?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6
[linux-2.6/linux-mips/linux-dm7025.git] / arch / powerpc / mm / numa.c
blobdc704da363eb77897d392781591bf87ac337a8dd
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>
20 #include <linux/lmb.h>
21 #include <linux/of.h>
22 #include <asm/sparsemem.h>
23 #include <asm/prom.h>
24 #include <asm/system.h>
25 #include <asm/smp.h>
26
27 static int numa_enabled = 1;
28
29 static char *cmdline __initdata;
30
31 static int numa_debug;
32 #define dbg(args...) if (numa_debug) { printk(KERN_INFO args); }
33
34 int numa_cpu_lookup_table[NR_CPUS];
35 cpumask_t numa_cpumask_lookup_table[MAX_NUMNODES];
36 struct pglist_data *node_data[MAX_NUMNODES];
37
38 EXPORT_SYMBOL(numa_cpu_lookup_table);
39 EXPORT_SYMBOL(numa_cpumask_lookup_table);
40 EXPORT_SYMBOL(node_data);
41
42 static bootmem_data_t __initdata plat_node_bdata[MAX_NUMNODES];
43 static int min_common_depth;
44 static int n_mem_addr_cells, n_mem_size_cells;
45
46 static int __cpuinit fake_numa_create_new_node(unsigned long end_pfn,
47 unsigned int *nid)
48 {
49 unsigned long long mem;
50 char *p = cmdline;
51 static unsigned int fake_nid;
52 static unsigned long long curr_boundary;
53
54 /*
55 * Modify node id, iff we started creating NUMA nodes
56 * We want to continue from where we left of the last time
57 */
58 if (fake_nid)
59 *nid = fake_nid;
60 /*
61 * In case there are no more arguments to parse, the
62 * node_id should be the same as the last fake node id
63 * (we've handled this above).
64 */
65 if (!p)
66 return 0;
67
68 mem = memparse(p, &p);
69 if (!mem)
70 return 0;
71
72 if (mem < curr_boundary)
73 return 0;
74
75 curr_boundary = mem;
76
77 if ((end_pfn << PAGE_SHIFT) > mem) {
78 /*
79 * Skip commas and spaces
80 */
81 while (*p == ',' || *p == ' ' || *p == '\t')
82 p++;
83
84 cmdline = p;
85 fake_nid++;
86 *nid = fake_nid;
87 dbg("created new fake_node with id %d\n", fake_nid);
88 return 1;
89 }
90 return 0;
91 }
92
93 static void __cpuinit map_cpu_to_node(int cpu, int node)
94 {
95 numa_cpu_lookup_table[cpu] = node;
96
97 dbg("adding cpu %d to node %d\n", cpu, node);
98
99 if (!(cpu_isset(cpu, numa_cpumask_lookup_table[node])))
100 cpu_set(cpu, numa_cpumask_lookup_table[node]);
101 }
102
103 #ifdef CONFIG_HOTPLUG_CPU
104 static void unmap_cpu_from_node(unsigned long cpu)
105 {
106 int node = numa_cpu_lookup_table[cpu];
107
108 dbg("removing cpu %lu from node %d\n", cpu, node);
109
110 if (cpu_isset(cpu, numa_cpumask_lookup_table[node])) {
111 cpu_clear(cpu, numa_cpumask_lookup_table[node]);
112 } else {
113 printk(KERN_ERR "WARNING: cpu %lu not found in node %d\n",
114 cpu, node);
115 }
116 }
117 #endif /* CONFIG_HOTPLUG_CPU */
118
119 static struct device_node * __cpuinit find_cpu_node(unsigned int cpu)
120 {
121 unsigned int hw_cpuid = get_hard_smp_processor_id(cpu);
122 struct device_node *cpu_node = NULL;
123 const unsigned int *interrupt_server, *reg;
124 int len;
125
126 while ((cpu_node = of_find_node_by_type(cpu_node, "cpu")) != NULL) {
127 /* Try interrupt server first */
128 interrupt_server = of_get_property(cpu_node,
129 "ibm,ppc-interrupt-server#s", &len);
130
131 len = len / sizeof(u32);
132
133 if (interrupt_server && (len > 0)) {
134 while (len--) {
135 if (interrupt_server[len] == hw_cpuid)
136 return cpu_node;
137 }
138 } else {
139 reg = of_get_property(cpu_node, "reg", &len);
140 if (reg && (len > 0) && (reg[0] == hw_cpuid))
141 return cpu_node;
142 }
143 }
144
145 return NULL;
146 }
147
148 /* must hold reference to node during call */
149 static const int *of_get_associativity(struct device_node *dev)
150 {
151 return of_get_property(dev, "ibm,associativity", NULL);
152 }
153
154 /* Returns nid in the range [0..MAX_NUMNODES-1], or -1 if no useful numa
155 * info is found.
