]>
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> | |
30c05350 | 25 | #include <linux/stop_machine.h> |
45fb6cea | 26 | #include <asm/sparsemem.h> |
d9b2b2a2 | 27 | #include <asm/prom.h> |
2249ca9d | 28 | #include <asm/smp.h> |
9eff1a38 JL |
29 | #include <asm/firmware.h> |
30 | #include <asm/paca.h> | |
39bf990e | 31 | #include <asm/hvcall.h> |
ae3a197e | 32 | #include <asm/setup.h> |
176bbf14 | 33 | #include <asm/vdso.h> |
1da177e4 LT |
34 | |
35 | static int numa_enabled = 1; | |
36 | ||
1daa6d08 BS |
37 | static char *cmdline __initdata; |
38 | ||
1da177e4 LT |
39 | static int numa_debug; |
40 | #define dbg(args...) if (numa_debug) { printk(KERN_INFO args); } | |
41 | ||
45fb6cea | 42 | int numa_cpu_lookup_table[NR_CPUS]; |
25863de0 | 43 | cpumask_var_t node_to_cpumask_map[MAX_NUMNODES]; |
1da177e4 | 44 | struct pglist_data *node_data[MAX_NUMNODES]; |
45fb6cea AB |
45 | |
46 | EXPORT_SYMBOL(numa_cpu_lookup_table); | |
25863de0 | 47 | EXPORT_SYMBOL(node_to_cpumask_map); |
45fb6cea AB |
48 | EXPORT_SYMBOL(node_data); |
49 | ||
1da177e4 | 50 | static int min_common_depth; |
237a0989 | 51 | static int n_mem_addr_cells, n_mem_size_cells; |
41eab6f8 AB |
52 | static int form1_affinity; |
53 | ||
54 | #define MAX_DISTANCE_REF_POINTS 4 | |
55 | static int distance_ref_points_depth; | |
56 | static const unsigned int *distance_ref_points; | |
57 | static int distance_lookup_table[MAX_NUMNODES][MAX_DISTANCE_REF_POINTS]; | |
1da177e4 | 58 | |
25863de0 AB |
59 | /* |
60 | * Allocate node_to_cpumask_map based on number of available nodes | |
61 | * Requires node_possible_map to be valid. | |
62 | * | |
9512938b | 63 | * Note: cpumask_of_node() is not valid until after this is done. |
25863de0 AB |
64 | */ |
65 | static void __init setup_node_to_cpumask_map(void) | |
66 | { | |
67 | unsigned int node, num = 0; | |
68 | ||
69 | /* setup nr_node_ids if not done yet */ | |
70 | if (nr_node_ids == MAX_NUMNODES) { | |
71 | for_each_node_mask(node, node_possible_map) | |
72 | num = node; | |
73 | nr_node_ids = num + 1; | |
74 | } | |
75 | ||
76 | /* allocate the map */ | |
77 | for (node = 0; node < nr_node_ids; node++) | |
78 | alloc_bootmem_cpumask_var(&node_to_cpumask_map[node]); | |
79 | ||
80 | /* cpumask_of_node() will now work */ | |
81 | dbg("Node to cpumask map for %d nodes\n", nr_node_ids); | |
82 | } | |
83 | ||
55671f3c | 84 | static int __init fake_numa_create_new_node(unsigned long end_pfn, |
1daa6d08 BS |
85 | unsigned int *nid) |
86 | { | |
87 | unsigned long long mem; | |
88 | char *p = cmdline; | |
89 | static unsigned int fake_nid; | |
90 | static unsigned long long curr_boundary; | |
91 | ||
92 | /* | |
93 | * Modify node id, iff we started creating NUMA nodes | |
94 | * We want to continue from where we left of the last time | |
95 | */ | |
96 | if (fake_nid) | |
97 | *nid = fake_nid; | |
98 | /* | |
99 | * In case there are no more arguments to parse, the | |
100 | * node_id should be the same as the last fake node id | |
101 | * (we've handled this above). | |
102 | */ | |
103 | if (!p) | |
104 | return 0; | |
105 | ||
106 | mem = memparse(p, &p); | |
107 | if (!mem) | |
108 | return 0; | |
109 | ||
110 | if (mem < curr_boundary) | |
111 | return 0; | |
112 | ||
113 | curr_boundary = mem; | |
114 | ||
115 | if ((end_pfn << PAGE_SHIFT) > mem) { | |
116 | /* | |
117 | * Skip commas and spaces | |
118 | */ | |
119 | while (*p == ',' || *p == ' ' || *p == '\t') | |
120 | p++; | |
121 | ||
122 | cmdline = p; | |
123 | fake_nid++; | |
124 | *nid = fake_nid; | |
125 | dbg("created new fake_node with id %d\n", fake_nid); | |
126 | return 1; | |
127 | } | |
128 | return 0; | |
129 | } | |
130 | ||
8f64e1f2 | 131 | /* |
5dfe8660 | 132 | * get_node_active_region - Return active region containing pfn |
e8170372 | 133 | * Active range returned is empty if none found. |
5dfe8660 TH |
134 | * @pfn: The page to return the region for |
135 | * @node_ar: Returned set to the active region containing @pfn | |
8f64e1f2 | 136 | */ |
5dfe8660 TH |
137 | static void __init get_node_active_region(unsigned long pfn, |
138 | struct node_active_region *node_ar) | |
8f64e1f2 | 139 | { |
5dfe8660 TH |
140 | unsigned long start_pfn, end_pfn; |
141 | int i, nid; | |
142 | ||
143 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { | |
144 | if (pfn >= start_pfn && pfn < end_pfn) { | |
145 | node_ar->nid = nid; | |
146 | node_ar->start_pfn = start_pfn; | |
147 | node_ar->end_pfn = end_pfn; | |
148 | break; | |
149 | } | |
150 | } | |
8f64e1f2 JT |
151 | } |
152 | ||
39bf990e | 153 | static void map_cpu_to_node(int cpu, int node) |
1da177e4 LT |
154 | { |
155 | numa_cpu_lookup_table[cpu] = node; | |
45fb6cea | 156 | |
bf4b85b0 NL |
157 | dbg("adding cpu %d to node %d\n", cpu, node); |
158 | ||
25863de0 AB |
159 | if (!(cpumask_test_cpu(cpu, node_to_cpumask_map[node]))) |
160 | cpumask_set_cpu(cpu, node_to_cpumask_map[node]); | |
1da177e4 LT |
161 | } |
162 | ||
39bf990e | 163 | #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_PPC_SPLPAR) |
1da177e4 LT |
164 | static void unmap_cpu_from_node(unsigned long cpu) |
165 | { | |
166 | int node = numa_cpu_lookup_table[cpu]; | |
167 | ||
168 | dbg("removing cpu %lu from node %d\n", cpu, node); | |
169 | ||
25863de0 | 170 | if (cpumask_test_cpu(cpu, node_to_cpumask_map[node])) { |
429f4d8d | 171 | cpumask_clear_cpu(cpu, node_to_cpumask_map[node]); |
1da177e4 LT |
172 | } else { |
173 | printk(KERN_ERR "WARNING: cpu %lu not found in node %d\n", | |
174 | cpu, node); | |
175 | } | |
176 | } | |
39bf990e | 177 | #endif /* CONFIG_HOTPLUG_CPU || CONFIG_PPC_SPLPAR */ |
1da177e4 | 178 | |
1da177e4 | 179 | /* must hold reference to node during call */ |
a7f67bdf | 180 | static const int *of_get_associativity(struct device_node *dev) |
1da177e4 | 181 | { |
e2eb6392 | 182 | return of_get_property(dev, "ibm,associativity", NULL); |
1da177e4 LT |
183 | } |
184 | ||
cf00085d C |
185 | /* |
186 | * Returns the property linux,drconf-usable-memory if | |
187 | * it exists (the property exists only in kexec/kdump kernels, | |
188 | * added by kexec-tools) | |
189 | */ | |
190 | static const u32 *of_get_usable_memory(struct device_node *memory) | |
191 | { | |
192 | const u32 *prop; | |
193 | u32 len; | |
194 | prop = of_get_property(memory, "linux,drconf-usable-memory", &len); | |
195 | if (!prop || len < sizeof(unsigned int)) | |
196 | return 0; | |
197 | return prop; | |
198 | } | |
199 | ||
41eab6f8 AB |
200 | int __node_distance(int a, int b) |
201 | { | |
202 | int i; | |
203 | int distance = LOCAL_DISTANCE; | |
204 | ||
205 | if (!form1_affinity) | |
7122beee | 206 | return ((a == b) ? LOCAL_DISTANCE : REMOTE_DISTANCE); |
41eab6f8 AB |
207 | |
208 | for (i = 0; i < distance_ref_points_depth; i++) { | |
209 | if (distance_lookup_table[a][i] == distance_lookup_table[b][i]) | |
210 | break; | |
211 | ||
212 | /* Double the distance for each NUMA level */ | |
213 | distance *= 2; | |
214 | } | |
215 | ||
216 | return distance; | |
217 | } | |
218 | ||
219 | static void initialize_distance_lookup_table(int nid, | |
220 | const unsigned int *associativity) | |
