Scheduling while atomic errors on swsusp resume

Scheduling while atomic errors on swsusp resume

[Matthew Garrett]

Using the current Ubuntu development kernel (2.6.10 with acpi and swsusp
stuff backported from 2.6.11), a user is getting the following trace on
resume. Passing noapic nolapic removes the APIC error, but the rest of
the trace is identical. This is reproducible, but only seems to happen
on this machine. Anyone have any idea what's going on before I head off
to try getting it reproduced with a stock kernel?

Stopping tasks:
=========================================================|
Freeing memory...  \b-\b\\b|\b/\b-\b\\b|\b/\b-\b\\b|\b/\b-\b\\b|\b/\b-\b\\b|\b/\b-\b\\b|\b/\b-\b\\b|\b/\b-\b\\b|\b/\b-\b\\b|\b/\b-\b\\b|\b/\b-\b\\b|\b/\b-\b\\b|\b/\b-\b\\b|\b/\b-\b\\b|\bdone (21866 pages freed)
.........................................swsusp: Need to copy 19917 pages
.swsusp: Restoring Highmem
APIC error on CPU0: 00(00)
ACPI: PCI interrupt 0000:00:11.1[A]: no GSI
scheduling while atomic: hibernate.sh/0x00000002/6955
 [<c029d72d>] schedule+0x52d/0x540
 [<f886c5e0>] task_no_data_intr+0x0/0xa0 [ide_core]
 [<c029d808>] wait_for_completion+0x78/0xd0
 [<c01161e0>] default_wake_function+0x0/0x20
 [<c01161e0>] default_wake_function+0x0/0x20
 [<c021a5f8>] __elv_add_request+0x78/0xc0
 [<f8868049>] ide_do_drive_cmd+0xf9/0x170 [ide_core]
 [<f8865343>] generic_ide_resume+0x93/0xc0 [ide_core]
 [<c0219d67>] resume_device+0x27/0x30
 [<c0219e14>] dpm_resume+0xa4/0xb0
 [<c0219e31>] device_resume+0x11/0x20
 [<c0135712>] finish+0x12/0x60
 [<c0135841>] pm_suspend_disk+0x41/0x80
 [<c01334dc>] enter_state+0x6c/0x70
 [<c0133620>] state_store+0xa0/0xa8
 [<c019563a>] subsys_attr_store+0x3a/0x40
 [<c01958de>] flush_write_buffer+0x3e/0x50
 [<c019595f>] sysfs_write_file+0x6f/0x80
 [<c01958f0>] sysfs_write_file+0x0/0x80
 [<c015831f>] vfs_write+0xbf/0x150
 [<c0158481>] sys_write+0x51/0x80
 [<c01030e9>] sysenter_past_esp+0x52/0x75
Restarting tasks...<3>scheduling while atomic: hibernate.sh/0x00000002/6955
 [<c029d72d>] schedule+0x52d/0x540
 [<c011593e>] wake_up_process+0x1e/0x20
 [<c0133944>] thaw_processes+0xa4/0xe0
 [<c0135720>] finish+0x20/0x60
 [<c0135841>] pm_suspend_disk+0x41/0x80
 [<c01334dc>] enter_state+0x6c/0x70
 [<c0133620>] state_store+0xa0/0xa8
 [<c019563a>] subsys_attr_store+0x3a/0x40
 [<c01958de>] flush_write_buffer+0x3e/0x50
 [<c019595f>] sysfs_write_file+0x6f/0x80
 [<c01958f0>] sysfs_write_file+0x0/0x80
 [<c015831f>] vfs_write+0xbf/0x150
 [<c0158481>] sys_write+0x51/0x80
 [<c01030e9>] sysenter_past_esp+0x52/0x75
 done
scheduling while atomic: hibernate.sh/0x00000001/6955
 [<c029d72d>] schedule+0x52d/0x540
 [<c0158481>] sys_write+0x51/0x80
 [<c0103166>] work_resched+0x5/0x16
scheduling while atomic: hibernate.sh/0x00000001/6955
 [<c029d72d>] schedule+0x52d/0x540
 [<c0116cd3>] sys_sched_yield+0x53/0x70
 [<c0164358>] coredump_wait+0x38/0xa0
 [<c0115904>] try_to_wake_up+0xa4/0xc0
 [<c016448d>] do_coredump+0xcd/0x1d6
 [<c01c93e4>] vgacon_scroll+0x144/0x230
 [<c0122dcf>] free_uid+0x1f/0x80
 [<c01236b5>] __dequeue_signal+0xe5/0x1a0
 [<c01237a5>] dequeue_signal+0x35/0x90
 [<c012547d>] get_signal_to_deliver+0x20d/0x300
 [<c0102f3d>] do_signal+0x9d/0x130
 [<c012ad6e>] __kernel_text_address+0x2e/0x40
 [<c012ad6e>] __kernel_text_address+0x2e/0x40
 [<c01156af>] recalc_task_prio+0x8f/0x190
 [<c029d4f1>] schedule+0x2f1/0x540
 [<c01142b0>] do_page_fault+0x0/0x5c7
 [<c0103007>] do_notify_resume+0x37/0x3c
 [<c010318a>] work_notifysig+0x13/0x15
note: hibernate.sh[6955] exited with preempt_count 1
-- 
Matthew Garrett | mjg59@srcf.ucam.org

Re: Scheduling while atomic errors on swsusp resume

[Pavel Machek]

Hi!

> Using the current Ubuntu development kernel (2.6.10 with acpi and swsusp
> stuff backported from 2.6.11), a user is getting the following trace on
> resume. Passing noapic nolapic removes the APIC error, but the rest of
> the trace is identical. This is reproducible, but only seems to happen
> on this machine. Anyone have any idea what's going on before I head off
> to try getting it reproduced with a stock kernel?

Well, those are warnings, so it still works, right? Aha, "exited with
preempt count 1" seems very wrong. Yes, please try this with
vanilla. I'm running 2.6.11 with 

CONFIG_PREEMPT=y
CONFIG_PREEMPT_BKL=y

and I certainly do not get these ugly warnings.
								Pavel

-- 
People were complaining that M$ turns users into beta-testers...
...jr ghea gurz vagb qrirybcref, naq gurl frrz gb yvxr vg gung jnl!

Re: Scheduling while atomic errors on swsusp resume

[Matthew Garrett]

On Fri, 2005-03-04 at 12:26 +0100, Pavel Machek wrote:

> Well, those are warnings, so it still works, right? Aha, "exited with
> preempt count 1" seems very wrong. Yes, please try this with
> vanilla. I'm running 2.6.11 with 

Yeah, the resume script crashes, which is a bit of a problem. I'll get
the user to try with a vanilla kernel, but I'm not having these problems
with an identical kernel - it seems to be something specific to his
setup.

-- 
Matthew Garrett | mjg59@srcf.ucam.org

Re: Scheduling while atomic errors on swsusp resume

[Pavel Machek]

Hi!

> > Well, those are warnings, so it still works, right? Aha, "exited with
> > preempt count 1" seems very wrong. Yes, please try this with
> > vanilla. I'm running 2.6.11 with 
> 
> Yeah, the resume script crashes, which is a bit of a problem. I'll get
> the user to try with a vanilla kernel, but I'm not having these problems
> with an identical kernel - it seems to be something specific to his
> setup.

Make him try it with minimal list of modules, then.
								Pavel
-- 
People were complaining that M$ turns users into beta-testers...
...jr ghea gurz vagb qrirybcref, naq gurl frrz gb yvxr vg gung jnl!