Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Mon, 25 May 2020 23:17:36 +0800
From:   Kent Gibson <warthog618@gmail.com>
To:     Linus Walleij <linus.walleij@linaro.org>
Cc:     Bartosz Golaszewski <brgl@bgdev.pl>,
        Hector Bujanda <hector.bujanda@digi.com>,
        "open list:GPIO SUBSYSTEM" <linux-gpio@vger.kernel.org>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH] gpiolib: add GPIO_SET_DEBOUNCE_IOCTL
Message-ID: <20200525151736.GA32461@sol>
References: <20200419001858.105281-1-hector.bujanda@digi.com>
 <CAMRc=MeHun_WEApEXP59ZszGa2n+wbU9qq3wU1VO9o590rO-Pw@mail.gmail.com>
 <CACRpkdaeXFW5K=Npy2ubWsffc7aepEQ5kSJ2HrkrESjaTy_psQ@mail.gmail.com>
 <20200525022252.GA22956@sol>
 <CACRpkdagkhbULGVGJqcS55m=X2EaH_iK0Khr8+6M7ATWrC3hOQ@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <CACRpkdagkhbULGVGJqcS55m=X2EaH_iK0Khr8+6M7ATWrC3hOQ@mail.gmail.com>
User-Agent: Mutt/1.9.4 (2018-02-28)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Mon, May 25, 2020 at 02:17:41PM +0200, Linus Walleij wrote:
> On Mon, May 25, 2020 at 4:22 AM Kent Gibson <warthog618@gmail.com> wrote:
> 
> > You mention timers for the gpio pins that cannot provide debounce,
> > so I'm confused. Could you clarify which strategy you have in mind?
> 
> My idea is that the callback gpiod_set_debounce() for in-kernel consumers
> should more or less always return success. Either the hardware does
> the debounce,  or gpiolib sets up a timer.
> 
> > I've also had a quick look at the possibility of providing the software
> > debounce for in-kernel consumers.
> 
> That is where I think it should start.
> 
> >  Are you anticipating new API for
> > that?  e.g. allowing consumers to request gpio events?  Cos, gpio_keys
> > grabs the irq itself - and that would bypass the software debouncer,
> > or even conflict with it.
> 
> It may be hard or impossible.
> 
> I suppose gpiolib would have to steal or intercept the interrupt
> by using e.g. IRQF_SHARED and then just return IRQ_HANDLED
> on the first IRQ so the underlying irq handler does not get called.
> 

And how would gpiolib ensure that it was first in the chain?

> After the timer times out it needs to retrigger the IRQ.
> 
> So the consuming driver would se a "debounced and ready"
> IRQ so when it gets this IRQ it knows for sure there is
> no bounciness on it because gpiolib already took care
> of that.
> 
> The above is in no way trivial, but it follows the design pattern
> of "narrow and deep" APIs.
> 

Totally agree with the concept - just trying to work out how to
implement it seemlessly given the existing API and usage, and given my
limited knowledge of the kernel internals.

> Failure is an option! Sorry if I push too complex ideas.
> 

I'm not as concerned about complexity as I am about fragility.

I don't see any problem adding debounce for gpiolib-cdev.
Adding a more complete solution to gpiolib itself is certainly
non-trivial, if it is possible at all. 

The path I'll probably be taking is adding a debouncer to gpiolib-cdev,
so at least we have a solution for userspace, then take a longer look at
the more general solution.

Cheers,
Kent.