]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Generic VM initialization for x86-64 NUMA setups. | |
3 | * Copyright 2002,2003 Andi Kleen, SuSE Labs. | |
4 | */ | |
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> | |
14 | ||
15 | #include <asm/e820.h> | |
16 | #include <asm/proto.h> | |
17 | #include <asm/dma.h> | |
18 | #include <asm/numa.h> | |
19 | #include <asm/acpi.h> | |
20 | ||
21 | #ifndef Dprintk | |
22 | #define Dprintk(x...) | |
23 | #endif | |
24 | ||
6c231b7b | 25 | struct pglist_data *node_data[MAX_NUMNODES] __read_mostly; |
1da177e4 LT |
26 | bootmem_data_t plat_node_bdata[MAX_NUMNODES]; |
27 | ||
dcf36bfa | 28 | struct memnode memnode; |
1da177e4 | 29 | |
3f098c26 AK |
30 | unsigned char cpu_to_node[NR_CPUS] __read_mostly = { |
31 | [0 ... NR_CPUS-1] = NUMA_NO_NODE | |
0b07e984 | 32 | }; |
3f098c26 AK |
33 | unsigned char apicid_to_node[MAX_LOCAL_APIC] __cpuinitdata = { |
34 | [0 ... MAX_LOCAL_APIC-1] = NUMA_NO_NODE | |
35 | }; | |
36 | cpumask_t node_to_cpumask[MAX_NUMNODES] __read_mostly; | |
1da177e4 LT |
37 | |
38 | int numa_off __initdata; | |
076422d2 AS |
39 | unsigned long __initdata nodemap_addr; |
40 | unsigned long __initdata nodemap_size; | |
1da177e4 | 41 | |
529a3404 ED |
42 | |
43 | /* | |
44 | * Given a shift value, try to populate memnodemap[] | |
45 | * Returns : | |
46 | * 1 if OK | |
47 | * 0 if memnodmap[] too small (of shift too small) | |
48 | * -1 if node overlap or lost ram (shift too big) | |
49 | */ | |
d18ff470 | 50 | static int __init |
abe059e7 | 51 | populate_memnodemap(const struct bootnode *nodes, int numnodes, int shift) |
1da177e4 LT |
52 | { |
53 | int i; | |
529a3404 ED |
54 | int res = -1; |
55 | unsigned long addr, end; | |
b684664f | 56 | |
076422d2 | 57 | memset(memnodemap, 0xff, memnodemapsize); |
b684664f | 58 | for (i = 0; i < numnodes; i++) { |
529a3404 ED |
59 | addr = nodes[i].start; |
60 | end = nodes[i].end; | |
61 | if (addr >= end) | |
b684664f | 62 | continue; |
076422d2 | 63 | if ((end >> shift) >= memnodemapsize) |
529a3404 ED |
64 | return 0; |
65 | do { | |
66 | if (memnodemap[addr >> shift] != 0xff) | |
b684664f | 67 | return -1; |
b684664f | 68 | memnodemap[addr >> shift] = i; |
076422d2 | 69 | addr += (1UL << shift); |
529a3404 ED |
70 | } while (addr < end); |
71 | res = 1; | |
1da177e4 | 72 | } |
529a3404 ED |
73 | return res; |
74 | } | |
75 | ||
076422d2 AS |
76 | static int __init allocate_cachealigned_memnodemap(void) |
77 | { | |
78 | unsigned long pad, pad_addr; | |
79 | ||
80 | memnodemap = memnode.embedded_map; | |
54413927 | 81 | if (memnodemapsize <= 48) |
076422d2 | 82 | return 0; |
076422d2 AS |
83 | |
84 | pad = L1_CACHE_BYTES - 1; | |
85 | pad_addr = 0x8000; | |
86 | nodemap_size = pad + memnodemapsize; | |
87 | nodemap_addr = find_e820_area(pad_addr, end_pfn<<PAGE_SHIFT, | |
88 | nodemap_size); | |
89 | if (nodemap_addr == -1UL) { | |
90 | printk(KERN_ERR | |
91 | "NUMA: Unable to allocate Memory to Node hash map\n"); | |
