[Regression] firewire_ohci causes system resume to fail

[Regression] firewire_ohci causes system resume to fail

[Rafael J. Wysocki]

Hi,

There is a regression I can reproduce 100% of the time on a AMD-based desktop
box that resume from hibernation (or suspend) fails because of the firewire_ohci
module.  The box just hangs when firewire_ohci resume runs, apparently, and it
is sufficient to do:

# echo devices > /sys/power/pm_test
# echo disk > /sys/power/state

to reproduce the issue.

It definitely works with 2.6.34, but it fails with 2.6.37-rc4+ (latest mainline).
I haven't tested the kernels in between, but I'm going to do that later today
or some time next week.

If you have any idea about what may be the reason, please let me know.

Thanks,
Rafael

Re: [Regression] firewire_ohci causes system resume to fail

[Stefan Richter]

On Dec 05 Rafael J. Wysocki wrote:
> There is a regression I can reproduce 100% of the time on a AMD-based desktop
> box that resume from hibernation (or suspend) fails because of the firewire_ohci
> module.  The box just hangs when firewire_ohci resume runs, apparently, and it
> is sufficient to do:
> 
> # echo devices > /sys/power/pm_test
> # echo disk > /sys/power/state
> 
> to reproduce the issue.

Do you have anything attached to the FireWire bus?  What controller do
you have?

> It definitely works with 2.6.34, but it fails with 2.6.37-rc4+ (latest mainline).
> I haven't tested the kernels in between, but I'm going to do that later today
> or some time next week.
> 
> If you have any idea about what may be the reason, please let me know.

As first quick thought:  While we of course need to have a look at
firewire-ohci changes, it could also be a higher-level issue in
firewire-core.  E.g. something with the 2.6.36 workqueue related core
kernel changes.  (I am only saying so because I noticed some module
unloading issues with the firewire stack lately that seem workqueue
related.)

I admit that PM suspend/resume or hibernate/restore are not part of my
personal firewire driver testing regime.  I need to address that.

I Cc'd Maxim Levitsky who has working firewire suspend/resume on
2.6.37-rc, although with some minor firewire-net related and
isochronous I/O related nits for which he sent fixes lately:
http://marc.info/?l=linux1394-devel&m=129099665721794
http://marc.info/?l=linux1394-devel&m=129099666021798
But I do not think that these could matter to your hang-at-resume issue.
-- 
Stefan Richter
-=====-==-=- ==-- --=-=
http://arcgraph.de/sr/

Re: [Regression] firewire_ohci causes system resume to fail

[Stefan Richter]

On Dec 05 Stefan Richter wrote:
> On Dec 05 Rafael J. Wysocki wrote:
> > There is a regression I can reproduce 100% of the time on a AMD-based desktop
> > box that resume from hibernation (or suspend) fails because of the firewire_ohci
> > module.  The box just hangs when firewire_ohci resume runs, apparently, and it
> > is sufficient to do:
> > 
> > # echo devices > /sys/power/pm_test
> > # echo disk > /sys/power/state
> > 
> > to reproduce the issue.
> 
> Do you have anything attached to the FireWire bus?  What controller do
> you have?

I just now tested the above on an Intel Mac mini, running
unmodified 2.6.37-rc4 x86-32, without and with a camera on the FireWire
port, and could not reproduce the problem.  I also got an AMD based PC
here with an array of different FireWire controllers which I shall make
available to the same test...
-- 
Stefan Richter
-=====-==-=- ==-- --=-=
http://arcgraph.de/sr/

Re: [Regression] firewire_ohci causes system resume to fail

[Rafael J. Wysocki]

On Sunday, December 05, 2010, Stefan Richter wrote:
> On Dec 05 Stefan Richter wrote:
> > On Dec 05 Rafael J. Wysocki wrote:
> > > There is a regression I can reproduce 100% of the time on a AMD-based desktop
> > > box that resume from hibernation (or suspend) fails because of the firewire_ohci
> > > module.  The box just hangs when firewire_ohci resume runs, apparently, and it
> > > is sufficient to do:
> > > 
> > > # echo devices > /sys/power/pm_test
> > > # echo disk > /sys/power/state
> > > 
> > > to reproduce the issue.
> > 
> > Do you have anything attached to the FireWire bus?  What controller do
> > you have?
> 
> I just now tested the above on an Intel Mac mini, running
> unmodified 2.6.37-rc4 x86-32, without and with a camera on the FireWire
> port, and could not reproduce the problem.  I also got an AMD based PC
> here with an array of different FireWire controllers which I shall make
> available to the same test...

I'm unable to reproduce the problem with 2.6.37-rc6, so presumably this was
a temporary glitch with the particular kernel build.

Thanks,
Rafael

Re: [Regression] firewire_ohci causes system resume to fail

[Stefan Richter]

On Dec 17 Rafael J. Wysocki wrote:
> On Sunday, December 05, 2010, Stefan Richter wrote:
> > On Dec 05 Stefan Richter wrote:
> > > On Dec 05 Rafael J. Wysocki wrote:
> > > > There is a regression I can reproduce 100% of the time on a AMD-based desktop
> > > > box that resume from hibernation (or suspend) fails because of the firewire_ohci
> > > > module.  The box just hangs when firewire_ohci resume runs, apparently, and it
> > > > is sufficient to do:
> > > > 
> > > > # echo devices > /sys/power/pm_test
> > > > # echo disk > /sys/power/state
> > > > 
> > > > to reproduce the issue.
> > > 
> > > Do you have anything attached to the FireWire bus?  What controller do
> > > you have?
> > 
> > I just now tested the above on an Intel Mac mini, running
> > unmodified 2.6.37-rc4 x86-32, without and with a camera on the FireWire
> > port, and could not reproduce the problem.  I also got an AMD based PC
> > here with an array of different FireWire controllers which I shall make
> > available to the same test...
> 
> I'm unable to reproduce the problem with 2.6.37-rc6, so presumably this was
> a temporary glitch with the particular kernel build.

Good.

I finally got around to test this on my AMD RS780 based PC with several
FireWire controllers in it.  It performs above PM test without any
apparent problem.
-- 
Stefan Richter
-=====-==-=- ==-- =---=
http://arcgraph.de/sr/