Re: [PATCH v2 2/7] drm/xe: Use usleep_range for accurate long-running workload timeslicing

["Thomas Hellström" <thomas.hellstrom@linux.intel.com>]

On Fri, 2025-12-12 at 10:28 -0800, Matthew Brost wrote:
> msleep is not very accurate in terms of how long it actually sleeps,
> whereas usleep_range is precise. Replace the timeslice sleep for
> long-running workloads with the more accurate usleep_range to avoid
> jitter if the sleep period is less than 20ms.
> 
> Fixes: dd08ebf6c352 ("drm/xe: Introduce a new DRM driver for Intel
> GPUs")
> Cc: stable@vger.kernel.org
> Signed-off-by: Matthew Brost <matthew.brost@intel.com>

Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>

> ---
>  drivers/gpu/drm/xe/xe_guc_submit.c | 20 +++++++++++++++++++-
>  1 file changed, 19 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/xe/xe_guc_submit.c
> b/drivers/gpu/drm/xe/xe_guc_submit.c
> index 21a8bd2ec672..18cac5594d6a 100644
> --- a/drivers/gpu/drm/xe/xe_guc_submit.c
> +++ b/drivers/gpu/drm/xe/xe_guc_submit.c
> @@ -990,6 +990,24 @@ static u32 wq_space_until_wrap(struct
> xe_exec_queue *q)
>  	return (WQ_SIZE - q->guc->wqi_tail);
>  }
>  
> +static inline void relaxed_ms_sleep(unsigned int delay_ms)
> +{
> +	unsigned long min_us, max_us;
> +
> +	if (!delay_ms)
> +		return;
> +
> +	if (delay_ms > 20) {
> +		msleep(delay_ms);
> +		return;
> +	}
> +
> +	min_us = mul_u32_u32(delay_ms, 1000);
> +	max_us = min_us + 500;
> +
> +	usleep_range(min_us, max_us);
> +}
> +
>  static int wq_wait_for_space(struct xe_exec_queue *q, u32 wqi_size)
>  {
>  	struct xe_guc *guc = exec_queue_to_guc(q);
> @@ -1903,7 +1921,7 @@ static void
> __guc_exec_queue_process_msg_suspend(struct xe_sched_msg *msg)
>  				since_resume_ms;
>  
>  			if (wait_ms > 0 && q->guc->resume_time)
> -				msleep(wait_ms);
> +				relaxed_ms_sleep(wait_ms);
>  
>  			set_exec_queue_suspended(q);
>  			disable_scheduling(q, false);