92 | nodemap_addr = nodemap_size = 0; | |
93 | return -1; | |
94 | } | |
95 | pad_addr = (nodemap_addr + pad) & ~pad; | |
96 | memnodemap = phys_to_virt(pad_addr); | |
97 | ||
98 | printk(KERN_DEBUG "NUMA: Allocated memnodemap from %lx - %lx\n", | |
99 | nodemap_addr, nodemap_addr + nodemap_size); | |
100 | return 0; | |
101 | } | |
102 | ||
103 | /* | |
104 | * The LSB of all start and end addresses in the node map is the value of the | |
105 | * maximum possible shift. | |
106 | */ | |
107 | static int __init | |
108 | extract_lsb_from_nodes (const struct bootnode *nodes, int numnodes) | |
529a3404 | 109 | { |
54413927 | 110 | int i, nodes_used = 0; |
076422d2 AS |
111 | unsigned long start, end; |
112 | unsigned long bitfield = 0, memtop = 0; | |
113 | ||
114 | for (i = 0; i < numnodes; i++) { | |
115 | start = nodes[i].start; | |
116 | end = nodes[i].end; | |
117 | if (start >= end) | |
118 | continue; | |
54413927 AS |
119 | bitfield |= start; |
120 | nodes_used++; | |
076422d2 AS |
121 | if (end > memtop) |
122 | memtop = end; | |
123 | } | |
54413927 AS |
124 | if (nodes_used <= 1) |
125 | i = 63; | |
126 | else | |
127 | i = find_first_bit(&bitfield, sizeof(unsigned long)*8); | |
076422d2 AS |
128 | memnodemapsize = (memtop >> i)+1; |
129 | return i; | |
130 | } | |
529a3404 | 131 | |
076422d2 AS |
132 | int __init compute_hash_shift(struct bootnode *nodes, int numnodes) |
133 | { | |
134 | int shift; | |
529a3404 | 135 | |
076422d2 AS |
136 | shift = extract_lsb_from_nodes(nodes, numnodes); |
137 | if (allocate_cachealigned_memnodemap()) | |
138 | return -1; | |
6b050f80 | 139 | printk(KERN_DEBUG "NUMA: Using %d for the hash shift.\n", |
529a3404 ED |
140 | shift); |
141 | ||
142 | if (populate_memnodemap(nodes, numnodes, shift) != 1) { | |
143 | printk(KERN_INFO | |
144 | "Your memory is not aligned you need to rebuild your kernel " | |
145 | "with a bigger NODEMAPSIZE shift=%d\n", | |
146 | shift); | |
147 | return -1; | |
148 | } | |
b684664f | 149 | return shift; |
1da177e4 LT |
150 | } |
151 | ||
bbfceef4 MT |
152 | #ifdef CONFIG_SPARSEMEM |
153 | int early_pfn_to_nid(unsigned long pfn) | |
154 | { | |
155 | return phys_to_nid(pfn << PAGE_SHIFT); | |
156 | } | |
157 | #endif | |
158 | ||
a8062231 AK |
159 | static void * __init |
160 | early_node_mem(int nodeid, unsigned long start, unsigned long end, | |
161 | unsigned long size) | |
162 | { | |
163 | unsigned long mem = find_e820_area(start, end, size); | |
164 | void *ptr; | |
165 | if (mem != -1L) | |
166 | return __va(mem); | |
167 | ptr = __alloc_bootmem_nopanic(size, | |
168 | SMP_CACHE_BYTES, __pa(MAX_DMA_ADDRESS)); | |
169 | if (ptr == 0) { | |
170 | printk(KERN_ERR "Cannot find %lu bytes in node %d\n", | |
171 | size, nodeid); | |
172 | return NULL; | |
173 | } | |
174 | return ptr; | |
175 | } | |
176 | ||
1da177e4 LT |
177 | /* Initialize bootmem allocator for a node */ |
178 | void __init setup_node_bootmem(int nodeid, unsigned long start, unsigned long end) | |
179 | { | |
