]>
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> |
45fb6cea | 22 | #include <asm/sparsemem.h> |
d9b2b2a2 | 23 | #include <asm/prom.h> |
cf00a8d1 | 24 | #include <asm/system.h> |
2249ca9d | 25 | #include <asm/smp.h> |
1da177e4 LT |
26 | |
27 | static int numa_enabled = 1; | |
28 | ||
1daa6d08 BS |
29 | static char *cmdline __initdata; |
30 | ||
1da177e4 LT |
31 | static int numa_debug; |
32 | #define dbg(args...) if (numa_debug) { printk(KERN_INFO args); } | |
33 | ||
45fb6cea | 34 | int numa_cpu_lookup_table[NR_CPUS]; |
1da177e4 | 35 | cpumask_t numa_cpumask_lookup_table[MAX_NUMNODES]; |
1da177e4 | 36 | struct pglist_data *node_data[MAX_NUMNODES]; |
45fb6cea AB |
37 | |
38 | EXPORT_SYMBOL(numa_cpu_lookup_table); | |
39 | EXPORT_SYMBOL(numa_cpumask_lookup_table); | |
40 | EXPORT_SYMBOL(node_data); | |
41 | ||
1da177e4 | 42 | static int min_common_depth; |
237a0989 | 43 | static int n_mem_addr_cells, n_mem_size_cells; |
1da177e4 | 44 | |
1daa6d08 BS |
45 | static int __cpuinit fake_numa_create_new_node(unsigned long end_pfn, |
46 | unsigned int *nid) | |
47 | { | |
48 | unsigned long long mem; | |
49 | char *p = cmdline; | |
50 | static unsigned int fake_nid; | |
51 | static unsigned long long curr_boundary; | |
52 | ||
53 | /* | |
54 | * Modify node id, iff we started creating NUMA nodes | |
55 | * We want to continue from where we left of the last time | |
56 | */ | |
57 | if (fake_nid) | |
58 | *nid = fake_nid; | |
59 | /* | |
60 | * In case there are no more arguments to parse, the | |
61 | * node_id should be the same as the last fake node id | |
62 | * (we've handled this above). | |
63 | */ | |
64 | if (!p) | |
65 | return 0; | |
66 | ||
67 | mem = memparse(p, &p); | |
68 | if (!mem) | |
69 | return 0; | |
70 | ||
71 | if (mem < curr_boundary) | |
72 | return 0; | |
73 | ||
74 | curr_boundary = mem; | |
75 | ||
76 | if ((end_pfn << PAGE_SHIFT) > mem) { | |
77 | /* | |
78 | * Skip commas and spaces | |
79 | */ | |
80 | while (*p == ',' || *p == ' ' || *p == '\t') | |
81 | p++; | |
82 | ||
83 | cmdline = p; | |
84 | fake_nid++; | |
85 | *nid = fake_nid; | |
86 | dbg("created new fake_node with id %d\n", fake_nid); | |
87 | return 1; | |
88 | } | |
89 | return 0; | |
90 | } | |
91 | ||
2e5ce39d | 92 | static void __cpuinit map_cpu_to_node(int cpu, int node) |
1da177e4 LT |
93 | { |
94 | numa_cpu_lookup_table[cpu] = node; | |
45fb6cea | 95 | |
bf4b85b0 NL |
96 | dbg("adding cpu %d to node %d\n", cpu, node); |
97 | ||
45fb6cea | 98 | if (!(cpu_isset(cpu, numa_cpumask_lookup_table[node]))) |
1da177e4 | 99 | cpu_set(cpu, numa_cpumask_lookup_table[node]); |
1da177e4 LT |
100 | } |
101 | ||
102 | #ifdef CONFIG_HOTPLUG_CPU | |
103 | static void unmap_cpu_from_node(unsigned long cpu) | |
104 | { | |
105 | int node = numa_cpu_lookup_table[cpu]; | |
106 | ||
107 | dbg("removing cpu %lu from node %d\n", cpu, node); | |
108 | ||
109 | if (cpu_isset(cpu, numa_cpumask_lookup_table[node])) { | |
110 | cpu_clear(cpu, numa_cpumask_lookup_table[node]); | |
1da177e4 LT |
111 | } else { |
112 | printk(KERN_ERR "WARNING: cpu %lu not found in node %d\n", | |
113 | cpu, node); | |
114 | } | |
115 | } | |
116 | #endif /* CONFIG_HOTPLUG_CPU */ | |
117 | ||
2e5ce39d | 118 | static struct device_node * __cpuinit find_cpu_node(unsigned int cpu) |
1da177e4 LT |
119 | { |
120 | unsigned int hw_cpuid = get_hard_smp_processor_id(cpu); | |
121 | struct device_node *cpu_node = NULL; | |
a7f67bdf | 122 | const unsigned int *interrupt_server, *reg; |
1da177e4 LT |
123 | int len; |
124 | ||
125 | while ((cpu_node = of_find_node_by_type(cpu_node, "cpu")) != NULL) { | |
126 | /* Try interrupt server first */ | |
e2eb6392 | 127 | interrupt_server = of_get_property(cpu_node, |
1da177e4 LT |
128 | "ibm,ppc-interrupt-server#s", &len); |
129 | ||
130 | len = len / sizeof(u32); | |
131 | ||
132 | if (interrupt_server && (len > 0)) { | |
133 | while (len--) { | |
134 | if (interrupt_server[len] == hw_cpuid) | |
135 | return cpu_node; | |
136 | } | |
137 | } else { | |
e2eb6392 | 138 | reg = of_get_property(cpu_node, "reg", &len); |
1da177e4 LT |
139 | if (reg && (len > 0) && (reg[0] == hw_cpuid)) |
140 | return cpu_node; | |
141 | } | |
142 | } | |
143 | ||
144 | return NULL; | |
145 | } | |
146 | ||
147 | /* must hold reference to node during call */ | |
a7f67bdf | 148 | static const int *of_get_associativity(struct device_node *dev) |
1da177e4 | 149 | { |
e2eb6392 | 150 | return of_get_property(dev, "ibm,associativity", NULL); |
1da177e4 LT |
151 | } |
152 | ||
cf00085d C |
153 | /* |
154 | * Returns the property linux,drconf-usable-memory if | |
