]>
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 | int x86_cpu_to_node_map_init[NR_CPUS] = { |
3f098c26 | 35 | [0 ... NR_CPUS-1] = NUMA_NO_NODE |
0b07e984 | 36 | }; |
df3825c5 | 37 | void *x86_cpu_to_node_map_early_ptr; |
43238382 | 38 | DEFINE_PER_CPU(int, x86_cpu_to_node_map) = NUMA_NO_NODE; |
df3825c5 | 39 | EXPORT_PER_CPU_SYMBOL(x86_cpu_to_node_map); |
c49a4955 | 40 | EXPORT_SYMBOL(x86_cpu_to_node_map_early_ptr); |
e3cfe529 | 41 | |
43238382 | 42 | s16 apicid_to_node[MAX_LOCAL_APIC] __cpuinitdata = { |
e3cfe529 | 43 | [0 ... MAX_LOCAL_APIC-1] = NUMA_NO_NODE |
3f098c26 | 44 | }; |
e3cfe529 | 45 | |
7462894a TG |
46 | cpumask_t node_to_cpumask_map[MAX_NUMNODES] __read_mostly; |
47 | EXPORT_SYMBOL(node_to_cpumask_map); | |
1da177e4 LT |
48 | |
49 | int numa_off __initdata; | |
076422d2 AS |
50 | unsigned long __initdata nodemap_addr; |
51 | unsigned long __initdata nodemap_size; | |
1da177e4 | 52 | |
529a3404 ED |
53 | /* |
54 | * Given a shift value, try to populate memnodemap[] | |
55 | * Returns : | |
56 | * 1 if OK | |
57 | * 0 if memnodmap[] too small (of shift too small) | |
58 | * -1 if node overlap or lost ram (shift too big) | |
59 | */ | |
e3cfe529 TG |
60 | static int __init populate_memnodemap(const struct bootnode *nodes, |
61 | int numnodes, int shift) | |
1da177e4 | 62 | { |
529a3404 | 63 | unsigned long addr, end; |
e3cfe529 | 64 | int i, res = -1; |
b684664f | 65 | |
43238382 | 66 | memset(memnodemap, 0xff, sizeof(s16)*memnodemapsize); |
b684664f | 67 | for (i = 0; i < numnodes; i++) { |
529a3404 ED |
68 | addr = nodes[i].start; |
69 | end = nodes[i].end; | |
70 | if (addr >= end) | |
b684664f | 71 | continue; |
076422d2 | 72 | if ((end >> shift) >= memnodemapsize) |
529a3404 ED |
73 | return 0; |
74 | do { | |
43238382 | 75 | if (memnodemap[addr >> shift] != NUMA_NO_NODE) |
b684664f | 76 | return -1; |
b684664f | 77 | memnodemap[addr >> shift] = i; |
076422d2 | 78 | addr += (1UL << shift); |
529a3404 ED |
79 | } while (addr < end); |
80 | res = 1; | |
e3cfe529 | 81 | } |
529a3404 ED |
82 | return res; |
83 | } | |
84 | ||
076422d2 AS |
85 | static int __init allocate_cachealigned_memnodemap(void) |
86 | { | |
24a5da73 | 87 | unsigned long addr; |
076422d2 AS |
88 | |
89 | memnodemap = memnode.embedded_map; | |
316390b0 | 90 | if (memnodemapsize <= ARRAY_SIZE(memnode.embedded_map)) |
076422d2 | 91 | return 0; |
076422d2 | 92 | |
24a5da73 YL |
93 | addr = 0x8000; |
94 | nodemap_size = round_up(sizeof(s16) * memnodemapsize, L1_CACHE_BYTES); | |
95 | nodemap_addr = find_e820_area(addr, end_pfn<<PAGE_SHIFT, | |
96 | nodemap_size, L1_CACHE_BYTES); | |
076422d2 AS |
97 | if (nodemap_addr == -1UL) { |
98 | printk(KERN_ERR | |
99 | "NUMA: Unable to allocate Memory to Node hash map\n"); | |
100 | nodemap_addr = nodemap_size = 0; | |
101 | return -1; | |
102 | } | |
24a5da73 | 103 | memnodemap = phys_to_virt(nodemap_addr); |
