]>
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 | 41 | |
2e5ce39d | 42 | static void __cpuinit map_cpu_to_node(int cpu, int node) |
1da177e4 LT |
43 | { |
44 | numa_cpu_lookup_table[cpu] = node; | |
45fb6cea | 45 | |
bf4b85b0 NL |
46 | dbg("adding cpu %d to node %d\n", cpu, node); |
47 | ||
45fb6cea | 48 | if (!(cpu_isset(cpu, numa_cpumask_lookup_table[node]))) |
1da177e4 | 49 | cpu_set(cpu, numa_cpumask_lookup_table[node]); |
1da177e4 LT |
50 | } |
51 | ||
52 | #ifdef CONFIG_HOTPLUG_CPU | |
53 | static void unmap_cpu_from_node(unsigned long cpu) | |
54 | { | |
55 | int node = numa_cpu_lookup_table[cpu]; | |
56 | ||
57 | dbg("removing cpu %lu from node %d\n", cpu, node); | |
58 | ||
59 | if (cpu_isset(cpu, numa_cpumask_lookup_table[node])) { | |
60 | cpu_clear(cpu, numa_cpumask_lookup_table[node]); | |
1da177e4 LT |
61 | } else { |
62 | printk(KERN_ERR "WARNING: cpu %lu not found in node %d\n", | |
63 | cpu, node); | |
64 | } | |
65 | } | |
66 | #endif /* CONFIG_HOTPLUG_CPU */ | |
67 | ||
2e5ce39d | 68 | static struct device_node * __cpuinit find_cpu_node(unsigned int cpu) |
1da177e4 LT |
69 | { |
70 | unsigned int hw_cpuid = get_hard_smp_processor_id(cpu); | |
71 | struct device_node *cpu_node = NULL; | |
a7f67bdf | 72 | const unsigned int *interrupt_server, *reg; |
1da177e4 LT |
73 | int len; |
74 | ||
75 | while ((cpu_node = of_find_node_by_type(cpu_node, "cpu")) != NULL) { | |
76 | /* Try interrupt server first */ | |
e2eb6392 | 77 | interrupt_server = of_get_property(cpu_node, |
1da177e4 LT |
78 | "ibm,ppc-interrupt-server#s", &len); |
79 | ||
80 | len = len / sizeof(u32); | |
81 | ||
82 | if (interrupt_server && (len > 0)) { | |
83 | while (len--) { | |
84 | if (interrupt_server[len] == hw_cpuid) | |
85 | return cpu_node; | |
86 | } | |
87 | } else { | |
e2eb6392 | 88 | reg = of_get_property(cpu_node, "reg", &len); |
1da177e4 LT |
89 | if (reg && (len > 0) && (reg[0] == hw_cpuid)) |
90 | return cpu_node; | |
91 | } | |
92 | } | |
93 | ||
94 | return NULL; | |
95 | } | |
96 | ||
97 | /* must hold reference to node during call */ | |
a7f67bdf | 98 | static const int *of_get_associativity(struct device_node *dev) |
1da177e4 | 99 | { |
e2eb6392 | 100 | return of_get_property(dev, "ibm,associativity", NULL); |
1da177e4 LT |
101 | } |
102 | ||
482ec7c4 NL |
103 | /* Returns nid in the range [0..MAX_NUMNODES-1], or -1 if no useful numa |
104 | * info is found. | |
105 | */ | |
953039c8 | 106 | static int of_node_to_nid_single(struct device_node *device) |
1da177e4 | 107 | { |
482ec7c4 | 108 | int nid = -1; |
a7f67bdf | 109 | const unsigned int *tmp; |
1da177e4 LT |
110 | |
111 | if (min_common_depth == -1) | |
482ec7c4 | 112 | goto out; |
1da177e4 LT |
113 | |
114 | tmp = of_get_associativity(device); | |
482ec7c4 NL |
115 | if (!tmp) |
116 | goto out; | |
117 | ||
118 | if (tmp[0] >= min_common_depth) | |
cf950b7a | 119 | nid = tmp[min_common_depth]; |
bc16a759 NL |
120 | |
121 | /* POWER4 LPAR uses 0xffff as invalid node */ | |
482ec7c4 NL |
122 | if (nid == 0xffff || nid >= MAX_NUMNODES) |
123 | nid = -1; | |
124 | out: | |
cf950b7a | 125 | return nid; |
1da177e4 LT |
126 | } |
127 | ||
953039c8 JK |
128 | /* Walk the device tree upwards, looking for an associativity id */ |
129 | int of_node_to_nid(struct device_node *device) | |
