Re: [RFC PATCH] can: add tx/rx led trigger support

[Fabio Baltieri <fabio.baltieri@gmail.com>]

On Thu, Apr 12, 2012 at 6:13 PM, Wolfgang Grandegger <wg@grandegger.com> wrote:
>> I think the original approach from Fabio was to follow the LED implementation
>> from mac80211:
>>
>> "The implementation is similar to the MAC80211_LEDS one, and takes quite a lot
>> of inspiration and some code from it."
>>
>> And IMHO we should follow this implementation as it is proved to be used in
>> high traffic netdev environments. I wonder if some 'generic gpio LED'
>> implementation will perform sufficiently.
>
> Yes, I agree. But that implementation does not use timers but toggles
> the LED on/off for each new packet.
>
> http://lxr.linux.no/#linux+v3.3.1/net/mac80211/led.c#L15

That's correct, but I think it works for 80211 because there you
always have some traffic. In the canbus case if implemented as
"toggle" you'll end up with a led full-on when removing the cable with
50% probability - which I think is quite misleading.

> I would vote for that solution as well (because it's very lite and we
> already have counters for rx and tx packets).

There are other implementation (which I took as example) in critical
paths which blink leds on one-time events with timers:

ledtrig-ide-disk (with fixed 10ms timeout):
http://lxr.linux.no/#linux+v3.3.1/drivers/leds/ledtrig-ide-disk.c#L28

netfilter's xt_LED trigger:
http://lxr.linux.no/#linux+v3.3.1/net/netfilter/xt_LED.c#L69

I think that with the timer called with a reasonably high delay
(20-30ms) it should be quite lightweight. After all, most of the times
embedded system's leds are gpio-mapped and will be handled by a bunch
of function call and single memory write...

Regards.

-- 
Fabio Baltieri