]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - arch/x86/mm/numa_64.c
uv_time: add parameter to uv_read_rtc()
[mirror_ubuntu-bionic-kernel.git] / arch / x86 / mm / numa_64.c
CommitLineData
e3cfe529 1/*
1da177e4
LT
2 * Generic VM initialization for x86-64 NUMA setups.
3 * Copyright 2002,2003 Andi Kleen, SuSE Labs.
e3cfe529 4 */
1da177e4
LT
5#include <linux/kernel.h>
6#include <linux/mm.h>
7#include <linux/string.h>
8#include <linux/init.h>
9#include <linux/bootmem.h>
10#include <linux/mmzone.h>
11#include <linux/ctype.h>
12#include <linux/module.h>
13#include <linux/nodemask.h>
3cc87e3f 14#include <linux/sched.h>
1da177e4
LT
15
16#include <asm/e820.h>
17#include <asm/proto.h>
18#include <asm/dma.h>
19#include <asm/numa.h>
20#include <asm/acpi.h>
c9ff0342 21#include <asm/k8.h>
1da177e4 22
6c231b7b 23struct pglist_data *node_data[MAX_NUMNODES] __read_mostly;
e3cfe529
TG
24EXPORT_SYMBOL(node_data);
25
dcf36bfa 26struct memnode memnode;
1da177e4 27
43238382 28s16 apicid_to_node[MAX_LOCAL_APIC] __cpuinitdata = {
e3cfe529 29 [0 ... MAX_LOCAL_APIC-1] = NUMA_NO_NODE
3f098c26 30};
e3cfe529 31
1da177e4 32int numa_off __initdata;
864fc31e
TG
33static unsigned long __initdata nodemap_addr;
34static unsigned long __initdata nodemap_size;
1da177e4 35
6470aff6
BG
36DEFINE_PER_CPU(int, node_number) = 0;
37EXPORT_PER_CPU_SYMBOL(node_number);
38
39/*
40 * Map cpu index to node index
41 */
42DEFINE_EARLY_PER_CPU(int, x86_cpu_to_node_map, NUMA_NO_NODE);
43EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_node_map);
44
529a3404
ED
45/*
46 * Given a shift value, try to populate memnodemap[]
47 * Returns :
48 * 1 if OK
49 * 0 if memnodmap[] too small (of shift too small)
50 * -1 if node overlap or lost ram (shift too big)
51 */
e3cfe529 52static int __init populate_memnodemap(const struct bootnode *nodes,
6ec6e0d9 53 int numnodes, int shift, int *nodeids)
1da177e4 54{
529a3404 55 unsigned long addr, end;
e3cfe529 56 int i, res = -1;
b684664f 57
43238382 58 memset(memnodemap, 0xff, sizeof(s16)*memnodemapsize);
b684664f 59 for (i = 0; i < numnodes; i++) {
529a3404
ED
60 addr = nodes[i].start;
61 end = nodes[i].end;
62 if (addr >= end)
b684664f 63 continue;
076422d2 64 if ((end >> shift) >= memnodemapsize)
529a3404
ED
65 return 0;
66 do {
43238382 67 if (memnodemap[addr >> shift] != NUMA_NO_NODE)
b684664f 68 return -1;
6ec6e0d9
SS
69
70 if (!nodeids)
71 memnodemap[addr >> shift] = i;
72 else
73 memnodemap[addr >> shift] = nodeids[i];
74
076422d2 75 addr += (1UL << shift);
529a3404
ED
76 } while (addr < end);
77 res = 1;
e3cfe529 78 }
529a3404
ED
79 return res;
80}
81
076422d2
AS
82static int __init allocate_cachealigned_memnodemap(void)
83{
24a5da73 84 unsigned long addr;
076422d2
AS
85
86 memnodemap = memnode.embedded_map;
316390b0 87 if (memnodemapsize <= ARRAY_SIZE(memnode.embedded_map))
076422d2 88 return 0;
076422d2 89
24a5da73 90 addr = 0x8000;
be3e89ee 91 nodemap_size = roundup(sizeof(s16) * memnodemapsize, L1_CACHE_BYTES);
c987d12f 92 nodemap_addr = find_e820_area(addr, max_pfn<<PAGE_SHIFT,
24a5da73 93 nodemap_size, L1_CACHE_BYTES);
076422d2
AS
94 if (nodemap_addr == -1UL) {
95 printk(KERN_ERR
96 "NUMA: Unable to allocate Memory to Node hash map\n");
97 nodemap_addr = nodemap_size = 0;
98 return -1;
99 }
24a5da73 100 memnodemap = phys_to_virt(nodemap_addr);
25eff8d4 101 reserve_early(nodemap_addr, nodemap_addr + nodemap_size, "MEMNODEMAP");
076422d2
AS
102
103 printk(KERN_DEBUG "NUMA: Allocated memnodemap from %lx - %lx\n",
104 nodemap_addr, nodemap_addr + nodemap_size);
105 return 0;
106}
107
108/*
109 * The LSB of all start and end addresses in the node map is the value of the
110 * maximum possible shift.
