]>
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> | |
16 | #include <linux/module.h> | |
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> |
45fb6cea | 23 | #include <asm/sparsemem.h> |
d9b2b2a2 | 24 | #include <asm/prom.h> |
cf00a8d1 | 25 | #include <asm/system.h> |
2249ca9d | 26 | #include <asm/smp.h> |
1da177e4 LT |
27 | |
28 | static int numa_enabled = 1; | |
29 | ||
1daa6d08 BS |
30 | static char *cmdline __initdata; |
31 | ||
1da177e4 LT |
32 | static int numa_debug; |
33 | #define dbg(args...) if (numa_debug) { printk(KERN_INFO args); } | |
34 | ||
45fb6cea | 35 | int numa_cpu_lookup_table[NR_CPUS]; |
25863de0 | 36 | cpumask_var_t node_to_cpumask_map[MAX_NUMNODES]; |
1da177e4 | 37 | struct pglist_data *node_data[MAX_NUMNODES]; |
45fb6cea AB |
38 | |
39 | EXPORT_SYMBOL(numa_cpu_lookup_table); | |
25863de0 | 40 | EXPORT_SYMBOL(node_to_cpumask_map); |
45fb6cea AB |
41 | EXPORT_SYMBOL(node_data); |
42 | ||
1da177e4 | 43 | static int min_common_depth; |
237a0989 | 44 | static int n_mem_addr_cells, n_mem_size_cells; |
1da177e4 | 45 | |
25863de0 AB |
46 | /* |
47 | * Allocate node_to_cpumask_map based on number of available nodes | |
48 | * Requires node_possible_map to be valid. | |
49 | * | |
50 | * Note: node_to_cpumask() is not valid until after this is done. | |
51 | */ | |
52 | static void __init setup_node_to_cpumask_map(void) | |
53 | { | |
54 | unsigned int node, num = 0; | |
55 | ||
56 | /* setup nr_node_ids if not done yet */ | |
57 | if (nr_node_ids == MAX_NUMNODES) { | |
58 | for_each_node_mask(node, node_possible_map) | |
59 | num = node; | |
60 | nr_node_ids = num + 1; | |
61 | } | |
62 | ||
63 | /* allocate the map */ | |
64 | for (node = 0; node < nr_node_ids; node++) | |
65 | alloc_bootmem_cpumask_var(&node_to_cpumask_map[node]); | |
66 | ||
67 | /* cpumask_of_node() will now work */ | |
68 | dbg("Node to cpumask map for %d nodes\n", nr_node_ids); | |
69 | } | |
70 | ||
1daa6d08 BS |
71 | static int __cpuinit fake_numa_create_new_node(unsigned long end_pfn, |
72 | unsigned int *nid) | |
73 | { | |
74 | unsigned long long mem; | |
75 | char *p = cmdline; | |
76 | static unsigned int fake_nid; | |
77 | static unsigned long long curr_boundary; | |
78 | ||
79 | /* | |
80 | * Modify node id, iff we started creating NUMA nodes | |
81 | * We want to continue from where we left of the last time | |
82 | */ | |
83 | if (fake_nid) | |
84 | *nid = fake_nid; | |
85 | /* | |
86 | * In case there are no more arguments to parse, the | |
87 | * node_id should be the same as the last fake node id | |
88 | * (we've handled this above). | |
89 | */ | |
90 | if (!p) | |
91 | return 0; | |
92 | ||
93 | mem = memparse(p, &p); | |
94 | if (!mem) | |
95 | return 0; | |
96 | ||
97 | if (mem < curr_boundary) | |
98 | return 0; | |
99 | ||
100 | curr_boundary = mem; | |
101 | ||
102 | if ((end_pfn << PAGE_SHIFT) > mem) { | |
103 | /* | |
104 | * Skip commas and spaces | |
105 | */ | |
106 | while (*p == ',' || *p == ' ' || *p == '\t') | |
107 | p++; | |
108 | ||
109 | cmdline = p; | |
110 | fake_nid++; | |
111 | *nid = fake_nid; | |
112 | dbg("created new fake_node with id %d\n", fake_nid); | |
113 | return 1; | |
114 | } | |
115 | return 0; | |
116 | } | |
117 | ||
8f64e1f2 JT |
118 | /* |
119 | * get_active_region_work_fn - A helper function for get_node_active_region | |
120 | * Returns datax set to the start_pfn and end_pfn if they contain | |
121 | * the initial value of datax->start_pfn between them | |
122 | * @start_pfn: start page(inclusive) of region to check | |
123 | * @end_pfn: end page(exclusive) of region to check | |
124 | * @datax: comes in with ->start_pfn set to value to search for and | |
125 | * goes out with active range if it contains it | |
126 | * Returns 1 if search value is in range else 0 | |
127 | */ | |
128 | static int __init get_active_region_work_fn(unsigned long start_pfn, | |
129 | unsigned long end_pfn, void *datax) | |
130 | { | |
131 | struct node_active_region *data; | |
132 | data = (struct node_active_region *)datax; | |
133 | ||
134 | if (start_pfn <= data->start_pfn && end_pfn > data->start_pfn) { | |
135 | data->start_pfn = start_pfn; | |
136 | data->end_pfn = end_pfn; | |
137 | return 1; | |
138 | } | |
139 | return 0; | |
140 | ||
141 | } | |
142 | ||
143 | /* | |
144 | * get_node_active_region - Return active region containing start_pfn | |
e8170372 | 145 | * Active range returned is empty if none found. |
8f64e1f2 JT |
146 | * @start_pfn: The page to return the region for. |
147 | * @node_ar: Returned set to the active region containing start_pfn | |
148 | */ | |
149 | static void __init get_node_active_region(unsigned long start_pfn, | |
150 | struct node_active_region *node_ar) | |
151 | { | |
152 | int nid = early_pfn_to_nid(start_pfn); | |
153 | ||
154 | node_ar->nid = nid; | |
155 | node_ar->start_pfn = start_pfn; | |
e8170372 | 156 | node_ar->end_pfn = start_pfn; |
8f64e1f2 JT |
157 | work_with_active_regions(nid, get_active_region_work_fn, node_ar); |
158 | } | |
159 | ||
2e5ce39d | 160 | static void __cpuinit map_cpu_to_node(int cpu, int node) |
1da177e4 LT |
161 | { |
162 | numa_cpu_lookup_table[cpu] = node; | |
45fb6cea | 163 | |
bf4b85b0 NL |
164 | dbg("adding cpu %d to node %d\n", cpu, node); |
165 | ||
25863de0 AB |
166 | if (!(cpumask_test_cpu(cpu, node_to_cpumask_map[node]))) |
