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 | 13 | #include <linux/threads.h> |
57c8a661 | 14 | #include <linux/memblock.h> |
1da177e4 LT |
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> | |
6df1646e | 22 | #include <linux/of.h> |
06eccea6 | 23 | #include <linux/pfn.h> |
9eff1a38 JL |
24 | #include <linux/cpuset.h> |
25 | #include <linux/node.h> | |
30c05350 | 26 | #include <linux/stop_machine.h> |
e04fa612 NF |
27 | #include <linux/proc_fs.h> |
28 | #include <linux/seq_file.h> | |
29 | #include <linux/uaccess.h> | |
191a7120 | 30 | #include <linux/slab.h> |
3be7db6a | 31 | #include <asm/cputhreads.h> |
45fb6cea | 32 | #include <asm/sparsemem.h> |
d9b2b2a2 | 33 | #include <asm/prom.h> |
2249ca9d | 34 | #include <asm/smp.h> |
d4edc5b6 | 35 | #include <asm/topology.h> |
9eff1a38 JL |
36 | #include <asm/firmware.h> |
37 | #include <asm/paca.h> | |
39bf990e | 38 | #include <asm/hvcall.h> |
ae3a197e | 39 | #include <asm/setup.h> |
176bbf14 | 40 | #include <asm/vdso.h> |
514a9cb3 | 41 | #include <asm/drmem.h> |
1da177e4 LT |
42 | |
43 | static int numa_enabled = 1; | |
44 | ||
1daa6d08 BS |
45 | static char *cmdline __initdata; |
46 | ||
1da177e4 LT |
47 | static int numa_debug; |
48 | #define dbg(args...) if (numa_debug) { printk(KERN_INFO args); } | |
49 | ||
45fb6cea | 50 | int numa_cpu_lookup_table[NR_CPUS]; |
25863de0 | 51 | cpumask_var_t node_to_cpumask_map[MAX_NUMNODES]; |
1da177e4 | 52 | struct pglist_data *node_data[MAX_NUMNODES]; |
45fb6cea AB |
53 | |
54 | EXPORT_SYMBOL(numa_cpu_lookup_table); | |
25863de0 | 55 | EXPORT_SYMBOL(node_to_cpumask_map); |
45fb6cea AB |
56 | EXPORT_SYMBOL(node_data); |
57 | ||
1da177e4 | 58 | static int min_common_depth; |
237a0989 | 59 | static int n_mem_addr_cells, n_mem_size_cells; |
41eab6f8 AB |
60 | static int form1_affinity; |
61 | ||
62 | #define MAX_DISTANCE_REF_POINTS 4 | |
63 | static int distance_ref_points_depth; | |
b08a2a12 | 64 | static const __be32 *distance_ref_points; |
41eab6f8 | 65 | static int distance_lookup_table[MAX_NUMNODES][MAX_DISTANCE_REF_POINTS]; |
1da177e4 | 66 | |
25863de0 AB |
67 | /* |
68 | * Allocate node_to_cpumask_map based on number of available nodes | |
69 | * Requires node_possible_map to be valid. | |
70 | * | |
9512938b | 71 | * Note: cpumask_of_node() is not valid until after this is done. |
25863de0 AB |
72 | */ |
73 | static void __init setup_node_to_cpumask_map(void) | |
74 | { | |
f9d531b8 | 75 | unsigned int node; |
25863de0 AB |
76 | |
77 | /* setup nr_node_ids if not done yet */ | |
f9d531b8 CS |
78 | if (nr_node_ids == MAX_NUMNODES) |
79 | setup_nr_node_ids(); | |
25863de0 AB |
80 | |
81 | /* allocate the map */ | |
c118baf8 | 82 | for_each_node(node) |
25863de0 AB |
83 | alloc_bootmem_cpumask_var(&node_to_cpumask_map[node]); |
84 | ||
85 | /* cpumask_of_node() will now work */ | |
b9726c26 | 86 | dbg("Node to cpumask map for %u nodes\n", nr_node_ids); |
25863de0 AB |
87 | } |
88 | ||
55671f3c | 89 | static int __init fake_numa_create_new_node(unsigned long end_pfn, |
1daa6d08 BS |
90 | unsigned int *nid) |
91 | { | |
92 | unsigned long long mem; | |
93 | char *p = cmdline; | |
94 | static unsigned int fake_nid; | |
95 | static unsigned long long curr_boundary; | |
96 | ||
97 | /* | |
98 | * Modify node id, iff we started creating NUMA nodes | |
99 | * We want to continue from where we left of the last time | |
100 | */ | |
101 | if (fake_nid) | |
102 | *nid = fake_nid; | |
103 | /* | |
104 | * In case there are no more arguments to parse, the | |
105 | * node_id should be the same as the last fake node id | |
106 | * (we've handled this above). | |
107 | */ | |
108 | if (!p) | |
109 | return 0; | |
110 | ||
111 | mem = memparse(p, &p); | |
112 | if (!mem) | |
113 | return 0; | |
114 | ||
115 | if (mem < curr_boundary) | |
116 | return 0; | |
117 | ||
118 | curr_boundary = mem; | |
119 | ||
120 | if ((end_pfn << PAGE_SHIFT) > mem) { | |
121 | /* | |
122 | * Skip commas and spaces | |
123 | */ | |
124 | while (*p == ',' || *p == ' ' || *p == '\t') | |
125 | p++; | |
126 | ||
127 | cmdline = p; | |
128 | fake_nid++; | |
129 | *nid = fake_nid; | |
130 | dbg("created new fake_node with id %d\n", fake_nid); | |
131 | return 1; | |
132 | } | |
133 | return 0; | |
134 | } | |
135 | ||
d4edc5b6 SB |
136 | static void reset_numa_cpu_lookup_table(void) |
137 | { | |
138 | unsigned int cpu; | |
139 | ||
140 | for_each_possible_cpu(cpu) | |
141 | numa_cpu_lookup_table[cpu] = -1; | |
142 | } | |
143 | ||
d4edc5b6 SB |
144 | static void map_cpu_to_node(int cpu, int node) |
145 | { | |
146 | update_numa_cpu_lookup_table(cpu, node); | |
45fb6cea | 147 | |
bf4b85b0 NL |
148 | dbg("adding cpu %d to node %d\n", cpu, node); |
149 | ||
25863de0 AB |
150 | if (!(cpumask_test_cpu(cpu, node_to_cpumask_map[node]))) |
151 | cpumask_set_cpu(cpu, node_to_cpumask_map[node]); | |
1da177e4 LT |
152 | } |
153 | ||
39bf990e | 154 | #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_PPC_SPLPAR) |
1da177e4 LT |
155 | static void unmap_cpu_from_node(unsigned long cpu) |
156 | { | |
157 | int node = numa_cpu_lookup_table[cpu]; | |
158 | ||
159 | dbg("removing cpu %lu from node %d\n", cpu, node); | |
160 | ||
25863de0 | 161 | if (cpumask_test_cpu(cpu, node_to_cpumask_map[node])) { |
429f4d8d | 162 | cpumask_clear_cpu(cpu, node_to_cpumask_map[node]); |
1da177e4 LT |
163 | } else { |
164 | printk(KERN_ERR "WARNING: cpu %lu not found in node %d\n", | |
165 | cpu, node); | |
166 | } | |
167 | } | |
39bf990e | 168 | #endif /* CONFIG_HOTPLUG_CPU || CONFIG_PPC_SPLPAR */ |
1da177e4 | 169 | |
d62c8dee NR |
170 | int cpu_distance(__be32 *cpu1_assoc, __be32 *cpu2_assoc) |
171 | { | |
172 | int dist = 0; | |
173 | ||
174 | int i, index; | |
175 | ||
176 | for (i = 0; i < distance_ref_points_depth; i++) { | |
177 | index = be32_to_cpu(distance_ref_points[i]); | |
178 | if (cpu1_assoc[index] == cpu2_assoc[index]) | |
179 | break; | |
180 | dist++; | |
181 | } | |
182 | ||
183 | return dist; | |
184 | } | |
185 | ||
1da177e4 | 186 | /* must hold reference to node during call */ |
b08a2a12 | 187 | static const __be32 *of_get_associativity(struct device_node *dev) |
1da177e4 | 188 | { |
e2eb6392 | 189 | return of_get_property(dev, "ibm,associativity", NULL); |
1da177e4 LT |
190 | } |
191 | ||
41eab6f8 AB |
192 | int __node_distance(int a, int b) |
193 | { | |
194 | int i; | |
195 | int distance = LOCAL_DISTANCE; | |
196 | ||
197 | if (!form1_affinity) | |
7122beee | 198 | return ((a == b) ? LOCAL_DISTANCE : REMOTE_DISTANCE); |
41eab6f8 AB |
199 | |
200 | for (i = 0; i < distance_ref_points_depth; i++) { | |
201 | if (distance_lookup_table[a][i] == distance_lookup_table[b][i]) | |
202 | break; | |
203 | ||
204 | /* Double the distance for each NUMA level */ | |
205 | distance *= 2; | |
206 | } | |
207 | ||
208 | return distance; | |
209 | } | |
12c743eb | 210 | EXPORT_SYMBOL(__node_distance); |
41eab6f8 AB |
211 | |
212 | static void initialize_distance_lookup_table(int nid, | |
b08a2a12 | 213 | const __be32 *associativity) |
41eab6f8 AB |
214 | { |
215 | int i; | |
216 | ||
217 | if (!form1_affinity) | |
218 | return; | |
219 | ||
220 | for (i = 0; i < distance_ref_points_depth; i++) { | |
b08a2a12 AP |
221 | const __be32 *entry; |
222 | ||
1d805440 | 223 | entry = &associativity[be32_to_cpu(distance_ref_points[i]) - 1]; |
b08a2a12 | 224 | distance_lookup_table[nid][i] = of_read_number(entry, 1); |
41eab6f8 AB |
225 | } |
226 | } | |
227 | ||
482ec7c4 NL |
228 | /* Returns nid in the range [0..MAX_NUMNODES-1], or -1 if no useful numa |
