1
2
3
4
5
6
7#define pr_fmt(fmt) "numa: " fmt
8
9#include <linux/threads.h>
10#include <linux/memblock.h>
11#include <linux/init.h>
12#include <linux/mm.h>
13#include <linux/mmzone.h>
14#include <linux/export.h>
15#include <linux/nodemask.h>
16#include <linux/cpu.h>
17#include <linux/notifier.h>
18#include <linux/of.h>
19#include <linux/pfn.h>
20#include <linux/cpuset.h>
21#include <linux/node.h>
22#include <linux/stop_machine.h>
23#include <linux/proc_fs.h>
24#include <linux/seq_file.h>
25#include <linux/uaccess.h>
26#include <linux/slab.h>
27#include <asm/cputhreads.h>
28#include <asm/sparsemem.h>
29#include <asm/prom.h>
30#include <asm/smp.h>
31#include <asm/topology.h>
32#include <asm/firmware.h>
33#include <asm/paca.h>
34#include <asm/hvcall.h>
35#include <asm/setup.h>
36#include <asm/vdso.h>
37#include <asm/drmem.h>
38
39static int numa_enabled = 1;
40
41static char *cmdline __initdata;
42
43static int numa_debug;
44#define dbg(args...) if (numa_debug) { printk(KERN_INFO args); }
45
46int numa_cpu_lookup_table[NR_CPUS];
47cpumask_var_t node_to_cpumask_map[MAX_NUMNODES];
48struct pglist_data *node_data[MAX_NUMNODES];
49
50EXPORT_SYMBOL(numa_cpu_lookup_table);
51EXPORT_SYMBOL(node_to_cpumask_map);
52EXPORT_SYMBOL(node_data);
53
54static int min_common_depth;
55static int n_mem_addr_cells, n_mem_size_cells;
56static int form1_affinity;
57
58#define MAX_DISTANCE_REF_POINTS 4
59static int distance_ref_points_depth;
60static const __be32 *distance_ref_points;
61static int distance_lookup_table[MAX_NUMNODES][MAX_DISTANCE_REF_POINTS];
62
63
64
65
66
67
68
69static void __init setup_node_to_cpumask_map(void)
70{
71 unsigned int node;
72
73
74 if (nr_node_ids == MAX_NUMNODES)
75 setup_nr_node_ids();
76
77
78 for_each_node(node)
79 alloc_bootmem_cpumask_var(&node_to_cpumask_map[node]);
80
81
82 dbg("Node to cpumask map for %u nodes\n", nr_node_ids);
83}
84
85static int __init fake_numa_create_new_node(unsigned long end_pfn,
86 unsigned int *nid)
87{
88 unsigned long long mem;
89 char *p = cmdline;
90 static unsigned int fake_nid;
91 static unsigned long long curr_boundary;
92
93
94
95
96
97 if (fake_nid)
98 *nid = fake_nid;
99
100
101
102
103
104 if (!p)
105 return 0;
106
107 mem = memparse(p, &p);
108 if (!mem)
109 return 0;
110
111 if (mem < curr_boundary)
112 return 0;
113
114 curr_boundary = mem;
115
116 if ((end_pfn << PAGE_SHIFT) > mem) {
117
118
119
120 while (*p == ',' || *p == ' ' || *p == '\t')
121 p++;
122
123 cmdline = p;
124 fake_nid++;
125 *nid = fake_nid;
126 dbg("created new fake_node with id %d\n", fake_nid);
127 return 1;
128 }
129 return 0;
130}
131
132static void reset_numa_cpu_lookup_table(void)
133{
134 unsigned int cpu;
135
136 for_each_possible_cpu(cpu)
137 numa_cpu_lookup_table[cpu] = -1;
138}
139
140static void map_cpu_to_node(int cpu, int node)
141{
142 update_numa_cpu_lookup_table(cpu, node);
143
144 dbg("adding cpu %d to node %d\n", cpu, node);
145
146 if (!(cpumask_test_cpu(cpu, node_to_cpumask_map[node])))
147 cpumask_set_cpu(cpu, node_to_cpumask_map[node]);
148}
149
150#if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_PPC_SPLPAR)
151static void unmap_cpu_from_node(unsigned long cpu)
