Re: [Intel-gfx] [PATCH v7 00/21] Move all drivers to a common dma-buf locking convention

[Dmitry Osipenko <dmitry.osipenko@collabora.com>]

On 10/18/22 14:34, Christian König wrote:
> Am 18.10.22 um 01:07 schrieb Dmitry Osipenko:
>> On 10/17/22 20:22, Dmitry Osipenko wrote:
>>> Hello,
>>>
>>> This series moves all drivers to a dynamic dma-buf locking
>>> specification.
>>>  From now on all dma-buf importers are made responsible for holding
>>> dma-buf's reservation lock around all operations performed over dma-bufs
>>> in accordance to the locking specification. This allows us to utilize
>>> reservation lock more broadly around kernel without fearing of a
>>> potential
>>> deadlocks.
>>>
>>> This patchset passes all i915 selftests. It was also tested using
>>> VirtIO,
>>> Panfrost, Lima, Tegra, udmabuf, AMDGPU and Nouveau drivers. I tested
>>> cases
>>> of display+GPU, display+V4L and GPU+V4L dma-buf sharing (where
>>> appropriate),
>>> which covers majority of kernel drivers since rest of the drivers share
>>> same or similar code paths.
>>>
>>> Changelog:
>>>
>>> v7: - Rebased on top of recent drm-misc-next.
>>>
>>>      - Added ack from Jason Gunthorpe to the RDMA patch.
>>>
>>>      - Added iosys_map_clear() to dma_buf_vmap_unlocked(), making it
>>> fully
>>>        consistent with dma_buf_vmap().
>>>
>>> v6: - Added r-b from Michael Ruhl to the i915 patch.
>>>
>>>      - Added acks from Sumit Semwal and updated commit message of the
>>>        "Move dma_buf_vmap() to dynamic locking specification" patch like
>>>        was suggested by Sumit.
>>>
>>>      - Added "!dmabuf" check to dma_buf_vmap_unlocked() to match the
>>> locked
>>>        variant of the function, for consistency.
>>>
>>> v5: - Added acks and r-bs that were given to v4.
>>>
>>>      - Changed i915 preparation patch like was suggested by Michael
>>> Ruhl.
>>>        The scope of reservation locking is smaller now.
>>>
>>> v4: - Added dma_buf_mmap() to the "locking convention" documentation,
>>>        which was missed by accident in v3.
>>>
>>>      - Added acks from Christian König, Tomasz Figa and Hans Verkuil
>>> that
>>>        they gave to couple v3 patches.
>>>
>>>      - Dropped the "_unlocked" postfix from function names that don't
>>> have
>>>        the locked variant, as was requested by Christian König.
>>>
>>>      - Factored out the per-driver preparations into separate patches
>>>        to ease reviewing of the changes, which is now doable without the
>>>        global dma-buf functions renaming.
>>>
>>>      - Factored out the dynamic locking convention enforcements into
>>> separate
>>>        patches which add the final dma_resv_assert_held(dmabuf->resv)
>>> to the
>>>        dma-buf API functions.
>>>
>>> v3: - Factored out dma_buf_mmap_unlocked() and attachment functions
>>>        into aseparate patches, like was suggested by Christian König.
>>>
>>>      - Corrected and factored out dma-buf locking documentation into
>>>        a separate patch, like was suggested by Christian König.
>>>
>>>      - Intel driver dropped the reservation locking fews days ago from
>>>        its BO-release code path, but we need that locking for the
>>> imported
>>>        GEMs because in the end that code path unmaps the imported GEM.
>>>        So I added back the locking needed by the imported GEMs, updating
>>>        the "dma-buf attachment locking specification" patch
>>> appropriately.
>>>
>>>      - Tested Nouveau+Intel dma-buf import/export combo.
>>>
>>>      - Tested udmabuf import to i915/Nouveau/AMDGPU.
>>>
>>>      - Fixed few places in Etnaviv, Panfrost and Lima drivers that I
>>> missed
>>>        to switch to locked dma-buf vmapping in the drm/gem: Take
>>> reservation
>>>        lock for vmap/vunmap operations" patch. In a result
>>> invalidated the
>>>        Christian's r-b that he gave to v2.
>>>
>>>      - Added locked dma-buf vmap/vunmap functions that are needed for
>>> fixing
>>>        vmappping of Etnaviv, Panfrost and Lima drivers mentioned above.
>>>        I actually had this change stashed for the drm-shmem shrinker
>>> patchset,
>>>        but then realized that it's already needed by the dma-buf
>>> patches.
>>>        Also improved my tests to better cover these code paths.