155 | * it exists (the property exists only in kexec/kdump kernels, | |
156 | * added by kexec-tools) | |
157 | */ | |
158 | static const u32 *of_get_usable_memory(struct device_node *memory) | |
159 | { | |
160 | const u32 *prop; | |
161 | u32 len; | |
162 | prop = of_get_property(memory, "linux,drconf-usable-memory", &len); | |
163 | if (!prop || len < sizeof(unsigned int)) | |
164 | return 0; | |
165 | return prop; | |
166 | } | |
167 | ||
482ec7c4 NL |
168 | /* Returns nid in the range [0..MAX_NUMNODES-1], or -1 if no useful numa |
169 | * info is found. | |
170 | */ | |
953039c8 | 171 | static int of_node_to_nid_single(struct device_node *device) |
1da177e4 | 172 | { |
482ec7c4 | 173 | int nid = -1; |
a7f67bdf | 174 | const unsigned int *tmp; |
1da177e4 LT |
175 | |
176 | if (min_common_depth == -1) | |
482ec7c4 | 177 | goto out; |
1da177e4 LT |
178 | |
179 | tmp = of_get_associativity(device); | |
482ec7c4 NL |
180 | if (!tmp) |
181 | goto out; | |
182 | ||
183 | if (tmp[0] >= min_common_depth) | |
cf950b7a | 184 | nid = tmp[min_common_depth]; |
bc16a759 NL |
185 | |
186 | /* POWER4 LPAR uses 0xffff as invalid node */ | |
482ec7c4 NL |
187 | if (nid == 0xffff || nid >= MAX_NUMNODES) |
188 | nid = -1; | |
189 | out: | |
cf950b7a | 190 | return nid; |
1da177e4 LT |
191 | } |
192 | ||
953039c8 JK |
193 | /* Walk the device tree upwards, looking for an associativity id */ |
194 | int of_node_to_nid(struct device_node *device) | |
195 | { | |
196 | struct device_node *tmp; | |
197 | int nid = -1; | |
198 | ||
199 | of_node_get(device); | |
200 | while (device) { | |
201 | nid = of_node_to_nid_single(device); | |
202 | if (nid != -1) | |
203 | break; | |
204 | ||
205 | tmp = device; | |
206 | device = of_get_parent(tmp); | |
207 | of_node_put(tmp); | |
208 | } | |
209 | of_node_put(device); | |
210 | ||
211 | return nid; | |
212 | } | |
213 | EXPORT_SYMBOL_GPL(of_node_to_nid); | |
214 | ||
1da177e4 LT |
215 | /* |
216 | * In theory, the "ibm,associativity" property may contain multiple | |
217 | * associativity lists because a resource may be multiply connected | |
218 | * into the machine. This resource then has different associativity | |
219 | * characteristics relative to its multiple connections. We ignore | |
220 | * this for now. We also assume that all cpu and memory sets have | |
221 | * their distances represented at a common level. This won't be | |
1b3c3714 | 222 | * true for hierarchical NUMA. |
1da177e4 LT |
223 | * |
224 | * In any case the ibm,associativity-reference-points should give | |
225 | * the correct depth for a normal NUMA system. | |
226 | * | |
227 | * - Dave Hansen <haveblue@us.ibm.com> | |
228 | */ | |
229 | static int __init find_min_common_depth(void) | |
230 | { | |
231 | int depth; | |
a7f67bdf | 232 | const unsigned int *ref_points; |
1da177e4 LT |
233 | struct device_node *rtas_root; |
234 | unsigned int len; | |
235 | ||
236 | rtas_root = of_find_node_by_path("/rtas"); | |
237 | ||
238 | if (!rtas_root) | |
239 | return -1; | |
240 | ||
241 | /* | |
242 | * this property is 2 32-bit integers, each representing a level of | |
243 | * depth in the associativity nodes. The first is for an SMP | |
244 | * configuration (should be all 0's) and the second is for a normal | |
245 | * NUMA configuration. | |
246 | */ | |
e2eb6392 | 247 | ref_points = of_get_property(rtas_root, |
1da177e4 LT |
248 | "ibm,associativity-reference-points", &len); |
249 | ||
250 | if ((len >= 1) && ref_points) { | |
251 | depth = ref_points[1]; | |
252 | } else { | |
bf4b85b0 | 253 | dbg("NUMA: ibm,associativity-reference-points not found.\n"); |
1da177e4 LT |
254 | depth = -1; |
255 | } | |
256 | of_node_put(rtas_root); | |
257 | ||
258 | return depth; | |
259 | } | |
260 | ||
84c9fdd1 | 261 | static void __init get_n_mem_cells(int *n_addr_cells, int *n_size_cells) |
1da177e4 LT |
262 | { |
263 | struct device_node *memory = NULL; | |
1da177e4 LT |
264 | |
265 | memory = of_find_node_by_type(memory, "memory"); | |
54c23310 | 266 | if (!memory) |
84c9fdd1 | 267 | panic("numa.c: No memory nodes found!"); |
54c23310 | 268 | |
a8bda5dd | 269 | *n_addr_cells = of_n_addr_cells(memory); |
9213feea | 270 | *n_size_cells = of_n_size_cells(memory); |
84c9fdd1 | 271 | of_node_put(memory); |
1da177e4 LT |
272 | } |
273 | ||
a7f67bdf | 274 | static unsigned long __devinit read_n_cells(int n, const unsigned int **buf) |
1da177e4 LT |
275 | { |
276 | unsigned long result = 0; | |
277 | ||
278 | while (n--) { | |
279 | result = (result << 32) | **buf; | |
280 | (*buf)++; | |
281 | } | |
282 | return result; | |
283 | } | |
284 | ||
