Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <b6a96cca17832219042c90baec2442b1a5bad7b7.camel@gmail.com>
Subject: Re: [PATCH v3] Fix IRQ issue by setting IRQ_DISABLE_UNLAZY flag
From:   Nuno =?ISO-8859-1?Q?S=E1?= <noname.nuno@gmail.com>
To:     Jonathan Cameron <jic23@kernel.org>,
        Masahiro Honda <honda@mechatrax.com>
Cc:     Lars-Peter Clausen <lars@metafoo.de>,
        Michael Hennerich <Michael.Hennerich@analog.com>,
        linux-iio@vger.kernel.org, linux-kernel@vger.kernel.org
Date:   Mon, 24 Apr 2023 11:09:34 +0200
In-Reply-To: <20230423121528.259d3cd4@jic23-huawei>
References: <20230420102316.757-1-honda@mechatrax.com>
         <20230423121528.259d3cd4@jic23-huawei>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
User-Agent: Evolution 3.46.4 
MIME-Version: 1.0
Precedence: bulk

Hi Jonathan,

On Sun, 2023-04-23 at 12:15 +0100, Jonathan Cameron wrote:
> On Thu, 20 Apr 2023 19:23:16 +0900
> Masahiro Honda <honda@mechatrax.com> wrote:
>=20
> > The Sigma-Delta ADCs supported by this driver can use SDO as an interru=
pt
> > line to indicate the completion of a conversion. However, some devices
> > cannot properly detect the completion of a conversion by an interrupt.
> > This is for the reason mentioned in the following commit.
> >=20
> > commit e9849777d0e2 ("genirq: Add flag to force mask in
> > =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=
=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 disable_irq[_nosy=
nc]()")
> >=20
> > A read operation is performed by an extra interrupt before the completi=
on
> > of a conversion. This patch fixes the issue by setting IRQ_DISABLE_UNLA=
ZY
> > flag.
> >=20
> > Signed-off-by: Masahiro Honda <honda@mechatrax.com>
> > ---
> > v3:
> > =C2=A0- Remove the Kconfig option.
> > v2:
> > https://lore.kernel.org/linux-iio/20230414102744.150-1-honda@mechatrax.=
com/
> > =C2=A0- Rework commit message.
> > =C2=A0- Add a new entry in the Kconfig.
> > =C2=A0- Call irq_clear_status_flags(irq, IRQ_DISABLE_UNLAZY) when freei=
ng the
> > IRQ.
> > v1:
> > https://lore.kernel.org/linux-iio/20230306044737.862-1-honda@mechatrax.=
com/
> >=20
> > =C2=A0drivers/iio/adc/ad_sigma_delta.c | 25 ++++++++++++++++++++-----
> > =C2=A01 file changed, 20 insertions(+), 5 deletions(-)
> >=20
> > diff --git a/drivers/iio/adc/ad_sigma_delta.c
> > b/drivers/iio/adc/ad_sigma_delta.c
> > index d8570f620..215ecbedb 100644
> > --- a/drivers/iio/adc/ad_sigma_delta.c
> > +++ b/drivers/iio/adc/ad_sigma_delta.c
> > @@ -565,6 +565,14 @@ int ad_sd_validate_trigger(struct iio_dev *indio_d=
ev,
> > struct iio_trigger *trig)
> > =C2=A0}
> > =C2=A0EXPORT_SYMBOL_NS_GPL(ad_sd_validate_trigger, IIO_AD_SIGMA_DELTA);
> > =C2=A0
> > +static void ad_sd_free_irq(void *sd)
> > +{
> > +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0struct ad_sigma_delta *sigma=
_delta =3D sd;
> > +
> > +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0irq_clear_status_flags(sigma=
_delta->spi->irq, IRQ_DISABLE_UNLAZY);
> > +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0free_irq(sigma_delta->spi->i=
rq, sigma_delta);
> > +}
>=20
> Don't fuse the two operations unwinding like this.=C2=A0 Just register a =
callback
> that only
> does the irq_clear_status_flags immediately after setting them.=C2=A0 The=
n leave

I was the one to propose fusing them together because I thought that we cou=
ld
have issues by clearing the flag after calling free_irq(). After looking ag=
ain
at the IRQ code, I can see that it is not up to free_irq() to free the allo=
cated
irq_descs (that might only happen when unmapping the virq) which means we s=
hould
be fine doing the normal way.

That said, looking at the only users that care to clear this flag, it looks=
 like
they do it before calling free_irq(). Hence, I'm not sure if there's anythi=
ng
subtle going on. In fact, looking at this line:

https://elixir.bootlin.com/linux/latest/source/kernel/irq/manage.c#L1909

I'm not so sure we actually need to clear the flag as for these devices, we
should only have one consumer/action per IRQ. Anyways, probably for correct=
ness
we should still explicitly clear it?

- Nuno S=C3=A1