130 | { | |
131 | struct device_node *tmp; | |
132 | int nid = -1; | |
133 | ||
134 | of_node_get(device); | |
135 | while (device) { | |
136 | nid = of_node_to_nid_single(device); | |
137 | if (nid != -1) | |
138 | break; | |
139 | ||
140 | tmp = device; | |
141 | device = of_get_parent(tmp); | |
142 | of_node_put(tmp); | |
143 | } | |
144 | of_node_put(device); | |
145 | ||
146 | return nid; | |
147 | } | |
148 | EXPORT_SYMBOL_GPL(of_node_to_nid); | |
149 | ||
1da177e4 LT |
150 | /* |
151 | * In theory, the "ibm,associativity" property may contain multiple | |
152 | * associativity lists because a resource may be multiply connected | |
153 | * into the machine. This resource then has different associativity | |
154 | * characteristics relative to its multiple connections. We ignore | |
155 | * this for now. We also assume that all cpu and memory sets have | |
156 | * their distances represented at a common level. This won't be | |
1b3c3714 | 157 | * true for hierarchical NUMA. |
1da177e4 LT |
158 | * |
159 | * In any case the ibm,associativity-reference-points should give | |
160 | * the correct depth for a normal NUMA system. | |
161 | * | |
162 | * - Dave Hansen <haveblue@us.ibm.com> | |
163 | */ | |
164 | static int __init find_min_common_depth(void) | |
165 | { | |
166 | int depth; | |
a7f67bdf | 167 | const unsigned int *ref_points; |
1da177e4 LT |
168 | struct device_node *rtas_root; |
169 | unsigned int len; | |
170 | ||
171 | rtas_root = of_find_node_by_path("/rtas"); | |
172 | ||
173 | if (!rtas_root) | |
174 | return -1; | |
175 | ||
176 | /* | |
177 | * this property is 2 32-bit integers, each representing a level of | |
178 | * depth in the associativity nodes. The first is for an SMP | |
179 | * configuration (should be all 0's) and the second is for a normal | |
180 | * NUMA configuration. | |
181 | */ | |
e2eb6392 | 182 | ref_points = of_get_property(rtas_root, |
1da177e4 LT |
183 | "ibm,associativity-reference-points", &len); |
184 | ||
185 | if ((len >= 1) && ref_points) { | |
186 | depth = ref_points[1]; | |
187 | } else { | |
bf4b85b0 | 188 | dbg("NUMA: ibm,associativity-reference-points not found.\n"); |
1da177e4 LT |
189 | depth = -1; |
190 | } | |
191 | of_node_put(rtas_root); | |
192 | ||
193 | return depth; | |
194 | } | |
195 | ||
84c9fdd1 | 196 | static void __init get_n_mem_cells(int *n_addr_cells, int *n_size_cells) |
1da177e4 LT |
197 | { |
198 | struct device_node *memory = NULL; | |
1da177e4 LT |
199 | |
200 | memory = of_find_node_by_type(memory, "memory"); | |
54c23310 | 201 | if (!memory) |
84c9fdd1 | 202 | panic("numa.c: No memory nodes found!"); |
54c23310 | 203 | |
a8bda5dd | 204 | *n_addr_cells = of_n_addr_cells(memory); |
9213feea | 205 | *n_size_cells = of_n_size_cells(memory); |
84c9fdd1 | 206 | of_node_put(memory); |
1da177e4 LT |
207 | } |
208 | ||
a7f67bdf | 209 | static unsigned long __devinit read_n_cells(int n, const unsigned int **buf) |
1da177e4 LT |
210 | { |
211 | unsigned long result = 0; | |
212 | ||
213 | while (n--) { | |
214 | result = (result << 32) | **buf; | |
215 | (*buf)++; | |
216 | } | |
217 | return result; | |
218 | } | |
219 | ||
220 | /* | |
221 | * Figure out to which domain a cpu belongs and stick it there. | |
222 | * Return the id of the domain used. | |
223 | */ | |
