]>
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 | */ | |
2d73bae1 NA |
11 | #define pr_fmt(fmt) "numa: " fmt |
12 | ||
1da177e4 LT |
13 | #include <linux/threads.h> |
14 | #include <linux/bootmem.h> | |
15 | #include <linux/init.h> | |
16 | #include <linux/mm.h> | |
17 | #include <linux/mmzone.h> | |
4b16f8e2 | 18 | #include <linux/export.h> |
1da177e4 LT |
19 | #include <linux/nodemask.h> |
20 | #include <linux/cpu.h> | |
21 | #include <linux/notifier.h> | |
95f72d1e | 22 | #include <linux/memblock.h> |
6df1646e | 23 | #include <linux/of.h> |
06eccea6 | 24 | #include <linux/pfn.h> |
9eff1a38 JL |
25 | #include <linux/cpuset.h> |
26 | #include <linux/node.h> | |
30c05350 | 27 | #include <linux/stop_machine.h> |
e04fa612 NF |
28 | #include <linux/proc_fs.h> |
29 | #include <linux/seq_file.h> | |
30 | #include <linux/uaccess.h> | |
191a7120 | 31 | #include <linux/slab.h> |
3be7db6a | 32 | #include <asm/cputhreads.h> |
45fb6cea | 33 | #include <asm/sparsemem.h> |
d9b2b2a2 | 34 | #include <asm/prom.h> |
2249ca9d | 35 | #include <asm/smp.h> |
d4edc5b6 SB |
36 | #include <asm/cputhreads.h> |
37 | #include <asm/topology.h> | |
9eff1a38 JL |
38 | #include <asm/firmware.h> |
39 | #include <asm/paca.h> | |
39bf990e | 40 | #include <asm/hvcall.h> |
ae3a197e | 41 | #include <asm/setup.h> |
176bbf14 | 42 | #include <asm/vdso.h> |
1da177e4 LT |
43 | |
44 | static int numa_enabled = 1; | |
45 | ||
1daa6d08 BS |
46 | static char *cmdline __initdata; |
47 | ||
1da177e4 LT |
48 | static int numa_debug; |
49 | #define dbg(args...) if (numa_debug) { printk(KERN_INFO args); } | |
50 | ||
45fb6cea | 51 | int numa_cpu_lookup_table[NR_CPUS]; |
25863de0 | 52 | cpumask_var_t node_to_cpumask_map[MAX_NUMNODES]; |
1da177e4 | 53 | struct pglist_data *node_data[MAX_NUMNODES]; |
45fb6cea AB |
54 | |
55 | EXPORT_SYMBOL(numa_cpu_lookup_table); | |
25863de0 | 56 | EXPORT_SYMBOL(node_to_cpumask_map); |
45fb6cea AB |
57 | EXPORT_SYMBOL(node_data); |
58 | ||
1da177e4 | 59 | static int min_common_depth; |
237a0989 | 60 | static int n_mem_addr_cells, n_mem_size_cells; |
41eab6f8 AB |
61 | static int form1_affinity; |
62 | ||
63 | #define MAX_DISTANCE_REF_POINTS 4 | |
64 | static int distance_ref_points_depth; | |
b08a2a12 | 65 | static const __be32 *distance_ref_points; |
41eab6f8 | 66 | static int distance_lookup_table[MAX_NUMNODES][MAX_DISTANCE_REF_POINTS]; |
1da177e4 | 67 | |
25863de0 AB |
68 | /* |
69 | * Allocate node_to_cpumask_map based on number of available nodes | |
70 | * Requires node_possible_map to be valid. | |
71 | * | |
9512938b | 72 | * Note: cpumask_of_node() is not valid until after this is done. |
25863de0 AB |
73 | */ |
74 | static void __init setup_node_to_cpumask_map(void) | |
75 | { | |
f9d531b8 | 76 | unsigned int node; |
25863de0 AB |
77 | |
78 | /* setup nr_node_ids if not done yet */ | |
f9d531b8 CS |
79 | if (nr_node_ids == MAX_NUMNODES) |
80 | setup_nr_node_ids(); | |
25863de0 AB |
81 | |
82 | /* allocate the map */ | |
c118baf8 | 83 | for_each_node(node) |
25863de0 AB |
84 | alloc_bootmem_cpumask_var(&node_to_cpumask_map[node]); |
85 | ||
86 | /* cpumask_of_node() will now work */ | |
87 | dbg("Node to cpumask map for %d nodes\n", nr_node_ids); | |
88 | } | |
89 | ||
55671f3c | 90 | static int __init fake_numa_create_new_node(unsigned long end_pfn, |
1daa6d08 BS |
91 | unsigned int *nid) |
92 | { | |
93 | unsigned long long mem; | |
94 | char *p = cmdline; | |
95 | static unsigned int fake_nid; | |
96 | static unsigned long long curr_boundary; | |
97 | ||
98 | /* | |
99 | * Modify node id, iff we started creating NUMA nodes | |
100 | * We want to continue from where we left of the last time | |
101 | */ | |
102 | if (fake_nid) | |
103 | *nid = fake_nid; | |
104 | /* | |
105 | * In case there are no more arguments to parse, the | |
106 | * node_id should be the same as the last fake node id | |
107 | * (we've handled this above). | |
108 | */ | |
109 | if (!p) | |
110 | return 0; | |
111 | ||
112 | mem = memparse(p, &p); | |
113 | if (!mem) | |
114 | return 0; | |
115 | ||
116 | if (mem < curr_boundary) | |
117 | return 0; | |
118 | ||
119 | curr_boundary = mem; | |
120 | ||
121 | if ((end_pfn << PAGE_SHIFT) > mem) { | |
122 | /* | |
123 | * Skip commas and spaces | |
124 | */ | |
125 | while (*p == ',' || *p == ' ' || *p == '\t') | |
126 | p++; | |
127 | ||
128 | cmdline = p; | |
129 | fake_nid++; | |
130 | *nid = fake_nid; | |
131 | dbg("created new fake_node with id %d\n", fake_nid); | |
132 | return 1; | |
133 | } | |
134 | return 0; | |
135 | } | |
136 | ||
d4edc5b6 SB |
137 | static void reset_numa_cpu_lookup_table(void) |
138 | { | |
139 | unsigned int cpu; | |
140 | ||
141 | for_each_possible_cpu(cpu) | |
142 | numa_cpu_lookup_table[cpu] = -1; | |
143 | } | |
144 | ||
145 | static void update_numa_cpu_lookup_table(unsigned int cpu, int node) | |
1da177e4 LT |
146 | { |
147 | numa_cpu_lookup_table[cpu] = node; | |
d4edc5b6 SB |
148 | } |
149 | ||
150 | static void map_cpu_to_node(int cpu, int node) | |
151 | { | |
152 | update_numa_cpu_lookup_table(cpu, node); | |
45fb6cea | 153 | |
bf4b85b0 NL |
154 | dbg("adding cpu %d to node %d\n", cpu, node); |
155 | ||
25863de0 AB |
156 | if (!(cpumask_test_cpu(cpu, node_to_cpumask_map[node]))) |
157 | cpumask_set_cpu(cpu, node_to_cpumask_map[node]); | |
1da177e4 LT |
158 | } |
159 | ||
39bf990e | 160 | #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_PPC_SPLPAR) |
1da177e4 LT |
161 | static void unmap_cpu_from_node(unsigned long cpu) |
162 | { | |
163 | int node = numa_cpu_lookup_table[cpu]; | |
164 | ||
165 | dbg("removing cpu %lu from node %d\n", cpu, node); | |
166 | ||
25863de0 | 167 | if (cpumask_test_cpu(cpu, node_to_cpumask_map[node])) { |
429f4d8d | 168 | cpumask_clear_cpu(cpu, node_to_cpumask_map[node]); |
1da177e4 LT |
169 | } else { |
170 | printk(KERN_ERR "WARNING: cpu %lu not found in node %d\n", | |
171 | cpu, node); | |
172 | } | |
173 | } | |
39bf990e | 174 | #endif /* CONFIG_HOTPLUG_CPU || CONFIG_PPC_SPLPAR */ |
1da177e4 | 175 | |
1da177e4 | 176 | /* must hold reference to node during call */ |
b08a2a12 | 177 | static const __be32 *of_get_associativity(struct device_node *dev) |
1da177e4 | 178 | { |
e2eb6392 | 179 | return of_get_property(dev, "ibm,associativity", NULL); |
1da177e4 LT |
180 | } |
181 | ||
cf00085d C |
182 | /* |
183 | * Returns the property linux,drconf-usable-memory if | |
184 | * it exists (the property exists only in kexec/kdump kernels, | |
185 | * added by kexec-tools) | |
186 | */ | |
b08a2a12 | 187 | static const __be32 *of_get_usable_memory(struct device_node *memory) |
cf00085d | 188 | { |
b08a2a12 | 189 | const __be32 *prop; |
cf00085d C |
190 | u32 len; |
191 | prop = of_get_property(memory, "linux,drconf-usable-memory", &len); | |
192 | if (!prop || len < sizeof(unsigned int)) | |
ec32dd66 | 193 | return NULL; |
cf00085d C |
194 | return prop; |
195 | } | |
196 | ||
41eab6f8 AB |
197 | int __node_distance(int a, int b) |
198 | { | |
199 | int i; | |
200 | int distance = LOCAL_DISTANCE; | |
201 | ||
202 | if (!form1_affinity) | |
7122beee | 203 | return ((a == b) ? LOCAL_DISTANCE : REMOTE_DISTANCE); |
41eab6f8 AB |
204 | |
205 | for (i = 0; i < distance_ref_points_depth; i++) { | |
206 | if (distance_lookup_table[a][i] == distance_lookup_table[b][i]) | |
207 | break; | |
208 | ||
209 | /* Double the distance for each NUMA level */ | |
210 | distance *= 2; | |
211 | } | |
212 | ||
213 | return distance; | |
214 | } | |
12c743eb | 215 | EXPORT_SYMBOL(__node_distance); |
41eab6f8 AB |
216 | |
217 | static void initialize_distance_lookup_table(int nid, | |
b08a2a12 | 218 | const __be32 *associativity) |
