numa.c 20.1 KB
Newer Older
L
Linus Torvalds 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
/*
 * pSeries NUMA support
 *
 * Copyright (C) 2002 Anton Blanchard <anton@au.ibm.com>, IBM
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License
 * as published by the Free Software Foundation; either version
 * 2 of the License, or (at your option) any later version.
 */
#include <linux/threads.h>
#include <linux/bootmem.h>
#include <linux/init.h>
#include <linux/mm.h>
#include <linux/mmzone.h>
#include <linux/module.h>
#include <linux/nodemask.h>
#include <linux/cpu.h>
#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>
P
Paul Mackerras 已提交
25
#include <asm/smp.h>
L
Linus Torvalds 已提交
26 27 28

static int numa_enabled = 1;

29 30
static char *cmdline __initdata;

L
Linus Torvalds 已提交
31 32 33
static int numa_debug;
#define dbg(args...) if (numa_debug) { printk(KERN_INFO args); }

34
int numa_cpu_lookup_table[NR_CPUS];
L
Linus Torvalds 已提交
35 36
cpumask_t numa_cpumask_lookup_table[MAX_NUMNODES];
struct pglist_data *node_data[MAX_NUMNODES];
37 38 39 40 41 42

EXPORT_SYMBOL(numa_cpu_lookup_table);
EXPORT_SYMBOL(numa_cpumask_lookup_table);
EXPORT_SYMBOL(node_data);

static bootmem_data_t __initdata plat_node_bdata[MAX_NUMNODES];
L
Linus Torvalds 已提交
43
static int min_common_depth;
44
static int n_mem_addr_cells, n_mem_size_cells;
L
Linus Torvalds 已提交
45

46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92
static int __cpuinit fake_numa_create_new_node(unsigned long end_pfn,
						unsigned int *nid)
{
	unsigned long long mem;
	char *p = cmdline;
	static unsigned int fake_nid;
	static unsigned long long curr_boundary;

	/*
	 * Modify node id, iff we started creating NUMA nodes
	 * We want to continue from where we left of the last time
	 */
	if (fake_nid)
		*nid = fake_nid;
	/*
	 * In case there are no more arguments to parse, the
	 * node_id should be the same as the last fake node id
	 * (we've handled this above).
	 */
	if (!p)
		return 0;

	mem = memparse(p, &p);
	if (!mem)
		return 0;

	if (mem < curr_boundary)
		return 0;

	curr_boundary = mem;

	if ((end_pfn << PAGE_SHIFT) > mem) {
		/*
		 * Skip commas and spaces
		 */
		while (*p == ',' || *p == ' ' || *p == '\t')
			p++;

		cmdline = p;
		fake_nid++;
		*nid = fake_nid;
		dbg("created new fake_node with id %d\n", fake_nid);
		return 1;
	}
	return 0;
}

93
static void __cpuinit map_cpu_to_node(int cpu, int node)
L
Linus Torvalds 已提交
94 95
{
	numa_cpu_lookup_table[cpu] = node;
96

97 98
	dbg("adding cpu %d to node %d\n", cpu, node);

99
	if (!(cpu_isset(cpu, numa_cpumask_lookup_table[node])))
L
Linus Torvalds 已提交
100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118
		cpu_set(cpu, numa_cpumask_lookup_table[node]);
}

#ifdef CONFIG_HOTPLUG_CPU
static void unmap_cpu_from_node(unsigned long cpu)
{
	int node = numa_cpu_lookup_table[cpu];

	dbg("removing cpu %lu from node %d\n", cpu, node);

	if (cpu_isset(cpu, numa_cpumask_lookup_table[node])) {
		cpu_clear(cpu, numa_cpumask_lookup_table[node]);
	} else {
		printk(KERN_ERR "WARNING: cpu %lu not found in node %d\n",
		       cpu, node);
	}
}
#endif /* CONFIG_HOTPLUG_CPU */

