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