]>
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> | |
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> | |
c9ff0342 | 20 | #include <asm/k8.h> |
1da177e4 LT |
21 | |
22 | #ifndef Dprintk | |
23 | #define Dprintk(x...) | |
24 | #endif | |
25 | ||
6c231b7b | 26 | struct pglist_data *node_data[MAX_NUMNODES] __read_mostly; |
e3cfe529 TG |
27 | EXPORT_SYMBOL(node_data); |
28 | ||
1da177e4 LT |
29 | bootmem_data_t plat_node_bdata[MAX_NUMNODES]; |
30 | ||
dcf36bfa | 31 | struct memnode memnode; |
1da177e4 | 32 | |
7462894a | 33 | int cpu_to_node_map[NR_CPUS] __read_mostly = { |
3f098c26 | 34 | [0 ... NR_CPUS-1] = NUMA_NO_NODE |
0b07e984 | 35 | }; |
7462894a | 36 | EXPORT_SYMBOL(cpu_to_node_map); |
e3cfe529 | 37 | |
3f098c26 | 38 | unsigned char apicid_to_node[MAX_LOCAL_APIC] __cpuinitdata = { |
e3cfe529 | 39 | [0 ... MAX_LOCAL_APIC-1] = NUMA_NO_NODE |
3f098c26 | 40 | }; |
e3cfe529 | 41 | |
7462894a TG |
42 | cpumask_t node_to_cpumask_map[MAX_NUMNODES] __read_mostly; |
43 | EXPORT_SYMBOL(node_to_cpumask_map); | |
1da177e4 LT |
44 | |
45 | int numa_off __initdata; | |
076422d2 AS |
46 | unsigned long __initdata nodemap_addr; |
47 | unsigned long __initdata nodemap_size; | |
1da177e4 | 48 | |
529a3404 ED |
49 | /* |
50 | * Given a shift value, try to populate memnodemap[] | |
51 | * Returns : | |
52 | * 1 if OK | |
53 | * 0 if memnodmap[] too small (of shift too small) | |
54 | * -1 if node overlap or lost ram (shift too big) | |
55 | */ | |
e3cfe529 TG |
56 | static int __init populate_memnodemap(const struct bootnode *nodes, |
57 | int numnodes, int shift) | |
1da177e4 | 58 | { |
529a3404 | 59 | unsigned long addr, end; |
e3cfe529 | 60 | int i, res = -1; |
b684664f | 61 | |
076422d2 | 62 | memset(memnodemap, 0xff, memnodemapsize); |
b684664f | 63 | for (i = 0; i < numnodes; i++) { |
529a3404 ED |
64 | addr = nodes[i].start; |
65 | end = nodes[i].end; | |
66 | if (addr >= end) | |
b684664f | 67 | continue; |
076422d2 | 68 | if ((end >> shift) >= memnodemapsize) |
529a3404 ED |
69 | return 0; |
70 | do { | |
71 | if (memnodemap[addr >> shift] != 0xff) | |
b684664f | 72 | return -1; |
b684664f | 73 | memnodemap[addr >> shift] = i; |
076422d2 | 74 | addr += (1UL << shift); |
529a3404 ED |
75 | } while (addr < end); |
76 | res = 1; | |
e3cfe529 | 77 | } |
529a3404 ED |
78 | return res; |
79 | } | |
80 | ||
076422d2 AS |
81 | static int __init allocate_cachealigned_memnodemap(void) |
82 | { | |
83 | unsigned long pad, pad_addr; | |
84 | ||
85 | memnodemap = memnode.embedded_map; | |
54413927 | 86 | if (memnodemapsize <= 48) |
076422d2 | 87 | return 0; |
076422d2 AS |
88 | |
89 | pad = L1_CACHE_BYTES - 1; | |
90 | pad_addr = 0x8000; | |
91 | nodemap_size = pad + memnodemapsize; | |
92 | nodemap_addr = find_e820_area(pad_addr, end_pfn<<PAGE_SHIFT, | |
93 | nodemap_size); | |
