Hi Thomas, On Mon, May 30, 2022 at 10:34 AM Geert Uytterhoeven <geert@xxxxxxxxxxxxxx> wrote: > On Mon, May 30, 2022 at 9:47 AM Thomas Zimmermann <tzimmermann@xxxxxxx> wrote: > > Am 28.05.22 um 22:17 schrieb Geert Uytterhoeven: > > > On Mon, Oct 14, 2019 at 4:05 PM Thomas Zimmermann <tzimmermann@xxxxxxx> wrote: > > >> The display mode is set by converting the DRM display mode to an > > >> fb_info state and handling it to the fbdev driver's fb_setvar() > > >> function. This also requires a color depth, which we take from the > > >> value of struct drm_mode_config.preferred_depth > > >> > > >> Signed-off-by: Thomas Zimmermann <tzimmermann@xxxxxxx> > > > > > >> --- a/drivers/gpu/drm/drm_fbconv_helper.c > > >> +++ b/drivers/gpu/drm/drm_fbconv_helper.c > > >> @@ -919,6 +919,24 @@ static int drm_fbconv_update_fb_var_screeninfo_from_framebuffer( > > >> return 0; > > >> } > > >> > > >> +static int drm_fbconv_update_fb_var_screeninfo_from_simple_display_pipe( > > >> + struct fb_var_screeninfo *fb_var, struct drm_simple_display_pipe *pipe) > > >> +{ > > >> + struct drm_plane *primary = pipe->crtc.primary; > > >> + struct drm_fbconv_modeset *modeset = drm_fbconv_modeset_of_pipe(pipe); > > >> + > > >> + if (primary && primary->state && primary->state->fb) > > >> + return drm_fbconv_update_fb_var_screeninfo_from_framebuffer( > > >> + fb_var, primary->state->fb, > > >> + modeset->fb_info->fix.smem_len); > > >> + > > >> + fb_var->xres_virtual = fb_var->xres; > > >> + fb_var->yres_virtual = fb_var->yres; > > >> + fb_var->bits_per_pixel = modeset->dev->mode_config.preferred_depth; > > > > > > This looks wrong to me: IMHO bits_per_pixel should be derived from > > > the fourcc format of the _new_ mode to be set... > > > > Indeed, this appears to be wrong. > > OK. > > > > > > > > >> + > > >> + return 0; > > >> +} > > >> + > > >> /** > > >> * drm_fbconv_simple_display_pipe_mode_valid - default implementation for > > >> * struct drm_simple_display_pipe_funcs.mode_valid > > >> @@ -950,6 +968,28 @@ bool drm_fbconv_simple_display_pipe_mode_fixup( > > >> struct drm_crtc *crtc, const struct drm_display_mode *mode, > > >> struct drm_display_mode *adjusted_mode) > > >> { > > >> + struct drm_simple_display_pipe *pipe = > > >> + container_of(crtc, struct drm_simple_display_pipe, crtc); > > >> + struct drm_fbconv_modeset *modeset = drm_fbconv_modeset_of_pipe(pipe); > > >> + struct fb_var_screeninfo fb_var; > > >> + int ret; > > >> + > > >> + if (!modeset->fb_info->fbops->fb_check_var) > > >> + return true; > > >> + > > >> + drm_fbconv_init_fb_var_screeninfo_from_mode(&fb_var, mode); > > >> + > > >> + ret = drm_fbconv_update_fb_var_screeninfo_from_simple_display_pipe( > > >> + &fb_var, &modeset->pipe); > > >> + if (ret) > > >> + return true; > > >> + > > >> + ret = modeset->fb_info->fbops->fb_check_var(&fb_var, modeset->fb_info); > > > > > > ... hence this fails if the requested mode is valid with the new > > > fourcc format, but invalid with the old (but preferred) depth. > > > E.g. due to bandwidth limitations, a high-resolution mode is valid > > > with a low color depth, while a high color depth is limited to lower > > > resolutions. > > > Unfortunately we do not know the new fourcc format here, as both > > > drm_simple_display_pipe_funcs.mode_{valid,fixup}() are passed > > > the mode (from drm_mode_set.mode), but not the new format (from > > > drm_mode_set.fb->format). > > > > > > Am I missing something? Is the new format available in some other way? > > > > We can always get the format from the new plane state of > > modeset->pipe->plane. We'd have this in the atomic_check call. And it > > appears that drm_fbconv_simple_display_pipe_check() is a better place > > for this code anyway. > > Thanks, I'll give that a try! Getting the format from the new plane state of pipe->plane doesn't work, as pipe->plane.state->fb = NULL. But it is indeed available in the drm_simple_display_pipe_funcs.check() callback, so that seems to work... Gr{oetje,eeting}s, Geert -- Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that. -- Linus Torvalds