]>
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> | |
45fb6cea | 20 | #include <asm/sparsemem.h> |
1da177e4 | 21 | #include <asm/lmb.h> |
cf00a8d1 | 22 | #include <asm/system.h> |
2249ca9d | 23 | #include <asm/smp.h> |
1da177e4 LT |
24 | |
25 | static int numa_enabled = 1; | |
26 | ||
27 | static int numa_debug; | |
28 | #define dbg(args...) if (numa_debug) { printk(KERN_INFO args); } | |
29 | ||
45fb6cea | 30 | int numa_cpu_lookup_table[NR_CPUS]; |
1da177e4 | 31 | cpumask_t numa_cpumask_lookup_table[MAX_NUMNODES]; |
1da177e4 | 32 | struct pglist_data *node_data[MAX_NUMNODES]; |
45fb6cea AB |
33 | |
34 | EXPORT_SYMBOL(numa_cpu_lookup_table); | |
35 | EXPORT_SYMBOL(numa_cpumask_lookup_table); | |
36 | EXPORT_SYMBOL(node_data); | |
37 | ||
38 | static bootmem_data_t __initdata plat_node_bdata[MAX_NUMNODES]; | |
1da177e4 | 39 | static int min_common_depth; |
237a0989 | 40 | static int n_mem_addr_cells, n_mem_size_cells; |
1da177e4 LT |
41 | |
42 | /* | |
45fb6cea | 43 | * We need somewhere to store start/end/node for each region until we have |
1da177e4 LT |
44 | * allocated the real node_data structures. |
45 | */ | |
45fb6cea | 46 | #define MAX_REGIONS (MAX_LMB_REGIONS*2) |
1da177e4 | 47 | static struct { |
45fb6cea AB |
48 | unsigned long start_pfn; |
49 | unsigned long end_pfn; | |
50 | int nid; | |
51 | } init_node_data[MAX_REGIONS] __initdata; | |
1da177e4 | 52 | |
45fb6cea AB |
53 | int __init early_pfn_to_nid(unsigned long pfn) |
54 | { | |
55 | unsigned int i; | |
56 | ||
57 | for (i = 0; init_node_data[i].end_pfn; i++) { | |
58 | unsigned long start_pfn = init_node_data[i].start_pfn; | |
59 | unsigned long end_pfn = init_node_data[i].end_pfn; | |
60 | ||
61 | if ((start_pfn <= pfn) && (pfn < end_pfn)) | |
62 | return init_node_data[i].nid; | |
63 | } | |
64 | ||
65 | return -1; | |
66 | } | |
67 | ||
68 | void __init add_region(unsigned int nid, unsigned long start_pfn, | |
69 | unsigned long pages) | |
70 | { | |
71 | unsigned int i; | |
72 | ||
73 | dbg("add_region nid %d start_pfn 0x%lx pages 0x%lx\n", | |
74 | nid, start_pfn, pages); | |
75 | ||
76 | for (i = 0; init_node_data[i].end_pfn; i++) { | |
77 | if (init_node_data[i].nid != nid) | |
78 | continue; | |
79 | if (init_node_data[i].end_pfn == start_pfn) { | |
80 | init_node_data[i].end_pfn += pages; | |
81 | return; | |
82 | } | |
83 | if (init_node_data[i].start_pfn == (start_pfn + pages)) { | |
84 | init_node_data[i].start_pfn -= pages; | |
85 | return; | |
86 | } | |
87 | } | |
88 | ||
89 | /* | |
90 | * Leave last entry NULL so we dont iterate off the end (we use | |
91 | * entry.end_pfn to terminate the walk). | |
92 | */ | |
93 | if (i >= (MAX_REGIONS - 1)) { | |
94 | printk(KERN_ERR "WARNING: too many memory regions in " | |
95 | "numa code, truncating\n"); | |
96 | return; | |
97 | } | |
98 | ||
99 | init_node_data[i].start_pfn = start_pfn; | |
100 | init_node_data[i].end_pfn = start_pfn + pages; | |
101 | init_node_data[i].nid = nid; | |
102 | } | |
103 | ||
104 | /* We assume init_node_data has no overlapping regions */ | |
105 | void __init get_region(unsigned int nid, unsigned long *start_pfn, | |
106 | unsigned long *end_pfn, unsigned long *pages_present) | |
107 | { | |
108 | unsigned int i; | |
109 | ||
110 | *start_pfn = -1UL; | |
111 | *end_pfn = *pages_present = 0; | |
112 | ||
113 | for (i = 0; init_node_data[i].end_pfn; i++) { | |
114 | if (init_node_data[i].nid != nid) | |
115 | continue; | |
116 | ||
117 | *pages_present += init_node_data[i].end_pfn - | |
118 | init_node_data[i].start_pfn; | |
119 | ||
120 | if (init_node_data[i].start_pfn < *start_pfn) | |