41eab6f8 AB |
219 | { |
220 | int i; | |
221 | ||
222 | if (!form1_affinity) | |
223 | return; | |
224 | ||
225 | for (i = 0; i < distance_ref_points_depth; i++) { | |
b08a2a12 AP |
226 | const __be32 *entry; |
227 | ||
1d805440 | 228 | entry = &associativity[be32_to_cpu(distance_ref_points[i]) - 1]; |
b08a2a12 | 229 | distance_lookup_table[nid][i] = of_read_number(entry, 1); |
41eab6f8 AB |
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 | */ | |
b08a2a12 | 236 | static int associativity_to_nid(const __be32 *associativity) |
1da177e4 | 237 | { |
482ec7c4 | 238 | int nid = -1; |
1da177e4 LT |
239 | |
240 | if (min_common_depth == -1) | |
482ec7c4 | 241 | goto out; |
1da177e4 | 242 | |
b08a2a12 AP |
243 | if (of_read_number(associativity, 1) >= min_common_depth) |
244 | nid = of_read_number(&associativity[min_common_depth], 1); | |
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 | |
b08a2a12 | 250 | if (nid > 0 && |
1d805440 ND |
251 | of_read_number(associativity, 1) >= distance_ref_points_depth) { |
252 | /* | |
253 | * Skip the length field and send start of associativity array | |
254 | */ | |
255 | initialize_distance_lookup_table(nid, associativity + 1); | |
256 | } | |
41eab6f8 | 257 | |
482ec7c4 | 258 | out: |
cf950b7a | 259 | return nid; |
1da177e4 LT |
260 | } |
261 | ||
9eff1a38 JL |
262 | /* Returns the nid associated with the given device tree node, |
263 | * or -1 if not found. | |
264 | */ | |
265 | static int of_node_to_nid_single(struct device_node *device) | |
266 | { | |
267 | int nid = -1; | |
b08a2a12 | 268 | const __be32 *tmp; |
9eff1a38 JL |
269 | |
270 | tmp = of_get_associativity(device); | |
271 | if (tmp) | |
272 | nid = associativity_to_nid(tmp); | |
273 | return nid; | |
274 | } | |
275 | ||
953039c8 JK |
276 | /* Walk the device tree upwards, looking for an associativity id */ |
277 | int of_node_to_nid(struct device_node *device) | |
278 | { | |
953039c8 JK |
279 | int nid = -1; |
280 | ||
281 | of_node_get(device); | |
282 | while (device) { | |
283 | nid = of_node_to_nid_single(device); | |
284 | if (nid != -1) | |
285 | break; | |
286 | ||
1def3758 | 287 | device = of_get_next_parent(device); |
953039c8 JK |
288 | } |
289 | of_node_put(device); | |
290 | ||
291 | return nid; | |
292 | } | |
be9ba9ff | 293 | EXPORT_SYMBOL(of_node_to_nid); |
953039c8 | 294 | |
1da177e4 LT |
295 | static int __init find_min_common_depth(void) |
296 | { | |
41eab6f8 | 297 | int depth; |
e70606eb | 298 | struct device_node *root; |
1da177e4 | 299 | |
1c8ee733 DS |
300 | if (firmware_has_feature(FW_FEATURE_OPAL)) |
301 | root = of_find_node_by_path("/ibm,opal"); | |
302 | else | |
303 | root = of_find_node_by_path("/rtas"); | |
e70606eb ME |
304 | if (!root) |
305 | root = of_find_node_by_path("/"); | |
1da177e4 LT |
306 | |
307 | /* | |
41eab6f8 AB |
308 | * This property is a set of 32-bit integers, each representing |
309 | * an index into the ibm,associativity nodes. | |
310 | * | |
311 | * With form 0 affinity the first integer is for an SMP configuration | |
312 | * (should be all 0's) and the second is for a normal NUMA | |
313 | * configuration. We have only one level of NUMA. | |
314 | * | |
315 | * With form 1 affinity the first integer is the most significant | |
316 | * NUMA boundary and the following are progressively less significant | |
317 | * boundaries. There can be more than one level of NUMA. | |
1da177e4 | 318 | */ |
e70606eb | 319 | distance_ref_points = of_get_property(root, |
41eab6f8 AB |
320 | "ibm,associativity-reference-points", |
321 | &distance_ref_points_depth); | |
322 | ||
323 | if (!distance_ref_points) { | |
324 | dbg("NUMA: ibm,associativity-reference-points not found.\n"); | |
325 | goto err; | |
326 | } | |
327 | ||
328 | distance_ref_points_depth /= sizeof(int); | |
1da177e4 | 329 | |
8002b0c5 NF |
330 | if (firmware_has_feature(FW_FEATURE_OPAL) || |
331 | firmware_has_feature(FW_FEATURE_TYPE1_AFFINITY)) { | |
332 | dbg("Using form 1 affinity\n"); | |
1c8ee733 | 333 | form1_affinity = 1; |
4b83c330 AB |
334 | } |
335 | ||
41eab6f8 | 336 | if (form1_affinity) { |
b08a2a12 | 337 | depth = of_read_number(distance_ref_points, 1); |
1da177e4 | 338 | } else { |
41eab6f8 AB |
339 | if (distance_ref_points_depth < 2) { |
340 | printk(KERN_WARNING "NUMA: " | |
341 | "short ibm,associativity-reference-points\n"); | |
342 | goto err; | |
343 | } | |
344 | ||
b08a2a12 | 345 | depth = of_read_number(&distance_ref_points[1], 1); |
1da177e4 | 346 | } |
1da177e4 | 347 | |
41eab6f8 AB |
348 | /* |
349 | * Warn and cap if the hardware supports more than | |
350 | * MAX_DISTANCE_REF_POINTS domains. | |
351 | */ | |
352 | if (distance_ref_points_depth > MAX_DISTANCE_REF_POINTS) { | |
353 | printk(KERN_WARNING "NUMA: distance array capped at " | |
354 | "%d entries\n", MAX_DISTANCE_REF_POINTS); | |
355 | distance_ref_points_depth = MAX_DISTANCE_REF_POINTS; | |
356 | } | |
357 | ||
e70606eb | 358 | of_node_put(root); |
1da177e4 | 359 | return depth; |
41eab6f8 AB |
360 | |
361 | err: | |
e70606eb | 362 | of_node_put(root); |
41eab6f8 | 363 | return -1; |
1da177e4 LT |
364 | } |
365 | ||
84c9fdd1 | 366 | static void __init get_n_mem_cells(int *n_addr_cells, int *n_size_cells) |
1da177e4 LT |
367 | { |
368 | struct device_node *memory = NULL; | |
1da177e4 LT |
369 | |
370 | memory = of_find_node_by_type(memory, "memory"); | |
54c23310 | 371 | if (!memory) |
84c9fdd1 | 372 | panic("numa.c: No memory nodes found!"); |
54c23310 | 373 | |
a8bda5dd | 374 | *n_addr_cells = of_n_addr_cells(memory); |
9213feea | 375 | *n_size_cells = of_n_size_cells(memory); |
84c9fdd1 | 376 | of_node_put(memory); |
1da177e4 LT |
377 | } |
378 | ||
b08a2a12 | 379 | static unsigned long read_n_cells(int n, const __be32 **buf) |
1da177e4 LT |
380 | { |
381 | unsigned long result = 0; | |
382 | ||
383 | while (n--) { | |
b08a2a12 | 384 | result = (result << 32) | of_read_number(*buf, 1); |
1da177e4 LT |
385 | (*buf)++; |
386 | } | |
387 | return result; | |
388 | } | |
389 | ||
8342681d | 390 | /* |
95f72d1e | 391 | * Read the next memblock list entry from the ibm,dynamic-memory property |
8342681d NF |
392 | * and return the information in the provided of_drconf_cell structure. |
393 | */ | |
b08a2a12 | 394 | static void read_drconf_cell(struct of_drconf_cell *drmem, const __be32 **cellp) |
8342681d | 395 | { |
b08a2a12 | 396 | const __be32 *cp; |
8342681d NF |
397 | |
398 | drmem->base_addr = read_n_cells(n_mem_addr_cells, cellp); | |
399 | ||
400 | cp = *cellp; | |
b08a2a12 AP |
401 | drmem->drc_index = of_read_number(cp, 1); |
402 | drmem->reserved = of_read_number(&cp[1], 1); | |
403 | drmem->aa_index = of_read_number(&cp[2], 1); | |
404 | drmem->flags = of_read_number(&cp[3], 1); | |
8342681d NF |
405 | |
406 | *cellp = cp + 4; | |
407 | } | |
408 | ||
409 | /* | |
25985edc | 410 | * Retrieve and validate the ibm,dynamic-memory property of the device tree. |
8342681d | 411 | * |
95f72d1e YL |
412 | * The layout of the ibm,dynamic-memory property is a number N of memblock |
413 | * list entries followed by N memblock list entries. Each memblock list entry | |
25985edc | 414 | * contains information as laid out in the of_drconf_cell struct above. |
8342681d | 415 | */ |
b08a2a12 | 416 | static int of_get_drconf_memory(struct device_node *memory, const __be32 **dm) |
8342681d | 417 | { |
b08a2a12 | 418 | const __be32 *prop; |
8342681d NF |
419 | u32 len, entries; |
420 | ||
421 | prop = of_get_property(memory, "ibm,dynamic-memory", &len); | |
422 | if (!prop || len < sizeof(unsigned int)) | |
423 | return 0; | |
424 | ||
b08a2a12 | 425 | entries = of_read_number(prop++, 1); |