111 */
e3cfe529
TG
112static int __init extract_lsb_from_nodes(const struct bootnode *nodes,
113 int numnodes)
529a3404 114{
54413927 115 int i, nodes_used = 0;
076422d2
AS
116 unsigned long start, end;
117 unsigned long bitfield = 0, memtop = 0;
118
119 for (i = 0; i < numnodes; i++) {
120 start = nodes[i].start;
121 end = nodes[i].end;
122 if (start >= end)
123 continue;
54413927
AS
124 bitfield |= start;
125 nodes_used++;
076422d2
AS
126 if (end > memtop)
127 memtop = end;
128 }
54413927
AS
129 if (nodes_used <= 1)
130 i = 63;
131 else
132 i = find_first_bit(&bitfield, sizeof(unsigned long)*8);
076422d2
AS
133 memnodemapsize = (memtop >> i)+1;
134 return i;
135}
529a3404 136
6ec6e0d9
SS
137int __init compute_hash_shift(struct bootnode *nodes, int numnodes,
138 int *nodeids)
076422d2
AS
139{
140 int shift;
529a3404 141
076422d2
AS
142 shift = extract_lsb_from_nodes(nodes, numnodes);
143 if (allocate_cachealigned_memnodemap())
144 return -1;
6b050f80 145 printk(KERN_DEBUG "NUMA: Using %d for the hash shift.\n",
529a3404
ED
146 shift);
147
6ec6e0d9 148 if (populate_memnodemap(nodes, numnodes, shift, nodeids) != 1) {
e3cfe529
TG
149 printk(KERN_INFO "Your memory is not aligned you need to "
150 "rebuild your kernel with a bigger NODEMAPSIZE "
151 "shift=%d\n", shift);
529a3404
ED
152 return -1;
153 }
b684664f 154 return shift;
1da177e4
LT
155}
156
f2dbcfa7 157int __meminit __early_pfn_to_nid(unsigned long pfn)
bbfceef4
MT
158{
159 return phys_to_nid(pfn << PAGE_SHIFT);
160}
bbfceef4 161
e3cfe529 162static void * __init early_node_mem(int nodeid, unsigned long start,
24a5da73
YL
163 unsigned long end, unsigned long size,
164 unsigned long align)
a8062231 165{
24a5da73 166 unsigned long mem = find_e820_area(start, end, size, align);
a8062231 167 void *ptr;
e3cfe529 168
9347e0b0 169 if (mem != -1L)
a8062231 170 return __va(mem);
9347e0b0 171
24a5da73 172 ptr = __alloc_bootmem_nopanic(size, align, __pa(MAX_DMA_ADDRESS));
83e83d54 173 if (ptr == NULL) {
a8062231 174 printk(KERN_ERR "Cannot find %lu bytes in node %d\n",
e3cfe529 175 size, nodeid);
a8062231
AK
176 return NULL;
177 }
178 return ptr;
179}
180
1da177e4 181/* Initialize bootmem allocator for a node */
e3cfe529
TG
182void __init setup_node_bootmem(int nodeid, unsigned long start,
183 unsigned long end)
184{
886533a3 185 unsigned long start_pfn, last_pfn, bootmap_pages, bootmap_size;
e3cfe529 186 unsigned long bootmap_start, nodedata_phys;
a8062231 187 void *bootmap;
be3e89ee 188 const int pgdat_size = roundup(sizeof(pg_data_t), PAGE_SIZE);
1a27fc0a 189 int nid;
1da177e4 190
be3e89ee 191 start = roundup(start, ZONE_ALIGN);
1da177e4 192
e3cfe529
TG
193 printk(KERN_INFO "Bootmem setup node %d %016lx-%016lx\n", nodeid,
194 start, end);
1da177e4
LT
195
196 start_pfn = start >> PAGE_SHIFT;
886533a3 197 last_pfn = end >> PAGE_SHIFT;
1da177e4 198
24a5da73
YL
199 node_data[nodeid] = early_node_mem(nodeid, start, end, pgdat_size,
200 SMP_CACHE_BYTES);
a8062231
AK
201 if (node_data[nodeid] == NULL)
202 return;
203 nodedata_phys = __pa(node_data[nodeid]);
6118f76f
YL
204 printk(KERN_INFO " NODE_DATA [%016lx - %016lx]\n", nodedata_phys,
