Re: [PATCH i-g-t 2/2] tests/intel/xe_pm: Add mocs suspend resume test

[Matt Roper <matthew.d.roper@intel.com>]

On Thu, Mar 21, 2024 at 08:46:37PM +0530, janga.rahul.kumar@intel.com wrote:
> From: Janga Rahul Kumar <janga.rahul.kumar@intel.com>
> 
> Add test to check mocs values are retained over suspend/resume.
> 
> Cc: Matt Roper <matthew.d.roper@intel.com>
> Signed-off-by: Janga Rahul Kumar <janga.rahul.kumar@intel.com>
> ---
>  tests/intel/xe_pm.c | 43 +++++++++++++++++++++++++++++++++++++++++++
>  1 file changed, 43 insertions(+)
> 
> diff --git a/tests/intel/xe_pm.c b/tests/intel/xe_pm.c
> index fcbed6249..0761d59c4 100644
> --- a/tests/intel/xe_pm.c
> +++ b/tests/intel/xe_pm.c
> @@ -560,6 +560,45 @@ static void test_mmap(device_t device, uint32_t placement, uint32_t flags)
>  	close(fw_handle);
>  }
>  
> +/**
> + * SUBTEST: mocs_suspend_resume
> + * Description:
> + *     Validate mocs register contents over suspend resume
> + *
> + * Functionality: mocs
> + * Run type: FULL
> + */
> +static void test_mocs_suspend_resume(device_t device)
> +{
> +	int gt;
> +
> +	xe_for_each_gt(device.fd_xe, gt) {
> +		char path[256];
> +
> +		// Mocs debugfs contents before and after suspend-resume
> +		char mocs_content_pre[4096], mocs_contents_post[4096];
> +
> +		sprintf(path, "gt%d/mocs", gt);
> +		igt_assert(igt_debugfs_exists(device.fd_xe, path, O_RDONLY));
> +		igt_debugfs_dump(device.fd_xe, path);
> +		igt_debugfs_read(device.fd_xe, path, mocs_content_pre);
> +
> +		fw_handle = igt_debugfs_open(device.fd_xe, "forcewake_all", O_RDONLY);
> +		igt_assert(fw_handle >= 0);
> +		igt_assert(igt_get_runtime_pm_status() == IGT_RUNTIME_PM_STATUS_ACTIVE);
> +
> +		/* Runtime suspend  */
> +		close(fw_handle);
> +		igt_assert(igt_wait_for_pm_status(IGT_RUNTIME_PM_STATUS_SUSPENDED));
> +
> +		igt_assert(igt_debugfs_exists(device.fd_xe, path, O_RDONLY));
> +		igt_debugfs_dump(device.fd_xe, path);
> +		igt_debugfs_read(device.fd_xe, path, mocs_contents_post);
> +
> +		igt_assert(strcmp(mocs_content_pre, mocs_contents_post) == 0);
> +	}
> +}

Assuming xe_pm has already ensured that anything else that might have
blocked runtime PM is out of the way, looks good to me.  But should we
also test S3 and S4 as well rather than just runtime suspend?  Those
follow different driver flows, and it's possible a driver bug could
cause of to fail to reprogram MOCS on an S3/S4 resume, even though
runtime PM resume works okay.

It would also be good to extend this same pattern to other kinds of
non-PM MOCS tests as well.  E.g., for reset IGTs it would be valuable to
have subtests like:

        a = capture
        trigger engine hang & reset
        b = capture
        fail if a != b

and

        a = capture
        trigger full GT reset
        b = capture
        fail if a != b


Matt

> +
>  igt_main
>  {
>  	struct drm_xe_engine_class_instance *hwe;
> @@ -701,6 +740,10 @@ igt_main
>  			dpms_on_off(device, DRM_MODE_DPMS_ON);
>  			igt_pm_set_autosuspend_delay(device.pci_xe, delay_ms);
>  		}
> +
> +		igt_subtest("mocs_suspend_resume")
> +			test_mocs_suspend_resume(device);
> +
>  	}
>  
>  	igt_fixture {
> -- 
> 2.25.1
> 

-- 
Matt Roper
Graphics Software Engineer
Linux GPU Platform Enablement
Intel Corporation