Re: [Qemu-devel] [PATCH qemu v15 11/17] vfio: spapr: Add SPAPR IOMMU v2 support (DMA memory preregistering)

[Alexey Kardashevskiy <aik@ozlabs.ru>]

On 04/06/2016 04:05 PM, David Gibson wrote:
> On Mon, Apr 04, 2016 at 07:33:40PM +1000, Alexey Kardashevskiy wrote:
>> This makes use of the new "memory registering" feature. The idea is
>> to provide the userspace ability to notify the host kernel about pages
>> which are going to be used for DMA. Having this information, the host
>> kernel can pin them all once per user process, do locked pages
>> accounting (once) and not spent time on doing that in real time with
>> possible failures which cannot be handled nicely in some cases.
>>
>> This adds a prereg memory listener which listens on address_space_memory
>> and notifies a VFIO container about memory which needs to be
>> pinned/unpinned. VFIO MMIO regions (i.e. "skip dump" regions) are skipped.
>>
>> As there is no per-IOMMU-type release() callback anymore, this stores
>> the IOMMU type in the container so vfio_listener_release() can device
>
> s/device/determine/ ?

That was supposed to be "decide" but "determine" will also do :)


>
>> if it needs to unregister @prereg_listener.
>>
>> The feature is only enabled for SPAPR IOMMU v2. The host kernel changes
>> are required. Since v2 does not need/support VFIO_IOMMU_ENABLE, this does
>> not call it when v2 is detected and enabled.
>>
>> This does not change the guest visible interface.
>>
>> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
>> ---
>> Changes:
>> v15:
>> * banned unaligned sections
>> * added an vfio_prereg_gpa_to_ua() helper
>>
>> v14:
>> * s/free_container_exit/listener_release_exit/g
>> * added "if memory_region_is_iommu()" to vfio_prereg_listener_skipped_section
>> ---
>>   hw/vfio/Makefile.objs         |   1 +
>>   hw/vfio/common.c              |  38 +++++++++---
>>   hw/vfio/prereg.c              | 138 ++++++++++++++++++++++++++++++++++++++++++
>>   include/hw/vfio/vfio-common.h |   4 ++
>>   trace-events                  |   2 +
>>   5 files changed, 173 insertions(+), 10 deletions(-)
>>   create mode 100644 hw/vfio/prereg.c
>>
>> diff --git a/hw/vfio/Makefile.objs b/hw/vfio/Makefile.objs
>> index ceddbb8..5800e0e 100644
>> --- a/hw/vfio/Makefile.objs
>> +++ b/hw/vfio/Makefile.objs
>> @@ -4,4 +4,5 @@ obj-$(CONFIG_PCI) += pci.o pci-quirks.o
>>   obj-$(CONFIG_SOFTMMU) += platform.o
>>   obj-$(CONFIG_SOFTMMU) += calxeda-xgmac.o
>>   obj-$(CONFIG_SOFTMMU) += amd-xgbe.o
>> +obj-$(CONFIG_SOFTMMU) += prereg.o
>>   endif
>> diff --git a/hw/vfio/common.c b/hw/vfio/common.c
>> index 6bec419..3e9c579 100644
>> --- a/hw/vfio/common.c
>> +++ b/hw/vfio/common.c
>> @@ -493,6 +493,9 @@ static const MemoryListener vfio_memory_listener = {
>>   static void vfio_listener_release(VFIOContainer *container)
>>   {
>>       memory_listener_unregister(&container->listener);
>> +    if (container->iommu_type == VFIO_SPAPR_TCE_v2_IOMMU) {
>> +        memory_listener_unregister(&container->prereg_listener);
>> +    }
>>   }
>>
>>   int vfio_region_setup(Object *obj, VFIODevice *vbasedev, VFIORegion *region,
>> @@ -800,8 +803,8 @@ static int vfio_connect_container(VFIOGroup *group, AddressSpace *as)
>>               goto free_container_exit;
>>           }
>>
>> -        ret = ioctl(fd, VFIO_SET_IOMMU,
>> -                    v2 ? VFIO_TYPE1v2_IOMMU : VFIO_TYPE1_IOMMU);
>> +        container->iommu_type = v2 ? VFIO_TYPE1v2_IOMMU : VFIO_TYPE1_IOMMU;
>> +        ret = ioctl(fd, VFIO_SET_IOMMU, container->iommu_type);
>>           if (ret) {
>>               error_report("vfio: failed to set iommu for container: %m");
>>               ret = -errno;
>> @@ -826,8 +829,10 @@ static int vfio_connect_container(VFIOGroup *group, AddressSpace *as)
>>           if ((ret == 0) && (info.flags & VFIO_IOMMU_INFO_PGSIZES)) {
>>               container->iova_pgsizes = info.iova_pgsizes;
>>           }
>> -    } else if (ioctl(fd, VFIO_CHECK_EXTENSION, VFIO_SPAPR_TCE_IOMMU)) {
>> +    } else if (ioctl(fd, VFIO_CHECK_EXTENSION, VFIO_SPAPR_TCE_IOMMU) ||
>> +               ioctl(fd, VFIO_CHECK_EXTENSION, VFIO_SPAPR_TCE_v2_IOMMU)) {
>>           struct vfio_iommu_spapr_tce_info info;
>> +        bool v2 = !!ioctl(fd, VFIO_CHECK_EXTENSION, VFIO_SPAPR_TCE_v2_IOMMU);
>>
>>           ret = ioctl(group->fd, VFIO_GROUP_SET_CONTAINER, &fd);
>>           if (ret) {
>> @@ -835,7 +840,9 @@ static int vfio_connect_container(VFIOGroup *group, AddressSpace *as)
>>               ret = -errno;
>>               goto free_container_exit;
>>           }
>> -        ret = ioctl(fd, VFIO_SET_IOMMU, VFIO_SPAPR_TCE_IOMMU);
