]>
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 | 39 | static int min_common_depth; |
237a0989 | 40 | static int n_mem_addr_cells, n_mem_size_cells; |
1da177e4 LT |
41 | |
42 | /* | |
45fb6cea | 43 | * We need somewhere to store start/end/node for each region until we have |
1da177e4 LT |
44 | * allocated the real node_data structures. |
45 | */ | |
45fb6cea | 46 | #define MAX_REGIONS (MAX_LMB_REGIONS*2) |
1da177e4 | 47 | static struct { |
45fb6cea AB |
48 | unsigned long start_pfn; |
49 | unsigned long end_pfn; | |
50 | int nid; | |
51 | } init_node_data[MAX_REGIONS] __initdata; | |
1da177e4 | 52 | |
45fb6cea AB |
53 | int __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 | ||
68 | void __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 */ | |
105 | void __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) | |
6d91bb93 | 129 | *start_pfn = 0; |
45fb6cea | 130 | } |
1da177e4 | 131 | |
2e5ce39d | 132 | static void __cpuinit map_cpu_to_node(int cpu, int node) |
1da177e4 LT |
133 | { |
134 | numa_cpu_lookup_table[cpu] = node; | |
45fb6cea | 135 | |
bf4b85b0 NL |
136 | dbg("adding cpu %d to node %d\n", cpu, node); |
137 | ||
45fb6cea | 138 | if (!(cpu_isset(cpu, numa_cpumask_lookup_table[node]))) |
1da177e4 | 139 | cpu_set(cpu, numa_cpumask_lookup_table[node]); |
1da177e4 LT |
140 | } |
141 | ||
142 | #ifdef CONFIG_HOTPLUG_CPU | |
143 | static 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]); | |
1da177e4 LT |
151 | } 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 | ||
2e5ce39d | 158 | static struct device_node * __cpuinit find_cpu_node(unsigned int cpu) |
1da177e4 LT |
159 | { |
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 */ | |
189 | static int *of_get_associativity(struct device_node *dev) | |
190 | { | |
191 | return (unsigned int *)get_property(dev, "ibm,associativity", NULL); | |
192 | } | |
193 | ||
482ec7c4 NL |
194 | /* Returns nid in the range [0..MAX_NUMNODES-1], or -1 if no useful numa |
195 | * info is found. | |
196 | */ | |
953039c8 | 197 | static int of_node_to_nid_single(struct device_node *device) |
1da177e4 | 198 | { |
482ec7c4 | 199 | int nid = -1; |
1da177e4 LT |
200 | unsigned int *tmp; |
201 | ||
202 | if (min_common_depth == -1) | |
482ec7c4 | 203 | goto out; |
1da177e4 LT |
204 | |
205 | tmp = of_get_associativity(device); | |
482ec7c4 NL |
206 | if (!tmp) |
207 | goto out; | |
208 | ||
209 | if (tmp[0] >= min_common_depth) | |
cf950b7a | 210 | nid = tmp[min_common_depth]; |
bc16a759 NL |
211 | |
212 | /* POWER4 LPAR uses 0xffff as invalid node */ | |
482ec7c4 NL |
213 | if (nid == 0xffff || nid >= MAX_NUMNODES) |
214 | nid = -1; | |
215 | out: | |
cf950b7a | 216 | return nid; |
1da177e4 LT |
217 | } |
218 | ||
953039c8 JK |
219 | /* Walk the device tree upwards, looking for an associativity id */ |
220 | int of_node_to_nid(struct device_node *device) | |
221 | { | |
222 | struct device_node *tmp; | |
223 | int nid = -1; | |
224 | ||
225 | of_node_get(device); | |
226 | while (device) { | |
227 | nid = of_node_to_nid_single(device); | |
