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> |
1da177e4 LT |
33 | |
34 | static int numa_enabled = 1; | |
35 | ||
1daa6d08 BS |
36 | static char *cmdline __initdata; |
37 | ||
1da177e4 LT |
38 | static int numa_debug; |
39 | #define dbg(args...) if (numa_debug) { printk(KERN_INFO args); } | |
40 | ||
45fb6cea | 41 | int numa_cpu_lookup_table[NR_CPUS]; |
25863de0 | 42 | cpumask_var_t node_to_cpumask_map[MAX_NUMNODES]; |
1da177e4 | 43 | struct pglist_data *node_data[MAX_NUMNODES]; |
45fb6cea AB |
44 | |
45 | EXPORT_SYMBOL(numa_cpu_lookup_table); | |
25863de0 | 46 | EXPORT_SYMBOL(node_to_cpumask_map); |
45fb6cea AB |
47 | EXPORT_SYMBOL(node_data); |
48 | ||
1da177e4 | 49 | static int min_common_depth; |
237a0989 | 50 | static int n_mem_addr_cells, n_mem_size_cells; |
41eab6f8 AB |
51 | static int form1_affinity; |
52 | ||
53 | #define MAX_DISTANCE_REF_POINTS 4 | |
54 | static int distance_ref_points_depth; | |
55 | static const unsigned int *distance_ref_points; | |
56 | static int distance_lookup_table[MAX_NUMNODES][MAX_DISTANCE_REF_POINTS]; | |
1da177e4 | 57 | |
25863de0 AB |
58 | /* |
59 | * Allocate node_to_cpumask_map based on number of available nodes | |
60 | * Requires node_possible_map to be valid. | |
61 | * | |
9512938b | 62 | * Note: cpumask_of_node() is not valid until after this is done. |
25863de0 AB |
63 | */ |
64 | static void __init setup_node_to_cpumask_map(void) | |
65 | { | |
66 | unsigned int node, num = 0; | |
67 | ||
68 | /* setup nr_node_ids if not done yet */ | |
69 | if (nr_node_ids == MAX_NUMNODES) { | |
70 | for_each_node_mask(node, node_possible_map) | |
71 | num = node; | |
72 | nr_node_ids = num + 1; | |
73 | } | |
74 | ||
75 | /* allocate the map */ | |
76 | for (node = 0; node < nr_node_ids; node++) | |
77 | alloc_bootmem_cpumask_var(&node_to_cpumask_map[node]); | |
78 | ||
79 | /* cpumask_of_node() will now work */ | |
80 | dbg("Node to cpumask map for %d nodes\n", nr_node_ids); | |
81 | } | |
82 | ||
55671f3c | 83 | static int __init fake_numa_create_new_node(unsigned long end_pfn, |
1daa6d08 BS |
84 | unsigned int *nid) |
85 | { | |
86 | unsigned long long mem; | |
87 | char *p = cmdline; | |
88 | static unsigned int fake_nid; | |
89 | static unsigned long long curr_boundary; | |
90 | ||
91 | /* | |
92 | * Modify node id, iff we started creating NUMA nodes | |
93 | * We want to continue from where we left of the last time | |
94 | */ | |
95 | if (fake_nid) | |
96 | *nid = fake_nid; | |
97 | /* | |
98 | * In case there are no more arguments to parse, the | |
99 | * node_id should be the same as the last fake node id | |
100 | * (we've handled this above). | |
101 | */ | |
102 | if (!p) | |
103 | return 0; | |
104 | ||
105 | mem = memparse(p, &p); | |
106 | if (!mem) | |
107 | return 0; | |
108 | ||
109 | if (mem < curr_boundary) | |
110 | return 0; | |
111 | ||
112 | curr_boundary = mem; | |
113 | ||
114 | if ((end_pfn << PAGE_SHIFT) > mem) { | |
115 | /* | |
116 | * Skip commas and spaces | |
117 | */ | |
118 | while (*p == ',' || *p == ' ' || *p == '\t') | |
119 | p++; | |
120 | ||
121 | cmdline = p; | |
122 | fake_nid++; | |
123 | *nid = fake_nid; | |
124 | dbg("created new fake_node with id %d\n", fake_nid); | |
125 | return 1; | |
126 | } | |
127 | return 0; | |
128 | } | |
129 | ||
8f64e1f2 | 130 | /* |
5dfe8660 | 131 | * get_node_active_region - Return active region containing pfn |
e8170372 | 132 | * Active range returned is empty if none found. |
5dfe8660 TH |
133 | * @pfn: The page to return the region for |
134 | * @node_ar: Returned set to the active region containing @pfn | |
8f64e1f2 | 135 | */ |
5dfe8660 TH |
136 | static void __init get_node_active_region(unsigned long pfn, |
137 | struct node_active_region *node_ar) | |
8f64e1f2 | 138 | { |
5dfe8660 TH |
139 | unsigned long start_pfn, end_pfn; |
140 | int i, nid; | |
141 | ||
142 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { | |
143 | if (pfn >= start_pfn && pfn < end_pfn) { | |
144 | node_ar->nid = nid; | |
145 | node_ar->start_pfn = start_pfn; | |
146 | node_ar->end_pfn = end_pfn; | |
147 | break; | |
148 | } | |
149 | } | |
8f64e1f2 JT |
150 | } |
151 | ||
39bf990e | 152 | static void map_cpu_to_node(int cpu, int node) |
1da177e4 LT |
153 | { |
154 | numa_cpu_lookup_table[cpu] = node; | |
45fb6cea | 155 | |
bf4b85b0 NL |
156 | dbg("adding cpu %d to node %d\n", cpu, node); |
157 | ||
25863de0 AB |
158 | if (!(cpumask_test_cpu(cpu, node_to_cpumask_map[node]))) |
159 | cpumask_set_cpu(cpu, node_to_cpumask_map[node]); | |
1da177e4 LT |
160 | } |
161 | ||
39bf990e | 162 | #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_PPC_SPLPAR) |
1da177e4 LT |
163 | static void unmap_cpu_from_node(unsigned long cpu) |
164 | { | |
165 | int node = numa_cpu_lookup_table[cpu]; | |
166 | ||
167 | dbg("removing cpu %lu from node %d\n", cpu, node); | |
168 | ||
25863de0 | 169 | if (cpumask_test_cpu(cpu, node_to_cpumask_map[node])) { |
429f4d8d | 170 | cpumask_clear_cpu(cpu, node_to_cpumask_map[node]); |
1da177e4 LT |
171 | } else { |
172 | printk(KERN_ERR "WARNING: cpu %lu not found in node %d\n", | |
173 | cpu, node); | |
174 | } | |
175 | } | |
39bf990e | 176 | #endif /* CONFIG_HOTPLUG_CPU || CONFIG_PPC_SPLPAR */ |
1da177e4 | 177 | |
1da177e4 | 178 | /* must hold reference to node during call */ |
a7f67bdf | 179 | static const int *of_get_associativity(struct device_node *dev) |
1da177e4 | 180 | { |
e2eb6392 | 181 | return of_get_property(dev, "ibm,associativity", NULL); |
1da177e4 LT |
182 | } |
183 | ||
cf00085d C |
184 | /* |
185 | * Returns the property linux,drconf-usable-memory if | |
186 | * it exists (the property exists only in kexec/kdump kernels, | |
187 | * added by kexec-tools) | |
188 | */ | |
189 | static const u32 *of_get_usable_memory(struct device_node *memory) | |
190 | { | |
191 | const u32 *prop; | |
192 | u32 len; | |
193 | prop = of_get_property(memory, "linux,drconf-usable-memory", &len); | |
194 | if (!prop || len < sizeof(unsigned int)) | |
195 | return 0; | |
196 | return prop; | |
197 | } | |
198 | ||
41eab6f8 AB |
199 | int __node_distance(int a, int b) |
200 | { | |
201 | int i; | |
202 | int distance = LOCAL_DISTANCE; | |
203 | ||
204 | if (!form1_affinity) | |
7122beee | 205 | return ((a == b) ? LOCAL_DISTANCE : REMOTE_DISTANCE); |
41eab6f8 AB |
206 | |
207 | for (i = 0; i < distance_ref_points_depth; i++) { | |
208 | if (distance_lookup_table[a][i] == distance_lookup_table[b][i]) | |
209 | break; | |
210 | ||
211 | /* Double the distance for each NUMA level */ | |
212 | distance *= 2; | |
213 | } | |
214 | ||
215 | return distance; | |
216 | } | |
217 | ||
218 | static void initialize_distance_lookup_table(int nid, | |
219 | const unsigned int *associativity) | |
220 | { | |
221 | int i; | |
222 | ||
223 | if (!form1_affinity) | |
224 | return; | |
225 | ||
226 | for (i = 0; i < distance_ref_points_depth; i++) { | |
227 | distance_lookup_table[nid][i] = | |
228 | associativity[distance_ref_points[i]]; | |
229 | } | |
230 | } | |
231 | ||
482ec7c4 NL |
232 | /* Returns nid in the range [0..MAX_NUMNODES-1], or -1 if no useful numa |
233 | * info is found. | |
234 | */ | |
