memory.c 17.3 KB
Newer Older
1
/*
2
 * Memory subsystem support
3 4 5 6 7 8 9 10 11 12 13 14 15
 *
 * Written by Matt Tolentino <matthew.e.tolentino@intel.com>
 *            Dave Hansen <haveblue@us.ibm.com>
 *
 * This file provides the necessary infrastructure to represent
 * a SPARSEMEM-memory-model system's physical memory in /sysfs.
 * All arch-independent code that assumes MEMORY_HOTPLUG requires
 * SPARSEMEM should be contained here, or in mm/memory_hotplug.c.
 */

#include <linux/module.h>
#include <linux/init.h>
#include <linux/topology.h>
16
#include <linux/capability.h>
17 18 19 20 21
#include <linux/device.h>
#include <linux/memory.h>
#include <linux/kobject.h>
#include <linux/memory_hotplug.h>
#include <linux/mm.h>
22
#include <linux/mutex.h>
23
#include <linux/stat.h>
24
#include <linux/slab.h>
25

Arun Sharma's avatar
Arun Sharma committed
26
#include <linux/atomic.h>
27 28
#include <asm/uaccess.h>

29 30
static DEFINE_MUTEX(mem_sysfs_mutex);

31
#define MEMORY_CLASS_NAME	"memory"
32 33 34 35 36 37 38

static int sections_per_block;

static inline int base_memory_block_id(int section_nr)
{
	return section_nr / sections_per_block;
}
39

40
static struct bus_type memory_subsys = {
41
	.name = MEMORY_CLASS_NAME,
42
	.dev_name = MEMORY_CLASS_NAME,
43 44
};

45
static BLOCKING_NOTIFIER_HEAD(memory_chain);
46

47
int register_memory_notifier(struct notifier_block *nb)
48
{
49
        return blocking_notifier_chain_register(&memory_chain, nb);
50
}
51
EXPORT_SYMBOL(register_memory_notifier);
52

53
void unregister_memory_notifier(struct notifier_block *nb)
54
{
55
        blocking_notifier_chain_unregister(&memory_chain, nb);
56
}
57
EXPORT_SYMBOL(unregister_memory_notifier);
58

59 60 61 62 63 64 65 66 67 68 69 70 71 72
static ATOMIC_NOTIFIER_HEAD(memory_isolate_chain);

int register_memory_isolate_notifier(struct notifier_block *nb)
{
	return atomic_notifier_chain_register(&memory_isolate_chain, nb);
}
EXPORT_SYMBOL(register_memory_isolate_notifier);

void unregister_memory_isolate_notifier(struct notifier_block *nb)
{
	atomic_notifier_chain_unregister(&memory_isolate_chain, nb);
}
EXPORT_SYMBOL(unregister_memory_isolate_notifier);

73 74 75
/*
 * register_memory - Setup a sysfs device for a memory block
 */
76
static
77
int register_memory(struct memory_block *memory)
78 79 80
{
	int error;

81 82
	memory->dev.bus = &memory_subsys;
	memory->dev.id = memory->start_section_nr / sections_per_block;
83

84
	error = device_register(&memory->dev);
85 86 87 88
	return error;
}

static void
89
unregister_memory(struct memory_block *memory)
90
{
91
	BUG_ON(memory->dev.bus != &memory_subsys);
92

93
	/* drop the ref. we got in remove_memory_block() */
94 95
	kobject_put(&memory->dev.kobj);
	device_unregister(&memory->dev);
96 97
}

98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117
unsigned long __weak memory_block_size_bytes(void)
{
	return MIN_MEMORY_BLOCK_SIZE;
}

static unsigned long get_memory_block_size(void)
{
	unsigned long block_sz;

	block_sz = memory_block_size_bytes();

	/* Validate blk_sz is a power of 2 and not less than section size */
	if ((block_sz & (block_sz - 1)) || (block_sz < MIN_MEMORY_BLOCK_SIZE)) {
		WARN_ON(1);
		block_sz = MIN_MEMORY_BLOCK_SIZE;
	}

	return block_sz;
}

118 119 120 121 122
/*
 * use this as the physical section index that this memsection
 * uses.
 */