152{
153 int node = numa_cpu_lookup_table[cpu];
154
155 dbg("removing cpu %lu from node %d\n", cpu, node);
156
157 if (cpumask_test_cpu(cpu, node_to_cpumask_map[node])) {
158 cpumask_clear_cpu(cpu, node_to_cpumask_map[node]);
159 } else {
160 printk(KERN_ERR "WARNING: cpu %lu not found in node %d\n",
161 cpu, node);
162 }
163}
164#endif
165
166int cpu_distance(__be32 *cpu1_assoc, __be32 *cpu2_assoc)
167{
168 int dist = 0;
169
170 int i, index;
171
172 for (i = 0; i < distance_ref_points_depth; i++) {
173 index = be32_to_cpu(distance_ref_points[i]);
174 if (cpu1_assoc[index] == cpu2_assoc[index])
175 break;
176 dist++;
177 }
178
179 return dist;
180}
181
182
183static const __be32 *of_get_associativity(struct device_node *dev)
184{
185 return of_get_property(dev, "ibm,associativity", NULL);
186}
187
188int __node_distance(int a, int b)
189{
190 int i;
191 int distance = LOCAL_DISTANCE;
192
193 if (!form1_affinity)
194 return ((a == b) ? LOCAL_DISTANCE : REMOTE_DISTANCE);
195
196 for (i = 0; i < distance_ref_points_depth; i++) {
197 if (distance_lookup_table[a][i] == distance_lookup_table[b][i])
198 break;
199
200
201 distance *= 2;
202 }
203
204 return distance;
205}
206EXPORT_SYMBOL(__node_distance);
207
208static void initialize_distance_lookup_table(int nid,
209 const __be32 *associativity)
210{
211 int i;
212
213 if (!form1_affinity)
214 return;
215
216 for (i = 0; i < distance_ref_points_depth; i++) {
217 const __be32 *entry;
218
219 entry = &associativity[be32_to_cpu(distance_ref_points[i]) - 1];
220 distance_lookup_table[nid][i] = of_read_number(entry, 1);
221 }
222}
223
224
225
226
227
228static int associativity_to_nid(const __be32 *associativity)
229{
230 int nid = NUMA_NO_NODE;
231
232 if (!numa_enabled)
233 goto out;
234
235 if (of_read_number(associativity, 1) >= min_common_depth)
236 nid = of_read_number(&associativity[min_common_depth], 1);
237
238
239 if (nid == 0xffff || nid >= nr_node_ids)
240 nid = NUMA_NO_NODE;
241
242 if (nid > 0 &&
243 of_read_number(associativity, 1) >= distance_ref_points_depth) {
244
245
246
247 initialize_distance_lookup_table(nid, associativity + 1);
248 }
249
250out:
251 return nid;
252}
253
254
255
256
257static int of_node_to_nid_single(struct device_node *device)
258{
259 int nid = NUMA_NO_NODE;
260 const __be32 *tmp;
261
262 tmp = of_get_associativity(device);
263 if (tmp)
264 nid = associativity_to_nid(tmp);
265 return nid;
266}
267
268
269int of_node_to_nid(struct device_node *device)
270{
271 int nid = NUMA_NO_NODE;
272
273 of_node_get(device);
274 while (device) {
275 nid = of_node_to_nid_single(device);
276 if (nid != -1)
277 break;
278
279 device = of_get_next_parent(device);
280 }
281 of_node_put(device);
282
283 return nid;
284}
285EXPORT_SYMBOL(of_node_to_nid);
286
287static int __init find_min_common_depth(void)
288{
289 int depth;
290 struct device_node *root;
291
292 if (firmware_has_feature(FW_FEATURE_OPAL))
293 root = of_find_node_by_path("/ibm,opal");
294 else
295 root = of_find_node_by_path("/rtas");
296 if (!root)
297 root = of_find_node_by_path("/");
298
299
300
301
302
303
304
305
306
307
308
309
310
311 distance_ref_points = of_get_property(root,