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 | } | |
100 | pad_addr = (nodemap_addr + pad) & ~pad; | |
101 | memnodemap = phys_to_virt(pad_addr); | |
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 |
112 | static 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 | |
076422d2 AS |
137 | int __init compute_hash_shift(struct bootnode *nodes, int numnodes) |
138 | { | |
139 | int shift; | |
529a3404 | 140 | |
076422d2 AS |
141 | shift = extract_lsb_from_nodes(nodes, numnodes); |
142 | if (allocate_cachealigned_memnodemap()) | |
143 | return -1; | |
6b050f80 | 144 | printk(KERN_DEBUG "NUMA: Using %d for the hash shift.\n", |
529a3404 ED |
145 | shift); |
146 | ||
147 | if (populate_memnodemap(nodes, numnodes, shift) != 1) { | |
e3cfe529 TG |
148 | printk(KERN_INFO "Your memory is not aligned you need to " |
149 | "rebuild your kernel with a bigger NODEMAPSIZE " | |
150 | "shift=%d\n", shift); | |
529a3404 ED |
151 | return -1; |
152 | } | |
b684664f | 153 | return shift; |
1da177e4 LT |
154 | } |
155 | ||
bbfceef4 MT |
156 | int early_pfn_to_nid(unsigned long pfn) |
157 | { | |
158 | return phys_to_nid(pfn << PAGE_SHIFT); | |
159 | } | |
bbfceef4 | 160 | |
e3cfe529 TG |
161 | static void * __init early_node_mem(int nodeid, unsigned long start, |
162 | unsigned long end, unsigned long size) | |
a8062231 AK |
163 | { |
164 | unsigned long mem = find_e820_area(start, end, size); | |
165 | void *ptr; | |
e3cfe529 | 166 | |
a8062231 AK |
167 | if (mem != -1L) |
168 | return __va(mem); | |
169 | ptr = __alloc_bootmem_nopanic(size, | |
170 | SMP_CACHE_BYTES, __pa(MAX_DMA_ADDRESS)); | |
83e83d54 | 171 | if (ptr == NULL) { |
a8062231 | 172 | printk(KERN_ERR "Cannot find %lu bytes in node %d\n", |
e3cfe529 | 173 | size, nodeid); |
a8062231 AK |
174 | return NULL; |
175 | } | |
176 | return ptr; | |
177 | } | |
178 | ||
1da177e4 | 179 | /* Initialize bootmem allocator for a node */ |
e3cfe529 TG |
180 | void __init setup_node_bootmem(int nodeid, unsigned long start, |
181 | unsigned long end) | |
182 | { | |
183 | unsigned long start_pfn, end_pfn, bootmap_pages, bootmap_size; | |
184 | unsigned long bootmap_start, nodedata_phys; | |
a8062231 | 185 | void *bootmap; |
1da177e4 LT |
186 | const int pgdat_size = round_up(sizeof(pg_data_t), PAGE_SIZE); |
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 | ||
a8062231 AK |
196 | node_data[nodeid] = early_node_mem(nodeid, start, end, pgdat_size); |
197 | if (node_data[nodeid] == NULL) | |
198 | return; | |
199 | nodedata_phys = __pa(node_data[nodeid]); | |
1da177e4 | 200 | |
1da177e4 LT |
201 | memset(NODE_DATA(nodeid), 0, sizeof(pg_data_t)); |
202 | NODE_DATA(nodeid)->bdata = &plat_node_bdata[nodeid]; | |
203 | NODE_DATA(nodeid)->node_start_pfn = start_pfn; | |
204 | NODE_DATA(nodeid)->node_spanned_pages = end_pfn - start_pfn; | |
205 | ||