180 | unsigned long start_pfn, end_pfn, bootmap_pages, bootmap_size, bootmap_start; | |
181 | unsigned long nodedata_phys; | |
a8062231 | 182 | void *bootmap; |
1da177e4 LT |
183 | const int pgdat_size = round_up(sizeof(pg_data_t), PAGE_SIZE); |
184 | ||
185 | start = round_up(start, ZONE_ALIGN); | |
186 | ||
6b050f80 | 187 | printk(KERN_INFO "Bootmem setup node %d %016lx-%016lx\n", nodeid, start, end); |
1da177e4 LT |
188 | |
189 | start_pfn = start >> PAGE_SHIFT; | |
190 | end_pfn = end >> PAGE_SHIFT; | |
191 | ||
a8062231 AK |
192 | node_data[nodeid] = early_node_mem(nodeid, start, end, pgdat_size); |
193 | if (node_data[nodeid] == NULL) | |
194 | return; | |
195 | nodedata_phys = __pa(node_data[nodeid]); | |
1da177e4 | 196 | |
1da177e4 LT |
197 | memset(NODE_DATA(nodeid), 0, sizeof(pg_data_t)); |
198 | NODE_DATA(nodeid)->bdata = &plat_node_bdata[nodeid]; | |
199 | NODE_DATA(nodeid)->node_start_pfn = start_pfn; | |
200 | NODE_DATA(nodeid)->node_spanned_pages = end_pfn - start_pfn; | |
201 | ||
202 | /* Find a place for the bootmem map */ | |
203 | bootmap_pages = bootmem_bootmap_pages(end_pfn - start_pfn); | |
204 | bootmap_start = round_up(nodedata_phys + pgdat_size, PAGE_SIZE); | |
a8062231 AK |
205 | bootmap = early_node_mem(nodeid, bootmap_start, end, |
206 | bootmap_pages<<PAGE_SHIFT); | |
207 | if (bootmap == NULL) { | |
208 | if (nodedata_phys < start || nodedata_phys >= end) | |
209 | free_bootmem((unsigned long)node_data[nodeid],pgdat_size); | |
210 | node_data[nodeid] = NULL; | |
211 | return; | |
212 | } | |
213 | bootmap_start = __pa(bootmap); | |
1da177e4 LT |
214 | Dprintk("bootmap start %lu pages %lu\n", bootmap_start, bootmap_pages); |
215 | ||
216 | bootmap_size = init_bootmem_node(NODE_DATA(nodeid), | |
217 | bootmap_start >> PAGE_SHIFT, | |
218 | start_pfn, end_pfn); | |
219 | ||
5cb248ab | 220 | free_bootmem_with_active_regions(nodeid, end); |
1da177e4 LT |
221 | |
222 | reserve_bootmem_node(NODE_DATA(nodeid), nodedata_phys, pgdat_size); | |
223 | reserve_bootmem_node(NODE_DATA(nodeid), bootmap_start, bootmap_pages<<PAGE_SHIFT); | |
68a3a7fe AK |
224 | #ifdef CONFIG_ACPI_NUMA |
225 | srat_reserve_add_area(nodeid); | |
226 | #endif | |
1da177e4 LT |
227 | node_set_online(nodeid); |
228 | } | |
229 | ||
230 | /* Initialize final allocator for a zone */ | |
231 | void __init setup_node_zones(int nodeid) | |
232 | { | |
267b4801 | 233 | unsigned long start_pfn, end_pfn, memmapsize, limit; |
1da177e4 | 234 | |
a2f1b424 AK |
235 | start_pfn = node_start_pfn(nodeid); |
236 | end_pfn = node_end_pfn(nodeid); | |
1da177e4 | 237 | |
5cb248ab | 238 | Dprintk(KERN_INFO "Setting up memmap for node %d %lx-%lx\n", |
a2f1b424 | 239 | nodeid, start_pfn, end_pfn); |
1da177e4 | 240 | |
267b4801 AK |
241 | /* Try to allocate mem_map at end to not fill up precious <4GB |
242 | memory. */ | |
243 | memmapsize = sizeof(struct page) * (end_pfn-start_pfn); | |
244 | limit = end_pfn << PAGE_SHIFT; | |
3b5fd59f | 245 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
