linux-kernel - Re: [PATCH] PM: Document rules on using pm_runtime

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: <20170920162731.wopbh7iwggu2xouk@sig21.net>
Date:   Wed, 20 Sep 2017 18:27:31 +0200
From:   Johannes Stezenbach <js@...21.net>
To:     "Rafael J. Wysocki" <rafael@...nel.org>
Cc:     Ulf Hansson <ulf.hansson@...aro.org>,
        "Rafael J. Wysocki" <rjw@...ysocki.net>,
        Linux PM <linux-pm@...r.kernel.org>,
        Alan Stern <stern@...land.harvard.edu>,
        Mika Westerberg <mika.westerberg@...ux.intel.com>,
        Andy Shevchenko <andriy.shevchenko@...ux.intel.com>,
        Kevin Hilman <khilman@...nel.org>,
        LKML <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] PM: Document rules on using pm_runtime_resume() in
 system suspend callbacks

On Wed, Sep 20, 2017 at 04:01:32PM +0200, Rafael J. Wysocki wrote:
> On Wed, Sep 20, 2017 at 2:28 PM, Ulf Hansson <ulf.hansson@...aro.org> wrote:
> > On 20 September 2017 at 02:26, Rafael J. Wysocki <rjw@...ysocki.net> wrote:
> >>
> >> Second, leaving devices in runtime suspend in the "suspend" phase of system
> >> suspend is fishy even when their runtime PM is disabled, because that doesn't
> >> guarantee anything regarding their children or possible consumers.  Runtime
> >> PM may still be enabled for those devices at that time and runtime resume may
> >> be triggered for them later, in which case it all quickly falls apart.
> >
> > This is true, although to me this is a about a different problem and
> > has very little to do with pm_runtime_force_suspend().
> >
> > More precisely, whether runtime PM becomes disabled in the suspend
> > phase or suspend_late phase, really doesn't matter. Because in the end
> > this is about suspending/resuming devices in the correct order.
> 
> Yes, it is, but this is not my point (I didn't make it clear enough I guess).
> 
> At the time you make the decision to disable runtime PM for a parent
> (say) and leave it in runtime suspend, all of its children are
> suspended just fine (otherwise the parent wouldn't have been suspended
> too).  However, you *also* need to make sure that there will be no
> attempts to resume any of them *after* that point, which practically
> means that either runtime PM has to have been disabled already for all
> of them at the time it is disabled for the parent, or there has to be
> another guarantee in place.
> 
> That's why the core tries to enforce the "runtime PM disabled for the
> entire hierarchy below" guarantee for the devices with direct_complete
> set, but that may just be overkill in many cases.  I guess it may be
> better to use WARN_ON() to catch the cases in which things may really
> go wrong.

I read this half a dozen times and I'm still confused.
Moreover, Documentation/driver-api/pm/devices.rst says:

    Runtime Power Management model:

        Devices may also be put into low-power states while the system is
        running, independently of other power management activity in principle.
        However, devices are not generally independent of each other (for
        example, a parent device cannot be suspended unless all of its child
        devices have been suspended).  ...

However, isn't this a fundamental difference of runtime suspend
vs. system suspend that parent devices *can* be runtime suspended
before their children?  E.g. an audio codec could keep running
while the i2c bus used to program its registers can be runtime suspended.
If this is correct I think it would be useful to spell it out explicitly
in the documentation.

During system suspend, pm core will suspend children first,
and if the child's ->suspend hook uses the i2c bus to access registers,
it will implicitly runtime resume the i2c bus (e.g. due to pm_runtime_get_sync()
in i2c_dw_xfer()).  Later pm core will ->suspend the i2c bus.

I have a hunch the root of the problem is that ->prepare walks the tree
in top-down order, and its return value is used to decide about
direct-complete.  Why does it do that?  Shouldn't pm core check
the direct_complete flag during ->suspend if the device
is in runtime suspend, to decide whether to skip runtime resume + ->suspend
for *this* device?


Johannes