]>
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 | |
1da177e4 | 170 | /* must hold reference to node during call */ |
b08a2a12 | 171 | static const __be32 *of_get_associativity(struct device_node *dev) |
1da177e4 | 172 | { |
e2eb6392 | 173 | return of_get_property(dev, "ibm,associativity", NULL); |
1da177e4 LT |
174 | } |
175 | ||
41eab6f8 AB |
176 | int __node_distance(int a, int b) |
177 | { | |
178 | int i; | |
179 | int distance = LOCAL_DISTANCE; | |
180 | ||
181 | if (!form1_affinity) | |
7122beee | 182 | return ((a == b) ? LOCAL_DISTANCE : REMOTE_DISTANCE); |
41eab6f8 AB |
183 | |
184 | for (i = 0; i < distance_ref_points_depth; i++) { | |
185 | if (distance_lookup_table[a][i] == distance_lookup_table[b][i]) | |
186 | break; | |
187 | ||
188 | /* Double the distance for each NUMA level */ | |
189 | distance *= 2; | |
190 | } | |
191 | ||
192 | return distance; | |
193 | } | |
12c743eb | 194 | EXPORT_SYMBOL(__node_distance); |
41eab6f8 AB |
195 | |
196 | static void initialize_distance_lookup_table(int nid, | |
b08a2a12 | 197 | const __be32 *associativity) |
41eab6f8 AB |
198 | { |
199 | int i; | |
200 | ||
201 | if (!form1_affinity) | |
202 | return; | |
203 | ||
204 | for (i = 0; i < distance_ref_points_depth; i++) { | |
b08a2a12 AP |
205 | const __be32 *entry; |
206 | ||
1d805440 | 207 | entry = &associativity[be32_to_cpu(distance_ref_points[i]) - 1]; |
b08a2a12 | 208 | distance_lookup_table[nid][i] = of_read_number(entry, 1); |
41eab6f8 AB |
209 | } |
210 | } | |
211 | ||
482ec7c4 NL |
212 | /* Returns nid in the range [0..MAX_NUMNODES-1], or -1 if no useful numa |
213 | * info is found. | |
214 | */ | |
b08a2a12 | 215 | static int associativity_to_nid(const __be32 *associativity) |
1da177e4 | 216 | { |
98fa15f3 | 217 | int nid = NUMA_NO_NODE; |
1da177e4 LT |
218 | |
219 | if (min_common_depth == -1) | |
482ec7c4 | 220 | goto out; |
1da177e4 | 221 | |
b08a2a12 AP |
222 | if (of_read_number(associativity, 1) >= min_common_depth) |
223 | nid = of_read_number(&associativity[min_common_depth], 1); | |
bc16a759 NL |
224 | |
225 | /* POWER4 LPAR uses 0xffff as invalid node */ | |
482ec7c4 | 226 | if (nid == 0xffff || nid >= MAX_NUMNODES) |
98fa15f3 | 227 | nid = NUMA_NO_NODE; |
41eab6f8 | 228 | |
b08a2a12 | 229 | if (nid > 0 && |
1d805440 ND |
230 | of_read_number(associativity, 1) >= distance_ref_points_depth) { |
231 | /* | |
232 | * Skip the length field and send start of associativity array | |
233 | */ | |
234 | initialize_distance_lookup_table(nid, associativity + 1); | |
235 | } | |
41eab6f8 | 236 | |
482ec7c4 | 237 | out: |
cf950b7a | 238 | return nid; |
1da177e4 LT |
239 | } |
240 | ||
9eff1a38 JL |
241 | /* Returns the nid associated with the given device tree node, |
242 | * or -1 if not found. | |
243 | */ | |
244 | static int of_node_to_nid_single(struct device_node *device) | |
245 | { | |
98fa15f3 | 246 | int nid = NUMA_NO_NODE; |
b08a2a12 | 247 | const __be32 *tmp; |
9eff1a38 JL |
248 | |
249 | tmp = of_get_associativity(device); | |
250 | if (tmp) | |
251 | nid = associativity_to_nid(tmp); | |
252 | return nid; | |
253 | } | |
254 | ||
953039c8 JK |
255 | /* Walk the device tree upwards, looking for an associativity id */ |
256 | int of_node_to_nid(struct device_node *device) | |
257 | { | |
98fa15f3 | 258 | int nid = NUMA_NO_NODE; |
953039c8 JK |
259 | |
260 | of_node_get(device); | |
261 | while (device) { | |
262 | nid = of_node_to_nid_single(device); | |
263 | if (nid != -1) | |
264 | break; | |
265 | ||
1def3758 | 266 | device = of_get_next_parent(device); |
953039c8 JK |
267 | } |
268 | of_node_put(device); | |
269 | ||
270 | return nid; | |
271 | } | |
be9ba9ff | 272 | EXPORT_SYMBOL(of_node_to_nid); |
953039c8 | 273 | |
1da177e4 LT |
274 | static int __init find_min_common_depth(void) |
275 | { | |
41eab6f8 | 276 | int depth; |
e70606eb | 277 | struct device_node *root; |
1da177e4 | 278 | |
1c8ee733 DS |
279 | if (firmware_has_feature(FW_FEATURE_OPAL)) |
280 | root = of_find_node_by_path("/ibm,opal"); | |
281 | else | |
282 | root = of_find_node_by_path("/rtas"); | |
e70606eb ME |
283 | if (!root) |
284 | root = of_find_node_by_path("/"); | |
1da177e4 LT |
285 | |
286 | /* | |
41eab6f8 AB |
287 | * This property is a set of 32-bit integers, each representing |
288 | * an index into the ibm,associativity nodes. | |
289 | * | |
290 | * With form 0 affinity the first integer is for an SMP configuration | |
291 | * (should be all 0's) and the second is for a normal NUMA | |
292 | * configuration. We have only one level of NUMA. | |
293 | * | |
294 | * With form 1 affinity the first integer is the most significant | |
295 | * NUMA boundary and the following are progressively less significant | |
296 | * boundaries. There can be more than one level of NUMA. | |
1da177e4 | 297 | */ |
e70606eb | 298 | distance_ref_points = of_get_property(root, |
41eab6f8 AB |
299 | "ibm,associativity-reference-points", |
300 | &distance_ref_points_depth); | |
301 | ||
302 | if (!distance_ref_points) { | |
303 | dbg("NUMA: ibm,associativity-reference-points not found.\n"); | |
304 | goto err; | |
305 | } | |
306 | ||
307 | distance_ref_points_depth /= sizeof(int); | |
1da177e4 | 308 | |
8002b0c5 NF |
309 | if (firmware_has_feature(FW_FEATURE_OPAL) || |
310 | firmware_has_feature(FW_FEATURE_TYPE1_AFFINITY)) { | |
311 | dbg("Using form 1 affinity\n"); | |
1c8ee733 | 312 | form1_affinity = 1; |
4b83c330 AB |
313 | } |
314 | ||
41eab6f8 | 315 | if (form1_affinity) { |
b08a2a12 | 316 | depth = of_read_number(distance_ref_points, 1); |
1da177e4 | 317 | } else { |
41eab6f8 AB |
318 | if (distance_ref_points_depth < 2) { |
319 | printk(KERN_WARNING "NUMA: " | |
320 | "short ibm,associativity-reference-points\n"); | |
321 | goto err; | |
322 | } | |
323 | ||
b08a2a12 | 324 | depth = of_read_number(&distance_ref_points[1], 1); |
1da177e4 | 325 | } |
1da177e4 | 326 | |
41eab6f8 AB |
327 | /* |
328 | * Warn and cap if the hardware supports more than | |
329 | * MAX_DISTANCE_REF_POINTS domains. | |
330 | */ | |
331 | if (distance_ref_points_depth > MAX_DISTANCE_REF_POINTS) { | |
332 | printk(KERN_WARNING "NUMA: distance array capped at " | |
333 | "%d entries\n", MAX_DISTANCE_REF_POINTS); | |
334 | distance_ref_points_depth = MAX_DISTANCE_REF_POINTS; | |
335 | } | |
336 | ||
e70606eb | 337 | of_node_put(root); |
1da177e4 | 338 | return depth; |
41eab6f8 AB |
339 | |
340 | err: | |
e70606eb | 341 | of_node_put(root); |
41eab6f8 | 342 | return -1; |
1da177e4 LT |
343 | } |
344 | ||
84c9fdd1 | 345 | static void __init get_n_mem_cells(int *n_addr_cells, int *n_size_cells) |
1da177e4 LT |
346 | { |
347 | struct device_node *memory = NULL; | |
1da177e4 LT |
348 | |
349 | memory = of_find_node_by_type(memory, "memory"); | |
54c23310 | 350 | if (!memory) |
