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 | ||
1da177e4 | 42 | static int min_common_depth; |
237a0989 | 43 | static int n_mem_addr_cells, n_mem_size_cells; |
1da177e4 | 44 | |
1daa6d08 BS |
45 | static int __cpuinit fake_numa_create_new_node(unsigned long end_pfn, |
46 | unsigned int *nid) | |
47 | { | |
48 | unsigned long long mem; | |
49 | char *p = cmdline; | |
50 | static unsigned int fake_nid; | |
51 | static unsigned long long curr_boundary; | |
52 | ||
53 | /* | |
54 | * Modify node id, iff we started creating NUMA nodes | |
55 | * We want to continue from where we left of the last time | |
56 | */ | |
57 | if (fake_nid) | |
58 | *nid = fake_nid; | |
59 | /* | |
60 | * In case there are no more arguments to parse, the | |
61 | * node_id should be the same as the last fake node id | |
62 | * (we've handled this above). | |
63 | */ | |
64 | if (!p) | |
65 | return 0; | |
66 | ||
67 | mem = memparse(p, &p); | |
68 | if (!mem) | |
69 | return 0; | |
70 | ||
71 | if (mem < curr_boundary) | |
72 | return 0; | |
73 | ||
74 | curr_boundary = mem; | |
75 | ||
76 | if ((end_pfn << PAGE_SHIFT) > mem) { | |
77 | /* | |
78 | * Skip commas and spaces | |
79 | */ | |
80 | while (*p == ',' || *p == ' ' || *p == '\t') | |
81 | p++; | |
82 | ||
83 | cmdline = p; | |
84 | fake_nid++; | |
85 | *nid = fake_nid; | |
86 | dbg("created new fake_node with id %d\n", fake_nid); | |
87 | return 1; | |
88 | } | |
89 | return 0; | |
90 | } | |
91 | ||
2e5ce39d | 92 | static void __cpuinit map_cpu_to_node(int cpu, int node) |
1da177e4 LT |
93 | { |
94 | numa_cpu_lookup_table[cpu] = node; | |
45fb6cea | 95 | |
bf4b85b0 NL |
96 | dbg("adding cpu %d to node %d\n", cpu, node); |
97 | ||
45fb6cea | 98 | if (!(cpu_isset(cpu, numa_cpumask_lookup_table[node]))) |
1da177e4 | 99 | cpu_set(cpu, numa_cpumask_lookup_table[node]); |
1da177e4 LT |
100 | } |
101 | ||
102 | #ifdef CONFIG_HOTPLUG_CPU | |
103 | static void unmap_cpu_from_node(unsigned long cpu) | |
104 | { | |
105 | int node = numa_cpu_lookup_table[cpu]; | |
106 | ||
107 | dbg("removing cpu %lu from node %d\n", cpu, node); | |
108 | ||
109 | if (cpu_isset(cpu, numa_cpumask_lookup_table[node])) { | |
110 | cpu_clear(cpu, numa_cpumask_lookup_table[node]); | |
1da177e4 LT |
111 | } else { |
112 | printk(KERN_ERR "WARNING: cpu %lu not found in node %d\n", | |
113 | cpu, node); | |
114 | } | |
115 | } | |
116 | #endif /* CONFIG_HOTPLUG_CPU */ | |
117 | ||
2e5ce39d | 118 | static struct device_node * __cpuinit find_cpu_node(unsigned int cpu) |
1da177e4 LT |
119 | { |
120 | unsigned int hw_cpuid = get_hard_smp_processor_id(cpu); | |
121 | struct device_node *cpu_node = NULL; | |
a7f67bdf | 122 | const unsigned int *interrupt_server, *reg; |
1da177e4 LT |
123 | int len; |
124 | ||
125 | while ((cpu_node = of_find_node_by_type(cpu_node, "cpu")) != NULL) { | |
126 | /* Try interrupt server first */ | |
e2eb6392 | 127 | interrupt_server = of_get_property(cpu_node, |
1da177e4 LT |
128 | "ibm,ppc-interrupt-server#s", &len); |
129 | ||
130 | len = len / sizeof(u32); | |
131 | ||
132 | if (interrupt_server && (len > 0)) { | |
133 | while (len--) { | |
134 | if (interrupt_server[len] == hw_cpuid) | |
135 | return cpu_node; | |
136 | } | |
137 | } else { | |
e2eb6392 | 138 | reg = of_get_property(cpu_node, "reg", &len); |
1da177e4 LT |
139 | if (reg && (len > 0) && (reg[0] == hw_cpuid)) |
140 | return cpu_node; | |
141 | } | |
142 | } | |
143 | ||
144 | return NULL; | |
145 | } | |
146 | ||
147 | /* must hold reference to node during call */ | |
a7f67bdf | 148 | static const int *of_get_associativity(struct device_node *dev) |
1da177e4 | 149 | { |
e2eb6392 | 150 | return of_get_property(dev, "ibm,associativity", NULL); |
1da177e4 LT |
151 | } |
152 | ||
482ec7c4 NL |
153 | /* Returns nid in the range [0..MAX_NUMNODES-1], or -1 if no useful numa |
154 | * info is found. | |
155 | */ | |
953039c8 | 156 | static int of_node_to_nid_single(struct device_node *device) |
1da177e4 | 157 | { |
482ec7c4 | 158 | int nid = -1; |
a7f67bdf | 159 | const unsigned int *tmp; |
1da177e4 LT |
160 | |
161 | if (min_common_depth == -1) | |
482ec7c4 | 162 | goto out; |
1da177e4 LT |
163 | |
164 | tmp = of_get_associativity(device); | |
482ec7c4 NL |
165 | if (!tmp) |
166 | goto out; | |
167 | ||