312 "ibm,associativity-reference-points",
313 &distance_ref_points_depth);
314
315 if (!distance_ref_points) {
316 dbg("NUMA: ibm,associativity-reference-points not found.\n");
317 goto err;
318 }
319
320 distance_ref_points_depth /= sizeof(int);
321
322 if (firmware_has_feature(FW_FEATURE_OPAL) ||
323 firmware_has_feature(FW_FEATURE_TYPE1_AFFINITY)) {
324 dbg("Using form 1 affinity\n");
325 form1_affinity = 1;
326 }
327
328 if (form1_affinity) {
329 depth = of_read_number(distance_ref_points, 1);
330 } else {
331 if (distance_ref_points_depth < 2) {
332 printk(KERN_WARNING "NUMA: "
333 "short ibm,associativity-reference-points\n");
334 goto err;
335 }
336
337 depth = of_read_number(&distance_ref_points[1], 1);
338 }
339
340
341
342
343
344 if (distance_ref_points_depth > MAX_DISTANCE_REF_POINTS) {
345 printk(KERN_WARNING "NUMA: distance array capped at "
346 "%d entries\n", MAX_DISTANCE_REF_POINTS);
347 distance_ref_points_depth = MAX_DISTANCE_REF_POINTS;
348 }
349
350 of_node_put(root);
351 return depth;
352
353err:
354 of_node_put(root);
355 return -1;
356}
357
358static void __init get_n_mem_cells(int *n_addr_cells, int *n_size_cells)
359{
360 struct device_node *memory = NULL;
361
362 memory = of_find_node_by_type(memory, "memory");
363 if (!memory)
364 panic("numa.c: No memory nodes found!");
365
366 *n_addr_cells = of_n_addr_cells(memory);
367 *n_size_cells = of_n_size_cells(memory);
368 of_node_put(memory);
369}
370
371static unsigned long read_n_cells(int n, const __be32 **buf)
372{
373 unsigned long result = 0;
374
375 while (n--) {
376 result = (result << 32) | of_read_number(*buf, 1);
377 (*buf)++;
378 }
379 return result;
380}
381
382struct assoc_arrays {
383 u32 n_arrays;
384 u32 array_sz;
385 const __be32 *arrays;
386};
387
388
389
390
391
392
393
394
395
396
397
398static int of_get_assoc_arrays(struct assoc_arrays *aa)
399{
400 struct device_node *memory;
401 const __be32 *prop;
402 u32 len;
403
404 memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory");
405 if (!memory)
406 return -1;
407
408 prop = of_get_property(memory, "ibm,associativity-lookup-arrays", &len);
409 if (!prop || len < 2 * sizeof(unsigned int)) {
410 of_node_put(memory);
411 return -1;
412 }
413
414 aa->n_arrays = of_read_number(prop++, 1);
415 aa->array_sz = of_read_number(prop++, 1);
416
417 of_node_put(memory);
418
419
420
421
422 if (len < (aa->n_arrays * aa->array_sz + 2) * sizeof(unsigned int))
423 return -1;
424
425 aa->arrays = prop;
426 return 0;
427}
428
429
430
431
432
433int of_drconf_to_nid_single(struct drmem_lmb *lmb)
434{
435 struct assoc_arrays aa = { .arrays = NULL };
436 int default_nid = NUMA_NO_NODE;
437 int nid = default_nid;
438 int rc, index;
439
440 if ((min_common_depth < 0) || !numa_enabled)
441 return default_nid;
442
443 rc = of_get_assoc_arrays(&aa);
444 if (rc)
445 return default_nid;
446
447 if (min_common_depth <= aa.array_sz &&
448 !(lmb->flags & DRCONF_MEM_AI_INVALID) && lmb->aa_index < aa.n_arrays) {
449 index = lmb->aa_index * aa.array_sz + min_common_depth - 1;
450 nid = of_read_number(&aa.arrays[index], 1);
451
452 if (nid == 0xffff || nid >= nr_node_ids)
