Announcement

Collapse
No announcement yet.

Sandy Bridge PCI Card Drivers Fail with "Disabling IRQ"

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Sandy Bridge PCI Card Drivers Fail with "Disabling IRQ"

    I wasn't aware of problems with PCI cards on Sandy Bridge (until I bought an Asus P8Z68-V LX) even though there seem to be quite a few reports out there. No fix appears to have been developed (I have tried 3.1-rc10). The only thing to help is booting with irqpoll kernel parameter which seems to reduce the number of occurrences.

    Most reports appear to be about Asus boards, but I did find one about an Intel DH67BL:
    http://lists.centos.org/pipermail/ce...ry/104590.html

    Some other reports, there are some links in them to bugzilla.kernel.org but of course they are impossible to read at the moment.
    https://lkml.org/lkml/2011/6/30/197
    https://lkml.org/lkml/2011/10/14/146
    https://bugzilla.redhat.com/show_bug.cgi?id=713351
    https://bugs.launchpad.net/ubuntu/+s...ux/+bug/877588
    http://ubuntuforums.org/showthread.php?t=1676491

  • #2
    Seems to be a more general regression introduced by IRQ code re-factoring in recent kernels:
    https://bugs.launchpad.net/linux/+bug/855199

    One fix already committed for 3.2:
    http://git.kernel.org/?p=linux/kerne...ea383621027bf2

    Is there more needed?:
    https://lkml.org/lkml/2011/11/19/189

    Thanks Edward!!!

    Comment


    • #3
      First fix now in 3.1.3 too and second accepted for 3.2:
      http://git.kernel.org/?p=linux/kerne...88d5818f57f62a

      But... although they plus irqpoll reduces the frequency it is still not completely fixed for me.

      Comment


      • #4
        Looks like suspicion is falling on the ASM1083 PCI Express-to-PCI bridge chip:
        https://lkml.org/lkml/2011/12/9/129

        Hopefully this interest means there will be a complete solution.

        Comment


        • #5
          The Sandy Bridge chipsets don't have PCI support anymore, that's why board vendors use a bridge chip. The ASM1083 is quite a common one but ASMedia doesn't seem to be very Linux-friendly. Usually PCI or PCI Express bridges just work, without needing a driver, but this chip is buggy and probably only tested with Windows.

          I hope motherboard vendors will start to use JMicron chips instead. JMicron supports Linux somehow, as they list the minimal kernel version required for a device to work on their website.

          Comment


          • #6
            The Intel DH67BL mentioned above uses an ITE IT8892E:
            http://www.lostcircuits.com/mambo//i...1&limitstart=4
            http://www.overclockersclub.com/vima...5_2500k/54.jpg

            Not that Intel claims Linux support:
            http://www.intel.com/support/motherb.../cs-008326.htm

            Gigabyte also appear to use ITE, with the GA-H67A-USB3-B3 manual saying it has an ITE IT8892.

            It would be interesting to hear from anyone that does have PCI working.

            Comment


            • #7
              That ITE chip is not the PCI bridge. It's the Super I/O chip which means it handles PS/2, parallel, serial, and floppy, and often also hardware monitoring, along with a lot of BIOS stuff. It's not related to PCI at all. You can't even see the Super I/O chip of a motherboard in the lspci output.

              The first link clearly mentions it as a Super I/O chip.

              Comment


              • #8
                Here's a picture of the ASMedia chip:
                http://www.modlabs.net/uploads/galle...ia-asm1083.jpg

                Comment


                • #9
                  Every link I can find says the ITE IT8892E is the PCI bridge - although I can't find an actual specification.

                  Comment


                  • #10
                    (please remove this post)
                    Last edited by AlbertP; 12-24-2011, 06:30 AM.

                    Comment


                    • #11
                      Including Gigabyte:
                      http://www.manualowl.com/m/Gigabyte/.../207430?page=8

                      Anyway, the real point was what is the PCI bridge on the Intel DH67BL? Are there more bridges than the ASMedia causing problems on Linux? Are there any PCI bridges that are working with Linux?

                      Comment


                      • #12
                        A PCI bridge shouldn't even need a driver. They just work, like the ASM1083 did in previous kernels. But the chip is buggy, so you have to copy Windows's behaviour on PCI bridges to get it working, instead of just following the specs.

                        Nearly every chipset has a PCI bridge built-in. I know two exceptions:
                        • Newer Intel chipsets lack PCI so they use a third-party chip.
                        • SiS has an entirely different configuration on their chipsets. The OS talks directly to the PCI devices, through the northbridge, and doesn't know that the southbridge's PCI bridge is in-between. This works fine on Linux, including IRQs.
                        All those PCI bridges in chipsets just work. Intel, AMD, nVidia, all working fine.
                        I'm sorry. Most ITE devices on motherboards are Super I/O chips but they also appear to sell PCI bridges. I had never seen those chips before while ITE I/O chips are quite common, also on motherboards where the PCI is built-in into the chipset.
                        The chip is, however, not listed on ITE's webpage. I couldn't find any official documentation on it.
                        I've removed my previous post.
                        Last edited by AlbertP; 12-24-2011, 06:53 AM.

                        Comment

                        Working...
                        X