206 | /* Find a place for the bootmem map */ | |
e3cfe529 | 207 | bootmap_pages = bootmem_bootmap_pages(end_pfn - start_pfn); |
1da177e4 | 208 | bootmap_start = round_up(nodedata_phys + pgdat_size, PAGE_SIZE); |
a8062231 AK |
209 | bootmap = early_node_mem(nodeid, bootmap_start, end, |
210 | bootmap_pages<<PAGE_SHIFT); | |
211 | if (bootmap == NULL) { | |
212 | if (nodedata_phys < start || nodedata_phys >= end) | |
e3cfe529 TG |
213 | free_bootmem((unsigned long)node_data[nodeid], |
214 | pgdat_size); | |
a8062231 AK |
215 | node_data[nodeid] = NULL; |
216 | return; | |
217 | } | |
218 | bootmap_start = __pa(bootmap); | |
e3cfe529 TG |
219 | Dprintk("bootmap start %lu pages %lu\n", bootmap_start, bootmap_pages); |
220 | ||
1da177e4 | 221 | bootmap_size = init_bootmem_node(NODE_DATA(nodeid), |
e3cfe529 TG |
222 | bootmap_start >> PAGE_SHIFT, |
223 | start_pfn, end_pfn); | |
1da177e4 | 224 | |
5cb248ab | 225 | free_bootmem_with_active_regions(nodeid, end); |
1da177e4 | 226 | |
e3cfe529 TG |
227 | reserve_bootmem_node(NODE_DATA(nodeid), nodedata_phys, pgdat_size); |
228 | reserve_bootmem_node(NODE_DATA(nodeid), bootmap_start, | |
229 | bootmap_pages<<PAGE_SHIFT); | |
68a3a7fe AK |
230 | #ifdef CONFIG_ACPI_NUMA |
231 | srat_reserve_add_area(nodeid); | |
232 | #endif | |
1da177e4 | 233 | node_set_online(nodeid); |
e3cfe529 | 234 | } |
1da177e4 | 235 | |
a261670a | 236 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
1da177e4 | 237 | /* Initialize final allocator for a zone */ |
a261670a | 238 | static void __init flat_setup_node_zones(int nodeid) |
e3cfe529 | 239 | { |
267b4801 | 240 | unsigned long start_pfn, end_pfn, memmapsize, limit; |
1da177e4 | 241 | |
e3cfe529 TG |
242 | start_pfn = node_start_pfn(nodeid); |
243 | end_pfn = node_end_pfn(nodeid); | |
1da177e4 | 244 | |
5cb248ab | 245 | Dprintk(KERN_INFO "Setting up memmap for node %d %lx-%lx\n", |
a2f1b424 | 246 | nodeid, start_pfn, end_pfn); |
1da177e4 | 247 | |
e3cfe529 TG |
248 | /* |
249 | * Try to allocate mem_map at end to not fill up precious <4GB | |
250 | * memory. | |
251 | */ | |
267b4801 AK |
252 | memmapsize = sizeof(struct page) * (end_pfn-start_pfn); |
253 | limit = end_pfn << PAGE_SHIFT; | |
a261670a | 254 | |
e3cfe529 TG |
255 | NODE_DATA(nodeid)->node_mem_map = |
256 | __alloc_bootmem_core(NODE_DATA(nodeid)->bdata, | |
257 | memmapsize, SMP_CACHE_BYTES, | |
258 | round_down(limit - memmapsize, PAGE_SIZE), | |
259 | limit); | |
e3cfe529 | 260 | } |
a261670a YL |
261 | #else |
262 | #define flat_setup_node_zones(i) do {} while (0) | |
263 | #endif | |
1da177e4 | 264 | |
e3cfe529 TG |
265 | /* |
266 | * There are unfortunately some poorly designed mainboards around that | |
267 | * only connect memory to a single CPU. This breaks the 1:1 cpu->node | |
268 | * mapping. To avoid this fill in the mapping for all possible CPUs, | |
