• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Memory subsystem support
3  *
4  * Written by Matt Tolentino <matthew.e.tolentino@intel.com>
5  *            Dave Hansen <haveblue@us.ibm.com>
6  *
7  * This file provides the necessary infrastructure to represent
8  * a SPARSEMEM-memory-model system's physical memory in /sysfs.
9  * All arch-independent code that assumes MEMORY_HOTPLUG requires
10  * SPARSEMEM should be contained here, or in mm/memory_hotplug.c.
11  */
12 
13 #include <linux/module.h>
14 #include <linux/init.h>
15 #include <linux/topology.h>
16 #include <linux/capability.h>
17 #include <linux/device.h>
18 #include <linux/memory.h>
19 #include <linux/memory_hotplug.h>
20 #include <linux/mm.h>
21 #include <linux/mutex.h>
22 #include <linux/stat.h>
23 #include <linux/slab.h>
24 
25 #include <linux/atomic.h>
26 #include <asm/uaccess.h>
27 
28 static DEFINE_MUTEX(mem_sysfs_mutex);
29 
30 #define MEMORY_CLASS_NAME	"memory"
31 
32 #define to_memory_block(dev) container_of(dev, struct memory_block, dev)
33 
34 static int sections_per_block;
35 
base_memory_block_id(int section_nr)36 static inline int base_memory_block_id(int section_nr)
37 {
38 	return section_nr / sections_per_block;
39 }
40 
41 static int memory_subsys_online(struct device *dev);
42 static int memory_subsys_offline(struct device *dev);
43 
44 static struct bus_type memory_subsys = {
45 	.name = MEMORY_CLASS_NAME,
46 	.dev_name = MEMORY_CLASS_NAME,
47 	.online = memory_subsys_online,
48 	.offline = memory_subsys_offline,
49 };
50 
51 static BLOCKING_NOTIFIER_HEAD(memory_chain);
52 
register_memory_notifier(struct notifier_block * nb)53 int register_memory_notifier(struct notifier_block *nb)
54 {
55 	return blocking_notifier_chain_register(&memory_chain, nb);
56 }
57 EXPORT_SYMBOL(register_memory_notifier);
58 
unregister_memory_notifier(struct notifier_block * nb)59 void unregister_memory_notifier(struct notifier_block *nb)
60 {
61 	blocking_notifier_chain_unregister(&memory_chain, nb);
62 }
63 EXPORT_SYMBOL(unregister_memory_notifier);
64 
65 static ATOMIC_NOTIFIER_HEAD(memory_isolate_chain);
66 
register_memory_isolate_notifier(struct notifier_block * nb)67 int register_memory_isolate_notifier(struct notifier_block *nb)
68 {
69 	return atomic_notifier_chain_register(&memory_isolate_chain, nb);
70 }
71 EXPORT_SYMBOL(register_memory_isolate_notifier);
72 
unregister_memory_isolate_notifier(struct notifier_block * nb)73 void unregister_memory_isolate_notifier(struct notifier_block *nb)
74 {
75 	atomic_notifier_chain_unregister(&memory_isolate_chain, nb);
76 }
77 EXPORT_SYMBOL(unregister_memory_isolate_notifier);
78 
memory_block_release(struct device * dev)79 static void memory_block_release(struct device *dev)
80 {
81 	struct memory_block *mem = to_memory_block(dev);
82 
83 	kfree(mem);
84 }
85 
memory_block_size_bytes(void)86 unsigned long __weak memory_block_size_bytes(void)
87 {
88 	return MIN_MEMORY_BLOCK_SIZE;
89 }
90 
get_memory_block_size(void)91 static unsigned long get_memory_block_size(void)
92 {
93 	unsigned long block_sz;
94 
95 	block_sz = memory_block_size_bytes();
96 
97 	/* Validate blk_sz is a power of 2 and not less than section size */
98 	if ((block_sz & (block_sz - 1)) || (block_sz < MIN_MEMORY_BLOCK_SIZE)) {
99 		WARN_ON(1);
100 		block_sz = MIN_MEMORY_BLOCK_SIZE;
101 	}
102 
103 	return block_sz;
104 }
105 
106 /*
107  * use this as the physical section index that this memsection
108  * uses.
