[linux-dvb] Re: [PATCH] fix for stream corruption on budget /Nova-T cards

[Jon Burgess <mplayer@jburgess.uklinux.net>]

Andreas Vierengel wrote:

> Am So, 2003-08-31 um 18.28 schrieb Jon Burgess:
> the vga-chip seems to be on bus 0. only onboard ethernet is on bus 1
> together with all external pci-cards. here is a lspci output. I don't
> know very much about today's motherboard internals, but are these 2
> busses are physically seperate ?
Yes the busses are separate, but at some point the data must get from 
the card to the RAM chips.

I've been reading some of the Intel docs for the 845G memory controlled 
(GMCH) and it confirms that has an arbitration engine which decides 
which devices gets its data to or from the memory.

In order to boost the performance they implement buffering so that small 
amounts of data can be queued while the RAM is actually doing a transfer 
for something else. Perhaps what is happening is that the larger burst 
sizes cause these buffers to fill completely and it halts the DMA 
transfer until it can empty the buffers to RAM. By making the burst size 
smaller, it all gets buffered by the memory controller and things work 
more smoothly.

> > A PCI bus analyzer would be needed to really understand what is going on.
> that would be really cool and certainly a nice addon to strace, tcpdump,
> etc :)
Unfortunately it requires an external logic analyser and costs many 
$1000's. Unless anyone knows of any performance statistics which can be 
read from the PCI bridge / GMCH?

	Jon




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