Re: PCI Interrupt and Acknowledgement

[Dimiter Popoff <tgi_earth@yahoo.com>]

Servan, Kevin,

>> My question is that is this time is normal or long? Is it possible to
>> shorten
>> this interrupt acknowledgement time?
>
>use another OS?
>
>Are you using some variant of Windows?


Wintel and interrupts - I can hardly think of another two less compatible words.

The 100 mS Kevin mentions are dated 1999; about another 3 or so years ago
another guy (working with NT) told me about 22 mS. Linear extrapolation
yields something like 200 to 500 mS today...

Talking interrups in milliseconds is of course done only in wintel land, to the
best of my knowledge.

Under DPS, a 25 MHz CPU32 has under 2 uS (miscorseconds) interrupt latencsy.
Probably less than that for a 250 MHz 8240 - I have no numbers here, I use
it since recently. A cache miss results indeed in refilling cachelines,
but this is a matter of nanoseconds rather than microseconds. Then there
is a way to lock some part(s) of the cache with the interrupt code
inside - though this is a resource which must be considered more
than once before usage, locking some cache 'forever' is no small thing.

The bottom line - in answer to the original question is:

Yes, you can achieve interrupt latencies well below 1 uS nowadays.
(But you have to limit your usage of wintel devices to office
applications, which is what they are meant to do).


Dimiter


--------------------------------------------------------------------
Dimiter Popoff
Transgalactic Instruments, Gourko Str. 25 b, 1000 Sofia, Bulgaria
http://transgalactic.freeyellow.com   <---- now with the new Nukeman
Email: tgi@cit.bg, tgi_earth@yahoo.com
Phone: 00359/2/566752