>> +        container->iommu_type =
>> +            v2 ? VFIO_SPAPR_TCE_v2_IOMMU : VFIO_SPAPR_TCE_IOMMU;
>> +        ret = ioctl(fd, VFIO_SET_IOMMU, container->iommu_type);
>>           if (ret) {
>>               error_report("vfio: failed to set iommu for container: %m");
>>               ret = -errno;
>> @@ -847,11 +854,22 @@ static int vfio_connect_container(VFIOGroup *group, AddressSpace *as)
>>            * when container fd is closed so we do not call it explicitly
>>            * in this file.
>>            */
>> -        ret = ioctl(fd, VFIO_IOMMU_ENABLE);
>> -        if (ret) {
>> -            error_report("vfio: failed to enable container: %m");
>> -            ret = -errno;
>> -            goto free_container_exit;
>> +        if (!v2) {
>> +            ret = ioctl(fd, VFIO_IOMMU_ENABLE);
>> +            if (ret) {
>> +                error_report("vfio: failed to enable container: %m");
>> +                ret = -errno;
>> +                goto free_container_exit;
>> +            }
>> +        } else {
>> +            container->prereg_listener = vfio_prereg_listener;
>> +
>> +            memory_listener_register(&container->prereg_listener,
>> +                                     &address_space_memory);
>> +            if (container->error) {
>> +                error_report("vfio: RAM memory listener initialization failed for container");
>> +                goto listener_release_exit;
>> +            }
>>           }
>>
>>           /*
>> @@ -864,7 +882,7 @@ static int vfio_connect_container(VFIOGroup *group, AddressSpace *as)
>>           if (ret) {
>>               error_report("vfio: VFIO_IOMMU_SPAPR_TCE_GET_INFO failed: %m");
>>               ret = -errno;
>> -            goto free_container_exit;
>> +            goto listener_release_exit;
>>           }
>>           container->min_iova = info.dma32_window_start;
>>           container->max_iova = container->min_iova + info.dma32_window_size - 1;
>> diff --git a/hw/vfio/prereg.c b/hw/vfio/prereg.c
>> new file mode 100644
>> index 0000000..5f7fa30
>> --- /dev/null
>> +++ b/hw/vfio/prereg.c
>> @@ -0,0 +1,138 @@
>> +/*
>> + * DMA memory preregistration
>> + *
>> + * Authors:
>> + *  Alexey Kardashevskiy <aik@ozlabs.ru>
>> + *
>> + * This work is licensed under the terms of the GNU GPL, version 2.  See
>> + * the COPYING file in the top-level directory.
>> + */
>> +
>> +#include "qemu/osdep.h"
>> +#include <sys/ioctl.h>
>> +#include <linux/vfio.h>
>> +
>> +#include "hw/vfio/vfio-common.h"
>> +#include "qemu/error-report.h"
>> +#include "trace.h"
>> +
>> +static bool vfio_prereg_listener_skipped_section(MemoryRegionSection *section)
>> +{
>> +    if (memory_region_is_iommu(section->mr)) {
>> +        error_report("Cannot possibly preregister IOMMU memory");
>> +        return true;
>> +    }
>> +
>> +    return !memory_region_is_ram(section->mr) ||
>> +            memory_region_is_skip_dump(section->mr);
>> +}
>> +
>> +static void *vfio_prereg_gpa_to_ua(MemoryRegionSection *section, hwaddr gpa)
>> +{
>> +    return memory_region_get_ram_ptr(section->mr) +
>> +        section->offset_within_region +
>> +        (gpa - section->offset_within_address_space);
>> +}
>> +
>> +static void vfio_prereg_listener_region_add(MemoryListener *listener,
>> +                                            MemoryRegionSection *section)
>> +{
>> +    VFIOContainer *container = container_of(listener, VFIOContainer,
>> +                                            prereg_listener);
>> +    const hwaddr gpa = section->offset_within_address_space;
>> +    Int128 llend;
>> +    int ret;
>> +    hwaddr page_mask = qemu_real_host_page_mask;
>> +    struct vfio_iommu_spapr_register_memory reg = {
>> +        .argsz = sizeof(reg),
>> +        .flags = 0,
>> +    };
>> +
>> +    if (vfio_prereg_listener_skipped_section(section)) {
>> +        trace_vfio_listener_region_add_skip(
>> +                section->offset_within_address_space,
>> +                section->offset_within_address_space +
>> +                int128_get64(int128_sub(section->size, int128_one())));
>> +        return;
>> +    }
>> +
>> +    if (unlikely((section->offset_within_address_space & ~page_mask) ||
>> +                 (section->offset_within_region & ~page_mask) ||
>> +                 (int128_get64(section->size) & ~page_mask))) {
>> +        error_report("%s received unaligned region", __func__);
>> +        return;
>> +    }
>
> You could round out the range to pre-register rather than just failing
> here, but that can be changed later if we need it.


