Linux Audio

Check our new training course

Loading...
v3.15
 
  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
121static ssize_t show_mem_end_phys_index(struct device *dev,
122			struct device_attribute *attr, char *buf)
123{
124	struct memory_block *mem = to_memory_block(dev);
125	unsigned long phys_index;
126
127	phys_index = mem->end_section_nr / sections_per_block;
128	return sprintf(buf, "%08lx\n", phys_index);
129}
130
131/*
132 * Show whether the section of memory is likely to be hot-removable
 
133 */
134static ssize_t show_mem_removable(struct device *dev,
135			struct device_attribute *attr, char *buf)
136{
137	unsigned long i, pfn;
138	int ret = 1;
139	struct memory_block *mem = to_memory_block(dev);
140
141	for (i = 0; i < sections_per_block; i++) {
142		if (!present_section_nr(mem->start_section_nr + i))
143			continue;
144		pfn = section_nr_to_pfn(mem->start_section_nr + i);
145		ret &= is_mem_section_removable(pfn, PAGES_PER_SECTION);
146	}
147
148	return sprintf(buf, "%d\n", ret);
149}
150
151/*
152 * online, offline, going offline, etc.
153 */
154static ssize_t show_mem_state(struct device *dev,
155			struct device_attribute *attr, char *buf)
156{
157	struct memory_block *mem = to_memory_block(dev);
158	ssize_t len = 0;
159
160	/*
161	 * We can probably put these states in a nice little array
162	 * so that they're not open-coded
163	 */
164	switch (mem->state) {
165		case MEM_ONLINE:
166			len = sprintf(buf, "online\n");
167			break;
168		case MEM_OFFLINE:
169			len = sprintf(buf, "offline\n");
170			break;
171		case MEM_GOING_OFFLINE:
172			len = sprintf(buf, "going-offline\n");
173			break;
174		default:
175			len = sprintf(buf, "ERROR-UNKNOWN-%ld\n",
176					mem->state);
177			WARN_ON(1);
178			break;
179	}
180
181	return len;
182}
183
184int memory_notify(unsigned long val, void *v)
185{
186	return blocking_notifier_call_chain(&memory_chain, val, v);
187}
188
189int memory_isolate_notify(unsigned long val, void *v)
190{
191	return atomic_notifier_call_chain(&memory_isolate_chain, val, v);
192}
 
 
 
193
194/*
195 * The probe routines leave the pages reserved, just as the bootmem code does.
196 * Make sure they're still that way.
197 */
198static bool pages_correctly_reserved(unsigned long start_pfn)
199{
200	int i, j;
201	struct page *page;
202	unsigned long pfn = start_pfn;
203
204	/*
205	 * memmap between sections is not contiguous except with
206	 * SPARSEMEM_VMEMMAP. We lookup the page once per section
207	 * and assume memmap is contiguous within each section
 
 
208	 */
209	for (i = 0; i < sections_per_block; i++, pfn += PAGES_PER_SECTION) {
210		if (WARN_ON_ONCE(!pfn_valid(pfn)))
211			return false;
212		page = pfn_to_page(pfn);
 
213
214		for (j = 0; j < PAGES_PER_SECTION; j++) {
215			if (PageReserved(page + j))
216				continue;
 
 
 
 
217
218			printk(KERN_WARNING "section number %ld page number %d "
219				"not reserved, was it already online?\n",
220				pfn_to_section_nr(pfn), j);
 
 
 
221
222			return false;
223		}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
224	}
225
226	return true;
 
 
 
227}
228
229/*
230 * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is
231 * OK to have direct references to sparsemem variables in here.
232 */
233static int
234memory_block_action(unsigned long phys_index, unsigned long action, int online_type)
235{
236	unsigned long start_pfn;
237	unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
238	struct page *first_page;
239	int ret;
240
241	first_page = pfn_to_page(phys_index << PFN_SECTION_SHIFT);
242	start_pfn = page_to_pfn(first_page);
243
244	switch (action) {
245		case MEM_ONLINE:
246			if (!pages_correctly_reserved(start_pfn))
247				return -EBUSY;
248
249			ret = online_pages(start_pfn, nr_pages, online_type);
250			break;
251		case MEM_OFFLINE:
252			ret = offline_pages(start_pfn, nr_pages);
253			break;
254		default:
255			WARN(1, KERN_WARNING "%s(%ld, %ld) unknown action: "
256			     "%ld\n", __func__, phys_index, action, action);
257			ret = -EINVAL;
258	}
259
260	return ret;
261}
262
263static int memory_block_change_state(struct memory_block *mem,
264		unsigned long to_state, unsigned long from_state_req)
265{
266	int ret = 0;
267
268	if (mem->state != from_state_req)
269		return -EINVAL;
270
271	if (to_state == MEM_OFFLINE)
272		mem->state = MEM_GOING_OFFLINE;
273
274	ret = memory_block_action(mem->start_section_nr, to_state,
275				mem->online_type);
276
277	mem->state = ret ? from_state_req : to_state;
278
279	return ret;
280}
281
282/* The device lock serializes operations on memory_subsys_[online|offline] */
283static int memory_subsys_online(struct device *dev)
284{
285	struct memory_block *mem = to_memory_block(dev);
286	int ret;
287
288	if (mem->state == MEM_ONLINE)
289		return 0;
290
291	/*
292	 * If we are called from store_mem_state(), online_type will be
293	 * set >= 0 Otherwise we were called from the device online
294	 * attribute and need to set the online_type.
295	 */
296	if (mem->online_type < 0)
297		mem->online_type = ONLINE_KEEP;
298
299	ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE);
300
301	/* clear online_type */
302	mem->online_type = -1;
303
304	return ret;
305}
306
307static int memory_subsys_offline(struct device *dev)
308{
309	struct memory_block *mem = to_memory_block(dev);
310
311	if (mem->state == MEM_OFFLINE)
312		return 0;
313
314	return memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE);
315}
316
317static ssize_t
318store_mem_state(struct device *dev,
319		struct device_attribute *attr, const char *buf, size_t count)
320{
 
