Linux Audio

Check our new training course

Loading...
v4.10.11
 
  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 <linux/uaccess.h>
 27
 28static 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
 34static int sections_per_block;
 35
 36static inline int base_memory_block_id(int section_nr)
 37{
 38	return section_nr / sections_per_block;
 39}
 40
 41static int memory_subsys_online(struct device *dev);
 42static int memory_subsys_offline(struct device *dev);
 43
 44static 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
 51static BLOCKING_NOTIFIER_HEAD(memory_chain);
 52
 53int register_memory_notifier(struct notifier_block *nb)
 54{
 55	return blocking_notifier_chain_register(&memory_chain, nb);
 56}
 57EXPORT_SYMBOL(register_memory_notifier);
 58
 59void unregister_memory_notifier(struct notifier_block *nb)
 60{
 61	blocking_notifier_chain_unregister(&memory_chain, nb);
 62}
 63EXPORT_SYMBOL(unregister_memory_notifier);
 64
 65static ATOMIC_NOTIFIER_HEAD(memory_isolate_chain);
 66
 67int register_memory_isolate_notifier(struct notifier_block *nb)
 68{
 69	return atomic_notifier_chain_register(&memory_isolate_chain, nb);
 70}
 71EXPORT_SYMBOL(register_memory_isolate_notifier);
 72
 73void unregister_memory_isolate_notifier(struct notifier_block *nb)
 74{
 75	atomic_notifier_chain_unregister(&memory_isolate_chain, nb);
 76}
 77EXPORT_SYMBOL(unregister_memory_isolate_notifier);
 78
 79static void memory_block_release(struct device *dev)
 80{
 81	struct memory_block *mem = to_memory_block(dev);
 82
 83	kfree(mem);
 84}
 85
 86unsigned long __weak memory_block_size_bytes(void)
 87{
 88	return MIN_MEMORY_BLOCK_SIZE;
 89}
 90
 91static 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
111static 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 */
124static 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 */
144static 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
174int memory_notify(unsigned long val, void *v)
175{
176	return blocking_notifier_call_chain(&memory_chain, val, v);
177}
178
179int 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 */
188static 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 */
224static int
225memory_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	int ret;
230
231	start_pfn = section_nr_to_pfn(phys_index);
232
233	switch (action) {
234	case MEM_ONLINE:
235		if (!pages_correctly_reserved(start_pfn))
236			return -EBUSY;
237
238		ret = online_pages(start_pfn, nr_pages, online_type);
239		break;
240	case MEM_OFFLINE:
241		ret = offline_pages(start_pfn, nr_pages);
242		break;
243	default:
244		WARN(1, KERN_WARNING "%s(%ld, %ld) unknown action: "
245		     "%ld\n", __func__, phys_index, action, action);
246		ret = -EINVAL;
247	}
248
249	return ret;
250}
251
252int memory_block_change_state(struct memory_block *mem,
253		unsigned long to_state, unsigned long from_state_req)
254{
255	int ret = 0;
256
257	if (mem->state != from_state_req)
258		return -EINVAL;
259
260	if (to_state == MEM_OFFLINE)
261		mem->state = MEM_GOING_OFFLINE;
262
263	ret = memory_block_action(mem->start_section_nr, to_state,
264				mem->online_type);
265
266	mem->state = ret ? from_state_req : to_state;
267
268	return ret;
269}
270
271/* The device lock serializes operations on memory_subsys_[online|offline] */
272static int memory_subsys_online(struct device *dev)
273{
274	struct memory_block *mem = to_memory_block(dev);
275	int ret;
276
277	if (mem->state == MEM_ONLINE)
278		return 0;
279
280	/*
281	 * If we are called from store_mem_state(), online_type will be
282	 * set >= 0 Otherwise we were called from the device online
283	 * attribute and need to set the online_type.
284	 */
285	if (mem->online_type < 0)
286		mem->online_type = MMOP_ONLINE_KEEP;
287
288	/* Already under protection of mem_hotplug_begin() */
289	ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE);
290
291	/* clear online_type */
292	mem->online_type = -1;
293
294	return ret;
295}
296
297static int memory_subsys_offline(struct device *dev)
298{
299	struct memory_block *mem = to_memory_block(dev);
300
301	if (mem->state == MEM_OFFLINE)
302		return 0;
303
304	/* Can't offline block with non-present sections */
305	if (mem->section_count != sections_per_block)
306		return -EINVAL;
307
308	return memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE);
309}
310
311static ssize_t
312store_mem_state(struct device *dev,
313		struct device_attribute *attr, const char *buf, size_t count)
314{
315	struct memory_block *mem = to_memory_block(dev);
316	int ret, online_type;
317
318	ret = lock_device_hotplug_sysfs();
319	if (ret)
320		return ret;
321
322	if (sysfs_streq(buf, "online_kernel"))
323		online_type = MMOP_ONLINE_KERNEL;
324	else if (sysfs_streq(buf, "online_movable"))
325		online_type = MMOP_ONLINE_MOVABLE;
326	else if (sysfs_streq(buf, "online"))
327		online_type = MMOP_ONLINE_KEEP;
328	else if (sysfs_streq(buf, "offline"))
329		online_type = MMOP_OFFLINE;
330	else {
331		ret = -EINVAL;
332		goto err;
333	}
334
335	/*
336	 * Memory hotplug needs to hold mem_hotplug_begin() for probe to find
337	 * the correct memory block to online before doing device_online(dev),
338	 * which will take dev->mutex.  Take the lock early to prevent an
339	 * inversion, memory_subsys_online() callbacks will be implemented by
340	 * assuming it's already protected.
341	 */
342	mem_hotplug_begin();
343
344	switch (online_type) {
345	case MMOP_ONLINE_KERNEL:
346	case MMOP_ONLINE_MOVABLE:
347	case MMOP_ONLINE_KEEP:
348		mem->online_type = online_type;
349		ret = device_online(&mem->dev);
350		break;
351	case MMOP_OFFLINE:
352		ret = device_offline(&mem->dev);
353		break;
354	default:
355		ret = -EINVAL; /* should never happen */
356	}
357
358	mem_hotplug_done();
359err:
360	unlock_device_hotplug();
361
362	if (ret < 0)
363		return ret;
364	if (ret)
365		return -EINVAL;
366
367	return count;
368}
369
370/*
371 * phys_device is a bad name for this.  What I really want
372 * is a way to differentiate between memory ranges that
373 * are part of physical devices that constitute
374 * a complete removable unit or fru.
375 * i.e. do these ranges belong to the same physical device,
376 * s.t. if I offline all of these sections I can then
377 * remove the physical device?
378 */
379static ssize_t show_phys_device(struct device *dev,
380				struct device_attribute *attr, char *buf)
381{
382	struct memory_block *mem = to_memory_block(dev);
383	return sprintf(buf, "%d\n", mem->phys_device);
384}
385
386#ifdef CONFIG_MEMORY_HOTREMOVE
 
