]>
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 | |
b447a468 | 34 | #ifdef CONFIG_SMP |
43238382 | 35 | int x86_cpu_to_node_map_init[NR_CPUS] = { |
3f098c26 | 36 | [0 ... NR_CPUS-1] = NUMA_NO_NODE |
0b07e984 | 37 | }; |
df3825c5 | 38 | void *x86_cpu_to_node_map_early_ptr; |
b447a468 MT |
39 | EXPORT_SYMBOL(x86_cpu_to_node_map_early_ptr); |
40 | #endif | |
43238382 | 41 | DEFINE_PER_CPU(int, x86_cpu_to_node_map) = NUMA_NO_NODE; |
df3825c5 | 42 | EXPORT_PER_CPU_SYMBOL(x86_cpu_to_node_map); |
e3cfe529 | 43 | |
43238382 | 44 | s16 apicid_to_node[MAX_LOCAL_APIC] __cpuinitdata = { |
e3cfe529 | 45 | [0 ... MAX_LOCAL_APIC-1] = NUMA_NO_NODE |
3f098c26 | 46 | }; |
e3cfe529 | 47 | |
7462894a TG |
48 | cpumask_t node_to_cpumask_map[MAX_NUMNODES] __read_mostly; |
49 | EXPORT_SYMBOL(node_to_cpumask_map); | |
1da177e4 LT |
50 | |
51 | int numa_off __initdata; | |
076422d2 AS |
52 | unsigned long __initdata nodemap_addr; |
53 | unsigned long __initdata nodemap_size; | |
1da177e4 | 54 | |
529a3404 ED |
55 | /* |
56 | * Given a shift value, try to populate memnodemap[] | |
57 | * Returns : | |
58 | * 1 if OK | |
59 | * 0 if memnodmap[] too small (of shift too small) | |
60 | * -1 if node overlap or lost ram (shift too big) | |
61 | */ | |
e3cfe529 | 62 | static int __init populate_memnodemap(const struct bootnode *nodes, |
6ec6e0d9 | 63 | int numnodes, int shift, int *nodeids) |
1da177e4 | 64 | { |
529a3404 | 65 | unsigned long addr, end; |
e3cfe529 | 66 | int i, res = -1; |
b684664f | 67 | |
43238382 | 68 | memset(memnodemap, 0xff, sizeof(s16)*memnodemapsize); |
b684664f | 69 | for (i = 0; i < numnodes; i++) { |
529a3404 ED |
70 | addr = nodes[i].start; |
71 | end = nodes[i].end; | |
72 | if (addr >= end) | |
b684664f | 73 | continue; |
076422d2 | 74 | if ((end >> shift) >= memnodemapsize) |
529a3404 ED |
75 | return 0; |
76 | do { | |
43238382 | 77 | if (memnodemap[addr >> shift] != NUMA_NO_NODE) |
b684664f | 78 | return -1; |
6ec6e0d9 SS |
79 | |
80 | if (!nodeids) | |
81 | memnodemap[addr >> shift] = i; | |
82 | else | |
83 | memnodemap[addr >> shift] = nodeids[i]; | |
84 | ||
076422d2 | 85 | addr += (1UL << shift); |
529a3404 ED |
86 | } while (addr < end); |
87 | res = 1; | |
e3cfe529 | 88 | } |
529a3404 ED |
89 | return res; |
90 | } | |
91 | ||
076422d2 AS |
92 | static int __init allocate_cachealigned_memnodemap(void) |
93 | { | |
24a5da73 | 94 | unsigned long addr; |
076422d2 AS |
95 | |
96 | memnodemap = memnode.embedded_map; | |
316390b0 | 97 | if (memnodemapsize <= ARRAY_SIZE(memnode.embedded_map)) |
076422d2 | 98 | return 0; |
076422d2 | 99 | |
24a5da73 YL |
100 | addr = 0x8000; |
101 | nodemap_size = round_up(sizeof(s16) * memnodemapsize, L1_CACHE_BYTES); | |
102 | nodemap_addr = find_e820_area(addr, end_pfn<<PAGE_SHIFT, | |
103 | nodemap_size, L1_CACHE_BYTES); | |
076422d2 AS |
104 | if (nodemap_addr == -1UL) { |
105 | printk(KERN_ERR | |
106 | "NUMA: Unable to allocate Memory to Node hash map\n"); | |
107 | nodemap_addr = nodemap_size = 0; | |
108 | return -1; | |
109 | } | |
24a5da73 | 110 | memnodemap = phys_to_virt(nodemap_addr); |
25eff8d4 | 111 | reserve_early(nodemap_addr, nodemap_addr + nodemap_size, "MEMNODEMAP"); |
