]>
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> | |
d9b2b2a2 | 20 | #include <linux/lmb.h> |
6df1646e | 21 | #include <linux/of.h> |
45fb6cea | 22 | #include <asm/sparsemem.h> |
d9b2b2a2 | 23 | #include <asm/prom.h> |
cf00a8d1 | 24 | #include <asm/system.h> |
2249ca9d | 25 | #include <asm/smp.h> |
1da177e4 LT |
26 | |
27 | static int numa_enabled = 1; | |
28 | ||
1daa6d08 BS |
29 | static char *cmdline __initdata; |
30 | ||
1da177e4 LT |
31 | static int numa_debug; |
32 | #define dbg(args...) if (numa_debug) { printk(KERN_INFO args); } | |
33 | ||
45fb6cea | 34 | int numa_cpu_lookup_table[NR_CPUS]; |
1da177e4 | 35 | cpumask_t numa_cpumask_lookup_table[MAX_NUMNODES]; |
1da177e4 | 36 | struct pglist_data *node_data[MAX_NUMNODES]; |
45fb6cea AB |
37 | |
38 | EXPORT_SYMBOL(numa_cpu_lookup_table); | |
39 | EXPORT_SYMBOL(numa_cpumask_lookup_table); | |
40 | EXPORT_SYMBOL(node_data); | |
41 | ||
42 | static bootmem_data_t __initdata plat_node_bdata[MAX_NUMNODES]; | |
1da177e4 | 43 | static int min_common_depth; |
237a0989 | 44 | static int n_mem_addr_cells, n_mem_size_cells; |
1da177e4 | 45 | |
1daa6d08 BS |
46 | static int __cpuinit fake_numa_create_new_node(unsigned long end_pfn, |
47 | unsigned int *nid) | |
48 | { | |
49 | unsigned long long mem; | |
50 | char *p = cmdline; | |
51 | static unsigned int fake_nid; | |
52 | static unsigned long long curr_boundary; | |
53 | ||
54 | /* | |
55 | * Modify node id, iff we started creating NUMA nodes | |
56 | * We want to continue from where we left of the last time | |
57 | */ | |
58 | if (fake_nid) | |
59 | *nid = fake_nid; | |
60 | /* | |
61 | * In case there are no more arguments to parse, the | |
62 | * node_id should be the same as the last fake node id | |
63 | * (we've handled this above). | |
64 | */ | |
65 | if (!p) | |
66 | return 0; | |
67 | ||
68 | mem = memparse(p, &p); | |
69 | if (!mem) | |
70 | return 0; | |
71 | ||
72 | if (mem < curr_boundary) | |
73 | return 0; | |
74 | ||
75 | curr_boundary = mem; | |
76 | ||
77 | if ((end_pfn << PAGE_SHIFT) > mem) { | |
78 | /* | |
79 | * Skip commas and spaces | |
80 | */ | |
81 | while (*p == ',' || *p == ' ' || *p == '\t') | |
82 | p++; | |
83 | ||
84 | cmdline = p; | |
85 | fake_nid++; | |
86 | *nid = fake_nid; | |
87 | dbg("created new fake_node with id %d\n", fake_nid); | |
88 | return 1; | |
89 | } | |
90 | return 0; | |
91 | } | |
92 | ||
2e5ce39d | 93 | static void __cpuinit map_cpu_to_node(int cpu, int node) |
1da177e4 LT |
94 | { |
95 | numa_cpu_lookup_table[cpu] = node; | |
45fb6cea | 96 | |
bf4b85b0 NL |
97 | dbg("adding cpu %d to node %d\n", cpu, node); |
98 | ||
45fb6cea | 99 | if (!(cpu_isset(cpu, numa_cpumask_lookup_table[node]))) |
1da177e4 | 100 | cpu_set(cpu, numa_cpumask_lookup_table[node]); |
1da177e4 LT |
101 | } |
102 | ||
103 | #ifdef CONFIG_HOTPLUG_CPU | |
104 | static void unmap_cpu_from_node(unsigned long cpu) | |
105 | { | |
106 | int node = numa_cpu_lookup_table[cpu]; | |
107 | ||
108 | dbg("removing cpu %lu from node %d\n", cpu, node); | |
109 | ||
110 | if (cpu_isset(cpu, numa_cpumask_lookup_table[node])) { | |
111 | cpu_clear(cpu, numa_cpumask_lookup_table[node]); | |
1da177e4 LT |
112 | } else { |
113 | printk(KERN_ERR "WARNING: cpu %lu not found in node %d\n", | |
114 | cpu, node); | |
115 | } | |
116 | } | |
117 | #endif /* CONFIG_HOTPLUG_CPU */ | |
118 | ||
2e5ce39d | 119 | static struct device_node * __cpuinit find_cpu_node(unsigned int cpu) |
1da177e4 LT |
120 | { |
121 | unsigned int hw_cpuid = get_hard_smp_processor_id(cpu); | |
122 | struct device_node *cpu_node = NULL; | |
a7f67bdf | 123 | const unsigned int *interrupt_server, *reg; |
1da177e4 LT |
124 | int len; |
125 | ||
126 | while ((cpu_node = of_find_node_by_type(cpu_node, "cpu")) != NULL) { | |
127 | /* Try interrupt server first */ | |
e2eb6392 | 128 | interrupt_server = of_get_property(cpu_node, |
1da177e4 LT |
129 | "ibm,ppc-interrupt-server#s", &len); |
130 | ||
131 | len = len / sizeof(u32); | |
132 | ||
133 | if (interrupt_server && (len > 0)) { | |
134 | while (len--) { | |
135 | if (interrupt_server[len] == hw_cpuid) | |
136 | return cpu_node; | |
137 | } | |
138 | } else { | |
