Skip to contentSkip to conversation

git-acpi breaks resume-from-ram on the Vaio

DISCUSSION Andrew Morton 5 participants 14 messages Sep 18, 2007 – Sep 19, 2007 ACPI

lore
Share & tools
lore
#0 lore raw

It suspends OK, then during resume it gets partway through it, then everything just stops.

Would prefer not to have to bisect this one - I've done enough bisecting this week to last a lifetime, and bisecting git-acpi is painful due to build bustage at various points (cpuidle).

#1 lore raw
Quoted 7 lines · On Mon, 17 Sep 2007 21:12:13 -0700 Andrew Morton <[email protected]>…

It suspends OK, then during resume it gets partway through it, then everything just stops.

Would prefer not to have to bisect this one - I've done enough bisecting this week to last a lifetime, and bisecting git-acpi is painful due to build bustage at various points (cpuidle).

And git-acpi breaks suspend-to-disk as well. It gets up to "Suspending console(s)" and then the cursor stops blinking at it wedges up.

#2 lore raw
Quoted 6 lines · Andrew Morton
Quoted 4 lines · On Mon, 17 Sep 2007 21:12:13 -0700 Andrew Morton <[email protected]>…

It suspends OK, then during resume it gets partway through it, then everything just stops.

Quoted 6 lines
Quoted 3 lines

Would prefer not to have to bisect this one - I've done enough bisecting this week to last a lifetime, and bisecting git-acpi is painful due to build bustage at various points (cpuidle).

And git-acpi breaks suspend-to-disk as well. It gets up to "Suspending console(s)" and then the cursor stops blinking at it wedges up.

Can you please compile with CONFIG_DISABLE_CONSOLE_SUSPEND set and try:

# echo test > /sys/power/disk
# echo disk > /sys/power/state

and if that hangs,

# echo testproc > /sys/power/disk
# echo disk > /sys/power/state
#3 lore raw
Quoted 7 lines · Rafael J. Wysocki
Quoted 6 lines · Andrew Morton
Quoted 4 lines · On Mon, 17 Sep 2007 21:12:13 -0700 Andrew Morton <[email protected]>…

It suspends OK, then during resume it gets partway through it, then everything just stops.

Quoted 8 lines
Quoted 6 lines
Quoted 3 lines

Would prefer not to have to bisect this one - I've done enough bisecting this week to last a lifetime, and bisecting git-acpi is painful due to build bustage at various points (cpuidle).

And git-acpi breaks suspend-to-disk as well. It gets up to "Suspending console(s)" and then the cursor stops blinking at it wedges up.

Can you please compile with CONFIG_DISABLE_CONSOLE_SUSPEND set and try:

Ah, that's -mm, sorry. Instead of setting CONFIG_DISABLE_CONSOLE_SUSPEND (which has been removed) please pass no_console_suspend in the command line.

Quoted 7 lines
# echo test > /sys/power/disk
# echo disk > /sys/power/state

and if that hangs,

# echo testproc > /sys/power/disk
# echo disk > /sys/power/state
#4 lore raw
Quoted 8 lines · On Tue, 18 Sep 2007 13:59:04 +0200 "Rafael J. Wysocki" <[email protected]> wrote:
Quoted 7 lines · Rafael J. Wysocki
Quoted 6 lines · Andrew Morton
Quoted 4 lines · On Mon, 17 Sep 2007 21:12:13 -0700 Andrew Morton <[email protected]>…

It suspends OK, then during resume it gets partway through it, then everything just stops.

Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 3 lines

Would prefer not to have to bisect this one - I've done enough bisecting this week to last a lifetime, and bisecting git-acpi is painful due to build bustage at various points (cpuidle).

And git-acpi breaks suspend-to-disk as well. It gets up to "Suspending console(s)" and then the cursor stops blinking at it wedges up.

Bisection shows that the resume-from-ram failure is caused by

commit 987196fa82d4db52c407e8c9d5dec884ba602183

Author: Venkatesh Pallipadi <[email protected]>
Date:   Thu Feb 22 13:52:57 2007 -0800

cpuidle take2: Core cpuidle infrastructure

Note that this is the patch which *fixed* resume-from-RAM prior to Thomas's git-hrt merge. Now it breaks it!?!?!