076422d2 AS |
112 | |
113 | printk(KERN_DEBUG "NUMA: Allocated memnodemap from %lx - %lx\n", | |
114 | nodemap_addr, nodemap_addr + nodemap_size); | |
115 | return 0; | |
116 | } | |
117 | ||
118 | /* | |
119 | * The LSB of all start and end addresses in the node map is the value of the | |
120 | * maximum possible shift. | |
121 | */ | |
e3cfe529 TG |
122 | static int __init extract_lsb_from_nodes(const struct bootnode *nodes, |
123 | int numnodes) | |
529a3404 | 124 | { |
54413927 | 125 | int i, nodes_used = 0; |
076422d2 AS |
126 | unsigned long start, end; |
127 | unsigned long bitfield = 0, memtop = 0; | |
128 | ||
129 | for (i = 0; i < numnodes; i++) { | |
130 | start = nodes[i].start; | |
131 | end = nodes[i].end; | |
132 | if (start >= end) | |
133 | continue; | |
54413927 AS |
134 | bitfield |= start; |
135 | nodes_used++; | |
076422d2 AS |
136 | if (end > memtop) |
137 | memtop = end; | |
138 | } | |
54413927 AS |
139 | if (nodes_used <= 1) |
140 | i = 63; | |
141 | else | |
142 | i = find_first_bit(&bitfield, sizeof(unsigned long)*8); | |
076422d2 AS |
143 | memnodemapsize = (memtop >> i)+1; |
144 | return i; | |
145 | } | |
529a3404 | 146 | |
6ec6e0d9 SS |
147 | int __init compute_hash_shift(struct bootnode *nodes, int numnodes, |
148 | int *nodeids) | |
076422d2 AS |
149 | { |
150 | int shift; | |
529a3404 | 151 | |
076422d2 AS |
152 | shift = extract_lsb_from_nodes(nodes, numnodes); |
153 | if (allocate_cachealigned_memnodemap()) | |
154 | return -1; | |
6b050f80 | 155 | printk(KERN_DEBUG "NUMA: Using %d for the hash shift.\n", |
529a3404 ED |
156 | shift); |
157 | ||
6ec6e0d9 | 158 | if (populate_memnodemap(nodes, numnodes, shift, nodeids) != 1) { |
e3cfe529 TG |
159 | printk(KERN_INFO "Your memory is not aligned you need to " |
160 | "rebuild your kernel with a bigger NODEMAPSIZE " | |
161 | "shift=%d\n", shift); | |
529a3404 ED |
162 | return -1; |
163 | } | |
b684664f | 164 | return shift; |
1da177e4 LT |
165 | } |
166 | ||
bbfceef4 MT |
167 | int early_pfn_to_nid(unsigned long pfn) |
168 | { | |
169 | return phys_to_nid(pfn << PAGE_SHIFT); | |
170 | } | |
bbfceef4 | 171 | |
e3cfe529 | 172 | static void * __init early_node_mem(int nodeid, unsigned long start, |
24a5da73 YL |
173 | unsigned long end, unsigned long size, |
174 | unsigned long align) | |
a8062231 | 175 | { |
24a5da73 | 176 | unsigned long mem = find_e820_area(start, end, size, align); |
a8062231 | 177 | void *ptr; |
e3cfe529 | 178 | |
9347e0b0 | 179 | if (mem != -1L) |
a8062231 | 180 | return __va(mem); |
9347e0b0 | 181 | |
24a5da73 | 182 | ptr = __alloc_bootmem_nopanic(size, align, __pa(MAX_DMA_ADDRESS)); |
83e83d54 | 183 | if (ptr == NULL) { |
a8062231 | 184 | printk(KERN_ERR "Cannot find %lu bytes in node %d\n", |
e3cfe529 | 185 | size, nodeid); |
a8062231 AK |
186 | return NULL; |
187 | } | |
188 | return ptr; | |
189 | } | |
190 | ||
1da177e4 | 191 | /* Initialize bootmem allocator for a node */ |
e3cfe529 TG |
192 | void __init setup_node_bootmem(int nodeid, unsigned long start, |
193 | unsigned long end) | |
194 | { | |
195 | unsigned long start_pfn, end_pfn, bootmap_pages, bootmap_size; | |
196 | unsigned long bootmap_start, nodedata_phys; | |