205 nodedata_phys + pgdat_size - 1);
1da177e4 206
1da177e4 207 memset(NODE_DATA(nodeid), 0, sizeof(pg_data_t));
b61bfa3c 208 NODE_DATA(nodeid)->bdata = &bootmem_node_data[nodeid];
1da177e4 209 NODE_DATA(nodeid)->node_start_pfn = start_pfn;
886533a3 210 NODE_DATA(nodeid)->node_spanned_pages = last_pfn - start_pfn;
1da177e4 211
1a27fc0a
YL
212 /*
213 * Find a place for the bootmem map
214 * nodedata_phys could be on other nodes by alloc_bootmem,
215 * so need to sure bootmap_start not to be small, otherwise
216 * early_node_mem will get that with find_e820_area instead
217 * of alloc_bootmem, that could clash with reserved range
218 */
886533a3 219 bootmap_pages = bootmem_bootmap_pages(last_pfn - start_pfn);
1a27fc0a
YL
220 nid = phys_to_nid(nodedata_phys);
221 if (nid == nodeid)
be3e89ee 222 bootmap_start = roundup(nodedata_phys + pgdat_size, PAGE_SIZE);
1a27fc0a 223 else
be3e89ee 224 bootmap_start = roundup(start, PAGE_SIZE);
24a5da73 225 /*
e9197bf0 226 * SMP_CACHE_BYTES could be enough, but init_bootmem_node like
24a5da73
YL
227 * to use that to align to PAGE_SIZE
228 */
a8062231 229 bootmap = early_node_mem(nodeid, bootmap_start, end,
24a5da73 230 bootmap_pages<<PAGE_SHIFT, PAGE_SIZE);
a8062231
AK
231 if (bootmap == NULL) {
232 if (nodedata_phys < start || nodedata_phys >= end)
37bff62e 233 free_bootmem(nodedata_phys, pgdat_size);
a8062231
AK
234 node_data[nodeid] = NULL;
235 return;
236 }
237 bootmap_start = __pa(bootmap);
e3cfe529 238
1da177e4 239 bootmap_size = init_bootmem_node(NODE_DATA(nodeid),
e3cfe529 240 bootmap_start >> PAGE_SHIFT,
886533a3 241 start_pfn, last_pfn);
1da177e4 242
6118f76f
YL
243 printk(KERN_INFO " bootmap [%016lx - %016lx] pages %lx\n",
244 bootmap_start, bootmap_start + bootmap_size - 1,
245 bootmap_pages);
246
5cb248ab 247 free_bootmem_with_active_regions(nodeid, end);
1da177e4 248
1a27fc0a
YL
249 /*
250 * convert early reserve to bootmem reserve earlier
251 * otherwise early_node_mem could use early reserved mem
252 * on previous node
253 */
254 early_res_to_bootmem(start, end);
255
256 /*
257 * in some case early_node_mem could use alloc_bootmem
258 * to get range on other node, don't reserve that again
259 */
260 if (nid != nodeid)
261 printk(KERN_INFO " NODE_DATA(%d) on node %d\n", nodeid, nid);
262 else
263 reserve_bootmem_node(NODE_DATA(nodeid), nodedata_phys,
264 pgdat_size, BOOTMEM_DEFAULT);
265 nid = phys_to_nid(bootmap_start);
266 if (nid != nodeid)
267 printk(KERN_INFO " bootmap(%d) on node %d\n", nodeid, nid);
268 else
269 reserve_bootmem_node(NODE_DATA(nodeid), bootmap_start,
270 bootmap_pages<<PAGE_SHIFT, BOOTMEM_DEFAULT);
271
68a3a7fe
AK
272#ifdef CONFIG_ACPI_NUMA
273 srat_reserve_add_area(nodeid);
274#endif
1da177e4 275 node_set_online(nodeid);
e3cfe529 276}
1da177e4 277
e3cfe529
TG
278/*
279 * There are unfortunately some poorly designed mainboards around that
280 * only connect memory to a single CPU. This breaks the 1:1 cpu->node
281 * mapping. To avoid this fill in the mapping for all possible CPUs,
282 * as the number of CPUs is not known yet. We round robin the existing
283 * nodes.