109  */
110 
show_mem_start_phys_index(struct device * dev,struct device_attribute * attr,char * buf)111 static ssize_t show_mem_start_phys_index(struct device *dev,
112 			struct device_attribute *attr, char *buf)
113 {
114 	struct memory_block *mem = to_memory_block(dev);
115 	unsigned long phys_index;
116 
117 	phys_index = mem->start_section_nr / sections_per_block;
118 	return sprintf(buf, "%08lx\n", phys_index);
119 }
120 
121 /*
122  * Show whether the section of memory is likely to be hot-removable
123  */
show_mem_removable(struct device * dev,struct device_attribute * attr,char * buf)124 static ssize_t show_mem_removable(struct device *dev,
125 			struct device_attribute *attr, char *buf)
126 {
127 	unsigned long i, pfn;
128 	int ret = 1;
129 	struct memory_block *mem = to_memory_block(dev);
130 
131 	for (i = 0; i < sections_per_block; i++) {
132 		if (!present_section_nr(mem->start_section_nr + i))
133 			continue;
134 		pfn = section_nr_to_pfn(mem->start_section_nr + i);
135 		ret &= is_mem_section_removable(pfn, PAGES_PER_SECTION);
136 	}
137 
138 	return sprintf(buf, "%d\n", ret);
139 }
140 
141 /*
142  * online, offline, going offline, etc.
143  */
show_mem_state(struct device * dev,struct device_attribute * attr,char * buf)144 static ssize_t show_mem_state(struct device *dev,
145 			struct device_attribute *attr, char *buf)
146 {
147 	struct memory_block *mem = to_memory_block(dev);
148 	ssize_t len = 0;
149 
150 	/*
151 	 * We can probably put these states in a nice little array
152 	 * so that they're not open-coded
153 	 */
154 	switch (mem->state) {
155 	case MEM_ONLINE:
156 		len = sprintf(buf, "online\n");
157 		break;
158 	case MEM_OFFLINE:
159 		len = sprintf(buf, "offline\n");
160 		break;
161 	case MEM_GOING_OFFLINE:
162 		len = sprintf(buf, "going-offline\n");
163 		break;
164 	default:
165 		len = sprintf(buf, "ERROR-UNKNOWN-%ld\n",
166 				mem->state);
167 		WARN_ON(1);
168 		break;
169 	}
170 
171 	return len;
172 }
173 
memory_notify(unsigned long val,void * v)174 int memory_notify(unsigned long val, void *v)
175 {
176 	return blocking_notifier_call_chain(&memory_chain, val, v);
177 }
178 
memory_isolate_notify(unsigned long val,void * v)179 int memory_isolate_notify(unsigned long val, void *v)
180 {
181 	return atomic_notifier_call_chain(&memory_isolate_chain, val, v);
182 }
183 
184 /*
185  * The probe routines leave the pages reserved, just as the bootmem code does.
186  * Make sure they're still that way.
187  */
pages_correctly_reserved(unsigned long start_pfn)188 static bool pages_correctly_reserved(unsigned long start_pfn)
189 {
190 	int i, j;
191 	struct page *page;
192 	unsigned long pfn = start_pfn;
193 
194 	/*
195 	 * memmap between sections is not contiguous except with
196 	 * SPARSEMEM_VMEMMAP. We lookup the page once per section
197 	 * and assume memmap is contiguous within each section
198 	 */
199 	for (i = 0; i < sections_per_block; i++, pfn += PAGES_PER_SECTION) {
200 		if (WARN_ON_ONCE(!pfn_valid(pfn)))
201 			return false;
202 		page = pfn_to_page(pfn);
203 
204 		for (j = 0; j < PAGES_PER_SECTION; j++) {
205 			if (PageReserved(page + j))
206 				continue;
207 
208 			printk(KERN_WARNING "section number %ld page number %d "
209 				"not reserved, was it already online?\n",
210 				pfn_to_section_nr(pfn), j);
211 
212 			return false;
213 		}
214 	}
215 
216 	return true;
217 }
218 
219 /*
220  * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is
221  * OK to have direct references to sparsemem variables in here.