229 | * info is found. | |
230 | */ | |
b08a2a12 | 231 | static int associativity_to_nid(const __be32 *associativity) |
1da177e4 | 232 | { |
98fa15f3 | 233 | int nid = NUMA_NO_NODE; |
1da177e4 | 234 | |
495c2ff4 | 235 | if (!numa_enabled) |
482ec7c4 | 236 | goto out; |
1da177e4 | 237 | |
b08a2a12 AP |
238 | if (of_read_number(associativity, 1) >= min_common_depth) |
239 | nid = of_read_number(&associativity[min_common_depth], 1); | |
bc16a759 NL |
240 | |
241 | /* POWER4 LPAR uses 0xffff as invalid node */ | |
482ec7c4 | 242 | if (nid == 0xffff || nid >= MAX_NUMNODES) |
98fa15f3 | 243 | nid = NUMA_NO_NODE; |
41eab6f8 | 244 | |
b08a2a12 | 245 | if (nid > 0 && |
1d805440 ND |
246 | of_read_number(associativity, 1) >= distance_ref_points_depth) { |
247 | /* | |
248 | * Skip the length field and send start of associativity array | |
249 | */ | |
250 | initialize_distance_lookup_table(nid, associativity + 1); | |
251 | } | |
41eab6f8 | 252 | |
482ec7c4 | 253 | out: |
cf950b7a | 254 | return nid; |
1da177e4 LT |
255 | } |
256 | ||
9eff1a38 JL |
257 | /* Returns the nid associated with the given device tree node, |
258 | * or -1 if not found. | |
259 | */ | |
260 | static int of_node_to_nid_single(struct device_node *device) | |
261 | { | |
98fa15f3 | 262 | int nid = NUMA_NO_NODE; |
b08a2a12 | 263 | const __be32 *tmp; |
9eff1a38 JL |
264 | |
265 | tmp = of_get_associativity(device); | |
266 | if (tmp) | |
267 | nid = associativity_to_nid(tmp); | |
268 | return nid; | |
269 | } | |
270 | ||
953039c8 JK |
271 | /* Walk the device tree upwards, looking for an associativity id */ |
272 | int of_node_to_nid(struct device_node *device) | |
273 | { | |
98fa15f3 | 274 | int nid = NUMA_NO_NODE; |
953039c8 JK |
275 | |
276 | of_node_get(device); | |
277 | while (device) { | |
278 | nid = of_node_to_nid_single(device); | |
279 | if (nid != -1) | |
280 | break; | |
281 | ||
1def3758 | 282 | device = of_get_next_parent(device); |
953039c8 JK |
283 | } |
284 | of_node_put(device); | |
285 | ||
286 | return nid; | |
287 | } | |
be9ba9ff | 288 | EXPORT_SYMBOL(of_node_to_nid); |
953039c8 | 289 | |
1da177e4 LT |
290 | static int __init find_min_common_depth(void) |
291 | { | |
41eab6f8 | 292 | int depth; |
e70606eb | 293 | struct device_node *root; |
1da177e4 | 294 | |
1c8ee733 DS |
295 | if (firmware_has_feature(FW_FEATURE_OPAL)) |
296 | root = of_find_node_by_path("/ibm,opal"); | |
297 | else | |
298 | root = of_find_node_by_path("/rtas"); | |
e70606eb ME |
299 | if (!root) |
300 | root = of_find_node_by_path("/"); | |
1da177e4 LT |
301 | |
302 | /* | |
41eab6f8 AB |
303 | * This property is a set of 32-bit integers, each representing |
304 | * an index into the ibm,associativity nodes. | |
305 | * | |
306 | * With form 0 affinity the first integer is for an SMP configuration | |
307 | * (should be all 0's) and the second is for a normal NUMA | |
308 | * configuration. We have only one level of NUMA. | |
309 | * | |
310 | * With form 1 affinity the first integer is the most significant | |
311 | * NUMA boundary and the following are progressively less significant | |
312 | * boundaries. There can be more than one level of NUMA. | |
1da177e4 | 313 | */ |
e70606eb | 314 | distance_ref_points = of_get_property(root, |
41eab6f8 AB |
315 | "ibm,associativity-reference-points", |
316 | &distance_ref_points_depth); | |
317 | ||
318 | if (!distance_ref_points) { | |
319 | dbg("NUMA: ibm,associativity-reference-points not found.\n"); | |
320 | goto err; | |
321 | } | |
322 | ||
323 | distance_ref_points_depth /= sizeof(int); | |
1da177e4 | 324 | |
8002b0c5 NF |
325 | if (firmware_has_feature(FW_FEATURE_OPAL) || |
326 | firmware_has_feature(FW_FEATURE_TYPE1_AFFINITY)) { | |
327 | dbg("Using form 1 affinity\n"); | |
1c8ee733 | 328 | form1_affinity = 1; |
4b83c330 AB |
329 | } |
330 | ||
41eab6f8 | 331 | if (form1_affinity) { |
b08a2a12 | 332 | depth = of_read_number(distance_ref_points, 1); |
1da177e4 | 333 | } else { |
41eab6f8 AB |
334 | if (distance_ref_points_depth < 2) { |
335 | printk(KERN_WARNING "NUMA: " | |
336 | "short ibm,associativity-reference-points\n"); | |
337 | goto err; | |
338 | } | |
339 | ||
b08a2a12 | 340 | depth = of_read_number(&distance_ref_points[1], 1); |
1da177e4 | 341 | } |
1da177e4 | 342 | |
41eab6f8 AB |
343 | /* |
344 | * Warn and cap if the hardware supports more than | |
345 | * MAX_DISTANCE_REF_POINTS domains. | |
346 | */ | |
347 | if (distance_ref_points_depth > MAX_DISTANCE_REF_POINTS) { | |
348 | printk(KERN_WARNING "NUMA: distance array capped at " | |
349 | "%d entries\n", MAX_DISTANCE_REF_POINTS); | |
350 | distance_ref_points_depth = MAX_DISTANCE_REF_POINTS; | |
351 | } | |
352 | ||
e70606eb | 353 | of_node_put(root); |
1da177e4 | 354 | return depth; |
41eab6f8 AB |
355 | |
356 | err: | |
e70606eb | 357 | of_node_put(root); |
41eab6f8 | 358 | return -1; |
1da177e4 LT |
359 | } |
360 | ||
84c9fdd1 | 361 | static void __init get_n_mem_cells(int *n_addr_cells, int *n_size_cells) |
1da177e4 LT |
362 | { |
363 | struct device_node *memory = NULL; | |
1da177e4 LT |
364 | |
365 | memory = of_find_node_by_type(memory, "memory"); | |
54c23310 | 366 | if (!memory) |
84c9fdd1 | 367 | panic("numa.c: No memory nodes found!"); |
54c23310 | 368 | |
a8bda5dd | 369 | *n_addr_cells = of_n_addr_cells(memory); |
9213feea | 370 | *n_size_cells = of_n_size_cells(memory); |
84c9fdd1 | 371 | of_node_put(memory); |
1da177e4 LT |
372 | } |
373 | ||
b08a2a12 | 374 | static unsigned long read_n_cells(int n, const __be32 **buf) |
1da177e4 LT |
375 | { |
376 | unsigned long result = 0; | |
377 | ||
378 | while (n--) { | |
b08a2a12 | 379 | result = (result << 32) | of_read_number(*buf, 1); |
1da177e4 LT |
380 | (*buf)++; |
381 | } | |
382 | return result; | |
383 | } | |
384 | ||
8342681d NF |
385 | struct assoc_arrays { |
386 | u32 n_arrays; | |
387 | u32 array_sz; | |
b08a2a12 | 388 | const __be32 *arrays; |
8342681d NF |
389 | }; |
390 | ||
391 | /* | |
25985edc | 392 | * Retrieve and validate the list of associativity arrays for drconf |
8342681d NF |
393 | * memory from the ibm,associativity-lookup-arrays property of the |
394 | * device tree.. | |
395 | * | |
396 | * The layout of the ibm,associativity-lookup-arrays property is a number N | |
397 | * indicating the number of associativity arrays, followed by a number M | |
398 | * indicating the size of each associativity array, followed by a list | |
399 | * of N associativity arrays. | |
400 | */ | |
35f80deb | 401 | static int of_get_assoc_arrays(struct assoc_arrays *aa) |
8342681d | 402 | { |
35f80deb | 403 | struct device_node *memory; |
b08a2a12 | 404 | const __be32 *prop; |
8342681d NF |
405 | u32 len; |
406 | ||
35f80deb NF |
407 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); |
408 | if (!memory) | |
409 | return -1; | |
410 | ||
8342681d | 411 | prop = of_get_property(memory, "ibm,associativity-lookup-arrays", &len); |
35f80deb NF |
412 | if (!prop || len < 2 * sizeof(unsigned int)) { |
413 | of_node_put(memory); | |
8342681d | 414 | return -1; |
35f80deb | 415 | } |
8342681d | 416 | |
b08a2a12 AP |
417 | aa->n_arrays = of_read_number(prop++, 1); |
418 | aa->array_sz = of_read_number(prop++, 1); | |
8342681d | 419 | |
35f80deb NF |
420 | of_node_put(memory); |
421 | ||
42b2aa86 | 422 | /* Now that we know the number of arrays and size of each array, |
8342681d NF |
423 | * revalidate the size of the property read in. |
424 | */ | |
