]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * pSeries NUMA support | |
3 | * | |
4 | * Copyright (C) 2002 Anton Blanchard <anton@au.ibm.com>, IBM | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or | |
7 | * modify it under the terms of the GNU General Public License | |
8 | * as published by the Free Software Foundation; either version | |
9 | * 2 of the License, or (at your option) any later version. | |
10 | */ | |
2d73bae1 NA |
11 | #define pr_fmt(fmt) "numa: " fmt |
12 | ||
1da177e4 LT |
13 | #include <linux/threads.h> |
14 | #include <linux/bootmem.h> | |
15 | #include <linux/init.h> | |
16 | #include <linux/mm.h> | |
17 | #include <linux/mmzone.h> | |
4b16f8e2 | 18 | #include <linux/export.h> |
1da177e4 LT |
19 | #include <linux/nodemask.h> |
20 | #include <linux/cpu.h> | |
21 | #include <linux/notifier.h> | |
95f72d1e | 22 | #include <linux/memblock.h> |
6df1646e | 23 | #include <linux/of.h> |
06eccea6 | 24 | #include <linux/pfn.h> |
9eff1a38 JL |
25 | #include <linux/cpuset.h> |
26 | #include <linux/node.h> | |
30c05350 | 27 | #include <linux/stop_machine.h> |
e04fa612 NF |
28 | #include <linux/proc_fs.h> |
29 | #include <linux/seq_file.h> | |
30 | #include <linux/uaccess.h> | |
191a7120 | 31 | #include <linux/slab.h> |
3be7db6a | 32 | #include <asm/cputhreads.h> |
45fb6cea | 33 | #include <asm/sparsemem.h> |
d9b2b2a2 | 34 | #include <asm/prom.h> |
2249ca9d | 35 | #include <asm/smp.h> |
d4edc5b6 SB |
36 | #include <asm/cputhreads.h> |
37 | #include <asm/topology.h> | |
9eff1a38 JL |
38 | #include <asm/firmware.h> |
39 | #include <asm/paca.h> | |
39bf990e | 40 | #include <asm/hvcall.h> |
ae3a197e | 41 | #include <asm/setup.h> |
176bbf14 | 42 | #include <asm/vdso.h> |
1da177e4 LT |
43 | |
44 | static int numa_enabled = 1; | |
45 | ||
1daa6d08 BS |
46 | static char *cmdline __initdata; |
47 | ||
1da177e4 LT |
48 | static int numa_debug; |
49 | #define dbg(args...) if (numa_debug) { printk(KERN_INFO args); } | |
50 | ||
45fb6cea | 51 | int numa_cpu_lookup_table[NR_CPUS]; |
25863de0 | 52 | cpumask_var_t node_to_cpumask_map[MAX_NUMNODES]; |
1da177e4 | 53 | struct pglist_data *node_data[MAX_NUMNODES]; |
45fb6cea AB |
54 | |
55 | EXPORT_SYMBOL(numa_cpu_lookup_table); | |
25863de0 | 56 | EXPORT_SYMBOL(node_to_cpumask_map); |
45fb6cea AB |
57 | EXPORT_SYMBOL(node_data); |
58 | ||
1da177e4 | 59 | static int min_common_depth; |
237a0989 | 60 | static int n_mem_addr_cells, n_mem_size_cells; |
41eab6f8 AB |
61 | static int form1_affinity; |
62 | ||
63 | #define MAX_DISTANCE_REF_POINTS 4 | |
64 | static int distance_ref_points_depth; | |
b08a2a12 | 65 | static const __be32 *distance_ref_points; |
41eab6f8 | 66 | static int distance_lookup_table[MAX_NUMNODES][MAX_DISTANCE_REF_POINTS]; |
1da177e4 | 67 | |
25863de0 AB |
68 | /* |
69 | * Allocate node_to_cpumask_map based on number of available nodes | |
70 | * Requires node_possible_map to be valid. | |
71 | * | |
9512938b | 72 | * Note: cpumask_of_node() is not valid until after this is done. |
25863de0 AB |
73 | */ |
74 | static void __init setup_node_to_cpumask_map(void) | |
75 | { | |
f9d531b8 | 76 | unsigned int node; |
25863de0 AB |
77 | |
78 | /* setup nr_node_ids if not done yet */ | |
f9d531b8 CS |
79 | if (nr_node_ids == MAX_NUMNODES) |
80 | setup_nr_node_ids(); | |
25863de0 AB |
81 | |
82 | /* allocate the map */ | |
83 | for (node = 0; node < nr_node_ids; node++) | |
84 | alloc_bootmem_cpumask_var(&node_to_cpumask_map[node]); | |
85 | ||
86 | /* cpumask_of_node() will now work */ | |
87 | dbg("Node to cpumask map for %d nodes\n", nr_node_ids); | |
88 | } | |
89 | ||
55671f3c | 90 | static int __init fake_numa_create_new_node(unsigned long end_pfn, |
1daa6d08 BS |
91 | unsigned int *nid) |
92 | { | |
93 | unsigned long long mem; | |
94 | char *p = cmdline; | |
95 | static unsigned int fake_nid; | |
96 | static unsigned long long curr_boundary; | |
97 | ||
98 | /* | |
99 | * Modify node id, iff we started creating NUMA nodes | |
100 | * We want to continue from where we left of the last time | |
101 | */ | |
102 | if (fake_nid) | |
103 | *nid = fake_nid; | |
104 | /* | |
105 | * In case there are no more arguments to parse, the | |
106 | * node_id should be the same as the last fake node id | |
107 | * (we've handled this above). | |
108 | */ | |
109 | if (!p) | |
110 | return 0; | |
111 | ||
112 | mem = memparse(p, &p); | |
113 | if (!mem) | |
114 | return 0; | |
115 | ||
116 | if (mem < curr_boundary) | |
117 | return 0; | |
118 | ||
119 | curr_boundary = mem; | |
120 | ||
121 | if ((end_pfn << PAGE_SHIFT) > mem) { | |
122 | /* | |
123 | * Skip commas and spaces | |
124 | */ | |
125 | while (*p == ',' || *p == ' ' || *p == '\t') | |
126 | p++; | |
127 | ||
128 | cmdline = p; | |
129 | fake_nid++; | |
130 | *nid = fake_nid; | |
131 | dbg("created new fake_node with id %d\n", fake_nid); | |
132 | return 1; | |
133 | } | |
134 | return 0; | |
135 | } | |
136 | ||
d4edc5b6 SB |
137 | static void reset_numa_cpu_lookup_table(void) |
138 | { | |
139 | unsigned int cpu; | |
140 | ||
141 | for_each_possible_cpu(cpu) | |
142 | numa_cpu_lookup_table[cpu] = -1; | |
143 | } | |
144 | ||
145 | static void update_numa_cpu_lookup_table(unsigned int cpu, int node) | |
1da177e4 LT |
146 | { |
147 | numa_cpu_lookup_table[cpu] = node; | |
d4edc5b6 SB |
148 | } |
149 | ||
150 | static void map_cpu_to_node(int cpu, int node) | |
151 | { | |
152 | update_numa_cpu_lookup_table(cpu, node); | |
45fb6cea | 153 | |
bf4b85b0 NL |
154 | dbg("adding cpu %d to node %d\n", cpu, node); |
155 | ||
25863de0 AB |
156 | if (!(cpumask_test_cpu(cpu, node_to_cpumask_map[node]))) |
157 | cpumask_set_cpu(cpu, node_to_cpumask_map[node]); | |
1da177e4 LT |
158 | } |
159 | ||
39bf990e | 160 | #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_PPC_SPLPAR) |
1da177e4 LT |
161 | static void unmap_cpu_from_node(unsigned long cpu) |
162 | { | |
163 | int node = numa_cpu_lookup_table[cpu]; | |
164 | ||
165 | dbg("removing cpu %lu from node %d\n", cpu, node); | |
166 | ||
25863de0 | 167 | if (cpumask_test_cpu(cpu, node_to_cpumask_map[node])) { |
429f4d8d | 168 | cpumask_clear_cpu(cpu, node_to_cpumask_map[node]); |
1da177e4 LT |
169 | } else { |
170 | printk(KERN_ERR "WARNING: cpu %lu not found in node %d\n", | |
171 | cpu, node); | |
172 | } | |
173 | } | |
39bf990e | 174 | #endif /* CONFIG_HOTPLUG_CPU || CONFIG_PPC_SPLPAR */ |
1da177e4 | 175 | |
1da177e4 | 176 | /* must hold reference to node during call */ |
b08a2a12 | 177 | static const __be32 *of_get_associativity(struct device_node *dev) |
1da177e4 | 178 | { |
e2eb6392 | 179 | return of_get_property(dev, "ibm,associativity", NULL); |
1da177e4 LT |
180 | } |
181 | ||
cf00085d C |
182 | /* |
183 | * Returns the property linux,drconf-usable-memory if | |
184 | * it exists (the property exists only in kexec/kdump kernels, | |
185 | * added by kexec-tools) | |
186 | */ | |
b08a2a12 | 187 | static const __be32 *of_get_usable_memory(struct device_node *memory) |
cf00085d | 188 | { |
b08a2a12 | 189 | const __be32 *prop; |
cf00085d C |
190 | u32 len; |
191 | prop = of_get_property(memory, "linux,drconf-usable-memory", &len); | |
192 | if (!prop || len < sizeof(unsigned int)) | |
ec32dd66 | 193 | return NULL; |
cf00085d C |
194 | return prop; |
195 | } | |
196 | ||
41eab6f8 AB |
197 | int __node_distance(int a, int b) |
198 | { | |
199 | int i; | |
200 | int distance = LOCAL_DISTANCE; | |
201 | ||
202 | if (!form1_affinity) | |
7122beee | 203 | return ((a == b) ? LOCAL_DISTANCE : REMOTE_DISTANCE); |
41eab6f8 AB |
204 | |
205 | for (i = 0; i < distance_ref_points_depth; i++) { | |
206 | if (distance_lookup_table[a][i] == distance_lookup_table[b][i]) | |
207 | break; | |
208 | ||
209 | /* Double the distance for each NUMA level */ | |
210 | distance *= 2; | |
211 | } | |
212 | ||
