kernels to test for 3ware problems
People who had problems with the SLES8-SP3 or 9.0 kernel and 3ware controllers please test if the kernels in ftp://ftp.suse.com/pub/people/ak/test11/* fix the problem. [they should work on both 9.0 and SLES8]. The test makes only sense when you have more than 3GB memory in your machine. -Andi
Andi Kleen wrote:
People who had problems with the SLES8-SP3 or 9.0 kernel and 3ware controllers please test if the kernels in
ftp://ftp.suse.com/pub/people/ak/test11/*
fix the problem. [they should work on both 9.0 and SLES8]. The test makes only sense when you have more than 3GB memory in your machine.
-Andi
I still get file corruption with this kernel and 8GB of memory. Mike
On Mon, Nov 24, 2003 at 02:03:26PM -0800, Michael Madore wrote:
Andi Kleen wrote:
People who had problems with the SLES8-SP3 or 9.0 kernel and 3ware controllers please test if the kernels in
ftp://ftp.suse.com/pub/people/ak/test11/*
fix the problem. [they should work on both 9.0 and SLES8]. The test makes only sense when you have more than 3GB memory in your machine.
-Andi
I still get file corruption with this kernel and 8GB of memory.
Thanks for testing. Can you please try test12/* in the same directory too? [mirror still running, may lag a bit] Thanks, -Andi
Andi Kleen wrote:
On Mon, Nov 24, 2003 at 02:03:26PM -0800, Michael Madore wrote:
Andi Kleen wrote:
People who had problems with the SLES8-SP3 or 9.0 kernel and 3ware controllers please test if the kernels in
ftp://ftp.suse.com/pub/people/ak/test11/*
fix the problem. [they should work on both 9.0 and SLES8]. The test makes only sense when you have more than 3GB memory in your machine.
-Andi
I still get file corruption with this kernel and 8GB of memory.
Thanks for testing. Can you please try test12/* in the same directory too? [mirror still running, may lag a bit]
Thanks, -Andi
This one seems to work. I copied the entire filesystem from /dev/hda over to the 3ware array and then ran rpm --verify on all the packages to make sure none of the binaries got corrupted. I did this twice. Mike
On Mon, 24 Nov 2003 17:54:48 -0800 Michael Madore <mmadore@aslab.com> wrote:
This one seems to work. I copied the entire filesystem from /dev/hda over to the 3ware array and then ran rpm --verify on all the packages to make sure none of the binaries got corrupted. I did this twice.
Ok. Thanks for testing. The kernel actually just undoes an IOMMU related optimization that was done between SP2 and SP3. The optimization seems to work fine on most systems (I ran all my systems for months with forced IOMMU and it in), but seems to cause problems on 3ware and qlogic. For qlogic it is normally not a problem in the production kernel because it supports 64bit IO, you can only see it in a debugging mode that forces the IOMMU for every IO. We didn't catch this in time. 3ware unfortunately doesn't support 64bit IO, so it requires the IOMMU code as soon as you have more than 3GB of main memory. What's actually wrong with the optimization is still unknown. It's probably some bad interaction with those drivers. -Andi
participants (2)
-
Andi Kleen -
Michael Madore