Linux Audio

Check our new training course

Loading...
v5.9
  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/stat.h>
 23#include <linux/slab.h>
 24#include <linux/xarray.h>
 25
 26#include <linux/atomic.h>
 27#include <linux/uaccess.h>
 28
 29#define MEMORY_CLASS_NAME	"memory"
 30
 31static const char *const online_type_to_str[] = {
 32	[MMOP_OFFLINE] = "offline",
 33	[MMOP_ONLINE] = "online",
 34	[MMOP_ONLINE_KERNEL] = "online_kernel",
 35	[MMOP_ONLINE_MOVABLE] = "online_movable",
 36};
 37
 38int memhp_online_type_from_str(const char *str)
 39{
 40	int i;
 41
 42	for (i = 0; i < ARRAY_SIZE(online_type_to_str); i++) {
 43		if (sysfs_streq(str, online_type_to_str[i]))
 44			return i;
 45	}
 46	return -EINVAL;
 47}
 48
 49#define to_memory_block(dev) container_of(dev, struct memory_block, dev)
 50
 51static int sections_per_block;
 52
 53static inline unsigned long memory_block_id(unsigned long section_nr)
 54{
 55	return section_nr / sections_per_block;
 56}
 57
 58static inline unsigned long pfn_to_block_id(unsigned long pfn)
 59{
 60	return memory_block_id(pfn_to_section_nr(pfn));
 61}
 62
 63static inline unsigned long phys_to_block_id(unsigned long phys)
 64{
 65	return pfn_to_block_id(PFN_DOWN(phys));
 66}
 67
 68static int memory_subsys_online(struct device *dev);
 69static int memory_subsys_offline(struct device *dev);
 70
 71static struct bus_type memory_subsys = {
 72	.name = MEMORY_CLASS_NAME,
 73	.dev_name = MEMORY_CLASS_NAME,
 74	.online = memory_subsys_online,
 75	.offline = memory_subsys_offline,
 76};
 77
 78/*
 79 * Memory blocks are cached in a local radix tree to avoid
 80 * a costly linear search for the corresponding device on
 81 * the subsystem bus.
 82 */
 83static DEFINE_XARRAY(memory_blocks);
 84
 85static BLOCKING_NOTIFIER_HEAD(memory_chain);
 86
 87int register_memory_notifier(struct notifier_block *nb)
 88{
 89	return blocking_notifier_chain_register(&memory_chain, nb);
 90}
 91EXPORT_SYMBOL(register_memory_notifier);
 92
 93void unregister_memory_notifier(struct notifier_block *nb)
 94{
 95	blocking_notifier_chain_unregister(&memory_chain, nb);
 96}
 97EXPORT_SYMBOL(unregister_memory_notifier);
 98
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 99static void memory_block_release(struct device *dev)
100{
101	struct memory_block *mem = to_memory_block(dev);
102
103	kfree(mem);
104}
105
106unsigned long __weak memory_block_size_bytes(void)
107{
108	return MIN_MEMORY_BLOCK_SIZE;
109}
110EXPORT_SYMBOL_GPL(memory_block_size_bytes);
111
112/*
113 * Show the first physical section index (number) of this memory block.
114 */
115static ssize_t phys_index_show(struct device *dev,
116			       struct device_attribute *attr, char *buf)
117{
118	struct memory_block *mem = to_memory_block(dev);
119	unsigned long phys_index;
120
121	phys_index = mem->start_section_nr / sections_per_block;
122	return sprintf(buf, "%08lx\n", phys_index);
123}
124
125/*
126 * Legacy interface that we cannot remove. Always indicate "removable"
127 * with CONFIG_MEMORY_HOTREMOVE - bad heuristic.
 
 
128 */
129static ssize_t removable_show(struct device *dev, struct device_attribute *attr,
130			      char *buf)
131{
132	return sprintf(buf, "%d\n", (int)IS_ENABLED(CONFIG_MEMORY_HOTREMOVE));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
133}
134
135/*
136 * online, offline, going offline, etc.
137 */
138static ssize_t state_show(struct device *dev, struct device_attribute *attr,
139			  char *buf)
140{
141	struct memory_block *mem = to_memory_block(dev);
142	ssize_t len = 0;
143
144	/*
145	 * We can probably put these states in a nice little array
146	 * so that they're not open-coded
147	 */
148	switch (mem->state) {
149	case MEM_ONLINE:
150		len = sprintf(buf, "online\n");
151		break;
152	case MEM_OFFLINE:
153		len = sprintf(buf, "offline\n");
154		break;
155	case MEM_GOING_OFFLINE:
156		len = sprintf(buf, "going-offline\n");
157		break;
158	default:
159		len = sprintf(buf, "ERROR-UNKNOWN-%ld\n",
160				mem->state);
161		WARN_ON(1);
162		break;
163	}
164
165	return len;
166}
167
168int memory_notify(unsigned long val, void *v)
169{
170	return blocking_notifier_call_chain(&memory_chain, val, v);
171}
172
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
173/*
174 * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is
175 * OK to have direct references to sparsemem variables in here.
176 */
177static int
178memory_block_action(unsigned long start_section_nr, unsigned long action,
179		    int online_type, int nid)
180{
181	unsigned long start_pfn;
182	unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
183	int ret;
184
185	start_pfn = section_nr_to_pfn(start_section_nr);
186
187	switch (action) {
188	case MEM_ONLINE:
189		ret = online_pages(start_pfn, nr_pages, online_type, nid);
 
 
 
190		break;
191	case MEM_OFFLINE:
192		ret = offline_pages(start_pfn, nr_pages);
193		break;
194	default:
195		WARN(1, KERN_WARNING "%s(%ld, %ld) unknown action: "
196		     "%ld\n", __func__, start_section_nr, action, action);
197		ret = -EINVAL;
198	}
199
200	return ret;
201}
202
203static int memory_block_change_state(struct memory_block *mem,
204		unsigned long to_state, unsigned long from_state_req)
205{
206	int ret = 0;
207
208	if (mem->state != from_state_req)
209		return -EINVAL;
210
211	if (to_state == MEM_OFFLINE)
212		mem->state = MEM_GOING_OFFLINE;
213
214	ret = memory_block_action(mem->start_section_nr, to_state,
215				  mem->online_type, mem->nid);
216
217	mem->state = ret ? from_state_req : to_state;
218
219	return ret;
220}
221
222/* The device lock serializes operations on memory_subsys_[online|offline] */
223static int memory_subsys_online(struct device *dev)
224{
225	struct memory_block *mem = to_memory_block(dev);
226	int ret;
227
228	if (mem->state == MEM_ONLINE)
229		return 0;
230
231	/*
232	 * When called via device_online() without configuring the online_type,
233	 * we want to default to MMOP_ONLINE.
 
