Commit | Line | Data |
---|---|---|
2874c5fd | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
1da177e4 LT |
2 | /* |
3 | * pSeries NUMA support | |
4 | * | |
5 | * Copyright (C) 2002 Anton Blanchard <anton@au.ibm.com>, IBM | |
1da177e4 | 6 | */ |
2d73bae1 NA |
7 | #define pr_fmt(fmt) "numa: " fmt |
8 | ||
1da177e4 | 9 | #include <linux/threads.h> |
57c8a661 | 10 | #include <linux/memblock.h> |
1da177e4 LT |
11 | #include <linux/init.h> |
12 | #include <linux/mm.h> | |
13 | #include <linux/mmzone.h> | |
4b16f8e2 | 14 | #include <linux/export.h> |
1da177e4 LT |
15 | #include <linux/nodemask.h> |
16 | #include <linux/cpu.h> | |
17 | #include <linux/notifier.h> | |
6df1646e | 18 | #include <linux/of.h> |
06eccea6 | 19 | #include <linux/pfn.h> |
9eff1a38 JL |
20 | #include <linux/cpuset.h> |
21 | #include <linux/node.h> | |
30c05350 | 22 | #include <linux/stop_machine.h> |
e04fa612 NF |
23 | #include <linux/proc_fs.h> |
24 | #include <linux/seq_file.h> | |
25 | #include <linux/uaccess.h> | |
191a7120 | 26 | #include <linux/slab.h> |
3be7db6a | 27 | #include <asm/cputhreads.h> |
45fb6cea | 28 | #include <asm/sparsemem.h> |
d9b2b2a2 | 29 | #include <asm/prom.h> |
2249ca9d | 30 | #include <asm/smp.h> |
d4edc5b6 | 31 | #include <asm/topology.h> |
9eff1a38 JL |
32 | #include <asm/firmware.h> |
33 | #include <asm/paca.h> | |
39bf990e | 34 | #include <asm/hvcall.h> |
ae3a197e | 35 | #include <asm/setup.h> |
176bbf14 | 36 | #include <asm/vdso.h> |
514a9cb3 | 37 | #include <asm/drmem.h> |
1da177e4 LT |
38 | |
39 | static int numa_enabled = 1; | |
40 | ||
1daa6d08 BS |
41 | static char *cmdline __initdata; |
42 | ||
1da177e4 LT |
43 | static int numa_debug; |
44 | #define dbg(args...) if (numa_debug) { printk(KERN_INFO args); } | |
45 | ||
45fb6cea | 46 | int numa_cpu_lookup_table[NR_CPUS]; |
25863de0 | 47 | cpumask_var_t node_to_cpumask_map[MAX_NUMNODES]; |
1da177e4 | 48 | struct pglist_data *node_data[MAX_NUMNODES]; |
45fb6cea AB |
49 | |
50 | EXPORT_SYMBOL(numa_cpu_lookup_table); | |
25863de0 | 51 | EXPORT_SYMBOL(node_to_cpumask_map); |
45fb6cea AB |
52 | EXPORT_SYMBOL(node_data); |
53 | ||
1da177e4 | 54 | static int min_common_depth; |
237a0989 | 55 | static int n_mem_addr_cells, n_mem_size_cells; |
41eab6f8 AB |
56 | static int form1_affinity; |
57 | ||
58 | #define MAX_DISTANCE_REF_POINTS 4 | |
59 | static int distance_ref_points_depth; | |
b08a2a12 | 60 | static const __be32 *distance_ref_points; |
41eab6f8 | 61 | static int distance_lookup_table[MAX_NUMNODES][MAX_DISTANCE_REF_POINTS]; |
1da177e4 | 62 | |
25863de0 AB |
63 | /* |
64 | * Allocate node_to_cpumask_map based on number of available nodes | |
65 | * Requires node_possible_map to be valid. | |
66 | * | |
9512938b | 67 | * Note: cpumask_of_node() is not valid until after this is done. |
25863de0 AB |
68 | */ |
69 | static void __init setup_node_to_cpumask_map(void) | |
70 | { | |
f9d531b8 | 71 | unsigned int node; |
25863de0 AB |
72 | |
73 | /* setup nr_node_ids if not done yet */ | |
f9d531b8 CS |
74 | if (nr_node_ids == MAX_NUMNODES) |
75 | setup_nr_node_ids(); | |
25863de0 AB |
76 | |
77 | /* allocate the map */ | |
c118baf8 | 78 | for_each_node(node) |
25863de0 AB |
79 | alloc_bootmem_cpumask_var(&node_to_cpumask_map[node]); |
80 | ||
81 | /* cpumask_of_node() will now work */ | |
b9726c26 | 82 | dbg("Node to cpumask map for %u nodes\n", nr_node_ids); |
25863de0 AB |
83 | } |
84 | ||
55671f3c | 85 | static int __init fake_numa_create_new_node(unsigned long end_pfn, |
1daa6d08 BS |
86 | unsigned int *nid) |
87 | { | |
88 | unsigned long long mem; | |
89 | char *p = cmdline; | |
90 | static unsigned int fake_nid; | |
91 | static unsigned long long curr_boundary; | |
92 | ||
93 | /* | |
94 | * Modify node id, iff we started creating NUMA nodes | |
95 | * We want to continue from where we left of the last time | |
96 | */ | |
97 | if (fake_nid) | |
98 | *nid = fake_nid; | |
99 | /* | |
100 | * In case there are no more arguments to parse, the | |
101 | * node_id should be the same as the last fake node id | |
102 | * (we've handled this above). | |
103 | */ | |
104 | if (!p) | |
105 | return 0; | |
106 | ||
107 | mem = memparse(p, &p); | |
108 | if (!mem) | |
109 | return 0; | |
110 | ||
111 | if (mem < curr_boundary) | |
112 | return 0; | |
113 | ||
114 | curr_boundary = mem; | |
115 | ||
116 | if ((end_pfn << PAGE_SHIFT) > mem) { | |
117 | /* | |
118 | * Skip commas and spaces | |
119 | */ | |
120 | while (*p == ',' || *p == ' ' || *p == '\t') | |
121 | p++; | |
122 | ||
123 | cmdline = p; | |
124 | fake_nid++; | |
125 | *nid = fake_nid; | |
126 | dbg("created new fake_node with id %d\n", fake_nid); | |
127 | return 1; | |
128 | } | |
129 | return 0; | |
130 | } | |
131 | ||
d4edc5b6 SB |
132 | static void reset_numa_cpu_lookup_table(void) |
133 | { | |
134 | unsigned int cpu; | |
135 | ||
136 | for_each_possible_cpu(cpu) | |
137 | numa_cpu_lookup_table[cpu] = -1; | |
138 | } | |
139 | ||
d4edc5b6 SB |
140 | static void map_cpu_to_node(int cpu, int node) |
141 | { | |
142 | update_numa_cpu_lookup_table(cpu, node); | |
45fb6cea | 143 | |
bf4b85b0 NL |
144 | dbg("adding cpu %d to node %d\n", cpu, node); |
145 | ||
25863de0 AB |
146 | if (!(cpumask_test_cpu(cpu, node_to_cpumask_map[node]))) |
147 | cpumask_set_cpu(cpu, node_to_cpumask_map[node]); | |
1da177e4 LT |
148 | } |
149 | ||
39bf990e | 150 | #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_PPC_SPLPAR) |
1da177e4 LT |
151 | static void unmap_cpu_from_node(unsigned long cpu) |
152 | { | |
153 | int node = numa_cpu_lookup_table[cpu]; | |
154 | ||
155 | dbg("removing cpu %lu from node %d\n", cpu, node); | |
156 | ||
25863de0 | 157 | if (cpumask_test_cpu(cpu, node_to_cpumask_map[node])) { |
429f4d8d | 158 | cpumask_clear_cpu(cpu, node_to_cpumask_map[node]); |
1da177e4 LT |
159 | } else { |
160 | printk(KERN_ERR "WARNING: cpu %lu not found in node %d\n", | |
161 | cpu, node); | |
162 | } | |
163 | } | |
39bf990e | 164 | #endif /* CONFIG_HOTPLUG_CPU || CONFIG_PPC_SPLPAR */ |
1da177e4 | 165 | |
1da177e4 | 166 | /* must hold reference to node during call */ |
b08a2a12 | 167 | static const __be32 *of_get_associativity(struct device_node *dev) |
1da177e4 | 168 | { |
e2eb6392 | 169 | return of_get_property(dev, "ibm,associativity", NULL); |
1da177e4 LT |
170 | } |
171 | ||
41eab6f8 AB |
172 | int __node_distance(int a, int b) |
173 | { | |
174 | int i; | |
175 | int distance = LOCAL_DISTANCE; | |
176 | ||
177 | if (!form1_affinity) | |
7122beee | 178 | return ((a == b) ? LOCAL_DISTANCE : REMOTE_DISTANCE); |
41eab6f8 AB |
179 | |
180 | for (i = 0; i < distance_ref_points_depth; i++) { | |
181 | if (distance_lookup_table[a][i] == distance_lookup_table[b][i]) | |
182 | break; | |
183 | ||
184 | /* Double the distance for each NUMA level */ | |
185 | distance *= 2; | |
186 | } | |
187 | ||
188 | return distance; | |
189 | } | |
12c743eb | 190 | EXPORT_SYMBOL(__node_distance); |
41eab6f8 AB |
191 | |
192 | static void initialize_distance_lookup_table(int nid, | |
b08a2a12 | 193 | const __be32 *associativity) |
41eab6f8 AB |
194 | { |
195 | int i; | |
196 | ||
197 | if (!form1_affinity) | |
198 | return; | |
199 | ||
200 | for (i = 0; i < distance_ref_points_depth; i++) { | |
b08a2a12 AP |
201 | const __be32 *entry; |
202 | ||
1d805440 | 203 | entry = &associativity[be32_to_cpu(distance_ref_points[i]) - 1]; |
b08a2a12 | 204 | distance_lookup_table[nid][i] = of_read_number(entry, 1); |
41eab6f8 AB |
205 | } |
206 | } | |
207 | ||
482ec7c4 NL |