228 | if (nid != -1) | |
229 | break; | |
230 | ||
231 | tmp = device; | |
232 | device = of_get_parent(tmp); | |
233 | of_node_put(tmp); | |
234 | } | |
235 | of_node_put(device); | |
236 | ||
237 | return nid; | |
238 | } | |
239 | EXPORT_SYMBOL_GPL(of_node_to_nid); | |
240 | ||
1da177e4 LT |
241 | /* |
242 | * In theory, the "ibm,associativity" property may contain multiple | |
243 | * associativity lists because a resource may be multiply connected | |
244 | * into the machine. This resource then has different associativity | |
245 | * characteristics relative to its multiple connections. We ignore | |
246 | * this for now. We also assume that all cpu and memory sets have | |
247 | * their distances represented at a common level. This won't be | |
248 | * true for heirarchical NUMA. | |
249 | * | |
250 | * In any case the ibm,associativity-reference-points should give | |
251 | * the correct depth for a normal NUMA system. | |
252 | * | |
253 | * - Dave Hansen <haveblue@us.ibm.com> | |
254 | */ | |
255 | static int __init find_min_common_depth(void) | |
256 | { | |
257 | int depth; | |
258 | unsigned int *ref_points; | |
259 | struct device_node *rtas_root; | |
260 | unsigned int len; | |
261 | ||
262 | rtas_root = of_find_node_by_path("/rtas"); | |
263 | ||
264 | if (!rtas_root) | |
265 | return -1; | |
266 | ||
267 | /* | |
268 | * this property is 2 32-bit integers, each representing a level of | |
269 | * depth in the associativity nodes. The first is for an SMP | |
270 | * configuration (should be all 0's) and the second is for a normal | |
271 | * NUMA configuration. | |
272 | */ | |
273 | ref_points = (unsigned int *)get_property(rtas_root, | |
274 | "ibm,associativity-reference-points", &len); | |
275 | ||
276 | if ((len >= 1) && ref_points) { | |
277 | depth = ref_points[1]; | |
278 | } else { | |
bf4b85b0 | 279 | dbg("NUMA: ibm,associativity-reference-points not found.\n"); |
1da177e4 LT |
280 | depth = -1; |
281 | } | |
282 | of_node_put(rtas_root); | |
283 | ||
284 | return depth; | |
285 | } | |
286 | ||
84c9fdd1 | 287 | static void __init get_n_mem_cells(int *n_addr_cells, int *n_size_cells) |
1da177e4 LT |
288 | { |
289 | struct device_node *memory = NULL; | |
1da177e4 LT |
290 | |
291 | memory = of_find_node_by_type(memory, "memory"); | |
54c23310 | 292 | if (!memory) |
84c9fdd1 | 293 | panic("numa.c: No memory nodes found!"); |
54c23310 | 294 | |
84c9fdd1 MK |
295 | *n_addr_cells = prom_n_addr_cells(memory); |
296 | *n_size_cells = prom_n_size_cells(memory); | |
297 | of_node_put(memory); | |
1da177e4 LT |
298 | } |
299 | ||
237a0989 | 300 | static unsigned long __devinit read_n_cells(int n, unsigned int **buf) |
1da177e4 LT |
301 | { |
302 | unsigned long result = 0; | |
303 | ||
304 | while (n--) { | |
305 | result = (result << 32) | **buf; | |
306 | (*buf)++; | |
307 | } | |
308 | return result; | |
309 | } | |
310 | ||
311 | /* | |
312 | * Figure out to which domain a cpu belongs and stick it there. | |
313 | * Return the id of the domain used. | |
314 | */ | |