119
static struct device_node * __cpuinit find_cpu_node(unsigned int cpu)
L
Linus Torvalds 已提交
120 121 122
{
	unsigned int hw_cpuid = get_hard_smp_processor_id(cpu);
	struct device_node *cpu_node = NULL;
123
	const unsigned int *interrupt_server, *reg;
L
Linus Torvalds 已提交
124 125 126 127
	int len;

	while ((cpu_node = of_find_node_by_type(cpu_node, "cpu")) != NULL) {
		/* Try interrupt server first */
128
		interrupt_server = of_get_property(cpu_node,
L
Linus Torvalds 已提交
129 130 131 132 133 134 135 136 137 138
					"ibm,ppc-interrupt-server#s", &len);

		len = len / sizeof(u32);

		if (interrupt_server && (len > 0)) {
			while (len--) {
				if (interrupt_server[len] == hw_cpuid)
					return cpu_node;
			}
		} else {
139
			reg = of_get_property(cpu_node, "reg", &len);
L
Linus Torvalds 已提交
140 141 142 143 144 145 146 147 148
			if (reg && (len > 0) && (reg[0] == hw_cpuid))
				return cpu_node;
		}
	}

	return NULL;
}

/* must hold reference to node during call */
149
static const int *of_get_associativity(struct device_node *dev)
L
Linus Torvalds 已提交
150
{
151
	return of_get_property(dev, "ibm,associativity", NULL);
L
Linus Torvalds 已提交
152 153
}

154 155 156
/* Returns nid in the range [0..MAX_NUMNODES-1], or -1 if no useful numa
 * info is found.
 */
157
static int of_node_to_nid_single(struct device_node *device)
L
Linus Torvalds 已提交
158
{
159
	int nid = -1;
160
	const unsigned int *tmp;
L
Linus Torvalds 已提交
161 162

	if (min_common_depth == -1)
163
		goto out;
L
Linus Torvalds 已提交
164 165

	tmp = of_get_associativity(device);
166 167 168 169
	if (!tmp)
		goto out;

	if (tmp[0] >= min_common_depth)
170
		nid = tmp[min_common_depth];
171 172

	/* POWER4 LPAR uses 0xffff as invalid node */
173 174 175
	if (nid == 0xffff || nid >= MAX_NUMNODES)
		nid = -1;
out:
176
	return nid;
L
Linus Torvalds 已提交
177 178
}

179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200
/* Walk the device tree upwards, looking for an associativity id */
int of_node_to_nid(struct device_node *device)
{
	struct device_node *tmp;
	int nid = -1;

	of_node_get(device);
	while (device) {
		nid = of_node_to_nid_single(device);
		if (nid != -1)
			break;

	        tmp = device;
		device = of_get_parent(tmp);
		of_node_put(tmp);
	}
	of_node_put(device);

	return nid;
}
EXPORT_SYMBOL_GPL(of_node_to_nid);

L
Linus Torvalds 已提交
201 202 203 204 205 206 207
/*
 * In theory, the "ibm,associativity" property may contain multiple
 * associativity lists because a resource may be multiply connected
 * into the machine.  This resource then has different associativity
 * characteristics relative to its multiple connections.  We ignore
 * this for now.  We also assume that all cpu and memory sets have
 * their distances represented at a common level.  This won't be
208
 * true for hierarchical NUMA.
L
Linus Torvalds 已提交
209 210 211 212 213 214 215 216 217
 *
 * In any case the ibm,associativity-reference-points should give
 * the correct depth for a normal NUMA system.
 *
 * - Dave Hansen <haveblue@us.ibm.com>
 */
static int __init find_min_common_depth(void)
{
	int depth;
218
	const unsigned int *ref_points;
L
Linus Torvalds 已提交
219 220 221 222 223 224 225 226 227 228 229 230 231 232
	struct device_node *rtas_root;
	unsigned int len;

	rtas_root = of_find_node_by_path("/rtas");

	if (!rtas_root)
		return -1;

	/*
	 * this property is 2 32-bit integers, each representing a level of
	 * depth in the associativity nodes.  The first is for an SMP
	 * configuration (should be all 0's) and the second is for a normal
	 * NUMA configuration.
	 */
233
	ref_points = of_get_property(rtas_root,
L
Linus Torvalds 已提交
234 235 236 237 238
			"ibm,associativity-reference-points", &len);

	if ((len >= 1) && ref_points) {
		depth = ref_points[1];
	} else {
239
		dbg("NUMA: ibm,associativity-reference-points not found.\n");
L
Linus Torvalds 已提交
240 241 242 243 244 245 246
		depth = -1;
	}
	of_node_put(rtas_root);

	return depth;
}

247
static void __init get_n_mem_cells(int *n_addr_cells, int *n_size_cells)
L
Linus Torvalds 已提交
248 249 250 251
{
	struct device_node *memory = NULL;

	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);
L
Linus Torvalds 已提交
258 259
}

