blob: 1fb11bbe1ace5ae3443bd6c8d2f1635128bf164f [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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>
Anton Blanchard45fb6ce2005-11-11 14:22:35 +110020#include <asm/sparsemem.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <asm/lmb.h>
Paul Mackerrascf00a8d2005-10-31 13:07:02 +110022#include <asm/system.h>
Paul Mackerras2249ca92005-11-07 13:18:13 +110023#include <asm/smp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024
25static int numa_enabled = 1;
26
27static int numa_debug;
28#define dbg(args...) if (numa_debug) { printk(KERN_INFO args); }
29
Anton Blanchard45fb6ce2005-11-11 14:22:35 +110030int numa_cpu_lookup_table[NR_CPUS];
Linus Torvalds1da177e2005-04-16 15:20:36 -070031cpumask_t numa_cpumask_lookup_table[MAX_NUMNODES];
Linus Torvalds1da177e2005-04-16 15:20:36 -070032struct pglist_data *node_data[MAX_NUMNODES];
Anton Blanchard45fb6ce2005-11-11 14:22:35 +110033
34EXPORT_SYMBOL(numa_cpu_lookup_table);
35EXPORT_SYMBOL(numa_cpumask_lookup_table);
36EXPORT_SYMBOL(node_data);
37
38static bootmem_data_t __initdata plat_node_bdata[MAX_NUMNODES];
Linus Torvalds1da177e2005-04-16 15:20:36 -070039static int min_common_depth;
Mike Kravetz237a0982005-12-05 12:06:42 -080040static int n_mem_addr_cells, n_mem_size_cells;
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
42/*
Anton Blanchard45fb6ce2005-11-11 14:22:35 +110043 * We need somewhere to store start/end/node for each region until we have
Linus Torvalds1da177e2005-04-16 15:20:36 -070044 * allocated the real node_data structures.
45 */
Anton Blanchard45fb6ce2005-11-11 14:22:35 +110046#define MAX_REGIONS (MAX_LMB_REGIONS*2)
Linus Torvalds1da177e2005-04-16 15:20:36 -070047static struct {
Anton Blanchard45fb6ce2005-11-11 14:22:35 +110048 unsigned long start_pfn;
49 unsigned long end_pfn;
50 int nid;
51} init_node_data[MAX_REGIONS] __initdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
Anton Blanchard45fb6ce2005-11-11 14:22:35 +110053int __init early_pfn_to_nid(unsigned long pfn)
54{
55 unsigned int i;
56
57 for (i = 0; init_node_data[i].end_pfn; i++) {
58 unsigned long start_pfn = init_node_data[i].start_pfn;
59 unsigned long end_pfn = init_node_data[i].end_pfn;
60
61 if ((start_pfn <= pfn) && (pfn < end_pfn))
62 return init_node_data[i].nid;
63 }
64
65 return -1;
66}
67
68void __init add_region(unsigned int nid, unsigned long start_pfn,
69 unsigned long pages)
70{
71 unsigned int i;
72
73 dbg("add_region nid %d start_pfn 0x%lx pages 0x%lx\n",
74 nid, start_pfn, pages);
75
76 for (i = 0; init_node_data[i].end_pfn; i++) {
77 if (init_node_data[i].nid != nid)
78 continue;
79 if (init_node_data[i].end_pfn == start_pfn) {
80 init_node_data[i].end_pfn += pages;
81 return;
82 }
83 if (init_node_data[i].start_pfn == (start_pfn + pages)) {
84 init_node_data[i].start_pfn -= pages;
85 return;
86 }
87 }
88
89 /*
90 * Leave last entry NULL so we dont iterate off the end (we use
91 * entry.end_pfn to terminate the walk).
