drm: Add support for GEM buffers backed by non-coherent memory

Having GEM buffers backed by non-coherent memory is interesting in the
particular case where it is faster to render to a non-coherent buffer
then sync the data cache, than to render to a write-combine buffer, and
(by extension) much faster than using a shadow buffer. This is true for
instance on some Ingenic SoCs, where even simple blits (e.g. memcpy)
are about three times faster using this method.

Add a 'map_noncoherent' flag to the drm_gem_cma_object structure, which
can be set by the drivers when they create the dumb buffer.

Since this really only applies to software rendering, disable this flag
as soon as the CMA objects are exported via PRIME.

v3: New patch. Now uses a simple 'map_noncoherent' flag to control how
the objects are mapped, and use the new dma_mmap_pages function.

v4: Make sure map_noncoherent is always disabled when creating GEM
objects meant to be used with dma-buf.

Signed-off-by: Paul Cercueil <[e-mail]>
Acked-by: Thomas Zimmermann <[e-mail]>
Link: https://patchwork.freedesktop.org/patch/msgid/[e-mail]

drivers/gpu/drm/drm_gem_cma_helper.c

···
  * __drm_gem_cma_create - Create a GEM CMA object without allocating memory
  * @drm: DRM device
  * @size: size of the object to allocate
+ * @private: true if used for internal purposes
  *
  * This function creates and initializes a GEM CMA object of the given size,
  * but doesn't allocate any memory to back the object.
···
  * error code on failure.
  */
 static struct drm_gem_cma_object *
-__drm_gem_cma_create(struct drm_device *drm, size_t size)
+__drm_gem_cma_create(struct drm_device *drm, size_t size, bool private)
 {
 	struct drm_gem_cma_object *cma_obj;
 	struct drm_gem_object *gem_obj;
-	int ret;
+	int ret = 0;
 
 	if (drm->driver->gem_create_object)
 		gem_obj = drm->driver->gem_create_object(drm, size);
···
 
 	cma_obj = container_of(gem_obj, struct drm_gem_cma_object, base);
 
-	ret = drm_gem_object_init(drm, gem_obj, size);
+	if (private) {
+		drm_gem_private_object_init(drm, gem_obj, size);
+
+		/* Always use writecombine for dma-buf mappings */
+		cma_obj->map_noncoherent = false;
+	} else {
+		ret = drm_gem_object_init(drm, gem_obj, size);
+	}
 	if (ret)
 		goto error;
 
···
 
 	size = round_up(size, PAGE_SIZE);
 
-	cma_obj = __drm_gem_cma_create(drm, size);
+	cma_obj = __drm_gem_cma_create(drm, size, false);
 	if (IS_ERR(cma_obj))
 		return cma_obj;
 
-	cma_obj->vaddr = dma_alloc_wc(drm->dev, size, &cma_obj->paddr,
-				      GFP_KERNEL | __GFP_NOWARN);
+	if (cma_obj->map_noncoherent) {
+		cma_obj->vaddr = dma_alloc_noncoherent(drm->dev, size,
+						       &cma_obj->paddr,
+						       DMA_TO_DEVICE,
+						       GFP_KERNEL | __GFP_NOWARN);
+	} else {
+		cma_obj->vaddr = dma_alloc_wc(drm->dev, size, &cma_obj->paddr,
+					      GFP_KERNEL | __GFP_NOWARN);
+	}
 	if (!cma_obj->vaddr) {
 		drm_dbg(drm, "failed to allocate buffer with size %zu\n",
 			 size);
···
 		return ERR_PTR(-EINVAL);
 
 	/* Create a CMA GEM buffer. */
-	cma_obj = __drm_gem_cma_create(dev, attach->dmabuf->size);
+	cma_obj = __drm_gem_cma_create(dev, attach->dmabuf->size, true);
 	if (IS_ERR(cma_obj))
 		return ERR_CAST(cma_obj);
 
···
 
 	cma_obj = to_drm_gem_cma_obj(obj);
 
-	ret = dma_mmap_wc(cma_obj->base.dev->dev, vma, cma_obj->vaddr,
-			  cma_obj->paddr, vma->vm_end - vma->vm_start);
+	vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
+	if (!cma_obj->map_noncoherent)
+		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
+
+	ret = dma_mmap_pages(cma_obj->base.dev->dev,
+			     vma, vma->vm_end - vma->vm_start,
+			     virt_to_page(cma_obj->vaddr));
 	if (ret)
 		drm_gem_vm_close(vma);
 

include/drm/drm_gem_cma_helper.h

···
  *       more than one entry but they are guaranteed to have contiguous
  *       DMA addresses.
  * @vaddr: kernel virtual address of the backing memory
+ * @map_noncoherent: if true, the GEM object is backed by non-coherent memory
  */
 struct drm_gem_cma_object {
 	struct drm_gem_object base;
···
 
 	/* For objects with DMA memory allocated by GEM CMA */
 	void *vaddr;
+
+	bool map_noncoherent;
 };
 
 #define to_drm_gem_cma_obj(gem_obj) \