8342681d NF |
285 | struct of_drconf_cell { |
286 | u64 base_addr; | |
287 | u32 drc_index; | |
288 | u32 reserved; | |
289 | u32 aa_index; | |
290 | u32 flags; | |
291 | }; | |
292 | ||
293 | #define DRCONF_MEM_ASSIGNED 0x00000008 | |
294 | #define DRCONF_MEM_AI_INVALID 0x00000040 | |
295 | #define DRCONF_MEM_RESERVED 0x00000080 | |
296 | ||
297 | /* | |
298 | * Read the next lmb list entry from the ibm,dynamic-memory property | |
299 | * and return the information in the provided of_drconf_cell structure. | |
300 | */ | |
301 | static void read_drconf_cell(struct of_drconf_cell *drmem, const u32 **cellp) | |
302 | { | |
303 | const u32 *cp; | |
304 | ||
305 | drmem->base_addr = read_n_cells(n_mem_addr_cells, cellp); | |
306 | ||
307 | cp = *cellp; | |
308 | drmem->drc_index = cp[0]; | |
309 | drmem->reserved = cp[1]; | |
310 | drmem->aa_index = cp[2]; | |
311 | drmem->flags = cp[3]; | |
312 | ||
313 | *cellp = cp + 4; | |
314 | } | |
315 | ||
316 | /* | |
317 | * Retreive and validate the ibm,dynamic-memory property of the device tree. | |
318 | * | |
319 | * The layout of the ibm,dynamic-memory property is a number N of lmb | |
320 | * list entries followed by N lmb list entries. Each lmb list entry | |
321 | * contains information as layed out in the of_drconf_cell struct above. | |
322 | */ | |
323 | static int of_get_drconf_memory(struct device_node *memory, const u32 **dm) | |
324 | { | |
325 | const u32 *prop; | |
326 | u32 len, entries; | |
327 | ||
328 | prop = of_get_property(memory, "ibm,dynamic-memory", &len); | |
329 | if (!prop || len < sizeof(unsigned int)) | |
330 | return 0; | |
331 | ||
332 | entries = *prop++; | |
333 | ||
334 | /* Now that we know the number of entries, revalidate the size | |
335 | * of the property read in to ensure we have everything | |
336 | */ | |
337 | if (len < (entries * (n_mem_addr_cells + 4) + 1) * sizeof(unsigned int)) | |
338 | return 0; | |
339 | ||
340 | *dm = prop; | |
341 | return entries; | |
342 | } | |
343 | ||
344 | /* | |
345 | * Retreive and validate the ibm,lmb-size property for drconf memory | |
346 | * from the device tree. | |
347 | */ | |
348 | static u64 of_get_lmb_size(struct device_node *memory) | |
349 | { | |
350 | const u32 *prop; | |
351 | u32 len; | |
352 | ||
353 | prop = of_get_property(memory, "ibm,lmb-size", &len); | |
354 | if (!prop || len < sizeof(unsigned int)) | |
355 | return 0; | |
356 | ||
357 | return read_n_cells(n_mem_size_cells, &prop); | |
358 | } | |
359 | ||
360 | struct assoc_arrays { | |
361 | u32 n_arrays; | |
362 | u32 array_sz; | |
363 | const u32 *arrays; | |
364 | }; | |
365 | ||
366 | /* | |
367 | * Retreive and validate the list of associativity arrays for drconf | |
368 | * memory from the ibm,associativity-lookup-arrays property of the | |
369 | * device tree.. | |
370 | * | |
371 | * The layout of the ibm,associativity-lookup-arrays property is a number N | |
372 | * indicating the number of associativity arrays, followed by a number M | |
373 | * indicating the size of each associativity array, followed by a list | |
374 | * of N associativity arrays. | |
375 | */ | |
376 | static int of_get_assoc_arrays(struct device_node *memory, | |
377 | struct assoc_arrays *aa) | |
378 | { | |
379 | const u32 *prop; | |
380 | u32 len; | |
381 | ||
382 | prop = of_get_property(memory, "ibm,associativity-lookup-arrays", &len); | |
383 | if (!prop || len < 2 * sizeof(unsigned int)) | |
384 | return -1; | |
385 | ||
386 | aa->n_arrays = *prop++; | |
387 | aa->array_sz = *prop++; | |
388 | ||
389 | /* Now that we know the number of arrrays and size of each array, | |
390 | * revalidate the size of the property read in. | |
391 | */ | |
392 | if (len < (aa->n_arrays * aa->array_sz + 2) * sizeof(unsigned int)) | |
393 | return -1; | |
394 | ||
395 | aa->arrays = prop; | |
396 | return 0; | |
397 | } | |
398 | ||
399 | /* | |
400 | * This is like of_node_to_nid_single() for memory represented in the | |
401 | * ibm,dynamic-reconfiguration-memory node. | |
402 | */ | |
403 | static int of_drconf_to_nid_single(struct of_drconf_cell *drmem, | |
404 | struct assoc_arrays *aa) | |
405 | { | |
406 | int default_nid = 0; | |
407 | int nid = default_nid; | |
408 | int index; | |
409 | ||
410 | if (min_common_depth > 0 && min_common_depth <= aa->array_sz && | |
411 | !(drmem->flags & DRCONF_MEM_AI_INVALID) && | |
412 | drmem->aa_index < aa->n_arrays) { | |
413 | index = drmem->aa_index * aa->array_sz + min_common_depth - 1; | |
414 | nid = aa->arrays[index]; | |
415 | ||
416 | if (nid == 0xffff || nid >= MAX_NUMNODES) | |
417 | nid = default_nid; | |
