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 | |
df3825c5 | 34 | u16 x86_cpu_to_node_map_init[NR_CPUS] __initdata = { |
3f098c26 | 35 | [0 ... NR_CPUS-1] = NUMA_NO_NODE |
0b07e984 | 36 | }; |
df3825c5 | 37 | void *x86_cpu_to_node_map_early_ptr; |
38 | EXPORT_SYMBOL(x86_cpu_to_node_map_init); | |
39 | EXPORT_SYMBOL(x86_cpu_to_node_map_early_ptr); | |
40 | DEFINE_PER_CPU(u16, x86_cpu_to_node_map) = NUMA_NO_NODE; | |
41 | EXPORT_PER_CPU_SYMBOL(x86_cpu_to_node_map); | |
e3cfe529 | 42 | |
3cc87e3f | 43 | u16 apicid_to_node[MAX_LOCAL_APIC] __cpuinitdata = { |
e3cfe529 | 44 | [0 ... MAX_LOCAL_APIC-1] = NUMA_NO_NODE |
3f098c26 | 45 | }; |
e3cfe529 | 46 | |
7462894a TG |
47 | cpumask_t node_to_cpumask_map[MAX_NUMNODES] __read_mostly; |
48 | EXPORT_SYMBOL(node_to_cpumask_map); | |
1da177e4 LT |
49 | |
50 | int numa_off __initdata; | |
076422d2 AS |
51 | unsigned long __initdata nodemap_addr; |
52 | unsigned long __initdata nodemap_size; | |
1da177e4 | 53 | |
529a3404 ED |
54 | /* |
55 | * Given a shift value, try to populate memnodemap[] | |
56 | * Returns : | |
57 | * 1 if OK | |
58 | * 0 if memnodmap[] too small (of shift too small) | |
59 | * -1 if node overlap or lost ram (shift too big) | |
60 | */ | |
e3cfe529 TG |
61 | static int __init populate_memnodemap(const struct bootnode *nodes, |
62 | int numnodes, int shift) | |
1da177e4 | 63 | { |
529a3404 | 64 | unsigned long addr, end; |
e3cfe529 | 65 | int i, res = -1; |
b684664f | 66 | |
076422d2 | 67 | memset(memnodemap, 0xff, memnodemapsize); |
b684664f | 68 | for (i = 0; i < numnodes; i++) { |
529a3404 ED |
69 | addr = nodes[i].start; |
70 | end = nodes[i].end; | |
71 | if (addr >= end) | |
b684664f | 72 | continue; |
076422d2 | 73 | if ((end >> shift) >= memnodemapsize) |
529a3404 ED |
74 | return 0; |
75 | do { | |
76 | if (memnodemap[addr >> shift] != 0xff) | |
b684664f | 77 | return -1; |
b684664f | 78 | memnodemap[addr >> shift] = i; |
076422d2 | 79 | addr += (1UL << shift); |
529a3404 ED |
80 | } while (addr < end); |
81 | res = 1; | |
e3cfe529 | 82 | } |
529a3404 ED |
83 | return res; |
84 | } | |
85 | ||
076422d2 AS |
86 | static int __init allocate_cachealigned_memnodemap(void) |
87 | { | |
88 | unsigned long pad, pad_addr; | |
89 | ||
90 | memnodemap = memnode.embedded_map; | |
54413927 | 91 | if (memnodemapsize <= 48) |
076422d2 | 92 | return 0; |
076422d2 AS |
93 | |
94 | pad = L1_CACHE_BYTES - 1; | |
95 | pad_addr = 0x8000; | |
96 | nodemap_size = pad + memnodemapsize; | |
97 | nodemap_addr = find_e820_area(pad_addr, end_pfn<<PAGE_SHIFT, | |
98 | nodemap_size); | |
99 | if (nodemap_addr == -1UL) { | |
100 | printk(KERN_ERR | |
101 | "NUMA: Unable to allocate Memory to Node hash map\n"); | |
102 | nodemap_addr = nodemap_size = 0; | |
103 | return -1; | |
104 | } | |
105 | pad_addr = (nodemap_addr + pad) & ~pad; | |
106 | memnodemap = phys_to_virt(pad_addr); | |
107 | ||
108 | printk(KERN_DEBUG "NUMA: Allocated memnodemap from %lx - %lx\n", | |
109 | nodemap_addr, nodemap_addr + nodemap_size); | |
110 | return 0; | |
111 | } | |