84c9fdd1 | 351 | panic("numa.c: No memory nodes found!"); |
54c23310 | 352 | |
a8bda5dd | 353 | *n_addr_cells = of_n_addr_cells(memory); |
9213feea | 354 | *n_size_cells = of_n_size_cells(memory); |
84c9fdd1 | 355 | of_node_put(memory); |
1da177e4 LT |
356 | } |
357 | ||
b08a2a12 | 358 | static unsigned long read_n_cells(int n, const __be32 **buf) |
1da177e4 LT |
359 | { |
360 | unsigned long result = 0; | |
361 | ||
362 | while (n--) { | |
b08a2a12 | 363 | result = (result << 32) | of_read_number(*buf, 1); |
1da177e4 LT |
364 | (*buf)++; |
365 | } | |
366 | return result; | |
367 | } | |
368 | ||
8342681d NF |
369 | struct assoc_arrays { |
370 | u32 n_arrays; | |
371 | u32 array_sz; | |
b08a2a12 | 372 | const __be32 *arrays; |
8342681d NF |
373 | }; |
374 | ||
375 | /* | |
25985edc | 376 | * Retrieve and validate the list of associativity arrays for drconf |
8342681d NF |
377 | * memory from the ibm,associativity-lookup-arrays property of the |
378 | * device tree.. | |
379 | * | |
380 | * The layout of the ibm,associativity-lookup-arrays property is a number N | |
381 | * indicating the number of associativity arrays, followed by a number M | |
382 | * indicating the size of each associativity array, followed by a list | |
383 | * of N associativity arrays. | |
384 | */ | |
35f80deb | 385 | static int of_get_assoc_arrays(struct assoc_arrays *aa) |
8342681d | 386 | { |
35f80deb | 387 | struct device_node *memory; |
b08a2a12 | 388 | const __be32 *prop; |
8342681d NF |
389 | u32 len; |
390 | ||
35f80deb NF |
391 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); |
392 | if (!memory) | |
393 | return -1; | |
394 | ||
8342681d | 395 | prop = of_get_property(memory, "ibm,associativity-lookup-arrays", &len); |
35f80deb NF |
396 | if (!prop || len < 2 * sizeof(unsigned int)) { |
397 | of_node_put(memory); | |
8342681d | 398 | return -1; |
35f80deb | 399 | } |
8342681d | 400 | |
b08a2a12 AP |
401 | aa->n_arrays = of_read_number(prop++, 1); |
402 | aa->array_sz = of_read_number(prop++, 1); | |
8342681d | 403 | |
35f80deb NF |
404 | of_node_put(memory); |
405 | ||
42b2aa86 | 406 | /* Now that we know the number of arrays and size of each array, |
8342681d NF |
407 | * revalidate the size of the property read in. |
408 | */ | |
409 | if (len < (aa->n_arrays * aa->array_sz + 2) * sizeof(unsigned int)) | |
410 | return -1; | |
411 | ||
412 | aa->arrays = prop; | |
413 | return 0; | |
414 | } | |
415 | ||
416 | /* | |
417 | * This is like of_node_to_nid_single() for memory represented in the | |
418 | * ibm,dynamic-reconfiguration-memory node. | |
419 | */ | |
514a9cb3 | 420 | static int of_drconf_to_nid_single(struct drmem_lmb *lmb) |
8342681d | 421 | { |
b88fc309 | 422 | struct assoc_arrays aa = { .arrays = NULL }; |
8342681d NF |
423 | int default_nid = 0; |
424 | int nid = default_nid; | |
b88fc309 NF |
425 | int rc, index; |
426 | ||
427 | rc = of_get_assoc_arrays(&aa); | |
428 | if (rc) | |
429 | return default_nid; | |
8342681d | 430 | |
b88fc309 | 431 | if (min_common_depth > 0 && min_common_depth <= aa.array_sz && |
514a9cb3 NF |
432 | !(lmb->flags & DRCONF_MEM_AI_INVALID) && |
433 | lmb->aa_index < aa.n_arrays) { | |
434 | index = lmb->aa_index * aa.array_sz + min_common_depth - 1; | |
b88fc309 | 435 | nid = of_read_number(&aa.arrays[index], 1); |
8342681d NF |
436 | |
437 | if (nid == 0xffff || nid >= MAX_NUMNODES) | |
438 | nid = default_nid; | |
1d805440 ND |
439 | |
440 | if (nid > 0) { | |
514a9cb3 | 441 | index = lmb->aa_index * aa.array_sz; |
1d805440 | 442 | initialize_distance_lookup_table(nid, |
b88fc309 | 443 | &aa.arrays[index]); |
1d805440 | 444 | } |
8342681d NF |
445 | } |
446 | ||
447 | return nid; | |
448 | } | |
449 | ||
1da177e4 LT |
450 | /* |
451 | * Figure out to which domain a cpu belongs and stick it there. | |
452 | * Return the id of the domain used. | |
453 | */ | |
061d19f2 | 454 | static int numa_setup_cpu(unsigned long lcpu) |
1da177e4 | 455 | { |
98fa15f3 | 456 | int nid = NUMA_NO_NODE; |
d4edc5b6 SB |
457 | struct device_node *cpu; |
458 | ||
459 | /* | |
460 | * If a valid cpu-to-node mapping is already available, use it | |
461 | * directly instead of querying the firmware, since it represents | |
462 | * the most recent mapping notified to us by the platform (eg: VPHN). | |
463 | */ | |
464 | if ((nid = numa_cpu_lookup_table[lcpu]) >= 0) { | |
465 | map_cpu_to_node(lcpu, nid); | |
466 | return nid; | |
467 | } | |
468 | ||
469 | cpu = of_get_cpu_node(lcpu, NULL); | |
1da177e4 LT |
470 | |
471 | if (!cpu) { | |
472 | WARN_ON(1); | |
297cf502 LZ |
473 | if (cpu_present(lcpu)) |
474 | goto out_present; | |
475 | else | |
476 | goto out; | |
1da177e4 LT |
477 | } |
478 | ||
953039c8 | 479 | nid = of_node_to_nid_single(cpu); |
1da177e4 | 480 | |
297cf502 | 481 | out_present: |
ea05ba7c | 482 | if (nid < 0 || !node_possible(nid)) |
72c33688 | 483 | nid = first_online_node; |
1da177e4 | 484 | |
297cf502 | 485 | map_cpu_to_node(lcpu, nid); |
1da177e4 | 486 | of_node_put(cpu); |
297cf502 | 487 | out: |
cf950b7a | 488 | return nid; |
1da177e4 LT |
489 | } |
490 | ||
68fb18aa SB |
491 | static void verify_cpu_node_mapping(int cpu, int node) |
492 | { | |
493 | int base, sibling, i; | |
494 | ||
495 | /* Verify that all the threads in the core belong to the same node */ | |
496 | base = cpu_first_thread_sibling(cpu); | |
497 | ||
498 | for (i = 0; i < threads_per_core; i++) { | |
499 | sibling = base + i; | |
500 | ||
501 | if (sibling == cpu || cpu_is_offline(sibling)) | |
502 | continue; | |
503 | ||
504 | if (cpu_to_node(sibling) != node) { | |
505 | WARN(1, "CPU thread siblings %d and %d don't belong" | |
506 | " to the same node!\n", cpu, sibling); | |
507 | break; | |
508 | } | |
509 | } | |
510 | } | |
511 | ||
bdab88e0 SAS |
512 | /* Must run before sched domains notifier. */ |
513 | static int ppc_numa_cpu_prepare(unsigned int cpu) | |
514 | { | |
515 | int nid; | |
516 | ||
517 | nid = numa_setup_cpu(cpu); | |
518 | verify_cpu_node_mapping(cpu, nid); | |
519 | return 0; | |
520 | } | |
521 | ||
522 | static int ppc_numa_cpu_dead(unsigned int cpu) | |
523 | { | |
1da177e4 | 524 | #ifdef CONFIG_HOTPLUG_CPU |
bdab88e0 | 525 | unmap_cpu_from_node(cpu); |
1da177e4 | 526 | #endif |
bdab88e0 | 527 | return 0; |
1da177e4 LT |
528 | } |
529 | ||
530 | /* | |
531 | * Check and possibly modify a memory region to enforce the memory limit. | |
532 | * | |
533 | * Returns the size the region should have to enforce the memory limit. | |
534 | * This will either be the original value of size, a truncated value, | |
535 | * or zero. If the returned value of size is 0 the region should be | |
25985edc | 536 | * discarded as it lies wholly above the memory limit. |
1da177e4 | 537 | */ |
45fb6cea AB |
