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> |
6df1646e | 21 | #include <linux/of.h> |
06eccea6 | 22 | #include <linux/pfn.h> |
45fb6cea | 23 | #include <asm/sparsemem.h> |
d9b2b2a2 | 24 | #include <asm/prom.h> |
cf00a8d1 | 25 | #include <asm/system.h> |
2249ca9d | 26 | #include <asm/smp.h> |
1da177e4 LT |
27 | |
28 | static int numa_enabled = 1; | |
29 | ||
1daa6d08 BS |
30 | static char *cmdline __initdata; |
31 | ||
1da177e4 LT |
32 | static int numa_debug; |
33 | #define dbg(args...) if (numa_debug) { printk(KERN_INFO args); } | |
34 | ||
45fb6cea | 35 | int numa_cpu_lookup_table[NR_CPUS]; |
1da177e4 | 36 | cpumask_t numa_cpumask_lookup_table[MAX_NUMNODES]; |
1da177e4 | 37 | struct pglist_data *node_data[MAX_NUMNODES]; |
45fb6cea AB |
38 | |
39 | EXPORT_SYMBOL(numa_cpu_lookup_table); | |
40 | EXPORT_SYMBOL(numa_cpumask_lookup_table); | |
41 | EXPORT_SYMBOL(node_data); | |
42 | ||
1da177e4 | 43 | static int min_common_depth; |
237a0989 | 44 | static int n_mem_addr_cells, n_mem_size_cells; |
1da177e4 | 45 | |
1daa6d08 BS |
46 | static int __cpuinit fake_numa_create_new_node(unsigned long end_pfn, |
47 | unsigned int *nid) | |
48 | { | |
49 | unsigned long long mem; | |
50 | char *p = cmdline; | |
51 | static unsigned int fake_nid; | |
52 | static unsigned long long curr_boundary; | |
53 | ||
54 | /* | |
55 | * Modify node id, iff we started creating NUMA nodes | |
56 | * We want to continue from where we left of the last time | |
57 | */ | |
58 | if (fake_nid) | |
59 | *nid = fake_nid; | |
60 | /* | |
61 | * In case there are no more arguments to parse, the | |
62 | * node_id should be the same as the last fake node id | |
63 | * (we've handled this above). | |
64 | */ | |
65 | if (!p) | |
66 | return 0; | |
67 | ||
68 | mem = memparse(p, &p); | |
69 | if (!mem) | |
70 | return 0; | |
71 | ||
72 | if (mem < curr_boundary) | |
73 | return 0; | |
74 | ||
75 | curr_boundary = mem; | |
76 | ||
77 | if ((end_pfn << PAGE_SHIFT) > mem) { | |
78 | /* | |
79 | * Skip commas and spaces | |
80 | */ | |
81 | while (*p == ',' || *p == ' ' || *p == '\t') | |
82 | p++; | |
83 | ||
84 | cmdline = p; | |
85 | fake_nid++; | |
86 | *nid = fake_nid; | |
87 | dbg("created new fake_node with id %d\n", fake_nid); | |
88 | return 1; | |
89 | } | |
90 | return 0; | |
91 | } | |
92 | ||
8f64e1f2 JT |
93 | /* |
94 | * get_active_region_work_fn - A helper function for get_node_active_region | |
95 | * Returns datax set to the start_pfn and end_pfn if they contain | |
96 | * the initial value of datax->start_pfn between them | |
97 | * @start_pfn: start page(inclusive) of region to check | |
98 | * @end_pfn: end page(exclusive) of region to check | |
99 | * @datax: comes in with ->start_pfn set to value to search for and | |
100 | * goes out with active range if it contains it | |
101 | * Returns 1 if search value is in range else 0 | |
102 | */ | |
103 | static int __init get_active_region_work_fn(unsigned long start_pfn, | |
104 | unsigned long end_pfn, void *datax) | |
105 | { | |
106 | struct node_active_region *data; | |
107 | data = (struct node_active_region *)datax; | |
108 | ||
109 | if (start_pfn <= data->start_pfn && end_pfn > data->start_pfn) { | |
110 | data->start_pfn = start_pfn; | |
111 | data->end_pfn = end_pfn; | |
112 | return 1; | |
113 | } | |
114 | return 0; | |
115 | ||
116 | } | |
117 | ||
118 | /* | |
119 | * get_node_active_region - Return active region containing start_pfn | |
e8170372 | 120 | * Active range returned is empty if none found. |
8f64e1f2 JT |
121 | * @start_pfn: The page to return the region for. |
122 | * @node_ar: Returned set to the active region containing start_pfn | |
123 | */ | |
124 | static void __init get_node_active_region(unsigned long start_pfn, | |
125 | struct node_active_region *node_ar) | |
126 | { | |
127 | int nid = early_pfn_to_nid(start_pfn); | |
128 | ||
129 | node_ar->nid = nid; | |
130 | node_ar->start_pfn = start_pfn; | |
e8170372 | 131 | node_ar->end_pfn = start_pfn; |
8f64e1f2 JT |
132 | work_with_active_regions(nid, get_active_region_work_fn, node_ar); |
133 | } | |
134 | ||
2e5ce39d | 135 | static void __cpuinit map_cpu_to_node(int cpu, int node) |
1da177e4 LT |
136 | { |
137 | numa_cpu_lookup_table[cpu] = node; | |
45fb6cea | 138 | |
bf4b85b0 NL |
139 | dbg("adding cpu %d to node %d\n", cpu, node); |
140 | ||
45fb6cea | 141 | if (!(cpu_isset(cpu, numa_cpumask_lookup_table[node]))) |
1da177e4 | 142 | cpu_set(cpu, numa_cpumask_lookup_table[node]); |
1da177e4 LT |
143 | } |
144 | ||
145 | #ifdef CONFIG_HOTPLUG_CPU | |
146 | static void unmap_cpu_from_node(unsigned long cpu) | |
147 | { | |
148 | int node = numa_cpu_lookup_table[cpu]; | |
149 | ||
150 | dbg("removing cpu %lu from node %d\n", cpu, node); | |
151 | ||
152 | if (cpu_isset(cpu, numa_cpumask_lookup_table[node])) { | |
153 | cpu_clear(cpu, numa_cpumask_lookup_table[node]); | |
1da177e4 LT |
154 | } else { |
155 | printk(KERN_ERR "WARNING: cpu %lu not found in node %d\n", | |
156 | cpu, node); | |
157 | } | |
158 | } | |
159 | #endif /* CONFIG_HOTPLUG_CPU */ | |
