Re: [PATCH 1/2] gpio/omap: convert gpio irq domain to linear mapping

[Jon Hunter <jon-hunter@ti.com>]

On 03/04/2013 11:11 AM, Felipe Balbi wrote:
> On Mon, Mar 04, 2013 at 11:06:12AM -0600, Jon Hunter wrote:
>>
>> On 03/02/2013 05:48 AM, Felipe Balbi wrote:
>>> On Fri, Mar 01, 2013 at 11:22:47AM -0600, Jon Hunter wrote:
>>>> Currently the OMAP GPIO driver uses a legacy mapping for the GPIO IRQ
>>>> domain. This is not necessary because we do not need to assign a
>>>> specific interrupt number to the GPIO IRQ domain. Therefore, convert
>>>> the OMAP GPIO driver to use a linear mapping instead.
>>>>
>>>> Please note that this also allows to simplify the logic in the OMAP
>>>> gpio_irq_handler() routine, by using irq_find_mapping() to obtain the
>>>> virtual irq number from the GPIO bank and bank index.
>>>>
>>>> Reported-by: Linus Walleij <linus.walleij@linaro.org>
>>>> Signed-off-by: Jon Hunter <jon-hunter@ti.com>
>>>
>>> Reviewed-by: Felipe Balbi <balbi@ti.com>
>>>
>>> Just one suggestion below for a later patch.
>>>
>>>> @@ -680,7 +686,7 @@ static void gpio_irq_handler(unsigned int irq, struct irq_desc *desc)
>>>>  {
>>>>  	void __iomem *isr_reg = NULL;
>>>>  	u32 isr;
>>>> -	unsigned int gpio_irq, gpio_index;
>>>> +	unsigned int i;
>>>>  	struct gpio_bank *bank;
>>>>  	int unmasked = 0;
>>>>  	struct irq_chip *chip = irq_desc_get_chip(desc);
>>>> @@ -721,15 +727,10 @@ static void gpio_irq_handler(unsigned int irq, struct irq_desc *desc)
>>>>  		if (!isr)
>>>>  			break;
>>>>  
>>>> -		gpio_irq = bank->irq_base;
>>>> -		for (; isr != 0; isr >>= 1, gpio_irq++) {
>>>> -			int gpio = irq_to_gpio(bank, gpio_irq);
>>>> -
>>>> +		for (i = 0; isr != 0; isr >>= 1, i++) {
>>>>  			if (!(isr & 1))
>>>>  				continue;
>>>
>>> this will iterate over all 32 GPIOs, a better way to handle this would
>>> be to have something like:
>>
>> Worse case, if only bit 31 was set then I agree this is not that
>> efficient. Or even if one bit is set. However, the loop itself will
>> iterate while isr != 0 so not always over each bit. No different to the
>> existing code.
>>
>>> while (isr) {
>>> 	unsigned long bit = __ffs(isr);
>>>
>>> 	/* clear this bit */
>>> 	isr &= ~bit;
>>>
>>> 	generic_handle_irq(irq_find_mapping(bank->domain, bit);
>>> }
>>>
>>> this way you will only iterate the amount of bits enabled in the isr
>>> register.
>>
>> Definitely cleaner but I am wondering which approach would be more
>> efficient from an instruction standpoint. This could definitely be much
>> more efficient if there is only a couple bits set.
> 
> __ffs() is done with CLZ instruction, so it's pretty fast.

Ok, yes I see that now for ARMv5 onwards. Grant has pushed the patch,
but may be we can update this as an optimisation separately.

Thanks
Jon