538 | static unsigned long __init numa_enforce_memory_limit(unsigned long start, |
539 | unsigned long size) | |
1da177e4 LT |
540 | { |
541 | /* | |
95f72d1e | 542 | * We use memblock_end_of_DRAM() in here instead of memory_limit because |
1da177e4 | 543 | * we've already adjusted it for the limit and it takes care of |
fe55249d MM |
544 | * having memory holes below the limit. Also, in the case of |
545 | * iommu_is_off, memory_limit is not set but is implicitly enforced. | |
1da177e4 | 546 | */ |
1da177e4 | 547 | |
95f72d1e | 548 | if (start + size <= memblock_end_of_DRAM()) |
1da177e4 LT |
549 | return size; |
550 | ||
95f72d1e | 551 | if (start >= memblock_end_of_DRAM()) |
1da177e4 LT |
552 | return 0; |
553 | ||
95f72d1e | 554 | return memblock_end_of_DRAM() - start; |
1da177e4 LT |
555 | } |
556 | ||
cf00085d C |
557 | /* |
558 | * Reads the counter for a given entry in | |
559 | * linux,drconf-usable-memory property | |
560 | */ | |
b08a2a12 | 561 | static inline int __init read_usm_ranges(const __be32 **usm) |
cf00085d C |
562 | { |
563 | /* | |
3fdfd990 | 564 | * For each lmb in ibm,dynamic-memory a corresponding |
cf00085d C |
565 | * entry in linux,drconf-usable-memory property contains |
566 | * a counter followed by that many (base, size) duple. | |
567 | * read the counter from linux,drconf-usable-memory | |
568 | */ | |
569 | return read_n_cells(n_mem_size_cells, usm); | |
570 | } | |
571 | ||
0204568a PM |
572 | /* |
573 | * Extract NUMA information from the ibm,dynamic-reconfiguration-memory | |
574 | * node. This assumes n_mem_{addr,size}_cells have been set. | |
575 | */ | |
514a9cb3 NF |
576 | static void __init numa_setup_drmem_lmb(struct drmem_lmb *lmb, |
577 | const __be32 **usm) | |
0204568a | 578 | { |
514a9cb3 NF |
579 | unsigned int ranges, is_kexec_kdump = 0; |
580 | unsigned long base, size, sz; | |
8342681d | 581 | int nid; |
8342681d | 582 | |
514a9cb3 NF |
583 | /* |
584 | * Skip this block if the reserved bit is set in flags (0x80) | |
585 | * or if the block is not assigned to this partition (0x8) | |
586 | */ | |
587 | if ((lmb->flags & DRCONF_MEM_RESERVED) | |
588 | || !(lmb->flags & DRCONF_MEM_ASSIGNED)) | |
8342681d NF |
589 | return; |
590 | ||
514a9cb3 | 591 | if (*usm) |
cf00085d C |
592 | is_kexec_kdump = 1; |
593 | ||
514a9cb3 NF |
594 | base = lmb->base_addr; |
595 | size = drmem_lmb_size(); | |
596 | ranges = 1; | |
8342681d | 597 | |
514a9cb3 NF |
598 | if (is_kexec_kdump) { |
599 | ranges = read_usm_ranges(usm); | |
600 | if (!ranges) /* there are no (base, size) duple */ | |
601 | return; | |
602 | } | |
8342681d | 603 | |
514a9cb3 | 604 | do { |
cf00085d | 605 | if (is_kexec_kdump) { |
514a9cb3 NF |
606 | base = read_n_cells(n_mem_addr_cells, usm); |
607 | size = read_n_cells(n_mem_size_cells, usm); | |
cf00085d | 608 | } |
514a9cb3 NF |
609 | |
610 | nid = of_drconf_to_nid_single(lmb); | |
611 | fake_numa_create_new_node(((base + size) >> PAGE_SHIFT), | |
612 | &nid); | |
613 | node_set_online(nid); | |
614 | sz = numa_enforce_memory_limit(base, size); | |
615 | if (sz) | |
616 | memblock_set_node(base, sz, &memblock.memory, nid); | |
617 | } while (--ranges); | |
0204568a PM |
618 | } |
619 | ||
1da177e4 LT |
620 | static int __init parse_numa_properties(void) |
621 | { | |
94db7c5e | 622 | struct device_node *memory; |
482ec7c4 | 623 | int default_nid = 0; |
1da177e4 LT |
624 | unsigned long i; |
625 | ||
626 | if (numa_enabled == 0) { | |
627 | printk(KERN_WARNING "NUMA disabled by user\n"); | |
628 | return -1; | |
629 | } | |
630 | ||
1da177e4 LT |
631 | min_common_depth = find_min_common_depth(); |
632 | ||
1da177e4 LT |
633 | if (min_common_depth < 0) |
634 | return min_common_depth; | |
635 | ||
bf4b85b0 NL |
636 | dbg("NUMA associativity depth for CPU/Memory: %d\n", min_common_depth); |
637 | ||
1da177e4 | 638 | /* |
482ec7c4 NL |
639 | * Even though we connect cpus to numa domains later in SMP |
640 | * init, we need to know the node ids now. This is because | |
641 | * each node to be onlined must have NODE_DATA etc backing it. | |
1da177e4 | 642 | */ |
482ec7c4 | 643 | for_each_present_cpu(i) { |
dfbe93a2 | 644 | struct device_node *cpu; |
cf950b7a | 645 | int nid; |
1da177e4 | 646 | |
8b16cd23 | 647 | cpu = of_get_cpu_node(i, NULL); |
482ec7c4 | 648 | BUG_ON(!cpu); |
953039c8 | 649 | nid = of_node_to_nid_single(cpu); |
482ec7c4 | 650 | of_node_put(cpu); |
1da177e4 | 651 | |
482ec7c4 NL |
652 | /* |
653 | * Don't fall back to default_nid yet -- we will plug | |
654 | * cpus into nodes once the memory scan has discovered | |
655 | * the topology. | |
656 | */ | |
657 | if (nid < 0) | |
658 | continue; | |
659 | node_set_online(nid); | |
1da177e4 LT |
660 | } |
661 | ||
237a0989 | 662 | get_n_mem_cells(&n_mem_addr_cells, &n_mem_size_cells); |
94db7c5e AB |
663 | |
664 | for_each_node_by_type(memory, "memory") { | |
1da177e4 LT |
665 | unsigned long start; |
666 | unsigned long size; | |
cf950b7a | 667 | int nid; |
1da177e4 | 668 | int ranges; |
b08a2a12 | 669 | const __be32 *memcell_buf; |
1da177e4 LT |
670 | unsigned int len; |
671 | ||
e2eb6392 | 672 | memcell_buf = of_get_property(memory, |
ba759485 ME |
673 | "linux,usable-memory", &len); |
674 | if (!memcell_buf || len <= 0) | |
e2eb6392 | 675 | memcell_buf = of_get_property(memory, "reg", &len); |
1da177e4 LT |
676 | if (!memcell_buf || len <= 0) |
677 | continue; | |
678 | ||
cc5d0189 BH |
679 | /* ranges in cell */ |
680 | ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells); | |
1da177e4 LT |
681 | new_range: |
682 | /* these are order-sensitive, and modify the buffer pointer */ | |
237a0989 MK |
683 | start = read_n_cells(n_mem_addr_cells, &memcell_buf); |
684 | size = read_n_cells(n_mem_size_cells, &memcell_buf); | |
1da177e4 | 685 | |
482ec7c4 NL |
686 | /* |
687 | * Assumption: either all memory nodes or none will | |
688 | * have associativity properties. If none, then | |
689 | * everything goes to default_nid. | |
690 | */ | |
953039c8 | 691 | nid = of_node_to_nid_single(memory); |
482ec7c4 NL |
692 | if (nid < 0) |
693 | nid = default_nid; | |
1daa6d08 BS |
694 | |
695 | fake_numa_create_new_node(((start + size) >> PAGE_SHIFT), &nid); | |
482ec7c4 | 696 | node_set_online(nid); |
1da177e4 | 697 | |
7656cd8e RA |
698 | size = numa_enforce_memory_limit(start, size); |
699 | if (size) | |
700 | memblock_set_node(start, size, &memblock.memory, nid); | |
1da177e4 LT |
701 | |
702 | if (--ranges) | |
703 | goto new_range; | |
704 | } | |
705 | ||
0204568a | 706 | /* |
dfbe93a2 AB |
707 | * Now do the same thing for each MEMBLOCK listed in the |
708 | * ibm,dynamic-memory property in the | |
709 | * ibm,dynamic-reconfiguration-memory node. | |
0204568a PM |
710 | */ |
711 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); | |
514a9cb3 NF |
