2 * Memory subsystem support
4 * Written by Matt Tolentino <matthew.e.tolentino@intel.com>
5 * Dave Hansen <haveblue@us.ibm.com>
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.
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/kobject.h>
20 #include <linux/memory_hotplug.h>
22 #include <linux/mutex.h>
23 #include <linux/stat.h>
24 #include <linux/slab.h>
26 #include <linux/atomic.h>
27 #include <asm/uaccess.h>
29 static DEFINE_MUTEX(mem_sysfs_mutex
);
31 #define MEMORY_CLASS_NAME "memory"
33 static int sections_per_block
;
35 static inline int base_memory_block_id(int section_nr
)
37 return section_nr
/ sections_per_block
;
40 static struct bus_type memory_subsys
= {
41 .name
= MEMORY_CLASS_NAME
,
42 .dev_name
= MEMORY_CLASS_NAME
,
45 static BLOCKING_NOTIFIER_HEAD(memory_chain
);
47 int register_memory_notifier(struct notifier_block
*nb
)
49 return blocking_notifier_chain_register(&memory_chain
, nb
);
51 EXPORT_SYMBOL(register_memory_notifier
);
53 void unregister_memory_notifier(struct notifier_block
*nb
)
55 blocking_notifier_chain_unregister(&memory_chain
, nb
);
57 EXPORT_SYMBOL(unregister_memory_notifier
);
59 static ATOMIC_NOTIFIER_HEAD(memory_isolate_chain
);
61 int register_memory_isolate_notifier(struct notifier_block
*nb
)
63 return atomic_notifier_chain_register(&memory_isolate_chain
, nb
);
65 EXPORT_SYMBOL(register_memory_isolate_notifier
);
67 void unregister_memory_isolate_notifier(struct notifier_block
*nb
)
69 atomic_notifier_chain_unregister(&memory_isolate_chain
, nb
);
71 EXPORT_SYMBOL(unregister_memory_isolate_notifier
);
73 static void memory_block_release(struct device
*dev
)
75 struct memory_block
*mem
= container_of(dev
, struct memory_block
, dev
);
81 * register_memory - Setup a sysfs device for a memory block
84 int register_memory(struct memory_block
*memory
)
88 memory
->dev
.bus
= &memory_subsys
;
89 memory
->dev
.id
= memory
->start_section_nr
/ sections_per_block
;
90 memory
->dev
.release
= memory_block_release
;
92 error
= device_register(&memory
->dev
);
97 unregister_memory(struct memory_block
*memory
)
99 BUG_ON(memory
->dev
.bus
!= &memory_subsys
);
101 /* drop the ref. we got in remove_memory_block() */
102 kobject_put(&memory
->dev
.kobj
);
103 device_unregister(&memory
->dev
);
106 unsigned long __weak
memory_block_size_bytes(void)
108 return MIN_MEMORY_BLOCK_SIZE
;
111 static unsigned long get_memory_block_size(void)
113 unsigned long block_sz
;
115 block_sz
= memory_block_size_bytes();
117 /* Validate blk_sz is a power of 2 and not less than section size */
118 if ((block_sz
& (block_sz
- 1)) || (block_sz
< MIN_MEMORY_BLOCK_SIZE
)) {
120 block_sz
= MIN_MEMORY_BLOCK_SIZE
;
127 * use this as the physical section index that this memsection
131 static ssize_t
show_mem_start_phys_index(struct device
*dev
,
132 struct device_attribute
*attr
, char *buf
)
134 struct memory_block
*mem
=
135 container_of(dev
, struct memory_block
, dev
);
136 unsigned long phys_index
;
138 phys_index
= mem
->start_section_nr
/ sections_per_block
;
139 return sprintf(buf
, "%08lx\n", phys_index
);
142 static ssize_t
show_mem_end_phys_index(struct device
*dev
,
143 struct device_attribute
*attr
, char *buf
)
145 struct memory_block
*mem
=
146 container_of(dev
, struct memory_block
, dev
);
147 unsigned long phys_index
;
149 phys_index
= mem
->end_section_nr
/ sections_per_block
;
150 return sprintf(buf
, "%08lx\n", phys_index
);
154 * Show whether the section of memory is likely to be hot-removable
156 static ssize_t
show_mem_removable(struct device
*dev
,
157 struct device_attribute
*attr
, char *buf
)
159 unsigned long i
, pfn
;
161 struct memory_block
*mem
=
162 container_of(dev
, struct memory_block
, dev
);
164 for (i
= 0; i
< sections_per_block
; i
++) {
165 pfn
= section_nr_to_pfn(mem
->start_section_nr
+ i
);
166 ret
&= is_mem_section_removable(pfn
, PAGES_PER_SECTION
);
169 return sprintf(buf
, "%d\n", ret
);
173 * online, offline, going offline, etc.