156 */
157 static int of_node_to_nid_single(struct device_node *device)
158 {
159 int nid = -1;
160 const unsigned int *tmp;
161
162 if (min_common_depth == -1)
163 goto out;
164
165 tmp = of_get_associativity(device);
166 if (!tmp)
167 goto out;
168
169 if (tmp[0] >= min_common_depth)
170 nid = tmp[min_common_depth];
171
172 /* POWER4 LPAR uses 0xffff as invalid node */
173 if (nid == 0xffff || nid >= MAX_NUMNODES)
174 nid = -1;
175 out:
176 return nid;
177 }
178
179 /* Walk the device tree upwards, looking for an associativity id */
180 int of_node_to_nid(struct device_node *device)
181 {
182 struct device_node *tmp;
183 int nid = -1;
184
185 of_node_get(device);
186 while (device) {
187 nid = of_node_to_nid_single(device);
188 if (nid != -1)
189 break;
190
191 tmp = device;
192 device = of_get_parent(tmp);
193 of_node_put(tmp);
194 }
195 of_node_put(device);
196
197 return nid;
198 }
199 EXPORT_SYMBOL_GPL(of_node_to_nid);
200
201 /*
202 * In theory, the "ibm,associativity" property may contain multiple
203 * associativity lists because a resource may be multiply connected
204 * into the machine. This resource then has different associativity
205 * characteristics relative to its multiple connections. We ignore
206 * this for now. We also assume that all cpu and memory sets have
207 * their distances represented at a common level. This won't be
208 * true for hierarchical NUMA.
209 *
210 * In any case the ibm,associativity-reference-points should give
211 * the correct depth for a normal NUMA system.
212 *
213 * - Dave Hansen <haveblue@us.ibm.com>
214 */
215 static int __init find_min_common_depth(void)
216 {
217 int depth;
218 const unsigned int *ref_points;
219 struct device_node *rtas_root;
220 unsigned int len;
221
222 rtas_root = of_find_node_by_path("/rtas");
223
224 if (!rtas_root)
225 return -1;
226
227 /*
228 * this property is 2 32-bit integers, each representing a level of
229 * depth in the associativity nodes. The first is for an SMP
230 * configuration (should be all 0's) and the second is for a normal
231 * NUMA configuration.
232 */
233 ref_points = of_get_property(rtas_root,
234 "ibm,associativity-reference-points", &len);
235
236 if ((len >= 1) && ref_points) {
237 depth = ref_points[1];
238 } else {
239 dbg("NUMA: ibm,associativity-reference-points not found.\n");
240 depth = -1;
241 }
242 of_node_put(rtas_root);
243
244 return depth;
245 }
246
247 static void __init get_n_mem_cells(int *n_addr_cells, int *n_size_cells)
248 {
249 struct device_node *memory = NULL;
250
251 memory = of_find_node_by_type(memory, "memory");
252 if (!memory)
253 panic("numa.c: No memory nodes found!");
254
255 *n_addr_cells = of_n_addr_cells(memory);
256 *n_size_cells = of_n_size_cells(memory);
257 of_node_put(memory);
258 }
259
260 static unsigned long __devinit read_n_cells(int n, const unsigned int **buf)
261 {
262 unsigned long result = 0;
263
264 while (n--) {
265 result = (result << 32) | **buf;
266 (*buf)++;
267 }
268 return result;
269 }
270
271 /*
272 * Figure out to which domain a cpu belongs and stick it there.