234	 */
235	if (mem->online_type == MMOP_OFFLINE)
236		mem->online_type = MMOP_ONLINE;
237
238	ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE);
239	mem->online_type = MMOP_OFFLINE;
 
 
240
241	return ret;
242}
243
244static int memory_subsys_offline(struct device *dev)
245{
246	struct memory_block *mem = to_memory_block(dev);
247
248	if (mem->state == MEM_OFFLINE)
249		return 0;
250
 
 
 
 
251	return memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE);
252}
253
254static ssize_t state_store(struct device *dev, struct device_attribute *attr,
255			   const char *buf, size_t count)
256{
257	const int online_type = memhp_online_type_from_str(buf);
258	struct memory_block *mem = to_memory_block(dev);
259	int ret;
260
261	if (online_type < 0)
262		return -EINVAL;
263
264	ret = lock_device_hotplug_sysfs();
265	if (ret)
266		return ret;
267
 
 
 
 
 
 
 
 
 
 
 
 
 
268	switch (online_type) {
269	case MMOP_ONLINE_KERNEL:
270	case MMOP_ONLINE_MOVABLE:
271	case MMOP_ONLINE:
272		/* mem->online_type is protected by device_hotplug_lock */
273		mem->online_type = online_type;
274		ret = device_online(&mem->dev);
275		break;
276	case MMOP_OFFLINE:
277		ret = device_offline(&mem->dev);
278		break;
279	default:
280		ret = -EINVAL; /* should never happen */
281	}
282
 
283	unlock_device_hotplug();
284
285	if (ret < 0)
286		return ret;
287	if (ret)
288		return -EINVAL;
289
290	return count;
291}
292
293/*
294 * phys_device is a bad name for this.  What I really want
295 * is a way to differentiate between memory ranges that
296 * are part of physical devices that constitute
297 * a complete removable unit or fru.
298 * i.e. do these ranges belong to the same physical device,
299 * s.t. if I offline all of these sections I can then
300 * remove the physical device?
301 */
302static ssize_t phys_device_show(struct device *dev,
303				struct device_attribute *attr, char *buf)
304{
305	struct memory_block *mem = to_memory_block(dev);
306	return sprintf(buf, "%d\n", mem->phys_device);
307}
308
309#ifdef CONFIG_MEMORY_HOTREMOVE
310static void print_allowed_zone(char *buf, int nid, unsigned long start_pfn,
311		unsigned long nr_pages, int online_type,
312		struct zone *default_zone)
313{
314	struct zone *zone;
315
316	zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages);
317	if (zone != default_zone) {
318		strcat(buf, " ");
319		strcat(buf, zone->name);
320	}
321}
322
323static ssize_t valid_zones_show(struct device *dev,
324				struct device_attribute *attr, char *buf)
325{
326	struct memory_block *mem = to_memory_block(dev);
327	unsigned long start_pfn = section_nr_to_pfn(mem->start_section_nr);
328	unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
 
329	struct zone *default_zone;
330	int nid;
331
332	/*
333	 * Check the existing zone. Make sure that we do that only on the
334	 * online nodes otherwise the page_zone is not reliable
335	 */
336	if (mem->state == MEM_ONLINE) {
337		/*
338		 * The block contains more than one zone can not be offlined.
339		 * This can happen e.g. for ZONE_DMA and ZONE_DMA32
340		 */
341		default_zone = test_pages_in_a_zone(start_pfn,
342						    start_pfn + nr_pages);
343		if (!default_zone)
344			return sprintf(buf, "none\n");
345		strcat(buf, default_zone->name);
 
346		goto out;
347	}
348
349	nid = mem->nid;
350	default_zone = zone_for_pfn_range(MMOP_ONLINE, nid, start_pfn,
351					  nr_pages);
352	strcat(buf, default_zone->name);
353
354	print_allowed_zone(buf, nid, start_pfn, nr_pages, MMOP_ONLINE_KERNEL,
355			default_zone);
356	print_allowed_zone(buf, nid, start_pfn, nr_pages, MMOP_ONLINE_MOVABLE,
357			default_zone);
358out:
359	strcat(buf, "\n");
360
361	return strlen(buf);
362}
363static DEVICE_ATTR_RO(valid_zones);
364#endif
365
366static DEVICE_ATTR_RO(phys_index);
367static DEVICE_ATTR_RW(state);
368static DEVICE_ATTR_RO(phys_device);
369static DEVICE_ATTR_RO(removable);
370
371/*
372 * Show the memory block size (shared by all memory blocks).
373 */
374static ssize_t block_size_bytes_show(struct device *dev,
375				     struct device_attribute *attr, char *buf)
376{
377	return sprintf(buf, "%lx\n", memory_block_size_bytes());
378}
379
380static DEVICE_ATTR_RO(block_size_bytes);
381
382/*
383 * Memory auto online policy.
384 */
385
386static ssize_t auto_online_blocks_show(struct device *dev,
387				       struct device_attribute *attr, char *buf)
388{
389	return sprintf(buf, "%s\n",
390		       online_type_to_str[memhp_default_online_type]);
 
 
391}
392
393static ssize_t auto_online_blocks_store(struct device *dev,
394					struct device_attribute *attr,
395					const char *buf, size_t count)
396{
397	const int online_type = memhp_online_type_from_str(buf);
398
399	if (online_type < 0)
 
 
400		return -EINVAL;
401
402	memhp_default_online_type = online_type;
403	return count;
404}
405
406static DEVICE_ATTR_RW(auto_online_blocks);
407
408/*
409 * Some architectures will have custom drivers to do this, and
410 * will not need to do it from userspace.  The fake hot-add code
411 * as well as ppc64 will do all of their discovery in userspace
412 * and will require this interface.
413 */
414#ifdef CONFIG_ARCH_MEMORY_PROBE
415static ssize_t probe_store(struct device *dev, struct device_attribute *attr,
416			   const char *buf, size_t count)
417{
418	u64 phys_addr;
419	int nid, ret;
420	unsigned long pages_per_block = PAGES_PER_SECTION * sections_per_block;
421
422	ret = kstrtoull(buf, 0, &phys_addr);
423	if (ret)
424		return ret;
425
426	if (phys_addr & ((pages_per_block << PAGE_SHIFT) - 1))
427		return -EINVAL;
428
429	ret = lock_device_hotplug_sysfs();
430	if (ret)
431		return ret;
432
433	nid = memory_add_physaddr_to_nid(phys_addr);
434	ret = __add_memory(nid, phys_addr,
435			   MIN_MEMORY_BLOCK_SIZE * sections_per_block);
436
437	if (ret)
438		goto out;
439
440	ret = count;
441out:
442	unlock_device_hotplug();
443	return ret;
444}
445
446static DEVICE_ATTR_WO(probe);
447#endif
448
449#ifdef CONFIG_MEMORY_FAILURE
450/*
451 * Support for offlining pages of memory
452 */
453
454/* Soft offline a page */
455static ssize_t soft_offline_page_store(struct device *dev,
456				       struct device_attribute *attr,
457				       const char *buf, size_t count)
458{
459	int ret;
460	u64 pfn;
461	if (!capable(CAP_SYS_ADMIN))
462		return -EPERM;
463	if (kstrtoull(buf, 0, &pfn) < 0)
464		return -EINVAL;
465	pfn >>= PAGE_SHIFT;
466	ret = soft_offline_page(pfn, 0);
 