269 | * as the number of CPUs is not known yet. We round robin the existing | |
270 | * nodes. | |
271 | */ | |
1da177e4 LT |
272 | void __init numa_init_array(void) |
273 | { | |
274 | int rr, i; | |
e3cfe529 | 275 | |
85cc5135 | 276 | rr = first_node(node_online_map); |
1da177e4 | 277 | for (i = 0; i < NR_CPUS; i++) { |
98c9e27a | 278 | if (cpu_to_node(i) != NUMA_NO_NODE) |
1da177e4 | 279 | continue; |
e3cfe529 | 280 | numa_set_node(i, rr); |
1da177e4 LT |
281 | rr = next_node(rr, node_online_map); |
282 | if (rr == MAX_NUMNODES) | |
283 | rr = first_node(node_online_map); | |
1da177e4 | 284 | } |
1da177e4 LT |
285 | } |
286 | ||
287 | #ifdef CONFIG_NUMA_EMU | |
53fee04f | 288 | /* Numa emulation */ |
8b8ca80e | 289 | char *cmdline __initdata; |
1da177e4 | 290 | |
53fee04f | 291 | /* |
e3cfe529 TG |
292 | * Setups up nid to range from addr to addr + size. If the end |
293 | * boundary is greater than max_addr, then max_addr is used instead. | |
294 | * The return value is 0 if there is additional memory left for | |
295 | * allocation past addr and -1 otherwise. addr is adjusted to be at | |
296 | * the end of the node. | |
53fee04f | 297 | */ |
8b8ca80e DR |
298 | static int __init setup_node_range(int nid, struct bootnode *nodes, u64 *addr, |
299 | u64 size, u64 max_addr) | |
53fee04f | 300 | { |
8b8ca80e | 301 | int ret = 0; |
e3cfe529 | 302 | |
8b8ca80e DR |
303 | nodes[nid].start = *addr; |
304 | *addr += size; | |
305 | if (*addr >= max_addr) { | |
306 | *addr = max_addr; | |
307 | ret = -1; | |
308 | } | |
309 | nodes[nid].end = *addr; | |
e3f1caee | 310 | node_set(nid, node_possible_map); |
8b8ca80e DR |
311 | printk(KERN_INFO "Faking node %d at %016Lx-%016Lx (%LuMB)\n", nid, |
312 | nodes[nid].start, nodes[nid].end, | |
313 | (nodes[nid].end - nodes[nid].start) >> 20); | |
314 | return ret; | |
53fee04f RS |
315 | } |
316 | ||
8b8ca80e DR |
317 | /* |
318 | * Splits num_nodes nodes up equally starting at node_start. The return value | |
319 | * is the number of nodes split up and addr is adjusted to be at the end of the | |
320 | * last node allocated. | |
321 | */ | |
322 | static int __init split_nodes_equally(struct bootnode *nodes, u64 *addr, | |
323 | u64 max_addr, int node_start, | |
324 | int num_nodes) | |
1da177e4 | 325 | { |
8b8ca80e DR |
326 | unsigned int big; |
327 | u64 size; | |
328 | int i; | |
53fee04f | 329 | |
8b8ca80e DR |
330 | if (num_nodes <= 0) |
331 | return -1; | |
332 | if (num_nodes > MAX_NUMNODES) | |
333 | num_nodes = MAX_NUMNODES; | |
a7e96629 | 334 | size = (max_addr - *addr - e820_hole_size(*addr, max_addr)) / |
8b8ca80e | 335 | num_nodes; |
53fee04f | 336 | /* |
8b8ca80e DR |
337 | * Calculate the number of big nodes that can be allocated as a result |
338 | * of consolidating the leftovers. | |
53fee04f | 339 | */ |
8b8ca80e DR |