 
 
 
 
 
 
 
 
 
 
 
 
387static ssize_t show_valid_zones(struct device *dev,
388				struct device_attribute *attr, char *buf)
389{
390	struct memory_block *mem = to_memory_block(dev);
391	unsigned long start_pfn, end_pfn;
392	unsigned long valid_start, valid_end, valid_pages;
393	unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
394	struct zone *zone;
395	int zone_shift = 0;
396
397	start_pfn = section_nr_to_pfn(mem->start_section_nr);
398	end_pfn = start_pfn + nr_pages;
399
400	/* The block contains more than one zone can not be offlined. */
401	if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start, &valid_end))
 
 
 
402		return sprintf(buf, "none\n");
403
404	zone = page_zone(pfn_to_page(valid_start));
405	valid_pages = valid_end - valid_start;
406
407	/* MMOP_ONLINE_KEEP */
408	sprintf(buf, "%s", zone->name);
409
410	/* MMOP_ONLINE_KERNEL */
411	zone_can_shift(valid_start, valid_pages, ZONE_NORMAL, &zone_shift);
412	if (zone_shift) {
413		strcat(buf, " ");
414		strcat(buf, (zone + zone_shift)->name);
415	}
416
417	/* MMOP_ONLINE_MOVABLE */
418	zone_can_shift(valid_start, valid_pages, ZONE_MOVABLE, &zone_shift);
419	if (zone_shift) {
420		strcat(buf, " ");
421		strcat(buf, (zone + zone_shift)->name);
 
 
422	}
423
 
 
 
 
 
 
 
 
 