453 nid = default_nid;
454
455 if (nid > 0) {
456 index = lmb->aa_index * aa.array_sz;
457 initialize_distance_lookup_table(nid,
458 &aa.arrays[index]);
459 }
460 }
461
462 return nid;
463}
464
465#ifdef CONFIG_PPC_SPLPAR
466static int vphn_get_nid(long lcpu)
467{
468 __be32 associativity[VPHN_ASSOC_BUFSIZE] = {0};
469 long rc, hwid;
470
471
472
473
474
475
476
477
478
479
480 if (firmware_has_feature(FW_FEATURE_VPHN)) {
481 if (cpu_to_phys_id)
482 hwid = cpu_to_phys_id[lcpu];
483 else
484 hwid = get_hard_smp_processor_id(lcpu);
485
486 rc = hcall_vphn(hwid, VPHN_FLAG_VCPU, associativity);
487 if (rc == H_SUCCESS)
488 return associativity_to_nid(associativity);
489 }
490
491 return NUMA_NO_NODE;
492}
493#else
494static int vphn_get_nid(long unused)
495{
496 return NUMA_NO_NODE;
497}
498#endif
499
500
501
502
503
504static int numa_setup_cpu(unsigned long lcpu)
505{
506 struct device_node *cpu;
507 int fcpu = cpu_first_thread_sibling(lcpu);
508 int nid = NUMA_NO_NODE;
509
510 if (!cpu_present(lcpu)) {
511 set_cpu_numa_node(lcpu, first_online_node);
512 return first_online_node;
513 }
514
515
516
517
518
519
520
521
522
523 nid = numa_cpu_lookup_table[fcpu];
524 if (nid >= 0) {
525 map_cpu_to_node(lcpu, nid);
526 return nid;
527 }
528
529 nid = vphn_get_nid(lcpu);
530 if (nid != NUMA_NO_NODE)
531 goto out_present;
532
533 cpu = of_get_cpu_node(lcpu, NULL);
534
535 if (!cpu) {
536 WARN_ON(1);
537 if (cpu_present(lcpu))
538 goto out_present;
539 else
540 goto out;
541 }
542
543 nid = of_node_to_nid_single(cpu);
544 of_node_put(cpu);
545
546out_present:
547 if (nid < 0 || !node_possible(nid))
548 nid = first_online_node;
549
550
551
552
553
554
555
556
557
558 if (fcpu != lcpu) {
559 WARN_ON(cpu_online(fcpu));
560 map_cpu_to_node(fcpu, nid);
561 }
562
563 map_cpu_to_node(lcpu, nid);
564out:
565 return nid;
566}
567
568static void verify_cpu_node_mapping(int cpu, int node)
569{
570 int base, sibling, i;
571
572
573 base = cpu_first_thread_sibling(cpu);
574
575 for (i = 0; i < threads_per_core; i++) {
576 sibling = base + i;
577
578 if (sibling == cpu || cpu_is_offline(sibling))
579 continue;
580
581 if (cpu_to_node(sibling) != node) {
582 WARN(1, "CPU thread siblings %d and %d don't belong"
583 " to the same node!\n", cpu, sibling);
584 break;
585 }
586 }
587}
588
589
590static int ppc_numa_cpu_prepare(unsigned int cpu)
591{
592 int nid;
593
594 nid = numa_setup_cpu(cpu);
595 verify_cpu_node_mapping(cpu, nid);
596 return 0;
597}
598
599static int ppc_numa_cpu_dead(unsigned int cpu)
600{
601#ifdef CONFIG_HOTPLUG_CPU
602 unmap_cpu_from_node(cpu);
603#endif
604 return 0;
605}
606
607
608
609
610
611
612
613
614
615static unsigned long __init numa_enforce_memory_limit(unsigned long start,
616 unsigned long size)
617{
618
619
620
621
622
623
624
625 if (start + size <= memblock_end_of_DRAM())
626 return size;
627
628 if (start >= memblock_end_of_DRAM())
629 return 0;
630
631 return memblock_end_of_DRAM() - start;
632}
633
634
635
636
637
638static inline int __init read_usm_ranges(const __be32 **usm)
639{
640
641
642
643
644
645
646 return read_n_cells(n_mem_size_cells, usm);
647}
648
649
650
651
652
