Version-Release number of selected component (if applicable): 2.6.18-92.1.1.el5 At boot time only two of the four ports are activated, with the diagnostic for other two ports being: eth0: No MII transceiver found, aborting. ASIC status ffffffff I found a LKML post describing the exact same symptoms, a diagnosis, and a partial workaround: http://lkml.org/lkml/2007/5/4/376 It's only a partial workaround at the moment because I cannot get all interfaces to come up without manual intervention. I've tried adding this to modprobe.conf: install sundance setpci -s 04:00.0 3e.b=00; /sbin/modprobe --ignore-install sundance That doesn't seem to have any effect at boot time, but if I then log in and do "rmmod sundance; modprobe sundance", the four interfaces all come up. I've also tried this in modprobe.conf to try to reproduce the manual step: install sundance setpci -s 04:00.0 3e.b=00; /sbin/modprobe --ignore-install sundance; /sbin/modprobe -r sundance; sleep 2; /sbin/modprobe --ignore-install sundance That didn't help either. I appreciate that this is probably a BIOS problem but a workaround that could bring all four interfaces up at boot time without manual intervention would be very helpful.
Created attachment 310402 [details] dmesg output immediately after booting
Created attachment 310403 [details] Output of lspci -vvxxx
Same thing happens with 2.6.18-128.el5.
Same thing happens with 2.6.18-194 Workaround for PCI-to-PCI Bridge NoISA bit also works.
Some notes for better understanding of the bug: The DFE-580TX card consists of 4 Sundance DL10050 chips on a PCI bus which is connected to the system's PCI bus via the PCI-to-PCI bridge (Intel 21152). A picture of the card confirms it: http://www.lanshop.co.uk/CategoryImages%5CDFE580TXl.jpg Datasheet of the Intel 21152 bridge: http://download.intel.com/products/bridge/datashts/278060.pdf Articles describing the ISA aliasing problem (i.e. why the "ISA enable" was introduced): http://www.microsoft.com/whdc/system/sysinternals/PartialAddress.mspx http://www.techarp.com/showarticle.aspx?artno=606 Does the system's BIOS have any options related to ISA aliasing?
Unfortunately no options exist about ISA Aliasing.Actually the card is Intel D410Pt and has the lightest BIOS features I have ever seen. http://www.intel.com/products/desktop/motherboards/D410PT/D410PT-overview.htm
Oguz Yilmaz, would you also attach the output of "dmesg" and "lspci -vvxxx" please?
My motherboard doesn't have any options relating to ISA aliasing either. I can attach dmesg and lspci output from an EL 5.5 kernel now if it's useful (original data in Comment #1 and Comment #2).
Created attachment 415642 [details] Dmesg output
Created attachment 415643 [details] lspci -vvxxx output
After investigating the issue and going through upstream commits and PCI code, I think that it affects only specific similar types of BIOS (which incorrectly configure the I/O range). I couldn't find fixes in upstream for this, however I could find similar issues dating back from 2002 which got custom workarounds but since they changed general functions for specific needs, they weren't integrated in upstream. Disabling the NoISA flag in BridgeCtl register before loading the module is a valid workaround.
Nikolay, thank you for the research about this problem. So it appears that the impact is limited to specific BIOSes and the past attempts to get workarounds for this into upstream failed. Therefore I'm closing this as WONTFIX.