]>
Commit | Line | Data |
---|---|---|
71ee73e7 | 1 | /* Common code for 32 and 64-bit NUMA */ |
a4106eae TH |
2 | #include <linux/kernel.h> |
3 | #include <linux/mm.h> | |
4 | #include <linux/string.h> | |
5 | #include <linux/init.h> | |
71ee73e7 | 6 | #include <linux/bootmem.h> |
a4106eae TH |
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> | |
14 | ||
15 | #include <asm/e820.h> | |
16 | #include <asm/proto.h> | |
17 | #include <asm/dma.h> | |
90321602 | 18 | #include <asm/acpi.h> |
a4106eae TH |
19 | #include <asm/amd_nb.h> |
20 | ||
21 | #include "numa_internal.h" | |
90321602 JB |
22 | |
23 | int __initdata numa_off; | |
e6df595b | 24 | nodemask_t numa_nodes_parsed __initdata; |
90321602 | 25 | |
a4106eae TH |
26 | struct pglist_data *node_data[MAX_NUMNODES] __read_mostly; |
27 | EXPORT_SYMBOL(node_data); | |
28 | ||
29 | static struct numa_meminfo numa_meminfo | |
30 | #ifndef CONFIG_MEMORY_HOTPLUG | |
31 | __initdata | |
32 | #endif | |
33 | ; | |
34 | ||
35 | static int numa_distance_cnt; | |
36 | static u8 *numa_distance; | |
a4106eae | 37 | |
90321602 JB |
38 | static __init int numa_setup(char *opt) |
39 | { | |
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; | |
53 | } | |
54 | early_param("numa", numa_setup); | |
71ee73e7 | 55 | |
71ee73e7 | 56 | /* |
bbc9e2f4 | 57 | * apicid, cpu, node mappings |
71ee73e7 | 58 | */ |
c4c60524 | 59 | s16 __apicid_to_node[MAX_LOCAL_APIC] = { |
bbc9e2f4 TH |
60 | [0 ... MAX_LOCAL_APIC-1] = NUMA_NO_NODE |
61 | }; | |
62 | ||
6bd26273 TH |
63 | int __cpuinit numa_cpu_node(int cpu) |
64 | { | |
65 | int apicid = early_per_cpu(x86_cpu_to_apicid, cpu); | |
66 | ||
67 | if (apicid != BAD_APICID) | |
68 | return __apicid_to_node[apicid]; | |
69 | return NUMA_NO_NODE; | |
70 | } | |
71 | ||
c032ef60 | 72 | cpumask_var_t node_to_cpumask_map[MAX_NUMNODES]; |
71ee73e7 RR |
73 | EXPORT_SYMBOL(node_to_cpumask_map); |
74 | ||
645a7919 TH |
75 | /* |
76 | * Map cpu index to node index | |
77 | */ | |
645a7919 | 78 | DEFINE_EARLY_PER_CPU(int, x86_cpu_to_node_map, NUMA_NO_NODE); |
645a7919 TH |
79 | EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_node_map); |
80 | ||
e13fe869 | 81 | void numa_set_node(int cpu, int node) |
645a7919 TH |
82 | { |
83 | int *cpu_to_node_map = early_per_cpu_ptr(x86_cpu_to_node_map); | |
84 | ||
85 | /* early setting, no percpu area yet */ | |
86 | if (cpu_to_node_map) { | |
87 | cpu_to_node_map[cpu] = node; | |
88 | return; | |
89 | } | |
90 | ||
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; | |
96 | } | |
97 | #endif | |
98 | per_cpu(x86_cpu_to_node_map, cpu) = node; | |
99 | ||
942670d0 | 100 | set_cpu_numa_node(cpu, node); |
645a7919 TH |
101 | } |
102 | ||
e13fe869 | 103 | void numa_clear_node(int cpu) |
645a7919 TH |
104 | { |
105 | numa_set_node(cpu, NUMA_NO_NODE); | |
106 | } | |
107 | ||
71ee73e7 RR |
108 | /* |
109 | * Allocate node_to_cpumask_map based on number of available nodes | |
110 | * Requires node_possible_map to be valid. | |
111 | * | |
9512938b | 112 | * Note: cpumask_of_node() is not valid until after this is done. |
71ee73e7 RR |
113 | * (Use CONFIG_DEBUG_PER_CPU_MAPS to check this.) |
114 | */ | |
115 | void __init setup_node_to_cpumask_map(void) | |
116 | { | |
117 | unsigned int node, num = 0; | |
71ee73e7 RR |
118 | |
119 | /* setup nr_node_ids if not done yet */ | |
120 | if (nr_node_ids == MAX_NUMNODES) { | |
121 | for_each_node_mask(node, node_possible_map) | |
122 | num = node; | |
123 | nr_node_ids = num + 1; | |
124 | } | |
125 | ||
126 | /* allocate the map */ | |
c032ef60 RR |
127 | for (node = 0; node < nr_node_ids; node++) |