2e5ce39d | 224 | static int __cpuinit numa_setup_cpu(unsigned long lcpu) |
1da177e4 | 225 | { |
cf950b7a | 226 | int nid = 0; |
1da177e4 LT |
227 | struct device_node *cpu = find_cpu_node(lcpu); |
228 | ||
229 | if (!cpu) { | |
230 | WARN_ON(1); | |
231 | goto out; | |
232 | } | |
233 | ||
953039c8 | 234 | nid = of_node_to_nid_single(cpu); |
1da177e4 | 235 | |
482ec7c4 NL |
236 | if (nid < 0 || !node_online(nid)) |
237 | nid = any_online_node(NODE_MASK_ALL); | |
1da177e4 | 238 | out: |
cf950b7a | 239 | map_cpu_to_node(lcpu, nid); |
1da177e4 LT |
240 | |
241 | of_node_put(cpu); | |
242 | ||
cf950b7a | 243 | return nid; |
1da177e4 LT |
244 | } |
245 | ||
74b85f37 | 246 | static int __cpuinit cpu_numa_callback(struct notifier_block *nfb, |
1da177e4 LT |
247 | unsigned long action, |
248 | void *hcpu) | |
249 | { | |
250 | unsigned long lcpu = (unsigned long)hcpu; | |
251 | int ret = NOTIFY_DONE; | |
252 | ||
253 | switch (action) { | |
254 | case CPU_UP_PREPARE: | |
8bb78442 | 255 | case CPU_UP_PREPARE_FROZEN: |
2b261227 | 256 | numa_setup_cpu(lcpu); |
1da177e4 LT |
257 | ret = NOTIFY_OK; |
258 | break; | |
259 | #ifdef CONFIG_HOTPLUG_CPU | |
260 | case CPU_DEAD: | |
8bb78442 | 261 | case CPU_DEAD_FROZEN: |
1da177e4 | 262 | case CPU_UP_CANCELED: |
8bb78442 | 263 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 LT |
264 | unmap_cpu_from_node(lcpu); |
265 | break; | |
266 | ret = NOTIFY_OK; | |
267 | #endif | |
268 | } | |
269 | return ret; | |
270 | } | |
271 | ||
272 | /* | |
273 | * Check and possibly modify a memory region to enforce the memory limit. | |
274 | * | |
275 | * Returns the size the region should have to enforce the memory limit. | |
276 | * This will either be the original value of size, a truncated value, | |
277 | * or zero. If the returned value of size is 0 the region should be | |
278 | * discarded as it lies wholy above the memory limit. | |
279 | */ | |
45fb6cea AB |
280 | static unsigned long __init numa_enforce_memory_limit(unsigned long start, |
281 | unsigned long size) | |
1da177e4 LT |
282 | { |
283 | /* | |
284 | * We use lmb_end_of_DRAM() in here instead of memory_limit because | |
285 | * we've already adjusted it for the limit and it takes care of | |
286 | * having memory holes below the limit. | |
287 | */ | |
1da177e4 LT |
288 | |
289 | if (! memory_limit) | |
290 | return size; | |
291 | ||
292 | if (start + size <= lmb_end_of_DRAM()) | |
293 | return size; | |
294 | ||
295 | if (start >= lmb_end_of_DRAM()) | |
296 | return 0; | |
297 | ||
298 | return lmb_end_of_DRAM() - start; | |
299 | } | |
300 | ||
0204568a PM |
301 | /* |
302 | * Extract NUMA information from the ibm,dynamic-reconfiguration-memory | |
303 | * node. This assumes n_mem_{addr,size}_cells have been set. | |
304 | */ | |
305 | static void __init parse_drconf_memory(struct device_node *memory) | |
306 | { | |
307 | const unsigned int *lm, *dm, *aa; | |
308 | unsigned int ls, ld, la; | |
309 | unsigned int n, aam, aalen; | |
310 | unsigned long lmb_size, size; | |
311 | int nid, default_nid = 0; | |
312 | unsigned int start, ai, flags; | |
313 | ||
e2eb6392 SR |
314 | lm = of_get_property(memory, "ibm,lmb-size", &ls); |
315 | dm = of_get_property(memory, "ibm,dynamic-memory", &ld); | |
316 | aa = of_get_property(memory, "ibm,associativity-lookup-arrays", &la); | |