213 | return distance; | |
214 | } | |
12c743eb | 215 | EXPORT_SYMBOL(__node_distance); |
41eab6f8 AB |
216 | |
217 | static void initialize_distance_lookup_table(int nid, | |
b08a2a12 | 218 | const __be32 *associativity) |
41eab6f8 AB |
219 | { |
220 | int i; | |
221 | ||
222 | if (!form1_affinity) | |
223 | return; | |
224 | ||
225 | for (i = 0; i < distance_ref_points_depth; i++) { | |
b08a2a12 AP |
226 | const __be32 *entry; |
227 | ||
228 | entry = &associativity[be32_to_cpu(distance_ref_points[i])]; | |
229 | distance_lookup_table[nid][i] = of_read_number(entry, 1); | |
41eab6f8 AB |
230 | } |
231 | } | |
232 | ||
482ec7c4 NL |
233 | /* Returns nid in the range [0..MAX_NUMNODES-1], or -1 if no useful numa |
234 | * info is found. | |
235 | */ | |
b08a2a12 | 236 | static int associativity_to_nid(const __be32 *associativity) |
1da177e4 | 237 | { |
482ec7c4 | 238 | int nid = -1; |
1da177e4 LT |
239 | |
240 | if (min_common_depth == -1) | |
482ec7c4 | 241 | goto out; |
1da177e4 | 242 | |
b08a2a12 AP |
243 | if (of_read_number(associativity, 1) >= min_common_depth) |
244 | nid = of_read_number(&associativity[min_common_depth], 1); | |
bc16a759 NL |
245 | |
246 | /* POWER4 LPAR uses 0xffff as invalid node */ | |
482ec7c4 NL |
247 | if (nid == 0xffff || nid >= MAX_NUMNODES) |
248 | nid = -1; | |
41eab6f8 | 249 | |
b08a2a12 AP |
250 | if (nid > 0 && |
251 | of_read_number(associativity, 1) >= distance_ref_points_depth) | |
9eff1a38 | 252 | initialize_distance_lookup_table(nid, associativity); |
41eab6f8 | 253 | |
482ec7c4 | 254 | out: |
cf950b7a | 255 | return nid; |
1da177e4 LT |
256 | } |
257 | ||
9eff1a38 JL |
258 | /* Returns the nid associated with the given device tree node, |
259 | * or -1 if not found. | |
260 | */ | |
261 | static int of_node_to_nid_single(struct device_node *device) | |
262 | { | |
263 | int nid = -1; | |
b08a2a12 | 264 | const __be32 *tmp; |
9eff1a38 JL |
265 | |
266 | tmp = of_get_associativity(device); | |
267 | if (tmp) | |
268 | nid = associativity_to_nid(tmp); | |
269 | return nid; | |
270 | } | |
271 | ||
953039c8 JK |
272 | /* Walk the device tree upwards, looking for an associativity id */ |
273 | int of_node_to_nid(struct device_node *device) | |
274 | { | |
275 | struct device_node *tmp; | |
276 | int nid = -1; | |
277 | ||
278 | of_node_get(device); | |
279 | while (device) { | |
280 | nid = of_node_to_nid_single(device); | |
281 | if (nid != -1) | |
282 | break; | |
283 | ||
284 | tmp = device; | |
285 | device = of_get_parent(tmp); | |
286 | of_node_put(tmp); | |
287 | } | |
288 | of_node_put(device); | |
289 | ||
290 | return nid; | |
291 | } | |
292 | EXPORT_SYMBOL_GPL(of_node_to_nid); | |
293 | ||
1da177e4 LT |
294 | static int __init find_min_common_depth(void) |
295 | { | |
41eab6f8 | 296 | int depth; |
e70606eb | 297 | struct device_node *root; |
1da177e4 | 298 | |
1c8ee733 DS |
299 | if (firmware_has_feature(FW_FEATURE_OPAL)) |
300 | root = of_find_node_by_path("/ibm,opal"); | |
301 | else | |
302 | root = of_find_node_by_path("/rtas"); | |
e70606eb ME |
303 | if (!root) |
304 | root = of_find_node_by_path("/"); | |
1da177e4 LT |
305 | |
306 | /* | |
41eab6f8 AB |
307 | * This property is a set of 32-bit integers, each representing |
308 | * an index into the ibm,associativity nodes. | |
309 | * | |
310 | * With form 0 affinity the first integer is for an SMP configuration | |
311 | * (should be all 0's) and the second is for a normal NUMA | |
312 | * configuration. We have only one level of NUMA. | |
313 | * | |
314 | * With form 1 affinity the first integer is the most significant | |
315 | * NUMA boundary and the following are progressively less significant | |
316 | * boundaries. There can be more than one level of NUMA. | |
1da177e4 | 317 | */ |
e70606eb | 318 | distance_ref_points = of_get_property(root, |
41eab6f8 AB |
319 | "ibm,associativity-reference-points", |
320 | &distance_ref_points_depth); | |
321 | ||
322 | if (!distance_ref_points) { | |
323 | dbg("NUMA: ibm,associativity-reference-points not found.\n"); | |
324 | goto err; | |
325 | } | |
326 | ||
327 | distance_ref_points_depth /= sizeof(int); | |
1da177e4 | 328 | |
8002b0c5 NF |
329 | if (firmware_has_feature(FW_FEATURE_OPAL) || |
330 | firmware_has_feature(FW_FEATURE_TYPE1_AFFINITY)) { | |
331 | dbg("Using form 1 affinity\n"); | |
1c8ee733 | 332 | form1_affinity = 1; |
4b83c330 AB |
333 | } |
334 | ||
41eab6f8 | 335 | if (form1_affinity) { |
b08a2a12 | 336 | depth = of_read_number(distance_ref_points, 1); |
1da177e4 | 337 | } else { |
41eab6f8 AB |
338 | if (distance_ref_points_depth < 2) { |
339 | printk(KERN_WARNING "NUMA: " | |
340 | "short ibm,associativity-reference-points\n"); | |
341 | goto err; | |
342 | } | |
343 | ||
b08a2a12 | 344 | depth = of_read_number(&distance_ref_points[1], 1); |
1da177e4 | 345 | } |
1da177e4 | 346 | |
41eab6f8 AB |
347 | /* |
348 | * Warn and cap if the hardware supports more than | |
349 | * MAX_DISTANCE_REF_POINTS domains. | |
350 | */ | |
351 | if (distance_ref_points_depth > MAX_DISTANCE_REF_POINTS) { | |
352 | printk(KERN_WARNING "NUMA: distance array capped at " | |
353 | "%d entries\n", MAX_DISTANCE_REF_POINTS); | |
354 | distance_ref_points_depth = MAX_DISTANCE_REF_POINTS; | |
355 | } | |
356 | ||
e70606eb | 357 | of_node_put(root); |
1da177e4 | 358 | return depth; |
41eab6f8 AB |
359 | |
360 | err: | |
e70606eb | 361 | of_node_put(root); |
41eab6f8 | 362 | return -1; |
1da177e4 LT |
363 | } |
364 | ||
84c9fdd1 | 365 | static void __init get_n_mem_cells(int *n_addr_cells, int *n_size_cells) |
1da177e4 LT |
366 | { |
367 | struct device_node *memory = NULL; | |
1da177e4 LT |
368 | |
369 | memory = of_find_node_by_type(memory, "memory"); | |
54c23310 | 370 | if (!memory) |
84c9fdd1 | 371 | panic("numa.c: No memory nodes found!"); |
54c23310 | 372 | |
a8bda5dd | 373 | *n_addr_cells = of_n_addr_cells(memory); |
9213feea | 374 | *n_size_cells = of_n_size_cells(memory); |
84c9fdd1 | 375 | of_node_put(memory); |
1da177e4 LT |
376 | } |
377 | ||
b08a2a12 | 378 | static unsigned long read_n_cells(int n, const __be32 **buf) |
1da177e4 LT |
379 | { |
380 | unsigned long result = 0; | |
381 | ||
382 | while (n--) { | |
b08a2a12 | 383 | result = (result << 32) | of_read_number(*buf, 1); |
1da177e4 LT |
384 | (*buf)++; |
385 | } | |
386 | return result; | |
387 | } | |
388 | ||
8342681d | 389 | /* |
95f72d1e | 390 | * Read the next memblock list entry from the ibm,dynamic-memory property |
8342681d NF |
391 | * and return the information in the provided of_drconf_cell structure. |
392 | */ | |
b08a2a12 | 393 | static void read_drconf_cell(struct of_drconf_cell *drmem, const __be32 **cellp) |
8342681d | 394 | { |
b08a2a12 | 395 | const __be32 *cp; |
8342681d NF |
396 | |
397 | drmem->base_addr = read_n_cells(n_mem_addr_cells, cellp); | |
398 | ||
399 | cp = *cellp; | |
b08a2a12 AP |
400 | drmem->drc_index = of_read_number(cp, 1); |
401 | drmem->reserved = of_read_number(&cp[1], 1); | |
402 | drmem->aa_index = of_read_number(&cp[2], 1); | |
403 | drmem->flags = of_read_number(&cp[3], 1); | |
8342681d NF |
404 | |
405 | *cellp = cp + 4; | |
406 | } | |
407 | ||
408 | /* | |
25985edc | 409 | * Retrieve and validate the ibm,dynamic-memory property of the device tree. |
8342681d | 410 | * |
95f72d1e YL |
411 | * The layout of the ibm,dynamic-memory property is a number N of memblock |
412 | * list entries followed by N memblock list entries. Each memblock list entry | |
25985edc | 413 | * contains information as laid out in the of_drconf_cell struct above. |
8342681d | 414 | */ |
b08a2a12 | 415 | static int of_get_drconf_memory(struct device_node *memory, const __be32 **dm) |
8342681d | 416 | { |
b08a2a12 | 417 | const __be32 *prop; |
8342681d NF |
418 | u32 len, entries; |
419 | ||