a8062231 | 197 | void *bootmap; |
1da177e4 | 198 | const int pgdat_size = round_up(sizeof(pg_data_t), PAGE_SIZE); |
1a27fc0a | 199 | int nid; |
1da177e4 | 200 | |
e3cfe529 | 201 | start = round_up(start, ZONE_ALIGN); |
1da177e4 | 202 | |
e3cfe529 TG |
203 | printk(KERN_INFO "Bootmem setup node %d %016lx-%016lx\n", nodeid, |
204 | start, end); | |
1da177e4 LT |
205 | |
206 | start_pfn = start >> PAGE_SHIFT; | |
207 | end_pfn = end >> PAGE_SHIFT; | |
208 | ||
24a5da73 YL |
209 | node_data[nodeid] = early_node_mem(nodeid, start, end, pgdat_size, |
210 | SMP_CACHE_BYTES); | |
a8062231 AK |
211 | if (node_data[nodeid] == NULL) |
212 | return; | |
213 | nodedata_phys = __pa(node_data[nodeid]); | |
6118f76f YL |
214 | printk(KERN_INFO " NODE_DATA [%016lx - %016lx]\n", nodedata_phys, |
215 | nodedata_phys + pgdat_size - 1); | |
1da177e4 | 216 | |
1da177e4 LT |
217 | memset(NODE_DATA(nodeid), 0, sizeof(pg_data_t)); |
218 | NODE_DATA(nodeid)->bdata = &plat_node_bdata[nodeid]; | |
219 | NODE_DATA(nodeid)->node_start_pfn = start_pfn; | |
220 | NODE_DATA(nodeid)->node_spanned_pages = end_pfn - start_pfn; | |
221 | ||
1a27fc0a YL |
222 | /* |
223 | * Find a place for the bootmem map | |
224 | * nodedata_phys could be on other nodes by alloc_bootmem, | |
225 | * so need to sure bootmap_start not to be small, otherwise | |
226 | * early_node_mem will get that with find_e820_area instead | |
227 | * of alloc_bootmem, that could clash with reserved range | |
228 | */ | |
e3cfe529 | 229 | bootmap_pages = bootmem_bootmap_pages(end_pfn - start_pfn); |
1a27fc0a YL |
230 | nid = phys_to_nid(nodedata_phys); |
231 | if (nid == nodeid) | |
232 | bootmap_start = round_up(nodedata_phys + pgdat_size, PAGE_SIZE); | |
233 | else | |
234 | bootmap_start = round_up(start, PAGE_SIZE); | |
24a5da73 YL |
235 | /* |
236 | * SMP_CAHCE_BYTES could be enough, but init_bootmem_node like | |
237 | * to use that to align to PAGE_SIZE | |
238 | */ | |
a8062231 | 239 | bootmap = early_node_mem(nodeid, bootmap_start, end, |
24a5da73 | 240 | bootmap_pages<<PAGE_SHIFT, PAGE_SIZE); |
a8062231 AK |
241 | if (bootmap == NULL) { |
242 | if (nodedata_phys < start || nodedata_phys >= end) | |
37bff62e | 243 | free_bootmem(nodedata_phys, pgdat_size); |
a8062231 AK |
244 | node_data[nodeid] = NULL; |
245 | return; | |
246 | } | |
247 | bootmap_start = __pa(bootmap); | |
e3cfe529 | 248 | |
1da177e4 | 249 | bootmap_size = init_bootmem_node(NODE_DATA(nodeid), |
e3cfe529 TG |
250 | bootmap_start >> PAGE_SHIFT, |
251 | start_pfn, end_pfn); | |
1da177e4 | 252 | |
6118f76f YL |
253 | printk(KERN_INFO " bootmap [%016lx - %016lx] pages %lx\n", |
254 | bootmap_start, bootmap_start + bootmap_size - 1, | |
255 | bootmap_pages); | |
256 | ||
5cb248ab | 257 | free_bootmem_with_active_regions(nodeid, end); |
1da177e4 | 258 | |
1a27fc0a YL |
259 | /* |
260 | * convert early reserve to bootmem reserve earlier | |
261 | * otherwise early_node_mem could use early reserved mem | |
262 | * on previous node | |
263 | */ | |
264 | early_res_to_bootmem(start, end); | |
265 | ||
266 | /* | |
267 | * in some case early_node_mem could use alloc_bootmem | |