128 | alloc_bootmem_cpumask_var(&node_to_cpumask_map[node]); | |
71ee73e7 | 129 | |
c032ef60 RR |
130 | /* cpumask_of_node() will now work */ |
131 | pr_debug("Node to cpumask map for %d nodes\n", nr_node_ids); | |
71ee73e7 RR |
132 | } |
133 | ||
a4106eae TH |
134 | static int __init numa_add_memblk_to(int nid, u64 start, u64 end, |
135 | struct numa_meminfo *mi) | |
136 | { | |
137 | /* ignore zero length blks */ | |
138 | if (start == end) | |
139 | return 0; | |
140 | ||
141 | /* whine about and ignore invalid blks */ | |
142 | if (start > end || nid < 0 || nid >= MAX_NUMNODES) { | |
365811d6 BH |
143 | pr_warning("NUMA: Warning: invalid memblk node %d [mem %#010Lx-%#010Lx]\n", |
144 | nid, start, end - 1); | |
a4106eae TH |
145 | return 0; |
146 | } | |
147 | ||
148 | if (mi->nr_blks >= NR_NODE_MEMBLKS) { | |
149 | pr_err("NUMA: too many memblk ranges\n"); | |
150 | return -EINVAL; | |
151 | } | |
152 | ||
153 | mi->blk[mi->nr_blks].start = start; | |
154 | mi->blk[mi->nr_blks].end = end; | |
155 | mi->blk[mi->nr_blks].nid = nid; | |
156 | mi->nr_blks++; | |
157 | return 0; | |
158 | } | |
159 | ||
160 | /** | |
161 | * numa_remove_memblk_from - Remove one numa_memblk from a numa_meminfo | |
162 | * @idx: Index of memblk to remove | |
163 | * @mi: numa_meminfo to remove memblk from | |
164 | * | |
165 | * Remove @idx'th numa_memblk from @mi by shifting @mi->blk[] and | |
166 | * decrementing @mi->nr_blks. | |
167 | */ | |
168 | void __init numa_remove_memblk_from(int idx, struct numa_meminfo *mi) | |
169 | { | |
170 | mi->nr_blks--; | |
171 | memmove(&mi->blk[idx], &mi->blk[idx + 1], | |
172 | (mi->nr_blks - idx) * sizeof(mi->blk[0])); | |
173 | } | |
174 | ||
175 | /** | |
176 | * numa_add_memblk - Add one numa_memblk to numa_meminfo | |
177 | * @nid: NUMA node ID of the new memblk | |
178 | * @start: Start address of the new memblk | |
179 | * @end: End address of the new memblk | |
180 | * | |
181 | * Add a new memblk to the default numa_meminfo. | |
182 | * | |
183 | * RETURNS: | |
184 | * 0 on success, -errno on failure. | |
185 | */ | |
186 | int __init numa_add_memblk(int nid, u64 start, u64 end) | |
187 | { | |
188 | return numa_add_memblk_to(nid, start, end, &numa_meminfo); | |
189 | } | |
190 | ||
a56bca80 YL |
191 | /* Initialize NODE_DATA for a node on the local memory */ |
192 | static void __init setup_node_data(int nid, u64 start, u64 end) | |
a4106eae | 193 | { |
a4106eae | 194 | const size_t nd_size = roundup(sizeof(pg_data_t), PAGE_SIZE); |
38f3e1ca | 195 | u64 nd_pa; |
7888e96b | 196 | void *nd; |
a4106eae TH |
197 | int tnid; |
198 | ||
199 | /* | |
200 | * Don't confuse VM with a node that doesn't have the | |
201 | * minimum amount of memory: | |
202 | */ | |
203 | if (end && (end - start) < NODE_MIN_SIZE) | |
204 | return; | |
205 | ||
206 | start = roundup(start, ZONE_ALIGN); | |
207 | ||
365811d6 BH |
208 | printk(KERN_INFO "Initmem setup node %d [mem %#010Lx-%#010Lx]\n", |
209 | nid, start, end - 1); | |
a4106eae TH |
210 | |
211 | /* | |
07f4207a PA |
212 | * Allocate node data. Try node-local memory and then any node. |
213 | * Never allocate in DMA zone. | |
a4106eae | 214 | */ |
20e6926d | 215 | nd_pa = memblock_alloc_nid(nd_size, SMP_CACHE_BYTES, nid); |
07f4207a | 216 | if (!nd_pa) { |
20e6926d YL |
217 | pr_err("Cannot find %zu bytes in node %d\n", |
218 | nd_size, nid); | |
07f4207a | 219 | return; |
a4106eae | 220 | } |
07f4207a | 221 | nd = __va(nd_pa); |
a4106eae TH |
222 | |
223 | /* report and initialize */ | |
07f4207a PA |
224 | printk(KERN_INFO " NODE_DATA [mem %#010Lx-%#010Lx]\n", |
