drm/xe: Use atomic instead of mutex for xe_device_mem_access_ongoing

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, and the atomic guarantees are good enough
to ensure xe->mem_access.hold_rpm is set before last ref is dropped.

As far as I can tell, the runtime ref in device access should be
killable, but don't dare to do it yet.

Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Acked-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

3 files changed, 13 insertions, 23 deletions

diff --git a/drivers/gpu/drm/xe/xe_device.c b/drivers/gpu/drm/xe/xe_device.c
index 49ce11fc1174..ffacf80c8942 100644
--- a/drivers/gpu/drm/xe/xe_device.c
+++ b/drivers/gpu/drm/xe/xe_device.c
@@ -206,8 +206,6 @@ struct xe_device *xe_device_create(struct pci_dev *pdev,
 	if (err)
 		goto err_put;
 
-	drmm_mutex_init(&xe->drm, &xe->mem_access.lock);
-
 	return xe;
 
 err_put:
@@ -354,25 +352,25 @@ u32 xe_device_ccs_bytes(struct xe_device *xe, u64 size)
 void xe_device_mem_access_get(struct xe_device *xe)
 {
 	bool resumed = xe_pm_runtime_resume_if_suspended(xe);
+	int ref = atomic_inc_return(&xe->mem_access.ref);
 
-	mutex_lock(&xe->mem_access.lock);
-	if (xe->mem_access.ref++ == 0)
+	if (ref == 1)
 		xe->mem_access.hold_rpm = xe_pm_runtime_get_if_active(xe);
-	mutex_unlock(&xe->mem_access.lock);
 
 	/* The usage counter increased if device was immediately resumed */
 	if (resumed)
 		xe_pm_runtime_put(xe);
 
-	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)
 {
-	mutex_lock(&xe->mem_access.lock);
-	if (--xe->mem_access.ref == 0 && xe->mem_access.hold_rpm)
+	bool hold = xe->mem_access.hold_rpm;
+	int ref = atomic_dec_return(&xe->mem_access.ref);
+
+	if (!ref && hold)
 		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 25c5087f5aad..d277f8985f7b 100644
--- a/drivers/gpu/drm/xe/xe_device.h
+++ b/drivers/gpu/drm/xe/xe_device.h
@@ -90,20 +90,14 @@ 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)
+static inline bool xe_device_mem_access_ongoing(struct xe_device *xe)
 {
-	XE_WARN_ON(!xe->mem_access.ref);
+	return atomic_read(&xe->mem_access.ref);
 }
 
-static inline bool xe_device_mem_access_ongoing(struct xe_device *xe)
+static inline void xe_device_assert_mem_access(struct xe_device *xe)
 {
-	bool ret;
-
-	mutex_lock(&xe->mem_access.lock);
-	ret = xe->mem_access.ref;
-	mutex_unlock(&xe->mem_access.lock);
-
-	return ret;
+	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 377a8979bc06..3917b9152eb9 100644
--- a/drivers/gpu/drm/xe/xe_device_types.h
+++ b/drivers/gpu/drm/xe/xe_device_types.h
@@ -184,10 +184,8 @@ struct xe_device {
 	 * triggering additional actions when they occur.
 	 */
 	struct {
-		/** @lock: protect the ref count */
-		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;