25eff8d4 | 104 | reserve_early(nodemap_addr, nodemap_addr + nodemap_size, "MEMNODEMAP"); |
076422d2 AS |
105 | |
106 | printk(KERN_DEBUG "NUMA: Allocated memnodemap from %lx - %lx\n", | |
107 | nodemap_addr, nodemap_addr + nodemap_size); | |
108 | return 0; | |
109 | } | |
110 | ||
111 | /* | |
112 | * The LSB of all start and end addresses in the node map is the value of the | |
113 | * maximum possible shift. | |
114 | */ | |
e3cfe529 TG |
115 | static int __init extract_lsb_from_nodes(const struct bootnode *nodes, |
116 | int numnodes) | |
529a3404 | 117 | { |
54413927 | 118 | int i, nodes_used = 0; |
076422d2 AS |
119 | unsigned long start, end; |
120 | unsigned long bitfield = 0, memtop = 0; | |
121 | ||
122 | for (i = 0; i < numnodes; i++) { | |
123 | start = nodes[i].start; | |
124 | end = nodes[i].end; | |
125 | if (start >= end) | |
126 | continue; | |
54413927 AS |
127 | bitfield |= start; |
128 | nodes_used++; | |
076422d2 AS |
129 | if (end > memtop) |
130 | memtop = end; | |
131 | } | |
54413927 AS |
132 | if (nodes_used <= 1) |
133 | i = 63; | |
134 | else | |
135 | i = find_first_bit(&bitfield, sizeof(unsigned long)*8); | |
076422d2 AS |
136 | memnodemapsize = (memtop >> i)+1; |
137 | return i; | |
138 | } | |
529a3404 | 139 | |
076422d2 AS |
140 | int __init compute_hash_shift(struct bootnode *nodes, int numnodes) |
141 | { | |
142 | int shift; | |
529a3404 | 143 | |
076422d2 AS |
144 | shift = extract_lsb_from_nodes(nodes, numnodes); |
145 | if (allocate_cachealigned_memnodemap()) | |
146 | return -1; | |
6b050f80 | 147 | printk(KERN_DEBUG "NUMA: Using %d for the hash shift.\n", |
529a3404 ED |
148 | shift); |
149 | ||
150 | if (populate_memnodemap(nodes, numnodes, shift) != 1) { | |
e3cfe529 TG |
151 | printk(KERN_INFO "Your memory is not aligned you need to " |
152 | "rebuild your kernel with a bigger NODEMAPSIZE " | |
153 | "shift=%d\n", shift); | |
529a3404 ED |
154 | return -1; |
155 | } | |
b684664f | 156 | return shift; |
1da177e4 LT |
157 | } |
158 | ||
bbfceef4 MT |
159 | int early_pfn_to_nid(unsigned long pfn) |
160 | { | |
161 | return phys_to_nid(pfn << PAGE_SHIFT); | |
162 | } | |
bbfceef4 | 163 | |
e3cfe529 | 164 | static void * __init early_node_mem(int nodeid, unsigned long start, |
24a5da73 YL |
165 | unsigned long end, unsigned long size, |
166 | unsigned long align) | |
a8062231 | 167 | { |
24a5da73 | 168 | unsigned long mem = find_e820_area(start, end, size, align); |
a8062231 | 169 | void *ptr; |
e3cfe529 | 170 | |
9347e0b0 | 171 | if (mem != -1L) |
a8062231 | 172 | return __va(mem); |
9347e0b0 | 173 | |
24a5da73 | 174 | ptr = __alloc_bootmem_nopanic(size, align, __pa(MAX_DMA_ADDRESS)); |
83e83d54 | 175 | if (ptr == NULL) { |
a8062231 | 176 | printk(KERN_ERR "Cannot find %lu bytes in node %d\n", |
e3cfe529 | 177 | size, nodeid); |
a8062231 AK |
178 | return NULL; |
179 | } | |
180 | return ptr; | |
