]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * pSeries NUMA support | |
3 | * | |
4 | * Copyright (C) 2002 Anton Blanchard <anton@au.ibm.com>, IBM | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or | |
7 | * modify it under the terms of the GNU General Public License | |
8 | * as published by the Free Software Foundation; either version | |
9 | * 2 of the License, or (at your option) any later version. | |
10 | */ | |
11 | #include <linux/threads.h> | |
12 | #include <linux/bootmem.h> | |
13 | #include <linux/init.h> | |
14 | #include <linux/mm.h> | |
15 | #include <linux/mmzone.h> | |
16 | #include <linux/module.h> | |
17 | #include <linux/nodemask.h> | |
18 | #include <linux/cpu.h> | |
19 | #include <linux/notifier.h> | |
45fb6cea | 20 | #include <asm/sparsemem.h> |
1da177e4 | 21 | #include <asm/lmb.h> |
cf00a8d1 | 22 | #include <asm/system.h> |
2249ca9d | 23 | #include <asm/smp.h> |
1da177e4 LT |
24 | |
25 | static int numa_enabled = 1; | |
26 | ||
27 | static int numa_debug; | |
28 | #define dbg(args...) if (numa_debug) { printk(KERN_INFO args); } | |
29 | ||
45fb6cea | 30 | int numa_cpu_lookup_table[NR_CPUS]; |
1da177e4 | 31 | cpumask_t numa_cpumask_lookup_table[MAX_NUMNODES]; |
1da177e4 | 32 | struct pglist_data *node_data[MAX_NUMNODES]; |
45fb6cea AB |
33 | |
34 | EXPORT_SYMBOL(numa_cpu_lookup_table); | |
35 | EXPORT_SYMBOL(numa_cpumask_lookup_table); | |
36 | EXPORT_SYMBOL(node_data); | |
37 | ||
38 | static bootmem_data_t __initdata plat_node_bdata[MAX_NUMNODES]; | |
1da177e4 LT |
39 | static int min_common_depth; |
40 | ||
41 | /* | |
45fb6cea | 42 | * We need somewhere to store start/end/node for each region until we have |
1da177e4 LT |
43 | * allocated the real node_data structures. |
44 | */ | |
45fb6cea | 45 | #define MAX_REGIONS (MAX_LMB_REGIONS*2) |
1da177e4 | 46 | static struct { |
45fb6cea AB |
47 | unsigned long start_pfn; |
48 | unsigned long end_pfn; | |
49 | int nid; | |
50 | } init_node_data[MAX_REGIONS] __initdata; | |
1da177e4 | 51 | |
45fb6cea AB |
52 | int __init early_pfn_to_nid(unsigned long pfn) |
53 | { | |
54 | unsigned int i; | |
55 | ||
56 | for (i = 0; init_node_data[i].end_pfn; i++) { | |
57 | unsigned long start_pfn = init_node_data[i].start_pfn; | |
58 | unsigned long end_pfn = init_node_data[i].end_pfn; | |
59 | ||
60 | if ((start_pfn <= pfn) && (pfn < end_pfn)) | |
61 | return init_node_data[i].nid; | |
62 | } | |
63 | ||
64 | return -1; | |
65 | } | |
66 | ||
67 | void __init add_region(unsigned int nid, unsigned long start_pfn, | |
68 | unsigned long pages) | |
69 | { | |
70 | unsigned int i; | |
71 | ||
72 | dbg("add_region nid %d start_pfn 0x%lx pages 0x%lx\n", | |
73 | nid, start_pfn, pages); | |
74 | ||
75 | for (i = 0; init_node_data[i].end_pfn; i++) { | |
76 | if (init_node_data[i].nid != nid) | |
77 | continue; | |
78 | if (init_node_data[i].end_pfn == start_pfn) { | |
79 | init_node_data[i].end_pfn += pages; | |
80 | return; | |
81 | } | |
82 | if (init_node_data[i].start_pfn == (start_pfn + pages)) { | |
83 | init_node_data[i].start_pfn -= pages; | |
84 | return; | |
85 | } | |
86 | } | |
87 | ||
