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> | |
4b16f8e2 | 16 | #include <linux/export.h> |
1da177e4 LT |
17 | #include <linux/nodemask.h> |
18 | #include <linux/cpu.h> | |
19 | #include <linux/notifier.h> | |
95f72d1e | 20 | #include <linux/memblock.h> |
6df1646e | 21 | #include <linux/of.h> |
06eccea6 | 22 | #include <linux/pfn.h> |
9eff1a38 JL |
23 | #include <linux/cpuset.h> |
24 | #include <linux/node.h> | |
45fb6cea | 25 | #include <asm/sparsemem.h> |
d9b2b2a2 | 26 | #include <asm/prom.h> |
2249ca9d | 27 | #include <asm/smp.h> |
9eff1a38 JL |
28 | #include <asm/firmware.h> |
29 | #include <asm/paca.h> | |
39bf990e | 30 | #include <asm/hvcall.h> |
ae3a197e | 31 | #include <asm/setup.h> |
1da177e4 LT |
32 | |
33 | static int numa_enabled = 1; | |
34 | ||
1daa6d08 BS |
35 | static char *cmdline __initdata; |
36 | ||
1da177e4 LT |
37 | static int numa_debug; |
38 | #define dbg(args...) if (numa_debug) { printk(KERN_INFO args); } | |
39 | ||
45fb6cea | 40 | int numa_cpu_lookup_table[NR_CPUS]; |
25863de0 | 41 | cpumask_var_t node_to_cpumask_map[MAX_NUMNODES]; |
1da177e4 | 42 | struct pglist_data *node_data[MAX_NUMNODES]; |
45fb6cea AB |
43 | |
44 | EXPORT_SYMBOL(numa_cpu_lookup_table); | |
25863de0 | 45 | EXPORT_SYMBOL(node_to_cpumask_map); |
45fb6cea AB |
46 | EXPORT_SYMBOL(node_data); |
47 | ||
1da177e4 | 48 | static int min_common_depth; |
237a0989 | 49 | static int n_mem_addr_cells, n_mem_size_cells; |
41eab6f8 AB |
50 | static int form1_affinity; |
51 | ||
52 | #define MAX_DISTANCE_REF_POINTS 4 | |
53 | static int distance_ref_points_depth; | |
54 | static const unsigned int *distance_ref_points; | |
55 | static int distance_lookup_table[MAX_NUMNODES][MAX_DISTANCE_REF_POINTS]; | |
1da177e4 | 56 | |
25863de0 AB |
57 | /* |
58 | * Allocate node_to_cpumask_map based on number of available nodes | |
59 | * Requires node_possible_map to be valid. | |
60 | * | |
9512938b | 61 | * Note: cpumask_of_node() is not valid until after this is done. |
25863de0 AB |
62 | */ |
63 | static void __init setup_node_to_cpumask_map(void) | |
64 | { | |
65 | unsigned int node, num = 0; | |
66 | ||
67 | /* setup nr_node_ids if not done yet */ | |
68 | if (nr_node_ids == MAX_NUMNODES) { | |
69 | for_each_node_mask(node, node_possible_map) | |
70 | num = node; | |
71 | nr_node_ids = num + 1; | |
72 | } | |
73 | ||
74 | /* allocate the map */ | |
75 | for (node = 0; node < nr_node_ids; node++) | |
76 | alloc_bootmem_cpumask_var(&node_to_cpumask_map[node]); | |
77 | ||
78 | /* cpumask_of_node() will now work */ | |
79 | dbg("Node to cpumask map for %d nodes\n", nr_node_ids); | |
80 | } | |
81 | ||
1daa6d08 BS |
82 | static int __cpuinit fake_numa_create_new_node(unsigned long end_pfn, |
83 | unsigned int *nid) | |
84 | { | |
85 | unsigned long long mem; | |
86 | char *p = cmdline; | |
87 | static unsigned int fake_nid; | |
88 | static unsigned long long curr_boundary; | |
89 | ||
90 | /* | |
91 | * Modify node id, iff we started creating NUMA nodes | |
92 | * We want to continue from where we left of the last time | |
93 | */ | |
94 | if (fake_nid) | |
95 | *nid = fake_nid; | |
96 | /* | |
97 | * In case there are no more arguments to parse, the | |
98 | * node_id should be the same as the last fake node id | |
99 | * (we've handled this above). | |
100 | */ | |
101 | if (!p) | |
102 | return 0; | |
103 | ||
104 | mem = memparse(p, &p); | |
105 | if (!mem) | |
106 | return 0; | |
107 | ||
108 | if (mem < curr_boundary) | |
109 | return 0; | |
110 | ||
111 | curr_boundary = mem; | |
112 | ||
113 | if ((end_pfn << PAGE_SHIFT) > mem) { | |
114 | /* | |
115 | * Skip commas and spaces | |
116 | */ | |
117 | while (*p == ',' || *p == ' ' || *p == '\t') | |
118 | p++; | |
119 | ||
120 | cmdline = p; | |
121 | fake_nid++; | |
122 | *nid = fake_nid; | |
123 | dbg("created new fake_node with id %d\n", fake_nid); | |
124 | return 1; | |
125 | } | |
126 | return 0; | |
127 | } | |
128 | ||
8f64e1f2 | 129 | /* |
5dfe8660 | 130 | * get_node_active_region - Return active region containing pfn |
e8170372 | 131 | * Active range returned is empty if none found. |
5dfe8660 TH |
132 | * @pfn: The page to return the region for |
133 | * @node_ar: Returned set to the active region containing @pfn | |
8f64e1f2 | 134 | */ |
5dfe8660 TH |
135 | static void __init get_node_active_region(unsigned long pfn, |
136 | struct node_active_region *node_ar) | |
8f64e1f2 | 137 | { |
5dfe8660 TH |
138 | unsigned long start_pfn, end_pfn; |
139 | int i, nid; | |
140 | ||
141 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { | |
142 | if (pfn >= start_pfn && pfn < end_pfn) { | |
143 | node_ar->nid = nid; | |
144 | node_ar->start_pfn = start_pfn; | |
145 | node_ar->end_pfn = end_pfn; | |
146 | break; | |
147 | } | |
148 | } | |
8f64e1f2 JT |
149 | } |
150 | ||
39bf990e | 151 | static void map_cpu_to_node(int cpu, int node) |
1da177e4 LT |
152 | { |
153 | numa_cpu_lookup_table[cpu] = node; | |
45fb6cea | 154 | |
bf4b85b0 NL |
155 | dbg("adding cpu %d to node %d\n", cpu, node); |
156 | ||
25863de0 AB |
157 | if (!(cpumask_test_cpu(cpu, node_to_cpumask_map[node]))) |
158 | cpumask_set_cpu(cpu, node_to_cpumask_map[node]); | |
1da177e4 LT |
159 | } |
160 | ||
39bf990e | 161 | #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_PPC_SPLPAR) |
1da177e4 LT |
162 | static void unmap_cpu_from_node(unsigned long cpu) |
163 | { | |
164 | int node = numa_cpu_lookup_table[cpu]; | |
165 | ||
166 | dbg("removing cpu %lu from node %d\n", cpu, node); | |
167 | ||
25863de0 | 168 | if (cpumask_test_cpu(cpu, node_to_cpumask_map[node])) { |
429f4d8d | 169 | cpumask_clear_cpu(cpu, node_to_cpumask_map[node]); |
1da177e4 LT |
170 | } else { |
171 | printk(KERN_ERR "WARNING: cpu %lu not found in node %d\n", | |
172 | cpu, node); | |
173 | } | |
174 | } | |
39bf990e | 175 | #endif /* CONFIG_HOTPLUG_CPU || CONFIG_PPC_SPLPAR */ |
1da177e4 | 176 | |
1da177e4 | 177 | /* must hold reference to node during call */ |
a7f67bdf | 178 | static const int *of_get_associativity(struct device_node *dev) |
1da177e4 | 179 | { |
e2eb6392 | 180 | return of_get_property(dev, "ibm,associativity", NULL); |
1da177e4 LT |
181 | } |
182 | ||
cf00085d C |
183 | /* |
184 | * Returns the property linux,drconf-usable-memory if | |
185 | * it exists (the property exists only in kexec/kdump kernels, | |
186 | * added by kexec-tools) | |
187 | */ | |
188 | static const u32 *of_get_usable_memory(struct device_node *memory) | |
189 | { | |
190 | const u32 *prop; | |
191 | u32 len; | |
192 | prop = of_get_property(memory, "linux,drconf-usable-memory", &len); | |
193 | if (!prop || len < sizeof(unsigned int)) | |
194 | return 0; | |
195 | return prop; | |
196 | } | |
197 | ||
41eab6f8 AB |
198 | int __node_distance(int a, int b) |
199 | { | |
200 | int i; | |
201 | int distance = LOCAL_DISTANCE; | |
202 | ||
203 | if (!form1_affinity) | |
