[Intel-gfx] [PATCH 2/2] drm/i915: Don't check for idleness before retiring after a GPU hang

[Chris Wilson]

When we force the cleanup after a GPU hang, we want to retire all
requests, or else we may leak them if truly wedged (and the GPU never
advances again). Converting to the active request helpers had the issue
of doing the check against busyness before reporting the request, so if
we claim the GPU had hung but this engine hadn't we could potential skip
the request cleanup - triggering the self-check BUG.

Fixes: dcff85c8443e ("drm/i915: Enable i915_gem_wait_for_idle() ...")
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen at linux.intel.com>
---
 drivers/gpu/drm/i915/i915_gem.c | 8 +++-----
 1 file changed, 3 insertions(+), 5 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index f4f8eaa90f2a..cd5229301aae 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -2423,15 +2423,11 @@ static void i915_gem_reset_engine_cleanup(struct intel_engine_cs *engine)
 	struct drm_i915_gem_request *request;
 	struct intel_ring *ring;
 
-	request = i915_gem_active_peek(&engine->last_request,
-				       &engine->i915->drm.struct_mutex);
-
 	/* Mark all pending requests as complete so that any concurrent
 	 * (lockless) lookup doesn't try and wait upon the request as we
 	 * reset it.
 	 */
-	if (request)
-		intel_engine_init_seqno(engine, request->fence.seqno);
+	intel_engine_init_seqno(engine, engine->last_submitted_seqno);
 
 	/*
 	 * Clear the execlists queue up before freeing the requests, as those
@@ -2453,6 +2449,8 @@ static void i915_gem_reset_engine_cleanup(struct intel_engine_cs *engine)
 	 * implicit references on things like e.g. ppgtt address spaces through
 	 * the request.
 	 */
+	request = i915_gem_active_raw(&engine->last_request,
+				      &engine->i915->drm.struct_mutex);
 	if (request)
 		i915_gem_request_retire_upto(request);
 	GEM_BUG_ON(intel_engine_is_active(engine));
-- 
2.8.1