92 */
93 if (i >= (MAX_REGIONS - 1)) {
94 printk(KERN_ERR "WARNING: too many memory regions in "
95 "numa code, truncating\n");
96 return;
97 }
98
99 init_node_data[i].start_pfn = start_pfn;
100 init_node_data[i].end_pfn = start_pfn + pages;
101 init_node_data[i].nid = nid;
102}
103
104/* We assume init_node_data has no overlapping regions */
105void __init get_region(unsigned int nid, unsigned long *start_pfn,
106 unsigned long *end_pfn, unsigned long *pages_present)
107{
108 unsigned int i;
109
110 *start_pfn = -1UL;
111 *end_pfn = *pages_present = 0;
112
113 for (i = 0; init_node_data[i].end_pfn; i++) {
114 if (init_node_data[i].nid != nid)
115 continue;
116
117 *pages_present += init_node_data[i].end_pfn -
118 init_node_data[i].start_pfn;
119
120 if (init_node_data[i].start_pfn < *start_pfn)
121 *start_pfn = init_node_data[i].start_pfn;
122
123 if (init_node_data[i].end_pfn > *end_pfn)
124 *end_pfn = init_node_data[i].end_pfn;
125 }
126
127 /* We didnt find a matching region, return start/end as 0 */
128 if (*start_pfn == -1UL)
Mike Kravetz6d91bb92005-12-07 13:07:23 -0800129 *start_pfn = 0;
Anton Blanchard45fb6ce2005-11-11 14:22:35 +1100130}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131
Nathan Lynch2e5ce392006-03-20 18:35:15 -0600132static void __cpuinit map_cpu_to_node(int cpu, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133{
134 numa_cpu_lookup_table[cpu] = node;
Anton Blanchard45fb6ce2005-11-11 14:22:35 +1100135
Nathan Lynchbf4b85b2006-03-20 18:34:45 -0600136 dbg("adding cpu %d to node %d\n", cpu, node);
137
Anton Blanchard45fb6ce2005-11-11 14:22:35 +1100138 if (!(cpu_isset(cpu, numa_cpumask_lookup_table[node])))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 cpu_set(cpu, numa_cpumask_lookup_table[node]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140}
141
142#ifdef CONFIG_HOTPLUG_CPU
143static void unmap_cpu_from_node(unsigned long cpu)
144{
145 int node = numa_cpu_lookup_table[cpu];
146
147 dbg("removing cpu %lu from node %d\n", cpu, node);
148
149 if (cpu_isset(cpu, numa_cpumask_lookup_table[node])) {
150 cpu_clear(cpu, numa_cpumask_lookup_table[node]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151 } else {
152 printk(KERN_ERR "WARNING: cpu %lu not found in node %d\n",
153 cpu, node);
154 }
155}
156#endif /* CONFIG_HOTPLUG_CPU */
157
Nathan Lynch2e5ce392006-03-20 18:35:15 -0600158static struct device_node * __cpuinit find_cpu_node(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159{
160 unsigned int hw_cpuid = get_hard_smp_processor_id(cpu);
161 struct device_node *cpu_node = NULL;
162 unsigned int *interrupt_server, *reg;
163 int len;
164
165 while ((cpu_node = of_find_node_by_type(cpu_node, "cpu")) != NULL) {
166 /* Try interrupt server first */
167 interrupt_server = (unsigned int *)get_property(cpu_node,
168 "ibm,ppc-interrupt-server#s", &len);
169
170 len = len / sizeof(u32);
171
172 if (interrupt_server && (len > 0)) {
173 while (len--) {
174 if (interrupt_server[len] == hw_cpuid)
175 return cpu_node;
176 }
177 } else {
178 reg = (unsigned int *)get_property(cpu_node,
179 "reg", &len);
180 if (reg && (len > 0) && (reg[0] == hw_cpuid))
181 return cpu_node;
182 }
183 }
184
185 return NULL;
186}
187
188/* must hold reference to node during call */
189static int *of_get_associativity(struct device_node *dev)
190{
191 return (unsigned int *)get_property(dev, "ibm,associativity", NULL);
192}
193
194static int of_node_numa_domain(struct device_node *device)
195{
196 int numa_domain;
197 unsigned int *tmp;
198
199 if (min_common_depth == -1)
200 return 0;
201
202 tmp = of_get_associativity(device);
203 if (tmp && (tmp[0] >= min_common_depth)) {
204 numa_domain = tmp[min_common_depth];
205 } else {
206 dbg("WARNING: no NUMA information for %s\n",
207 device->full_name);
208 numa_domain = 0;
209 }
210 return numa_domain;
211}
212
213/*
214 * In theory, the "ibm,associativity" property may contain multiple
215 * associativity lists because a resource may be multiply connected
216 * into the machine. This resource then has different associativity
217 * characteristics relative to its multiple connections. We ignore
218 * this for now. We also assume that all cpu and memory sets have
219 * their distances represented at a common level. This won't be
220 * true for heirarchical NUMA.
