2 #include <linux/mmzone.h>
3 #include <linux/bootmem.h>
4 #include <linux/bit_spinlock.h>
5 #include <linux/page_cgroup.h>
6 #include <linux/hash.h>
7 #include <linux/slab.h>
8 #include <linux/memory.h>
9 #include <linux/vmalloc.h>
10 #include <linux/cgroup.h>
11 #include <linux/swapops.h>
12 #include <linux/kmemleak.h>
14 static void __meminit
init_page_cgroup(struct page_cgroup
*pc
, unsigned long id
)
17 set_page_cgroup_array_id(pc
, id
);
18 pc
->mem_cgroup
= NULL
;
20 static unsigned long total_usage
;
22 #if !defined(CONFIG_SPARSEMEM)
25 void __meminit
pgdat_page_cgroup_init(struct pglist_data
*pgdat
)
27 pgdat
->node_page_cgroup
= NULL
;
30 struct page_cgroup
*lookup_page_cgroup(struct page
*page
)
32 unsigned long pfn
= page_to_pfn(page
);
34 struct page_cgroup
*base
;
36 base
= NODE_DATA(page_to_nid(page
))->node_page_cgroup
;
40 offset
= pfn
- NODE_DATA(page_to_nid(page
))->node_start_pfn
;
44 struct page
*lookup_cgroup_page(struct page_cgroup
*pc
)
50 pgdat
= NODE_DATA(page_cgroup_array_id(pc
));
51 pfn
= pc
- pgdat
->node_page_cgroup
+ pgdat
->node_start_pfn
;
52 page
= pfn_to_page(pfn
);
53 VM_BUG_ON(pc
!= lookup_page_cgroup(page
));
57 static int __init
alloc_node_page_cgroup(int nid
)
59 struct page_cgroup
*base
, *pc
;
60 unsigned long table_size
;
61 unsigned long start_pfn
, nr_pages
, index
;
63 start_pfn
= NODE_DATA(nid
)->node_start_pfn
;
64 nr_pages
= NODE_DATA(nid
)->node_spanned_pages
;
69 table_size
= sizeof(struct page_cgroup
) * nr_pages
;
71 base
= __alloc_bootmem_node_nopanic(NODE_DATA(nid
),
72 table_size
, PAGE_SIZE
, __pa(MAX_DMA_ADDRESS
));
75 for (index
= 0; index
< nr_pages
; index
++) {
77 init_page_cgroup(pc
, nid
);
79 NODE_DATA(nid
)->node_page_cgroup
= base
;
80 total_usage
+= table_size
;
84 void __init
page_cgroup_init_flatmem(void)
89 if (mem_cgroup_disabled())
92 for_each_online_node(nid
) {
93 fail
= alloc_node_page_cgroup(nid
);
97 printk(KERN_INFO
"allocated %ld bytes of page_cgroup\n", total_usage
);
98 printk(KERN_INFO
"please try 'cgroup_disable=memory' option if you"
99 " don't want memory cgroups\n");
102 printk(KERN_CRIT
"allocation of page_cgroup failed.\n");
103 printk(KERN_CRIT
"please try 'cgroup_disable=memory' boot option\n");
104 panic("Out of memory");
107 #else /* CONFIG_FLAT_NODE_MEM_MAP */
109 struct page_cgroup
*lookup_page_cgroup(struct page
*page
)
111 unsigned long pfn
= page_to_pfn(page
);
112 struct mem_section
*section
= __pfn_to_section(pfn
);
114 if (!section
->page_cgroup
)
116 return section
->page_cgroup
+ pfn
;
119 struct page
*lookup_cgroup_page(struct page_cgroup
*pc
)
121 struct mem_section
*section
;
125 nr
= page_cgroup_array_id(pc
);
126 section
= __nr_to_section(nr
);
127 page
= pfn_to_page(pc
- section
->page_cgroup
);
128 VM_BUG_ON(pc
!= lookup_page_cgroup(page
));
132 static void *__meminit
alloc_page_cgroup(size_t size
, int nid
)
135 gfp_t flags
= GFP_KERNEL
| __GFP_NOWARN
;
137 addr
= alloc_pages_exact_nid(nid
, size
, flags
);
139 kmemleak_alloc(addr
, size
, 1, flags
);
143 if (node_state(nid
, N_HIGH_MEMORY
))
144 addr
= vmalloc_node(size
, nid
);
146 addr
= vmalloc(size
);
151 #ifdef CONFIG_MEMORY_HOTPLUG
152 static void free_page_cgroup(void *addr
)
154 if (is_vmalloc_addr(addr
)) {
157 struct page
*page
= virt_to_page(addr
);
159 sizeof(struct page_cgroup