267b4801 AK |
246 | NODE_DATA(nodeid)->node_mem_map = |
247 | __alloc_bootmem_core(NODE_DATA(nodeid)->bdata, | |
248 | memmapsize, SMP_CACHE_BYTES, | |
249 | round_down(limit - memmapsize, PAGE_SIZE), | |
250 | limit); | |
3b5fd59f | 251 | #endif |
1da177e4 LT |
252 | } |
253 | ||
254 | void __init numa_init_array(void) | |
255 | { | |
256 | int rr, i; | |
257 | /* There are unfortunately some poorly designed mainboards around | |
258 | that only connect memory to a single CPU. This breaks the 1:1 cpu->node | |
259 | mapping. To avoid this fill in the mapping for all possible | |
260 | CPUs, as the number of CPUs is not known yet. | |
261 | We round robin the existing nodes. */ | |
85cc5135 | 262 | rr = first_node(node_online_map); |
1da177e4 LT |
263 | for (i = 0; i < NR_CPUS; i++) { |
264 | if (cpu_to_node[i] != NUMA_NO_NODE) | |
265 | continue; | |
69d81fcd | 266 | numa_set_node(i, rr); |
1da177e4 LT |
267 | rr = next_node(rr, node_online_map); |
268 | if (rr == MAX_NUMNODES) | |
269 | rr = first_node(node_online_map); | |
1da177e4 LT |
270 | } |
271 | ||
1da177e4 LT |
272 | } |
273 | ||
274 | #ifdef CONFIG_NUMA_EMU | |
53fee04f | 275 | /* Numa emulation */ |
8b8ca80e DR |
276 | #define E820_ADDR_HOLE_SIZE(start, end) \ |
277 | (e820_hole_size((start) >> PAGE_SHIFT, (end) >> PAGE_SHIFT) << \ | |
278 | PAGE_SHIFT) | |
279 | char *cmdline __initdata; | |
1da177e4 | 280 | |
53fee04f | 281 | /* |
8b8ca80e DR |
282 | * Setups up nid to range from addr to addr + size. If the end boundary is |
283 | * greater than max_addr, then max_addr is used instead. The return value is 0 | |
284 | * if there is additional memory left for allocation past addr and -1 otherwise. | |
285 | * addr is adjusted to be at the end of the node. | |
53fee04f | 286 | */ |
8b8ca80e DR |
287 | static int __init setup_node_range(int nid, struct bootnode *nodes, u64 *addr, |
288 | u64 size, u64 max_addr) | |
53fee04f | 289 | { |
8b8ca80e DR |
290 | int ret = 0; |
291 | nodes[nid].start = *addr; | |
292 | *addr += size; | |
293 | if (*addr >= max_addr) { | |
294 | *addr = max_addr; | |
295 | ret = -1; | |
296 | } | |
297 | nodes[nid].end = *addr; | |
e3f1caee | 298 | node_set(nid, node_possible_map); |
8b8ca80e DR |
299 | printk(KERN_INFO "Faking node %d at %016Lx-%016Lx (%LuMB)\n", nid, |
300 | nodes[nid].start, nodes[nid].end, | |
301 | (nodes[nid].end - nodes[nid].start) >> 20); | |
302 | return ret; | |
53fee04f RS |
303 | } |
304 | ||
8b8ca80e DR |
305 | /* |
306 | * Splits num_nodes nodes up equally starting at node_start. The return value | |
307 | * is the number of nodes split up and addr is adjusted to be at the end of the | |
308 | * last node allocated. | |
309 | */ | |
310 | static int __init split_nodes_equally(struct bootnode *nodes, u64 *addr, | |
311 | u64 max_addr, int node_start, | |
312 | int num_nodes) | |
1da177e4 | 313 | { |
8b8ca80e DR |
314 | unsigned int big; |
315 | u64 size; | |
316 | int i; | |
53fee04f | 317 | |
8b8ca80e DR |
318 | if (num_nodes <= 0) |