221 | { | |
222 | int i; | |
223 | ||
224 | if (!form1_affinity) | |
225 | return; | |
226 | ||
227 | for (i = 0; i < distance_ref_points_depth; i++) { | |
228 | distance_lookup_table[nid][i] = | |
229 | associativity[distance_ref_points[i]]; | |
230 | } | |
231 | } | |
232 | ||
482ec7c4 NL |
233 | /* Returns nid in the range [0..MAX_NUMNODES-1], or -1 if no useful numa |
234 | * info is found. | |
235 | */ | |
9eff1a38 | 236 | static int associativity_to_nid(const unsigned int *associativity) |
1da177e4 | 237 | { |
482ec7c4 | 238 | int nid = -1; |
1da177e4 LT |
239 | |
240 | if (min_common_depth == -1) | |
482ec7c4 | 241 | goto out; |
1da177e4 | 242 | |
9eff1a38 JL |
243 | if (associativity[0] >= min_common_depth) |
244 | nid = associativity[min_common_depth]; | |
bc16a759 NL |
245 | |
246 | /* POWER4 LPAR uses 0xffff as invalid node */ | |
482ec7c4 NL |
247 | if (nid == 0xffff || nid >= MAX_NUMNODES) |
248 | nid = -1; | |
41eab6f8 | 249 | |
9eff1a38 JL |
250 | if (nid > 0 && associativity[0] >= distance_ref_points_depth) |
251 | initialize_distance_lookup_table(nid, associativity); | |
41eab6f8 | 252 | |
482ec7c4 | 253 | out: |
cf950b7a | 254 | return nid; |
1da177e4 LT |
255 | } |
256 | ||
9eff1a38 JL |
257 | /* Returns the nid associated with the given device tree node, |
258 | * or -1 if not found. | |
259 | */ | |
260 | static int of_node_to_nid_single(struct device_node *device) | |
261 | { | |
262 | int nid = -1; | |
263 | const unsigned int *tmp; | |
264 | ||
265 | tmp = of_get_associativity(device); | |
266 | if (tmp) | |
267 | nid = associativity_to_nid(tmp); | |
268 | return nid; | |
269 | } | |
270 | ||
953039c8 JK |
271 | /* Walk the device tree upwards, looking for an associativity id */ |
272 | int of_node_to_nid(struct device_node *device) | |
273 | { | |
274 | struct device_node *tmp; | |
275 | int nid = -1; | |
276 | ||
277 | of_node_get(device); | |
278 | while (device) { | |
279 | nid = of_node_to_nid_single(device); | |
280 | if (nid != -1) | |
281 | break; | |
282 | ||
283 | tmp = device; | |
284 | device = of_get_parent(tmp); | |
285 | of_node_put(tmp); | |
286 | } | |
287 | of_node_put(device); | |
288 | ||
289 | return nid; | |
290 | } | |
291 | EXPORT_SYMBOL_GPL(of_node_to_nid); | |
292 | ||
1da177e4 LT |
293 | static int __init find_min_common_depth(void) |
294 | { | |
41eab6f8 | 295 | int depth; |
e70606eb | 296 | struct device_node *root; |
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 | |
8002b0c5 NF |
328 | if (firmware_has_feature(FW_FEATURE_OPAL) || |
329 | firmware_has_feature(FW_FEATURE_TYPE1_AFFINITY)) { | |
330 | dbg("Using form 1 affinity\n"); | |
1c8ee733 | 331 | form1_affinity = 1; |
4b83c330 AB |
332 | } |
333 | ||
41eab6f8 AB |
334 | if (form1_affinity) { |
335 | depth = distance_ref_points[0]; | |
1da177e4 | 336 | } else { |
41eab6f8 AB |
337 | if (distance_ref_points_depth < 2) { |
338 | printk(KERN_WARNING "NUMA: " | |
339 | "short ibm,associativity-reference-points\n"); | |
340 | goto err; | |
341 | } | |
342 | ||
343 | depth = distance_ref_points[1]; | |
1da177e4 | 344 | } |
1da177e4 | 345 | |
41eab6f8 AB |
346 | /* |
347 | * Warn and cap if the hardware supports more than | |
348 | * MAX_DISTANCE_REF_POINTS domains. | |
349 | */ | |
350 | if (distance_ref_points_depth > MAX_DISTANCE_REF_POINTS) { | |
351 | printk(KERN_WARNING "NUMA: distance array capped at " | |
352 | "%d entries\n", MAX_DISTANCE_REF_POINTS); | |
353 | distance_ref_points_depth = MAX_DISTANCE_REF_POINTS; | |
354 | } | |
355 | ||
e70606eb | 356 | of_node_put(root); |
1da177e4 | 357 | return depth; |
41eab6f8 AB |
358 | |
359 | err: | |
e70606eb | 360 | of_node_put(root); |
41eab6f8 | 361 | return -1; |
1da177e4 LT |
362 | } |
363 | ||
84c9fdd1 | 364 | static void __init get_n_mem_cells(int *n_addr_cells, int *n_size_cells) |
1da177e4 LT |
365 | { |
366 | struct device_node *memory = NULL; | |
1da177e4 LT |
367 | |
368 | memory = of_find_node_by_type(memory, "memory"); | |
54c23310 | 369 | if (!memory) |
84c9fdd1 | 370 | panic("numa.c: No memory nodes found!"); |
54c23310 | 371 | |
a8bda5dd | 372 | *n_addr_cells = of_n_addr_cells(memory); |
9213feea | 373 | *n_size_cells = of_n_size_cells(memory); |
84c9fdd1 | 374 | of_node_put(memory); |
1da177e4 LT |
375 | } |
376 | ||
2011b1d0 | 377 | static unsigned long read_n_cells(int n, const unsigned int **buf) |
1da177e4 LT |
378 | { |
379 | unsigned long result = 0; | |
380 | ||
381 | while (n--) { | |
382 | result = (result << 32) | **buf; | |
383 | (*buf)++; | |
384 | } | |
385 | return result; | |
386 | } | |
387 | ||
8342681d | 388 | /* |
95f72d1e | 389 | * Read the next memblock list entry from the ibm,dynamic-memory property |
8342681d NF |
390 | * and return the information in the provided of_drconf_cell structure. |
391 | */ | |
392 | static void read_drconf_cell(struct of_drconf_cell *drmem, const u32 **cellp) | |
393 | { | |
394 | const u32 *cp; | |
395 | ||
396 | drmem->base_addr = read_n_cells(n_mem_addr_cells, cellp); | |
397 | ||
398 | cp = *cellp; | |
399 | drmem->drc_index = cp[0]; | |
400 | drmem->reserved = cp[1]; | |
401 | drmem->aa_index = cp[2]; | |
402 | drmem->flags = cp[3]; | |
403 | ||
404 | *cellp = cp + 4; | |
405 | } | |
406 | ||
407 | /* | |
25985edc | 408 | * Retrieve and validate the ibm,dynamic-memory property of the device tree. |
8342681d | 409 | * |
95f72d1e YL |
410 | * The layout of the ibm,dynamic-memory property is a number N of memblock |
411 | * list entries followed by N memblock list entries. Each memblock list entry | |
25985edc | 412 | * contains information as laid out in the of_drconf_cell struct above. |
8342681d NF |
413 | */ |
414 | static int of_get_drconf_memory(struct device_node *memory, const u32 **dm) | |
415 | { | |
416 | const u32 *prop; | |
417 | u32 len, entries; | |
418 | ||
419 | prop = of_get_property(memory, "ibm,dynamic-memory", &len); | |
420 | if (!prop || len < sizeof(unsigned int)) | |
421 | return 0; | |
422 | ||
423 | entries = *prop++; | |
424 | ||
425 | /* Now that we know the number of entries, revalidate the size | |
426 | * of the property read in to ensure we have everything | |
427 | */ | |
428 | if (len < (entries * (n_mem_addr_cells + 4) + 1) * sizeof(unsigned int)) | |
429 | return 0; | |
430 | ||
431 | *dm = prop; | |
432 | return entries; | |
433 | } | |
434 | ||
435 | /* | |
25985edc | 436 | * Retrieve and validate the ibm,lmb-size property for drconf memory |
8342681d NF |
437 | * from the device tree. |
438 | */ | |
3fdfd990 | 439 | static u64 of_get_lmb_size(struct device_node *memory) |
8342681d NF |
440 | { |
441 | const u32 *prop; | |
442 | u32 len; | |
443 | ||
3fdfd990 | 444 | prop = of_get_property(memory, "ibm,lmb-size", &len); |
8342681d NF |
445 | if (!prop || len < sizeof(unsigned int)) |
446 | return 0; | |
447 | ||
448 | return read_n_cells(n_mem_size_cells, &prop); | |
449 | } | |
450 | ||
451 | struct assoc_arrays { | |
452 | u32 n_arrays; | |
453 | u32 array_sz; | |
454 | const u32 *arrays; | |
455 | }; | |
456 | ||
457 | /* | |
25985edc | 458 | * Retrieve and validate the list of associativity arrays for drconf |