123 124
static ssize_t show_mem_start_phys_index(struct device *dev,
			struct device_attribute *attr, char *buf)
125 126
{
	struct memory_block *mem =
127
		container_of(dev, struct memory_block, dev);
128 129 130 131 132 133
	unsigned long phys_index;

	phys_index = mem->start_section_nr / sections_per_block;
	return sprintf(buf, "%08lx\n", phys_index);
}

134 135
static ssize_t show_mem_end_phys_index(struct device *dev,
			struct device_attribute *attr, char *buf)
136 137
{
	struct memory_block *mem =
138
		container_of(dev, struct memory_block, dev);
139 140 141 142
	unsigned long phys_index;

	phys_index = mem->end_section_nr / sections_per_block;
	return sprintf(buf, "%08lx\n", phys_index);
143 144
}

145 146 147
/*
 * Show whether the section of memory is likely to be hot-removable
 */
148 149
static ssize_t show_mem_removable(struct device *dev,
			struct device_attribute *attr, char *buf)
150
{
151 152
	unsigned long i, pfn;
	int ret = 1;
153
	struct memory_block *mem =
154
		container_of(dev, struct memory_block, dev);
155

156
	for (i = 0; i < sections_per_block; i++) {
157
		pfn = section_nr_to_pfn(mem->start_section_nr + i);
158 159 160
		ret &= is_mem_section_removable(pfn, PAGES_PER_SECTION);
	}

161 162 163
	return sprintf(buf, "%d\n", ret);
}

164 165 166
/*
 * online, offline, going offline, etc.
 */
167 168
static ssize_t show_mem_state(struct device *dev,
			struct device_attribute *attr, char *buf)
169 170
{
	struct memory_block *mem =
171
		container_of(dev, struct memory_block, dev);
172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197
	ssize_t len = 0;

	/*
	 * We can probably put these states in a nice little array
	 * so that they're not open-coded
	 */
	switch (mem->state) {
		case MEM_ONLINE:
			len = sprintf(buf, "online\n");
			break;
		case MEM_OFFLINE:
			len = sprintf(buf, "offline\n");
			break;
		case MEM_GOING_OFFLINE:
			len = sprintf(buf, "going-offline\n");
			break;
		default:
			len = sprintf(buf, "ERROR-UNKNOWN-%ld\n",
					mem->state);
			WARN_ON(1);
			break;
	}

	return len;
}

198
int memory_notify(unsigned long val, void *v)
199
{
200
	return blocking_notifier_call_chain(&memory_chain, val, v);
201 202
}

203 204 205 206 207
int memory_isolate_notify(unsigned long val, void *v)
{
	return atomic_notifier_call_chain(&memory_isolate_chain, val, v);
}

208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243
/*
 * The probe routines leave the pages reserved, just as the bootmem code does.
 * Make sure they're still that way.
 */
static bool pages_correctly_reserved(unsigned long start_pfn,
					unsigned long nr_pages)
{
	int i, j;
	struct page *page;
	unsigned long pfn = start_pfn;

	/*
	 * memmap between sections is not contiguous except with
	 * SPARSEMEM_VMEMMAP. We lookup the page once per section
	 * and assume memmap is contiguous within each section
	 */
	for (i = 0; i < sections_per_block; i++, pfn += PAGES_PER_SECTION) {
		if (WARN_ON_ONCE(!pfn_valid(pfn)))
			return false;
		page = pfn_to_page(pfn);

		for (j = 0; j < PAGES_PER_SECTION; j++) {
			if (PageReserved(page + j))
				continue;

			printk(KERN_WARNING "section number %ld page number %d "
				"not reserved, was it already online?\n",
				pfn_to_section_nr(pfn), j);

			return false;
		}
	}

	return true;
}

244 245 246 247 248
/*
 * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is
 * OK to have direct references to sparsemem variables in here.
 */
static int
249
memory_block_action(unsigned long phys_index, unsigned long action)
250
{
251
	unsigned long start_pfn;
252
	unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
253
	struct page *first_page;
254 255
	int ret;