9eff1a38 | 235 | static int associativity_to_nid(const unsigned int *associativity) |
1da177e4 | 236 | { |
482ec7c4 | 237 | int nid = -1; |
1da177e4 LT |
238 | |
239 | if (min_common_depth == -1) | |
482ec7c4 | 240 | goto out; |
1da177e4 | 241 | |
9eff1a38 JL |
242 | if (associativity[0] >= min_common_depth) |
243 | nid = associativity[min_common_depth]; | |
bc16a759 NL |
244 | |
245 | /* POWER4 LPAR uses 0xffff as invalid node */ | |
482ec7c4 NL |
246 | if (nid == 0xffff || nid >= MAX_NUMNODES) |
247 | nid = -1; | |
41eab6f8 | 248 | |
9eff1a38 JL |
249 | if (nid > 0 && associativity[0] >= distance_ref_points_depth) |
250 | initialize_distance_lookup_table(nid, associativity); | |
41eab6f8 | 251 | |
482ec7c4 | 252 | out: |
cf950b7a | 253 | return nid; |
1da177e4 LT |
254 | } |
255 | ||
9eff1a38 JL |
256 | /* Returns the nid associated with the given device tree node, |
257 | * or -1 if not found. | |
258 | */ | |
259 | static int of_node_to_nid_single(struct device_node *device) | |
260 | { | |
261 | int nid = -1; | |
262 | const unsigned int *tmp; | |
263 | ||
264 | tmp = of_get_associativity(device); | |
265 | if (tmp) | |
266 | nid = associativity_to_nid(tmp); | |
267 | return nid; | |
268 | } | |
269 | ||
953039c8 JK |
270 | /* Walk the device tree upwards, looking for an associativity id */ |
271 | int of_node_to_nid(struct device_node *device) | |
272 | { | |
273 | struct device_node *tmp; | |
274 | int nid = -1; | |
275 | ||
276 | of_node_get(device); | |
277 | while (device) { | |
278 | nid = of_node_to_nid_single(device); | |
279 | if (nid != -1) | |
280 | break; | |
281 | ||
282 | tmp = device; | |
283 | device = of_get_parent(tmp); | |
284 | of_node_put(tmp); | |
285 | } | |
286 | of_node_put(device); | |
287 | ||
288 | return nid; | |
289 | } | |
290 | EXPORT_SYMBOL_GPL(of_node_to_nid); | |
291 | ||
1da177e4 LT |
292 | static int __init find_min_common_depth(void) |
293 | { | |
41eab6f8 | 294 | int depth; |
e70606eb | 295 | struct device_node *root; |
1da177e4 | 296 | |
1c8ee733 DS |
297 | if (firmware_has_feature(FW_FEATURE_OPAL)) |
298 | root = of_find_node_by_path("/ibm,opal"); | |
299 | else | |
300 | root = of_find_node_by_path("/rtas"); | |
e70606eb ME |
301 | if (!root) |
302 | root = of_find_node_by_path("/"); | |
1da177e4 LT |
303 | |
304 | /* | |
41eab6f8 AB |
305 | * This property is a set of 32-bit integers, each representing |
306 | * an index into the ibm,associativity nodes. | |
307 | * | |
308 | * With form 0 affinity the first integer is for an SMP configuration | |
309 | * (should be all 0's) and the second is for a normal NUMA | |
310 | * configuration. We have only one level of NUMA. | |
311 | * | |
312 | * With form 1 affinity the first integer is the most significant | |
313 | * NUMA boundary and the following are progressively less significant | |
314 | * boundaries. There can be more than one level of NUMA. | |
1da177e4 | 315 | */ |
e70606eb | 316 | distance_ref_points = of_get_property(root, |
41eab6f8 AB |
317 | "ibm,associativity-reference-points", |
318 | &distance_ref_points_depth); | |
319 | ||
320 | if (!distance_ref_points) { | |
321 | dbg("NUMA: ibm,associativity-reference-points not found.\n"); | |
322 | goto err; | |
323 | } | |
324 | ||
325 | distance_ref_points_depth /= sizeof(int); | |
1da177e4 | 326 | |
8002b0c5 NF |
327 | if (firmware_has_feature(FW_FEATURE_OPAL) || |
328 | firmware_has_feature(FW_FEATURE_TYPE1_AFFINITY)) { | |
329 | dbg("Using form 1 affinity\n"); | |
1c8ee733 | 330 | form1_affinity = 1; |
4b83c330 AB |
331 | } |
332 | ||
41eab6f8 AB |
333 | if (form1_affinity) { |
334 | depth = distance_ref_points[0]; | |
1da177e4 | 335 | } else { |
41eab6f8 AB |
336 | if (distance_ref_points_depth < 2) { |
337 | printk(KERN_WARNING "NUMA: " | |
338 | "short ibm,associativity-reference-points\n"); | |
339 | goto err; | |
340 | } | |
341 | ||
342 | depth = distance_ref_points[1]; | |
1da177e4 | 343 | } |
1da177e4 | 344 | |
41eab6f8 AB |
345 | /* |
346 | * Warn and cap if the hardware supports more than | |
347 | * MAX_DISTANCE_REF_POINTS domains. | |
348 | */ | |
349 | if (distance_ref_points_depth > MAX_DISTANCE_REF_POINTS) { | |
350 | printk(KERN_WARNING "NUMA: distance array capped at " | |
351 | "%d entries\n", MAX_DISTANCE_REF_POINTS); | |
352 | distance_ref_points_depth = MAX_DISTANCE_REF_POINTS; | |
353 | } | |
354 | ||
e70606eb | 355 | of_node_put(root); |
1da177e4 | 356 | return depth; |
41eab6f8 AB |
357 | |
358 | err: | |
e70606eb | 359 | of_node_put(root); |
41eab6f8 | 360 | return -1; |
1da177e4 LT |
361 | } |
362 | ||
84c9fdd1 | 363 | static void __init get_n_mem_cells(int *n_addr_cells, int *n_size_cells) |
1da177e4 LT |
364 | { |
365 | struct device_node *memory = NULL; | |
1da177e4 LT |
366 | |
367 | memory = of_find_node_by_type(memory, "memory"); | |
54c23310 | 368 | if (!memory) |
84c9fdd1 | 369 | panic("numa.c: No memory nodes found!"); |
54c23310 | 370 | |
a8bda5dd | 371 | *n_addr_cells = of_n_addr_cells(memory); |
9213feea | 372 | *n_size_cells = of_n_size_cells(memory); |
84c9fdd1 | 373 | of_node_put(memory); |
1da177e4 LT |
374 | } |
375 | ||
2011b1d0 | 376 | static unsigned long read_n_cells(int n, const unsigned int **buf) |
1da177e4 LT |
377 | { |
378 | unsigned long result = 0; | |
379 | ||
380 | while (n--) { | |
381 | result = (result << 32) | **buf; | |
382 | (*buf)++; | |
383 | } | |
384 | return result; | |
385 | } | |
386 | ||
8342681d | 387 | /* |
95f72d1e | 388 | * Read the next memblock list entry from the ibm,dynamic-memory property |
8342681d NF |
389 | * and return the information in the provided of_drconf_cell structure. |
390 | */ | |
391 | static void read_drconf_cell(struct of_drconf_cell *drmem, const u32 **cellp) | |
392 | { | |
393 | const u32 *cp; | |
394 | ||
395 | drmem->base_addr = read_n_cells(n_mem_addr_cells, cellp); | |
396 | ||
397 | cp = *cellp; | |
398 | drmem->drc_index = cp[0]; | |
399 | drmem->reserved = cp[1]; | |
400 | drmem->aa_index = cp[2]; | |
401 | drmem->flags = cp[3]; | |
402 | ||
403 | *cellp = cp + 4; | |
404 | } | |
405 | ||
406 | /* | |
25985edc | 407 | * Retrieve and validate the ibm,dynamic-memory property of the device tree. |
8342681d | 408 | * |
95f72d1e YL |
409 | * The layout of the ibm,dynamic-memory property is a number N of memblock |
410 | * list entries followed by N memblock list entries. Each memblock list entry | |
25985edc | 411 | * contains information as laid out in the of_drconf_cell struct above. |
8342681d NF |
412 | */ |
413 | static int of_get_drconf_memory(struct device_node *memory, const u32 **dm) | |
414 | { | |
415 | const u32 *prop; | |
416 | u32 len, entries; | |
417 | ||
418 | prop = of_get_property(memory, "ibm,dynamic-memory", &len); | |
419 | if (!prop || len < sizeof(unsigned int)) | |
420 | return 0; | |
421 | ||
422 | entries = *prop++; | |
423 | ||
424 | /* Now that we know the number of entries, revalidate the size | |
425 | * of the property read in to ensure we have everything | |
426 | */ | |
427 | if (len < (entries * (n_mem_addr_cells + 4) + 1) * sizeof(unsigned int)) | |
428 | return 0; | |
429 | ||
430 | *dm = prop; | |
431 | return entries; | |
432 | } | |
433 | ||
434 | /* | |
