]>
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 | |
3f098c26 AK |
33 | unsigned char cpu_to_node[NR_CPUS] __read_mostly = { |
34 | [0 ... NR_CPUS-1] = NUMA_NO_NODE | |
0b07e984 | 35 | }; |
e3cfe529 TG |
36 | EXPORT_SYMBOL(cpu_to_node); |
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 | |
3f098c26 | 42 | cpumask_t node_to_cpumask[MAX_NUMNODES] __read_mostly; |
e3cfe529 | 43 | EXPORT_SYMBOL(node_to_cpumask); |
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 | #ifdef CONFIG_SPARSEMEM |
157 | int early_pfn_to_nid(unsigned long pfn) | |
158 | { | |
159 | return phys_to_nid(pfn << PAGE_SHIFT); | |
160 | } | |
161 | #endif | |
162 | ||
e3cfe529 TG |
163 | static void * __init early_node_mem(int nodeid, unsigned long start, |
164 | unsigned long end, unsigned long size) | |
a8062231 AK |
165 | { |
166 | unsigned long mem = find_e820_area(start, end, size); | |
167 | void *ptr; | |
e3cfe529 | 168 | |
a8062231 AK |
169 | if (mem != -1L) |
170 | return __va(mem); | |
171 | ptr = __alloc_bootmem_nopanic(size, | |
172 | SMP_CACHE_BYTES, __pa(MAX_DMA_ADDRESS)); | |
83e83d54 | 173 | if (ptr == NULL) { |
a8062231 | 174 | printk(KERN_ERR "Cannot find %lu bytes in node %d\n", |
e3cfe529 | 175 | size, nodeid); |
a8062231 AK |
176 | return NULL; |
177 | } | |
178 | return ptr; | |
179 | } | |
180 | ||
1da177e4 | 181 | /* Initialize bootmem allocator for a node */ |
e3cfe529 TG |
182 | void __init setup_node_bootmem(int nodeid, unsigned long start, |
183 | unsigned long end) | |
184 | { | |
185 | unsigned long start_pfn, end_pfn, bootmap_pages, bootmap_size; | |
186 | unsigned long bootmap_start, nodedata_phys; | |
a8062231 | 187 | void *bootmap; |
1da177e4 LT |
188 | const int pgdat_size = round_up(sizeof(pg_data_t), PAGE_SIZE); |
189 | ||
e3cfe529 | 190 | start = round_up(start, ZONE_ALIGN); |
1da177e4 | 191 | |
e3cfe529 TG |
192 | printk(KERN_INFO "Bootmem setup node %d %016lx-%016lx\n", nodeid, |
193 | start, end); | |
1da177e4 LT |
194 | |
195 | start_pfn = start >> PAGE_SHIFT; | |
196 | end_pfn = end >> PAGE_SHIFT; | |
197 | ||
a8062231 AK |
198 | node_data[nodeid] = early_node_mem(nodeid, start, end, pgdat_size); |
199 | if (node_data[nodeid] == NULL) | |
200 | return; | |
201 | nodedata_phys = __pa(node_data[nodeid]); | |
1da177e4 | 202 | |
1da177e4 LT |
203 | memset(NODE_DATA(nodeid), 0, sizeof(pg_data_t)); |
204 | NODE_DATA(nodeid)->bdata = &plat_node_bdata[nodeid]; | |
205 | NODE_DATA(nodeid)->node_start_pfn = start_pfn; | |
206 | NODE_DATA(nodeid)->node_spanned_pages = end_pfn - start_pfn; | |
207 | ||
208 | /* Find a place for the bootmem map */ | |
e3cfe529 | 209 | bootmap_pages = bootmem_bootmap_pages(end_pfn - start_pfn); |
1da177e4 | 210 | bootmap_start = round_up(nodedata_phys + pgdat_size, PAGE_SIZE); |
a8062231 AK |
211 | bootmap = early_node_mem(nodeid, bootmap_start, end, |
212 | bootmap_pages<<PAGE_SHIFT); | |
213 | if (bootmap == NULL) { | |
214 | if (nodedata_phys < start || nodedata_phys >= end) | |
e3cfe529 TG |
215 | free_bootmem((unsigned long)node_data[nodeid], |
216 | pgdat_size); | |
a8062231 AK |
217 | node_data[nodeid] = NULL; |
218 | return; | |
219 | } | |
220 | bootmap_start = __pa(bootmap); | |
e3cfe529 TG |
221 | Dprintk("bootmap start %lu pages %lu\n", bootmap_start, bootmap_pages); |