0204568a PM |
317 | if (!lm || !dm || !aa || |
318 | ls < sizeof(unsigned int) || ld < sizeof(unsigned int) || | |
319 | la < 2 * sizeof(unsigned int)) | |
320 | return; | |
321 | ||
322 | lmb_size = read_n_cells(n_mem_size_cells, &lm); | |
323 | n = *dm++; /* number of LMBs */ | |
324 | aam = *aa++; /* number of associativity lists */ | |
325 | aalen = *aa++; /* length of each associativity list */ | |
326 | if (ld < (n * (n_mem_addr_cells + 4) + 1) * sizeof(unsigned int) || | |
327 | la < (aam * aalen + 2) * sizeof(unsigned int)) | |
328 | return; | |
329 | ||
330 | for (; n != 0; --n) { | |
331 | start = read_n_cells(n_mem_addr_cells, &dm); | |
332 | ai = dm[2]; | |
333 | flags = dm[3]; | |
334 | dm += 4; | |
335 | /* 0x80 == reserved, 0x8 = assigned to us */ | |
336 | if ((flags & 0x80) || !(flags & 0x8)) | |
337 | continue; | |
338 | nid = default_nid; | |
339 | /* flags & 0x40 means associativity index is invalid */ | |
340 | if (min_common_depth > 0 && min_common_depth <= aalen && | |
341 | (flags & 0x40) == 0 && ai < aam) { | |
342 | /* this is like of_node_to_nid_single */ | |
343 | nid = aa[ai * aalen + min_common_depth - 1]; | |
344 | if (nid == 0xffff || nid >= MAX_NUMNODES) | |
345 | nid = default_nid; | |
346 | } | |
347 | node_set_online(nid); | |
348 | ||
349 | size = numa_enforce_memory_limit(start, lmb_size); | |
350 | if (!size) | |
351 | continue; | |
352 | ||
353 | add_active_range(nid, start >> PAGE_SHIFT, | |
354 | (start >> PAGE_SHIFT) + (size >> PAGE_SHIFT)); | |
355 | } | |
356 | } | |
357 | ||
1da177e4 LT |
358 | static int __init parse_numa_properties(void) |
359 | { | |
360 | struct device_node *cpu = NULL; | |
361 | struct device_node *memory = NULL; | |
482ec7c4 | 362 | int default_nid = 0; |
1da177e4 LT |
363 | unsigned long i; |
364 | ||
365 | if (numa_enabled == 0) { | |
366 | printk(KERN_WARNING "NUMA disabled by user\n"); | |
367 | return -1; | |
368 | } | |
369 | ||
1da177e4 LT |
370 | min_common_depth = find_min_common_depth(); |
371 | ||
1da177e4 LT |
372 | if (min_common_depth < 0) |
373 | return min_common_depth; | |
374 | ||
bf4b85b0 NL |
375 | dbg("NUMA associativity depth for CPU/Memory: %d\n", min_common_depth); |
376 | ||
1da177e4 | 377 | /* |
482ec7c4 NL |
378 | * Even though we connect cpus to numa domains later in SMP |
379 | * init, we need to know the node ids now. This is because | |
380 | * each node to be onlined must have NODE_DATA etc backing it. | |
1da177e4 | 381 | */ |
482ec7c4 | 382 | for_each_present_cpu(i) { |
cf950b7a | 383 | int nid; |
1da177e4 LT |
384 | |
385 | cpu = find_cpu_node(i); | |
482ec7c4 | 386 | BUG_ON(!cpu); |
953039c8 | 387 | nid = of_node_to_nid_single(cpu); |
482ec7c4 | 388 | of_node_put(cpu); |
1da177e4 | 389 | |
482ec7c4 NL |
390 | /* |
391 | * Don't fall back to default_nid yet -- we will plug | |
392 | * cpus into nodes once the memory scan has discovered | |
393 | * the topology. | |
394 | */ | |
395 | if (nid < 0) | |
396 | continue; | |
397 | node_set_online(nid); | |
1da177e4 LT |
398 | } |
399 | ||
237a0989 | 400 | get_n_mem_cells(&n_mem_addr_cells, &n_mem_size_cells); |
1da177e4 LT |
401 | memory = NULL; |
402 | while ((memory = of_find_node_by_type(memory, "memory")) != NULL) { | |
403 | unsigned long start; | |
404 | unsigned long size; | |