425 | if (len < (aa->n_arrays * aa->array_sz + 2) * sizeof(unsigned int)) | |
426 | return -1; | |
427 | ||
428 | aa->arrays = prop; | |
429 | return 0; | |
430 | } | |
431 | ||
432 | /* | |
433 | * This is like of_node_to_nid_single() for memory represented in the | |
434 | * ibm,dynamic-reconfiguration-memory node. | |
435 | */ | |
514a9cb3 | 436 | static int of_drconf_to_nid_single(struct drmem_lmb *lmb) |
8342681d | 437 | { |
b88fc309 | 438 | struct assoc_arrays aa = { .arrays = NULL }; |
ea9f5b70 | 439 | int default_nid = NUMA_NO_NODE; |
8342681d | 440 | int nid = default_nid; |
b88fc309 NF |
441 | int rc, index; |
442 | ||
f52741c4 | 443 | if ((min_common_depth < 0) || !numa_enabled) |
ea9f5b70 AK |
444 | return default_nid; |
445 | ||
b88fc309 NF |
446 | rc = of_get_assoc_arrays(&aa); |
447 | if (rc) | |
448 | return default_nid; | |
8342681d | 449 | |
ea9f5b70 AK |
450 | if (min_common_depth <= aa.array_sz && |
451 | !(lmb->flags & DRCONF_MEM_AI_INVALID) && lmb->aa_index < aa.n_arrays) { | |
514a9cb3 | 452 | index = lmb->aa_index * aa.array_sz + min_common_depth - 1; |
b88fc309 | 453 | nid = of_read_number(&aa.arrays[index], 1); |
8342681d NF |
454 | |
455 | if (nid == 0xffff || nid >= MAX_NUMNODES) | |
456 | nid = default_nid; | |
1d805440 ND |
457 | |
458 | if (nid > 0) { | |
514a9cb3 | 459 | index = lmb->aa_index * aa.array_sz; |
1d805440 | 460 | initialize_distance_lookup_table(nid, |
b88fc309 | 461 | &aa.arrays[index]); |
1d805440 | 462 | } |
8342681d NF |
463 | } |
464 | ||
465 | return nid; | |
466 | } | |
467 | ||
1da177e4 LT |
468 | /* |
469 | * Figure out to which domain a cpu belongs and stick it there. | |
470 | * Return the id of the domain used. | |
471 | */ | |
061d19f2 | 472 | static int numa_setup_cpu(unsigned long lcpu) |
1da177e4 | 473 | { |
98fa15f3 | 474 | int nid = NUMA_NO_NODE; |
d4edc5b6 SB |
475 | struct device_node *cpu; |
476 | ||
477 | /* | |
478 | * If a valid cpu-to-node mapping is already available, use it | |
479 | * directly instead of querying the firmware, since it represents | |
480 | * the most recent mapping notified to us by the platform (eg: VPHN). | |
481 | */ | |
482 | if ((nid = numa_cpu_lookup_table[lcpu]) >= 0) { | |
483 | map_cpu_to_node(lcpu, nid); | |
484 | return nid; | |
485 | } | |
486 | ||
487 | cpu = of_get_cpu_node(lcpu, NULL); | |
1da177e4 LT |
488 | |
489 | if (!cpu) { | |
490 | WARN_ON(1); | |
297cf502 LZ |
491 | if (cpu_present(lcpu)) |
492 | goto out_present; | |
493 | else | |
494 | goto out; | |
1da177e4 LT |
495 | } |
496 | ||
953039c8 | 497 | nid = of_node_to_nid_single(cpu); |
1da177e4 | 498 | |
297cf502 | 499 | out_present: |
ea05ba7c | 500 | if (nid < 0 || !node_possible(nid)) |
72c33688 | 501 | nid = first_online_node; |
1da177e4 | 502 | |
297cf502 | 503 | map_cpu_to_node(lcpu, nid); |
1da177e4 | 504 | of_node_put(cpu); |
297cf502 | 505 | out: |
cf950b7a | 506 | return nid; |
1da177e4 LT |
507 | } |
508 | ||
68fb18aa SB |
509 | static void verify_cpu_node_mapping(int cpu, int node) |
510 | { | |
511 | int base, sibling, i; | |
512 | ||
513 | /* Verify that all the threads in the core belong to the same node */ | |
514 | base = cpu_first_thread_sibling(cpu); | |
515 | ||
516 | for (i = 0; i < threads_per_core; i++) { | |
517 | sibling = base + i; | |
518 | ||
519 | if (sibling == cpu || cpu_is_offline(sibling)) | |
520 | continue; | |
521 | ||
522 | if (cpu_to_node(sibling) != node) { | |
523 | WARN(1, "CPU thread siblings %d and %d don't belong" | |
524 | " to the same node!\n", cpu, sibling); | |
525 | break; | |
526 | } | |
527 | } | |
528 | } | |
529 | ||
bdab88e0 SAS |
530 | /* Must run before sched domains notifier. */ |
531 | static int ppc_numa_cpu_prepare(unsigned int cpu) | |
532 | { | |
533 | int nid; | |
534 | ||
535 | nid = numa_setup_cpu(cpu); | |
536 | verify_cpu_node_mapping(cpu, nid); | |
537 | return 0; | |
538 | } | |
539 | ||
540 | static int ppc_numa_cpu_dead(unsigned int cpu) | |
541 | { | |
1da177e4 | 542 | #ifdef CONFIG_HOTPLUG_CPU |
bdab88e0 | 543 | unmap_cpu_from_node(cpu); |
1da177e4 | 544 | #endif |
bdab88e0 | 545 | return 0; |
1da177e4 LT |
546 | } |
547 | ||
548 | /* | |
549 | * Check and possibly modify a memory region to enforce the memory limit. | |
550 | * | |
551 | * Returns the size the region should have to enforce the memory limit. | |
552 | * This will either be the original value of size, a truncated value, | |
553 | * or zero. If the returned value of size is 0 the region should be | |
25985edc | 554 | * discarded as it lies wholly above the memory limit. |
1da177e4 | 555 | */ |
45fb6cea AB |
556 | static unsigned long __init numa_enforce_memory_limit(unsigned long start, |
557 | unsigned long size) | |
1da177e4 LT |
558 | { |
559 | /* | |
95f72d1e | 560 | * We use memblock_end_of_DRAM() in here instead of memory_limit because |
1da177e4 | 561 | * we've already adjusted it for the limit and it takes care of |
fe55249d MM |
562 | * having memory holes below the limit. Also, in the case of |
563 | * iommu_is_off, memory_limit is not set but is implicitly enforced. | |
1da177e4 | 564 | */ |
1da177e4 | 565 | |
95f72d1e | 566 | if (start + size <= memblock_end_of_DRAM()) |
1da177e4 LT |
567 | return size; |
568 | ||
95f72d1e | 569 | if (start >= memblock_end_of_DRAM()) |
1da177e4 LT |
570 | return 0; |
571 | ||
95f72d1e | 572 | return memblock_end_of_DRAM() - start; |
1da177e4 LT |
573 | } |
574 | ||
cf00085d C |
575 | /* |
576 | * Reads the counter for a given entry in | |
577 | * linux,drconf-usable-memory property | |
578 | */ | |
b08a2a12 | 579 | static inline int __init read_usm_ranges(const __be32 **usm) |
cf00085d C |
580 | { |
581 | /* | |
3fdfd990 | 582 | * For each lmb in ibm,dynamic-memory a corresponding |
cf00085d C |
583 | * entry in linux,drconf-usable-memory property contains |
584 | * a counter followed by that many (base, size) duple. | |
585 | * read the counter from linux,drconf-usable-memory | |
586 | */ | |
587 | return read_n_cells(n_mem_size_cells, usm); | |
588 | } | |
589 | ||
0204568a PM |
590 | /* |
591 | * Extract NUMA information from the ibm,dynamic-reconfiguration-memory | |
592 | * node. This assumes n_mem_{addr,size}_cells have been set. | |
593 | */ | |
514a9cb3 NF |
594 | static void __init numa_setup_drmem_lmb(struct drmem_lmb *lmb, |
595 | const __be32 **usm) | |
0204568a | 596 | { |
514a9cb3 NF |
597 | unsigned int ranges, is_kexec_kdump = 0; |
598 | unsigned long base, size, sz; | |
8342681d | 599 | int nid; |
8342681d | 600 | |
514a9cb3 NF |
601 | /* |
602 | * Skip this block if the reserved bit is set in flags (0x80) | |
603 | * or if the block is not assigned to this partition (0x8) | |
604 | */ | |
605 | if ((lmb->flags & DRCONF_MEM_RESERVED) | |
606 | || !(lmb->flags & DRCONF_MEM_ASSIGNED)) | |
8342681d NF |
607 | return; |
608 | ||
514a9cb3 | 609 | if (*usm) |
cf00085d C |
610 | is_kexec_kdump = 1; |
611 | ||
514a9cb3 NF |
612 | base = lmb->base_addr; |
613 | size = drmem_lmb_size(); | |
614 | ranges = 1; | |
8342681d | 615 | |
514a9cb3 NF |
616 | if (is_kexec_kdump) { |
617 | ranges = read_usm_ranges(usm); | |
618 | if (!ranges) /* there are no (base, size) duple */ | |
619 | return; | |
620 | } | |
8342681d | 621 | |
514a9cb3 | 622 | do { |
cf00085d | 623 | if (is_kexec_kdump) { |
514a9cb3 NF |
624 | base = read_n_cells(n_mem_addr_cells, usm); |
625 | size = read_n_cells(n_mem_size_cells, usm); | |
cf00085d | 626 | } |
514a9cb3 NF |
627 | |
628 | nid = of_drconf_to_nid_single(lmb); | |
629 | fake_numa_create_new_node(((base + size) >> PAGE_SHIFT), | |
