Re: 5 PCI slots, 5 cards: Are IRQ conflicts inevitable?



clowes_ian@xxxxxxxxxxx wrote:

Hi

I've never quite understood the PCI IRQ sharing thing, but am assuming
it's causing a problem I'm seeing on my 4 PCI, 1 ISA, 1 shared slot
mobo.

Until recently I ran my server with the following cards:

- Graphics (PCI)
- Network 1 (PCI)
- Network 2 (PCI)
- SCSI (PCI)
- Backup modem (ISA)

When I added a PCI DVB-T card one of the network cards stopped working
(media errors on the console). I've found that removing the ISA card
and moving the PCI cards around improves things cosmetically (clean
boots, no media errors) but I never got all the cards working.

I've also tried disabling the onboard serial and EPP cards, but this
didn't seem to bring any new IRQs into use.

In short, I found that I had to remove the SCSI card. This is no big
problem, since it was mainly used for a SCSI /tmp device which I've
moved to IDE and a 'legacy' Jaz drize last used about 6 years ago, but
it would potentially be handly to keep the SCSI available.

Is it reasonable to want to put these 5 PCI cards in, or is there some
design limitation that means it'll never work? Are there more
'standard' BIOS tricks that I should be pulling, or indeed some kernel
configuratiuon?

Four problems. One, there aren't enough interrupts available to handle the
number of devices to be attached. Two, Plug and Play is an attempt to
idiot proof a system by replacing the human idiot with an artificial
idiot--it has no means of optimizing IRQ assignments based on knowledge of
the anticipated load on each device, nor can it prioritize based on the
desired boot order because it doesn't know that either--one hopes that
there is a special Hell for whoever decided to implement Plug and Play with
no manual override. Three, on most boards two slots are hard wired to the
same interrupt trace for some unfathomable reason--you'll need to read the
docs for the board to determine which, if the docs even say. Four,
interrupt sharing doesn't work well when both devices sharing the interrupt
generate a high number of interrupts, as is likely the case with a disk
controller or network interface.

Your ISA modem adds to the problem by removing an interrupt from the
sharable pool--ISA devices cannot in general share interrupts.

You're pretty much going to have to play musical boards until you either
find an ordering that works or have tried all 120 or so possible orderings
and found that none work.


--
--John
to email, dial "usenet" and validate
(was jclarke at eye bee em dot net)
.



Relevant Pages

  • Re: ISDN testing
    ... ITK ix1-micro 2.1 ISA ... Advantage of using the two cards is that it covers both PCI and ISA. ... But when I tried to modprobe hisax, ...
    (Linux-Kernel)
  • Re: [PATCH] yenta: irq-routing for TI bridges - take 2
    ... I understand you are trying serial ISA interrupts ... The known problems are with PCI ... which have PCI interrupts and don't need anything else. ... may be there are cards out there that need an exclusive interrupt. ...
    (Linux-Kernel)
  • Re: ATI Radeon 9600se VS ATI X300se
    ... if Phil's motherboard only has PCI Express slot he should be aware ... higher end PCIex cards burn much hotter- think about system cooling. ... > computer that uses AGP graphics cards. ... Note that computers that use PCI Express 16 video cards ...
    (microsoft.public.windowsxp.hardware)
  • Re: SCO Technical Articles to say "tata".
    ... > cards in a machine that needs to perform well. ... It's good that ISA ... The pci version are also much ... When they ported to sco, ...
    (comp.unix.sco.misc)
  • Re: What motherboards are based on Intel 915 and Intel 945 chipsets
    ... The vendors of the cards prefer Intel because in ... known to work well with PV149 or Kodicom series DVR cards I'll be glad ... Is 1 PCI bus enough? ... I suspect that if you are using 5 capture cards, ...
    (comp.sys.ibm.pc.hardware.chips)