github-mirror/linux
1
0
Fork 0
mirror of https://github.com/torvalds/linux.git synced 2026-07-28 01:55:51 +02:00
Code Issues Actions Packages Projects Releases Wiki Activity Graph

drm/amdkfd: suspend scheduler during PTL re-enabling

Browse Source 
Stop the scheduler before releasing the PTL disable request to ensure
the GPU is quiescent during the PTL state transition. This prevents
potential queue preemption failures and GPU resets caused by modifying
PTL state while waves are executing

v1->v2:
only stop/start the scheduler when the PTL state actually needs to transition(Yifan)

Signed-off-by: Perry Yuan <perry.yuan@amd.com>
Reviewed-by: Yifan Zhang <yifan1.zhang@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This commit is contained in:
Perry Yuan 2026-02-09 00:42:08 +08:00 committed by Alex Deucher
parent dd61e27535
commit 022f071db5
1 changed files with 28 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

30
drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
View File

@ -1789,10 +1789,17 @@ static int kfd_ptl_control(struct kfd_process_device *pdd, bool enable)
if (!pdd->dev->kfd2kgd || !pdd->dev->kfd2kgd->ptl_ctrl)
return -EOPNOTSUPP;
if (adev->kfd.init_complete)
amdgpu_amdkfd_stop_sched(adev, pdd->dev->node_id);
ret = pdd->dev->kfd2kgd->ptl_ctrl(adev, PSP_PTL_PERF_MON_SET,
&ptl_state,
&pref_format1,
&pref_format2);
if (adev->kfd.init_complete)
amdgpu_amdkfd_start_sched(adev, pdd->dev->node_id);
return ret;
}
@ -3322,6 +3329,7 @@ static inline uint32_t profile_lock_device(struct kfd_process *p,
struct kfd_process_device *pdd;
struct kfd_dev *kfd;
int status = -EINVAL;
struct amdgpu_ptl *ptl;
if (!p)
return -EINVAL;
@ -3334,13 +3342,22 @@ static inline uint32_t profile_lock_device(struct kfd_process *p,
return -EINVAL;
kfd = pdd->dev->kfd;
ptl = &pdd->dev->adev->psp.ptl;
mutex_lock(&kfd->profiler_lock);
if (op == 1) {
if (!kfd->profiler_process) {
kfd->profiler_process = p;
status = 0;
kfd_ptl_disable_request(pdd, p);
mutex_unlock(&kfd->profiler_lock);
if (ptl->hw_supported) {
status = kfd_ptl_disable_request(pdd, p);
if (status != 0)
dev_err(kfd_device,
"Failed to lock device %d for profiling, error %d\n",
gpu_id, status);
}
return status;
} else if (kfd->profiler_process == p) {
status = -EALREADY;
} else {
@ -3349,7 +3366,16 @@ static inline uint32_t profile_lock_device(struct kfd_process *p,
} else if (op == 0 && kfd->profiler_process == p) {
kfd->profiler_process = NULL;
status = 0;
kfd_ptl_disable_release(pdd, p);
mutex_unlock(&kfd->profiler_lock);
if (ptl->hw_supported) {
status = kfd_ptl_disable_release(pdd, p);
if (status)
dev_err(kfd_device,
"Failed to unlock device %d for profiling, error %d\n",
gpu_id, status);
}
return status;
}
mutex_unlock(&kfd->profiler_lock);