181 | } | |
182 | ||
1da177e4 | 183 | /* Initialize bootmem allocator for a node */ |
e3cfe529 TG |
184 | void __init setup_node_bootmem(int nodeid, unsigned long start, |
185 | unsigned long end) | |
186 | { | |
187 | unsigned long start_pfn, end_pfn, bootmap_pages, bootmap_size; | |
188 | unsigned long bootmap_start, nodedata_phys; | |
a8062231 | 189 | void *bootmap; |
1da177e4 LT |
190 | const int pgdat_size = round_up(sizeof(pg_data_t), PAGE_SIZE); |
191 | ||
e3cfe529 | 192 | start = round_up(start, ZONE_ALIGN); |
1da177e4 | 193 | |
e3cfe529 TG |
194 | printk(KERN_INFO "Bootmem setup node %d %016lx-%016lx\n", nodeid, |
195 | start, end); | |
1da177e4 LT |
196 | |
197 | start_pfn = start >> PAGE_SHIFT; | |
198 | end_pfn = end >> PAGE_SHIFT; | |
199 | ||
24a5da73 YL |
200 | node_data[nodeid] = early_node_mem(nodeid, start, end, pgdat_size, |
201 | SMP_CACHE_BYTES); | |
a8062231 AK |
202 | if (node_data[nodeid] == NULL) |
203 | return; | |
204 | nodedata_phys = __pa(node_data[nodeid]); | |
6118f76f YL |
205 | printk(KERN_INFO " NODE_DATA [%016lx - %016lx]\n", nodedata_phys, |
206 | nodedata_phys + pgdat_size - 1); | |
1da177e4 | 207 | |
1da177e4 LT |
208 | memset(NODE_DATA(nodeid), 0, sizeof(pg_data_t)); |
209 | NODE_DATA(nodeid)->bdata = &plat_node_bdata[nodeid]; | |
210 | NODE_DATA(nodeid)->node_start_pfn = start_pfn; | |
211 | NODE_DATA(nodeid)->node_spanned_pages = end_pfn - start_pfn; | |
212 | ||
213 | /* Find a place for the bootmem map */ | |
e3cfe529 | 214 | bootmap_pages = bootmem_bootmap_pages(end_pfn - start_pfn); |
1da177e4 | 215 | bootmap_start = round_up(nodedata_phys + pgdat_size, PAGE_SIZE); |
24a5da73 YL |
216 | /* |
217 | * SMP_CAHCE_BYTES could be enough, but init_bootmem_node like | |
218 | * to use that to align to PAGE_SIZE | |
219 | */ | |
a8062231 | 220 | bootmap = early_node_mem(nodeid, bootmap_start, end, |
24a5da73 | 221 | bootmap_pages<<PAGE_SHIFT, PAGE_SIZE); |
a8062231 AK |
222 | if (bootmap == NULL) { |
223 | if (nodedata_phys < start || nodedata_phys >= end) | |
e3cfe529 TG |
224 | free_bootmem((unsigned long)node_data[nodeid], |
225 | pgdat_size); | |
a8062231 AK |
226 | node_data[nodeid] = NULL; |
227 | return; | |
228 | } | |
229 | bootmap_start = __pa(bootmap); | |
e3cfe529 | 230 | |
1da177e4 | 231 | bootmap_size = init_bootmem_node(NODE_DATA(nodeid), |
e3cfe529 TG |
232 | bootmap_start >> PAGE_SHIFT, |
233 | start_pfn, end_pfn); | |
1da177e4 | 234 | |
6118f76f YL |
235 | printk(KERN_INFO " bootmap [%016lx - %016lx] pages %lx\n", |
236 | bootmap_start, bootmap_start + bootmap_size - 1, | |
237 | bootmap_pages); | |
238 | ||
5cb248ab | 239 | free_bootmem_with_active_regions(nodeid, end); |
1da177e4 | 240 | |
72a7fe39 BW |
241 | reserve_bootmem_node(NODE_DATA(nodeid), nodedata_phys, pgdat_size, |
242 | BOOTMEM_DEFAULT); | |
e3cfe529 | 243 | reserve_bootmem_node(NODE_DATA(nodeid), bootmap_start, |
