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 | */ | |
11 | #include <linux/threads.h> | |
12 | #include <linux/bootmem.h> | |
13 | #include <linux/init.h> | |
14 | #include <linux/mm.h> | |
15 | #include <linux/mmzone.h> | |
4b16f8e2 | 16 | #include <linux/export.h> |
1da177e4 LT |
17 | #include <linux/nodemask.h> |
18 | #include <linux/cpu.h> | |
19 | #include <linux/notifier.h> | |
95f72d1e | 20 | #include <linux/memblock.h> |
6df1646e | 21 | #include <linux/of.h> |
06eccea6 | 22 | #include <linux/pfn.h> |
9eff1a38 JL |
23 | #include <linux/cpuset.h> |
24 | #include <linux/node.h> | |
30c05350 | 25 | #include <linux/stop_machine.h> |
e04fa612 NF |
26 | #include <linux/proc_fs.h> |
27 | #include <linux/seq_file.h> | |
28 | #include <linux/uaccess.h> | |
9f3a90e8 | 29 | #include <linux/slab.h> |
45fb6cea | 30 | #include <asm/sparsemem.h> |
d9b2b2a2 | 31 | #include <asm/prom.h> |
2249ca9d | 32 | #include <asm/smp.h> |
9eff1a38 JL |
33 | #include <asm/firmware.h> |
34 | #include <asm/paca.h> | |
39bf990e | 35 | #include <asm/hvcall.h> |
ae3a197e | 36 | #include <asm/setup.h> |
176bbf14 | 37 | #include <asm/vdso.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; | |
60 | static const unsigned int *distance_ref_points; | |
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 | { | |
71 | unsigned int node, num = 0; | |
72 | ||
73 | /* setup nr_node_ids if not done yet */ | |
74 | if (nr_node_ids == MAX_NUMNODES) { | |
75 | for_each_node_mask(node, node_possible_map) | |
76 | num = node; | |
77 | nr_node_ids = num + 1; | |
78 | } | |
79 | ||
80 | /* allocate the map */ | |
81 | for (node = 0; node < nr_node_ids; node++) | |
82 | alloc_bootmem_cpumask_var(&node_to_cpumask_map[node]); | |
83 | ||
84 | /* cpumask_of_node() will now work */ | |
85 | dbg("Node to cpumask map for %d nodes\n", nr_node_ids); | |
86 | } | |
87 | ||
55671f3c | 88 | static int __init fake_numa_create_new_node(unsigned long end_pfn, |
1daa6d08 BS |
89 | unsigned int *nid) |
90 | { | |
91 | unsigned long long mem; | |
92 | char *p = cmdline; | |
93 | static unsigned int fake_nid; | |
94 | static unsigned long long curr_boundary; | |
95 | ||
96 | /* | |
97 | * Modify node id, iff we started creating NUMA nodes | |
98 | * We want to continue from where we left of the last time | |
99 | */ | |
100 | if (fake_nid) | |
101 | *nid = fake_nid; | |
102 | /* | |
103 | * In case there are no more arguments to parse, the | |
104 | * node_id should be the same as the last fake node id | |
105 | * (we've handled this above). | |
106 | */ | |
107 | if (!p) | |
108 | return 0; | |
109 | ||
110 | mem = memparse(p, &p); | |
111 | if (!mem) | |
112 | return 0; | |
113 | ||
114 | if (mem < curr_boundary) | |
115 | return 0; | |
116 | ||
117 | curr_boundary = mem; | |
118 | ||
119 | if ((end_pfn << PAGE_SHIFT) > mem) { | |
120 | /* | |
121 | * Skip commas and spaces | |
122 | */ | |
123 | while (*p == ',' || *p == ' ' || *p == '\t') | |
124 | p++; | |
125 | ||
126 | cmdline = p; | |
127 | fake_nid++; | |
128 | *nid = fake_nid; | |
129 | dbg("created new fake_node with id %d\n", fake_nid); | |
130 | return 1; | |
131 | } | |
132 | return 0; | |
133 | } | |
134 | ||
8f64e1f2 | 135 | /* |
5dfe8660 | 136 | * get_node_active_region - Return active region containing pfn |
e8170372 | 137 | * Active range returned is empty if none found. |
5dfe8660 TH |
138 | * @pfn: The page to return the region for |
139 | * @node_ar: Returned set to the active region containing @pfn | |
8f64e1f2 | 140 | */ |
5dfe8660 TH |
141 | static void __init get_node_active_region(unsigned long pfn, |
142 | struct node_active_region *node_ar) | |
8f64e1f2 | 143 | { |
5dfe8660 TH |
144 | unsigned long start_pfn, end_pfn; |
145 | int i, nid; | |
146 | ||
147 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { | |
148 | if (pfn >= start_pfn && pfn < end_pfn) { | |
149 | node_ar->nid = nid; | |
150 | node_ar->start_pfn = start_pfn; | |
151 | node_ar->end_pfn = end_pfn; | |
152 | break; | |
153 | } | |
154 | } | |
8f64e1f2 JT |
155 | } |
156 | ||
39bf990e | 157 | static void map_cpu_to_node(int cpu, int node) |
1da177e4 LT |
158 | { |
159 | numa_cpu_lookup_table[cpu] = node; | |
45fb6cea | 160 | |
bf4b85b0 NL |
161 | dbg("adding cpu %d to node %d\n", cpu, node); |
162 | ||
25863de0 AB |
163 | if (!(cpumask_test_cpu(cpu, node_to_cpumask_map[node]))) |
164 | cpumask_set_cpu(cpu, node_to_cpumask_map[node]); | |
1da177e4 LT |
165 | } |
166 | ||
39bf990e | 167 | #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_PPC_SPLPAR) |
1da177e4 LT |
168 | static void unmap_cpu_from_node(unsigned long cpu) |
169 | { | |
170 | int node = numa_cpu_lookup_table[cpu]; | |
171 | ||
172 | dbg("removing cpu %lu from node %d\n", cpu, node); | |
173 | ||
25863de0 | 174 | if (cpumask_test_cpu(cpu, node_to_cpumask_map[node])) { |
429f4d8d | 175 | cpumask_clear_cpu(cpu, node_to_cpumask_map[node]); |
1da177e4 LT |
176 | } else { |
177 | printk(KERN_ERR "WARNING: cpu %lu not found in node %d\n", | |
178 | cpu, node); | |
179 | } | |
180 | } | |
39bf990e | 181 | #endif /* CONFIG_HOTPLUG_CPU || CONFIG_PPC_SPLPAR */ |
1da177e4 | 182 | |
1da177e4 | 183 | /* must hold reference to node during call */ |
a7f67bdf | 184 | static const int *of_get_associativity(struct device_node *dev) |
1da177e4 | 185 | { |
e2eb6392 | 186 | return of_get_property(dev, "ibm,associativity", NULL); |
1da177e4 LT |
187 | } |
188 | ||
cf00085d C |
189 | /* |
190 | * Returns the property linux,drconf-usable-memory if | |
191 | * it exists (the property exists only in kexec/kdump kernels, | |
192 | * added by kexec-tools) | |
193 | */ | |
194 | static const u32 *of_get_usable_memory(struct device_node *memory) | |
195 | { | |
196 | const u32 *prop; | |
197 | u32 len; | |
198 | prop = of_get_property(memory, "linux,drconf-usable-memory", &len); | |
199 | if (!prop || len < sizeof(unsigned int)) | |
200 | return 0; | |
201 | return prop; | |
202 | } | |
203 | ||
41eab6f8 AB |
204 | int __node_distance(int a, int b) |
205 | { | |
206 | int i; | |
207 | int distance = LOCAL_DISTANCE; | |
208 | ||
209 | if (!form1_affinity) | |
7122beee | 210 | return ((a == b) ? LOCAL_DISTANCE : REMOTE_DISTANCE); |
41eab6f8 AB |
211 | |
212 | for (i = 0; i < distance_ref_points_depth; i++) { | |
213 | if (distance_lookup_table[a][i] == distance_lookup_table[b][i]) | |
214 | break; | |
215 | ||
216 | /* Double the distance for each NUMA level */ | |
217 | distance *= 2; | |
218 | } | |
219 | ||
220 | return distance; | |
221 | } | |
222 | ||
223 | static void initialize_distance_lookup_table(int nid, | |
224 | const unsigned int *associativity) | |
225 | { | |
226 | int i; | |
227 | ||
228 | if (!form1_affinity) | |
229 | return; | |
230 | ||
231 | for (i = 0; i < distance_ref_points_depth; i++) { | |
232 | distance_lookup_table[nid][i] = | |
233 | associativity[distance_ref_points[i]]; | |
234 | } | |
235 | } | |
236 | ||
482ec7c4 NL |
237 | /* Returns nid in the range [0..MAX_NUMNODES-1], or -1 if no useful numa |
238 | * info is found. | |
239 | */ | |
9eff1a38 | 240 | static int associativity_to_nid(const unsigned int *associativity) |
1da177e4 | 241 | { |
482ec7c4 | 242 | int nid = -1; |
1da177e4 LT |
243 | |
244 | if (min_common_depth == -1) | |
482ec7c4 | 245 | goto out; |
1da177e4 | 246 | |
9eff1a38 JL |
247 | if (associativity[0] >= min_common_depth) |
248 | nid = associativity[min_common_depth]; | |
bc16a759 NL |