208 | /* Returns nid in the range [0..MAX_NUMNODES-1], or -1 if no useful numa |
209 | * info is found. | |
210 | */ | |
b08a2a12 | 211 | static int associativity_to_nid(const __be32 *associativity) |
1da177e4 | 212 | { |
98fa15f3 | 213 | int nid = NUMA_NO_NODE; |
1da177e4 LT |
214 | |
215 | if (min_common_depth == -1) | |
482ec7c4 | 216 | goto out; |
1da177e4 | 217 | |
b08a2a12 AP |
218 | if (of_read_number(associativity, 1) >= min_common_depth) |
219 | nid = of_read_number(&associativity[min_common_depth], 1); | |
bc16a759 NL |
220 | |
221 | /* POWER4 LPAR uses 0xffff as invalid node */ | |
482ec7c4 | 222 | if (nid == 0xffff || nid >= MAX_NUMNODES) |
98fa15f3 | 223 | nid = NUMA_NO_NODE; |
41eab6f8 | 224 | |
b08a2a12 | 225 | if (nid > 0 && |
1d805440 ND |
226 | of_read_number(associativity, 1) >= distance_ref_points_depth) { |
227 | /* | |
228 | * Skip the length field and send start of associativity array | |
229 | */ | |
230 | initialize_distance_lookup_table(nid, associativity + 1); | |
231 | } | |
41eab6f8 | 232 | |
482ec7c4 | 233 | out: |
cf950b7a | 234 | return nid; |
1da177e4 LT |
235 | } |
236 | ||
9eff1a38 JL |
237 | /* Returns the nid associated with the given device tree node, |
238 | * or -1 if not found. | |
239 | */ | |
240 | static int of_node_to_nid_single(struct device_node *device) | |
241 | { | |
98fa15f3 | 242 | int nid = NUMA_NO_NODE; |
b08a2a12 | 243 | const __be32 *tmp; |
9eff1a38 JL |
244 | |
245 | tmp = of_get_associativity(device); | |
246 | if (tmp) | |
247 | nid = associativity_to_nid(tmp); | |
248 | return nid; | |
249 | } | |
250 | ||
953039c8 JK |
251 | /* Walk the device tree upwards, looking for an associativity id */ |
252 | int of_node_to_nid(struct device_node *device) | |
253 | { | |
98fa15f3 | 254 | int nid = NUMA_NO_NODE; |
953039c8 JK |
255 | |
256 | of_node_get(device); | |
257 | while (device) { | |
258 | nid = of_node_to_nid_single(device); | |
259 | if (nid != -1) | |
260 | break; | |
261 | ||
1def3758 | 262 | device = of_get_next_parent(device); |
953039c8 JK |
263 | } |
264 | of_node_put(device); | |
265 | ||
266 | return nid; | |
267 | } | |
be9ba9ff | 268 | EXPORT_SYMBOL(of_node_to_nid); |
953039c8 | 269 | |
1da177e4 LT |
270 | static int __init find_min_common_depth(void) |
271 | { | |
41eab6f8 | 272 | int depth; |
e70606eb | 273 | struct device_node *root; |
1da177e4 | 274 | |
1c8ee733 DS |
275 | if (firmware_has_feature(FW_FEATURE_OPAL)) |
276 | root = of_find_node_by_path("/ibm,opal"); | |
277 | else | |
278 | root = of_find_node_by_path("/rtas"); | |
e70606eb ME |
279 | if (!root) |
280 | root = of_find_node_by_path("/"); | |
1da177e4 LT |
281 | |
282 | /* | |
41eab6f8 AB |
283 | * This property is a set of 32-bit integers, each representing |
284 | * an index into the ibm,associativity nodes. | |
285 | * | |
286 | * With form 0 affinity the first integer is for an SMP configuration | |
287 | * (should be all 0's) and the second is for a normal NUMA | |
288 | * configuration. We have only one level of NUMA. | |
289 | * | |
290 | * With form 1 affinity the first integer is the most significant | |
291 | * NUMA boundary and the following are progressively less significant | |
292 | * boundaries. There can be more than one level of NUMA. | |
1da177e4 | 293 | */ |
e70606eb | 294 | distance_ref_points = of_get_property(root, |
41eab6f8 AB |
295 | "ibm,associativity-reference-points", |
296 | &distance_ref_points_depth); | |
297 | ||
298 | if (!distance_ref_points) { | |
299 | dbg("NUMA: ibm,associativity-reference-points not found.\n"); | |
300 | goto err; | |
301 | } | |
302 | ||
303 | distance_ref_points_depth /= sizeof(int); | |
1da177e4 | 304 | |
8002b0c5 NF |
305 | if (firmware_has_feature(FW_FEATURE_OPAL) || |
306 | firmware_has_feature(FW_FEATURE_TYPE1_AFFINITY)) { | |
307 | dbg("Using form 1 affinity\n"); | |
1c8ee733 | 308 | form1_affinity = 1; |
4b83c330 AB |
309 | } |
310 | ||
41eab6f8 | 311 | if (form1_affinity) { |
b08a2a12 | 312 | depth = of_read_number(distance_ref_points, 1); |
1da177e4 | 313 | } else { |
41eab6f8 AB |
314 | if (distance_ref_points_depth < 2) { |
315 | printk(KERN_WARNING "NUMA: " | |
316 | "short ibm,associativity-reference-points\n"); | |
317 | goto err; | |
318 | } | |
319 | ||
b08a2a12 | 320 | depth = of_read_number(&distance_ref_points[1], 1); |
1da177e4 | 321 | } |
1da177e4 | 322 | |
41eab6f8 AB |
323 | /* |
324 | * Warn and cap if the hardware supports more than | |
325 | * MAX_DISTANCE_REF_POINTS domains. | |
326 | */ | |
327 | if (distance_ref_points_depth > MAX_DISTANCE_REF_POINTS) { | |
328 | printk(KERN_WARNING "NUMA: distance array capped at " | |
329 | "%d entries\n", MAX_DISTANCE_REF_POINTS); | |
330 | distance_ref_points_depth = MAX_DISTANCE_REF_POINTS; | |
331 | } | |
332 | ||
e70606eb | 333 | of_node_put(root); |
1da177e4 | 334 | return depth; |
41eab6f8 AB |
335 | |
336 | err: | |
e70606eb | 337 | of_node_put(root); |
41eab6f8 | 338 | return -1; |
1da177e4 LT |
339 | } |
340 | ||
84c9fdd1 | 341 | static void __init get_n_mem_cells(int *n_addr_cells, int *n_size_cells) |
1da177e4 LT |
342 | { |
343 | struct device_node *memory = NULL; | |
1da177e4 LT |
344 | |
345 | memory = of_find_node_by_type(memory, "memory"); | |
54c23310 | 346 | if (!memory) |
84c9fdd1 | 347 | panic("numa.c: No memory nodes found!"); |
54c23310 | 348 | |
a8bda5dd | 349 | *n_addr_cells = of_n_addr_cells(memory); |
9213feea | 350 | *n_size_cells = of_n_size_cells(memory); |
84c9fdd1 | 351 | of_node_put(memory); |
1da177e4 LT |
352 | } |
353 | ||
b08a2a12 | 354 | static unsigned long read_n_cells(int n, const __be32 **buf) |
1da177e4 LT |
355 | { |
356 | unsigned long result = 0; | |
357 | ||
358 | while (n--) { | |
b08a2a12 | 359 | result = (result << 32) | of_read_number(*buf, 1); |
1da177e4 LT |
360 | (*buf)++; |
361 | } | |
362 | return result; | |
363 | } | |
364 | ||
8342681d NF |
365 | struct assoc_arrays { |
366 | u32 n_arrays; | |
367 | u32 array_sz; | |
b08a2a12 | 368 | const __be32 *arrays; |
8342681d NF |
369 | }; |
370 | ||
371 | /* | |
25985edc | 372 | * Retrieve and validate the list of associativity arrays for drconf |
8342681d NF |
373 | * memory from the ibm,associativity-lookup-arrays property of the |
374 | * device tree.. | |
375 | * | |
376 | * The layout of the ibm,associativity-lookup-arrays property is a number N | |
377 | * indicating the number of associativity arrays, followed by a number M | |
378 | * indicating the size of each associativity array, followed by a list | |
379 | * of N associativity arrays. | |
380 | */ | |
35f80deb | 381 | static int of_get_assoc_arrays(struct assoc_arrays *aa) |
8342681d | 382 | { |
35f80deb | 383 | struct device_node *memory; |
b08a2a12 | 384 | const __be32 *prop; |
8342681d NF |
385 | u32 len; |
386 | ||
35f80deb NF |
387 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); |
388 | if (!memory) | |
389 | return -1; | |
390 | ||
8342681d | 391 | prop = of_get_property(memory, "ibm,associativity-lookup-arrays", &len); |
35f80deb NF |
392 | if (!prop || len < 2 * sizeof(unsigned int)) { |
393 | of_node_put(memory); | |
8342681d | 394 | return -1; |
35f80deb | 395 | } |
8342681d | 396 | |
b08a2a12 AP |
397 | aa->n_arrays = of_read_number(prop++, 1); |
398 | aa->array_sz = of_read_number(prop++, 1); | |
8342681d | 399 | |
35f80deb NF |
400 | of_node_put(memory); |
401 | ||
42b2aa86 | 402 | /* Now that we know the number of arrays and size of each array, |
8342681d NF |
403 | * revalidate the size of the property read in. |
404 | */ | |
405 | if (len < (aa->n_arrays * aa->array_sz + 2) * sizeof(unsigned int)) | |
