Mailing List archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[linux-dvb] Re: System load caused by I2C (was: Re: Re: Announcement:new saa7146 driver core for the av7110 dvb driver)



Hello Johannes,

 > Usually there's no need to to capture vbi on a DVB card.
 > One exception might be the analog extension for the
 > Fujitsu-Seimens DVB-C card, but I'm not shure if that is
 > currently supported at all.

I think that something like "vbi-insertion" is totally superfluous, 
because you have the data most likely already in the "digital form".

(Ok, you can run you legacy videotext application...)

 > It would be nice if we could use RPS1 for I2C if RPS1 is available,
 > and fall back to the current I2C mode if RPS1 is used for vbi.
 > But like I said, it's not terribly important. Maybe it's
 > better to slow down kdvb-fe to lower the system load after
 > one round of zig-zag scanning failed.

I didn't reply in the proper thread, so here it comes again:

-------------------------------------------------------------------------------
As a first attempt, I changed i2c_writeout() in CVS to use a waitqueue 
and the irq handler instead of busy waiting. Currently it does this for 
each i2c writeout, not for the whole transaction.

It decreases the load dramatically for long i2c transactions in my tests.

(Of course it would be better to write something like:
- prepare the i2c message, then fire it off
- let the irq handler directly handle all following i2c transactions
- only wake up the waiting process if the whole transaction is finished
)
-------------------------------------------------------------------------------

Using the RPS is probably total overkill, the irq + waitqueue approach 
should be ok, I think.

Please test it. 8-)

> Ciao,
> Johannes

CU
Michael.




-- 
Info:
To unsubscribe send a mail to listar@linuxtv.org with "unsubscribe linux-dvb" as subject.



Home | Main Index | Thread Index