quoted

quoted

-	irq_set_chained_handler(irq, dwapb_irq_handler);
-	irq_set_handler_data(irq, gpio);
+	if (!pp->irq_shared) {
+		irq_set_chained_handler(pp->irq, dwapb_irq_handler);
+		irq_set_handler_data(pp->irq, gpio);
+	} else {
+		/*
+		 * Request a shared IRQ since where MFD would have devices
+		 * using the same irq pin
+		 */
+		err = devm_request_irq(gpio->dev, pp->irq,
+				       dwapb_irq_handler_mfd,
+				       IRQF_SHARED, "gpio-dwapb-mfd", gpio);
+		if (err) {
+			dev_err(gpio->dev, "error requesting IRQ\n");
+			irq_domain_remove(gpio->domain);
+			gpio->domain = NULL;
+			return;
+		}
+	}

I think this need some better documentation. Why is it safe to use
devm_request_irq rather than irq_set_chained_handler here?

Usually it is preferred to use irq_set_chained_handler() for the chained handler
so the handler does not show up in /proc/interrupts.
This requires an exclusive non-shared handler which is not the case on the intel
platform. So they have to use devm_request_irq() instead.

Yes, for Intel Quark, it has a single PCI function exporting a GPIO and I2C controller, and
the irq is shared by GPIO and I2C, so we need shared irq as the comments said.