25985edc | 435 | * Retrieve and validate the ibm,lmb-size property for drconf memory |
8342681d NF |
436 | * from the device tree. |
437 | */ | |
3fdfd990 | 438 | static u64 of_get_lmb_size(struct device_node *memory) |
8342681d NF |
439 | { |
440 | const u32 *prop; | |
441 | u32 len; | |
442 | ||
3fdfd990 | 443 | prop = of_get_property(memory, "ibm,lmb-size", &len); |
8342681d NF |
444 | if (!prop || len < sizeof(unsigned int)) |
445 | return 0; | |
446 | ||
447 | return read_n_cells(n_mem_size_cells, &prop); | |
448 | } | |
449 | ||
450 | struct assoc_arrays { | |
451 | u32 n_arrays; | |
452 | u32 array_sz; | |
453 | const u32 *arrays; | |
454 | }; | |
455 | ||
456 | /* | |
25985edc | 457 | * Retrieve and validate the list of associativity arrays for drconf |
8342681d NF |
458 | * memory from the ibm,associativity-lookup-arrays property of the |
459 | * device tree.. | |
460 | * | |
461 | * The layout of the ibm,associativity-lookup-arrays property is a number N | |
462 | * indicating the number of associativity arrays, followed by a number M | |
463 | * indicating the size of each associativity array, followed by a list | |
464 | * of N associativity arrays. | |
465 | */ | |
466 | static int of_get_assoc_arrays(struct device_node *memory, | |
467 | struct assoc_arrays *aa) | |
468 | { | |
469 | const u32 *prop; | |
470 | u32 len; | |
471 | ||
472 | prop = of_get_property(memory, "ibm,associativity-lookup-arrays", &len); | |
473 | if (!prop || len < 2 * sizeof(unsigned int)) | |
474 | return -1; | |
475 | ||
476 | aa->n_arrays = *prop++; | |
477 | aa->array_sz = *prop++; | |
478 | ||
42b2aa86 | 479 | /* Now that we know the number of arrays and size of each array, |
8342681d NF |
480 | * revalidate the size of the property read in. |
481 | */ | |
482 | if (len < (aa->n_arrays * aa->array_sz + 2) * sizeof(unsigned int)) | |
483 | return -1; | |
484 | ||
485 | aa->arrays = prop; | |
486 | return 0; | |
487 | } | |
488 | ||
489 | /* | |
490 | * This is like of_node_to_nid_single() for memory represented in the | |
491 | * ibm,dynamic-reconfiguration-memory node. | |
492 | */ | |
493 | static int of_drconf_to_nid_single(struct of_drconf_cell *drmem, | |
494 | struct assoc_arrays *aa) | |
495 | { | |
496 | int default_nid = 0; | |
497 | int nid = default_nid; | |
498 | int index; | |
499 | ||
500 | if (min_common_depth > 0 && min_common_depth <= aa->array_sz && | |
501 | !(drmem->flags & DRCONF_MEM_AI_INVALID) && | |
502 | drmem->aa_index < aa->n_arrays) { | |
503 | index = drmem->aa_index * aa->array_sz + min_common_depth - 1; | |
504 | nid = aa->arrays[index]; | |
505 | ||
506 | if (nid == 0xffff || nid >= MAX_NUMNODES) | |
507 | nid = default_nid; | |
508 | } | |
509 | ||
510 | return nid; | |
511 | } | |
512 | ||
1da177e4 LT |
513 | /* |
514 | * Figure out to which domain a cpu belongs and stick it there. | |
515 | * Return the id of the domain used. | |
516 | */ | |
2e5ce39d | 517 | static int __cpuinit numa_setup_cpu(unsigned long lcpu) |
1da177e4 | 518 | { |
cf950b7a | 519 | int nid = 0; |
8b16cd23 | 520 | struct device_node *cpu = of_get_cpu_node(lcpu, NULL); |
1da177e4 LT |
521 | |
522 | if (!cpu) { | |
523 | WARN_ON(1); | |
524 | goto out; | |
525 | } | |
526 | ||
953039c8 | 527 | nid = of_node_to_nid_single(cpu); |
1da177e4 | 528 | |
482ec7c4 | 529 | if (nid < 0 || !node_online(nid)) |
72c33688 | 530 | nid = first_online_node; |
1da177e4 | 531 | out: |
cf950b7a | 532 | map_cpu_to_node(lcpu, nid); |
1da177e4 LT |
533 | |
534 | of_node_put(cpu); | |
535 | ||
cf950b7a | 536 | return nid; |
1da177e4 LT |
537 | } |
538 | ||
74b85f37 | 539 | static int __cpuinit cpu_numa_callback(struct notifier_block *nfb, |
1da177e4 LT |
540 | unsigned long action, |
541 | void *hcpu) | |
542 | { | |
543 | unsigned long lcpu = (unsigned long)hcpu; | |
544 | int ret = NOTIFY_DONE; | |
545 | ||
546 | switch (action) { | |
547 | case CPU_UP_PREPARE: | |
8bb78442 | 548 | case CPU_UP_PREPARE_FROZEN: |
2b261227 | 549 | numa_setup_cpu(lcpu); |
1da177e4 LT |
550 | ret = NOTIFY_OK; |
551 | break; | |
552 | #ifdef CONFIG_HOTPLUG_CPU | |
553 | case CPU_DEAD: | |
8bb78442 | 554 | case CPU_DEAD_FROZEN: |
1da177e4 | 555 | case CPU_UP_CANCELED: |
8bb78442 | 556 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 LT |
557 | unmap_cpu_from_node(lcpu); |
558 | break; | |
559 | ret = NOTIFY_OK; | |
560 | #endif | |
561 | } | |
562 | return ret; | |
563 | } | |
564 | ||
565 | /* | |
566 | * Check and possibly modify a memory region to enforce the memory limit. | |
567 | * | |
568 | * Returns the size the region should have to enforce the memory limit. | |
569 | * This will either be the original value of size, a truncated value, | |
570 | * or zero. If the returned value of size is 0 the region should be | |
25985edc | 571 | * discarded as it lies wholly above the memory limit. |
1da177e4 | 572 | */ |
45fb6cea AB |
573 | static unsigned long __init numa_enforce_memory_limit(unsigned long start, |
574 | unsigned long size) | |
1da177e4 LT |
575 | { |
576 | /* | |
95f72d1e | 577 | * We use memblock_end_of_DRAM() in here instead of memory_limit because |
1da177e4 | 578 | * we've already adjusted it for the limit and it takes care of |
fe55249d MM |
579 | * having memory holes below the limit. Also, in the case of |
580 | * iommu_is_off, memory_limit is not set but is implicitly enforced. | |
1da177e4 | 581 | */ |
1da177e4 | 582 | |
95f72d1e | 583 | if (start + size <= memblock_end_of_DRAM()) |
1da177e4 LT |
584 | return size; |
585 | ||
95f72d1e | 586 | if (start >= memblock_end_of_DRAM()) |
1da177e4 LT |
587 | return 0; |
588 | ||
95f72d1e | 589 | return memblock_end_of_DRAM() - start; |
1da177e4 LT |
590 | } |
591 | ||
cf00085d C |
592 | /* |
593 | * Reads the counter for a given entry in | |
594 | * linux,drconf-usable-memory property | |
595 | */ | |
596 | static inline int __init read_usm_ranges(const u32 **usm) | |
597 | { | |
598 | /* | |
3fdfd990 | 599 | * For each lmb in ibm,dynamic-memory a corresponding |
cf00085d C |
600 | * entry in linux,drconf-usable-memory property contains |
601 | * a counter followed by that many (base, size) duple. | |
602 | * read the counter from linux,drconf-usable-memory | |
603 | */ | |
604 | return read_n_cells(n_mem_size_cells, usm); | |
605 | } | |
606 | ||
0204568a PM |
607 | /* |
608 | * Extract NUMA information from the ibm,dynamic-reconfiguration-memory | |
609 | * node. This assumes n_mem_{addr,size}_cells have been set. | |
610 | */ | |
611 | static void __init parse_drconf_memory(struct device_node *memory) | |
612 | { | |
82b2521d | 613 | const u32 *uninitialized_var(dm), *usm; |
cf00085d | 614 | unsigned int n, rc, ranges, is_kexec_kdump = 0; |
3fdfd990 | 615 | unsigned long lmb_size, base, size, sz; |
8342681d | 616 | int nid; |
aa709f3b | 617 | struct assoc_arrays aa = { .arrays = NULL }; |
8342681d NF |
618 | |
619 | n = of_get_drconf_memory(memory, &dm); | |
620 | if (!n) | |
0204568a PM |
621 | return; |
622 | ||
3fdfd990 BH |
623 | lmb_size = of_get_lmb_size(memory); |
624 | if (!lmb_size) | |
8342681d NF |
625 | return; |
626 | ||
627 | rc = of_get_assoc_arrays(memory, &aa); | |
628 | if (rc) | |
0204568a PM |
629 | return; |
630 | ||
cf00085d C |
