irqchip/meson-gpio: Allow the GPIO interrupts to wake the system

meson_gpio_irq_chip has neither .irq_set_wake nor IRQCHIP_SKIP_SET_WAKE,
so enable_irq_wake() fails with -ENXIO for every interrupt routed through
it, and such an interrupt never gets IRQD_WAKEUP_STATE.

That matters for suspend-to-idle. suspend_device_irqs() keeps only the
wakeup interrupts enabled; the rest are suspended and masked by the flow
handler when they first fire. A GPIO interrupt that fires during s2idle
therefore brings a CPU out of idle once, gets masked, and the system
sleeps on. For mem the firmware chooses the wakeup sources, and this
flag changes nothing there.

Set IRQCHIP_SKIP_SET_WAKE, like the other hierarchical irqchips with a
GIC parent. The GIC sets the same flag, so irq_chip_set_wake_parent()
would end there anyway.

Tested on ODROID-HC4 with the PCF8563 alarm interrupt routed through this
controller: with the flag the alarm wakes the board from s2idle 4/4, and
/sys/power/pm_wakeup_irq names it; without the flag the board stayed in
s2idle past the alarm until a power cycle (0/5).

Signed-off-by: Igor Velkov <iav@iav.lv>
Signed-off-by: Thomas Gleixner <tglx@kernel.org>
Assisted-by: LLM
Reviewed-by: Neil Armstrong <neil.armstrong@linaro.org>
Link: https://patch.msgid.link/20261010-meson-gpio-wake-v2-1-790ecc6df792@iav.lv

drivers/irqchip/irq-meson-gpio.c

@@ -538,7 +538,7 @@ static struct irq_chip meson_gpio_irq_chip = {
 #ifdef CONFIG_SMP
 	.irq_set_affinity	= irq_chip_set_affinity_parent,
 #endif
-	.flags			= IRQCHIP_SET_TYPE_MASKED,
+	.flags			= IRQCHIP_SET_TYPE_MASKED | IRQCHIP_SKIP_SET_WAKE,
 };
 
 static int meson_gpio_irq_domain_translate(struct irq_domain *domain,