On 8/29/06, Lourens Veen <[EMAIL PROTECTED]> wrote:
How are we on hogging the PCI bus? Are the low-latency Linux people (you
know, the ones who build digital audio workstations and hard disc
recorders and hack on Ardour and Jack and LADSPA) going to complain
about us messing up the latency to their PCI sound cards?

One of the major complaints in the past came from this:  Most graphics
chips have a way to "throttle" the bus when writing to the drawing
engine.  That is, if you don't check the command queue to make sure
there's room in it and just go ahead and write commands, that's "okay"
because the GPU will tell the PCI bus to wait.  Some driver developers
took advantage of this to get a few % more performance, but the wait
periods on the PCI bus got to be so long that other devices were not
able to use the bus to service interrupts.  OGA simply will not
provide a mechanism to throttle the bus when using PIO to write to the
command queue, instead, you'll get an undefined result.  We discourae
the use of PIO anyhow.

The second way we can hog the bus is to do long DMA bursts.  In the
interest of fairness, we'll always limit our bursts to reasonable
lengths of time (16 or 32 words, IIRC).
_______________________________________________
Open-graphics mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-graphics
List service provided by Duskglow Consulting, LLC (www.duskglow.com)

Reply via email to