[LAU] irq sharing

Mark Knecht markknecht at gmail.com
Tue Aug 24 19:42:53 UTC 2010

On Tue, Aug 24, 2010 at 8:55 AM, David Santamauro
<david.santamauro at gmail.com> wrote:
>> David,
>>    Well, at first blush that implies to me this has nothing to do with
>> interrupts. Is the any card good? Have you tried it in another system?
> the card works fine on the same hardware under 64-bit windows7. I'm
> trying to get it working 100% in fedora 12 64-bit with an rt-kernel
> (multi-os machine). I agree, interrupts are probably not the issue.
> Last time I was fiddling with this problem I had suspected 64-bit
> linux drivers as it works in the 32-bit machine I have.
>>    This was a long time ago in my chip design architect history but I
>> helped write one of the early versions of the PCI-x spec for bridging
>> devices. IIRC PCI-x host controllers were supposed to correctly handle
>> both 32-bit and 64-bit PCI cards when plugged into those slots so
>> (according to the original spec written maybe 12 years ago) if your
>> card physically plugs into whatever connectors your MB provides it
>> should work. (I.e. - PCI-x slows down to become PCI.) However if you
>> had any PCI-x cards they would slow down also. Not a problem in your
>> case it seems.
>>    Obviously we don't want to damage anything so I'd check your MB
>> manual on this, as well as looking at any BIOS for any settings or
>> clues about allowing PCI cards in PCI-x slots. You'll find a
>> supporting position here:
>> http://en.wikipedia.org/wiki/Pci-x
> I read that as well, but my MB pci-x slots are (apparently) backwards
> (pardon my ignorance)
> see page 7
> http://www.tyan.com/manuals/m_s5396_120.pdf
> ... backwards, meaning, I'd have to stick the card in backwards.
> David

No, the PCI-X slots aren't backwards but you __MIGHT__ have a card
that cannot be inserted safely.

Look at page 30. Notice that the PCI connector (very bottom) matches
up against the __length__ of the first two sections of  PCI-X 64-bit
connector. (Just above it) That's where the PCI-X spec expects the PCI
card to be placed.

The reason you are thinking it's backwards is because of the way the
smallest section of each connector is positioned. However that's not
what's really going on with the cards.

There are two kinds of PCI cards - those with a single indentation in
the gold connector area on the card and those with two indentations.
Cards with a single indentation required 5 volts (IIRC) and those with
2 indentations did not. A PCI card with two indentations only used 3.3
volts and can be safely inserted in the PCI-X connector because that
second indentation allows it to fit. If the card doesn't have two
indentations then that piece of plastic in the connector (which is
making you think things are backward) will keep the card out and thus
protect it from voltages that it's not designed to use.

Again, go slowly and maybe find a copy of the spec to read. I am doing
this from memory circa 1997 so it's very vague and should be verified
somewhere. I've never actually owned a PCI-X machine - the spec never
because important in the retail market but is much loved in the server
market where lots of RAID controllers do PCI-X.

Hope this helps, and again, BE CAREFUL!

- Mark

More information about the Linux-audio-user mailing list