aboutsummaryrefslogtreecommitdiffstats
path: root/mm
diff options
context:
space:
mode:
Diffstat (limited to 'mm')
-rw-r--r--mm/Kconfig14
-rw-r--r--mm/Makefile8
-rw-r--r--mm/filemap.c6
-rw-r--r--mm/hugetlb.c12
-rw-r--r--mm/hwpoison-inject.c41
-rw-r--r--mm/ksm.c14
-rw-r--r--mm/madvise.c30
-rw-r--r--mm/memcontrol.c737
-rw-r--r--mm/memory-failure.c832
-rw-r--r--mm/memory.c86
-rw-r--r--mm/memory_hotplug.c6
-rw-r--r--mm/migrate.c2
-rw-r--r--mm/mremap.c4
-rw-r--r--mm/nommu.c85
-rw-r--r--mm/page-writeback.c27
-rw-r--r--mm/page_alloc.c44
-rw-r--r--mm/quicklist.c3
-rw-r--r--mm/rmap.c60
-rw-r--r--mm/shmem.c5
-rw-r--r--mm/swapfile.c4
-rw-r--r--mm/truncate.c136
-rw-r--r--mm/vmalloc.c2
-rw-r--r--mm/vmscan.c51
23 files changed, 1901 insertions, 308 deletions
diff --git a/mm/Kconfig b/mm/Kconfig
index 71eb0b4cce8d..247760729593 100644
--- a/mm/Kconfig
+++ b/mm/Kconfig
@@ -245,6 +245,20 @@ config DEFAULT_MMAP_MIN_ADDR
245 /proc/sys/vm/mmap_min_addr tunable. 245 /proc/sys/vm/mmap_min_addr tunable.
246 246
247 247
248config MEMORY_FAILURE
249 depends on MMU
250 depends on X86_MCE
251 bool "Enable recovery from hardware memory errors"
252 help
253 Enables code to recover from some memory failures on systems
254 with MCA recovery. This allows a system to continue running
255 even when some of its memory has uncorrected errors. This requires
256 special hardware support and typically ECC memory.
257
258config HWPOISON_INJECT
259 tristate "Poison pages injector"
260 depends on MEMORY_FAILURE && DEBUG_KERNEL
261
248config NOMMU_INITIAL_TRIM_EXCESS 262config NOMMU_INITIAL_TRIM_EXCESS
249 int "Turn on mmap() excess space trimming before booting" 263 int "Turn on mmap() excess space trimming before booting"
250 depends on !MMU 264 depends on !MMU
diff --git a/mm/Makefile b/mm/Makefile
index 728a9fde49d1..ebf849042ed3 100644
--- a/mm/Makefile
+++ b/mm/Makefile
@@ -5,16 +5,16 @@
5mmu-y := nommu.o 5mmu-y := nommu.o
6mmu-$(CONFIG_MMU) := fremap.o highmem.o madvise.o memory.o mincore.o \ 6mmu-$(CONFIG_MMU) := fremap.o highmem.o madvise.o memory.o mincore.o \
7 mlock.o mmap.o mprotect.o mremap.o msync.o rmap.o \ 7 mlock.o mmap.o mprotect.o mremap.o msync.o rmap.o \
8 vmalloc.o 8 vmalloc.o pagewalk.o
9 9
10obj-y := bootmem.o filemap.o mempool.o oom_kill.o fadvise.o \ 10obj-y := bootmem.o filemap.o mempool.o oom_kill.o fadvise.o \
11 maccess.o page_alloc.o page-writeback.o \ 11 maccess.o page_alloc.o page-writeback.o \
12 readahead.o swap.o truncate.o vmscan.o shmem.o \ 12 readahead.o swap.o truncate.o vmscan.o shmem.o \
13 prio_tree.o util.o mmzone.o vmstat.o backing-dev.o \ 13 prio_tree.o util.o mmzone.o vmstat.o backing-dev.o \
14 page_isolation.o mm_init.o mmu_context.o $(mmu-y) 14 page_isolation.o mm_init.o mmu_context.o \
15 $(mmu-y)
15obj-y += init-mm.o 16obj-y += init-mm.o
16 17
17obj-$(CONFIG_PROC_PAGE_MONITOR) += pagewalk.o
18obj-$(CONFIG_BOUNCE) += bounce.o 18obj-$(CONFIG_BOUNCE) += bounce.o
19obj-$(CONFIG_SWAP) += page_io.o swap_state.o swapfile.o thrash.o 19obj-$(CONFIG_SWAP) += page_io.o swap_state.o swapfile.o thrash.o
20obj-$(CONFIG_HAS_DMA) += dmapool.o 20obj-$(CONFIG_HAS_DMA) += dmapool.o
@@ -41,5 +41,7 @@ obj-$(CONFIG_SMP) += allocpercpu.o
41endif 41endif
42obj-$(CONFIG_QUICKLIST) += quicklist.o 42obj-$(CONFIG_QUICKLIST) += quicklist.o
43obj-$(CONFIG_CGROUP_MEM_RES_CTLR) += memcontrol.o page_cgroup.o 43obj-$(CONFIG_CGROUP_MEM_RES_CTLR) += memcontrol.o page_cgroup.o
44obj-$(CONFIG_MEMORY_FAILURE) += memory-failure.o
45obj-$(CONFIG_HWPOISON_INJECT) += hwpoison-inject.o
44obj-$(CONFIG_DEBUG_KMEMLEAK) += kmemleak.o 46obj-$(CONFIG_DEBUG_KMEMLEAK) += kmemleak.o
45obj-$(CONFIG_DEBUG_KMEMLEAK_TEST) += kmemleak-test.o 47obj-$(CONFIG_DEBUG_KMEMLEAK_TEST) += kmemleak-test.o
diff --git a/mm/filemap.c b/mm/filemap.c
index bcc7372aebbc..6c84e598b4a9 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -58,7 +58,7 @@
58/* 58/*