121 | *start_pfn = init_node_data[i].start_pfn; | |
122 | ||
123 | if (init_node_data[i].end_pfn > *end_pfn) | |
124 | *end_pfn = init_node_data[i].end_pfn; | |
125 | } | |
126 | ||
127 | /* We didnt find a matching region, return start/end as 0 */ | |
128 | if (*start_pfn == -1UL) | |
6d91bb93 | 129 | *start_pfn = 0; |
45fb6cea | 130 | } |
1da177e4 | 131 | |
2e5ce39d | 132 | static void __cpuinit map_cpu_to_node(int cpu, int node) |
1da177e4 LT |
133 | { |
134 | numa_cpu_lookup_table[cpu] = node; | |
45fb6cea | 135 | |
bf4b85b0 NL |
136 | dbg("adding cpu %d to node %d\n", cpu, node); |
137 | ||
45fb6cea | 138 | if (!(cpu_isset(cpu, numa_cpumask_lookup_table[node]))) |
1da177e4 | 139 | cpu_set(cpu, numa_cpumask_lookup_table[node]); |
1da177e4 LT |
140 | } |
141 | ||
142 | #ifdef CONFIG_HOTPLUG_CPU | |
143 | static void unmap_cpu_from_node(unsigned long cpu) | |
144 | { | |
145 | int node = numa_cpu_lookup_table[cpu]; | |
146 | ||
147 | dbg("removing cpu %lu from node %d\n", cpu, node); | |
148 | ||
149 | if (cpu_isset(cpu, numa_cpumask_lookup_table[node])) { | |
150 | cpu_clear(cpu, numa_cpumask_lookup_table[node]); | |
1da177e4 LT |
151 | } else { |
152 | printk(KERN_ERR "WARNING: cpu %lu not found in node %d\n", | |
153 | cpu, node); | |
154 | } | |
155 | } | |
156 | #endif /* CONFIG_HOTPLUG_CPU */ | |
157 | ||
2e5ce39d | 158 | static struct device_node * __cpuinit find_cpu_node(unsigned int cpu) |
1da177e4 LT |
159 | { |
160 | unsigned int hw_cpuid = get_hard_smp_processor_id(cpu); | |
161 | struct device_node *cpu_node = NULL; | |
a7f67bdf | 162 | const unsigned int *interrupt_server, *reg; |
1da177e4 LT |
163 | int len; |
164 | ||
165 | while ((cpu_node = of_find_node_by_type(cpu_node, "cpu")) != NULL) { | |
166 | /* Try interrupt server first */ | |
a7f67bdf | 167 | interrupt_server = get_property(cpu_node, |
1da177e4 LT |
168 | "ibm,ppc-interrupt-server#s", &len); |
169 | ||
170 | len = len / sizeof(u32); | |
171 | ||
172 | if (interrupt_server && (len > 0)) { | |
173 | while (len--) { | |
174 | if (interrupt_server[len] == hw_cpuid) | |
175 | return cpu_node; | |
176 | } | |
177 | } else { | |
a7f67bdf | 178 | reg = get_property(cpu_node, "reg", &len); |
1da177e4 LT |
179 | if (reg && (len > 0) && (reg[0] == hw_cpuid)) |
180 | return cpu_node; | |
181 | } | |
182 | } | |
183 | ||
184 | return NULL; | |
185 | } | |
186 | ||
187 | /* must hold reference to node during call */ | |
a7f67bdf | 188 | static const int *of_get_associativity(struct device_node *dev) |
1da177e4 | 189 | { |
a7f67bdf | 190 | return get_property(dev, "ibm,associativity", NULL); |
1da177e4 LT |
191 | } |
192 | ||
482ec7c4 NL |
193 | /* Returns nid in the range [0..MAX_NUMNODES-1], or -1 if no useful numa |
194 | * info is found. | |
195 | */ | |
953039c8 | 196 | static int of_node_to_nid_single(struct device_node *device) |
1da177e4 | 197 | { |
482ec7c4 | 198 | int nid = -1; |
a7f67bdf | 199 | const unsigned int *tmp; |
1da177e4 LT |
200 | |
201 | if (min_common_depth == -1) | |
482ec7c4 | 202 | goto out; |
1da177e4 LT |
203 | |
204 | tmp = of_get_associativity(device); | |
482ec7c4 NL |
205 | if (!tmp) |
206 | goto out; | |
207 | ||
208 | if (tmp[0] >= min_common_depth) | |
cf950b7a | 209 | nid = tmp[min_common_depth]; |
bc16a759 NL |
210 | |
211 | /* POWER4 LPAR uses 0xffff as invalid node */ | |
482ec7c4 NL |
212 | if (nid == 0xffff || nid >= MAX_NUMNODES) |
213 | nid = -1; | |
214 | out: | |
cf950b7a | 215 | return nid; |
1da177e4 LT |
216 | } |
217 | ||
953039c8 JK |
218 | /* Walk the device tree upwards, looking for an associativity id */ |
