Hi Linus,
On Sun, May 1, 2016 at 10:48 AM, Linus Walleij <linus.walleij@xxxxxxxxxx> wrote:
>> Currrently the gpio_chip.to_irq() callback returns -ENOSYS on error,
>> which causes bad interactions with the serial_mctrl_gpio helpers.
>>
>> mctrl_gpio_init() returns -ENOSYS if GPIOLIB is not enabled, which is
>> intended to be ignored by its callers. However, ignoring -ENOSYS when it
>> was caused by a gpiod_to_irq() failure will lead to a crash later:
>>
>> Unable to handle kernel paging request at virtual address ffffffde
>> ...
>> PC is at mctrl_gpio_set+0x14/0x78
>>
>> Fix this by returning -ENXIO instead.
>>
>> Signed-off-by: Geert Uytterhoeven <geert+renesas@xxxxxxxxx>
>> ---
>> Is -ENXIO the right error code?
>
> I would say that whatever the generic helpers in drivers/gpio/gpiolib.c
> returns should be the norm. However the guy who wrote them (me)
> isn't very smart either. It looks like so:
>
> static int gpiochip_to_irq(struct gpio_chip *chip, unsigned offset)
> {
> return irq_find_mapping(chip->irqdomain, offset);
> }
>
> That returns 0 (NO_IRQ) on failure.
>
> And as you say:
>
>> - Drivers that call irq_find_mapping(), irq_create_mapping(), or
>> irq_create_fwspec_mapping() return zero! This also applies to the
>> core helper gpiochip_to_irq().
>
> Zero means NO_IRQ.
>
> Reminder:
> http://lwn.net/Articles/470820/
>
> What we should do is patch all drivers to return 0 on failure, and
> patch any consumers like mctrl_gpio_init() to handle that correctly.
That's the Long Term Plan. There are still too many non-zero NO_IRQ
definitions in use...
Is -ENXIO acceptable for the short term?
Thanks!
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