8342681d NF |
426 | |
427 | /* Now that we know the number of entries, revalidate the size | |
428 | * of the property read in to ensure we have everything | |
429 | */ | |
430 | if (len < (entries * (n_mem_addr_cells + 4) + 1) * sizeof(unsigned int)) | |
431 | return 0; | |
432 | ||
433 | *dm = prop; | |
434 | return entries; | |
435 | } | |
436 | ||
437 | /* | |
25985edc | 438 | * Retrieve and validate the ibm,lmb-size property for drconf memory |
8342681d NF |
439 | * from the device tree. |
440 | */ | |
3fdfd990 | 441 | static u64 of_get_lmb_size(struct device_node *memory) |
8342681d | 442 | { |
b08a2a12 | 443 | const __be32 *prop; |
8342681d NF |
444 | u32 len; |
445 | ||
3fdfd990 | 446 | prop = of_get_property(memory, "ibm,lmb-size", &len); |
8342681d NF |
447 | if (!prop || len < sizeof(unsigned int)) |
448 | return 0; | |
449 | ||
450 | return read_n_cells(n_mem_size_cells, &prop); | |
451 | } | |
452 | ||
453 | struct assoc_arrays { | |
454 | u32 n_arrays; | |
455 | u32 array_sz; | |
b08a2a12 | 456 | const __be32 *arrays; |
8342681d NF |
457 | }; |
458 | ||
459 | /* | |
25985edc | 460 | * Retrieve and validate the list of associativity arrays for drconf |
8342681d NF |
461 | * memory from the ibm,associativity-lookup-arrays property of the |
462 | * device tree.. | |
463 | * | |
464 | * The layout of the ibm,associativity-lookup-arrays property is a number N | |
465 | * indicating the number of associativity arrays, followed by a number M | |
466 | * indicating the size of each associativity array, followed by a list | |
467 | * of N associativity arrays. | |
468 | */ | |
469 | static int of_get_assoc_arrays(struct device_node *memory, | |
470 | struct assoc_arrays *aa) | |
471 | { | |
b08a2a12 | 472 | const __be32 *prop; |
8342681d NF |
473 | u32 len; |
474 | ||
475 | prop = of_get_property(memory, "ibm,associativity-lookup-arrays", &len); | |
476 | if (!prop || len < 2 * sizeof(unsigned int)) | |
477 | return -1; | |
478 | ||
b08a2a12 AP |
479 | aa->n_arrays = of_read_number(prop++, 1); |
480 | aa->array_sz = of_read_number(prop++, 1); | |
8342681d | 481 | |
42b2aa86 | 482 | /* Now that we know the number of arrays and size of each array, |
8342681d NF |
483 | * revalidate the size of the property read in. |
484 | */ | |
485 | if (len < (aa->n_arrays * aa->array_sz + 2) * sizeof(unsigned int)) | |
486 | return -1; | |
487 | ||
488 | aa->arrays = prop; | |
489 | return 0; | |
490 | } | |
491 | ||
492 | /* | |
493 | * This is like of_node_to_nid_single() for memory represented in the | |
494 | * ibm,dynamic-reconfiguration-memory node. | |
495 | */ | |
496 | static int of_drconf_to_nid_single(struct of_drconf_cell *drmem, | |
497 | struct assoc_arrays *aa) | |
498 | { | |
499 | int default_nid = 0; | |
500 | int nid = default_nid; | |
501 | int index; | |
502 | ||
503 | if (min_common_depth > 0 && min_common_depth <= aa->array_sz && | |
504 | !(drmem->flags & DRCONF_MEM_AI_INVALID) && | |
505 | drmem->aa_index < aa->n_arrays) { | |
506 | index = drmem->aa_index * aa->array_sz + min_common_depth - 1; | |
b08a2a12 | 507 | nid = of_read_number(&aa->arrays[index], 1); |
8342681d NF |
508 | |
509 | if (nid == 0xffff || nid >= MAX_NUMNODES) | |
510 | nid = default_nid; | |
1d805440 ND |
511 | |
512 | if (nid > 0) { | |
513 | index = drmem->aa_index * aa->array_sz; | |
514 | initialize_distance_lookup_table(nid, | |
515 | &aa->arrays[index]); | |
516 | } | |
8342681d NF |
517 | } |
518 | ||
519 | return nid; | |
520 | } | |
521 | ||
1da177e4 LT |
522 | /* |
523 | * Figure out to which domain a cpu belongs and stick it there. | |
524 | * Return the id of the domain used. | |
525 | */ | |
061d19f2 | 526 | static int numa_setup_cpu(unsigned long lcpu) |
1da177e4 | 527 | { |
297cf502 | 528 | int nid = -1; |
d4edc5b6 SB |
529 | struct device_node *cpu; |
530 | ||
531 | /* | |
532 | * If a valid cpu-to-node mapping is already available, use it | |
533 | * directly instead of querying the firmware, since it represents | |
534 | * the most recent mapping notified to us by the platform (eg: VPHN). | |
535 | */ | |
536 | if ((nid = numa_cpu_lookup_table[lcpu]) >= 0) { | |
537 | map_cpu_to_node(lcpu, nid); | |
538 | return nid; | |
539 | } | |
540 | ||
541 | cpu = of_get_cpu_node(lcpu, NULL); | |
1da177e4 LT |
542 | |
543 | if (!cpu) { | |
544 | WARN_ON(1); | |
297cf502 LZ |
545 | if (cpu_present(lcpu)) |
546 | goto out_present; | |
547 | else | |
548 | goto out; | |
1da177e4 LT |
549 | } |
550 | ||
953039c8 | 551 | nid = of_node_to_nid_single(cpu); |
1da177e4 | 552 | |
297cf502 | 553 | out_present: |
482ec7c4 | 554 | if (nid < 0 || !node_online(nid)) |
72c33688 | 555 | nid = first_online_node; |
1da177e4 | 556 | |
297cf502 | 557 | map_cpu_to_node(lcpu, nid); |
1da177e4 | 558 | of_node_put(cpu); |
297cf502 | 559 | out: |
cf950b7a | 560 | return nid; |
1da177e4 LT |
561 | } |
562 | ||
68fb18aa SB |
563 | static void verify_cpu_node_mapping(int cpu, int node) |
564 | { | |
565 | int base, sibling, i; | |
566 | ||
567 | /* Verify that all the threads in the core belong to the same node */ | |
568 | base = cpu_first_thread_sibling(cpu); | |
569 | ||
570 | for (i = 0; i < threads_per_core; i++) { | |
571 | sibling = base + i; | |
572 | ||
573 | if (sibling == cpu || cpu_is_offline(sibling)) | |
574 | continue; | |
575 | ||
576 | if (cpu_to_node(sibling) != node) { | |
577 | WARN(1, "CPU thread siblings %d and %d don't belong" | |
578 | " to the same node!\n", cpu, sibling); | |
579 | break; | |
580 | } | |
581 | } | |
582 | } | |
583 | ||
bdab88e0 SAS |
584 | /* Must run before sched domains notifier. */ |
585 | static int ppc_numa_cpu_prepare(unsigned int cpu) | |
586 | { | |
587 | int nid; | |
588 | ||
589 | nid = numa_setup_cpu(cpu); | |
590 | verify_cpu_node_mapping(cpu, nid); | |
591 | return 0; | |
592 | } | |
593 | ||
594 | static int ppc_numa_cpu_dead(unsigned int cpu) | |
595 | { | |
1da177e4 | 596 | #ifdef CONFIG_HOTPLUG_CPU |
bdab88e0 | 597 | unmap_cpu_from_node(cpu); |
1da177e4 | 598 | #endif |
bdab88e0 | 599 | return 0; |
1da177e4 LT |
600 | } |
601 | ||
602 | /* | |
603 | * Check and possibly modify a memory region to enforce the memory limit. | |
604 | * | |
605 | * Returns the size the region should have to enforce the memory limit. | |
606 | * This will either be the original value of size, a truncated value, | |
607 | * or zero. If the returned value of size is 0 the region should be | |
25985edc | 608 | * discarded as it lies wholly above the memory limit. |
1da177e4 | 609 | */ |
45fb6cea AB |
610 | static unsigned long __init numa_enforce_memory_limit(unsigned long start, |
611 | unsigned long size) | |
1da177e4 LT |
612 | { |
613 | /* | |
95f72d1e | 614 | * We use memblock_end_of_DRAM() in here instead of memory_limit because |
1da177e4 | 615 | * we've already adjusted it for the limit and it takes care of |
fe55249d MM |
616 | * having memory holes below the limit. Also, in the case of |
617 | * iommu_is_off, memory_limit is not set but is implicitly enforced. | |
1da177e4 | 618 | */ |
1da177e4 | 619 | |
95f72d1e | 620 | if (start + size <= memblock_end_of_DRAM()) |
1da177e4 LT |
621 | return size; |
622 | ||
95f72d1e | 623 | if (start >= memblock_end_of_DRAM()) |
1da177e4 LT |
624 | return 0; |
625 | ||
95f72d1e | 626 | return memblock_end_of_DRAM() - start; |
1da177e4 LT |
627 | } |
628 | ||
cf00085d C |
629 | /* |
630 | * Reads the counter for a given entry in | |
631 | * linux,drconf-usable-memory property | |
632 | */ | |
b08a2a12 | 633 | static inline int __init read_usm_ranges(const __be32 **usm) |
cf00085d C |
634 | { |
635 | /* | |