712 | if (memory) { |
713 | walk_drmem_lmbs(memory, numa_setup_drmem_lmb); | |
714 | of_node_put(memory); | |
715 | } | |
0204568a | 716 | |
1da177e4 LT |
717 | return 0; |
718 | } | |
719 | ||
720 | static void __init setup_nonnuma(void) | |
721 | { | |
95f72d1e YL |
722 | unsigned long top_of_ram = memblock_end_of_DRAM(); |
723 | unsigned long total_ram = memblock_phys_mem_size(); | |
c67c3cb4 | 724 | unsigned long start_pfn, end_pfn; |
28be7072 BH |
725 | unsigned int nid = 0; |
726 | struct memblock_region *reg; | |
1da177e4 | 727 | |
e110b281 | 728 | printk(KERN_DEBUG "Top of RAM: 0x%lx, Total RAM: 0x%lx\n", |
1da177e4 | 729 | top_of_ram, total_ram); |
e110b281 | 730 | printk(KERN_DEBUG "Memory hole size: %ldMB\n", |
1da177e4 LT |
731 | (top_of_ram - total_ram) >> 20); |
732 | ||
28be7072 | 733 | for_each_memblock(memory, reg) { |
c7fc2de0 YL |
734 | start_pfn = memblock_region_memory_base_pfn(reg); |
735 | end_pfn = memblock_region_memory_end_pfn(reg); | |
1daa6d08 BS |
736 | |
737 | fake_numa_create_new_node(end_pfn, &nid); | |
1d7cfe18 | 738 | memblock_set_node(PFN_PHYS(start_pfn), |
e7e8de59 TC |
739 | PFN_PHYS(end_pfn - start_pfn), |
740 | &memblock.memory, nid); | |
1daa6d08 | 741 | node_set_online(nid); |
c67c3cb4 | 742 | } |
1da177e4 LT |
743 | } |
744 | ||
4b703a23 AB |
745 | void __init dump_numa_cpu_topology(void) |
746 | { | |
747 | unsigned int node; | |
748 | unsigned int cpu, count; | |
749 | ||
750 | if (min_common_depth == -1 || !numa_enabled) | |
751 | return; | |
752 | ||
753 | for_each_online_node(node) { | |
8467801c | 754 | pr_info("Node %d CPUs:", node); |
4b703a23 AB |
755 | |
756 | count = 0; | |
757 | /* | |
758 | * If we used a CPU iterator here we would miss printing | |
759 | * the holes in the cpumap. | |
760 | */ | |
25863de0 AB |
761 | for (cpu = 0; cpu < nr_cpu_ids; cpu++) { |
762 | if (cpumask_test_cpu(cpu, | |
763 | node_to_cpumask_map[node])) { | |
4b703a23 | 764 | if (count == 0) |
8467801c | 765 | pr_cont(" %u", cpu); |
4b703a23 AB |
766 | ++count; |
767 | } else { | |
768 | if (count > 1) | |
8467801c | 769 | pr_cont("-%u", cpu - 1); |
4b703a23 AB |
770 | count = 0; |
771 | } | |
772 | } | |
773 | ||
774 | if (count > 1) | |
8467801c AK |
775 | pr_cont("-%u", nr_cpu_ids - 1); |
776 | pr_cont("\n"); | |
4b703a23 AB |
777 | } |
778 | } | |
779 | ||
10239733 AB |
780 | /* Initialize NODE_DATA for a node on the local memory */ |
781 | static void __init setup_node_data(int nid, u64 start_pfn, u64 end_pfn) | |
4a618669 | 782 | { |
10239733 AB |
783 | u64 spanned_pages = end_pfn - start_pfn; |
784 | const size_t nd_size = roundup(sizeof(pg_data_t), SMP_CACHE_BYTES); | |
785 | u64 nd_pa; | |
786 | void *nd; | |
787 | int tnid; | |
4a618669 | 788 | |
9a8dd708 | 789 | nd_pa = memblock_phys_alloc_try_nid(nd_size, SMP_CACHE_BYTES, nid); |
33755574 MR |
790 | if (!nd_pa) |
791 | panic("Cannot allocate %zu bytes for node %d data\n", | |
792 | nd_size, nid); | |
793 | ||
10239733 | 794 | nd = __va(nd_pa); |
4a618669 | 795 | |
10239733 AB |
796 | /* report and initialize */ |
797 | pr_info(" NODE_DATA [mem %#010Lx-%#010Lx]\n", | |
798 | nd_pa, nd_pa + nd_size - 1); | |
799 | tnid = early_pfn_to_nid(nd_pa >> PAGE_SHIFT); | |
800 | if (tnid != nid) | |
801 | pr_info(" NODE_DATA(%d) on node %d\n", nid, tnid); | |
4a618669 | 802 | |
10239733 AB |
803 | node_data[nid] = nd; |
804 | memset(NODE_DATA(nid), 0, sizeof(pg_data_t)); | |
805 | NODE_DATA(nid)->node_id = nid; | |
806 | NODE_DATA(nid)->node_start_pfn = start_pfn; | |
807 | NODE_DATA(nid)->node_spanned_pages = spanned_pages; | |
808 | } | |
4a618669 | 809 | |
a346137e MB |
810 | static void __init find_possible_nodes(void) |
811 | { | |
812 | struct device_node *rtas; | |
813 | u32 numnodes, i; | |
814 | ||
815 | if (min_common_depth <= 0) | |
816 | return; | |
817 | ||
818 | rtas = of_find_node_by_path("/rtas"); | |
819 | if (!rtas) | |
820 | return; | |
821 | ||
822 | if (of_property_read_u32_index(rtas, | |
823 | "ibm,max-associativity-domains", | |
824 | min_common_depth, &numnodes)) | |
825 | goto out; | |
826 | ||
827 | for (i = 0; i < numnodes; i++) { | |
ea05ba7c | 828 | if (!node_possible(i)) |
a346137e | 829 | node_set(i, node_possible_map); |
a346137e MB |
830 | } |
831 | ||
832 | out: | |
833 | of_node_put(rtas); | |
834 | } | |
835 | ||
9bd9be00 | 836 | void __init mem_topology_setup(void) |
1da177e4 | 837 | { |
9bd9be00 | 838 | int cpu; |
1da177e4 LT |
839 | |
840 | if (parse_numa_properties()) | |
841 | setup_nonnuma(); | |
1da177e4 | 842 | |
3af229f2 | 843 | /* |
a346137e MB |
844 | * Modify the set of possible NUMA nodes to reflect information |
845 | * available about the set of online nodes, and the set of nodes | |
846 | * that we expect to make use of for this platform's affinity | |
847 | * calculations. | |
3af229f2 NA |
848 | */ |
849 | nodes_and(node_possible_map, node_possible_map, node_online_map); | |
850 | ||
a346137e MB |
851 | find_possible_nodes(); |
852 | ||
9bd9be00 NP |
853 | setup_node_to_cpumask_map(); |
854 | ||
855 | reset_numa_cpu_lookup_table(); | |
856 | ||
857 | for_each_present_cpu(cpu) | |
858 | numa_setup_cpu(cpu); | |
859 | } | |
860 | ||
861 | void __init initmem_init(void) | |
862 | { | |
863 | int nid; | |
864 | ||
865 | max_low_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT; | |
866 | max_pfn = max_low_pfn; | |
867 | ||
868 | memblock_dump_all(); | |
869 | ||
1da177e4 | 870 | for_each_online_node(nid) { |
c67c3cb4 | 871 | unsigned long start_pfn, end_pfn; |
1da177e4 | 872 | |
c67c3cb4 | 873 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); |
10239733 | 874 | setup_node_data(nid, start_pfn, end_pfn); |
8f64e1f2 | 875 | sparse_memory_present_with_active_regions(nid); |
4a618669 | 876 | } |
d3f6204a | 877 | |
21098b9e | 878 | sparse_init(); |
25863de0 | 879 | |
2fabf084 NA |
880 | /* |
881 | * We need the numa_cpu_lookup_table to be accurate for all CPUs, | |
882 | * even before we online them, so that we can use cpu_to_{node,mem} | |
883 | * early in boot, cf. smp_prepare_cpus(). | |
bdab88e0 SAS |
884 | * _nocalls() + manual invocation is used because cpuhp is not yet |
885 | * initialized for the boot CPU. | |
2fabf084 | 886 | */ |
73c1b41e | 887 | cpuhp_setup_state_nocalls(CPUHP_POWER_NUMA_PREPARE, "powerpc/numa:prepare", |
bdab88e0 | 888 | ppc_numa_cpu_prepare, ppc_numa_cpu_dead); |
1da177e4 LT |
889 | } |
890 | ||
1da177e4 LT |
891 | static int __init early_numa(char *p) |
892 | { | |
893 | if (!p) | |
894 | return 0; | |
895 | ||
896 | if (strstr(p, "off")) | |
897 | numa_enabled = 0; | |
898 | ||
899 | if (strstr(p, "debug")) | |
900 | numa_debug = 1; | |
901 | ||
1daa6d08 BS |
902 | p = strstr(p, "fake="); |
903 | if (p) | |
904 | cmdline = p + strlen("fake="); | |
905 | ||
1da177e4 LT |
906 | return 0; |
