ARM: 7409/1: Do not call flush_cache_user_range with mmap_sem held
[linux/fpc-iii.git] / arch / x86 / mm / numa.c
blobf5510d889a226d4d38bbf529330fcc829beb8e51
1 /* Common code for 32 and 64-bit NUMA */
2 #include <linux/kernel.h>
3 #include <linux/mm.h>
4 #include <linux/string.h>
5 #include <linux/init.h>
6 #include <linux/bootmem.h>
7 #include <linux/memblock.h>
8 #include <linux/mmzone.h>
9 #include <linux/ctype.h>
10 #include <linux/module.h>
11 #include <linux/nodemask.h>
12 #include <linux/sched.h>
13 #include <linux/topology.h>
15 #include <asm/e820.h>
16 #include <asm/proto.h>
17 #include <asm/dma.h>
18 #include <asm/acpi.h>
19 #include <asm/amd_nb.h>
21 #include "numa_internal.h"
23 int __initdata numa_off;
24 nodemask_t numa_nodes_parsed __initdata;
26 struct pglist_data *node_data[MAX_NUMNODES] __read_mostly;
27 EXPORT_SYMBOL(node_data);
29 static struct numa_meminfo numa_meminfo
30 #ifndef CONFIG_MEMORY_HOTPLUG
31 __initdata
32 #endif
35 static int numa_distance_cnt;
36 static u8 *numa_distance;
38 static __init int numa_setup(char *opt)
40 if (!opt)
41 return -EINVAL;
42 if (!strncmp(opt, "off", 3))
43 numa_off = 1;
44 #ifdef CONFIG_NUMA_EMU
45 if (!strncmp(opt, "fake=", 5))
46 numa_emu_cmdline(opt + 5);
47 #endif
48 #ifdef CONFIG_ACPI_NUMA
49 if (!strncmp(opt, "noacpi", 6))
50 acpi_numa = -1;
51 #endif
52 return 0;
54 early_param("numa", numa_setup);
57 * apicid, cpu, node mappings
59 s16 __apicid_to_node[MAX_LOCAL_APIC] __cpuinitdata = {
60 [0 ... MAX_LOCAL_APIC-1] = NUMA_NO_NODE
63 int __cpuinit numa_cpu_node(int cpu)
65 int apicid = early_per_cpu(x86_cpu_to_apicid, cpu);
67 if (apicid != BAD_APICID)
68 return __apicid_to_node[apicid];
69 return NUMA_NO_NODE;
72 cpumask_var_t node_to_cpumask_map[MAX_NUMNODES];
73 EXPORT_SYMBOL(node_to_cpumask_map);
76 * Map cpu index to node index
78 DEFINE_EARLY_PER_CPU(int, x86_cpu_to_node_map, NUMA_NO_NODE);
79 EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_node_map);
81 void __cpuinit numa_set_node(int cpu, int node)
83 int *cpu_to_node_map = early_per_cpu_ptr(x86_cpu_to_node_map);
85 /* early setting, no percpu area yet */
86 if (cpu_to_node_map) {
87 cpu_to_node_map[cpu] = node;
88 return;
91 #ifdef CONFIG_DEBUG_PER_CPU_MAPS
92 if (cpu >= nr_cpu_ids || !cpu_possible(cpu)) {
93 printk(KERN_ERR "numa_set_node: invalid cpu# (%d)\n", cpu);
94 dump_stack();
95 return;
97 #endif
98 per_cpu(x86_cpu_to_node_map, cpu) = node;
100 if (node != NUMA_NO_NODE)
101 set_cpu_numa_node(cpu, node);
104 void __cpuinit numa_clear_node(int cpu)
106 numa_set_node(cpu, NUMA_NO_NODE);
110 * Allocate node_to_cpumask_map based on number of available nodes
111 * Requires node_possible_map to be valid.
113 * Note: node_to_cpumask() is not valid until after this is done.
114 * (Use CONFIG_DEBUG_PER_CPU_MAPS to check this.)