406 | return -1; | |
407 | ||
408 | aa->arrays = prop; | |
409 | return 0; | |
410 | } | |
411 | ||
412 | /* | |
413 | * This is like of_node_to_nid_single() for memory represented in the | |
414 | * ibm,dynamic-reconfiguration-memory node. | |
415 | */ | |
514a9cb3 | 416 | static int of_drconf_to_nid_single(struct drmem_lmb *lmb) |
8342681d | 417 | { |
b88fc309 | 418 | struct assoc_arrays aa = { .arrays = NULL }; |
8342681d NF |
419 | int default_nid = 0; |
420 | int nid = default_nid; | |
b88fc309 NF |
421 | int rc, index; |
422 | ||
423 | rc = of_get_assoc_arrays(&aa); | |
424 | if (rc) | |
425 | return default_nid; | |
8342681d | 426 | |
b88fc309 | 427 | if (min_common_depth > 0 && min_common_depth <= aa.array_sz && |
514a9cb3 NF |
428 | !(lmb->flags & DRCONF_MEM_AI_INVALID) && |
429 | lmb->aa_index < aa.n_arrays) { | |
430 | index = lmb->aa_index * aa.array_sz + min_common_depth - 1; | |
b88fc309 | 431 | nid = of_read_number(&aa.arrays[index], 1); |
8342681d NF |
432 | |
433 | if (nid == 0xffff || nid >= MAX_NUMNODES) | |
434 | nid = default_nid; | |
1d805440 ND |
435 | |
436 | if (nid > 0) { | |
514a9cb3 | 437 | index = lmb->aa_index * aa.array_sz; |
1d805440 | 438 | initialize_distance_lookup_table(nid, |
b88fc309 | 439 | &aa.arrays[index]); |
1d805440 | 440 | } |
8342681d NF |
441 | } |
442 | ||
443 | return nid; | |
444 | } | |
445 | ||
1da177e4 LT |
446 | /* |
447 | * Figure out to which domain a cpu belongs and stick it there. | |
448 | * Return the id of the domain used. | |
449 | */ | |
061d19f2 | 450 | static int numa_setup_cpu(unsigned long lcpu) |
1da177e4 | 451 | { |
98fa15f3 | 452 | int nid = NUMA_NO_NODE; |
d4edc5b6 SB |
453 | struct device_node *cpu; |
454 | ||
455 | /* | |
456 | * If a valid cpu-to-node mapping is already available, use it | |
457 | * directly instead of querying the firmware, since it represents | |
458 | * the most recent mapping notified to us by the platform (eg: VPHN). | |
459 | */ | |
460 | if ((nid = numa_cpu_lookup_table[lcpu]) >= 0) { | |
461 | map_cpu_to_node(lcpu, nid); | |
462 | return nid; | |
463 | } | |
464 | ||
465 | cpu = of_get_cpu_node(lcpu, NULL); | |
1da177e4 LT |
466 | |
467 | if (!cpu) { | |
468 | WARN_ON(1); | |
297cf502 LZ |
469 | if (cpu_present(lcpu)) |
470 | goto out_present; | |
471 | else | |
472 | goto out; | |
1da177e4 LT |
473 | } |
474 | ||
953039c8 | 475 | nid = of_node_to_nid_single(cpu); |
1da177e4 | 476 | |
297cf502 | 477 | out_present: |
ea05ba7c | 478 | if (nid < 0 || !node_possible(nid)) |
72c33688 | 479 | nid = first_online_node; |
1da177e4 | 480 | |
297cf502 | 481 | map_cpu_to_node(lcpu, nid); |
1da177e4 | 482 | of_node_put(cpu); |
297cf502 | 483 | out: |
cf950b7a | 484 | return nid; |
1da177e4 LT |
485 | } |
486 | ||
68fb18aa SB |
487 | static void verify_cpu_node_mapping(int cpu, int node) |
488 | { | |
489 | int base, sibling, i; | |
490 | ||
491 | /* Verify that all the threads in the core belong to the same node */ | |
492 | base = cpu_first_thread_sibling(cpu); | |
493 | ||
494 | for (i = 0; i < threads_per_core; i++) { | |
495 | sibling = base + i; | |
496 | ||
497 | if (sibling == cpu || cpu_is_offline(sibling)) | |
498 | continue; | |
499 | ||
500 | if (cpu_to_node(sibling) != node) { | |
501 | WARN(1, "CPU thread siblings %d and %d don't belong" | |
502 | " to the same node!\n", cpu, sibling); | |
503 | break; | |
504 | } | |
505 | } | |
506 | } | |
507 | ||
bdab88e0 SAS |
508 | /* Must run before sched domains notifier. */ |
509 | static int ppc_numa_cpu_prepare(unsigned int cpu) | |
510 | { | |
511 | int nid; | |
512 | ||
513 | nid = numa_setup_cpu(cpu); | |
514 | verify_cpu_node_mapping(cpu, nid); | |
515 | return 0; | |
516 | } | |
517 | ||
518 | static int ppc_numa_cpu_dead(unsigned int cpu) | |
519 | { | |
1da177e4 | 520 | #ifdef CONFIG_HOTPLUG_CPU |
bdab88e0 | 521 | unmap_cpu_from_node(cpu); |
1da177e4 | 522 | #endif |
bdab88e0 | 523 | return 0; |
1da177e4 LT |
524 | } |
525 | ||
526 | /* | |
527 | * Check and possibly modify a memory region to enforce the memory limit. | |
528 | * | |
529 | * Returns the size the region should have to enforce the memory limit. | |
530 | * This will either be the original value of size, a truncated value, | |
531 | * or zero. If the returned value of size is 0 the region should be | |
25985edc | 532 | * discarded as it lies wholly above the memory limit. |
1da177e4 | 533 | */ |
45fb6cea AB |
534 | static unsigned long __init numa_enforce_memory_limit(unsigned long start, |
535 | unsigned long size) | |
1da177e4 LT |
536 | { |
537 | /* | |
95f72d1e | 538 | * We use memblock_end_of_DRAM() in here instead of memory_limit because |
1da177e4 | 539 | * we've already adjusted it for the limit and it takes care of |
fe55249d MM |
540 | * having memory holes below the limit. Also, in the case of |
541 | * iommu_is_off, memory_limit is not set but is implicitly enforced. | |
1da177e4 | 542 | */ |
1da177e4 | 543 | |
95f72d1e | 544 | if (start + size <= memblock_end_of_DRAM()) |
1da177e4 LT |
545 | return size; |
546 | ||
95f72d1e | 547 | if (start >= memblock_end_of_DRAM()) |
1da177e4 LT |
548 | return 0; |
549 | ||
95f72d1e | 550 | return memblock_end_of_DRAM() - start; |
1da177e4 LT |
551 | } |
552 | ||
cf00085d C |
553 | /* |
554 | * Reads the counter for a given entry in | |
555 | * linux,drconf-usable-memory property | |
556 | */ | |
b08a2a12 | 557 | static inline int __init read_usm_ranges(const __be32 **usm) |
cf00085d C |
558 | { |
559 | /* | |
3fdfd990 | 560 | * For each lmb in ibm,dynamic-memory a corresponding |
cf00085d C |
561 | * entry in linux,drconf-usable-memory property contains |
562 | * a counter followed by that many (base, size) duple. | |
563 | * read the counter from linux,drconf-usable-memory | |
564 | */ | |
565 | return read_n_cells(n_mem_size_cells, usm); | |
566 | } | |
567 | ||
0204568a PM |
568 | /* |
569 | * Extract NUMA information from the ibm,dynamic-reconfiguration-memory | |
570 | * node. This assumes n_mem_{addr,size}_cells have been set. | |
571 | */ | |
514a9cb3 NF |
572 | static void __init numa_setup_drmem_lmb(struct drmem_lmb *lmb, |
573 | const __be32 **usm) | |
0204568a | 574 | { |
514a9cb3 NF |
575 | unsigned int ranges, is_kexec_kdump = 0; |
576 | unsigned long base, size, sz; | |
8342681d | 577 | int nid; |
8342681d | 578 | |
514a9cb3 NF |
579 | /* |
580 | * Skip this block if the reserved bit is set in flags (0x80) | |
581 | * or if the block is not assigned to this partition (0x8) | |
582 | */ | |
583 | if ((lmb->flags & DRCONF_MEM_RESERVED) | |
584 | || !(lmb->flags & DRCONF_MEM_ASSIGNED)) | |
8342681d NF |
585 | return; |
586 | ||
514a9cb3 | 587 | if (*usm) |
cf00085d C |
588 | is_kexec_kdump = 1; |
589 | ||
514a9cb3 NF |
590 | base = lmb->base_addr; |
591 | size = drmem_lmb_size(); | |
592 | ranges = 1; | |
8342681d | 593 | |
514a9cb3 NF |
594 | if (is_kexec_kdump) { |
595 | ranges = read_usm_ranges(usm); | |
596 | if (!ranges) /* there are no (base, size) duple */ | |
597 | return; | |
598 | } | |
8342681d | 599 | |
514a9cb3 | 600 | do { |
cf00085d | 601 | if (is_kexec_kdump) { |
514a9cb3 NF |
602 | base = read_n_cells(n_mem_addr_cells, usm); |
603 | size = read_n_cells(n_mem_size_cells, usm); | |
cf00085d | 604 | } |
514a9cb3 NF |
605 | |
606 | nid = of_drconf_to_nid_single(lmb); | |
607 | fake_numa_create_new_node(((base + size) >> PAGE_SHIFT), | |
608 | &nid); | |
609 | node_set_online(nid); | |
610 | sz = numa_enforce_memory_limit(base, size); | |
611 | if (sz) | |