222 | ||
1da177e4 | 223 | bootmap_size = init_bootmem_node(NODE_DATA(nodeid), |
e3cfe529 TG |
224 | bootmap_start >> PAGE_SHIFT, |
225 | start_pfn, end_pfn); | |
1da177e4 | 226 | |
5cb248ab | 227 | free_bootmem_with_active_regions(nodeid, end); |
1da177e4 | 228 | |
e3cfe529 TG |
229 | reserve_bootmem_node(NODE_DATA(nodeid), nodedata_phys, pgdat_size); |
230 | reserve_bootmem_node(NODE_DATA(nodeid), bootmap_start, | |
231 | bootmap_pages<<PAGE_SHIFT); | |
68a3a7fe AK |
232 | #ifdef CONFIG_ACPI_NUMA |
233 | srat_reserve_add_area(nodeid); | |
234 | #endif | |
1da177e4 | 235 | node_set_online(nodeid); |
e3cfe529 | 236 | } |
1da177e4 LT |
237 | |
238 | /* Initialize final allocator for a zone */ | |
239 | void __init setup_node_zones(int nodeid) | |
e3cfe529 | 240 | { |
267b4801 | 241 | unsigned long start_pfn, end_pfn, memmapsize, limit; |
1da177e4 | 242 | |
e3cfe529 TG |
243 | start_pfn = node_start_pfn(nodeid); |
244 | end_pfn = node_end_pfn(nodeid); | |
1da177e4 | 245 | |
5cb248ab | 246 | Dprintk(KERN_INFO "Setting up memmap for node %d %lx-%lx\n", |
a2f1b424 | 247 | nodeid, start_pfn, end_pfn); |
1da177e4 | 248 | |
e3cfe529 TG |
249 | /* |
250 | * Try to allocate mem_map at end to not fill up precious <4GB | |
251 | * memory. | |
252 | */ | |
267b4801 AK |
253 | memmapsize = sizeof(struct page) * (end_pfn-start_pfn); |
254 | limit = end_pfn << PAGE_SHIFT; | |
3b5fd59f | 255 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
e3cfe529 TG |
256 | NODE_DATA(nodeid)->node_mem_map = |
257 | __alloc_bootmem_core(NODE_DATA(nodeid)->bdata, | |
258 | memmapsize, SMP_CACHE_BYTES, | |
259 | round_down(limit - memmapsize, PAGE_SIZE), | |
260 | limit); | |
3b5fd59f | 261 | #endif |
e3cfe529 | 262 | } |
1da177e4 | 263 | |
e3cfe529 TG |
264 | /* |
265 | * There are unfortunately some poorly designed mainboards around that | |
266 | * only connect memory to a single CPU. This breaks the 1:1 cpu->node | |
267 | * mapping. To avoid this fill in the mapping for all possible CPUs, | |
268 | * as the number of CPUs is not known yet. We round robin the existing | |
269 | * nodes. | |
270 | */ | |
1da177e4 LT |
271 | void __init numa_init_array(void) |
272 | { | |
273 | int rr, i; | |
e3cfe529 | 274 | |
85cc5135 | 275 | rr = first_node(node_online_map); |
1da177e4 | 276 | for (i = 0; i < NR_CPUS; i++) { |
98c9e27a | 277 | if (cpu_to_node(i) != NUMA_NO_NODE) |
1da177e4 | 278 | continue; |
e3cfe529 | 279 | numa_set_node(i, rr); |
1da177e4 LT |
280 | rr = next_node(rr, node_online_map); |
281 | if (rr == MAX_NUMNODES) | |
282 | rr = first_node(node_online_map); | |
1da177e4 | 283 | } |
1da177e4 LT |
284 | } |
285 | ||
286 | #ifdef CONFIG_NUMA_EMU | |
53fee04f | 287 | /* Numa emulation */ |
8b8ca80e | 288 | char *cmdline __initdata; |
1da177e4 | 289 | |
53fee04f | 290 | /* |
e3cfe529 TG |
291 | * Setups up nid to range from addr to addr + size. If the end |
292 | * boundary is greater than max_addr, then max_addr is used instead. | |
293 | * The return value is 0 if there is additional memory left for | |
294 | * allocation past addr and -1 otherwise. addr is adjusted to be at | |
295 | * the end of the node. | |
53fee04f | 296 | */ |
8b8ca80e DR |
297 | static int __init setup_node_range(int nid, struct bootnode *nodes, u64 *addr, |