222  * Must already be protected by mem_hotplug_begin().
223  */
224 static int
memory_block_action(unsigned long phys_index,unsigned long action,int online_type)225 memory_block_action(unsigned long phys_index, unsigned long action, int online_type)
226 {
227 	unsigned long start_pfn;
228 	unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
229 	struct page *first_page;
230 	int ret;
231 
232 	start_pfn = section_nr_to_pfn(phys_index);
233 	first_page = pfn_to_page(start_pfn);
234 
235 	switch (action) {
236 	case MEM_ONLINE:
237 		if (!pages_correctly_reserved(start_pfn))
238 			return -EBUSY;
239 
240 		ret = online_pages(start_pfn, nr_pages, online_type);
241 		break;
242 	case MEM_OFFLINE:
243 		ret = offline_pages(start_pfn, nr_pages);
244 		break;
245 	default:
246 		WARN(1, KERN_WARNING "%s(%ld, %ld) unknown action: "
247 		     "%ld\n", __func__, phys_index, action, action);
248 		ret = -EINVAL;
249 	}
250 
251 	return ret;
252 }
253 
memory_block_change_state(struct memory_block * mem,unsigned long to_state,unsigned long from_state_req)254 static int memory_block_change_state(struct memory_block *mem,
255 		unsigned long to_state, unsigned long from_state_req)
256 {
257 	int ret = 0;
258 
259 	if (mem->state != from_state_req)
260 		return -EINVAL;
261 
262 	if (to_state == MEM_OFFLINE)
263 		mem->state = MEM_GOING_OFFLINE;
264 
265 	ret = memory_block_action(mem->start_section_nr, to_state,
266 				mem->online_type);
267 
268 	mem->state = ret ? from_state_req : to_state;
269 
270 	return ret;
271 }
272 
273 /* The device lock serializes operations on memory_subsys_[online|offline] */
memory_subsys_online(struct device * dev)274 static int memory_subsys_online(struct device *dev)
275 {
276 	struct memory_block *mem = to_memory_block(dev);
277 	int ret;
278 
279 	if (mem->state == MEM_ONLINE)
280 		return 0;
281 
282 	/*
283 	 * If we are called from store_mem_state(), online_type will be
284 	 * set >= 0 Otherwise we were called from the device online
285 	 * attribute and need to set the online_type.
286 	 */
287 	if (mem->online_type < 0)
288 		mem->online_type = MMOP_ONLINE_KEEP;
289 
290 	/* Already under protection of mem_hotplug_begin() */
291 	ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE);
292 
293 	/* clear online_type */
294 	mem->online_type = -1;
295 
296 	return ret;
297 }
298 
memory_subsys_offline(struct device * dev)299 static int memory_subsys_offline(struct device *dev)
300 {
301 	struct memory_block *mem = to_memory_block(dev);
302 
303 	if (mem->state == MEM_OFFLINE)
304 		return 0;
305 
306 	/* Can't offline block with non-present sections */
307 	if (mem->section_count != sections_per_block)
308 		return -EINVAL;
309 
310 	return memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE);
311 }
312 
313 static ssize_t
store_mem_state(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)314 store_mem_state(struct device *dev,
315 		struct device_attribute *attr, const char *buf, size_t count)
316 {
317 	struct memory_block *mem = to_memory_block(dev);
318 	int ret, online_type;
319 
320 	ret = lock_device_hotplug_sysfs();
321 	if (ret)
322 		return ret;
323 
324 	if (sysfs_streq(buf, "online_kernel"))
325 		online_type = MMOP_ONLINE_KERNEL;
326 	else if (sysfs_streq(buf, "online_movable"))
327 		online_type = MMOP_ONLINE_MOVABLE;
328 	else if (sysfs_streq(buf, "online"))
329 		online_type = MMOP_ONLINE_KEEP;
330 	else if (sysfs_streq(buf, "offline"))
331 		online_type = MMOP_OFFLINE;
332 	else {
333 		ret = -EINVAL;
334 		goto err;
335 	}
336 
337 	/*
338 	 * Memory hotplug needs to hold mem_hotplug_begin() for probe to find
339 	 * the correct memory block to online before doing device_online(dev),
340 	 * which will take dev->mutex.  Take the lock early to prevent an
341 	 * inversion, memory_subsys_online() callbacks will be implemented by
342 	 * assuming it's already protected.
