Re: [PATCH 4/5] iio: adc: vf610_adc: Replace indio_dev->mlock with own device lock

From: Alexandru Ardelean <ardeleanalex@gmail.com>
To: Jonathan Cameron <jic23@kernel.org>
Cc: Mircea Caprioru <mircea.caprioru@analog.com>,
	"Hennerich, Michael" <Michael.Hennerich@analog.com>,
	Alexandru Ardelean <alexandru.ardelean@analog.com>,
	Lars-Peter Clausen <lars@metafoo.de>,
	Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
	LKML <linux-kernel@vger.kernel.org>,
	linux-iio <linux-iio@vger.kernel.org>,
	Sergiu Cuciurean <sergiu.cuciurean@analog.com>
Subject: Re: [PATCH 4/5] iio: adc: vf610_adc: Replace indio_dev->mlock with own device lock
Date: Wed, 30 Sep 2020 08:57:55 +0300	[thread overview]
Message-ID: <CA+U=DsqDymuWht_H1z+TnTvcXbnuJ9tm45w4g_U6g=qWz+fGvQ@mail.gmail.com> (raw)
In-Reply-To: <20200929171441.5b4ff8c8@archlinux>

On Tue, Sep 29, 2020 at 7:15 PM Jonathan Cameron <jic23@kernel.org> wrote:
>
> On Mon, 28 Sep 2020 16:13:32 +0300
> Mircea Caprioru <mircea.caprioru@analog.com> wrote:
>
> > From: Sergiu Cuciurean <sergiu.cuciurean@analog.com>
> >
> > As part of the general cleanup of indio_dev->mlock, this change replaces
> > it with a local lock on the device's state structure.
> >
> > This is part of a bigger cleanup.
> > Link: https://lore.kernel.org/linux-iio/CA+U=Dsoo6YABe5ODLp+eFNPGFDjk5ZeQEceGkqjxXcVEhLWubw@mail.gmail.com/
> >
> > Signed-off-by: Sergiu Cuciurean <sergiu.cuciurean@analog.com>
> > Signed-off-by: Mircea Caprioru <mircea.caprioru@analog.com>
>
> There are more problems in the locking in here than just this one.
> See below.  The taking of mlock like this was what originally motivated
> the efforts to hide it away from drivers.
>
> In this particular case I don't think a local lock is the correct solution.
>
> Thanks,
>
> Jonathan
>
>
> > ---
> >  drivers/iio/adc/vf610_adc.c | 28 ++++++++++++++++++++--------
> >  1 file changed, 20 insertions(+), 8 deletions(-)
> >
> > diff --git a/drivers/iio/adc/vf610_adc.c b/drivers/iio/adc/vf610_adc.c
> > index 1d794cf3e3f1..b7d583993f0b 100644
> > --- a/drivers/iio/adc/vf610_adc.c
> > +++ b/drivers/iio/adc/vf610_adc.c
> > @@ -168,6 +168,15 @@ struct vf610_adc {
> >
> >       struct completion completion;
> >       u16 buffer[8];
>
> Side note.  That buffer isn't correctly aligned.  I'll add this one to
> my next series fixing those.
>
> > +     /*
> > +      * Lock to protect the device state during a potential concurrent
> > +      * read access from userspace. Reading a raw value requires a sequence
> > +      * of register writes, then a wait for a completion callback,
> > +      * and finally a register read, during which userspace could issue
> > +      * another read request. This lock protects a read access from
> > +      * ocurring before another one has finished.
> > +      */
> > +     struct mutex lock;
> >  };
> >
> >  static const u32 vf610_hw_avgs[] = { 1, 4, 8, 16, 32 };
> > @@ -464,11 +473,11 @@ static int vf610_set_conversion_mode(struct iio_dev *indio_dev,
> >  {
> >       struct vf610_adc *info = iio_priv(indio_dev);
> >
> > -     mutex_lock(&indio_dev->mlock);
> > +     mutex_lock(&info->lock);
> Hmm. So there is a bit of a question on what the locking here is doing.
> (see below for a different use of mlock).
>
> What it will do currently is to prevent the conversion mode changing whilst
> we are in buffered mode.  It will also protect against concurrent
> calls of this function.
>
> I would replace this with iio_device_claim_direct_mode() rather than a
> local lock.

This raises a new question: if there's any drivers that we missed [for
iio_device_claim_direct_mode()].
While I was aware of iio_device_claim_direct_mode(), I missed this
fact when pushing the mlock cleanup.

Oh well, I'll do a quick audit over the current drivers that were converted.
Hopefully I don't find anything :P

>
> >       info->adc_feature.conv_mode = mode;
> >       vf610_adc_calculate_rates(info);
> >       vf610_adc_hw_init(info);
> > -     mutex_unlock(&indio_dev->mlock);
> > +     mutex_unlock(&info->lock);
> >
> >       return 0;
> >  }
> > @@ -632,9 +641,9 @@ static int vf610_read_raw(struct iio_dev *indio_dev,
> >       switch (mask) {
> >       case IIO_CHAN_INFO_RAW:
> >       case IIO_CHAN_INFO_PROCESSED:
> > -             mutex_lock(&indio_dev->mlock);
> > +             mutex_lock(&info->lock);
> >               if (iio_buffer_enabled(indio_dev)) {
> > -                     mutex_unlock(&indio_dev->mlock);
> > +                     mutex_unlock(&info->lock);
>
> Should be use iio_device_claim_direct_mode()
>
> mlock is being taken here to stop us entering buffered mode.
>
> Whilst I'd rather a driver didn't rely on internal details of
> IIO, it is rather fiddly to get the locking right when there is a completion
> going on, so I think here you are safe to do so.
>
> >                       return -EBUSY;
> >               }
> >
> > @@ -645,11 +654,11 @@ static int vf610_read_raw(struct iio_dev *indio_dev,
> >               ret = wait_for_completion_interruptible_timeout
> >                               (&info->completion, VF610_ADC_TIMEOUT);
> >               if (ret == 0) {
> > -                     mutex_unlock(&indio_dev->mlock);
> > +                     mutex_unlock(&info->lock);
> >                       return -ETIMEDOUT;
> >               }
> >               if (ret < 0) {
> > -                     mutex_unlock(&indio_dev->mlock);
> > +                     mutex_unlock(&info->lock);
> >                       return ret;
> >               }
> >
> > @@ -668,11 +677,11 @@ static int vf610_read_raw(struct iio_dev *indio_dev,
> >
> >                       break;
> >               default:
> > -                     mutex_unlock(&indio_dev->mlock);
> > +                     mutex_unlock(&info->lock);
> >                       return -EINVAL;
> >               }
> >
> > -             mutex_unlock(&indio_dev->mlock);
> > +             mutex_unlock(&info->lock);
> >               return IIO_VAL_INT;
> >
> >       case IIO_CHAN_INFO_SCALE:
> > @@ -807,6 +816,9 @@ static int vf610_adc_probe(struct platform_device *pdev)
> >       }
> >
> >       info = iio_priv(indio_dev);
> > +
> > +     mutex_init(&info->lock);
> > +
> >       info->dev = &pdev->dev;
> >
> >       info->regs = devm_platform_ioremap_resource(pdev, 0);
>