8342681d NF |
459 | * memory from the ibm,associativity-lookup-arrays property of the |
460 | * device tree.. | |
461 | * | |
462 | * The layout of the ibm,associativity-lookup-arrays property is a number N | |
463 | * indicating the number of associativity arrays, followed by a number M | |
464 | * indicating the size of each associativity array, followed by a list | |
465 | * of N associativity arrays. | |
466 | */ | |
467 | static int of_get_assoc_arrays(struct device_node *memory, | |
468 | struct assoc_arrays *aa) | |
469 | { | |
470 | const u32 *prop; | |
471 | u32 len; | |
472 | ||
473 | prop = of_get_property(memory, "ibm,associativity-lookup-arrays", &len); | |
474 | if (!prop || len < 2 * sizeof(unsigned int)) | |
475 | return -1; | |
476 | ||
477 | aa->n_arrays = *prop++; | |
478 | aa->array_sz = *prop++; | |
479 | ||
42b2aa86 | 480 | /* Now that we know the number of arrays and size of each array, |
8342681d NF |
481 | * revalidate the size of the property read in. |
482 | */ | |
483 | if (len < (aa->n_arrays * aa->array_sz + 2) * sizeof(unsigned int)) | |
484 | return -1; | |
485 | ||
486 | aa->arrays = prop; | |
487 | return 0; | |
488 | } | |
489 | ||
490 | /* | |
491 | * This is like of_node_to_nid_single() for memory represented in the | |
492 | * ibm,dynamic-reconfiguration-memory node. | |
493 | */ | |
494 | static int of_drconf_to_nid_single(struct of_drconf_cell *drmem, | |
495 | struct assoc_arrays *aa) | |
496 | { | |
497 | int default_nid = 0; | |
498 | int nid = default_nid; | |
499 | int index; | |
500 | ||
501 | if (min_common_depth > 0 && min_common_depth <= aa->array_sz && | |
502 | !(drmem->flags & DRCONF_MEM_AI_INVALID) && | |
503 | drmem->aa_index < aa->n_arrays) { | |
504 | index = drmem->aa_index * aa->array_sz + min_common_depth - 1; | |
505 | nid = aa->arrays[index]; | |
506 | ||
507 | if (nid == 0xffff || nid >= MAX_NUMNODES) | |
508 | nid = default_nid; | |
509 | } | |
510 | ||
511 | return nid; | |
512 | } | |
513 | ||
1da177e4 LT |
514 | /* |
515 | * Figure out to which domain a cpu belongs and stick it there. | |
516 | * Return the id of the domain used. | |
517 | */ | |
2e5ce39d | 518 | static int __cpuinit numa_setup_cpu(unsigned long lcpu) |
1da177e4 | 519 | { |
cf950b7a | 520 | int nid = 0; |
8b16cd23 | 521 | struct device_node *cpu = of_get_cpu_node(lcpu, NULL); |
1da177e4 LT |
522 | |
523 | if (!cpu) { | |
524 | WARN_ON(1); | |
525 | goto out; | |
526 | } | |
527 | ||
953039c8 | 528 | nid = of_node_to_nid_single(cpu); |
1da177e4 | 529 | |
482ec7c4 | 530 | if (nid < 0 || !node_online(nid)) |
72c33688 | 531 | nid = first_online_node; |
1da177e4 | 532 | out: |
cf950b7a | 533 | map_cpu_to_node(lcpu, nid); |
1da177e4 LT |
534 | |
535 | of_node_put(cpu); | |
536 | ||
cf950b7a | 537 | return nid; |
1da177e4 LT |
538 | } |
539 | ||
74b85f37 | 540 | static int __cpuinit cpu_numa_callback(struct notifier_block *nfb, |
1da177e4 LT |
541 | unsigned long action, |
542 | void *hcpu) | |
543 | { | |
544 | unsigned long lcpu = (unsigned long)hcpu; | |
545 | int ret = NOTIFY_DONE; | |
546 | ||
547 | switch (action) { | |
548 | case CPU_UP_PREPARE: | |
8bb78442 | 549 | case CPU_UP_PREPARE_FROZEN: |
2b261227 | 550 | numa_setup_cpu(lcpu); |
1da177e4 LT |
551 | ret = NOTIFY_OK; |
552 | break; | |
553 | #ifdef CONFIG_HOTPLUG_CPU | |
554 | case CPU_DEAD: | |
8bb78442 | 555 | case CPU_DEAD_FROZEN: |
1da177e4 | 556 | case CPU_UP_CANCELED: |
8bb78442 | 557 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 LT |
558 | unmap_cpu_from_node(lcpu); |
559 | break; | |
560 | ret = NOTIFY_OK; | |
561 | #endif | |
562 | } | |
563 | return ret; | |
564 | } | |
565 | ||
566 | /* | |
567 | * Check and possibly modify a memory region to enforce the memory limit. | |
568 | * | |
569 | * Returns the size the region should have to enforce the memory limit. | |
570 | * This will either be the original value of size, a truncated value, | |
571 | * or zero. If the returned value of size is 0 the region should be | |
25985edc | 572 | * discarded as it lies wholly above the memory limit. |
1da177e4 | 573 | */ |
45fb6cea AB |
574 | static unsigned long __init numa_enforce_memory_limit(unsigned long start, |
575 | unsigned long size) | |
1da177e4 LT |
576 | { |
577 | /* | |
95f72d1e | 578 | * We use memblock_end_of_DRAM() in here instead of memory_limit because |
1da177e4 | 579 | * we've already adjusted it for the limit and it takes care of |
fe55249d MM |
580 | * having memory holes below the limit. Also, in the case of |
581 | * iommu_is_off, memory_limit is not set but is implicitly enforced. | |
1da177e4 | 582 | */ |
1da177e4 | 583 | |
95f72d1e | 584 | if (start + size <= memblock_end_of_DRAM()) |
1da177e4 LT |
585 | return size; |
586 | ||
95f72d1e | 587 | if (start >= memblock_end_of_DRAM()) |
1da177e4 LT |
588 | return 0; |
589 | ||
95f72d1e | 590 | return memblock_end_of_DRAM() - start; |
1da177e4 LT |
591 | } |
592 | ||
cf00085d C |
593 | /* |
594 | * Reads the counter for a given entry in | |
595 | * linux,drconf-usable-memory property | |
596 | */ | |
597 | static inline int __init read_usm_ranges(const u32 **usm) | |
598 | { | |
599 | /* | |
3fdfd990 | 600 | * For each lmb in ibm,dynamic-memory a corresponding |
cf00085d C |
601 | * entry in linux,drconf-usable-memory property contains |
602 | * a counter followed by that many (base, size) duple. | |
603 | * read the counter from linux,drconf-usable-memory | |
604 | */ | |
605 | return read_n_cells(n_mem_size_cells, usm); | |
606 | } | |
607 | ||
0204568a PM |
608 | /* |
609 | * Extract NUMA information from the ibm,dynamic-reconfiguration-memory | |
610 | * node. This assumes n_mem_{addr,size}_cells have been set. | |
611 | */ | |
612 | static void __init parse_drconf_memory(struct device_node *memory) | |
613 | { | |
82b2521d | 614 | const u32 *uninitialized_var(dm), *usm; |
cf00085d | 615 | unsigned int n, rc, ranges, is_kexec_kdump = 0; |
3fdfd990 | 616 | unsigned long lmb_size, base, size, sz; |
8342681d | 617 | int nid; |
aa709f3b | 618 | struct assoc_arrays aa = { .arrays = NULL }; |
8342681d NF |
619 | |
620 | n = of_get_drconf_memory(memory, &dm); | |
621 | if (!n) | |
0204568a PM |
622 | return; |
623 | ||
3fdfd990 BH |
624 | lmb_size = of_get_lmb_size(memory); |
625 | if (!lmb_size) | |
8342681d NF |
626 | return; |
627 | ||
628 | rc = of_get_assoc_arrays(memory, &aa); | |
629 | if (rc) | |
0204568a PM |
630 | return; |
631 | ||
cf00085d C |
632 | /* check if this is a kexec/kdump kernel */ |
633 | usm = of_get_usable_memory(memory); | |
634 | if (usm != NULL) | |
635 | is_kexec_kdump = 1; | |
636 | ||
0204568a | 637 | for (; n != 0; --n) { |
8342681d NF |
638 | struct of_drconf_cell drmem; |
639 | ||
640 | read_drconf_cell(&drmem, &dm); | |
641 | ||
642 | /* skip this block if the reserved bit is set in flags (0x80) | |
643 | or if the block is not assigned to this partition (0x8) */ | |
644 | if ((drmem.flags & DRCONF_MEM_RESERVED) | |
645 | || !(drmem.flags & DRCONF_MEM_ASSIGNED)) | |
0204568a | 646 | continue; |
1daa6d08 | 647 | |
cf00085d | 648 | base = drmem.base_addr; |