88 | /* | |
89 | * Leave last entry NULL so we dont iterate off the end (we use | |
90 | * entry.end_pfn to terminate the walk). | |
91 | */ | |
92 | if (i >= (MAX_REGIONS - 1)) { | |
93 | printk(KERN_ERR "WARNING: too many memory regions in " | |
94 | "numa code, truncating\n"); | |
95 | return; | |
96 | } | |
97 | ||
98 | init_node_data[i].start_pfn = start_pfn; | |
99 | init_node_data[i].end_pfn = start_pfn + pages; | |
100 | init_node_data[i].nid = nid; | |
101 | } | |
102 | ||
103 | /* We assume init_node_data has no overlapping regions */ | |
104 | void __init get_region(unsigned int nid, unsigned long *start_pfn, | |
105 | unsigned long *end_pfn, unsigned long *pages_present) | |
106 | { | |
107 | unsigned int i; | |
108 | ||
109 | *start_pfn = -1UL; | |
110 | *end_pfn = *pages_present = 0; | |
111 | ||
112 | for (i = 0; init_node_data[i].end_pfn; i++) { | |
113 | if (init_node_data[i].nid != nid) | |
114 | continue; | |
115 | ||
116 | *pages_present += init_node_data[i].end_pfn - | |
117 | init_node_data[i].start_pfn; | |
118 | ||
119 | if (init_node_data[i].start_pfn < *start_pfn) | |
120 | *start_pfn = init_node_data[i].start_pfn; | |
121 | ||
122 | if (init_node_data[i].end_pfn > *end_pfn) | |
123 | *end_pfn = init_node_data[i].end_pfn; | |
124 | } | |
125 | ||
126 | /* We didnt find a matching region, return start/end as 0 */ | |
127 | if (*start_pfn == -1UL) | |
128 | start_pfn = 0; | |
129 | } | |
1da177e4 LT |
130 | |
131 | static inline void map_cpu_to_node(int cpu, int node) | |
132 | { | |
133 | numa_cpu_lookup_table[cpu] = node; | |
45fb6cea AB |
134 | |
135 | if (!(cpu_isset(cpu, numa_cpumask_lookup_table[node]))) | |
1da177e4 | 136 | cpu_set(cpu, numa_cpumask_lookup_table[node]); |
1da177e4 LT |
137 | } |
138 | ||
139 | #ifdef CONFIG_HOTPLUG_CPU | |
140 | static void unmap_cpu_from_node(unsigned long cpu) | |
141 | { | |
142 | int node = numa_cpu_lookup_table[cpu]; | |
143 | ||
144 | dbg("removing cpu %lu from node %d\n", cpu, node); | |
145 | ||
146 | if (cpu_isset(cpu, numa_cpumask_lookup_table[node])) { | |
147 | cpu_clear(cpu, numa_cpumask_lookup_table[node]); | |
1da177e4 LT |
148 | } else { |
149 | printk(KERN_ERR "WARNING: cpu %lu not found in node %d\n", | |
150 | cpu, node); | |
151 | } | |
152 | } | |
153 | #endif /* CONFIG_HOTPLUG_CPU */ | |
154 | ||
45fb6cea | 155 | static struct device_node *find_cpu_node(unsigned int cpu) |
1da177e4 LT |
156 | { |
157 | unsigned int hw_cpuid = get_hard_smp_processor_id(cpu); | |
158 | struct device_node *cpu_node = NULL; | |
159 | unsigned int *interrupt_server, *reg; | |
160 | int len; | |
161 | ||
162 | while ((cpu_node = of_find_node_by_type(cpu_node, "cpu")) != NULL) { | |
163 | /* Try interrupt server first */ | |
164 | interrupt_server = (unsigned int *)get_property(cpu_node, | |
165 | "ibm,ppc-interrupt-server#s", &len); | |
166 | ||
167 | len = len / sizeof(u32); | |
168 | ||
169 | if (interrupt_server && (len > 0)) { | |
170 | while (len--) { | |
171 | if (interrupt_server[len] == hw_cpuid) | |
172 | return cpu_node; | |
173 | } | |
174 | } else { | |
175 | reg = (unsigned int *)get_property(cpu_node, | |