e2eb6392 | 139 | reg = of_get_property(cpu_node, "reg", &len); |
1da177e4 LT |
140 | if (reg && (len > 0) && (reg[0] == hw_cpuid)) |
141 | return cpu_node; | |
142 | } | |
143 | } | |
144 | ||
145 | return NULL; | |
146 | } | |
147 | ||
148 | /* must hold reference to node during call */ | |
a7f67bdf | 149 | static const int *of_get_associativity(struct device_node *dev) |
1da177e4 | 150 | { |
e2eb6392 | 151 | return of_get_property(dev, "ibm,associativity", NULL); |
1da177e4 LT |
152 | } |
153 | ||
482ec7c4 NL |
154 | /* Returns nid in the range [0..MAX_NUMNODES-1], or -1 if no useful numa |
155 | * info is found. | |
156 | */ | |
953039c8 | 157 | static int of_node_to_nid_single(struct device_node *device) |
1da177e4 | 158 | { |
482ec7c4 | 159 | int nid = -1; |
a7f67bdf | 160 | const unsigned int *tmp; |
1da177e4 LT |
161 | |
162 | if (min_common_depth == -1) | |
482ec7c4 | 163 | goto out; |
1da177e4 LT |
164 | |
165 | tmp = of_get_associativity(device); | |
482ec7c4 NL |
166 | if (!tmp) |
167 | goto out; | |
168 | ||
169 | if (tmp[0] >= min_common_depth) | |
cf950b7a | 170 | nid = tmp[min_common_depth]; |
bc16a759 NL |
171 | |
172 | /* POWER4 LPAR uses 0xffff as invalid node */ | |
482ec7c4 NL |
173 | if (nid == 0xffff || nid >= MAX_NUMNODES) |
174 | nid = -1; | |
175 | out: | |
cf950b7a | 176 | return nid; |
1da177e4 LT |
177 | } |
178 | ||
953039c8 JK |
179 | /* Walk the device tree upwards, looking for an associativity id */ |
180 | int of_node_to_nid(struct device_node *device) | |
181 | { | |
182 | struct device_node *tmp; | |
183 | int nid = -1; | |
184 | ||
185 | of_node_get(device); | |
186 | while (device) { | |
187 | nid = of_node_to_nid_single(device); | |
188 | if (nid != -1) | |
189 | break; | |
190 | ||
191 | tmp = device; | |
192 | device = of_get_parent(tmp); | |
193 | of_node_put(tmp); | |
194 | } | |
195 | of_node_put(device); | |
196 | ||
197 | return nid; | |
198 | } | |
199 | EXPORT_SYMBOL_GPL(of_node_to_nid); | |
200 | ||
1da177e4 LT |
201 | /* |
202 | * In theory, the "ibm,associativity" property may contain multiple | |
203 | * associativity lists because a resource may be multiply connected | |
204 | * into the machine. This resource then has different associativity | |
205 | * characteristics relative to its multiple connections. We ignore | |
206 | * this for now. We also assume that all cpu and memory sets have | |
207 | * their distances represented at a common level. This won't be | |
1b3c3714 | 208 | * true for hierarchical NUMA. |
1da177e4 LT |
209 | * |
210 | * In any case the ibm,associativity-reference-points should give | |
211 | * the correct depth for a normal NUMA system. | |
212 | * | |
213 | * - Dave Hansen <haveblue@us.ibm.com> | |
214 | */ | |
215 | static int __init find_min_common_depth(void) | |
216 | { | |
217 | int depth; | |
a7f67bdf | 218 | const unsigned int *ref_points; |
1da177e4 LT |
219 | struct device_node *rtas_root; |
220 | unsigned int len; | |
221 | ||
222 | rtas_root = of_find_node_by_path("/rtas"); | |
223 | ||
224 | if (!rtas_root) | |
225 | return -1; | |
226 | ||
227 | /* | |
228 | * this property is 2 32-bit integers, each representing a level of | |
229 | * depth in the associativity nodes. The first is for an SMP | |
230 | * configuration (should be all 0's) and the second is for a normal | |
231 | * NUMA configuration. | |
232 | */ | |
e2eb6392 | 233 | ref_points = of_get_property(rtas_root, |
1da177e4 LT |
234 | "ibm,associativity-reference-points", &len); |
235 | ||
236 | if ((len >= 1) && ref_points) { | |
237 | depth = ref_points[1]; | |
238 | } else { | |
bf4b85b0 | 239 | dbg("NUMA: ibm,associativity-reference-points not found.\n"); |
1da177e4 LT |
240 | depth = -1; |
241 | } | |
242 | of_node_put(rtas_root); | |
243 | ||
244 | return depth; | |
245 | } | |
246 | ||
84c9fdd1 | 247 | static void __init get_n_mem_cells(int *n_addr_cells, int *n_size_cells) |
1da177e4 LT |
248 | { |
249 | struct device_node *memory = NULL; | |
1da177e4 LT |
250 | |
251 | memory = of_find_node_by_type(memory, "memory"); | |
54c23310 | 252 | if (!memory) |
84c9fdd1 | 253 | panic("numa.c: No memory nodes found!"); |
54c23310 | 254 | |