cf950b7a | 405 | int nid; |
1da177e4 | 406 | int ranges; |
a7f67bdf | 407 | const unsigned int *memcell_buf; |
1da177e4 LT |
408 | unsigned int len; |
409 | ||
e2eb6392 | 410 | memcell_buf = of_get_property(memory, |
ba759485 ME |
411 | "linux,usable-memory", &len); |
412 | if (!memcell_buf || len <= 0) | |
e2eb6392 | 413 | memcell_buf = of_get_property(memory, "reg", &len); |
1da177e4 LT |
414 | if (!memcell_buf || len <= 0) |
415 | continue; | |
416 | ||
cc5d0189 BH |
417 | /* ranges in cell */ |
418 | ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells); | |
1da177e4 LT |
419 | new_range: |
420 | /* these are order-sensitive, and modify the buffer pointer */ | |
237a0989 MK |
421 | start = read_n_cells(n_mem_addr_cells, &memcell_buf); |
422 | size = read_n_cells(n_mem_size_cells, &memcell_buf); | |
1da177e4 | 423 | |
482ec7c4 NL |
424 | /* |
425 | * Assumption: either all memory nodes or none will | |
426 | * have associativity properties. If none, then | |
427 | * everything goes to default_nid. | |
428 | */ | |
953039c8 | 429 | nid = of_node_to_nid_single(memory); |
482ec7c4 NL |
430 | if (nid < 0) |
431 | nid = default_nid; | |
432 | node_set_online(nid); | |
1da177e4 | 433 | |
45fb6cea | 434 | if (!(size = numa_enforce_memory_limit(start, size))) { |
1da177e4 LT |
435 | if (--ranges) |
436 | goto new_range; | |
437 | else | |
438 | continue; | |
439 | } | |
440 | ||
c67c3cb4 MG |
441 | add_active_range(nid, start >> PAGE_SHIFT, |
442 | (start >> PAGE_SHIFT) + (size >> PAGE_SHIFT)); | |
1da177e4 LT |
443 | |
444 | if (--ranges) | |
445 | goto new_range; | |
446 | } | |
447 | ||
0204568a PM |
448 | /* |
449 | * Now do the same thing for each LMB listed in the ibm,dynamic-memory | |
450 | * property in the ibm,dynamic-reconfiguration-memory node. | |
451 | */ | |
452 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); | |
453 | if (memory) | |
454 | parse_drconf_memory(memory); | |
455 | ||
1da177e4 LT |
456 | return 0; |
457 | } | |
458 | ||
459 | static void __init setup_nonnuma(void) | |
460 | { | |
461 | unsigned long top_of_ram = lmb_end_of_DRAM(); | |
462 | unsigned long total_ram = lmb_phys_mem_size(); | |
c67c3cb4 | 463 | unsigned long start_pfn, end_pfn; |
fb6d73d3 | 464 | unsigned int i; |
1da177e4 | 465 | |
e110b281 | 466 | printk(KERN_DEBUG "Top of RAM: 0x%lx, Total RAM: 0x%lx\n", |
1da177e4 | 467 | top_of_ram, total_ram); |
e110b281 | 468 | printk(KERN_DEBUG "Memory hole size: %ldMB\n", |
1da177e4 LT |
469 | (top_of_ram - total_ram) >> 20); |
470 | ||
c67c3cb4 MG |
471 | for (i = 0; i < lmb.memory.cnt; ++i) { |
472 | start_pfn = lmb.memory.region[i].base >> PAGE_SHIFT; | |
473 | end_pfn = start_pfn + lmb_size_pages(&lmb.memory, i); | |
474 | add_active_range(0, start_pfn, end_pfn); | |
475 | } | |
1da177e4 | 476 | node_set_online(0); |
1da177e4 LT |
477 | } |
478 | ||
4b703a23 AB |
479 | void __init dump_numa_cpu_topology(void) |
480 | { | |
481 | unsigned int node; | |
482 | unsigned int cpu, count; | |
483 | ||
484 | if (min_common_depth == -1 || !numa_enabled) | |
485 | return; | |
486 | ||
487 | for_each_online_node(node) { | |
e110b281 | 488 | printk(KERN_DEBUG "Node %d CPUs:", node); |
4b703a23 AB |
489 | |
490 | count = 0; | |
491 | /* | |
492 | * If we used a CPU iterator here we would miss printing | |
493 | * the holes in the cpumap. | |
