[Intel-gfx] [PATCH] drm/i915: allocate large pointer arrays with vmalloc

[Jesse Barnes]

For awhile now, many of the GEM code paths have allocated page or
object arrays with the slab allocator.  This is nice and fast, but
won't work well if memory is fragmented, since the slab allocator works
with physically contiguous memory (i.e. order > 2 allocations are
likely to fail fairly early after booting and doing some work).

This patch works around the issue by falling back to vmalloc for
>PAGE_SIZE allocations.  This is ugly, but much less work than chaining
a bunch of pages together by hand (suprisingly there's not a bunch of
generic kernel helpers for this yet afaik).  vmalloc space is somewhat
precious on 32 bit kernels, but our allocations shouldn't be big enough
to cause problems, though they're routinely more than a page.

Note that this patch doesn't address the unchecked
alloc-based-on-ioctl-args in GEM; that needs to be fixed in a separate
patch.

Also, I've deliberately ignored the DRM's "area" junk.  I don't think
anyone actually uses it anymore and I'm hoping it gets ripped out soon.

Signed-off-by: Jesse Barnes <jbarnes at virtuousgeek.org>

diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index ee896d9..5f01e1d 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -349,7 +349,7 @@ i915_gem_shmem_pread_slow(struct drm_device *dev, struct drm_gem_object *obj,
 	last_data_page = (data_ptr + args->size - 1) / PAGE_SIZE;
 	num_pages = last_data_page - first_data_page + 1;
 
-	user_pages = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
+	user_pages = drm_calloc_large(num_pages, sizeof(struct page *));
 	if (user_pages == NULL)
 		return -ENOMEM;
 
@@ -429,7 +429,7 @@ fail_put_user_pages:
 		SetPageDirty(user_pages[i]);
 		page_cache_release(user_pages[i]);
 	}
-	kfree(user_pages);
+	drm_free_large(user_pages, num_pages * sizeof(struct page *));
 
 	return ret;
 }
@@ -649,7 +649,7 @@ i915_gem_gtt_pwrite_slow(struct drm_device *dev, struct drm_gem_object *obj,
 	last_data_page = (data_ptr + args->size - 1) / PAGE_SIZE;
 	num_pages = last_data_page - first_data_page + 1;
 
-	user_pages = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
+	user_pages = drm_calloc_large(num_pages, sizeof(struct page *));
 	if (user_pages == NULL)
 		return -ENOMEM;
 
@@ -719,7 +719,7 @@ out_unlock:
 out_unpin_pages:
 	for (i = 0; i < pinned_pages; i++)
 		page_cache_release(user_pages[i]);
-	kfree(user_pages);
+	drm_free_large(user_pages, num_pages * sizeof(struct page *));
 
 	return ret;
 }
@@ -824,7 +824,7 @@ i915_gem_shmem_pwrite_slow(struct drm_device *dev, struct drm_gem_object *obj,
 	last_data_page = (data_ptr + args->size - 1) / PAGE_SIZE;
 	num_pages = last_data_page - first_data_page + 1;
 
-	user_pages = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
+	user_pages = drm_calloc_large(num_pages, sizeof(struct page *));
 	if (user_pages == NULL)
 		return -ENOMEM;
 
@@ -902,7 +902,7 @@ fail_unlock:
 fail_put_user_pages:
 	for (i = 0; i < pinned_pages; i++)
 		page_cache_release(user_pages[i]);
-	kfree(user_pages);
+	drm_free_large(user_pages, num_pages * sizeof(struct page *));
 
 	return ret;
 }
@@ -1408,9 +1408,8 @@ i915_gem_object_put_pages(struct drm_gem_object *obj)
 		}
 	obj_priv->dirty = 0;
 
-	drm_free(obj_priv->pages,
-		 page_count * sizeof(struct page *),
-		 DRM_MEM_DRIVER);
+	drm_free_large(obj_priv->pages,
+		       page_count * sizeof(struct page *));
 	obj_priv->pages = NULL;
 }
 