 
 
 
 
467	return ret == 0 ? count : ret;
468}
469
470/* Forcibly offline a page, including killing processes. */
471static ssize_t hard_offline_page_store(struct device *dev,
472				       struct device_attribute *attr,
473				       const char *buf, size_t count)
474{
475	int ret;
476	u64 pfn;
477	if (!capable(CAP_SYS_ADMIN))
478		return -EPERM;
479	if (kstrtoull(buf, 0, &pfn) < 0)
480		return -EINVAL;
481	pfn >>= PAGE_SHIFT;
482	ret = memory_failure(pfn, 0);
483	return ret ? ret : count;
484}
485
486static DEVICE_ATTR_WO(soft_offline_page);
487static DEVICE_ATTR_WO(hard_offline_page);
488#endif
489
490/*
491 * Note that phys_device is optional.  It is here to allow for
492 * differentiation between which *physical* devices each
493 * section belongs to...
494 */
495int __weak arch_get_memory_phys_device(unsigned long start_pfn)
496{
497	return 0;
498}
499
500/*
501 * A reference for the returned memory block device is acquired.
502 *
503 * Called under device_hotplug_lock.
504 */
505static struct memory_block *find_memory_block_by_id(unsigned long block_id)
506{
507	struct memory_block *mem;
508
509	mem = xa_load(&memory_blocks, block_id);
510	if (mem)
511		get_device(&mem->dev);
512	return mem;
513}
514
515/*
516 * Called under device_hotplug_lock.
 
 
 
 
 
517 */
518struct memory_block *find_memory_block(struct mem_section *section)
519{
520	unsigned long block_id = memory_block_id(__section_nr(section));
521
522	return find_memory_block_by_id(block_id);
523}
524
525static struct attribute *memory_memblk_attrs[] = {
526	&dev_attr_phys_index.attr,
527	&dev_attr_state.attr,
528	&dev_attr_phys_device.attr,
529	&dev_attr_removable.attr,
530#ifdef CONFIG_MEMORY_HOTREMOVE
531	&dev_attr_valid_zones.attr,
532#endif
533	NULL
534};
535
536static struct attribute_group memory_memblk_attr_group = {
537	.attrs = memory_memblk_attrs,
538};
539
540static const struct attribute_group *memory_memblk_attr_groups[] = {
541	&memory_memblk_attr_group,
542	NULL,
543};
544
545/*
546 * register_memory - Setup a sysfs device for a memory block
547 */
548static
549int register_memory(struct memory_block *memory)
550{
551	int ret;
552
553	memory->dev.bus = &memory_subsys;
554	memory->dev.id = memory->start_section_nr / sections_per_block;
555	memory->dev.release = memory_block_release;
556	memory->dev.groups = memory_memblk_attr_groups;
557	memory->dev.offline = memory->state == MEM_OFFLINE;
558
559	ret = device_register(&memory->dev);
560	if (ret) {
561		put_device(&memory->dev);
562		return ret;
563	}
564	ret = xa_err(xa_store(&memory_blocks, memory->dev.id, memory,
565			      GFP_KERNEL));
566	if (ret) {
567		put_device(&memory->dev);
568		device_unregister(&memory->dev);
569	}
570	return ret;
571}
572
573static int init_memory_block(unsigned long block_id, unsigned long state)
 
574{
575	struct memory_block *mem;
576	unsigned long start_pfn;
577	int ret = 0;
578
579	mem = find_memory_block_by_id(block_id);
580	if (mem) {
581		put_device(&mem->dev);
582		return -EEXIST;
583	}
584	mem = kzalloc(sizeof(*mem), GFP_KERNEL);
585	if (!mem)
586		return -ENOMEM;
587
588	mem->start_section_nr = block_id * sections_per_block;
589	mem->state = state;
590	start_pfn = section_nr_to_pfn(mem->start_section_nr);
591	mem->phys_device = arch_get_memory_phys_device(start_pfn);
592	mem->nid = NUMA_NO_NODE;
593
594	ret = register_memory(mem);
595
 
596	return ret;
597}
598
599static int add_memory_block(unsigned long base_section_nr)
600{
601	int section_count = 0;
 
602	unsigned long nr;
603
604	for (nr = base_section_nr; nr < base_section_nr + sections_per_block;
605	     nr++)
606		if (present_section_nr(nr))
607			section_count++;
608
609	if (section_count == 0)
610		return 0;
611	return init_memory_block(memory_block_id(base_section_nr),
612				 MEM_ONLINE);
 
 
 
 
613}
614
615static void unregister_memory(struct memory_block *memory)
616{
617	if (WARN_ON_ONCE(memory->dev.bus != &memory_subsys))
618		return;
619
620	WARN_ON(xa_erase(&memory_blocks, memory->dev.id) == NULL);
621
622	/* drop the ref. we got via find_memory_block() */
623	put_device(&memory->dev);
624	device_unregister(&memory->dev);
625}
626
627/*
628 * Create memory block devices for the given memory area. Start and size
629 * have to be aligned to memory block granularity. Memory block devices
630 * will be initialized as offline.
631 *
632 * Called under device_hotplug_lock.
633 */
634int create_memory_block_devices(unsigned long start, unsigned long size)
635{
636	const unsigned long start_block_id = pfn_to_block_id(PFN_DOWN(start));
637	unsigned long end_block_id = pfn_to_block_id(PFN_DOWN(start + size));
638	struct memory_block *mem;
639	unsigned long block_id;
640	int ret = 0;
641
642	if (WARN_ON_ONCE(!IS_ALIGNED(start, memory_block_size_bytes()) ||
643			 !IS_ALIGNED(size, memory_block_size_bytes())))
644		return -EINVAL;
645
 
646	for (block_id = start_block_id; block_id != end_block_id; block_id++) {
647		ret = init_memory_block(block_id, MEM_OFFLINE);
648		if (ret)
649			break;
 
650	}
651	if (ret) {
652		end_block_id = block_id;
653		for (block_id = start_block_id; block_id != end_block_id;
654		     block_id++) {
655			mem = find_memory_block_by_id(block_id);
656			if (WARN_ON_ONCE(!mem))
657				continue;
658			unregister_memory(mem);
659		}
660	}
 