116 void __init setup_node_to_cpumask_map(void)
118 unsigned int node, num = 0;
120 /* setup nr_node_ids if not done yet */
121 if (nr_node_ids == MAX_NUMNODES) {
122 for_each_node_mask(node, node_possible_map)
123 num = node;
124 nr_node_ids = num + 1;
127 /* allocate the map */
128 for (node = 0; node < nr_node_ids; node++)
129 alloc_bootmem_cpumask_var(&node_to_cpumask_map[node]);
131 /* cpumask_of_node() will now work */
132 pr_debug("Node to cpumask map for %d nodes\n", nr_node_ids);
135 static int __init numa_add_memblk_to(int nid, u64 start, u64 end,
136 struct numa_meminfo *mi)
138 /* ignore zero length blks */
139 if (start == end)
140 return 0;
142 /* whine about and ignore invalid blks */
143 if (start > end || nid < 0 || nid >= MAX_NUMNODES) {
144 pr_warning("NUMA: Warning: invalid memblk node %d (%Lx-%Lx)\n",
145 nid, start, end);
146 return 0;
149 if (mi->nr_blks >= NR_NODE_MEMBLKS) {
150 pr_err("NUMA: too many memblk ranges\n");
151 return -EINVAL;
154 mi->blk[mi->nr_blks].start = start;
155 mi->blk[mi->nr_blks].end = end;
156 mi->blk[mi->nr_blks].nid = nid;
157 mi->nr_blks++;
158 return 0;
162 * numa_remove_memblk_from - Remove one numa_memblk from a numa_meminfo
163 * @idx: Index of memblk to remove
164 * @mi: numa_meminfo to remove memblk from
166 * Remove @idx'th numa_memblk from @mi by shifting @mi->blk[] and
167 * decrementing @mi->nr_blks.
169 void __init numa_remove_memblk_from(int idx, struct numa_meminfo *mi)
171 mi->nr_blks--;
172 memmove(&mi->blk[idx], &mi->blk[idx + 1],
173 (mi->nr_blks - idx) * sizeof(mi->blk[0]));
177 * numa_add_memblk - Add one numa_memblk to numa_meminfo
178 * @nid: NUMA node ID of the new memblk
179 * @start: Start address of the new memblk
180 * @end: End address of the new memblk
182 * Add a new memblk to the default numa_meminfo.
184 * RETURNS:
185 * 0 on success, -errno on failure.
187 int __init numa_add_memblk(int nid, u64 start, u64 end)
189 return numa_add_memblk_to(nid, start, end, &numa_meminfo);
192 /* Initialize NODE_DATA for a node on the local memory */
193 static void __init setup_node_data(int nid, u64 start, u64 end)
195 const u64 nd_low = PFN_PHYS(MAX_DMA_PFN);
196 const u64 nd_high = PFN_PHYS(max_pfn_mapped);
197 const size_t nd_size = roundup(sizeof(pg_data_t), PAGE_SIZE);
198 bool remapped = false;
199 u64 nd_pa;
200 void *nd;
201 int tnid;
204 * Don't confuse VM with a node that doesn't have the
205 * minimum amount of memory:
207 if (end && (end - start) < NODE_MIN_SIZE)
208 return;
210 /* initialize remap allocator before aligning to ZONE_ALIGN */
211 init_alloc_remap(nid, start, end);
213 start = roundup(start, ZONE_ALIGN);
215 printk(KERN_INFO "Initmem setup node %d %016Lx-%016Lx\n",
216 nid, start, end);
219 * Allocate node data. Try remap allocator first, node-local
220 * memory and then any node. Never allocate in DMA zone.