160 | ||
1da177e4 | 161 | /* must hold reference to node during call */ |
a7f67bdf | 162 | static const int *of_get_associativity(struct device_node *dev) |
1da177e4 | 163 | { |
e2eb6392 | 164 | return of_get_property(dev, "ibm,associativity", NULL); |
1da177e4 LT |
165 | } |
166 | ||
cf00085d C |
167 | /* |
168 | * Returns the property linux,drconf-usable-memory if | |
169 | * it exists (the property exists only in kexec/kdump kernels, | |
170 | * added by kexec-tools) | |
171 | */ | |
172 | static const u32 *of_get_usable_memory(struct device_node *memory) | |
173 | { | |
174 | const u32 *prop; | |
175 | u32 len; | |
176 | prop = of_get_property(memory, "linux,drconf-usable-memory", &len); | |
177 | if (!prop || len < sizeof(unsigned int)) | |
178 | return 0; | |
179 | return prop; | |
180 | } | |
181 | ||
482ec7c4 NL |
182 | /* Returns nid in the range [0..MAX_NUMNODES-1], or -1 if no useful numa |
183 | * info is found. | |
184 | */ | |
953039c8 | 185 | static int of_node_to_nid_single(struct device_node *device) |
1da177e4 | 186 | { |
482ec7c4 | 187 | int nid = -1; |
a7f67bdf | 188 | const unsigned int *tmp; |
1da177e4 LT |
189 | |
190 | if (min_common_depth == -1) | |
482ec7c4 | 191 | goto out; |
1da177e4 LT |
192 | |
193 | tmp = of_get_associativity(device); | |
482ec7c4 NL |
194 | if (!tmp) |
195 | goto out; | |
196 | ||
197 | if (tmp[0] >= min_common_depth) | |
cf950b7a | 198 | nid = tmp[min_common_depth]; |
bc16a759 NL |
199 | |
200 | /* POWER4 LPAR uses 0xffff as invalid node */ | |
482ec7c4 NL |
201 | if (nid == 0xffff || nid >= MAX_NUMNODES) |
202 | nid = -1; | |
203 | out: | |
cf950b7a | 204 | return nid; |
1da177e4 LT |
205 | } |
206 | ||
953039c8 JK |
207 | /* Walk the device tree upwards, looking for an associativity id */ |
208 | int of_node_to_nid(struct device_node *device) | |
209 | { | |
210 | struct device_node *tmp; | |
211 | int nid = -1; | |
212 | ||
213 | of_node_get(device); | |
214 | while (device) { | |
215 | nid = of_node_to_nid_single(device); | |
216 | if (nid != -1) | |
217 | break; | |
218 | ||
219 | tmp = device; | |
220 | device = of_get_parent(tmp); | |
221 | of_node_put(tmp); | |
222 | } | |
223 | of_node_put(device); | |
224 | ||
225 | return nid; | |
226 | } | |
227 | EXPORT_SYMBOL_GPL(of_node_to_nid); | |
228 | ||
1da177e4 LT |
229 | /* |
230 | * In theory, the "ibm,associativity" property may contain multiple | |
231 | * associativity lists because a resource may be multiply connected | |
232 | * into the machine. This resource then has different associativity | |
233 | * characteristics relative to its multiple connections. We ignore | |
234 | * this for now. We also assume that all cpu and memory sets have | |
235 | * their distances represented at a common level. This won't be | |
1b3c3714 | 236 | * true for hierarchical NUMA. |
1da177e4 LT |
237 | * |
238 | * In any case the ibm,associativity-reference-points should give | |
239 | * the correct depth for a normal NUMA system. | |
240 | * | |
241 | * - Dave Hansen <haveblue@us.ibm.com> | |
242 | */ | |
243 | static int __init find_min_common_depth(void) | |
244 | { | |
245 | int depth; | |
a7f67bdf | 246 | const unsigned int *ref_points; |
1da177e4 LT |
247 | struct device_node *rtas_root; |
248 | unsigned int len; | |
249 | ||
250 | rtas_root = of_find_node_by_path("/rtas"); | |
251 | ||
252 | if (!rtas_root) | |
253 | return -1; | |
254 | ||
255 | /* | |
256 | * this property is 2 32-bit integers, each representing a level of | |
257 | * depth in the associativity nodes. The first is for an SMP | |
258 | * configuration (should be all 0's) and the second is for a normal | |
259 | * NUMA configuration. | |
260 | */ | |
e2eb6392 | 261 | ref_points = of_get_property(rtas_root, |
1da177e4 LT |
262 | "ibm,associativity-reference-points", &len); |
263 | ||
20fcefe5 | 264 | if ((len >= 2 * sizeof(unsigned int)) && ref_points) { |
1da177e4 LT |
265 | depth = ref_points[1]; |
266 | } else { | |
bf4b85b0 | 267 | dbg("NUMA: ibm,associativity-reference-points not found.\n"); |
1da177e4 LT |
268 | depth = -1; |
269 | } | |
270 | of_node_put(rtas_root); | |
271 | ||
272 | return depth; | |
273 | } | |
274 | ||
84c9fdd1 | 275 | static void __init get_n_mem_cells(int *n_addr_cells, int *n_size_cells) |
1da177e4 LT |
276 | { |
277 | struct device_node *memory = NULL; | |
1da177e4 LT |
278 | |
279 | memory = of_find_node_by_type(memory, "memory"); | |
54c23310 | 280 | if (!memory) |
84c9fdd1 | 281 | panic("numa.c: No memory nodes found!"); |
54c23310 | 282 | |
a8bda5dd | 283 | *n_addr_cells = of_n_addr_cells(memory); |
9213feea | 284 | *n_size_cells = of_n_size_cells(memory); |
84c9fdd1 | 285 | of_node_put(memory); |
1da177e4 LT |
286 | } |
287 | ||
a7f67bdf | 288 | static unsigned long __devinit read_n_cells(int n, const unsigned int **buf) |
1da177e4 LT |
289 | { |
290 | unsigned long result = 0; | |
291 | ||
292 | while (n--) { | |
293 | result = (result << 32) | **buf; | |
294 | (*buf)++; | |
295 | } | |
296 | return result; | |
297 | } | |
298 | ||