260
static unsigned long __devinit read_n_cells(int n, const unsigned int **buf)
L
Linus Torvalds 已提交
261 262 263 264 265 266 267 268 269 270 271 272 273 274
{
	unsigned long result = 0;

	while (n--) {
		result = (result << 32) | **buf;
		(*buf)++;
	}
	return result;
}

/*
 * Figure out to which domain a cpu belongs and stick it there.
 * Return the id of the domain used.
 */
275
static int __cpuinit numa_setup_cpu(unsigned long lcpu)
L
Linus Torvalds 已提交
276
{
277
	int nid = 0;
L
Linus Torvalds 已提交
278 279 280 281 282 283 284
	struct device_node *cpu = find_cpu_node(lcpu);

	if (!cpu) {
		WARN_ON(1);
		goto out;
	}

285
	nid = of_node_to_nid_single(cpu);
L
Linus Torvalds 已提交
286

287 288
	if (nid < 0 || !node_online(nid))
		nid = any_online_node(NODE_MASK_ALL);
L
Linus Torvalds 已提交
289
out:
290
	map_cpu_to_node(lcpu, nid);
L
Linus Torvalds 已提交
291 292 293

	of_node_put(cpu);

294
	return nid;
L
Linus Torvalds 已提交
295 296
}

297
static int __cpuinit cpu_numa_callback(struct notifier_block *nfb,
L
Linus Torvalds 已提交
298 299 300 301 302 303 304 305
			     unsigned long action,
			     void *hcpu)
{
	unsigned long lcpu = (unsigned long)hcpu;
	int ret = NOTIFY_DONE;

	switch (action) {
	case CPU_UP_PREPARE:
306
	case CPU_UP_PREPARE_FROZEN:
307
		numa_setup_cpu(lcpu);
L
Linus Torvalds 已提交
308 309 310 311
		ret = NOTIFY_OK;
		break;
#ifdef CONFIG_HOTPLUG_CPU
	case CPU_DEAD:
312
	case CPU_DEAD_FROZEN:
L
Linus Torvalds 已提交
313
	case CPU_UP_CANCELED:
314
	case CPU_UP_CANCELED_FROZEN:
L
Linus Torvalds 已提交
315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330
		unmap_cpu_from_node(lcpu);
		break;
		ret = NOTIFY_OK;
#endif
	}
	return ret;
}

/*
 * Check and possibly modify a memory region to enforce the memory limit.
 *
 * Returns the size the region should have to enforce the memory limit.
 * This will either be the original value of size, a truncated value,
 * or zero. If the returned value of size is 0 the region should be
 * discarded as it lies wholy above the memory limit.
 */
331 332
static unsigned long __init numa_enforce_memory_limit(unsigned long start,
						      unsigned long size)
L
Linus Torvalds 已提交
333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351
{
	/*
	 * We use lmb_end_of_DRAM() in here instead of memory_limit because
	 * we've already adjusted it for the limit and it takes care of
	 * having memory holes below the limit.
	 */

	if (! memory_limit)
		return size;

	if (start + size <= lmb_end_of_DRAM())
		return size;

	if (start >= lmb_end_of_DRAM())
		return 0;

	return lmb_end_of_DRAM() - start;
}

352 353 354 355 356 357 358 359 360
/*
 * Extract NUMA information from the ibm,dynamic-reconfiguration-memory
 * node.  This assumes n_mem_{addr,size}_cells have been set.
 */
static void __init parse_drconf_memory(struct device_node *memory)
{
	const unsigned int *lm, *dm, *aa;
	unsigned int ls, ld, la;
	unsigned int n, aam, aalen;
361
	unsigned long lmb_size, size, start;
362
	int nid, default_nid = 0;
363
	unsigned int ai, flags;
364