222 nd = alloc_remap(nid, nd_size);
223 if (nd) {
224 nd_pa = __pa(nd);
225 remapped = true;
226 } else {
227 nd_pa = memblock_x86_find_in_range_node(nid, nd_low, nd_high,
228 nd_size, SMP_CACHE_BYTES);
229 if (nd_pa == MEMBLOCK_ERROR)
230 nd_pa = memblock_find_in_range(nd_low, nd_high,
231 nd_size, SMP_CACHE_BYTES);
232 if (nd_pa == MEMBLOCK_ERROR) {
233 pr_err("Cannot find %zu bytes in node %d\n",
234 nd_size, nid);
235 return;
237 memblock_x86_reserve_range(nd_pa, nd_pa + nd_size, "NODE_DATA");
238 nd = __va(nd_pa);
241 /* report and initialize */
242 printk(KERN_INFO " NODE_DATA [%016Lx - %016Lx]%s\n",
243 nd_pa, nd_pa + nd_size - 1, remapped ? " (remapped)" : "");
244 tnid = early_pfn_to_nid(nd_pa >> PAGE_SHIFT);
245 if (!remapped && tnid != nid)
246 printk(KERN_INFO " NODE_DATA(%d) on node %d\n", nid, tnid);
248 node_data[nid] = nd;
249 memset(NODE_DATA(nid), 0, sizeof(pg_data_t));
250 NODE_DATA(nid)->node_id = nid;
251 NODE_DATA(nid)->node_start_pfn = start >> PAGE_SHIFT;
252 NODE_DATA(nid)->node_spanned_pages = (end - start) >> PAGE_SHIFT;
254 node_set_online(nid);
258 * numa_cleanup_meminfo - Cleanup a numa_meminfo
259 * @mi: numa_meminfo to clean up
261 * Sanitize @mi by merging and removing unncessary memblks. Also check for
262 * conflicts and clear unused memblks.
264 * RETURNS:
265 * 0 on success, -errno on failure.
267 int __init numa_cleanup_meminfo(struct numa_meminfo *mi)
269 const u64 low = 0;
270 const u64 high = PFN_PHYS(max_pfn);
271 int i, j, k;
273 /* first, trim all entries */
274 for (i = 0; i < mi->nr_blks; i++) {
275 struct numa_memblk *bi = &mi->blk[i];
277 /* make sure all blocks are inside the limits */
278 bi->start = max(bi->start, low);
279 bi->end = min(bi->end, high);
281 /* and there's no empty block */
282 if (bi->start >= bi->end)
283 numa_remove_memblk_from(i--, mi);
286 /* merge neighboring / overlapping entries */
287 for (i = 0; i < mi->nr_blks; i++) {
288 struct numa_memblk *bi = &mi->blk[i];
290 for (j = i + 1; j < mi->nr_blks; j++) {
291 struct numa_memblk *bj = &mi->blk[j];
292 u64 start, end;
295 * See whether there are overlapping blocks. Whine
296 * about but allow overlaps of the same nid. They
297 * will be merged below.
299 if (bi->end > bj->start && bi->start < bj->end) {
300 if (bi->nid != bj->nid) {
301 pr_err("NUMA: node %d (%Lx-%Lx) overlaps with node %d (%Lx-%Lx)\n",
302 bi->nid, bi->start, bi->end,
303 bj->nid, bj->start, bj->end);
304 return -EINVAL;
306 pr_warning("NUMA: Warning: node %d (%Lx-%Lx) overlaps with itself (%Lx-%Lx)\n",
307 bi->nid, bi->start, bi->end,
308 bj->start, bj->end);
312 * Join together blocks on the same node, holes
313 * between which don't overlap with memory on other
314 * nodes.
316 if (bi->nid != bj->nid)
317 continue;
318 start = min(bi->start, bj->start);
319 end = max(bi->end, bj->end);
320 for (k = 0; k < mi->nr_blks; k++) {
321 struct numa_memblk *bk = &mi->blk[k];
323 if (bi->nid == bk->nid)
324 continue;
325 if (start < bk->end && end > bk->start)
326 break;
328 if (k < mi->nr_blks)
329 continue;
330 printk(KERN_INFO "NUMA: Node %d [%Lx,%Lx) + [%Lx,%Lx) -> [%Lx,%Lx)\n",
331 bi->nid, bi->start, bi->end, bj->start, bj->end,
332 start, end);
333 bi->start = start;
334 bi->end = end;
335 numa_remove_memblk_from(j--, mi);
339 /* clear unused ones */
340 for (i = mi->nr_blks; i < ARRAY_SIZE(mi->blk); i++) {
341 mi->blk[i].start = mi->blk[i].end = 0;
342 mi->blk[i].nid = NUMA_NO_NODE;
345 return 0;
349 * Set nodes, which have memory in @mi, in *@nodemask.