424	strcat(buf, "\n");
425
426	return strlen(buf);
427}
428static DEVICE_ATTR(valid_zones, 0444, show_valid_zones, NULL);
429#endif
430
431static DEVICE_ATTR(phys_index, 0444, show_mem_start_phys_index, NULL);
432static DEVICE_ATTR(state, 0644, show_mem_state, store_mem_state);
433static DEVICE_ATTR(phys_device, 0444, show_phys_device, NULL);
434static DEVICE_ATTR(removable, 0444, show_mem_removable, NULL);
435
436/*
437 * Block size attribute stuff
438 */
439static ssize_t
440print_block_size(struct device *dev, struct device_attribute *attr,
441		 char *buf)
442{
443	return sprintf(buf, "%lx\n", get_memory_block_size());
444}
445
446static DEVICE_ATTR(block_size_bytes, 0444, print_block_size, NULL);
447
448/*
449 * Memory auto online policy.
450 */
451
452static ssize_t
453show_auto_online_blocks(struct device *dev, struct device_attribute *attr,
454			char *buf)
455{
456	if (memhp_auto_online)
457		return sprintf(buf, "online\n");
458	else
459		return sprintf(buf, "offline\n");
460}
461
462static ssize_t
463store_auto_online_blocks(struct device *dev, struct device_attribute *attr,
464			 const char *buf, size_t count)
465{
466	if (sysfs_streq(buf, "online"))
467		memhp_auto_online = true;
468	else if (sysfs_streq(buf, "offline"))
469		memhp_auto_online = false;
470	else
471		return -EINVAL;
472
473	return count;
474}
475
476static DEVICE_ATTR(auto_online_blocks, 0644, show_auto_online_blocks,
477		   store_auto_online_blocks);
478
479/*
480 * Some architectures will have custom drivers to do this, and
481 * will not need to do it from userspace.  The fake hot-add code
482 * as well as ppc64 will do all of their discovery in userspace
483 * and will require this interface.
484 */
485#ifdef CONFIG_ARCH_MEMORY_PROBE
486static ssize_t
487memory_probe_store(struct device *dev, struct device_attribute *attr,
488		   const char *buf, size_t count)
489{
490	u64 phys_addr;
491	int nid, ret;
492	unsigned long pages_per_block = PAGES_PER_SECTION * sections_per_block;
493
494	ret = kstrtoull(buf, 0, &phys_addr);
495	if (ret)
496		return ret;
497
498	if (phys_addr & ((pages_per_block << PAGE_SHIFT) - 1))
499		return -EINVAL;
500
501	nid = memory_add_physaddr_to_nid(phys_addr);
502	ret = add_memory(nid, phys_addr,
503			 MIN_MEMORY_BLOCK_SIZE * sections_per_block);
504
505	if (ret)
506		goto out;
507
508	ret = count;
509out:
510	return ret;
511}
512
513static DEVICE_ATTR(probe, S_IWUSR, NULL, memory_probe_store);
514#endif
515
516#ifdef CONFIG_MEMORY_FAILURE
517/*
518 * Support for offlining pages of memory
519 */
520
521/* Soft offline a page */
522static ssize_t
523store_soft_offline_page(struct device *dev,
524			struct device_attribute *attr,
525			const char *buf, size_t count)
526{
527	int ret;
528	u64 pfn;
529	if (!capable(CAP_SYS_ADMIN))
530		return -EPERM;
531	if (kstrtoull(buf, 0, &pfn) < 0)
532		return -EINVAL;
533	pfn >>= PAGE_SHIFT;
534	if (!pfn_valid(pfn))
535		return -ENXIO;
536	ret = soft_offline_page(pfn_to_page(pfn), 0);
537	return ret == 0 ? count : ret;
538}
539
540/* Forcibly offline a page, including killing processes. */
541static ssize_t
542store_hard_offline_page(struct device *dev,
543			struct device_attribute *attr,
544			const char *buf, size_t count)
545{
546	int ret;
547	u64 pfn;
548	if (!capable(CAP_SYS_ADMIN))
549		return -EPERM;
550	if (kstrtoull(buf, 0, &pfn) < 0)
551		return -EINVAL;
552	pfn >>= PAGE_SHIFT;
553	ret = memory_failure(pfn, 0, 0);
554	return ret ? ret : count;
555}
556
557static DEVICE_ATTR(soft_offline_page, S_IWUSR, NULL, store_soft_offline_page);
558static DEVICE_ATTR(hard_offline_page, S_IWUSR, NULL, store_hard_offline_page);
559#endif
560
561/*
562 * Note that phys_device is optional.  It is here to allow for
563 * differentiation between which *physical* devices each
564 * section belongs to...
565 */
566int __weak arch_get_memory_phys_device(unsigned long start_pfn)
567{
568	return 0;
569}
570
571/*
572 * A reference for the returned object is held and the reference for the
573 * hinted object is released.
574 */
575struct memory_block *find_memory_block_hinted(struct mem_section *section,
576					      struct memory_block *hint)
577{
578	int block_id = base_memory_block_id(__section_nr(section));
579	struct device *hintdev = hint ? &hint->dev : NULL;
580	struct device *dev;
581
582	dev = subsys_find_device_by_id(&memory_subsys, block_id, hintdev);
583	if (hint)
584		put_device(&hint->dev);
585	if (!dev)
586		return NULL;
587	return to_memory_block(dev);
588}
589
590/*
591 * For now, we have a linear search to go find the appropriate
592 * memory_block corresponding to a particular phys_index. If
593 * this gets to be a real problem, we can always use a radix
594 * tree or something here.
595 *
596 * This could be made generic for all device subsystems.
597 */
598struct memory_block *find_memory_block(struct mem_section *section)
599{
600	return find_memory_block_hinted(section, NULL);
601}
602
603static struct attribute *memory_memblk_attrs[] = {
604	&dev_attr_phys_index.attr,
605	&dev_attr_state.attr,
606	&dev_attr_phys_device.attr,
607	&dev_attr_removable.attr,
608#ifdef CONFIG_MEMORY_HOTREMOVE
609	&dev_attr_valid_zones.attr,
610#endif
611	NULL
612};
613
614static struct attribute_group memory_memblk_attr_group = {
615	.attrs = memory_memblk_attrs,
616};
617
618static const struct attribute_group *memory_memblk_attr_groups[] = {
619	&memory_memblk_attr_group,
620	NULL,
621};
622
623/*
624 * register_memory - Setup a sysfs device for a memory block
625 */
626static
627int register_memory(struct memory_block *memory)
628{
629	memory->dev.bus = &memory_subsys;
630	memory->dev.id = memory->start_section_nr / sections_per_block;
631	memory->dev.release = memory_block_release;
632	memory->dev.groups = memory_memblk_attr_groups;
633	memory->dev.offline = memory->state == MEM_OFFLINE;
634
635	return device_register(&memory->dev);
636}
637
638static int init_memory_block(struct memory_block **memory,
639			     struct mem_section *section, unsigned long state)
640{
641	struct memory_block *mem;
642	unsigned long start_pfn;
643	int scn_nr;
644	int ret = 0;
645
646	mem = kzalloc(sizeof(*mem), GFP_KERNEL);
647	if (!mem)
648		return -ENOMEM;
649
650	scn_nr = __section_nr(section);
651	mem->start_section_nr =
652			base_memory_block_id(scn_nr) * sections_per_block;
653	mem->end_section_nr = mem->start_section_nr + sections_per_block - 1;
654	mem->state = state;
655	start_pfn = section_nr_to_pfn(mem->start_section_nr);
656	mem->phys_device = arch_get_memory_phys_device(start_pfn);
657
658	ret = register_memory(mem);
659
660	*memory = mem;
661	return ret;
662}
663
664static int add_memory_block(int base_section_nr)
665{
666	struct memory_block *mem;
667	int i, ret, section_count = 0, section_nr;
668
669	for (i = base_section_nr;
670	     (i < base_section_nr + sections_per_block) && i < NR_MEM_SECTIONS;
671	     i++) {
672		if (!present_section_nr(i))
673			continue;
674		if (section_count == 0)
675			section_nr = i;
676		section_count++;
677	}
678
679	if (section_count == 0)
680		return 0;
681	ret = init_memory_block(&mem, __nr_to_section(section_nr), MEM_ONLINE);
682	if (ret)
683		return ret;
684	mem->section_count = section_count;
685	return 0;
686}
687
688static bool is_zone_device_section(struct mem_section *ms)
689{
690	struct page *page;
691
692	page = sparse_decode_mem_map(ms->section_mem_map, __section_nr(ms));
693	return is_zone_device_page(page);
694}
695
696/*
697 * need an interface for the VM to add new memory regions,
698 * but without onlining it.
699 */
700int register_new_memory(int nid, struct mem_section *section)
701{
702	int ret = 0;
703	struct memory_block *mem;
704
705	if (is_zone_device_section(section))
706		return 0;
707
708	mutex_lock(&mem_sysfs_mutex);
709
710	mem = find_memory_block(section);
711	if (mem) {
712		mem->section_count++;
713		put_device(&mem->dev);
714	} else {
715		ret = init_memory_block(&mem, section, MEM_OFFLINE);
716		if (ret)
717			goto out;
718		mem->section_count++;
719	}
720
721	if (mem->section_count == sections_per_block)
722		ret = register_mem_sect_under_node(mem, nid);
723out:
724	mutex_unlock(&mem_sysfs_mutex);
725	return ret;
726}
727
728#ifdef CONFIG_MEMORY_HOTREMOVE
729static void
730unregister_memory(struct memory_block *memory)
731{
732	BUG_ON(memory->dev.bus != &memory_subsys);
733
734	/* drop the ref. we got in remove_memory_block() */
735	put_device(&memory->dev);
736	device_unregister(&memory->dev);
737}
738
739static int remove_memory_section(unsigned long node_id,
740			       struct mem_section *section, int phys_device)
741{
742	struct memory_block *mem;
743
744	if (is_zone_device_section(section))
745		return 0;
746
747	mutex_lock(&mem_sysfs_mutex);
 
