Re: [PATCH v5 3/5] iio:temperature:mlx90632: Convert polling while loop to do-while

From: Andy Shevchenko <andy.shevchenko@gmail.com>
To: Crt Mori <cmo@melexis.com>
Cc: Jonathan Cameron <jic23@kernel.org>,
	linux-iio <linux-iio@vger.kernel.org>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH v5 3/5] iio:temperature:mlx90632: Convert polling while loop to do-while
Date: Thu, 13 Aug 2020 22:40:47 +0300	[thread overview]
Message-ID: <CAHp75VeH5SA2KeiTSN__8ndj1v_SEb7mEWPG1p2Lz-tATDWi8A@mail.gmail.com> (raw)
In-Reply-To: <CAKv63usrjEHTmYtahedqULnqu-fM3TFC8HJ=S4h+w=UTv5sd-Q@mail.gmail.com>

On Thu, Aug 13, 2020 at 4:04 PM Crt Mori <cmo@melexis.com> wrote:
> On Thu, 13 Aug 2020 at 13:24, Andy Shevchenko <andy.shevchenko@gmail.com> wrote:
> > On Thu, Aug 13, 2020 at 2:14 PM Crt Mori <cmo@melexis.com> wrote:
> > > On Thu, 13 Aug 2020 at 13:03, Andy Shevchenko <andy.shevchenko@gmail.com> wrote:
> > > > On Thu, Aug 13, 2020 at 10:53 AM Crt Mori <cmo@melexis.com> wrote:

...

> > > > I don't see how it prevents using iopoll.h. It uses usleep_range()
> > > > under the hood in the same way you did here, but open coded.
> > > >
> > >
> > > One loop is indeed 10ms and that is not the problem, the problem is
> > > that timeout is at least 3 calls of this data ready (3 channels), so
> > > that is at minimum 30ms of timeout, or it could even be 4 in worse
> > > case scenario and that is outside of the range for usleep to measure.
> > > So in case of the other loop, where we wait 200ms for channel refresh
> > > it is also out of scope. Timeout should be in number of tries or in
> > > msleep range if you ask me.
> >
> > I still didn't buy it. You have in both cases usleep_range(). Why in
> > your case it's okay and in regmap_read_poll_timeout() is not?
> >
>
> I tried and it did not work, so then I read the manual. Looking into
>
> * regmap_read_poll_timeout_atomic - Poll until a condition is met or a
> timeout occurs

Why _atomic?!

> ...
>  * @delay_us: Time to udelay between reads in us (0 tight-loops).
>  *            Should be less than ~10us since udelay is used
>  *            (see Documentation/timers/timers-howto.rst).
>  * @timeout_us: Timeout in us, 0 means never timeout
>
>
> So I went to read Documentation/timers/timers-howto.rst
>
> SLEEPING FOR ~USECS OR SMALL MSECS ( 10us - 20ms):
> * Use usleep_range
>
> - Why not msleep for (1ms - 20ms)?
> Explained originally here:
> http://lkml.org/lkml/2007/8/3/250
>
> msleep(1~20) may not do what the caller intends, and
> will often sleep longer (~20 ms actual sleep for any
> value given in the 1~20ms range). In many cases this
> is not the desired behavior.
>
> Since I am above the 20ms range, it is too much for usleep_range and
> that proved to be a case as I got -ETIMEOUT and the desired channels
> were not read.
> > > > ...
> > > >
> > > > > -       while (tries-- > 0) {
> > > > > +       do {
> > > > >                 ret = regmap_read(data->regmap, MLX90632_REG_STATUS,
> > > > >                                   &reg_status);
> > > > >                 if (ret < 0)
> > > > >                         return ret;
> > > > > -               if (reg_status & MLX90632_STAT_DATA_RDY)
> > > > > -                       break;
> > > > >                 usleep_range(10000, 11000);

You use here usleep_range(). The same is used for
regmap_read_poll_timeout(). What's the difference?

Since it uses 1/4 of the range you probably need to update tries and
timeout_us to make it work.

> > > > > -       }
> > > > > +       } while (!(reg_status & MLX90632_STAT_DATA_RDY) && tries--);
> > > > >
> > > > >         if (tries < 0) {
> > > > >                 dev_err(&data->client->dev, "data not ready");

-- 
With Best Regards,
Andy Shevchenko