On 11 September 2015 at 11:57, Mark Brown <broonie@xxxxxxxxxx> wrote:
> On Mon, Sep 07, 2015 at 02:31:06PM +0200, Tomeu Vizoso wrote:
>> On 11 August 2015 at 11:37, Tomeu Vizoso <tomeu.vizoso@xxxxxxxxxxxxx> wrote:
>> > On 7 August 2015 at 14:19, Mark Brown <broonie@xxxxxxxxxx> wrote:
>
>> >> This sounds like it's going to break in the case where we have MFDs that
>> >> represent their functions in DT (not a pattern I'm a fan of but it's a
>> >> thing people do). We'll walk back to the platform device for the MFD
>> >> function, try to probe it and then give up. Perhaps that's good enough
>> >> anyway but it's not clear to me why we don't just try every parent we
>> >> find?
>
>> > Agreed. In the attempt at probing dependencies before a device is
>> > probed, I considered that a device's parent is also a dependency and
>
>> Actually I'm not sure how we could probe the ascendants on demand, as
>> currently the parent's device lock is taken when probing so trying to
>> probe a sibling from within a probe callback will cause a deadlock.
>
> How do silbilings come into this? There is an issue there but it's
> going to happen anyway.
Once a platform device (with the platform bus as its parent) is
retrieved from the deferred queue, both the parent and the device in
question are locked (because of the USB stuff mentioned below). If
that device depends on another device whose parent is the platform bus
and we try to probe it (useless, but I don't see a good way of
avoiding it) then we'll deadlock when device_attach locks that device.
>> AFAICS this is only needed for USB interface devices and this
>> behaviour could be limited to them, but I don't like much assuming
>> that no USB device will ever have a dependency on a sibling (though
>> that probably won't happen ever).
>
> I don't see the connection with USB here, sorry - my initial thought was
> about MFDs?
This commit introduced locking of a device's parent before it's
probed, mainly for USB interfaces:
bf74ad5bc417 ("[PATCH] Hold the device's parent's lock during probe and remove")
Regards,
Tomeu
--
To unsubscribe from this list: send the line "unsubscribe linux-acpi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
