[Intel-gfx] [igt-dev] [PATCH i-g-t] i915/perf: Skip OA testing on systems too old

[Chris Wilson]

Quoting Lionel Landwerlin (2019-12-16 10:06:53)
> On 16/12/2019 11:56, Chris Wilson wrote:
> 
>     Quoting Lionel Landwerlin (2019-12-16 09:46:56)
> 
>         On 16/12/2019 11:34, Chris Wilson wrote:
> 
>             Don't flat out fail if the system doesn't support OA, just skip.
> 
>             Closes: https://gitlab.freedesktop.org/drm/intel/issues/834
>             Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
>             ---
>               tests/perf.c | 4 +---
>               1 file changed, 1 insertion(+), 3 deletions(-)
> 
>             diff --git a/tests/perf.c b/tests/perf.c
>             index f5dd6051e..12f552743 100644
>             --- a/tests/perf.c
>             +++ b/tests/perf.c
>             @@ -884,11 +884,9 @@ init_sys_info(void)
>                   const char *test_set_uuid = NULL;
>                   char buf[256];
> 
>             -     igt_assert_neq(devid, 0);
>             -
>                   timestamp_frequency = get_cs_timestamp_frequency();
>                   igt_debug("timestamp_frequency = %lu\n", timestamp_frequency);
>             -     igt_assert_neq(timestamp_frequency, 0);
>             +     igt_require(timestamp_frequency);
> 
> 
>         This requires a kernel version more recent (4.16) than when perf support
>         was added (4.13).
> 
>         Is this what you intended?
> 
>     You have a fatal assert there. I am just changing it so that it skips
>     when not supported as no testing is being performed.
>     -Chris
> 
> I think there might be a problem in i915 if this returns 0.

It should return 0 for gen3 before Pineview.

However, since it returns i915->rawclk_freq on pnv and g4x, it should
have a value except that i915->cs_timestamp_freq is set in
intel_device_info_runtime_init (i915_driver_hw_probe) before the
rawclk_freq is set (i915_driver_modeset_probe).

Not sure the best approach to straighten out that mess... Just delaying
setting cs_timestamp_freq to i915_driver_register seems to be the best
idea.
-Chris