8342681d NF |
299 | struct of_drconf_cell { |
300 | u64 base_addr; | |
301 | u32 drc_index; | |
302 | u32 reserved; | |
303 | u32 aa_index; | |
304 | u32 flags; | |
305 | }; | |
306 | ||
307 | #define DRCONF_MEM_ASSIGNED 0x00000008 | |
308 | #define DRCONF_MEM_AI_INVALID 0x00000040 | |
309 | #define DRCONF_MEM_RESERVED 0x00000080 | |
310 | ||
311 | /* | |
312 | * Read the next lmb list entry from the ibm,dynamic-memory property | |
313 | * and return the information in the provided of_drconf_cell structure. | |
314 | */ | |
315 | static void read_drconf_cell(struct of_drconf_cell *drmem, const u32 **cellp) | |
316 | { | |
317 | const u32 *cp; | |
318 | ||
319 | drmem->base_addr = read_n_cells(n_mem_addr_cells, cellp); | |
320 | ||
321 | cp = *cellp; | |
322 | drmem->drc_index = cp[0]; | |
323 | drmem->reserved = cp[1]; | |
324 | drmem->aa_index = cp[2]; | |
325 | drmem->flags = cp[3]; | |
326 | ||
327 | *cellp = cp + 4; | |
328 | } | |
329 | ||
330 | /* | |
331 | * Retreive and validate the ibm,dynamic-memory property of the device tree. | |
332 | * | |
333 | * The layout of the ibm,dynamic-memory property is a number N of lmb | |
334 | * list entries followed by N lmb list entries. Each lmb list entry | |
335 | * contains information as layed out in the of_drconf_cell struct above. | |
336 | */ | |
337 | static int of_get_drconf_memory(struct device_node *memory, const u32 **dm) | |
338 | { | |
339 | const u32 *prop; | |
340 | u32 len, entries; | |
341 | ||
342 | prop = of_get_property(memory, "ibm,dynamic-memory", &len); | |
343 | if (!prop || len < sizeof(unsigned int)) | |
344 | return 0; | |
345 | ||
346 | entries = *prop++; | |
347 | ||
348 | /* Now that we know the number of entries, revalidate the size | |
349 | * of the property read in to ensure we have everything | |
350 | */ | |
351 | if (len < (entries * (n_mem_addr_cells + 4) + 1) * sizeof(unsigned int)) | |
352 | return 0; | |
353 | ||
354 | *dm = prop; | |
355 | return entries; | |
356 | } | |
357 | ||
358 | /* | |
359 | * Retreive and validate the ibm,lmb-size property for drconf memory | |
360 | * from the device tree. | |
361 | */ | |
362 | static u64 of_get_lmb_size(struct device_node *memory) | |
363 | { | |
364 | const u32 *prop; | |
365 | u32 len; | |
366 | ||
367 | prop = of_get_property(memory, "ibm,lmb-size", &len); | |
368 | if (!prop || len < sizeof(unsigned int)) | |
369 | return 0; | |
370 | ||
371 | return read_n_cells(n_mem_size_cells, &prop); | |
372 | } | |
373 | ||
374 | struct assoc_arrays { | |
375 | u32 n_arrays; | |
376 | u32 array_sz; | |
377 | const u32 *arrays; | |
378 | }; | |
379 | ||
380 | /* | |
381 | * Retreive and validate the list of associativity arrays for drconf | |
382 | * memory from the ibm,associativity-lookup-arrays property of the | |
383 | * device tree.. | |
384 | * | |
385 | * The layout of the ibm,associativity-lookup-arrays property is a number N | |
386 | * indicating the number of associativity arrays, followed by a number M | |
387 | * indicating the size of each associativity array, followed by a list | |
388 | * of N associativity arrays. | |
389 | */ | |
390 | static int of_get_assoc_arrays(struct device_node *memory, | |
391 | struct assoc_arrays *aa) | |
392 | { | |
393 | const u32 *prop; | |
394 | u32 len; | |
395 | ||
396 | prop = of_get_property(memory, "ibm,associativity-lookup-arrays", &len); | |
397 | if (!prop || len < 2 * sizeof(unsigned int)) | |
398 | return -1; | |
399 | ||
400 | aa->n_arrays = *prop++; | |
401 | aa->array_sz = *prop++; | |
402 | ||
403 | /* Now that we know the number of arrrays and size of each array, | |
404 | * revalidate the size of the property read in. | |
405 | */ | |
406 | if (len < (aa->n_arrays * aa->array_sz + 2) * sizeof(unsigned int)) | |
407 | return -1; | |
408 | ||
409 | aa->arrays = prop; | |
410 | return 0; | |
411 | } | |
412 | ||
413 | /* | |
414 | * This is like of_node_to_nid_single() for memory represented in the | |
415 | * ibm,dynamic-reconfiguration-memory node. | |
416 | */ | |
417 | static int of_drconf_to_nid_single(struct of_drconf_cell *drmem, | |
418 | struct assoc_arrays *aa) | |
419 | { | |
420 | int default_nid = 0; | |
421 | int nid = default_nid; | |
422 | int index; | |
423 | ||
424 | if (min_common_depth > 0 && min_common_depth <= aa->array_sz && | |
425 | !(drmem->flags & DRCONF_MEM_AI_INVALID) && | |
426 | drmem->aa_index < aa->n_arrays) { | |
427 | index = drmem->aa_index * aa->array_sz + min_common_depth - 1; | |
428 | nid = aa->arrays[index]; | |
429 | ||
430 | if (nid == 0xffff || nid >= MAX_NUMNODES) | |
431 | nid = default_nid; | |
432 | } | |
433 | ||
434 | return nid; | |
435 | } | |
436 | ||
1da177e4 LT |
437 | /* |
438 | * Figure out to which domain a cpu belongs and stick it there. | |
439 | * Return the id of the domain used. | |
440 | */ | |
2e5ce39d | 441 | static int __cpuinit numa_setup_cpu(unsigned long lcpu) |
1da177e4 | 442 | { |
cf950b7a | 443 | int nid = 0; |
8b16cd23 | 444 | struct device_node *cpu = of_get_cpu_node(lcpu, NULL); |
1da177e4 LT |