365 366 367
	lm = of_get_property(memory, "ibm,lmb-size", &ls);
	dm = of_get_property(memory, "ibm,dynamic-memory", &ld);
	aa = of_get_property(memory, "ibm,associativity-lookup-arrays", &la);
368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397
	if (!lm || !dm || !aa ||
	    ls < sizeof(unsigned int) || ld < sizeof(unsigned int) ||
	    la < 2 * sizeof(unsigned int))
		return;

	lmb_size = read_n_cells(n_mem_size_cells, &lm);
	n = *dm++;		/* number of LMBs */
	aam = *aa++;		/* number of associativity lists */
	aalen = *aa++;		/* length of each associativity list */
	if (ld < (n * (n_mem_addr_cells + 4) + 1) * sizeof(unsigned int) ||
	    la < (aam * aalen + 2) * sizeof(unsigned int))
		return;

	for (; n != 0; --n) {
		start = read_n_cells(n_mem_addr_cells, &dm);
		ai = dm[2];
		flags = dm[3];
		dm += 4;
		/* 0x80 == reserved, 0x8 = assigned to us */
		if ((flags & 0x80) || !(flags & 0x8))
			continue;
		nid = default_nid;
		/* flags & 0x40 means associativity index is invalid */
		if (min_common_depth > 0 && min_common_depth <= aalen &&
		    (flags & 0x40) == 0 && ai < aam) {
			/* this is like of_node_to_nid_single */
			nid = aa[ai * aalen + min_common_depth - 1];
			if (nid == 0xffff || nid >= MAX_NUMNODES)
				nid = default_nid;
		}
398 399 400

		fake_numa_create_new_node(((start + lmb_size) >> PAGE_SHIFT),
						&nid);
401 402 403 404 405 406 407 408 409 410 411
		node_set_online(nid);

		size = numa_enforce_memory_limit(start, lmb_size);
		if (!size)
			continue;

		add_active_range(nid, start >> PAGE_SHIFT,
				 (start >> PAGE_SHIFT) + (size >> PAGE_SHIFT));
	}
}

L
Linus Torvalds 已提交
412 413 414 415
static int __init parse_numa_properties(void)
{
	struct device_node *cpu = NULL;
	struct device_node *memory = NULL;
416
	int default_nid = 0;
L
Linus Torvalds 已提交
417 418 419 420 421 422 423 424 425 426 427 428
	unsigned long i;

	if (numa_enabled == 0) {
		printk(KERN_WARNING "NUMA disabled by user\n");
		return -1;
	}

	min_common_depth = find_min_common_depth();

	if (min_common_depth < 0)
		return min_common_depth;

429 430
	dbg("NUMA associativity depth for CPU/Memory: %d\n", min_common_depth);

L
Linus Torvalds 已提交
431
	/*
432 433 434
	 * Even though we connect cpus to numa domains later in SMP
	 * init, we need to know the node ids now. This is because
	 * each node to be onlined must have NODE_DATA etc backing it.
L
Linus Torvalds 已提交
435
	 */
436
	for_each_present_cpu(i) {
437
		int nid;
L
Linus Torvalds 已提交
438 439

		cpu = find_cpu_node(i);
440
		BUG_ON(!cpu);
441
		nid = of_node_to_nid_single(cpu);
442
		of_node_put(cpu);
L
Linus Torvalds 已提交
443

444 445 446 447 448 449 450 451
		/*
		 * Don't fall back to default_nid yet -- we will plug
		 * cpus into nodes once the memory scan has discovered
		 * the topology.
		 */
		if (nid < 0)
			continue;
		node_set_online(nid);
L
Linus Torvalds 已提交
452 453
	}

454
	get_n_mem_cells(&n_mem_addr_cells, &n_mem_size_cells);
L
Linus Torvalds 已提交
455 456 457 458
	memory = NULL;
	while ((memory = of_find_node_by_type(memory, "memory")) != NULL) {
		unsigned long start;
		unsigned long size;
459
		int nid;
L
Linus Torvalds 已提交
460
		int ranges;
461
		const unsigned int *memcell_buf;
L
Linus Torvalds 已提交
462 463
		unsigned int len;

464
		memcell_buf = of_get_property(memory,
465 466
			"linux,usable-memory", &len);
		if (!memcell_buf || len <= 0)
467
			memcell_buf = of_get_property(memory, "reg", &len);
L
Linus Torvalds 已提交
468 469 470
		if (!memcell_buf || len <= 0)
			continue;

471 472
		/* ranges in cell */
		ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells);
L
Linus Torvalds 已提交
473 474
new_range:
		/* these are order-sensitive, and modify the buffer pointer */
475 476
		start = read_n_cells(n_mem_addr_cells, &memcell_buf);
		size = read_n_cells(n_mem_size_cells, &memcell_buf);
L
Linus Torvalds 已提交
477