273 * Return the id of the domain used.
274 */
275 static int __cpuinit numa_setup_cpu(unsigned long lcpu)
276 {
277 int nid = 0;
278 struct device_node *cpu = find_cpu_node(lcpu);
279
280 if (!cpu) {
281 WARN_ON(1);
282 goto out;
283 }
284
285 nid = of_node_to_nid_single(cpu);
286
287 if (nid < 0 || !node_online(nid))
288 nid = any_online_node(NODE_MASK_ALL);
289 out:
290 map_cpu_to_node(lcpu, nid);
291
292 of_node_put(cpu);
293
294 return nid;
295 }
296
297 static int __cpuinit cpu_numa_callback(struct notifier_block *nfb,
298 unsigned long action,
299 void *hcpu)
300 {
301 unsigned long lcpu = (unsigned long)hcpu;
302 int ret = NOTIFY_DONE;
303
304 switch (action) {
305 case CPU_UP_PREPARE:
306 case CPU_UP_PREPARE_FROZEN:
307 numa_setup_cpu(lcpu);
308 ret = NOTIFY_OK;
309 break;
310 #ifdef CONFIG_HOTPLUG_CPU
311 case CPU_DEAD:
312 case CPU_DEAD_FROZEN:
313 case CPU_UP_CANCELED:
314 case CPU_UP_CANCELED_FROZEN:
315 unmap_cpu_from_node(lcpu);
316 break;
317 ret = NOTIFY_OK;
318 #endif
319 }
320 return ret;
321 }
322
323 /*
324 * Check and possibly modify a memory region to enforce the memory limit.
325 *
326 * Returns the size the region should have to enforce the memory limit.
327 * This will either be the original value of size, a truncated value,
328 * or zero. If the returned value of size is 0 the region should be
329 * discarded as it lies wholy above the memory limit.
330 */
331 static unsigned long __init numa_enforce_memory_limit(unsigned long start,
332 unsigned long size)
333 {
334 /*
335 * We use lmb_end_of_DRAM() in here instead of memory_limit because
336 * we've already adjusted it for the limit and it takes care of
337 * having memory holes below the limit.
338 */
339
340 if (! memory_limit)
341 return size;
342
343 if (start + size <= lmb_end_of_DRAM())
344 return size;
345
346 if (start >= lmb_end_of_DRAM())
347 return 0;
348
349 return lmb_end_of_DRAM() - start;
350 }
351
352 /*
353 * Extract NUMA information from the ibm,dynamic-reconfiguration-memory
354 * node. This assumes n_mem_{addr,size}_cells have been set.