340 | big = ((size & ~FAKE_NODE_MIN_HASH_MASK) * num_nodes) / |
341 | FAKE_NODE_MIN_SIZE; | |
342 | ||
343 | /* Round down to nearest FAKE_NODE_MIN_SIZE. */ | |
344 | size &= FAKE_NODE_MIN_HASH_MASK; | |
345 | if (!size) { | |
346 | printk(KERN_ERR "Not enough memory for each node. " | |
347 | "NUMA emulation disabled.\n"); | |
348 | return -1; | |
53fee04f | 349 | } |
8b8ca80e DR |
350 | |
351 | for (i = node_start; i < num_nodes + node_start; i++) { | |
352 | u64 end = *addr + size; | |
e3cfe529 | 353 | |
53fee04f RS |
354 | if (i < big) |
355 | end += FAKE_NODE_MIN_SIZE; | |
356 | /* | |
8b8ca80e DR |
357 | * The final node can have the remaining system RAM. Other |
358 | * nodes receive roughly the same amount of available pages. | |
53fee04f | 359 | */ |
8b8ca80e DR |
360 | if (i == num_nodes + node_start - 1) |
361 | end = max_addr; | |
362 | else | |
a7e96629 | 363 | while (end - *addr - e820_hole_size(*addr, end) < |
8b8ca80e DR |
364 | size) { |
365 | end += FAKE_NODE_MIN_SIZE; | |
366 | if (end > max_addr) { | |
367 | end = max_addr; | |
368 | break; | |
369 | } | |
370 | } | |
371 | if (setup_node_range(i, nodes, addr, end - *addr, max_addr) < 0) | |
372 | break; | |
373 | } | |
374 | return i - node_start + 1; | |
375 | } | |
376 | ||
382591d5 DR |
377 | /* |
378 | * Splits the remaining system RAM into chunks of size. The remaining memory is | |
379 | * always assigned to a final node and can be asymmetric. Returns the number of | |
380 | * nodes split. | |
381 | */ | |
382 | static int __init split_nodes_by_size(struct bootnode *nodes, u64 *addr, | |
383 | u64 max_addr, int node_start, u64 size) | |
384 | { | |
385 | int i = node_start; | |
386 | size = (size << 20) & FAKE_NODE_MIN_HASH_MASK; | |
387 | while (!setup_node_range(i++, nodes, addr, size, max_addr)) | |
388 | ; | |
389 | return i - node_start; | |
390 | } | |
391 | ||
8b8ca80e DR |
392 | /* |
393 | * Sets up the system RAM area from start_pfn to end_pfn according to the | |
394 | * numa=fake command-line option. | |
395 | */ | |
396 | static int __init numa_emulation(unsigned long start_pfn, unsigned long end_pfn) | |
397 | { | |
398 | struct bootnode nodes[MAX_NUMNODES]; | |
e3cfe529 | 399 | u64 size, addr = start_pfn << PAGE_SHIFT; |
8b8ca80e | 400 | u64 max_addr = end_pfn << PAGE_SHIFT; |
e3cfe529 | 401 | int num_nodes = 0, num = 0, coeff_flag, coeff = -1, i; |
8b8ca80e DR |
402 | |
403 | memset(&nodes, 0, sizeof(nodes)); | |
404 | /* | |
405 | * If the numa=fake command-line is just a single number N, split the | |
406 | * system RAM into N fake nodes. | |
407 | */ | |
408 | if (!strchr(cmdline, '*') && !strchr(cmdline, ',')) { | |
e3cfe529 TG |
409 | long n = simple_strtol(cmdline, NULL, 0); |
410 | ||
411 | num_nodes = split_nodes_equally(nodes, &addr, max_addr, 0, n); | |
8b8ca80e DR |
412 | if (num_nodes < 0) |
413 | return num_nodes; | |
