* Thomas Zimmermann <tzimmermann@xxxxxxx> [240226 09:10]: > Hi > > Am 26.02.24 um 10:01 schrieb Tomi Valkeinen: > > On 26/02/2024 10:26, Tomi Valkeinen wrote: > > > Hi Tony, > > > > > > On 25/02/2024 08:46, Tony Lindgren wrote: > > > > Here are two fixes for omapdrm console. > > > > > > How is it broken? I don't usually use the console (or fbdev) but > > > enabling it now, it seems to work fine for me, on DRA76 EVM with > > > HDMI output. > > Omapdrm implements drm_framebuffer_funcs.dirty withomap_framebuffer_dirty(). > AFAIK DRM semantics requires to run the dirty helper after writing to the > framebuffer's memory. Userspace does this via the DIRTYFB ioctl. [1] But (at > least) for correctness the console needs to do the same. > > [1] https://elixir.bootlin.com/linux/v6.7.6/source/drivers/gpu/drm/drm_ioctl.c#L679 Yes I noticed console not updating and bisected it down to the two commits listed. I did the bisect on a droid4 though with command mode LCD. I did not test with HDMI, will give that a try too. > > After applying your patches, I see a lot of cache-related artifacts on > > the screen when updating the fb. > > I guess we might need a dma-specific mmap helper to make this work > correctly. I can easily test this if you have some suggested patch to try. Hmm so I wonder if we now have double updates happening on HDMI? Regards, Tony
