Mali: midgard: add an error handling pass in kbase_mmu_interrupt()

For RK redmine Defect #168230.

Change-Id: I3cd6544dd23b833138e4cc700a8f2cdd627ff592
Signed-off-by: Zhen Chen <chenzhen@rock-chips.com>

4 files changed, 26 insertions, 7 deletions

diff --git a/drivers/gpu/arm/midgard/backend/gpu/mali_kbase_mmu_hw_direct.c b/drivers/gpu/arm/midgard/backend/gpu/mali_kbase_mmu_hw_direct.c
index aa1817c8bca9..ba826184dd3f 100644
--- a/drivers/gpu/arm/midgard/backend/gpu/mali_kbase_mmu_hw_direct.c
+++ b/drivers/gpu/arm/midgard/backend/gpu/mali_kbase_mmu_hw_direct.c
@@ -14,6 +14,8 @@
  */
 
 
+/* #define ENABLE_DEBUG_LOG */
+#include "../../platform/rk/custom_log.h"
 
 #include <linux/bitops.h>
 
@@ -164,6 +166,10 @@ void kbase_mmu_interrupt(struct kbase_device *kbdev, u32 irq_stat)
 		 * the MMU is updated
 		 */
 		kctx = kbasep_js_runpool_lookup_ctx(kbdev, as_no);
+		if (!kctx) {
+			E("fail to lookup ctx, to break out.");
+			break;
+		}
 
 
 		/* find faulting address */
diff --git a/drivers/gpu/arm/midgard/mali_kbase_ctx_sched.c b/drivers/gpu/arm/midgard/mali_kbase_ctx_sched.c
index e2f7baabad43..ce004841403f 100644
--- a/drivers/gpu/arm/midgard/mali_kbase_ctx_sched.c
+++ b/drivers/gpu/arm/midgard/mali_kbase_ctx_sched.c
@@ -131,16 +131,21 @@ int kbase_ctx_sched_retain_ctx(struct kbase_context *kctx)
 	return kctx->as_nr;
 }
 
-void kbase_ctx_sched_retain_ctx_refcount(struct kbase_context *kctx)
+int kbase_ctx_sched_retain_ctx_refcount(struct kbase_context *kctx)
 {
 	struct kbase_device *const kbdev = kctx->kbdev;
 
 	lockdep_assert_held(&kbdev->hwaccess_lock);
 	WARN_ON(atomic_read(&kctx->refcount) == 0);
+	if (atomic_read(&kctx->refcount) == 0)
+		return -1;
+
 	WARN_ON(kctx->as_nr == KBASEP_AS_NR_INVALID);
 	WARN_ON(kbdev->as_to_kctx[kctx->as_nr] != kctx);
 
 	atomic_inc(&kctx->refcount);
+
+	return 0;
 }
 
 void kbase_ctx_sched_release_ctx(struct kbase_context *kctx)
diff --git a/drivers/gpu/arm/midgard/mali_kbase_ctx_sched.h b/drivers/gpu/arm/midgard/mali_kbase_ctx_sched.h
index e55152547c31..47474fecc2a9 100644
--- a/drivers/gpu/arm/midgard/mali_kbase_ctx_sched.h
+++ b/drivers/gpu/arm/midgard/mali_kbase_ctx_sched.h
@@ -87,8 +87,11 @@ int kbase_ctx_sched_retain_ctx(struct kbase_context *kctx);
  * it doesn't get descheduled.
  *
  * The kbase_device::hwaccess_lock must be held whilst calling this function
+ * @return
+ *	若成功, 返回 0;
+ *	若 *kctx 状态异常, 返回 -1.
  */
-void kbase_ctx_sched_retain_ctx_refcount(struct kbase_context *kctx);
+int kbase_ctx_sched_retain_ctx_refcount(struct kbase_context *kctx);
 
 /* kbase_ctx_sched_release_ctx - Release a reference to the @ref kbase_context
  *
diff --git a/drivers/gpu/arm/midgard/mali_kbase_js.c b/drivers/gpu/arm/midgard/mali_kbase_js.c
index b053ce1f22a1..10a1d5909bbe 100644
--- a/drivers/gpu/arm/midgard/mali_kbase_js.c
+++ b/drivers/gpu/arm/midgard/mali_kbase_js.c
@@ -13,9 +13,8 @@
  *
  */
 
-
-
-
+/* #define ENABLE_DEBUG_LOG */
+#include "./platform/rk/custom_log.h"
 
 /*
  * Job Scheduler Implementation
@@ -1245,6 +1244,7 @@ bool kbasep_js_runpool_retain_ctx(struct kbase_device *kbdev,
 struct kbase_context *kbasep_js_runpool_lookup_ctx(struct kbase_device *kbdev,
 		int as_nr)
 {
+	int ret = 0;
 	unsigned long flags;
 	struct kbasep_js_device_data *js_devdata;
 	struct kbase_context *found_kctx = NULL;
@@ -1257,8 +1257,13 @@ struct kbase_context *kbasep_js_runpool_lookup_ctx(struct kbase_device *kbdev,
 
 	found_kctx = kbdev->as_to_kctx[as_nr];
 
-	if (found_kctx != NULL)
-		kbase_ctx_sched_retain_ctx_refcount(found_kctx);
+	if (found_kctx != NULL) {
+		ret = kbase_ctx_sched_retain_ctx_refcount(found_kctx);
+		if (ret != 0) {
+			E("fail to retain ctx_refcount, ret : %d.", ret);
+			found_kctx = NULL;
+		}
+	}
 
 	spin_unlock_irqrestore(&kbdev->hwaccess_lock, flags);