420 | prop = of_get_property(memory, "ibm,dynamic-memory", &len); | |
421 | if (!prop || len < sizeof(unsigned int)) | |
422 | return 0; | |
423 | ||
b08a2a12 | 424 | entries = of_read_number(prop++, 1); |
8342681d NF |
425 | |
426 | /* Now that we know the number of entries, revalidate the size | |
427 | * of the property read in to ensure we have everything | |
428 | */ | |
429 | if (len < (entries * (n_mem_addr_cells + 4) + 1) * sizeof(unsigned int)) | |
430 | return 0; | |
431 | ||
432 | *dm = prop; | |
433 | return entries; | |
434 | } | |
435 | ||
436 | /* | |
25985edc | 437 | * Retrieve and validate the ibm,lmb-size property for drconf memory |
8342681d NF |
438 | * from the device tree. |
439 | */ | |
3fdfd990 | 440 | static u64 of_get_lmb_size(struct device_node *memory) |
8342681d | 441 | { |
b08a2a12 | 442 | const __be32 *prop; |
8342681d NF |
443 | u32 len; |
444 | ||
3fdfd990 | 445 | prop = of_get_property(memory, "ibm,lmb-size", &len); |
8342681d NF |
446 | if (!prop || len < sizeof(unsigned int)) |
447 | return 0; | |
448 | ||
449 | return read_n_cells(n_mem_size_cells, &prop); | |
450 | } | |
451 | ||
452 | struct assoc_arrays { | |
453 | u32 n_arrays; | |
454 | u32 array_sz; | |
b08a2a12 | 455 | const __be32 *arrays; |
8342681d NF |
456 | }; |
457 | ||
458 | /* | |
25985edc | 459 | * Retrieve and validate the list of associativity arrays for drconf |
8342681d NF |
460 | * memory from the ibm,associativity-lookup-arrays property of the |
461 | * device tree.. | |
462 | * | |
463 | * The layout of the ibm,associativity-lookup-arrays property is a number N | |
464 | * indicating the number of associativity arrays, followed by a number M | |
465 | * indicating the size of each associativity array, followed by a list | |
466 | * of N associativity arrays. | |
467 | */ | |
468 | static int of_get_assoc_arrays(struct device_node *memory, | |
469 | struct assoc_arrays *aa) | |
470 | { | |
b08a2a12 | 471 | const __be32 *prop; |
8342681d NF |
472 | u32 len; |
473 | ||
474 | prop = of_get_property(memory, "ibm,associativity-lookup-arrays", &len); | |
475 | if (!prop || len < 2 * sizeof(unsigned int)) | |
476 | return -1; | |
477 | ||
b08a2a12 AP |
478 | aa->n_arrays = of_read_number(prop++, 1); |
479 | aa->array_sz = of_read_number(prop++, 1); | |
8342681d | 480 | |
42b2aa86 | 481 | /* Now that we know the number of arrays and size of each array, |
8342681d NF |
482 | * revalidate the size of the property read in. |
483 | */ | |
484 | if (len < (aa->n_arrays * aa->array_sz + 2) * sizeof(unsigned int)) | |
485 | return -1; | |
486 | ||
487 | aa->arrays = prop; | |
488 | return 0; | |
489 | } | |
490 | ||
491 | /* | |
492 | * This is like of_node_to_nid_single() for memory represented in the | |
493 | * ibm,dynamic-reconfiguration-memory node. | |
494 | */ | |
495 | static int of_drconf_to_nid_single(struct of_drconf_cell *drmem, | |
496 | struct assoc_arrays *aa) | |
497 | { | |
498 | int default_nid = 0; | |
499 | int nid = default_nid; | |
500 | int index; | |
501 | ||
502 | if (min_common_depth > 0 && min_common_depth <= aa->array_sz && | |
503 | !(drmem->flags & DRCONF_MEM_AI_INVALID) && | |
504 | drmem->aa_index < aa->n_arrays) { | |
505 | index = drmem->aa_index * aa->array_sz + min_common_depth - 1; | |
b08a2a12 | 506 | nid = of_read_number(&aa->arrays[index], 1); |
8342681d NF |
507 | |
508 | if (nid == 0xffff || nid >= MAX_NUMNODES) | |
509 | nid = default_nid; | |
510 | } | |
511 | ||
512 | return nid; | |
513 | } | |
514 | ||
1da177e4 LT |
515 | /* |
516 | * Figure out to which domain a cpu belongs and stick it there. | |
517 | * Return the id of the domain used. | |
518 | */ | |
061d19f2 | 519 | static int numa_setup_cpu(unsigned long lcpu) |
1da177e4 | 520 | { |
297cf502 | 521 | int nid = -1; |
d4edc5b6 SB |
522 | struct device_node *cpu; |
523 | ||
524 | /* | |
525 | * If a valid cpu-to-node mapping is already available, use it | |
526 | * directly instead of querying the firmware, since it represents | |
527 | * the most recent mapping notified to us by the platform (eg: VPHN). | |
528 | */ | |
529 | if ((nid = numa_cpu_lookup_table[lcpu]) >= 0) { | |
530 | map_cpu_to_node(lcpu, nid); | |
531 | return nid; | |
532 | } | |
533 | ||
534 | cpu = of_get_cpu_node(lcpu, NULL); | |
1da177e4 LT |
535 | |
536 | if (!cpu) { | |
537 | WARN_ON(1); | |
297cf502 LZ |
538 | if (cpu_present(lcpu)) |
539 | goto out_present; | |
540 | else | |
541 | goto out; | |
1da177e4 LT |
542 | } |
543 | ||
953039c8 | 544 | nid = of_node_to_nid_single(cpu); |
1da177e4 | 545 | |
297cf502 | 546 | out_present: |
482ec7c4 | 547 | if (nid < 0 || !node_online(nid)) |
72c33688 | 548 | nid = first_online_node; |
1da177e4 | 549 | |
297cf502 | 550 | map_cpu_to_node(lcpu, nid); |
1da177e4 | 551 | of_node_put(cpu); |
297cf502 | 552 | out: |
cf950b7a | 553 | return nid; |
1da177e4 LT |
554 | } |
555 | ||
68fb18aa SB |
556 | static void verify_cpu_node_mapping(int cpu, int node) |
557 | { | |
558 | int base, sibling, i; | |
559 | ||
560 | /* Verify that all the threads in the core belong to the same node */ | |
561 | base = cpu_first_thread_sibling(cpu); | |
562 | ||
563 | for (i = 0; i < threads_per_core; i++) { | |
564 | sibling = base + i; | |
565 | ||
566 | if (sibling == cpu || cpu_is_offline(sibling)) | |
567 | continue; | |
568 | ||
569 | if (cpu_to_node(sibling) != node) { | |
570 | WARN(1, "CPU thread siblings %d and %d don't belong" | |
571 | " to the same node!\n", cpu, sibling); | |
572 | break; | |
573 | } | |
574 | } | |
575 | } | |
576 | ||
061d19f2 | 577 | static int cpu_numa_callback(struct notifier_block *nfb, unsigned long action, |
1da177e4 LT |
578 | void *hcpu) |
579 | { | |
580 | unsigned long lcpu = (unsigned long)hcpu; | |
68fb18aa | 581 | int ret = NOTIFY_DONE, nid; |
1da177e4 LT |
582 | |
583 | switch (action) { | |
584 | case CPU_UP_PREPARE: | |
8bb78442 | 585 | case CPU_UP_PREPARE_FROZEN: |
68fb18aa SB |
586 | nid = numa_setup_cpu(lcpu); |
587 | verify_cpu_node_mapping((int)lcpu, nid); | |
1da177e4 LT |
588 | ret = NOTIFY_OK; |
589 | break; | |
590 | #ifdef CONFIG_HOTPLUG_CPU | |
591 | case CPU_DEAD: | |
8bb78442 | 592 | case CPU_DEAD_FROZEN: |
1da177e4 | 593 | case CPU_UP_CANCELED: |
8bb78442 | 594 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 | 595 | unmap_cpu_from_node(lcpu); |
1da177e4 | 596 | ret = NOTIFY_OK; |
b00fc6ec | 597 | break; |
1da177e4 LT |
598 | #endif |
599 | } | |
600 | return ret; | |
601 | } | |
602 | ||
603 | /* | |
604 | * Check and possibly modify a memory region to enforce the memory limit. | |
605 | * | |
606 | * Returns the size the region should have to enforce the memory limit. | |
607 | * This will either be the original value of size, a truncated value, | |
608 | * or zero. If the returned value of size is 0 the region should be | |
25985edc | 609 | * discarded as it lies wholly above the memory limit. |
1da177e4 | 610 | */ |
45fb6cea AB |
611 | static unsigned long __init numa_enforce_memory_limit(unsigned long start, |
612 | unsigned long size) | |
1da177e4 LT |
613 | { |
614 | /* | |
95f72d1e | 615 | * We use memblock_end_of_DRAM() in here instead of memory_limit because |
1da177e4 | 616 | * we've already adjusted it for the limit and it takes care of |
fe55249d MM |
617 | * having memory holes below the limit. Also, in the case of |
618 | * iommu_is_off, memory_limit is not set but is implicitly enforced. | |
1da177e4 | 619 | */ |
1da177e4 | 620 | |
95f72d1e | 621 | if (start + size <= memblock_end_of_DRAM()) |
1da177e4 LT |
622 | return size; |
623 | ||
95f72d1e | 624 | if (start >= memblock_end_of_DRAM()) |
1da177e4 LT |
625 | return 0; |
626 | ||
95f72d1e | 627 | return memblock_end_of_DRAM() - start; |
1da177e4 LT |
628 | } |
629 | ||
cf00085d C |
630 | /* |
631 | * Reads the counter for a given entry in | |
