[PATCH] tests/intel/xe_waitfence: Set ABSTIME flag for abstime subtest

[Nareshkumar Gollakoti <naresh.kumar.g@intel.com>]

The abstime subtest computes an absolute CLOCK_MONOTONIC deadline, but
passes flags=0 to DRM_IOCTL_XE_WAIT_USER_FENCE. Without
DRM_XE_UFENCE_WAIT_FLAG_ABSTIME, the kernel interprets the value as a
relative timeout.

Since the absolute monotonic timestamp is much larger than the intended
10 ms wait, the ioctl can sleep until IGT's inactivity timeout kills the
test if the user fence has not already been signaled.

This was observed as:
  igt@xe_waitfence@abstime - timeout -
  Inactivity timeout exceeded. Killing the current test with SIGQUIT.

Pass DRM_XE_UFENCE_WAIT_FLAG_ABSTIME for the abstime subtest, matching the
timeout value being supplied.

Assisted-by: GitHub Copilot:Claude Opus 4.6
Signed-off-by: Nareshkumar Gollakoti <naresh.kumar.g@intel.com>
---
 tests/intel/xe_waitfence.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/tests/intel/xe_waitfence.c b/tests/intel/xe_waitfence.c
index ecdce9f93..f453a4a2c 100644
--- a/tests/intel/xe_waitfence.c
+++ b/tests/intel/xe_waitfence.c
@@ -130,7 +130,8 @@ waitfence(int fd, enum waittype wt)
 		current = ts.tv_sec * 1e9 + ts.tv_nsec;
 		timeout = current + 10 * NSEC_PER_MSEC;
 		signalled = wait_ufence_abstime(fd, &wait_fence, 7, 0,
-						   timeout, 0);
+						   timeout,
+						   DRM_XE_UFENCE_WAIT_FLAG_ABSTIME);
 		igt_debug("wait type: ABSTIME - timeout: %" PRId64
 			  ", signalled: %" PRId64
 			  ", elapsed: %" PRId64 "\n",
-- 
2.43.0