Skip to content

Improvements for IB handling V8

Alex Deucher amd-gfx 1 version In review

Get series with b4
b4 am [email protected]

Versions newest first

  1. v1 Latest12 patches 11 repliesIn review

Patches in v1 No applies recorded yet

#Subject and filesSizeRepliesApplied
1/12drm/amdgpu: re-add the bad job to the pending list for ring resetsdrivers/gpu/drm/amd/amdgpu/amdgpu_job.c drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c+8 -51 reply
2/12drm/amdgpu/job: use GFP_ATOMIC while in gpu resetdrivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c +4 more+16 -120 replies
3/12drm/amdgpu: switch all IPs to using job for IBsdrivers/gpu/drm/amd/amdgpu/amdgpu_vpe.c drivers/gpu/drm/amd/amdgpu/cik_sdma.c drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +19 more+500 -4275 replies
4/12drm/amdgpu: require a job to schedule an IBdrivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c drivers/gpu/drm/amd/amdgpu/amdgpu_job.c +1 more+44 -770 replies
5/12drm/amdgpu: don't call drm_sched_stop/start() in asic resetdrivers/gpu/drm/amd/amdgpu/amdgpu_device.c drivers/gpu/drm/amd/amdgpu/amdgpu_job.c+4 -63 replies
6/12drm/amdgpu/cs: return -ETIME for guilty contextsdrivers/gpu/drm/amd/amdgpu/amdgpu_cs.c+1 -10 replies
7/12drm/amdgpu: plumb timedout fence through to force completiondrivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c drivers/gpu/drm/amd/amdgpu/amdgpu_device.c drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +6 more+49 -210 replies
8/12drm/amdgpu: simplify VCN reset helperdrivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c+13 -280 replies
9/12drm/amdgpu: Call drm_sched_increase_karma() for ring resetsdrivers/gpu/drm/amd/amdgpu/amdgpu_job.c+2 -00 replies
10/12drm/amdgpu: reorder IB schedule sequencedrivers/gpu/drm/amd/amdgpu/amdgpu_ib.c drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c+18 -111 reply
11/12drm/amdgpu: add a helper to calculate ring distancedrivers/gpu/drm/amd/amdgpu/amdgpu_ring.h+12 -61 reply
12/12drm/amdgpu: rework ring reset backup and reemit v8drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c +1 more+72 -1170 replies

Reviews trailers per patch

All 12 patches have a Reviewed-by or Acked-by.

PatchJesse ZhangChristian KönigPierre-Eric Pelloux-Prayer
1/12A-b
2/12R-b
3/12A-b
4/12R-b
5/12A-b
6/12R-b
7/12R-b
8/12R-b
9/12R-b
10/12A-b
11/12R-b
12/12R-b

Trailers given in replies are filled; those carried in the patches from earlier versions are outlined with dashes. Scroll the table sideways for all 3 reviewers.

Read the v1 thread

Cover letter [PATCH 00/12] Improvements for IB handling V8

This set contains a number of bug fixes and cleanups for IB handling that I worked on over the holidays. The first the three patches from V1 are already reviewed, so I didn't include them in V2 or V3. The first patches from V3 have been reviewed so I did not include them either.

Patch 1:
Fix for ring resets. In V8, return the proper job status so scheduler re-adds the job to the pending list itself.

Patches 2-4:
Removes the direct submit path for IBs and requires that all IB submissions use a job structure. This greatly simplifies the IB submission code. V2 uses GFP_ATOMIC when in reset. V3 sqaushes all of the IP changes into one patch. Not sure there is much value breaking this out per IP. V4 uses the pool type to determine the alloc flags. No changes in V8.

Patches 5-9:
Improvements for adapter resets. Properly add the bad job back to the pending list so we can stop calling drm_sched_stop/start(). Return the proper error number in the CS IOCTL when the context is guilty. Now that that is handled properly we can call drm_sched_increase_karma() for ring resets as well. Properly set the error on the the timedout fence for adapter resets. These changes also resulted in a small clean up the the VCN reset helper. In V8, patch 5 is updated to return the proper job status so the scheduler adds the job back to the pending list itself.

Patch 10-12:
Rework the backup and reemit code for per ring reset so that we can safely reemit repeatedly. This removes the current single reemit limit currently in place. V4 saves and restores the fences rather than emitting them directly and also saves and reemits the vm flush contents. V5 simplifies and further cleans things up. V6 reworks the IB schedule sequence to cleanly split the vm fence state from the IB fence state. This greatly simplifies the reemit code as the VM state is no longer intermixed with the IB state. V7 adds some minor improvements to the IB bookkeeping in the fences. No changes in V8.

Git tree available as well:
https://gitlab.freedesktop.org/agd5f/linux/-/commits/ib_improvements8?ref_type=heads

Alex Deucher (12):
drm/amdgpu: re-add the bad job to the pending list for ring resets drm/amdgpu/job: use GFP_ATOMIC while in gpu reset drm/amdgpu: switch all IPs to using job for IBs drm/amdgpu: require a job to schedule an IB
drm/amdgpu: don't call drm_sched_stop/start() in asic reset drm/amdgpu/cs: return -ETIME for guilty contexts drm/amdgpu: plumb timedout fence through to force completion drm/amdgpu: simplify VCN reset helper
drm/amdgpu: Call drm_sched_increase_karma() for ring resets drm/amdgpu: reorder IB schedule sequence
drm/amdgpu: add a helper to calculate ring distance drm/amdgpu: rework ring reset backup and reemit v8

 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c  |   4 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c      |   4 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c |   2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c  |  10 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c   | 124 ++++++++--------
 drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c      | 152 +++++++++-----------
 drivers/gpu/drm/amd/amdgpu/amdgpu_job.c     |  27 ++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_job.h     |   2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.h  |   2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c    |  50 ++-----
 drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h    |  44 +++---
 drivers/gpu/drm/amd/amdgpu/amdgpu_sa.c      |   5 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.c    |   4 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c     |   2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c     |  52 +++----
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c      |   7 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_vpe.c     |  37 ++---
 drivers/gpu/drm/amd/amdgpu/cik_sdma.c       |  31 ++--
 drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c      |  29 ++--
 drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c      |  29 ++--
 drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c      |  29 ++--
 drivers/gpu/drm/amd/amdgpu/gfx_v12_1.c      |  29 ++--
 drivers/gpu/drm/amd/amdgpu/gfx_v6_0.c       |  24 ++--
 drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c       |  25 ++--
 drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c       | 139 +++++++++---------
 drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c       | 138 +++++++++---------
 drivers/gpu/drm/amd/amdgpu/gfx_v9_4_2.c     |  26 ++--
 drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c     |  29 ++--
 drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c      |  38 ++---
 drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c      |  38 ++---
 drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c      |  38 ++---
 drivers/gpu/drm/amd/amdgpu/sdma_v4_4_2.c    |  38 ++---
 drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c      |  37 ++---
 drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c      |  36 ++---
 drivers/gpu/drm/amd/amdgpu/sdma_v6_0.c      |  36 ++---
 drivers/gpu/drm/amd/amdgpu/sdma_v7_0.c      |  36 ++---
 drivers/gpu/drm/amd/amdgpu/sdma_v7_1.c      |  36 ++---
 drivers/gpu/drm/amd/amdgpu/si_dma.c         |  29 ++--
 drivers/gpu/drm/amd/amdgpu/vcn_v4_0_3.c     |   2 +-
 drivers/gpu/drm/amd/amdgpu/vcn_v5_0_1.c     |   2 +-
 40 files changed, 725 insertions(+), 697 deletions(-)
Signature · 1 line
2.52.0