268 | * to get range on other node, don't reserve that again | |
269 | */ | |
270 | if (nid != nodeid) | |
271 | printk(KERN_INFO " NODE_DATA(%d) on node %d\n", nodeid, nid); | |
272 | else | |
273 | reserve_bootmem_node(NODE_DATA(nodeid), nodedata_phys, | |
274 | pgdat_size, BOOTMEM_DEFAULT); | |
275 | nid = phys_to_nid(bootmap_start); | |
276 | if (nid != nodeid) | |
277 | printk(KERN_INFO " bootmap(%d) on node %d\n", nodeid, nid); | |
278 | else | |
279 | reserve_bootmem_node(NODE_DATA(nodeid), bootmap_start, | |
280 | bootmap_pages<<PAGE_SHIFT, BOOTMEM_DEFAULT); | |
281 | ||
68a3a7fe AK |
282 | #ifdef CONFIG_ACPI_NUMA |
283 | srat_reserve_add_area(nodeid); | |
284 | #endif | |
1da177e4 | 285 | node_set_online(nodeid); |
e3cfe529 | 286 | } |
1da177e4 | 287 | |
e3cfe529 TG |
288 | /* |
289 | * There are unfortunately some poorly designed mainboards around that | |
290 | * only connect memory to a single CPU. This breaks the 1:1 cpu->node | |
291 | * mapping. To avoid this fill in the mapping for all possible CPUs, | |
292 | * as the number of CPUs is not known yet. We round robin the existing | |
293 | * nodes. | |
294 | */ | |
1da177e4 LT |
295 | void __init numa_init_array(void) |
296 | { | |
297 | int rr, i; | |
e3cfe529 | 298 | |
85cc5135 | 299 | rr = first_node(node_online_map); |
1da177e4 | 300 | for (i = 0; i < NR_CPUS; i++) { |
1ce35712 | 301 | if (early_cpu_to_node(i) != NUMA_NO_NODE) |
1da177e4 | 302 | continue; |
e3cfe529 | 303 | numa_set_node(i, rr); |
1da177e4 LT |
304 | rr = next_node(rr, node_online_map); |
305 | if (rr == MAX_NUMNODES) | |
306 | rr = first_node(node_online_map); | |
1da177e4 | 307 | } |
1da177e4 LT |
308 | } |
309 | ||
310 | #ifdef CONFIG_NUMA_EMU | |
53fee04f | 311 | /* Numa emulation */ |
8b8ca80e | 312 | char *cmdline __initdata; |
1da177e4 | 313 | |
53fee04f | 314 | /* |
e3cfe529 TG |
315 | * Setups up nid to range from addr to addr + size. If the end |
316 | * boundary is greater than max_addr, then max_addr is used instead. | |
317 | * The return value is 0 if there is additional memory left for | |
318 | * allocation past addr and -1 otherwise. addr is adjusted to be at | |
319 | * the end of the node. | |
53fee04f | 320 | */ |
8b8ca80e DR |
321 | static int __init setup_node_range(int nid, struct bootnode *nodes, u64 *addr, |
322 | u64 size, u64 max_addr) | |
53fee04f | 323 | { |
8b8ca80e | 324 | int ret = 0; |
e3cfe529 | 325 | |
8b8ca80e DR |
326 | nodes[nid].start = *addr; |
327 | *addr += size; | |
328 | if (*addr >= max_addr) { | |
329 | *addr = max_addr; | |
330 | ret = -1; | |
331 | } | |
332 | nodes[nid].end = *addr; | |
e3f1caee | 333 | node_set(nid, node_possible_map); |
8b8ca80e DR |
334 | printk(KERN_INFO "Faking node %d at %016Lx-%016Lx (%LuMB)\n", nid, |
335 | nodes[nid].start, nodes[nid].end, | |
336 | (nodes[nid].end - nodes[nid].start) >> 20); | |
337 | return ret; | |
53fee04f RS |
338 | } |
339 | ||
8b8ca80e DR |
340 | /* |
341 | * Splits num_nodes nodes up equally starting at node_start. The return value | |
342 | * is the number of nodes split up and addr is adjusted to be at the end of the | |
343 | * last node allocated. | |
344 | */ | |
345 | static int __init split_nodes_equally(struct bootnode *nodes, u64 *addr, | |