59 * Lock ordering: 59 * Lock ordering:
60 * 60 *
61 * ->i_mmap_lock (vmtruncate) 61 * ->i_mmap_lock (truncate_pagecache)
62 * ->private_lock (__free_pte->__set_page_dirty_buffers) 62 * ->private_lock (__free_pte->__set_page_dirty_buffers)
63 * ->swap_lock (exclusive_swap_page, others) 63 * ->swap_lock (exclusive_swap_page, others)
64 * ->mapping->tree_lock 64 * ->mapping->tree_lock
@@ -104,6 +104,10 @@
104 * 104 *
105 * ->task->proc_lock 105 * ->task->proc_lock
106 * ->dcache_lock (proc_pid_lookup) 106 * ->dcache_lock (proc_pid_lookup)
107 *
108 * (code doesn't rely on that order, so you could switch it around)
109 * ->tasklist_lock (memory_failure, collect_procs_ao)
110 * ->i_mmap_lock
107 */ 111 */
108 112
109/* 113/*
diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index 815dbd4a6dcb..6f048fcc749c 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -1537,7 +1537,7 @@ static unsigned int cpuset_mems_nr(unsigned int *array)
1537 1537
1538#ifdef CONFIG_SYSCTL 1538#ifdef CONFIG_SYSCTL
1539int hugetlb_sysctl_handler(struct ctl_table *table, int write, 1539int hugetlb_sysctl_handler(struct ctl_table *table, int write,
1540 struct file *file, void __user *buffer, 1540 void __user *buffer,
1541 size_t *length, loff_t *ppos) 1541 size_t *length, loff_t *ppos)
1542{ 1542{
1543 struct hstate *h = &default_hstate; 1543 struct hstate *h = &default_hstate;
@@ -1548,7 +1548,7 @@ int hugetlb_sysctl_handler(struct ctl_table *table, int write,
1548 1548
1549 table->data = &tmp; 1549 table->data = &tmp;
1550 table->maxlen = sizeof(unsigned long); 1550 table->maxlen = sizeof(unsigned long);
1551 proc_doulongvec_minmax(table, write, file, buffer, length, ppos); 1551 proc_doulongvec_minmax(table, write, buffer, length, ppos);
1552 1552
1553 if (write) 1553 if (write)
1554 h->max_huge_pages = set_max_huge_pages(h, tmp); 1554 h->max_huge_pages = set_max_huge_pages(h, tmp);
@@ -1557,10 +1557,10 @@ int hugetlb_sysctl_handler(struct ctl_table *table, int write,
1557} 1557}
1558 1558
1559int hugetlb_treat_movable_handler(struct ctl_table *table, int write, 1559int hugetlb_treat_movable_handler(struct ctl_table *table, int write,
1560 struct file *file, void __user *buffer, 1560 void __user *buffer,
1561 size_t *length, loff_t *ppos) 1561 size_t *length, loff_t *ppos)
1562{ 1562{
1563 proc_dointvec(table, write, file, buffer, length, ppos); 1563 proc_dointvec(table, write, buffer, length, ppos);
1564 if (hugepages_treat_as_movable) 1564 if (hugepages_treat_as_movable)
1565 htlb_alloc_mask = GFP_HIGHUSER_MOVABLE; 1565 htlb_alloc_mask = GFP_HIGHUSER_MOVABLE;
1566 else 1566 else
@@ -1569,7 +1569,7 @@ int hugetlb_treat_movable_handler(struct ctl_table *table, int write,
1569} 1569}
1570 1570
1571int hugetlb_overcommit_handler(struct ctl_table *table, int write, 1571int hugetlb_overcommit_handler(struct ctl_table *table, int write,
1572 struct file *file, void __user *buffer, 1572 void __user *buffer,
1573 size_t *length, loff_t *ppos) 1573 size_t *length, loff_t *ppos)
1574{ 1574{
1575 struct hstate *h = &default_hstate; 1575 struct hstate *h = &default_hstate;
@@ -1580,7 +1580,7 @@ int hugetlb_overcommit_handler(struct ctl_table *table, int write,
1580 1580
1581 table->data = &tmp; 1581 table->data = &tmp;
1582 table->maxlen = sizeof(unsigned long); 1582 table->maxlen = sizeof(unsigned long);
1583 proc_doulongvec_minmax(table, write, file, buffer, length, ppos); 1583 proc_doulongvec_minmax(table, write, buffer, length, ppos);
1584 1584
1585 if (write) { 1585 if (write) {
1586 spin_lock(&hugetlb_lock); 1586 spin_lock(&hugetlb_lock);
diff --git a/mm/hwpoison-inject.c b/mm/hwpoison-inject.c
new file mode 100644
index 000000000000..e1d85137f086
--- /dev/null
+++ b/mm/hwpoison-inject.c
@@ -0,0 +1,41 @@
1/* Inject a hwpoison memory failure on a arbitary pfn */
2#include <linux/module.h>
3#include <linux/debugfs.h>
4#include <linux/kernel.h>
5#include <linux/mm.h>