321	struct memory_block *mem = to_memory_block(dev);
322	int ret, online_type;
 
 
 
323
324	ret = lock_device_hotplug_sysfs();
325	if (ret)
326		return ret;
327
328	if (!strncmp(buf, "online_kernel", min_t(int, count, 13)))
329		online_type = ONLINE_KERNEL;
330	else if (!strncmp(buf, "online_movable", min_t(int, count, 14)))
331		online_type = ONLINE_MOVABLE;
332	else if (!strncmp(buf, "online", min_t(int, count, 6)))
333		online_type = ONLINE_KEEP;
334	else if (!strncmp(buf, "offline", min_t(int, count, 7)))
335		online_type = -1;
336	else {
337		ret = -EINVAL;
338		goto err;
339	}
340
341	switch (online_type) {
342	case ONLINE_KERNEL:
343	case ONLINE_MOVABLE:
344	case ONLINE_KEEP:
345		/*
346		 * mem->online_type is not protected so there can be a
347		 * race here.  However, when racing online, the first
348		 * will succeed and the second will just return as the
349		 * block will already be online.  The online type
350		 * could be either one, but that is expected.
351		 */
352		mem->online_type = online_type;
353		ret = device_online(&mem->dev);
354		break;
355	case -1:
356		ret = device_offline(&mem->dev);
357		break;
358	default:
359		ret = -EINVAL; /* should never happen */
360	}
361
362err:
363	unlock_device_hotplug();
364
365	if (ret)
366		return ret;
 
 
 
367	return count;
368}
369
370/*
371 * phys_device is a bad name for this.  What I really want
372 * is a way to differentiate between memory ranges that
373 * are part of physical devices that constitute
374 * a complete removable unit or fru.
375 * i.e. do these ranges belong to the same physical device,
376 * s.t. if I offline all of these sections I can then
377 * remove the physical device?
378 */
379static ssize_t show_phys_device(struct device *dev,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
380				struct device_attribute *attr, char *buf)
381{
382	struct memory_block *mem = to_memory_block(dev);
383	return sprintf(buf, "%d\n", mem->phys_device);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
384}
 
 
385
386static DEVICE_ATTR(phys_index, 0444, show_mem_start_phys_index, NULL);
387static DEVICE_ATTR(end_phys_index, 0444, show_mem_end_phys_index, NULL);
388static DEVICE_ATTR(state, 0644, show_mem_state, store_mem_state);
389static DEVICE_ATTR(phys_device, 0444, show_phys_device, NULL);
390static DEVICE_ATTR(removable, 0444, show_mem_removable, NULL);
391
392/*
393 * Block size attribute stuff
394 */
395static ssize_t
396print_block_size(struct device *dev, struct device_attribute *attr,
397		 char *buf)
398{
399	return sprintf(buf, "%lx\n", get_memory_block_size());
400}
401
402static DEVICE_ATTR(block_size_bytes, 0444, print_block_size, NULL);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
403
404/*
405 * Some architectures will have custom drivers to do this, and
406 * will not need to do it from userspace.  The fake hot-add code
407 * as well as ppc64 will do all of their discovery in userspace
408 * and will require this interface.
409 */
410#ifdef CONFIG_ARCH_MEMORY_PROBE
411static ssize_t
412memory_probe_store(struct device *dev, struct device_attribute *attr,
413		   const char *buf, size_t count)
414{
415	u64 phys_addr;
416	int nid;
417	int i, ret;
418	unsigned long pages_per_block = PAGES_PER_SECTION * sections_per_block;
419
420	phys_addr = simple_strtoull(buf, NULL, 0);
 
 
421
422	if (phys_addr & ((pages_per_block << PAGE_SHIFT) - 1))
423		return -EINVAL;
424
425	for (i = 0; i < sections_per_block; i++) {
426		nid = memory_add_physaddr_to_nid(phys_addr);
427		ret = add_memory(nid, phys_addr,
428				 PAGES_PER_SECTION << PAGE_SHIFT);
429		if (ret)
430			goto out;
431
432		phys_addr += MIN_MEMORY_BLOCK_SIZE;
433	}
 
 
 
 
 
434
435	ret = count;
436out:
 
437	return ret;
438}
439
440static DEVICE_ATTR(probe, S_IWUSR, NULL, memory_probe_store);
441#endif
442
443#ifdef CONFIG_MEMORY_FAILURE
444/*
445 * Support for offlining pages of memory
446 */
447
448/* Soft offline a page */
449static ssize_t
450store_soft_offline_page(struct device *dev,
451			struct device_attribute *attr,
452			const char *buf, size_t count)
453{
454	int ret;
455	u64 pfn;
456	if (!capable(CAP_SYS_ADMIN))
457		return -EPERM;
458	if (kstrtoull(buf, 0, &pfn) < 0)
459		return -EINVAL;
460	pfn >>= PAGE_SHIFT;
461	if (!pfn_valid(pfn))
462		return -ENXIO;
463	ret = soft_offline_page(pfn_to_page(pfn), 0);
464	return ret == 0 ? count : ret;
465}
466
467/* Forcibly offline a page, including killing processes. */
468static ssize_t
469store_hard_offline_page(struct device *dev,
470			struct device_attribute *attr,
471			const char *buf, size_t count)
472{
473	int ret;
474	u64 pfn;
475	if (!capable(CAP_SYS_ADMIN))
476		return -EPERM;
477	if (kstrtoull(buf, 0, &pfn) < 0)
478		return -EINVAL;
479	pfn >>= PAGE_SHIFT;
480	ret = memory_failure(pfn, 0, 0);
481	return ret ? ret : count;
482}
483
484static DEVICE_ATTR(soft_offline_page, S_IWUSR, NULL, store_soft_offline_page);
485static DEVICE_ATTR(hard_offline_page, S_IWUSR, NULL, store_hard_offline_page);
486#endif
487
488/*
489 * Note that phys_device is optional.  It is here to allow for
490 * differentiation between which *physical* devices each
491 * section belongs to...
492 */
493int __weak arch_get_memory_phys_device(unsigned long start_pfn)
494{
495	return 0;
496}
497
498/*
499 * A reference for the returned object is held and the reference for the
500 * hinted object is released.
 
