Package: linux-image-2.6.26-2-686-bigmem
Version: 2.6.26-15
Hello,
The Linux kernel sources upon which
linux-image-2.6.26-2-686-bigmem is based are suffering from a
bug, which only affects the (relatively rare) summit SMP
architecture. It is found in IBM xSeries servers such as the
x440 or x445.
On these servers, linux-image-2.6.26-2-686-bigmem only correctly
identifies and uses one CPU, regardless of the number of CPUs
actually installed.
The bug is documented and resolved on linux-kernel:
http://lkml.indiana.edu/hypermail/linux/kernel/0807.3/0493.html
The patch responsible for the resolution is the following:
--- a/include/asm-x86/mach-summit/mach_apic.h 2008-07-24 18:06:39.0
-0700
+++ b/include/asm-x86/mach-summit/mach_apic.h 2008-07-24 18:09:29.0
-0700
@@ -122,7 +122,7 @@ static inline physid_mask_t ioapic_phys_
static inline physid_mask_t apicid_to_cpu_present(int apicid)
{
- return physid_mask_of_physid(0);
+ return physid_mask_of_physid(apicid);
}
I report the bug against a lenny kernel, as I don't have any such system
running
testing or unstable. The current linux kernel source in unstable,
linux-2.6_2.6.29-5,
appears to have the file structure reorganized. The relevant file appears to be
arch/x86/include/asm/summit/apic.h, and the bug appears still to be present:
apicid_to_cpu_present still returns physid_mask_of_physid(0) instead of
(apicid).
However, currently I cannot test whether the fix is still relevant.
If it is of help, I could try to boot an unpatched linux-2.6_2.6.29-5 kernel on
a
summit machine and compare the number of identified CPUs with a patched
linux-2.6_2.6.29-5
kernel.
In any case, it could probably be a good idea to include the fix in the next
lenny
kernel update, as it does render those multi-CPU machines relatively useless,
and
the fix does not affect anyone else.
With kind regards,
--
Andreas Trottmann
--
To UNSUBSCRIBE, email to debian-kernel-requ...@lists.debian.org
with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org