284 */
1da177e4
LT
285void __init numa_init_array(void)
286{
287 int rr, i;
e3cfe529 288
85cc5135 289 rr = first_node(node_online_map);
168ef543 290 for (i = 0; i < nr_cpu_ids; i++) {
1ce35712 291 if (early_cpu_to_node(i) != NUMA_NO_NODE)
1da177e4 292 continue;
e3cfe529 293 numa_set_node(i, rr);
1da177e4
LT
294 rr = next_node(rr, node_online_map);
295 if (rr == MAX_NUMNODES)
296 rr = first_node(node_online_map);
1da177e4 297 }
1da177e4
LT
298}
299
300#ifdef CONFIG_NUMA_EMU
53fee04f 301/* Numa emulation */
864fc31e 302static char *cmdline __initdata;
1da177e4 303
53fee04f 304/*
e3cfe529
TG
305 * Setups up nid to range from addr to addr + size. If the end
306 * boundary is greater than max_addr, then max_addr is used instead.
307 * The return value is 0 if there is additional memory left for
308 * allocation past addr and -1 otherwise. addr is adjusted to be at
309 * the end of the node.
53fee04f 310 */
8b8ca80e
DR
311static int __init setup_node_range(int nid, struct bootnode *nodes, u64 *addr,
312 u64 size, u64 max_addr)
53fee04f 313{
8b8ca80e 314 int ret = 0;
e3cfe529 315
8b8ca80e
DR
316 nodes[nid].start = *addr;
317 *addr += size;
318 if (*addr >= max_addr) {
319 *addr = max_addr;
320 ret = -1;
321 }
322 nodes[nid].end = *addr;
e3f1caee 323 node_set(nid, node_possible_map);
8b8ca80e
DR
324 printk(KERN_INFO "Faking node %d at %016Lx-%016Lx (%LuMB)\n", nid,
325 nodes[nid].start, nodes[nid].end,
326 (nodes[nid].end - nodes[nid].start) >> 20);
327 return ret;
53fee04f
RS
328}
329
8b8ca80e
DR
330/*
331 * Splits num_nodes nodes up equally starting at node_start. The return value
332 * is the number of nodes split up and addr is adjusted to be at the end of the
333 * last node allocated.
334 */
335static int __init split_nodes_equally(struct bootnode *nodes, u64 *addr,
336 u64 max_addr, int node_start,
337 int num_nodes)
1da177e4 338{
8b8ca80e
DR
339 unsigned int big;
340 u64 size;
341 int i;
53fee04f 342
8b8ca80e
DR
343 if (num_nodes <= 0)
344 return -1;
345 if (num_nodes > MAX_NUMNODES)
346 num_nodes = MAX_NUMNODES;
a7e96629 347 size = (max_addr - *addr - e820_hole_size(*addr, max_addr)) /
8b8ca80e 348 num_nodes;
53fee04f 349 /*
8b8ca80e
DR
350 * Calculate the number of big nodes that can be allocated as a result
351 * of consolidating the leftovers.
53fee04f 352 */
8b8ca80e
DR
353 big = ((size & ~FAKE_NODE_MIN_HASH_MASK) * num_nodes) /
354 FAKE_NODE_MIN_SIZE;
355
356 /* Round down to nearest FAKE_NODE_MIN_SIZE. */
357 size &= FAKE_NODE_MIN_HASH_MASK;
358 if (!size) {
359 printk(KERN_ERR "Not enough memory for each node. "
360 "NUMA emulation disabled.\n");
361 return -1;
53fee04f 362 }
8b8ca80e
DR
363
364 for (i = node_start; i < num_nodes + node_start; i++) {
365 u64 end = *addr + size;
e3cfe529 366
53fee04f
RS
367 if (i < big)
368 end += FAKE_NODE_MIN_SIZE;
369 /*
8b8ca80e
DR
370 * The final node can have the remaining system RAM. Other
371 * nodes receive roughly the same amount of available pages.