343 	 */
344 	mem_hotplug_begin();
345 
346 	switch (online_type) {
347 	case MMOP_ONLINE_KERNEL:
348 	case MMOP_ONLINE_MOVABLE:
349 	case MMOP_ONLINE_KEEP:
350 		mem->online_type = online_type;
351 		ret = device_online(&mem->dev);
352 		break;
353 	case MMOP_OFFLINE:
354 		ret = device_offline(&mem->dev);
355 		break;
356 	default:
357 		ret = -EINVAL; /* should never happen */
358 	}
359 
360 	mem_hotplug_done();
361 err:
362 	unlock_device_hotplug();
363 
364 	if (ret)
365 		return ret;
366 	return count;
367 }
368 
369 /*
370  * phys_device is a bad name for this.  What I really want
371  * is a way to differentiate between memory ranges that
372  * are part of physical devices that constitute
373  * a complete removable unit or fru.
374  * i.e. do these ranges belong to the same physical device,
375  * s.t. if I offline all of these sections I can then
376  * remove the physical device?
377  */
show_phys_device(struct device * dev,struct device_attribute * attr,char * buf)378 static ssize_t show_phys_device(struct device *dev,
379 				struct device_attribute *attr, char *buf)
380 {
381 	struct memory_block *mem = to_memory_block(dev);
382 	return sprintf(buf, "%d\n", mem->phys_device);
383 }
384 
385 #ifdef CONFIG_MEMORY_HOTREMOVE
show_valid_zones(struct device * dev,struct device_attribute * attr,char * buf)386 static ssize_t show_valid_zones(struct device *dev,
387 				struct device_attribute *attr, char *buf)
388 {
389 	struct memory_block *mem = to_memory_block(dev);
390 	unsigned long start_pfn, end_pfn;
391 	unsigned long valid_start, valid_end;
392 	unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
393 	struct zone *zone;
394 
395 	start_pfn = section_nr_to_pfn(mem->start_section_nr);
396 	end_pfn = start_pfn + nr_pages;
397 
398 	/* The block contains more than one zone can not be offlined. */
399 	if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start, &valid_end))
400 		return sprintf(buf, "none\n");
401 
402 	zone = page_zone(pfn_to_page(valid_start));
403 
404 	if (zone_idx(zone) == ZONE_MOVABLE - 1) {
405 		/*The mem block is the last memoryblock of this zone.*/
406 		if (valid_end == zone_end_pfn(zone))
407 			return sprintf(buf, "%s %s\n",
408 					zone->name, (zone + 1)->name);
409 	}
410 
411 	if (zone_idx(zone) == ZONE_MOVABLE) {
412 		/*The mem block is the first memoryblock of ZONE_MOVABLE.*/
413 		if (valid_start == zone->zone_start_pfn)
414 			return sprintf(buf, "%s %s\n",
415 					zone->name, (zone - 1)->name);
416 	}
417 
418 	return sprintf(buf, "%s\n", zone->name);
419 }
420 static DEVICE_ATTR(valid_zones, 0444, show_valid_zones, NULL);
421 #endif
422 
423 static DEVICE_ATTR(phys_index, 0444, show_mem_start_phys_index, NULL);
424 static DEVICE_ATTR(state, 0644, show_mem_state, store_mem_state);
425 static DEVICE_ATTR(phys_device, 0444, show_phys_device, NULL);
426 static DEVICE_ATTR(removable, 0444, show_mem_removable, NULL);
427 
428 /*
429  * Block size attribute stuff
430  */
431 static ssize_t
print_block_size(struct device * dev,struct device_attribute * attr,char * buf)432 print_block_size(struct device *dev, struct device_attribute *attr,
433 		 char *buf)
434 {
435 	return sprintf(buf, "%lx\n", get_memory_block_size());
436 }
437 
438 static DEVICE_ATTR(block_size_bytes, 0444, print_block_size, NULL);
439 
440 /*
441  * Some architectures will have custom drivers to do this, and
442  * will not need to do it from userspace.  The fake hot-add code
443  * as well as ppc64 will do all of their discovery in userspace
444  * and will require this interface.
