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> | |
d9b2b2a2 | 20 | #include <linux/lmb.h> |
45fb6cea | 21 | #include <asm/sparsemem.h> |
d9b2b2a2 | 22 | #include <asm/prom.h> |
cf00a8d1 | 23 | #include <asm/system.h> |
2249ca9d | 24 | #include <asm/smp.h> |
1da177e4 LT |
25 | |
26 | static int numa_enabled = 1; | |
27 | ||
1daa6d08 BS |
28 | static char *cmdline __initdata; |
29 | ||
1da177e4 LT |
30 | static int numa_debug; |
31 | #define dbg(args...) if (numa_debug) { printk(KERN_INFO args); } | |
32 | ||
45fb6cea | 33 | int numa_cpu_lookup_table[NR_CPUS]; |
1da177e4 | 34 | cpumask_t numa_cpumask_lookup_table[MAX_NUMNODES]; |
1da177e4 | 35 | struct pglist_data *node_data[MAX_NUMNODES]; |
45fb6cea AB |
36 | |
37 | EXPORT_SYMBOL(numa_cpu_lookup_table); | |
38 | EXPORT_SYMBOL(numa_cpumask_lookup_table); | |
39 | EXPORT_SYMBOL(node_data); | |
40 | ||
41 | static bootmem_data_t __initdata plat_node_bdata[MAX_NUMNODES]; | |
1da177e4 | 42 | static int min_common_depth; |
237a0989 | 43 | static int n_mem_addr_cells, n_mem_size_cells; |
1da177e4 | 44 | |
1daa6d08 BS |
45 | static int __cpuinit fake_numa_create_new_node(unsigned long end_pfn, |
46 | unsigned int *nid) | |
47 | { | |
48 | unsigned long long mem; | |
49 | char *p = cmdline; | |
50 | static unsigned int fake_nid; | |
51 | static unsigned long long curr_boundary; | |
52 | ||
53 | /* | |
54 | * Modify node id, iff we started creating NUMA nodes | |
55 | * We want to continue from where we left of the last time | |
56 | */ | |
57 | if (fake_nid) | |
58 | *nid = fake_nid; | |
59 | /* | |
60 | * In case there are no more arguments to parse, the | |
61 | * node_id should be the same as the last fake node id | |
62 | * (we've handled this above). | |
63 | */ | |
64 | if (!p) | |
65 | return 0; | |
66 | ||
67 | mem = memparse(p, &p); | |
68 | if (!mem) | |
69 | return 0; | |
70 | ||
71 | if (mem < curr_boundary) | |
72 | return 0; | |
73 | ||
74 | curr_boundary = mem; | |
75 | ||
76 | if ((end_pfn << PAGE_SHIFT) > mem) { | |
77 | /* | |
78 | * Skip commas and spaces | |
79 | */ | |
80 | while (*p == ',' || *p == ' ' || *p == '\t') | |
81 | p++; | |
82 | ||
83 | cmdline = p; | |
84 | fake_nid++; | |
85 | *nid = fake_nid; | |
86 | dbg("created new fake_node with id %d\n", fake_nid); | |
87 | return 1; | |
88 | } | |
89 | return 0; | |
90 | } | |
91 | ||
2e5ce39d | 92 | static void __cpuinit map_cpu_to_node(int cpu, int node) |
1da177e4 LT |
93 | { |
94 | numa_cpu_lookup_table[cpu] = node; | |
45fb6cea | 95 | |
bf4b85b0 NL |
96 | dbg("adding cpu %d to node %d\n", cpu, node); |
97 | ||
45fb6cea | 98 | if (!(cpu_isset(cpu, numa_cpumask_lookup_table[node]))) |
1da177e4 | 99 | cpu_set(cpu, numa_cpumask_lookup_table[node]); |
1da177e4 LT |
100 | } |
101 | ||
102 | #ifdef CONFIG_HOTPLUG_CPU | |
103 | static void unmap_cpu_from_node(unsigned long cpu) | |
104 | { | |
105 | int node = numa_cpu_lookup_table[cpu]; | |
106 | ||
107 | dbg("removing cpu %lu from node %d\n", cpu, node); | |
108 | ||
109 | if (cpu_isset(cpu, numa_cpumask_lookup_table[node])) { | |
110 | cpu_clear(cpu, numa_cpumask_lookup_table[node]); | |
1da177e4 LT |
111 | } else { |
112 | printk(KERN_ERR "WARNING: cpu %lu not found in node %d\n", | |
113 | cpu, node); | |
114 | } | |
115 | } | |
116 | #endif /* CONFIG_HOTPLUG_CPU */ | |
117 | ||
2e5ce39d | 118 | static struct device_node * __cpuinit find_cpu_node(unsigned int cpu) |
1da177e4 LT |
119 | { |
120 | unsigned int hw_cpuid = get_hard_smp_processor_id(cpu); | |
121 | struct device_node *cpu_node = NULL; | |
a7f67bdf | 122 | const unsigned int *interrupt_server, *reg; |
1da177e4 LT |
123 | int len; |
124 | ||