176 | "reg", &len); | |
177 | if (reg && (len > 0) && (reg[0] == hw_cpuid)) | |
178 | return cpu_node; | |
179 | } | |
180 | } | |
181 | ||
182 | return NULL; | |
183 | } | |
184 | ||
185 | /* must hold reference to node during call */ | |
186 | static int *of_get_associativity(struct device_node *dev) | |
187 | { | |
188 | return (unsigned int *)get_property(dev, "ibm,associativity", NULL); | |
189 | } | |
190 | ||
191 | static int of_node_numa_domain(struct device_node *device) | |
192 | { | |
193 | int numa_domain; | |
194 | unsigned int *tmp; | |
195 | ||
196 | if (min_common_depth == -1) | |
197 | return 0; | |
198 | ||
199 | tmp = of_get_associativity(device); | |
200 | if (tmp && (tmp[0] >= min_common_depth)) { | |
201 | numa_domain = tmp[min_common_depth]; | |
202 | } else { | |
203 | dbg("WARNING: no NUMA information for %s\n", | |
204 | device->full_name); | |
205 | numa_domain = 0; | |
206 | } | |
207 | return numa_domain; | |
208 | } | |
209 | ||
210 | /* | |
211 | * In theory, the "ibm,associativity" property may contain multiple | |
212 | * associativity lists because a resource may be multiply connected | |
213 | * into the machine. This resource then has different associativity | |
214 | * characteristics relative to its multiple connections. We ignore | |
215 | * this for now. We also assume that all cpu and memory sets have | |
216 | * their distances represented at a common level. This won't be | |
217 | * true for heirarchical NUMA. | |
218 | * | |
219 | * In any case the ibm,associativity-reference-points should give | |
220 | * the correct depth for a normal NUMA system. | |
221 | * | |
222 | * - Dave Hansen <haveblue@us.ibm.com> | |
223 | */ | |
224 | static int __init find_min_common_depth(void) | |
225 | { | |
226 | int depth; | |
227 | unsigned int *ref_points; | |
228 | struct device_node *rtas_root; | |
229 | unsigned int len; | |
230 | ||
231 | rtas_root = of_find_node_by_path("/rtas"); | |
232 | ||
233 | if (!rtas_root) | |
234 | return -1; | |
235 | ||
236 | /* | |
237 | * this property is 2 32-bit integers, each representing a level of | |
238 | * depth in the associativity nodes. The first is for an SMP | |
239 | * configuration (should be all 0's) and the second is for a normal | |
240 | * NUMA configuration. | |
241 | */ | |
242 | ref_points = (unsigned int *)get_property(rtas_root, | |
243 | "ibm,associativity-reference-points", &len); | |
244 | ||
245 | if ((len >= 1) && ref_points) { | |
246 | depth = ref_points[1]; | |
247 | } else { | |
248 | dbg("WARNING: could not find NUMA " | |
249 | "associativity reference point\n"); | |
250 | depth = -1; | |
251 | } | |
252 | of_node_put(rtas_root); | |
253 | ||
254 | return depth; | |
255 | } | |
256 | ||
257 | static int __init get_mem_addr_cells(void) | |
258 | { | |
259 | struct device_node *memory = NULL; | |
260 | int rc; | |
261 | ||
262 | memory = of_find_node_by_type(memory, "memory"); | |
263 | if (!memory) | |
264 | return 0; /* it won't matter */ | |
265 | ||
266 | rc = prom_n_addr_cells(memory); | |
267 | return rc; | |
268 | } | |
269 | ||