175 static ssize_t
show_mem_state(struct device
*dev
,
176 struct device_attribute
*attr
, char *buf
)
178 struct memory_block
*mem
=
179 container_of(dev
, struct memory_block
, dev
);
183 * We can probably put these states in a nice little array
184 * so that they're not open-coded
186 switch (mem
->state
) {
188 len
= sprintf(buf
, "online\n");
191 len
= sprintf(buf
, "offline\n");
193 case MEM_GOING_OFFLINE
:
194 len
= sprintf(buf
, "going-offline\n");
197 len
= sprintf(buf
, "ERROR-UNKNOWN-%ld\n",
206 int memory_notify(unsigned long val
, void *v
)
208 return blocking_notifier_call_chain(&memory_chain
, val
, v
);
211 int memory_isolate_notify(unsigned long val
, void *v
)
213 return atomic_notifier_call_chain(&memory_isolate_chain
, val
, v
);
217 * The probe routines leave the pages reserved, just as the bootmem code does.
218 * Make sure they're still that way.
220 static bool pages_correctly_reserved(unsigned long start_pfn
,
221 unsigned long nr_pages
)
225 unsigned long pfn
= start_pfn
;
228 * memmap between sections is not contiguous except with
229 * SPARSEMEM_VMEMMAP. We lookup the page once per section
230 * and assume memmap is contiguous within each section
232 for (i
= 0; i
< sections_per_block
; i
++, pfn
+= PAGES_PER_SECTION
) {
233 if (WARN_ON_ONCE(!pfn_valid(pfn
)))
235 page
= pfn_to_page(pfn
);
237 for (j
= 0; j
< PAGES_PER_SECTION
; j
++) {
238 if (PageReserved(page
+ j
))
241 printk(KERN_WARNING
"section number %ld page number %d "
242 "not reserved, was it already online?\n",
243 pfn_to_section_nr(pfn
), j
);
253 * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is
254 * OK to have direct references to sparsemem variables in here.
257 memory_block_action(unsigned long phys_index
, unsigned long action
, int online_type
)
259 unsigned long start_pfn
;
260 unsigned long nr_pages
= PAGES_PER_SECTION
* sections_per_block
;
261 struct page
*first_page
;
264 first_page
= pfn_to_page(phys_index
<< PFN_SECTION_SHIFT
);
265 start_pfn
= page_to_pfn(first_page
);
269 if (!pages_correctly_reserved(start_pfn
, nr_pages
))
272 ret
= online_pages(start_pfn
, nr_pages
, online_type
);
275 ret
= offline_pages(start_pfn
, nr_pages
);
278 WARN(1, KERN_WARNING
"%s(%ld, %ld) unknown action: "
279 "%ld\n", __func__
, phys_index
, action
, action
);
286 static int __memory_block_change_state(struct memory_block
*mem
,
287 unsigned long to_state
, unsigned long from_state_req
,
292 if (mem
->state
!= from_state_req
) {
297 if (to_state
== MEM_OFFLINE
)
298 mem
->state
= MEM_GOING_OFFLINE
;
300 ret
= memory_block_action(mem
->start_section_nr
, to_state
, online_type
);
303 mem
->state
= from_state_req
;
307 mem
->state
= to_state
;
308 switch (mem
->state
) {
310 kobject_uevent(&mem
->dev
.kobj
, KOBJ_OFFLINE
);
313 kobject_uevent(&mem
->dev
.kobj
, KOBJ_ONLINE
);
