_group_bind(int container_fd, int iommu_group_num); -static int vfio_container_group_unbind(int container_fd, int iommu_group_num); - -/* IOMMU types we support */ -static const struct vfio_iommu_type iommu_types[] = { - /* x86 IOMMU, otherwise known as type 1 */ - { - .type_id = VFIO_TYPE1_IOMMU, - .name = "Type 1", - .partial_unmap = false, - .dma_map_func = &vfio_type1_dma_map, - .dma_user_map_func = &vfio_type1_dma_mem_map - }, - /* ppc64 IOMMU, otherwise known as spapr */ - { - .type_id = VFIO_SPAPR_TCE_v2_IOMMU, - .name = "sPAPR", - .partial_unmap = true, - .dma_map_func = &vfio_spapr_dma_map, - .dma_user_map_func = &vfio_spapr_dma_mem_map - }, - /* IOMMU-less mode */ - { - .type_id = VFIO_NOIOMMU_IOMMU, - .name = "No-IOMMU", - .partial_unmap = true, - .dma_map_func = &vfio_noiommu_dma_map, - .dma_user_map_func = &vfio_noiommu_dma_mem_map - }, -}; - static int is_null_map(const struct user_mem_map *map) { @@ -353,279 +313,127 @@ compact_user_maps(struct user_mem_maps *user_mem_maps) sizeof(user_mem_maps->maps[0]), user_mem_map_cmp); } -static int -vfio_open_group_fd(int iommu_group_num, bool mp_request) +/* + * we will rely on kernel to not allow user to assign the same device to different containers, but + * kernel will not prevent mapping the same device twice using two different fd's, so we need to + * deduplicate our internal config to make sure we only store unique device fd's. + */ +static bool +fd_is_same(int fd1, int fd2) { - int vfio_group_fd; - char filename[PATH_MAX]; - struct rte_mp_msg mp_req, *mp_rep; - struct rte_mp_reply mp_reply = {0}; - struct timespec ts = {.tv_sec = 5, .tv_nsec = 0}; - struct vfio_mp_param *p = (struct vfio_mp_param *)mp_req.param; + struct stat st1, st2; - /* if not requesting via mp, open the group locally */ - if (!mp_request) { - /* try regular group format */ - snprintf(filename, sizeof(filename), RTE_VFIO_GROUP_FMT, iommu_group_num); - vfio_group_fd = open(filename, O_RDWR); - if (vfio_group_fd < 0) { - /* if file not found, it's not an error */ - if (errno != ENOENT) { - EAL_LOG(ERR, "Cannot open %s: %s", - filename, strerror(errno)); - return -1; - } + if (fd1 < 0 || fd2 < 0) + return false; - /* special case: try no-IOMMU path as well */ - snprintf(filename, sizeof(filename), RTE_VFIO_NOIOMMU_GROUP_FMT, - iommu_group_num); - vfio_group_fd = open(filename, O_RDWR); - if (vfio_group_fd < 0) { - if (errno != ENOENT) { - EAL_LOG(ERR, - "Cannot open %s: %s", - filename, strerror(errno)); - return -1; - } - return -ENOENT; - } - /* noiommu group found */ - } + if (fstat(fd1, &st1) < 0) + return false; + if (fstat(fd2, &st2) < 0) + return false; - return vfio_group_fd; - } - /* if we're in a secondary process, request group fd from the primary - * process via mp channel. - */ - p->req = SOCKET_REQ_GROUP; - p->group_num = iommu_group_num; - strcpy(mp_req.name, EAL_VFIO_MP); - mp_req.len_param = sizeof(*p); - mp_req.num_fds = 0; - - vfio_group_fd = -1; - if (rte_mp_request_sync(&mp_req, &mp_reply, &ts) == 0 && - mp_reply.nb_received == 1) { - mp_rep = &mp_reply.msgs[0]; - p = (struct vfio_mp_param *)mp_rep->param; - if (p->result == SOCKET_OK && mp_rep->num_fds == 1) { - vfio_group_fd = mp_rep->fds[0]; - } else if (p->result == SOCKET_NO_FD) { - EAL_LOG(ERR, "Bad VFIO group fd"); - vfio_group_fd = -ENOENT; - } - } - - free(mp_reply.msgs); - if (vfio_group_fd < 0 && vfio_group_fd != -ENOENT) - EAL_LOG(ERR, "Cannot request VFIO group fd"); - return vfio_group_fd; -} - -static struct vfio_config * -get_vfio_cfg_by_group_num(int iommu_group_num) -{ - struct vfio_config *vfio_cfg; - unsigned int i, j; - - for (i = 0; i < RTE_DIM(vfio_cfgs); i++) { - vfio_cfg = &vfio_cfgs[i]; - for (j = 0; j < RTE_DIM(vfio_cfg->vfio_groups); j++) { - if (vfio_cfg->vfio_groups[j].group_num == - iommu_group_num) - return vfio_cfg; - } - } - - return NULL; -} - -static int -vfio_get_group_fd(struct vfio_config *vfio_cfg, - int iommu_group_num) -{ - struct vfio_group *cur_grp = NULL; - int vfio_group_fd; - unsigned int i; - - /* check if we already have the group descriptor open */ - for (i = 0; i < RTE_DIM(vfio_cfg->vfio_groups); i++) - if (vfio_cfg->vfio_groups[i].group_num == iommu_group_num) - return vfio_cfg->vfio_groups[i].fd; - - /* Lets see first if there is room for a new group */ - if (vfio_cfg->vfio_active_groups == RTE_DIM(vfio_cfg->vfio_groups)) { - EAL_LOG(ERR, "Maximum number of VFIO groups reached!"); - return -1; - } - - /* Now lets get an index for the new group */ - for (i = 0; i < RTE_DIM(vfio_cfg->vfio_groups); i++) - if (vfio_cfg->vfio_groups[i].group_num == -1) { - cur_grp = &vfio_cfg->vfio_groups[i]; - break; - } - - /* This should not happen */ - if (cur_grp == NULL) { - EAL_LOG(ERR, "No VFIO group free slot found"); - return -1; - } - - /* - * When opening a group fd, we need to decide whether to open it locally - * or request it from the primary process via mp_sync. - * - * For the default container, secondary processes use mp_sync so that - * the primary process tracks the group fd and maintains VFIO state - * across all processes. - * - * For custom containers, we open the group fd locally in each process - * since custom containers are process-local and the primary has no - * knowledge of them. Requesting a group fd from the primary for a - * container it doesn't know about would be incorrect. - */ - const struct internal_config *internal_conf = eal_get_internal_configuration(); - bool mp_request = (internal_conf->process_type == RTE_PROC_SECONDARY) && - (vfio_cfg == default_vfio_cfg); - - vfio_group_fd = vfio_open_group_fd(iommu_group_num, mp_request); - if (vfio_group_fd < 0) { - EAL_LOG(ERR, "Failed to open VFIO group %d", - iommu_group_num); - return vfio_group_fd; - } - - cur_grp->group_num = iommu_group_num; - cur_grp->fd = vfio_group_fd; - vfio_cfg->vfio_active_groups++; - - return vfio_group_fd; + return st1.st_dev == st2.st_dev && st1.st_ino == st2.st_ino; } -static struct vfio_config * -get_vfio_cfg_by_group_fd(int vfio_group_fd) +bool +vfio_container_is_default(struct container *cfg) { - struct vfio_config *vfio_cfg; - unsigned int i, j; - - for (i = 0; i < RTE_DIM(vfio_cfgs); i++) { - vfio_cfg = &vfio_cfgs[i]; - for (j = 0; j < RTE_DIM(vfio_cfg->vfio_groups); j++) - if (vfio_cfg->vfio_groups[j].fd == vfio_group_fd) - return vfio_cfg; - } - - return NULL; + return cfg == vfio_cfg.default_cfg; } -static struct vfio_config * -get_vfio_cfg_by_container_fd(int container_fd) +static struct container * +vfio_container_get_by_fd(int container_fd) { - unsigned int i; + struct container *cfg; if (container_fd == RTE_VFIO_DEFAULT_CONTAINER_FD) - return default_vfio_cfg; + return vfio_cfg.default_cfg; - for (i = 0; i < RTE_DIM(vfio_cfgs); i++) { - if (vfio_cfgs[i].vfio_container_fd == container_fd) - return &vfio_cfgs[i]; + CONTAINER_FOREACH_ACTIVE(cfg) { + if (cfg->container_fd == container_fd) + return cfg; } - return NULL; } -int -vfio_get_group_fd_by_num(int iommu_group_num) +static struct container * +vfio_container_get_by_group_num(int group_num) { - struct vfio_config *vfio_cfg; - - /* get the vfio_config it belongs to */ - vfio_cfg = get_vfio_cfg_by_group_num(iommu_group_num); - vfio_cfg = vfio_cfg ? vfio_cfg : default_vfio_cfg; + struct container *cfg; + struct vfio_group *grp; - return vfio_get_group_fd(vfio_cfg, iommu_group_num); + CONTAINER_FOREACH_ACTIVE(cfg) { + GROUP_FOREACH_ACTIVE(cfg, grp) + if (grp->group_num == group_num) + return cfg; + } + return NULL; } -static int -get_vfio_group_idx(int vfio_group_fd) +static struct container * +vfio_container_create(void) { - struct vfio_config *vfio_cfg; - unsigned int i, j; - - for (i = 0; i < RTE_DIM(vfio_cfgs); i++) { - vfio_cfg = &vfio_cfgs[i]; - for (j = 0; j < RTE_DIM(vfio_cfg->vfio_groups); j++) - if (vfio_cfg->vfio_groups[j].fd == vfio_group_fd) - return j; + struct container *cfg; + + /* find an unused container config */ + CONTAINER_FOREACH(cfg) { + if (!cfg->active) { + *cfg = CONTAINER_INITIALIZER; + cfg->active = true; + return cfg; + } } - - return -1; + /* no space */ + return NULL; } static void -vfio_group_device_get(int vfio_group_fd) +vfio_container_erase(struct container *cfg) { - struct vfio_config *vfio_cfg; - int i; + if (cfg->container_fd >= 0 && close(cfg->container_fd)) + EAL_LOG(ERR, "Error when closing container, %d (%s)", errno, strerror(errno)); - vfio_cfg = get_vfio_cfg_by_group_fd(vfio_group_fd); - if (vfio_cfg == NULL) { - EAL_LOG(ERR, "Invalid VFIO group fd!"); - return; - } - - i = get_vfio_group_idx(vfio_group_fd); - if (i < 0) - EAL_LOG(ERR, "Wrong VFIO group index (%d)", i); - else - vfio_cfg->vfio_groups[i].devices++; + *cfg = (struct container){0}; } -static void -vfio_group_device_put(int vfio_group_fd) +static struct vfio_device * +vfio_device_create(struct container *cfg) { - struct vfio_config *vfio_cfg; - int i; + struct vfio_device *dev; - vfio_cfg = get_vfio_cfg_by_group_fd(vfio_group_fd); - if (vfio_cfg == NULL) { - EAL_LOG(ERR, "Invalid VFIO group fd!"); - return; - } + /* is there space? */ + if (cfg->n_devices == RTE_DIM(cfg->devices)) + return NULL; - i = get_vfio_group_idx(vfio_group_fd); - if (i < 0) - EAL_LOG(ERR, "Wrong VFIO group index (%d)", i); - else - vfio_cfg->vfio_groups[i].devices--; -} + DEVICE_FOREACH(cfg, dev) { + if (dev->active) + continue; + dev->active = true; + /* set to invalid fd */ + dev->fd = -1; -static int -vfio_group_device_count(int vfio_group_fd) -{ - struct vfio_config *vfio_cfg; - int i; - - vfio_cfg = get_vfio_cfg_by_group_fd(vfio_group_fd); - if (vfio_cfg == NULL) { - EAL_LOG(ERR, "Invalid VFIO group fd!"); - return -1; + cfg->n_devices++; + return dev; } + /* should not happen */ + EAL_LOG(WARNING, "Could not find space in device list for container"); + return NULL; +} - i = get_vfio_group_idx(vfio_group_fd); - if (i < 0) { - EAL_LOG(ERR, "Wrong VFIO group index (%d)", i); - return -1; - } +static void +vfio_device_erase(struct container *cfg, struct vfio_device *dev) +{ + if (dev->fd >= 0 && close(dev->fd)) + EAL_LOG(ERR, "Error when closing device, %d (%s)", errno, strerror(errno)); - return vfio_cfg->vfio_groups[i].devices; + *dev = (struct vfio_device){0}; + cfg->n_devices--; } static void vfio_mem_event_callback(enum rte_mem_event type, const void *addr, size_t len, void *arg __rte_unused) { + struct container *cfg = vfio_cfg.default_cfg; struct rte_memseg_list *msl; struct rte_memseg *ms; size_t cur_len = 0; @@ -640,11 +448,9 @@ vfio_mem_event_callback(enum rte_mem_event type, const void *addr, size_t len, /* Maintain granularity of DMA map/unmap to memseg size */ for (; cur_len < len; cur_len += page_sz) { if (type == RTE_MEM_EVENT_ALLOC) - vfio_dma_mem_map(default_vfio_cfg, vfio_va, - vfio_va, page_sz, 1); + vfio_dma_mem_map(cfg, vfio_va, vfio_va, page_sz, 1); else - vfio_dma_mem_map(default_vfio_cfg, vfio_va, - vfio_va, page_sz, 0); + vfio_dma_mem_map(cfg, vfio_va, vfio_va, page_sz, 0); vfio_va += page_sz; } @@ -662,11 +468,9 @@ vfio_mem_event_callback(enum rte_mem_event type, const void *addr, size_t len, goto next; } if (type == RTE_MEM_EVENT_ALLOC) - vfio_dma_mem_map(default_vfio_cfg, ms->addr_64, - ms->iova, ms->len, 1); + vfio_dma_mem_map(cfg, ms->addr_64, ms->iova, ms->len, 1); else - vfio_dma_mem_map(default_vfio_cfg, ms->addr_64, - ms->iova, ms->len, 0); + vfio_dma_mem_map(cfg, ms->addr_64, ms->iova, ms->len, 0); next: cur_len += ms->len; ++ms; @@ -674,445 +478,546 @@ vfio_mem_event_callback(enum rte_mem_event type, const void *addr, size_t len, } static int -vfio_sync_default_container(void) +vfio_register_mem_event_callback(void) { - struct rte_mp_msg mp_req, *mp_rep; - struct rte_mp_reply mp_reply = {0}; - struct timespec ts = {.tv_sec = 5, .tv_nsec = 0}; - struct vfio_mp_param *p = (struct vfio_mp_param *)mp_req.param; - int iommu_type_id; - unsigned int i; + int ret; - /* cannot be called from primary */ - if (rte_eal_process_type() != RTE_PROC_SECONDARY) - return -1; + ret = rte_mem_event_callback_register(VFIO_MEM_EVENT_CLB_NAME, + vfio_mem_event_callback, NULL); - /* default container fd should have been opened in rte_vfio_enable() */ - if (!default_vfio_cfg->vfio_enabled || - default_vfio_cfg->vfio_container_fd < 0) { - EAL_LOG(ERR, "VFIO support is not initialized"); + if (ret && rte_errno != ENOTSUP) { + EAL_LOG(ERR, "Could not install memory event callback for VFIO"); return -1; } + if (ret) + EAL_LOG(DEBUG, "Memory event callbacks not supported"); + else + EAL_LOG(DEBUG, "Installed memory event callback for VFIO"); - /* find default container's IOMMU type */ - p->req = SOCKET_REQ_IOMMU_TYPE; - strcpy(mp_req.name, EAL_VFIO_MP); - mp_req.len_param = sizeof(*p); - mp_req.num_fds = 0; - - iommu_type_id = -1; - if (rte_mp_request_sync(&mp_req, &mp_reply, &ts) == 0 && - mp_reply.nb_received == 1) { - mp_rep = &mp_reply.msgs[0]; - p = (struct vfio_mp_param *)mp_rep->param; - if (p->result == SOCKET_OK) - iommu_type_id = p->iommu_type_id; - } - free(mp_reply.msgs); - if (iommu_type_id < 0) { - EAL_LOG(ERR, - "Could not get IOMMU type for default container"); - return -1; - } - - /* we now have an fd for default container, as well as its IOMMU type. - * now, set up default VFIO container config to match. - */ - for (i = 0; i < RTE_DIM(iommu_types); i++) { - const struct vfio_iommu_type *t = &iommu_types[i]; - if (t->type_id != iommu_type_id) - continue; - - /* we found our IOMMU type */ - default_vfio_cfg->vfio_iommu_type = t; - - return 0; - } - EAL_LOG(ERR, "Could not find IOMMU type id (%i)", - iommu_type_id); - return -1; + return 0; } static int -vfio_clear_group(int vfio_group_fd) +vfio_setup_dma_mem(struct container *cfg) { - int i; - struct vfio_config *vfio_cfg; + struct user_mem_maps *user_mem_maps = &cfg->mem_maps; + int i, ret; - vfio_cfg = get_vfio_cfg_by_group_fd(vfio_group_fd); - if (vfio_cfg == NULL) { - EAL_LOG(ERR, "Invalid VFIO group fd!"); + /* do we need to map DPDK-managed memory? */ + if (vfio_container_is_default(cfg) && rte_eal_process_type() == RTE_PROC_PRIMARY) + ret = vfio_cfg.ops->dma_map_func(cfg); + else + ret = 0; + if (ret) { + EAL_LOG(ERR, "DMA remapping failed, error %i (%s)", + errno, strerror(errno)); return -1; } - i = get_vfio_group_idx(vfio_group_fd); - if (i < 0) - return -1; - vfio_cfg->vfio_groups[i].group_num = -1; - vfio_cfg->vfio_groups[i].fd = -1; - vfio_cfg->vfio_groups[i].devices = 0; - vfio_cfg->vfio_active_groups--; + /* + * not all IOMMU types support DMA mapping, but if we have mappings in the list - that + * means we have previously mapped something successfully, so we can be sure that DMA + * mapping is supported. + */ + for (i = 0; i < user_mem_maps->n_maps; i++) { + struct user_mem_map *map; + map = &user_mem_maps->maps[i]; + + ret = vfio_cfg.ops->dma_user_map_func(cfg, map->addr, map->iova, map->len, 1); + if (ret) { + EAL_LOG(ERR, "Couldn't map user memory for DMA: " + "va: 0x%" PRIx64 " " + "iova: 0x%" PRIx64 " " + "len: 0x%" PRIu64, + map->addr, map->iova, + map->len); + return -1; + } + } return 0; } +static enum vfio_result +vfio_group_assign_device(struct container *cfg, const char *sysfs_base, + const char *dev_addr, struct vfio_device **out_dev) +{ + struct vfio_group_config *group_cfg = &cfg->group_cfg; + struct vfio_group *grp; + struct vfio_device *idev, *dev; + int iommu_group_num; + enum vfio_result res; + int ret; + + /* allocate new device in config */ + dev = vfio_device_create(cfg); + if (dev == NULL) { + EAL_LOG(ERR, "No space to track new VFIO device"); + return VFIO_NO_SPACE; + } + + /* remember to register mem event callback for default container in primary */ + bool need_clb = vfio_container_is_default(cfg) && + rte_eal_process_type() == RTE_PROC_PRIMARY; + + /* get group number for this device */ + ret = vfio_group_get_num(sysfs_base, dev_addr, &iommu_group_num); + if (ret < 0) { + EAL_LOG(ERR, "Cannot get IOMMU group for %s", dev_addr); + res = VFIO_ERROR; + goto device_erase; + } else if (ret == 0) { + res = VFIO_NOT_MANAGED; + goto device_erase; + } + + /* group may already exist as multiple devices may share group */ + grp = vfio_group_get_by_num(cfg, iommu_group_num); + if (grp == NULL) { + /* no device currently uses this group, create it */ + grp = vfio_group_create(cfg, iommu_group_num); + if (grp == NULL) { + EAL_LOG(ERR, "Cannot allocate group for device %s", dev_addr); + res = VFIO_NO_SPACE; + goto device_erase; + } + + /* open group fd */ + ret = vfio_group_open_fd(cfg, grp); + if (ret == -ENOENT) { + EAL_LOG(DEBUG, "Device %s (IOMMU group %d) not managed by VFIO", + dev_addr, iommu_group_num); + res = VFIO_NOT_MANAGED; + goto group_erase; + } else if (ret < 0) { + EAL_LOG(ERR, "Cannot open VFIO group %d for device %s", + iommu_group_num, dev_addr); + res = VFIO_ERROR; + goto group_erase; + } + + /* prepare group (viability + container attach) */ + ret = vfio_group_prepare(cfg, grp); + if (ret < 0) { + res = VFIO_ERROR; + goto group_erase; + } + + /* set up IOMMU type once per container */ + if (!group_cfg->iommu_type_set) { + ret = vfio_group_setup_iommu(cfg); + if (ret < 0) { + res = VFIO_ERROR; + goto group_erase; + } + group_cfg->iommu_type_set = true; + } + + /* set up DMA memory once per container */ + if (!group_cfg->dma_setup_done) { + rte_spinlock_recursive_lock(&cfg->mem_maps.lock); + ret = vfio_setup_dma_mem(cfg); + rte_spinlock_recursive_unlock(&cfg->mem_maps.lock); + if (ret < 0) { + EAL_LOG(ERR, "DMA remapping for %s failed", dev_addr); + res = VFIO_ERROR; + goto group_erase; + } + group_cfg->dma_setup_done = true; + } + + /* set up mem event callback if needed */ + if (need_clb && !group_cfg->mem_event_clb_set) { + ret = vfio_register_mem_event_callback(); + if (ret < 0) { + res = VFIO_ERROR; + goto group_erase; + } + group_cfg->mem_event_clb_set = true; + } + } + + /* open dev fd */ + ret = vfio_group_setup_device_fd(dev_addr, grp, dev); + if (ret < 0) { + EAL_LOG(ERR, "Cannot open VFIO device %s, error %i (%s)", + dev_addr, errno, strerror(errno)); + res = VFIO_ERROR; + goto group_erase; + } + + /* + * we want to prevent user from assigning devices twice to prevent resource leaks, but for + * group mode this is not trivial, as there is no direct way to know which fd belongs to + * which group/device, except for directly comparing fd's with stat. so, that's what we're + * going to do. we do not need to look in other configs as if we were to attempt to use a + * different container, the kernel wouldn't have allowed us to bind the group to the + * container in the first place. + */ + DEVICE_FOREACH_ACTIVE(cfg, idev) { + if (idev != dev && fd_is_same(idev->fd, dev->fd)) { + EAL_LOG(ERR, "Device %s already assigned to this container", + dev_addr); + res = VFIO_EXISTS; + *out_dev = idev; + goto dev_remove; + } + } + *out_dev = dev; + return VFIO_SUCCESS; +dev_remove: + /* device will be closed, but we still need to keep the group consistent */ + grp->n_devices--; +group_erase: + /* this may be a pre-existing group so only erase it if it has no devices */ + if (grp->n_devices == 0) + vfio_group_erase(cfg, grp); + /* if we registered callback, unregister it */ + if (group_cfg->n_groups == 0 && group_cfg->mem_event_clb_set) { + rte_mem_event_callback_unregister(VFIO_MEM_EVENT_CLB_NAME, NULL); + group_cfg->mem_event_clb_set = false; + } +device_erase: + vfio_device_erase(cfg, dev); + return res; +} + +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_assign_device) +int +rte_vfio_container_assign_device(int container_fd, const char *sysfs_base, const char *dev_addr) +{ + struct container *cfg; + enum vfio_result res; + struct vfio_device *dev; + + if (sysfs_base == NULL || dev_addr == NULL) { + rte_errno = EINVAL; + return -1; + } + + if (vfio_cfg.mode == RTE_VFIO_MODE_NONE) { + EAL_LOG(ERR, "VFIO support not initialized"); + rte_errno = ENXIO; + return -1; + } + + cfg = vfio_container_get_by_fd(container_fd); + if (cfg == NULL) { + EAL_LOG(ERR, "Invalid VFIO container fd"); + rte_errno = EINVAL; + return -1; + } + /* protect memory configuration while setting up IOMMU/DMA */ + rte_mcfg_mem_read_lock(); + + switch (vfio_cfg.mode) { + case RTE_VFIO_MODE_GROUP: + case RTE_VFIO_MODE_NOIOMMU: + res = vfio_group_assign_device(cfg, sysfs_base, dev_addr, &dev); + break; + default: + EAL_LOG(ERR, "Unsupported VFIO mode"); + res = VFIO_NOT_SUPPORTED; + break; + } + rte_mcfg_mem_read_unlock(); + + switch (res) { + case VFIO_SUCCESS: + return 0; + case VFIO_EXISTS: + rte_errno = EEXIST; + return -1; + case VFIO_NOT_MANAGED: + EAL_LOG(DEBUG, "Device %s not managed by VFIO", dev_addr); + rte_errno = ENODEV; + return -1; + case VFIO_NO_SPACE: + EAL_LOG(ERR, "No space in VFIO container to assign device %s", dev_addr); + rte_errno = ENOSPC; + return -1; + default: + EAL_LOG(ERR, "Error assigning device %s to container", dev_addr); + rte_errno = EIO; + return -1; + } +} + RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_setup_device) int rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, int *vfio_dev_fd) { - struct vfio_group_status group_status = { - .argsz = sizeof(group_status) - }; - struct vfio_config *vfio_cfg; - struct user_mem_maps *user_mem_maps; - int vfio_container_fd; - int vfio_group_fd; - int iommu_group_num; - rte_uuid_t vf_token; - int i, ret; - const struct internal_config *internal_conf = - eal_get_internal_configuration(); - - /* get group number */ - ret = rte_vfio_get_group_num(sysfs_base, dev_addr, &iommu_group_num); - if (ret == 0) { - EAL_LOG(NOTICE, - "%s not managed by VFIO driver, skipping", - dev_addr); - return 1; - } - - /* if negative, something failed */ - if (ret < 0) - return -1; - - /* get the actual group fd */ - vfio_group_fd = vfio_get_group_fd_by_num(iommu_group_num); - if (vfio_group_fd < 0 && vfio_group_fd != -ENOENT) - return -1; - - /* - * if vfio_group_fd == -ENOENT, that means the device - * isn't managed by VFIO - */ - if (vfio_group_fd == -ENOENT) { - EAL_LOG(NOTICE, - "%s not managed by VFIO driver, skipping", - dev_addr); - return 1; - } - - /* - * at this point, we know that this group is viable (meaning, all devices - * are either bound to VFIO or not bound to anything) - */ - - /* check if the group is viable */ - ret = ioctl(vfio_group_fd, VFIO_GROUP_GET_STATUS, &group_status); - if (ret) { - EAL_LOG(ERR, "%s cannot get VFIO group status, " - "error %i (%s)", dev_addr, errno, strerror(errno)); - close(vfio_group_fd); - vfio_clear_group(vfio_group_fd); - return -1; - } else if (!(group_status.flags & VFIO_GROUP_FLAGS_VIABLE)) { - EAL_LOG(ERR, "%s VFIO group is not viable! " - "Not all devices in IOMMU group bound to VFIO or unbound", - dev_addr); - close(vfio_group_fd); - vfio_clear_group(vfio_group_fd); - return -1; - } - - /* get the vfio_config it belongs to */ - vfio_cfg = get_vfio_cfg_by_group_num(iommu_group_num); - vfio_cfg = vfio_cfg ? vfio_cfg : default_vfio_cfg; - vfio_container_fd = vfio_cfg->vfio_container_fd; - user_mem_maps = &vfio_cfg->mem_maps; - - /* check if group does not have a container yet */ - if (!(group_status.flags & VFIO_GROUP_FLAGS_CONTAINER_SET)) { - - /* add group to a container */ - ret = ioctl(vfio_group_fd, VFIO_GROUP_SET_CONTAINER, - &vfio_container_fd); - if (ret) { - EAL_LOG(ERR, - "%s cannot add VFIO group to container, error " - "%i (%s)", dev_addr, errno, strerror(errno)); - close(vfio_group_fd); - vfio_clear_group(vfio_group_fd); - return -1; - } - - /* - * pick an IOMMU type and set up DMA mappings for container - * - * needs to be done only once, only when first group is - * assigned to a container and only in primary process. - * Note this can happen several times with the hotplug - * functionality. - */ - if (internal_conf->process_type == RTE_PROC_PRIMARY && - vfio_cfg->vfio_active_groups == 1 && - vfio_group_device_count(vfio_group_fd) == 0) { - const struct vfio_iommu_type *t; - - /* select an IOMMU type which we will be using */ - t = vfio_set_iommu_type(vfio_container_fd); - if (!t) { - EAL_LOG(ERR, - "%s failed to select IOMMU type", - dev_addr); - close(vfio_group_fd); - vfio_clear_group(vfio_group_fd); - return -1; - } - /* lock memory hotplug before mapping and release it - * after registering callback, to prevent races - */ - rte_mcfg_mem_read_lock(); - if (vfio_cfg == default_vfio_cfg) - ret = t->dma_map_func(vfio_container_fd); - else - ret = 0; - if (ret) { - EAL_LOG(ERR, - "%s DMA remapping failed, error " - "%i (%s)", - dev_addr, errno, strerror(errno)); - close(vfio_group_fd); - vfio_clear_group(vfio_group_fd); - rte_mcfg_mem_read_unlock(); - return -1; - } - - vfio_cfg->vfio_iommu_type = t; - - /* re-map all user-mapped segments */ - rte_spinlock_recursive_lock(&user_mem_maps->lock); - - /* this IOMMU type may not support DMA mapping, but - * if we have mappings in the list - that means we have - * previously mapped something successfully, so we can - * be sure that DMA mapping is supported. - */ - for (i = 0; i < user_mem_maps->n_maps; i++) { - struct user_mem_map *map; - map = &user_mem_maps->maps[i]; - - ret = t->dma_user_map_func( - vfio_container_fd, - map->addr, map->iova, map->len, - 1); - if (ret) { - EAL_LOG(ERR, "Couldn't map user memory for DMA: " - "va: 0x%" PRIx64 " " - "iova: 0x%" PRIx64 " " - "len: 0x%" PRIu64, - map->addr, map->iova, - map->len); - rte_spinlock_recursive_unlock( - &user_mem_maps->lock); - rte_mcfg_mem_read_unlock(); - return -1; - } - } - rte_spinlock_recursive_unlock(&user_mem_maps->lock); - - /* register callback for mem events */ - if (vfio_cfg == default_vfio_cfg) - ret = rte_mem_event_callback_register( - VFIO_MEM_EVENT_CLB_NAME, - vfio_mem_event_callback, NULL); - else - ret = 0; - /* unlock memory hotplug */ - rte_mcfg_mem_read_unlock(); - - if (ret && rte_errno != ENOTSUP) { - EAL_LOG(ERR, "Could not install memory event callback for VFIO"); - return -1; - } - if (ret) - EAL_LOG(DEBUG, "Memory event callbacks not supported"); - else - EAL_LOG(DEBUG, "Installed memory event callback for VFIO"); - } - } else if (rte_eal_process_type() != RTE_PROC_PRIMARY && - vfio_cfg == default_vfio_cfg && - vfio_cfg->vfio_iommu_type == NULL) { - /* if we're not a primary process, we do not set up the VFIO - * container because it's already been set up by the primary - * process. instead, we simply ask the primary about VFIO type - * we are using, and set the VFIO config up appropriately. - */ - ret = vfio_sync_default_container(); - if (ret < 0) { - EAL_LOG(ERR, "Could not sync default VFIO container"); - close(vfio_group_fd); - vfio_clear_group(vfio_group_fd); - return -1; - } - /* we have successfully initialized VFIO, notify user */ - const struct vfio_iommu_type *t = - default_vfio_cfg->vfio_iommu_type; - EAL_LOG(INFO, "Using IOMMU type %d (%s)", - t->type_id, t->name); - } - - rte_eal_vfio_get_vf_token(vf_token); - - /* get a file descriptor for the device with VF token firstly */ - if (!rte_uuid_is_null(vf_token)) { - char vf_token_str[RTE_UUID_STRLEN]; - char dev[PATH_MAX]; - - rte_uuid_unparse(vf_token, vf_token_str, sizeof(vf_token_str)); - snprintf(dev, sizeof(dev), - "%s vf_token=%s", dev_addr, vf_token_str); - - *vfio_dev_fd = ioctl(vfio_group_fd, VFIO_GROUP_GET_DEVICE_FD, - dev); - if (*vfio_dev_fd >= 0) - goto out; - } - - /* get a file descriptor for the device */ - *vfio_dev_fd = ioctl(vfio_group_fd, VFIO_GROUP_GET_DEVICE_FD, dev_addr); - if (*vfio_dev_fd < 0) { - /* if we cannot get a device fd, this implies a problem with - * the VFIO group or the container not having IOMMU configured. - */ - - EAL_LOG(WARNING, "Getting a vfio_dev_fd for %s failed", - dev_addr); - close(vfio_group_fd); - vfio_clear_group(vfio_group_fd); - return -1; - } - - /* device is now set up */ -out: - vfio_group_device_get(vfio_group_fd); - - return 0; -} - -RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_release_device) -int -rte_vfio_release_device(const char *sysfs_base, const char *dev_addr, - int vfio_dev_fd) -{ - struct vfio_config *vfio_cfg; - int vfio_group_fd; - int iommu_group_num; + struct container *cfg; + struct vfio_device *dev; + enum vfio_result res; int ret; - /* we don't want any DMA mapping messages to come while we're detaching - * VFIO device, because this might be the last device and we might need - * to unregister the callback. - */ + if (sysfs_base == NULL || dev_addr == NULL || vfio_dev_fd == NULL) { + rte_errno = EINVAL; + return -1; + } + + if (vfio_cfg.mode == RTE_VFIO_MODE_NONE) { + EAL_LOG(ERR, "VFIO support not initialized"); + rte_errno = ENXIO; + return -1; + } + rte_mcfg_mem_read_lock(); - /* get group number */ - ret = rte_vfio_get_group_num(sysfs_base, dev_addr, &iommu_group_num); - if (ret <= 0) { - EAL_LOG(WARNING, "%s not managed by VFIO driver", - dev_addr); - /* This is an error at this point. */ - ret = -1; - goto out; - } - - /* get the actual group fd */ - vfio_group_fd = vfio_get_group_fd_by_num(iommu_group_num); - if (vfio_group_fd < 0) { - EAL_LOG(INFO, "vfio_get_group_fd_by_num failed for %s", - dev_addr); - ret = vfio_group_fd; - goto out; + switch (vfio_cfg.mode) { + case RTE_VFIO_MODE_GROUP: + case RTE_VFIO_MODE_NOIOMMU: + { + int iommu_group_num; + + /* find group number */ + ret = vfio_group_get_num(sysfs_base, dev_addr, &iommu_group_num); + if (ret < 0) + goto assign_fail; + else if (ret == 0) + goto not_managed; + + /* find config by group */ + cfg = vfio_container_get_by_group_num(iommu_group_num); + if (cfg == NULL) + cfg = vfio_cfg.default_cfg; + + res = vfio_group_assign_device(cfg, sysfs_base, dev_addr, &dev); + break; } - - /* get the vfio_config it belongs to */ - vfio_cfg = get_vfio_cfg_by_group_num(iommu_group_num); - vfio_cfg = vfio_cfg ? vfio_cfg : default_vfio_cfg; - - /* At this point we got an active group. Closing it will make the - * container detachment. If this is the last active group, VFIO kernel - * code will unset the container and the IOMMU mappings. - */ - - /* Closing a device */ - if (close(vfio_dev_fd) < 0) { - EAL_LOG(INFO, "Error when closing vfio_dev_fd for %s", - dev_addr); + default: + EAL_LOG(ERR, "Unsupported VFIO mode"); + rte_errno = ENOTSUP; ret = -1; - goto out; + goto unlock; } - /* An VFIO group can have several devices attached. Just when there is - * no devices remaining should the group be closed. - */ - vfio_group_device_put(vfio_group_fd); - if (!vfio_group_device_count(vfio_group_fd)) { - - if (close(vfio_group_fd) < 0) { - EAL_LOG(INFO, "Error when closing vfio_group_fd for %s", - dev_addr); - ret = -1; - goto out; - } - - if (vfio_clear_group(vfio_group_fd) < 0) { - EAL_LOG(INFO, "Error when clearing group for %s", - dev_addr); - ret = -1; - goto out; - } + switch (res) { + case VFIO_NOT_MANAGED: +not_managed: + EAL_LOG(DEBUG, "Device %s not managed by VFIO", dev_addr); + rte_errno = ENODEV; + ret = -1; + goto unlock; + case VFIO_SUCCESS: + case VFIO_EXISTS: + break; + case VFIO_NO_SPACE: + EAL_LOG(ERR, "No space in VFIO container to assign device %s", dev_addr); + rte_errno = ENOSPC; + ret = -1; + goto unlock; + default: +assign_fail: + EAL_LOG(ERR, "Error assigning device %s to container", dev_addr); + rte_errno = EIO; + ret = -1; + goto unlock; } - - /* if there are no active device groups, unregister the callback to - * avoid spurious attempts to map/unmap memory from VFIO. - */ - if (vfio_cfg == default_vfio_cfg && vfio_cfg->vfio_active_groups == 0 && - rte_eal_process_type() != RTE_PROC_SECONDARY) - rte_mem_event_callback_unregister(VFIO_MEM_EVENT_CLB_NAME, - NULL); + *vfio_dev_fd = dev->fd; /* success */ ret = 0; -out: +unlock: rte_mcfg_mem_read_unlock(); + return ret; } +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_release_device) +int +rte_vfio_release_device(const char *sysfs_base __rte_unused, + const char *dev_addr, int vfio_dev_fd) +{ + struct container *cfg = NULL, *icfg; + struct vfio_device *dev = NULL, *idev; + int ret; + + if (sysfs_base == NULL || dev_addr == NULL) { + rte_errno = EINVAL; + return -1; + } + + if (vfio_cfg.mode == RTE_VFIO_MODE_NONE) { + EAL_LOG(ERR, "VFIO support not initialized"); + rte_errno = ENXIO; + return -1; + } + + rte_mcfg_mem_read_lock(); + + /* we need to find both config and device */ + CONTAINER_FOREACH_ACTIVE(icfg) { + DEVICE_FOREACH_ACTIVE(icfg, idev) { + if (idev->fd != vfio_dev_fd) + continue; + cfg = icfg; + dev = idev; + goto found; + } + } +found: + if (dev == NULL) { + EAL_LOG(ERR, "Device %s not managed by any container", dev_addr); + rte_errno = ENOENT; + ret = -1; + goto unlock; + } + + switch (vfio_cfg.mode) { + case RTE_VFIO_MODE_GROUP: + case RTE_VFIO_MODE_NOIOMMU: + { + int iommu_group_num = dev->group; + struct vfio_group_config *group_cfg = &cfg->group_cfg; + struct vfio_group *grp; + + bool need_clb = vfio_container_is_default(cfg) && + rte_eal_process_type() == RTE_PROC_PRIMARY; + + /* find the group */ + grp = vfio_group_get_by_num(cfg, iommu_group_num); + if (grp == NULL) { + /* shouldn't happen because we already know the device is valid */ + EAL_LOG(ERR, "IOMMU group %d not found in container", + iommu_group_num); + rte_errno = EIO; + ret = -1; + goto unlock; + } + + /* close device handle */ + vfio_device_erase(cfg, dev); + + /* remove device from group */ + grp->n_devices--; + + /* was this the last device? */ + if (grp->n_devices == 0) + vfio_group_erase(cfg, grp); + + /* if no more groups left, remove callback */ + if (need_clb && group_cfg->n_groups == 0 && group_cfg->mem_event_clb_set) { + rte_mem_event_callback_unregister(VFIO_MEM_EVENT_CLB_NAME, NULL); + group_cfg->mem_event_clb_set = false; + } + break; + } + default: + EAL_LOG(ERR, "Unsupported VFIO mode"); + rte_errno = ENOTSUP; + ret = -1; + goto unlock; + } + ret = 0; +unlock: + rte_mcfg_mem_read_unlock(); + + return ret; +} + +static int +vfio_sync_mode(struct container *cfg, enum rte_vfio_mode *mode) +{ + struct vfio_mp_param *p; + struct rte_mp_msg mp_req = {0}; + struct rte_mp_reply mp_reply = {0}; + struct timespec ts = {5, 0}; + + /* request iommufd from primary via mp_sync */ + rte_strscpy(mp_req.name, EAL_VFIO_MP, sizeof(mp_req.name)); + mp_req.len_param = sizeof(*p); + mp_req.num_fds = 0; + p = (struct vfio_mp_param *)mp_req.param; + p->req = SOCKET_REQ_CONTAINER; + + if (rte_mp_request_sync(&mp_req, &mp_reply, &ts) == 0 && + mp_reply.nb_received == 1) { + struct rte_mp_msg *mp_rep; + mp_rep = &mp_reply.msgs[0]; + p = (struct vfio_mp_param *)mp_rep->param; + if (p->result == SOCKET_OK && mp_rep->num_fds == 1) { + cfg->container_fd = mp_rep->fds[0]; + *mode = p->mode; + free(mp_reply.msgs); + return 0; + } + } + + free(mp_reply.msgs); + EAL_LOG(ERR, "Cannot request container_fd"); + return -1; +} + +static enum rte_vfio_mode +vfio_select_mode(void) +{ + struct container *cfg; + enum rte_vfio_mode mode = RTE_VFIO_MODE_NONE; + + cfg = vfio_container_create(); + /* cannot happen */ + if (cfg == NULL || cfg != vfio_cfg.default_cfg) { + EAL_LOG(ERR, "Unexpected VFIO config structure"); + return RTE_VFIO_MODE_NONE; + } + + /* for secondary, just ask the primary for the container and mode */ + if (rte_eal_process_type() != RTE_PROC_PRIMARY) { + struct vfio_group_config *group_cfg = &cfg->group_cfg; + + if (vfio_sync_mode(cfg, &mode) < 0) + goto err; + + /* primary handles DMA setup for default containers */ + group_cfg->dma_setup_done = true; + return mode; + } + /* if we failed mp sync setup, we cannot initialize VFIO */ + if (vfio_mp_sync_setup() < 0) + return RTE_VFIO_MODE_NONE; + + /* try group mode first */ + if (vfio_group_enable(cfg) == 0) { + /* check for noiommu */ + int ret = vfio_group_noiommu_is_enabled(); + if (ret < 0) + goto err_mpsync; + else if (ret == 1) + return RTE_VFIO_MODE_NOIOMMU; + return RTE_VFIO_MODE_GROUP; + } +err_mpsync: + vfio_mp_sync_cleanup(); +err: + vfio_container_erase(cfg); + + return RTE_VFIO_MODE_NONE; +} + +static const char * +vfio_mode_to_str(enum rte_vfio_mode mode) +{ + switch (mode) { + case RTE_VFIO_MODE_GROUP: return "group"; + case RTE_VFIO_MODE_NOIOMMU: return "noiommu"; + default: return "not initialized"; + } +} + RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_enable) int rte_vfio_enable(const char *modname) { - /* initialize group list */ - unsigned int i, j; int vfio_available; - DIR *dir; - const struct internal_config *internal_conf = - eal_get_internal_configuration(); + enum rte_vfio_mode mode = RTE_VFIO_MODE_NONE; - rte_spinlock_recursive_t lock = RTE_SPINLOCK_RECURSIVE_INITIALIZER; - - for (i = 0; i < RTE_DIM(vfio_cfgs); i++) { - vfio_cfgs[i].vfio_container_fd = -1; - vfio_cfgs[i].vfio_active_groups = 0; - vfio_cfgs[i].vfio_iommu_type = NULL; - vfio_cfgs[i].mem_maps.lock = lock; - - for (j = 0; j < RTE_DIM(vfio_cfgs[i].vfio_groups); j++) { - vfio_cfgs[i].vfio_groups[j].fd = -1; - vfio_cfgs[i].vfio_groups[j].group_num = -1; - vfio_cfgs[i].vfio_groups[j].devices = 0; - } + if (modname == NULL) { + rte_errno = EINVAL; + return -1; } EAL_LOG(DEBUG, "Probing VFIO support..."); @@ -1132,36 +1037,16 @@ rte_vfio_enable(const char *modname) "VFIO modules not loaded, skipping VFIO support..."); return 0; } + EAL_LOG(DEBUG, "VFIO module '%s' loaded, attempting to initialize VFIO...", modname); + mode = vfio_select_mode(); - /* VFIO directory might not exist (e.g., unprivileged containers) */ - dir = opendir(RTE_VFIO_DIR); - if (dir == NULL) { - EAL_LOG(DEBUG, - "VFIO directory does not exist, skipping VFIO support..."); - return 0; - } - closedir(dir); - - if (internal_conf->process_type == RTE_PROC_PRIMARY) { - if (vfio_mp_sync_setup() == -1) { - default_vfio_cfg->vfio_container_fd = -1; - } else { - /* open a default container */ - default_vfio_cfg->vfio_container_fd = vfio_open_container_fd(false); - } - } else { - /* get the default container from the primary process */ - default_vfio_cfg->vfio_container_fd = - vfio_open_container_fd(true); - } - - /* check if we have VFIO driver enabled */ - if (default_vfio_cfg->vfio_container_fd != -1) { - EAL_LOG(INFO, "VFIO support initialized"); - default_vfio_cfg->vfio_enabled = 1; - } else { + /* have we initialized anything? */ + if (mode == RTE_VFIO_MODE_NONE) EAL_LOG(NOTICE, "VFIO support could not be initialized"); - } + else + EAL_LOG(NOTICE, "VFIO support initialized: %s mode", vfio_mode_to_str(mode)); + + vfio_cfg.mode = mode; return 0; } @@ -1170,40 +1055,17 @@ RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_is_enabled) int rte_vfio_is_enabled(const char *modname) { - const int mod_available = rte_eal_check_module(modname) > 0; - return default_vfio_cfg->vfio_enabled && mod_available; + const int mod_available = modname ? rte_eal_check_module(modname) > 0 : 0; + return vfio_cfg.default_cfg->active && mod_available; } int vfio_get_iommu_type(void) { - if (default_vfio_cfg->vfio_iommu_type == NULL) + if (vfio_cfg.ops == NULL) return -1; - return default_vfio_cfg->vfio_iommu_type->type_id; -} - -const struct vfio_iommu_type * -vfio_set_iommu_type(int vfio_container_fd) -{ - unsigned idx; - for (idx = 0; idx < RTE_DIM(iommu_types); idx++) { - const struct vfio_iommu_type *t = &iommu_types[idx]; - - int ret = ioctl(vfio_container_fd, VFIO_SET_IOMMU, - t->type_id); - if (!ret) { - EAL_LOG(INFO, "Using IOMMU type %d (%s)", - t->type_id, t->name); - return t; - } - /* not an error, there may be more supported IOMMU types */ - EAL_LOG(DEBUG, "Set IOMMU type %d (%s) failed, error " - "%i (%s)", t->type_id, t->name, errno, - strerror(errno)); - } - /* if we didn't find a suitable IOMMU type, fail */ - return NULL; + return vfio_cfg.ops->type_id; } RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_device_info) @@ -1212,126 +1074,27 @@ rte_vfio_get_device_info(int vfio_dev_fd, struct vfio_device_info *device_info) { int ret; - if (device_info == NULL || vfio_dev_fd < 0) + if (device_info == NULL) { + rte_errno = EINVAL; return -1; + } + + if (vfio_cfg.mode == RTE_VFIO_MODE_NONE) { + EAL_LOG(ERR, "VFIO support not initialized"); + rte_errno = ENXIO; + return -1; + } ret = ioctl(vfio_dev_fd, VFIO_DEVICE_GET_INFO, device_info); if (ret) { - EAL_LOG(ERR, "Cannot get device info, error %i (%s)", - errno, strerror(errno)); + EAL_LOG(ERR, "Cannot get device info, error %d (%s)", errno, strerror(errno)); + rte_errno = errno; return -1; } return 0; } -int -vfio_has_supported_extensions(int vfio_container_fd) -{ - int ret; - unsigned idx, n_extensions = 0; - for (idx = 0; idx < RTE_DIM(iommu_types); idx++) { - const struct vfio_iommu_type *t = &iommu_types[idx]; - - ret = ioctl(vfio_container_fd, VFIO_CHECK_EXTENSION, - t->type_id); - if (ret < 0) { - EAL_LOG(ERR, "Could not get IOMMU type, error " - "%i (%s)", errno, strerror(errno)); - close(vfio_container_fd); - return -1; - } else if (ret == 1) { - /* we found a supported extension */ - n_extensions++; - } - EAL_LOG(DEBUG, "IOMMU type %d (%s) is %s", - t->type_id, t->name, - ret ? "supported" : "not supported"); - } - - /* if we didn't find any supported IOMMU types, fail */ - if (!n_extensions) { - close(vfio_container_fd); - return -1; - } - - return 0; -} - -/* - * Open a new VFIO container fd. - * - * If mp_request is true, requests a new container fd from the primary process - * via mp channel (for secondary processes that need to open the default container). - * - * Otherwise, opens a new container fd locally by opening /dev/vfio/vfio. - */ -int -vfio_open_container_fd(bool mp_request) -{ - int ret, vfio_container_fd; - struct rte_mp_msg mp_req, *mp_rep; - struct rte_mp_reply mp_reply = {0}; - struct timespec ts = {.tv_sec = 5, .tv_nsec = 0}; - struct vfio_mp_param *p = (struct vfio_mp_param *)mp_req.param; - - /* if not requesting via mp, open a new container locally */ - if (!mp_request) { - vfio_container_fd = open(RTE_VFIO_CONTAINER_PATH, O_RDWR); - if (vfio_container_fd < 0) { - EAL_LOG(ERR, "Cannot open VFIO container %s, error %i (%s)", - RTE_VFIO_CONTAINER_PATH, errno, strerror(errno)); - return -1; - } - - /* check VFIO API version */ - ret = ioctl(vfio_container_fd, VFIO_GET_API_VERSION); - if (ret != VFIO_API_VERSION) { - if (ret < 0) - EAL_LOG(ERR, - "Could not get VFIO API version, error " - "%i (%s)", errno, strerror(errno)); - else - EAL_LOG(ERR, "Unsupported VFIO API version!"); - close(vfio_container_fd); - return -1; - } - - ret = vfio_has_supported_extensions(vfio_container_fd); - if (ret) { - EAL_LOG(ERR, - "No supported IOMMU extensions found!"); - return -1; - } - - return vfio_container_fd; - } - /* - * if we're in a secondary process, request container fd from the - * primary process via mp channel - */ - p->req = SOCKET_REQ_CONTAINER; - strcpy(mp_req.name, EAL_VFIO_MP); - mp_req.len_param = sizeof(*p); - mp_req.num_fds = 0; - - vfio_container_fd = -1; - if (rte_mp_request_sync(&mp_req, &mp_reply, &ts) == 0 && - mp_reply.nb_received == 1) { - mp_rep = &mp_reply.msgs[0]; - p = (struct vfio_mp_param *)mp_rep->param; - if (p->result == SOCKET_OK && mp_rep->num_fds == 1) { - vfio_container_fd = mp_rep->fds[0]; - free(mp_reply.msgs); - return vfio_container_fd; - } - } - - free(mp_reply.msgs); - EAL_LOG(ERR, "Cannot request VFIO container fd"); - return -1; -} - RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_container_fd) int rte_vfio_get_container_fd(void) @@ -1340,511 +1103,54 @@ rte_vfio_get_container_fd(void) * The default container is set up during rte_vfio_enable(). * This function does not create a new container. */ - if (!default_vfio_cfg->vfio_enabled) - return -1; + if (vfio_cfg.mode != RTE_VFIO_MODE_NONE) + return vfio_cfg.default_cfg->container_fd; - return default_vfio_cfg->vfio_container_fd; + EAL_LOG(ERR, "VFIO support not initialized"); + rte_errno = ENXIO; + return -1; } RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_group_num) int -rte_vfio_get_group_num(const char *sysfs_base, - const char *dev_addr, int *iommu_group_num) +rte_vfio_get_group_num(const char *sysfs_base, const char *dev_addr, int *iommu_group_num) { - char linkname[PATH_MAX]; - char filename[PATH_MAX]; - char *tok[16], *group_tok, *end; int ret; - memset(linkname, 0, sizeof(linkname)); - memset(filename, 0, sizeof(filename)); - - /* try to find out IOMMU group for this device */ - snprintf(linkname, sizeof(linkname), - "%s/%s/iommu_group", sysfs_base, dev_addr); - - ret = readlink(linkname, filename, sizeof(filename)); - - /* if the link doesn't exist, no VFIO for us */ - if (ret < 0) - return 0; - - ret = rte_strsplit(filename, sizeof(filename), - tok, RTE_DIM(tok), '/'); - - if (ret <= 0) { - EAL_LOG(ERR, "%s cannot get IOMMU group", dev_addr); - return -1; - } - - /* IOMMU group is always the last token */ - errno = 0; - group_tok = tok[ret - 1]; - end = group_tok; - *iommu_group_num = strtol(group_tok, &end, 10); - if ((end != group_tok && *end != '\0') || errno != 0) { - EAL_LOG(ERR, "%s error parsing IOMMU number!", dev_addr); - return -1; - } - - return 1; -} - -static int -type1_map(const struct rte_memseg_list *msl, const struct rte_memseg *ms, - void *arg) -{ - int *vfio_container_fd = arg; - - /* skip external memory that isn't a heap */ - if (msl->external && !msl->heap) - return 0; - - /* skip any segments with invalid IOVA addresses */ - if (ms->iova == RTE_BAD_IOVA) - return 0; - - return vfio_type1_dma_mem_map(*vfio_container_fd, ms->addr_64, ms->iova, - ms->len, 1); -} - -static int -vfio_type1_dma_mem_map(int vfio_container_fd, uint64_t vaddr, uint64_t iova, - uint64_t len, int do_map) -{ - struct vfio_iommu_type1_dma_map dma_map; - struct vfio_iommu_type1_dma_unmap dma_unmap; - int ret; - - if (do_map != 0) { - memset(&dma_map, 0, sizeof(dma_map)); - dma_map.argsz = sizeof(struct vfio_iommu_type1_dma_map); - dma_map.vaddr = vaddr; - dma_map.size = len; - dma_map.iova = iova; - dma_map.flags = VFIO_DMA_MAP_FLAG_READ | - VFIO_DMA_MAP_FLAG_WRITE; - - ret = ioctl(vfio_container_fd, VFIO_IOMMU_MAP_DMA, &dma_map); - if (ret) { - /** - * In case the mapping was already done EEXIST will be - * returned from kernel. - */ - if (errno == EEXIST) { - EAL_LOG(DEBUG, - "Memory segment is already mapped, skipping"); - } else { - EAL_LOG(ERR, - "Cannot set up DMA remapping, error " - "%i (%s)", errno, strerror(errno)); - return -1; - } - } - } else { - memset(&dma_unmap, 0, sizeof(dma_unmap)); - dma_unmap.argsz = sizeof(struct vfio_iommu_type1_dma_unmap); - dma_unmap.size = len; - dma_unmap.iova = iova; - - ret = ioctl(vfio_container_fd, VFIO_IOMMU_UNMAP_DMA, - &dma_unmap); - if (ret) { - EAL_LOG(ERR, "Cannot clear DMA remapping, error " - "%i (%s)", errno, strerror(errno)); - return -1; - } else if (dma_unmap.size != len) { - EAL_LOG(ERR, "Unexpected size %"PRIu64 - " of DMA remapping cleared instead of %"PRIu64, - (uint64_t)dma_unmap.size, len); - rte_errno = EIO; - return -1; - } - } - - return 0; -} - -static int -vfio_type1_dma_map(int vfio_container_fd) -{ - return rte_memseg_walk(type1_map, &vfio_container_fd); -} - -/* Track the size of the statically allocated DMA window for SPAPR */ -uint64_t spapr_dma_win_len; -uint64_t spapr_dma_win_page_sz; - -static int -vfio_spapr_dma_do_map(int vfio_container_fd, uint64_t vaddr, uint64_t iova, - uint64_t len, int do_map) -{ - struct vfio_iommu_spapr_register_memory reg = { - .argsz = sizeof(reg), - .vaddr = (uintptr_t) vaddr, - .size = len, - .flags = 0 - }; - int ret; - - if (do_map != 0) { - struct vfio_iommu_type1_dma_map dma_map; - - if (iova + len > spapr_dma_win_len) { - EAL_LOG(ERR, "DMA map attempt outside DMA window"); - return -1; - } - - ret = ioctl(vfio_container_fd, - VFIO_IOMMU_SPAPR_REGISTER_MEMORY, ®); - if (ret) { - EAL_LOG(ERR, - "Cannot register vaddr for IOMMU, error " - "%i (%s)", errno, strerror(errno)); - return -1; - } - - memset(&dma_map, 0, sizeof(dma_map)); - dma_map.argsz = sizeof(struct vfio_iommu_type1_dma_map); - dma_map.vaddr = vaddr; - dma_map.size = len; - dma_map.iova = iova; - dma_map.flags = VFIO_DMA_MAP_FLAG_READ | - VFIO_DMA_MAP_FLAG_WRITE; - - ret = ioctl(vfio_container_fd, VFIO_IOMMU_MAP_DMA, &dma_map); - if (ret) { - EAL_LOG(ERR, "Cannot map vaddr for IOMMU, error " - "%i (%s)", errno, strerror(errno)); - return -1; - } - - } else { - struct vfio_iommu_type1_dma_map dma_unmap; - - memset(&dma_unmap, 0, sizeof(dma_unmap)); - dma_unmap.argsz = sizeof(struct vfio_iommu_type1_dma_unmap); - dma_unmap.size = len; - dma_unmap.iova = iova; - - ret = ioctl(vfio_container_fd, VFIO_IOMMU_UNMAP_DMA, - &dma_unmap); - if (ret) { - EAL_LOG(ERR, "Cannot unmap vaddr for IOMMU, error " - "%i (%s)", errno, strerror(errno)); - return -1; - } - - ret = ioctl(vfio_container_fd, - VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY, ®); - if (ret) { - EAL_LOG(ERR, - "Cannot unregister vaddr for IOMMU, error " - "%i (%s)", errno, strerror(errno)); - return -1; - } - } - - return ret; -} - -static int -vfio_spapr_map_walk(const struct rte_memseg_list *msl, - const struct rte_memseg *ms, void *arg) -{ - int *vfio_container_fd = arg; - - /* skip external memory that isn't a heap */ - if (msl->external && !msl->heap) - return 0; - - /* skip any segments with invalid IOVA addresses */ - if (ms->iova == RTE_BAD_IOVA) - return 0; - - return vfio_spapr_dma_do_map(*vfio_container_fd, - ms->addr_64, ms->iova, ms->len, 1); -} - -struct spapr_size_walk_param { - uint64_t max_va; - uint64_t page_sz; - bool is_user_managed; -}; - -/* - * In order to set the DMA window size required for the SPAPR IOMMU - * we need to walk the existing virtual memory allocations as well as - * find the hugepage size used. - */ -static int -vfio_spapr_size_walk(const struct rte_memseg_list *msl, void *arg) -{ - struct spapr_size_walk_param *param = arg; - uint64_t max = (uint64_t) msl->base_va + (uint64_t) msl->len; - - if (msl->external && !msl->heap) { - /* ignore user managed external memory */ - param->is_user_managed = true; - return 0; - } - - if (max > param->max_va) { - param->page_sz = msl->page_sz; - param->max_va = max; - } - - return 0; -} - -/* - * Find the highest memory address used in physical or virtual address - * space and use that as the top of the DMA window. - */ -static int -find_highest_mem_addr(struct spapr_size_walk_param *param) -{ - /* find the maximum IOVA address for setting the DMA window size */ - if (rte_eal_iova_mode() == RTE_IOVA_PA) { - static const char proc_iomem[] = "/proc/iomem"; - static const char str_sysram[] = "System RAM"; - uint64_t start, end, max = 0; - char *line = NULL; - char *dash, *space; - size_t line_len; - - /* - * Example "System RAM" in /proc/iomem: - * 00000000-1fffffffff : System RAM - * 200000000000-201fffffffff : System RAM - */ - FILE *fd = fopen(proc_iomem, "r"); - if (fd == NULL) { - EAL_LOG(ERR, "Cannot open %s", proc_iomem); - return -1; - } - /* Scan /proc/iomem for the highest PA in the system */ - while (getline(&line, &line_len, fd) != -1) { - if (strstr(line, str_sysram) == NULL) - continue; - - space = strstr(line, " "); - dash = strstr(line, "-"); - - /* Validate the format of the memory string */ - if (space == NULL || dash == NULL || space < dash) { - EAL_LOG(ERR, "Can't parse line \"%s\" in file %s", - line, proc_iomem); - continue; - } - - start = strtoull(line, NULL, 16); - end = strtoull(dash + 1, NULL, 16); - EAL_LOG(DEBUG, "Found system RAM from 0x%" PRIx64 - " to 0x%" PRIx64, start, end); - if (end > max) - max = end; - } - free(line); - fclose(fd); - - if (max == 0) { - EAL_LOG(ERR, "Failed to find valid \"System RAM\" " - "entry in file %s", proc_iomem); - return -1; - } - - spapr_dma_win_len = rte_align64pow2(max + 1); - return 0; - } else if (rte_eal_iova_mode() == RTE_IOVA_VA) { - EAL_LOG(DEBUG, "Highest VA address in memseg list is 0x%" - PRIx64, param->max_va); - spapr_dma_win_len = rte_align64pow2(param->max_va); - return 0; - } - - spapr_dma_win_len = 0; - EAL_LOG(ERR, "Unsupported IOVA mode"); - return -1; -} - - -/* - * The SPAPRv2 IOMMU supports 2 DMA windows with starting - * address at 0 or 1<<59. By default, a DMA window is set - * at address 0, 2GB long, with a 4KB page. For DPDK we - * must remove the default window and setup a new DMA window - * based on the hugepage size and memory requirements of - * the application before we can map memory for DMA. - */ -static int -spapr_dma_win_size(void) -{ - struct spapr_size_walk_param param; - - /* only create DMA window once */ - if (spapr_dma_win_len > 0) - return 0; - - /* walk the memseg list to find the page size/max VA address */ - memset(¶m, 0, sizeof(param)); - if (rte_memseg_list_walk(vfio_spapr_size_walk, ¶m) < 0) { - EAL_LOG(ERR, "Failed to walk memseg list for DMA window size"); + if (sysfs_base == NULL || dev_addr == NULL || iommu_group_num == NULL) { + rte_errno = EINVAL; return -1; } - /* we can't be sure if DMA window covers external memory */ - if (param.is_user_managed) - EAL_LOG(WARNING, "Detected user managed external memory which may not be managed by the IOMMU"); - - /* check physical/virtual memory size */ - if (find_highest_mem_addr(¶m) < 0) + if (vfio_cfg.mode == RTE_VFIO_MODE_NONE) { + EAL_LOG(ERR, "VFIO support not initialized"); + rte_errno = ENXIO; return -1; - EAL_LOG(DEBUG, "Setting DMA window size to 0x%" PRIx64, - spapr_dma_win_len); - spapr_dma_win_page_sz = param.page_sz; - rte_mem_set_dma_mask(rte_ctz64(spapr_dma_win_len)); - return 0; -} - -static int -vfio_spapr_create_dma_window(int vfio_container_fd) -{ - struct vfio_iommu_spapr_tce_create create = { - .argsz = sizeof(create), }; - struct vfio_iommu_spapr_tce_remove remove = { - .argsz = sizeof(remove), }; - struct vfio_iommu_spapr_tce_info info = { - .argsz = sizeof(info), }; - int ret; - - ret = spapr_dma_win_size(); - if (ret < 0) - return ret; - - ret = ioctl(vfio_container_fd, VFIO_IOMMU_SPAPR_TCE_GET_INFO, &info); - if (ret) { - EAL_LOG(ERR, "Cannot get IOMMU info, error %i (%s)", - errno, strerror(errno)); - return -1; - } - - /* - * sPAPR v1/v2 IOMMU always has a default 1G DMA window set. The window - * can't be changed for v1 but it can be changed for v2. Since DPDK only - * supports v2, remove the default DMA window so it can be resized. - */ - remove.start_addr = info.dma32_window_start; - ret = ioctl(vfio_container_fd, VFIO_IOMMU_SPAPR_TCE_REMOVE, &remove); - if (ret) - return -1; - - /* create a new DMA window (start address is not