3fdfd990 | 649 | size = lmb_size; |
cf00085d | 650 | ranges = 1; |
8342681d | 651 | |
cf00085d C |
652 | if (is_kexec_kdump) { |
653 | ranges = read_usm_ranges(&usm); | |
654 | if (!ranges) /* there are no (base, size) duple */ | |
655 | continue; | |
656 | } | |
657 | do { | |
658 | if (is_kexec_kdump) { | |
659 | base = read_n_cells(n_mem_addr_cells, &usm); | |
660 | size = read_n_cells(n_mem_size_cells, &usm); | |
661 | } | |
662 | nid = of_drconf_to_nid_single(&drmem, &aa); | |
663 | fake_numa_create_new_node( | |
664 | ((base + size) >> PAGE_SHIFT), | |
8342681d | 665 | &nid); |
cf00085d C |
666 | node_set_online(nid); |
667 | sz = numa_enforce_memory_limit(base, size); | |
668 | if (sz) | |
1d7cfe18 | 669 | memblock_set_node(base, sz, nid); |
cf00085d | 670 | } while (--ranges); |
0204568a PM |
671 | } |
672 | } | |
673 | ||
1da177e4 LT |
674 | static int __init parse_numa_properties(void) |
675 | { | |
94db7c5e | 676 | struct device_node *memory; |
482ec7c4 | 677 | int default_nid = 0; |
1da177e4 LT |
678 | unsigned long i; |
679 | ||
680 | if (numa_enabled == 0) { | |
681 | printk(KERN_WARNING "NUMA disabled by user\n"); | |
682 | return -1; | |
683 | } | |
684 | ||
1da177e4 LT |
685 | min_common_depth = find_min_common_depth(); |
686 | ||
1da177e4 LT |
687 | if (min_common_depth < 0) |
688 | return min_common_depth; | |
689 | ||
bf4b85b0 NL |
690 | dbg("NUMA associativity depth for CPU/Memory: %d\n", min_common_depth); |
691 | ||
1da177e4 | 692 | /* |
482ec7c4 NL |
693 | * Even though we connect cpus to numa domains later in SMP |
694 | * init, we need to know the node ids now. This is because | |
695 | * each node to be onlined must have NODE_DATA etc backing it. | |
1da177e4 | 696 | */ |
482ec7c4 | 697 | for_each_present_cpu(i) { |
dfbe93a2 | 698 | struct device_node *cpu; |
cf950b7a | 699 | int nid; |
1da177e4 | 700 | |
8b16cd23 | 701 | cpu = of_get_cpu_node(i, NULL); |
482ec7c4 | 702 | BUG_ON(!cpu); |
953039c8 | 703 | nid = of_node_to_nid_single(cpu); |
482ec7c4 | 704 | of_node_put(cpu); |
1da177e4 | 705 | |
482ec7c4 NL |
706 | /* |
707 | * Don't fall back to default_nid yet -- we will plug | |
708 | * cpus into nodes once the memory scan has discovered | |
709 | * the topology. | |
710 | */ | |
711 | if (nid < 0) | |
712 | continue; | |
713 | node_set_online(nid); | |
1da177e4 LT |
714 | } |
715 | ||
237a0989 | 716 | get_n_mem_cells(&n_mem_addr_cells, &n_mem_size_cells); |
94db7c5e AB |
717 | |
718 | for_each_node_by_type(memory, "memory") { | |
1da177e4 LT |
719 | unsigned long start; |
720 | unsigned long size; | |
cf950b7a | 721 | int nid; |
1da177e4 | 722 | int ranges; |
a7f67bdf | 723 | const unsigned int *memcell_buf; |
1da177e4 LT |
724 | unsigned int len; |
725 | ||
e2eb6392 | 726 | memcell_buf = of_get_property(memory, |
ba759485 ME |
727 | "linux,usable-memory", &len); |
728 | if (!memcell_buf || len <= 0) | |
e2eb6392 | 729 | memcell_buf = of_get_property(memory, "reg", &len); |
1da177e4 LT |
730 | if (!memcell_buf || len <= 0) |
731 | continue; | |
732 | ||
cc5d0189 BH |
733 | /* ranges in cell */ |
734 | ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells); | |
1da177e4 LT |
735 | new_range: |
736 | /* these are order-sensitive, and modify the buffer pointer */ | |
237a0989 MK |
737 | start = read_n_cells(n_mem_addr_cells, &memcell_buf); |
738 | size = read_n_cells(n_mem_size_cells, &memcell_buf); | |
1da177e4 | 739 | |
482ec7c4 NL |
740 | /* |
741 | * Assumption: either all memory nodes or none will | |
742 | * have associativity properties. If none, then | |
743 | * everything goes to default_nid. | |
744 | */ | |
953039c8 | 745 | nid = of_node_to_nid_single(memory); |
482ec7c4 NL |
746 | if (nid < 0) |
747 | nid = default_nid; | |
1daa6d08 BS |
748 | |
749 | fake_numa_create_new_node(((start + size) >> PAGE_SHIFT), &nid); | |
482ec7c4 | 750 | node_set_online(nid); |
1da177e4 | 751 | |
45fb6cea | 752 | if (!(size = numa_enforce_memory_limit(start, size))) { |
1da177e4 LT |
753 | if (--ranges) |
754 | goto new_range; | |
755 | else | |
756 | continue; | |
757 | } | |
758 | ||
1d7cfe18 | 759 | memblock_set_node(start, size, nid); |
1da177e4 LT |
760 | |
761 | if (--ranges) | |
762 | goto new_range; | |
763 | } | |
764 | ||
0204568a | 765 | /* |
dfbe93a2 AB |
766 | * Now do the same thing for each MEMBLOCK listed in the |
767 | * ibm,dynamic-memory property in the | |
768 | * ibm,dynamic-reconfiguration-memory node. | |
0204568a PM |
769 | */ |
770 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); | |
771 | if (memory) | |
772 | parse_drconf_memory(memory); | |
773 | ||
1da177e4 LT |
774 | return 0; |
775 | } | |
776 | ||
777 | static void __init setup_nonnuma(void) | |
778 | { | |
95f72d1e YL |
779 | unsigned long top_of_ram = memblock_end_of_DRAM(); |
780 | unsigned long total_ram = memblock_phys_mem_size(); | |
c67c3cb4 | 781 | unsigned long start_pfn, end_pfn; |
28be7072 BH |
782 | unsigned int nid = 0; |
783 | struct memblock_region *reg; | |
1da177e4 | 784 | |
e110b281 | 785 | printk(KERN_DEBUG "Top of RAM: 0x%lx, Total RAM: 0x%lx\n", |
1da177e4 | 786 | top_of_ram, total_ram); |
e110b281 | 787 | printk(KERN_DEBUG "Memory hole size: %ldMB\n", |
1da177e4 LT |
788 | (top_of_ram - total_ram) >> 20); |
789 | ||
28be7072 | 790 | for_each_memblock(memory, reg) { |
c7fc2de0 YL |
791 | start_pfn = memblock_region_memory_base_pfn(reg); |
792 | end_pfn = memblock_region_memory_end_pfn(reg); | |
1daa6d08 BS |
793 | |
794 | fake_numa_create_new_node(end_pfn, &nid); | |
1d7cfe18 TH |
795 | memblock_set_node(PFN_PHYS(start_pfn), |
796 | PFN_PHYS(end_pfn - start_pfn), nid); | |
1daa6d08 | 797 | node_set_online(nid); |
c67c3cb4 | 798 | } |
1da177e4 LT |
799 | } |
800 | ||
4b703a23 AB |
801 | void __init dump_numa_cpu_topology(void) |
802 | { | |
803 | unsigned int node; | |
804 | unsigned int cpu, count; | |
805 | ||
806 | if (min_common_depth == -1 || !numa_enabled) | |
807 | return; | |
808 | ||
809 | for_each_online_node(node) { | |
e110b281 | 810 | printk(KERN_DEBUG "Node %d CPUs:", node); |
4b703a23 AB |
811 | |
812 | count = 0; | |
813 | /* | |
814 | * If we used a CPU iterator here we would miss printing | |
815 | * the holes in the cpumap. | |
816 | */ | |
25863de0 AB |
817 | for (cpu = 0; cpu < nr_cpu_ids; cpu++) { |
818 | if (cpumask_test_cpu(cpu, | |
819 | node_to_cpumask_map[node])) { | |
4b703a23 AB |
820 | if (count == 0) |
821 | printk(" %u", cpu); | |
822 | ++count; | |
823 | } else { | |
824 | if (count > 1) | |
825 | printk("-%u", cpu - 1); | |
826 | count = 0; | |
827 | } | |
828 | } | |
829 | ||
830 | if (count > 1) | |
25863de0 | 831 | printk("-%u", nr_cpu_ids - 1); |
4b703a23 AB |
832 | printk("\n"); |
833 | } | |
834 | } | |
835 | ||
836 | static void __init dump_numa_memory_topology(void) | |
1da177e4 LT |
837 | { |
838 | unsigned int node; | |
839 | unsigned int count; | |
840 | ||
841 | if (min_common_depth == -1 || !numa_enabled) | |
842 | return; | |
843 | ||
844 | for_each_online_node(node) { | |
845 | unsigned long i; | |
846 | ||
e110b281 | 847 | printk(KERN_DEBUG "Node %d Memory:", node); |
1da177e4 LT |
