On 2023-12-05 17:20, Felix Kuehling wrote:
Properly mark kfd_process->ef as __rcu and consistently access it with
rcu_dereference_protected.
Reported-by: kernel test robot <lkp@xxxxxxxxx>
Closes: https://lore.kernel.org/oe-kbuild-all/202312052245.yFpBSgNH-lkp@xxxxxxxxx/
Signed-off-by: Felix Kuehling <Felix.Kuehling@xxxxxxx>
ping.
Christian, would you review this patch, please?
Thanks,
Felix
---
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h | 2 +-
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 4 ++--
drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 2 +-
drivers/gpu/drm/amd/amdkfd/kfd_process.c | 6 ++++--
4 files changed, 8 insertions(+), 6 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
index f2e920734c98..20cb266dcedd 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
@@ -314,7 +314,7 @@ void amdgpu_amdkfd_gpuvm_unmap_gtt_bo_from_kernel(struct kgd_mem *mem);
int amdgpu_amdkfd_map_gtt_bo_to_gart(struct amdgpu_device *adev, struct amdgpu_bo *bo);
int amdgpu_amdkfd_gpuvm_restore_process_bos(void *process_info,
- struct dma_fence **ef);
+ struct dma_fence __rcu **ef);
int amdgpu_amdkfd_gpuvm_get_vm_fault_info(struct amdgpu_device *adev,
struct kfd_vm_fault_info *info);
int amdgpu_amdkfd_gpuvm_import_dmabuf_fd(struct amdgpu_device *adev, int fd,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
index 7d91f99acb59..8ba6f6c8363d 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
@@ -2806,7 +2806,7 @@ static void amdgpu_amdkfd_restore_userptr_worker(struct work_struct *work)
put_task_struct(usertask);
}
-static void replace_eviction_fence(struct dma_fence **ef,
+static void replace_eviction_fence(struct dma_fence __rcu **ef,
struct dma_fence *new_ef)
{
struct dma_fence *old_ef = rcu_replace_pointer(*ef, new_ef, true
@@ -2841,7 +2841,7 @@ static void replace_eviction_fence(struct dma_fence **ef,
* 7. Add fence to all PD and PT BOs.
* 8. Unreserve all BOs
*/
-int amdgpu_amdkfd_gpuvm_restore_process_bos(void *info, struct dma_fence **ef)
+int amdgpu_amdkfd_gpuvm_restore_process_bos(void *info, struct dma_fence __rcu **ef)
{
struct amdkfd_process_info *process_info = info;
struct amdgpu_vm *peer_vm;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
index 45366b4ca976..5a24097a9f28 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
@@ -917,7 +917,7 @@ struct kfd_process {
* fence will be triggered during eviction and new one will be created
* during restore
*/
- struct dma_fence *ef;
+ struct dma_fence __rcu *ef;
/* Work items for evicting and restoring BOs */
struct delayed_work eviction_work;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
index 71df51fcc1b0..14b11d61f8dd 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
@@ -1110,6 +1110,8 @@ static void kfd_process_wq_release(struct work_struct *work)
{
struct kfd_process *p = container_of(work, struct kfd_process,
release_work);
+ struct dma_fence *ef = rcu_dereference_protected(p->ef,
+ kref_read(&p->ref) == 0);
kfd_process_dequeue_from_all_devices(p);
pqm_uninit(&p->pqm);
@@ -1118,7 +1120,7 @@ static void kfd_process_wq_release(struct work_struct *work)
* destroyed. This allows any BOs to be freed without
* triggering pointless evictions or waiting for fences.
*/
- dma_fence_signal(p->ef);
+ dma_fence_signal(ef);
kfd_process_remove_sysfs(p);
@@ -1127,7 +1129,7 @@ static void kfd_process_wq_release(struct work_struct *work)
svm_range_list_fini(p);
kfd_process_destroy_pdds(p);
- dma_fence_put(p->ef);
+ dma_fence_put(ef);
kfd_event_free_process(p);
