linux-kernel - Re: lockdep warning in urb.c:363 usb_submit

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CAJZ5v0j8jaQPOPEjPJS7=i8ZXOZuevh83XGTyrQnYPyE9y9DPA@mail.gmail.com>
Date:   Sat, 18 Apr 2020 11:40:11 +0200
From:   "Rafael J. Wysocki" <rafael@...nel.org>
To:     Alan Stern <stern@...land.harvard.edu>
Cc:     "Rafael J. Wysocki" <rjw@...ysocki.net>,
        "Rafael J. Wysocki" <rafael@...nel.org>,
        Qais Yousef <qais.yousef@....com>,
        USB list <linux-usb@...r.kernel.org>,
        Linux-pm mailing list <linux-pm@...r.kernel.org>,
        Kernel development list <linux-kernel@...r.kernel.org>
Subject: Re: lockdep warning in urb.c:363 usb_submit_urb

On Sat, Apr 18, 2020 at 1:37 AM Alan Stern <stern@...land.harvard.edu> wrote:
>
> On Fri, 17 Apr 2020, Rafael J. Wysocki wrote:
>
> > There is one detail here that I missed, sorry about that.
> >
> > Actually, the core can only set the runtime status to "active" for
> > devices where dev_pm_skip_suspend() returns 'true'.
> >
> > First, if the device is not "suspended", its status is "active" already
> > anyway.
> >
> > Second, if the device has SMART_SUSPEND clear, the driver may not expect
> > its runtime status to change from "suspended" to "active" during system-wide
> > resume-type transitions (the driver's system-wide PM callbacks may use
> > the runtime status to determine what to do and changing the status this
> > way may confuse that).
> >
> > [Actually, the drivers that set neither SMART_SUSPEND nor MAY_SKIP_RESUME
> >  may not expect the runtime status to change during system-wide resume-type
> >  transitions at all, but there is the corner case when the driver can set
> >  MAY_SKIP_RESUME without setting SMART_SUSPEND.  In that case its "noirq"
> >  and "early" resume callbacks may be skipped and then it should expect
> >  the runtime status to sometimes change from "active" to "suspended" during
> >  RESUME transitions, but it may still not expect to see changes the other way
> >  around, as in that case all of its callbacks are going to be invoked and
> >  apply the internal runtime status handling mentioned above.]
> >
> > So overall:
> >
> >   At the start of the {resume,thaw,restore}_noirq phase, if
> >   dev_pm_skip_resume() returns true ,then the core will set the
> >   runtime status to "suspended".  Otherwise, if dev_pm_skip_suspend()
> >   also returns true, then the core will set the runtime status to "active".
> >   If this is not what the subsystem or driver wants, it must update the
> >   runtime status itself.
>
> Sigh.  The bug which prompted this whole thread was when I forgot to
> set the runtime PM status back to "active" in one of my drivers.  I was
> hoping the core could handle it for me automatically.
>
> I guess the answer is always to set the SMART_SUSPEND flag.

I would say so. :-)

> > > > > For this to work properly, we will have to rely on subsystems/drivers
> > > > > to call pm_runtime_resume() during the suspend/freeze transition if
> > > > > SMART_SUSPEND is clear.
> > > >
> > > > That has been the case forever, though.
> > >
> > > I'm not so sure about that.  The existing PM core code doesn't ever get
> > > into a situation where it tries to set a device's runtime status to
> > > "active" while the parent's status is "suspended".
> >
> > I'm assuming that you refer to the scenario below.
> >
> > > > > Otherwise we could have the following scenario:
> > > > >
> > > > > Device A has a child B, and both are runtime suspended when hibernation
> > > > > starts.  Suppose that the SMART_SUSPEND flag is set for A but not for
> > > > > B, and suppose that B's subsystem/driver neglects to call
> > > > > pm_runtime_resume() during the FREEZE transition.  Then during the THAW
> > > > > transition, dev_pm_skip_resume() will return "true" for A and "false"
> > > > > for B.  This will lead to an error when the core tries to set B's
> > > > > runtime status to "active" while A's status is "suspended".
> >
> > That cannot happen, because dev_pm_smart_suspend() also returns 'false' for B
> > and so its runtime status will not be changed to "active".
>
> Yes, your change to dev_pm_skip_resume() will prevent the problem from
> arising.
>
>
> > BTW, I have updated my pm-sleep-core branch to reflect what appears to be
> > the current state-of-the-art to me.
> >
> > I'm going to post a v2 of this patch series over the weekend for reference.
>
> Okay, I'll check it out.
>
> By the way, if you don't mind I may want to do some editing of
> devices.rst.

Sure, please feel free to do that.

Cheers!