630 | &nid); | |
631 | node_set_online(nid); | |
632 | sz = numa_enforce_memory_limit(base, size); | |
633 | if (sz) | |
634 | memblock_set_node(base, sz, &memblock.memory, nid); | |
635 | } while (--ranges); | |
0204568a PM |
636 | } |
637 | ||
1da177e4 LT |
638 | static int __init parse_numa_properties(void) |
639 | { | |
94db7c5e | 640 | struct device_node *memory; |
482ec7c4 | 641 | int default_nid = 0; |
1da177e4 LT |
642 | unsigned long i; |
643 | ||
644 | if (numa_enabled == 0) { | |
645 | printk(KERN_WARNING "NUMA disabled by user\n"); | |
646 | return -1; | |
647 | } | |
648 | ||
1da177e4 LT |
649 | min_common_depth = find_min_common_depth(); |
650 | ||
495c2ff4 AK |
651 | if (min_common_depth < 0) { |
652 | /* | |
653 | * if we fail to parse min_common_depth from device tree | |
654 | * mark the numa disabled, boot with numa disabled. | |
655 | */ | |
656 | numa_enabled = false; | |
1da177e4 | 657 | return min_common_depth; |
495c2ff4 | 658 | } |
1da177e4 | 659 | |
bf4b85b0 NL |
660 | dbg("NUMA associativity depth for CPU/Memory: %d\n", min_common_depth); |
661 | ||
1da177e4 | 662 | /* |
482ec7c4 NL |
663 | * Even though we connect cpus to numa domains later in SMP |
664 | * init, we need to know the node ids now. This is because | |
665 | * each node to be onlined must have NODE_DATA etc backing it. | |
1da177e4 | 666 | */ |
482ec7c4 | 667 | for_each_present_cpu(i) { |
dfbe93a2 | 668 | struct device_node *cpu; |
cf950b7a | 669 | int nid; |
1da177e4 | 670 | |
8b16cd23 | 671 | cpu = of_get_cpu_node(i, NULL); |
482ec7c4 | 672 | BUG_ON(!cpu); |
953039c8 | 673 | nid = of_node_to_nid_single(cpu); |
482ec7c4 | 674 | of_node_put(cpu); |
1da177e4 | 675 | |
482ec7c4 NL |
676 | /* |
677 | * Don't fall back to default_nid yet -- we will plug | |
678 | * cpus into nodes once the memory scan has discovered | |
679 | * the topology. | |
680 | */ | |
681 | if (nid < 0) | |
682 | continue; | |
683 | node_set_online(nid); | |
1da177e4 LT |
684 | } |
685 | ||
237a0989 | 686 | get_n_mem_cells(&n_mem_addr_cells, &n_mem_size_cells); |
94db7c5e AB |
687 | |
688 | for_each_node_by_type(memory, "memory") { | |
1da177e4 LT |
689 | unsigned long start; |
690 | unsigned long size; | |
cf950b7a | 691 | int nid; |
1da177e4 | 692 | int ranges; |
b08a2a12 | 693 | const __be32 *memcell_buf; |
1da177e4 LT |
694 | unsigned int len; |
695 | ||
e2eb6392 | 696 | memcell_buf = of_get_property(memory, |
ba759485 ME |
697 | "linux,usable-memory", &len); |
698 | if (!memcell_buf || len <= 0) | |
e2eb6392 | 699 | memcell_buf = of_get_property(memory, "reg", &len); |
1da177e4 LT |
700 | if (!memcell_buf || len <= 0) |
701 | continue; | |
702 | ||
cc5d0189 BH |
703 | /* ranges in cell */ |
704 | ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells); | |
1da177e4 LT |
705 | new_range: |
706 | /* these are order-sensitive, and modify the buffer pointer */ | |
237a0989 MK |
707 | start = read_n_cells(n_mem_addr_cells, &memcell_buf); |
708 | size = read_n_cells(n_mem_size_cells, &memcell_buf); | |
1da177e4 | 709 | |
482ec7c4 NL |
710 | /* |
711 | * Assumption: either all memory nodes or none will | |
712 | * have associativity properties. If none, then | |
713 | * everything goes to default_nid. | |
714 | */ | |
953039c8 | 715 | nid = of_node_to_nid_single(memory); |
482ec7c4 NL |
716 | if (nid < 0) |
717 | nid = default_nid; | |
1daa6d08 BS |
718 | |
719 | fake_numa_create_new_node(((start + size) >> PAGE_SHIFT), &nid); | |
482ec7c4 | 720 | node_set_online(nid); |
1da177e4 | 721 | |
7656cd8e RA |
722 | size = numa_enforce_memory_limit(start, size); |
723 | if (size) | |
724 | memblock_set_node(start, size, &memblock.memory, nid); | |
1da177e4 LT |
725 | |
726 | if (--ranges) | |
727 | goto new_range; | |
728 | } | |
729 | ||
0204568a | 730 | /* |
dfbe93a2 AB |
731 | * Now do the same thing for each MEMBLOCK listed in the |
732 | * ibm,dynamic-memory property in the | |
733 | * ibm,dynamic-reconfiguration-memory node. | |
0204568a PM |
734 | */ |
735 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); | |
514a9cb3 NF |
736 | if (memory) { |
737 | walk_drmem_lmbs(memory, numa_setup_drmem_lmb); | |
738 | of_node_put(memory); | |
739 | } | |
0204568a | 740 | |
1da177e4 LT |
741 | return 0; |
742 | } | |
743 | ||
744 | static void __init setup_nonnuma(void) | |
745 | { | |
95f72d1e YL |
746 | unsigned long top_of_ram = memblock_end_of_DRAM(); |
747 | unsigned long total_ram = memblock_phys_mem_size(); | |
c67c3cb4 | 748 | unsigned long start_pfn, end_pfn; |
28be7072 BH |
749 | unsigned int nid = 0; |
750 | struct memblock_region *reg; | |
1da177e4 | 751 | |
e110b281 | 752 | printk(KERN_DEBUG "Top of RAM: 0x%lx, Total RAM: 0x%lx\n", |
1da177e4 | 753 | top_of_ram, total_ram); |
e110b281 | 754 | printk(KERN_DEBUG "Memory hole size: %ldMB\n", |
1da177e4 LT |
755 | (top_of_ram - total_ram) >> 20); |
756 | ||
28be7072 | 757 | for_each_memblock(memory, reg) { |
c7fc2de0 YL |
758 | start_pfn = memblock_region_memory_base_pfn(reg); |
759 | end_pfn = memblock_region_memory_end_pfn(reg); | |
1daa6d08 BS |
760 | |
761 | fake_numa_create_new_node(end_pfn, &nid); | |
1d7cfe18 | 762 | memblock_set_node(PFN_PHYS(start_pfn), |
e7e8de59 TC |
763 | PFN_PHYS(end_pfn - start_pfn), |
764 | &memblock.memory, nid); | |
1daa6d08 | 765 | node_set_online(nid); |
c67c3cb4 | 766 | } |
1da177e4 LT |
767 | } |
768 | ||
4b703a23 AB |
769 | void __init dump_numa_cpu_topology(void) |
770 | { | |
771 | unsigned int node; | |
772 | unsigned int cpu, count; | |
773 | ||
495c2ff4 | 774 | if (!numa_enabled) |
4b703a23 AB |
775 | return; |
776 | ||
777 | for_each_online_node(node) { | |
8467801c | 778 | pr_info("Node %d CPUs:", node); |
4b703a23 AB |
779 | |
780 | count = 0; | |
781 | /* | |
782 | * If we used a CPU iterator here we would miss printing | |
783 | * the holes in the cpumap. | |
784 | */ | |
25863de0 AB |
785 | for (cpu = 0; cpu < nr_cpu_ids; cpu++) { |
786 | if (cpumask_test_cpu(cpu, | |
787 | node_to_cpumask_map[node])) { | |
4b703a23 | 788 | if (count == 0) |
8467801c | 789 | pr_cont(" %u", cpu); |
4b703a23 AB |
790 | ++count; |
791 | } else { | |
792 | if (count > 1) | |
8467801c | 793 | pr_cont("-%u", cpu - 1); |
4b703a23 AB |
794 | count = 0; |
795 | } | |
796 | } | |
797 | ||
798 | if (count > 1) | |
8467801c AK |
799 | pr_cont("-%u", nr_cpu_ids - 1); |
800 | pr_cont("\n"); | |
4b703a23 AB |
801 | } |
802 | } | |
803 | ||
10239733 AB |
804 | /* Initialize NODE_DATA for a node on the local memory */ |
805 | static void __init setup_node_data(int nid, u64 start_pfn, u64 end_pfn) | |
4a618669 | 806 | { |
10239733 AB |
807 | u64 spanned_pages = end_pfn - start_pfn; |
808 | const size_t nd_size = roundup(sizeof(pg_data_t), SMP_CACHE_BYTES); | |
809 | u64 nd_pa; | |
810 | void *nd; | |
811 | int tnid; | |
4a618669 | 812 | |
9a8dd708 | 813 | nd_pa = memblock_phys_alloc_try_nid(nd_size, SMP_CACHE_BYTES, nid); |
33755574 MR |
814 | if (!nd_pa) |
815 | panic("Cannot allocate %zu bytes for node %d data\n", | |
816 | nd_size, nid); | |
817 | ||
10239733 | 818 | nd = __va(nd_pa); |
4a618669 | 819 | |
10239733 AB |
820 | /* report and initialize */ |
821 | pr_info(" NODE_DATA [mem %#010Lx-%#010Lx]\n", | |
822 | nd_pa, nd_pa + nd_size - 1); | |
823 | tnid = early_pfn_to_nid(nd_pa >> PAGE_SHIFT); | |
824 | if (tnid != nid) | |
825 | pr_info(" NODE_DATA(%d) on node %d\n", nid, tnid); | |
4a618669 | 826 | |
10239733 AB |
827 | node_data[nid] = nd; |
828 | memset(NODE_DATA(nid), 0, sizeof(pg_data_t)); | |