53fee04f 372 */
8b8ca80e
DR
373 if (i == num_nodes + node_start - 1)
374 end = max_addr;
375 else
a7e96629 376 while (end - *addr - e820_hole_size(*addr, end) <
8b8ca80e
DR
377 size) {
378 end += FAKE_NODE_MIN_SIZE;
379 if (end > max_addr) {
380 end = max_addr;
381 break;
382 }
383 }
384 if (setup_node_range(i, nodes, addr, end - *addr, max_addr) < 0)
385 break;
386 }
387 return i - node_start + 1;
388}
389
382591d5
DR
390/*
391 * Splits the remaining system RAM into chunks of size. The remaining memory is
392 * always assigned to a final node and can be asymmetric. Returns the number of
393 * nodes split.
394 */
395static int __init split_nodes_by_size(struct bootnode *nodes, u64 *addr,
396 u64 max_addr, int node_start, u64 size)
397{
398 int i = node_start;
399 size = (size << 20) & FAKE_NODE_MIN_HASH_MASK;
400 while (!setup_node_range(i++, nodes, addr, size, max_addr))
401 ;
402 return i - node_start;
403}
404
8b8ca80e 405/*
886533a3 406 * Sets up the system RAM area from start_pfn to last_pfn according to the
8b8ca80e
DR
407 * numa=fake command-line option.
408 */
f46bdf2d
MT
409static struct bootnode nodes[MAX_NUMNODES] __initdata;
410
886533a3 411static int __init numa_emulation(unsigned long start_pfn, unsigned long last_pfn)
8b8ca80e 412{
e3cfe529 413 u64 size, addr = start_pfn << PAGE_SHIFT;
886533a3 414 u64 max_addr = last_pfn << PAGE_SHIFT;
e3cfe529 415 int num_nodes = 0, num = 0, coeff_flag, coeff = -1, i;
8b8ca80e
DR
416
417 memset(&nodes, 0, sizeof(nodes));
418 /*
419 * If the numa=fake command-line is just a single number N, split the
420 * system RAM into N fake nodes.
421 */
422 if (!strchr(cmdline, '*') && !strchr(cmdline, ',')) {
e3cfe529
TG
423 long n = simple_strtol(cmdline, NULL, 0);
424
425 num_nodes = split_nodes_equally(nodes, &addr, max_addr, 0, n);
8b8ca80e
DR
426 if (num_nodes < 0)
427 return num_nodes;
428 goto out;
429 }
430
431 /* Parse the command line. */
382591d5 432 for (coeff_flag = 0; ; cmdline++) {
8b8ca80e
DR
433 if (*cmdline && isdigit(*cmdline)) {
434 num = num * 10 + *cmdline - '0';
435 continue;
53fee04f 436 }
382591d5
DR
437 if (*cmdline == '*') {
438 if (num > 0)
439 coeff = num;
440 coeff_flag = 1;
441 }
8b8ca80e 442 if (!*cmdline || *cmdline == ',') {
382591d5
DR
443 if (!coeff_flag)
444 coeff = 1;
8b8ca80e
DR
445 /*
446 * Round down to the nearest FAKE_NODE_MIN_SIZE.
447 * Command-line coefficients are in megabytes.