112 | ||
113 | /* | |
114 | * The LSB of all start and end addresses in the node map is the value of the | |
115 | * maximum possible shift. | |
116 | */ | |
e3cfe529 TG |
117 | static int __init extract_lsb_from_nodes(const struct bootnode *nodes, |
118 | int numnodes) | |
529a3404 | 119 | { |
54413927 | 120 | int i, nodes_used = 0; |
076422d2 AS |
121 | unsigned long start, end; |
122 | unsigned long bitfield = 0, memtop = 0; | |
123 | ||
124 | for (i = 0; i < numnodes; i++) { | |
125 | start = nodes[i].start; | |
126 | end = nodes[i].end; | |
127 | if (start >= end) | |
128 | continue; | |
54413927 AS |
129 | bitfield |= start; |
130 | nodes_used++; | |
076422d2 AS |
131 | if (end > memtop) |
132 | memtop = end; | |
133 | } | |
54413927 AS |
134 | if (nodes_used <= 1) |
135 | i = 63; | |
136 | else | |
137 | i = find_first_bit(&bitfield, sizeof(unsigned long)*8); | |
076422d2 AS |
138 | memnodemapsize = (memtop >> i)+1; |
139 | return i; | |
140 | } | |
529a3404 | 141 | |
076422d2 AS |
142 | int __init compute_hash_shift(struct bootnode *nodes, int numnodes) |
143 | { | |
144 | int shift; | |
529a3404 | 145 | |
076422d2 AS |
146 | shift = extract_lsb_from_nodes(nodes, numnodes); |
147 | if (allocate_cachealigned_memnodemap()) | |
148 | return -1; | |
6b050f80 | 149 | printk(KERN_DEBUG "NUMA: Using %d for the hash shift.\n", |
529a3404 ED |
150 | shift); |
151 | ||
152 | if (populate_memnodemap(nodes, numnodes, shift) != 1) { | |
e3cfe529 TG |
153 | printk(KERN_INFO "Your memory is not aligned you need to " |
154 | "rebuild your kernel with a bigger NODEMAPSIZE " | |
155 | "shift=%d\n", shift); | |
529a3404 ED |
156 | return -1; |
157 | } | |
b684664f | 158 | return shift; |
1da177e4 LT |
159 | } |
160 | ||
bbfceef4 MT |
161 | int early_pfn_to_nid(unsigned long pfn) |
162 | { | |
163 | return phys_to_nid(pfn << PAGE_SHIFT); | |
164 | } | |
bbfceef4 | 165 | |
e3cfe529 TG |
166 | static void * __init early_node_mem(int nodeid, unsigned long start, |
167 | unsigned long end, unsigned long size) | |
a8062231 AK |
168 | { |
169 | unsigned long mem = find_e820_area(start, end, size); | |
170 | void *ptr; | |
e3cfe529 | 171 | |
a8062231 AK |
172 | if (mem != -1L) |
173 | return __va(mem); | |
174 | ptr = __alloc_bootmem_nopanic(size, | |
175 | SMP_CACHE_BYTES, __pa(MAX_DMA_ADDRESS)); | |
83e83d54 | 176 | if (ptr == NULL) { |
a8062231 | 177 | printk(KERN_ERR "Cannot find %lu bytes in node %d\n", |
e3cfe529 | 178 | size, nodeid); |
a8062231 AK |
179 | return NULL; |
180 | } | |
181 | return ptr; | |
182 | } | |
183 | ||
1da177e4 | 184 | /* Initialize bootmem allocator for a node */ |
e3cfe529 TG |
185 | void __init setup_node_bootmem(int nodeid, unsigned long start, |
186 | unsigned long end) | |
187 | { | |
188 | unsigned long start_pfn, end_pfn, bootmap_pages, bootmap_size; | |
189 | unsigned long bootmap_start, nodedata_phys; | |
a8062231 | 190 | void *bootmap; |
1da177e4 LT |
191 | const int pgdat_size = round_up(sizeof(pg_data_t), PAGE_SIZE); |
192 | ||
e3cfe529 | 193 | start = round_up(start, ZONE_ALIGN); |