445 | |
446 | if (!cpu) { | |
447 | WARN_ON(1); | |
448 | goto out; | |
449 | } | |
450 | ||
953039c8 | 451 | nid = of_node_to_nid_single(cpu); |
1da177e4 | 452 | |
482ec7c4 NL |
453 | if (nid < 0 || !node_online(nid)) |
454 | nid = any_online_node(NODE_MASK_ALL); | |
1da177e4 | 455 | out: |
cf950b7a | 456 | map_cpu_to_node(lcpu, nid); |
1da177e4 LT |
457 | |
458 | of_node_put(cpu); | |
459 | ||
cf950b7a | 460 | return nid; |
1da177e4 LT |
461 | } |
462 | ||
74b85f37 | 463 | static int __cpuinit cpu_numa_callback(struct notifier_block *nfb, |
1da177e4 LT |
464 | unsigned long action, |
465 | void *hcpu) | |
466 | { | |
467 | unsigned long lcpu = (unsigned long)hcpu; | |
468 | int ret = NOTIFY_DONE; | |
469 | ||
470 | switch (action) { | |
471 | case CPU_UP_PREPARE: | |
8bb78442 | 472 | case CPU_UP_PREPARE_FROZEN: |
2b261227 | 473 | numa_setup_cpu(lcpu); |
1da177e4 LT |
474 | ret = NOTIFY_OK; |
475 | break; | |
476 | #ifdef CONFIG_HOTPLUG_CPU | |
477 | case CPU_DEAD: | |
8bb78442 | 478 | case CPU_DEAD_FROZEN: |
1da177e4 | 479 | case CPU_UP_CANCELED: |
8bb78442 | 480 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 LT |
481 | unmap_cpu_from_node(lcpu); |
482 | break; | |
483 | ret = NOTIFY_OK; | |
484 | #endif | |
485 | } | |
486 | return ret; | |
487 | } | |
488 | ||
489 | /* | |
490 | * Check and possibly modify a memory region to enforce the memory limit. | |
491 | * | |
492 | * Returns the size the region should have to enforce the memory limit. | |
493 | * This will either be the original value of size, a truncated value, | |
494 | * or zero. If the returned value of size is 0 the region should be | |
495 | * discarded as it lies wholy above the memory limit. | |
496 | */ | |
45fb6cea AB |
497 | static unsigned long __init numa_enforce_memory_limit(unsigned long start, |
498 | unsigned long size) | |
1da177e4 LT |
499 | { |
500 | /* | |
501 | * We use lmb_end_of_DRAM() in here instead of memory_limit because | |
502 | * we've already adjusted it for the limit and it takes care of | |
fe55249d MM |
503 | * having memory holes below the limit. Also, in the case of |
504 | * iommu_is_off, memory_limit is not set but is implicitly enforced. | |
1da177e4 | 505 | */ |
1da177e4 | 506 | |
1da177e4 LT |
507 | if (start + size <= lmb_end_of_DRAM()) |
508 | return size; | |
509 | ||
510 | if (start >= lmb_end_of_DRAM()) | |
511 | return 0; | |
512 | ||
513 | return lmb_end_of_DRAM() - start; | |
514 | } | |
515 | ||
cf00085d C |
516 | /* |
517 | * Reads the counter for a given entry in | |
518 | * linux,drconf-usable-memory property | |
519 | */ | |
520 | static inline int __init read_usm_ranges(const u32 **usm) | |
521 | { | |
522 | /* | |
523 | * For each lmb in ibm,dynamic-memory a corresponding | |
524 | * entry in linux,drconf-usable-memory property contains | |
525 | * a counter followed by that many (base, size) duple. | |
526 | * read the counter from linux,drconf-usable-memory | |
527 | */ | |
528 | return read_n_cells(n_mem_size_cells, usm); | |
529 | } | |
530 | ||
0204568a PM |
531 | /* |
532 | * Extract NUMA information from the ibm,dynamic-reconfiguration-memory | |
533 | * node. This assumes n_mem_{addr,size}_cells have been set. | |
534 | */ | |
535 | static void __init parse_drconf_memory(struct device_node *memory) | |
536 | { | |
cf00085d C |
537 | const u32 *dm, *usm; |
538 | unsigned int n, rc, ranges, is_kexec_kdump = 0; | |
539 | unsigned long lmb_size, base, size, sz; | |
8342681d NF |
540 | int nid; |
541 | struct assoc_arrays aa; | |
542 | ||
543 | n = of_get_drconf_memory(memory, &dm); | |
544 | if (!n) | |
0204568a PM |
545 | return; |
546 | ||
8342681d NF |
547 | lmb_size = of_get_lmb_size(memory); |
548 | if (!lmb_size) | |
549 | return; | |
550 | ||
551 | rc = of_get_assoc_arrays(memory, &aa); | |
552 | if (rc) | |
0204568a PM |
553 | return; |
554 | ||
cf00085d C |
555 | /* check if this is a kexec/kdump kernel */ |
556 | usm = of_get_usable_memory(memory); | |
557 | if (usm != NULL) | |
558 | is_kexec_kdump = 1; | |
559 | ||
0204568a | 560 | for (; n != 0; --n) { |
8342681d NF |
561 | struct of_drconf_cell drmem; |
562 | ||
563 | read_drconf_cell(&drmem, &dm); | |
564 | ||
565 | /* skip this block if the reserved bit is set in flags (0x80) | |
566 | or if the block is not assigned to this partition (0x8) */ | |
567 | if ((drmem.flags & DRCONF_MEM_RESERVED) | |
568 | || !(drmem.flags & DRCONF_MEM_ASSIGNED)) | |
0204568a | 569 | continue; |
1daa6d08 | 570 | |
cf00085d C |
571 | base = drmem.base_addr; |
572 | size = lmb_size; | |
573 | ranges = 1; | |
8342681d | 574 | |
cf00085d C |
575 | if (is_kexec_kdump) { |
576 | ranges = read_usm_ranges(&usm); | |
577 | if (!ranges) /* there are no (base, size) duple */ | |
578 | continue; | |
579 | } | |
580 | do { | |
581 | if (is_kexec_kdump) { | |
582 | base = read_n_cells(n_mem_addr_cells, &usm); | |
583 | size = read_n_cells(n_mem_size_cells, &usm); | |
584 | } | |
585 | nid = of_drconf_to_nid_single(&drmem, &aa); | |
586 | fake_numa_create_new_node( | |