168 | if (tmp[0] >= min_common_depth) | |
cf950b7a | 169 | nid = tmp[min_common_depth]; |
bc16a759 NL |
170 | |
171 | /* POWER4 LPAR uses 0xffff as invalid node */ | |
482ec7c4 NL |
172 | if (nid == 0xffff || nid >= MAX_NUMNODES) |
173 | nid = -1; | |
174 | out: | |
cf950b7a | 175 | return nid; |
1da177e4 LT |
176 | } |
177 | ||
953039c8 JK |
178 | /* Walk the device tree upwards, looking for an associativity id */ |
179 | int of_node_to_nid(struct device_node *device) | |
180 | { | |
181 | struct device_node *tmp; | |
182 | int nid = -1; | |
183 | ||
184 | of_node_get(device); | |
185 | while (device) { | |
186 | nid = of_node_to_nid_single(device); | |
187 | if (nid != -1) | |
188 | break; | |
189 | ||
190 | tmp = device; | |
191 | device = of_get_parent(tmp); | |
192 | of_node_put(tmp); | |
193 | } | |
194 | of_node_put(device); | |
195 | ||
196 | return nid; | |
197 | } | |
198 | EXPORT_SYMBOL_GPL(of_node_to_nid); | |
199 | ||
1da177e4 LT |
200 | /* |
201 | * In theory, the "ibm,associativity" property may contain multiple | |
202 | * associativity lists because a resource may be multiply connected | |
203 | * into the machine. This resource then has different associativity | |
204 | * characteristics relative to its multiple connections. We ignore | |
205 | * this for now. We also assume that all cpu and memory sets have | |
206 | * their distances represented at a common level. This won't be | |
1b3c3714 | 207 | * true for hierarchical NUMA. |
1da177e4 LT |
208 | * |
209 | * In any case the ibm,associativity-reference-points should give | |
210 | * the correct depth for a normal NUMA system. | |
211 | * | |
212 | * - Dave Hansen <haveblue@us.ibm.com> | |
213 | */ | |
214 | static int __init find_min_common_depth(void) | |
215 | { | |
216 | int depth; | |
a7f67bdf | 217 | const unsigned int *ref_points; |
1da177e4 LT |
218 | struct device_node *rtas_root; |
219 | unsigned int len; | |
220 | ||
221 | rtas_root = of_find_node_by_path("/rtas"); | |
222 | ||
223 | if (!rtas_root) | |
224 | return -1; | |
225 | ||
226 | /* | |
227 | * this property is 2 32-bit integers, each representing a level of | |
228 | * depth in the associativity nodes. The first is for an SMP | |
229 | * configuration (should be all 0's) and the second is for a normal | |
230 | * NUMA configuration. | |
231 | */ | |
e2eb6392 | 232 | ref_points = of_get_property(rtas_root, |
1da177e4 LT |
233 | "ibm,associativity-reference-points", &len); |
234 | ||
235 | if ((len >= 1) && ref_points) { | |
236 | depth = ref_points[1]; | |
237 | } else { | |
bf4b85b0 | 238 | dbg("NUMA: ibm,associativity-reference-points not found.\n"); |
1da177e4 LT |
239 | depth = -1; |
240 | } | |
241 | of_node_put(rtas_root); | |
242 | ||
243 | return depth; | |
244 | } | |
245 | ||
84c9fdd1 | 246 | static void __init get_n_mem_cells(int *n_addr_cells, int *n_size_cells) |
1da177e4 LT |
247 | { |
248 | struct device_node *memory = NULL; | |
1da177e4 LT |
249 | |
250 | memory = of_find_node_by_type(memory, "memory"); | |
54c23310 | 251 | if (!memory) |
84c9fdd1 | 252 | panic("numa.c: No memory nodes found!"); |
54c23310 | 253 | |
a8bda5dd | 254 | *n_addr_cells = of_n_addr_cells(memory); |
9213feea | 255 | *n_size_cells = of_n_size_cells(memory); |
84c9fdd1 | 256 | of_node_put(memory); |
1da177e4 LT |
257 | } |
258 | ||
a7f67bdf | 259 | static unsigned long __devinit read_n_cells(int n, const unsigned int **buf) |
1da177e4 LT |
260 | { |
261 | unsigned long result = 0; | |
262 | ||
263 | while (n--) { | |
264 | result = (result << 32) | **buf; | |
265 | (*buf)++; | |
266 | } | |
267 | return result; | |
268 | } | |
269 | ||
8342681d NF |
270 | struct of_drconf_cell { |
271 | u64 base_addr; | |
272 | u32 drc_index; | |
273 | u32 reserved; | |
274 | u32 aa_index; | |
275 | u32 flags; | |
276 | }; | |
277 | ||
278 | #define DRCONF_MEM_ASSIGNED 0x00000008 | |
279 | #define DRCONF_MEM_AI_INVALID 0x00000040 | |
280 | #define DRCONF_MEM_RESERVED 0x00000080 | |
281 | ||
282 | /* | |
283 | * Read the next lmb list entry from the ibm,dynamic-memory property | |
284 | * and return the information in the provided of_drconf_cell structure. | |
285 | */ | |
286 | static void read_drconf_cell(struct of_drconf_cell *drmem, const u32 **cellp) | |
287 | { | |
288 | const u32 *cp; | |
289 | ||
