[PATCH v1] gpio: mmio: restroe get multiple gpio mask

[PATCH v1] gpio: mmio: restroe get multiple gpio mask

[Yan Wang]

Simplify the code,should not modify its logic.

Fixes: 761b5c30c206 ("gpio: mmio: replace open-coded for_each_set_bit()")
Signed-off-by: Yan Wang <rk.code@outlook.com>
---
 drivers/gpio/gpio-mmio.c | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)

diff --git a/drivers/gpio/gpio-mmio.c b/drivers/gpio/gpio-mmio.c
index d9dff3dc92ae..c2347ef3a4df 100644
--- a/drivers/gpio/gpio-mmio.c
+++ b/drivers/gpio/gpio-mmio.c
@@ -271,10 +271,14 @@ static void bgpio_multiple_get_masks(struct gpio_chip *gc,
 	*clear_mask = 0;
 
 	for_each_set_bit(i, mask, gc->bgpio_bits) {
-		if (test_bit(i, bits))
-			*set_mask |= bgpio_line2mask(gc, i);
-		else
-			*clear_mask |= bgpio_line2mask(gc, i);
+		if (*mask == 0)
+			break;
+		if (__test_and_clear_bit(i, mask)) {
+			if (test_bit(i, bits))
+				*set_mask |= bgpio_line2mask(gc, i);
+			else
+				*clear_mask |= bgpio_line2mask(gc, i);
+		}
 	}
 }
 
-- 
2.17.1

Re: [PATCH v1] gpio: mmio: restroe get multiple gpio mask

[Linus Walleij]

Hi Yan,

thanks for your patch!

On Sun, Apr 23, 2023 at 11:07 AM Yan Wang <rk.code@outlook.com> wrote:

> Simplify the code,should not modify its logic.

I don't see how it simplifies the code, it seems to me that it is
making it more complex?

> Fixes: 761b5c30c206 ("gpio: mmio: replace open-coded for_each_set_bit()")
> Signed-off-by: Yan Wang <rk.code@outlook.com>
> ---
>  drivers/gpio/gpio-mmio.c | 12 ++++++++----
>  1 file changed, 8 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/gpio/gpio-mmio.c b/drivers/gpio/gpio-mmio.c
> index d9dff3dc92ae..c2347ef3a4df 100644
> --- a/drivers/gpio/gpio-mmio.c
> +++ b/drivers/gpio/gpio-mmio.c
> @@ -271,10 +271,14 @@ static void bgpio_multiple_get_masks(struct gpio_chip *gc,
>         *clear_mask = 0;
>
>         for_each_set_bit(i, mask, gc->bgpio_bits) {
> -               if (test_bit(i, bits))
> -                       *set_mask |= bgpio_line2mask(gc, i);
> -               else
> -                       *clear_mask |= bgpio_line2mask(gc, i);
> +               if (*mask == 0)
> +                       break;
> +               if (__test_and_clear_bit(i, mask)) {
> +                       if (test_bit(i, bits))
> +                               *set_mask |= bgpio_line2mask(gc, i);
> +                       else
> +                               *clear_mask |= bgpio_line2mask(gc, i);
> +               }

The intention of the change seems to be to break out of the loop
when all set bits are handled, by successively masking off one
bit at a time from mask. So this is intended as an optimization,
not a simplification.

I think for_each_set_bit() is already skipping over every bit
that is zero, see include/linux/find.h.

So this optimization gains us nothing.

Yours,
Linus Walleij

Re: [PATCH v1] gpio: mmio: restroe get multiple gpio mask

[Yan Wang]

On 4/24/2023 2:56 PM, Linus Walleij wrote:
> Hi Yan,
>
> thanks for your patch!
>
> On Sun, Apr 23, 2023 at 11:07 AM Yan Wang <rk.code@outlook.com> wrote:
>
>> Simplify the code,should not modify its logic.
> I don't see how it simplifies the code, it seems to me that it is
> making it more complex?
yes ,it is .
The description is wrong.
>> Fixes: 761b5c30c206 ("gpio: mmio: replace open-coded for_each_set_bit()")
>> Signed-off-by: Yan Wang <rk.code@outlook.com>
>> ---
>>   drivers/gpio/gpio-mmio.c | 12 ++++++++----
>>   1 file changed, 8 insertions(+), 4 deletions(-)
>>
>> diff --git a/drivers/gpio/gpio-mmio.c b/drivers/gpio/gpio-mmio.c
>> index d9dff3dc92ae..c2347ef3a4df 100644
>> --- a/drivers/gpio/gpio-mmio.c
>> +++ b/drivers/gpio/gpio-mmio.c
>> @@ -271,10 +271,14 @@ static void bgpio_multiple_get_masks(struct gpio_chip *gc,
>>          *clear_mask = 0;
>>
>>          for_each_set_bit(i, mask, gc->bgpio_bits) {
>> -               if (test_bit(i, bits))
>> -                       *set_mask |= bgpio_line2mask(gc, i);
>> -               else
>> -                       *clear_mask |= bgpio_line2mask(gc, i);
>> +               if (*mask == 0)
>> +                       break;
>> +               if (__test_and_clear_bit(i, mask)) {
>> +                       if (test_bit(i, bits))
>> +                               *set_mask |= bgpio_line2mask(gc, i);
>> +                       else
>> +                               *clear_mask |= bgpio_line2mask(gc, i);
>> +               }
> The intention of the change seems to be to break out of the loop
> when all set bits are handled, by successively masking off one
> bit at a time from mask. So this is intended as an optimization,
> not a simplification.
Because my description is wrong and there is a difference in 
understanding between us.
I think using for_each_set_bit() , but the __test_and_clear_bit() should 
not remove.
> I think for_each_set_bit() is already skipping over every bit
> that is zero, see include/linux/find.h.
>
> So this optimization gains us nothing.
this a patch that only restores to original logic.the 
__test_and_clear_bit() clear mask
then get a new set_mask. this logic can affect functions of 
gpiod_set_raw_array_value ().