632 | * linux,drconf-usable-memory property | |
633 | */ | |
b08a2a12 | 634 | static inline int __init read_usm_ranges(const __be32 **usm) |
cf00085d C |
635 | { |
636 | /* | |
3fdfd990 | 637 | * For each lmb in ibm,dynamic-memory a corresponding |
cf00085d C |
638 | * entry in linux,drconf-usable-memory property contains |
639 | * a counter followed by that many (base, size) duple. | |
640 | * read the counter from linux,drconf-usable-memory | |
641 | */ | |
642 | return read_n_cells(n_mem_size_cells, usm); | |
643 | } | |
644 | ||
0204568a PM |
645 | /* |
646 | * Extract NUMA information from the ibm,dynamic-reconfiguration-memory | |
647 | * node. This assumes n_mem_{addr,size}_cells have been set. | |
648 | */ | |
649 | static void __init parse_drconf_memory(struct device_node *memory) | |
650 | { | |
b08a2a12 | 651 | const __be32 *uninitialized_var(dm), *usm; |
cf00085d | 652 | unsigned int n, rc, ranges, is_kexec_kdump = 0; |
3fdfd990 | 653 | unsigned long lmb_size, base, size, sz; |
8342681d | 654 | int nid; |
aa709f3b | 655 | struct assoc_arrays aa = { .arrays = NULL }; |
8342681d NF |
656 | |
657 | n = of_get_drconf_memory(memory, &dm); | |
658 | if (!n) | |
0204568a PM |
659 | return; |
660 | ||
3fdfd990 BH |
661 | lmb_size = of_get_lmb_size(memory); |
662 | if (!lmb_size) | |
8342681d NF |
663 | return; |
664 | ||
665 | rc = of_get_assoc_arrays(memory, &aa); | |
666 | if (rc) | |
0204568a PM |
667 | return; |
668 | ||
cf00085d C |
669 | /* check if this is a kexec/kdump kernel */ |
670 | usm = of_get_usable_memory(memory); | |
671 | if (usm != NULL) | |
672 | is_kexec_kdump = 1; | |
673 | ||
0204568a | 674 | for (; n != 0; --n) { |
8342681d NF |
675 | struct of_drconf_cell drmem; |
676 | ||
677 | read_drconf_cell(&drmem, &dm); | |
678 | ||
679 | /* skip this block if the reserved bit is set in flags (0x80) | |
680 | or if the block is not assigned to this partition (0x8) */ | |
681 | if ((drmem.flags & DRCONF_MEM_RESERVED) | |
682 | || !(drmem.flags & DRCONF_MEM_ASSIGNED)) | |
0204568a | 683 | continue; |
1daa6d08 | 684 | |
cf00085d | 685 | base = drmem.base_addr; |
3fdfd990 | 686 | size = lmb_size; |
cf00085d | 687 | ranges = 1; |
8342681d | 688 | |
cf00085d C |
689 | if (is_kexec_kdump) { |
690 | ranges = read_usm_ranges(&usm); | |
691 | if (!ranges) /* there are no (base, size) duple */ | |
692 | continue; | |
693 | } | |
694 | do { | |
695 | if (is_kexec_kdump) { | |
696 | base = read_n_cells(n_mem_addr_cells, &usm); | |
697 | size = read_n_cells(n_mem_size_cells, &usm); | |
698 | } | |
699 | nid = of_drconf_to_nid_single(&drmem, &aa); | |
700 | fake_numa_create_new_node( | |
701 | ((base + size) >> PAGE_SHIFT), | |
8342681d | 702 | &nid); |
cf00085d C |
703 | node_set_online(nid); |
704 | sz = numa_enforce_memory_limit(base, size); | |
705 | if (sz) | |
e7e8de59 TC |
706 | memblock_set_node(base, sz, |
707 | &memblock.memory, nid); | |
cf00085d | 708 | } while (--ranges); |
0204568a PM |
709 | } |
710 | } | |
711 | ||
1da177e4 LT |
712 | static int __init parse_numa_properties(void) |
713 | { | |
94db7c5e | 714 | struct device_node *memory; |
482ec7c4 | 715 | int default_nid = 0; |
1da177e4 LT |
716 | unsigned long i; |
717 | ||
718 | if (numa_enabled == 0) { | |
719 | printk(KERN_WARNING "NUMA disabled by user\n"); | |
720 | return -1; | |
721 | } | |
722 | ||
1da177e4 LT |
723 | min_common_depth = find_min_common_depth(); |
724 | ||
1da177e4 LT |
725 | if (min_common_depth < 0) |
726 | return min_common_depth; | |
727 | ||
bf4b85b0 NL |
728 | dbg("NUMA associativity depth for CPU/Memory: %d\n", min_common_depth); |
729 | ||
1da177e4 | 730 | /* |
482ec7c4 NL |
731 | * Even though we connect cpus to numa domains later in SMP |
732 | * init, we need to know the node ids now. This is because | |
733 | * each node to be onlined must have NODE_DATA etc backing it. | |
1da177e4 | 734 | */ |
482ec7c4 | 735 | for_each_present_cpu(i) { |
dfbe93a2 | 736 | struct device_node *cpu; |
cf950b7a | 737 | int nid; |
1da177e4 | 738 | |
8b16cd23 | 739 | cpu = of_get_cpu_node(i, NULL); |
482ec7c4 | 740 | BUG_ON(!cpu); |
953039c8 | 741 | nid = of_node_to_nid_single(cpu); |
482ec7c4 | 742 | of_node_put(cpu); |
1da177e4 | 743 | |
482ec7c4 NL |
744 | /* |
745 | * Don't fall back to default_nid yet -- we will plug | |
746 | * cpus into nodes once the memory scan has discovered | |
747 | * the topology. | |
748 | */ | |
749 | if (nid < 0) | |
750 | continue; | |
751 | node_set_online(nid); | |
1da177e4 LT |
752 | } |
753 | ||
237a0989 | 754 | get_n_mem_cells(&n_mem_addr_cells, &n_mem_size_cells); |
94db7c5e AB |
755 | |
756 | for_each_node_by_type(memory, "memory") { | |
1da177e4 LT |
757 | unsigned long start; |
758 | unsigned long size; | |
cf950b7a | 759 | int nid; |
1da177e4 | 760 | int ranges; |
b08a2a12 | 761 | const __be32 *memcell_buf; |
1da177e4 LT |
762 | unsigned int len; |
763 | ||
e2eb6392 | 764 | memcell_buf = of_get_property(memory, |
ba759485 ME |
765 | "linux,usable-memory", &len); |
766 | if (!memcell_buf || len <= 0) | |
e2eb6392 | 767 | memcell_buf = of_get_property(memory, "reg", &len); |
1da177e4 LT |
768 | if (!memcell_buf || len <= 0) |
769 | continue; | |
770 | ||
cc5d0189 BH |
771 | /* ranges in cell */ |
772 | ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells); | |
1da177e4 LT |
773 | new_range: |
774 | /* these are order-sensitive, and modify the buffer pointer */ | |
237a0989 MK |
775 | start = read_n_cells(n_mem_addr_cells, &memcell_buf); |
776 | size = read_n_cells(n_mem_size_cells, &memcell_buf); | |
1da177e4 | 777 | |
482ec7c4 NL |
778 | /* |
779 | * Assumption: either all memory nodes or none will | |
780 | * have associativity properties. If none, then | |
781 | * everything goes to default_nid. | |
782 | */ | |
953039c8 | 783 | nid = of_node_to_nid_single(memory); |
482ec7c4 NL |
784 | if (nid < 0) |
785 | nid = default_nid; | |
1daa6d08 BS |
786 | |
787 | fake_numa_create_new_node(((start + size) >> PAGE_SHIFT), &nid); | |
482ec7c4 | 788 | node_set_online(nid); |
1da177e4 | 789 | |
45fb6cea | 790 | if (!(size = numa_enforce_memory_limit(start, size))) { |
1da177e4 LT |
791 | if (--ranges) |
792 | goto new_range; | |
793 | else | |
794 | continue; | |
795 | } | |
796 | ||
e7e8de59 | 797 | memblock_set_node(start, size, &memblock.memory, nid); |
1da177e4 LT |
798 | |
799 | if (--ranges) | |
800 | goto new_range; | |
801 | } | |
802 | ||
0204568a | 803 | /* |
dfbe93a2 AB |
804 | * Now do the same thing for each MEMBLOCK listed in the |
805 | * ibm,dynamic-memory property in the | |
806 | * ibm,dynamic-reconfiguration-memory node. | |
0204568a PM |
807 | */ |
808 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); | |
809 | if (memory) | |
810 | parse_drconf_memory(memory); | |
811 | ||
1da177e4 LT |
812 | return 0; |
813 | } | |
814 | ||
815 | static void __init setup_nonnuma(void) | |
816 | { | |
95f72d1e YL |
817 | unsigned long top_of_ram = memblock_end_of_DRAM(); |
818 | unsigned long total_ram = memblock_phys_mem_size(); | |
c67c3cb4 | 819 | unsigned long start_pfn, end_pfn; |
28be7072 BH |
820 | unsigned int nid = 0; |
821 | struct memblock_region *reg; | |
1da177e4 | 822 | |
e110b281 | 823 | printk(KERN_DEBUG "Top of RAM: 0x%lx, Total RAM: 0x%lx\n", |
1da177e4 | 824 | top_of_ram, total_ram); |
e110b281 | 825 | printk(KERN_DEBUG "Memory hole size: %ldMB\n", |
1da177e4 LT |
826 | (top_of_ram - total_ram) >> 20); |
827 | ||
28be7072 | 828 | for_each_memblock(memory, reg) { |
c7fc2de0 YL |
829 | start_pfn = memblock_region_memory_base_pfn(reg); |
830 | end_pfn = memblock_region_memory_end_pfn(reg); | |
1daa6d08 BS |
831 | |
832 | fake_numa_create_new_node(end_pfn, &nid); | |
1d7cfe18 | 833 | memblock_set_node(PFN_PHYS(start_pfn), |
e7e8de59 TC |
834 | PFN_PHYS(end_pfn - start_pfn), |
835 | &memblock.memory, nid); | |