290 | drmem->base_addr = read_n_cells(n_mem_addr_cells, cellp); | |
291 | ||
292 | cp = *cellp; | |
293 | drmem->drc_index = cp[0]; | |
294 | drmem->reserved = cp[1]; | |
295 | drmem->aa_index = cp[2]; | |
296 | drmem->flags = cp[3]; | |
297 | ||
298 | *cellp = cp + 4; | |
299 | } | |
300 | ||
301 | /* | |
302 | * Retreive and validate the ibm,dynamic-memory property of the device tree. | |
303 | * | |
304 | * The layout of the ibm,dynamic-memory property is a number N of lmb | |
305 | * list entries followed by N lmb list entries. Each lmb list entry | |
306 | * contains information as layed out in the of_drconf_cell struct above. | |
307 | */ | |
308 | static int of_get_drconf_memory(struct device_node *memory, const u32 **dm) | |
309 | { | |
310 | const u32 *prop; | |
311 | u32 len, entries; | |
312 | ||
313 | prop = of_get_property(memory, "ibm,dynamic-memory", &len); | |
314 | if (!prop || len < sizeof(unsigned int)) | |
315 | return 0; | |
316 | ||
317 | entries = *prop++; | |
318 | ||
319 | /* Now that we know the number of entries, revalidate the size | |
320 | * of the property read in to ensure we have everything | |
321 | */ | |
322 | if (len < (entries * (n_mem_addr_cells + 4) + 1) * sizeof(unsigned int)) | |
323 | return 0; | |
324 | ||
325 | *dm = prop; | |
326 | return entries; | |
327 | } | |
328 | ||
329 | /* | |
330 | * Retreive and validate the ibm,lmb-size property for drconf memory | |
331 | * from the device tree. | |
332 | */ | |
333 | static u64 of_get_lmb_size(struct device_node *memory) | |
334 | { | |
335 | const u32 *prop; | |
336 | u32 len; | |
337 | ||
338 | prop = of_get_property(memory, "ibm,lmb-size", &len); | |
339 | if (!prop || len < sizeof(unsigned int)) | |
340 | return 0; | |
341 | ||
342 | return read_n_cells(n_mem_size_cells, &prop); | |
343 | } | |
344 | ||
345 | struct assoc_arrays { | |
346 | u32 n_arrays; | |
347 | u32 array_sz; | |
348 | const u32 *arrays; | |
349 | }; | |
350 | ||
351 | /* | |
352 | * Retreive and validate the list of associativity arrays for drconf | |
353 | * memory from the ibm,associativity-lookup-arrays property of the | |
354 | * device tree.. | |
355 | * | |
356 | * The layout of the ibm,associativity-lookup-arrays property is a number N | |
357 | * indicating the number of associativity arrays, followed by a number M | |
358 | * indicating the size of each associativity array, followed by a list | |
359 | * of N associativity arrays. | |
360 | */ | |
361 | static int of_get_assoc_arrays(struct device_node *memory, | |
362 | struct assoc_arrays *aa) | |
363 | { | |
364 | const u32 *prop; | |
365 | u32 len; | |
366 | ||
367 | prop = of_get_property(memory, "ibm,associativity-lookup-arrays", &len); | |
368 | if (!prop || len < 2 * sizeof(unsigned int)) | |
369 | return -1; | |
370 | ||
371 | aa->n_arrays = *prop++; | |
372 | aa->array_sz = *prop++; | |
373 | ||
374 | /* Now that we know the number of arrrays and size of each array, | |
375 | * revalidate the size of the property read in. | |
376 | */ | |
377 | if (len < (aa->n_arrays * aa->array_sz + 2) * sizeof(unsigned int)) | |
378 | return -1; | |
379 | ||
380 | aa->arrays = prop; | |
381 | return 0; | |
382 | } | |
383 | ||
384 | /* | |
385 | * This is like of_node_to_nid_single() for memory represented in the | |
386 | * ibm,dynamic-reconfiguration-memory node. | |
387 | */ | |
388 | static int of_drconf_to_nid_single(struct of_drconf_cell *drmem, | |
389 | struct assoc_arrays *aa) | |
390 | { | |
391 | int default_nid = 0; | |
392 | int nid = default_nid; | |
393 | int index; | |
394 | ||
395 | if (min_common_depth > 0 && min_common_depth <= aa->array_sz && | |
396 | !(drmem->flags & DRCONF_MEM_AI_INVALID) && | |
397 | drmem->aa_index < aa->n_arrays) { | |
398 | index = drmem->aa_index * aa->array_sz + min_common_depth - 1; | |
399 | nid = aa->arrays[index]; | |
400 | ||
401 | if (nid == 0xffff || nid >= MAX_NUMNODES) | |
402 | nid = default_nid; | |
403 | } | |
404 | ||
405 | return nid; | |
406 | } | |
407 | ||
1da177e4 LT |
408 | /* |
409 | * Figure out to which domain a cpu belongs and stick it there. | |
410 | * Return the id of the domain used. | |
411 | */ | |
2e5ce39d | 412 | static int __cpuinit numa_setup_cpu(unsigned long lcpu) |
1da177e4 | 413 | { |
cf950b7a | 414 | int nid = 0; |
1da177e4 LT |
