diff mbox

drm/i915: Actually flush interrupts on reset not just wedging

Message ID 20180323090806.13096-1-chris@chris-wilson.co.uk (mailing list archive)
State New, archived
Headers show

Commit Message

Chris Wilson March 23, 2018, 9:08 a.m. UTC
Commit 0f36a85c3bd5 ("drm/i915: Flush pending interrupt following a GPU
reset") got confused and only applied the flush to the set-wedge path
(which itself is proving troublesome), but we also need the
serialisation on the regular reset path. Oops.

Move the interrupt into reset_irq() and make it common to the reset and
final set-wedge.

References: 0f36a85c3bd5 ("drm/i915: Flush pending interrupt following a GPU reset")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Cc: Michel Thierry <michel.thierry@intel.com>
Cc: MichaƂ Winiarski <michal.winiarski@intel.com>
Cc: Jeff McGee <jeff.mcgee@intel.com>
---
 drivers/gpu/drm/i915/intel_lrc.c | 105 ++++++++++++++++++++-------------------
 1 file changed, 53 insertions(+), 52 deletions(-)

Comments

Chris Wilson March 23, 2018, 12:06 p.m. UTC | #1
Quoting Patchwork (2018-03-23 11:56:41)
> == Series Details ==
> 
> Series: drm/i915: Actually flush interrupts on reset not just wedging (rev2)
> URL   : https://patchwork.freedesktop.org/series/40550/
> State : failure
> 
> == Summary ==
> 
> Series 40550v2 drm/i915: Actually flush interrupts on reset not just wedging
> https://patchwork.freedesktop.org/api/1.0/series/40550/revisions/2/mbox/
> 
> ---- Possible new issues:
> 
> Test kms_pipe_crc_basic:
>         Subgroup nonblocking-crc-pipe-a:
>                 pass       -> INCOMPLETE (fi-elk-e7500)

You call that a failure! I call that an outright success after last
time!
-Chris
diff mbox

Patch

diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
index ce09c5ad334f..2e53882b826b 100644
--- a/drivers/gpu/drm/i915/intel_lrc.c
+++ b/drivers/gpu/drm/i915/intel_lrc.c
@@ -740,6 +740,57 @@  execlists_cancel_port_requests(struct intel_engine_execlists * const execlists)
 	}
 }
 
+static void clear_gtiir(struct intel_engine_cs *engine)
+{
+	static const u8 gtiir[] = {
+		[RCS] = 0,
+		[BCS] = 0,
+		[VCS] = 1,
+		[VCS2] = 1,
+		[VECS] = 3,
+	};
+	struct drm_i915_private *dev_priv = engine->i915;
+	int i;
+
+	/* TODO: correctly reset irqs for gen11 */
+	if (WARN_ON_ONCE(INTEL_GEN(engine->i915) >= 11))
+		return;
+
+	GEM_BUG_ON(engine->id >= ARRAY_SIZE(gtiir));
+
+	/*
+	 * Clear any pending interrupt state.
+	 *
+	 * We do it twice out of paranoia that some of the IIR are
+	 * double buffered, and so if we only reset it once there may
+	 * still be an interrupt pending.
+	 */
+	for (i = 0; i < 2; i++) {
+		I915_WRITE(GEN8_GT_IIR(gtiir[engine->id]),
+			   engine->irq_keep_mask);
+		POSTING_READ(GEN8_GT_IIR(gtiir[engine->id]));
+	}
+	GEM_BUG_ON(I915_READ(GEN8_GT_IIR(gtiir[engine->id])) &
+		   engine->irq_keep_mask);
+}
+
+static void reset_irq(struct intel_engine_cs *engine)
+{
+	/* Mark all CS interrupts as complete */
+	smp_store_mb(engine->execlists.active, 0);
+	synchronize_hardirq(engine->i915->drm.irq);
+
+	clear_gtiir(engine);
+
+	/*
+	 * The port is checked prior to scheduling a tasklet, but
+	 * just in case we have suspended the tasklet to do the
+	 * wedging make sure that when it wakes, it decides there
+	 * is no work to do by clearing the irq_posted bit.
+	 */
+	clear_bit(ENGINE_IRQ_EXECLIST, &engine->irq_posted);
+}
+
 static void execlists_cancel_requests(struct intel_engine_cs *engine)
 {
 	struct intel_engine_execlists * const execlists = &engine->execlists;
@@ -765,6 +816,8 @@  static void execlists_cancel_requests(struct intel_engine_cs *engine)
 	 */
 	local_irq_save(flags);
 
+	reset_irq(engine);
+
 	/* Cancel the requests on the HW and clear the ELSP tracker. */
 	execlists_cancel_port_requests(execlists);
 
@@ -805,18 +858,6 @@  static void execlists_cancel_requests(struct intel_engine_cs *engine)
 
 	spin_unlock(&engine->timeline->lock);
 
-	/* Mark all CS interrupts as complete */
-	smp_store_mb(execlists->active, 0);
-	synchronize_hardirq(engine->i915->drm.irq);
-
-	/*
-	 * The port is checked prior to scheduling a tasklet, but
-	 * just in case we have suspended the tasklet to do the
-	 * wedging make sure that when it wakes, it decides there
-	 * is no work to do by clearing the irq_posted bit.
-	 */
-	clear_bit(ENGINE_IRQ_EXECLIST, &engine->irq_posted);
-
 	local_irq_restore(flags);
 }
 
@@ -1566,14 +1607,6 @@  static int intel_init_workaround_bb(struct intel_engine_cs *engine)
 	return ret;
 }
 
-static u8 gtiir[] = {
-	[RCS] = 0,
-	[BCS] = 0,
-	[VCS] = 1,
-	[VCS2] = 1,
-	[VECS] = 3,
-};
-
 static void enable_execlists(struct intel_engine_cs *engine)
 {
 	struct drm_i915_private *dev_priv = engine->i915;
@@ -1657,35 +1690,6 @@  static int gen9_init_render_ring(struct intel_engine_cs *engine)
 	return init_workarounds_ring(engine);
 }
 
-static void reset_irq(struct intel_engine_cs *engine)
-{
-	struct drm_i915_private *dev_priv = engine->i915;
-	int i;
-
-	/* TODO: correctly reset irqs for gen11 */
-	if (WARN_ON_ONCE(INTEL_GEN(engine->i915) >= 11))
-		return;
-
-	GEM_BUG_ON(engine->id >= ARRAY_SIZE(gtiir));
-
-	/*
-	 * Clear any pending interrupt state.
-	 *
-	 * We do it twice out of paranoia that some of the IIR are double
-	 * buffered, and if we only reset it once there may still be
-	 * an interrupt pending.
-	 */
-	for (i = 0; i < 2; i++) {
-		I915_WRITE(GEN8_GT_IIR(gtiir[engine->id]),
-			   engine->irq_keep_mask);
-		POSTING_READ(GEN8_GT_IIR(gtiir[engine->id]));
-	}
-	GEM_BUG_ON(I915_READ(GEN8_GT_IIR(gtiir[engine->id])) &
-		   engine->irq_keep_mask);
-
-	clear_bit(ENGINE_IRQ_EXECLIST, &engine->irq_posted);
-}
-
 static void reset_common_ring(struct intel_engine_cs *engine,
 			      struct i915_request *request)
 {
@@ -1717,9 +1721,6 @@  static void reset_common_ring(struct intel_engine_cs *engine,
 	__unwind_incomplete_requests(engine);
 	spin_unlock(&engine->timeline->lock);
 
-	/* Mark all CS interrupts as complete */
-	execlists->active = 0;
-
 	local_irq_restore(flags);
 
 	/*