1daa6d08 | 836 | node_set_online(nid); |
c67c3cb4 | 837 | } |
1da177e4 LT |
838 | } |
839 | ||
4b703a23 AB |
840 | void __init dump_numa_cpu_topology(void) |
841 | { | |
842 | unsigned int node; | |
843 | unsigned int cpu, count; | |
844 | ||
845 | if (min_common_depth == -1 || !numa_enabled) | |
846 | return; | |
847 | ||
848 | for_each_online_node(node) { | |
e110b281 | 849 | printk(KERN_DEBUG "Node %d CPUs:", node); |
4b703a23 AB |
850 | |
851 | count = 0; | |
852 | /* | |
853 | * If we used a CPU iterator here we would miss printing | |
854 | * the holes in the cpumap. | |
855 | */ | |
25863de0 AB |
856 | for (cpu = 0; cpu < nr_cpu_ids; cpu++) { |
857 | if (cpumask_test_cpu(cpu, | |
858 | node_to_cpumask_map[node])) { | |
4b703a23 AB |
859 | if (count == 0) |
860 | printk(" %u", cpu); | |
861 | ++count; | |
862 | } else { | |
863 | if (count > 1) | |
864 | printk("-%u", cpu - 1); | |
865 | count = 0; | |
866 | } | |
867 | } | |
868 | ||
869 | if (count > 1) | |
25863de0 | 870 | printk("-%u", nr_cpu_ids - 1); |
4b703a23 AB |
871 | printk("\n"); |
872 | } | |
873 | } | |
874 | ||
875 | static void __init dump_numa_memory_topology(void) | |
1da177e4 LT |
876 | { |
877 | unsigned int node; | |
878 | unsigned int count; | |
879 | ||
880 | if (min_common_depth == -1 || !numa_enabled) | |
881 | return; | |
882 | ||
883 | for_each_online_node(node) { | |
884 | unsigned long i; | |
885 | ||
e110b281 | 886 | printk(KERN_DEBUG "Node %d Memory:", node); |
1da177e4 LT |
887 | |
888 | count = 0; | |
889 | ||
95f72d1e | 890 | for (i = 0; i < memblock_end_of_DRAM(); |
45fb6cea AB |
891 | i += (1 << SECTION_SIZE_BITS)) { |
892 | if (early_pfn_to_nid(i >> PAGE_SHIFT) == node) { | |
1da177e4 LT |
893 | if (count == 0) |
894 | printk(" 0x%lx", i); | |
895 | ++count; | |
896 | } else { | |
897 | if (count > 0) | |
898 | printk("-0x%lx", i); | |
899 | count = 0; | |
900 | } | |
901 | } | |
902 | ||
903 | if (count > 0) | |
904 | printk("-0x%lx", i); | |
905 | printk("\n"); | |
906 | } | |
1da177e4 LT |
907 | } |
908 | ||
061d19f2 | 909 | static struct notifier_block ppc64_numa_nb = { |
74b85f37 CS |
910 | .notifier_call = cpu_numa_callback, |
911 | .priority = 1 /* Must run before sched domains notifier. */ | |
912 | }; | |
913 | ||
10239733 AB |
914 | /* Initialize NODE_DATA for a node on the local memory */ |
915 | static void __init setup_node_data(int nid, u64 start_pfn, u64 end_pfn) | |
4a618669 | 916 | { |
10239733 AB |
917 | u64 spanned_pages = end_pfn - start_pfn; |
918 | const size_t nd_size = roundup(sizeof(pg_data_t), SMP_CACHE_BYTES); | |
919 | u64 nd_pa; | |
920 | void *nd; | |
921 | int tnid; | |
4a618669 | 922 | |
10239733 AB |
923 | if (spanned_pages) |
924 | pr_info("Initmem setup node %d [mem %#010Lx-%#010Lx]\n", | |
925 | nid, start_pfn << PAGE_SHIFT, | |
926 | (end_pfn << PAGE_SHIFT) - 1); | |
927 | else | |
928 | pr_info("Initmem setup node %d\n", nid); | |
4a618669 | 929 | |
10239733 AB |
930 | nd_pa = memblock_alloc_try_nid(nd_size, SMP_CACHE_BYTES, nid); |
931 | nd = __va(nd_pa); | |
4a618669 | 932 | |
10239733 AB |
933 | /* report and initialize */ |
934 | pr_info(" NODE_DATA [mem %#010Lx-%#010Lx]\n", | |
935 | nd_pa, nd_pa + nd_size - 1); | |
936 | tnid = early_pfn_to_nid(nd_pa >> PAGE_SHIFT); | |
937 | if (tnid != nid) | |
938 | pr_info(" NODE_DATA(%d) on node %d\n", nid, tnid); | |
4a618669 | 939 | |
10239733 AB |
940 | node_data[nid] = nd; |
941 | memset(NODE_DATA(nid), 0, sizeof(pg_data_t)); | |
942 | NODE_DATA(nid)->node_id = nid; | |
943 | NODE_DATA(nid)->node_start_pfn = start_pfn; | |
944 | NODE_DATA(nid)->node_spanned_pages = spanned_pages; | |
945 | } | |
4a618669 | 946 | |
10239733 | 947 | void __init initmem_init(void) |
1da177e4 | 948 | { |
2fabf084 | 949 | int nid, cpu; |
1da177e4 | 950 | |
95f72d1e | 951 | max_low_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT; |
1da177e4 LT |
952 | max_pfn = max_low_pfn; |
953 | ||
954 | if (parse_numa_properties()) | |
955 | setup_nonnuma(); | |
956 | else | |
4b703a23 | 957 | dump_numa_memory_topology(); |
1da177e4 | 958 | |
10239733 AB |
959 | memblock_dump_all(); |
960 | ||
1da177e4 | 961 | for_each_online_node(nid) { |
c67c3cb4 | 962 | unsigned long start_pfn, end_pfn; |
1da177e4 | 963 | |
c67c3cb4 | 964 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); |
10239733 | 965 | setup_node_data(nid, start_pfn, end_pfn); |
8f64e1f2 | 966 | sparse_memory_present_with_active_regions(nid); |
4a618669 | 967 | } |
d3f6204a | 968 | |
25863de0 AB |
969 | setup_node_to_cpumask_map(); |
970 | ||
d4edc5b6 | 971 | reset_numa_cpu_lookup_table(); |
25863de0 | 972 | register_cpu_notifier(&ppc64_numa_nb); |
2fabf084 NA |
973 | /* |
974 | * We need the numa_cpu_lookup_table to be accurate for all CPUs, | |
975 | * even before we online them, so that we can use cpu_to_{node,mem} | |
976 | * early in boot, cf. smp_prepare_cpus(). | |
977 | */ | |
bc3c4327 | 978 | for_each_present_cpu(cpu) { |
70ad2375 | 979 | numa_setup_cpu((unsigned long)cpu); |
2fabf084 | 980 | } |
1da177e4 LT |
981 | } |
982 | ||
1da177e4 LT |
983 | static int __init early_numa(char *p) |
984 | { | |
985 | if (!p) | |
986 | return 0; | |
987 | ||
988 | if (strstr(p, "off")) | |
989 | numa_enabled = 0; | |
990 | ||
991 | if (strstr(p, "debug")) | |
992 | numa_debug = 1; | |
993 | ||
1daa6d08 BS |
994 | p = strstr(p, "fake="); |
995 | if (p) | |
996 | cmdline = p + strlen("fake="); | |
997 | ||
1da177e4 LT |
998 | return 0; |
999 | } | |
1000 | early_param("numa", early_numa); | |
237a0989 | 1001 | |
2d73bae1 NA |
1002 | static bool topology_updates_enabled = true; |
1003 | ||
1004 | static int __init early_topology_updates(char *p) | |
1005 | { | |
1006 | if (!p) | |
1007 | return 0; | |
1008 | ||
1009 | if (!strcmp(p, "off")) { | |
1010 | pr_info("Disabling topology updates\n"); | |
1011 | topology_updates_enabled = false; | |
1012 | } | |
1013 | ||
1014 | return 0; | |
1015 | } | |
1016 | early_param("topology_updates", early_topology_updates); | |
1017 | ||
237a0989 | 1018 | #ifdef CONFIG_MEMORY_HOTPLUG |
0db9360a | 1019 | /* |
0f16ef7f NF |
1020 | * Find the node associated with a hot added memory section for |
1021 | * memory represented in the device tree by the property | |
1022 | * ibm,dynamic-reconfiguration-memory/ibm,dynamic-memory. | |
0db9360a NF |
1023 | */ |
1024 | static int hot_add_drconf_scn_to_nid(struct device_node *memory, | |
1025 | unsigned long scn_addr) | |
1026 | { | |
b08a2a12 | 1027 | const __be32 *dm; |
0f16ef7f | 1028 | unsigned int drconf_cell_cnt, rc; |
3fdfd990 | 1029 | unsigned long lmb_size; |
0db9360a | 1030 | struct assoc_arrays aa; |
0f16ef7f | 1031 | int nid = -1; |
0db9360a | 1032 | |
0f16ef7f NF |
1033 | drconf_cell_cnt = of_get_drconf_memory(memory, &dm); |
1034 | if (!drconf_cell_cnt) | |
1035 | return -1; | |
0db9360a | 1036 | |
3fdfd990 BH |
1037 | lmb_size = of_get_lmb_size(memory); |
1038 | if (!lmb_size) | |
0f16ef7f | 1039 | return -1; |
0db9360a NF |
1040 | |
1041 | rc = of_get_assoc_arrays(memory, &aa); | |
1042 | if (rc) | |
0f16ef7f | 1043 | return -1; |
0db9360a | 1044 | |
0f16ef7f | 1045 | for (; drconf_cell_cnt != 0; --drconf_cell_cnt) { |
0db9360a NF |
1046 | struct of_drconf_cell drmem; |
1047 | ||
1048 | read_drconf_cell(&drmem, &dm); | |
1049 | ||
1050 | /* skip this block if it is reserved or not assigned to | |
1051 | * this partition */ | |
1052 | if ((drmem.flags & DRCONF_MEM_RESERVED) | |
1053 | || !(drmem.flags & DRCONF_MEM_ASSIGNED)) | |
1054 | continue; | |
1055 | ||
0f16ef7f | 1056 | if ((scn_addr < drmem.base_addr) |
3fdfd990 | 1057 | || (scn_addr >= (drmem.base_addr + lmb_size))) |
0f16ef7f NF |
1058 | continue; |
1059 | ||
0db9360a | 1060 | nid = of_drconf_to_nid_single(&drmem, &aa); |
0f16ef7f NF |
1061 | break; |
1062 | } | |
1063 | ||
1064 | return nid; | |
1065 | } | |