256
	first_page = pfn_to_page(phys_index << PFN_SECTION_SHIFT);
257
	start_pfn = page_to_pfn(first_page);
258

259 260
	switch (action) {
		case MEM_ONLINE:
261 262 263
			if (!pages_correctly_reserved(start_pfn, nr_pages))
				return -EBUSY;

264
			ret = online_pages(start_pfn, nr_pages);
265 266
			break;
		case MEM_OFFLINE:
267
			ret = offline_pages(start_pfn, nr_pages);
268 269
			break;
		default:
270 271
			WARN(1, KERN_WARNING "%s(%ld, %ld) unknown action: "
			     "%ld\n", __func__, phys_index, action, action);
272 273 274 275 276 277
			ret = -EINVAL;
	}

	return ret;
}

278
static int __memory_block_change_state(struct memory_block *mem,
279 280
		unsigned long to_state, unsigned long from_state_req)
{
281
	int ret = 0;
282

283 284 285 286 287
	if (mem->state != from_state_req) {
		ret = -EINVAL;
		goto out;
	}

288 289 290
	if (to_state == MEM_OFFLINE)
		mem->state = MEM_GOING_OFFLINE;

291
	ret = memory_block_action(mem->start_section_nr, to_state);
292

293
	if (ret) {
294
		mem->state = from_state_req;
295 296
		goto out;
	}
297

298 299 300 301 302 303 304 305 306 307 308
	mem->state = to_state;
	switch (mem->state) {
	case MEM_OFFLINE:
		kobject_uevent(&mem->dev.kobj, KOBJ_OFFLINE);
		break;
	case MEM_ONLINE:
		kobject_uevent(&mem->dev.kobj, KOBJ_ONLINE);
		break;
	default:
		break;
	}
309 310 311 312
out:
	return ret;
}

313 314 315 316 317 318 319 320 321 322 323
static int memory_block_change_state(struct memory_block *mem,
		unsigned long to_state, unsigned long from_state_req)
{
	int ret;

	mutex_lock(&mem->state_mutex);
	ret = __memory_block_change_state(mem, to_state, from_state_req);
	mutex_unlock(&mem->state_mutex);

	return ret;
}
324
static ssize_t
325 326
store_mem_state(struct device *dev,
		struct device_attribute *attr, const char *buf, size_t count)
327 328 329 330
{
	struct memory_block *mem;
	int ret = -EINVAL;

331
	mem = container_of(dev, struct memory_block, dev);
332 333 334 335 336

	if (!strncmp(buf, "online", min((int)count, 6)))
		ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE);
	else if(!strncmp(buf, "offline", min((int)count, 7)))
		ret = memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE);
337

338 339 340 341 342 343 344 345 346 347 348 349 350 351
	if (ret)
		return ret;
	return count;
}

/*
 * phys_device is a bad name for this.  What I really want
 * is a way to differentiate between memory ranges that
 * are part of physical devices that constitute
 * a complete removable unit or fru.
 * i.e. do these ranges belong to the same physical device,
 * s.t. if I offline all of these sections I can then
 * remove the physical device?
 */
352 353
static ssize_t show_phys_device(struct device *dev,
				struct device_attribute *attr, char *buf)
354 355
{
	struct memory_block *mem =
356
		container_of(dev, struct memory_block, dev);
357 358 359
	return sprintf(buf, "%d\n", mem->phys_device);
}

360 361 362 363 364
static DEVICE_ATTR(phys_index, 0444, show_mem_start_phys_index, NULL);
static DEVICE_ATTR(end_phys_index, 0444, show_mem_end_phys_index, NULL);
static DEVICE_ATTR(state, 0644, show_mem_state, store_mem_state);
static DEVICE_ATTR(phys_device, 0444, show_phys_device, NULL);
static DEVICE_ATTR(removable, 0444, show_mem_removable, NULL);
365 366

#define mem_create_simple_file(mem, attr_name)	\
367
	device_create_file(&mem->dev, &dev_attr_##attr_name)
368
#define mem_remove_simple_file(mem, attr_name)	\
369
	device_remove_file(&mem->dev, &dev_attr_##attr_name)
370 371 372 373 374

/*
 * Block size attribute stuff
 */
static ssize_t
375
print_block_size(struct device *dev, struct device_attribute *attr,
376
		 char *buf)
377
{
378
	return sprintf(buf, "%lx\n", get_memory_block_size());
379 380
}

381
static DEVICE_ATTR(block_size_bytes, 0444, print_block_size, NULL);
382 383 384

static int block_size_init(void)
{
385 386
	return device_create_file(memory_subsys.dev_root,
				  &dev_attr_block_size_bytes);
387 388 389 390 391 392 393 394 395 396
}

/*
 * Some architectures will have custom drivers to do this, and
 * will not need to do it from userspace.  The fake hot-add code
 * as well as ppc64 will do all of their discovery in userspace
 * and will require this interface.
 */
#ifdef CONFIG_ARCH_MEMORY_PROBE
static ssize_t
397
memory_probe_store(struct device *dev, struct device_attribute *attr,
398
		   const char *buf, size_t count)
399 400
{
	u64 phys_addr;
401
	int nid;
402
	int i, ret;
403
	unsigned long pages_per_block = PAGES_PER_SECTION * sections_per_block;
404 405 406

	phys_addr = simple_strtoull(buf, NULL, 0);

407 408 409
	if (phys_addr & ((pages_per_block << PAGE_SHIFT) - 1))
		return -EINVAL;

410 411 412 413 414
	for (i = 0; i < sections_per_block; i++) {
		nid = memory_add_physaddr_to_nid(phys_addr);
		ret = add_memory(nid, phys_addr,
				 PAGES_PER_SECTION << PAGE_SHIFT);
		if (ret)
415
			goto out;
416 417 418

		phys_addr += MIN_MEMORY_BLOCK_SIZE;
	}
419

420 421 422
	ret = count;
out:
	return ret;
423
}
424
static DEVICE_ATTR(probe, S_IWUSR, NULL, memory_probe_store);
425 426 427

static int memory_probe_init(void)
{
428
	return device_create_file(memory_subsys.dev_root, &dev_attr_probe);
429 430
}
#else
431 432 433 434
static inline int memory_probe_init(void)
{
	return 0;
}
435 436
#endif

437 438 439 440 441 442 443
#ifdef CONFIG_MEMORY_FAILURE
/*
 * Support for offlining pages of memory
 */

/* Soft offline a page */
static ssize_t
444 445
store_soft_offline_page(struct device *dev,
			struct device_attribute *attr,
446
			const char *buf, size_t count)
447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462
{
	int ret;
	u64 pfn;
	if (!capable(CAP_SYS_ADMIN))
		return -EPERM;
	if (strict_strtoull(buf, 0, &pfn) < 0)
		return -EINVAL;
	pfn >>= PAGE_SHIFT;
	if (!pfn_valid(pfn))
		return -ENXIO;
	ret = soft_offline_page(pfn_to_page(pfn), 0);
	return ret == 0 ? count : ret;
}

/* Forcibly offline a page, including killing processes. */
static ssize_t
463 464
store_hard_offline_page(struct device *dev,
			struct device_attribute *attr,
465
			const char *buf, size_t count)
466 467 468 469 470 471 472 473
{
	int ret;
	u64 pfn;
	if (!capable(CAP_SYS_ADMIN))
		return -EPERM;
	if (strict_strtoull(buf, 0, &pfn) < 0)
		return -EINVAL;
	pfn >>= PAGE_SHIFT;
474
	ret = memory_failure(pfn, 0, 0);
475 476 477
	return ret ? ret : count;
}

478 479
static DEVICE_ATTR(soft_offline_page, 0644, NULL, store_soft_offline_page);
static DEVICE_ATTR(hard_offline_page, 0644, NULL, store_hard_offline_page);
480 481 482 483 484

static __init int memory_fail_init(void)
{
	int err;