221 *
222 * In any case the ibm,associativity-reference-points should give
223 * the correct depth for a normal NUMA system.
224 *
225 * - Dave Hansen <haveblue@us.ibm.com>
226 */
227static int __init find_min_common_depth(void)
228{
229 int depth;
230 unsigned int *ref_points;
231 struct device_node *rtas_root;
232 unsigned int len;
233
234 rtas_root = of_find_node_by_path("/rtas");
235
236 if (!rtas_root)
237 return -1;
238
239 /*
240 * this property is 2 32-bit integers, each representing a level of
241 * depth in the associativity nodes. The first is for an SMP
242 * configuration (should be all 0's) and the second is for a normal
243 * NUMA configuration.
244 */
245 ref_points = (unsigned int *)get_property(rtas_root,
246 "ibm,associativity-reference-points", &len);
247
248 if ((len >= 1) && ref_points) {
249 depth = ref_points[1];
250 } else {
Nathan Lynchbf4b85b2006-03-20 18:34:45 -0600251 dbg("NUMA: ibm,associativity-reference-points not found.\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 depth = -1;
253 }
254 of_node_put(rtas_root);
255
256 return depth;
257}
258
Mike Kravetz84c9fdd2005-11-30 13:47:23 -0800259static void __init get_n_mem_cells(int *n_addr_cells, int *n_size_cells)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260{
261 struct device_node *memory = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262
263 memory = of_find_node_by_type(memory, "memory");
Paul Mackerras54c23312005-12-05 15:50:39 +1100264 if (!memory)
Mike Kravetz84c9fdd2005-11-30 13:47:23 -0800265 panic("numa.c: No memory nodes found!");
Paul Mackerras54c23312005-12-05 15:50:39 +1100266
Mike Kravetz84c9fdd2005-11-30 13:47:23 -0800267 *n_addr_cells = prom_n_addr_cells(memory);
268 *n_size_cells = prom_n_size_cells(memory);
269 of_node_put(memory);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270}
271
Mike Kravetz237a0982005-12-05 12:06:42 -0800272static unsigned long __devinit read_n_cells(int n, unsigned int **buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273{
274 unsigned long result = 0;
275
276 while (n--) {
277 result = (result << 32) | **buf;
278 (*buf)++;
279 }
280 return result;
281}
282
283/*
284 * Figure out to which domain a cpu belongs and stick it there.
285 * Return the id of the domain used.
286 */
Nathan Lynch2e5ce392006-03-20 18:35:15 -0600287static int __cpuinit numa_setup_cpu(unsigned long lcpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288{
289 int numa_domain = 0;
290 struct device_node *cpu = find_cpu_node(lcpu);
291
292 if (!cpu) {
293 WARN_ON(1);
294 goto out;
295 }
296
297 numa_domain = of_node_numa_domain(cpu);
298
299 if (numa_domain >= num_online_nodes()) {
300 /*
301 * POWER4 LPAR uses 0xffff as invalid node,
302 * dont warn in this case.
303 */
304 if (numa_domain != 0xffff)
305 printk(KERN_ERR "WARNING: cpu %ld "
306 "maps to invalid NUMA node %d\n",
307 lcpu, numa_domain);
308 numa_domain = 0;
309 }
310out:
311 node_set_online(numa_domain);
312
313 map_cpu_to_node(lcpu, numa_domain);
314
315 of_node_put(cpu);
316
317 return numa_domain;
318}
319
320static int cpu_numa_callback(struct notifier_block *nfb,
321 unsigned long action,
322 void *hcpu)
323{
324 unsigned long lcpu = (unsigned long)hcpu;
325 int ret = NOTIFY_DONE;
326
327 switch (action) {
328 case CPU_UP_PREPARE:
329 if (min_common_depth == -1 || !numa_enabled)
330 map_cpu_to_node(lcpu, 0);
331 else
332 numa_setup_cpu(lcpu);
333 ret = NOTIFY_OK;
334 break;
335#ifdef CONFIG_HOTPLUG_CPU
336 case CPU_DEAD:
337 case CPU_UP_CANCELED:
338 unmap_cpu_from_node(lcpu);
339 break;
340 ret = NOTIFY_OK;
341#endif
342 }
343 return ret;
344}
345
346/*
347 * Check and possibly modify a memory region to enforce the memory limit.
