PCI: derive nearby CPUs from device's instead of bus' NUMA information
authorAndreas Herrmann <andreas.herrmann3@amd.com>
Fri, 17 Apr 2009 10:01:55 +0000 (12:01 +0200)
committerJesse Barnes <jbarnes@virtuousgeek.org>
Fri, 6 Nov 2009 22:09:15 +0000 (14:09 -0800)
In case of AMD CPU northbridge functions this NUMA information might
differ.  Here is an example from a 4-socket system.

Currently Linux shows

  root@hagen:/sys/devices/pci0000:00/0000:00:1a.4# cat numa_node
  0
  root@hagen:/sys/devices/pci0000:00/0000:00:1a.4# cat local_cpu*
  0-3
  00000000,0000000f

which is not correct for northbridge functions as the local CPUs
are those of the same socket.

With this patch and a quirk for AMD CPU NB functions Linux can
do better and correctly show

  root@hagen:/sys/devices/pci0000:00/0000:00:1a.4# cat numa_node
  2
  root@hagen:/sys/devices/pci0000:00/0000:00:1a.4# cat local_cpu*
  8-11
  00000000,00000f00

Signed-off-by: Andreas Herrmann <andreas.herrmann3@amd.com>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
drivers/pci/pci-sysfs.c

index 0f6382f..0fa707e 100644 (file)
@@ -74,7 +74,11 @@ static ssize_t local_cpus_show(struct device *dev,
        const struct cpumask *mask;
        int len;
 
+#ifdef CONFIG_NUMA
+       mask = cpumask_of_node(dev_to_node(dev));
+#else
        mask = cpumask_of_pcibus(to_pci_dev(dev)->bus);
+#endif
        len = cpumask_scnprintf(buf, PAGE_SIZE-2, mask);
        buf[len++] = '\n';
        buf[len] = '\0';
@@ -88,7 +92,11 @@ static ssize_t local_cpulist_show(struct device *dev,
        const struct cpumask *mask;
        int len;
 
+#ifdef CONFIG_NUMA
+       mask = cpumask_of_node(dev_to_node(dev));
+#else
        mask = cpumask_of_pcibus(to_pci_dev(dev)->bus);
+#endif
        len = cpulist_scnprintf(buf, PAGE_SIZE-2, mask);
        buf[len++] = '\n';
        buf[len] = '\0';