Re: [PATCH 2/2] media: i2c: Add driver for ST VD56G3 camera sensor

[Sakari Ailus <sakari.ailus@iki.fi>]

Hi Benjamin,

On Mon, May 27, 2024 at 09:44:15AM +0200, Benjamin Mugnier wrote:
> Hi Sakari,
> 
> On 5/27/24 09:31, Sakari Ailus wrote:
> > Hi Benjamin,
> > 
> > On Mon, May 13, 2024 at 03:37:25PM +0200, Benjamin Mugnier wrote:
> >>>>> +static int vd56g3_read_expo_cluster(struct vd56g3 *sensor, bool force_cur_val)
> >>>>> +{
> >>>>> +	u64 exposure = 0;
> >>>>> +	u64 again = 0;
> >>>>> +	u64 dgain = 0;
> >>>>
> >>>> Do you need the assignments? The values are assigned by cci_read() below,
> >>>> right?
> >>>
> >>> Well initially, I didn't have those assignments and some checker
> >>> complains (Honestly I didn't investigate the warning).
> >>> I will double check and see if they are necessary.
> >>>
> >>
> >> Since I have smatch ready, I ran it without these initialization against
> >> the latest media tree. I got these :
> >>
> >> drivers/media/i2c/st-vd56g3.c:403 vd56g3_read_expo_cluster() error:
> >> uninitialized symbol 'exposure'.
> >> drivers/media/i2c/st-vd56g3.c:404 vd56g3_read_expo_cluster() error:
> >> uninitialized symbol 'again'.
> >> drivers/media/i2c/st-vd56g3.c:405 vd56g3_read_expo_cluster() error:
> >> uninitialized symbol 'dgain'.
> >> drivers/media/i2c/st-vd56g3.c:407 vd56g3_read_expo_cluster() error:
> >> uninitialized symbol 'exposure'.
> >> drivers/media/i2c/st-vd56g3.c:408 vd56g3_read_expo_cluster() error:
> >> uninitialized symbol 'again'.
> >> drivers/media/i2c/st-vd56g3.c:409 vd56g3_read_expo_cluster() error:
> >> uninitialized symbol 'dgain'.
> >>
> >> cci_read() is indeed not modifying its 'var' argument in the default
> >> case of the switch [1]. Spawning these errors.
> >> Adding a '*val = 0' in the default case would also fix it, but will
> >> change the function behavior.
> > 
> > I think I'd add the assignment to cci_read(). This isn't the only driver
> > that's affected. In the best case smatch would be fixed to figure this out.
> > 
> > Any thoughts?
> 
> IMHO adding the assignment to cci_read() is the cleanest way to handle it.

A fix in smatch would be the best in my opinion as it could catch the use
of the value in case of an error. But it's not going to be a short term
solution, hence I'd also assign *val to 0 now.

> 
> > 
> > Cc also Hans and Laurent.
> > 
> 

-- 
Regards,

Sakari Ailus