661	return ret;
662}
663
664/*
665 * Remove memory block devices for the given memory area. Start and size
666 * have to be aligned to memory block granularity. Memory block devices
667 * have to be offline.
668 *
669 * Called under device_hotplug_lock.
670 */
671void remove_memory_block_devices(unsigned long start, unsigned long size)
672{
673	const unsigned long start_block_id = pfn_to_block_id(PFN_DOWN(start));
674	const unsigned long end_block_id = pfn_to_block_id(PFN_DOWN(start + size));
675	struct memory_block *mem;
676	unsigned long block_id;
677
678	if (WARN_ON_ONCE(!IS_ALIGNED(start, memory_block_size_bytes()) ||
679			 !IS_ALIGNED(size, memory_block_size_bytes())))
680		return;
681
 
682	for (block_id = start_block_id; block_id != end_block_id; block_id++) {
683		mem = find_memory_block_by_id(block_id);
684		if (WARN_ON_ONCE(!mem))
685			continue;
 
686		unregister_memory_block_under_nodes(mem);
687		unregister_memory(mem);
688	}
 
689}
690
691/* return true if the memory block is offlined, otherwise, return false */
692bool is_memblock_offlined(struct memory_block *mem)
693{
694	return mem->state == MEM_OFFLINE;
695}
696
697static struct attribute *memory_root_attrs[] = {
698#ifdef CONFIG_ARCH_MEMORY_PROBE
699	&dev_attr_probe.attr,
700#endif
701
702#ifdef CONFIG_MEMORY_FAILURE
703	&dev_attr_soft_offline_page.attr,
704	&dev_attr_hard_offline_page.attr,
705#endif
706
707	&dev_attr_block_size_bytes.attr,
708	&dev_attr_auto_online_blocks.attr,
709	NULL
710};
711
712static struct attribute_group memory_root_attr_group = {
713	.attrs = memory_root_attrs,
714};
715
716static const struct attribute_group *memory_root_attr_groups[] = {
717	&memory_root_attr_group,
718	NULL,
719};
720
721/*
722 * Initialize the sysfs support for memory devices. At the time this function
723 * is called, we cannot have concurrent creation/deletion of memory block
724 * devices, the device_hotplug_lock is not needed.
725 */
726void __init memory_dev_init(void)
727{
728	int ret;
 
729	unsigned long block_sz, nr;
730
731	/* Validate the configured memory block size */
732	block_sz = memory_block_size_bytes();
733	if (!is_power_of_2(block_sz) || block_sz < MIN_MEMORY_BLOCK_SIZE)
734		panic("Memory block size not suitable: 0x%lx\n", block_sz);
735	sections_per_block = block_sz / MIN_MEMORY_BLOCK_SIZE;
736
737	ret = subsys_system_register(&memory_subsys, memory_root_attr_groups);
738	if (ret)
739		panic("%s() failed to register subsystem: %d\n", __func__, ret);
740
741	/*
742	 * Create entries for memory sections that were found
743	 * during boot and have been initialized
744	 */
 
745	for (nr = 0; nr <= __highest_present_section_nr;
746	     nr += sections_per_block) {
747		ret = add_memory_block(nr);
748		if (ret)
749			panic("%s() failed to add memory block: %d\n", __func__,
750			      ret);
751	}
 
 
 
 
 
752}
753
754/**
755 * walk_memory_blocks - walk through all present memory blocks overlapped
756 *			by the range [start, start + size)
757 *
758 * @start: start address of the memory range
759 * @size: size of the memory range
760 * @arg: argument passed to func
761 * @func: callback for each memory section walked
762 *
763 * This function walks through all present memory blocks overlapped by the
764 * range [start, start + size), calling func on each memory block.
765 *
766 * In case func() returns an error, walking is aborted and the error is
767 * returned.
768 *
769 * Called under device_hotplug_lock.
770 */
771int walk_memory_blocks(unsigned long start, unsigned long size,
772		       void *arg, walk_memory_blocks_func_t func)
773{
774	const unsigned long start_block_id = phys_to_block_id(start);
775	const unsigned long end_block_id = phys_to_block_id(start + size - 1);
776	struct memory_block *mem;
777	unsigned long block_id;
778	int ret = 0;
779
780	if (!size)
781		return 0;
782
783	for (block_id = start_block_id; block_id <= end_block_id; block_id++) {
784		mem = find_memory_block_by_id(block_id);
785		if (!mem)
786			continue;
787
788		ret = func(mem, arg);
789		put_device(&mem->dev);
790		if (ret)
791			break;
792	}
793	return ret;
794}
795
796struct for_each_memory_block_cb_data {
797	walk_memory_blocks_func_t func;
798	void *arg;
799};
800
801static int for_each_memory_block_cb(struct device *dev, void *data)
802{
803	struct memory_block *mem = to_memory_block(dev);
804	struct for_each_memory_block_cb_data *cb_data = data;
805
806	return cb_data->func(mem, cb_data->arg);
807}
808
809/**
810 * for_each_memory_block - walk through all present memory blocks
811 *
812 * @arg: argument passed to func
813 * @func: callback for each memory block walked
814 *
815 * This function walks through all present memory blocks, calling func on
816 * each memory block.
817 *
818 * In case func() returns an error, walking is aborted and the error is
819 * returned.
820 */
821int for_each_memory_block(void *arg, walk_memory_blocks_func_t func)
822{
823	struct for_each_memory_block_cb_data cb_data = {
824		.func = func,
825		.arg = arg,
826	};
827
828	return bus_for_each_dev(&memory_subsys, NULL, &cb_data,
829				for_each_memory_block_cb);
830}
v5.4
  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 unsigned long base_memory_block_id(unsigned long section_nr)
 38{
 39	return section_nr / sections_per_block;
 40}
 41
 42static inline unsigned long pfn_to_block_id(unsigned long pfn)
 43{
 44	return base_memory_block_id(pfn_to_section_nr(pfn));
 45}
 46
 47static inline unsigned long phys_to_block_id(unsigned long phys)
 48{
 49	return pfn_to_block_id(PFN_DOWN(phys));
 50}
 51
 52static int memory_subsys_online(struct device *dev);
 53static int memory_subsys_offline(struct device *dev);
 54
 55static struct bus_type memory_subsys = {
 56	.name = MEMORY_CLASS_NAME,
 57	.dev_name = MEMORY_CLASS_NAME,
 58	.online = memory_subsys_online,
 59	.offline = memory_subsys_offline,
 60};
 61
 
 
 
 
 
 
 