348 *
349 * Returns the size the region should have to enforce the memory limit.
350 * This will either be the original value of size, a truncated value,
351 * or zero. If the returned value of size is 0 the region should be
352 * discarded as it lies wholy above the memory limit.
353 */
Anton Blanchard45fb6ce2005-11-11 14:22:35 +1100354static unsigned long __init numa_enforce_memory_limit(unsigned long start,
355 unsigned long size)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356{
357 /*
358 * We use lmb_end_of_DRAM() in here instead of memory_limit because
359 * we've already adjusted it for the limit and it takes care of
360 * having memory holes below the limit.
361 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362
363 if (! memory_limit)
364 return size;
365
366 if (start + size <= lmb_end_of_DRAM())
367 return size;
368
369 if (start >= lmb_end_of_DRAM())
370 return 0;
371
372 return lmb_end_of_DRAM() - start;
373}
374
375static int __init parse_numa_properties(void)
376{
377 struct device_node *cpu = NULL;
378 struct device_node *memory = NULL;
Nathan Lynchc08888c2006-03-20 18:34:15 -0600379 int max_domain = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 unsigned long i;
381
382 if (numa_enabled == 0) {
383 printk(KERN_WARNING "NUMA disabled by user\n");
384 return -1;
385 }
386
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 min_common_depth = find_min_common_depth();
388
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 if (min_common_depth < 0)
390 return min_common_depth;
391
Nathan Lynchbf4b85b2006-03-20 18:34:45 -0600392 dbg("NUMA associativity depth for CPU/Memory: %d\n", min_common_depth);
393
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 /*
395 * Even though we connect cpus to numa domains later in SMP init,
396 * we need to know the maximum node id now. This is because each
397 * node id must have NODE_DATA etc backing it.
398 * As a result of hotplug we could still have cpus appear later on
399 * with larger node ids. In that case we force the cpu into node 0.
400 */
401 for_each_cpu(i) {
402 int numa_domain;
403
404 cpu = find_cpu_node(i);
405
406 if (cpu) {
407 numa_domain = of_node_numa_domain(cpu);
408 of_node_put(cpu);
409
410 if (numa_domain < MAX_NUMNODES &&
411 max_domain < numa_domain)
412 max_domain = numa_domain;
413 }
414 }
415
Mike Kravetz237a0982005-12-05 12:06:42 -0800416 get_n_mem_cells(&n_mem_addr_cells, &n_mem_size_cells);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 memory = NULL;
418 while ((memory = of_find_node_by_type(memory, "memory")) != NULL) {
419 unsigned long start;
420 unsigned long size;
421 int numa_domain;
422 int ranges;
423 unsigned int *memcell_buf;
424 unsigned int len;
425
Michael Ellermanba759482005-12-04 18:39:55 +1100426 memcell_buf = (unsigned int *)get_property(memory,
427 "linux,usable-memory", &len);
428 if (!memcell_buf || len <= 0)
429 memcell_buf =
430 (unsigned int *)get_property(memory, "reg",
431 &len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 if (!memcell_buf || len <= 0)
433 continue;
434
Benjamin Herrenschmidtcc5d0182005-12-13 18:01:21 +1100435 /* ranges in cell */
436 ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437new_range:
438 /* these are order-sensitive, and modify the buffer pointer */
Mike Kravetz237a0982005-12-05 12:06:42 -0800439 start = read_n_cells(n_mem_addr_cells, &memcell_buf);
440 size = read_n_cells(n_mem_size_cells, &memcell_buf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 numa_domain = of_node_numa_domain(memory);
443
444 if (numa_domain >= MAX_NUMNODES) {