 
 
 
 
748	mem = find_memory_block(section);
 
 
 
749	unregister_mem_sect_under_nodes(mem, __section_nr(section));
750
751	mem->section_count--;
752	if (mem->section_count == 0)
753		unregister_memory(mem);
754	else
755		put_device(&mem->dev);
756
 
757	mutex_unlock(&mem_sysfs_mutex);
758	return 0;
759}
760
761int unregister_memory_section(struct mem_section *section)
762{
763	if (!present_section(section))
764		return -EINVAL;
765
766	return remove_memory_section(0, section, 0);
767}
768#endif /* CONFIG_MEMORY_HOTREMOVE */
769
770/* return true if the memory block is offlined, otherwise, return false */
771bool is_memblock_offlined(struct memory_block *mem)
772{
773	return mem->state == MEM_OFFLINE;
774}
775
776static struct attribute *memory_root_attrs[] = {
777#ifdef CONFIG_ARCH_MEMORY_PROBE
778	&dev_attr_probe.attr,
779#endif
780
781#ifdef CONFIG_MEMORY_FAILURE
782	&dev_attr_soft_offline_page.attr,
783	&dev_attr_hard_offline_page.attr,
784#endif
785
786	&dev_attr_block_size_bytes.attr,
787	&dev_attr_auto_online_blocks.attr,
788	NULL
789};
790
791static struct attribute_group memory_root_attr_group = {
792	.attrs = memory_root_attrs,
793};
794
795static const struct attribute_group *memory_root_attr_groups[] = {
796	&memory_root_attr_group,
797	NULL,
798};
799
800/*
801 * Initialize the sysfs support for memory devices...
802 */
803int __init memory_dev_init(void)
804{
805	unsigned int i;
806	int ret;
807	int err;
808	unsigned long block_sz;
809
810	ret = subsys_system_register(&memory_subsys, memory_root_attr_groups);
811	if (ret)
812		goto out;
813
814	block_sz = get_memory_block_size();
815	sections_per_block = block_sz / MIN_MEMORY_BLOCK_SIZE;
816
817	/*
818	 * Create entries for memory sections that were found
819	 * during boot and have been initialized
820	 */
821	mutex_lock(&mem_sysfs_mutex);
822	for (i = 0; i < NR_MEM_SECTIONS; i += sections_per_block) {
 