249 | |
250 | /* POWER4 LPAR uses 0xffff as invalid node */ | |
482ec7c4 NL |
251 | if (nid == 0xffff || nid >= MAX_NUMNODES) |
252 | nid = -1; | |
41eab6f8 | 253 | |
9eff1a38 JL |
254 | if (nid > 0 && associativity[0] >= distance_ref_points_depth) |
255 | initialize_distance_lookup_table(nid, associativity); | |
41eab6f8 | 256 | |
482ec7c4 | 257 | out: |
cf950b7a | 258 | return nid; |
1da177e4 LT |
259 | } |
260 | ||
9eff1a38 JL |
261 | /* Returns the nid associated with the given device tree node, |
262 | * or -1 if not found. | |
263 | */ | |
264 | static int of_node_to_nid_single(struct device_node *device) | |
265 | { | |
266 | int nid = -1; | |
267 | const unsigned int *tmp; | |
268 | ||
269 | tmp = of_get_associativity(device); | |
270 | if (tmp) | |
271 | nid = associativity_to_nid(tmp); | |
272 | return nid; | |
273 | } | |
274 | ||
953039c8 JK |
275 | /* Walk the device tree upwards, looking for an associativity id */ |
276 | int of_node_to_nid(struct device_node *device) | |
277 | { | |
278 | struct device_node *tmp; | |
279 | int nid = -1; | |
280 | ||
281 | of_node_get(device); | |
282 | while (device) { | |
283 | nid = of_node_to_nid_single(device); | |
284 | if (nid != -1) | |
285 | break; | |
286 | ||
287 | tmp = device; | |
288 | device = of_get_parent(tmp); | |
289 | of_node_put(tmp); | |
290 | } | |
291 | of_node_put(device); | |
292 | ||
293 | return nid; | |
294 | } | |
295 | EXPORT_SYMBOL_GPL(of_node_to_nid); | |
296 | ||
1da177e4 LT |
297 | static int __init find_min_common_depth(void) |
298 | { | |
41eab6f8 | 299 | int depth; |
e70606eb | 300 | struct device_node *root; |
1da177e4 | 301 | |
1c8ee733 DS |
302 | if (firmware_has_feature(FW_FEATURE_OPAL)) |
303 | root = of_find_node_by_path("/ibm,opal"); | |
304 | else | |
305 | root = of_find_node_by_path("/rtas"); | |
e70606eb ME |
306 | if (!root) |
307 | root = of_find_node_by_path("/"); | |
1da177e4 LT |
308 | |
309 | /* | |
41eab6f8 AB |
310 | * This property is a set of 32-bit integers, each representing |
311 | * an index into the ibm,associativity nodes. | |
312 | * | |
313 | * With form 0 affinity the first integer is for an SMP configuration | |
314 | * (should be all 0's) and the second is for a normal NUMA | |
315 | * configuration. We have only one level of NUMA. | |
316 | * | |
317 | * With form 1 affinity the first integer is the most significant | |
318 | * NUMA boundary and the following are progressively less significant | |
319 | * boundaries. There can be more than one level of NUMA. | |
1da177e4 | 320 | */ |
e70606eb | 321 | distance_ref_points = of_get_property(root, |
41eab6f8 AB |
322 | "ibm,associativity-reference-points", |
323 | &distance_ref_points_depth); | |
324 | ||
325 | if (!distance_ref_points) { | |
326 | dbg("NUMA: ibm,associativity-reference-points not found.\n"); | |
327 | goto err; | |
328 | } | |
329 | ||
330 | distance_ref_points_depth /= sizeof(int); | |
1da177e4 | 331 | |
8002b0c5 NF |
332 | if (firmware_has_feature(FW_FEATURE_OPAL) || |
333 | firmware_has_feature(FW_FEATURE_TYPE1_AFFINITY)) { | |
334 | dbg("Using form 1 affinity\n"); | |
1c8ee733 | 335 | form1_affinity = 1; |
4b83c330 AB |
336 | } |
337 | ||
41eab6f8 AB |
338 | if (form1_affinity) { |
339 | depth = distance_ref_points[0]; | |
1da177e4 | 340 | } else { |
41eab6f8 AB |
341 | if (distance_ref_points_depth < 2) { |
342 | printk(KERN_WARNING "NUMA: " | |
343 | "short ibm,associativity-reference-points\n"); | |
344 | goto err; | |
345 | } | |
346 | ||
347 | depth = distance_ref_points[1]; | |
1da177e4 | 348 | } |
1da177e4 | 349 | |
41eab6f8 AB |
350 | /* |
351 | * Warn and cap if the hardware supports more than | |
352 | * MAX_DISTANCE_REF_POINTS domains. | |
353 | */ | |
354 | if (distance_ref_points_depth > MAX_DISTANCE_REF_POINTS) { | |
355 | printk(KERN_WARNING "NUMA: distance array capped at " | |
356 | "%d entries\n", MAX_DISTANCE_REF_POINTS); | |
357 | distance_ref_points_depth = MAX_DISTANCE_REF_POINTS; | |
358 | } | |
359 | ||
e70606eb | 360 | of_node_put(root); |
1da177e4 | 361 | return depth; |
41eab6f8 AB |
362 | |
363 | err: | |
e70606eb | 364 | of_node_put(root); |
41eab6f8 | 365 | return -1; |
1da177e4 LT |
366 | } |
367 | ||
84c9fdd1 | 368 | static void __init get_n_mem_cells(int *n_addr_cells, int *n_size_cells) |
1da177e4 LT |
369 | { |
370 | struct device_node *memory = NULL; | |
1da177e4 LT |
371 | |
372 | memory = of_find_node_by_type(memory, "memory"); | |
54c23310 | 373 | if (!memory) |
84c9fdd1 | 374 | panic("numa.c: No memory nodes found!"); |
54c23310 | 375 | |
a8bda5dd | 376 | *n_addr_cells = of_n_addr_cells(memory); |
9213feea | 377 | *n_size_cells = of_n_size_cells(memory); |
84c9fdd1 | 378 | of_node_put(memory); |
1da177e4 LT |
379 | } |
380 | ||
2011b1d0 | 381 | static unsigned long read_n_cells(int n, const unsigned int **buf) |
1da177e4 LT |
382 | { |
383 | unsigned long result = 0; | |
384 | ||
385 | while (n--) { | |
386 | result = (result << 32) | **buf; | |
387 | (*buf)++; | |
388 | } | |
389 | return result; | |
390 | } | |
391 | ||
8342681d | 392 | /* |
95f72d1e | 393 | * Read the next memblock list entry from the ibm,dynamic-memory property |
8342681d NF |
394 | * and return the information in the provided of_drconf_cell structure. |
395 | */ | |
396 | static void read_drconf_cell(struct of_drconf_cell *drmem, const u32 **cellp) | |
397 | { | |
398 | const u32 *cp; | |
399 | ||
400 | drmem->base_addr = read_n_cells(n_mem_addr_cells, cellp); | |
401 | ||
402 | cp = *cellp; | |
403 | drmem->drc_index = cp[0]; | |
404 | drmem->reserved = cp[1]; | |
405 | drmem->aa_index = cp[2]; | |
406 | drmem->flags = cp[3]; | |
407 | ||
408 | *cellp = cp + 4; | |
409 | } | |
410 | ||
411 | /* | |
25985edc | 412 | * Retrieve and validate the ibm,dynamic-memory property of the device tree. |
8342681d | 413 | * |
95f72d1e YL |
414 | * The layout of the ibm,dynamic-memory property is a number N of memblock |
415 | * list entries followed by N memblock list entries. Each memblock list entry | |
25985edc | 416 | * contains information as laid out in the of_drconf_cell struct above. |
8342681d NF |
417 | */ |
418 | static int of_get_drconf_memory(struct device_node *memory, const u32 **dm) | |
419 | { | |
420 | const u32 *prop; | |
421 | u32 len, entries; | |
422 | ||
423 | prop = of_get_property(memory, "ibm,dynamic-memory", &len); | |
424 | if (!prop || len < sizeof(unsigned int)) | |
425 | return 0; | |
426 | ||
427 | entries = *prop++; | |
428 | ||
429 | /* Now that we know the number of entries, revalidate the size | |
430 | * of the property read in to ensure we have everything | |
431 | */ | |
432 | if (len < (entries * (n_mem_addr_cells + 4) + 1) * sizeof(unsigned int)) | |
433 | return 0; | |
434 | ||
435 | *dm = prop; | |
436 | return entries; | |
437 | } | |
438 | ||
439 | /* | |
25985edc | 440 | * Retrieve and validate the ibm,lmb-size property for drconf memory |
8342681d NF |
441 | * from the device tree. |
442 | */ | |
3fdfd990 | 443 | static u64 of_get_lmb_size(struct device_node *memory) |
8342681d NF |
444 | { |
445 | const u32 *prop; | |
446 | u32 len; | |
447 | ||
3fdfd990 | 448 | prop = of_get_property(memory, "ibm,lmb-size", &len); |
8342681d NF |
449 | if (!prop || len < sizeof(unsigned int)) |
450 | return 0; | |
451 | ||
452 | return read_n_cells(n_mem_size_cells, &prop); | |
453 | } | |
454 | ||
455 | struct assoc_arrays { | |
456 | u32 n_arrays; | |
