drm/i915: Serialize per-engine resets against new requests
authorChris Wilson <chris@chris-wilson.co.uk>
Fri, 21 Jul 2017 12:32:21 +0000 (13:32 +0100)
committerDaniel Vetter <daniel.vetter@ffwll.ch>
Thu, 27 Jul 2017 07:38:44 +0000 (09:38 +0200)
We rely on disabling the execlists (by stopping the tasklet) to prevent
new requests from submitting to the engine ELSP before we are ready.
However, we re-enable the engine before we call init_hw which gives
userspace the opportunity to subit a new request which is then
overwritten by init_hw -- but not before the HW may have started
executing. The subsequent out-of-order CSB is detected by our sanity
checks in intel_lrc_irq_handler().

Fixes: a1ef70e14453 ("drm/i915: Add support for per engine reset recovery")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Michel Thierry <michel.thierry@intel.com>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Reviewed-by: Michel Thierry <michel.thierry@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20170721123238.16428-3-chris@chris-wilson.co.uk
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
drivers/gpu/drm/i915/i915_drv.c

index 853d674..6753fc8 100644 (file)
@@ -1951,6 +1951,12 @@ int i915_reset_engine(struct intel_engine_cs *engine)
        }
 
        ret = intel_gpu_reset(engine->i915, intel_engine_flag(engine));
+       if (ret) {
+               /* If we fail here, we expect to fallback to a global reset */
+               DRM_DEBUG_DRIVER("Failed to reset %s, ret=%d\n",
+                                engine->name, ret);
+               goto out;
+       }
 
        /*
         * The request that caused the hang is stuck on elsp, we know the
@@ -1959,15 +1965,6 @@ int i915_reset_engine(struct intel_engine_cs *engine)
         */
        i915_gem_reset_engine(engine, active_request);
 
-       i915_gem_reset_finish_engine(engine);
-
-       if (ret) {
-               /* If we fail here, we expect to fallback to a global reset */
-               DRM_DEBUG_DRIVER("Failed to reset %s, ret=%d\n",
-                                engine->name, ret);
-               goto out;
-       }
-
        /*
         * The engine and its registers (and workarounds in case of render)
         * have been reset to their default values. Follow the init_ring
@@ -1979,6 +1976,7 @@ int i915_reset_engine(struct intel_engine_cs *engine)
 
        error->reset_engine_count[engine->id]++;
 out:
+       i915_gem_reset_finish_engine(engine);
        return ret;
 }