On Sun, Nov 22, 2026 at 09:31:43AM -0000, Arun Iyengar wrote:
> On Sat, Nov 14, 2026 at 06:00:39AM +0100, Sebastian Reichel wrote:
> > @@ -1491,6 +1493,12 @@ static ssize_t dw_dp_aux_transfer(struct drm_dp_aux 
> > *aux,
> >     status = wait_for_completion_timeout(&dp->complete, timeout);
> >     if (!status) {
> >             dev_err(dp->dev, "timeout waiting for AUX reply\n");
> 
> This still drops a reply the sink did send when the IRQ thread runs
> more than 10 ms late. On RK3588 (7.3, KASAN on, v11 of this series:
> same 06/23), with a SCHED_FIFO hog at the IRQ thread's priority on its
> CPU (busy 40 ms of every 200 ms), 1410 DPCD reads gave 183 timeouts.
> Checking for the reply once more before giving up gave 0 timeouts,
> with 127 late replies taken:
> 
> --- a/drivers/gpu/drm/bridge/synopsys/dw-dp.c
> +++ b/drivers/gpu/drm/bridge/synopsys/dw-dp.c
> @@ -1493,2 +1493,6 @@ static ssize_t dw_dp_aux_transfer(struct drm_dp_aux 
> *aux,
>       status = wait_for_completion_timeout(&dp->complete, timeout);
> +     if (!status) {
> +             synchronize_irq(dp->irq);
> +             status = try_wait_for_completion(&dp->complete);
> +     }
>       if (!status) {
> 
> Feel free to fold it in.
> 
> Signed-off-by: Arun Iyengar <[email protected]>
> 
> Claude (Anthropic) wrote this change; I reviewed and tested it.

There's something missing here. You may have reviewed (whatever that
means) and tested it (probably), it doesn't seem you've understood it.
All of this screams of a big hack.

-- 
Regards,

Laurent Pinchart

Reply via email to