[PATCH 12/18] drm/i915: Park before resetting the submission backend

[Chris Wilson]

As different backends may have different park/unpark callbacks, we
should only ever switch backends (reset_default_submission on wedge
recovery, or on enabling the guc) while parked.

v2: Remove the assert from the guc code, as we are currently trying to
modify the engine vfuncs pointer on a live system after reset (not just
wedging). We will just have to hope that the system is balanced.

Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
Cc: Michal Wajdeczko <michal.wajdeczko at intel.com>
Cc: Sagar Arun Kamble <sagar.a.kamble at intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
Cc: Mika Kuoppala <mika.kuoppala at linux.intel.com>
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20180406205548.6952-6-michal.wajdeczko@intel.com
---
 drivers/gpu/drm/i915/i915_gem.c        | 14 ++++++++++++--
 drivers/gpu/drm/i915/intel_engine_cs.c |  3 +++
 2 files changed, 15 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 9f4d46e9f1c6..cf72defc948a 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -146,8 +146,6 @@ static u32 __i915_gem_park(struct drm_i915_private *i915)
 	if (!i915->gt.awake)
 		return I915_EPOCH_INVALID;
 
-	GEM_BUG_ON(i915->gt.epoch == I915_EPOCH_INVALID);
-
 	/*
 	 * Be paranoid and flush a concurrent interrupt to make sure
 	 * we don't reactivate any irq tasklets after parking.
@@ -176,6 +174,7 @@ static u32 __i915_gem_park(struct drm_i915_private *i915)
 
 	intel_runtime_pm_put(i915);
 
+	GEM_BUG_ON(i915->gt.epoch == I915_EPOCH_INVALID);
 	return i915->gt.epoch;
 }
 
@@ -3383,6 +3382,17 @@ bool i915_gem_unset_wedged(struct drm_i915_private *i915)
 	i915_retire_requests(i915);
 	GEM_BUG_ON(i915->gt.active_requests);
 
+	/*
+	 * Park before disengaging the old submit mechanism as different
+	 * backends may have different park/unpack callbacks.
+	 *
+	 * We are idle; the idle-worker will be queued, but we need to run
+	 * it now. As we already hold the struct mutex, we can get park
+	 * the GPU right away, letting the lazy worker see that we are
+	 * already active again by the time it acquires the mutex.
+	 */
+	i915_gem_park(i915);
+
 	/*
 	 * Undo nop_submit_request. We prevent all new i915 requests from
 	 * being queued (by disallowing execbuf whilst wedged) so having
diff --git a/drivers/gpu/drm/i915/intel_engine_cs.c b/drivers/gpu/drm/i915/intel_engine_cs.c
index 8c795f854c9b..978d8af3c451 100644
--- a/drivers/gpu/drm/i915/intel_engine_cs.c
+++ b/drivers/gpu/drm/i915/intel_engine_cs.c
@@ -1040,6 +1040,9 @@ void intel_engines_reset_default_submission(struct drm_i915_private *i915)
 	struct intel_engine_cs *engine;
 	enum intel_engine_id id;
 
+	/* Must be parked first! */
+	GEM_BUG_ON(i915->gt.awake);
+
 	for_each_engine(engine, i915, id)
 		engine->set_default_submission(engine);
 }
-- 
2.17.0