[PATCH] gpio: mvebu: clear irq in edge cause register before unmask edge irq

[PATCH] gpio: mvebu: clear irq in edge cause register before unmask edge irq

[Maxim]

From: Maxim Kiselev <bigunclemax@gmail.com>

When input GPIO set from 0 to 1, the interrupt bit asserted in the GPIO
Interrupt Cause Register (ICR) even if the corresponding interrupt
masked in the GPIO Interrupt Mask Register.

Because interrupt mask register only affects assertion of the interrupt
bits in Main Interrupt Cause Register and it does not affect the
setting of bits in the GPIO ICR.

So, there is problem, when we unmask interrupt with already
asserted bit in the GPIO ICR, then false interrupt immediately occurs
even if GPIO don't change their value since last unmask.

Signed-off-by: Maxim Kiselev <bigunclemax@gmail.com>
---
 drivers/gpio/gpio-mvebu.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/gpio/gpio-mvebu.c b/drivers/gpio/gpio-mvebu.c
index 993bbeb3c006..bdae0c08f239 100644
--- a/drivers/gpio/gpio-mvebu.c
+++ b/drivers/gpio/gpio-mvebu.c
@@ -432,6 +432,7 @@ static void mvebu_gpio_edge_irq_unmask(struct irq_data *d)
 	u32 mask = d->mask;
 
 	irq_gc_lock(gc);
+	mvebu_gpio_write_edge_cause(mvchip, ~mask);
 	ct->mask_cache_priv |= mask;
 	mvebu_gpio_write_edge_mask(mvchip, ct->mask_cache_priv);
 	irq_gc_unlock(gc);
-- 
2.24.1

[PATCH] gpio: mvebu: clear irq in edge cause register before unmask edge irq

[Maxim]

From: Maxim Kiselev <bigunclemax@gmail.com>

When input GPIO set from 0 to 1, the interrupt bit asserted in the GPIO
Interrupt Cause Register (ICR) even if the corresponding interrupt
masked in the GPIO Interrupt Mask Register.

Because interrupt mask register only affects assertion of the interrupt
bits in Main Interrupt Cause Register and it does not affect the
setting of bits in the GPIO ICR.

So, there is problem, when we unmask interrupt with already
asserted bit in the GPIO ICR, then false interrupt immediately occurs
even if GPIO don't change their value since last unmask.

Signed-off-by: Maxim Kiselev <bigunclemax@gmail.com>
---
 drivers/gpio/gpio-mvebu.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/gpio/gpio-mvebu.c b/drivers/gpio/gpio-mvebu.c
index 993bbeb3c006..bdae0c08f239 100644
--- a/drivers/gpio/gpio-mvebu.c
+++ b/drivers/gpio/gpio-mvebu.c
@@ -432,6 +432,7 @@ static void mvebu_gpio_edge_irq_unmask(struct irq_data *d)
 	u32 mask = d->mask;
 
 	irq_gc_lock(gc);
+	mvebu_gpio_write_edge_cause(mvchip, ~mask);
 	ct->mask_cache_priv |= mask;
 	mvebu_gpio_write_edge_mask(mvchip, ct->mask_cache_priv);
 	irq_gc_unlock(gc);
-- 
2.24.1

Re: [PATCH] gpio: mvebu: clear irq in edge cause register before unmask edge irq

[Linus Walleij]

On Wed, Jan 15, 2020 at 8:40 AM Maxim <bigunclemax@gmail.com> wrote:

> From: Maxim Kiselev <bigunclemax@gmail.com>
>
> When input GPIO set from 0 to 1, the interrupt bit asserted in the GPIO
> Interrupt Cause Register (ICR) even if the corresponding interrupt
> masked in the GPIO Interrupt Mask Register.
>
> Because interrupt mask register only affects assertion of the interrupt
> bits in Main Interrupt Cause Register and it does not affect the
> setting of bits in the GPIO ICR.
>
> So, there is problem, when we unmask interrupt with already
> asserted bit in the GPIO ICR, then false interrupt immediately occurs
> even if GPIO don't change their value since last unmask.
>
> Signed-off-by: Maxim Kiselev <bigunclemax@gmail.com>

This looks correct to me, a review from the MVEBU maintainers
would be appreciated.

- Should it be applied to fixes as a regression fix?

- Should it be tagged for stable?

Yours.
Linus Walleij

Re: [PATCH] gpio: mvebu: clear irq in edge cause register before unmask edge irq

[Linus Walleij]

On Wed, Jan 15, 2020 at 8:40 AM Maxim <bigunclemax@gmail.com> wrote:

> From: Maxim Kiselev <bigunclemax@gmail.com>
>
> When input GPIO set from 0 to 1, the interrupt bit asserted in the GPIO
> Interrupt Cause Register (ICR) even if the corresponding interrupt
> masked in the GPIO Interrupt Mask Register.
>
> Because interrupt mask register only affects assertion of the interrupt
> bits in Main Interrupt Cause Register and it does not affect the
> setting of bits in the GPIO ICR.
>
> So, there is problem, when we unmask interrupt with already
> asserted bit in the GPIO ICR, then false interrupt immediately occurs
> even if GPIO don't change their value since last unmask.
>
> Signed-off-by: Maxim Kiselev <bigunclemax@gmail.com>

Since there is no feedback from the MVEBU maintainers I have
tentatively applied the patch for v5.6 so it gets some testing.

Yours,
Linus Walleij