457 | u32 array_sz; | |
458 | const u32 *arrays; | |
459 | }; | |
460 | ||
461 | /* | |
25985edc | 462 | * Retrieve and validate the list of associativity arrays for drconf |
8342681d NF |
463 | * memory from the ibm,associativity-lookup-arrays property of the |
464 | * device tree.. | |
465 | * | |
466 | * The layout of the ibm,associativity-lookup-arrays property is a number N | |
467 | * indicating the number of associativity arrays, followed by a number M | |
468 | * indicating the size of each associativity array, followed by a list | |
469 | * of N associativity arrays. | |
470 | */ | |
471 | static int of_get_assoc_arrays(struct device_node *memory, | |
472 | struct assoc_arrays *aa) | |
473 | { | |
474 | const u32 *prop; | |
475 | u32 len; | |
476 | ||
477 | prop = of_get_property(memory, "ibm,associativity-lookup-arrays", &len); | |
478 | if (!prop || len < 2 * sizeof(unsigned int)) | |
479 | return -1; | |
480 | ||
481 | aa->n_arrays = *prop++; | |
482 | aa->array_sz = *prop++; | |
483 | ||
42b2aa86 | 484 | /* Now that we know the number of arrays and size of each array, |
8342681d NF |
485 | * revalidate the size of the property read in. |
486 | */ | |
487 | if (len < (aa->n_arrays * aa->array_sz + 2) * sizeof(unsigned int)) | |
488 | return -1; | |
489 | ||
490 | aa->arrays = prop; | |
491 | return 0; | |
492 | } | |
493 | ||
494 | /* | |
495 | * This is like of_node_to_nid_single() for memory represented in the | |
496 | * ibm,dynamic-reconfiguration-memory node. | |
497 | */ | |
498 | static int of_drconf_to_nid_single(struct of_drconf_cell *drmem, | |
499 | struct assoc_arrays *aa) | |
500 | { | |
501 | int default_nid = 0; | |
502 | int nid = default_nid; | |
503 | int index; | |
504 | ||
505 | if (min_common_depth > 0 && min_common_depth <= aa->array_sz && | |
506 | !(drmem->flags & DRCONF_MEM_AI_INVALID) && | |
507 | drmem->aa_index < aa->n_arrays) { | |
508 | index = drmem->aa_index * aa->array_sz + min_common_depth - 1; | |
509 | nid = aa->arrays[index]; | |
510 | ||
511 | if (nid == 0xffff || nid >= MAX_NUMNODES) | |
512 | nid = default_nid; | |
513 | } | |
514 | ||
515 | return nid; | |
516 | } | |
517 | ||
1da177e4 LT |
518 | /* |
519 | * Figure out to which domain a cpu belongs and stick it there. | |
520 | * Return the id of the domain used. | |
521 | */ | |
2e5ce39d | 522 | static int __cpuinit numa_setup_cpu(unsigned long lcpu) |
1da177e4 | 523 | { |
cf950b7a | 524 | int nid = 0; |
8b16cd23 | 525 | struct device_node *cpu = of_get_cpu_node(lcpu, NULL); |
1da177e4 LT |
526 | |
527 | if (!cpu) { | |
528 | WARN_ON(1); | |
529 | goto out; | |
530 | } | |
531 | ||
953039c8 | 532 | nid = of_node_to_nid_single(cpu); |
1da177e4 | 533 | |
482ec7c4 | 534 | if (nid < 0 || !node_online(nid)) |
72c33688 | 535 | nid = first_online_node; |
1da177e4 | 536 | out: |
cf950b7a | 537 | map_cpu_to_node(lcpu, nid); |
1da177e4 LT |
538 | |
539 | of_node_put(cpu); | |
540 | ||
cf950b7a | 541 | return nid; |
1da177e4 LT |
542 | } |
543 | ||
74b85f37 | 544 | static int __cpuinit cpu_numa_callback(struct notifier_block *nfb, |
1da177e4 LT |
545 | unsigned long action, |
546 | void *hcpu) | |
547 | { | |
548 | unsigned long lcpu = (unsigned long)hcpu; | |
549 | int ret = NOTIFY_DONE; | |
550 | ||
551 | switch (action) { | |
552 | case CPU_UP_PREPARE: | |
8bb78442 | 553 | case CPU_UP_PREPARE_FROZEN: |
2b261227 | 554 | numa_setup_cpu(lcpu); |
1da177e4 LT |
555 | ret = NOTIFY_OK; |
556 | break; | |
557 | #ifdef CONFIG_HOTPLUG_CPU | |
558 | case CPU_DEAD: | |
8bb78442 | 559 | case CPU_DEAD_FROZEN: |
1da177e4 | 560 | case CPU_UP_CANCELED: |
8bb78442 | 561 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 LT |
562 | unmap_cpu_from_node(lcpu); |
563 | break; | |
564 | ret = NOTIFY_OK; | |
565 | #endif | |
566 | } | |
567 | return ret; | |
568 | } | |
569 | ||
570 | /* | |
571 | * Check and possibly modify a memory region to enforce the memory limit. | |
572 | * | |
573 | * Returns the size the region should have to enforce the memory limit. | |
574 | * This will either be the original value of size, a truncated value, | |
575 | * or zero. If the returned value of size is 0 the region should be | |
25985edc | 576 | * discarded as it lies wholly above the memory limit. |
1da177e4 | 577 | */ |
45fb6cea AB |
578 | static unsigned long __init numa_enforce_memory_limit(unsigned long start, |
579 | unsigned long size) | |
1da177e4 LT |
580 | { |
581 | /* | |
95f72d1e | 582 | * We use memblock_end_of_DRAM() in here instead of memory_limit because |
1da177e4 | 583 | * we've already adjusted it for the limit and it takes care of |
fe55249d MM |
584 | * having memory holes below the limit. Also, in the case of |
585 | * iommu_is_off, memory_limit is not set but is implicitly enforced. | |
1da177e4 | 586 | */ |
1da177e4 | 587 | |
95f72d1e | 588 | if (start + size <= memblock_end_of_DRAM()) |
1da177e4 LT |
589 | return size; |
590 | ||
95f72d1e | 591 | if (start >= memblock_end_of_DRAM()) |
1da177e4 LT |
592 | return 0; |
593 | ||
95f72d1e | 594 | return memblock_end_of_DRAM() - start; |
1da177e4 LT |
595 | } |
596 | ||
cf00085d C |
597 | /* |
598 | * Reads the counter for a given entry in | |
599 | * linux,drconf-usable-memory property | |
600 | */ | |
601 | static inline int __init read_usm_ranges(const u32 **usm) | |
602 | { | |
603 | /* | |
3fdfd990 | 604 | * For each lmb in ibm,dynamic-memory a corresponding |
cf00085d C |
605 | * entry in linux,drconf-usable-memory property contains |
606 | * a counter followed by that many (base, size) duple. | |
607 | * read the counter from linux,drconf-usable-memory | |
608 | */ | |
609 | return read_n_cells(n_mem_size_cells, usm); | |
610 | } | |
611 | ||
0204568a PM |
612 | /* |
613 | * Extract NUMA information from the ibm,dynamic-reconfiguration-memory | |
614 | * node. This assumes n_mem_{addr,size}_cells have been set. | |
615 | */ | |
616 | static void __init parse_drconf_memory(struct device_node *memory) | |
617 | { | |
82b2521d | 618 | const u32 *uninitialized_var(dm), *usm; |
cf00085d | 619 | unsigned int n, rc, ranges, is_kexec_kdump = 0; |
3fdfd990 | 620 | unsigned long lmb_size, base, size, sz; |
8342681d | 621 | int nid; |
aa709f3b | 622 | struct assoc_arrays aa = { .arrays = NULL }; |
8342681d NF |
623 | |
624 | n = of_get_drconf_memory(memory, &dm); | |
625 | if (!n) | |
0204568a PM |
626 | return; |
627 | ||
3fdfd990 BH |
628 | lmb_size = of_get_lmb_size(memory); |
629 | if (!lmb_size) | |
8342681d NF |
630 | return; |
631 | ||
632 | rc = of_get_assoc_arrays(memory, &aa); | |
633 | if (rc) | |
0204568a PM |
634 | return; |
635 | ||
cf00085d C |
636 | /* check if this is a kexec/kdump kernel */ |
637 | usm = of_get_usable_memory(memory); | |
638 | if (usm != NULL) | |
639 | is_kexec_kdump = 1; | |
640 | ||
0204568a | 641 | for (; n != 0; --n) { |
8342681d NF |
642 | struct of_drconf_cell drmem; |
643 | ||
644 | read_drconf_cell(&drmem, &dm); | |
645 | ||
646 | /* skip this block if the reserved bit is set in flags (0x80) | |
647 | or if the block is not assigned to this partition (0x8) */ | |
648 | if ((drmem.flags & DRCONF_MEM_RESERVED) | |
649 | || !(drmem.flags & DRCONF_MEM_ASSIGNED)) | |
0204568a | 650 | continue; |
1daa6d08 | 651 | |
cf00085d | 652 | base = drmem.base_addr; |
3fdfd990 | 653 | size = lmb_size; |
cf00085d | 654 | ranges = 1; |
8342681d | 655 | |
cf00085d C |
656 | if (is_kexec_kdump) { |