167 | cpumask_set_cpu(cpu, node_to_cpumask_map[node]); | |
1da177e4 LT |
168 | } |
169 | ||
170 | #ifdef CONFIG_HOTPLUG_CPU | |
171 | static void unmap_cpu_from_node(unsigned long cpu) | |
172 | { | |
173 | int node = numa_cpu_lookup_table[cpu]; | |
174 | ||
175 | dbg("removing cpu %lu from node %d\n", cpu, node); | |
176 | ||
25863de0 AB |
177 | if (cpumask_test_cpu(cpu, node_to_cpumask_map[node])) { |
178 | cpumask_set_cpu(cpu, node_to_cpumask_map[node]); | |
1da177e4 LT |
179 | } else { |
180 | printk(KERN_ERR "WARNING: cpu %lu not found in node %d\n", | |
181 | cpu, node); | |
182 | } | |
183 | } | |
184 | #endif /* CONFIG_HOTPLUG_CPU */ | |
185 | ||
1da177e4 | 186 | /* must hold reference to node during call */ |
a7f67bdf | 187 | static const int *of_get_associativity(struct device_node *dev) |
1da177e4 | 188 | { |
e2eb6392 | 189 | return of_get_property(dev, "ibm,associativity", NULL); |
1da177e4 LT |
190 | } |
191 | ||
cf00085d C |
192 | /* |
193 | * Returns the property linux,drconf-usable-memory if | |
194 | * it exists (the property exists only in kexec/kdump kernels, | |
195 | * added by kexec-tools) | |
196 | */ | |
197 | static const u32 *of_get_usable_memory(struct device_node *memory) | |
198 | { | |
199 | const u32 *prop; | |
200 | u32 len; | |
201 | prop = of_get_property(memory, "linux,drconf-usable-memory", &len); | |
202 | if (!prop || len < sizeof(unsigned int)) | |
203 | return 0; | |
204 | return prop; | |
205 | } | |
206 | ||
482ec7c4 NL |
207 | /* Returns nid in the range [0..MAX_NUMNODES-1], or -1 if no useful numa |
208 | * info is found. | |
209 | */ | |
953039c8 | 210 | static int of_node_to_nid_single(struct device_node *device) |
1da177e4 | 211 | { |
482ec7c4 | 212 | int nid = -1; |
a7f67bdf | 213 | const unsigned int *tmp; |
1da177e4 LT |
214 | |
215 | if (min_common_depth == -1) | |
482ec7c4 | 216 | goto out; |
1da177e4 LT |
217 | |
218 | tmp = of_get_associativity(device); | |
482ec7c4 NL |
219 | if (!tmp) |
220 | goto out; | |
221 | ||
222 | if (tmp[0] >= min_common_depth) | |
cf950b7a | 223 | nid = tmp[min_common_depth]; |
bc16a759 NL |
224 | |
225 | /* POWER4 LPAR uses 0xffff as invalid node */ | |
482ec7c4 NL |
226 | if (nid == 0xffff || nid >= MAX_NUMNODES) |
227 | nid = -1; | |
228 | out: | |
cf950b7a | 229 | return nid; |
1da177e4 LT |
230 | } |
231 | ||
953039c8 JK |
232 | /* Walk the device tree upwards, looking for an associativity id */ |
233 | int of_node_to_nid(struct device_node *device) | |
234 | { | |
235 | struct device_node *tmp; | |
236 | int nid = -1; | |
237 | ||
238 | of_node_get(device); | |
239 | while (device) { | |
240 | nid = of_node_to_nid_single(device); | |
241 | if (nid != -1) | |
242 | break; | |
243 | ||
244 | tmp = device; | |
245 | device = of_get_parent(tmp); | |
246 | of_node_put(tmp); | |
247 | } | |
248 | of_node_put(device); | |
249 | ||
250 | return nid; | |
251 | } | |
252 | EXPORT_SYMBOL_GPL(of_node_to_nid); | |
253 | ||
1da177e4 LT |
254 | /* |
255 | * In theory, the "ibm,associativity" property may contain multiple | |
256 | * associativity lists because a resource may be multiply connected | |
257 | * into the machine. This resource then has different associativity | |
258 | * characteristics relative to its multiple connections. We ignore | |
259 | * this for now. We also assume that all cpu and memory sets have | |
260 | * their distances represented at a common level. This won't be | |
1b3c3714 | 261 | * true for hierarchical NUMA. |
1da177e4 LT |
262 | * |
263 | * In any case the ibm,associativity-reference-points should give | |
264 | * the correct depth for a normal NUMA system. | |
265 | * | |
266 | * - Dave Hansen <haveblue@us.ibm.com> | |
267 | */ | |
268 | static int __init find_min_common_depth(void) | |
269 | { | |
4b83c330 | 270 | int depth, index; |
a7f67bdf | 271 | const unsigned int *ref_points; |
1da177e4 LT |
272 | struct device_node *rtas_root; |
273 | unsigned int len; | |
bc8449cc AB |
274 | struct device_node *chosen; |
275 | const char *vec5; | |
1da177e4 LT |
276 | |
277 | rtas_root = of_find_node_by_path("/rtas"); | |
278 | ||
279 | if (!rtas_root) | |
280 | return -1; | |
281 | ||
282 | /* | |
283 | * this property is 2 32-bit integers, each representing a level of | |
284 | * depth in the associativity nodes. The first is for an SMP | |
285 | * configuration (should be all 0's) and the second is for a normal | |
286 | * NUMA configuration. | |
287 | */ | |
4b83c330 | 288 | index = 1; |
e2eb6392 | 289 | ref_points = of_get_property(rtas_root, |
1da177e4 LT |
290 | "ibm,associativity-reference-points", &len); |
291 | ||
4b83c330 | 292 | /* |
bc8449cc | 293 | * For form 1 affinity information we want the first field |
4b83c330 | 294 | */ |
bc8449cc AB |
295 | #define VEC5_AFFINITY_BYTE 5 |
296 | #define VEC5_AFFINITY 0x80 | |
297 | chosen = of_find_node_by_path("/chosen"); | |
298 | if (chosen) { | |
299 | vec5 = of_get_property(chosen, "ibm,architecture-vec-5", NULL); | |
300 | if (vec5 && (vec5[VEC5_AFFINITY_BYTE] & VEC5_AFFINITY)) { | |
301 | dbg("Using form 1 affinity\n"); | |
302 | index = 0; | |
303 | } | |
4b83c330 AB |
304 | } |
305 | ||
20fcefe5 | 306 | if ((len >= 2 * sizeof(unsigned int)) && ref_points) { |
4b83c330 | 307 | depth = ref_points[index]; |
1da177e4 | 308 | } else { |
bf4b85b0 | 309 | dbg("NUMA: ibm,associativity-reference-points not found.\n"); |
1da177e4 LT |