587 | ((base + size) >> PAGE_SHIFT), | |
8342681d | 588 | &nid); |
cf00085d C |
589 | node_set_online(nid); |
590 | sz = numa_enforce_memory_limit(base, size); | |
591 | if (sz) | |
592 | add_active_range(nid, base >> PAGE_SHIFT, | |
593 | (base >> PAGE_SHIFT) | |
594 | + (sz >> PAGE_SHIFT)); | |
595 | } while (--ranges); | |
0204568a PM |
596 | } |
597 | } | |
598 | ||
1da177e4 LT |
599 | static int __init parse_numa_properties(void) |
600 | { | |
601 | struct device_node *cpu = NULL; | |
602 | struct device_node *memory = NULL; | |
482ec7c4 | 603 | int default_nid = 0; |
1da177e4 LT |
604 | unsigned long i; |
605 | ||
606 | if (numa_enabled == 0) { | |
607 | printk(KERN_WARNING "NUMA disabled by user\n"); | |
608 | return -1; | |
609 | } | |
610 | ||
1da177e4 LT |
611 | min_common_depth = find_min_common_depth(); |
612 | ||
1da177e4 LT |
613 | if (min_common_depth < 0) |
614 | return min_common_depth; | |
615 | ||
bf4b85b0 NL |
616 | dbg("NUMA associativity depth for CPU/Memory: %d\n", min_common_depth); |
617 | ||
1da177e4 | 618 | /* |
482ec7c4 NL |
619 | * Even though we connect cpus to numa domains later in SMP |
620 | * init, we need to know the node ids now. This is because | |
621 | * each node to be onlined must have NODE_DATA etc backing it. | |
1da177e4 | 622 | */ |
482ec7c4 | 623 | for_each_present_cpu(i) { |
cf950b7a | 624 | int nid; |
1da177e4 | 625 | |
8b16cd23 | 626 | cpu = of_get_cpu_node(i, NULL); |
482ec7c4 | 627 | BUG_ON(!cpu); |
953039c8 | 628 | nid = of_node_to_nid_single(cpu); |
482ec7c4 | 629 | of_node_put(cpu); |
1da177e4 | 630 | |
482ec7c4 NL |
631 | /* |
632 | * Don't fall back to default_nid yet -- we will plug | |
633 | * cpus into nodes once the memory scan has discovered | |
634 | * the topology. | |
635 | */ | |
636 | if (nid < 0) | |
637 | continue; | |
638 | node_set_online(nid); | |
1da177e4 LT |
639 | } |
640 | ||
237a0989 | 641 | get_n_mem_cells(&n_mem_addr_cells, &n_mem_size_cells); |
1da177e4 LT |
642 | memory = NULL; |
643 | while ((memory = of_find_node_by_type(memory, "memory")) != NULL) { | |
644 | unsigned long start; | |
645 | unsigned long size; | |
cf950b7a | 646 | int nid; |
1da177e4 | 647 | int ranges; |
a7f67bdf | 648 | const unsigned int *memcell_buf; |
1da177e4 LT |
649 | unsigned int len; |
650 | ||
e2eb6392 | 651 | memcell_buf = of_get_property(memory, |
ba759485 ME |
652 | "linux,usable-memory", &len); |
653 | if (!memcell_buf || len <= 0) | |
e2eb6392 | 654 | memcell_buf = of_get_property(memory, "reg", &len); |
1da177e4 LT |
655 | if (!memcell_buf || len <= 0) |
656 | continue; | |
657 | ||
cc5d0189 BH |
658 | /* ranges in cell */ |
659 | ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells); | |
1da177e4 LT |
660 | new_range: |
661 | /* these are order-sensitive, and modify the buffer pointer */ | |
237a0989 MK |
662 | start = read_n_cells(n_mem_addr_cells, &memcell_buf); |
663 | size = read_n_cells(n_mem_size_cells, &memcell_buf); | |
1da177e4 | 664 | |
482ec7c4 NL |
665 | /* |
666 | * Assumption: either all memory nodes or none will | |
667 | * have associativity properties. If none, then | |
668 | * everything goes to default_nid. | |
669 | */ | |
953039c8 | 670 | nid = of_node_to_nid_single(memory); |
482ec7c4 NL |
671 | if (nid < 0) |
672 | nid = default_nid; | |
1daa6d08 BS |
673 | |
674 | fake_numa_create_new_node(((start + size) >> PAGE_SHIFT), &nid); | |
482ec7c4 | 675 | node_set_online(nid); |
1da177e4 | 676 | |
45fb6cea | 677 | if (!(size = numa_enforce_memory_limit(start, size))) { |
1da177e4 LT |
678 | if (--ranges) |
679 | goto new_range; | |
680 | else | |
681 | continue; | |
682 | } | |
683 | ||
c67c3cb4 MG |
684 | add_active_range(nid, start >> PAGE_SHIFT, |
685 | (start >> PAGE_SHIFT) + (size >> PAGE_SHIFT)); | |
1da177e4 LT |
686 | |
687 | if (--ranges) | |
688 | goto new_range; | |
689 | } | |
690 | ||
0204568a PM |
691 | /* |
692 | * Now do the same thing for each LMB listed in the ibm,dynamic-memory | |
693 | * property in the ibm,dynamic-reconfiguration-memory node. | |
694 | */ | |
695 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); | |
696 | if (memory) | |
697 | parse_drconf_memory(memory); | |
698 | ||
1da177e4 LT |
699 | return 0; |
700 | } | |
701 | ||
702 | static void __init setup_nonnuma(void) | |
703 | { | |
704 | unsigned long top_of_ram = lmb_end_of_DRAM(); | |
705 | unsigned long total_ram = lmb_phys_mem_size(); | |
c67c3cb4 | 706 | unsigned long start_pfn, end_pfn; |
1daa6d08 | 707 | unsigned int i, nid = 0; |
1da177e4 | 708 | |
e110b281 | 709 | printk(KERN_DEBUG "Top of RAM: 0x%lx, Total RAM: 0x%lx\n", |
1da177e4 | 710 | top_of_ram, total_ram); |
e110b281 | 711 | printk(KERN_DEBUG "Memory hole size: %ldMB\n", |
1da177e4 LT |
712 | (top_of_ram - total_ram) >> 20); |
713 | ||
c67c3cb4 MG |
714 | for (i = 0; i < lmb.memory.cnt; ++i) { |
715 | start_pfn = lmb.memory.region[i].base >> PAGE_SHIFT; | |
716 | end_pfn = start_pfn + lmb_size_pages(&lmb.memory, i); | |
1daa6d08 BS |
717 | |
718 | fake_numa_create_new_node(end_pfn, &nid); | |
719 | add_active_range(nid, start_pfn, end_pfn); | |