657 | ranges = read_usm_ranges(&usm); | |
658 | if (!ranges) /* there are no (base, size) duple */ | |
659 | continue; | |
660 | } | |
661 | do { | |
662 | if (is_kexec_kdump) { | |
663 | base = read_n_cells(n_mem_addr_cells, &usm); | |
664 | size = read_n_cells(n_mem_size_cells, &usm); | |
665 | } | |
666 | nid = of_drconf_to_nid_single(&drmem, &aa); | |
667 | fake_numa_create_new_node( | |
668 | ((base + size) >> PAGE_SHIFT), | |
8342681d | 669 | &nid); |
cf00085d C |
670 | node_set_online(nid); |
671 | sz = numa_enforce_memory_limit(base, size); | |
672 | if (sz) | |
1d7cfe18 | 673 | memblock_set_node(base, sz, nid); |
cf00085d | 674 | } while (--ranges); |
0204568a PM |
675 | } |
676 | } | |
677 | ||
1da177e4 LT |
678 | static int __init parse_numa_properties(void) |
679 | { | |
94db7c5e | 680 | struct device_node *memory; |
482ec7c4 | 681 | int default_nid = 0; |
1da177e4 LT |
682 | unsigned long i; |
683 | ||
684 | if (numa_enabled == 0) { | |
685 | printk(KERN_WARNING "NUMA disabled by user\n"); | |
686 | return -1; | |
687 | } | |
688 | ||
1da177e4 LT |
689 | min_common_depth = find_min_common_depth(); |
690 | ||
1da177e4 LT |
691 | if (min_common_depth < 0) |
692 | return min_common_depth; | |
693 | ||
bf4b85b0 NL |
694 | dbg("NUMA associativity depth for CPU/Memory: %d\n", min_common_depth); |
695 | ||
1da177e4 | 696 | /* |
482ec7c4 NL |
697 | * Even though we connect cpus to numa domains later in SMP |
698 | * init, we need to know the node ids now. This is because | |
699 | * each node to be onlined must have NODE_DATA etc backing it. | |
1da177e4 | 700 | */ |
482ec7c4 | 701 | for_each_present_cpu(i) { |
dfbe93a2 | 702 | struct device_node *cpu; |
cf950b7a | 703 | int nid; |
1da177e4 | 704 | |
8b16cd23 | 705 | cpu = of_get_cpu_node(i, NULL); |
482ec7c4 | 706 | BUG_ON(!cpu); |
953039c8 | 707 | nid = of_node_to_nid_single(cpu); |
482ec7c4 | 708 | of_node_put(cpu); |
1da177e4 | 709 | |
482ec7c4 NL |
710 | /* |
711 | * Don't fall back to default_nid yet -- we will plug | |
712 | * cpus into nodes once the memory scan has discovered | |
713 | * the topology. | |
714 | */ | |
715 | if (nid < 0) | |
716 | continue; | |
717 | node_set_online(nid); | |
1da177e4 LT |
718 | } |
719 | ||
237a0989 | 720 | get_n_mem_cells(&n_mem_addr_cells, &n_mem_size_cells); |
94db7c5e AB |
721 | |
722 | for_each_node_by_type(memory, "memory") { | |
1da177e4 LT |
723 | unsigned long start; |
724 | unsigned long size; | |
cf950b7a | 725 | int nid; |
1da177e4 | 726 | int ranges; |
a7f67bdf | 727 | const unsigned int *memcell_buf; |
1da177e4 LT |
728 | unsigned int len; |
729 | ||
e2eb6392 | 730 | memcell_buf = of_get_property(memory, |
ba759485 ME |
731 | "linux,usable-memory", &len); |
732 | if (!memcell_buf || len <= 0) | |
e2eb6392 | 733 | memcell_buf = of_get_property(memory, "reg", &len); |
1da177e4 LT |
734 | if (!memcell_buf || len <= 0) |
735 | continue; | |
736 | ||
cc5d0189 BH |
737 | /* ranges in cell */ |
738 | ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells); | |
1da177e4 LT |
739 | new_range: |
740 | /* these are order-sensitive, and modify the buffer pointer */ | |
237a0989 MK |
741 | start = read_n_cells(n_mem_addr_cells, &memcell_buf); |
742 | size = read_n_cells(n_mem_size_cells, &memcell_buf); | |
1da177e4 | 743 | |
482ec7c4 NL |
744 | /* |
745 | * Assumption: either all memory nodes or none will | |
746 | * have associativity properties. If none, then | |
747 | * everything goes to default_nid. | |
748 | */ | |
953039c8 | 749 | nid = of_node_to_nid_single(memory); |
482ec7c4 NL |
750 | if (nid < 0) |
751 | nid = default_nid; | |
1daa6d08 BS |
752 | |
753 | fake_numa_create_new_node(((start + size) >> PAGE_SHIFT), &nid); | |
482ec7c4 | 754 | node_set_online(nid); |
1da177e4 | 755 | |
45fb6cea | 756 | if (!(size = numa_enforce_memory_limit(start, size))) { |
1da177e4 LT |
757 | if (--ranges) |
758 | goto new_range; | |
759 | else | |
760 | continue; | |
761 | } | |
762 | ||
1d7cfe18 | 763 | memblock_set_node(start, size, nid); |
1da177e4 LT |
764 | |
765 | if (--ranges) | |
766 | goto new_range; | |
767 | } | |
768 | ||
0204568a | 769 | /* |
dfbe93a2 AB |
770 | * Now do the same thing for each MEMBLOCK listed in the |
771 | * ibm,dynamic-memory property in the | |
772 | * ibm,dynamic-reconfiguration-memory node. | |
0204568a PM |
773 | */ |
774 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); | |
775 | if (memory) | |
776 | parse_drconf_memory(memory); | |
777 | ||
1da177e4 LT |
778 | return 0; |
779 | } | |
780 | ||
781 | static void __init setup_nonnuma(void) | |
782 | { | |
95f72d1e YL |
783 | unsigned long top_of_ram = memblock_end_of_DRAM(); |
784 | unsigned long total_ram = memblock_phys_mem_size(); | |
c67c3cb4 | 785 | unsigned long start_pfn, end_pfn; |
28be7072 BH |
786 | unsigned int nid = 0; |
787 | struct memblock_region *reg; | |
1da177e4 | 788 | |
e110b281 | 789 | printk(KERN_DEBUG "Top of RAM: 0x%lx, Total RAM: 0x%lx\n", |
1da177e4 | 790 | top_of_ram, total_ram); |
e110b281 | 791 | printk(KERN_DEBUG "Memory hole size: %ldMB\n", |
1da177e4 LT |
792 | (top_of_ram - total_ram) >> 20); |
793 | ||
28be7072 | 794 | for_each_memblock(memory, reg) { |
c7fc2de0 YL |
795 | start_pfn = memblock_region_memory_base_pfn(reg); |
796 | end_pfn = memblock_region_memory_end_pfn(reg); | |
1daa6d08 BS |
797 | |
798 | fake_numa_create_new_node(end_pfn, &nid); | |
1d7cfe18 TH |
799 | memblock_set_node(PFN_PHYS(start_pfn), |
800 | PFN_PHYS(end_pfn - start_pfn), nid); | |
1daa6d08 | 801 | node_set_online(nid); |
c67c3cb4 | 802 | } |
1da177e4 LT |
803 | } |
804 | ||
4b703a23 AB |
805 | void __init dump_numa_cpu_topology(void) |
806 | { | |
807 | unsigned int node; | |
808 | unsigned int cpu, count; | |
809 | ||
810 | if (min_common_depth == -1 || !numa_enabled) | |
811 | return; | |
812 | ||
813 | for_each_online_node(node) { | |
e110b281 | 814 | printk(KERN_DEBUG "Node %d CPUs:", node); |
4b703a23 AB |
815 | |
816 | count = 0; | |
817 | /* | |
818 | * If we used a CPU iterator here we would miss printing | |
819 | * the holes in the cpumap. | |
820 | */ | |
25863de0 AB |
821 | for (cpu = 0; cpu < nr_cpu_ids; cpu++) { |
822 | if (cpumask_test_cpu(cpu, | |
823 | node_to_cpumask_map[node])) { | |
4b703a23 AB |
824 | if (count == 0) |
825 | printk(" %u", cpu); | |
826 | ++count; | |
827 | } else { | |
828 | if (count > 1) | |
829 | printk("-%u", cpu - 1); | |
830 | count = 0; | |
831 | } | |
832 | } | |
833 | ||
834 | if (count > 1) | |
25863de0 | 835 | printk("-%u", nr_cpu_ids - 1); |
4b703a23 AB |
836 | printk("\n"); |
837 | } | |
838 | } | |
839 | ||
840 | static void __init dump_numa_memory_topology(void) | |
1da177e4 LT |
841 | { |
842 | unsigned int node; | |
843 | unsigned int count; | |
844 | ||
845 | if (min_common_depth == -1 || !numa_enabled) | |
846 | return; | |
847 | ||
848 | for_each_online_node(node) { | |
849 | unsigned long i; | |
850 | ||
e110b281 | 851 | printk(KERN_DEBUG "Node %d Memory:", node); |
1da177e4 LT |
852 | |
853 | count = 0; | |
854 | ||
95f72d1e | 855 | for (i = 0; i < memblock_end_of_DRAM(); |
45fb6cea AB |
856 | i += (1 << SECTION_SIZE_BITS)) { |
857 | if (early_pfn_to_nid(i >> PAGE_SHIFT) == node) { | |
1da177e4 LT |
858 | if (count == 0) |
859 | printk(" 0x%lx", i); | |
860 | ++count; | |
861 | } else { | |
862 | if (count > 0) | |
863 | printk("-0x%lx", i); | |
864 | count = 0; | |
865 | } | |
866 | } | |
867 | ||