631 | /* check if this is a kexec/kdump kernel */ |
632 | usm = of_get_usable_memory(memory); | |
633 | if (usm != NULL) | |
634 | is_kexec_kdump = 1; | |
635 | ||
0204568a | 636 | for (; n != 0; --n) { |
8342681d NF |
637 | struct of_drconf_cell drmem; |
638 | ||
639 | read_drconf_cell(&drmem, &dm); | |
640 | ||
641 | /* skip this block if the reserved bit is set in flags (0x80) | |
642 | or if the block is not assigned to this partition (0x8) */ | |
643 | if ((drmem.flags & DRCONF_MEM_RESERVED) | |
644 | || !(drmem.flags & DRCONF_MEM_ASSIGNED)) | |
0204568a | 645 | continue; |
1daa6d08 | 646 | |
cf00085d | 647 | base = drmem.base_addr; |
3fdfd990 | 648 | size = lmb_size; |
cf00085d | 649 | ranges = 1; |
8342681d | 650 | |
cf00085d C |
651 | if (is_kexec_kdump) { |
652 | ranges = read_usm_ranges(&usm); | |
653 | if (!ranges) /* there are no (base, size) duple */ | |
654 | continue; | |
655 | } | |
656 | do { | |
657 | if (is_kexec_kdump) { | |
658 | base = read_n_cells(n_mem_addr_cells, &usm); | |
659 | size = read_n_cells(n_mem_size_cells, &usm); | |
660 | } | |
661 | nid = of_drconf_to_nid_single(&drmem, &aa); | |
662 | fake_numa_create_new_node( | |
663 | ((base + size) >> PAGE_SHIFT), | |
8342681d | 664 | &nid); |
cf00085d C |
665 | node_set_online(nid); |
666 | sz = numa_enforce_memory_limit(base, size); | |
667 | if (sz) | |
1d7cfe18 | 668 | memblock_set_node(base, sz, nid); |
cf00085d | 669 | } while (--ranges); |
0204568a PM |
670 | } |
671 | } | |
672 | ||
1da177e4 LT |
673 | static int __init parse_numa_properties(void) |
674 | { | |
94db7c5e | 675 | struct device_node *memory; |
482ec7c4 | 676 | int default_nid = 0; |
1da177e4 LT |
677 | unsigned long i; |
678 | ||
679 | if (numa_enabled == 0) { | |
680 | printk(KERN_WARNING "NUMA disabled by user\n"); | |
681 | return -1; | |
682 | } | |
683 | ||
1da177e4 LT |
684 | min_common_depth = find_min_common_depth(); |
685 | ||
1da177e4 LT |
686 | if (min_common_depth < 0) |
687 | return min_common_depth; | |
688 | ||
bf4b85b0 NL |
689 | dbg("NUMA associativity depth for CPU/Memory: %d\n", min_common_depth); |
690 | ||
1da177e4 | 691 | /* |
482ec7c4 NL |
692 | * Even though we connect cpus to numa domains later in SMP |
693 | * init, we need to know the node ids now. This is because | |
694 | * each node to be onlined must have NODE_DATA etc backing it. | |
1da177e4 | 695 | */ |
482ec7c4 | 696 | for_each_present_cpu(i) { |
dfbe93a2 | 697 | struct device_node *cpu; |
cf950b7a | 698 | int nid; |
1da177e4 | 699 | |
8b16cd23 | 700 | cpu = of_get_cpu_node(i, NULL); |
482ec7c4 | 701 | BUG_ON(!cpu); |
953039c8 | 702 | nid = of_node_to_nid_single(cpu); |
482ec7c4 | 703 | of_node_put(cpu); |
1da177e4 | 704 | |
482ec7c4 NL |
705 | /* |
706 | * Don't fall back to default_nid yet -- we will plug | |
707 | * cpus into nodes once the memory scan has discovered | |
708 | * the topology. | |
709 | */ | |
710 | if (nid < 0) | |
711 | continue; | |
712 | node_set_online(nid); | |
1da177e4 LT |
713 | } |
714 | ||
237a0989 | 715 | get_n_mem_cells(&n_mem_addr_cells, &n_mem_size_cells); |
94db7c5e AB |
716 | |
717 | for_each_node_by_type(memory, "memory") { | |
1da177e4 LT |
718 | unsigned long start; |
719 | unsigned long size; | |
cf950b7a | 720 | int nid; |
1da177e4 | 721 | int ranges; |
a7f67bdf | 722 | const unsigned int *memcell_buf; |
1da177e4 LT |
723 | unsigned int len; |
724 | ||
e2eb6392 | 725 | memcell_buf = of_get_property(memory, |
ba759485 ME |
726 | "linux,usable-memory", &len); |
727 | if (!memcell_buf || len <= 0) | |
e2eb6392 | 728 | memcell_buf = of_get_property(memory, "reg", &len); |
1da177e4 LT |
729 | if (!memcell_buf || len <= 0) |
730 | continue; | |
731 | ||
cc5d0189 BH |
732 | /* ranges in cell */ |
733 | ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells); | |
1da177e4 LT |
734 | new_range: |
735 | /* these are order-sensitive, and modify the buffer pointer */ | |
237a0989 MK |
736 | start = read_n_cells(n_mem_addr_cells, &memcell_buf); |
737 | size = read_n_cells(n_mem_size_cells, &memcell_buf); | |
1da177e4 | 738 | |
482ec7c4 NL |
739 | /* |
740 | * Assumption: either all memory nodes or none will | |
741 | * have associativity properties. If none, then | |
742 | * everything goes to default_nid. | |
743 | */ | |
953039c8 | 744 | nid = of_node_to_nid_single(memory); |
482ec7c4 NL |
745 | if (nid < 0) |
746 | nid = default_nid; | |
1daa6d08 BS |
747 | |
748 | fake_numa_create_new_node(((start + size) >> PAGE_SHIFT), &nid); | |
482ec7c4 | 749 | node_set_online(nid); |
1da177e4 | 750 | |
45fb6cea | 751 | if (!(size = numa_enforce_memory_limit(start, size))) { |
1da177e4 LT |
752 | if (--ranges) |
753 | goto new_range; | |
754 | else | |
755 | continue; | |
756 | } | |
757 | ||
1d7cfe18 | 758 | memblock_set_node(start, size, nid); |
1da177e4 LT |
759 | |
760 | if (--ranges) | |
761 | goto new_range; | |
762 | } | |
763 | ||
0204568a | 764 | /* |
dfbe93a2 AB |
765 | * Now do the same thing for each MEMBLOCK listed in the |
766 | * ibm,dynamic-memory property in the | |
767 | * ibm,dynamic-reconfiguration-memory node. | |
0204568a PM |
768 | */ |
769 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); | |
770 | if (memory) | |
771 | parse_drconf_memory(memory); | |
772 | ||
1da177e4 LT |
773 | return 0; |
774 | } | |
775 | ||
776 | static void __init setup_nonnuma(void) | |
777 | { | |
95f72d1e YL |
778 | unsigned long top_of_ram = memblock_end_of_DRAM(); |
779 | unsigned long total_ram = memblock_phys_mem_size(); | |
c67c3cb4 | 780 | unsigned long start_pfn, end_pfn; |
28be7072 BH |
781 | unsigned int nid = 0; |
782 | struct memblock_region *reg; | |
1da177e4 | 783 | |
e110b281 | 784 | printk(KERN_DEBUG "Top of RAM: 0x%lx, Total RAM: 0x%lx\n", |
1da177e4 | 785 | top_of_ram, total_ram); |
e110b281 | 786 | printk(KERN_DEBUG "Memory hole size: %ldMB\n", |
1da177e4 LT |
787 | (top_of_ram - total_ram) >> 20); |
788 | ||
28be7072 | 789 | for_each_memblock(memory, reg) { |
c7fc2de0 YL |
790 | start_pfn = memblock_region_memory_base_pfn(reg); |
791 | end_pfn = memblock_region_memory_end_pfn(reg); | |
1daa6d08 BS |
792 | |
793 | fake_numa_create_new_node(end_pfn, &nid); | |
1d7cfe18 TH |
794 | memblock_set_node(PFN_PHYS(start_pfn), |
795 | PFN_PHYS(end_pfn - start_pfn), nid); | |
1daa6d08 | 796 | node_set_online(nid); |
c67c3cb4 | 797 | } |
1da177e4 LT |
798 | } |
799 | ||
4b703a23 AB |
800 | void __init dump_numa_cpu_topology(void) |
801 | { | |
802 | unsigned int node; | |
803 | unsigned int cpu, count; | |
804 | ||
805 | if (min_common_depth == -1 || !numa_enabled) | |
806 | return; | |
807 | ||
808 | for_each_online_node(node) { | |
e110b281 | 809 | printk(KERN_DEBUG "Node %d CPUs:", node); |
4b703a23 AB |
810 | |
811 | count = 0; | |
812 | /* | |
813 | * If we used a CPU iterator here we would miss printing | |
814 | * the holes in the cpumap. | |
815 | */ | |
25863de0 AB |
816 | for (cpu = 0; cpu < nr_cpu_ids; cpu++) { |
817 | if (cpumask_test_cpu(cpu, | |
818 | node_to_cpumask_map[node])) { | |
4b703a23 AB |
819 | if (count == 0) |
820 | printk(" %u", cpu); | |
821 | ++count; | |
822 | } else { | |
823 | if (count > 1) | |
824 | printk("-%u", cpu - 1); | |
825 | count = 0; | |