2e5ce39d | 315 | static int __cpuinit numa_setup_cpu(unsigned long lcpu) |
1da177e4 | 316 | { |
cf950b7a | 317 | int nid = 0; |
1da177e4 LT |
318 | struct device_node *cpu = find_cpu_node(lcpu); |
319 | ||
320 | if (!cpu) { | |
321 | WARN_ON(1); | |
322 | goto out; | |
323 | } | |
324 | ||
953039c8 | 325 | nid = of_node_to_nid_single(cpu); |
1da177e4 | 326 | |
482ec7c4 NL |
327 | if (nid < 0 || !node_online(nid)) |
328 | nid = any_online_node(NODE_MASK_ALL); | |
1da177e4 | 329 | out: |
cf950b7a | 330 | map_cpu_to_node(lcpu, nid); |
1da177e4 LT |
331 | |
332 | of_node_put(cpu); | |
333 | ||
cf950b7a | 334 | return nid; |
1da177e4 LT |
335 | } |
336 | ||
74b85f37 | 337 | static int __cpuinit cpu_numa_callback(struct notifier_block *nfb, |
1da177e4 LT |
338 | unsigned long action, |
339 | void *hcpu) | |
340 | { | |
341 | unsigned long lcpu = (unsigned long)hcpu; | |
342 | int ret = NOTIFY_DONE; | |
343 | ||
344 | switch (action) { | |
345 | case CPU_UP_PREPARE: | |
2b261227 | 346 | numa_setup_cpu(lcpu); |
1da177e4 LT |
347 | ret = NOTIFY_OK; |
348 | break; | |
349 | #ifdef CONFIG_HOTPLUG_CPU | |
350 | case CPU_DEAD: | |
351 | case CPU_UP_CANCELED: | |
352 | unmap_cpu_from_node(lcpu); | |
353 | break; | |
354 | ret = NOTIFY_OK; | |
355 | #endif | |
356 | } | |
357 | return ret; | |
358 | } | |
359 | ||
360 | /* | |
361 | * Check and possibly modify a memory region to enforce the memory limit. | |
362 | * | |
363 | * Returns the size the region should have to enforce the memory limit. | |
364 | * This will either be the original value of size, a truncated value, | |
365 | * or zero. If the returned value of size is 0 the region should be | |
366 | * discarded as it lies wholy above the memory limit. | |
367 | */ | |
45fb6cea AB |
368 | static unsigned long __init numa_enforce_memory_limit(unsigned long start, |
369 | unsigned long size) | |
1da177e4 LT |
370 | { |
371 | /* | |
372 | * We use lmb_end_of_DRAM() in here instead of memory_limit because | |
373 | * we've already adjusted it for the limit and it takes care of | |
374 | * having memory holes below the limit. | |
375 | */ | |
1da177e4 LT |
376 | |
377 | if (! memory_limit) | |
378 | return size; | |
379 | ||
380 | if (start + size <= lmb_end_of_DRAM()) | |
381 | return size; | |
382 | ||
383 | if (start >= lmb_end_of_DRAM()) | |
384 | return 0; | |
385 | ||
386 | return lmb_end_of_DRAM() - start; | |
387 | } | |
388 | ||
389 | static int __init parse_numa_properties(void) | |
390 | { | |
391 | struct device_node *cpu = NULL; | |
392 | struct device_node *memory = NULL; | |
482ec7c4 | 393 | int default_nid = 0; |
1da177e4 LT |
394 | unsigned long i; |
395 | ||
396 | if (numa_enabled == 0) { | |
397 | printk(KERN_WARNING "NUMA disabled by user\n"); | |
398 | return -1; | |
399 | } | |
400 | ||
1da177e4 LT |
401 | min_common_depth = find_min_common_depth(); |
402 | ||
1da177e4 LT |
403 | if (min_common_depth < 0) |
404 | return min_common_depth; | |
405 | ||
bf4b85b0 NL |
406 | dbg("NUMA associativity depth for CPU/Memory: %d\n", min_common_depth); |
407 | ||
1da177e4 | 408 | /* |