1da177e4 | 194 | |
e3cfe529 TG |
195 | printk(KERN_INFO "Bootmem setup node %d %016lx-%016lx\n", nodeid, |
196 | start, end); | |
1da177e4 LT |
197 | |
198 | start_pfn = start >> PAGE_SHIFT; | |
199 | end_pfn = end >> PAGE_SHIFT; | |
200 | ||
a8062231 AK |
201 | node_data[nodeid] = early_node_mem(nodeid, start, end, pgdat_size); |
202 | if (node_data[nodeid] == NULL) | |
203 | return; | |
204 | nodedata_phys = __pa(node_data[nodeid]); | |
1da177e4 | 205 | |
1da177e4 LT |
206 | memset(NODE_DATA(nodeid), 0, sizeof(pg_data_t)); |
207 | NODE_DATA(nodeid)->bdata = &plat_node_bdata[nodeid]; | |
208 | NODE_DATA(nodeid)->node_start_pfn = start_pfn; | |
209 | NODE_DATA(nodeid)->node_spanned_pages = end_pfn - start_pfn; | |
210 | ||
211 | /* Find a place for the bootmem map */ | |
e3cfe529 | 212 | bootmap_pages = bootmem_bootmap_pages(end_pfn - start_pfn); |
1da177e4 | 213 | bootmap_start = round_up(nodedata_phys + pgdat_size, PAGE_SIZE); |
a8062231 AK |
214 | bootmap = early_node_mem(nodeid, bootmap_start, end, |
215 | bootmap_pages<<PAGE_SHIFT); | |
216 | if (bootmap == NULL) { | |
217 | if (nodedata_phys < start || nodedata_phys >= end) | |
e3cfe529 TG |
218 | free_bootmem((unsigned long)node_data[nodeid], |
219 | pgdat_size); | |
a8062231 AK |
220 | node_data[nodeid] = NULL; |
221 | return; | |
222 | } | |
223 | bootmap_start = __pa(bootmap); | |
e3cfe529 TG |
224 | Dprintk("bootmap start %lu pages %lu\n", bootmap_start, bootmap_pages); |
225 | ||
1da177e4 | 226 | bootmap_size = init_bootmem_node(NODE_DATA(nodeid), |
e3cfe529 TG |
227 | bootmap_start >> PAGE_SHIFT, |
228 | start_pfn, end_pfn); | |
1da177e4 | 229 | |
5cb248ab | 230 | free_bootmem_with_active_regions(nodeid, end); |
1da177e4 | 231 | |
e3cfe529 TG |
232 | reserve_bootmem_node(NODE_DATA(nodeid), nodedata_phys, pgdat_size); |
233 | reserve_bootmem_node(NODE_DATA(nodeid), bootmap_start, | |
234 | bootmap_pages<<PAGE_SHIFT); | |
68a3a7fe AK |
235 | #ifdef CONFIG_ACPI_NUMA |
236 | srat_reserve_add_area(nodeid); | |
237 | #endif | |
1da177e4 | 238 | node_set_online(nodeid); |
e3cfe529 | 239 | } |
1da177e4 | 240 | |
a261670a | 241 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
1da177e4 | 242 | /* Initialize final allocator for a zone */ |
a261670a | 243 | static void __init flat_setup_node_zones(int nodeid) |
e3cfe529 | 244 | { |
267b4801 | 245 | unsigned long start_pfn, end_pfn, memmapsize, limit; |
1da177e4 | 246 | |
e3cfe529 TG |
247 | start_pfn = node_start_pfn(nodeid); |
248 | end_pfn = node_end_pfn(nodeid); | |
1da177e4 | 249 | |
5cb248ab | 250 | Dprintk(KERN_INFO "Setting up memmap for node %d %lx-%lx\n", |
a2f1b424 | 251 | nodeid, start_pfn, end_pfn); |
1da177e4 | 252 | |
e3cfe529 TG |
253 | /* |
254 | * Try to allocate mem_map at end to not fill up precious <4GB | |
255 | * memory. | |
256 | */ | |
267b4801 AK |
257 | memmapsize = sizeof(struct page) * (end_pfn-start_pfn); |
258 | limit = end_pfn << PAGE_SHIFT; | |
a261670a | 259 | |
e3cfe529 TG |
260 | NODE_DATA(nodeid)->node_mem_map = |