298 | u64 size, u64 max_addr) | |
53fee04f | 299 | { |
8b8ca80e | 300 | int ret = 0; |
e3cfe529 | 301 | |
8b8ca80e DR |
302 | nodes[nid].start = *addr; |
303 | *addr += size; | |
304 | if (*addr >= max_addr) { | |
305 | *addr = max_addr; | |
306 | ret = -1; | |
307 | } | |
308 | nodes[nid].end = *addr; | |
e3f1caee | 309 | node_set(nid, node_possible_map); |
8b8ca80e DR |
310 | printk(KERN_INFO "Faking node %d at %016Lx-%016Lx (%LuMB)\n", nid, |
311 | nodes[nid].start, nodes[nid].end, | |
312 | (nodes[nid].end - nodes[nid].start) >> 20); | |
313 | return ret; | |
53fee04f RS |
314 | } |
315 | ||
8b8ca80e DR |
316 | /* |
317 | * Splits num_nodes nodes up equally starting at node_start. The return value | |
318 | * is the number of nodes split up and addr is adjusted to be at the end of the | |
319 | * last node allocated. | |
320 | */ | |
321 | static int __init split_nodes_equally(struct bootnode *nodes, u64 *addr, | |
322 | u64 max_addr, int node_start, | |
323 | int num_nodes) | |
1da177e4 | 324 | { |
8b8ca80e DR |
325 | unsigned int big; |
326 | u64 size; | |
327 | int i; | |
53fee04f | 328 | |
8b8ca80e DR |
329 | if (num_nodes <= 0) |
330 | return -1; | |
331 | if (num_nodes > MAX_NUMNODES) | |
332 | num_nodes = MAX_NUMNODES; | |
a7e96629 | 333 | size = (max_addr - *addr - e820_hole_size(*addr, max_addr)) / |
8b8ca80e | 334 | num_nodes; |
53fee04f | 335 | /* |
8b8ca80e DR |
336 | * Calculate the number of big nodes that can be allocated as a result |
337 | * of consolidating the leftovers. | |
53fee04f | 338 | */ |
8b8ca80e DR |
339 | big = ((size & ~FAKE_NODE_MIN_HASH_MASK) * num_nodes) / |
340 | FAKE_NODE_MIN_SIZE; | |
341 | ||
342 | /* Round down to nearest FAKE_NODE_MIN_SIZE. */ | |
343 | size &= FAKE_NODE_MIN_HASH_MASK; | |
344 | if (!size) { | |
345 | printk(KERN_ERR "Not enough memory for each node. " | |
346 | "NUMA emulation disabled.\n"); | |
347 | return -1; | |
53fee04f | 348 | } |
8b8ca80e DR |
349 | |
350 | for (i = node_start; i < num_nodes + node_start; i++) { | |
351 | u64 end = *addr + size; | |
e3cfe529 | 352 | |
53fee04f RS |
353 | if (i < big) |
354 | end += FAKE_NODE_MIN_SIZE; | |
355 | /* | |
8b8ca80e DR |
356 | * The final node can have the remaining system RAM. Other |
357 | * nodes receive roughly the same amount of available pages. | |
53fee04f | 358 | */ |
8b8ca80e DR |
359 | if (i == num_nodes + node_start - 1) |
360 | end = max_addr; | |
361 | else | |
a7e96629 | 362 | while (end - *addr - e820_hole_size(*addr, end) < |
8b8ca80e DR |
363 | size) { |
364 | end += FAKE_NODE_MIN_SIZE; | |
365 | if (end > max_addr) { | |
366 | end = max_addr; | |
367 | break; | |
368 | } | |
369 | } | |
370 | if (setup_node_range(i, nodes, addr, end - *addr, max_addr) < 0) | |
371 | break; | |
372 | } | |
373 | return i - node_start + 1; | |
374 | } | |
375 | ||
382591d5 DR |
376 | /* |
377 | * Splits the remaining system RAM into chunks of size. The remaining memory is | |
378 | * always assigned to a final node and can be asymmetric. Returns the number of | |
379 | * nodes split. | |
380 | */ | |
381 | static int __init split_nodes_by_size(struct bootnode *nodes, u64 *addr, | |
382 | u64 max_addr, int node_start, u64 size) | |
383 | { | |
384 | int i = node_start; | |
