Re: [PATCH v2 1/5] dvb-core: add a new tuner ops to dvb_frontend for APIv5

[Antti Palosaari <crope@iki.fi>]

On 09/06/2014 06:17 AM, Mauro Carvalho Chehab wrote:
> Em Sat, 06 Sep 2014 06:10:01 +0300
> Antti Palosaari <crope@iki.fi> escreveu:

>> ... I simply don't understand why you want hook that RF strength call
>> via demod? The frontend cache is shared between demod and tuner. We use
>> it for tuner driver as well demod driver. Let the tuner driver make RSSI
>> calculation independently without any unneeded relation to demod driver.
>
> Well, adding kthreads has a cost, with is a way higher than just
> calling a callback function.
>
> Also, it makes a way more complicated to implement several tasks.
>
> For example, devices with kthreads need to stop them during suspend,
> and restart them at resume time, or otherwise suspend and/or resume
> may not work.
>
> Also, the power consumption increases with kthread, as the CPU need
> to be periodically waken.
>
> I'm not saying we shouldn't use kthreads at driver level, but
> the best is to avoid when there are some other simpler ways of doing it.

And small reality check, how much you think one kthreads, that polls 
once per 2 second or so causes, in a case when you are *already 
streaming* 20-80 Mbit/sec data stream :) I think CPU does not need wake 
up to execute one thread as there is a lot of other interrupts happening 
in that use case anyway.

We have a remote controller which is polled often as 100ms and it 
happens even when device is "sleeping".

regards
Antti

-- 
http://palosaari.fi/