482ec7c4 NL |
409 | * Even though we connect cpus to numa domains later in SMP |
410 | * init, we need to know the node ids now. This is because | |
411 | * each node to be onlined must have NODE_DATA etc backing it. | |
1da177e4 | 412 | */ |
482ec7c4 | 413 | for_each_present_cpu(i) { |
cf950b7a | 414 | int nid; |
1da177e4 LT |
415 | |
416 | cpu = find_cpu_node(i); | |
482ec7c4 | 417 | BUG_ON(!cpu); |
953039c8 | 418 | nid = of_node_to_nid_single(cpu); |
482ec7c4 | 419 | of_node_put(cpu); |
1da177e4 | 420 | |
482ec7c4 NL |
421 | /* |
422 | * Don't fall back to default_nid yet -- we will plug | |
423 | * cpus into nodes once the memory scan has discovered | |
424 | * the topology. | |
425 | */ | |
426 | if (nid < 0) | |
427 | continue; | |
428 | node_set_online(nid); | |
1da177e4 LT |
429 | } |
430 | ||
237a0989 | 431 | get_n_mem_cells(&n_mem_addr_cells, &n_mem_size_cells); |
1da177e4 LT |
432 | memory = NULL; |
433 | while ((memory = of_find_node_by_type(memory, "memory")) != NULL) { | |
434 | unsigned long start; | |
435 | unsigned long size; | |
cf950b7a | 436 | int nid; |
1da177e4 LT |
437 | int ranges; |
438 | unsigned int *memcell_buf; | |
439 | unsigned int len; | |
440 | ||
ba759485 ME |
441 | memcell_buf = (unsigned int *)get_property(memory, |
442 | "linux,usable-memory", &len); | |
443 | if (!memcell_buf || len <= 0) | |
444 | memcell_buf = | |
445 | (unsigned int *)get_property(memory, "reg", | |
446 | &len); | |
1da177e4 LT |
447 | if (!memcell_buf || len <= 0) |
448 | continue; | |
449 | ||
cc5d0189 BH |
450 | /* ranges in cell */ |
451 | ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells); | |
1da177e4 LT |
452 | new_range: |
453 | /* these are order-sensitive, and modify the buffer pointer */ | |
237a0989 MK |
454 | start = read_n_cells(n_mem_addr_cells, &memcell_buf); |
455 | size = read_n_cells(n_mem_size_cells, &memcell_buf); | |
1da177e4 | 456 | |
482ec7c4 NL |
457 | /* |
458 | * Assumption: either all memory nodes or none will | |
459 | * have associativity properties. If none, then | |
460 | * everything goes to default_nid. | |
461 | */ | |
953039c8 | 462 | nid = of_node_to_nid_single(memory); |
482ec7c4 NL |
463 | if (nid < 0) |
464 | nid = default_nid; | |
465 | node_set_online(nid); | |
1da177e4 | 466 | |
45fb6cea | 467 | if (!(size = numa_enforce_memory_limit(start, size))) { |
1da177e4 LT |
468 | if (--ranges) |
469 | goto new_range; | |
470 | else | |
471 | continue; | |
472 | } | |
473 | ||
cf950b7a | 474 | add_region(nid, start >> PAGE_SHIFT, |
45fb6cea | 475 | size >> PAGE_SHIFT); |
1da177e4 LT |
476 | |
477 | if (--ranges) | |
478 | goto new_range; | |
479 | } | |
480 | ||
1da177e4 LT |
481 | return 0; |
482 | } | |
483 | ||
484 | static void __init setup_nonnuma(void) | |
485 | { | |
486 | unsigned long top_of_ram = lmb_end_of_DRAM(); | |
487 | unsigned long total_ram = lmb_phys_mem_size(); | |
fb6d73d3 | 488 | unsigned int i; |
1da177e4 | 489 | |
e110b281 | 490 | printk(KERN_DEBUG "Top of RAM: 0x%lx, Total RAM: 0x%lx\n", |
1da177e4 | 491 | top_of_ram, total_ram); |
