On Tue, 8 Jul 2008 00:07:03 -0700 Yinghai Lu wrote: > On Mon, Jul 7, 2008 at 10:28 PM, Randy Dunlap wrote: > > On Mon, 7 Jul 2008 17:24:16 -0700 Yinghai Lu wrote: > > > >> On Mon, Jul 7, 2008 at 3:40 PM, Randy Dunlap wrote: > >> > On Mon, 7 Jul 2008 11:39:17 -0700 Randy Dunlap wrote: > >> > > >> >> On Mon, 7 Jul 2008 08:32:18 +0200 Ingo Molnar wrote: > >> >> > >> >> > > >> >> > * Linus Torvalds wrote: > >> >> > > >> >> > > > This still happens with 2.6.26-rc9. Using CONFIG_NUMA=y boots OK. > >> >> > > > >> >> > > Ok, then it wasn't the nr_zones thing. > >> >> > > > >> >> > > Since it seems to be repeatable for you, can you bisect it? > >> >> > > >> >> > one guess would be: > >> >> > > >> >> > | commit e8ee6f0ae5cd860e8e6c02807edfa3c1fa01bcb5 > >> >> > | Author: Yinghai Lu > >> >> > | Date: Sun Apr 13 01:41:58 2008 -0700 > >> >> > | > >> >> > | x86: work around io allocation overlap of HT links > >> >> > > >> >> > but ... since CONFIG_NUMA makes it work, i'm not sure about that. > >> >> > > >> >> > Randy, could you post the full CONFIG_NUMA bootlog as well, does it show > >> >> > any difference in resource allocations? > >> >> > >> >> Good and bad boot logs are attached. There are several differences, but I don't > >> >> see any that are significant. > >> >> > >> >> I've started bisecting with: > >> >> > >> >> $ git bisect start > >> >> $ git bisect bad v2.6.26-rc1 > >> >> $ git bisect good v2.6.25 > >> >> > >> >> That's only about 1.29M lines of changes. > >> > > >> > git bisect and normal rebooting did not find a problem. > >> > > >> > I'll repeat this using kexec to boot the new kernel and see if that > >> > locates any issues... since I normally use kexec to load/test new kernels > >> > and that was how the failure occurred (occurs). > >> > > >> > >> same NON-NUMA kernel kexec NON-NUMA kernel? > >> > >> or other kernel kexex it? > > > > Ah. Good question. I hadn't noticed that. > > NUMA kernel kexec-ing a non-NUMA kernel now fails, but it worked in 2.6.25. > > > > can you resend out that two config? The host/first kernel that loads the second/failing kernel uses config-2625-work. The second kernel that hangs during boot uses kconfig.numa.bad . (both attached) --- ~Randy Linux Plumbers Conference, 17-19 September 2008, Portland, Oregon USA http://linuxplumbersconf.org/