Re: [PATCH v4 17/17] staging: iio: resolver: ad2s1210: simplify code with guard(mutex)

From: David Lechner <dlechner@baylibre.com>
To: Jonathan Cameron <jic23@kernel.org>
Cc: linux-iio@vger.kernel.org, linux-staging@lists.linux.dev,
	"Michael Hennerich" <Michael.Hennerich@analog.com>,
	"Nuno Sá" <nuno.sa@analog.com>,
	"Axel Haslam" <ahaslam@baylibre.com>,
	"Philip Molloy" <pmolloy@baylibre.com>,
	linux-kernel@vger.kernel.org
Subject: Re: [PATCH v4 17/17] staging: iio: resolver: ad2s1210: simplify code with guard(mutex)
Date: Tue, 10 Oct 2023 12:40:03 -0500	[thread overview]
Message-ID: <CAMknhBFP1Or+06rAMNOAU1Dc3sx1QgO44-N5xpsE-54DVOHYSQ@mail.gmail.com> (raw)
In-Reply-To: <20231010171738.5a23e66e@jic23-huawei>

On Tue, Oct 10, 2023 at 11:17 AM Jonathan Cameron <jic23@kernel.org> wrote:
>
> On Thu,  5 Oct 2023 19:50:34 -0500
> David Lechner <dlechner@baylibre.com> wrote:
>
> > We can simplify the code and get rid of most of the gotos by using
> > guard(mutex) from cleanup.h.
> You could consider scoped_guard() for a few cases in here, but perhaps
> it's better to be consistent and always use the guard() version.

Yes, there it doesn't look like there are any cases where there is any
long-running operation that could be done after unlocking the mutex,
so I went with the simpler approach everywhere.

>
> There is a small timing question wrt to the gpio manipulation inline.
>
> >
> > Signed-off-by: David Lechner <dlechner@baylibre.com>
> > ---
> >
> > v4 changes: New patch in v4.
> >
> >  drivers/staging/iio/resolver/ad2s1210.c | 157 ++++++++++----------------------
> >  1 file changed, 50 insertions(+), 107 deletions(-)
> >
> > diff --git a/drivers/staging/iio/resolver/ad2s1210.c b/drivers/staging/iio/resolver/ad2s1210.c
> > index c4e1bc22e8b0..c4e0ffa92dc2 100644
> > --- a/drivers/staging/iio/resolver/ad2s1210.c
> > +++ b/drivers/staging/iio/resolver/ad2s1210.c
> > @@ -47,6 +47,7 @@
> >
> >  #include <linux/bitfield.h>
> >  #include <linux/bits.h>
> > +#include <linux/cleanup.h>
> >  #include <linux/clk.h>
> >  #include <linux/delay.h>
> >  #include <linux/device.h>
> > @@ -404,11 +405,13 @@ static int ad2s1210_single_conversion(struct iio_dev *indio_dev,
> >       s64 timestamp;
> >       int ret;
> >
> > -     mutex_lock(&st->lock);
> > +     guard(mutex)(&st->lock);
> > +
> >       gpiod_set_value(st->sample_gpio, 1);
> >       timestamp = iio_get_time_ns(indio_dev);
> >       /* delay (6 * tck + 20) nano seconds */
> >       udelay(1);
> > +     gpiod_set_value(st->sample_gpio, 0);
> >
> >       switch (chan->type) {
> >       case IIO_ANGL:
> > @@ -418,14 +421,13 @@ static int ad2s1210_single_conversion(struct iio_dev *indio_dev,
> >               ret = ad2s1210_set_mode(st, MOD_VEL);
> >               break;
> >       default:
> > -             ret = -EINVAL;
> > -             break;
> > +             return -EINVAL;
> >       }
> >       if (ret < 0)
> > -             goto error_ret;
> > +             return ret;
> >       ret = spi_read(st->sdev, &st->sample, 3);
> >       if (ret < 0)
> > -             goto error_ret;
> > +             return ret;
> >
> >       switch (chan->type) {
> >       case IIO_ANGL:
> > @@ -437,17 +439,11 @@ static int ad2s1210_single_conversion(struct iio_dev *indio_dev,
> >               ret = IIO_VAL_INT;
> >               break;
> >       default:
> > -             ret = -EINVAL;
> > -             break;
> > +             return -EINVAL;
> >       }
> >
> >       ad2s1210_push_events(indio_dev, st->sample.fault, timestamp);
> >
> > -error_ret:
> > -     gpiod_set_value(st->sample_gpio, 0);
> > -     /* delay (2 * tck + 20) nano seconds */
> > -     udelay(1);
>
> Dropping this delay isn't obviously safe (though it probably is given stuff done before we exit).
> I assume there are no rules on holding the gpio down for the register read.

Correct. The SAMPLE gpio only needs to be held for a short time (~350
nanoseconds) to latch in the current values, then it doesn't matter
when it is released. (Figure 35 in datasheet)

>
> If nothing else I think the patch description needs to made an argument for why it is fine.

The longest possible delay needed after releasing the SAMPLE line
before reasserting is ~350 nanoseconds. Is there a rule of thumb for
deciding when there are enough instructions that no processor could
execute them faster than this vs. when we should add an explicit
delay?

I think I will consider adding a patch in the next round to refactor
the SAMPLE toggle to a separate function so we can be sure it is
handled the same in all cases.

>
> > -     mutex_unlock(&st->lock);
> >       return ret;
> >  }
> >