346 | u64 max_addr, int node_start, | |
347 | int num_nodes) | |
1da177e4 | 348 | { |
8b8ca80e DR |
349 | unsigned int big; |
350 | u64 size; | |
351 | int i; | |
53fee04f | 352 | |
8b8ca80e DR |
353 | if (num_nodes <= 0) |
354 | return -1; | |
355 | if (num_nodes > MAX_NUMNODES) | |
356 | num_nodes = MAX_NUMNODES; | |
a7e96629 | 357 | size = (max_addr - *addr - e820_hole_size(*addr, max_addr)) / |
8b8ca80e | 358 | num_nodes; |
53fee04f | 359 | /* |
8b8ca80e DR |
360 | * Calculate the number of big nodes that can be allocated as a result |
361 | * of consolidating the leftovers. | |
53fee04f | 362 | */ |
8b8ca80e DR |
363 | big = ((size & ~FAKE_NODE_MIN_HASH_MASK) * num_nodes) / |
364 | FAKE_NODE_MIN_SIZE; | |
365 | ||
366 | /* Round down to nearest FAKE_NODE_MIN_SIZE. */ | |
367 | size &= FAKE_NODE_MIN_HASH_MASK; | |
368 | if (!size) { | |
369 | printk(KERN_ERR "Not enough memory for each node. " | |
370 | "NUMA emulation disabled.\n"); | |
371 | return -1; | |
53fee04f | 372 | } |
8b8ca80e DR |
373 | |
374 | for (i = node_start; i < num_nodes + node_start; i++) { | |
375 | u64 end = *addr + size; | |
e3cfe529 | 376 | |
53fee04f RS |
377 | if (i < big) |
378 | end += FAKE_NODE_MIN_SIZE; | |
379 | /* | |
8b8ca80e DR |
380 | * The final node can have the remaining system RAM. Other |
381 | * nodes receive roughly the same amount of available pages. | |
53fee04f | 382 | */ |
8b8ca80e DR |
383 | if (i == num_nodes + node_start - 1) |
384 | end = max_addr; | |
385 | else | |
a7e96629 | 386 | while (end - *addr - e820_hole_size(*addr, end) < |
8b8ca80e DR |
387 | size) { |
388 | end += FAKE_NODE_MIN_SIZE; | |
389 | if (end > max_addr) { | |
390 | end = max_addr; | |
391 | break; | |
392 | } | |
393 | } | |
394 | if (setup_node_range(i, nodes, addr, end - *addr, max_addr) < 0) | |
395 | break; | |
396 | } | |
397 | return i - node_start + 1; | |
398 | } | |
399 | ||
382591d5 DR |
400 | /* |
401 | * Splits the remaining system RAM into chunks of size. The remaining memory is | |
402 | * always assigned to a final node and can be asymmetric. Returns the number of | |
403 | * nodes split. | |
404 | */ | |
405 | static int __init split_nodes_by_size(struct bootnode *nodes, u64 *addr, | |
406 | u64 max_addr, int node_start, u64 size) | |
407 | { | |
408 | int i = node_start; | |
409 | size = (size << 20) & FAKE_NODE_MIN_HASH_MASK; | |
410 | while (!setup_node_range(i++, nodes, addr, size, max_addr)) | |
411 | ; | |
412 | return i - node_start; | |
413 | } | |
414 | ||
8b8ca80e DR |
415 | /* |
416 | * Sets up the system RAM area from start_pfn to end_pfn according to the | |
417 | * numa=fake command-line option. | |
418 | */ | |
f46bdf2d MT |
419 | static struct bootnode nodes[MAX_NUMNODES] __initdata; |
420 | ||
8b8ca80e DR |
421 | static int __init numa_emulation(unsigned long start_pfn, unsigned long end_pfn) |
422 | { | |
e3cfe529 | 423 | u64 size, addr = start_pfn << PAGE_SHIFT; |
8b8ca80e | 424 | u64 max_addr = end_pfn << PAGE_SHIFT; |
e3cfe529 | 425 | int num_nodes = 0, num = 0, coeff_flag, coeff = -1, i; |
8b8ca80e DR |
426 | |
427 | memset(&nodes, 0, sizeof(nodes)); | |
428 | /* | |