6
7static struct dentry *hwpoison_dir, *corrupt_pfn;
8
9static int hwpoison_inject(void *data, u64 val)
10{
11 if (!capable(CAP_SYS_ADMIN))
12 return -EPERM;
13 printk(KERN_INFO "Injecting memory failure at pfn %Lx\n", val);
14 return __memory_failure(val, 18, 0);
15}
16
17DEFINE_SIMPLE_ATTRIBUTE(hwpoison_fops, NULL, hwpoison_inject, "%lli\n");
18
19static void pfn_inject_exit(void)
20{
21 if (hwpoison_dir)
22 debugfs_remove_recursive(hwpoison_dir);
23}
24
25static int pfn_inject_init(void)
26{
27 hwpoison_dir = debugfs_create_dir("hwpoison", NULL);
28 if (hwpoison_dir == NULL)
29 return -ENOMEM;
30 corrupt_pfn = debugfs_create_file("corrupt-pfn", 0600, hwpoison_dir,
31 NULL, &hwpoison_fops);
32 if (corrupt_pfn == NULL) {
33 pfn_inject_exit();
34 return -ENOMEM;
35 }
36 return 0;
37}
38
39module_init(pfn_inject_init);
40module_exit(pfn_inject_exit);
41MODULE_LICENSE("GPL");
diff --git a/mm/ksm.c b/mm/ksm.c
index 37cc37325094..f7edac356f46 100644
--- a/mm/ksm.c
+++ b/mm/ksm.c
@@ -30,6 +30,7 @@
30#include <linux/slab.h> 30#include <linux/slab.h>
31#include <linux/rbtree.h> 31#include <linux/rbtree.h>
32#include <linux/mmu_notifier.h> 32#include <linux/mmu_notifier.h>
33#include <linux/swap.h>
33#include <linux/ksm.h> 34#include <linux/ksm.h>
34 35
35#include <asm/tlbflush.h> 36#include <asm/tlbflush.h>
@@ -162,10 +163,10 @@ static unsigned long ksm_pages_unshared;
162static unsigned long ksm_rmap_items; 163static unsigned long ksm_rmap_items;
163 164
164/* Limit on the number of unswappable pages used */ 165/* Limit on the number of unswappable pages used */
165static unsigned long ksm_max_kernel_pages = 2000; 166static unsigned long ksm_max_kernel_pages;
166 167
167/* Number of pages ksmd should scan in one batch */ 168/* Number of pages ksmd should scan in one batch */
168static unsigned int ksm_thread_pages_to_scan = 200; 169static unsigned int ksm_thread_pages_to_scan = 100;
169 170
170/* Milliseconds ksmd should sleep between batches */ 171/* Milliseconds ksmd should sleep between batches */
171static unsigned int ksm_thread_sleep_millisecs = 20; 172static unsigned int ksm_thread_sleep_millisecs = 20;
@@ -173,7 +174,7 @@ static unsigned int ksm_thread_sleep_millisecs = 20;
173#define KSM_RUN_STOP 0 174#define KSM_RUN_STOP 0
174#define KSM_RUN_MERGE 1 175#define KSM_RUN_MERGE 1
175#define KSM_RUN_UNMERGE 2 176#define KSM_RUN_UNMERGE 2
176static unsigned int ksm_run = KSM_RUN_MERGE; 177static unsigned int ksm_run = KSM_RUN_STOP;
177 178
178static DECLARE_WAIT_QUEUE_HEAD(ksm_thread_wait); 179static DECLARE_WAIT_QUEUE_HEAD(ksm_thread_wait);
179static DEFINE_MUTEX(ksm_thread_mutex); 180static DEFINE_MUTEX(ksm_thread_mutex);
@@ -183,6 +184,11 @@ static DEFINE_SPINLOCK(ksm_mmlist_lock);
183 sizeof(struct __struct), __alignof__(struct __struct),\ 184 sizeof(struct __struct), __alignof__(struct __struct),\
184 (__flags), NULL) 185 (__flags), NULL)
185 186
187static void __init ksm_init_max_kernel_pages(void)
188{
189 ksm_max_kernel_pages = nr_free_buffer_pages() / 4;
190}
191
186static int __init ksm_slab_init(void) 192static int __init ksm_slab_init(void)
187{ 193{
188 rmap_item_cache = KSM_KMEM_CACHE(rmap_item, 0); 194 rmap_item_cache = KSM_KMEM_CACHE(rmap_item, 0);
@@ -1667,6 +1673,8 @@ static int __init ksm_init(void)
1667 struct task_struct *ksm_thread; 1673 struct task_struct *ksm_thread;
1668 int err; 1674 int err;
1669 1675
1676 ksm_init_max_kernel_pages();
1677
1670 err = ksm_slab_init(); 1678 err = ksm_slab_init();
1671 if (err) 1679 if (err)
1672 goto out; 1680 goto out;
diff --git a/mm/madvise.c b/mm/madvise.c
index d9ae2067952e..35b1479b7c9d 100644
--- a/mm/madvise.c
+++ b/mm/madvise.c
@@ -218,6 +218,32 @@ static long madvise_remove(struct vm_area_struct *vma,
218 return error; 218 return error;
219} 219}
220 220
221#ifdef CONFIG_MEMORY_FAILURE
222/*
223 * Error injection support for memory error handling.
