Hi
Am 06.05.19 um 15:17 schrieb Gerd Hoffmann:
Hi,
This misses the call to drm_gem_vram_placement(), where drm_gem_vram_push_to_system() enforces placement in system memory.
Ah, missed that detail.
We could build a common implementation out of both interfaces, but that would obfuscate the code IMHO. I'd just leave it as it is.
Ok.
+struct drm_gem_vram_object {
- /* Supported placements are %TTM_PL_VRAM and %TTM_PL_SYSTEM */
- struct ttm_placement placement;
- struct ttm_place placements[3];
placements[2] should be enough I guess?
TTM_PL_VRAM has index 2 and TTM_PL_SYSTEM has index 0. There's TTM_PL_TT at index 1. We don't use all three array entries here, but I'm not sure if something in TTM does. I took the line from the drivers and didn't change it for that reason.
TTM_PL_* isn't an index into that array. See drm_gem_vram_placement() which fills that array. It'll use one or two entries of that array.
The field 'placements' is assigned to 'placement.placements' and 'placement.busy_placement'. The placement field is later given to ttm_bo_validate() for internal use by TTM. From reading ttm_bo.c, that's apparently not a problem. I'm just being defensive here, but let's remove the additional entry if it's just overhead.
Best regards Thomas
cheers, Gerd
dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel