Re: [PATCH 1/3] media: i2c: imx334: uninitialized variable in imx334_update_exp_gain()

[Tarang Raval <tarang.raval@siliconsignals.io>]

> On Wed, Apr 30, 2025 at 09:47:44AM +0000, Tarang Raval wrote:
> > Hi Dan,
> >
> > > The "ret" variable is not initialized on the success path.  Set it to
> > > zero.
> > >
> > > Fixes: 7b19b0fc8ac8 ("media: i2c: imx334: Convert to CCI register access helpers")
> > > Signed-off-by: Dan Carpenter <dan.carpenter@linaro.org>
> > > ---
> > >  drivers/media/i2c/imx334.c | 3 ++-
> > >  1 file changed, 2 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/drivers/media/i2c/imx334.c b/drivers/media/i2c/imx334.c
> > > index fc875072f859..846b9928d4e8 100644
> > > --- a/drivers/media/i2c/imx334.c
> > > +++ b/drivers/media/i2c/imx334.c
> > > @@ -536,7 +536,8 @@ static int imx334_update_controls(struct imx334 *imx334,
> > >  static int imx334_update_exp_gain(struct imx334 *imx334, u32 exposure, u32 gain)
> > >  {
> > >         u32 lpfr, shutter;
> > > -       int ret, ret_hold;
> > > +       int ret_hold;
> > > +       int ret = 0;
> >
> > I think this initialization may not really be necessary.
> >
> > If all of those cci_write are skipped, then yes, using ret uninitialized
> > would be a problem.
> >
> > However, I don’t see any case where they would be skipped in the
> > current implementation.
> >
> > So, is this initialization really needed?
> >
> > Best Regards,
> > Tarang
> 
> This is a new bug that was introduced in linux-next...
> 
> drivers/media/i2c/imx334.c
>    536  static int imx334_update_exp_gain(struct imx334 *imx334, u32 exposure, u32 gain)
>    537  {
>    538          u32 lpfr, shutter;
>    539          int ret, ret_hold;
>    540
>    541          lpfr = imx334->vblank + imx334->cur_mode->height;
>    542          shutter = lpfr - exposure;
>    543
>    544          dev_dbg(imx334->dev, "Set long exp %u analog gain %u sh0 %u lpfr %u\n",
>    545                  exposure, gain, shutter, lpfr);
>    546
>    547          cci_write(imx334->cci, IMX334_REG_HOLD, 1, &ret);
> 
> This first call will do an unitialized read of ret to check if it holds
> an error code.
> 
>    548          cci_write(imx334->cci, IMX334_REG_VMAX, lpfr, &ret);
>    549          cci_write(imx334->cci, IMX334_REG_SHUTTER, shutter, &ret);
>    550          cci_write(imx334->cci, IMX334_REG_AGAIN, gain, &ret);
> 
> cci_write() is designed to preserve the error codes from previous calls.
> It will only write error codes to "ret", it will not write success.  If
> everything succeeds then "ret" is uninitialized.

Yes, correct
 
>    551
>    552          ret_hold = cci_write(imx334->cci, IMX334_REG_HOLD, 0, NULL);
>    553          if (ret_hold)
>    554                  return ret_hold;
>    555
>    556          return ret;
>    557  }
> 
> In production then everyone should run with INIT_STACK_ALL_ZERO.  In that
> case everything works fine.  However some older distributions do not use
> this option.  Also in testing, I would encourage everyone to run with
> INIT_STACK_ALL_PATTERN.

I am not aware of the INIT_STACK_ALL_ZERO and INIT_STACK_ALL_PATTERN options.
I’ll read more about them

Best Regards,
Tarang