disable_irq() should wait for all running handlers to complete
before returning. As such, if it's used to disable an interrupt
from that interrupt's handler it will deadlock. This replaces
the dangerous instances with the _nosync() variant which doesn't
have this problem.
This patch may be overly paranoid, it'd be cleaner to just
disable all the row irqs _nosync() but it wouldn't be as
correct. Up to $MAINTAINER :-)
Signed-off-by: Ben Nizette <[email protected]>
---
diff --git a/drivers/input/keyboard/omap-keypad.c b/drivers/input/keyboard/omap-keypad.c
index 058fa8b..9606a1c 100644
--- a/drivers/input/keyboard/omap-keypad.c
+++ b/drivers/input/keyboard/omap-keypad.c
@@ -100,8 +100,16 @@ static irqreturn_t omap_kp_interrupt(int irq, void *dev_id)
/* disable keyboard interrupt and schedule for handling */
if (cpu_is_omap24xx()) {
int i;
- for (i = 0; i < omap_kp->rows; i++)
- disable_irq(gpio_to_irq(row_gpios[i]));
+ for (i = 0; i < omap_kp->rows; i++) {
+ /* The interrupt which we're currently handling should
+ * be disabled _nosync() to avoid deadlocks waiting
+ * for this handler to complete. All others should
+ * be disabled the regular way for SMP safety. */
+ if (gpio_to_irq(row_gpios[i]) == irq)
+ disable_irq_nosync(gpio_to_irq(row_gpios[i]));
+ else
+ disable_irq(gpio_to_irq(row_gpios[i]));
+ }
} else
/* disable keyboard interrupt and schedule for handling */
omap_writew(1, OMAP_MPUIO_BASE + OMAP_MPUIO_KBD_MASKIT);
--
1.6.0.2