[Intel-gfx] [PATCH i-g-t v2] lib/igt_core: Prefer CLOCK_MONOTONIC_RAW

[Chris Wilson]

On Tue, Nov 17, 2015 at 02:18:09PM +0200, Joonas Lahtinen wrote:
> CLOCK_MONOTONIC_RAW is not affected by NTP, so it should be THE clock
> used for timing execution of tests.
> 
> v2:
> - Cache the used clock (Chris)
> - Do not change the clock during execution
> - Spit out and error if monotonic time can not be read
> 
> Cc: Thomas Wood <thomas.wood at intel.com>
> Cc: Chris Wilson <chris at chris-wilson.co.uk>
> Signed-off-by: Joonas Lahtinen <joonas.lahtinen at linux.intel.com>
> ---
>  lib/igt_core.c | 20 ++++++++++++++++++--
>  1 file changed, 18 insertions(+), 2 deletions(-)
> 
> diff --git a/lib/igt_core.c b/lib/igt_core.c
> index 04a0ab2..c56bc15 100644
> --- a/lib/igt_core.c
> +++ b/lib/igt_core.c
> @@ -220,6 +220,7 @@ static char *run_single_subtest = NULL;
>  static bool run_single_subtest_found = false;
>  static const char *in_subtest = NULL;
>  static struct timespec subtest_time;
> +static clockid_t igt_clock = (clockid_t)-1;
>  static bool in_fixture = false;
>  static bool test_with_subtests = false;
>  static bool in_atexit_handler = false;
> @@ -341,10 +342,25 @@ static void gettime(struct timespec *ts)
>  {
>  	memset(ts, 0, sizeof(*ts));
>  
> +	// Stay on the same clock for consistency.
> +	if (igt_clock != (clockid_t)-1) {
> +		clock_gettime(igt_clock, ts);
> +		return;
> +	}
> +
> +#ifdef CLOCK_MONOTONIC_RAW
> +	if (!clock_gettime(igt_clock = CLOCK_MONOTONIC_RAW, ts))
> +		return;
> +#endif
>  #ifdef CLOCK_MONOTONIC_COARSE
> -	if (clock_gettime(CLOCK_MONOTONIC_COARSE, ts))
> +	if (!clock_gettime(igt_clock = CLOCK_MONOTONIC_COARSE, ts))
> +		return;
>  #endif
> -		clock_gettime(CLOCK_MONOTONIC, ts);
> +	if (!clock_gettime(igt_clock = CLOCK_MONOTONIC, ts))
> +		return;
> +
> +	igt_warn("Unable to get monotonic time!\n");
> +	exit(IGT_EXIT_FAILURE);

Hmm. gettime is used inside the subtest execution, so I'm not too sure
what should be done here, but definitely not call exit() itself!

I think probably return -errno is best, and move the igt_assert() into
the callers (as whatever is appropriate).

A second patch to then export

void igt_gettime(struct timespec *ts)
{
	igt_assert_f(gettime(ts) == 0,
		     "Unable to get monotonic time\n");
}

and replace all of the mismash in the tests would be nice.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre