[PATCH i-g-t] tests/xe_eudebug_online: Ignore stale attention events during single-step

Manszewski, Christoph christoph.manszewski at intel.com
Thu Feb 6 16:02:53 UTC 2025


Hi Dominik,

On 6.02.2025 14:14, Dominik Grzegorzek wrote:
> Signle step testcases are not filtering stale attention events in
> resume trigger, as a result those expect fixed number of attention
> events to come. If for some reason attention wasn't handeled before the
> next one came test would fail. Filter out all atentions with seqno
> lower than last eu control call.
> 
> Signed-off-by: Dominik Grzegorzek <dominik.grzegorzek at intel.com>

Reviewed-by: Christoph Manszewski <christoph.manszewski at intel.com>

Thanks,
Christoph

> ---
>   tests/intel/xe_eudebug_online.c | 9 ++++++---
>   1 file changed, 6 insertions(+), 3 deletions(-)
> 
> diff --git a/tests/intel/xe_eudebug_online.c b/tests/intel/xe_eudebug_online.c
> index ecc32ee97..c98f1ef1f 100644
> --- a/tests/intel/xe_eudebug_online.c
> +++ b/tests/intel/xe_eudebug_online.c
> @@ -694,6 +694,9 @@ static void eu_attention_resume_single_step_trigger(struct xe_eudebug_debugger *
>   
>   	get_aips_offset_table(data, threads);
>   
> +	if (data->last_eu_control_seqno > att->base.seqno)
> +		return;
> +
>   	if (d->flags & TRIGGER_RESUME_PARALLEL_WALK) {
>   		if (data->stepped_threads_count != -1)
>   			if (data->steps_done < SINGLE_STEP_COUNT) {
> @@ -766,9 +769,9 @@ static void eu_attention_resume_single_step_trigger(struct xe_eudebug_debugger *
>   			     data->target_offset + steering_offset(threads)), sz);
>   	fsync(data->vm_fd);
>   
> -	eu_ctl_resume(d->master_fd, d->fd, att->client_handle,
> -		      att->exec_queue_handle, att->lrc_handle,
> -		      att->bitmask, att->bitmask_size);
> +	data->last_eu_control_seqno = eu_ctl_resume(d->master_fd, d->fd, att->client_handle,
> +						    att->exec_queue_handle, att->lrc_handle,
> +						    att->bitmask, att->bitmask_size);
>   
>   	if (data->single_step_bitmask)
>   		for (int i = 0; i < att->bitmask_size; i++)


More information about the igt-dev mailing list