[Intel-xe] [PATCH v3 6/6] drm/xe: Use atomic for mem_access.ref

[Matthew Auld]

From: Maarten Lankhorst <maarten.lankhorst at linux.intel.com>

xe_guc_ct_fast_path() is called from an irq context, and cannot lock
the mutex used by xe_device_mem_access_ongoing().

Fortunately it is easy to fix if we switch the ref over to an atomic.

Signed-off-by: Maarten Lankhorst <maarten.lankhorst at linux.intel.com>
Signed-off-by: Matthew Auld <matthew.auld at intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi at intel.com>
Cc: Thomas Hellström <thomas.hellstrom at linux.intel.com>
Cc: Matthew Brost <matthew.brost at intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi at intel.com>
---
 drivers/gpu/drm/xe/xe_device.c       | 15 ++++++++++-----
 drivers/gpu/drm/xe/xe_device.h       | 19 +++++--------------
 drivers/gpu/drm/xe/xe_device_types.h |  4 ++--
 3 files changed, 17 insertions(+), 21 deletions(-)

diff --git a/drivers/gpu/drm/xe/xe_device.c b/drivers/gpu/drm/xe/xe_device.c
index 123343bc33b0..9509565a28d7 100644
--- a/drivers/gpu/drm/xe/xe_device.c
+++ b/drivers/gpu/drm/xe/xe_device.c
@@ -405,6 +405,8 @@ u32 xe_device_ccs_bytes(struct xe_device *xe, u64 size)
 
 void xe_device_mem_access_get(struct xe_device *xe)
 {
+	int ref;
+
 	/*
 	 * This looks racy, but should be fine since the pm_callback_task only
 	 * transitions from NULL -> current (and back to NULL again), during the
@@ -424,23 +426,26 @@ void xe_device_mem_access_get(struct xe_device *xe)
 	 * for example just on the 0 -> 1.
 	 */
 	mutex_lock(&xe->mem_access.lock);
-	if (xe->mem_access.ref == 0)
+	if (atomic_read(&xe->mem_access.ref) == 0)
 		xe->mem_access.hold_rpm = xe_pm_runtime_resume_and_get(xe);
-	xe->mem_access.ref++;
+	ref = atomic_inc_return(&xe->mem_access.ref);
 	mutex_unlock(&xe->mem_access.lock);
 
-	XE_WARN_ON(xe->mem_access.ref == S32_MAX);
+	XE_WARN_ON(ref == S32_MAX);
 }
 
 void xe_device_mem_access_put(struct xe_device *xe)
 {
+	int ref;
+
 	if (READ_ONCE(xe->pm_callback_task) == current)
 		return;
 
 	mutex_lock(&xe->mem_access.lock);
-	if (--xe->mem_access.ref == 0 && xe->mem_access.hold_rpm)
+	ref = atomic_dec_return(&xe->mem_access.ref);
+	if (ref == 0 && xe->mem_access.hold_rpm)
 		xe_pm_runtime_put(xe);
 	mutex_unlock(&xe->mem_access.lock);
 
-	XE_WARN_ON(xe->mem_access.ref < 0);
+	XE_WARN_ON(ref < 0);
 }
diff --git a/drivers/gpu/drm/xe/xe_device.h b/drivers/gpu/drm/xe/xe_device.h
index 7a4ba5212d9c..4aab3658e0ae 100644
--- a/drivers/gpu/drm/xe/xe_device.h
+++ b/drivers/gpu/drm/xe/xe_device.h
@@ -90,26 +90,17 @@ static inline struct xe_force_wake * gt_to_fw(struct xe_gt *gt)
 void xe_device_mem_access_get(struct xe_device *xe);
 void xe_device_mem_access_put(struct xe_device *xe);
 
-static inline void xe_device_assert_mem_access(struct xe_device *xe)
-{
-	if (READ_ONCE(xe->pm_callback_task) == current)
-		return;
-
-	XE_WARN_ON(!xe->mem_access.ref);
-}
-
 static inline bool xe_device_mem_access_ongoing(struct xe_device *xe)
 {
-	bool ret;
-
 	if (READ_ONCE(xe->pm_callback_task) == current)
 		return true;
 
-	mutex_lock(&xe->mem_access.lock);
-	ret = xe->mem_access.ref;
-	mutex_unlock(&xe->mem_access.lock);
+	return atomic_read(&xe->mem_access.ref);
+}
 
-	return ret;
+static inline void xe_device_assert_mem_access(struct xe_device *xe)
+{
+	XE_WARN_ON(!xe_device_mem_access_ongoing(xe));
 }
 
 static inline bool xe_device_in_fault_mode(struct xe_device *xe)
diff --git a/drivers/gpu/drm/xe/xe_device_types.h b/drivers/gpu/drm/xe/xe_device_types.h
index 0f4f00fa4d18..c291fe66b1c1 100644
--- a/drivers/gpu/drm/xe/xe_device_types.h
+++ b/drivers/gpu/drm/xe/xe_device_types.h
@@ -258,11 +258,11 @@ struct xe_device {
 	struct {
 		/**
 		 * @lock: Serialize xe_device_mem_access users,
-		 * and protect the below internal state, like @ref.
+		 * and protect the below internal state.
 		 */
 		struct mutex lock;
 		/** @ref: ref count of memory accesses */
-		s32 ref;
+		atomic_t ref;
 		/** @hold_rpm: need to put rpm ref back at the end */
 		bool hold_rpm;
 	} mem_access;
-- 
2.40.1