448 */
449 size = ((u64)num << 20) & FAKE_NODE_MIN_HASH_MASK;
382591d5 450 if (size)
8b8ca80e
DR
451 for (i = 0; i < coeff; i++, num_nodes++)
452 if (setup_node_range(num_nodes, nodes,
453 &addr, size, max_addr) < 0)
454 goto done;
382591d5
DR
455 if (!*cmdline)
456 break;
457 coeff_flag = 0;
458 coeff = -1;
53fee04f 459 }
8b8ca80e
DR
460 num = 0;
461 }
462done:
463 if (!num_nodes)
464 return -1;
14694d73 465 /* Fill remainder of system RAM, if appropriate. */
8b8ca80e 466 if (addr < max_addr) {
382591d5
DR
467 if (coeff_flag && coeff < 0) {
468 /* Split remaining nodes into num-sized chunks */
469 num_nodes += split_nodes_by_size(nodes, &addr, max_addr,
470 num_nodes, num);
471 goto out;
472 }
14694d73
DR
473 switch (*(cmdline - 1)) {
474 case '*':
475 /* Split remaining nodes into coeff chunks */
476 if (coeff <= 0)
477 break;
478 num_nodes += split_nodes_equally(nodes, &addr, max_addr,
479 num_nodes, coeff);
480 break;
481 case ',':
482 /* Do not allocate remaining system RAM */
483 break;
484 default:
485 /* Give one final node */
486 setup_node_range(num_nodes, nodes, &addr,
487 max_addr - addr, max_addr);
488 num_nodes++;
489 }
8b8ca80e
DR
490 }
491out:
6ec6e0d9 492 memnode_shift = compute_hash_shift(nodes, num_nodes, NULL);
8b8ca80e
DR
493 if (memnode_shift < 0) {
494 memnode_shift = 0;
495 printk(KERN_ERR "No NUMA hash function found. NUMA emulation "
496 "disabled.\n");
497 return -1;
498 }
499
500 /*
501 * We need to vacate all active ranges that may have been registered by
1c05f093
DR
502 * SRAT and set acpi_numa to -1 so that srat_disabled() always returns
503 * true. NUMA emulation has succeeded so we will not scan ACPI nodes.
8b8ca80e
DR
504 */
505 remove_all_active_ranges();
1c05f093
DR
506#ifdef CONFIG_ACPI_NUMA
507 acpi_numa = -1;
508#endif
e3f1caee 509 for_each_node_mask(i, node_possible_map) {
5cb248ab
MG
510 e820_register_active_regions(i, nodes[i].start >> PAGE_SHIFT,
511 nodes[i].end >> PAGE_SHIFT);
e3cfe529 512 setup_node_bootmem(i, nodes[i].start, nodes[i].end);
5cb248ab 513 }
3484d798 514 acpi_fake_nodes(nodes, num_nodes);
e3cfe529
TG
515 numa_init_array();
516 return 0;
1da177e4 517}
8b8ca80e 518#endif /* CONFIG_NUMA_EMU */
1da177e4 519
1f75d7e3 520void __init initmem_init(unsigned long start_pfn, unsigned long last_pfn)
e3cfe529 521{
1da177e4
LT
522 int i;
523
e3f1caee 524 nodes_clear(node_possible_map);
b7ad149d 525 nodes_clear(node_online_map);
e3f1caee 526
1da177e4 527#ifdef CONFIG_NUMA_EMU
886533a3 528 if (cmdline && !numa_emulation(start_pfn, last_pfn))
e3cfe529 529 return;
e3f1caee 530 nodes_clear(node_possible_map);
b7ad149d 531 nodes_clear(node_online_map);
1da177e4
LT
532#endif
533
534#ifdef CONFIG_ACPI_NUMA
535 if (!numa_off && !acpi_scan_nodes(start_pfn << PAGE_SHIFT,
886533a3 536 last_pfn << PAGE_SHIFT))
e3cfe529 537 return;
e3f1caee 538 nodes_clear(node_possible_map);
b7ad149d 539 nodes_clear(node_online_map);
1da177e4
LT
540#endif
541
542#ifdef CONFIG_K8_NUMA
e3cfe529 543 if (!numa_off && !k8_scan_nodes(start_pfn<<PAGE_SHIFT,