868 | if (count > 0) | |
869 | printk("-0x%lx", i); | |
870 | printk("\n"); | |
871 | } | |
1da177e4 LT |
872 | } |
873 | ||
874 | /* | |
95f72d1e | 875 | * Allocate some memory, satisfying the memblock or bootmem allocator where |
1da177e4 LT |
876 | * required. nid is the preferred node and end is the physical address of |
877 | * the highest address in the node. | |
878 | * | |
0be210fd | 879 | * Returns the virtual address of the memory. |
1da177e4 | 880 | */ |
893473df | 881 | static void __init *careful_zallocation(int nid, unsigned long size, |
45fb6cea AB |
882 | unsigned long align, |
883 | unsigned long end_pfn) | |
1da177e4 | 884 | { |
0be210fd | 885 | void *ret; |
45fb6cea | 886 | int new_nid; |
0be210fd DH |
887 | unsigned long ret_paddr; |
888 | ||
95f72d1e | 889 | ret_paddr = __memblock_alloc_base(size, align, end_pfn << PAGE_SHIFT); |
1da177e4 LT |
890 | |
891 | /* retry over all memory */ | |
0be210fd | 892 | if (!ret_paddr) |
95f72d1e | 893 | ret_paddr = __memblock_alloc_base(size, align, memblock_end_of_DRAM()); |
1da177e4 | 894 | |
0be210fd | 895 | if (!ret_paddr) |
5d21ea2b | 896 | panic("numa.c: cannot allocate %lu bytes for node %d", |
1da177e4 LT |
897 | size, nid); |
898 | ||
0be210fd DH |
899 | ret = __va(ret_paddr); |
900 | ||
1da177e4 | 901 | /* |
c555e520 | 902 | * We initialize the nodes in numeric order: 0, 1, 2... |
95f72d1e | 903 | * and hand over control from the MEMBLOCK allocator to the |
c555e520 DH |
904 | * bootmem allocator. If this function is called for |
905 | * node 5, then we know that all nodes <5 are using the | |
95f72d1e | 906 | * bootmem allocator instead of the MEMBLOCK allocator. |
c555e520 DH |
907 | * |
908 | * So, check the nid from which this allocation came | |
909 | * and double check to see if we need to use bootmem | |
95f72d1e | 910 | * instead of the MEMBLOCK. We don't free the MEMBLOCK memory |
c555e520 | 911 | * since it would be useless. |
1da177e4 | 912 | */ |
0be210fd | 913 | new_nid = early_pfn_to_nid(ret_paddr >> PAGE_SHIFT); |
45fb6cea | 914 | if (new_nid < nid) { |
0be210fd | 915 | ret = __alloc_bootmem_node(NODE_DATA(new_nid), |
1da177e4 LT |
916 | size, align, 0); |
917 | ||
0be210fd | 918 | dbg("alloc_bootmem %p %lx\n", ret, size); |
1da177e4 LT |
919 | } |
920 | ||
893473df | 921 | memset(ret, 0, size); |
0be210fd | 922 | return ret; |
1da177e4 LT |
923 | } |
924 | ||
74b85f37 CS |
925 | static struct notifier_block __cpuinitdata ppc64_numa_nb = { |
926 | .notifier_call = cpu_numa_callback, | |
927 | .priority = 1 /* Must run before sched domains notifier. */ | |
928 | }; | |
929 | ||
28e86bdb | 930 | static void __init mark_reserved_regions_for_nid(int nid) |
4a618669 DH |
931 | { |
932 | struct pglist_data *node = NODE_DATA(nid); | |
28be7072 | 933 | struct memblock_region *reg; |
4a618669 | 934 | |
28be7072 BH |
935 | for_each_memblock(reserved, reg) { |
936 | unsigned long physbase = reg->base; | |
937 | unsigned long size = reg->size; | |
4a618669 | 938 | unsigned long start_pfn = physbase >> PAGE_SHIFT; |
06eccea6 | 939 | unsigned long end_pfn = PFN_UP(physbase + size); |
4a618669 DH |
940 | struct node_active_region node_ar; |
941 | unsigned long node_end_pfn = node->node_start_pfn + | |
942 | node->node_spanned_pages; | |
943 | ||
944 | /* | |
95f72d1e | 945 | * Check to make sure that this memblock.reserved area is |
4a618669 DH |
946 | * within the bounds of the node that we care about. |
947 | * Checking the nid of the start and end points is not | |
948 | * sufficient because the reserved area could span the | |
949 | * entire node. | |
950 | */ | |
951 | if (end_pfn <= node->node_start_pfn || | |
952 | start_pfn >= node_end_pfn) | |
953 | continue; | |
954 | ||
955 | get_node_active_region(start_pfn, &node_ar); | |
956 | while (start_pfn < end_pfn && | |
957 | node_ar.start_pfn < node_ar.end_pfn) { | |
958 | unsigned long reserve_size = size; | |
959 | /* | |
960 | * if reserved region extends past active region | |
961 | * then trim size to active region | |
962 | */ | |
963 | if (end_pfn > node_ar.end_pfn) | |
964 | reserve_size = (node_ar.end_pfn << PAGE_SHIFT) | |
06eccea6 | 965 | - physbase; |
a4c74ddd DH |
966 | /* |
967 | * Only worry about *this* node, others may not | |
968 | * yet have valid NODE_DATA(). | |
969 | */ | |
970 | if (node_ar.nid == nid) { | |
971 | dbg("reserve_bootmem %lx %lx nid=%d\n", | |
972 | physbase, reserve_size, node_ar.nid); | |
973 | reserve_bootmem_node(NODE_DATA(node_ar.nid), | |
974 | physbase, reserve_size, | |
975 | BOOTMEM_DEFAULT); | |
976 | } | |
4a618669 DH |
977 | /* |
978 | * if reserved region is contained in the active region | |
979 | * then done. | |
980 | */ | |
981 | if (end_pfn <= node_ar.end_pfn) | |
982 | break; | |
983 | ||
984 | /* | |
985 | * reserved region extends past the active region | |
986 | * get next active region that contains this | |
987 | * reserved region | |
988 | */ | |
989 | start_pfn = node_ar.end_pfn; | |
990 | physbase = start_pfn << PAGE_SHIFT; | |
991 | size = size - reserve_size; | |
992 | get_node_active_region(start_pfn, &node_ar); | |
993 | } | |
994 | } | |
995 | } | |
996 | ||
997 | ||
1da177e4 LT |
998 | void __init do_init_bootmem(void) |
999 | { | |
1000 | int nid; | |
1da177e4 LT |
1001 | |
1002 | min_low_pfn = 0; | |
95f72d1e | 1003 | max_low_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT; |
1da177e4 LT |
1004 | max_pfn = max_low_pfn; |
1005 | ||
1006 | if (parse_numa_properties()) | |
1007 | setup_nonnuma(); | |
1008 | else | |
4b703a23 | 1009 | dump_numa_memory_topology(); |
1da177e4 | 1010 | |
1da177e4 | 1011 | for_each_online_node(nid) { |
c67c3cb4 | 1012 | unsigned long start_pfn, end_pfn; |
0be210fd | 1013 | void *bootmem_vaddr; |
1da177e4 LT |
1014 | unsigned long bootmap_pages; |
1015 | ||
c67c3cb4 | 1016 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); |
1da177e4 | 1017 | |
4a618669 DH |
1018 | /* |
1019 | * Allocate the node structure node local if possible | |
1020 | * | |
1021 | * Be careful moving this around, as it relies on all | |
1022 | * previous nodes' bootmem to be initialized and have | |
1023 | * all reserved areas marked. | |
1024 | */ | |
893473df | 1025 | NODE_DATA(nid) = careful_zallocation(nid, |
1da177e4 | 1026 | sizeof(struct pglist_data), |
45fb6cea | 1027 | SMP_CACHE_BYTES, end_pfn); |
1da177e4 LT |
1028 | |
1029 | dbg("node %d\n", nid); | |
1030 | dbg("NODE_DATA() = %p\n", NODE_DATA(nid)); | |
1031 | ||
b61bfa3c | 1032 | NODE_DATA(nid)->bdata = &bootmem_node_data[nid]; |
45fb6cea AB |
1033 | NODE_DATA(nid)->node_start_pfn = start_pfn; |
1034 | NODE_DATA(nid)->node_spanned_pages = end_pfn - start_pfn; | |
1da177e4 LT |
1035 | |
1036 | if (NODE_DATA(nid)->node_spanned_pages == 0) | |
1037 | continue; | |
1038 | ||
45fb6cea AB |
1039 | dbg("start_paddr = %lx\n", start_pfn << PAGE_SHIFT); |
1040 | dbg("end_paddr = %lx\n", end_pfn << PAGE_SHIFT); | |
1da177e4 | 1041 | |
45fb6cea | 1042 | bootmap_pages = bootmem_bootmap_pages(end_pfn - start_pfn); |
893473df | 1043 | bootmem_vaddr = careful_zallocation(nid, |
45fb6cea AB |
1044 | bootmap_pages << PAGE_SHIFT, |
1045 | PAGE_SIZE, end_pfn); | |
1da177e4 | 1046 | |
0be210fd | 1047 | dbg("bootmap_vaddr = %p\n", bootmem_vaddr); |
1da177e4 | 1048 | |
0be210fd DH |
1049 | init_bootmem_node(NODE_DATA(nid), |
1050 | __pa(bootmem_vaddr) >> PAGE_SHIFT, | |
45fb6cea | 1051 | start_pfn, end_pfn); |
1da177e4 | 1052 | |