848 | |
849 | count = 0; | |
850 | ||
95f72d1e | 851 | for (i = 0; i < memblock_end_of_DRAM(); |
45fb6cea AB |
852 | i += (1 << SECTION_SIZE_BITS)) { |
853 | if (early_pfn_to_nid(i >> PAGE_SHIFT) == node) { | |
1da177e4 LT |
854 | if (count == 0) |
855 | printk(" 0x%lx", i); | |
856 | ++count; | |
857 | } else { | |
858 | if (count > 0) | |
859 | printk("-0x%lx", i); | |
860 | count = 0; | |
861 | } | |
862 | } | |
863 | ||
864 | if (count > 0) | |
865 | printk("-0x%lx", i); | |
866 | printk("\n"); | |
867 | } | |
1da177e4 LT |
868 | } |
869 | ||
870 | /* | |
95f72d1e | 871 | * Allocate some memory, satisfying the memblock or bootmem allocator where |
1da177e4 LT |
872 | * required. nid is the preferred node and end is the physical address of |
873 | * the highest address in the node. | |
874 | * | |
0be210fd | 875 | * Returns the virtual address of the memory. |
1da177e4 | 876 | */ |
893473df | 877 | static void __init *careful_zallocation(int nid, unsigned long size, |
45fb6cea AB |
878 | unsigned long align, |
879 | unsigned long end_pfn) | |
1da177e4 | 880 | { |
0be210fd | 881 | void *ret; |
45fb6cea | 882 | int new_nid; |
0be210fd DH |
883 | unsigned long ret_paddr; |
884 | ||
95f72d1e | 885 | ret_paddr = __memblock_alloc_base(size, align, end_pfn << PAGE_SHIFT); |
1da177e4 LT |
886 | |
887 | /* retry over all memory */ | |
0be210fd | 888 | if (!ret_paddr) |
95f72d1e | 889 | ret_paddr = __memblock_alloc_base(size, align, memblock_end_of_DRAM()); |
1da177e4 | 890 | |
0be210fd | 891 | if (!ret_paddr) |
5d21ea2b | 892 | panic("numa.c: cannot allocate %lu bytes for node %d", |
1da177e4 LT |
893 | size, nid); |
894 | ||
0be210fd DH |
895 | ret = __va(ret_paddr); |
896 | ||
1da177e4 | 897 | /* |
c555e520 | 898 | * We initialize the nodes in numeric order: 0, 1, 2... |
95f72d1e | 899 | * and hand over control from the MEMBLOCK allocator to the |
c555e520 DH |
900 | * bootmem allocator. If this function is called for |
901 | * node 5, then we know that all nodes <5 are using the | |
95f72d1e | 902 | * bootmem allocator instead of the MEMBLOCK allocator. |
c555e520 DH |
903 | * |
904 | * So, check the nid from which this allocation came | |
905 | * and double check to see if we need to use bootmem | |
95f72d1e | 906 | * instead of the MEMBLOCK. We don't free the MEMBLOCK memory |
c555e520 | 907 | * since it would be useless. |
1da177e4 | 908 | */ |
0be210fd | 909 | new_nid = early_pfn_to_nid(ret_paddr >> PAGE_SHIFT); |
45fb6cea | 910 | if (new_nid < nid) { |
0be210fd | 911 | ret = __alloc_bootmem_node(NODE_DATA(new_nid), |
1da177e4 LT |
912 | size, align, 0); |
913 | ||
0be210fd | 914 | dbg("alloc_bootmem %p %lx\n", ret, size); |
1da177e4 LT |
915 | } |
916 | ||
893473df | 917 | memset(ret, 0, size); |
0be210fd | 918 | return ret; |
1da177e4 LT |
919 | } |
920 | ||
74b85f37 CS |
921 | static struct notifier_block __cpuinitdata ppc64_numa_nb = { |
922 | .notifier_call = cpu_numa_callback, | |
923 | .priority = 1 /* Must run before sched domains notifier. */ | |
924 | }; | |
925 | ||
28e86bdb | 926 | static void __init mark_reserved_regions_for_nid(int nid) |
4a618669 DH |
927 | { |
928 | struct pglist_data *node = NODE_DATA(nid); | |
28be7072 | 929 | struct memblock_region *reg; |
4a618669 | 930 | |
28be7072 BH |
931 | for_each_memblock(reserved, reg) { |
932 | unsigned long physbase = reg->base; | |
933 | unsigned long size = reg->size; | |
4a618669 | 934 | unsigned long start_pfn = physbase >> PAGE_SHIFT; |
06eccea6 | 935 | unsigned long end_pfn = PFN_UP(physbase + size); |
4a618669 DH |
936 | struct node_active_region node_ar; |
937 | unsigned long node_end_pfn = node->node_start_pfn + | |
938 | node->node_spanned_pages; | |
939 | ||
940 | /* | |
95f72d1e | 941 | * Check to make sure that this memblock.reserved area is |
4a618669 DH |
942 | * within the bounds of the node that we care about. |
943 | * Checking the nid of the start and end points is not | |
944 | * sufficient because the reserved area could span the | |
945 | * entire node. | |
946 | */ | |
947 | if (end_pfn <= node->node_start_pfn || | |
948 | start_pfn >= node_end_pfn) | |
949 | continue; | |
950 | ||
951 | get_node_active_region(start_pfn, &node_ar); | |
952 | while (start_pfn < end_pfn && | |
953 | node_ar.start_pfn < node_ar.end_pfn) { | |
954 | unsigned long reserve_size = size; | |
955 | /* | |
956 | * if reserved region extends past active region | |
957 | * then trim size to active region | |
958 | */ | |
959 | if (end_pfn > node_ar.end_pfn) | |
960 | reserve_size = (node_ar.end_pfn << PAGE_SHIFT) | |
06eccea6 | 961 | - physbase; |
a4c74ddd DH |
962 | /* |
963 | * Only worry about *this* node, others may not | |
964 | * yet have valid NODE_DATA(). | |
965 | */ | |
966 | if (node_ar.nid == nid) { | |
967 | dbg("reserve_bootmem %lx %lx nid=%d\n", | |
968 | physbase, reserve_size, node_ar.nid); | |
969 | reserve_bootmem_node(NODE_DATA(node_ar.nid), | |
970 | physbase, reserve_size, | |
971 | BOOTMEM_DEFAULT); | |
972 | } | |
4a618669 DH |
973 | /* |
974 | * if reserved region is contained in the active region | |
975 | * then done. | |
976 | */ | |
977 | if (end_pfn <= node_ar.end_pfn) | |
978 | break; | |
979 | ||
980 | /* | |
981 | * reserved region extends past the active region | |
982 | * get next active region that contains this | |
983 | * reserved region | |
984 | */ | |
985 | start_pfn = node_ar.end_pfn; | |
986 | physbase = start_pfn << PAGE_SHIFT; | |
987 | size = size - reserve_size; | |
988 | get_node_active_region(start_pfn, &node_ar); | |
989 | } | |
990 | } | |
991 | } | |
992 | ||
993 | ||
1da177e4 LT |
994 | void __init do_init_bootmem(void) |
995 | { | |
996 | int nid; | |
1da177e4 LT |
997 | |
998 | min_low_pfn = 0; | |
95f72d1e | 999 | max_low_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT; |
1da177e4 LT |
1000 | max_pfn = max_low_pfn; |
1001 | ||
1002 | if (parse_numa_properties()) | |
1003 | setup_nonnuma(); | |
1004 | else | |
4b703a23 | 1005 | dump_numa_memory_topology(); |
1da177e4 | 1006 | |
1da177e4 | 1007 | for_each_online_node(nid) { |
c67c3cb4 | 1008 | unsigned long start_pfn, end_pfn; |
0be210fd | 1009 | void *bootmem_vaddr; |
1da177e4 LT |
1010 | unsigned long bootmap_pages; |
1011 | ||
c67c3cb4 | 1012 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); |
1da177e4 | 1013 | |
4a618669 DH |
1014 | /* |
1015 | * Allocate the node structure node local if possible | |
1016 | * | |
1017 | * Be careful moving this around, as it relies on all | |
1018 | * previous nodes' bootmem to be initialized and have | |
1019 | * all reserved areas marked. | |
1020 | */ | |
893473df | 1021 | NODE_DATA(nid) = careful_zallocation(nid, |
1da177e4 | 1022 | sizeof(struct pglist_data), |
45fb6cea | 1023 | SMP_CACHE_BYTES, end_pfn); |
1da177e4 LT |
1024 | |
1025 | dbg("node %d\n", nid); | |
1026 | dbg("NODE_DATA() = %p\n", NODE_DATA(nid)); | |
1027 | ||
b61bfa3c | 1028 | NODE_DATA(nid)->bdata = &bootmem_node_data[nid]; |
45fb6cea AB |
1029 | NODE_DATA(nid)->node_start_pfn = start_pfn; |
1030 | NODE_DATA(nid)->node_spanned_pages = end_pfn - start_pfn; | |
1da177e4 LT |
1031 | |