72a7fe39 | 244 | bootmap_pages<<PAGE_SHIFT, BOOTMEM_DEFAULT); |
68a3a7fe AK |
245 | #ifdef CONFIG_ACPI_NUMA |
246 | srat_reserve_add_area(nodeid); | |
247 | #endif | |
1da177e4 | 248 | node_set_online(nodeid); |
e3cfe529 | 249 | } |
1da177e4 | 250 | |
e3cfe529 TG |
251 | /* |
252 | * There are unfortunately some poorly designed mainboards around that | |
253 | * only connect memory to a single CPU. This breaks the 1:1 cpu->node | |
254 | * mapping. To avoid this fill in the mapping for all possible CPUs, | |
255 | * as the number of CPUs is not known yet. We round robin the existing | |
256 | * nodes. | |
257 | */ | |
1da177e4 LT |
258 | void __init numa_init_array(void) |
259 | { | |
260 | int rr, i; | |
e3cfe529 | 261 | |
85cc5135 | 262 | rr = first_node(node_online_map); |
1da177e4 | 263 | for (i = 0; i < NR_CPUS; i++) { |
1ce35712 | 264 | if (early_cpu_to_node(i) != NUMA_NO_NODE) |
1da177e4 | 265 | continue; |
e3cfe529 | 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 | 270 | } |
1da177e4 LT |
271 | } |
272 | ||
273 | #ifdef CONFIG_NUMA_EMU | |
53fee04f | 274 | /* Numa emulation */ |
8b8ca80e | 275 | char *cmdline __initdata; |
1da177e4 | 276 | |
53fee04f | 277 | /* |
e3cfe529 TG |
278 | * Setups up nid to range from addr to addr + size. If the end |
279 | * boundary is greater than max_addr, then max_addr is used instead. | |
280 | * The return value is 0 if there is additional memory left for | |
281 | * allocation past addr and -1 otherwise. addr is adjusted to be at | |
282 | * the end of the node. | |
53fee04f | 283 | */ |
8b8ca80e DR |
284 | static int __init setup_node_range(int nid, struct bootnode *nodes, u64 *addr, |
285 | u64 size, u64 max_addr) | |
53fee04f | 286 | { |
8b8ca80e | 287 | int ret = 0; |
e3cfe529 | 288 | |
8b8ca80e DR |
289 | nodes[nid].start = *addr; |
290 | *addr += size; | |
291 | if (*addr >= max_addr) { | |
292 | *addr = max_addr; | |
293 | ret = -1; | |
294 | } | |
295 | nodes[nid].end = *addr; | |
e3f1caee | 296 | node_set(nid, node_possible_map); |
8b8ca80e DR |
297 | printk(KERN_INFO "Faking node %d at %016Lx-%016Lx (%LuMB)\n", nid, |
298 | nodes[nid].start, nodes[nid].end, | |
299 | (nodes[nid].end - nodes[nid].start) >> 20); | |
300 | return ret; | |
53fee04f RS |
301 | } |
302 | ||
8b8ca80e DR |
303 | /* |
304 | * Splits num_nodes nodes up equally starting at node_start. The return value | |
305 | * is the number of nodes split up and addr is adjusted to be at the end of the | |
306 | * last node allocated. | |
307 | */ | |
308 | static int __init split_nodes_equally(struct bootnode *nodes, u64 *addr, | |
309 | u64 max_addr, int node_start, | |
310 | int num_nodes) | |
1da177e4 | 311 | { |
8b8ca80e DR |
312 | unsigned int big; |
313 | u64 size; | |
314 | int i; | |
53fee04f | 315 | |
8b8ca80e DR |
316 | if (num_nodes <= 0) |
317 | return -1; | |
318 | if (num_nodes > MAX_NUMNODES) | |
