Re: [PATCH v3 2/2] Add GPIO support for Realtek DHC(Digital Home Center) RTD SoCs.

[Andy Shevchenko]

+Cc: Michael (GPIO regmap maintainer)

On Thu, Dec 14, 2023 at 07:57:55AM +0000, TY_Chang[張子逸] wrote:
> >On Tue, Dec 12, 2023 at 09:55:59AM +0000, TY_Chang[張子逸] wrote:
> >> >On Thu, Dec 07, 2023 at 06:07:23PM +0800, TY Chang wrote:

...

> >> >> This driver enables configuration of GPIO direction, GPIO values,
> >> >> GPIO debounce settings and handles GPIO interrupts.
> >> >
> >> >Why gpio-regmap can't be used?
> >>
> >> I will try to use gpio-remap in the next version.
> >
> >If it appears that it makes code uglier / complicated, please add the note
> >somewhere to answer the above question.
> 
> I've traced the gpio-regmap.c file. It appears that for the driver to register
> gpio_irq_chip, it must create the irq_domain and add it into gpio_regmap_config.
> Additionally, the driver needs to register the irq handler by itself.
> However, this process can be managed by the gpiolib if the driver fills in the struct
> gpio_irq_chip inside struct gpio_chip before invoking gpiochip_add_data.

Hmm... I thought this is solvable issue.
Michael, is there a limitation in GPIO regmap that this driver can't be converted?

> Moreover, apart from managing the registers for gpio direction and value, there
> are several other registers that require access(interrupt enable, debounce...).
> The GPIO IRQ status registers are located at different base addresses and are
> not contiguous. It may need to create an additional regmap and assign the access
> table to this regmap.

AFAIK this is not a problem as you can provide your own xlate function that
will take care about register mapping.

> With the above consideration, I tend to keep using the existing method.

I would like to hear from Michael if it's indeed a big obstacle.

...

> >> >> +     if (irq == data->irqs[0])
> >> >> +             get_reg_offset = &rtd_gpio_gpa_offset;
> >> >> +     else if (irq == data->irqs[1])
> >> >> +             get_reg_offset = &rtd_gpio_gpda_offset;
> >> >
> >> >Can't it be done before entering into chained IRQ handler?
> >>
> >> I will revise it.
> >
> >Thinking about this more, perhaps you can register two IRQ chips with
> >different functions, so this won't be part of the very critical interrupt
> >handler (as we all want to reduce overhead in it as much as possible).
> >Anyway, think about this and try different options, choose the one you think
> >the best.
> 
> In the previous patch (v1), I had registered two IRQ chips with different
> handlers. However, these two handlers appeared quite similar and the
> gpio_irq_chip only allows the registration of a single handler. Therefore,
> I ended up registering one handler for both IRQs and included conditional
> checks within the handler to differentiate between the two.

What is the performance impact that you have that condition in the interrupt
handler?

-- 
With Best Regards,
Andy Shevchenko