1032 | if (NODE_DATA(nid)->node_spanned_pages == 0) | |
1033 | continue; | |
1034 | ||
45fb6cea AB |
1035 | dbg("start_paddr = %lx\n", start_pfn << PAGE_SHIFT); |
1036 | dbg("end_paddr = %lx\n", end_pfn << PAGE_SHIFT); | |
1da177e4 | 1037 | |
45fb6cea | 1038 | bootmap_pages = bootmem_bootmap_pages(end_pfn - start_pfn); |
893473df | 1039 | bootmem_vaddr = careful_zallocation(nid, |
45fb6cea AB |
1040 | bootmap_pages << PAGE_SHIFT, |
1041 | PAGE_SIZE, end_pfn); | |
1da177e4 | 1042 | |
0be210fd | 1043 | dbg("bootmap_vaddr = %p\n", bootmem_vaddr); |
1da177e4 | 1044 | |
0be210fd DH |
1045 | init_bootmem_node(NODE_DATA(nid), |
1046 | __pa(bootmem_vaddr) >> PAGE_SHIFT, | |
45fb6cea | 1047 | start_pfn, end_pfn); |
1da177e4 | 1048 | |
c67c3cb4 | 1049 | free_bootmem_with_active_regions(nid, end_pfn); |
4a618669 DH |
1050 | /* |
1051 | * Be very careful about moving this around. Future | |
893473df | 1052 | * calls to careful_zallocation() depend on this getting |
4a618669 DH |
1053 | * done correctly. |
1054 | */ | |
1055 | mark_reserved_regions_for_nid(nid); | |
8f64e1f2 | 1056 | sparse_memory_present_with_active_regions(nid); |
4a618669 | 1057 | } |
d3f6204a BH |
1058 | |
1059 | init_bootmem_done = 1; | |
25863de0 AB |
1060 | |
1061 | /* | |
1062 | * Now bootmem is initialised we can create the node to cpumask | |
1063 | * lookup tables and setup the cpu callback to populate them. | |
1064 | */ | |
1065 | setup_node_to_cpumask_map(); | |
1066 | ||
1067 | register_cpu_notifier(&ppc64_numa_nb); | |
1068 | cpu_numa_callback(&ppc64_numa_nb, CPU_UP_PREPARE, | |
1069 | (void *)(unsigned long)boot_cpuid); | |
1da177e4 LT |
1070 | } |
1071 | ||
1072 | void __init paging_init(void) | |
1073 | { | |
6391af17 MG |
1074 | unsigned long max_zone_pfns[MAX_NR_ZONES]; |
1075 | memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); | |
95f72d1e | 1076 | max_zone_pfns[ZONE_DMA] = memblock_end_of_DRAM() >> PAGE_SHIFT; |
c67c3cb4 | 1077 | free_area_init_nodes(max_zone_pfns); |
1da177e4 LT |
1078 | } |
1079 | ||
1080 | static int __init early_numa(char *p) | |
1081 | { | |
1082 | if (!p) | |
1083 | return 0; | |
1084 | ||
1085 | if (strstr(p, "off")) | |
1086 | numa_enabled = 0; | |
1087 | ||
1088 | if (strstr(p, "debug")) | |
1089 | numa_debug = 1; | |
1090 | ||
1daa6d08 BS |
1091 | p = strstr(p, "fake="); |
1092 | if (p) | |
1093 | cmdline = p + strlen("fake="); | |
1094 | ||
1da177e4 LT |
1095 | return 0; |
1096 | } | |
1097 | early_param("numa", early_numa); | |
237a0989 MK |
1098 | |
1099 | #ifdef CONFIG_MEMORY_HOTPLUG | |
0db9360a | 1100 | /* |
0f16ef7f NF |
1101 | * Find the node associated with a hot added memory section for |
1102 | * memory represented in the device tree by the property | |
1103 | * ibm,dynamic-reconfiguration-memory/ibm,dynamic-memory. | |
0db9360a NF |
1104 | */ |
1105 | static int hot_add_drconf_scn_to_nid(struct device_node *memory, | |
1106 | unsigned long scn_addr) | |
1107 | { | |
1108 | const u32 *dm; | |
0f16ef7f | 1109 | unsigned int drconf_cell_cnt, rc; |
3fdfd990 | 1110 | unsigned long lmb_size; |
0db9360a | 1111 | struct assoc_arrays aa; |
0f16ef7f | 1112 | int nid = -1; |
0db9360a | 1113 | |
0f16ef7f NF |
1114 | drconf_cell_cnt = of_get_drconf_memory(memory, &dm); |
1115 | if (!drconf_cell_cnt) | |
1116 | return -1; | |
0db9360a | 1117 | |
3fdfd990 BH |
1118 | lmb_size = of_get_lmb_size(memory); |
1119 | if (!lmb_size) | |
0f16ef7f | 1120 | return -1; |
0db9360a NF |
1121 | |
1122 | rc = of_get_assoc_arrays(memory, &aa); | |
1123 | if (rc) | |
0f16ef7f | 1124 | return -1; |
0db9360a | 1125 | |
0f16ef7f | 1126 | for (; drconf_cell_cnt != 0; --drconf_cell_cnt) { |
0db9360a NF |
1127 | struct of_drconf_cell drmem; |
1128 | ||
1129 | read_drconf_cell(&drmem, &dm); | |
1130 | ||
1131 | /* skip this block if it is reserved or not assigned to | |
1132 | * this partition */ | |
1133 | if ((drmem.flags & DRCONF_MEM_RESERVED) | |
1134 | || !(drmem.flags & DRCONF_MEM_ASSIGNED)) | |
1135 | continue; | |
1136 | ||
0f16ef7f | 1137 | if ((scn_addr < drmem.base_addr) |
3fdfd990 | 1138 | || (scn_addr >= (drmem.base_addr + lmb_size))) |
0f16ef7f NF |
1139 | continue; |
1140 | ||
0db9360a | 1141 | nid = of_drconf_to_nid_single(&drmem, &aa); |
0f16ef7f NF |
1142 | break; |
1143 | } | |
1144 | ||
1145 | return nid; | |
1146 | } | |
1147 | ||
1148 | /* | |
1149 | * Find the node associated with a hot added memory section for memory | |
1150 | * represented in the device tree as a node (i.e. memory@XXXX) for | |
95f72d1e | 1151 | * each memblock. |
0f16ef7f NF |
1152 | */ |
1153 | int hot_add_node_scn_to_nid(unsigned long scn_addr) | |
1154 | { | |
94db7c5e | 1155 | struct device_node *memory; |
0f16ef7f NF |
1156 | int nid = -1; |
1157 | ||
94db7c5e | 1158 | for_each_node_by_type(memory, "memory") { |
0f16ef7f NF |
1159 | unsigned long start, size; |
1160 | int ranges; | |
1161 | const unsigned int *memcell_buf; | |
1162 | unsigned int len; | |
1163 | ||
1164 | memcell_buf = of_get_property(memory, "reg", &len); | |
1165 | if (!memcell_buf || len <= 0) | |
1166 | continue; | |
1167 | ||
1168 | /* ranges in cell */ | |
1169 | ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells); | |
1170 | ||
1171 | while (ranges--) { | |
1172 | start = read_n_cells(n_mem_addr_cells, &memcell_buf); | |
1173 | size = read_n_cells(n_mem_size_cells, &memcell_buf); | |
1174 | ||
1175 | if ((scn_addr < start) || (scn_addr >= (start + size))) | |
1176 | continue; | |
1177 | ||
1178 | nid = of_node_to_nid_single(memory); | |
1179 | break; | |
1180 | } | |
0db9360a | 1181 | |
0f16ef7f NF |
1182 | if (nid >= 0) |
1183 | break; | |
0db9360a NF |
1184 | } |
1185 | ||
60831842 AB |
1186 | of_node_put(memory); |
1187 | ||
0f16ef7f | 1188 | return nid; |
0db9360a NF |
1189 | } |
1190 | ||
237a0989 MK |
1191 | /* |
1192 | * Find the node associated with a hot added memory section. Section | |
95f72d1e YL |
1193 | * corresponds to a SPARSEMEM section, not an MEMBLOCK. It is assumed that |
1194 | * sections are fully contained within a single MEMBLOCK. | |
237a0989 MK |
1195 | */ |
1196 | int hot_add_scn_to_nid(unsigned long scn_addr) | |
1197 | { | |
1198 | struct device_node *memory = NULL; | |
0f16ef7f | 1199 | int nid, found = 0; |
237a0989 MK |
1200 | |
1201 | if (!numa_enabled || (min_common_depth < 0)) | |
72c33688 | 1202 | return first_online_node; |
0db9360a NF |
1203 | |
1204 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); | |
1205 | if (memory) { | |
1206 | nid = hot_add_drconf_scn_to_nid(memory, scn_addr); | |
1207 | of_node_put(memory); | |
0f16ef7f NF |
1208 | } else { |
1209 | nid = hot_add_node_scn_to_nid(scn_addr); | |
0db9360a | 1210 | } |
237a0989 | 1211 | |
0f16ef7f | 1212 | if (nid < 0 || !node_online(nid)) |
72c33688 | 1213 | nid = first_online_node; |
237a0989 | 1214 | |
0f16ef7f NF |
1215 | if (NODE_DATA(nid)->node_spanned_pages) |
1216 | return nid; | |
237a0989 | 1217 | |
0f16ef7f NF |
1218 | for_each_online_node(nid) { |
1219 | if (NODE_DATA(nid)->node_spanned_pages) { | |
1220 | found = 1; | |
1221 | break; | |
237a0989 | 1222 | } |
237a0989 | 1223 | } |
0f16ef7f NF |