) * PAGES_PER_SECTION
;
161 BUG_ON(PageReserved(page
));
162 free_pages_exact(addr
, table_size
);
167 static int __meminit
init_section_page_cgroup(unsigned long pfn
, int nid
)
169 struct page_cgroup
*base
, *pc
;
170 struct mem_section
*section
;
171 unsigned long table_size
;
175 nr
= pfn_to_section_nr(pfn
);
176 section
= __nr_to_section(nr
);
178 if (section
->page_cgroup
)
181 table_size
= sizeof(struct page_cgroup
) * PAGES_PER_SECTION
;
182 base
= alloc_page_cgroup(table_size
, nid
);
185 * The value stored in section->page_cgroup is (base - pfn)
186 * and it does not point to the memory block allocated above,
187 * causing kmemleak false positives.
189 kmemleak_not_leak(base
);
192 printk(KERN_ERR
"page cgroup allocation failure\n");
196 for (index
= 0; index
< PAGES_PER_SECTION
; index
++) {
198 init_page_cgroup(pc
, nr
);
201 * The passed "pfn" may not be aligned to SECTION. For the calculation
202 * we need to apply a mask.
204 pfn
&= PAGE_SECTION_MASK
;
205 section
->page_cgroup
= base
- pfn
;
206 total_usage
+= table_size
;
209 #ifdef CONFIG_MEMORY_HOTPLUG
210 void __free_page_cgroup(unsigned long pfn
)
212 struct mem_section
*ms
;
213 struct page_cgroup
*base
;
215 ms
= __pfn_to_section(pfn
);
216 if (!ms
|| !ms
->page_cgroup
)
218 base
= ms
->page_cgroup
+ pfn
;
219 free_page_cgroup(base
);
220 ms
->page_cgroup
= NULL
;
223 int __meminit
online_page_cgroup(unsigned long start_pfn
,
224 unsigned long nr_pages
,
227 unsigned long start
, end
, pfn
;
230 start
= SECTION_ALIGN_DOWN(start_pfn
);
231 end
= SECTION_ALIGN_UP(start_pfn
+ nr_pages
);
235 * In this case, "nid" already exists and contains valid memory.
236 * "start_pfn" passed to us is a pfn which is an arg for
237 * online__pages(), and start_pfn should exist.
239 nid
= pfn_to_nid(start_pfn
);
240 VM_BUG_ON(!node_state(nid
, N_ONLINE
));
243 for (pfn
= start
; !fail
&& pfn
< end
; pfn
+= PAGES_PER_SECTION
) {
244 if (!pfn_present(pfn
))
246 fail
= init_section_page_cgroup(pfn
, nid
);
252 for (pfn
= start
; pfn
< end
; pfn
+= PAGES_PER_SECTION
)
253 __free_page_cgroup(pfn
);
258 int __meminit
offline_page_cgroup(unsigned long start_pfn
,
259 unsigned long nr_pages
, int nid
)
261 unsigned long start
, end
, pfn
;
263 start
= SECTION_ALIGN_DOWN(start_pfn
);
264 end
= SECTION_ALIGN_UP(start_pfn
+ nr_pages
);
266 for (pfn
= start
; pfn
< end
; pfn
+= PAGES_PER_SECTION
)
267 __free_page_cgroup(pfn
);
272 static int __meminit
page_cgroup_callback(struct notifier_block
*self
,
273 unsigned long action
, void *arg
)
275 struct memory_notify
*mn
= arg
;
278 case MEM_GOING_ONLINE
:
279 ret
= online_page_cgroup(mn
->start_pfn
,
280 mn
->nr_pages
, mn
->status_change_nid
);
283 offline_page_cgroup(mn
->start_pfn
,
284 mn
->nr_pages
, mn
->status_change_nid
);
286 case MEM_CANCEL_ONLINE
:
287 case MEM_GOING_OFFLINE
:
290 case MEM_CANCEL_OFFLINE
:
294 return notifier_from_errno(ret
);
299 void __init
page_cgroup_init(void)
304 if (mem_cgroup_disabled())
307 for_each_node_state(nid
, N_HIGH_MEMORY
) {
308 unsigned long start_pfn
, end_pfn
;
310 start_pfn
= node_start_pfn(nid
);
311 end_pfn
= node_end_pfn(nid
);
313 * start_pfn and end_pfn may not be aligned to SECTION and the
314 * page->flags of out of node pages are not initialized. So we
315 * scan [start_pfn, the biggest section's pfn < end_pfn) here.