a8bda5dd | 255 | *n_addr_cells = of_n_addr_cells(memory); |
9213feea | 256 | *n_size_cells = of_n_size_cells(memory); |
84c9fdd1 | 257 | of_node_put(memory); |
1da177e4 LT |
258 | } |
259 | ||
a7f67bdf | 260 | static unsigned long __devinit read_n_cells(int n, const unsigned int **buf) |
1da177e4 LT |
261 | { |
262 | unsigned long result = 0; | |
263 | ||
264 | while (n--) { | |
265 | result = (result << 32) | **buf; | |
266 | (*buf)++; | |
267 | } | |
268 | return result; | |
269 | } | |
270 | ||
8342681d NF |
271 | struct of_drconf_cell { |
272 | u64 base_addr; | |
273 | u32 drc_index; | |
274 | u32 reserved; | |
275 | u32 aa_index; | |
276 | u32 flags; | |
277 | }; | |
278 | ||
279 | #define DRCONF_MEM_ASSIGNED 0x00000008 | |
280 | #define DRCONF_MEM_AI_INVALID 0x00000040 | |
281 | #define DRCONF_MEM_RESERVED 0x00000080 | |
282 | ||
283 | /* | |
284 | * Read the next lmb list entry from the ibm,dynamic-memory property | |
285 | * and return the information in the provided of_drconf_cell structure. | |
286 | */ | |
287 | static void read_drconf_cell(struct of_drconf_cell *drmem, const u32 **cellp) | |
288 | { | |
289 | const u32 *cp; | |
290 | ||
291 | drmem->base_addr = read_n_cells(n_mem_addr_cells, cellp); | |
292 | ||
293 | cp = *cellp; | |
294 | drmem->drc_index = cp[0]; | |
295 | drmem->reserved = cp[1]; | |
296 | drmem->aa_index = cp[2]; | |
297 | drmem->flags = cp[3]; | |
298 | ||
299 | *cellp = cp + 4; | |
300 | } | |
301 | ||
302 | /* | |
303 | * Retreive and validate the ibm,dynamic-memory property of the device tree. | |
304 | * | |
305 | * The layout of the ibm,dynamic-memory property is a number N of lmb | |
306 | * list entries followed by N lmb list entries. Each lmb list entry | |
307 | * contains information as layed out in the of_drconf_cell struct above. | |
308 | */ | |
309 | static int of_get_drconf_memory(struct device_node *memory, const u32 **dm) | |
310 | { | |
311 | const u32 *prop; | |
312 | u32 len, entries; | |
313 | ||
314 | prop = of_get_property(memory, "ibm,dynamic-memory", &len); | |
315 | if (!prop || len < sizeof(unsigned int)) | |
316 | return 0; | |
317 | ||
318 | entries = *prop++; | |
319 | ||
320 | /* Now that we know the number of entries, revalidate the size | |
321 | * of the property read in to ensure we have everything | |
322 | */ | |
323 | if (len < (entries * (n_mem_addr_cells + 4) + 1) * sizeof(unsigned int)) | |
324 | return 0; | |
325 | ||
326 | *dm = prop; | |
327 | return entries; | |
328 | } | |
329 | ||
330 | /* | |
331 | * Retreive and validate the ibm,lmb-size property for drconf memory | |
332 | * from the device tree. | |
333 | */ | |
334 | static u64 of_get_lmb_size(struct device_node *memory) | |
335 | { | |
336 | const u32 *prop; | |
337 | u32 len; | |
338 | ||
339 | prop = of_get_property(memory, "ibm,lmb-size", &len); | |
340 | if (!prop || len < sizeof(unsigned int)) | |
341 | return 0; | |
342 | ||
343 | return read_n_cells(n_mem_size_cells, &prop); | |
344 | } | |
345 | ||
346 | struct assoc_arrays { | |
347 | u32 n_arrays; | |
348 | u32 array_sz; | |
349 | const u32 *arrays; | |
350 | }; | |
351 | ||
352 | /* | |
353 | * Retreive and validate the list of associativity arrays for drconf | |
354 | * memory from the ibm,associativity-lookup-arrays property of the | |
355 | * device tree.. | |
356 | * | |
357 | * The layout of the ibm,associativity-lookup-arrays property is a number N | |
358 | * indicating the number of associativity arrays, followed by a number M | |
359 | * indicating the size of each associativity array, followed by a list | |
360 | * of N associativity arrays. | |
361 | */ | |
362 | static int of_get_assoc_arrays(struct device_node *memory, | |
363 | struct assoc_arrays *aa) | |
364 | { | |
365 | const u32 *prop; | |
366 | u32 len; | |
367 | ||
368 | prop = of_get_property(memory, "ibm,associativity-lookup-arrays", &len); | |
369 | if (!prop || len < 2 * sizeof(unsigned int)) | |
370 | return -1; | |
371 | ||
372 | aa->n_arrays = *prop++; | |
373 | aa->array_sz = *prop++; | |
374 | ||
375 | /* Now that we know the number of arrrays and size of each array, | |
376 | * revalidate the size of the property read in. | |
377 | */ | |