204 | return distance; | |
205 | ||
206 | for (i = 0; i < distance_ref_points_depth; i++) { | |
207 | if (distance_lookup_table[a][i] == distance_lookup_table[b][i]) | |
208 | break; | |
209 | ||
210 | /* Double the distance for each NUMA level */ | |
211 | distance *= 2; | |
212 | } | |
213 | ||
214 | return distance; | |
215 | } | |
216 | ||
217 | static void initialize_distance_lookup_table(int nid, | |
218 | const unsigned int *associativity) | |
219 | { | |
220 | int i; | |
221 | ||
222 | if (!form1_affinity) | |
223 | return; | |
224 | ||
225 | for (i = 0; i < distance_ref_points_depth; i++) { | |
226 | distance_lookup_table[nid][i] = | |
227 | associativity[distance_ref_points[i]]; | |
228 | } | |
229 | } | |
230 | ||
482ec7c4 NL |
231 | /* Returns nid in the range [0..MAX_NUMNODES-1], or -1 if no useful numa |
232 | * info is found. | |
233 | */ | |
9eff1a38 | 234 | static int associativity_to_nid(const unsigned int *associativity) |
1da177e4 | 235 | { |
482ec7c4 | 236 | int nid = -1; |
1da177e4 LT |
237 | |
238 | if (min_common_depth == -1) | |
482ec7c4 | 239 | goto out; |
1da177e4 | 240 | |
9eff1a38 JL |
241 | if (associativity[0] >= min_common_depth) |
242 | nid = associativity[min_common_depth]; | |
bc16a759 NL |
243 | |
244 | /* POWER4 LPAR uses 0xffff as invalid node */ | |
482ec7c4 NL |
245 | if (nid == 0xffff || nid >= MAX_NUMNODES) |
246 | nid = -1; | |
41eab6f8 | 247 | |
9eff1a38 JL |
248 | if (nid > 0 && associativity[0] >= distance_ref_points_depth) |
249 | initialize_distance_lookup_table(nid, associativity); | |
41eab6f8 | 250 | |
482ec7c4 | 251 | out: |
cf950b7a | 252 | return nid; |
1da177e4 LT |
253 | } |
254 | ||
9eff1a38 JL |
255 | /* Returns the nid associated with the given device tree node, |
256 | * or -1 if not found. | |
257 | */ | |
258 | static int of_node_to_nid_single(struct device_node *device) | |
259 | { | |
260 | int nid = -1; | |
261 | const unsigned int *tmp; | |
262 | ||
263 | tmp = of_get_associativity(device); | |
264 | if (tmp) | |
265 | nid = associativity_to_nid(tmp); | |
266 | return nid; | |
267 | } | |
268 | ||
953039c8 JK |
269 | /* Walk the device tree upwards, looking for an associativity id */ |
270 | int of_node_to_nid(struct device_node *device) | |
271 | { | |
272 | struct device_node *tmp; | |
273 | int nid = -1; | |
274 | ||
275 | of_node_get(device); | |
276 | while (device) { | |
277 | nid = of_node_to_nid_single(device); | |
278 | if (nid != -1) | |
279 | break; | |
280 | ||
281 | tmp = device; | |
282 | device = of_get_parent(tmp); | |
283 | of_node_put(tmp); | |
284 | } | |
285 | of_node_put(device); | |
286 | ||
287 | return nid; | |
288 | } | |
289 | EXPORT_SYMBOL_GPL(of_node_to_nid); | |
290 | ||
1da177e4 LT |
291 | static int __init find_min_common_depth(void) |
292 | { | |
41eab6f8 | 293 | int depth; |
bc8449cc | 294 | struct device_node *chosen; |
e70606eb | 295 | struct device_node *root; |
bc8449cc | 296 | const char *vec5; |
1da177e4 | 297 | |
1c8ee733 DS |
298 | if (firmware_has_feature(FW_FEATURE_OPAL)) |
299 | root = of_find_node_by_path("/ibm,opal"); | |
300 | else | |
301 | root = of_find_node_by_path("/rtas"); | |
e70606eb ME |
302 | if (!root) |
303 | root = of_find_node_by_path("/"); | |
1da177e4 LT |
304 | |
305 | /* | |
41eab6f8 AB |
306 | * This property is a set of 32-bit integers, each representing |
307 | * an index into the ibm,associativity nodes. | |
308 | * | |
309 | * With form 0 affinity the first integer is for an SMP configuration | |
310 | * (should be all 0's) and the second is for a normal NUMA | |
311 | * configuration. We have only one level of NUMA. | |
312 | * | |
313 | * With form 1 affinity the first integer is the most significant | |
314 | * NUMA boundary and the following are progressively less significant | |
315 | * boundaries. There can be more than one level of NUMA. | |
1da177e4 | 316 | */ |
e70606eb | 317 | distance_ref_points = of_get_property(root, |
41eab6f8 AB |
318 | "ibm,associativity-reference-points", |
319 | &distance_ref_points_depth); | |
320 | ||
321 | if (!distance_ref_points) { | |
322 | dbg("NUMA: ibm,associativity-reference-points not found.\n"); | |
323 | goto err; | |
324 | } | |
325 | ||
326 | distance_ref_points_depth /= sizeof(int); | |
1da177e4 | 327 | |
bc8449cc AB |
328 | #define VEC5_AFFINITY_BYTE 5 |
329 | #define VEC5_AFFINITY 0x80 | |
1c8ee733 DS |
330 | |
331 | if (firmware_has_feature(FW_FEATURE_OPAL)) | |
332 | form1_affinity = 1; | |
333 | else { | |
334 | chosen = of_find_node_by_path("/chosen"); | |
335 | if (chosen) { | |
336 | vec5 = of_get_property(chosen, | |
337 | "ibm,architecture-vec-5", NULL); | |
338 | if (vec5 && (vec5[VEC5_AFFINITY_BYTE] & | |
339 | VEC5_AFFINITY)) { | |
340 | dbg("Using form 1 affinity\n"); | |
341 | form1_affinity = 1; | |
342 | } | |
5b958a7e GS |
343 | |
344 | of_node_put(chosen); | |
bc8449cc | 345 | } |
4b83c330 AB |
346 | } |
347 | ||
41eab6f8 AB |
348 | if (form1_affinity) { |
349 | depth = distance_ref_points[0]; | |
1da177e4 | 350 | } else { |
41eab6f8 AB |
351 | if (distance_ref_points_depth < 2) { |
352 | printk(KERN_WARNING "NUMA: " | |
353 | "short ibm,associativity-reference-points\n"); | |
354 | goto err; | |
355 | } | |
356 | ||
357 | depth = distance_ref_points[1]; | |
1da177e4 | 358 | } |
1da177e4 | 359 | |
41eab6f8 AB |
360 | /* |
361 | * Warn and cap if the hardware supports more than | |
362 | * MAX_DISTANCE_REF_POINTS domains. | |
363 | */ | |
364 | if (distance_ref_points_depth > MAX_DISTANCE_REF_POINTS) { | |
365 | printk(KERN_WARNING "NUMA: distance array capped at " | |
366 | "%d entries\n", MAX_DISTANCE_REF_POINTS); | |
367 | distance_ref_points_depth = MAX_DISTANCE_REF_POINTS; | |
368 | } | |
369 | ||
e70606eb | 370 | of_node_put(root); |
1da177e4 | 371 | return depth; |
41eab6f8 AB |
372 | |
373 | err: | |
e70606eb | 374 | of_node_put(root); |
41eab6f8 | 375 | return -1; |
1da177e4 LT |
376 | } |
377 | ||
84c9fdd1 | 378 | static void __init get_n_mem_cells(int *n_addr_cells, int *n_size_cells) |
1da177e4 LT |
379 | { |
380 | struct device_node *memory = NULL; | |
1da177e4 LT |
381 | |
382 | memory = of_find_node_by_type(memory, "memory"); | |
54c23310 | 383 | if (!memory) |
84c9fdd1 | 384 | panic("numa.c: No memory nodes found!"); |
54c23310 | 385 | |
a8bda5dd | 386 | *n_addr_cells = of_n_addr_cells(memory); |
9213feea | 387 | *n_size_cells = of_n_size_cells(memory); |
84c9fdd1 | 388 | of_node_put(memory); |
1da177e4 LT |
389 | } |
390 | ||
2011b1d0 | 391 | static unsigned long read_n_cells(int n, const unsigned int **buf) |
1da177e4 LT |
392 | { |
393 | unsigned long result = 0; | |
394 | ||
395 | while (n--) { | |
396 | result = (result << 32) | **buf; | |
397 | (*buf)++; | |
398 | } | |
399 | return result; | |
400 | } | |
401 | ||
8342681d | 402 | /* |
95f72d1e | 403 | * Read the next memblock list entry from the ibm,dynamic-memory property |