1066 | ||
1067 | /* | |
1068 | * Find the node associated with a hot added memory section for memory | |
1069 | * represented in the device tree as a node (i.e. memory@XXXX) for | |
95f72d1e | 1070 | * each memblock. |
0f16ef7f | 1071 | */ |
ec32dd66 | 1072 | static int hot_add_node_scn_to_nid(unsigned long scn_addr) |
0f16ef7f | 1073 | { |
94db7c5e | 1074 | struct device_node *memory; |
0f16ef7f NF |
1075 | int nid = -1; |
1076 | ||
94db7c5e | 1077 | for_each_node_by_type(memory, "memory") { |
0f16ef7f NF |
1078 | unsigned long start, size; |
1079 | int ranges; | |
b08a2a12 | 1080 | const __be32 *memcell_buf; |
0f16ef7f NF |
1081 | unsigned int len; |
1082 | ||
1083 | memcell_buf = of_get_property(memory, "reg", &len); | |
1084 | if (!memcell_buf || len <= 0) | |
1085 | continue; | |
1086 | ||
1087 | /* ranges in cell */ | |
1088 | ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells); | |
1089 | ||
1090 | while (ranges--) { | |
1091 | start = read_n_cells(n_mem_addr_cells, &memcell_buf); | |
1092 | size = read_n_cells(n_mem_size_cells, &memcell_buf); | |
1093 | ||
1094 | if ((scn_addr < start) || (scn_addr >= (start + size))) | |
1095 | continue; | |
1096 | ||
1097 | nid = of_node_to_nid_single(memory); | |
1098 | break; | |
1099 | } | |
0db9360a | 1100 | |
0f16ef7f NF |
1101 | if (nid >= 0) |
1102 | break; | |
0db9360a NF |
1103 | } |
1104 | ||
60831842 AB |
1105 | of_node_put(memory); |
1106 | ||
0f16ef7f | 1107 | return nid; |
0db9360a NF |
1108 | } |
1109 | ||
237a0989 MK |
1110 | /* |
1111 | * Find the node associated with a hot added memory section. Section | |
95f72d1e YL |
1112 | * corresponds to a SPARSEMEM section, not an MEMBLOCK. It is assumed that |
1113 | * sections are fully contained within a single MEMBLOCK. | |
237a0989 MK |
1114 | */ |
1115 | int hot_add_scn_to_nid(unsigned long scn_addr) | |
1116 | { | |
1117 | struct device_node *memory = NULL; | |
0f16ef7f | 1118 | int nid, found = 0; |
237a0989 MK |
1119 | |
1120 | if (!numa_enabled || (min_common_depth < 0)) | |
72c33688 | 1121 | return first_online_node; |
0db9360a NF |
1122 | |
1123 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); | |
1124 | if (memory) { | |
1125 | nid = hot_add_drconf_scn_to_nid(memory, scn_addr); | |
1126 | of_node_put(memory); | |
0f16ef7f NF |
1127 | } else { |
1128 | nid = hot_add_node_scn_to_nid(scn_addr); | |
0db9360a | 1129 | } |
237a0989 | 1130 | |
0f16ef7f | 1131 | if (nid < 0 || !node_online(nid)) |
72c33688 | 1132 | nid = first_online_node; |
237a0989 | 1133 | |
0f16ef7f NF |
1134 | if (NODE_DATA(nid)->node_spanned_pages) |
1135 | return nid; | |
237a0989 | 1136 | |
0f16ef7f NF |
1137 | for_each_online_node(nid) { |
1138 | if (NODE_DATA(nid)->node_spanned_pages) { | |
1139 | found = 1; | |
1140 | break; | |
237a0989 | 1141 | } |
237a0989 | 1142 | } |
0f16ef7f NF |
1143 | |
1144 | BUG_ON(!found); | |
1145 | return nid; | |
237a0989 | 1146 | } |
0f16ef7f | 1147 | |
cd34206e NA |
1148 | static u64 hot_add_drconf_memory_max(void) |
1149 | { | |
1150 | struct device_node *memory = NULL; | |
1151 | unsigned int drconf_cell_cnt = 0; | |
1152 | u64 lmb_size = 0; | |
ec32dd66 | 1153 | const __be32 *dm = NULL; |
cd34206e NA |
1154 | |
1155 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); | |
1156 | if (memory) { | |
1157 | drconf_cell_cnt = of_get_drconf_memory(memory, &dm); | |
1158 | lmb_size = of_get_lmb_size(memory); | |
1159 | of_node_put(memory); | |
1160 | } | |
1161 | return lmb_size * drconf_cell_cnt; | |
1162 | } | |
1163 | ||
1164 | /* | |
1165 | * memory_hotplug_max - return max address of memory that may be added | |
1166 | * | |
1167 | * This is currently only used on systems that support drconfig memory | |
1168 | * hotplug. | |
1169 | */ | |
1170 | u64 memory_hotplug_max(void) | |
1171 | { | |
1172 | return max(hot_add_drconf_memory_max(), memblock_end_of_DRAM()); | |
1173 | } | |
237a0989 | 1174 | #endif /* CONFIG_MEMORY_HOTPLUG */ |
9eff1a38 | 1175 | |
bd03403a | 1176 | /* Virtual Processor Home Node (VPHN) support */ |
39bf990e | 1177 | #ifdef CONFIG_PPC_SPLPAR |
30c05350 NF |
1178 | struct topology_update_data { |
1179 | struct topology_update_data *next; | |
1180 | unsigned int cpu; | |
1181 | int old_nid; | |
1182 | int new_nid; | |
1183 | }; | |
1184 | ||
5de16699 | 1185 | static u8 vphn_cpu_change_counts[NR_CPUS][MAX_DISTANCE_REF_POINTS]; |
9eff1a38 JL |
1186 | static cpumask_t cpu_associativity_changes_mask; |
1187 | static int vphn_enabled; | |
5d88aa85 JL |
1188 | static int prrn_enabled; |
1189 | static void reset_topology_timer(void); | |
9eff1a38 JL |
1190 | |
1191 | /* | |
1192 | * Store the current values of the associativity change counters in the | |
1193 | * hypervisor. | |
1194 | */ | |
1195 | static void setup_cpu_associativity_change_counters(void) | |
1196 | { | |
cd9d6cc7 | 1197 | int cpu; |
9eff1a38 | 1198 | |
5de16699 AB |
1199 | /* The VPHN feature supports a maximum of 8 reference points */ |
1200 | BUILD_BUG_ON(MAX_DISTANCE_REF_POINTS > 8); | |
1201 | ||
9eff1a38 | 1202 | for_each_possible_cpu(cpu) { |
cd9d6cc7 | 1203 | int i; |
9eff1a38 JL |
1204 | u8 *counts = vphn_cpu_change_counts[cpu]; |
1205 | volatile u8 *hypervisor_counts = lppaca[cpu].vphn_assoc_counts; | |
1206 | ||
5de16699 | 1207 | for (i = 0; i < distance_ref_points_depth; i++) |
9eff1a38 | 1208 | counts[i] = hypervisor_counts[i]; |
9eff1a38 JL |
1209 | } |
1210 | } | |
1211 | ||
1212 | /* | |
1213 | * The hypervisor maintains a set of 8 associativity change counters in | |
1214 | * the VPA of each cpu that correspond to the associativity levels in the | |
1215 | * ibm,associativity-reference-points property. When an associativity | |
1216 | * level changes, the corresponding counter is incremented. | |
1217 | * | |
1218 | * Set a bit in cpu_associativity_changes_mask for each cpu whose home | |
1219 | * node associativity levels have changed. | |
1220 | * | |
1221 | * Returns the number of cpus with unhandled associativity changes. | |
1222 | */ | |
1223 | static int update_cpu_associativity_changes_mask(void) | |
1224 | { | |
5d88aa85 | 1225 | int cpu; |
9eff1a38 JL |
1226 | cpumask_t *changes = &cpu_associativity_changes_mask; |
1227 | ||
9eff1a38 JL |
1228 | for_each_possible_cpu(cpu) { |
1229 | int i, changed = 0; | |
1230 | u8 *counts = vphn_cpu_change_counts[cpu]; | |
1231 | volatile u8 *hypervisor_counts = lppaca[cpu].vphn_assoc_counts; | |
1232 | ||
5de16699 | 1233 | for (i = 0; i < distance_ref_points_depth; i++) { |
d69043e8 | 1234 | if (hypervisor_counts[i] != counts[i]) { |
9eff1a38 JL |
1235 | counts[i] = hypervisor_counts[i]; |
1236 | changed = 1; | |
1237 | } | |
1238 | } | |
1239 | if (changed) { | |
3be7db6a RJ |
1240 | cpumask_or(changes, changes, cpu_sibling_mask(cpu)); |
1241 | cpu = cpu_last_thread_sibling(cpu); | |
9eff1a38 JL |
1242 | } |
1243 | } | |
1244 | ||
5d88aa85 | 1245 | return cpumask_weight(changes); |
9eff1a38 JL |
1246 | } |
1247 | ||
c0e5e46f AB |
1248 | /* |
1249 | * 6 64-bit registers unpacked into 12 32-bit associativity values. To form | |
1250 | * the complete property we have to add the length in the first cell. | |
1251 | */ | |
1252 | #define VPHN_ASSOC_BUFSIZE (6*sizeof(u64)/sizeof(u32) + 1) | |
9eff1a38 JL |
1253 | |
1254 | /* | |
1255 | * Convert the associativity domain numbers returned from the hypervisor | |
1256 | * to the sequence they would appear in the ibm,associativity property. | |
1257 | */ | |
b08a2a12 | 1258 | static int vphn_unpack_associativity(const long *packed, __be32 *unpacked) |
9eff1a38 | 1259 | { |
cd9d6cc7 | 1260 | int i, nr_assoc_doms = 0; |
b08a2a12 | 1261 | const __be16 *field = (const __be16 *) packed; |
9eff1a38 JL |
1262 | |
1263 | #define VPHN_FIELD_UNUSED (0xffff) | |
1264 | #define VPHN_FIELD_MSB (0x8000) | |
1265 | #define VPHN_FIELD_MASK (~VPHN_FIELD_MSB) | |
