Re: [PATCH v4 0/5] drm/vc4: dsi: Make failed reads visible and recoverable
From: Henry Zhang
Date: Wed Nov 18 2026 - 11:41:05 EST
On Wed, Nov 18, 2026 at 6:35 AM Maíra Canal <mcanal@xxxxxxxxxx> wrote:
> Applied the first patch (1/5) to drm-misc-fixes, thanks!
>
> I'll leave the review of the rest of the patches for Dave, as he has
> more expertise on the DSI side. Anyway, you should consider addressing
> Sashiko's comment on 4/5.
Thanks, Maíra.
Sashiko's comment is on 5/5. It concerns the single reg_dma_mem buffer
used by the DMA write workaround on BCM2835 DSI1. The threaded
interrupt handler writes INT_STAT through the same buffer as the
transfer path, without serialization, so a concurrent write to CTRL in
the timeout path can receive the wrong value.
This race predates the series and also affects the existing timeout
path. I sent a separate fix the same day:
[PATCH] drm/vc4: dsi: Serialize register writes through the DMA workaround
https://lore.kernel.org/all/20261028181720.3476629-1-henryzhang62@xxxxxxxxx/
That patch serializes the shared-buffer writes, while 5/5 keeps the
interrupt handler out during the reset itself. The separate fix is
compile-tested only; the RP4 tests do not exercise the BCM2835
DMA-workaround path.
If you prefer, I can resend the remaining patches as v5 with the
DMA-workaround fix first, making the dependency explicit and omitting
the already-applied 1/5. I'll wait for Dave's review before doing so.
Thanks,
Henry