612 | memblock_set_node(base, sz, &memblock.memory, nid); | |
613 | } while (--ranges); | |
0204568a PM |
614 | } |
615 | ||
1da177e4 LT |
616 | static int __init parse_numa_properties(void) |
617 | { | |
94db7c5e | 618 | struct device_node *memory; |
482ec7c4 | 619 | int default_nid = 0; |
1da177e4 LT |
620 | unsigned long i; |
621 | ||
622 | if (numa_enabled == 0) { | |
623 | printk(KERN_WARNING "NUMA disabled by user\n"); | |
624 | return -1; | |
625 | } | |
626 | ||
1da177e4 LT |
627 | min_common_depth = find_min_common_depth(); |
628 | ||
1da177e4 LT |
629 | if (min_common_depth < 0) |
630 | return min_common_depth; | |
631 | ||
bf4b85b0 NL |
632 | dbg("NUMA associativity depth for CPU/Memory: %d\n", min_common_depth); |
633 | ||
1da177e4 | 634 | /* |
482ec7c4 NL |
635 | * Even though we connect cpus to numa domains later in SMP |
636 | * init, we need to know the node ids now. This is because | |
637 | * each node to be onlined must have NODE_DATA etc backing it. | |
1da177e4 | 638 | */ |
482ec7c4 | 639 | for_each_present_cpu(i) { |
dfbe93a2 | 640 | struct device_node *cpu; |
cf950b7a | 641 | int nid; |
1da177e4 | 642 | |
8b16cd23 | 643 | cpu = of_get_cpu_node(i, NULL); |
482ec7c4 | 644 | BUG_ON(!cpu); |
953039c8 | 645 | nid = of_node_to_nid_single(cpu); |
482ec7c4 | 646 | of_node_put(cpu); |
1da177e4 | 647 | |
482ec7c4 NL |
648 | /* |
649 | * Don't fall back to default_nid yet -- we will plug | |
650 | * cpus into nodes once the memory scan has discovered | |
651 | * the topology. | |
652 | */ | |
653 | if (nid < 0) | |
654 | continue; | |
655 | node_set_online(nid); | |
1da177e4 LT |
656 | } |
657 | ||
237a0989 | 658 | get_n_mem_cells(&n_mem_addr_cells, &n_mem_size_cells); |
94db7c5e AB |
659 | |
660 | for_each_node_by_type(memory, "memory") { | |
1da177e4 LT |
661 | unsigned long start; |
662 | unsigned long size; | |
cf950b7a | 663 | int nid; |
1da177e4 | 664 | int ranges; |
b08a2a12 | 665 | const __be32 *memcell_buf; |
1da177e4 LT |
666 | unsigned int len; |
667 | ||
e2eb6392 | 668 | memcell_buf = of_get_property(memory, |
ba759485 ME |
669 | "linux,usable-memory", &len); |
670 | if (!memcell_buf || len <= 0) | |
e2eb6392 | 671 | memcell_buf = of_get_property(memory, "reg", &len); |
1da177e4 LT |
672 | if (!memcell_buf || len <= 0) |
673 | continue; | |
674 | ||
cc5d0189 BH |
675 | /* ranges in cell */ |
676 | ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells); | |
1da177e4 LT |
677 | new_range: |
678 | /* these are order-sensitive, and modify the buffer pointer */ | |
237a0989 MK |
679 | start = read_n_cells(n_mem_addr_cells, &memcell_buf); |
680 | size = read_n_cells(n_mem_size_cells, &memcell_buf); | |
1da177e4 | 681 | |
482ec7c4 NL |
682 | /* |
683 | * Assumption: either all memory nodes or none will | |
684 | * have associativity properties. If none, then | |
685 | * everything goes to default_nid. | |
686 | */ | |
953039c8 | 687 | nid = of_node_to_nid_single(memory); |
482ec7c4 NL |
688 | if (nid < 0) |
689 | nid = default_nid; | |
1daa6d08 BS |
690 | |
691 | fake_numa_create_new_node(((start + size) >> PAGE_SHIFT), &nid); | |
482ec7c4 | 692 | node_set_online(nid); |
1da177e4 | 693 | |
7656cd8e RA |
694 | size = numa_enforce_memory_limit(start, size); |
695 | if (size) | |
696 | memblock_set_node(start, size, &memblock.memory, nid); | |
1da177e4 LT |
697 | |
698 | if (--ranges) | |
699 | goto new_range; | |
700 | } | |
701 | ||
0204568a | 702 | /* |
dfbe93a2 AB |
703 | * Now do the same thing for each MEMBLOCK listed in the |
704 | * ibm,dynamic-memory property in the | |
705 | * ibm,dynamic-reconfiguration-memory node. | |
0204568a PM |
706 | */ |
707 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); | |
514a9cb3 NF |
708 | if (memory) { |
709 | walk_drmem_lmbs(memory, numa_setup_drmem_lmb); | |
710 | of_node_put(memory); | |
711 | } | |
0204568a | 712 | |
1da177e4 LT |
713 | return 0; |
714 | } | |
715 | ||
716 | static void __init setup_nonnuma(void) | |
717 | { | |
95f72d1e YL |
718 | unsigned long top_of_ram = memblock_end_of_DRAM(); |
719 | unsigned long total_ram = memblock_phys_mem_size(); | |
c67c3cb4 | 720 | unsigned long start_pfn, end_pfn; |
28be7072 BH |
721 | unsigned int nid = 0; |
722 | struct memblock_region *reg; | |
1da177e4 | 723 | |
e110b281 | 724 | printk(KERN_DEBUG "Top of RAM: 0x%lx, Total RAM: 0x%lx\n", |
1da177e4 | 725 | top_of_ram, total_ram); |
e110b281 | 726 | printk(KERN_DEBUG "Memory hole size: %ldMB\n", |
1da177e4 LT |
727 | (top_of_ram - total_ram) >> 20); |
728 | ||
28be7072 | 729 | for_each_memblock(memory, reg) { |
c7fc2de0 YL |
730 | start_pfn = memblock_region_memory_base_pfn(reg); |
731 | end_pfn = memblock_region_memory_end_pfn(reg); | |
1daa6d08 BS |
732 | |
733 | fake_numa_create_new_node(end_pfn, &nid); | |
1d7cfe18 | 734 | memblock_set_node(PFN_PHYS(start_pfn), |
e7e8de59 TC |
735 | PFN_PHYS(end_pfn - start_pfn), |
736 | &memblock.memory, nid); | |
1daa6d08 | 737 | node_set_online(nid); |
c67c3cb4 | 738 | } |
1da177e4 LT |
739 | } |
740 | ||
4b703a23 AB |
741 | void __init dump_numa_cpu_topology(void) |
742 | { | |
743 | unsigned int node; | |
744 | unsigned int cpu, count; | |
745 | ||
746 | if (min_common_depth == -1 || !numa_enabled) | |
747 | return; | |
748 | ||
749 | for_each_online_node(node) { | |
8467801c | 750 | pr_info("Node %d CPUs:", node); |
4b703a23 AB |
751 | |
752 | count = 0; | |
753 | /* | |
754 | * If we used a CPU iterator here we would miss printing | |
755 | * the holes in the cpumap. | |
756 | */ | |
25863de0 AB |
757 | for (cpu = 0; cpu < nr_cpu_ids; cpu++) { |
758 | if (cpumask_test_cpu(cpu, | |
759 | node_to_cpumask_map[node])) { | |
4b703a23 | 760 | if (count == 0) |
8467801c | 761 | pr_cont(" %u", cpu); |
4b703a23 AB |
762 | ++count; |
763 | } else { | |
764 | if (count > 1) | |
8467801c | 765 | pr_cont("-%u", cpu - 1); |
4b703a23 AB |
766 | count = 0; |
767 | } | |
768 | } | |
769 | ||
770 | if (count > 1) | |
8467801c AK |
771 | pr_cont("-%u", nr_cpu_ids - 1); |
772 | pr_cont("\n"); | |
4b703a23 AB |
773 | } |
774 | } | |
775 | ||
10239733 AB |
776 | /* Initialize NODE_DATA for a node on the local memory */ |
777 | static void __init setup_node_data(int nid, u64 start_pfn, u64 end_pfn) | |
4a618669 | 778 | { |
10239733 AB |
779 | u64 spanned_pages = end_pfn - start_pfn; |
780 | const size_t nd_size = roundup(sizeof(pg_data_t), SMP_CACHE_BYTES); | |
781 | u64 nd_pa; | |
782 | void *nd; | |
783 | int tnid; | |
4a618669 | 784 | |
9a8dd708 | 785 | nd_pa = memblock_phys_alloc_try_nid(nd_size, SMP_CACHE_BYTES, nid); |
33755574 MR |
786 | if (!nd_pa) |
787 | panic("Cannot allocate %zu bytes for node %d data\n", | |
788 | nd_size, nid); | |
789 | ||
10239733 | 790 | nd = __va(nd_pa); |
4a618669 | 791 | |
10239733 AB |
792 | /* report and initialize */ |
793 | pr_info(" NODE_DATA [mem %#010Lx-%#010Lx]\n", | |
794 | nd_pa, nd_pa + nd_size - 1); | |
795 | tnid = early_pfn_to_nid(nd_pa >> PAGE_SHIFT); | |
796 | if (tnid != nid) | |
797 | pr_info(" NODE_DATA(%d) on node %d\n", nid, tnid); | |
4a618669 | 798 | |
10239733 AB |
799 | node_data[nid] = nd; |
800 | memset(NODE_DATA(nid), 0, sizeof(pg_data_t)); | |
801 | NODE_DATA(nid)->node_id = nid; | |
802 | NODE_DATA(nid)->node_start_pfn = start_pfn; | |
803 | NODE_DATA(nid)->node_spanned_pages = spanned_pages; | |
804 | } | |
4a618669 | 805 | |
a346137e MB |
806 | static void __init find_possible_nodes(void) |
807 | { | |
808 | struct device_node *rtas; | |
809 | u32 numnodes, i; | |
810 | ||
811 | if (min_common_depth <= 0) | |