c67c3cb4 | 1053 | free_bootmem_with_active_regions(nid, end_pfn); |
4a618669 DH |
1054 | /* |
1055 | * Be very careful about moving this around. Future | |
893473df | 1056 | * calls to careful_zallocation() depend on this getting |
4a618669 DH |
1057 | * done correctly. |
1058 | */ | |
1059 | mark_reserved_regions_for_nid(nid); | |
8f64e1f2 | 1060 | sparse_memory_present_with_active_regions(nid); |
4a618669 | 1061 | } |
d3f6204a BH |
1062 | |
1063 | init_bootmem_done = 1; | |
25863de0 AB |
1064 | |
1065 | /* | |
1066 | * Now bootmem is initialised we can create the node to cpumask | |
1067 | * lookup tables and setup the cpu callback to populate them. | |
1068 | */ | |
1069 | setup_node_to_cpumask_map(); | |
1070 | ||
1071 | register_cpu_notifier(&ppc64_numa_nb); | |
1072 | cpu_numa_callback(&ppc64_numa_nb, CPU_UP_PREPARE, | |
1073 | (void *)(unsigned long)boot_cpuid); | |
1da177e4 LT |
1074 | } |
1075 | ||
1076 | void __init paging_init(void) | |
1077 | { | |
6391af17 MG |
1078 | unsigned long max_zone_pfns[MAX_NR_ZONES]; |
1079 | memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); | |
95f72d1e | 1080 | max_zone_pfns[ZONE_DMA] = memblock_end_of_DRAM() >> PAGE_SHIFT; |
c67c3cb4 | 1081 | free_area_init_nodes(max_zone_pfns); |
1da177e4 LT |
1082 | } |
1083 | ||
1084 | static int __init early_numa(char *p) | |
1085 | { | |
1086 | if (!p) | |
1087 | return 0; | |
1088 | ||
1089 | if (strstr(p, "off")) | |
1090 | numa_enabled = 0; | |
1091 | ||
1092 | if (strstr(p, "debug")) | |
1093 | numa_debug = 1; | |
1094 | ||
1daa6d08 BS |
1095 | p = strstr(p, "fake="); |
1096 | if (p) | |
1097 | cmdline = p + strlen("fake="); | |
1098 | ||
1da177e4 LT |
1099 | return 0; |
1100 | } | |
1101 | early_param("numa", early_numa); | |
237a0989 MK |
1102 | |
1103 | #ifdef CONFIG_MEMORY_HOTPLUG | |
0db9360a | 1104 | /* |
0f16ef7f NF |
1105 | * Find the node associated with a hot added memory section for |
1106 | * memory represented in the device tree by the property | |
1107 | * ibm,dynamic-reconfiguration-memory/ibm,dynamic-memory. | |
0db9360a NF |
1108 | */ |
1109 | static int hot_add_drconf_scn_to_nid(struct device_node *memory, | |
1110 | unsigned long scn_addr) | |
1111 | { | |
1112 | const u32 *dm; | |
0f16ef7f | 1113 | unsigned int drconf_cell_cnt, rc; |
3fdfd990 | 1114 | unsigned long lmb_size; |
0db9360a | 1115 | struct assoc_arrays aa; |
0f16ef7f | 1116 | int nid = -1; |
0db9360a | 1117 | |
0f16ef7f NF |
1118 | drconf_cell_cnt = of_get_drconf_memory(memory, &dm); |
1119 | if (!drconf_cell_cnt) | |
1120 | return -1; | |
0db9360a | 1121 | |
3fdfd990 BH |
1122 | lmb_size = of_get_lmb_size(memory); |
1123 | if (!lmb_size) | |
0f16ef7f | 1124 | return -1; |
0db9360a NF |
1125 | |
1126 | rc = of_get_assoc_arrays(memory, &aa); | |
1127 | if (rc) | |
0f16ef7f | 1128 | return -1; |
0db9360a | 1129 | |
0f16ef7f | 1130 | for (; drconf_cell_cnt != 0; --drconf_cell_cnt) { |
0db9360a NF |
1131 | struct of_drconf_cell drmem; |
1132 | ||
1133 | read_drconf_cell(&drmem, &dm); | |
1134 | ||
1135 | /* skip this block if it is reserved or not assigned to | |
1136 | * this partition */ | |
1137 | if ((drmem.flags & DRCONF_MEM_RESERVED) | |
1138 | || !(drmem.flags & DRCONF_MEM_ASSIGNED)) | |
1139 | continue; | |
1140 | ||
0f16ef7f | 1141 | if ((scn_addr < drmem.base_addr) |
3fdfd990 | 1142 | || (scn_addr >= (drmem.base_addr + lmb_size))) |
0f16ef7f NF |
1143 | continue; |
1144 | ||
0db9360a | 1145 | nid = of_drconf_to_nid_single(&drmem, &aa); |
0f16ef7f NF |
1146 | break; |
1147 | } | |
1148 | ||
1149 | return nid; | |
1150 | } | |
1151 | ||
1152 | /* | |
1153 | * Find the node associated with a hot added memory section for memory | |
1154 | * represented in the device tree as a node (i.e. memory@XXXX) for | |
95f72d1e | 1155 | * each memblock. |
0f16ef7f NF |
1156 | */ |
1157 | int hot_add_node_scn_to_nid(unsigned long scn_addr) | |
1158 | { | |
94db7c5e | 1159 | struct device_node *memory; |
0f16ef7f NF |
1160 | int nid = -1; |
1161 | ||
94db7c5e | 1162 | for_each_node_by_type(memory, "memory") { |
0f16ef7f NF |
1163 | unsigned long start, size; |
1164 | int ranges; | |
1165 | const unsigned int *memcell_buf; | |
1166 | unsigned int len; | |
1167 | ||
1168 | memcell_buf = of_get_property(memory, "reg", &len); | |
1169 | if (!memcell_buf || len <= 0) | |
1170 | continue; | |
1171 | ||
1172 | /* ranges in cell */ | |
1173 | ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells); | |
1174 | ||
1175 | while (ranges--) { | |
1176 | start = read_n_cells(n_mem_addr_cells, &memcell_buf); | |
1177 | size = read_n_cells(n_mem_size_cells, &memcell_buf); | |
1178 | ||
1179 | if ((scn_addr < start) || (scn_addr >= (start + size))) | |
1180 | continue; | |
1181 | ||
1182 | nid = of_node_to_nid_single(memory); | |
1183 | break; | |
1184 | } | |
0db9360a | 1185 | |
0f16ef7f NF |
1186 | if (nid >= 0) |
1187 | break; | |
0db9360a NF |
1188 | } |
1189 | ||
60831842 AB |
1190 | of_node_put(memory); |
1191 | ||
0f16ef7f | 1192 | return nid; |
0db9360a NF |
1193 | } |
1194 | ||
237a0989 MK |
1195 | /* |
1196 | * Find the node associated with a hot added memory section. Section | |
95f72d1e YL |
1197 | * corresponds to a SPARSEMEM section, not an MEMBLOCK. It is assumed that |
1198 | * sections are fully contained within a single MEMBLOCK. | |
237a0989 MK |
1199 | */ |
1200 | int hot_add_scn_to_nid(unsigned long scn_addr) | |
1201 | { | |
1202 | struct device_node *memory = NULL; | |
0f16ef7f | 1203 | int nid, found = 0; |
237a0989 MK |
1204 | |
1205 | if (!numa_enabled || (min_common_depth < 0)) | |
72c33688 | 1206 | return first_online_node; |
0db9360a NF |
1207 | |
1208 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); | |
1209 | if (memory) { | |
1210 | nid = hot_add_drconf_scn_to_nid(memory, scn_addr); | |
1211 | of_node_put(memory); | |
0f16ef7f NF |
1212 | } else { |
1213 | nid = hot_add_node_scn_to_nid(scn_addr); | |
0db9360a | 1214 | } |
237a0989 | 1215 | |
0f16ef7f | 1216 | if (nid < 0 || !node_online(nid)) |
72c33688 | 1217 | nid = first_online_node; |
237a0989 | 1218 | |
0f16ef7f NF |
1219 | if (NODE_DATA(nid)->node_spanned_pages) |
1220 | return nid; | |
237a0989 | 1221 | |
0f16ef7f NF |
1222 | for_each_online_node(nid) { |
1223 | if (NODE_DATA(nid)->node_spanned_pages) { | |
1224 | found = 1; | |
1225 | break; | |
237a0989 | 1226 | } |
237a0989 | 1227 | } |
0f16ef7f NF |
1228 | |
1229 | BUG_ON(!found); | |
1230 | return nid; | |
237a0989 | 1231 | } |
0f16ef7f | 1232 | |
cd34206e NA |
1233 | static u64 hot_add_drconf_memory_max(void) |
1234 | { | |
1235 | struct device_node *memory = NULL; | |
1236 | unsigned int drconf_cell_cnt = 0; | |
1237 | u64 lmb_size = 0; | |
1238 | const u32 *dm = 0; | |
1239 | ||
1240 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); | |
1241 | if (memory) { | |
1242 | drconf_cell_cnt = of_get_drconf_memory(memory, &dm); | |
1243 | lmb_size = of_get_lmb_size(memory); | |
1244 | of_node_put(memory); | |
1245 | } | |
1246 | return lmb_size * drconf_cell_cnt; | |
1247 | } | |
1248 | ||
1249 | /* | |
1250 | * memory_hotplug_max - return max address of memory that may be added | |
1251 | * | |
1252 | * This is currently only used on systems that support drconfig memory | |
1253 | * hotplug. | |
1254 | */ | |
1255 | u64 memory_hotplug_max(void) | |
1256 | { | |
1257 | return max(hot_add_drconf_memory_max(), memblock_end_of_DRAM()); | |
1258 | } | |