445 if (numa_domain != 0xffff)
446 printk(KERN_ERR "WARNING: memory at %lx maps "
447 "to invalid NUMA node %d\n", start,
448 numa_domain);
449 numa_domain = 0;
450 }
451
452 if (max_domain < numa_domain)
453 max_domain = numa_domain;
454
Anton Blanchard45fb6ce2005-11-11 14:22:35 +1100455 if (!(size = numa_enforce_memory_limit(start, size))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 if (--ranges)
457 goto new_range;
458 else
459 continue;
460 }
461
Anton Blanchard45fb6ce2005-11-11 14:22:35 +1100462 add_region(numa_domain, start >> PAGE_SHIFT,
463 size >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464
465 if (--ranges)
466 goto new_range;
467 }
468
469 for (i = 0; i <= max_domain; i++)
470 node_set_online(i);
471
Nathan Lynchc08888c2006-03-20 18:34:15 -0600472 max_domain = numa_setup_cpu(boot_cpuid);
473
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 return 0;
475}
476
477static void __init setup_nonnuma(void)
478{
479 unsigned long top_of_ram = lmb_end_of_DRAM();
480 unsigned long total_ram = lmb_phys_mem_size();
Paul Mackerrasfb6d73d2005-11-16 11:43:26 +1100481 unsigned int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482
483 printk(KERN_INFO "Top of RAM: 0x%lx, Total RAM: 0x%lx\n",
484 top_of_ram, total_ram);
485 printk(KERN_INFO "Memory hole size: %ldMB\n",
486 (top_of_ram - total_ram) >> 20);
487
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 map_cpu_to_node(boot_cpuid, 0);
Paul Mackerrasfb6d73d2005-11-16 11:43:26 +1100489 for (i = 0; i < lmb.memory.cnt; ++i)
490 add_region(0, lmb.memory.region[i].base >> PAGE_SHIFT,
491 lmb_size_pages(&lmb.memory, i));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 node_set_online(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493}
494
Anton Blanchard4b703a22005-12-13 06:56:47 +1100495void __init dump_numa_cpu_topology(void)
496{
497 unsigned int node;
498 unsigned int cpu, count;
499
500 if (min_common_depth == -1 || !numa_enabled)
501 return;
502
503 for_each_online_node(node) {
504 printk(KERN_INFO "Node %d CPUs:", node);
505
506 count = 0;
507 /*
508 * If we used a CPU iterator here we would miss printing
509 * the holes in the cpumap.
510 */
511 for (cpu = 0; cpu < NR_CPUS; cpu++) {
512 if (cpu_isset(cpu, numa_cpumask_lookup_table[node])) {
513 if (count == 0)
514 printk(" %u", cpu);
515 ++count;
516 } else {
517 if (count > 1)
518 printk("-%u", cpu - 1);
519 count = 0;
520 }
521 }
522
523 if (count > 1)
524 printk("-%u", NR_CPUS - 1);
525 printk("\n");
526 }
527}
528
529static void __init dump_numa_memory_topology(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530{
531 unsigned int node;
532 unsigned int count;
533
534 if (min_common_depth == -1 || !numa_enabled)
535 return;
536
537 for_each_online_node(node) {
538 unsigned long i;
539
540 printk(KERN_INFO "Node %d Memory:", node);
541
542 count = 0;
543
Anton Blanchard45fb6ce2005-11-11 14:22:35 +1100544 for (i = 0; i < lmb_end_of_DRAM();
545 i += (1 << SECTION_SIZE_BITS)) {
546 if (early_pfn_to_nid(i >> PAGE_SHIFT) == node) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 if (count == 0)
548 printk(" 0x%lx", i);
549 ++count;
550 } else {
551 if (count > 0)
552 printk("-0x%lx", i);
553 count = 0;
554 }
555 }
556
557 if (count > 0)
558 printk("-0x%lx", i);
559 printk("\n");
560 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561}
562
563/*
564 * Allocate some memory, satisfying the lmb or bootmem allocator where
565 * required. nid is the preferred node and end is the physical address of
566 * the highest address in the node.
567 *
568 * Returns the physical address of the memory.