 62static BLOCKING_NOTIFIER_HEAD(memory_chain);
 63
 64int register_memory_notifier(struct notifier_block *nb)
 65{
 66	return blocking_notifier_chain_register(&memory_chain, nb);
 67}
 68EXPORT_SYMBOL(register_memory_notifier);
 69
 70void unregister_memory_notifier(struct notifier_block *nb)
 71{
 72	blocking_notifier_chain_unregister(&memory_chain, nb);
 73}
 74EXPORT_SYMBOL(unregister_memory_notifier);
 75
 76static ATOMIC_NOTIFIER_HEAD(memory_isolate_chain);
 77
 78int register_memory_isolate_notifier(struct notifier_block *nb)
 79{
 80	return atomic_notifier_chain_register(&memory_isolate_chain, nb);
 81}
 82EXPORT_SYMBOL(register_memory_isolate_notifier);
 83
 84void unregister_memory_isolate_notifier(struct notifier_block *nb)
 85{
 86	atomic_notifier_chain_unregister(&memory_isolate_chain, nb);
 87}
 88EXPORT_SYMBOL(unregister_memory_isolate_notifier);
 89
 90static void memory_block_release(struct device *dev)
 91{
 92	struct memory_block *mem = to_memory_block(dev);
 93
 94	kfree(mem);
 95}
 96
 97unsigned long __weak memory_block_size_bytes(void)
 98{
 99	return MIN_MEMORY_BLOCK_SIZE;
100}
101EXPORT_SYMBOL_GPL(memory_block_size_bytes);
102
103/*
104 * Show the first physical section index (number) of this memory block.
105 */
106static ssize_t phys_index_show(struct device *dev,
107			       struct device_attribute *attr, char *buf)
108{
109	struct memory_block *mem = to_memory_block(dev);
110	unsigned long phys_index;
111
112	phys_index = mem->start_section_nr / sections_per_block;
113	return sprintf(buf, "%08lx\n", phys_index);
114}
115
116/*
117 * Show whether the memory block is likely to be offlineable (or is already
118 * offline). Once offline, the memory block could be removed. The return
119 * value does, however, not indicate that there is a way to remove the
120 * memory block.
121 */
122static ssize_t removable_show(struct device *dev, struct device_attribute *attr,
123			      char *buf)
124{
125	struct memory_block *mem = to_memory_block(dev);
126	unsigned long pfn;
127	int ret = 1, i;
128
129	if (mem->state != MEM_ONLINE)
130		goto out;
131
132	for (i = 0; i < sections_per_block; i++) {
133		if (!present_section_nr(mem->start_section_nr + i))
134			continue;
135		pfn = section_nr_to_pfn(mem->start_section_nr + i);
136		ret &= is_mem_section_removable(pfn, PAGES_PER_SECTION);
137	}
138
139out:
140	return sprintf(buf, "%d\n", ret);
141}
142
143/*
144 * online, offline, going offline, etc.
145 */
146static ssize_t state_show(struct device *dev, struct device_attribute *attr,
147			  char *buf)
148{
149	struct memory_block *mem = to_memory_block(dev);
150	ssize_t len = 0;
151
152	/*
153	 * We can probably put these states in a nice little array
154	 * so that they're not open-coded
155	 */
156	switch (mem->state) {
157	case MEM_ONLINE:
158		len = sprintf(buf, "online\n");
159		break;
160	case MEM_OFFLINE:
161		len = sprintf(buf, "offline\n");
162		break;
163	case MEM_GOING_OFFLINE:
164		len = sprintf(buf, "going-offline\n");
165		break;
166	default:
167		len = sprintf(buf, "ERROR-UNKNOWN-%ld\n",
168				mem->state);
169		WARN_ON(1);
170		break;
171	}
172
173	return len;
174}
175
176int memory_notify(unsigned long val, void *v)
177{
178	return blocking_notifier_call_chain(&memory_chain, val, v);
179}
180
181int memory_isolate_notify(unsigned long val, void *v)
182{
183	return atomic_notifier_call_chain(&memory_isolate_chain, val, v);
184}
185
186/*
187 * The probe routines leave the pages uninitialized, just as the bootmem code
188 * does. Make sure we do not access them, but instead use only information from
189 * within sections.
190 */
191static bool pages_correctly_probed(unsigned long start_pfn)
192{
193	unsigned long section_nr = pfn_to_section_nr(start_pfn);
194	unsigned long section_nr_end = section_nr + sections_per_block;
195	unsigned long pfn = start_pfn;
196
197	/*
198	 * memmap between sections is not contiguous except with
199	 * SPARSEMEM_VMEMMAP. We lookup the page once per section
200	 * and assume memmap is contiguous within each section
201	 */
202	for (; section_nr < section_nr_end; section_nr++) {
203		if (WARN_ON_ONCE(!pfn_valid(pfn)))
204			return false;
205
206		if (!present_section_nr(section_nr)) {
207			pr_warn("section %ld pfn[%lx, %lx) not present\n",
208				section_nr, pfn, pfn + PAGES_PER_SECTION);
209			return false;
210		} else if (!valid_section_nr(section_nr)) {
211			pr_warn("section %ld pfn[%lx, %lx) no valid memmap\n",
212				section_nr, pfn, pfn + PAGES_PER_SECTION);
213			return false;
214		} else if (online_section_nr(section_nr)) {
215			pr_warn("section %ld pfn[%lx, %lx) is already online\n",
216				section_nr, pfn, pfn + PAGES_PER_SECTION);
217			return false;
218		}
219		pfn += PAGES_PER_SECTION;
220	}
221
222	return true;
223}
224
225/*
226 * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is
227 * OK to have direct references to sparsemem variables in here.
228 */
229static int
230memory_block_action(unsigned long start_section_nr, unsigned long action,
231		    int online_type)
232{
233	unsigned long start_pfn;
234	unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
235	int ret;
236
237	start_pfn = section_nr_to_pfn(start_section_nr);
238
239	switch (action) {
240	case MEM_ONLINE:
241		if (!pages_correctly_probed(start_pfn))
242			return -EBUSY;
243
244		ret = online_pages(start_pfn, nr_pages, online_type);
245		break;
246	case MEM_OFFLINE:
247		ret = offline_pages(start_pfn, nr_pages);
248		break;
249	default:
250		WARN(1, KERN_WARNING "%s(%ld, %ld) unknown action: "
251		     "%ld\n", __func__, start_section_nr, action, action);
252		ret = -EINVAL;
253	}
254
255	return ret;
256}
257
258static int memory_block_change_state(struct memory_block *mem,
259		unsigned long to_state, unsigned long from_state_req)
260{
261	int ret = 0;
262
263	if (mem->state != from_state_req)
264		return -EINVAL;
265
266	if (to_state == MEM_OFFLINE)
267		mem->state = MEM_GOING_OFFLINE;
268
269	ret = memory_block_action(mem->start_section_nr, to_state,
270				mem->online_type);
271
272	mem->state = ret ? from_state_req : to_state;
273
274	return ret;
275}
276
277/* The device lock serializes operations on memory_subsys_[online|offline] */
278static int memory_subsys_online(struct device *dev)
279{
280	struct memory_block *mem = to_memory_block(dev);
281	int ret;
282
283	if (mem->state == MEM_ONLINE)
284		return 0;
285
286	/*
287	 * If we are called from state_store(), online_type will be
288	 * set >= 0 Otherwise we were called from the device online
289	 * attribute and need to set the online_type.
290	 */
291	if (mem->online_type < 0)
292		mem->online_type = MMOP_ONLINE_KEEP;
293
294	ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE);
295
296	/* clear online_type */
297	mem->online_type = -1;
298
299	return ret;
300}
301
302static int memory_subsys_offline(struct device *dev)
303{
304	struct memory_block *mem = to_memory_block(dev);
305
306	if (mem->state == MEM_OFFLINE)
307		return 0;
308
309	/* Can't offline block with non-present sections */
310	if (mem->section_count != sections_per_block)
311		return -EINVAL;
312
313	return memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE);
314}
315
316static ssize_t state_store(struct device *dev, struct device_attribute *attr,
317			   const char *buf, size_t count)
318{
 
