drm/i915: Move idle checks before intel_engine_init_global_seqno()
authorChris Wilson <chris@chris-wilson.co.uk>
Fri, 21 Jul 2017 12:32:26 +0000 (13:32 +0100)
committerDaniel Vetter <daniel.vetter@ffwll.ch>
Thu, 27 Jul 2017 07:38:46 +0000 (09:38 +0200)
intel_engine_init_globa_seqno() may be called from an uncontrolled
set-wedged path where we have given up waiting for broken hw and declare
it defunct. Along that path, any sanity checks that the hw is idle
before we adjust its state will expectedly fail, so we simply cannot.
Instead of asserting inside init_global_seqno, we move them to the
normal caller reset_all_global_seqno() as it handles runtime seqno
wraparound.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20170721123238.16428-8-chris@chris-wilson.co.uk
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
drivers/gpu/drm/i915/i915_gem_request.c
drivers/gpu/drm/i915/intel_engine_cs.c

index 483af8921060bfcd963b062bf0845a94ce3f14d9..d93a185c0f0a855fd9b2c1b4e3338e1ceb16398f 100644 (file)
@@ -213,6 +213,10 @@ static int reset_all_global_seqno(struct drm_i915_private *i915, u32 seqno)
                                cond_resched();
                }
 
+               /* Check we are idle before we fiddle with hw state! */
+               GEM_BUG_ON(!intel_engine_is_idle(engine));
+               GEM_BUG_ON(i915_gem_active_isset(&engine->timeline->last_request));
+
                /* Finally reset hw state */
                intel_engine_init_global_seqno(engine, seqno);
                tl->seqno = seqno;
index c6ebfe7fb4f15d5e2ff8e3ef2fb4560260efbb11..9ab5969413722a5999a4266629ea2ba0fc2305f0 100644 (file)
@@ -337,9 +337,6 @@ void intel_engine_init_global_seqno(struct intel_engine_cs *engine, u32 seqno)
 {
        struct drm_i915_private *dev_priv = engine->i915;
 
-       GEM_BUG_ON(!intel_engine_is_idle(engine));
-       GEM_BUG_ON(i915_gem_active_isset(&engine->timeline->last_request));
-
        /* Our semaphore implementation is strictly monotonic (i.e. we proceed
         * so long as the semaphore value in the register/page is greater
         * than the sync value), so whenever we reset the seqno,