270 | static int __init get_mem_size_cells(void) | |
271 | { | |
272 | struct device_node *memory = NULL; | |
273 | int rc; | |
274 | ||
275 | memory = of_find_node_by_type(memory, "memory"); | |
276 | if (!memory) | |
277 | return 0; /* it won't matter */ | |
278 | rc = prom_n_size_cells(memory); | |
279 | return rc; | |
280 | } | |
281 | ||
45fb6cea | 282 | static unsigned long __init read_n_cells(int n, unsigned int **buf) |
1da177e4 LT |
283 | { |
284 | unsigned long result = 0; | |
285 | ||
286 | while (n--) { | |
287 | result = (result << 32) | **buf; | |
288 | (*buf)++; | |
289 | } | |
290 | return result; | |
291 | } | |
292 | ||
293 | /* | |
294 | * Figure out to which domain a cpu belongs and stick it there. | |
295 | * Return the id of the domain used. | |
296 | */ | |
297 | static int numa_setup_cpu(unsigned long lcpu) | |
298 | { | |
299 | int numa_domain = 0; | |
300 | struct device_node *cpu = find_cpu_node(lcpu); | |
301 | ||
302 | if (!cpu) { | |
303 | WARN_ON(1); | |
304 | goto out; | |
305 | } | |
306 | ||
307 | numa_domain = of_node_numa_domain(cpu); | |
308 | ||
309 | if (numa_domain >= num_online_nodes()) { | |
310 | /* | |
311 | * POWER4 LPAR uses 0xffff as invalid node, | |
312 | * dont warn in this case. | |
313 | */ | |
314 | if (numa_domain != 0xffff) | |
315 | printk(KERN_ERR "WARNING: cpu %ld " | |
316 | "maps to invalid NUMA node %d\n", | |
317 | lcpu, numa_domain); | |
318 | numa_domain = 0; | |
319 | } | |
320 | out: | |
321 | node_set_online(numa_domain); | |
322 | ||
323 | map_cpu_to_node(lcpu, numa_domain); | |
324 | ||
325 | of_node_put(cpu); | |
326 | ||
327 | return numa_domain; | |
328 | } | |
329 | ||
330 | static int cpu_numa_callback(struct notifier_block *nfb, | |
331 | unsigned long action, | |
332 | void *hcpu) | |
333 | { | |
334 | unsigned long lcpu = (unsigned long)hcpu; | |
335 | int ret = NOTIFY_DONE; | |
336 | ||
337 | switch (action) { | |
338 | case CPU_UP_PREPARE: | |
339 | if (min_common_depth == -1 || !numa_enabled) | |
340 | map_cpu_to_node(lcpu, 0); | |
341 | else | |
342 | numa_setup_cpu(lcpu); | |
343 | ret = NOTIFY_OK; | |
344 | break; | |
345 | #ifdef CONFIG_HOTPLUG_CPU | |
346 | case CPU_DEAD: | |
347 | case CPU_UP_CANCELED: | |
348 | unmap_cpu_from_node(lcpu); | |
349 | break; | |
350 | ret = NOTIFY_OK; | |
351 | #endif | |
352 | } | |
353 | return ret; | |
354 | } | |
355 | ||
356 | /* | |
357 | * Check and possibly modify a memory region to enforce the memory limit. | |
358 | * | |
359 | * Returns the size the region should have to enforce the memory limit. | |
360 | * This will either be the original value of size, a truncated value, | |
361 | * or zero. If the returned value of size is 0 the region should be | |
362 | * discarded as it lies wholy above the memory limit. | |
363 | */ | |
45fb6cea AB |
364 | static unsigned long __init numa_enforce_memory_limit(unsigned long start, |
365 | unsigned long size) | |
1da177e4 LT |
366 | { |
367 | /* | |
368 | * We use lmb_end_of_DRAM() in here instead of memory_limit because | |
369 | * we've already adjusted it for the limit and it takes care of | |
370 | * having memory holes below the limit. | |