Do we really want to support non-host-page-aligned guest RAM size?

I'd prefer less automated decisions at this stage really, regardless.


>
>> +    llend = int128_make64(section->offset_within_address_space);
>> +    llend = int128_add(llend, section->size);
>> +
>> +    g_assert(!int128_ge(int128_make64(gpa), llend));
>> +
>> +    memory_region_ref(section->mr);
>> +
>> +    reg.vaddr = (__u64) vfio_prereg_gpa_to_ua(section, gpa);
>> +    reg.size = int128_get64(llend) - gpa;
>> +
>> +    ret = ioctl(container->fd, VFIO_IOMMU_SPAPR_REGISTER_MEMORY, &reg);
>> +    trace_vfio_ram_register(reg.vaddr, reg.size, ret ? -errno : 0);
>> +    if (ret) {
>> +        /*
>> +         * On the initfn path, store the first error in the container so we
>> +         * can gracefully fail.  Runtime, there's not much we can do other
>> +         * than throw a hardware error.
>> +         */
>> +        if (!container->initialized) {
>> +            if (!container->error) {
>> +                container->error = ret;
>> +            }
>> +        } else {
>> +            hw_error("vfio: Memory registering failed, unable to continue");
>> +        }
>> +    }
>> +}
>> +
>> +static void vfio_prereg_listener_region_del(MemoryListener *listener,
>> +                                            MemoryRegionSection *section)
>> +{
>> +    VFIOContainer *container = container_of(listener, VFIOContainer,
>> +                                            prereg_listener);
>> +    const hwaddr gpa = section->offset_within_address_space;
>> +    hwaddr end;
>> +    int ret;
>> +    hwaddr page_mask = qemu_real_host_page_mask;
>> +    struct vfio_iommu_spapr_register_memory reg = {
>> +        .argsz = sizeof(reg),
>> +        .flags = 0,
>> +    };
>> +
>> +    if (vfio_prereg_listener_skipped_section(section)) {
>> +        trace_vfio_listener_region_del_skip(
>> +                section->offset_within_address_space,
>> +                section->offset_within_address_space +
>> +                int128_get64(int128_sub(section->size, int128_one())));
>> +        return;
>> +    }
>> +
>> +    if (unlikely((section->offset_within_address_space & ~page_mask) ||
>> +                 (section->offset_within_region & ~page_mask) ||
>> +                 (int128_get64(section->size) & ~page_mask))) {
>> +        error_report("%s received unaligned region", __func__);
>> +        return;
>> +    }
>> +
>> +    end = section->offset_within_address_space + int128_get64(section->size);
>
> Why 64-bit math here, but 128 bit math in the region_add path?