125 | while ((cpu_node = of_find_node_by_type(cpu_node, "cpu")) != NULL) { | |
126 | /* Try interrupt server first */ | |
e2eb6392 | 127 | interrupt_server = of_get_property(cpu_node, |
1da177e4 LT |
128 | "ibm,ppc-interrupt-server#s", &len); |
129 | ||
130 | len = len / sizeof(u32); | |
131 | ||
132 | if (interrupt_server && (len > 0)) { | |
133 | while (len--) { | |
134 | if (interrupt_server[len] == hw_cpuid) | |
135 | return cpu_node; | |
136 | } | |
137 | } else { | |
e2eb6392 | 138 | reg = of_get_property(cpu_node, "reg", &len); |
1da177e4 LT |
139 | if (reg && (len > 0) && (reg[0] == hw_cpuid)) |
140 | return cpu_node; | |
141 | } | |
142 | } | |
143 | ||
144 | return NULL; | |
145 | } | |
146 | ||
147 | /* must hold reference to node during call */ | |
a7f67bdf | 148 | static const int *of_get_associativity(struct device_node *dev) |
1da177e4 | 149 | { |
e2eb6392 | 150 | return of_get_property(dev, "ibm,associativity", NULL); |
1da177e4 LT |
151 | } |
152 | ||
482ec7c4 NL |
153 | /* Returns nid in the range [0..MAX_NUMNODES-1], or -1 if no useful numa |
154 | * info is found. | |
155 | */ | |
953039c8 | 156 | static int of_node_to_nid_single(struct device_node *device) |
1da177e4 | 157 | { |
482ec7c4 | 158 | int nid = -1; |
a7f67bdf | 159 | const unsigned int *tmp; |
1da177e4 LT |
160 | |
161 | if (min_common_depth == -1) | |
482ec7c4 | 162 | goto out; |
1da177e4 LT |
163 | |
164 | tmp = of_get_associativity(device); | |
482ec7c4 NL |
165 | if (!tmp) |
166 | goto out; | |
167 | ||
168 | if (tmp[0] >= min_common_depth) | |
cf950b7a | 169 | nid = tmp[min_common_depth]; |
bc16a759 NL |
170 | |
171 | /* POWER4 LPAR uses 0xffff as invalid node */ | |
482ec7c4 NL |
172 | if (nid == 0xffff || nid >= MAX_NUMNODES) |
173 | nid = -1; | |
174 | out: | |
cf950b7a | 175 | return nid; |
1da177e4 LT |
176 | } |
177 | ||
953039c8 JK |
178 | /* Walk the device tree upwards, looking for an associativity id */ |
179 | int of_node_to_nid(struct device_node *device) | |
180 | { | |
181 | struct device_node *tmp; | |
182 | int nid = -1; | |
183 | ||
184 | of_node_get(device); | |
185 | while (device) { | |
186 | nid = of_node_to_nid_single(device); | |
187 | if (nid != -1) | |
188 | break; | |
189 | ||
190 | tmp = device; | |
191 | device = of_get_parent(tmp); | |
192 | of_node_put(tmp); | |
193 | } | |
194 | of_node_put(device); | |
195 | ||
196 | return nid; | |
197 | } | |
198 | EXPORT_SYMBOL_GPL(of_node_to_nid); | |
199 | ||
1da177e4 LT |
200 | /* |
201 | * In theory, the "ibm,associativity" property may contain multiple | |
202 | * associativity lists because a resource may be multiply connected | |
203 | * into the machine. This resource then has different associativity | |
204 | * characteristics relative to its multiple connections. We ignore | |
205 | * this for now. We also assume that all cpu and memory sets have | |
206 | * their distances represented at a common level. This won't be | |
1b3c3714 | 207 | * true for hierarchical NUMA. |
1da177e4 LT |
208 | * |
209 | * In any case the ibm,associativity-reference-points should give | |
210 | * the correct depth for a normal NUMA system. | |
211 | * | |
212 | * - Dave Hansen <haveblue@us.ibm.com> | |
213 | */ | |
214 | static int __init find_min_common_depth(void) | |
215 | { | |
216 | int depth; | |
a7f67bdf | 217 | const unsigned int *ref_points; |
1da177e4 LT |
218 | struct device_node *rtas_root; |
219 | unsigned int len; | |
220 | ||
221 | rtas_root = of_find_node_by_path("/rtas"); | |
222 | ||
223 | if (!rtas_root) | |
224 | return -1; | |
225 | ||
226 | /* | |
227 | * this property is 2 32-bit integers, each representing a level of | |