219 | int of_node_to_nid(struct device_node *device) | |
220 | { | |
221 | struct device_node *tmp; | |
222 | int nid = -1; | |
223 | ||
224 | of_node_get(device); | |
225 | while (device) { | |
226 | nid = of_node_to_nid_single(device); | |
227 | if (nid != -1) | |
228 | break; | |
229 | ||
230 | tmp = device; | |
231 | device = of_get_parent(tmp); | |
232 | of_node_put(tmp); | |
233 | } | |
234 | of_node_put(device); | |
235 | ||
236 | return nid; | |
237 | } | |
238 | EXPORT_SYMBOL_GPL(of_node_to_nid); | |
239 | ||
1da177e4 LT |
240 | /* |
241 | * In theory, the "ibm,associativity" property may contain multiple | |
242 | * associativity lists because a resource may be multiply connected | |
243 | * into the machine. This resource then has different associativity | |
244 | * characteristics relative to its multiple connections. We ignore | |
245 | * this for now. We also assume that all cpu and memory sets have | |
246 | * their distances represented at a common level. This won't be | |
247 | * true for heirarchical NUMA. | |
248 | * | |
249 | * In any case the ibm,associativity-reference-points should give | |
250 | * the correct depth for a normal NUMA system. | |
251 | * | |
252 | * - Dave Hansen <haveblue@us.ibm.com> | |
253 | */ | |
254 | static int __init find_min_common_depth(void) | |
255 | { | |
256 | int depth; | |
a7f67bdf | 257 | const unsigned int *ref_points; |
1da177e4 LT |
258 | struct device_node *rtas_root; |
259 | unsigned int len; | |
260 | ||
261 | rtas_root = of_find_node_by_path("/rtas"); | |
262 | ||
263 | if (!rtas_root) | |
264 | return -1; | |
265 | ||
266 | /* | |
267 | * this property is 2 32-bit integers, each representing a level of | |
268 | * depth in the associativity nodes. The first is for an SMP | |
269 | * configuration (should be all 0's) and the second is for a normal | |
270 | * NUMA configuration. | |
271 | */ | |
a7f67bdf | 272 | ref_points = get_property(rtas_root, |
1da177e4 LT |
273 | "ibm,associativity-reference-points", &len); |
274 | ||
275 | if ((len >= 1) && ref_points) { | |
276 | depth = ref_points[1]; | |
277 | } else { | |
bf4b85b0 | 278 | dbg("NUMA: ibm,associativity-reference-points not found.\n"); |
1da177e4 LT |
279 | depth = -1; |
280 | } | |
281 | of_node_put(rtas_root); | |
282 | ||
283 | return depth; | |
284 | } | |
285 | ||
84c9fdd1 | 286 | static void __init get_n_mem_cells(int *n_addr_cells, int *n_size_cells) |
1da177e4 LT |
287 | { |
288 | struct device_node *memory = NULL; | |
1da177e4 LT |
289 | |
290 | memory = of_find_node_by_type(memory, "memory"); | |
54c23310 | 291 | if (!memory) |
84c9fdd1 | 292 | panic("numa.c: No memory nodes found!"); |
54c23310 | 293 | |
84c9fdd1 MK |
294 | *n_addr_cells = prom_n_addr_cells(memory); |
295 | *n_size_cells = prom_n_size_cells(memory); | |
296 | of_node_put(memory); | |
1da177e4 LT |
297 | } |
298 | ||
a7f67bdf | 299 | static unsigned long __devinit read_n_cells(int n, const unsigned int **buf) |
1da177e4 LT |
300 | { |
301 | unsigned long result = 0; | |
302 | ||
303 | while (n--) { | |
304 | result = (result << 32) | **buf; | |
305 | (*buf)++; | |
306 | } | |
307 | return result; | |
308 | } | |
309 | ||
310 | /* | |
311 | * Figure out to which domain a cpu belongs and stick it there. | |
312 | * Return the id of the domain used. | |
313 | */ | |
2e5ce39d | 314 | static int __cpuinit numa_setup_cpu(unsigned long lcpu) |
1da177e4 | 315 | { |
cf950b7a | 316 | int nid = 0; |
1da177e4 LT |
317 | struct device_node *cpu = find_cpu_node(lcpu); |
318 | ||
319 | if (!cpu) { | |
320 | WARN_ON(1); | |
321 | goto out; | |
322 | } | |
323 | ||
953039c8 | 324 | nid = of_node_to_nid_single(cpu); |