823		err = add_memory_block(i);
824		if (!ret)
825			ret = err;
826	}
827	mutex_unlock(&mem_sysfs_mutex);
828
829out:
830	if (ret)
831		printk(KERN_ERR "%s() failed: %d\n", __func__, ret);
832	return ret;
833}
v4.17
  1// SPDX-License-Identifier: GPL-2.0
  2/*
  3 * Memory subsystem support
  4 *
  5 * Written by Matt Tolentino <matthew.e.tolentino@intel.com>
  6 *            Dave Hansen <haveblue@us.ibm.com>
  7 *
  8 * This file provides the necessary infrastructure to represent
  9 * a SPARSEMEM-memory-model system's physical memory in /sysfs.
 10 * All arch-independent code that assumes MEMORY_HOTPLUG requires
 11 * SPARSEMEM should be contained here, or in mm/memory_hotplug.c.
 12 */
 13
 14#include <linux/module.h>
 15#include <linux/init.h>
 16#include <linux/topology.h>
 17#include <linux/capability.h>
 18#include <linux/device.h>
 19#include <linux/memory.h>
 20#include <linux/memory_hotplug.h>
 21#include <linux/mm.h>
 22#include <linux/mutex.h>
 23#include <linux/stat.h>
 24#include <linux/slab.h>
 25
 26#include <linux/atomic.h>
 27#include <linux/uaccess.h>
 28
 29static DEFINE_MUTEX(mem_sysfs_mutex);
 30
 31#define MEMORY_CLASS_NAME	"memory"
 32
 33#define to_memory_block(dev) container_of(dev, struct memory_block, dev)
 34
 35static int sections_per_block;
 36
 37static inline int base_memory_block_id(int section_nr)
 38{
 39	return section_nr / sections_per_block;
 40}
 41
 42static int memory_subsys_online(struct device *dev);
 43static int memory_subsys_offline(struct device *dev);
 44
 45static struct bus_type memory_subsys = {
 46	.name = MEMORY_CLASS_NAME,
 47	.dev_name = MEMORY_CLASS_NAME,
 48	.online = memory_subsys_online,
 49	.offline = memory_subsys_offline,
 50};
 51
 52static BLOCKING_NOTIFIER_HEAD(memory_chain);
 53
 54int register_memory_notifier(struct notifier_block *nb)
 55{
 56	return blocking_notifier_chain_register(&memory_chain, nb);
 57}
 58EXPORT_SYMBOL(register_memory_notifier);
 59
 60void unregister_memory_notifier(struct notifier_block *nb)
 61{
 62	blocking_notifier_chain_unregister(&memory_chain, nb);
 63}
 64EXPORT_SYMBOL(unregister_memory_notifier);
 65
 66static ATOMIC_NOTIFIER_HEAD(memory_isolate_chain);
 67
 68int register_memory_isolate_notifier(struct notifier_block *nb)
 69{
 70	return atomic_notifier_chain_register(&memory_isolate_chain, nb);
 71}
 72EXPORT_SYMBOL(register_memory_isolate_notifier);
 73
 74void unregister_memory_isolate_notifier(struct notifier_block *nb)
 75{
 76	atomic_notifier_chain_unregister(&memory_isolate_chain, nb);
 77}
 78EXPORT_SYMBOL(unregister_memory_isolate_notifier);
 79
 80static void memory_block_release(struct device *dev)
 81{
 82	struct memory_block *mem = to_memory_block(dev);
 83
 84	kfree(mem);
 85}
 86
 87unsigned long __weak memory_block_size_bytes(void)
 88{
 89	return MIN_MEMORY_BLOCK_SIZE;
 90}
 91
 92static unsigned long get_memory_block_size(void)
 93{
 94	unsigned long block_sz;
 95
 96	block_sz = memory_block_size_bytes();
 97
 98	/* Validate blk_sz is a power of 2 and not less than section size */
 99	if ((block_sz & (block_sz - 1)) || (block_sz < MIN_MEMORY_BLOCK_SIZE)) {
100		WARN_ON(1);
101		block_sz = MIN_MEMORY_BLOCK_SIZE;
102	}
103
104	return block_sz;
105}
106
107/*
108 * use this as the physical section index that this memsection
109 * uses.
110 */
111
112static ssize_t show_mem_start_phys_index(struct device *dev,
113			struct device_attribute *attr, char *buf)
114{
115	struct memory_block *mem = to_memory_block(dev);
116	unsigned long phys_index;
117
118	phys_index = mem->start_section_nr / sections_per_block;
119	return sprintf(buf, "%08lx\n", phys_index);
120}
121
122/*
123 * Show whether the section of memory is likely to be hot-removable
124 */
125static ssize_t show_mem_removable(struct device *dev,
126			struct device_attribute *attr, char *buf)
127{
128	unsigned long i, pfn;
129	int ret = 1;
130	struct memory_block *mem = to_memory_block(dev);
131
132	if (mem->state != MEM_ONLINE)
133		goto out;
134
135	for (i = 0; i < sections_per_block; i++) {
136		if (!present_section_nr(mem->start_section_nr + i))
137			continue;
138		pfn = section_nr_to_pfn(mem->start_section_nr + i);
139		ret &= is_mem_section_removable(pfn, PAGES_PER_SECTION);
140	}
141
142out:
143	return sprintf(buf, "%d\n", ret);
144}
145
146/*
147 * online, offline, going offline, etc.
148 */
149static ssize_t show_mem_state(struct device *dev,
150			struct device_attribute *attr, char *buf)
151{
152	struct memory_block *mem = to_memory_block(dev);
153	ssize_t len = 0;
154
155	/*
156	 * We can probably put these states in a nice little array
157	 * so that they're not open-coded
158	 */
159	switch (mem->state) {
160	case MEM_ONLINE:
161		len = sprintf(buf, "online\n");
162		break;
163	case MEM_OFFLINE:
164		len = sprintf(buf, "offline\n");
165		break;
166	case MEM_GOING_OFFLINE:
167		len = sprintf(buf, "going-offline\n");
168		break;
169	default:
170		len = sprintf(buf, "ERROR-UNKNOWN-%ld\n",
171				mem->state);
172		WARN_ON(1);
173		break;
174	}
175
176	return len;
177}
178
179int memory_notify(unsigned long val, void *v)
180{
181	return blocking_notifier_call_chain(&memory_chain, val, v);
182}
183
184int memory_isolate_notify(unsigned long val, void *v)
185{
186	return atomic_notifier_call_chain(&memory_isolate_chain, val, v);
187}
188
189/*
190 * The probe routines leave the pages uninitialized, just as the bootmem code
191 * does. Make sure we do not access them, but instead use only information from
192 * within sections.
193 */
194static bool pages_correctly_probed(unsigned long start_pfn)
195{
196	unsigned long section_nr = pfn_to_section_nr(start_pfn);
197	unsigned long section_nr_end = section_nr + sections_per_block;
198	unsigned long pfn = start_pfn;
199
200	/*
201	 * memmap between sections is not contiguous except with
202	 * SPARSEMEM_VMEMMAP. We lookup the page once per section
203	 * and assume memmap is contiguous within each section
204	 */
205	for (; section_nr < section_nr_end; section_nr++) {
206		if (WARN_ON_ONCE(!pfn_valid(pfn)))
207			return false;
 
 
 
 
 
 
 
 
 