907 | } | |
908 | early_param("numa", early_numa); | |
237a0989 | 909 | |
558f8649 NL |
910 | /* |
911 | * The platform can inform us through one of several mechanisms | |
912 | * (post-migration device tree updates, PRRN or VPHN) that the NUMA | |
913 | * assignment of a resource has changed. This controls whether we act | |
914 | * on that. Disabled by default. | |
915 | */ | |
916 | static bool topology_updates_enabled; | |
2d73bae1 NA |
917 | |
918 | static int __init early_topology_updates(char *p) | |
919 | { | |
920 | if (!p) | |
921 | return 0; | |
922 | ||
558f8649 NL |
923 | if (!strcmp(p, "on")) { |
924 | pr_warn("Caution: enabling topology updates\n"); | |
925 | topology_updates_enabled = true; | |
2d73bae1 NA |
926 | } |
927 | ||
928 | return 0; | |
929 | } | |
930 | early_param("topology_updates", early_topology_updates); | |
931 | ||
237a0989 | 932 | #ifdef CONFIG_MEMORY_HOTPLUG |
0db9360a | 933 | /* |
0f16ef7f NF |
934 | * Find the node associated with a hot added memory section for |
935 | * memory represented in the device tree by the property | |
936 | * ibm,dynamic-reconfiguration-memory/ibm,dynamic-memory. | |
0db9360a | 937 | */ |
514a9cb3 | 938 | static int hot_add_drconf_scn_to_nid(unsigned long scn_addr) |
0db9360a | 939 | { |
514a9cb3 | 940 | struct drmem_lmb *lmb; |
3fdfd990 | 941 | unsigned long lmb_size; |
98fa15f3 | 942 | int nid = NUMA_NO_NODE; |
0db9360a | 943 | |
514a9cb3 | 944 | lmb_size = drmem_lmb_size(); |
0db9360a | 945 | |
514a9cb3 | 946 | for_each_drmem_lmb(lmb) { |
0db9360a NF |
947 | /* skip this block if it is reserved or not assigned to |
948 | * this partition */ | |
514a9cb3 NF |
949 | if ((lmb->flags & DRCONF_MEM_RESERVED) |
950 | || !(lmb->flags & DRCONF_MEM_ASSIGNED)) | |
0db9360a NF |
951 | continue; |
952 | ||
514a9cb3 NF |
953 | if ((scn_addr < lmb->base_addr) |
954 | || (scn_addr >= (lmb->base_addr + lmb_size))) | |
0f16ef7f NF |
955 | continue; |
956 | ||
514a9cb3 | 957 | nid = of_drconf_to_nid_single(lmb); |
0f16ef7f NF |
958 | break; |
959 | } | |
960 | ||
961 | return nid; | |
962 | } | |
963 | ||
964 | /* | |
965 | * Find the node associated with a hot added memory section for memory | |
966 | * represented in the device tree as a node (i.e. memory@XXXX) for | |
95f72d1e | 967 | * each memblock. |
0f16ef7f | 968 | */ |
ec32dd66 | 969 | static int hot_add_node_scn_to_nid(unsigned long scn_addr) |
0f16ef7f | 970 | { |
94db7c5e | 971 | struct device_node *memory; |
98fa15f3 | 972 | int nid = NUMA_NO_NODE; |
0f16ef7f | 973 | |
94db7c5e | 974 | for_each_node_by_type(memory, "memory") { |
0f16ef7f NF |
975 | unsigned long start, size; |
976 | int ranges; | |
b08a2a12 | 977 | const __be32 *memcell_buf; |
0f16ef7f NF |
978 | unsigned int len; |
979 | ||
980 | memcell_buf = of_get_property(memory, "reg", &len); | |
981 | if (!memcell_buf || len <= 0) | |
982 | continue; | |
983 | ||
984 | /* ranges in cell */ | |
985 | ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells); | |
986 | ||
987 | while (ranges--) { | |
988 | start = read_n_cells(n_mem_addr_cells, &memcell_buf); | |
989 | size = read_n_cells(n_mem_size_cells, &memcell_buf); | |
990 | ||
991 | if ((scn_addr < start) || (scn_addr >= (start + size))) | |
992 | continue; | |
993 | ||
994 | nid = of_node_to_nid_single(memory); | |
995 | break; | |
996 | } | |
0db9360a | 997 | |
0f16ef7f NF |
998 | if (nid >= 0) |
999 | break; | |
0db9360a NF |
1000 | } |
1001 | ||
60831842 AB |
1002 | of_node_put(memory); |
1003 | ||
0f16ef7f | 1004 | return nid; |
0db9360a NF |
1005 | } |
1006 | ||
237a0989 MK |
1007 | /* |
1008 | * Find the node associated with a hot added memory section. Section | |
95f72d1e YL |
1009 | * corresponds to a SPARSEMEM section, not an MEMBLOCK. It is assumed that |
1010 | * sections are fully contained within a single MEMBLOCK. | |
237a0989 MK |
1011 | */ |
1012 | int hot_add_scn_to_nid(unsigned long scn_addr) | |
1013 | { | |
1014 | struct device_node *memory = NULL; | |
4a3bac4e | 1015 | int nid; |
237a0989 MK |
1016 | |
1017 | if (!numa_enabled || (min_common_depth < 0)) | |
72c33688 | 1018 | return first_online_node; |
0db9360a NF |
1019 | |
1020 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); | |
1021 | if (memory) { | |
514a9cb3 | 1022 | nid = hot_add_drconf_scn_to_nid(scn_addr); |
0db9360a | 1023 | of_node_put(memory); |
0f16ef7f NF |
1024 | } else { |
1025 | nid = hot_add_node_scn_to_nid(scn_addr); | |
0db9360a | 1026 | } |
237a0989 | 1027 | |
2a8628d4 | 1028 | if (nid < 0 || !node_possible(nid)) |
72c33688 | 1029 | nid = first_online_node; |
237a0989 | 1030 | |
0f16ef7f | 1031 | return nid; |
237a0989 | 1032 | } |
0f16ef7f | 1033 | |
cd34206e NA |
1034 | static u64 hot_add_drconf_memory_max(void) |
1035 | { | |
e70bd3ae | 1036 | struct device_node *memory = NULL; |
45b64ee6 | 1037 | struct device_node *dn = NULL; |
45b64ee6 | 1038 | const __be64 *lrdr = NULL; |
45b64ee6 BR |
1039 | |
1040 | dn = of_find_node_by_path("/rtas"); | |
1041 | if (dn) { | |
1042 | lrdr = of_get_property(dn, "ibm,lrdr-capacity", NULL); | |
1043 | of_node_put(dn); | |
1044 | if (lrdr) | |
1045 | return be64_to_cpup(lrdr); | |
1046 | } | |
cd34206e | 1047 | |
e70bd3ae BR |
1048 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); |
1049 | if (memory) { | |
e70bd3ae | 1050 | of_node_put(memory); |
514a9cb3 | 1051 | return drmem_lmb_memory_max(); |
e70bd3ae | 1052 | } |
45b64ee6 | 1053 | return 0; |
cd34206e NA |
1054 | } |
1055 | ||
1056 | /* | |
1057 | * memory_hotplug_max - return max address of memory that may be added | |
1058 | * | |
1059 | * This is currently only used on systems that support drconfig memory | |
1060 | * hotplug. | |
1061 | */ | |
1062 | u64 memory_hotplug_max(void) | |
1063 | { | |
1064 | return max(hot_add_drconf_memory_max(), memblock_end_of_DRAM()); | |
1065 | } | |
237a0989 | 1066 | #endif /* CONFIG_MEMORY_HOTPLUG */ |
9eff1a38 | 1067 | |
bd03403a | 1068 | /* Virtual Processor Home Node (VPHN) support */ |
39bf990e | 1069 | #ifdef CONFIG_PPC_SPLPAR |
4b6cfb2a | 1070 | |
47d99948 | 1071 | #include "book3s64/vphn.h" |
4b6cfb2a | 1072 | |
30c05350 NF |
1073 | struct topology_update_data { |
1074 | struct topology_update_data *next; | |
1075 | unsigned int cpu; | |
1076 | int old_nid; | |
1077 | int new_nid; | |
1078 | }; | |
1079 | ||
cee5405d MB |
1080 | #define TOPOLOGY_DEF_TIMER_SECS 60 |
1081 | ||
5de16699 | 1082 | static u8 vphn_cpu_change_counts[NR_CPUS][MAX_DISTANCE_REF_POINTS]; |
9eff1a38 JL |
1083 | static cpumask_t cpu_associativity_changes_mask; |
1084 | static int vphn_enabled; | |
5d88aa85 JL |
1085 | static int prrn_enabled; |
1086 | static void reset_topology_timer(void); | |
cee5405d | 1087 | static int topology_timer_secs = 1; |
17f444c0 | 1088 | static int topology_inited; |