e110b281 | 492 | printk(KERN_DEBUG "Memory hole size: %ldMB\n", |
1da177e4 LT |
493 | (top_of_ram - total_ram) >> 20); |
494 | ||
fb6d73d3 PM |
495 | for (i = 0; i < lmb.memory.cnt; ++i) |
496 | add_region(0, lmb.memory.region[i].base >> PAGE_SHIFT, | |
497 | lmb_size_pages(&lmb.memory, i)); | |
1da177e4 | 498 | node_set_online(0); |
1da177e4 LT |
499 | } |
500 | ||
4b703a23 AB |
501 | void __init dump_numa_cpu_topology(void) |
502 | { | |
503 | unsigned int node; | |
504 | unsigned int cpu, count; | |
505 | ||
506 | if (min_common_depth == -1 || !numa_enabled) | |
507 | return; | |
508 | ||
509 | for_each_online_node(node) { | |
e110b281 | 510 | printk(KERN_DEBUG "Node %d CPUs:", node); |
4b703a23 AB |
511 | |
512 | count = 0; | |
513 | /* | |
514 | * If we used a CPU iterator here we would miss printing | |
515 | * the holes in the cpumap. | |
516 | */ | |
517 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
518 | if (cpu_isset(cpu, numa_cpumask_lookup_table[node])) { | |
519 | if (count == 0) | |
520 | printk(" %u", cpu); | |
521 | ++count; | |
522 | } else { | |
523 | if (count > 1) | |
524 | printk("-%u", cpu - 1); | |
525 | count = 0; | |
526 | } | |
527 | } | |
528 | ||
529 | if (count > 1) | |
530 | printk("-%u", NR_CPUS - 1); | |
531 | printk("\n"); | |
532 | } | |
533 | } | |
534 | ||
535 | static void __init dump_numa_memory_topology(void) | |
1da177e4 LT |
536 | { |
537 | unsigned int node; | |
538 | unsigned int count; | |
539 | ||
540 | if (min_common_depth == -1 || !numa_enabled) | |
541 | return; | |
542 | ||
543 | for_each_online_node(node) { | |
544 | unsigned long i; | |
545 | ||
e110b281 | 546 | printk(KERN_DEBUG "Node %d Memory:", node); |
1da177e4 LT |
547 | |
548 | count = 0; | |
549 | ||
45fb6cea AB |
550 | for (i = 0; i < lmb_end_of_DRAM(); |
551 | i += (1 << SECTION_SIZE_BITS)) { | |
552 | if (early_pfn_to_nid(i >> PAGE_SHIFT) == node) { | |
1da177e4 LT |
553 | if (count == 0) |
554 | printk(" 0x%lx", i); | |
555 | ++count; | |
556 | } else { | |
557 | if (count > 0) | |
558 | printk("-0x%lx", i); | |
559 | count = 0; | |
560 | } | |
561 | } | |
562 | ||
563 | if (count > 0) | |
564 | printk("-0x%lx", i); | |
565 | printk("\n"); | |
566 | } | |
1da177e4 LT |
567 | } |
568 | ||
569 | /* | |
570 | * Allocate some memory, satisfying the lmb or bootmem allocator where | |
571 | * required. nid is the preferred node and end is the physical address of | |
572 | * the highest address in the node. | |
573 | * | |
574 | * Returns the physical address of the memory. | |
575 | */ | |
45fb6cea AB |
576 | static void __init *careful_allocation(int nid, unsigned long size, |
577 | unsigned long align, | |
578 | unsigned long end_pfn) | |
1da177e4 | 579 | { |
45fb6cea | 580 | int new_nid; |
d7a5b2ff | 581 | unsigned long ret = __lmb_alloc_base(size, align, end_pfn << PAGE_SHIFT); |
1da177e4 LT |
582 | |
583 | /* retry over all memory */ | |
584 | if (!ret) | |
d7a5b2ff | 585 | ret = __lmb_alloc_base(size, align, lmb_end_of_DRAM()); |
1da177e4 LT |
586 | |
587 | if (!ret) | |