>>>
>>> v2: - Changed locking specification to avoid problems with a
>>> cross-driver
>>>        ww locking, like was suggested by Christian König. Now the
>>> attach/detach
>>>        callbacks are invoked without the held lock and exporter
>>> should take the
>>>        lock.
>>>
>>>      - Added "locking convention" documentation that explains which
>>> dma-buf
>>>        functions and callbacks are locked/unlocked for importers and
>>> exporters,
>>>        which was requested by Christian König.
>>>
>>>      - Added ack from Tomasz Figa to the V4L patches that he gave to v1.
>>>
>>> Dmitry Osipenko (21):
>>>    dma-buf: Add unlocked variant of vmapping functions
>>>    dma-buf: Add unlocked variant of attachment-mapping functions
>>>    drm/gem: Take reservation lock for vmap/vunmap operations
>>>    drm/prime: Prepare to dynamic dma-buf locking specification
>>>    drm/armada: Prepare to dynamic dma-buf locking specification
>>>    drm/i915: Prepare to dynamic dma-buf locking specification
>>>    drm/omapdrm: Prepare to dynamic dma-buf locking specification
>>>    drm/tegra: Prepare to dynamic dma-buf locking specification
>>>    drm/etnaviv: Prepare to dynamic dma-buf locking specification
>>>    RDMA/umem: Prepare to dynamic dma-buf locking specification
>>>    misc: fastrpc: Prepare to dynamic dma-buf locking specification
>>>    xen/gntdev: Prepare to dynamic dma-buf locking specification
>>>    media: videobuf2: Prepare to dynamic dma-buf locking specification
>>>    media: tegra-vde: Prepare to dynamic dma-buf locking specification
>>>    dma-buf: Move dma_buf_vmap() to dynamic locking specification
>>>    dma-buf: Move dma_buf_attach() to dynamic locking specification
>>>    dma-buf: Move dma_buf_map_attachment() to dynamic locking
>>>      specification
>>>    dma-buf: Move dma_buf_mmap() to dynamic locking specification
>>>    dma-buf: Document dynamic locking convention
>>>    media: videobuf2: Stop using internal dma-buf lock
>>>    dma-buf: Remove obsoleted internal lock
>>>
>>>   Documentation/driver-api/dma-buf.rst          |   6 +
>>>   drivers/dma-buf/dma-buf.c                     | 216 +++++++++++++++---
>>>   drivers/gpu/drm/armada/armada_gem.c           |   8 +-
>>>   drivers/gpu/drm/drm_client.c                  |   4 +-
>>>   drivers/gpu/drm/drm_gem.c                     |  24 ++
>>>   drivers/gpu/drm/drm_gem_dma_helper.c          |   6 +-
>>>   drivers/gpu/drm/drm_gem_framebuffer_helper.c  |   6 +-
>>>   drivers/gpu/drm/drm_gem_ttm_helper.c          |   9 +-
>>>   drivers/gpu/drm/drm_prime.c                   |   6 +-
>>>   drivers/gpu/drm/etnaviv/etnaviv_gem_prime.c   |   2 +-
>>>   drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c    |   2 +-
>>>   drivers/gpu/drm/i915/gem/i915_gem_object.c    |  14 ++
>>>   .../drm/i915/gem/selftests/i915_gem_dmabuf.c  |  16 +-
>>>   drivers/gpu/drm/lima/lima_sched.c             |   4 +-
>>>   drivers/gpu/drm/omapdrm/omap_gem_dmabuf.c     |   4 +-
>>>   drivers/gpu/drm/panfrost/panfrost_dump.c      |   4 +-
>>>   drivers/gpu/drm/panfrost/panfrost_perfcnt.c   |   6 +-
>>>   drivers/gpu/drm/qxl/qxl_object.c              |  17 +-
>>>   drivers/gpu/drm/qxl/qxl_prime.c               |   4 +-
>>>   drivers/gpu/drm/tegra/gem.c                   |  17 +-
>>>   drivers/infiniband/core/umem_dmabuf.c         |   7 +-
>>>   .../common/videobuf2/videobuf2-dma-contig.c   |  22 +-
>>>   .../media/common/videobuf2/videobuf2-dma-sg.c |  19 +-
>>>   .../common/videobuf2/videobuf2-vmalloc.c      |  17 +-
>>>   .../platform/nvidia/tegra-vde/dmabuf-cache.c  |   6 +-
>>>   drivers/misc/fastrpc.c                        |   6 +-
>>>   drivers/xen/gntdev-dmabuf.c                   |   8 +-
>>>   include/drm/drm_gem.h                         |   3 +
>>>   include/linux/dma-buf.h                       |  17 +-
>>>   29 files changed, 325 insertions(+), 155 deletions(-)
>>>
>> Applied to drm-misc-next
> 
> Finally! Fingers crossed that all corner cases where fixed during the
> review.
> 
> But if anything shows up feel free to loop me in to help fixing things.
> 
> Thanks a lot for doing this,
> Christian.

I'll ping you if anything will show up. Thank you!

-- 
Best regards,
Dmitry