9eff1a38 | 1089 | |
cee5405d MB |
1090 | /* |
1091 | * Change polling interval for associativity changes. | |
1092 | */ | |
1093 | int timed_topology_update(int nsecs) | |
1094 | { | |
1095 | if (vphn_enabled) { | |
1096 | if (nsecs > 0) | |
1097 | topology_timer_secs = nsecs; | |
1098 | else | |
1099 | topology_timer_secs = TOPOLOGY_DEF_TIMER_SECS; | |
1100 | ||
1101 | reset_topology_timer(); | |
1102 | } | |
1103 | ||
1104 | return 0; | |
1105 | } | |
9eff1a38 JL |
1106 | |
1107 | /* | |
1108 | * Store the current values of the associativity change counters in the | |
1109 | * hypervisor. | |
1110 | */ | |
1111 | static void setup_cpu_associativity_change_counters(void) | |
1112 | { | |
cd9d6cc7 | 1113 | int cpu; |
9eff1a38 | 1114 | |
5de16699 AB |
1115 | /* The VPHN feature supports a maximum of 8 reference points */ |
1116 | BUILD_BUG_ON(MAX_DISTANCE_REF_POINTS > 8); | |
1117 | ||
9eff1a38 | 1118 | for_each_possible_cpu(cpu) { |
cd9d6cc7 | 1119 | int i; |
9eff1a38 | 1120 | u8 *counts = vphn_cpu_change_counts[cpu]; |
499dcd41 | 1121 | volatile u8 *hypervisor_counts = lppaca_of(cpu).vphn_assoc_counts; |
9eff1a38 | 1122 | |
5de16699 | 1123 | for (i = 0; i < distance_ref_points_depth; i++) |
9eff1a38 | 1124 | counts[i] = hypervisor_counts[i]; |
9eff1a38 JL |
1125 | } |
1126 | } | |
1127 | ||
1128 | /* | |
1129 | * The hypervisor maintains a set of 8 associativity change counters in | |
1130 | * the VPA of each cpu that correspond to the associativity levels in the | |
1131 | * ibm,associativity-reference-points property. When an associativity | |
1132 | * level changes, the corresponding counter is incremented. | |
1133 | * | |
1134 | * Set a bit in cpu_associativity_changes_mask for each cpu whose home | |
1135 | * node associativity levels have changed. | |
1136 | * | |
1137 | * Returns the number of cpus with unhandled associativity changes. | |
1138 | */ | |
1139 | static int update_cpu_associativity_changes_mask(void) | |
1140 | { | |
5d88aa85 | 1141 | int cpu; |
9eff1a38 JL |
1142 | cpumask_t *changes = &cpu_associativity_changes_mask; |
1143 | ||
9eff1a38 JL |
1144 | for_each_possible_cpu(cpu) { |
1145 | int i, changed = 0; | |
1146 | u8 *counts = vphn_cpu_change_counts[cpu]; | |
499dcd41 | 1147 | volatile u8 *hypervisor_counts = lppaca_of(cpu).vphn_assoc_counts; |
9eff1a38 | 1148 | |
5de16699 | 1149 | for (i = 0; i < distance_ref_points_depth; i++) { |
d69043e8 | 1150 | if (hypervisor_counts[i] != counts[i]) { |
9eff1a38 JL |
1151 | counts[i] = hypervisor_counts[i]; |
1152 | changed = 1; | |
1153 | } | |
1154 | } | |
1155 | if (changed) { | |
3be7db6a RJ |
1156 | cpumask_or(changes, changes, cpu_sibling_mask(cpu)); |
1157 | cpu = cpu_last_thread_sibling(cpu); | |
9eff1a38 JL |
1158 | } |
1159 | } | |
1160 | ||
5d88aa85 | 1161 | return cpumask_weight(changes); |
9eff1a38 JL |
1162 | } |
1163 | ||
9eff1a38 JL |
1164 | /* |
1165 | * Retrieve the new associativity information for a virtual processor's | |
1166 | * home node. | |
1167 | */ | |
b08a2a12 | 1168 | static long hcall_vphn(unsigned long cpu, __be32 *associativity) |
9eff1a38 | 1169 | { |
cd9d6cc7 | 1170 | long rc; |
9eff1a38 JL |
1171 | long retbuf[PLPAR_HCALL9_BUFSIZE] = {0}; |
1172 | u64 flags = 1; | |
1173 | int hwcpu = get_hard_smp_processor_id(cpu); | |
1174 | ||
1175 | rc = plpar_hcall9(H_HOME_NODE_ASSOCIATIVITY, retbuf, flags, hwcpu); | |
1176 | vphn_unpack_associativity(retbuf, associativity); | |
1177 | ||
1178 | return rc; | |
1179 | } | |
1180 | ||
1181 | static long vphn_get_associativity(unsigned long cpu, | |
b08a2a12 | 1182 | __be32 *associativity) |
9eff1a38 | 1183 | { |
cd9d6cc7 | 1184 | long rc; |
9eff1a38 JL |
1185 | |
1186 | rc = hcall_vphn(cpu, associativity); | |
1187 | ||
1188 | switch (rc) { | |
1189 | case H_FUNCTION: | |
437ccdc8 | 1190 | printk_once(KERN_INFO |
9eff1a38 JL |
1191 | "VPHN is not supported. Disabling polling...\n"); |
1192 | stop_topology_update(); | |
1193 | break; | |
1194 | case H_HARDWARE: | |
1195 | printk(KERN_ERR | |
1196 | "hcall_vphn() experienced a hardware fault " | |
1197 | "preventing VPHN. Disabling polling...\n"); | |
1198 | stop_topology_update(); | |
17f444c0 MB |
1199 | break; |
1200 | case H_SUCCESS: | |
1201 | dbg("VPHN hcall succeeded. Reset polling...\n"); | |
cee5405d | 1202 | timed_topology_update(0); |
17f444c0 | 1203 | break; |
9eff1a38 JL |
1204 | } |
1205 | ||
1206 | return rc; | |
1207 | } | |
1208 | ||
e67e02a5 | 1209 | int find_and_online_cpu_nid(int cpu) |
ea05ba7c MB |
1210 | { |
1211 | __be32 associativity[VPHN_ASSOC_BUFSIZE] = {0}; | |
1212 | int new_nid; | |
1213 | ||
1214 | /* Use associativity from first thread for all siblings */ | |
2483ef05 SD |
1215 | if (vphn_get_associativity(cpu, associativity)) |
1216 | return cpu_to_node(cpu); | |
1217 | ||
ea05ba7c MB |
1218 | new_nid = associativity_to_nid(associativity); |
1219 | if (new_nid < 0 || !node_possible(new_nid)) | |
1220 | new_nid = first_online_node; | |
1221 | ||
1222 | if (NODE_DATA(new_nid) == NULL) { | |
1223 | #ifdef CONFIG_MEMORY_HOTPLUG | |
1224 | /* | |
1225 | * Need to ensure that NODE_DATA is initialized for a node from | |
1226 | * available memory (see memblock_alloc_try_nid). If unable to | |
1227 | * init the node, then default to nearest node that has memory | |
ac1788cc SD |
1228 | * installed. Skip onlining a node if the subsystems are not |
1229 | * yet initialized. | |
ea05ba7c | 1230 | */ |
ac1788cc | 1231 | if (!topology_inited || try_online_node(new_nid)) |
ea05ba7c MB |
1232 | new_nid = first_online_node; |
1233 | #else | |
1234 | /* | |
1235 | * Default to using the nearest node that has memory installed. | |
1236 | * Otherwise, it would be necessary to patch the kernel MM code | |
1237 | * to deal with more memoryless-node error conditions. | |
1238 | */ | |
1239 | new_nid = first_online_node; | |
1240 | #endif | |
1241 | } | |
1242 | ||
e67e02a5 MB |
1243 | pr_debug("%s:%d cpu %d nid %d\n", __FUNCTION__, __LINE__, |
1244 | cpu, new_nid); | |
ea05ba7c MB |
1245 | return new_nid; |
1246 | } | |
1247 | ||
30c05350 NF |
1248 | /* |
1249 | * Update the CPU maps and sysfs entries for a single CPU when its NUMA | |
1250 | * characteristics change. This function doesn't perform any locking and is | |
1251 | * only safe to call from stop_machine(). | |
1252 | */ | |
1253 | static int update_cpu_topology(void *data) | |
1254 | { | |
1255 | struct topology_update_data *update; | |
1256 | unsigned long cpu; | |
1257 | ||
1258 | if (!data) | |
1259 | return -EINVAL; | |
1260 | ||
3be7db6a | 1261 | cpu = smp_processor_id(); |
30c05350 NF |
1262 | |
1263 | for (update = data; update; update = update->next) { | |
2c0a33f9 | 1264 | int new_nid = update->new_nid; |
30c05350 NF |
1265 | if (cpu != update->cpu) |
1266 | continue; | |
1267 | ||