8342681d NF |
404 | * and return the information in the provided of_drconf_cell structure. |
405 | */ | |
406 | static void read_drconf_cell(struct of_drconf_cell *drmem, const u32 **cellp) | |
407 | { | |
408 | const u32 *cp; | |
409 | ||
410 | drmem->base_addr = read_n_cells(n_mem_addr_cells, cellp); | |
411 | ||
412 | cp = *cellp; | |
413 | drmem->drc_index = cp[0]; | |
414 | drmem->reserved = cp[1]; | |
415 | drmem->aa_index = cp[2]; | |
416 | drmem->flags = cp[3]; | |
417 | ||
418 | *cellp = cp + 4; | |
419 | } | |
420 | ||
421 | /* | |
25985edc | 422 | * Retrieve and validate the ibm,dynamic-memory property of the device tree. |
8342681d | 423 | * |
95f72d1e YL |
424 | * The layout of the ibm,dynamic-memory property is a number N of memblock |
425 | * list entries followed by N memblock list entries. Each memblock list entry | |
25985edc | 426 | * contains information as laid out in the of_drconf_cell struct above. |
8342681d NF |
427 | */ |
428 | static int of_get_drconf_memory(struct device_node *memory, const u32 **dm) | |
429 | { | |
430 | const u32 *prop; | |
431 | u32 len, entries; | |
432 | ||
433 | prop = of_get_property(memory, "ibm,dynamic-memory", &len); | |
434 | if (!prop || len < sizeof(unsigned int)) | |
435 | return 0; | |
436 | ||
437 | entries = *prop++; | |
438 | ||
439 | /* Now that we know the number of entries, revalidate the size | |
440 | * of the property read in to ensure we have everything | |
441 | */ | |
442 | if (len < (entries * (n_mem_addr_cells + 4) + 1) * sizeof(unsigned int)) | |
443 | return 0; | |
444 | ||
445 | *dm = prop; | |
446 | return entries; | |
447 | } | |
448 | ||
449 | /* | |
25985edc | 450 | * Retrieve and validate the ibm,lmb-size property for drconf memory |
8342681d NF |
451 | * from the device tree. |
452 | */ | |
3fdfd990 | 453 | static u64 of_get_lmb_size(struct device_node *memory) |
8342681d NF |
454 | { |
455 | const u32 *prop; | |
456 | u32 len; | |
457 | ||
3fdfd990 | 458 | prop = of_get_property(memory, "ibm,lmb-size", &len); |
8342681d NF |
459 | if (!prop || len < sizeof(unsigned int)) |
460 | return 0; | |
461 | ||
462 | return read_n_cells(n_mem_size_cells, &prop); | |
463 | } | |
464 | ||
465 | struct assoc_arrays { | |
466 | u32 n_arrays; | |
467 | u32 array_sz; | |
468 | const u32 *arrays; | |
469 | }; | |
470 | ||
471 | /* | |
25985edc | 472 | * Retrieve and validate the list of associativity arrays for drconf |
8342681d NF |
473 | * memory from the ibm,associativity-lookup-arrays property of the |
474 | * device tree.. | |
475 | * | |
476 | * The layout of the ibm,associativity-lookup-arrays property is a number N | |
477 | * indicating the number of associativity arrays, followed by a number M | |
478 | * indicating the size of each associativity array, followed by a list | |
479 | * of N associativity arrays. | |
480 | */ | |
481 | static int of_get_assoc_arrays(struct device_node *memory, | |
482 | struct assoc_arrays *aa) | |
483 | { | |
484 | const u32 *prop; | |
485 | u32 len; | |
486 | ||
487 | prop = of_get_property(memory, "ibm,associativity-lookup-arrays", &len); | |
488 | if (!prop || len < 2 * sizeof(unsigned int)) | |
489 | return -1; | |
490 | ||
491 | aa->n_arrays = *prop++; | |
492 | aa->array_sz = *prop++; | |
493 | ||
42b2aa86 | 494 | /* Now that we know the number of arrays and size of each array, |
8342681d NF |
495 | * revalidate the size of the property read in. |
496 | */ | |
497 | if (len < (aa->n_arrays * aa->array_sz + 2) * sizeof(unsigned int)) | |
498 | return -1; | |
499 | ||
500 | aa->arrays = prop; | |
501 | return 0; | |
502 | } | |
503 | ||
504 | /* | |
505 | * This is like of_node_to_nid_single() for memory represented in the | |
506 | * ibm,dynamic-reconfiguration-memory node. | |
507 | */ | |
508 | static int of_drconf_to_nid_single(struct of_drconf_cell *drmem, | |
509 | struct assoc_arrays *aa) | |
510 | { | |
511 | int default_nid = 0; | |
512 | int nid = default_nid; | |
513 | int index; | |
514 | ||
515 | if (min_common_depth > 0 && min_common_depth <= aa->array_sz && | |
516 | !(drmem->flags & DRCONF_MEM_AI_INVALID) && | |
517 | drmem->aa_index < aa->n_arrays) { | |
518 | index = drmem->aa_index * aa->array_sz + min_common_depth - 1; | |
519 | nid = aa->arrays[index]; | |
520 | ||
521 | if (nid == 0xffff || nid >= MAX_NUMNODES) | |
522 | nid = default_nid; | |
523 | } | |
524 | ||
525 | return nid; | |
526 | } | |
527 | ||
1da177e4 LT |
528 | /* |
529 | * Figure out to which domain a cpu belongs and stick it there. | |
530 | * Return the id of the domain used. | |
531 | */ | |
2e5ce39d | 532 | static int __cpuinit numa_setup_cpu(unsigned long lcpu) |
1da177e4 | 533 | { |
cf950b7a | 534 | int nid = 0; |
8b16cd23 | 535 | struct device_node *cpu = of_get_cpu_node(lcpu, NULL); |
1da177e4 LT |
536 | |
537 | if (!cpu) { | |
538 | WARN_ON(1); | |
539 | goto out; | |
540 | } | |
541 | ||
953039c8 | 542 | nid = of_node_to_nid_single(cpu); |
1da177e4 | 543 | |
482ec7c4 | 544 | if (nid < 0 || !node_online(nid)) |
72c33688 | 545 | nid = first_online_node; |
1da177e4 | 546 | out: |
cf950b7a | 547 | map_cpu_to_node(lcpu, nid); |
1da177e4 LT |
548 | |
549 | of_node_put(cpu); | |
550 | ||
cf950b7a | 551 | return nid; |
1da177e4 LT |
552 | } |
553 | ||
74b85f37 | 554 | static int __cpuinit cpu_numa_callback(struct notifier_block *nfb, |
1da177e4 LT |
555 | unsigned long action, |
556 | void *hcpu) | |
557 | { | |
558 | unsigned long lcpu = (unsigned long)hcpu; | |
559 | int ret = NOTIFY_DONE; | |
560 | ||
561 | switch (action) { | |
562 | case CPU_UP_PREPARE: | |
8bb78442 | 563 | case CPU_UP_PREPARE_FROZEN: |
2b261227 | 564 | numa_setup_cpu(lcpu); |
1da177e4 LT |
565 | ret = NOTIFY_OK; |
566 | break; | |
567 | #ifdef CONFIG_HOTPLUG_CPU | |
568 | case CPU_DEAD: | |
8bb78442 | 569 | case CPU_DEAD_FROZEN: |
1da177e4 | 570 | case CPU_UP_CANCELED: |
8bb78442 | 571 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 LT |
572 | unmap_cpu_from_node(lcpu); |
573 | break; | |
574 | ret = NOTIFY_OK; | |
575 | #endif | |
576 | } | |
577 | return ret; | |
578 | } | |
579 | ||
580 | /* | |
581 | * Check and possibly modify a memory region to enforce the memory limit. | |
582 | * | |
583 | * Returns the size the region should have to enforce the memory limit. | |
584 | * This will either be the original value of size, a truncated value, | |
585 | * or zero. If the returned value of size is 0 the region should be | |
25985edc | 586 | * discarded as it lies wholly above the memory limit. |
1da177e4 | 587 | */ |
45fb6cea AB |
588 | static unsigned long __init numa_enforce_memory_limit(unsigned long start, |
589 | unsigned long size) | |
1da177e4 LT |
590 | { |
591 | /* | |
95f72d1e | 592 | * We use memblock_end_of_DRAM() in here instead of memory_limit because |
1da177e4 | 593 | * we've already adjusted it for the limit and it takes care of |