The effect of this patch is as follows:
gpiod_set_raw_array_value->
     gpiod_set_array_value_complex->
{
     ..
     if (array_info && array_info->desc == desc_array &&
         array_size <= array_info->size &&
         (void *)array_info == desc_array + array_info->size) {
         if (!can_sleep)
             WARN_ON(array_info->chip->can_sleep);

         if (!raw && !bitmap_empty(array_info->invert_mask, array_size))
             bitmap_xor(value_bitmap, value_bitmap,
                    array_info->invert_mask, array_size);

         gpio_chip_set_multiple(array_info->chip, array_info->set_mask,
                        value_bitmap);

         i = find_first_zero_bit(array_info->set_mask, array_size);
         if (i == array_size)
             return 0;
     } else {
         array_info = NULL;
     }

    -> while (i < array_size) {
         struct gpio_chip *gc = desc_array[i]->gdev->chip;
         DECLARE_BITMAP(fastpath_mask, FASTPATH_NGPIO);
         DECLARE_BITMAP(fastpath_bits, FASTPATH_NGPIO);
         unsigned long *mask, *bits;
         int count = 0;

          ..............
         if (count != 0)
             gpio_chip_set_multiple(gc, mask, bits);

         ..............
     }
     return 0;
}

Due to __test_and_clear_bit() clear array_info->set_mask,the i < 
array_size condition holds.
then calculate a new mask based on the GPIO number of the hardware.

I reconfirmed it,Although it works, it should not be modified in the place.
I think have a wrong of gpiod_get_array() and not 
bgpio_multiple_get_masks().

Link: 
https://lore.kernel.org/linux-gpio/KL1PR01MB5448327326B6EDA8001AF714E6669@KL1PR01MB5448.apcprd01.prod.exchangelabs.com/

this is a new patch. can you take a look ?

Thank you.
> Yours,
> Linus Walleij

Re: [PATCH v1] gpio: mmio: restroe get multiple gpio mask

[Andy Shevchenko]

On Sun, Apr 23, 2023 at 05:06:48PM +0800, Yan Wang wrote:
> Simplify the code,should not modify its logic.

> Fixes: 761b5c30c206 ("gpio: mmio: replace open-coded for_each_set_bit()")

What does it fix?

...

>  	for_each_set_bit(i, mask, gc->bgpio_bits) {
> -		if (test_bit(i, bits))
> -			*set_mask |= bgpio_line2mask(gc, i);
> -		else
> -			*clear_mask |= bgpio_line2mask(gc, i);
> +		if (*mask == 0)
> +			break;

Huh?!

We never enter here if mask is 0. So, do not add a dead code, please.

Moreover, in principle mask can be longer than 1 long, this code simply wrong.

NAK

> +		if (__test_and_clear_bit(i, mask)) {
> +			if (test_bit(i, bits))
> +				*set_mask |= bgpio_line2mask(gc, i);
> +			else
> +				*clear_mask |= bgpio_line2mask(gc, i);
> +		}
>  	}

-- 
With Best Regards,
Andy Shevchenko

Re: [PATCH v1] gpio: mmio: restroe get multiple gpio mask

[Yan Wang]

> On Apr 24, 2023, at 21:31, Andy Shevchenko <andriy.shevchenko@linux.intel.com> wrote:
> 
> On Sun, Apr 23, 2023 at 05:06:48PM +0800, Yan Wang wrote:
>> Simplify the code,should not modify its logic.
> 
>> Fixes: 761b5c30c206 ("gpio: mmio: replace open-coded for_each_set_bit()")
> 
> What does it fix?
> 
> ...
> 
>> 	for_each_set_bit(i, mask, gc->bgpio_bits) {
>> -		if (test_bit(i, bits))
>> -			*set_mask |= bgpio_line2mask(gc, i);
>> -		else
>> -			*clear_mask |= bgpio_line2mask(gc, i);
>> +		if (*mask == 0)
>> +			break;
> 
> Huh?!
> 
> We never enter here if mask is 0. So, do not add a dead code, please.
> 
> Moreover, in principle mask can be longer than 1 long, this code simply wrong.

You are right.

Because  I use gpiod_set_array_value_cansleep() to set multiple gpios
occur wrong . I restored logic of the original code and found it to be effective. 
So,I try to modify it.

I  recheck logic of  this position that it’s correct. I think  there should be a error in 
Gpiolib. 


> NAK
> 
>> +		if (__test_and_clear_bit(i, mask)) {
>> +			if (test_bit(i, bits))
>> +				*set_mask |= bgpio_line2mask(gc, i);
>> +			else
>> +				*clear_mask |= bgpio_line2mask(gc, i);
>> +		}
>> 	}
> 
> -- 
> With Best Regards,
> Andy Shevchenko
> 
>