494 | */ | |
495 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
496 | if (cpu_isset(cpu, numa_cpumask_lookup_table[node])) { | |
497 | if (count == 0) | |
498 | printk(" %u", cpu); | |
499 | ++count; | |
500 | } else { | |
501 | if (count > 1) | |
502 | printk("-%u", cpu - 1); | |
503 | count = 0; | |
504 | } | |
505 | } | |
506 | ||
507 | if (count > 1) | |
508 | printk("-%u", NR_CPUS - 1); | |
509 | printk("\n"); | |
510 | } | |
511 | } | |
512 | ||
513 | static void __init dump_numa_memory_topology(void) | |
1da177e4 LT |
514 | { |
515 | unsigned int node; | |
516 | unsigned int count; | |
517 | ||
518 | if (min_common_depth == -1 || !numa_enabled) | |
519 | return; | |
520 | ||
521 | for_each_online_node(node) { | |
522 | unsigned long i; | |
523 | ||
e110b281 | 524 | printk(KERN_DEBUG "Node %d Memory:", node); |
1da177e4 LT |
525 | |
526 | count = 0; | |
527 | ||
45fb6cea AB |
528 | for (i = 0; i < lmb_end_of_DRAM(); |
529 | i += (1 << SECTION_SIZE_BITS)) { | |
530 | if (early_pfn_to_nid(i >> PAGE_SHIFT) == node) { | |
1da177e4 LT |
531 | if (count == 0) |
532 | printk(" 0x%lx", i); | |
533 | ++count; | |
534 | } else { | |
535 | if (count > 0) | |
536 | printk("-0x%lx", i); | |
537 | count = 0; | |
538 | } | |
539 | } | |
540 | ||
541 | if (count > 0) | |
542 | printk("-0x%lx", i); | |
543 | printk("\n"); | |
544 | } | |
1da177e4 LT |
545 | } |
546 | ||
547 | /* | |
548 | * Allocate some memory, satisfying the lmb or bootmem allocator where | |
549 | * required. nid is the preferred node and end is the physical address of | |
550 | * the highest address in the node. | |
551 | * | |
552 | * Returns the physical address of the memory. | |
553 | */ | |
45fb6cea AB |
554 | static void __init *careful_allocation(int nid, unsigned long size, |
555 | unsigned long align, | |
556 | unsigned long end_pfn) | |
1da177e4 | 557 | { |
45fb6cea | 558 | int new_nid; |
d7a5b2ff | 559 | unsigned long ret = __lmb_alloc_base(size, align, end_pfn << PAGE_SHIFT); |
1da177e4 LT |
560 | |
561 | /* retry over all memory */ | |
562 | if (!ret) | |
d7a5b2ff | 563 | ret = __lmb_alloc_base(size, align, lmb_end_of_DRAM()); |
1da177e4 LT |
564 | |
565 | if (!ret) | |
566 | panic("numa.c: cannot allocate %lu bytes on node %d", | |
567 | size, nid); | |
568 | ||
569 | /* | |
570 | * If the memory came from a previously allocated node, we must | |
571 | * retry with the bootmem allocator. | |
572 | */ | |
45fb6cea AB |
573 | new_nid = early_pfn_to_nid(ret >> PAGE_SHIFT); |
574 | if (new_nid < nid) { | |
575 | ret = (unsigned long)__alloc_bootmem_node(NODE_DATA(new_nid), | |
1da177e4 LT |
576 | size, align, 0); |
577 | ||
578 | if (!ret) | |
579 | panic("numa.c: cannot allocate %lu bytes on node %d", | |
45fb6cea | 580 | size, new_nid); |
1da177e4 | 581 | |
45fb6cea | 582 | ret = __pa(ret); |
1da177e4 LT |
583 | |
584 | dbg("alloc_bootmem %lx %lx\n", ret, size); | |
585 | } | |
586 | ||
45fb6cea | 587 | return (void *)ret; |
1da177e4 LT |
588 | } |
589 | ||
74b85f37 CS |
590 | static struct notifier_block __cpuinitdata ppc64_numa_nb = { |
591 | .notifier_call = cpu_numa_callback, | |
592 | .priority = 1 /* Must run before sched domains notifier. */ | |
593 | }; | |
594 | ||
1da177e4 LT |
595 | void __init do_init_bootmem(void) |
596 | { | |
597 | int nid; | |