588 | panic("numa.c: cannot allocate %lu bytes on node %d", | |
589 | size, nid); | |
590 | ||
591 | /* | |
592 | * If the memory came from a previously allocated node, we must | |
593 | * retry with the bootmem allocator. | |
594 | */ | |
45fb6cea AB |
595 | new_nid = early_pfn_to_nid(ret >> PAGE_SHIFT); |
596 | if (new_nid < nid) { | |
597 | ret = (unsigned long)__alloc_bootmem_node(NODE_DATA(new_nid), | |
1da177e4 LT |
598 | size, align, 0); |
599 | ||
600 | if (!ret) | |
601 | panic("numa.c: cannot allocate %lu bytes on node %d", | |
45fb6cea | 602 | size, new_nid); |
1da177e4 | 603 | |
45fb6cea | 604 | ret = __pa(ret); |
1da177e4 LT |
605 | |
606 | dbg("alloc_bootmem %lx %lx\n", ret, size); | |
607 | } | |
608 | ||
45fb6cea | 609 | return (void *)ret; |
1da177e4 LT |
610 | } |
611 | ||
74b85f37 CS |
612 | static struct notifier_block __cpuinitdata ppc64_numa_nb = { |
613 | .notifier_call = cpu_numa_callback, | |
614 | .priority = 1 /* Must run before sched domains notifier. */ | |
615 | }; | |
616 | ||
1da177e4 LT |
617 | void __init do_init_bootmem(void) |
618 | { | |
619 | int nid; | |
45fb6cea | 620 | unsigned int i; |
1da177e4 LT |
621 | |
622 | min_low_pfn = 0; | |
623 | max_low_pfn = lmb_end_of_DRAM() >> PAGE_SHIFT; | |
624 | max_pfn = max_low_pfn; | |
625 | ||
626 | if (parse_numa_properties()) | |
627 | setup_nonnuma(); | |
628 | else | |
4b703a23 | 629 | dump_numa_memory_topology(); |
1da177e4 LT |
630 | |
631 | register_cpu_notifier(&ppc64_numa_nb); | |
2b261227 NL |
632 | cpu_numa_callback(&ppc64_numa_nb, CPU_UP_PREPARE, |
633 | (void *)(unsigned long)boot_cpuid); | |
1da177e4 LT |
634 | |
635 | for_each_online_node(nid) { | |
45fb6cea | 636 | unsigned long start_pfn, end_pfn, pages_present; |
1da177e4 LT |
637 | unsigned long bootmem_paddr; |
638 | unsigned long bootmap_pages; | |
639 | ||
45fb6cea | 640 | get_region(nid, &start_pfn, &end_pfn, &pages_present); |
1da177e4 LT |
641 | |
642 | /* Allocate the node structure node local if possible */ | |
45fb6cea | 643 | NODE_DATA(nid) = careful_allocation(nid, |
1da177e4 | 644 | sizeof(struct pglist_data), |
45fb6cea AB |
645 | SMP_CACHE_BYTES, end_pfn); |
646 | NODE_DATA(nid) = __va(NODE_DATA(nid)); | |
1da177e4 LT |
647 | memset(NODE_DATA(nid), 0, sizeof(struct pglist_data)); |
648 | ||
649 | dbg("node %d\n", nid); | |
650 | dbg("NODE_DATA() = %p\n", NODE_DATA(nid)); | |
651 | ||
652 | NODE_DATA(nid)->bdata = &plat_node_bdata[nid]; | |
45fb6cea AB |
653 | NODE_DATA(nid)->node_start_pfn = start_pfn; |
654 | NODE_DATA(nid)->node_spanned_pages = end_pfn - start_pfn; | |
1da177e4 LT |
655 | |
656 | if (NODE_DATA(nid)->node_spanned_pages == 0) | |
657 | continue; | |
658 | ||
45fb6cea AB |
659 | dbg("start_paddr = %lx\n", start_pfn << PAGE_SHIFT); |
660 | dbg("end_paddr = %lx\n", end_pfn << PAGE_SHIFT); | |
1da177e4 | 661 | |
45fb6cea AB |
662 | bootmap_pages = bootmem_bootmap_pages(end_pfn - start_pfn); |
663 | bootmem_paddr = (unsigned long)careful_allocation(nid, | |
664 | bootmap_pages << PAGE_SHIFT, | |
665 | PAGE_SIZE, end_pfn); | |
