Re: Is consumer documentation correct?

[Andy Shevchenko <andy.shevchenko@xxxxxxxxx>]

On Wed, Oct 28, 2020 at 6:15 PM Linus Walleij <linus.walleij@xxxxxxxxxx> wrote:
>
> On Thu, Oct 22, 2020 at 6:26 PM Andy Shevchenko
> <andy.shevchenko@xxxxxxxxx> wrote:
>
> > It _seems_ between
> >   79a9becda894 ("gpiolib: export descriptor-based GPIO interface")
> > and
> >   14e85c0e69d5 ("gpio: remove gpio_descs global array")
> > the "Interacting With the Legacy GPIO Subsystem" of GPIO consumer
> > interface is correct.
> >
> > But is it the case afterwards?
> >
> > Details of the question are here [1].
> >
> > [1]: https://stackoverflow.com/q/64455505/2511795
>
> I don't know if I even understand the question right, but gpio_to_desc()
> can be called at any time as long as there is a struct gpio_chip with
> the global GPIO number assigned to it. It will just index into the
> array of descriptors stored inside the chip and return a pointer.
> The allocation of the memory lives with the struct gpio_chip.
>
> If for example that chip is on USB and you unplug that USB
> cable, naturally referencing the descriptor after that point
> will cause disasters.
>
> The gpio_request()/free() doesn't really allocate or free
> any memory at all. This is why we call gpio_to_desc() first
> thing in the legacy functions.

The section titled "Interacting With the Legacy GPIO Subsystem"
describes as far as I got it the interaction of gpio_to_desc() and
desc_to_gpio() with new / legacy APIs along with their scope. But is
that description correct after the above mentioned commit, i.e.
14e85c0e69d5 ("gpio: remove gpio_descs global array") when we lose the
always-present data structure (if I'm not mistaken)?

-- 
With Best Regards,
Andy Shevchenko