351 static void __init numa_nodemask_from_meminfo(nodemask_t *nodemask,
352 const struct numa_meminfo *mi)
354 int i;
356 for (i = 0; i < ARRAY_SIZE(mi->blk); i++)
357 if (mi->blk[i].start != mi->blk[i].end &&
358 mi->blk[i].nid != NUMA_NO_NODE)
359 node_set(mi->blk[i].nid, *nodemask);
363 * numa_reset_distance - Reset NUMA distance table
365 * The current table is freed. The next numa_set_distance() call will
366 * create a new one.
368 void __init numa_reset_distance(void)
370 size_t size = numa_distance_cnt * numa_distance_cnt * sizeof(numa_distance[0]);
372 /* numa_distance could be 1LU marking allocation failure, test cnt */
373 if (numa_distance_cnt)
374 memblock_x86_free_range(__pa(numa_distance),
375 __pa(numa_distance) + size);
376 numa_distance_cnt = 0;
377 numa_distance = NULL; /* enable table creation */
380 static int __init numa_alloc_distance(void)
382 nodemask_t nodes_parsed;
383 size_t size;
384 int i, j, cnt = 0;
385 u64 phys;
387 /* size the new table and allocate it */
388 nodes_parsed = numa_nodes_parsed;
389 numa_nodemask_from_meminfo(&nodes_parsed, &numa_meminfo);
391 for_each_node_mask(i, nodes_parsed)
392 cnt = i;
393 cnt++;
394 size = cnt * cnt * sizeof(numa_distance[0]);
396 phys = memblock_find_in_range(0, PFN_PHYS(max_pfn_mapped),
397 size, PAGE_SIZE);
398 if (phys == MEMBLOCK_ERROR) {
399 pr_warning("NUMA: Warning: can't allocate distance table!\n");
400 /* don't retry until explicitly reset */
401 numa_distance = (void *)1LU;
402 return -ENOMEM;
404 memblock_x86_reserve_range(phys, phys + size, "NUMA DIST");
406 numa_distance = __va(phys);
407 numa_distance_cnt = cnt;
409 /* fill with the default distances */
410 for (i = 0; i < cnt; i++)
411 for (j = 0; j < cnt; j++)
412 numa_distance[i * cnt + j] = i == j ?
413 LOCAL_DISTANCE : REMOTE_DISTANCE;
414 printk(KERN_DEBUG "NUMA: Initialized distance table, cnt=%d\n", cnt);
416 return 0;
420 * numa_set_distance - Set NUMA distance from one NUMA to another
421 * @from: the 'from' node to set distance
422 * @to: the 'to' node to set distance
423 * @distance: NUMA distance
425 * Set the distance from node @from to @to to @distance. If distance table
426 * doesn't exist, one which is large enough to accommodate all the currently
427 * known nodes will be created.
429 * If such table cannot be allocated, a warning is printed and further
430 * calls are ignored until the distance table is reset with
431 * numa_reset_distance().
433 * If @from or @to is higher than the highest known node at the time of
434 * table creation or @distance doesn't make sense, the call is ignored.
435 * This is to allow simplification of specific NUMA config implementations.