418 | } | |
419 | ||
420 | return nid; | |
421 | } | |
422 | ||
1da177e4 LT |
423 | /* |
424 | * Figure out to which domain a cpu belongs and stick it there. | |
425 | * Return the id of the domain used. | |
426 | */ | |
2e5ce39d | 427 | static int __cpuinit numa_setup_cpu(unsigned long lcpu) |
1da177e4 | 428 | { |
cf950b7a | 429 | int nid = 0; |
1da177e4 LT |
430 | struct device_node *cpu = find_cpu_node(lcpu); |
431 | ||
432 | if (!cpu) { | |
433 | WARN_ON(1); | |
434 | goto out; | |
435 | } | |
436 | ||
953039c8 | 437 | nid = of_node_to_nid_single(cpu); |
1da177e4 | 438 | |
482ec7c4 NL |
439 | if (nid < 0 || !node_online(nid)) |
440 | nid = any_online_node(NODE_MASK_ALL); | |
1da177e4 | 441 | out: |
cf950b7a | 442 | map_cpu_to_node(lcpu, nid); |
1da177e4 LT |
443 | |
444 | of_node_put(cpu); | |
445 | ||
cf950b7a | 446 | return nid; |
1da177e4 LT |
447 | } |
448 | ||
74b85f37 | 449 | static int __cpuinit cpu_numa_callback(struct notifier_block *nfb, |
1da177e4 LT |
450 | unsigned long action, |
451 | void *hcpu) | |
452 | { | |
453 | unsigned long lcpu = (unsigned long)hcpu; | |
454 | int ret = NOTIFY_DONE; | |
455 | ||
456 | switch (action) { | |
457 | case CPU_UP_PREPARE: | |
8bb78442 | 458 | case CPU_UP_PREPARE_FROZEN: |
2b261227 | 459 | numa_setup_cpu(lcpu); |
1da177e4 LT |
460 | ret = NOTIFY_OK; |
461 | break; | |
462 | #ifdef CONFIG_HOTPLUG_CPU | |
463 | case CPU_DEAD: | |
8bb78442 | 464 | case CPU_DEAD_FROZEN: |
1da177e4 | 465 | case CPU_UP_CANCELED: |
8bb78442 | 466 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 LT |
467 | unmap_cpu_from_node(lcpu); |
468 | break; | |
469 | ret = NOTIFY_OK; | |
470 | #endif | |
471 | } | |
472 | return ret; | |
473 | } | |
474 | ||
475 | /* | |
476 | * Check and possibly modify a memory region to enforce the memory limit. | |
477 | * | |
478 | * Returns the size the region should have to enforce the memory limit. | |
479 | * This will either be the original value of size, a truncated value, | |
480 | * or zero. If the returned value of size is 0 the region should be | |
481 | * discarded as it lies wholy above the memory limit. | |
482 | */ | |
45fb6cea AB |
483 | static unsigned long __init numa_enforce_memory_limit(unsigned long start, |
484 | unsigned long size) | |
1da177e4 LT |
485 | { |
486 | /* | |
487 | * We use lmb_end_of_DRAM() in here instead of memory_limit because | |
488 | * we've already adjusted it for the limit and it takes care of | |
489 | * having memory holes below the limit. | |
490 | */ | |
1da177e4 LT |
491 | |
492 | if (! memory_limit) | |
493 | return size; | |
494 | ||
495 | if (start + size <= lmb_end_of_DRAM()) | |
496 | return size; | |
497 | ||
498 | if (start >= lmb_end_of_DRAM()) | |
499 | return 0; | |
500 | ||
501 | return lmb_end_of_DRAM() - start; | |
502 | } | |
503 | ||
cf00085d C |
504 | /* |
505 | * Reads the counter for a given entry in | |
506 | * linux,drconf-usable-memory property | |
507 | */ | |
508 | static inline int __init read_usm_ranges(const u32 **usm) | |
509 | { | |
510 | /* | |
511 | * For each lmb in ibm,dynamic-memory a corresponding | |
512 | * entry in linux,drconf-usable-memory property contains | |
513 | * a counter followed by that many (base, size) duple. | |
514 | * read the counter from linux,drconf-usable-memory | |
515 | */ | |
516 | return read_n_cells(n_mem_size_cells, usm); | |
517 | } | |
518 | ||
0204568a PM |
519 | /* |
520 | * Extract NUMA information from the ibm,dynamic-reconfiguration-memory | |
521 | * node. This assumes n_mem_{addr,size}_cells have been set. | |
522 | */ | |
523 | static void __init parse_drconf_memory(struct device_node *memory) | |
524 | { | |
cf00085d C |
525 | const u32 *dm, *usm; |
526 | unsigned int n, rc, ranges, is_kexec_kdump = 0; | |
527 | unsigned long lmb_size, base, size, sz; | |
8342681d NF |
528 | int nid; |
529 | struct assoc_arrays aa; | |
530 | ||
531 | n = of_get_drconf_memory(memory, &dm); | |
532 | if (!n) | |
0204568a PM |
533 | return; |
534 | ||
8342681d NF |
535 | lmb_size = of_get_lmb_size(memory); |
536 | if (!lmb_size) | |
537 | return; | |
538 | ||
539 | rc = of_get_assoc_arrays(memory, &aa); | |
540 | if (rc) | |
0204568a PM |
541 | return; |
542 | ||
cf00085d C |
543 | /* check if this is a kexec/kdump kernel */ |
544 | usm = of_get_usable_memory(memory); | |
545 | if (usm != NULL) | |
546 | is_kexec_kdump = 1; | |
547 | ||
0204568a | 548 | for (; n != 0; --n) { |
8342681d NF |
549 | struct of_drconf_cell drmem; |
550 | ||
551 | read_drconf_cell(&drmem, &dm); | |
552 | ||
553 | /* skip this block if the reserved bit is set in flags (0x80) | |