237a0989 | 1259 | #endif /* CONFIG_MEMORY_HOTPLUG */ |
9eff1a38 | 1260 | |
bd03403a | 1261 | /* Virtual Processor Home Node (VPHN) support */ |
39bf990e | 1262 | #ifdef CONFIG_PPC_SPLPAR |
30c05350 NF |
1263 | struct topology_update_data { |
1264 | struct topology_update_data *next; | |
1265 | unsigned int cpu; | |
1266 | int old_nid; | |
1267 | int new_nid; | |
1268 | }; | |
1269 | ||
5de16699 | 1270 | static u8 vphn_cpu_change_counts[NR_CPUS][MAX_DISTANCE_REF_POINTS]; |
9eff1a38 JL |
1271 | static cpumask_t cpu_associativity_changes_mask; |
1272 | static int vphn_enabled; | |
5d88aa85 JL |
1273 | static int prrn_enabled; |
1274 | static void reset_topology_timer(void); | |
9eff1a38 JL |
1275 | |
1276 | /* | |
1277 | * Store the current values of the associativity change counters in the | |
1278 | * hypervisor. | |
1279 | */ | |
1280 | static void setup_cpu_associativity_change_counters(void) | |
1281 | { | |
cd9d6cc7 | 1282 | int cpu; |
9eff1a38 | 1283 | |
5de16699 AB |
1284 | /* The VPHN feature supports a maximum of 8 reference points */ |
1285 | BUILD_BUG_ON(MAX_DISTANCE_REF_POINTS > 8); | |
1286 | ||
9eff1a38 | 1287 | for_each_possible_cpu(cpu) { |
cd9d6cc7 | 1288 | int i; |
9eff1a38 JL |
1289 | u8 *counts = vphn_cpu_change_counts[cpu]; |
1290 | volatile u8 *hypervisor_counts = lppaca[cpu].vphn_assoc_counts; | |
1291 | ||
5de16699 | 1292 | for (i = 0; i < distance_ref_points_depth; i++) |
9eff1a38 | 1293 | counts[i] = hypervisor_counts[i]; |
9eff1a38 JL |
1294 | } |
1295 | } | |
1296 | ||
1297 | /* | |
1298 | * The hypervisor maintains a set of 8 associativity change counters in | |
1299 | * the VPA of each cpu that correspond to the associativity levels in the | |
1300 | * ibm,associativity-reference-points property. When an associativity | |
1301 | * level changes, the corresponding counter is incremented. | |
1302 | * | |
1303 | * Set a bit in cpu_associativity_changes_mask for each cpu whose home | |
1304 | * node associativity levels have changed. | |
1305 | * | |
1306 | * Returns the number of cpus with unhandled associativity changes. | |
1307 | */ | |
1308 | static int update_cpu_associativity_changes_mask(void) | |
1309 | { | |
5d88aa85 | 1310 | int cpu; |
9eff1a38 JL |
1311 | cpumask_t *changes = &cpu_associativity_changes_mask; |
1312 | ||
9eff1a38 JL |
1313 | for_each_possible_cpu(cpu) { |
1314 | int i, changed = 0; | |
1315 | u8 *counts = vphn_cpu_change_counts[cpu]; | |
1316 | volatile u8 *hypervisor_counts = lppaca[cpu].vphn_assoc_counts; | |
1317 | ||
5de16699 | 1318 | for (i = 0; i < distance_ref_points_depth; i++) { |
d69043e8 | 1319 | if (hypervisor_counts[i] != counts[i]) { |
9eff1a38 JL |
1320 | counts[i] = hypervisor_counts[i]; |
1321 | changed = 1; | |
1322 | } | |
1323 | } | |
1324 | if (changed) { | |
1325 | cpumask_set_cpu(cpu, changes); | |
9eff1a38 JL |
1326 | } |
1327 | } | |
1328 | ||
5d88aa85 | 1329 | return cpumask_weight(changes); |
9eff1a38 JL |
1330 | } |
1331 | ||
c0e5e46f AB |
1332 | /* |
1333 | * 6 64-bit registers unpacked into 12 32-bit associativity values. To form | |
1334 | * the complete property we have to add the length in the first cell. | |
1335 | */ | |
1336 | #define VPHN_ASSOC_BUFSIZE (6*sizeof(u64)/sizeof(u32) + 1) | |
9eff1a38 JL |
1337 | |
1338 | /* | |
1339 | * Convert the associativity domain numbers returned from the hypervisor | |
1340 | * to the sequence they would appear in the ibm,associativity property. | |
1341 | */ | |
1342 | static int vphn_unpack_associativity(const long *packed, unsigned int *unpacked) | |
1343 | { | |
cd9d6cc7 | 1344 | int i, nr_assoc_doms = 0; |
9eff1a38 JL |
1345 | const u16 *field = (const u16*) packed; |
1346 | ||
1347 | #define VPHN_FIELD_UNUSED (0xffff) | |
1348 | #define VPHN_FIELD_MSB (0x8000) | |
1349 | #define VPHN_FIELD_MASK (~VPHN_FIELD_MSB) | |
1350 | ||
c0e5e46f | 1351 | for (i = 1; i < VPHN_ASSOC_BUFSIZE; i++) { |
9eff1a38 JL |
1352 | if (*field == VPHN_FIELD_UNUSED) { |
1353 | /* All significant fields processed, and remaining | |
1354 | * fields contain the reserved value of all 1's. | |
1355 | * Just store them. | |
1356 | */ | |
1357 | unpacked[i] = *((u32*)field); | |
1358 | field += 2; | |
7639adaa | 1359 | } else if (*field & VPHN_FIELD_MSB) { |
9eff1a38 JL |
1360 | /* Data is in the lower 15 bits of this field */ |
1361 | unpacked[i] = *field & VPHN_FIELD_MASK; | |
1362 | field++; | |
1363 | nr_assoc_doms++; | |
7639adaa | 1364 | } else { |
9eff1a38 JL |
1365 | /* Data is in the lower 15 bits of this field |
1366 | * concatenated with the next 16 bit field | |
1367 | */ | |
1368 | unpacked[i] = *((u32*)field); | |
1369 | field += 2; | |
1370 | nr_assoc_doms++; | |
1371 | } | |
1372 | } | |
1373 | ||
c0e5e46f AB |
1374 | /* The first cell contains the length of the property */ |
1375 | unpacked[0] = nr_assoc_doms; | |
1376 | ||
9eff1a38 JL |
1377 | return nr_assoc_doms; |
1378 | } | |
1379 | ||
1380 | /* | |
1381 | * Retrieve the new associativity information for a virtual processor's | |
1382 | * home node. | |
1383 | */ | |
1384 | static long hcall_vphn(unsigned long cpu, unsigned int *associativity) | |
1385 | { | |
cd9d6cc7 | 1386 | long rc; |
9eff1a38 JL |
1387 | long retbuf[PLPAR_HCALL9_BUFSIZE] = {0}; |
1388 | u64 flags = 1; | |
1389 | int hwcpu = get_hard_smp_processor_id(cpu); | |
1390 | ||
1391 | rc = plpar_hcall9(H_HOME_NODE_ASSOCIATIVITY, retbuf, flags, hwcpu); | |
1392 | vphn_unpack_associativity(retbuf, associativity); | |
1393 | ||
1394 | return rc; | |
1395 | } | |
1396 | ||
1397 | static long vphn_get_associativity(unsigned long cpu, | |
1398 | unsigned int *associativity) | |
1399 | { | |
cd9d6cc7 | 1400 | long rc; |
9eff1a38 JL |
1401 | |
1402 | rc = hcall_vphn(cpu, associativity); | |
1403 | ||
1404 | switch (rc) { | |
1405 | case H_FUNCTION: | |
1406 | printk(KERN_INFO | |
1407 | "VPHN is not supported. Disabling polling...\n"); | |
1408 | stop_topology_update(); | |
1409 | break; | |
1410 | case H_HARDWARE: | |
1411 | printk(KERN_ERR | |
1412 | "hcall_vphn() experienced a hardware fault " | |
1413 | "preventing VPHN. Disabling polling...\n"); | |
1414 | stop_topology_update(); | |
1415 | } | |
1416 | ||
1417 | return rc; | |
1418 | } | |
1419 | ||
30c05350 NF |
1420 | /* |
1421 | * Update the CPU maps and sysfs entries for a single CPU when its NUMA | |
1422 | * characteristics change. This function doesn't perform any locking and is | |
1423 | * only safe to call from stop_machine(). | |
1424 | */ | |
1425 | static int update_cpu_topology(void *data) | |
1426 | { | |
1427 | struct topology_update_data *update; | |
1428 | unsigned long cpu; | |
1429 | ||
1430 | if (!data) | |
1431 | return -EINVAL; | |
1432 | ||
1433 | cpu = get_cpu(); | |
1434 | ||
1435 | for (update = data; update; update = update->next) { | |
1436 | if (cpu != update->cpu) | |
1437 | continue; | |
1438 | ||
1439 | unregister_cpu_under_node(update->cpu, update->old_nid); | |
1440 | unmap_cpu_from_node(update->cpu); | |
1441 | map_cpu_to_node(update->cpu, update->new_nid); | |
176bbf14 | 1442 | vdso_getcpu_init(); |
30c05350 NF |
1443 | register_cpu_under_node(update->cpu, update->new_nid); |
1444 | } | |
1445 | ||
1446 | return 0; | |
1447 | } | |
1448 | ||
9eff1a38 JL |
1449 | /* |
1450 | * Update the node maps and sysfs entries for each cpu whose home node | |
79c5fceb | 1451 | * has changed. Returns 1 when the topology has changed, and 0 otherwise. |
9eff1a38 JL |
1452 | */ |