829 | NODE_DATA(nid)->node_id = nid; | |
830 | NODE_DATA(nid)->node_start_pfn = start_pfn; | |
831 | NODE_DATA(nid)->node_spanned_pages = spanned_pages; | |
832 | } | |
4a618669 | 833 | |
a346137e MB |
834 | static void __init find_possible_nodes(void) |
835 | { | |
836 | struct device_node *rtas; | |
837 | u32 numnodes, i; | |
838 | ||
495c2ff4 | 839 | if (!numa_enabled) |
a346137e MB |
840 | return; |
841 | ||
842 | rtas = of_find_node_by_path("/rtas"); | |
843 | if (!rtas) | |
844 | return; | |
845 | ||
846 | if (of_property_read_u32_index(rtas, | |
847 | "ibm,max-associativity-domains", | |
848 | min_common_depth, &numnodes)) | |
849 | goto out; | |
850 | ||
851 | for (i = 0; i < numnodes; i++) { | |
ea05ba7c | 852 | if (!node_possible(i)) |
a346137e | 853 | node_set(i, node_possible_map); |
a346137e MB |
854 | } |
855 | ||
856 | out: | |
857 | of_node_put(rtas); | |
858 | } | |
859 | ||
9bd9be00 | 860 | void __init mem_topology_setup(void) |
1da177e4 | 861 | { |
9bd9be00 | 862 | int cpu; |
1da177e4 LT |
863 | |
864 | if (parse_numa_properties()) | |
865 | setup_nonnuma(); | |
1da177e4 | 866 | |
3af229f2 | 867 | /* |
a346137e MB |
868 | * Modify the set of possible NUMA nodes to reflect information |
869 | * available about the set of online nodes, and the set of nodes | |
870 | * that we expect to make use of for this platform's affinity | |
871 | * calculations. | |
3af229f2 NA |
872 | */ |
873 | nodes_and(node_possible_map, node_possible_map, node_online_map); | |
874 | ||
a346137e MB |
875 | find_possible_nodes(); |
876 | ||
9bd9be00 NP |
877 | setup_node_to_cpumask_map(); |
878 | ||
879 | reset_numa_cpu_lookup_table(); | |
880 | ||
881 | for_each_present_cpu(cpu) | |
882 | numa_setup_cpu(cpu); | |
883 | } | |
884 | ||
885 | void __init initmem_init(void) | |
886 | { | |
887 | int nid; | |
888 | ||
889 | max_low_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT; | |
890 | max_pfn = max_low_pfn; | |
891 | ||
892 | memblock_dump_all(); | |
893 | ||
1da177e4 | 894 | for_each_online_node(nid) { |
c67c3cb4 | 895 | unsigned long start_pfn, end_pfn; |
1da177e4 | 896 | |
c67c3cb4 | 897 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); |
10239733 | 898 | setup_node_data(nid, start_pfn, end_pfn); |
8f64e1f2 | 899 | sparse_memory_present_with_active_regions(nid); |
4a618669 | 900 | } |
d3f6204a | 901 | |
21098b9e | 902 | sparse_init(); |
25863de0 | 903 | |
2fabf084 NA |
904 | /* |
905 | * We need the numa_cpu_lookup_table to be accurate for all CPUs, | |
906 | * even before we online them, so that we can use cpu_to_{node,mem} | |
907 | * early in boot, cf. smp_prepare_cpus(). | |
bdab88e0 SAS |
908 | * _nocalls() + manual invocation is used because cpuhp is not yet |
909 | * initialized for the boot CPU. | |
2fabf084 | 910 | */ |
73c1b41e | 911 | cpuhp_setup_state_nocalls(CPUHP_POWER_NUMA_PREPARE, "powerpc/numa:prepare", |
bdab88e0 | 912 | ppc_numa_cpu_prepare, ppc_numa_cpu_dead); |
1da177e4 LT |
913 | } |
914 | ||
1da177e4 LT |
915 | static int __init early_numa(char *p) |
916 | { | |
917 | if (!p) | |
918 | return 0; | |
919 | ||
920 | if (strstr(p, "off")) | |
921 | numa_enabled = 0; | |
922 | ||
923 | if (strstr(p, "debug")) | |
924 | numa_debug = 1; | |
925 | ||
1daa6d08 BS |
926 | p = strstr(p, "fake="); |
927 | if (p) | |
928 | cmdline = p + strlen("fake="); | |
929 | ||
1da177e4 LT |
930 | return 0; |
931 | } | |
932 | early_param("numa", early_numa); | |
237a0989 | 933 | |
558f8649 NL |
934 | /* |
935 | * The platform can inform us through one of several mechanisms | |
936 | * (post-migration device tree updates, PRRN or VPHN) that the NUMA | |
937 | * assignment of a resource has changed. This controls whether we act | |
938 | * on that. Disabled by default. | |
939 | */ | |
940 | static bool topology_updates_enabled; | |
2d73bae1 NA |
941 | |
942 | static int __init early_topology_updates(char *p) | |
943 | { | |
944 | if (!p) | |
945 | return 0; | |
946 | ||
558f8649 NL |
947 | if (!strcmp(p, "on")) { |
948 | pr_warn("Caution: enabling topology updates\n"); | |
949 | topology_updates_enabled = true; | |
2d73bae1 NA |
950 | } |
951 | ||
952 | return 0; | |
953 | } | |
954 | early_param("topology_updates", early_topology_updates); | |
955 | ||
237a0989 | 956 | #ifdef CONFIG_MEMORY_HOTPLUG |
0db9360a | 957 | /* |
0f16ef7f NF |
958 | * Find the node associated with a hot added memory section for |
959 | * memory represented in the device tree by the property | |
960 | * ibm,dynamic-reconfiguration-memory/ibm,dynamic-memory. | |
0db9360a | 961 | */ |
514a9cb3 | 962 | static int hot_add_drconf_scn_to_nid(unsigned long scn_addr) |
0db9360a | 963 | { |
514a9cb3 | 964 | struct drmem_lmb *lmb; |
3fdfd990 | 965 | unsigned long lmb_size; |
98fa15f3 | 966 | int nid = NUMA_NO_NODE; |
0db9360a | 967 | |
514a9cb3 | 968 | lmb_size = drmem_lmb_size(); |
0db9360a | 969 | |
514a9cb3 | 970 | for_each_drmem_lmb(lmb) { |
0db9360a NF |
971 | /* skip this block if it is reserved or not assigned to |
972 | * this partition */ | |
514a9cb3 NF |
973 | if ((lmb->flags & DRCONF_MEM_RESERVED) |
974 | || !(lmb->flags & DRCONF_MEM_ASSIGNED)) | |
0db9360a NF |
975 | continue; |
976 | ||
514a9cb3 NF |
977 | if ((scn_addr < lmb->base_addr) |
978 | || (scn_addr >= (lmb->base_addr + lmb_size))) | |
0f16ef7f NF |
979 | continue; |
980 | ||
514a9cb3 | 981 | nid = of_drconf_to_nid_single(lmb); |
0f16ef7f NF |
982 | break; |
983 | } | |
984 | ||
985 | return nid; | |
986 | } | |
987 | ||
988 | /* | |
989 | * Find the node associated with a hot added memory section for memory | |
990 | * represented in the device tree as a node (i.e. memory@XXXX) for | |
95f72d1e | 991 | * each memblock. |
0f16ef7f | 992 | */ |
ec32dd66 | 993 | static int hot_add_node_scn_to_nid(unsigned long scn_addr) |
0f16ef7f | 994 | { |
94db7c5e | 995 | struct device_node *memory; |
98fa15f3 | 996 | int nid = NUMA_NO_NODE; |
0f16ef7f | 997 | |
94db7c5e | 998 | for_each_node_by_type(memory, "memory") { |
0f16ef7f NF |
999 | unsigned long start, size; |
1000 | int ranges; | |
b08a2a12 | 1001 | const __be32 *memcell_buf; |
0f16ef7f NF |
1002 | unsigned int len; |
1003 | ||
1004 | memcell_buf = of_get_property(memory, "reg", &len); | |
1005 | if (!memcell_buf || len <= 0) | |
1006 | continue; | |
1007 | ||
1008 | /* ranges in cell */ | |
1009 | ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells); | |
1010 | ||
1011 | while (ranges--) { | |
1012 | start = read_n_cells(n_mem_addr_cells, &memcell_buf); | |
1013 | size = read_n_cells(n_mem_size_cells, &memcell_buf); | |
1014 | ||
1015 | if ((scn_addr < start) || (scn_addr >= (start + size))) | |
1016 | continue; | |
1017 | ||
1018 | nid = of_node_to_nid_single(memory); | |
1019 | break; | |
1020 | } | |
0db9360a | 1021 | |
0f16ef7f NF |
1022 | if (nid >= 0) |
1023 | break; | |
0db9360a NF |
1024 | } |
1025 | ||
60831842 AB |
1026 | of_node_put(memory); |
1027 | ||
0f16ef7f | 1028 | return nid; |
0db9360a NF |
1029 | } |
1030 | ||
237a0989 MK |
1031 | /* |
1032 | * Find the node associated with a hot added memory section. Section | |
95f72d1e YL |
1033 | * corresponds to a SPARSEMEM section, not an MEMBLOCK. It is assumed that |
1034 | * sections are fully contained within a single MEMBLOCK. | |
237a0989 MK |
1035 | */ |
1036 | int hot_add_scn_to_nid(unsigned long scn_addr) | |
1037 | { | |
1038 | struct device_node *memory = NULL; | |
4a3bac4e | 1039 | int nid; |
237a0989 | 1040 | |
495c2ff4 | 1041 | if (!numa_enabled) |
72c33688 | 1042 | return first_online_node; |
