drm/amdgpu: fix userptr HMM range handling v2

The basic problem here is that it's not allowed to page fault while
holding the reservation lock.

So it can happen that multiple processes try to validate an userptr
at the same time.

Work around that by putting the HMM range object into the mutex
protected bo list for now.

v2: make sure range is set to NULL in case of an error

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
CC: stable@vger.kernel.org
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

1 files changed, 15 insertions, 38 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 02a8f25ae134..965808f4f682 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -643,9 +643,6 @@ struct amdgpu_ttm_tt {
 	struct task_struct	*usertask;
 	uint32_t		userflags;
 	bool			bound;
-#if IS_ENABLED(CONFIG_DRM_AMDGPU_USERPTR)
-	struct hmm_range	*range;
-#endif
 };
 
 #define ttm_to_amdgpu_ttm_tt(ptr)	container_of(ptr, struct amdgpu_ttm_tt, ttm)
@@ -658,7 +655,8 @@ struct amdgpu_ttm_tt {
  * Calling function must call amdgpu_ttm_tt_userptr_range_done() once and only
  * once afterwards to stop HMM tracking
  */
-int amdgpu_ttm_tt_get_user_pages(struct amdgpu_bo *bo, struct page **pages)
+int amdgpu_ttm_tt_get_user_pages(struct amdgpu_bo *bo, struct page **pages,
+				 struct hmm_range **range)
 {
 	struct ttm_tt *ttm = bo->tbo.ttm;
 	struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(ttm);
@@ -668,16 +666,15 @@ int amdgpu_ttm_tt_get_user_pages(struct amdgpu_bo *bo, struct page **pages)
 	bool readonly;
 	int r = 0;
 
+	/* Make sure get_user_pages_done() can cleanup gracefully */
+	*range = NULL;
+
 	mm = bo->notifier.mm;
 	if (unlikely(!mm)) {
 		DRM_DEBUG_DRIVER("BO is not registered?\n");
 		return -EFAULT;
 	}
 
-	/* Another get_user_pages is running at the same time?? */
-	if (WARN_ON(gtt->range))
-		return -EFAULT;
-
 	if (!mmget_not_zero(mm)) /* Happens during process shutdown */
 		return -ESRCH;
 
@@ -695,7 +692,7 @@ int amdgpu_ttm_tt_get_user_pages(struct amdgpu_bo *bo, struct page **pages)
 
 	readonly = amdgpu_ttm_tt_is_readonly(ttm);
 	r = amdgpu_hmm_range_get_pages(&bo->notifier, mm, pages, start,
-				       ttm->num_pages, &gtt->range, readonly,
+				       ttm->num_pages, range, readonly,
 				       true, NULL);
 out_unlock:
 	mmap_read_unlock(mm);
@@ -713,30 +710,24 @@ out_unlock:
  *
  * Returns: true if pages are still valid
  */
-bool amdgpu_ttm_tt_get_user_pages_done(struct ttm_tt *ttm)
+bool amdgpu_ttm_tt_get_user_pages_done(struct ttm_tt *ttm,
+				       struct hmm_range *range)
 {
 	struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(ttm);
-	bool r = false;
 
-	if (!gtt || !gtt->userptr)
+	if (!gtt || !gtt->userptr || !range)
 		return false;
 
 	DRM_DEBUG_DRIVER("user_pages_done 0x%llx pages 0x%x\n",
 		gtt->userptr, ttm->num_pages);
 
-	WARN_ONCE(!gtt->range || !gtt->range->hmm_pfns,
-		"No user pages to check\n");
+	WARN_ONCE(!range->hmm_pfns, "No user pages to check\n");
 
-	if (gtt->range) {
-		/*
-		 * FIXME: Must always hold notifier_lock for this, and must
-		 * not ignore the return code.
-		 */
-		r = amdgpu_hmm_range_get_pages_done(gtt->range);
-		gtt->range = NULL;
-	}
-
-	return !r;
+	/*
+	 * FIXME: Must always hold notifier_lock for this, and must
+	 * not ignore the return code.
+	 */
+	return !amdgpu_hmm_range_get_pages_done(range);
 }
 #endif
 
@@ -813,20 +804,6 @@ static void amdgpu_ttm_tt_unpin_userptr(struct ttm_device *bdev,
 	/* unmap the pages mapped to the device */
 	dma_unmap_sgtable(adev->dev, ttm->sg, direction, 0);
 	sg_free_table(ttm->sg);
-
-#if IS_ENABLED(CONFIG_DRM_AMDGPU_USERPTR)
-	if (gtt->range) {
-		unsigned long i;
-
-		for (i = 0; i < ttm->num_pages; i++) {
-			if (ttm->pages[i] !=
-			    hmm_pfn_to_page(gtt->range->hmm_pfns[i]))
-				break;
-		}
-
-		WARN((i == ttm->num_pages), "Missing get_user_page_done\n");
-	}
-#endif
 }
 
 static void amdgpu_ttm_gart_bind(struct amdgpu_device *adev,