Date: Thu, 20 Aug 2009 10:10:34 +1000
From: Anton Blanchard <anton@samba.org>
To: linux-numa@vger.kernel.org
Cc: cpw@sgi.com, ak@linux.intel.com
Subject: [PATCH] libnuma: Fix issue with numactl --hardware on sparse cpumaps

On a machine with sparse cpu ids, numactl fails to print the right cpus:

# numactl --hardware | grep cpus
node 0 cpus: 0 2 4 6
node 1 cpus:

It turns out we were iterating through 0..nr_cpus, not 0..max_cpuid. The
following patch fixes it by using a cpumask instead of an open coded bitmask,
and looping from 0 to the size of the cpumask.

# numactl --hardware | grep cpus
node 0 cpus: 0 2 4 6
node 1 cpus: 8 10 12 14

Signed-off-by: Anton Blanchard <anton@samba.org>
---

diff -ru numactl-2.0.3~/numactl.c numactl-2.0.3/numactl.c
--- numactl-2.0.3~/numactl.c	2009-06-10 07:30:03.000000000 -0500
+++ numactl-2.0.3/numactl.c	2009-08-19 18:57:10.064311316 -0500
@@ -200,14 +200,13 @@
 
 void print_node_cpus(int node)
 {
-	int conf_cpus = numa_num_configured_cpus();
 	int i, err;
 	struct bitmask *cpus;
 
-	cpus = numa_bitmask_alloc(conf_cpus);
+	cpus = numa_allocate_cpumask();
 	err = numa_node_to_cpus(node, cpus);
 	if (err >= 0) 
-		for (i = 0; i < conf_cpus; i++) 
+		for (i = 0; i < cpus->size; i++) 
 			if (numa_bitmask_isbitset(cpus, i))
 				printf(" %d", i);
 	putchar('\n');