261 | __alloc_bootmem_core(NODE_DATA(nodeid)->bdata, | |
262 | memmapsize, SMP_CACHE_BYTES, | |
263 | round_down(limit - memmapsize, PAGE_SIZE), | |
264 | limit); | |
e3cfe529 | 265 | } |
a261670a YL |
266 | #else |
267 | #define flat_setup_node_zones(i) do {} while (0) | |
268 | #endif | |
1da177e4 | 269 | |
e3cfe529 TG |
270 | /* |
271 | * There are unfortunately some poorly designed mainboards around that | |
272 | * only connect memory to a single CPU. This breaks the 1:1 cpu->node | |
273 | * mapping. To avoid this fill in the mapping for all possible CPUs, | |
274 | * as the number of CPUs is not known yet. We round robin the existing | |
275 | * nodes. | |
276 | */ | |
1da177e4 LT |
277 | void __init numa_init_array(void) |
278 | { | |
279 | int rr, i; | |
e3cfe529 | 280 | |
85cc5135 | 281 | rr = first_node(node_online_map); |
1da177e4 | 282 | for (i = 0; i < NR_CPUS; i++) { |
98c9e27a | 283 | if (cpu_to_node(i) != NUMA_NO_NODE) |
1da177e4 | 284 | continue; |
e3cfe529 | 285 | numa_set_node(i, rr); |
1da177e4 LT |
286 | rr = next_node(rr, node_online_map); |
287 | if (rr == MAX_NUMNODES) | |
288 | rr = first_node(node_online_map); | |
1da177e4 | 289 | } |
1da177e4 LT |
290 | } |
291 | ||
292 | #ifdef CONFIG_NUMA_EMU | |
53fee04f | 293 | /* Numa emulation */ |
8b8ca80e | 294 | char *cmdline __initdata; |
1da177e4 | 295 | |
53fee04f | 296 | /* |
e3cfe529 TG |
297 | * Setups up nid to range from addr to addr + size. If the end |
298 | * boundary is greater than max_addr, then max_addr is used instead. | |
299 | * The return value is 0 if there is additional memory left for | |
300 | * allocation past addr and -1 otherwise. addr is adjusted to be at | |
301 | * the end of the node. | |
53fee04f | 302 | */ |
8b8ca80e DR |
303 | static int __init setup_node_range(int nid, struct bootnode *nodes, u64 *addr, |
304 | u64 size, u64 max_addr) | |
53fee04f | 305 | { |
8b8ca80e | 306 | int ret = 0; |
e3cfe529 | 307 | |
8b8ca80e DR |
308 | nodes[nid].start = *addr; |
309 | *addr += size; | |
310 | if (*addr >= max_addr) { | |
311 | *addr = max_addr; | |
312 | ret = -1; | |
313 | } | |
314 | nodes[nid].end = *addr; | |
e3f1caee | 315 | node_set(nid, node_possible_map); |
8b8ca80e DR |
316 | printk(KERN_INFO "Faking node %d at %016Lx-%016Lx (%LuMB)\n", nid, |
317 | nodes[nid].start, nodes[nid].end, | |
318 | (nodes[nid].end - nodes[nid].start) >> 20); | |
319 | return ret; | |
53fee04f RS |
320 | } |
321 | ||
8b8ca80e DR |
322 | /* |
323 | * Splits num_nodes nodes up equally starting at node_start. The return value | |
324 | * is the number of nodes split up and addr is adjusted to be at the end of the | |
325 | * last node allocated. | |
326 | */ | |
327 | static int __init split_nodes_equally(struct bootnode *nodes, u64 *addr, | |
328 | u64 max_addr, int node_start, | |
329 | int num_nodes) | |
1da177e4 | 330 | { |
8b8ca80e DR |
331 | unsigned int big; |
332 | u64 size; | |
333 | int i; | |
53fee04f | 334 | |
8b8ca80e DR |
335 | if (num_nodes <= 0) |
336 | return -1; | |
337 | if (num_nodes > MAX_NUMNODES) | |