554 | or if the block is not assigned to this partition (0x8) */ | |
555 | if ((drmem.flags & DRCONF_MEM_RESERVED) | |
556 | || !(drmem.flags & DRCONF_MEM_ASSIGNED)) | |
0204568a | 557 | continue; |
1daa6d08 | 558 | |
cf00085d C |
559 | base = drmem.base_addr; |
560 | size = lmb_size; | |
561 | ranges = 1; | |
8342681d | 562 | |
cf00085d C |
563 | if (is_kexec_kdump) { |
564 | ranges = read_usm_ranges(&usm); | |
565 | if (!ranges) /* there are no (base, size) duple */ | |
566 | continue; | |
567 | } | |
568 | do { | |
569 | if (is_kexec_kdump) { | |
570 | base = read_n_cells(n_mem_addr_cells, &usm); | |
571 | size = read_n_cells(n_mem_size_cells, &usm); | |
572 | } | |
573 | nid = of_drconf_to_nid_single(&drmem, &aa); | |
574 | fake_numa_create_new_node( | |
575 | ((base + size) >> PAGE_SHIFT), | |
8342681d | 576 | &nid); |
cf00085d C |
577 | node_set_online(nid); |
578 | sz = numa_enforce_memory_limit(base, size); | |
579 | if (sz) | |
580 | add_active_range(nid, base >> PAGE_SHIFT, | |
581 | (base >> PAGE_SHIFT) | |
582 | + (sz >> PAGE_SHIFT)); | |
583 | } while (--ranges); | |
0204568a PM |
584 | } |
585 | } | |
586 | ||
1da177e4 LT |
587 | static int __init parse_numa_properties(void) |
588 | { | |
589 | struct device_node *cpu = NULL; | |
590 | struct device_node *memory = NULL; | |
482ec7c4 | 591 | int default_nid = 0; |
1da177e4 LT |
592 | unsigned long i; |
593 | ||
594 | if (numa_enabled == 0) { | |
595 | printk(KERN_WARNING "NUMA disabled by user\n"); | |
596 | return -1; | |
597 | } | |
598 | ||
1da177e4 LT |
599 | min_common_depth = find_min_common_depth(); |
600 | ||
1da177e4 LT |
601 | if (min_common_depth < 0) |
602 | return min_common_depth; | |
603 | ||
bf4b85b0 NL |
604 | dbg("NUMA associativity depth for CPU/Memory: %d\n", min_common_depth); |
605 | ||
1da177e4 | 606 | /* |
482ec7c4 NL |
607 | * Even though we connect cpus to numa domains later in SMP |
608 | * init, we need to know the node ids now. This is because | |
609 | * each node to be onlined must have NODE_DATA etc backing it. | |
1da177e4 | 610 | */ |
482ec7c4 | 611 | for_each_present_cpu(i) { |
cf950b7a | 612 | int nid; |
1da177e4 LT |
613 | |
614 | cpu = find_cpu_node(i); | |
482ec7c4 | 615 | BUG_ON(!cpu); |
953039c8 | 616 | nid = of_node_to_nid_single(cpu); |
482ec7c4 | 617 | of_node_put(cpu); |
1da177e4 | 618 | |
482ec7c4 NL |
619 | /* |
620 | * Don't fall back to default_nid yet -- we will plug | |
621 | * cpus into nodes once the memory scan has discovered | |
622 | * the topology. | |
623 | */ | |
624 | if (nid < 0) | |
625 | continue; | |
626 | node_set_online(nid); | |
1da177e4 LT |
627 | } |
628 | ||
237a0989 | 629 | get_n_mem_cells(&n_mem_addr_cells, &n_mem_size_cells); |
1da177e4 LT |
630 | memory = NULL; |
631 | while ((memory = of_find_node_by_type(memory, "memory")) != NULL) { | |
632 | unsigned long start; | |
633 | unsigned long size; | |
cf950b7a | 634 | int nid; |
1da177e4 | 635 | int ranges; |
a7f67bdf | 636 | const unsigned int *memcell_buf; |
1da177e4 LT |
637 | unsigned int len; |
638 | ||
e2eb6392 | 639 | memcell_buf = of_get_property(memory, |
ba759485 ME |
640 | "linux,usable-memory", &len); |
641 | if (!memcell_buf || len <= 0) | |
e2eb6392 | 642 | memcell_buf = of_get_property(memory, "reg", &len); |
1da177e4 LT |
643 | if (!memcell_buf || len <= 0) |
644 | continue; | |
645 | ||
cc5d0189 BH |
646 | /* ranges in cell */ |
647 | ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells); | |
1da177e4 LT |
648 | new_range: |
649 | /* these are order-sensitive, and modify the buffer pointer */ | |
237a0989 MK |
650 | start = read_n_cells(n_mem_addr_cells, &memcell_buf); |
651 | size = read_n_cells(n_mem_size_cells, &memcell_buf); | |
1da177e4 | 652 | |
482ec7c4 NL |
653 | /* |
654 | * Assumption: either all memory nodes or none will | |
655 | * have associativity properties. If none, then | |
656 | * everything goes to default_nid. | |
657 | */ | |
953039c8 | 658 | nid = of_node_to_nid_single(memory); |
482ec7c4 NL |
659 | if (nid < 0) |
660 | nid = default_nid; | |
1daa6d08 BS |
661 | |
662 | fake_numa_create_new_node(((start + size) >> PAGE_SHIFT), &nid); | |
482ec7c4 | 663 | node_set_online(nid); |
1da177e4 | 664 | |
45fb6cea | 665 | if (!(size = numa_enforce_memory_limit(start, size))) { |
1da177e4 LT |
666 | if (--ranges) |
667 | goto new_range; | |
668 | else | |
669 | continue; | |
670 | } | |
671 | ||
c67c3cb4 MG |
672 | add_active_range(nid, start >> PAGE_SHIFT, |
673 | (start >> PAGE_SHIFT) + (size >> PAGE_SHIFT)); | |
1da177e4 LT |
674 | |
675 | if (--ranges) | |
676 | goto new_range; | |
677 | } | |