478 479 480 481 482
		/*
		 * Assumption: either all memory nodes or none will
		 * have associativity properties.  If none, then
		 * everything goes to default_nid.
		 */
483
		nid = of_node_to_nid_single(memory);
484 485
		if (nid < 0)
			nid = default_nid;
486 487

		fake_numa_create_new_node(((start + size) >> PAGE_SHIFT), &nid);
488
		node_set_online(nid);
L
Linus Torvalds 已提交
489

490
		if (!(size = numa_enforce_memory_limit(start, size))) {
L
Linus Torvalds 已提交
491 492 493 494 495 496
			if (--ranges)
				goto new_range;
			else
				continue;
		}

497 498
		add_active_range(nid, start >> PAGE_SHIFT,
				(start >> PAGE_SHIFT) + (size >> PAGE_SHIFT));
L
Linus Torvalds 已提交
499 500 501 502 503

		if (--ranges)
			goto new_range;
	}

504 505 506 507 508 509 510 511
	/*
	 * Now do the same thing for each LMB listed in the ibm,dynamic-memory
	 * property in the ibm,dynamic-reconfiguration-memory node.
	 */
	memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory");
	if (memory)
		parse_drconf_memory(memory);

L
Linus Torvalds 已提交
512 513 514 515 516 517 518
	return 0;
}

static void __init setup_nonnuma(void)
{
	unsigned long top_of_ram = lmb_end_of_DRAM();
	unsigned long total_ram = lmb_phys_mem_size();
519
	unsigned long start_pfn, end_pfn;
520
	unsigned int i, nid = 0;
L
Linus Torvalds 已提交
521

522
	printk(KERN_DEBUG "Top of RAM: 0x%lx, Total RAM: 0x%lx\n",
L
Linus Torvalds 已提交
523
	       top_of_ram, total_ram);
524
	printk(KERN_DEBUG "Memory hole size: %ldMB\n",
L
Linus Torvalds 已提交
525 526
	       (top_of_ram - total_ram) >> 20);

527 528 529
	for (i = 0; i < lmb.memory.cnt; ++i) {
		start_pfn = lmb.memory.region[i].base >> PAGE_SHIFT;
		end_pfn = start_pfn + lmb_size_pages(&lmb.memory, i);
530 531 532 533

		fake_numa_create_new_node(end_pfn, &nid);
		add_active_range(nid, start_pfn, end_pfn);
		node_set_online(nid);
534
	}
L
Linus Torvalds 已提交
535 536
}

537 538 539 540 541 542 543 544 545
void __init dump_numa_cpu_topology(void)
{
	unsigned int node;
	unsigned int cpu, count;

	if (min_common_depth == -1 || !numa_enabled)
		return;

	for_each_online_node(node) {
546
		printk(KERN_DEBUG "Node %d CPUs:", node);
547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571

		count = 0;
		/*
		 * If we used a CPU iterator here we would miss printing
		 * the holes in the cpumap.
		 */
		for (cpu = 0; cpu < NR_CPUS; cpu++) {
			if (cpu_isset(cpu, numa_cpumask_lookup_table[node])) {
				if (count == 0)
					printk(" %u", cpu);
				++count;
			} else {
				if (count > 1)
					printk("-%u", cpu - 1);
				count = 0;
			}
		}

		if (count > 1)
			printk("-%u", NR_CPUS - 1);
		printk("\n");
	}
}

static void __init dump_numa_memory_topology(void)
L
Linus Torvalds 已提交
572 573 574 575 576 577 578 579 580 581
{
	unsigned int node;
	unsigned int count;

	if (min_common_depth == -1 || !numa_enabled)
		return;

	for_each_online_node(node) {
		unsigned long i;

582
		printk(KERN_DEBUG "Node %d Memory:", node);
L
Linus Torvalds 已提交
583 584 585

		count = 0;

586 587 588
		for (i = 0; i < lmb_end_of_DRAM();
		     i += (1 << SECTION_SIZE_BITS)) {
			if (early_pfn_to_nid(i >> PAGE_SHIFT) == node) {
L
Linus Torvalds 已提交
589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611
				if (count == 0)
					printk(" 0x%lx", i);
				++count;
			} else {
				if (count > 0)
					printk("-0x%lx", i);
				count = 0;
			}
		}

		if (count > 0)
			printk("-0x%lx", i);
		printk("\n");
	}
}

/*
 * Allocate some memory, satisfying the lmb or bootmem allocator where
 * required. nid is the preferred node and end is the physical address of
 * the highest address in the node.
 *
 * Returns the physical address of the memory.
 */
612 613 614
static void __init *careful_allocation(int nid, unsigned long size,
				       unsigned long align,
				       unsigned long end_pfn)
L
Linus Torvalds 已提交
615
{
616
	int new_nid;
617
	unsigned long ret = __lmb_alloc_base(size, align, end_pfn << PAGE_SHIFT);
L
Linus Torvalds 已提交
618 619 620

