Bug 5945 - "scheduling while atomic" during wakeup from swsusp
Summary: "scheduling while atomic" during wakeup from swsusp
Alias: None
Product: Power Management
Classification: Unclassified
Component: Other (show other bugs)
Hardware: i386 Linux
: P2 normal
Assignee: power-management_other
Depends on:
Reported: 2006-01-23 10:53 UTC by mrsteven
Modified: 2007-04-28 16:00 UTC (History)
6 users (show)

See Also:
Kernel Version:
Tree: Mainline
Regression: ---

syslog log file of the session (40.72 KB, text/plain)
2006-01-23 11:00 UTC, mrsteven
lspci -vvv (7.91 KB, text/plain)
2006-01-23 11:03 UTC, mrsteven
hibernate.conf (4.95 KB, text/plain)
2006-01-23 11:08 UTC, mrsteven
lsmod (1.23 KB, text/plain)
2006-01-23 11:09 UTC, mrsteven
.config (27.11 KB, text/plain)
2006-01-23 11:14 UTC, mrsteven
syslog log file of the session with pavel's patch (40.38 KB, text/plain)
2006-02-07 03:06 UTC, mrsteven
dmesg of another session with pavel's patch (14.90 KB, text/plain)
2006-02-07 04:57 UTC, mrsteven

Description mrsteven 2006-01-23 10:53:45 UTC
Most recent kernel where this bug did not occur: ???  
Distribution: Gentoo Linux with vanilla-sources-  
Hardware Environment: AMD Athlon 950MHz 512MB RAM, Logitech Optical PS2/USB  
Mouse on PS2  
Problem Description:  
When I use swsusp for suspend to disk the system suspends normally, but upon  
resume I get warnings about "scheduling while atomic" (see attachments for 
details). After that, the mouse  
driver gets confused, i.e. the cursor jumps wildly on the screen and generates  
random mouse click events when I move the mouse. I think everything else works  
fine after resuming.  
Steps to reproduce:  
Use swsusp and then resume.
Comment 1 mrsteven 2006-01-23 11:00:40 UTC
Created attachment 7104 [details]
syslog log file of the session

This is the /var/log/messages file created by syslog-ng.
Comment 2 mrsteven 2006-01-23 11:03:17 UTC
Created attachment 7105 [details]
lspci -vvv

Output of lspci -vvv
Comment 3 mrsteven 2006-01-23 11:08:01 UTC
Created attachment 7106 [details]

/etc/hibernate/hibernate.conf for hibernate-script-1.12
Comment 4 mrsteven 2006-01-23 11:09:39 UTC
Created attachment 7107 [details]

lsmod output
Comment 5 mrsteven 2006-01-23 11:14:45 UTC
Created attachment 7108 [details]

The config file of the kernel.

If more information is needed, feel free to ask... ;-)
Comment 6 Pavel Machek 2006-01-23 14:19:51 UTC
This is input subsystem in both cases, AFAICS. Scheduling in atomic happens on
something called from PS/2 code.
Comment 7 Dmitry Torokhov 2006-01-23 14:40:25 UTC
Pavel, did you make device_resume atomic again?
Comment 8 Dmitry Torokhov 2006-01-23 16:21:47 UTC
Please note that it is not only input layer but also IDE... Generally speaking 
anything upder device_resume() should be allowed to sleep and schedule. 
Comment 9 mrsteven 2006-01-24 08:53:19 UTC
I think it has something to do with the mouse on PS/2. On my laptop the whole 
thing works fine, as I've just tested now. It has a synaptics touchpad and I've 
plugged in exactly the same mouse at USB, of course without the USB to PS/2 
Comment 10 Anonymous Emailer 2006-01-24 11:34:38 UTC
Reply-To: pavel@ucw.cz

Can you apply this and see if it triggers?

diff --git a/drivers/base/power/resume.c b/drivers/base/power/resume.c
index 0a7aa07..d9c9d78 100644
--- a/drivers/base/power/resume.c
+++ b/drivers/base/power/resume.c
@@ -72,6 +72,7 @@ void dpm_resume(void)
 void device_resume(void)
+	might_sleep();
diff --git a/drivers/base/power/suspend.c b/drivers/base/power/suspend.c
index 5050176..815e2f1 100644
--- a/drivers/base/power/suspend.c
+++ b/drivers/base/power/suspend.c
@@ -83,6 +83,7 @@ int device_suspend(pm_message_t state)
 	int error = 0;
+	might_sleep();
 	while (!list_empty(&dpm_active) && error == 0) {

Comment 11 mrsteven 2006-02-07 03:06:48 UTC
Created attachment 7264 [details]
syslog log file of the session with pavel's patch

Suspending with pavel's patch.
Comment 12 mrsteven 2006-02-07 03:11:38 UTC
The mouse kept in sync this time, but I didn't test very long. 
What I forgot to say (that might help): I tried swsusp with kernel 2.6.14 on 
the same machine some time ago and I also had the same problem. 
Comment 13 Pavel Machek 2006-02-07 03:13:53 UTC
Feb  7 11:51:11 athlon950 ACPI: PCI interrupt for device 0000:00:07.2 disabled
Feb  7 11:52:11 athlon950 Stopping tasks: 75a>] ps2_sendbyte+0x12a/0x150
Feb  7 11:52:11 athlon950 [<c012e0e0>] autoremove_wake_function+0x0/0x60
Feb  7 11:52:11 athlon950 [<c024da2c>] ps2_command+0xdc/0x340
Feb  7 11:52:11 athlon950 [<c024da01>] ps2_command+0xb1/0x340

..unfortunately log is truncated at the most interresting place :-(
Comment 14 mrsteven 2006-02-07 03:37:37 UTC
That must be a problem with your browser (works fine with konqueror). The last 
line in the logfile is:  
"Feb  7 11:55:25 athlon950 syslog-ng[6108]: syslog-ng version 1.6.9 going down"  
But I could send you this log via e-mail if you want.  
Comment 15 Anonymous Emailer 2006-02-07 03:40:54 UTC
Reply-To: pavel@ucw.cz

> ------- Additional Comments From mrsteven@gmx.de  2006-02-07 03:37 -------
> That must be a problem with your browser (works fine with konqueror). The last 
> line in the logfile is:  
> "Feb  7 11:55:25 athlon950 syslog-ng[6108]: syslog-ng version 1.6.9 going down"  
> But I could send you this log via e-mail if you want.  

No, problem is in the middle. Part just after Freezing processes: is


Comment 16 mrsteven 2006-02-07 04:57:44 UTC
Created attachment 7265 [details]
dmesg of another session with pavel's patch

Here's a dmesg output of a session with pavel's patch. As you see, there's not
enough space in the dmesg buffer to get all messages.
Comment 17 Pavel Machek 2006-07-11 01:37:10 UTC
Does it still happen in 2.6.18-rc1? Dmitry, can you take a look what is going on?
Comment 18 Pavel Machek 2006-09-27 01:41:40 UTC
If it still happens with 2.6.18, reopen.
Comment 19 mrsteven 2007-04-28 14:32:49 UTC
Works perfectly here since months with newer kernels (2.6.18 and up), so I 
mark this as closed!
Comment 20 Rafael J. Wysocki 2007-04-28 16:00:35 UTC

Note You need to log in before you can comment on or make changes to this bug.