415 | struct device_node *cpu = find_cpu_node(lcpu); |
416 | ||
417 | if (!cpu) { | |
418 | WARN_ON(1); | |
419 | goto out; | |
420 | } | |
421 | ||
953039c8 | 422 | nid = of_node_to_nid_single(cpu); |
1da177e4 | 423 | |
482ec7c4 NL |
424 | if (nid < 0 || !node_online(nid)) |
425 | nid = any_online_node(NODE_MASK_ALL); | |
1da177e4 | 426 | out: |
cf950b7a | 427 | map_cpu_to_node(lcpu, nid); |
1da177e4 LT |
428 | |
429 | of_node_put(cpu); | |
430 | ||
cf950b7a | 431 | return nid; |
1da177e4 LT |
432 | } |
433 | ||
74b85f37 | 434 | static int __cpuinit cpu_numa_callback(struct notifier_block *nfb, |
1da177e4 LT |
435 | unsigned long action, |
436 | void *hcpu) | |
437 | { | |
438 | unsigned long lcpu = (unsigned long)hcpu; | |
439 | int ret = NOTIFY_DONE; | |
440 | ||
441 | switch (action) { | |
442 | case CPU_UP_PREPARE: | |
8bb78442 | 443 | case CPU_UP_PREPARE_FROZEN: |
2b261227 | 444 | numa_setup_cpu(lcpu); |
1da177e4 LT |
445 | ret = NOTIFY_OK; |
446 | break; | |
447 | #ifdef CONFIG_HOTPLUG_CPU | |
448 | case CPU_DEAD: | |
8bb78442 | 449 | case CPU_DEAD_FROZEN: |
1da177e4 | 450 | case CPU_UP_CANCELED: |
8bb78442 | 451 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 LT |
452 | unmap_cpu_from_node(lcpu); |
453 | break; | |
454 | ret = NOTIFY_OK; | |
455 | #endif | |
456 | } | |
457 | return ret; | |
458 | } | |
459 | ||
460 | /* | |
461 | * Check and possibly modify a memory region to enforce the memory limit. | |
462 | * | |
463 | * Returns the size the region should have to enforce the memory limit. | |
464 | * This will either be the original value of size, a truncated value, | |
465 | * or zero. If the returned value of size is 0 the region should be | |
466 | * discarded as it lies wholy above the memory limit. | |
467 | */ | |
45fb6cea AB |
468 | static unsigned long __init numa_enforce_memory_limit(unsigned long start, |
469 | unsigned long size) | |
1da177e4 LT |
470 | { |
471 | /* | |
472 | * We use lmb_end_of_DRAM() in here instead of memory_limit because | |
473 | * we've already adjusted it for the limit and it takes care of | |
474 | * having memory holes below the limit. | |
475 | */ | |
1da177e4 LT |
476 | |
477 | if (! memory_limit) | |
478 | return size; | |
479 | ||
480 | if (start + size <= lmb_end_of_DRAM()) | |
481 | return size; | |
482 | ||
483 | if (start >= lmb_end_of_DRAM()) | |
484 | return 0; | |
485 | ||
486 | return lmb_end_of_DRAM() - start; | |
487 | } | |
488 | ||
0204568a PM |
489 | /* |
490 | * Extract NUMA information from the ibm,dynamic-reconfiguration-memory | |
491 | * node. This assumes n_mem_{addr,size}_cells have been set. | |
492 | */ | |
493 | static void __init parse_drconf_memory(struct device_node *memory) | |
494 | { | |
8342681d NF |
495 | const u32 *dm; |
496 | unsigned int n, rc; | |
497 | unsigned long lmb_size, size; | |
498 | int nid; | |
499 | struct assoc_arrays aa; | |
500 | ||
501 | n = of_get_drconf_memory(memory, &dm); | |
502 | if (!n) | |
0204568a PM |
503 | return; |
504 | ||
8342681d NF |
505 | lmb_size = of_get_lmb_size(memory); |
506 | if (!lmb_size) | |
507 | return; | |
508 | ||
509 | rc = of_get_assoc_arrays(memory, &aa); | |
510 | if (rc) | |
0204568a PM |
511 | return; |
512 | ||
513 | for (; n != 0; --n) { | |
8342681d NF |
514 | struct of_drconf_cell drmem; |
515 | ||
516 | read_drconf_cell(&drmem, &dm); | |
517 | ||
518 | /* skip this block if the reserved bit is set in flags (0x80) | |
519 | or if the block is not assigned to this partition (0x8) */ | |
520 | if ((drmem.flags & DRCONF_MEM_RESERVED) | |
521 | || !(drmem.flags & DRCONF_MEM_ASSIGNED)) | |
0204568a | 522 | continue; |
1daa6d08 | 523 | |
8342681d NF |
524 | nid = of_drconf_to_nid_single(&drmem, &aa); |
525 | ||
526 | fake_numa_create_new_node( | |
527 | ((drmem.base_addr + lmb_size) >> PAGE_SHIFT), | |
528 | &nid); | |
529 | ||
0204568a PM |
530 | node_set_online(nid); |
531 | ||
8342681d | 532 | size = numa_enforce_memory_limit(drmem.base_addr, lmb_size); |
0204568a PM |
533 | if (!size) |
534 | continue; | |
535 | ||
8342681d NF |
536 | add_active_range(nid, drmem.base_addr >> PAGE_SHIFT, |
537 | (drmem.base_addr >> PAGE_SHIFT) | |
538 | + (size >> PAGE_SHIFT)); | |
0204568a PM |
539 | } |
540 | } | |
541 | ||
1da177e4 LT |
542 | static int __init parse_numa_properties(void) |
