Re: [PATCH] drm/amdgpu: Avoid extra evict-restore process.

[Philip Yang]

On 2025-07-08 16:14, Gang Ba wrote:
If vm belongs to another process, this is fclose after fork,
wait may enable signaling KFD eviction fence and cause parent process queue 
evicted.

Signed-off-by: Gang Ba <gang...@amd.com>
---
  drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 7 +++++++
  1 file changed, 7 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index f042372d9f2e..8ee1b7e62dee 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -2410,6 +2410,13 @@ void amdgpu_vm_adjust_size(struct amdgpu_device *adev, 
uint32_t min_vm_size,
   */
  long amdgpu_vm_wait_idle(struct amdgpu_vm *vm, long timeout)
  {
+       /* If vm belongs to another process, this is fclose after fork,
+        * wait may enable signaling KFD eviction fence and cause parent 
process queue evicted.
+        */
+       if (vm->task_info->tgid &&
+           vm->task_info->tgid != current->group_leader->pid)
+             return 0;
+

Only check this for KFD vm, in case this may cause gfx test regression.

    if (vm->is_compute_context && vm->task_info->tgid != 
current->group_leader->pid)

          return 0;


Regards,

Philip

        timeout = dma_resv_wait_timeout(vm->root.bo->tbo.base.resv,
                                        DMA_RESV_USAGE_BOOKKEEP,
                                        true, timeout);