spin_lock(&engine->breadcrumbs.irq_lock);
wait = engine->breadcrumbs.irq_wait;
if (wait) {
+ bool wakeup = engine->irq_seqno_barrier;
+
/* We use a callback from the dma-fence to submit
* requests after waiting on our own requests. To
* ensure minimum delay in queuing the next request to
* and many waiters.
*/
if (i915_seqno_passed(intel_engine_get_seqno(engine),
- wait->seqno) &&
- !test_bit(DMA_FENCE_FLAG_SIGNALED_BIT,
- &wait->request->fence.flags))
- rq = i915_gem_request_get(wait->request);
+ wait->seqno)) {
+ wakeup = true;
+ if (!test_bit(DMA_FENCE_FLAG_SIGNALED_BIT,
+ &wait->request->fence.flags))
+ rq = i915_gem_request_get(wait->request);
+ }
- wake_up_process(wait->tsk);
+ if (wakeup)
+ wake_up_process(wait->tsk);
} else {
__intel_engine_disarm_breadcrumbs(engine);
}