310 | depth = -1; |
311 | } | |
312 | of_node_put(rtas_root); | |
313 | ||
314 | return depth; | |
315 | } | |
316 | ||
84c9fdd1 | 317 | static void __init get_n_mem_cells(int *n_addr_cells, int *n_size_cells) |
1da177e4 LT |
318 | { |
319 | struct device_node *memory = NULL; | |
1da177e4 LT |
320 | |
321 | memory = of_find_node_by_type(memory, "memory"); | |
54c23310 | 322 | if (!memory) |
84c9fdd1 | 323 | panic("numa.c: No memory nodes found!"); |
54c23310 | 324 | |
a8bda5dd | 325 | *n_addr_cells = of_n_addr_cells(memory); |
9213feea | 326 | *n_size_cells = of_n_size_cells(memory); |
84c9fdd1 | 327 | of_node_put(memory); |
1da177e4 LT |
328 | } |
329 | ||
a7f67bdf | 330 | static unsigned long __devinit read_n_cells(int n, const unsigned int **buf) |
1da177e4 LT |
331 | { |
332 | unsigned long result = 0; | |
333 | ||
334 | while (n--) { | |
335 | result = (result << 32) | **buf; | |
336 | (*buf)++; | |
337 | } | |
338 | return result; | |
339 | } | |
340 | ||
8342681d NF |
341 | struct of_drconf_cell { |
342 | u64 base_addr; | |
343 | u32 drc_index; | |
344 | u32 reserved; | |
345 | u32 aa_index; | |
346 | u32 flags; | |
347 | }; | |
348 | ||
349 | #define DRCONF_MEM_ASSIGNED 0x00000008 | |
350 | #define DRCONF_MEM_AI_INVALID 0x00000040 | |
351 | #define DRCONF_MEM_RESERVED 0x00000080 | |
352 | ||
353 | /* | |
95f72d1e | 354 | * Read the next memblock list entry from the ibm,dynamic-memory property |
8342681d NF |
355 | * and return the information in the provided of_drconf_cell structure. |
356 | */ | |
357 | static void read_drconf_cell(struct of_drconf_cell *drmem, const u32 **cellp) | |
358 | { | |
359 | const u32 *cp; | |
360 | ||
361 | drmem->base_addr = read_n_cells(n_mem_addr_cells, cellp); | |
362 | ||
363 | cp = *cellp; | |
364 | drmem->drc_index = cp[0]; | |
365 | drmem->reserved = cp[1]; | |
366 | drmem->aa_index = cp[2]; | |
367 | drmem->flags = cp[3]; | |
368 | ||
369 | *cellp = cp + 4; | |
370 | } | |
371 | ||
372 | /* | |
373 | * Retreive and validate the ibm,dynamic-memory property of the device tree. | |
374 | * | |
95f72d1e YL |
375 | * The layout of the ibm,dynamic-memory property is a number N of memblock |
376 | * list entries followed by N memblock list entries. Each memblock list entry | |
8342681d NF |
377 | * contains information as layed out in the of_drconf_cell struct above. |
378 | */ | |
379 | static int of_get_drconf_memory(struct device_node *memory, const u32 **dm) | |
380 | { | |
381 | const u32 *prop; | |
382 | u32 len, entries; | |
383 | ||
384 | prop = of_get_property(memory, "ibm,dynamic-memory", &len); | |
385 | if (!prop || len < sizeof(unsigned int)) | |
386 | return 0; | |
387 | ||
388 | entries = *prop++; | |
389 | ||
390 | /* Now that we know the number of entries, revalidate the size | |
391 | * of the property read in to ensure we have everything | |
392 | */ | |
393 | if (len < (entries * (n_mem_addr_cells + 4) + 1) * sizeof(unsigned int)) | |
394 | return 0; | |
395 | ||
396 | *dm = prop; | |
397 | return entries; | |
398 | } | |
399 | ||
400 | /* | |
3fdfd990 | 401 | * Retreive and validate the ibm,lmb-size property for drconf memory |
8342681d NF |
402 | * from the device tree. |
403 | */ | |
3fdfd990 | 404 | static u64 of_get_lmb_size(struct device_node *memory) |
8342681d NF |
405 | { |
406 | const u32 *prop; | |
407 | u32 len; | |
408 | ||
3fdfd990 | 409 | prop = of_get_property(memory, "ibm,lmb-size", &len); |
8342681d NF |
410 | if (!prop || len < sizeof(unsigned int)) |
411 | return 0; | |
412 | ||
413 | return read_n_cells(n_mem_size_cells, &prop); | |
414 | } | |
415 | ||
416 | struct assoc_arrays { | |
417 | u32 n_arrays; | |
418 | u32 array_sz; | |
419 | const u32 *arrays; | |
420 | }; | |
421 | ||
422 | /* | |
423 | * Retreive and validate the list of associativity arrays for drconf | |
424 | * memory from the ibm,associativity-lookup-arrays property of the | |
425 | * device tree.. | |
426 | * | |
427 | * The layout of the ibm,associativity-lookup-arrays property is a number N | |
428 | * indicating the number of associativity arrays, followed by a number M | |
429 | * indicating the size of each associativity array, followed by a list | |
430 | * of N associativity arrays. | |
431 | */ | |
432 | static int of_get_assoc_arrays(struct device_node *memory, | |
433 | struct assoc_arrays *aa) | |
434 | { | |
435 | const u32 *prop; | |
436 | u32 len; | |
437 | ||
438 | prop = of_get_property(memory, "ibm,associativity-lookup-arrays", &len); | |
439 | if (!prop || len < 2 * sizeof(unsigned int)) | |
440 | return -1; | |
441 | ||
442 | aa->n_arrays = *prop++; | |
443 | aa->array_sz = *prop++; | |
444 | ||
445 | /* Now that we know the number of arrrays and size of each array, | |
446 | * revalidate the size of the property read in. | |
447 | */ | |
448 | if (len < (aa->n_arrays * aa->array_sz + 2) * sizeof(unsigned int)) | |
449 | return -1; | |
450 | ||
451 | aa->arrays = prop; | |
452 | return 0; | |
453 | } | |
454 | ||
455 | /* | |
456 | * This is like of_node_to_nid_single() for memory represented in the | |
457 | * ibm,dynamic-reconfiguration-memory node. | |
458 | */ | |
459 | static int of_drconf_to_nid_single(struct of_drconf_cell *drmem, | |
460 | struct assoc_arrays *aa) | |
461 | { | |
462 | int default_nid = 0; | |
463 | int nid = default_nid; | |
464 | int index; | |
465 | ||
