1
2
3
4
5
6
7
8
9
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>
20#include <asm/sparsemem.h>
21#include <asm/lmb.h>
22#include <asm/system.h>
23#include <asm/smp.h>
24
25static int numa_enabled = 1;
26
27static int numa_debug;
28#define dbg(args...) if (numa_debug) { printk(KERN_INFO args); }
29
30int numa_cpu_lookup_table[NR_CPUS];
31cpumask_t numa_cpumask_lookup_table[MAX_NUMNODES];
32struct pglist_data *node_data[MAX_NUMNODES];
33
34EXPORT_SYMBOL(numa_cpu_lookup_table);
35EXPORT_SYMBOL(numa_cpumask_lookup_table);
36EXPORT_SYMBOL(node_data);
37
38static bootmem_data_t __initdata plat_node_bdata[MAX_NUMNODES];
39static int min_common_depth;
40static int n_mem_addr_cells, n_mem_size_cells;
41
42static void __cpuinit map_cpu_to_node(int cpu, int node)
43{
44 numa_cpu_lookup_table[cpu] = node;
45
46 dbg("adding cpu %d to node %d\n", cpu, node);
47
48 if (!(cpu_isset(cpu, numa_cpumask_lookup_table[node])))
49 cpu_set(cpu, numa_cpumask_lookup_table[node]);
50}
51
52#ifdef CONFIG_HOTPLUG_CPU
53static void unmap_cpu_from_node(unsigned long cpu)
54{
55 int node = numa_cpu_lookup_table[cpu];
56
57 dbg("removing cpu %lu from node %d\n", cpu, node);
58
59 if (cpu_isset(cpu, numa_cpumask_lookup_table[node])) {
60 cpu_clear(cpu, numa_cpumask_lookup_table[node]);
61 } else {
62 printk(KERN_ERR "WARNING: cpu %lu not found in node %d\n",
63 cpu, node);
64 }
65}
66#endif
67
68static struct device_node * __cpuinit find_cpu_node(unsigned int cpu)
69{
70 unsigned int hw_cpuid = get_hard_smp_processor_id(cpu);
71 struct device_node *cpu_node = NULL;
72 const unsigned int *interrupt_server, *reg;
73 int len;
74
75 while ((cpu_node = of_find_node_by_type(cpu_node, "cpu")) != NULL) {
76
77 interrupt_server = of_get_property(cpu_node,
78 "ibm,ppc-interrupt-server#s", &len);
79
80 len = len / sizeof(u32);
81
82 if (interrupt_server && (len > 0)) {
83 while (len--) {
84 if (interrupt_server[len] == hw_cpuid)
85 return cpu_node;
86 }
87 } else {
88 reg = of_get_property(cpu_node, "reg", &len);
89 if (reg && (len > 0) && (reg[0] == hw_cpuid))
90 return cpu_node;
91 }
92 }
93
94 return NULL;
95}
96
97
98static const int *of_get_associativity(struct device_node *dev)
99{
100 return of_get_property(dev, "ibm,associativity", NULL);
101}
102
103
104
105
106static int of_node_to_nid_single(struct device_node *device)
107{
108 int nid = -1;
109 const unsigned int *tmp;
110
111 if (min_common_depth == -1)
112 goto out;
113
114 tmp = of_get_associativity(device);
115 if (!tmp)
116 goto out;
117
118 if (tmp[0] >= min_common_depth)
119 nid = tmp[min_common_depth];
120
121
122 if (nid == 0xffff || nid >= MAX_NUMNODES)
123 nid = -1;
124out:
125 return nid;
126}
127
128
129int of_node_to_nid(struct device_node *device)
130{
131 struct device_node *tmp;
132 int nid = -1;
133
134 of_node_get(device);
135 while (device) {
136 nid = of_node_to_nid_single(device);
137 if (nid != -1)
138 break;
139
140 tmp = device;
141 device = of_get_parent(tmp);
142 of_node_put(tmp);
143 }
144 of_node_put(device);
145
146 return nid;
147}
148EXPORT_SYMBOL_GPL(of_node_to_nid);
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164static int __init find_min_common_depth(void)
165{
166 int depth;
167 const unsigned int *ref_points;
168 struct device_node *rtas_root;
169 unsigned int len;
170