49f8d8c0 | 1268 | unmap_cpu_from_node(cpu); |
2c0a33f9 NA |
1269 | map_cpu_to_node(cpu, new_nid); |
1270 | set_cpu_numa_node(cpu, new_nid); | |
1271 | set_cpu_numa_mem(cpu, local_memory_node(new_nid)); | |
176bbf14 | 1272 | vdso_getcpu_init(); |
30c05350 NF |
1273 | } |
1274 | ||
1275 | return 0; | |
1276 | } | |
1277 | ||
d4edc5b6 SB |
1278 | static int update_lookup_table(void *data) |
1279 | { | |
1280 | struct topology_update_data *update; | |
1281 | ||
1282 | if (!data) | |
1283 | return -EINVAL; | |
1284 | ||
1285 | /* | |
1286 | * Upon topology update, the numa-cpu lookup table needs to be updated | |
1287 | * for all threads in the core, including offline CPUs, to ensure that | |
1288 | * future hotplug operations respect the cpu-to-node associativity | |
1289 | * properly. | |
1290 | */ | |
1291 | for (update = data; update; update = update->next) { | |
1292 | int nid, base, j; | |
1293 | ||
1294 | nid = update->new_nid; | |
1295 | base = cpu_first_thread_sibling(update->cpu); | |
1296 | ||
1297 | for (j = 0; j < threads_per_core; j++) { | |
1298 | update_numa_cpu_lookup_table(base + j, nid); | |
1299 | } | |
1300 | } | |
1301 | ||
1302 | return 0; | |
1303 | } | |
1304 | ||
9eff1a38 JL |
1305 | /* |
1306 | * Update the node maps and sysfs entries for each cpu whose home node | |
79c5fceb | 1307 | * has changed. Returns 1 when the topology has changed, and 0 otherwise. |
3e401f7a TJB |
1308 | * |
1309 | * cpus_locked says whether we already hold cpu_hotplug_lock. | |
9eff1a38 | 1310 | */ |
3e401f7a | 1311 | int numa_update_cpu_topology(bool cpus_locked) |
9eff1a38 | 1312 | { |
3be7db6a | 1313 | unsigned int cpu, sibling, changed = 0; |
30c05350 | 1314 | struct topology_update_data *updates, *ud; |
176bbf14 | 1315 | cpumask_t updated_cpus; |
8a25a2fd | 1316 | struct device *dev; |
3be7db6a | 1317 | int weight, new_nid, i = 0; |
9eff1a38 | 1318 | |
2ea62630 | 1319 | if (!prrn_enabled && !vphn_enabled && topology_inited) |
2d73bae1 NA |
1320 | return 0; |
1321 | ||
30c05350 NF |
1322 | weight = cpumask_weight(&cpu_associativity_changes_mask); |
1323 | if (!weight) | |
1324 | return 0; | |
1325 | ||
6396bb22 | 1326 | updates = kcalloc(weight, sizeof(*updates), GFP_KERNEL); |
30c05350 NF |
1327 | if (!updates) |
1328 | return 0; | |
9eff1a38 | 1329 | |
176bbf14 JL |
1330 | cpumask_clear(&updated_cpus); |
1331 | ||
5d88aa85 | 1332 | for_each_cpu(cpu, &cpu_associativity_changes_mask) { |
3be7db6a RJ |
1333 | /* |
1334 | * If siblings aren't flagged for changes, updates list | |
1335 | * will be too short. Skip on this update and set for next | |
1336 | * update. | |
1337 | */ | |
1338 | if (!cpumask_subset(cpu_sibling_mask(cpu), | |
1339 | &cpu_associativity_changes_mask)) { | |
1340 | pr_info("Sibling bits not set for associativity " | |
1341 | "change, cpu%d\n", cpu); | |
1342 | cpumask_or(&cpu_associativity_changes_mask, | |
1343 | &cpu_associativity_changes_mask, | |
1344 | cpu_sibling_mask(cpu)); | |
1345 | cpu = cpu_last_thread_sibling(cpu); | |
1346 | continue; | |
1347 | } | |
9eff1a38 | 1348 | |
ea05ba7c | 1349 | new_nid = find_and_online_cpu_nid(cpu); |
3be7db6a RJ |
1350 | |
1351 | if (new_nid == numa_cpu_lookup_table[cpu]) { | |
1352 | cpumask_andnot(&cpu_associativity_changes_mask, | |
1353 | &cpu_associativity_changes_mask, | |
1354 | cpu_sibling_mask(cpu)); | |
17f444c0 MB |
1355 | dbg("Assoc chg gives same node %d for cpu%d\n", |
1356 | new_nid, cpu); | |
3be7db6a RJ |
1357 | cpu = cpu_last_thread_sibling(cpu); |
1358 | continue; | |
1359 | } | |
9eff1a38 | 1360 | |
3be7db6a RJ |
1361 | for_each_cpu(sibling, cpu_sibling_mask(cpu)) { |
1362 | ud = &updates[i++]; | |
8bc93149 | 1363 | ud->next = &updates[i]; |
3be7db6a RJ |
1364 | ud->cpu = sibling; |
1365 | ud->new_nid = new_nid; | |
1366 | ud->old_nid = numa_cpu_lookup_table[sibling]; | |
1367 | cpumask_set_cpu(sibling, &updated_cpus); | |
3be7db6a RJ |
1368 | } |
1369 | cpu = cpu_last_thread_sibling(cpu); | |
30c05350 NF |
1370 | } |
1371 | ||
8bc93149 MB |
1372 | /* |
1373 | * Prevent processing of 'updates' from overflowing array | |
1374 | * where last entry filled in a 'next' pointer. | |
1375 | */ | |
1376 | if (i) | |
1377 | updates[i-1].next = NULL; | |
1378 | ||
2d73bae1 NA |
1379 | pr_debug("Topology update for the following CPUs:\n"); |
1380 | if (cpumask_weight(&updated_cpus)) { | |
1381 | for (ud = &updates[0]; ud; ud = ud->next) { | |
1382 | pr_debug("cpu %d moving from node %d " | |
1383 | "to %d\n", ud->cpu, | |
1384 | ud->old_nid, ud->new_nid); | |
1385 | } | |
1386 | } | |
1387 | ||
9a013361 MW |
1388 | /* |
1389 | * In cases where we have nothing to update (because the updates list | |
1390 | * is too short or because the new topology is same as the old one), | |
1391 | * skip invoking update_cpu_topology() via stop-machine(). This is | |
1392 | * necessary (and not just a fast-path optimization) since stop-machine | |
1393 | * can end up electing a random CPU to run update_cpu_topology(), and | |
1394 | * thus trick us into setting up incorrect cpu-node mappings (since | |
1395 | * 'updates' is kzalloc()'ed). | |
1396 | * | |
1397 | * And for the similar reason, we will skip all the following updating. | |
1398 | */ | |
1399 | if (!cpumask_weight(&updated_cpus)) | |
1400 | goto out; | |
1401 | ||
3e401f7a TJB |
1402 | if (cpus_locked) |
1403 | stop_machine_cpuslocked(update_cpu_topology, &updates[0], | |
1404 | &updated_cpus); | |
1405 | else | |
1406 | stop_machine(update_cpu_topology, &updates[0], &updated_cpus); | |
30c05350 | 1407 | |
d4edc5b6 SB |
1408 | /* |
1409 | * Update the numa-cpu lookup table with the new mappings, even for | |
1410 | * offline CPUs. It is best to perform this update from the stop- | |
1411 | * machine context. | |
1412 | */ | |
3e401f7a TJB |
1413 | if (cpus_locked) |
1414 | stop_machine_cpuslocked(update_lookup_table, &updates[0], | |
d4edc5b6 | 1415 | cpumask_of(raw_smp_processor_id())); |
3e401f7a TJB |
1416 | else |
1417 | stop_machine(update_lookup_table, &updates[0], | |
1418 | cpumask_of(raw_smp_processor_id())); | |
d4edc5b6 | 1419 | |
30c05350 | 1420 | for (ud = &updates[0]; ud; ud = ud->next) { |
dd023217 NF |
1421 | unregister_cpu_under_node(ud->cpu, ud->old_nid); |
1422 | register_cpu_under_node(ud->cpu, ud->new_nid); | |
1423 | ||
30c05350 | 1424 | dev = get_cpu_device(ud->cpu); |
8a25a2fd KS |
1425 | if (dev) |
1426 | kobject_uevent(&dev->kobj, KOBJ_CHANGE); | |
30c05350 | 1427 | cpumask_clear_cpu(ud->cpu, &cpu_associativity_changes_mask); |
79c5fceb | 1428 | changed = 1; |
9eff1a38 JL |
1429 | } |
1430 | ||
9a013361 | 1431 | out: |
30c05350 | 1432 | kfree(updates); |
79c5fceb | 1433 | return changed; |
9eff1a38 JL |
1434 | } |
1435 | ||
3e401f7a TJB |
1436 | int arch_update_cpu_topology(void) |
1437 | { | |
3e401f7a TJB |