228 | * depth in the associativity nodes. The first is for an SMP | |
229 | * configuration (should be all 0's) and the second is for a normal | |
230 | * NUMA configuration. | |
231 | */ | |
e2eb6392 | 232 | ref_points = of_get_property(rtas_root, |
1da177e4 LT |
233 | "ibm,associativity-reference-points", &len); |
234 | ||
235 | if ((len >= 1) && ref_points) { | |
236 | depth = ref_points[1]; | |
237 | } else { | |
bf4b85b0 | 238 | dbg("NUMA: ibm,associativity-reference-points not found.\n"); |
1da177e4 LT |
239 | depth = -1; |
240 | } | |
241 | of_node_put(rtas_root); | |
242 | ||
243 | return depth; | |
244 | } | |
245 | ||
84c9fdd1 | 246 | static void __init get_n_mem_cells(int *n_addr_cells, int *n_size_cells) |
1da177e4 LT |
247 | { |
248 | struct device_node *memory = NULL; | |
1da177e4 LT |
249 | |
250 | memory = of_find_node_by_type(memory, "memory"); | |
54c23310 | 251 | if (!memory) |
84c9fdd1 | 252 | panic("numa.c: No memory nodes found!"); |
54c23310 | 253 | |
a8bda5dd | 254 | *n_addr_cells = of_n_addr_cells(memory); |
9213feea | 255 | *n_size_cells = of_n_size_cells(memory); |
84c9fdd1 | 256 | of_node_put(memory); |
1da177e4 LT |
257 | } |
258 | ||
a7f67bdf | 259 | static unsigned long __devinit read_n_cells(int n, const unsigned int **buf) |
1da177e4 LT |
260 | { |
261 | unsigned long result = 0; | |
262 | ||
263 | while (n--) { | |
264 | result = (result << 32) | **buf; | |
265 | (*buf)++; | |
266 | } | |
267 | return result; | |
268 | } | |
269 | ||
270 | /* | |
271 | * Figure out to which domain a cpu belongs and stick it there. | |
272 | * Return the id of the domain used. | |
273 | */ | |
2e5ce39d | 274 | static int __cpuinit numa_setup_cpu(unsigned long lcpu) |
1da177e4 | 275 | { |
cf950b7a | 276 | int nid = 0; |
1da177e4 LT |
277 | struct device_node *cpu = find_cpu_node(lcpu); |
278 | ||
279 | if (!cpu) { | |
280 | WARN_ON(1); | |
281 | goto out; | |
282 | } | |
283 | ||
953039c8 | 284 | nid = of_node_to_nid_single(cpu); |
1da177e4 | 285 | |
482ec7c4 NL |
286 | if (nid < 0 || !node_online(nid)) |
287 | nid = any_online_node(NODE_MASK_ALL); | |
1da177e4 | 288 | out: |
cf950b7a | 289 | map_cpu_to_node(lcpu, nid); |
1da177e4 LT |
290 | |
291 | of_node_put(cpu); | |
292 | ||
cf950b7a | 293 | return nid; |
1da177e4 LT |
294 | } |
295 | ||
74b85f37 | 296 | static int __cpuinit cpu_numa_callback(struct notifier_block *nfb, |
1da177e4 LT |
297 | unsigned long action, |
298 | void *hcpu) | |
299 | { | |
300 | unsigned long lcpu = (unsigned long)hcpu; | |
301 | int ret = NOTIFY_DONE; | |
302 | ||
303 | switch (action) { | |
304 | case CPU_UP_PREPARE: | |
8bb78442 | 305 | case CPU_UP_PREPARE_FROZEN: |
2b261227 | 306 | numa_setup_cpu(lcpu); |
1da177e4 LT |
307 | ret = NOTIFY_OK; |
308 | break; | |
309 | #ifdef CONFIG_HOTPLUG_CPU | |
310 | case CPU_DEAD: | |
8bb78442 | 311 | case CPU_DEAD_FROZEN: |
1da177e4 | 312 | case CPU_UP_CANCELED: |
8bb78442 | 313 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 LT |
314 | unmap_cpu_from_node(lcpu); |
315 | break; | |
316 | ret = NOTIFY_OK; | |
317 | #endif | |
318 | } | |
319 | return ret; | |
320 | } | |
321 | ||
322 | /* | |
323 | * Check and possibly modify a memory region to enforce the memory limit. | |
324 | * | |
325 | * Returns the size the region should have to enforce the memory limit. | |
326 | * This will either be the original value of size, a truncated value, | |
327 | * or zero. If the returned value of size is 0 the region should be | |
328 | * discarded as it lies wholy above the memory limit. | |
329 | */ | |
45fb6cea AB |
330 | static unsigned long __init numa_enforce_memory_limit(unsigned long start, |
331 | unsigned long size) | |
1da177e4 LT |
332 | { |