317 for (pfn
= start_pfn
;
319 pfn
= ALIGN(pfn
+ 1, PAGES_PER_SECTION
)) {
324 * Nodes's pfns can be overlapping.
325 * We know some arch can have a nodes layout such as
326 * -------------pfn-------------->
327 * N0 | N1 | N2 | N0 | N1 | N2|....
329 if (pfn_to_nid(pfn
) != nid
)
331 if (init_section_page_cgroup(pfn
, nid
))
335 hotplug_memory_notifier(page_cgroup_callback
, 0);
336 printk(KERN_INFO
"allocated %ld bytes of page_cgroup\n", total_usage
);
337 printk(KERN_INFO
"please try 'cgroup_disable=memory' option if you "
338 "don't want memory cgroups\n");
341 printk(KERN_CRIT
"try 'cgroup_disable=memory' boot option\n");
342 panic("Out of memory");
345 void __meminit
pgdat_page_cgroup_init(struct pglist_data
*pgdat
)
353 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
355 static DEFINE_MUTEX(swap_cgroup_mutex
);
356 struct swap_cgroup_ctrl
{
358 unsigned long length
;
362 static struct swap_cgroup_ctrl swap_cgroup_ctrl
[MAX_SWAPFILES
];
367 #define SC_PER_PAGE (PAGE_SIZE/sizeof(struct swap_cgroup))
368 #define SC_POS_MASK (SC_PER_PAGE - 1)
371 * SwapCgroup implements "lookup" and "exchange" operations.
372 * In typical usage, this swap_cgroup is accessed via memcg's charge/uncharge
373 * against SwapCache. At swap_free(), this is accessed directly from swap.
376 * - we have no race in "exchange" when we're accessed via SwapCache because
377 * SwapCache(and its swp_entry) is under lock.
378 * - When called via swap_free(), there is no user of this entry and no race.
379 * Then, we don't need lock around "exchange".
381 * TODO: we can push these buffers out to HIGHMEM.
385 * allocate buffer for swap_cgroup.
387 static int swap_cgroup_prepare(int type
)
390 struct swap_cgroup_ctrl
*ctrl
;
391 unsigned long idx
, max
;
393 ctrl
= &swap_cgroup_ctrl
[type
];
395 for (idx
= 0; idx
< ctrl
->length
; idx
++) {
396 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
398 goto not_enough_page
;
399 ctrl
->map
[idx
] = page
;
404 for (idx
= 0; idx
< max
; idx
++)
405 __free_page(ctrl
->map
[idx
]);
411 * swap_cgroup_cmpxchg - cmpxchg mem_cgroup's id for this swp_entry.
412 * @end: swap entry to be cmpxchged
416 * Returns old id at success, 0 at failure.