812 | return; | |
813 | ||
814 | rtas = of_find_node_by_path("/rtas"); | |
815 | if (!rtas) | |
816 | return; | |
817 | ||
818 | if (of_property_read_u32_index(rtas, | |
819 | "ibm,max-associativity-domains", | |
820 | min_common_depth, &numnodes)) | |
821 | goto out; | |
822 | ||
823 | for (i = 0; i < numnodes; i++) { | |
ea05ba7c | 824 | if (!node_possible(i)) |
a346137e | 825 | node_set(i, node_possible_map); |
a346137e MB |
826 | } |
827 | ||
828 | out: | |
829 | of_node_put(rtas); | |
830 | } | |
831 | ||
9bd9be00 | 832 | void __init mem_topology_setup(void) |
1da177e4 | 833 | { |
9bd9be00 | 834 | int cpu; |
1da177e4 LT |
835 | |
836 | if (parse_numa_properties()) | |
837 | setup_nonnuma(); | |
1da177e4 | 838 | |
3af229f2 | 839 | /* |
a346137e MB |
840 | * Modify the set of possible NUMA nodes to reflect information |
841 | * available about the set of online nodes, and the set of nodes | |
842 | * that we expect to make use of for this platform's affinity | |
843 | * calculations. | |
3af229f2 NA |
844 | */ |
845 | nodes_and(node_possible_map, node_possible_map, node_online_map); | |
846 | ||
a346137e MB |
847 | find_possible_nodes(); |
848 | ||
9bd9be00 NP |
849 | setup_node_to_cpumask_map(); |
850 | ||
851 | reset_numa_cpu_lookup_table(); | |
852 | ||
853 | for_each_present_cpu(cpu) | |
854 | numa_setup_cpu(cpu); | |
855 | } | |
856 | ||
857 | void __init initmem_init(void) | |
858 | { | |
859 | int nid; | |
860 | ||
861 | max_low_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT; | |
862 | max_pfn = max_low_pfn; | |
863 | ||
864 | memblock_dump_all(); | |
865 | ||
1da177e4 | 866 | for_each_online_node(nid) { |
c67c3cb4 | 867 | unsigned long start_pfn, end_pfn; |
1da177e4 | 868 | |
c67c3cb4 | 869 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); |
10239733 | 870 | setup_node_data(nid, start_pfn, end_pfn); |
8f64e1f2 | 871 | sparse_memory_present_with_active_regions(nid); |
4a618669 | 872 | } |
d3f6204a | 873 | |
21098b9e | 874 | sparse_init(); |
25863de0 | 875 | |
2fabf084 NA |
876 | /* |
877 | * We need the numa_cpu_lookup_table to be accurate for all CPUs, | |
878 | * even before we online them, so that we can use cpu_to_{node,mem} | |
879 | * early in boot, cf. smp_prepare_cpus(). | |
bdab88e0 SAS |
880 | * _nocalls() + manual invocation is used because cpuhp is not yet |
881 | * initialized for the boot CPU. | |
2fabf084 | 882 | */ |
73c1b41e | 883 | cpuhp_setup_state_nocalls(CPUHP_POWER_NUMA_PREPARE, "powerpc/numa:prepare", |
bdab88e0 | 884 | ppc_numa_cpu_prepare, ppc_numa_cpu_dead); |
1da177e4 LT |
885 | } |
886 | ||
1da177e4 LT |
887 | static int __init early_numa(char *p) |
888 | { | |
889 | if (!p) | |
890 | return 0; | |
891 | ||
892 | if (strstr(p, "off")) | |
893 | numa_enabled = 0; | |
894 | ||
895 | if (strstr(p, "debug")) | |
896 | numa_debug = 1; | |
897 | ||
1daa6d08 BS |
898 | p = strstr(p, "fake="); |
899 | if (p) | |
900 | cmdline = p + strlen("fake="); | |
901 | ||
1da177e4 LT |
902 | return 0; |
903 | } | |
904 | early_param("numa", early_numa); | |
237a0989 | 905 | |
558f8649 NL |
906 | /* |
907 | * The platform can inform us through one of several mechanisms | |
908 | * (post-migration device tree updates, PRRN or VPHN) that the NUMA | |
909 | * assignment of a resource has changed. This controls whether we act | |
910 | * on that. Disabled by default. | |
911 | */ | |
912 | static bool topology_updates_enabled; | |
2d73bae1 NA |
913 | |
914 | static int __init early_topology_updates(char *p) | |
915 | { | |
916 | if (!p) | |
917 | return 0; | |
918 | ||
558f8649 NL |
919 | if (!strcmp(p, "on")) { |
920 | pr_warn("Caution: enabling topology updates\n"); | |
921 | topology_updates_enabled = true; | |
2d73bae1 NA |
922 | } |
923 | ||
924 | return 0; | |
925 | } | |
926 | early_param("topology_updates", early_topology_updates); | |
927 | ||
237a0989 | 928 | #ifdef CONFIG_MEMORY_HOTPLUG |
0db9360a | 929 | /* |
0f16ef7f NF |
930 | * Find the node associated with a hot added memory section for |
931 | * memory represented in the device tree by the property | |
932 | * ibm,dynamic-reconfiguration-memory/ibm,dynamic-memory. | |
0db9360a | 933 | */ |
514a9cb3 | 934 | static int hot_add_drconf_scn_to_nid(unsigned long scn_addr) |
0db9360a | 935 | { |
514a9cb3 | 936 | struct drmem_lmb *lmb; |
3fdfd990 | 937 | unsigned long lmb_size; |
98fa15f3 | 938 | int nid = NUMA_NO_NODE; |
0db9360a | 939 | |
514a9cb3 | 940 | lmb_size = drmem_lmb_size(); |
0db9360a | 941 | |
514a9cb3 | 942 | for_each_drmem_lmb(lmb) { |
0db9360a NF |
943 | /* skip this block if it is reserved or not assigned to |
944 | * this partition */ | |
514a9cb3 NF |
945 | if ((lmb->flags & DRCONF_MEM_RESERVED) |
946 | || !(lmb->flags & DRCONF_MEM_ASSIGNED)) | |
0db9360a NF |
947 | continue; |
948 | ||
514a9cb3 NF |
949 | if ((scn_addr < lmb->base_addr) |
950 | || (scn_addr >= (lmb->base_addr + lmb_size))) | |
0f16ef7f NF |
951 | continue; |
952 | ||
514a9cb3 | 953 | nid = of_drconf_to_nid_single(lmb); |
0f16ef7f NF |
954 | break; |
955 | } | |
956 | ||
957 | return nid; | |
958 | } | |
959 | ||
960 | /* | |
961 | * Find the node associated with a hot added memory section for memory | |
962 | * represented in the device tree as a node (i.e. memory@XXXX) for | |
95f72d1e | 963 | * each memblock. |
0f16ef7f | 964 | */ |
ec32dd66 | 965 | static int hot_add_node_scn_to_nid(unsigned long scn_addr) |
0f16ef7f | 966 | { |
94db7c5e | 967 | struct device_node *memory; |
98fa15f3 | 968 | int nid = NUMA_NO_NODE; |
0f16ef7f | 969 | |
94db7c5e | 970 | for_each_node_by_type(memory, "memory") { |
0f16ef7f NF |
971 | unsigned long start, size; |
972 | int ranges; | |
b08a2a12 | 973 | const __be32 *memcell_buf; |
0f16ef7f NF |
974 | unsigned int len; |
975 | ||
976 | memcell_buf = of_get_property(memory, "reg", &len); | |
977 | if (!memcell_buf || len <= 0) | |
978 | continue; | |
979 | ||
980 | /* ranges in cell */ | |
981 | ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells); | |
982 | ||
983 | while (ranges--) { | |
984 | start = read_n_cells(n_mem_addr_cells, &memcell_buf); | |
985 | size = read_n_cells(n_mem_size_cells, &memcell_buf); | |
986 | ||
987 | if ((scn_addr < start) || (scn_addr >= (start + size))) | |
988 | continue; | |
989 | ||
990 | nid = of_node_to_nid_single(memory); | |
991 | break; | |
992 | } | |
0db9360a | 993 | |
0f16ef7f NF |
994 | if (nid >= 0) |
995 | break; | |
0db9360a NF |
996 | } |
997 | ||
60831842 AB |
998 | of_node_put(memory); |
999 | ||
0f16ef7f | 1000 | return nid; |
0db9360a NF |
1001 | } |
1002 | ||
237a0989 MK |
1003 | /* |
1004 | * Find the node associated with a hot added memory section. Section | |
95f72d1e YL |
1005 | * corresponds to a SPARSEMEM section, not an MEMBLOCK. It is assumed that |
1006 | * sections are fully contained within a single MEMBLOCK. | |
237a0989 MK |
1007 | */ |
1008 | int hot_add_scn_to_nid(unsigned long scn_addr) | |
1009 | { | |
1010 | struct device_node *memory = NULL; | |
4a3bac4e | 1011 | int nid; |
237a0989 MK |
1012 | |
1013 | if (!numa_enabled || (min_common_depth < 0)) | |
72c33688 | 1014 | return first_online_node; |
0db9360a NF |
1015 | |
1016 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); | |
1017 | if (memory) { | |
514a9cb3 | 1018 | nid = hot_add_drconf_scn_to_nid(scn_addr); |
0db9360a | 1019 | of_node_put(memory); |
0f16ef7f NF |
1020 | } else { |
1021 | nid = hot_add_node_scn_to_nid(scn_addr); | |
0db9360a | 1022 | } |
237a0989 | 1023 | |
2a8628d4 | 1024 | if (nid < 0 || !node_possible(nid)) |