0db9360a NF |
1043 | |
1044 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); | |
1045 | if (memory) { | |
514a9cb3 | 1046 | nid = hot_add_drconf_scn_to_nid(scn_addr); |
0db9360a | 1047 | of_node_put(memory); |
0f16ef7f NF |
1048 | } else { |
1049 | nid = hot_add_node_scn_to_nid(scn_addr); | |
0db9360a | 1050 | } |
237a0989 | 1051 | |
2a8628d4 | 1052 | if (nid < 0 || !node_possible(nid)) |
72c33688 | 1053 | nid = first_online_node; |
237a0989 | 1054 | |
0f16ef7f | 1055 | return nid; |
237a0989 | 1056 | } |
0f16ef7f | 1057 | |
cd34206e NA |
1058 | static u64 hot_add_drconf_memory_max(void) |
1059 | { | |
e70bd3ae | 1060 | struct device_node *memory = NULL; |
45b64ee6 | 1061 | struct device_node *dn = NULL; |
45b64ee6 | 1062 | const __be64 *lrdr = NULL; |
45b64ee6 BR |
1063 | |
1064 | dn = of_find_node_by_path("/rtas"); | |
1065 | if (dn) { | |
1066 | lrdr = of_get_property(dn, "ibm,lrdr-capacity", NULL); | |
1067 | of_node_put(dn); | |
1068 | if (lrdr) | |
1069 | return be64_to_cpup(lrdr); | |
1070 | } | |
cd34206e | 1071 | |
e70bd3ae BR |
1072 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); |
1073 | if (memory) { | |
e70bd3ae | 1074 | of_node_put(memory); |
514a9cb3 | 1075 | return drmem_lmb_memory_max(); |
e70bd3ae | 1076 | } |
45b64ee6 | 1077 | return 0; |
cd34206e NA |
1078 | } |
1079 | ||
1080 | /* | |
1081 | * memory_hotplug_max - return max address of memory that may be added | |
1082 | * | |
1083 | * This is currently only used on systems that support drconfig memory | |
1084 | * hotplug. | |
1085 | */ | |
1086 | u64 memory_hotplug_max(void) | |
1087 | { | |
1088 | return max(hot_add_drconf_memory_max(), memblock_end_of_DRAM()); | |
1089 | } | |
237a0989 | 1090 | #endif /* CONFIG_MEMORY_HOTPLUG */ |
9eff1a38 | 1091 | |
bd03403a | 1092 | /* Virtual Processor Home Node (VPHN) support */ |
39bf990e | 1093 | #ifdef CONFIG_PPC_SPLPAR |
30c05350 NF |
1094 | struct topology_update_data { |
1095 | struct topology_update_data *next; | |
1096 | unsigned int cpu; | |
1097 | int old_nid; | |
1098 | int new_nid; | |
1099 | }; | |
1100 | ||
cee5405d MB |
1101 | #define TOPOLOGY_DEF_TIMER_SECS 60 |
1102 | ||
5de16699 | 1103 | static u8 vphn_cpu_change_counts[NR_CPUS][MAX_DISTANCE_REF_POINTS]; |
9eff1a38 JL |
1104 | static cpumask_t cpu_associativity_changes_mask; |
1105 | static int vphn_enabled; | |
5d88aa85 JL |
1106 | static int prrn_enabled; |
1107 | static void reset_topology_timer(void); | |
cee5405d | 1108 | static int topology_timer_secs = 1; |
17f444c0 | 1109 | static int topology_inited; |
9eff1a38 | 1110 | |
cee5405d MB |
1111 | /* |
1112 | * Change polling interval for associativity changes. | |
1113 | */ | |
1114 | int timed_topology_update(int nsecs) | |
1115 | { | |
1116 | if (vphn_enabled) { | |
1117 | if (nsecs > 0) | |
1118 | topology_timer_secs = nsecs; | |
1119 | else | |
1120 | topology_timer_secs = TOPOLOGY_DEF_TIMER_SECS; | |
1121 | ||
1122 | reset_topology_timer(); | |
1123 | } | |
1124 | ||
1125 | return 0; | |
1126 | } | |
9eff1a38 JL |
1127 | |
1128 | /* | |
1129 | * Store the current values of the associativity change counters in the | |
1130 | * hypervisor. | |
1131 | */ | |
1132 | static void setup_cpu_associativity_change_counters(void) | |
1133 | { | |
cd9d6cc7 | 1134 | int cpu; |
9eff1a38 | 1135 | |
5de16699 AB |
1136 | /* The VPHN feature supports a maximum of 8 reference points */ |
1137 | BUILD_BUG_ON(MAX_DISTANCE_REF_POINTS > 8); | |
1138 | ||
9eff1a38 | 1139 | for_each_possible_cpu(cpu) { |
cd9d6cc7 | 1140 | int i; |
9eff1a38 | 1141 | u8 *counts = vphn_cpu_change_counts[cpu]; |
499dcd41 | 1142 | volatile u8 *hypervisor_counts = lppaca_of(cpu).vphn_assoc_counts; |
9eff1a38 | 1143 | |
5de16699 | 1144 | for (i = 0; i < distance_ref_points_depth; i++) |
9eff1a38 | 1145 | counts[i] = hypervisor_counts[i]; |
9eff1a38 JL |
1146 | } |
1147 | } | |
1148 | ||
1149 | /* | |
1150 | * The hypervisor maintains a set of 8 associativity change counters in | |
1151 | * the VPA of each cpu that correspond to the associativity levels in the | |
1152 | * ibm,associativity-reference-points property. When an associativity | |
1153 | * level changes, the corresponding counter is incremented. | |
1154 | * | |
1155 | * Set a bit in cpu_associativity_changes_mask for each cpu whose home | |
1156 | * node associativity levels have changed. | |
1157 | * | |
1158 | * Returns the number of cpus with unhandled associativity changes. | |
1159 | */ | |
1160 | static int update_cpu_associativity_changes_mask(void) | |
1161 | { | |
5d88aa85 | 1162 | int cpu; |
9eff1a38 JL |
1163 | cpumask_t *changes = &cpu_associativity_changes_mask; |
1164 | ||
9eff1a38 JL |
1165 | for_each_possible_cpu(cpu) { |
1166 | int i, changed = 0; | |
1167 | u8 *counts = vphn_cpu_change_counts[cpu]; | |
499dcd41 | 1168 | volatile u8 *hypervisor_counts = lppaca_of(cpu).vphn_assoc_counts; |
9eff1a38 | 1169 | |
5de16699 | 1170 | for (i = 0; i < distance_ref_points_depth; i++) { |
d69043e8 | 1171 | if (hypervisor_counts[i] != counts[i]) { |
9eff1a38 JL |
1172 | counts[i] = hypervisor_counts[i]; |
1173 | changed = 1; | |
1174 | } | |
1175 | } | |
1176 | if (changed) { | |
3be7db6a RJ |
1177 | cpumask_or(changes, changes, cpu_sibling_mask(cpu)); |
1178 | cpu = cpu_last_thread_sibling(cpu); | |
9eff1a38 JL |
1179 | } |
1180 | } | |
1181 | ||
5d88aa85 | 1182 | return cpumask_weight(changes); |
9eff1a38 JL |
1183 | } |
1184 | ||
9eff1a38 JL |
1185 | /* |
1186 | * Retrieve the new associativity information for a virtual processor's | |
1187 | * home node. | |
1188 | */ | |
9eff1a38 | 1189 | static long vphn_get_associativity(unsigned long cpu, |
b08a2a12 | 1190 | __be32 *associativity) |
9eff1a38 | 1191 | { |
cd9d6cc7 | 1192 | long rc; |
9eff1a38 | 1193 | |
ef34e0ef NR |
1194 | rc = hcall_vphn(get_hard_smp_processor_id(cpu), |
1195 | VPHN_FLAG_VCPU, associativity); | |
9eff1a38 JL |
1196 | |
1197 | switch (rc) { | |
1198 | case H_FUNCTION: | |
437ccdc8 | 1199 | printk_once(KERN_INFO |
9eff1a38 JL |
1200 | "VPHN is not supported. Disabling polling...\n"); |
1201 | stop_topology_update(); | |
1202 | break; | |
1203 | case H_HARDWARE: | |
1204 | printk(KERN_ERR | |
1205 | "hcall_vphn() experienced a hardware fault " | |
1206 | "preventing VPHN. Disabling polling...\n"); | |
1207 | stop_topology_update(); | |
17f444c0 MB |
1208 | break; |
1209 | case H_SUCCESS: | |
1210 | dbg("VPHN hcall succeeded. Reset polling...\n"); | |
cee5405d | 1211 | timed_topology_update(0); |
17f444c0 | 1212 | break; |
9eff1a38 JL |
1213 | } |
1214 | ||
1215 | return rc; | |
1216 | } | |
1217 | ||
e67e02a5 | 1218 | int find_and_online_cpu_nid(int cpu) |
ea05ba7c MB |
1219 | { |
1220 | __be32 associativity[VPHN_ASSOC_BUFSIZE] = {0}; | |
1221 | int new_nid; | |
1222 | ||
1223 | /* Use associativity from first thread for all siblings */ | |
2483ef05 SD |
1224 | if (vphn_get_associativity(cpu, associativity)) |
1225 | return cpu_to_node(cpu); | |
1226 | ||
ea05ba7c MB |
1227 | new_nid = associativity_to_nid(associativity); |
1228 | if (new_nid < 0 || !node_possible(new_nid)) | |
1229 | new_nid = first_online_node; | |
1230 | ||
1231 | if (NODE_DATA(new_nid) == NULL) { | |
1232 | #ifdef CONFIG_MEMORY_HOTPLUG | |
1233 | /* | |
1234 | * Need to ensure that NODE_DATA is initialized for a node from | |
1235 | * available memory (see memblock_alloc_try_nid). If unable to | |
1236 | * init the node, then default to nearest node that has memory | |
ac1788cc SD |
1237 | * installed. Skip onlining a node if the subsystems are not |