666 | memset(__va(bootmem_paddr), 0, bootmap_pages << PAGE_SHIFT); | |
1da177e4 | 667 | |
1da177e4 LT |
668 | dbg("bootmap_paddr = %lx\n", bootmem_paddr); |
669 | ||
670 | init_bootmem_node(NODE_DATA(nid), bootmem_paddr >> PAGE_SHIFT, | |
45fb6cea | 671 | start_pfn, end_pfn); |
1da177e4 | 672 | |
45fb6cea AB |
673 | /* Add free regions on this node */ |
674 | for (i = 0; init_node_data[i].end_pfn; i++) { | |
675 | unsigned long start, end; | |
1da177e4 | 676 | |
45fb6cea | 677 | if (init_node_data[i].nid != nid) |
1da177e4 LT |
678 | continue; |
679 | ||
45fb6cea AB |
680 | start = init_node_data[i].start_pfn << PAGE_SHIFT; |
681 | end = init_node_data[i].end_pfn << PAGE_SHIFT; | |
1da177e4 | 682 | |
45fb6cea AB |
683 | dbg("free_bootmem %lx %lx\n", start, end - start); |
684 | free_bootmem_node(NODE_DATA(nid), start, end - start); | |
1da177e4 LT |
685 | } |
686 | ||
45fb6cea | 687 | /* Mark reserved regions on this node */ |
1da177e4 | 688 | for (i = 0; i < lmb.reserved.cnt; i++) { |
180379dc | 689 | unsigned long physbase = lmb.reserved.region[i].base; |
1da177e4 | 690 | unsigned long size = lmb.reserved.region[i].size; |
45fb6cea AB |
691 | unsigned long start_paddr = start_pfn << PAGE_SHIFT; |
692 | unsigned long end_paddr = end_pfn << PAGE_SHIFT; | |
1da177e4 | 693 | |
45fb6cea AB |
694 | if (early_pfn_to_nid(physbase >> PAGE_SHIFT) != nid && |
695 | early_pfn_to_nid((physbase+size-1) >> PAGE_SHIFT) != nid) | |
1da177e4 LT |
696 | continue; |
697 | ||
698 | if (physbase < end_paddr && | |
699 | (physbase+size) > start_paddr) { | |
700 | /* overlaps */ | |
701 | if (physbase < start_paddr) { | |
702 | size -= start_paddr - physbase; | |
703 | physbase = start_paddr; | |
704 | } | |
705 | ||
706 | if (size > end_paddr - physbase) | |
707 | size = end_paddr - physbase; | |
708 | ||
709 | dbg("reserve_bootmem %lx %lx\n", physbase, | |
710 | size); | |
711 | reserve_bootmem_node(NODE_DATA(nid), physbase, | |
712 | size); | |
713 | } | |
714 | } | |
802f192e | 715 | |
45fb6cea AB |
716 | /* Add regions into sparsemem */ |
717 | for (i = 0; init_node_data[i].end_pfn; i++) { | |
718 | unsigned long start, end; | |
719 | ||
720 | if (init_node_data[i].nid != nid) | |
802f192e BP |
721 | continue; |
722 | ||
45fb6cea AB |
723 | start = init_node_data[i].start_pfn; |
724 | end = init_node_data[i].end_pfn; | |
802f192e | 725 | |
45fb6cea | 726 | memory_present(nid, start, end); |
802f192e | 727 | } |
1da177e4 LT |
728 | } |
729 | } | |
730 | ||
731 | void __init paging_init(void) | |
732 | { | |
733 | unsigned long zones_size[MAX_NR_ZONES]; | |
734 | unsigned long zholes_size[MAX_NR_ZONES]; | |
735 | int nid; | |
736 | ||
737 | memset(zones_size, 0, sizeof(zones_size)); | |
738 | memset(zholes_size, 0, sizeof(zholes_size)); | |
739 | ||
740 | for_each_online_node(nid) { | |
45fb6cea | 741 | unsigned long start_pfn, end_pfn, pages_present; |
1da177e4 | 742 | |
45fb6cea | 743 | get_region(nid, &start_pfn, &end_pfn, &pages_present); |
1da177e4 LT |
744 | |
745 | zones_size[ZONE_DMA] = end_pfn - start_pfn; | |