fe55249d MM |
594 | * having memory holes below the limit. Also, in the case of |
595 | * iommu_is_off, memory_limit is not set but is implicitly enforced. | |
1da177e4 | 596 | */ |
1da177e4 | 597 | |
95f72d1e | 598 | if (start + size <= memblock_end_of_DRAM()) |
1da177e4 LT |
599 | return size; |
600 | ||
95f72d1e | 601 | if (start >= memblock_end_of_DRAM()) |
1da177e4 LT |
602 | return 0; |
603 | ||
95f72d1e | 604 | return memblock_end_of_DRAM() - start; |
1da177e4 LT |
605 | } |
606 | ||
cf00085d C |
607 | /* |
608 | * Reads the counter for a given entry in | |
609 | * linux,drconf-usable-memory property | |
610 | */ | |
611 | static inline int __init read_usm_ranges(const u32 **usm) | |
612 | { | |
613 | /* | |
3fdfd990 | 614 | * For each lmb in ibm,dynamic-memory a corresponding |
cf00085d C |
615 | * entry in linux,drconf-usable-memory property contains |
616 | * a counter followed by that many (base, size) duple. | |
617 | * read the counter from linux,drconf-usable-memory | |
618 | */ | |
619 | return read_n_cells(n_mem_size_cells, usm); | |
620 | } | |
621 | ||
0204568a PM |
622 | /* |
623 | * Extract NUMA information from the ibm,dynamic-reconfiguration-memory | |
624 | * node. This assumes n_mem_{addr,size}_cells have been set. | |
625 | */ | |
626 | static void __init parse_drconf_memory(struct device_node *memory) | |
627 | { | |
82b2521d | 628 | const u32 *uninitialized_var(dm), *usm; |
cf00085d | 629 | unsigned int n, rc, ranges, is_kexec_kdump = 0; |
3fdfd990 | 630 | unsigned long lmb_size, base, size, sz; |
8342681d | 631 | int nid; |
aa709f3b | 632 | struct assoc_arrays aa = { .arrays = NULL }; |
8342681d NF |
633 | |
634 | n = of_get_drconf_memory(memory, &dm); | |
635 | if (!n) | |
0204568a PM |
636 | return; |
637 | ||
3fdfd990 BH |
638 | lmb_size = of_get_lmb_size(memory); |
639 | if (!lmb_size) | |
8342681d NF |
640 | return; |
641 | ||
642 | rc = of_get_assoc_arrays(memory, &aa); | |
643 | if (rc) | |
0204568a PM |
644 | return; |
645 | ||
cf00085d C |
646 | /* check if this is a kexec/kdump kernel */ |
647 | usm = of_get_usable_memory(memory); | |
648 | if (usm != NULL) | |
649 | is_kexec_kdump = 1; | |
650 | ||
0204568a | 651 | for (; n != 0; --n) { |
8342681d NF |
652 | struct of_drconf_cell drmem; |
653 | ||
654 | read_drconf_cell(&drmem, &dm); | |
655 | ||
656 | /* skip this block if the reserved bit is set in flags (0x80) | |
657 | or if the block is not assigned to this partition (0x8) */ | |
658 | if ((drmem.flags & DRCONF_MEM_RESERVED) | |
659 | || !(drmem.flags & DRCONF_MEM_ASSIGNED)) | |
0204568a | 660 | continue; |
1daa6d08 | 661 | |
cf00085d | 662 | base = drmem.base_addr; |
3fdfd990 | 663 | size = lmb_size; |
cf00085d | 664 | ranges = 1; |
8342681d | 665 | |
cf00085d C |
666 | if (is_kexec_kdump) { |
667 | ranges = read_usm_ranges(&usm); | |
668 | if (!ranges) /* there are no (base, size) duple */ | |
669 | continue; | |
670 | } | |
671 | do { | |
672 | if (is_kexec_kdump) { | |
673 | base = read_n_cells(n_mem_addr_cells, &usm); | |
674 | size = read_n_cells(n_mem_size_cells, &usm); | |
675 | } | |
676 | nid = of_drconf_to_nid_single(&drmem, &aa); | |
677 | fake_numa_create_new_node( | |
678 | ((base + size) >> PAGE_SHIFT), | |
8342681d | 679 | &nid); |
cf00085d C |
680 | node_set_online(nid); |
681 | sz = numa_enforce_memory_limit(base, size); | |
682 | if (sz) | |
1d7cfe18 | 683 | memblock_set_node(base, sz, nid); |
cf00085d | 684 | } while (--ranges); |
0204568a PM |
685 | } |
686 | } | |
687 | ||
1da177e4 LT |
688 | static int __init parse_numa_properties(void) |
689 | { | |
94db7c5e | 690 | struct device_node *memory; |
482ec7c4 | 691 | int default_nid = 0; |
1da177e4 LT |
692 | unsigned long i; |
693 | ||
694 | if (numa_enabled == 0) { | |
695 | printk(KERN_WARNING "NUMA disabled by user\n"); | |
696 | return -1; | |
697 | } | |
698 | ||
1da177e4 LT |
699 | min_common_depth = find_min_common_depth(); |
700 | ||
1da177e4 LT |
701 | if (min_common_depth < 0) |
702 | return min_common_depth; | |
703 | ||
bf4b85b0 NL |
704 | dbg("NUMA associativity depth for CPU/Memory: %d\n", min_common_depth); |
705 | ||
1da177e4 | 706 | /* |
482ec7c4 NL |
707 | * Even though we connect cpus to numa domains later in SMP |
708 | * init, we need to know the node ids now. This is because | |
709 | * each node to be onlined must have NODE_DATA etc backing it. | |
1da177e4 | 710 | */ |
482ec7c4 | 711 | for_each_present_cpu(i) { |
dfbe93a2 | 712 | struct device_node *cpu; |
cf950b7a | 713 | int nid; |
1da177e4 | 714 | |
8b16cd23 | 715 | cpu = of_get_cpu_node(i, NULL); |
482ec7c4 | 716 | BUG_ON(!cpu); |
953039c8 | 717 | nid = of_node_to_nid_single(cpu); |
482ec7c4 | 718 | of_node_put(cpu); |
1da177e4 | 719 | |
482ec7c4 NL |
720 | /* |
721 | * Don't fall back to default_nid yet -- we will plug | |
722 | * cpus into nodes once the memory scan has discovered | |
723 | * the topology. | |
724 | */ | |
725 | if (nid < 0) | |
726 | continue; | |
727 | node_set_online(nid); | |
1da177e4 LT |
728 | } |
729 | ||
237a0989 | 730 | get_n_mem_cells(&n_mem_addr_cells, &n_mem_size_cells); |
94db7c5e AB |
731 | |
732 | for_each_node_by_type(memory, "memory") { | |
1da177e4 LT |
733 | unsigned long start; |
734 | unsigned long size; | |
cf950b7a | 735 | int nid; |
1da177e4 | 736 | int ranges; |
a7f67bdf | 737 | const unsigned int *memcell_buf; |
1da177e4 LT |
738 | unsigned int len; |
739 | ||
e2eb6392 | 740 | memcell_buf = of_get_property(memory, |
ba759485 ME |
741 | "linux,usable-memory", &len); |
742 | if (!memcell_buf || len <= 0) | |
e2eb6392 | 743 | memcell_buf = of_get_property(memory, "reg", &len); |
1da177e4 LT |
744 | if (!memcell_buf || len <= 0) |
745 | continue; | |
746 | ||
cc5d0189 BH |
747 | /* ranges in cell */ |
748 | ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells); | |
1da177e4 LT |
749 | new_range: |
750 | /* these are order-sensitive, and modify the buffer pointer */ | |
237a0989 MK |
751 | start = read_n_cells(n_mem_addr_cells, &memcell_buf); |
752 | size = read_n_cells(n_mem_size_cells, &memcell_buf); | |
1da177e4 | 753 | |
482ec7c4 NL |
754 | /* |
755 | * Assumption: either all memory nodes or none will | |
756 | * have associativity properties. If none, then | |
757 | * everything goes to default_nid. | |
758 | */ | |
953039c8 | 759 | nid = of_node_to_nid_single(memory); |
482ec7c4 NL |
760 | if (nid < 0) |
761 | nid = default_nid; | |
1daa6d08 BS |
762 | |
763 | fake_numa_create_new_node(((start + size) >> PAGE_SHIFT), &nid); | |
482ec7c4 | 764 | node_set_online(nid); |
1da177e4 | 765 | |
45fb6cea | 766 | if (!(size = numa_enforce_memory_limit(start, size))) { |
1da177e4 LT |
767 | if (--ranges) |
768 | goto new_range; | |
769 | else | |
770 | continue; | |
771 | } | |
772 | ||
1d7cfe18 | 773 | memblock_set_node(start, size, nid); |
1da177e4 LT |
774 | |
775 | if (--ranges) | |