1da177e4 | 325 | |
482ec7c4 NL |
326 | if (nid < 0 || !node_online(nid)) |
327 | nid = any_online_node(NODE_MASK_ALL); | |
1da177e4 | 328 | out: |
cf950b7a | 329 | map_cpu_to_node(lcpu, nid); |
1da177e4 LT |
330 | |
331 | of_node_put(cpu); | |
332 | ||
cf950b7a | 333 | return nid; |
1da177e4 LT |
334 | } |
335 | ||
74b85f37 | 336 | static int __cpuinit cpu_numa_callback(struct notifier_block *nfb, |
1da177e4 LT |
337 | unsigned long action, |
338 | void *hcpu) | |
339 | { | |
340 | unsigned long lcpu = (unsigned long)hcpu; | |
341 | int ret = NOTIFY_DONE; | |
342 | ||
343 | switch (action) { | |
344 | case CPU_UP_PREPARE: | |
2b261227 | 345 | numa_setup_cpu(lcpu); |
1da177e4 LT |
346 | ret = NOTIFY_OK; |
347 | break; | |
348 | #ifdef CONFIG_HOTPLUG_CPU | |
349 | case CPU_DEAD: | |
350 | case CPU_UP_CANCELED: | |
351 | unmap_cpu_from_node(lcpu); | |
352 | break; | |
353 | ret = NOTIFY_OK; | |
354 | #endif | |
355 | } | |
356 | return ret; | |
357 | } | |
358 | ||
359 | /* | |
360 | * Check and possibly modify a memory region to enforce the memory limit. | |
361 | * | |
362 | * Returns the size the region should have to enforce the memory limit. | |
363 | * This will either be the original value of size, a truncated value, | |
364 | * or zero. If the returned value of size is 0 the region should be | |
365 | * discarded as it lies wholy above the memory limit. | |
366 | */ | |
45fb6cea AB |
367 | static unsigned long __init numa_enforce_memory_limit(unsigned long start, |
368 | unsigned long size) | |
1da177e4 LT |
369 | { |
370 | /* | |
371 | * We use lmb_end_of_DRAM() in here instead of memory_limit because | |
372 | * we've already adjusted it for the limit and it takes care of | |
373 | * having memory holes below the limit. | |
374 | */ | |
1da177e4 LT |
375 | |
376 | if (! memory_limit) | |
377 | return size; | |
378 | ||
379 | if (start + size <= lmb_end_of_DRAM()) | |
380 | return size; | |
381 | ||
382 | if (start >= lmb_end_of_DRAM()) | |
383 | return 0; | |
384 | ||
385 | return lmb_end_of_DRAM() - start; | |
386 | } | |
387 | ||
388 | static int __init parse_numa_properties(void) | |
389 | { | |
390 | struct device_node *cpu = NULL; | |
391 | struct device_node *memory = NULL; | |
482ec7c4 | 392 | int default_nid = 0; |
1da177e4 LT |
393 | unsigned long i; |
394 | ||
395 | if (numa_enabled == 0) { | |
396 | printk(KERN_WARNING "NUMA disabled by user\n"); | |
397 | return -1; | |
398 | } | |
399 | ||
1da177e4 LT |
400 | min_common_depth = find_min_common_depth(); |
401 | ||
1da177e4 LT |
402 | if (min_common_depth < 0) |
403 | return min_common_depth; | |
404 | ||
bf4b85b0 NL |
405 | dbg("NUMA associativity depth for CPU/Memory: %d\n", min_common_depth); |
406 | ||
1da177e4 | 407 | /* |
482ec7c4 NL |
408 | * Even though we connect cpus to numa domains later in SMP |
409 | * init, we need to know the node ids now. This is because | |
410 | * each node to be onlined must have NODE_DATA etc backing it. | |
1da177e4 | 411 | */ |
482ec7c4 | 412 | for_each_present_cpu(i) { |
cf950b7a | 413 | int nid; |
1da177e4 LT |
414 | |
415 | cpu = find_cpu_node(i); | |
482ec7c4 | 416 | BUG_ON(!cpu); |
953039c8 | 417 | nid = of_node_to_nid_single(cpu); |
482ec7c4 | 418 | of_node_put(cpu); |
1da177e4 | 419 | |
482ec7c4 NL |
420 | /* |
421 | * Don't fall back to default_nid yet -- we will plug | |
422 | * cpus into nodes once the memory scan has discovered | |
423 | * the topology. | |
424 | */ | |
425 | if (nid < 0) | |
426 | continue; | |
427 | node_set_online(nid); | |
1da177e4 LT |
428 | } |
429 | ||
237a0989 | 430 | get_n_mem_cells(&n_mem_addr_cells, &n_mem_size_cells); |
