Hi Marc,
On Mon, Apr 11, 2016 at 6:55 PM, Marc Zyngier <marc.zyngier@xxxxxxx> wrote:
> On 11/04/16 17:26, Laurent Pinchart wrote:
>> On Friday 19 Feb 2016 11:59:40 Marc Zyngier wrote:
>>> On 19/02/16 09:18, Linus Walleij wrote:
>>>> Top-quoting so everyone on the new To:-line gets the context.
>>>>
>>>> I definately need an indication from an irqchip maintainer like tglx or
>>>> Marc Z before I merge this. Also, as in reply to the previous letter,
>>>> coordinate efforts with Jon Hunters similar problem space.
>>>
>>> Seems pretty straightforward to me.
>>>
>>> Acked-by: Marc Zyngier <marc.zyngier@xxxxxxx>
>>
>> Too straightforward to be correct :-/
>>
>> [ 6.232681] BUG: sleeping function called from invalid context at /home/laurent/src/iob/renesas/linux/drivers/base/power/runtime.c:955
>> [ 6.244795] in_atomic(): 1, irqs_disabled(): 128, pid: 658, name: udevd
>> [ 6.251429] CPU: 3 PID: 658 Comm: udevd Tainted: P 4.6.0-rc3 #756
>> [ 6.258844] Hardware name: Generic R8A7790 (Flattened Device Tree)
>
> [...]
>
> Ah! That will teach me a lesson.
>
>> The .irq_request_resources() handler is called with a spinlock held, it thus
>> can't call the synchronous version of the PM runtime functions.
>
> OK, so we're back to square one. Is that just a matter of calling the
> non-synchronous version? My hunch is that it is not that simple...
>
> Geert?
Unfortunately it's not that simple.
The irqchip must be runtime-resumed before we can access its registers.
I'm afraid we'll have to keep gpio-rcar runtime-resumed all the time,
i.e. revert both of
b26a719bdb ("gpio: rcar: Add Runtime PM handling for interrupts")
65194cb174 ("gpio: rcar: Fine-grained Runtime PM support")
until Jon Hunter's "genirq: Add runtime power management support for IRQ chips"
(and perhaps a few more patches from his series) is applied.
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx
In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds
