Bug 78234

Created attachment 85666 [details]
my 2.4.20-rc2 config

I an attest that 2.4.20-rc4 does in fact work with L440GX+ board with a single
CPU installed.  I also have a DAC960 installed as well, root on raid5.

Here is what I did.

Boot net (dhcp) via "linux rescue"
chroot /mnt/sysimage
get linux-2.4.19.tar.gz via kernel.org
get patch-2.4.20-rc4.gz via kernel.org

patch /usr/src/linux-2.4.19 to make 2.4.20-rc4
rename /usr/src/linux-2.4.19 to /usr/src/linux-2.4.20-rc4

recompile kernel via (make oldconfig) using config-2.4.18-18smp as basis.

mkinitrd -f /boot/initrd-2.4.20-rc4smp.img 2.4.20-rc4
update lilo.conf, etc.

install and boot

I an attest that 2.4.20-rc4 does in fact work with L440GX+ board with a single
CPU installed.  I also have a DAC960 installed as well, root on raid5.

Here is what I did.

Boot net (dhcp) via "linux rescue"
chroot /mnt/sysimage
get linux-2.4.19.tar.gz via kernel.org
get patch-2.4.20-rc4.gz via kernel.org

patch /usr/src/linux-2.4.19 to make 2.4.20-rc4
rename /usr/src/linux-2.4.19 to /usr/src/linux-2.4.20-rc4

recompile kernel via (make oldconfig) using config-2.4.18-18smp as basis.

mkinitrd -f /boot/initrd-2.4.20-rc4smp.img 2.4.20-rc4
update lilo.conf, etc.

install and boot

I'd like to report having success with downloading 2.4.20 from kernel.org, 
copying RedHat's 2.4.18-19 .config, using make menuconfig to set APIC and 
IOAPIC and building/installing/booting 2.4.20.  It seems to work, so far!

This is an EMC AV1400 (a 440GX mb).

Tim

FWIW If you recompile the redhat kernels with local APIC support for 
uniprocessors and IO-APIC support for uniprocessors enabled (You do not need 
to make any other changes) the std RedHat kernels work. I even built them as 
rpms because I have several machines with the 440 chipset and it is easier for 
me to keep track that way. Arjan has said in the past that the reason they 
(Red Hat) will not build their kernels that way was because this breaks too 
many laptops.

I tried this and I was still unable to get the machine to boot.  In my case I 
do have an additional Adaptec PCI SCSI board installed, maybe that had 
something to do with it.

So if this is the only problem, maybe a switch or something can be added to 
enable this at boot time? (like "linux apic", but that only seems to work on 
the BOOT kernel).

What works depends on a lot of random factors. Intel didn't want to provide the
needed info to do something about this on the 440GX. That seems to be changing,
fingers crossed.

I'd be interested to know if the boot option  "pci=biosirq" works

pci=biosirq doesn't help me on my L440GX+, bios 11.1.

I saw this in an announcment on lkml for a recent ac kernel: 
Improved 440GX bios workarounds                 (Arjan van de Ven) 
        | Thanks to the guys at Intel for hints on this 
 
Is there any time line for getting these fixes either into an official errata 
kernel or rawhide?? 

We have an L440GX+ board that exhibits this problem.  The various "pci=", SMP
kernel, and "apic" workarounds did not help.  Fortunately, the system we were
installing to had a RAID controller (used a different driver), and an IDE
CD-ROM.  So, we did a "noprobe" install, skipped the "aic7xxx" driver, and got a
working system.  We then downloaded kernel 2.4.20 (final release, pristine
sources), did a custom build, and tried that.  It worked perfectly.  It appears
that those fixes Alan Cox mentioned did the trick.

Actually we know that you can recompile your kernel such that the bug doesn't
show. However we can't do that since it breaks tons of others machines.

There was once a precomplied kernel to download for 7.3 to fix this issue are 
you still not providing that for 8.0 now as well?

there was no such kernel for RHL7.3, and there won't be one for RHL8.0
For RHL9 most of this should just work for anyone who has submitted their bios
ID to us.

Hummm, How does one tell if "their bios id" was submitted?? I would like to 
have an idea if this is going to work before doing the upgrade. Is there a 
list I can look at to compare to my bios id?? Is the bios id something I get 
with dmidecode?? If it is not submitted is submitting it here sufficient to 
get in included in future errata kernels?? 

> How does one tell if "their bios id" was submitted??

if you added your dmidecode to a bug in bugzilla I have it added.

dmidecode will show you your bios ID's....
If it starts with L440GX0.86B then it'll work most likely
There is no way to do an erratum for this, since you can't even install if it's
not working so you can't actually get to the point where you could use the erratum.

Your correct it was not 7.3 it was 7.1 where Redhat provided 3 different 
boot.img files for us to use instead.


So regarding that you still providing those specific boot disks?  And please 
read the old bugzilla 29555 cause there were more helpful people there when I 
had this issue.

Arjan van de Ven/Red Hat:

Could you please elaborate on the comment, "Actually we know that you can
recompile your kernel such that the bug doesn't show. However we can't do that
since it breaks tons of others machines."  If you are referring to the "enable
APIC on uni-processor" option fix, please note that it did *not* work for our
system.  See comment #5 of bug 77090 and comment #3 of bug 55358.

If you are referring to a *different* kernel recompile fix, please let us know
what it is.

For completeness, all of the following bugs appear to be related to this one:
Bug 29555 (RHL 7.1), bug 55358 (RHL 7.2), bug 64971 (RHL 7.3), bug 77090 (RHL
8.0), and bug 77643 (RHL 8.0).  The "APIC fix" has worked for some but not everyone.

WHOOHOOOOO!! I just upgraded my 440 to kernel-2.4.20-13.8. It boots and so far 
runs just fine (uptime=39 minutes but I am VERY hopeful). This is a RHL 8.0 
system. Looks like the fixes worked!! Can anyone else verify my experience?? 

yep, the new 2.4.20 kernel spits out a nice little message saying the bios is
known to have issues and talks about using smp as a workaround, but then it goes
ahead and actually works.  prior to this, the scsi was finding irq 11, but now
it finds irq 10 and things actually seem to work.  it resync'ed the raid1
mirrors and did all the upgrades (for 7.2) without problem.  intel should be
ashamed.