319 | return -1; | |
320 | if (num_nodes > MAX_NUMNODES) | |
321 | num_nodes = MAX_NUMNODES; | |
322 | size = (max_addr - *addr - E820_ADDR_HOLE_SIZE(*addr, max_addr)) / | |
323 | num_nodes; | |
53fee04f | 324 | /* |
8b8ca80e DR |
325 | * Calculate the number of big nodes that can be allocated as a result |
326 | * of consolidating the leftovers. | |
53fee04f | 327 | */ |
8b8ca80e DR |
328 | big = ((size & ~FAKE_NODE_MIN_HASH_MASK) * num_nodes) / |
329 | FAKE_NODE_MIN_SIZE; | |
330 | ||
331 | /* Round down to nearest FAKE_NODE_MIN_SIZE. */ | |
332 | size &= FAKE_NODE_MIN_HASH_MASK; | |
333 | if (!size) { | |
334 | printk(KERN_ERR "Not enough memory for each node. " | |
335 | "NUMA emulation disabled.\n"); | |
336 | return -1; | |
53fee04f | 337 | } |
8b8ca80e DR |
338 | |
339 | for (i = node_start; i < num_nodes + node_start; i++) { | |
340 | u64 end = *addr + size; | |
53fee04f RS |
341 | if (i < big) |
342 | end += FAKE_NODE_MIN_SIZE; | |
343 | /* | |
8b8ca80e DR |
344 | * The final node can have the remaining system RAM. Other |
345 | * nodes receive roughly the same amount of available pages. | |
53fee04f | 346 | */ |
8b8ca80e DR |
347 | if (i == num_nodes + node_start - 1) |
348 | end = max_addr; | |
349 | else | |
350 | while (end - *addr - E820_ADDR_HOLE_SIZE(*addr, end) < | |
351 | size) { | |
352 | end += FAKE_NODE_MIN_SIZE; | |
353 | if (end > max_addr) { | |
354 | end = max_addr; | |
355 | break; | |
356 | } | |
357 | } | |
358 | if (setup_node_range(i, nodes, addr, end - *addr, max_addr) < 0) | |
359 | break; | |
360 | } | |
361 | return i - node_start + 1; | |
362 | } | |
363 | ||
382591d5 DR |
364 | /* |
365 | * Splits the remaining system RAM into chunks of size. The remaining memory is | |
366 | * always assigned to a final node and can be asymmetric. Returns the number of | |
367 | * nodes split. | |
368 | */ | |
369 | static int __init split_nodes_by_size(struct bootnode *nodes, u64 *addr, | |
370 | u64 max_addr, int node_start, u64 size) | |
371 | { | |
372 | int i = node_start; | |
373 | size = (size << 20) & FAKE_NODE_MIN_HASH_MASK; | |
374 | while (!setup_node_range(i++, nodes, addr, size, max_addr)) | |
375 | ; | |
376 | return i - node_start; | |
377 | } | |
378 | ||
8b8ca80e DR |
379 | /* |
380 | * Sets up the system RAM area from start_pfn to end_pfn according to the | |
381 | * numa=fake command-line option. | |
382 | */ | |
383 | static int __init numa_emulation(unsigned long start_pfn, unsigned long end_pfn) | |
384 | { | |
385 | struct bootnode nodes[MAX_NUMNODES]; | |
386 | u64 addr = start_pfn << PAGE_SHIFT; | |
387 | u64 max_addr = end_pfn << PAGE_SHIFT; | |
8b8ca80e | 388 | int num_nodes = 0; |
382591d5 DR |
389 | int coeff_flag; |
390 | int coeff = -1; | |
391 | int num = 0; | |
8b8ca80e DR |
392 | u64 size; |
393 | int i; | |
394 | ||
395 | memset(&nodes, 0, sizeof(nodes)); | |
396 | /* | |
397 | * If the numa=fake command-line is just a single number N, split the | |
398 | * system RAM into N fake nodes. | |