886533a3 544 last_pfn<<PAGE_SHIFT))
1da177e4 545 return;
e3f1caee 546 nodes_clear(node_possible_map);
b7ad149d 547 nodes_clear(node_online_map);
1da177e4
LT
548#endif
549 printk(KERN_INFO "%s\n",
550 numa_off ? "NUMA turned off" : "No NUMA configuration found");
551
e3cfe529 552 printk(KERN_INFO "Faking a node at %016lx-%016lx\n",
1da177e4 553 start_pfn << PAGE_SHIFT,
886533a3 554 last_pfn << PAGE_SHIFT);
e3cfe529
TG
555 /* setup dummy node covering all memory */
556 memnode_shift = 63;
076422d2 557 memnodemap = memnode.embedded_map;
1da177e4 558 memnodemap[0] = 0;
1da177e4 559 node_set_online(0);
e3f1caee 560 node_set(0, node_possible_map);
168ef543 561 for (i = 0; i < nr_cpu_ids; i++)
69d81fcd 562 numa_set_node(i, 0);
886533a3
TG
563 e820_register_active_regions(0, start_pfn, last_pfn);
564 setup_node_bootmem(0, start_pfn << PAGE_SHIFT, last_pfn << PAGE_SHIFT);
69d81fcd
AK
565}
566
e3cfe529
TG
567unsigned long __init numa_free_all_bootmem(void)
568{
1da177e4 569 unsigned long pages = 0;
e3cfe529
TG
570 int i;
571
572 for_each_online_node(i)
1da177e4 573 pages += free_all_bootmem_node(NODE_DATA(i));
e3cfe529 574
1da177e4 575 return pages;
e3cfe529 576}
1da177e4
LT
577
578void __init paging_init(void)
e3cfe529 579{
6391af17 580 unsigned long max_zone_pfns[MAX_NR_ZONES];
e3cfe529 581
6391af17
MG
582 memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
583 max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN;
584 max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN;
c987d12f 585 max_zone_pfns[ZONE_NORMAL] = max_pfn;
d3ee871e 586
f0a5a58a
BP
587 sparse_memory_present_with_active_regions(MAX_NUMNODES);
588 sparse_init();
d3ee871e 589
5cb248ab 590 free_area_init_nodes(max_zone_pfns);
e3cfe529 591}
1da177e4 592
2c8c0e6b 593static __init int numa_setup(char *opt)
e3cfe529 594{
2c8c0e6b
AK
595 if (!opt)
596 return -EINVAL;
e3cfe529 597 if (!strncmp(opt, "off", 3))
1da177e4
LT
598 numa_off = 1;
599#ifdef CONFIG_NUMA_EMU
8b8ca80e
DR
600 if (!strncmp(opt, "fake=", 5))
601 cmdline = opt + 5;
1da177e4
LT
602#endif
603#ifdef CONFIG_ACPI_NUMA
e3cfe529
TG
604 if (!strncmp(opt, "noacpi", 6))
605 acpi_numa = -1;
606 if (!strncmp(opt, "hotadd=", 7))
68a3a7fe 607 hotadd_percent = simple_strtoul(opt+7, NULL, 10);
1da177e4 608#endif
2c8c0e6b 609 return 0;
e3cfe529 610}
2c8c0e6b
AK
611early_param("numa", numa_setup);
612
23ca4bba 613#ifdef CONFIG_NUMA
05b3cbd8
RT
614/*
615 * Setup early cpu_to_node.
616 *
617 * Populate cpu_to_node[] only if x86_cpu_to_apicid[],
618 * and apicid_to_node[] tables have valid entries for a CPU.
619 * This means we skip cpu_to_node[] initialisation for NUMA
620 * emulation and faking node case (when running a kernel compiled
621 * for NUMA on a non NUMA box), which is OK as cpu_to_node[]
622 * is already initialized in a round robin manner at numa_init_array,
623 * prior to this call, and this initialization is good enough
624 * for the fake NUMA cases.
23ca4bba
MT
625 *
626 * Called before the per_cpu areas are setup.
