[v3] gpio:mcp23s08 Fixed missing interrupts

From: Robert Middleton <robert.middleton@rm5248.com>

When an interrupt occurs on an MCP23S08 chip, the INTF register will only
contain one bit as causing the interrupt.  If more than two pins change at
the same time on the chip, this causes one of the pins to not be reported.
This patch fixes the logic for checking if a pin has changed, so that
multiple pins will always cause more than one change.

Signed-off-by: Robert Middleton <robert.middleton@rm5248.com>
---
 drivers/gpio/gpio-mcp23s08.c | 65 ++++++++++++++++++++++++++++++++++++++++----
 1 file changed, 60 insertions(+), 5 deletions(-)

On 16/03/2017 04:56, robert.middleton@rm5248.com wrote:
> From: Robert Middleton <robert.middleton@rm5248.com>
>
> When an interrupt occurs on an MCP23S08 chip, the INTF register will only
> contain one bit as causing the interrupt.  If more than two pins change at
> the same time on the chip, this causes one of the pins to not be reported.
> This patch fixes the logic for checking if a pin has changed, so that
> multiple pins will always cause more than one change.
>
> Signed-off-by: Robert Middleton <robert.middleton@rm5248.com>
> ---
>  drivers/gpio/gpio-mcp23s08.c | 65 ++++++++++++++++++++++++++++++++++++++++----
>  1 file changed, 60 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/gpio/gpio-mcp23s08.c b/drivers/gpio/gpio-mcp23s08.c
> index bdb6923..2a57d024 100644
> --- a/drivers/gpio/gpio-mcp23s08.c
> +++ b/drivers/gpio/gpio-mcp23s08.c
> @@ -270,8 +270,10 @@ mcp23s08_direction_output(struct gpio_chip *chip, unsigned offset, int value)
>  static irqreturn_t mcp23s08_irq(int irq, void *data)
>  {
>  	struct mcp23s08 *mcp = data;
> -	int intcap, intf, i;
> +	int intcap, intf, i, gpio, gpio_orig, intcap_mask;
>  	unsigned int child_irq;
> +	bool intf_set, intcap_changed, gpio_bit_changed,
> +		defval_changed, gpio_set;
>
>  	mutex_lock(&mcp->lock);
>  	if (mcp_read(mcp, MCP_INTF, &intf) < 0) {
> @@ -287,14 +289,67 @@ static irqreturn_t mcp23s08_irq(int irq, void *data)
>  	}
>
>  	mcp->cache[MCP_INTCAP] = intcap;
> +
> +	/* This clears the interrupt(configurable on S18) */
> +	if (mcp_read(mcp, MCP_GPIO, &gpio) < 0) {
> +		mutex_unlock(&mcp->lock);
> +		return IRQ_HANDLED;
> +	}
> +	gpio_orig = mcp->cache[MCP_GPIO];
> +	mcp->cache[MCP_GPIO] = gpio;
>  	mutex_unlock(&mcp->lock);
>
> +	if (mcp->cache[MCP_INTF] == 0) {
> +		/* There is no interrupt pending */
> +		return IRQ_HANDLED;
> +	}
> +
> +	dev_dbg(mcp->chip.parent,
> +		"intcap 0x%04X intf 0x%04X gpio_orig 0x%04X gpio 0x%04X\n",
> +		intcap, intf, gpio_orig, gpio);
>
>  	for (i = 0; i < mcp->chip.ngpio; i++) {
> -		if ((BIT(i) & mcp->cache[MCP_INTF]) &&
> -		    ((BIT(i) & intcap & mcp->irq_rise) ||
> -		     (mcp->irq_fall & ~intcap & BIT(i)) ||
> -		     (BIT(i) & mcp->cache[MCP_INTCON]))) {
> +		/* We must check all of the inputs on the chip,
> +		 * otherwise we may not notice a change on >=2 pins.
> +		 *
> +		 * On at least the mcp23s17, INTCAP is only updated
> +		 * one byte at a time(INTCAPA and INTCAPB are
> +		 * not written to at the same time - only on a per-bank
> +		 * basis).
> +		 *
> +		 * INTF only contains the single bit that caused the
> +		 * interrupt per-bank.  On the mcp23s17, there is
> +		 * INTFA and INTFB.  If two pins are changed on the A
> +		 * side at the same time, INTF will only have one bit
> +		 * set.  If one pin on the A side and one pin on the B
> +		 * side are changed at the same time, INTF will have
> +		 * two bits set.  Thus, INTF can't be the only check
> +		 * to see if the input has changed.
> +		 */
> +
> +		intf_set = BIT(i) & mcp->cache[MCP_INTF];
> +		if (i < 8 && intf_set)
> +			intcap_mask = 0x00FF;
> +		else if (i >= 8 && intf_set)
> +			intcap_mask = 0xFF00;
> +		else
> +			intcap_mask = 0x00;
> +
> +		intcap_changed = (intcap_mask &
> +			(BIT(i) & mcp->cache[MCP_INTCAP])) !=
> +			(intcap_mask & (BIT(i) & gpio_orig));
> +		gpio_set = BIT(i) & mcp->cache[MCP_GPIO];
> +		gpio_bit_changed = (BIT(i) & gpio_orig) !=
> +			(BIT(i) & mcp->cache[MCP_GPIO]);
> +		defval_changed = (BIT(i) & mcp->cache[MCP_INTCON]) &&
> +			((BIT(i) & mcp->cache[MCP_GPIO]) !=
> +			(BIT(i) & mcp->cache[MCP_DEFVAL]));
> +
> +		if (((gpio_bit_changed || intcap_changed) &&
> +			(BIT(i) & mcp->irq_rise) && gpio_set) ||
> +		    ((gpio_bit_changed || intcap_changed) &&
> +			(BIT(i) & mcp->irq_fall) && !gpio_set) ||
> +		    defval_changed) {
>  			child_irq = irq_find_mapping(mcp->chip.irqdomain, i);
>  			handle_nested_irq(child_irq);
>  		}
>

I tested this on my system based on 4.9 branch. It required minor changes to the
register read calls to apply to that branch as it didn't have regmap changes.
all works for me.

Tested-by: Phil Reid <preid@electromag.com.au>

On Wed, Mar 15, 2017 at 9:56 PM,  <robert.middleton@rm5248.com> wrote:

> From: Robert Middleton <robert.middleton@rm5248.com>
>
> When an interrupt occurs on an MCP23S08 chip, the INTF register will only
> contain one bit as causing the interrupt.  If more than two pins change at
> the same time on the chip, this causes one of the pins to not be reported.
> This patch fixes the logic for checking if a pin has changed, so that
> multiple pins will always cause more than one change.
>
> Signed-off-by: Robert Middleton <robert.middleton@rm5248.com>

Patch applied for fixes with Phil's Tested-by and tagged for stable.

Yours,
Linus Walleij
--
To unsubscribe from this list: send the line "unsubscribe linux-gpio" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html