Re: [PATCH v2] tests/intel/xe_pm_residency: Fix suspend duration calculation

["Tauro, Riana" <riana.tauro@intel.com>]

On 3/26/2026 2:59 PM, Sk Anirban wrote:
> Calculate actual suspend duration using CLOCK_BOOTTIME instead of
> relying on the autoresume delay approximation. This ensures accurate
> validation of idle residency measurements during suspend cycles.
>
> v2: Use CLOCK_BOOTTIME (Ville)
>      Use Fixes (Jani)
>
> Fixes: 5c85b336f75b ("tests/xe: Add a test that validates residency during s2idle")
> Signed-off-by: Sk Anirban <sk.anirban@intel.com>
> ---
>   tests/intel/xe_pm_residency.c | 19 ++++++++++++++-----
>   1 file changed, 14 insertions(+), 5 deletions(-)
>
> diff --git a/tests/intel/xe_pm_residency.c b/tests/intel/xe_pm_residency.c
> index 61b63ed4f..f2b802273 100644
> --- a/tests/intel/xe_pm_residency.c
> +++ b/tests/intel/xe_pm_residency.c
> @@ -215,18 +215,27 @@ static unsigned long read_idle_residency(int fd, int gt)
>   static void test_idle_residency(int fd, int gt, enum test_type flag)
>   {
>   	unsigned long elapsed_ms, residency_start, residency_end;
> +	struct timespec ts_start, ts_end;
> +	long elapsed_sec, elapsed_nsec;
>   
>   	igt_assert_f(igt_wait(xe_gt_is_in_c6(fd, gt), 1000, 1), "GT %d not in C6\n", gt);
>   
>   	if (flag == TEST_S2IDLE) {
> -		/*
> -		 * elapsed time during suspend is approximately equal to autoresume delay
> -		 * when a full suspend cycle(SUSPEND_TEST_NONE) is used.
> -		 */
> -		elapsed_ms = igt_get_autoresume_delay(SUSPEND_STATE_FREEZE);
> +		clock_gettime(CLOCK_BOOTTIME, &ts_start);
>   		residency_start = read_idle_residency(fd, gt);
>   		igt_system_suspend_autoresume(SUSPEND_STATE_FREEZE, SUSPEND_TEST_NONE);
>   		residency_end = read_idle_residency(fd, gt);
> +		clock_gettime(CLOCK_BOOTTIME, &ts_end);
> +
> +		elapsed_sec = ts_end.tv_sec - ts_start.tv_sec;
> +		elapsed_nsec = ts_end.tv_nsec - ts_start.tv_nsec;
> +
> +		if (elapsed_nsec < 0) {
> +			elapsed_sec--;
> +			elapsed_nsec += NSEC_PER_SEC;
>
> You could use igt_time_elapsed. Also remove the below check that hid the bug.
> It shouldn't be needed when boot time is used
>
> Thanks
> Riana
>
> +		}
> +
> +		elapsed_ms = (elapsed_sec * MSEC_PER_SEC) + (elapsed_nsec / NSEC_PER_MSEC);
>   	}
>   
>   	if (flag == TEST_IDLE) {