Loading...
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}
1/*
2 * Memory subsystem support
3 *
4 * Written by Matt Tolentino <matthew.e.tolentino@intel.com>
5 * Dave Hansen <haveblue@us.ibm.com>
6 *
7 * This file provides the necessary infrastructure to represent
8 * a SPARSEMEM-memory-model system's physical memory in /sysfs.
9 * All arch-independent code that assumes MEMORY_HOTPLUG requires
10 * SPARSEMEM should be contained here, or in mm/memory_hotplug.c.
11 */
12
13#include <linux/module.h>
14#include <linux/init.h>
15#include <linux/topology.h>
16#include <linux/capability.h>
17#include <linux/device.h>
18#include <linux/memory.h>
19#include <linux/memory_hotplug.h>
20#include <linux/mm.h>
21#include <linux/mutex.h>
22#include <linux/stat.h>
23#include <linux/slab.h>
24
25#include <linux/atomic.h>
26#include <asm/uaccess.h>
27
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 struct page *first_page;
230 int ret;
231
232 start_pfn = section_nr_to_pfn(phys_index);
233 first_page = pfn_to_page(start_pfn);
234
235 switch (action) {
236 case MEM_ONLINE:
237 if (!pages_correctly_reserved(start_pfn))
238 return -EBUSY;
239
240 ret = online_pages(start_pfn, nr_pages, online_type);
241 break;
242 case MEM_OFFLINE:
243 ret = offline_pages(start_pfn, nr_pages);
244 break;
245 default:
246 WARN(1, KERN_WARNING "%s(%ld, %ld) unknown action: "
247 "%ld\n", __func__, phys_index, action, action);
248 ret = -EINVAL;
249 }
250
251 return ret;
252}
253
254int memory_block_change_state(struct memory_block *mem,
255 unsigned long to_state, unsigned long from_state_req)
256{
257 int ret = 0;
258
259 if (mem->state != from_state_req)
260 return -EINVAL;
261
262 if (to_state == MEM_OFFLINE)
263 mem->state = MEM_GOING_OFFLINE;
264
265 ret = memory_block_action(mem->start_section_nr, to_state,
266 mem->online_type);
267
268 mem->state = ret ? from_state_req : to_state;
269
270 return ret;
271}
272
273/* The device lock serializes operations on memory_subsys_[online|offline] */
274static int memory_subsys_online(struct device *dev)
275{
276 struct memory_block *mem = to_memory_block(dev);
277 int ret;
278
279 if (mem->state == MEM_ONLINE)
280 return 0;
281
282 /*
283 * If we are called from store_mem_state(), online_type will be
284 * set >= 0 Otherwise we were called from the device online
285 * attribute and need to set the online_type.
286 */
287 if (mem->online_type < 0)
288 mem->online_type = MMOP_ONLINE_KEEP;
289
290 /* Already under protection of mem_hotplug_begin() */
291 ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE);
292
293 /* clear online_type */
294 mem->online_type = -1;
295
296 return ret;
297}
298
299static int memory_subsys_offline(struct device *dev)
300{
301 struct memory_block *mem = to_memory_block(dev);
302
303 if (mem->state == MEM_OFFLINE)
304 return 0;
305
306 /* Can't offline block with non-present sections */
307 if (mem->section_count != sections_per_block)
308 return -EINVAL;
309
310 return memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE);
311}
312
313static ssize_t
314store_mem_state(struct device *dev,
315 struct device_attribute *attr, const char *buf, size_t count)
316{
317 struct memory_block *mem = to_memory_block(dev);
318 int ret, online_type;
319
320 ret = lock_device_hotplug_sysfs();
321 if (ret)
322 return ret;
323
324 if (sysfs_streq(buf, "online_kernel"))
325 online_type = MMOP_ONLINE_KERNEL;
326 else if (sysfs_streq(buf, "online_movable"))
327 online_type = MMOP_ONLINE_MOVABLE;
328 else if (sysfs_streq(buf, "online"))
329 online_type = MMOP_ONLINE_KEEP;
330 else if (sysfs_streq(buf, "offline"))
331 online_type = MMOP_OFFLINE;
332 else {
333 ret = -EINVAL;
334 goto err;
335 }
336
337 /*
338 * Memory hotplug needs to hold mem_hotplug_begin() for probe to find
339 * the correct memory block to online before doing device_online(dev),
340 * which will take dev->mutex. Take the lock early to prevent an
341 * inversion, memory_subsys_online() callbacks will be implemented by
342 * assuming it's already protected.