653static int __init numa_setup_drmem_lmb(struct drmem_lmb *lmb,
654 const __be32 **usm,
655 void *data)
656{
657 unsigned int ranges, is_kexec_kdump = 0;
658 unsigned long base, size, sz;
659 int nid;
660
661
662
663
664
665 if ((lmb->flags & DRCONF_MEM_RESERVED)
666 || !(lmb->flags & DRCONF_MEM_ASSIGNED))
667 return 0;
668
669 if (*usm)
670 is_kexec_kdump = 1;
671
672 base = lmb->base_addr;
673 size = drmem_lmb_size();
674 ranges = 1;
675
676 if (is_kexec_kdump) {
677 ranges = read_usm_ranges(usm);
678 if (!ranges)
679 return 0;
680 }
681
682 do {
683 if (is_kexec_kdump) {
684 base = read_n_cells(n_mem_addr_cells, usm);
685 size = read_n_cells(n_mem_size_cells, usm);
686 }
687
688 nid = of_drconf_to_nid_single(lmb);
689 fake_numa_create_new_node(((base + size) >> PAGE_SHIFT),
690 &nid);
691 node_set_online(nid);
692 sz = numa_enforce_memory_limit(base, size);
693 if (sz)
694 memblock_set_node(base, sz, &memblock.memory, nid);
695 } while (--ranges);
696
697 return 0;
698}
699
700static int __init parse_numa_properties(void)
701{
702 struct device_node *memory;
703 int default_nid = 0;
704 unsigned long i;
705
706 if (numa_enabled == 0) {
707 printk(KERN_WARNING "NUMA disabled by user\n");
708 return -1;
709 }
710
711 min_common_depth = find_min_common_depth();
712
713 if (min_common_depth < 0) {
714
715
716
717
718 numa_enabled = false;
719 return min_common_depth;
720 }
721
722 dbg("NUMA associativity depth for CPU/Memory: %d\n", min_common_depth);
723
724
725
726
727
728
729 for_each_present_cpu(i) {
730 struct device_node *cpu;
731 int nid = vphn_get_nid(i);
732
733
734
735
736
737
738 if (nid == NUMA_NO_NODE) {
739 cpu = of_get_cpu_node(i, NULL);
740 BUG_ON(!cpu);
741 nid = of_node_to_nid_single(cpu);
742 of_node_put(cpu);
743 }
744
745 node_set_online(nid);
746 }
747
748 get_n_mem_cells(&n_mem_addr_cells, &n_mem_size_cells);
749
750 for_each_node_by_type(memory, "memory") {
751 unsigned long start;
752 unsigned long size;
753 int nid;
754 int ranges;
755 const __be32 *memcell_buf;
756 unsigned int len;
757
758 memcell_buf = of_get_property(memory,
759 "linux,usable-memory", &len);
760 if (!memcell_buf || len <= 0)
761 memcell_buf = of_get_property(memory, "reg", &len);
762 if (!memcell_buf || len <= 0)
763 continue;
764
765
766 ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells);
767new_range:
768
769 start = read_n_cells(n_mem_addr_cells, &memcell_buf);
770 size = read_n_cells(n_mem_size_cells, &memcell_buf);
771
772
773
774
775
776
777 nid = of_node_to_nid_single(memory);
778 if (nid < 0)
779 nid = default_nid;
780
781 fake_numa_create_new_node(((start + size) >> PAGE_SHIFT), &nid);
782 node_set_online(nid);
783
784 size = numa_enforce_memory_limit(start, size);
785 if (size)
786 memblock_set_node(start, size, &memblock.memory, nid);
787
788 if (--ranges)
789 goto new_range;
790 }
791
792
793
794
795
796
797 memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory");
798 if (memory) {
799 walk_drmem_lmbs(memory, NULL, numa_setup_drmem_lmb);
800 of_node_put(memory);
801 }
802
803 return 0;
804}
805
806static void __init setup_nonnuma(void)
807{
808 unsigned long top_of_ram = memblock_end_of_DRAM();
