In article <1993Jun24.153619.26394@tpl68k0.tplrd.tpl.oz.au> charlieb@tplrd.tpl.oz.au (Charlie Brady) writes:
quoted text
>
>There is a new set of code for a 386bsd driver (not, unsurprisingly, written
>by the appalling Jesus Monroy, jnr) which can stream most of the time, albiet
>without error detection/correction. The comments state clearly that a timer
>with microsecond resolution, and low latency interrupts are a must. Are
>either of these a problem under Linux? Sorry for my ignorance of the kernel
>internals.

Interrupt latency: if it works under 386bsd, it should work under linux
as well unless they have seriously cleaned up their interrupt problems. 
The linux interrupt latency is reasonably good: it could probably be
improved upon, but it should be acceptable. 

As to a microsecond timer..  Hardly.  The current linux timer mechanism
isn't even millisecond accurate (it uses the 100Hz ticks), and greater
accuracy isn't expected.  Microsecond _delays_ are possible, but will
require busy looping - not that much looping is needed on a AT bus as a
single IO instruction takes about a microsecond if I remember correctly. 
I assume that is how the 386bsd crowd did it too. 

                Linus