466 | if (min_common_depth > 0 && min_common_depth <= aa->array_sz && | |
467 | !(drmem->flags & DRCONF_MEM_AI_INVALID) && | |
468 | drmem->aa_index < aa->n_arrays) { | |
469 | index = drmem->aa_index * aa->array_sz + min_common_depth - 1; | |
470 | nid = aa->arrays[index]; | |
471 | ||
472 | if (nid == 0xffff || nid >= MAX_NUMNODES) | |
473 | nid = default_nid; | |
474 | } | |
475 | ||
476 | return nid; | |
477 | } | |
478 | ||
1da177e4 LT |
479 | /* |
480 | * Figure out to which domain a cpu belongs and stick it there. | |
481 | * Return the id of the domain used. | |
482 | */ | |
2e5ce39d | 483 | static int __cpuinit numa_setup_cpu(unsigned long lcpu) |
1da177e4 | 484 | { |
cf950b7a | 485 | int nid = 0; |
8b16cd23 | 486 | struct device_node *cpu = of_get_cpu_node(lcpu, NULL); |
1da177e4 LT |
487 | |
488 | if (!cpu) { | |
489 | WARN_ON(1); | |
490 | goto out; | |
491 | } | |
492 | ||
953039c8 | 493 | nid = of_node_to_nid_single(cpu); |
1da177e4 | 494 | |
482ec7c4 | 495 | if (nid < 0 || !node_online(nid)) |
72c33688 | 496 | nid = first_online_node; |
1da177e4 | 497 | out: |
cf950b7a | 498 | map_cpu_to_node(lcpu, nid); |
1da177e4 LT |
499 | |
500 | of_node_put(cpu); | |
501 | ||
cf950b7a | 502 | return nid; |
1da177e4 LT |
503 | } |
504 | ||
74b85f37 | 505 | static int __cpuinit cpu_numa_callback(struct notifier_block *nfb, |
1da177e4 LT |
506 | unsigned long action, |
507 | void *hcpu) | |
508 | { | |
509 | unsigned long lcpu = (unsigned long)hcpu; | |
510 | int ret = NOTIFY_DONE; | |
511 | ||
512 | switch (action) { | |
513 | case CPU_UP_PREPARE: | |
8bb78442 | 514 | case CPU_UP_PREPARE_FROZEN: |
2b261227 | 515 | numa_setup_cpu(lcpu); |
1da177e4 LT |
516 | ret = NOTIFY_OK; |
517 | break; | |
518 | #ifdef CONFIG_HOTPLUG_CPU | |
519 | case CPU_DEAD: | |
8bb78442 | 520 | case CPU_DEAD_FROZEN: |
1da177e4 | 521 | case CPU_UP_CANCELED: |
8bb78442 | 522 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 LT |
523 | unmap_cpu_from_node(lcpu); |
524 | break; | |
525 | ret = NOTIFY_OK; | |
526 | #endif | |
527 | } | |
528 | return ret; | |
529 | } | |
530 | ||
531 | /* | |
532 | * Check and possibly modify a memory region to enforce the memory limit. | |
533 | * | |
534 | * Returns the size the region should have to enforce the memory limit. | |
535 | * This will either be the original value of size, a truncated value, | |
536 | * or zero. If the returned value of size is 0 the region should be | |
537 | * discarded as it lies wholy above the memory limit. | |
538 | */ | |
45fb6cea AB |
539 | static unsigned long __init numa_enforce_memory_limit(unsigned long start, |
540 | unsigned long size) | |
1da177e4 LT |
541 | { |
542 | /* | |
95f72d1e | 543 | * We use memblock_end_of_DRAM() in here instead of memory_limit because |
1da177e4 | 544 | * we've already adjusted it for the limit and it takes care of |
fe55249d MM |
545 | * having memory holes below the limit. Also, in the case of |
546 | * iommu_is_off, memory_limit is not set but is implicitly enforced. | |
1da177e4 | 547 | */ |
1da177e4 | 548 | |
95f72d1e | 549 | if (start + size <= memblock_end_of_DRAM()) |
1da177e4 LT |
550 | return size; |
551 | ||
95f72d1e | 552 | if (start >= memblock_end_of_DRAM()) |
1da177e4 LT |
553 | return 0; |
554 | ||
95f72d1e | 555 | return memblock_end_of_DRAM() - start; |
1da177e4 LT |
556 | } |
557 | ||
cf00085d C |
558 | /* |
559 | * Reads the counter for a given entry in | |
560 | * linux,drconf-usable-memory property | |
561 | */ | |
562 | static inline int __init read_usm_ranges(const u32 **usm) | |
563 | { | |
564 | /* | |
3fdfd990 | 565 | * For each lmb in ibm,dynamic-memory a corresponding |
cf00085d C |
566 | * entry in linux,drconf-usable-memory property contains |
567 | * a counter followed by that many (base, size) duple. | |
568 | * read the counter from linux,drconf-usable-memory | |
569 | */ | |
570 | return read_n_cells(n_mem_size_cells, usm); | |
571 | } | |
572 | ||
0204568a PM |
573 | /* |
574 | * Extract NUMA information from the ibm,dynamic-reconfiguration-memory | |
575 | * node. This assumes n_mem_{addr,size}_cells have been set. | |
576 | */ | |
577 | static void __init parse_drconf_memory(struct device_node *memory) | |
578 | { | |
cf00085d C |
579 | const u32 *dm, *usm; |
580 | unsigned int n, rc, ranges, is_kexec_kdump = 0; | |
3fdfd990 | 581 | unsigned long lmb_size, base, size, sz; |
8342681d NF |
582 | int nid; |
583 | struct assoc_arrays aa; | |
584 | ||
585 | n = of_get_drconf_memory(memory, &dm); | |
586 | if (!n) | |
0204568a PM |
587 | return; |
588 | ||
3fdfd990 BH |
589 | lmb_size = of_get_lmb_size(memory); |
590 | if (!lmb_size) | |
8342681d NF |
591 | return; |
592 | ||
593 | rc = of_get_assoc_arrays(memory, &aa); | |
594 | if (rc) | |
0204568a PM |
595 | return; |
596 | ||
cf00085d C |
597 | /* check if this is a kexec/kdump kernel */ |
598 | usm = of_get_usable_memory(memory); | |
599 | if (usm != NULL) | |
600 | is_kexec_kdump = 1; | |
601 | ||
0204568a | 602 | for (; n != 0; --n) { |
8342681d NF |
603 | struct of_drconf_cell drmem; |
604 | ||
605 | read_drconf_cell(&drmem, &dm); | |
606 | ||
607 | /* skip this block if the reserved bit is set in flags (0x80) | |
608 | or if the block is not assigned to this partition (0x8) */ | |
609 | if ((drmem.flags & DRCONF_MEM_RESERVED) | |
610 | || !(drmem.flags & DRCONF_MEM_ASSIGNED)) | |
0204568a | 611 | continue; |
1daa6d08 | 612 | |
