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