445  */
446 #ifdef CONFIG_ARCH_MEMORY_PROBE
447 static ssize_t
memory_probe_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)448 memory_probe_store(struct device *dev, struct device_attribute *attr,
449 		   const char *buf, size_t count)
450 {
451 	u64 phys_addr;
452 	int nid;
453 	int i, ret;
454 	unsigned long pages_per_block = PAGES_PER_SECTION * sections_per_block;
455 
456 	ret = kstrtoull(buf, 0, &phys_addr);
457 	if (ret)
458 		return ret;
459 
460 	if (phys_addr & ((pages_per_block << PAGE_SHIFT) - 1))
461 		return -EINVAL;
462 
463 	for (i = 0; i < sections_per_block; i++) {
464 		nid = memory_add_physaddr_to_nid(phys_addr);
465 		ret = add_memory(nid, phys_addr,
466 				 PAGES_PER_SECTION << PAGE_SHIFT);
467 		if (ret)
468 			goto out;
469 
470 		phys_addr += MIN_MEMORY_BLOCK_SIZE;
471 	}
472 
473 	ret = count;
474 out:
475 	return ret;
476 }
477 
478 static DEVICE_ATTR(probe, S_IWUSR, NULL, memory_probe_store);
479 #endif
480 
481 #ifdef CONFIG_MEMORY_FAILURE
482 /*
483  * Support for offlining pages of memory
484  */
485 
486 /* Soft offline a page */
487 static ssize_t
store_soft_offline_page(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)488 store_soft_offline_page(struct device *dev,
489 			struct device_attribute *attr,
490 			const char *buf, size_t count)
491 {
492 	int ret;
493 	u64 pfn;
494 	if (!capable(CAP_SYS_ADMIN))
495 		return -EPERM;
496 	if (kstrtoull(buf, 0, &pfn) < 0)
497 		return -EINVAL;
498 	pfn >>= PAGE_SHIFT;
499 	if (!pfn_valid(pfn))
500 		return -ENXIO;
501 	ret = soft_offline_page(pfn_to_page(pfn), 0);
502 	return ret == 0 ? count : ret;
503 }
504 
505 /* Forcibly offline a page, including killing processes. */
506 static ssize_t
store_hard_offline_page(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)507 store_hard_offline_page(struct device *dev,
508 			struct device_attribute *attr,
509 			const char *buf, size_t count)
510 {
511 	int ret;
512 	u64 pfn;
513 	if (!capable(CAP_SYS_ADMIN))
514 		return -EPERM;
515 	if (kstrtoull(buf, 0, &pfn) < 0)
516 		return -EINVAL;
517 	pfn >>= PAGE_SHIFT;
518 	ret = memory_failure(pfn, 0, 0);
519 	return ret ? ret : count;
520 }
521 
522 static DEVICE_ATTR(soft_offline_page, S_IWUSR, NULL, store_soft_offline_page);
523 static DEVICE_ATTR(hard_offline_page, S_IWUSR, NULL, store_hard_offline_page);
524 #endif
525 
526 /*
527  * Note that phys_device is optional.  It is here to allow for
528  * differentiation between which *physical* devices each
529  * section belongs to...
530  */
arch_get_memory_phys_device(unsigned long start_pfn)531 int __weak arch_get_memory_phys_device(unsigned long start_pfn)
532 {
533 	return 0;
534 }
535 
536 /*
537  * A reference for the returned object is held and the reference for the
538  * hinted object is released.