414 | goto out; | |
415 | } | |
416 | ||
417 | /* Parse the command line. */ | |
382591d5 | 418 | for (coeff_flag = 0; ; cmdline++) { |
8b8ca80e DR |
419 | if (*cmdline && isdigit(*cmdline)) { |
420 | num = num * 10 + *cmdline - '0'; | |
421 | continue; | |
53fee04f | 422 | } |
382591d5 DR |
423 | if (*cmdline == '*') { |
424 | if (num > 0) | |
425 | coeff = num; | |
426 | coeff_flag = 1; | |
427 | } | |
8b8ca80e | 428 | if (!*cmdline || *cmdline == ',') { |
382591d5 DR |
429 | if (!coeff_flag) |
430 | coeff = 1; | |
8b8ca80e DR |
431 | /* |
432 | * Round down to the nearest FAKE_NODE_MIN_SIZE. | |
433 | * Command-line coefficients are in megabytes. | |
434 | */ | |
435 | size = ((u64)num << 20) & FAKE_NODE_MIN_HASH_MASK; | |
382591d5 | 436 | if (size) |
8b8ca80e DR |
437 | for (i = 0; i < coeff; i++, num_nodes++) |
438 | if (setup_node_range(num_nodes, nodes, | |
439 | &addr, size, max_addr) < 0) | |
440 | goto done; | |
382591d5 DR |
441 | if (!*cmdline) |
442 | break; | |
443 | coeff_flag = 0; | |
444 | coeff = -1; | |
53fee04f | 445 | } |
8b8ca80e DR |
446 | num = 0; |
447 | } | |
448 | done: | |
449 | if (!num_nodes) | |
450 | return -1; | |
14694d73 | 451 | /* Fill remainder of system RAM, if appropriate. */ |
8b8ca80e | 452 | if (addr < max_addr) { |
382591d5 DR |
453 | if (coeff_flag && coeff < 0) { |
454 | /* Split remaining nodes into num-sized chunks */ | |
455 | num_nodes += split_nodes_by_size(nodes, &addr, max_addr, | |
456 | num_nodes, num); | |
457 | goto out; | |
458 | } | |
14694d73 DR |
459 | switch (*(cmdline - 1)) { |
460 | case '*': | |
461 | /* Split remaining nodes into coeff chunks */ | |
462 | if (coeff <= 0) | |
463 | break; | |
464 | num_nodes += split_nodes_equally(nodes, &addr, max_addr, | |
465 | num_nodes, coeff); | |
466 | break; | |
467 | case ',': | |
468 | /* Do not allocate remaining system RAM */ | |
469 | break; | |
470 | default: | |
471 | /* Give one final node */ | |
472 | setup_node_range(num_nodes, nodes, &addr, | |
473 | max_addr - addr, max_addr); | |
474 | num_nodes++; | |
475 | } | |
8b8ca80e DR |
476 | } |
477 | out: | |
478 | memnode_shift = compute_hash_shift(nodes, num_nodes); | |
479 | if (memnode_shift < 0) { | |
480 | memnode_shift = 0; | |
481 | printk(KERN_ERR "No NUMA hash function found. NUMA emulation " | |
482 | "disabled.\n"); | |
483 | return -1; | |
484 | } | |
485 | ||
486 | /* | |
487 | * We need to vacate all active ranges that may have been registered by | |
1c05f093 DR |
488 | * SRAT and set acpi_numa to -1 so that srat_disabled() always returns |
489 | * true. NUMA emulation has succeeded so we will not scan ACPI nodes. | |
8b8ca80e DR |
490 | */ |
491 | remove_all_active_ranges(); | |
1c05f093 DR |
492 | #ifdef CONFIG_ACPI_NUMA |
493 | acpi_numa = -1; | |
494 | #endif | |
e3f1caee | 495 | for_each_node_mask(i, node_possible_map) { |