208
209		if (!present_section_nr(section_nr)) {
210			pr_warn("section %ld pfn[%lx, %lx) not present",
211				section_nr, pfn, pfn + PAGES_PER_SECTION);
212			return false;
213		} else if (!valid_section_nr(section_nr)) {
214			pr_warn("section %ld pfn[%lx, %lx) no valid memmap",
215				section_nr, pfn, pfn + PAGES_PER_SECTION);
216			return false;
217		} else if (online_section_nr(section_nr)) {
218			pr_warn("section %ld pfn[%lx, %lx) is already online",
219				section_nr, pfn, pfn + PAGES_PER_SECTION);
220			return false;
221		}
222		pfn += PAGES_PER_SECTION;
223	}
224
225	return true;
226}
227
228/*
229 * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is
230 * OK to have direct references to sparsemem variables in here.
231 * Must already be protected by mem_hotplug_begin().
232 */
233static int
234memory_block_action(unsigned long phys_index, unsigned long action, int online_type)
235{
236	unsigned long start_pfn;
237	unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
238	int ret;
239
240	start_pfn = section_nr_to_pfn(phys_index);
241
242	switch (action) {
243	case MEM_ONLINE:
244		if (!pages_correctly_probed(start_pfn))
245			return -EBUSY;
246
247		ret = online_pages(start_pfn, nr_pages, online_type);
248		break;
249	case MEM_OFFLINE:
250		ret = offline_pages(start_pfn, nr_pages);
251		break;
252	default:
253		WARN(1, KERN_WARNING "%s(%ld, %ld) unknown action: "
254		     "%ld\n", __func__, phys_index, action, action);
255		ret = -EINVAL;
256	}
257
258	return ret;
259}
260
261static int memory_block_change_state(struct memory_block *mem,
262		unsigned long to_state, unsigned long from_state_req)
263{
264	int ret = 0;
265
266	if (mem->state != from_state_req)
267		return -EINVAL;
268
269	if (to_state == MEM_OFFLINE)
270		mem->state = MEM_GOING_OFFLINE;
271
272	ret = memory_block_action(mem->start_section_nr, to_state,
273				mem->online_type);
274
275	mem->state = ret ? from_state_req : to_state;
276
277	return ret;
278}
279
280/* The device lock serializes operations on memory_subsys_[online|offline] */
281static int memory_subsys_online(struct device *dev)
282{
283	struct memory_block *mem = to_memory_block(dev);
284	int ret;
285
286	if (mem->state == MEM_ONLINE)
287		return 0;
288
289	/*
290	 * If we are called from store_mem_state(), online_type will be
291	 * set >= 0 Otherwise we were called from the device online
292	 * attribute and need to set the online_type.
293	 */
294	if (mem->online_type < 0)
295		mem->online_type = MMOP_ONLINE_KEEP;
296
297	/* Already under protection of mem_hotplug_begin() */
298	ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE);
299
300	/* clear online_type */
301	mem->online_type = -1;
302
303	return ret;
304}
305
306static int memory_subsys_offline(struct device *dev)
307{
308	struct memory_block *mem = to_memory_block(dev);
309
310	if (mem->state == MEM_OFFLINE)
311		return 0;
312
313	/* Can't offline block with non-present sections */
314	if (mem->section_count != sections_per_block)
315		return -EINVAL;
316
317	return memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE);
318}
319
320static ssize_t
321store_mem_state(struct device *dev,
322		struct device_attribute *attr, const char *buf, size_t count)
323{
324	struct memory_block *mem = to_memory_block(dev);
325	int ret, online_type;
326
327	ret = lock_device_hotplug_sysfs();
328	if (ret)
329		return ret;
330
331	if (sysfs_streq(buf, "online_kernel"))
332		online_type = MMOP_ONLINE_KERNEL;
333	else if (sysfs_streq(buf, "online_movable"))
334		online_type = MMOP_ONLINE_MOVABLE;
335	else if (sysfs_streq(buf, "online"))
336		online_type = MMOP_ONLINE_KEEP;
337	else if (sysfs_streq(buf, "offline"))
338		online_type = MMOP_OFFLINE;
339	else {
340		ret = -EINVAL;
341		goto err;
342	}
343
344	/*
345	 * Memory hotplug needs to hold mem_hotplug_begin() for probe to find
346	 * the correct memory block to online before doing device_online(dev),
347	 * which will take dev->mutex.  Take the lock early to prevent an
348	 * inversion, memory_subsys_online() callbacks will be implemented by
349	 * assuming it's already protected.
350	 */
351	mem_hotplug_begin();
352
353	switch (online_type) {
354	case MMOP_ONLINE_KERNEL:
355	case MMOP_ONLINE_MOVABLE:
356	case MMOP_ONLINE_KEEP:
357		mem->online_type = online_type;
358		ret = device_online(&mem->dev);
359		break;
360	case MMOP_OFFLINE:
361		ret = device_offline(&mem->dev);
362		break;
363	default:
364		ret = -EINVAL; /* should never happen */
365	}
366
367	mem_hotplug_done();
368err:
369	unlock_device_hotplug();
370
371	if (ret < 0)
372		return ret;
373	if (ret)
374		return -EINVAL;
375
376	return count;
377}
378
379/*
380 * phys_device is a bad name for this.  What I really want
381 * is a way to differentiate between memory ranges that
382 * are part of physical devices that constitute
383 * a complete removable unit or fru.
384 * i.e. do these ranges belong to the same physical device,
385 * s.t. if I offline all of these sections I can then
386 * remove the physical device?
387 */
388static ssize_t show_phys_device(struct device *dev,
389				struct device_attribute *attr, char *buf)
390{
391	struct memory_block *mem = to_memory_block(dev);
392	return sprintf(buf, "%d\n", mem->phys_device);
393}
394
395#ifdef CONFIG_MEMORY_HOTREMOVE
396static void print_allowed_zone(char *buf, int nid, unsigned long start_pfn,
397		unsigned long nr_pages, int online_type,
398		struct zone *default_zone)
399{
400	struct zone *zone;
401
402	zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages);
403	if (zone != default_zone) {
404		strcat(buf, " ");
405		strcat(buf, zone->name);
406	}
407}
408
409static ssize_t show_valid_zones(struct device *dev,
410				struct device_attribute *attr, char *buf)
411{
412	struct memory_block *mem = to_memory_block(dev);
413	unsigned long start_pfn = section_nr_to_pfn(mem->start_section_nr);
 
414	unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
415	unsigned long valid_start_pfn, valid_end_pfn;
416	struct zone *default_zone;
417	int nid;
 
 
418
419	/*
420	 * The block contains more than one zone can not be offlined.
421	 * This can happen e.g. for ZONE_DMA and ZONE_DMA32
422	 */
423	if (!test_pages_in_a_zone(start_pfn, start_pfn + nr_pages, &valid_start_pfn, &valid_end_pfn))
424		return sprintf(buf, "none\n");
425
426	start_pfn = valid_start_pfn;
427	nr_pages = valid_end_pfn - start_pfn;
 
 
 
 
 
 
 
 
 