543 | { | |
544 | struct device_node *cpu = NULL; | |
545 | struct device_node *memory = NULL; | |
482ec7c4 | 546 | int default_nid = 0; |
1da177e4 LT |
547 | unsigned long i; |
548 | ||
549 | if (numa_enabled == 0) { | |
550 | printk(KERN_WARNING "NUMA disabled by user\n"); | |
551 | return -1; | |
552 | } | |
553 | ||
1da177e4 LT |
554 | min_common_depth = find_min_common_depth(); |
555 | ||
1da177e4 LT |
556 | if (min_common_depth < 0) |
557 | return min_common_depth; | |
558 | ||
bf4b85b0 NL |
559 | dbg("NUMA associativity depth for CPU/Memory: %d\n", min_common_depth); |
560 | ||
1da177e4 | 561 | /* |
482ec7c4 NL |
562 | * Even though we connect cpus to numa domains later in SMP |
563 | * init, we need to know the node ids now. This is because | |
564 | * each node to be onlined must have NODE_DATA etc backing it. | |
1da177e4 | 565 | */ |
482ec7c4 | 566 | for_each_present_cpu(i) { |
cf950b7a | 567 | int nid; |
1da177e4 LT |
568 | |
569 | cpu = find_cpu_node(i); | |
482ec7c4 | 570 | BUG_ON(!cpu); |
953039c8 | 571 | nid = of_node_to_nid_single(cpu); |
482ec7c4 | 572 | of_node_put(cpu); |
1da177e4 | 573 | |
482ec7c4 NL |
574 | /* |
575 | * Don't fall back to default_nid yet -- we will plug | |
576 | * cpus into nodes once the memory scan has discovered | |
577 | * the topology. | |
578 | */ | |
579 | if (nid < 0) | |
580 | continue; | |
581 | node_set_online(nid); | |
1da177e4 LT |
582 | } |
583 | ||
237a0989 | 584 | get_n_mem_cells(&n_mem_addr_cells, &n_mem_size_cells); |
1da177e4 LT |
585 | memory = NULL; |
586 | while ((memory = of_find_node_by_type(memory, "memory")) != NULL) { | |
587 | unsigned long start; | |
588 | unsigned long size; | |
cf950b7a | 589 | int nid; |
1da177e4 | 590 | int ranges; |
a7f67bdf | 591 | const unsigned int *memcell_buf; |
1da177e4 LT |
592 | unsigned int len; |
593 | ||
e2eb6392 | 594 | memcell_buf = of_get_property(memory, |
ba759485 ME |
595 | "linux,usable-memory", &len); |
596 | if (!memcell_buf || len <= 0) | |
e2eb6392 | 597 | memcell_buf = of_get_property(memory, "reg", &len); |
1da177e4 LT |
598 | if (!memcell_buf || len <= 0) |
599 | continue; | |
600 | ||
cc5d0189 BH |
601 | /* ranges in cell */ |
602 | ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells); | |
1da177e4 LT |
603 | new_range: |
604 | /* these are order-sensitive, and modify the buffer pointer */ | |
237a0989 MK |
605 | start = read_n_cells(n_mem_addr_cells, &memcell_buf); |
606 | size = read_n_cells(n_mem_size_cells, &memcell_buf); | |
1da177e4 | 607 | |
482ec7c4 NL |
608 | /* |
609 | * Assumption: either all memory nodes or none will | |
610 | * have associativity properties. If none, then | |
611 | * everything goes to default_nid. | |
612 | */ | |
953039c8 | 613 | nid = of_node_to_nid_single(memory); |
482ec7c4 NL |
614 | if (nid < 0) |
615 | nid = default_nid; | |
1daa6d08 BS |
616 | |
617 | fake_numa_create_new_node(((start + size) >> PAGE_SHIFT), &nid); | |
482ec7c4 | 618 | node_set_online(nid); |
1da177e4 | 619 | |
45fb6cea | 620 | if (!(size = numa_enforce_memory_limit(start, size))) { |
1da177e4 LT |
621 | if (--ranges) |
622 | goto new_range; | |
623 | else | |
624 | continue; | |
625 | } | |
626 | ||
c67c3cb4 MG |
627 | add_active_range(nid, start >> PAGE_SHIFT, |
628 | (start >> PAGE_SHIFT) + (size >> PAGE_SHIFT)); | |
1da177e4 LT |
629 | |
630 | if (--ranges) | |
631 | goto new_range; | |
632 | } | |
633 | ||
0204568a PM |
634 | /* |
635 | * Now do the same thing for each LMB listed in the ibm,dynamic-memory | |
636 | * property in the ibm,dynamic-reconfiguration-memory node. | |
637 | */ | |
638 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); | |
639 | if (memory) | |
640 | parse_drconf_memory(memory); | |
641 | ||
1da177e4 LT |
642 | return 0; |
643 | } | |
644 | ||
645 | static void __init setup_nonnuma(void) | |
646 | { | |
647 | unsigned long top_of_ram = lmb_end_of_DRAM(); | |
648 | unsigned long total_ram = lmb_phys_mem_size(); | |
c67c3cb4 | 649 | unsigned long start_pfn, end_pfn; |
1daa6d08 | 650 | unsigned int i, nid = 0; |
1da177e4 | 651 | |
e110b281 | 652 | printk(KERN_DEBUG "Top of RAM: 0x%lx, Total RAM: 0x%lx\n", |
1da177e4 | 653 | top_of_ram, total_ram); |
e110b281 | 654 | printk(KERN_DEBUG "Memory hole size: %ldMB\n", |
1da177e4 LT |
655 | (top_of_ram - total_ram) >> 20); |