776 | goto new_range; | |
777 | } | |
778 | ||
0204568a | 779 | /* |
dfbe93a2 AB |
780 | * Now do the same thing for each MEMBLOCK listed in the |
781 | * ibm,dynamic-memory property in the | |
782 | * ibm,dynamic-reconfiguration-memory node. | |
0204568a PM |
783 | */ |
784 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); | |
785 | if (memory) | |
786 | parse_drconf_memory(memory); | |
787 | ||
1da177e4 LT |
788 | return 0; |
789 | } | |
790 | ||
791 | static void __init setup_nonnuma(void) | |
792 | { | |
95f72d1e YL |
793 | unsigned long top_of_ram = memblock_end_of_DRAM(); |
794 | unsigned long total_ram = memblock_phys_mem_size(); | |
c67c3cb4 | 795 | unsigned long start_pfn, end_pfn; |
28be7072 BH |
796 | unsigned int nid = 0; |
797 | struct memblock_region *reg; | |
1da177e4 | 798 | |
e110b281 | 799 | printk(KERN_DEBUG "Top of RAM: 0x%lx, Total RAM: 0x%lx\n", |
1da177e4 | 800 | top_of_ram, total_ram); |
e110b281 | 801 | printk(KERN_DEBUG "Memory hole size: %ldMB\n", |
1da177e4 LT |
802 | (top_of_ram - total_ram) >> 20); |
803 | ||
28be7072 | 804 | for_each_memblock(memory, reg) { |
c7fc2de0 YL |
805 | start_pfn = memblock_region_memory_base_pfn(reg); |
806 | end_pfn = memblock_region_memory_end_pfn(reg); | |
1daa6d08 BS |
807 | |
808 | fake_numa_create_new_node(end_pfn, &nid); | |
1d7cfe18 TH |
809 | memblock_set_node(PFN_PHYS(start_pfn), |
810 | PFN_PHYS(end_pfn - start_pfn), nid); | |
1daa6d08 | 811 | node_set_online(nid); |
c67c3cb4 | 812 | } |
1da177e4 LT |
813 | } |
814 | ||
4b703a23 AB |
815 | void __init dump_numa_cpu_topology(void) |
816 | { | |
817 | unsigned int node; | |
818 | unsigned int cpu, count; | |
819 | ||
820 | if (min_common_depth == -1 || !numa_enabled) | |
821 | return; | |
822 | ||
823 | for_each_online_node(node) { | |
e110b281 | 824 | printk(KERN_DEBUG "Node %d CPUs:", node); |
4b703a23 AB |
825 | |
826 | count = 0; | |
827 | /* | |
828 | * If we used a CPU iterator here we would miss printing | |
829 | * the holes in the cpumap. | |
830 | */ | |
25863de0 AB |
831 | for (cpu = 0; cpu < nr_cpu_ids; cpu++) { |
832 | if (cpumask_test_cpu(cpu, | |
833 | node_to_cpumask_map[node])) { | |
4b703a23 AB |
834 | if (count == 0) |
835 | printk(" %u", cpu); | |
836 | ++count; | |
837 | } else { | |
838 | if (count > 1) | |
839 | printk("-%u", cpu - 1); | |
840 | count = 0; | |
841 | } | |
842 | } | |
843 | ||
844 | if (count > 1) | |
25863de0 | 845 | printk("-%u", nr_cpu_ids - 1); |
4b703a23 AB |
846 | printk("\n"); |
847 | } | |
848 | } | |
849 | ||
850 | static void __init dump_numa_memory_topology(void) | |
1da177e4 LT |
851 | { |
852 | unsigned int node; | |
853 | unsigned int count; | |
854 | ||
855 | if (min_common_depth == -1 || !numa_enabled) | |
856 | return; | |
857 | ||
858 | for_each_online_node(node) { | |
859 | unsigned long i; | |
860 | ||
e110b281 | 861 | printk(KERN_DEBUG "Node %d Memory:", node); |
1da177e4 LT |
862 | |
863 | count = 0; | |
864 | ||
95f72d1e | 865 | for (i = 0; i < memblock_end_of_DRAM(); |
45fb6cea AB |
866 | i += (1 << SECTION_SIZE_BITS)) { |
867 | if (early_pfn_to_nid(i >> PAGE_SHIFT) == node) { | |
1da177e4 LT |
868 | if (count == 0) |
869 | printk(" 0x%lx", i); | |
870 | ++count; | |
871 | } else { | |
872 | if (count > 0) | |
873 | printk("-0x%lx", i); | |
874 | count = 0; | |
875 | } | |
876 | } | |
877 | ||
878 | if (count > 0) | |
879 | printk("-0x%lx", i); | |
880 | printk("\n"); | |
881 | } | |
1da177e4 LT |
882 | } |
883 | ||
884 | /* | |
95f72d1e | 885 | * Allocate some memory, satisfying the memblock or bootmem allocator where |
1da177e4 LT |
886 | * required. nid is the preferred node and end is the physical address of |
887 | * the highest address in the node. | |
888 | * | |
0be210fd | 889 | * Returns the virtual address of the memory. |
1da177e4 | 890 | */ |
893473df | 891 | static void __init *careful_zallocation(int nid, unsigned long size, |
45fb6cea AB |
892 | unsigned long align, |
893 | unsigned long end_pfn) | |
1da177e4 | 894 | { |
0be210fd | 895 | void *ret; |
45fb6cea | 896 | int new_nid; |
0be210fd DH |
897 | unsigned long ret_paddr; |
898 | ||
95f72d1e | 899 | ret_paddr = __memblock_alloc_base(size, align, end_pfn << PAGE_SHIFT); |
1da177e4 LT |
900 | |
901 | /* retry over all memory */ | |
0be210fd | 902 | if (!ret_paddr) |
95f72d1e | 903 | ret_paddr = __memblock_alloc_base(size, align, memblock_end_of_DRAM()); |
1da177e4 | 904 | |
0be210fd | 905 | if (!ret_paddr) |
5d21ea2b | 906 | panic("numa.c: cannot allocate %lu bytes for node %d", |
1da177e4 LT |
907 | size, nid); |
908 | ||
0be210fd DH |
909 | ret = __va(ret_paddr); |
910 | ||
1da177e4 | 911 | /* |
c555e520 | 912 | * We initialize the nodes in numeric order: 0, 1, 2... |
95f72d1e | 913 | * and hand over control from the MEMBLOCK allocator to the |
c555e520 DH |
914 | * bootmem allocator. If this function is called for |
915 | * node 5, then we know that all nodes <5 are using the | |
95f72d1e | 916 | * bootmem allocator instead of the MEMBLOCK allocator. |
c555e520 DH |
917 | * |
918 | * So, check the nid from which this allocation came | |
919 | * and double check to see if we need to use bootmem | |
95f72d1e | 920 | * instead of the MEMBLOCK. We don't free the MEMBLOCK memory |
c555e520 | 921 | * since it would be useless. |
1da177e4 | 922 | */ |
0be210fd | 923 | new_nid = early_pfn_to_nid(ret_paddr >> PAGE_SHIFT); |
45fb6cea | 924 | if (new_nid < nid) { |
0be210fd | 925 | ret = __alloc_bootmem_node(NODE_DATA(new_nid), |
1da177e4 LT |
926 | size, align, 0); |
927 | ||
0be210fd | 928 | dbg("alloc_bootmem %p %lx\n", ret, size); |
1da177e4 LT |
929 | } |
930 | ||
893473df | 931 | memset(ret, 0, size); |
0be210fd | 932 | return ret; |
1da177e4 LT |
933 | } |
934 | ||
74b85f37 CS |
935 | static struct notifier_block __cpuinitdata ppc64_numa_nb = { |
936 | .notifier_call = cpu_numa_callback, | |
937 | .priority = 1 /* Must run before sched domains notifier. */ | |
938 | }; | |
939 | ||
28e86bdb | 940 | static void __init mark_reserved_regions_for_nid(int nid) |
4a618669 DH |
941 | { |
942 | struct pglist_data *node = NODE_DATA(nid); | |
28be7072 | 943 | struct memblock_region *reg; |
4a618669 | 944 | |
28be7072 BH |
945 | for_each_memblock(reserved, reg) { |
946 | unsigned long physbase = reg->base; | |
947 | unsigned long size = reg->size; | |
4a618669 | 948 | unsigned long start_pfn = physbase >> PAGE_SHIFT; |
06eccea6 | 949 | unsigned long end_pfn = PFN_UP(physbase + size); |
4a618669 DH |
950 | struct node_active_region node_ar; |
951 | unsigned long node_end_pfn = node->node_start_pfn + | |
952 | node->node_spanned_pages; | |
953 | ||
954 | /* | |
95f72d1e | 955 | * Check to make sure that this memblock.reserved area is |
4a618669 DH |
956 | * within the bounds of the node that we care about. |
957 | * Checking the nid of the start and end points is not | |