417 * (There is no mem_cgroup using 0 as its id)
419 unsigned short swap_cgroup_cmpxchg(swp_entry_t ent
,
420 unsigned short old
, unsigned short new)
422 int type
= swp_type(ent
);
423 unsigned long offset
= swp_offset(ent
);
424 unsigned long idx
= offset
/ SC_PER_PAGE
;
425 unsigned long pos
= offset
& SC_POS_MASK
;
426 struct swap_cgroup_ctrl
*ctrl
;
427 struct page
*mappage
;
428 struct swap_cgroup
*sc
;
430 unsigned short retval
;
432 ctrl
= &swap_cgroup_ctrl
[type
];
434 mappage
= ctrl
->map
[idx
];
435 sc
= page_address(mappage
);
437 spin_lock_irqsave(&ctrl
->lock
, flags
);
443 spin_unlock_irqrestore(&ctrl
->lock
, flags
);
448 * swap_cgroup_record - record mem_cgroup for this swp_entry.
449 * @ent: swap entry to be recorded into
450 * @mem: mem_cgroup to be recorded
452 * Returns old value at success, 0 at failure.
453 * (Of course, old value can be 0.)
455 unsigned short swap_cgroup_record(swp_entry_t ent
, unsigned short id
)
457 int type
= swp_type(ent
);
458 unsigned long offset
= swp_offset(ent
);
459 unsigned long idx
= offset
/ SC_PER_PAGE
;
460 unsigned long pos
= offset
& SC_POS_MASK
;
461 struct swap_cgroup_ctrl
*ctrl
;
462 struct page
*mappage
;
463 struct swap_cgroup
*sc
;
467 ctrl
= &swap_cgroup_ctrl
[type
];
469 mappage
= ctrl
->map
[idx
];
470 sc
= page_address(mappage
);
472 spin_lock_irqsave(&ctrl
->lock
, flags
);
475 spin_unlock_irqrestore(&ctrl
->lock
, flags
);
481 * lookup_swap_cgroup - lookup mem_cgroup tied to swap entry
482 * @ent: swap entry to be looked up.
484 * Returns CSS ID of mem_cgroup at success. 0 at failure. (0 is invalid ID)
486 unsigned short lookup_swap_cgroup(swp_entry_t ent
)
488 int type
= swp_type(ent
);
489 unsigned long offset
= swp_offset(ent
);
490 unsigned long idx
= offset
/ SC_PER_PAGE
;
491 unsigned long pos
= offset
& SC_POS_MASK
;
492 struct swap_cgroup_ctrl
*ctrl
;
493 struct page
*mappage
;
494 struct swap_cgroup
*sc
;
497 ctrl
= &swap_cgroup_ctrl
[type
];
498 mappage
= ctrl
->map
[idx
];
499 sc
= page_address(mappage
);
505 int swap_cgroup_swapon(int type
, unsigned long max_pages
)
508 unsigned long array_size
;
509 unsigned long length
;
510 struct swap_cgroup_ctrl
*ctrl
;
512 if (!do_swap_account
)
515 length
= DIV_ROUND_UP(max_pages
, SC_PER_PAGE
);
516 array_size
= length
* sizeof(void *);
518 array
= vzalloc(array_size
);
522 ctrl
= &swap_cgroup_ctrl
[type
];
523 mutex_lock(&swap_cgroup_mutex
);
524 ctrl
->length
= length
;
526 spin_lock_init(&ctrl
->lock
);
527 if (swap_cgroup_prepare(type
)) {
528 /* memory shortage */
531 mutex_unlock(&swap_cgroup_mutex
);
535 mutex_unlock(&swap_cgroup_mutex
);
539 printk(KERN_INFO
"couldn't allocate enough memory for swap_cgroup.\n");
541 "swap_cgroup can be disabled by swapaccount=0 boot option\n");
545 void swap_cgroup_swapoff(int type
)
548 unsigned long i
, length
;
549 struct swap_cgroup_ctrl
*ctrl
;
551 if (!do_swap_account
)
554 mutex_lock(&swap_cgroup_mutex
);
555 ctrl
= &swap_cgroup_ctrl
[type
];
557 length
= ctrl
->length
;
560 mutex_unlock(&swap_cgroup_mutex
);
563 for (i
= 0; i
< length
; i
++) {
564 struct page
*page
= map
[i
];