1238 | * yet initialized. | |
ea05ba7c | 1239 | */ |
ac1788cc | 1240 | if (!topology_inited || try_online_node(new_nid)) |
ea05ba7c MB |
1241 | new_nid = first_online_node; |
1242 | #else | |
1243 | /* | |
1244 | * Default to using the nearest node that has memory installed. | |
1245 | * Otherwise, it would be necessary to patch the kernel MM code | |
1246 | * to deal with more memoryless-node error conditions. | |
1247 | */ | |
1248 | new_nid = first_online_node; | |
1249 | #endif | |
1250 | } | |
1251 | ||
e67e02a5 MB |
1252 | pr_debug("%s:%d cpu %d nid %d\n", __FUNCTION__, __LINE__, |
1253 | cpu, new_nid); | |
ea05ba7c MB |
1254 | return new_nid; |
1255 | } | |
1256 | ||
30c05350 NF |
1257 | /* |
1258 | * Update the CPU maps and sysfs entries for a single CPU when its NUMA | |
1259 | * characteristics change. This function doesn't perform any locking and is | |
1260 | * only safe to call from stop_machine(). | |
1261 | */ | |
1262 | static int update_cpu_topology(void *data) | |
1263 | { | |
1264 | struct topology_update_data *update; | |
1265 | unsigned long cpu; | |
1266 | ||
1267 | if (!data) | |
1268 | return -EINVAL; | |
1269 | ||
3be7db6a | 1270 | cpu = smp_processor_id(); |
30c05350 NF |
1271 | |
1272 | for (update = data; update; update = update->next) { | |
2c0a33f9 | 1273 | int new_nid = update->new_nid; |
30c05350 NF |
1274 | if (cpu != update->cpu) |
1275 | continue; | |
1276 | ||
49f8d8c0 | 1277 | unmap_cpu_from_node(cpu); |
2c0a33f9 NA |
1278 | map_cpu_to_node(cpu, new_nid); |
1279 | set_cpu_numa_node(cpu, new_nid); | |
1280 | set_cpu_numa_mem(cpu, local_memory_node(new_nid)); | |
176bbf14 | 1281 | vdso_getcpu_init(); |
30c05350 NF |
1282 | } |
1283 | ||
1284 | return 0; | |
1285 | } | |
1286 | ||
d4edc5b6 SB |
1287 | static int update_lookup_table(void *data) |
1288 | { | |
1289 | struct topology_update_data *update; | |
1290 | ||
1291 | if (!data) | |
1292 | return -EINVAL; | |
1293 | ||
1294 | /* | |
1295 | * Upon topology update, the numa-cpu lookup table needs to be updated | |
1296 | * for all threads in the core, including offline CPUs, to ensure that | |
1297 | * future hotplug operations respect the cpu-to-node associativity | |
1298 | * properly. | |
1299 | */ | |
1300 | for (update = data; update; update = update->next) { | |
1301 | int nid, base, j; | |
1302 | ||
1303 | nid = update->new_nid; | |
1304 | base = cpu_first_thread_sibling(update->cpu); | |
1305 | ||
1306 | for (j = 0; j < threads_per_core; j++) { | |
1307 | update_numa_cpu_lookup_table(base + j, nid); | |
1308 | } | |
1309 | } | |
1310 | ||
1311 | return 0; | |
1312 | } | |
1313 | ||
9eff1a38 JL |
1314 | /* |
1315 | * Update the node maps and sysfs entries for each cpu whose home node | |
79c5fceb | 1316 | * has changed. Returns 1 when the topology has changed, and 0 otherwise. |
3e401f7a TJB |
1317 | * |
1318 | * cpus_locked says whether we already hold cpu_hotplug_lock. | |
9eff1a38 | 1319 | */ |
3e401f7a | 1320 | int numa_update_cpu_topology(bool cpus_locked) |
9eff1a38 | 1321 | { |
3be7db6a | 1322 | unsigned int cpu, sibling, changed = 0; |
30c05350 | 1323 | struct topology_update_data *updates, *ud; |
176bbf14 | 1324 | cpumask_t updated_cpus; |
8a25a2fd | 1325 | struct device *dev; |
3be7db6a | 1326 | int weight, new_nid, i = 0; |
9eff1a38 | 1327 | |
2ea62630 | 1328 | if (!prrn_enabled && !vphn_enabled && topology_inited) |
2d73bae1 NA |
1329 | return 0; |
1330 | ||
30c05350 NF |
1331 | weight = cpumask_weight(&cpu_associativity_changes_mask); |
1332 | if (!weight) | |
1333 | return 0; | |
1334 | ||
6396bb22 | 1335 | updates = kcalloc(weight, sizeof(*updates), GFP_KERNEL); |
30c05350 NF |
1336 | if (!updates) |
1337 | return 0; | |
9eff1a38 | 1338 | |
176bbf14 JL |
1339 | cpumask_clear(&updated_cpus); |
1340 | ||
5d88aa85 | 1341 | for_each_cpu(cpu, &cpu_associativity_changes_mask) { |
3be7db6a RJ |
1342 | /* |
1343 | * If siblings aren't flagged for changes, updates list | |
1344 | * will be too short. Skip on this update and set for next | |
1345 | * update. | |
1346 | */ | |
1347 | if (!cpumask_subset(cpu_sibling_mask(cpu), | |
1348 | &cpu_associativity_changes_mask)) { | |
1349 | pr_info("Sibling bits not set for associativity " | |
1350 | "change, cpu%d\n", cpu); | |
1351 | cpumask_or(&cpu_associativity_changes_mask, | |
1352 | &cpu_associativity_changes_mask, | |
1353 | cpu_sibling_mask(cpu)); | |
1354 | cpu = cpu_last_thread_sibling(cpu); | |
1355 | continue; | |
1356 | } | |
9eff1a38 | 1357 | |
ea05ba7c | 1358 | new_nid = find_and_online_cpu_nid(cpu); |
3be7db6a RJ |
1359 | |
1360 | if (new_nid == numa_cpu_lookup_table[cpu]) { | |
1361 | cpumask_andnot(&cpu_associativity_changes_mask, | |
1362 | &cpu_associativity_changes_mask, | |
1363 | cpu_sibling_mask(cpu)); | |
17f444c0 MB |
1364 | dbg("Assoc chg gives same node %d for cpu%d\n", |
1365 | new_nid, cpu); | |
3be7db6a RJ |
1366 | cpu = cpu_last_thread_sibling(cpu); |
1367 | continue; | |
1368 | } | |
9eff1a38 | 1369 | |
3be7db6a RJ |
1370 | for_each_cpu(sibling, cpu_sibling_mask(cpu)) { |
1371 | ud = &updates[i++]; | |
8bc93149 | 1372 | ud->next = &updates[i]; |
3be7db6a RJ |
1373 | ud->cpu = sibling; |
1374 | ud->new_nid = new_nid; | |
1375 | ud->old_nid = numa_cpu_lookup_table[sibling]; | |
1376 | cpumask_set_cpu(sibling, &updated_cpus); | |
3be7db6a RJ |
1377 | } |
1378 | cpu = cpu_last_thread_sibling(cpu); | |
30c05350 NF |
1379 | } |
1380 | ||
8bc93149 MB |
1381 | /* |
1382 | * Prevent processing of 'updates' from overflowing array | |
1383 | * where last entry filled in a 'next' pointer. | |
1384 | */ | |
1385 | if (i) | |
1386 | updates[i-1].next = NULL; | |
1387 | ||
2d73bae1 NA |
1388 | pr_debug("Topology update for the following CPUs:\n"); |
1389 | if (cpumask_weight(&updated_cpus)) { | |
1390 | for (ud = &updates[0]; ud; ud = ud->next) { | |
1391 | pr_debug("cpu %d moving from node %d " | |
1392 | "to %d\n", ud->cpu, | |
1393 | ud->old_nid, ud->new_nid); | |
1394 | } | |
1395 | } | |
1396 | ||
9a013361 MW |
1397 | /* |
1398 | * In cases where we have nothing to update (because the updates list | |
1399 | * is too short or because the new topology is same as the old one), | |
1400 | * skip invoking update_cpu_topology() via stop-machine(). This is | |
1401 | * necessary (and not just a fast-path optimization) since stop-machine | |
1402 | * can end up electing a random CPU to run update_cpu_topology(), and | |
1403 | * thus trick us into setting up incorrect cpu-node mappings (since | |
1404 | * 'updates' is kzalloc()'ed). | |
1405 | * | |
1406 | * And for the similar reason, we will skip all the following updating. | |
1407 | */ | |
1408 | if (!cpumask_weight(&updated_cpus)) | |
1409 | goto out; | |
1410 | ||
3e401f7a TJB |
1411 | if (cpus_locked) |
1412 | stop_machine_cpuslocked(update_cpu_topology, &updates[0], | |
1413 | &updated_cpus); | |
1414 | else | |
1415 | stop_machine(update_cpu_topology, &updates[0], &updated_cpus); | |
30c05350 | 1416 | |
d4edc5b6 SB |
1417 | /* |
1418 | * Update the numa-cpu lookup table with the new mappings, even for | |
1419 | * offline CPUs. It is best to perform this update from the stop- | |
1420 | * machine context. | |
1421 | */ | |
3e401f7a TJB |
1422 | if (cpus_locked) |
1423 | stop_machine_cpuslocked(update_lookup_table, &updates[0], | |
d4edc5b6 | 1424 | cpumask_of(raw_smp_processor_id())); |
3e401f7a TJB |
1425 | else |
1426 | stop_machine(update_lookup_table, &updates[0], | |
1427 | cpumask_of(raw_smp_processor_id())); | |