338 | num_nodes = MAX_NUMNODES; | |
a7e96629 | 339 | size = (max_addr - *addr - e820_hole_size(*addr, max_addr)) / |
8b8ca80e | 340 | num_nodes; |
53fee04f | 341 | /* |
8b8ca80e DR |
342 | * Calculate the number of big nodes that can be allocated as a result |
343 | * of consolidating the leftovers. | |
53fee04f | 344 | */ |
8b8ca80e DR |
345 | big = ((size & ~FAKE_NODE_MIN_HASH_MASK) * num_nodes) / |
346 | FAKE_NODE_MIN_SIZE; | |
347 | ||
348 | /* Round down to nearest FAKE_NODE_MIN_SIZE. */ | |
349 | size &= FAKE_NODE_MIN_HASH_MASK; | |
350 | if (!size) { | |
351 | printk(KERN_ERR "Not enough memory for each node. " | |
352 | "NUMA emulation disabled.\n"); | |
353 | return -1; | |
53fee04f | 354 | } |
8b8ca80e DR |
355 | |
356 | for (i = node_start; i < num_nodes + node_start; i++) { | |
357 | u64 end = *addr + size; | |
e3cfe529 | 358 | |
53fee04f RS |
359 | if (i < big) |
360 | end += FAKE_NODE_MIN_SIZE; | |
361 | /* | |
8b8ca80e DR |
362 | * The final node can have the remaining system RAM. Other |
363 | * nodes receive roughly the same amount of available pages. | |
53fee04f | 364 | */ |
8b8ca80e DR |
365 | if (i == num_nodes + node_start - 1) |
366 | end = max_addr; | |
367 | else | |
a7e96629 | 368 | while (end - *addr - e820_hole_size(*addr, end) < |
8b8ca80e DR |
369 | size) { |
370 | end += FAKE_NODE_MIN_SIZE; | |
371 | if (end > max_addr) { | |
372 | end = max_addr; | |
373 | break; | |
374 | } | |
375 | } | |
376 | if (setup_node_range(i, nodes, addr, end - *addr, max_addr) < 0) | |
377 | break; | |
378 | } | |
379 | return i - node_start + 1; | |
380 | } | |
381 | ||
382591d5 DR |
382 | /* |
383 | * Splits the remaining system RAM into chunks of size. The remaining memory is | |
384 | * always assigned to a final node and can be asymmetric. Returns the number of | |
385 | * nodes split. | |
386 | */ | |
387 | static int __init split_nodes_by_size(struct bootnode *nodes, u64 *addr, | |
388 | u64 max_addr, int node_start, u64 size) | |
389 | { | |
390 | int i = node_start; | |
391 | size = (size << 20) & FAKE_NODE_MIN_HASH_MASK; | |
392 | while (!setup_node_range(i++, nodes, addr, size, max_addr)) | |
393 | ; | |
394 | return i - node_start; | |
395 | } | |
396 | ||
8b8ca80e DR |
397 | /* |
398 | * Sets up the system RAM area from start_pfn to end_pfn according to the | |
399 | * numa=fake command-line option. | |
400 | */ | |
401 | static int __init numa_emulation(unsigned long start_pfn, unsigned long end_pfn) | |
402 | { | |
403 | struct bootnode nodes[MAX_NUMNODES]; | |
e3cfe529 | 404 | u64 size, addr = start_pfn << PAGE_SHIFT; |
8b8ca80e | 405 | u64 max_addr = end_pfn << PAGE_SHIFT; |
e3cfe529 | 406 | int num_nodes = 0, num = 0, coeff_flag, coeff = -1, i; |
8b8ca80e DR |
407 | |
408 | memset(&nodes, 0, sizeof(nodes)); | |
409 | /* | |
410 | * If the numa=fake command-line is just a single number N, split the | |
411 | * system RAM into N fake nodes. | |
412 | */ | |
413 | if (!strchr(cmdline, '*') && !strchr(cmdline, ',')) { | |
e3cfe529 TG |
414 | long n = simple_strtol(cmdline, NULL, 0); |