72c33688 | 1025 | nid = first_online_node; |
237a0989 | 1026 | |
0f16ef7f | 1027 | return nid; |
237a0989 | 1028 | } |
0f16ef7f | 1029 | |
cd34206e NA |
1030 | static u64 hot_add_drconf_memory_max(void) |
1031 | { | |
e70bd3ae | 1032 | struct device_node *memory = NULL; |
45b64ee6 | 1033 | struct device_node *dn = NULL; |
45b64ee6 | 1034 | const __be64 *lrdr = NULL; |
45b64ee6 BR |
1035 | |
1036 | dn = of_find_node_by_path("/rtas"); | |
1037 | if (dn) { | |
1038 | lrdr = of_get_property(dn, "ibm,lrdr-capacity", NULL); | |
1039 | of_node_put(dn); | |
1040 | if (lrdr) | |
1041 | return be64_to_cpup(lrdr); | |
1042 | } | |
cd34206e | 1043 | |
e70bd3ae BR |
1044 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); |
1045 | if (memory) { | |
e70bd3ae | 1046 | of_node_put(memory); |
514a9cb3 | 1047 | return drmem_lmb_memory_max(); |
e70bd3ae | 1048 | } |
45b64ee6 | 1049 | return 0; |
cd34206e NA |
1050 | } |
1051 | ||
1052 | /* | |
1053 | * memory_hotplug_max - return max address of memory that may be added | |
1054 | * | |
1055 | * This is currently only used on systems that support drconfig memory | |
1056 | * hotplug. | |
1057 | */ | |
1058 | u64 memory_hotplug_max(void) | |
1059 | { | |
1060 | return max(hot_add_drconf_memory_max(), memblock_end_of_DRAM()); | |
1061 | } | |
237a0989 | 1062 | #endif /* CONFIG_MEMORY_HOTPLUG */ |
9eff1a38 | 1063 | |
bd03403a | 1064 | /* Virtual Processor Home Node (VPHN) support */ |
39bf990e | 1065 | #ifdef CONFIG_PPC_SPLPAR |
4b6cfb2a | 1066 | |
47d99948 | 1067 | #include "book3s64/vphn.h" |
4b6cfb2a | 1068 | |
30c05350 NF |
1069 | struct topology_update_data { |
1070 | struct topology_update_data *next; | |
1071 | unsigned int cpu; | |
1072 | int old_nid; | |
1073 | int new_nid; | |
1074 | }; | |
1075 | ||
cee5405d MB |
1076 | #define TOPOLOGY_DEF_TIMER_SECS 60 |
1077 | ||
5de16699 | 1078 | static u8 vphn_cpu_change_counts[NR_CPUS][MAX_DISTANCE_REF_POINTS]; |
9eff1a38 JL |
1079 | static cpumask_t cpu_associativity_changes_mask; |
1080 | static int vphn_enabled; | |
5d88aa85 JL |
1081 | static int prrn_enabled; |
1082 | static void reset_topology_timer(void); | |
cee5405d | 1083 | static int topology_timer_secs = 1; |
17f444c0 | 1084 | static int topology_inited; |
9eff1a38 | 1085 | |
cee5405d MB |
1086 | /* |
1087 | * Change polling interval for associativity changes. | |
1088 | */ | |
1089 | int timed_topology_update(int nsecs) | |
1090 | { | |
1091 | if (vphn_enabled) { | |
1092 | if (nsecs > 0) | |
1093 | topology_timer_secs = nsecs; | |
1094 | else | |
1095 | topology_timer_secs = TOPOLOGY_DEF_TIMER_SECS; | |
1096 | ||
1097 | reset_topology_timer(); | |
1098 | } | |
1099 | ||
1100 | return 0; | |
1101 | } | |
9eff1a38 JL |
1102 | |
1103 | /* | |
1104 | * Store the current values of the associativity change counters in the | |
1105 | * hypervisor. | |
1106 | */ | |
1107 | static void setup_cpu_associativity_change_counters(void) | |
1108 | { | |
cd9d6cc7 | 1109 | int cpu; |
9eff1a38 | 1110 | |
5de16699 AB |
1111 | /* The VPHN feature supports a maximum of 8 reference points */ |
1112 | BUILD_BUG_ON(MAX_DISTANCE_REF_POINTS > 8); | |
1113 | ||
9eff1a38 | 1114 | for_each_possible_cpu(cpu) { |
cd9d6cc7 | 1115 | int i; |
9eff1a38 | 1116 | u8 *counts = vphn_cpu_change_counts[cpu]; |
499dcd41 | 1117 | volatile u8 *hypervisor_counts = lppaca_of(cpu).vphn_assoc_counts; |
9eff1a38 | 1118 | |
5de16699 | 1119 | for (i = 0; i < distance_ref_points_depth; i++) |
9eff1a38 | 1120 | counts[i] = hypervisor_counts[i]; |
9eff1a38 JL |
1121 | } |
1122 | } | |
1123 | ||
1124 | /* | |
1125 | * The hypervisor maintains a set of 8 associativity change counters in | |
1126 | * the VPA of each cpu that correspond to the associativity levels in the | |
1127 | * ibm,associativity-reference-points property. When an associativity | |
1128 | * level changes, the corresponding counter is incremented. | |
1129 | * | |
1130 | * Set a bit in cpu_associativity_changes_mask for each cpu whose home | |
1131 | * node associativity levels have changed. | |
1132 | * | |
1133 | * Returns the number of cpus with unhandled associativity changes. | |
1134 | */ | |
1135 | static int update_cpu_associativity_changes_mask(void) | |
1136 | { | |
5d88aa85 | 1137 | int cpu; |
9eff1a38 JL |
1138 | cpumask_t *changes = &cpu_associativity_changes_mask; |
1139 | ||
9eff1a38 JL |
1140 | for_each_possible_cpu(cpu) { |
1141 | int i, changed = 0; | |
1142 | u8 *counts = vphn_cpu_change_counts[cpu]; | |
499dcd41 | 1143 | volatile u8 *hypervisor_counts = lppaca_of(cpu).vphn_assoc_counts; |
9eff1a38 | 1144 | |
5de16699 | 1145 | for (i = 0; i < distance_ref_points_depth; i++) { |
d69043e8 | 1146 | if (hypervisor_counts[i] != counts[i]) { |
9eff1a38 JL |
1147 | counts[i] = hypervisor_counts[i]; |
1148 | changed = 1; | |
1149 | } | |
1150 | } | |
1151 | if (changed) { | |
3be7db6a RJ |
1152 | cpumask_or(changes, changes, cpu_sibling_mask(cpu)); |
1153 | cpu = cpu_last_thread_sibling(cpu); | |
9eff1a38 JL |
1154 | } |
1155 | } | |
1156 | ||
5d88aa85 | 1157 | return cpumask_weight(changes); |
9eff1a38 JL |
1158 | } |
1159 | ||
9eff1a38 JL |
1160 | /* |
1161 | * Retrieve the new associativity information for a virtual processor's | |
1162 | * home node. | |
1163 | */ | |
b08a2a12 | 1164 | static long hcall_vphn(unsigned long cpu, __be32 *associativity) |
9eff1a38 | 1165 | { |
cd9d6cc7 | 1166 | long rc; |
9eff1a38 JL |
1167 | long retbuf[PLPAR_HCALL9_BUFSIZE] = {0}; |
1168 | u64 flags = 1; | |
1169 | int hwcpu = get_hard_smp_processor_id(cpu); | |
1170 | ||
1171 | rc = plpar_hcall9(H_HOME_NODE_ASSOCIATIVITY, retbuf, flags, hwcpu); | |
1172 | vphn_unpack_associativity(retbuf, associativity); | |
1173 | ||
1174 | return rc; | |
1175 | } | |
1176 | ||
1177 | static long vphn_get_associativity(unsigned long cpu, | |
b08a2a12 | 1178 | __be32 *associativity) |
9eff1a38 | 1179 | { |
cd9d6cc7 | 1180 | long rc; |
9eff1a38 JL |
1181 | |
1182 | rc = hcall_vphn(cpu, associativity); | |
1183 | ||
1184 | switch (rc) { | |
1185 | case H_FUNCTION: | |
437ccdc8 | 1186 | printk_once(KERN_INFO |
9eff1a38 JL |
1187 | "VPHN is not supported. Disabling polling...\n"); |
1188 | stop_topology_update(); | |
1189 | break; | |
1190 | case H_HARDWARE: | |
1191 | printk(KERN_ERR | |
1192 | "hcall_vphn() experienced a hardware fault " | |
1193 | "preventing VPHN. Disabling polling...\n"); | |
1194 | stop_topology_update(); | |
17f444c0 MB |
1195 | break; |
1196 | case H_SUCCESS: | |
1197 | dbg("VPHN hcall succeeded. Reset polling...\n"); | |
cee5405d | 1198 | timed_topology_update(0); |
17f444c0 | 1199 | break; |
9eff1a38 JL |
1200 | } |
1201 | ||
1202 | return rc; | |
1203 | } | |
1204 | ||
e67e02a5 | 1205 | int find_and_online_cpu_nid(int cpu) |
ea05ba7c MB |
1206 | { |
1207 | __be32 associativity[VPHN_ASSOC_BUFSIZE] = {0}; | |
1208 | int new_nid; | |
1209 | ||
1210 | /* Use associativity from first thread for all siblings */ | |
2483ef05 SD |
1211 | if (vphn_get_associativity(cpu, associativity)) |
1212 | return cpu_to_node(cpu); | |
1213 | ||
ea05ba7c MB |
1214 | new_nid = associativity_to_nid(associativity); |
1215 | if (new_nid < 0 || !node_possible(new_nid)) | |
1216 | new_nid = first_online_node; | |
1217 | ||
1218 | if (NODE_DATA(new_nid) == NULL) { | |
1219 | #ifdef CONFIG_MEMORY_HOTPLUG | |
1220 | /* | |
1221 | * Need to ensure that NODE_DATA is initialized for a node from | |
1222 | * available memory (see memblock_alloc_try_nid). If unable to | |
1223 | * init the node, then default to nearest node that has memory | |
ac1788cc SD |