225 | nd_pa, nd_pa + nd_size - 1); | |
a4106eae | 226 | tnid = early_pfn_to_nid(nd_pa >> PAGE_SHIFT); |
07f4207a | 227 | if (tnid != nid) |
a4106eae TH |
228 | printk(KERN_INFO " NODE_DATA(%d) on node %d\n", nid, tnid); |
229 | ||
7888e96b | 230 | node_data[nid] = nd; |
a4106eae TH |
231 | memset(NODE_DATA(nid), 0, sizeof(pg_data_t)); |
232 | NODE_DATA(nid)->node_id = nid; | |
233 | NODE_DATA(nid)->node_start_pfn = start >> PAGE_SHIFT; | |
234 | NODE_DATA(nid)->node_spanned_pages = (end - start) >> PAGE_SHIFT; | |
235 | ||
236 | node_set_online(nid); | |
237 | } | |
238 | ||
239 | /** | |
240 | * numa_cleanup_meminfo - Cleanup a numa_meminfo | |
241 | * @mi: numa_meminfo to clean up | |
242 | * | |
243 | * Sanitize @mi by merging and removing unncessary memblks. Also check for | |
244 | * conflicts and clear unused memblks. | |
245 | * | |
246 | * RETURNS: | |
247 | * 0 on success, -errno on failure. | |
248 | */ | |
249 | int __init numa_cleanup_meminfo(struct numa_meminfo *mi) | |
250 | { | |
251 | const u64 low = 0; | |
38f3e1ca | 252 | const u64 high = PFN_PHYS(max_pfn); |
a4106eae TH |
253 | int i, j, k; |
254 | ||
e5a10c1b | 255 | /* first, trim all entries */ |
a4106eae TH |
256 | for (i = 0; i < mi->nr_blks; i++) { |
257 | struct numa_memblk *bi = &mi->blk[i]; | |
258 | ||
259 | /* make sure all blocks are inside the limits */ | |
260 | bi->start = max(bi->start, low); | |
261 | bi->end = min(bi->end, high); | |
262 | ||
263 | /* and there's no empty block */ | |
e5a10c1b | 264 | if (bi->start >= bi->end) |
a4106eae | 265 | numa_remove_memblk_from(i--, mi); |
e5a10c1b YL |
266 | } |
267 | ||
268 | /* merge neighboring / overlapping entries */ | |
269 | for (i = 0; i < mi->nr_blks; i++) { | |
270 | struct numa_memblk *bi = &mi->blk[i]; | |
a4106eae TH |
271 | |
272 | for (j = i + 1; j < mi->nr_blks; j++) { | |
273 | struct numa_memblk *bj = &mi->blk[j]; | |
38f3e1ca | 274 | u64 start, end; |
a4106eae TH |
275 | |
276 | /* | |
277 | * See whether there are overlapping blocks. Whine | |
278 | * about but allow overlaps of the same nid. They | |
279 | * will be merged below. | |
280 | */ | |
281 | if (bi->end > bj->start && bi->start < bj->end) { | |
282 | if (bi->nid != bj->nid) { | |
365811d6 BH |
283 | pr_err("NUMA: node %d [mem %#010Lx-%#010Lx] overlaps with node %d [mem %#010Lx-%#010Lx]\n", |
284 | bi->nid, bi->start, bi->end - 1, | |
285 | bj->nid, bj->start, bj->end - 1); | |
a4106eae TH |
286 | return -EINVAL; |
287 | } | |
365811d6 BH |
288 | pr_warning("NUMA: Warning: node %d [mem %#010Lx-%#010Lx] overlaps with itself [mem %#010Lx-%#010Lx]\n", |
289 | bi->nid, bi->start, bi->end - 1, | |
290 | bj->start, bj->end - 1); | |
a4106eae TH |
291 | } |
292 | ||
293 | /* | |
294 | * Join together blocks on the same node, holes | |
295 | * between which don't overlap with memory on other | |
296 | * nodes. | |
297 | */ | |
298 | if (bi->nid != bj->nid) | |
299 | continue; | |
e5a10c1b YL |
300 | start = min(bi->start, bj->start); |
301 | end = max(bi->end, bj->end); | |
a4106eae TH |
302 | for (k = 0; k < mi->nr_blks; k++) { |
303 | struct numa_memblk *bk = &mi->blk[k]; | |
304 | ||
305 | if (bi->nid == bk->nid) | |
306 | continue; | |
307 | if (start < bk->end && end > bk->start) | |
308 | break; | |
309 | } | |
310 | if (k < mi->nr_blks) | |
311 | continue; | |
365811d6 BH |
312 | printk(KERN_INFO "NUMA: Node %d [mem %#010Lx-%#010Lx] + [mem %#010Lx-%#010Lx] -> [mem %#010Lx-%#010Lx]\n", |
313 | bi->nid, bi->start, bi->end - 1, bj->start, | |
314 | bj->end - 1, start, end - 1); | |
a4106eae TH |