224 */
225static int madvise_hwpoison(unsigned long start, unsigned long end)
226{
227 int ret = 0;
228
229 if (!capable(CAP_SYS_ADMIN))
230 return -EPERM;
231 for (; start < end; start += PAGE_SIZE) {
232 struct page *p;
233 int ret = get_user_pages(current, current->mm, start, 1,
234 0, 0, &p, NULL);
235 if (ret != 1)
236 return ret;
237 printk(KERN_INFO "Injecting memory failure for page %lx at %lx\n",
238 page_to_pfn(p), start);
239 /* Ignore return value for now */
240 __memory_failure(page_to_pfn(p), 0, 1);
241 put_page(p);
242 }
243 return ret;
244}
245#endif
246
221static long 247static long
222madvise_vma(struct vm_area_struct *vma, struct vm_area_struct **prev, 248madvise_vma(struct vm_area_struct *vma, struct vm_area_struct **prev,
223 unsigned long start, unsigned long end, int behavior) 249 unsigned long start, unsigned long end, int behavior)
@@ -308,6 +334,10 @@ SYSCALL_DEFINE3(madvise, unsigned long, start, size_t, len_in, int, behavior)
308 int write; 334 int write;
309 size_t len; 335 size_t len;
310 336
337#ifdef CONFIG_MEMORY_FAILURE
338 if (behavior == MADV_HWPOISON)
339 return madvise_hwpoison(start, start+len_in);
340#endif
311 if (!madvise_behavior_valid(behavior)) 341 if (!madvise_behavior_valid(behavior))
312 return error; 342 return error;
313 343
diff --git a/mm/memcontrol.c b/mm/memcontrol.c
index 9b10d8753784..e2b98a6875c0 100644
--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c
@@ -29,6 +29,7 @@
29#include <linux/rcupdate.h> 29#include <linux/rcupdate.h>
30#include <linux/limits.h> 30#include <linux/limits.h>
31#include <linux/mutex.h> 31#include <linux/mutex.h>
32#include <linux/rbtree.h>
32#include <linux/slab.h> 33#include <linux/slab.h>
33#include <linux/swap.h> 34#include <linux/swap.h>
34#include <linux/spinlock.h> 35#include <linux/spinlock.h>
@@ -43,6 +44,7 @@
43 44
44struct cgroup_subsys mem_cgroup_subsys __read_mostly; 45struct cgroup_subsys mem_cgroup_subsys __read_mostly;
45#define MEM_CGROUP_RECLAIM_RETRIES 5 46#define MEM_CGROUP_RECLAIM_RETRIES 5
47struct mem_cgroup *root_mem_cgroup __read_mostly;
46 48
47#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP 49#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
48/* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */ 50/* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
@@ -53,6 +55,7 @@ static int really_do_swap_account __initdata = 1; /* for remember boot option*/
53#endif 55#endif
54 56
55static DEFINE_MUTEX(memcg_tasklist); /* can be hold under cgroup_mutex */ 57static DEFINE_MUTEX(memcg_tasklist); /* can be hold under cgroup_mutex */
58#define SOFTLIMIT_EVENTS_THRESH (1000)
56 59
57/* 60/*
58 * Statistics for memory cgroup. 61 * Statistics for memory cgroup.