826 | } | |
827 | } | |
828 | ||
829 | if (count > 1) | |
25863de0 | 830 | printk("-%u", nr_cpu_ids - 1); |
4b703a23 AB |
831 | printk("\n"); |
832 | } | |
833 | } | |
834 | ||
835 | static void __init dump_numa_memory_topology(void) | |
1da177e4 LT |
836 | { |
837 | unsigned int node; | |
838 | unsigned int count; | |
839 | ||
840 | if (min_common_depth == -1 || !numa_enabled) | |
841 | return; | |
842 | ||
843 | for_each_online_node(node) { | |
844 | unsigned long i; | |
845 | ||
e110b281 | 846 | printk(KERN_DEBUG "Node %d Memory:", node); |
1da177e4 LT |
847 | |
848 | count = 0; | |
849 | ||
95f72d1e | 850 | for (i = 0; i < memblock_end_of_DRAM(); |
45fb6cea AB |
851 | i += (1 << SECTION_SIZE_BITS)) { |
852 | if (early_pfn_to_nid(i >> PAGE_SHIFT) == node) { | |
1da177e4 LT |
853 | if (count == 0) |
854 | printk(" 0x%lx", i); | |
855 | ++count; | |
856 | } else { | |
857 | if (count > 0) | |
858 | printk("-0x%lx", i); | |
859 | count = 0; | |
860 | } | |
861 | } | |
862 | ||
863 | if (count > 0) | |
864 | printk("-0x%lx", i); | |
865 | printk("\n"); | |
866 | } | |
1da177e4 LT |
867 | } |
868 | ||
869 | /* | |
95f72d1e | 870 | * Allocate some memory, satisfying the memblock or bootmem allocator where |
1da177e4 LT |
871 | * required. nid is the preferred node and end is the physical address of |
872 | * the highest address in the node. | |
873 | * | |
0be210fd | 874 | * Returns the virtual address of the memory. |
1da177e4 | 875 | */ |
893473df | 876 | static void __init *careful_zallocation(int nid, unsigned long size, |
45fb6cea AB |
877 | unsigned long align, |
878 | unsigned long end_pfn) | |
1da177e4 | 879 | { |
0be210fd | 880 | void *ret; |
45fb6cea | 881 | int new_nid; |
0be210fd DH |
882 | unsigned long ret_paddr; |
883 | ||
95f72d1e | 884 | ret_paddr = __memblock_alloc_base(size, align, end_pfn << PAGE_SHIFT); |
1da177e4 LT |
885 | |
886 | /* retry over all memory */ | |
0be210fd | 887 | if (!ret_paddr) |
95f72d1e | 888 | ret_paddr = __memblock_alloc_base(size, align, memblock_end_of_DRAM()); |
1da177e4 | 889 | |
0be210fd | 890 | if (!ret_paddr) |
5d21ea2b | 891 | panic("numa.c: cannot allocate %lu bytes for node %d", |
1da177e4 LT |
892 | size, nid); |
893 | ||
0be210fd DH |
894 | ret = __va(ret_paddr); |
895 | ||
1da177e4 | 896 | /* |
c555e520 | 897 | * We initialize the nodes in numeric order: 0, 1, 2... |
95f72d1e | 898 | * and hand over control from the MEMBLOCK allocator to the |
c555e520 DH |
899 | * bootmem allocator. If this function is called for |
900 | * node 5, then we know that all nodes <5 are using the | |
95f72d1e | 901 | * bootmem allocator instead of the MEMBLOCK allocator. |
c555e520 DH |
902 | * |
903 | * So, check the nid from which this allocation came | |
904 | * and double check to see if we need to use bootmem | |
95f72d1e | 905 | * instead of the MEMBLOCK. We don't free the MEMBLOCK memory |
c555e520 | 906 | * since it would be useless. |
1da177e4 | 907 | */ |
0be210fd | 908 | new_nid = early_pfn_to_nid(ret_paddr >> PAGE_SHIFT); |
45fb6cea | 909 | if (new_nid < nid) { |
0be210fd | 910 | ret = __alloc_bootmem_node(NODE_DATA(new_nid), |
1da177e4 LT |
911 | size, align, 0); |
912 | ||
0be210fd | 913 | dbg("alloc_bootmem %p %lx\n", ret, size); |
1da177e4 LT |
914 | } |
915 | ||
893473df | 916 | memset(ret, 0, size); |
0be210fd | 917 | return ret; |
1da177e4 LT |
918 | } |
919 | ||
74b85f37 CS |
920 | static struct notifier_block __cpuinitdata ppc64_numa_nb = { |
921 | .notifier_call = cpu_numa_callback, | |
922 | .priority = 1 /* Must run before sched domains notifier. */ | |
923 | }; | |
924 | ||
28e86bdb | 925 | static void __init mark_reserved_regions_for_nid(int nid) |
4a618669 DH |
926 | { |
927 | struct pglist_data *node = NODE_DATA(nid); | |
28be7072 | 928 | struct memblock_region *reg; |
4a618669 | 929 | |
28be7072 BH |
930 | for_each_memblock(reserved, reg) { |
931 | unsigned long physbase = reg->base; | |
932 | unsigned long size = reg->size; | |
4a618669 | 933 | unsigned long start_pfn = physbase >> PAGE_SHIFT; |
06eccea6 | 934 | unsigned long end_pfn = PFN_UP(physbase + size); |
4a618669 DH |
935 | struct node_active_region node_ar; |
936 | unsigned long node_end_pfn = node->node_start_pfn + | |
937 | node->node_spanned_pages; | |
938 | ||
939 | /* | |
95f72d1e | 940 | * Check to make sure that this memblock.reserved area is |
4a618669 DH |
941 | * within the bounds of the node that we care about. |
942 | * Checking the nid of the start and end points is not | |
943 | * sufficient because the reserved area could span the | |
944 | * entire node. | |
945 | */ | |
946 | if (end_pfn <= node->node_start_pfn || | |
947 | start_pfn >= node_end_pfn) | |
948 | continue; | |
949 | ||
950 | get_node_active_region(start_pfn, &node_ar); | |
951 | while (start_pfn < end_pfn && | |
952 | node_ar.start_pfn < node_ar.end_pfn) { | |
953 | unsigned long reserve_size = size; | |
954 | /* | |
955 | * if reserved region extends past active region | |
956 | * then trim size to active region | |
957 | */ | |
958 | if (end_pfn > node_ar.end_pfn) | |
959 | reserve_size = (node_ar.end_pfn << PAGE_SHIFT) | |
06eccea6 | 960 | - physbase; |
a4c74ddd DH |
961 | /* |
962 | * Only worry about *this* node, others may not | |
963 | * yet have valid NODE_DATA(). | |
964 | */ | |
965 | if (node_ar.nid == nid) { | |
966 | dbg("reserve_bootmem %lx %lx nid=%d\n", | |
967 | physbase, reserve_size, node_ar.nid); | |
968 | reserve_bootmem_node(NODE_DATA(node_ar.nid), | |
969 | physbase, reserve_size, | |
970 | BOOTMEM_DEFAULT); | |
971 | } | |
4a618669 DH |
972 | /* |
973 | * if reserved region is contained in the active region | |
974 | * then done. | |
975 | */ | |
976 | if (end_pfn <= node_ar.end_pfn) | |
977 | break; | |
978 | ||
979 | /* | |
980 | * reserved region extends past the active region | |
981 | * get next active region that contains this | |
982 | * reserved region | |
983 | */ | |
984 | start_pfn = node_ar.end_pfn; | |
985 | physbase = start_pfn << PAGE_SHIFT; | |
986 | size = size - reserve_size; | |
987 | get_node_active_region(start_pfn, &node_ar); | |
988 | } | |
989 | } | |
990 | } | |
991 | ||
992 | ||
1da177e4 LT |
993 | void __init do_init_bootmem(void) |
994 | { | |
995 | int nid; | |
1da177e4 LT |
996 | |
997 | min_low_pfn = 0; | |
95f72d1e | 998 | max_low_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT; |
1da177e4 LT |
999 | max_pfn = max_low_pfn; |
1000 | ||
1001 | if (parse_numa_properties()) | |
1002 | setup_nonnuma(); | |
1003 | else | |
4b703a23 | 1004 | dump_numa_memory_topology(); |
1da177e4 | 1005 | |
1da177e4 | 1006 | for_each_online_node(nid) { |
c67c3cb4 | 1007 | unsigned long start_pfn, end_pfn; |
0be210fd | 1008 | void *bootmem_vaddr; |
1da177e4 LT |
1009 | unsigned long bootmap_pages; |
1010 | ||
c67c3cb4 | 1011 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); |
1da177e4 | 1012 | |
4a618669 DH |
1013 | /* |
1014 | * Allocate the node structure node local if possible | |
1015 | * | |
1016 | * Be careful moving this around, as it relies on all | |
1017 | * previous nodes' bootmem to be initialized and have | |
1018 | * all reserved areas marked. | |
1019 | */ | |
