On 2023-01-09 14:27, Eric Huang wrote:
There will be data corruption on vram allocated by svm
if initialization is not being done. Adding sync is to
resolve this issue.

Signed-off-by: Eric Huang <jinhuieric.hu...@amd.com>
---
  drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 7 +++++++
  1 file changed, 7 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c 
b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
index b8c9753a4818..344e20306635 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
@@ -574,6 +574,13 @@ svm_range_vram_node_new(struct amdgpu_device *adev, struct 
svm_range *prange,
                goto reserve_bo_failed;
        }

Thanks for catching this bug, we could optimize as clear VRAM is only needed for partial migration, ex pass the clear flag clear = (cpages != npages) from svm_migrate_vma_to_vram -> svm_migrate_copy_to_vram -> svm_range_vram_node_new.

Regards,

Philip

+       r = amdgpu_bo_sync_wait(bo, AMDGPU_FENCE_OWNER_KFD, false);
+       if (r) {
+               pr_debug("failed %d to sync bo\n", r);
+               amdgpu_bo_unreserve(bo);
+               goto reserve_bo_failed;
+       }
+
        r = dma_resv_reserve_fences(amdkcl_ttm_resvp(&bo->tbo), 1);
        if (r) {
                pr_debug("failed %d to reserve bo\n", r);

Reply via email to