429 | * If the numa=fake command-line is just a single number N, split the | |
430 | * system RAM into N fake nodes. | |
431 | */ | |
432 | if (!strchr(cmdline, '*') && !strchr(cmdline, ',')) { | |
e3cfe529 TG |
433 | long n = simple_strtol(cmdline, NULL, 0); |
434 | ||
435 | num_nodes = split_nodes_equally(nodes, &addr, max_addr, 0, n); | |
8b8ca80e DR |
436 | if (num_nodes < 0) |
437 | return num_nodes; | |
438 | goto out; | |
439 | } | |
440 | ||
441 | /* Parse the command line. */ | |
382591d5 | 442 | for (coeff_flag = 0; ; cmdline++) { |
8b8ca80e DR |
443 | if (*cmdline && isdigit(*cmdline)) { |
444 | num = num * 10 + *cmdline - '0'; | |
445 | continue; | |
53fee04f | 446 | } |
382591d5 DR |
447 | if (*cmdline == '*') { |
448 | if (num > 0) | |
449 | coeff = num; | |
450 | coeff_flag = 1; | |
451 | } | |
8b8ca80e | 452 | if (!*cmdline || *cmdline == ',') { |
382591d5 DR |
453 | if (!coeff_flag) |
454 | coeff = 1; | |
8b8ca80e DR |
455 | /* |
456 | * Round down to the nearest FAKE_NODE_MIN_SIZE. | |
457 | * Command-line coefficients are in megabytes. | |
458 | */ | |
459 | size = ((u64)num << 20) & FAKE_NODE_MIN_HASH_MASK; | |
382591d5 | 460 | if (size) |
8b8ca80e DR |
461 | for (i = 0; i < coeff; i++, num_nodes++) |
462 | if (setup_node_range(num_nodes, nodes, | |
463 | &addr, size, max_addr) < 0) | |
464 | goto done; | |
382591d5 DR |
465 | if (!*cmdline) |
466 | break; | |
467 | coeff_flag = 0; | |
468 | coeff = -1; | |
53fee04f | 469 | } |
8b8ca80e DR |
470 | num = 0; |
471 | } | |
472 | done: | |
473 | if (!num_nodes) | |
474 | return -1; | |
14694d73 | 475 | /* Fill remainder of system RAM, if appropriate. */ |
8b8ca80e | 476 | if (addr < max_addr) { |
382591d5 DR |
477 | if (coeff_flag && coeff < 0) { |
478 | /* Split remaining nodes into num-sized chunks */ | |
479 | num_nodes += split_nodes_by_size(nodes, &addr, max_addr, | |
480 | num_nodes, num); | |
481 | goto out; | |
482 | } | |
14694d73 DR |
483 | switch (*(cmdline - 1)) { |
484 | case '*': | |
485 | /* Split remaining nodes into coeff chunks */ | |
486 | if (coeff <= 0) | |
487 | break; | |
488 | num_nodes += split_nodes_equally(nodes, &addr, max_addr, | |
489 | num_nodes, coeff); | |
490 | break; | |
491 | case ',': | |
492 | /* Do not allocate remaining system RAM */ | |
493 | break; | |
494 | default: | |
495 | /* Give one final node */ | |
496 | setup_node_range(num_nodes, nodes, &addr, | |
497 | max_addr - addr, max_addr); | |
498 | num_nodes++; | |
499 | } | |
8b8ca80e DR |
500 | } |
501 | out: | |
6ec6e0d9 | 502 | memnode_shift = compute_hash_shift(nodes, num_nodes, NULL); |
8b8ca80e DR |
503 | if (memnode_shift < 0) { |
504 | memnode_shift = 0; | |
505 | printk(KERN_ERR "No NUMA hash function found. NUMA emulation " | |
506 | "disabled.\n"); | |
507 | return -1; | |
508 | } | |
509 | ||
510 | /* | |
511 | * We need to vacate all active ranges that may have been registered by | |
1c05f093 DR |
512 | * SRAT and set acpi_numa to -1 so that srat_disabled() always returns |
513 | * true. NUMA emulation has succeeded so we will not scan ACPI nodes. | |
8b8ca80e DR |
514 | */ |
515 | remove_all_active_ranges(); | |
1c05f093 DR |
516 | #ifdef CONFIG_ACPI_NUMA |