656 | ||
c67c3cb4 MG |
657 | for (i = 0; i < lmb.memory.cnt; ++i) { |
658 | start_pfn = lmb.memory.region[i].base >> PAGE_SHIFT; | |
659 | end_pfn = start_pfn + lmb_size_pages(&lmb.memory, i); | |
1daa6d08 BS |
660 | |
661 | fake_numa_create_new_node(end_pfn, &nid); | |
662 | add_active_range(nid, start_pfn, end_pfn); | |
663 | node_set_online(nid); | |
c67c3cb4 | 664 | } |
1da177e4 LT |
665 | } |
666 | ||
4b703a23 AB |
667 | void __init dump_numa_cpu_topology(void) |
668 | { | |
669 | unsigned int node; | |
670 | unsigned int cpu, count; | |
671 | ||
672 | if (min_common_depth == -1 || !numa_enabled) | |
673 | return; | |
674 | ||
675 | for_each_online_node(node) { | |
e110b281 | 676 | printk(KERN_DEBUG "Node %d CPUs:", node); |
4b703a23 AB |
677 | |
678 | count = 0; | |
679 | /* | |
680 | * If we used a CPU iterator here we would miss printing | |
681 | * the holes in the cpumap. | |
682 | */ | |
683 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
684 | if (cpu_isset(cpu, numa_cpumask_lookup_table[node])) { | |
685 | if (count == 0) | |
686 | printk(" %u", cpu); | |
687 | ++count; | |
688 | } else { | |
689 | if (count > 1) | |
690 | printk("-%u", cpu - 1); | |
691 | count = 0; | |
692 | } | |
693 | } | |
694 | ||
695 | if (count > 1) | |
696 | printk("-%u", NR_CPUS - 1); | |
697 | printk("\n"); | |
698 | } | |
699 | } | |
700 | ||
701 | static void __init dump_numa_memory_topology(void) | |
1da177e4 LT |
702 | { |
703 | unsigned int node; | |
704 | unsigned int count; | |
705 | ||
706 | if (min_common_depth == -1 || !numa_enabled) | |
707 | return; | |
708 | ||
709 | for_each_online_node(node) { | |
710 | unsigned long i; | |
711 | ||
e110b281 | 712 | printk(KERN_DEBUG "Node %d Memory:", node); |
1da177e4 LT |
713 | |
714 | count = 0; | |
715 | ||
45fb6cea AB |
716 | for (i = 0; i < lmb_end_of_DRAM(); |
717 | i += (1 << SECTION_SIZE_BITS)) { | |
718 | if (early_pfn_to_nid(i >> PAGE_SHIFT) == node) { | |
1da177e4 LT |
719 | if (count == 0) |
720 | printk(" 0x%lx", i); | |
721 | ++count; | |
722 | } else { | |
723 | if (count > 0) | |
724 | printk("-0x%lx", i); | |
725 | count = 0; | |
726 | } | |
727 | } | |
728 | ||
729 | if (count > 0) | |
730 | printk("-0x%lx", i); | |
731 | printk("\n"); | |
732 | } | |
1da177e4 LT |
733 | } |
734 | ||
735 | /* | |
736 | * Allocate some memory, satisfying the lmb or bootmem allocator where | |
737 | * required. nid is the preferred node and end is the physical address of | |
738 | * the highest address in the node. | |
739 | * | |
740 | * Returns the physical address of the memory. | |
741 | */ | |
45fb6cea AB |
742 | static void __init *careful_allocation(int nid, unsigned long size, |
743 | unsigned long align, | |
744 | unsigned long end_pfn) | |
1da177e4 | 745 | { |
45fb6cea | 746 | int new_nid; |
d7a5b2ff | 747 | unsigned long ret = __lmb_alloc_base(size, align, end_pfn << PAGE_SHIFT); |
1da177e4 LT |
748 | |
749 | /* retry over all memory */ | |
750 | if (!ret) | |
d7a5b2ff | 751 | ret = __lmb_alloc_base(size, align, lmb_end_of_DRAM()); |
1da177e4 LT |
752 | |
753 | if (!ret) | |
754 | panic("numa.c: cannot allocate %lu bytes on node %d", | |
755 | size, nid); | |
756 | ||
757 | /* | |
758 | * If the memory came from a previously allocated node, we must | |
759 | * retry with the bootmem allocator. | |
760 | */ | |
45fb6cea AB |
761 | new_nid = early_pfn_to_nid(ret >> PAGE_SHIFT); |
762 | if (new_nid < nid) { | |
763 | ret = (unsigned long)__alloc_bootmem_node(NODE_DATA(new_nid), | |
1da177e4 LT |
764 | size, align, 0); |
765 | ||
766 | if (!ret) | |
767 | panic("numa.c: cannot allocate %lu bytes on node %d", | |
45fb6cea | 768 | size, new_nid); |
1da177e4 | 769 | |
45fb6cea | 770 | ret = __pa(ret); |
1da177e4 LT |
771 | |
772 | dbg("alloc_bootmem %lx %lx\n", ret, size); | |
773 | } | |
774 | ||
45fb6cea | 775 | return (void *)ret; |
1da177e4 LT |
776 | } |
777 | ||
74b85f37 CS |
778 | static struct notifier_block __cpuinitdata ppc64_numa_nb = { |
779 | .notifier_call = cpu_numa_callback, | |
780 | .priority = 1 /* Must run before sched domains notifier. */ | |
781 | }; | |
782 | ||
1da177e4 LT |
783 | void __init do_init_bootmem(void) |
784 | { | |
785 | int nid; | |
45fb6cea | 786 | unsigned int i; |
1da177e4 LT |
787 | |
788 | min_low_pfn = 0; | |