501 */
502struct memory_block *find_memory_block_hinted(struct mem_section *section,
503					      struct memory_block *hint)
504{
505	int block_id = base_memory_block_id(__section_nr(section));
506	struct device *hintdev = hint ? &hint->dev : NULL;
507	struct device *dev;
508
509	dev = subsys_find_device_by_id(&memory_subsys, block_id, hintdev);
510	if (hint)
511		put_device(&hint->dev);
512	if (!dev)
513		return NULL;
514	return to_memory_block(dev);
515}
516
517/*
518 * For now, we have a linear search to go find the appropriate
519 * memory_block corresponding to a particular phys_index. If
520 * this gets to be a real problem, we can always use a radix
521 * tree or something here.
522 *
523 * This could be made generic for all device subsystems.
524 */
525struct memory_block *find_memory_block(struct mem_section *section)
526{
527	return find_memory_block_hinted(section, NULL);
 
 
528}
529
530static struct attribute *memory_memblk_attrs[] = {
531	&dev_attr_phys_index.attr,
532	&dev_attr_end_phys_index.attr,
533	&dev_attr_state.attr,
534	&dev_attr_phys_device.attr,
535	&dev_attr_removable.attr,
 
 
 
536	NULL
537};
538
539static struct attribute_group memory_memblk_attr_group = {
540	.attrs = memory_memblk_attrs,
541};
542
543static const struct attribute_group *memory_memblk_attr_groups[] = {
544	&memory_memblk_attr_group,
545	NULL,
546};
547
548/*
549 * register_memory - Setup a sysfs device for a memory block
550 */
551static
552int register_memory(struct memory_block *memory)
553{
 
 
554	memory->dev.bus = &memory_subsys;
555	memory->dev.id = memory->start_section_nr / sections_per_block;
556	memory->dev.release = memory_block_release;
557	memory->dev.groups = memory_memblk_attr_groups;
558	memory->dev.offline = memory->state == MEM_OFFLINE;
559
560	return device_register(&memory->dev);
 
 
 
 
 
 
 
 
 
 
 
561}
562
563static int init_memory_block(struct memory_block **memory,
564			     struct mem_section *section, unsigned long state)
565{
566	struct memory_block *mem;
567	unsigned long start_pfn;
568	int scn_nr;
569	int ret = 0;
570
 
 
 
 
 
571	mem = kzalloc(sizeof(*mem), GFP_KERNEL);
572	if (!mem)
573		return -ENOMEM;
574
575	scn_nr = __section_nr(section);
576	mem->start_section_nr =
577			base_memory_block_id(scn_nr) * sections_per_block;
578	mem->end_section_nr = mem->start_section_nr + sections_per_block - 1;
579	mem->state = state;
580	mem->section_count++;
581	start_pfn = section_nr_to_pfn(mem->start_section_nr);
582	mem->phys_device = arch_get_memory_phys_device(start_pfn);
583
584	ret = register_memory(mem);
585
586	*memory = mem;
587	return ret;
588}
589
590static int add_memory_block(int base_section_nr)
591{
592	struct memory_block *mem;
593	int i, ret, section_count = 0, section_nr;
594
595	for (i = base_section_nr;
596	     (i < base_section_nr + sections_per_block) && i < NR_MEM_SECTIONS;
597	     i++) {
598		if (!present_section_nr(i))
599			continue;
600		if (section_count == 0)
601			section_nr = i;
602		section_count++;
603	}
604
605	if (section_count == 0)
606		return 0;
607	ret = init_memory_block(&mem, __nr_to_section(section_nr), MEM_ONLINE);
608	if (ret)
609		return ret;
610	mem->section_count = section_count;
611	return 0;
612}
613
 
 
 
 
 
 
 
 
 
 
 
614
615/*
616 * need an interface for the VM to add new memory regions,
617 * but without onlining it.
 
 
 
618 */
619int register_new_memory(int nid, struct mem_section *section)
 
620{
621	int ret = 0;
 
622	struct memory_block *mem;
 
 
623
624	mutex_lock(&mem_sysfs_mutex);
 
 
625
626	mem = find_memory_block(section);
627	if (mem) {
628		mem->section_count++;
629		put_device(&mem->dev);
630	} else {
631		ret = init_memory_block(&mem, section, MEM_OFFLINE);
632		if (ret)
633			goto out;
 
 
 
 
 
 
 
 
 
 
634	}
635
636	if (mem->section_count == sections_per_block)
637		ret = register_mem_sect_under_node(mem, nid);
638out:
639	mutex_unlock(&mem_sysfs_mutex);
640	return ret;
641}
642
643#ifdef CONFIG_MEMORY_HOTREMOVE
644static void
645unregister_memory(struct memory_block *memory)
646{
647	BUG_ON(memory->dev.bus != &memory_subsys);
648
649	/* drop the ref. we got in remove_memory_block() */
650	put_device(&memory->dev);
651	device_unregister(&memory->dev);
652}
653
654static int remove_memory_block(unsigned long node_id,
655			       struct mem_section *section, int phys_device)
656{
 
