On Tue, 11 Jul 2017, Linus Torvalds wrote:
> On Tue, Jul 11, 2017 at 9:19 AM, Thomas Gleixner <t...@linutronix.de> wrote:
> >
> > What I do not understand here is that we have already power management
> > around all of that.
> >
> > irq_chip_pm_get(&desc->irq_data);
> > ...
> > chip_bus_lock(desc);
> > ...
> > chip_bus_unlock_sync(desc);
> > ...
> > irq_chip_pm_put(&desc->irq_data);
> >
> > So why is that not sufficient and needs extra magic in that GPIO driver?
>
> Well, irq_chip_pm_get/put() isn't called just over the operation, it's
> called over the *whole* sequence of the irq being enabled at all.
>
> So the different (right now) is that
>
> - chip_bus_lock/unlock_sync() is purely done around the actual
> operations to set up and tear down the irq data.
>
> So this just covers the very short setup/teardown.
>
> - irq_chip_pm_get/put() is called around the *whole* "irqs can be active"
> block
>
> This covers the whole lifetime of the irq, from setup to free.
>
> Very different.
>
> I'd really prefer my simple patch for now, leaving everything working
> the way it used to work. I *think* it's ok for RT too. Yes?
Not completely, because of the free path issues. See the other mail. Tony
confirmed that it works. I wait for Sebastian and queue it with a proper
changelog, ok?
Thanks,
tglx
