[Intel-gfx] [PATCH] drm/i915/gt: Be paranoid and reset the GPU before release

[Chris Wilson]

Just in the very unlikely case we have not stopped the GPU before we
return the pages being used by the GPU to the system, force a reset.

Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld at intel.com>
---
 drivers/gpu/drm/i915/gt/intel_engine_cs.c | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)

diff --git a/drivers/gpu/drm/i915/gt/intel_engine_cs.c b/drivers/gpu/drm/i915/gt/intel_engine_cs.c
index 2d0ef5fc6748..b4b69a3fa25c 100644
--- a/drivers/gpu/drm/i915/gt/intel_engine_cs.c
+++ b/drivers/gpu/drm/i915/gt/intel_engine_cs.c
@@ -35,6 +35,7 @@
 #include "intel_engine_user.h"
 #include "intel_gt.h"
 #include "intel_gt_requests.h"
+#include "intel_gt_pm.h"
 #include "intel_lrc.h"
 #include "intel_reset.h"
 #include "intel_ring.h"
@@ -392,8 +393,23 @@ void intel_engines_release(struct intel_gt *gt)
 	struct intel_engine_cs *engine;
 	enum intel_engine_id id;
 
+	/*
+	 * Before we release the resources held by engine, we must be certain
+	 * that the HW is no longer accessing them -- having the GPU scribble
+	 * to or read from a page being used for something else causes no end
+	 * of fun.
+	 *
+	 * The GPU should be reset by this point, but assume the worst just
+	 * in case we aborted before completely initialising the engines.
+	 */
+	GEM_BUG_ON(intel_gt_pm_is_awake(gt));
+	if (!INTEL_INFO(gt->i915)->gpu_reset_clobbers_display)
+		__intel_gt_reset(gt, ALL_ENGINES);
+
 	/* Decouple the backend; but keep the layout for late GPU resets */
 	for_each_engine(engine, gt, id) {
+		GEM_BUG_ON(intel_engine_pm_is_awake(engine));
+
 		if (!engine->release)
 			continue;
 
-- 
2.25.0