On 2022-09-14 05:10, Christian König wrote:
Good point, will change to flush_work, and move it outside amdgpu_bo_reserve(root), as amdgpu_vm_pt_free_work takes amdgpu_bo_reserve(root), this will cause deadlock.Am 13.09.22 um 22:05 schrieb Philip Yang:Free page table BO from vm resv unlocked context generate below warnings. Add a pt_free_work in vm to free page table BO from vm->pt_freed list. pass vm resv unlock status from page table update caller, and add vm_bo entry to vm->pt_freed list and schedule the pt_free_work if calling with vm resv unlocked. WARNING: CPU: 12 PID: 3238 at drivers/gpu/drm/ttm/ttm_bo.c:106 ttm_bo_set_bulk_move+0xa1/0xc0 Call Trace: amdgpu_vm_pt_free+0x42/0xd0 [amdgpu] amdgpu_vm_pt_free_dfs+0xb3/0xf0 [amdgpu] amdgpu_vm_ptes_update+0x52d/0x850 [amdgpu] amdgpu_vm_update_range+0x2a6/0x640 [amdgpu] svm_range_unmap_from_gpus+0x110/0x300 [amdgpu] svm_range_cpu_invalidate_pagetables+0x535/0x600 [amdgpu] __mmu_notifier_invalidate_range_start+0x1cd/0x230 unmap_vmas+0x9d/0x140 unmap_region+0xa8/0x110 Signed-off-by: Philip Yang <Philip.Yang@xxxxxxx> --- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 4 ++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 6 +++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c | 47 ++++++++++++++++++++--- 3 files changed, 51 insertions(+), 6 deletions(-)diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.cindex 59cac347baa3..27e6155053b6 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c@@ -2022,6 +2022,9 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm)spin_lock_init(&vm->invalidated_lock); INIT_LIST_HEAD(&vm->freed); INIT_LIST_HEAD(&vm->done); + INIT_LIST_HEAD(&vm->pt_freed); + INIT_WORK(&vm->pt_free_work, amdgpu_vm_pt_free_work); + spin_lock_init(&vm->pt_free_lock); /* create scheduler entities for page table updates */r = drm_sched_entity_init(&vm->immediate, DRM_SCHED_PRIORITY_NORMAL, @@ -2244,6 +2247,7 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm)amdgpu_vm_free_mapping(adev, vm, mapping, NULL); } + cancel_work_sync(&vm->pt_free_work);We should probably not cancel the work here, but rather just wait for it to finish.
amdgpu_vm_pt_free_root(adev, vm); amdgpu_bo_unreserve(root); amdgpu_bo_unref(&root);diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.hindex 9ecb7f663e19..b77fe838c327 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h @@ -276,6 +276,11 @@ struct amdgpu_vm { /* BOs which are invalidated, has been updated in the PTs */ struct list_head done;+ /* PT BOs scheduled to free and fill with zero if vm_resv is not hold */+ struct list_head pt_freed; + struct work_struct pt_free_work; + spinlock_t pt_free_lock; + /* contains the page directory */ struct amdgpu_vm_bo_base root; struct dma_fence *last_update;@@ -471,6 +476,7 @@ int amdgpu_vm_pde_update(struct amdgpu_vm_update_params *params,int amdgpu_vm_ptes_update(struct amdgpu_vm_update_params *params, uint64_t start, uint64_t end, uint64_t dst, uint64_t flags); +void amdgpu_vm_pt_free_work(struct work_struct *work); #if defined(CONFIG_DEBUG_FS)void amdgpu_debugfs_vm_bo_info(struct amdgpu_vm *vm, struct seq_file *m); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.cindex 88de9f0d4728..c6f91731ecfb 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c@@ -624,12 +624,22 @@ static int amdgpu_vm_pt_alloc(struct amdgpu_device *adev,* * @entry: PDE to free */ -static void amdgpu_vm_pt_free(struct amdgpu_vm_bo_base *entry)+static void amdgpu_vm_pt_free(struct amdgpu_vm_bo_base *entry, bool unlocked){ struct amdgpu_bo *shadow; if (!entry->bo) return; + + if (unlocked) { + spin_lock(&entry->vm->pt_free_lock); + list_move(&entry->vm_status, &entry->vm->pt_freed); + spin_unlock(&entry->vm->pt_free_lock); + + schedule_work(&entry->vm->pt_free_work); + return; + } +I strongly prefer to handle this in amdgpu_vm_pt_free_dfs() or even the caller of this instead and keep this here the low level free function.
Agree. Will send out v5 patch. Philip
shadow = amdgpu_bo_shadowed(entry->bo); if (shadow) { ttm_bo_set_bulk_move(&shadow->tbo, NULL);@@ -641,6 +651,29 @@ static void amdgpu_vm_pt_free(struct amdgpu_vm_bo_base *entry)amdgpu_bo_unref(&entry->bo); } +void amdgpu_vm_pt_free_work(struct work_struct *work)I think we could make this one here static if we introduce a amdgpu_vm_pt_init()/amdgpu_vm_pt_fini() function. But not a must have.Christian.+{ + struct amdgpu_vm_bo_base *entry, *next; + struct amdgpu_vm *vm; + struct amdgpu_bo *root; + LIST_HEAD(pt_freed); + + vm = container_of(work, struct amdgpu_vm, pt_free_work); + + spin_lock(&vm->pt_free_lock); + list_splice_init(&vm->pt_freed, &pt_freed); + spin_unlock(&vm->pt_free_lock); + + root = amdgpu_bo_ref(vm->root.bo); + amdgpu_bo_reserve(root, true); + + list_for_each_entry_safe(entry, next, &pt_freed, vm_status) + amdgpu_vm_pt_free(entry, false); + + amdgpu_bo_unreserve(root); + amdgpu_bo_unref(&root); +} + /** * amdgpu_vm_pt_free_dfs - free PD/PT levels *@@ -652,16 +685,17 @@ static void amdgpu_vm_pt_free(struct amdgpu_vm_bo_base *entry)*/ static void amdgpu_vm_pt_free_dfs(struct amdgpu_device *adev, struct amdgpu_vm *vm, - struct amdgpu_vm_pt_cursor *start) + struct amdgpu_vm_pt_cursor *start, + bool unlocked) { struct amdgpu_vm_pt_cursor cursor; struct amdgpu_vm_bo_base *entry; for_each_amdgpu_vm_pt_dfs_safe(adev, vm, start, cursor, entry) - amdgpu_vm_pt_free(entry); + amdgpu_vm_pt_free(entry, unlocked); if (start) - amdgpu_vm_pt_free(start->entry); + amdgpu_vm_pt_free(start->entry, unlocked); } /**@@ -673,7 +707,7 @@ static void amdgpu_vm_pt_free_dfs(struct amdgpu_device *adev,*/void amdgpu_vm_pt_free_root(struct amdgpu_device *adev, struct amdgpu_vm *vm){ - amdgpu_vm_pt_free_dfs(adev, vm, NULL); + amdgpu_vm_pt_free_dfs(adev, vm, NULL, false); } /**@@ -966,7 +1000,8 @@ int amdgpu_vm_ptes_update(struct amdgpu_vm_update_params *params,if (cursor.entry->bo) { params->table_freed = true; amdgpu_vm_pt_free_dfs(adev, params->vm, - &cursor); + &cursor, + params->unlocked); } amdgpu_vm_pt_next(adev, &cursor); }