378 | if (len < (aa->n_arrays * aa->array_sz + 2) * sizeof(unsigned int)) | |
379 | return -1; | |
380 | ||
381 | aa->arrays = prop; | |
382 | return 0; | |
383 | } | |
384 | ||
385 | /* | |
386 | * This is like of_node_to_nid_single() for memory represented in the | |
387 | * ibm,dynamic-reconfiguration-memory node. | |
388 | */ | |
389 | static int of_drconf_to_nid_single(struct of_drconf_cell *drmem, | |
390 | struct assoc_arrays *aa) | |
391 | { | |
392 | int default_nid = 0; | |
393 | int nid = default_nid; | |
394 | int index; | |
395 | ||
396 | if (min_common_depth > 0 && min_common_depth <= aa->array_sz && | |
397 | !(drmem->flags & DRCONF_MEM_AI_INVALID) && | |
398 | drmem->aa_index < aa->n_arrays) { | |
399 | index = drmem->aa_index * aa->array_sz + min_common_depth - 1; | |
400 | nid = aa->arrays[index]; | |
401 | ||
402 | if (nid == 0xffff || nid >= MAX_NUMNODES) | |
403 | nid = default_nid; | |
404 | } | |
405 | ||
406 | return nid; | |
407 | } | |
408 | ||
1da177e4 LT |
409 | /* |
410 | * Figure out to which domain a cpu belongs and stick it there. | |
411 | * Return the id of the domain used. | |
412 | */ | |
2e5ce39d | 413 | static int __cpuinit numa_setup_cpu(unsigned long lcpu) |
1da177e4 | 414 | { |
cf950b7a | 415 | int nid = 0; |
1da177e4 LT |
416 | struct device_node *cpu = find_cpu_node(lcpu); |
417 | ||
418 | if (!cpu) { | |
419 | WARN_ON(1); | |
420 | goto out; | |
421 | } | |
422 | ||
953039c8 | 423 | nid = of_node_to_nid_single(cpu); |
1da177e4 | 424 | |
482ec7c4 NL |
425 | if (nid < 0 || !node_online(nid)) |
426 | nid = any_online_node(NODE_MASK_ALL); | |
1da177e4 | 427 | out: |
cf950b7a | 428 | map_cpu_to_node(lcpu, nid); |
1da177e4 LT |
429 | |
430 | of_node_put(cpu); | |
431 | ||
cf950b7a | 432 | return nid; |
1da177e4 LT |
433 | } |
434 | ||
74b85f37 | 435 | static int __cpuinit cpu_numa_callback(struct notifier_block *nfb, |
1da177e4 LT |
436 | unsigned long action, |
437 | void *hcpu) | |
438 | { | |
439 | unsigned long lcpu = (unsigned long)hcpu; | |
440 | int ret = NOTIFY_DONE; | |
441 | ||
442 | switch (action) { | |
443 | case CPU_UP_PREPARE: | |
8bb78442 | 444 | case CPU_UP_PREPARE_FROZEN: |
2b261227 | 445 | numa_setup_cpu(lcpu); |
1da177e4 LT |
446 | ret = NOTIFY_OK; |
447 | break; | |
448 | #ifdef CONFIG_HOTPLUG_CPU | |
449 | case CPU_DEAD: | |
8bb78442 | 450 | case CPU_DEAD_FROZEN: |
1da177e4 | 451 | case CPU_UP_CANCELED: |
8bb78442 | 452 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 LT |
453 | unmap_cpu_from_node(lcpu); |
454 | break; | |
455 | ret = NOTIFY_OK; | |
456 | #endif | |
457 | } | |
458 | return ret; | |
459 | } | |
460 | ||
461 | /* | |
462 | * Check and possibly modify a memory region to enforce the memory limit. | |
463 | * | |
464 | * Returns the size the region should have to enforce the memory limit. | |
465 | * This will either be the original value of size, a truncated value, | |
466 | * or zero. If the returned value of size is 0 the region should be | |
467 | * discarded as it lies wholy above the memory limit. | |
468 | */ | |
45fb6cea AB |
469 | static unsigned long __init numa_enforce_memory_limit(unsigned long start, |
470 | unsigned long size) | |
1da177e4 LT |
471 | { |
472 | /* | |
473 | * We use lmb_end_of_DRAM() in here instead of memory_limit because | |
474 | * we've already adjusted it for the limit and it takes care of | |
475 | * having memory holes below the limit. | |
476 | */ | |
1da177e4 LT |
477 | |
478 | if (! memory_limit) | |
479 | return size; | |
480 | ||
481 | if (start + size <= lmb_end_of_DRAM()) | |
482 | return size; | |
483 | ||
484 | if (start >= lmb_end_of_DRAM()) | |
485 | return 0; | |
486 | ||
487 | return lmb_end_of_DRAM() - start; | |
488 | } | |
489 | ||
0204568a PM |
490 | /* |
491 | * Extract NUMA information from the ibm,dynamic-reconfiguration-memory | |
492 | * node. This assumes n_mem_{addr,size}_cells have been set. | |
493 | */ | |
494 | static void __init parse_drconf_memory(struct device_node *memory) | |
495 | { | |
8342681d NF |
496 | const u32 *dm; |
497 | unsigned int n, rc; | |
498 | unsigned long lmb_size, size; | |
499 | int nid; | |
500 | struct assoc_arrays aa; | |