1da177e4 LT |
431 | memory = NULL; |
432 | while ((memory = of_find_node_by_type(memory, "memory")) != NULL) { | |
433 | unsigned long start; | |
434 | unsigned long size; | |
cf950b7a | 435 | int nid; |
1da177e4 | 436 | int ranges; |
a7f67bdf | 437 | const unsigned int *memcell_buf; |
1da177e4 LT |
438 | unsigned int len; |
439 | ||
a7f67bdf | 440 | memcell_buf = get_property(memory, |
ba759485 ME |
441 | "linux,usable-memory", &len); |
442 | if (!memcell_buf || len <= 0) | |
a7f67bdf | 443 | memcell_buf = get_property(memory, "reg", &len); |
1da177e4 LT |
444 | if (!memcell_buf || len <= 0) |
445 | continue; | |
446 | ||
cc5d0189 BH |
447 | /* ranges in cell */ |
448 | ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells); | |
1da177e4 LT |
449 | new_range: |
450 | /* these are order-sensitive, and modify the buffer pointer */ | |
237a0989 MK |
451 | start = read_n_cells(n_mem_addr_cells, &memcell_buf); |
452 | size = read_n_cells(n_mem_size_cells, &memcell_buf); | |
1da177e4 | 453 | |
482ec7c4 NL |
454 | /* |
455 | * Assumption: either all memory nodes or none will | |
456 | * have associativity properties. If none, then | |
457 | * everything goes to default_nid. | |
458 | */ | |
953039c8 | 459 | nid = of_node_to_nid_single(memory); |
482ec7c4 NL |
460 | if (nid < 0) |
461 | nid = default_nid; | |
462 | node_set_online(nid); | |
1da177e4 | 463 | |
45fb6cea | 464 | if (!(size = numa_enforce_memory_limit(start, size))) { |
1da177e4 LT |
465 | if (--ranges) |
466 | goto new_range; | |
467 | else | |
468 | continue; | |
469 | } | |
470 | ||
cf950b7a | 471 | add_region(nid, start >> PAGE_SHIFT, |
45fb6cea | 472 | size >> PAGE_SHIFT); |
1da177e4 LT |
473 | |
474 | if (--ranges) | |
475 | goto new_range; | |
476 | } | |
477 | ||
1da177e4 LT |
478 | return 0; |
479 | } | |
480 | ||
481 | static void __init setup_nonnuma(void) | |
482 | { | |
483 | unsigned long top_of_ram = lmb_end_of_DRAM(); | |
484 | unsigned long total_ram = lmb_phys_mem_size(); | |
fb6d73d3 | 485 | unsigned int i; |
1da177e4 | 486 | |
e110b281 | 487 | printk(KERN_DEBUG "Top of RAM: 0x%lx, Total RAM: 0x%lx\n", |
1da177e4 | 488 | top_of_ram, total_ram); |
e110b281 | 489 | printk(KERN_DEBUG "Memory hole size: %ldMB\n", |
1da177e4 LT |
490 | (top_of_ram - total_ram) >> 20); |
491 | ||
fb6d73d3 PM |
492 | for (i = 0; i < lmb.memory.cnt; ++i) |
493 | add_region(0, lmb.memory.region[i].base >> PAGE_SHIFT, | |
494 | lmb_size_pages(&lmb.memory, i)); | |
1da177e4 | 495 | node_set_online(0); |
1da177e4 LT |
496 | } |
497 | ||
4b703a23 AB |
498 | void __init dump_numa_cpu_topology(void) |
499 | { | |
500 | unsigned int node; | |
501 | unsigned int cpu, count; | |
502 | ||
503 | if (min_common_depth == -1 || !numa_enabled) | |
504 | return; | |
505 | ||
506 | for_each_online_node(node) { | |
e110b281 | 507 | printk(KERN_DEBUG "Node %d CPUs:", node); |
4b703a23 AB |
508 | |
509 | count = 0; | |
510 | /* | |
511 | * If we used a CPU iterator here we would miss printing | |
512 | * the holes in the cpumap. | |
513 | */ | |
514 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
515 | if (cpu_isset(cpu, numa_cpumask_lookup_table[node])) { | |
516 | if (count == 0) | |
517 | printk(" %u", cpu); | |
518 | ++count; | |
519 | } else { | |
520 | if (count > 1) | |
521 | printk("-%u", cpu - 1); | |
522 | count = 0; | |
523 | } | |
524 | } | |
525 | ||
526 | if (count > 1) | |
527 | printk("-%u", NR_CPUS - 1); | |
528 | printk("\n"); | |
529 | } | |
530 | } | |
531 | ||
532 | static void __init dump_numa_memory_topology(void) | |
1da177e4 LT |
533 | { |