893473df | 1020 | NODE_DATA(nid) = careful_zallocation(nid, |
1da177e4 | 1021 | sizeof(struct pglist_data), |
45fb6cea | 1022 | SMP_CACHE_BYTES, end_pfn); |
1da177e4 LT |
1023 | |
1024 | dbg("node %d\n", nid); | |
1025 | dbg("NODE_DATA() = %p\n", NODE_DATA(nid)); | |
1026 | ||
b61bfa3c | 1027 | NODE_DATA(nid)->bdata = &bootmem_node_data[nid]; |
45fb6cea AB |
1028 | NODE_DATA(nid)->node_start_pfn = start_pfn; |
1029 | NODE_DATA(nid)->node_spanned_pages = end_pfn - start_pfn; | |
1da177e4 LT |
1030 | |
1031 | if (NODE_DATA(nid)->node_spanned_pages == 0) | |
1032 | continue; | |
1033 | ||
45fb6cea AB |
1034 | dbg("start_paddr = %lx\n", start_pfn << PAGE_SHIFT); |
1035 | dbg("end_paddr = %lx\n", end_pfn << PAGE_SHIFT); | |
1da177e4 | 1036 | |
45fb6cea | 1037 | bootmap_pages = bootmem_bootmap_pages(end_pfn - start_pfn); |
893473df | 1038 | bootmem_vaddr = careful_zallocation(nid, |
45fb6cea AB |
1039 | bootmap_pages << PAGE_SHIFT, |
1040 | PAGE_SIZE, end_pfn); | |
1da177e4 | 1041 | |
0be210fd | 1042 | dbg("bootmap_vaddr = %p\n", bootmem_vaddr); |
1da177e4 | 1043 | |
0be210fd DH |
1044 | init_bootmem_node(NODE_DATA(nid), |
1045 | __pa(bootmem_vaddr) >> PAGE_SHIFT, | |
45fb6cea | 1046 | start_pfn, end_pfn); |
1da177e4 | 1047 | |
c67c3cb4 | 1048 | free_bootmem_with_active_regions(nid, end_pfn); |
4a618669 DH |
1049 | /* |
1050 | * Be very careful about moving this around. Future | |
893473df | 1051 | * calls to careful_zallocation() depend on this getting |
4a618669 DH |
1052 | * done correctly. |
1053 | */ | |
1054 | mark_reserved_regions_for_nid(nid); | |
8f64e1f2 | 1055 | sparse_memory_present_with_active_regions(nid); |
4a618669 | 1056 | } |
d3f6204a BH |
1057 | |
1058 | init_bootmem_done = 1; | |
25863de0 AB |
1059 | |
1060 | /* | |
1061 | * Now bootmem is initialised we can create the node to cpumask | |
1062 | * lookup tables and setup the cpu callback to populate them. | |
1063 | */ | |
1064 | setup_node_to_cpumask_map(); | |
1065 | ||
1066 | register_cpu_notifier(&ppc64_numa_nb); | |
1067 | cpu_numa_callback(&ppc64_numa_nb, CPU_UP_PREPARE, | |
1068 | (void *)(unsigned long)boot_cpuid); | |
1da177e4 LT |
1069 | } |
1070 | ||
1071 | void __init paging_init(void) | |
1072 | { | |
6391af17 MG |
1073 | unsigned long max_zone_pfns[MAX_NR_ZONES]; |
1074 | memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); | |
95f72d1e | 1075 | max_zone_pfns[ZONE_DMA] = memblock_end_of_DRAM() >> PAGE_SHIFT; |
c67c3cb4 | 1076 | free_area_init_nodes(max_zone_pfns); |
1da177e4 LT |
1077 | } |
1078 | ||
1079 | static int __init early_numa(char *p) | |
1080 | { | |
1081 | if (!p) | |
1082 | return 0; | |
1083 | ||
1084 | if (strstr(p, "off")) | |
1085 | numa_enabled = 0; | |
1086 | ||
1087 | if (strstr(p, "debug")) | |
1088 | numa_debug = 1; | |
1089 | ||
1daa6d08 BS |
1090 | p = strstr(p, "fake="); |
1091 | if (p) | |
1092 | cmdline = p + strlen("fake="); | |
1093 | ||
1da177e4 LT |
1094 | return 0; |
1095 | } | |
1096 | early_param("numa", early_numa); | |
237a0989 MK |
1097 | |
1098 | #ifdef CONFIG_MEMORY_HOTPLUG | |
0db9360a | 1099 | /* |
0f16ef7f NF |
1100 | * Find the node associated with a hot added memory section for |
1101 | * memory represented in the device tree by the property | |
1102 | * ibm,dynamic-reconfiguration-memory/ibm,dynamic-memory. | |
0db9360a NF |
1103 | */ |
1104 | static int hot_add_drconf_scn_to_nid(struct device_node *memory, | |
1105 | unsigned long scn_addr) | |
1106 | { | |
1107 | const u32 *dm; | |
0f16ef7f | 1108 | unsigned int drconf_cell_cnt, rc; |
3fdfd990 | 1109 | unsigned long lmb_size; |
0db9360a | 1110 | struct assoc_arrays aa; |
0f16ef7f | 1111 | int nid = -1; |
0db9360a | 1112 | |
0f16ef7f NF |
1113 | drconf_cell_cnt = of_get_drconf_memory(memory, &dm); |
1114 | if (!drconf_cell_cnt) | |
1115 | return -1; | |
0db9360a | 1116 | |
3fdfd990 BH |
1117 | lmb_size = of_get_lmb_size(memory); |
1118 | if (!lmb_size) | |
0f16ef7f | 1119 | return -1; |
0db9360a NF |
1120 | |
1121 | rc = of_get_assoc_arrays(memory, &aa); | |
1122 | if (rc) | |
0f16ef7f | 1123 | return -1; |
0db9360a | 1124 | |
0f16ef7f | 1125 | for (; drconf_cell_cnt != 0; --drconf_cell_cnt) { |
0db9360a NF |
1126 | struct of_drconf_cell drmem; |
1127 | ||
1128 | read_drconf_cell(&drmem, &dm); | |
1129 | ||
1130 | /* skip this block if it is reserved or not assigned to | |
1131 | * this partition */ | |
1132 | if ((drmem.flags & DRCONF_MEM_RESERVED) | |
1133 | || !(drmem.flags & DRCONF_MEM_ASSIGNED)) | |
1134 | continue; | |
1135 | ||
0f16ef7f | 1136 | if ((scn_addr < drmem.base_addr) |
3fdfd990 | 1137 | || (scn_addr >= (drmem.base_addr + lmb_size))) |
0f16ef7f NF |
1138 | continue; |
1139 | ||
0db9360a | 1140 | nid = of_drconf_to_nid_single(&drmem, &aa); |
0f16ef7f NF |
1141 | break; |
1142 | } | |
1143 | ||
1144 | return nid; | |
1145 | } | |
1146 | ||
1147 | /* | |
1148 | * Find the node associated with a hot added memory section for memory | |
1149 | * represented in the device tree as a node (i.e. memory@XXXX) for | |
95f72d1e | 1150 | * each memblock. |
0f16ef7f NF |
1151 | */ |
1152 | int hot_add_node_scn_to_nid(unsigned long scn_addr) | |
1153 | { | |
94db7c5e | 1154 | struct device_node *memory; |
0f16ef7f NF |
1155 | int nid = -1; |
1156 | ||
94db7c5e | 1157 | for_each_node_by_type(memory, "memory") { |
0f16ef7f NF |
1158 | unsigned long start, size; |
1159 | int ranges; | |
1160 | const unsigned int *memcell_buf; | |
1161 | unsigned int len; | |
1162 | ||
1163 | memcell_buf = of_get_property(memory, "reg", &len); | |
1164 | if (!memcell_buf || len <= 0) | |
1165 | continue; | |
1166 | ||
1167 | /* ranges in cell */ | |
1168 | ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells); | |
1169 | ||
1170 | while (ranges--) { | |
1171 | start = read_n_cells(n_mem_addr_cells, &memcell_buf); | |
1172 | size = read_n_cells(n_mem_size_cells, &memcell_buf); | |
1173 | ||
1174 | if ((scn_addr < start) || (scn_addr >= (start + size))) | |
1175 | continue; | |
1176 | ||
1177 | nid = of_node_to_nid_single(memory); | |
1178 | break; | |
1179 | } | |
0db9360a | 1180 | |
0f16ef7f NF |
1181 | if (nid >= 0) |
1182 | break; | |
0db9360a NF |
1183 | } |
1184 | ||
60831842 AB |
1185 | of_node_put(memory); |
1186 | ||
0f16ef7f | 1187 | return nid; |
0db9360a NF |
1188 | } |
1189 | ||
237a0989 MK |
1190 | /* |
1191 | * Find the node associated with a hot added memory section. Section | |
95f72d1e YL |
1192 | * corresponds to a SPARSEMEM section, not an MEMBLOCK. It is assumed that |
1193 | * sections are fully contained within a single MEMBLOCK. | |
237a0989 MK |
1194 | */ |
1195 | int hot_add_scn_to_nid(unsigned long scn_addr) | |
1196 | { | |
1197 | struct device_node *memory = NULL; | |
0f16ef7f | 1198 | int nid, found = 0; |
237a0989 MK |
1199 | |
1200 | if (!numa_enabled || (min_common_depth < 0)) | |
72c33688 | 1201 | return first_online_node; |
0db9360a NF |
1202 | |
1203 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); | |
1204 | if (memory) { | |
1205 | nid = hot_add_drconf_scn_to_nid(memory, scn_addr); | |
1206 | of_node_put(memory); | |
0f16ef7f NF |
1207 | } else { |
1208 | nid = hot_add_node_scn_to_nid(scn_addr); | |
0db9360a | 1209 | } |
237a0989 | 1210 | |
0f16ef7f | 1211 | if (nid < 0 || !node_online(nid)) |