809 unsigned long total_ram = memblock_phys_mem_size();
810 unsigned long start_pfn, end_pfn;
811 unsigned int nid = 0;
812 int i;
813
814 printk(KERN_DEBUG "Top of RAM: 0x%lx, Total RAM: 0x%lx\n",
815 top_of_ram, total_ram);
816 printk(KERN_DEBUG "Memory hole size: %ldMB\n",
817 (top_of_ram - total_ram) >> 20);
818
819 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, NULL) {
820 fake_numa_create_new_node(end_pfn, &nid);
821 memblock_set_node(PFN_PHYS(start_pfn),
822 PFN_PHYS(end_pfn - start_pfn),
823 &memblock.memory, nid);
824 node_set_online(nid);
825 }
826}
827
828void __init dump_numa_cpu_topology(void)
829{
830 unsigned int node;
831 unsigned int cpu, count;
832
833 if (!numa_enabled)
834 return;
835
836 for_each_online_node(node) {
837 pr_info("Node %d CPUs:", node);
838
839 count = 0;
840
841
842
843
844 for (cpu = 0; cpu < nr_cpu_ids; cpu++) {
845 if (cpumask_test_cpu(cpu,
846 node_to_cpumask_map[node])) {
847 if (count == 0)
848 pr_cont(" %u", cpu);
849 ++count;
850 } else {
851 if (count > 1)
852 pr_cont("-%u", cpu - 1);
853 count = 0;
854 }
855 }
856
857 if (count > 1)
858 pr_cont("-%u", nr_cpu_ids - 1);
859 pr_cont("\n");
860 }
861}
862
863
864static void __init setup_node_data(int nid, u64 start_pfn, u64 end_pfn)
865{
866 u64 spanned_pages = end_pfn - start_pfn;
867 const size_t nd_size = roundup(sizeof(pg_data_t), SMP_CACHE_BYTES);
868 u64 nd_pa;
869 void *nd;
870 int tnid;
871
872 nd_pa = memblock_phys_alloc_try_nid(nd_size, SMP_CACHE_BYTES, nid);
873 if (!nd_pa)
874 panic("Cannot allocate %zu bytes for node %d data\n",
875 nd_size, nid);
876
877 nd = __va(nd_pa);
878
879
880 pr_info(" NODE_DATA [mem %#010Lx-%#010Lx]\n",
881 nd_pa, nd_pa + nd_size - 1);
882 tnid = early_pfn_to_nid(nd_pa >> PAGE_SHIFT);
883 if (tnid != nid)
884 pr_info(" NODE_DATA(%d) on node %d\n", nid, tnid);
885
886 node_data[nid] = nd;
887 memset(NODE_DATA(nid), 0, sizeof(pg_data_t));
888 NODE_DATA(nid)->node_id = nid;
889 NODE_DATA(nid)->node_start_pfn = start_pfn;
890 NODE_DATA(nid)->node_spanned_pages = spanned_pages;
891}
892
893static void __init find_possible_nodes(void)
894{
895 struct device_node *rtas;
896 const __be32 *domains;
897 int prop_length, max_nodes;
898 u32 i;
899
900 if (!numa_enabled)
901 return;
902
903 rtas = of_find_node_by_path("/rtas");
904 if (!rtas)
905 return;
906
907
908
909
910
911
912
913 domains = of_get_property(rtas, "ibm,current-associativity-domains",
914 &prop_length);
915 if (!domains) {
916 domains = of_get_property(rtas, "ibm,max-associativity-domains",
917 &prop_length);
918 if (!domains)
919 goto out;
920 }
921
922 max_nodes = of_read_number(&domains[min_common_depth], 1);
923 for (i = 0; i < max_nodes; i++) {
924 if (!node_possible(i))
925 node_set(i, node_possible_map);
926 }
927
928 prop_length /= sizeof(int);
929 if (prop_length > min_common_depth + 2)
930 coregroup_enabled = 1;
931
932out:
933 of_node_put(rtas);
934}
935
936void __init mem_topology_setup(void)
937{
938 int cpu;
939
940
941
942
943
944
945
946
947
948 node_set_offline(0);
949
950 if (parse_numa_properties())
951 setup_nonnuma();
952
953
954
955
956
957
958
959 nodes_and(node_possible_map, node_possible_map, node_online_map);