3fdfd990 | 636 | * For each lmb in ibm,dynamic-memory a corresponding |
cf00085d C |
637 | * entry in linux,drconf-usable-memory property contains |
638 | * a counter followed by that many (base, size) duple. | |
639 | * read the counter from linux,drconf-usable-memory | |
640 | */ | |
641 | return read_n_cells(n_mem_size_cells, usm); | |
642 | } | |
643 | ||
0204568a PM |
644 | /* |
645 | * Extract NUMA information from the ibm,dynamic-reconfiguration-memory | |
646 | * node. This assumes n_mem_{addr,size}_cells have been set. | |
647 | */ | |
648 | static void __init parse_drconf_memory(struct device_node *memory) | |
649 | { | |
b08a2a12 | 650 | const __be32 *uninitialized_var(dm), *usm; |
cf00085d | 651 | unsigned int n, rc, ranges, is_kexec_kdump = 0; |
3fdfd990 | 652 | unsigned long lmb_size, base, size, sz; |
8342681d | 653 | int nid; |
aa709f3b | 654 | struct assoc_arrays aa = { .arrays = NULL }; |
8342681d NF |
655 | |
656 | n = of_get_drconf_memory(memory, &dm); | |
657 | if (!n) | |
0204568a PM |
658 | return; |
659 | ||
3fdfd990 BH |
660 | lmb_size = of_get_lmb_size(memory); |
661 | if (!lmb_size) | |
8342681d NF |
662 | return; |
663 | ||
664 | rc = of_get_assoc_arrays(memory, &aa); | |
665 | if (rc) | |
0204568a PM |
666 | return; |
667 | ||
cf00085d C |
668 | /* check if this is a kexec/kdump kernel */ |
669 | usm = of_get_usable_memory(memory); | |
670 | if (usm != NULL) | |
671 | is_kexec_kdump = 1; | |
672 | ||
0204568a | 673 | for (; n != 0; --n) { |
8342681d NF |
674 | struct of_drconf_cell drmem; |
675 | ||
676 | read_drconf_cell(&drmem, &dm); | |
677 | ||
678 | /* skip this block if the reserved bit is set in flags (0x80) | |
679 | or if the block is not assigned to this partition (0x8) */ | |
680 | if ((drmem.flags & DRCONF_MEM_RESERVED) | |
681 | || !(drmem.flags & DRCONF_MEM_ASSIGNED)) | |
0204568a | 682 | continue; |
1daa6d08 | 683 | |
cf00085d | 684 | base = drmem.base_addr; |
3fdfd990 | 685 | size = lmb_size; |
cf00085d | 686 | ranges = 1; |
8342681d | 687 | |
cf00085d C |
688 | if (is_kexec_kdump) { |
689 | ranges = read_usm_ranges(&usm); | |
690 | if (!ranges) /* there are no (base, size) duple */ | |
691 | continue; | |
692 | } | |
693 | do { | |
694 | if (is_kexec_kdump) { | |
695 | base = read_n_cells(n_mem_addr_cells, &usm); | |
696 | size = read_n_cells(n_mem_size_cells, &usm); | |
697 | } | |
698 | nid = of_drconf_to_nid_single(&drmem, &aa); | |
699 | fake_numa_create_new_node( | |
700 | ((base + size) >> PAGE_SHIFT), | |
8342681d | 701 | &nid); |
cf00085d C |
702 | node_set_online(nid); |
703 | sz = numa_enforce_memory_limit(base, size); | |
704 | if (sz) | |
e7e8de59 TC |
705 | memblock_set_node(base, sz, |
706 | &memblock.memory, nid); | |
cf00085d | 707 | } while (--ranges); |
0204568a PM |
708 | } |
709 | } | |
710 | ||
1da177e4 LT |
711 | static int __init parse_numa_properties(void) |
712 | { | |
94db7c5e | 713 | struct device_node *memory; |
482ec7c4 | 714 | int default_nid = 0; |
1da177e4 LT |
715 | unsigned long i; |
716 | ||
717 | if (numa_enabled == 0) { | |
718 | printk(KERN_WARNING "NUMA disabled by user\n"); | |
719 | return -1; | |
720 | } | |
721 | ||
1da177e4 LT |
722 | min_common_depth = find_min_common_depth(); |
723 | ||
1da177e4 LT |
724 | if (min_common_depth < 0) |
725 | return min_common_depth; | |
726 | ||
bf4b85b0 NL |
727 | dbg("NUMA associativity depth for CPU/Memory: %d\n", min_common_depth); |
728 | ||
1da177e4 | 729 | /* |
482ec7c4 NL |
730 | * Even though we connect cpus to numa domains later in SMP |
731 | * init, we need to know the node ids now. This is because | |
732 | * each node to be onlined must have NODE_DATA etc backing it. | |
1da177e4 | 733 | */ |
482ec7c4 | 734 | for_each_present_cpu(i) { |
dfbe93a2 | 735 | struct device_node *cpu; |
cf950b7a | 736 | int nid; |
1da177e4 | 737 | |
8b16cd23 | 738 | cpu = of_get_cpu_node(i, NULL); |
482ec7c4 | 739 | BUG_ON(!cpu); |
953039c8 | 740 | nid = of_node_to_nid_single(cpu); |
482ec7c4 | 741 | of_node_put(cpu); |
1da177e4 | 742 | |
482ec7c4 NL |
743 | /* |
744 | * Don't fall back to default_nid yet -- we will plug | |
745 | * cpus into nodes once the memory scan has discovered | |
746 | * the topology. | |
747 | */ | |
748 | if (nid < 0) | |
749 | continue; | |
750 | node_set_online(nid); | |
1da177e4 LT |
751 | } |
752 | ||
237a0989 | 753 | get_n_mem_cells(&n_mem_addr_cells, &n_mem_size_cells); |
94db7c5e AB |
754 | |
755 | for_each_node_by_type(memory, "memory") { | |
1da177e4 LT |
756 | unsigned long start; |
757 | unsigned long size; | |
cf950b7a | 758 | int nid; |
1da177e4 | 759 | int ranges; |
b08a2a12 | 760 | const __be32 *memcell_buf; |
1da177e4 LT |
761 | unsigned int len; |
762 | ||
e2eb6392 | 763 | memcell_buf = of_get_property(memory, |
ba759485 ME |
764 | "linux,usable-memory", &len); |
765 | if (!memcell_buf || len <= 0) | |
e2eb6392 | 766 | memcell_buf = of_get_property(memory, "reg", &len); |
1da177e4 LT |
767 | if (!memcell_buf || len <= 0) |
768 | continue; | |
769 | ||
cc5d0189 BH |
770 | /* ranges in cell */ |
771 | ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells); | |
1da177e4 LT |
772 | new_range: |
773 | /* these are order-sensitive, and modify the buffer pointer */ | |
237a0989 MK |
774 | start = read_n_cells(n_mem_addr_cells, &memcell_buf); |
775 | size = read_n_cells(n_mem_size_cells, &memcell_buf); | |
1da177e4 | 776 | |
482ec7c4 NL |
777 | /* |
778 | * Assumption: either all memory nodes or none will | |
779 | * have associativity properties. If none, then | |
780 | * everything goes to default_nid. | |
781 | */ | |
953039c8 | 782 | nid = of_node_to_nid_single(memory); |
482ec7c4 NL |
783 | if (nid < 0) |
784 | nid = default_nid; | |
1daa6d08 BS |
785 | |
786 | fake_numa_create_new_node(((start + size) >> PAGE_SHIFT), &nid); | |
482ec7c4 | 787 | node_set_online(nid); |
1da177e4 | 788 | |
7656cd8e RA |
789 | size = numa_enforce_memory_limit(start, size); |
790 | if (size) | |
791 | memblock_set_node(start, size, &memblock.memory, nid); | |
1da177e4 LT |
792 | |
793 | if (--ranges) | |
794 | goto new_range; | |
795 | } | |
796 | ||
0204568a | 797 | /* |
dfbe93a2 AB |
798 | * Now do the same thing for each MEMBLOCK listed in the |
799 | * ibm,dynamic-memory property in the | |
800 | * ibm,dynamic-reconfiguration-memory node. | |
0204568a PM |
801 | */ |
802 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); | |
803 | if (memory) | |
804 | parse_drconf_memory(memory); | |
805 | ||
1da177e4 LT |
806 | return 0; |
807 | } | |
808 | ||
809 | static void __init setup_nonnuma(void) | |
810 | { | |
95f72d1e YL |
811 | unsigned long top_of_ram = memblock_end_of_DRAM(); |
812 | unsigned long total_ram = memblock_phys_mem_size(); | |
c67c3cb4 | 813 | unsigned long start_pfn, end_pfn; |
28be7072 BH |
814 | unsigned int nid = 0; |
815 | struct memblock_region *reg; | |
1da177e4 | 816 | |
e110b281 | 817 | printk(KERN_DEBUG "Top of RAM: 0x%lx, Total RAM: 0x%lx\n", |
1da177e4 | 818 | top_of_ram, total_ram); |
e110b281 | 819 | printk(KERN_DEBUG "Memory hole size: %ldMB\n", |
1da177e4 LT |
820 | (top_of_ram - total_ram) >> 20); |
821 | ||
28be7072 | 822 | for_each_memblock(memory, reg) { |
c7fc2de0 YL |
823 | start_pfn = memblock_region_memory_base_pfn(reg); |
824 | end_pfn = memblock_region_memory_end_pfn(reg); | |
1daa6d08 BS |
825 | |
826 | fake_numa_create_new_node(end_pfn, &nid); | |
1d7cfe18 | 827 | memblock_set_node(PFN_PHYS(start_pfn), |
e7e8de59 TC |
828 | PFN_PHYS(end_pfn - start_pfn), |
829 | &memblock.memory, nid); | |
1daa6d08 | 830 | node_set_online(nid); |
c67c3cb4 | 831 | } |
1da177e4 LT |
832 | } |
833 | ||
4b703a23 AB |