534 | unsigned int node; | |
535 | unsigned int count; | |
536 | ||
537 | if (min_common_depth == -1 || !numa_enabled) | |
538 | return; | |
539 | ||
540 | for_each_online_node(node) { | |
541 | unsigned long i; | |
542 | ||
e110b281 | 543 | printk(KERN_DEBUG "Node %d Memory:", node); |
1da177e4 LT |
544 | |
545 | count = 0; | |
546 | ||
45fb6cea AB |
547 | for (i = 0; i < lmb_end_of_DRAM(); |
548 | i += (1 << SECTION_SIZE_BITS)) { | |
549 | if (early_pfn_to_nid(i >> PAGE_SHIFT) == node) { | |
1da177e4 LT |
550 | if (count == 0) |
551 | printk(" 0x%lx", i); | |
552 | ++count; | |
553 | } else { | |
554 | if (count > 0) | |
555 | printk("-0x%lx", i); | |
556 | count = 0; | |
557 | } | |
558 | } | |
559 | ||
560 | if (count > 0) | |
561 | printk("-0x%lx", i); | |
562 | printk("\n"); | |
563 | } | |
1da177e4 LT |
564 | } |
565 | ||
566 | /* | |
567 | * Allocate some memory, satisfying the lmb or bootmem allocator where | |
568 | * required. nid is the preferred node and end is the physical address of | |
569 | * the highest address in the node. | |
570 | * | |
571 | * Returns the physical address of the memory. | |
572 | */ | |
45fb6cea AB |
573 | static void __init *careful_allocation(int nid, unsigned long size, |
574 | unsigned long align, | |
575 | unsigned long end_pfn) | |
1da177e4 | 576 | { |
45fb6cea | 577 | int new_nid; |
d7a5b2ff | 578 | unsigned long ret = __lmb_alloc_base(size, align, end_pfn << PAGE_SHIFT); |
1da177e4 LT |
579 | |
580 | /* retry over all memory */ | |
581 | if (!ret) | |
d7a5b2ff | 582 | ret = __lmb_alloc_base(size, align, lmb_end_of_DRAM()); |
1da177e4 LT |
583 | |
584 | if (!ret) | |
585 | panic("numa.c: cannot allocate %lu bytes on node %d", | |
586 | size, nid); | |
587 | ||
588 | /* | |
589 | * If the memory came from a previously allocated node, we must | |
590 | * retry with the bootmem allocator. | |
591 | */ | |
45fb6cea AB |
592 | new_nid = early_pfn_to_nid(ret >> PAGE_SHIFT); |
593 | if (new_nid < nid) { | |
594 | ret = (unsigned long)__alloc_bootmem_node(NODE_DATA(new_nid), | |
1da177e4 LT |
595 | size, align, 0); |
596 | ||
597 | if (!ret) | |
598 | panic("numa.c: cannot allocate %lu bytes on node %d", | |
45fb6cea | 599 | size, new_nid); |
1da177e4 | 600 | |
45fb6cea | 601 | ret = __pa(ret); |
1da177e4 LT |
602 | |
603 | dbg("alloc_bootmem %lx %lx\n", ret, size); | |
604 | } | |
605 | ||
45fb6cea | 606 | return (void *)ret; |
1da177e4 LT |
607 | } |
608 | ||
74b85f37 CS |
609 | static struct notifier_block __cpuinitdata ppc64_numa_nb = { |
610 | .notifier_call = cpu_numa_callback, | |
611 | .priority = 1 /* Must run before sched domains notifier. */ | |
612 | }; | |
613 | ||
1da177e4 LT |
614 | void __init do_init_bootmem(void) |
615 | { | |
616 | int nid; | |
45fb6cea | 617 | unsigned int i; |
1da177e4 LT |
618 | |
619 | min_low_pfn = 0; | |
620 | max_low_pfn = lmb_end_of_DRAM() >> PAGE_SHIFT; | |
621 | max_pfn = max_low_pfn; | |
622 | ||
623 | if (parse_numa_properties()) | |
624 | setup_nonnuma(); | |
625 | else | |
4b703a23 | 626 | dump_numa_memory_topology(); |
1da177e4 LT |
627 | |
628 | register_cpu_notifier(&ppc64_numa_nb); | |
2b261227 NL |
629 | cpu_numa_callback(&ppc64_numa_nb, CPU_UP_PREPARE, |
630 | (void *)(unsigned long)boot_cpuid); | |
1da177e4 LT |
631 | |
632 | for_each_online_node(nid) { | |
45fb6cea | 633 | unsigned long start_pfn, end_pfn, pages_present; |
1da177e4 LT |
634 | unsigned long bootmem_paddr; |
635 | unsigned long bootmap_pages; | |
636 | ||
45fb6cea | 637 | get_region(nid, &start_pfn, &end_pfn, &pages_present); |
1da177e4 LT |
638 | |
639 | /* Allocate the node structure node local if possible */ | |
