Re: [PATCH] iio: imu: st_lsm6dsx: fix possible lockup during FIFO read

[Silvano Seva]

On Fri, Mar 7, 2025 at 3:40 PM Lorenzo Bianconi <lorenzo@xxxxxxxxxx> wrote:
>
> > On Thu, Mar 6, 2025 at 4:48 PM Lorenzo Bianconi <lorenzo@xxxxxxxxxx> wrote:
> > >
> > > > Prevent st_lsm6dsx_read_fifo and st_lsm6dsx_read_tagged_fifo functions
> > > > from falling in an infinite loop in case pattern_len is equal to zero and
> > > > the device FIFO is not empty.
> > > >
> > > > Signed-off-by: Silvano Seva <s.seva@xxxxxxxxx>
> > > > ---
> > > >  drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c | 8 ++++++++
> > > >  1 file changed, 8 insertions(+)
> > > >
> > > > diff --git a/drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c b/drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c
> > > > index 0a7cd8c1aa33..7f343614f8a5 100644
> > > > --- a/drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c
> > > > +++ b/drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c
> > > > @@ -395,12 +395,17 @@ int st_lsm6dsx_read_fifo(struct st_lsm6dsx_hw *hw)
> > > >       fifo_len = (le16_to_cpu(fifo_status) & fifo_diff_mask) *
> > > >                  ST_LSM6DSX_CHAN_SIZE;
> > > >       fifo_len = (fifo_len / pattern_len) * pattern_len;
> > > > +     if (!fifo_len)
> > > > +             return 0;
> > >
> > > I do not think this check is necessary since if fifo_len is 0 we will not run
> > > the for loop, right?
> >
> > This check is present in the st_lsm6dsx_read_tagged_fifo() function, i
> > added it here for consistency. I agree with you that is not strictly
> > necessary.
> >
> > >
> > > >
> > > >       acc_sensor = iio_priv(hw->iio_devs[ST_LSM6DSX_ID_ACC]);
> > > >       gyro_sensor = iio_priv(hw->iio_devs[ST_LSM6DSX_ID_GYRO]);
> > > >       if (hw->iio_devs[ST_LSM6DSX_ID_EXT0])
> > > >               ext_sensor = iio_priv(hw->iio_devs[ST_LSM6DSX_ID_EXT0]);
> > > >
> > > > +     if (!pattern_len)
> > > > +             pattern_len = ST_LSM6DSX_SAMPLE_SIZE;
> > >
> > > same here, I do not think pattern_len can be 0 since hw->sip must be greater
> > > than 0 in order to enable the FIFO. Moreover, this check should be some lines
> > > above since we have already divided fifo_len by pattern_len here.
> > >
> >
> > There is a situation causing the subsequent for loop to never
> > terminate, hanging the kernel boot process: given a system which
> > doesn't have an hardware reset line allowing the kernel to
> > re-initialize the IMU hardware, in case of an hot reboot the driver
> > probe() function attempts to flush the FIFO, which may contain some
> > data, while the hw->sip is zero.
> > The complete execution path is the following:
> > - call of st_lsm6dsx_probe();
> > - allocation of the st_lsm6dsx_hw structure via the devm_kzalloc,
> > zero-initializing the sip field;
> > - call of st_lsm6dsx_init_device();
> > - call of st_lsm6dsx_reset_device();
> > - call of st_lsm6dsx_flush_fifo();
> > - call of st_lsm6dsx_read_fifo/st_lsm6dsx_read_tagged_fifo via the
> > fifo_ops function pointer.
>
> ack, I can see the issue now, thx for pointing this out. I should we should set
> pattern_len to ST_LSM6DSX_SAMPLE_SIZE or ST_LSM6DSX_TAGGED_SAMPLE_SIZE if it is
> 0. Can you please move the check before updating fifo_len in
> st_lsm6dsx_read_fifo()?
>
> Regards,
> Lorenzo
>

Changes done, I sent the updated patch in a separate reply email: I
hope that doing so didn't cause any trouble.

Thank you,
Silvano.