1266 | ||
c0e5e46f | 1267 | for (i = 1; i < VPHN_ASSOC_BUFSIZE; i++) { |
b08a2a12 | 1268 | if (be16_to_cpup(field) == VPHN_FIELD_UNUSED) { |
9eff1a38 JL |
1269 | /* All significant fields processed, and remaining |
1270 | * fields contain the reserved value of all 1's. | |
1271 | * Just store them. | |
1272 | */ | |
b08a2a12 | 1273 | unpacked[i] = *((__be32 *)field); |
9eff1a38 | 1274 | field += 2; |
b08a2a12 | 1275 | } else if (be16_to_cpup(field) & VPHN_FIELD_MSB) { |
9eff1a38 | 1276 | /* Data is in the lower 15 bits of this field */ |
b08a2a12 AP |
1277 | unpacked[i] = cpu_to_be32( |
1278 | be16_to_cpup(field) & VPHN_FIELD_MASK); | |
9eff1a38 JL |
1279 | field++; |
1280 | nr_assoc_doms++; | |
7639adaa | 1281 | } else { |
9eff1a38 JL |
1282 | /* Data is in the lower 15 bits of this field |
1283 | * concatenated with the next 16 bit field | |
1284 | */ | |
b08a2a12 | 1285 | unpacked[i] = *((__be32 *)field); |
9eff1a38 JL |
1286 | field += 2; |
1287 | nr_assoc_doms++; | |
1288 | } | |
1289 | } | |
1290 | ||
c0e5e46f | 1291 | /* The first cell contains the length of the property */ |
b08a2a12 | 1292 | unpacked[0] = cpu_to_be32(nr_assoc_doms); |
c0e5e46f | 1293 | |
9eff1a38 JL |
1294 | return nr_assoc_doms; |
1295 | } | |
1296 | ||
1297 | /* | |
1298 | * Retrieve the new associativity information for a virtual processor's | |
1299 | * home node. | |
1300 | */ | |
b08a2a12 | 1301 | static long hcall_vphn(unsigned long cpu, __be32 *associativity) |
9eff1a38 | 1302 | { |
cd9d6cc7 | 1303 | long rc; |
9eff1a38 JL |
1304 | long retbuf[PLPAR_HCALL9_BUFSIZE] = {0}; |
1305 | u64 flags = 1; | |
1306 | int hwcpu = get_hard_smp_processor_id(cpu); | |
5c9fb189 | 1307 | int i; |
9eff1a38 JL |
1308 | |
1309 | rc = plpar_hcall9(H_HOME_NODE_ASSOCIATIVITY, retbuf, flags, hwcpu); | |
5c9fb189 GK |
1310 | for (i = 0; i < 6; i++) |
1311 | retbuf[i] = cpu_to_be64(retbuf[i]); | |
9eff1a38 JL |
1312 | vphn_unpack_associativity(retbuf, associativity); |
1313 | ||
1314 | return rc; | |
1315 | } | |
1316 | ||
1317 | static long vphn_get_associativity(unsigned long cpu, | |
b08a2a12 | 1318 | __be32 *associativity) |
9eff1a38 | 1319 | { |
cd9d6cc7 | 1320 | long rc; |
9eff1a38 JL |
1321 | |
1322 | rc = hcall_vphn(cpu, associativity); | |
1323 | ||
1324 | switch (rc) { | |
1325 | case H_FUNCTION: | |
1326 | printk(KERN_INFO | |
1327 | "VPHN is not supported. Disabling polling...\n"); | |
1328 | stop_topology_update(); | |
1329 | break; | |
1330 | case H_HARDWARE: | |
1331 | printk(KERN_ERR | |
1332 | "hcall_vphn() experienced a hardware fault " | |
1333 | "preventing VPHN. Disabling polling...\n"); | |
1334 | stop_topology_update(); | |
1335 | } | |
1336 | ||
1337 | return rc; | |
1338 | } | |
1339 | ||
30c05350 NF |
1340 | /* |
1341 | * Update the CPU maps and sysfs entries for a single CPU when its NUMA | |
1342 | * characteristics change. This function doesn't perform any locking and is | |
1343 | * only safe to call from stop_machine(). | |
1344 | */ | |
1345 | static int update_cpu_topology(void *data) | |
1346 | { | |
1347 | struct topology_update_data *update; | |
1348 | unsigned long cpu; | |
1349 | ||
1350 | if (!data) | |
1351 | return -EINVAL; | |
1352 | ||
3be7db6a | 1353 | cpu = smp_processor_id(); |
30c05350 NF |
1354 | |
1355 | for (update = data; update; update = update->next) { | |
2c0a33f9 | 1356 | int new_nid = update->new_nid; |
30c05350 NF |
1357 | if (cpu != update->cpu) |
1358 | continue; | |
1359 | ||
49f8d8c0 | 1360 | unmap_cpu_from_node(cpu); |
2c0a33f9 NA |
1361 | map_cpu_to_node(cpu, new_nid); |
1362 | set_cpu_numa_node(cpu, new_nid); | |
1363 | set_cpu_numa_mem(cpu, local_memory_node(new_nid)); | |
176bbf14 | 1364 | vdso_getcpu_init(); |
30c05350 NF |
1365 | } |
1366 | ||
1367 | return 0; | |
1368 | } | |
1369 | ||
d4edc5b6 SB |
1370 | static int update_lookup_table(void *data) |
1371 | { | |
1372 | struct topology_update_data *update; | |
1373 | ||
1374 | if (!data) | |
1375 | return -EINVAL; | |
1376 | ||
1377 | /* | |
1378 | * Upon topology update, the numa-cpu lookup table needs to be updated | |
1379 | * for all threads in the core, including offline CPUs, to ensure that | |
1380 | * future hotplug operations respect the cpu-to-node associativity | |
1381 | * properly. | |
1382 | */ | |
1383 | for (update = data; update; update = update->next) { | |
1384 | int nid, base, j; | |
1385 | ||
1386 | nid = update->new_nid; | |
1387 | base = cpu_first_thread_sibling(update->cpu); | |
1388 | ||
1389 | for (j = 0; j < threads_per_core; j++) { | |
1390 | update_numa_cpu_lookup_table(base + j, nid); | |
1391 | } | |
1392 | } | |
1393 | ||
1394 | return 0; | |
1395 | } | |
1396 | ||
9eff1a38 JL |
1397 | /* |
1398 | * Update the node maps and sysfs entries for each cpu whose home node | |
79c5fceb | 1399 | * has changed. Returns 1 when the topology has changed, and 0 otherwise. |
9eff1a38 JL |
1400 | */ |
1401 | int arch_update_cpu_topology(void) | |
1402 | { | |
3be7db6a | 1403 | unsigned int cpu, sibling, changed = 0; |
30c05350 | 1404 | struct topology_update_data *updates, *ud; |
b08a2a12 | 1405 | __be32 associativity[VPHN_ASSOC_BUFSIZE] = {0}; |
176bbf14 | 1406 | cpumask_t updated_cpus; |
8a25a2fd | 1407 | struct device *dev; |
3be7db6a | 1408 | int weight, new_nid, i = 0; |
9eff1a38 | 1409 | |
2d73bae1 NA |
1410 | if (!prrn_enabled && !vphn_enabled) |
1411 | return 0; | |
1412 | ||
30c05350 NF |
1413 | weight = cpumask_weight(&cpu_associativity_changes_mask); |
1414 | if (!weight) | |
1415 | return 0; | |
1416 | ||
1417 | updates = kzalloc(weight * (sizeof(*updates)), GFP_KERNEL); | |
1418 | if (!updates) | |
1419 | return 0; | |
9eff1a38 | 1420 | |
176bbf14 JL |
1421 | cpumask_clear(&updated_cpus); |
1422 | ||
5d88aa85 | 1423 | for_each_cpu(cpu, &cpu_associativity_changes_mask) { |
3be7db6a RJ |
1424 | /* |
1425 | * If siblings aren't flagged for changes, updates list | |
1426 | * will be too short. Skip on this update and set for next | |
1427 | * update. | |
1428 | */ | |
1429 | if (!cpumask_subset(cpu_sibling_mask(cpu), | |
1430 | &cpu_associativity_changes_mask)) { | |
1431 | pr_info("Sibling bits not set for associativity " | |
1432 | "change, cpu%d\n", cpu); | |
1433 | cpumask_or(&cpu_associativity_changes_mask, | |
1434 | &cpu_associativity_changes_mask, | |
1435 | cpu_sibling_mask(cpu)); | |
1436 | cpu = cpu_last_thread_sibling(cpu); | |
1437 | continue; | |
1438 | } | |
9eff1a38 | 1439 | |
3be7db6a RJ |
1440 | /* Use associativity from first thread for all siblings */ |
1441 | vphn_get_associativity(cpu, associativity); | |
1442 | new_nid = associativity_to_nid(associativity); | |
1443 | if (new_nid < 0 || !node_online(new_nid)) | |
1444 | new_nid = first_online_node; | |
1445 | ||
1446 | if (new_nid == numa_cpu_lookup_table[cpu]) { | |
1447 | cpumask_andnot(&cpu_associativity_changes_mask, | |
1448 | &cpu_associativity_changes_mask, | |
1449 | cpu_sibling_mask(cpu)); | |
1450 | cpu = cpu_last_thread_sibling(cpu); | |
1451 | continue; | |
1452 | } | |
9eff1a38 | 1453 | |
3be7db6a RJ |
1454 | for_each_cpu(sibling, cpu_sibling_mask(cpu)) { |
1455 | ud = &updates[i++]; | |
1456 | ud->cpu = sibling; | |
1457 | ud->new_nid = new_nid; | |
1458 | ud->old_nid = numa_cpu_lookup_table[sibling]; | |
1459 | cpumask_set_cpu(sibling, &updated_cpus); | |
1460 | if (i < weight) | |
1461 | ud->next = &updates[i]; | |
1462 | } | |
1463 | cpu = cpu_last_thread_sibling(cpu); | |
30c05350 NF |
1464 | } |
1465 | ||
2d73bae1 NA |
1466 | pr_debug("Topology update for the following CPUs:\n"); |
1467 | if (cpumask_weight(&updated_cpus)) { | |
1468 | for (ud = &updates[0]; ud; ud = ud->next) { | |
1469 | pr_debug("cpu %d moving from node %d " | |
1470 | "to %d\n", ud->cpu, | |
1471 | ud->old_nid, ud->new_nid); | |
1472 | } | |
1473 | } | |
1474 | ||