355 */
356 static void __init parse_drconf_memory(struct device_node *memory)
357 {
358 const unsigned int *lm, *dm, *aa;
359 unsigned int ls, ld, la;
360 unsigned int n, aam, aalen;
361 unsigned long lmb_size, size, start;
362 int nid, default_nid = 0;
363 unsigned int ai, flags;
364
365 lm = of_get_property(memory, "ibm,lmb-size", &ls);
366 dm = of_get_property(memory, "ibm,dynamic-memory", &ld);
367 aa = of_get_property(memory, "ibm,associativity-lookup-arrays", &la);
368 if (!lm || !dm || !aa ||
369 ls < sizeof(unsigned int) || ld < sizeof(unsigned int) ||
370 la < 2 * sizeof(unsigned int))
371 return;
372
373 lmb_size = read_n_cells(n_mem_size_cells, &lm);
374 n = *dm++; /* number of LMBs */
375 aam = *aa++; /* number of associativity lists */
376 aalen = *aa++; /* length of each associativity list */
377 if (ld < (n * (n_mem_addr_cells + 4) + 1) * sizeof(unsigned int) ||
378 la < (aam * aalen + 2) * sizeof(unsigned int))
379 return;
380
381 for (; n != 0; --n) {
382 start = read_n_cells(n_mem_addr_cells, &dm);
383 ai = dm[2];
384 flags = dm[3];
385 dm += 4;
386 /* 0x80 == reserved, 0x8 = assigned to us */
387 if ((flags & 0x80) || !(flags & 0x8))
388 continue;
389 nid = default_nid;
390 /* flags & 0x40 means associativity index is invalid */
391 if (min_common_depth > 0 && min_common_depth <= aalen &&
392 (flags & 0x40) == 0 && ai < aam) {
393 /* this is like of_node_to_nid_single */
394 nid = aa[ai * aalen + min_common_depth - 1];
395 if (nid == 0xffff || nid >= MAX_NUMNODES)
396 nid = default_nid;
397 }
398
399 fake_numa_create_new_node(((start + lmb_size) >> PAGE_SHIFT),
400 &nid);
401 node_set_online(nid);
402
403 size = numa_enforce_memory_limit(start, lmb_size);
404 if (!size)
405 continue;
406
407 add_active_range(nid, start >> PAGE_SHIFT,
408 (start >> PAGE_SHIFT) + (size >> PAGE_SHIFT));
409 }
410 }
411
412 static int __init parse_numa_properties(void)
413 {
414 struct device_node *cpu = NULL;
415 struct device_node *memory = NULL;
416 int default_nid = 0;
417 unsigned long i;
418
419 if (numa_enabled == 0) {
420 printk(KERN_WARNING "NUMA disabled by user\n");
421 return -1;
422 }
423
424 min_common_depth = find_min_common_depth();
425
426 if (min_common_depth < 0)
427 return min_common_depth;
428
429 dbg("NUMA associativity depth for CPU/Memory: %d\n", min_common_depth);
430
431 /*
432 * Even though we connect cpus to numa domains later in SMP
433 * init, we need to know the node ids now. This is because
434 * each node to be onlined must have NODE_DATA etc backing it.
435 */
436 for_each_present_cpu(i) {
437 int nid;
438
439 cpu = find_cpu_node(i);
440 BUG_ON(!cpu);
441 nid = of_node_to_nid_single(cpu);
442 of_node_put(cpu);
443
444 /*
445 * Don't fall back to default_nid yet -- we will plug
446 * cpus into nodes once the memory scan has discovered
447 * the topology.
448 */
449 if (nid < 0)
450 continue;
451 node_set_online(nid);
452 }
453
454 get_n_mem_cells(&n_mem_addr_cells, &n_mem_size_cells);
455 memory = NULL;
456 while ((memory = of_find_node_by_type(memory, "memory")) != NULL) {
457 unsigned long start;
458 unsigned long size;
459 int nid;
460 int ranges;
461 const unsigned int *memcell_buf;
462 unsigned int len;
463
464 memcell_buf = of_get_property(memory,
465 "linux,usable-memory", &len);
466 if (!memcell_buf || len <= 0)
467 memcell_buf = of_get_property(memory, "reg", &len);
468 if (!memcell_buf || len <= 0)
469 continue;
470
471 /* ranges in cell */
472 ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells);
473 new_range:
474 /* these are order-sensitive, and modify the buffer pointer */
475 start = read_n_cells(n_mem_addr_cells, &memcell_buf);
476 size = read_n_cells(n_mem_size_cells, &memcell_buf);
477
478 /*
479 * Assumption: either all memory nodes or none will
480 * have associativity properties. If none, then
481 * everything goes to default_nid.