333 | /* | |
334 | * We use lmb_end_of_DRAM() in here instead of memory_limit because | |
335 | * we've already adjusted it for the limit and it takes care of | |
336 | * having memory holes below the limit. | |
337 | */ | |
1da177e4 LT |
338 | |
339 | if (! memory_limit) | |
340 | return size; | |
341 | ||
342 | if (start + size <= lmb_end_of_DRAM()) | |
343 | return size; | |
344 | ||
345 | if (start >= lmb_end_of_DRAM()) | |
346 | return 0; | |
347 | ||
348 | return lmb_end_of_DRAM() - start; | |
349 | } | |
350 | ||
0204568a PM |
351 | /* |
352 | * Extract NUMA information from the ibm,dynamic-reconfiguration-memory | |
353 | * node. This assumes n_mem_{addr,size}_cells have been set. | |
354 | */ | |
355 | static void __init parse_drconf_memory(struct device_node *memory) | |
356 | { | |
357 | const unsigned int *lm, *dm, *aa; | |
358 | unsigned int ls, ld, la; | |
359 | unsigned int n, aam, aalen; | |
b9c3fdb0 | 360 | unsigned long lmb_size, size, start; |
0204568a | 361 | int nid, default_nid = 0; |
b9c3fdb0 | 362 | unsigned int ai, flags; |
0204568a | 363 | |
e2eb6392 SR |
364 | lm = of_get_property(memory, "ibm,lmb-size", &ls); |
365 | dm = of_get_property(memory, "ibm,dynamic-memory", &ld); | |
366 | aa = of_get_property(memory, "ibm,associativity-lookup-arrays", &la); | |
0204568a PM |
367 | if (!lm || !dm || !aa || |
368 | ls < sizeof(unsigned int) || ld < sizeof(unsigned int) || | |
369 | la < 2 * sizeof(unsigned int)) | |
370 | return; | |
371 | ||
372 | lmb_size = read_n_cells(n_mem_size_cells, &lm); | |
373 | n = *dm++; /* number of LMBs */ | |
374 | aam = *aa++; /* number of associativity lists */ | |
375 | aalen = *aa++; /* length of each associativity list */ | |
376 | if (ld < (n * (n_mem_addr_cells + 4) + 1) * sizeof(unsigned int) || | |
377 | la < (aam * aalen + 2) * sizeof(unsigned int)) | |
378 | return; | |
379 | ||
380 | for (; n != 0; --n) { | |
381 | start = read_n_cells(n_mem_addr_cells, &dm); | |
382 | ai = dm[2]; | |
383 | flags = dm[3]; | |
384 | dm += 4; | |
385 | /* 0x80 == reserved, 0x8 = assigned to us */ | |
386 | if ((flags & 0x80) || !(flags & 0x8)) | |
387 | continue; | |
388 | nid = default_nid; | |
389 | /* flags & 0x40 means associativity index is invalid */ | |
390 | if (min_common_depth > 0 && min_common_depth <= aalen && | |
391 | (flags & 0x40) == 0 && ai < aam) { | |
392 | /* this is like of_node_to_nid_single */ | |
393 | nid = aa[ai * aalen + min_common_depth - 1]; | |
394 | if (nid == 0xffff || nid >= MAX_NUMNODES) | |
395 | nid = default_nid; | |
396 | } | |
1daa6d08 BS |
397 | |
398 | fake_numa_create_new_node(((start + lmb_size) >> PAGE_SHIFT), | |
399 | &nid); | |
0204568a PM |
400 | node_set_online(nid); |
401 | ||
402 | size = numa_enforce_memory_limit(start, lmb_size); | |
403 | if (!size) | |
404 | continue; | |
405 | ||
406 | add_active_range(nid, start >> PAGE_SHIFT, | |
407 | (start >> PAGE_SHIFT) + (size >> PAGE_SHIFT)); | |
408 | } | |
409 | } | |
410 | ||
1da177e4 LT |
411 | static int __init parse_numa_properties(void) |
412 | { | |
413 | struct device_node *cpu = NULL; | |
414 | struct device_node *memory = NULL; | |
482ec7c4 | 415 | int default_nid = 0; |
1da177e4 LT |
416 | unsigned long i; |
417 | ||
418 | if (numa_enabled == 0) { | |
419 | printk(KERN_WARNING "NUMA disabled by user\n"); | |
420 | return -1; | |
421 | } | |
422 | ||
1da177e4 LT |
423 | min_common_depth = find_min_common_depth(); |
424 | ||
1da177e4 LT |
425 | if (min_common_depth < 0) |
426 | return min_common_depth; | |
427 | ||
bf4b85b0 NL |
428 | dbg("NUMA associativity depth for CPU/Memory: %d\n", min_common_depth); |
429 | ||
1da177e4 | 430 | /* |