9a013361 MW |
1475 | /* |
1476 | * In cases where we have nothing to update (because the updates list | |
1477 | * is too short or because the new topology is same as the old one), | |
1478 | * skip invoking update_cpu_topology() via stop-machine(). This is | |
1479 | * necessary (and not just a fast-path optimization) since stop-machine | |
1480 | * can end up electing a random CPU to run update_cpu_topology(), and | |
1481 | * thus trick us into setting up incorrect cpu-node mappings (since | |
1482 | * 'updates' is kzalloc()'ed). | |
1483 | * | |
1484 | * And for the similar reason, we will skip all the following updating. | |
1485 | */ | |
1486 | if (!cpumask_weight(&updated_cpus)) | |
1487 | goto out; | |
1488 | ||
176bbf14 | 1489 | stop_machine(update_cpu_topology, &updates[0], &updated_cpus); |
30c05350 | 1490 | |
d4edc5b6 SB |
1491 | /* |
1492 | * Update the numa-cpu lookup table with the new mappings, even for | |
1493 | * offline CPUs. It is best to perform this update from the stop- | |
1494 | * machine context. | |
1495 | */ | |
1496 | stop_machine(update_lookup_table, &updates[0], | |
1497 | cpumask_of(raw_smp_processor_id())); | |
1498 | ||
30c05350 | 1499 | for (ud = &updates[0]; ud; ud = ud->next) { |
dd023217 NF |
1500 | unregister_cpu_under_node(ud->cpu, ud->old_nid); |
1501 | register_cpu_under_node(ud->cpu, ud->new_nid); | |
1502 | ||
30c05350 | 1503 | dev = get_cpu_device(ud->cpu); |
8a25a2fd KS |
1504 | if (dev) |
1505 | kobject_uevent(&dev->kobj, KOBJ_CHANGE); | |
30c05350 | 1506 | cpumask_clear_cpu(ud->cpu, &cpu_associativity_changes_mask); |
79c5fceb | 1507 | changed = 1; |
9eff1a38 JL |
1508 | } |
1509 | ||
9a013361 | 1510 | out: |
30c05350 | 1511 | kfree(updates); |
79c5fceb | 1512 | return changed; |
9eff1a38 JL |
1513 | } |
1514 | ||
1515 | static void topology_work_fn(struct work_struct *work) | |
1516 | { | |
1517 | rebuild_sched_domains(); | |
1518 | } | |
1519 | static DECLARE_WORK(topology_work, topology_work_fn); | |
1520 | ||
ec32dd66 | 1521 | static void topology_schedule_update(void) |
9eff1a38 JL |
1522 | { |
1523 | schedule_work(&topology_work); | |
1524 | } | |
1525 | ||
1526 | static void topology_timer_fn(unsigned long ignored) | |
1527 | { | |
5d88aa85 | 1528 | if (prrn_enabled && cpumask_weight(&cpu_associativity_changes_mask)) |
9eff1a38 | 1529 | topology_schedule_update(); |
5d88aa85 JL |
1530 | else if (vphn_enabled) { |
1531 | if (update_cpu_associativity_changes_mask() > 0) | |
1532 | topology_schedule_update(); | |
1533 | reset_topology_timer(); | |
1534 | } | |
9eff1a38 JL |
1535 | } |
1536 | static struct timer_list topology_timer = | |
1537 | TIMER_INITIALIZER(topology_timer_fn, 0, 0); | |
1538 | ||
5d88aa85 | 1539 | static void reset_topology_timer(void) |
9eff1a38 JL |
1540 | { |
1541 | topology_timer.data = 0; | |
1542 | topology_timer.expires = jiffies + 60 * HZ; | |
5d88aa85 | 1543 | mod_timer(&topology_timer, topology_timer.expires); |
9eff1a38 JL |
1544 | } |
1545 | ||
601abdc3 NF |
1546 | #ifdef CONFIG_SMP |
1547 | ||
5d88aa85 JL |
1548 | static void stage_topology_update(int core_id) |
1549 | { | |
1550 | cpumask_or(&cpu_associativity_changes_mask, | |
1551 | &cpu_associativity_changes_mask, cpu_sibling_mask(core_id)); | |
1552 | reset_topology_timer(); | |
1553 | } | |
1554 | ||
1555 | static int dt_update_callback(struct notifier_block *nb, | |
1556 | unsigned long action, void *data) | |
1557 | { | |
1558 | struct of_prop_reconfig *update; | |
1559 | int rc = NOTIFY_DONE; | |
1560 | ||
1561 | switch (action) { | |
5d88aa85 JL |
1562 | case OF_RECONFIG_UPDATE_PROPERTY: |
1563 | update = (struct of_prop_reconfig *)data; | |
30c05350 NF |
1564 | if (!of_prop_cmp(update->dn->type, "cpu") && |
1565 | !of_prop_cmp(update->prop->name, "ibm,associativity")) { | |
5d88aa85 JL |
1566 | u32 core_id; |
1567 | of_property_read_u32(update->dn, "reg", &core_id); | |
1568 | stage_topology_update(core_id); | |
1569 | rc = NOTIFY_OK; | |
1570 | } | |
1571 | break; | |
1572 | } | |
1573 | ||
1574 | return rc; | |
9eff1a38 JL |
1575 | } |
1576 | ||
5d88aa85 JL |
1577 | static struct notifier_block dt_update_nb = { |
1578 | .notifier_call = dt_update_callback, | |
1579 | }; | |
1580 | ||
601abdc3 NF |
1581 | #endif |
1582 | ||
9eff1a38 | 1583 | /* |
5d88aa85 | 1584 | * Start polling for associativity changes. |
9eff1a38 JL |
1585 | */ |
1586 | int start_topology_update(void) | |
1587 | { | |
1588 | int rc = 0; | |
1589 | ||
5d88aa85 JL |
1590 | if (firmware_has_feature(FW_FEATURE_PRRN)) { |
1591 | if (!prrn_enabled) { | |
1592 | prrn_enabled = 1; | |
1593 | vphn_enabled = 0; | |
601abdc3 | 1594 | #ifdef CONFIG_SMP |
5d88aa85 | 1595 | rc = of_reconfig_notifier_register(&dt_update_nb); |
601abdc3 | 1596 | #endif |
5d88aa85 | 1597 | } |
b7abef04 | 1598 | } else if (firmware_has_feature(FW_FEATURE_VPHN) && |
f13c13a0 | 1599 | lppaca_shared_proc(get_lppaca())) { |
5d88aa85 JL |
1600 | if (!vphn_enabled) { |
1601 | prrn_enabled = 0; | |
1602 | vphn_enabled = 1; | |
1603 | setup_cpu_associativity_change_counters(); | |
1604 | init_timer_deferrable(&topology_timer); | |
1605 | reset_topology_timer(); | |
1606 | } | |
9eff1a38 JL |
1607 | } |
1608 | ||
1609 | return rc; | |
1610 | } | |
9eff1a38 JL |
1611 | |
1612 | /* | |
1613 | * Disable polling for VPHN associativity changes. | |
1614 | */ | |
1615 | int stop_topology_update(void) | |
1616 | { | |
5d88aa85 JL |
1617 | int rc = 0; |
1618 | ||
1619 | if (prrn_enabled) { | |
1620 | prrn_enabled = 0; | |
601abdc3 | 1621 | #ifdef CONFIG_SMP |
5d88aa85 | 1622 | rc = of_reconfig_notifier_unregister(&dt_update_nb); |
601abdc3 | 1623 | #endif |
5d88aa85 JL |
1624 | } else if (vphn_enabled) { |
1625 | vphn_enabled = 0; | |
1626 | rc = del_timer_sync(&topology_timer); | |
1627 | } | |
1628 | ||
1629 | return rc; | |
9eff1a38 | 1630 | } |
e04fa612 NF |
1631 | |
1632 | int prrn_is_enabled(void) | |
1633 | { | |
1634 | return prrn_enabled; | |
1635 | } | |
1636 | ||
1637 | static int topology_read(struct seq_file *file, void *v) | |
1638 | { | |
1639 | if (vphn_enabled || prrn_enabled) | |
1640 | seq_puts(file, "on\n"); | |
1641 | else | |
1642 | seq_puts(file, "off\n"); | |
1643 | ||
1644 | return 0; | |
1645 | } | |
1646 | ||
1647 | static int topology_open(struct inode *inode, struct file *file) | |
1648 | { | |
1649 | return single_open(file, topology_read, NULL); | |
1650 | } | |
1651 | ||
1652 | static ssize_t topology_write(struct file *file, const char __user *buf, | |
1653 | size_t count, loff_t *off) | |
1654 | { | |
1655 | char kbuf[4]; /* "on" or "off" plus null. */ | |
1656 | int read_len; | |
1657 | ||
1658 | read_len = count < 3 ? count : 3; | |
1659 | if (copy_from_user(kbuf, buf, read_len)) | |
1660 | return -EINVAL; | |
1661 | ||
1662 | kbuf[read_len] = '\0'; | |
1663 | ||
1664 | if (!strncmp(kbuf, "on", 2)) | |
1665 | start_topology_update(); | |
1666 | else if (!strncmp(kbuf, "off", 3)) | |
1667 | stop_topology_update(); | |
1668 | else | |
1669 | return -EINVAL; | |
1670 | ||
1671 | return count; | |
1672 | } | |
1673 | ||
1674 | static const struct file_operations topology_ops = { | |
1675 | .read = seq_read, | |
1676 | .write = topology_write, | |
1677 | .open = topology_open, | |
1678 | .release = single_release | |
1679 | }; | |
1680 | ||
1681 | static int topology_update_init(void) | |
1682 | { | |
2d73bae1 NA |
1683 | /* Do not poll for changes if disabled at boot */ |
1684 | if (topology_updates_enabled) | |
1685 | start_topology_update(); | |
1686 | ||
2d15b9b4 NA |
1687 | if (!proc_create("powerpc/topology_updates", 0644, NULL, &topology_ops)) |
1688 | return -ENOMEM; | |
e04fa612 NF |
1689 | |
1690 | return 0; | |
9eff1a38 | 1691 | } |
e04fa612 | 1692 | device_initcall(topology_update_init); |
39bf990e | 1693 | #endif /* CONFIG_PPC_SPLPAR */ |