437 void __init numa_set_distance(int from, int to, int distance)
439 if (!numa_distance && numa_alloc_distance() < 0)
440 return;
442 if (from >= numa_distance_cnt || to >= numa_distance_cnt) {
443 printk_once(KERN_DEBUG "NUMA: Debug: distance out of bound, from=%d to=%d distance=%d\n",
444 from, to, distance);
445 return;
448 if ((u8)distance != distance ||
449 (from == to && distance != LOCAL_DISTANCE)) {
450 pr_warn_once("NUMA: Warning: invalid distance parameter, from=%d to=%d distance=%d\n",
451 from, to, distance);
452 return;
455 numa_distance[from * numa_distance_cnt + to] = distance;
458 int __node_distance(int from, int to)
460 if (from >= numa_distance_cnt || to >= numa_distance_cnt)
461 return from == to ? LOCAL_DISTANCE : REMOTE_DISTANCE;
462 return numa_distance[from * numa_distance_cnt + to];
464 EXPORT_SYMBOL(__node_distance);
467 * Sanity check to catch more bad NUMA configurations (they are amazingly
468 * common). Make sure the nodes cover all memory.
470 static bool __init numa_meminfo_cover_memory(const struct numa_meminfo *mi)
472 u64 numaram, e820ram;
473 int i;
475 numaram = 0;
476 for (i = 0; i < mi->nr_blks; i++) {
477 u64 s = mi->blk[i].start >> PAGE_SHIFT;
478 u64 e = mi->blk[i].end >> PAGE_SHIFT;
479 numaram += e - s;
480 numaram -= __absent_pages_in_range(mi->blk[i].nid, s, e);
481 if ((s64)numaram < 0)
482 numaram = 0;
485 e820ram = max_pfn - (memblock_x86_hole_size(0,
486 PFN_PHYS(max_pfn)) >> PAGE_SHIFT);
487 /* We seem to lose 3 pages somewhere. Allow 1M of slack. */
488 if ((s64)(e820ram - numaram) >= (1 << (20 - PAGE_SHIFT))) {
489 printk(KERN_ERR "NUMA: nodes only cover %LuMB of your %LuMB e820 RAM. Not used.\n",
490 (numaram << PAGE_SHIFT) >> 20,
491 (e820ram << PAGE_SHIFT) >> 20);
492 return false;
494 return true;
497 static int __init numa_register_memblks(struct numa_meminfo *mi)
499 int i, nid;
501 /* Account for nodes with cpus and no memory */
502 node_possible_map = numa_nodes_parsed;
503 numa_nodemask_from_meminfo(&node_possible_map, mi);
504 if (WARN_ON(nodes_empty(node_possible_map)))
505 return -EINVAL;
507 for (i = 0; i < mi->nr_blks; i++)
508 memblock_x86_register_active_regions(mi->blk[i].nid,
509 mi->blk[i].start >> PAGE_SHIFT,
510 mi->blk[i].end >> PAGE_SHIFT);
512 /* for out of order entries */
513 sort_node_map();
514 if (!numa_meminfo_cover_memory(mi))
515 return -EINVAL;
517 /* Finally register nodes. */
518 for_each_node_mask(nid, node_possible_map) {
519 u64 start = PFN_PHYS(max_pfn);
520 u64 end = 0;
522 for (i = 0; i < mi->nr_blks; i++) {
523 if (nid != mi->blk[i].nid)
524 continue;
525 start = min(mi->blk[i].start, start);
526 end = max(mi->blk[i].end, end);
529 if (start < end)
530 setup_node_data(nid, start, end);
533 return 0;
537 * There are unfortunately some poorly designed mainboards around that
538 * only connect memory to a single CPU. This breaks the 1:1 cpu->node
539 * mapping. To avoid this fill in the mapping for all possible CPUs,
540 * as the number of CPUs is not known yet. We round robin the existing
541 * nodes.