	/* retry over all memory */
	if (!ret)
621
		ret = __lmb_alloc_base(size, align, lmb_end_of_DRAM());
L
Linus Torvalds 已提交
622 623 624 625 626 627 628 629 630

	if (!ret)
		panic("numa.c: cannot allocate %lu bytes on node %d",
		      size, nid);

	/*
	 * If the memory came from a previously allocated node, we must
	 * retry with the bootmem allocator.
	 */
631 632 633
	new_nid = early_pfn_to_nid(ret >> PAGE_SHIFT);
	if (new_nid < nid) {
		ret = (unsigned long)__alloc_bootmem_node(NODE_DATA(new_nid),
L
Linus Torvalds 已提交
634 635 636 637
				size, align, 0);

		if (!ret)
			panic("numa.c: cannot allocate %lu bytes on node %d",
638
			      size, new_nid);
L
Linus Torvalds 已提交
639

640
		ret = __pa(ret);
L
Linus Torvalds 已提交
641 642 643 644

		dbg("alloc_bootmem %lx %lx\n", ret, size);
	}

645
	return (void *)ret;
L
Linus Torvalds 已提交
646 647
}

648 649 650 651 652
static struct notifier_block __cpuinitdata ppc64_numa_nb = {
	.notifier_call = cpu_numa_callback,
	.priority = 1 /* Must run before sched domains notifier. */
};

L
Linus Torvalds 已提交
653 654 655
void __init do_init_bootmem(void)
{
	int nid;
656
	unsigned int i;
L
Linus Torvalds 已提交
657 658 659 660 661 662 663 664

	min_low_pfn = 0;
	max_low_pfn = lmb_end_of_DRAM() >> PAGE_SHIFT;
	max_pfn = max_low_pfn;

	if (parse_numa_properties())
		setup_nonnuma();
	else
665
		dump_numa_memory_topology();
L
Linus Torvalds 已提交
666 667

	register_cpu_notifier(&ppc64_numa_nb);
668 669
	cpu_numa_callback(&ppc64_numa_nb, CPU_UP_PREPARE,
			  (void *)(unsigned long)boot_cpuid);
L
Linus Torvalds 已提交
670 671

	for_each_online_node(nid) {
672
		unsigned long start_pfn, end_pfn;
L
Linus Torvalds 已提交
673 674 675
		unsigned long bootmem_paddr;
		unsigned long bootmap_pages;

676
		get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
L
Linus Torvalds 已提交
677 678

		/* Allocate the node structure node local if possible */
679
		NODE_DATA(nid) = careful_allocation(nid,
L
Linus Torvalds 已提交
680
					sizeof(struct pglist_data),
681 682
					SMP_CACHE_BYTES, end_pfn);
		NODE_DATA(nid) = __va(NODE_DATA(nid));
L
Linus Torvalds 已提交
683 684 685 686 687 688
		memset(NODE_DATA(nid), 0, sizeof(struct pglist_data));

  		dbg("node %d\n", nid);
		dbg("NODE_DATA() = %p\n", NODE_DATA(nid));

		NODE_DATA(nid)->bdata = &plat_node_bdata[nid];
689 690
		NODE_DATA(nid)->node_start_pfn = start_pfn;
		NODE_DATA(nid)->node_spanned_pages = end_pfn - start_pfn;
L
Linus Torvalds 已提交
691 692 693 694

		if (NODE_DATA(nid)->node_spanned_pages == 0)
  			continue;

695 696
  		dbg("start_paddr = %lx\n", start_pfn << PAGE_SHIFT);
  		dbg("end_paddr = %lx\n", end_pfn << PAGE_SHIFT);
L
Linus Torvalds 已提交
697

