Re: [PATCH i-g-t v4 3/3] tests/intel/kms_pipe_stress: Fit display modes to MST link budget

[Jani Nikula <jani.nikula@intel.com>]

On Mon, 29 Jun 2026, Sowmiya S <sowmiya.s@intel.com> wrote:
> When multiple displays share a single MST link,
> running them at their highest modes can exceed the
> available bandwidth.This causes the stress loop to
> fail with -ENOSPC when the driver rejects the commit.
> Before allocating framebuffers,check that the combined
> mode bandwidth fits the link and scale down modes if
> needed,so stress loop never commits an impossible config.

I don't see any reason in the patch to require the use of enum pipe. It
should be avoided if it's at all possible.

>
> Signed-off-by: Sowmiya S <sowmiya.s@intel.com>
> ---
>  tests/intel/kms_pipe_stress.c | 51 +++++++++++++++++++++++++++++++++++
>  1 file changed, 51 insertions(+)
>
> diff --git a/tests/intel/kms_pipe_stress.c b/tests/intel/kms_pipe_stress.c
> index 36e3842c6..84dcba11f 100644
> --- a/tests/intel/kms_pipe_stress.c
> +++ b/tests/intel/kms_pipe_stress.c
> @@ -763,6 +763,8 @@ static void prepare_test(struct data *data)
>  	int i, j;
>  	int num_connectors;
>  	int num_cpus = (int) sysconf(_SC_NPROCESSORS_ONLN);
> +	igt_output_t *output;
> +	enum pipe pipe = PIPE_A;
>  
>  	data->number_of_cores = min(num_cpus, MAX_CORES);
>  
> @@ -820,6 +822,55 @@ static void prepare_test(struct data *data)
>  			 data->num_planes[i], i);
>  	}
>  
> +	/*
> +	 * Assign all connected outputs to their pipes and fit the modes
> +	 * within the available link bandwidth.  On shared-link MST
> +	 * configurations such as a TBT dock (e.g. 2x4K displays)
> +	 * the combined DP stream bandwidth can exceed the link capacity
> +	 * and cause subsequent commits to fail with -ENOSPC.
> +	 * igt_require_mst_link_bw() detects this and selects the highest
> +	 * modes that fit the link budget; for non-MST setups it is a
> +	 * no-op.  Update data->highest_mode[] before create_framebuffers()
> +	 * so that FBs are allocated at the BW-constrained dimensions and
> +	 * the stress loop commits with a mode the driver will accept.
> +	 */
> +	igt_display_reset(display);
> +	for_each_connected_output(display, output) {
> +		igt_crtc_t *crtc = igt_crtc_for_pipe(display, pipe);
> +
> +		if (!crtc || !data->highest_mode[pipe]) {
> +			if (++pipe >= IGT_MAX_PIPES)
> +				break;
> +			continue;
> +		}
> +
> +		igt_output_set_crtc(output, crtc);
> +		igt_output_override_mode(output, data->highest_mode[pipe]);
> +		pipe++;
> +		if (pipe >= IGT_MAX_PIPES)
> +			break;
> +	}
> +
> +	igt_require_mst_link_bw(display);
> +
> +	/* Update highest_mode[] with the (possibly fitted) modes. */
> +	for_each_connected_output(display, output) {
> +		igt_crtc_t *crtc = output->pending_crtc;
> +		enum pipe p;
> +
> +		if (!crtc)
> +			continue;
> +
> +		p = crtc->pipe;
> +		if (!data->highest_mode[p])
> +			continue;
> +
> +		data->highest_mode[p] = igt_output_get_mode(output);
> +		igt_assert(data->highest_mode[p]);
> +		igt_info("Mode for pipe %d after BW fitting:\n", (int)p);

There's a helper for printing pipe name. (But hopefully you can just
switch to using igt_crtc_t and CRTC index.)

> +		kmstest_dump_mode(data->highest_mode[p]);
> +	}
> +
>  	create_framebuffers(data);
>  
>  	if (intel_gen(intel_get_drm_devid(data->drm_fd)) > 9)

-- 
Jani Nikula, Intel