543 static void __init numa_init_array(void)
545 int rr, i;
547 rr = first_node(node_online_map);
548 for (i = 0; i < nr_cpu_ids; i++) {
549 if (early_cpu_to_node(i) != NUMA_NO_NODE)
550 continue;
551 numa_set_node(i, rr);
552 rr = next_node(rr, node_online_map);
553 if (rr == MAX_NUMNODES)
554 rr = first_node(node_online_map);
558 static int __init numa_init(int (*init_func)(void))
560 int i;
561 int ret;
563 for (i = 0; i < MAX_LOCAL_APIC; i++)
564 set_apicid_to_node(i, NUMA_NO_NODE);
566 nodes_clear(numa_nodes_parsed);
567 nodes_clear(node_possible_map);
568 nodes_clear(node_online_map);
569 memset(&numa_meminfo, 0, sizeof(numa_meminfo));
570 remove_all_active_ranges();
571 numa_reset_distance();
573 ret = init_func();
574 if (ret < 0)
575 return ret;
576 ret = numa_cleanup_meminfo(&numa_meminfo);
577 if (ret < 0)
578 return ret;
580 numa_emulation(&numa_meminfo, numa_distance_cnt);
582 ret = numa_register_memblks(&numa_meminfo);
583 if (ret < 0)
584 return ret;
586 for (i = 0; i < nr_cpu_ids; i++) {
587 int nid = early_cpu_to_node(i);
589 if (nid == NUMA_NO_NODE)
590 continue;
591 if (!node_online(nid))
592 numa_clear_node(i);
594 numa_init_array();
595 return 0;
599 * dummy_numa_init - Fallback dummy NUMA init
601 * Used if there's no underlying NUMA architecture, NUMA initialization
602 * fails, or NUMA is disabled on the command line.
604 * Must online at least one node and add memory blocks that cover all
605 * allowed memory. This function must not fail.
607 static int __init dummy_numa_init(void)
609 printk(KERN_INFO "%s\n",
610 numa_off ? "NUMA turned off" : "No NUMA configuration found");
611 printk(KERN_INFO "Faking a node at %016Lx-%016Lx\n",
612 0LLU, PFN_PHYS(max_pfn));
614 node_set(0, numa_nodes_parsed);
615 numa_add_memblk(0, 0, PFN_PHYS(max_pfn));
617 return 0;
621 * x86_numa_init - Initialize NUMA
623 * Try each configured NUMA initialization method until one succeeds. The
624 * last fallback is dummy single node config encomapssing whole memory and
625 * never fails.
627 void __init x86_numa_init(void)
629 if (!numa_off) {
630 #ifdef CONFIG_X86_NUMAQ
631 if (!numa_init(numaq_numa_init))
632 return;
633 #endif
634 #ifdef CONFIG_ACPI_NUMA
635 if (!numa_init(x86_acpi_numa_init))
636 return;
637 #endif
638 #ifdef CONFIG_AMD_NUMA
639 if (!numa_init(amd_numa_init))
640 return;
641 #endif
644 numa_init(dummy_numa_init);
647 static __init int find_near_online_node(int node)
649 int n, val;
650 int min_val = INT_MAX;
651 int best_node = -1;
653 for_each_online_node(n) {
654 val = node_distance(node, n);
656 if (val < min_val) {
657 min_val = val;
658 best_node = n;
662 return best_node;
666 * Setup early cpu_to_node.
668 * Populate cpu_to_node[] only if x86_cpu_to_apicid[],
669 * and apicid_to_node[] tables have valid entries for a CPU.
670 * This means we skip cpu_to_node[] initialisation for NUMA
671 * emulation and faking node case (when running a kernel compiled
672 * for NUMA on a non NUMA box), which is OK as cpu_to_node[]
673 * is already initialized in a round robin manner at numa_init_array,
674 * prior to this call, and this initialization is good enough
675 * for the fake NUMA cases.
677 * Called before the per_cpu areas are setup.
