irqchip/bcm2836: Fix IPI acknowledgement after conversion to handle_percpu_devid_irq

It appears that despite its name, the bcm2836_arm_irqchip_ipi_eoi()
callback is an acknowledgement, and not an EOI. This means that
we lose IPIs that are made pending between the handling of the
IPI and the write to LOCAL_MAILBOX0_CLR0. With the right timing,
things fail nicely.

This used to work with handle_percpu_devid_fasteoi_ipi(), which
started by eoi-ing the interrupt. With the standard fasteoi flow,
this doesn't work anymore.

So let's use this callback for what it is, an ack. Your favourite
RPi-2/3 is back up and running.

Fixes: ffdad79 ("irqchip/bcm2836: Make IPIs use handle_percpu_devid_irq()")
Cc: Valentin Schneider <[email protected]>
Reported-by: Guenter Roeck <[email protected]>
Tested-by: Guenter Roeck <[email protected]>
Signed-off-by: Marc Zyngier <[email protected]>
Link: https://lore.kernel.org/r/[email protected]

Author: Marc Zyngier

drivers/irqchip/irq-bcm2836.c

@@ -167,7 +167,7 @@ static void bcm2836_arm_irqchip_handle_ipi(struct irq_desc *desc)
 	chained_irq_exit(chip, desc);
 }
 
-static void bcm2836_arm_irqchip_ipi_eoi(struct irq_data *d)
+static void bcm2836_arm_irqchip_ipi_ack(struct irq_data *d)
 {
 	int cpu = smp_processor_id();
 
@@ -195,7 +195,7 @@ static struct irq_chip bcm2836_arm_irqchip_ipi = {
 	.name		= "IPI",
 	.irq_mask	= bcm2836_arm_irqchip_dummy_op,
 	.irq_unmask	= bcm2836_arm_irqchip_dummy_op,
-	.irq_eoi	= bcm2836_arm_irqchip_ipi_eoi,
+	.irq_ack	= bcm2836_arm_irqchip_ipi_ack,
 	.ipi_send_mask	= bcm2836_arm_irqchip_ipi_send_mask,
 };