Received-SPF: pass (google.com: domain of linux-kernel+bounces-4034-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45d1:ec00::1 as permitted sender) client-ip=2604:1380:45d1:ec00::1;
Date: Tue, 19 Dec 2023 00:08:23 +0800
From: Kent Gibson <warthog618@gmail.com>
To: Bartosz Golaszewski <brgl@bgdev.pl>
Cc: linux-kernel@vger.kernel.org, linux-gpio@vger.kernel.org,
	linus.walleij@linaro.org, andy@kernel.org
Subject: Re: [PATCH v4 1/5] gpiolib: cdev: relocate debounce_period_us from
 struct gpio_desc
Message-ID: <ZYBud4tobG3x9Zhh@rigel>
References: <20231216001652.56276-1-warthog618@gmail.com>
 <20231216001652.56276-2-warthog618@gmail.com>
 <CAMRc=McBVeQ=yRpGRsnPEULfPx15PBO3kiGscdS4s6-d0URc3w@mail.gmail.com>
 <ZYBoA25z76uutBBI@rigel>
 <CAMRc=MdKfs1ok2KgkO0C64cA9k8bOupxsjReBMQSdZbP+MQMCQ@mail.gmail.com>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <CAMRc=MdKfs1ok2KgkO0C64cA9k8bOupxsjReBMQSdZbP+MQMCQ@mail.gmail.com>

On Mon, Dec 18, 2023 at 05:01:32PM +0100, Bartosz Golaszewski wrote:
> On Mon, Dec 18, 2023 at 4:40 PM Kent Gibson <warthog618@gmail.com> wrote:
> >
> > On Mon, Dec 18, 2023 at 04:24:50PM +0100, Bartosz Golaszewski wrote:
> > > On Sat, Dec 16, 2023 at 1:17 AM Kent Gibson <warthog618@gmail.com> wrote:
> > > >

> > > > +static void line_set_debounce_period(struct line *line,
> > > > +                                    unsigned int debounce_period_us)
> > > > +{
> > > > +       bool was_suppl = line_is_supplemental(line);
> > > > +
> > > > +       WRITE_ONCE(line->debounce_period_us, debounce_period_us);
> > > > +
> > > > +       if (line_is_supplemental(line) == was_suppl)
> > > > +               return;
> > > > +
> > > > +       if (was_suppl)
> > > > +               supinfo_erase(line);
> > > > +       else
> > > > +               supinfo_insert(line);
> > >
> > > Could you add a comment here saying it's called with the config mutex
> > > taken as at first glance it looks racy but actually isn't?
> > >
> >
> > Sure.  Though it is also covered by the gdev->sem you added, right?
> > So the config_mutex is now redundant?
> > Should I document it is covered by both?
> > Or drop the config_mutex entirely?
> >
>
> No! The semaphore is a read-write semaphore and we can have multiple
> readers at once. This semaphore only protects us from the chip being
> set to NULL during a system call. It will also go away once I get the
> descriptor access serialized (or not - I'm figuring out a lockless
> approach) and finally use SRCU to protect gdev->chip.
>

Ah, so it is.

> > And you wanted some comments to explain the logic?
> > I thought this is a common "has it changed" pattern, and so didn't
> > require additional explanation, but I guess not as common as I thought.
> >
>
> If line_set_debounce_period() could be called concurrently for the
> same line, this approach would be racy. It cannot but I want a comment
> here as I fear that if in the future we add some more attributes that
> constitute "supplemental info" and which may be changed outside of the
> config lock then this would in fact become racy.
>

If any line config is going to be changed from the userspace side then
it will be by the SET_CONFIG ioctl, and so be covered by the config_mutex,
but it can't hurt to explicitly document it here as well.

Cheers,
Kent.