834 | void __init dump_numa_cpu_topology(void) |
835 | { | |
836 | unsigned int node; | |
837 | unsigned int cpu, count; | |
838 | ||
839 | if (min_common_depth == -1 || !numa_enabled) | |
840 | return; | |
841 | ||
842 | for_each_online_node(node) { | |
8467801c | 843 | pr_info("Node %d CPUs:", node); |
4b703a23 AB |
844 | |
845 | count = 0; | |
846 | /* | |
847 | * If we used a CPU iterator here we would miss printing | |
848 | * the holes in the cpumap. | |
849 | */ | |
25863de0 AB |
850 | for (cpu = 0; cpu < nr_cpu_ids; cpu++) { |
851 | if (cpumask_test_cpu(cpu, | |
852 | node_to_cpumask_map[node])) { | |
4b703a23 | 853 | if (count == 0) |
8467801c | 854 | pr_cont(" %u", cpu); |
4b703a23 AB |
855 | ++count; |
856 | } else { | |
857 | if (count > 1) | |
8467801c | 858 | pr_cont("-%u", cpu - 1); |
4b703a23 AB |
859 | count = 0; |
860 | } | |
861 | } | |
862 | ||
863 | if (count > 1) | |
8467801c AK |
864 | pr_cont("-%u", nr_cpu_ids - 1); |
865 | pr_cont("\n"); | |
4b703a23 AB |
866 | } |
867 | } | |
868 | ||
10239733 AB |
869 | /* Initialize NODE_DATA for a node on the local memory */ |
870 | static void __init setup_node_data(int nid, u64 start_pfn, u64 end_pfn) | |
4a618669 | 871 | { |
10239733 AB |
872 | u64 spanned_pages = end_pfn - start_pfn; |
873 | const size_t nd_size = roundup(sizeof(pg_data_t), SMP_CACHE_BYTES); | |
874 | u64 nd_pa; | |
875 | void *nd; | |
876 | int tnid; | |
4a618669 | 877 | |
10239733 AB |
878 | nd_pa = memblock_alloc_try_nid(nd_size, SMP_CACHE_BYTES, nid); |
879 | nd = __va(nd_pa); | |
4a618669 | 880 | |
10239733 AB |
881 | /* report and initialize */ |
882 | pr_info(" NODE_DATA [mem %#010Lx-%#010Lx]\n", | |
883 | nd_pa, nd_pa + nd_size - 1); | |
884 | tnid = early_pfn_to_nid(nd_pa >> PAGE_SHIFT); | |
885 | if (tnid != nid) | |
886 | pr_info(" NODE_DATA(%d) on node %d\n", nid, tnid); | |
4a618669 | 887 | |
10239733 AB |
888 | node_data[nid] = nd; |
889 | memset(NODE_DATA(nid), 0, sizeof(pg_data_t)); | |
890 | NODE_DATA(nid)->node_id = nid; | |
891 | NODE_DATA(nid)->node_start_pfn = start_pfn; | |
892 | NODE_DATA(nid)->node_spanned_pages = spanned_pages; | |
893 | } | |
4a618669 | 894 | |
10239733 | 895 | void __init initmem_init(void) |
1da177e4 | 896 | { |
2fabf084 | 897 | int nid, cpu; |
1da177e4 | 898 | |
95f72d1e | 899 | max_low_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT; |
1da177e4 LT |
900 | max_pfn = max_low_pfn; |
901 | ||
902 | if (parse_numa_properties()) | |
903 | setup_nonnuma(); | |
1da177e4 | 904 | |
10239733 AB |
905 | memblock_dump_all(); |
906 | ||
3af229f2 NA |
907 | /* |
908 | * Reduce the possible NUMA nodes to the online NUMA nodes, | |
909 | * since we do not support node hotplug. This ensures that we | |
910 | * lower the maximum NUMA node ID to what is actually present. | |
911 | */ | |
912 | nodes_and(node_possible_map, node_possible_map, node_online_map); | |
913 | ||
1da177e4 | 914 | for_each_online_node(nid) { |
c67c3cb4 | 915 | unsigned long start_pfn, end_pfn; |
1da177e4 | 916 | |
c67c3cb4 | 917 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); |
10239733 | 918 | setup_node_data(nid, start_pfn, end_pfn); |
8f64e1f2 | 919 | sparse_memory_present_with_active_regions(nid); |
4a618669 | 920 | } |
d3f6204a | 921 | |
21098b9e | 922 | sparse_init(); |
25863de0 | 923 | |
25863de0 AB |
924 | setup_node_to_cpumask_map(); |
925 | ||
d4edc5b6 | 926 | reset_numa_cpu_lookup_table(); |
bdab88e0 | 927 | |
2fabf084 NA |
928 | /* |
929 | * We need the numa_cpu_lookup_table to be accurate for all CPUs, | |
930 | * even before we online them, so that we can use cpu_to_{node,mem} | |
931 | * early in boot, cf. smp_prepare_cpus(). | |
bdab88e0 SAS |
932 | * _nocalls() + manual invocation is used because cpuhp is not yet |
933 | * initialized for the boot CPU. | |
2fabf084 | 934 | */ |
73c1b41e | 935 | cpuhp_setup_state_nocalls(CPUHP_POWER_NUMA_PREPARE, "powerpc/numa:prepare", |
bdab88e0 SAS |
936 | ppc_numa_cpu_prepare, ppc_numa_cpu_dead); |
937 | for_each_present_cpu(cpu) | |
938 | numa_setup_cpu(cpu); | |
1da177e4 LT |
939 | } |
940 | ||
1da177e4 LT |
941 | static int __init early_numa(char *p) |
942 | { | |
943 | if (!p) | |
944 | return 0; | |
945 | ||
946 | if (strstr(p, "off")) | |
947 | numa_enabled = 0; | |
948 | ||
949 | if (strstr(p, "debug")) | |
950 | numa_debug = 1; | |
951 | ||
1daa6d08 BS |
952 | p = strstr(p, "fake="); |
953 | if (p) | |
954 | cmdline = p + strlen("fake="); | |
955 | ||
1da177e4 LT |
956 | return 0; |
957 | } | |
958 | early_param("numa", early_numa); | |
237a0989 | 959 | |
2d73bae1 NA |
960 | static bool topology_updates_enabled = true; |
961 | ||
962 | static int __init early_topology_updates(char *p) | |
963 | { | |
964 | if (!p) | |
965 | return 0; | |
966 | ||
967 | if (!strcmp(p, "off")) { | |
968 | pr_info("Disabling topology updates\n"); | |
969 | topology_updates_enabled = false; | |
970 | } | |
971 | ||
972 | return 0; | |
973 | } | |
974 | early_param("topology_updates", early_topology_updates); | |
975 | ||
237a0989 | 976 | #ifdef CONFIG_MEMORY_HOTPLUG |
0db9360a | 977 | /* |
0f16ef7f NF |
978 | * Find the node associated with a hot added memory section for |
979 | * memory represented in the device tree by the property | |
980 | * ibm,dynamic-reconfiguration-memory/ibm,dynamic-memory. | |
0db9360a NF |
981 | */ |
982 | static int hot_add_drconf_scn_to_nid(struct device_node *memory, | |
983 | unsigned long scn_addr) | |
984 | { | |
b08a2a12 | 985 | const __be32 *dm; |
0f16ef7f | 986 | unsigned int drconf_cell_cnt, rc; |
3fdfd990 | 987 | unsigned long lmb_size; |
0db9360a | 988 | struct assoc_arrays aa; |
0f16ef7f | 989 | int nid = -1; |
0db9360a | 990 | |
0f16ef7f NF |
991 | drconf_cell_cnt = of_get_drconf_memory(memory, &dm); |
992 | if (!drconf_cell_cnt) | |
993 | return -1; | |
0db9360a | 994 | |
3fdfd990 BH |
995 | lmb_size = of_get_lmb_size(memory); |
996 | if (!lmb_size) | |
0f16ef7f | 997 | return -1; |
0db9360a NF |
998 | |
999 | rc = of_get_assoc_arrays(memory, &aa); | |
1000 | if (rc) | |
0f16ef7f | 1001 | return -1; |
0db9360a | 1002 | |
0f16ef7f | 1003 | for (; drconf_cell_cnt != 0; --drconf_cell_cnt) { |
0db9360a NF |
1004 | struct of_drconf_cell drmem; |
1005 | ||
1006 | read_drconf_cell(&drmem, &dm); | |
1007 | ||
1008 | /* skip this block if it is reserved or not assigned to | |
1009 | * this partition */ | |
1010 | if ((drmem.flags & DRCONF_MEM_RESERVED) | |
1011 | || !(drmem.flags & DRCONF_MEM_ASSIGNED)) | |
1012 | continue; | |
1013 | ||
0f16ef7f | 1014 | if ((scn_addr < drmem.base_addr) |
3fdfd990 | 1015 | || (scn_addr >= (drmem.base_addr + lmb_size))) |
0f16ef7f NF |
1016 | continue; |
1017 | ||
0db9360a | 1018 | nid = of_drconf_to_nid_single(&drmem, &aa); |
0f16ef7f NF |
1019 | break; |
1020 | } | |
1021 | ||
1022 | return nid; | |
1023 | } | |
1024 | ||
1025 | /* | |
1026 | * Find the node associated with a hot added memory section for memory | |
1027 | * represented in the device tree as a node (i.e. memory@XXXX) for | |
95f72d1e | 1028 | * each memblock. |
0f16ef7f | 1029 | */ |
ec32dd66 | 1030 | static int hot_add_node_scn_to_nid(unsigned long scn_addr) |
0f16ef7f | 1031 | { |
94db7c5e | 1032 | struct device_node *memory; |
0f16ef7f NF |
1033 | int nid = -1; |
1034 | ||
94db7c5e | 1035 | for_each_node_by_type(memory, "memory") { |
0f16ef7f NF |
1036 | unsigned long start, size; |
1037 | int ranges; | |
b08a2a12 | 1038 | const __be32 *memcell_buf; |
0f16ef7f NF |
1039 | unsigned int len; |
1040 | ||
1041 | memcell_buf = of_get_property(memory, "reg", &len); | |
1042 | if (!memcell_buf || len <= 0) | |
1043 | continue; | |
1044 | ||
1045 | /* ranges in cell */ | |