@@ -66,6 +69,8 @@ enum mem_cgroup_stat_index {
66 MEM_CGROUP_STAT_MAPPED_FILE, /* # of pages charged as file rss */ 69 MEM_CGROUP_STAT_MAPPED_FILE, /* # of pages charged as file rss */
67 MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */ 70 MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */
68 MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */ 71 MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */
72 MEM_CGROUP_STAT_EVENTS, /* sum of pagein + pageout for internal use */
73 MEM_CGROUP_STAT_SWAPOUT, /* # of pages, swapped out */
69 74
70 MEM_CGROUP_STAT_NSTATS, 75 MEM_CGROUP_STAT_NSTATS,
71}; 76};
@@ -78,6 +83,20 @@ struct mem_cgroup_stat {
78 struct mem_cgroup_stat_cpu cpustat[0]; 83 struct mem_cgroup_stat_cpu cpustat[0];
79}; 84};
80 85
86static inline void
87__mem_cgroup_stat_reset_safe(struct mem_cgroup_stat_cpu *stat,
88 enum mem_cgroup_stat_index idx)
89{
90 stat->count[idx] = 0;
91}
92
93static inline s64
94__mem_cgroup_stat_read_local(struct mem_cgroup_stat_cpu *stat,
95 enum mem_cgroup_stat_index idx)
96{
97 return stat->count[idx];
98}
99
81/* 100/*
82 * For accounting under irq disable, no need for increment preempt count. 101 * For accounting under irq disable, no need for increment preempt count.
83 */ 102 */
@@ -117,6 +136,12 @@ struct mem_cgroup_per_zone {
117 unsigned long count[NR_LRU_LISTS]; 136 unsigned long count[NR_LRU_LISTS];
118 137
119 struct zone_reclaim_stat reclaim_stat; 138 struct zone_reclaim_stat reclaim_stat;
139 struct rb_node tree_node; /* RB tree node */
140 unsigned long long usage_in_excess;/* Set to the value by which */
141 /* the soft limit is exceeded*/
142 bool on_tree;
143 struct mem_cgroup *mem; /* Back pointer, we cannot */
144 /* use container_of */
120}; 145};
121/* Macro for accessing counter */ 146/* Macro for accessing counter */
122#define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)]) 147#define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
@@ -130,6 +155,26 @@ struct mem_cgroup_lru_info {
130}; 155};
131 156
132/* 157/*
158 * Cgroups above their limits are maintained in a RB-Tree, independent of
159 * their hierarchy representation
160 */
161
162struct mem_cgroup_tree_per_zone {
163 struct rb_root rb_root;
164 spinlock_t lock;
165};
166
167struct mem_cgroup_tree_per_node {
168 struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
169};
170
171struct mem_cgroup_tree {
172 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
173};
174
175static struct mem_cgroup_tree soft_limit_tree __read_mostly;
176
177/*
133 * The memory controller data structure. The memory controller controls both 178 * The memory controller data structure. The memory controller controls both
134 * page cache and RSS per cgroup. We would eventually like to provide 179 * page cache and RSS per cgroup. We would eventually like to provide
135 * statistics based on the statistics developed by Rik Van Riel for clock-pro, 180 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
@@ -186,6 +231,13 @@ struct mem_cgroup {
186 struct mem_cgroup_stat stat; 231 struct mem_cgroup_stat stat;
187}; 232};
188 233
234/*
235 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
236 * limit reclaim to prevent infinite loops, if they ever occur.
237 */
238#define MEM_CGROUP_MAX_RECLAIM_LOOPS (100)
239#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2)
240
189enum charge_type { 241enum charge_type {
190 MEM_CGROUP_CHARGE_TYPE_CACHE = 0, 242 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
191 MEM_CGROUP_CHARGE_TYPE_MAPPED, 243 MEM_CGROUP_CHARGE_TYPE_MAPPED,
@@ -200,13 +252,8 @@ enum charge_type {
200#define PCGF_CACHE (1UL << PCG_CACHE) 252#define PCGF_CACHE (1UL << PCG_CACHE)
201#define PCGF_USED (1UL << PCG_USED) 253#define PCGF_USED (1UL << PCG_USED)
202#define PCGF_LOCK (1UL << PCG_LOCK) 254#define PCGF_LOCK (1UL << PCG_LOCK)
203static const unsigned long 255/* Not used, but added here for completeness */
204pcg_default_flags[NR_CHARGE_TYPE] = { 256#define PCGF_ACCT (1UL << PCG_ACCT)
205 PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* File Cache */
206 PCGF_USED | PCGF_LOCK, /* Anon */
207 PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* Shmem */
208 0, /* FORCE */
209};
210 257
211/* for encoding cft->private value on file */ 258/* for encoding cft->private value on file */
212#define _MEM (0) 259#define _MEM (0)
@@ -215,15 +262,241 @@ pcg_default_flags[NR_CHARGE_TYPE] = {
