Re: [PATCH v1 1/2] leds: rt4505: Add support for Richtek RT4505 flash led controller

[Jacek Anaszewski]

On 10/28/20 5:57 AM, ChiYuan Huang wrote:
> Hi,
>
> Jacek Anaszewski <jacek.anaszewski@xxxxxxxxx> 於 2020年10月28日 週三 上午12:40寫道：
> > Hi Pavel, ChiYuan,
> >
> > On 10/27/20 9:29 AM, Pavel Machek wrote:
> > > Hi!
> > >
> > > > From: ChiYuan Huang <cy_huang@xxxxxxxxxxx>
> > > >
> > > > Add support for RT4505 flash led controller. It can support up to 1.5A
> > > > flash current with hardware timeout and low input voltage
> > > > protection.
> > >
> > > Please use upper-case "LED" everywhere.
> > >
> > > This should be 2nd in the series, after DT changes.
> > >
> > > > Signed-off-by: ChiYuan Huang <cy_huang@xxxxxxxxxxx>
> > > > ---
> > > >    drivers/leds/Kconfig       |  11 ++
> > > >    drivers/leds/Makefile      |   1 +
> > > >    drivers/leds/leds-rt4505.c | 397 +++++++++++++++++++++++++++++++++++++++++++++
> > > >    3 files changed, 409 insertions(+)
> > > >    create mode 100644 drivers/leds/leds-rt4505.c
> > [...]
> > > > +static int rt4505_torch_brightness_set(struct led_classdev *lcdev, enum led_brightness level)
> > > > +{
> > >
> > > 80 columns, where easy.
> > >
> > > > +    struct rt4505_priv *priv = container_of(lcdev, struct rt4505_priv, flash.led_cdev);
> > > > +    u32 val = 0;
> > > > +    int ret;
> > > > +
> > > > +    mutex_lock(&priv->lock);
> > > > +
> > > > +    if (level != LED_OFF) {
> > > > +            ret = regmap_update_bits(priv->regmap, RT4505_REG_ILED, RT4505_ITORCH_MASK,
> > > > +                                     (level - 1) << RT4505_ITORCH_SHIFT);
> > > > +            if (ret)
> > > > +                    goto unlock;
> > > > +
> > > > +            val = RT4505_TORCH_SET;
> > > > +    }
> > > > +
> > > > +    ret = regmap_update_bits(priv->regmap, RT4505_REG_ENABLE, RT4505_ENABLE_MASK, val);
> > > > +
> > > > +unlock:
> > > > +    mutex_unlock(&priv->lock);
> > > > +    return ret;
> > > > +}
> > >
> > > Why is the locking needed? What will the /sys/class/leds interface
> > > look like on system with your flash?
> >
> > The locking is needed since this can be called via led_set_brightness()
> > from any place in the kernel, and especially from triggers.
> > From this case, It means only led classdev
> brihtness_get/brightness_set need to be protected.
> I search led_flash_classdev, it only can be controlled via sysfs or V4l2.
> Like as described in last mail, flash related operation is protected
> by led access_lock and v4l2 framework.
> I'll keep the locking only in led classdev brightness_get/brightness_set API.
> If I misunderstand something, please help to point out.

Locking have to be used consistently for each access to the resource
being protected with the lock. Otherwise you can end up in a situation
when rt4505_torch_brightness_set and rt4505_flash_brightness_set will
try concurrently alter hardware state. Regardless of how harmful could
it be in case of this particular device it is certainly against
programming rules.

--
Best regards,
Jacek Anaszewski