 
657	struct memory_block *mem;
 
658
659	mutex_lock(&mem_sysfs_mutex);
660	mem = find_memory_block(section);
661	unregister_mem_sect_under_nodes(mem, __section_nr(section));
662
663	mem->section_count--;
664	if (mem->section_count == 0)
 
 
 
665		unregister_memory(mem);
666	else
667		put_device(&mem->dev);
668
669	mutex_unlock(&mem_sysfs_mutex);
670	return 0;
671}
672
673int unregister_memory_section(struct mem_section *section)
674{
675	if (!present_section(section))
676		return -EINVAL;
677
678	return remove_memory_block(0, section, 0);
679}
680#endif /* CONFIG_MEMORY_HOTREMOVE */
681
682/* return true if the memory block is offlined, otherwise, return false */
683bool is_memblock_offlined(struct memory_block *mem)
684{
685	return mem->state == MEM_OFFLINE;
686}
687
688static struct attribute *memory_root_attrs[] = {
689#ifdef CONFIG_ARCH_MEMORY_PROBE
690	&dev_attr_probe.attr,
691#endif
692
693#ifdef CONFIG_MEMORY_FAILURE
694	&dev_attr_soft_offline_page.attr,
695	&dev_attr_hard_offline_page.attr,
696#endif
697
698	&dev_attr_block_size_bytes.attr,
 
699	NULL
700};
701
702static struct attribute_group memory_root_attr_group = {
703	.attrs = memory_root_attrs,
704};
705
706static const struct attribute_group *memory_root_attr_groups[] = {
707	&memory_root_attr_group,
708	NULL,
709};
710
711/*
712 * Initialize the sysfs support for memory devices...
 
 
713 */
714int __init memory_dev_init(void)
715{
716	unsigned int i;
717	int ret;
718	int err;
719	unsigned long block_sz;
 
 
 
 
 
720
721	ret = subsys_system_register(&memory_subsys, memory_root_attr_groups);
722	if (ret)
723		goto out;
724
725	block_sz = get_memory_block_size();
726	sections_per_block = block_sz / MIN_MEMORY_BLOCK_SIZE;
727
728	/*
729	 * Create entries for memory sections that were found
730	 * during boot and have been initialized
731	 */
732	mutex_lock(&mem_sysfs_mutex);
733	for (i = 0; i < NR_MEM_SECTIONS; i += sections_per_block) {
734		err = add_memory_block(i);
735		if (!ret)
736			ret = err;
 
737	}
738	mutex_unlock(&mem_sysfs_mutex);
739
740out:
741	if (ret)
742		printk(KERN_ERR "%s() failed: %d\n", __func__, ret);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
743	return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
744}
v5.14.15
  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 mhp_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
123	return sysfs_emit(buf, "%08lx\n", phys_index);
 
 
 
 
 
 
 
124}
125
126/*
127 * Legacy interface that we cannot remove. Always indicate "removable"
128 * with CONFIG_MEMORY_HOTREMOVE - bad heuristic.
129 */
130static ssize_t removable_show(struct device *dev, struct device_attribute *attr,
131			      char *buf)
132{
133	return sysfs_emit(buf, "%d\n", (int)IS_ENABLED(CONFIG_MEMORY_HOTREMOVE));
 
 
 
 
 
 
 
 
 
 
 
134}
135
136/*
137 * online, offline, going offline, etc.
138 */
139static ssize_t state_show(struct device *dev, struct device_attribute *attr,
140			  char *buf)
141{
142	struct memory_block *mem = to_memory_block(dev);
143	const char *output;
144
145	/*
146	 * We can probably put these states in a nice little array
147	 * so that they're not open-coded
148	 */
149	switch (mem->state) {
150	case MEM_ONLINE:
151		output = "online";
152		break;
153	case MEM_OFFLINE:
154		output = "offline";
155		break;
156	case MEM_GOING_OFFLINE:
157		output = "going-offline";
158		break;
159	default:
160		WARN_ON(1);
161		return sysfs_emit(buf, "ERROR-UNKNOWN-%ld\n", mem->state);
 
 
162	}
163
164	return sysfs_emit(buf, "%s\n", output);
165}
166
167int memory_notify(unsigned long val, void *v)
168{
169	return blocking_notifier_call_chain(&memory_chain, val, v);
170}
171
172static int memory_block_online(struct memory_block *mem)
173{
174	unsigned long start_pfn = section_nr_to_pfn(mem->start_section_nr);
175	unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
176	unsigned long nr_vmemmap_pages = mem->nr_vmemmap_pages;
177	struct zone *zone;
178	int ret;
179
180	zone = zone_for_pfn_range(mem->online_type, mem->nid, start_pfn, nr_pages);
 
 
 
 
 
 
 