315 | bi->start = start; |
316 | bi->end = end; | |
317 | numa_remove_memblk_from(j--, mi); | |
318 | } | |
319 | } | |
320 | ||
e5a10c1b | 321 | /* clear unused ones */ |
a4106eae TH |
322 | for (i = mi->nr_blks; i < ARRAY_SIZE(mi->blk); i++) { |
323 | mi->blk[i].start = mi->blk[i].end = 0; | |
324 | mi->blk[i].nid = NUMA_NO_NODE; | |
325 | } | |
326 | ||
327 | return 0; | |
328 | } | |
329 | ||
330 | /* | |
331 | * Set nodes, which have memory in @mi, in *@nodemask. | |
332 | */ | |
333 | static void __init numa_nodemask_from_meminfo(nodemask_t *nodemask, | |
334 | const struct numa_meminfo *mi) | |
335 | { | |
336 | int i; | |
337 | ||
338 | for (i = 0; i < ARRAY_SIZE(mi->blk); i++) | |
339 | if (mi->blk[i].start != mi->blk[i].end && | |
340 | mi->blk[i].nid != NUMA_NO_NODE) | |
341 | node_set(mi->blk[i].nid, *nodemask); | |
342 | } | |
343 | ||
344 | /** | |
345 | * numa_reset_distance - Reset NUMA distance table | |
346 | * | |
347 | * The current table is freed. The next numa_set_distance() call will | |
348 | * create a new one. | |
349 | */ | |
350 | void __init numa_reset_distance(void) | |
351 | { | |
352 | size_t size = numa_distance_cnt * numa_distance_cnt * sizeof(numa_distance[0]); | |
353 | ||
354 | /* numa_distance could be 1LU marking allocation failure, test cnt */ | |
355 | if (numa_distance_cnt) | |
24aa0788 | 356 | memblock_free(__pa(numa_distance), size); |
a4106eae TH |
357 | numa_distance_cnt = 0; |
358 | numa_distance = NULL; /* enable table creation */ | |
359 | } | |
360 | ||
361 | static int __init numa_alloc_distance(void) | |
362 | { | |
363 | nodemask_t nodes_parsed; | |
364 | size_t size; | |
365 | int i, j, cnt = 0; | |
366 | u64 phys; | |
367 | ||
368 | /* size the new table and allocate it */ | |
369 | nodes_parsed = numa_nodes_parsed; | |
370 | numa_nodemask_from_meminfo(&nodes_parsed, &numa_meminfo); | |
371 | ||
372 | for_each_node_mask(i, nodes_parsed) | |
373 | cnt = i; | |
374 | cnt++; | |
375 | size = cnt * cnt * sizeof(numa_distance[0]); | |
376 | ||
38f3e1ca | 377 | phys = memblock_find_in_range(0, PFN_PHYS(max_pfn_mapped), |
a4106eae | 378 | size, PAGE_SIZE); |
1f5026a7 | 379 | if (!phys) { |
a4106eae TH |
380 | pr_warning("NUMA: Warning: can't allocate distance table!\n"); |
381 | /* don't retry until explicitly reset */ | |
382 | numa_distance = (void *)1LU; | |
383 | return -ENOMEM; | |
384 | } | |
24aa0788 | 385 | memblock_reserve(phys, size); |
a4106eae TH |
386 | |
387 | numa_distance = __va(phys); | |
388 | numa_distance_cnt = cnt; | |
389 | ||
390 | /* fill with the default distances */ | |
391 | for (i = 0; i < cnt; i++) | |
392 | for (j = 0; j < cnt; j++) | |
393 | numa_distance[i * cnt + j] = i == j ? | |
394 | LOCAL_DISTANCE : REMOTE_DISTANCE; | |
395 | printk(KERN_DEBUG "NUMA: Initialized distance table, cnt=%d\n", cnt); | |
396 | ||
397 | return 0; | |
398 | } | |
399 | ||
400 | /** | |
401 | * numa_set_distance - Set NUMA distance from one NUMA to another | |
402 | * @from: the 'from' node to set distance | |
403 | * @to: the 'to' node to set distance | |
404 | * @distance: NUMA distance | |
405 | * | |
406 | * Set the distance from node @from to @to to @distance. If distance table | |
407 | * doesn't exist, one which is large enough to accommodate all the currently | |
408 | * known nodes will be created. | |
409 | * | |
410 | * If such table cannot be allocated, a warning is printed and further | |
411 | * calls are ignored until the distance table is reset with | |
412 | * numa_reset_distance(). | |
413 | * | |
54eed6cb PH |
414 | * If @from or @to is higher than the highest known node or lower than zero |