 
428
429	/*
430	 * Check the existing zone. Make sure that we do that only on the
431	 * online nodes otherwise the page_zone is not reliable
432	 */
433	if (mem->state == MEM_ONLINE) {
434		strcat(buf, page_zone(pfn_to_page(start_pfn))->name);
435		goto out;
436	}
437
438	nid = pfn_to_nid(start_pfn);
439	default_zone = zone_for_pfn_range(MMOP_ONLINE_KEEP, nid, start_pfn, nr_pages);
440	strcat(buf, default_zone->name);
441
442	print_allowed_zone(buf, nid, start_pfn, nr_pages, MMOP_ONLINE_KERNEL,
443			default_zone);
444	print_allowed_zone(buf, nid, start_pfn, nr_pages, MMOP_ONLINE_MOVABLE,
445			default_zone);
446out:
447	strcat(buf, "\n");
448
449	return strlen(buf);
450}
451static DEVICE_ATTR(valid_zones, 0444, show_valid_zones, NULL);
452#endif
453
454static DEVICE_ATTR(phys_index, 0444, show_mem_start_phys_index, NULL);
455static DEVICE_ATTR(state, 0644, show_mem_state, store_mem_state);
456static DEVICE_ATTR(phys_device, 0444, show_phys_device, NULL);
457static DEVICE_ATTR(removable, 0444, show_mem_removable, NULL);
458
459/*
460 * Block size attribute stuff
461 */
462static ssize_t
463print_block_size(struct device *dev, struct device_attribute *attr,
464		 char *buf)
465{
466	return sprintf(buf, "%lx\n", get_memory_block_size());
467}
468
469static DEVICE_ATTR(block_size_bytes, 0444, print_block_size, NULL);
470
471/*
472 * Memory auto online policy.
473 */
474
475static ssize_t
476show_auto_online_blocks(struct device *dev, struct device_attribute *attr,
477			char *buf)
478{
479	if (memhp_auto_online)
480		return sprintf(buf, "online\n");
481	else
482		return sprintf(buf, "offline\n");
483}
484
485static ssize_t
486store_auto_online_blocks(struct device *dev, struct device_attribute *attr,
487			 const char *buf, size_t count)
488{
489	if (sysfs_streq(buf, "online"))
490		memhp_auto_online = true;
491	else if (sysfs_streq(buf, "offline"))
492		memhp_auto_online = false;
493	else
494		return -EINVAL;
495
496	return count;
497}
498
499static DEVICE_ATTR(auto_online_blocks, 0644, show_auto_online_blocks,
500		   store_auto_online_blocks);
501
502/*
503 * Some architectures will have custom drivers to do this, and
504 * will not need to do it from userspace.  The fake hot-add code
505 * as well as ppc64 will do all of their discovery in userspace
506 * and will require this interface.
507 */
508#ifdef CONFIG_ARCH_MEMORY_PROBE
509static ssize_t
510memory_probe_store(struct device *dev, struct device_attribute *attr,
511		   const char *buf, size_t count)
512{
513	u64 phys_addr;
514	int nid, ret;
515	unsigned long pages_per_block = PAGES_PER_SECTION * sections_per_block;
516
517	ret = kstrtoull(buf, 0, &phys_addr);
518	if (ret)
519		return ret;
520
521	if (phys_addr & ((pages_per_block << PAGE_SHIFT) - 1))
522		return -EINVAL;
523
524	nid = memory_add_physaddr_to_nid(phys_addr);
525	ret = add_memory(nid, phys_addr,
526			 MIN_MEMORY_BLOCK_SIZE * sections_per_block);
527
528	if (ret)
529		goto out;
530
531	ret = count;
532out:
533	return ret;
534}
535
536static DEVICE_ATTR(probe, S_IWUSR, NULL, memory_probe_store);
537#endif
538
539#ifdef CONFIG_MEMORY_FAILURE
540/*
541 * Support for offlining pages of memory
542 */
543
544/* Soft offline a page */
545static ssize_t
546store_soft_offline_page(struct device *dev,
547			struct device_attribute *attr,
548			const char *buf, size_t count)
549{
550	int ret;
551	u64 pfn;
552	if (!capable(CAP_SYS_ADMIN))
553		return -EPERM;
554	if (kstrtoull(buf, 0, &pfn) < 0)
555		return -EINVAL;
556	pfn >>= PAGE_SHIFT;
557	if (!pfn_valid(pfn))
558		return -ENXIO;
559	ret = soft_offline_page(pfn_to_page(pfn), 0);
560	return ret == 0 ? count : ret;
561}
562
563/* Forcibly offline a page, including killing processes. */
564static ssize_t
565store_hard_offline_page(struct device *dev,
566			struct device_attribute *attr,
567			const char *buf, size_t count)
568{
569	int ret;
570	u64 pfn;
571	if (!capable(CAP_SYS_ADMIN))
572		return -EPERM;
573	if (kstrtoull(buf, 0, &pfn) < 0)
574		return -EINVAL;
575	pfn >>= PAGE_SHIFT;
576	ret = memory_failure(pfn, 0);
577	return ret ? ret : count;
578}
579
580static DEVICE_ATTR(soft_offline_page, S_IWUSR, NULL, store_soft_offline_page);
581static DEVICE_ATTR(hard_offline_page, S_IWUSR, NULL, store_hard_offline_page);
582#endif
583
584/*
585 * Note that phys_device is optional.  It is here to allow for
586 * differentiation between which *physical* devices each
587 * section belongs to...
588 */
589int __weak arch_get_memory_phys_device(unsigned long start_pfn)
590{
591	return 0;
592}
593
594/*
595 * A reference for the returned object is held and the reference for the
596 * hinted object is released.
597 */
598struct memory_block *find_memory_block_hinted(struct mem_section *section,
599					      struct memory_block *hint)
600{
601	int block_id = base_memory_block_id(__section_nr(section));
602	struct device *hintdev = hint ? &hint->dev : NULL;
603	struct device *dev;
604
605	dev = subsys_find_device_by_id(&memory_subsys, block_id, hintdev);
606	if (hint)
607		put_device(&hint->dev);
608	if (!dev)
609		return NULL;
610	return to_memory_block(dev);
611}
612
613/*
614 * For now, we have a linear search to go find the appropriate
615 * memory_block corresponding to a particular phys_index. If
616 * this gets to be a real problem, we can always use a radix
617 * tree or something here.
618 *
619 * This could be made generic for all device subsystems.
620 */
621struct memory_block *find_memory_block(struct mem_section *section)
622{
623	return find_memory_block_hinted(section, NULL);
624}
625
626static struct attribute *memory_memblk_attrs[] = {
627	&dev_attr_phys_index.attr,
628	&dev_attr_state.attr,
629	&dev_attr_phys_device.attr,
630	&dev_attr_removable.attr,
631#ifdef CONFIG_MEMORY_HOTREMOVE
632	&dev_attr_valid_zones.attr,
633#endif
634	NULL
635};
636
637static struct attribute_group memory_memblk_attr_group = {
638	.attrs = memory_memblk_attrs,
639};
640
641static const struct attribute_group *memory_memblk_attr_groups[] = {
642	&memory_memblk_attr_group,
643	NULL,
644};
645
646/*
647 * register_memory - Setup a sysfs device for a memory block
648 */
649static
650int register_memory(struct memory_block *memory)
651{
652	memory->dev.bus = &memory_subsys;
653	memory->dev.id = memory->start_section_nr / sections_per_block;
654	memory->dev.release = memory_block_release;
655	memory->dev.groups = memory_memblk_attr_groups;
656	memory->dev.offline = memory->state == MEM_OFFLINE;
657
658	return device_register(&memory->dev);
659}
660
661static int init_memory_block(struct memory_block **memory,
662			     struct mem_section *section, unsigned long state)
663{
664	struct memory_block *mem;
665	unsigned long start_pfn;
666	int scn_nr;
667	int ret = 0;
668
669	mem = kzalloc(sizeof(*mem), GFP_KERNEL);
670	if (!mem)
671		return -ENOMEM;
672
673	scn_nr = __section_nr(section);
674	mem->start_section_nr =
675			base_memory_block_id(scn_nr) * sections_per_block;
676	mem->end_section_nr = mem->start_section_nr + sections_per_block - 1;
677	mem->state = state;
678	start_pfn = section_nr_to_pfn(mem->start_section_nr);
679	mem->phys_device = arch_get_memory_phys_device(start_pfn);
680
681	ret = register_memory(mem);
682
683	*memory = mem;
684	return ret;
685}
686
687static int add_memory_block(int base_section_nr)
688{
689	struct memory_block *mem;
690	int i, ret, section_count = 0, section_nr;
691
692	for (i = base_section_nr;
693	     (i < base_section_nr + sections_per_block) && i < NR_MEM_SECTIONS;
694	     i++) {
695		if (!present_section_nr(i))
696			continue;
697		if (section_count == 0)
698			section_nr = i;
699		section_count++;
700	}
701
702	if (section_count == 0)
703		return 0;
704	ret = init_memory_block(&mem, __nr_to_section(section_nr), MEM_ONLINE);
705	if (ret)
706		return ret;
707	mem->section_count = section_count;
708	return 0;
709}
710
 
