Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
References: <20190809030352.8387-1-hslester96@gmail.com> <20190809151114.GD3963@sirena.co.uk>
 <CANhBUQ09+q9_=7nMs63w4KRLGOhW1=z-AnuwOzAnUrWRY6uC6A@mail.gmail.com> <20190812110719.GE4592@sirena.co.uk>
In-Reply-To: <20190812110719.GE4592@sirena.co.uk>
From:   Chuhong Yuan <hslester96@gmail.com>
Date:   Mon, 12 Aug 2019 20:51:25 +0800
Message-ID: <CANhBUQ2XWAJBgfzbiiffaJ60wnoP__kDPHOF4d+Z_1b1HzSpPQ@mail.gmail.com>
Subject: Re: [PATCH] regulator: core: Add devres versions of regulator_enable/disable
To:     Mark Brown <broonie@kernel.org>
Cc:     Liam Girdwood <lgirdwood@gmail.com>,
        LKML <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Mon, Aug 12, 2019 at 7:07 PM Mark Brown <broonie@kernel.org> wrote:
>
> On Sat, Aug 10, 2019 at 09:44:45AM +0800, Chuhong Yuan wrote:
> > On Fri, Aug 9, 2019 at 11:11 PM Mark Brown <broonie@kernel.org> wrote:
>
> > > I'm not super keen on managed versions of these functions since they're
> > > very likely to cause reference counting issues between the probe/remove
> > > path and the suspend/resume path which aren't obvious from the code, I'm
> > > especially worried about double frees on release.
>
> > I find that 29 of 31 cases I found call regulator_disable() only when encounter
> > probe failure or in .remove.
> > So I think the devm versions of regulator_enable/disable() will not cause big
> > problems.
>
> There's way more drivers using regulators than that...
>

I wrote a new coccinelle script to detect all regulator_disable() in .remove,
101 drivers are found in total.
Within them, 25 drivers cannot benefit from devres version of regulator_enable()
since they have additional non-devm operations after
regulator_disable() in .remove.
Within the left 76 cases, 60 drivers (79%) only use
regulator_disable() when encounter
probe failure or in .remove.
The left 16 cases mostly use regulator_disable() in _suspend().
Furthermore, 3 cases of 76 are found to forget to disable regulator
when fail in probe.
So I think a devres version of regulator_enable/disable() has more
benefits than potential
risk.

> > I even found a driver to forget to disable regulator when encounter
> > probe failure,
> > which is drivers/iio/adc/ti-adc128s052.c.
> > And a devm version of regulator_enable() can prevent such mistakes.
>
> Yes, it's useful for that.