517 | acpi_numa = -1; | |
518 | #endif | |
e3f1caee | 519 | for_each_node_mask(i, node_possible_map) { |
5cb248ab MG |
520 | e820_register_active_regions(i, nodes[i].start >> PAGE_SHIFT, |
521 | nodes[i].end >> PAGE_SHIFT); | |
e3cfe529 | 522 | setup_node_bootmem(i, nodes[i].start, nodes[i].end); |
5cb248ab | 523 | } |
3484d798 | 524 | acpi_fake_nodes(nodes, num_nodes); |
e3cfe529 TG |
525 | numa_init_array(); |
526 | return 0; | |
1da177e4 | 527 | } |
8b8ca80e | 528 | #endif /* CONFIG_NUMA_EMU */ |
1da177e4 LT |
529 | |
530 | void __init numa_initmem_init(unsigned long start_pfn, unsigned long end_pfn) | |
e3cfe529 | 531 | { |
1da177e4 LT |
532 | int i; |
533 | ||
e3f1caee | 534 | nodes_clear(node_possible_map); |
b7ad149d | 535 | nodes_clear(node_online_map); |
e3f1caee | 536 | |
1da177e4 | 537 | #ifdef CONFIG_NUMA_EMU |
8b8ca80e | 538 | if (cmdline && !numa_emulation(start_pfn, end_pfn)) |
e3cfe529 | 539 | return; |
e3f1caee | 540 | nodes_clear(node_possible_map); |
b7ad149d | 541 | nodes_clear(node_online_map); |
1da177e4 LT |
542 | #endif |
543 | ||
544 | #ifdef CONFIG_ACPI_NUMA | |
545 | if (!numa_off && !acpi_scan_nodes(start_pfn << PAGE_SHIFT, | |
546 | end_pfn << PAGE_SHIFT)) | |
e3cfe529 | 547 | return; |
e3f1caee | 548 | nodes_clear(node_possible_map); |
b7ad149d | 549 | nodes_clear(node_online_map); |
1da177e4 LT |
550 | #endif |
551 | ||
552 | #ifdef CONFIG_K8_NUMA | |
e3cfe529 TG |
553 | if (!numa_off && !k8_scan_nodes(start_pfn<<PAGE_SHIFT, |
554 | end_pfn<<PAGE_SHIFT)) | |
1da177e4 | 555 | return; |
e3f1caee | 556 | nodes_clear(node_possible_map); |
b7ad149d | 557 | nodes_clear(node_online_map); |
1da177e4 LT |
558 | #endif |
559 | printk(KERN_INFO "%s\n", | |
560 | numa_off ? "NUMA turned off" : "No NUMA configuration found"); | |
561 | ||
e3cfe529 | 562 | printk(KERN_INFO "Faking a node at %016lx-%016lx\n", |
1da177e4 | 563 | start_pfn << PAGE_SHIFT, |
e3cfe529 TG |
564 | end_pfn << PAGE_SHIFT); |
565 | /* setup dummy node covering all memory */ | |
566 | memnode_shift = 63; | |
076422d2 | 567 | memnodemap = memnode.embedded_map; |
1da177e4 | 568 | memnodemap[0] = 0; |
1da177e4 | 569 | node_set_online(0); |
e3f1caee | 570 | node_set(0, node_possible_map); |
1da177e4 | 571 | for (i = 0; i < NR_CPUS; i++) |
69d81fcd | 572 | numa_set_node(i, 0); |
df3825c5 | 573 | /* cpumask_of_cpu() may not be available during early startup */ |
3cc87e3f | 574 | memset(&node_to_cpumask_map[0], 0, sizeof(node_to_cpumask_map[0])); |
575 | cpu_set(0, node_to_cpumask_map[0]); | |
5cb248ab | 576 | e820_register_active_regions(0, start_pfn, end_pfn); |
1da177e4 LT |
577 | setup_node_bootmem(0, start_pfn << PAGE_SHIFT, end_pfn << PAGE_SHIFT); |
578 | } | |
579 | ||
e6982c67 | 580 | __cpuinit void numa_add_cpu(int cpu) |
1da177e4 | 581 | { |
1ce35712 | 582 | set_bit(cpu, |
583 | (unsigned long *)&node_to_cpumask_map[early_cpu_to_node(cpu)]); | |
e3cfe529 | 584 | } |
1da177e4 | 585 | |
69d81fcd AK |
586 | void __cpuinit numa_set_node(int cpu, int node) |
587 | { | |
43238382 | 588 | int *cpu_to_node_map = x86_cpu_to_node_map_early_ptr; |
df3825c5 | 589 | |
df3825c5 | 590 | if(cpu_to_node_map) |
591 | cpu_to_node_map[cpu] = node; | |
