drm: Allow determining if current task is output poll worker
authorLukas Wunner <lukas@wunner.de>
Wed, 14 Feb 2018 05:41:25 +0000 (06:41 +0100)
committerBen Hutchings <ben@decadent.org.uk>
Sat, 16 Jun 2018 21:22:18 +0000 (22:22 +0100)
commit 25c058ccaf2ebbc3e250ec1e199e161f91fe27d4 upstream.

Introduce a helper to determine if the current task is an output poll
worker.

This allows us to fix a long-standing deadlock in several DRM drivers
wherein the ->runtime_suspend callback waits for the output poll worker
to finish and the worker in turn calls a ->detect callback which waits
for runtime suspend to finish.  The ->detect callback is invoked from
multiple call sites and waiting for runtime suspend to finish is the
correct thing to do except if it's executing in the context of the
worker.

v2: Expand kerneldoc to specifically mention deadlock between
    output poll worker and autosuspend worker as use case. (Lyude)

Cc: Dave Airlie <airlied@redhat.com>
Cc: Ben Skeggs <bskeggs@redhat.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
Signed-off-by: Lukas Wunner <lukas@wunner.de>
Link: https://patchwork.freedesktop.org/patch/msgid/3549ce32e7f1467102e70d3e9cbf70c46bfe108e.1518593424.git.lukas@wunner.de
[bwh: Backported to 3.16: adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
drivers/gpu/drm/drm_probe_helper.c
include/drm/drm_crtc_helper.h

index d22676b89cbb337147ec319fe5af4963e5e9f1d8..79bc67cda504f1cb1e9e9d5a2856d528496119b3 100644 (file)
@@ -302,6 +302,26 @@ static void output_poll_execute(struct work_struct *work)
                schedule_delayed_work(delayed_work, DRM_OUTPUT_POLL_PERIOD);
 }
 
+/**
+ * drm_kms_helper_is_poll_worker - is %current task an output poll worker?
+ *
+ * Determine if %current task is an output poll worker.  This can be used
+ * to select distinct code paths for output polling versus other contexts.
+ *
+ * One use case is to avoid a deadlock between the output poll worker and
+ * the autosuspend worker wherein the latter waits for polling to finish
+ * upon calling drm_kms_helper_poll_disable(), while the former waits for
+ * runtime suspend to finish upon calling pm_runtime_get_sync() in a
+ * connector ->detect hook.
+ */
+bool drm_kms_helper_is_poll_worker(void)
+{
+       struct work_struct *work = current_work();
+
+       return work && work->func == output_poll_execute;
+}
+EXPORT_SYMBOL(drm_kms_helper_is_poll_worker);
+
 /**
  * drm_kms_helper_poll_disable - disable output polling
  * @dev: drm_device
index a3d75fefd01020b520be0d9b35958f9462b4aec4..9dbc9f7b77ca590e7df93b04e8b132d79b9fdea2 100644 (file)
@@ -176,5 +176,6 @@ extern void drm_kms_helper_hotplug_event(struct drm_device *dev);
 
 extern void drm_kms_helper_poll_disable(struct drm_device *dev);
 extern void drm_kms_helper_poll_enable(struct drm_device *dev);
+bool drm_kms_helper_is_poll_worker(void);
 
 #endif