399 | */ | |
400 | if (!strchr(cmdline, '*') && !strchr(cmdline, ',')) { | |
401 | num_nodes = split_nodes_equally(nodes, &addr, max_addr, 0, | |
402 | simple_strtol(cmdline, NULL, 0)); | |
403 | if (num_nodes < 0) | |
404 | return num_nodes; | |
405 | goto out; | |
406 | } | |
407 | ||
408 | /* Parse the command line. */ | |
382591d5 | 409 | for (coeff_flag = 0; ; cmdline++) { |
8b8ca80e DR |
410 | if (*cmdline && isdigit(*cmdline)) { |
411 | num = num * 10 + *cmdline - '0'; | |
412 | continue; | |
53fee04f | 413 | } |
382591d5 DR |
414 | if (*cmdline == '*') { |
415 | if (num > 0) | |
416 | coeff = num; | |
417 | coeff_flag = 1; | |
418 | } | |
8b8ca80e | 419 | if (!*cmdline || *cmdline == ',') { |
382591d5 DR |
420 | if (!coeff_flag) |
421 | coeff = 1; | |
8b8ca80e DR |
422 | /* |
423 | * Round down to the nearest FAKE_NODE_MIN_SIZE. | |
424 | * Command-line coefficients are in megabytes. | |
425 | */ | |
426 | size = ((u64)num << 20) & FAKE_NODE_MIN_HASH_MASK; | |
382591d5 | 427 | if (size) |
8b8ca80e DR |
428 | for (i = 0; i < coeff; i++, num_nodes++) |
429 | if (setup_node_range(num_nodes, nodes, | |
430 | &addr, size, max_addr) < 0) | |
431 | goto done; | |
382591d5 DR |
432 | if (!*cmdline) |
433 | break; | |
434 | coeff_flag = 0; | |
435 | coeff = -1; | |
53fee04f | 436 | } |
8b8ca80e DR |
437 | num = 0; |
438 | } | |
439 | done: | |
440 | if (!num_nodes) | |
441 | return -1; | |
14694d73 | 442 | /* Fill remainder of system RAM, if appropriate. */ |
8b8ca80e | 443 | if (addr < max_addr) { |
382591d5 DR |
444 | if (coeff_flag && coeff < 0) { |
445 | /* Split remaining nodes into num-sized chunks */ | |
446 | num_nodes += split_nodes_by_size(nodes, &addr, max_addr, | |
447 | num_nodes, num); | |
448 | goto out; | |
449 | } | |
14694d73 DR |
450 | switch (*(cmdline - 1)) { |
451 | case '*': | |
452 | /* Split remaining nodes into coeff chunks */ | |
453 | if (coeff <= 0) | |
454 | break; | |
455 | num_nodes += split_nodes_equally(nodes, &addr, max_addr, | |
456 | num_nodes, coeff); | |
457 | break; | |
458 | case ',': | |
459 | /* Do not allocate remaining system RAM */ | |
460 | break; | |
461 | default: | |
462 | /* Give one final node */ | |
463 | setup_node_range(num_nodes, nodes, &addr, | |
464 | max_addr - addr, max_addr); | |
465 | num_nodes++; | |
466 | } | |
8b8ca80e DR |
467 | } |
468 | out: | |
469 | memnode_shift = compute_hash_shift(nodes, num_nodes); | |
470 | if (memnode_shift < 0) { | |
471 | memnode_shift = 0; | |
472 | printk(KERN_ERR "No NUMA hash function found. NUMA emulation " | |
473 | "disabled.\n"); | |
474 | return -1; | |
475 | } | |
476 | ||
477 | /* | |
478 | * We need to vacate all active ranges that may have been registered by | |
479 | * SRAT. | |
480 | */ | |
481 | remove_all_active_ranges(); | |
e3f1caee | 482 | for_each_node_mask(i, node_possible_map) { |
5cb248ab MG |
483 | e820_register_active_regions(i, nodes[i].start >> PAGE_SHIFT, |
484 | nodes[i].end >> PAGE_SHIFT); | |