 
 
 
 
 
 
 
711/*
712 * need an interface for the VM to add new memory regions,
713 * but without onlining it.
714 */
715int hotplug_memory_register(int nid, struct mem_section *section)
716{
717	int ret = 0;
718	struct memory_block *mem;
719
 
 
 
720	mutex_lock(&mem_sysfs_mutex);
721
722	mem = find_memory_block(section);
723	if (mem) {
724		mem->section_count++;
725		put_device(&mem->dev);
726	} else {
727		ret = init_memory_block(&mem, section, MEM_OFFLINE);
728		if (ret)
729			goto out;
730		mem->section_count++;
731	}
732
733	if (mem->section_count == sections_per_block)
734		ret = register_mem_sect_under_node(mem, nid, false);
735out:
736	mutex_unlock(&mem_sysfs_mutex);
737	return ret;
738}
739
740#ifdef CONFIG_MEMORY_HOTREMOVE
741static void
742unregister_memory(struct memory_block *memory)
743{
744	BUG_ON(memory->dev.bus != &memory_subsys);
745
746	/* drop the ref. we got in remove_memory_block() */
747	put_device(&memory->dev);
748	device_unregister(&memory->dev);
749}
750
751static int remove_memory_section(unsigned long node_id,
752			       struct mem_section *section, int phys_device)
753{
754	struct memory_block *mem;
755
 
 
 
756	mutex_lock(&mem_sysfs_mutex);
757
758	/*
759	 * Some users of the memory hotplug do not want/need memblock to
760	 * track all sections. Skip over those.
761	 */
762	mem = find_memory_block(section);
763	if (!mem)
764		goto out_unlock;
765
766	unregister_mem_sect_under_nodes(mem, __section_nr(section));
767
768	mem->section_count--;
769	if (mem->section_count == 0)
770		unregister_memory(mem);
771	else
772		put_device(&mem->dev);
773
774out_unlock:
775	mutex_unlock(&mem_sysfs_mutex);
776	return 0;
777}
778
779int unregister_memory_section(struct mem_section *section)
780{
781	if (!present_section(section))
782		return -EINVAL;
783
784	return remove_memory_section(0, section, 0);
785}
786#endif /* CONFIG_MEMORY_HOTREMOVE */
787
788/* return true if the memory block is offlined, otherwise, return false */
789bool is_memblock_offlined(struct memory_block *mem)
790{
791	return mem->state == MEM_OFFLINE;
792}
793
794static struct attribute *memory_root_attrs[] = {
795#ifdef CONFIG_ARCH_MEMORY_PROBE
796	&dev_attr_probe.attr,
797#endif
798
799#ifdef CONFIG_MEMORY_FAILURE
800	&dev_attr_soft_offline_page.attr,
801	&dev_attr_hard_offline_page.attr,
802#endif
803
804	&dev_attr_block_size_bytes.attr,
805	&dev_attr_auto_online_blocks.attr,
806	NULL
807};
808
809static struct attribute_group memory_root_attr_group = {
810	.attrs = memory_root_attrs,
811};
812
813static const struct attribute_group *memory_root_attr_groups[] = {
814	&memory_root_attr_group,
815	NULL,
816};
817
818/*
819 * Initialize the sysfs support for memory devices...
820 */
821int __init memory_dev_init(void)
822{
823	unsigned int i;
824	int ret;
825	int err;
826	unsigned long block_sz;
827
828	ret = subsys_system_register(&memory_subsys, memory_root_attr_groups);
829	if (ret)
830		goto out;
831
832	block_sz = get_memory_block_size();
833	sections_per_block = block_sz / MIN_MEMORY_BLOCK_SIZE;
834
835	/*
836	 * Create entries for memory sections that were found
837	 * during boot and have been initialized
838	 */
839	mutex_lock(&mem_sysfs_mutex);
840	for (i = 0; i <= __highest_present_section_nr;
841		i += sections_per_block) {
842		err = add_memory_block(i);
843		if (!ret)
844			ret = err;
845	}
846	mutex_unlock(&mem_sysfs_mutex);
847
848out:
849	if (ret)
850		printk(KERN_ERR "%s() failed: %d\n", __func__, ret);
851	return ret;
852}