Re: [PATCH] gpio: Allow user to customise maximum number of GPIOs

[Christophe Leroy <christophe.leroy@xxxxxxxxxx>]

Le 17/08/2022 à 19:46, Arnd Bergmann a écrit :
> On Tue, Aug 9, 2022 at 12:40 PM Christophe Leroy
> <christophe.leroy@xxxxxxxxxx> wrote:
>>
>> At the time being, the default maximum number of GPIOs is set to 512
>> and can only get customised via an architecture specific
>> CONFIG_ARCH_NR_GPIO.
>>
>> The maximum number of GPIOs might be dependent on the number of
>> interface boards and is somewhat independent of architecture.
>>
>> Allow the user to select that maximum number outside of any
>> architecture configuration. To enable that, re-define a
>> core CONFIG_ARCH_NR_GPIO for architectures which don't already
>> define one. Guard it with a new hidden CONFIG_ARCH_HAS_NR_GPIO.
>>
>> Only two architectures will need CONFIG_ARCH_HAS_NR_GPIO: x86 and arm.
>>
>> On arm, do like x86 and set 512 as the default instead of 0, that
>> allows simplifying the logic in asm-generic/gpio.h
>>
>> Signed-off-by: Christophe Leroy <christophe.leroy@xxxxxxxxxx>
>> ---
>>   Documentation/driver-api/gpio/legacy.rst |  2 +-
>>   arch/arm/Kconfig                         |  3 ++-
>>   arch/arm/include/asm/gpio.h              |  1 -
>>   arch/x86/Kconfig                         |  1 +
>>   drivers/gpio/Kconfig                     | 14 ++++++++++++++
>>   include/asm-generic/gpio.h               |  6 ------
>>   6 files changed, 18 insertions(+), 9 deletions(-)
>>
>> diff --git a/Documentation/driver-api/gpio/legacy.rst b/Documentation/driver-api/gpio/legacy.rst
>> index 9b12eeb89170..566b06a584cf 100644
>> --- a/Documentation/driver-api/gpio/legacy.rst
>> +++ b/Documentation/driver-api/gpio/legacy.rst
>> @@ -558,7 +558,7 @@ Platform Support
>>   To force-enable this framework, a platform's Kconfig will "select" GPIOLIB,
>>   else it is up to the user to configure support for GPIO.
>>
>> -It may also provide a custom value for ARCH_NR_GPIOS, so that it better
>> +It may also provide a custom value for CONFIG_ARCH_NR_GPIO, so that it better
>>   reflects the number of GPIOs in actual use on that platform, without
>>   wasting static table space.  (It should count both built-in/SoC GPIOs and
>>   also ones on GPIO expanders.
>> diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
>> index 53e6a1da9af5..e55b6560fe4f 100644
>> --- a/arch/arm/Kconfig
>> +++ b/arch/arm/Kconfig
>> @@ -14,6 +14,7 @@ config ARM
>>          select ARCH_HAS_KCOV
>>          select ARCH_HAS_MEMBARRIER_SYNC_CORE
>>          select ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
>> +       select ARCH_HAS_NR_GPIO
>>          select ARCH_HAS_PTE_SPECIAL if ARM_LPAE
>>          select ARCH_HAS_PHYS_TO_DMA
>>          select ARCH_HAS_SETUP_DMA_OPS
>> @@ -1243,7 +1244,7 @@ config ARCH_NR_GPIO
>>          default 352 if ARCH_VT8500
>>          default 288 if ARCH_ROCKCHIP
>>          default 264 if MACH_H4700
>> -       default 0
>> +       default 512
> 
> This list should be kept sorted, otherwise you still get e.g. the '264' default
> value. If you have a GPIO extender that provides hardcoded GPIO
> numbers on your machine, there should be a configuration option for
> that driver.

I don't want to change the behaviour for existing configurations. If the 
unconditional default goes before conditional ones, then all following 
defaults will be ignored and you'll get 512 instead of 264 if MAC_H4700 
is selected for instance.

At the time being, you get 0 only when no other default was selected, 
then that 0 implies 512 in asm-generic/gpio.h by:

#if defined(CONFIG_ARCH_NR_GPIO) && CONFIG_ARCH_NR_GPIO > 0
#define ARCH_NR_GPIOS CONFIG_ARCH_NR_GPIO
#else
#define ARCH_NR_GPIOS		512
#endif

> 
> Which driver is it that needs extra hardcoded GPIO numbers for you?
> Have you tried converting it to use GPIO descriptors so you don't
> need the number assignment?

It is a max7301 (drivers/gpio/gpio-max730x.c) but I can't understand 
what you mean. GPIO descriptors are for consumers, aren't they ?

During boot I get :

[    0.601942] gpiochip_find_base: found new base at 496
[    0.606337] gpiochip_find_base: found new base at 464
[    0.616408] gpiochip_find_base: found new base at 448
[    0.621826] gpiochip_find_base: found new base at 432
[    0.627228] gpiochip_find_base: found new base at 400
[    0.660984] gpiochip_find_base: found new base at 384
[    0.669631] gpiochip_find_base: found new base at 368
[    0.672713] gpiochip_find_base: found new base at 352
[    0.675805] gpiochip_find_base: found new base at 336
[    0.678885] gpiochip_find_base: found new base at 320
[    0.682178] gpiochip_find_base: found new base at 304
[    0.685275] gpiochip_find_base: found new base at 288
[    0.688366] gpiochip_find_base: found new base at 272
[    0.691678] gpiochip_find_base: found new base at 256
[    0.694762] gpiochip_find_base: found new base at 240
[    0.697847] gpiochip_find_base: found new base at 224
[    0.701441] gpiochip_find_base: found new base at 208
[    0.709427] gpiochip_find_base: found new base at 192
[    0.713859] gpiochip_find_base: found new base at 176
[    0.718002] gpiochip_find_base: found new base at 160
[    0.723316] gpiochip_find_base: found new base at 144
[    0.731105] gpiochip_find_base: found new base at 128
[    0.737403] gpiochip_find_base: found new base at 112
[    0.740614] gpiochip_find_base: found new base at 96
[    0.743701] gpiochip_find_base: found new base at 80
[    0.747246] gpiochip_find_base: found new base at 64
[    4.663677] gpiochip_find_base: found new base at 36
[    5.050792] gpiochip_find_base: found new base at 16
[    5.064892] gpiochip_find_base: cannot find free range
[    5.095527] gpiochip_find_base: cannot find free range

gpiochip_find_base() is called for any GPIO driver, by gpiochip_add() / 
gpiochip_add_data_with_key(), and there is the following comment:

	/*
	 * TODO: this allocates a Linux GPIO number base in the global
	 * GPIO numberspace for this chip. In the long run we want to
	 * get *rid* of this numberspace and use only descriptors, but
	 * it may be a pipe dream. It will not happen before we get rid
	 * of the sysfs interface anyways.
	 */

So, what did I miss ?

Thanks
Christophe