343 */
344 mem_hotplug_begin();
345
346 switch (online_type) {
347 case MMOP_ONLINE_KERNEL:
348 case MMOP_ONLINE_MOVABLE:
349 case MMOP_ONLINE_KEEP:
350 mem->online_type = online_type;
351 ret = device_online(&mem->dev);
352 break;
353 case MMOP_OFFLINE:
354 ret = device_offline(&mem->dev);
355 break;
356 default:
357 ret = -EINVAL; /* should never happen */
358 }
359
360 mem_hotplug_done();
361err:
362 unlock_device_hotplug();
363
364 if (ret)
365 return ret;
366 return count;
367}
368
369/*
370 * phys_device is a bad name for this. What I really want
371 * is a way to differentiate between memory ranges that
372 * are part of physical devices that constitute
373 * a complete removable unit or fru.
374 * i.e. do these ranges belong to the same physical device,
375 * s.t. if I offline all of these sections I can then
376 * remove the physical device?
377 */
378static ssize_t show_phys_device(struct device *dev,
379 struct device_attribute *attr, char *buf)
380{
381 struct memory_block *mem = to_memory_block(dev);
382 return sprintf(buf, "%d\n", mem->phys_device);
383}
384
385#ifdef CONFIG_MEMORY_HOTREMOVE
386static ssize_t show_valid_zones(struct device *dev,
387 struct device_attribute *attr, char *buf)
388{
389 struct memory_block *mem = to_memory_block(dev);
390 unsigned long start_pfn, end_pfn;
391 unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
392 struct page *first_page;
393 struct zone *zone;
394
395 start_pfn = section_nr_to_pfn(mem->start_section_nr);
396 end_pfn = start_pfn + nr_pages;
397 first_page = pfn_to_page(start_pfn);
398
399 /* The block contains more than one zone can not be offlined. */
400 if (!test_pages_in_a_zone(start_pfn, end_pfn))
401 return sprintf(buf, "none\n");
402
403 zone = page_zone(first_page);
404
405 if (zone_idx(zone) == ZONE_MOVABLE - 1) {
406 /*The mem block is the last memoryblock of this zone.*/
407 if (end_pfn == zone_end_pfn(zone))
408 return sprintf(buf, "%s %s\n",
409 zone->name, (zone + 1)->name);
410 }
411
412 if (zone_idx(zone) == ZONE_MOVABLE) {
413 /*The mem block is the first memoryblock of ZONE_MOVABLE.*/
414 if (start_pfn == zone->zone_start_pfn)
415 return sprintf(buf, "%s %s\n",
416 zone->name, (zone - 1)->name);
417 }
418
419 return sprintf(buf, "%s\n", zone->name);
420}
421static DEVICE_ATTR(valid_zones, 0444, show_valid_zones, NULL);
422#endif
423
424static DEVICE_ATTR(phys_index, 0444, show_mem_start_phys_index, NULL);
425static DEVICE_ATTR(state, 0644, show_mem_state, store_mem_state);
426static DEVICE_ATTR(phys_device, 0444, show_phys_device, NULL);
427static DEVICE_ATTR(removable, 0444, show_mem_removable, NULL);
428
429/*
430 * Block size attribute stuff
431 */
432static ssize_t
433print_block_size(struct device *dev, struct device_attribute *attr,
434 char *buf)
435{
436 return sprintf(buf, "%lx\n", get_memory_block_size());
437}
438
439static DEVICE_ATTR(block_size_bytes, 0444, print_block_size, NULL);
440
441/*
442 * Memory auto online policy.
