On Wed, Dec 10, 2014 at 2:36 PM, Geert Uytterhoeven
[off-list ref] wrote:

The pcf857x GPIO and interrupt controller uses dummy_irq_chip, which
does not implement irq_chip.irq_set_wake() and does not set
IRQCHIP_SKIP_SET_WAKE.

This causes two s2ram issues if wake-up is enabled for the pcf857x GPIO
pins:
  1. During resume from s2ram, the following warning is printed:

     WARNING: CPU: 0 PID: 1046 at kernel/irq/manage.c:537 irq_set_irq_wake+0x9c/0xf8()
     Unbalanced IRQ 113 wake disable

  2. Wake-up through the pcf857x GPIO pins may fail, as the parent
     interrupt controller may be suspended.

Migrate the pcf857x GPIO and interrupt controller from dummy_irq_chip to
its own irq_chip. This irq chip implements irq_chip.irq_set_wake() to
propagate its wake-up setting to the parent interrupt controller.

This fixes wake-up through gpio-keys on sh73a0/kzm9g, where the pcf857x
interrupt is cascaded to irq-renesas-intc-irqpin, and the latter must
not be suspended when wake-up is enabled.

Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>

(...)

+static struct irq_chip pcf857x_irq_chip = {
+       .name           = "pcf857x",
+       .irq_startup    = noop_ret,
+       .irq_shutdown   = noop,
+       .irq_enable     = noop,
+       .irq_disable    = noop,
+       .irq_ack        = noop,
+       .irq_mask       = noop,
+       .irq_unmask     = noop,
+       .irq_set_wake   = pcf857x_irq_set_wake,
+};

Argh this is so simplistic ... any GPIO irqchip worthy of it's name
shoult at least call gpiochip_lock_as_irq() in .irq_request_resources() and vice
versa mutatis mutandis in .irq_release_resources().

Isn't it possible to also migrate this to GPIOLIB_IRQCHIP when
you're at it can also cut out a big chunk of rusty code?

Yours,
Linus Walleij