Re: [PATCH v2] iio: light: vcnl4000: Fix WARN_ON on uninitialized lock

[Mårten Lindahl <martenli@xxxxxxxx>]

On 1/31/23 13:46, Cai Huoqing wrote:
> On 31 1月 23 11:29:51, Mårten Lindahl wrote:
> > There are different init functions for the sensors in this driver in
> > which only one initialize the generic vcnl4000_lock. With commit
> > e21b5b1f2669 ("iio: light: vcnl4000: Preserve conf bits when toggle power")
> > the vcnl4040 sensor started to depend on the lock, but it was missed to
> > initialize it in vcnl4040's init function. This has not been visible
> > until we run lockdep on it:
> >
> >    DEBUG_LOCKS_WARN_ON(lock->magic != lock)
> >    at kernel/locking/mutex.c:575 __mutex_lock+0x4f8/0x890
> >    Call trace:
> >    __mutex_lock
> >    mutex_lock_nested
> >    vcnl4200_set_power_state
> >    vcnl4200_init
> >    vcnl4000_probe
> >    i2c_device_probe
> >    really_probe
> >    __driver_probe_device
> >    driver_probe_device
> >    __driver_attach
> >    bus_for_each_dev
> >    driver_attach
> >    bus_add_driver
> >    driver_register
> >    i2c_register_driver
> >    vcnl4000_driver_init
> >    do_one_initcall
> >    do_init_module
> >    load_module
> >    __do_sys_finit_module
> >
> > Fix this by initializing the lock in the probe function instead of doing
> > it in the chip specific init functions.
> >
> > Fixes: e21b5b1f2669 ("iio: light: vcnl4000: Preserve conf bits when toggle power")
> > Signed-off-by: Mårten Lindahl <marten.lindahl@xxxxxxxx>
> > ---
> >
> > v2:
> >   - Trimmed backtrace in commit message
> >   - Have 12 digit sha-1 id in Fixes tag
> >   - Make the lock initialization in probe instead of in _init function
> >
> >   drivers/iio/light/vcnl4000.c | 2 +-
> >   1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/drivers/iio/light/vcnl4000.c b/drivers/iio/light/vcnl4000.c
> > index b5d398228289..caa2fff9f486 100644
> > --- a/drivers/iio/light/vcnl4000.c
> > +++ b/drivers/iio/light/vcnl4000.c
> > @@ -208,7 +208,6 @@ static int vcnl4000_init(struct vcnl4000_data *data)
> >   
> >   	data->rev = ret & 0xf;
> >   	data->al_scale = 250000;
> > -	mutex_init(&data->vcnl4000_lock);
> >   
> >   	return data->chip_spec->set_power_state(data, true);
> >   };
> > @@ -1366,6 +1365,7 @@ static int vcnl4000_probe(struct i2c_client *client,
> >   	data->client = client;
> >   	data->id = id->driver_data;
> >   	data->chip_spec = &vcnl4000_chip_spec_cfg[data->id];
> > +	mutex_init(&data->vcnl4000_lock);
> >   
> >   	ret = data->chip_spec->init(data);
> >   	if (ret < 0)
> Why not add mutex_init(&data->vcnl4000_lock) in vcnl4200_init.
> like this
>
> @@ -330,6 +330,7 @@ static int vcnl4200_init(struct vcnl4000_data *data)
>          }
>          mutex_init(&data->vcnl4200_al.lock);
>          mutex_init(&data->vcnl4200_ps.lock);
> +       mutex_init(&data->vcnl4000_lock);
>
>          ret = data->chip_spec->set_power_state(data, true);
>          if (ret < 0)
>
> Perfer adding mutex_init to vcnl4200_init.

Hi Cai!

This is what I did in v1, but I got a suggestion to move it to the probe 
instead.

Having it in probe takes one mutex_init. Otherwise it needs to be in two 
places (both init functions).

Kind regards

Mårten

> > --
> > 2.30.2