Re: SMP Panic caused by [PATCH] sched: consolidate sched domains

[Jesse Barnes <jbarnes@engr.sgi.com>]

[-- Attachment #1: Type: text/plain, Size: 881 bytes --]

On Sunday, August 29, 2004 9:58 am, James Bottomley wrote:
> On Sun, 2004-08-29 at 12:48, Jesse Barnes wrote:
> > But I think this breaks what the code is supposed to do.  You're right
> > that we shouldn't use cpu_online_map, but we should leave the nodemask in
> > there and fix the code that sets it in the non-NUMA case instead.
>
> Well, let's say it puts back the original behaviour. If you look at even
> the NUMA code before these changes, it had cpu_possible_map in there.
>
> I totally agree about fixing NUMA, it looks completely broken to me in
> the way it handles cpu maps because node_to_cpumask(i) needs to expand
> to cpu_possible_map for initialisation and cpu_online_map for
> operation.  Has anyone ever checked NUMA for hotplug CPU?

I've up and downed a few CPUs on an Altix, and it seems to work ok, but that's 
a pretty basic test.  How about this?

Jesse


[-- Attachment #2: node-to-cpumask-fix.patch --]
[-- Type: text/x-diff, Size: 634 bytes --]

===== include/asm-generic/topology.h 1.6 vs edited =====
--- 1.6/include/asm-generic/topology.h	2004-02-03 21:35:17 -08:00
+++ edited/include/asm-generic/topology.h	2004-08-29 10:06:17 -07:00
@@ -36,13 +36,13 @@
 #define parent_node(node)	(0)
 #endif
 #ifndef node_to_cpumask
-#define node_to_cpumask(node)	(cpu_online_map)
+#define node_to_cpumask(node)	(cpu_possible_map)
 #endif
 #ifndef node_to_first_cpu
 #define node_to_first_cpu(node)	(0)
 #endif
 #ifndef pcibus_to_cpumask
-#define pcibus_to_cpumask(bus)	(cpu_online_map)
+#define pcibus_to_cpumask(bus)	(cpu_possible_map)
 #endif
 
 /* Cross-node load balancing interval. */