1046 | ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells); | |
1047 | ||
1048 | while (ranges--) { | |
1049 | start = read_n_cells(n_mem_addr_cells, &memcell_buf); | |
1050 | size = read_n_cells(n_mem_size_cells, &memcell_buf); | |
1051 | ||
1052 | if ((scn_addr < start) || (scn_addr >= (start + size))) | |
1053 | continue; | |
1054 | ||
1055 | nid = of_node_to_nid_single(memory); | |
1056 | break; | |
1057 | } | |
0db9360a | 1058 | |
0f16ef7f NF |
1059 | if (nid >= 0) |
1060 | break; | |
0db9360a NF |
1061 | } |
1062 | ||
60831842 AB |
1063 | of_node_put(memory); |
1064 | ||
0f16ef7f | 1065 | return nid; |
0db9360a NF |
1066 | } |
1067 | ||
237a0989 MK |
1068 | /* |
1069 | * Find the node associated with a hot added memory section. Section | |
95f72d1e YL |
1070 | * corresponds to a SPARSEMEM section, not an MEMBLOCK. It is assumed that |
1071 | * sections are fully contained within a single MEMBLOCK. | |
237a0989 MK |
1072 | */ |
1073 | int hot_add_scn_to_nid(unsigned long scn_addr) | |
1074 | { | |
1075 | struct device_node *memory = NULL; | |
4a3bac4e | 1076 | int nid; |
237a0989 MK |
1077 | |
1078 | if (!numa_enabled || (min_common_depth < 0)) | |
72c33688 | 1079 | return first_online_node; |
0db9360a NF |
1080 | |
1081 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); | |
1082 | if (memory) { | |
1083 | nid = hot_add_drconf_scn_to_nid(memory, scn_addr); | |
1084 | of_node_put(memory); | |
0f16ef7f NF |
1085 | } else { |
1086 | nid = hot_add_node_scn_to_nid(scn_addr); | |
0db9360a | 1087 | } |
237a0989 | 1088 | |
2a8628d4 | 1089 | if (nid < 0 || !node_possible(nid)) |
72c33688 | 1090 | nid = first_online_node; |
237a0989 | 1091 | |
0f16ef7f | 1092 | return nid; |
237a0989 | 1093 | } |
0f16ef7f | 1094 | |
cd34206e NA |
1095 | static u64 hot_add_drconf_memory_max(void) |
1096 | { | |
e70bd3ae | 1097 | struct device_node *memory = NULL; |
45b64ee6 | 1098 | struct device_node *dn = NULL; |
e70bd3ae BR |
1099 | unsigned int drconf_cell_cnt = 0; |
1100 | u64 lmb_size = 0; | |
ec32dd66 | 1101 | const __be32 *dm = NULL; |
45b64ee6 BR |
1102 | const __be64 *lrdr = NULL; |
1103 | struct of_drconf_cell drmem; | |
1104 | ||
1105 | dn = of_find_node_by_path("/rtas"); | |
1106 | if (dn) { | |
1107 | lrdr = of_get_property(dn, "ibm,lrdr-capacity", NULL); | |
1108 | of_node_put(dn); | |
1109 | if (lrdr) | |
1110 | return be64_to_cpup(lrdr); | |
1111 | } | |
cd34206e | 1112 | |
e70bd3ae BR |
1113 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); |
1114 | if (memory) { | |
1115 | drconf_cell_cnt = of_get_drconf_memory(memory, &dm); | |
1116 | lmb_size = of_get_lmb_size(memory); | |
cd34206e | 1117 | |
45b64ee6 BR |
1118 | /* Advance to the last cell, each cell has 6 32 bit integers */ |
1119 | dm += (drconf_cell_cnt - 1) * 6; | |
1120 | read_drconf_cell(&drmem, &dm); | |
e70bd3ae | 1121 | of_node_put(memory); |
45b64ee6 | 1122 | return drmem.base_addr + lmb_size; |
e70bd3ae | 1123 | } |
45b64ee6 | 1124 | return 0; |
cd34206e NA |
1125 | } |
1126 | ||
1127 | /* | |
1128 | * memory_hotplug_max - return max address of memory that may be added | |
1129 | * | |
1130 | * This is currently only used on systems that support drconfig memory | |
1131 | * hotplug. | |
1132 | */ | |
1133 | u64 memory_hotplug_max(void) | |
1134 | { | |
1135 | return max(hot_add_drconf_memory_max(), memblock_end_of_DRAM()); | |
1136 | } | |
237a0989 | 1137 | #endif /* CONFIG_MEMORY_HOTPLUG */ |
9eff1a38 | 1138 | |
bd03403a | 1139 | /* Virtual Processor Home Node (VPHN) support */ |
39bf990e | 1140 | #ifdef CONFIG_PPC_SPLPAR |
4b6cfb2a GK |
1141 | |
1142 | #include "vphn.h" | |
1143 | ||
30c05350 NF |
1144 | struct topology_update_data { |
1145 | struct topology_update_data *next; | |
1146 | unsigned int cpu; | |
1147 | int old_nid; | |
1148 | int new_nid; | |
1149 | }; | |
1150 | ||
cee5405d MB |
1151 | #define TOPOLOGY_DEF_TIMER_SECS 60 |
1152 | ||
5de16699 | 1153 | static u8 vphn_cpu_change_counts[NR_CPUS][MAX_DISTANCE_REF_POINTS]; |
9eff1a38 JL |
1154 | static cpumask_t cpu_associativity_changes_mask; |
1155 | static int vphn_enabled; | |
5d88aa85 JL |
1156 | static int prrn_enabled; |
1157 | static void reset_topology_timer(void); | |
cee5405d | 1158 | static int topology_timer_secs = 1; |
17f444c0 MB |
1159 | static int topology_inited; |
1160 | static int topology_update_needed; | |
9eff1a38 | 1161 | |
cee5405d MB |
1162 | /* |
1163 | * Change polling interval for associativity changes. | |
1164 | */ | |
1165 | int timed_topology_update(int nsecs) | |
1166 | { | |
1167 | if (vphn_enabled) { | |
1168 | if (nsecs > 0) | |
1169 | topology_timer_secs = nsecs; | |
1170 | else | |
1171 | topology_timer_secs = TOPOLOGY_DEF_TIMER_SECS; | |
1172 | ||
1173 | reset_topology_timer(); | |
1174 | } | |
1175 | ||
1176 | return 0; | |
1177 | } | |
9eff1a38 JL |
1178 | |
1179 | /* | |
1180 | * Store the current values of the associativity change counters in the | |
1181 | * hypervisor. | |
1182 | */ | |
1183 | static void setup_cpu_associativity_change_counters(void) | |
1184 | { | |
cd9d6cc7 | 1185 | int cpu; |
9eff1a38 | 1186 | |
5de16699 AB |
1187 | /* The VPHN feature supports a maximum of 8 reference points */ |
1188 | BUILD_BUG_ON(MAX_DISTANCE_REF_POINTS > 8); | |
1189 | ||
9eff1a38 | 1190 | for_each_possible_cpu(cpu) { |
cd9d6cc7 | 1191 | int i; |
9eff1a38 JL |
1192 | u8 *counts = vphn_cpu_change_counts[cpu]; |
1193 | volatile u8 *hypervisor_counts = lppaca[cpu].vphn_assoc_counts; | |
1194 | ||
5de16699 | 1195 | for (i = 0; i < distance_ref_points_depth; i++) |
9eff1a38 | 1196 | counts[i] = hypervisor_counts[i]; |
9eff1a38 JL |
1197 | } |
1198 | } | |
1199 | ||
1200 | /* | |
1201 | * The hypervisor maintains a set of 8 associativity change counters in | |
1202 | * the VPA of each cpu that correspond to the associativity levels in the | |
1203 | * ibm,associativity-reference-points property. When an associativity | |
1204 | * level changes, the corresponding counter is incremented. | |
1205 | * | |
1206 | * Set a bit in cpu_associativity_changes_mask for each cpu whose home | |
1207 | * node associativity levels have changed. | |
1208 | * | |
1209 | * Returns the number of cpus with unhandled associativity changes. | |
1210 | */ | |
1211 | static int update_cpu_associativity_changes_mask(void) | |
1212 | { | |
5d88aa85 | 1213 | int cpu; |
9eff1a38 JL |
1214 | cpumask_t *changes = &cpu_associativity_changes_mask; |
1215 | ||
9eff1a38 JL |
1216 | for_each_possible_cpu(cpu) { |
1217 | int i, changed = 0; | |
1218 | u8 *counts = vphn_cpu_change_counts[cpu]; | |
1219 | volatile u8 *hypervisor_counts = lppaca[cpu].vphn_assoc_counts; | |
1220 | ||
5de16699 | 1221 | for (i = 0; i < distance_ref_points_depth; i++) { |
d69043e8 | 1222 | if (hypervisor_counts[i] != counts[i]) { |
9eff1a38 JL |
1223 | counts[i] = hypervisor_counts[i]; |
1224 | changed = 1; | |
1225 | } | |
1226 | } | |
1227 | if (changed) { | |
3be7db6a RJ |
1228 | cpumask_or(changes, changes, cpu_sibling_mask(cpu)); |
1229 | cpu = cpu_last_thread_sibling(cpu); | |
9eff1a38 JL |
1230 | } |
1231 | } | |
1232 | ||
5d88aa85 | 1233 | return cpumask_weight(changes); |
9eff1a38 JL |
1234 | } |
1235 | ||
9eff1a38 JL |
1236 | /* |
1237 | * Retrieve the new associativity information for a virtual processor's | |
1238 | * home node. | |
1239 | */ | |
b08a2a12 | 1240 | static long hcall_vphn(unsigned long cpu, __be32 *associativity) |
9eff1a38 | 1241 | { |
cd9d6cc7 | 1242 | long rc; |
9eff1a38 JL |
1243 | long retbuf[PLPAR_HCALL9_BUFSIZE] = {0}; |
1244 | u64 flags = 1; | |
1245 | int hwcpu = get_hard_smp_processor_id(cpu); | |
1246 | ||
1247 | rc = plpar_hcall9(H_HOME_NODE_ASSOCIATIVITY, retbuf, flags, hwcpu); | |
