Re: [PATCH v7 00/13] blkio: add libblkio BlockDriver

[Stefan Hajnoczi <stefanha@redhat.com>]

[-- Attachment #1: Type: text/plain, Size: 8867 bytes --]

On Thu, Oct 13, 2022 at 02:58:55PM -0400, Stefan Hajnoczi wrote:
> v7:
> - Add nvme-io_uring and virtio-blk-vhost-user syntax examples to commit description [Markus]
> - Add missing nvme-io_uring QAPI [Markus, Alberto]
> - Rename mem-regions-pinned to may-pin-mem-regions [Alberto]
> - Fix value/bs->bl.max_iov mix-up [Stefano]
> v6:
> - Add untested nvme-io_uring driver. Please test in your nested NVMe environment, Alberto. [Alberto]
> - Map blkio mem regions only when necessary to reduce conflicts with RAM discard [Alberto]
> - Reduce duplication by having a single blkio_virtio_blk_common_open() function [Alberto]
> - Avoid duplication in BlockDriver definitions using a macro [Alberto]
> - Avoid ram block registrar segfault [Stefano]
> - Use QLIST_FOREACH_SAFE() in ram block notifier code so callbacks can remove themselves
> v5:
> - Drop "RFC" since libblkio 1.0 has been released and the library API is stable
> - Disable BDRV_REQ_REGISTERED_BUF if we run out of blkio_mem_regions. The
>   bounce buffer slow path is taken when there are not enough blkio_mem_regions
>   to cover guest RAM. [Hanna & David Hildenbrand]
> - Call ram_block_discard_disable() when mem-region-pinned property is true or
>   absent [David Hildenbrand]
> - Use a bounce buffer pool instead of allocating/freeing a buffer for each
>   request. This reduces the number of blkio_mem_regions required for bounce
>   buffers to 1 and avoids frequent blkio_mem_region_map/unmap() calls.
> - Switch to .bdrv_co_*() instead of .bdrv_aio_*(). Needed for the bounce buffer
>   pool's CoQueue.
> v4:
> - Patch 1:
>   - Add virtio-blk-vhost-user driver [Kevin]
>   - Drop .bdrv_parse_filename() and .bdrv_needs_filename for virtio-blk-vhost-vdpa [Stefano]
>   - Add copyright and license header [Hanna]
>   - Drop .bdrv_parse_filename() in favor of --blockdev or json: [Hanna]
>   - Clarify that "filename" is always non-NULL for io_uring [Hanna]
>   - Check that virtio-blk-vhost-vdpa "path" option is non-NULL [Hanna]
>   - Fix virtio-blk-vhost-vdpa cache.direct=off logic [Hanna]
>   - Use macros for driver names [Hanna]
>   - Assert that the driver name is valid [Hanna]
>   - Update "readonly" property name to "read-only" [Hanna]
>   - Call blkio_detach_aio_context() in blkio_close() [Hanna]
>   - Avoid uint32_t * to int * casts in blkio_refresh_limits() [Hanna]
>   - Remove write zeroes and discard from the todo list [Hanna]
>   - Use PRIu32 instead of %d for uint32_t [Hanna]
>   - Fix error messages with buf-alignment instead of optimal-io-size [Hanna]
>   - Call map/unmap APIs since libblkio alloc/free APIs no longer do that
>   - Update QAPI schema QEMU version to 7.2
> - Patch 5:
>   - Expand the BDRV_REQ_REGISTERED_BUF flag passthrough and drop assert(!flags)
>     in drivers [Hanna]
> - Patch 7:
>   - Fix BLK->BDRV typo [Hanna]
>   - Make BlockRAMRegistrar handle failure [Hanna]
> - Patch 8:
>   - Replace memory_region_get_fd() approach with qemu_ram_get_fd()
> - Patch 10:
>   - Use (void)ret; to discard unused return value [Hanna]
>   - libblkio's blkio_unmap_mem_region() API no longer has a return value
>   - Check for registered bufs that cross RAMBlocks [Hanna]
> - Patch 11:
>   - Handle bdrv_register_buf() errors [Hanna]
> v3:
> - Add virtio-blk-vhost-vdpa for vdpa-blk devices including VDUSE
> - Add discard and write zeroes support
> - Rebase and adopt latest libblkio APIs
> v2:
> - Add BDRV_REQ_REGISTERED_BUF to bs.supported_write_flags [Stefano]
> - Use new blkioq_get_num_completions() API
> - Implement .bdrv_refresh_limits()
> 
> This patch series adds a QEMU BlockDriver for libblkio
> (https://gitlab.com/libblkio/libblkio/), a library for high-performance block
> device I/O. This work was presented at KVM Forum 2022 and slides are available
> here:
> https://static.sched.com/hosted_files/kvmforum2022/8c/libblkio-kvm-forum-2022.pdf
> 
> The second patch adds the core BlockDriver and most of the libblkio API usage.
> Three libblkio drivers are included:
> - io_uring
> - virtio-blk-vhost-user
> - virtio-blk-vhost-vdpa
> 
> The remainder of the patch series reworks the existing QEMU bdrv_register_buf()
> API so virtio-blk emulation efficiently map guest RAM for libblkio - some
> libblkio drivers require that I/O buffer memory is pre-registered (think VFIO,
