VM might already be freed when amdgpu_vm_tlb_seq_cb() is called.
We see the calltrace below.
Fix it by keeping the last flush fence around and wait for it to signal
BUG kmalloc-4k (Not tainted): Poison overwritten
0xffff9c88630414e8-0xffff9c88630414e8 @offset=5352. First byte 0x6c
instead of 0x6b Allocated in amdgpu_driver_open_kms+0x9d/0x360 [amdgpu]
age=44 cpu=0 pid=2343
__slab_alloc.isra.0+0x4f/0x90
kmem_cache_alloc_trace+0x6b8/0x7a0
amdgpu_driver_open_kms+0x9d/0x360 [amdgpu]
drm_file_alloc+0x222/0x3e0 [drm]
drm_open+0x11d/0x410 [drm]
drm_stub_open+0xdc/0x230 [drm]
chrdev_open+0xa5/0x1e0
do_dentry_open+0x16c/0x3c0
vfs_open+0x2d/0x30
path_openat+0x70a/0xa90
do_filp_open+0xb2/0x120
do_sys_openat2+0x245/0x330
do_sys_open+0x46/0x80
__x64_sys_openat+0x20/0x30
do_syscall_64+0x38/0xc0
entry_SYSCALL_64_after_hwframe+0x44/0xae
Freed in amdgpu_driver_postclose_kms+0x3e9/0x550 [amdgpu] age=22 cpu=1
pid=2485
kfree+0x4a2/0x580
amdgpu_driver_postclose_kms+0x3e9/0x550 [amdgpu]
drm_file_free+0x24e/0x3c0 [drm]
drm_close_helper.isra.0+0x90/0xb0 [drm]
drm_release+0x97/0x1a0 [drm]
__fput+0xb6/0x280
____fput+0xe/0x10
task_work_run+0x64/0xb0
do_exit+0x406/0xcf0
do_group_exit+0x50/0xc0
__x64_sys_exit_group+0x18/0x20
do_syscall_64+0x38/0xc0
entry_SYSCALL_64_after_hwframe+0x44/0xae
Suggested-by: Christian König <christian.koenig@xxxxxxx>
Signed-off-by: xinhui pan <xinhui.pan@xxxxxxx>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 22 +++++++++++++++++++---
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 1 +
2 files changed, 20 insertions(+), 3 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 645ce28277c2..e2486e95ca69 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -932,9 +932,12 @@ int amdgpu_vm_update_range(struct amdgpu_device *adev, struct amdgpu_vm *vm,
if (flush_tlb || params.table_freed) {
tlb_cb->vm = vm;
- if (!fence || !*fence ||
- dma_fence_add_callback(*fence, &tlb_cb->cb,
- amdgpu_vm_tlb_seq_cb))
+ if (fence && *fence &&
+ !dma_fence_add_callback(*fence, &tlb_cb->cb,
+ amdgpu_vm_tlb_seq_cb)) {
+ dma_fence_put(vm->last_delayed_tlb_flush);
+ vm->last_delayed_tlb_flush = dma_fence_get(*fence);
+ } else
amdgpu_vm_tlb_seq_cb(NULL, &tlb_cb->cb);
tlb_cb = NULL;
}
@@ -2258,6 +2261,19 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm)
dma_fence_wait(vm->last_unlocked, false);
dma_fence_put(vm->last_unlocked);
+ if (vm->last_delayed_tlb_flush) {
+ /* Wait until fence is signaled.
+ * But must double check to make sure fence cb is called.
+ * As dma_fence_default_wait checks DMA_FENCE_FLAG_SIGNALED_BIT without
+ * holding fence lock(the first test_bit).
+ * So call dma_fence_get_status which will hold the fence lock.
+ * Then we can make sure fence cb has been called.
+ */
+ (void)dma_fence_wait(vm->last_delayed_tlb_flush, false);
+ (void)dma_fence_get_status(vm->last_delayed_tlb_flush);
+ dma_fence_put(vm->last_delayed_tlb_flush);
+ }
+
list_for_each_entry_safe(mapping, tmp, &vm->freed, list) {
if (mapping->flags & AMDGPU_PTE_PRT && prt_fini_needed) {
amdgpu_vm_prt_fini(adev, vm);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
index 1a814fbffff8..c1a48f5c1019 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
@@ -286,6 +286,7 @@ struct amdgpu_vm {
/* Last finished delayed update */
atomic64_t tlb_seq;
+ struct dma_fence *last_delayed_tlb_flush;
/* Last unlocked submission to the scheduler entities */
struct dma_fence *last_unlocked;
--
2.25.1