371 | */ | |
1da177e4 LT |
372 | |
373 | if (! memory_limit) | |
374 | return size; | |
375 | ||
376 | if (start + size <= lmb_end_of_DRAM()) | |
377 | return size; | |
378 | ||
379 | if (start >= lmb_end_of_DRAM()) | |
380 | return 0; | |
381 | ||
382 | return lmb_end_of_DRAM() - start; | |
383 | } | |
384 | ||
385 | static int __init parse_numa_properties(void) | |
386 | { | |
387 | struct device_node *cpu = NULL; | |
388 | struct device_node *memory = NULL; | |
389 | int addr_cells, size_cells; | |
45fb6cea | 390 | int max_domain; |
1da177e4 LT |
391 | unsigned long i; |
392 | ||
393 | if (numa_enabled == 0) { | |
394 | printk(KERN_WARNING "NUMA disabled by user\n"); | |
395 | return -1; | |
396 | } | |
397 | ||
1da177e4 LT |
398 | min_common_depth = find_min_common_depth(); |
399 | ||
400 | dbg("NUMA associativity depth for CPU/Memory: %d\n", min_common_depth); | |
401 | if (min_common_depth < 0) | |
402 | return min_common_depth; | |
403 | ||
404 | max_domain = numa_setup_cpu(boot_cpuid); | |
405 | ||
406 | /* | |
407 | * Even though we connect cpus to numa domains later in SMP init, | |
408 | * we need to know the maximum node id now. This is because each | |
409 | * node id must have NODE_DATA etc backing it. | |
410 | * As a result of hotplug we could still have cpus appear later on | |
411 | * with larger node ids. In that case we force the cpu into node 0. | |
412 | */ | |
413 | for_each_cpu(i) { | |
414 | int numa_domain; | |
415 | ||
416 | cpu = find_cpu_node(i); | |
417 | ||
418 | if (cpu) { | |
419 | numa_domain = of_node_numa_domain(cpu); | |
420 | of_node_put(cpu); | |
421 | ||
422 | if (numa_domain < MAX_NUMNODES && | |
423 | max_domain < numa_domain) | |
424 | max_domain = numa_domain; | |
425 | } | |
426 | } | |
427 | ||
428 | addr_cells = get_mem_addr_cells(); | |
429 | size_cells = get_mem_size_cells(); | |
430 | memory = NULL; | |
431 | while ((memory = of_find_node_by_type(memory, "memory")) != NULL) { | |
432 | unsigned long start; | |
433 | unsigned long size; | |
434 | int numa_domain; | |
435 | int ranges; | |
436 | unsigned int *memcell_buf; | |
437 | unsigned int len; | |
438 | ||
439 | memcell_buf = (unsigned int *)get_property(memory, "reg", &len); | |
440 | if (!memcell_buf || len <= 0) | |
441 | continue; | |
442 | ||
443 | ranges = memory->n_addrs; | |
444 | new_range: | |
445 | /* these are order-sensitive, and modify the buffer pointer */ | |
446 | start = read_n_cells(addr_cells, &memcell_buf); | |
447 | size = read_n_cells(size_cells, &memcell_buf); | |
448 | ||
1da177e4 LT |
449 | numa_domain = of_node_numa_domain(memory); |
450 | ||
451 | if (numa_domain >= MAX_NUMNODES) { | |
452 | if (numa_domain != 0xffff) | |
453 | printk(KERN_ERR "WARNING: memory at %lx maps " | |
454 | "to invalid NUMA node %d\n", start, | |
455 | numa_domain); | |
456 | numa_domain = 0; | |
457 | } | |
458 | ||
459 | if (max_domain < numa_domain) | |
460 | max_domain = numa_domain; | |
461 | ||
45fb6cea | 462 | if (!(size = numa_enforce_memory_limit(start, size))) { |
1da177e4 LT |
463 | if (--ranges) |
464 | goto new_range; | |