selectable) */ - create.window_size = spapr_dma_win_len; - create.page_shift = rte_ctz64(spapr_dma_win_page_sz); - create.levels = 1; - ret = ioctl(vfio_container_fd, VFIO_IOMMU_SPAPR_TCE_CREATE, &create); -#ifdef VFIO_IOMMU_SPAPR_INFO_DDW - /* - * The vfio_iommu_spapr_tce_info structure was modified in - * Linux kernel 4.2.0 to add support for the - * vfio_iommu_spapr_tce_ddw_info structure needed to try - * multiple table levels. Skip the attempt if running with - * an older kernel. - */ - if (ret) { - /* if at first we don't succeed, try more levels */ - uint32_t levels; - - for (levels = create.levels + 1; - ret && levels <= info.ddw.levels; levels++) { - create.levels = levels; - ret = ioctl(vfio_container_fd, - VFIO_IOMMU_SPAPR_TCE_CREATE, &create); - } } -#endif /* VFIO_IOMMU_SPAPR_INFO_DDW */ - if (ret) { - EAL_LOG(ERR, "Cannot create new DMA window, error " - "%i (%s)", errno, strerror(errno)); - EAL_LOG(ERR, - "Consider using a larger hugepage size if supported by the system"); + if (vfio_cfg.mode != RTE_VFIO_MODE_GROUP && vfio_cfg.mode != RTE_VFIO_MODE_NOIOMMU) { + EAL_LOG(ERR, "VFIO not initialized in group mode"); + rte_errno = ENOTSUP; return -1; } - - /* verify the start address */ - if (create.start_addr != 0) { - EAL_LOG(ERR, "Received unsupported start address 0x%" - PRIx64, (uint64_t)create.start_addr); + ret = vfio_group_get_num(sysfs_base, dev_addr, iommu_group_num); + if (ret < 0) { + rte_errno = EINVAL; return -1; - } - return ret; -} - -static int -vfio_spapr_dma_mem_map(int vfio_container_fd, uint64_t vaddr, - uint64_t iova, uint64_t len, int do_map) -{ - int ret = 0; - - if (do_map) { - if (vfio_spapr_dma_do_map(vfio_container_fd, - vaddr, iova, len, 1)) { - EAL_LOG(ERR, "Failed to map DMA"); - ret = -1; - } - } else { - if (vfio_spapr_dma_do_map(vfio_container_fd, - vaddr, iova, len, 0)) { - EAL_LOG(ERR, "Failed to unmap DMA"); - ret = -1; - } - } - - return ret; -} - -static int -vfio_spapr_dma_map(int vfio_container_fd) -{ - if (vfio_spapr_create_dma_window(vfio_container_fd) < 0) { - EAL_LOG(ERR, "Could not create new DMA window!"); + } else if (ret == 0) { + rte_errno = ENODEV; return -1; } - - /* map all existing DPDK segments for DMA */ - if (rte_memseg_walk(vfio_spapr_map_walk, &vfio_container_fd) < 0) - return -1; - - return 0; -} - -static int -vfio_noiommu_dma_map(int __rte_unused vfio_container_fd) -{ - /* No-IOMMU mode does not need DMA mapping */ - return 0; -} - -static int -vfio_noiommu_dma_mem_map(int __rte_unused vfio_container_fd, - uint64_t __rte_unused vaddr, - uint64_t __rte_unused iova, uint64_t __rte_unused len, - int __rte_unused do_map) -{ - /* No-IOMMU mode does not need DMA mapping */ return 0; } static int -vfio_dma_mem_map(struct vfio_config *vfio_cfg, uint64_t vaddr, uint64_t iova, +vfio_dma_mem_map(struct container *cfg, uint64_t vaddr, uint64_t iova, uint64_t len, int do_map) { - const struct vfio_iommu_type *t = vfio_cfg->vfio_iommu_type; + const struct vfio_iommu_ops *t = vfio_cfg.ops; if (!t) { EAL_LOG(ERR, "VFIO support not initialized"); - rte_errno = ENODEV; return -1; } @@ -1852,16 +1158,14 @@ vfio_dma_mem_map(struct vfio_config *vfio_cfg, uint64_t vaddr, uint64_t iova, EAL_LOG(ERR, "VFIO custom DMA region mapping not supported by IOMMU %s", t->name); - rte_errno = ENOTSUP; return -1; } - return t->dma_user_map_func(vfio_cfg->vfio_container_fd, vaddr, iova, - len, do_map); + return t->dma_user_map_func(cfg, vaddr, iova, len, do_map); } static int -container_dma_map(struct vfio_config *vfio_cfg, uint64_t vaddr, uint64_t iova, +container_dma_map(struct container *cfg, uint64_t vaddr, uint64_t iova, uint64_t len) { struct user_mem_map *new_map; @@ -1869,16 +1173,15 @@ container_dma_map(struct vfio_config *vfio_cfg, uint64_t vaddr, uint64_t iova, bool has_partial_unmap; int ret = 0; - user_mem_maps = &vfio_cfg->mem_maps; + user_mem_maps = &cfg->mem_maps; rte_spinlock_recursive_lock(&user_mem_maps->lock); if (user_mem_maps->n_maps == RTE_DIM(user_mem_maps->maps)) { EAL_LOG(ERR, "No more space for user mem maps"); - rte_errno = ENOMEM; ret = -1; goto out; } /* map the entry */ - if (vfio_dma_mem_map(vfio_cfg, vaddr, iova, len, 1)) { + if (vfio_dma_mem_map(cfg, vaddr, iova, len, 1)) { /* technically, this will fail if there are currently no devices * plugged in, even if a device were added later, this mapping * might have succeeded. however, since we cannot verify if this @@ -1891,7 +1194,7 @@ container_dma_map(struct vfio_config *vfio_cfg, uint64_t vaddr, uint64_t iova, goto out; } /* do we have partial unmap support? */ - has_partial_unmap = vfio_cfg->vfio_iommu_type->partial_unmap; + has_partial_unmap = vfio_cfg.ops->partial_unmap; /* create new user mem map entry */ new_map = &user_mem_maps->maps[user_mem_maps->n_maps++]; @@ -1908,17 +1211,17 @@ container_dma_map(struct vfio_config *vfio_cfg, uint64_t vaddr, uint64_t iova, } static int -container_dma_unmap(struct vfio_config *vfio_cfg, uint64_t vaddr, uint64_t iova, +container_dma_unmap(struct container *cfg, uint64_t vaddr, uint64_t iova, uint64_t len) { - struct user_mem_map orig_maps[RTE_DIM(vfio_cfg->mem_maps.maps)]; + struct user_mem_map orig_maps[RTE_DIM(cfg->mem_maps.maps)]; struct user_mem_map new_maps[2]; /* can be at most 2 */ struct user_mem_maps *user_mem_maps; int n_orig, n_new, ret = 0; bool has_partial_unmap; unsigned int newlen; - user_mem_maps = &vfio_cfg->mem_maps; + user_mem_maps = &cfg->mem_maps; rte_spinlock_recursive_lock(&user_mem_maps->lock); /* @@ -1944,13 +1247,12 @@ container_dma_unmap(struct vfio_config *vfio_cfg, uint64_t vaddr, uint64_t iova, /* did we find anything? */ if (n_orig < 0) { EAL_LOG(ERR, "Couldn't find previously mapped region"); - rte_errno = EINVAL; ret = -1; goto out; } /* do we have partial unmap capability? */ - has_partial_unmap = vfio_cfg->vfio_iommu_type->partial_unmap; + has_partial_unmap = vfio_cfg.ops->partial_unmap; /* * if we don't support partial unmap, we must check if start and end of @@ -1966,7 +1268,6 @@ container_dma_unmap(struct vfio_config *vfio_cfg, uint64_t vaddr, uint64_t iova, if (!start_aligned || !end_aligned) { EAL_LOG(DEBUG, "DMA partial unmap unsupported"); - rte_errno = ENOTSUP; ret = -1; goto out; } @@ -1984,28 +1285,20 @@ container_dma_unmap(struct vfio_config *vfio_cfg, uint64_t vaddr, uint64_t iova, newlen = (user_mem_maps->n_maps - n_orig) + n_new; if (newlen >= RTE_DIM(user_mem_maps->maps)) { EAL_LOG(ERR, "Not enough space to store partial mapping"); - rte_errno = ENOMEM; ret = -1; goto out; } /* unmap the entry */ - if (vfio_dma_mem_map(vfio_cfg, vaddr, iova, len, 0)) { + if (vfio_dma_mem_map(cfg, vaddr, iova, len, 0)) { /* there may not be any devices plugged in, so unmapping will - * fail with ENODEV/ENOTSUP rte_errno values, but that doesn't - * stop us from removing the mapping, as the assumption is we - * won't be needing this memory any more and thus will want to - * prevent it from being remapped again on hotplug. so, only - * fail if we indeed failed to unmap (e.g. if the mapping was - * within our mapped range but had invalid alignment). + * fail, but that doesn't stop us from removing the mapping, + * as the assumption is we won't be needing this memory any + * more and thus will want to prevent it from being remapped + * again on hotplug. Ignore the error and proceed with + * removing the mapping from our records. */ - if (rte_errno != ENODEV && rte_errno != ENOTSUP) { - EAL_LOG(ERR, "Couldn't unmap region for DMA"); - ret = -1; - goto out; - } else { - EAL_LOG(DEBUG, "DMA unmapping failed, but removing mappings anyway"); - } + EAL_LOG(DEBUG, "DMA unmapping failed, but removing mappings anyway"); } /* we have unmapped the region, so now update the maps */ @@ -2021,212 +1314,178 @@ RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_noiommu_is_enabled) int rte_vfio_noiommu_is_enabled(void) { - int fd; - ssize_t cnt; - char c; - - fd = open(RTE_VFIO_NOIOMMU_MODE, O_RDONLY); - if (fd < 0) { - if (errno != ENOENT) { - EAL_LOG(ERR, "Cannot open VFIO noiommu file " - "%i (%s)", errno, strerror(errno)); - return -1; - } - /* - * else the file does not exists - * i.e. noiommu is not enabled - */ - return 0; - } - - cnt = read(fd, &c, 1); - close(fd); - if (cnt != 1) { - EAL_LOG(ERR, "Unable to read from VFIO noiommu file " - "%i (%s)", errno, strerror(errno)); - return -1; - } - - return c == 'Y'; + return vfio_cfg.mode == RTE_VFIO_MODE_NOIOMMU; } RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_create) int rte_vfio_container_create(void) { - unsigned int i; + struct container *cfg; + int container_fd; - /* Find an empty slot to store new vfio config */ - for (i = 1; i < RTE_DIM(vfio_cfgs); i++) { - if (vfio_cfgs[i].vfio_container_fd == -1) - break; - } - - if (i == RTE_DIM(vfio_cfgs)) { - EAL_LOG(ERR, "Exceed max VFIO container limit"); + if (vfio_cfg.mode == RTE_VFIO_MODE_NONE) { + EAL_LOG(ERR, "VFIO not initialized"); + rte_errno = ENXIO; return -1; } - - /* Create a new container fd */ - vfio_cfgs[i].vfio_container_fd = vfio_open_container_fd(false); - if (vfio_cfgs[i].vfio_container_fd < 0) { - EAL_LOG(NOTICE, "Fail to create a new VFIO container"); + cfg = vfio_container_create(); + if (cfg == NULL) { + EAL_LOG(ERR, "Reached VFIO container limit"); + rte_errno = ENOSPC; return -1; } - return vfio_cfgs[i].vfio_container_fd; + switch (vfio_cfg.mode) { + case RTE_VFIO_MODE_GROUP: + case RTE_VFIO_MODE_NOIOMMU: + { + container_fd = vfio_group_open_container_fd(); + if (container_fd < 0) { + EAL_LOG(ERR, "Fail to create a new VFIO container"); + rte_errno = EIO; + goto err; + } + cfg->container_fd = container_fd; + break; + } + default: + EAL_LOG(NOTICE, "Unsupported VFIO mode"); + rte_errno = ENOTSUP; + goto err; + } + return container_fd; +err: + vfio_container_erase(cfg); + return -1; } RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_destroy) int rte_vfio_container_destroy(int container_fd) { - struct vfio_config *vfio_cfg; - unsigned int i; + struct container *cfg; + struct vfio_device *dev; - vfio_cfg = get_vfio_cfg_by_container_fd(container_fd); - if (vfio_cfg == NULL) { - EAL_LOG(ERR, "Invalid VFIO container fd"); + if (vfio_cfg.mode == RTE_VFIO_MODE_NONE) { + EAL_LOG(ERR, "VFIO not initialized"); + rte_errno = ENXIO; return -1; } - for (i = 0; i < RTE_DIM(vfio_cfg->vfio_groups); i++) - if (vfio_cfg->vfio_groups[i].group_num != -1) - vfio_container_group_unbind(container_fd, - vfio_cfg->vfio_groups[i].group_num); - - close(container_fd); - vfio_cfg->vfio_container_fd = -1; - vfio_cfg->vfio_active_groups = 0; - vfio_cfg->vfio_iommu_type = NULL; - - return 0; -} - -RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_assign_device) -int -rte_vfio_container_assign_device(int vfio_container_fd, const char *sysfs_base, - const char *dev_addr) -{ - int iommu_group_num; - int ret; - - ret = rte_vfio_get_group_num(sysfs_base, dev_addr, &iommu_group_num); - if (ret < 0) { - EAL_LOG(ERR, "Cannot get IOMMU group number for device %s", - dev_addr); - return -1; - } else if (ret == 0) { - EAL_LOG(ERR, - "Device %s is not assigned to any IOMMU group", - dev_addr); - return -1; - } - - ret = vfio_container_group_bind(vfio_container_fd, - iommu_group_num); - if (ret < 0) { - EAL_LOG(ERR, - "Cannot bind IOMMU group %d for device %s", - iommu_group_num, dev_addr); - return -1; - } - - return 0; -} - -static int -vfio_container_group_bind(int container_fd, int iommu_group_num) -{ - struct vfio_config *vfio_cfg; - - vfio_cfg = get_vfio_cfg_by_container_fd(container_fd); - if (vfio_cfg == NULL) { - EAL_LOG(ERR, "Invalid VFIO container fd"); + cfg = vfio_container_get_by_fd(container_fd); + if (cfg == NULL) { + EAL_LOG(ERR, "VFIO container fd not managed by VFIO"); + rte_errno = ENODEV; return -1; } - - return vfio_get_group_fd(vfio_cfg, iommu_group_num); -} - -static int -vfio_container_group_unbind(int container_fd, int iommu_group_num) -{ - struct vfio_group *cur_grp = NULL; - struct vfio_config *vfio_cfg; - unsigned int i; - - vfio_cfg = get_vfio_cfg_by_container_fd(container_fd); - if (vfio_cfg == NULL) { - EAL_LOG(ERR, "Invalid VFIO container fd"); + /* forbid destroying default container */ + if (vfio_container_is_default(cfg)) { + EAL_LOG(ERR, "Cannot destroy default VFIO container"); + rte_errno = EINVAL; return -1; } - for (i = 0; i < RTE_DIM(vfio_cfg->vfio_groups); i++) { - if (vfio_cfg->vfio_groups[i].group_num == iommu_group_num) { - cur_grp = &vfio_cfg->vfio_groups[i]; - break; + switch (vfio_cfg.mode) { + case RTE_VFIO_MODE_GROUP: + case RTE_VFIO_MODE_NOIOMMU: + /* erase all devices */ + DEVICE_FOREACH_ACTIVE(cfg, dev) { + EAL_LOG(DEBUG, "Device in IOMMU group %d still open, closing", dev->group); + /* + * technically we could've done back-reference lookup and closed our groups + * following a device close, but since we're closing and erasing all groups + * anyway, we can afford to not bother. + */ + vfio_device_erase(cfg, dev); } - } - /* This should not happen */ - if (cur_grp == NULL) { - EAL_LOG(ERR, "Specified VFIO group number not found"); + /* erase all groups */ + struct vfio_group *grp; + GROUP_FOREACH_ACTIVE(cfg, grp) { + EAL_LOG(DEBUG, "IOMMU group %d still open, closing", grp->group_num); + vfio_group_erase(cfg, grp); + } + break; + default: + EAL_LOG(ERR, "Unsupported VFIO mode"); + rte_errno = ENOTSUP; return -1; } - if (cur_grp->fd >= 0 && close(cur_grp->fd) < 0) { - EAL_LOG(ERR, - "Error when closing vfio_group_fd for iommu_group_num " - "%d", iommu_group_num); - return -1; - } - cur_grp->group_num = -1; - cur_grp->fd = -1; - cur_grp->devices = 0; - vfio_cfg->vfio_active_groups--; + /* erase entire config */ + vfio_container_erase(cfg); return 0; } RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_dma_map) int -rte_vfio_container_dma_map(int container_fd, uint64_t vaddr, uint64_t iova, - uint64_t len) +rte_vfio_container_dma_map(int container_fd, uint64_t vaddr, uint64_t iova, uint64_t len) { - struct vfio_config *vfio_cfg; + struct container *cfg; if (len == 0) { rte_errno = EINVAL; return -1; } - vfio_cfg = get_vfio_cfg_by_container_fd(container_fd); - if (vfio_cfg == NULL) { + if (vfio_cfg.mode == RTE_VFIO_MODE_NONE) { + EAL_LOG(ERR, "VFIO support not initialized"); + rte_errno = ENXIO; + return -1; + } + + cfg = vfio_container_get_by_fd(container_fd); + if (cfg == NULL) { EAL_LOG(ERR, "Invalid VFIO container fd"); + rte_errno = EINVAL; return -1; } - return container_dma_map(vfio_cfg, vaddr, iova, len); + if (container_dma_map(cfg, vaddr, iova, len) < 0) { + rte_errno = EIO; + return -1; + } + + return 0; } RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_dma_unmap) int -rte_vfio_container_dma_unmap(int container_fd, uint64_t vaddr, uint64_t iova, - uint64_t len) +rte_vfio_container_dma_unmap(int container_fd, uint64_t vaddr, uint64_t iova, uint64_t len) { - struct vfio_config *vfio_cfg; + struct container *cfg; if (len == 0) { rte_errno = EINVAL; return -1; } - vfio_cfg = get_vfio_cfg_by_container_fd(container_fd); - if (vfio_cfg == NULL) { + if (vfio_cfg.mode == RTE_VFIO_MODE_NONE) { + EAL_LOG(ERR, "VFIO support not initialized"); + rte_errno = ENXIO; + return -1; + } + + cfg = vfio_container_get_by_fd(container_fd); + if (cfg == NULL) { EAL_LOG(ERR, "Invalid VFIO container fd"); + rte_errno = EINVAL; return -1; } - return container_dma_unmap(vfio_cfg, vaddr, iova, len); + if (container_dma_unmap(cfg, vaddr, iova, len) < 0) { + rte_errno = EIO; + return -1; + } + + return 0; +} + +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_mode) +enum rte_vfio_mode +rte_vfio_get_mode(void) +{ + return vfio_cfg.mode; } diff --git a/lib/eal/linux/eal_vfio.h b/lib/eal/linux/eal_vfio.h index 30389fb274..68d3a3ec6e 100644 --- a/lib/eal/linux/eal_vfio.h +++ b/lib/eal/linux/eal_vfio.h @@ -6,60 +6,161 @@ #define EAL_VFIO_H_ #include <rte_common.h> +#include <rte_spinlock.h> #include <stdint.h> +#include <rte_vfio.h> + +/* hot plug/unplug of VFIO groups may cause all DMA maps to be dropped. we can + * recreate the mappings for DPDK segments, but we cannot do so for memory that + * was registered by the user themselves, so we need to store the user mappings + * somewhere, to recreate them later. + */ +#define EAL_VFIO_MAX_USER_MEM_MAPS 256 + +/* user memory map entry */ +struct user_mem_map { + uint64_t addr; /**< start VA */ + uint64_t iova; /**< start IOVA */ + uint64_t len; /**< total length of the mapping */ + uint64_t chunk; /**< this mapping can be split in chunks of this size */ +}; + +/* user memory maps container (common for all API modes) */ +struct user_mem_maps { + rte_spinlock_recursive_t lock; + int n_maps; + struct user_mem_map maps[EAL_VFIO_MAX_USER_MEM_MAPS]; +}; + /* * we don't need to store device fd's anywhere since they can be obtained from * the group fd via an ioctl() call. */ struct vfio_group { + bool active; int group_num; int fd; - int devices; + int n_devices; +}; + +/* device tracking (common for group and cdev modes) */ +struct vfio_device { + bool active; + int group; /**< back-reference to group list (group mode) */ + int fd; +}; + +/* group mode specific configuration */ +struct vfio_group_config { + bool dma_setup_done; + bool iommu_type_set; + bool mem_event_clb_set; + size_t n_groups; + struct vfio_group groups[RTE_MAX_VFIO_GROUPS]; +}; + +/* per-container configuration */ +struct container { + bool active; + int container_fd; + struct user_mem_maps mem_maps; + struct vfio_group_config group_cfg; + int n_devices; + struct vfio_device devices[RTE_MAX_VFIO_DEVICES]; }; /* DMA mapping function prototype. - * Takes VFIO container fd as a parameter. + * Takes VFIO container config as a parameter. * Returns 0 on success, -1 on error. */ -typedef int (*vfio_dma_func_t)(int); +typedef int (*dma_func_t)(struct container *cfg); /* Custom memory region DMA mapping function prototype. - * Takes VFIO container fd, virtual address, physical address, length and + * Takes VFIO container config, virtual address, physical address, length and * operation type (0 to unmap 1 for map) as a parameters. * Returns 0 on success, -1 on error. */ -typedef int (*vfio_dma_user_func_t)(int fd, uint64_t vaddr, uint64_t iova, - uint64_t len, int do_map); +typedef int (*dma_user_func_t)(struct container *cfg, uint64_t vaddr, + uint64_t iova, uint64_t len, int do_map); -struct vfio_iommu_type { +/* mode-independent ops */ +struct vfio_iommu_ops { int type_id; const char *name; bool partial_unmap; - vfio_dma_user_func_t dma_user_map_func; - vfio_dma_func_t dma_map_func; + dma_user_func_t dma_user_map_func; + dma_func_t dma_map_func; }; -/* get the vfio container that devices are bound to by default */ -int vfio_open_container_fd(bool mp_request); +/* global configuration */ +struct vfio_config { + struct container *default_cfg; + enum rte_vfio_mode mode; + const struct vfio_iommu_ops *ops; +}; + +/* per-process, per-container data */ +extern struct container containers[RTE_MAX_VFIO_CONTAINERS]; + +/* current configuration */ +extern struct vfio_config vfio_cfg; + +#define CONTAINER_FOREACH(cfg) \ + for ((cfg) = &containers[0]; \ + (cfg) < &containers[RTE_DIM(containers)]; \ + (cfg)++) + +#define CONTAINER_FOREACH_ACTIVE(cfg) \ + CONTAINER_FOREACH((cfg)) \ + if (((cfg)->active)) + +#define GROUP_FOREACH(cfg, grp) \ + for ((grp) = &((cfg)->group_cfg.groups[0]); \ + (grp) < &((cfg)->group_cfg.groups[RTE_DIM((cfg)->group_cfg.groups)]); \ + (grp)++) -/* pick IOMMU type. returns a pointer to vfio_iommu_type or NULL for error */ -const struct vfio_iommu_type * -vfio_set_iommu_type(int vfio_container_fd); +#define GROUP_FOREACH_ACTIVE(cfg, grp) \ + GROUP_FOREACH((cfg), (grp)) \ + if ((grp)->active) -int -vfio_get_iommu_type(void); +#define DEVICE_FOREACH(cfg, dev) \ + for ((dev) = &((cfg)->devices[0]); \ + (dev) < &((cfg)->devices[RTE_DIM((cfg)->devices)]); \ + (dev)++) -int vfio_get_group_fd_by_num(int iommu_group_num); +#define DEVICE_FOREACH_ACTIVE(cfg, dev) \ + DEVICE_FOREACH((cfg), (dev)) \ + if ((dev)->active) -/* check if we have any supported extensions */ -int -vfio_has_supported_extensions(int vfio_container_fd); +/* for containers, we only need to initialize the lock in mem maps */ +#define CONTAINER_INITIALIZER \ + ((struct container){ \ + .mem_maps = {.lock = RTE_SPINLOCK_RECURSIVE_INITIALIZER,}, \ + }) +int vfio_get_iommu_type(void); int vfio_mp_sync_setup(void); void vfio_mp_sync_cleanup(void); +bool vfio_container_is_default(struct container *cfg); +/* group mode functions */ +int vfio_group_enable(struct container *cfg); +int vfio_group_open_container_fd(void); +int vfio_group_noiommu_is_enabled(void); +int vfio_group_get_num(const char *sysfs_base, const char *dev_addr, + int *iommu_group_num); +struct vfio_group *vfio_group_get_by_num(struct container *cfg, int iommu_group); +struct vfio_group *vfio_group_create(struct container *cfg, int iommu_group); +void vfio_group_erase(struct container *cfg, struct vfio_group *grp); +int vfio_group_open_fd(struct container *cfg, struct vfio_group *grp); +int vfio_group_prepare(struct container *cfg, struct vfio_group *grp); +int vfio_group_setup_iommu(struct container *cfg); +int vfio_group_setup_device_fd(const char *dev_addr, + struct vfio_group *grp, struct vfio_device *dev); + +#define VFIO_MEM_EVENT_CLB_NAME "vfio_mem_event_clb" #define EAL_VFIO_MP "eal_vfio_mp_sync" #define SOCKET_REQ_CONTAINER 0x100 @@ -75,6 +176,7 @@ struct vfio_mp_param { union { int group_num; int iommu_type_id; + enum rte_vfio_mode mode; }; }; diff --git a/lib/eal/linux/eal_vfio_group.c b/lib/eal/linux/eal_vfio_group.c new file mode 100644 index 0000000000..6123a83aec --- /dev/null +++ b/lib/eal/linux/eal_vfio_group.c @@ -0,0 +1,983 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright(c) 2010-2025 Intel Corporation + */ + +#include <dirent.h> +#include <fcntl.h> +#include <inttypes.h> +#include <stdbool.h> +#include <stdint.h> +#include <sys/ioctl.h> +#include <unistd.h> + +#include <uapi/linux/vfio.h> + +#include <rte_log.h> +#include <rte_errno.h> +#include <rte_eal_memconfig.h> +#include <rte_memory.h> +#include <rte_string_fns.h> +#include <rte_vfio.h> + +#include "eal_vfio.h" +#include "eal_private.h" +#include "eal_internal_cfg.h" + +static int vfio_type1_dma_map(struct container *); +static int vfio_type1_dma_mem_map(struct container *, uint64_t, uint64_t, uint64_t, int); +static int vfio_spapr_dma_map(struct container *); +static int vfio_spapr_dma_mem_map(struct container *, uint64_t, uint64_t, uint64_t, int); +static int vfio_noiommu_dma_map(struct container *); +static int vfio_noiommu_dma_mem_map(struct container *, uint64_t, uint64_t, uint64_t, int); + +/* IOMMU types we support */ +static const struct vfio_iommu_ops iommu_types[] = { + /* x86 IOMMU, otherwise known as type 1 */ + { + .type_id = VFIO_TYPE1_IOMMU, + .name = "Type 1", + .partial_unmap = false, + .dma_map_func = &vfio_type1_dma_map, + .dma_user_map_func = &vfio_type1_dma_mem_map + }, + /* ppc64 IOMMU, otherwise known as spapr */ + { + .type_id = VFIO_SPAPR_TCE_v2_IOMMU, + .name = "sPAPR", + .partial_unmap = true, + .dma_map_func = &vfio_spapr_dma_map, + .dma_user_map_func = &vfio_spapr_dma_mem_map + }, + /* IOMMU-less mode */ + { + .type_id = VFIO_NOIOMMU_IOMMU, + .name = "No-IOMMU", + .partial_unmap = true, + .dma_map_func = &vfio_noiommu_dma_map, + .dma_user_map_func = &vfio_noiommu_dma_mem_map + }, +}; + +static const struct vfio_iommu_ops * +vfio_group_set_iommu_type(int vfio_container_fd) +{ + unsigned int idx; + for (idx = 0; idx < RTE_DIM(iommu_types); idx++) { + const struct vfio_iommu_ops *t = &iommu_types[idx]; + + int ret = ioctl(vfio_container_fd, VFIO_SET_IOMMU, t->type_id); + if (ret == 0) + return t; + /* not an error, there may be more supported IOMMU types */ + EAL_LOG(DEBUG, "Set IOMMU type %d (%s) failed, error " + "%i (%s)", t->type_id, t->name, errno, + strerror(errno)); + } + /* if we didn't find a suitable IOMMU type, fail */ + return NULL; +} + +static int +type1_map(const struct rte_memseg_list *msl, const struct rte_memseg *ms, + void *arg) +{ + struct container *cfg = arg; + + /* skip external memory that isn't a heap */ + if (msl->external && !msl->heap) + return 0; + + /* skip any segments with invalid IOVA addresses */ + if (ms->iova == RTE_BAD_IOVA) + return 0; + + return vfio_type1_dma_mem_map(cfg, ms->addr_64, ms->iova, ms->len, 1); +} + +static int +vfio_type1_dma_mem_map(struct container *cfg, uint64_t vaddr, uint64_t iova, + uint64_t len, int do_map) +{ + struct vfio_iommu_type1_dma_map dma_map; + struct vfio_iommu_type1_dma_unmap dma_unmap; + int ret; + + if (do_map != 0) { + memset(&dma_map, 0, sizeof(dma_map)); + dma_map.argsz = sizeof(struct vfio_iommu_type1_dma_map); + dma_map.vaddr = vaddr; + dma_map.size = len; + dma_map.iova = iova; + dma_map.flags = VFIO_DMA_MAP_FLAG_READ | + VFIO_DMA_MAP_FLAG_WRITE; + + ret = ioctl(cfg->container_fd, VFIO_IOMMU_MAP_DMA, &dma_map); + if (ret) { + /** + * In case the mapping was already done EEXIST will be + * returned from kernel. + */ + if (errno == EEXIST) { + EAL_LOG(DEBUG, + "Memory segment is already mapped, skipping"); + } else { + EAL_LOG(ERR, + "Cannot set up DMA remapping, error " + "%i (%s)", errno, strerror(errno)); + return -1; + } + } + } else { + memset(&dma_unmap, 0, sizeof(dma_unmap)); + dma_unmap.argsz = sizeof(struct vfio_iommu_type1_dma_unmap); + dma_unmap.size = len; + dma_unmap.iova = iova; + + ret = ioctl(cfg->container_fd, VFIO_IOMMU_UNMAP_DMA, + &dma_unmap); + if (ret) { + EAL_LOG(ERR, "Cannot clear DMA remapping, error " + "%i (%s)", errno, strerror(errno)); + return -1; + } else if (dma_unmap.size != len) { + EAL_LOG(ERR, "Unexpected size %"PRIu64 + " of DMA remapping cleared instead of %"PRIu64, + (uint64_t)dma_unmap.size, len); + return -1; + } + } + + return 0; +} + +static int +vfio_type1_dma_map(struct container *cfg) +{ + return rte_memseg_walk(type1_map, cfg); +} + +/* Track the size of the statically allocated DMA window for SPAPR */ +uint64_t spapr_dma_win_len; +uint64_t spapr_dma_win_page_sz; + +static int +vfio_spapr_dma_do_map(struct container *cfg, uint64_t vaddr, uint64_t iova, + uint64_t len, int do_map) +{ + struct vfio_iommu_spapr_register_memory reg = { + .argsz = sizeof(reg), + .vaddr = (uintptr_t) vaddr, + .size = len, + .flags = 0 + }; + int ret; + + if (do_map != 0) { + struct vfio_iommu_type1_dma_map dma_map; + + if (iova + len > spapr_dma_win_len) { + EAL_LOG(ERR, "DMA map attempt outside DMA window"); + return -1; + } + + ret = ioctl(cfg->container_fd, + VFIO_IOMMU_SPAPR_REGISTER_MEMORY, ®); + if (ret) { + EAL_LOG(ERR, + "Cannot register vaddr for IOMMU, error " + "%i (%s)", errno, strerror(errno)); + return -1; + } + + memset(&dma_map, 0, sizeof(dma_map)); + dma_map.argsz = sizeof(struct vfio_iommu_type1_dma_map); + dma_map.vaddr = vaddr; + dma_map.size = len; + dma_map.iova = iova; + dma_map.flags = VFIO_DMA_MAP_FLAG_READ | VFIO_DMA_MAP_FLAG_WRITE; + + ret = ioctl(cfg->container_fd, VFIO_IOMMU_MAP_DMA, &dma_map); + if (ret) { + EAL_LOG(ERR, "Cannot map vaddr for IOMMU, error " + "%i (%s)", errno, strerror(errno)); + return -1; + } + + } else { + struct vfio_iommu_type1_dma_map dma_unmap; + + memset(&dma_unmap, 0, sizeof(dma_unmap)); + dma_unmap.argsz = sizeof(struct vfio_iommu_type1_dma_unmap); + dma_unmap.size = len; + dma_unmap.iova = iova; + + ret = ioctl(cfg->container_fd, VFIO_IOMMU_UNMAP_DMA, + &dma_unmap); + if (ret) { + EAL_LOG(ERR, "Cannot unmap vaddr for IOMMU, error " + "%i (%s)", errno, strerror(errno)); + return -1; + } + + ret = ioctl(cfg->container_fd, + VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY, ®); + if (ret) { + EAL_LOG(ERR, + "Cannot unregister vaddr for IOMMU, error " + "%i (%s)", errno, strerror(errno)); + return -1; + } + } + + return ret; +} + +static int +vfio_spapr_map_walk(const struct rte_memseg_list *msl, + const struct rte_memseg *ms, void *arg) +{ + struct container *cfg = arg; + + /* skip external memory that isn't a heap */ + if (msl->external && !msl->heap) + return 0; + + /* skip any segments with invalid IOVA addresses */ + if (ms->iova == RTE_BAD_IOVA) + return 0; + + return vfio_spapr_dma_do_map(cfg, ms->addr_64, ms->iova, ms->len, 1); +} + +struct spapr_size_walk_param { + uint64_t max_va; + uint64_t page_sz; + bool is_user_managed; +}; + +/* + * In order to set the DMA window size required for the SPAPR IOMMU + * we need to walk the existing virtual memory allocations as well as + * find the hugepage size used. + */ +static int +vfio_spapr_size_walk(const struct rte_memseg_list *msl, void *arg) +{ + struct spapr_size_walk_param *param = arg; + uint64_t max = (uint64_t) msl->base_va + (uint64_t) msl->len; + + if (msl->external && !msl->heap) { + /* ignore user managed external memory */ + param->is_user_managed = true; + return 0; + } + + if (max > param->max_va) { + param->page_sz = msl->page_sz; + param->max_va = max; + } + + return 0; +} + +/* + * Find the highest memory address used in physical or virtual address + * space and use that as the top of the DMA window. + */ +static int +find_highest_mem_addr(struct spapr_size_walk_param *param) +{ + /* find the maximum IOVA address for setting the DMA window size */ + if (rte_eal_iova_mode() == RTE_IOVA_PA) { + static const char proc_iomem[] = "/proc/iomem"; + static const char str_sysram[] = "System RAM"; + uint64_t start, end, max = 0; + char *line = NULL; + char *dash, *space; + size_t line_len; + + /* + * Example "System RAM" in /proc/iomem: + * 00000000-1fffffffff : System RAM + * 200000000000-201fffffffff : System RAM + */ + FILE *fd = fopen(proc_iomem, "r"); + if (fd == NULL) { + EAL_LOG(ERR, "Cannot open %s", proc_iomem); + return -1; + } + /* Scan /proc/iomem for the highest PA in the system */ + while (getline(&line, &line_len, fd) != -1) { + if (strstr(line, str_sysram) == NULL) + continue; + + space = strstr(line, " "); + dash = strstr(line, "-"); + + /* Validate the format of the memory string */ + if (space == NULL || dash == NULL || space < dash) { + EAL_LOG(ERR, "Can't parse line \"%s\" in file %s", + line, proc_iomem); + continue; + } + + start = strtoull(line, NULL, 16); + end = strtoull(dash + 1, NULL, 16); + EAL_LOG(DEBUG, "Found system RAM from 0x%" PRIx64 + " to 0x%" PRIx64, start, end); + if (end > max) + max = end; + } + free(line); + fclose(fd); + + if (max == 0) { + EAL_LOG(ERR, "Failed to find valid \"System RAM\" " + "entry in file %s", proc_iomem); + return -1; + } + + spapr_dma_win_len = rte_align64pow2(max + 1); + return 0; + } else if (rte_eal_iova_mode() == RTE_IOVA_VA) { + EAL_LOG(DEBUG, "Highest VA address in memseg list is 0x%" + PRIx64, param->max_va); + spapr_dma_win_len = rte_align64pow2(param->max_va); + return 0; + } + + spapr_dma_win_len = 0; + EAL_LOG(ERR, "Unsupported IOVA mode"); + return -1; +} + + +/* + * The SPAPRv2 IOMMU supports 2 DMA windows with starting + * address at 0 or 1<<59. By default, a DMA window is set + * at address 0, 2GB long, with a 4KB page. For DPDK we + * must remove the default window and setup a new DMA window + * based on the hugepage size and memory requirements of + * the application before we can map memory for DMA. + */ +static int +spapr_dma_win_size(void) +{ + struct spapr_size_walk_param param; + + /* only create DMA window once */ + if (spapr_dma_win_len > 0) + return 0; + + /* walk the memseg list to find the page size/max VA address */ + memset(¶m, 0, sizeof(param)); + if (rte_memseg_list_walk(vfio_spapr_size_walk, ¶m) < 0) { + EAL_LOG(ERR, "Failed to walk memseg list for DMA window size"); + return -1; + } + + /* we can't be sure if DMA window covers external memory */ + if (param.is_user_managed) + EAL_LOG(WARNING, "Detected user managed external memory which may not be managed by the IOMMU"); + + /* check physical/virtual memory size */ + if (find_highest_mem_addr(¶m) < 0) + return -1; + EAL_LOG(DEBUG, "Setting DMA window size to 0x%" PRIx64, + spapr_dma_win_len); + spapr_dma_win_page_sz = param.page_sz; + rte_mem_set_dma_mask(rte_ctz64(spapr_dma_win_len)); + return 0; +} + +static int +vfio_spapr_create_dma_window(struct container *cfg) +{ + struct vfio_iommu_spapr_tce_create create = { + .argsz = sizeof(create), }; + struct vfio_iommu_spapr_tce_remove remove = { + .argsz = sizeof(remove), }; + struct vfio_iommu_spapr_tce_info info = { + .argsz = sizeof(info), }; + int ret; + + ret = spapr_dma_win_size(); + if (ret < 0) + return ret; + + ret = ioctl(cfg->container_fd, VFIO_IOMMU_SPAPR_TCE_GET_INFO, &info); + if (ret) { + EAL_LOG(ERR, "Cannot get IOMMU info, error %i (%s)", + errno, strerror(errno)); + return -1; + } + + /* + * sPAPR v1/v2 IOMMU always has a default 1G DMA window set. The window + * can't be changed for v1 but it can be changed for v2. Since DPDK only + * supports v2, remove the default DMA window so it can be resized. + */ + remove.start_addr = info.dma32_window_start; + ret = ioctl(cfg->container_fd, VFIO_IOMMU_SPAPR_TCE_REMOVE, &remove); + if (ret) + return -1; + + /* create a new DMA window (start address is not selectable) */ + create.window_size = spapr_dma_win_len; + create.page_shift = rte_ctz64(spapr_dma_win_page_sz); + create.levels = 1; + ret = ioctl(cfg->container_fd, VFIO_IOMMU_SPAPR_TCE_CREATE, &create); + /* + * The vfio_iommu_spapr_tce_info structure was modified in + * Linux kernel 4.2.0 to add support for the + * vfio_iommu_spapr_tce_ddw_info structure needed to try + * multiple table levels. Skip the attempt if running with + * an older kernel. + */ + if (ret) { + /* if at first we don't succeed, try more levels */ + uint32_t levels; + + for (levels = create.levels + 1; + ret && levels <= info.ddw.levels; levels++) { + create.levels = levels; + ret = ioctl(cfg->container_fd, + VFIO_IOMMU_SPAPR_TCE_CREATE, &create); + } + } + if (ret) { + EAL_LOG(ERR, "Cannot create new DMA window, error " + "%i (%s)", errno, strerror(errno)); + EAL_LOG(ERR, + "Consider using a larger hugepage size if supported by the system"); + return -1; + } + + /* verify the start address */ + if (create.start_addr != 0) { + EAL_LOG(ERR, "Received unsupported start address 0x%" + PRIx64, (uint64_t)create.start_addr); + return -1; + } + return ret; +} + +static int +vfio_spapr_dma_mem_map(struct container *cfg, uint64_t vaddr, + uint64_t iova, uint64_t len, int do_map) +{ + int ret = 0; + + if (do_map) { + if (vfio_spapr_dma_do_map(cfg, vaddr, iova, len, 1)) { + EAL_LOG(ERR, "Failed to map DMA"); + ret = -1; + } + } else { + if (vfio_spapr_dma_do_map(cfg, vaddr, iova, len, 0)) { + EAL_LOG(ERR, "Failed to unmap DMA"); + ret = -1; + } + } + + return ret; +} + +static int +vfio_spapr_dma_map(struct container *cfg) +{ + if (vfio_spapr_create_dma_window(cfg) < 0) { + EAL_LOG(ERR, "Could not create new DMA window!"); + return -1; + } + + /* map all existing DPDK segments for DMA */ + if (rte_memseg_walk(vfio_spapr_map_walk, cfg) < 0) + return -1; + + return 0; +} + +static int +vfio_noiommu_dma_map(struct container *cfg __rte_unused) +{ + /* No-IOMMU mode does not need DMA mapping */ + return 0; +} + +static int +vfio_noiommu_dma_mem_map(struct container *cfg __rte_unused, + uint64_t vaddr __rte_unused, + uint64_t iova __rte_unused, uint64_t len __rte_unused, + int do_map __rte_unused) +{ + /* No-IOMMU mode does not need DMA mapping */ + return 0; +} + +struct vfio_group * +vfio_group_create(struct container *cfg, int iommu_group) +{ + struct vfio_group *grp; + + if (cfg->group_cfg.n_groups >= RTE_DIM(cfg->group_cfg.groups)) { + EAL_LOG(ERR, "Cannot add more VFIO groups to container"); + return NULL; + } + GROUP_FOREACH(cfg, grp) { + if (grp->active) + continue; + cfg->group_cfg.n_groups++; + grp->active = true; + grp->group_num = iommu_group; + grp->fd = -1; + return grp; + } + /* should not happen */ + return NULL; +} + +void +vfio_group_erase(struct container *cfg, struct vfio_group *grp) +{ + struct vfio_group_config *group_cfg = &cfg->group_cfg; + + if (grp->fd >= 0 && close(grp->fd) < 0) + EAL_LOG(ERR, "Error when closing group fd %d", grp->fd); + + *grp = (struct vfio_group){0}; + group_cfg->n_groups--; + + /* if this was the last group in config, erase IOMMU setup and unregister callback */ + if (group_cfg->n_groups == 0) { + group_cfg->dma_setup_done = false; + group_cfg->iommu_type_set = false; + } +} + +struct vfio_group * +vfio_group_get_by_num(struct container *cfg, int iommu_group) +{ + struct vfio_group *grp; + + GROUP_FOREACH_ACTIVE(cfg, grp) { + if (grp->group_num == iommu_group) + return grp; + } + return NULL; +} + +static int +vfio_open_group_sysfs(int iommu_group_num) +{ + char filename[PATH_MAX]; + int fd; + + if (vfio_cfg.mode == RTE_VFIO_MODE_GROUP) + snprintf(filename, sizeof(filename), RTE_VFIO_GROUP_FMT, iommu_group_num); + else if (vfio_cfg.mode == RTE_VFIO_MODE_NOIOMMU) + snprintf(filename, sizeof(filename), RTE_VFIO_NOIOMMU_GROUP_FMT, iommu_group_num); + + /* reset errno before open to differentiate errors */ + errno = 0; + fd = open(filename, O_RDWR); + + /* we have to differentiate between failed open and non-existence */ + if (errno == ENOENT) + return -ENOENT; + return fd; +} + +static int +vfio_group_request_fd(int iommu_group_num) +{ + struct rte_mp_msg mp_req, *mp_rep; + struct rte_mp_reply mp_reply = {0}; + struct timespec ts = {.tv_sec = 5, .tv_nsec = 0}; + struct vfio_mp_param *p = (struct vfio_mp_param *)mp_req.param; + int vfio_group_fd = -1; + + p->req = SOCKET_REQ_GROUP; + p->group_num = iommu_group_num; + rte_strscpy(mp_req.name, EAL_VFIO_MP, sizeof(mp_req.name)); + mp_req.len_param = sizeof(*p); + mp_req.num_fds = 0; + + if (rte_mp_request_sync(&mp_req, &mp_reply, &ts) == 0 && mp_reply.nb_received == 1) { + mp_rep = &mp_reply.msgs[0]; + p = (struct vfio_mp_param *)mp_rep->param; + if (p->result == SOCKET_OK && mp_rep->num_fds == 1) { + vfio_group_fd = mp_rep->fds[0]; + } else if (p->result == SOCKET_NO_FD) { + EAL_LOG(ERR, "Bad VFIO group fd"); + vfio_group_fd = -ENOENT; + } + } + + free(mp_reply.msgs); + return vfio_group_fd; +} + +int +vfio_group_open_fd(struct container *cfg, struct vfio_group *grp) +{ + int vfio_group_fd; + + /* we make multiprocess request only in secondary processes for default config */ + if ((rte_eal_process_type() != RTE_PROC_PRIMARY) && (vfio_container_is_default(cfg))) + vfio_group_fd = vfio_group_request_fd(grp->group_num); + else + vfio_group_fd = vfio_open_group_sysfs(grp->group_num); + + /* pass the non-existence up the chain */ + if (vfio_group_fd == -ENOENT) + return vfio_group_fd; + else if (vfio_group_fd < 0) { + EAL_LOG(ERR, "Failed to open VFIO group %d", grp->group_num); + return vfio_group_fd; + } + grp->fd = vfio_group_fd; + return 0; +} + +static const struct vfio_iommu_ops * +vfio_group_sync_iommu_ops(void) +{ + struct rte_mp_msg mp_req, *mp_rep; + struct rte_mp_reply mp_reply = {0}; + struct timespec ts = {.tv_sec = 5, .tv_nsec = 0}; + struct vfio_mp_param *p = (struct vfio_mp_param *)mp_req.param; + int iommu_type_id; + unsigned int i; + + /* find default container's IOMMU type */ + p->req = SOCKET_REQ_IOMMU_TYPE; + rte_strscpy(mp_req.name, EAL_VFIO_MP, sizeof(mp_req.name)); + mp_req.len_param = sizeof(*p); + mp_req.num_fds = 0; + + iommu_type_id = -1; + if (rte_mp_request_sync(&mp_req, &mp_reply, &ts) == 0 && + mp_reply.nb_received == 1) { + mp_rep = &mp_reply.msgs[0]; + p = (struct vfio_mp_param *)mp_rep->param; + if (p->result == SOCKET_OK) + iommu_type_id = p->iommu_type_id; + } + free(mp_reply.msgs); + if (iommu_type_id < 0) { + EAL_LOG(ERR, "Could not get IOMMU type from primary process"); + return NULL; + } + + /* we now have an fd for default container, as well as its IOMMU type. + * now, set up default VFIO container config to match. + */ + for (i = 0; i < RTE_DIM(iommu_types); i++) { + const struct vfio_iommu_ops *t = &iommu_types[i]; + if (t->type_id != iommu_type_id) + continue; + + return t; + } + EAL_LOG(ERR, "Could not find IOMMU type id (%i)", iommu_type_id); + return NULL; +} + +int +vfio_group_noiommu_is_enabled(void) +{ + int fd; + ssize_t cnt; + char c; + + fd = open(RTE_VFIO_NOIOMMU_MODE, O_RDONLY); + if (fd < 0) { + if (errno != ENOENT) { + EAL_LOG(ERR, "Cannot open VFIO noiommu file " + "%i (%s)", errno, strerror(errno)); + return -1; + } + /* + * else the file does not exists + * i.e. noiommu is not enabled + */ + return 0; + } + + cnt = read(fd, &c, 1); + close(fd); + if (cnt != 1) { + EAL_LOG(ERR, "Unable to read from VFIO noiommu file " + "%i (%s)", errno, strerror(errno)); + return -1; + } + + return c == 'Y'; +} + +static int +vfio_has_supported_extensions(int vfio_container_fd) +{ + int ret; + unsigned int idx, n_extensions = 0; + for (idx = 0; idx < RTE_DIM(iommu_types); idx++) { + const struct vfio_iommu_ops *t = &iommu_types[idx]; + + ret = ioctl(vfio_container_fd, VFIO_CHECK_EXTENSION, + t->type_id); + if (ret < 0) { + EAL_LOG(ERR, "Could not get IOMMU type, error " + "%i (%s)", errno, strerror(errno)); + close(vfio_container_fd); + return -1; + } else if (ret == 1) { + /* we found a supported extension */ + n_extensions++; + } + EAL_LOG(DEBUG, "IOMMU type %d (%s) is %s", + t->type_id, t->name, + ret ? "supported" : "not supported"); + } + + /* if we didn't find any supported IOMMU types, fail */ + if (!n_extensions) + return -1; + + return 0; +} + +int +vfio_group_open_container_fd(void) +{ + int ret, vfio_container_fd; + + vfio_container_fd = open(RTE_VFIO_CONTAINER_PATH, O_RDWR); + if (vfio_container_fd < 0) { + EAL_LOG(DEBUG, "Cannot open VFIO container %s, error %i (%s)", + RTE_VFIO_CONTAINER_PATH, errno, strerror(errno)); + return -1; + } + + /* check VFIO API version */ + ret = ioctl(vfio_container_fd, VFIO_GET_API_VERSION); + if (ret != VFIO_API_VERSION) { + if (ret < 0) + EAL_LOG(DEBUG, + "Could not get VFIO API version, error " + "%i (%s)", errno, strerror(errno)); + else + EAL_LOG(DEBUG, "Unsupported VFIO API version!"); + close(vfio_container_fd); + return -1; + } + + ret = vfio_has_supported_extensions(vfio_container_fd); + if (ret) { + EAL_LOG(DEBUG, + "No supported IOMMU extensions found!"); + close(vfio_container_fd); + return -1; + } + + return vfio_container_fd; +} + +int +vfio_group_enable(struct container *cfg) +{ + int container_fd; + DIR *dir; + + /* VFIO directory might not exist (e.g., unprivileged containers) */ + dir = opendir(RTE_VFIO_DIR); + if (dir == NULL) { + EAL_LOG(DEBUG, + "VFIO directory does not exist, skipping VFIO group support..."); + return 1; + } + closedir(dir); + + /* open a default container */ + container_fd = vfio_group_open_container_fd(); + if (container_fd < 0) + return -1; + + cfg->container_fd = container_fd; + return 0; +} + +int +vfio_group_prepare(struct container *cfg, struct vfio_group *grp) +{ + struct vfio_group_status group_status = { + .argsz = sizeof(group_status)}; + int ret; + + /* + * We need to assign group to a container and check if it is viable, but there are cases + * where we don't need to do that. + * + * For default container, we need to set up the group only in primary process, as secondary + * process would have requested group fd over IPC, which implies it would have already been + * set up by the primary. + * + * For custom containers, every process sets up its own groups. + */ + if (vfio_container_is_default(cfg) && rte_eal_process_type() != RTE_PROC_PRIMARY) { + EAL_LOG(DEBUG, "Skipping setup for VFIO group %d", grp->group_num); + return 0; + } + + /* check if the group is viable */ + ret = ioctl(grp->fd, VFIO_GROUP_GET_STATUS, &group_status); + if (ret) { + EAL_LOG(ERR, "Cannot get VFIO group status for group %d, error %i (%s)", + grp->group_num, errno, strerror(errno)); + return -1; + } + + if ((group_status.flags & VFIO_GROUP_FLAGS_VIABLE) == 0) { + EAL_LOG(ERR, "VFIO group %d is not viable! " + "Not all devices in IOMMU group bound to VFIO or unbound", + grp->group_num); + return -1; + } + + /* set container for group if necessary */ + if ((group_status.flags & VFIO_GROUP_FLAGS_CONTAINER_SET) == 0) { + /* add group to a container */ + ret = ioctl(grp->fd, VFIO_GROUP_SET_CONTAINER, &cfg->container_fd); + if (ret) { + EAL_LOG(ERR, "Cannot add VFIO group %d to container, error %i (%s)", + grp->group_num, errno, strerror(errno)); + return -1; + } + } else { + /* group is already added to a container - this should not happen */ + EAL_LOG(ERR, "VFIO group %d is already assigned to a container", grp->group_num); + return -1; + } + return 0; +} + +int +vfio_group_setup_iommu(struct container *cfg) +{ + const struct vfio_iommu_ops *ops; + + /* + * Setting IOMMU type is a per-container operation (via ioctl on container fd), but the ops + * structure is global and shared across all containers. + * + * For secondary processes with default container, we sync ops from primary. For all other + * cases (primary, or secondary with custom containers), we set IOMMU type on the container + * which also discovers the ops. + */ + if (vfio_container_is_default(cfg) && rte_eal_process_type() != RTE_PROC_PRIMARY) { + /* Secondary process: sync ops from primary for default container */ + ops = vfio_group_sync_iommu_ops(); + if (ops == NULL) + return -1; + } else { + /* Primary process OR custom container: set IOMMU type on container */ + ops = vfio_group_set_iommu_type(cfg->container_fd); + if (ops == NULL) + return -1; + } + + /* Set or verify global ops */ + if (vfio_cfg.ops == NULL) { + vfio_cfg.ops = ops; + EAL_LOG(INFO, "IOMMU type set to %d (%s)", ops->type_id, ops->name); + } else if (vfio_cfg.ops != ops) { + /* This shouldn't happen on the same machine, but log it */ + EAL_LOG(WARNING, + "Container has different IOMMU type (%d - %s) than previously set (%d - %s)", + ops->type_id, ops->name, vfio_cfg.ops->type_id, vfio_cfg.ops->name); + } + + return 0; +} + +int +vfio_group_setup_device_fd(const char *dev_addr, struct vfio_group *grp, struct vfio_device *dev) +{ + rte_uuid_t vf_token; + int fd; + + rte_eal_vfio_get_vf_token(vf_token); + + if (!rte_uuid_is_null(vf_token)) { + char vf_token_str[RTE_UUID_STRLEN]; + char dev[PATH_MAX]; + + rte_uuid_unparse(vf_token, vf_token_str, sizeof(vf_token_str)); + snprintf(dev, sizeof(dev), + "%s vf_token=%s", dev_addr, vf_token_str); + + fd = ioctl(grp->fd, VFIO_GROUP_GET_DEVICE_FD, dev); + if (fd >= 0) + goto out; + } + /* get a file descriptor for the device */ + fd = ioctl(grp->fd, VFIO_GROUP_GET_DEVICE_FD, dev_addr); + if (fd < 0) { + /* + * if we cannot get a device fd, this implies a problem with the VFIO group or the + * container not having IOMMU configured. + */ + EAL_LOG(WARNING, "Getting a vfio_dev_fd for %s failed", dev_addr); + return -1; + } +out: + dev->fd = fd; + /* store backreference to group */ + dev->group = grp->group_num; + /* increment number of devices in group */ + grp->n_devices++; + return 0; +} + +int +vfio_group_get_num(const char *sysfs_base, const char *dev_addr, int *iommu_group_num) +{ + char linkname[PATH_MAX]; + char filename[PATH_MAX]; + char *tok[16], *group_tok, *end; + int ret, group_num; + + memset(linkname, 0, sizeof(linkname)); + memset(filename, 0, sizeof(filename)); + + /* try to find out IOMMU group for this device */ + snprintf(linkname, sizeof(linkname), + "%s/%s/iommu_group", sysfs_base, dev_addr); + + ret = readlink(linkname, filename, sizeof(filename)); + + /* if the link doesn't exist, no VFIO for us */ + if (ret < 0) + return 0; + + ret = rte_strsplit(filename, sizeof(filename), + tok, RTE_DIM(tok), '/'); + + if (ret <= 0) { + EAL_LOG(ERR, "%s cannot get IOMMU group", dev_addr); + return -1; + } + + /* IOMMU group is always the last token */ + errno = 0; + group_tok = tok[ret - 1]; + end = group_tok; + group_num = strtol(group_tok, &end, 10); + if (end == group_tok || *end != '\0' || errno != 0) { + EAL_LOG(ERR, "%s error parsing IOMMU number!", dev_addr); + return -1; + } + *iommu_group_num = group_num; + + return 1; +} diff --git a/lib/eal/linux/eal_vfio_mp_sync.c b/lib/eal/linux/eal_vfio_mp_sync.c index 3eaeef2fc8..9a07d35023 100644 --- a/lib/eal/linux/eal_vfio_mp_sync.c +++ b/lib/eal/linux/eal_vfio_mp_sync.c @@ -32,21 +32,32 @@ vfio_mp_primary(const struct rte_mp_msg *msg, const void *peer) switch (m->req) { case SOCKET_REQ_GROUP: + { + struct container *cfg = vfio_cfg.default_cfg; + struct vfio_group *grp; + + if (vfio_cfg.mode != RTE_VFIO_MODE_GROUP && + vfio_cfg.mode != RTE_VFIO_MODE_NOIOMMU) { + EAL_LOG(ERR, "VFIO not initialized in group mode"); + r->result = SOCKET_ERR; + break; + } + r->req = SOCKET_REQ_GROUP; r->group_num = m->group_num; - fd = vfio_get_group_fd_by_num(m->group_num); - if (fd < 0 && fd != -ENOENT) - r->result = SOCKET_ERR; - else if (fd == -ENOENT) - /* if VFIO group exists but isn't bound to VFIO driver */ + grp = vfio_group_get_by_num(cfg, m->group_num); + if (grp == NULL) { + /* group doesn't exist in primary */ r->result = SOCKET_NO_FD; - else { - /* if group exists and is bound to VFIO driver */ + } else { + /* group exists and is bound to VFIO driver */ + fd = grp->fd; r->result = SOCKET_OK; reply.num_fds = 1; reply.fds[0] = fd; } break; + } case SOCKET_REQ_CONTAINER: r->req = SOCKET_REQ_CONTAINER; fd = rte_vfio_get_container_fd(); @@ -54,6 +65,7 @@ vfio_mp_primary(const struct rte_mp_msg *msg, const void *peer) r->result = SOCKET_ERR; else { r->result = SOCKET_OK; + r->mode = vfio_cfg.mode; reply.num_fds = 1; reply.fds[0] = fd; } @@ -62,6 +74,13 @@ vfio_mp_primary(const struct rte_mp_msg *msg, const void *peer) { int iommu_type_id; + if (vfio_cfg.mode != RTE_VFIO_MODE_GROUP && + vfio_cfg.mode != RTE_VFIO_MODE_NOIOMMU) { + EAL_LOG(ERR, "VFIO not initialized in group mode"); + r->result = SOCKET_ERR; + break; + } + r->req = SOCKET_REQ_IOMMU_TYPE; iommu_type_id = vfio_get_iommu_type(); @@ -90,8 +109,11 @@ vfio_mp_sync_setup(void) { if (rte_eal_process_type() == RTE_PROC_PRIMARY) { int ret = rte_mp_action_register(EAL_VFIO_MP, vfio_mp_primary); - if (ret && rte_errno != ENOTSUP) + if (ret && rte_errno != ENOTSUP) { + EAL_LOG(ERR, "Multiprocess sync setup failed: %d (%s)", + rte_errno, rte_strerror(rte_errno)); return -1; + } } return 0; diff --git a/lib/eal/linux/meson.build b/lib/eal/linux/meson.build index e99ebed256..75a9afdd03 100644 --- a/lib/eal/linux/meson.build +++ b/lib/eal/linux/meson.build @@ -16,6 +16,7 @@ sources += files( 'eal_thread.c', 'eal_timer.c', 'eal_vfio.c', + 'eal_vfio_group.c', 'eal_vfio_mp_sync.c', ) -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* [PATCH v3 15/20] bus/pci: use the new VFIO mode API 2025-11-18 16:29 ` [PATCH v3 00/20] Support VFIO cdev API in DPDK Anatoly Burakov ` (13 preceding siblings ...) 2025-11-18 16:29 ` [PATCH v3 14/20] vfio: cleanup and refactor Anatoly Burakov @ 2025-11-18 16:29 ` Anatoly Burakov 2025-11-18 16:29 ` [PATCH v3 16/20] bus/fslmc: " Anatoly Burakov ` (4 subsequent siblings) 19 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2025-11-18 16:29 UTC (permalink / raw) To: dev, Chenbo Xia, Nipun Gupta Use the new VFIO mode API to query no-IOMMU status. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- drivers/bus/pci/linux/pci.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/bus/pci/linux/pci.c b/drivers/bus/pci/linux/pci.c index 66d7e09a6e..6f7b67e140 100644 --- a/drivers/bus/pci/linux/pci.c +++ b/drivers/bus/pci/linux/pci.c @@ -598,7 +598,7 @@ pci_device_iova_mode(const struct rte_pci_driver *pdrv, static int is_vfio_noiommu_enabled = -1; if (is_vfio_noiommu_enabled == -1) { - if (rte_vfio_noiommu_is_enabled() == 1) + if (rte_vfio_get_mode() == RTE_VFIO_MODE_NOIOMMU) is_vfio_noiommu_enabled = 1; else is_vfio_noiommu_enabled = 0; -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* [PATCH v3 16/20] bus/fslmc: use the new VFIO mode API 2025-11-18 16:29 ` [PATCH v3 00/20] Support VFIO cdev API in DPDK Anatoly Burakov ` (14 preceding siblings ...) 2025-11-18 16:29 ` [PATCH v3 15/20] bus/pci: use the new VFIO mode API Anatoly Burakov @ 2025-11-18 16:29 ` Anatoly Burakov 2025-11-18 16:29 ` [PATCH v3 17/20] net/hinic3: " Anatoly Burakov ` (3 subsequent siblings) 19 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2025-11-18 16:29 UTC (permalink / raw) To: dev, Hemant Agrawal, Sachin Saxena FSLMC bus only supports operating in group mode, and relies on no-IOMMU mode checks. Use the new VFIO API's to query no-IOMMU status, as well as protect the bus from initializing in non-group mode. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- drivers/bus/fslmc/fslmc_bus.c | 10 +++++++++- drivers/bus/fslmc/fslmc_vfio.c | 2 +- 2 files changed, 10 insertions(+), 2 deletions(-) diff --git a/drivers/bus/fslmc/fslmc_bus.c b/drivers/bus/fslmc/fslmc_bus.c index 4529ec5085..82b53bcc54 100644 --- a/drivers/bus/fslmc/fslmc_bus.c +++ b/drivers/bus/fslmc/fslmc_bus.c @@ -335,6 +335,13 @@ rte_fslmc_scan(void) goto scan_fail; } + /* for container groups to work, VFIO must be in group mode */ + if (rte_vfio_get_mode() != RTE_VFIO_MODE_GROUP && + rte_vfio_get_mode() != RTE_VFIO_MODE_NOIOMMU) { + ret = -EINVAL; + goto scan_fail; + } + ret = fslmc_get_container_group(group_name, &groupid); if (ret != 0) goto scan_fail; @@ -582,7 +589,8 @@ rte_dpaa2_get_iommu_class(void) return RTE_IOVA_DC; /* check if all devices on the bus support Virtual addressing or not */ - if (fslmc_all_device_support_iova() != 0 && rte_vfio_noiommu_is_enabled() == 0) + if (fslmc_all_device_support_iova() != 0 && + rte_vfio_get_mode() != RTE_VFIO_MODE_NOIOMMU) return RTE_IOVA_VA; return RTE_IOVA_PA; diff --git a/drivers/bus/fslmc/fslmc_vfio.c b/drivers/bus/fslmc/fslmc_vfio.c index 4b11b38322..afde967e34 100644 --- a/drivers/bus/fslmc/fslmc_vfio.c +++ b/drivers/bus/fslmc/fslmc_vfio.c @@ -192,7 +192,7 @@ fslmc_vfio_add_group(int vfio_group_fd, group->fd = vfio_group_fd; group->groupid = iommu_group_num; rte_strscpy(group->group_name, group_name, sizeof(group->group_name)); - if (rte_vfio_noiommu_is_enabled() > 0) + if (rte_vfio_get_mode() == RTE_VFIO_MODE_NOIOMMU) group->iommu_type = VFIO_NOIOMMU_IOMMU; else group->iommu_type = VFIO_TYPE1_IOMMU; -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* [PATCH v3 17/20] net/hinic3: use the new VFIO mode API 2025-11-18 16:29 ` [PATCH v3 00/20] Support VFIO cdev API in DPDK Anatoly Burakov ` (15 preceding siblings ...) 