171 rtas_root = of_find_node_by_path("/rtas");
172
173 if (!rtas_root)
174 return -1;
175
176
177
178
179
180
181
182 ref_points = of_get_property(rtas_root,
183 "ibm,associativity-reference-points", &len);
184
185 if ((len >= 1) && ref_points) {
186 depth = ref_points[1];
187 } else {
188 dbg("NUMA: ibm,associativity-reference-points not found.\n");
189 depth = -1;
190 }
191 of_node_put(rtas_root);
192
193 return depth;
194}
195
196static void __init get_n_mem_cells(int *n_addr_cells, int *n_size_cells)
197{
198 struct device_node *memory = NULL;
199
200 memory = of_find_node_by_type(memory, "memory");
201 if (!memory)
202 panic("numa.c: No memory nodes found!");
203
204 *n_addr_cells = of_n_addr_cells(memory);
205 *n_size_cells = of_n_size_cells(memory);
206 of_node_put(memory);
207}
208
209static unsigned long __devinit read_n_cells(int n, const unsigned int **buf)
210{
211 unsigned long result = 0;
212
213 while (n--) {
214 result = (result << 32) | **buf;
215 (*buf)++;
216 }
217 return result;
218}
219
220
221
222
223
224static int __cpuinit numa_setup_cpu(unsigned long lcpu)
225{
226 int nid = 0;
227 struct device_node *cpu = find_cpu_node(lcpu);
228
229 if (!cpu) {
230 WARN_ON(1);
231 goto out;
232 }
233
234 nid = of_node_to_nid_single(cpu);
235
236 if (nid < 0 || !node_online(nid))
237 nid = any_online_node(NODE_MASK_ALL);
238out:
239 map_cpu_to_node(lcpu, nid);
240
241 of_node_put(cpu);
242
243 return nid;
244}
245
246static int __cpuinit cpu_numa_callback(struct notifier_block *nfb,
247 unsigned long action,
248 void *hcpu)
249{
250 unsigned long lcpu = (unsigned long)hcpu;
251 int ret = NOTIFY_DONE;
252
253 switch (action) {
254 case CPU_UP_PREPARE:
255 case CPU_UP_PREPARE_FROZEN:
256 numa_setup_cpu(lcpu);
257 ret = NOTIFY_OK;
258 break;
259#ifdef CONFIG_HOTPLUG_CPU
260 case CPU_DEAD:
261 case CPU_DEAD_FROZEN:
262 case CPU_UP_CANCELED:
263 case CPU_UP_CANCELED_FROZEN:
264 unmap_cpu_from_node(lcpu);
265 break;
266 ret = NOTIFY_OK;
267#endif
268 }
269 return ret;
270}
271
272
273
274
275
276
277
278
279
280static unsigned long __init numa_enforce_memory_limit(unsigned long start,
281 unsigned long size)
282{
283
284
285
286
287
288
289 if (! memory_limit)
290 return size;
291
292 if (start + size <= lmb_end_of_DRAM())
293 return size;
294
295 if (start >= lmb_end_of_DRAM())
296 return 0;
297
298 return lmb_end_of_DRAM() - start;
299}
300
301
302
303
304
305static void __init parse_drconf_memory(struct device_node *memory)
306{
307 const unsigned int *lm, *dm, *aa;
308 unsigned int ls, ld, la;
309 unsigned int n, aam, aalen;
310 unsigned long lmb_size, size, start;
311 int nid, default_nid = 0;
312 unsigned int ai, flags;
313
314 lm = of_get_property(memory, "ibm,lmb-size", &ls);
315 dm = of_get_property(memory, "ibm,dynamic-memory", &ld);
316 aa = of_get_property(memory, "ibm,associativity-lookup-arrays", &la);
317 if (!lm || !dm || !aa ||
318 ls < sizeof(unsigned int) || ld < sizeof(unsigned int) ||
319 la < 2 * sizeof(unsigned int))
320 return;
321
322 lmb_size = read_n_cells(n_mem_size_cells, &lm);
323 n = *dm++;
324 aam = *aa++;
325 aalen = *aa++;
326 if (ld < (n * (n_mem_addr_cells + 4) + 1) * sizeof(unsigned int) ||
327 la < (aam * aalen + 2) * sizeof(unsigned int))
328 return;
329
330 for (; n != 0; --n) {
331 start = read_n_cells(n_mem_addr_cells, &dm);
