[PATCH 3/6] drm/i915: No TLB invalidation on wedged or suspended GT

Andi Shyti andi.shyti at linux.intel.com
Thu Oct 5 06:50:45 UTC 2023


From: Jonathan Cavitt <jonathan.cavitt at intel.com>

In case of GT is suspended or wedged, don't allow submission of new TLB
invalidation request and cancel all pending requests. The TLB entries
will be invalidated either during GuC reload or on system resume.

Signed-off-by: Fei Yang <fei.yang at intel.com>
Signed-off-by: Jonathan Cavitt <jonathan.cavitt at intel.com>
CC: John Harrison <john.c.harrison at intel.com>
---
 drivers/gpu/drm/i915/gt/intel_gt.h            |  1 +
 drivers/gpu/drm/i915/gt/intel_tlb.c           | 25 ++++++++++++
 drivers/gpu/drm/i915/gt/intel_tlb.h           |  3 ++
 drivers/gpu/drm/i915/gt/uc/intel_guc.h        |  1 +
 .../gpu/drm/i915/gt/uc/intel_guc_submission.c | 39 +++++++++++++++----
 drivers/gpu/drm/i915/i915_driver.c            |  6 +++
 6 files changed, 68 insertions(+), 7 deletions(-)

diff --git a/drivers/gpu/drm/i915/gt/intel_gt.h b/drivers/gpu/drm/i915/gt/intel_gt.h
index 970bedf6b78a..4e3bb221d2f4 100644
--- a/drivers/gpu/drm/i915/gt/intel_gt.h
+++ b/drivers/gpu/drm/i915/gt/intel_gt.h
@@ -179,4 +179,5 @@ enum i915_map_type intel_gt_coherent_map_type(struct intel_gt *gt,
 void intel_gt_bind_context_set_ready(struct intel_gt *gt);
 void intel_gt_bind_context_set_unready(struct intel_gt *gt);
 bool intel_gt_is_bind_context_ready(struct intel_gt *gt);
+
 #endif /* __INTEL_GT_H__ */
diff --git a/drivers/gpu/drm/i915/gt/intel_tlb.c b/drivers/gpu/drm/i915/gt/intel_tlb.c
index a84563c178bc..c6bb1b34abda 100644
--- a/drivers/gpu/drm/i915/gt/intel_tlb.c
+++ b/drivers/gpu/drm/i915/gt/intel_tlb.c
@@ -166,6 +166,31 @@ void intel_gt_fini_tlb(struct intel_gt *gt)
 	mutex_destroy(&gt->tlb.invalidate_lock);
 }
 
+void intel_gt_tlb_suspend_all(struct drm_i915_private *i915)
+{
+	struct intel_gt *gt;
+	int i;
+
+	if (!HAS_GUC_TLB_INVALIDATION(i915))
+		return;
+	for_each_gt(gt, i915, i)
+		wake_up_all_tlb_invalidate(&gt->uc.guc);
+}
+
+void intel_gt_tlb_resume_all(struct drm_i915_private *i915)
+{
+	struct intel_gt *gt;
+	int i;
+
+	if (!HAS_GUC_TLB_INVALIDATION(i915))
+		return;
+	for_each_gt(gt, i915, i) {
+		/* Perform tlb invalidation on both GT and GuC, in that order. */
+		intel_guc_invalidate_tlb_full(&gt->uc.guc);
+		intel_guc_invalidate_tlb(&gt->uc.guc);
+	}
+}
+
 #if IS_ENABLED(CONFIG_DRM_I915_SELFTEST)
 #include "selftest_tlb.c"
 #endif
diff --git a/drivers/gpu/drm/i915/gt/intel_tlb.h b/drivers/gpu/drm/i915/gt/intel_tlb.h
index 337327af92ac..1a34ec0c447f 100644
--- a/drivers/gpu/drm/i915/gt/intel_tlb.h
+++ b/drivers/gpu/drm/i915/gt/intel_tlb.h
@@ -26,4 +26,7 @@ static inline u32 intel_gt_next_invalidate_tlb_full(const struct intel_gt *gt)
 	return intel_gt_tlb_seqno(gt) | 1;
 }
 
+void intel_gt_tlb_suspend_all(struct drm_i915_private *i915);
+void intel_gt_tlb_resume_all(struct drm_i915_private *i915);
+
 #endif /* INTEL_TLB_H */
diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc.h b/drivers/gpu/drm/i915/gt/uc/intel_guc.h
index d7203ba88b0c..e76b26095eef 100644
--- a/drivers/gpu/drm/i915/gt/uc/intel_guc.h
+++ b/drivers/gpu/drm/i915/gt/uc/intel_guc.h
@@ -536,4 +536,5 @@ void intel_guc_dump_time_info(struct intel_guc *guc, struct drm_printer *p);
 
 int intel_guc_sched_disable_gucid_threshold_max(struct intel_guc *guc);
 
+void wake_up_all_tlb_invalidate(struct intel_guc *guc);
 #endif
diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c b/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
index 7b484d1573f1..7172335199c7 100644
--- a/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
+++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
@@ -32,6 +32,7 @@
 
 #include "i915_drv.h"
 #include "i915_reg.h"
+#include "i915_irq.h"
 #include "i915_trace.h"
 
