Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.34 as permitted sender) client-ip=23.128.96.34;
MIME-Version: 1.0
References: <20231212054253.50094-1-warthog618@gmail.com> <CAMRc=Me90Lu7Duc8-4xSfDcHQd6M7+0t0O8FAa6jiizp-OO5=Q@mail.gmail.com>
 <ZXjzjOtKFoMRhKA-@rigel>
In-Reply-To: <ZXjzjOtKFoMRhKA-@rigel>
From:   Bartosz Golaszewski <brgl@bgdev.pl>
Date:   Wed, 13 Dec 2023 11:03:40 +0100
Message-ID: <CAMRc=Me1czOqnJUG3sth6kZh=G+iXAHp7HHL1u-Oy3=MwkCPug@mail.gmail.com>
Subject: Re: [PATCH 0/4] gpiolib: cdev: relocate debounce_period_us
To:     Kent Gibson <warthog618@gmail.com>
Cc:     linux-kernel@vger.kernel.org, linux-gpio@vger.kernel.org,
        linus.walleij@linaro.org, andy@kernel.org
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Wed, Dec 13, 2023 at 12:58=E2=80=AFAM Kent Gibson <warthog618@gmail.com>=
 wrote:
>
> On Tue, Dec 12, 2023 at 06:09:00PM +0100, Bartosz Golaszewski wrote:

[snip]

> >
> > Patches 2-4 look fine, I was about to review patch 1 in detail but I
> > thought I'd just throw this one in here before we commit to a specific
> > solution.
> >
> > For some reason I thought this would not work but I'm now considering
> > it as an alternative approach: is there anything wrong with adding
> > struct kref to struct line, allocating it separately per-line when
> > gpio_chardev_data is created, referencing it from struct linereq when
> > the line is being requested, and dropping the reference from
> > gpio_chardev_data and linereq when either is being removed? Other than
> > the increased number of allocations?
> >
>
> The collection of struct line always has to be global, right, as both
> gpio_chardev_data and linereq are ephemeral.  e.g. if one process request=
s
> a line and another checks the lineinfo, those will have distinct
> gpio_chardev_data.
>

Strictly speaking at least the supplemental info has to be globally
accessible. But I get your point.

> But the key issue is that the linereq and struct line lifetimes are
> strictly tied - a struct line does not live beyond the containing linereq=
.

I was thinking about decoupling one from the other actually.

> Leaving the struct line alive after the linereq is released is just wrong=
.
> The line has been released back to gpiolib so there can be no
> supplemental info left.

Indeed.

> If you want any such info to persist beyond the line release then it
> should be located in gpiolib itself, not cdev.
>

Right, we even zero debounce_period_us anyway on line release - just
as if we released struct line.

Bart