215#define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff) 262#define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
216#define MEMFILE_ATTR(val) ((val) & 0xffff) 263#define MEMFILE_ATTR(val) ((val) & 0xffff)
217 264
265/*
266 * Reclaim flags for mem_cgroup_hierarchical_reclaim
267 */
268#define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
269#define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
270#define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
271#define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
272#define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2
273#define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT)
274
218static void mem_cgroup_get(struct mem_cgroup *mem); 275static void mem_cgroup_get(struct mem_cgroup *mem);
219static void mem_cgroup_put(struct mem_cgroup *mem); 276static void mem_cgroup_put(struct mem_cgroup *mem);
220static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem); 277static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem);
221 278
279static struct mem_cgroup_per_zone *
280mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid)
281{
282 return &mem->info.nodeinfo[nid]->zoneinfo[zid];
283}
284
285static struct mem_cgroup_per_zone *
286page_cgroup_zoneinfo(struct page_cgroup *pc)
287{
288 struct mem_cgroup *mem = pc->mem_cgroup;
289 int nid = page_cgroup_nid(pc);
290 int zid = page_cgroup_zid(pc);
291
292 if (!mem)
293 return NULL;
294
295 return mem_cgroup_zoneinfo(mem, nid, zid);
296}
297
298static struct mem_cgroup_tree_per_zone *
299soft_limit_tree_node_zone(int nid, int zid)
300{
301 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
302}
303
304static struct mem_cgroup_tree_per_zone *
305soft_limit_tree_from_page(struct page *page)
306{
307 int nid = page_to_nid(page);
308 int zid = page_zonenum(page);
309
310 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
311}
312
313static void
314__mem_cgroup_insert_exceeded(struct mem_cgroup *mem,
315 struct mem_cgroup_per_zone *mz,
316 struct mem_cgroup_tree_per_zone *mctz)
317{
318 struct rb_node **p = &mctz->rb_root.rb_node;
319 struct rb_node *parent = NULL;
320 struct mem_cgroup_per_zone *mz_node;
321
322 if (mz->on_tree)
323 return;
324
325 mz->usage_in_excess = res_counter_soft_limit_excess(&mem->res);
326 while (*p) {
327 parent = *p;
328 mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
329 tree_node);
330 if (mz->usage_in_excess < mz_node->usage_in_excess)
331 p = &(*p)->rb_left;
332 /*
333 * We can't avoid mem cgroups that are over their soft
334 * limit by the same amount
335 */
336 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
337 p = &(*p)->rb_right;
338 }
339 rb_link_node(&mz->tree_node, parent, p);
340 rb_insert_color(&mz->tree_node, &mctz->rb_root);
341 mz->on_tree = true;
342}
343
344static void
345__mem_cgroup_remove_exceeded(struct mem_cgroup *mem,
346 struct mem_cgroup_per_zone *mz,
347 struct mem_cgroup_tree_per_zone *mctz)
348{
349 if (!mz->on_tree)
350 return;
351 rb_erase(&mz->tree_node, &mctz->rb_root);
352 mz->on_tree = false;
353}
354
355static void
356mem_cgroup_insert_exceeded(struct mem_cgroup *mem,
357 struct mem_cgroup_per_zone *mz,
358 struct mem_cgroup_tree_per_zone *mctz)
359{
360 spin_lock(&mctz->lock);
361 __mem_cgroup_insert_exceeded(mem, mz, mctz);
362 spin_unlock(&mctz->lock);
363}
364
365static void
366mem_cgroup_remove_exceeded(struct mem_cgroup *mem,
367 struct mem_cgroup_per_zone *mz,
368 struct mem_cgroup_tree_per_zone *mctz)
369{
370 spin_lock(&mctz->lock);
371 __mem_cgroup_remove_exceeded(mem, mz, mctz);
372 spin_unlock(&mctz->lock);
373}
374
375static bool mem_cgroup_soft_limit_check(struct mem_cgroup *mem)
376{
377 bool ret = false;
378 int cpu;
379 s64 val;
380 struct mem_cgroup_stat_cpu *cpustat;
381
382 cpu = get_cpu();
383 cpustat = &mem->stat.cpustat[cpu];
384 val = __mem_cgroup_stat_read_local(cpustat, MEM_CGROUP_STAT_EVENTS);
385 if (unlikely(val > SOFTLIMIT_EVENTS_THRESH)) {
386 __mem_cgroup_stat_reset_safe(cpustat, MEM_CGROUP_STAT_EVENTS);
387 ret = true;
388 }
389 put_cpu();
390 return ret;
391}
392
393static void mem_cgroup_update_tree(struct mem_cgroup *mem, struct page *page)
394{
395 unsigned long long prev_usage_in_excess, new_usage_in_excess;
396 bool updated_tree = false;
397 struct mem_cgroup_per_zone *mz;
398 struct mem_cgroup_tree_per_zone *mctz;
399
400 mz = mem_cgroup_zoneinfo(mem, page_to_nid(page), page_zonenum(page));
401 mctz = soft_limit_tree_from_page(page);
402
403 /*
404 * We do updates in lazy mode, mem's are removed