45fb6cea | 598 | unsigned int i; |
1da177e4 LT |
599 | |
600 | min_low_pfn = 0; | |
601 | max_low_pfn = lmb_end_of_DRAM() >> PAGE_SHIFT; | |
602 | max_pfn = max_low_pfn; | |
603 | ||
604 | if (parse_numa_properties()) | |
605 | setup_nonnuma(); | |
606 | else | |
4b703a23 | 607 | dump_numa_memory_topology(); |
1da177e4 LT |
608 | |
609 | register_cpu_notifier(&ppc64_numa_nb); | |
2b261227 NL |
610 | cpu_numa_callback(&ppc64_numa_nb, CPU_UP_PREPARE, |
611 | (void *)(unsigned long)boot_cpuid); | |
1da177e4 LT |
612 | |
613 | for_each_online_node(nid) { | |
c67c3cb4 | 614 | unsigned long start_pfn, end_pfn; |
1da177e4 LT |
615 | unsigned long bootmem_paddr; |
616 | unsigned long bootmap_pages; | |
617 | ||
c67c3cb4 | 618 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); |
1da177e4 LT |
619 | |
620 | /* Allocate the node structure node local if possible */ | |
45fb6cea | 621 | NODE_DATA(nid) = careful_allocation(nid, |
1da177e4 | 622 | sizeof(struct pglist_data), |
45fb6cea AB |
623 | SMP_CACHE_BYTES, end_pfn); |
624 | NODE_DATA(nid) = __va(NODE_DATA(nid)); | |
1da177e4 LT |
625 | memset(NODE_DATA(nid), 0, sizeof(struct pglist_data)); |
626 | ||
627 | dbg("node %d\n", nid); | |
628 | dbg("NODE_DATA() = %p\n", NODE_DATA(nid)); | |
629 | ||
630 | NODE_DATA(nid)->bdata = &plat_node_bdata[nid]; | |
45fb6cea AB |
631 | NODE_DATA(nid)->node_start_pfn = start_pfn; |
632 | NODE_DATA(nid)->node_spanned_pages = end_pfn - start_pfn; | |
1da177e4 LT |
633 | |
634 | if (NODE_DATA(nid)->node_spanned_pages == 0) | |
635 | continue; | |
636 | ||
45fb6cea AB |
637 | dbg("start_paddr = %lx\n", start_pfn << PAGE_SHIFT); |
638 | dbg("end_paddr = %lx\n", end_pfn << PAGE_SHIFT); | |
1da177e4 | 639 | |
45fb6cea AB |
640 | bootmap_pages = bootmem_bootmap_pages(end_pfn - start_pfn); |
641 | bootmem_paddr = (unsigned long)careful_allocation(nid, | |
642 | bootmap_pages << PAGE_SHIFT, | |
643 | PAGE_SIZE, end_pfn); | |
644 | memset(__va(bootmem_paddr), 0, bootmap_pages << PAGE_SHIFT); | |
1da177e4 | 645 | |
1da177e4 LT |
646 | dbg("bootmap_paddr = %lx\n", bootmem_paddr); |
647 | ||
648 | init_bootmem_node(NODE_DATA(nid), bootmem_paddr >> PAGE_SHIFT, | |
45fb6cea | 649 | start_pfn, end_pfn); |
1da177e4 | 650 | |
c67c3cb4 | 651 | free_bootmem_with_active_regions(nid, end_pfn); |
1da177e4 | 652 | |
45fb6cea | 653 | /* Mark reserved regions on this node */ |
1da177e4 | 654 | for (i = 0; i < lmb.reserved.cnt; i++) { |
180379dc | 655 | unsigned long physbase = lmb.reserved.region[i].base; |
1da177e4 | 656 | unsigned long size = lmb.reserved.region[i].size; |
45fb6cea AB |
657 | unsigned long start_paddr = start_pfn << PAGE_SHIFT; |
658 | unsigned long end_paddr = end_pfn << PAGE_SHIFT; | |
1da177e4 | 659 | |
45fb6cea AB |
660 | if (early_pfn_to_nid(physbase >> PAGE_SHIFT) != nid && |
661 | early_pfn_to_nid((physbase+size-1) >> PAGE_SHIFT) != nid) | |
1da177e4 LT |
662 | continue; |
663 | ||
664 | if (physbase < end_paddr && | |
665 | (physbase+size) > start_paddr) { | |
666 | /* overlaps */ | |
667 | if (physbase < start_paddr) { | |
668 | size -= start_paddr - physbase; | |
669 | physbase = start_paddr; | |
670 | } | |
671 | ||
672 | if (size > end_paddr - physbase) | |
673 | size = end_paddr - physbase; | |
674 | ||
675 | dbg("reserve_bootmem %lx %lx\n", physbase, | |