1224 | * installed. Skip onlining a node if the subsystems are not |
1225 | * yet initialized. | |
ea05ba7c | 1226 | */ |
ac1788cc | 1227 | if (!topology_inited || try_online_node(new_nid)) |
ea05ba7c MB |
1228 | new_nid = first_online_node; |
1229 | #else | |
1230 | /* | |
1231 | * Default to using the nearest node that has memory installed. | |
1232 | * Otherwise, it would be necessary to patch the kernel MM code | |
1233 | * to deal with more memoryless-node error conditions. | |
1234 | */ | |
1235 | new_nid = first_online_node; | |
1236 | #endif | |
1237 | } | |
1238 | ||
e67e02a5 MB |
1239 | pr_debug("%s:%d cpu %d nid %d\n", __FUNCTION__, __LINE__, |
1240 | cpu, new_nid); | |
ea05ba7c MB |
1241 | return new_nid; |
1242 | } | |
1243 | ||
30c05350 NF |
1244 | /* |
1245 | * Update the CPU maps and sysfs entries for a single CPU when its NUMA | |
1246 | * characteristics change. This function doesn't perform any locking and is | |
1247 | * only safe to call from stop_machine(). | |
1248 | */ | |
1249 | static int update_cpu_topology(void *data) | |
1250 | { | |
1251 | struct topology_update_data *update; | |
1252 | unsigned long cpu; | |
1253 | ||
1254 | if (!data) | |
1255 | return -EINVAL; | |
1256 | ||
3be7db6a | 1257 | cpu = smp_processor_id(); |
30c05350 NF |
1258 | |
1259 | for (update = data; update; update = update->next) { | |
2c0a33f9 | 1260 | int new_nid = update->new_nid; |
30c05350 NF |
1261 | if (cpu != update->cpu) |
1262 | continue; | |
1263 | ||
49f8d8c0 | 1264 | unmap_cpu_from_node(cpu); |
2c0a33f9 NA |
1265 | map_cpu_to_node(cpu, new_nid); |
1266 | set_cpu_numa_node(cpu, new_nid); | |
1267 | set_cpu_numa_mem(cpu, local_memory_node(new_nid)); | |
176bbf14 | 1268 | vdso_getcpu_init(); |
30c05350 NF |
1269 | } |
1270 | ||
1271 | return 0; | |
1272 | } | |
1273 | ||
d4edc5b6 SB |
1274 | static int update_lookup_table(void *data) |
1275 | { | |
1276 | struct topology_update_data *update; | |
1277 | ||
1278 | if (!data) | |
1279 | return -EINVAL; | |
1280 | ||
1281 | /* | |
1282 | * Upon topology update, the numa-cpu lookup table needs to be updated | |
1283 | * for all threads in the core, including offline CPUs, to ensure that | |
1284 | * future hotplug operations respect the cpu-to-node associativity | |
1285 | * properly. | |
1286 | */ | |
1287 | for (update = data; update; update = update->next) { | |
1288 | int nid, base, j; | |
1289 | ||
1290 | nid = update->new_nid; | |
1291 | base = cpu_first_thread_sibling(update->cpu); | |
1292 | ||
1293 | for (j = 0; j < threads_per_core; j++) { | |
1294 | update_numa_cpu_lookup_table(base + j, nid); | |
1295 | } | |
1296 | } | |
1297 | ||
1298 | return 0; | |
1299 | } | |
1300 | ||
9eff1a38 JL |
1301 | /* |
1302 | * Update the node maps and sysfs entries for each cpu whose home node | |
79c5fceb | 1303 | * has changed. Returns 1 when the topology has changed, and 0 otherwise. |
3e401f7a TJB |
1304 | * |
1305 | * cpus_locked says whether we already hold cpu_hotplug_lock. | |
9eff1a38 | 1306 | */ |
3e401f7a | 1307 | int numa_update_cpu_topology(bool cpus_locked) |
9eff1a38 | 1308 | { |
3be7db6a | 1309 | unsigned int cpu, sibling, changed = 0; |
30c05350 | 1310 | struct topology_update_data *updates, *ud; |
176bbf14 | 1311 | cpumask_t updated_cpus; |
8a25a2fd | 1312 | struct device *dev; |
3be7db6a | 1313 | int weight, new_nid, i = 0; |
9eff1a38 | 1314 | |
2ea62630 | 1315 | if (!prrn_enabled && !vphn_enabled && topology_inited) |
2d73bae1 NA |
1316 | return 0; |
1317 | ||
30c05350 NF |
1318 | weight = cpumask_weight(&cpu_associativity_changes_mask); |
1319 | if (!weight) | |
1320 | return 0; | |
1321 | ||
6396bb22 | 1322 | updates = kcalloc(weight, sizeof(*updates), GFP_KERNEL); |
30c05350 NF |
1323 | if (!updates) |
1324 | return 0; | |
9eff1a38 | 1325 | |
176bbf14 JL |
1326 | cpumask_clear(&updated_cpus); |
1327 | ||
5d88aa85 | 1328 | for_each_cpu(cpu, &cpu_associativity_changes_mask) { |
3be7db6a RJ |
1329 | /* |
1330 | * If siblings aren't flagged for changes, updates list | |
1331 | * will be too short. Skip on this update and set for next | |
1332 | * update. | |
1333 | */ | |
1334 | if (!cpumask_subset(cpu_sibling_mask(cpu), | |
1335 | &cpu_associativity_changes_mask)) { | |
1336 | pr_info("Sibling bits not set for associativity " | |
1337 | "change, cpu%d\n", cpu); | |
1338 | cpumask_or(&cpu_associativity_changes_mask, | |
1339 | &cpu_associativity_changes_mask, | |
1340 | cpu_sibling_mask(cpu)); | |
1341 | cpu = cpu_last_thread_sibling(cpu); | |
1342 | continue; | |
1343 | } | |
9eff1a38 | 1344 | |
ea05ba7c | 1345 | new_nid = find_and_online_cpu_nid(cpu); |
3be7db6a RJ |
1346 | |
1347 | if (new_nid == numa_cpu_lookup_table[cpu]) { | |
1348 | cpumask_andnot(&cpu_associativity_changes_mask, | |
1349 | &cpu_associativity_changes_mask, | |
1350 | cpu_sibling_mask(cpu)); | |
17f444c0 MB |
1351 | dbg("Assoc chg gives same node %d for cpu%d\n", |
1352 | new_nid, cpu); | |
3be7db6a RJ |
1353 | cpu = cpu_last_thread_sibling(cpu); |
1354 | continue; | |
1355 | } | |
9eff1a38 | 1356 | |
3be7db6a RJ |
1357 | for_each_cpu(sibling, cpu_sibling_mask(cpu)) { |
1358 | ud = &updates[i++]; | |
8bc93149 | 1359 | ud->next = &updates[i]; |
3be7db6a RJ |
1360 | ud->cpu = sibling; |
1361 | ud->new_nid = new_nid; | |
1362 | ud->old_nid = numa_cpu_lookup_table[sibling]; | |
1363 | cpumask_set_cpu(sibling, &updated_cpus); | |
3be7db6a RJ |
1364 | } |
1365 | cpu = cpu_last_thread_sibling(cpu); | |
30c05350 NF |
1366 | } |
1367 | ||
8bc93149 MB |
1368 | /* |
1369 | * Prevent processing of 'updates' from overflowing array | |
1370 | * where last entry filled in a 'next' pointer. | |
1371 | */ | |
1372 | if (i) | |
1373 | updates[i-1].next = NULL; | |
1374 | ||
2d73bae1 NA |
1375 | pr_debug("Topology update for the following CPUs:\n"); |
1376 | if (cpumask_weight(&updated_cpus)) { | |
1377 | for (ud = &updates[0]; ud; ud = ud->next) { | |
1378 | pr_debug("cpu %d moving from node %d " | |
1379 | "to %d\n", ud->cpu, | |
1380 | ud->old_nid, ud->new_nid); | |
1381 | } | |
1382 | } | |
1383 | ||
9a013361 MW |
1384 | /* |
1385 | * In cases where we have nothing to update (because the updates list | |
1386 | * is too short or because the new topology is same as the old one), | |
1387 | * skip invoking update_cpu_topology() via stop-machine(). This is | |
1388 | * necessary (and not just a fast-path optimization) since stop-machine | |
1389 | * can end up electing a random CPU to run update_cpu_topology(), and | |
1390 | * thus trick us into setting up incorrect cpu-node mappings (since | |
1391 | * 'updates' is kzalloc()'ed). | |
1392 | * | |
1393 | * And for the similar reason, we will skip all the following updating. | |
1394 | */ | |
1395 | if (!cpumask_weight(&updated_cpus)) | |
1396 | goto out; | |
1397 | ||
3e401f7a TJB |
1398 | if (cpus_locked) |
1399 | stop_machine_cpuslocked(update_cpu_topology, &updates[0], | |
1400 | &updated_cpus); | |
1401 | else | |
1402 | stop_machine(update_cpu_topology, &updates[0], &updated_cpus); | |
30c05350 | 1403 | |
d4edc5b6 SB |
1404 | /* |
1405 | * Update the numa-cpu lookup table with the new mappings, even for | |
1406 | * offline CPUs. It is best to perform this update from the stop- | |
1407 | * machine context. | |
1408 | */ | |
3e401f7a TJB |
1409 | if (cpus_locked) |
1410 | stop_machine_cpuslocked(update_lookup_table, &updates[0], | |
d4edc5b6 | 1411 | cpumask_of(raw_smp_processor_id())); |
3e401f7a TJB |
1412 | else |
1413 | stop_machine(update_lookup_table, &updates[0], | |
1414 | cpumask_of(raw_smp_processor_id())); | |