678 | ||
0204568a PM |
679 | /* |
680 | * Now do the same thing for each LMB listed in the ibm,dynamic-memory | |
681 | * property in the ibm,dynamic-reconfiguration-memory node. | |
682 | */ | |
683 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); | |
684 | if (memory) | |
685 | parse_drconf_memory(memory); | |
686 | ||
1da177e4 LT |
687 | return 0; |
688 | } | |
689 | ||
690 | static void __init setup_nonnuma(void) | |
691 | { | |
692 | unsigned long top_of_ram = lmb_end_of_DRAM(); | |
693 | unsigned long total_ram = lmb_phys_mem_size(); | |
c67c3cb4 | 694 | unsigned long start_pfn, end_pfn; |
1daa6d08 | 695 | unsigned int i, nid = 0; |
1da177e4 | 696 | |
e110b281 | 697 | printk(KERN_DEBUG "Top of RAM: 0x%lx, Total RAM: 0x%lx\n", |
1da177e4 | 698 | top_of_ram, total_ram); |
e110b281 | 699 | printk(KERN_DEBUG "Memory hole size: %ldMB\n", |
1da177e4 LT |
700 | (top_of_ram - total_ram) >> 20); |
701 | ||
c67c3cb4 MG |
702 | for (i = 0; i < lmb.memory.cnt; ++i) { |
703 | start_pfn = lmb.memory.region[i].base >> PAGE_SHIFT; | |
704 | end_pfn = start_pfn + lmb_size_pages(&lmb.memory, i); | |
1daa6d08 BS |
705 | |
706 | fake_numa_create_new_node(end_pfn, &nid); | |
707 | add_active_range(nid, start_pfn, end_pfn); | |
708 | node_set_online(nid); | |
c67c3cb4 | 709 | } |
1da177e4 LT |
710 | } |
711 | ||
4b703a23 AB |
712 | void __init dump_numa_cpu_topology(void) |
713 | { | |
714 | unsigned int node; | |
715 | unsigned int cpu, count; | |
716 | ||
717 | if (min_common_depth == -1 || !numa_enabled) | |
718 | return; | |
719 | ||
720 | for_each_online_node(node) { | |
e110b281 | 721 | printk(KERN_DEBUG "Node %d CPUs:", node); |
4b703a23 AB |
722 | |
723 | count = 0; | |
724 | /* | |
725 | * If we used a CPU iterator here we would miss printing | |
726 | * the holes in the cpumap. | |
727 | */ | |
728 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
729 | if (cpu_isset(cpu, numa_cpumask_lookup_table[node])) { | |
730 | if (count == 0) | |
731 | printk(" %u", cpu); | |
732 | ++count; | |
733 | } else { | |
734 | if (count > 1) | |
735 | printk("-%u", cpu - 1); | |
736 | count = 0; | |
737 | } | |
738 | } | |
739 | ||
740 | if (count > 1) | |
741 | printk("-%u", NR_CPUS - 1); | |
742 | printk("\n"); | |
743 | } | |
744 | } | |
745 | ||
746 | static void __init dump_numa_memory_topology(void) | |
1da177e4 LT |
747 | { |
748 | unsigned int node; | |
749 | unsigned int count; | |
750 | ||
751 | if (min_common_depth == -1 || !numa_enabled) | |
752 | return; | |
753 | ||
754 | for_each_online_node(node) { | |
755 | unsigned long i; | |
756 | ||
e110b281 | 757 | printk(KERN_DEBUG "Node %d Memory:", node); |
1da177e4 LT |
758 | |
759 | count = 0; | |
760 | ||
45fb6cea AB |
761 | for (i = 0; i < lmb_end_of_DRAM(); |
762 | i += (1 << SECTION_SIZE_BITS)) { | |
763 | if (early_pfn_to_nid(i >> PAGE_SHIFT) == node) { | |
1da177e4 LT |
764 | if (count == 0) |
765 | printk(" 0x%lx", i); | |
766 | ++count; | |
767 | } else { | |
768 | if (count > 0) | |
769 | printk("-0x%lx", i); | |
770 | count = 0; | |
771 | } | |
772 | } | |
773 | ||
774 | if (count > 0) | |
775 | printk("-0x%lx", i); | |
776 | printk("\n"); | |
777 | } | |
1da177e4 LT |
778 | } |
779 | ||
780 | /* | |
781 | * Allocate some memory, satisfying the lmb or bootmem allocator where | |
782 | * required. nid is the preferred node and end is the physical address of | |
783 | * the highest address in the node. | |
784 | * | |
785 | * Returns the physical address of the memory. | |
786 | */ | |
45fb6cea AB |
787 | static void __init *careful_allocation(int nid, unsigned long size, |
788 | unsigned long align, | |
789 | unsigned long end_pfn) | |
1da177e4 | 790 | { |
45fb6cea | 791 | int new_nid; |
d7a5b2ff | 792 | unsigned long ret = __lmb_alloc_base(size, align, end_pfn << PAGE_SHIFT); |
1da177e4 LT |
793 | |
794 | /* retry over all memory */ | |
795 | if (!ret) | |
d7a5b2ff | 796 | ret = __lmb_alloc_base(size, align, lmb_end_of_DRAM()); |
1da177e4 LT |
797 | |
798 | if (!ret) | |
799 | panic("numa.c: cannot allocate %lu bytes on node %d", | |
800 | size, nid); | |
801 | ||
802 | /* | |
803 | * If the memory came from a previously allocated node, we must | |
804 | * retry with the bootmem allocator. | |
805 | */ | |
45fb6cea AB |
806 | new_nid = early_pfn_to_nid(ret >> PAGE_SHIFT); |
807 | if (new_nid < nid) { | |
808 | ret = (unsigned long)__alloc_bootmem_node(NODE_DATA(new_nid), | |
1da177e4 LT |
809 | size, align, 0); |
810 | ||
811 | if (!ret) | |
812 | panic("numa.c: cannot allocate %lu bytes on node %d", | |
45fb6cea | 813 | size, new_nid); |
1da177e4 | 814 | |