415 | * at the time of table creation or @distance doesn't make sense, the call | |
416 | * is ignored. | |
a4106eae TH |
417 | * This is to allow simplification of specific NUMA config implementations. |
418 | */ | |
419 | void __init numa_set_distance(int from, int to, int distance) | |
420 | { | |
421 | if (!numa_distance && numa_alloc_distance() < 0) | |
422 | return; | |
423 | ||
54eed6cb PH |
424 | if (from >= numa_distance_cnt || to >= numa_distance_cnt || |
425 | from < 0 || to < 0) { | |
426 | pr_warn_once("NUMA: Warning: node ids are out of bound, from=%d to=%d distance=%d\n", | |
a4106eae TH |
427 | from, to, distance); |
428 | return; | |
429 | } | |
430 | ||
431 | if ((u8)distance != distance || | |
432 | (from == to && distance != LOCAL_DISTANCE)) { | |
433 | pr_warn_once("NUMA: Warning: invalid distance parameter, from=%d to=%d distance=%d\n", | |
434 | from, to, distance); | |
435 | return; | |
436 | } | |
437 | ||
438 | numa_distance[from * numa_distance_cnt + to] = distance; | |
439 | } | |
440 | ||
441 | int __node_distance(int from, int to) | |
442 | { | |
443 | if (from >= numa_distance_cnt || to >= numa_distance_cnt) | |
444 | return from == to ? LOCAL_DISTANCE : REMOTE_DISTANCE; | |
445 | return numa_distance[from * numa_distance_cnt + to]; | |
446 | } | |
447 | EXPORT_SYMBOL(__node_distance); | |
448 | ||
449 | /* | |
450 | * Sanity check to catch more bad NUMA configurations (they are amazingly | |
451 | * common). Make sure the nodes cover all memory. | |
452 | */ | |
453 | static bool __init numa_meminfo_cover_memory(const struct numa_meminfo *mi) | |
454 | { | |
38f3e1ca | 455 | u64 numaram, e820ram; |
a4106eae TH |
456 | int i; |
457 | ||
458 | numaram = 0; | |
459 | for (i = 0; i < mi->nr_blks; i++) { | |
38f3e1ca TH |
460 | u64 s = mi->blk[i].start >> PAGE_SHIFT; |
461 | u64 e = mi->blk[i].end >> PAGE_SHIFT; | |
a4106eae TH |
462 | numaram += e - s; |
463 | numaram -= __absent_pages_in_range(mi->blk[i].nid, s, e); | |
38f3e1ca | 464 | if ((s64)numaram < 0) |
a4106eae TH |
465 | numaram = 0; |
466 | } | |
467 | ||
474b881b TH |
468 | e820ram = max_pfn - absent_pages_in_range(0, max_pfn); |
469 | ||
a4106eae | 470 | /* We seem to lose 3 pages somewhere. Allow 1M of slack. */ |
38f3e1ca TH |
471 | if ((s64)(e820ram - numaram) >= (1 << (20 - PAGE_SHIFT))) { |
472 | printk(KERN_ERR "NUMA: nodes only cover %LuMB of your %LuMB e820 RAM. Not used.\n", | |
a4106eae TH |
473 | (numaram << PAGE_SHIFT) >> 20, |
474 | (e820ram << PAGE_SHIFT) >> 20); | |
475 | return false; | |
476 | } | |
477 | return true; | |
478 | } | |
479 | ||
480 | static int __init numa_register_memblks(struct numa_meminfo *mi) | |
481 | { | |
1e01979c | 482 | unsigned long uninitialized_var(pfn_align); |
a4106eae TH |
483 | int i, nid; |
484 | ||
485 | /* Account for nodes with cpus and no memory */ | |
486 | node_possible_map = numa_nodes_parsed; | |
487 | numa_nodemask_from_meminfo(&node_possible_map, mi); | |
488 | if (WARN_ON(nodes_empty(node_possible_map))) | |
489 | return -EINVAL; | |
490 | ||
0608f70c TH |
491 | for (i = 0; i < mi->nr_blks; i++) { |
492 | struct numa_memblk *mb = &mi->blk[i]; | |
493 | memblock_set_node(mb->start, mb->end - mb->start, mb->nid); | |
494 | } | |
1e01979c TH |
495 | |
496 | /* | |
497 | * If sections array is gonna be used for pfn -> nid mapping, check | |
498 | * whether its granularity is fine enough. | |
499 | */ | |
500 | #ifdef NODE_NOT_IN_PAGE_FLAGS | |
501 | pfn_align = node_map_pfn_alignment(); | |
502 | if (pfn_align && pfn_align < PAGES_PER_SECTION) { | |