539  */
find_memory_block_hinted(struct mem_section * section,struct memory_block * hint)540 struct memory_block *find_memory_block_hinted(struct mem_section *section,
541 					      struct memory_block *hint)
542 {
543 	int block_id = base_memory_block_id(__section_nr(section));
544 	struct device *hintdev = hint ? &hint->dev : NULL;
545 	struct device *dev;
546 
547 	dev = subsys_find_device_by_id(&memory_subsys, block_id, hintdev);
548 	if (hint)
549 		put_device(&hint->dev);
550 	if (!dev)
551 		return NULL;
552 	return to_memory_block(dev);
553 }
554 
555 /*
556  * For now, we have a linear search to go find the appropriate
557  * memory_block corresponding to a particular phys_index. If
558  * this gets to be a real problem, we can always use a radix
559  * tree or something here.
560  *
561  * This could be made generic for all device subsystems.
562  */
find_memory_block(struct mem_section * section)563 struct memory_block *find_memory_block(struct mem_section *section)
564 {
565 	return find_memory_block_hinted(section, NULL);
566 }
567 
568 static struct attribute *memory_memblk_attrs[] = {
569 	&dev_attr_phys_index.attr,
570 	&dev_attr_state.attr,
571 	&dev_attr_phys_device.attr,
572 	&dev_attr_removable.attr,
573 #ifdef CONFIG_MEMORY_HOTREMOVE
574 	&dev_attr_valid_zones.attr,
575 #endif
576 	NULL
577 };
578 
579 static struct attribute_group memory_memblk_attr_group = {
580 	.attrs = memory_memblk_attrs,
581 };
582 
583 static const struct attribute_group *memory_memblk_attr_groups[] = {
584 	&memory_memblk_attr_group,
585 	NULL,
586 };
587 
588 /*
589  * register_memory - Setup a sysfs device for a memory block
590  */
591 static
register_memory(struct memory_block * memory)592 int register_memory(struct memory_block *memory)
593 {
594 	memory->dev.bus = &memory_subsys;
595 	memory->dev.id = memory->start_section_nr / sections_per_block;
596 	memory->dev.release = memory_block_release;
597 	memory->dev.groups = memory_memblk_attr_groups;
598 	memory->dev.offline = memory->state == MEM_OFFLINE;
599 
600 	return device_register(&memory->dev);
601 }
602 
init_memory_block(struct memory_block ** memory,struct mem_section * section,unsigned long state)603 static int init_memory_block(struct memory_block **memory,
604 			     struct mem_section *section, unsigned long state)
605 {
606 	struct memory_block *mem;
607 	unsigned long start_pfn;
608 	int scn_nr;
609 	int ret = 0;
610 
611 	mem = kzalloc(sizeof(*mem), GFP_KERNEL);
612 	if (!mem)
613 		return -ENOMEM;
614 
615 	scn_nr = __section_nr(section);
616 	mem->start_section_nr =
617 			base_memory_block_id(scn_nr) * sections_per_block;
618 	mem->end_section_nr = mem->start_section_nr + sections_per_block - 1;
619 	mem->state = state;
620 	mem->section_count++;
621 	start_pfn = section_nr_to_pfn(mem->start_section_nr);
622 	mem->phys_device = arch_get_memory_phys_device(start_pfn);
623 
624 	ret = register_memory(mem);
625 
626 	*memory = mem;
627 	return ret;
628 }
629 
add_memory_block(int base_section_nr)630 static int add_memory_block(int base_section_nr)
631 {
632 	struct memory_block *mem;
633 	int i, ret, section_count = 0, section_nr;
634 
635 	for (i = base_section_nr;
636 	     (i < base_section_nr + sections_per_block) && i < NR_MEM_SECTIONS;
637 	     i++) {
638 		if (!present_section_nr(i))
639 			continue;
640 		if (section_count == 0)
641 			section_nr = i;
642 		section_count++;
643 	}
644 
645 	if (section_count == 0)
646 		return 0;
647 	ret = init_memory_block(&mem, __nr_to_section(section_nr), MEM_ONLINE);
648 	if (ret)
649 		return ret;
650 	mem->section_count = section_count;
651 	return 0;
652 }
653 
654 
655 /*
656  * need an interface for the VM to add new memory regions,
657  * but without onlining it.
