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;
MIME-Version: 1.0
References: <20200623040107.22270-1-warthog618@gmail.com> <20200623040107.22270-9-warthog618@gmail.com>
 <CAMpxmJVbiH9mh2c8aAS+GG9a76w9c0Nrrv-VDVhpo+wSyQ2NeQ@mail.gmail.com> <20200624140806.GA7569@sol>
In-Reply-To: <20200624140806.GA7569@sol>
From:   Bartosz Golaszewski <bgolaszewski@baylibre.com>
Date:   Thu, 25 Jun 2020 11:44:30 +0200
Message-ID: <CAMpxmJXbzFvR9h4fXvCwDMz4t6N16o2Y88K94aPwu6MFuCzqrg@mail.gmail.com>
Subject: Re: [PATCH 08/22] gpiolib: cdev: complete the irq/thread timestamp handshake
To:     Kent Gibson <warthog618@gmail.com>
Cc:     LKML <linux-kernel@vger.kernel.org>,
        linux-gpio <linux-gpio@vger.kernel.org>,
        Linus Walleij <linus.walleij@linaro.org>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Wed, Jun 24, 2020 at 4:08 PM Kent Gibson <warthog618@gmail.com> wrote:
>
> On Wed, Jun 24, 2020 at 04:00:42PM +0200, Bartosz Golaszewski wrote:
> > wt., 23 cze 2020 o 06:02 Kent Gibson <warthog618@gmail.com> napisa=C5=
=82(a):
> > >
> > > Reset the timestamp field to 0 after using it in lineevent_irq_thread=
.
> > >
> > > The timestamp is set by lineevent_irq_handler and is tested by
> > > lineevent_irq_thread to determine if it is called from a nested thead=
ed
> > > interrupt.
> > > lineevent_irq_thread is assuming that the nested, or otherwise, statu=
s
> > > of the IRQ is static, i.e. it is either always nested or never nested=
.
> > > This change removes that assumption, resetting the timestamp so it ca=
n
> > > be re-used to determine the nested state of subsequent interrupts.
> > >
> > > Signed-off-by: Kent Gibson <warthog618@gmail.com>
> > >
> >
> > This change makes sense to me but I'm having a hard time processing
> > the explanation. If we're requesting the interrupt and allocating the
> > lineevent state in the same function - how can we run into a situation
> > here the status of the irq would change like what you describe?
> >
>
> I'm not totally sure myself, as my understanding of how interrupts are
> shared in the kernel is pretty sketchy, but my concern is that if we
> are sharing the irq then whoever we are sharing with may release the irq
> and we go from nested to unnested.  Or vice versa.  Not sure if that is
> valid, but that was my concern, and it seemed like a minor change to
> cover it just in case.
>

It's my understanding that a shared interrupt must be explicitly
requested as shared by all previous users or request_irq() will fail.
In this case: we call request_threaded_irq() without the IRQF_SHARED
flag so it's never a shared interrupt. Even if someone previously
requested it as shared - our call will simply fail.

I still think that resetting the timestamp is fine because it's not
being set to 0 in hardirq context. We just need a different
explanation.

Bart