503 | printk(KERN_WARNING "Node alignment %LuMB < min %LuMB, rejecting NUMA config\n", | |
504 | PFN_PHYS(pfn_align) >> 20, | |
505 | PFN_PHYS(PAGES_PER_SECTION) >> 20); | |
506 | return -EINVAL; | |
507 | } | |
508 | #endif | |
a4106eae TH |
509 | if (!numa_meminfo_cover_memory(mi)) |
510 | return -EINVAL; | |
511 | ||
512 | /* Finally register nodes. */ | |
513 | for_each_node_mask(nid, node_possible_map) { | |
38f3e1ca | 514 | u64 start = PFN_PHYS(max_pfn); |
a4106eae TH |
515 | u64 end = 0; |
516 | ||
517 | for (i = 0; i < mi->nr_blks; i++) { | |
518 | if (nid != mi->blk[i].nid) | |
519 | continue; | |
520 | start = min(mi->blk[i].start, start); | |
521 | end = max(mi->blk[i].end, end); | |
522 | } | |
523 | ||
524 | if (start < end) | |
a56bca80 | 525 | setup_node_data(nid, start, end); |
a4106eae TH |
526 | } |
527 | ||
0608f70c TH |
528 | /* Dump memblock with node info and return. */ |
529 | memblock_dump_all(); | |
a4106eae TH |
530 | return 0; |
531 | } | |
a4106eae | 532 | |
8db78cc4 TH |
533 | /* |
534 | * There are unfortunately some poorly designed mainboards around that | |
535 | * only connect memory to a single CPU. This breaks the 1:1 cpu->node | |
536 | * mapping. To avoid this fill in the mapping for all possible CPUs, | |
537 | * as the number of CPUs is not known yet. We round robin the existing | |
538 | * nodes. | |
539 | */ | |
752d4f37 | 540 | static void __init numa_init_array(void) |
8db78cc4 TH |
541 | { |
542 | int rr, i; | |
543 | ||
544 | rr = first_node(node_online_map); | |
545 | for (i = 0; i < nr_cpu_ids; i++) { | |
546 | if (early_cpu_to_node(i) != NUMA_NO_NODE) | |
547 | continue; | |
548 | numa_set_node(i, rr); | |
549 | rr = next_node(rr, node_online_map); | |
550 | if (rr == MAX_NUMNODES) | |
551 | rr = first_node(node_online_map); | |
552 | } | |
553 | } | |
554 | ||
a4106eae TH |
555 | static int __init numa_init(int (*init_func)(void)) |
556 | { | |
557 | int i; | |
558 | int ret; | |
559 | ||
560 | for (i = 0; i < MAX_LOCAL_APIC; i++) | |
561 | set_apicid_to_node(i, NUMA_NO_NODE); | |
562 | ||
20e6926d | 563 | nodes_clear(numa_nodes_parsed); |
a4106eae TH |
564 | nodes_clear(node_possible_map); |
565 | nodes_clear(node_online_map); | |
20e6926d | 566 | memset(&numa_meminfo, 0, sizeof(numa_meminfo)); |
0608f70c | 567 | WARN_ON(memblock_set_node(0, ULLONG_MAX, MAX_NUMNODES)); |
a4106eae TH |
568 | numa_reset_distance(); |
569 | ||
570 | ret = init_func(); | |
571 | if (ret < 0) | |
572 | return ret; | |
573 | ret = numa_cleanup_meminfo(&numa_meminfo); | |
574 | if (ret < 0) | |
575 | return ret; | |
576 | ||
577 | numa_emulation(&numa_meminfo, numa_distance_cnt); | |
578 | ||
579 | ret = numa_register_memblks(&numa_meminfo); | |
580 | if (ret < 0) | |
581 | return ret; | |
582 | ||
583 | for (i = 0; i < nr_cpu_ids; i++) { | |
584 | int nid = early_cpu_to_node(i); | |
585 | ||
586 | if (nid == NUMA_NO_NODE) | |
587 | continue; | |
588 | if (!node_online(nid)) | |
589 | numa_clear_node(i); | |
590 | } | |
591 | numa_init_array(); | |
592 | return 0; | |
593 | } | |
594 | ||
595 | /** | |
596 | * dummy_numa_init - Fallback dummy NUMA init | |
597 | * | |
598 | * Used if there's no underlying NUMA architecture, NUMA initialization | |
599 | * fails, or NUMA is disabled on the command line. | |
600 | * | |
601 | * Must online at least one node and add memory blocks that cover all | |
602 | * allowed memory. This function must not fail. | |
603 | */ | |
604 | static int __init dummy_numa_init(void) | |
605 | { | |
606 | printk(KERN_INFO "%s\n", | |
607 | numa_off ? "NUMA turned off" : "No NUMA configuration found"); | |