Also, I'm seeing this, in mainline, when I do echo mem > /sys/power/state

[   26.316685] ipw2200: Failed to send WEP_KEY: Aborted due to RF kill switch.
[   33.044631] ipw2200: Failed to send WEP_KEY: Command timed out.
[   39.862821] ipw2200: Failed to send WEP_KEY: Command timed out.
[   68.962896] PM: Preparing system for mem sleep
[   68.979485] Stopping tasks ... WARNING: at kernel/lockdep.c:2658 check_flags()
[   68.980154]  [<c0104ea4>] show_trace_log_lvl+0x1a/0x2f
[   68.980295]  [<c0105a44>] show_trace+0x12/0x14
[   68.980416]  [<c0105a5b>] dump_stack+0x15/0x17
[   68.980535]  [<c0137276>] check_flags+0x93/0x13d
[   68.980663]  [<c013a8ab>] lock_acquire+0x3a/0x91
[   68.980789]  [<c031948b>] _spin_lock+0x38/0x62
[   68.980913]  [<c0142a4d>] refrigerator+0x13/0xc2
[   68.981040]  [<c01283a2>] get_signal_to_deliver+0x32/0x405
[   68.981188]  [<c01035f4>] do_notify_resume+0x91/0x69f
[   68.981323]  [<c010402d>] work_notifysig+0x13/0x1a
[   68.981453]  =======================
[   68.981542] irq event stamp: 1511
[   68.981624] hardirqs last  enabled at (1511): [<c010408d>] syscall_exit_work+0x11/0x26
[   68.981834] hardirqs last disabled at (1510): [<c0103f63>] syscall_exit+0x9/0x1a
[   68.982031] softirqs last  enabled at (1418): [<c03104c3>] unix_accept+0xe5/0xfb
[   68.982230] softirqs last disabled at (1416): [<c031958d>] _write_lock_bh+0xf/0x67
Quoted 5 lines
Quoted 1 line

Can you please compile with CONFIG_DISABLE_CONSOLE_SUSPEND set and try:

Ah, that's -mm, sorry. Instead of setting CONFIG_DISABLE_CONSOLE_SUSPEND (which has been removed) please pass no_console_suspend in the command line.

Didn't appear to change anything. Bear in mind that e100 netconsole is a bit busted across resume anyway, and the video display on this machine has never ever survived resume-from-ram (but the X server can bring it back).

#5 lore raw
Quoted 20 lines · Andrew Morton
Quoted 8 lines · On Tue, 18 Sep 2007 13:59:04 +0200 "Rafael J. Wysocki" <[email protected]> wrote:
Quoted 7 lines · Rafael J. Wysocki
Quoted 6 lines · Andrew Morton
Quoted 4 lines · On Mon, 17 Sep 2007 21:12:13 -0700 Andrew Morton <[email protected]>…

It suspends OK, then during resume it gets partway through it, then everything just stops.

Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 3 lines

Would prefer not to have to bisect this one - I've done enough bisecting this week to last a lifetime, and bisecting git-acpi is painful due to build bustage at various points (cpuidle).

And git-acpi breaks suspend-to-disk as well. It gets up to "Suspending console(s)" and then the cursor stops blinking at it wedges up.

Bisection shows that the resume-from-ram failure is caused by

commit 987196fa82d4db52c407e8c9d5dec884ba602183

Quoted 8 lines
Author: Venkatesh Pallipadi <[email protected]>
Date:   Thu Feb 22 13:52:57 2007 -0800

cpuidle take2: Core cpuidle infrastructure

Note that this is the patch which *fixed* resume-from-RAM prior to Thomas's git-hrt merge. Now it breaks it!?!?!

Beats me hands down. :-(

I guess Thomas and Venki should look into it.

Quoted 15 lines

Also, I'm seeing this, in mainline, when I do echo mem > /sys/power/state

[   26.316685] ipw2200: Failed to send WEP_KEY: Aborted due to RF kill switch.
[   33.044631] ipw2200: Failed to send WEP_KEY: Command timed out.
[   39.862821] ipw2200: Failed to send WEP_KEY: Command timed out.
[   68.962896] PM: Preparing system for mem sleep
[   68.979485] Stopping tasks ... WARNING: at kernel/lockdep.c:2658 check_flags()
[   68.980154]  [<c0104ea4>] show_trace_log_lvl+0x1a/0x2f
[   68.980295]  [<c0105a44>] show_trace+0x12/0x14
[   68.980416]  [<c0105a5b>] dump_stack+0x15/0x17
[   68.980535]  [<c0137276>] check_flags+0x93/0x13d
[   68.980663]  [<c013a8ab>] lock_acquire+0x3a/0x91
[   68.980789]  [<c031948b>] _spin_lock+0x38/0x62
[   68.980913]  [<c0142a4d>] refrigerator+0x13/0xc2
[   68.981040]  [<c01283a2>] get_signal_to_deliver+0x32/0x405
Quoted 8 lines
[   68.981188]  [<c01035f4>] do_notify_resume+0x91/0x69f
[   68.981323]  [<c010402d>] work_notifysig+0x13/0x1a
[   68.981453]  =======================
[   68.981542] irq event stamp: 1511
[   68.981624] hardirqs last  enabled at (1511): [<c010408d>] syscall_exit_work+0x11/0x26
[   68.981834] hardirqs last disabled at (1510): [<c0103f63>] syscall_exit+0x9/0x1a
[   68.982031] softirqs last  enabled at (1418): [<c03104c3>] unix_accept+0xe5/0xfb
[   68.982230] softirqs last disabled at (1416): [<c031958d>] _write_lock_bh+0xf/0x67