569 */
Anton Blanchard45fb6ce2005-11-11 14:22:35 +1100570static void __init *careful_allocation(int nid, unsigned long size,
571 unsigned long align,
572 unsigned long end_pfn)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573{
Anton Blanchard45fb6ce2005-11-11 14:22:35 +1100574 int new_nid;
Michael Ellermand7a5b2f2006-01-25 21:31:28 +1300575 unsigned long ret = __lmb_alloc_base(size, align, end_pfn << PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576
577 /* retry over all memory */
578 if (!ret)
Michael Ellermand7a5b2f2006-01-25 21:31:28 +1300579 ret = __lmb_alloc_base(size, align, lmb_end_of_DRAM());
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580
581 if (!ret)
582 panic("numa.c: cannot allocate %lu bytes on node %d",
583 size, nid);
584
585 /*
586 * If the memory came from a previously allocated node, we must
587 * retry with the bootmem allocator.
588 */
Anton Blanchard45fb6ce2005-11-11 14:22:35 +1100589 new_nid = early_pfn_to_nid(ret >> PAGE_SHIFT);
590 if (new_nid < nid) {
591 ret = (unsigned long)__alloc_bootmem_node(NODE_DATA(new_nid),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 size, align, 0);
593
594 if (!ret)
595 panic("numa.c: cannot allocate %lu bytes on node %d",
Anton Blanchard45fb6ce2005-11-11 14:22:35 +1100596 size, new_nid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597
Anton Blanchard45fb6ce2005-11-11 14:22:35 +1100598 ret = __pa(ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599
600 dbg("alloc_bootmem %lx %lx\n", ret, size);
601 }
602
Anton Blanchard45fb6ce2005-11-11 14:22:35 +1100603 return (void *)ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604}
605
606void __init do_init_bootmem(void)
607{
608 int nid;
Anton Blanchard45fb6ce2005-11-11 14:22:35 +1100609 unsigned int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 static struct notifier_block ppc64_numa_nb = {
611 .notifier_call = cpu_numa_callback,
612 .priority = 1 /* Must run before sched domains notifier. */
613 };
614
615 min_low_pfn = 0;
616 max_low_pfn = lmb_end_of_DRAM() >> PAGE_SHIFT;
617 max_pfn = max_low_pfn;
618
619 if (parse_numa_properties())
620 setup_nonnuma();
621 else
Anton Blanchard4b703a22005-12-13 06:56:47 +1100622 dump_numa_memory_topology();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623
624 register_cpu_notifier(&ppc64_numa_nb);
625
626 for_each_online_node(nid) {
Anton Blanchard45fb6ce2005-11-11 14:22:35 +1100627 unsigned long start_pfn, end_pfn, pages_present;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 unsigned long bootmem_paddr;
629 unsigned long bootmap_pages;
630
Anton Blanchard45fb6ce2005-11-11 14:22:35 +1100631 get_region(nid, &start_pfn, &end_pfn, &pages_present);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632
633 /* Allocate the node structure node local if possible */
Anton Blanchard45fb6ce2005-11-11 14:22:35 +1100634 NODE_DATA(nid) = careful_allocation(nid,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 sizeof(struct pglist_data),
Anton Blanchard45fb6ce2005-11-11 14:22:35 +1100636 SMP_CACHE_BYTES, end_pfn);
637 NODE_DATA(nid) = __va(NODE_DATA(nid));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 memset(NODE_DATA(nid), 0, sizeof(struct pglist_data));
639
640 dbg("node %d\n", nid);
641 dbg("NODE_DATA() = %p\n", NODE_DATA(nid));
642
643 NODE_DATA(nid)->bdata = &plat_node_bdata[nid];
Anton Blanchard45fb6ce2005-11-11 14:22:35 +1100644 NODE_DATA(nid)->node_start_pfn = start_pfn;
645 NODE_DATA(nid)->node_spanned_pages = end_pfn - start_pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646
647 if (NODE_DATA(nid)->node_spanned_pages == 0)
648 continue;