 
181
182	/*
183	 * Although vmemmap pages have a different lifecycle than the pages
184	 * they describe (they remain until the memory is unplugged), doing
185	 * their initialization and accounting at memory onlining/offlining
186	 * stage helps to keep accounting easier to follow - e.g vmemmaps
187	 * belong to the same zone as the memory they backed.
188	 */
189	if (nr_vmemmap_pages) {
190		ret = mhp_init_memmap_on_memory(start_pfn, nr_vmemmap_pages, zone);
191		if (ret)
192			return ret;
193	}
194
195	ret = online_pages(start_pfn + nr_vmemmap_pages,
196			   nr_pages - nr_vmemmap_pages, zone);
197	if (ret) {
198		if (nr_vmemmap_pages)
199			mhp_deinit_memmap_on_memory(start_pfn, nr_vmemmap_pages);
200		return ret;
201	}
202
203	/*
204	 * Account once onlining succeeded. If the zone was unpopulated, it is
205	 * now already properly populated.
206	 */
207	if (nr_vmemmap_pages)
208		adjust_present_page_count(zone, nr_vmemmap_pages);
209
210	return ret;
211}
212
213static int memory_block_offline(struct memory_block *mem)
214{
215	unsigned long start_pfn = section_nr_to_pfn(mem->start_section_nr);
216	unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
217	unsigned long nr_vmemmap_pages = mem->nr_vmemmap_pages;
218	struct zone *zone;
219	int ret;
220
221	/*
222	 * Unaccount before offlining, such that unpopulated zone and kthreads
223	 * can properly be torn down in offline_pages().
224	 */
225	if (nr_vmemmap_pages) {
226		zone = page_zone(pfn_to_page(start_pfn));
227		adjust_present_page_count(zone, -nr_vmemmap_pages);
228	}
229
230	ret = offline_pages(start_pfn + nr_vmemmap_pages,
231			    nr_pages - nr_vmemmap_pages);
232	if (ret) {
233		/* offline_pages() failed. Account back. */
234		if (nr_vmemmap_pages)
235			adjust_present_page_count(zone, nr_vmemmap_pages);
236		return ret;
237	}
238
239	if (nr_vmemmap_pages)
240		mhp_deinit_memmap_on_memory(start_pfn, nr_vmemmap_pages);
241
242	return ret;
243}
244
245/*
246 * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is
247 * OK to have direct references to sparsemem variables in here.
248 */
249static int
250memory_block_action(struct memory_block *mem, unsigned long action)
251{
 
 
 
252	int ret;
253
 
 
 
254	switch (action) {
255	case MEM_ONLINE:
256		ret = memory_block_online(mem);
257		break;
258	case MEM_OFFLINE:
259		ret = memory_block_offline(mem);
260		break;
261	default:
262		WARN(1, KERN_WARNING "%s(%ld, %ld) unknown action: "
263		     "%ld\n", __func__, mem->start_section_nr, action, action);
264		ret = -EINVAL;
 
 
 
265	}
266
267	return ret;
268}
269
270static int memory_block_change_state(struct memory_block *mem,
271		unsigned long to_state, unsigned long from_state_req)
272{
273	int ret = 0;
274
275	if (mem->state != from_state_req)
276		return -EINVAL;
277
278	if (to_state == MEM_OFFLINE)
279		mem->state = MEM_GOING_OFFLINE;
280
281	ret = memory_block_action(mem, to_state);
 
 
282	mem->state = ret ? from_state_req : to_state;
283
284	return ret;
285}
286
287/* The device lock serializes operations on memory_subsys_[online|offline] */
288static int memory_subsys_online(struct device *dev)
289{
290	struct memory_block *mem = to_memory_block(dev);
291	int ret;
292
293	if (mem->state == MEM_ONLINE)
294		return 0;
295
296	/*
297	 * When called via device_online() without configuring the online_type,
298	 * we want to default to MMOP_ONLINE.
 
299	 */
300	if (mem->online_type == MMOP_OFFLINE)
301		mem->online_type = MMOP_ONLINE;
302
303	ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE);
304	mem->online_type = MMOP_OFFLINE;
 
 
305
306	return ret;
307}
308
309static int memory_subsys_offline(struct device *dev)
310{
311	struct memory_block *mem = to_memory_block(dev);
312
313	if (mem->state == MEM_OFFLINE)
314		return 0;
315
316	return memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE);
317}
318
319static ssize_t state_store(struct device *dev, struct device_attribute *attr,
320			   const char *buf, size_t count)
 
321{
322	const int online_type = mhp_online_type_from_str(buf);
323	struct memory_block *mem = to_memory_block(dev);
324	int ret;
325
326	if (online_type < 0)
327		return -EINVAL;
328
329	ret = lock_device_hotplug_sysfs();
330	if (ret)
331		return ret;
332
 
 
 
 
 
 
 
 
 
 
 
 
 
333	switch (online_type) {
334	case MMOP_ONLINE_KERNEL:
335	case MMOP_ONLINE_MOVABLE:
336	case MMOP_ONLINE:
337		/* mem->online_type is protected by device_hotplug_lock */
 
 
 
 
 
 
338		mem->online_type = online_type;
339		ret = device_online(&mem->dev);
340		break;
341	case MMOP_OFFLINE:
342		ret = device_offline(&mem->dev);
343		break;
344	default:
345		ret = -EINVAL; /* should never happen */
346	}
347
 
348	unlock_device_hotplug();
349
350	if (ret < 0)
351		return ret;
352	if (ret)
353		return -EINVAL;
354
355	return count;
356}
357
358/*
359 * Legacy interface that we cannot remove: s390x exposes the storage increment
360 * covered by a memory block, allowing for identifying which memory blocks
361 * comprise a storage increment. Since a memory block spans complete
362 * storage increments nowadays, this interface is basically unused. Other
363 * archs never exposed != 0.
 