482 */
483 nid = of_node_to_nid_single(memory);
484 if (nid < 0)
485 nid = default_nid;
486
487 fake_numa_create_new_node(((start + size) >> PAGE_SHIFT), &nid);
488 node_set_online(nid);
489
490 if (!(size = numa_enforce_memory_limit(start, size))) {
491 if (--ranges)
492 goto new_range;
493 else
494 continue;
495 }
496
497 add_active_range(nid, start >> PAGE_SHIFT,
498 (start >> PAGE_SHIFT) + (size >> PAGE_SHIFT));
499
500 if (--ranges)
501 goto new_range;
502 }
503
504 /*
505 * Now do the same thing for each LMB listed in the ibm,dynamic-memory
506 * property in the ibm,dynamic-reconfiguration-memory node.
507 */
508 memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory");
509 if (memory)
510 parse_drconf_memory(memory);
511
512 return 0;
513 }
514
515 static void __init setup_nonnuma(void)
516 {
517 unsigned long top_of_ram = lmb_end_of_DRAM();
518 unsigned long total_ram = lmb_phys_mem_size();
519 unsigned long start_pfn, end_pfn;
520 unsigned int i, nid = 0;
521
522 printk(KERN_DEBUG "Top of RAM: 0x%lx, Total RAM: 0x%lx\n",
523 top_of_ram, total_ram);
524 printk(KERN_DEBUG "Memory hole size: %ldMB\n",
525 (top_of_ram - total_ram) >> 20);
526
527 for (i = 0; i < lmb.memory.cnt; ++i) {
528 start_pfn = lmb.memory.region[i].base >> PAGE_SHIFT;
529 end_pfn = start_pfn + lmb_size_pages(&lmb.memory, i);
530
531 fake_numa_create_new_node(end_pfn, &nid);
532 add_active_range(nid, start_pfn, end_pfn);
533 node_set_online(nid);
534 }
535 }
536
537 void __init dump_numa_cpu_topology(void)
538 {
539 unsigned int node;
540 unsigned int cpu, count;
541
542 if (min_common_depth == -1 || !numa_enabled)
543 return;
544
545 for_each_online_node(node) {
546 printk(KERN_DEBUG "Node %d CPUs:", node);
547
548 count = 0;
549 /*
550 * If we used a CPU iterator here we would miss printing
551 * the holes in the cpumap.
552 */
553 for (cpu = 0; cpu < NR_CPUS; cpu++) {
554 if (cpu_isset(cpu, numa_cpumask_lookup_table[node])) {
555 if (count == 0)
556 printk(" %u", cpu);
557 ++count;
558 } else {
559 if (count > 1)
560 printk("-%u", cpu - 1);
561 count = 0;
562 }
563 }
564
565 if (count > 1)
566 printk("-%u", NR_CPUS - 1);
567 printk("\n");
568 }
569 }
570
571 static void __init dump_numa_memory_topology(void)
572 {
573 unsigned int node;
574 unsigned int count;
575
576 if (min_common_depth == -1 || !numa_enabled)
577 return;
578
579 for_each_online_node(node) {
580 unsigned long i;
581
582 printk(KERN_DEBUG "Node %d Memory:", node);
583
584 count = 0;
585
586 for (i = 0; i < lmb_end_of_DRAM();
587 i += (1 << SECTION_SIZE_BITS)) {
588 if (early_pfn_to_nid(i >> PAGE_SHIFT) == node) {
589 if (count == 0)
590 printk(" 0x%lx", i);
591 ++count;
592 } else {
593 if (count > 0)
594 printk("-0x%lx", i);
595 count = 0;
596 }
597 }
598
599 if (count > 0)
600 printk("-0x%lx", i);
601 printk("\n");
602 }
603 }
604
605 /*
606 * Allocate some memory, satisfying the lmb or bootmem allocator where
607 * required. nid is the preferred node and end is the physical address of
608 * the highest address in the node.
609 *
610 * Returns the physical address of the memory.