1248 | vphn_unpack_associativity(retbuf, associativity); | |
1249 | ||
1250 | return rc; | |
1251 | } | |
1252 | ||
1253 | static long vphn_get_associativity(unsigned long cpu, | |
b08a2a12 | 1254 | __be32 *associativity) |
9eff1a38 | 1255 | { |
cd9d6cc7 | 1256 | long rc; |
9eff1a38 JL |
1257 | |
1258 | rc = hcall_vphn(cpu, associativity); | |
1259 | ||
1260 | switch (rc) { | |
1261 | case H_FUNCTION: | |
1262 | printk(KERN_INFO | |
1263 | "VPHN is not supported. Disabling polling...\n"); | |
1264 | stop_topology_update(); | |
1265 | break; | |
1266 | case H_HARDWARE: | |
1267 | printk(KERN_ERR | |
1268 | "hcall_vphn() experienced a hardware fault " | |
1269 | "preventing VPHN. Disabling polling...\n"); | |
1270 | stop_topology_update(); | |
17f444c0 MB |
1271 | break; |
1272 | case H_SUCCESS: | |
1273 | dbg("VPHN hcall succeeded. Reset polling...\n"); | |
cee5405d | 1274 | timed_topology_update(0); |
17f444c0 | 1275 | break; |
9eff1a38 JL |
1276 | } |
1277 | ||
1278 | return rc; | |
1279 | } | |
1280 | ||
30c05350 NF |
1281 | /* |
1282 | * Update the CPU maps and sysfs entries for a single CPU when its NUMA | |
1283 | * characteristics change. This function doesn't perform any locking and is | |
1284 | * only safe to call from stop_machine(). | |
1285 | */ | |
1286 | static int update_cpu_topology(void *data) | |
1287 | { | |
1288 | struct topology_update_data *update; | |
1289 | unsigned long cpu; | |
1290 | ||
1291 | if (!data) | |
1292 | return -EINVAL; | |
1293 | ||
3be7db6a | 1294 | cpu = smp_processor_id(); |
30c05350 NF |
1295 | |
1296 | for (update = data; update; update = update->next) { | |
2c0a33f9 | 1297 | int new_nid = update->new_nid; |
30c05350 NF |
1298 | if (cpu != update->cpu) |
1299 | continue; | |
1300 | ||
49f8d8c0 | 1301 | unmap_cpu_from_node(cpu); |
2c0a33f9 NA |
1302 | map_cpu_to_node(cpu, new_nid); |
1303 | set_cpu_numa_node(cpu, new_nid); | |
1304 | set_cpu_numa_mem(cpu, local_memory_node(new_nid)); | |
176bbf14 | 1305 | vdso_getcpu_init(); |
30c05350 NF |
1306 | } |
1307 | ||
1308 | return 0; | |
1309 | } | |
1310 | ||
d4edc5b6 SB |
1311 | static int update_lookup_table(void *data) |
1312 | { | |
1313 | struct topology_update_data *update; | |
1314 | ||
1315 | if (!data) | |
1316 | return -EINVAL; | |
1317 | ||
1318 | /* | |
1319 | * Upon topology update, the numa-cpu lookup table needs to be updated | |
1320 | * for all threads in the core, including offline CPUs, to ensure that | |
1321 | * future hotplug operations respect the cpu-to-node associativity | |
1322 | * properly. | |
1323 | */ | |
1324 | for (update = data; update; update = update->next) { | |
1325 | int nid, base, j; | |
1326 | ||
1327 | nid = update->new_nid; | |
1328 | base = cpu_first_thread_sibling(update->cpu); | |
1329 | ||
1330 | for (j = 0; j < threads_per_core; j++) { | |
1331 | update_numa_cpu_lookup_table(base + j, nid); | |
1332 | } | |
1333 | } | |
1334 | ||
1335 | return 0; | |
1336 | } | |
1337 | ||
9eff1a38 JL |
1338 | /* |
1339 | * Update the node maps and sysfs entries for each cpu whose home node | |
79c5fceb | 1340 | * has changed. Returns 1 when the topology has changed, and 0 otherwise. |
3e401f7a TJB |
1341 | * |
1342 | * cpus_locked says whether we already hold cpu_hotplug_lock. | |
9eff1a38 | 1343 | */ |
3e401f7a | 1344 | int numa_update_cpu_topology(bool cpus_locked) |
9eff1a38 | 1345 | { |
3be7db6a | 1346 | unsigned int cpu, sibling, changed = 0; |
30c05350 | 1347 | struct topology_update_data *updates, *ud; |
b08a2a12 | 1348 | __be32 associativity[VPHN_ASSOC_BUFSIZE] = {0}; |
176bbf14 | 1349 | cpumask_t updated_cpus; |
8a25a2fd | 1350 | struct device *dev; |
3be7db6a | 1351 | int weight, new_nid, i = 0; |
9eff1a38 | 1352 | |
17f444c0 MB |
1353 | if (!prrn_enabled && !vphn_enabled) { |
1354 | if (!topology_inited) | |
1355 | topology_update_needed = 1; | |
2d73bae1 | 1356 | return 0; |
17f444c0 | 1357 | } |
2d73bae1 | 1358 | |
30c05350 NF |
1359 | weight = cpumask_weight(&cpu_associativity_changes_mask); |
1360 | if (!weight) | |
1361 | return 0; | |
1362 | ||
1363 | updates = kzalloc(weight * (sizeof(*updates)), GFP_KERNEL); | |
1364 | if (!updates) | |
1365 | return 0; | |
9eff1a38 | 1366 | |
176bbf14 JL |
1367 | cpumask_clear(&updated_cpus); |
1368 | ||
5d88aa85 | 1369 | for_each_cpu(cpu, &cpu_associativity_changes_mask) { |
3be7db6a RJ |
1370 | /* |
1371 | * If siblings aren't flagged for changes, updates list | |
1372 | * will be too short. Skip on this update and set for next | |
1373 | * update. | |
1374 | */ | |
1375 | if (!cpumask_subset(cpu_sibling_mask(cpu), | |
1376 | &cpu_associativity_changes_mask)) { | |
1377 | pr_info("Sibling bits not set for associativity " | |
1378 | "change, cpu%d\n", cpu); | |
1379 | cpumask_or(&cpu_associativity_changes_mask, | |
1380 | &cpu_associativity_changes_mask, | |
1381 | cpu_sibling_mask(cpu)); | |
1382 | cpu = cpu_last_thread_sibling(cpu); | |
1383 | continue; | |
1384 | } | |
9eff1a38 | 1385 | |
3be7db6a RJ |
1386 | /* Use associativity from first thread for all siblings */ |
1387 | vphn_get_associativity(cpu, associativity); | |
1388 | new_nid = associativity_to_nid(associativity); | |
1389 | if (new_nid < 0 || !node_online(new_nid)) | |
1390 | new_nid = first_online_node; | |
1391 | ||
1392 | if (new_nid == numa_cpu_lookup_table[cpu]) { | |
1393 | cpumask_andnot(&cpu_associativity_changes_mask, | |
1394 | &cpu_associativity_changes_mask, | |
1395 | cpu_sibling_mask(cpu)); | |
17f444c0 MB |
1396 | dbg("Assoc chg gives same node %d for cpu%d\n", |
1397 | new_nid, cpu); | |
3be7db6a RJ |
1398 | cpu = cpu_last_thread_sibling(cpu); |
1399 | continue; | |
1400 | } | |
9eff1a38 | 1401 | |
3be7db6a RJ |
1402 | for_each_cpu(sibling, cpu_sibling_mask(cpu)) { |
1403 | ud = &updates[i++]; | |
8bc93149 | 1404 | ud->next = &updates[i]; |
3be7db6a RJ |
1405 | ud->cpu = sibling; |
1406 | ud->new_nid = new_nid; | |
1407 | ud->old_nid = numa_cpu_lookup_table[sibling]; | |
1408 | cpumask_set_cpu(sibling, &updated_cpus); | |
3be7db6a RJ |
1409 | } |
1410 | cpu = cpu_last_thread_sibling(cpu); | |
30c05350 NF |
1411 | } |
1412 | ||
8bc93149 MB |
1413 | /* |
1414 | * Prevent processing of 'updates' from overflowing array | |
1415 | * where last entry filled in a 'next' pointer. | |
1416 | */ | |
1417 | if (i) | |
1418 | updates[i-1].next = NULL; | |
1419 | ||
2d73bae1 NA |
1420 | pr_debug("Topology update for the following CPUs:\n"); |
1421 | if (cpumask_weight(&updated_cpus)) { | |
1422 | for (ud = &updates[0]; ud; ud = ud->next) { | |
1423 | pr_debug("cpu %d moving from node %d " | |
1424 | "to %d\n", ud->cpu, | |
1425 | ud->old_nid, ud->new_nid); | |
1426 | } | |
1427 | } | |
1428 | ||
9a013361 MW |
1429 | /* |
1430 | * In cases where we have nothing to update (because the updates list | |
1431 | * is too short or because the new topology is same as the old one), | |
1432 | * skip invoking update_cpu_topology() via stop-machine(). This is | |
1433 | * necessary (and not just a fast-path optimization) since stop-machine | |
1434 | * can end up electing a random CPU to run update_cpu_topology(), and | |
1435 | * thus trick us into setting up incorrect cpu-node mappings (since | |
1436 | * 'updates' is kzalloc()'ed). | |
1437 | * | |
1438 | * And for the similar reason, we will skip all the following updating. | |
1439 | */ | |
1440 | if (!cpumask_weight(&updated_cpus)) | |
1441 | goto out; | |
1442 | ||
3e401f7a TJB |
1443 | if (cpus_locked) |
1444 | stop_machine_cpuslocked(update_cpu_topology, &updates[0], | |
1445 | &updated_cpus); | |
1446 | else | |
1447 | stop_machine(update_cpu_topology, &updates[0], &updated_cpus); | |