d4edc5b6 | 1428 | |
30c05350 | 1429 | for (ud = &updates[0]; ud; ud = ud->next) { |
dd023217 NF |
1430 | unregister_cpu_under_node(ud->cpu, ud->old_nid); |
1431 | register_cpu_under_node(ud->cpu, ud->new_nid); | |
1432 | ||
30c05350 | 1433 | dev = get_cpu_device(ud->cpu); |
8a25a2fd KS |
1434 | if (dev) |
1435 | kobject_uevent(&dev->kobj, KOBJ_CHANGE); | |
30c05350 | 1436 | cpumask_clear_cpu(ud->cpu, &cpu_associativity_changes_mask); |
79c5fceb | 1437 | changed = 1; |
9eff1a38 JL |
1438 | } |
1439 | ||
9a013361 | 1440 | out: |
30c05350 | 1441 | kfree(updates); |
79c5fceb | 1442 | return changed; |
9eff1a38 JL |
1443 | } |
1444 | ||
3e401f7a TJB |
1445 | int arch_update_cpu_topology(void) |
1446 | { | |
3e401f7a TJB |
1447 | return numa_update_cpu_topology(true); |
1448 | } | |
1449 | ||
9eff1a38 JL |
1450 | static void topology_work_fn(struct work_struct *work) |
1451 | { | |
1452 | rebuild_sched_domains(); | |
1453 | } | |
1454 | static DECLARE_WORK(topology_work, topology_work_fn); | |
1455 | ||
ec32dd66 | 1456 | static void topology_schedule_update(void) |
9eff1a38 JL |
1457 | { |
1458 | schedule_work(&topology_work); | |
1459 | } | |
1460 | ||
df7e828c | 1461 | static void topology_timer_fn(struct timer_list *unused) |
9eff1a38 | 1462 | { |
5d88aa85 | 1463 | if (prrn_enabled && cpumask_weight(&cpu_associativity_changes_mask)) |
9eff1a38 | 1464 | topology_schedule_update(); |
5d88aa85 JL |
1465 | else if (vphn_enabled) { |
1466 | if (update_cpu_associativity_changes_mask() > 0) | |
1467 | topology_schedule_update(); | |
1468 | reset_topology_timer(); | |
1469 | } | |
9eff1a38 | 1470 | } |
df7e828c | 1471 | static struct timer_list topology_timer; |
9eff1a38 | 1472 | |
5d88aa85 | 1473 | static void reset_topology_timer(void) |
9eff1a38 | 1474 | { |
8604895a MB |
1475 | if (vphn_enabled) |
1476 | mod_timer(&topology_timer, jiffies + topology_timer_secs * HZ); | |
9eff1a38 JL |
1477 | } |
1478 | ||
601abdc3 NF |
1479 | #ifdef CONFIG_SMP |
1480 | ||
5d88aa85 JL |
1481 | static int dt_update_callback(struct notifier_block *nb, |
1482 | unsigned long action, void *data) | |
1483 | { | |
f5242e5a | 1484 | struct of_reconfig_data *update = data; |
5d88aa85 JL |
1485 | int rc = NOTIFY_DONE; |
1486 | ||
1487 | switch (action) { | |
5d88aa85 | 1488 | case OF_RECONFIG_UPDATE_PROPERTY: |
e5480bdc | 1489 | if (of_node_is_type(update->dn, "cpu") && |
30c05350 | 1490 | !of_prop_cmp(update->prop->name, "ibm,associativity")) { |
5d88aa85 JL |
1491 | u32 core_id; |
1492 | of_property_read_u32(update->dn, "reg", &core_id); | |
81b61324 | 1493 | rc = dlpar_cpu_readd(core_id); |
5d88aa85 JL |
1494 | rc = NOTIFY_OK; |
1495 | } | |
1496 | break; | |
1497 | } | |
1498 | ||
1499 | return rc; | |
9eff1a38 JL |
1500 | } |
1501 | ||
5d88aa85 JL |
1502 | static struct notifier_block dt_update_nb = { |
1503 | .notifier_call = dt_update_callback, | |
1504 | }; | |
1505 | ||
601abdc3 NF |
1506 | #endif |
1507 | ||
9eff1a38 | 1508 | /* |
5d88aa85 | 1509 | * Start polling for associativity changes. |
9eff1a38 JL |
1510 | */ |
1511 | int start_topology_update(void) | |
1512 | { | |
1513 | int rc = 0; | |
1514 | ||
2d4d9b30 NL |
1515 | if (!topology_updates_enabled) |
1516 | return 0; | |
1517 | ||
5d88aa85 JL |
1518 | if (firmware_has_feature(FW_FEATURE_PRRN)) { |
1519 | if (!prrn_enabled) { | |
1520 | prrn_enabled = 1; | |
601abdc3 | 1521 | #ifdef CONFIG_SMP |
5d88aa85 | 1522 | rc = of_reconfig_notifier_register(&dt_update_nb); |
601abdc3 | 1523 | #endif |
5d88aa85 | 1524 | } |
a3496e91 MB |
1525 | } |
1526 | if (firmware_has_feature(FW_FEATURE_VPHN) && | |
f13c13a0 | 1527 | lppaca_shared_proc(get_lppaca())) { |
5d88aa85 | 1528 | if (!vphn_enabled) { |
5d88aa85 JL |
1529 | vphn_enabled = 1; |
1530 | setup_cpu_associativity_change_counters(); | |
df7e828c KC |
1531 | timer_setup(&topology_timer, topology_timer_fn, |
1532 | TIMER_DEFERRABLE); | |
5d88aa85 JL |
1533 | reset_topology_timer(); |
1534 | } | |
9eff1a38 JL |
1535 | } |
1536 | ||
65b9fdad MB |
1537 | pr_info("Starting topology update%s%s\n", |
1538 | (prrn_enabled ? " prrn_enabled" : ""), | |
1539 | (vphn_enabled ? " vphn_enabled" : "")); | |
1540 | ||
9eff1a38 JL |
1541 | return rc; |
1542 | } | |
9eff1a38 JL |
1543 | |
1544 | /* | |
1545 | * Disable polling for VPHN associativity changes. | |
1546 | */ | |
1547 | int stop_topology_update(void) | |
1548 | { | |
5d88aa85 JL |
1549 | int rc = 0; |
1550 | ||
2d4d9b30 NL |
1551 | if (!topology_updates_enabled) |
1552 | return 0; | |
1553 | ||
5d88aa85 JL |
1554 | if (prrn_enabled) { |
1555 | prrn_enabled = 0; | |
601abdc3 | 1556 | #ifdef CONFIG_SMP |
5d88aa85 | 1557 | rc = of_reconfig_notifier_unregister(&dt_update_nb); |
601abdc3 | 1558 | #endif |
a3496e91 MB |
1559 | } |
1560 | if (vphn_enabled) { | |
5d88aa85 JL |
1561 | vphn_enabled = 0; |
1562 | rc = del_timer_sync(&topology_timer); | |
1563 | } | |
1564 | ||
65b9fdad MB |
1565 | pr_info("Stopping topology update\n"); |
1566 | ||
5d88aa85 | 1567 | return rc; |
9eff1a38 | 1568 | } |
e04fa612 NF |
1569 | |
1570 | int prrn_is_enabled(void) | |
1571 | { | |
1572 | return prrn_enabled; | |
1573 | } | |
1574 | ||
2ea62630 SD |
1575 | void __init shared_proc_topology_init(void) |
1576 | { | |
1577 | if (lppaca_shared_proc(get_lppaca())) { | |
1578 | bitmap_fill(cpumask_bits(&cpu_associativity_changes_mask), | |
1579 | nr_cpumask_bits); | |
1580 | numa_update_cpu_topology(false); | |
1581 | } | |
1582 | } | |
1583 | ||
e04fa612 NF |
1584 | static int topology_read(struct seq_file *file, void *v) |
1585 | { | |
1586 | if (vphn_enabled || prrn_enabled) | |
1587 | seq_puts(file, "on\n"); | |
1588 | else | |
1589 | seq_puts(file, "off\n"); | |
1590 | ||
1591 | return 0; | |
1592 | } | |
1593 | ||
1594 | static int topology_open(struct inode *inode, struct file *file) | |
1595 | { | |
1596 | return single_open(file, topology_read, NULL); | |
1597 | } | |
1598 | ||
1599 | static ssize_t topology_write(struct file *file, const char __user *buf, | |
1600 | size_t count, loff_t *off) | |
1601 | { | |
1602 | char kbuf[4]; /* "on" or "off" plus null. */ | |
1603 | int read_len; | |
1604 | ||
1605 | read_len = count < 3 ? count : 3; | |
1606 | if (copy_from_user(kbuf, buf, read_len)) | |
1607 | return -EINVAL; | |
1608 | ||
1609 | kbuf[read_len] = '\0'; | |
1610 | ||
2d4d9b30 NL |
1611 | if (!strncmp(kbuf, "on", 2)) { |
1612 | topology_updates_enabled = true; | |
e04fa612 | 1613 | start_topology_update(); |
2d4d9b30 | 1614 | } else if (!strncmp(kbuf, "off", 3)) { |
e04fa612 | 1615 | stop_topology_update(); |
2d4d9b30 NL |
1616 | topology_updates_enabled = false; |
1617 | } else | |
e04fa612 NF |
1618 | return -EINVAL; |
1619 | ||
1620 | return count; | |
1621 | } | |
1622 | ||
1623 | static const struct file_operations topology_ops = { | |
1624 | .read = seq_read, | |
1625 | .write = topology_write, | |
1626 | .open = topology_open, | |
1627 | .release = single_release | |
1628 | }; | |
1629 | ||
1630 | static int topology_update_init(void) | |
1631 | { | |
2d4d9b30 | 1632 | start_topology_update(); |
2d73bae1 | 1633 | |
17f444c0 MB |
1634 | if (vphn_enabled) |
1635 | topology_schedule_update(); | |
1636 | ||
2d15b9b4 NA |
1637 | if (!proc_create("powerpc/topology_updates", 0644, NULL, &topology_ops)) |
1638 | return -ENOMEM; | |
e04fa612 | 1639 | |
17f444c0 | 1640 | topology_inited = 1; |
e04fa612 | 1641 | return 0; |
9eff1a38 | 1642 | } |
e04fa612 | 1643 | device_initcall(topology_update_init); |
39bf990e | 1644 | #endif /* CONFIG_PPC_SPLPAR */ |