405 * lazily from the per-zone, per-node rb tree
406 */
407 prev_usage_in_excess = mz->usage_in_excess;
408
409 new_usage_in_excess = res_counter_soft_limit_excess(&mem->res);
410 if (prev_usage_in_excess) {
411 mem_cgroup_remove_exceeded(mem, mz, mctz);
412 updated_tree = true;
413 }
414 if (!new_usage_in_excess)
415 goto done;
416 mem_cgroup_insert_exceeded(mem, mz, mctz);
417
418done:
419 if (updated_tree) {
420 spin_lock(&mctz->lock);
421 mz->usage_in_excess = new_usage_in_excess;
422 spin_unlock(&mctz->lock);
423 }
424}
425
426static void mem_cgroup_remove_from_trees(struct mem_cgroup *mem)
427{
428 int node, zone;
429 struct mem_cgroup_per_zone *mz;
430 struct mem_cgroup_tree_per_zone *mctz;
431
432 for_each_node_state(node, N_POSSIBLE) {
433 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
434 mz = mem_cgroup_zoneinfo(mem, node, zone);
435 mctz = soft_limit_tree_node_zone(node, zone);
436 mem_cgroup_remove_exceeded(mem, mz, mctz);
437 }
438 }
439}
440
441static inline unsigned long mem_cgroup_get_excess(struct mem_cgroup *mem)
442{
443 return res_counter_soft_limit_excess(&mem->res) >> PAGE_SHIFT;
444}
445
446static struct mem_cgroup_per_zone *
447__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
448{
449 struct rb_node *rightmost = NULL;
450 struct mem_cgroup_per_zone *mz = NULL;
451
452retry:
453 rightmost = rb_last(&mctz->rb_root);
454 if (!rightmost)
455 goto done; /* Nothing to reclaim from */
456
457 mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
458 /*
459 * Remove the node now but someone else can add it back,
460 * we will to add it back at the end of reclaim to its correct
461 * position in the tree.
462 */
463 __mem_cgroup_remove_exceeded(mz->mem, mz, mctz);
464 if (!res_counter_soft_limit_excess(&mz->mem->res) ||
465 !css_tryget(&mz->mem->css))
466 goto retry;
467done:
468 return mz;
469}
470
471static struct mem_cgroup_per_zone *
472mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
473{
474 struct mem_cgroup_per_zone *mz;
475
476 spin_lock(&mctz->lock);
477 mz = __mem_cgroup_largest_soft_limit_node(mctz);
478 spin_unlock(&mctz->lock);
479 return mz;
480}
481
482static void mem_cgroup_swap_statistics(struct mem_cgroup *mem,
483 bool charge)
484{
485 int val = (charge) ? 1 : -1;
486 struct mem_cgroup_stat *stat = &mem->stat;
487 struct mem_cgroup_stat_cpu *cpustat;
488 int cpu = get_cpu();
489
490 cpustat = &stat->cpustat[cpu];
491 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_SWAPOUT, val);
492 put_cpu();
493}
494
222static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, 495static void mem_cgroup_charge_statistics(struct mem_cgroup *mem,
223 struct page_cgroup *pc, 496 struct page_cgroup *pc,
224 bool charge) 497 bool charge)
225{ 498{
226 int val = (charge)? 1 : -1; 499 int val = (charge) ? 1 : -1;
227 struct mem_cgroup_stat *stat = &mem->stat; 500 struct mem_cgroup_stat *stat = &mem->stat;
228 struct mem_cgroup_stat_cpu *cpustat; 501 struct mem_cgroup_stat_cpu *cpustat;
229 int cpu = get_cpu(); 502 int cpu = get_cpu();
@@ -240,28 +513,10 @@ static void mem_cgroup_charge_statistics(struct mem_cgroup *mem,
240 else 513 else
241 __mem_cgroup_stat_add_safe(cpustat, 514 __mem_cgroup_stat_add_safe(cpustat,
242 MEM_CGROUP_STAT_PGPGOUT_COUNT, 1); 515 MEM_CGROUP_STAT_PGPGOUT_COUNT, 1);
516 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_EVENTS, 1);
243 put_cpu(); 517 put_cpu();
244} 518}
245 519
246static struct mem_cgroup_per_zone *
247mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid)
248{
249 return &mem->info.nodeinfo[nid]->zoneinfo[zid];
250}
251
252static struct mem_cgroup_per_zone *
253page_cgroup_zoneinfo(struct page_cgroup *pc)
254{
255 struct mem_cgroup *mem = pc->mem_cgroup;
256 int nid = page_cgroup_nid(pc);
257 int zid = page_cgroup_zid(pc);
258
259 if (!mem)
260 return NULL;
261
262 return mem_cgroup_zoneinfo(mem, nid, zid);
263}
264
265static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem, 520static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem,
266 enum lru_list idx) 521 enum lru_list idx)
267{ 522{
@@ -354,6 +609,11 @@ static int mem_cgroup_walk_tree(struct mem_cgroup *root, void *data,
354 return ret; 609 return ret;
355} 610}
356 611
612static inline bool mem_cgroup_is_root(struct mem_cgroup *mem)
613{
614 return (mem == root_mem_cgroup);
615}
616
357/* 617/*
358 * Following LRU functions are allowed to be used without PCG_LOCK. 618 * Following LRU functions are allowed to be used without PCG_LOCK.