 
364 */
365static ssize_t phys_device_show(struct device *dev,
366				struct device_attribute *attr, char *buf)
367{
368	struct memory_block *mem = to_memory_block(dev);
369	unsigned long start_pfn = section_nr_to_pfn(mem->start_section_nr);
370
371	return sysfs_emit(buf, "%d\n",
372			  arch_get_memory_phys_device(start_pfn));
373}
374
375#ifdef CONFIG_MEMORY_HOTREMOVE
376static int print_allowed_zone(char *buf, int len, int nid,
377			      unsigned long start_pfn, unsigned long nr_pages,
378			      int online_type, struct zone *default_zone)
379{
380	struct zone *zone;
381
382	zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages);
383	if (zone == default_zone)
384		return 0;
385
386	return sysfs_emit_at(buf, len, " %s", zone->name);
387}
388
389static ssize_t valid_zones_show(struct device *dev,
390				struct device_attribute *attr, char *buf)
391{
392	struct memory_block *mem = to_memory_block(dev);
393	unsigned long start_pfn = section_nr_to_pfn(mem->start_section_nr);
394	unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
395	struct zone *default_zone;
396	int len = 0;
397	int nid;
398
399	/*
400	 * Check the existing zone. Make sure that we do that only on the
401	 * online nodes otherwise the page_zone is not reliable
402	 */
403	if (mem->state == MEM_ONLINE) {
404		/*
405		 * The block contains more than one zone can not be offlined.
406		 * This can happen e.g. for ZONE_DMA and ZONE_DMA32
407		 */
408		default_zone = test_pages_in_a_zone(start_pfn,
409						    start_pfn + nr_pages);
410		if (!default_zone)
411			return sysfs_emit(buf, "%s\n", "none");
412		len += sysfs_emit_at(buf, len, "%s", default_zone->name);
413		goto out;
414	}
415
416	nid = mem->nid;
417	default_zone = zone_for_pfn_range(MMOP_ONLINE, nid, start_pfn,
418					  nr_pages);
419
420	len += sysfs_emit_at(buf, len, "%s", default_zone->name);
421	len += print_allowed_zone(buf, len, nid, start_pfn, nr_pages,
422				  MMOP_ONLINE_KERNEL, default_zone);
423	len += print_allowed_zone(buf, len, nid, start_pfn, nr_pages,
424				  MMOP_ONLINE_MOVABLE, default_zone);
425out:
426	len += sysfs_emit_at(buf, len, "\n");
427	return len;
428}
429static DEVICE_ATTR_RO(valid_zones);
430#endif
431
432static DEVICE_ATTR_RO(phys_index);
433static DEVICE_ATTR_RW(state);
434static DEVICE_ATTR_RO(phys_device);
435static DEVICE_ATTR_RO(removable);
 
436
437/*
438 * Show the memory block size (shared by all memory blocks).
439 */
440static ssize_t block_size_bytes_show(struct device *dev,
441				     struct device_attribute *attr, char *buf)
 
442{
443	return sysfs_emit(buf, "%lx\n", memory_block_size_bytes());
444}
445
446static DEVICE_ATTR_RO(block_size_bytes);
447
448/*
449 * Memory auto online policy.
450 */
451
452static ssize_t auto_online_blocks_show(struct device *dev,
453				       struct device_attribute *attr, char *buf)
454{
455	return sysfs_emit(buf, "%s\n",
456			  online_type_to_str[mhp_default_online_type]);
457}
458
459static ssize_t auto_online_blocks_store(struct device *dev,
460					struct device_attribute *attr,
461					const char *buf, size_t count)
462{
463	const int online_type = mhp_online_type_from_str(buf);
464
465	if (online_type < 0)
466		return -EINVAL;
467
468	mhp_default_online_type = online_type;
469	return count;
470}
471
472static DEVICE_ATTR_RW(auto_online_blocks);
473
474/*
475 * Some architectures will have custom drivers to do this, and
476 * will not need to do it from userspace.  The fake hot-add code
477 * as well as ppc64 will do all of their discovery in userspace
478 * and will require this interface.
479 */
480#ifdef CONFIG_ARCH_MEMORY_PROBE
481static ssize_t probe_store(struct device *dev, struct device_attribute *attr,
482			   const char *buf, size_t count)
 
483{
484	u64 phys_addr;
485	int nid, ret;
 
486	unsigned long pages_per_block = PAGES_PER_SECTION * sections_per_block;
487
488	ret = kstrtoull(buf, 0, &phys_addr);
489	if (ret)
490		return ret;
491
492	if (phys_addr & ((pages_per_block << PAGE_SHIFT) - 1))
493		return -EINVAL;
494
495	ret = lock_device_hotplug_sysfs();
496	if (ret)
497		return ret;
 
 
 
498
499	nid = memory_add_physaddr_to_nid(phys_addr);
500	ret = __add_memory(nid, phys_addr,
501			   MIN_MEMORY_BLOCK_SIZE * sections_per_block,
502			   MHP_NONE);
503
504	if (ret)
505		goto out;
506
507	ret = count;
508out:
509	unlock_device_hotplug();
510	return ret;
511}
512
513static DEVICE_ATTR_WO(probe);
514#endif
515
516#ifdef CONFIG_MEMORY_FAILURE
517/*
518 * Support for offlining pages of memory
519 */
520
521/* Soft offline a page */
522static ssize_t soft_offline_page_store(struct device *dev,
523				       struct device_attribute *attr,
524				       const char *buf, size_t count)
 
525{
526	int ret;
527	u64 pfn;
528	if (!capable(CAP_SYS_ADMIN))
529		return -EPERM;
530	if (kstrtoull(buf, 0, &pfn) < 0)
531		return -EINVAL;
532	pfn >>= PAGE_SHIFT;
533	ret = soft_offline_page(pfn, 0);
 
 
534	return ret == 0 ? count : ret;
535}
536
537/* Forcibly offline a page, including killing processes. */
538static ssize_t hard_offline_page_store(struct device *dev,
539				       struct device_attribute *attr,
540				       const char *buf, size_t count)
 
