On Wed, Feb 10, 2016 at 4:09 PM, Marc Zyngier [off-list ref] wrote:

On 29/01/16 04:28, Quan Nguyen wrote:

quoted

+static void xgene_gpio_sb_domain_deactivate(struct irq_domain *d,
+             struct irq_data *irq_data)
+{
+     struct xgene_gpio_sb *priv = d->host_data;
+     u32 gpio = HWIRQ_TO_GPIO(priv, irq_data->hwirq);
+
+     gpiochip_unlock_as_irq(&priv->gc, gpio);

Is it right to do the unlock both in irq_shutdown and domain_deactivate?
This seems a bit odd to me to have such an inbalance. My hunch is that
you should either implement irq_startup, do the locking there and drop
the unlock drop deactivate, or kill irq_shutdown.

Linus, what do you think?

Those functions should be called in .irq_alloc/release_resources(), that is
why Gleixner added those callbacks in the first place.

Yours,
Linus Walleij