Re: [REGRESSION]: hibernate/sleep regression w/ bisection

On Tue, Nov 01, 2011 at 07:48:19AM -0500, Andrew Watts wrote:

Hi.

Hibernate/sleep (echo disk/mem > /sys/power/state) has presented problems
for me starting with 2.6.39. Kernel 2.6.37.6 was the last completely bug-free
version I used (I skipped the 2.6.38 branch entirely).

The symptoms are that upon resume (from sleep/hibernate) there is no video
nor any keyboard input with the exception of sysrq.

It has been a frustrating bug to hunt down because it is not easily
reproduced; sometimes the bug doesn't pop up until after a long sequence of
hibernate/sleep cycles.

I successfully bisected the problem to: 8ee294cd9def000.

=======
Commit: 8ee294cd9def0004887da7f44b80563493b0a097
Author: Dmitry Torokhov [off-list ref]
Date:   Mon Nov 15 01:39:57 2010 -0800
Input: serio - convert to common workqueue instead of a thread
=======

Backing out 8ee294cd9def000 (which requires reversing part of
1d64b655dc083df also) fixes this particular problem on 2.6.39.4,
3.0.8, and 3.1.

Unfortunately, in 3.0.8 and 3.1 I have other suspend/hibernate problems
that I will investigate next and detail in a different regression report.

Does the following patch fix the problem?

Thanks.

diff --git a/drivers/input/serio/serio.c b/drivers/input/serio/serio.c
index ba70058..95aeebd 100644
--- a/drivers/input/serio/serio.c
+++ b/drivers/input/serio/serio.c

@@ -949,6 +949,7 @@ static int serio_resume(struct device *dev)
 	 * deal with it.
 	 */
 	serio_queue_event(serio, NULL, SERIO_RECONNECT_PORT);
+	queue_work(system_long_wq, &serio_event_work);
 
 	return 0;
 }