[05/15] drm: drop modeset_lock_all from drm_state_info

Commit Message

Daniel Vetter April 3, 2017, 8:32 a.m. UTC

If we push the locks down we don't have to take them all at the same
time.

Aside: Making dump_info fully safe should be fairly simple, if we
protect the ->state pointers with rcu. Simply putting a
synchronize_rcu() into the drm_atomic_state free function should be
all that's roughly needed. Well except we shouldn't block in there, so
better to put that into a work_struct. But I've not set out to fix
that little issue.

Cc: Rob Clark <[e-mail]>
Signed-off-by: Daniel Vetter <[e-mail]>
---
 drivers/gpu/drm/drm_atomic.c | 60 ++++++++++++++++++++++++++++----------------
 1 file changed, 39 insertions(+), 21 deletions(-)

Patch

diff --git a/drivers/gpu/drm/drm_atomic.c b/drivers/gpu/drm/drm_atomic.c
index 345310213820..9afb14371ce0 100644
--- a/drivers/gpu/drm/drm_atomic.c
+++ b/drivers/gpu/drm/drm_atomic.c
@@ -1676,22 +1676,8 @@  static void drm_atomic_print_state(const struct drm_atomic_state *state)
 		drm_atomic_connector_print_state(&p, connector_state);
 }
 
-/**
- * drm_state_dump - dump entire device atomic state
- * @dev: the drm device
- * @p: where to print the state to
- *
- * Just for debugging.  Drivers might want an option to dump state
- * to dmesg in case of error irq's.  (Hint, you probably want to
- * ratelimit this!)
- *
- * The caller must drm_modeset_lock_all(), or if this is called
- * from error irq handler, it should not be enabled by default.
- * (Ie. if you are debugging errors you might not care that this
- * is racey.  But calling this without all modeset locks held is
- * not inherently safe.)
- */
-void drm_state_dump(struct drm_device *dev, struct drm_printer *p)
+static void __drm_state_dump(struct drm_device *dev, struct drm_printer *p,
+			     bool take_locks)
 {
 	struct drm_mode_config *config = &dev->mode_config;
 	struct drm_plane *plane;
@@ -1702,17 +1688,51 @@  void drm_state_dump(struct drm_device *dev, struct drm_printer *p)
 	if (!drm_core_check_feature(dev, DRIVER_ATOMIC))
 		return;
 
-	list_for_each_entry(plane, &config->plane_list, head)
+	list_for_each_entry(plane, &config->plane_list, head) {
+		if (take_locks)
+			drm_modeset_lock(&plane->mutex, NULL);
 		drm_atomic_plane_print_state(p, plane->state);
+		if (take_locks)
+			drm_modeset_unlock(&plane->mutex);
+	}
 
-	list_for_each_entry(crtc, &config->crtc_list, head)
+	list_for_each_entry(crtc, &config->crtc_list, head) {
+		if (take_locks)
+			drm_modeset_lock(&crtc->mutex, NULL);
 		drm_atomic_crtc_print_state(p, crtc->state);
+		if (take_locks)
+			drm_modeset_unlock(&crtc->mutex);
+	}
 
 	drm_connector_list_iter_begin(dev, &conn_iter);
+	if (take_locks)
+		drm_modeset_lock(&dev->mode_config.connection_mutex, NULL);
 	drm_for_each_connector_iter(connector, &conn_iter)
 		drm_atomic_connector_print_state(p, connector->state);
+	if (take_locks)
+		drm_modeset_unlock(&dev->mode_config.connection_mutex);
 	drm_connector_list_iter_end(&conn_iter);
 }
+
+/**
+ * drm_state_dump - dump entire device atomic state
+ * @dev: the drm device
+ * @p: where to print the state to
+ *
+ * Just for debugging.  Drivers might want an option to dump state
+ * to dmesg in case of error irq's.  (Hint, you probably want to
+ * ratelimit this!)
+ *
+ * The caller must drm_modeset_lock_all(), or if this is called
+ * from error irq handler, it should not be enabled by default.
+ * (Ie. if you are debugging errors you might not care that this
+ * is racey.  But calling this without all modeset locks held is
+ * not inherently safe.)
+ */
+void drm_state_dump(struct drm_device *dev, struct drm_printer *p)
+{
+	__drm_state_dump(dev, p, false);
+}
 EXPORT_SYMBOL(drm_state_dump);
 
 #ifdef CONFIG_DEBUG_FS
@@ -1722,9 +1742,7 @@  static int drm_state_info(struct seq_file *m, void *data)
 	struct drm_device *dev = node->minor->dev;
 	struct drm_printer p = drm_seq_file_printer(m);
 
-	drm_modeset_lock_all(dev);
-	drm_state_dump(dev, &p);
-	drm_modeset_unlock_all(dev);
+	__drm_state_dump(dev, &p, true);
 
 	return 0;
 }