1453 | int arch_update_cpu_topology(void) | |
1454 | { | |
30c05350 NF |
1455 | unsigned int cpu, changed = 0; |
1456 | struct topology_update_data *updates, *ud; | |
9eff1a38 | 1457 | unsigned int associativity[VPHN_ASSOC_BUFSIZE] = {0}; |
176bbf14 | 1458 | cpumask_t updated_cpus; |
8a25a2fd | 1459 | struct device *dev; |
30c05350 NF |
1460 | int weight, i = 0; |
1461 | ||
1462 | weight = cpumask_weight(&cpu_associativity_changes_mask); | |
1463 | if (!weight) | |
1464 | return 0; | |
1465 | ||
1466 | updates = kzalloc(weight * (sizeof(*updates)), GFP_KERNEL); | |
1467 | if (!updates) | |
1468 | return 0; | |
9eff1a38 | 1469 | |
176bbf14 JL |
1470 | cpumask_clear(&updated_cpus); |
1471 | ||
5d88aa85 | 1472 | for_each_cpu(cpu, &cpu_associativity_changes_mask) { |
30c05350 NF |
1473 | ud = &updates[i++]; |
1474 | ud->cpu = cpu; | |
9eff1a38 | 1475 | vphn_get_associativity(cpu, associativity); |
30c05350 | 1476 | ud->new_nid = associativity_to_nid(associativity); |
9eff1a38 | 1477 | |
30c05350 NF |
1478 | if (ud->new_nid < 0 || !node_online(ud->new_nid)) |
1479 | ud->new_nid = first_online_node; | |
9eff1a38 | 1480 | |
30c05350 | 1481 | ud->old_nid = numa_cpu_lookup_table[cpu]; |
176bbf14 | 1482 | cpumask_set_cpu(cpu, &updated_cpus); |
9eff1a38 | 1483 | |
30c05350 NF |
1484 | if (i < weight) |
1485 | ud->next = &updates[i]; | |
1486 | } | |
1487 | ||
176bbf14 | 1488 | stop_machine(update_cpu_topology, &updates[0], &updated_cpus); |
30c05350 NF |
1489 | |
1490 | for (ud = &updates[0]; ud; ud = ud->next) { | |
1491 | dev = get_cpu_device(ud->cpu); | |
8a25a2fd KS |
1492 | if (dev) |
1493 | kobject_uevent(&dev->kobj, KOBJ_CHANGE); | |
30c05350 | 1494 | cpumask_clear_cpu(ud->cpu, &cpu_associativity_changes_mask); |
79c5fceb | 1495 | changed = 1; |
9eff1a38 JL |
1496 | } |
1497 | ||
30c05350 | 1498 | kfree(updates); |
79c5fceb | 1499 | return changed; |
9eff1a38 JL |
1500 | } |
1501 | ||
1502 | static void topology_work_fn(struct work_struct *work) | |
1503 | { | |
1504 | rebuild_sched_domains(); | |
1505 | } | |
1506 | static DECLARE_WORK(topology_work, topology_work_fn); | |
1507 | ||
1508 | void topology_schedule_update(void) | |
1509 | { | |
1510 | schedule_work(&topology_work); | |
1511 | } | |
1512 | ||
1513 | static void topology_timer_fn(unsigned long ignored) | |
1514 | { | |
5d88aa85 | 1515 | if (prrn_enabled && cpumask_weight(&cpu_associativity_changes_mask)) |
9eff1a38 | 1516 | topology_schedule_update(); |
5d88aa85 JL |
1517 | else if (vphn_enabled) { |
1518 | if (update_cpu_associativity_changes_mask() > 0) | |
1519 | topology_schedule_update(); | |
1520 | reset_topology_timer(); | |
1521 | } | |
9eff1a38 JL |
1522 | } |
1523 | static struct timer_list topology_timer = | |
1524 | TIMER_INITIALIZER(topology_timer_fn, 0, 0); | |
1525 | ||
5d88aa85 | 1526 | static void reset_topology_timer(void) |
9eff1a38 JL |
1527 | { |
1528 | topology_timer.data = 0; | |
1529 | topology_timer.expires = jiffies + 60 * HZ; | |
5d88aa85 JL |
1530 | mod_timer(&topology_timer, topology_timer.expires); |
1531 | } | |
1532 | ||
1533 | static void stage_topology_update(int core_id) | |
1534 | { | |
1535 | cpumask_or(&cpu_associativity_changes_mask, | |
1536 | &cpu_associativity_changes_mask, cpu_sibling_mask(core_id)); | |
1537 | reset_topology_timer(); | |
1538 | } | |
1539 | ||
1540 | static int dt_update_callback(struct notifier_block *nb, | |
1541 | unsigned long action, void *data) | |
1542 | { | |
1543 | struct of_prop_reconfig *update; | |
1544 | int rc = NOTIFY_DONE; | |
1545 | ||
1546 | switch (action) { | |
5d88aa85 JL |
1547 | case OF_RECONFIG_UPDATE_PROPERTY: |
1548 | update = (struct of_prop_reconfig *)data; | |
30c05350 NF |
1549 | if (!of_prop_cmp(update->dn->type, "cpu") && |
1550 | !of_prop_cmp(update->prop->name, "ibm,associativity")) { | |
5d88aa85 JL |
1551 | u32 core_id; |
1552 | of_property_read_u32(update->dn, "reg", &core_id); | |
1553 | stage_topology_update(core_id); | |
1554 | rc = NOTIFY_OK; | |
1555 | } | |
1556 | break; | |
1557 | } | |
1558 | ||
1559 | return rc; | |
9eff1a38 JL |
1560 | } |
1561 | ||
5d88aa85 JL |
1562 | static struct notifier_block dt_update_nb = { |
1563 | .notifier_call = dt_update_callback, | |
1564 | }; | |
1565 | ||
9eff1a38 | 1566 | /* |
5d88aa85 | 1567 | * Start polling for associativity changes. |
9eff1a38 JL |
1568 | */ |
1569 | int start_topology_update(void) | |
1570 | { | |
1571 | int rc = 0; | |
1572 | ||
5d88aa85 JL |
1573 | if (firmware_has_feature(FW_FEATURE_PRRN)) { |
1574 | if (!prrn_enabled) { | |
1575 | prrn_enabled = 1; | |
1576 | vphn_enabled = 0; | |
1577 | rc = of_reconfig_notifier_register(&dt_update_nb); | |
1578 | } | |
b7abef04 | 1579 | } else if (firmware_has_feature(FW_FEATURE_VPHN) && |
5d88aa85 | 1580 | get_lppaca()->shared_proc) { |
5d88aa85 JL |
1581 | if (!vphn_enabled) { |
1582 | prrn_enabled = 0; | |
1583 | vphn_enabled = 1; | |
1584 | setup_cpu_associativity_change_counters(); | |
1585 | init_timer_deferrable(&topology_timer); | |
1586 | reset_topology_timer(); | |
1587 | } | |
9eff1a38 JL |
1588 | } |
1589 | ||
1590 | return rc; | |
1591 | } | |
9eff1a38 JL |
1592 | |
1593 | /* | |
1594 | * Disable polling for VPHN associativity changes. | |
1595 | */ | |
1596 | int stop_topology_update(void) | |
1597 | { | |
5d88aa85 JL |
1598 | int rc = 0; |
1599 | ||
1600 | if (prrn_enabled) { | |
1601 | prrn_enabled = 0; | |
1602 | rc = of_reconfig_notifier_unregister(&dt_update_nb); | |
1603 | } else if (vphn_enabled) { | |
1604 | vphn_enabled = 0; | |
1605 | rc = del_timer_sync(&topology_timer); | |
1606 | } | |
1607 | ||
1608 | return rc; | |
9eff1a38 | 1609 | } |
e04fa612 NF |
1610 | |
1611 | int prrn_is_enabled(void) | |
1612 | { | |
1613 | return prrn_enabled; | |
1614 | } | |
1615 | ||
1616 | static int topology_read(struct seq_file *file, void *v) | |
1617 | { | |
1618 | if (vphn_enabled || prrn_enabled) | |
1619 | seq_puts(file, "on\n"); | |
1620 | else | |
1621 | seq_puts(file, "off\n"); | |
1622 | ||
1623 | return 0; | |
1624 | } | |
1625 | ||
1626 | static int topology_open(struct inode *inode, struct file *file) | |
1627 | { | |
1628 | return single_open(file, topology_read, NULL); | |
1629 | } | |
1630 | ||
1631 | static ssize_t topology_write(struct file *file, const char __user *buf, | |
1632 | size_t count, loff_t *off) | |
1633 | { | |
1634 | char kbuf[4]; /* "on" or "off" plus null. */ | |
1635 | int read_len; | |
1636 | ||
1637 | read_len = count < 3 ? count : 3; | |
1638 | if (copy_from_user(kbuf, buf, read_len)) | |
1639 | return -EINVAL; | |
1640 | ||
1641 | kbuf[read_len] = '\0'; | |
1642 | ||
1643 | if (!strncmp(kbuf, "on", 2)) | |
1644 | start_topology_update(); | |
1645 | else if (!strncmp(kbuf, "off", 3)) | |
1646 | stop_topology_update(); | |
1647 | else | |
1648 | return -EINVAL; | |
1649 | ||
1650 | return count; | |
1651 | } | |
1652 | ||
1653 | static const struct file_operations topology_ops = { | |
1654 | .read = seq_read, | |
1655 | .write = topology_write, | |
1656 | .open = topology_open, | |
1657 | .release = single_release | |
1658 | }; | |
1659 | ||
1660 | static int topology_update_init(void) | |
1661 | { | |
1662 | start_topology_update(); | |
1663 | proc_create("powerpc/topology_updates", 644, NULL, &topology_ops); | |
1664 | ||
1665 | return 0; | |
1666 | } | |
1667 | device_initcall(topology_update_init); | |
39bf990e | 1668 | #endif /* CONFIG_PPC_SPLPAR */ |