465 | else | |
466 | continue; | |
467 | } | |
468 | ||
45fb6cea AB |
469 | add_region(numa_domain, start >> PAGE_SHIFT, |
470 | size >> PAGE_SHIFT); | |
1da177e4 LT |
471 | |
472 | if (--ranges) | |
473 | goto new_range; | |
474 | } | |
475 | ||
476 | for (i = 0; i <= max_domain; i++) | |
477 | node_set_online(i); | |
478 | ||
479 | return 0; | |
480 | } | |
481 | ||
482 | static void __init setup_nonnuma(void) | |
483 | { | |
484 | unsigned long top_of_ram = lmb_end_of_DRAM(); | |
485 | unsigned long total_ram = lmb_phys_mem_size(); | |
fb6d73d3 | 486 | unsigned int i; |
1da177e4 LT |
487 | |
488 | printk(KERN_INFO "Top of RAM: 0x%lx, Total RAM: 0x%lx\n", | |
489 | top_of_ram, total_ram); | |
490 | printk(KERN_INFO "Memory hole size: %ldMB\n", | |
491 | (top_of_ram - total_ram) >> 20); | |
492 | ||
1da177e4 | 493 | map_cpu_to_node(boot_cpuid, 0); |
fb6d73d3 PM |
494 | for (i = 0; i < lmb.memory.cnt; ++i) |
495 | add_region(0, lmb.memory.region[i].base >> PAGE_SHIFT, | |
496 | lmb_size_pages(&lmb.memory, i)); | |
1da177e4 | 497 | node_set_online(0); |
1da177e4 LT |
498 | } |
499 | ||
500 | static void __init dump_numa_topology(void) | |
501 | { | |
502 | unsigned int node; | |
503 | unsigned int count; | |
504 | ||
505 | if (min_common_depth == -1 || !numa_enabled) | |
506 | return; | |
507 | ||
508 | for_each_online_node(node) { | |
509 | unsigned long i; | |
510 | ||
511 | printk(KERN_INFO "Node %d Memory:", node); | |
512 | ||
513 | count = 0; | |
514 | ||
45fb6cea AB |
515 | for (i = 0; i < lmb_end_of_DRAM(); |
516 | i += (1 << SECTION_SIZE_BITS)) { | |
517 | if (early_pfn_to_nid(i >> PAGE_SHIFT) == node) { | |
1da177e4 LT |
518 | if (count == 0) |
519 | printk(" 0x%lx", i); | |
520 | ++count; | |
521 | } else { | |
522 | if (count > 0) | |
523 | printk("-0x%lx", i); | |
524 | count = 0; | |
525 | } | |
526 | } | |
527 | ||
528 | if (count > 0) | |
529 | printk("-0x%lx", i); | |
530 | printk("\n"); | |
531 | } | |
532 | return; | |
533 | } | |
534 | ||
535 | /* | |
536 | * Allocate some memory, satisfying the lmb or bootmem allocator where | |
537 | * required. nid is the preferred node and end is the physical address of | |
538 | * the highest address in the node. | |
539 | * | |
540 | * Returns the physical address of the memory. | |
541 | */ | |
45fb6cea AB |
542 | static void __init *careful_allocation(int nid, unsigned long size, |
543 | unsigned long align, | |
544 | unsigned long end_pfn) | |
1da177e4 | 545 | { |
45fb6cea AB |
546 | int new_nid; |
547 | unsigned long ret = lmb_alloc_base(size, align, end_pfn << PAGE_SHIFT); | |
1da177e4 LT |
548 | |
549 | /* retry over all memory */ | |
550 | if (!ret) | |
551 | ret = lmb_alloc_base(size, align, lmb_end_of_DRAM()); | |
552 | ||
553 | if (!ret) | |
554 | panic("numa.c: cannot allocate %lu bytes on node %d", | |
555 | size, nid); | |
556 | ||
557 | /* | |
558 | * If the memory came from a previously allocated node, we must | |
559 | * retry with the bootmem allocator. | |
560 | */ | |
45fb6cea AB |
561 | new_nid = early_pfn_to_nid(ret >> PAGE_SHIFT); |
562 | if (new_nid < nid) { | |