415 | ||
416 | num_nodes = split_nodes_equally(nodes, &addr, max_addr, 0, n); | |
8b8ca80e DR |
417 | if (num_nodes < 0) |
418 | return num_nodes; | |
419 | goto out; | |
420 | } | |
421 | ||
422 | /* Parse the command line. */ | |
382591d5 | 423 | for (coeff_flag = 0; ; cmdline++) { |
8b8ca80e DR |
424 | if (*cmdline && isdigit(*cmdline)) { |
425 | num = num * 10 + *cmdline - '0'; | |
426 | continue; | |
53fee04f | 427 | } |
382591d5 DR |
428 | if (*cmdline == '*') { |
429 | if (num > 0) | |
430 | coeff = num; | |
431 | coeff_flag = 1; | |
432 | } | |
8b8ca80e | 433 | if (!*cmdline || *cmdline == ',') { |
382591d5 DR |
434 | if (!coeff_flag) |
435 | coeff = 1; | |
8b8ca80e DR |
436 | /* |
437 | * Round down to the nearest FAKE_NODE_MIN_SIZE. | |
438 | * Command-line coefficients are in megabytes. | |
439 | */ | |
440 | size = ((u64)num << 20) & FAKE_NODE_MIN_HASH_MASK; | |
382591d5 | 441 | if (size) |
8b8ca80e DR |
442 | for (i = 0; i < coeff; i++, num_nodes++) |
443 | if (setup_node_range(num_nodes, nodes, | |
444 | &addr, size, max_addr) < 0) | |
445 | goto done; | |
382591d5 DR |
446 | if (!*cmdline) |
447 | break; | |
448 | coeff_flag = 0; | |
449 | coeff = -1; | |
53fee04f | 450 | } |
8b8ca80e DR |
451 | num = 0; |
452 | } | |
453 | done: | |
454 | if (!num_nodes) | |
455 | return -1; | |
14694d73 | 456 | /* Fill remainder of system RAM, if appropriate. */ |
8b8ca80e | 457 | if (addr < max_addr) { |
382591d5 DR |
458 | if (coeff_flag && coeff < 0) { |
459 | /* Split remaining nodes into num-sized chunks */ | |
460 | num_nodes += split_nodes_by_size(nodes, &addr, max_addr, | |
461 | num_nodes, num); | |
462 | goto out; | |
463 | } | |
14694d73 DR |
464 | switch (*(cmdline - 1)) { |
465 | case '*': | |
466 | /* Split remaining nodes into coeff chunks */ | |
467 | if (coeff <= 0) | |
468 | break; | |
469 | num_nodes += split_nodes_equally(nodes, &addr, max_addr, | |
470 | num_nodes, coeff); | |
471 | break; | |
472 | case ',': | |
473 | /* Do not allocate remaining system RAM */ | |
474 | break; | |
475 | default: | |
476 | /* Give one final node */ | |
477 | setup_node_range(num_nodes, nodes, &addr, | |
478 | max_addr - addr, max_addr); | |
479 | num_nodes++; | |
480 | } | |
8b8ca80e DR |
481 | } |
482 | out: | |
483 | memnode_shift = compute_hash_shift(nodes, num_nodes); | |
484 | if (memnode_shift < 0) { | |
485 | memnode_shift = 0; | |
486 | printk(KERN_ERR "No NUMA hash function found. NUMA emulation " | |
487 | "disabled.\n"); | |
488 | return -1; | |
489 | } | |
490 | ||
491 | /* | |
492 | * We need to vacate all active ranges that may have been registered by | |
1c05f093 DR |
493 | * SRAT and set acpi_numa to -1 so that srat_disabled() always returns |
494 | * true. NUMA emulation has succeeded so we will not scan ACPI nodes. | |
8b8ca80e DR |
495 | */ |
496 | remove_all_active_ranges(); | |
1c05f093 DR |
497 | #ifdef CONFIG_ACPI_NUMA |
498 | acpi_numa = -1; | |
499 | #endif | |
e3f1caee | 500 | for_each_node_mask(i, node_possible_map) { |
5cb248ab MG |