72c33688 | 1212 | nid = first_online_node; |
237a0989 | 1213 | |
0f16ef7f NF |
1214 | if (NODE_DATA(nid)->node_spanned_pages) |
1215 | return nid; | |
237a0989 | 1216 | |
0f16ef7f NF |
1217 | for_each_online_node(nid) { |
1218 | if (NODE_DATA(nid)->node_spanned_pages) { | |
1219 | found = 1; | |
1220 | break; | |
237a0989 | 1221 | } |
237a0989 | 1222 | } |
0f16ef7f NF |
1223 | |
1224 | BUG_ON(!found); | |
1225 | return nid; | |
237a0989 | 1226 | } |
0f16ef7f | 1227 | |
cd34206e NA |
1228 | static u64 hot_add_drconf_memory_max(void) |
1229 | { | |
1230 | struct device_node *memory = NULL; | |
1231 | unsigned int drconf_cell_cnt = 0; | |
1232 | u64 lmb_size = 0; | |
1233 | const u32 *dm = 0; | |
1234 | ||
1235 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); | |
1236 | if (memory) { | |
1237 | drconf_cell_cnt = of_get_drconf_memory(memory, &dm); | |
1238 | lmb_size = of_get_lmb_size(memory); | |
1239 | of_node_put(memory); | |
1240 | } | |
1241 | return lmb_size * drconf_cell_cnt; | |
1242 | } | |
1243 | ||
1244 | /* | |
1245 | * memory_hotplug_max - return max address of memory that may be added | |
1246 | * | |
1247 | * This is currently only used on systems that support drconfig memory | |
1248 | * hotplug. | |
1249 | */ | |
1250 | u64 memory_hotplug_max(void) | |
1251 | { | |
1252 | return max(hot_add_drconf_memory_max(), memblock_end_of_DRAM()); | |
1253 | } | |
237a0989 | 1254 | #endif /* CONFIG_MEMORY_HOTPLUG */ |
9eff1a38 | 1255 | |
bd03403a | 1256 | /* Virtual Processor Home Node (VPHN) support */ |
39bf990e | 1257 | #ifdef CONFIG_PPC_SPLPAR |
30c05350 NF |
1258 | struct topology_update_data { |
1259 | struct topology_update_data *next; | |
1260 | unsigned int cpu; | |
1261 | int old_nid; | |
1262 | int new_nid; | |
1263 | }; | |
1264 | ||
5de16699 | 1265 | static u8 vphn_cpu_change_counts[NR_CPUS][MAX_DISTANCE_REF_POINTS]; |
9eff1a38 JL |
1266 | static cpumask_t cpu_associativity_changes_mask; |
1267 | static int vphn_enabled; | |
5d88aa85 JL |
1268 | static int prrn_enabled; |
1269 | static void reset_topology_timer(void); | |
9eff1a38 JL |
1270 | |
1271 | /* | |
1272 | * Store the current values of the associativity change counters in the | |
1273 | * hypervisor. | |
1274 | */ | |
1275 | static void setup_cpu_associativity_change_counters(void) | |
1276 | { | |
cd9d6cc7 | 1277 | int cpu; |
9eff1a38 | 1278 | |
5de16699 AB |
1279 | /* The VPHN feature supports a maximum of 8 reference points */ |
1280 | BUILD_BUG_ON(MAX_DISTANCE_REF_POINTS > 8); | |
1281 | ||
9eff1a38 | 1282 | for_each_possible_cpu(cpu) { |
cd9d6cc7 | 1283 | int i; |
9eff1a38 JL |
1284 | u8 *counts = vphn_cpu_change_counts[cpu]; |
1285 | volatile u8 *hypervisor_counts = lppaca[cpu].vphn_assoc_counts; | |
1286 | ||
5de16699 | 1287 | for (i = 0; i < distance_ref_points_depth; i++) |
9eff1a38 | 1288 | counts[i] = hypervisor_counts[i]; |
9eff1a38 JL |
1289 | } |
1290 | } | |
1291 | ||
1292 | /* | |
1293 | * The hypervisor maintains a set of 8 associativity change counters in | |
1294 | * the VPA of each cpu that correspond to the associativity levels in the | |
1295 | * ibm,associativity-reference-points property. When an associativity | |
1296 | * level changes, the corresponding counter is incremented. | |
1297 | * | |
1298 | * Set a bit in cpu_associativity_changes_mask for each cpu whose home | |
1299 | * node associativity levels have changed. | |
1300 | * | |
1301 | * Returns the number of cpus with unhandled associativity changes. | |
1302 | */ | |
1303 | static int update_cpu_associativity_changes_mask(void) | |
1304 | { | |
5d88aa85 | 1305 | int cpu; |
9eff1a38 JL |
1306 | cpumask_t *changes = &cpu_associativity_changes_mask; |
1307 | ||
9eff1a38 JL |
1308 | for_each_possible_cpu(cpu) { |
1309 | int i, changed = 0; | |
1310 | u8 *counts = vphn_cpu_change_counts[cpu]; | |
1311 | volatile u8 *hypervisor_counts = lppaca[cpu].vphn_assoc_counts; | |
1312 | ||
5de16699 | 1313 | for (i = 0; i < distance_ref_points_depth; i++) { |
d69043e8 | 1314 | if (hypervisor_counts[i] != counts[i]) { |
9eff1a38 JL |
1315 | counts[i] = hypervisor_counts[i]; |
1316 | changed = 1; | |
1317 | } | |
1318 | } | |
1319 | if (changed) { | |
1320 | cpumask_set_cpu(cpu, changes); | |
9eff1a38 JL |
1321 | } |
1322 | } | |
1323 | ||
5d88aa85 | 1324 | return cpumask_weight(changes); |
9eff1a38 JL |
1325 | } |
1326 | ||
c0e5e46f AB |
1327 | /* |
1328 | * 6 64-bit registers unpacked into 12 32-bit associativity values. To form | |
1329 | * the complete property we have to add the length in the first cell. | |
1330 | */ | |
1331 | #define VPHN_ASSOC_BUFSIZE (6*sizeof(u64)/sizeof(u32) + 1) | |
9eff1a38 JL |
1332 | |
1333 | /* | |
1334 | * Convert the associativity domain numbers returned from the hypervisor | |
1335 | * to the sequence they would appear in the ibm,associativity property. | |
1336 | */ | |
1337 | static int vphn_unpack_associativity(const long *packed, unsigned int *unpacked) | |
1338 | { | |
cd9d6cc7 | 1339 | int i, nr_assoc_doms = 0; |
9eff1a38 JL |
1340 | const u16 *field = (const u16*) packed; |
1341 | ||
1342 | #define VPHN_FIELD_UNUSED (0xffff) | |
1343 | #define VPHN_FIELD_MSB (0x8000) | |
1344 | #define VPHN_FIELD_MASK (~VPHN_FIELD_MSB) | |
1345 | ||
c0e5e46f | 1346 | for (i = 1; i < VPHN_ASSOC_BUFSIZE; i++) { |
9eff1a38 JL |
1347 | if (*field == VPHN_FIELD_UNUSED) { |
1348 | /* All significant fields processed, and remaining | |
1349 | * fields contain the reserved value of all 1's. | |
1350 | * Just store them. | |
1351 | */ | |
1352 | unpacked[i] = *((u32*)field); | |
1353 | field += 2; | |
7639adaa | 1354 | } else if (*field & VPHN_FIELD_MSB) { |
9eff1a38 JL |
1355 | /* Data is in the lower 15 bits of this field */ |
1356 | unpacked[i] = *field & VPHN_FIELD_MASK; | |
1357 | field++; | |
1358 | nr_assoc_doms++; | |
7639adaa | 1359 | } else { |
9eff1a38 JL |
1360 | /* Data is in the lower 15 bits of this field |
1361 | * concatenated with the next 16 bit field | |
1362 | */ | |
1363 | unpacked[i] = *((u32*)field); | |
1364 | field += 2; | |
1365 | nr_assoc_doms++; | |
1366 | } | |
1367 | } | |
1368 | ||
c0e5e46f AB |
1369 | /* The first cell contains the length of the property */ |
1370 | unpacked[0] = nr_assoc_doms; | |
1371 | ||
9eff1a38 JL |
1372 | return nr_assoc_doms; |
1373 | } | |
1374 | ||
1375 | /* | |
1376 | * Retrieve the new associativity information for a virtual processor's | |
1377 | * home node. | |
1378 | */ | |
1379 | static long hcall_vphn(unsigned long cpu, unsigned int *associativity) | |
1380 | { | |
cd9d6cc7 | 1381 | long rc; |
9eff1a38 JL |
1382 | long retbuf[PLPAR_HCALL9_BUFSIZE] = {0}; |
1383 | u64 flags = 1; | |
1384 | int hwcpu = get_hard_smp_processor_id(cpu); | |
1385 | ||
1386 | rc = plpar_hcall9(H_HOME_NODE_ASSOCIATIVITY, retbuf, flags, hwcpu); | |
1387 | vphn_unpack_associativity(retbuf, associativity); | |
1388 | ||
1389 | return rc; | |
1390 | } | |
1391 | ||
1392 | static long vphn_get_associativity(unsigned long cpu, | |
1393 | unsigned int *associativity) | |
1394 | { | |
cd9d6cc7 | 1395 | long rc; |
9eff1a38 JL |
1396 | |
1397 | rc = hcall_vphn(cpu, associativity); | |
1398 | ||
1399 | switch (rc) { | |