332 ai = dm[2];
333 flags = dm[3];
334 dm += 4;
335
336 if ((flags & 0x80) || !(flags & 0x8))
337 continue;
338 nid = default_nid;
339
340 if (min_common_depth > 0 && min_common_depth <= aalen &&
341 (flags & 0x40) == 0 && ai < aam) {
342
343 nid = aa[ai * aalen + min_common_depth - 1];
344 if (nid == 0xffff || nid >= MAX_NUMNODES)
345 nid = default_nid;
346 }
347 node_set_online(nid);
348
349 size = numa_enforce_memory_limit(start, lmb_size);
350 if (!size)
351 continue;
352
353 add_active_range(nid, start >> PAGE_SHIFT,
354 (start >> PAGE_SHIFT) + (size >> PAGE_SHIFT));
355 }
356}
357
358static int __init parse_numa_properties(void)
359{
360 struct device_node *cpu = NULL;
361 struct device_node *memory = NULL;
362 int default_nid = 0;
363 unsigned long i;
364
365 if (numa_enabled == 0) {
366 printk(KERN_WARNING "NUMA disabled by user\n");
367 return -1;
368 }
369
370 min_common_depth = find_min_common_depth();
371
372 if (min_common_depth < 0)
373 return min_common_depth;
374
375 dbg("NUMA associativity depth for CPU/Memory: %d\n", min_common_depth);
376
377
378
379
380
381
382 for_each_present_cpu(i) {
383 int nid;
384
385 cpu = find_cpu_node(i);
386 BUG_ON(!cpu);
387 nid = of_node_to_nid_single(cpu);
388 of_node_put(cpu);
389
390
391
392
393
394
395 if (nid < 0)
396 continue;
397 node_set_online(nid);
398 }
399
400 get_n_mem_cells(&n_mem_addr_cells, &n_mem_size_cells);
401 memory = NULL;
402 while ((memory = of_find_node_by_type(memory, "memory")) != NULL) {
403 unsigned long start;
404 unsigned long size;
405 int nid;
406 int ranges;
407 const unsigned int *memcell_buf;
408 unsigned int len;
409
410 memcell_buf = of_get_property(memory,
411 "linux,usable-memory", &len);
412 if (!memcell_buf || len <= 0)
413 memcell_buf = of_get_property(memory, "reg", &len);
414 if (!memcell_buf || len <= 0)
415 continue;
416
417
418 ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells);
419new_range:
420
421 start = read_n_cells(n_mem_addr_cells, &memcell_buf);
422 size = read_n_cells(n_mem_size_cells, &memcell_buf);
423
424
425
426
427
428
429 nid = of_node_to_nid_single(memory);
430 if (nid < 0)
431 nid = default_nid;
432 node_set_online(nid);
433
434 if (!(size = numa_enforce_memory_limit(start, size))) {
435 if (--ranges)
436 goto new_range;
437 else
438 continue;
439 }
440
441 add_active_range(nid, start >> PAGE_SHIFT,
442 (start >> PAGE_SHIFT) + (size >> PAGE_SHIFT));
443
444 if (--ranges)
445 goto new_range;
446 }
447
448
449
450
451
452 memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory");
453 if (memory)
454 parse_drconf_memory(memory);
455
456 return 0;
457}
458
459static void __init setup_nonnuma(void)
460{
461 unsigned long top_of_ram = lmb_end_of_DRAM();
462 unsigned long total_ram = lmb_phys_mem_size();
463 unsigned long start_pfn, end_pfn;
464 unsigned int i;
465
466 printk(KERN_DEBUG "Top of RAM: 0x%lx, Total RAM: 0x%lx\n",
467 top_of_ram, total_ram);
468 printk(KERN_DEBUG "Memory hole size: %ldMB\n",
469 (top_of_ram - total_ram) >> 20);
470
471 for (i = 0; i < lmb.memory.cnt; ++i) {
472 start_pfn = lmb.memory.region[i].base >> PAGE_SHIFT;
473 end_pfn = start_pfn + lmb_size_pages(&lmb.memory, i);
474 add_active_range(0, start_pfn, end_pfn);
475 }
476 node_set_online(0);
477}
478