I guess task_lock(current) in refrigerator() triggers this, which probably means that task_lock(current) has been taken with interrupts disabled somewhere.

I'm not sure if my interpretation is correct, though.

I also don't really know what I'm supposed to do about it ...

Quoted 9 lines
Quoted 5 lines
Quoted 1 line

Can you please compile with CONFIG_DISABLE_CONSOLE_SUSPEND set and try:

Ah, that's -mm, sorry. Instead of setting CONFIG_DISABLE_CONSOLE_SUSPEND (which has been removed) please pass no_console_suspend in the command line.

Didn't appear to change anything. Bear in mind that e100 netconsole is a bit busted across resume anyway, and the video display on this machine has never ever survived resume-from-ram (but the X server can bring it back).

s2ram might help here (http://en.opensuse.org/s2ram), but I guess the Vaio is not on the whitelist ...

#6 lore raw
Quoted 12 lines · Rafael J. Wysocki
Quoted 12 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

And git-acpi breaks suspend-to-disk as well. It gets up to "Suspending console(s)" and then the cursor stops blinking at it wedges up.

Bisection shows that the resume-from-ram failure is caused by

commit 987196fa82d4db52c407e8c9d5dec884ba602183

Author: Venkatesh Pallipadi <[email protected]>
Date:   Thu Feb 22 13:52:57 2007 -0800

cpuidle take2: Core cpuidle infrastructure

Quoted 6 lines
Quoted 2 lines

Note that this is the patch which *fixed* resume-from-RAM prior to Thomas's git-hrt merge. Now it breaks it!?!?!

Beats me hands down. :-(

I guess Thomas and Venki should look into it.

Yeah, I twisted my brain already. I checked the difference of the cpuidle stuff, which I have in my 2.6.23-rc6-hrt2 queue (including the mainline clockevent fixes). It's basically zero.

My own incarnation of a jinxed VAIO is showing the same problem with -hrt2. rc6-mm1 is not booting at all on that box.

I'm going back into the dark fishing grounds of suspend resume debugging.

tglx

reviewer
#7 lore raw
In reply to Andrew Morton
Quoted 3 lines · Andrew Morton

It suspends OK, then during resume it gets partway through it, then everything just stops.

Hmmm, i thought your vaio resume problems were in linus' tree -- did that go away -- or maybe it was in -mm and not in linus' tree all along?

Quoted 3 lines

Would prefer not to have to bisect this one - I've done enough bisecting this week to last a lifetime, and bisecting git-acpi is painful due to build bustage at various points (cpuidle).

bisecting the acpi git three should work well, because it uses topic branches. This includes cpuidle. ie. unless the breakage is within the cpuidle branch itself, the whole cpuidle topic can be removed by skipping the single merge commit that pulls it in.

If you like, I can build the topic branch heads and export plain patches for you to try. Also, I'll probably be re-writing the test branch history later today -- so let me know if you dig into it and i'll delay that until you're done.

thanks,
-Len

#8 lore raw

Bissecting acpi merges (was Re: git-acpi breaks resume-from-ram on the Vaio)

Quoted 4 lines · Len Brown

bisecting the acpi git three should work well, because it uses topic branches. This includes cpuidle. ie. unless the breakage is within the cpuidle branch itself, the whole cpuidle topic can be removed by skipping the single merge commit that pulls it in.

The issue I had sometimes while bissecting ACPI commits was that the topic branches are often rooted on much earlier versions of the kernel tree. So you are near rcY, and suddenly you are way back to rc(Y-2) while inside one of the ACPI merges... depending on what happened in the large number of non-acpi commits this reverts, that can lead to some annoyance with the resulting kernel.

I don't recall if those were bissects of acpi-test, or of mainline though.

Maybe my git-fu is simply not up to it, and the above is not a problem for those with better git skills.