482ec7c4 NL |
431 | * Even though we connect cpus to numa domains later in SMP |
432 | * init, we need to know the node ids now. This is because | |
433 | * each node to be onlined must have NODE_DATA etc backing it. | |
1da177e4 | 434 | */ |
482ec7c4 | 435 | for_each_present_cpu(i) { |
cf950b7a | 436 | int nid; |
1da177e4 LT |
437 | |
438 | cpu = find_cpu_node(i); | |
482ec7c4 | 439 | BUG_ON(!cpu); |
953039c8 | 440 | nid = of_node_to_nid_single(cpu); |
482ec7c4 | 441 | of_node_put(cpu); |
1da177e4 | 442 | |
482ec7c4 NL |
443 | /* |
444 | * Don't fall back to default_nid yet -- we will plug | |
445 | * cpus into nodes once the memory scan has discovered | |
446 | * the topology. | |
447 | */ | |
448 | if (nid < 0) | |
449 | continue; | |
450 | node_set_online(nid); | |
1da177e4 LT |
451 | } |
452 | ||
237a0989 | 453 | get_n_mem_cells(&n_mem_addr_cells, &n_mem_size_cells); |
1da177e4 LT |
454 | memory = NULL; |
455 | while ((memory = of_find_node_by_type(memory, "memory")) != NULL) { | |
456 | unsigned long start; | |
457 | unsigned long size; | |
cf950b7a | 458 | int nid; |
1da177e4 | 459 | int ranges; |
a7f67bdf | 460 | const unsigned int *memcell_buf; |
1da177e4 LT |
461 | unsigned int len; |
462 | ||
e2eb6392 | 463 | memcell_buf = of_get_property(memory, |
ba759485 ME |
464 | "linux,usable-memory", &len); |
465 | if (!memcell_buf || len <= 0) | |
e2eb6392 | 466 | memcell_buf = of_get_property(memory, "reg", &len); |
1da177e4 LT |
467 | if (!memcell_buf || len <= 0) |
468 | continue; | |
469 | ||
cc5d0189 BH |
470 | /* ranges in cell */ |
471 | ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells); | |
1da177e4 LT |
472 | new_range: |
473 | /* these are order-sensitive, and modify the buffer pointer */ | |
237a0989 MK |
474 | start = read_n_cells(n_mem_addr_cells, &memcell_buf); |
475 | size = read_n_cells(n_mem_size_cells, &memcell_buf); | |
1da177e4 | 476 | |
482ec7c4 NL |
477 | /* |
478 | * Assumption: either all memory nodes or none will | |
479 | * have associativity properties. If none, then | |
480 | * everything goes to default_nid. | |
481 | */ | |
953039c8 | 482 | nid = of_node_to_nid_single(memory); |
482ec7c4 NL |
483 | if (nid < 0) |
484 | nid = default_nid; | |
1daa6d08 BS |
485 | |
486 | fake_numa_create_new_node(((start + size) >> PAGE_SHIFT), &nid); | |
482ec7c4 | 487 | node_set_online(nid); |
1da177e4 | 488 | |
45fb6cea | 489 | if (!(size = numa_enforce_memory_limit(start, size))) { |
1da177e4 LT |
490 | if (--ranges) |
491 | goto new_range; | |
492 | else | |
493 | continue; | |
494 | } | |
495 | ||
c67c3cb4 MG |
496 | add_active_range(nid, start >> PAGE_SHIFT, |
497 | (start >> PAGE_SHIFT) + (size >> PAGE_SHIFT)); | |
1da177e4 LT |
498 | |
499 | if (--ranges) | |
500 | goto new_range; | |
501 | } | |
502 | ||
0204568a PM |
503 | /* |
504 | * Now do the same thing for each LMB listed in the ibm,dynamic-memory | |
505 | * property in the ibm,dynamic-reconfiguration-memory node. | |
506 | */ | |
507 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); | |
508 | if (memory) | |
509 | parse_drconf_memory(memory); | |
510 | ||
1da177e4 LT |
511 | return 0; |
512 | } | |
513 | ||
514 | static void __init setup_nonnuma(void) | |
515 | { | |
516 | unsigned long top_of_ram = lmb_end_of_DRAM(); | |
517 | unsigned long total_ram = lmb_phys_mem_size(); | |
c67c3cb4 | 518 | unsigned long start_pfn, end_pfn; |
1daa6d08 | 519 | unsigned int i, nid = 0; |
1da177e4 | 520 | |
e110b281 | 521 | printk(KERN_DEBUG "Top of RAM: 0x%lx, Total RAM: 0x%lx\n", |
1da177e4 | 522 | top_of_ram, total_ram); |
e110b281 | 523 | printk(KERN_DEBUG "Memory hole size: %ldMB\n", |
1da177e4 LT |