05b3cbd8
RT
627 */
628void __init init_cpu_to_node(void)
629{
23ca4bba
MT
630 int cpu;
631 u16 *cpu_to_apicid = early_per_cpu_ptr(x86_cpu_to_apicid);
e3cfe529 632
23ca4bba
MT
633 BUG_ON(cpu_to_apicid == NULL);
634
635 for_each_possible_cpu(cpu) {
7c9e92b6 636 int node;
23ca4bba 637 u16 apicid = cpu_to_apicid[cpu];
e3cfe529 638
05b3cbd8
RT
639 if (apicid == BAD_APICID)
640 continue;
7c9e92b6
YL
641 node = apicid_to_node[apicid];
642 if (node == NUMA_NO_NODE)
05b3cbd8 643 continue;
7c9e92b6
YL
644 if (!node_online(node))
645 continue;
23ca4bba 646 numa_set_node(cpu, node);
05b3cbd8
RT
647 }
648}
23ca4bba 649#endif
05b3cbd8 650
cf050132 651
6470aff6
BG
652void __cpuinit numa_set_node(int cpu, int node)
653{
654 int *cpu_to_node_map = early_per_cpu_ptr(x86_cpu_to_node_map);
655
656 /* early setting, no percpu area yet */
657 if (cpu_to_node_map) {
658 cpu_to_node_map[cpu] = node;
659 return;
660 }
661
662#ifdef CONFIG_DEBUG_PER_CPU_MAPS
44581a28 663 if (cpu >= nr_cpu_ids || !cpu_possible(cpu)) {
6470aff6
BG
664 printk(KERN_ERR "numa_set_node: invalid cpu# (%d)\n", cpu);
665 dump_stack();
666 return;
667 }
668#endif
669 per_cpu(x86_cpu_to_node_map, cpu) = node;
670
671 if (node != NUMA_NO_NODE)
672 per_cpu(node_number, cpu) = node;
673}
674
675void __cpuinit numa_clear_node(int cpu)
676{
677 numa_set_node(cpu, NUMA_NO_NODE);
678}
679
680#ifndef CONFIG_DEBUG_PER_CPU_MAPS
681
682void __cpuinit numa_add_cpu(int cpu)
683{
c032ef60 684 cpumask_set_cpu(cpu, node_to_cpumask_map[early_cpu_to_node(cpu)]);
6470aff6
BG
685}
686
687void __cpuinit numa_remove_cpu(int cpu)
688{
c032ef60 689 cpumask_clear_cpu(cpu, node_to_cpumask_map[early_cpu_to_node(cpu)]);
6470aff6
BG
690}
691
692#else /* CONFIG_DEBUG_PER_CPU_MAPS */
693
694/*
695 * --------- debug versions of the numa functions ---------
696 */
697static void __cpuinit numa_set_cpumask(int cpu, int enable)
698{
699 int node = early_cpu_to_node(cpu);
73e907de 700 struct cpumask *mask;
6470aff6
BG
701 char buf[64];
702
c032ef60
RR
703 mask = node_to_cpumask_map[node];
704 if (mask == NULL) {
705 printk(KERN_ERR "node_to_cpumask_map[%i] NULL\n", node);
6470aff6
BG
706 dump_stack();
707 return;
708 }
709
6470aff6 710 if (enable)
c032ef60 711 cpumask_set_cpu(cpu, mask);
6470aff6 712 else
c032ef60 713 cpumask_clear_cpu(cpu, mask);
6470aff6
BG
714
715 cpulist_scnprintf(buf, sizeof(buf), mask);
716 printk(KERN_DEBUG "%s cpu %d node %d: mask now %s\n",
717 enable ? "numa_add_cpu" : "numa_remove_cpu", cpu, node, buf);
718}
719
720void __cpuinit numa_add_cpu(int cpu)
721{
722 numa_set_cpumask(cpu, 1);
723}
724
725void __cpuinit numa_remove_cpu(int cpu)
726{
727 numa_set_cpumask(cpu, 0);
728}
729
730int cpu_to_node(int cpu)
731{
732 if (early_per_cpu_ptr(x86_cpu_to_node_map)) {
733 printk(KERN_WARNING
734 "cpu_to_node(%d): usage too early!\n", cpu);
735 dump_stack();
736 return early_per_cpu_ptr(x86_cpu_to_node_map)[cpu];
737 }
738 return per_cpu(x86_cpu_to_node_map, cpu);
739}
740EXPORT_SYMBOL(cpu_to_node);
741
742/*
743 * Same function as cpu_to_node() but used if called before the
744 * per_cpu areas are setup.
745 */
746int early_cpu_to_node(int cpu)
747{
748 if (early_per_cpu_ptr(x86_cpu_to_node_map))
749 return early_per_cpu_ptr(x86_cpu_to_node_map)[cpu];
750
44581a28 751 if (!cpu_possible(cpu)) {
6470aff6
BG
752 printk(KERN_WARNING
753 "early_cpu_to_node(%d): no per_cpu area!\n", cpu);
754 dump_stack();
755 return NUMA_NO_NODE;
756 }
757 return per_cpu(x86_cpu_to_node_map, cpu);
758}
759
6470aff6
BG
760/*
761 * --------- end of debug versions of the numa functions ---------
762 */
763
764#endif /* CONFIG_DEBUG_PER_CPU_MAPS */