[SRU][Q][PATCH 0/3] Kernel lockup on Lenovo P14s Gen 6 AMD Ryzen AI 7 PRO 350

[Aaron Ma]

BugLink: https://bugs.launchpad.net/bugs/2148538

SRU Justification:

[ Impact ]
On Lenovo ThinkPad P14s Gen 6 systems with AMD Ryzen AI 7 PRO 350,
the system can lock up under graphical workloads. Kernel logs show
the AMDGPU MES ring buffer becoming full and hung task reports
involving amdgpu_tlb_fence_work.

The issue is caused by commit f3854e04b708 ("drm/amdgpu: attach tlb
fence to the PTs update") which attaches TLB fences too broadly and
can flood KIQ/MES TLB invalidation work.

[ Fix ]
Backport upstream fixes:
- d967509651601 ("drm/amdgpu: make sure userqs are enabled in userq IOCTLs")
- 9163fe4d790f (Revert "drm/amdgpu: don't attach the tlb fence for SI")
- e9f58ff991dd ("drm/amdgpu: rework how we handle TLB fences")

[ Test Plan ]
1. Boot the affected Lenovo P14s Gen 6 AMD Ryzen AI 7 PRO 350 system.
2. Exercise the graphical workload that previously reproduced the lockup.
3. Verify the system remains responsive and dmesg does not show MES ring
   buffer full or amdgpu_tlb_fence_work hung task messages.

A test kernel with this fix series ran for 12 hours on the affected
system without reproducing the lockup (LP #2148538 comment #29).

[ Where problems could occur ]
Regressions could appear as GPU hangs, failed VM updates, TLB flush
failures, or user queue/KFD workload failures on AMD GPUs.

[ Other Info ]
Noble linux is not affected (offending commit absent).
Resolute linux is not affected (fixes already present).

Alex Deucher (2):
  drm/amdgpu: make sure userqs are enabled in userq IOCTLs
  drm/amdgpu: rework how we handle TLB fences

Prike Liang (1):
  Revert "drm/amdgpu: don't attach the tlb fence for SI"

 drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c       | 16 ++++++++++++++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_userq.h       |  1 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_userq_fence.c |  6 ++++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c          |  9 ++++++---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h          |  2 ++
 5 files changed, 31 insertions(+), 3 deletions(-)

-- 
2.43.0