541{
542	int ret;
543	u64 pfn;
544	if (!capable(CAP_SYS_ADMIN))
545		return -EPERM;
546	if (kstrtoull(buf, 0, &pfn) < 0)
547		return -EINVAL;
548	pfn >>= PAGE_SHIFT;
549	ret = memory_failure(pfn, 0);
550	return ret ? ret : count;
551}
552
553static DEVICE_ATTR_WO(soft_offline_page);
554static DEVICE_ATTR_WO(hard_offline_page);
555#endif
556
557/* See phys_device_show(). */
 
 
 
 
558int __weak arch_get_memory_phys_device(unsigned long start_pfn)
559{
560	return 0;
561}
562
563/*
564 * A reference for the returned memory block device is acquired.
565 *
566 * Called under device_hotplug_lock.
567 */
568static struct memory_block *find_memory_block_by_id(unsigned long block_id)
 
569{
570	struct memory_block *mem;
 
 
571
572	mem = xa_load(&memory_blocks, block_id);
573	if (mem)
574		get_device(&mem->dev);
575	return mem;
 
 
576}
577
578/*
579 * Called under device_hotplug_lock.
 
 
 
 
 
580 */
581struct memory_block *find_memory_block(struct mem_section *section)
582{
583	unsigned long block_id = memory_block_id(__section_nr(section));
584
585	return find_memory_block_by_id(block_id);
586}
587
588static struct attribute *memory_memblk_attrs[] = {
589	&dev_attr_phys_index.attr,
 
590	&dev_attr_state.attr,
591	&dev_attr_phys_device.attr,
592	&dev_attr_removable.attr,
593#ifdef CONFIG_MEMORY_HOTREMOVE
594	&dev_attr_valid_zones.attr,
595#endif
596	NULL
597};
598
599static const struct attribute_group memory_memblk_attr_group = {
600	.attrs = memory_memblk_attrs,
601};
602
603static const struct attribute_group *memory_memblk_attr_groups[] = {
604	&memory_memblk_attr_group,
605	NULL,
606};
607
608/*
609 * register_memory - Setup a sysfs device for a memory block
610 */
611static
612int register_memory(struct memory_block *memory)
613{
614	int ret;
615
616	memory->dev.bus = &memory_subsys;
617	memory->dev.id = memory->start_section_nr / sections_per_block;
618	memory->dev.release = memory_block_release;
619	memory->dev.groups = memory_memblk_attr_groups;
620	memory->dev.offline = memory->state == MEM_OFFLINE;
621
622	ret = device_register(&memory->dev);
623	if (ret) {
624		put_device(&memory->dev);
625		return ret;
626	}
627	ret = xa_err(xa_store(&memory_blocks, memory->dev.id, memory,
628			      GFP_KERNEL));
629	if (ret) {
630		put_device(&memory->dev);
631		device_unregister(&memory->dev);
632	}
633	return ret;
634}
635
636static int init_memory_block(unsigned long block_id, unsigned long state,
637			     unsigned long nr_vmemmap_pages)
638{
639	struct memory_block *mem;
 
 
640	int ret = 0;
641
642	mem = find_memory_block_by_id(block_id);
643	if (mem) {
644		put_device(&mem->dev);
645		return -EEXIST;
646	}
647	mem = kzalloc(sizeof(*mem), GFP_KERNEL);
648	if (!mem)
649		return -ENOMEM;
650
651	mem->start_section_nr = block_id * sections_per_block;
 
 
 
652	mem->state = state;
653	mem->nid = NUMA_NO_NODE;
654	mem->nr_vmemmap_pages = nr_vmemmap_pages;
 
655
656	ret = register_memory(mem);
657
 
658	return ret;
659}
660
661static int add_memory_block(unsigned long base_section_nr)
662{
663	int section_count = 0;
664	unsigned long nr;
665
666	for (nr = base_section_nr; nr < base_section_nr + sections_per_block;
667	     nr++)
668		if (present_section_nr(nr))
669			section_count++;
 
 
 
 
 
670
671	if (section_count == 0)
672		return 0;
673	return init_memory_block(memory_block_id(base_section_nr),
674				 MEM_ONLINE, 0);
 
 
 
675}
676
677static void unregister_memory(struct memory_block *memory)
678{
679	if (WARN_ON_ONCE(memory->dev.bus != &memory_subsys))
680		return;
681
682	WARN_ON(xa_erase(&memory_blocks, memory->dev.id) == NULL);
683
684	/* drop the ref. we got via find_memory_block() */
685	put_device(&memory->dev);
686	device_unregister(&memory->dev);
687}
688
689/*
690 * Create memory block devices for the given memory area. Start and size
691 * have to be aligned to memory block granularity. Memory block devices
692 * will be initialized as offline.
693 *
694 * Called under device_hotplug_lock.
695 */
696int create_memory_block_devices(unsigned long start, unsigned long size,
697				unsigned long vmemmap_pages)
698{
699	const unsigned long start_block_id = pfn_to_block_id(PFN_DOWN(start));
700	unsigned long end_block_id = pfn_to_block_id(PFN_DOWN(start + size));
701	struct memory_block *mem;
702	unsigned long block_id;
703	int ret = 0;
704
705	if (WARN_ON_ONCE(!IS_ALIGNED(start, memory_block_size_bytes()) ||
706			 !IS_ALIGNED(size, memory_block_size_bytes())))
707		return -EINVAL;
708
709	for (block_id = start_block_id; block_id != end_block_id; block_id++) {
710		ret = init_memory_block(block_id, MEM_OFFLINE, vmemmap_pages);
 
 
 
 
711		if (ret)
712			break;
713	}
714	if (ret) {
715		end_block_id = block_id;
716		for (block_id = start_block_id; block_id != end_block_id;
717		     block_id++) {
718			mem = find_memory_block_by_id(block_id);
719			if (WARN_ON_ONCE(!mem))
720				continue;
721			unregister_memory(mem);
722		}
723	}
 