611 */
612 static void __init *careful_allocation(int nid, unsigned long size,
613 unsigned long align,
614 unsigned long end_pfn)
615 {
616 int new_nid;
617 unsigned long ret = __lmb_alloc_base(size, align, end_pfn << PAGE_SHIFT);
618
619 /* retry over all memory */
620 if (!ret)
621 ret = __lmb_alloc_base(size, align, lmb_end_of_DRAM());
622
623 if (!ret)
624 panic("numa.c: cannot allocate %lu bytes on node %d",
625 size, nid);
626
627 /*
628 * If the memory came from a previously allocated node, we must
629 * retry with the bootmem allocator.
630 */
631 new_nid = early_pfn_to_nid(ret >> PAGE_SHIFT);
632 if (new_nid < nid) {
633 ret = (unsigned long)__alloc_bootmem_node(NODE_DATA(new_nid),
634 size, align, 0);
635
636 if (!ret)
637 panic("numa.c: cannot allocate %lu bytes on node %d",
638 size, new_nid);
639
640 ret = __pa(ret);
641
642 dbg("alloc_bootmem %lx %lx\n", ret, size);
643 }
644
645 return (void *)ret;
646 }
647
648 static struct notifier_block __cpuinitdata ppc64_numa_nb = {
649 .notifier_call = cpu_numa_callback,
650 .priority = 1 /* Must run before sched domains notifier. */
651 };
652
653 void __init do_init_bootmem(void)
654 {
655 int nid;
656 unsigned int i;
657
658 min_low_pfn = 0;
659 max_low_pfn = lmb_end_of_DRAM() >> PAGE_SHIFT;
660 max_pfn = max_low_pfn;
661
662 if (parse_numa_properties())
663 setup_nonnuma();
664 else
665 dump_numa_memory_topology();
666
667 register_cpu_notifier(&ppc64_numa_nb);
668 cpu_numa_callback(&ppc64_numa_nb, CPU_UP_PREPARE,
669 (void *)(unsigned long)boot_cpuid);
670
671 for_each_online_node(nid) {
672 unsigned long start_pfn, end_pfn;
673 unsigned long bootmem_paddr;
674 unsigned long bootmap_pages;
675
676 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
677
678 /* Allocate the node structure node local if possible */
679 NODE_DATA(nid) = careful_allocation(nid,
680 sizeof(struct pglist_data),
681 SMP_CACHE_BYTES, end_pfn);
682 NODE_DATA(nid) = __va(NODE_DATA(nid));
683 memset(NODE_DATA(nid), 0, sizeof(struct pglist_data));
684
685 dbg("node %d\n", nid);
686 dbg("NODE_DATA() = %p\n", NODE_DATA(nid));
687
688 NODE_DATA(nid)->bdata = &plat_node_bdata[nid];
689 NODE_DATA(nid)->node_start_pfn = start_pfn;
690 NODE_DATA(nid)->node_spanned_pages = end_pfn - start_pfn;
691
692 if (NODE_DATA(nid)->node_spanned_pages == 0)
693 continue;
694
695 dbg("start_paddr = %lx\n", start_pfn << PAGE_SHIFT);
696 dbg("end_paddr = %lx\n", end_pfn << PAGE_SHIFT);
697
698 bootmap_pages = bootmem_bootmap_pages(end_pfn - start_pfn);
699 bootmem_paddr = (unsigned long)careful_allocation(nid,
700 bootmap_pages << PAGE_SHIFT,
701 PAGE_SIZE, end_pfn);
702 memset(__va(bootmem_paddr), 0, bootmap_pages << PAGE_SHIFT);
703
704 dbg("bootmap_paddr = %lx\n", bootmem_paddr);
705
706 init_bootmem_node(NODE_DATA(nid), bootmem_paddr >> PAGE_SHIFT,
707 start_pfn, end_pfn);
708
709 free_bootmem_with_active_regions(nid, end_pfn);
710
711 /* Mark reserved regions on this node */
712 for (i = 0; i < lmb.reserved.cnt; i++) {
713 unsigned long physbase = lmb.reserved.region[i].base;
714 unsigned long size = lmb.reserved.region[i].size;
