Commit 97e8cecb authored by Matthew Brost's avatar Matthew Brost
Browse files

drm/xe: Add xe_ttm_access_memory 



Non-contiguous VRAM cannot easily be mapped in TTM nor can non-visible
VRAM easily be accessed. Add xe_ttm_access_memory which hooks into
ttm_bo_access to access such memory.

v4:
 - Assert memory access rather than taking RPM ref (Thomas / Auld)
 - Fix warning on xe_res_cursor.h for non-zero offset (Mika)

Reported-by: default avatarChristoph Manszewski <christoph.manszewski@intel.com>
Suggested-by: default avatarThomas Hellström <thomas.hellstrom@linux.intel.com>
Signed-off-by: default avatarMatthew Brost <matthew.brost@intel.com>
Reviewed-by: default avatarMatthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20241126174615.2665852-4-matthew.brost@intel.com
parent 7d08df5d

drivers/gpu/drm/xe/xe_bo.c

+56 −3
Original line number Diff line number Diff line
@@ -442,6 +442,14 @@ static void xe_ttm_tt_destroy(struct ttm_device *ttm_dev, struct ttm_tt *tt) 
	kfree(tt);

}



static bool xe_ttm_resource_visible(struct ttm_resource *mem)

{

	struct xe_ttm_vram_mgr_resource *vres =

		to_xe_ttm_vram_mgr_resource(mem);



	return vres->used_visible_size == mem->size;

}



static int xe_ttm_io_mem_reserve(struct ttm_device *bdev,

				 struct ttm_resource *mem)

{
@@ -453,11 +461,9 @@ static int xe_ttm_io_mem_reserve(struct ttm_device *bdev, 
		return 0;

	case XE_PL_VRAM0:

	case XE_PL_VRAM1: {

		struct xe_ttm_vram_mgr_resource *vres =

			to_xe_ttm_vram_mgr_resource(mem);

		struct xe_mem_region *vram = res_to_mem_region(mem);



		if (vres->used_visible_size < mem->size)

		if (!xe_ttm_resource_visible(mem))

			return -EINVAL;



		mem->bus.offset = mem->start << PAGE_SHIFT;
@@ -1121,6 +1127,52 @@ static void xe_ttm_bo_swap_notify(struct ttm_buffer_object *ttm_bo) 
	}

}



static int xe_ttm_access_memory(struct ttm_buffer_object *ttm_bo,

				unsigned long offset, void *buf, int len,

				int write)

{

	struct xe_bo *bo = ttm_to_xe_bo(ttm_bo);

	struct xe_device *xe = ttm_to_xe_device(ttm_bo->bdev);

	struct iosys_map vmap;

	struct xe_res_cursor cursor;

	struct xe_mem_region *vram;

	int bytes_left = len;



	xe_bo_assert_held(bo);

	xe_device_assert_mem_access(xe);



	if (!mem_type_is_vram(ttm_bo->resource->mem_type))

		return -EIO;



	/* FIXME: Use GPU for non-visible VRAM */

	if (!xe_ttm_resource_visible(ttm_bo->resource))

		return -EIO;



	vram = res_to_mem_region(ttm_bo->resource);

	xe_res_first(ttm_bo->resource, offset & PAGE_MASK,

		     bo->size - (offset & PAGE_MASK), &cursor);



	do {

		unsigned long page_offset = (offset & ~PAGE_MASK);

		int byte_count = min((int)(PAGE_SIZE - page_offset), bytes_left);



		iosys_map_set_vaddr_iomem(&vmap, (u8 __iomem *)vram->mapping +

					  cursor.start);

		if (write)

			xe_map_memcpy_to(xe, &vmap, page_offset, buf, byte_count);

		else

			xe_map_memcpy_from(xe, buf, &vmap, page_offset, byte_count);



		buf += byte_count;

		offset += byte_count;

		bytes_left -= byte_count;

		if (bytes_left)

			xe_res_next(&cursor, PAGE_SIZE);

	} while (bytes_left);



	return len;

}



const struct ttm_device_funcs xe_ttm_funcs = {

	.ttm_tt_create = xe_ttm_tt_create,

	.ttm_tt_populate = xe_ttm_tt_populate,
@@ -1130,6 +1182,7 @@ const struct ttm_device_funcs xe_ttm_funcs = { 
	.move = xe_bo_move,

	.io_mem_reserve = xe_ttm_io_mem_reserve,

	.io_mem_pfn = xe_ttm_io_mem_pfn,

	.access_memory = xe_ttm_access_memory,

	.release_notify = xe_ttm_bo_release_notify,

	.eviction_valuable = ttm_bo_eviction_valuable,

	.delete_mem_notify = xe_ttm_bo_delete_mem_notify,