5cb248ab MG |
496 | e820_register_active_regions(i, nodes[i].start >> PAGE_SHIFT, |
497 | nodes[i].end >> PAGE_SHIFT); | |
e3cfe529 | 498 | setup_node_bootmem(i, nodes[i].start, nodes[i].end); |
5cb248ab | 499 | } |
3484d798 | 500 | acpi_fake_nodes(nodes, num_nodes); |
e3cfe529 TG |
501 | numa_init_array(); |
502 | return 0; | |
1da177e4 | 503 | } |
8b8ca80e | 504 | #endif /* CONFIG_NUMA_EMU */ |
1da177e4 LT |
505 | |
506 | void __init numa_initmem_init(unsigned long start_pfn, unsigned long end_pfn) | |
e3cfe529 | 507 | { |
1da177e4 LT |
508 | int i; |
509 | ||
e3f1caee SS |
510 | nodes_clear(node_possible_map); |
511 | ||
1da177e4 | 512 | #ifdef CONFIG_NUMA_EMU |
8b8ca80e | 513 | if (cmdline && !numa_emulation(start_pfn, end_pfn)) |
e3cfe529 | 514 | return; |
e3f1caee | 515 | nodes_clear(node_possible_map); |
1da177e4 LT |
516 | #endif |
517 | ||
518 | #ifdef CONFIG_ACPI_NUMA | |
519 | if (!numa_off && !acpi_scan_nodes(start_pfn << PAGE_SHIFT, | |
520 | end_pfn << PAGE_SHIFT)) | |
e3cfe529 | 521 | return; |
e3f1caee | 522 | nodes_clear(node_possible_map); |
1da177e4 LT |
523 | #endif |
524 | ||
525 | #ifdef CONFIG_K8_NUMA | |
e3cfe529 TG |
526 | if (!numa_off && !k8_scan_nodes(start_pfn<<PAGE_SHIFT, |
527 | end_pfn<<PAGE_SHIFT)) | |
1da177e4 | 528 | return; |
e3f1caee | 529 | nodes_clear(node_possible_map); |
1da177e4 LT |
530 | #endif |
531 | printk(KERN_INFO "%s\n", | |
532 | numa_off ? "NUMA turned off" : "No NUMA configuration found"); | |
533 | ||
e3cfe529 | 534 | printk(KERN_INFO "Faking a node at %016lx-%016lx\n", |
1da177e4 | 535 | start_pfn << PAGE_SHIFT, |
e3cfe529 TG |
536 | end_pfn << PAGE_SHIFT); |
537 | /* setup dummy node covering all memory */ | |
538 | memnode_shift = 63; | |
076422d2 | 539 | memnodemap = memnode.embedded_map; |
1da177e4 LT |
540 | memnodemap[0] = 0; |
541 | nodes_clear(node_online_map); | |
542 | node_set_online(0); | |
e3f1caee | 543 | node_set(0, node_possible_map); |
1da177e4 | 544 | for (i = 0; i < NR_CPUS; i++) |
69d81fcd | 545 | numa_set_node(i, 0); |
7462894a | 546 | node_to_cpumask_map[0] = cpumask_of_cpu(0); |
5cb248ab | 547 | e820_register_active_regions(0, start_pfn, end_pfn); |
1da177e4 LT |
548 | setup_node_bootmem(0, start_pfn << PAGE_SHIFT, end_pfn << PAGE_SHIFT); |
549 | } | |
550 | ||
e6982c67 | 551 | __cpuinit void numa_add_cpu(int cpu) |
1da177e4 | 552 | { |
5548fecd | 553 | set_bit(cpu, (unsigned long *)&node_to_cpumask_map[cpu_to_node(cpu)]); |
e3cfe529 | 554 | } |
1da177e4 | 555 | |
69d81fcd AK |
556 | void __cpuinit numa_set_node(int cpu, int node) |
557 | { | |
df79efde | 558 | cpu_pda(cpu)->nodenumber = node; |
7462894a | 559 | cpu_to_node_map[cpu] = node; |
69d81fcd AK |
560 | } |
561 | ||
e3cfe529 TG |
562 | unsigned long __init numa_free_all_bootmem(void) |
563 | { | |
1da177e4 | 564 | unsigned long pages = 0; |