789 | max_low_pfn = lmb_end_of_DRAM() >> PAGE_SHIFT; | |
790 | max_pfn = max_low_pfn; | |
791 | ||
792 | if (parse_numa_properties()) | |
793 | setup_nonnuma(); | |
794 | else | |
4b703a23 | 795 | dump_numa_memory_topology(); |
1da177e4 LT |
796 | |
797 | register_cpu_notifier(&ppc64_numa_nb); | |
2b261227 NL |
798 | cpu_numa_callback(&ppc64_numa_nb, CPU_UP_PREPARE, |
799 | (void *)(unsigned long)boot_cpuid); | |
1da177e4 LT |
800 | |
801 | for_each_online_node(nid) { | |
c67c3cb4 | 802 | unsigned long start_pfn, end_pfn; |
1da177e4 LT |
803 | unsigned long bootmem_paddr; |
804 | unsigned long bootmap_pages; | |
805 | ||
c67c3cb4 | 806 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); |
1da177e4 LT |
807 | |
808 | /* Allocate the node structure node local if possible */ | |
45fb6cea | 809 | NODE_DATA(nid) = careful_allocation(nid, |
1da177e4 | 810 | sizeof(struct pglist_data), |
45fb6cea AB |
811 | SMP_CACHE_BYTES, end_pfn); |
812 | NODE_DATA(nid) = __va(NODE_DATA(nid)); | |
1da177e4 LT |
813 | memset(NODE_DATA(nid), 0, sizeof(struct pglist_data)); |
814 | ||
815 | dbg("node %d\n", nid); | |
816 | dbg("NODE_DATA() = %p\n", NODE_DATA(nid)); | |
817 | ||
b61bfa3c | 818 | NODE_DATA(nid)->bdata = &bootmem_node_data[nid]; |
45fb6cea AB |
819 | NODE_DATA(nid)->node_start_pfn = start_pfn; |
820 | NODE_DATA(nid)->node_spanned_pages = end_pfn - start_pfn; | |
1da177e4 LT |
821 | |
822 | if (NODE_DATA(nid)->node_spanned_pages == 0) | |
823 | continue; | |
824 | ||
45fb6cea AB |
825 | dbg("start_paddr = %lx\n", start_pfn << PAGE_SHIFT); |
826 | dbg("end_paddr = %lx\n", end_pfn << PAGE_SHIFT); | |
1da177e4 | 827 | |
45fb6cea AB |
828 | bootmap_pages = bootmem_bootmap_pages(end_pfn - start_pfn); |
829 | bootmem_paddr = (unsigned long)careful_allocation(nid, | |
830 | bootmap_pages << PAGE_SHIFT, | |
831 | PAGE_SIZE, end_pfn); | |
832 | memset(__va(bootmem_paddr), 0, bootmap_pages << PAGE_SHIFT); | |
1da177e4 | 833 | |
1da177e4 LT |
834 | dbg("bootmap_paddr = %lx\n", bootmem_paddr); |
835 | ||
836 | init_bootmem_node(NODE_DATA(nid), bootmem_paddr >> PAGE_SHIFT, | |
45fb6cea | 837 | start_pfn, end_pfn); |
1da177e4 | 838 | |
c67c3cb4 | 839 | free_bootmem_with_active_regions(nid, end_pfn); |
1da177e4 | 840 | |
45fb6cea | 841 | /* Mark reserved regions on this node */ |
1da177e4 | 842 | for (i = 0; i < lmb.reserved.cnt; i++) { |
180379dc | 843 | unsigned long physbase = lmb.reserved.region[i].base; |
1da177e4 | 844 | unsigned long size = lmb.reserved.region[i].size; |
45fb6cea AB |
845 | unsigned long start_paddr = start_pfn << PAGE_SHIFT; |
846 | unsigned long end_paddr = end_pfn << PAGE_SHIFT; | |
1da177e4 | 847 | |
45fb6cea AB |
848 | if (early_pfn_to_nid(physbase >> PAGE_SHIFT) != nid && |
849 | early_pfn_to_nid((physbase+size-1) >> PAGE_SHIFT) != nid) | |
1da177e4 LT |
850 | continue; |
851 | ||
852 | if (physbase < end_paddr && | |
853 | (physbase+size) > start_paddr) { | |
854 | /* overlaps */ | |
855 | if (physbase < start_paddr) { | |
856 | size -= start_paddr - physbase; | |
857 | physbase = start_paddr; | |
858 | } | |
859 | ||
860 | if (size > end_paddr - physbase) | |
861 | size = end_paddr - physbase; | |
862 | ||
863 | dbg("reserve_bootmem %lx %lx\n", physbase, | |
864 | size); | |
865 | reserve_bootmem_node(NODE_DATA(nid), physbase, | |
72a7fe39 | 866 | size, BOOTMEM_DEFAULT); |
1da177e4 LT |
867 | } |
868 | } | |
802f192e | 869 | |
c67c3cb4 | 870 | sparse_memory_present_with_active_regions(nid); |
1da177e4 LT |
871 | } |
872 | } | |
873 | ||
874 | void __init paging_init(void) | |
875 | { | |
6391af17 MG |
876 | unsigned long max_zone_pfns[MAX_NR_ZONES]; |
877 | memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); | |
878 | max_zone_pfns[ZONE_DMA] = lmb_end_of_DRAM() >> PAGE_SHIFT; | |
c67c3cb4 | 879 | free_area_init_nodes(max_zone_pfns); |
1da177e4 LT |
880 | } |
881 | ||
882 | static int __init early_numa(char *p) | |
883 | { | |
884 | if (!p) | |
885 | return 0; | |
886 | ||
887 | if (strstr(p, "off")) | |
888 | numa_enabled = 0; | |
889 | ||
890 | if (strstr(p, "debug")) | |
891 | numa_debug = 1; | |
892 | ||
1daa6d08 BS |
893 | p = strstr(p, "fake="); |
894 | if (p) | |
895 | cmdline = p + strlen("fake="); | |
896 | ||
1da177e4 LT |
897 | return 0; |
898 | } | |
899 | early_param("numa", early_numa); | |