365811d6 BH |
608 | printk(KERN_INFO "Faking a node at [mem %#018Lx-%#018Lx]\n", |
609 | 0LLU, PFN_PHYS(max_pfn) - 1); | |
a4106eae TH |
610 | |
611 | node_set(0, numa_nodes_parsed); | |
38f3e1ca | 612 | numa_add_memblk(0, 0, PFN_PHYS(max_pfn)); |
a4106eae TH |
613 | |
614 | return 0; | |
615 | } | |
616 | ||
617 | /** | |
618 | * x86_numa_init - Initialize NUMA | |
619 | * | |
620 | * Try each configured NUMA initialization method until one succeeds. The | |
621 | * last fallback is dummy single node config encomapssing whole memory and | |
622 | * never fails. | |
623 | */ | |
624 | void __init x86_numa_init(void) | |
625 | { | |
626 | if (!numa_off) { | |
bd6709a9 TH |
627 | #ifdef CONFIG_X86_NUMAQ |
628 | if (!numa_init(numaq_numa_init)) | |
629 | return; | |
630 | #endif | |
a4106eae TH |
631 | #ifdef CONFIG_ACPI_NUMA |
632 | if (!numa_init(x86_acpi_numa_init)) | |
633 | return; | |
634 | #endif | |
635 | #ifdef CONFIG_AMD_NUMA | |
636 | if (!numa_init(amd_numa_init)) | |
637 | return; | |
638 | #endif | |
639 | } | |
640 | ||
641 | numa_init(dummy_numa_init); | |
642 | } | |
a4106eae | 643 | |
8db78cc4 TH |
644 | static __init int find_near_online_node(int node) |
645 | { | |
646 | int n, val; | |
647 | int min_val = INT_MAX; | |
648 | int best_node = -1; | |
649 | ||
650 | for_each_online_node(n) { | |
651 | val = node_distance(node, n); | |
652 | ||
653 | if (val < min_val) { | |
654 | min_val = val; | |
655 | best_node = n; | |
656 | } | |
657 | } | |
658 | ||
659 | return best_node; | |
660 | } | |
661 | ||
662 | /* | |
663 | * Setup early cpu_to_node. | |
664 | * | |
665 | * Populate cpu_to_node[] only if x86_cpu_to_apicid[], | |
666 | * and apicid_to_node[] tables have valid entries for a CPU. | |
667 | * This means we skip cpu_to_node[] initialisation for NUMA | |
668 | * emulation and faking node case (when running a kernel compiled | |
669 | * for NUMA on a non NUMA box), which is OK as cpu_to_node[] | |
670 | * is already initialized in a round robin manner at numa_init_array, | |
671 | * prior to this call, and this initialization is good enough | |
672 | * for the fake NUMA cases. | |
673 | * | |
674 | * Called before the per_cpu areas are setup. | |
675 | */ | |
676 | void __init init_cpu_to_node(void) | |
677 | { | |
678 | int cpu; | |
679 | u16 *cpu_to_apicid = early_per_cpu_ptr(x86_cpu_to_apicid); | |
680 | ||
681 | BUG_ON(cpu_to_apicid == NULL); | |
682 | ||
683 | for_each_possible_cpu(cpu) { | |
684 | int node = numa_cpu_node(cpu); | |
685 | ||
686 | if (node == NUMA_NO_NODE) | |
687 | continue; | |
688 | if (!node_online(node)) | |
689 | node = find_near_online_node(node); | |
690 | numa_set_node(cpu, node); | |
691 | } | |
692 | } | |
693 | ||
de2d9445 TH |
694 | #ifndef CONFIG_DEBUG_PER_CPU_MAPS |
695 | ||
696 | # ifndef CONFIG_NUMA_EMU | |
697 | void __cpuinit numa_add_cpu(int cpu) | |
698 | { | |
699 | cpumask_set_cpu(cpu, node_to_cpumask_map[early_cpu_to_node(cpu)]); | |
700 | } | |
701 | ||
702 | void __cpuinit numa_remove_cpu(int cpu) | |
703 | { | |
704 | cpumask_clear_cpu(cpu, node_to_cpumask_map[early_cpu_to_node(cpu)]); | |
705 | } | |
706 | # endif /* !CONFIG_NUMA_EMU */ | |
707 | ||
708 | #else /* !CONFIG_DEBUG_PER_CPU_MAPS */ | |
645a7919 TH |
709 | |
710 | int __cpu_to_node(int cpu) | |
711 | { | |
712 | if (early_per_cpu_ptr(x86_cpu_to_node_map)) { | |
713 | printk(KERN_WARNING | |
714 | "cpu_to_node(%d): usage too early!\n", cpu); | |
715 | dump_stack(); | |
716 | return early_per_cpu_ptr(x86_cpu_to_node_map)[cpu]; | |
717 | } | |
718 | return per_cpu(x86_cpu_to_node_map, cpu); | |
