[PATCH 4.9 053/164] drm/nouveau/tmr: avoid processing completed alarms when adding a new one

From: Greg Kroah-Hartman
Date: Tue May 23 2017 - 16:27:00 EST


4.9-stable review patch. If anyone has any objections, please let me know.

------------------

From: Ben Skeggs <bskeggs@xxxxxxxxxx>

commit 330bdf62fe6a6c5b99a647f7bf7157107c9348b3 upstream.

The idea here was to avoid having to "manually" program the HW if there's
a new earliest alarm. This was lazy and bad, as it leads to loads of fun
races between inter-related callers (ie. therm).

Turns out, it's not so difficult after all. Go figure ;)

Signed-off-by: Ben Skeggs <bskeggs@xxxxxxxxxx>
Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>

---
drivers/gpu/drm/nouveau/nvkm/subdev/timer/base.c | 16 +++++++++++++---
1 file changed, 13 insertions(+), 3 deletions(-)

--- a/drivers/gpu/drm/nouveau/nvkm/subdev/timer/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/timer/base.c
@@ -80,12 +80,22 @@ nvkm_timer_alarm(struct nvkm_timer *tmr,
if (list->timestamp > alarm->timestamp)
break;
}
+
list_add_tail(&alarm->head, &list->head);
+
+ /* Update HW if this is now the earliest alarm. */
+ list = list_first_entry(&tmr->alarms, typeof(*list), head);
+ if (list == alarm) {
+ tmr->func->alarm_init(tmr, alarm->timestamp);
+ /* This shouldn't happen if callers aren't stupid.
+ *
+ * Worst case scenario is that it'll take roughly
+ * 4 seconds for the next alarm to trigger.
+ */
+ WARN_ON(alarm->timestamp <= nvkm_timer_read(tmr));
+ }
}
spin_unlock_irqrestore(&tmr->lock, flags);
-
- /* process pending alarms */
- nvkm_timer_alarm_trigger(tmr);
}

void