443 */
444
445static ssize_t
446show_auto_online_blocks(struct device *dev, struct device_attribute *attr,
447 char *buf)
448{
449 if (memhp_auto_online)
450 return sprintf(buf, "online\n");
451 else
452 return sprintf(buf, "offline\n");
453}
454
455static ssize_t
456store_auto_online_blocks(struct device *dev, struct device_attribute *attr,
457 const char *buf, size_t count)
458{
459 if (sysfs_streq(buf, "online"))
460 memhp_auto_online = true;
461 else if (sysfs_streq(buf, "offline"))
462 memhp_auto_online = false;
463 else
464 return -EINVAL;
465
466 return count;
467}
468
469static DEVICE_ATTR(auto_online_blocks, 0644, show_auto_online_blocks,
470 store_auto_online_blocks);
471
472/*
473 * Some architectures will have custom drivers to do this, and
474 * will not need to do it from userspace. The fake hot-add code
475 * as well as ppc64 will do all of their discovery in userspace
476 * and will require this interface.
477 */
478#ifdef CONFIG_ARCH_MEMORY_PROBE
479static ssize_t
480memory_probe_store(struct device *dev, struct device_attribute *attr,
481 const char *buf, size_t count)
482{
483 u64 phys_addr;
484 int nid, ret;
485 unsigned long pages_per_block = PAGES_PER_SECTION * sections_per_block;
486
487 ret = kstrtoull(buf, 0, &phys_addr);
488 if (ret)
489 return ret;
490
491 if (phys_addr & ((pages_per_block << PAGE_SHIFT) - 1))
492 return -EINVAL;
493
494 nid = memory_add_physaddr_to_nid(phys_addr);
495 ret = add_memory(nid, phys_addr,
496 MIN_MEMORY_BLOCK_SIZE * sections_per_block);
497
498 if (ret)
499 goto out;
500
501 ret = count;
502out:
503 return ret;
504}
505
506static DEVICE_ATTR(probe, S_IWUSR, NULL, memory_probe_store);
507#endif
508
509#ifdef CONFIG_MEMORY_FAILURE
510/*
511 * Support for offlining pages of memory
512 */
513
514/* Soft offline a page */
515static ssize_t
516store_soft_offline_page(struct device *dev,
517 struct device_attribute *attr,
518 const char *buf, size_t count)
519{
520 int ret;
521 u64 pfn;
522 if (!capable(CAP_SYS_ADMIN))
523 return -EPERM;
524 if (kstrtoull(buf, 0, &pfn) < 0)
525 return -EINVAL;
526 pfn >>= PAGE_SHIFT;
527 if (!pfn_valid(pfn))
528 return -ENXIO;
529 ret = soft_offline_page(pfn_to_page(pfn), 0);
530 return ret == 0 ? count : ret;
531}
532
533/* Forcibly offline a page, including killing processes. */
534static ssize_t
535store_hard_offline_page(struct device *dev,
536 struct device_attribute *attr,
537 const char *buf, size_t count)
538{
539 int ret;
540 u64 pfn;
541 if (!capable(CAP_SYS_ADMIN))
542 return -EPERM;
543 if (kstrtoull(buf, 0, &pfn) < 0)
544 return -EINVAL;
545 pfn >>= PAGE_SHIFT;
546 ret = memory_failure(pfn, 0, 0);
547 return ret ? ret : count;
548}
549
550static DEVICE_ATTR(soft_offline_page, S_IWUSR, NULL, store_soft_offline_page);
551static DEVICE_ATTR(hard_offline_page, S_IWUSR, NULL, store_hard_offline_page);
552#endif
553
554/*
555 * Note that phys_device is optional. It is here to allow for
556 * differentiation between which *physical* devices each
557 * section belongs to...
558 */
559int __weak arch_get_memory_phys_device(unsigned long start_pfn)
560{
561 return 0;
562}
563
564/*
565 * A reference for the returned object is held and the reference for the
566 * hinted object is released.