cf00085d | 613 | base = drmem.base_addr; |
3fdfd990 | 614 | size = lmb_size; |
cf00085d | 615 | ranges = 1; |
8342681d | 616 | |
cf00085d C |
617 | if (is_kexec_kdump) { |
618 | ranges = read_usm_ranges(&usm); | |
619 | if (!ranges) /* there are no (base, size) duple */ | |
620 | continue; | |
621 | } | |
622 | do { | |
623 | if (is_kexec_kdump) { | |
624 | base = read_n_cells(n_mem_addr_cells, &usm); | |
625 | size = read_n_cells(n_mem_size_cells, &usm); | |
626 | } | |
627 | nid = of_drconf_to_nid_single(&drmem, &aa); | |
628 | fake_numa_create_new_node( | |
629 | ((base + size) >> PAGE_SHIFT), | |
8342681d | 630 | &nid); |
cf00085d C |
631 | node_set_online(nid); |
632 | sz = numa_enforce_memory_limit(base, size); | |
633 | if (sz) | |
634 | add_active_range(nid, base >> PAGE_SHIFT, | |
635 | (base >> PAGE_SHIFT) | |
636 | + (sz >> PAGE_SHIFT)); | |
637 | } while (--ranges); | |
0204568a PM |
638 | } |
639 | } | |
640 | ||
1da177e4 LT |
641 | static int __init parse_numa_properties(void) |
642 | { | |
643 | struct device_node *cpu = NULL; | |
644 | struct device_node *memory = NULL; | |
482ec7c4 | 645 | int default_nid = 0; |
1da177e4 LT |
646 | unsigned long i; |
647 | ||
648 | if (numa_enabled == 0) { | |
649 | printk(KERN_WARNING "NUMA disabled by user\n"); | |
650 | return -1; | |
651 | } | |
652 | ||
1da177e4 LT |
653 | min_common_depth = find_min_common_depth(); |
654 | ||
1da177e4 LT |
655 | if (min_common_depth < 0) |
656 | return min_common_depth; | |
657 | ||
bf4b85b0 NL |
658 | dbg("NUMA associativity depth for CPU/Memory: %d\n", min_common_depth); |
659 | ||
1da177e4 | 660 | /* |
482ec7c4 NL |
661 | * Even though we connect cpus to numa domains later in SMP |
662 | * init, we need to know the node ids now. This is because | |
663 | * each node to be onlined must have NODE_DATA etc backing it. | |
1da177e4 | 664 | */ |
482ec7c4 | 665 | for_each_present_cpu(i) { |
cf950b7a | 666 | int nid; |
1da177e4 | 667 | |
8b16cd23 | 668 | cpu = of_get_cpu_node(i, NULL); |
482ec7c4 | 669 | BUG_ON(!cpu); |
953039c8 | 670 | nid = of_node_to_nid_single(cpu); |
482ec7c4 | 671 | of_node_put(cpu); |
1da177e4 | 672 | |
482ec7c4 NL |
673 | /* |
674 | * Don't fall back to default_nid yet -- we will plug | |
675 | * cpus into nodes once the memory scan has discovered | |
676 | * the topology. | |
677 | */ | |
678 | if (nid < 0) | |
679 | continue; | |
680 | node_set_online(nid); | |
1da177e4 LT |
681 | } |
682 | ||
237a0989 | 683 | get_n_mem_cells(&n_mem_addr_cells, &n_mem_size_cells); |
1da177e4 LT |
684 | memory = NULL; |
685 | while ((memory = of_find_node_by_type(memory, "memory")) != NULL) { | |
686 | unsigned long start; | |
687 | unsigned long size; | |
cf950b7a | 688 | int nid; |
1da177e4 | 689 | int ranges; |
a7f67bdf | 690 | const unsigned int *memcell_buf; |
1da177e4 LT |
691 | unsigned int len; |
692 | ||
e2eb6392 | 693 | memcell_buf = of_get_property(memory, |
ba759485 ME |
694 | "linux,usable-memory", &len); |
695 | if (!memcell_buf || len <= 0) | |
e2eb6392 | 696 | memcell_buf = of_get_property(memory, "reg", &len); |
1da177e4 LT |
697 | if (!memcell_buf || len <= 0) |
698 | continue; | |
699 | ||
cc5d0189 BH |
700 | /* ranges in cell */ |
701 | ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells); | |
1da177e4 LT |
702 | new_range: |
703 | /* these are order-sensitive, and modify the buffer pointer */ | |
237a0989 MK |
704 | start = read_n_cells(n_mem_addr_cells, &memcell_buf); |
705 | size = read_n_cells(n_mem_size_cells, &memcell_buf); | |
1da177e4 | 706 | |
482ec7c4 NL |
707 | /* |
708 | * Assumption: either all memory nodes or none will | |
709 | * have associativity properties. If none, then | |
710 | * everything goes to default_nid. | |
711 | */ | |
953039c8 | 712 | nid = of_node_to_nid_single(memory); |
482ec7c4 NL |
713 | if (nid < 0) |
714 | nid = default_nid; | |
1daa6d08 BS |
715 | |
716 | fake_numa_create_new_node(((start + size) >> PAGE_SHIFT), &nid); | |
482ec7c4 | 717 | node_set_online(nid); |
1da177e4 | 718 | |
45fb6cea | 719 | if (!(size = numa_enforce_memory_limit(start, size))) { |
1da177e4 LT |
720 | if (--ranges) |
721 | goto new_range; | |
722 | else | |
723 | continue; | |
724 | } | |
725 | ||
c67c3cb4 MG |
726 | add_active_range(nid, start >> PAGE_SHIFT, |
727 | (start >> PAGE_SHIFT) + (size >> PAGE_SHIFT)); | |
1da177e4 LT |
728 | |
729 | if (--ranges) | |
730 | goto new_range; | |
731 | } | |
732 | ||
0204568a | 733 | /* |
95f72d1e | 734 | * Now do the same thing for each MEMBLOCK listed in the ibm,dynamic-memory |
0204568a PM |
735 | * property in the ibm,dynamic-reconfiguration-memory node. |
736 | */ | |
737 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); | |
738 | if (memory) | |
739 | parse_drconf_memory(memory); | |
740 | ||
1da177e4 LT |
741 | return 0; |
742 | } | |
743 | ||
744 | static void __init setup_nonnuma(void) | |
745 | { | |
95f72d1e YL |
746 | unsigned long top_of_ram = memblock_end_of_DRAM(); |
747 | unsigned long total_ram = memblock_phys_mem_size(); | |
c67c3cb4 | 748 | unsigned long start_pfn, end_pfn; |
1daa6d08 | 749 | unsigned int i, nid = 0; |
1da177e4 | 750 | |
e110b281 | 751 | printk(KERN_DEBUG "Top of RAM: 0x%lx, Total RAM: 0x%lx\n", |
1da177e4 | 752 | top_of_ram, total_ram); |
e110b281 | 753 | printk(KERN_DEBUG "Memory hole size: %ldMB\n", |
1da177e4 LT |
754 | (top_of_ram - total_ram) >> 20); |
