On Wed, 02 Jan 2002 11:13:42 -0500,
Ken Murchison <ken@xxxxxxxxxx> wrote:
>Mike Baptiste wrote:
>> The problem is this: When booting either the smp or enterprise kernels
>> (stock RH 2.4.9-XFS from the XFS installer CD), I get random kernel
>> panics. Maybe 3 out of 4 boots, at varying points in Interactive boot
>> (from LVM activiation forward), I'll get a Kernel Panic from the DAC960
>> driver:
>>
>> DAC960#0: SegmentNumber != SegmentCount
>
>I have the same problem with my quad Xeon. No problem with RH
>2.4.3-SGI_XFS_1.0.1 or 2.4.14-SGI_XFS_1.0.2. I haven't been able to try
>a stock RH 2.4.9-13 (without the XFS bits) because all of my filesystems
>are XFS
The SMP kernels for 2.4.9-13SGI_XFS_PR1 are broken. I just installed a
new machine, install (UP) was fine, booting the SMP kernel broke at
random in the ncr53c8xx SCSI driver. Booting the UP kernel avoided the
oops, but then I hit Al Viro's "corrupt hard links and super blocks"
bug. These are all problems caused by bugs in Linus's kernel.
I used another machine to compile 2.4.17-XFS from the CVS tree and it
runs fine in SMP on the machine that was failing. Installing was
"interesting" because links did not work, preventing the use of rsync
to copy the new kernel to the target machine. In the end I created a
tarball of module and kernel, transferred it by ftp and unpacked on the
target system. Then sync several times, boot into single user mode on
2.4.17-XFS and run e2fsck or xfs_repair on each partition to find any
corrupt inodes left by Al Viro's 2.4.9 bug.
I don't trust any kernel between about 2.4.9 and 2.4.16. There were
too many bugs in the base kernel code, not XFS. 2.4.17-XFS from CVS is
looking nice and stable.
|