1224 | |
1225 | BUG_ON(!found); | |
1226 | return nid; | |
237a0989 | 1227 | } |
0f16ef7f | 1228 | |
cd34206e NA |
1229 | static u64 hot_add_drconf_memory_max(void) |
1230 | { | |
1231 | struct device_node *memory = NULL; | |
1232 | unsigned int drconf_cell_cnt = 0; | |
1233 | u64 lmb_size = 0; | |
1234 | const u32 *dm = 0; | |
1235 | ||
1236 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); | |
1237 | if (memory) { | |
1238 | drconf_cell_cnt = of_get_drconf_memory(memory, &dm); | |
1239 | lmb_size = of_get_lmb_size(memory); | |
1240 | of_node_put(memory); | |
1241 | } | |
1242 | return lmb_size * drconf_cell_cnt; | |
1243 | } | |
1244 | ||
1245 | /* | |
1246 | * memory_hotplug_max - return max address of memory that may be added | |
1247 | * | |
1248 | * This is currently only used on systems that support drconfig memory | |
1249 | * hotplug. | |
1250 | */ | |
1251 | u64 memory_hotplug_max(void) | |
1252 | { | |
1253 | return max(hot_add_drconf_memory_max(), memblock_end_of_DRAM()); | |
1254 | } | |
237a0989 | 1255 | #endif /* CONFIG_MEMORY_HOTPLUG */ |
9eff1a38 | 1256 | |
bd03403a | 1257 | /* Virtual Processor Home Node (VPHN) support */ |
39bf990e | 1258 | #ifdef CONFIG_PPC_SPLPAR |
30c05350 NF |
1259 | struct topology_update_data { |
1260 | struct topology_update_data *next; | |
1261 | unsigned int cpu; | |
1262 | int old_nid; | |
1263 | int new_nid; | |
1264 | }; | |
1265 | ||
5de16699 | 1266 | static u8 vphn_cpu_change_counts[NR_CPUS][MAX_DISTANCE_REF_POINTS]; |
9eff1a38 JL |
1267 | static cpumask_t cpu_associativity_changes_mask; |
1268 | static int vphn_enabled; | |
5d88aa85 JL |
1269 | static int prrn_enabled; |
1270 | static void reset_topology_timer(void); | |
9eff1a38 JL |
1271 | |
1272 | /* | |
1273 | * Store the current values of the associativity change counters in the | |
1274 | * hypervisor. | |
1275 | */ | |
1276 | static void setup_cpu_associativity_change_counters(void) | |
1277 | { | |
cd9d6cc7 | 1278 | int cpu; |
9eff1a38 | 1279 | |
5de16699 AB |
1280 | /* The VPHN feature supports a maximum of 8 reference points */ |
1281 | BUILD_BUG_ON(MAX_DISTANCE_REF_POINTS > 8); | |
1282 | ||
9eff1a38 | 1283 | for_each_possible_cpu(cpu) { |
cd9d6cc7 | 1284 | int i; |
9eff1a38 JL |
1285 | u8 *counts = vphn_cpu_change_counts[cpu]; |
1286 | volatile u8 *hypervisor_counts = lppaca[cpu].vphn_assoc_counts; | |
1287 | ||
5de16699 | 1288 | for (i = 0; i < distance_ref_points_depth; i++) |
9eff1a38 | 1289 | counts[i] = hypervisor_counts[i]; |
9eff1a38 JL |
1290 | } |
1291 | } | |
1292 | ||
1293 | /* | |
1294 | * The hypervisor maintains a set of 8 associativity change counters in | |
1295 | * the VPA of each cpu that correspond to the associativity levels in the | |
1296 | * ibm,associativity-reference-points property. When an associativity | |
1297 | * level changes, the corresponding counter is incremented. | |
1298 | * | |
1299 | * Set a bit in cpu_associativity_changes_mask for each cpu whose home | |
1300 | * node associativity levels have changed. | |
1301 | * | |
1302 | * Returns the number of cpus with unhandled associativity changes. | |
1303 | */ | |
1304 | static int update_cpu_associativity_changes_mask(void) | |
1305 | { | |
5d88aa85 | 1306 | int cpu; |
9eff1a38 JL |
1307 | cpumask_t *changes = &cpu_associativity_changes_mask; |
1308 | ||
9eff1a38 JL |
1309 | for_each_possible_cpu(cpu) { |
1310 | int i, changed = 0; | |
1311 | u8 *counts = vphn_cpu_change_counts[cpu]; | |
1312 | volatile u8 *hypervisor_counts = lppaca[cpu].vphn_assoc_counts; | |
1313 | ||
5de16699 | 1314 | for (i = 0; i < distance_ref_points_depth; i++) { |
d69043e8 | 1315 | if (hypervisor_counts[i] != counts[i]) { |
9eff1a38 JL |
1316 | counts[i] = hypervisor_counts[i]; |
1317 | changed = 1; | |
1318 | } | |
1319 | } | |
1320 | if (changed) { | |
1321 | cpumask_set_cpu(cpu, changes); | |
9eff1a38 JL |
1322 | } |
1323 | } | |
1324 | ||
5d88aa85 | 1325 | return cpumask_weight(changes); |
9eff1a38 JL |
1326 | } |
1327 | ||
c0e5e46f AB |
1328 | /* |
1329 | * 6 64-bit registers unpacked into 12 32-bit associativity values. To form | |
1330 | * the complete property we have to add the length in the first cell. | |
1331 | */ | |
1332 | #define VPHN_ASSOC_BUFSIZE (6*sizeof(u64)/sizeof(u32) + 1) | |
9eff1a38 JL |
1333 | |
1334 | /* | |
1335 | * Convert the associativity domain numbers returned from the hypervisor | |
1336 | * to the sequence they would appear in the ibm,associativity property. | |
1337 | */ | |
1338 | static int vphn_unpack_associativity(const long *packed, unsigned int *unpacked) | |
1339 | { | |
cd9d6cc7 | 1340 | int i, nr_assoc_doms = 0; |
9eff1a38 JL |
1341 | const u16 *field = (const u16*) packed; |
1342 | ||
1343 | #define VPHN_FIELD_UNUSED (0xffff) | |
1344 | #define VPHN_FIELD_MSB (0x8000) | |
1345 | #define VPHN_FIELD_MASK (~VPHN_FIELD_MSB) | |
1346 | ||
c0e5e46f | 1347 | for (i = 1; i < VPHN_ASSOC_BUFSIZE; i++) { |
9eff1a38 JL |
1348 | if (*field == VPHN_FIELD_UNUSED) { |
1349 | /* All significant fields processed, and remaining | |
1350 | * fields contain the reserved value of all 1's. | |
1351 | * Just store them. | |
1352 | */ | |
1353 | unpacked[i] = *((u32*)field); | |
1354 | field += 2; | |
7639adaa | 1355 | } else if (*field & VPHN_FIELD_MSB) { |
9eff1a38 JL |
1356 | /* Data is in the lower 15 bits of this field */ |
1357 | unpacked[i] = *field & VPHN_FIELD_MASK; | |
1358 | field++; | |
1359 | nr_assoc_doms++; | |
7639adaa | 1360 | } else { |
9eff1a38 JL |
1361 | /* Data is in the lower 15 bits of this field |
1362 | * concatenated with the next 16 bit field | |
1363 | */ | |
1364 | unpacked[i] = *((u32*)field); | |
1365 | field += 2; | |
1366 | nr_assoc_doms++; | |
1367 | } | |
1368 | } | |
1369 | ||
c0e5e46f AB |
1370 | /* The first cell contains the length of the property */ |
1371 | unpacked[0] = nr_assoc_doms; | |
1372 | ||
9eff1a38 JL |
1373 | return nr_assoc_doms; |
1374 | } | |
1375 | ||
1376 | /* | |
1377 | * Retrieve the new associativity information for a virtual processor's | |
1378 | * home node. | |
1379 | */ | |
1380 | static long hcall_vphn(unsigned long cpu, unsigned int *associativity) | |
1381 | { | |
cd9d6cc7 | 1382 | long rc; |
9eff1a38 JL |
1383 | long retbuf[PLPAR_HCALL9_BUFSIZE] = {0}; |
1384 | u64 flags = 1; | |
1385 | int hwcpu = get_hard_smp_processor_id(cpu); | |
1386 | ||
1387 | rc = plpar_hcall9(H_HOME_NODE_ASSOCIATIVITY, retbuf, flags, hwcpu); | |
1388 | vphn_unpack_associativity(retbuf, associativity); | |
1389 | ||
1390 | return rc; | |
1391 | } | |
1392 | ||
1393 | static long vphn_get_associativity(unsigned long cpu, | |
1394 | unsigned int *associativity) | |
1395 | { | |
cd9d6cc7 | 1396 | long rc; |
9eff1a38 JL |
1397 | |
1398 | rc = hcall_vphn(cpu, associativity); | |
1399 | ||
1400 | switch (rc) { | |
1401 | case H_FUNCTION: | |
1402 | printk(KERN_INFO | |
1403 | "VPHN is not supported. Disabling polling...\n"); | |
1404 | stop_topology_update(); | |
1405 | break; | |
1406 | case H_HARDWARE: | |
1407 | printk(KERN_ERR | |
1408 | "hcall_vphn() experienced a hardware fault " | |
1409 | "preventing VPHN. Disabling polling...\n"); | |
