gpio: rcar: Fix level interrupt handling

According to the manual, if a port is set for level detection using
the corresponding bit in the edge/level select register and an external
level interrupt signal is asserted, the corresponding bit in INTDT
does not use the FF to hold the input.
Thus, writing 1 to the corresponding bits in INTCLR cannot clear the
corresponding bits in the INTDT register. Instead, when an external
input signal is stopped, the corresponding bit in INTDT is cleared
automatically.

Since the INTDT bit cannot be cleared for the level interrupts until
the interrupt signal is stopped, we end up with the infinite loop
when using deferred (threaded) IRQ handling.

Since a deferred interrupt is disabled by the low-level handler and
re-enabled only when the deferred handler is completed, Fix the issue
by dropping disabled interrupts from the pending mask as suggested by
Laurent Pinchart <[email protected]>

Changes in V2:
* Drop disabled interrupts from pending mask altogether instead of
  dropping level interrupts one by one once they get handled.

Signed-off-by: Valentine Barshak <[email protected]>
Acked-by: Laurent Pinchart <[email protected]>
Acked-by: Magnus Damm <[email protected]>
Signed-off-by: Linus Walleij <[email protected]>

diff --git a/drivers/gpio/gpio-rcar.c b/drivers/gpio/gpio-rcar.c
index fe088a30567ac63325fd02be82b8e682aa2323eb..8b7e719a68c3662f454c76b3e82b9569abeb22f8 100644 (file)
--- a/drivers/gpio/gpio-rcar.c
+++ b/drivers/gpio/gpio-rcar.c
@@ -169,7 +169,8 @@ static irqreturn_t gpio_rcar_irq_handler(int irq, void *dev_id)
        u32 pending;
        unsigned int offset, irqs_handled = 0;
 
-       while ((pending = gpio_rcar_read(p, INTDT))) {
+       while ((pending = gpio_rcar_read(p, INTDT) &
+                         gpio_rcar_read(p, INTMSK))) {
                offset = __ffs(pending);
                gpio_rcar_write(p, INTCLR, BIT(offset));
                generic_handle_irq(irq_find_mapping(p->irq_domain, offset));