319 | num_nodes = MAX_NUMNODES; | |
a7e96629 | 320 | size = (max_addr - *addr - e820_hole_size(*addr, max_addr)) / |
8b8ca80e | 321 | num_nodes; |
53fee04f | 322 | /* |
8b8ca80e DR |
323 | * Calculate the number of big nodes that can be allocated as a result |
324 | * of consolidating the leftovers. | |
53fee04f | 325 | */ |
8b8ca80e DR |
326 | big = ((size & ~FAKE_NODE_MIN_HASH_MASK) * num_nodes) / |
327 | FAKE_NODE_MIN_SIZE; | |
328 | ||
329 | /* Round down to nearest FAKE_NODE_MIN_SIZE. */ | |
330 | size &= FAKE_NODE_MIN_HASH_MASK; | |
331 | if (!size) { | |
332 | printk(KERN_ERR "Not enough memory for each node. " | |
333 | "NUMA emulation disabled.\n"); | |
334 | return -1; | |
53fee04f | 335 | } |
8b8ca80e DR |
336 | |
337 | for (i = node_start; i < num_nodes + node_start; i++) { | |
338 | u64 end = *addr + size; | |
e3cfe529 | 339 | |
53fee04f RS |
340 | if (i < big) |
341 | end += FAKE_NODE_MIN_SIZE; | |
342 | /* | |
8b8ca80e DR |
343 | * The final node can have the remaining system RAM. Other |
344 | * nodes receive roughly the same amount of available pages. | |
53fee04f | 345 | */ |
8b8ca80e DR |
346 | if (i == num_nodes + node_start - 1) |
347 | end = max_addr; | |
348 | else | |
a7e96629 | 349 | while (end - *addr - e820_hole_size(*addr, end) < |
8b8ca80e DR |
350 | size) { |
351 | end += FAKE_NODE_MIN_SIZE; | |
352 | if (end > max_addr) { | |
353 | end = max_addr; | |
354 | break; | |
355 | } | |
356 | } | |
357 | if (setup_node_range(i, nodes, addr, end - *addr, max_addr) < 0) | |
358 | break; | |
359 | } | |
360 | return i - node_start + 1; | |
361 | } | |
362 | ||
382591d5 DR |
363 | /* |
364 | * Splits the remaining system RAM into chunks of size. The remaining memory is | |
365 | * always assigned to a final node and can be asymmetric. Returns the number of | |
366 | * nodes split. | |
367 | */ | |
368 | static int __init split_nodes_by_size(struct bootnode *nodes, u64 *addr, | |
369 | u64 max_addr, int node_start, u64 size) | |
370 | { | |
371 | int i = node_start; | |
372 | size = (size << 20) & FAKE_NODE_MIN_HASH_MASK; | |
373 | while (!setup_node_range(i++, nodes, addr, size, max_addr)) | |
374 | ; | |
375 | return i - node_start; | |
376 | } | |
377 | ||
8b8ca80e DR |
378 | /* |
379 | * Sets up the system RAM area from start_pfn to end_pfn according to the | |
380 | * numa=fake command-line option. | |
381 | */ | |
382 | static int __init numa_emulation(unsigned long start_pfn, unsigned long end_pfn) | |
383 | { | |
384 | struct bootnode nodes[MAX_NUMNODES]; | |
e3cfe529 | 385 | u64 size, addr = start_pfn << PAGE_SHIFT; |
8b8ca80e | 386 | u64 max_addr = end_pfn << PAGE_SHIFT; |
e3cfe529 | 387 | int num_nodes = 0, num = 0, coeff_flag, coeff = -1, i; |
8b8ca80e DR |
388 | |
389 | memset(&nodes, 0, sizeof(nodes)); | |
390 | /* | |
391 | * If the numa=fake command-line is just a single number N, split the | |
392 | * system RAM into N fake nodes. | |