958 | * sufficient because the reserved area could span the | |
959 | * entire node. | |
960 | */ | |
961 | if (end_pfn <= node->node_start_pfn || | |
962 | start_pfn >= node_end_pfn) | |
963 | continue; | |
964 | ||
965 | get_node_active_region(start_pfn, &node_ar); | |
966 | while (start_pfn < end_pfn && | |
967 | node_ar.start_pfn < node_ar.end_pfn) { | |
968 | unsigned long reserve_size = size; | |
969 | /* | |
970 | * if reserved region extends past active region | |
971 | * then trim size to active region | |
972 | */ | |
973 | if (end_pfn > node_ar.end_pfn) | |
974 | reserve_size = (node_ar.end_pfn << PAGE_SHIFT) | |
06eccea6 | 975 | - physbase; |
a4c74ddd DH |
976 | /* |
977 | * Only worry about *this* node, others may not | |
978 | * yet have valid NODE_DATA(). | |
979 | */ | |
980 | if (node_ar.nid == nid) { | |
981 | dbg("reserve_bootmem %lx %lx nid=%d\n", | |
982 | physbase, reserve_size, node_ar.nid); | |
983 | reserve_bootmem_node(NODE_DATA(node_ar.nid), | |
984 | physbase, reserve_size, | |
985 | BOOTMEM_DEFAULT); | |
986 | } | |
4a618669 DH |
987 | /* |
988 | * if reserved region is contained in the active region | |
989 | * then done. | |
990 | */ | |
991 | if (end_pfn <= node_ar.end_pfn) | |
992 | break; | |
993 | ||
994 | /* | |
995 | * reserved region extends past the active region | |
996 | * get next active region that contains this | |
997 | * reserved region | |
998 | */ | |
999 | start_pfn = node_ar.end_pfn; | |
1000 | physbase = start_pfn << PAGE_SHIFT; | |
1001 | size = size - reserve_size; | |
1002 | get_node_active_region(start_pfn, &node_ar); | |
1003 | } | |
1004 | } | |
1005 | } | |
1006 | ||
1007 | ||
1da177e4 LT |
1008 | void __init do_init_bootmem(void) |
1009 | { | |
1010 | int nid; | |
1da177e4 LT |
1011 | |
1012 | min_low_pfn = 0; | |
95f72d1e | 1013 | max_low_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT; |
1da177e4 LT |
1014 | max_pfn = max_low_pfn; |
1015 | ||
1016 | if (parse_numa_properties()) | |
1017 | setup_nonnuma(); | |
1018 | else | |
4b703a23 | 1019 | dump_numa_memory_topology(); |
1da177e4 | 1020 | |
1da177e4 | 1021 | for_each_online_node(nid) { |
c67c3cb4 | 1022 | unsigned long start_pfn, end_pfn; |
0be210fd | 1023 | void *bootmem_vaddr; |
1da177e4 LT |
1024 | unsigned long bootmap_pages; |
1025 | ||
c67c3cb4 | 1026 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); |
1da177e4 | 1027 | |
4a618669 DH |
1028 | /* |
1029 | * Allocate the node structure node local if possible | |
1030 | * | |
1031 | * Be careful moving this around, as it relies on all | |
1032 | * previous nodes' bootmem to be initialized and have | |
1033 | * all reserved areas marked. | |
1034 | */ | |
893473df | 1035 | NODE_DATA(nid) = careful_zallocation(nid, |
1da177e4 | 1036 | sizeof(struct pglist_data), |
45fb6cea | 1037 | SMP_CACHE_BYTES, end_pfn); |
1da177e4 LT |
1038 | |
1039 | dbg("node %d\n", nid); | |
1040 | dbg("NODE_DATA() = %p\n", NODE_DATA(nid)); | |
1041 | ||
b61bfa3c | 1042 | NODE_DATA(nid)->bdata = &bootmem_node_data[nid]; |
45fb6cea AB |
1043 | NODE_DATA(nid)->node_start_pfn = start_pfn; |
1044 | NODE_DATA(nid)->node_spanned_pages = end_pfn - start_pfn; | |
1da177e4 LT |
1045 | |
1046 | if (NODE_DATA(nid)->node_spanned_pages == 0) | |
1047 | continue; | |
1048 | ||
45fb6cea AB |
1049 | dbg("start_paddr = %lx\n", start_pfn << PAGE_SHIFT); |
1050 | dbg("end_paddr = %lx\n", end_pfn << PAGE_SHIFT); | |
1da177e4 | 1051 | |
45fb6cea | 1052 | bootmap_pages = bootmem_bootmap_pages(end_pfn - start_pfn); |
893473df | 1053 | bootmem_vaddr = careful_zallocation(nid, |
45fb6cea AB |
1054 | bootmap_pages << PAGE_SHIFT, |
1055 | PAGE_SIZE, end_pfn); | |
1da177e4 | 1056 | |
0be210fd | 1057 | dbg("bootmap_vaddr = %p\n", bootmem_vaddr); |
1da177e4 | 1058 | |
0be210fd DH |
1059 | init_bootmem_node(NODE_DATA(nid), |
1060 | __pa(bootmem_vaddr) >> PAGE_SHIFT, | |
45fb6cea | 1061 | start_pfn, end_pfn); |
1da177e4 | 1062 | |
c67c3cb4 | 1063 | free_bootmem_with_active_regions(nid, end_pfn); |
4a618669 DH |
1064 | /* |
1065 | * Be very careful about moving this around. Future | |
893473df | 1066 | * calls to careful_zallocation() depend on this getting |
4a618669 DH |
1067 | * done correctly. |
1068 | */ | |
1069 | mark_reserved_regions_for_nid(nid); | |
8f64e1f2 | 1070 | sparse_memory_present_with_active_regions(nid); |
4a618669 | 1071 | } |
d3f6204a BH |
1072 | |
1073 | init_bootmem_done = 1; | |
25863de0 AB |
1074 | |
1075 | /* | |
1076 | * Now bootmem is initialised we can create the node to cpumask | |
1077 | * lookup tables and setup the cpu callback to populate them. | |
1078 | */ | |
1079 | setup_node_to_cpumask_map(); | |
1080 | ||
1081 | register_cpu_notifier(&ppc64_numa_nb); | |
1082 | cpu_numa_callback(&ppc64_numa_nb, CPU_UP_PREPARE, | |
1083 | (void *)(unsigned long)boot_cpuid); | |
1da177e4 LT |
1084 | } |
1085 | ||
1086 | void __init paging_init(void) | |
1087 | { | |
6391af17 MG |
1088 | unsigned long max_zone_pfns[MAX_NR_ZONES]; |
1089 | memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); | |
95f72d1e | 1090 | max_zone_pfns[ZONE_DMA] = memblock_end_of_DRAM() >> PAGE_SHIFT; |
c67c3cb4 | 1091 | free_area_init_nodes(max_zone_pfns); |
1da177e4 LT |
1092 | } |
1093 | ||
1094 | static int __init early_numa(char *p) | |
1095 | { | |
1096 | if (!p) | |
1097 | return 0; | |
1098 | ||
1099 | if (strstr(p, "off")) | |
1100 | numa_enabled = 0; | |
1101 | ||
1102 | if (strstr(p, "debug")) | |
1103 | numa_debug = 1; | |
1104 | ||
1daa6d08 BS |
1105 | p = strstr(p, "fake="); |
1106 | if (p) | |
1107 | cmdline = p + strlen("fake="); | |
1108 | ||
1da177e4 LT |
1109 | return 0; |
1110 | } | |
1111 | early_param("numa", early_numa); | |
237a0989 MK |
1112 | |
1113 | #ifdef CONFIG_MEMORY_HOTPLUG | |
0db9360a | 1114 | /* |
0f16ef7f NF |
1115 | * Find the node associated with a hot added memory section for |
1116 | * memory represented in the device tree by the property | |
1117 | * ibm,dynamic-reconfiguration-memory/ibm,dynamic-memory. | |
0db9360a NF |
1118 | */ |
1119 | static int hot_add_drconf_scn_to_nid(struct device_node *memory, | |
1120 | unsigned long scn_addr) | |
1121 | { | |
1122 | const u32 *dm; | |
0f16ef7f | 1123 | unsigned int drconf_cell_cnt, rc; |
3fdfd990 | 1124 | unsigned long lmb_size; |
0db9360a | 1125 | struct assoc_arrays aa; |
0f16ef7f | 1126 | int nid = -1; |
0db9360a | 1127 | |
0f16ef7f NF |
1128 | drconf_cell_cnt = of_get_drconf_memory(memory, &dm); |
1129 | if (!drconf_cell_cnt) | |
1130 | return -1; | |
0db9360a | 1131 | |
3fdfd990 BH |
1132 | lmb_size = of_get_lmb_size(memory); |
1133 | if (!lmb_size) | |
0f16ef7f | 1134 | return -1; |
0db9360a NF |
1135 | |
1136 | rc = of_get_assoc_arrays(memory, &aa); | |
1137 | if (rc) | |
0f16ef7f | 1138 | return -1; |
0db9360a | 1139 | |
0f16ef7f | 1140 | for (; drconf_cell_cnt != 0; --drconf_cell_cnt) { |
0db9360a NF |
1141 | struct of_drconf_cell drmem; |