485 486
	err = device_create_file(memory_subsys.dev_root,
				&dev_attr_soft_offline_page);
487
	if (!err)
488 489
		err = device_create_file(memory_subsys.dev_root,
				&dev_attr_hard_offline_page);
490 491 492 493 494 495 496 497 498
	return err;
}
#else
static inline int memory_fail_init(void)
{
	return 0;
}
#endif

499 500 501 502 503
/*
 * Note that phys_device is optional.  It is here to allow for
 * differentiation between which *physical* devices each
 * section belongs to...
 */
504 505 506 507
int __weak arch_get_memory_phys_device(unsigned long start_pfn)
{
	return 0;
}
508

509 510 511 512
/*
 * A reference for the returned object is held and the reference for the
 * hinted object is released.
 */
513 514
struct memory_block *find_memory_block_hinted(struct mem_section *section,
					      struct memory_block *hint)
515
{
516
	int block_id = base_memory_block_id(__section_nr(section));
517 518
	struct device *hintdev = hint ? &hint->dev : NULL;
	struct device *dev;
519

520 521 522 523
	dev = subsys_find_device_by_id(&memory_subsys, block_id, hintdev);
	if (hint)
		put_device(&hint->dev);
	if (!dev)
524
		return NULL;
525
	return container_of(dev, struct memory_block, dev);
526 527
}

528 529 530 531 532 533
/*
 * For now, we have a linear search to go find the appropriate
 * memory_block corresponding to a particular phys_index. If
 * this gets to be a real problem, we can always use a radix
 * tree or something here.
 *
534
 * This could be made generic for all device subsystems.
535 536 537 538 539 540
 */
struct memory_block *find_memory_block(struct mem_section *section)
{
	return find_memory_block_hinted(section, NULL);
}

541 542
static int init_memory_block(struct memory_block **memory,
			     struct mem_section *section, unsigned long state)
543
{
544
	struct memory_block *mem;
545
	unsigned long start_pfn;
546
	int scn_nr;
547 548
	int ret = 0;

549
	mem = kzalloc(sizeof(*mem), GFP_KERNEL);
550 551 552
	if (!mem)
		return -ENOMEM;

553
	scn_nr = __section_nr(section);
554 555 556
	mem->start_section_nr =
			base_memory_block_id(scn_nr) * sections_per_block;
	mem->end_section_nr = mem->start_section_nr + sections_per_block - 1;
557
	mem->state = state;
558
	mem->section_count++;
559
	mutex_init(&mem->state_mutex);
560
	start_pfn = section_nr_to_pfn(mem->start_section_nr);
561 562
	mem->phys_device = arch_get_memory_phys_device(start_pfn);

563
	ret = register_memory(mem);
564 565
	if (!ret)
		ret = mem_create_simple_file(mem, phys_index);
566 567
	if (!ret)
		ret = mem_create_simple_file(mem, end_phys_index);
568 569 570 571 572 573
	if (!ret)
		ret = mem_create_simple_file(mem, state);
	if (!ret)
		ret = mem_create_simple_file(mem, phys_device);
	if (!ret)
		ret = mem_create_simple_file(mem, removable);
574 575 576 577 578 579

	*memory = mem;
	return ret;
}

static int add_memory_section(int nid, struct mem_section *section,
580
			struct memory_block **mem_p,
581 582
			unsigned long state, enum mem_add_context context)
{
583 584
	struct memory_block *mem = NULL;
	int scn_nr = __section_nr(section);
585 586 587 588
	int ret = 0;

	mutex_lock(&mem_sysfs_mutex);

589 590 591 592 593 594 595 596 597 598 599
	if (context == BOOT) {
		/* same memory block ? */
		if (mem_p && *mem_p)
			if (scn_nr >= (*mem_p)->start_section_nr &&
			    scn_nr <= (*mem_p)->end_section_nr) {
				mem = *mem_p;
				kobject_get(&mem->dev.kobj);
			}
	} else
		mem = find_memory_block(section);