393 | */ | |
394 | if (!strchr(cmdline, '*') && !strchr(cmdline, ',')) { | |
e3cfe529 TG |
395 | long n = simple_strtol(cmdline, NULL, 0); |
396 | ||
397 | num_nodes = split_nodes_equally(nodes, &addr, max_addr, 0, n); | |
8b8ca80e DR |
398 | if (num_nodes < 0) |
399 | return num_nodes; | |
400 | goto out; | |
401 | } | |
402 | ||
403 | /* Parse the command line. */ | |
382591d5 | 404 | for (coeff_flag = 0; ; cmdline++) { |
8b8ca80e DR |
405 | if (*cmdline && isdigit(*cmdline)) { |
406 | num = num * 10 + *cmdline - '0'; | |
407 | continue; | |
53fee04f | 408 | } |
382591d5 DR |
409 | if (*cmdline == '*') { |
410 | if (num > 0) | |
411 | coeff = num; | |
412 | coeff_flag = 1; | |
413 | } | |
8b8ca80e | 414 | if (!*cmdline || *cmdline == ',') { |
382591d5 DR |
415 | if (!coeff_flag) |
416 | coeff = 1; | |
8b8ca80e DR |
417 | /* |
418 | * Round down to the nearest FAKE_NODE_MIN_SIZE. | |
419 | * Command-line coefficients are in megabytes. | |
420 | */ | |
421 | size = ((u64)num << 20) & FAKE_NODE_MIN_HASH_MASK; | |
382591d5 | 422 | if (size) |
8b8ca80e DR |
423 | for (i = 0; i < coeff; i++, num_nodes++) |
424 | if (setup_node_range(num_nodes, nodes, | |
425 | &addr, size, max_addr) < 0) | |
426 | goto done; | |
382591d5 DR |
427 | if (!*cmdline) |
428 | break; | |
429 | coeff_flag = 0; | |
430 | coeff = -1; | |
53fee04f | 431 | } |
8b8ca80e DR |
432 | num = 0; |
433 | } | |
434 | done: | |
435 | if (!num_nodes) | |
436 | return -1; | |
14694d73 | 437 | /* Fill remainder of system RAM, if appropriate. */ |
8b8ca80e | 438 | if (addr < max_addr) { |
382591d5 DR |
439 | if (coeff_flag && coeff < 0) { |
440 | /* Split remaining nodes into num-sized chunks */ | |
441 | num_nodes += split_nodes_by_size(nodes, &addr, max_addr, | |
442 | num_nodes, num); | |
443 | goto out; | |
444 | } | |
14694d73 DR |
445 | switch (*(cmdline - 1)) { |
446 | case '*': | |
447 | /* Split remaining nodes into coeff chunks */ | |
448 | if (coeff <= 0) | |
449 | break; | |
450 | num_nodes += split_nodes_equally(nodes, &addr, max_addr, | |
451 | num_nodes, coeff); | |
452 | break; | |
453 | case ',': | |
454 | /* Do not allocate remaining system RAM */ | |
455 | break; | |
456 | default: | |
457 | /* Give one final node */ | |
458 | setup_node_range(num_nodes, nodes, &addr, | |
459 | max_addr - addr, max_addr); | |
460 | num_nodes++; | |
461 | } | |
8b8ca80e DR |
462 | } |
463 | out: | |
464 | memnode_shift = compute_hash_shift(nodes, num_nodes); | |
465 | if (memnode_shift < 0) { | |
466 | memnode_shift = 0; | |
467 | printk(KERN_ERR "No NUMA hash function found. NUMA emulation " | |
468 | "disabled.\n"); | |
469 | return -1; | |
470 | } | |
471 | ||
472 | /* | |
473 | * We need to vacate all active ranges that may have been registered by | |
1c05f093 DR |
474 | * SRAT and set acpi_numa to -1 so that srat_disabled() always returns |