1142 | ||
1143 | read_drconf_cell(&drmem, &dm); | |
1144 | ||
1145 | /* skip this block if it is reserved or not assigned to | |
1146 | * this partition */ | |
1147 | if ((drmem.flags & DRCONF_MEM_RESERVED) | |
1148 | || !(drmem.flags & DRCONF_MEM_ASSIGNED)) | |
1149 | continue; | |
1150 | ||
0f16ef7f | 1151 | if ((scn_addr < drmem.base_addr) |
3fdfd990 | 1152 | || (scn_addr >= (drmem.base_addr + lmb_size))) |
0f16ef7f NF |
1153 | continue; |
1154 | ||
0db9360a | 1155 | nid = of_drconf_to_nid_single(&drmem, &aa); |
0f16ef7f NF |
1156 | break; |
1157 | } | |
1158 | ||
1159 | return nid; | |
1160 | } | |
1161 | ||
1162 | /* | |
1163 | * Find the node associated with a hot added memory section for memory | |
1164 | * represented in the device tree as a node (i.e. memory@XXXX) for | |
95f72d1e | 1165 | * each memblock. |
0f16ef7f NF |
1166 | */ |
1167 | int hot_add_node_scn_to_nid(unsigned long scn_addr) | |
1168 | { | |
94db7c5e | 1169 | struct device_node *memory; |
0f16ef7f NF |
1170 | int nid = -1; |
1171 | ||
94db7c5e | 1172 | for_each_node_by_type(memory, "memory") { |
0f16ef7f NF |
1173 | unsigned long start, size; |
1174 | int ranges; | |
1175 | const unsigned int *memcell_buf; | |
1176 | unsigned int len; | |
1177 | ||
1178 | memcell_buf = of_get_property(memory, "reg", &len); | |
1179 | if (!memcell_buf || len <= 0) | |
1180 | continue; | |
1181 | ||
1182 | /* ranges in cell */ | |
1183 | ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells); | |
1184 | ||
1185 | while (ranges--) { | |
1186 | start = read_n_cells(n_mem_addr_cells, &memcell_buf); | |
1187 | size = read_n_cells(n_mem_size_cells, &memcell_buf); | |
1188 | ||
1189 | if ((scn_addr < start) || (scn_addr >= (start + size))) | |
1190 | continue; | |
1191 | ||
1192 | nid = of_node_to_nid_single(memory); | |
1193 | break; | |
1194 | } | |
0db9360a | 1195 | |
0f16ef7f NF |
1196 | if (nid >= 0) |
1197 | break; | |
0db9360a NF |
1198 | } |
1199 | ||
60831842 AB |
1200 | of_node_put(memory); |
1201 | ||
0f16ef7f | 1202 | return nid; |
0db9360a NF |
1203 | } |
1204 | ||
237a0989 MK |
1205 | /* |
1206 | * Find the node associated with a hot added memory section. Section | |
95f72d1e YL |
1207 | * corresponds to a SPARSEMEM section, not an MEMBLOCK. It is assumed that |
1208 | * sections are fully contained within a single MEMBLOCK. | |
237a0989 MK |
1209 | */ |
1210 | int hot_add_scn_to_nid(unsigned long scn_addr) | |
1211 | { | |
1212 | struct device_node *memory = NULL; | |
0f16ef7f | 1213 | int nid, found = 0; |
237a0989 MK |
1214 | |
1215 | if (!numa_enabled || (min_common_depth < 0)) | |
72c33688 | 1216 | return first_online_node; |
0db9360a NF |
1217 | |
1218 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); | |
1219 | if (memory) { | |
1220 | nid = hot_add_drconf_scn_to_nid(memory, scn_addr); | |
1221 | of_node_put(memory); | |
0f16ef7f NF |
1222 | } else { |
1223 | nid = hot_add_node_scn_to_nid(scn_addr); | |
0db9360a | 1224 | } |
237a0989 | 1225 | |
0f16ef7f | 1226 | if (nid < 0 || !node_online(nid)) |
72c33688 | 1227 | nid = first_online_node; |
237a0989 | 1228 | |
0f16ef7f NF |
1229 | if (NODE_DATA(nid)->node_spanned_pages) |
1230 | return nid; | |
237a0989 | 1231 | |
0f16ef7f NF |
1232 | for_each_online_node(nid) { |
1233 | if (NODE_DATA(nid)->node_spanned_pages) { | |
1234 | found = 1; | |
1235 | break; | |
237a0989 | 1236 | } |
237a0989 | 1237 | } |
0f16ef7f NF |
1238 | |
1239 | BUG_ON(!found); | |
1240 | return nid; | |
237a0989 | 1241 | } |
0f16ef7f | 1242 | |
cd34206e NA |
1243 | static u64 hot_add_drconf_memory_max(void) |
1244 | { | |
1245 | struct device_node *memory = NULL; | |
1246 | unsigned int drconf_cell_cnt = 0; | |
1247 | u64 lmb_size = 0; | |
1248 | const u32 *dm = 0; | |
1249 | ||
1250 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); | |
1251 | if (memory) { | |
1252 | drconf_cell_cnt = of_get_drconf_memory(memory, &dm); | |
1253 | lmb_size = of_get_lmb_size(memory); | |
1254 | of_node_put(memory); | |
1255 | } | |
1256 | return lmb_size * drconf_cell_cnt; | |
1257 | } | |
1258 | ||
1259 | /* | |
1260 | * memory_hotplug_max - return max address of memory that may be added | |
1261 | * | |
1262 | * This is currently only used on systems that support drconfig memory | |
1263 | * hotplug. | |
1264 | */ | |
1265 | u64 memory_hotplug_max(void) | |
1266 | { | |
1267 | return max(hot_add_drconf_memory_max(), memblock_end_of_DRAM()); | |
1268 | } | |
237a0989 | 1269 | #endif /* CONFIG_MEMORY_HOTPLUG */ |
9eff1a38 | 1270 | |
bd03403a | 1271 | /* Virtual Processor Home Node (VPHN) support */ |
39bf990e | 1272 | #ifdef CONFIG_PPC_SPLPAR |
5de16699 | 1273 | static u8 vphn_cpu_change_counts[NR_CPUS][MAX_DISTANCE_REF_POINTS]; |
9eff1a38 JL |
1274 | static cpumask_t cpu_associativity_changes_mask; |
1275 | static int vphn_enabled; | |
1276 | static void set_topology_timer(void); | |
9eff1a38 JL |
1277 | |
1278 | /* | |
1279 | * Store the current values of the associativity change counters in the | |
1280 | * hypervisor. | |
1281 | */ | |
1282 | static void setup_cpu_associativity_change_counters(void) | |
1283 | { | |
cd9d6cc7 | 1284 | int cpu; |
9eff1a38 | 1285 | |
5de16699 AB |
1286 | /* The VPHN feature supports a maximum of 8 reference points */ |
1287 | BUILD_BUG_ON(MAX_DISTANCE_REF_POINTS > 8); | |
1288 | ||
9eff1a38 | 1289 | for_each_possible_cpu(cpu) { |
cd9d6cc7 | 1290 | int i; |
9eff1a38 JL |
1291 | u8 *counts = vphn_cpu_change_counts[cpu]; |
1292 | volatile u8 *hypervisor_counts = lppaca[cpu].vphn_assoc_counts; | |
1293 | ||
5de16699 | 1294 | for (i = 0; i < distance_ref_points_depth; i++) |
9eff1a38 | 1295 | counts[i] = hypervisor_counts[i]; |
9eff1a38 JL |
1296 | } |
1297 | } | |
1298 | ||
1299 | /* | |
1300 | * The hypervisor maintains a set of 8 associativity change counters in | |
1301 | * the VPA of each cpu that correspond to the associativity levels in the | |
1302 | * ibm,associativity-reference-points property. When an associativity | |
1303 | * level changes, the corresponding counter is incremented. | |
1304 | * | |
1305 | * Set a bit in cpu_associativity_changes_mask for each cpu whose home | |
1306 | * node associativity levels have changed. | |
1307 | * | |
1308 | * Returns the number of cpus with unhandled associativity changes. | |
1309 | */ | |
1310 | static int update_cpu_associativity_changes_mask(void) | |
1311 | { | |
cd9d6cc7 | 1312 | int cpu, nr_cpus = 0; |
9eff1a38 JL |
1313 | cpumask_t *changes = &cpu_associativity_changes_mask; |
1314 | ||
1315 | cpumask_clear(changes); | |
1316 | ||
1317 | for_each_possible_cpu(cpu) { | |
1318 | int i, changed = 0; | |
1319 | u8 *counts = vphn_cpu_change_counts[cpu]; | |
1320 | volatile u8 *hypervisor_counts = lppaca[cpu].vphn_assoc_counts; | |
1321 | ||
5de16699 | 1322 | for (i = 0; i < distance_ref_points_depth; i++) { |
d69043e8 | 1323 | if (hypervisor_counts[i] != counts[i]) { |
