[PATCH] powerpc numa: Get rid of "numa domain" terminology
authorNathan Lynch <nathanl@austin.ibm.com>
Tue, 21 Mar 2006 00:35:45 +0000 (18:35 -0600)
committerPaul Mackerras <paulus@samba.org>
Wed, 22 Mar 2006 04:03:52 +0000 (15:03 +1100)
Since we effectively treat the domain ids given to us by firmare as
logical node ids, make this explicit (basically s/numa_domain/nid/).

No functional changes, only variable and function names are modified.

Signed-off-by: Nathan Lynch <nathanl@austin.ibm.com>
Signed-off-by: Paul Mackerras <paulus@samba.org>
arch/powerpc/mm/numa.c

index 1fb11bb..a5286a6 100644 (file)
@@ -191,9 +191,9 @@ static int *of_get_associativity(struct device_node *dev)
        return (unsigned int *)get_property(dev, "ibm,associativity", NULL);
 }
 
-static int of_node_numa_domain(struct device_node *device)
+static int of_node_to_nid(struct device_node *device)
 {
-       int numa_domain;
+       int nid;
        unsigned int *tmp;
 
        if (min_common_depth == -1)
@@ -201,13 +201,13 @@ static int of_node_numa_domain(struct device_node *device)
 
        tmp = of_get_associativity(device);
        if (tmp && (tmp[0] >= min_common_depth)) {
-               numa_domain = tmp[min_common_depth];
+               nid = tmp[min_common_depth];
        } else {
                dbg("WARNING: no NUMA information for %s\n",
                    device->full_name);
-               numa_domain = 0;
+               nid = 0;
        }
-       return numa_domain;
+       return nid;
 }
 
 /*
@@ -286,7 +286,7 @@ static unsigned long __devinit read_n_cells(int n, unsigned int **buf)
  */
 static int __cpuinit numa_setup_cpu(unsigned long lcpu)
 {
-       int numa_domain = 0;
+       int nid = 0;
        struct device_node *cpu = find_cpu_node(lcpu);
 
        if (!cpu) {
@@ -294,27 +294,27 @@ static int __cpuinit numa_setup_cpu(unsigned long lcpu)
                goto out;
        }
 
-       numa_domain = of_node_numa_domain(cpu);
+       nid = of_node_to_nid(cpu);
 
-       if (numa_domain >= num_online_nodes()) {
+       if (nid >= num_online_nodes()) {
                /*
                 * POWER4 LPAR uses 0xffff as invalid node,
                 * dont warn in this case.
                 */
-               if (numa_domain != 0xffff)
+               if (nid != 0xffff)
                        printk(KERN_ERR "WARNING: cpu %ld "
                               "maps to invalid NUMA node %d\n",
-                              lcpu, numa_domain);
-               numa_domain = 0;
+                              lcpu, nid);
+               nid = 0;
        }
 out:
-       node_set_online(numa_domain);
+       node_set_online(nid);
 
-       map_cpu_to_node(lcpu, numa_domain);
+       map_cpu_to_node(lcpu, nid);
 
        of_node_put(cpu);
 
-       return numa_domain;
+       return nid;
 }
 
 static int cpu_numa_callback(struct notifier_block *nfb,
@@ -399,17 +399,17 @@ static int __init parse_numa_properties(void)
         * with larger node ids. In that case we force the cpu into node 0.
         */
        for_each_cpu(i) {
-               int numa_domain;
+               int nid;
 
                cpu = find_cpu_node(i);
 
                if (cpu) {
-                       numa_domain = of_node_numa_domain(cpu);
+                       nid = of_node_to_nid(cpu);
                        of_node_put(cpu);
 
-                       if (numa_domain < MAX_NUMNODES &&
-                           max_domain < numa_domain)
-                               max_domain = numa_domain;
+                       if (nid < MAX_NUMNODES &&
+                           max_domain < nid)
+                               max_domain = nid;
                }
        }
 
@@ -418,7 +418,7 @@ static int __init parse_numa_properties(void)
        while ((memory = of_find_node_by_type(memory, "memory")) != NULL) {
                unsigned long start;
                unsigned long size;
-               int numa_domain;
+               int nid;
                int ranges;
                unsigned int *memcell_buf;
                unsigned int len;
@@ -439,18 +439,18 @@ new_range:
                start = read_n_cells(n_mem_addr_cells, &memcell_buf);
                size = read_n_cells(n_mem_size_cells, &memcell_buf);
 
-               numa_domain = of_node_numa_domain(memory);
+               nid = of_node_to_nid(memory);
 
-               if (numa_domain >= MAX_NUMNODES) {
-                       if (numa_domain != 0xffff)
+               if (nid >= MAX_NUMNODES) {
+                       if (nid != 0xffff)
                                printk(KERN_ERR "WARNING: memory at %lx maps "
                                       "to invalid NUMA node %d\n", start,
-                                      numa_domain);
-                       numa_domain = 0;
+                                      nid);
+                       nid = 0;
                }
 
-               if (max_domain < numa_domain)
-                       max_domain = numa_domain;
+               if (max_domain < nid)
+                       max_domain = nid;
 
                if (!(size = numa_enforce_memory_limit(start, size))) {
                        if (--ranges)
@@ -459,7 +459,7 @@ new_range:
                                continue;
                }
 
-               add_region(numa_domain, start >> PAGE_SHIFT,
+               add_region(nid, start >> PAGE_SHIFT,
                           size >> PAGE_SHIFT);
 
                if (--ranges)
@@ -769,10 +769,10 @@ int hot_add_scn_to_nid(unsigned long scn_addr)
 {
        struct device_node *memory = NULL;
        nodemask_t nodes;
-       int numa_domain = 0;
+       int nid = 0;
 
        if (!numa_enabled || (min_common_depth < 0))
-               return numa_domain;
+               return nid;
 
        while ((memory = of_find_node_by_type(memory, "memory")) != NULL) {
                unsigned long start, size;
@@ -789,15 +789,15 @@ int hot_add_scn_to_nid(unsigned long scn_addr)
 ha_new_range:
                start = read_n_cells(n_mem_addr_cells, &memcell_buf);
                size = read_n_cells(n_mem_size_cells, &memcell_buf);
-               numa_domain = of_node_numa_domain(memory);
+               nid = of_node_to_nid(memory);
 
                /* Domains not present at boot default to 0 */
-               if (!node_online(numa_domain))
-                       numa_domain = any_online_node(NODE_MASK_ALL);
+               if (!node_online(nid))
+                       nid = any_online_node(NODE_MASK_ALL);
 
                if ((scn_addr >= start) && (scn_addr < (start + size))) {
                        of_node_put(memory);
-                       goto got_numa_domain;
+                       goto got_nid;
                }
 
                if (--ranges)           /* process all ranges in cell */
@@ -806,12 +806,12 @@ ha_new_range:
        BUG();  /* section address should be found above */
 
        /* Temporary code to ensure that returned node is not empty */
-got_numa_domain:
+got_nid:
        nodes_setall(nodes);
-       while (NODE_DATA(numa_domain)->node_spanned_pages == 0) {
-               node_clear(numa_domain, nodes);
-               numa_domain = any_online_node(nodes);
+       while (NODE_DATA(nid)->node_spanned_pages == 0) {
+               node_clear(nid, nodes);
+               nid = any_online_node(nodes);
        }
-       return numa_domain;
+       return nid;
 }
 #endif /* CONFIG_MEMORY_HOTPLUG */