755 | ||
95f72d1e YL |
756 | for (i = 0; i < memblock.memory.cnt; ++i) { |
757 | start_pfn = memblock.memory.region[i].base >> PAGE_SHIFT; | |
758 | end_pfn = start_pfn + memblock_size_pages(&memblock.memory, i); | |
1daa6d08 BS |
759 | |
760 | fake_numa_create_new_node(end_pfn, &nid); | |
761 | add_active_range(nid, start_pfn, end_pfn); | |
762 | node_set_online(nid); | |
c67c3cb4 | 763 | } |
1da177e4 LT |
764 | } |
765 | ||
4b703a23 AB |
766 | void __init dump_numa_cpu_topology(void) |
767 | { | |
768 | unsigned int node; | |
769 | unsigned int cpu, count; | |
770 | ||
771 | if (min_common_depth == -1 || !numa_enabled) | |
772 | return; | |
773 | ||
774 | for_each_online_node(node) { | |
e110b281 | 775 | printk(KERN_DEBUG "Node %d CPUs:", node); |
4b703a23 AB |
776 | |
777 | count = 0; | |
778 | /* | |
779 | * If we used a CPU iterator here we would miss printing | |
780 | * the holes in the cpumap. | |
781 | */ | |
25863de0 AB |
782 | for (cpu = 0; cpu < nr_cpu_ids; cpu++) { |
783 | if (cpumask_test_cpu(cpu, | |
784 | node_to_cpumask_map[node])) { | |
4b703a23 AB |
785 | if (count == 0) |
786 | printk(" %u", cpu); | |
787 | ++count; | |
788 | } else { | |
789 | if (count > 1) | |
790 | printk("-%u", cpu - 1); | |
791 | count = 0; | |
792 | } | |
793 | } | |
794 | ||
795 | if (count > 1) | |
25863de0 | 796 | printk("-%u", nr_cpu_ids - 1); |
4b703a23 AB |
797 | printk("\n"); |
798 | } | |
799 | } | |
800 | ||
801 | static void __init dump_numa_memory_topology(void) | |
1da177e4 LT |
802 | { |
803 | unsigned int node; | |
804 | unsigned int count; | |
805 | ||
806 | if (min_common_depth == -1 || !numa_enabled) | |
807 | return; | |
808 | ||
809 | for_each_online_node(node) { | |
810 | unsigned long i; | |
811 | ||
e110b281 | 812 | printk(KERN_DEBUG "Node %d Memory:", node); |
1da177e4 LT |
813 | |
814 | count = 0; | |
815 | ||
95f72d1e | 816 | for (i = 0; i < memblock_end_of_DRAM(); |
45fb6cea AB |
817 | i += (1 << SECTION_SIZE_BITS)) { |
818 | if (early_pfn_to_nid(i >> PAGE_SHIFT) == node) { | |
1da177e4 LT |
819 | if (count == 0) |
820 | printk(" 0x%lx", i); | |
821 | ++count; | |
822 | } else { | |
823 | if (count > 0) | |
824 | printk("-0x%lx", i); | |
825 | count = 0; | |
826 | } | |
827 | } | |
828 | ||
829 | if (count > 0) | |
830 | printk("-0x%lx", i); | |
831 | printk("\n"); | |
832 | } | |
1da177e4 LT |
833 | } |
834 | ||
835 | /* | |
95f72d1e | 836 | * Allocate some memory, satisfying the memblock or bootmem allocator where |
1da177e4 LT |
837 | * required. nid is the preferred node and end is the physical address of |
838 | * the highest address in the node. | |
839 | * | |
0be210fd | 840 | * Returns the virtual address of the memory. |
1da177e4 | 841 | */ |
893473df | 842 | static void __init *careful_zallocation(int nid, unsigned long size, |
45fb6cea AB |
843 | unsigned long align, |
844 | unsigned long end_pfn) | |
1da177e4 | 845 | { |
0be210fd | 846 | void *ret; |
45fb6cea | 847 | int new_nid; |
0be210fd DH |
848 | unsigned long ret_paddr; |
849 | ||
95f72d1e | 850 | ret_paddr = __memblock_alloc_base(size, align, end_pfn << PAGE_SHIFT); |
1da177e4 LT |
851 | |
852 | /* retry over all memory */ | |
0be210fd | 853 | if (!ret_paddr) |
95f72d1e | 854 | ret_paddr = __memblock_alloc_base(size, align, memblock_end_of_DRAM()); |
1da177e4 | 855 | |
0be210fd | 856 | if (!ret_paddr) |
5d21ea2b | 857 | panic("numa.c: cannot allocate %lu bytes for node %d", |
1da177e4 LT |
858 | size, nid); |
859 | ||
0be210fd DH |
860 | ret = __va(ret_paddr); |
861 | ||
1da177e4 | 862 | /* |
c555e520 | 863 | * We initialize the nodes in numeric order: 0, 1, 2... |
95f72d1e | 864 | * and hand over control from the MEMBLOCK allocator to the |
c555e520 DH |
865 | * bootmem allocator. If this function is called for |
866 | * node 5, then we know that all nodes <5 are using the | |
95f72d1e | 867 | * bootmem allocator instead of the MEMBLOCK allocator. |
c555e520 DH |
868 | * |
869 | * So, check the nid from which this allocation came | |
870 | * and double check to see if we need to use bootmem | |
95f72d1e | 871 | * instead of the MEMBLOCK. We don't free the MEMBLOCK memory |
c555e520 | 872 | * since it would be useless. |
1da177e4 | 873 | */ |
0be210fd | 874 | new_nid = early_pfn_to_nid(ret_paddr >> PAGE_SHIFT); |
45fb6cea | 875 | if (new_nid < nid) { |
0be210fd | 876 | ret = __alloc_bootmem_node(NODE_DATA(new_nid), |
1da177e4 LT |
877 | size, align, 0); |
878 | ||
0be210fd | 879 | dbg("alloc_bootmem %p %lx\n", ret, size); |
1da177e4 LT |
880 | } |
881 | ||
893473df | 882 | memset(ret, 0, size); |
0be210fd | 883 | return ret; |
1da177e4 LT |
884 | } |
885 | ||
74b85f37 CS |
886 | static struct notifier_block __cpuinitdata ppc64_numa_nb = { |
887 | .notifier_call = cpu_numa_callback, | |
888 | .priority = 1 /* Must run before sched domains notifier. */ | |
889 | }; | |
890 | ||
4a618669 DH |
891 | static void mark_reserved_regions_for_nid(int nid) |
892 | { | |
893 | struct pglist_data *node = NODE_DATA(nid); | |
894 | int i; | |
895 | ||
95f72d1e YL |
896 | for (i = 0; i < memblock.reserved.cnt; i++) { |
897 | unsigned long physbase = memblock.reserved.region[i].base; | |
898 | unsigned long size = memblock.reserved.region[i].size; | |
4a618669 | 899 | unsigned long start_pfn = physbase >> PAGE_SHIFT; |
06eccea6 | 900 | unsigned long end_pfn = PFN_UP(physbase + size); |
4a618669 DH |
