Re: [linux-pm] Fundamental flaw in system suspend, exposed by freezer removal

["Rafael J. Wysocki" <rjw@sisk.pl>]

On Thursday, 28 of February 2008, Alan Stern wrote:
> Rafael:

Hi,

> Here's my patch.  It doesn't include the timers for deadlock debugging, 
> but it does include all the other stuff we've been talking about.  My 
> base probably isn't quite in sync with yours, so this may not apply 
> cleanly on your system.  But the divergences should be small.

No big deal.  In fact we're working with Alex on a patch that modifies main.c
too, so I'll have to resolve merging conflicts anyway. ;-)

> Incidentally, there seemed to be a bug in your dpm_suspend() -- the 
> dpm_list_mtx needs to be reacquired before the error checking.

Oh, yes.  Will you please send a fix when
http://marc.info/?l=linux-acpi&m=120389632114090&w=4 is merged?

> This patch fixes that.  It also removes pm_sleep_rwsem, which isn't used 
> any more.

Yeah.  I left pm_sleep_rwsem, because I wasn't sure it would be necessary
in the future.

> We should think about device_pm_schedule_removal().  It won't work
> right if a suspend method calls it for the device being suspended, 
> because the device gets moved to the dpm_off list after the method 
> runs.

I think we should remove device_pm_schedule_removal() early in the 2.6.26
cycle.

I'll review the patch more thoroughly tomorrow, since I have some new material
for the regressions list I need to take care of.

Thanks,
Rafael