479void __init dump_numa_cpu_topology(void)
480{
481 unsigned int node;
482 unsigned int cpu, count;
483
484 if (min_common_depth == -1 || !numa_enabled)
485 return;
486
487 for_each_online_node(node) {
488 printk(KERN_DEBUG "Node %d CPUs:", node);
489
490 count = 0;
491
492
493
494
495 for (cpu = 0; cpu < NR_CPUS; cpu++) {
496 if (cpu_isset(cpu, numa_cpumask_lookup_table[node])) {
497 if (count == 0)
498 printk(" %u", cpu);
499 ++count;
500 } else {
501 if (count > 1)
502 printk("-%u", cpu - 1);
503 count = 0;
504 }
505 }
506
507 if (count > 1)
508 printk("-%u", NR_CPUS - 1);
509 printk("\n");
510 }
511}
512
513static void __init dump_numa_memory_topology(void)
514{
515 unsigned int node;
516 unsigned int count;
517
518 if (min_common_depth == -1 || !numa_enabled)
519 return;
520
521 for_each_online_node(node) {
522 unsigned long i;
523
524 printk(KERN_DEBUG "Node %d Memory:", node);
525
526 count = 0;
527
528 for (i = 0; i < lmb_end_of_DRAM();
529 i += (1 << SECTION_SIZE_BITS)) {
530 if (early_pfn_to_nid(i >> PAGE_SHIFT) == node) {
531 if (count == 0)
532 printk(" 0x%lx", i);
533 ++count;
534 } else {
535 if (count > 0)
536 printk("-0x%lx", i);
537 count = 0;
538 }
539 }
540
541 if (count > 0)
542 printk("-0x%lx", i);
543 printk("\n");
544 }
545}
546
547
548
549
550
551
552
553
554static void __init *careful_allocation(int nid, unsigned long size,
555 unsigned long align,
556 unsigned long end_pfn)
557{
558 int new_nid;
559 unsigned long ret = __lmb_alloc_base(size, align, end_pfn << PAGE_SHIFT);
560
561
562 if (!ret)
563 ret = __lmb_alloc_base(size, align, lmb_end_of_DRAM());
564
565 if (!ret)
566 panic("numa.c: cannot allocate %lu bytes on node %d",
567 size, nid);
568
569
570
571
572
573 new_nid = early_pfn_to_nid(ret >> PAGE_SHIFT);
574 if (new_nid < nid) {
575 ret = (unsigned long)__alloc_bootmem_node(NODE_DATA(new_nid),
576 size, align, 0);
577
578 if (!ret)
579 panic("numa.c: cannot allocate %lu bytes on node %d",
580 size, new_nid);
581
582 ret = __pa(ret);
583
584 dbg("alloc_bootmem %lx %lx\n", ret, size);
585 }
586
587 return (void *)ret;
588}
589
590static struct notifier_block __cpuinitdata ppc64_numa_nb = {
591 .notifier_call = cpu_numa_callback,
592 .priority = 1
593};
594
595void __init do_init_bootmem(void)
596{
597 int nid;
598 unsigned int i;
599
600 min_low_pfn = 0;
601 max_low_pfn = lmb_end_of_DRAM() >> PAGE_SHIFT;
602 max_pfn = max_low_pfn;
603
604 if (parse_numa_properties())
605 setup_nonnuma();
606 else
607 dump_numa_memory_topology();
608
609 register_cpu_notifier(&ppc64_numa_nb);
610 cpu_numa_callback(&ppc64_numa_nb, CPU_UP_PREPARE,
611 (void *)(unsigned long)boot_cpuid);
612
613 for_each_online_node(nid) {
614 unsigned long start_pfn, end_pfn;
615 unsigned long bootmem_paddr;
616 unsigned long bootmap_pages;
617
618 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
619
620
621 NODE_DATA(nid) = careful_allocation(nid,
622 sizeof(struct pglist_data),
623 SMP_CACHE_BYTES, end_pfn);
624 NODE_DATA(nid) = __va(NODE_DATA(nid));
625 memset(NODE_DATA(nid), 0, sizeof(struct pglist_data));
626
627 dbg("node %d\n", nid);
628 dbg("NODE_DATA() = %p\n", NODE_DATA(nid));
629
630 NODE_DATA(nid)->bdata = &plat_node_bdata[nid];
631 NODE_DATA(nid)->node_start_pfn = start_pfn;
632 NODE_DATA(nid)->node_spanned_pages = end_pfn - start_pfn;
633