1438 | return numa_update_cpu_topology(true); |
1439 | } | |
1440 | ||
9eff1a38 JL |
1441 | static void topology_work_fn(struct work_struct *work) |
1442 | { | |
1443 | rebuild_sched_domains(); | |
1444 | } | |
1445 | static DECLARE_WORK(topology_work, topology_work_fn); | |
1446 | ||
ec32dd66 | 1447 | static void topology_schedule_update(void) |
9eff1a38 JL |
1448 | { |
1449 | schedule_work(&topology_work); | |
1450 | } | |
1451 | ||
df7e828c | 1452 | static void topology_timer_fn(struct timer_list *unused) |
9eff1a38 | 1453 | { |
5d88aa85 | 1454 | if (prrn_enabled && cpumask_weight(&cpu_associativity_changes_mask)) |
9eff1a38 | 1455 | topology_schedule_update(); |
5d88aa85 JL |
1456 | else if (vphn_enabled) { |
1457 | if (update_cpu_associativity_changes_mask() > 0) | |
1458 | topology_schedule_update(); | |
1459 | reset_topology_timer(); | |
1460 | } | |
9eff1a38 | 1461 | } |
df7e828c | 1462 | static struct timer_list topology_timer; |
9eff1a38 | 1463 | |
5d88aa85 | 1464 | static void reset_topology_timer(void) |
9eff1a38 | 1465 | { |
8604895a MB |
1466 | if (vphn_enabled) |
1467 | mod_timer(&topology_timer, jiffies + topology_timer_secs * HZ); | |
9eff1a38 JL |
1468 | } |
1469 | ||
601abdc3 NF |
1470 | #ifdef CONFIG_SMP |
1471 | ||
5d88aa85 JL |
1472 | static int dt_update_callback(struct notifier_block *nb, |
1473 | unsigned long action, void *data) | |
1474 | { | |
f5242e5a | 1475 | struct of_reconfig_data *update = data; |
5d88aa85 JL |
1476 | int rc = NOTIFY_DONE; |
1477 | ||
1478 | switch (action) { | |
5d88aa85 | 1479 | case OF_RECONFIG_UPDATE_PROPERTY: |
e5480bdc | 1480 | if (of_node_is_type(update->dn, "cpu") && |
30c05350 | 1481 | !of_prop_cmp(update->prop->name, "ibm,associativity")) { |
5d88aa85 JL |
1482 | u32 core_id; |
1483 | of_property_read_u32(update->dn, "reg", &core_id); | |
81b61324 | 1484 | rc = dlpar_cpu_readd(core_id); |
5d88aa85 JL |
1485 | rc = NOTIFY_OK; |
1486 | } | |
1487 | break; | |
1488 | } | |
1489 | ||
1490 | return rc; | |
9eff1a38 JL |
1491 | } |
1492 | ||
5d88aa85 JL |
1493 | static struct notifier_block dt_update_nb = { |
1494 | .notifier_call = dt_update_callback, | |
1495 | }; | |
1496 | ||
601abdc3 NF |
1497 | #endif |
1498 | ||
9eff1a38 | 1499 | /* |
5d88aa85 | 1500 | * Start polling for associativity changes. |
9eff1a38 JL |
1501 | */ |
1502 | int start_topology_update(void) | |
1503 | { | |
1504 | int rc = 0; | |
1505 | ||
2d4d9b30 NL |
1506 | if (!topology_updates_enabled) |
1507 | return 0; | |
1508 | ||
5d88aa85 JL |
1509 | if (firmware_has_feature(FW_FEATURE_PRRN)) { |
1510 | if (!prrn_enabled) { | |
1511 | prrn_enabled = 1; | |
601abdc3 | 1512 | #ifdef CONFIG_SMP |
5d88aa85 | 1513 | rc = of_reconfig_notifier_register(&dt_update_nb); |
601abdc3 | 1514 | #endif |
5d88aa85 | 1515 | } |
a3496e91 MB |
1516 | } |
1517 | if (firmware_has_feature(FW_FEATURE_VPHN) && | |
f13c13a0 | 1518 | lppaca_shared_proc(get_lppaca())) { |
5d88aa85 | 1519 | if (!vphn_enabled) { |
5d88aa85 JL |
1520 | vphn_enabled = 1; |
1521 | setup_cpu_associativity_change_counters(); | |
df7e828c KC |
1522 | timer_setup(&topology_timer, topology_timer_fn, |
1523 | TIMER_DEFERRABLE); | |
5d88aa85 JL |
1524 | reset_topology_timer(); |
1525 | } | |
9eff1a38 JL |
1526 | } |
1527 | ||
65b9fdad MB |
1528 | pr_info("Starting topology update%s%s\n", |
1529 | (prrn_enabled ? " prrn_enabled" : ""), | |
1530 | (vphn_enabled ? " vphn_enabled" : "")); | |
1531 | ||
9eff1a38 JL |
1532 | return rc; |
1533 | } | |
9eff1a38 JL |
1534 | |
1535 | /* | |
1536 | * Disable polling for VPHN associativity changes. | |
1537 | */ | |
1538 | int stop_topology_update(void) | |
1539 | { | |
5d88aa85 JL |
1540 | int rc = 0; |
1541 | ||
2d4d9b30 NL |
1542 | if (!topology_updates_enabled) |
1543 | return 0; | |
1544 | ||
5d88aa85 JL |
1545 | if (prrn_enabled) { |
1546 | prrn_enabled = 0; | |
601abdc3 | 1547 | #ifdef CONFIG_SMP |
5d88aa85 | 1548 | rc = of_reconfig_notifier_unregister(&dt_update_nb); |
601abdc3 | 1549 | #endif |
a3496e91 MB |
1550 | } |
1551 | if (vphn_enabled) { | |
5d88aa85 JL |
1552 | vphn_enabled = 0; |
1553 | rc = del_timer_sync(&topology_timer); | |
1554 | } | |
1555 | ||
65b9fdad MB |
1556 | pr_info("Stopping topology update\n"); |
1557 | ||
5d88aa85 | 1558 | return rc; |
9eff1a38 | 1559 | } |
e04fa612 NF |
1560 | |
1561 | int prrn_is_enabled(void) | |
1562 | { | |
1563 | return prrn_enabled; | |
1564 | } | |
1565 | ||
2ea62630 SD |
1566 | void __init shared_proc_topology_init(void) |
1567 | { | |
1568 | if (lppaca_shared_proc(get_lppaca())) { | |
1569 | bitmap_fill(cpumask_bits(&cpu_associativity_changes_mask), | |
1570 | nr_cpumask_bits); | |
1571 | numa_update_cpu_topology(false); | |
1572 | } | |
1573 | } | |
1574 | ||
e04fa612 NF |
1575 | static int topology_read(struct seq_file *file, void *v) |
1576 | { | |
1577 | if (vphn_enabled || prrn_enabled) | |
1578 | seq_puts(file, "on\n"); | |
1579 | else | |
1580 | seq_puts(file, "off\n"); | |
1581 | ||
1582 | return 0; | |
1583 | } | |
1584 | ||
1585 | static int topology_open(struct inode *inode, struct file *file) | |
1586 | { | |
1587 | return single_open(file, topology_read, NULL); | |
1588 | } | |
1589 | ||
1590 | static ssize_t topology_write(struct file *file, const char __user *buf, | |
1591 | size_t count, loff_t *off) | |
1592 | { | |
1593 | char kbuf[4]; /* "on" or "off" plus null. */ | |
1594 | int read_len; | |
1595 | ||
1596 | read_len = count < 3 ? count : 3; | |
1597 | if (copy_from_user(kbuf, buf, read_len)) | |
1598 | return -EINVAL; | |
1599 | ||
1600 | kbuf[read_len] = '\0'; | |
1601 | ||
2d4d9b30 NL |
1602 | if (!strncmp(kbuf, "on", 2)) { |
1603 | topology_updates_enabled = true; | |
e04fa612 | 1604 | start_topology_update(); |
2d4d9b30 | 1605 | } else if (!strncmp(kbuf, "off", 3)) { |
e04fa612 | 1606 | stop_topology_update(); |
2d4d9b30 NL |
1607 | topology_updates_enabled = false; |
1608 | } else | |
e04fa612 NF |
1609 | return -EINVAL; |
1610 | ||
1611 | return count; | |
1612 | } | |
1613 | ||
1614 | static const struct file_operations topology_ops = { | |
1615 | .read = seq_read, | |
1616 | .write = topology_write, | |
1617 | .open = topology_open, | |
1618 | .release = single_release | |
1619 | }; | |
1620 | ||
1621 | static int topology_update_init(void) | |
1622 | { | |
2d4d9b30 | 1623 | start_topology_update(); |
2d73bae1 | 1624 | |
17f444c0 MB |
1625 | if (vphn_enabled) |
1626 | topology_schedule_update(); | |
1627 | ||
2d15b9b4 NA |
1628 | if (!proc_create("powerpc/topology_updates", 0644, NULL, &topology_ops)) |
1629 | return -ENOMEM; | |
e04fa612 | 1630 | |
17f444c0 | 1631 | topology_inited = 1; |
e04fa612 | 1632 | return 0; |
9eff1a38 | 1633 | } |
e04fa612 | 1634 | device_initcall(topology_update_init); |
39bf990e | 1635 | #endif /* CONFIG_PPC_SPLPAR */ |