Mailing List archive

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

[linux-dvb] Re: new. perfect fidbirq()



HI!!

fidbirq() gave us much hassle...
you fixed it to work with his setup of channel broadcast, then it didn't 
work on my setup with network reception. I fixed it for network, then it 
didn't work for yours. 

Now your latest change to vpeirq() works for both of us.

I think that is the revolutionary approach that made it finally
working well on both heavy-duty setups

> and using vpeirq() instead. The vpeirq() calls are directly related to
> DMApos, not to VSYNC on other chip, so the DMApos I read during
> vpeirq() seems to be _always_ reliable. Moreover, I've setup the registers 

My tests on DVB tree show that vpeirq() performed for me with
zero packet loss. mplayer worked well and more important, network 
reception worked as well too. 

vpeirq() 4x interrupt technical solution gives the best reliability
and has capability to recover from even 2 lost interrupts.

It has excellent packet reception and zero packet loss for different 
mission-critical setups, 

> in order to have 4 interrupts during one buffer cycle, which means if we miss 
> one (or even two), nothing bad happens, since DMA is writing into other 1/4 of
> the whole buffer.

This is so far the best way to avoid DMA overwriting.

>   It survived 24 hours of 16 TV channel broadcast form 3 cards without a
> single discontinuity problem. Since it uses DMApos, I believe it should
> work fine on all setups.

I will port it to dvb-kernel. From my tests, it can be applied to 
DVB tree.

>   I'm attaching the patch against DVB tree, could you try it?

It was my pleasure to test it. Everything worked well

Emard


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



Home | Main Index | Thread Index