1da177e4 | 485 | setup_node_bootmem(i, nodes[i].start, nodes[i].end); |
5cb248ab | 486 | } |
1da177e4 LT |
487 | numa_init_array(); |
488 | return 0; | |
489 | } | |
8b8ca80e DR |
490 | #undef E820_ADDR_HOLE_SIZE |
491 | #endif /* CONFIG_NUMA_EMU */ | |
1da177e4 LT |
492 | |
493 | void __init numa_initmem_init(unsigned long start_pfn, unsigned long end_pfn) | |
494 | { | |
495 | int i; | |
496 | ||
e3f1caee SS |
497 | nodes_clear(node_possible_map); |
498 | ||
1da177e4 | 499 | #ifdef CONFIG_NUMA_EMU |
8b8ca80e | 500 | if (cmdline && !numa_emulation(start_pfn, end_pfn)) |
1da177e4 | 501 | return; |
e3f1caee | 502 | nodes_clear(node_possible_map); |
1da177e4 LT |
503 | #endif |
504 | ||
505 | #ifdef CONFIG_ACPI_NUMA | |
506 | if (!numa_off && !acpi_scan_nodes(start_pfn << PAGE_SHIFT, | |
507 | end_pfn << PAGE_SHIFT)) | |
508 | return; | |
e3f1caee | 509 | nodes_clear(node_possible_map); |
1da177e4 LT |
510 | #endif |
511 | ||
512 | #ifdef CONFIG_K8_NUMA | |
513 | if (!numa_off && !k8_scan_nodes(start_pfn<<PAGE_SHIFT, end_pfn<<PAGE_SHIFT)) | |
514 | return; | |
e3f1caee | 515 | nodes_clear(node_possible_map); |
1da177e4 LT |
516 | #endif |
517 | printk(KERN_INFO "%s\n", | |
518 | numa_off ? "NUMA turned off" : "No NUMA configuration found"); | |
519 | ||
520 | printk(KERN_INFO "Faking a node at %016lx-%016lx\n", | |
521 | start_pfn << PAGE_SHIFT, | |
522 | end_pfn << PAGE_SHIFT); | |
523 | /* setup dummy node covering all memory */ | |
524 | memnode_shift = 63; | |
076422d2 | 525 | memnodemap = memnode.embedded_map; |
1da177e4 LT |
526 | memnodemap[0] = 0; |
527 | nodes_clear(node_online_map); | |
528 | node_set_online(0); | |
e3f1caee | 529 | node_set(0, node_possible_map); |
1da177e4 | 530 | for (i = 0; i < NR_CPUS; i++) |
69d81fcd | 531 | numa_set_node(i, 0); |
1da177e4 | 532 | node_to_cpumask[0] = cpumask_of_cpu(0); |
5cb248ab | 533 | e820_register_active_regions(0, start_pfn, end_pfn); |
1da177e4 LT |
534 | setup_node_bootmem(0, start_pfn << PAGE_SHIFT, end_pfn << PAGE_SHIFT); |
535 | } | |
536 | ||
e6982c67 | 537 | __cpuinit void numa_add_cpu(int cpu) |
1da177e4 | 538 | { |
e6a045a5 | 539 | set_bit(cpu, &node_to_cpumask[cpu_to_node(cpu)]); |
1da177e4 LT |
540 | } |
541 | ||
69d81fcd AK |
542 | void __cpuinit numa_set_node(int cpu, int node) |
543 | { | |
df79efde | 544 | cpu_pda(cpu)->nodenumber = node; |
69d81fcd AK |
545 | cpu_to_node[cpu] = node; |
546 | } | |
547 | ||
1da177e4 LT |
548 | unsigned long __init numa_free_all_bootmem(void) |
549 | { | |
550 | int i; | |
551 | unsigned long pages = 0; | |
552 | for_each_online_node(i) { | |
553 | pages += free_all_bootmem_node(NODE_DATA(i)); | |
554 | } | |
555 | return pages; | |
556 | } | |
557 | ||
558 | void __init paging_init(void) | |
559 | { | |
560 | int i; | |
6391af17 MG |
561 | unsigned long max_zone_pfns[MAX_NR_ZONES]; |
562 | memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); | |
563 | max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN; | |
564 | max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN; | |