319	struct memory_block *mem = to_memory_block(dev);
320	int ret, online_type;
 
 
 
321
322	ret = lock_device_hotplug_sysfs();
323	if (ret)
324		return ret;
325
326	if (sysfs_streq(buf, "online_kernel"))
327		online_type = MMOP_ONLINE_KERNEL;
328	else if (sysfs_streq(buf, "online_movable"))
329		online_type = MMOP_ONLINE_MOVABLE;
330	else if (sysfs_streq(buf, "online"))
331		online_type = MMOP_ONLINE_KEEP;
332	else if (sysfs_streq(buf, "offline"))
333		online_type = MMOP_OFFLINE;
334	else {
335		ret = -EINVAL;
336		goto err;
337	}
338
339	switch (online_type) {
340	case MMOP_ONLINE_KERNEL:
341	case MMOP_ONLINE_MOVABLE:
342	case MMOP_ONLINE_KEEP:
343		/* mem->online_type is protected by device_hotplug_lock */
344		mem->online_type = online_type;
345		ret = device_online(&mem->dev);
346		break;
347	case MMOP_OFFLINE:
348		ret = device_offline(&mem->dev);
349		break;
350	default:
351		ret = -EINVAL; /* should never happen */
352	}
353
354err:
355	unlock_device_hotplug();
356
357	if (ret < 0)
358		return ret;
359	if (ret)
360		return -EINVAL;
361
362	return count;
363}
364
365/*
366 * phys_device is a bad name for this.  What I really want
367 * is a way to differentiate between memory ranges that
368 * are part of physical devices that constitute
369 * a complete removable unit or fru.
370 * i.e. do these ranges belong to the same physical device,
371 * s.t. if I offline all of these sections I can then
372 * remove the physical device?
373 */
374static ssize_t phys_device_show(struct device *dev,
375				struct device_attribute *attr, char *buf)
376{
377	struct memory_block *mem = to_memory_block(dev);
378	return sprintf(buf, "%d\n", mem->phys_device);
379}
380
381#ifdef CONFIG_MEMORY_HOTREMOVE
382static void print_allowed_zone(char *buf, int nid, unsigned long start_pfn,
383		unsigned long nr_pages, int online_type,
384		struct zone *default_zone)
385{
386	struct zone *zone;
387
388	zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages);
389	if (zone != default_zone) {
390		strcat(buf, " ");
391		strcat(buf, zone->name);
392	}
393}
394
395static ssize_t valid_zones_show(struct device *dev,
396				struct device_attribute *attr, char *buf)
397{
398	struct memory_block *mem = to_memory_block(dev);
399	unsigned long start_pfn = section_nr_to_pfn(mem->start_section_nr);
400	unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
401	unsigned long valid_start_pfn, valid_end_pfn;
402	struct zone *default_zone;
403	int nid;
404
405	/*
406	 * Check the existing zone. Make sure that we do that only on the
407	 * online nodes otherwise the page_zone is not reliable
408	 */
409	if (mem->state == MEM_ONLINE) {
410		/*
411		 * The block contains more than one zone can not be offlined.
412		 * This can happen e.g. for ZONE_DMA and ZONE_DMA32
413		 */
414		if (!test_pages_in_a_zone(start_pfn, start_pfn + nr_pages,
415					  &valid_start_pfn, &valid_end_pfn))
 
416			return sprintf(buf, "none\n");
417		start_pfn = valid_start_pfn;
418		strcat(buf, page_zone(pfn_to_page(start_pfn))->name);
419		goto out;
420	}
421
422	nid = mem->nid;
423	default_zone = zone_for_pfn_range(MMOP_ONLINE_KEEP, nid, start_pfn, nr_pages);
 
424	strcat(buf, default_zone->name);
425
426	print_allowed_zone(buf, nid, start_pfn, nr_pages, MMOP_ONLINE_KERNEL,
427			default_zone);
428	print_allowed_zone(buf, nid, start_pfn, nr_pages, MMOP_ONLINE_MOVABLE,
429			default_zone);
430out:
431	strcat(buf, "\n");
432
433	return strlen(buf);
434}
435static DEVICE_ATTR_RO(valid_zones);
436#endif
437
438static DEVICE_ATTR_RO(phys_index);
439static DEVICE_ATTR_RW(state);
440static DEVICE_ATTR_RO(phys_device);
441static DEVICE_ATTR_RO(removable);
442
443/*
444 * Show the memory block size (shared by all memory blocks).
445 */
446static ssize_t block_size_bytes_show(struct device *dev,
447				     struct device_attribute *attr, char *buf)
448{
449	return sprintf(buf, "%lx\n", memory_block_size_bytes());
450}
451
452static DEVICE_ATTR_RO(block_size_bytes);
453
454/*
455 * Memory auto online policy.
456 */
457
458static ssize_t auto_online_blocks_show(struct device *dev,
459				       struct device_attribute *attr, char *buf)
460{
461	if (memhp_auto_online)
462		return sprintf(buf, "online\n");
463	else
464		return sprintf(buf, "offline\n");
465}
466
467static ssize_t auto_online_blocks_store(struct device *dev,
468					struct device_attribute *attr,
469					const char *buf, size_t count)
470{
471	if (sysfs_streq(buf, "online"))
472		memhp_auto_online = true;
473	else if (sysfs_streq(buf, "offline"))
474		memhp_auto_online = false;
475	else
476		return -EINVAL;
477
 