563 | ret = (unsigned long)__alloc_bootmem_node(NODE_DATA(new_nid), | |
1da177e4 LT |
564 | size, align, 0); |
565 | ||
566 | if (!ret) | |
567 | panic("numa.c: cannot allocate %lu bytes on node %d", | |
45fb6cea | 568 | size, new_nid); |
1da177e4 | 569 | |
45fb6cea | 570 | ret = __pa(ret); |
1da177e4 LT |
571 | |
572 | dbg("alloc_bootmem %lx %lx\n", ret, size); | |
573 | } | |
574 | ||
45fb6cea | 575 | return (void *)ret; |
1da177e4 LT |
576 | } |
577 | ||
578 | void __init do_init_bootmem(void) | |
579 | { | |
580 | int nid; | |
45fb6cea | 581 | unsigned int i; |
1da177e4 LT |
582 | static struct notifier_block ppc64_numa_nb = { |
583 | .notifier_call = cpu_numa_callback, | |
584 | .priority = 1 /* Must run before sched domains notifier. */ | |
585 | }; | |
586 | ||
587 | min_low_pfn = 0; | |
588 | max_low_pfn = lmb_end_of_DRAM() >> PAGE_SHIFT; | |
589 | max_pfn = max_low_pfn; | |
590 | ||
591 | if (parse_numa_properties()) | |
592 | setup_nonnuma(); | |
593 | else | |
594 | dump_numa_topology(); | |
595 | ||
596 | register_cpu_notifier(&ppc64_numa_nb); | |
597 | ||
598 | for_each_online_node(nid) { | |
45fb6cea | 599 | unsigned long start_pfn, end_pfn, pages_present; |
1da177e4 LT |
600 | unsigned long bootmem_paddr; |
601 | unsigned long bootmap_pages; | |
602 | ||
45fb6cea | 603 | get_region(nid, &start_pfn, &end_pfn, &pages_present); |
1da177e4 LT |
604 | |
605 | /* Allocate the node structure node local if possible */ | |
45fb6cea | 606 | NODE_DATA(nid) = careful_allocation(nid, |
1da177e4 | 607 | sizeof(struct pglist_data), |
45fb6cea AB |
608 | SMP_CACHE_BYTES, end_pfn); |
609 | NODE_DATA(nid) = __va(NODE_DATA(nid)); | |
1da177e4 LT |
610 | memset(NODE_DATA(nid), 0, sizeof(struct pglist_data)); |
611 | ||
612 | dbg("node %d\n", nid); | |
613 | dbg("NODE_DATA() = %p\n", NODE_DATA(nid)); | |
614 | ||
615 | NODE_DATA(nid)->bdata = &plat_node_bdata[nid]; | |
45fb6cea AB |
616 | NODE_DATA(nid)->node_start_pfn = start_pfn; |
617 | NODE_DATA(nid)->node_spanned_pages = end_pfn - start_pfn; | |
1da177e4 LT |
618 | |
619 | if (NODE_DATA(nid)->node_spanned_pages == 0) | |
620 | continue; | |
621 | ||
45fb6cea AB |
622 | dbg("start_paddr = %lx\n", start_pfn << PAGE_SHIFT); |
623 | dbg("end_paddr = %lx\n", end_pfn << PAGE_SHIFT); | |
1da177e4 | 624 | |
45fb6cea AB |
625 | bootmap_pages = bootmem_bootmap_pages(end_pfn - start_pfn); |
626 | bootmem_paddr = (unsigned long)careful_allocation(nid, | |
627 | bootmap_pages << PAGE_SHIFT, | |
628 | PAGE_SIZE, end_pfn); | |
629 | memset(__va(bootmem_paddr), 0, bootmap_pages << PAGE_SHIFT); | |
1da177e4 | 630 | |
1da177e4 LT |
631 | dbg("bootmap_paddr = %lx\n", bootmem_paddr); |
632 | ||
633 | init_bootmem_node(NODE_DATA(nid), bootmem_paddr >> PAGE_SHIFT, | |
45fb6cea | 634 | start_pfn, end_pfn); |
1da177e4 | 635 | |
45fb6cea AB |
636 | /* Add free regions on this node */ |
637 | for (i = 0; init_node_data[i].end_pfn; i++) { | |
638 | unsigned long start, end; | |
1da177e4 | 639 | |
45fb6cea | 640 | if (init_node_data[i].nid != nid) |
1da177e4 LT |
641 | continue; |
642 | ||
45fb6cea AB |
643 | start = init_node_data[i].start_pfn << PAGE_SHIFT; |