634 if (NODE_DATA(nid)->node_spanned_pages == 0)
635 continue;
636
637 dbg("start_paddr = %lx\n", start_pfn << PAGE_SHIFT);
638 dbg("end_paddr = %lx\n", end_pfn << PAGE_SHIFT);
639
640 bootmap_pages = bootmem_bootmap_pages(end_pfn - start_pfn);
641 bootmem_paddr = (unsigned long)careful_allocation(nid,
642 bootmap_pages << PAGE_SHIFT,
643 PAGE_SIZE, end_pfn);
644 memset(__va(bootmem_paddr), 0, bootmap_pages << PAGE_SHIFT);
645
646 dbg("bootmap_paddr = %lx\n", bootmem_paddr);
647
648 init_bootmem_node(NODE_DATA(nid), bootmem_paddr >> PAGE_SHIFT,
649 start_pfn, end_pfn);
650
651 free_bootmem_with_active_regions(nid, end_pfn);
652
653
654 for (i = 0; i < lmb.reserved.cnt; i++) {
655 unsigned long physbase = lmb.reserved.region[i].base;
656 unsigned long size = lmb.reserved.region[i].size;
657 unsigned long start_paddr = start_pfn << PAGE_SHIFT;
658 unsigned long end_paddr = end_pfn << PAGE_SHIFT;
659
660 if (early_pfn_to_nid(physbase >> PAGE_SHIFT) != nid &&
661 early_pfn_to_nid((physbase+size-1) >> PAGE_SHIFT) != nid)
662 continue;
663
664 if (physbase < end_paddr &&
665 (physbase+size) > start_paddr) {
666
667 if (physbase < start_paddr) {
668 size -= start_paddr - physbase;
669 physbase = start_paddr;
670 }
671
672 if (size > end_paddr - physbase)
673 size = end_paddr - physbase;
674
675 dbg("reserve_bootmem %lx %lx\n", physbase,
676 size);
677 reserve_bootmem_node(NODE_DATA(nid), physbase,
678 size);
679 }
680 }
681
682 sparse_memory_present_with_active_regions(nid);
683 }
684}
685
686void __init paging_init(void)
687{
688 unsigned long max_zone_pfns[MAX_NR_ZONES];
689 memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
690 max_zone_pfns[ZONE_DMA] = lmb_end_of_DRAM() >> PAGE_SHIFT;
691 free_area_init_nodes(max_zone_pfns);
692}
693
694static int __init early_numa(char *p)
695{
696 if (!p)
697 return 0;
698
699 if (strstr(p, "off"))
700 numa_enabled = 0;
701
702 if (strstr(p, "debug"))
703 numa_debug = 1;
704
705 return 0;
706}
707early_param("numa", early_numa);
708
709#ifdef CONFIG_MEMORY_HOTPLUG
710
711
712
713
714
715int hot_add_scn_to_nid(unsigned long scn_addr)
716{
717 struct device_node *memory = NULL;
718 nodemask_t nodes;
719 int default_nid = any_online_node(NODE_MASK_ALL);
720 int nid;
721
722 if (!numa_enabled || (min_common_depth < 0))
723 return default_nid;
724
725 while ((memory = of_find_node_by_type(memory, "memory")) != NULL) {
726 unsigned long start, size;
727 int ranges;
728 const unsigned int *memcell_buf;
729 unsigned int len;
730
731 memcell_buf = of_get_property(memory, "reg", &len);
732 if (!memcell_buf || len <= 0)
733 continue;
734
735
736 ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells);
737ha_new_range:
738 start = read_n_cells(n_mem_addr_cells, &memcell_buf);
739 size = read_n_cells(n_mem_size_cells, &memcell_buf);
740 nid = of_node_to_nid_single(memory);
741
742
743 if (nid < 0 || !node_online(nid))
744 nid = default_nid;
745
746 if ((scn_addr >= start) && (scn_addr < (start + size))) {
747 of_node_put(memory);
748 goto got_nid;
749 }
750
751 if (--ranges)
752 goto ha_new_range;
753 }
754 BUG();
755 return 0;
756
757
758got_nid:
759 nodes_setall(nodes);
760 while (NODE_DATA(nid)->node_spanned_pages == 0) {
761 node_clear(nid, nodes);
762 nid = any_online_node(nodes);
763 }
764 return nid;
765}
766#endif
767