478	return count;
479}
480
481static DEVICE_ATTR_RW(auto_online_blocks);
482
483/*
484 * Some architectures will have custom drivers to do this, and
485 * will not need to do it from userspace.  The fake hot-add code
486 * as well as ppc64 will do all of their discovery in userspace
487 * and will require this interface.
488 */
489#ifdef CONFIG_ARCH_MEMORY_PROBE
490static ssize_t probe_store(struct device *dev, struct device_attribute *attr,
491			   const char *buf, size_t count)
492{
493	u64 phys_addr;
494	int nid, ret;
495	unsigned long pages_per_block = PAGES_PER_SECTION * sections_per_block;
496
497	ret = kstrtoull(buf, 0, &phys_addr);
498	if (ret)
499		return ret;
500
501	if (phys_addr & ((pages_per_block << PAGE_SHIFT) - 1))
502		return -EINVAL;
503
504	ret = lock_device_hotplug_sysfs();
505	if (ret)
506		return ret;
507
508	nid = memory_add_physaddr_to_nid(phys_addr);
509	ret = __add_memory(nid, phys_addr,
510			   MIN_MEMORY_BLOCK_SIZE * sections_per_block);
511
512	if (ret)
513		goto out;
514
515	ret = count;
516out:
517	unlock_device_hotplug();
518	return ret;
519}
520
521static DEVICE_ATTR_WO(probe);
522#endif
523
524#ifdef CONFIG_MEMORY_FAILURE
525/*
526 * Support for offlining pages of memory
527 */
528
529/* Soft offline a page */
530static ssize_t soft_offline_page_store(struct device *dev,
531				       struct device_attribute *attr,
532				       const char *buf, size_t count)
533{
534	int ret;
535	u64 pfn;
536	if (!capable(CAP_SYS_ADMIN))
537		return -EPERM;
538	if (kstrtoull(buf, 0, &pfn) < 0)
539		return -EINVAL;
540	pfn >>= PAGE_SHIFT;
541	if (!pfn_valid(pfn))
542		return -ENXIO;
543	/* Only online pages can be soft-offlined (esp., not ZONE_DEVICE). */
544	if (!pfn_to_online_page(pfn))
545		return -EIO;
546	ret = soft_offline_page(pfn_to_page(pfn), 0);
547	return ret == 0 ? count : ret;
548}
549
550/* Forcibly offline a page, including killing processes. */
551static ssize_t hard_offline_page_store(struct device *dev,
552				       struct device_attribute *attr,
553				       const char *buf, size_t count)
554{
555	int ret;
556	u64 pfn;
557	if (!capable(CAP_SYS_ADMIN))
558		return -EPERM;
559	if (kstrtoull(buf, 0, &pfn) < 0)
560		return -EINVAL;
561	pfn >>= PAGE_SHIFT;
562	ret = memory_failure(pfn, 0);
563	return ret ? ret : count;
564}
565
566static DEVICE_ATTR_WO(soft_offline_page);
567static DEVICE_ATTR_WO(hard_offline_page);
568#endif
569
570/*
571 * Note that phys_device is optional.  It is here to allow for
572 * differentiation between which *physical* devices each
573 * section belongs to...
574 */
575int __weak arch_get_memory_phys_device(unsigned long start_pfn)
576{
577	return 0;
578}
579
580/* A reference for the returned memory block device is acquired. */
 
 
 
 
581static struct memory_block *find_memory_block_by_id(unsigned long block_id)
582{
583	struct device *dev;
584
585	dev = subsys_find_device_by_id(&memory_subsys, block_id, NULL);
586	return dev ? to_memory_block(dev) : NULL;
 
 
587}
588
589/*
590 * For now, we have a linear search to go find the appropriate
591 * memory_block corresponding to a particular phys_index. If
592 * this gets to be a real problem, we can always use a radix
593 * tree or something here.
594 *
595 * This could be made generic for all device subsystems.
596 */
597struct memory_block *find_memory_block(struct mem_section *section)
598{
599	unsigned long block_id = base_memory_block_id(__section_nr(section));
600
601	return find_memory_block_by_id(block_id);
602}
603
604static struct attribute *memory_memblk_attrs[] = {
605	&dev_attr_phys_index.attr,
606	&dev_attr_state.attr,
607	&dev_attr_phys_device.attr,
608	&dev_attr_removable.attr,
609#ifdef CONFIG_MEMORY_HOTREMOVE
610	&dev_attr_valid_zones.attr,
611#endif
612	NULL
613};
614
615static struct attribute_group memory_memblk_attr_group = {
616	.attrs = memory_memblk_attrs,
617};
618
619static const struct attribute_group *memory_memblk_attr_groups[] = {
620	&memory_memblk_attr_group,
621	NULL,
622};
623
624/*
625 * register_memory - Setup a sysfs device for a memory block
626 */
627static
628int register_memory(struct memory_block *memory)
629{
630	int ret;
631
632	memory->dev.bus = &memory_subsys;
633	memory->dev.id = memory->start_section_nr / sections_per_block;
634	memory->dev.release = memory_block_release;
635	memory->dev.groups = memory_memblk_attr_groups;
636	memory->dev.offline = memory->state == MEM_OFFLINE;
637
638	ret = device_register(&memory->dev);
639	if (ret)
 
 
 
 
 
 
640		put_device(&memory->dev);
641
 
642	return ret;
643}
644
645static int init_memory_block(struct memory_block **memory,
646			     unsigned long block_id, unsigned long state)
647{
648	struct memory_block *mem;
649	unsigned long start_pfn;
650	int ret = 0;
651
652	mem = find_memory_block_by_id(block_id);
653	if (mem) {
654		put_device(&mem->dev);
655		return -EEXIST;
656	}
657	mem = kzalloc(sizeof(*mem), GFP_KERNEL);
658	if (!mem)
659		return -ENOMEM;
660
661	mem->start_section_nr = block_id * sections_per_block;
662	mem->state = state;
663	start_pfn = section_nr_to_pfn(mem->start_section_nr);
664	mem->phys_device = arch_get_memory_phys_device(start_pfn);
665	mem->nid = NUMA_NO_NODE;
666
667	ret = register_memory(mem);
668
669	*memory = mem;
670	return ret;
671}
672
673static int add_memory_block(unsigned long base_section_nr)
674{
675	int ret, section_count = 0;
676	struct memory_block *mem;
677	unsigned long nr;
678
679	for (nr = base_section_nr; nr < base_section_nr + sections_per_block;
680	     nr++)
681		if (present_section_nr(nr))
682			section_count++;
683
684	if (section_count == 0)
685		return 0;
686	ret = init_memory_block(&mem, base_memory_block_id(base_section_nr),
687				MEM_ONLINE);
688	if (ret)
689		return ret;
690	mem->section_count = section_count;
691	return 0;
692}
693
694static void unregister_memory(struct memory_block *memory)
695{
696	if (WARN_ON_ONCE(memory->dev.bus != &memory_subsys))
697		return;
698
 
 
699	/* drop the ref. we got via find_memory_block() */
700	put_device(&memory->dev);
701	device_unregister(&memory->dev);
702}
703
704/*
705 * Create memory block devices for the given memory area. Start and size
706 * have to be aligned to memory block granularity. Memory block devices
707 * will be initialized as offline.
 