501 | ||
502 | n = of_get_drconf_memory(memory, &dm); | |
503 | if (!n) | |
0204568a PM |
504 | return; |
505 | ||
8342681d NF |
506 | lmb_size = of_get_lmb_size(memory); |
507 | if (!lmb_size) | |
508 | return; | |
509 | ||
510 | rc = of_get_assoc_arrays(memory, &aa); | |
511 | if (rc) | |
0204568a PM |
512 | return; |
513 | ||
514 | for (; n != 0; --n) { | |
8342681d NF |
515 | struct of_drconf_cell drmem; |
516 | ||
517 | read_drconf_cell(&drmem, &dm); | |
518 | ||
519 | /* skip this block if the reserved bit is set in flags (0x80) | |
520 | or if the block is not assigned to this partition (0x8) */ | |
521 | if ((drmem.flags & DRCONF_MEM_RESERVED) | |
522 | || !(drmem.flags & DRCONF_MEM_ASSIGNED)) | |
0204568a | 523 | continue; |
1daa6d08 | 524 | |
8342681d NF |
525 | nid = of_drconf_to_nid_single(&drmem, &aa); |
526 | ||
527 | fake_numa_create_new_node( | |
528 | ((drmem.base_addr + lmb_size) >> PAGE_SHIFT), | |
529 | &nid); | |
530 | ||
0204568a PM |
531 | node_set_online(nid); |
532 | ||
8342681d | 533 | size = numa_enforce_memory_limit(drmem.base_addr, lmb_size); |
0204568a PM |
534 | if (!size) |
535 | continue; | |
536 | ||
8342681d NF |
537 | add_active_range(nid, drmem.base_addr >> PAGE_SHIFT, |
538 | (drmem.base_addr >> PAGE_SHIFT) | |
539 | + (size >> PAGE_SHIFT)); | |
0204568a PM |
540 | } |
541 | } | |
542 | ||
1da177e4 LT |
543 | static int __init parse_numa_properties(void) |
544 | { | |
545 | struct device_node *cpu = NULL; | |
546 | struct device_node *memory = NULL; | |
482ec7c4 | 547 | int default_nid = 0; |
1da177e4 LT |
548 | unsigned long i; |
549 | ||
550 | if (numa_enabled == 0) { | |
551 | printk(KERN_WARNING "NUMA disabled by user\n"); | |
552 | return -1; | |
553 | } | |
554 | ||
1da177e4 LT |
555 | min_common_depth = find_min_common_depth(); |
556 | ||
1da177e4 LT |
557 | if (min_common_depth < 0) |
558 | return min_common_depth; | |
559 | ||
bf4b85b0 NL |
560 | dbg("NUMA associativity depth for CPU/Memory: %d\n", min_common_depth); |
561 | ||
1da177e4 | 562 | /* |
482ec7c4 NL |
563 | * Even though we connect cpus to numa domains later in SMP |
564 | * init, we need to know the node ids now. This is because | |
565 | * each node to be onlined must have NODE_DATA etc backing it. | |
1da177e4 | 566 | */ |
482ec7c4 | 567 | for_each_present_cpu(i) { |
cf950b7a | 568 | int nid; |
1da177e4 LT |
569 | |
570 | cpu = find_cpu_node(i); | |
482ec7c4 | 571 | BUG_ON(!cpu); |
953039c8 | 572 | nid = of_node_to_nid_single(cpu); |
482ec7c4 | 573 | of_node_put(cpu); |
1da177e4 | 574 | |
482ec7c4 NL |
575 | /* |
576 | * Don't fall back to default_nid yet -- we will plug | |
577 | * cpus into nodes once the memory scan has discovered | |
578 | * the topology. | |
579 | */ | |
580 | if (nid < 0) | |
581 | continue; | |
582 | node_set_online(nid); | |
1da177e4 LT |
583 | } |
584 | ||
237a0989 | 585 | get_n_mem_cells(&n_mem_addr_cells, &n_mem_size_cells); |
1da177e4 LT |
586 | memory = NULL; |
587 | while ((memory = of_find_node_by_type(memory, "memory")) != NULL) { | |
588 | unsigned long start; | |
589 | unsigned long size; | |
cf950b7a | 590 | int nid; |
1da177e4 | 591 | int ranges; |
a7f67bdf | 592 | const unsigned int *memcell_buf; |
1da177e4 LT |
593 | unsigned int len; |
594 | ||
e2eb6392 | 595 | memcell_buf = of_get_property(memory, |
ba759485 ME |
596 | "linux,usable-memory", &len); |
597 | if (!memcell_buf || len <= 0) | |
e2eb6392 | 598 | memcell_buf = of_get_property(memory, "reg", &len); |
1da177e4 LT |
599 | if (!memcell_buf || len <= 0) |
600 | continue; | |
601 | ||
cc5d0189 BH |
602 | /* ranges in cell */ |
603 | ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells); | |
1da177e4 LT |
604 | new_range: |
605 | /* these are order-sensitive, and modify the buffer pointer */ | |
237a0989 MK |
606 | start = read_n_cells(n_mem_addr_cells, &memcell_buf); |
607 | size = read_n_cells(n_mem_size_cells, &memcell_buf); | |
1da177e4 | 608 | |
482ec7c4 NL |
609 | /* |
610 | * Assumption: either all memory nodes or none will | |
611 | * have associativity properties. If none, then | |
612 | * everything goes to default_nid. | |
613 | */ | |
953039c8 | 614 | nid = of_node_to_nid_single(memory); |
