Received-SPF: pass (google.com: domain of linux-kernel+bounces-6901-linux.lists.archive=gmail.com@vger.kernel.org designates 139.178.88.99 as permitted sender) client-ip=139.178.88.99;
Precedence: bulk
MIME-Version: 1.0
References: <20231219125706.23284-1-brgl@bgdev.pl> <ZYGyumQXaM8PXzcz@smile.fi.intel.com>
 <CAMRc=McFsswSjCU5m+_wmmxQh+T1DkwnrBsYp_KsQfOSwENvLw@mail.gmail.com>
In-Reply-To: <CAMRc=McFsswSjCU5m+_wmmxQh+T1DkwnrBsYp_KsQfOSwENvLw@mail.gmail.com>
From: Bartosz Golaszewski <brgl@bgdev.pl>
Date: Wed, 20 Dec 2023 13:35:37 +0100
Message-ID: <CAMRc=MesmR+=gMsk2y0Y7jiwtFpcHCofYsi3mxWFrdY9=1Y46A@mail.gmail.com>
Subject: Re: [PATCH] gpiolib: use __counted_by() for GPIO descriptors
To: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Cc: Linus Walleij <linus.walleij@linaro.org>, Kent Gibson <warthog618@gmail.com>, 
	linux-gpio@vger.kernel.org, linux-kernel@vger.kernel.org, 
	Bartosz Golaszewski <bartosz.golaszewski@linaro.org>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Wed, Dec 20, 2023 at 9:16=E2=80=AFAM Bartosz Golaszewski <brgl@bgdev.pl>=
 wrote:
>
> On Tue, Dec 19, 2023 at 4:12=E2=80=AFPM Andy Shevchenko
> <andriy.shevchenko@linux.intel.com> wrote:
> >
> > On Tue, Dec 19, 2023 at 01:57:06PM +0100, Bartosz Golaszewski wrote:
> > > From: Bartosz Golaszewski <bartosz.golaszewski@linaro.org>
> > >
> > > Pull the array of GPIO descriptors into struct gpio_device as a flexi=
ble
> > > array and use __counted_by() to control its size.
> >
> > How big is the struct gpio_device? Unifying like this might provoke sub=
tle
> > errors on very fragmented memory, where k*alloc() might not find enough=
 free
> > space. Note, k*alloc() guarantees finding only memory for a single page=
.
> > With PAGE_SIZE =3D 4k, this might be an issue.
> >
> > I would suggest, if nothing prevents us from switching, to use kvmalloc=
().
> >
>
> That's a good point but there's another thing. We need to call
> gpiochip_get_ngpios() before the allocation. I need to revisit this
> one.
>

It's trickier than I thought. We need the struct device to exist (have
its software node assigned) before we check the property but we cannot
allocate gdev (embedding struct device) + descs before we have read
it. Eh, maybe it's not worth it and let's keep the two allocations
separate.

Bart

> Bart
>
> > --
> > With Best Regards,
> > Andy Shevchenko
> >
> >