45fb6cea | 815 | ret = __pa(ret); |
1da177e4 LT |
816 | |
817 | dbg("alloc_bootmem %lx %lx\n", ret, size); | |
818 | } | |
819 | ||
45fb6cea | 820 | return (void *)ret; |
1da177e4 LT |
821 | } |
822 | ||
74b85f37 CS |
823 | static struct notifier_block __cpuinitdata ppc64_numa_nb = { |
824 | .notifier_call = cpu_numa_callback, | |
825 | .priority = 1 /* Must run before sched domains notifier. */ | |
826 | }; | |
827 | ||
1da177e4 LT |
828 | void __init do_init_bootmem(void) |
829 | { | |
830 | int nid; | |
45fb6cea | 831 | unsigned int i; |
1da177e4 LT |
832 | |
833 | min_low_pfn = 0; | |
834 | max_low_pfn = lmb_end_of_DRAM() >> PAGE_SHIFT; | |
835 | max_pfn = max_low_pfn; | |
836 | ||
837 | if (parse_numa_properties()) | |
838 | setup_nonnuma(); | |
839 | else | |
4b703a23 | 840 | dump_numa_memory_topology(); |
1da177e4 LT |
841 | |
842 | register_cpu_notifier(&ppc64_numa_nb); | |
2b261227 NL |
843 | cpu_numa_callback(&ppc64_numa_nb, CPU_UP_PREPARE, |
844 | (void *)(unsigned long)boot_cpuid); | |
1da177e4 LT |
845 | |
846 | for_each_online_node(nid) { | |
c67c3cb4 | 847 | unsigned long start_pfn, end_pfn; |
1da177e4 LT |
848 | unsigned long bootmem_paddr; |
849 | unsigned long bootmap_pages; | |
850 | ||
c67c3cb4 | 851 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); |
1da177e4 LT |
852 | |
853 | /* Allocate the node structure node local if possible */ | |
45fb6cea | 854 | NODE_DATA(nid) = careful_allocation(nid, |
1da177e4 | 855 | sizeof(struct pglist_data), |
45fb6cea AB |
856 | SMP_CACHE_BYTES, end_pfn); |
857 | NODE_DATA(nid) = __va(NODE_DATA(nid)); | |
1da177e4 LT |
858 | memset(NODE_DATA(nid), 0, sizeof(struct pglist_data)); |
859 | ||
860 | dbg("node %d\n", nid); | |
861 | dbg("NODE_DATA() = %p\n", NODE_DATA(nid)); | |
862 | ||
b61bfa3c | 863 | NODE_DATA(nid)->bdata = &bootmem_node_data[nid]; |
45fb6cea AB |
864 | NODE_DATA(nid)->node_start_pfn = start_pfn; |
865 | NODE_DATA(nid)->node_spanned_pages = end_pfn - start_pfn; | |
1da177e4 LT |
866 | |
867 | if (NODE_DATA(nid)->node_spanned_pages == 0) | |
868 | continue; | |
869 | ||
45fb6cea AB |
870 | dbg("start_paddr = %lx\n", start_pfn << PAGE_SHIFT); |
871 | dbg("end_paddr = %lx\n", end_pfn << PAGE_SHIFT); | |
1da177e4 | 872 | |
45fb6cea AB |
873 | bootmap_pages = bootmem_bootmap_pages(end_pfn - start_pfn); |
874 | bootmem_paddr = (unsigned long)careful_allocation(nid, | |
875 | bootmap_pages << PAGE_SHIFT, | |
876 | PAGE_SIZE, end_pfn); | |
877 | memset(__va(bootmem_paddr), 0, bootmap_pages << PAGE_SHIFT); | |
1da177e4 | 878 | |
1da177e4 LT |
879 | dbg("bootmap_paddr = %lx\n", bootmem_paddr); |
880 | ||
881 | init_bootmem_node(NODE_DATA(nid), bootmem_paddr >> PAGE_SHIFT, | |
45fb6cea | 882 | start_pfn, end_pfn); |
1da177e4 | 883 | |
c67c3cb4 | 884 | free_bootmem_with_active_regions(nid, end_pfn); |
1da177e4 | 885 | |
45fb6cea | 886 | /* Mark reserved regions on this node */ |
1da177e4 | 887 | for (i = 0; i < lmb.reserved.cnt; i++) { |
180379dc | 888 | unsigned long physbase = lmb.reserved.region[i].base; |
1da177e4 | 889 | unsigned long size = lmb.reserved.region[i].size; |
45fb6cea AB |
890 | unsigned long start_paddr = start_pfn << PAGE_SHIFT; |
891 | unsigned long end_paddr = end_pfn << PAGE_SHIFT; | |
1da177e4 | 892 | |
45fb6cea AB |
893 | if (early_pfn_to_nid(physbase >> PAGE_SHIFT) != nid && |
894 | early_pfn_to_nid((physbase+size-1) >> PAGE_SHIFT) != nid) | |
1da177e4 LT |
895 | continue; |
896 | ||
897 | if (physbase < end_paddr && | |
898 | (physbase+size) > start_paddr) { | |
899 | /* overlaps */ | |
900 | if (physbase < start_paddr) { | |
901 | size -= start_paddr - physbase; | |
902 | physbase = start_paddr; | |
903 | } | |
904 | ||
905 | if (size > end_paddr - physbase) | |
906 | size = end_paddr - physbase; | |
907 | ||
908 | dbg("reserve_bootmem %lx %lx\n", physbase, | |
909 | size); | |
910 | reserve_bootmem_node(NODE_DATA(nid), physbase, | |
72a7fe39 | 911 | size, BOOTMEM_DEFAULT); |
1da177e4 LT |
912 | } |
913 | } | |
802f192e | 914 | |
c67c3cb4 | 915 | sparse_memory_present_with_active_regions(nid); |
1da177e4 LT |
916 | } |
917 | } | |
918 | ||
919 | void __init paging_init(void) | |
920 | { | |
6391af17 MG |
921 | unsigned long max_zone_pfns[MAX_NR_ZONES]; |
922 | memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); | |
923 | max_zone_pfns[ZONE_DMA] = lmb_end_of_DRAM() >> PAGE_SHIFT; | |
c67c3cb4 | 924 | free_area_init_nodes(max_zone_pfns); |
1da177e4 LT |
925 | } |
926 | ||
927 | static int __init early_numa(char *p) | |
928 | { | |
929 | if (!p) | |
930 | return 0; | |
931 | ||
932 | if (strstr(p, "off")) | |
933 | numa_enabled = 0; | |