676 | size); | |
677 | reserve_bootmem_node(NODE_DATA(nid), physbase, | |
678 | size); | |
679 | } | |
680 | } | |
802f192e | 681 | |
c67c3cb4 | 682 | sparse_memory_present_with_active_regions(nid); |
1da177e4 LT |
683 | } |
684 | } | |
685 | ||
686 | void __init paging_init(void) | |
687 | { | |
6391af17 MG |
688 | unsigned long max_zone_pfns[MAX_NR_ZONES]; |
689 | memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); | |
690 | max_zone_pfns[ZONE_DMA] = lmb_end_of_DRAM() >> PAGE_SHIFT; | |
c67c3cb4 | 691 | free_area_init_nodes(max_zone_pfns); |
1da177e4 LT |
692 | } |
693 | ||
694 | static int __init early_numa(char *p) | |
695 | { | |
696 | if (!p) | |
697 | return 0; | |
698 | ||
699 | if (strstr(p, "off")) | |
700 | numa_enabled = 0; | |
701 | ||
702 | if (strstr(p, "debug")) | |
703 | numa_debug = 1; | |
704 | ||
705 | return 0; | |
706 | } | |
707 | early_param("numa", early_numa); | |
237a0989 MK |
708 | |
709 | #ifdef CONFIG_MEMORY_HOTPLUG | |
710 | /* | |
711 | * Find the node associated with a hot added memory section. Section | |
712 | * corresponds to a SPARSEMEM section, not an LMB. It is assumed that | |
713 | * sections are fully contained within a single LMB. | |
714 | */ | |
715 | int hot_add_scn_to_nid(unsigned long scn_addr) | |
716 | { | |
717 | struct device_node *memory = NULL; | |
b226e462 | 718 | nodemask_t nodes; |
482ec7c4 | 719 | int default_nid = any_online_node(NODE_MASK_ALL); |
069007ae | 720 | int nid; |
237a0989 MK |
721 | |
722 | if (!numa_enabled || (min_common_depth < 0)) | |
482ec7c4 | 723 | return default_nid; |
237a0989 MK |
724 | |
725 | while ((memory = of_find_node_by_type(memory, "memory")) != NULL) { | |
726 | unsigned long start, size; | |
b226e462 | 727 | int ranges; |
a7f67bdf | 728 | const unsigned int *memcell_buf; |
237a0989 MK |
729 | unsigned int len; |
730 | ||
e2eb6392 | 731 | memcell_buf = of_get_property(memory, "reg", &len); |
237a0989 MK |
732 | if (!memcell_buf || len <= 0) |
733 | continue; | |
734 | ||
cc5d0189 BH |
735 | /* ranges in cell */ |
736 | ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells); | |
237a0989 MK |
737 | ha_new_range: |
738 | start = read_n_cells(n_mem_addr_cells, &memcell_buf); | |
739 | size = read_n_cells(n_mem_size_cells, &memcell_buf); | |
953039c8 | 740 | nid = of_node_to_nid_single(memory); |
237a0989 MK |
741 | |
742 | /* Domains not present at boot default to 0 */ | |
482ec7c4 NL |
743 | if (nid < 0 || !node_online(nid)) |
744 | nid = default_nid; | |
237a0989 MK |
745 | |
746 | if ((scn_addr >= start) && (scn_addr < (start + size))) { | |
747 | of_node_put(memory); | |
cf950b7a | 748 | goto got_nid; |
237a0989 MK |
749 | } |
750 | ||
751 | if (--ranges) /* process all ranges in cell */ | |
752 | goto ha_new_range; | |
753 | } | |
237a0989 | 754 | BUG(); /* section address should be found above */ |
069007ae | 755 | return 0; |
b226e462 MK |
756 | |
757 | /* Temporary code to ensure that returned node is not empty */ | |
cf950b7a | 758 | got_nid: |
b226e462 | 759 | nodes_setall(nodes); |
cf950b7a NL |
760 | while (NODE_DATA(nid)->node_spanned_pages == 0) { |
761 | node_clear(nid, nodes); | |
762 | nid = any_online_node(nodes); | |
b226e462 | 763 | } |
cf950b7a | 764 | return nid; |
237a0989 MK |
765 | } |
766 | #endif /* CONFIG_MEMORY_HOTPLUG */ |