237a0989 MK |
900 | |
901 | #ifdef CONFIG_MEMORY_HOTPLUG | |
0db9360a NF |
902 | /* |
903 | * Validate the node associated with the memory section we are | |
904 | * trying to add. | |
905 | */ | |
906 | int valid_hot_add_scn(int *nid, unsigned long start, u32 lmb_size, | |
907 | unsigned long scn_addr) | |
908 | { | |
909 | nodemask_t nodes; | |
910 | ||
911 | if (*nid < 0 || !node_online(*nid)) | |
912 | *nid = any_online_node(NODE_MASK_ALL); | |
913 | ||
914 | if ((scn_addr >= start) && (scn_addr < (start + lmb_size))) { | |
915 | nodes_setall(nodes); | |
916 | while (NODE_DATA(*nid)->node_spanned_pages == 0) { | |
917 | node_clear(*nid, nodes); | |
918 | *nid = any_online_node(nodes); | |
919 | } | |
920 | ||
921 | return 1; | |
922 | } | |
923 | ||
924 | return 0; | |
925 | } | |
926 | ||
927 | /* | |
928 | * Find the node associated with a hot added memory section represented | |
929 | * by the ibm,dynamic-reconfiguration-memory node. | |
930 | */ | |
931 | static int hot_add_drconf_scn_to_nid(struct device_node *memory, | |
932 | unsigned long scn_addr) | |
933 | { | |
934 | const u32 *dm; | |
935 | unsigned int n, rc; | |
936 | unsigned long lmb_size; | |
937 | int default_nid = any_online_node(NODE_MASK_ALL); | |
938 | int nid; | |
939 | struct assoc_arrays aa; | |
940 | ||
941 | n = of_get_drconf_memory(memory, &dm); | |
942 | if (!n) | |
943 | return default_nid;; | |
944 | ||
945 | lmb_size = of_get_lmb_size(memory); | |
946 | if (!lmb_size) | |
947 | return default_nid; | |
948 | ||
949 | rc = of_get_assoc_arrays(memory, &aa); | |
950 | if (rc) | |
951 | return default_nid; | |
952 | ||
953 | for (; n != 0; --n) { | |
954 | struct of_drconf_cell drmem; | |
955 | ||
956 | read_drconf_cell(&drmem, &dm); | |
957 | ||
958 | /* skip this block if it is reserved or not assigned to | |
959 | * this partition */ | |
960 | if ((drmem.flags & DRCONF_MEM_RESERVED) | |
961 | || !(drmem.flags & DRCONF_MEM_ASSIGNED)) | |
962 | continue; | |
963 | ||
964 | nid = of_drconf_to_nid_single(&drmem, &aa); | |
965 | ||
966 | if (valid_hot_add_scn(&nid, drmem.base_addr, lmb_size, | |
967 | scn_addr)) | |
968 | return nid; | |
969 | } | |
970 | ||
971 | BUG(); /* section address should be found above */ | |
972 | return 0; | |
973 | } | |
974 | ||
237a0989 MK |
975 | /* |
976 | * Find the node associated with a hot added memory section. Section | |
977 | * corresponds to a SPARSEMEM section, not an LMB. It is assumed that | |
978 | * sections are fully contained within a single LMB. | |
979 | */ | |
980 | int hot_add_scn_to_nid(unsigned long scn_addr) | |
981 | { | |
982 | struct device_node *memory = NULL; | |
069007ae | 983 | int nid; |
237a0989 MK |
984 | |
985 | if (!numa_enabled || (min_common_depth < 0)) | |
0db9360a NF |
986 | return any_online_node(NODE_MASK_ALL); |
987 | ||
988 | memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory"); | |
989 | if (memory) { | |
990 | nid = hot_add_drconf_scn_to_nid(memory, scn_addr); | |
991 | of_node_put(memory); | |
992 | return nid; | |
993 | } | |
237a0989 MK |
994 | |
995 | while ((memory = of_find_node_by_type(memory, "memory")) != NULL) { | |
996 | unsigned long start, size; | |
b226e462 | 997 | int ranges; |
a7f67bdf | 998 | const unsigned int *memcell_buf; |
237a0989 MK |
999 | unsigned int len; |
1000 | ||
e2eb6392 | 1001 | memcell_buf = of_get_property(memory, "reg", &len); |
237a0989 MK |
1002 | if (!memcell_buf || len <= 0) |
1003 | continue; | |
1004 | ||
cc5d0189 BH |
1005 | /* ranges in cell */ |
1006 | ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells); | |
237a0989 MK |
1007 | ha_new_range: |
1008 | start = read_n_cells(n_mem_addr_cells, &memcell_buf); | |
1009 | size = read_n_cells(n_mem_size_cells, &memcell_buf); | |
953039c8 | 1010 | nid = of_node_to_nid_single(memory); |
237a0989 | 1011 | |
0db9360a | 1012 | if (valid_hot_add_scn(&nid, start, size, scn_addr)) { |
237a0989 | 1013 | of_node_put(memory); |
0db9360a | 1014 | return nid; |
237a0989 MK |
1015 | } |
1016 | ||
1017 | if (--ranges) /* process all ranges in cell */ | |
1018 | goto ha_new_range; | |
1019 | } | |
237a0989 | 1020 | BUG(); /* section address should be found above */ |
069007ae | 1021 | return 0; |
237a0989 MK |
1022 | } |
1023 | #endif /* CONFIG_MEMORY_HOTPLUG */ |