720 | node_set_online(nid); | |
c67c3cb4 | 721 | } |
1da177e4 LT |
722 | } |
723 | ||
4b703a23 AB |
724 | void __init dump_numa_cpu_topology(void) |
725 | { | |
726 | unsigned int node; | |
727 | unsigned int cpu, count; | |
728 | ||
729 | if (min_common_depth == -1 || !numa_enabled) | |
730 | return; | |
731 | ||
732 | for_each_online_node(node) { | |
e110b281 | 733 | printk(KERN_DEBUG "Node %d CPUs:", node); |
4b703a23 AB |
734 | |
735 | count = 0; | |
736 | /* | |
737 | * If we used a CPU iterator here we would miss printing | |
738 | * the holes in the cpumap. | |
739 | */ | |
740 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
741 | if (cpu_isset(cpu, numa_cpumask_lookup_table[node])) { | |
742 | if (count == 0) | |
743 | printk(" %u", cpu); | |
744 | ++count; | |
745 | } else { | |
746 | if (count > 1) | |
747 | printk("-%u", cpu - 1); | |
748 | count = 0; | |
749 | } | |
750 | } | |
751 | ||
752 | if (count > 1) | |
753 | printk("-%u", NR_CPUS - 1); | |
754 | printk("\n"); | |
755 | } | |
756 | } | |
757 | ||
758 | static void __init dump_numa_memory_topology(void) | |
1da177e4 LT |
759 | { |
760 | unsigned int node; | |
761 | unsigned int count; | |
762 | ||
763 | if (min_common_depth == -1 || !numa_enabled) | |
764 | return; | |
765 | ||
766 | for_each_online_node(node) { | |
767 | unsigned long i; | |
768 | ||
e110b281 | 769 | printk(KERN_DEBUG "Node %d Memory:", node); |
1da177e4 LT |
770 | |
771 | count = 0; | |
772 | ||
45fb6cea AB |
773 | for (i = 0; i < lmb_end_of_DRAM(); |
774 | i += (1 << SECTION_SIZE_BITS)) { | |
775 | if (early_pfn_to_nid(i >> PAGE_SHIFT) == node) { | |
1da177e4 LT |
776 | if (count == 0) |
777 | printk(" 0x%lx", i); | |
778 | ++count; | |
779 | } else { | |
780 | if (count > 0) | |
781 | printk("-0x%lx", i); | |
782 | count = 0; | |
783 | } | |
784 | } | |
785 | ||
786 | if (count > 0) | |
787 | printk("-0x%lx", i); | |
788 | printk("\n"); | |
789 | } | |
1da177e4 LT |
790 | } |
791 | ||
792 | /* | |
793 | * Allocate some memory, satisfying the lmb or bootmem allocator where | |
794 | * required. nid is the preferred node and end is the physical address of | |
795 | * the highest address in the node. | |
796 | * | |
0be210fd | 797 | * Returns the virtual address of the memory. |
1da177e4 | 798 | */ |
893473df | 799 | static void __init *careful_zallocation(int nid, unsigned long size, |
45fb6cea AB |
800 | unsigned long align, |
801 | unsigned long end_pfn) | |
1da177e4 | 802 | { |
0be210fd | 803 | void *ret; |
45fb6cea | 804 | int new_nid; |
0be210fd DH |
805 | unsigned long ret_paddr; |
806 | ||
807 | ret_paddr = __lmb_alloc_base(size, align, end_pfn << PAGE_SHIFT); | |
1da177e4 LT |
808 | |
809 | /* retry over all memory */ | |
0be210fd DH |
810 | if (!ret_paddr) |
811 | ret_paddr = __lmb_alloc_base(size, align, lmb_end_of_DRAM()); | |
1da177e4 | 812 | |
0be210fd | 813 | if (!ret_paddr) |
5d21ea2b | 814 | panic("numa.c: cannot allocate %lu bytes for node %d", |
1da177e4 LT |
815 | size, nid); |
816 | ||
0be210fd DH |
817 | ret = __va(ret_paddr); |
818 | ||
1da177e4 | 819 | /* |
c555e520 DH |
820 | * We initialize the nodes in numeric order: 0, 1, 2... |
821 | * and hand over control from the LMB allocator to the | |
822 | * bootmem allocator. If this function is called for | |
823 | * node 5, then we know that all nodes <5 are using the | |
824 | * bootmem allocator instead of the LMB allocator. | |
825 | * | |
826 | * So, check the nid from which this allocation came | |
827 | * and double check to see if we need to use bootmem | |
828 | * instead of the LMB. We don't free the LMB memory | |
829 | * since it would be useless. | |
1da177e4 | 830 | */ |
0be210fd | 831 | new_nid = early_pfn_to_nid(ret_paddr >> PAGE_SHIFT); |
45fb6cea | 832 | if (new_nid < nid) { |
0be210fd | 833 | ret = __alloc_bootmem_node(NODE_DATA(new_nid), |
1da177e4 LT |
834 | size, align, 0); |
835 | ||
0be210fd | 836 | dbg("alloc_bootmem %p %lx\n", ret, size); |
1da177e4 LT |
837 | } |
838 | ||
893473df | 839 | memset(ret, 0, size); |
0be210fd | 840 | return ret; |
1da177e4 LT |
841 | } |
842 | ||
74b85f37 CS |
843 | static struct notifier_block __cpuinitdata ppc64_numa_nb = { |
844 | .notifier_call = cpu_numa_callback, | |
845 | .priority = 1 /* Must run before sched domains notifier. */ | |
846 | }; | |
847 | ||
4a618669 DH |
848 | static void mark_reserved_regions_for_nid(int nid) |
849 | { | |
850 | struct pglist_data *node = NODE_DATA(nid); | |
851 | int i; | |
852 | ||
853 | for (i = 0; i < lmb.reserved.cnt; i++) { | |
854 | unsigned long physbase = lmb.reserved.region[i].base; | |
855 | unsigned long size = lmb.reserved.region[i].size; | |
856 | unsigned long start_pfn = physbase >> PAGE_SHIFT; | |
06eccea6 | 857 | unsigned long end_pfn = PFN_UP(physbase + size); |
4a618669 DH |
858 | struct node_active_region node_ar; |
859 | unsigned long node_end_pfn = node->node_start_pfn + | |
860 | node->node_spanned_pages; | |
861 | ||
862 | /* | |
863 | * Check to make sure that this lmb.reserved area is | |