960
961 find_possible_nodes();
962
963 setup_node_to_cpumask_map();
964
965 reset_numa_cpu_lookup_table();
966
967 for_each_possible_cpu(cpu) {
968
969
970
971
972
973
974
975
976 numa_setup_cpu(cpu);
977 }
978}
979
980void __init initmem_init(void)
981{
982 int nid;
983
984 max_low_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
985 max_pfn = max_low_pfn;
986
987 memblock_dump_all();
988
989 for_each_online_node(nid) {
990 unsigned long start_pfn, end_pfn;
991
992 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
993 setup_node_data(nid, start_pfn, end_pfn);
994 }
995
996 sparse_init();
997
998
999
1000
1001
1002
1003
1004
1005 cpuhp_setup_state_nocalls(CPUHP_POWER_NUMA_PREPARE, "powerpc/numa:prepare",
1006 ppc_numa_cpu_prepare, ppc_numa_cpu_dead);
1007}
1008
1009static int __init early_numa(char *p)
1010{
1011 if (!p)
1012 return 0;
1013
1014 if (strstr(p, "off"))
1015 numa_enabled = 0;
1016
1017 if (strstr(p, "debug"))
1018 numa_debug = 1;
1019
1020 p = strstr(p, "fake=");
1021 if (p)
1022 cmdline = p + strlen("fake=");
1023
1024 return 0;
1025}
1026early_param("numa", early_numa);
1027
1028#ifdef CONFIG_MEMORY_HOTPLUG
1029
1030
1031
1032
1033
1034static int hot_add_drconf_scn_to_nid(unsigned long scn_addr)
1035{
1036 struct drmem_lmb *lmb;
1037 unsigned long lmb_size;
1038 int nid = NUMA_NO_NODE;
1039
1040 lmb_size = drmem_lmb_size();
1041
1042 for_each_drmem_lmb(lmb) {
1043
1044
1045 if ((lmb->flags & DRCONF_MEM_RESERVED)
1046 || !(lmb->flags & DRCONF_MEM_ASSIGNED))
1047 continue;
1048
1049 if ((scn_addr < lmb->base_addr)
1050 || (scn_addr >= (lmb->base_addr + lmb_size)))
1051 continue;
1052
1053 nid = of_drconf_to_nid_single(lmb);
1054 break;
1055 }
1056
1057 return nid;
1058}
1059
1060
1061
1062
1063
1064
1065static int hot_add_node_scn_to_nid(unsigned long scn_addr)
1066{
1067 struct device_node *memory;
1068 int nid = NUMA_NO_NODE;
1069
1070 for_each_node_by_type(memory, "memory") {
1071 unsigned long start, size;
1072 int ranges;
1073 const __be32 *memcell_buf;
1074 unsigned int len;
1075
1076 memcell_buf = of_get_property(memory, "reg", &len);
1077 if (!memcell_buf || len <= 0)
1078 continue;
1079
1080
1081 ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells);
1082
1083 while (ranges--) {
1084 start = read_n_cells(n_mem_addr_cells, &memcell_buf);
1085 size = read_n_cells(n_mem_size_cells, &memcell_buf);
1086
1087 if ((scn_addr < start) || (scn_addr >= (start + size)))
1088 continue;
1089
1090 nid = of_node_to_nid_single(memory);
1091 break;
1092 }
1093
1094 if (nid >= 0)
1095 break;
1096 }
1097
1098 of_node_put(memory);
1099
1100 return nid;
1101}
1102
1103
1104
1105
1106
1107
1108int hot_add_scn_to_nid(unsigned long scn_addr)
1109{
1110 struct device_node *memory = NULL;
1111 int nid;
1112
1113 if (!numa_enabled)
1114 return first_online_node;
1115
1116 memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory");
1117 if (memory) {
1118 nid = hot_add_drconf_scn_to_nid(scn_addr);
1119 of_node_put(memory);
1120 } else {
1121 nid = hot_add_node_scn_to_nid(scn_addr);
1122 }
1123
1124 if (nid < 0 || !node_possible(nid))
1125 nid = first_online_node;
1126
1127 return nid;