322 static int memory_block_change_state(struct memory_block
*mem
,
323 unsigned long to_state
, unsigned long from_state_req
,
328 mutex_lock(&mem
->state_mutex
);
329 ret
= __memory_block_change_state(mem
, to_state
, from_state_req
,
331 mutex_unlock(&mem
->state_mutex
);
336 store_mem_state(struct device
*dev
,
337 struct device_attribute
*attr
, const char *buf
, size_t count
)
339 struct memory_block
*mem
;
342 mem
= container_of(dev
, struct memory_block
, dev
);
344 if (!strncmp(buf
, "online_kernel", min_t(int, count
, 13)))
345 ret
= memory_block_change_state(mem
, MEM_ONLINE
,
346 MEM_OFFLINE
, ONLINE_KERNEL
);
347 else if (!strncmp(buf
, "online_movable", min_t(int, count
, 14)))
348 ret
= memory_block_change_state(mem
, MEM_ONLINE
,
349 MEM_OFFLINE
, ONLINE_MOVABLE
);
350 else if (!strncmp(buf
, "online", min_t(int, count
, 6)))
351 ret
= memory_block_change_state(mem
, MEM_ONLINE
,
352 MEM_OFFLINE
, ONLINE_KEEP
);
353 else if(!strncmp(buf
, "offline", min_t(int, count
, 7)))
354 ret
= memory_block_change_state(mem
, MEM_OFFLINE
,
363 * phys_device is a bad name for this. What I really want
364 * is a way to differentiate between memory ranges that
365 * are part of physical devices that constitute
366 * a complete removable unit or fru.
367 * i.e. do these ranges belong to the same physical device,
368 * s.t. if I offline all of these sections I can then
369 * remove the physical device?
371 static ssize_t
show_phys_device(struct device
*dev
,
372 struct device_attribute
*attr
, char *buf
)
374 struct memory_block
*mem
=
375 container_of(dev
, struct memory_block
, dev
);
376 return sprintf(buf
, "%d\n", mem
->phys_device
);
379 static DEVICE_ATTR(phys_index
, 0444, show_mem_start_phys_index
, NULL
);
380 static DEVICE_ATTR(end_phys_index
, 0444, show_mem_end_phys_index
, NULL
);
381 static DEVICE_ATTR(state
, 0644, show_mem_state
, store_mem_state
);
382 static DEVICE_ATTR(phys_device
, 0444, show_phys_device
, NULL
);
383 static DEVICE_ATTR(removable
, 0444, show_mem_removable
, NULL
);
385 #define mem_create_simple_file(mem, attr_name) \
386 device_create_file(&mem->dev, &dev_attr_##attr_name)
387 #define mem_remove_simple_file(mem, attr_name) \
388 device_remove_file(&mem->dev, &dev_attr_##attr_name)
391 * Block size attribute stuff
394 print_block_size(struct device
*dev
, struct device_attribute
*attr
,
397 return sprintf(buf
, "%lx\n", get_memory_block_size());
400 static DEVICE_ATTR(block_size_bytes
, 0444, print_block_size
, NULL
);
402 static int block_size_init(void)
404 return device_create_file(memory_subsys
.dev_root
,
405 &dev_attr_block_size_bytes
);
409 * Some architectures will have custom drivers to do this, and
410 * will not need to do it from userspace. The fake hot-add code
411 * as well as ppc64 will do all of their discovery in userspace
412 * and will require this interface.