 
 
 
 
724	return ret;
725}
726
727/*
728 * Remove memory block devices for the given memory area. Start and size
729 * have to be aligned to memory block granularity. Memory block devices
730 * have to be offline.
731 *
732 * Called under device_hotplug_lock.
733 */
734void remove_memory_block_devices(unsigned long start, unsigned long size)
 
 
 
 
 
735{
736	const unsigned long start_block_id = pfn_to_block_id(PFN_DOWN(start));
737	const unsigned long end_block_id = pfn_to_block_id(PFN_DOWN(start + size));
738	struct memory_block *mem;
739	unsigned long block_id;
740
741	if (WARN_ON_ONCE(!IS_ALIGNED(start, memory_block_size_bytes()) ||
742			 !IS_ALIGNED(size, memory_block_size_bytes())))
743		return;
744
745	for (block_id = start_block_id; block_id != end_block_id; block_id++) {
746		mem = find_memory_block_by_id(block_id);
747		if (WARN_ON_ONCE(!mem))
748			continue;
749		unregister_memory_block_under_nodes(mem);
750		unregister_memory(mem);
751	}
 
 
 
 
 
 
 
 
 
 
 
 
752}
 
753
754/* return true if the memory block is offlined, otherwise, return false */
755bool is_memblock_offlined(struct memory_block *mem)
756{
757	return mem->state == MEM_OFFLINE;
758}
759
760static struct attribute *memory_root_attrs[] = {
761#ifdef CONFIG_ARCH_MEMORY_PROBE
762	&dev_attr_probe.attr,
763#endif
764
765#ifdef CONFIG_MEMORY_FAILURE
766	&dev_attr_soft_offline_page.attr,
767	&dev_attr_hard_offline_page.attr,
768#endif
769
770	&dev_attr_block_size_bytes.attr,
771	&dev_attr_auto_online_blocks.attr,
772	NULL
773};
774
775static const struct attribute_group memory_root_attr_group = {
776	.attrs = memory_root_attrs,
777};
778
779static const struct attribute_group *memory_root_attr_groups[] = {
780	&memory_root_attr_group,
781	NULL,
782};
783
784/*
785 * Initialize the sysfs support for memory devices. At the time this function
786 * is called, we cannot have concurrent creation/deletion of memory block
787 * devices, the device_hotplug_lock is not needed.
788 */
789void __init memory_dev_init(void)
790{
 
791	int ret;
792	unsigned long block_sz, nr;
793
794	/* Validate the configured memory block size */
795	block_sz = memory_block_size_bytes();
796	if (!is_power_of_2(block_sz) || block_sz < MIN_MEMORY_BLOCK_SIZE)
797		panic("Memory block size not suitable: 0x%lx\n", block_sz);
798	sections_per_block = block_sz / MIN_MEMORY_BLOCK_SIZE;
799
800	ret = subsys_system_register(&memory_subsys, memory_root_attr_groups);
801	if (ret)
802		panic("%s() failed to register subsystem: %d\n", __func__, ret);
 
 
 
803
804	/*
805	 * Create entries for memory sections that were found
806	 * during boot and have been initialized
807	 */
808	for (nr = 0; nr <= __highest_present_section_nr;
809	     nr += sections_per_block) {
810		ret = add_memory_block(nr);
811		if (ret)
812			panic("%s() failed to add memory block: %d\n", __func__,
813			      ret);
814	}
815}
816
817/**
818 * walk_memory_blocks - walk through all present memory blocks overlapped
819 *			by the range [start, start + size)
820 *
821 * @start: start address of the memory range
822 * @size: size of the memory range
823 * @arg: argument passed to func
824 * @func: callback for each memory section walked
825 *
826 * This function walks through all present memory blocks overlapped by the
827 * range [start, start + size), calling func on each memory block.
828 *
829 * In case func() returns an error, walking is aborted and the error is
830 * returned.
831 *
832 * Called under device_hotplug_lock.
833 */
834int walk_memory_blocks(unsigned long start, unsigned long size,
835		       void *arg, walk_memory_blocks_func_t func)
836{
837	const unsigned long start_block_id = phys_to_block_id(start);
838	const unsigned long end_block_id = phys_to_block_id(start + size - 1);
839	struct memory_block *mem;
840	unsigned long block_id;
841	int ret = 0;
842
843	if (!size)
844		return 0;
845
846	for (block_id = start_block_id; block_id <= end_block_id; block_id++) {
847		mem = find_memory_block_by_id(block_id);
848		if (!mem)
849			continue;
850
851		ret = func(mem, arg);
852		put_device(&mem->dev);
853		if (ret)
854			break;
855	}
856	return ret;
857}
858
859struct for_each_memory_block_cb_data {
860	walk_memory_blocks_func_t func;
861	void *arg;
862};
863
864static int for_each_memory_block_cb(struct device *dev, void *data)
865{
866	struct memory_block *mem = to_memory_block(dev);
867	struct for_each_memory_block_cb_data *cb_data = data;
868
869	return cb_data->func(mem, cb_data->arg);
870}
871
872/**
873 * for_each_memory_block - walk through all present memory blocks
874 *
875 * @arg: argument passed to func
876 * @func: callback for each memory block walked
877 *
878 * This function walks through all present memory blocks, calling func on
879 * each memory block.
880 *
881 * In case func() returns an error, walking is aborted and the error is
882 * returned.
883 */
884int for_each_memory_block(void *arg, walk_memory_blocks_func_t func)
885{
886	struct for_each_memory_block_cb_data cb_data = {
887		.func = func,
888		.arg = arg,
889	};
890
891	return bus_for_each_dev(&memory_subsys, NULL, &cb_data,
892				for_each_memory_block_cb);
893}