9eff1a38 JL |
1324 | counts[i] = hypervisor_counts[i]; |
1325 | changed = 1; | |
1326 | } | |
1327 | } | |
1328 | if (changed) { | |
1329 | cpumask_set_cpu(cpu, changes); | |
1330 | nr_cpus++; | |
1331 | } | |
1332 | } | |
1333 | ||
1334 | return nr_cpus; | |
1335 | } | |
1336 | ||
c0e5e46f AB |
1337 | /* |
1338 | * 6 64-bit registers unpacked into 12 32-bit associativity values. To form | |
1339 | * the complete property we have to add the length in the first cell. | |
1340 | */ | |
1341 | #define VPHN_ASSOC_BUFSIZE (6*sizeof(u64)/sizeof(u32) + 1) | |
9eff1a38 JL |
1342 | |
1343 | /* | |
1344 | * Convert the associativity domain numbers returned from the hypervisor | |
1345 | * to the sequence they would appear in the ibm,associativity property. | |
1346 | */ | |
1347 | static int vphn_unpack_associativity(const long *packed, unsigned int *unpacked) | |
1348 | { | |
cd9d6cc7 | 1349 | int i, nr_assoc_doms = 0; |
9eff1a38 JL |
1350 | const u16 *field = (const u16*) packed; |
1351 | ||
1352 | #define VPHN_FIELD_UNUSED (0xffff) | |
1353 | #define VPHN_FIELD_MSB (0x8000) | |
1354 | #define VPHN_FIELD_MASK (~VPHN_FIELD_MSB) | |
1355 | ||
c0e5e46f | 1356 | for (i = 1; i < VPHN_ASSOC_BUFSIZE; i++) { |
9eff1a38 JL |
1357 | if (*field == VPHN_FIELD_UNUSED) { |
1358 | /* All significant fields processed, and remaining | |
1359 | * fields contain the reserved value of all 1's. | |
1360 | * Just store them. | |
1361 | */ | |
1362 | unpacked[i] = *((u32*)field); | |
1363 | field += 2; | |
7639adaa | 1364 | } else if (*field & VPHN_FIELD_MSB) { |
9eff1a38 JL |
1365 | /* Data is in the lower 15 bits of this field */ |
1366 | unpacked[i] = *field & VPHN_FIELD_MASK; | |
1367 | field++; | |
1368 | nr_assoc_doms++; | |
7639adaa | 1369 | } else { |
9eff1a38 JL |
1370 | /* Data is in the lower 15 bits of this field |
1371 | * concatenated with the next 16 bit field | |
1372 | */ | |
1373 | unpacked[i] = *((u32*)field); | |
1374 | field += 2; | |
1375 | nr_assoc_doms++; | |
1376 | } | |
1377 | } | |
1378 | ||
c0e5e46f AB |
1379 | /* The first cell contains the length of the property */ |
1380 | unpacked[0] = nr_assoc_doms; | |
1381 | ||
9eff1a38 JL |
1382 | return nr_assoc_doms; |
1383 | } | |
1384 | ||
1385 | /* | |
1386 | * Retrieve the new associativity information for a virtual processor's | |
1387 | * home node. | |
1388 | */ | |
1389 | static long hcall_vphn(unsigned long cpu, unsigned int *associativity) | |
1390 | { | |
cd9d6cc7 | 1391 | long rc; |
9eff1a38 JL |
1392 | long retbuf[PLPAR_HCALL9_BUFSIZE] = {0}; |
1393 | u64 flags = 1; | |
1394 | int hwcpu = get_hard_smp_processor_id(cpu); | |
1395 | ||
1396 | rc = plpar_hcall9(H_HOME_NODE_ASSOCIATIVITY, retbuf, flags, hwcpu); | |
1397 | vphn_unpack_associativity(retbuf, associativity); | |
1398 | ||
1399 | return rc; | |
1400 | } | |
1401 | ||
1402 | static long vphn_get_associativity(unsigned long cpu, | |
1403 | unsigned int *associativity) | |
1404 | { | |
cd9d6cc7 | 1405 | long rc; |
9eff1a38 JL |
1406 | |
1407 | rc = hcall_vphn(cpu, associativity); | |
1408 | ||
1409 | switch (rc) { | |
1410 | case H_FUNCTION: | |
1411 | printk(KERN_INFO | |
1412 | "VPHN is not supported. Disabling polling...\n"); | |
1413 | stop_topology_update(); | |
1414 | break; | |
1415 | case H_HARDWARE: | |
1416 | printk(KERN_ERR | |
1417 | "hcall_vphn() experienced a hardware fault " | |
1418 | "preventing VPHN. Disabling polling...\n"); | |
1419 | stop_topology_update(); | |
1420 | } | |
1421 | ||
1422 | return rc; | |
1423 | } | |
1424 | ||
1425 | /* | |
1426 | * Update the node maps and sysfs entries for each cpu whose home node | |
79c5fceb | 1427 | * has changed. Returns 1 when the topology has changed, and 0 otherwise. |
9eff1a38 JL |
1428 | */ |
1429 | int arch_update_cpu_topology(void) | |
1430 | { | |
79c5fceb | 1431 | int cpu, nid, old_nid, changed = 0; |
9eff1a38 | 1432 | unsigned int associativity[VPHN_ASSOC_BUFSIZE] = {0}; |
8a25a2fd | 1433 | struct device *dev; |
9eff1a38 | 1434 | |
104699c0 | 1435 | for_each_cpu(cpu,&cpu_associativity_changes_mask) { |
9eff1a38 JL |
1436 | vphn_get_associativity(cpu, associativity); |
1437 | nid = associativity_to_nid(associativity); | |
1438 | ||
1439 | if (nid < 0 || !node_online(nid)) | |
1440 | nid = first_online_node; | |
1441 | ||
1442 | old_nid = numa_cpu_lookup_table[cpu]; | |
1443 | ||
1444 | /* Disable hotplug while we update the cpu | |
1445 | * masks and sysfs. | |
1446 | */ | |
1447 | get_online_cpus(); | |
1448 | unregister_cpu_under_node(cpu, old_nid); | |
1449 | unmap_cpu_from_node(cpu); | |
1450 | map_cpu_to_node(cpu, nid); | |
1451 | register_cpu_under_node(cpu, nid); | |
1452 | put_online_cpus(); | |
1453 | ||
8a25a2fd KS |
1454 | dev = get_cpu_device(cpu); |
1455 | if (dev) | |
1456 | kobject_uevent(&dev->kobj, KOBJ_CHANGE); | |
79c5fceb | 1457 | changed = 1; |
9eff1a38 JL |
1458 | } |
1459 | ||
79c5fceb | 1460 | return changed; |
9eff1a38 JL |
1461 | } |
1462 | ||
1463 | static void topology_work_fn(struct work_struct *work) | |
1464 | { | |
1465 | rebuild_sched_domains(); | |
1466 | } | |
1467 | static DECLARE_WORK(topology_work, topology_work_fn); | |
1468 | ||
1469 | void topology_schedule_update(void) | |
1470 | { | |
1471 | schedule_work(&topology_work); | |
1472 | } | |
1473 | ||
1474 | static void topology_timer_fn(unsigned long ignored) | |
1475 | { | |
1476 | if (!vphn_enabled) | |
1477 | return; | |
1478 | if (update_cpu_associativity_changes_mask() > 0) | |
1479 | topology_schedule_update(); | |
1480 | set_topology_timer(); | |
1481 | } | |
1482 | static struct timer_list topology_timer = | |
1483 | TIMER_INITIALIZER(topology_timer_fn, 0, 0); | |
1484 | ||
1485 | static void set_topology_timer(void) | |
1486 | { | |
1487 | topology_timer.data = 0; | |
1488 | topology_timer.expires = jiffies + 60 * HZ; | |
1489 | add_timer(&topology_timer); | |
1490 | } | |
1491 | ||
1492 | /* | |
1493 | * Start polling for VPHN associativity changes. | |
1494 | */ | |
1495 | int start_topology_update(void) | |
1496 | { | |
1497 | int rc = 0; | |
1498 | ||
36e8695c BH |
1499 | /* Disabled until races with load balancing are fixed */ |
1500 | if (0 && firmware_has_feature(FW_FEATURE_VPHN) && | |
fe5cfd63 | 1501 | get_lppaca()->shared_proc) { |
9eff1a38 JL |
1502 | vphn_enabled = 1; |
1503 | setup_cpu_associativity_change_counters(); | |
1504 | init_timer_deferrable(&topology_timer); | |
1505 | set_topology_timer(); | |
1506 | rc = 1; | |
1507 | } | |
1508 | ||
1509 | return rc; | |
1510 | } | |
1511 | __initcall(start_topology_update); | |
1512 | ||
1513 | /* | |
1514 | * Disable polling for VPHN associativity changes. | |
1515 | */ | |
1516 | int stop_topology_update(void) | |
1517 | { | |
1518 | vphn_enabled = 0; | |
1519 | return del_timer_sync(&topology_timer); | |
1520 | } | |
39bf990e | 1521 | #endif /* CONFIG_PPC_SPLPAR */ |