@@ -2015,8 +2014,7 @@ i915_gem_object_get_pages(struct drm_gem_object *obj)
 	 */
 	page_count = obj->size / PAGE_SIZE;
 	BUG_ON(obj_priv->pages != NULL);
-	obj_priv->pages = drm_calloc(page_count, sizeof(struct page *),
-				     DRM_MEM_DRIVER);
+	obj_priv->pages = drm_calloc_large(page_count, sizeof(struct page *));
 	if (obj_priv->pages == NULL) {
 		DRM_ERROR("Faled to allocate page list\n");
 		obj_priv->pages_refcount--;
@@ -3102,7 +3100,7 @@ i915_gem_get_relocs_from_user(struct drm_i915_gem_exec_object *exec_list,
 		reloc_count += exec_list[i].relocation_count;
 	}
 
-	*relocs = drm_calloc(reloc_count, sizeof(**relocs), DRM_MEM_DRIVER);
+	*relocs = drm_calloc_large(reloc_count, sizeof(**relocs));
 	if (*relocs == NULL)
 		return -ENOMEM;
 
@@ -3116,8 +3114,7 @@ i915_gem_get_relocs_from_user(struct drm_i915_gem_exec_object *exec_list,
 				     exec_list[i].relocation_count *
 				     sizeof(**relocs));
 		if (ret != 0) {
-			drm_free(*relocs, reloc_count * sizeof(**relocs),
-				 DRM_MEM_DRIVER);
+			drm_free_large(*relocs, reloc_count * sizeof(**relocs));
 			*relocs = NULL;
 			return -EFAULT;
 		}
@@ -3156,7 +3153,7 @@ i915_gem_put_relocs_to_user(struct drm_i915_gem_exec_object *exec_list,
 	}
 
 err:
-	drm_free(relocs, reloc_count * sizeof(*relocs), DRM_MEM_DRIVER);
+	drm_free_large(relocs, reloc_count * sizeof(*relocs));
 
 	return ret;
 }
@@ -3189,10 +3186,8 @@ i915_gem_execbuffer(struct drm_device *dev, void *data,
 		return -EINVAL;
 	}
 	/* Copy in the exec list from userland */
-	exec_list = drm_calloc(sizeof(*exec_list), args->buffer_count,
-			       DRM_MEM_DRIVER);
-	object_list = drm_calloc(sizeof(*object_list), args->buffer_count,
-				 DRM_MEM_DRIVER);
+	exec_list = drm_calloc_large(sizeof(*exec_list), args->buffer_count);
+	object_list = drm_calloc_large(sizeof(*object_list), args->buffer_count);
 	if (exec_list == NULL || object_list == NULL) {
 		DRM_ERROR("Failed to allocate exec or object list "
 			  "for %d buffers\n",
@@ -3453,10 +3448,8 @@ err:
 	}
 
 pre_mutex_err:
-	drm_free(object_list, sizeof(*object_list) * args->buffer_count,
-		 DRM_MEM_DRIVER);
-	drm_free(exec_list, sizeof(*exec_list) * args->buffer_count,
-		 DRM_MEM_DRIVER);
+	drm_free_large(object_list, sizeof(*object_list) * args->buffer_count);
+	drm_free_large(exec_list, sizeof(*exec_list) * args->buffer_count);
 	drm_free(cliprects, sizeof(*cliprects) * args->num_cliprects,
 		 DRM_MEM_DRIVER);
 
diff --git a/include/drm/drmP.h b/include/drm/drmP.h
index c8c4221..3ffea26 100644
--- a/include/drm/drmP.h
+++ b/include/drm/drmP.h
@@ -1519,6 +1519,30 @@ static __inline__ void *drm_calloc(size_t nmemb, size_t size, int area)
 {
 	return kcalloc(nmemb, size, GFP_KERNEL);
 }
+
+static __inline__ void *drm_calloc_large(size_t nmemb, size_t size)
+{
+	u8 *addr;
+
+	if (size <= PAGE_SIZE)
+	    return kcalloc(nmemb, size, GFP_KERNEL);
+
+	addr = vmalloc(nmemb * size);
+	if (!addr)
+		return NULL;
+
+	memset(addr, 0, nmemb * size);
+
+	return addr;
+}
+
+static __inline void drm_free_large(void *ptr, int size)
+{
+	if (size <= PAGE_SIZE)
+		return kfree(ptr);
+
+	vfree(ptr);
+}
 #else
 extern void *drm_alloc(size_t size, int area);
 extern void drm_free(void *pt, size_t size, int area);