567 */
568struct memory_block *find_memory_block_hinted(struct mem_section *section,
569 struct memory_block *hint)
570{
571 int block_id = base_memory_block_id(__section_nr(section));
572 struct device *hintdev = hint ? &hint->dev : NULL;
573 struct device *dev;
574
575 dev = subsys_find_device_by_id(&memory_subsys, block_id, hintdev);
576 if (hint)
577 put_device(&hint->dev);
578 if (!dev)
579 return NULL;
580 return to_memory_block(dev);
581}
582
583/*
584 * For now, we have a linear search to go find the appropriate
585 * memory_block corresponding to a particular phys_index. If
586 * this gets to be a real problem, we can always use a radix
587 * tree or something here.
588 *
589 * This could be made generic for all device subsystems.
590 */
591struct memory_block *find_memory_block(struct mem_section *section)
592{
593 return find_memory_block_hinted(section, NULL);
594}
595
596static struct attribute *memory_memblk_attrs[] = {
597 &dev_attr_phys_index.attr,
598 &dev_attr_state.attr,
599 &dev_attr_phys_device.attr,
600 &dev_attr_removable.attr,
601#ifdef CONFIG_MEMORY_HOTREMOVE
602 &dev_attr_valid_zones.attr,
603#endif
604 NULL
605};
606
607static struct attribute_group memory_memblk_attr_group = {
608 .attrs = memory_memblk_attrs,
609};
610
611static const struct attribute_group *memory_memblk_attr_groups[] = {
612 &memory_memblk_attr_group,
613 NULL,
614};
615
616/*
617 * register_memory - Setup a sysfs device for a memory block
618 */
619static
620int register_memory(struct memory_block *memory)
621{
622 memory->dev.bus = &memory_subsys;
623 memory->dev.id = memory->start_section_nr / sections_per_block;
624 memory->dev.release = memory_block_release;
625 memory->dev.groups = memory_memblk_attr_groups;
626 memory->dev.offline = memory->state == MEM_OFFLINE;
627
628 return device_register(&memory->dev);
629}
630
631static int init_memory_block(struct memory_block **memory,
632 struct mem_section *section, unsigned long state)
633{
634 struct memory_block *mem;
635 unsigned long start_pfn;
636 int scn_nr;
637 int ret = 0;
638
639 mem = kzalloc(sizeof(*mem), GFP_KERNEL);
640 if (!mem)
641 return -ENOMEM;
642
643 scn_nr = __section_nr(section);
644 mem->start_section_nr =
645 base_memory_block_id(scn_nr) * sections_per_block;
646 mem->end_section_nr = mem->start_section_nr + sections_per_block - 1;
647 mem->state = state;
648 start_pfn = section_nr_to_pfn(mem->start_section_nr);
649 mem->phys_device = arch_get_memory_phys_device(start_pfn);
650
651 ret = register_memory(mem);
652
653 *memory = mem;
654 return ret;
655}
656
657static int add_memory_block(int base_section_nr)
658{
659 struct memory_block *mem;
660 int i, ret, section_count = 0, section_nr;
661
662 for (i = base_section_nr;
663 (i < base_section_nr + sections_per_block) && i < NR_MEM_SECTIONS;
664 i++) {
665 if (!present_section_nr(i))
666 continue;
667 if (section_count == 0)
668 section_nr = i;
669 section_count++;
670 }
671
672 if (section_count == 0)
673 return 0;
674 ret = init_memory_block(&mem, __nr_to_section(section_nr), MEM_ONLINE);
675 if (ret)
676 return ret;
677 mem->section_count = section_count;
678 return 0;
679}
680
681static bool is_zone_device_section(struct mem_section *ms)
682{
683 struct page *page;
684
685 page = sparse_decode_mem_map(ms->section_mem_map, __section_nr(ms));
686 return is_zone_device_page(page);
687}
688
689/*
690 * need an interface for the VM to add new memory regions,
691 * but without onlining it.