> vhost, etc).
> 
> Vladimir requested performance results that show the effect of the
> BDRV_REQ_REGISTERED_BUF flag. I ran the patches against qemu-storage-daemon's
> vhost-user-blk export with iodepth=1 bs=512 to see the per-request overhead due
> to bounce buffer allocation/mapping:
> 
> Name                                   IOPS   Error
> bounce-buf                          4373.81 ± 0.01%
> registered-buf                     13062.80 ± 0.67%
> 
> The BDRV_REQ_REGISTERED_BUF optimization version is about 3x faster.
> 
> See the BlockDriver struct in block/blkio.c for a list of APIs that still need
> to be implemented. The core functionality is covered.
> 
> Regarding the design: each libblkio driver is a separately named BlockDriver.
> That means there is an "io_uring" BlockDriver and not a generic "libblkio"
> BlockDriver. This way QAPI and open parameters are type-safe and mandatory
> parameters can be checked by QEMU.
> 
> Stefan Hajnoczi (13):
>   coroutine: add flag to re-queue at front of CoQueue
>   blkio: add libblkio block driver
>   numa: call ->ram_block_removed() in ram_block_notifer_remove()
>   block: pass size to bdrv_unregister_buf()
>   block: use BdrvRequestFlags type for supported flag fields
>   block: add BDRV_REQ_REGISTERED_BUF request flag
>   block: return errors from bdrv_register_buf()
>   numa: use QLIST_FOREACH_SAFE() for RAM block notifiers
>   block: add BlockRAMRegistrar
>   exec/cpu-common: add qemu_ram_get_fd()
>   stubs: add qemu_ram_block_from_host() and qemu_ram_get_fd()
>   blkio: implement BDRV_REQ_REGISTERED_BUF optimization
>   virtio-blk: use BDRV_REQ_REGISTERED_BUF optimization hint
> 
>  MAINTAINERS                                 |    7 +
>  meson_options.txt                           |    2 +
>  qapi/block-core.json                        |   77 +-
>  meson.build                                 |    9 +
>  include/block/block-common.h                |    9 +
>  include/block/block-global-state.h          |   10 +-
>  include/block/block_int-common.h            |   15 +-
>  include/exec/cpu-common.h                   |    1 +
>  include/hw/virtio/virtio-blk.h              |    2 +
>  include/qemu/coroutine.h                    |   15 +-
>  include/sysemu/block-backend-global-state.h |    4 +-
>  include/sysemu/block-ram-registrar.h        |   37 +
>  block.c                                     |   14 +
>  block/blkio.c                               | 1008 +++++++++++++++++++
>  block/blkverify.c                           |    4 +-
>  block/block-backend.c                       |    8 +-
>  block/block-ram-registrar.c                 |   58 ++
>  block/crypto.c                              |    4 +-
>  block/file-posix.c                          |    1 -
>  block/gluster.c                             |    1 -
>  block/io.c                                  |  101 +-
>  block/mirror.c                              |    2 +
>  block/nbd.c                                 |    1 -
>  block/nvme.c                                |   20 +-
>  block/parallels.c                           |    1 -
>  block/qcow.c                                |    2 -
>  block/qed.c                                 |    1 -
>  block/raw-format.c                          |    2 +
>  block/replication.c                         |    1 -
>  block/ssh.c                                 |    1 -
>  block/vhdx.c                                |    1 -
>  hw/block/virtio-blk.c                       |   39 +-
>  hw/core/numa.c                              |   26 +-
>  qemu-img.c                                  |    6 +-
>  softmmu/physmem.c                           |    5 +
>  stubs/physmem.c                             |   13 +
>  tests/qtest/modules-test.c                  |    3 +
>  util/qemu-coroutine-lock.c                  |    9 +-
>  util/vfio-helpers.c                         |    5 +-
>  block/meson.build                           |    2 +
>  scripts/meson-buildoptions.sh               |    3 +
>  stubs/meson.build                           |    1 +
>  42 files changed, 1435 insertions(+), 96 deletions(-)
>  create mode 100644 include/sysemu/block-ram-registrar.h
>  create mode 100644 block/blkio.c
>  create mode 100644 block/block-ram-registrar.c
>  create mode 100644 stubs/physmem.c
> 
> -- 
> 2.37.3
> 

Thanks, applied to my block tree:
https://gitlab.com/stefanha/qemu/commits/block

Stefan

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]