385 | size = (size << 20) & FAKE_NODE_MIN_HASH_MASK; | |
386 | while (!setup_node_range(i++, nodes, addr, size, max_addr)) | |
387 | ; | |
388 | return i - node_start; | |
389 | } | |
390 | ||
8b8ca80e DR |
391 | /* |
392 | * Sets up the system RAM area from start_pfn to end_pfn according to the | |
393 | * numa=fake command-line option. | |
394 | */ | |
395 | static int __init numa_emulation(unsigned long start_pfn, unsigned long end_pfn) | |
396 | { | |
397 | struct bootnode nodes[MAX_NUMNODES]; | |
e3cfe529 | 398 | u64 size, addr = start_pfn << PAGE_SHIFT; |
8b8ca80e | 399 | u64 max_addr = end_pfn << PAGE_SHIFT; |
e3cfe529 | 400 | int num_nodes = 0, num = 0, coeff_flag, coeff = -1, i; |
8b8ca80e DR |
401 | |
402 | memset(&nodes, 0, sizeof(nodes)); | |
403 | /* | |
404 | * If the numa=fake command-line is just a single number N, split the | |
405 | * system RAM into N fake nodes. | |
406 | */ | |
407 | if (!strchr(cmdline, '*') && !strchr(cmdline, ',')) { | |
e3cfe529 TG |
408 | long n = simple_strtol(cmdline, NULL, 0); |
409 | ||
410 | num_nodes = split_nodes_equally(nodes, &addr, max_addr, 0, n); | |
8b8ca80e DR |
411 | if (num_nodes < 0) |
412 | return num_nodes; | |
413 | goto out; | |
414 | } | |
415 | ||
416 | /* Parse the command line. */ | |
382591d5 | 417 | for (coeff_flag = 0; ; cmdline++) { |
8b8ca80e DR |
418 | if (*cmdline && isdigit(*cmdline)) { |
419 | num = num * 10 + *cmdline - '0'; | |
420 | continue; | |
53fee04f | 421 | } |
382591d5 DR |
422 | if (*cmdline == '*') { |
423 | if (num > 0) | |
424 | coeff = num; | |
425 | coeff_flag = 1; | |
426 | } | |
8b8ca80e | 427 | if (!*cmdline || *cmdline == ',') { |
382591d5 DR |
428 | if (!coeff_flag) |
429 | coeff = 1; | |
8b8ca80e DR |
430 | /* |
431 | * Round down to the nearest FAKE_NODE_MIN_SIZE. | |
432 | * Command-line coefficients are in megabytes. | |
433 | */ | |
434 | size = ((u64)num << 20) & FAKE_NODE_MIN_HASH_MASK; | |
382591d5 | 435 | if (size) |
8b8ca80e DR |
436 | for (i = 0; i < coeff; i++, num_nodes++) |
437 | if (setup_node_range(num_nodes, nodes, | |
438 | &addr, size, max_addr) < 0) | |
439 | goto done; | |
382591d5 DR |
440 | if (!*cmdline) |
441 | break; | |
442 | coeff_flag = 0; | |
443 | coeff = -1; | |
53fee04f | 444 | } |
8b8ca80e DR |
445 | num = 0; |
446 | } | |
447 | done: | |
448 | if (!num_nodes) | |
449 | return -1; | |
14694d73 | 450 | /* Fill remainder of system RAM, if appropriate. */ |
8b8ca80e | 451 | if (addr < max_addr) { |
382591d5 DR |
452 | if (coeff_flag && coeff < 0) { |
453 | /* Split remaining nodes into num-sized chunks */ | |
454 | num_nodes += split_nodes_by_size(nodes, &addr, max_addr, | |
455 | num_nodes, num); | |
456 | goto out; | |
457 | } | |
14694d73 DR |
458 | switch (*(cmdline - 1)) { |
459 | case '*': | |
460 | /* Split remaining nodes into coeff chunks */ | |
461 | if (coeff <= 0) | |
462 | break; | |
463 | num_nodes += split_nodes_equally(nodes, &addr, max_addr, | |
464 | num_nodes, coeff); | |
465 | break; | |
466 | case ',': | |
467 | /* Do not allocate remaining system RAM */ | |
468 | break; | |
469 | default: | |
470 | /* Give one final node */ | |
471 | setup_node_range(num_nodes, nodes, &addr, | |
472 | max_addr - addr, max_addr); | |
473 | num_nodes++; | |
