Re: [PATCH] drm/msm/dsi: Restore widebus calculation for CMDMode panels

[Dmitry Baryshkov]

On Thu, Mar 12, 2026 at 01:10:07PM +0800, Pengyu Luo wrote:
> On Thu, Mar 12, 2026 at 6:31 AM Marijn Suijten
> <marijn.suijten@xxxxxxxxxxxxxx> wrote:
> >
> > Commit ac47870fd795 ("drm/msm/dsi: fix hdisplay calculation when
> > programming dsi registers") makes a false and ungrounded statement that
> > "Since CMD mode does not use this, we can remove !(msm_host->mode_flags
> > & MIPI_DSI_MODE_VIDEO) safely." which isn't the case at all.
> > dsi_timing_setup() affects both command mode and video mode panels, and
> > by no longer having any path that sets bits_per_pclk = 48 (contrary to
> > the updated code-comment) all DSI DSC panels on SM8350 and above (i.e.
> > with widebus support) regress thanks to this patch.
> >
> > The entire reason that video mode was originally omitted from this code
> > path is because it was never tested before; any change that enables
> > widebus for video mode panels should not regress the command mode path.
> >
> > Thus add back the path allows 6 bytes or 48 bits to be sent per pclk
> > on command mode DSI panels with widebus, restoring the panel on devices
> > like the Sony Xperia 1 III and upwards.
> >
> > Fixes: ac47870fd795 ("drm/msm/dsi: fix hdisplay calculation when programming dsi registers")
> > Signed-off-by: Marijn Suijten <marijn.suijten@xxxxxxxxxxxxxx>
> > ---
> 
> Apologies, I messed up, I had sent the same fixes days ago.
> https://lore.kernel.org/linux-arm-msm/20260307111250.105772-2-mitltlatltl@xxxxxxxxx/
> 
> > In addition I can't say I understand the original commit message
> > at all; it mentions a BPC=10 mode however the highest format that
> > mipi_dsi_pixel_format_to_bpp supports is RGB888 thus it won't
> > ever return anything above 24, which is the original amount the
> > non-command-mode path defaulted to (regardless of widebus)...  Was that
> > patch doing anything for video mode at all?
> >
> 
> RGB888 is the dst bpc, which is tied to qcom,mdss-dsi-bpp in the downstream.
> 
> Actually, we should use src bpc here, another fixe
> https://lore.kernel.org/linux-arm-msm/20260307111250.105772-1-mitltlatltl@xxxxxxxxx/

I'm really torn between your two patchsets here.

Marijn, Pengyu, what are your testing platforms and are you testing
video or command mode panels?

> > It feels like the conditional introduced here is only making things more
> > confusing, but I don't have enough input to confirm what the video-mode
> > path should be doing in widebus mode (multiply BPC by the number of
> > components and by 2 in case of widebus?).
> 
> I left a comment. For CMD mode, it consumes 6 bytes, for Video mode,
> 
>   * DPU sends 3 bytes per pclk cycle to DSI. If widebus is
> - * enabled, bus width is extended to 6 bytes.
> + * enabled, MDP always sends out 48-bit compressed data per
> + * pclk and on average, DSI consumes an amount of compressed
> + * data equivalent to the uncompressed pixel depth per pclk.

I think this comment more or less matches what I see in the docs.

-- 
With best wishes
Dmitry