600 601
	if (mem) {
		mem->section_count++;
602
		kobject_put(&mem->dev.kobj);
603
	} else {
604
		ret = init_memory_block(&mem, section, state);
605 606 607 608 609
		/* store memory_block pointer for next loop */
		if (!ret && context == BOOT)
			if (mem_p)
				*mem_p = mem;
	}
610

611
	if (!ret) {
612 613
		if (context == HOTPLUG &&
		    mem->section_count == sections_per_block)
614 615 616
			ret = register_mem_sect_under_node(mem, nid);
	}

617
	mutex_unlock(&mem_sysfs_mutex);
618 619 620
	return ret;
}

621 622 623 624 625
int remove_memory_block(unsigned long node_id, struct mem_section *section,
		int phys_device)
{
	struct memory_block *mem;

626
	mutex_lock(&mem_sysfs_mutex);
627
	mem = find_memory_block(section);
628
	unregister_mem_sect_under_nodes(mem, __section_nr(section));
629 630 631 632

	mem->section_count--;
	if (mem->section_count == 0) {
		mem_remove_simple_file(mem, phys_index);
633
		mem_remove_simple_file(mem, end_phys_index);
634 635 636
		mem_remove_simple_file(mem, state);
		mem_remove_simple_file(mem, phys_device);
		mem_remove_simple_file(mem, removable);
637 638 639
		unregister_memory(mem);
		kfree(mem);
	} else
640
		kobject_put(&mem->dev.kobj);
641

642
	mutex_unlock(&mem_sysfs_mutex);
643 644 645 646 647 648 649
	return 0;
}

/*
 * need an interface for the VM to add new memory regions,
 * but without onlining it.
 */
650
int register_new_memory(int nid, struct mem_section *section)
651
{
652
	return add_memory_section(nid, section, NULL, MEM_OFFLINE, HOTPLUG);
653 654 655 656
}

int unregister_memory_section(struct mem_section *section)
{
657
	if (!present_section(section))
658 659 660 661 662
		return -EINVAL;

	return remove_memory_block(0, section, 0);
}

663 664 665 666 667 668 669 670 671 672 673 674 675 676 677
/*
 * offline one memory block. If the memory block has been offlined, do nothing.
 */
int offline_memory_block(struct memory_block *mem)
{
	int ret = 0;

	mutex_lock(&mem->state_mutex);
	if (mem->state != MEM_OFFLINE)
		ret = __memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE);
	mutex_unlock(&mem->state_mutex);

	return ret;
}

678 679 680 681 682 683 684
/*
 * Initialize the sysfs support for memory devices...
 */
int __init memory_dev_init(void)
{
	unsigned int i;
	int ret;
685
	int err;
686
	unsigned long block_sz;
687
	struct memory_block *mem = NULL;
688

689
	ret = subsys_system_register(&memory_subsys, NULL);
690 691
	if (ret)
		goto out;
692

693 694 695
	block_sz = get_memory_block_size();
	sections_per_block = block_sz / MIN_MEMORY_BLOCK_SIZE;

696 697 698 699 700
	/*
	 * Create entries for memory sections that were found
	 * during boot and have been initialized
	 */
	for (i = 0; i < NR_MEM_SECTIONS; i++) {
701
		if (!present_section_nr(i))
702
			continue;
703 704 705 706
		/* don't need to reuse memory_block if only one per block */
		err = add_memory_section(0, __nr_to_section(i),
				 (sections_per_block == 1) ? NULL : &mem,
					 MEM_ONLINE,
707
					 BOOT);
708 709
		if (!ret)
			ret = err;
710 711
	}

712
	err = memory_probe_init();
713 714 715
	if (!ret)
		ret = err;
	err = memory_fail_init();
716 717 718 719 720 721 722
	if (!ret)
		ret = err;
	err = block_size_init();
	if (!ret)
		ret = err;
out:
	if (ret)
723
		printk(KERN_ERR "%s() failed: %d\n", __func__, ret);
724 725
	return ret;
}