649
Anton Blanchard45fb6ce2005-11-11 14:22:35 +1100650 dbg("start_paddr = %lx\n", start_pfn << PAGE_SHIFT);
651 dbg("end_paddr = %lx\n", end_pfn << PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652
Anton Blanchard45fb6ce2005-11-11 14:22:35 +1100653 bootmap_pages = bootmem_bootmap_pages(end_pfn - start_pfn);
654 bootmem_paddr = (unsigned long)careful_allocation(nid,
655 bootmap_pages << PAGE_SHIFT,
656 PAGE_SIZE, end_pfn);
657 memset(__va(bootmem_paddr), 0, bootmap_pages << PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 dbg("bootmap_paddr = %lx\n", bootmem_paddr);
660
661 init_bootmem_node(NODE_DATA(nid), bootmem_paddr >> PAGE_SHIFT,
Anton Blanchard45fb6ce2005-11-11 14:22:35 +1100662 start_pfn, end_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663
Anton Blanchard45fb6ce2005-11-11 14:22:35 +1100664 /* Add free regions on this node */
665 for (i = 0; init_node_data[i].end_pfn; i++) {
666 unsigned long start, end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667
Anton Blanchard45fb6ce2005-11-11 14:22:35 +1100668 if (init_node_data[i].nid != nid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 continue;
670
Anton Blanchard45fb6ce2005-11-11 14:22:35 +1100671 start = init_node_data[i].start_pfn << PAGE_SHIFT;
672 end = init_node_data[i].end_pfn << PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673
Anton Blanchard45fb6ce2005-11-11 14:22:35 +1100674 dbg("free_bootmem %lx %lx\n", start, end - start);
675 free_bootmem_node(NODE_DATA(nid), start, end - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 }
677
Anton Blanchard45fb6ce2005-11-11 14:22:35 +1100678 /* Mark reserved regions on this node */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 for (i = 0; i < lmb.reserved.cnt; i++) {
Michael Ellerman180379d2005-08-03 20:21:26 +1000680 unsigned long physbase = lmb.reserved.region[i].base;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 unsigned long size = lmb.reserved.region[i].size;
Anton Blanchard45fb6ce2005-11-11 14:22:35 +1100682 unsigned long start_paddr = start_pfn << PAGE_SHIFT;
683 unsigned long end_paddr = end_pfn << PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684
Anton Blanchard45fb6ce2005-11-11 14:22:35 +1100685 if (early_pfn_to_nid(physbase >> PAGE_SHIFT) != nid &&
686 early_pfn_to_nid((physbase+size-1) >> PAGE_SHIFT) != nid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 continue;
688
689 if (physbase < end_paddr &&
690 (physbase+size) > start_paddr) {
691 /* overlaps */
692 if (physbase < start_paddr) {
693 size -= start_paddr - physbase;
694 physbase = start_paddr;
695 }
696
697 if (size > end_paddr - physbase)
698 size = end_paddr - physbase;
699
700 dbg("reserve_bootmem %lx %lx\n", physbase,
701 size);
702 reserve_bootmem_node(NODE_DATA(nid), physbase,
703 size);
704 }
705 }
Bob Picco802f1922005-09-03 15:54:26 -0700706
Anton Blanchard45fb6ce2005-11-11 14:22:35 +1100707 /* Add regions into sparsemem */
708 for (i = 0; init_node_data[i].end_pfn; i++) {
709 unsigned long start, end;
710
711 if (init_node_data[i].nid != nid)
Bob Picco802f1922005-09-03 15:54:26 -0700712 continue;
713
Anton Blanchard45fb6ce2005-11-11 14:22:35 +1100714 start = init_node_data[i].start_pfn;
715 end = init_node_data[i].end_pfn;
Bob Picco802f1922005-09-03 15:54:26 -0700716
Anton Blanchard45fb6ce2005-11-11 14:22:35 +1100717 memory_present(nid, start, end);
Bob Picco802f1922005-09-03 15:54:26 -0700718 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 }
720}
721
722void __init paging_init(void)
723{
724 unsigned long zones_size[MAX_NR_ZONES];
725 unsigned long zholes_size[MAX_NR_ZONES];
726 int nid;
