OSDN Git Service

drm/i915/execlists: Process interrupted context on reset
authorChris Wilson <chris@chris-wilson.co.uk>
Tue, 16 Jul 2019 12:49:29 +0000 (13:49 +0100)
committerChris Wilson <chris@chris-wilson.co.uk>
Wed, 17 Jul 2019 13:44:57 +0000 (14:44 +0100)
By stopping the rings, we may trigger an arbitration point resulting in
a premature context-switch (i.e. a completion event before the request
is actually complete). This clears the active context before the reset,
but we must remember to rewind the incomplete context for replay upon
resume.

Fixes: 1863e3020ab5 ("drm/i915/execlists: Always reset the context's RING registers")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190716124931.5870-3-chris@chris-wilson.co.uk
drivers/gpu/drm/i915/gt/intel_lrc.c

index f35a57d..6564b5d 100644 (file)
@@ -1425,7 +1425,8 @@ static void process_csb(struct intel_engine_cs *engine)
                         * coherent (visible from the CPU) before the
                         * user interrupt and CSB is processed.
                         */
-                       GEM_BUG_ON(!i915_request_completed(*execlists->active));
+                       GEM_BUG_ON(!i915_request_completed(*execlists->active) &&
+                                  !reset_in_progress(execlists));
                        execlists_schedule_out(*execlists->active++);
 
                        GEM_BUG_ON(execlists->active - execlists->inflight >
@@ -2251,7 +2252,7 @@ static void __execlists_reset(struct intel_engine_cs *engine, bool stalled)
         */
        rq = execlists_active(execlists);
        if (!rq)
-               return;
+               goto unwind;
 
        ce = rq->hw_context;
        GEM_BUG_ON(i915_active_is_idle(&ce->active));
@@ -2328,6 +2329,7 @@ out_replay:
        intel_ring_update_space(ce->ring);
        __execlists_update_reg_state(ce, engine);
 
+unwind:
        /* Push back any incomplete requests for replay after the reset. */
        __unwind_incomplete_requests(engine);
 }