1128}
1129
1130static u64 hot_add_drconf_memory_max(void)
1131{
1132 struct device_node *memory = NULL;
1133 struct device_node *dn = NULL;
1134 const __be64 *lrdr = NULL;
1135
1136 dn = of_find_node_by_path("/rtas");
1137 if (dn) {
1138 lrdr = of_get_property(dn, "ibm,lrdr-capacity", NULL);
1139 of_node_put(dn);
1140 if (lrdr)
1141 return be64_to_cpup(lrdr);
1142 }
1143
1144 memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory");
1145 if (memory) {
1146 of_node_put(memory);
1147 return drmem_lmb_memory_max();
1148 }
1149 return 0;
1150}
1151
1152
1153
1154
1155
1156
1157
1158u64 memory_hotplug_max(void)
1159{
1160 return max(hot_add_drconf_memory_max(), memblock_end_of_DRAM());
1161}
1162#endif
1163
1164
1165#ifdef CONFIG_PPC_SPLPAR
1166static int topology_inited;
1167
1168
1169
1170
1171
1172static long vphn_get_associativity(unsigned long cpu,
1173 __be32 *associativity)
1174{
1175 long rc;
1176
1177 rc = hcall_vphn(get_hard_smp_processor_id(cpu),
1178 VPHN_FLAG_VCPU, associativity);
1179
1180 switch (rc) {
1181 case H_SUCCESS:
1182 dbg("VPHN hcall succeeded. Reset polling...\n");
1183 goto out;
1184
1185 case H_FUNCTION:
1186 pr_err_ratelimited("VPHN unsupported. Disabling polling...\n");
1187 break;
1188 case H_HARDWARE:
1189 pr_err_ratelimited("hcall_vphn() experienced a hardware fault "
1190 "preventing VPHN. Disabling polling...\n");
1191 break;
1192 case H_PARAMETER:
1193 pr_err_ratelimited("hcall_vphn() was passed an invalid parameter. "
1194 "Disabling polling...\n");
1195 break;
1196 default:
1197 pr_err_ratelimited("hcall_vphn() returned %ld. Disabling polling...\n"
1198 , rc);
1199 break;
1200 }
1201out:
1202 return rc;
1203}
1204
1205int find_and_online_cpu_nid(int cpu)
1206{
1207 __be32 associativity[VPHN_ASSOC_BUFSIZE] = {0};
1208 int new_nid;
1209
1210
1211 if (vphn_get_associativity(cpu, associativity))
1212 return cpu_to_node(cpu);
1213
1214 new_nid = associativity_to_nid(associativity);
1215 if (new_nid < 0 || !node_possible(new_nid))
1216 new_nid = first_online_node;
1217
1218 if (NODE_DATA(new_nid) == NULL) {
1219#ifdef CONFIG_MEMORY_HOTPLUG
1220
1221
1222
1223
1224
1225
1226
1227 if (!topology_inited || try_online_node(new_nid))
1228 new_nid = first_online_node;
1229#else
1230
1231
1232
1233
1234
1235 new_nid = first_online_node;
1236#endif
1237 }
1238
1239 pr_debug("%s:%d cpu %d nid %d\n", __FUNCTION__, __LINE__,
1240 cpu, new_nid);
1241 return new_nid;
1242}
1243
1244int cpu_to_coregroup_id(int cpu)
1245{
1246 __be32 associativity[VPHN_ASSOC_BUFSIZE] = {0};
1247 int index;
1248
1249 if (cpu < 0 || cpu > nr_cpu_ids)
1250 return -1;
1251
1252 if (!coregroup_enabled)
1253 goto out;
1254
1255 if (!firmware_has_feature(FW_FEATURE_VPHN))
1256 goto out;
1257
1258 if (vphn_get_associativity(cpu, associativity))
1259 goto out;
1260
1261 index = of_read_number(associativity, 1);
1262 if (index > min_common_depth + 1)
1263 return of_read_number(&associativity[index - 1], 1);
1264
1265out:
1266 return cpu_to_core_id(cpu);
1267}
1268
1269static int topology_update_init(void)
1270{
1271 topology_inited = 1;
1272 return 0;
1273}
1274device_initcall(topology_update_init);
1275#endif
1276