OSDN Git Service

drm/amdgpu: Reset CP_VMID_PREEMPT after trailing fence signaled
authorJiadong Zhu <Jiadong.Zhu@amd.com>
Wed, 24 May 2023 03:42:19 +0000 (11:42 +0800)
committerAlex Deucher <alexander.deucher@amd.com>
Fri, 9 Jun 2023 15:02:05 +0000 (11:02 -0400)
commitb7941e2fef13baabd3eade31601e70adf729e887
treeb80bc3043b0bdc624ccddb5918e6d7adfae35666
parent24bc366a4309f407ea77110ba15e3581005def6b
drm/amdgpu: Reset CP_VMID_PREEMPT after trailing fence signaled

When MEC executes unmap_queue for mid command buffer preemption, it will
kick the write pointer of the gfx ring, set CP_VMID_PREEMPT to trigger the
preemption and wait for CP_VMID_PREEMPT becomes zero after the preemption
done. There is a race condition that PFP may excute the resetting command
before MEC set CP_VMID_PREEMPT. As a result, hang happens as
CP_VMID_PREEMPT is always 0xffff.

To avoid this, we send resetting CP_VMID_PREEMPT command after the trailing
fence is siganled and update gfx write pointer explicitly.

Signed-off-by: Jiadong Zhu <Jiadong.Zhu@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c