Re: [PATCH 1/2] pinctrl: renesas: rzg2l: Add suspend/resume support

[Geert Uytterhoeven <geert@xxxxxxxxxxxxxx>]

Hi Claudiu,

On Thu, Feb 15, 2024 at 10:15 AM claudiu beznea
<claudiu.beznea@xxxxxxxxx> wrote:
> On 12.02.2024 17:35, claudiu beznea wrote:
> >>>  static const u16 available_ps[] = { 1800, 2500, 3300 };
> >>> @@ -1880,6 +1938,19 @@ static void rzg2l_gpio_irq_print_chip(struct irq_data *data, struct seq_file *p)
> >>>         seq_printf(p, dev_name(gc->parent));
> >>>  }
> >>>
> >>> +static int rzg2l_gpio_irq_set_wake(struct irq_data *data, unsigned int on)
> >>> +{
> >>> +       struct gpio_chip *gc = irq_data_get_irq_chip_data(data);
> >>> +       struct rzg2l_pinctrl *pctrl = container_of(gc, struct rzg2l_pinctrl, gpio_chip);
> >>> +
> >> I think you also have to call irq_set_irq_wake(pctrl->hwirq[...]) here.
> >> Cfr. drivers/gpio/gpio-rcar.c (which is simpler, as it has a single interrupt
> >> parent, instead of a parent irq_domain with multiple interrupts).
> > I had it in my initial implementation (done long time ago) but I don't
> > remember why I removed it. I'll re-add it anyway.
>
> I did some investigation on this. It seems adding irq_set_irq_wake() is not
> necessary as the pinctrl has no virq requested on behalf of itself.
>
> With this irqchip hierarchy (pinctrl-rzg2l -> irq-renesas-rzg2l -> gic) if
> an IRQ consumer, e.g., the gpio-keys, request an interrupt then it may call
> irq_set_irq_wake(virq) (gpio-keys does that).
>
> irq_set_irq_wake(virq) is forwarded to pinctrl as follows:
>
> irq_set_irq_wake(virq, on) ->
>     set_irq_wake_real(virq, ono) ->
>         rzg2l_gpio_irq_set_wake(irq, on)
>
> As the irq_set_irq_wake() gets a virq as argument and as we have no virq
> requested by pinctrl driver there is no need to call irq_set_irq_wake(), as
> of my investigation. Calling it with hwirq will return with -22 and calling
> it with virq received as argument leads to deadlock (as it's the same virq
> that consumer already is configuring with irq_set_irq_wake()) due the
> following line from irq_set_irq_wake():
>
> struct irq_desc *desc = irq_get_desc_buslock(irq, &flags,
> IRQ_GET_DESC_CHECK_GLOBAL);
>
> What we can do is to forward irq_set_wake() to the parent IRQ chip
> (irq-renesas-rzg2l) with irq_chip_set_wake_parent() to let him set its
> wakeup_path, if any. But, at the moment the irq-renesas-rzg2l has
> IRQCHIP_SKIP_SET_WAKE thus the irq_chip_set_wake_parent() does nothing (but
> it can be updated for that). Now I remember that irq_chip_set_wake_parent()
> is what I've called in my initial implementation and removed it due to
> IRQCHIP_SKIP_SET_WAKE.
>
> Please let me know if you are OK to add irq_chip_set_wake_parent() and
> update the irq-renesas-rzg2l driver.

I think calling irq_chip_set_wake_parent() regardless is a good thing
to do.  Whether the irq-renesas-rzg2l needs an update for wake-up
handling, I don't know (and that is orthogonal to the above).

If you haven't already done so, you may want to browse the wake-related
git history of e.g. drivers/gpio/gpio-rcar.c.

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