45fb6cea | 746 | zholes_size[ZONE_DMA] = zones_size[ZONE_DMA] - pages_present; |
1da177e4 LT |
747 | |
748 | dbg("free_area_init node %d %lx %lx (hole: %lx)\n", nid, | |
749 | zones_size[ZONE_DMA], start_pfn, zholes_size[ZONE_DMA]); | |
750 | ||
45fb6cea AB |
751 | free_area_init_node(nid, NODE_DATA(nid), zones_size, start_pfn, |
752 | zholes_size); | |
1da177e4 LT |
753 | } |
754 | } | |
755 | ||
756 | static int __init early_numa(char *p) | |
757 | { | |
758 | if (!p) | |
759 | return 0; | |
760 | ||
761 | if (strstr(p, "off")) | |
762 | numa_enabled = 0; | |
763 | ||
764 | if (strstr(p, "debug")) | |
765 | numa_debug = 1; | |
766 | ||
767 | return 0; | |
768 | } | |
769 | early_param("numa", early_numa); | |
237a0989 MK |
770 | |
771 | #ifdef CONFIG_MEMORY_HOTPLUG | |
772 | /* | |
773 | * Find the node associated with a hot added memory section. Section | |
774 | * corresponds to a SPARSEMEM section, not an LMB. It is assumed that | |
775 | * sections are fully contained within a single LMB. | |
776 | */ | |
777 | int hot_add_scn_to_nid(unsigned long scn_addr) | |
778 | { | |
779 | struct device_node *memory = NULL; | |
b226e462 | 780 | nodemask_t nodes; |
482ec7c4 | 781 | int default_nid = any_online_node(NODE_MASK_ALL); |
069007ae | 782 | int nid; |
237a0989 MK |
783 | |
784 | if (!numa_enabled || (min_common_depth < 0)) | |
482ec7c4 | 785 | return default_nid; |
237a0989 MK |
786 | |
787 | while ((memory = of_find_node_by_type(memory, "memory")) != NULL) { | |
788 | unsigned long start, size; | |
b226e462 | 789 | int ranges; |
237a0989 MK |
790 | unsigned int *memcell_buf; |
791 | unsigned int len; | |
792 | ||
793 | memcell_buf = (unsigned int *)get_property(memory, "reg", &len); | |
794 | if (!memcell_buf || len <= 0) | |
795 | continue; | |
796 | ||
cc5d0189 BH |
797 | /* ranges in cell */ |
798 | ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells); | |
237a0989 MK |
799 | ha_new_range: |
800 | start = read_n_cells(n_mem_addr_cells, &memcell_buf); | |
801 | size = read_n_cells(n_mem_size_cells, &memcell_buf); | |
953039c8 | 802 | nid = of_node_to_nid_single(memory); |
237a0989 MK |
803 | |
804 | /* Domains not present at boot default to 0 */ | |
482ec7c4 NL |
805 | if (nid < 0 || !node_online(nid)) |
806 | nid = default_nid; | |
237a0989 MK |
807 | |
808 | if ((scn_addr >= start) && (scn_addr < (start + size))) { | |
809 | of_node_put(memory); | |
cf950b7a | 810 | goto got_nid; |
237a0989 MK |
811 | } |
812 | ||
813 | if (--ranges) /* process all ranges in cell */ | |
814 | goto ha_new_range; | |
815 | } | |
237a0989 | 816 | BUG(); /* section address should be found above */ |
069007ae | 817 | return 0; |
b226e462 MK |
818 | |
819 | /* Temporary code to ensure that returned node is not empty */ | |
cf950b7a | 820 | got_nid: |
b226e462 | 821 | nodes_setall(nodes); |
cf950b7a NL |
822 | while (NODE_DATA(nid)->node_spanned_pages == 0) { |
823 | node_clear(nid, nodes); | |
824 | nid = any_online_node(nodes); | |
b226e462 | 825 | } |
cf950b7a | 826 | return nid; |
237a0989 MK |
827 | } |
828 | #endif /* CONFIG_MEMORY_HOTPLUG */ |