692 */
693int register_new_memory(int nid, struct mem_section *section)
694{
695 int ret = 0;
696 struct memory_block *mem;
697
698 if (is_zone_device_section(section))
699 return 0;
700
701 mutex_lock(&mem_sysfs_mutex);
702
703 mem = find_memory_block(section);
704 if (mem) {
705 mem->section_count++;
706 put_device(&mem->dev);
707 } else {
708 ret = init_memory_block(&mem, section, MEM_OFFLINE);
709 if (ret)
710 goto out;
711 mem->section_count++;
712 }
713
714 if (mem->section_count == sections_per_block)
715 ret = register_mem_sect_under_node(mem, nid);
716out:
717 mutex_unlock(&mem_sysfs_mutex);
718 return ret;
719}
720
721#ifdef CONFIG_MEMORY_HOTREMOVE
722static void
723unregister_memory(struct memory_block *memory)
724{
725 BUG_ON(memory->dev.bus != &memory_subsys);
726
727 /* drop the ref. we got in remove_memory_block() */
728 put_device(&memory->dev);
729 device_unregister(&memory->dev);
730}
731
732static int remove_memory_section(unsigned long node_id,
733 struct mem_section *section, int phys_device)
734{
735 struct memory_block *mem;
736
737 if (is_zone_device_section(section))
738 return 0;
739
740 mutex_lock(&mem_sysfs_mutex);
741 mem = find_memory_block(section);
742 unregister_mem_sect_under_nodes(mem, __section_nr(section));
743
744 mem->section_count--;
745 if (mem->section_count == 0)
746 unregister_memory(mem);
747 else
748 put_device(&mem->dev);
749
750 mutex_unlock(&mem_sysfs_mutex);
751 return 0;
752}
753
754int unregister_memory_section(struct mem_section *section)
755{
756 if (!present_section(section))
757 return -EINVAL;
758
759 return remove_memory_section(0, section, 0);
760}
761#endif /* CONFIG_MEMORY_HOTREMOVE */
762
763/* return true if the memory block is offlined, otherwise, return false */
764bool is_memblock_offlined(struct memory_block *mem)
765{
766 return mem->state == MEM_OFFLINE;
767}
768
769static struct attribute *memory_root_attrs[] = {
770#ifdef CONFIG_ARCH_MEMORY_PROBE
771 &dev_attr_probe.attr,
772#endif
773
774#ifdef CONFIG_MEMORY_FAILURE
775 &dev_attr_soft_offline_page.attr,
776 &dev_attr_hard_offline_page.attr,
777#endif
778
779 &dev_attr_block_size_bytes.attr,
780 &dev_attr_auto_online_blocks.attr,
781 NULL
782};
783
784static struct attribute_group memory_root_attr_group = {
785 .attrs = memory_root_attrs,
786};
787
788static const struct attribute_group *memory_root_attr_groups[] = {
789 &memory_root_attr_group,
790 NULL,
791};
792
793/*
794 * Initialize the sysfs support for memory devices...
795 */
796int __init memory_dev_init(void)
797{
798 unsigned int i;
799 int ret;
800 int err;
801 unsigned long block_sz;
802
803 ret = subsys_system_register(&memory_subsys, memory_root_attr_groups);
804 if (ret)
805 goto out;
806
807 block_sz = get_memory_block_size();
808 sections_per_block = block_sz / MIN_MEMORY_BLOCK_SIZE;
809
810 /*
811 * Create entries for memory sections that were found
812 * during boot and have been initialized
813 */
814 mutex_lock(&mem_sysfs_mutex);
815 for (i = 0; i < NR_MEM_SECTIONS; i += sections_per_block) {
816 err = add_memory_block(i);
817 if (!ret)
818 ret = err;
819 }
820 mutex_unlock(&mem_sysfs_mutex);
821
822out:
823 if (ret)
824 printk(KERN_ERR "%s() failed: %d\n", __func__, ret);
825 return ret;
826}