475 | * true. NUMA emulation has succeeded so we will not scan ACPI nodes. | |
8b8ca80e DR |
476 | */ |
477 | remove_all_active_ranges(); | |
1c05f093 DR |
478 | #ifdef CONFIG_ACPI_NUMA |
479 | acpi_numa = -1; | |
480 | #endif | |
e3f1caee | 481 | for_each_node_mask(i, node_possible_map) { |
5cb248ab MG |
482 | e820_register_active_regions(i, nodes[i].start >> PAGE_SHIFT, |
483 | nodes[i].end >> PAGE_SHIFT); | |
e3cfe529 | 484 | setup_node_bootmem(i, nodes[i].start, nodes[i].end); |
5cb248ab | 485 | } |
3484d798 | 486 | acpi_fake_nodes(nodes, num_nodes); |
e3cfe529 TG |
487 | numa_init_array(); |
488 | return 0; | |
1da177e4 | 489 | } |
8b8ca80e | 490 | #endif /* CONFIG_NUMA_EMU */ |
1da177e4 LT |
491 | |
492 | void __init numa_initmem_init(unsigned long start_pfn, unsigned long end_pfn) | |
e3cfe529 | 493 | { |
1da177e4 LT |
494 | int i; |
495 | ||
e3f1caee SS |
496 | nodes_clear(node_possible_map); |
497 | ||
1da177e4 | 498 | #ifdef CONFIG_NUMA_EMU |
8b8ca80e | 499 | if (cmdline && !numa_emulation(start_pfn, end_pfn)) |
e3cfe529 | 500 | return; |
e3f1caee | 501 | nodes_clear(node_possible_map); |
1da177e4 LT |
502 | #endif |
503 | ||
504 | #ifdef CONFIG_ACPI_NUMA | |
505 | if (!numa_off && !acpi_scan_nodes(start_pfn << PAGE_SHIFT, | |
506 | end_pfn << PAGE_SHIFT)) | |
e3cfe529 | 507 | return; |
e3f1caee | 508 | nodes_clear(node_possible_map); |
1da177e4 LT |
509 | #endif |
510 | ||
511 | #ifdef CONFIG_K8_NUMA | |
e3cfe529 TG |
512 | if (!numa_off && !k8_scan_nodes(start_pfn<<PAGE_SHIFT, |
513 | end_pfn<<PAGE_SHIFT)) | |
1da177e4 | 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 | ||
e3cfe529 | 520 | printk(KERN_INFO "Faking a node at %016lx-%016lx\n", |
1da177e4 | 521 | start_pfn << PAGE_SHIFT, |
e3cfe529 TG |
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); |
df3825c5 | 532 | /* cpumask_of_cpu() may not be available during early startup */ |
3cc87e3f | 533 | memset(&node_to_cpumask_map[0], 0, sizeof(node_to_cpumask_map[0])); |
534 | cpu_set(0, node_to_cpumask_map[0]); | |
5cb248ab | 535 | e820_register_active_regions(0, start_pfn, end_pfn); |
1da177e4 LT |
536 | setup_node_bootmem(0, start_pfn << PAGE_SHIFT, end_pfn << PAGE_SHIFT); |
537 | } | |
538 | ||
e6982c67 | 539 | __cpuinit void numa_add_cpu(int cpu) |
1da177e4 | 540 | { |
1ce35712 | 541 | set_bit(cpu, |
542 | (unsigned long *)&node_to_cpumask_map[early_cpu_to_node(cpu)]); | |
e3cfe529 | 543 | } |
1da177e4 | 544 | |
69d81fcd AK |
545 | void __cpuinit numa_set_node(int cpu, int node) |
546 | { | |
43238382 | 547 | int *cpu_to_node_map = x86_cpu_to_node_map_early_ptr; |
df3825c5 | 548 | |
df79efde | 549 | cpu_pda(cpu)->nodenumber = node; |
df3825c5 | 550 | |
551 | if(cpu_to_node_map) | |
552 | cpu_to_node_map[cpu] = node; | |
553 | else if(per_cpu_offset(cpu)) | |
554 | per_cpu(x86_cpu_to_node_map, cpu) = node; | |
555 | else | |
556 | Dprintk(KERN_INFO "Setting node for non-present cpu %d\n", cpu); | |