501 | e820_register_active_regions(i, nodes[i].start >> PAGE_SHIFT, |
502 | nodes[i].end >> PAGE_SHIFT); | |
e3cfe529 | 503 | setup_node_bootmem(i, nodes[i].start, nodes[i].end); |
5cb248ab | 504 | } |
3484d798 | 505 | acpi_fake_nodes(nodes, num_nodes); |
e3cfe529 TG |
506 | numa_init_array(); |
507 | return 0; | |
1da177e4 | 508 | } |
8b8ca80e | 509 | #endif /* CONFIG_NUMA_EMU */ |
1da177e4 LT |
510 | |
511 | void __init numa_initmem_init(unsigned long start_pfn, unsigned long end_pfn) | |
e3cfe529 | 512 | { |
1da177e4 LT |
513 | int i; |
514 | ||
e3f1caee SS |
515 | nodes_clear(node_possible_map); |
516 | ||
1da177e4 | 517 | #ifdef CONFIG_NUMA_EMU |
8b8ca80e | 518 | if (cmdline && !numa_emulation(start_pfn, end_pfn)) |
e3cfe529 | 519 | return; |
e3f1caee | 520 | nodes_clear(node_possible_map); |
1da177e4 LT |
521 | #endif |
522 | ||
523 | #ifdef CONFIG_ACPI_NUMA | |
524 | if (!numa_off && !acpi_scan_nodes(start_pfn << PAGE_SHIFT, | |
525 | end_pfn << PAGE_SHIFT)) | |
e3cfe529 | 526 | return; |
e3f1caee | 527 | nodes_clear(node_possible_map); |
1da177e4 LT |
528 | #endif |
529 | ||
530 | #ifdef CONFIG_K8_NUMA | |
e3cfe529 TG |
531 | if (!numa_off && !k8_scan_nodes(start_pfn<<PAGE_SHIFT, |
532 | end_pfn<<PAGE_SHIFT)) | |
1da177e4 | 533 | return; |
e3f1caee | 534 | nodes_clear(node_possible_map); |
1da177e4 LT |
535 | #endif |
536 | printk(KERN_INFO "%s\n", | |
537 | numa_off ? "NUMA turned off" : "No NUMA configuration found"); | |
538 | ||
e3cfe529 | 539 | printk(KERN_INFO "Faking a node at %016lx-%016lx\n", |
1da177e4 | 540 | start_pfn << PAGE_SHIFT, |
e3cfe529 TG |
541 | end_pfn << PAGE_SHIFT); |
542 | /* setup dummy node covering all memory */ | |
543 | memnode_shift = 63; | |
076422d2 | 544 | memnodemap = memnode.embedded_map; |
1da177e4 LT |
545 | memnodemap[0] = 0; |
546 | nodes_clear(node_online_map); | |
547 | node_set_online(0); | |
e3f1caee | 548 | node_set(0, node_possible_map); |
1da177e4 | 549 | for (i = 0; i < NR_CPUS; i++) |
69d81fcd | 550 | numa_set_node(i, 0); |
df3825c5 | 551 | /* cpumask_of_cpu() may not be available during early startup */ |
3cc87e3f | 552 | memset(&node_to_cpumask_map[0], 0, sizeof(node_to_cpumask_map[0])); |
553 | cpu_set(0, node_to_cpumask_map[0]); | |
5cb248ab | 554 | e820_register_active_regions(0, start_pfn, end_pfn); |
1da177e4 LT |
555 | setup_node_bootmem(0, start_pfn << PAGE_SHIFT, end_pfn << PAGE_SHIFT); |
556 | } | |
557 | ||
e6982c67 | 558 | __cpuinit void numa_add_cpu(int cpu) |
1da177e4 | 559 | { |
5548fecd | 560 | set_bit(cpu, (unsigned long *)&node_to_cpumask_map[cpu_to_node(cpu)]); |
e3cfe529 | 561 | } |
1da177e4 | 562 | |
69d81fcd AK |
563 | void __cpuinit numa_set_node(int cpu, int node) |
564 | { | |
df3825c5 | 565 | u16 *cpu_to_node_map = (u16 *)x86_cpu_to_node_map_early_ptr; |
566 | ||
df79efde | 567 | cpu_pda(cpu)->nodenumber = node; |
df3825c5 | 568 | |
569 | if(cpu_to_node_map) | |
570 | cpu_to_node_map[cpu] = node; | |
571 | else if(per_cpu_offset(cpu)) | |
572 | per_cpu(x86_cpu_to_node_map, cpu) = node; | |
573 | else | |