Hm. Good question. It has been copied from the common VFIO code where it 
was since 7532d3cbf1, back those days I did not even think of VFIO devices 
hot(un)plugging so region_del() was never executed so this seems to be a 
bug there and here. Yet another patch is coming then.


>
>> +    if (gpa >= end) {
>> +        return;
>> +    }
>> +
>> +    reg.vaddr = (__u64) vfio_prereg_gpa_to_ua(section, gpa);
>> +    reg.size = end - gpa;
>> +
>> +    ret = ioctl(container->fd, VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY, &reg);
>> +    trace_vfio_ram_unregister(reg.vaddr, reg.size, ret ? -errno : 0);
>> +}
>> +
>> +const MemoryListener vfio_prereg_listener = {
>> +    .region_add = vfio_prereg_listener_region_add,
>> +    .region_del = vfio_prereg_listener_region_del,
>> +};
>> diff --git a/include/hw/vfio/vfio-common.h b/include/hw/vfio/vfio-common.h
>> index c9b6622..c72e45a 100644
>> --- a/include/hw/vfio/vfio-common.h
>> +++ b/include/hw/vfio/vfio-common.h
>> @@ -73,6 +73,8 @@ typedef struct VFIOContainer {
>>       VFIOAddressSpace *space;
>>       int fd; /* /dev/vfio/vfio, empowered by the attached groups */
>>       MemoryListener listener;
>> +    MemoryListener prereg_listener;
>> +    unsigned iommu_type;
>>       int error;
>>       bool initialized;
>>       /*
>> @@ -156,4 +158,6 @@ extern QLIST_HEAD(vfio_as_head, VFIOAddressSpace) vfio_address_spaces;
>>   int vfio_get_region_info(VFIODevice *vbasedev, int index,
>>                            struct vfio_region_info **info);
>>   #endif
>> +extern const MemoryListener vfio_prereg_listener;
>> +
>>   #endif /* !HW_VFIO_VFIO_COMMON_H */
>> diff --git a/trace-events b/trace-events
>> index 4335b9b..23ca0b9 100644
>> --- a/trace-events
>> +++ b/trace-events
>> @@ -1736,6 +1736,8 @@ vfio_region_mmap(const char *name, unsigned long offset, unsigned long end) "Reg
>>   vfio_region_exit(const char *name, int index) "Device %s, region %d"
>>   vfio_region_finalize(const char *name, int index) "Device %s, region %d"
>>   vfio_region_mmaps_set_enabled(const char *name, bool enabled) "Region %s mmaps enabled: %d"
>> +vfio_ram_register(uint64_t va, uint64_t size, int ret) "va=%"PRIx64" size=%"PRIx64" ret=%d"
>> +vfio_ram_unregister(uint64_t va, uint64_t size, int ret) "va=%"PRIx64" size=%"PRIx64" ret=%d"
>>
>>   # hw/vfio/platform.c
>>   vfio_platform_base_device_init(char *name, int groupid) "%s belongs to group #%d"
>


-- 
Alexey