719 | } | |
720 | EXPORT_SYMBOL(__cpu_to_node); | |
721 | ||
722 | /* | |
723 | * Same function as cpu_to_node() but used if called before the | |
724 | * per_cpu areas are setup. | |
725 | */ | |
726 | int early_cpu_to_node(int cpu) | |
727 | { | |
728 | if (early_per_cpu_ptr(x86_cpu_to_node_map)) | |
729 | return early_per_cpu_ptr(x86_cpu_to_node_map)[cpu]; | |
730 | ||
731 | if (!cpu_possible(cpu)) { | |
732 | printk(KERN_WARNING | |
733 | "early_cpu_to_node(%d): no per_cpu area!\n", cpu); | |
734 | dump_stack(); | |
735 | return NUMA_NO_NODE; | |
736 | } | |
737 | return per_cpu(x86_cpu_to_node_map, cpu); | |
738 | } | |
739 | ||
7a6c6547 | 740 | void debug_cpumask_set_cpu(int cpu, int node, bool enable) |
de2d9445 | 741 | { |
de2d9445 TH |
742 | struct cpumask *mask; |
743 | char buf[64]; | |
744 | ||
14392fd3 DR |
745 | if (node == NUMA_NO_NODE) { |
746 | /* early_cpu_to_node() already emits a warning and trace */ | |
7a6c6547 | 747 | return; |
14392fd3 | 748 | } |
de2d9445 TH |
749 | mask = node_to_cpumask_map[node]; |
750 | if (!mask) { | |
751 | pr_err("node_to_cpumask_map[%i] NULL\n", node); | |
752 | dump_stack(); | |
7a6c6547 | 753 | return; |
de2d9445 TH |
754 | } |
755 | ||
7a6c6547 DR |
756 | if (enable) |
757 | cpumask_set_cpu(cpu, mask); | |
758 | else | |
759 | cpumask_clear_cpu(cpu, mask); | |
760 | ||
de2d9445 TH |
761 | cpulist_scnprintf(buf, sizeof(buf), mask); |
762 | printk(KERN_DEBUG "%s cpu %d node %d: mask now %s\n", | |
763 | enable ? "numa_add_cpu" : "numa_remove_cpu", | |
764 | cpu, node, buf); | |
7a6c6547 | 765 | return; |
de2d9445 TH |
766 | } |
767 | ||
768 | # ifndef CONFIG_NUMA_EMU | |
7a6c6547 | 769 | static void __cpuinit numa_set_cpumask(int cpu, bool enable) |
de2d9445 | 770 | { |
7a6c6547 | 771 | debug_cpumask_set_cpu(cpu, early_cpu_to_node(cpu), enable); |
de2d9445 TH |
772 | } |
773 | ||
774 | void __cpuinit numa_add_cpu(int cpu) | |
775 | { | |
7a6c6547 | 776 | numa_set_cpumask(cpu, true); |
de2d9445 TH |
777 | } |
778 | ||
779 | void __cpuinit numa_remove_cpu(int cpu) | |
780 | { | |
7a6c6547 | 781 | numa_set_cpumask(cpu, false); |
de2d9445 TH |
782 | } |
783 | # endif /* !CONFIG_NUMA_EMU */ | |
784 | ||
71ee73e7 RR |
785 | /* |
786 | * Returns a pointer to the bitmask of CPUs on Node 'node'. | |
787 | */ | |
73e907de | 788 | const struct cpumask *cpumask_of_node(int node) |
71ee73e7 | 789 | { |
71ee73e7 RR |
790 | if (node >= nr_node_ids) { |
791 | printk(KERN_WARNING | |
792 | "cpumask_of_node(%d): node > nr_node_ids(%d)\n", | |
793 | node, nr_node_ids); | |
794 | dump_stack(); | |
795 | return cpu_none_mask; | |
796 | } | |
c032ef60 RR |
797 | if (node_to_cpumask_map[node] == NULL) { |
798 | printk(KERN_WARNING | |
799 | "cpumask_of_node(%d): no node_to_cpumask_map!\n", | |
800 | node); | |
801 | dump_stack(); | |
802 | return cpu_online_mask; | |
803 | } | |
0b966252 | 804 | return node_to_cpumask_map[node]; |
71ee73e7 RR |
805 | } |
806 | EXPORT_SYMBOL(cpumask_of_node); | |
645a7919 | 807 | |
de2d9445 | 808 | #endif /* !CONFIG_DEBUG_PER_CPU_MAPS */ |
a4106eae | 809 | |
bd6709a9 | 810 | #ifdef CONFIG_MEMORY_HOTPLUG |
a4106eae TH |
811 | int memory_add_physaddr_to_nid(u64 start) |
812 | { | |
813 | struct numa_meminfo *mi = &numa_meminfo; | |
814 | int nid = mi->blk[0].nid; | |
815 | int i; | |
816 | ||
817 | for (i = 0; i < mi->nr_blks; i++) | |
818 | if (mi->blk[i].start <= start && mi->blk[i].end > start) | |
819 | nid = mi->blk[i].nid; | |
820 | return nid; | |
821 | } | |
822 | EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid); | |
823 | #endif |