414 #ifdef CONFIG_ARCH_MEMORY_PROBE
416 memory_probe_store(struct device
*dev
, struct device_attribute
*attr
,
417 const char *buf
, size_t count
)
422 unsigned long pages_per_block
= PAGES_PER_SECTION
* sections_per_block
;
424 phys_addr
= simple_strtoull(buf
, NULL
, 0);
426 if (phys_addr
& ((pages_per_block
<< PAGE_SHIFT
) - 1))
429 for (i
= 0; i
< sections_per_block
; i
++) {
430 nid
= memory_add_physaddr_to_nid(phys_addr
);
431 ret
= add_memory(nid
, phys_addr
,
432 PAGES_PER_SECTION
<< PAGE_SHIFT
);
436 phys_addr
+= MIN_MEMORY_BLOCK_SIZE
;
443 static DEVICE_ATTR(probe
, S_IWUSR
, NULL
, memory_probe_store
);
445 static int memory_probe_init(void)
447 return device_create_file(memory_subsys
.dev_root
, &dev_attr_probe
);
450 static inline int memory_probe_init(void)
456 #ifdef CONFIG_MEMORY_FAILURE
458 * Support for offlining pages of memory
461 /* Soft offline a page */
463 store_soft_offline_page(struct device
*dev
,
464 struct device_attribute
*attr
,
465 const char *buf
, size_t count
)
469 if (!capable(CAP_SYS_ADMIN
))
471 if (strict_strtoull(buf
, 0, &pfn
) < 0)
476 ret
= soft_offline_page(pfn_to_page(pfn
), 0);
477 return ret
== 0 ? count
: ret
;
480 /* Forcibly offline a page, including killing processes. */
482 store_hard_offline_page(struct device
*dev
,
483 struct device_attribute
*attr
,
484 const char *buf
, size_t count
)
488 if (!capable(CAP_SYS_ADMIN
))
490 if (strict_strtoull(buf
, 0, &pfn
) < 0)
493 ret
= memory_failure(pfn
, 0, 0);
494 return ret
? ret
: count
;
497 static DEVICE_ATTR(soft_offline_page
, 0644, NULL
, store_soft_offline_page
);
498 static DEVICE_ATTR(hard_offline_page
, 0644, NULL
, store_hard_offline_page
);
500 static __init
int memory_fail_init(void)
504 err
= device_create_file(memory_subsys
.dev_root
,
505 &dev_attr_soft_offline_page
);
507 err
= device_create_file(memory_subsys
.dev_root
,
508 &dev_attr_hard_offline_page
);
512 static inline int memory_fail_init(void)
519 * Note that phys_device is optional. It is here to allow for
520 * differentiation between which *physical* devices each
521 * section belongs to...
523 int __weak
arch_get_memory_phys_device(unsigned long start_pfn
)
529 * A reference for the returned object is held and the reference for the
530 * hinted object is released.
532 struct memory_block
*find_memory_block_hinted(struct mem_section
*section
,
533 struct memory_block
*hint
)
535 int block_id
= base_memory_block_id(__section_nr(section
));
536 struct device
*hintdev
= hint
? &hint
->dev
: NULL
;
539 dev
= subsys_find_device_by_id(&memory_subsys
, block_id
, hintdev
);
541 put_device(&hint
->dev
);
544 return container_of(dev
, struct memory_block
, dev
);
548 * For now, we have a linear search to go find the appropriate
549 * memory_block corresponding to a particular phys_index. If
550 * this gets to be a real problem, we can always use a radix
551 * tree or something here.
553 * This could be made generic for all device subsystems.
