linux-kernel - Re: [PATCH 3/3] ACPI / PCI / PM: Rework acpi_pci_propagate

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]

Date:   Fri, 21 Jul 2017 22:44:30 +0200
From:   "Rafael J. Wysocki" <rjw@...ysocki.net>
To:     Andy Shevchenko <andy.shevchenko@...il.com>
Cc:     Linux PCI <linux-pci@...r.kernel.org>,
        Linux ACPI <linux-acpi@...r.kernel.org>,
        Linux PM <linux-pm@...r.kernel.org>,
        LKML <linux-kernel@...r.kernel.org>,
        Mika Westerberg <mika.westerberg@...ux.intel.com>,
        Bjorn Helgaas <bhelgaas@...gle.com>
Subject: Re: [PATCH 3/3] ACPI / PCI / PM: Rework acpi_pci_propagate_wakeup()

On Friday, July 21, 2017 06:45:03 PM Andy Shevchenko wrote:
> On Fri, Jul 21, 2017 at 3:42 PM, Rafael J. Wysocki <rjw@...ysocki.net> wrote:
> 
> > The acpi_pci_propagate_wakeup() routine is there to handle cases in
> > which PCI bridges (or PCIe ports) are expected to signal wakeup
> > for devices below them, but currently it doesn't do that correctly.
> >
> > The problem is that acpi_pci_propagate_wakeup() uses
> > acpi_pm_set_device_wakeup() for bridges and if that routine is
> > called for multiple times to disable wakeup for the same device,
> > it will disable it on the first invocation and the next calls
> > will have no effect (it works analogously when called to enable
> > wakeup, but that is not a problem).
> >
> > Now, say acpi_pci_propagate_wakeup() has been called for two
> > different devices under the same bridge and it has called
> > acpi_pm_set_device_wakeup() for that bridge each time.  The
> > bridge is now enabled to generate wakeup signals.  Next,
> > suppose that one of the devices below it resumes and
> > acpi_pci_propagate_wakeup() is called to disable wakeup for that
> > device.  It will then call acpi_pm_set_device_wakeup() for the bridge
> > and that will effectively disable remote wakeup for all devices under
> > it even though some of them may still be suspended and remote wakeup
> > may be expected to work for them.
> >
> > To address this (arguably theoretical) issue, allow
> > wakeup.enable_count under struct acpi_device to grow beyond 1 in
> > certain situations.  In particular, allow that to happen in
> > acpi_pci_propagate_wakeup() when wakeup is enabled or disabled
> > for PCI bridges, so that wakeup is actually disabled for the
> > bridge when all devices under it resume and not when just one
> > of them does that.
> 
> > -       if (wakeup->enable_count > 0)
> > -               goto out;
> > +       if (wakeup->enable_count > 0) {
> > +               if (wakeup->enable_count < max_count)
> > +                       goto inc;
> > +               else
> > +                       goto out;
> > +       }
> 
> Wouldn't be simpler

I'm not really sure what you mean.

In general, ->

>     if (wakeup->enable_count >= max_count)
>       goto out;

-> this is unlikely and ->>

>     if (wakeup->enable_count > 0)
>       goto inc;

->> this isn't.

Why would checking an unlikely condition before a likely one covering it
ever be better?

> If max_count can be <= 0,

No, it can't be.

Thanks,
Rafael