[Intel-xe] [PATCH v9 8/8] drm/xe: Use atomic for mem_access.ref

[Matthew Auld <matthew.auld@intel.com>]

From: Maarten Lankhorst <maarten.lankhorst@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@linux.intel.com>
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
---
 drivers/gpu/drm/xe/xe_device.c       | 34 ++++++++++++----------------
 drivers/gpu/drm/xe/xe_device_types.h |  6 ++---
 2 files changed, 18 insertions(+), 22 deletions(-)

diff --git a/drivers/gpu/drm/xe/xe_device.c b/drivers/gpu/drm/xe/xe_device.c
index ac11cd7c3b67..b69fefbac845 100644
--- a/drivers/gpu/drm/xe/xe_device.c
+++ b/drivers/gpu/drm/xe/xe_device.c
@@ -412,30 +412,23 @@ u32 xe_device_ccs_bytes(struct xe_device *xe, u64 size)
 		DIV_ROUND_UP(size, NUM_BYTES_PER_CCS_BYTE) : 0;
 }
 
-void xe_device_assert_mem_access(struct xe_device *xe)
-{
-	if (xe_pm_read_callback_task(xe) == current)
-		return;
-
-	XE_WARN_ON(!xe->mem_access.ref);
-}
-
 bool xe_device_mem_access_ongoing(struct xe_device *xe)
 {
-	bool ret;
-
 	if (xe_pm_read_callback_task(xe) == 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;
+void xe_device_assert_mem_access(struct xe_device *xe)
+{
+	XE_WARN_ON(!xe_device_mem_access_ongoing(xe));
 }
 
 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
@@ -455,23 +448,26 @@ 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 (xe_pm_read_callback_task(xe) == 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_types.h b/drivers/gpu/drm/xe/xe_device_types.h
index b5a56e112c0a..f1515aff8437 100644
--- a/drivers/gpu/drm/xe/xe_device_types.h
+++ b/drivers/gpu/drm/xe/xe_device_types.h
@@ -261,9 +261,9 @@ struct xe_device {
 	 */
 	struct {
 		/**
-		 * @lock: Protects @ref and serializes multiple callers.
+		 * @lock: Serializes multiple callers.
 		 *
-		 * Also used as lockdep aid to hopefully ensure that it sees
+		 * But mainly used as lockdep aid to hopefully ensure that it sees
 		 * that the runtime_pm callbacks can be called while holding
 		 * @lock, and therefore any locks that the caller is holding
 		 * also have a clear dependency on the @lock and any locks the
@@ -274,7 +274,7 @@ struct xe_device {
 		 */
 		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