524 | (top_of_ram - total_ram) >> 20); |
525 | ||
c67c3cb4 MG |
526 | for (i = 0; i < lmb.memory.cnt; ++i) { |
527 | start_pfn = lmb.memory.region[i].base >> PAGE_SHIFT; | |
528 | end_pfn = start_pfn + lmb_size_pages(&lmb.memory, i); | |
1daa6d08 BS |
529 | |
530 | fake_numa_create_new_node(end_pfn, &nid); | |
531 | add_active_range(nid, start_pfn, end_pfn); | |
532 | node_set_online(nid); | |
c67c3cb4 | 533 | } |
1da177e4 LT |
534 | } |
535 | ||
4b703a23 AB |
536 | void __init dump_numa_cpu_topology(void) |
537 | { | |
538 | unsigned int node; | |
539 | unsigned int cpu, count; | |
540 | ||
541 | if (min_common_depth == -1 || !numa_enabled) | |
542 | return; | |
543 | ||
544 | for_each_online_node(node) { | |
e110b281 | 545 | printk(KERN_DEBUG "Node %d CPUs:", node); |
4b703a23 AB |
546 | |
547 | count = 0; | |
548 | /* | |
549 | * If we used a CPU iterator here we would miss printing | |
550 | * the holes in the cpumap. | |
551 | */ | |
552 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
553 | if (cpu_isset(cpu, numa_cpumask_lookup_table[node])) { | |
554 | if (count == 0) | |
555 | printk(" %u", cpu); | |
556 | ++count; | |
557 | } else { | |
558 | if (count > 1) | |
559 | printk("-%u", cpu - 1); | |
560 | count = 0; | |
561 | } | |
562 | } | |
563 | ||
564 | if (count > 1) | |
565 | printk("-%u", NR_CPUS - 1); | |
566 | printk("\n"); | |
567 | } | |
568 | } | |
569 | ||
570 | static void __init dump_numa_memory_topology(void) | |
1da177e4 LT |
571 | { |
572 | unsigned int node; | |
573 | unsigned int count; | |
574 | ||
575 | if (min_common_depth == -1 || !numa_enabled) | |
576 | return; | |
577 | ||
578 | for_each_online_node(node) { | |
579 | unsigned long i; | |
580 | ||
e110b281 | 581 | printk(KERN_DEBUG "Node %d Memory:", node); |
1da177e4 LT |
582 | |
583 | count = 0; | |
584 | ||
45fb6cea AB |
585 | for (i = 0; i < lmb_end_of_DRAM(); |
586 | i += (1 << SECTION_SIZE_BITS)) { | |
587 | if (early_pfn_to_nid(i >> PAGE_SHIFT) == node) { | |
1da177e4 LT |
588 | if (count == 0) |
589 | printk(" 0x%lx", i); | |
590 | ++count; | |
591 | } else { | |
592 | if (count > 0) | |
593 | printk("-0x%lx", i); | |
594 | count = 0; | |
595 | } | |
596 | } | |
597 | ||
598 | if (count > 0) | |
599 | printk("-0x%lx", i); | |
600 | printk("\n"); | |
601 | } | |
1da177e4 LT |
602 | } |
603 | ||
604 | /* | |
605 | * Allocate some memory, satisfying the lmb or bootmem allocator where | |
606 | * required. nid is the preferred node and end is the physical address of | |
607 | * the highest address in the node. | |
608 | * | |
609 | * Returns the physical address of the memory. | |
610 | */ | |
45fb6cea AB |
611 | static void __init *careful_allocation(int nid, unsigned long size, |
612 | unsigned long align, | |
613 | unsigned long end_pfn) | |
1da177e4 | 614 | { |
45fb6cea | 615 | int new_nid; |
d7a5b2ff | 616 | unsigned long ret = __lmb_alloc_base(size, align, end_pfn << PAGE_SHIFT); |
1da177e4 LT |
617 | |
618 | /* retry over all memory */ | |
619 | if (!ret) | |
d7a5b2ff | 620 | ret = __lmb_alloc_base(size, align, lmb_end_of_DRAM()); |
1da177e4 LT |
621 | |
622 | if (!ret) | |
623 | panic("numa.c: cannot allocate %lu bytes on node %d", | |
624 | size, nid); | |
625 | ||
626 | /* | |
627 | * If the memory came from a previously allocated node, we must | |
628 | * retry with the bootmem allocator. | |
629 | */ | |
45fb6cea AB |
630 | new_nid = early_pfn_to_nid(ret >> PAGE_SHIFT); |
631 | if (new_nid < nid) { | |
632 | ret = (unsigned long)__alloc_bootmem_node(NODE_DATA(new_nid), | |
1da177e4 LT |
633 | size, align, 0); |
634 | ||
635 | if (!ret) | |
636 | panic("numa.c: cannot allocate %lu bytes on node %d", | |