555 struct memory_block
*find_memory_block(struct mem_section
*section
)
557 return find_memory_block_hinted(section
, NULL
);
560 static int init_memory_block(struct memory_block
**memory
,
561 struct mem_section
*section
, unsigned long state
)
563 struct memory_block
*mem
;
564 unsigned long start_pfn
;
568 mem
= kzalloc(sizeof(*mem
), GFP_KERNEL
);
572 scn_nr
= __section_nr(section
);
573 mem
->start_section_nr
=
574 base_memory_block_id(scn_nr
) * sections_per_block
;
575 mem
->end_section_nr
= mem
->start_section_nr
+ sections_per_block
- 1;
577 mem
->section_count
++;
578 mutex_init(&mem
->state_mutex
);
579 start_pfn
= section_nr_to_pfn(mem
->start_section_nr
);
580 mem
->phys_device
= arch_get_memory_phys_device(start_pfn
);
582 ret
= register_memory(mem
);
584 ret
= mem_create_simple_file(mem
, phys_index
);
586 ret
= mem_create_simple_file(mem
, end_phys_index
);
588 ret
= mem_create_simple_file(mem
, state
);
590 ret
= mem_create_simple_file(mem
, phys_device
);
592 ret
= mem_create_simple_file(mem
, removable
);
598 static int add_memory_section(int nid
, struct mem_section
*section
,
599 struct memory_block
**mem_p
,
600 unsigned long state
, enum mem_add_context context
)
602 struct memory_block
*mem
= NULL
;
603 int scn_nr
= __section_nr(section
);
606 mutex_lock(&mem_sysfs_mutex
);
608 if (context
== BOOT
) {
609 /* same memory block ? */
611 if (scn_nr
>= (*mem_p
)->start_section_nr
&&
612 scn_nr
<= (*mem_p
)->end_section_nr
) {
614 kobject_get(&mem
->dev
.kobj
);
617 mem
= find_memory_block(section
);
620 mem
->section_count
++;
621 kobject_put(&mem
->dev
.kobj
);
623 ret
= init_memory_block(&mem
, section
, state
);
624 /* store memory_block pointer for next loop */
625 if (!ret
&& context
== BOOT
)
631 if (context
== HOTPLUG
&&
632 mem
->section_count
== sections_per_block
)
633 ret
= register_mem_sect_under_node(mem
, nid
);
636 mutex_unlock(&mem_sysfs_mutex
);
640 int remove_memory_block(unsigned long node_id
, struct mem_section
*section
,
643 struct memory_block
*mem
;
645 mutex_lock(&mem_sysfs_mutex
);
646 mem
= find_memory_block(section
);
647 unregister_mem_sect_under_nodes(mem
, __section_nr(section
));
649 mem
->section_count
--;
650 if (mem
->section_count
== 0) {
651 mem_remove_simple_file(mem
, phys_index
);
652 mem_remove_simple_file(mem
, end_phys_index
);
653 mem_remove_simple_file(mem
, state
);
654 mem_remove_simple_file(mem
, phys_device
);
655 mem_remove_simple_file(mem
, removable
);
656 unregister_memory(mem
);
658 kobject_put(&mem
->dev
.kobj
);
660 mutex_unlock(&mem_sysfs_mutex
);
665 * need an interface for the VM to add new memory regions,
666 * but without onlining it.
668 int register_new_memory(int nid
, struct mem_section
*section
)
670 return add_memory_section(nid
, section
, NULL
, MEM_OFFLINE
, HOTPLUG
);
673 int unregister_memory_section(struct mem_section
*section
)
675 if (!present_section(section
))
678 return remove_memory_block(0, section
, 0);
682 * offline one memory block. If the memory block has been offlined, do nothing.
684 int offline_memory_block(struct memory_block
*mem
)
688 mutex_lock(&mem
->state_mutex
);
689 if (mem
->state
!= MEM_OFFLINE
)
690 ret
= __memory_block_change_state(mem
, MEM_OFFLINE
, MEM_ONLINE
, -1);
691 mutex_unlock(&mem
->state_mutex
);
697 * Initialize the sysfs support for memory devices...
699 int __init
memory_dev_init(void)
704 unsigned long block_sz
;
705 struct memory_block
*mem
= NULL
;
707 ret
= subsys_system_register(&memory_subsys
, NULL
);
711 block_sz
= get_memory_block_size();
712 sections_per_block
= block_sz
/ MIN_MEMORY_BLOCK_SIZE
;
715 * Create entries for memory sections that were found
716 * during boot and have been initialized
718 for (i
= 0; i
< NR_MEM_SECTIONS
; i
++) {
719 if (!present_section_nr(i
))
721 /* don't need to reuse memory_block if only one per block */
722 err
= add_memory_section(0, __nr_to_section(i
),
723 (sections_per_block
== 1) ? NULL
: &mem
,
730 err
= memory_probe_init();
733 err
= memory_fail_init();
736 err
= block_size_init();
741 printk(KERN_ERR
"%s() failed: %d\n", __func__
, ret
);