574 | Dprintk(KERN_INFO "Setting node for non-present cpu %d\n", cpu); | |
69d81fcd AK |
575 | } |
576 | ||
e3cfe529 TG |
577 | unsigned long __init numa_free_all_bootmem(void) |
578 | { | |
1da177e4 | 579 | unsigned long pages = 0; |
e3cfe529 TG |
580 | int i; |
581 | ||
582 | for_each_online_node(i) | |
1da177e4 | 583 | pages += free_all_bootmem_node(NODE_DATA(i)); |
e3cfe529 | 584 | |
1da177e4 | 585 | return pages; |
e3cfe529 | 586 | } |
1da177e4 LT |
587 | |
588 | void __init paging_init(void) | |
e3cfe529 | 589 | { |
6391af17 | 590 | unsigned long max_zone_pfns[MAX_NR_ZONES]; |
e3cfe529 TG |
591 | int i; |
592 | ||
6391af17 MG |
593 | memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); |
594 | max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN; | |
595 | max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN; | |
596 | max_zone_pfns[ZONE_NORMAL] = end_pfn; | |
d3ee871e | 597 | |
f0a5a58a BP |
598 | sparse_memory_present_with_active_regions(MAX_NUMNODES); |
599 | sparse_init(); | |
d3ee871e | 600 | |
e3cfe529 | 601 | for_each_online_node(i) |
a261670a | 602 | flat_setup_node_zones(i); |
5cb248ab MG |
603 | |
604 | free_area_init_nodes(max_zone_pfns); | |
e3cfe529 | 605 | } |
1da177e4 | 606 | |
2c8c0e6b | 607 | static __init int numa_setup(char *opt) |
e3cfe529 | 608 | { |
2c8c0e6b AK |
609 | if (!opt) |
610 | return -EINVAL; | |
e3cfe529 | 611 | if (!strncmp(opt, "off", 3)) |
1da177e4 LT |
612 | numa_off = 1; |
613 | #ifdef CONFIG_NUMA_EMU | |
8b8ca80e DR |
614 | if (!strncmp(opt, "fake=", 5)) |
615 | cmdline = opt + 5; | |
1da177e4 LT |
616 | #endif |
617 | #ifdef CONFIG_ACPI_NUMA | |
e3cfe529 TG |
618 | if (!strncmp(opt, "noacpi", 6)) |
619 | acpi_numa = -1; | |
620 | if (!strncmp(opt, "hotadd=", 7)) | |
68a3a7fe | 621 | hotadd_percent = simple_strtoul(opt+7, NULL, 10); |
1da177e4 | 622 | #endif |
2c8c0e6b | 623 | return 0; |
e3cfe529 | 624 | } |
2c8c0e6b AK |
625 | early_param("numa", numa_setup); |
626 | ||
05b3cbd8 RT |
627 | /* |
628 | * Setup early cpu_to_node. | |
629 | * | |
630 | * Populate cpu_to_node[] only if x86_cpu_to_apicid[], | |
631 | * and apicid_to_node[] tables have valid entries for a CPU. | |
632 | * This means we skip cpu_to_node[] initialisation for NUMA | |
633 | * emulation and faking node case (when running a kernel compiled | |
634 | * for NUMA on a non NUMA box), which is OK as cpu_to_node[] | |
635 | * is already initialized in a round robin manner at numa_init_array, | |
636 | * prior to this call, and this initialization is good enough | |
637 | * for the fake NUMA cases. | |
638 | */ | |
639 | void __init init_cpu_to_node(void) | |
640 | { | |
641 | int i; | |
e3cfe529 TG |
642 | |
643 | for (i = 0; i < NR_CPUS; i++) { | |
ef97001f | 644 | u16 apicid = x86_cpu_to_apicid_init[i]; |
e3cfe529 | 645 | |
05b3cbd8 RT |
646 | if (apicid == BAD_APICID) |
647 | continue; | |
648 | if (apicid_to_node[apicid] == NUMA_NO_NODE) | |
649 | continue; | |
e3cfe529 | 650 | numa_set_node(i, apicid_to_node[apicid]); |
05b3cbd8 RT |
651 | } |
652 | } | |
653 | ||
cf050132 | 654 |