482ec7c4 NL |
615 | if (nid < 0) |
616 | nid = default_nid; | |
1daa6d08 BS |
617 | |
618 | fake_numa_create_new_node(((start + size) >> PAGE_SHIFT), &nid); | |
482ec7c4 | 619 | node_set_online(nid); |
1da177e4 | 620 | |
45fb6cea | 621 | if (!(size = numa_enforce_memory_limit(start, size))) { |
1da177e4 LT |
622 | if (--ranges) |
623 | goto new_range; | |
624 | else | |
625 | continue; | |
626 | } | |
627 | ||
c67c3cb4 MG |
628 | add_active_range(nid, start >> PAGE_SHIFT, |
629 | (start >> PAGE_SHIFT) + (size >> PAGE_SHIFT)); | |
1da177e4 LT |
630 | |
631 | if (--ranges) | |
632 | goto new_range; | |
633 | } | |
634 | ||
0204568a PM |
635 | /* |
636 | * Now do the same thing for each LMB listed in the ibm,dynamic-memory | |
637 | * property in the ibm,dynamic-reconfiguration-memory node. | |
638 | */ | |
639 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); | |
640 | if (memory) | |
641 | parse_drconf_memory(memory); | |
642 | ||
1da177e4 LT |
643 | return 0; |
644 | } | |
645 | ||
646 | static void __init setup_nonnuma(void) | |
647 | { | |
648 | unsigned long top_of_ram = lmb_end_of_DRAM(); | |
649 | unsigned long total_ram = lmb_phys_mem_size(); | |
c67c3cb4 | 650 | unsigned long start_pfn, end_pfn; |
1daa6d08 | 651 | unsigned int i, nid = 0; |
1da177e4 | 652 | |
e110b281 | 653 | printk(KERN_DEBUG "Top of RAM: 0x%lx, Total RAM: 0x%lx\n", |
1da177e4 | 654 | top_of_ram, total_ram); |
e110b281 | 655 | printk(KERN_DEBUG "Memory hole size: %ldMB\n", |
1da177e4 LT |
656 | (top_of_ram - total_ram) >> 20); |
657 | ||
c67c3cb4 MG |
658 | for (i = 0; i < lmb.memory.cnt; ++i) { |
659 | start_pfn = lmb.memory.region[i].base >> PAGE_SHIFT; | |
660 | end_pfn = start_pfn + lmb_size_pages(&lmb.memory, i); | |
1daa6d08 BS |
661 | |
662 | fake_numa_create_new_node(end_pfn, &nid); | |
663 | add_active_range(nid, start_pfn, end_pfn); | |
664 | node_set_online(nid); | |
c67c3cb4 | 665 | } |
1da177e4 LT |
666 | } |
667 | ||
4b703a23 AB |
668 | void __init dump_numa_cpu_topology(void) |
669 | { | |
670 | unsigned int node; | |
671 | unsigned int cpu, count; | |
672 | ||
673 | if (min_common_depth == -1 || !numa_enabled) | |
674 | return; | |
675 | ||
676 | for_each_online_node(node) { | |
e110b281 | 677 | printk(KERN_DEBUG "Node %d CPUs:", node); |
4b703a23 AB |
678 | |
679 | count = 0; | |
680 | /* | |
681 | * If we used a CPU iterator here we would miss printing | |
682 | * the holes in the cpumap. | |
683 | */ | |
684 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
685 | if (cpu_isset(cpu, numa_cpumask_lookup_table[node])) { | |
686 | if (count == 0) | |
687 | printk(" %u", cpu); | |
688 | ++count; | |
689 | } else { | |
690 | if (count > 1) | |
691 | printk("-%u", cpu - 1); | |
692 | count = 0; | |
693 | } | |
694 | } | |
695 | ||
696 | if (count > 1) | |
697 | printk("-%u", NR_CPUS - 1); | |
698 | printk("\n"); | |
699 | } | |
700 | } | |
701 | ||
702 | static void __init dump_numa_memory_topology(void) | |
1da177e4 LT |
703 | { |
704 | unsigned int node; | |
705 | unsigned int count; | |
706 | ||
707 | if (min_common_depth == -1 || !numa_enabled) | |
708 | return; | |
709 | ||
710 | for_each_online_node(node) { | |
711 | unsigned long i; | |
712 | ||
e110b281 | 713 | printk(KERN_DEBUG "Node %d Memory:", node); |
1da177e4 LT |
714 | |
715 | count = 0; | |
716 | ||
45fb6cea AB |
717 | for (i = 0; i < lmb_end_of_DRAM(); |
718 | i += (1 << SECTION_SIZE_BITS)) { | |
719 | if (early_pfn_to_nid(i >> PAGE_SHIFT) == node) { | |
1da177e4 LT |
720 | if (count == 0) |
721 | printk(" 0x%lx", i); | |
722 | ++count; | |
723 | } else { | |
724 | if (count > 0) | |
725 | printk("-0x%lx", i); | |
726 | count = 0; | |
727 | } | |
728 | } | |
729 | ||
730 | if (count > 0) | |
731 | printk("-0x%lx", i); | |
732 | printk("\n"); | |
733 | } | |
1da177e4 LT |
734 | } |
735 | ||
736 | /* | |
737 | * Allocate some memory, satisfying the lmb or bootmem allocator where | |
738 | * required. nid is the preferred node and end is the physical address of | |
739 | * the highest address in the node. | |
740 | * | |
741 | * Returns the physical address of the memory. | |
742 | */ | |
45fb6cea AB |
