Re: [PATCH 2/2] tests/intel/xe_exec_queue_property: Sync to latest KMD properties

[Matthew Brost <matthew.brost@intel.com>]

On Mon, Dec 01, 2025 at 12:08:07PM -0800, Matt Roper wrote:
> On Mon, Dec 01, 2025 at 10:21:14AM -0800, Matthew Brost wrote:
> > DRM_XE_EXEC_QUEUE_SET_HANG_REPLAY_STATE was added in b80961a86b40
> > ("drm/xe/uapi: Add DRM_XE_EXEC_QUEUE_SET_HANG_REPLAY_STATE"), update
> > invalidate property test to reflect latest KMD exec queue properties.
> > 
> > Signed-off-by: Matthew Brost <matthew.brost@intel.com>
> > ---
> >  tests/intel/xe_exec_queue_property.c | 3 ++-
> >  1 file changed, 2 insertions(+), 1 deletion(-)
> > 
> > diff --git a/tests/intel/xe_exec_queue_property.c b/tests/intel/xe_exec_queue_property.c
> > index 4dfcf09e26..7a2fbdeec8 100644
> > --- a/tests/intel/xe_exec_queue_property.c
> > +++ b/tests/intel/xe_exec_queue_property.c
> > @@ -172,6 +172,7 @@ static void basic_get_property(int xe)
> >  static void invalid_property(int xe)
> >  {
> >  	uint32_t valid_property = DRM_XE_EXEC_QUEUE_SET_PROPERTY_PRIORITY;
> > +	uint32_t max_property = DRM_XE_EXEC_QUEUE_SET_HANG_REPLAY_STATE;
> >  	struct drm_xe_engine_class_instance instance = {
> >  			.engine_class = DRM_XE_ENGINE_CLASS_VM_BIND,
> >  	};
> > @@ -189,7 +190,7 @@ static void invalid_property(int xe)
> >  
> >  	/* This will fail as soon as a new property is introduced. It is
> >  	 * expected and the test will have to be updated. */
> > -	for (int i = 3; i < 16; i++ ) {
> > +	for (int i = max_property + 1; i < 16; i++ ) {
> 
> Is this loop actually providing us any value?  It seems the main thing

I don't see a ton of value in the loop.

> we're trying to check is that the error value for invalid properties
> comes back as EINVAL instead of something else, and that setting a valid
> property after an invalid one stil works; it doesn't seem like a loop is
> really necessary to check either of those things.  Would we lose
> anything if we dropped the loop and instead just checked a single
> (really high, more future-proof) property ID?
>

+1, let me just update this test to do this so as we add properties CI
doesn't keep on breaking.

Matt

> 
> Matt
> 
> >  		ext.property = i;
> >  		igt_assert_eq(__xe_exec_queue_create(xe, vm, 1, 1, &instance,
> >  						     to_user_pointer(&ext), &exec_queue_id), -EINVAL);
> > -- 
> > 2.34.1
> > 
> 
> -- 
> Matt Roper
> Graphics Software Engineer
> Linux GPU Platform Enablement
> Intel Corporation