1410 | stop_topology_update(); | |
1411 | } | |
1412 | ||
1413 | return rc; | |
1414 | } | |
1415 | ||
30c05350 NF |
1416 | /* |
1417 | * Update the CPU maps and sysfs entries for a single CPU when its NUMA | |
1418 | * characteristics change. This function doesn't perform any locking and is | |
1419 | * only safe to call from stop_machine(). | |
1420 | */ | |
1421 | static int update_cpu_topology(void *data) | |
1422 | { | |
1423 | struct topology_update_data *update; | |
1424 | unsigned long cpu; | |
1425 | ||
1426 | if (!data) | |
1427 | return -EINVAL; | |
1428 | ||
1429 | cpu = get_cpu(); | |
1430 | ||
1431 | for (update = data; update; update = update->next) { | |
1432 | if (cpu != update->cpu) | |
1433 | continue; | |
1434 | ||
1435 | unregister_cpu_under_node(update->cpu, update->old_nid); | |
1436 | unmap_cpu_from_node(update->cpu); | |
1437 | map_cpu_to_node(update->cpu, update->new_nid); | |
176bbf14 | 1438 | vdso_getcpu_init(); |
30c05350 NF |
1439 | register_cpu_under_node(update->cpu, update->new_nid); |
1440 | } | |
1441 | ||
1442 | return 0; | |
1443 | } | |
1444 | ||
9eff1a38 JL |
1445 | /* |
1446 | * Update the node maps and sysfs entries for each cpu whose home node | |
79c5fceb | 1447 | * has changed. Returns 1 when the topology has changed, and 0 otherwise. |
9eff1a38 JL |
1448 | */ |
1449 | int arch_update_cpu_topology(void) | |
1450 | { | |
30c05350 NF |
1451 | unsigned int cpu, changed = 0; |
1452 | struct topology_update_data *updates, *ud; | |
9eff1a38 | 1453 | unsigned int associativity[VPHN_ASSOC_BUFSIZE] = {0}; |
176bbf14 | 1454 | cpumask_t updated_cpus; |
8a25a2fd | 1455 | struct device *dev; |
30c05350 NF |
1456 | int weight, i = 0; |
1457 | ||
1458 | weight = cpumask_weight(&cpu_associativity_changes_mask); | |
1459 | if (!weight) | |
1460 | return 0; | |
1461 | ||
1462 | updates = kzalloc(weight * (sizeof(*updates)), GFP_KERNEL); | |
1463 | if (!updates) | |
1464 | return 0; | |
9eff1a38 | 1465 | |
176bbf14 JL |
1466 | cpumask_clear(&updated_cpus); |
1467 | ||
5d88aa85 | 1468 | for_each_cpu(cpu, &cpu_associativity_changes_mask) { |
30c05350 NF |
1469 | ud = &updates[i++]; |
1470 | ud->cpu = cpu; | |
9eff1a38 | 1471 | vphn_get_associativity(cpu, associativity); |
30c05350 | 1472 | ud->new_nid = associativity_to_nid(associativity); |
9eff1a38 | 1473 | |
30c05350 NF |
1474 | if (ud->new_nid < 0 || !node_online(ud->new_nid)) |
1475 | ud->new_nid = first_online_node; | |
9eff1a38 | 1476 | |
30c05350 | 1477 | ud->old_nid = numa_cpu_lookup_table[cpu]; |
176bbf14 | 1478 | cpumask_set_cpu(cpu, &updated_cpus); |
9eff1a38 | 1479 | |
30c05350 NF |
1480 | if (i < weight) |
1481 | ud->next = &updates[i]; | |
1482 | } | |
1483 | ||
176bbf14 | 1484 | stop_machine(update_cpu_topology, &updates[0], &updated_cpus); |
30c05350 NF |
1485 | |
1486 | for (ud = &updates[0]; ud; ud = ud->next) { | |
1487 | dev = get_cpu_device(ud->cpu); | |
8a25a2fd KS |
1488 | if (dev) |
1489 | kobject_uevent(&dev->kobj, KOBJ_CHANGE); | |
30c05350 | 1490 | cpumask_clear_cpu(ud->cpu, &cpu_associativity_changes_mask); |
79c5fceb | 1491 | changed = 1; |
9eff1a38 JL |
1492 | } |
1493 | ||
30c05350 | 1494 | kfree(updates); |
79c5fceb | 1495 | return changed; |
9eff1a38 JL |
1496 | } |
1497 | ||
1498 | static void topology_work_fn(struct work_struct *work) | |
1499 | { | |
1500 | rebuild_sched_domains(); | |
1501 | } | |
1502 | static DECLARE_WORK(topology_work, topology_work_fn); | |
1503 | ||
1504 | void topology_schedule_update(void) | |
1505 | { | |
1506 | schedule_work(&topology_work); | |
1507 | } | |
1508 | ||
1509 | static void topology_timer_fn(unsigned long ignored) | |
1510 | { | |
5d88aa85 | 1511 | if (prrn_enabled && cpumask_weight(&cpu_associativity_changes_mask)) |
9eff1a38 | 1512 | topology_schedule_update(); |
5d88aa85 JL |
1513 | else if (vphn_enabled) { |
1514 | if (update_cpu_associativity_changes_mask() > 0) | |
1515 | topology_schedule_update(); | |
1516 | reset_topology_timer(); | |
1517 | } | |
9eff1a38 JL |
1518 | } |
1519 | static struct timer_list topology_timer = | |
1520 | TIMER_INITIALIZER(topology_timer_fn, 0, 0); | |
1521 | ||
5d88aa85 | 1522 | static void reset_topology_timer(void) |
9eff1a38 JL |
1523 | { |
1524 | topology_timer.data = 0; | |
1525 | topology_timer.expires = jiffies + 60 * HZ; | |
5d88aa85 JL |
1526 | mod_timer(&topology_timer, topology_timer.expires); |
1527 | } | |
1528 | ||
1529 | static void stage_topology_update(int core_id) | |
1530 | { | |
1531 | cpumask_or(&cpu_associativity_changes_mask, | |
1532 | &cpu_associativity_changes_mask, cpu_sibling_mask(core_id)); | |
1533 | reset_topology_timer(); | |
1534 | } | |
1535 | ||
1536 | static int dt_update_callback(struct notifier_block *nb, | |
1537 | unsigned long action, void *data) | |
1538 | { | |
1539 | struct of_prop_reconfig *update; | |
1540 | int rc = NOTIFY_DONE; | |
1541 | ||
1542 | switch (action) { | |
5d88aa85 JL |
1543 | case OF_RECONFIG_UPDATE_PROPERTY: |
1544 | update = (struct of_prop_reconfig *)data; | |
30c05350 NF |
1545 | if (!of_prop_cmp(update->dn->type, "cpu") && |
1546 | !of_prop_cmp(update->prop->name, "ibm,associativity")) { | |
5d88aa85 JL |
1547 | u32 core_id; |
1548 | of_property_read_u32(update->dn, "reg", &core_id); | |
1549 | stage_topology_update(core_id); | |
1550 | rc = NOTIFY_OK; | |
1551 | } | |
1552 | break; | |
1553 | } | |
1554 | ||
1555 | return rc; | |
9eff1a38 JL |
1556 | } |
1557 | ||
5d88aa85 JL |
1558 | static struct notifier_block dt_update_nb = { |
1559 | .notifier_call = dt_update_callback, | |
1560 | }; | |
1561 | ||
9eff1a38 | 1562 | /* |
5d88aa85 | 1563 | * Start polling for associativity changes. |
9eff1a38 JL |
1564 | */ |
1565 | int start_topology_update(void) | |
1566 | { | |
1567 | int rc = 0; | |
1568 | ||
5d88aa85 JL |
1569 | if (firmware_has_feature(FW_FEATURE_PRRN)) { |
1570 | if (!prrn_enabled) { | |
1571 | prrn_enabled = 1; | |
1572 | vphn_enabled = 0; | |
1573 | rc = of_reconfig_notifier_register(&dt_update_nb); | |
1574 | } | |
1575 | } else if (0 && firmware_has_feature(FW_FEATURE_VPHN) && | |
1576 | get_lppaca()->shared_proc) { | |
1577 | /* Disabled until races with load balancing are fixed */ | |
1578 | if (!vphn_enabled) { | |
1579 | prrn_enabled = 0; | |
1580 | vphn_enabled = 1; | |
1581 | setup_cpu_associativity_change_counters(); | |
1582 | init_timer_deferrable(&topology_timer); | |
1583 | reset_topology_timer(); | |
1584 | } | |
9eff1a38 JL |
1585 | } |
1586 | ||
1587 | return rc; | |
1588 | } | |
1589 | __initcall(start_topology_update); | |
1590 | ||
1591 | /* | |
1592 | * Disable polling for VPHN associativity changes. | |
1593 | */ | |
1594 | int stop_topology_update(void) | |
1595 | { | |
5d88aa85 JL |
1596 | int rc = 0; |
1597 | ||
1598 | if (prrn_enabled) { | |
1599 | prrn_enabled = 0; | |
1600 | rc = of_reconfig_notifier_unregister(&dt_update_nb); | |
1601 | } else if (vphn_enabled) { | |
1602 | vphn_enabled = 0; | |
1603 | rc = del_timer_sync(&topology_timer); | |
1604 | } | |
1605 | ||
1606 | return rc; | |
9eff1a38 | 1607 | } |
39bf990e | 1608 | #endif /* CONFIG_PPC_SPLPAR */ |