474 | } | |
8b8ca80e DR |
475 | } |
476 | out: | |
477 | memnode_shift = compute_hash_shift(nodes, num_nodes); | |
478 | if (memnode_shift < 0) { | |
479 | memnode_shift = 0; | |
480 | printk(KERN_ERR "No NUMA hash function found. NUMA emulation " | |
481 | "disabled.\n"); | |
482 | return -1; | |
483 | } | |
484 | ||
485 | /* | |
486 | * We need to vacate all active ranges that may have been registered by | |
1c05f093 DR |
487 | * SRAT and set acpi_numa to -1 so that srat_disabled() always returns |
488 | * true. NUMA emulation has succeeded so we will not scan ACPI nodes. | |
8b8ca80e DR |
489 | */ |
490 | remove_all_active_ranges(); | |
1c05f093 DR |
491 | #ifdef CONFIG_ACPI_NUMA |
492 | acpi_numa = -1; | |
493 | #endif | |
e3f1caee | 494 | for_each_node_mask(i, node_possible_map) { |
5cb248ab MG |
495 | e820_register_active_regions(i, nodes[i].start >> PAGE_SHIFT, |
496 | nodes[i].end >> PAGE_SHIFT); | |
e3cfe529 | 497 | setup_node_bootmem(i, nodes[i].start, nodes[i].end); |
5cb248ab | 498 | } |
3484d798 | 499 | acpi_fake_nodes(nodes, num_nodes); |
e3cfe529 TG |
500 | numa_init_array(); |
501 | return 0; | |
1da177e4 | 502 | } |
8b8ca80e | 503 | #endif /* CONFIG_NUMA_EMU */ |
1da177e4 LT |
504 | |
505 | void __init numa_initmem_init(unsigned long start_pfn, unsigned long end_pfn) | |
e3cfe529 | 506 | { |
1da177e4 LT |
507 | int i; |
508 | ||
e3f1caee SS |
509 | nodes_clear(node_possible_map); |
510 | ||
1da177e4 | 511 | #ifdef CONFIG_NUMA_EMU |
8b8ca80e | 512 | if (cmdline && !numa_emulation(start_pfn, end_pfn)) |
e3cfe529 | 513 | return; |
e3f1caee | 514 | nodes_clear(node_possible_map); |
1da177e4 LT |
515 | #endif |
516 | ||
517 | #ifdef CONFIG_ACPI_NUMA | |
518 | if (!numa_off && !acpi_scan_nodes(start_pfn << PAGE_SHIFT, | |
519 | end_pfn << PAGE_SHIFT)) | |
e3cfe529 | 520 | return; |
e3f1caee | 521 | nodes_clear(node_possible_map); |
1da177e4 LT |
522 | #endif |
523 | ||
524 | #ifdef CONFIG_K8_NUMA | |
e3cfe529 TG |
525 | if (!numa_off && !k8_scan_nodes(start_pfn<<PAGE_SHIFT, |
526 | end_pfn<<PAGE_SHIFT)) | |
1da177e4 | 527 | return; |
e3f1caee | 528 | nodes_clear(node_possible_map); |
1da177e4 LT |
529 | #endif |
530 | printk(KERN_INFO "%s\n", | |
531 | numa_off ? "NUMA turned off" : "No NUMA configuration found"); | |
532 | ||
e3cfe529 | 533 | printk(KERN_INFO "Faking a node at %016lx-%016lx\n", |
1da177e4 | 534 | start_pfn << PAGE_SHIFT, |
e3cfe529 TG |
535 | end_pfn << PAGE_SHIFT); |
536 | /* setup dummy node covering all memory */ | |
537 | memnode_shift = 63; | |
076422d2 | 538 | memnodemap = memnode.embedded_map; |
1da177e4 LT |
539 | memnodemap[0] = 0; |
540 | nodes_clear(node_online_map); | |
541 | node_set_online(0); | |
e3f1caee | 542 | node_set(0, node_possible_map); |
1da177e4 | 543 | for (i = 0; i < NR_CPUS; i++) |
69d81fcd | 544 | numa_set_node(i, 0); |
1da177e4 | 545 | node_to_cpumask[0] = cpumask_of_cpu(0); |
5cb248ab | 546 | e820_register_active_regions(0, start_pfn, end_pfn); |
1da177e4 LT |
547 | setup_node_bootmem(0, start_pfn << PAGE_SHIFT, end_pfn << PAGE_SHIFT); |
548 | } | |
549 | ||
e6982c67 | 550 | __cpuinit void numa_add_cpu(int cpu) |
1da177e4 | 551 | { |
e6a045a5 | 552 | set_bit(cpu, &node_to_cpumask[cpu_to_node(cpu)]); |
e3cfe529 | 553 | } |
1da177e4 | 554 | |