743 | static void __init *careful_allocation(int nid, unsigned long size, |
744 | unsigned long align, | |
745 | unsigned long end_pfn) | |
1da177e4 | 746 | { |
45fb6cea | 747 | int new_nid; |
d7a5b2ff | 748 | unsigned long ret = __lmb_alloc_base(size, align, end_pfn << PAGE_SHIFT); |
1da177e4 LT |
749 | |
750 | /* retry over all memory */ | |
751 | if (!ret) | |
d7a5b2ff | 752 | ret = __lmb_alloc_base(size, align, lmb_end_of_DRAM()); |
1da177e4 LT |
753 | |
754 | if (!ret) | |
755 | panic("numa.c: cannot allocate %lu bytes on node %d", | |
756 | size, nid); | |
757 | ||
758 | /* | |
759 | * If the memory came from a previously allocated node, we must | |
760 | * retry with the bootmem allocator. | |
761 | */ | |
45fb6cea AB |
762 | new_nid = early_pfn_to_nid(ret >> PAGE_SHIFT); |
763 | if (new_nid < nid) { | |
764 | ret = (unsigned long)__alloc_bootmem_node(NODE_DATA(new_nid), | |
1da177e4 LT |
765 | size, align, 0); |
766 | ||
767 | if (!ret) | |
768 | panic("numa.c: cannot allocate %lu bytes on node %d", | |
45fb6cea | 769 | size, new_nid); |
1da177e4 | 770 | |
45fb6cea | 771 | ret = __pa(ret); |
1da177e4 LT |
772 | |
773 | dbg("alloc_bootmem %lx %lx\n", ret, size); | |
774 | } | |
775 | ||
45fb6cea | 776 | return (void *)ret; |
1da177e4 LT |
777 | } |
778 | ||
74b85f37 CS |
779 | static struct notifier_block __cpuinitdata ppc64_numa_nb = { |
780 | .notifier_call = cpu_numa_callback, | |
781 | .priority = 1 /* Must run before sched domains notifier. */ | |
782 | }; | |
783 | ||
1da177e4 LT |
784 | void __init do_init_bootmem(void) |
785 | { | |
786 | int nid; | |
45fb6cea | 787 | unsigned int i; |
1da177e4 LT |
788 | |
789 | min_low_pfn = 0; | |
790 | max_low_pfn = lmb_end_of_DRAM() >> PAGE_SHIFT; | |
791 | max_pfn = max_low_pfn; | |
792 | ||
793 | if (parse_numa_properties()) | |
794 | setup_nonnuma(); | |
795 | else | |
4b703a23 | 796 | dump_numa_memory_topology(); |
1da177e4 LT |
797 | |
798 | register_cpu_notifier(&ppc64_numa_nb); | |
2b261227 NL |
799 | cpu_numa_callback(&ppc64_numa_nb, CPU_UP_PREPARE, |
800 | (void *)(unsigned long)boot_cpuid); | |
1da177e4 LT |
801 | |
802 | for_each_online_node(nid) { | |
c67c3cb4 | 803 | unsigned long start_pfn, end_pfn; |
1da177e4 LT |
804 | unsigned long bootmem_paddr; |
805 | unsigned long bootmap_pages; | |
806 | ||
c67c3cb4 | 807 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); |
1da177e4 LT |
808 | |
809 | /* Allocate the node structure node local if possible */ | |
45fb6cea | 810 | NODE_DATA(nid) = careful_allocation(nid, |
1da177e4 | 811 | sizeof(struct pglist_data), |
45fb6cea AB |
812 | SMP_CACHE_BYTES, end_pfn); |
813 | NODE_DATA(nid) = __va(NODE_DATA(nid)); | |
1da177e4 LT |
814 | memset(NODE_DATA(nid), 0, sizeof(struct pglist_data)); |
815 | ||
816 | dbg("node %d\n", nid); | |
817 | dbg("NODE_DATA() = %p\n", NODE_DATA(nid)); | |
818 | ||
819 | NODE_DATA(nid)->bdata = &plat_node_bdata[nid]; | |
45fb6cea AB |
820 | NODE_DATA(nid)->node_start_pfn = start_pfn; |
821 | NODE_DATA(nid)->node_spanned_pages = end_pfn - start_pfn; | |
1da177e4 LT |
822 | |
823 | if (NODE_DATA(nid)->node_spanned_pages == 0) | |
824 | continue; | |
825 | ||
45fb6cea AB |
826 | dbg("start_paddr = %lx\n", start_pfn << PAGE_SHIFT); |
827 | dbg("end_paddr = %lx\n", end_pfn << PAGE_SHIFT); | |
1da177e4 | 828 | |
45fb6cea AB |
829 | bootmap_pages = bootmem_bootmap_pages(end_pfn - start_pfn); |
830 | bootmem_paddr = (unsigned long)careful_allocation(nid, | |
831 | bootmap_pages << PAGE_SHIFT, | |
832 | PAGE_SIZE, end_pfn); | |
833 | memset(__va(bootmem_paddr), 0, bootmap_pages << PAGE_SHIFT); | |
1da177e4 | 834 | |
1da177e4 LT |
835 | dbg("bootmap_paddr = %lx\n", bootmem_paddr); |
836 | ||
837 | init_bootmem_node(NODE_DATA(nid), bootmem_paddr >> PAGE_SHIFT, | |
45fb6cea | 838 | start_pfn, end_pfn); |
1da177e4 | 839 | |
c67c3cb4 | 840 | free_bootmem_with_active_regions(nid, end_pfn); |
1da177e4 | 841 | |
45fb6cea | 842 | /* Mark reserved regions on this node */ |
1da177e4 | 843 | for (i = 0; i < lmb.reserved.cnt; i++) { |
180379dc | 844 | unsigned long physbase = lmb.reserved.region[i].base; |
1da177e4 | 845 | unsigned long size = lmb.reserved.region[i].size; |
45fb6cea AB |