2025-11-18 16:29 ` [PATCH v3 16/20] bus/fslmc: " Anatoly Burakov @ 2025-11-18 16:29 ` Anatoly Burakov 2025-11-18 16:29 ` [PATCH v3 18/20] net/ntnic: " Anatoly Burakov ` (2 subsequent siblings) 19 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2025-11-18 16:29 UTC (permalink / raw) To: dev, Feifei Wang Use new VFIO mode API to query no-IOMMU status. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- drivers/net/hinic3/base/hinic3_hwdev.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/net/hinic3/base/hinic3_hwdev.c b/drivers/net/hinic3/base/hinic3_hwdev.c index 668bbf4a0e..c5bf2bc8a4 100644 --- a/drivers/net/hinic3/base/hinic3_hwdev.c +++ b/drivers/net/hinic3/base/hinic3_hwdev.c @@ -77,7 +77,7 @@ bool hinic3_is_vfio_iommu_enable(const struct rte_eth_dev *rte_dev) { return ((RTE_ETH_DEV_TO_PCI(rte_dev)->kdrv == RTE_PCI_KDRV_VFIO) && - (rte_vfio_noiommu_is_enabled() != 1)); + (rte_vfio_get_mode() != RTE_VFIO_MODE_NOIOMMU)); } int -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* [PATCH v3 18/20] net/ntnic: use the new VFIO mode API 2025-11-18 16:29 ` [PATCH v3 00/20] Support VFIO cdev API in DPDK Anatoly Burakov ` (16 preceding siblings ...) 2025-11-18 16:29 ` [PATCH v3 17/20] net/hinic3: " Anatoly Burakov @ 2025-11-18 16:29 ` Anatoly Burakov 2025-11-18 16:29 ` [PATCH v3 19/20] vfio: remove no-IOMMU check API Anatoly Burakov 2025-11-18 16:29 ` [PATCH v3 20/20] vfio: introduce cdev mode Anatoly Burakov 19 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2025-11-18 16:29 UTC (permalink / raw) To: dev, Christian Koue Muf, Serhii Iliushyk Use new VFIO mode API to query no-IOMMU status. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- drivers/net/ntnic/ntnic_ethdev.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/net/ntnic/ntnic_ethdev.c b/drivers/net/ntnic/ntnic_ethdev.c index 23c3452d2d..c1195e5fbe 100644 --- a/drivers/net/ntnic/ntnic_ethdev.c +++ b/drivers/net/ntnic/ntnic_ethdev.c @@ -2692,7 +2692,7 @@ nthw_pci_probe(struct rte_pci_driver *pci_drv __rte_unused, (pci_dev->device.devargs->data ? pci_dev->device.devargs->data : "NULL")); } - const int n_rte_vfio_no_io_mmu_enabled = rte_vfio_noiommu_is_enabled(); + const int n_rte_vfio_no_io_mmu_enabled = rte_vfio_get_mode() == RTE_VFIO_MODE_NOIOMMU; NT_LOG(DBG, NTNIC, "vfio_no_iommu_enabled=%d", n_rte_vfio_no_io_mmu_enabled); if (n_rte_vfio_no_io_mmu_enabled) { -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* [PATCH v3 19/20] vfio: remove no-IOMMU check API 2025-11-18 16:29 ` [PATCH v3 00/20] Support VFIO cdev API in DPDK Anatoly Burakov ` (17 preceding siblings ...) 2025-11-18 16:29 ` [PATCH v3 18/20] net/ntnic: " Anatoly Burakov @ 2025-11-18 16:29 ` Anatoly Burakov 2025-11-18 16:29 ` [PATCH v3 20/20] vfio: introduce cdev mode Anatoly Burakov 19 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2025-11-18 16:29 UTC (permalink / raw) To: dev, Bruce Richardson, Tyler Retzlaff The `rte_vfio_noiommu_is_enabled()` check has now been replaced by the new mode API, so remove it. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- lib/eal/freebsd/eal.c | 6 ------ lib/eal/include/rte_vfio.h | 14 -------------- lib/eal/linux/eal_vfio.c | 7 ------- 3 files changed, 27 deletions(-) diff --git a/lib/eal/freebsd/eal.c b/lib/eal/freebsd/eal.c index 7b3ce875a8..d850ef2c4f 100644 --- a/lib/eal/freebsd/eal.c +++ b/lib/eal/freebsd/eal.c @@ -842,12 +842,6 @@ int rte_vfio_is_enabled(__rte_unused const char *modname) return 0; } -RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_noiommu_is_enabled) -int rte_vfio_noiommu_is_enabled(void) -{ - return 0; -} - RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_group_num) int rte_vfio_get_group_num(__rte_unused const char *sysfs_base, diff --git a/lib/eal/include/rte_vfio.h b/lib/eal/include/rte_vfio.h index b6d7b4591f..2730cfa67d 100644 --- a/lib/eal/include/rte_vfio.h +++ b/lib/eal/include/rte_vfio.h @@ -170,20 +170,6 @@ __rte_internal enum rte_vfio_mode rte_vfio_get_mode(void); -/** - * @internal - * Check if VFIO NOIOMMU mode is enabled. - * - * This function is only relevant on Linux in group mode. - * - * @return - * 1 if enabled. - * 0 if not enabled or not supported. - */ -__rte_internal -int -rte_vfio_noiommu_is_enabled(void); - /** * @internal * Parse IOMMU group number for a device. diff --git a/lib/eal/linux/eal_vfio.c b/lib/eal/linux/eal_vfio.c index 8545552308..2e9ad7dbb4 100644 --- a/lib/eal/linux/eal_vfio.c +++ b/lib/eal/linux/eal_vfio.c @@ -1310,13 +1310,6 @@ container_dma_unmap(struct container *cfg, uint64_t vaddr, uint64_t iova, return ret; } -RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_noiommu_is_enabled) -int -rte_vfio_noiommu_is_enabled(void) -{ - return vfio_cfg.mode == RTE_VFIO_MODE_NOIOMMU; -} - RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_create) int rte_vfio_container_create(void) -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* [PATCH v3 20/20] vfio: introduce cdev mode 2025-11-18 16:29 ` [PATCH v3 00/20] Support VFIO cdev API in DPDK Anatoly Burakov ` (18 preceding siblings ...) 2025-11-18 16:29 ` [PATCH v3 19/20] vfio: remove no-IOMMU check API Anatoly Burakov @ 2025-11-18 16:29 ` Anatoly Burakov 19 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2025-11-18 16:29 UTC (permalink / raw) To: dev, Bruce Richardson, Tyler Retzlaff Add support for VFIO cdev (also known as IOMMUFD) API. The group API is now considered legacy in the kernel, and all further development is expected to happen in IOMMUFD infrastructure. To assist any future use of VFIO cdev mode for custom behavior, also introduce "get device number" API, which is kind-of-but-not-really similar to the concept of IOMMU group. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- lib/eal/freebsd/eal.c | 10 + lib/eal/include/rte_vfio.h | 31 +++ lib/eal/linux/eal_vfio.c | 210 +++++++++++++++++ lib/eal/linux/eal_vfio.h | 29 ++- lib/eal/linux/eal_vfio_cdev.c | 389 +++++++++++++++++++++++++++++++ lib/eal/linux/eal_vfio_mp_sync.c | 42 ++++ lib/eal/linux/meson.build | 1 + 7 files changed, 710 insertions(+), 2 deletions(-) create mode 100644 lib/eal/linux/eal_vfio_cdev.c diff --git a/lib/eal/freebsd/eal.c b/lib/eal/freebsd/eal.c index d850ef2c4f..c993d383bf 100644 --- a/lib/eal/freebsd/eal.c +++ b/lib/eal/freebsd/eal.c @@ -923,3 +923,13 @@ rte_vfio_get_mode(void) { return RTE_VFIO_MODE_NONE; } + +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_device_num) +int +rte_vfio_get_device_num(__rte_unused const char *sysfs_base, + __rte_unused const char *dev_addr, + __rte_unused int *vfio_device_num) +{ + rte_errno = ENOTSUP; + return -1; +} diff --git a/lib/eal/include/rte_vfio.h b/lib/eal/include/rte_vfio.h index 2730cfa67d..385e09b5ba 100644 --- a/lib/eal/include/rte_vfio.h +++ b/lib/eal/include/rte_vfio.h @@ -28,6 +28,8 @@ extern "C" { #define RTE_VFIO_DIR "/dev/vfio" #define RTE_VFIO_CONTAINER_PATH "/dev/vfio/vfio" +#define RTE_VFIO_IOMMUFD_PATH "/dev/iommu" +#define RTE_VFIO_CDEV_DEVICES_PATH "/dev/vfio/devices" #define RTE_VFIO_GROUP_FMT "/dev/vfio/%u" #define RTE_VFIO_NOIOMMU_GROUP_FMT "/dev/vfio/noiommu-%u" #define RTE_VFIO_NOIOMMU_MODE "/sys/module/vfio/parameters/enable_unsafe_noiommu_mode" @@ -48,11 +50,13 @@ struct vfio_device_info; * - RTE_VFIO_MODE_NONE: VFIO is not enabled. * - RTE_VFIO_MODE_GROUP: Legacy group mode. * - RTE_VFIO_MODE_NOIOMMU: Unsafe no-IOMMU mode. + * - RTE_VFIO_MODE_CDEV: Character device mode. */ enum rte_vfio_mode { RTE_VFIO_MODE_NONE = 0, /**< VFIO not enabled */ RTE_VFIO_MODE_GROUP, /**< Group mode */ RTE_VFIO_MODE_NOIOMMU, /**< Group mode with no IOMMU protection */ + RTE_VFIO_MODE_CDEV, /**< Device mode */ }; /** @@ -197,6 +201,33 @@ __rte_internal int rte_vfio_get_group_num(const char *sysfs_base, const char *dev_addr, int *iommu_group_num); +/** + * @internal + * Parse VFIO cdev device number for a device. + * + * This function is only relevant on Linux in cdev mode. + * + * @param sysfs_base + * Sysfs path prefix. + * @param dev_addr + * Device identifier. + * @param vfio_device_num + * Pointer to where VFIO cdev device number will be stored. + * + * @return + * 0 on success. + * <0 on failure, rte_errno is set. + * + * Possible rte_errno values include: + * - ENODEV - Device not managed by VFIO. + * - EINVAL - Invalid parameters. + * - ENXIO - VFIO support not initialized. + * - ENOTSUP - Unsupported VFIO mode. + */ +__rte_internal +int +rte_vfio_get_device_num(const char *sysfs_base, const char *dev_addr, int *vfio_device_num); + /** * @internal * Get device information. diff --git a/lib/eal/linux/eal_vfio.c b/lib/eal/linux/eal_vfio.c index 2e9ad7dbb4..e1d1de494b 100644 --- a/lib/eal/linux/eal_vfio.c +++ b/lib/eal/linux/eal_vfio.c @@ -369,6 +369,20 @@ vfio_container_get_by_group_num(int group_num) return NULL; } +static struct container * +vfio_container_get_by_dev_num(int dev_num) +{ + struct container *cfg; + struct vfio_device *dev; + + CONTAINER_FOREACH_ACTIVE(cfg) { + DEVICE_FOREACH_ACTIVE(cfg, dev) + if (dev->dev_num == dev_num) + return cfg; + } + return NULL; +} + static struct container * vfio_container_create(void) { @@ -538,6 +552,55 @@ vfio_setup_dma_mem(struct container *cfg) return 0; } +static enum vfio_result +vfio_cdev_assign_device(struct container *cfg, const char *sysfs_base, + const char *dev_addr, struct vfio_device **out_dev) +{ + struct vfio_device *dev, *found_dev; + enum vfio_result res; + int dev_num, ret; + + /* get the cdev device number from sysfs */ + ret = vfio_cdev_get_device_num(sysfs_base, dev_addr, &dev_num); + if (ret < 0) { + EAL_LOG(ERR, "Failed to get cdev device number for %s", dev_addr); + return VFIO_ERROR; + } else if (ret == 0) { + EAL_LOG(ERR, "Device %s not bound to vfio-pci cdev", dev_addr); + return VFIO_NOT_MANAGED; + } + + /* do we already have this device? */ + found_dev = vfio_cdev_get_dev_by_num(cfg, dev_num); + if (found_dev != NULL) { + EAL_LOG(ERR, "Device %s already assigned to this container", dev_addr); + *out_dev = found_dev; + return VFIO_EXISTS; + } + /* create new device structure */ + dev = vfio_device_create(cfg); + if (dev == NULL) { + EAL_LOG(ERR, "No space to track new VFIO cdev device"); + return VFIO_NO_SPACE; + } + /* store device number */ + dev->dev_num = dev_num; + + /* set up our device now and store it in config */ + ret = vfio_cdev_setup_device(cfg, dev); + if (ret < 0) { + EAL_LOG(ERR, "Cannot setup cdev device %s", dev_addr); + res = VFIO_ERROR; + goto err; + } + *out_dev = dev; + return VFIO_SUCCESS; + +err: + vfio_device_erase(cfg, dev); + return res; +} + static enum vfio_result vfio_group_assign_device(struct container *cfg, const char *sysfs_base, const char *dev_addr, struct vfio_device **out_dev) @@ -695,6 +758,49 @@ rte_vfio_container_assign_device(int container_fd, const char *sysfs_base, const return -1; } + /* + * The device-to-container assignment is a complex problem to solve, for the following + * reasons: + * + * 1. PCI infrastructure is decoupled from VFIO, so PCI does not know anything about VFIO + * + * This means that while 99% of VFIO usage is PCI-related, we cannot communicate to PCI that + * we want to map a particular device using a particular container. Previously, this was + * achieved using back-channel communication via IOMMU group binding, so that whenever PCI + * map actually happens, VFIO knows which container to use, so this is roughly the model we + * are going with. + * + * 2. VFIO cannot depend on PCI because VFIO is in EAL + * + * We cannot "assign" a PCI device to container using rte_pci_device pointer because VFIO + * cannot depend on PCI definitions, nor can't we even assume that our device is in fact a + * PCI device, even though in practice this is true (at the time of this writing, FSLMC is + * the only bus doing non-PCI VFIO mappings, but FSLMC manages all VFIO infrastructure by + * itself, so in practice even counting FSLMC bus, we're always dealing with PCI devices). + * + * 3. The "assignment" means different things for group and cdev mode + * + * In group mode, to "bind" a device to a specific container, it is enough to bind its + * IOMMU group, so that when rte_vfio_setup_device() is called, we simply retrieve already + * existing group, and through that we figure out which container to use. + * + * For cdev mode, there are no "groups", so "assignment" either means we store some kind of + * uniquely identifying token (such as device number, or an opaque pointer), or we simply + * open the device straight away, and when rte_vfio_setup_device() comes we simply return + * the fd that was already opened at assign. + * + * Doing it the latter way (opening the device at assign for both group and cdev modes) + * actually solves all of these problems, so that's what we're going to do - the device + * setup API call will actually just assign the device to default container, while release + * will automatically cleanup and unassign anything that needs unassigned. There will be no + * "unassign" call, as it is not necessary. + * + * There is one downside for group mode when adding duplicate devices: to get to device fd, + * we need to go through the entire codepath before we arrive at fd only to realize it was + * already opened earlier, but this is acceptable compromise for unifying the API around + * device assignment. + */ + if (vfio_cfg.mode == RTE_VFIO_MODE_NONE) { EAL_LOG(ERR, "VFIO support not initialized"); rte_errno = ENXIO; @@ -715,6 +821,9 @@ rte_vfio_container_assign_device(int container_fd, const char *sysfs_base, const case RTE_VFIO_MODE_NOIOMMU: res = vfio_group_assign_device(cfg, sysfs_base, dev_addr, &dev); break; + case RTE_VFIO_MODE_CDEV: + res = vfio_cdev_assign_device(cfg, sysfs_base, dev_addr, &dev); + break; default: EAL_LOG(ERR, "Unsupported VFIO mode"); res = VFIO_NOT_SUPPORTED; @@ -787,6 +896,24 @@ rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, res = vfio_group_assign_device(cfg, sysfs_base, dev_addr, &dev); break; } + case RTE_VFIO_MODE_CDEV: + { + int dev_num; + + /* find device number */ + ret = vfio_cdev_get_device_num(sysfs_base, dev_addr, &dev_num); + if (ret < 0) + goto assign_fail; + else if (ret == 0) + goto not_managed; + + cfg = vfio_container_get_by_dev_num(dev_num); + if (cfg == NULL) + cfg = vfio_cfg.default_cfg; + + res = vfio_cdev_assign_device(cfg, sysfs_base, dev_addr, &dev); + break; + } default: EAL_LOG(ERR, "Unsupported VFIO mode"); rte_errno = ENOTSUP; @@ -906,6 +1033,12 @@ rte_vfio_release_device(const char *sysfs_base __rte_unused, } break; } + case RTE_VFIO_MODE_CDEV: + { + /* for cdev, just erase the device and we're done */ + vfio_device_erase(cfg, dev); + break; + } default: EAL_LOG(ERR, "Unsupported VFIO mode"); rte_errno = ENOTSUP; @@ -971,6 +1104,9 @@ vfio_select_mode(void) if (vfio_sync_mode(cfg, &mode) < 0) goto err; + /* if primary is in cdev mode, we need to sync ioas as well */ + if (mode == RTE_VFIO_MODE_CDEV && vfio_cdev_sync_ioas(cfg) < 0) + goto err; /* primary handles DMA setup for default containers */ group_cfg->dma_setup_done = true; @@ -990,6 +1126,19 @@ vfio_select_mode(void) return RTE_VFIO_MODE_NOIOMMU; return RTE_VFIO_MODE_GROUP; } + EAL_LOG(DEBUG, "VFIO group mode not available, trying cdev mode..."); + /* try cdev mode */ + if (vfio_cdev_enable(cfg) == 0) { + if (vfio_cdev_setup_ioas(cfg) < 0) + goto err_mpsync; + if (vfio_setup_dma_mem(cfg) < 0) + goto err_mpsync; + if (vfio_register_mem_event_callback() < 0) + goto err_mpsync; + + return RTE_VFIO_MODE_CDEV; + } + EAL_LOG(DEBUG, "VFIO cdev mode not available"); err_mpsync: vfio_mp_sync_cleanup(); err: @@ -1004,6 +1153,7 @@ vfio_mode_to_str(enum rte_vfio_mode mode) switch (mode) { case RTE_VFIO_MODE_GROUP: return "group"; case RTE_VFIO_MODE_NOIOMMU: return "noiommu"; + case RTE_VFIO_MODE_CDEV: return "cdev"; default: return "not initialized"; } } @@ -1143,6 +1293,40 @@ rte_vfio_get_group_num(const char *sysfs_base, const char *dev_addr, int *iommu_ return 0; } +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_device_num) +int +rte_vfio_get_device_num(const char *sysfs_base, const char *dev_addr, int *device_num) +{ + int ret; + + if (sysfs_base == NULL || dev_addr == NULL || device_num == NULL) { + rte_errno = EINVAL; + return -1; + } + + if (vfio_cfg.mode == RTE_VFIO_MODE_NONE) { + EAL_LOG(ERR, "VFIO support not initialized"); + rte_errno = ENXIO; + return -1; + } + + if (vfio_cfg.mode != RTE_VFIO_MODE_CDEV) { + EAL_LOG(ERR, "VFIO not initialized in cdev mode"); + rte_errno = ENOTSUP; + return -1; + } + + ret = vfio_cdev_get_device_num(sysfs_base, dev_addr, device_num); + if (ret < 0) { + rte_errno = EINVAL; + return -1; + } else if (ret == 0) { + rte_errno = ENODEV; + return -1; + } + return 0; +} + static int vfio_dma_mem_map(struct container *cfg, uint64_t vaddr, uint64_t iova, uint64_t len, int do_map) @@ -1342,6 +1526,25 @@ rte_vfio_container_create(void) cfg->container_fd = container_fd; break; } + case RTE_VFIO_MODE_CDEV: + { + /* Open new iommufd for custom container */ + container_fd = vfio_cdev_get_iommufd(); + if (container_fd < 0) { + EAL_LOG(ERR, "Cannot open iommufd for cdev container"); + rte_errno = EIO; + goto err; + } + cfg->container_fd = container_fd; + + /* Set up IOAS for this container */ + if (vfio_cdev_setup_ioas(cfg) < 0) { + EAL_LOG(ERR, "Cannot setup IOAS for cdev container"); + rte_errno = EIO; + goto err; + } + break; + } default: EAL_LOG(NOTICE, "Unsupported VFIO mode"); rte_errno = ENOTSUP; @@ -1400,6 +1603,13 @@ rte_vfio_container_destroy(int container_fd) vfio_group_erase(cfg, grp); } break; + case RTE_VFIO_MODE_CDEV: + /* erase all devices */ + DEVICE_FOREACH_ACTIVE(cfg, dev) { + EAL_LOG(DEBUG, "Device vfio%d still open, closing", dev->dev_num); + vfio_device_erase(cfg, dev); + } + break; default: EAL_LOG(ERR, "Unsupported VFIO mode"); rte_errno = ENOTSUP; diff --git a/lib/eal/linux/eal_vfio.h b/lib/eal/linux/eal_vfio.h index 68d3a3ec6e..52cb3a0e08 100644 --- a/lib/eal/linux/eal_vfio.h +++ b/lib/eal/linux/eal_vfio.h @@ -48,7 +48,10 @@ struct vfio_group { /* device tracking (common for group and cdev modes) */ struct vfio_device { bool active; - int group; /**< back-reference to group list (group mode) */ + union { + int group; /**< back-reference to group list (group mode) */ + int dev_num; /**< device number, e.g., X in /dev/vfio/devices/vfioX (cdev mode) */ + }; int fd; }; @@ -61,12 +64,20 @@ struct vfio_group_config { struct vfio_group groups[RTE_MAX_VFIO_GROUPS]; }; +/* cdev mode specific configuration */ +struct vfio_cdev_config { + uint32_t ioas_id; +}; + /* per-container configuration */ struct container { bool active; int container_fd; struct user_mem_maps mem_maps; - struct vfio_group_config group_cfg; + union { + struct vfio_group_config group_cfg; + struct vfio_cdev_config cdev_cfg; + }; int n_devices; struct vfio_device devices[RTE_MAX_VFIO_DEVICES]; }; @@ -160,12 +171,24 @@ int vfio_group_setup_iommu(struct container *cfg); int vfio_group_setup_device_fd(const char *dev_addr, struct vfio_group *grp, struct vfio_device *dev); +/* cdev mode functions */ +int vfio_cdev_enable(struct container *cfg); +int vfio_cdev_setup_ioas(struct container *cfg); +int vfio_cdev_sync_ioas(struct container *cfg); +int vfio_cdev_get_iommufd(void); +int vfio_cdev_get_device_num(const char *sysfs_base, const char *dev_addr, + int *cdev_dev_num); +struct vfio_device *vfio_cdev_get_dev_by_num(struct container *cfg, int cdev_dev_num); +int vfio_cdev_setup_device(struct container *cfg, struct vfio_device *dev); + #define VFIO_MEM_EVENT_CLB_NAME "vfio_mem_event_clb" #define EAL_VFIO_MP "eal_vfio_mp_sync" #define SOCKET_REQ_CONTAINER 0x100 #define SOCKET_REQ_GROUP 0x200 #define SOCKET_REQ_IOMMU_TYPE 0x400 +#define SOCKET_REQ_CDEV 0x800 +#define SOCKET_REQ_IOAS_ID 0x1000 #define SOCKET_OK 0x0 #define SOCKET_NO_FD 0x1 #define SOCKET_ERR 0xFF @@ -176,6 +199,8 @@ struct vfio_mp_param { union { int group_num; int iommu_type_id; + int cdev_dev_num; + int ioas_id; enum rte_vfio_mode mode; }; }; diff --git a/lib/eal/linux/eal_vfio_cdev.c b/lib/eal/linux/eal_vfio_cdev.c new file mode 100644 index 0000000000..d2d48faa8e --- /dev/null +++ b/lib/eal/linux/eal_vfio_cdev.c @@ -0,0 +1,389 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright(c) 2010-2025 Intel Corporation + */ + +#include <dirent.h> +#include <fcntl.h> +#include <string.h> +#include <unistd.h> +#include <sys/ioctl.h> +#include <limits.h> + +#include <uapi/linux/iommufd.h> +#include <uapi/linux/vfio.h> + +#include <rte_log.h> +#include <rte_errno.h> +#include <rte_memory.h> +#include <rte_string_fns.h> + +#include "eal_vfio.h" +#include "eal_private.h" +#include "eal_internal_cfg.h" + +static int vfio_cdev_dma_map(struct container *cfg); +static int vfio_cdev_dma_mem_map(struct container *cfg, uint64_t vaddr, + uint64_t iova, uint64_t len, int do_map); + +/* IOMMUFD cdev mode IOMMU operations */ +static const struct vfio_iommu_ops iommufd_ops = { + .type_id = 0, /* cdev mode doesn't use type_id */ + .name = "IOMMUFD", + .partial_unmap = false, + .dma_map_func = &vfio_cdev_dma_map, + .dma_user_map_func = &vfio_cdev_dma_mem_map +}; + +static int +vfio_cdev_dma_mem_map(struct container *cfg, uint64_t vaddr, uint64_t iova, + uint64_t len, int do_map) +{ + struct iommu_ioas_map ioas_map; + struct iommu_ioas_unmap ioas_unmap; + int ret; + + if (do_map != 0) { + memset(&ioas_map, 0, sizeof(ioas_map)); + ioas_map.size = sizeof(struct iommu_ioas_map); + ioas_map.flags = IOMMU_IOAS_MAP_FIXED_IOVA | + IOMMU_IOAS_MAP_READABLE | + IOMMU_IOAS_MAP_WRITEABLE; + ioas_map.ioas_id = cfg->cdev_cfg.ioas_id; + ioas_map.user_va = vaddr; + ioas_map.length = len; + ioas_map.iova = iova; + + ret = ioctl(cfg->container_fd, IOMMU_IOAS_MAP, &ioas_map); + if (ret) { + /** + * In case the mapping was already done EEXIST will be + * returned from kernel. + */ + if (errno == EEXIST) { + EAL_LOG(DEBUG, + "Memory segment is already mapped, skipping"); + } else { + EAL_LOG(ERR, + "Cannot set up DMA remapping, error " + "%i (%s)", errno, strerror(errno)); + return -1; + } + } + } else { + memset(&ioas_unmap, 0, sizeof(ioas_unmap)); + ioas_unmap.size = sizeof(struct iommu_ioas_unmap); + ioas_unmap.ioas_id = cfg->cdev_cfg.ioas_id; + ioas_unmap.length = len; + ioas_unmap.iova = iova; + + ret = ioctl(cfg->container_fd, IOMMU_IOAS_UNMAP, &ioas_unmap); + if (ret) { + EAL_LOG(ERR, "Cannot clear DMA remapping, error " + "%i (%s)", errno, strerror(errno)); + return -1; + } + } + + return 0; +} + +static int +cdev_map(const struct rte_memseg_list *msl, const struct rte_memseg *ms, + void *arg) +{ + struct container *cfg = arg; + + /* skip external memory that isn't a heap */ + if (msl->external && !msl->heap) + return 0; + + /* skip any segments with invalid IOVA addresses */ + if (ms->iova == RTE_BAD_IOVA) + return 0; + + return vfio_cdev_dma_mem_map(cfg, ms->addr_64, ms->iova, ms->len, 1); +} + +static int +vfio_cdev_dma_map(struct container *cfg) +{ + return rte_memseg_walk(cdev_map, cfg); +} + +int +vfio_cdev_sync_ioas(struct container *cfg) +{ + struct rte_mp_msg mp_req, *mp_rep; + struct rte_mp_reply mp_reply = {0}; + struct timespec ts = {.tv_sec = 5, .tv_nsec = 0}; + struct vfio_mp_param *p = (struct vfio_mp_param *)mp_req.param; + + p->req = SOCKET_REQ_IOAS_ID; + rte_strscpy(mp_req.name, EAL_VFIO_MP, sizeof(mp_req.name)); + mp_req.len_param = sizeof(*p); + mp_req.num_fds = 0; + + if (rte_mp_request_sync(&mp_req, &mp_reply, &ts) == 0 && mp_reply.nb_received == 1) { + mp_rep = &mp_reply.msgs[0]; + p = (struct vfio_mp_param *)mp_rep->param; + if (p->result == SOCKET_OK && mp_rep->num_fds == 0) { + cfg->cdev_cfg.ioas_id = p->ioas_id; + free(mp_reply.msgs); + return 0; + } + } + + free(mp_reply.msgs); + EAL_LOG(ERR, "Cannot request ioas_id"); + return -1; +} + +int +vfio_cdev_setup_ioas(struct container *cfg) +{ + struct iommu_ioas_alloc ioas_alloc; + int ret; + + /* Allocate an IOAS */ + memset(&ioas_alloc, 0, sizeof(ioas_alloc)); + ioas_alloc.size = sizeof(struct iommu_ioas_alloc); + ioas_alloc.flags = 0; + + ret = ioctl(cfg->container_fd, IOMMU_IOAS_ALLOC, &ioas_alloc); + if (ret) { + EAL_LOG(ERR, "Cannot allocate IOAS, error %i (%s)", + errno, strerror(errno)); + return -1; + } + cfg->cdev_cfg.ioas_id = ioas_alloc.out_ioas_id; + + EAL_LOG(DEBUG, "Allocated IOAS with ID %u", cfg->cdev_cfg.ioas_id); + return 0; +} + +int +vfio_cdev_get_iommufd(void) +{ + int iommufd; + + /* if not requesting via mp, open iommufd locally */ + iommufd = open(RTE_VFIO_IOMMUFD_PATH, O_RDWR); + if (iommufd < 0) { + EAL_LOG(ERR, "Cannot open %s: %s", + RTE_VFIO_IOMMUFD_PATH, strerror(errno)); + return -1; + } + + return iommufd; +} + +int +vfio_cdev_enable(struct container *cfg) +{ + int iommufd; + + /* Check if iommufd device exists */ + if (access(RTE_VFIO_IOMMUFD_PATH, F_OK) != 0) { + EAL_LOG(DEBUG, + "IOMMUFD device does not exist, skipping VFIO cdev support..."); + return 1; + } + + /* open iommufd */ + iommufd = vfio_cdev_get_iommufd(); + if (iommufd < 0) + return -1; + + /* cdev mode does not have different IOMMU ops */ + vfio_cfg.ops = &iommufd_ops; + + cfg->container_fd = iommufd; + return 0; +} + +int +vfio_cdev_get_device_num(const char *sysfs_base, const char *dev_addr, int *cdev_dev_num) +{ + char linkname[PATH_MAX]; + char filename[PATH_MAX]; + char *dev_tok, *end; + int dev_num; + DIR *dir; + struct dirent *entry; + + memset(linkname, 0, sizeof(linkname)); + memset(filename, 0, sizeof(filename)); + + /* check if vfio-dev directory exists for this device */ + snprintf(linkname, sizeof(linkname), + "%s/%s/vfio-dev", sysfs_base, dev_addr); + + dir = opendir(linkname); + if (dir == NULL) { + /* device doesn't have vfio-dev, not bound to vfio-pci cdev */ + return 0; + } + + /* find vfioX entry in vfio-dev directory */ + while ((entry = readdir(dir)) != NULL) { + if (strncmp(entry->d_name, "vfio", 4) == 0) { + /* parse device number from vfioX */ + errno = 0; + dev_tok = entry->d_name + 4; /* skip "vfio" prefix */ + end = dev_tok; + dev_num = strtol(dev_tok, &end, 10); + if (end == dev_tok || *end != '\0' || errno != 0) { + EAL_LOG(ERR, "%s error parsing VFIO cdev device number!", + dev_addr); + closedir(dir); + return -1; + } + *cdev_dev_num = dev_num; + closedir(dir); + return 1; + } + } + + closedir(dir); + /* no vfio device found */ + return 0; +} + +struct vfio_device * +vfio_cdev_get_dev_by_num(struct container *cfg, int cdev_dev_num) +{ + struct vfio_device *dev; + /* find device handle */ + DEVICE_FOREACH_ACTIVE(cfg, dev) { + if (dev->dev_num != cdev_dev_num) + continue; + return dev; + } + return NULL; +} + +static int +cdev_open_device_fd(int cdev_dev_num) +{ + char devname[PATH_MAX] = {0}; + int dev_fd; + + snprintf(devname, sizeof(devname), "%s/vfio%d", + RTE_VFIO_CDEV_DEVICES_PATH, cdev_dev_num); + + dev_fd = open(devname, O_RDWR); + if (dev_fd < 0) { + EAL_LOG(ERR, "Cannot open %s: %s", devname, strerror(errno)); + return -1; + } + + return dev_fd; +} + +static int +cdev_attach_device_to_iommufd(struct container *cfg, struct vfio_device *dev) +{ + struct vfio_device_bind_iommufd bind = {0}; + struct vfio_device_attach_iommufd_pt attach = {0}; + rte_uuid_t vf_token; + + rte_eal_vfio_get_vf_token(vf_token); + + /* try with token first */ + if (!rte_uuid_is_null(vf_token)) { + bind.flags = VFIO_DEVICE_BIND_FLAG_TOKEN; + bind.token_uuid_ptr = (uintptr_t)&vf_token; + bind.argsz = sizeof(bind); + bind.iommufd = cfg->container_fd; + + /* this may fail because the kernel is too old */ + if (ioctl(dev->fd, VFIO_DEVICE_BIND_IOMMUFD, &bind) < 0) { + EAL_LOG(DEBUG, "Failed to bind device %d with VF token", dev->dev_num); + EAL_LOG(NOTICE, "Unable to use VF tokens with current kernel version."); + EAL_LOG(NOTICE, "Please use kernel >=6.17 or use group mode."); + /* erase the bind structure */ + bind = (struct vfio_device_bind_iommufd){0}; + } else { + goto attach; + } + } + bind.flags = 0; + bind.argsz = sizeof(bind); + bind.iommufd = cfg->container_fd; + + if (ioctl(dev->fd, VFIO_DEVICE_BIND_IOMMUFD, &bind) < 0) { + EAL_LOG(ERR, "Cannot bind device to IOMMUFD, error %i (%s)", + errno, strerror(errno)); + return -1; + } + +attach: + /* attach device to IOAS */ + attach.argsz = sizeof(attach); + attach.flags = 0; + attach.pt_id = cfg->cdev_cfg.ioas_id; + + if (ioctl(dev->fd, VFIO_DEVICE_ATTACH_IOMMUFD_PT, &attach) < 0) { + EAL_LOG(ERR, "Cannot attach device to IOAS, error %i (%s)", + errno, strerror(errno)); + return -1; + } + + return 0; +} + +static int +vfio_cdev_request_dev_fd(struct vfio_device *dev) +{ + struct rte_mp_msg mp_req, *mp_rep; + struct rte_mp_reply mp_reply = {0}; + struct timespec ts = {.tv_sec = 5, .tv_nsec = 0}; + struct vfio_mp_param *p = (struct vfio_mp_param *)mp_req.param; + int device_fd = -1; + + /* secondary process requests device fd from primary */ + p->req = SOCKET_REQ_CDEV; + p->cdev_dev_num = dev->dev_num; + rte_strscpy(mp_req.name, EAL_VFIO_MP, sizeof(mp_req.name)); + mp_req.len_param = sizeof(*p); + mp_req.num_fds = 0; + + if (rte_mp_request_sync(&mp_req, &mp_reply, &ts) == 0 && + mp_reply.nb_received == 1) { + mp_rep = &mp_reply.msgs[0]; + p = (struct vfio_mp_param *)mp_rep->param; + if (p->result == SOCKET_OK && mp_rep->num_fds == 1) + device_fd = mp_rep->fds[0]; + } + + free(mp_reply.msgs); + + if (device_fd < 0) { + EAL_LOG(ERR, "Cannot request device fd for vfio%d", dev->dev_num); + return -1; + } + dev->fd = device_fd; + + return 0; +} + +int +vfio_cdev_setup_device(struct container *cfg, struct vfio_device *dev) +{ + int device_fd; + + /* get device fd - primary or custom container opens it, secondary requests from primary */ + if (rte_eal_process_type() == RTE_PROC_PRIMARY || !vfio_container_is_default(cfg)) { + device_fd = cdev_open_device_fd(dev->dev_num); + if (device_fd < 0) + return -1; + dev->fd = device_fd; + + /* attach device to iommufd - only in primary */ + if (cdev_attach_device_to_iommufd(cfg, dev) < 0) + return -1; + } else if (vfio_cdev_request_dev_fd(dev) < 0) { + return -1; + } + return 0; +} diff --git a/lib/eal/linux/eal_vfio_mp_sync.c b/lib/eal/linux/eal_vfio_mp_sync.c index 9a07d35023..6d94f44af8 100644 --- a/lib/eal/linux/eal_vfio_mp_sync.c +++ b/lib/eal/linux/eal_vfio_mp_sync.c @@ -93,6 +93,48 @@ vfio_mp_primary(const struct rte_mp_msg *msg, const void *peer) } break; } + case SOCKET_REQ_CDEV: + { + struct container *cfg; + struct vfio_device *dev; + + if (vfio_cfg.mode != RTE_VFIO_MODE_CDEV) { + EAL_LOG(ERR, "VFIO not initialized in cdev mode"); + r->result = SOCKET_ERR; + break; + } + + r->req = SOCKET_REQ_CDEV; + r->cdev_dev_num = m->cdev_dev_num; + + cfg = vfio_cfg.default_cfg; + dev = vfio_cdev_get_dev_by_num(cfg, m->cdev_dev_num); + if (dev == NULL) { + r->result = SOCKET_NO_FD; + } else { + r->result = SOCKET_OK; + reply.num_fds = 1; + reply.fds[0] = dev->fd; + } + break; + } + case SOCKET_REQ_IOAS_ID: + { + struct container *cfg; + + if (vfio_cfg.mode != RTE_VFIO_MODE_CDEV) { + EAL_LOG(ERR, "VFIO not initialized in cdev mode"); + r->result = SOCKET_ERR; + break; + } + + r->req = SOCKET_REQ_IOAS_ID; + cfg = vfio_cfg.default_cfg; + r->ioas_id = cfg->cdev_cfg.ioas_id; + + r->result = SOCKET_OK; + break; + } default: EAL_LOG(ERR, "vfio received invalid message!"); return -1; diff --git a/lib/eal/linux/meson.build b/lib/eal/linux/meson.build index 75a9afdd03..adf043d24c 100644 --- a/lib/eal/linux/meson.build +++ b/lib/eal/linux/meson.build @@ -16,6 +16,7 @@ sources += files( 'eal_thread.c', 'eal_timer.c', 'eal_vfio.c', + 'eal_vfio_cdev.c', 'eal_vfio_group.c', 'eal_vfio_mp_sync.c', ) -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* [PATCH v7 00/18] Support VFIO cdev API in DPDK [not found] <cover.1763141462.git.anatoly.burakov@intel.com> 2025-11-18 16:29 ` [PATCH v3 00/20] Support VFIO cdev API in DPDK Anatoly Burakov @ 2026-02-26 14:17 ` Anatoly Burakov 2026-02-26 14:17 ` [PATCH v7 01/18] uapi: update to v6.17 and add iommufd.h Anatoly Burakov ` (18 more replies) 1 sibling, 19 replies; 42+ messages in thread From: Anatoly Burakov @ 2026-02-26 14:17 UTC (permalink / raw) To: dev This patchset introduces a major refactor of the VFIO subsystem in DPDK to support character device (cdev) interface introduced in Linux kernel, as well as make the API more streamlined and useful. The goal is to simplify device management, improve compatibility, and clarify API responsibilities. The following sections outline the key issues addressed by this patchset and the corresponding changes introduced. 1. Only group mode is supported =============================== Since kernel version 4.14.327 (LTS), VFIO supports the new character device (cdev)-based way of working with VFIO devices (otherwise known as IOMMUFD). This is a device-centric mode and does away with all the complexity regarding groups and IOMMU types, delegating it all to the kernel, and exposes a much simpler interface to userspace. The old group interface is still around, and will need to be kept in DPDK both for compatibility reasons, as well as supporting special cases (FSLMC bus, NBL driver, etc.). To enable this, VFIO is heavily refactored, so that the code can support both modes while relying on (mostly) common infrastructure. Note that the existing `rte_vfio_device_setup/release` model is fundamentally incompatible with cdev mode, because for custom container cases, the expected flow is that the user binds the IOMMU group (and thus, implicitly, the device itself) to a specific container using `rte_vfio_container_group_bind`, whereas this step is not needed for cdev as the device fd is assigned to the container straight away. Therefore, what we do instead is introduce a new API for container device assignment which, semantically, will assign a device to specified container, so that when it is mapped using `rte_pci_map_device`, the appropriate container is selected. Under the hood though, we essentially transition to getting device fd straight away at assign stage, so that by the time the PCI bus attempts to map the device, it is already mapped and we just return an fd. There is no "unassign" API because `release_device` already performs that function. Additionally, a new `rte_vfio_get_mode` API is added for those cases that need some introspection into VFIO's internals, with three new modes: group (old-style), no-iommu (old-style but without IOMMU), and cdev (the new mode). Although no-IOMMU is technically a variant of group mode, the distinction is largely irrelevant to the user, as all usages of noiommu checks in our codebase are for deciding whether to use IOVA or PA, not anything to do with managing groups. The current plan for kernel community is to *not* introduce no-IOMMU cdev implementation, which is why this will be kept for compatibility for these use cases. There were other users of VFIO which relied on group API but only for convenience purposes; no actual VFIO functionality depended on those API's. Therefore, group API's are removed and, where appropriate, replaced with the new API's. List of removed API's: * `rte_vfio_get_group_fd` * `rte_vfio_clear_group` * `rte_vfio_container_group_bind` (replaced by container assign API) * `rte_vfio_container_group_unbind` * `rte_vfio_noiommu_is_enabled` (replaced by new mode API) 2. The API responsibilities aren't clear and bleed into each other ================================================================== Some API's do multiple things at once. In particular: * `rte_vfio_get_device_info` will setup the device * `rte_vfio_setup_device` will get device info These API's have been adjusted to do one thing only. v7: - Rebase - Added removal of deprecation notices - Fixed implicit numeric comparison in patch 12 v6: - Fixed missing header include in vfio cdev file v5: - Added back missing uapi patch v4: - Fixed issues with documenting rte_vfio_mode enum - Separated deprecation notices into a separate patchset v3: - Make API removal cleaner - Fix `get_group_num` usages to align with new API - Fix issues with function exports - Fix issues with `setup_device` returning old-style values in some cases v2: - Make the entire API internal - More aggressive API pruning, complete removal of group API - Fixed a bug in group mode where device could not be used - Better documentation and deprecation notice patches - Moved doc patches to beginning of patchset Anatoly Burakov (18): uapi: update to v6.17 and add iommufd.h vfio: make all functions internal vfio: split get device info from setup vfio: add container device assignment API net/nbl: do not use VFIO group bind API net/ntnic: use container device assignment API vdpa/ifc: use container device assignment API vdpa/nfp: use container device assignment API vdpa/sfc: use container device assignment API vhost: remove group-related API from drivers vfio: remove group-based API vfio: cleanup and refactor bus/pci: use the new VFIO mode API bus/fslmc: use the new VFIO mode API net/hinic3: use the new VFIO mode API net/ntnic: use the new VFIO mode API vfio: remove no-IOMMU check API vfio: introduce cdev mode config/arm/meson.build | 1 + config/meson.build | 1 + doc/guides/prog_guide/vhost_lib.rst | 4 - doc/guides/rel_notes/deprecation.rst | 7 - drivers/bus/cdx/cdx_vfio.c | 25 +- drivers/bus/fslmc/fslmc_bus.c | 10 +- drivers/bus/fslmc/fslmc_vfio.c | 6 +- drivers/bus/pci/linux/pci.c | 2 +- drivers/bus/pci/linux/pci_vfio.c | 33 +- drivers/bus/platform/platform.c | 9 +- drivers/crypto/bcmfs/bcmfs_vfio.c | 14 +- drivers/net/hinic3/base/hinic3_hwdev.c | 2 +- drivers/net/nbl/nbl_common/nbl_userdev.c | 20 +- drivers/net/nbl/nbl_include/nbl_include.h | 1 + drivers/net/ntnic/ntnic_ethdev.c | 2 +- drivers/net/ntnic/ntnic_vfio.c | 30 +- drivers/vdpa/ifc/ifcvf_vdpa.c | 34 +- drivers/vdpa/mlx5/mlx5_vdpa.c | 1 - drivers/vdpa/nfp/nfp_vdpa.c | 37 +- drivers/vdpa/sfc/sfc_vdpa.c | 39 +- drivers/vdpa/sfc/sfc_vdpa.h | 2 - kernel/linux/uapi/linux/iommufd.h | 1292 +++++++++++ kernel/linux/uapi/linux/vduse.h | 2 +- kernel/linux/uapi/linux/vfio.h | 12 +- kernel/linux/uapi/version | 2 +- lib/eal/freebsd/eal.c | 98 +- lib/eal/include/rte_vfio.h | 387 ++-- lib/eal/linux/eal_vfio.c | 2433 ++++++++------------- lib/eal/linux/eal_vfio.h | 167 +- lib/eal/linux/eal_vfio_cdev.c | 390 ++++ lib/eal/linux/eal_vfio_group.c | 984 +++++++++ lib/eal/linux/eal_vfio_mp_sync.c | 80 +- lib/eal/linux/meson.build | 2 + lib/eal/windows/eal.c | 4 +- lib/vhost/vdpa_driver.h | 3 - 35 files changed, 4261 insertions(+), 1875 deletions(-) create mode 100644 kernel/linux/uapi/linux/iommufd.h create mode 100644 lib/eal/linux/eal_vfio_cdev.c create mode 100644 lib/eal/linux/eal_vfio_group.c -- 2.47.3 ^ permalink raw reply [flat|nested] 42+ messages in thread
* [PATCH v7 01/18] uapi: update to v6.17 and add iommufd.h 2026-02-26 14:17 ` [PATCH v7 00/18] Support VFIO cdev API in DPDK Anatoly Burakov @ 2026-02-26 14:17 ` Anatoly Burakov 2026-02-26 14:17 ` [PATCH v7 02/18] vfio: make all functions internal Anatoly Burakov ` (17 subsequent siblings) 18 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2026-02-26 14:17 UTC (permalink / raw) To: dev, Maxime Coquelin In order to support VF tokens for cdev-based VFIO mode, kernel v6.17 is required. Update internal headers to version v6.17 and include iommufd.h. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Acked-by: Stephen Hemminger <stephen@networkplumber.org> --- Depends-on: series-37486 ("doc: add deprecation notice for VFIO API") kernel/linux/uapi/linux/iommufd.h | 1292 +++++++++++++++++++++++++++++ kernel/linux/uapi/linux/vduse.h | 2 +- kernel/linux/uapi/linux/vfio.h | 12 +- kernel/linux/uapi/version | 2 +- 4 files changed, 1305 insertions(+), 3 deletions(-) create mode 100644 kernel/linux/uapi/linux/iommufd.h diff --git a/kernel/linux/uapi/linux/iommufd.h b/kernel/linux/uapi/linux/iommufd.h new file mode 100644 index 0000000000..2105a03955 --- /dev/null +++ b/kernel/linux/uapi/linux/iommufd.h @@ -0,0 +1,1292 @@ +/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */ +/* Copyright (c) 2021-2022, NVIDIA CORPORATION & AFFILIATES. + */ +#ifndef _IOMMUFD_H +#define _IOMMUFD_H + +#include <linux/ioctl.h> +#include <linux/types.h> + +#define IOMMUFD_TYPE (';') + +/** + * DOC: General ioctl format + * + * The ioctl interface follows a general format to allow for extensibility. Each + * ioctl is passed in a structure pointer as the argument providing the size of + * the structure in the first u32. The kernel checks that any structure space + * beyond what it understands is 0. This allows userspace to use the backward + * compatible portion while consistently using the newer, larger, structures. + * + * ioctls use a standard meaning for common errnos: + * + * - ENOTTY: The IOCTL number itself is not supported at all + * - E2BIG: The IOCTL number is supported, but the provided structure has + * non-zero in a part the kernel does not understand. + * - EOPNOTSUPP: The IOCTL number is supported, and the structure is + * understood, however a known field has a value the kernel does not + * understand or support. + * - EINVAL: Everything about the IOCTL was understood, but a field is not + * correct. + * - ENOENT: An ID or IOVA provided does not exist. + * - ENOMEM: Out of memory. + * - EOVERFLOW: Mathematics overflowed. + * + * As well as additional errnos, within specific ioctls. + */ +enum { + IOMMUFD_CMD_BASE = 0x80, + IOMMUFD_CMD_DESTROY = IOMMUFD_CMD_BASE, + IOMMUFD_CMD_IOAS_ALLOC = 0x81, + IOMMUFD_CMD_IOAS_ALLOW_IOVAS = 0x82, + IOMMUFD_CMD_IOAS_COPY = 0x83, + IOMMUFD_CMD_IOAS_IOVA_RANGES = 0x84, + IOMMUFD_CMD_IOAS_MAP = 0x85, + IOMMUFD_CMD_IOAS_UNMAP = 0x86, + IOMMUFD_CMD_OPTION = 0x87, + IOMMUFD_CMD_VFIO_IOAS = 0x88, + IOMMUFD_CMD_HWPT_ALLOC = 0x89, + IOMMUFD_CMD_GET_HW_INFO = 0x8a, + IOMMUFD_CMD_HWPT_SET_DIRTY_TRACKING = 0x8b, + IOMMUFD_CMD_HWPT_GET_DIRTY_BITMAP = 0x8c, + IOMMUFD_CMD_HWPT_INVALIDATE = 0x8d, + IOMMUFD_CMD_FAULT_QUEUE_ALLOC = 0x8e, + IOMMUFD_CMD_IOAS_MAP_FILE = 0x8f, + IOMMUFD_CMD_VIOMMU_ALLOC = 0x90, + IOMMUFD_CMD_VDEVICE_ALLOC = 0x91, + IOMMUFD_CMD_IOAS_CHANGE_PROCESS = 0x92, + IOMMUFD_CMD_VEVENTQ_ALLOC = 0x93, + IOMMUFD_CMD_HW_QUEUE_ALLOC = 0x94, +}; + +/** + * struct iommu_destroy - ioctl(IOMMU_DESTROY) + * @size: sizeof(struct iommu_destroy) + * @id: iommufd object ID to destroy. Can be any destroyable object type. + * + * Destroy any object held within iommufd. + */ +struct iommu_destroy { + __u32 size; + __u32 id; +}; +#define IOMMU_DESTROY _IO(IOMMUFD_TYPE, IOMMUFD_CMD_DESTROY) + +/** + * struct iommu_ioas_alloc - ioctl(IOMMU_IOAS_ALLOC) + * @size: sizeof(struct iommu_ioas_alloc) + * @flags: Must be 0 + * @out_ioas_id: Output IOAS ID for the allocated object + * + * Allocate an IO Address Space (IOAS) which holds an IO Virtual Address (IOVA) + * to memory mapping. + */ +struct iommu_ioas_alloc { + __u32 size; + __u32 flags; + __u32 out_ioas_id; +}; +#define IOMMU_IOAS_ALLOC _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_ALLOC) + +/** + * struct iommu_iova_range - ioctl(IOMMU_IOVA_RANGE) + * @start: First IOVA + * @last: Inclusive last IOVA + * + * An interval in IOVA space. + */ +struct iommu_iova_range { + __aligned_u64 start; + __aligned_u64 last; +}; + +/** + * struct iommu_ioas_iova_ranges - ioctl(IOMMU_IOAS_IOVA_RANGES) + * @size: sizeof(struct iommu_ioas_iova_ranges) + * @ioas_id: IOAS ID to read ranges from + * @num_iovas: Input/Output total number of ranges in the IOAS + * @__reserved: Must be 0 + * @allowed_iovas: Pointer to the output array of struct iommu_iova_range + * @out_iova_alignment: Minimum alignment required for mapping IOVA + * + * Query an IOAS for ranges of allowed IOVAs. Mapping IOVA outside these ranges + * is not allowed. num_iovas will be set to the total number of iovas and + * the allowed_iovas[] will be filled in as space permits. + * + * The allowed ranges are dependent on the HW path the DMA operation takes, and + * can change during the lifetime of the IOAS. A fresh empty IOAS will have a + * full range, and each attached device will narrow the ranges based on that + * device's HW restrictions. Detaching a device can widen the ranges. Userspace + * should query ranges after every attach/detach to know what IOVAs are valid + * for mapping. + * + * On input num_iovas is the length of the allowed_iovas array. On output it is + * the total number of iovas filled in. The ioctl will return -EMSGSIZE and set + * num_iovas to the required value if num_iovas is too small. In this case the + * caller should allocate a larger output array and re-issue the ioctl. + * + * out_iova_alignment returns the minimum IOVA alignment that can be given + * to IOMMU_IOAS_MAP/COPY. IOVA's must satisfy:: + * + * starting_iova % out_iova_alignment == 0 + * (starting_iova + length) % out_iova_alignment == 0 + * + * out_iova_alignment can be 1 indicating any IOVA is allowed. It cannot + * be higher than the system PAGE_SIZE. + */ +struct iommu_ioas_iova_ranges { + __u32 size; + __u32 ioas_id; + __u32 num_iovas; + __u32 __reserved; + __aligned_u64 allowed_iovas; + __aligned_u64 out_iova_alignment; +}; +#define IOMMU_IOAS_IOVA_RANGES _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_IOVA_RANGES) + +/** + * struct iommu_ioas_allow_iovas - ioctl(IOMMU_IOAS_ALLOW_IOVAS) + * @size: sizeof(struct iommu_ioas_allow_iovas) + * @ioas_id: IOAS ID to allow IOVAs from + * @num_iovas: Input/Output total number of ranges in the IOAS + * @__reserved: Must be 0 + * @allowed_iovas: Pointer to array of struct iommu_iova_range + * + * Ensure a range of IOVAs are always available for allocation. If this call + * succeeds then IOMMU_IOAS_IOVA_RANGES will never return a list of IOVA ranges + * that are narrower than the ranges provided here. This call will fail if + * IOMMU_IOAS_IOVA_RANGES is currently narrower than the given ranges. + * + * When an IOAS is first created the IOVA_RANGES will be maximally sized, and as + * devices are attached the IOVA will narrow based on the device restrictions. + * When an allowed range is specified any narrowing will be refused, ie device + * attachment can fail if the device requires limiting within the allowed range. + * + * Automatic IOVA allocation is also impacted by this call. MAP will only + * allocate within the allowed IOVAs if they are present. + * + * This call replaces the entire allowed list with the given list. + */ +struct iommu_ioas_allow_iovas { + __u32 size; + __u32 ioas_id; + __u32 num_iovas; + __u32 __reserved; + __aligned_u64 allowed_iovas; +}; +#define IOMMU_IOAS_ALLOW_IOVAS _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_ALLOW_IOVAS) + +/** + * enum iommufd_ioas_map_flags - Flags for map and copy + * @IOMMU_IOAS_MAP_FIXED_IOVA: If clear the kernel will compute an appropriate + * IOVA to place the mapping at + * @IOMMU_IOAS_MAP_WRITEABLE: DMA is allowed to write to this mapping + * @IOMMU_IOAS_MAP_READABLE: DMA is allowed to read from this mapping + */ +enum iommufd_ioas_map_flags { + IOMMU_IOAS_MAP_FIXED_IOVA = 1 << 0, + IOMMU_IOAS_MAP_WRITEABLE = 1 << 1, + IOMMU_IOAS_MAP_READABLE = 1 << 2, +}; + +/** + * struct iommu_ioas_map - ioctl(IOMMU_IOAS_MAP) + * @size: sizeof(struct iommu_ioas_map) + * @flags: Combination of enum iommufd_ioas_map_flags + * @ioas_id: IOAS ID to change the mapping of + * @__reserved: Must be 0 + * @user_va: Userspace pointer to start mapping from + * @length: Number of bytes to map + * @iova: IOVA the mapping was placed at. If IOMMU_IOAS_MAP_FIXED_IOVA is set + * then this must be provided as input. + * + * Set an IOVA mapping from a user pointer. If FIXED_IOVA is specified then the + * mapping will be established at iova, otherwise a suitable location based on + * the reserved and allowed lists will be automatically selected and returned in + * iova. + * + * If IOMMU_IOAS_MAP_FIXED_IOVA is specified then the iova range must currently + * be unused, existing IOVA cannot be replaced. + */ +struct iommu_ioas_map { + __u32 size; + __u32 flags; + __u32 ioas_id; + __u32 __reserved; + __aligned_u64 user_va; + __aligned_u64 length; + __aligned_u64 iova; +}; +#define IOMMU_IOAS_MAP _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_MAP) + +/** + * struct iommu_ioas_map_file - ioctl(IOMMU_IOAS_MAP_FILE) + * @size: sizeof(struct iommu_ioas_map_file) + * @flags: same as for iommu_ioas_map + * @ioas_id: same as for iommu_ioas_map + * @fd: the memfd to map + * @start: byte offset from start of file to map from + * @length: same as for iommu_ioas_map + * @iova: same as for iommu_ioas_map + * + * Set an IOVA mapping from a memfd file. All other arguments and semantics + * match those of IOMMU_IOAS_MAP. + */ +struct iommu_ioas_map_file { + __u32 size; + __u32 flags; + __u32 ioas_id; + __s32 fd; + __aligned_u64 start; + __aligned_u64 length; + __aligned_u64 iova; +}; +#define IOMMU_IOAS_MAP_FILE _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_MAP_FILE) + +/** + * struct iommu_ioas_copy - ioctl(IOMMU_IOAS_COPY) + * @size: sizeof(struct iommu_ioas_copy) + * @flags: Combination of enum iommufd_ioas_map_flags + * @dst_ioas_id: IOAS ID to change the mapping of + * @src_ioas_id: IOAS ID to copy from + * @length: Number of bytes to copy and map + * @dst_iova: IOVA the mapping was placed at. If IOMMU_IOAS_MAP_FIXED_IOVA is + * set then this must be provided as input. + * @src_iova: IOVA to start the copy + * + * Copy an already existing mapping from src_ioas_id and establish it in + * dst_ioas_id. The src iova/length must exactly match a range used with + * IOMMU_IOAS_MAP. + * + * This may be used to efficiently clone a subset of an IOAS to another, or as a + * kind of 'cache' to speed up mapping. Copy has an efficiency advantage over + * establishing equivalent new mappings, as internal resources are shared, and + * the kernel will pin the user memory only once. + */ +struct iommu_ioas_copy { + __u32 size; + __u32 flags; + __u32 dst_ioas_id; + __u32 src_ioas_id; + __aligned_u64 length; + __aligned_u64 dst_iova; + __aligned_u64 src_iova; +}; +#define IOMMU_IOAS_COPY _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_COPY) + +/** + * struct iommu_ioas_unmap - ioctl(IOMMU_IOAS_UNMAP) + * @size: sizeof(struct iommu_ioas_unmap) + * @ioas_id: IOAS ID to change the mapping of + * @iova: IOVA to start the unmapping at + * @length: Number of bytes to unmap, and return back the bytes unmapped + * + * Unmap an IOVA range. The iova/length must be a superset of a previously + * mapped range used with IOMMU_IOAS_MAP or IOMMU_IOAS_COPY. Splitting or + * truncating ranges is not allowed. The values 0 to U64_MAX will unmap + * everything. + */ +struct iommu_ioas_unmap { + __u32 size; + __u32 ioas_id; + __aligned_u64 iova; + __aligned_u64 length; +}; +#define IOMMU_IOAS_UNMAP _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_UNMAP) + +/** + * enum iommufd_option - ioctl(IOMMU_OPTION_RLIMIT_MODE) and + * ioctl(IOMMU_OPTION_HUGE_PAGES) + * @IOMMU_OPTION_RLIMIT_MODE: + * Change how RLIMIT_MEMLOCK accounting works. The caller must have privilege + * to invoke this. Value 0 (default) is user based accounting, 1 uses process + * based accounting. Global option, object_id must be 0 + * @IOMMU_OPTION_HUGE_PAGES: + * Value 1 (default) allows contiguous pages to be combined when generating + * iommu mappings. Value 0 disables combining, everything is mapped to + * PAGE_SIZE. This can be useful for benchmarking. This is a per-IOAS + * option, the object_id must be the IOAS ID. + */ +enum iommufd_option { + IOMMU_OPTION_RLIMIT_MODE = 0, + IOMMU_OPTION_HUGE_PAGES = 1, +}; + +/** + * enum iommufd_option_ops - ioctl(IOMMU_OPTION_OP_SET) and + * ioctl(IOMMU_OPTION_OP_GET) + * @IOMMU_OPTION_OP_SET: Set the option's value + * @IOMMU_OPTION_OP_GET: Get the option's value + */ +enum iommufd_option_ops { + IOMMU_OPTION_OP_SET = 0, + IOMMU_OPTION_OP_GET = 1, +}; + +/** + * struct iommu_option - iommu option multiplexer + * @size: sizeof(struct iommu_option) + * @option_id: One of enum iommufd_option + * @op: One of enum iommufd_option_ops + * @__reserved: Must be 0 + * @object_id: ID of the object if required + * @val64: Option value to set or value returned on get + * + * Change a simple option value. This multiplexor allows controlling options + * on objects. IOMMU_OPTION_OP_SET will load an option and IOMMU_OPTION_OP_GET + * will return the current value. + */ +struct iommu_option { + __u32 size; + __u32 option_id; + __u16 op; + __u16 __reserved; + __u32 object_id; + __aligned_u64 val64; +}; +#define IOMMU_OPTION _IO(IOMMUFD_TYPE, IOMMUFD_CMD_OPTION) + +/** + * enum iommufd_vfio_ioas_op - IOMMU_VFIO_IOAS_* ioctls + * @IOMMU_VFIO_IOAS_GET: Get the current compatibility IOAS + * @IOMMU_VFIO_IOAS_SET: Change the current compatibility IOAS + * @IOMMU_VFIO_IOAS_CLEAR: Disable VFIO compatibility + */ +enum iommufd_vfio_ioas_op { + IOMMU_VFIO_IOAS_GET = 0, + IOMMU_VFIO_IOAS_SET = 1, + IOMMU_VFIO_IOAS_CLEAR = 2, +}; + +/** + * struct iommu_vfio_ioas - ioctl(IOMMU_VFIO_IOAS) + * @size: sizeof(struct iommu_vfio_ioas) + * @ioas_id: For IOMMU_VFIO_IOAS_SET the input IOAS ID to set + * For IOMMU_VFIO_IOAS_GET will output the IOAS ID + * @op: One of enum iommufd_vfio_ioas_op + * @__reserved: Must be 0 + * + * The VFIO compatibility support uses a single ioas because VFIO APIs do not + * support the ID field. Set or Get the IOAS that VFIO compatibility will use. + * When VFIO_GROUP_SET_CONTAINER is used on an iommufd it will get the + * compatibility ioas, either by taking what is already set, or auto creating + * one. From then on VFIO will continue to use that ioas and is not effected by + * this ioctl. SET or CLEAR does not destroy any auto-created IOAS. + */ +struct iommu_vfio_ioas { + __u32 size; + __u32 ioas_id; + __u16 op; + __u16 __reserved; +}; +#define IOMMU_VFIO_IOAS _IO(IOMMUFD_TYPE, IOMMUFD_CMD_VFIO_IOAS) + +/** + * enum iommufd_hwpt_alloc_flags - Flags for HWPT allocation + * @IOMMU_HWPT_ALLOC_NEST_PARENT: If set, allocate a HWPT that can serve as + * the parent HWPT in a nesting configuration. + * @IOMMU_HWPT_ALLOC_DIRTY_TRACKING: Dirty tracking support for device IOMMU is + * enforced on device