Following on from the scenario Tvrtko envision to explain a hard-to-hit
race with multiple first waiters, we could also then race in the
__i915_request_irq_complete() and the bottom-half may miss the vital
irq-seqno barrier and so go to sleep not noticing their seqno is
complete.
Signed-off-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
Cc: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx>
---
drivers/gpu/drm/i915/i915_drv.h | 20 ++++++++++++++++++++
1 file changed, 20 insertions(+)
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index c269e0a..9a5498c 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -3999,6 +3999,8 @@ static inline bool __i915_request_irq_complete(struct drm_i915_gem_request *req)
*/
if (engine->irq_seqno_barrier &&
cmpxchg_relaxed(&engine->irq_posted, 1, 0)) {
+ struct task_struct *tsk;
+
/* The ordering of irq_posted versus applying the barrier
* is crucial. The clearing of the current irq_posted must
* be visible before we perform the barrier operation,
@@ -4012,6 +4014,24 @@ static inline bool __i915_request_irq_complete(struct drm_i915_gem_request *req)
* seqno update.
*/
engine->irq_seqno_barrier(engine);
+
+ /* If we consume the irq, but we are no longer the bottom-half,
+ * the real bottom-half may not have serialised their own
+ * seqno check with the irq-barrier (i.e. may have inspected
+ * the seqno before we believe it coherent).
+ */
+ rcu_read_lock();
+ tsk = READ_ONCE(engine->breadcrumbs.tasklet);
+ if (tsk != current)
+ /* Note that if the bottom-half is changed as we
+ * are sending the wake-up, the new bottom-half will
+ * be woken by whomever made the change. We only have
+ * to worry about when we steal the irq-posted for
+ * ourself.
+ */
+ wake_up_process(tsk);
+ rcu_read_lock();
+
if (i915_gem_request_completed(req))
return true;
}
--
2.8.1
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