644 | end = init_node_data[i].end_pfn << PAGE_SHIFT; | |
1da177e4 | 645 | |
45fb6cea AB |
646 | dbg("free_bootmem %lx %lx\n", start, end - start); |
647 | free_bootmem_node(NODE_DATA(nid), start, end - start); | |
1da177e4 LT |
648 | } |
649 | ||
45fb6cea | 650 | /* Mark reserved regions on this node */ |
1da177e4 | 651 | for (i = 0; i < lmb.reserved.cnt; i++) { |
180379dc | 652 | unsigned long physbase = lmb.reserved.region[i].base; |
1da177e4 | 653 | unsigned long size = lmb.reserved.region[i].size; |
45fb6cea AB |
654 | unsigned long start_paddr = start_pfn << PAGE_SHIFT; |
655 | unsigned long end_paddr = end_pfn << PAGE_SHIFT; | |
1da177e4 | 656 | |
45fb6cea AB |
657 | if (early_pfn_to_nid(physbase >> PAGE_SHIFT) != nid && |
658 | early_pfn_to_nid((physbase+size-1) >> PAGE_SHIFT) != nid) | |
1da177e4 LT |
659 | continue; |
660 | ||
661 | if (physbase < end_paddr && | |
662 | (physbase+size) > start_paddr) { | |
663 | /* overlaps */ | |
664 | if (physbase < start_paddr) { | |
665 | size -= start_paddr - physbase; | |
666 | physbase = start_paddr; | |
667 | } | |
668 | ||
669 | if (size > end_paddr - physbase) | |
670 | size = end_paddr - physbase; | |
671 | ||
672 | dbg("reserve_bootmem %lx %lx\n", physbase, | |
673 | size); | |
674 | reserve_bootmem_node(NODE_DATA(nid), physbase, | |
675 | size); | |
676 | } | |
677 | } | |
802f192e | 678 | |
45fb6cea AB |
679 | /* Add regions into sparsemem */ |
680 | for (i = 0; init_node_data[i].end_pfn; i++) { | |
681 | unsigned long start, end; | |
682 | ||
683 | if (init_node_data[i].nid != nid) | |
802f192e BP |
684 | continue; |
685 | ||
45fb6cea AB |
686 | start = init_node_data[i].start_pfn; |
687 | end = init_node_data[i].end_pfn; | |
802f192e | 688 | |
45fb6cea | 689 | memory_present(nid, start, end); |
802f192e | 690 | } |
1da177e4 LT |
691 | } |
692 | } | |
693 | ||
694 | void __init paging_init(void) | |
695 | { | |
696 | unsigned long zones_size[MAX_NR_ZONES]; | |
697 | unsigned long zholes_size[MAX_NR_ZONES]; | |
698 | int nid; | |
699 | ||
700 | memset(zones_size, 0, sizeof(zones_size)); | |
701 | memset(zholes_size, 0, sizeof(zholes_size)); | |
702 | ||
703 | for_each_online_node(nid) { | |
45fb6cea | 704 | unsigned long start_pfn, end_pfn, pages_present; |
1da177e4 | 705 | |
45fb6cea | 706 | get_region(nid, &start_pfn, &end_pfn, &pages_present); |
1da177e4 LT |
707 | |
708 | zones_size[ZONE_DMA] = end_pfn - start_pfn; | |
45fb6cea | 709 | zholes_size[ZONE_DMA] = zones_size[ZONE_DMA] - pages_present; |
1da177e4 LT |
710 | |
711 | dbg("free_area_init node %d %lx %lx (hole: %lx)\n", nid, | |
712 | zones_size[ZONE_DMA], start_pfn, zholes_size[ZONE_DMA]); | |
713 | ||
45fb6cea AB |
714 | free_area_init_node(nid, NODE_DATA(nid), zones_size, start_pfn, |
715 | zholes_size); | |
1da177e4 LT |
716 | } |
717 | } | |
718 | ||
719 | static int __init early_numa(char *p) | |
720 | { | |
721 | if (!p) | |
722 | return 0; | |
723 | ||
724 | if (strstr(p, "off")) | |
725 | numa_enabled = 0; | |
726 | ||
727 | if (strstr(p, "debug")) | |
728 | numa_debug = 1; | |
729 | ||
730 | return 0; | |
731 | } | |
732 | early_param("numa", early_numa); |