698 699 700 701 702
		bootmap_pages = bootmem_bootmap_pages(end_pfn - start_pfn);
		bootmem_paddr = (unsigned long)careful_allocation(nid,
					bootmap_pages << PAGE_SHIFT,
					PAGE_SIZE, end_pfn);
		memset(__va(bootmem_paddr), 0, bootmap_pages << PAGE_SHIFT);
L
Linus Torvalds 已提交
703 704 705 706

		dbg("bootmap_paddr = %lx\n", bootmem_paddr);

		init_bootmem_node(NODE_DATA(nid), bootmem_paddr >> PAGE_SHIFT,
707
				  start_pfn, end_pfn);
L
Linus Torvalds 已提交
708

709
		free_bootmem_with_active_regions(nid, end_pfn);
L
Linus Torvalds 已提交
710

711
		/* Mark reserved regions on this node */
L
Linus Torvalds 已提交
712
		for (i = 0; i < lmb.reserved.cnt; i++) {
713
			unsigned long physbase = lmb.reserved.region[i].base;
L
Linus Torvalds 已提交
714
			unsigned long size = lmb.reserved.region[i].size;
715 716
			unsigned long start_paddr = start_pfn << PAGE_SHIFT;
			unsigned long end_paddr = end_pfn << PAGE_SHIFT;
L
Linus Torvalds 已提交
717

718 719
			if (early_pfn_to_nid(physbase >> PAGE_SHIFT) != nid &&
			    early_pfn_to_nid((physbase+size-1) >> PAGE_SHIFT) != nid)
L
Linus Torvalds 已提交
720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735
				continue;

			if (physbase < end_paddr &&
			    (physbase+size) > start_paddr) {
				/* overlaps */
				if (physbase < start_paddr) {
					size -= start_paddr - physbase;
					physbase = start_paddr;
				}

				if (size > end_paddr - physbase)
					size = end_paddr - physbase;

				dbg("reserve_bootmem %lx %lx\n", physbase,
				    size);
				reserve_bootmem_node(NODE_DATA(nid), physbase,
736
						     size, BOOTMEM_DEFAULT);
L
Linus Torvalds 已提交
737 738
			}
		}
B
Bob Picco 已提交
739

740
		sparse_memory_present_with_active_regions(nid);
L
Linus Torvalds 已提交
741 742 743 744 745
	}
}

void __init paging_init(void)
{
746 747 748
	unsigned long max_zone_pfns[MAX_NR_ZONES];
	memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
	max_zone_pfns[ZONE_DMA] = lmb_end_of_DRAM() >> PAGE_SHIFT;
749
	free_area_init_nodes(max_zone_pfns);
L
Linus Torvalds 已提交
750 751 752 753 754 755 756 757 758 759 760 761 762
}

static int __init early_numa(char *p)
{
	if (!p)
		return 0;

	if (strstr(p, "off"))
		numa_enabled = 0;

	if (strstr(p, "debug"))
		numa_debug = 1;

763 764 765 766
	p = strstr(p, "fake=");
	if (p)
		cmdline = p + strlen("fake=");

L
Linus Torvalds 已提交
767 768 769
	return 0;
}
early_param("numa", early_numa);
770 771 772 773 774 775 776 777 778 779

#ifdef CONFIG_MEMORY_HOTPLUG
/*
 * Find the node associated with a hot added memory section.  Section
 * corresponds to a SPARSEMEM section, not an LMB.  It is assumed that
 * sections are fully contained within a single LMB.
 */
int hot_add_scn_to_nid(unsigned long scn_addr)
{
	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;
786 787 788

	while ((memory = of_find_node_by_type(memory, "memory")) != NULL) {
		unsigned long start, size;
789
		int ranges;
790
		const unsigned int *memcell_buf;
791 792
		unsigned int len;

793
		memcell_buf = of_get_property(memory, "reg", &len);
794 795 796
		if (!memcell_buf || len <= 0)
			continue;

797 798
		/* ranges in cell */
		ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells);
799 800 801
ha_new_range:
		start = read_n_cells(n_mem_addr_cells, &memcell_buf);
		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 */
805 806
		if (nid < 0 || !node_online(nid))
			nid = default_nid;
807 808 809

		if ((scn_addr >= start) && (scn_addr < (start + size))) {
			of_node_put(memory);
810
			goto got_nid;
811 812 813 814 815 816
		}

		if (--ranges)		/* process all ranges in cell */
			goto ha_new_range;
	}
	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);
822 823 824
	while (NODE_DATA(nid)->node_spanned_pages == 0) {
		node_clear(nid, nodes);
		nid = any_online_node(nodes);
825
	}
826
	return nid;
827 828
}
#endif /* CONFIG_MEMORY_HOTPLUG */