drm/i915/gem: Limit lmem scatterlist elements to UINT_MAX

Adhere to the i915_sg_max_segment() limit on the lengths of individual
scatterlist elements, and in doing so split up very large chunks of lmem
into manageable pieces for the dma-mapping backend.
Reported-by: Venkata Sandeep Dhanalakota <venkata.s.dhanalakota@intel.com>
Suggested-by: Matthew Auld <matthew.auld@intel.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Venkata Sandeep Dhanalakota <venkata.s.dhanalakota@intel.com>
Cc: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20201202173444.14903-1-chris@chris-wilson.co.uk

drivers/gpu/drm/i915/gem/i915_gem_region.c

@@ -22,6 +22,7 @@ i915_gem_object_put_pages_buddy(struct drm_i915_gem_object *obj,
 int
 i915_gem_object_get_pages_buddy(struct drm_i915_gem_object *obj)
 {
+	const u64 max_segment = i915_sg_segment_size();
 	struct intel_memory_region *mem = obj->mm.region;
 	struct list_head *blocks = &obj->mm.blocks;
 	resource_size_t size = obj->base.size;
@@ -37,7 +38,7 @@ i915_gem_object_get_pages_buddy(struct drm_i915_gem_object *obj)
 	if (!st)
 		return -ENOMEM;
 
-	if (sg_alloc_table(st, size >> ilog2(mem->mm.chunk_size), GFP_KERNEL)) {
+	if (sg_alloc_table(st, size >> PAGE_SHIFT, GFP_KERNEL)) {
 		kfree(st);
 		return -ENOMEM;
 	}
@@ -64,27 +65,30 @@ i915_gem_object_get_pages_buddy(struct drm_i915_gem_object *obj)
 				   i915_buddy_block_size(&mem->mm, block));
 		offset = i915_buddy_block_offset(block);
 
-		GEM_BUG_ON(overflows_type(block_size, sg->length));
+		while (block_size) {
+			u64 len;
 
-		if (offset != prev_end ||
-		    add_overflows_t(typeof(sg->length), sg->length, block_size)) {
+			if (offset != prev_end || sg->length >= max_segment) {
 				if (st->nents) {
 					sg_page_sizes |= sg->length;
 					sg = __sg_next(sg);
 				}
 
 				sg_dma_address(sg) = mem->region.start + offset;
-			sg_dma_len(sg) = block_size;
-
-			sg->length = block_size;
-
+				sg_dma_len(sg) = 0;
+				sg->length = 0;
 				st->nents++;
-		} else {
-			sg->length += block_size;
-			sg_dma_len(sg) += block_size;
 			}
 
-		prev_end = offset + block_size;
+			len = min(block_size, max_segment - sg->length);
+			sg->length += len;
+			sg_dma_len(sg) += len;
+
+			offset += len;
+			block_size -= len;
+
+			prev_end = offset;
+		}
 	}
 
 	sg_page_sizes |= sg->length;

drivers/gpu/drm/i915/selftests/intel_memory_region.c

@@ -129,6 +129,21 @@ static void igt_object_release(struct drm_i915_gem_object *obj)
 	i915_gem_object_put(obj);
 }
 
+static bool is_contiguous(struct drm_i915_gem_object *obj)
+{
+	struct scatterlist *sg;
+	dma_addr_t addr = -1;
+
+	for (sg = obj->mm.pages->sgl; sg; sg = sg_next(sg)) {
+		if (addr != -1 && sg_dma_address(sg) != addr)
+			return false;
+
+		addr = sg_dma_address(sg) + sg_dma_len(sg);
+	}
+
+	return true;
+}
+
 static int igt_mock_contiguous(void *arg)
 {
 	struct intel_memory_region *mem = arg;
@@ -150,8 +165,8 @@ static int igt_mock_contiguous(void *arg)
 	if (IS_ERR(obj))
 		return PTR_ERR(obj);
 
-	if (obj->mm.pages->nents != 1) {
-		pr_err("%s min object spans multiple sg entries\n", __func__);
+	if (!is_contiguous(obj)) {
+		pr_err("%s min object spans disjoint sg entries\n", __func__);
 		err = -EINVAL;
 		goto err_close_objects;
 	}
@@ -163,8 +178,8 @@ static int igt_mock_contiguous(void *arg)
 	if (IS_ERR(obj))
 		return PTR_ERR(obj);
 
-	if (obj->mm.pages->nents != 1) {
-		pr_err("%s max object spans multiple sg entries\n", __func__);
+	if (!is_contiguous(obj)) {
+		pr_err("%s max object spans disjoint sg entries\n", __func__);
 		err = -EINVAL;
 		goto err_close_objects;
 	}
@@ -189,8 +204,8 @@ static int igt_mock_contiguous(void *arg)
 		goto err_close_objects;
 	}
 
-	if (obj->mm.pages->nents != 1) {
-		pr_err("%s object spans multiple sg entries\n", __func__);
+	if (!is_contiguous(obj)) {
+		pr_err("%s object spans disjoint sg entries\n", __func__);
 		err = -EINVAL;
 		goto err_close_objects;
 	}