 
708 */
709int create_memory_block_devices(unsigned long start, unsigned long size)
710{
711	const unsigned long start_block_id = pfn_to_block_id(PFN_DOWN(start));
712	unsigned long end_block_id = pfn_to_block_id(PFN_DOWN(start + size));
713	struct memory_block *mem;
714	unsigned long block_id;
715	int ret = 0;
716
717	if (WARN_ON_ONCE(!IS_ALIGNED(start, memory_block_size_bytes()) ||
718			 !IS_ALIGNED(size, memory_block_size_bytes())))
719		return -EINVAL;
720
721	mutex_lock(&mem_sysfs_mutex);
722	for (block_id = start_block_id; block_id != end_block_id; block_id++) {
723		ret = init_memory_block(&mem, block_id, MEM_OFFLINE);
724		if (ret)
725			break;
726		mem->section_count = sections_per_block;
727	}
728	if (ret) {
729		end_block_id = block_id;
730		for (block_id = start_block_id; block_id != end_block_id;
731		     block_id++) {
732			mem = find_memory_block_by_id(block_id);
733			mem->section_count = 0;
 
734			unregister_memory(mem);
735		}
736	}
737	mutex_unlock(&mem_sysfs_mutex);
738	return ret;
739}
740
741/*
742 * Remove memory block devices for the given memory area. Start and size
743 * have to be aligned to memory block granularity. Memory block devices
744 * have to be offline.
 
 
745 */
746void remove_memory_block_devices(unsigned long start, unsigned long size)
747{
748	const unsigned long start_block_id = pfn_to_block_id(PFN_DOWN(start));
749	const unsigned long end_block_id = pfn_to_block_id(PFN_DOWN(start + size));
750	struct memory_block *mem;
751	unsigned long block_id;
752
753	if (WARN_ON_ONCE(!IS_ALIGNED(start, memory_block_size_bytes()) ||
754			 !IS_ALIGNED(size, memory_block_size_bytes())))
755		return;
756
757	mutex_lock(&mem_sysfs_mutex);
758	for (block_id = start_block_id; block_id != end_block_id; block_id++) {
759		mem = find_memory_block_by_id(block_id);
760		if (WARN_ON_ONCE(!mem))
761			continue;
762		mem->section_count = 0;
763		unregister_memory_block_under_nodes(mem);
764		unregister_memory(mem);
765	}
766	mutex_unlock(&mem_sysfs_mutex);
767}
768
769/* return true if the memory block is offlined, otherwise, return false */
770bool is_memblock_offlined(struct memory_block *mem)
771{
772	return mem->state == MEM_OFFLINE;
773}
774
775static struct attribute *memory_root_attrs[] = {
776#ifdef CONFIG_ARCH_MEMORY_PROBE
777	&dev_attr_probe.attr,
778#endif
779
780#ifdef CONFIG_MEMORY_FAILURE
781	&dev_attr_soft_offline_page.attr,
782	&dev_attr_hard_offline_page.attr,
783#endif
784
785	&dev_attr_block_size_bytes.attr,
786	&dev_attr_auto_online_blocks.attr,
787	NULL
788};
789
790static struct attribute_group memory_root_attr_group = {
791	.attrs = memory_root_attrs,
792};
793
794static const struct attribute_group *memory_root_attr_groups[] = {
795	&memory_root_attr_group,
796	NULL,
797};
798
799/*
800 * Initialize the sysfs support for memory devices...
 
 
801 */
802void __init memory_dev_init(void)
803{
804	int ret;
805	int err;
806	unsigned long block_sz, nr;
807
808	/* Validate the configured memory block size */
809	block_sz = memory_block_size_bytes();
810	if (!is_power_of_2(block_sz) || block_sz < MIN_MEMORY_BLOCK_SIZE)
811		panic("Memory block size not suitable: 0x%lx\n", block_sz);
812	sections_per_block = block_sz / MIN_MEMORY_BLOCK_SIZE;
813
814	ret = subsys_system_register(&memory_subsys, memory_root_attr_groups);
815	if (ret)
816		goto out;
817
818	/*
819	 * Create entries for memory sections that were found
820	 * during boot and have been initialized
821	 */
822	mutex_lock(&mem_sysfs_mutex);
823	for (nr = 0; nr <= __highest_present_section_nr;
824	     nr += sections_per_block) {
825		err = add_memory_block(nr);
826		if (!ret)
827			ret = err;
 
828	}
829	mutex_unlock(&mem_sysfs_mutex);
830
831out:
832	if (ret)
833		panic("%s() failed: %d\n", __func__, ret);
834}
835
836/**
837 * walk_memory_blocks - walk through all present memory blocks overlapped
838 *			by the range [start, start + size)
839 *
840 * @start: start address of the memory range
841 * @size: size of the memory range
842 * @arg: argument passed to func
843 * @func: callback for each memory section walked
844 *
845 * This function walks through all present memory blocks overlapped by the
846 * range [start, start + size), calling func on each memory block.
847 *
848 * In case func() returns an error, walking is aborted and the error is
849 * returned.
 
 
850 */
851int walk_memory_blocks(unsigned long start, unsigned long size,
852		       void *arg, walk_memory_blocks_func_t func)
853{
854	const unsigned long start_block_id = phys_to_block_id(start);
855	const unsigned long end_block_id = phys_to_block_id(start + size - 1);
856	struct memory_block *mem;
857	unsigned long block_id;
858	int ret = 0;
859
860	if (!size)
861		return 0;
862
863	for (block_id = start_block_id; block_id <= end_block_id; block_id++) {
864		mem = find_memory_block_by_id(block_id);
865		if (!mem)
866			continue;
867
868		ret = func(mem, arg);
869		put_device(&mem->dev);
870		if (ret)
871			break;
872	}
873	return ret;
874}
875
876struct for_each_memory_block_cb_data {
877	walk_memory_blocks_func_t func;
878	void *arg;
879};
880
881static int for_each_memory_block_cb(struct device *dev, void *data)
882{
883	struct memory_block *mem = to_memory_block(dev);
884	struct for_each_memory_block_cb_data *cb_data = data;
885
886	return cb_data->func(mem, cb_data->arg);
887}
888
889/**
890 * for_each_memory_block - walk through all present memory blocks
891 *
892 * @arg: argument passed to func
893 * @func: callback for each memory block walked
894 *
895 * This function walks through all present memory blocks, calling func on
896 * each memory block.
897 *
898 * In case func() returns an error, walking is aborted and the error is
899 * returned.
900 */
901int for_each_memory_block(void *arg, walk_memory_blocks_func_t func)
902{
903	struct for_each_memory_block_cb_data cb_data = {
904		.func = func,
905		.arg = arg,
906	};
907
908	return bus_for_each_dev(&memory_subsys, NULL, &cb_data,
909				for_each_memory_block_cb);
910}