On 11/5/25 07:16, Prike Liang wrote:
> The userq TLB flush should be performed after
> the vm update complete and input fence signal.
>
> Signed-off-by: Prike Liang <[email protected]>
> ---
> drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c | 1 +
> drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 5 +++--
> 2 files changed, 4 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c
> b/drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c
> index a8bb3f1b8649..01707d72d982 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c
> @@ -384,6 +384,7 @@ amdgpu_userq_wait_for_last_fence(struct amdgpu_userq_mgr
> *uq_mgr,
> return -ETIME;
> }
> }
> + amdgpu_gmc_flush_gpu_tlb_pasid(uq_mgr->adev, queue->vm->pasid, 2, true,
> 0);
Please drop that one, it doesn't seem to make any sense at all.
>
> return ret;
> }
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> index 48a64d828eb8..1537f0b6b712 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> @@ -1062,7 +1062,8 @@ amdgpu_vm_tlb_flush(struct amdgpu_vm_update_params
> *params,
> }
>
> /* Prepare a TLB flush fence to be attached to PTs */
> - if (!params->unlocked && vm->is_compute_context) {
> + if (!params->unlocked && (vm->is_compute_context ||
> + vm->is_userq_context)) {
Just drop the is_compute_context check here, don't add any new checks.
> amdgpu_vm_tlb_fence_create(params->adev, vm, fence);
>
> /* Makes sure no PD/PT is freed before the flush */
> @@ -2792,7 +2793,7 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct
> amdgpu_vm *vm)
> amdgpu_bo_unreserve(root);
> amdgpu_bo_unref(&root);
> WARN_ON(vm->root.bo);
> -
> + amdgpu_gmc_flush_gpu_tlb_pasid(adev, vm->pasid, 2, true, 0);
Please drop that one as well, it doesn't seem to make sense.
Regards,
Christian.
> amdgpu_vm_fini_entities(vm);
>
> if (!RB_EMPTY_ROOT(&vm->va.rb_root)) {