679 void __init init_cpu_to_node(void)
681 int cpu;
682 u16 *cpu_to_apicid = early_per_cpu_ptr(x86_cpu_to_apicid);
684 BUG_ON(cpu_to_apicid == NULL);
686 for_each_possible_cpu(cpu) {
687 int node = numa_cpu_node(cpu);
689 if (node == NUMA_NO_NODE)
690 continue;
691 if (!node_online(node))
692 node = find_near_online_node(node);
693 numa_set_node(cpu, node);
697 #ifndef CONFIG_DEBUG_PER_CPU_MAPS
699 # ifndef CONFIG_NUMA_EMU
700 void __cpuinit numa_add_cpu(int cpu)
702 cpumask_set_cpu(cpu, node_to_cpumask_map[early_cpu_to_node(cpu)]);
705 void __cpuinit numa_remove_cpu(int cpu)
707 cpumask_clear_cpu(cpu, node_to_cpumask_map[early_cpu_to_node(cpu)]);
709 # endif /* !CONFIG_NUMA_EMU */
711 #else /* !CONFIG_DEBUG_PER_CPU_MAPS */
713 int __cpu_to_node(int cpu)
715 if (early_per_cpu_ptr(x86_cpu_to_node_map)) {
716 printk(KERN_WARNING
717 "cpu_to_node(%d): usage too early!\n", cpu);
718 dump_stack();
719 return early_per_cpu_ptr(x86_cpu_to_node_map)[cpu];
721 return per_cpu(x86_cpu_to_node_map, cpu);
723 EXPORT_SYMBOL(__cpu_to_node);
726 * Same function as cpu_to_node() but used if called before the
727 * per_cpu areas are setup.
729 int early_cpu_to_node(int cpu)
731 if (early_per_cpu_ptr(x86_cpu_to_node_map))
732 return early_per_cpu_ptr(x86_cpu_to_node_map)[cpu];
734 if (!cpu_possible(cpu)) {
735 printk(KERN_WARNING
736 "early_cpu_to_node(%d): no per_cpu area!\n", cpu);
737 dump_stack();
738 return NUMA_NO_NODE;
740 return per_cpu(x86_cpu_to_node_map, cpu);
743 void debug_cpumask_set_cpu(int cpu, int node, bool enable)
745 struct cpumask *mask;
746 char buf[64];
748 if (node == NUMA_NO_NODE) {
749 /* early_cpu_to_node() already emits a warning and trace */
750 return;
752 mask = node_to_cpumask_map[node];
753 if (!mask) {
754 pr_err("node_to_cpumask_map[%i] NULL\n", node);
755 dump_stack();
756 return;
759 if (enable)
760 cpumask_set_cpu(cpu, mask);
761 else
762 cpumask_clear_cpu(cpu, mask);
764 cpulist_scnprintf(buf, sizeof(buf), mask);
765 printk(KERN_DEBUG "%s cpu %d node %d: mask now %s\n",
766 enable ? "numa_add_cpu" : "numa_remove_cpu",
767 cpu, node, buf);
768 return;
771 # ifndef CONFIG_NUMA_EMU
772 static void __cpuinit numa_set_cpumask(int cpu, bool enable)
774 debug_cpumask_set_cpu(cpu, early_cpu_to_node(cpu), enable);
777 void __cpuinit numa_add_cpu(int cpu)
779 numa_set_cpumask(cpu, true);
782 void __cpuinit numa_remove_cpu(int cpu)
784 numa_set_cpumask(cpu, false);
786 # endif /* !CONFIG_NUMA_EMU */
789 * Returns a pointer to the bitmask of CPUs on Node 'node'.
791 const struct cpumask *cpumask_of_node(int node)
793 if (node >= nr_node_ids) {
794 printk(KERN_WARNING
795 "cpumask_of_node(%d): node > nr_node_ids(%d)\n",
796 node, nr_node_ids);
797 dump_stack();
798 return cpu_none_mask;
800 if (node_to_cpumask_map[node] == NULL) {
801 printk(KERN_WARNING
802 "cpumask_of_node(%d): no node_to_cpumask_map!\n",
803 node);
804 dump_stack();
805 return cpu_online_mask;
807 return node_to_cpumask_map[node];
809 EXPORT_SYMBOL(cpumask_of_node);
811 #endif /* !CONFIG_DEBUG_PER_CPU_MAPS */
813 #ifdef CONFIG_MEMORY_HOTPLUG
814 int memory_add_physaddr_to_nid(u64 start)
816 struct numa_meminfo *mi = &numa_meminfo;
817 int nid = mi->blk[0].nid;
818 int i;
820 for (i = 0; i < mi->nr_blks; i++)
821 if (mi->blk[i].start <= start && mi->blk[i].end > start)
822 nid = mi->blk[i].nid;
823 return nid;
825 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
826 #endif