[bug report] net/mlx5: E-Switch, Protect changing mode while adding rules

[bug report] net/mlx5: E-Switch, Protect changing mode while adding rules

[Dan Carpenter]

Hello Roi Dayan,

The patch 7dc84de98bab: "net/mlx5: E-Switch, Protect changing mode
while adding rules" from Sep 16, 2020, leads to the following Smatch
static checker warning:

	drivers/net/ethernet/mellanox/mlx5/core/eswitch.c:2000 mlx5_esw_unlock()
	warn: inconsistent returns '&esw->mode_lock'.

drivers/net/ethernet/mellanox/mlx5/core/eswitch.c
    1996 void mlx5_esw_unlock(struct mlx5_eswitch *esw)
    1997 {
    1998         if (!mlx5_esw_allowed(esw))
    1999                 return;

Smatch is complaining because how will the caller know if we dropped
the lock or not.  I thought, "Hm.  I guess the lock function has a
similar check?  Although, how does that work that mlx5_esw_allowed()
means that it doesn't need locking?"

But then when I looked at the lock function, mlx5_esw_try_lock(), and it
does *NOT* have a similar check.  This probably works because it's
checked in different layers and this is just a duplicative (layering
violation) check which is ugly but harmless.

--> 2000         up_write(&esw->mode_lock);
    2001 }

regards,
dan carpenter

Re: [bug report] net/mlx5: E-Switch, Protect changing mode while adding rules

[Leon Romanovsky]

On Mon, May 16, 2022 at 10:04:05AM +0300, Dan Carpenter wrote:
> Hello Roi Dayan,
> 
> The patch 7dc84de98bab: "net/mlx5: E-Switch, Protect changing mode
> while adding rules" from Sep 16, 2020, leads to the following Smatch
> static checker warning:
> 
> 	drivers/net/ethernet/mellanox/mlx5/core/eswitch.c:2000 mlx5_esw_unlock()
> 	warn: inconsistent returns '&esw->mode_lock'.
> 
> drivers/net/ethernet/mellanox/mlx5/core/eswitch.c
>     1996 void mlx5_esw_unlock(struct mlx5_eswitch *esw)
>     1997 {
>     1998         if (!mlx5_esw_allowed(esw))
>     1999                 return;
> 
> Smatch is complaining because how will the caller know if we dropped
> the lock or not.  I thought, "Hm.  I guess the lock function has a
> similar check?  Although, how does that work that mlx5_esw_allowed()
> means that it doesn't need locking?"
> 
> But then when I looked at the lock function, mlx5_esw_try_lock(), and it
> does *NOT* have a similar check.  This probably works because it's
> checked in different layers and this is just a duplicative (layering
> violation) check which is ugly but harmless.

Your analysis is correct and I agree with you, the check should be removed.
However the "problematic" commit is ec2fa47d7b98 ("net/mlx5: Lag, use lag lock"),
where mlx5_esw_lock() was removed.

Thanks

> 
> --> 2000         up_write(&esw->mode_lock);
>     2001 }
> 
> regards,
> dan carpenter