45fb6cea | 640 | NODE_DATA(nid) = careful_allocation(nid, |
1da177e4 | 641 | sizeof(struct pglist_data), |
45fb6cea AB |
642 | SMP_CACHE_BYTES, end_pfn); |
643 | NODE_DATA(nid) = __va(NODE_DATA(nid)); | |
1da177e4 LT |
644 | memset(NODE_DATA(nid), 0, sizeof(struct pglist_data)); |
645 | ||
646 | dbg("node %d\n", nid); | |
647 | dbg("NODE_DATA() = %p\n", NODE_DATA(nid)); | |
648 | ||
649 | NODE_DATA(nid)->bdata = &plat_node_bdata[nid]; | |
45fb6cea AB |
650 | NODE_DATA(nid)->node_start_pfn = start_pfn; |
651 | NODE_DATA(nid)->node_spanned_pages = end_pfn - start_pfn; | |
1da177e4 LT |
652 | |
653 | if (NODE_DATA(nid)->node_spanned_pages == 0) | |
654 | continue; | |
655 | ||
45fb6cea AB |
656 | dbg("start_paddr = %lx\n", start_pfn << PAGE_SHIFT); |
657 | dbg("end_paddr = %lx\n", end_pfn << PAGE_SHIFT); | |
1da177e4 | 658 | |
45fb6cea AB |
659 | bootmap_pages = bootmem_bootmap_pages(end_pfn - start_pfn); |
660 | bootmem_paddr = (unsigned long)careful_allocation(nid, | |
661 | bootmap_pages << PAGE_SHIFT, | |
662 | PAGE_SIZE, end_pfn); | |
663 | memset(__va(bootmem_paddr), 0, bootmap_pages << PAGE_SHIFT); | |
1da177e4 | 664 | |
1da177e4 LT |
665 | dbg("bootmap_paddr = %lx\n", bootmem_paddr); |
666 | ||
667 | init_bootmem_node(NODE_DATA(nid), bootmem_paddr >> PAGE_SHIFT, | |
45fb6cea | 668 | start_pfn, end_pfn); |
1da177e4 | 669 | |
45fb6cea AB |
670 | /* Add free regions on this node */ |
671 | for (i = 0; init_node_data[i].end_pfn; i++) { | |
672 | unsigned long start, end; | |
1da177e4 | 673 | |
45fb6cea | 674 | if (init_node_data[i].nid != nid) |
1da177e4 LT |
675 | continue; |
676 | ||
45fb6cea AB |
677 | start = init_node_data[i].start_pfn << PAGE_SHIFT; |
678 | end = init_node_data[i].end_pfn << PAGE_SHIFT; | |
1da177e4 | 679 | |
45fb6cea AB |
680 | dbg("free_bootmem %lx %lx\n", start, end - start); |
681 | free_bootmem_node(NODE_DATA(nid), start, end - start); | |
1da177e4 LT |
682 | } |
683 | ||
45fb6cea | 684 | /* Mark reserved regions on this node */ |
1da177e4 | 685 | for (i = 0; i < lmb.reserved.cnt; i++) { |
180379dc | 686 | unsigned long physbase = lmb.reserved.region[i].base; |
1da177e4 | 687 | unsigned long size = lmb.reserved.region[i].size; |
45fb6cea AB |
688 | unsigned long start_paddr = start_pfn << PAGE_SHIFT; |
689 | unsigned long end_paddr = end_pfn << PAGE_SHIFT; | |
1da177e4 | 690 | |
45fb6cea AB |
691 | if (early_pfn_to_nid(physbase >> PAGE_SHIFT) != nid && |
692 | early_pfn_to_nid((physbase+size-1) >> PAGE_SHIFT) != nid) | |
1da177e4 LT |
693 | continue; |
694 | ||
695 | if (physbase < end_paddr && | |
696 | (physbase+size) > start_paddr) { | |
697 | /* overlaps */ | |
698 | if (physbase < start_paddr) { | |
699 | size -= start_paddr - physbase; | |
700 | physbase = start_paddr; | |
701 | } | |
702 | ||
703 | if (size > end_paddr - physbase) | |
704 | size = end_paddr - physbase; | |
705 | ||
706 | dbg("reserve_bootmem %lx %lx\n", physbase, | |
707 | size); | |
708 | reserve_bootmem_node(NODE_DATA(nid), physbase, | |
709 | size); | |
710 | } | |
711 | } | |
802f192e | 712 | |
45fb6cea AB |
713 | /* Add regions into sparsemem */ |
714 | for (i = 0; init_node_data[i].end_pfn; i++) { | |
715 | unsigned long start, end; | |
716 | ||
717 | if (init_node_data[i].nid != nid) | |
802f192e BP |
718 | continue; |
719 | ||
45fb6cea AB |
720 | start = init_node_data[i].start_pfn; |
721 | end = init_node_data[i].end_pfn; | |
802f192e | 722 | |
45fb6cea | 723 | memory_present(nid, start, end); |
802f192e | 724 | } |
1da177e4 LT |
725 | } |
726 | } | |
727 | ||
728 | void __init paging_init(void) | |
729 | { | |