 /**
@@ -1796,13 +1797,23 @@ static void __guc_reset_context(struct intel_context *ce, intel_engine_mask_t st
 	intel_context_put(parent);
 }
 
-void intel_guc_submission_reset(struct intel_guc *guc, intel_engine_mask_t stalled)
+void wake_up_all_tlb_invalidate(struct intel_guc *guc)
 {
 	struct intel_guc_tlb_wait *wait;
+	unsigned long i;
+
+	if (!HAS_GUC_TLB_INVALIDATION(guc_to_gt(guc)->i915))
+		return;
+	xa_for_each(&guc->tlb_lookup, i, wait) {
+		wake_up(&wait->wq);
+	}
+}
+
+void intel_guc_submission_reset(struct intel_guc *guc, intel_engine_mask_t stalled)
+{
 	struct intel_context *ce;
 	unsigned long index;
 	unsigned long flags;
-	unsigned long i;
 
 	if (unlikely(!guc_submission_initialized(guc))) {
 		/* Reset called during driver load? GuC not yet initialised! */
@@ -1833,9 +1844,7 @@ void intel_guc_submission_reset(struct intel_guc *guc, intel_engine_mask_t stall
 	 * The full GT reset will have cleared the TLB caches and flushed the
 	 * G2H message queue; we can release all the blocked waiters.
 	 */
-	xa_for_each(&guc->tlb_lookup, i, wait) {
-		wake_up(&wait->wq);
-	}
+	wake_up_all_tlb_invalidate(guc);
 }
 
 static void guc_cancel_context_requests(struct intel_context *ce)
@@ -1931,6 +1940,12 @@ void intel_guc_submission_cancel_requests(struct intel_guc *guc)
 
 	/* GuC is blown away, drop all references to contexts */
 	xa_destroy(&guc->context_lookup);
+
+	/*
+	 * Wedged GT won't respond to any TLB invalidation request. Simply
+	 * release all the blocked waiters.
+	 */
+	wake_up_all_tlb_invalidate(guc);
 }
 
 void intel_guc_submission_reset_finish(struct intel_guc *guc)
@@ -4746,6 +4761,14 @@ static long must_wait_woken(struct wait_queue_entry *wq_entry, long timeout)
 	return timeout;
 }
 
+static bool intel_gt_is_enabled(const struct intel_gt *gt)
+{
+	/* Check if GT is wedged or suspended */
+	if (intel_gt_is_wedged(gt) || !intel_irqs_enabled(gt->i915))
+		return false;
+	return true;
+}
+
 static int guc_send_invalidate_tlb(struct intel_guc *guc, u32 type)
 {
 	struct intel_guc_tlb_wait _wq, *wq = &_wq;
@@ -4763,7 +4786,8 @@ static int guc_send_invalidate_tlb(struct intel_guc *guc, u32 type)
 	};
 	u32 size = ARRAY_SIZE(action);
 
-	if (!intel_guc_ct_enabled(&guc->ct))
+	if (!intel_gt_is_enabled(gt) ||
+	    !intel_guc_ct_enabled(&guc->ct))
 		return -EINVAL;
 
 	init_waitqueue_head(&_wq.wq);
@@ -4806,7 +4830,8 @@ static int guc_send_invalidate_tlb(struct intel_guc *guc, u32 type)
 	 * can be queued in CT buffer.
 	 */
 #define OUTSTANDING_GUC_TIMEOUT_PERIOD  (HZ * 2)
-	if (!must_wait_woken(&wait, OUTSTANDING_GUC_TIMEOUT_PERIOD)) {
+	if (intel_gt_is_enabled(gt) &&
+	    !must_wait_woken(&wait, OUTSTANDING_GUC_TIMEOUT_PERIOD)) {
 		gt_err(gt,
 		       "TLB invalidation response timed out for seqno %u\n", seqno);
 		err = -ETIME;
diff --git a/drivers/gpu/drm/i915/i915_driver.c b/drivers/gpu/drm/i915/i915_driver.c
index ccbb2834cde0..0c9d9826d2f4 100644
--- a/drivers/gpu/drm/i915/i915_driver.c
+++ b/drivers/gpu/drm/i915/i915_driver.c
@@ -72,6 +72,7 @@
 #include "gt/intel_gt.h"
 #include "gt/intel_gt_pm.h"
 #include "gt/intel_rc6.h"
+#include "gt/intel_tlb.h"
 
 #include "pxp/intel_pxp.h"
 #include "pxp/intel_pxp_debugfs.h"
@@ -1093,6 +1094,9 @@ static int i915_drm_suspend(struct drm_device *dev)
 	intel_dp_mst_suspend(dev_priv);
 
 	intel_runtime_pm_disable_interrupts(dev_priv);
+
+	intel_gt_tlb_suspend_all(dev_priv);
+
 	intel_hpd_cancel_work(dev_priv);
 
 	intel_suspend_encoders(dev_priv);
@@ -1264,6 +1268,8 @@ static int i915_drm_resume(struct drm_device *dev)
 
 	intel_gvt_resume(dev_priv);
 
+	intel_gt_tlb_resume_all(dev_priv);
+
 	enable_rpm_wakeref_asserts(&dev_priv->runtime_pm);
 
 	return 0;
-- 
2.40.1



More information about the Intel-gfx-trybot mailing list