69d81fcd AK |
555 | void __cpuinit numa_set_node(int cpu, int node) |
556 | { | |
df79efde | 557 | cpu_pda(cpu)->nodenumber = node; |
98c9e27a | 558 | cpu_to_node(cpu) = node; |
69d81fcd AK |
559 | } |
560 | ||
e3cfe529 TG |
561 | unsigned long __init numa_free_all_bootmem(void) |
562 | { | |
1da177e4 | 563 | unsigned long pages = 0; |
e3cfe529 TG |
564 | int i; |
565 | ||
566 | for_each_online_node(i) | |
1da177e4 | 567 | pages += free_all_bootmem_node(NODE_DATA(i)); |
e3cfe529 | 568 | |
1da177e4 | 569 | return pages; |
e3cfe529 | 570 | } |
1da177e4 LT |
571 | |
572 | void __init paging_init(void) | |
e3cfe529 | 573 | { |
6391af17 | 574 | unsigned long max_zone_pfns[MAX_NR_ZONES]; |
e3cfe529 TG |
575 | int i; |
576 | ||
6391af17 MG |
577 | memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); |
578 | max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN; | |
579 | max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN; | |
580 | max_zone_pfns[ZONE_NORMAL] = end_pfn; | |
d3ee871e | 581 | |
f0a5a58a BP |
582 | sparse_memory_present_with_active_regions(MAX_NUMNODES); |
583 | sparse_init(); | |
d3ee871e | 584 | |
e3cfe529 TG |
585 | for_each_online_node(i) |
586 | setup_node_zones(i); | |
5cb248ab MG |
587 | |
588 | free_area_init_nodes(max_zone_pfns); | |
e3cfe529 | 589 | } |
1da177e4 | 590 | |
2c8c0e6b | 591 | static __init int numa_setup(char *opt) |
e3cfe529 | 592 | { |
2c8c0e6b AK |
593 | if (!opt) |
594 | return -EINVAL; | |
e3cfe529 | 595 | if (!strncmp(opt, "off", 3)) |
1da177e4 LT |
596 | numa_off = 1; |
597 | #ifdef CONFIG_NUMA_EMU | |
8b8ca80e DR |
598 | if (!strncmp(opt, "fake=", 5)) |
599 | cmdline = opt + 5; | |
1da177e4 LT |
600 | #endif |
601 | #ifdef CONFIG_ACPI_NUMA | |
e3cfe529 TG |
602 | if (!strncmp(opt, "noacpi", 6)) |
603 | acpi_numa = -1; | |
604 | if (!strncmp(opt, "hotadd=", 7)) | |
68a3a7fe | 605 | hotadd_percent = simple_strtoul(opt+7, NULL, 10); |
1da177e4 | 606 | #endif |
2c8c0e6b | 607 | return 0; |
e3cfe529 | 608 | } |
2c8c0e6b AK |
609 | early_param("numa", numa_setup); |
610 | ||
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. | |
622 | */ | |
623 | void __init init_cpu_to_node(void) | |
624 | { | |
625 | int i; | |
e3cfe529 TG |
626 | |
627 | for (i = 0; i < NR_CPUS; i++) { | |
71fff5e6 | 628 | u8 apicid = x86_cpu_to_apicid_init[i]; |
e3cfe529 | 629 | |
05b3cbd8 RT |
630 | if (apicid == BAD_APICID) |
631 | continue; | |
632 | if (apicid_to_node[apicid] == NUMA_NO_NODE) | |
633 | continue; | |
e3cfe529 | 634 | numa_set_node(i, apicid_to_node[apicid]); |
05b3cbd8 RT |
635 | } |
636 | } | |
637 | ||
cf050132 AK |
638 | #ifdef CONFIG_DISCONTIGMEM |
639 | /* | |
640 | * Functions to convert PFNs from/to per node page addresses. | |
641 | * These are out of line because they are quite big. | |
642 | * They could be all tuned by pre caching more state. | |
643 | * Should do that. | |
644 | */ | |
645 | ||
cf050132 AK |
646 | int pfn_valid(unsigned long pfn) |
647 | { | |
648 | unsigned nid; | |
649 | if (pfn >= num_physpages) | |
650 | return 0; | |
651 | nid = pfn_to_nid(pfn); | |
652 | if (nid == 0xff) | |
653 | return 0; | |
654 | return pfn >= node_start_pfn(nid) && (pfn) < node_end_pfn(nid); | |
655 | } | |
656 | EXPORT_SYMBOL(pfn_valid); | |
657 | #endif |