658  */
register_new_memory(int nid,struct mem_section * section)659 int register_new_memory(int nid, struct mem_section *section)
660 {
661 	int ret = 0;
662 	struct memory_block *mem;
663 
664 	mutex_lock(&mem_sysfs_mutex);
665 
666 	mem = find_memory_block(section);
667 	if (mem) {
668 		mem->section_count++;
669 		put_device(&mem->dev);
670 	} else {
671 		ret = init_memory_block(&mem, section, MEM_OFFLINE);
672 		if (ret)
673 			goto out;
674 	}
675 
676 	if (mem->section_count == sections_per_block)
677 		ret = register_mem_sect_under_node(mem, nid);
678 out:
679 	mutex_unlock(&mem_sysfs_mutex);
680 	return ret;
681 }
682 
683 #ifdef CONFIG_MEMORY_HOTREMOVE
684 static void
unregister_memory(struct memory_block * memory)685 unregister_memory(struct memory_block *memory)
686 {
687 	BUG_ON(memory->dev.bus != &memory_subsys);
688 
689 	/* drop the ref. we got in remove_memory_block() */
690 	put_device(&memory->dev);
691 	device_unregister(&memory->dev);
692 }
693 
remove_memory_block(unsigned long node_id,struct mem_section * section,int phys_device)694 static int remove_memory_block(unsigned long node_id,
695 			       struct mem_section *section, int phys_device)
696 {
697 	struct memory_block *mem;
698 
699 	mutex_lock(&mem_sysfs_mutex);
700 	mem = find_memory_block(section);
701 	unregister_mem_sect_under_nodes(mem, __section_nr(section));
702 
703 	mem->section_count--;
704 	if (mem->section_count == 0)
705 		unregister_memory(mem);
706 	else
707 		put_device(&mem->dev);
708 
709 	mutex_unlock(&mem_sysfs_mutex);
710 	return 0;
711 }
712 
unregister_memory_section(struct mem_section * section)713 int unregister_memory_section(struct mem_section *section)
714 {
715 	if (!present_section(section))
716 		return -EINVAL;
717 
718 	return remove_memory_block(0, section, 0);
719 }
720 #endif /* CONFIG_MEMORY_HOTREMOVE */
721 
722 /* return true if the memory block is offlined, otherwise, return false */
is_memblock_offlined(struct memory_block * mem)723 bool is_memblock_offlined(struct memory_block *mem)
724 {
725 	return mem->state == MEM_OFFLINE;
726 }
727 
728 static struct attribute *memory_root_attrs[] = {
729 #ifdef CONFIG_ARCH_MEMORY_PROBE
730 	&dev_attr_probe.attr,
731 #endif
732 
733 #ifdef CONFIG_MEMORY_FAILURE
734 	&dev_attr_soft_offline_page.attr,
735 	&dev_attr_hard_offline_page.attr,
736 #endif
737 
738 	&dev_attr_block_size_bytes.attr,
739 	NULL
740 };
741 
742 static struct attribute_group memory_root_attr_group = {
743 	.attrs = memory_root_attrs,
744 };
745 
746 static const struct attribute_group *memory_root_attr_groups[] = {
747 	&memory_root_attr_group,
748 	NULL,
749 };
750 
751 /*
752  * Initialize the sysfs support for memory devices...
753  */
memory_dev_init(void)754 int __init memory_dev_init(void)
755 {
756 	unsigned int i;
757 	int ret;
758 	int err;
759 	unsigned long block_sz;
760 
761 	ret = subsys_system_register(&memory_subsys, memory_root_attr_groups);
762 	if (ret)
763 		goto out;
764 
765 	block_sz = get_memory_block_size();
766 	sections_per_block = block_sz / MIN_MEMORY_BLOCK_SIZE;
767 
768 	/*
769 	 * Create entries for memory sections that were found
770 	 * during boot and have been initialized
771 	 */
772 	mutex_lock(&mem_sysfs_mutex);
773 	for (i = 0; i < NR_MEM_SECTIONS; i += sections_per_block) {
774 		err = add_memory_block(i);
775 		if (!ret)
776 			ret = err;
777 	}
778 	mutex_unlock(&mem_sysfs_mutex);
779 
780 out:
781 	if (ret)
782 		printk(KERN_ERR "%s() failed: %d\n", __func__, ret);
783 	return ret;
784 }
785