846 | unsigned long start_paddr = start_pfn << PAGE_SHIFT; |
847 | unsigned long end_paddr = end_pfn << PAGE_SHIFT; | |
1da177e4 | 848 | |
45fb6cea AB |
849 | if (early_pfn_to_nid(physbase >> PAGE_SHIFT) != nid && |
850 | early_pfn_to_nid((physbase+size-1) >> PAGE_SHIFT) != nid) | |
1da177e4 LT |
851 | continue; |
852 | ||
853 | if (physbase < end_paddr && | |
854 | (physbase+size) > start_paddr) { | |
855 | /* overlaps */ | |
856 | if (physbase < start_paddr) { | |
857 | size -= start_paddr - physbase; | |
858 | physbase = start_paddr; | |
859 | } | |
860 | ||
861 | if (size > end_paddr - physbase) | |
862 | size = end_paddr - physbase; | |
863 | ||
864 | dbg("reserve_bootmem %lx %lx\n", physbase, | |
865 | size); | |
866 | reserve_bootmem_node(NODE_DATA(nid), physbase, | |
72a7fe39 | 867 | size, BOOTMEM_DEFAULT); |
1da177e4 LT |
868 | } |
869 | } | |
802f192e | 870 | |
c67c3cb4 | 871 | sparse_memory_present_with_active_regions(nid); |
1da177e4 LT |
872 | } |
873 | } | |
874 | ||
875 | void __init paging_init(void) | |
876 | { | |
6391af17 MG |
877 | unsigned long max_zone_pfns[MAX_NR_ZONES]; |
878 | memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); | |
879 | max_zone_pfns[ZONE_DMA] = lmb_end_of_DRAM() >> PAGE_SHIFT; | |
c67c3cb4 | 880 | free_area_init_nodes(max_zone_pfns); |
1da177e4 LT |
881 | } |
882 | ||
883 | static int __init early_numa(char *p) | |
884 | { | |
885 | if (!p) | |
886 | return 0; | |
887 | ||
888 | if (strstr(p, "off")) | |
889 | numa_enabled = 0; | |
890 | ||
891 | if (strstr(p, "debug")) | |
892 | numa_debug = 1; | |
893 | ||
1daa6d08 BS |
894 | p = strstr(p, "fake="); |
895 | if (p) | |
896 | cmdline = p + strlen("fake="); | |
897 | ||
1da177e4 LT |
898 | return 0; |
899 | } | |
900 | early_param("numa", early_numa); | |
237a0989 MK |
901 | |
902 | #ifdef CONFIG_MEMORY_HOTPLUG | |
903 | /* | |
904 | * Find the node associated with a hot added memory section. Section | |
905 | * corresponds to a SPARSEMEM section, not an LMB. It is assumed that | |
906 | * sections are fully contained within a single LMB. | |
907 | */ | |
908 | int hot_add_scn_to_nid(unsigned long scn_addr) | |
909 | { | |
910 | struct device_node *memory = NULL; | |
b226e462 | 911 | nodemask_t nodes; |
482ec7c4 | 912 | int default_nid = any_online_node(NODE_MASK_ALL); |
069007ae | 913 | int nid; |
237a0989 MK |
914 | |
915 | if (!numa_enabled || (min_common_depth < 0)) | |
482ec7c4 | 916 | return default_nid; |
237a0989 MK |
917 | |
918 | while ((memory = of_find_node_by_type(memory, "memory")) != NULL) { | |
919 | unsigned long start, size; | |
b226e462 | 920 | int ranges; |
a7f67bdf | 921 | const unsigned int *memcell_buf; |
237a0989 MK |
922 | unsigned int len; |
923 | ||
e2eb6392 | 924 | memcell_buf = of_get_property(memory, "reg", &len); |
237a0989 MK |
925 | if (!memcell_buf || len <= 0) |
926 | continue; | |
927 | ||
cc5d0189 BH |
928 | /* ranges in cell */ |
929 | ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells); | |
237a0989 MK |
930 | ha_new_range: |
931 | start = read_n_cells(n_mem_addr_cells, &memcell_buf); | |
932 | size = read_n_cells(n_mem_size_cells, &memcell_buf); | |
953039c8 | 933 | nid = of_node_to_nid_single(memory); |
237a0989 MK |
934 | |
935 | /* Domains not present at boot default to 0 */ | |
482ec7c4 NL |
936 | if (nid < 0 || !node_online(nid)) |
937 | nid = default_nid; | |
237a0989 MK |
938 | |
939 | if ((scn_addr >= start) && (scn_addr < (start + size))) { | |
940 | of_node_put(memory); | |
cf950b7a | 941 | goto got_nid; |
237a0989 MK |
942 | } |
943 | ||
944 | if (--ranges) /* process all ranges in cell */ | |
945 | goto ha_new_range; | |
946 | } | |
237a0989 | 947 | BUG(); /* section address should be found above */ |
069007ae | 948 | return 0; |
b226e462 MK |
949 | |
950 | /* Temporary code to ensure that returned node is not empty */ | |
cf950b7a | 951 | got_nid: |
b226e462 | 952 | nodes_setall(nodes); |
cf950b7a NL |
953 | while (NODE_DATA(nid)->node_spanned_pages == 0) { |
954 | node_clear(nid, nodes); | |
955 | nid = any_online_node(nodes); | |
b226e462 | 956 | } |
cf950b7a | 957 | return nid; |
237a0989 MK |
958 | } |
959 | #endif /* CONFIG_MEMORY_HOTPLUG */ |