[PATCH 0/2] USB PM and PM QoS fixes (Re: gpf in pm_qos_remote_wakeup_show)

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

On Sunday, March 31, 2013 03:41:11 AM Rafael J. Wysocki wrote:
> [Moving the thread to the LKML.]
> 
> On Saturday, March 30, 2013 06:41:16 PM Sasha Levin wrote:
> > On 03/15/2013 01:06 PM, Rafael J. Wysocki wrote:
> [...]
> > >> Rafael, Is there anything you would like me to test?
> > > 
> > > Please just test 3.9-rc2 (or later).
> > 
> > Hi Rafael,
> 
> Hi,
> 
> > I got this after a bit of fuzzing, it looks related to the fix:
> 
> So the complaint is that we shouldn't call pm_qos_sysfs_remove_flags() under
> dev_pm_qos_mtx, because then it may deadlock with dev_pm_qos_update_flags()
> called from pm_qos_remote_wakeup_store(), for example.  This appears to be a
> valid one.
> 
> To avoid that, we can use a separate mutex for exposing/hiding the flags
> (and the latency limit too) that won't be acquired by dev_pm_qos_update_flags()
> or dev_pm_qos_update_request().
> 
> Can you please try the patch below?

Never mind, I have reproduced the lockdep splat and the patch fixes it for me.

Moreover, I've discovered that we call dev_pm_qos_hide_flags() from
usb_port_device_release(), which is totally incorrect.

So, I have two patches (on top of the Linus' tree) that will follow shortly:

[1/2] USB / PM: Don't try to hide PM QoS flags from usb_port_device_release()
[2/2] PM / QoS: Avoid possible deadlock related to sysfs access

Thanks,
Rafael


-- 
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.