45fb6cea | 637 | size, new_nid); |
1da177e4 | 638 | |
45fb6cea | 639 | ret = __pa(ret); |
1da177e4 LT |
640 | |
641 | dbg("alloc_bootmem %lx %lx\n", ret, size); | |
642 | } | |
643 | ||
45fb6cea | 644 | return (void *)ret; |
1da177e4 LT |
645 | } |
646 | ||
74b85f37 CS |
647 | static struct notifier_block __cpuinitdata ppc64_numa_nb = { |
648 | .notifier_call = cpu_numa_callback, | |
649 | .priority = 1 /* Must run before sched domains notifier. */ | |
650 | }; | |
651 | ||
1da177e4 LT |
652 | void __init do_init_bootmem(void) |
653 | { | |
654 | int nid; | |
45fb6cea | 655 | unsigned int i; |
1da177e4 LT |
656 | |
657 | min_low_pfn = 0; | |
658 | max_low_pfn = lmb_end_of_DRAM() >> PAGE_SHIFT; | |
659 | max_pfn = max_low_pfn; | |
660 | ||
661 | if (parse_numa_properties()) | |
662 | setup_nonnuma(); | |
663 | else | |
4b703a23 | 664 | dump_numa_memory_topology(); |
1da177e4 LT |
665 | |
666 | register_cpu_notifier(&ppc64_numa_nb); | |
2b261227 NL |
667 | cpu_numa_callback(&ppc64_numa_nb, CPU_UP_PREPARE, |
668 | (void *)(unsigned long)boot_cpuid); | |
1da177e4 LT |
669 | |
670 | for_each_online_node(nid) { | |
c67c3cb4 | 671 | unsigned long start_pfn, end_pfn; |
1da177e4 LT |
672 | unsigned long bootmem_paddr; |
673 | unsigned long bootmap_pages; | |
674 | ||
c67c3cb4 | 675 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); |
1da177e4 LT |
676 | |
677 | /* Allocate the node structure node local if possible */ | |
45fb6cea | 678 | NODE_DATA(nid) = careful_allocation(nid, |
1da177e4 | 679 | sizeof(struct pglist_data), |
45fb6cea AB |
680 | SMP_CACHE_BYTES, end_pfn); |
681 | NODE_DATA(nid) = __va(NODE_DATA(nid)); | |
1da177e4 LT |
682 | memset(NODE_DATA(nid), 0, sizeof(struct pglist_data)); |
683 | ||
684 | dbg("node %d\n", nid); | |
685 | dbg("NODE_DATA() = %p\n", NODE_DATA(nid)); | |
686 | ||
687 | NODE_DATA(nid)->bdata = &plat_node_bdata[nid]; | |
45fb6cea AB |
688 | NODE_DATA(nid)->node_start_pfn = start_pfn; |
689 | NODE_DATA(nid)->node_spanned_pages = end_pfn - start_pfn; | |
1da177e4 LT |
690 | |
691 | if (NODE_DATA(nid)->node_spanned_pages == 0) | |
692 | continue; | |
693 | ||
45fb6cea AB |
694 | dbg("start_paddr = %lx\n", start_pfn << PAGE_SHIFT); |
695 | dbg("end_paddr = %lx\n", end_pfn << PAGE_SHIFT); | |
1da177e4 | 696 | |
45fb6cea AB |
697 | bootmap_pages = bootmem_bootmap_pages(end_pfn - start_pfn); |
698 | bootmem_paddr = (unsigned long)careful_allocation(nid, | |
699 | bootmap_pages << PAGE_SHIFT, | |
700 | PAGE_SIZE, end_pfn); | |
701 | memset(__va(bootmem_paddr), 0, bootmap_pages << PAGE_SHIFT); | |
1da177e4 | 702 | |
1da177e4 LT |
703 | dbg("bootmap_paddr = %lx\n", bootmem_paddr); |
704 | ||
705 | init_bootmem_node(NODE_DATA(nid), bootmem_paddr >> PAGE_SHIFT, | |
45fb6cea | 706 | start_pfn, end_pfn); |
1da177e4 | 707 | |
c67c3cb4 | 708 | free_bootmem_with_active_regions(nid, end_pfn); |
1da177e4 | 709 | |
45fb6cea | 710 | /* Mark reserved regions on this node */ |
1da177e4 | 711 | for (i = 0; i < lmb.reserved.cnt; i++) { |
180379dc | 712 | unsigned long physbase = lmb.reserved.region[i].base; |
1da177e4 | 713 | unsigned long size = lmb.reserved.region[i].size; |
45fb6cea AB |
714 | unsigned long start_paddr = start_pfn << PAGE_SHIFT; |
715 | unsigned long end_paddr = end_pfn << PAGE_SHIFT; | |
1da177e4 | 716 | |
45fb6cea AB |
717 | if (early_pfn_to_nid(physbase >> PAGE_SHIFT) != nid && |
718 | early_pfn_to_nid((physbase+size-1) >> PAGE_SHIFT) != nid) | |
1da177e4 LT |
719 | continue; |
720 | ||
721 | if (physbase < end_paddr && | |
722 | (physbase+size) > start_paddr) { | |
723 | /* overlaps */ | |
724 | if (physbase < start_paddr) { | |
725 | size -= start_paddr - physbase; | |