901 | struct node_active_region node_ar; |
902 | unsigned long node_end_pfn = node->node_start_pfn + | |
903 | node->node_spanned_pages; | |
904 | ||
905 | /* | |
95f72d1e | 906 | * Check to make sure that this memblock.reserved area is |
4a618669 DH |
907 | * within the bounds of the node that we care about. |
908 | * Checking the nid of the start and end points is not | |
909 | * sufficient because the reserved area could span the | |
910 | * entire node. | |
911 | */ | |
912 | if (end_pfn <= node->node_start_pfn || | |
913 | start_pfn >= node_end_pfn) | |
914 | continue; | |
915 | ||
916 | get_node_active_region(start_pfn, &node_ar); | |
917 | while (start_pfn < end_pfn && | |
918 | node_ar.start_pfn < node_ar.end_pfn) { | |
919 | unsigned long reserve_size = size; | |
920 | /* | |
921 | * if reserved region extends past active region | |
922 | * then trim size to active region | |
923 | */ | |
924 | if (end_pfn > node_ar.end_pfn) | |
925 | reserve_size = (node_ar.end_pfn << PAGE_SHIFT) | |
06eccea6 | 926 | - physbase; |
a4c74ddd DH |
927 | /* |
928 | * Only worry about *this* node, others may not | |
929 | * yet have valid NODE_DATA(). | |
930 | */ | |
931 | if (node_ar.nid == nid) { | |
932 | dbg("reserve_bootmem %lx %lx nid=%d\n", | |
933 | physbase, reserve_size, node_ar.nid); | |
934 | reserve_bootmem_node(NODE_DATA(node_ar.nid), | |
935 | physbase, reserve_size, | |
936 | BOOTMEM_DEFAULT); | |
937 | } | |
4a618669 DH |
938 | /* |
939 | * if reserved region is contained in the active region | |
940 | * then done. | |
941 | */ | |
942 | if (end_pfn <= node_ar.end_pfn) | |
943 | break; | |
944 | ||
945 | /* | |
946 | * reserved region extends past the active region | |
947 | * get next active region that contains this | |
948 | * reserved region | |
949 | */ | |
950 | start_pfn = node_ar.end_pfn; | |
951 | physbase = start_pfn << PAGE_SHIFT; | |
952 | size = size - reserve_size; | |
953 | get_node_active_region(start_pfn, &node_ar); | |
954 | } | |
955 | } | |
956 | } | |
957 | ||
958 | ||
1da177e4 LT |
959 | void __init do_init_bootmem(void) |
960 | { | |
961 | int nid; | |
1da177e4 LT |
962 | |
963 | min_low_pfn = 0; | |
95f72d1e | 964 | max_low_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT; |
1da177e4 LT |
965 | max_pfn = max_low_pfn; |
966 | ||
967 | if (parse_numa_properties()) | |
968 | setup_nonnuma(); | |
969 | else | |
4b703a23 | 970 | dump_numa_memory_topology(); |
1da177e4 | 971 | |
1da177e4 | 972 | for_each_online_node(nid) { |
c67c3cb4 | 973 | unsigned long start_pfn, end_pfn; |
0be210fd | 974 | void *bootmem_vaddr; |
1da177e4 LT |
975 | unsigned long bootmap_pages; |
976 | ||
c67c3cb4 | 977 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); |
1da177e4 | 978 | |
4a618669 DH |
979 | /* |
980 | * Allocate the node structure node local if possible | |
981 | * | |
982 | * Be careful moving this around, as it relies on all | |
983 | * previous nodes' bootmem to be initialized and have | |
984 | * all reserved areas marked. | |
985 | */ | |
893473df | 986 | NODE_DATA(nid) = careful_zallocation(nid, |
1da177e4 | 987 | sizeof(struct pglist_data), |
45fb6cea | 988 | SMP_CACHE_BYTES, end_pfn); |
1da177e4 LT |
989 | |
990 | dbg("node %d\n", nid); | |
991 | dbg("NODE_DATA() = %p\n", NODE_DATA(nid)); | |
992 | ||
b61bfa3c | 993 | NODE_DATA(nid)->bdata = &bootmem_node_data[nid]; |
45fb6cea AB |
994 | NODE_DATA(nid)->node_start_pfn = start_pfn; |
995 | NODE_DATA(nid)->node_spanned_pages = end_pfn - start_pfn; | |
1da177e4 LT |
996 | |
997 | if (NODE_DATA(nid)->node_spanned_pages == 0) | |
998 | continue; | |
999 | ||
45fb6cea AB |
1000 | dbg("start_paddr = %lx\n", start_pfn << PAGE_SHIFT); |
1001 | dbg("end_paddr = %lx\n", end_pfn << PAGE_SHIFT); | |
1da177e4 | 1002 | |
45fb6cea | 1003 | bootmap_pages = bootmem_bootmap_pages(end_pfn - start_pfn); |
893473df | 1004 | bootmem_vaddr = careful_zallocation(nid, |
45fb6cea AB |
1005 | bootmap_pages << PAGE_SHIFT, |
1006 | PAGE_SIZE, end_pfn); | |
1da177e4 | 1007 | |
0be210fd | 1008 | dbg("bootmap_vaddr = %p\n", bootmem_vaddr); |
1da177e4 | 1009 | |
0be210fd DH |
1010 | init_bootmem_node(NODE_DATA(nid), |
1011 | __pa(bootmem_vaddr) >> PAGE_SHIFT, | |
45fb6cea | 1012 | start_pfn, end_pfn); |
1da177e4 | 1013 | |
c67c3cb4 | 1014 | free_bootmem_with_active_regions(nid, end_pfn); |
4a618669 DH |
1015 | /* |
1016 | * Be very careful about moving this around. Future | |
893473df | 1017 | * calls to careful_zallocation() depend on this getting |
4a618669 DH |
1018 | * done correctly. |
1019 | */ | |
1020 | mark_reserved_regions_for_nid(nid); | |
8f64e1f2 | 1021 | sparse_memory_present_with_active_regions(nid); |
4a618669 | 1022 | } |
d3f6204a BH |
1023 | |
1024 | init_bootmem_done = 1; | |
25863de0 AB |
1025 | |
1026 | /* | |
1027 | * Now bootmem is initialised we can create the node to cpumask | |
1028 | * lookup tables and setup the cpu callback to populate them. | |
1029 | */ | |
1030 | setup_node_to_cpumask_map(); | |
1031 | ||
1032 | register_cpu_notifier(&ppc64_numa_nb); | |
1033 | cpu_numa_callback(&ppc64_numa_nb, CPU_UP_PREPARE, | |
1034 | (void *)(unsigned long)boot_cpuid); | |
1da177e4 LT |
1035 | } |
1036 | ||
1037 | void __init paging_init(void) | |
1038 | { | |
6391af17 MG |
1039 | unsigned long max_zone_pfns[MAX_NR_ZONES]; |
1040 | memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); | |
