[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20200118112906.0ad6dd6b@archlinux>
Date: Sat, 18 Jan 2020 11:29:06 +0000
From: Jonathan Cameron <jic23@...nel.org>
To: "Hennerich, Michael" <Michael.Hennerich@...log.com>
Cc: "Bia, Beniamin" <Beniamin.Bia@...log.com>,
"lars@...afoo.de" <lars@...afoo.de>,
"pmeerw@...erw.net" <pmeerw@...erw.net>,
"linux-iio@...r.kernel.org" <linux-iio@...r.kernel.org>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
"biabeniamin@...look.com" <biabeniamin@...look.com>,
"knaack.h@....de" <knaack.h@....de>
Subject: Re: [PATCH] iio: frequency: adf4371: Fix divide by zero exception
bug
On Tue, 14 Jan 2020 16:19:28 +0000
"Hennerich, Michael" <Michael.Hennerich@...log.com> wrote:
> > -----Original Message-----
> > From: Jonathan Cameron <jic23@...nel.org>
> > Sent: Samstag, 11. Januar 2020 12:09
> > To: Bia, Beniamin <Beniamin.Bia@...log.com>
> > Cc: lars@...afoo.de; Hennerich, Michael <Michael.Hennerich@...log.com>;
> > pmeerw@...erw.net; linux-iio@...r.kernel.org; linux-
> > kernel@...r.kernel.org; biabeniamin@...look.com; knaack.h@....de
> > Subject: Re: [PATCH] iio: frequency: adf4371: Fix divide by zero exception bug
> >
> > On Tue, 7 Jan 2020 15:15:59 +0200
> > Beniamin Bia <beniamin.bia@...log.com> wrote:
> >
> > > From: Michael Hennerich <michael.hennerich@...log.com>
> > >
> > > During initialization adf4371_pll_fract_n_get_rate() is called on all
> > > output channels to determine if the device was setup. In this case
> > > mod2 is zero which can cause a divide by zero exception.
> > > Return before that can happen.
> > I'm confused by this description vs the code.
> >
> > As far as I can see fract_n_get_rate is only called on a sysfs read of the
> > frequency.
>
> That's not the case. The failure mechanism comes via adf4371_channel_config().
> It calls adf4371_pll_fract_n_get_rate() prior adf4371_set_freq() which initializes
> st->mod2 via adf4371_pll_fract_n_compute(). This only happens the first time
> during probe and setup. So the solution was to return 0 if st->mod2 == 0.
Not in mainline it doesn't. I took a look at the ADI git hub and I see it does
there. I think we are missing a precursor patch.
Probably this one from blame...
https://github.com/analogdevicesinc/linux/commit/c8e6b341abf749f78e00326cb92b365d90d9de1f
Jonathan
>
> -Michael
>
> >
> > mod2 is set when fract_n_compute is called in the relevant set_freq calls.
> > This seems to occur on a sysfs set frequency call.
> >
> > So the issue here is that a sysfs read before a write of the frequency will cause a
> > div zero? If so, is there a sane set of initial values we can put in mod2 and
> > friends before exposing them via the device register?
> >
> > If mod2==0 is a valid value and indicates for example that the channel is turned
> > off, then the description should make that clear.
> >
> > Jonathan
> >
> > >
> > > Fixes: 7f699bd149134 ("iio: frequency: adf4371: Add support for
> > > ADF4371 PLL")
> > > Signed-off-by: Michael Hennerich <michael.hennerich@...log.com>
> > > Signed-off-by: Beniamin Bia <beniamin.bia@...log.com>
> > > ---
> > > drivers/iio/frequency/adf4371.c | 3 +++
> > > 1 file changed, 3 insertions(+)
> > >
> > > diff --git a/drivers/iio/frequency/adf4371.c
> > > b/drivers/iio/frequency/adf4371.c index e2a599b912e5..c21462238314
> > > 100644
> > > --- a/drivers/iio/frequency/adf4371.c
> > > +++ b/drivers/iio/frequency/adf4371.c
> > > @@ -191,6 +191,9 @@ static unsigned long long
> > adf4371_pll_fract_n_get_rate(struct adf4371_state *st,
> > > unsigned long long val, tmp;
> > > unsigned int ref_div_sel;
> > >
> > > + if (st->mod2 == 0)
> > > + return 0;
> > > +
> > > val = (((u64)st->integer * ADF4371_MODULUS1) + st->fract1) * st-
> > >fpfd;
> > > tmp = (u64)st->fract2 * st->fpfd;
> > > do_div(tmp, st->mod2);
>
Powered by blists - more mailing lists