730 | unsigned long zones_size[MAX_NR_ZONES]; | |
731 | unsigned long zholes_size[MAX_NR_ZONES]; | |
732 | int nid; | |
733 | ||
734 | memset(zones_size, 0, sizeof(zones_size)); | |
735 | memset(zholes_size, 0, sizeof(zholes_size)); | |
736 | ||
737 | for_each_online_node(nid) { | |
45fb6cea | 738 | unsigned long start_pfn, end_pfn, pages_present; |
1da177e4 | 739 | |
45fb6cea | 740 | get_region(nid, &start_pfn, &end_pfn, &pages_present); |
1da177e4 LT |
741 | |
742 | zones_size[ZONE_DMA] = end_pfn - start_pfn; | |
45fb6cea | 743 | zholes_size[ZONE_DMA] = zones_size[ZONE_DMA] - pages_present; |
1da177e4 LT |
744 | |
745 | dbg("free_area_init node %d %lx %lx (hole: %lx)\n", nid, | |
746 | zones_size[ZONE_DMA], start_pfn, zholes_size[ZONE_DMA]); | |
747 | ||
45fb6cea AB |
748 | free_area_init_node(nid, NODE_DATA(nid), zones_size, start_pfn, |
749 | zholes_size); | |
1da177e4 LT |
750 | } |
751 | } | |
752 | ||
753 | static int __init early_numa(char *p) | |
754 | { | |
755 | if (!p) | |
756 | return 0; | |
757 | ||
758 | if (strstr(p, "off")) | |
759 | numa_enabled = 0; | |
760 | ||
761 | if (strstr(p, "debug")) | |
762 | numa_debug = 1; | |
763 | ||
764 | return 0; | |
765 | } | |
766 | early_param("numa", early_numa); | |
237a0989 MK |
767 | |
768 | #ifdef CONFIG_MEMORY_HOTPLUG | |
769 | /* | |
770 | * Find the node associated with a hot added memory section. Section | |
771 | * corresponds to a SPARSEMEM section, not an LMB. It is assumed that | |
772 | * sections are fully contained within a single LMB. | |
773 | */ | |
774 | int hot_add_scn_to_nid(unsigned long scn_addr) | |
775 | { | |
776 | struct device_node *memory = NULL; | |
b226e462 | 777 | nodemask_t nodes; |
482ec7c4 | 778 | int default_nid = any_online_node(NODE_MASK_ALL); |
069007ae | 779 | int nid; |
237a0989 MK |
780 | |
781 | if (!numa_enabled || (min_common_depth < 0)) | |
482ec7c4 | 782 | return default_nid; |
237a0989 MK |
783 | |
784 | while ((memory = of_find_node_by_type(memory, "memory")) != NULL) { | |
785 | unsigned long start, size; | |
b226e462 | 786 | int ranges; |
a7f67bdf | 787 | const unsigned int *memcell_buf; |
237a0989 MK |
788 | unsigned int len; |
789 | ||
a7f67bdf | 790 | memcell_buf = get_property(memory, "reg", &len); |
237a0989 MK |
791 | if (!memcell_buf || len <= 0) |
792 | continue; | |
793 | ||
cc5d0189 BH |
794 | /* ranges in cell */ |
795 | ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells); | |
237a0989 MK |
796 | ha_new_range: |
797 | start = read_n_cells(n_mem_addr_cells, &memcell_buf); | |
798 | size = read_n_cells(n_mem_size_cells, &memcell_buf); | |
953039c8 | 799 | nid = of_node_to_nid_single(memory); |
237a0989 MK |
800 | |
801 | /* Domains not present at boot default to 0 */ | |
482ec7c4 NL |
802 | if (nid < 0 || !node_online(nid)) |
803 | nid = default_nid; | |
237a0989 MK |
804 | |
805 | if ((scn_addr >= start) && (scn_addr < (start + size))) { | |
806 | of_node_put(memory); | |
cf950b7a | 807 | goto got_nid; |
237a0989 MK |
808 | } |
809 | ||
810 | if (--ranges) /* process all ranges in cell */ | |
811 | goto ha_new_range; | |
812 | } | |
237a0989 | 813 | BUG(); /* section address should be found above */ |
069007ae | 814 | return 0; |
b226e462 MK |
815 | |
816 | /* Temporary code to ensure that returned node is not empty */ | |
cf950b7a | 817 | got_nid: |
b226e462 | 818 | nodes_setall(nodes); |
cf950b7a NL |
819 | while (NODE_DATA(nid)->node_spanned_pages == 0) { |
820 | node_clear(nid, nodes); | |
821 | nid = any_online_node(nodes); | |
b226e462 | 822 | } |
cf950b7a | 823 | return nid; |
237a0989 MK |
824 | } |
825 | #endif /* CONFIG_MEMORY_HOTPLUG */ |