Re: [PATCH v5 3/3] leds: trigger: implement a tty trigger

[Uwe Kleine-König]

Hello Pavel,

On Sat, Dec 21, 2019 at 07:40:47PM +0100, Pavel Machek wrote:
> > +++ b/Documentation/ABI/testing/sysfs-class-led-trigger-tty
> > @@ -0,0 +1,6 @@
> > +What:		/sys/class/leds/<led>/dev
> > +Date:		Dec 2019
> > +KernelVersion:	5.6
> > +Contact:	linux-leds@xxxxxxxxxxxxxxx
> > +Description:
> > +		Specifies $major:$minor of the triggering tty
> 
> Ok, sounds reasonable.
> 
> > +static ssize_t dev_store(struct device *dev,
> > +			 struct device_attribute *attr, const char *buf,
> > +			 size_t size)
> > +{
> > +	struct ledtrig_tty_data *trigger_data = led_trigger_get_drvdata(dev);
> > +	struct tty_struct *tty;
> > +	unsigned major, minor;
> > +	int ret;
> > +
> > +	if (size == 0 || (size == 1 && buf[0] == '\n')) {
> > +		tty = NULL;
> > +	} else {
> > +		ret = sscanf(buf, "%u:%u", &major, &minor);
> > +		if (ret < 2)
> > +			return -EINVAL;
> 
> If user writes 1:2:badparsingofdata into the file, it will pass, right?

Yes, and it will have the same effect as writing 1:2. I wonder if this
is bad.

> > +		tty = tty_kopen_shared(MKDEV(major, minor));
> > +		if (IS_ERR(tty))
> > +			return PTR_ERR(tty);
> > +	}
> 
> Do you need to do some kind of tty_kclose()? What happens if the
> device disappears, for example because the USB modem is unplugged?

Only tty_kref_put is needed to close.

> > +static void ledtrig_tty_work(struct work_struct *work)
> > +{
> > +	struct ledtrig_tty_data *trigger_data =
> > +		container_of(work, struct ledtrig_tty_data, dwork.work);
> > +	struct serial_icounter_struct icount;
> > +	int ret;
> > +
> > +	if (!trigger_data->tty) {
> > +		led_set_brightness(trigger_data->led_cdev, LED_OFF);
> > +		return;
> > +	}
> > +
> > +	ret = tty_get_icount(trigger_data->tty, &icount);
> > +	if (ret)
> > +		return;
> > +
> > +	if (icount.rx != trigger_data->rx ||
> > +	    icount.tx != trigger_data->tx) {
> > +		unsigned long delay_on = 100, delay_off = 100;
> > +
> > +		led_blink_set_oneshot(trigger_data->led_cdev,
> > +				      &delay_on, &delay_off, 0);
> > +
> > +		trigger_data->rx = icount.rx;
> > +		trigger_data->tx = icount.tx;
> > +	}
> 
> Since you are polling this, anyway, can you just manipulate brightness
> directly instead of using _oneshot()? _oneshot() will likely invoke
> another set of workqueues.

I copied that from the netdev trigger. I failed to find a suitable
helper function, did I miss that or does it need creating?
 
> LED triggers were meant to operate directly from the events, not based
> on statistics like this.

Ditto; just copied from the netdev trigger. I tried to find a suitable
place to add a trigger in the core, but this is hard without having to
modify all drivers; additionally this is in thier hot path. So I
considered using statistics a good idea. Greg also liked it and someone
before us for the network trigger, too ...

Best regards
Uwe

-- 
Pengutronix e.K.                           | Uwe Kleine-König            |
Industrial Linux Solutions                 | https://www.pengutronix.de/ |