30c05350 | 1448 | |
d4edc5b6 SB |
1449 | /* |
1450 | * Update the numa-cpu lookup table with the new mappings, even for | |
1451 | * offline CPUs. It is best to perform this update from the stop- | |
1452 | * machine context. | |
1453 | */ | |
3e401f7a TJB |
1454 | if (cpus_locked) |
1455 | stop_machine_cpuslocked(update_lookup_table, &updates[0], | |
d4edc5b6 | 1456 | cpumask_of(raw_smp_processor_id())); |
3e401f7a TJB |
1457 | else |
1458 | stop_machine(update_lookup_table, &updates[0], | |
1459 | cpumask_of(raw_smp_processor_id())); | |
d4edc5b6 | 1460 | |
30c05350 | 1461 | for (ud = &updates[0]; ud; ud = ud->next) { |
dd023217 NF |
1462 | unregister_cpu_under_node(ud->cpu, ud->old_nid); |
1463 | register_cpu_under_node(ud->cpu, ud->new_nid); | |
1464 | ||
30c05350 | 1465 | dev = get_cpu_device(ud->cpu); |
8a25a2fd KS |
1466 | if (dev) |
1467 | kobject_uevent(&dev->kobj, KOBJ_CHANGE); | |
30c05350 | 1468 | cpumask_clear_cpu(ud->cpu, &cpu_associativity_changes_mask); |
79c5fceb | 1469 | changed = 1; |
9eff1a38 JL |
1470 | } |
1471 | ||
9a013361 | 1472 | out: |
30c05350 | 1473 | kfree(updates); |
17f444c0 | 1474 | topology_update_needed = 0; |
79c5fceb | 1475 | return changed; |
9eff1a38 JL |
1476 | } |
1477 | ||
3e401f7a TJB |
1478 | int arch_update_cpu_topology(void) |
1479 | { | |
3e401f7a TJB |
1480 | return numa_update_cpu_topology(true); |
1481 | } | |
1482 | ||
9eff1a38 JL |
1483 | static void topology_work_fn(struct work_struct *work) |
1484 | { | |
1485 | rebuild_sched_domains(); | |
1486 | } | |
1487 | static DECLARE_WORK(topology_work, topology_work_fn); | |
1488 | ||
ec32dd66 | 1489 | static void topology_schedule_update(void) |
9eff1a38 JL |
1490 | { |
1491 | schedule_work(&topology_work); | |
1492 | } | |
1493 | ||
df7e828c | 1494 | static void topology_timer_fn(struct timer_list *unused) |
9eff1a38 | 1495 | { |
5d88aa85 | 1496 | if (prrn_enabled && cpumask_weight(&cpu_associativity_changes_mask)) |
9eff1a38 | 1497 | topology_schedule_update(); |
5d88aa85 JL |
1498 | else if (vphn_enabled) { |
1499 | if (update_cpu_associativity_changes_mask() > 0) | |
1500 | topology_schedule_update(); | |
1501 | reset_topology_timer(); | |
1502 | } | |
9eff1a38 | 1503 | } |
df7e828c | 1504 | static struct timer_list topology_timer; |
9eff1a38 | 1505 | |
5d88aa85 | 1506 | static void reset_topology_timer(void) |
9eff1a38 | 1507 | { |
5b0e2cb0 | 1508 | mod_timer(&topology_timer, jiffies + topology_timer_secs * HZ); |
9eff1a38 JL |
1509 | } |
1510 | ||
601abdc3 NF |
1511 | #ifdef CONFIG_SMP |
1512 | ||
5d88aa85 JL |
1513 | static void stage_topology_update(int core_id) |
1514 | { | |
1515 | cpumask_or(&cpu_associativity_changes_mask, | |
1516 | &cpu_associativity_changes_mask, cpu_sibling_mask(core_id)); | |
1517 | reset_topology_timer(); | |
1518 | } | |
1519 | ||
1520 | static int dt_update_callback(struct notifier_block *nb, | |
1521 | unsigned long action, void *data) | |
1522 | { | |
f5242e5a | 1523 | struct of_reconfig_data *update = data; |
5d88aa85 JL |
1524 | int rc = NOTIFY_DONE; |
1525 | ||
1526 | switch (action) { | |
5d88aa85 | 1527 | case OF_RECONFIG_UPDATE_PROPERTY: |
30c05350 NF |
1528 | if (!of_prop_cmp(update->dn->type, "cpu") && |
1529 | !of_prop_cmp(update->prop->name, "ibm,associativity")) { | |
5d88aa85 JL |
1530 | u32 core_id; |
1531 | of_property_read_u32(update->dn, "reg", &core_id); | |
1532 | stage_topology_update(core_id); | |
1533 | rc = NOTIFY_OK; | |
1534 | } | |
1535 | break; | |
1536 | } | |
1537 | ||
1538 | return rc; | |
9eff1a38 JL |
1539 | } |
1540 | ||
5d88aa85 JL |
1541 | static struct notifier_block dt_update_nb = { |
1542 | .notifier_call = dt_update_callback, | |
1543 | }; | |
1544 | ||
601abdc3 NF |
1545 | #endif |
1546 | ||
9eff1a38 | 1547 | /* |
5d88aa85 | 1548 | * Start polling for associativity changes. |
9eff1a38 JL |
1549 | */ |
1550 | int start_topology_update(void) | |
1551 | { | |
1552 | int rc = 0; | |
1553 | ||
5d88aa85 JL |
1554 | if (firmware_has_feature(FW_FEATURE_PRRN)) { |
1555 | if (!prrn_enabled) { | |
1556 | prrn_enabled = 1; | |
601abdc3 | 1557 | #ifdef CONFIG_SMP |
5d88aa85 | 1558 | rc = of_reconfig_notifier_register(&dt_update_nb); |
601abdc3 | 1559 | #endif |
5d88aa85 | 1560 | } |
a3496e91 MB |
1561 | } |
1562 | if (firmware_has_feature(FW_FEATURE_VPHN) && | |
f13c13a0 | 1563 | lppaca_shared_proc(get_lppaca())) { |
5d88aa85 | 1564 | if (!vphn_enabled) { |
5d88aa85 JL |
1565 | vphn_enabled = 1; |
1566 | setup_cpu_associativity_change_counters(); | |
df7e828c KC |
1567 | timer_setup(&topology_timer, topology_timer_fn, |
1568 | TIMER_DEFERRABLE); | |
5d88aa85 JL |
1569 | reset_topology_timer(); |
1570 | } | |
9eff1a38 JL |
1571 | } |
1572 | ||
1573 | return rc; | |
1574 | } | |
9eff1a38 JL |
1575 | |
1576 | /* | |
1577 | * Disable polling for VPHN associativity changes. | |
1578 | */ | |
1579 | int stop_topology_update(void) | |
1580 | { | |
5d88aa85 JL |
1581 | int rc = 0; |
1582 | ||
1583 | if (prrn_enabled) { | |
1584 | prrn_enabled = 0; | |
601abdc3 | 1585 | #ifdef CONFIG_SMP |
5d88aa85 | 1586 | rc = of_reconfig_notifier_unregister(&dt_update_nb); |
601abdc3 | 1587 | #endif |
a3496e91 MB |
1588 | } |
1589 | if (vphn_enabled) { | |
5d88aa85 JL |
1590 | vphn_enabled = 0; |
1591 | rc = del_timer_sync(&topology_timer); | |
1592 | } | |
1593 | ||
1594 | return rc; | |
9eff1a38 | 1595 | } |
e04fa612 NF |
1596 | |
1597 | int prrn_is_enabled(void) | |
1598 | { | |
1599 | return prrn_enabled; | |
1600 | } | |
1601 | ||
1602 | static int topology_read(struct seq_file *file, void *v) | |
1603 | { | |
1604 | if (vphn_enabled || prrn_enabled) | |
1605 | seq_puts(file, "on\n"); | |
1606 | else | |
1607 | seq_puts(file, "off\n"); | |
1608 | ||
1609 | return 0; | |
1610 | } | |
1611 | ||
1612 | static int topology_open(struct inode *inode, struct file *file) | |
1613 | { | |
1614 | return single_open(file, topology_read, NULL); | |
1615 | } | |
1616 | ||
1617 | static ssize_t topology_write(struct file *file, const char __user *buf, | |
1618 | size_t count, loff_t *off) | |
1619 | { | |
1620 | char kbuf[4]; /* "on" or "off" plus null. */ | |
1621 | int read_len; | |
1622 | ||
1623 | read_len = count < 3 ? count : 3; | |
1624 | if (copy_from_user(kbuf, buf, read_len)) | |
1625 | return -EINVAL; | |
1626 | ||
1627 | kbuf[read_len] = '\0'; | |
1628 | ||
1629 | if (!strncmp(kbuf, "on", 2)) | |
1630 | start_topology_update(); | |
1631 | else if (!strncmp(kbuf, "off", 3)) | |
1632 | stop_topology_update(); | |
1633 | else | |
1634 | return -EINVAL; | |
1635 | ||
1636 | return count; | |
1637 | } | |
1638 | ||
1639 | static const struct file_operations topology_ops = { | |
1640 | .read = seq_read, | |
1641 | .write = topology_write, | |
1642 | .open = topology_open, | |
1643 | .release = single_release | |
1644 | }; | |
1645 | ||
1646 | static int topology_update_init(void) | |
1647 | { | |
2d73bae1 NA |
1648 | /* Do not poll for changes if disabled at boot */ |
1649 | if (topology_updates_enabled) | |
1650 | start_topology_update(); | |
1651 | ||
17f444c0 MB |
1652 | if (vphn_enabled) |
1653 | topology_schedule_update(); | |
1654 | ||
2d15b9b4 NA |
1655 | if (!proc_create("powerpc/topology_updates", 0644, NULL, &topology_ops)) |
1656 | return -ENOMEM; | |
e04fa612 | 1657 | |
17f444c0 MB |
1658 | topology_inited = 1; |
1659 | if (topology_update_needed) | |
1660 | bitmap_fill(cpumask_bits(&cpu_associativity_changes_mask), | |
1661 | nr_cpumask_bits); | |
1662 | ||
e04fa612 | 1663 | return 0; |
9eff1a38 | 1664 | } |
e04fa612 | 1665 | device_initcall(topology_update_init); |
39bf990e | 1666 | #endif /* CONFIG_PPC_SPLPAR */ |