attachment + * @IOMMU_HWPT_FAULT_ID_VALID: The fault_id field of hwpt allocation data is + * valid. + * @IOMMU_HWPT_ALLOC_PASID: Requests a domain that can be used with PASID. The + * domain can be attached to any PASID on the device. + * Any domain attached to the non-PASID part of the + * device must also be flagged, otherwise attaching a + * PASID will blocked. + * For the user that wants to attach PASID, ioas is + * not recommended for both the non-PASID part + * and PASID part of the device. + * If IOMMU does not support PASID it will return + * error (-EOPNOTSUPP). + */ +enum iommufd_hwpt_alloc_flags { + IOMMU_HWPT_ALLOC_NEST_PARENT = 1 << 0, + IOMMU_HWPT_ALLOC_DIRTY_TRACKING = 1 << 1, + IOMMU_HWPT_FAULT_ID_VALID = 1 << 2, + IOMMU_HWPT_ALLOC_PASID = 1 << 3, +}; + +/** + * enum iommu_hwpt_vtd_s1_flags - Intel VT-d stage-1 page table + * entry attributes + * @IOMMU_VTD_S1_SRE: Supervisor request + * @IOMMU_VTD_S1_EAFE: Extended access enable + * @IOMMU_VTD_S1_WPE: Write protect enable + */ +enum iommu_hwpt_vtd_s1_flags { + IOMMU_VTD_S1_SRE = 1 << 0, + IOMMU_VTD_S1_EAFE = 1 << 1, + IOMMU_VTD_S1_WPE = 1 << 2, +}; + +/** + * struct iommu_hwpt_vtd_s1 - Intel VT-d stage-1 page table + * info (IOMMU_HWPT_DATA_VTD_S1) + * @flags: Combination of enum iommu_hwpt_vtd_s1_flags + * @pgtbl_addr: The base address of the stage-1 page table. + * @addr_width: The address width of the stage-1 page table + * @__reserved: Must be 0 + */ +struct iommu_hwpt_vtd_s1 { + __aligned_u64 flags; + __aligned_u64 pgtbl_addr; + __u32 addr_width; + __u32 __reserved; +}; + +/** + * struct iommu_hwpt_arm_smmuv3 - ARM SMMUv3 nested STE + * (IOMMU_HWPT_DATA_ARM_SMMUV3) + * + * @ste: The first two double words of the user space Stream Table Entry for + * the translation. Must be little-endian. + * Allowed fields: (Refer to "5.2 Stream Table Entry" in SMMUv3 HW Spec) + * - word-0: V, Cfg, S1Fmt, S1ContextPtr, S1CDMax + * - word-1: EATS, S1DSS, S1CIR, S1COR, S1CSH, S1STALLD + * + * -EIO will be returned if @ste is not legal or contains any non-allowed field. + * Cfg can be used to select a S1, Bypass or Abort configuration. A Bypass + * nested domain will translate the same as the nesting parent. The S1 will + * install a Context Descriptor Table pointing at userspace memory translated + * by the nesting parent. + */ +struct iommu_hwpt_arm_smmuv3 { + __aligned_le64 ste[2]; +}; + +/** + * enum iommu_hwpt_data_type - IOMMU HWPT Data Type + * @IOMMU_HWPT_DATA_NONE: no data + * @IOMMU_HWPT_DATA_VTD_S1: Intel VT-d stage-1 page table + * @IOMMU_HWPT_DATA_ARM_SMMUV3: ARM SMMUv3 Context Descriptor Table + */ +enum iommu_hwpt_data_type { + IOMMU_HWPT_DATA_NONE = 0, + IOMMU_HWPT_DATA_VTD_S1 = 1, + IOMMU_HWPT_DATA_ARM_SMMUV3 = 2, +}; + +/** + * struct iommu_hwpt_alloc - ioctl(IOMMU_HWPT_ALLOC) + * @size: sizeof(struct iommu_hwpt_alloc) + * @flags: Combination of enum iommufd_hwpt_alloc_flags + * @dev_id: The device to allocate this HWPT for + * @pt_id: The IOAS or HWPT or vIOMMU to connect this HWPT to + * @out_hwpt_id: The ID of the new HWPT + * @__reserved: Must be 0 + * @data_type: One of enum iommu_hwpt_data_type + * @data_len: Length of the type specific data + * @data_uptr: User pointer to the type specific data + * @fault_id: The ID of IOMMUFD_FAULT object. Valid only if flags field of + * IOMMU_HWPT_FAULT_ID_VALID is set. + * @__reserved2: Padding to 64-bit alignment. Must be 0. + * + * Explicitly allocate a hardware page table object. This is the same object + * type that is returned by iommufd_device_attach() and represents the + * underlying iommu driver's iommu_domain kernel object. + * + * A kernel-managed HWPT will be created with the mappings from the given + * IOAS via the @pt_id. The @data_type for this allocation must be set to + * IOMMU_HWPT_DATA_NONE. The HWPT can be allocated as a parent HWPT for a + * nesting configuration by passing IOMMU_HWPT_ALLOC_NEST_PARENT via @flags. + * + * A user-managed nested HWPT will be created from a given vIOMMU (wrapping a + * parent HWPT) or a parent HWPT via @pt_id, in which the parent HWPT must be + * allocated previously via the same ioctl from a given IOAS (@pt_id). In this + * case, the @data_type must be set to a pre-defined type corresponding to an + * I/O page table type supported by the underlying IOMMU hardware. The device + * via @dev_id and the vIOMMU via @pt_id must be associated to the same IOMMU + * instance. + * + * If the @data_type is set to IOMMU_HWPT_DATA_NONE, @data_len and + * @data_uptr should be zero. Otherwise, both @data_len and @data_uptr + * must be given. + */ +struct iommu_hwpt_alloc { + __u32 size; + __u32 flags; + __u32 dev_id; + __u32 pt_id; + __u32 out_hwpt_id; + __u32 __reserved; + __u32 data_type; + __u32 data_len; + __aligned_u64 data_uptr; + __u32 fault_id; + __u32 __reserved2; +}; +#define IOMMU_HWPT_ALLOC _IO(IOMMUFD_TYPE, IOMMUFD_CMD_HWPT_ALLOC) + +/** + * enum iommu_hw_info_vtd_flags - Flags for VT-d hw_info + * @IOMMU_HW_INFO_VTD_ERRATA_772415_SPR17: If set, disallow read-only mappings + * on a nested_parent domain. + * https://www.intel.com/content/www/us/en/content-details/772415/content-details.html + */ +enum iommu_hw_info_vtd_flags { + IOMMU_HW_INFO_VTD_ERRATA_772415_SPR17 = 1 << 0, +}; + +/** + * struct iommu_hw_info_vtd - Intel VT-d hardware information + * + * @flags: Combination of enum iommu_hw_info_vtd_flags + * @__reserved: Must be 0 + * + * @cap_reg: Value of Intel VT-d capability register defined in VT-d spec + * section 11.4.2 Capability Register. + * @ecap_reg: Value of Intel VT-d capability register defined in VT-d spec + * section 11.4.3 Extended Capability Register. + * + * User needs to understand the Intel VT-d specification to decode the + * register value. + */ +struct iommu_hw_info_vtd { + __u32 flags; + __u32 __reserved; + __aligned_u64 cap_reg; + __aligned_u64 ecap_reg; +}; + +/** + * struct iommu_hw_info_arm_smmuv3 - ARM SMMUv3 hardware information + * (IOMMU_HW_INFO_TYPE_ARM_SMMUV3) + * + * @flags: Must be set to 0 + * @__reserved: Must be 0 + * @idr: Implemented features for ARM SMMU Non-secure programming interface + * @iidr: Information about the implementation and implementer of ARM SMMU, + * and architecture version supported + * @aidr: ARM SMMU architecture version + * + * For the details of @idr, @iidr and @aidr, please refer to the chapters + * from 6.3.1 to 6.3.6 in the SMMUv3 Spec. + * + * This reports the raw HW capability, and not all bits are meaningful to be + * read by userspace. Only the following fields should be used: + * + * idr[0]: ST_LEVEL, TERM_MODEL, STALL_MODEL, TTENDIAN , CD2L, ASID16, TTF + * idr[1]: SIDSIZE, SSIDSIZE + * idr[3]: BBML, RIL + * idr[5]: VAX, GRAN64K, GRAN16K, GRAN4K + * + * - S1P should be assumed to be true if a NESTED HWPT can be created + * - VFIO/iommufd only support platforms with COHACC, it should be assumed to be + * true. + * - ATS is a per-device property. If the VMM describes any devices as ATS + * capable in ACPI/DT it should set the corresponding idr. + * + * This list may expand in future (eg E0PD, AIE, PBHA, D128, DS etc). It is + * important that VMMs do not read bits outside the list to allow for + * compatibility with future kernels. Several features in the SMMUv3 + * architecture are not currently supported by the kernel for nesting: HTTU, + * BTM, MPAM and others. + */ +struct iommu_hw_info_arm_smmuv3 { + __u32 flags; + __u32 __reserved; + __u32 idr[6]; + __u32 iidr; + __u32 aidr; +}; + +/** + * struct iommu_hw_info_tegra241_cmdqv - NVIDIA Tegra241 CMDQV Hardware + * Information (IOMMU_HW_INFO_TYPE_TEGRA241_CMDQV) + * + * @flags: Must be 0 + * @version: Version number for the CMDQ-V HW for PARAM bits[03:00] + * @log2vcmdqs: Log2 of the total number of VCMDQs for PARAM bits[07:04] + * @log2vsids: Log2 of the total number of SID replacements for PARAM bits[15:12] + * @__reserved: Must be 0 + * + * VMM can use these fields directly in its emulated global PARAM register. Note + * that only one Virtual Interface (VINTF) should be exposed to a VM, i.e. PARAM + * bits[11:08] should be set to 0 for log2 of the total number of VINTFs. + */ +struct iommu_hw_info_tegra241_cmdqv { + __u32 flags; + __u8 version; + __u8 log2vcmdqs; + __u8 log2vsids; + __u8 __reserved; +}; + +/** + * enum iommu_hw_info_type - IOMMU Hardware Info Types + * @IOMMU_HW_INFO_TYPE_NONE: Output by the drivers that do not report hardware + * info + * @IOMMU_HW_INFO_TYPE_DEFAULT: Input to request for a default type + * @IOMMU_HW_INFO_TYPE_INTEL_VTD: Intel VT-d iommu info type + * @IOMMU_HW_INFO_TYPE_ARM_SMMUV3: ARM SMMUv3 iommu info type + * @IOMMU_HW_INFO_TYPE_TEGRA241_CMDQV: NVIDIA Tegra241 CMDQV (extension for ARM + * SMMUv3) info type + */ +enum iommu_hw_info_type { + IOMMU_HW_INFO_TYPE_NONE = 0, + IOMMU_HW_INFO_TYPE_DEFAULT = 0, + IOMMU_HW_INFO_TYPE_INTEL_VTD = 1, + IOMMU_HW_INFO_TYPE_ARM_SMMUV3 = 2, + IOMMU_HW_INFO_TYPE_TEGRA241_CMDQV = 3, +}; + +/** + * enum iommufd_hw_capabilities + * @IOMMU_HW_CAP_DIRTY_TRACKING: IOMMU hardware support for dirty tracking + * If available, it means the following APIs + * are supported: + * + * IOMMU_HWPT_GET_DIRTY_BITMAP + * IOMMU_HWPT_SET_DIRTY_TRACKING + * + * @IOMMU_HW_CAP_PCI_PASID_EXEC: Execute Permission Supported, user ignores it + * when the struct + * iommu_hw_info::out_max_pasid_log2 is zero. + * @IOMMU_HW_CAP_PCI_PASID_PRIV: Privileged Mode Supported, user ignores it + * when the struct + * iommu_hw_info::out_max_pasid_log2 is zero. + */ +enum iommufd_hw_capabilities { + IOMMU_HW_CAP_DIRTY_TRACKING = 1 << 0, + IOMMU_HW_CAP_PCI_PASID_EXEC = 1 << 1, + IOMMU_HW_CAP_PCI_PASID_PRIV = 1 << 2, +}; + +/** + * enum iommufd_hw_info_flags - Flags for iommu_hw_info + * @IOMMU_HW_INFO_FLAG_INPUT_TYPE: If set, @in_data_type carries an input type + * for user space to request for a specific info + */ +enum iommufd_hw_info_flags { + IOMMU_HW_INFO_FLAG_INPUT_TYPE = 1 << 0, +}; + +/** + * struct iommu_hw_info - ioctl(IOMMU_GET_HW_INFO) + * @size: sizeof(struct iommu_hw_info) + * @flags: Must be 0 + * @dev_id: The device bound to the iommufd + * @data_len: Input the length of a user buffer in bytes. Output the length of + * data that kernel supports + * @data_uptr: User pointer to a user-space buffer used by the kernel to fill + * the iommu type specific hardware information data + * @in_data_type: This shares the same field with @out_data_type, making it be + * a bidirectional field. When IOMMU_HW_INFO_FLAG_INPUT_TYPE is + * set, an input type carried via this @in_data_type field will + * be valid, requesting for the info data to the given type. If + * IOMMU_HW_INFO_FLAG_INPUT_TYPE is unset, any input value will + * be seen as IOMMU_HW_INFO_TYPE_DEFAULT + * @out_data_type: Output the iommu hardware info type as defined in the enum + * iommu_hw_info_type. + * @out_capabilities: Output the generic iommu capability info type as defined + * in the enum iommu_hw_capabilities. + * @out_max_pasid_log2: Output the width of PASIDs. 0 means no PASID support. + * PCI devices turn to out_capabilities to check if the + * specific capabilities is supported or not. + * @__reserved: Must be 0 + * + * Query an iommu type specific hardware information data from an iommu behind + * a given device that has been bound to iommufd. This hardware info data will + * be used to sync capabilities between the virtual iommu and the physical + * iommu, e.g. a nested translation setup needs to check the hardware info, so + * a guest stage-1 page table can be compatible with the physical iommu. + * + * To capture an iommu type specific hardware information data, @data_uptr and + * its length @data_len must be provided. Trailing bytes will be zeroed if the + * user buffer is larger than the data that kernel has. Otherwise, kernel only + * fills the buffer using the given length in @data_len. If the ioctl succeeds, + * @data_len will be updated to the length that kernel actually supports, + * @out_data_type will be filled to decode the data filled in the buffer + * pointed by @data_uptr. Input @data_len == zero is allowed. + */ +struct iommu_hw_info { + __u32 size; + __u32 flags; + __u32 dev_id; + __u32 data_len; + __aligned_u64 data_uptr; + union { + __u32 in_data_type; + __u32 out_data_type; + }; + __u8 out_max_pasid_log2; + __u8 __reserved[3]; + __aligned_u64 out_capabilities; +}; +#define IOMMU_GET_HW_INFO _IO(IOMMUFD_TYPE, IOMMUFD_CMD_GET_HW_INFO) + +/* + * enum iommufd_hwpt_set_dirty_tracking_flags - Flags for steering dirty + * tracking + * @IOMMU_HWPT_DIRTY_TRACKING_ENABLE: Enable dirty tracking + */ +enum iommufd_hwpt_set_dirty_tracking_flags { + IOMMU_HWPT_DIRTY_TRACKING_ENABLE = 1, +}; + +/** + * struct iommu_hwpt_set_dirty_tracking - ioctl(IOMMU_HWPT_SET_DIRTY_TRACKING) + * @size: sizeof(struct iommu_hwpt_set_dirty_tracking) + * @flags: Combination of enum iommufd_hwpt_set_dirty_tracking_flags + * @hwpt_id: HW pagetable ID that represents the IOMMU domain + * @__reserved: Must be 0 + * + * Toggle dirty tracking on an HW pagetable. + */ +struct iommu_hwpt_set_dirty_tracking { + __u32 size; + __u32 flags; + __u32 hwpt_id; + __u32 __reserved; +}; +#define IOMMU_HWPT_SET_DIRTY_TRACKING _IO(IOMMUFD_TYPE, \ + IOMMUFD_CMD_HWPT_SET_DIRTY_TRACKING) + +/** + * enum iommufd_hwpt_get_dirty_bitmap_flags - Flags for getting dirty bits + * @IOMMU_HWPT_GET_DIRTY_BITMAP_NO_CLEAR: Just read the PTEs without clearing + * any dirty bits metadata. This flag + * can be passed in the expectation + * where the next operation is an unmap + * of the same IOVA range. + * + */ +enum iommufd_hwpt_get_dirty_bitmap_flags { + IOMMU_HWPT_GET_DIRTY_BITMAP_NO_CLEAR = 1, +}; + +/** + * struct iommu_hwpt_get_dirty_bitmap - ioctl(IOMMU_HWPT_GET_DIRTY_BITMAP) + * @size: sizeof(struct iommu_hwpt_get_dirty_bitmap) + * @hwpt_id: HW pagetable ID that represents the IOMMU domain + * @flags: Combination of enum iommufd_hwpt_get_dirty_bitmap_flags + * @__reserved: Must be 0 + * @iova: base IOVA of the bitmap first bit + * @length: IOVA range size + * @page_size: page size granularity of each bit in the bitmap + * @data: bitmap where to set the dirty bits. The bitmap bits each + * represent a page_size which you deviate from an arbitrary iova. + * + * Checking a given IOVA is dirty: + * + * data[(iova / page_size) / 64] & (1ULL << ((iova / page_size) % 64)) + * + * Walk the IOMMU pagetables for a given IOVA range to return a bitmap + * with the dirty IOVAs. In doing so it will also by default clear any + * dirty bit metadata set in the IOPTE. + */ +struct iommu_hwpt_get_dirty_bitmap { + __u32 size; + __u32 hwpt_id; + __u32 flags; + __u32 __reserved; + __aligned_u64 iova; + __aligned_u64 length; + __aligned_u64 page_size; + __aligned_u64 data; +}; +#define IOMMU_HWPT_GET_DIRTY_BITMAP _IO(IOMMUFD_TYPE, \ + IOMMUFD_CMD_HWPT_GET_DIRTY_BITMAP) + +/** + * enum iommu_hwpt_invalidate_data_type - IOMMU HWPT Cache Invalidation + * Data Type + * @IOMMU_HWPT_INVALIDATE_DATA_VTD_S1: Invalidation data for VTD_S1 + * @IOMMU_VIOMMU_INVALIDATE_DATA_ARM_SMMUV3: Invalidation data for ARM SMMUv3 + */ +enum iommu_hwpt_invalidate_data_type { + IOMMU_HWPT_INVALIDATE_DATA_VTD_S1 = 0, + IOMMU_VIOMMU_INVALIDATE_DATA_ARM_SMMUV3 = 1, +}; + +/** + * enum iommu_hwpt_vtd_s1_invalidate_flags - Flags for Intel VT-d + * stage-1 cache invalidation + * @IOMMU_VTD_INV_FLAGS_LEAF: Indicates whether the invalidation applies + * to all-levels page structure cache or just + * the leaf PTE cache. + */ +enum iommu_hwpt_vtd_s1_invalidate_flags { + IOMMU_VTD_INV_FLAGS_LEAF = 1 << 0, +}; + +/** + * struct iommu_hwpt_vtd_s1_invalidate - Intel VT-d cache invalidation + * (IOMMU_HWPT_INVALIDATE_DATA_VTD_S1) + * @addr: The start address of the range to be invalidated. It needs to + * be 4KB aligned. + * @npages: Number of contiguous 4K pages to be invalidated. + * @flags: Combination of enum iommu_hwpt_vtd_s1_invalidate_flags + * @__reserved: Must be 0 + * + * The Intel VT-d specific invalidation data for user-managed stage-1 cache + * invalidation in nested translation. Userspace uses this structure to + * tell the impacted cache scope after modifying the stage-1 page table. + * + * Invalidating all the caches related to the page table by setting @addr + * to be 0 and @npages to be U64_MAX. + * + * The device TLB will be invalidated automatically if ATS is enabled. + */ +struct iommu_hwpt_vtd_s1_invalidate { + __aligned_u64 addr; + __aligned_u64 npages; + __u32 flags; + __u32 __reserved; +}; + +/** + * struct iommu_viommu_arm_smmuv3_invalidate - ARM SMMUv3 cache invalidation + * (IOMMU_VIOMMU_INVALIDATE_DATA_ARM_SMMUV3) + * @cmd: 128-bit cache invalidation command that runs in SMMU CMDQ. + * Must be little-endian. + * + * Supported command list only when passing in a vIOMMU via @hwpt_id: + * CMDQ_OP_TLBI_NSNH_ALL + * CMDQ_OP_TLBI_NH_VA + * CMDQ_OP_TLBI_NH_VAA + * CMDQ_OP_TLBI_NH_ALL + * CMDQ_OP_TLBI_NH_ASID + * CMDQ_OP_ATC_INV + * CMDQ_OP_CFGI_CD + * CMDQ_OP_CFGI_CD_ALL + * + * -EIO will be returned if the command is not supported. + */ +struct iommu_viommu_arm_smmuv3_invalidate { + __aligned_le64 cmd[2]; +}; + +/** + * struct iommu_hwpt_invalidate - ioctl(IOMMU_HWPT_INVALIDATE) + * @size: sizeof(struct iommu_hwpt_invalidate) + * @hwpt_id: ID of a nested HWPT or a vIOMMU, for cache invalidation + * @data_uptr: User pointer to an array of driver-specific cache invalidation + * data. + * @data_type: One of enum iommu_hwpt_invalidate_data_type, defining the data + * type of all the entries in the invalidation request array. It + * should be a type supported by the hwpt pointed by @hwpt_id. + * @entry_len: Length (in bytes) of a request entry in the request array + * @entry_num: Input the number of cache invalidation requests in the array. + * Output the number of requests successfully handled by kernel. + * @__reserved: Must be 0. + * + * Invalidate iommu cache for user-managed page table or vIOMMU. Modifications + * on a user-managed page table should be followed by this operation, if a HWPT + * is passed in via @hwpt_id. Other caches, such as device cache or descriptor + * cache can be flushed if a vIOMMU is passed in via the @hwpt_id field. + * + * Each ioctl can support one or more cache invalidation requests in the array + * that has a total size of @entry_len * @entry_num. + * + * An empty invalidation request array by setting @entry_num==0 is allowed, and + * @entry_len and @data_uptr would be ignored in this case. This can be used to + * check if the given @data_type is supported or not by kernel. + */ +struct iommu_hwpt_invalidate { + __u32 size; + __u32 hwpt_id; + __aligned_u64 data_uptr; + __u32 data_type; + __u32 entry_len; + __u32 entry_num; + __u32 __reserved; +}; +#define IOMMU_HWPT_INVALIDATE _IO(IOMMUFD_TYPE, IOMMUFD_CMD_HWPT_INVALIDATE) + +/** + * enum iommu_hwpt_pgfault_flags - flags for struct iommu_hwpt_pgfault + * @IOMMU_PGFAULT_FLAGS_PASID_VALID: The pasid field of the fault data is + * valid. + * @IOMMU_PGFAULT_FLAGS_LAST_PAGE: It's the last fault of a fault group. + */ +enum iommu_hwpt_pgfault_flags { + IOMMU_PGFAULT_FLAGS_PASID_VALID = (1 << 0), + IOMMU_PGFAULT_FLAGS_LAST_PAGE = (1 << 1), +}; + +/** + * enum iommu_hwpt_pgfault_perm - perm bits for struct iommu_hwpt_pgfault + * @IOMMU_PGFAULT_PERM_READ: request for read permission + * @IOMMU_PGFAULT_PERM_WRITE: request for write permission + * @IOMMU_PGFAULT_PERM_EXEC: (PCIE 10.4.1) request with a PASID that has the + * Execute Requested bit set in PASID TLP Prefix. + * @IOMMU_PGFAULT_PERM_PRIV: (PCIE 10.4.1) request with a PASID that has the + * Privileged Mode Requested bit set in PASID TLP + * Prefix. + */ +enum iommu_hwpt_pgfault_perm { + IOMMU_PGFAULT_PERM_READ = (1 << 0), + IOMMU_PGFAULT_PERM_WRITE = (1 << 1), + IOMMU_PGFAULT_PERM_EXEC = (1 << 2), + IOMMU_PGFAULT_PERM_PRIV = (1 << 3), +}; + +/** + * struct iommu_hwpt_pgfault - iommu page fault data + * @flags: Combination of enum iommu_hwpt_pgfault_flags + * @dev_id: id of the originated device + * @pasid: Process Address Space ID + * @grpid: Page Request Group Index + * @perm: Combination of enum iommu_hwpt_pgfault_perm + * @__reserved: Must be 0. + * @addr: Fault address + * @length: a hint of how much data the requestor is expecting to fetch. For + * example, if the PRI initiator knows it is going to do a 10MB + * transfer, it could fill in 10MB and the OS could pre-fault in + * 10MB of IOVA. It's default to 0 if there's no such hint. + * @cookie: kernel-managed cookie identifying a group of fault messages. The + * cookie number encoded in the last page fault of the group should + * be echoed back in the response message. + */ +struct iommu_hwpt_pgfault { + __u32 flags; + __u32 dev_id; + __u32 pasid; + __u32 grpid; + __u32 perm; + __u32 __reserved; + __aligned_u64 addr; + __u32 length; + __u32 cookie; +}; + +/** + * enum iommufd_page_response_code - Return status of fault handlers + * @IOMMUFD_PAGE_RESP_SUCCESS: Fault has been handled and the page tables + * populated, retry the access. This is the + * "Success" defined in PCI 10.4.2.1. + * @IOMMUFD_PAGE_RESP_INVALID: Could not handle this fault, don't retry the + * access. This is the "Invalid Request" in PCI + * 10.4.2.1. + */ +enum iommufd_page_response_code { + IOMMUFD_PAGE_RESP_SUCCESS = 0, + IOMMUFD_PAGE_RESP_INVALID = 1, +}; + +/** + * struct iommu_hwpt_page_response - IOMMU page fault response + * @cookie: The kernel-managed cookie reported in the fault message. + * @code: One of response code in enum iommufd_page_response_code. + */ +struct iommu_hwpt_page_response { + __u32 cookie; + __u32 code; +}; + +/** + * struct iommu_fault_alloc - ioctl(IOMMU_FAULT_QUEUE_ALLOC) + * @size: sizeof(struct iommu_fault_alloc) + * @flags: Must be 0 + * @out_fault_id: The ID of the new FAULT + * @out_fault_fd: The fd of the new FAULT + * + * Explicitly allocate a fault handling object. + */ +struct iommu_fault_alloc { + __u32 size; + __u32 flags; + __u32 out_fault_id; + __u32 out_fault_fd; +}; +#define IOMMU_FAULT_QUEUE_ALLOC _IO(IOMMUFD_TYPE, IOMMUFD_CMD_FAULT_QUEUE_ALLOC) + +/** + * enum iommu_viommu_type - Virtual IOMMU Type + * @IOMMU_VIOMMU_TYPE_DEFAULT: Reserved for future use + * @IOMMU_VIOMMU_TYPE_ARM_SMMUV3: ARM SMMUv3 driver specific type + * @IOMMU_VIOMMU_TYPE_TEGRA241_CMDQV: NVIDIA Tegra241 CMDQV (extension for ARM + * SMMUv3) enabled ARM SMMUv3 type + */ +enum iommu_viommu_type { + IOMMU_VIOMMU_TYPE_DEFAULT = 0, + IOMMU_VIOMMU_TYPE_ARM_SMMUV3 = 1, + IOMMU_VIOMMU_TYPE_TEGRA241_CMDQV = 2, +}; + +/** + * struct iommu_viommu_tegra241_cmdqv - NVIDIA Tegra241 CMDQV Virtual Interface + * (IOMMU_VIOMMU_TYPE_TEGRA241_CMDQV) + * @out_vintf_mmap_offset: mmap offset argument for VINTF's page0 + * @out_vintf_mmap_length: mmap length argument for VINTF's page0 + * + * Both @out_vintf_mmap_offset and @out_vintf_mmap_length are reported by kernel + * for user space to mmap the VINTF page0 from the host physical address space + * to the guest physical address space so that a guest kernel can directly R/W + * access to the VINTF page0 in order to control its virtual command queues. + */ +struct iommu_viommu_tegra241_cmdqv { + __aligned_u64 out_vintf_mmap_offset; + __aligned_u64 out_vintf_mmap_length; +}; + +/** + * struct iommu_viommu_alloc - ioctl(IOMMU_VIOMMU_ALLOC) + * @size: sizeof(struct iommu_viommu_alloc) + * @flags: Must be 0 + * @type: Type of the virtual IOMMU. Must be defined in enum iommu_viommu_type + * @dev_id: The device's physical IOMMU will be used to back the virtual IOMMU + * @hwpt_id: ID of a nesting parent HWPT to associate to + * @out_viommu_id: Output virtual IOMMU ID for the allocated object + * @data_len: Length of the type specific data + * @__reserved: Must be 0 + * @data_uptr: User pointer to a driver-specific virtual IOMMU data + * + * Allocate a virtual IOMMU object, representing the underlying physical IOMMU's + * virtualization support that is a security-isolated slice of the real IOMMU HW + * that is unique to a specific VM. Operations global to the IOMMU are connected + * to the vIOMMU, such as: + * - Security namespace for guest owned ID, e.g. guest-controlled cache tags + * - Non-device-affiliated event reporting, e.g. invalidation queue errors + * - Access to a sharable nesting parent pagetable across physical IOMMUs + * - Virtualization of various platforms IDs, e.g. RIDs and others + * - Delivery of paravirtualized invalidation + * - Direct assigned invalidation queues + * - Direct assigned interrupts + */ +struct iommu_viommu_alloc { + __u32 size; + __u32 flags; + __u32 type; + __u32 dev_id; + __u32 hwpt_id; + __u32 out_viommu_id; + __u32 data_len; + __u32 __reserved; + __aligned_u64 data_uptr; +}; +#define IOMMU_VIOMMU_ALLOC _IO(IOMMUFD_TYPE, IOMMUFD_CMD_VIOMMU_ALLOC) + +/** + * struct iommu_vdevice_alloc - ioctl(IOMMU_VDEVICE_ALLOC) + * @size: sizeof(struct iommu_vdevice_alloc) + * @viommu_id: vIOMMU ID to associate with the virtual device + * @dev_id: The physical device to allocate a virtual instance on the vIOMMU + * @out_vdevice_id: Object handle for the vDevice. Pass to IOMMU_DESTORY + * @virt_id: Virtual device ID per vIOMMU, e.g. vSID of ARM SMMUv3, vDeviceID + * of AMD IOMMU, and vRID of Intel VT-d + * + * Allocate a virtual device instance (for a physical device) against a vIOMMU. + * This instance holds the device's information (related to its vIOMMU) in a VM. + * User should use IOMMU_DESTROY to destroy the virtual device before + * destroying the physical device (by closing vfio_cdev fd). Otherwise the + * virtual device would be forcibly destroyed on physical device destruction, + * its vdevice_id would be permanently leaked (unremovable & unreusable) until + * iommu fd closed. + */ +struct iommu_vdevice_alloc { + __u32 size; + __u32 viommu_id; + __u32 dev_id; + __u32 out_vdevice_id; + __aligned_u64 virt_id; +}; +#define IOMMU_VDEVICE_ALLOC _IO(IOMMUFD_TYPE, IOMMUFD_CMD_VDEVICE_ALLOC) + +/** + * struct iommu_ioas_change_process - ioctl(VFIO_IOAS_CHANGE_PROCESS) + * @size: sizeof(struct iommu_ioas_change_process) + * @__reserved: Must be 0 + * + * This transfers pinned memory counts for every memory map in every IOAS + * in the context to the current process. This only supports maps created + * with IOMMU_IOAS_MAP_FILE, and returns EINVAL if other maps are present. + * If the ioctl returns a failure status, then nothing is changed. + * + * This API is useful for transferring operation of a device from one process + * to another, such as during userland live update. + */ +struct iommu_ioas_change_process { + __u32 size; + __u32 __reserved; +}; + +#define IOMMU_IOAS_CHANGE_PROCESS \ + _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_CHANGE_PROCESS) + +/** + * enum iommu_veventq_flag - flag for struct iommufd_vevent_header + * @IOMMU_VEVENTQ_FLAG_LOST_EVENTS: vEVENTQ has lost vEVENTs + */ +enum iommu_veventq_flag { + IOMMU_VEVENTQ_FLAG_LOST_EVENTS = (1U << 0), +}; + +/** + * struct iommufd_vevent_header - Virtual Event Header for a vEVENTQ Status + * @flags: Combination of enum iommu_veventq_flag + * @sequence: The sequence index of a vEVENT in the vEVENTQ, with a range of + * [0, INT_MAX] where the following index of INT_MAX is 0 + * + * Each iommufd_vevent_header reports a sequence index of the following vEVENT: + * + * +----------------------+-------+----------------------+-------+---+-------+ + * | header0 {sequence=0} | data0 | header1 {sequence=1} | data1 |...| dataN | + * +----------------------+-------+----------------------+-------+---+-------+ + * + * And this sequence index is expected to be monotonic to the sequence index of + * the previous vEVENT. If two adjacent sequence indexes has a delta larger than + * 1, it means that delta - 1 number of vEVENTs has lost, e.g. two lost vEVENTs: + * + * +-----+----------------------+-------+----------------------+-------+-----+ + * | ... | header3 {sequence=3} | data3 | header6 {sequence=6} | data6 | ... | + * +-----+----------------------+-------+----------------------+-------+-----+ + * + * If a vEVENT lost at the tail of the vEVENTQ and there is no following vEVENT + * providing the next sequence index, an IOMMU_VEVENTQ_FLAG_LOST_EVENTS header + * would be added to the tail, and no data would follow this header: + * + * +--+----------------------+-------+-----------------------------------------+ + * |..| header3 {sequence=3} | data3 | header4 {flags=LOST_EVENTS, sequence=4} | + * +--+----------------------+-------+-----------------------------------------+ + */ +struct iommufd_vevent_header { + __u32 flags; + __u32 sequence; +}; + +/** + * enum iommu_veventq_type - Virtual Event Queue Type + * @IOMMU_VEVENTQ_TYPE_DEFAULT: Reserved for future use + * @IOMMU_VEVENTQ_TYPE_ARM_SMMUV3: ARM SMMUv3 Virtual Event Queue + * @IOMMU_VEVENTQ_TYPE_TEGRA241_CMDQV: NVIDIA Tegra241 CMDQV Extension IRQ + */ +enum iommu_veventq_type { + IOMMU_VEVENTQ_TYPE_DEFAULT = 0, + IOMMU_VEVENTQ_TYPE_ARM_SMMUV3 = 1, + IOMMU_VEVENTQ_TYPE_TEGRA241_CMDQV = 2, +}; + +/** + * struct iommu_vevent_arm_smmuv3 - ARM SMMUv3 Virtual Event + * (IOMMU_VEVENTQ_TYPE_ARM_SMMUV3) + * @evt: 256-bit ARM SMMUv3 Event record, little-endian. + * Reported event records: (Refer to "7.3 Event records" in SMMUv3 HW Spec) + * - 0x04 C_BAD_STE + * - 0x06 F_STREAM_DISABLED + * - 0x08 C_BAD_SUBSTREAMID + * - 0x0a C_BAD_CD + * - 0x10 F_TRANSLATION + * - 0x11 F_ADDR_SIZE + * - 0x12 F_ACCESS + * - 0x13 F_PERMISSION + * + * StreamID field reports a virtual device ID. To receive a virtual event for a + * device, a vDEVICE must be allocated via IOMMU_VDEVICE_ALLOC. + */ +struct iommu_vevent_arm_smmuv3 { + __aligned_le64 evt[4]; +}; + +/** + * struct iommu_vevent_tegra241_cmdqv - Tegra241 CMDQV IRQ + * (IOMMU_VEVENTQ_TYPE_TEGRA241_CMDQV) + * @lvcmdq_err_map: 128-bit logical vcmdq error map, little-endian. + * (Refer to register LVCMDQ_ERR_MAPs per VINTF ) + * + * The 128-bit register value from HW exclusively reflect the error bits for a + * Virtual Interface represented by a vIOMMU object. Read and report directly. + */ +struct iommu_vevent_tegra241_cmdqv { + __aligned_le64 lvcmdq_err_map[2]; +}; + +/** + * struct iommu_veventq_alloc - ioctl(IOMMU_VEVENTQ_ALLOC) + * @size: sizeof(struct iommu_veventq_alloc) + * @flags: Must be 0 + * @viommu_id: virtual IOMMU ID to associate the vEVENTQ with + * @type: Type of the vEVENTQ. Must be defined in enum iommu_veventq_type + * @veventq_depth: Maximum number of events in the vEVENTQ + * @out_veventq_id: The ID of the new vEVENTQ + * @out_veventq_fd: The fd of the new vEVENTQ. User space must close the + * successfully returned fd after using it + * @__reserved: Must be 0 + * + * Explicitly allocate a virtual event queue interface for a vIOMMU. A vIOMMU + * can have multiple FDs for different types, but is confined to one per @type. + * User space should open the @out_veventq_fd to read vEVENTs out of a vEVENTQ, + * if there are vEVENTs available. A vEVENTQ will lose events due to overflow, + * if the number of the vEVENTs hits @veventq_depth. + * + * Each vEVENT in a vEVENTQ encloses a struct iommufd_vevent_header followed by + * a type-specific data structure, in a normal case: + * + * +-+---------+-------+---------+-------+-----+---------+-------+-+ + * | | header0 | data0 | header1 | data1 | ... | headerN | dataN | | + * +-+---------+-------+---------+-------+-----+---------+-------+-+ + * + * unless a tailing IOMMU_VEVENTQ_FLAG_LOST_EVENTS header is logged (refer to + * struct iommufd_vevent_header). + */ +struct iommu_veventq_alloc { + __u32 size; + __u32 flags; + __u32 viommu_id; + __u32 type; + __u32 veventq_depth; + __u32 out_veventq_id; + __u32 out_veventq_fd; + __u32 __reserved; +}; +#define IOMMU_VEVENTQ_ALLOC _IO(IOMMUFD_TYPE, IOMMUFD_CMD_VEVENTQ_ALLOC) + +/** + * enum iommu_hw_queue_type - HW Queue Type + * @IOMMU_HW_QUEUE_TYPE_DEFAULT: Reserved for future use + * @IOMMU_HW_QUEUE_TYPE_TEGRA241_CMDQV: NVIDIA Tegra241 CMDQV (extension for ARM + * SMMUv3) Virtual Command Queue (VCMDQ) + */ +enum iommu_hw_queue_type { + IOMMU_HW_QUEUE_TYPE_DEFAULT = 0, + /* + * TEGRA241_CMDQV requirements (otherwise, allocation will fail) + * - alloc starts from the lowest @index=0 in ascending order + * - destroy starts from the last allocated @index in descending order + * - @base_addr must be aligned to @length in bytes and mapped in IOAS + * - @length must be a power of 2, with a minimum 32 bytes and a maximum + * 2 ^ idr[1].CMDQS * 16 bytes (use GET_HW_INFO call to read idr[1] + * from struct iommu_hw_info_arm_smmuv3) + * - suggest to back the queue memory with contiguous physical pages or + * a single huge page with alignment of the queue size, and limit the + * emulated vSMMU's IDR1.CMDQS to log2(huge page size / 16 bytes) + */ + IOMMU_HW_QUEUE_TYPE_TEGRA241_CMDQV = 1, +}; + +/** + * struct iommu_hw_queue_alloc - ioctl(IOMMU_HW_QUEUE_ALLOC) + * @size: sizeof(struct iommu_hw_queue_alloc) + * @flags: Must be 0 + * @viommu_id: Virtual IOMMU ID to associate the HW queue with + * @type: One of enum iommu_hw_queue_type + * @index: The logical index to the HW queue per virtual IOMMU for a multi-queue + * model + * @out_hw_queue_id: The ID of the new HW queue + * @nesting_parent_iova: Base address of the queue memory in the guest physical + * address space + * @length: Length of the queue memory + * + * Allocate a HW queue object for a vIOMMU-specific HW-accelerated queue, which + * allows HW to access a guest queue memory described using @nesting_parent_iova + * and @length. + * + * A vIOMMU can allocate multiple queues, but it must use a different @index per + * type to separate each allocation, e.g:: + * + * Type1 HW queue0, Type1 HW queue1, Type2 HW queue0, ... + */ +struct iommu_hw_queue_alloc { + __u32 size; + __u32 flags; + __u32 viommu_id; + __u32 type; + __u32 index; + __u32 out_hw_queue_id; + __aligned_u64 nesting_parent_iova; + __aligned_u64 length; +}; +#define IOMMU_HW_QUEUE_ALLOC _IO(IOMMUFD_TYPE, IOMMUFD_CMD_HW_QUEUE_ALLOC) +#endif diff --git a/kernel/linux/uapi/linux/vduse.h b/kernel/linux/uapi/linux/vduse.h index f46269af34..da6ac89af1 100644 --- a/kernel/linux/uapi/linux/vduse.h +++ b/kernel/linux/uapi/linux/vduse.h @@ -237,7 +237,7 @@ struct vduse_iova_umem { * struct vduse_iova_info - information of one IOVA region * @start: start of the IOVA region * @last: last of the IOVA region - * @capability: capability of the IOVA regsion + * @capability: capability of the IOVA region * @reserved: for future use, needs to be initialized to zero * * Structure used by VDUSE_IOTLB_GET_INFO ioctl to get information of diff --git a/kernel/linux/uapi/linux/vfio.h b/kernel/linux/uapi/linux/vfio.h index 79bf8c0cc5..4d96d1fc12 100644 --- a/kernel/linux/uapi/linux/vfio.h +++ b/kernel/linux/uapi/linux/vfio.h @@ -905,10 +905,12 @@ struct vfio_device_feature { * VFIO_DEVICE_BIND_IOMMUFD - _IOR(VFIO_TYPE, VFIO_BASE + 18, * struct vfio_device_bind_iommufd) * @argsz: User filled size of this data. - * @flags: Must be 0. + * @flags: Must be 0 or a bit flags of VFIO_DEVICE_BIND_* * @iommufd: iommufd to bind. * @out_devid: The device id generated by this bind. devid is a handle for * this device/iommufd bond and can be used in IOMMUFD commands. + * @token_uuid_ptr: Valid if VFIO_DEVICE_BIND_FLAG_TOKEN. Points to a 16 byte + * UUID in the same format as VFIO_DEVICE_FEATURE_PCI_VF_TOKEN. * * Bind a vfio_device to the specified iommufd. * @@ -917,13 +919,21 @@ struct vfio_device_feature { * * Unbind is automatically conducted when device fd is closed. * + * A token is sometimes required to open the device, unless this is known to be + * needed VFIO_DEVICE_BIND_FLAG_TOKEN should not be set and token_uuid_ptr is + * ignored. The only case today is a PF/VF relationship where the VF bind must + * be provided the same token as VFIO_DEVICE_FEATURE_PCI_VF_TOKEN provided to + * the PF. + * * Return: 0 on success, -errno on failure. */ struct vfio_device_bind_iommufd { __u32 argsz; __u32 flags; +#define VFIO_DEVICE_BIND_FLAG_TOKEN (1 << 0) __s32 iommufd; __u32 out_devid; + __aligned_u64 token_uuid_ptr; }; #define VFIO_DEVICE_BIND_IOMMUFD _IO(VFIO_TYPE, VFIO_BASE + 18) diff --git a/kernel/linux/uapi/version b/kernel/linux/uapi/version index 966a998301..d9e789dade 100644 --- a/kernel/linux/uapi/version +++ b/kernel/linux/uapi/version @@ -1 +1 @@ -v6.16 +v6.17 -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* [PATCH v7 02/18] vfio: make all functions internal 2026-02-26 14:17 ` [PATCH v7 00/18] Support VFIO cdev API in DPDK Anatoly Burakov 2026-02-26 14:17 ` [PATCH v7 01/18] uapi: update to v6.17 and add iommufd.h Anatoly Burakov @ 2026-02-26 14:17 ` Anatoly Burakov 2026-02-26 14:17 ` [PATCH v7 03/18] vfio: split get device info from setup Anatoly Burakov ` (16 subsequent siblings) 18 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2026-02-26 14:17 UTC (permalink / raw) To: dev, Bruce Richardson, Dmitry Kozlyuk The VFIO API is an externally exported API because the original intent was to offer DMA mapping facilities to applications. However, practical usage of this API seems to be centered around drivers, so keeping this API exported to applications only creates needless API/ABI stability surface that has no added value. Make the entire VFIO API internal-only and visible only to drivers. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- doc/guides/rel_notes/deprecation.rst | 4 --- lib/eal/freebsd/eal.c | 30 +++++++++--------- lib/eal/include/rte_vfio.h | 47 ++++++++++++++++++++++++---- lib/eal/linux/eal_vfio.c | 32 +++++++++---------- lib/eal/windows/eal.c | 4 +-- 5 files changed, 74 insertions(+), 43 deletions(-) diff --git a/doc/guides/rel_notes/deprecation.rst b/doc/guides/rel_notes/deprecation.rst index 4bfbb7468e..b137ed0445 100644 --- a/doc/guides/rel_notes/deprecation.rst +++ b/doc/guides/rel_notes/deprecation.rst @@ -140,9 +140,5 @@ Deprecation Notices ``drivers/bus/vmbus/rte_bus_vmbus.h`` will become internal to DPDK. Those API functions are used internally by DPDK core and netvsc PMD. -* eal: The entire VFIO API (``rte_vfio_*``) will be made internal only, and will only be available to EAL and drivers. - Group-based API (``rte_vfio_*_group_*``) will be removed and replaced with unified container device assignment API. - This change will be made in 26.11 release. - * vdpa: vDPA driver API will no longer offer ``get_vfio_group_fd`` as part of its internal API. All drivers will be adjusted to use the new unified VFIO container device assignment API. diff --git a/lib/eal/freebsd/eal.c b/lib/eal/freebsd/eal.c index 60f5e676a8..4cc80a2107 100644 --- a/lib/eal/freebsd/eal.c +++ b/lib/eal/freebsd/eal.c @@ -813,7 +813,7 @@ rte_eal_vfio_get_vf_token(__rte_unused rte_uuid_t vf_token) { } -RTE_EXPORT_SYMBOL(rte_vfio_setup_device) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_setup_device) int rte_vfio_setup_device(__rte_unused const char *sysfs_base, __rte_unused const char *dev_addr, __rte_unused int *vfio_dev_fd, @@ -823,7 +823,7 @@ int rte_vfio_setup_device(__rte_unused const char *sysfs_base, return -1; } -RTE_EXPORT_SYMBOL(rte_vfio_release_device) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_release_device) int rte_vfio_release_device(__rte_unused const char *sysfs_base, __rte_unused const char *dev_addr, __rte_unused int fd) @@ -832,33 +832,33 @@ int rte_vfio_release_device(__rte_unused const char *sysfs_base, return -1; } -RTE_EXPORT_SYMBOL(rte_vfio_enable) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_enable) int rte_vfio_enable(__rte_unused const char *modname) { rte_errno = ENOTSUP; return -1; } -RTE_EXPORT_SYMBOL(rte_vfio_is_enabled) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_is_enabled) int rte_vfio_is_enabled(__rte_unused const char *modname) { return 0; } -RTE_EXPORT_SYMBOL(rte_vfio_noiommu_is_enabled) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_noiommu_is_enabled) int rte_vfio_noiommu_is_enabled(void) { return 0; } -RTE_EXPORT_SYMBOL(rte_vfio_clear_group) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_clear_group) int rte_vfio_clear_group(__rte_unused int vfio_group_fd) { rte_errno = ENOTSUP; return -1; } -RTE_EXPORT_SYMBOL(rte_vfio_get_group_num) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_group_num) int rte_vfio_get_group_num(__rte_unused const char *sysfs_base, __rte_unused const char *dev_addr, @@ -868,7 +868,7 @@ rte_vfio_get_group_num(__rte_unused const char *sysfs_base, return -1; } -RTE_EXPORT_SYMBOL(rte_vfio_get_container_fd) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_container_fd) int rte_vfio_get_container_fd(void) { @@ -876,7 +876,7 @@ rte_vfio_get_container_fd(void) return -1; } -RTE_EXPORT_SYMBOL(rte_vfio_get_group_fd) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_group_fd) int rte_vfio_get_group_fd(__rte_unused int iommu_group_num) { @@ -884,7 +884,7 @@ rte_vfio_get_group_fd(__rte_unused int iommu_group_num) return -1; } -RTE_EXPORT_SYMBOL(rte_vfio_container_create) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_create) int rte_vfio_container_create(void) { @@ -892,7 +892,7 @@ rte_vfio_container_create(void) return -1; } -RTE_EXPORT_SYMBOL(rte_vfio_container_destroy) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_destroy) int rte_vfio_container_destroy(__rte_unused int container_fd) { @@ -900,7 +900,7 @@ rte_vfio_container_destroy(__rte_unused int container_fd) return -1; } -RTE_EXPORT_SYMBOL(rte_vfio_container_group_bind) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_group_bind) int rte_vfio_container_group_bind(__rte_unused int container_fd, __rte_unused int iommu_group_num) @@ -909,7 +909,7 @@ rte_vfio_container_group_bind(__rte_unused int container_fd, return -1; } -RTE_EXPORT_SYMBOL(rte_vfio_container_group_unbind) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_group_unbind) int rte_vfio_container_group_unbind(__rte_unused int container_fd, __rte_unused int iommu_group_num) @@ -918,7 +918,7 @@ rte_vfio_container_group_unbind(__rte_unused int container_fd, return -1; } -RTE_EXPORT_SYMBOL(rte_vfio_container_dma_map) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_dma_map) int rte_vfio_container_dma_map(__rte_unused int container_fd, __rte_unused uint64_t vaddr, @@ -929,7 +929,7 @@ rte_vfio_container_dma_map(__rte_unused int container_fd, return -1; } -RTE_EXPORT_SYMBOL(rte_vfio_container_dma_unmap) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_dma_unmap) int rte_vfio_container_dma_unmap(__rte_unused int container_fd, __rte_unused uint64_t vaddr, diff --git a/lib/eal/include/rte_vfio.h b/lib/eal/include/rte_vfio.h index d1e8bce56b..0ddeb08f94 100644 --- a/lib/eal/include/rte_vfio.h +++ b/lib/eal/include/rte_vfio.h @@ -7,7 +7,11 @@ /** * @file - * RTE VFIO. This library provides various VFIO related utility functions. + * @internal + * + * RTE VFIO internal API. + * + * This library provides VFIO related utility functions for use by drivers. */ #include <stdbool.h> @@ -36,6 +40,7 @@ struct vfio_device_info; #define RTE_VFIO_DEFAULT_CONTAINER_FD (-1) /** + * @internal * Setup vfio_cfg for the device identified by its address. * It discovers the configured I/O MMU groups or sets a new one for the device. * If a new groups is assigned, the DMA mapping is performed. @@ -60,10 +65,12 @@ struct vfio_device_info; * <0 on failure. * >1 if the device cannot be managed this way. */ +__rte_internal int rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, int *vfio_dev_fd, struct vfio_device_info *device_info); /** + * @internal * Release a device mapped to a VFIO-managed I/O MMU group. * * This function is only relevant to linux and will return @@ -82,9 +89,11 @@ int rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, * 0 on success. * <0 on failure. */ +__rte_internal int rte_vfio_release_device(const char *sysfs_base, const char *dev_addr, int fd); /** + * @internal * Enable a VFIO-related kmod. * * This function is only relevant to linux and will return @@ -97,9 +106,11 @@ int rte_vfio_release_device(const char *sysfs_base, const char *dev_addr, int fd * 0 on success. * <0 on failure. */ +__rte_internal int rte_vfio_enable(const char *modname); /** + * @internal * Check whether a VFIO-related kmod is enabled. * * This function is only relevant to Linux. @@ -111,9 +122,11 @@ int rte_vfio_enable(const char *modname); * 1 if true. * 0 otherwise. */ +__rte_internal int rte_vfio_is_enabled(const char *modname); /** + * @internal * Whether VFIO NOIOMMU mode is enabled. * * This function is only relevant to Linux. @@ -123,10 +136,12 @@ int rte_vfio_is_enabled(const char *modname); * 0 if false. * <0 for errors. */ +__rte_internal int rte_vfio_noiommu_is_enabled(void); /** - * Remove group fd from internal VFIO group fd array/ + * @internal + * Remove group fd from internal VFIO group fd array. * * This function is only relevant to linux and will return * an error on BSD. @@ -138,11 +153,13 @@ int rte_vfio_noiommu_is_enabled(void); * 0 on success. * <0 on failure. */ +__rte_internal int rte_vfio_clear_group(int vfio_group_fd); /** - * Parse IOMMU group number for a device + * @internal + * Parse IOMMU group number for a device. * * This function is only relevant to linux and will return * an error on BSD. @@ -161,12 +178,14 @@ rte_vfio_clear_group(int vfio_group_fd); * 0 for non-existent group or VFIO * <0 for errors */ +__rte_internal int rte_vfio_get_group_num(const char *sysfs_base, const char *dev_addr, int *iommu_group_num); /** - * Get device information + * @internal + * Get device information. * * This function is only relevant to Linux and will return an error on BSD. * @@ -186,12 +205,13 @@ rte_vfio_get_group_num(const char *sysfs_base, * 0 on success. * <0 on failure. */ -__rte_experimental +__rte_internal int rte_vfio_get_device_info(const char *sysfs_base, const char *dev_addr, int *vfio_dev_fd, struct vfio_device_info *device_info); /** + * @internal * Get the default VFIO container fd * * This function is only relevant to linux and will return @@ -201,11 +221,13 @@ rte_vfio_get_device_info(const char *sysfs_base, const char *dev_addr, * > 0 default container fd * < 0 if VFIO is not enabled or not supported */ +__rte_internal int rte_vfio_get_container_fd(void); /** - * Open VFIO group fd or get an existing one + * @internal + * Open VFIO group fd or get an existing one. * * This function is only relevant to linux and will return * an error on BSD. @@ -217,10 +239,12 @@ rte_vfio_get_container_fd(void); * > 0 group fd * < 0 for errors */ +__rte_internal int rte_vfio_get_group_fd(int iommu_group_num); /** + * @internal * Create a new container for device binding. * * @note Any newly allocated DPDK memory will not be mapped into these @@ -235,10 +259,12 @@ rte_vfio_get_group_fd(int iommu_group_num); * the container fd if successful * <0 if failed */ +__rte_internal int rte_vfio_container_create(void); /** + * @internal * Destroy the container, unbind all vfio groups within it. * * @param container_fd @@ -248,10 +274,12 @@ rte_vfio_container_create(void); * 0 if successful * <0 if failed */ +__rte_internal int rte_vfio_container_destroy(int container_fd); /** + * @internal * Bind a IOMMU group to a container. * * @param container_fd @@ -264,10 +292,12 @@ rte_vfio_container_destroy(int container_fd); * group fd if successful * <0 if failed */ +__rte_internal int rte_vfio_container_group_bind(int container_fd, int iommu_group_num); /** + * @internal * Unbind a IOMMU group from a container. * * @param container_fd @@ -280,10 +310,12 @@ rte_vfio_container_group_bind(int container_fd, int iommu_group_num); * 0 if successful * <0 if failed */ +__rte_internal int rte_vfio_container_group_unbind(int container_fd, int iommu_group_num); /** + * @internal * Perform DMA mapping for devices in a container. * * @param container_fd @@ -303,11 +335,13 @@ rte_vfio_container_group_unbind(int container_fd, int iommu_group_num); * 0 if successful * <0 if failed */ +__rte_internal int rte_vfio_container_dma_map(int container_fd, uint64_t vaddr, uint64_t iova, uint64_t len); /** + * @internal * Perform DMA unmapping for devices in a container. * * @param container_fd @@ -327,6 +361,7 @@ rte_vfio_container_dma_map(int container_fd, uint64_t vaddr, * 0 if successful * <0 if failed */ +__rte_internal int rte_vfio_container_dma_unmap(int container_fd, uint64_t vaddr, uint64_t iova, uint64_t len); diff --git a/lib/eal/linux/eal_vfio.c b/lib/eal/linux/eal_vfio.c index f1050ffa60..33fa04feaf 100644 --- a/lib/eal/linux/eal_vfio.c +++ b/lib/eal/linux/eal_vfio.c @@ -532,7 +532,7 @@ get_vfio_cfg_by_container_fd(int container_fd) return NULL; } -RTE_EXPORT_SYMBOL(rte_vfio_get_group_fd) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_group_fd) int rte_vfio_get_group_fd(int iommu_group_num) { @@ -731,7 +731,7 @@ vfio_sync_default_container(void) return -1; } -RTE_EXPORT_SYMBOL(rte_vfio_clear_group) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_clear_group) int rte_vfio_clear_group(int vfio_group_fd) { @@ -755,7 +755,7 @@ rte_vfio_clear_group(int vfio_group_fd) return 0; } -RTE_EXPORT_SYMBOL(rte_vfio_setup_device) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_setup_device) int rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, int *vfio_dev_fd, struct vfio_device_info *device_info) @@ -1009,7 +1009,7 @@ rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, return 0; } -RTE_EXPORT_SYMBOL(rte_vfio_release_device) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_release_device) int rte_vfio_release_device(const char *sysfs_base, const char *dev_addr, int vfio_dev_fd) @@ -1098,7 +1098,7 @@ rte_vfio_release_device(const char *sysfs_base, const char *dev_addr, return ret; } -RTE_EXPORT_SYMBOL(rte_vfio_enable) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_enable) int rte_vfio_enable(const char *modname) { @@ -1175,7 +1175,7 @@ rte_vfio_enable(const char *modname) return 0; } -RTE_EXPORT_SYMBOL(rte_vfio_is_enabled) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_is_enabled) int rte_vfio_is_enabled(const char *modname) { @@ -1215,7 +1215,7 @@ vfio_set_iommu_type(int vfio_container_fd) return NULL; } -RTE_EXPORT_EXPERIMENTAL_SYMBOL(rte_vfio_get_device_info, 24.03) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_device_info) int rte_vfio_get_device_info(const char *sysfs_base, const char *dev_addr, int *vfio_dev_fd, struct vfio_device_info *device_info) @@ -1349,7 +1349,7 @@ vfio_open_container_fd(bool mp_request) return -1; } -RTE_EXPORT_SYMBOL(rte_vfio_get_container_fd) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_container_fd) int rte_vfio_get_container_fd(void) { @@ -1363,7 +1363,7 @@ rte_vfio_get_container_fd(void) return default_vfio_cfg->vfio_container_fd; } -RTE_EXPORT_SYMBOL(rte_vfio_get_group_num) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_group_num) int rte_vfio_get_group_num(const char *sysfs_base, const char *dev_addr, int *iommu_group_num) @@ -2034,7 +2034,7 @@ container_dma_unmap(struct vfio_config *vfio_cfg, uint64_t vaddr, uint64_t iova, return ret; } -RTE_EXPORT_SYMBOL(rte_vfio_noiommu_is_enabled) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_noiommu_is_enabled) int rte_vfio_noiommu_is_enabled(void) { @@ -2067,7 +2067,7 @@ rte_vfio_noiommu_is_enabled(void) return c == 'Y'; } -RTE_EXPORT_SYMBOL(rte_vfio_container_create) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_create) int rte_vfio_container_create(void) { @@ -2094,7 +2094,7 @@ rte_vfio_container_create(void) return vfio_cfgs[i].vfio_container_fd; } -RTE_EXPORT_SYMBOL(rte_vfio_container_destroy) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_destroy) int rte_vfio_container_destroy(int container_fd) { @@ -2120,7 +2120,7 @@ rte_vfio_container_destroy(int container_fd) return 0; } -RTE_EXPORT_SYMBOL(rte_vfio_container_group_bind) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_group_bind) int rte_vfio_container_group_bind(int container_fd, int iommu_group_num) { @@ -2135,7 +2135,7 @@ rte_vfio_container_group_bind(int container_fd, int iommu_group_num) return vfio_get_group_fd(vfio_cfg, iommu_group_num); } -RTE_EXPORT_SYMBOL(rte_vfio_container_group_unbind) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_group_unbind) int rte_vfio_container_group_unbind(int container_fd, int iommu_group_num) { @@ -2176,7 +2176,7 @@ rte_vfio_container_group_unbind(int container_fd, int iommu_group_num) return 0; } -RTE_EXPORT_SYMBOL(rte_vfio_container_dma_map) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_dma_map) int rte_vfio_container_dma_map(int container_fd, uint64_t vaddr, uint64_t iova, uint64_t len) @@ -2197,7 +2197,7 @@ rte_vfio_container_dma_map(int container_fd, uint64_t vaddr, uint64_t iova, return container_dma_map(vfio_cfg, vaddr, iova, len); } -RTE_EXPORT_SYMBOL(rte_vfio_container_dma_unmap) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_dma_unmap) int rte_vfio_container_dma_unmap(int container_fd, uint64_t vaddr, uint64_t iova, uint64_t len) diff --git a/lib/eal/windows/eal.c b/lib/eal/windows/eal.c index f06375a624..d2179a9569 100644 --- a/lib/eal/windows/eal.c +++ b/lib/eal/windows/eal.c @@ -447,7 +447,7 @@ eal_asprintf(char **buffer, const char *format, ...) return ret; } -RTE_EXPORT_SYMBOL(rte_vfio_container_dma_map) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_dma_map) int rte_vfio_container_dma_map(__rte_unused int container_fd, __rte_unused uint64_t vaddr, @@ -458,7 +458,7 @@ rte_vfio_container_dma_map(__rte_unused int container_fd, return -1; } -RTE_EXPORT_SYMBOL(rte_vfio_container_dma_unmap) +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_dma_unmap) int rte_vfio_container_dma_unmap(__rte_unused int container_fd, __rte_unused uint64_t vaddr, -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* [PATCH v7 03/18] vfio: split get device info from setup 2026-02-26 14:17 ` [PATCH v7 00/18] Support VFIO cdev API in DPDK Anatoly Burakov 2026-02-26 14:17 ` [PATCH v7 01/18] uapi: update to v6.17 and add iommufd.h Anatoly Burakov 2026-02-26 14:17 ` [PATCH v7 02/18] vfio: make all functions internal Anatoly Burakov @ 2026-02-26 14:17 ` Anatoly Burakov 2026-02-26 14:17 ` [PATCH v7 04/18] vfio: add container device assignment API Anatoly Burakov ` (15 subsequent siblings) 18 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2026-02-26 14:17 UTC (permalink / raw) To: dev, Nipun Gupta, Nikhil Agarwal, Chenbo Xia, Tomasz Duszynski, Ajit Khaparde, Vikas Gupta, Bruce Richardson Currently, setup gets device info as part of setup, while the separate get device info API also calls setup if the fd is zero. Untangle these two APIs and make each do one thing, and adjust all existing callers. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- drivers/bus/cdx/cdx_vfio.c | 12 ++++++++-- drivers/bus/pci/linux/pci_vfio.c | 18 ++++++++++---- drivers/bus/platform/platform.c | 9 ++++++- drivers/crypto/bcmfs/bcmfs_vfio.c | 8 ++++++- lib/eal/freebsd/eal.c | 3 +-- lib/eal/include/rte_vfio.h | 23 +++++++----------- lib/eal/linux/eal_vfio.c | 40 +++++++++---------------------- 7 files changed, 59 insertions(+), 54 deletions(-) diff --git a/drivers/bus/cdx/cdx_vfio.c b/drivers/bus/cdx/cdx_vfio.c index 11fe3265d2..9bae264409 100644 --- a/drivers/bus/cdx/cdx_vfio.c +++ b/drivers/bus/cdx/cdx_vfio.c @@ -401,10 +401,14 @@ cdx_vfio_map_resource_primary(struct rte_cdx_device *dev) return -1; ret = rte_vfio_setup_device(RTE_CDX_BUS_DEVICES_PATH, dev_name, - &vfio_dev_fd, &device_info); + &vfio_dev_fd); if (ret) return ret; + ret = rte_vfio_get_device_info(vfio_dev_fd, &device_info); + if (ret) + goto err_vfio_dev_fd; + /* allocate vfio_res and get region info */ vfio_res = rte_zmalloc("VFIO_RES", sizeof(*vfio_res), 0); if (vfio_res == NULL) { @@ -510,10 +514,14 @@ cdx_vfio_map_resource_secondary(struct rte_cdx_device *dev) } ret = rte_vfio_setup_device(RTE_CDX_BUS_DEVICES_PATH, dev_name, - &vfio_dev_fd, &device_info); + &vfio_dev_fd); if (ret) return ret; + ret = rte_vfio_get_device_info(vfio_dev_fd, &device_info); + if (ret) + goto err_vfio_dev_fd; + /* map MMIO regions */ maps = vfio_res->maps; diff --git a/drivers/bus/pci/linux/pci_vfio.c b/drivers/bus/pci/linux/pci_vfio.c index bc5c5c2499..54e9506058 100644 --- a/drivers/bus/pci/linux/pci_vfio.c +++ b/drivers/bus/pci/linux/pci_vfio.c @@ -753,10 +753,14 @@ pci_vfio_map_resource_primary(struct rte_pci_device *dev) loc->domain, loc->bus, loc->devid, loc->function); ret = rte_vfio_setup_device(rte_pci_get_sysfs_path(), pci_addr, - &vfio_dev_fd, &device_info); + &vfio_dev_fd); if (ret) return ret; + ret = rte_vfio_get_device_info(vfio_dev_fd, &device_info); + if (ret) + goto err_vfio_dev_fd; + if (rte_intr_dev_fd_set(dev->intr_handle, vfio_dev_fd)) goto err_vfio_dev_fd; @@ -962,10 +966,14 @@ pci_vfio_map_resource_secondary(struct rte_pci_device *dev) } ret = rte_vfio_setup_device(rte_pci_get_sysfs_path(), pci_addr, - &vfio_dev_fd, &device_info); + &vfio_dev_fd); if (ret) return ret; + ret = rte_vfio_get_device_info(vfio_dev_fd, &device_info); + if (ret) + goto err_vfio_dev_fd; + ret = pci_vfio_fill_regions(dev, vfio_dev_fd, &device_info); if (ret) goto err_vfio_dev; @@ -1195,12 +1203,14 @@ pci_vfio_ioport_map(struct rte_pci_device *dev, int bar, if (vfio_dev_fd < 0) { return -1; } else if (vfio_dev_fd == 0) { - if (rte_vfio_get_device_info(rte_pci_get_sysfs_path(), pci_addr, - &vfio_dev_fd, &device_info) != 0) + if (rte_vfio_setup_device(rte_pci_get_sysfs_path(), pci_addr, + &vfio_dev_fd) != 0) return -1; /* save vfio_dev_fd so it can be used during release */ if (rte_intr_dev_fd_set(dev->intr_handle, vfio_dev_fd) != 0) return -1; + if (rte_vfio_get_device_info(vfio_dev_fd, &device_info) != 0) + return -1; if (pci_vfio_fill_regions(dev, vfio_dev_fd, &device_info) != 0) return -1; diff --git a/drivers/bus/platform/platform.c b/drivers/bus/platform/platform.c index f6673cf181..9500b8aff5 100644 --- a/drivers/bus/platform/platform.c +++ b/drivers/bus/platform/platform.c @@ -328,12 +328,19 @@ device_setup(struct rte_platform_device *pdev) const char *name = pdev->name; int ret; - ret = rte_vfio_setup_device(PLATFORM_BUS_DEVICES_PATH, name, &pdev->dev_fd, &dev_info); + ret = rte_vfio_setup_device(PLATFORM_BUS_DEVICES_PATH, name, &pdev->dev_fd); if (ret) { PLATFORM_LOG_LINE(ERR, "failed to setup %s", name); return -ENODEV; } + ret = rte_vfio_get_device_info(pdev->dev_fd, &dev_info); + if (ret) { + PLATFORM_LOG_LINE(ERR, "failed to get device info for %s", name); + ret = -ENODEV; + goto out; + } + /* This is an extra check to confirm that platform device was initialized * by a kernel vfio-platform driver. On kernels that predate vfio-platform * driver this flag obviously does not exist. In such scenarios this diff --git a/drivers/crypto/bcmfs/bcmfs_vfio.c b/drivers/crypto/bcmfs/bcmfs_vfio.c index e7f7ed994c..d00aaf1bb7 100644 --- a/drivers/crypto/bcmfs/bcmfs_vfio.c +++ b/drivers/crypto/bcmfs/bcmfs_vfio.c @@ -25,12 +25,18 @@ vfio_map_dev_obj(const char *path, const char *dev_obj, struct vfio_device_info d_info = { .argsz = sizeof(d_info) }; struct vfio_region_info reg_info = { .argsz = sizeof(reg_info) }; - ret = rte_vfio_setup_device(path, dev_obj, dev_fd, &d_info); + ret = rte_vfio_setup_device(path, dev_obj, dev_fd); if (ret) { BCMFS_LOG(ERR, "VFIO Setting for device failed"); return ret; } + ret = rte_vfio_get_device_info(*dev_fd, &d_info); + if (ret) { + BCMFS_LOG(ERR, "VFIO Getting device info failed"); + goto map_failed; + } + /* getting device region info*/ ret = ioctl(*dev_fd, VFIO_DEVICE_GET_REGION_INFO, ®_info); if (ret < 0) { diff --git a/lib/eal/freebsd/eal.c b/lib/eal/freebsd/eal.c index 4cc80a2107..63eaa734eb 100644 --- a/lib/eal/freebsd/eal.c +++ b/lib/eal/freebsd/eal.c @@ -816,8 +816,7 @@ rte_eal_vfio_get_vf_token(__rte_unused rte_uuid_t vf_token) RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_setup_device) int rte_vfio_setup_device(__rte_unused const char *sysfs_base, __rte_unused const char *dev_addr, - __rte_unused int *vfio_dev_fd, - __rte_unused struct vfio_device_info *device_info) + __rte_unused int *vfio_dev_fd) { rte_errno = ENOTSUP; return -1; diff --git a/lib/eal/include/rte_vfio.h b/lib/eal/include/rte_vfio.h index 0ddeb08f94..fb666141f6 100644 --- a/lib/eal/include/rte_vfio.h +++ b/lib/eal/include/rte_vfio.h @@ -55,10 +55,7 @@ struct vfio_device_info; * device location. * * @param vfio_dev_fd - * VFIO fd. - * - * @param device_info - * Device information. + * Pointer to VFIO fd, will be set to the opened device fd on success. * * @return * 0 on success. @@ -67,7 +64,7 @@ struct vfio_device_info; */ __rte_internal int rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, - int *vfio_dev_fd, struct vfio_device_info *device_info); + int *vfio_dev_fd); /** * @internal @@ -187,19 +184,16 @@ rte_vfio_get_group_num(const char *sysfs_base, * @internal * Get device information. * + * This function retrieves VFIO device information from an already opened + * device. The device must be opened with rte_vfio_setup_device() first. + * * This function is only relevant to Linux and will return an error on BSD. * - * @param sysfs_base - * sysfs path prefix. - * - * @param dev_addr - * device location. - * * @param vfio_dev_fd - * VFIO fd. + * VFIO device fd (must be a valid, already opened fd). * * @param device_info - * Device information. + * Pointer to device information structure to be filled. * * @return * 0 on success. @@ -207,8 +201,7 @@ rte_vfio_get_group_num(const char *sysfs_base, */ __rte_internal int -rte_vfio_get_device_info(const char *sysfs_base, const char *dev_addr, - int *vfio_dev_fd, struct vfio_device_info *device_info); +rte_vfio_get_device_info(int vfio_dev_fd, struct vfio_device_info *device_info); /** * @internal diff --git a/lib/eal/linux/eal_vfio.c b/lib/eal/linux/eal_vfio.c index 33fa04feaf..47c973e49a 100644 --- a/lib/eal/linux/eal_vfio.c +++ b/lib/eal/linux/eal_vfio.c @@ -758,7 +758,7 @@ rte_vfio_clear_group(int vfio_group_fd) RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_setup_device) int rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, - int *vfio_dev_fd, struct vfio_device_info *device_info) + int *vfio_dev_fd) { struct vfio_group_status group_status = { .argsz = sizeof(group_status) @@ -975,7 +975,7 @@ rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, *vfio_dev_fd = ioctl(vfio_group_fd, VFIO_GROUP_GET_DEVICE_FD, dev); if (*vfio_dev_fd >= 0) - goto dev_get_info; + goto out; } /* get a file descriptor for the device */ @@ -992,18 +992,8 @@ rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, return -1; } - /* test and setup the device */ -dev_get_info: - ret = ioctl(*vfio_dev_fd, VFIO_DEVICE_GET_INFO, device_info); - if (ret) { - EAL_LOG(ERR, "%s cannot get device info, " - "error %i (%s)", dev_addr, errno, - strerror(errno)); - close(*vfio_dev_fd); - close(vfio_group_fd); - rte_vfio_clear_group(vfio_group_fd); - return -1; - } + /* device is now set up */ +out: vfio_group_device_get(vfio_group_fd); return 0; @@ -1217,26 +1207,18 @@ vfio_set_iommu_type(int vfio_container_fd) RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_device_info) int -rte_vfio_get_device_info(const char *sysfs_base, const char *dev_addr, - int *vfio_dev_fd, struct vfio_device_info *device_info) +rte_vfio_get_device_info(int vfio_dev_fd, struct vfio_device_info *device_info) { int ret; - if (device_info == NULL || *vfio_dev_fd < 0) + if (device_info == NULL || vfio_dev_fd < 0) return -1; - if (*vfio_dev_fd == 0) { - ret = rte_vfio_setup_device(sysfs_base, dev_addr, - vfio_dev_fd, device_info); - if (ret) - return -1; - } else { - ret = ioctl(*vfio_dev_fd, VFIO_DEVICE_GET_INFO, device_info); - if (ret) { - EAL_LOG(ERR, "%s cannot get device info, error %i (%s)", - dev_addr, errno, strerror(errno)); - return -1; - } + ret = ioctl(vfio_dev_fd, VFIO_DEVICE_GET_INFO, device_info); + if (ret) { + EAL_LOG(ERR, "Cannot get device info, error %i (%s)", + errno, strerror(errno)); + return -1; } return 0; -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* [PATCH v7 04/18] vfio: add container device assignment API 2026-02-26 14:17 ` [PATCH v7 00/18] Support VFIO cdev API in DPDK Anatoly Burakov ` (2 preceding siblings ...) 2026-02-26 14:17 ` [PATCH v7 03/18] vfio: split get device info from setup Anatoly Burakov @ 2026-02-26 14:17 ` Anatoly Burakov 2026-02-26 14:17 ` [PATCH v7 05/18] net/nbl: do not use VFIO group bind API Anatoly Burakov ` (14 subsequent siblings) 18 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2026-02-26 14:17 UTC (permalink / raw) To: dev, Bruce Richardson Currently, VFIO has explicit group bind API's, but the way they're used is such that no one actually cares about VFIO groups, and the real goal of everyone using VFIO group bind API is to bind devices to particular VFIO container, such that when `rte_vfio_setup_device` call eventually comes, VFIO will pick up the correct container. To remove dependency on group API's, add a new "container assign device" API call that will do the same thing, but will not depend on managing VFIO group fd's. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- lib/eal/freebsd/eal.c | 10 ++++++++++ lib/eal/include/rte_vfio.h | 26 ++++++++++++++++++++++++++ lib/eal/linux/eal_vfio.c | 32 ++++++++++++++++++++++++++++++++ 3 files changed, 68 insertions(+) diff --git a/lib/eal/freebsd/eal.c b/lib/eal/freebsd/eal.c index 63eaa734eb..0adb01032d 100644 --- a/lib/eal/freebsd/eal.c +++ b/lib/eal/freebsd/eal.c @@ -938,3 +938,13 @@ rte_vfio_container_dma_unmap(__rte_unused int container_fd, rte_errno = ENOTSUP; return -1; } + +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_assign_device) +int +rte_vfio_container_assign_device(__rte_unused int vfio_container_fd, + __rte_unused const char *sysfs_base, + __rte_unused const char *dev_addr) +{ + rte_errno = ENOTSUP; + return -1; +} diff --git a/lib/eal/include/rte_vfio.h b/lib/eal/include/rte_vfio.h index fb666141f6..e7e2ee950b 100644 --- a/lib/eal/include/rte_vfio.h +++ b/lib/eal/include/rte_vfio.h @@ -271,6 +271,32 @@ __rte_internal int rte_vfio_container_destroy(int container_fd); +/** + * @internal + * + * Assign a device to a VFIO container. + * + * Doing so will cause `rte_vfio_setup_device()` to set up the device with the VFIO container + * specified in this assign operation. + * + * This function is only relevant on Linux. + * + * @param vfio_container_fd + * VFIO container file descriptor. + * @param sysfs_base + * Sysfs path prefix. + * @param dev_addr + * Device identifier. + * + * @return + * 0 on success. + * <0 on failure, rte_errno is set. + */ +__rte_internal +int +rte_vfio_container_assign_device(int vfio_container_fd, const char *sysfs_base, + const char *dev_addr); + /** * @internal * Bind a IOMMU group to a container. diff --git a/lib/eal/linux/eal_vfio.c b/lib/eal/linux/eal_vfio.c index 47c973e49a..02fec64658 100644 --- a/lib/eal/linux/eal_vfio.c +++ b/lib/eal/linux/eal_vfio.c @@ -2102,6 +2102,38 @@ rte_vfio_container_destroy(int container_fd) return 0; } +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_assign_device) +int +rte_vfio_container_assign_device(int vfio_container_fd, const char *sysfs_base, + const char *dev_addr) +{ + int iommu_group_num; + int ret; + + ret = rte_vfio_get_group_num(sysfs_base, dev_addr, &iommu_group_num); + if (ret < 0) { + EAL_LOG(ERR, "Cannot get IOMMU group number for device %s", + dev_addr); + return -1; + } else if (ret == 0) { + EAL_LOG(ERR, + "Device %s is not assigned to any IOMMU group", + dev_addr); + return -1; + } + + ret = rte_vfio_container_group_bind(vfio_container_fd, + iommu_group_num); + if (ret < 0) { + EAL_LOG(ERR, + "Cannot bind IOMMU group %d for device %s", + iommu_group_num, dev_addr); + return -1; + } + + return 0; +} + RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_group_bind) int rte_vfio_container_group_bind(int container_fd, int iommu_group_num) -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* [PATCH v7 05/18] net/nbl: do not use VFIO group bind API 2026-02-26 14:17 ` [PATCH v7 00/18] Support VFIO cdev API in DPDK Anatoly Burakov ` (3 preceding siblings ...) 2026-02-26 14:17 ` [PATCH v7 04/18] vfio: add container device assignment API Anatoly Burakov @ 2026-02-26 14:17 ` Anatoly Burakov 2026-02-26 14:17 ` [PATCH v7 06/18] net/ntnic: use container device assignment API Anatoly Burakov ` (13 subsequent siblings) 18 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2026-02-26 14:17 UTC (permalink / raw) To: dev, Dimon Zhao, Kyo Liu, Leon Yu, Sam Chen The NBL driver currently uses group bind API, but it is using it only to get group fd and nothing else. In context of NBL driver, this is the only usage of VFIO API's in the driver, and it is not necessary to use it for what NBL driver is trying to accomplish. Use a direct `open()` call instead, and store the group fd in common structure. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- drivers/net/nbl/nbl_common/nbl_userdev.c | 18 +++++++++++------- drivers/net/nbl/nbl_include/nbl_include.h | 1 + 2 files changed, 12 insertions(+), 7 deletions(-) diff --git a/drivers/net/nbl/nbl_common/nbl_userdev.c b/drivers/net/nbl/nbl_common/nbl_userdev.c index 96f0d2e264..fb256e543f 100644 --- a/drivers/net/nbl/nbl_common/nbl_userdev.c +++ b/drivers/net/nbl/nbl_common/nbl_userdev.c @@ -387,6 +387,13 @@ nbl_userdev_mem_event_callback(enum rte_mem_event type, const void *addr, size_t } } +static int nbl_open_group_fd(int iommu_group_num) +{ + char path[PATH_MAX]; + snprintf(path, sizeof(path), RTE_VFIO_GROUP_FMT, iommu_group_num); + return open(path, O_RDWR); +} + static int nbl_mdev_map_device(struct nbl_adapter *adapter) { const struct rte_pci_device *pci_dev = adapter->pci_dev; @@ -424,11 +431,12 @@ static int nbl_mdev_map_device(struct nbl_adapter *adapter) } NBL_LOG(DEBUG, "nbl vfio container %d", container); - vfio_group_fd = rte_vfio_container_group_bind(container, common->iommu_group_num); + vfio_group_fd = nbl_open_group_fd(common->iommu_group_num); if (vfio_group_fd < 0) { NBL_LOG(ERR, "nbl vfio group bind failed, %d", vfio_group_fd); goto free_container; } + common->groupfd = vfio_group_fd; /* check if the group is viable */ ret = ioctl(vfio_group_fd, VFIO_GROUP_GET_STATUS, &group_status); @@ -535,7 +543,6 @@ static int nbl_mdev_map_device(struct nbl_adapter *adapter) } free_group: close(vfio_group_fd); - rte_vfio_clear_group(vfio_group_fd); free_container: if (container_create) rte_vfio_container_destroy(container); @@ -549,17 +556,14 @@ static int nbl_mdev_unmap_device(struct nbl_adapter *adapter) close(common->devfd); rte_mcfg_mem_read_lock(); - vfio_group_fd = rte_vfio_container_group_bind(nbl_default_container, - common->iommu_group_num); + vfio_group_fd = common->groupfd; NBL_LOG(DEBUG, "close vfio_group_fd %d", vfio_group_fd); ret = ioctl(vfio_group_fd, VFIO_GROUP_UNSET_CONTAINER, &nbl_default_container); if (ret) NBL_LOG(ERR, "unset container, error %i (%s) %d", errno, strerror(errno), ret); nbl_group_count--; - ret = rte_vfio_container_group_unbind(nbl_default_container, common->iommu_group_num); - if (ret) - NBL_LOG(ERR, "vfio container group unbind failed %d", ret); + close(vfio_group_fd); if (!nbl_group_count) { rte_mem_event_callback_unregister(NBL_USERDEV_EVENT_CLB_NAME, NULL); nbl_userdev_dma_free(); diff --git a/drivers/net/nbl/nbl_include/nbl_include.h b/drivers/net/nbl/nbl_include/nbl_include.h index eeae6a3301..ba99b9f8e7 100644 --- a/drivers/net/nbl/nbl_include/nbl_include.h +++ b/drivers/net/nbl/nbl_include/nbl_include.h @@ -132,6 +132,7 @@ struct nbl_common_info { u16 vsi_id; u16 instance_id; int devfd; + int groupfd; int eventfd; int ifindex; int iommu_group_num; -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* [PATCH v7 06/18] net/ntnic: use container device assignment API 2026-02-26 14:17 ` [PATCH v7 00/18] Support VFIO cdev API in DPDK Anatoly Burakov ` (4 preceding siblings ...) 2026-02-26 14:17 ` [PATCH v7 05/18] net/nbl: do not use VFIO group bind API Anatoly Burakov @ 2026-02-26 14:17 ` Anatoly Burakov 2026-02-26 14:17 ` [PATCH v7 07/18] vdpa/ifc: " Anatoly Burakov ` (12 subsequent siblings) 18 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2026-02-26 14:17 UTC (permalink / raw) To: dev, Christian Koue Muf, Serhii Iliushyk The NTNIC driver uses VFIO group bind/unbind functionality for container device assignment purposes. Use the new container device assignment API instead to provide clearer semantics. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- drivers/net/ntnic/ntnic_vfio.c | 30 +++++++++--------------------- 1 file changed, 9 insertions(+), 21 deletions(-) diff --git a/drivers/net/ntnic/ntnic_vfio.c b/drivers/net/ntnic/ntnic_vfio.c index 439468b3a2..c746b300b2 100644 --- a/drivers/net/ntnic/ntnic_vfio.c +++ b/drivers/net/ntnic/ntnic_vfio.c @@ -28,7 +28,6 @@ nt_vfio_vf_num(const struct rte_pci_device *pdev) /* Internal API */ struct vfio_dev { int container_fd; - int group_fd; int dev_fd; uint64_t iova_addr; }; @@ -50,7 +49,6 @@ nthw_vfio_setup(struct rte_pci_device *dev) { int ret; char devname[RTE_DEV_NAME_MAX_LEN] = { 0 }; - int iommu_group_num; int vf_num; struct vfio_dev *vfio; @@ -66,14 +64,9 @@ nthw_vfio_setup(struct rte_pci_device *dev) } vfio->dev_fd = -1; - vfio->group_fd = -1; vfio->iova_addr = START_VF_IOVA; rte_pci_device_name(&dev->addr, devname, RTE_DEV_NAME_MAX_LEN); - ret = rte_vfio_get_group_num(rte_pci_get_sysfs_path(), devname, &iommu_group_num); - if (ret <= 0) - return -1; - if (vf_num == 0) { /* use default container for pf0 */ vfio->container_fd = RTE_VFIO_DEFAULT_CONTAINER_FD; @@ -86,17 +79,14 @@ nthw_vfio_setup(struct rte_pci_device *dev) "VFIO device setup failed. VFIO container creation failed."); return -1; } - } + ret = rte_vfio_container_assign_device(vfio->container_fd, + rte_pci_get_sysfs_path(), devname); + if (ret < 0) { + NT_LOG(ERR, NTNIC, + "VFIO device setup failed. Assign device to container failed."); + goto err; + } - vfio->group_fd = rte_vfio_container_group_bind(vfio->container_fd, iommu_group_num); - - if (vfio->group_fd < 0) { - NT_LOG(ERR, NTNIC, - "VFIO device setup failed. VFIO container group bind failed."); - goto err; - } - - if (vf_num > 0) { if (rte_pci_map_device(dev)) { NT_LOG(ERR, NTNIC, "Map VFIO device failed. is the vfio-pci driver loaded?"); @@ -106,10 +96,8 @@ nthw_vfio_setup(struct rte_pci_device *dev) vfio->dev_fd = rte_intr_dev_fd_get(dev->intr_handle); - NT_LOG(DBG, NTNIC, - "%s: VFIO id=%d, dev_fd=%d, container_fd=%d, group_fd=%d, iommu_group_num=%d", - dev->name, vf_num, vfio->dev_fd, vfio->container_fd, vfio->group_fd, - iommu_group_num); + NT_LOG(DBG, NTNIC, "%s: VFIO id=%d, dev_fd=%d, container_fd=%d", + dev->name, vf_num, vfio->dev_fd, vfio->container_fd); return vf_num; -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* [PATCH v7 07/18] vdpa/ifc: use container device assignment API 2026-02-26 14:17 ` [PATCH v7 00/18] Support VFIO cdev API in DPDK Anatoly Burakov ` (5 preceding siblings ...) 2026-02-26 14:17 ` [PATCH v7 06/18] net/ntnic: use container device assignment API Anatoly Burakov @ 2026-02-26 14:17 ` Anatoly Burakov 2026-02-26 14:17 ` [PATCH v7 08/18] vdpa/nfp: " Anatoly Burakov ` (11 subsequent siblings) 18 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2026-02-26 14:17 UTC (permalink / raw) To: dev The IFC vDPA driver uses VFIO group bind/unbind functionality for container device assignment purposes. Use the new container device assignment API instead to provide clearer semantics. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- drivers/vdpa/ifc/ifcvf_vdpa.c | 15 +++------------ 1 file changed, 3 insertions(+), 12 deletions(-) diff --git a/drivers/vdpa/ifc/ifcvf_vdpa.c b/drivers/vdpa/ifc/ifcvf_vdpa.c index f319d455ba..6f1c050787 100644 --- a/drivers/vdpa/ifc/ifcvf_vdpa.c +++ b/drivers/vdpa/ifc/ifcvf_vdpa.c @@ -174,28 +174,19 @@ ifcvf_vfio_setup(struct ifcvf_internal *internal) { struct rte_pci_device *dev = internal->pdev; char devname[RTE_DEV_NAME_MAX_LEN] = {0}; - int iommu_group_num; - int i, ret; + int i; internal->vfio_dev_fd = -1; - internal->vfio_group_fd = -1; internal->vfio_container_fd = -1; rte_pci_device_name(&dev->addr, devname, RTE_DEV_NAME_MAX_LEN); - ret = rte_vfio_get_group_num(rte_pci_get_sysfs_path(), devname, - &iommu_group_num); - if (ret <= 0) { - DRV_LOG(ERR, "%s failed to get IOMMU group", devname); - return -1; - } internal->vfio_container_fd = rte_vfio_container_create(); if (internal->vfio_container_fd < 0) return -1; - internal->vfio_group_fd = rte_vfio_container_group_bind( - internal->vfio_container_fd, iommu_group_num); - if (internal->vfio_group_fd < 0) + if (rte_vfio_container_assign_device(internal->vfio_container_fd, + rte_pci_get_sysfs_path(), devname) < 0) goto err; if (rte_pci_map_device(dev)) -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* [PATCH v7 08/18] vdpa/nfp: use container device assignment API 2026-02-26 14:17 ` [PATCH v7 00/18] Support VFIO cdev API in DPDK Anatoly Burakov ` (6 preceding siblings ...) 2026-02-26 14:17 ` [PATCH v7 07/18] vdpa/ifc: " Anatoly Burakov @ 2026-02-26 14:17 ` Anatoly Burakov 2026-02-26 14:17 ` [PATCH v7 09/18] vdpa/sfc: " Anatoly Burakov ` (10 subsequent siblings) 18 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2026-02-26 14:17 UTC (permalink / raw) To: dev, Chaoyong He The NFP vDPA driver uses VFIO group bind/unbind functionality for container device assignment purposes. Use the new container device assignment API instead to provide clearer semantics. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- drivers/vdpa/nfp/nfp_vdpa.c | 17 +++++------------ 1 file changed, 5 insertions(+), 12 deletions(-) diff --git a/drivers/vdpa/nfp/nfp_vdpa.c b/drivers/vdpa/nfp/nfp_vdpa.c index f4fd5c92ec..4885fa5cbc 100644 --- a/drivers/vdpa/nfp/nfp_vdpa.c +++ b/drivers/vdpa/nfp/nfp_vdpa.c @@ -122,33 +122,26 @@ nfp_vdpa_vfio_setup(struct nfp_vdpa_dev *device) rte_pci_unmap_device(pci_dev); rte_pci_device_name(&pci_dev->addr, dev_name, RTE_DEV_NAME_MAX_LEN); - ret = rte_vfio_get_group_num(rte_pci_get_sysfs_path(), dev_name, - &device->iommu_group); - if (ret <= 0) - return -1; device->vfio_container_fd = rte_vfio_container_create(); if (device->vfio_container_fd < 0) return -1; - device->vfio_group_fd = rte_vfio_container_group_bind( - device->vfio_container_fd, device->iommu_group); - if (device->vfio_group_fd < 0) + ret = rte_vfio_container_assign_device(device->vfio_container_fd, + rte_pci_get_sysfs_path(), dev_name); + if (ret < 0) goto container_destroy; - DRV_VDPA_LOG(DEBUG, "The container_fd=%d, group_fd=%d.", - device->vfio_container_fd, device->vfio_group_fd); + DRV_VDPA_LOG(DEBUG, "container_fd=%d", device->vfio_container_fd); ret = rte_pci_map_device(pci_dev); if (ret != 0) - goto group_unbind; + goto container_destroy; device->vfio_dev_fd = rte_intr_dev_fd_get(pci_dev->intr_handle); return 0; -group_unbind: - rte_vfio_container_group_unbind(device->vfio_container_fd, device->iommu_group); container_destroy: rte_vfio_container_destroy(device->vfio_container_fd); -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* [PATCH v7 09/18] vdpa/sfc: use container device assignment API 2026-02-26 14:17 ` [PATCH v7 00/18] Support VFIO cdev API in DPDK Anatoly Burakov ` (7 preceding siblings ...) 2026-02-26 14:17 ` [PATCH v7 08/18] vdpa/nfp: " Anatoly Burakov @ 2026-02-26 14:17 ` Anatoly Burakov 2026-02-26 14:17 ` [PATCH v7 10/18] vhost: remove group-related API from drivers Anatoly Burakov ` (9 subsequent siblings) 18 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2026-02-26 14:17 UTC (permalink / raw) To: dev, Vijay Kumar Srivastava The SFC vDPA driver uses VFIO group bind/unbind functionality for container device assignment purposes. Use the new container device assignment API instead to provide clearer semantics. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- drivers/vdpa/sfc/sfc_vdpa.c | 39 +++++++------------------------------ drivers/vdpa/sfc/sfc_vdpa.h | 2 -- 2 files changed, 7 insertions(+), 34 deletions(-) diff --git a/drivers/vdpa/sfc/sfc_vdpa.c b/drivers/vdpa/sfc/sfc_vdpa.c index eda111954f..99b4ced3f4 100644 --- a/drivers/vdpa/sfc/sfc_vdpa.c +++ b/drivers/vdpa/sfc/sfc_vdpa.c @@ -80,22 +80,12 @@ sfc_vdpa_vfio_setup(struct sfc_vdpa_adapter *sva) goto fail_container_create; } - rc = rte_vfio_get_group_num(rte_pci_get_sysfs_path(), dev_name, - &sva->iommu_group_num); - if (rc <= 0) { - sfc_vdpa_err(sva, "failed to get IOMMU group for %s : %s", - dev_name, rte_strerror(-rc)); - goto fail_get_group_num; - } - - sva->vfio_group_fd = - rte_vfio_container_group_bind(sva->vfio_container_fd, - sva->iommu_group_num); - if (sva->vfio_group_fd < 0) { - sfc_vdpa_err(sva, - "failed to bind IOMMU group %d to container %d", - sva->iommu_group_num, sva->vfio_container_fd); - goto fail_group_bind; + rc = rte_vfio_container_assign_device(sva->vfio_container_fd, + rte_pci_get_sysfs_path(), dev_name); + if (rc < 0) { + sfc_vdpa_err(sva, "failed to assign device %s to container %d", + dev_name, sva->vfio_container_fd); + goto fail_device_assign; } if (rte_pci_map_device(dev) != 0) { @@ -109,15 +99,7 @@ sfc_vdpa_vfio_setup(struct sfc_vdpa_adapter *sva) return 0; fail_pci_map_device: - if (rte_vfio_container_group_unbind(sva->vfio_container_fd, - sva->iommu_group_num) != 0) { - sfc_vdpa_err(sva, - "failed to unbind IOMMU group %d from container %d", - sva->iommu_group_num, sva->vfio_container_fd); - } - -fail_group_bind: -fail_get_group_num: +fail_device_assign: if (rte_vfio_container_destroy(sva->vfio_container_fd) != 0) { sfc_vdpa_err(sva, "failed to destroy container %d", sva->vfio_container_fd); @@ -132,13 +114,6 @@ sfc_vdpa_vfio_teardown(struct sfc_vdpa_adapter *sva) { rte_pci_unmap_device(sva->pdev); - if (rte_vfio_container_group_unbind(sva->vfio_container_fd, - sva->iommu_group_num) != 0) { - sfc_vdpa_err(sva, - "failed to unbind IOMMU group %d from container %d", - sva->iommu_group_num, sva->vfio_container_fd); - } - if (rte_vfio_container_destroy(sva->vfio_container_fd) != 0) { sfc_vdpa_err(sva, "failed to destroy container %d", diff --git a/drivers/vdpa/sfc/sfc_vdpa.h b/drivers/vdpa/sfc/sfc_vdpa.h index 2b843e563d..99a81fd1b0 100644 --- a/drivers/vdpa/sfc/sfc_vdpa.h +++ b/drivers/vdpa/sfc/sfc_vdpa.h @@ -70,10 +70,8 @@ struct sfc_vdpa_adapter { sfc_vdpa_filter_t filters; - int vfio_group_fd; int vfio_dev_fd; int vfio_container_fd; - int iommu_group_num; struct sfc_vdpa_ops_data *ops_data; }; -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* [PATCH v7 10/18] vhost: remove group-related API from drivers 2026-02-26 14:17 ` [PATCH v7 00/18] Support VFIO cdev API in DPDK Anatoly Burakov ` (8 preceding siblings ...) 2026-02-26 14:17 ` [PATCH v7 09/18] vdpa/sfc: " Anatoly Burakov @ 2026-02-26 14:17 ` Anatoly Burakov 2026-02-26 14:17 ` [PATCH v7 11/18] vfio: remove group-based API Anatoly Burakov ` (8 subsequent siblings) 18 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2026-02-26 14:17 UTC (permalink / raw) To: dev, Maxime Coquelin, Chenbo Xia, Matan Azrad, Viacheslav Ovsiienko, Chaoyong He Some vDPA drivers have "get_vfio_group_fd" call in their internal driver API structure, but it is not used for anything beyond device assignment to containers which can now be achieved via other means, so remove this API and all its usages. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- doc/guides/prog_guide/vhost_lib.rst | 4 ---- doc/guides/rel_notes/deprecation.rst | 3 --- drivers/vdpa/ifc/ifcvf_vdpa.c | 19 ------------------- drivers/vdpa/mlx5/mlx5_vdpa.c | 1 - drivers/vdpa/nfp/nfp_vdpa.c | 20 -------------------- lib/vhost/vdpa_driver.h | 3 --- 6 files changed, 50 deletions(-) diff --git a/doc/guides/prog_guide/vhost_lib.rst b/doc/guides/prog_guide/vhost_lib.rst index 0c2b4d020a..2f80cf4072 100644 --- a/doc/guides/prog_guide/vhost_lib.rst +++ b/doc/guides/prog_guide/vhost_lib.rst @@ -471,10 +471,6 @@ Finally, a set of device ops is defined for device specific operations: Called to allow the device to response to RARP sending. -* ``get_vfio_group_fd`` - - Called to get the VFIO group fd of the device. - * ``get_vfio_device_fd`` Called to get the VFIO device fd of the device. diff --git a/doc/guides/rel_notes/deprecation.rst b/doc/guides/rel_notes/deprecation.rst index b137ed0445..ac667e91a6 100644 --- a/doc/guides/rel_notes/deprecation.rst +++ b/doc/guides/rel_notes/deprecation.rst @@ -139,6 +139,3 @@ Deprecation Notices * bus/vmbus: Starting DPDK 25.11, all the vmbus API defined in ``drivers/bus/vmbus/rte_bus_vmbus.h`` will become internal to DPDK. Those API functions are used internally by DPDK core and netvsc PMD. - -* vdpa: vDPA driver API will no longer offer ``get_vfio_group_fd`` as part of its internal API. - All drivers will be adjusted to use the new unified VFIO container device assignment API. diff --git a/drivers/vdpa/ifc/ifcvf_vdpa.c b/drivers/vdpa/ifc/ifcvf_vdpa.c index 6f1c050787..63f4172da5 100644 --- a/drivers/vdpa/ifc/ifcvf_vdpa.c +++ b/drivers/vdpa/ifc/ifcvf_vdpa.c @@ -58,7 +58,6 @@ struct ifcvf_internal { struct ifcvf_hw hw; int configured; int vfio_container_fd; - int vfio_group_fd; int vfio_dev_fd; rte_thread_t tid; /* thread for notify relay */ rte_thread_t intr_tid; /* thread for config space change interrupt relay */ @@ -1204,22 +1203,6 @@ ifcvf_set_features(int vid) return 0; } -static int -ifcvf_get_vfio_group_fd(int vid) -{ - struct rte_vdpa_device *vdev; - struct internal_list *list; - - vdev = rte_vhost_get_vdpa_device(vid); - list = find_internal_resource_by_vdev(vdev); - if (list == NULL) { - DRV_LOG(ERR, "Invalid vDPA device: %p", vdev); - return -1; - } - - return list->internal->vfio_group_fd; -} - static int ifcvf_get_vfio_device_fd(int vid) { @@ -1465,7 +1448,6 @@ static struct rte_vdpa_dev_ops ifcvf_net_ops = { .set_vring_state = ifcvf_set_vring_state, .set_features = ifcvf_set_features, .migration_done = NULL, - .get_vfio_group_fd = ifcvf_get_vfio_group_fd, .get_vfio_device_fd = ifcvf_get_vfio_device_fd, .get_notify_area = ifcvf_get_notify_area, .get_dev_type = ifcvf_get_device_type, @@ -1596,7 +1578,6 @@ static struct rte_vdpa_dev_ops ifcvf_blk_ops = { .dev_close = ifcvf_dev_close, .set_vring_state = ifcvf_set_vring_state, .migration_done = NULL, - .get_vfio_group_fd = ifcvf_get_vfio_group_fd, .get_vfio_device_fd = ifcvf_get_vfio_device_fd, .get_notify_area = ifcvf_get_notify_area, .get_config = ifcvf_blk_get_config, diff --git a/drivers/vdpa/mlx5/mlx5_vdpa.c b/drivers/vdpa/mlx5/mlx5_vdpa.c index 11708e2005..1bb5de51b6 100644 --- a/drivers/vdpa/mlx5/mlx5_vdpa.c +++ b/drivers/vdpa/mlx5/mlx5_vdpa.c @@ -523,7 +523,6 @@ static struct rte_vdpa_dev_ops mlx5_vdpa_ops = { .set_vring_state = mlx5_vdpa_set_vring_state, .set_features = mlx5_vdpa_features_set, .migration_done = NULL, - .get_vfio_group_fd = NULL, .get_vfio_device_fd = mlx5_vdpa_get_device_fd, .get_notify_area = mlx5_vdpa_get_notify_area, .get_stats_names = mlx5_vdpa_get_stats_names, diff --git a/drivers/vdpa/nfp/nfp_vdpa.c b/drivers/vdpa/nfp/nfp_vdpa.c index 4885fa5cbc..efd7ee7d95 100644 --- a/drivers/vdpa/nfp/nfp_vdpa.c +++ b/drivers/vdpa/nfp/nfp_vdpa.c @@ -36,9 +36,7 @@ struct nfp_vdpa_dev { struct nfp_vdpa_hw hw; int vfio_container_fd; - int vfio_group_fd; int vfio_dev_fd; - int iommu_group; rte_thread_t tid; /**< Thread for notify relay */ int epoll_fd; @@ -152,7 +150,6 @@ static void nfp_vdpa_vfio_teardown(struct nfp_vdpa_dev *device) { rte_pci_unmap_device(device->pci_dev); - rte_vfio_container_group_unbind(device->vfio_container_fd, device->iommu_group); rte_vfio_container_destroy(device->vfio_container_fd); } @@ -1018,22 +1015,6 @@ nfp_vdpa_dev_close(int vid) return 0; } -static int -nfp_vdpa_get_vfio_group_fd(int vid) -{ - struct rte_vdpa_device *vdev; - struct nfp_vdpa_dev_node *node; - - vdev = rte_vhost_get_vdpa_device(vid); - node = nfp_vdpa_find_node_by_vdev(vdev); - if (node == NULL) { - DRV_VDPA_LOG(ERR, "Invalid vDPA device: %p.", vdev); - return -ENODEV; - } - - return node->device->vfio_group_fd; -} - static int nfp_vdpa_get_vfio_device_fd(int vid) { @@ -1185,7 +1166,6 @@ struct rte_vdpa_dev_ops nfp_vdpa_ops = { .dev_close = nfp_vdpa_dev_close, .set_vring_state = nfp_vdpa_set_vring_state, .set_features = nfp_vdpa_set_features, - .get_vfio_group_fd = nfp_vdpa_get_vfio_group_fd, .get_vfio_device_fd = nfp_vdpa_get_vfio_device_fd, .get_notify_area = nfp_vdpa_get_notify_area, }; diff --git a/lib/vhost/vdpa_driver.h b/lib/vhost/vdpa_driver.h index 42392a0d14..c7b9be09fb 100644 --- a/lib/vhost/vdpa_driver.h +++ b/lib/vhost/vdpa_driver.h @@ -50,9 +50,6 @@ struct rte_vdpa_dev_ops { /** Destination operations when migration done */ int (*migration_done)(int vid); - /** Get the vfio group fd */ - int (*get_vfio_group_fd)(int vid); - /** Get the vfio device fd */ int (*get_vfio_device_fd)(int vid); -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* [PATCH v7 11/18] vfio: remove group-based API 2026-02-26 14:17 ` [PATCH v7 00/18] Support VFIO cdev API in DPDK Anatoly Burakov ` (9 preceding siblings ...) 2026-02-26 14:17 ` [PATCH v7 10/18] vhost: remove group-related API from drivers Anatoly Burakov @ 2026-02-26 14:17 ` Anatoly Burakov 2026-02-26 14:17 ` [PATCH v7 12/18] vfio: cleanup and refactor Anatoly Burakov ` (7 subsequent siblings) 18 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2026-02-26 14:17 UTC (permalink / raw) To: dev, Bruce Richardson All drivers have been adjusted to not use the VFIO group API directly and instead rely on container device assignment model, so the group API is no longer useful and can be removed. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- lib/eal/freebsd/eal.c | 33 --------------- lib/eal/include/rte_vfio.h | 72 -------------------------------- lib/eal/linux/eal_vfio.c | 47 ++++++++++----------- lib/eal/linux/eal_vfio.h | 2 + lib/eal/linux/eal_vfio_mp_sync.c | 2 +- 5 files changed, 26 insertions(+), 130 deletions(-) diff --git a/lib/eal/freebsd/eal.c b/lib/eal/freebsd/eal.c index 0adb01032d..da59e3d5cf 100644 --- a/lib/eal/freebsd/eal.c +++ b/lib/eal/freebsd/eal.c @@ -850,13 +850,6 @@ int rte_vfio_noiommu_is_enabled(void) return 0; } -RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_clear_group) -int rte_vfio_clear_group(__rte_unused int vfio_group_fd) -{ - rte_errno = ENOTSUP; - return -1; -} - RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_group_num) int rte_vfio_get_group_num(__rte_unused const char *sysfs_base, @@ -875,14 +868,6 @@ rte_vfio_get_container_fd(void) return -1; } -RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_group_fd) -int -rte_vfio_get_group_fd(__rte_unused int iommu_group_num) -{ - rte_errno = ENOTSUP; - return -1; -} - RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_create) int rte_vfio_container_create(void) @@ -899,24 +884,6 @@ rte_vfio_container_destroy(__rte_unused int container_fd) return -1; } -RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_group_bind) -int -rte_vfio_container_group_bind(__rte_unused int container_fd, - __rte_unused int iommu_group_num) -{ - rte_errno = ENOTSUP; - return -1; -} - -RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_group_unbind) -int -rte_vfio_container_group_unbind(__rte_unused int container_fd, - __rte_unused int iommu_group_num) -{ - rte_errno = ENOTSUP; - return -1; -} - RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_dma_map) int rte_vfio_container_dma_map(__rte_unused int container_fd, diff --git a/lib/eal/include/rte_vfio.h b/lib/eal/include/rte_vfio.h index e7e2ee950b..941b7d0541 100644 --- a/lib/eal/include/rte_vfio.h +++ b/lib/eal/include/rte_vfio.h @@ -136,24 +136,6 @@ int rte_vfio_is_enabled(const char *modname); __rte_internal int rte_vfio_noiommu_is_enabled(void); -/** - * @internal - * Remove group fd from internal VFIO group fd array. - * - * This function is only relevant to linux and will return - * an error on BSD. - * - * @param vfio_group_fd - * VFIO Group FD. - * - * @return - * 0 on success. - * <0 on failure. - */ -__rte_internal -int -rte_vfio_clear_group(int vfio_group_fd); - /** * @internal * Parse IOMMU group number for a device. @@ -218,24 +200,6 @@ __rte_internal int rte_vfio_get_container_fd(void); -/** - * @internal - * Open VFIO group fd or get an existing one. - * - * This function is only relevant to linux and will return - * an error on BSD. - * - * @param iommu_group_num - * iommu group number - * - * @return - * > 0 group fd - * < 0 for errors - */ -__rte_internal -int -rte_vfio_get_group_fd(int iommu_group_num); - /** * @internal * Create a new container for device binding. @@ -297,42 +261,6 @@ int rte_vfio_container_assign_device(int vfio_container_fd, const char *sysfs_base, const char *dev_addr); -/** - * @internal - * Bind a IOMMU group to a container. - * - * @param container_fd - * the container's fd - * - * @param iommu_group_num - * the iommu group number to bind to container - * - * @return - * group fd if successful - * <0 if failed - */ -__rte_internal -int -rte_vfio_container_group_bind(int container_fd, int iommu_group_num); - -/** - * @internal - * Unbind a IOMMU group from a container. - * - * @param container_fd - * the container fd of container - * - * @param iommu_group_num - * the iommu group number to delete from container - * - * @return - * 0 if successful - * <0 if failed - */ -__rte_internal -int -rte_vfio_container_group_unbind(int container_fd, int iommu_group_num); - /** * @internal * Perform DMA mapping for devices in a container. diff --git a/lib/eal/linux/eal_vfio.c b/lib/eal/linux/eal_vfio.c index 02fec64658..7893d334eb 100644 --- a/lib/eal/linux/eal_vfio.c +++ b/lib/eal/linux/eal_vfio.c @@ -67,6 +67,9 @@ static int vfio_noiommu_dma_mem_map(int, uint64_t, uint64_t, uint64_t, int); static int vfio_dma_mem_map(struct vfio_config *vfio_cfg, uint64_t vaddr, uint64_t iova, uint64_t len, int do_map); +static int vfio_container_group_bind(int container_fd, int iommu_group_num); +static int vfio_container_group_unbind(int container_fd, int iommu_group_num); + /* IOMMU types we support */ static const struct vfio_iommu_type iommu_types[] = { /* x86 IOMMU, otherwise known as type 1 */ @@ -532,9 +535,8 @@ get_vfio_cfg_by_container_fd(int container_fd) return NULL; } -RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_group_fd) int -rte_vfio_get_group_fd(int iommu_group_num) +vfio_get_group_fd_by_num(int iommu_group_num) { struct vfio_config *vfio_cfg; @@ -731,9 +733,8 @@ vfio_sync_default_container(void) return -1; } -RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_clear_group) -int -rte_vfio_clear_group(int vfio_group_fd) +static int +vfio_clear_group(int vfio_group_fd) { int i; struct vfio_config *vfio_cfg; @@ -787,7 +788,7 @@ rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, return -1; /* get the actual group fd */ - vfio_group_fd = rte_vfio_get_group_fd(iommu_group_num); + vfio_group_fd = vfio_get_group_fd_by_num(iommu_group_num); if (vfio_group_fd < 0 && vfio_group_fd != -ENOENT) return -1; @@ -813,14 +814,14 @@ rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, EAL_LOG(ERR, "%s cannot get VFIO group status, " "error %i (%s)", dev_addr, errno, strerror(errno)); close(vfio_group_fd); - rte_vfio_clear_group(vfio_group_fd); + vfio_clear_group(vfio_group_fd); return -1; } else if (!(group_status.flags & VFIO_GROUP_FLAGS_VIABLE)) { EAL_LOG(ERR, "%s VFIO group is not viable! " "Not all devices in IOMMU group bound to VFIO or unbound", dev_addr); close(vfio_group_fd); - rte_vfio_clear_group(vfio_group_fd); + vfio_clear_group(vfio_group_fd); return -1; } @@ -841,7 +842,7 @@ rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, "%s cannot add VFIO group to container, error " "%i (%s)", dev_addr, errno, strerror(errno)); close(vfio_group_fd); - rte_vfio_clear_group(vfio_group_fd); + vfio_clear_group(vfio_group_fd); return -1; } @@ -865,7 +866,7 @@ rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, "%s failed to select IOMMU type", dev_addr); close(vfio_group_fd); - rte_vfio_clear_group(vfio_group_fd); + vfio_clear_group(vfio_group_fd); return -1; } /* lock memory hotplug before mapping and release it @@ -882,7 +883,7 @@ rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, "%i (%s)", dev_addr, errno, strerror(errno)); close(vfio_group_fd); - rte_vfio_clear_group(vfio_group_fd); + vfio_clear_group(vfio_group_fd); rte_mcfg_mem_read_unlock(); return -1; } @@ -951,7 +952,7 @@ rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, if (ret < 0) { EAL_LOG(ERR, "Could not sync default VFIO container"); close(vfio_group_fd); - rte_vfio_clear_group(vfio_group_fd); + vfio_clear_group(vfio_group_fd); return -1; } /* we have successfully initialized VFIO, notify user */ @@ -988,7 +989,7 @@ rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, EAL_LOG(WARNING, "Getting a vfio_dev_fd for %s failed", dev_addr); close(vfio_group_fd); - rte_vfio_clear_group(vfio_group_fd); + vfio_clear_group(vfio_group_fd); return -1; } @@ -1026,9 +1027,9 @@ rte_vfio_release_device(const char *sysfs_base, const char *dev_addr, } /* get the actual group fd */ - vfio_group_fd = rte_vfio_get_group_fd(iommu_group_num); + vfio_group_fd = vfio_get_group_fd_by_num(iommu_group_num); if (vfio_group_fd < 0) { - EAL_LOG(INFO, "rte_vfio_get_group_fd failed for %s", + EAL_LOG(INFO, "vfio_get_group_fd_by_num failed for %s", dev_addr); ret = vfio_group_fd; goto out; @@ -1064,7 +1065,7 @@ rte_vfio_release_device(const char *sysfs_base, const char *dev_addr, goto out; } - if (rte_vfio_clear_group(vfio_group_fd) < 0) { + if (vfio_clear_group(vfio_group_fd) < 0) { EAL_LOG(INFO, "Error when clearing group for %s", dev_addr); ret = -1; @@ -2091,7 +2092,7 @@ rte_vfio_container_destroy(int container_fd) for (i = 0; i < RTE_DIM(vfio_cfg->vfio_groups); i++) if (vfio_cfg->vfio_groups[i].group_num != -1) - rte_vfio_container_group_unbind(container_fd, + vfio_container_group_unbind(container_fd, vfio_cfg->vfio_groups[i].group_num); close(container_fd); @@ -2122,7 +2123,7 @@ rte_vfio_container_assign_device(int vfio_container_fd, const char *sysfs_base, return -1; } - ret = rte_vfio_container_group_bind(vfio_container_fd, + ret = vfio_container_group_bind(vfio_container_fd, iommu_group_num); if (ret < 0) { EAL_LOG(ERR, @@ -2134,9 +2135,8 @@ rte_vfio_container_assign_device(int vfio_container_fd, const char *sysfs_base, return 0; } -RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_group_bind) -int -rte_vfio_container_group_bind(int container_fd, int iommu_group_num) +static int +vfio_container_group_bind(int container_fd, int iommu_group_num) { struct vfio_config *vfio_cfg; @@ -2149,9 +2149,8 @@ rte_vfio_container_group_bind(int container_fd, int iommu_group_num) return vfio_get_group_fd(vfio_cfg, iommu_group_num); } -RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_group_unbind) -int -rte_vfio_container_group_unbind(int container_fd, int iommu_group_num) +static int +vfio_container_group_unbind(int container_fd, int iommu_group_num) { struct vfio_group *cur_grp = NULL; struct vfio_config *vfio_cfg; diff --git a/lib/eal/linux/eal_vfio.h b/lib/eal/linux/eal_vfio.h index 89c4b5ba45..30389fb274 100644 --- a/lib/eal/linux/eal_vfio.h +++ b/lib/eal/linux/eal_vfio.h @@ -51,6 +51,8 @@ vfio_set_iommu_type(int vfio_container_fd); int vfio_get_iommu_type(void); +int vfio_get_group_fd_by_num(int iommu_group_num); + /* check if we have any supported extensions */ int vfio_has_supported_extensions(int vfio_container_fd); diff --git a/lib/eal/linux/eal_vfio_mp_sync.c b/lib/eal/linux/eal_vfio_mp_sync.c index 22136f2e8b..3eaeef2fc8 100644 --- a/lib/eal/linux/eal_vfio_mp_sync.c +++ b/lib/eal/linux/eal_vfio_mp_sync.c @@ -34,7 +34,7 @@ vfio_mp_primary(const struct rte_mp_msg *msg, const void *peer) case SOCKET_REQ_GROUP: r->req = SOCKET_REQ_GROUP; r->group_num = m->group_num; - fd = rte_vfio_get_group_fd(m->group_num); + fd = vfio_get_group_fd_by_num(m->group_num); if (fd < 0 && fd != -ENOENT) r->result = SOCKET_ERR; else if (fd == -ENOENT) -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* [PATCH v7 12/18] vfio: cleanup and refactor 2026-02-26 14:17 ` [PATCH v7 00/18] Support VFIO cdev API in DPDK Anatoly Burakov ` (10 preceding siblings ...) 2026-02-26 14:17 ` [PATCH v7 11/18] vfio: remove group-based API Anatoly Burakov @ 2026-02-26 14:17 ` Anatoly Burakov 2026-02-26 14:17 ` [PATCH v7 13/18] bus/pci: use the new VFIO mode API Anatoly Burakov ` (6 subsequent siblings) 18 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2026-02-26 14:17 UTC (permalink / raw) To: dev, Wathsala Vithanage, Bruce Richardson, Nipun Gupta, Nikhil Agarwal, Hemant Agrawal, Sachin Saxena, Chenbo Xia, Ajit Khaparde, Vikas Gupta, Dimon Zhao, Kyo Liu, Leon Yu, Sam Chen Currently, VFIO code is a bit of an incoherent mess internally, with API's bleeding into each other, inconsistent returns, and a certain amount of spaghetti stemming from organic growth. Refactor VFIO code to achieve the following goals: - Make all error handling consistent, and provide/document rte_errno values returned from API's to indicate various conditions. - Introduce new "VFIO mode" concept. This new API will tell caller if VFIO is enabled, and whether it is using group API, and whether it is running in no-IOMMU mode. - Decouple rte_vfio_setup_device semantics from PCI bus return convention. Currently, when device is not managed by VFIO, rte_vfio_setup_device will return 1, which is bus speak for "skip this device", however VFIO has nothing to do with PCI bus and should not follow its API conventions. - Perform device setup in device assign, and make device setup use shared code path with device assign and explicitly assuming default container. This is technically not necessary for group mode as device set up is a two-step process in that mode, but coming cdev mode will have a single-step device setup, and it would be easier if the worked the same way under the hood. - Make VFIO internals more readable. Introduce a lot of infrastructure and more explicit validation, rather than over-reliance on sentinel values and implicit assumptions. This will also make it easier to integrate cdev mode down the line, as it will rely on most of this infrastructure. This will change behavior of the following functions: - `rte_vfio_setup_device` - when the device is not managed by VFIO, the function will now return -1 with `rte_errno` set to ENODEV - `rte_vfio_get_group_num` - when the device is not managed by VFIO, the function will now return -1 with `rte_errno` set to ENODEV - `rte_vfio_container_destroy` - the function will now release and close all group and device resources associated with the container being destroyed by this call All users of `rte_vfio_setup_device` and `rte_vfio_get_group_num` have been adjusted to account for API change. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com> --- config/arm/meson.build | 1 + config/meson.build | 1 + drivers/bus/cdx/cdx_vfio.c | 17 +- drivers/bus/fslmc/fslmc_vfio.c | 4 +- drivers/bus/pci/linux/pci_vfio.c | 19 +- drivers/crypto/bcmfs/bcmfs_vfio.c | 6 +- drivers/net/nbl/nbl_common/nbl_userdev.c | 2 +- lib/eal/freebsd/eal.c | 16 + lib/eal/include/rte_vfio.h | 250 ++- lib/eal/linux/eal_vfio.c | 2285 ++++++++-------------- lib/eal/linux/eal_vfio.h | 142 +- lib/eal/linux/eal_vfio_group.c | 984 ++++++++++ lib/eal/linux/eal_vfio_mp_sync.c | 38 +- lib/eal/linux/meson.build | 1 + 14 files changed, 2130 insertions(+), 1636 deletions(-) create mode 100644 lib/eal/linux/eal_vfio_group.c diff --git a/config/arm/meson.build b/config/arm/meson.build index 523b0fc0ed..0f7206a31d 100644 --- a/config/arm/meson.build +++ b/config/arm/meson.build @@ -148,6 +148,7 @@ implementer_cavium = { 'description': 'Cavium', 'flags': [ ['RTE_MAX_VFIO_GROUPS', 128], + ['RTE_MAX_VFIO_DEVICES', 256], ['RTE_MAX_LCORE', 96], ['RTE_MAX_NUMA_NODES', 2] ], diff --git a/config/meson.build b/config/meson.build index 02e2798cca..070a6badf9 100644 --- a/config/meson.build +++ b/config/meson.build @@ -381,6 +381,7 @@ dpdk_conf.set('RTE_ENABLE_TRACE_FP', get_option('enable_trace_fp')) dpdk_conf.set('RTE_PKTMBUF_HEADROOM', get_option('pkt_mbuf_headroom')) # values which have defaults which may be overridden dpdk_conf.set('RTE_MAX_VFIO_GROUPS', 64) +dpdk_conf.set('RTE_MAX_VFIO_DEVICES', 256) dpdk_conf.set('RTE_DRIVER_MEMPOOL_BUCKET_SIZE_KB', 64) dpdk_conf.set('RTE_LIBRTE_DPAA2_USE_PHYS_IOVA', true) if dpdk_conf.get('RTE_ARCH_64') diff --git a/drivers/bus/cdx/cdx_vfio.c b/drivers/bus/cdx/cdx_vfio.c index 9bae264409..873f0f3460 100644 --- a/drivers/bus/cdx/cdx_vfio.c +++ b/drivers/bus/cdx/cdx_vfio.c @@ -22,6 +22,7 @@ #include <eal_export.h> #include <rte_eal_paging.h> +#include <rte_errno.h> #include <rte_malloc.h> #include <rte_vfio.h> @@ -402,8 +403,12 @@ cdx_vfio_map_resource_primary(struct rte_cdx_device *dev) ret = rte_vfio_setup_device(RTE_CDX_BUS_DEVICES_PATH, dev_name, &vfio_dev_fd); - if (ret) + if (ret < 0) { + /* Device not managed by VFIO - skip */ + if (rte_errno == ENODEV) + ret = 1; return ret; + } ret = rte_vfio_get_device_info(vfio_dev_fd, &device_info); if (ret) @@ -513,11 +518,13 @@ cdx_vfio_map_resource_secondary(struct rte_cdx_device *dev) return -1; } - ret = rte_vfio_setup_device(RTE_CDX_BUS_DEVICES_PATH, dev_name, - &vfio_dev_fd); - if (ret) + ret = rte_vfio_setup_device(RTE_CDX_BUS_DEVICES_PATH, dev_name, &vfio_dev_fd); + if (ret < 0) { + /* Device not managed by VFIO - skip */ + if (rte_errno == ENODEV) + ret = 1; return ret; - + } ret = rte_vfio_get_device_info(vfio_dev_fd, &device_info); if (ret) goto err_vfio_dev_fd; diff --git a/drivers/bus/fslmc/fslmc_vfio.c b/drivers/bus/fslmc/fslmc_vfio.c index 550d4e0e8d..4b11b38322 100644 --- a/drivers/bus/fslmc/fslmc_vfio.c +++ b/drivers/bus/fslmc/fslmc_vfio.c @@ -364,9 +364,9 @@ fslmc_get_group_id(const char *group_name, /* get group number */ ret = rte_vfio_get_group_num(SYSFS_FSL_MC_DEVICES, group_name, groupid); - if (ret <= 0) { + if (ret < 0) { DPAA2_BUS_ERR("Find %s IOMMU group", group_name); - if (ret < 0) + if (rte_errno != ENODEV) return ret; return -EIO; diff --git a/drivers/bus/pci/linux/pci_vfio.c b/drivers/bus/pci/linux/pci_vfio.c index 54e9506058..0d30b1cdf1 100644 --- a/drivers/bus/pci/linux/pci_vfio.c +++ b/drivers/bus/pci/linux/pci_vfio.c @@ -20,6 +20,7 @@ #include <rte_malloc.h> #include <rte_vfio.h> #include <rte_eal.h> +#include <rte_errno.h> #include <bus_driver.h> #include <rte_spinlock.h> #include <rte_tailq.h> @@ -752,10 +753,13 @@ pci_vfio_map_resource_primary(struct rte_pci_device *dev) snprintf(pci_addr, sizeof(pci_addr), PCI_PRI_FMT, loc->domain, loc->bus, loc->devid, loc->function); - ret = rte_vfio_setup_device(rte_pci_get_sysfs_path(), pci_addr, - &vfio_dev_fd); - if (ret) + ret = rte_vfio_setup_device(rte_pci_get_sysfs_path(), pci_addr, &vfio_dev_fd); + if (ret < 0) { + /* Device not managed by VFIO - skip */ + if (rte_errno == ENODEV) + ret = 1; return ret; + } ret = rte_vfio_get_device_info(vfio_dev_fd, &device_info); if (ret) @@ -965,10 +969,13 @@ pci_vfio_map_resource_secondary(struct rte_pci_device *dev) return -1; } - ret = rte_vfio_setup_device(rte_pci_get_sysfs_path(), pci_addr, - &vfio_dev_fd); - if (ret) + ret = rte_vfio_setup_device(rte_pci_get_sysfs_path(), pci_addr, &vfio_dev_fd); + if (ret < 0) { + /* Device not managed by VFIO - skip */ + if (rte_errno == ENODEV) + ret = 1; return ret; + } ret = rte_vfio_get_device_info(vfio_dev_fd, &device_info); if (ret) diff --git a/drivers/crypto/bcmfs/bcmfs_vfio.c b/drivers/crypto/bcmfs/bcmfs_vfio.c index d00aaf1bb7..92d8de4443 100644 --- a/drivers/crypto/bcmfs/bcmfs_vfio.c +++ b/drivers/crypto/bcmfs/bcmfs_vfio.c @@ -9,6 +9,7 @@ #include <sys/mman.h> #include <sys/ioctl.h> +#include <rte_errno.h> #include <rte_vfio.h> #include "bcmfs_device.h" @@ -26,7 +27,10 @@ vfio_map_dev_obj(const char *path, const char *dev_obj, struct vfio_region_info reg_info = { .argsz = sizeof(reg_info) }; ret = rte_vfio_setup_device(path, dev_obj, dev_fd); - if (ret) { + if (ret < 0) { + /* Device not managed by VFIO - skip */ + if (rte_errno == ENODEV) + ret = 1; BCMFS_LOG(ERR, "VFIO Setting for device failed"); return ret; } diff --git a/drivers/net/nbl/nbl_common/nbl_userdev.c b/drivers/net/nbl/nbl_common/nbl_userdev.c index fb256e543f..9aacf7438c 100644 --- a/drivers/net/nbl/nbl_common/nbl_userdev.c +++ b/drivers/net/nbl/nbl_common/nbl_userdev.c @@ -413,7 +413,7 @@ static int nbl_mdev_map_device(struct nbl_adapter *adapter) "%s/%s/", rte_pci_get_sysfs_path(), dev_name); ret = rte_vfio_get_group_num(pathname, dev_name, &common->iommu_group_num); - if (ret <= 0) { + if (ret < 0) { NBL_LOG(ERR, "nbl vfio group number failed"); return -1; } diff --git a/lib/eal/freebsd/eal.c b/lib/eal/freebsd/eal.c index da59e3d5cf..86ac148c88 100644 --- a/lib/eal/freebsd/eal.c +++ b/lib/eal/freebsd/eal.c @@ -915,3 +915,19 @@ rte_vfio_container_assign_device(__rte_unused int vfio_container_fd, rte_errno = ENOTSUP; return -1; } + +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_device_info) +int +rte_vfio_get_device_info(__rte_unused int vfio_dev_fd, + __rte_unused struct vfio_device_info *device_info) +{ + rte_errno = ENOTSUP; + return -1; +} + +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_mode) +enum rte_vfio_mode +rte_vfio_get_mode(void) +{ + return RTE_VFIO_MODE_NONE; +} diff --git a/lib/eal/include/rte_vfio.h b/lib/eal/include/rte_vfio.h index 941b7d0541..0af41c3610 100644 --- a/lib/eal/include/rte_vfio.h +++ b/lib/eal/include/rte_vfio.h @@ -18,6 +18,7 @@ #include <stdint.h> #include <rte_compat.h> +#include <rte_common.h> #ifdef __cplusplus extern "C" { @@ -29,8 +30,7 @@ extern "C" { #define RTE_VFIO_CONTAINER_PATH "/dev/vfio/vfio" #define RTE_VFIO_GROUP_FMT "/dev/vfio/%u" #define RTE_VFIO_NOIOMMU_GROUP_FMT "/dev/vfio/noiommu-%u" -#define RTE_VFIO_NOIOMMU_MODE \ - "/sys/module/vfio/parameters/enable_unsafe_noiommu_mode" +#define RTE_VFIO_NOIOMMU_MODE "/sys/module/vfio/parameters/enable_unsafe_noiommu_mode" #endif /* RTE_EXEC_ENV_LINUX */ @@ -39,28 +39,49 @@ struct vfio_device_info; #define RTE_VFIO_DEFAULT_CONTAINER_FD (-1) +/** + * @enum rte_vfio_mode + * Enumeration of VFIO operational modes. + * + * These modes define how VFIO devices are accessed and managed: + * + * - RTE_VFIO_MODE_NONE: VFIO is not enabled. + * - RTE_VFIO_MODE_GROUP: Legacy group mode. + * - RTE_VFIO_MODE_NOIOMMU: Unsafe no-IOMMU mode. + */ +enum rte_vfio_mode { + RTE_VFIO_MODE_NONE = 0, /**< VFIO not enabled */ + RTE_VFIO_MODE_GROUP, /**< Group mode */ + RTE_VFIO_MODE_NOIOMMU, /**< Group mode with no IOMMU protection */ +}; + /** * @internal - * Setup vfio_cfg for the device identified by its address. - * It discovers the configured I/O MMU groups or sets a new one for the device. - * If a new groups is assigned, the DMA mapping is performed. + * Set up a device managed by VFIO driver. * - * This function is only relevant to linux and will return - * an error on BSD. + * If the device was not previously assigned to a container using + * `rte_vfio_container_assign_device()`, default container will be used. + * + * This function is only relevant on Linux. * * @param sysfs_base - * sysfs path prefix. - * + * Sysfs path prefix. * @param dev_addr - * device location. - * + * Device identifier. * @param vfio_dev_fd - * Pointer to VFIO fd, will be set to the opened device fd on success. + * Pointer to where VFIO device file descriptor will be stored. * * @return * 0 on success. - * <0 on failure. - * >1 if the device cannot be managed this way. + * <0 on failure, rte_errno is set. + * + * Possible rte_errno values include: + * - ENODEV - Device not managed by VFIO. + * - ENOSPC - No space in VFIO container to track the device. + * - EINVAL - Invalid parameters. + * - EIO - Error during underlying VFIO operations. + * - ENXIO - VFIO support not initialized. + * - ENOTSUP - Unsupported VFIO mode. */ __rte_internal int rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, @@ -68,99 +89,127 @@ int rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, /** * @internal - * Release a device mapped to a VFIO-managed I/O MMU group. + * Release a device managed by VFIO driver. * - * This function is only relevant to linux and will return - * an error on BSD. + * This function is only relevant on Linux. + * + * @note As a result of this function, all internal resources used by the device will be released, + * so if the device was using a non-default container, it will need to be reassigned to the + * container before it can be used again. * * @param sysfs_base - * sysfs path prefix. - * + * Sysfs path prefix. * @param dev_addr - * device location. - * + * Device identifier. * @param fd - * VFIO fd. + * A previously set up VFIO file descriptor. * * @return * 0 on success. - * <0 on failure. + * <0 on failure, rte_errno is set. + * + * Possible rte_errno values include: + * - ENOENT - Device not found in any container. + * - EINVAL - Invalid parameters. + * - EIO - Error during underlying VFIO operations. + * - ENXIO - VFIO support not initialized. + * - ENOTSUP - Unsupported VFIO mode. */ __rte_internal int rte_vfio_release_device(const char *sysfs_base, const char *dev_addr, int fd); /** * @internal - * Enable a VFIO-related kmod. + * Enable VFIO subsystem and check if specified kernel module is loaded. * - * This function is only relevant to linux and will return - * an error on BSD. + * In case of success, `rte_vfio_get_mode()` can be used to retrieve the VFIO mode in use. + * + * This function is only relevant on Linux. * * @param modname - * kernel module name. + * Kernel module name. * * @return * 0 on success. - * <0 on failure. + * <0 on failure, rte_errno is set. + * + * Possible rte_errno values include: + * - EINVAL - Invalid parameters. + * - ENXIO - VFIO support not initialized. + * - ENOTSUP - Operation not supported. */ __rte_internal int rte_vfio_enable(const char *modname); /** * @internal - * Check whether a VFIO-related kmod is enabled. + * Check if VFIO subsystem is initialized and a specified kernel module is loaded. * - * This function is only relevant to Linux. + * This function is only relevant on Linux. * * @param modname - * kernel module name. + * Kernel module name. * * @return - * 1 if true. - * 0 otherwise. + * 1 if enabled. + * 0 if not enabled or not supported. */ __rte_internal int rte_vfio_is_enabled(const char *modname); /** * @internal - * Whether VFIO NOIOMMU mode is enabled. + * Get current VFIO mode. * - * This function is only relevant to Linux. + * This function is only relevant on Linux. * * @return - * 1 if true. - * 0 if false. - * <0 for errors. + * VFIO mode currently in use. */ __rte_internal -int rte_vfio_noiommu_is_enabled(void); +enum rte_vfio_mode +rte_vfio_get_mode(void); + +/** + * @internal + * Check if VFIO NOIOMMU mode is enabled. + * + * This function is only relevant on Linux in group mode. + * + * @return + * 1 if enabled. + * 0 if not enabled or not supported. + */ +__rte_internal +int +rte_vfio_noiommu_is_enabled(void); /** * @internal * Parse IOMMU group number for a device. * - * This function is only relevant to linux and will return - * an error on BSD. + * This function is only relevant on Linux in group mode. * * @param sysfs_base - * sysfs path prefix. - * + * Sysfs path prefix. * @param dev_addr - * device location. - * + * Device identifier. * @param iommu_group_num - * iommu group number + * Pointer to where IOMMU group number will be stored. * * @return - * >0 on success - * 0 for non-existent group or VFIO - * <0 for errors + * 0 on success. + * <0 on failure, rte_errno is set. + * + * Possible rte_errno values include: + * - ENODEV - Device not managed by VFIO. + * - EINVAL - Invalid parameters. + * - ENXIO - VFIO support not initialized. + * - ENOTSUP - Unsupported VFIO mode. */ __rte_internal int -rte_vfio_get_group_num(const char *sysfs_base, - const char *dev_addr, int *iommu_group_num); +rte_vfio_get_group_num(const char *sysfs_base, const char *dev_addr, int *iommu_group_num); /** * @internal @@ -179,7 +228,12 @@ rte_vfio_get_group_num(const char *sysfs_base, * * @return * 0 on success. - * <0 on failure. + * <0 on failure, rte_errno is set. + * + * Possible rte_errno values include: + * - EINVAL - Invalid parameters. + * - ENXIO - VFIO support not initialized. + * - ENOTSUP - Unsupported VFIO mode. */ __rte_internal int @@ -187,14 +241,17 @@ rte_vfio_get_device_info(int vfio_dev_fd, struct vfio_device_info *device_info); /** * @internal - * Get the default VFIO container fd + * Get the default VFIO container file descriptor. * - * This function is only relevant to linux and will return - * an error on BSD. + * This function is only relevant on Linux. * * @return - * > 0 default container fd - * < 0 if VFIO is not enabled or not supported + * Non-negative container file descriptor on success. + * <0 on failure, rte_errno is set. + * + * Possible rte_errno values include: + * - ENXIO - VFIO support not initialized. + * - ENOTSUP - Unsupported VFIO mode. */ __rte_internal int @@ -202,7 +259,9 @@ rte_vfio_get_container_fd(void); /** * @internal - * Create a new container for device binding. + * Create a new VFIO container for device assignment and DMA mapping. + * + * This function is only relevant on Linux. * * @note Any newly allocated DPDK memory will not be mapped into these * containers by default, user needs to manage DMA mappings for @@ -213,8 +272,14 @@ rte_vfio_get_container_fd(void); * devices between multiple processes is not supported. * * @return - * the container fd if successful - * <0 if failed + * Non-negative container file descriptor on success. + * <0 on failure, rte_errno is set. + * + * Possible rte_errno values include: + * - ENOSPC - Maximum number of containers reached. + * - EIO - Underlying VFIO operation failed. + * - ENXIO - VFIO support not initialized. + * - ENOTSUP - Unsupported VFIO mode. */ __rte_internal int @@ -222,14 +287,22 @@ rte_vfio_container_create(void); /** * @internal - * Destroy the container, unbind all vfio groups within it. + * Destroy a VFIO container and unmap all devices assigned to it. + * + * This function is only relevant on Linux. * * @param container_fd - * the container fd to destroy + * File descriptor of container to destroy. * * @return - * 0 if successful - * <0 if failed + * 0 on success. + * <0 on failure, rte_errno is set. + * + * Possible rte_errno values include: + * - ENODEV - Container not managed by VFIO. + * - EINVAL - Invalid container file descriptor. + * - ENXIO - VFIO support not initialized. + * - ENOTSUP - Unsupported VFIO mode. */ __rte_internal int @@ -255,32 +328,45 @@ rte_vfio_container_destroy(int container_fd); * @return * 0 on success. * <0 on failure, rte_errno is set. + * + * Possible rte_errno values include: + * - ENODEV - Device not managed by VFIO. + * - EEXIST - Device already assigned to the container. + * - ENOSPC - No space in VFIO container to assign device. + * - EINVAL - Invalid container file descriptor. + * - EIO - Error during underlying VFIO operations. + * - ENXIO - VFIO support not initialized. + * - ENOTSUP - Unsupported VFIO mode. */ __rte_internal int -rte_vfio_container_assign_device(int vfio_container_fd, const char *sysfs_base, - const char *dev_addr); +rte_vfio_container_assign_device(int vfio_container_fd, + const char *sysfs_base, const char *dev_addr); /** * @internal * Perform DMA mapping for devices in a container. * - * @param container_fd - * the specified container fd. Use RTE_VFIO_DEFAULT_CONTAINER_FD to - * use the default container. + * This function is only relevant on Linux. * + * @param container_fd + * Container file descriptor. Use RTE_VFIO_DEFAULT_CONTAINER_FD to use the default container. * @param vaddr * Starting virtual address of memory to be mapped. - * * @param iova * Starting IOVA address of memory to be mapped. - * * @param len * Length of memory segment being mapped. * * @return - * 0 if successful - * <0 if failed + * 0 on success. + * <0 on failure, rte_errno is set. + * + * Possible rte_errno values include: + * - EIO - DMA mapping operation failed. + * - EINVAL - Invalid parameters. + * - ENXIO - VFIO support not initialized. + * - ENOTSUP - Unsupported VFIO mode. */ __rte_internal int @@ -291,22 +377,26 @@ rte_vfio_container_dma_map(int container_fd, uint64_t vaddr, * @internal * Perform DMA unmapping for devices in a container. * - * @param container_fd - * the specified container fd. Use RTE_VFIO_DEFAULT_CONTAINER_FD to - * use the default container. + * This function is only relevant on Linux. * + * @param container_fd + * Container file descriptor. Use RTE_VFIO_DEFAULT_CONTAINER_FD to use the default container. * @param vaddr * Starting virtual address of memory to be unmapped. - * * @param iova * Starting IOVA address of memory to be unmapped. - * * @param len * Length of memory segment being unmapped. * * @return - * 0 if successful - * <0 if failed + * 0 on success. + * <0 on failure, rte_errno is set. + * + * Possible rte_errno values include: + * - EIO - DMA unmapping operation failed. + * - EINVAL - Invalid parameters. + * - ENXIO - VFIO support not initialized. + * - ENOTSUP - Unsupported VFIO mode. */ __rte_internal int diff --git a/lib/eal/linux/eal_vfio.c b/lib/eal/linux/eal_vfio.c index 7893d334eb..8545552308 100644 --- a/lib/eal/linux/eal_vfio.c +++ b/lib/eal/linux/eal_vfio.c @@ -9,6 +9,7 @@ #include <fcntl.h> #include <unistd.h> #include <sys/ioctl.h> +#include <sys/stat.h> #include <dirent.h> #include <rte_errno.h> @@ -24,80 +25,39 @@ #include "eal_private.h" #include "eal_internal_cfg.h" -#define VFIO_MEM_EVENT_CLB_NAME "vfio_mem_event_clb" - -/* hot plug/unplug of VFIO groups may cause all DMA maps to be dropped. we can - * recreate the mappings for DPDK segments, but we cannot do so for memory that - * was registered by the user themselves, so we need to store the user mappings - * somewhere, to recreate them later. +/* + * rte_errno convention: + * + * - EINVAL: invalid parameters + * - ENOTSUP: current mode does not support this operation + * - ENOXIO: VFIO not initialized + * - ENODEV: device not managed by VFIO + * - ENOSPC: no space in config + * - EEXIST: device already assigned + * - ENOENT: group or device not found + * - EIO: underlying VFIO operation failed */ -#define EAL_VFIO_MAX_USER_MEM_MAPS 256 -struct user_mem_map { - uint64_t addr; /**< start VA */ - uint64_t iova; /**< start IOVA */ - uint64_t len; /**< total length of the mapping */ - uint64_t chunk; /**< this mapping can be split in chunks of this size */ -}; -struct user_mem_maps { - rte_spinlock_recursive_t lock; - int n_maps; - struct user_mem_map maps[EAL_VFIO_MAX_USER_MEM_MAPS]; +/* functions can fail for multiple reasons, and errno is tedious */ +enum vfio_result { + VFIO_SUCCESS, + VFIO_ERROR, + VFIO_EXISTS, + VFIO_NOT_SUPPORTED, + VFIO_NOT_MANAGED, + VFIO_NOT_FOUND, + VFIO_NO_SPACE, }; -struct vfio_config { - int vfio_enabled; - int vfio_container_fd; - int vfio_active_groups; - const struct vfio_iommu_type *vfio_iommu_type; - struct vfio_group vfio_groups[RTE_MAX_VFIO_GROUPS]; - struct user_mem_maps mem_maps; +struct container containers[RTE_MAX_VFIO_CONTAINERS] = {0}; +struct vfio_config vfio_cfg = { + .mode = RTE_VFIO_MODE_NONE, + .default_cfg = &containers[0] }; -/* per-process VFIO config */ -static struct vfio_config vfio_cfgs[RTE_MAX_VFIO_CONTAINERS]; -static struct vfio_config *default_vfio_cfg = &vfio_cfgs[0]; - -static int vfio_type1_dma_map(int); -static int vfio_type1_dma_mem_map(int, uint64_t, uint64_t, uint64_t, int); -static int vfio_spapr_dma_map(int); -static int vfio_spapr_dma_mem_map(int, uint64_t, uint64_t, uint64_t, int); -static int vfio_noiommu_dma_map(int); -static int vfio_noiommu_dma_mem_map(int, uint64_t, uint64_t, uint64_t, int); -static int vfio_dma_mem_map(struct vfio_config *vfio_cfg, uint64_t vaddr, +static int vfio_dma_mem_map(struct container *cfg, uint64_t vaddr, uint64_t iova, uint64_t len, int do_map); -static int vfio_container_group_bind(int container_fd, int iommu_group_num); -static int vfio_container_group_unbind(int container_fd, int iommu_group_num); - -/* IOMMU types we support */ -static const struct vfio_iommu_type iommu_types[] = { - /* x86 IOMMU, otherwise known as type 1 */ - { - .type_id = VFIO_TYPE1_IOMMU, - .name = "Type 1", - .partial_unmap = false, - .dma_map_func = &vfio_type1_dma_map, - .dma_user_map_func = &vfio_type1_dma_mem_map - }, - /* ppc64 IOMMU, otherwise known as spapr */ - { - .type_id = VFIO_SPAPR_TCE_v2_IOMMU, - .name = "sPAPR", - .partial_unmap = true, - .dma_map_func = &vfio_spapr_dma_map, - .dma_user_map_func = &vfio_spapr_dma_mem_map - }, - /* IOMMU-less mode */ - { - .type_id = VFIO_NOIOMMU_IOMMU, - .name = "No-IOMMU", - .partial_unmap = true, - .dma_map_func = &vfio_noiommu_dma_map, - .dma_user_map_func = &vfio_noiommu_dma_mem_map - }, -}; - static int is_null_map(const struct user_mem_map *map) { @@ -353,279 +313,127 @@ compact_user_maps(struct user_mem_maps *user_mem_maps) sizeof(user_mem_maps->maps[0]), user_mem_map_cmp); } -static int -vfio_open_group_fd(int iommu_group_num, bool mp_request) +/* + * we will rely on kernel to not allow user to assign the same device to different containers, but + * kernel will not prevent mapping the same device twice using two different fd's, so we need to + * deduplicate our internal config to make sure we only store unique device fd's. + */ +static bool +fd_is_same(int fd1, int fd2) { - int vfio_group_fd; - char filename[PATH_MAX]; - struct rte_mp_msg mp_req, *mp_rep; - struct rte_mp_reply mp_reply = {0}; - struct timespec ts = {.tv_sec = 5, .tv_nsec = 0}; - struct vfio_mp_param *p = (struct vfio_mp_param *)mp_req.param; + struct stat st1, st2; - /* if not requesting via mp, open the group locally */ - if (!mp_request) { - /* try regular group format */ - snprintf(filename, sizeof(filename), RTE_VFIO_GROUP_FMT, iommu_group_num); - vfio_group_fd = open(filename, O_RDWR); - if (vfio_group_fd < 0) { - /* if file not found, it's not an error */ - if (errno != ENOENT) { - EAL_LOG(ERR, "Cannot open %s: %s", - filename, strerror(errno)); - return -1; - } + if (fd1 < 0 || fd2 < 0) + return false; - /* special case: try no-IOMMU path as well */ - snprintf(filename, sizeof(filename), RTE_VFIO_NOIOMMU_GROUP_FMT, - iommu_group_num); - vfio_group_fd = open(filename, O_RDWR); - if (vfio_group_fd < 0) { - if (errno != ENOENT) { - EAL_LOG(ERR, - "Cannot open %s: %s", - filename, strerror(errno)); - return -1; - } - return -ENOENT; - } - /* noiommu group found */ - } + if (fstat(fd1, &st1) < 0) + return false; + if (fstat(fd2, &st2) < 0) + return false; - return vfio_group_fd; - } - /* if we're in a secondary process, request group fd from the primary - * process via mp channel. - */ - p->req = SOCKET_REQ_GROUP; - p->group_num = iommu_group_num; - strcpy(mp_req.name, EAL_VFIO_MP); - mp_req.len_param = sizeof(*p); - mp_req.num_fds = 0; - - vfio_group_fd = -1; - if (rte_mp_request_sync(&mp_req, &mp_reply, &ts) == 0 && - mp_reply.nb_received == 1) { - mp_rep = &mp_reply.msgs[0]; - p = (struct vfio_mp_param *)mp_rep->param; - if (p->result == SOCKET_OK && mp_rep->num_fds == 1) { - vfio_group_fd = mp_rep->fds[0]; - } else if (p->result == SOCKET_NO_FD) { - EAL_LOG(ERR, "Bad VFIO group fd"); - vfio_group_fd = -ENOENT; - } - } - - free(mp_reply.msgs); - if (vfio_group_fd < 0 && vfio_group_fd != -ENOENT) - EAL_LOG(ERR, "Cannot request VFIO group fd"); - return vfio_group_fd; -} - -static struct vfio_config * -get_vfio_cfg_by_group_num(int iommu_group_num) -{ - struct vfio_config *vfio_cfg; - unsigned int i, j; - - for (i = 0; i < RTE_DIM(vfio_cfgs); i++) { - vfio_cfg = &vfio_cfgs[i]; - for (j = 0; j < RTE_DIM(vfio_cfg->vfio_groups); j++) { - if (vfio_cfg->vfio_groups[j].group_num == - iommu_group_num) - return vfio_cfg; - } - } - - return NULL; -} - -static int -vfio_get_group_fd(struct vfio_config *vfio_cfg, - int iommu_group_num) -{ - struct vfio_group *cur_grp = NULL; - int vfio_group_fd; - unsigned int i; - - /* check if we already have the group descriptor open */ - for (i = 0; i < RTE_DIM(vfio_cfg->vfio_groups); i++) - if (vfio_cfg->vfio_groups[i].group_num == iommu_group_num) - return vfio_cfg->vfio_groups[i].fd; - - /* Lets see first if there is room for a new group */ - if (vfio_cfg->vfio_active_groups == RTE_DIM(vfio_cfg->vfio_groups)) { - EAL_LOG(ERR, "Maximum number of VFIO groups reached!"); - return -1; - } - - /* Now lets get an index for the new group */ - for (i = 0; i < RTE_DIM(vfio_cfg->vfio_groups); i++) - if (vfio_cfg->vfio_groups[i].group_num == -1) { - cur_grp = &vfio_cfg->vfio_groups[i]; - break; - } - - /* This should not happen */ - if (cur_grp == NULL) { - EAL_LOG(ERR, "No VFIO group free slot found"); - return -1; - } - - /* - * When opening a group fd, we need to decide whether to open it locally - * or request it from the primary process via mp_sync. - * - * For the default container, secondary processes use mp_sync so that - * the primary process tracks the group fd and maintains VFIO state - * across all processes. - * - * For custom containers, we open the group fd locally in each process - * since custom containers are process-local and the primary has no - * knowledge of them. Requesting a group fd from the primary for a - * container it doesn't know about would be incorrect. - */ - const struct internal_config *internal_conf = eal_get_internal_configuration(); - bool mp_request = (internal_conf->process_type == RTE_PROC_SECONDARY) && - (vfio_cfg == default_vfio_cfg); - - vfio_group_fd = vfio_open_group_fd(iommu_group_num, mp_request); - if (vfio_group_fd < 0) { - EAL_LOG(ERR, "Failed to open VFIO group %d", - iommu_group_num); - return vfio_group_fd; - } - - cur_grp->group_num = iommu_group_num; - cur_grp->fd = vfio_group_fd; - vfio_cfg->vfio_active_groups++; - - return vfio_group_fd; + return st1.st_dev == st2.st_dev && st1.st_ino == st2.st_ino; } -static struct vfio_config * -get_vfio_cfg_by_group_fd(int vfio_group_fd) +bool +vfio_container_is_default(struct container *cfg) { - struct vfio_config *vfio_cfg; - unsigned int i, j; - - for (i = 0; i < RTE_DIM(vfio_cfgs); i++) { - vfio_cfg = &vfio_cfgs[i]; - for (j = 0; j < RTE_DIM(vfio_cfg->vfio_groups); j++) - if (vfio_cfg->vfio_groups[j].fd == vfio_group_fd) - return vfio_cfg; - } - - return NULL; + return cfg == vfio_cfg.default_cfg; } -static struct vfio_config * -get_vfio_cfg_by_container_fd(int container_fd) +static struct container * +vfio_container_get_by_fd(int container_fd) { - unsigned int i; + struct container *cfg; if (container_fd == RTE_VFIO_DEFAULT_CONTAINER_FD) - return default_vfio_cfg; + return vfio_cfg.default_cfg; - for (i = 0; i < RTE_DIM(vfio_cfgs); i++) { - if (vfio_cfgs[i].vfio_container_fd == container_fd) - return &vfio_cfgs[i]; + CONTAINER_FOREACH_ACTIVE(cfg) { + if (cfg->container_fd == container_fd) + return cfg; } - return NULL; } -int -vfio_get_group_fd_by_num(int iommu_group_num) +static struct container * +vfio_container_get_by_group_num(int group_num) { - struct vfio_config *vfio_cfg; - - /* get the vfio_config it belongs to */ - vfio_cfg = get_vfio_cfg_by_group_num(iommu_group_num); - vfio_cfg = vfio_cfg ? vfio_cfg : default_vfio_cfg; + struct container *cfg; + struct vfio_group *grp; - return vfio_get_group_fd(vfio_cfg, iommu_group_num); + CONTAINER_FOREACH_ACTIVE(cfg) { + GROUP_FOREACH_ACTIVE(cfg, grp) + if (grp->group_num == group_num) + return cfg; + } + return NULL; } -static int -get_vfio_group_idx(int vfio_group_fd) +static struct container * +vfio_container_create(void) { - struct vfio_config *vfio_cfg; - unsigned int i, j; - - for (i = 0; i < RTE_DIM(vfio_cfgs); i++) { - vfio_cfg = &vfio_cfgs[i]; - for (j = 0; j < RTE_DIM(vfio_cfg->vfio_groups); j++) - if (vfio_cfg->vfio_groups[j].fd == vfio_group_fd) - return j; + struct container *cfg; + + /* find an unused container config */ + CONTAINER_FOREACH(cfg) { + if (!cfg->active) { + *cfg = CONTAINER_INITIALIZER; + cfg->active = true; + return cfg; + } } - - return -1; + /* no space */ + return NULL; } static void -vfio_group_device_get(int vfio_group_fd) +vfio_container_erase(struct container *cfg) { - struct vfio_config *vfio_cfg; - int i; + if (cfg->container_fd >= 0 && close(cfg->container_fd)) + EAL_LOG(ERR, "Error when closing container, %d (%s)", errno, strerror(errno)); - vfio_cfg = get_vfio_cfg_by_group_fd(vfio_group_fd); - if (vfio_cfg == NULL) { - EAL_LOG(ERR, "Invalid VFIO group fd!"); - return; - } - - i = get_vfio_group_idx(vfio_group_fd); - if (i < 0) - EAL_LOG(ERR, "Wrong VFIO group index (%d)", i); - else - vfio_cfg->vfio_groups[i].devices++; + *cfg = (struct container){0}; } -static void -vfio_group_device_put(int vfio_group_fd) +static struct vfio_device * +vfio_device_create(struct container *cfg) { - struct vfio_config *vfio_cfg; - int i; + struct vfio_device *dev; - vfio_cfg = get_vfio_cfg_by_group_fd(vfio_group_fd); - if (vfio_cfg == NULL) { - EAL_LOG(ERR, "Invalid VFIO group fd!"); - return; - } + /* is there space? */ + if (cfg->n_devices == RTE_DIM(cfg->devices)) + return NULL; - i = get_vfio_group_idx(vfio_group_fd); - if (i < 0) - EAL_LOG(ERR, "Wrong VFIO group index (%d)", i); - else - vfio_cfg->vfio_groups[i].devices--; -} + DEVICE_FOREACH(cfg, dev) { + if (dev->active) + continue; + dev->active = true; + /* set to invalid fd */ + dev->fd = -1; -static int -vfio_group_device_count(int vfio_group_fd) -{ - struct vfio_config *vfio_cfg; - int i; - - vfio_cfg = get_vfio_cfg_by_group_fd(vfio_group_fd); - if (vfio_cfg == NULL) { - EAL_LOG(ERR, "Invalid VFIO group fd!"); - return -1; + cfg->n_devices++; + return dev; } + /* should not happen */ + EAL_LOG(WARNING, "Could not find space in device list for container"); + return NULL; +} - i = get_vfio_group_idx(vfio_group_fd); - if (i < 0) { - EAL_LOG(ERR, "Wrong VFIO group index (%d)", i); - return -1; - } +static void +vfio_device_erase(struct container *cfg, struct vfio_device *dev) +{ + if (dev->fd >= 0 && close(dev->fd)) + EAL_LOG(ERR, "Error when closing device, %d (%s)", errno, strerror(errno)); - return vfio_cfg->vfio_groups[i].devices; + *dev = (struct vfio_device){0}; + cfg->n_devices--; } static void vfio_mem_event_callback(enum rte_mem_event type, const void *addr, size_t len, void *arg __rte_unused) { + struct container *cfg = vfio_cfg.default_cfg; struct rte_memseg_list *msl; struct rte_memseg *ms; size_t cur_len = 0; @@ -640,11 +448,9 @@ vfio_mem_event_callback(enum rte_mem_event type, const void *addr, size_t len, /* Maintain granularity of DMA map/unmap to memseg size */ for (; cur_len < len; cur_len += page_sz) { if (type == RTE_MEM_EVENT_ALLOC) - vfio_dma_mem_map(default_vfio_cfg, vfio_va, - vfio_va, page_sz, 1); + vfio_dma_mem_map(cfg, vfio_va, vfio_va, page_sz, 1); else - vfio_dma_mem_map(default_vfio_cfg, vfio_va, - vfio_va, page_sz, 0); + vfio_dma_mem_map(cfg, vfio_va, vfio_va, page_sz, 0); vfio_va += page_sz; } @@ -662,11 +468,9 @@ vfio_mem_event_callback(enum rte_mem_event type, const void *addr, size_t len, goto next; } if (type == RTE_MEM_EVENT_ALLOC) - vfio_dma_mem_map(default_vfio_cfg, ms->addr_64, - ms->iova, ms->len, 1); + vfio_dma_mem_map(cfg, ms->addr_64, ms->iova, ms->len, 1); else - vfio_dma_mem_map(default_vfio_cfg, ms->addr_64, - ms->iova, ms->len, 0); + vfio_dma_mem_map(cfg, ms->addr_64, ms->iova, ms->len, 0); next: cur_len += ms->len; ++ms; @@ -674,445 +478,546 @@ vfio_mem_event_callback(enum rte_mem_event type, const void *addr, size_t len, } static int -vfio_sync_default_container(void) +vfio_register_mem_event_callback(void) { - struct rte_mp_msg mp_req, *mp_rep; - struct rte_mp_reply mp_reply = {0}; - struct timespec ts = {.tv_sec = 5, .tv_nsec = 0}; - struct vfio_mp_param *p = (struct vfio_mp_param *)mp_req.param; - int iommu_type_id; - unsigned int i; + int ret; - /* cannot be called from primary */ - if (rte_eal_process_type() != RTE_PROC_SECONDARY) - return -1; + ret = rte_mem_event_callback_register(VFIO_MEM_EVENT_CLB_NAME, + vfio_mem_event_callback, NULL); - /* default container fd should have been opened in rte_vfio_enable() */ - if (!default_vfio_cfg->vfio_enabled || - default_vfio_cfg->vfio_container_fd < 0) { - EAL_LOG(ERR, "VFIO support is not initialized"); + if (ret && rte_errno != ENOTSUP) { + EAL_LOG(ERR, "Could not install memory event callback for VFIO"); return -1; } + if (ret) + EAL_LOG(DEBUG, "Memory event callbacks not supported"); + else + EAL_LOG(DEBUG, "Installed memory event callback for VFIO"); - /* find default container's IOMMU type */ - p->req = SOCKET_REQ_IOMMU_TYPE; - strcpy(mp_req.name, EAL_VFIO_MP); - mp_req.len_param = sizeof(*p); - mp_req.num_fds = 0; - - iommu_type_id = -1; - if (rte_mp_request_sync(&mp_req, &mp_reply, &ts) == 0 && - mp_reply.nb_received == 1) { - mp_rep = &mp_reply.msgs[0]; - p = (struct vfio_mp_param *)mp_rep->param; - if (p->result == SOCKET_OK) - iommu_type_id = p->iommu_type_id; - } - free(mp_reply.msgs); - if (iommu_type_id < 0) { - EAL_LOG(ERR, - "Could not get IOMMU type for default container"); - return -1; - } - - /* we now have an fd for default container, as well as its IOMMU type. - * now, set up default VFIO container config to match. - */ - for (i = 0; i < RTE_DIM(iommu_types); i++) { - const struct vfio_iommu_type *t = &iommu_types[i]; - if (t->type_id != iommu_type_id) - continue; - - /* we found our IOMMU type */ - default_vfio_cfg->vfio_iommu_type = t; - - return 0; - } - EAL_LOG(ERR, "Could not find IOMMU type id (%i)", - iommu_type_id); - return -1; + return 0; } static int -vfio_clear_group(int vfio_group_fd) +vfio_setup_dma_mem(struct container *cfg) { - int i; - struct vfio_config *vfio_cfg; + struct user_mem_maps *user_mem_maps = &cfg->mem_maps; + int i, ret; - vfio_cfg = get_vfio_cfg_by_group_fd(vfio_group_fd); - if (vfio_cfg == NULL) { - EAL_LOG(ERR, "Invalid VFIO group fd!"); + /* do we need to map DPDK-managed memory? */ + if (vfio_container_is_default(cfg) && rte_eal_process_type() == RTE_PROC_PRIMARY) + ret = vfio_cfg.ops->dma_map_func(cfg); + else + ret = 0; + if (ret) { + EAL_LOG(ERR, "DMA remapping failed, error %i (%s)", + errno, strerror(errno)); return -1; } - i = get_vfio_group_idx(vfio_group_fd); - if (i < 0) - return -1; - vfio_cfg->vfio_groups[i].group_num = -1; - vfio_cfg->vfio_groups[i].fd = -1; - vfio_cfg->vfio_groups[i].devices = 0; - vfio_cfg->vfio_active_groups--; + /* + * not all IOMMU types support DMA mapping, but if we have mappings in the list - that + * means we have previously mapped something successfully, so we can be sure that DMA + * mapping is supported. + */ + for (i = 0; i < user_mem_maps->n_maps; i++) { + struct user_mem_map *map; + map = &user_mem_maps->maps[i]; + + ret = vfio_cfg.ops->dma_user_map_func(cfg, map->addr, map->iova, map->len, 1); + if (ret) { + EAL_LOG(ERR, "Couldn't map user memory for DMA: " + "va: 0x%" PRIx64 " " + "iova: 0x%" PRIx64 " " + "len: 0x%" PRIu64, + map->addr, map->iova, + map->len); + return -1; + } + } return 0; } +static enum vfio_result +vfio_group_assign_device(struct container *cfg, const char *sysfs_base, + const char *dev_addr, struct vfio_device **out_dev) +{ + struct vfio_group_config *group_cfg = &cfg->group_cfg; + struct vfio_group *grp; + struct vfio_device *idev, *dev; + int iommu_group_num; + enum vfio_result res; + int ret; + + /* allocate new device in config */ + dev = vfio_device_create(cfg); + if (dev == NULL) { + EAL_LOG(ERR, "No space to track new VFIO device"); + return VFIO_NO_SPACE; + } + + /* remember to register mem event callback for default container in primary */ + bool need_clb = vfio_container_is_default(cfg) && + rte_eal_process_type() == RTE_PROC_PRIMARY; + + /* get group number for this device */ + ret = vfio_group_get_num(sysfs_base, dev_addr, &iommu_group_num); + if (ret < 0) { + EAL_LOG(ERR, "Cannot get IOMMU group for %s", dev_addr); + res = VFIO_ERROR; + goto device_erase; + } else if (ret == 0) { + res = VFIO_NOT_MANAGED; + goto device_erase; + } + + /* group may already exist as multiple devices may share group */ + grp = vfio_group_get_by_num(cfg, iommu_group_num); + if (grp == NULL) { + /* no device currently uses this group, create it */ + grp = vfio_group_create(cfg, iommu_group_num); + if (grp == NULL) { + EAL_LOG(ERR, "Cannot allocate group for device %s", dev_addr); + res = VFIO_NO_SPACE; + goto device_erase; + } + + /* open group fd */ + ret = vfio_group_open_fd(cfg, grp); + if (ret == -ENOENT) { + EAL_LOG(DEBUG, "Device %s (IOMMU group %d) not managed by VFIO", + dev_addr, iommu_group_num); + res = VFIO_NOT_MANAGED; + goto group_erase; + } else if (ret < 0) { + EAL_LOG(ERR, "Cannot open VFIO group %d for device %s", + iommu_group_num, dev_addr); + res = VFIO_ERROR; + goto group_erase; + } + + /* prepare group (viability + container attach) */ + ret = vfio_group_prepare(cfg, grp); + if (ret < 0) { + res = VFIO_ERROR; + goto group_erase; + } + + /* set up IOMMU type once per container */ + if (!group_cfg->iommu_type_set) { + ret = vfio_group_setup_iommu(cfg); + if (ret < 0) { + res = VFIO_ERROR; + goto group_erase; + } + group_cfg->iommu_type_set = true; + } + + /* set up DMA memory once per container */ + if (!group_cfg->dma_setup_done) { + rte_spinlock_recursive_lock(&cfg->mem_maps.lock); + ret = vfio_setup_dma_mem(cfg); + rte_spinlock_recursive_unlock(&cfg->mem_maps.lock); + if (ret < 0) { + EAL_LOG(ERR, "DMA remapping for %s failed", dev_addr); + res = VFIO_ERROR; + goto group_erase; + } + group_cfg->dma_setup_done = true; + } + + /* set up mem event callback if needed */ + if (need_clb && !group_cfg->mem_event_clb_set) { + ret = vfio_register_mem_event_callback(); + if (ret < 0) { + res = VFIO_ERROR; + goto group_erase; + } + group_cfg->mem_event_clb_set = true; + } + } + + /* open dev fd */ + ret = vfio_group_setup_device_fd(dev_addr, grp, dev); + if (ret < 0) { + EAL_LOG(ERR, "Cannot open VFIO device %s, error %i (%s)", + dev_addr, errno, strerror(errno)); + res = VFIO_ERROR; + goto group_erase; + } + + /* + * we want to prevent user from assigning devices twice to prevent resource leaks, but for + * group mode this is not trivial, as there is no direct way to know which fd belongs to + * which group/device, except for directly comparing fd's with stat. so, that's what we're + * going to do. we do not need to look in other configs as if we were to attempt to use a + * different container, the kernel wouldn't have allowed us to bind the group to the + * container in the first place. + */ + DEVICE_FOREACH_ACTIVE(cfg, idev) { + if (idev != dev && fd_is_same(idev->fd, dev->fd)) { + EAL_LOG(ERR, "Device %s already assigned to this container", + dev_addr); + res = VFIO_EXISTS; + *out_dev = idev; + goto dev_remove; + } + } + *out_dev = dev; + return VFIO_SUCCESS; +dev_remove: + /* device will be closed, but we still need to keep the group consistent */ + grp->n_devices--; +group_erase: + /* this may be a pre-existing group so only erase it if it has no devices */ + if (grp->n_devices == 0) + vfio_group_erase(cfg, grp); + /* if we registered callback, unregister it */ + if (group_cfg->n_groups == 0 && group_cfg->mem_event_clb_set) { + rte_mem_event_callback_unregister(VFIO_MEM_EVENT_CLB_NAME, NULL); + group_cfg->mem_event_clb_set = false; + } +device_erase: + vfio_device_erase(cfg, dev); + return res; +} + +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_assign_device) +int +rte_vfio_container_assign_device(int container_fd, const char *sysfs_base, const char *dev_addr) +{ + struct container *cfg; + enum vfio_result res; + struct vfio_device *dev; + + if (sysfs_base == NULL || dev_addr == NULL) { + rte_errno = EINVAL; + return -1; + } + + if (vfio_cfg.mode == RTE_VFIO_MODE_NONE) { + EAL_LOG(ERR, "VFIO support not initialized"); + rte_errno = ENXIO; + return -1; + } + + cfg = vfio_container_get_by_fd(container_fd); + if (cfg == NULL) { + EAL_LOG(ERR, "Invalid VFIO container fd"); + rte_errno = EINVAL; + return -1; + } + /* protect memory configuration while setting up IOMMU/DMA */ + rte_mcfg_mem_read_lock(); + + switch (vfio_cfg.mode) { + case RTE_VFIO_MODE_GROUP: + case RTE_VFIO_MODE_NOIOMMU: + res = vfio_group_assign_device(cfg, sysfs_base, dev_addr, &dev); + break; + default: + EAL_LOG(ERR, "Unsupported VFIO mode"); + res = VFIO_NOT_SUPPORTED; + break; + } + rte_mcfg_mem_read_unlock(); + + switch (res) { + case VFIO_SUCCESS: + return 0; + case VFIO_EXISTS: + rte_errno = EEXIST; + return -1; + case VFIO_NOT_MANAGED: + EAL_LOG(DEBUG, "Device %s not managed by VFIO", dev_addr); + rte_errno = ENODEV; + return -1; + case VFIO_NO_SPACE: + EAL_LOG(ERR, "No space in VFIO container to assign device %s", dev_addr); + rte_errno = ENOSPC; + return -1; + default: + EAL_LOG(ERR, "Error assigning device %s to container", dev_addr); + rte_errno = EIO; + return -1; + } +} + RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_setup_device) int rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, int *vfio_dev_fd) { - struct vfio_group_status group_status = { - .argsz = sizeof(group_status) - }; - struct vfio_config *vfio_cfg; - struct user_mem_maps *user_mem_maps; - int vfio_container_fd; - int vfio_group_fd; - int iommu_group_num; - rte_uuid_t vf_token; - int i, ret; - const struct internal_config *internal_conf = - eal_get_internal_configuration(); - - /* get group number */ - ret = rte_vfio_get_group_num(sysfs_base, dev_addr, &iommu_group_num); - if (ret == 0) { - EAL_LOG(NOTICE, - "%s not managed by VFIO driver, skipping", - dev_addr); - return 1; - } - - /* if negative, something failed */ - if (ret < 0) - return -1; - - /* get the actual group fd */ - vfio_group_fd = vfio_get_group_fd_by_num(iommu_group_num); - if (vfio_group_fd < 0 && vfio_group_fd != -ENOENT) - return -1; - - /* - * if vfio_group_fd == -ENOENT, that means the device - * isn't managed by VFIO - */ - if (vfio_group_fd == -ENOENT) { - EAL_LOG(NOTICE, - "%s not managed by VFIO driver, skipping", - dev_addr); - return 1; - } - - /* - * at this point, we know that this group is viable (meaning, all devices - * are either bound to VFIO or not bound to anything) - */ - - /* check if the group is viable */ - ret = ioctl(vfio_group_fd, VFIO_GROUP_GET_STATUS, &group_status); - if (ret) { - EAL_LOG(ERR, "%s cannot get VFIO group status, " - "error %i (%s)", dev_addr, errno, strerror(errno)); - close(vfio_group_fd); - vfio_clear_group(vfio_group_fd); - return -1; - } else if (!(group_status.flags & VFIO_GROUP_FLAGS_VIABLE)) { - EAL_LOG(ERR, "%s VFIO group is not viable! " - "Not all devices in IOMMU group bound to VFIO or unbound", - dev_addr); - close(vfio_group_fd); - vfio_clear_group(vfio_group_fd); - return -1; - } - - /* get the vfio_config it belongs to */ - vfio_cfg = get_vfio_cfg_by_group_num(iommu_group_num); - vfio_cfg = vfio_cfg ? vfio_cfg : default_vfio_cfg; - vfio_container_fd = vfio_cfg->vfio_container_fd; - user_mem_maps = &vfio_cfg->mem_maps; - - /* check if group does not have a container yet */ - if (!(group_status.flags & VFIO_GROUP_FLAGS_CONTAINER_SET)) { - - /* add group to a container */ - ret = ioctl(vfio_group_fd, VFIO_GROUP_SET_CONTAINER, - &vfio_container_fd); - if (ret) { - EAL_LOG(ERR, - "%s cannot add VFIO group to container, error " - "%i (%s)", dev_addr, errno, strerror(errno)); - close(vfio_group_fd); - vfio_clear_group(vfio_group_fd); - return -1; - } - - /* - * pick an IOMMU type and set up DMA mappings for container - * - * needs to be done only once, only when first group is - * assigned to a container and only in primary process. - * Note this can happen several times with the hotplug - * functionality. - */ - if (internal_conf->process_type == RTE_PROC_PRIMARY && - vfio_cfg->vfio_active_groups == 1 && - vfio_group_device_count(vfio_group_fd) == 0) { - const struct vfio_iommu_type *t; - - /* select an IOMMU type which we will be using */ - t = vfio_set_iommu_type(vfio_container_fd); - if (!t) { - EAL_LOG(ERR, - "%s failed to select IOMMU type", - dev_addr); - close(vfio_group_fd); - vfio_clear_group(vfio_group_fd); - return -1; - } - /* lock memory hotplug before mapping and release it - * after registering callback, to prevent races - */ - rte_mcfg_mem_read_lock(); - if (vfio_cfg == default_vfio_cfg) - ret = t->dma_map_func(vfio_container_fd); - else - ret = 0; - if (ret) { - EAL_LOG(ERR, - "%s DMA remapping failed, error " - "%i (%s)", - dev_addr, errno, strerror(errno)); - close(vfio_group_fd); - vfio_clear_group(vfio_group_fd); - rte_mcfg_mem_read_unlock(); - return -1; - } - - vfio_cfg->vfio_iommu_type = t; - - /* re-map all user-mapped segments */ - rte_spinlock_recursive_lock(&user_mem_maps->lock); - - /* this IOMMU type may not support DMA mapping, but - * if we have mappings in the list - that means we have - * previously mapped something successfully, so we can - * be sure that DMA mapping is supported. - */ - for (i = 0; i < user_mem_maps->n_maps; i++) { - struct user_mem_map *map; - map = &user_mem_maps->maps[i]; - - ret = t->dma_user_map_func( - vfio_container_fd, - map->addr, map->iova, map->len, - 1); - if (ret) { - EAL_LOG(ERR, "Couldn't map user memory for DMA: " - "va: 0x%" PRIx64 " " - "iova: 0x%" PRIx64 " " - "len: 0x%" PRIu64, - map->addr, map->iova, - map->len); - rte_spinlock_recursive_unlock( - &user_mem_maps->lock); - rte_mcfg_mem_read_unlock(); - return -1; - } - } - rte_spinlock_recursive_unlock(&user_mem_maps->lock); - - /* register callback for mem events */ - if (vfio_cfg == default_vfio_cfg) - ret = rte_mem_event_callback_register( - VFIO_MEM_EVENT_CLB_NAME, - vfio_mem_event_callback, NULL); - else - ret = 0; - /* unlock memory hotplug */ - rte_mcfg_mem_read_unlock(); - - if (ret && rte_errno != ENOTSUP) { - EAL_LOG(ERR, "Could not install memory event callback for VFIO"); - return -1; - } - if (ret) - EAL_LOG(DEBUG, "Memory event callbacks not supported"); - else - EAL_LOG(DEBUG, "Installed memory event callback for VFIO"); - } - } else if (rte_eal_process_type() != RTE_PROC_PRIMARY && - vfio_cfg == default_vfio_cfg && - vfio_cfg->vfio_iommu_type == NULL) { - /* if we're not a primary process, we do not set up the VFIO - * container because it's already been set up by the primary - * process. instead, we simply ask the primary about VFIO type - * we are using, and set the VFIO config up appropriately. - */ - ret = vfio_sync_default_container(); - if (ret < 0) { - EAL_LOG(ERR, "Could not sync default VFIO container"); - close(vfio_group_fd); - vfio_clear_group(vfio_group_fd); - return -1; - } - /* we have successfully initialized VFIO, notify user */ - const struct vfio_iommu_type *t = - default_vfio_cfg->vfio_iommu_type; - EAL_LOG(INFO, "Using IOMMU type %d (%s)", - t->type_id, t->name); - } - - rte_eal_vfio_get_vf_token(vf_token); - - /* get a file descriptor for the device with VF token firstly */ - if (!rte_uuid_is_null(vf_token)) { - char vf_token_str[RTE_UUID_STRLEN]; - char dev[PATH_MAX]; - - rte_uuid_unparse(vf_token, vf_token_str, sizeof(vf_token_str)); - snprintf(dev, sizeof(dev), - "%s vf_token=%s", dev_addr, vf_token_str); - - *vfio_dev_fd = ioctl(vfio_group_fd, VFIO_GROUP_GET_DEVICE_FD, - dev); - if (*vfio_dev_fd >= 0) - goto out; - } - - /* get a file descriptor for the device */ - *vfio_dev_fd = ioctl(vfio_group_fd, VFIO_GROUP_GET_DEVICE_FD, dev_addr); - if (*vfio_dev_fd < 0) { - /* if we cannot get a device fd, this implies a problem with - * the VFIO group or the container not having IOMMU configured. - */ - - EAL_LOG(WARNING, "Getting a vfio_dev_fd for %s failed", - dev_addr); - close(vfio_group_fd); - vfio_clear_group(vfio_group_fd); - return -1; - } - - /* device is now set up */ -out: - vfio_group_device_get(vfio_group_fd); - - return 0; -} - -RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_release_device) -int -rte_vfio_release_device(const char *sysfs_base, const char *dev_addr, - int vfio_dev_fd) -{ - struct vfio_config *vfio_cfg; - int vfio_group_fd; - int iommu_group_num; + struct container *cfg; + struct vfio_device *dev; + enum vfio_result res; int ret; - /* we don't want any DMA mapping messages to come while we're detaching - * VFIO device, because this might be the last device and we might need - * to unregister the callback. - */ + if (sysfs_base == NULL || dev_addr == NULL || vfio_dev_fd == NULL) { + rte_errno = EINVAL; + return -1; + } + + if (vfio_cfg.mode == RTE_VFIO_MODE_NONE) { + EAL_LOG(ERR, "VFIO support not initialized"); + rte_errno = ENXIO; + return -1; + } + rte_mcfg_mem_read_lock(); - /* get group number */ - ret = rte_vfio_get_group_num(sysfs_base, dev_addr, &iommu_group_num); - if (ret <= 0) { - EAL_LOG(WARNING, "%s not managed by VFIO driver", - dev_addr); - /* This is an error at this point. */ - ret = -1; - goto out; - } - - /* get the actual group fd */ - vfio_group_fd = vfio_get_group_fd_by_num(iommu_group_num); - if (vfio_group_fd < 0) { - EAL_LOG(INFO, "vfio_get_group_fd_by_num failed for %s", - dev_addr); - ret = vfio_group_fd; - goto out; + switch (vfio_cfg.mode) { + case RTE_VFIO_MODE_GROUP: + case RTE_VFIO_MODE_NOIOMMU: + { + int iommu_group_num; + + /* find group number */ + ret = vfio_group_get_num(sysfs_base, dev_addr, &iommu_group_num); + if (ret < 0) + goto assign_fail; + else if (ret == 0) + goto not_managed; + + /* find config by group */ + cfg = vfio_container_get_by_group_num(iommu_group_num); + if (cfg == NULL) + cfg = vfio_cfg.default_cfg; + + res = vfio_group_assign_device(cfg, sysfs_base, dev_addr, &dev); + break; } - - /* get the vfio_config it belongs to */ - vfio_cfg = get_vfio_cfg_by_group_num(iommu_group_num); - vfio_cfg = vfio_cfg ? vfio_cfg : default_vfio_cfg; - - /* At this point we got an active group. Closing it will make the - * container detachment. If this is the last active group, VFIO kernel - * code will unset the container and the IOMMU mappings. - */ - - /* Closing a device */ - if (close(vfio_dev_fd) < 0) { - EAL_LOG(INFO, "Error when closing vfio_dev_fd for %s", - dev_addr); + default: + EAL_LOG(ERR, "Unsupported VFIO mode"); + rte_errno = ENOTSUP; ret = -1; - goto out; + goto unlock; } - /* An VFIO group can have several devices attached. Just when there is - * no devices remaining should the group be closed. - */ - vfio_group_device_put(vfio_group_fd); - if (!vfio_group_device_count(vfio_group_fd)) { - - if (close(vfio_group_fd) < 0) { - EAL_LOG(INFO, "Error when closing vfio_group_fd for %s", - dev_addr); - ret = -1; - goto out; - } - - if (vfio_clear_group(vfio_group_fd) < 0) { - EAL_LOG(INFO, "Error when clearing group for %s", - dev_addr); - ret = -1; - goto out; - } + switch (res) { + case VFIO_NOT_MANAGED: +not_managed: + EAL_LOG(DEBUG, "Device %s not managed by VFIO", dev_addr); + rte_errno = ENODEV; + ret = -1; + goto unlock; + case VFIO_SUCCESS: + case VFIO_EXISTS: + break; + case VFIO_NO_SPACE: + EAL_LOG(ERR, "No space in VFIO container to assign device %s", dev_addr); + rte_errno = ENOSPC; + ret = -1; + goto unlock; + default: +assign_fail: + EAL_LOG(ERR, "Error assigning device %s to container", dev_addr); + rte_errno = EIO; + ret = -1; + goto unlock; } - - /* if there are no active device groups, unregister the callback to - * avoid spurious attempts to map/unmap memory from VFIO. - */ - if (vfio_cfg == default_vfio_cfg && vfio_cfg->vfio_active_groups == 0 && - rte_eal_process_type() != RTE_PROC_SECONDARY) - rte_mem_event_callback_unregister(VFIO_MEM_EVENT_CLB_NAME, - NULL); + *vfio_dev_fd = dev->fd; /* success */ ret = 0; -out: +unlock: rte_mcfg_mem_read_unlock(); + return ret; } +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_release_device) +int +rte_vfio_release_device(const char *sysfs_base __rte_unused, + const char *dev_addr, int vfio_dev_fd) +{ + struct container *cfg = NULL, *icfg; + struct vfio_device *dev = NULL, *idev; + int ret; + + if (sysfs_base == NULL || dev_addr == NULL) { + rte_errno = EINVAL; + return -1; + } + + if (vfio_cfg.mode == RTE_VFIO_MODE_NONE) { + EAL_LOG(ERR, "VFIO support not initialized"); + rte_errno = ENXIO; + return -1; + } + + rte_mcfg_mem_read_lock(); + + /* we need to find both config and device */ + CONTAINER_FOREACH_ACTIVE(icfg) { + DEVICE_FOREACH_ACTIVE(icfg, idev) { + if (idev->fd != vfio_dev_fd) + continue; + cfg = icfg; + dev = idev; + goto found; + } + } +found: + if (dev == NULL) { + EAL_LOG(ERR, "Device %s not managed by any container", dev_addr); + rte_errno = ENOENT; + ret = -1; + goto unlock; + } + + switch (vfio_cfg.mode) { + case RTE_VFIO_MODE_GROUP: + case RTE_VFIO_MODE_NOIOMMU: + { + int iommu_group_num = dev->group; + struct vfio_group_config *group_cfg = &cfg->group_cfg; + struct vfio_group *grp; + + bool need_clb = vfio_container_is_default(cfg) && + rte_eal_process_type() == RTE_PROC_PRIMARY; + + /* find the group */ + grp = vfio_group_get_by_num(cfg, iommu_group_num); + if (grp == NULL) { + /* shouldn't happen because we already know the device is valid */ + EAL_LOG(ERR, "IOMMU group %d not found in container", + iommu_group_num); + rte_errno = EIO; + ret = -1; + goto unlock; + } + + /* close device handle */ + vfio_device_erase(cfg, dev); + + /* remove device from group */ + grp->n_devices--; + + /* was this the last device? */ + if (grp->n_devices == 0) + vfio_group_erase(cfg, grp); + + /* if no more groups left, remove callback */ + if (need_clb && group_cfg->n_groups == 0 && group_cfg->mem_event_clb_set) { + rte_mem_event_callback_unregister(VFIO_MEM_EVENT_CLB_NAME, NULL); + group_cfg->mem_event_clb_set = false; + } + break; + } + default: + EAL_LOG(ERR, "Unsupported VFIO mode"); + rte_errno = ENOTSUP; + ret = -1; + goto unlock; + } + ret = 0; +unlock: + rte_mcfg_mem_read_unlock(); + + return ret; +} + +static int +vfio_sync_mode(struct container *cfg, enum rte_vfio_mode *mode) +{ + struct vfio_mp_param *p; + struct rte_mp_msg mp_req = {0}; + struct rte_mp_reply mp_reply = {0}; + struct timespec ts = {5, 0}; + + /* request iommufd from primary via mp_sync */ + rte_strscpy(mp_req.name, EAL_VFIO_MP, sizeof(mp_req.name)); + mp_req.len_param = sizeof(*p); + mp_req.num_fds = 0; + p = (struct vfio_mp_param *)mp_req.param; + p->req = SOCKET_REQ_CONTAINER; + + if (rte_mp_request_sync(&mp_req, &mp_reply, &ts) == 0 && + mp_reply.nb_received == 1) { + struct rte_mp_msg *mp_rep; + mp_rep = &mp_reply.msgs[0]; + p = (struct vfio_mp_param *)mp_rep->param; + if (p->result == SOCKET_OK && mp_rep->num_fds == 1) { + cfg->container_fd = mp_rep->fds[0]; + *mode = p->mode; + free(mp_reply.msgs); + return 0; + } + } + + free(mp_reply.msgs); + EAL_LOG(ERR, "Cannot request container_fd"); + return -1; +} + +static enum rte_vfio_mode +vfio_select_mode(void) +{ + struct container *cfg; + enum rte_vfio_mode mode = RTE_VFIO_MODE_NONE; + + cfg = vfio_container_create(); + /* cannot happen */ + if (cfg == NULL || cfg != vfio_cfg.default_cfg) { + EAL_LOG(ERR, "Unexpected VFIO config structure"); + return RTE_VFIO_MODE_NONE; + } + + /* for secondary, just ask the primary for the container and mode */ + if (rte_eal_process_type() != RTE_PROC_PRIMARY) { + struct vfio_group_config *group_cfg = &cfg->group_cfg; + + if (vfio_sync_mode(cfg, &mode) < 0) + goto err; + + /* primary handles DMA setup for default containers */ + group_cfg->dma_setup_done = true; + return mode; + } + /* if we failed mp sync setup, we cannot initialize VFIO */ + if (vfio_mp_sync_setup() < 0) + return RTE_VFIO_MODE_NONE; + + /* try group mode first */ + if (vfio_group_enable(cfg) == 0) { + /* check for noiommu */ + int ret = vfio_group_noiommu_is_enabled(); + if (ret < 0) + goto err_mpsync; + else if (ret == 1) + return RTE_VFIO_MODE_NOIOMMU; + return RTE_VFIO_MODE_GROUP; + } +err_mpsync: + vfio_mp_sync_cleanup(); +err: + vfio_container_erase(cfg); + + return RTE_VFIO_MODE_NONE; +} + +static const char * +vfio_mode_to_str(enum rte_vfio_mode mode) +{ + switch (mode) { + case RTE_VFIO_MODE_GROUP: return "group"; + case RTE_VFIO_MODE_NOIOMMU: return "noiommu"; + default: return "not initialized"; + } +} + RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_enable) int rte_vfio_enable(const char *modname) { - /* initialize group list */ - unsigned int i, j; int vfio_available; - DIR *dir; - const struct internal_config *internal_conf = - eal_get_internal_configuration(); + enum rte_vfio_mode mode = RTE_VFIO_MODE_NONE; - rte_spinlock_recursive_t lock = RTE_SPINLOCK_RECURSIVE_INITIALIZER; - - for (i = 0; i < RTE_DIM(vfio_cfgs); i++) { - vfio_cfgs[i].vfio_container_fd = -1; - vfio_cfgs[i].vfio_active_groups = 0; - vfio_cfgs[i].vfio_iommu_type = NULL; - vfio_cfgs[i].mem_maps.lock = lock; - - for (j = 0; j < RTE_DIM(vfio_cfgs[i].vfio_groups); j++) { - vfio_cfgs[i].vfio_groups[j].fd = -1; - vfio_cfgs[i].vfio_groups[j].group_num = -1; - vfio_cfgs[i].vfio_groups[j].devices = 0; - } + if (modname == NULL) { + rte_errno = EINVAL; + return -1; } EAL_LOG(DEBUG, "Probing VFIO support..."); @@ -1132,36 +1037,16 @@ rte_vfio_enable(const char *modname) "VFIO modules not loaded, skipping VFIO support..."); return 0; } + EAL_LOG(DEBUG, "VFIO module '%s' loaded, attempting to initialize VFIO...", modname); + mode = vfio_select_mode(); - /* VFIO directory might not exist (e.g., unprivileged containers) */ - dir = opendir(RTE_VFIO_DIR); - if (dir == NULL) { - EAL_LOG(DEBUG, - "VFIO directory does not exist, skipping VFIO support..."); - return 0; - } - closedir(dir); - - if (internal_conf->process_type == RTE_PROC_PRIMARY) { - if (vfio_mp_sync_setup() == -1) { - default_vfio_cfg->vfio_container_fd = -1; - } else { - /* open a default container */ - default_vfio_cfg->vfio_container_fd = vfio_open_container_fd(false); - } - } else { - /* get the default container from the primary process */ - default_vfio_cfg->vfio_container_fd = - vfio_open_container_fd(true); - } - - /* check if we have VFIO driver enabled */ - if (default_vfio_cfg->vfio_container_fd != -1) { - EAL_LOG(INFO, "VFIO support initialized"); - default_vfio_cfg->vfio_enabled = 1; - } else { + /* have we initialized anything? */ + if (mode == RTE_VFIO_MODE_NONE) EAL_LOG(NOTICE, "VFIO support could not be initialized"); - } + else + EAL_LOG(NOTICE, "VFIO support initialized: %s mode", vfio_mode_to_str(mode)); + + vfio_cfg.mode = mode; return 0; } @@ -1170,40 +1055,17 @@ RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_is_enabled) int rte_vfio_is_enabled(const char *modname) { - const int mod_available = rte_eal_check_module(modname) > 0; - return default_vfio_cfg->vfio_enabled && mod_available; + const int mod_available = modname ? rte_eal_check_module(modname) > 0 : 0; + return vfio_cfg.default_cfg->active && mod_available; } int vfio_get_iommu_type(void) { - if (default_vfio_cfg->vfio_iommu_type == NULL) + if (vfio_cfg.ops == NULL) return -1; - return default_vfio_cfg->vfio_iommu_type->type_id; -} - -const struct vfio_iommu_type * -vfio_set_iommu_type(int vfio_container_fd) -{ - unsigned idx; - for (idx = 0; idx < RTE_DIM(iommu_types); idx++) { - const struct vfio_iommu_type *t = &iommu_types[idx]; - - int ret = ioctl(vfio_container_fd, VFIO_SET_IOMMU, - t->type_id); - if (!ret) { - EAL_LOG(INFO, "Using IOMMU type %d (%s)", - t->type_id, t->name); - return t; - } - /* not an error, there may be more supported IOMMU types */ - EAL_LOG(DEBUG, "Set IOMMU type %d (%s) failed, error " - "%i (%s)", t->type_id, t->name, errno, - strerror(errno)); - } - /* if we didn't find a suitable IOMMU type, fail */ - return NULL; + return vfio_cfg.ops->type_id; } RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_device_info) @@ -1212,126 +1074,27 @@ rte_vfio_get_device_info(int vfio_dev_fd, struct vfio_device_info *device_info) { int ret; - if (device_info == NULL || vfio_dev_fd < 0) + if (device_info == NULL) { + rte_errno = EINVAL; return -1; + } + + if (vfio_cfg.mode == RTE_VFIO_MODE_NONE) { + EAL_LOG(ERR, "VFIO support not initialized"); + rte_errno = ENXIO; + return -1; + } ret = ioctl(vfio_dev_fd, VFIO_DEVICE_GET_INFO, device_info); if (ret) { - EAL_LOG(ERR, "Cannot get device info, error %i (%s)", - errno, strerror(errno)); + EAL_LOG(ERR, "Cannot get device info, error %d (%s)", errno, strerror(errno)); + rte_errno = errno; return -1; } return 0; } -int -vfio_has_supported_extensions(int vfio_container_fd) -{ - int ret; - unsigned idx, n_extensions = 0; - for (idx = 0; idx < RTE_DIM(iommu_types); idx++) { - const struct vfio_iommu_type *t = &iommu_types[idx]; - - ret = ioctl(vfio_container_fd, VFIO_CHECK_EXTENSION, - t->type_id); - if (ret < 0) { - EAL_LOG(ERR, "Could not get IOMMU type, error " - "%i (%s)", errno, strerror(errno)); - close(vfio_container_fd); - return -1; - } else if (ret == 1) { - /* we found a supported extension */ - n_extensions++; - } - EAL_LOG(DEBUG, "IOMMU type %d (%s) is %s", - t->type_id, t->name, - ret ? "supported" : "not supported"); - } - - /* if we didn't find any supported IOMMU types, fail */ - if (!n_extensions) { - close(vfio_container_fd); - return -1; - } - - return 0; -} - -/* - * Open a new VFIO container fd. - * - * If mp_request is true, requests a new container fd from the primary process - * via mp channel (for secondary processes that need to open the default container). - * - * Otherwise, opens a new container fd locally by opening /dev/vfio/vfio. - */ -int -vfio_open_container_fd(bool mp_request) -{ - int ret, vfio_container_fd; - struct rte_mp_msg mp_req, *mp_rep; - struct rte_mp_reply mp_reply = {0}; - struct timespec ts = {.tv_sec = 5, .tv_nsec = 0}; - struct vfio_mp_param *p = (struct vfio_mp_param *)mp_req.param; - - /* if not requesting via mp, open a new container locally */ - if (!mp_request) { - vfio_container_fd = open(RTE_VFIO_CONTAINER_PATH, O_RDWR); - if (vfio_container_fd < 0) { - EAL_LOG(ERR, "Cannot open VFIO container %s, error %i (%s)", - RTE_VFIO_CONTAINER_PATH, errno, strerror(errno)); - return -1; - } - - /* check VFIO API version */ - ret = ioctl(vfio_container_fd, VFIO_GET_API_VERSION); - if (ret != VFIO_API_VERSION) { - if (ret < 0) - EAL_LOG(ERR, - "Could not get VFIO API version, error " - "%i (%s)", errno, strerror(errno)); - else - EAL_LOG(ERR, "Unsupported VFIO API version!"); - close(vfio_container_fd); - return -1; - } - - ret = vfio_has_supported_extensions(vfio_container_fd); - if (ret) { - EAL_LOG(ERR, - "No supported IOMMU extensions found!"); - return -1; - } - - return vfio_container_fd; - } - /* - * if we're in a secondary process, request container fd from the - * primary process via mp channel - */ - p->req = SOCKET_REQ_CONTAINER; - strcpy(mp_req.name, EAL_VFIO_MP); - mp_req.len_param = sizeof(*p); - mp_req.num_fds = 0; - - vfio_container_fd = -1; - if (rte_mp_request_sync(&mp_req, &mp_reply, &ts) == 0 && - mp_reply.nb_received == 1) { - mp_rep = &mp_reply.msgs[0]; - p = (struct vfio_mp_param *)mp_rep->param; - if (p->result == SOCKET_OK && mp_rep->num_fds == 1) { - vfio_container_fd = mp_rep->fds[0]; - free(mp_reply.msgs); - return vfio_container_fd; - } - } - - free(mp_reply.msgs); - EAL_LOG(ERR, "Cannot request VFIO container fd"); - return -1; -} - RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_container_fd) int rte_vfio_get_container_fd(void) @@ -1340,511 +1103,54 @@ rte_vfio_get_container_fd(void) * The default container is set up during rte_vfio_enable(). * This function does not create a new container. */ - if (!default_vfio_cfg->vfio_enabled) - return -1; + if (vfio_cfg.mode != RTE_VFIO_MODE_NONE) + return vfio_cfg.default_cfg->container_fd; - return default_vfio_cfg->vfio_container_fd; + EAL_LOG(ERR, "VFIO support not initialized"); + rte_errno = ENXIO; + return -1; } RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_group_num) int -rte_vfio_get_group_num(const char *sysfs_base, - const char *dev_addr, int *iommu_group_num) +rte_vfio_get_group_num(const char *sysfs_base, const char *dev_addr, int *iommu_group_num) { - char linkname[PATH_MAX]; - char filename[PATH_MAX]; - char *tok[16], *group_tok, *end; int ret; - memset(linkname, 0, sizeof(linkname)); - memset(filename, 0, sizeof(filename)); - - /* try to find out IOMMU group for this device */ - snprintf(linkname, sizeof(linkname), - "%s/%s/iommu_group", sysfs_base, dev_addr); - - ret = readlink(linkname, filename, sizeof(filename)); - - /* if the link doesn't exist, no VFIO for us */ - if (ret < 0) - return 0; - - ret = rte_strsplit(filename, sizeof(filename), - tok, RTE_DIM(tok), '/'); - - if (ret <= 0) { - EAL_LOG(ERR, "%s cannot get IOMMU group", dev_addr); - return -1; - } - - /* IOMMU group is always the last token */ - errno = 0; - group_tok = tok[ret - 1]; - end = group_tok; - *iommu_group_num = strtol(group_tok, &end, 10); - if ((end != group_tok && *end != '\0') || errno != 0) { - EAL_LOG(ERR, "%s error parsing IOMMU number!", dev_addr); - return -1; - } - - return 1; -} - -static int -type1_map(const struct rte_memseg_list *msl, const struct rte_memseg *ms, - void *arg) -{ - int *vfio_container_fd = arg; - - /* skip external memory that isn't a heap */ - if (msl->external && !msl->heap) - return 0; - - /* skip any segments with invalid IOVA addresses */ - if (ms->iova == RTE_BAD_IOVA) - return 0; - - return vfio_type1_dma_mem_map(*vfio_container_fd, ms->addr_64, ms->iova, - ms->len, 1); -} - -static int -vfio_type1_dma_mem_map(int vfio_container_fd, uint64_t vaddr, uint64_t iova, - uint64_t len, int do_map) -{ - struct vfio_iommu_type1_dma_map dma_map; - struct vfio_iommu_type1_dma_unmap dma_unmap; - int ret; - - if (do_map != 0) { - memset(&dma_map, 0, sizeof(dma_map)); - dma_map.argsz = sizeof(struct vfio_iommu_type1_dma_map); - dma_map.vaddr = vaddr; - dma_map.size = len; - dma_map.iova = iova; - dma_map.flags = VFIO_DMA_MAP_FLAG_READ | - VFIO_DMA_MAP_FLAG_WRITE; - - ret = ioctl(vfio_container_fd, VFIO_IOMMU_MAP_DMA, &dma_map); - if (ret) { - /** - * In case the mapping was already done EEXIST will be - * returned from kernel. - */ - if (errno == EEXIST) { - EAL_LOG(DEBUG, - "Memory segment is already mapped, skipping"); - } else { - EAL_LOG(ERR, - "Cannot set up DMA remapping, error " - "%i (%s)", errno, strerror(errno)); - return -1; - } - } - } else { - memset(&dma_unmap, 0, sizeof(dma_unmap)); - dma_unmap.argsz = sizeof(struct vfio_iommu_type1_dma_unmap); - dma_unmap.size = len; - dma_unmap.iova = iova; - - ret = ioctl(vfio_container_fd, VFIO_IOMMU_UNMAP_DMA, - &dma_unmap); - if (ret) { - EAL_LOG(ERR, "Cannot clear DMA remapping, error " - "%i (%s)", errno, strerror(errno)); - return -1; - } else if (dma_unmap.size != len) { - EAL_LOG(ERR, "Unexpected size %"PRIu64 - " of DMA remapping cleared instead of %"PRIu64, - (uint64_t)dma_unmap.size, len); - rte_errno = EIO; - return -1; - } - } - - return 0; -} - -static int -vfio_type1_dma_map(int vfio_container_fd) -{ - return rte_memseg_walk(type1_map, &vfio_container_fd); -} - -/* Track the size of the statically allocated DMA window for SPAPR */ -uint64_t spapr_dma_win_len; -uint64_t spapr_dma_win_page_sz; - -static int -vfio_spapr_dma_do_map(int vfio_container_fd, uint64_t vaddr, uint64_t iova, - uint64_t len, int do_map) -{ - struct vfio_iommu_spapr_register_memory reg = { - .argsz = sizeof(reg), - .vaddr = (uintptr_t) vaddr, - .size = len, - .flags = 0 - }; - int ret; - - if (do_map != 0) { - struct vfio_iommu_type1_dma_map dma_map; - - if (iova + len > spapr_dma_win_len) { - EAL_LOG(ERR, "DMA map attempt outside DMA window"); - return -1; - } - - ret = ioctl(vfio_container_fd, - VFIO_IOMMU_SPAPR_REGISTER_MEMORY, ®); - if (ret) { - EAL_LOG(ERR, - "Cannot register vaddr for IOMMU, error " - "%i (%s)", errno, strerror(errno)); - return -1; - } - - memset(&dma_map, 0, sizeof(dma_map)); - dma_map.argsz = sizeof(struct vfio_iommu_type1_dma_map); - dma_map.vaddr = vaddr; - dma_map.size = len; - dma_map.iova = iova; - dma_map.flags = VFIO_DMA_MAP_FLAG_READ | - VFIO_DMA_MAP_FLAG_WRITE; - - ret = ioctl(vfio_container_fd, VFIO_IOMMU_MAP_DMA, &dma_map); - if (ret) { - EAL_LOG(ERR, "Cannot map vaddr for IOMMU, error " - "%i (%s)", errno, strerror(errno)); - return -1; - } - - } else { - struct vfio_iommu_type1_dma_map dma_unmap; - - memset(&dma_unmap, 0, sizeof(dma_unmap)); - dma_unmap.argsz = sizeof(struct vfio_iommu_type1_dma_unmap); - dma_unmap.size = len; - dma_unmap.iova = iova; - - ret = ioctl(vfio_container_fd, VFIO_IOMMU_UNMAP_DMA, - &dma_unmap); - if (ret) { - EAL_LOG(ERR, "Cannot unmap vaddr for IOMMU, error " - "%i (%s)", errno, strerror(errno)); - return -1; - } - - ret = ioctl(vfio_container_fd, - VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY, ®); - if (ret) { - EAL_LOG(ERR, - "Cannot unregister vaddr for IOMMU, error " - "%i (%s)", errno, strerror(errno)); - return -1; - } - } - - return ret; -} - -static int -vfio_spapr_map_walk(const struct rte_memseg_list *msl, - const struct rte_memseg *ms, void *arg) -{ - int *vfio_container_fd = arg; - - /* skip external memory that isn't a heap */ - if (msl->external && !msl->heap) - return 0; - - /* skip any segments with invalid IOVA addresses */ - if (ms->iova == RTE_BAD_IOVA) - return 0; - - return vfio_spapr_dma_do_map(*vfio_container_fd, - ms->addr_64, ms->iova, ms->len, 1); -} - -struct spapr_size_walk_param { - uint64_t max_va; - uint64_t page_sz; - bool is_user_managed; -}; - -/* - * In order to set the DMA window size required for the SPAPR IOMMU - * we need to walk the existing virtual memory allocations as well as - * find the hugepage size used. - */ -static int -vfio_spapr_size_walk(const struct rte_memseg_list *msl, void *arg) -{ - struct spapr_size_walk_param *param = arg; - uint64_t max = (uint64_t) msl->base_va + (uint64_t) msl->len; - - if (msl->external && !msl->heap) { - /* ignore user managed external memory */ - param->is_user_managed = true; - return 0; - } - - if (max > param->max_va) { - param->page_sz = msl->page_sz; - param->max_va = max; - } - - return 0; -} - -/* - * Find the highest memory address used in physical or virtual address - * space and use that as the top of the DMA window. - */ -static int -find_highest_mem_addr(struct spapr_size_walk_param *param) -{ - /* find the maximum IOVA address for setting the DMA window size */ - if (rte_eal_iova_mode() == RTE_IOVA_PA) { - static const char proc_iomem[] = "/proc/iomem"; - static const char str_sysram[] = "System RAM"; - uint64_t start, end, max = 0; - char *line = NULL; - char *dash, *space; - size_t line_len; - - /* - * Example "System RAM" in /proc/iomem: - * 00000000-1fffffffff : System RAM - * 200000000000-201fffffffff : System RAM - */ - FILE *fd = fopen(proc_iomem, "r"); - if (fd == NULL) { - EAL_LOG(ERR, "Cannot open %s", proc_iomem); - return -1; - } - /* Scan /proc/iomem for the highest PA in the system */ - while (getline(&line, &line_len, fd) != -1) { - if (strstr(line, str_sysram) == NULL) - continue; - - space = strstr(line, " "); - dash = strstr(line, "-"); - - /* Validate the format of the memory string */ - if (space == NULL || dash == NULL || space < dash) { - EAL_LOG(ERR, "Can't parse line \"%s\" in file %s", - line, proc_iomem); - continue; - } - - start = strtoull(line, NULL, 16); - end = strtoull(dash + 1, NULL, 16); - EAL_LOG(DEBUG, "Found system RAM from 0x%" PRIx64 - " to 0x%" PRIx64, start, end); - if (end > max) - max = end; - } - free(line); - fclose(fd); - - if (max == 0) { - EAL_LOG(ERR, "Failed to find valid \"System RAM\" " - "entry in file %s", proc_iomem); - return -1; - } - - spapr_dma_win_len = rte_align64pow2(max + 1); - return 0; - } else if (rte_eal_iova_mode() == RTE_IOVA_VA) { - EAL_LOG(DEBUG, "Highest VA address in memseg list is 0x%" - PRIx64, param->max_va); - spapr_dma_win_len = rte_align64pow2(param->max_va); - return 0; - } - - spapr_dma_win_len = 0; - EAL_LOG(ERR, "Unsupported IOVA mode"); - return -1; -} - - -/* - * The SPAPRv2 IOMMU supports 2 DMA windows with starting - * address at 0 or 1<<59. By default, a DMA window is set - * at address 0, 2GB long, with a 4KB page. For DPDK we - * must remove the default window and setup a new DMA window - * based on the hugepage size and memory requirements of - * the application before we can map memory for DMA. - */ -static int -spapr_dma_win_size(void) -{ - struct spapr_size_walk_param param; - - /* only create DMA window once */ - if (spapr_dma_win_len > 0) - return 0; - - /* walk the memseg list to find the page size/max VA address */ - memset(¶m, 0, sizeof(param)); - if (rte_memseg_list_walk(vfio_spapr_size_walk, ¶m) < 0) { - EAL_LOG(ERR, "Failed to walk memseg list for DMA window size"); + if (sysfs_base == NULL || dev_addr == NULL || iommu_group_num == NULL) { + rte_errno = EINVAL; return -1; } - /* we can't be sure if DMA window covers external memory */ - if (param.is_user_managed) - EAL_LOG(WARNING, "Detected user managed external memory which may not be managed by the IOMMU"); - - /* check physical/virtual memory size */ - if (find_highest_mem_addr(¶m) < 0) + if (vfio_cfg.mode == RTE_VFIO_MODE_NONE) { + EAL_LOG(ERR, "VFIO support not initialized"); + rte_errno = ENXIO; return -1; - EAL_LOG(DEBUG, "Setting DMA window size to 0x%" PRIx64, - spapr_dma_win_len); - spapr_dma_win_page_sz = param.page_sz; - rte_mem_set_dma_mask(rte_ctz64(spapr_dma_win_len)); - return 0; -} - -static int -vfio_spapr_create_dma_window(int vfio_container_fd) -{ - struct vfio_iommu_spapr_tce_create create = { - .argsz = sizeof(create), }; - struct vfio_iommu_spapr_tce_remove remove = { - .argsz = sizeof(remove), }; - struct vfio_iommu_spapr_tce_info info = { - .argsz = sizeof(info), }; - int ret; - - ret = spapr_dma_win_size(); - if (ret < 0) - return ret; - - ret = ioctl(vfio_container_fd, VFIO_IOMMU_SPAPR_TCE_GET_INFO, &info); - if (ret) { - EAL_LOG(ERR, "Cannot get IOMMU info, error %i (%s)", - errno, strerror(errno)); - return -1; - } - - /* - * sPAPR v1/v2 IOMMU always has a default 1G DMA window set. The window - * can't be changed for v1 but it can be changed for v2. Since DPDK only - * supports v2, remove the default DMA window so it can be resized. - */ - remove.start_addr = info.dma32_window_start; - ret = ioctl(vfio_container_fd, VFIO_IOMMU_SPAPR_TCE_REMOVE, &remove); - if (ret) - return -1; - - /* create a new DMA window (start address is not selectable) */ - create.window_size = spapr_dma_win_len; - create.page_shift = rte_ctz64(spapr_dma_win_page_sz); - create.levels = 1; - ret = ioctl(vfio_container_fd, VFIO_IOMMU_SPAPR_TCE_CREATE, &create); -#ifdef VFIO_IOMMU_SPAPR_INFO_DDW - /* - * The vfio_iommu_spapr_tce_info structure was modified in - * Linux kernel 4.2.0 to add support for the - * vfio_iommu_spapr_tce_ddw_info structure needed to try - * multiple table levels. Skip the attempt if running with - * an older kernel. - */ - if (ret) { - /* if at first we don't succeed, try more levels */ - uint32_t levels; - - for (levels = create.levels + 1; - ret && levels <= info.ddw.levels; levels++) { - create.levels = levels; - ret = ioctl(vfio_container_fd, - VFIO_IOMMU_SPAPR_TCE_CREATE, &create); - } } -#endif /* VFIO_IOMMU_SPAPR_INFO_DDW */ - if (ret) { - EAL_LOG(ERR, "Cannot create new DMA window, error " - "%i (%s)", errno, strerror(errno)); - EAL_LOG(ERR, - "Consider using a larger hugepage size if supported by the system"); + if (vfio_cfg.mode != RTE_VFIO_MODE_GROUP && vfio_cfg.mode != RTE_VFIO_MODE_NOIOMMU) { + EAL_LOG(ERR, "VFIO not initialized in group mode"); + rte_errno = ENOTSUP; return -1; } - - /* verify the start address */ - if (create.start_addr != 0) { - EAL_LOG(ERR, "Received unsupported start address 0x%" - PRIx64, (uint64_t)create.start_addr); + ret = vfio_group_get_num(sysfs_base, dev_addr, iommu_group_num); + if (ret < 0) { + rte_errno = EINVAL; return -1; - } - return ret; -} - -static int -vfio_spapr_dma_mem_map(int vfio_container_fd, uint64_t vaddr, - uint64_t iova, uint64_t len, int do_map) -{ - int ret = 0; - - if (do_map) { - if (vfio_spapr_dma_do_map(vfio_container_fd, - vaddr, iova, len, 1)) { - EAL_LOG(ERR, "Failed to map DMA"); - ret = -1; - } - } else { - if (vfio_spapr_dma_do_map(vfio_container_fd, - vaddr, iova, len, 0)) { - EAL_LOG(ERR, "Failed to unmap DMA"); - ret = -1; - } - } - - return ret; -} - -static int -vfio_spapr_dma_map(int vfio_container_fd) -{ - if (vfio_spapr_create_dma_window(vfio_container_fd) < 0) { - EAL_LOG(ERR, "Could not create new DMA window!"); + } else if (ret == 0) { + rte_errno = ENODEV; return -1; } - - /* map all existing DPDK segments for DMA */ - if (rte_memseg_walk(vfio_spapr_map_walk, &vfio_container_fd) < 0) - return -1; - - return 0; -} - -static int -vfio_noiommu_dma_map(int __rte_unused vfio_container_fd) -{ - /* No-IOMMU mode does not need DMA mapping */ - return 0; -} - -static int -vfio_noiommu_dma_mem_map(int __rte_unused vfio_container_fd, - uint64_t __rte_unused vaddr, - uint64_t __rte_unused iova, uint64_t __rte_unused len, - int __rte_unused do_map) -{ - /* No-IOMMU mode does not need DMA mapping */ return 0; } static int -vfio_dma_mem_map(struct vfio_config *vfio_cfg, uint64_t vaddr, uint64_t iova, +vfio_dma_mem_map(struct container *cfg, uint64_t vaddr, uint64_t iova, uint64_t len, int do_map) { - const struct vfio_iommu_type *t = vfio_cfg->vfio_iommu_type; + const struct vfio_iommu_ops *t = vfio_cfg.ops; if (!t) { EAL_LOG(ERR, "VFIO support not initialized"); - rte_errno = ENODEV; return -1; } @@ -1852,16 +1158,14 @@ vfio_dma_mem_map(struct vfio_config *vfio_cfg, uint64_t vaddr, uint64_t iova, EAL_LOG(ERR, "VFIO custom DMA region mapping not supported by IOMMU %s", t->name); - rte_errno = ENOTSUP; return -1; } - return t->dma_user_map_func(vfio_cfg->vfio_container_fd, vaddr, iova, - len, do_map); + return t->dma_user_map_func(cfg, vaddr, iova, len, do_map); } static int -container_dma_map(struct vfio_config *vfio_cfg, uint64_t vaddr, uint64_t iova, +container_dma_map(struct container *cfg, uint64_t vaddr, uint64_t iova, uint64_t len) { struct user_mem_map *new_map; @@ -1869,16 +1173,15 @@ container_dma_map(struct vfio_config *vfio_cfg, uint64_t vaddr, uint64_t iova, bool has_partial_unmap; int ret = 0; - user_mem_maps = &vfio_cfg->mem_maps; + user_mem_maps = &cfg->mem_maps; rte_spinlock_recursive_lock(&user_mem_maps->lock); if (user_mem_maps->n_maps == RTE_DIM(user_mem_maps->maps)) { EAL_LOG(ERR, "No more space for user mem maps"); - rte_errno = ENOMEM; ret = -1; goto out; } /* map the entry */ - if (vfio_dma_mem_map(vfio_cfg, vaddr, iova, len, 1)) { + if (vfio_dma_mem_map(cfg, vaddr, iova, len, 1)) { /* technically, this will fail if there are currently no devices * plugged in, even if a device were added later, this mapping * might have succeeded. however, since we cannot verify if this @@ -1891,7 +1194,7 @@ container_dma_map(struct vfio_config *vfio_cfg, uint64_t vaddr, uint64_t iova, goto out; } /* do we have partial unmap support? */ - has_partial_unmap = vfio_cfg->vfio_iommu_type->partial_unmap; + has_partial_unmap = vfio_cfg.ops->partial_unmap; /* create new user mem map entry */ new_map = &user_mem_maps->maps[user_mem_maps->n_maps++]; @@ -1908,17 +1211,17 @@ container_dma_map(struct vfio_config *vfio_cfg, uint64_t vaddr, uint64_t iova, } static int -container_dma_unmap(struct vfio_config *vfio_cfg, uint64_t vaddr, uint64_t iova, +container_dma_unmap(struct container *cfg, uint64_t vaddr, uint64_t iova, uint64_t len) { - struct user_mem_map orig_maps[RTE_DIM(vfio_cfg->mem_maps.maps)]; + struct user_mem_map orig_maps[RTE_DIM(cfg->mem_maps.maps)]; struct user_mem_map new_maps[2]; /* can be at most 2 */ struct user_mem_maps *user_mem_maps; int n_orig, n_new, ret = 0; bool has_partial_unmap; unsigned int newlen; - user_mem_maps = &vfio_cfg->mem_maps; + user_mem_maps = &cfg->mem_maps; rte_spinlock_recursive_lock(&user_mem_maps->lock); /* @@ -1944,13 +1247,12 @@ container_dma_unmap(struct vfio_config *vfio_cfg, uint64_t vaddr, uint64_t iova, /* did we find anything? */ if (n_orig < 0) { EAL_LOG(ERR, "Couldn't find previously mapped region"); - rte_errno = EINVAL; ret = -1; goto out; } /* do we have partial unmap capability? */ - has_partial_unmap = vfio_cfg->vfio_iommu_type->partial_unmap; + has_partial_unmap = vfio_cfg.ops->partial_unmap; /* * if we don't support partial unmap, we must check if start and end of @@ -1966,7 +1268,6 @@ container_dma_unmap(struct vfio_config *vfio_cfg, uint64_t vaddr, uint64_t iova, if (!start_aligned || !end_aligned) { EAL_LOG(DEBUG, "DMA partial unmap unsupported"); - rte_errno = ENOTSUP; ret = -1; goto out; } @@ -1984,28 +1285,20 @@ container_dma_unmap(struct vfio_config *vfio_cfg, uint64_t vaddr, uint64_t iova, newlen = (user_mem_maps->n_maps - n_orig) + n_new; if (newlen >= RTE_DIM(user_mem_maps->maps)) { EAL_LOG(ERR, "Not enough space to store partial mapping"); - rte_errno = ENOMEM; ret = -1; goto out; } /* unmap the entry */ - if (vfio_dma_mem_map(vfio_cfg, vaddr, iova, len, 0)) { + if (vfio_dma_mem_map(cfg, vaddr, iova, len, 0)) { /* there may not be any devices plugged in, so unmapping will - * fail with ENODEV/ENOTSUP rte_errno values, but that doesn't - * stop us from removing the mapping, as the assumption is we - * won't be needing this memory any more and thus will want to - * prevent it from being remapped again on hotplug. so, only - * fail if we indeed failed to unmap (e.g. if the mapping was - * within our mapped range but had invalid alignment). + * fail, but that doesn't stop us from removing the mapping, + * as the assumption is we won't be needing this memory any + * more and thus will want to prevent it from being remapped + * again on hotplug. Ignore the error and proceed with + * removing the mapping from our records. */ - if (rte_errno != ENODEV && rte_errno != ENOTSUP) { - EAL_LOG(ERR, "Couldn't unmap region for DMA"); - ret = -1; - goto out; - } else { - EAL_LOG(DEBUG, "DMA unmapping failed, but removing mappings anyway"); - } + EAL_LOG(DEBUG, "DMA unmapping failed, but removing mappings anyway"); } /* we have unmapped the region, so now update the maps */ @@ -2021,212 +1314,178 @@ RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_noiommu_is_enabled) int rte_vfio_noiommu_is_enabled(void) { - int fd; - ssize_t cnt; - char c; - - fd = open(RTE_VFIO_NOIOMMU_MODE, O_RDONLY); - if (fd < 0) { - if (errno != ENOENT) { - EAL_LOG(ERR, "Cannot open VFIO noiommu file " - "%i (%s)", errno, strerror(errno)); - return -1; - } - /* - * else the file does not exists - * i.e. noiommu is not enabled - */ - return 0; - } - - cnt = read(fd, &c, 1); - close(fd); - if (cnt != 1) { - EAL_LOG(ERR, "Unable to read from VFIO noiommu file " - "%i (%s)", errno, strerror(errno)); - return -1; - } - - return c == 'Y'; + return vfio_cfg.mode == RTE_VFIO_MODE_NOIOMMU; } RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_create) int rte_vfio_container_create(void) { - unsigned int i; + struct container *cfg; + int container_fd; - /* Find an empty slot to store new vfio config */ - for (i = 1; i < RTE_DIM(vfio_cfgs); i++) { - if (vfio_cfgs[i].vfio_container_fd == -1) - break; - } - - if (i == RTE_DIM(vfio_cfgs)) { - EAL_LOG(ERR, "Exceed max VFIO container limit"); + if (vfio_cfg.mode == RTE_VFIO_MODE_NONE) { + EAL_LOG(ERR, "VFIO not initialized"); + rte_errno = ENXIO; return -1; } - - /* Create a new container fd */ - vfio_cfgs[i].vfio_container_fd = vfio_open_container_fd(false); - if (vfio_cfgs[i].vfio_container_fd < 0) { - EAL_LOG(NOTICE, "Fail to create a new VFIO container"); + cfg = vfio_container_create(); + if (cfg == NULL) { + EAL_LOG(ERR, "Reached VFIO container limit"); + rte_errno = ENOSPC; return -1; } - return vfio_cfgs[i].vfio_container_fd; + switch (vfio_cfg.mode) { + case RTE_VFIO_MODE_GROUP: + case RTE_VFIO_MODE_NOIOMMU: + { + container_fd = vfio_group_open_container_fd(); + if (container_fd < 0) { + EAL_LOG(ERR, "Fail to create a new VFIO container"); + rte_errno = EIO; + goto err; + } + cfg->container_fd = container_fd; + break; + } + default: + EAL_LOG(NOTICE, "Unsupported VFIO mode"); + rte_errno = ENOTSUP; + goto err; + } + return container_fd; +err: + vfio_container_erase(cfg); + return -1; } RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_destroy) int rte_vfio_container_destroy(int container_fd) { - struct vfio_config *vfio_cfg; - unsigned int i; + struct container *cfg; + struct vfio_device *dev; - vfio_cfg = get_vfio_cfg_by_container_fd(container_fd); - if (vfio_cfg == NULL) { - EAL_LOG(ERR, "Invalid VFIO container fd"); + if (vfio_cfg.mode == RTE_VFIO_MODE_NONE) { + EAL_LOG(ERR, "VFIO not initialized"); + rte_errno = ENXIO; return -1; } - for (i = 0; i < RTE_DIM(vfio_cfg->vfio_groups); i++) - if (vfio_cfg->vfio_groups[i].group_num != -1) - vfio_container_group_unbind(container_fd, - vfio_cfg->vfio_groups[i].group_num); - - close(container_fd); - vfio_cfg->vfio_container_fd = -1; - vfio_cfg->vfio_active_groups = 0; - vfio_cfg->vfio_iommu_type = NULL; - - return 0; -} - -RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_assign_device) -int -rte_vfio_container_assign_device(int vfio_container_fd, const char *sysfs_base, - const char *dev_addr) -{ - int iommu_group_num; - int ret; - - ret = rte_vfio_get_group_num(sysfs_base, dev_addr, &iommu_group_num); - if (ret < 0) { - EAL_LOG(ERR, "Cannot get IOMMU group number for device %s", - dev_addr); - return -1; - } else if (ret == 0) { - EAL_LOG(ERR, - "Device %s is not assigned to any IOMMU group", - dev_addr); - return -1; - } - - ret = vfio_container_group_bind(vfio_container_fd, - iommu_group_num); - if (ret < 0) { - EAL_LOG(ERR, - "Cannot bind IOMMU group %d for device %s", - iommu_group_num, dev_addr); - return -1; - } - - return 0; -} - -static int -vfio_container_group_bind(int container_fd, int iommu_group_num) -{ - struct vfio_config *vfio_cfg; - - vfio_cfg = get_vfio_cfg_by_container_fd(container_fd); - if (vfio_cfg == NULL) { - EAL_LOG(ERR, "Invalid VFIO container fd"); + cfg = vfio_container_get_by_fd(container_fd); + if (cfg == NULL) { + EAL_LOG(ERR, "VFIO container fd not managed by VFIO"); + rte_errno = ENODEV; return -1; } - - return vfio_get_group_fd(vfio_cfg, iommu_group_num); -} - -static int -vfio_container_group_unbind(int container_fd, int iommu_group_num) -{ - struct vfio_group *cur_grp = NULL; - struct vfio_config *vfio_cfg; - unsigned int i; - - vfio_cfg = get_vfio_cfg_by_container_fd(container_fd); - if (vfio_cfg == NULL) { - EAL_LOG(ERR, "Invalid VFIO container fd"); + /* forbid destroying default container */ + if (vfio_container_is_default(cfg)) { + EAL_LOG(ERR, "Cannot destroy default VFIO container"); + rte_errno = EINVAL; return -1; } - for (i = 0; i < RTE_DIM(vfio_cfg->vfio_groups); i++) { - if (vfio_cfg->vfio_groups[i].group_num == iommu_group_num) { - cur_grp = &vfio_cfg->vfio_groups[i]; - break; + switch (vfio_cfg.mode) { + case RTE_VFIO_MODE_GROUP: + case RTE_VFIO_MODE_NOIOMMU: + /* erase all devices */ + DEVICE_FOREACH_ACTIVE(cfg, dev) { + EAL_LOG(DEBUG, "Device in IOMMU group %d still open, closing", dev->group); + /* + * technically we could've done back-reference lookup and closed our groups + * following a device close, but since we're closing and erasing all groups + * anyway, we can afford to not bother. + */ + vfio_device_erase(cfg, dev); } - } - /* This should not happen */ - if (cur_grp == NULL) { - EAL_LOG(ERR, "Specified VFIO group number not found"); + /* erase all groups */ + struct vfio_group *grp; + GROUP_FOREACH_ACTIVE(cfg, grp) { + EAL_LOG(DEBUG, "IOMMU group %d still open, closing", grp->group_num); + vfio_group_erase(cfg, grp); + } + break; + default: + EAL_LOG(ERR, "Unsupported VFIO mode"); + rte_errno = ENOTSUP; return -1; } - if (cur_grp->fd >= 0 && close(cur_grp->fd) < 0) { - EAL_LOG(ERR, - "Error when closing vfio_group_fd for iommu_group_num " - "%d", iommu_group_num); - return -1; - } - cur_grp->group_num = -1; - cur_grp->fd = -1; - cur_grp->devices = 0; - vfio_cfg->vfio_active_groups--; + /* erase entire config */ + vfio_container_erase(cfg); return 0; } RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_dma_map) int -rte_vfio_container_dma_map(int container_fd, uint64_t vaddr, uint64_t iova, - uint64_t len) +rte_vfio_container_dma_map(int container_fd, uint64_t vaddr, uint64_t iova, uint64_t len) { - struct vfio_config *vfio_cfg; + struct container *cfg; if (len == 0) { rte_errno = EINVAL; return -1; } - vfio_cfg = get_vfio_cfg_by_container_fd(container_fd); - if (vfio_cfg == NULL) { + if (vfio_cfg.mode == RTE_VFIO_MODE_NONE) { + EAL_LOG(ERR, "VFIO support not initialized"); + rte_errno = ENXIO; + return -1; + } + + cfg = vfio_container_get_by_fd(container_fd); + if (cfg == NULL) { EAL_LOG(ERR, "Invalid VFIO container fd"); + rte_errno = EINVAL; return -1; } - return container_dma_map(vfio_cfg, vaddr, iova, len); + if (container_dma_map(cfg, vaddr, iova, len) < 0) { + rte_errno = EIO; + return -1; + } + + return 0; } RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_dma_unmap) int -rte_vfio_container_dma_unmap(int container_fd, uint64_t vaddr, uint64_t iova, - uint64_t len) +rte_vfio_container_dma_unmap(int container_fd, uint64_t vaddr, uint64_t iova, uint64_t len) { - struct vfio_config *vfio_cfg; + struct container *cfg; if (len == 0) { rte_errno = EINVAL; return -1; } - vfio_cfg = get_vfio_cfg_by_container_fd(container_fd); - if (vfio_cfg == NULL) { + if (vfio_cfg.mode == RTE_VFIO_MODE_NONE) { + EAL_LOG(ERR, "VFIO support not initialized"); + rte_errno = ENXIO; + return -1; + } + + cfg = vfio_container_get_by_fd(container_fd); + if (cfg == NULL) { EAL_LOG(ERR, "Invalid VFIO container fd"); + rte_errno = EINVAL; return -1; } - return container_dma_unmap(vfio_cfg, vaddr, iova, len); + if (container_dma_unmap(cfg, vaddr, iova, len) < 0) { + rte_errno = EIO; + return -1; + } + + return 0; +} + +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_mode) +enum rte_vfio_mode +rte_vfio_get_mode(void) +{ + return vfio_cfg.mode; } diff --git a/lib/eal/linux/eal_vfio.h b/lib/eal/linux/eal_vfio.h index 30389fb274..68d3a3ec6e 100644 --- a/lib/eal/linux/eal_vfio.h +++ b/lib/eal/linux/eal_vfio.h @@ -6,60 +6,161 @@ #define EAL_VFIO_H_ #include <rte_common.h> +#include <rte_spinlock.h> #include <stdint.h> +#include <rte_vfio.h> + +/* hot plug/unplug of VFIO groups may cause all DMA maps to be dropped. we can + * recreate the mappings for DPDK segments, but we cannot do so for memory that + * was registered by the user themselves, so we need to store the user mappings + * somewhere, to recreate them later. + */ +#define EAL_VFIO_MAX_USER_MEM_MAPS 256 + +/* user memory map entry */ +struct user_mem_map { + uint64_t addr; /**< start VA */ + uint64_t iova; /**< start IOVA */ + uint64_t len; /**< total length of the mapping */ + uint64_t chunk; /**< this mapping can be split in chunks of this size */ +}; + +/* user memory maps container (common for all API modes) */ +struct user_mem_maps { + rte_spinlock_recursive_t lock; + int n_maps; + struct user_mem_map maps[EAL_VFIO_MAX_USER_MEM_MAPS]; +}; + /* * we don't need to store device fd's anywhere since they can be obtained from * the group fd via an ioctl() call. */ struct vfio_group { + bool active; int group_num; int fd; - int devices; + int n_devices; +}; + +/* device tracking (common for group and cdev modes) */ +struct vfio_device { + bool active; + int group; /**< back-reference to group list (group mode) */ + int fd; +}; + +/* group mode specific configuration */ +struct vfio_group_config { + bool dma_setup_done; + bool iommu_type_set; + bool mem_event_clb_set; + size_t n_groups; + struct vfio_group groups[RTE_MAX_VFIO_GROUPS]; +}; + +/* per-container configuration */ +struct container { + bool active; + int container_fd; + struct user_mem_maps mem_maps; + struct vfio_group_config group_cfg; + int n_devices; + struct vfio_device devices[RTE_MAX_VFIO_DEVICES]; }; /* DMA mapping function prototype. - * Takes VFIO container fd as a parameter. + * Takes VFIO container config as a parameter. * Returns 0 on success, -1 on error. */ -typedef int (*vfio_dma_func_t)(int); +typedef int (*dma_func_t)(struct container *cfg); /* Custom memory region DMA mapping function prototype. - * Takes VFIO container fd, virtual address, physical address, length and + * Takes VFIO container config, virtual address, physical address, length and * operation type (0 to unmap 1 for map) as a parameters. * Returns 0 on success, -1 on error. */ -typedef int (*vfio_dma_user_func_t)(int fd, uint64_t vaddr, uint64_t iova, - uint64_t len, int do_map); +typedef int (*dma_user_func_t)(struct container *cfg, uint64_t vaddr, + uint64_t iova, uint64_t len, int do_map); -struct vfio_iommu_type { +/* mode-independent ops */ +struct vfio_iommu_ops { int type_id; const char *name; bool partial_unmap; - vfio_dma_user_func_t dma_user_map_func; - vfio_dma_func_t dma_map_func; + dma_user_func_t dma_user_map_func; + dma_func_t dma_map_func; }; -/* get the vfio container that devices are bound to by default */ -int vfio_open_container_fd(bool mp_request); +/* global configuration */ +struct vfio_config { + struct container *default_cfg; + enum rte_vfio_mode mode; + const struct vfio_iommu_ops *ops; +}; + +/* per-process, per-container data */ +extern struct container containers[RTE_MAX_VFIO_CONTAINERS]; + +/* current configuration */ +extern struct vfio_config vfio_cfg; + +#define CONTAINER_FOREACH(cfg) \ + for ((cfg) = &containers[0]; \ + (cfg) < &containers[RTE_DIM(containers)]; \ + (cfg)++) + +#define CONTAINER_FOREACH_ACTIVE(cfg) \ + CONTAINER_FOREACH((cfg)) \ + if (((cfg)->active)) + +#define GROUP_FOREACH(cfg, grp) \ + for ((grp) = &((cfg)->group_cfg.groups[0]); \ + (grp) < &((cfg)->group_cfg.groups[RTE_DIM((cfg)->group_cfg.groups)]); \ + (grp)++) -/* pick IOMMU type. returns a pointer to vfio_iommu_type or NULL for error */ -const struct vfio_iommu_type * -vfio_set_iommu_type(int vfio_container_fd); +#define GROUP_FOREACH_ACTIVE(cfg, grp) \ + GROUP_FOREACH((cfg), (grp)) \ + if ((grp)->active) -int -vfio_get_iommu_type(void); +#define DEVICE_FOREACH(cfg, dev) \ + for ((dev) = &((cfg)->devices[0]); \ + (dev) < &((cfg)->devices[RTE_DIM((cfg)->devices)]); \ + (dev)++) -int vfio_get_group_fd_by_num(int iommu_group_num); +#define DEVICE_FOREACH_ACTIVE(cfg, dev) \ + DEVICE_FOREACH((cfg), (dev)) \ + if ((dev)->active) -/* check if we have any supported extensions */ -int -vfio_has_supported_extensions(int vfio_container_fd); +/* for containers, we only need to initialize the lock in mem maps */ +#define CONTAINER_INITIALIZER \ + ((struct container){ \ + .mem_maps = {.lock = RTE_SPINLOCK_RECURSIVE_INITIALIZER,}, \ + }) +int vfio_get_iommu_type(void); int vfio_mp_sync_setup(void); void vfio_mp_sync_cleanup(void); +bool vfio_container_is_default(struct container *cfg); +/* group mode functions */ +int vfio_group_enable(struct container *cfg); +int vfio_group_open_container_fd(void); +int vfio_group_noiommu_is_enabled(void); +int vfio_group_get_num(const char *sysfs_base, const char *dev_addr, + int *iommu_group_num); +struct vfio_group *vfio_group_get_by_num(struct container *cfg, int iommu_group); +struct vfio_group *vfio_group_create(struct container *cfg, int iommu_group); +void vfio_group_erase(struct container *cfg, struct vfio_group *grp); +int vfio_group_open_fd(struct container *cfg, struct vfio_group *grp); +int vfio_group_prepare(struct container *cfg, struct vfio_group *grp); +int vfio_group_setup_iommu(struct container *cfg); +int vfio_group_setup_device_fd(const char *dev_addr, + struct vfio_group *grp, struct vfio_device *dev); + +#define VFIO_MEM_EVENT_CLB_NAME "vfio_mem_event_clb" #define EAL_VFIO_MP "eal_vfio_mp_sync" #define SOCKET_REQ_CONTAINER 0x100 @@ -75,6 +176,7 @@ struct vfio_mp_param { union { int group_num; int iommu_type_id; + enum rte_vfio_mode mode; }; }; diff --git a/lib/eal/linux/eal_vfio_group.c b/lib/eal/linux/eal_vfio_group.c new file mode 100644 index 0000000000..1f1b3084c2 --- /dev/null +++ b/lib/eal/linux/eal_vfio_group.c @@ -0,0 +1,984 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright(c) 2010-2025 Intel Corporation + */ + +#include <dirent.h> +#include <fcntl.h> +#include <inttypes.h> +#include <stdbool.h> +#include <stdint.h> +#include <stdlib.h> +#include <sys/ioctl.h> +#include <unistd.h> + +#include <uapi/linux/vfio.h> + +#include <rte_log.h> +#include <rte_errno.h> +#include <rte_eal_memconfig.h> +#include <rte_memory.h> +#include <rte_string_fns.h> +#include <rte_vfio.h> + +#include "eal_vfio.h" +#include "eal_private.h" +#include "eal_internal_cfg.h" + +static int vfio_type1_dma_map(struct container *); +static int vfio_type1_dma_mem_map(struct container *, uint64_t, uint64_t, uint64_t, int); +static int vfio_spapr_dma_map(struct container *); +static int vfio_spapr_dma_mem_map(struct container *, uint64_t, uint64_t, uint64_t, int); +static int vfio_noiommu_dma_map(struct container *); +static int vfio_noiommu_dma_mem_map(struct container *, uint64_t, uint64_t, uint64_t, int); + +/* IOMMU types we support */ +static const struct vfio_iommu_ops iommu_types[] = { + /* x86 IOMMU, otherwise known as type 1 */ + { + .type_id = VFIO_TYPE1_IOMMU, + .name = "Type 1", + .partial_unmap = false, + .dma_map_func = &vfio_type1_dma_map, + .dma_user_map_func = &vfio_type1_dma_mem_map + }, + /* ppc64 IOMMU, otherwise known as spapr */ + { + .type_id = VFIO_SPAPR_TCE_v2_IOMMU, + .name = "sPAPR", + .partial_unmap = true, + .dma_map_func = &vfio_spapr_dma_map, + .dma_user_map_func = &vfio_spapr_dma_mem_map + }, + /* IOMMU-less mode */ + { + .type_id = VFIO_NOIOMMU_IOMMU, + .name = "No-IOMMU", + .partial_unmap = true, + .dma_map_func = &vfio_noiommu_dma_map, + .dma_user_map_func = &vfio_noiommu_dma_mem_map + }, +}; + +static const struct vfio_iommu_ops * +vfio_group_set_iommu_type(int vfio_container_fd) +{ + unsigned int idx; + for (idx = 0; idx < RTE_DIM(iommu_types); idx++) { + const struct vfio_iommu_ops *t = &iommu_types[idx]; + + int ret = ioctl(vfio_container_fd, VFIO_SET_IOMMU, t->type_id); + if (ret == 0) + return t; + /* not an error, there may be more supported IOMMU types */ + EAL_LOG(DEBUG, "Set IOMMU type %d (%s) failed, error " + "%i (%s)", t->type_id, t->name, errno, + strerror(errno)); + } + /* if we didn't find a suitable IOMMU type, fail */ + return NULL; +} + +static int +type1_map(const struct rte_memseg_list *msl, const struct rte_memseg *ms, + void *arg) +{ + struct container *cfg = arg; + + /* skip external memory that isn't a heap */ + if (msl->external && !msl->heap) + return 0; + + /* skip any segments with invalid IOVA addresses */ + if (ms->iova == RTE_BAD_IOVA) + return 0; + + return vfio_type1_dma_mem_map(cfg, ms->addr_64, ms->iova, ms->len, 1); +} + +static int +vfio_type1_dma_mem_map(struct container *cfg, uint64_t vaddr, uint64_t iova, + uint64_t len, int do_map) +{ + struct vfio_iommu_type1_dma_map dma_map; + struct vfio_iommu_type1_dma_unmap dma_unmap; + int ret; + + if (do_map != 0) { + memset(&dma_map, 0, sizeof(dma_map)); + dma_map.argsz = sizeof(struct vfio_iommu_type1_dma_map); + dma_map.vaddr = vaddr; + dma_map.size = len; + dma_map.iova = iova; + dma_map.flags = VFIO_DMA_MAP_FLAG_READ | + VFIO_DMA_MAP_FLAG_WRITE; + + ret = ioctl(cfg->container_fd, VFIO_IOMMU_MAP_DMA, &dma_map); + if (ret) { + /** + * In case the mapping was already done EEXIST will be + * returned from kernel. + */ + if (errno == EEXIST) { + EAL_LOG(DEBUG, + "Memory segment is already mapped, skipping"); + } else { + EAL_LOG(ERR, + "Cannot set up DMA remapping, error " + "%i (%s)", errno, strerror(errno)); + return -1; + } + } + } else { + memset(&dma_unmap, 0, sizeof(dma_unmap)); + dma_unmap.argsz = sizeof(struct vfio_iommu_type1_dma_unmap); + dma_unmap.size = len; + dma_unmap.iova = iova; + + ret = ioctl(cfg->container_fd, VFIO_IOMMU_UNMAP_DMA, + &dma_unmap); + if (ret) { + EAL_LOG(ERR, "Cannot clear DMA remapping, error " + "%i (%s)", errno, strerror(errno)); + return -1; + } else if (dma_unmap.size != len) { + EAL_LOG(ERR, "Unexpected size %"PRIu64 + " of DMA remapping cleared instead of %"PRIu64, + (uint64_t)dma_unmap.size, len); + return -1; + } + } + + return 0; +} + +static int +vfio_type1_dma_map(struct container *cfg) +{ + return rte_memseg_walk(type1_map, cfg); +} + +/* Track the size of the statically allocated DMA window for SPAPR */ +uint64_t spapr_dma_win_len; +uint64_t spapr_dma_win_page_sz; + +static int +vfio_spapr_dma_do_map(struct container *cfg, uint64_t vaddr, uint64_t iova, + uint64_t len, int do_map) +{ + struct vfio_iommu_spapr_register_memory reg = { + .argsz = sizeof(reg), + .vaddr = (uintptr_t) vaddr, + .size = len, + .flags = 0 + }; + int ret; + + if (do_map != 0) { + struct vfio_iommu_type1_dma_map dma_map; + + if (iova + len > spapr_dma_win_len) { + EAL_LOG(ERR, "DMA map attempt outside DMA window"); + return -1; + } + + ret = ioctl(cfg->container_fd, + VFIO_IOMMU_SPAPR_REGISTER_MEMORY, ®); + if (ret) { + EAL_LOG(ERR, + "Cannot register vaddr for IOMMU, error " + "%i (%s)", errno, strerror(errno)); + return -1; + } + + memset(&dma_map, 0, sizeof(dma_map)); + dma_map.argsz = sizeof(struct vfio_iommu_type1_dma_map); + dma_map.vaddr = vaddr; + dma_map.size = len; + dma_map.iova = iova; + dma_map.flags = VFIO_DMA_MAP_FLAG_READ | VFIO_DMA_MAP_FLAG_WRITE; + + ret = ioctl(cfg->container_fd, VFIO_IOMMU_MAP_DMA, &dma_map); + if (ret) { + EAL_LOG(ERR, "Cannot map vaddr for IOMMU, error " + "%i (%s)", errno, strerror(errno)); + return -1; + } + + } else { + struct vfio_iommu_type1_dma_map dma_unmap; + + memset(&dma_unmap, 0, sizeof(dma_unmap)); + dma_unmap.argsz = sizeof(struct vfio_iommu_type1_dma_unmap); + dma_unmap.size = len; + dma_unmap.iova = iova; + + ret = ioctl(cfg->container_fd, VFIO_IOMMU_UNMAP_DMA, + &dma_unmap); + if (ret) { + EAL_LOG(ERR, "Cannot unmap vaddr for IOMMU, error " + "%i (%s)", errno, strerror(errno)); + return -1; + } + + ret = ioctl(cfg->container_fd, + VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY, ®); + if (ret) { + EAL_LOG(ERR, + "Cannot unregister vaddr for IOMMU, error " + "%i (%s)", errno, strerror(errno)); + return -1; + } + } + + return ret; +} + +static int +vfio_spapr_map_walk(const struct rte_memseg_list *msl, + const struct rte_memseg *ms, void *arg) +{ + struct container *cfg = arg; + + /* skip external memory that isn't a heap */ + if (msl->external && !msl->heap) + return 0; + + /* skip any segments with invalid IOVA addresses */ + if (ms->iova == RTE_BAD_IOVA) + return 0; + + return vfio_spapr_dma_do_map(cfg, ms->addr_64, ms->iova, ms->len, 1); +} + +struct spapr_size_walk_param { + uint64_t max_va; + uint64_t page_sz; + bool is_user_managed; +}; + +/* + * In order to set the DMA window size required for the SPAPR IOMMU + * we need to walk the existing virtual memory allocations as well as + * find the hugepage size used. + */ +static int +vfio_spapr_size_walk(const struct rte_memseg_list *msl, void *arg) +{ + struct spapr_size_walk_param *param = arg; + uint64_t max = (uint64_t) msl->base_va + (uint64_t) msl->len; + + if (msl->external && !msl->heap) { + /* ignore user managed external memory */ + param->is_user_managed = true; + return 0; + } + + if (max > param->max_va) { + param->page_sz = msl->page_sz; + param->max_va = max; + } + + return 0; +} + +/* + * Find the highest memory address used in physical or virtual address + * space and use that as the top of the DMA window. + */ +static int +find_highest_mem_addr(struct spapr_size_walk_param *param) +{ + /* find the maximum IOVA address for setting the DMA window size */ + if (rte_eal_iova_mode() == RTE_IOVA_PA) { + static const char proc_iomem[] = "/proc/iomem"; + static const char str_sysram[] = "System RAM"; + uint64_t start, end, max = 0; + char *line = NULL; + char *dash, *space; + size_t line_len; + + /* + * Example "System RAM" in /proc/iomem: + * 00000000-1fffffffff : System RAM + * 200000000000-201fffffffff : System RAM + */ + FILE *fd = fopen(proc_iomem, "r"); + if (fd == NULL) { + EAL_LOG(ERR, "Cannot open %s", proc_iomem); + return -1; + } + /* Scan /proc/iomem for the highest PA in the system */ + while (getline(&line, &line_len, fd) != -1) { + if (strstr(line, str_sysram) == NULL) + continue; + + space = strstr(line, " "); + dash = strstr(line, "-"); + + /* Validate the format of the memory string */ + if (space == NULL || dash == NULL || space < dash) { + EAL_LOG(ERR, "Can't parse line \"%s\" in file %s", + line, proc_iomem); + continue; + } + + start = strtoull(line, NULL, 16); + end = strtoull(dash + 1, NULL, 16); + EAL_LOG(DEBUG, "Found system RAM from 0x%" PRIx64 + " to 0x%" PRIx64, start, end); + if (end > max) + max = end; + } + free(line); + fclose(fd); + + if (max == 0) { + EAL_LOG(ERR, "Failed to find valid \"System RAM\" " + "entry in file %s", proc_iomem); + return -1; + } + + spapr_dma_win_len = rte_align64pow2(max + 1); + return 0; + } else if (rte_eal_iova_mode() == RTE_IOVA_VA) { + EAL_LOG(DEBUG, "Highest VA address in memseg list is 0x%" + PRIx64, param->max_va); + spapr_dma_win_len = rte_align64pow2(param->max_va); + return 0; + } + + spapr_dma_win_len = 0; + EAL_LOG(ERR, "Unsupported IOVA mode"); + return -1; +} + + +/* + * The SPAPRv2 IOMMU supports 2 DMA windows with starting + * address at 0 or 1<<59. By default, a DMA window is set + * at address 0, 2GB long, with a 4KB page. For DPDK we + * must remove the default window and setup a new DMA window + * based on the hugepage size and memory requirements of + * the application before we can map memory for DMA. + */ +static int +spapr_dma_win_size(void) +{ + struct spapr_size_walk_param param; + + /* only create DMA window once */ + if (spapr_dma_win_len > 0) + return 0; + + /* walk the memseg list to find the page size/max VA address */ + memset(¶m, 0, sizeof(param)); + if (rte_memseg_list_walk(vfio_spapr_size_walk, ¶m) < 0) { + EAL_LOG(ERR, "Failed to walk memseg list for DMA window size"); + return -1; + } + + /* we can't be sure if DMA window covers external memory */ + if (param.is_user_managed) + EAL_LOG(WARNING, "Detected user managed external memory which may not be managed by the IOMMU"); + + /* check physical/virtual memory size */ + if (find_highest_mem_addr(¶m) < 0) + return -1; + EAL_LOG(DEBUG, "Setting DMA window size to 0x%" PRIx64, + spapr_dma_win_len); + spapr_dma_win_page_sz = param.page_sz; + rte_mem_set_dma_mask(rte_ctz64(spapr_dma_win_len)); + return 0; +} + +static int +vfio_spapr_create_dma_window(struct container *cfg) +{ + struct vfio_iommu_spapr_tce_create create = { + .argsz = sizeof(create), }; + struct vfio_iommu_spapr_tce_remove remove = { + .argsz = sizeof(remove), }; + struct vfio_iommu_spapr_tce_info info = { + .argsz = sizeof(info), }; + int ret; + + ret = spapr_dma_win_size(); + if (ret < 0) + return ret; + + ret = ioctl(cfg->container_fd, VFIO_IOMMU_SPAPR_TCE_GET_INFO, &info); + if (ret) { + EAL_LOG(ERR, "Cannot get IOMMU info, error %i (%s)", + errno, strerror(errno)); + return -1; + } + + /* + * sPAPR v1/v2 IOMMU always has a default 1G DMA window set. The window + * can't be changed for v1 but it can be changed for v2. Since DPDK only + * supports v2, remove the default DMA window so it can be resized. + */ + remove.start_addr = info.dma32_window_start; + ret = ioctl(cfg->container_fd, VFIO_IOMMU_SPAPR_TCE_REMOVE, &remove); + if (ret) + return -1; + + /* create a new DMA window (start address is not selectable) */ + create.window_size = spapr_dma_win_len; + create.page_shift = rte_ctz64(spapr_dma_win_page_sz); + create.levels = 1; + ret = ioctl(cfg->container_fd, VFIO_IOMMU_SPAPR_TCE_CREATE, &create); + /* + * The vfio_iommu_spapr_tce_info structure was modified in + * Linux kernel 4.2.0 to add support for the + * vfio_iommu_spapr_tce_ddw_info structure needed to try + * multiple table levels. Skip the attempt if running with + * an older kernel. + */ + if (ret) { + /* if at first we don't succeed, try more levels */ + uint32_t levels; + + for (levels = create.levels + 1; + ret && levels <= info.ddw.levels; levels++) { + create.levels = levels; + ret = ioctl(cfg->container_fd, + VFIO_IOMMU_SPAPR_TCE_CREATE, &create); + } + } + if (ret) { + EAL_LOG(ERR, "Cannot create new DMA window, error " + "%i (%s)", errno, strerror(errno)); + EAL_LOG(ERR, + "Consider using a larger hugepage size if supported by the system"); + return -1; + } + + /* verify the start address */ + if (create.start_addr != 0) { + EAL_LOG(ERR, "Received unsupported start address 0x%" + PRIx64, (uint64_t)create.start_addr); + return -1; + } + return ret; +} + +static int +vfio_spapr_dma_mem_map(struct container *cfg, uint64_t vaddr, + uint64_t iova, uint64_t len, int do_map) +{ + int ret = 0; + + if (do_map) { + if (vfio_spapr_dma_do_map(cfg, vaddr, iova, len, 1)) { + EAL_LOG(ERR, "Failed to map DMA"); + ret = -1; + } + } else { + if (vfio_spapr_dma_do_map(cfg, vaddr, iova, len, 0)) { + EAL_LOG(ERR, "Failed to unmap DMA"); + ret = -1; + } + } + + return ret; +} + +static int +vfio_spapr_dma_map(struct container *cfg) +{ + if (vfio_spapr_create_dma_window(cfg) < 0) { + EAL_LOG(ERR, "Could not create new DMA window!"); + return -1; + } + + /* map all existing DPDK segments for DMA */ + if (rte_memseg_walk(vfio_spapr_map_walk, cfg) < 0) + return -1; + + return 0; +} + +static int +vfio_noiommu_dma_map(struct container *cfg __rte_unused) +{ + /* No-IOMMU mode does not need DMA mapping */ + return 0; +} + +static int +vfio_noiommu_dma_mem_map(struct container *cfg __rte_unused, + uint64_t vaddr __rte_unused, + uint64_t iova __rte_unused, uint64_t len __rte_unused, + int do_map __rte_unused) +{ + /* No-IOMMU mode does not need DMA mapping */ + return 0; +} + +struct vfio_group * +vfio_group_create(struct container *cfg, int iommu_group) +{ + struct vfio_group *grp; + + if (cfg->group_cfg.n_groups >= RTE_DIM(cfg->group_cfg.groups)) { + EAL_LOG(ERR, "Cannot add more VFIO groups to container"); + return NULL; + } + GROUP_FOREACH(cfg, grp) { + if (grp->active) + continue; + cfg->group_cfg.n_groups++; + grp->active = true; + grp->group_num = iommu_group; + grp->fd = -1; + return grp; + } + /* should not happen */ + return NULL; +} + +void +vfio_group_erase(struct container *cfg, struct vfio_group *grp) +{ + struct vfio_group_config *group_cfg = &cfg->group_cfg; + + if (grp->fd >= 0 && close(grp->fd) < 0) + EAL_LOG(ERR, "Error when closing group fd %d", grp->fd); + + *grp = (struct vfio_group){0}; + group_cfg->n_groups--; + + /* if this was the last group in config, erase IOMMU setup and unregister callback */ + if (group_cfg->n_groups == 0) { + group_cfg->dma_setup_done = false; + group_cfg->iommu_type_set = false; + } +} + +struct vfio_group * +vfio_group_get_by_num(struct container *cfg, int iommu_group) +{ + struct vfio_group *grp; + + GROUP_FOREACH_ACTIVE(cfg, grp) { + if (grp->group_num == iommu_group) + return grp; + } + return NULL; +} + +static int +vfio_open_group_sysfs(int iommu_group_num) +{ + char filename[PATH_MAX]; + int fd; + + if (vfio_cfg.mode == RTE_VFIO_MODE_GROUP) + snprintf(filename, sizeof(filename), RTE_VFIO_GROUP_FMT, iommu_group_num); + else if (vfio_cfg.mode == RTE_VFIO_MODE_NOIOMMU) + snprintf(filename, sizeof(filename), RTE_VFIO_NOIOMMU_GROUP_FMT, iommu_group_num); + + /* reset errno before open to differentiate errors */ + errno = 0; + fd = open(filename, O_RDWR); + + /* we have to differentiate between failed open and non-existence */ + if (errno == ENOENT) + return -ENOENT; + return fd; +} + +static int +vfio_group_request_fd(int iommu_group_num) +{ + struct rte_mp_msg mp_req, *mp_rep; + struct rte_mp_reply mp_reply = {0}; + struct timespec ts = {.tv_sec = 5, .tv_nsec = 0}; + struct vfio_mp_param *p = (struct vfio_mp_param *)mp_req.param; + int vfio_group_fd = -1; + + p->req = SOCKET_REQ_GROUP; + p->group_num = iommu_group_num; + rte_strscpy(mp_req.name, EAL_VFIO_MP, sizeof(mp_req.name)); + mp_req.len_param = sizeof(*p); + mp_req.num_fds = 0; + + if (rte_mp_request_sync(&mp_req, &mp_reply, &ts) == 0 && mp_reply.nb_received == 1) { + mp_rep = &mp_reply.msgs[0]; + p = (struct vfio_mp_param *)mp_rep->param; + if (p->result == SOCKET_OK && mp_rep->num_fds == 1) { + vfio_group_fd = mp_rep->fds[0]; + } else if (p->result == SOCKET_NO_FD) { + EAL_LOG(ERR, "Bad VFIO group fd"); + vfio_group_fd = -ENOENT; + } + } + + free(mp_reply.msgs); + return vfio_group_fd; +} + +int +vfio_group_open_fd(struct container *cfg, struct vfio_group *grp) +{ + int vfio_group_fd; + + /* we make multiprocess request only in secondary processes for default config */ + if ((rte_eal_process_type() != RTE_PROC_PRIMARY) && (vfio_container_is_default(cfg))) + vfio_group_fd = vfio_group_request_fd(grp->group_num); + else + vfio_group_fd = vfio_open_group_sysfs(grp->group_num); + + /* pass the non-existence up the chain */ + if (vfio_group_fd == -ENOENT) + return vfio_group_fd; + else if (vfio_group_fd < 0) { + EAL_LOG(ERR, "Failed to open VFIO group %d", grp->group_num); + return vfio_group_fd; + } + grp->fd = vfio_group_fd; + return 0; +} + +static const struct vfio_iommu_ops * +vfio_group_sync_iommu_ops(void) +{ + struct rte_mp_msg mp_req, *mp_rep; + struct rte_mp_reply mp_reply = {0}; + struct timespec ts = {.tv_sec = 5, .tv_nsec = 0}; + struct vfio_mp_param *p = (struct vfio_mp_param *)mp_req.param; + int iommu_type_id; + unsigned int i; + + /* find default container's IOMMU type */ + p->req = SOCKET_REQ_IOMMU_TYPE; + rte_strscpy(mp_req.name, EAL_VFIO_MP, sizeof(mp_req.name)); + mp_req.len_param = sizeof(*p); + mp_req.num_fds = 0; + + iommu_type_id = -1; + if (rte_mp_request_sync(&mp_req, &mp_reply, &ts) == 0 && + mp_reply.nb_received == 1) { + mp_rep = &mp_reply.msgs[0]; + p = (struct vfio_mp_param *)mp_rep->param; + if (p->result == SOCKET_OK) + iommu_type_id = p->iommu_type_id; + } + free(mp_reply.msgs); + if (iommu_type_id < 0) { + EAL_LOG(ERR, "Could not get IOMMU type from primary process"); + return NULL; + } + + /* we now have an fd for default container, as well as its IOMMU type. + * now, set up default VFIO container config to match. + */ + for (i = 0; i < RTE_DIM(iommu_types); i++) { + const struct vfio_iommu_ops *t = &iommu_types[i]; + if (t->type_id != iommu_type_id) + continue; + + return t; + } + EAL_LOG(ERR, "Could not find IOMMU type id (%i)", iommu_type_id); + return NULL; +} + +int +vfio_group_noiommu_is_enabled(void) +{ + int fd; + ssize_t cnt; + char c; + + fd = open(RTE_VFIO_NOIOMMU_MODE, O_RDONLY); + if (fd < 0) { + if (errno != ENOENT) { + EAL_LOG(ERR, "Cannot open VFIO noiommu file " + "%i (%s)", errno, strerror(errno)); + return -1; + } + /* + * else the file does not exists + * i.e. noiommu is not enabled + */ + return 0; + } + + cnt = read(fd, &c, 1); + close(fd); + if (cnt != 1) { + EAL_LOG(ERR, "Unable to read from VFIO noiommu file " + "%i (%s)", errno, strerror(errno)); + return -1; + } + + return c == 'Y'; +} + +static int +vfio_has_supported_extensions(int vfio_container_fd) +{ + int ret; + unsigned int idx, n_extensions = 0; + for (idx = 0; idx < RTE_DIM(iommu_types); idx++) { + const struct vfio_iommu_ops *t = &iommu_types[idx]; + + ret = ioctl(vfio_container_fd, VFIO_CHECK_EXTENSION, + t->type_id); + if (ret < 0) { + EAL_LOG(ERR, "Could not get IOMMU type, error " + "%i (%s)", errno, strerror(errno)); + close(vfio_container_fd); + return -1; + } else if (ret == 1) { + /* we found a supported extension */ + n_extensions++; + } + EAL_LOG(DEBUG, "IOMMU type %d (%s) is %s", + t->type_id, t->name, + ret ? "supported" : "not supported"); + } + + /* if we didn't find any supported IOMMU types, fail */ + if (n_extensions == 0) + return -1; + + return 0; +} + +int +vfio_group_open_container_fd(void) +{ + int ret, vfio_container_fd; + + vfio_container_fd = open(RTE_VFIO_CONTAINER_PATH, O_RDWR); + if (vfio_container_fd < 0) { + EAL_LOG(DEBUG, "Cannot open VFIO container %s, error %i (%s)", + RTE_VFIO_CONTAINER_PATH, errno, strerror(errno)); + return -1; + } + + /* check VFIO API version */ + ret = ioctl(vfio_container_fd, VFIO_GET_API_VERSION); + if (ret != VFIO_API_VERSION) { + if (ret < 0) + EAL_LOG(DEBUG, + "Could not get VFIO API version, error " + "%i (%s)", errno, strerror(errno)); + else + EAL_LOG(DEBUG, "Unsupported VFIO API version!"); + close(vfio_container_fd); + return -1; + } + + ret = vfio_has_supported_extensions(vfio_container_fd); + if (ret) { + EAL_LOG(DEBUG, + "No supported IOMMU extensions found!"); + close(vfio_container_fd); + return -1; + } + + return vfio_container_fd; +} + +int +vfio_group_enable(struct container *cfg) +{ + int container_fd; + DIR *dir; + + /* VFIO directory might not exist (e.g., unprivileged containers) */ + dir = opendir(RTE_VFIO_DIR); + if (dir == NULL) { + EAL_LOG(DEBUG, + "VFIO directory does not exist, skipping VFIO group support..."); + return 1; + } + closedir(dir); + + /* open a default container */ + container_fd = vfio_group_open_container_fd(); + if (container_fd < 0) + return -1; + + cfg->container_fd = container_fd; + return 0; +} + +int +vfio_group_prepare(struct container *cfg, struct vfio_group *grp) +{ + struct vfio_group_status group_status = { + .argsz = sizeof(group_status)}; + int ret; + + /* + * We need to assign group to a container and check if it is viable, but there are cases + * where we don't need to do that. + * + * For default container, we need to set up the group only in primary process, as secondary + * process would have requested group fd over IPC, which implies it would have already been + * set up by the primary. + * + * For custom containers, every process sets up its own groups. + */ + if (vfio_container_is_default(cfg) && rte_eal_process_type() != RTE_PROC_PRIMARY) { + EAL_LOG(DEBUG, "Skipping setup for VFIO group %d", grp->group_num); + return 0; + } + + /* check if the group is viable */ + ret = ioctl(grp->fd, VFIO_GROUP_GET_STATUS, &group_status); + if (ret) { + EAL_LOG(ERR, "Cannot get VFIO group status for group %d, error %i (%s)", + grp->group_num, errno, strerror(errno)); + return -1; + } + + if ((group_status.flags & VFIO_GROUP_FLAGS_VIABLE) == 0) { + EAL_LOG(ERR, "VFIO group %d is not viable! " + "Not all devices in IOMMU group bound to VFIO or unbound", + grp->group_num); + return -1; + } + + /* set container for group if necessary */ + if ((group_status.flags & VFIO_GROUP_FLAGS_CONTAINER_SET) == 0) { + /* add group to a container */ + ret = ioctl(grp->fd, VFIO_GROUP_SET_CONTAINER, &cfg->container_fd); + if (ret) { + EAL_LOG(ERR, "Cannot add VFIO group %d to container, error %i (%s)", + grp->group_num, errno, strerror(errno)); + return -1; + } + } else { + /* group is already added to a container - this should not happen */ + EAL_LOG(ERR, "VFIO group %d is already assigned to a container", grp->group_num); + return -1; + } + return 0; +} + +int +vfio_group_setup_iommu(struct container *cfg) +{ + const struct vfio_iommu_ops *ops; + + /* + * Setting IOMMU type is a per-container operation (via ioctl on container fd), but the ops + * structure is global and shared across all containers. + * + * For secondary processes with default container, we sync ops from primary. For all other + * cases (primary, or secondary with custom containers), we set IOMMU type on the container + * which also discovers the ops. + */ + if (vfio_container_is_default(cfg) && rte_eal_process_type() != RTE_PROC_PRIMARY) { + /* Secondary process: sync ops from primary for default container */ + ops = vfio_group_sync_iommu_ops(); + if (ops == NULL) + return -1; + } else { + /* Primary process OR custom container: set IOMMU type on container */ + ops = vfio_group_set_iommu_type(cfg->container_fd); + if (ops == NULL) + return -1; + } + + /* Set or verify global ops */ + if (vfio_cfg.ops == NULL) { + vfio_cfg.ops = ops; + EAL_LOG(INFO, "IOMMU type set to %d (%s)", ops->type_id, ops->name); + } else if (vfio_cfg.ops != ops) { + /* This shouldn't happen on the same machine, but log it */ + EAL_LOG(WARNING, + "Container has different IOMMU type (%d - %s) than previously set (%d - %s)", + ops->type_id, ops->name, vfio_cfg.ops->type_id, vfio_cfg.ops->name); + } + + return 0; +} + +int +vfio_group_setup_device_fd(const char *dev_addr, struct vfio_group *grp, struct vfio_device *dev) +{ + rte_uuid_t vf_token; + int fd; + + rte_eal_vfio_get_vf_token(vf_token); + + if (!rte_uuid_is_null(vf_token)) { + char vf_token_str[RTE_UUID_STRLEN]; + char dev[PATH_MAX]; + + rte_uuid_unparse(vf_token, vf_token_str, sizeof(vf_token_str)); + snprintf(dev, sizeof(dev), + "%s vf_token=%s", dev_addr, vf_token_str); + + fd = ioctl(grp->fd, VFIO_GROUP_GET_DEVICE_FD, dev); + if (fd >= 0) + goto out; + } + /* get a file descriptor for the device */ + fd = ioctl(grp->fd, VFIO_GROUP_GET_DEVICE_FD, dev_addr); + if (fd < 0) { + /* + * if we cannot get a device fd, this implies a problem with the VFIO group or the + * container not having IOMMU configured. + */ + EAL_LOG(WARNING, "Getting a vfio_dev_fd for %s failed", dev_addr); + return -1; + } +out: + dev->fd = fd; + /* store backreference to group */ + dev->group = grp->group_num; + /* increment number of devices in group */ + grp->n_devices++; + return 0; +} + +int +vfio_group_get_num(const char *sysfs_base, const char *dev_addr, int *iommu_group_num) +{ + char linkname[PATH_MAX]; + char filename[PATH_MAX]; + char *tok[16], *group_tok, *end; + int ret, group_num; + + memset(linkname, 0, sizeof(linkname)); + memset(filename, 0, sizeof(filename)); + + /* try to find out IOMMU group for this device */ + snprintf(linkname, sizeof(linkname), + "%s/%s/iommu_group", sysfs_base, dev_addr); + + ret = readlink(linkname, filename, sizeof(filename)); + + /* if the link doesn't exist, no VFIO for us */ + if (ret < 0) + return 0; + + ret = rte_strsplit(filename, sizeof(filename), + tok, RTE_DIM(tok), '/'); + + if (ret <= 0) { + EAL_LOG(ERR, "%s cannot get IOMMU group", dev_addr); + return -1; + } + + /* IOMMU group is always the last token */ + errno = 0; + group_tok = tok[ret - 1]; + end = group_tok; + group_num = strtol(group_tok, &end, 10); + if (end == group_tok || *end != '\0' || errno != 0) { + EAL_LOG(ERR, "%s error parsing IOMMU number!", dev_addr); + return -1; + } + *iommu_group_num = group_num; + + return 1; +} diff --git a/lib/eal/linux/eal_vfio_mp_sync.c b/lib/eal/linux/eal_vfio_mp_sync.c index 3eaeef2fc8..9a07d35023 100644 --- a/lib/eal/linux/eal_vfio_mp_sync.c +++ b/lib/eal/linux/eal_vfio_mp_sync.c @@ -32,21 +32,32 @@ vfio_mp_primary(const struct rte_mp_msg *msg, const void *peer) switch (m->req) { case SOCKET_REQ_GROUP: + { + struct container *cfg = vfio_cfg.default_cfg; + struct vfio_group *grp; + + if (vfio_cfg.mode != RTE_VFIO_MODE_GROUP && + vfio_cfg.mode != RTE_VFIO_MODE_NOIOMMU) { + EAL_LOG(ERR, "VFIO not initialized in group mode"); + r->result = SOCKET_ERR; + break; + } + r->req = SOCKET_REQ_GROUP; r->group_num = m->group_num; - fd = vfio_get_group_fd_by_num(m->group_num); - if (fd < 0 && fd != -ENOENT) - r->result = SOCKET_ERR; - else if (fd == -ENOENT) - /* if VFIO group exists but isn't bound to VFIO driver */ + grp = vfio_group_get_by_num(cfg, m->group_num); + if (grp == NULL) { + /* group doesn't exist in primary */ r->result = SOCKET_NO_FD; - else { - /* if group exists and is bound to VFIO driver */ + } else { + /* group exists and is bound to VFIO driver */ + fd = grp->fd; r->result = SOCKET_OK; reply.num_fds = 1; reply.fds[0] = fd; } break; + } case SOCKET_REQ_CONTAINER: r->req = SOCKET_REQ_CONTAINER; fd = rte_vfio_get_container_fd(); @@ -54,6 +65,7 @@ vfio_mp_primary(const struct rte_mp_msg *msg, const void *peer) r->result = SOCKET_ERR; else { r->result = SOCKET_OK; + r->mode = vfio_cfg.mode; reply.num_fds = 1; reply.fds[0] = fd; } @@ -62,6 +74,13 @@ vfio_mp_primary(const struct rte_mp_msg *msg, const void *peer) { int iommu_type_id; + if (vfio_cfg.mode != RTE_VFIO_MODE_GROUP && + vfio_cfg.mode != RTE_VFIO_MODE_NOIOMMU) { + EAL_LOG(ERR, "VFIO not initialized in group mode"); + r->result = SOCKET_ERR; + break; + } + r->req = SOCKET_REQ_IOMMU_TYPE; iommu_type_id = vfio_get_iommu_type(); @@ -90,8 +109,11 @@ vfio_mp_sync_setup(void) { if (rte_eal_process_type() == RTE_PROC_PRIMARY) { int ret = rte_mp_action_register(EAL_VFIO_MP, vfio_mp_primary); - if (ret && rte_errno != ENOTSUP) + if (ret && rte_errno != ENOTSUP) { + EAL_LOG(ERR, "Multiprocess sync setup failed: %d (%s)", + rte_errno, rte_strerror(rte_errno)); return -1; + } } return 0; diff --git a/lib/eal/linux/meson.build b/lib/eal/linux/meson.build index e99ebed256..75a9afdd03 100644 --- a/lib/eal/linux/meson.build +++ b/lib/eal/linux/meson.build @@ -16,6 +16,7 @@ sources += files( 'eal_thread.c', 'eal_timer.c', 'eal_vfio.c', + 'eal_vfio_group.c', 'eal_vfio_mp_sync.c', ) -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* [PATCH v7 13/18] bus/pci: use the new VFIO mode API 2026-02-26 14:17 ` [PATCH v7 00/18] Support VFIO cdev API in DPDK Anatoly Burakov ` (11 preceding siblings ...) 2026-02-26 14:17 ` [PATCH v7 12/18] vfio: cleanup and refactor Anatoly Burakov @ 2026-02-26 14:17 ` Anatoly Burakov 2026-02-26 14:17 ` [PATCH v7 14/18] bus/fslmc: " Anatoly Burakov ` (5 subsequent siblings) 18 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2026-02-26 14:17 UTC (permalink / raw) To: dev, Chenbo Xia, Nipun Gupta Use the new VFIO mode API to query no-IOMMU status. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- drivers/bus/pci/linux/pci.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/bus/pci/linux/pci.c b/drivers/bus/pci/linux/pci.c index 2ffac82e94..6a5035ac62 100644 --- a/drivers/bus/pci/linux/pci.c +++ b/drivers/bus/pci/linux/pci.c @@ -597,7 +597,7 @@ pci_device_iova_mode(const struct rte_pci_driver *pdrv, static int is_vfio_noiommu_enabled = -1; if (is_vfio_noiommu_enabled == -1) { - if (rte_vfio_noiommu_is_enabled() == 1) + if (rte_vfio_get_mode() == RTE_VFIO_MODE_NOIOMMU) is_vfio_noiommu_enabled = 1; else is_vfio_noiommu_enabled = 0; -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* [PATCH v7 14/18] bus/fslmc: use the new VFIO mode API 2026-02-26 14:17 ` [PATCH v7 00/18] Support VFIO cdev API in DPDK Anatoly Burakov ` (12 preceding siblings ...) 2026-02-26 14:17 ` [PATCH v7 13/18] bus/pci: use the new VFIO mode API Anatoly Burakov @ 2026-02-26 14:17 ` Anatoly Burakov 2026-02-26 14:17 ` [PATCH v7 15/18] net/hinic3: " Anatoly Burakov ` (4 subsequent siblings) 18 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2026-02-26 14:17 UTC (permalink / raw) To: dev, Hemant Agrawal, Sachin Saxena FSLMC bus only supports operating in group mode, and relies on no-IOMMU mode checks. Use the new VFIO API's to query no-IOMMU status, as well as protect the bus from initializing in non-group mode. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com> --- drivers/bus/fslmc/fslmc_bus.c | 10 +++++++++- drivers/bus/fslmc/fslmc_vfio.c | 2 +- 2 files changed, 10 insertions(+), 2 deletions(-) diff --git a/drivers/bus/fslmc/fslmc_bus.c b/drivers/bus/fslmc/fslmc_bus.c index abdb0ad50d..d486c41ef1 100644 --- a/drivers/bus/fslmc/fslmc_bus.c +++ b/drivers/bus/fslmc/fslmc_bus.c @@ -335,6 +335,13 @@ rte_fslmc_scan(void) goto scan_fail; } + /* for container groups to work, VFIO must be in group mode */ + if (rte_vfio_get_mode() != RTE_VFIO_MODE_GROUP && + rte_vfio_get_mode() != RTE_VFIO_MODE_NOIOMMU) { + ret = -EINVAL; + goto scan_fail; + } + ret = fslmc_get_container_group(group_name, &groupid); if (ret != 0) goto scan_fail; @@ -582,7 +589,8 @@ rte_dpaa2_get_iommu_class(void) return RTE_IOVA_DC; /* check if all devices on the bus support Virtual addressing or not */ - if (fslmc_all_device_support_iova() != 0 && rte_vfio_noiommu_is_enabled() == 0) + if (fslmc_all_device_support_iova() != 0 && + rte_vfio_get_mode() != RTE_VFIO_MODE_NOIOMMU) return RTE_IOVA_VA; return RTE_IOVA_PA; diff --git a/drivers/bus/fslmc/fslmc_vfio.c b/drivers/bus/fslmc/fslmc_vfio.c index 4b11b38322..afde967e34 100644 --- a/drivers/bus/fslmc/fslmc_vfio.c +++ b/drivers/bus/fslmc/fslmc_vfio.c @@ -192,7 +192,7 @@ fslmc_vfio_add_group(int vfio_group_fd, group->fd = vfio_group_fd; group->groupid = iommu_group_num; rte_strscpy(group->group_name, group_name, sizeof(group->group_name)); - if (rte_vfio_noiommu_is_enabled() > 0) + if (rte_vfio_get_mode() == RTE_VFIO_MODE_NOIOMMU) group->iommu_type = VFIO_NOIOMMU_IOMMU; else group->iommu_type = VFIO_TYPE1_IOMMU; -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* [PATCH v7 15/18] net/hinic3: use the new VFIO mode API 2026-02-26 14:17 ` [PATCH v7 00/18] Support VFIO cdev API in DPDK Anatoly Burakov ` (13 preceding siblings ...) 2026-02-26 14:17 ` [PATCH v7 14/18] bus/fslmc: " Anatoly Burakov @ 2026-02-26 14:17 ` Anatoly Burakov 2026-02-26 14:17 ` [PATCH v7 16/18] net/ntnic: " Anatoly Burakov ` (3 subsequent siblings) 18 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2026-02-26 14:17 UTC (permalink / raw) To: dev, Feifei Wang Use new VFIO mode API to query no-IOMMU status. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- drivers/net/hinic3/base/hinic3_hwdev.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/net/hinic3/base/hinic3_hwdev.c b/drivers/net/hinic3/base/hinic3_hwdev.c index 668bbf4a0e..c5bf2bc8a4 100644 --- a/drivers/net/hinic3/base/hinic3_hwdev.c +++ b/drivers/net/hinic3/base/hinic3_hwdev.c @@ -77,7 +77,7 @@ bool hinic3_is_vfio_iommu_enable(const struct rte_eth_dev *rte_dev) { return ((RTE_ETH_DEV_TO_PCI(rte_dev)->kdrv == RTE_PCI_KDRV_VFIO) && - (rte_vfio_noiommu_is_enabled() != 1)); + (rte_vfio_get_mode() != RTE_VFIO_MODE_NOIOMMU)); } int -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* [PATCH v7 16/18] net/ntnic: use the new VFIO mode API 2026-02-26 14:17 ` [PATCH v7 00/18] Support VFIO cdev API in DPDK Anatoly Burakov ` (14 preceding siblings ...) 2026-02-26 14:17 ` [PATCH v7 15/18] net/hinic3: " Anatoly Burakov @ 2026-02-26 14:17 ` Anatoly Burakov 2026-02-26 14:17 ` [PATCH v7 17/18] vfio: remove no-IOMMU check API Anatoly Burakov ` (2 subsequent siblings) 18 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2026-02-26 14:17 UTC (permalink / raw) To: dev, Christian Koue Muf, Serhii Iliushyk Use new VFIO mode API to query no-IOMMU status. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- drivers/net/ntnic/ntnic_ethdev.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/net/ntnic/ntnic_ethdev.c b/drivers/net/ntnic/ntnic_ethdev.c index 89e751e7e0..7fda61d3ff 100644 --- a/drivers/net/ntnic/ntnic_ethdev.c +++ b/drivers/net/ntnic/ntnic_ethdev.c @@ -2691,7 +2691,7 @@ nthw_pci_probe(struct rte_pci_driver *pci_drv __rte_unused, (pci_dev->device.devargs->data ? pci_dev->device.devargs->data : "NULL")); } - const int n_rte_vfio_no_io_mmu_enabled = rte_vfio_noiommu_is_enabled(); + const int n_rte_vfio_no_io_mmu_enabled = rte_vfio_get_mode() == RTE_VFIO_MODE_NOIOMMU; NT_LOG(DBG, NTNIC, "vfio_no_iommu_enabled=%d", n_rte_vfio_no_io_mmu_enabled); if (n_rte_vfio_no_io_mmu_enabled) { -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* [PATCH v7 17/18] vfio: remove no-IOMMU check API 2026-02-26 14:17 ` [PATCH v7 00/18] Support VFIO cdev API in DPDK Anatoly Burakov ` (15 preceding siblings ...) 2026-02-26 14:17 ` [PATCH v7 16/18] net/ntnic: " Anatoly Burakov @ 2026-02-26 14:17 ` Anatoly Burakov 2026-02-26 14:17 ` [PATCH v7 18/18] vfio: introduce cdev mode Anatoly Burakov 2026-05-01 22:45 ` [PATCH v7 00/18] Support VFIO cdev API in DPDK Stephen Hemminger 18 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2026-02-26 14:17 UTC (permalink / raw) To: dev, Bruce Richardson The `rte_vfio_noiommu_is_enabled()` check has now been replaced by the new mode API, so remove it. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- lib/eal/freebsd/eal.c | 6 ------ lib/eal/include/rte_vfio.h | 14 -------------- lib/eal/linux/eal_vfio.c | 7 ------- 3 files changed, 27 deletions(-) diff --git a/lib/eal/freebsd/eal.c b/lib/eal/freebsd/eal.c index 86ac148c88..84bb618484 100644 --- a/lib/eal/freebsd/eal.c +++ b/lib/eal/freebsd/eal.c @@ -844,12 +844,6 @@ int rte_vfio_is_enabled(__rte_unused const char *modname) return 0; } -RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_noiommu_is_enabled) -int rte_vfio_noiommu_is_enabled(void) -{ - return 0; -} - RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_group_num) int rte_vfio_get_group_num(__rte_unused const char *sysfs_base, diff --git a/lib/eal/include/rte_vfio.h b/lib/eal/include/rte_vfio.h index 0af41c3610..c4ba0e5cda 100644 --- a/lib/eal/include/rte_vfio.h +++ b/lib/eal/include/rte_vfio.h @@ -170,20 +170,6 @@ __rte_internal enum rte_vfio_mode rte_vfio_get_mode(void); -/** - * @internal - * Check if VFIO NOIOMMU mode is enabled. - * - * This function is only relevant on Linux in group mode. - * - * @return - * 1 if enabled. - * 0 if not enabled or not supported. - */ -__rte_internal -int -rte_vfio_noiommu_is_enabled(void); - /** * @internal * Parse IOMMU group number for a device. diff --git a/lib/eal/linux/eal_vfio.c b/lib/eal/linux/eal_vfio.c index 8545552308..2e9ad7dbb4 100644 --- a/lib/eal/linux/eal_vfio.c +++ b/lib/eal/linux/eal_vfio.c @@ -1310,13 +1310,6 @@ container_dma_unmap(struct container *cfg, uint64_t vaddr, uint64_t iova, return ret; } -RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_noiommu_is_enabled) -int -rte_vfio_noiommu_is_enabled(void) -{ - return vfio_cfg.mode == RTE_VFIO_MODE_NOIOMMU; -} - RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_container_create) int rte_vfio_container_create(void) -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* [PATCH v7 18/18] vfio: introduce cdev mode 2026-02-26 14:17 ` [PATCH v7 00/18] Support VFIO cdev API in DPDK Anatoly Burakov ` (16 preceding siblings ...) 2026-02-26 14:17 ` [PATCH v7 17/18] vfio: remove no-IOMMU check API Anatoly Burakov @ 2026-02-26 14:17 ` Anatoly Burakov 2026-05-01 22:45 ` [PATCH v7 00/18] Support VFIO cdev API in DPDK Stephen Hemminger 18 siblings, 0 replies; 42+ messages in thread From: Anatoly Burakov @ 2026-02-26 14:17 UTC (permalink / raw) To: dev, Bruce Richardson Add support for VFIO cdev (also known as IOMMUFD) API. The group API is now considered legacy in the kernel, and all further development is expected to happen in IOMMUFD infrastructure. To assist any future use of VFIO cdev mode for custom behavior, also introduce "get device number" API, which is kind-of-but-not-really similar to the concept of IOMMU group. Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- lib/eal/freebsd/eal.c | 10 + lib/eal/include/rte_vfio.h | 31 +++ lib/eal/linux/eal_vfio.c | 210 +++++++++++++++++ lib/eal/linux/eal_vfio.h | 29 ++- lib/eal/linux/eal_vfio_cdev.c | 390 +++++++++++++++++++++++++++++++ lib/eal/linux/eal_vfio_mp_sync.c | 42 ++++ lib/eal/linux/meson.build | 1 + 7 files changed, 711 insertions(+), 2 deletions(-) create mode 100644 lib/eal/linux/eal_vfio_cdev.c diff --git a/lib/eal/freebsd/eal.c b/lib/eal/freebsd/eal.c index 84bb618484..a49a8a2b4c 100644 --- a/lib/eal/freebsd/eal.c +++ b/lib/eal/freebsd/eal.c @@ -925,3 +925,13 @@ rte_vfio_get_mode(void) { return RTE_VFIO_MODE_NONE; } + +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_device_num) +int +rte_vfio_get_device_num(__rte_unused const char *sysfs_base, + __rte_unused const char *dev_addr, + __rte_unused int *vfio_device_num) +{ + rte_errno = ENOTSUP; + return -1; +} diff --git a/lib/eal/include/rte_vfio.h b/lib/eal/include/rte_vfio.h index c4ba0e5cda..502a68c948 100644 --- a/lib/eal/include/rte_vfio.h +++ b/lib/eal/include/rte_vfio.h @@ -28,6 +28,8 @@ extern "C" { #define RTE_VFIO_DIR "/dev/vfio" #define RTE_VFIO_CONTAINER_PATH "/dev/vfio/vfio" +#define RTE_VFIO_IOMMUFD_PATH "/dev/iommu" +#define RTE_VFIO_CDEV_DEVICES_PATH "/dev/vfio/devices" #define RTE_VFIO_GROUP_FMT "/dev/vfio/%u" #define RTE_VFIO_NOIOMMU_GROUP_FMT "/dev/vfio/noiommu-%u" #define RTE_VFIO_NOIOMMU_MODE "/sys/module/vfio/parameters/enable_unsafe_noiommu_mode" @@ -48,11 +50,13 @@ struct vfio_device_info; * - RTE_VFIO_MODE_NONE: VFIO is not enabled. * - RTE_VFIO_MODE_GROUP: Legacy group mode. * - RTE_VFIO_MODE_NOIOMMU: Unsafe no-IOMMU mode. + * - RTE_VFIO_MODE_CDEV: Character device mode. */ enum rte_vfio_mode { RTE_VFIO_MODE_NONE = 0, /**< VFIO not enabled */ RTE_VFIO_MODE_GROUP, /**< Group mode */ RTE_VFIO_MODE_NOIOMMU, /**< Group mode with no IOMMU protection */ + RTE_VFIO_MODE_CDEV, /**< Device mode */ }; /** @@ -197,6 +201,33 @@ __rte_internal int rte_vfio_get_group_num(const char *sysfs_base, const char *dev_addr, int *iommu_group_num); +/** + * @internal + * Parse VFIO cdev device number for a device. + * + * This function is only relevant on Linux in cdev mode. + * + * @param sysfs_base + * Sysfs path prefix. + * @param dev_addr + * Device identifier. + * @param vfio_device_num + * Pointer to where VFIO cdev device number will be stored. + * + * @return + * 0 on success. + * <0 on failure, rte_errno is set. + * + * Possible rte_errno values include: + * - ENODEV - Device not managed by VFIO. + * - EINVAL - Invalid parameters. + * - ENXIO - VFIO support not initialized. + * - ENOTSUP - Unsupported VFIO mode. + */ +__rte_internal +int +rte_vfio_get_device_num(const char *sysfs_base, const char *dev_addr, int *vfio_device_num); + /** * @internal * Get device information. diff --git a/lib/eal/linux/eal_vfio.c b/lib/eal/linux/eal_vfio.c index 2e9ad7dbb4..e1d1de494b 100644 --- a/lib/eal/linux/eal_vfio.c +++ b/lib/eal/linux/eal_vfio.c @@ -369,6 +369,20 @@ vfio_container_get_by_group_num(int group_num) return NULL; } +static struct container * +vfio_container_get_by_dev_num(int dev_num) +{ + struct container *cfg; + struct vfio_device *dev; + + CONTAINER_FOREACH_ACTIVE(cfg) { + DEVICE_FOREACH_ACTIVE(cfg, dev) + if (dev->dev_num == dev_num) + return cfg; + } + return NULL; +} + static struct container * vfio_container_create(void) { @@ -538,6 +552,55 @@ vfio_setup_dma_mem(struct container *cfg) return 0; } +static enum vfio_result +vfio_cdev_assign_device(struct container *cfg, const char *sysfs_base, + const char *dev_addr, struct vfio_device **out_dev) +{ + struct vfio_device *dev, *found_dev; + enum vfio_result res; + int dev_num, ret; + + /* get the cdev device number from sysfs */ + ret = vfio_cdev_get_device_num(sysfs_base, dev_addr, &dev_num); + if (ret < 0) { + EAL_LOG(ERR, "Failed to get cdev device number for %s", dev_addr); + return VFIO_ERROR; + } else if (ret == 0) { + EAL_LOG(ERR, "Device %s not bound to vfio-pci cdev", dev_addr); + return VFIO_NOT_MANAGED; + } + + /* do we already have this device? */ + found_dev = vfio_cdev_get_dev_by_num(cfg, dev_num); + if (found_dev != NULL) { + EAL_LOG(ERR, "Device %s already assigned to this container", dev_addr); + *out_dev = found_dev; + return VFIO_EXISTS; + } + /* create new device structure */ + dev = vfio_device_create(cfg); + if (dev == NULL) { + EAL_LOG(ERR, "No space to track new VFIO cdev device"); + return VFIO_NO_SPACE; + } + /* store device number */ + dev->dev_num = dev_num; + + /* set up our device now and store it in config */ + ret = vfio_cdev_setup_device(cfg, dev); + if (ret < 0) { + EAL_LOG(ERR, "Cannot setup cdev device %s", dev_addr); + res = VFIO_ERROR; + goto err; + } + *out_dev = dev; + return VFIO_SUCCESS; + +err: + vfio_device_erase(cfg, dev); + return res; +} + static enum vfio_result vfio_group_assign_device(struct container *cfg, const char *sysfs_base, const char *dev_addr, struct vfio_device **out_dev) @@ -695,6 +758,49 @@ rte_vfio_container_assign_device(int container_fd, const char *sysfs_base, const return -1; } + /* + * The device-to-container assignment is a complex problem to solve, for the following + * reasons: + * + * 1. PCI infrastructure is decoupled from VFIO, so PCI does not know anything about VFIO + * + * This means that while 99% of VFIO usage is PCI-related, we cannot communicate to PCI that + * we want to map a particular device using a particular container. Previously, this was + * achieved using back-channel communication via IOMMU group binding, so that whenever PCI + * map actually happens, VFIO knows which container to use, so this is roughly the model we + * are going with. + * + * 2. VFIO cannot depend on PCI because VFIO is in EAL + * + * We cannot "assign" a PCI device to container using rte_pci_device pointer because VFIO + * cannot depend on PCI definitions, nor can't we even assume that our device is in fact a + * PCI device, even though in practice this is true (at the time of this writing, FSLMC is + * the only bus doing non-PCI VFIO mappings, but FSLMC manages all VFIO infrastructure by + * itself, so in practice even counting FSLMC bus, we're always dealing with PCI devices). + * + * 3. The "assignment" means different things for group and cdev mode + * + * In group mode, to "bind" a device to a specific container, it is enough to bind its + * IOMMU group, so that when rte_vfio_setup_device() is called, we simply retrieve already + * existing group, and through that we figure out which container to use. + * + * For cdev mode, there are no "groups", so "assignment" either means we store some kind of + * uniquely identifying token (such as device number, or an opaque pointer), or we simply + * open the device straight away, and when rte_vfio_setup_device() comes we simply return + * the fd that was already opened at assign. + * + * Doing it the latter way (opening the device at assign for both group and cdev modes) + * actually solves all of these problems, so that's what we're going to do - the device + * setup API call will actually just assign the device to default container, while release + * will automatically cleanup and unassign anything that needs unassigned. There will be no + * "unassign" call, as it is not necessary. + * + * There is one downside for group mode when adding duplicate devices: to get to device fd, + * we need to go through the entire codepath before we arrive at fd only to realize it was + * already opened earlier, but this is acceptable compromise for unifying the API around + * device assignment. + */ + if (vfio_cfg.mode == RTE_VFIO_MODE_NONE) { EAL_LOG(ERR, "VFIO support not initialized"); rte_errno = ENXIO; @@ -715,6 +821,9 @@ rte_vfio_container_assign_device(int container_fd, const char *sysfs_base, const case RTE_VFIO_MODE_NOIOMMU: res = vfio_group_assign_device(cfg, sysfs_base, dev_addr, &dev); break; + case RTE_VFIO_MODE_CDEV: + res = vfio_cdev_assign_device(cfg, sysfs_base, dev_addr, &dev); + break; default: EAL_LOG(ERR, "Unsupported VFIO mode"); res = VFIO_NOT_SUPPORTED; @@ -787,6 +896,24 @@ rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr, res = vfio_group_assign_device(cfg, sysfs_base, dev_addr, &dev); break; } + case RTE_VFIO_MODE_CDEV: + { + int dev_num; + + /* find device number */ + ret = vfio_cdev_get_device_num(sysfs_base, dev_addr, &dev_num); + if (ret < 0) + goto assign_fail; + else if (ret == 0) + goto not_managed; + + cfg = vfio_container_get_by_dev_num(dev_num); + if (cfg == NULL) + cfg = vfio_cfg.default_cfg; + + res = vfio_cdev_assign_device(cfg, sysfs_base, dev_addr, &dev); + break; + } default: EAL_LOG(ERR, "Unsupported VFIO mode"); rte_errno = ENOTSUP; @@ -906,6 +1033,12 @@ rte_vfio_release_device(const char *sysfs_base __rte_unused, } break; } + case RTE_VFIO_MODE_CDEV: + { + /* for cdev, just erase the device and we're done */ + vfio_device_erase(cfg, dev); + break; + } default: EAL_LOG(ERR, "Unsupported VFIO mode"); rte_errno = ENOTSUP; @@ -971,6 +1104,9 @@ vfio_select_mode(void) if (vfio_sync_mode(cfg, &mode) < 0) goto err; + /* if primary is in cdev mode, we need to sync ioas as well */ + if (mode == RTE_VFIO_MODE_CDEV && vfio_cdev_sync_ioas(cfg) < 0) + goto err; /* primary handles DMA setup for default containers */ group_cfg->dma_setup_done = true; @@ -990,6 +1126,19 @@ vfio_select_mode(void) return RTE_VFIO_MODE_NOIOMMU; return RTE_VFIO_MODE_GROUP; } + EAL_LOG(DEBUG, "VFIO group mode not available, trying cdev mode..."); + /* try cdev mode */ + if (vfio_cdev_enable(cfg) == 0) { + if (vfio_cdev_setup_ioas(cfg) < 0) + goto err_mpsync; + if (vfio_setup_dma_mem(cfg) < 0) + goto err_mpsync; + if (vfio_register_mem_event_callback() < 0) + goto err_mpsync; + + return RTE_VFIO_MODE_CDEV; + } + EAL_LOG(DEBUG, "VFIO cdev mode not available"); err_mpsync: vfio_mp_sync_cleanup(); err: @@ -1004,6 +1153,7 @@ vfio_mode_to_str(enum rte_vfio_mode mode) switch (mode) { case RTE_VFIO_MODE_GROUP: return "group"; case RTE_VFIO_MODE_NOIOMMU: return "noiommu"; + case RTE_VFIO_MODE_CDEV: return "cdev"; default: return "not initialized"; } } @@ -1143,6 +1293,40 @@ rte_vfio_get_group_num(const char *sysfs_base, const char *dev_addr, int *iommu_ return 0; } +RTE_EXPORT_INTERNAL_SYMBOL(rte_vfio_get_device_num) +int +rte_vfio_get_device_num(const char *sysfs_base, const char *dev_addr, int *device_num) +{ + int ret; + + if (sysfs_base == NULL || dev_addr == NULL || device_num == NULL) { + rte_errno = EINVAL; + return -1; + } + + if (vfio_cfg.mode == RTE_VFIO_MODE_NONE) { + EAL_LOG(ERR, "VFIO support not initialized"); + rte_errno = ENXIO; + return -1; + } + + if (vfio_cfg.mode != RTE_VFIO_MODE_CDEV) { + EAL_LOG(ERR, "VFIO not initialized in cdev mode"); + rte_errno = ENOTSUP; + return -1; + } + + ret = vfio_cdev_get_device_num(sysfs_base, dev_addr, device_num); + if (ret < 0) { + rte_errno = EINVAL; + return -1; + } else if (ret == 0) { + rte_errno = ENODEV; + return -1; + } + return 0; +} + static int vfio_dma_mem_map(struct container *cfg, uint64_t vaddr, uint64_t iova, uint64_t len, int do_map) @@ -1342,6 +1526,25 @@ rte_vfio_container_create(void) cfg->container_fd = container_fd; break; } + case RTE_VFIO_MODE_CDEV: + { + /* Open new iommufd for custom container */ + container_fd = vfio_cdev_get_iommufd(); + if (container_fd < 0) { + EAL_LOG(ERR, "Cannot open iommufd for cdev container"); + rte_errno = EIO; + goto err; + } + cfg->container_fd = container_fd; + + /* Set up IOAS for this container */ + if (vfio_cdev_setup_ioas(cfg) < 0) { + EAL_LOG(ERR, "Cannot setup IOAS for cdev container"); + rte_errno = EIO; + goto err; + } + break; + } default: EAL_LOG(NOTICE, "Unsupported VFIO mode"); rte_errno = ENOTSUP; @@ -1400,6 +1603,13 @@ rte_vfio_container_destroy(int container_fd) vfio_group_erase(cfg, grp); } break; + case RTE_VFIO_MODE_CDEV: + /* erase all devices */ + DEVICE_FOREACH_ACTIVE(cfg, dev) { + EAL_LOG(DEBUG, "Device vfio%d still open, closing", dev->dev_num); + vfio_device_erase(cfg, dev); + } + break; default: EAL_LOG(ERR, "Unsupported VFIO mode"); rte_errno = ENOTSUP; diff --git a/lib/eal/linux/eal_vfio.h b/lib/eal/linux/eal_vfio.h index 68d3a3ec6e..52cb3a0e08 100644 --- a/lib/eal/linux/eal_vfio.h +++ b/lib/eal/linux/eal_vfio.h @@ -48,7 +48,10 @@ struct vfio_group { /* device tracking (common for group and cdev modes) */ struct vfio_device { bool active; - int group; /**< back-reference to group list (group mode) */ + union { + int group; /**< back-reference to group list (group mode) */ + int dev_num; /**< device number, e.g., X in /dev/vfio/devices/vfioX (cdev mode) */ + }; int fd; }; @@ -61,12 +64,20 @@ struct vfio_group_config { struct vfio_group groups[RTE_MAX_VFIO_GROUPS]; }; +/* cdev mode specific configuration */ +struct vfio_cdev_config { + uint32_t ioas_id; +}; + /* per-container configuration */ struct container { bool active; int container_fd; struct user_mem_maps mem_maps; - struct vfio_group_config group_cfg; + union { + struct vfio_group_config group_cfg; + struct vfio_cdev_config cdev_cfg; + }; int n_devices; struct vfio_device devices[RTE_MAX_VFIO_DEVICES]; }; @@ -160,12 +171,24 @@ int vfio_group_setup_iommu(struct container *cfg); int vfio_group_setup_device_fd(const char *dev_addr, struct vfio_group *grp, struct vfio_device *dev); +/* cdev mode functions */ +int vfio_cdev_enable(struct container *cfg); +int vfio_cdev_setup_ioas(struct container *cfg); +int vfio_cdev_sync_ioas(struct container *cfg); +int vfio_cdev_get_iommufd(void); +int vfio_cdev_get_device_num(const char *sysfs_base, const char *dev_addr, + int *cdev_dev_num); +struct vfio_device *vfio_cdev_get_dev_by_num(struct container *cfg, int cdev_dev_num); +int vfio_cdev_setup_device(struct container *cfg, struct vfio_device *dev); + #define VFIO_MEM_EVENT_CLB_NAME "vfio_mem_event_clb" #define EAL_VFIO_MP "eal_vfio_mp_sync" #define SOCKET_REQ_CONTAINER 0x100 #define SOCKET_REQ_GROUP 0x200 #define SOCKET_REQ_IOMMU_TYPE 0x400 +#define SOCKET_REQ_CDEV 0x800 +#define SOCKET_REQ_IOAS_ID 0x1000 #define SOCKET_OK 0x0 #define SOCKET_NO_FD 0x1 #define SOCKET_ERR 0xFF @@ -176,6 +199,8 @@ struct vfio_mp_param { union { int group_num; int iommu_type_id; + int cdev_dev_num; + int ioas_id; enum rte_vfio_mode mode; }; }; diff --git a/lib/eal/linux/eal_vfio_cdev.c b/lib/eal/linux/eal_vfio_cdev.c new file mode 100644 index 0000000000..ce61a97853 --- /dev/null +++ b/lib/eal/linux/eal_vfio_cdev.c @@ -0,0 +1,390 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright(c) 2010-2025 Intel Corporation + */ + +#include <dirent.h> +#include <fcntl.h> +#include <limits.h> +#include <stdlib.h> +#include <string.h> +#include <sys/ioctl.h> +#include <unistd.h> + +#include <uapi/linux/iommufd.h> +#include <uapi/linux/vfio.h> + +#include <rte_log.h> +#include <rte_errno.h> +#include <rte_memory.h> +#include <rte_string_fns.h> + +#include "eal_vfio.h" +#include "eal_private.h" +#include "eal_internal_cfg.h" + +static int vfio_cdev_dma_map(struct container *cfg); +static int vfio_cdev_dma_mem_map(struct container *cfg, uint64_t vaddr, + uint64_t iova, uint64_t len, int do_map); + +/* IOMMUFD cdev mode IOMMU operations */ +static const struct vfio_iommu_ops iommufd_ops = { + .type_id = 0, /* cdev mode doesn't use type_id */ + .name = "IOMMUFD", + .partial_unmap = false, + .dma_map_func = &vfio_cdev_dma_map, + .dma_user_map_func = &vfio_cdev_dma_mem_map +}; + +static int +vfio_cdev_dma_mem_map(struct container *cfg, uint64_t vaddr, uint64_t iova, + uint64_t len, int do_map) +{ + struct iommu_ioas_map ioas_map; + struct iommu_ioas_unmap ioas_unmap; + int ret; + + if (do_map != 0) { + memset(&ioas_map, 0, sizeof(ioas_map)); + ioas_map.size = sizeof(struct iommu_ioas_map); + ioas_map.flags = IOMMU_IOAS_MAP_FIXED_IOVA | + IOMMU_IOAS_MAP_READABLE | + IOMMU_IOAS_MAP_WRITEABLE; + ioas_map.ioas_id = cfg->cdev_cfg.ioas_id; + ioas_map.user_va = vaddr; + ioas_map.length = len; + ioas_map.iova = iova; + + ret = ioctl(cfg->container_fd, IOMMU_IOAS_MAP, &ioas_map); + if (ret) { + /** + * In case the mapping was already done EEXIST will be + * returned from kernel. + */ + if (errno == EEXIST) { + EAL_LOG(DEBUG, + "Memory segment is already mapped, skipping"); + } else { + EAL_LOG(ERR, + "Cannot set up DMA remapping, error " + "%i (%s)", errno, strerror(errno)); + return -1; + } + } + } else { + memset(&ioas_unmap, 0, sizeof(ioas_unmap)); + ioas_unmap.size = sizeof(struct iommu_ioas_unmap); + ioas_unmap.ioas_id = cfg->cdev_cfg.ioas_id; + ioas_unmap.length = len; + ioas_unmap.iova = iova; + + ret = ioctl(cfg->container_fd, IOMMU_IOAS_UNMAP, &ioas_unmap); + if (ret) { + EAL_LOG(ERR, "Cannot clear DMA remapping, error " + "%i (%s)", errno, strerror(errno)); + return -1; + } + } + + return 0; +} + +static int +cdev_map(const struct rte_memseg_list *msl, const struct rte_memseg *ms, + void *arg) +{ + struct container *cfg = arg; + + /* skip external memory that isn't a heap */ + if (msl->external && !msl->heap) + return 0; + + /* skip any segments with invalid IOVA addresses */ + if (ms->iova == RTE_BAD_IOVA) + return 0; + + return vfio_cdev_dma_mem_map(cfg, ms->addr_64, ms->iova, ms->len, 1); +} + +static int +vfio_cdev_dma_map(struct container *cfg) +{ + return rte_memseg_walk(cdev_map, cfg); +} + +int +vfio_cdev_sync_ioas(struct container *cfg) +{ + struct rte_mp_msg mp_req, *mp_rep; + struct rte_mp_reply mp_reply = {0}; + struct timespec ts = {.tv_sec = 5, .tv_nsec = 0}; + struct vfio_mp_param *p = (struct vfio_mp_param *)mp_req.param; + + p->req = SOCKET_REQ_IOAS_ID; + rte_strscpy(mp_req.name, EAL_VFIO_MP, sizeof(mp_req.name)); + mp_req.len_param = sizeof(*p); + mp_req.num_fds = 0; + + if (rte_mp_request_sync(&mp_req, &mp_reply, &ts) == 0 && mp_reply.nb_received == 1) { + mp_rep = &mp_reply.msgs[0]; + p = (struct vfio_mp_param *)mp_rep->param; + if (p->result == SOCKET_OK && mp_rep->num_fds == 0) { + cfg->cdev_cfg.ioas_id = p->ioas_id; + free(mp_reply.msgs); + return 0; + } + } + + free(mp_reply.msgs); + EAL_LOG(ERR, "Cannot request ioas_id"); + return -1; +} + +int +vfio_cdev_setup_ioas(struct container *cfg) +{ + struct iommu_ioas_alloc ioas_alloc; + int ret; + + /* Allocate an IOAS */ + memset(&ioas_alloc, 0, sizeof(ioas_alloc)); + ioas_alloc.size = sizeof(struct iommu_ioas_alloc); + ioas_alloc.flags = 0; + + ret = ioctl(cfg->container_fd, IOMMU_IOAS_ALLOC, &ioas_alloc); + if (ret) { + EAL_LOG(ERR, "Cannot allocate IOAS, error %i (%s)", + errno, strerror(errno)); + return -1; + } + cfg->cdev_cfg.ioas_id = ioas_alloc.out_ioas_id; + + EAL_LOG(DEBUG, "Allocated IOAS with ID %u", cfg->cdev_cfg.ioas_id); + return 0; +} + +int +vfio_cdev_get_iommufd(void) +{ + int iommufd; + + /* if not requesting via mp, open iommufd locally */ + iommufd = open(RTE_VFIO_IOMMUFD_PATH, O_RDWR); + if (iommufd < 0) { + EAL_LOG(ERR, "Cannot open %s: %s", + RTE_VFIO_IOMMUFD_PATH, strerror(errno)); + return -1; + } + + return iommufd; +} + +int +vfio_cdev_enable(struct container *cfg) +{ + int iommufd; + + /* Check if iommufd device exists */ + if (access(RTE_VFIO_IOMMUFD_PATH, F_OK) != 0) { + EAL_LOG(DEBUG, + "IOMMUFD device does not exist, skipping VFIO cdev support..."); + return 1; + } + + /* open iommufd */ + iommufd = vfio_cdev_get_iommufd(); + if (iommufd < 0) + return -1; + + /* cdev mode does not have different IOMMU ops */ + vfio_cfg.ops = &iommufd_ops; + + cfg->container_fd = iommufd; + return 0; +} + +int +vfio_cdev_get_device_num(const char *sysfs_base, const char *dev_addr, int *cdev_dev_num) +{ + char linkname[PATH_MAX]; + char filename[PATH_MAX]; + char *dev_tok, *end; + int dev_num; + DIR *dir; + struct dirent *entry; + + memset(linkname, 0, sizeof(linkname)); + memset(filename, 0, sizeof(filename)); + + /* check if vfio-dev directory exists for this device */ + snprintf(linkname, sizeof(linkname), + "%s/%s/vfio-dev", sysfs_base, dev_addr); + + dir = opendir(linkname); + if (dir == NULL) { + /* device doesn't have vfio-dev, not bound to vfio-pci cdev */ + return 0; + } + + /* find vfioX entry in vfio-dev directory */ + while ((entry = readdir(dir)) != NULL) { + if (strncmp(entry->d_name, "vfio", 4) == 0) { + /* parse device number from vfioX */ + errno = 0; + dev_tok = entry->d_name + 4; /* skip "vfio" prefix */ + end = dev_tok; + dev_num = strtol(dev_tok, &end, 10); + if (end == dev_tok || *end != '\0' || errno != 0) { + EAL_LOG(ERR, "%s error parsing VFIO cdev device number!", + dev_addr); + closedir(dir); + return -1; + } + *cdev_dev_num = dev_num; + closedir(dir); + return 1; + } + } + + closedir(dir); + /* no vfio device found */ + return 0; +} + +struct vfio_device * +vfio_cdev_get_dev_by_num(struct container *cfg, int cdev_dev_num) +{ + struct vfio_device *dev; + /* find device handle */ + DEVICE_FOREACH_ACTIVE(cfg, dev) { + if (dev->dev_num != cdev_dev_num) + continue; + return dev; + } + return NULL; +} + +static int +cdev_open_device_fd(int cdev_dev_num) +{ + char devname[PATH_MAX] = {0}; + int dev_fd; + + snprintf(devname, sizeof(devname), "%s/vfio%d", + RTE_VFIO_CDEV_DEVICES_PATH, cdev_dev_num); + + dev_fd = open(devname, O_RDWR); + if (dev_fd < 0) { + EAL_LOG(ERR, "Cannot open %s: %s", devname, strerror(errno)); + return -1; + } + + return dev_fd; +} + +static int +cdev_attach_device_to_iommufd(struct container *cfg, struct vfio_device *dev) +{ + struct vfio_device_bind_iommufd bind = {0}; + struct vfio_device_attach_iommufd_pt attach = {0}; + rte_uuid_t vf_token; + + rte_eal_vfio_get_vf_token(vf_token); + + /* try with token first */ + if (!rte_uuid_is_null(vf_token)) { + bind.flags = VFIO_DEVICE_BIND_FLAG_TOKEN; + bind.token_uuid_ptr = (uintptr_t)&vf_token; + bind.argsz = sizeof(bind); + bind.iommufd = cfg->container_fd; + + /* this may fail because the kernel is too old */ + if (ioctl(dev->fd, VFIO_DEVICE_BIND_IOMMUFD, &bind) < 0) { + EAL_LOG(DEBUG, "Failed to bind device %d with VF token", dev->dev_num); + EAL_LOG(NOTICE, "Unable to use VF tokens with current kernel version."); + EAL_LOG(NOTICE, "Please use kernel >=6.17 or use group mode."); + /* erase the bind structure */ + bind = (struct vfio_device_bind_iommufd){0}; + } else { + goto attach; + } + } + bind.flags = 0; + bind.argsz = sizeof(bind); + bind.iommufd = cfg->container_fd; + + if (ioctl(dev->fd, VFIO_DEVICE_BIND_IOMMUFD, &bind) < 0) { + EAL_LOG(ERR, "Cannot bind device to IOMMUFD, error %i (%s)", + errno, strerror(errno)); + return -1; + } + +attach: + /* attach device to IOAS */ + attach.argsz = sizeof(attach); + attach.flags = 0; + attach.pt_id = cfg->cdev_cfg.ioas_id; + + if (ioctl(dev->fd, VFIO_DEVICE_ATTACH_IOMMUFD_PT, &attach) < 0) { + EAL_LOG(ERR, "Cannot attach device to IOAS, error %i (%s)", + errno, strerror(errno)); + return -1; + } + + return 0; +} + +static int +vfio_cdev_request_dev_fd(struct vfio_device *dev) +{ + struct rte_mp_msg mp_req, *mp_rep; + struct rte_mp_reply mp_reply = {0}; + struct timespec ts = {.tv_sec = 5, .tv_nsec = 0}; + struct vfio_mp_param *p = (struct vfio_mp_param *)mp_req.param; + int device_fd = -1; + + /* secondary process requests device fd from primary */ + p->req = SOCKET_REQ_CDEV; + p->cdev_dev_num = dev->dev_num; + rte_strscpy(mp_req.name, EAL_VFIO_MP, sizeof(mp_req.name)); + mp_req.len_param = sizeof(*p); + mp_req.num_fds = 0; + + if (rte_mp_request_sync(&mp_req, &mp_reply, &ts) == 0 && + mp_reply.nb_received == 1) { + mp_rep = &mp_reply.msgs[0]; + p = (struct vfio_mp_param *)mp_rep->param; + if (p->result == SOCKET_OK && mp_rep->num_fds == 1) + device_fd = mp_rep->fds[0]; + } + + free(mp_reply.msgs); + + if (device_fd < 0) { + EAL_LOG(ERR, "Cannot request device fd for vfio%d", dev->dev_num); + return -1; + } + dev->fd = device_fd; + + return 0; +} + +int +vfio_cdev_setup_device(struct container *cfg, struct vfio_device *dev) +{ + int device_fd; + + /* get device fd - primary or custom container opens it, secondary requests from primary */ + if (rte_eal_process_type() == RTE_PROC_PRIMARY || !vfio_container_is_default(cfg)) { + device_fd = cdev_open_device_fd(dev->dev_num); + if (device_fd < 0) + return -1; + dev->fd = device_fd; + + /* attach device to iommufd - only in primary */ + if (cdev_attach_device_to_iommufd(cfg, dev) < 0) + return -1; + } else if (vfio_cdev_request_dev_fd(dev) < 0) { + return -1; + } + return 0; +} diff --git a/lib/eal/linux/eal_vfio_mp_sync.c b/lib/eal/linux/eal_vfio_mp_sync.c index 9a07d35023..6d94f44af8 100644 --- a/lib/eal/linux/eal_vfio_mp_sync.c +++ b/lib/eal/linux/eal_vfio_mp_sync.c @@ -93,6 +93,48 @@ vfio_mp_primary(const struct rte_mp_msg *msg, const void *peer) } break; } + case SOCKET_REQ_CDEV: + { + struct container *cfg; + struct vfio_device *dev; + + if (vfio_cfg.mode != RTE_VFIO_MODE_CDEV) { + EAL_LOG(ERR, "VFIO not initialized in cdev mode"); + r->result = SOCKET_ERR; + break; + } + + r->req = SOCKET_REQ_CDEV; + r->cdev_dev_num = m->cdev_dev_num; + + cfg = vfio_cfg.default_cfg; + dev = vfio_cdev_get_dev_by_num(cfg, m->cdev_dev_num); + if (dev == NULL) { + r->result = SOCKET_NO_FD; + } else { + r->result = SOCKET_OK; + reply.num_fds = 1; + reply.fds[0] = dev->fd; + } + break; + } + case SOCKET_REQ_IOAS_ID: + { + struct container *cfg; + + if (vfio_cfg.mode != RTE_VFIO_MODE_CDEV) { + EAL_LOG(ERR, "VFIO not initialized in cdev mode"); + r->result = SOCKET_ERR; + break; + } + + r->req = SOCKET_REQ_IOAS_ID; + cfg = vfio_cfg.default_cfg; + r->ioas_id = cfg->cdev_cfg.ioas_id; + + r->result = SOCKET_OK; + break; + } default: EAL_LOG(ERR, "vfio received invalid message!"); return -1; diff --git a/lib/eal/linux/meson.build b/lib/eal/linux/meson.build index 75a9afdd03..adf043d24c 100644 --- a/lib/eal/linux/meson.build +++ b/lib/eal/linux/meson.build @@ -16,6 +16,7 @@ sources += files( 'eal_thread.c', 'eal_timer.c', 'eal_vfio.c', + 'eal_vfio_cdev.c', 'eal_vfio_group.c', 'eal_vfio_mp_sync.c', ) -- 2.47.3 ^ permalink raw reply related [flat|nested] 42+ messages in thread
* Re: [PATCH v7 00/18] Support VFIO cdev API in DPDK 2026-02-26 14:17 ` [PATCH v7 00/18] Support VFIO cdev API in DPDK Anatoly Burakov ` (17 preceding siblings ...) 2026-02-26 14:17 ` [PATCH v7 18/18] vfio: introduce cdev mode Anatoly Burakov @ 2026-05-01 22:45 ` Stephen Hemminger 18 siblings, 0 replies; 42+ messages in thread From: Stephen Hemminger @ 2026-05-01 22:45 UTC (permalink / raw) To: Anatoly Burakov; +Cc: dev On Thu, 26 Feb 2026 14:17:23 +0000 Anatoly Burakov <anatoly.burakov@intel.com> wrote: > This patchset introduces a major refactor of the VFIO subsystem in DPDK to > support character device (cdev) interface introduced in Linux kernel, as well as > make the API more streamlined and useful. The goal is to simplify device > management, improve compatibility, and clarify API responsibilities. > > The following sections outline the key issues addressed by this patchset and the > corresponding changes introduced. The patch set is still interesting but lots of build failures. You need to rebase and resubmit. Look for shadow variable warnings etc. ccache clang -Ilib/librte_eal.a.p -Ilib -I../lib -Ilib/eal/common -I../lib/eal/common -I. -I.. -Iconfig -I../config -Ilib/eal/include -I../lib/eal/include -Ilib/eal/linux/include -I../lib/eal/linux/include -Ilib/eal/x86/include -I../lib/eal/x86/include -I../kernel/linux -Ilib/eal -I../lib/eal -Ilib/kvargs -I../lib/kvargs -Ilib/log -I../lib/log -Ilib/metrics -I../lib/metrics -Ilib/telemetry -I../lib/telemetry -Ilib/argparse -I../lib/argparse -Xclang -fcolor-diagnostics -pipe -D_FILE_OFFSET_BITS=64 -Wall -Winvalid-pch -Wextra -Werror -std=c11 -O2 -g -include rte_config.h -Wvla -Wcast-qual -Wcomma -Wdeprecated -Wformat -Wformat-nonliteral -Wformat-security -Wmissing-declarations -Wmissing-prototypes -Wnested-externs -Wold-style-definition -Wpointer-arith -Wshadow -Wsign-compare -Wstrict-prototypes -Wundef -Wwrite-strings -Wno-missing-field-initializers -D_GNU_SOURCE -fPIC -march=corei7 -mrtm -DALLOW_EXPERIMENTAL_API -DALLOW_INTERNAL_API '-DABI_VERSION="26.1"' -DRTE_EAL_PTHREAD_ATTR_SETAFFINITY_NP -DRTE_LOG_DEFAULT_LOGTYPE=lib.eal -DRTE_ANNOTATE_LOCKS -Wthread-safety -MD -MQ lib/librte_eal.a.p/eal_linux_eal_vfio_group.c.o -MF lib/librte_eal.a.p/eal_linux_eal_vfio_group.c.o.d -o lib/librte_eal.a.p/eal_linux_eal_vfio_group.c.o -c ../lib/eal/linux/eal_vfio_group.c ../lib/eal/linux/eal_vfio_group.c:914:8: error: declaration shadows a local variable [-Werror,-Wshadow] 914 | char dev[PATH_MAX]; | ^ ../lib/eal/linux/eal_vfio_group.c:905:94: note: previous declaration is here 905 | vfio_group_setup_device_fd(const char *dev_addr, struct vfio_group *grp, struct vfio_device *dev) | ^ 1 error generated. [97/7766] Generating ^ permalink raw reply [flat|nested] 42+ messages in thread
end of thread, other threads:[~2026-05-01 22:45 UTC | newest]
Thread overview: 42+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
[not found] <cover.1763141462.git.anatoly.burakov@intel.com>
2025-11-18 16:29 ` [PATCH v3 00/20] Support VFIO cdev API in DPDK Anatoly Burakov
2025-11-18 16:29 ` [PATCH v3 01/20] doc: add deprecation notice for VFIO API Anatoly Burakov
2025-11-18 16:29 ` [PATCH v3 02/20] doc: add deprecation notice for vDPA driver API Anatoly Burakov
2025-11-18 16:29 ` [PATCH v3 03/20] uapi: update to v6.17 and add iommufd.h Anatoly Burakov
2025-11-18 17:36 ` Stephen Hemminger
2025-11-18 16:29 ` [PATCH v3 04/20] vfio: make all functions internal Anatoly Burakov
2025-11-18 16:29 ` [PATCH v3 05/20] vfio: split get device info from setup Anatoly Burakov
2025-11-18 16:29 ` [PATCH v3 06/20] vfio: add container device assignment API Anatoly Burakov
2025-11-18 16:29 ` [PATCH v3 07/20] net/nbl: do not use VFIO group bind API Anatoly Burakov
2025-11-18 16:29 ` [PATCH v3 08/20] net/ntnic: use container device assignment API Anatoly Burakov
2025-11-18 16:29 ` [PATCH v3 09/20] vdpa/ifc: " Anatoly Burakov
2025-11-18 16:29 ` [PATCH v3 10/20] vdpa/nfp: " Anatoly Burakov
2025-11-18 16:29 ` [PATCH v3 11/20] vdpa/sfc: " Anatoly Burakov
2025-11-18 16:29 ` [PATCH v3 12/20] vhost: remove group-related API from drivers Anatoly Burakov
2025-11-18 16:29 ` [PATCH v3 13/20] vfio: remove group-based API Anatoly Burakov
2025-11-18 16:29 ` [PATCH v3 14/20] vfio: cleanup and refactor Anatoly Burakov
2025-11-18 16:29 ` [PATCH v3 15/20] bus/pci: use the new VFIO mode API Anatoly Burakov
2025-11-18 16:29 ` [PATCH v3 16/20] bus/fslmc: " Anatoly Burakov
2025-11-18 16:29 ` [PATCH v3 17/20] net/hinic3: " Anatoly Burakov
2025-11-18 16:29 ` [PATCH v3 18/20] net/ntnic: " Anatoly Burakov
2025-11-18 16:29 ` [PATCH v3 19/20] vfio: remove no-IOMMU check API Anatoly Burakov
2025-11-18 16:29 ` [PATCH v3 20/20] vfio: introduce cdev mode Anatoly Burakov
2026-02-26 14:17 ` [PATCH v7 00/18] Support VFIO cdev API in DPDK Anatoly Burakov
2026-02-26 14:17 ` [PATCH v7 01/18] uapi: update to v6.17 and add iommufd.h Anatoly Burakov
2026-02-26 14:17 ` [PATCH v7 02/18] vfio: make all functions internal Anatoly Burakov
2026-02-26 14:17 ` [PATCH v7 03/18] vfio: split get device info from setup Anatoly Burakov
2026-02-26 14:17 ` [PATCH v7 04/18] vfio: add container device assignment API Anatoly Burakov
2026-02-26 14:17 ` [PATCH v7 05/18] net/nbl: do not use VFIO group bind API Anatoly Burakov
2026-02-26 14:17 ` [PATCH v7 06/18] net/ntnic: use container device assignment API Anatoly Burakov
2026-02-26 14:17 ` [PATCH v7 07/18] vdpa/ifc: " Anatoly Burakov
2026-02-26 14:17 ` [PATCH v7 08/18] vdpa/nfp: " Anatoly Burakov
2026-02-26 14:17 ` [PATCH v7 09/18] vdpa/sfc: " Anatoly Burakov
2026-02-26 14:17 ` [PATCH v7 10/18] vhost: remove group-related API from drivers Anatoly Burakov
2026-02-26 14:17 ` [PATCH v7 11/18] vfio: remove group-based API Anatoly Burakov
2026-02-26 14:17 ` [PATCH v7 12/18] vfio: cleanup and refactor Anatoly Burakov
2026-02-26 14:17 ` [PATCH v7 13/18] bus/pci: use the new VFIO mode API Anatoly Burakov
2026-02-26 14:17 ` [PATCH v7 14/18] bus/fslmc: " Anatoly Burakov
2026-02-26 14:17 ` [PATCH v7 15/18] net/hinic3: " Anatoly Burakov
2026-02-26 14:17 ` [PATCH v7 16/18] net/ntnic: " Anatoly Burakov
2026-02-26 14:17 ` [PATCH v7 17/18] vfio: remove no-IOMMU check API Anatoly Burakov
2026-02-26 14:17 ` [PATCH v7 18/18] vfio: introduce cdev mode Anatoly Burakov
2026-05-01 22:45 ` [PATCH v7 00/18] Support VFIO cdev API in DPDK Stephen Hemminger
This is an external index of several public inboxes, see mirroring instructions on how to clone and mirror all data and code used by this external index.