715 unsigned long start_paddr = start_pfn << PAGE_SHIFT;
716 unsigned long end_paddr = end_pfn << PAGE_SHIFT;
717
718 if (early_pfn_to_nid(physbase >> PAGE_SHIFT) != nid &&
719 early_pfn_to_nid((physbase+size-1) >> PAGE_SHIFT) != nid)
720 continue;
721
722 if (physbase < end_paddr &&
723 (physbase+size) > start_paddr) {
724 /* overlaps */
725 if (physbase < start_paddr) {
726 size -= start_paddr - physbase;
727 physbase = start_paddr;
728 }
729
730 if (size > end_paddr - physbase)
731 size = end_paddr - physbase;
732
733 dbg("reserve_bootmem %lx %lx\n", physbase,
734 size);
735 reserve_bootmem_node(NODE_DATA(nid), physbase,
736 size, BOOTMEM_DEFAULT);
737 }
738 }
739
740 sparse_memory_present_with_active_regions(nid);
741 }
742 }
743
744 void __init paging_init(void)
745 {
746 unsigned long max_zone_pfns[MAX_NR_ZONES];
747 memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
748 max_zone_pfns[ZONE_DMA] = lmb_end_of_DRAM() >> PAGE_SHIFT;
749 free_area_init_nodes(max_zone_pfns);
750 }
751
752 static int __init early_numa(char *p)
753 {
754 if (!p)
755 return 0;
756
757 if (strstr(p, "off"))
758 numa_enabled = 0;
759
760 if (strstr(p, "debug"))
761 numa_debug = 1;
762
763 p = strstr(p, "fake=");
764 if (p)
765 cmdline = p + strlen("fake=");
766
767 return 0;
768 }
769 early_param("numa", early_numa);
770
771 #ifdef CONFIG_MEMORY_HOTPLUG
772 /*
773 * Find the node associated with a hot added memory section. Section
774 * corresponds to a SPARSEMEM section, not an LMB. It is assumed that
775 * sections are fully contained within a single LMB.
776 */
777 int hot_add_scn_to_nid(unsigned long scn_addr)
778 {
779 struct device_node *memory = NULL;
780 nodemask_t nodes;
781 int default_nid = any_online_node(NODE_MASK_ALL);
782 int nid;
783
784 if (!numa_enabled || (min_common_depth < 0))
785 return default_nid;
786
787 while ((memory = of_find_node_by_type(memory, "memory")) != NULL) {
788 unsigned long start, size;
789 int ranges;
790 const unsigned int *memcell_buf;
791 unsigned int len;
792
793 memcell_buf = of_get_property(memory, "reg", &len);
794 if (!memcell_buf || len <= 0)
795 continue;
796
797 /* ranges in cell */
798 ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells);
799 ha_new_range:
800 start = read_n_cells(n_mem_addr_cells, &memcell_buf);
801 size = read_n_cells(n_mem_size_cells, &memcell_buf);
802 nid = of_node_to_nid_single(memory);
803
804 /* Domains not present at boot default to 0 */
805 if (nid < 0 || !node_online(nid))
806 nid = default_nid;
807
808 if ((scn_addr >= start) && (scn_addr < (start + size))) {
809 of_node_put(memory);
810 goto got_nid;
811 }
812
813 if (--ranges) /* process all ranges in cell */
814 goto ha_new_range;
815 }
816 BUG(); /* section address should be found above */
817 return 0;
818
819 /* Temporary code to ensure that returned node is not empty */
820 got_nid:
821 nodes_setall(nodes);
822 while (NODE_DATA(nid)->node_spanned_pages == 0) {
823 node_clear(nid, nodes);
824 nid = any_online_node(nodes);
825 }
826 return nid;
827 }
828 #endif /* CONFIG_MEMORY_HOTPLUG */
A Kernel for the Dreambox dm7025.