1400 | case H_FUNCTION: | |
1401 | printk(KERN_INFO | |
1402 | "VPHN is not supported. Disabling polling...\n"); | |
1403 | stop_topology_update(); | |
1404 | break; | |
1405 | case H_HARDWARE: | |
1406 | printk(KERN_ERR | |
1407 | "hcall_vphn() experienced a hardware fault " | |
1408 | "preventing VPHN. Disabling polling...\n"); | |
1409 | stop_topology_update(); | |
1410 | } | |
1411 | ||
1412 | return rc; | |
1413 | } | |
1414 | ||
30c05350 NF |
1415 | /* |
1416 | * Update the CPU maps and sysfs entries for a single CPU when its NUMA | |
1417 | * characteristics change. This function doesn't perform any locking and is | |
1418 | * only safe to call from stop_machine(). | |
1419 | */ | |
1420 | static int update_cpu_topology(void *data) | |
1421 | { | |
1422 | struct topology_update_data *update; | |
1423 | unsigned long cpu; | |
1424 | ||
1425 | if (!data) | |
1426 | return -EINVAL; | |
1427 | ||
1428 | cpu = get_cpu(); | |
1429 | ||
1430 | for (update = data; update; update = update->next) { | |
1431 | if (cpu != update->cpu) | |
1432 | continue; | |
1433 | ||
1434 | unregister_cpu_under_node(update->cpu, update->old_nid); | |
1435 | unmap_cpu_from_node(update->cpu); | |
1436 | map_cpu_to_node(update->cpu, update->new_nid); | |
1437 | register_cpu_under_node(update->cpu, update->new_nid); | |
1438 | } | |
1439 | ||
1440 | return 0; | |
1441 | } | |
1442 | ||
9eff1a38 JL |
1443 | /* |
1444 | * Update the node maps and sysfs entries for each cpu whose home node | |
79c5fceb | 1445 | * has changed. Returns 1 when the topology has changed, and 0 otherwise. |
9eff1a38 JL |
1446 | */ |
1447 | int arch_update_cpu_topology(void) | |
1448 | { | |
30c05350 NF |
1449 | unsigned int cpu, changed = 0; |
1450 | struct topology_update_data *updates, *ud; | |
9eff1a38 | 1451 | unsigned int associativity[VPHN_ASSOC_BUFSIZE] = {0}; |
8a25a2fd | 1452 | struct device *dev; |
30c05350 NF |
1453 | int weight, i = 0; |
1454 | ||
1455 | weight = cpumask_weight(&cpu_associativity_changes_mask); | |
1456 | if (!weight) | |
1457 | return 0; | |
1458 | ||
1459 | updates = kzalloc(weight * (sizeof(*updates)), GFP_KERNEL); | |
1460 | if (!updates) | |
1461 | return 0; | |
9eff1a38 | 1462 | |
5d88aa85 | 1463 | for_each_cpu(cpu, &cpu_associativity_changes_mask) { |
30c05350 NF |
1464 | ud = &updates[i++]; |
1465 | ud->cpu = cpu; | |
9eff1a38 | 1466 | vphn_get_associativity(cpu, associativity); |
30c05350 | 1467 | ud->new_nid = associativity_to_nid(associativity); |
9eff1a38 | 1468 | |
30c05350 NF |
1469 | if (ud->new_nid < 0 || !node_online(ud->new_nid)) |
1470 | ud->new_nid = first_online_node; | |
9eff1a38 | 1471 | |
30c05350 | 1472 | ud->old_nid = numa_cpu_lookup_table[cpu]; |
9eff1a38 | 1473 | |
30c05350 NF |
1474 | if (i < weight) |
1475 | ud->next = &updates[i]; | |
1476 | } | |
1477 | ||
1478 | stop_machine(update_cpu_topology, &updates[0], cpu_online_mask); | |
1479 | ||
1480 | for (ud = &updates[0]; ud; ud = ud->next) { | |
1481 | dev = get_cpu_device(ud->cpu); | |
8a25a2fd KS |
1482 | if (dev) |
1483 | kobject_uevent(&dev->kobj, KOBJ_CHANGE); | |
30c05350 | 1484 | cpumask_clear_cpu(ud->cpu, &cpu_associativity_changes_mask); |
79c5fceb | 1485 | changed = 1; |
9eff1a38 JL |
1486 | } |
1487 | ||
30c05350 | 1488 | kfree(updates); |
79c5fceb | 1489 | return changed; |
9eff1a38 JL |
1490 | } |
1491 | ||
1492 | static void topology_work_fn(struct work_struct *work) | |
1493 | { | |
1494 | rebuild_sched_domains(); | |
1495 | } | |
1496 | static DECLARE_WORK(topology_work, topology_work_fn); | |
1497 | ||
1498 | void topology_schedule_update(void) | |
1499 | { | |
1500 | schedule_work(&topology_work); | |
1501 | } | |
1502 | ||
1503 | static void topology_timer_fn(unsigned long ignored) | |
1504 | { | |
5d88aa85 | 1505 | if (prrn_enabled && cpumask_weight(&cpu_associativity_changes_mask)) |
9eff1a38 | 1506 | topology_schedule_update(); |
5d88aa85 JL |
1507 | else if (vphn_enabled) { |
1508 | if (update_cpu_associativity_changes_mask() > 0) | |
1509 | topology_schedule_update(); | |
1510 | reset_topology_timer(); | |
1511 | } | |
9eff1a38 JL |
1512 | } |
1513 | static struct timer_list topology_timer = | |
1514 | TIMER_INITIALIZER(topology_timer_fn, 0, 0); | |
1515 | ||
5d88aa85 | 1516 | static void reset_topology_timer(void) |
9eff1a38 JL |
1517 | { |
1518 | topology_timer.data = 0; | |
1519 | topology_timer.expires = jiffies + 60 * HZ; | |
5d88aa85 JL |
1520 | mod_timer(&topology_timer, topology_timer.expires); |
1521 | } | |
1522 | ||
1523 | static void stage_topology_update(int core_id) | |
1524 | { | |
1525 | cpumask_or(&cpu_associativity_changes_mask, | |
1526 | &cpu_associativity_changes_mask, cpu_sibling_mask(core_id)); | |
1527 | reset_topology_timer(); | |
1528 | } | |
1529 | ||
1530 | static int dt_update_callback(struct notifier_block *nb, | |
1531 | unsigned long action, void *data) | |
1532 | { | |
1533 | struct of_prop_reconfig *update; | |
1534 | int rc = NOTIFY_DONE; | |
1535 | ||
1536 | switch (action) { | |
5d88aa85 JL |
1537 | case OF_RECONFIG_UPDATE_PROPERTY: |
1538 | update = (struct of_prop_reconfig *)data; | |
30c05350 NF |
1539 | if (!of_prop_cmp(update->dn->type, "cpu") && |
1540 | !of_prop_cmp(update->prop->name, "ibm,associativity")) { | |
5d88aa85 JL |
1541 | u32 core_id; |
1542 | of_property_read_u32(update->dn, "reg", &core_id); | |
1543 | stage_topology_update(core_id); | |
1544 | rc = NOTIFY_OK; | |
1545 | } | |
1546 | break; | |
1547 | } | |
1548 | ||
1549 | return rc; | |
9eff1a38 JL |
1550 | } |
1551 | ||
5d88aa85 JL |
1552 | static struct notifier_block dt_update_nb = { |
1553 | .notifier_call = dt_update_callback, | |
1554 | }; | |
1555 | ||
9eff1a38 | 1556 | /* |
5d88aa85 | 1557 | * Start polling for associativity changes. |
9eff1a38 JL |
1558 | */ |
1559 | int start_topology_update(void) | |
1560 | { | |
1561 | int rc = 0; | |
1562 | ||
5d88aa85 JL |
1563 | if (firmware_has_feature(FW_FEATURE_PRRN)) { |
1564 | if (!prrn_enabled) { | |
1565 | prrn_enabled = 1; | |
1566 | vphn_enabled = 0; | |
1567 | rc = of_reconfig_notifier_register(&dt_update_nb); | |
1568 | } | |
1569 | } else if (0 && firmware_has_feature(FW_FEATURE_VPHN) && | |
1570 | get_lppaca()->shared_proc) { | |
1571 | /* Disabled until races with load balancing are fixed */ | |
1572 | if (!vphn_enabled) { | |
1573 | prrn_enabled = 0; | |
1574 | vphn_enabled = 1; | |
1575 | setup_cpu_associativity_change_counters(); | |
1576 | init_timer_deferrable(&topology_timer); | |
1577 | reset_topology_timer(); | |
1578 | } | |
9eff1a38 JL |
1579 | } |
1580 | ||
1581 | return rc; | |
1582 | } | |
1583 | __initcall(start_topology_update); | |
1584 | ||
1585 | /* | |
1586 | * Disable polling for VPHN associativity changes. | |
1587 | */ | |
1588 | int stop_topology_update(void) | |
1589 | { | |
5d88aa85 JL |
1590 | int rc = 0; |
1591 | ||
1592 | if (prrn_enabled) { | |
1593 | prrn_enabled = 0; | |
1594 | rc = of_reconfig_notifier_unregister(&dt_update_nb); | |
1595 | } else if (vphn_enabled) { | |
1596 | vphn_enabled = 0; | |
1597 | rc = del_timer_sync(&topology_timer); | |
1598 | } | |
1599 | ||
1600 | return rc; | |
9eff1a38 | 1601 | } |
39bf990e | 1602 | #endif /* CONFIG_PPC_SPLPAR */ |