On 2016-01-26 13:18, Emil Velikov wrote:
On 14 January 2016 at 08:23, Meng Yi meng.yi@nxp.com wrote:
switch (fb->pixel_format) { case DRM_FORMAT_RGB565: case DRM_FORMAT_RGB888:@@ -85,9 +88,6 @@ static void fsl_dcu_drm_plane_atomic_update(struct
drm_plane *plane,
unsigned int alpha, bpp; int index, ret;
if (!fb)return;... which no longer has the !fb check, and we'll crash with null deref a few lines below ?
If there is a legitimate situation where fb is null which also ultimately leads to a atomic_commit, I guess we should keep the return here...
I think I made a mistake here, fb check should not be removed . As Stefan mentioned, if fb check in fsl_dcu_drm_plane_atomic_check return 0, fsl_dcu_drm_plane_atomic_update will ultimately called, and we'll crash since plane->state->fb is NULL.
I believe you meant "Emil" in the above ;-) But seriously... afaict a fair few drivers do a similar !fb (even !state->crtc) check(s)... which makes me wonder if core DRM isn't the better place for it ? Or perhaps that's intentional as core provides the flexibility for each driver to mangle with the fb between .check and .disable ?
There seem to be a consensus to check crtc and fb on atomic_check.
However, in atomic_update, some drives do a NULL check on crtc only, and some on both, crtc and fb.
The comment in drm_atomic_plane_disabling says CRTC and FB should always be NULL together... So I guess it does not really matter all that much, unless there is anyway a bug.
Furthermore, it seems that in the null case, atomic_disable is called anyway (which this driver implements). Not sure if there is another case in which either of this two could be NULL and atomic_update could be called.
Since this patch is mainly addressing the null check in atomic_check, I will apply it without the change in atomic_update for now.
-- Stefan