565 | max_zone_pfns[ZONE_NORMAL] = end_pfn; | |
d3ee871e | 566 | |
f0a5a58a BP |
567 | sparse_memory_present_with_active_regions(MAX_NUMNODES); |
568 | sparse_init(); | |
d3ee871e | 569 | |
1da177e4 LT |
570 | for_each_online_node(i) { |
571 | setup_node_zones(i); | |
572 | } | |
5cb248ab MG |
573 | |
574 | free_area_init_nodes(max_zone_pfns); | |
1da177e4 LT |
575 | } |
576 | ||
2c8c0e6b | 577 | static __init int numa_setup(char *opt) |
1da177e4 | 578 | { |
2c8c0e6b AK |
579 | if (!opt) |
580 | return -EINVAL; | |
1da177e4 LT |
581 | if (!strncmp(opt,"off",3)) |
582 | numa_off = 1; | |
583 | #ifdef CONFIG_NUMA_EMU | |
8b8ca80e DR |
584 | if (!strncmp(opt, "fake=", 5)) |
585 | cmdline = opt + 5; | |
1da177e4 LT |
586 | #endif |
587 | #ifdef CONFIG_ACPI_NUMA | |
588 | if (!strncmp(opt,"noacpi",6)) | |
589 | acpi_numa = -1; | |
68a3a7fe AK |
590 | if (!strncmp(opt,"hotadd=", 7)) |
591 | hotadd_percent = simple_strtoul(opt+7, NULL, 10); | |
1da177e4 | 592 | #endif |
2c8c0e6b | 593 | return 0; |
1da177e4 LT |
594 | } |
595 | ||
2c8c0e6b AK |
596 | early_param("numa", numa_setup); |
597 | ||
05b3cbd8 RT |
598 | /* |
599 | * Setup early cpu_to_node. | |
600 | * | |
601 | * Populate cpu_to_node[] only if x86_cpu_to_apicid[], | |
602 | * and apicid_to_node[] tables have valid entries for a CPU. | |
603 | * This means we skip cpu_to_node[] initialisation for NUMA | |
604 | * emulation and faking node case (when running a kernel compiled | |
605 | * for NUMA on a non NUMA box), which is OK as cpu_to_node[] | |
606 | * is already initialized in a round robin manner at numa_init_array, | |
607 | * prior to this call, and this initialization is good enough | |
608 | * for the fake NUMA cases. | |
609 | */ | |
610 | void __init init_cpu_to_node(void) | |
611 | { | |
612 | int i; | |
613 | for (i = 0; i < NR_CPUS; i++) { | |
614 | u8 apicid = x86_cpu_to_apicid[i]; | |
615 | if (apicid == BAD_APICID) | |
616 | continue; | |
617 | if (apicid_to_node[apicid] == NUMA_NO_NODE) | |
618 | continue; | |
d1db4ec8 | 619 | numa_set_node(i,apicid_to_node[apicid]); |
05b3cbd8 RT |
620 | } |
621 | } | |
622 | ||
1da177e4 LT |
623 | EXPORT_SYMBOL(cpu_to_node); |
624 | EXPORT_SYMBOL(node_to_cpumask); | |
dcf36bfa | 625 | EXPORT_SYMBOL(memnode); |
1da177e4 | 626 | EXPORT_SYMBOL(node_data); |
cf050132 AK |
627 | |
628 | #ifdef CONFIG_DISCONTIGMEM | |
629 | /* | |
630 | * Functions to convert PFNs from/to per node page addresses. | |
631 | * These are out of line because they are quite big. | |
632 | * They could be all tuned by pre caching more state. | |
633 | * Should do that. | |
634 | */ | |
635 | ||
cf050132 AK |
636 | int pfn_valid(unsigned long pfn) |
637 | { | |
638 | unsigned nid; | |
639 | if (pfn >= num_physpages) | |
640 | return 0; | |
641 | nid = pfn_to_nid(pfn); | |
642 | if (nid == 0xff) | |
643 | return 0; | |
644 | return pfn >= node_start_pfn(nid) && (pfn) < node_end_pfn(nid); | |
645 | } | |
646 | EXPORT_SYMBOL(pfn_valid); | |
647 | #endif |