592 | else if(per_cpu_offset(cpu)) | |
593 | per_cpu(x86_cpu_to_node_map, cpu) = node; | |
594 | else | |
595 | Dprintk(KERN_INFO "Setting node for non-present cpu %d\n", cpu); | |
69d81fcd AK |
596 | } |
597 | ||
e3cfe529 TG |
598 | unsigned long __init numa_free_all_bootmem(void) |
599 | { | |
1da177e4 | 600 | unsigned long pages = 0; |
e3cfe529 TG |
601 | int i; |
602 | ||
603 | for_each_online_node(i) | |
1da177e4 | 604 | pages += free_all_bootmem_node(NODE_DATA(i)); |
e3cfe529 | 605 | |
1da177e4 | 606 | return pages; |
e3cfe529 | 607 | } |
1da177e4 LT |
608 | |
609 | void __init paging_init(void) | |
e3cfe529 | 610 | { |
6391af17 | 611 | unsigned long max_zone_pfns[MAX_NR_ZONES]; |
e3cfe529 | 612 | |
6391af17 MG |
613 | memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); |
614 | max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN; | |
615 | max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN; | |
616 | max_zone_pfns[ZONE_NORMAL] = end_pfn; | |
d3ee871e | 617 | |
f0a5a58a BP |
618 | sparse_memory_present_with_active_regions(MAX_NUMNODES); |
619 | sparse_init(); | |
d3ee871e | 620 | |
5cb248ab | 621 | free_area_init_nodes(max_zone_pfns); |
e3cfe529 | 622 | } |
1da177e4 | 623 | |
2c8c0e6b | 624 | static __init int numa_setup(char *opt) |
e3cfe529 | 625 | { |
2c8c0e6b AK |
626 | if (!opt) |
627 | return -EINVAL; | |
e3cfe529 | 628 | if (!strncmp(opt, "off", 3)) |
1da177e4 LT |
629 | numa_off = 1; |
630 | #ifdef CONFIG_NUMA_EMU | |
8b8ca80e DR |
631 | if (!strncmp(opt, "fake=", 5)) |
632 | cmdline = opt + 5; | |
1da177e4 LT |
633 | #endif |
634 | #ifdef CONFIG_ACPI_NUMA | |
e3cfe529 TG |
635 | if (!strncmp(opt, "noacpi", 6)) |
636 | acpi_numa = -1; | |
637 | if (!strncmp(opt, "hotadd=", 7)) | |
68a3a7fe | 638 | hotadd_percent = simple_strtoul(opt+7, NULL, 10); |
1da177e4 | 639 | #endif |
2c8c0e6b | 640 | return 0; |
e3cfe529 | 641 | } |
2c8c0e6b AK |
642 | early_param("numa", numa_setup); |
643 | ||
05b3cbd8 RT |
644 | /* |
645 | * Setup early cpu_to_node. | |
646 | * | |
647 | * Populate cpu_to_node[] only if x86_cpu_to_apicid[], | |
648 | * and apicid_to_node[] tables have valid entries for a CPU. | |
649 | * This means we skip cpu_to_node[] initialisation for NUMA | |
650 | * emulation and faking node case (when running a kernel compiled | |
651 | * for NUMA on a non NUMA box), which is OK as cpu_to_node[] | |
652 | * is already initialized in a round robin manner at numa_init_array, | |
653 | * prior to this call, and this initialization is good enough | |
654 | * for the fake NUMA cases. | |
655 | */ | |
656 | void __init init_cpu_to_node(void) | |
657 | { | |
658 | int i; | |
e3cfe529 TG |
659 | |
660 | for (i = 0; i < NR_CPUS; i++) { | |
7c9e92b6 | 661 | int node; |
ef97001f | 662 | u16 apicid = x86_cpu_to_apicid_init[i]; |
e3cfe529 | 663 | |
05b3cbd8 RT |
664 | if (apicid == BAD_APICID) |
665 | continue; | |
7c9e92b6 YL |
666 | node = apicid_to_node[apicid]; |
667 | if (node == NUMA_NO_NODE) | |
05b3cbd8 | 668 | continue; |
7c9e92b6 YL |
669 | if (!node_online(node)) |
670 | continue; | |
671 | numa_set_node(i, node); | |
05b3cbd8 RT |
672 | } |
673 | } | |
674 | ||
cf050132 | 675 |