On Mon, 26 Jul 2021 20:03:14 +0800
hui.liu <hui.liu@xxxxxxxxxxxx> wrote:
> On Sat, 2021-07-24 at 18:30 +0100, Jonathan Cameron wrote:
> > On Fri, 23 Jul 2021 20:21:15 +0800
> > hui.liu <hui.liu@xxxxxxxxxxxx> wrote:
> >
> > > On Sat, 2021-07-17 at 17:44 +0100, Jonathan Cameron wrote:
> > > > On Thu, 15 Jul 2021 17:35:23 +0800
> > > > Hui Liu <hui.liu@xxxxxxxxxxxx> wrote:
> > > >
> > > > > Add mutex_destroy when probe fail and remove device.
> > > > >
> > > > > Signed-off-by: Hui Liu <hui.liu@xxxxxxxxxxxx>
> > > > Hi Hui Liu,
> > > >
> > > > We very very rarely bother to call mutex_destroy(). The reason is
> > > > that it is only a non noop in when mutex debugging is enabled and
> > > > that is only useful if there is a plausible route in which it could
> > > > be used after the mutex_destroy. Given these are both at the ends
> > > > of removal paths, I don't think this is useful. That's why you will
> > > > rarely find mutex_destroy() being called.
> > > >
> > > > Thanks,
> > > >
> > > > Jonathan
> > >
> > > Hi Jonathon,
> > >
> > > I think this patch could assurance the integrity of code.
> > > mutex_init will be called when driver probe. If driver probe fail or
> > > device removed, mutex_destroy could set lock->magic to NULL.
> >
> > I'm not seeing the use case here given the location doesn't leave
> > a huge amount of code that could have such a bug. There might have been
> > something if we had any route to increment the reference count of the
> > structure this mutex is ultimately embedded in and so have it outlast
> > the remove function or error path. In this driver it looks like there is
> > no such path. Hence you are protecting against a automated
> > cleanup of core code (nothing in the driver itself) which is obviously
> > not going to try taking a driver specific mutex.
> >
> > A few side notes:
> >
> > You are calling it wrong place in remove. The ordering in remove
> > should be the opposite of that in probe so the mutex_destroy should either
> > be a few lines earlier, or you should have a comment there to say why you
> > are doing it where you have chosen to do so.
> >
> > The style of this probe is to do error handling in a block at the end.
> > So this handling should be there, not in the if statement.
> >
> > Jonathan
> >
> >
> Hi Jonathon,
>
> Base on your helpful opinion, We will to do two changes in patch v2.
> 1. In probe: move mutex_destroy from the if statement to error handling
> path(err_power_off).
> 2. In remove: calling mutex_destroy right after iio_device_unregister.
>
> Do we need some more change? Thanks.
Ah. Sorry I missed this in the flood of emails during the week.
Anyhow, I've replied to the v1 posting.
Jonathan
> >
> > >
> > > Thanks.
> > > Hui
> > >
> > > >
> > > > > ---
> > > > > drivers/iio/adc/mt6577_auxadc.c | 2 ++
> > > > > 1 file changed, 2 insertions(+)
> > > > >
> > > > > diff --git a/drivers/iio/adc/mt6577_auxadc.c b/drivers/iio/adc/mt6577_auxadc.c
> > > > > index 79c1dd68b909..d57243037ad6 100644
> > > > > --- a/drivers/iio/adc/mt6577_auxadc.c
> > > > > +++ b/drivers/iio/adc/mt6577_auxadc.c
> > > > > @@ -289,6 +289,7 @@ static int mt6577_auxadc_probe(struct platform_device *pdev)
> > > > > ret = iio_device_register(indio_dev);
> > > > > if (ret < 0) {
> > > > > dev_err(&pdev->dev, "failed to register iio device\n");
> > > > > + mutex_destroy(&adc_dev->lock);
> > > > > goto err_power_off;
> > > > > }
> > > > >
> > > > > @@ -313,6 +314,7 @@ static int mt6577_auxadc_remove(struct platform_device *pdev)
> > > > > 0, MT6577_AUXADC_PDN_EN);
> > > > >
> > > > > clk_disable_unprepare(adc_dev->adc_clk);
> > > > > + mutex_destroy(&adc_dev->lock);
> > > > >
> > > > > return 0;
> > > > > }
> > > >
> > >
> >
>
