On Fri, Aug 28, 2009 at 03:19:25AM +0400, Anton Vorontsov wrote:
[...]
> > That is why platform code should device_init_wakeup() and
> > drivers should check device_can_wakeup(dev) ...
>
> They should (and do) check may_wakeup() (i.e. should_wakeup) before
> suspending, not can_wakeup().
>
> static int ds1374_suspend(struct i2c_client *client, pm_message_t state)
> {
> if (client->irq >= 0 && device_may_wakeup(&client->dev))
> enable_irq_wake(client->irq);
> return 0;
> }
>
> (quite funny, they issue enable_irq_wake(), assuming that otherwise
> IRQ line won't trigger CPU wakeup. But in reality, there are interrupt
> controllers that you can't control in that regard: any IRQ activity
> will always resume CPU. And so 'echo disable > /sys/.../wakeup' won't
> guarantee anything. Unreliable, nasty? Could be.)
BTW, of course we can fix this by masking interrupts before
suspending, but nobody actually do this (but should, I think).
And if RTC's IRQ is wired to power switch you're in trouble
without any way to fix this.
--
Anton Vorontsov
email: cbouatmailru@xxxxxxxxx
irc://irc.freenode.net/bd2
_______________________________________________
linux-pm mailing list
linux-pm@xxxxxxxxxxxxxxxxxxxxxxxxxx
https://lists.linux-foundation.org/mailman/listinfo/linux-pm