727
728 memset(zones_size, 0, sizeof(zones_size));
729 memset(zholes_size, 0, sizeof(zholes_size));
730
731 for_each_online_node(nid) {
Anton Blanchard45fb6ce2005-11-11 14:22:35 +1100732 unsigned long start_pfn, end_pfn, pages_present;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733
Anton Blanchard45fb6ce2005-11-11 14:22:35 +1100734 get_region(nid, &start_pfn, &end_pfn, &pages_present);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735
736 zones_size[ZONE_DMA] = end_pfn - start_pfn;
Anton Blanchard45fb6ce2005-11-11 14:22:35 +1100737 zholes_size[ZONE_DMA] = zones_size[ZONE_DMA] - pages_present;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738
739 dbg("free_area_init node %d %lx %lx (hole: %lx)\n", nid,
740 zones_size[ZONE_DMA], start_pfn, zholes_size[ZONE_DMA]);
741
Anton Blanchard45fb6ce2005-11-11 14:22:35 +1100742 free_area_init_node(nid, NODE_DATA(nid), zones_size, start_pfn,
743 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 }
745}
746
747static int __init early_numa(char *p)
748{
749 if (!p)
750 return 0;
751
752 if (strstr(p, "off"))
753 numa_enabled = 0;
754
755 if (strstr(p, "debug"))
756 numa_debug = 1;
757
758 return 0;
759}
760early_param("numa", early_numa);
Mike Kravetz237a0982005-12-05 12:06:42 -0800761
762#ifdef CONFIG_MEMORY_HOTPLUG
763/*
764 * Find the node associated with a hot added memory section. Section
765 * corresponds to a SPARSEMEM section, not an LMB. It is assumed that
766 * sections are fully contained within a single LMB.
767 */
768int hot_add_scn_to_nid(unsigned long scn_addr)
769{
770 struct device_node *memory = NULL;
Mike Kravetzb226e462005-12-16 14:30:35 -0800771 nodemask_t nodes;
772 int numa_domain = 0;
Mike Kravetz237a0982005-12-05 12:06:42 -0800773
774 if (!numa_enabled || (min_common_depth < 0))
Mike Kravetzb226e462005-12-16 14:30:35 -0800775 return numa_domain;
Mike Kravetz237a0982005-12-05 12:06:42 -0800776
777 while ((memory = of_find_node_by_type(memory, "memory")) != NULL) {
778 unsigned long start, size;
Mike Kravetzb226e462005-12-16 14:30:35 -0800779 int ranges;
Mike Kravetz237a0982005-12-05 12:06:42 -0800780 unsigned int *memcell_buf;
781 unsigned int len;
782
783 memcell_buf = (unsigned int *)get_property(memory, "reg", &len);
784 if (!memcell_buf || len <= 0)
785 continue;
786
Benjamin Herrenschmidtcc5d0182005-12-13 18:01:21 +1100787 /* ranges in cell */
788 ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells);
Mike Kravetz237a0982005-12-05 12:06:42 -0800789ha_new_range:
790 start = read_n_cells(n_mem_addr_cells, &memcell_buf);
791 size = read_n_cells(n_mem_size_cells, &memcell_buf);
792 numa_domain = of_node_numa_domain(memory);
793
794 /* Domains not present at boot default to 0 */
795 if (!node_online(numa_domain))
796 numa_domain = any_online_node(NODE_MASK_ALL);
797
798 if ((scn_addr >= start) && (scn_addr < (start + size))) {
799 of_node_put(memory);
Mike Kravetzb226e462005-12-16 14:30:35 -0800800 goto got_numa_domain;
Mike Kravetz237a0982005-12-05 12:06:42 -0800801 }
802
803 if (--ranges) /* process all ranges in cell */
804 goto ha_new_range;
805 }
Mike Kravetz237a0982005-12-05 12:06:42 -0800806 BUG(); /* section address should be found above */
Mike Kravetzb226e462005-12-16 14:30:35 -0800807
808 /* Temporary code to ensure that returned node is not empty */
809got_numa_domain:
810 nodes_setall(nodes);
811 while (NODE_DATA(numa_domain)->node_spanned_pages == 0) {
812 node_clear(numa_domain, nodes);
813 numa_domain = any_online_node(nodes);
814 }
815 return numa_domain;
Mike Kravetz237a0982005-12-05 12:06:42 -0800816}
817#endif /* CONFIG_MEMORY_HOTPLUG */