Re: [PATCH v2 0/8] Remove adis_initial_startup usage

["Nuno Sá" <noname.nuno@gmail.com>]

On Sat, 2022-11-05 at 15:06 +0000, Jonathan Cameron wrote:
> On Thu, 3 Nov 2022 12:35:31 +0000
> "Sa, Nuno" <Nuno.Sa@analog.com> wrote:
> 
> > > From: Ramona Bolboaca <ramona.bolboaca@analog.com>
> > > Sent: Thursday, November 3, 2022 9:09 AM
> > > To: jic23@kernel.org; linux-iio@vger.kernel.org; linux-
> > > kernel@vger.kernel.org
> > > Cc: Bolboaca, Ramona <Ramona.Bolboaca@analog.com>
> > > Subject: [PATCH v2 0/8] Remove adis_initial_startup usage
> > > 
> > > 
> > > Remove 'adis_initial_startup()' usage due to the fact that it
> > > leads to a
> > > deadlock.
> > > The same mutex is acquired twice, without releasing it, once
> > > inside
> > > 'adis_initial_startup()' and once inside 'adis_enable_irq()'.
> > > Instead of 'adis_initial_startup()', use
> > > '__adis_initial_startup()'.
> > > 
> > > Ramona Bolboaca (8):
> > >   iio: accel: adis16201: Fix deadlock in probe
> > >   iio: accel: adis16209: Fix deadlock in probe
> > >   iio: gyro: adis16136: Fix deadlock in probe
> > >   iio: gyro: adis16260: Fix deadlock in probe
> > >   iio: imu: adis16400: Fix deadlock in probe
> > >   staging: iio: accel: adis16203: Fix deadlock in probe
> > >   staging: iio: accel: adis16240: Fix deadlock in probe
> > >   iio: imu: adis: Remove adis_initial_startup function
> > > 
> > >  drivers/iio/accel/adis16201.c         |  2 +-
> > >  drivers/iio/accel/adis16209.c         |  2 +-
> > >  drivers/iio/gyro/adis16136.c          |  2 +-
> > >  drivers/iio/gyro/adis16260.c          |  2 +-
> > >  drivers/iio/imu/adis16400.c           |  2 +-
> > >  drivers/staging/iio/accel/adis16203.c |  2 +-
> > >  drivers/staging/iio/accel/adis16240.c |  2 +-
> > >  include/linux/iio/imu/adis.h          | 12 ------------
> > >  8 files changed, 7 insertions(+), 19 deletions(-)
> > >   
> > 
> > You could have placed your v2 changelog in the cover letter.
> > Moreover it's the same for all patches... Anyways: 
> > 
> > Reviewed-by: Nuno Sá <nuno.sa@analog.com>
> 
> This feels a little backwards.  Normally we'd expect the
> outer function to take the lock and the inner call to not
> do so.  Now it's fine to not take the lock here at all because
> the outer function call is in probe anyway, before we reach
> the point where there should be an concurrency.
> 
> I wonder if we should instead do this by having
> an unlocked __adis_enable_irq() that is always called
> by __adis_initial_startup().  That would be the fix that
> then needs backporting.
> 

I did mentioned the same thing in the first version of the series but
did not really pushed for it. Now that you mention, I agree it feels
weird (and wrong from a design perspective) to have the lock,
"silently", taken inside a function starting with double underscore
(which should mean unlocked call).

> Switching the calls from adis_initial_startup() to
> __adis_initial_startup() would then just be a trivial
> optimization to not take locks before they should ever matter.
> 
> This all hinges on my assumption that the lock isn't useful.
> Am I right on that?
> 

I think so as all the calls happen during probe before registering the
userspace interface.

- Nuno Sá