726 | physbase = start_paddr; | |
727 | } | |
728 | ||
729 | if (size > end_paddr - physbase) | |
730 | size = end_paddr - physbase; | |
731 | ||
732 | dbg("reserve_bootmem %lx %lx\n", physbase, | |
733 | size); | |
734 | reserve_bootmem_node(NODE_DATA(nid), physbase, | |
72a7fe39 | 735 | size, BOOTMEM_DEFAULT); |
1da177e4 LT |
736 | } |
737 | } | |
802f192e | 738 | |
c67c3cb4 | 739 | sparse_memory_present_with_active_regions(nid); |
1da177e4 LT |
740 | } |
741 | } | |
742 | ||
743 | void __init paging_init(void) | |
744 | { | |
6391af17 MG |
745 | unsigned long max_zone_pfns[MAX_NR_ZONES]; |
746 | memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); | |
747 | max_zone_pfns[ZONE_DMA] = lmb_end_of_DRAM() >> PAGE_SHIFT; | |
c67c3cb4 | 748 | free_area_init_nodes(max_zone_pfns); |
1da177e4 LT |
749 | } |
750 | ||
751 | static int __init early_numa(char *p) | |
752 | { | |
753 | if (!p) | |
754 | return 0; | |
755 | ||
756 | if (strstr(p, "off")) | |
757 | numa_enabled = 0; | |
758 | ||
759 | if (strstr(p, "debug")) | |
760 | numa_debug = 1; | |
761 | ||
1daa6d08 BS |
762 | p = strstr(p, "fake="); |
763 | if (p) | |
764 | cmdline = p + strlen("fake="); | |
765 | ||
1da177e4 LT |
766 | return 0; |
767 | } | |
768 | early_param("numa", early_numa); | |
237a0989 MK |
769 | |
770 | #ifdef CONFIG_MEMORY_HOTPLUG | |
771 | /* | |
772 | * Find the node associated with a hot added memory section. Section | |
773 | * corresponds to a SPARSEMEM section, not an LMB. It is assumed that | |
774 | * sections are fully contained within a single LMB. | |
775 | */ | |
776 | int hot_add_scn_to_nid(unsigned long scn_addr) | |
777 | { | |
778 | struct device_node *memory = NULL; | |
b226e462 | 779 | nodemask_t nodes; |
482ec7c4 | 780 | int default_nid = any_online_node(NODE_MASK_ALL); |
069007ae | 781 | int nid; |
237a0989 MK |
782 | |
783 | if (!numa_enabled || (min_common_depth < 0)) | |
482ec7c4 | 784 | return default_nid; |
237a0989 MK |
785 | |
786 | while ((memory = of_find_node_by_type(memory, "memory")) != NULL) { | |
787 | unsigned long start, size; | |
b226e462 | 788 | int ranges; |
a7f67bdf | 789 | const unsigned int *memcell_buf; |
237a0989 MK |
790 | unsigned int len; |
791 | ||
e2eb6392 | 792 | memcell_buf = of_get_property(memory, "reg", &len); |
237a0989 MK |
793 | if (!memcell_buf || len <= 0) |
794 | continue; | |
795 | ||
cc5d0189 BH |
796 | /* ranges in cell */ |
797 | ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells); | |
237a0989 MK |
798 | ha_new_range: |
799 | start = read_n_cells(n_mem_addr_cells, &memcell_buf); | |
800 | size = read_n_cells(n_mem_size_cells, &memcell_buf); | |
953039c8 | 801 | nid = of_node_to_nid_single(memory); |
237a0989 MK |
802 | |
803 | /* Domains not present at boot default to 0 */ | |
482ec7c4 NL |
804 | if (nid < 0 || !node_online(nid)) |
805 | nid = default_nid; | |
237a0989 MK |
806 | |
807 | if ((scn_addr >= start) && (scn_addr < (start + size))) { | |
808 | of_node_put(memory); | |
cf950b7a | 809 | goto got_nid; |
237a0989 MK |
810 | } |
811 | ||
812 | if (--ranges) /* process all ranges in cell */ | |
813 | goto ha_new_range; | |
814 | } | |
237a0989 | 815 | BUG(); /* section address should be found above */ |
069007ae | 816 | return 0; |
b226e462 MK |
817 | |
818 | /* Temporary code to ensure that returned node is not empty */ | |
cf950b7a | 819 | got_nid: |
b226e462 | 820 | nodes_setall(nodes); |
cf950b7a NL |
821 | while (NODE_DATA(nid)->node_spanned_pages == 0) { |
822 | node_clear(nid, nodes); | |
823 | nid = any_online_node(nodes); | |
b226e462 | 824 | } |
cf950b7a | 825 | return nid; |
237a0989 MK |
826 | } |
827 | #endif /* CONFIG_MEMORY_HOTPLUG */ |