e3cfe529 TG |
565 | int i; |
566 | ||
567 | for_each_online_node(i) | |
1da177e4 | 568 | pages += free_all_bootmem_node(NODE_DATA(i)); |
e3cfe529 | 569 | |
1da177e4 | 570 | return pages; |
e3cfe529 | 571 | } |
1da177e4 LT |
572 | |
573 | void __init paging_init(void) | |
e3cfe529 | 574 | { |
6391af17 | 575 | unsigned long max_zone_pfns[MAX_NR_ZONES]; |
e3cfe529 TG |
576 | int i; |
577 | ||
6391af17 MG |
578 | memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); |
579 | max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN; | |
580 | max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN; | |
581 | max_zone_pfns[ZONE_NORMAL] = end_pfn; | |
d3ee871e | 582 | |
f0a5a58a BP |
583 | sparse_memory_present_with_active_regions(MAX_NUMNODES); |
584 | sparse_init(); | |
d3ee871e | 585 | |
e3cfe529 | 586 | for_each_online_node(i) |
a261670a | 587 | flat_setup_node_zones(i); |
5cb248ab MG |
588 | |
589 | free_area_init_nodes(max_zone_pfns); | |
e3cfe529 | 590 | } |
1da177e4 | 591 | |
2c8c0e6b | 592 | static __init int numa_setup(char *opt) |
e3cfe529 | 593 | { |
2c8c0e6b AK |
594 | if (!opt) |
595 | return -EINVAL; | |
e3cfe529 | 596 | if (!strncmp(opt, "off", 3)) |
1da177e4 LT |
597 | numa_off = 1; |
598 | #ifdef CONFIG_NUMA_EMU | |
8b8ca80e DR |
599 | if (!strncmp(opt, "fake=", 5)) |
600 | cmdline = opt + 5; | |
1da177e4 LT |
601 | #endif |
602 | #ifdef CONFIG_ACPI_NUMA | |
e3cfe529 TG |
603 | if (!strncmp(opt, "noacpi", 6)) |
604 | acpi_numa = -1; | |
605 | if (!strncmp(opt, "hotadd=", 7)) | |
68a3a7fe | 606 | hotadd_percent = simple_strtoul(opt+7, NULL, 10); |
1da177e4 | 607 | #endif |
2c8c0e6b | 608 | return 0; |
e3cfe529 | 609 | } |
2c8c0e6b AK |
610 | early_param("numa", numa_setup); |
611 | ||
05b3cbd8 RT |
612 | /* |
613 | * Setup early cpu_to_node. | |
614 | * | |
615 | * Populate cpu_to_node[] only if x86_cpu_to_apicid[], | |
616 | * and apicid_to_node[] tables have valid entries for a CPU. | |
617 | * This means we skip cpu_to_node[] initialisation for NUMA | |
618 | * emulation and faking node case (when running a kernel compiled | |
619 | * for NUMA on a non NUMA box), which is OK as cpu_to_node[] | |
620 | * is already initialized in a round robin manner at numa_init_array, | |
621 | * prior to this call, and this initialization is good enough | |
622 | * for the fake NUMA cases. | |
623 | */ | |
624 | void __init init_cpu_to_node(void) | |
625 | { | |
626 | int i; | |
e3cfe529 TG |
627 | |
628 | for (i = 0; i < NR_CPUS; i++) { | |
ef97001f | 629 | u16 apicid = x86_cpu_to_apicid_init[i]; |
e3cfe529 | 630 | |
05b3cbd8 RT |
631 | if (apicid == BAD_APICID) |
632 | continue; | |
633 | if (apicid_to_node[apicid] == NUMA_NO_NODE) | |
634 | continue; | |
e3cfe529 | 635 | numa_set_node(i, apicid_to_node[apicid]); |
05b3cbd8 RT |
636 | } |
637 | } | |
638 | ||
cf050132 | 639 |