864 | * within the bounds of the node that we care about. | |
865 | * Checking the nid of the start and end points is not | |
866 | * sufficient because the reserved area could span the | |
867 | * entire node. | |
868 | */ | |
869 | if (end_pfn <= node->node_start_pfn || | |
870 | start_pfn >= node_end_pfn) | |
871 | continue; | |
872 | ||
873 | get_node_active_region(start_pfn, &node_ar); | |
874 | while (start_pfn < end_pfn && | |
875 | node_ar.start_pfn < node_ar.end_pfn) { | |
876 | unsigned long reserve_size = size; | |
877 | /* | |
878 | * if reserved region extends past active region | |
879 | * then trim size to active region | |
880 | */ | |
881 | if (end_pfn > node_ar.end_pfn) | |
882 | reserve_size = (node_ar.end_pfn << PAGE_SHIFT) | |
06eccea6 | 883 | - physbase; |
a4c74ddd DH |
884 | /* |
885 | * Only worry about *this* node, others may not | |
886 | * yet have valid NODE_DATA(). | |
887 | */ | |
888 | if (node_ar.nid == nid) { | |
889 | dbg("reserve_bootmem %lx %lx nid=%d\n", | |
890 | physbase, reserve_size, node_ar.nid); | |
891 | reserve_bootmem_node(NODE_DATA(node_ar.nid), | |
892 | physbase, reserve_size, | |
893 | BOOTMEM_DEFAULT); | |
894 | } | |
4a618669 DH |
895 | /* |
896 | * if reserved region is contained in the active region | |
897 | * then done. | |
898 | */ | |
899 | if (end_pfn <= node_ar.end_pfn) | |
900 | break; | |
901 | ||
902 | /* | |
903 | * reserved region extends past the active region | |
904 | * get next active region that contains this | |
905 | * reserved region | |
906 | */ | |
907 | start_pfn = node_ar.end_pfn; | |
908 | physbase = start_pfn << PAGE_SHIFT; | |
909 | size = size - reserve_size; | |
910 | get_node_active_region(start_pfn, &node_ar); | |
911 | } | |
912 | } | |
913 | } | |
914 | ||
915 | ||
1da177e4 LT |
916 | void __init do_init_bootmem(void) |
917 | { | |
918 | int nid; | |
1da177e4 LT |
919 | |
920 | min_low_pfn = 0; | |
921 | max_low_pfn = lmb_end_of_DRAM() >> PAGE_SHIFT; | |
922 | max_pfn = max_low_pfn; | |
923 | ||
924 | if (parse_numa_properties()) | |
925 | setup_nonnuma(); | |
926 | else | |
4b703a23 | 927 | dump_numa_memory_topology(); |
1da177e4 LT |
928 | |
929 | register_cpu_notifier(&ppc64_numa_nb); | |
2b261227 NL |
930 | cpu_numa_callback(&ppc64_numa_nb, CPU_UP_PREPARE, |
931 | (void *)(unsigned long)boot_cpuid); | |
1da177e4 LT |
932 | |
933 | for_each_online_node(nid) { | |
c67c3cb4 | 934 | unsigned long start_pfn, end_pfn; |
0be210fd | 935 | void *bootmem_vaddr; |
1da177e4 LT |
936 | unsigned long bootmap_pages; |
937 | ||
c67c3cb4 | 938 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); |
1da177e4 | 939 | |
4a618669 DH |
940 | /* |
941 | * Allocate the node structure node local if possible | |
942 | * | |
943 | * Be careful moving this around, as it relies on all | |
944 | * previous nodes' bootmem to be initialized and have | |
945 | * all reserved areas marked. | |
946 | */ | |
893473df | 947 | NODE_DATA(nid) = careful_zallocation(nid, |
1da177e4 | 948 | sizeof(struct pglist_data), |
45fb6cea | 949 | SMP_CACHE_BYTES, end_pfn); |
1da177e4 LT |
950 | |
951 | dbg("node %d\n", nid); | |
952 | dbg("NODE_DATA() = %p\n", NODE_DATA(nid)); | |
953 | ||
b61bfa3c | 954 | NODE_DATA(nid)->bdata = &bootmem_node_data[nid]; |
45fb6cea AB |
955 | NODE_DATA(nid)->node_start_pfn = start_pfn; |
956 | NODE_DATA(nid)->node_spanned_pages = end_pfn - start_pfn; | |
1da177e4 LT |
957 | |
958 | if (NODE_DATA(nid)->node_spanned_pages == 0) | |
959 | continue; | |
960 | ||
45fb6cea AB |
961 | dbg("start_paddr = %lx\n", start_pfn << PAGE_SHIFT); |
962 | dbg("end_paddr = %lx\n", end_pfn << PAGE_SHIFT); | |
1da177e4 | 963 | |
45fb6cea | 964 | bootmap_pages = bootmem_bootmap_pages(end_pfn - start_pfn); |
893473df | 965 | bootmem_vaddr = careful_zallocation(nid, |
45fb6cea AB |
966 | bootmap_pages << PAGE_SHIFT, |
967 | PAGE_SIZE, end_pfn); | |
1da177e4 | 968 | |
0be210fd | 969 | dbg("bootmap_vaddr = %p\n", bootmem_vaddr); |
1da177e4 | 970 | |
0be210fd DH |
971 | init_bootmem_node(NODE_DATA(nid), |
972 | __pa(bootmem_vaddr) >> PAGE_SHIFT, | |
45fb6cea | 973 | start_pfn, end_pfn); |
1da177e4 | 974 | |
c67c3cb4 | 975 | free_bootmem_with_active_regions(nid, end_pfn); |
4a618669 DH |
976 | /* |
977 | * Be very careful about moving this around. Future | |
893473df | 978 | * calls to careful_zallocation() depend on this getting |
4a618669 DH |
979 | * done correctly. |
980 | */ | |
981 | mark_reserved_regions_for_nid(nid); | |
8f64e1f2 | 982 | sparse_memory_present_with_active_regions(nid); |
4a618669 | 983 | } |
d3f6204a BH |
984 | |
985 | init_bootmem_done = 1; | |
1da177e4 LT |
986 | } |
987 | ||
988 | void __init paging_init(void) | |
989 | { | |
6391af17 MG |
990 | unsigned long max_zone_pfns[MAX_NR_ZONES]; |
991 | memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); | |
992 | max_zone_pfns[ZONE_DMA] = lmb_end_of_DRAM() >> PAGE_SHIFT; | |
c67c3cb4 | 993 | free_area_init_nodes(max_zone_pfns); |
1da177e4 LT |
994 | } |
995 | ||
996 | static int __init early_numa(char *p) | |
997 | { | |
998 | if (!p) | |