934 | ||
935 | if (strstr(p, "debug")) | |
936 | numa_debug = 1; | |
937 | ||
1daa6d08 BS |
938 | p = strstr(p, "fake="); |
939 | if (p) | |
940 | cmdline = p + strlen("fake="); | |
941 | ||
1da177e4 LT |
942 | return 0; |
943 | } | |
944 | early_param("numa", early_numa); | |
237a0989 MK |
945 | |
946 | #ifdef CONFIG_MEMORY_HOTPLUG | |
0db9360a NF |
947 | /* |
948 | * Validate the node associated with the memory section we are | |
949 | * trying to add. | |
950 | */ | |
951 | int valid_hot_add_scn(int *nid, unsigned long start, u32 lmb_size, | |
952 | unsigned long scn_addr) | |
953 | { | |
954 | nodemask_t nodes; | |
955 | ||
956 | if (*nid < 0 || !node_online(*nid)) | |
957 | *nid = any_online_node(NODE_MASK_ALL); | |
958 | ||
959 | if ((scn_addr >= start) && (scn_addr < (start + lmb_size))) { | |
960 | nodes_setall(nodes); | |
961 | while (NODE_DATA(*nid)->node_spanned_pages == 0) { | |
962 | node_clear(*nid, nodes); | |
963 | *nid = any_online_node(nodes); | |
964 | } | |
965 | ||
966 | return 1; | |
967 | } | |
968 | ||
969 | return 0; | |
970 | } | |
971 | ||
972 | /* | |
973 | * Find the node associated with a hot added memory section represented | |
974 | * by the ibm,dynamic-reconfiguration-memory node. | |
975 | */ | |
976 | static int hot_add_drconf_scn_to_nid(struct device_node *memory, | |
977 | unsigned long scn_addr) | |
978 | { | |
979 | const u32 *dm; | |
980 | unsigned int n, rc; | |
981 | unsigned long lmb_size; | |
982 | int default_nid = any_online_node(NODE_MASK_ALL); | |
983 | int nid; | |
984 | struct assoc_arrays aa; | |
985 | ||
986 | n = of_get_drconf_memory(memory, &dm); | |
987 | if (!n) | |
988 | return default_nid;; | |
989 | ||
990 | lmb_size = of_get_lmb_size(memory); | |
991 | if (!lmb_size) | |
992 | return default_nid; | |
993 | ||
994 | rc = of_get_assoc_arrays(memory, &aa); | |
995 | if (rc) | |
996 | return default_nid; | |
997 | ||
998 | for (; n != 0; --n) { | |
999 | struct of_drconf_cell drmem; | |
1000 | ||
1001 | read_drconf_cell(&drmem, &dm); | |
1002 | ||
1003 | /* skip this block if it is reserved or not assigned to | |
1004 | * this partition */ | |
1005 | if ((drmem.flags & DRCONF_MEM_RESERVED) | |
1006 | || !(drmem.flags & DRCONF_MEM_ASSIGNED)) | |
1007 | continue; | |
1008 | ||
1009 | nid = of_drconf_to_nid_single(&drmem, &aa); | |
1010 | ||
1011 | if (valid_hot_add_scn(&nid, drmem.base_addr, lmb_size, | |
1012 | scn_addr)) | |
1013 | return nid; | |
1014 | } | |
1015 | ||
1016 | BUG(); /* section address should be found above */ | |
1017 | return 0; | |
1018 | } | |
1019 | ||
237a0989 MK |
1020 | /* |
1021 | * Find the node associated with a hot added memory section. Section | |
1022 | * corresponds to a SPARSEMEM section, not an LMB. It is assumed that | |
1023 | * sections are fully contained within a single LMB. | |
1024 | */ | |
1025 | int hot_add_scn_to_nid(unsigned long scn_addr) | |
1026 | { | |
1027 | struct device_node *memory = NULL; | |
069007ae | 1028 | int nid; |
237a0989 MK |
1029 | |
1030 | if (!numa_enabled || (min_common_depth < 0)) | |
0db9360a NF |
1031 | return any_online_node(NODE_MASK_ALL); |
1032 | ||
1033 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); | |
1034 | if (memory) { | |
1035 | nid = hot_add_drconf_scn_to_nid(memory, scn_addr); | |
1036 | of_node_put(memory); | |
1037 | return nid; | |
1038 | } | |
237a0989 MK |
1039 | |
1040 | while ((memory = of_find_node_by_type(memory, "memory")) != NULL) { | |
1041 | unsigned long start, size; | |
b226e462 | 1042 | int ranges; |
a7f67bdf | 1043 | const unsigned int *memcell_buf; |
237a0989 MK |
1044 | unsigned int len; |
1045 | ||
e2eb6392 | 1046 | memcell_buf = of_get_property(memory, "reg", &len); |
237a0989 MK |
1047 | if (!memcell_buf || len <= 0) |
1048 | continue; | |
1049 | ||
cc5d0189 BH |
1050 | /* ranges in cell */ |
1051 | ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells); | |
237a0989 MK |
1052 | ha_new_range: |
1053 | start = read_n_cells(n_mem_addr_cells, &memcell_buf); | |
1054 | size = read_n_cells(n_mem_size_cells, &memcell_buf); | |
953039c8 | 1055 | nid = of_node_to_nid_single(memory); |
237a0989 | 1056 | |
0db9360a | 1057 | if (valid_hot_add_scn(&nid, start, size, scn_addr)) { |
237a0989 | 1058 | of_node_put(memory); |
0db9360a | 1059 | return nid; |
237a0989 MK |
1060 | } |
1061 | ||
1062 | if (--ranges) /* process all ranges in cell */ | |
1063 | goto ha_new_range; | |
1064 | } | |
237a0989 | 1065 | BUG(); /* section address should be found above */ |
069007ae | 1066 | return 0; |
237a0989 MK |
1067 | } |
1068 | #endif /* CONFIG_MEMORY_HOTPLUG */ |