69d81fcd AK |
557 | } |
558 | ||
e3cfe529 TG |
559 | unsigned long __init numa_free_all_bootmem(void) |
560 | { | |
1da177e4 | 561 | unsigned long pages = 0; |
e3cfe529 TG |
562 | int i; |
563 | ||
564 | for_each_online_node(i) | |
1da177e4 | 565 | pages += free_all_bootmem_node(NODE_DATA(i)); |
e3cfe529 | 566 | |
1da177e4 | 567 | return pages; |
e3cfe529 | 568 | } |
1da177e4 LT |
569 | |
570 | void __init paging_init(void) | |
e3cfe529 | 571 | { |
6391af17 | 572 | unsigned long max_zone_pfns[MAX_NR_ZONES]; |
e3cfe529 | 573 | |
6391af17 MG |
574 | memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); |
575 | max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN; | |
576 | max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN; | |
577 | max_zone_pfns[ZONE_NORMAL] = end_pfn; | |
d3ee871e | 578 | |
f0a5a58a BP |
579 | sparse_memory_present_with_active_regions(MAX_NUMNODES); |
580 | sparse_init(); | |
d3ee871e | 581 | |
5cb248ab | 582 | free_area_init_nodes(max_zone_pfns); |
e3cfe529 | 583 | } |
1da177e4 | 584 | |
2c8c0e6b | 585 | static __init int numa_setup(char *opt) |
e3cfe529 | 586 | { |
2c8c0e6b AK |
587 | if (!opt) |
588 | return -EINVAL; | |
e3cfe529 | 589 | if (!strncmp(opt, "off", 3)) |
1da177e4 LT |
590 | numa_off = 1; |
591 | #ifdef CONFIG_NUMA_EMU | |
8b8ca80e DR |
592 | if (!strncmp(opt, "fake=", 5)) |
593 | cmdline = opt + 5; | |
1da177e4 LT |
594 | #endif |
595 | #ifdef CONFIG_ACPI_NUMA | |
e3cfe529 TG |
596 | if (!strncmp(opt, "noacpi", 6)) |
597 | acpi_numa = -1; | |
598 | if (!strncmp(opt, "hotadd=", 7)) | |
68a3a7fe | 599 | hotadd_percent = simple_strtoul(opt+7, NULL, 10); |
1da177e4 | 600 | #endif |
2c8c0e6b | 601 | return 0; |
e3cfe529 | 602 | } |
2c8c0e6b AK |
603 | early_param("numa", numa_setup); |
604 | ||
05b3cbd8 RT |
605 | /* |
606 | * Setup early cpu_to_node. | |
607 | * | |
608 | * Populate cpu_to_node[] only if x86_cpu_to_apicid[], | |
609 | * and apicid_to_node[] tables have valid entries for a CPU. | |
610 | * This means we skip cpu_to_node[] initialisation for NUMA | |
611 | * emulation and faking node case (when running a kernel compiled | |
612 | * for NUMA on a non NUMA box), which is OK as cpu_to_node[] | |
613 | * is already initialized in a round robin manner at numa_init_array, | |
614 | * prior to this call, and this initialization is good enough | |
615 | * for the fake NUMA cases. | |
616 | */ | |
617 | void __init init_cpu_to_node(void) | |
618 | { | |
619 | int i; | |
e3cfe529 TG |
620 | |
621 | for (i = 0; i < NR_CPUS; i++) { | |
ef97001f | 622 | u16 apicid = x86_cpu_to_apicid_init[i]; |
e3cfe529 | 623 | |
05b3cbd8 RT |
624 | if (apicid == BAD_APICID) |
625 | continue; | |
626 | if (apicid_to_node[apicid] == NUMA_NO_NODE) | |
627 | continue; | |
e3cfe529 | 628 | numa_set_node(i, apicid_to_node[apicid]); |
05b3cbd8 RT |
629 | } |
630 | } | |
631 | ||
cf050132 | 632 |