999 | return 0; | |
1000 | ||
1001 | if (strstr(p, "off")) | |
1002 | numa_enabled = 0; | |
1003 | ||
1004 | if (strstr(p, "debug")) | |
1005 | numa_debug = 1; | |
1006 | ||
1daa6d08 BS |
1007 | p = strstr(p, "fake="); |
1008 | if (p) | |
1009 | cmdline = p + strlen("fake="); | |
1010 | ||
1da177e4 LT |
1011 | return 0; |
1012 | } | |
1013 | early_param("numa", early_numa); | |
237a0989 MK |
1014 | |
1015 | #ifdef CONFIG_MEMORY_HOTPLUG | |
0db9360a | 1016 | /* |
0f16ef7f NF |
1017 | * Find the node associated with a hot added memory section for |
1018 | * memory represented in the device tree by the property | |
1019 | * ibm,dynamic-reconfiguration-memory/ibm,dynamic-memory. | |
0db9360a NF |
1020 | */ |
1021 | static int hot_add_drconf_scn_to_nid(struct device_node *memory, | |
1022 | unsigned long scn_addr) | |
1023 | { | |
1024 | const u32 *dm; | |
0f16ef7f | 1025 | unsigned int drconf_cell_cnt, rc; |
0db9360a | 1026 | unsigned long lmb_size; |
0db9360a | 1027 | struct assoc_arrays aa; |
0f16ef7f | 1028 | int nid = -1; |
0db9360a | 1029 | |
0f16ef7f NF |
1030 | drconf_cell_cnt = of_get_drconf_memory(memory, &dm); |
1031 | if (!drconf_cell_cnt) | |
1032 | return -1; | |
0db9360a NF |
1033 | |
1034 | lmb_size = of_get_lmb_size(memory); | |
1035 | if (!lmb_size) | |
0f16ef7f | 1036 | return -1; |
0db9360a NF |
1037 | |
1038 | rc = of_get_assoc_arrays(memory, &aa); | |
1039 | if (rc) | |
0f16ef7f | 1040 | return -1; |
0db9360a | 1041 | |
0f16ef7f | 1042 | for (; drconf_cell_cnt != 0; --drconf_cell_cnt) { |
0db9360a NF |
1043 | struct of_drconf_cell drmem; |
1044 | ||
1045 | read_drconf_cell(&drmem, &dm); | |
1046 | ||
1047 | /* skip this block if it is reserved or not assigned to | |
1048 | * this partition */ | |
1049 | if ((drmem.flags & DRCONF_MEM_RESERVED) | |
1050 | || !(drmem.flags & DRCONF_MEM_ASSIGNED)) | |
1051 | continue; | |
1052 | ||
0f16ef7f NF |
1053 | if ((scn_addr < drmem.base_addr) |
1054 | || (scn_addr >= (drmem.base_addr + lmb_size))) | |
1055 | continue; | |
1056 | ||
0db9360a | 1057 | nid = of_drconf_to_nid_single(&drmem, &aa); |
0f16ef7f NF |
1058 | break; |
1059 | } | |
1060 | ||
1061 | return nid; | |
1062 | } | |
1063 | ||
1064 | /* | |
1065 | * Find the node associated with a hot added memory section for memory | |
1066 | * represented in the device tree as a node (i.e. memory@XXXX) for | |
1067 | * each lmb. | |
1068 | */ | |
1069 | int hot_add_node_scn_to_nid(unsigned long scn_addr) | |
1070 | { | |
1071 | struct device_node *memory = NULL; | |
1072 | int nid = -1; | |
1073 | ||
1074 | while ((memory = of_find_node_by_type(memory, "memory")) != NULL) { | |
1075 | unsigned long start, size; | |
1076 | int ranges; | |
1077 | const unsigned int *memcell_buf; | |
1078 | unsigned int len; | |
1079 | ||
1080 | memcell_buf = of_get_property(memory, "reg", &len); | |
1081 | if (!memcell_buf || len <= 0) | |
1082 | continue; | |
1083 | ||
1084 | /* ranges in cell */ | |
1085 | ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells); | |
1086 | ||
1087 | while (ranges--) { | |
1088 | start = read_n_cells(n_mem_addr_cells, &memcell_buf); | |
1089 | size = read_n_cells(n_mem_size_cells, &memcell_buf); | |
1090 | ||
1091 | if ((scn_addr < start) || (scn_addr >= (start + size))) | |
1092 | continue; | |
1093 | ||
1094 | nid = of_node_to_nid_single(memory); | |
1095 | break; | |
1096 | } | |
0db9360a | 1097 | |
0f16ef7f NF |
1098 | of_node_put(memory); |
1099 | if (nid >= 0) | |
1100 | break; | |
0db9360a NF |
1101 | } |
1102 | ||
0f16ef7f | 1103 | return nid; |
0db9360a NF |
1104 | } |
1105 | ||
237a0989 MK |
1106 | /* |
1107 | * Find the node associated with a hot added memory section. Section | |
1108 | * corresponds to a SPARSEMEM section, not an LMB. It is assumed that | |
1109 | * sections are fully contained within a single LMB. | |
1110 | */ | |
1111 | int hot_add_scn_to_nid(unsigned long scn_addr) | |
1112 | { | |
1113 | struct device_node *memory = NULL; | |
0f16ef7f | 1114 | int nid, found = 0; |
237a0989 MK |
1115 | |
1116 | if (!numa_enabled || (min_common_depth < 0)) | |
0db9360a NF |
1117 | return any_online_node(NODE_MASK_ALL); |
1118 | ||
1119 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); | |
1120 | if (memory) { | |
1121 | nid = hot_add_drconf_scn_to_nid(memory, scn_addr); | |
1122 | of_node_put(memory); | |
0f16ef7f NF |
1123 | } else { |
1124 | nid = hot_add_node_scn_to_nid(scn_addr); | |
0db9360a | 1125 | } |
237a0989 | 1126 | |
0f16ef7f NF |
1127 | if (nid < 0 || !node_online(nid)) |
1128 | nid = any_online_node(NODE_MASK_ALL); | |
237a0989 | 1129 | |
0f16ef7f NF |
1130 | if (NODE_DATA(nid)->node_spanned_pages) |
1131 | return nid; | |
237a0989 | 1132 | |
0f16ef7f NF |
1133 | for_each_online_node(nid) { |
1134 | if (NODE_DATA(nid)->node_spanned_pages) { | |
1135 | found = 1; | |
1136 | break; | |
237a0989 | 1137 | } |
237a0989 | 1138 | } |
0f16ef7f NF |
1139 | |
1140 | BUG_ON(!found); | |
1141 | return nid; | |
237a0989 | 1142 | } |
0f16ef7f | 1143 | |
237a0989 | 1144 | #endif /* CONFIG_MEMORY_HOTPLUG */ |