Re: [PATCH] drm/i915/dp: On DPCD read failure skip LT

[Imre Deak <imre.deak@intel.com>]

On Mon, May 11, 2026 at 08:50:07AM +0530, Arun R Murthy wrote:
> The commit e42e7e585984 ("drm/i915: Fix modesetting in case of
> unexpected AUX timeouts") continue the LT even if DPCD read failed.
> 
> But doing so will blot the dmesg with multiple AUX Tx errors, so skip
> the LT but continue the modeset seq for the reason said in the above
> mentioned patch.
> 
> Signed-off-by: Arun R Murthy <arun.r.murthy@intel.com>
> ---
>  drivers/gpu/drm/i915/display/intel_dp_link_training.c | 10 +++++++---
>  1 file changed, 7 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/display/intel_dp_link_training.c b/drivers/gpu/drm/i915/display/intel_dp_link_training.c
> index a26094223f78..2598e4ba9712 100644
> --- a/drivers/gpu/drm/i915/display/intel_dp_link_training.c
> +++ b/drivers/gpu/drm/i915/display/intel_dp_link_training.c
> @@ -1776,9 +1776,12 @@ void intel_dp_start_link_train(struct intel_atomic_state *state,
>  
>  	lttpr_count = intel_dp_init_lttpr_and_dprx_caps(intel_dp);
>  
> -	if (lttpr_count < 0)
> -		/* Still continue with enabling the port and link training. */
> -		lttpr_count = 0;
> +	if (lttpr_count < 0) {
> +		/* skip LT but complete the modeset seq */
> +		lt_dbg(intel_dp, DP_PHY_DPRX,
> +		       "Reading LTTPR/DPRX caps failed, skip LT\n");
> +		goto retry;

This does break the regular "link training sequence" referred to by
e42e7e585984, reintroducing the original problem fixed by that commit.

Avoiding AUX failures this way is also too ad-hoc: the sink
disconnecting at any other point of time than right here before you
check for it would still result in AUX transfer failures being logged.
So I don't think it's a good idea to special case this particular
failure point to avoid these logs.

> +	}
>  
>  	intel_dp_prepare_link_train(intel_dp, crtc_state);
>  
> @@ -1795,6 +1798,7 @@ void intel_dp_start_link_train(struct intel_atomic_state *state,
>  		return;
>  	}
>  
> +retry:
>  	intel_dp->link.seq_train_failures++;
>  
>  	/*
> -- 
> 2.25.1
>