d4edc5b6 | 1415 | |
30c05350 | 1416 | for (ud = &updates[0]; ud; ud = ud->next) { |
dd023217 NF |
1417 | unregister_cpu_under_node(ud->cpu, ud->old_nid); |
1418 | register_cpu_under_node(ud->cpu, ud->new_nid); | |
1419 | ||
30c05350 | 1420 | dev = get_cpu_device(ud->cpu); |
8a25a2fd KS |
1421 | if (dev) |
1422 | kobject_uevent(&dev->kobj, KOBJ_CHANGE); | |
30c05350 | 1423 | cpumask_clear_cpu(ud->cpu, &cpu_associativity_changes_mask); |
79c5fceb | 1424 | changed = 1; |
9eff1a38 JL |
1425 | } |
1426 | ||
9a013361 | 1427 | out: |
30c05350 | 1428 | kfree(updates); |
79c5fceb | 1429 | return changed; |
9eff1a38 JL |
1430 | } |
1431 | ||
3e401f7a TJB |
1432 | int arch_update_cpu_topology(void) |
1433 | { | |
3e401f7a TJB |
1434 | return numa_update_cpu_topology(true); |
1435 | } | |
1436 | ||
9eff1a38 JL |
1437 | static void topology_work_fn(struct work_struct *work) |
1438 | { | |
1439 | rebuild_sched_domains(); | |
1440 | } | |
1441 | static DECLARE_WORK(topology_work, topology_work_fn); | |
1442 | ||
ec32dd66 | 1443 | static void topology_schedule_update(void) |
9eff1a38 JL |
1444 | { |
1445 | schedule_work(&topology_work); | |
1446 | } | |
1447 | ||
df7e828c | 1448 | static void topology_timer_fn(struct timer_list *unused) |
9eff1a38 | 1449 | { |
5d88aa85 | 1450 | if (prrn_enabled && cpumask_weight(&cpu_associativity_changes_mask)) |
9eff1a38 | 1451 | topology_schedule_update(); |
5d88aa85 JL |
1452 | else if (vphn_enabled) { |
1453 | if (update_cpu_associativity_changes_mask() > 0) | |
1454 | topology_schedule_update(); | |
1455 | reset_topology_timer(); | |
1456 | } | |
9eff1a38 | 1457 | } |
df7e828c | 1458 | static struct timer_list topology_timer; |
9eff1a38 | 1459 | |
5d88aa85 | 1460 | static void reset_topology_timer(void) |
9eff1a38 | 1461 | { |
8604895a MB |
1462 | if (vphn_enabled) |
1463 | mod_timer(&topology_timer, jiffies + topology_timer_secs * HZ); | |
9eff1a38 JL |
1464 | } |
1465 | ||
601abdc3 NF |
1466 | #ifdef CONFIG_SMP |
1467 | ||
5d88aa85 JL |
1468 | static int dt_update_callback(struct notifier_block *nb, |
1469 | unsigned long action, void *data) | |
1470 | { | |
f5242e5a | 1471 | struct of_reconfig_data *update = data; |
5d88aa85 JL |
1472 | int rc = NOTIFY_DONE; |
1473 | ||
1474 | switch (action) { | |
5d88aa85 | 1475 | case OF_RECONFIG_UPDATE_PROPERTY: |
e5480bdc | 1476 | if (of_node_is_type(update->dn, "cpu") && |
30c05350 | 1477 | !of_prop_cmp(update->prop->name, "ibm,associativity")) { |
5d88aa85 JL |
1478 | u32 core_id; |
1479 | of_property_read_u32(update->dn, "reg", &core_id); | |
81b61324 | 1480 | rc = dlpar_cpu_readd(core_id); |
5d88aa85 JL |
1481 | rc = NOTIFY_OK; |
1482 | } | |
1483 | break; | |
1484 | } | |
1485 | ||
1486 | return rc; | |
9eff1a38 JL |
1487 | } |
1488 | ||
5d88aa85 JL |
1489 | static struct notifier_block dt_update_nb = { |
1490 | .notifier_call = dt_update_callback, | |
1491 | }; | |
1492 | ||
601abdc3 NF |
1493 | #endif |
1494 | ||
9eff1a38 | 1495 | /* |
5d88aa85 | 1496 | * Start polling for associativity changes. |
9eff1a38 JL |
1497 | */ |
1498 | int start_topology_update(void) | |
1499 | { | |
1500 | int rc = 0; | |
1501 | ||
2d4d9b30 NL |
1502 | if (!topology_updates_enabled) |
1503 | return 0; | |
1504 | ||
5d88aa85 JL |
1505 | if (firmware_has_feature(FW_FEATURE_PRRN)) { |
1506 | if (!prrn_enabled) { | |
1507 | prrn_enabled = 1; | |
601abdc3 | 1508 | #ifdef CONFIG_SMP |
5d88aa85 | 1509 | rc = of_reconfig_notifier_register(&dt_update_nb); |
601abdc3 | 1510 | #endif |
5d88aa85 | 1511 | } |
a3496e91 MB |
1512 | } |
1513 | if (firmware_has_feature(FW_FEATURE_VPHN) && | |
f13c13a0 | 1514 | lppaca_shared_proc(get_lppaca())) { |
5d88aa85 | 1515 | if (!vphn_enabled) { |
5d88aa85 JL |
1516 | vphn_enabled = 1; |
1517 | setup_cpu_associativity_change_counters(); | |
df7e828c KC |
1518 | timer_setup(&topology_timer, topology_timer_fn, |
1519 | TIMER_DEFERRABLE); | |
5d88aa85 JL |
1520 | reset_topology_timer(); |
1521 | } | |
9eff1a38 JL |
1522 | } |
1523 | ||
65b9fdad MB |
1524 | pr_info("Starting topology update%s%s\n", |
1525 | (prrn_enabled ? " prrn_enabled" : ""), | |
1526 | (vphn_enabled ? " vphn_enabled" : "")); | |
1527 | ||
9eff1a38 JL |
1528 | return rc; |
1529 | } | |
9eff1a38 JL |
1530 | |
1531 | /* | |
1532 | * Disable polling for VPHN associativity changes. | |
1533 | */ | |
1534 | int stop_topology_update(void) | |
1535 | { | |
5d88aa85 JL |
1536 | int rc = 0; |
1537 | ||
2d4d9b30 NL |
1538 | if (!topology_updates_enabled) |
1539 | return 0; | |
1540 | ||
5d88aa85 JL |
1541 | if (prrn_enabled) { |
1542 | prrn_enabled = 0; | |
601abdc3 | 1543 | #ifdef CONFIG_SMP |
5d88aa85 | 1544 | rc = of_reconfig_notifier_unregister(&dt_update_nb); |
601abdc3 | 1545 | #endif |
a3496e91 MB |
1546 | } |
1547 | if (vphn_enabled) { | |
5d88aa85 JL |
1548 | vphn_enabled = 0; |
1549 | rc = del_timer_sync(&topology_timer); | |
1550 | } | |
1551 | ||
65b9fdad MB |
1552 | pr_info("Stopping topology update\n"); |
1553 | ||
5d88aa85 | 1554 | return rc; |
9eff1a38 | 1555 | } |
e04fa612 NF |
1556 | |
1557 | int prrn_is_enabled(void) | |
1558 | { | |
1559 | return prrn_enabled; | |
1560 | } | |
1561 | ||
2ea62630 SD |
1562 | void __init shared_proc_topology_init(void) |
1563 | { | |
1564 | if (lppaca_shared_proc(get_lppaca())) { | |
1565 | bitmap_fill(cpumask_bits(&cpu_associativity_changes_mask), | |
1566 | nr_cpumask_bits); | |
1567 | numa_update_cpu_topology(false); | |
1568 | } | |
1569 | } | |
1570 | ||
e04fa612 NF |
1571 | static int topology_read(struct seq_file *file, void *v) |
1572 | { | |
1573 | if (vphn_enabled || prrn_enabled) | |
1574 | seq_puts(file, "on\n"); | |
1575 | else | |
1576 | seq_puts(file, "off\n"); | |
1577 | ||
1578 | return 0; | |
1579 | } | |
1580 | ||
1581 | static int topology_open(struct inode *inode, struct file *file) | |
1582 | { | |
1583 | return single_open(file, topology_read, NULL); | |
1584 | } | |
1585 | ||
1586 | static ssize_t topology_write(struct file *file, const char __user *buf, | |
1587 | size_t count, loff_t *off) | |
1588 | { | |
1589 | char kbuf[4]; /* "on" or "off" plus null. */ | |
1590 | int read_len; | |
1591 | ||
1592 | read_len = count < 3 ? count : 3; | |
1593 | if (copy_from_user(kbuf, buf, read_len)) | |
1594 | return -EINVAL; | |
1595 | ||
1596 | kbuf[read_len] = '\0'; | |
1597 | ||
2d4d9b30 NL |
1598 | if (!strncmp(kbuf, "on", 2)) { |
1599 | topology_updates_enabled = true; | |
e04fa612 | 1600 | start_topology_update(); |
2d4d9b30 | 1601 | } else if (!strncmp(kbuf, "off", 3)) { |
e04fa612 | 1602 | stop_topology_update(); |
2d4d9b30 NL |
1603 | topology_updates_enabled = false; |
1604 | } else | |
e04fa612 NF |
1605 | return -EINVAL; |
1606 | ||
1607 | return count; | |
1608 | } | |
1609 | ||
1610 | static const struct file_operations topology_ops = { | |
1611 | .read = seq_read, | |
1612 | .write = topology_write, | |
1613 | .open = topology_open, | |
1614 | .release = single_release | |
1615 | }; | |
1616 | ||
1617 | static int topology_update_init(void) | |
1618 | { | |
2d4d9b30 | 1619 | start_topology_update(); |
2d73bae1 | 1620 | |
17f444c0 MB |
1621 | if (vphn_enabled) |
1622 | topology_schedule_update(); | |
1623 | ||
2d15b9b4 NA |
1624 | if (!proc_create("powerpc/topology_updates", 0644, NULL, &topology_ops)) |
1625 | return -ENOMEM; | |
e04fa612 | 1626 | |
17f444c0 | 1627 | topology_inited = 1; |
e04fa612 | 1628 | return 0; |
9eff1a38 | 1629 | } |
e04fa612 | 1630 | device_initcall(topology_update_init); |
39bf990e | 1631 | #endif /* CONFIG_PPC_SPLPAR */ |