Red Hat Bugzilla – Bug 26634
[noapic] Kernel Bug, 64bit slot, adaptec ctrlr, DL360
Last modified: 2005-10-31 17:00:50 EST
Red Hat Linux Beta - beta3
Feb 7, 2001
Failing HW: DL360
Kernel Bug, 64bit slot, adaptec ctrlr, DL360, when installed in 64bit slot.
When installing on a DL360 with a Compaq UW SCSI controller (Adaptec
chipset) installed in a 64bit slot the server generates the following
error on the first reboot, after a successful install:
aic7xxx_abort returns 8194
kernel BUG at /usr/src/bs/BUILD/kernel-2.4.0/linux/include/asm/pci.h:61!
invalid operand 0000
When the controller is installed in a 32bit slot you do not get this error.
Compaq Bug #: 194235
Compaq consideres this defect a MUST-FIX for Florence.
Assigning to correct component.
This defect is considered MUST-FIX for Florence Gold release
The aic7xxx driver has been changed between beta 3 and current. Please try with
the new aic7xxx driver (it can be identified by the version string, if you do a
cat of /proc/scsi/aic7xxx/0 the new driver version is 5.2.3 and the known broken
driver versions are 6.0.8BETA and possibly 6.1.0). If the problem still exists,
then reopen this bug report. Note, kernel builds 2.4.1-0.1 and later have the
5.2.3 driver in them. Other kernels you will need to check individually to see
what aic7xxx driver is in them. If it has one of the 6.x.x aic7xxx driver
versions, then you can load the aic7xxx_old module instead to get the 5.2.1
aic7xxx driver version. The 5.2.1 driver is slightly outdated, but might work
where a 6.x.x driver does not.
Where can we get the driver to test with.
I dont get a kernel panic now but i continually get scsi bus timeouts / resets
testing on Red Hat 7.1 RC1.
What is a DL360? Is it x86 or alpha or ia64? If it's x86, then are you getting
continuous timeouts/resets during install or only after rebooting into the
a DL360 is an i386 system with a ROC chip installed (array controller down).
The scsi bus resets happen after reboot. It is ok during the install.
I'm assuming this is an SMP machine. Try booting the newly installed system
with the option noapic. Passing this option to the kernel command line will
change the IRQ routing of the system and from the sounds of it, that's what the
problem it. If it starts working when you boot with this option, then the bug
is in the MP/ACPI parsing code and the IRQ routing table, not the aic7xxx
The following SCSI issues should prove to be dependent upon the APIC fix found
in Linux Kernel 2.4.2-patch5. These SCSI issues are being tested at this time,
now that the latest beta seems to have the needed patch.
30170 - Full-Table-Mapped APIC mode causes a problem with installing Kernel
Linux version 2.4.1-0.1.9smp or later.
26632 - Cannot install to a Compaq array controller - hang
26633 - SCSI Timeout Errors on NCR53c1510D / sym53c1510D
26634 - [aic7xxx] Kernel Bug, 64bit slot, adaptec ctrlr, DL360
qa0309 does not have the fix needed. Drivers still error out (timeouts,
This information was already forwarded to RedHat. Just wanted to update
I think weve found the problem thats breaking SMP/APIC support on ServerWorks
It appears that when the 2.4.0 kernel was released (on Jan. 4), that an old
version of drivers/ide/osb4.c crept back in. Whenever we build a kernel with
this suspect version enabled, it breaks SMP APIC interrupts.
If you do a diff of osb4.c from 2.4.0-prerelease and 2.4.0, youll find a ton
of changes. Our guess is that its one of these changes (probably chipset
initialization) thats breaking APIC support with SMP kernels.
Here is some more information from our developers.
The below fix works!
I see 2 basic ways for this bug to be fixed:
1) Make sure that the
line in the kernel .config file is NOT set, or
2) Replace the osb4.c file with the osb4.c file from the 2.4.0-prerelease
(this is probably a better solution)
This is fixed in our kernels as of version 2.4.2-0.1.35
If this or newer kernels don't solve the problem, please reopen the bug.