Signature · 4 lines
  "One disk to rule them all, One disk to find them. One disk to bring
  them all and in the darkness grind them. In the Land of Redmond
  where the shadows lie." -- The Silicon Valley Tarot
  Henrique Holschuh
reviewer
#9 lore raw

Re: Bissecting acpi merges (was Re: git-acpi breaks resume-from-ram on the Vaio)

Quoted 6 lines · Henrique de Moraes Holschuh
Quoted 4 lines · Len Brown

bisecting the acpi git three should work well, because it uses topic branches. This includes cpuidle. ie. unless the breakage is within the cpuidle branch itself, the whole cpuidle topic can be removed by skipping the single merge commit that pulls it in.

Quoted 6 lines

The issue I had sometimes while bissecting ACPI commits was that the topic branches are often rooted on much earlier versions of the kernel tree. So you are near rcY, and suddenly you are way back to rc(Y-2) while inside one of the ACPI merges... depending on what happened in the large number of non-acpi commits this reverts, that can lead to some annoyance with the resulting kernel.

yes, that can be a problem when you get pulled down into a topic branch with an old root. However, as the topic branches are all pulled into they tree one after another, it should be very quick to determine which one of them is the cause.

Topic branches with old roots are also more work for me b/c upstream diverges and I have to repeat merges. for this reason I re-based the cpuidle branch a while back -- as that is by far the oldest one.

-Len

#10 lore raw
In reply to Len Brown
Quoted 7 lines · On Tue, 18 Sep 2007 11:21:41 -0400 Len Brown <[email protected]> wrote:
Quoted 3 lines · Andrew Morton

It suspends OK, then during resume it gets partway through it, then everything just stops.

Hmmm, i thought your vaio resume problems were in linus' tree -- did that go away -- or maybe it was in -mm and not in linus' tree all along?

It got fixed in Linus's tree by tglx's recent timers merge.

This breakage in git-acpi is new, I think. Four or five days ago, git-acpi _fixed_ suspend/resume on Linus's tree. Now it breaks it.

But I don't think there have been changes in got-acpi since then, so perhaps the breakage is due to interaction with Thomas's recent merge.

Quoted 8 lines
Quoted 3 lines

Would prefer not to have to bisect this one - I've done enough bisecting this week to last a lifetime, and bisecting git-acpi is painful due to build bustage at various points (cpuidle).

bisecting the acpi git three should work well, because it uses topic branches. This includes cpuidle. ie. unless the breakage is within the cpuidle branch itself, the whole cpuidle topic can be removed by skipping the single merge commit that pulls it in.

The breakage is within the cpuidle branch.

Quoted 4 lines

If you like, I can build the topic branch heads and export plain patches for you to try. Also, I'll probably be re-writing the test branch history later today -- so let me know if you dig into it and i'll delay that until you're done.

I won't be doing anything here for a day or so.

I think what needs to happen is that the commits in the cpuidle code need to be folded together. iirc the first couple of commits don't compile, and the third one fixes them. Or something like that.

reviewer
#11 lore raw
Quoted 4 lines

I think what needs to happen is that the commits in the cpuidle code need to be folded together. iirc the first couple of commits don't compile, and the third one fixes them. Or something like that.

yes, i think the older history has probably outlived its utility and can be folded entirely if it is okay with adam/venki/shaohua.

cheers,
-Len

reviewer
#12 lore raw
In reply to Andrew Morton
Quoted 5 lines · Andrew Morton

This breakage in git-acpi is new, I think. Four or five days ago, git-acpi _fixed_ suspend/resume on Linus's tree. Now it breaks it.

But I don't think there have been changes in got-acpi since then, so perhaps the breakage is due to interaction with Thomas's recent merge.

hmmm, maybe.

Perhaps if you are including sony-laptop, you can try excluding it -- since these reverse engineered things are always fraught with peril....

I think the next thing to test is to revert cpuidle -- let me make sure that my cpuidle branch produces a patch that will revert cleanly...

-Len

#13 lore raw
Quoted 8 lines · On Tue, 18 Sep 2007 15:58:18 -0400 Len Brown <[email protected]> wrote:
Quoted 5 lines · Andrew Morton

This breakage in git-acpi is new, I think. Four or five days ago, git-acpi _fixed_ suspend/resume on Linus's tree. Now it breaks it.

But I don't think there have been changes in got-acpi since then, so perhaps the breakage is due to interaction with Thomas's recent merge.

Quoted 9 lines

hmmm, maybe.

Perhaps if you are including sony-laptop, you can try excluding it -- since these reverse engineered things are always fraught with peril....

I think the next thing to test is to revert cpuidle -- let me make sure that my cpuidle branch produces a patch that will revert cleanly...

ho hum.

I still have the git-acpi tree as a quilt series so this evening I'll bisect these two failures down to a particular git commit.