95f72d1e | 1041 | max_zone_pfns[ZONE_DMA] = memblock_end_of_DRAM() >> PAGE_SHIFT; |
c67c3cb4 | 1042 | free_area_init_nodes(max_zone_pfns); |
1da177e4 LT |
1043 | } |
1044 | ||
1045 | static int __init early_numa(char *p) | |
1046 | { | |
1047 | if (!p) | |
1048 | return 0; | |
1049 | ||
1050 | if (strstr(p, "off")) | |
1051 | numa_enabled = 0; | |
1052 | ||
1053 | if (strstr(p, "debug")) | |
1054 | numa_debug = 1; | |
1055 | ||
1daa6d08 BS |
1056 | p = strstr(p, "fake="); |
1057 | if (p) | |
1058 | cmdline = p + strlen("fake="); | |
1059 | ||
1da177e4 LT |
1060 | return 0; |
1061 | } | |
1062 | early_param("numa", early_numa); | |
237a0989 MK |
1063 | |
1064 | #ifdef CONFIG_MEMORY_HOTPLUG | |
0db9360a | 1065 | /* |
0f16ef7f NF |
1066 | * Find the node associated with a hot added memory section for |
1067 | * memory represented in the device tree by the property | |
1068 | * ibm,dynamic-reconfiguration-memory/ibm,dynamic-memory. | |
0db9360a NF |
1069 | */ |
1070 | static int hot_add_drconf_scn_to_nid(struct device_node *memory, | |
1071 | unsigned long scn_addr) | |
1072 | { | |
1073 | const u32 *dm; | |
0f16ef7f | 1074 | unsigned int drconf_cell_cnt, rc; |
3fdfd990 | 1075 | unsigned long lmb_size; |
0db9360a | 1076 | struct assoc_arrays aa; |
0f16ef7f | 1077 | int nid = -1; |
0db9360a | 1078 | |
0f16ef7f NF |
1079 | drconf_cell_cnt = of_get_drconf_memory(memory, &dm); |
1080 | if (!drconf_cell_cnt) | |
1081 | return -1; | |
0db9360a | 1082 | |
3fdfd990 BH |
1083 | lmb_size = of_get_lmb_size(memory); |
1084 | if (!lmb_size) | |
0f16ef7f | 1085 | return -1; |
0db9360a NF |
1086 | |
1087 | rc = of_get_assoc_arrays(memory, &aa); | |
1088 | if (rc) | |
0f16ef7f | 1089 | return -1; |
0db9360a | 1090 | |
0f16ef7f | 1091 | for (; drconf_cell_cnt != 0; --drconf_cell_cnt) { |
0db9360a NF |
1092 | struct of_drconf_cell drmem; |
1093 | ||
1094 | read_drconf_cell(&drmem, &dm); | |
1095 | ||
1096 | /* skip this block if it is reserved or not assigned to | |
1097 | * this partition */ | |
1098 | if ((drmem.flags & DRCONF_MEM_RESERVED) | |
1099 | || !(drmem.flags & DRCONF_MEM_ASSIGNED)) | |
1100 | continue; | |
1101 | ||
0f16ef7f | 1102 | if ((scn_addr < drmem.base_addr) |
3fdfd990 | 1103 | || (scn_addr >= (drmem.base_addr + lmb_size))) |
0f16ef7f NF |
1104 | continue; |
1105 | ||
0db9360a | 1106 | nid = of_drconf_to_nid_single(&drmem, &aa); |
0f16ef7f NF |
1107 | break; |
1108 | } | |
1109 | ||
1110 | return nid; | |
1111 | } | |
1112 | ||
1113 | /* | |
1114 | * Find the node associated with a hot added memory section for memory | |
1115 | * represented in the device tree as a node (i.e. memory@XXXX) for | |
95f72d1e | 1116 | * each memblock. |
0f16ef7f NF |
1117 | */ |
1118 | int hot_add_node_scn_to_nid(unsigned long scn_addr) | |
1119 | { | |
1120 | struct device_node *memory = NULL; | |
1121 | int nid = -1; | |
1122 | ||
1123 | while ((memory = of_find_node_by_type(memory, "memory")) != NULL) { | |
1124 | unsigned long start, size; | |
1125 | int ranges; | |
1126 | const unsigned int *memcell_buf; | |
1127 | unsigned int len; | |
1128 | ||
1129 | memcell_buf = of_get_property(memory, "reg", &len); | |
1130 | if (!memcell_buf || len <= 0) | |
1131 | continue; | |
1132 | ||
1133 | /* ranges in cell */ | |
1134 | ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells); | |
1135 | ||
1136 | while (ranges--) { | |
1137 | start = read_n_cells(n_mem_addr_cells, &memcell_buf); | |
1138 | size = read_n_cells(n_mem_size_cells, &memcell_buf); | |
1139 | ||
1140 | if ((scn_addr < start) || (scn_addr >= (start + size))) | |
1141 | continue; | |
1142 | ||
1143 | nid = of_node_to_nid_single(memory); | |
1144 | break; | |
1145 | } | |
0db9360a | 1146 | |
0f16ef7f NF |
1147 | of_node_put(memory); |
1148 | if (nid >= 0) | |
1149 | break; | |
0db9360a NF |
1150 | } |
1151 | ||
0f16ef7f | 1152 | return nid; |
0db9360a NF |
1153 | } |
1154 | ||
237a0989 MK |
1155 | /* |
1156 | * Find the node associated with a hot added memory section. Section | |
95f72d1e YL |
1157 | * corresponds to a SPARSEMEM section, not an MEMBLOCK. It is assumed that |
1158 | * sections are fully contained within a single MEMBLOCK. | |
237a0989 MK |
1159 | */ |
1160 | int hot_add_scn_to_nid(unsigned long scn_addr) | |
1161 | { | |
1162 | struct device_node *memory = NULL; | |
0f16ef7f | 1163 | int nid, found = 0; |
237a0989 MK |
1164 | |
1165 | if (!numa_enabled || (min_common_depth < 0)) | |
72c33688 | 1166 | return first_online_node; |
0db9360a NF |
1167 | |
1168 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); | |
1169 | if (memory) { | |
1170 | nid = hot_add_drconf_scn_to_nid(memory, scn_addr); | |
1171 | of_node_put(memory); | |
0f16ef7f NF |
1172 | } else { |
1173 | nid = hot_add_node_scn_to_nid(scn_addr); | |
0db9360a | 1174 | } |
237a0989 | 1175 | |
0f16ef7f | 1176 | if (nid < 0 || !node_online(nid)) |
72c33688 | 1177 | nid = first_online_node; |
237a0989 | 1178 | |
0f16ef7f NF |
1179 | if (NODE_DATA(nid)->node_spanned_pages) |
1180 | return nid; | |
237a0989 | 1181 | |
0f16ef7f NF |
1182 | for_each_online_node(nid) { |
1183 | if (NODE_DATA(nid)->node_spanned_pages) { | |
1184 | found = 1; | |
1185 | break; | |
237a0989 | 1186 | } |
237a0989 | 1187 | } |
0f16ef7f NF |
1188 | |
1189 | BUG_ON(!found); | |
1190 | return nid; | |
237a0989 | 1191 | } |
0f16ef7f | 1192 | |
237a0989 | 1193 | #endif /* CONFIG_MEMORY_HOTPLUG */ |