359 * Operations are called by routine of global LRU independently from memcg. 619 * Operations are called by routine of global LRU independently from memcg.
@@ -371,22 +631,24 @@ static int mem_cgroup_walk_tree(struct mem_cgroup *root, void *data,
371void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru) 631void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru)
372{ 632{
373 struct page_cgroup *pc; 633 struct page_cgroup *pc;
374 struct mem_cgroup *mem;
375 struct mem_cgroup_per_zone *mz; 634 struct mem_cgroup_per_zone *mz;
376 635
377 if (mem_cgroup_disabled()) 636 if (mem_cgroup_disabled())
378 return; 637 return;
379 pc = lookup_page_cgroup(page); 638 pc = lookup_page_cgroup(page);
380 /* can happen while we handle swapcache. */ 639 /* can happen while we handle swapcache. */
381 if (list_empty(&pc->lru) || !pc->mem_cgroup) 640 if (!TestClearPageCgroupAcctLRU(pc))
382 return; 641 return;
642 VM_BUG_ON(!pc->mem_cgroup);
383 /* 643 /*
384 * We don't check PCG_USED bit. It's cleared when the "page" is finally 644 * We don't check PCG_USED bit. It's cleared when the "page" is finally
385 * removed from global LRU. 645 * removed from global LRU.
386 */ 646 */
387 mz = page_cgroup_zoneinfo(pc); 647 mz = page_cgroup_zoneinfo(pc);
388 mem = pc->mem_cgroup;
389 MEM_CGROUP_ZSTAT(mz, lru) -= 1; 648 MEM_CGROUP_ZSTAT(mz, lru) -= 1;
649 if (mem_cgroup_is_root(pc->mem_cgroup))
650 return;
651 VM_BUG_ON(list_empty(&pc->lru));
390 list_del_init(&pc->lru); 652 list_del_init(&pc->lru);
391 return; 653 return;
392} 654}
@@ -410,8 +672,8 @@ void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru)
410 * For making pc->mem_cgroup visible, insert smp_rmb() here. 672 * For making pc->mem_cgroup visible, insert smp_rmb() here.
411 */ 673 */
412 smp_rmb(); 674 smp_rmb();
413 /* unused page is not rotated. */ 675 /* unused or root page is not rotated. */
414 if (!PageCgroupUsed(pc)) 676 if (!PageCgroupUsed(pc) || mem_cgroup_is_root(pc->mem_cgroup))
415 return; 677 return;
416 mz = page_cgroup_zoneinfo(pc); 678 mz = page_cgroup_zoneinfo(pc);
417 list_move(&pc->lru, &mz->lists[lru]); 679 list_move(&pc->lru, &mz->lists[lru]);
@@ -425,6 +687,7 @@ void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru)
425 if (mem_cgroup_disabled()) 687 if (mem_cgroup_disabled())
426 return; 688 return;
427 pc = lookup_page_cgroup(page); 689 pc = lookup_page_cgroup(page);
690 VM_BUG_ON(PageCgroupAcctLRU(pc));
428 /* 691 /*
429 * Used bit is set without atomic ops but after smp_wmb(). 692 * Used bit is set without atomic ops but after smp_wmb().
430 * For making pc->mem_cgroup visible, insert smp_rmb() here. 693 * For making pc->mem_cgroup visible, insert smp_rmb() here.
@@ -435,6 +698,9 @@ void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru)
435 698
436 mz = page_cgroup_zoneinfo(pc); 699 mz = page_cgroup_zoneinfo(pc);
437 MEM_CGROUP_ZSTAT(mz, lru) += 1; 700 MEM_CGROUP_ZSTAT(mz, lru) += 1;
701 SetPageCgroupAcctLRU(pc);
702 if (mem_cgroup_is_root(pc->mem_cgroup))
703 return;
438 list_add(&pc->lru, &mz->lists[lru]); 704 list_add(&pc->lru, &mz->lists[lru]);
439} 705}
440 706
@@ -469,7 +735,7 @@ static void mem_cgroup_lru_add_after_commit_swapcache(struct page *page)
469 735
470 spin_lock_irqsave(&zone->lru_lock, flags); 736 spin_lock_irqsave(&zone->lru_lock, flags);
471 /* link when the page is linked to LRU but page_cgroup isn't */ 737 /* link when the page is linked to LRU but page_cgroup isn't */
472 if (PageLRU(page) && list_empty(&pc->lru)) 738 if (PageLRU(page) && !PageCgroupAcctLRU(pc))
473 mem_cgroup_add_lru_list(page, page_lru(page)); 739 mem_cgroup_add_lru_list(page, page_lru(page));
474 spin_unlock_irqrestore(&zone->lru_lock, flags); 740 spin_unlock_irqrestore(&zone->lru_lock, flags);
475} 741}
@@ -855,28 +1121,62 @@ mem_cgroup_select_victim(struct mem_cgroup *root_mem)
855 * If shrink==true, for avoiding to free too much, this returns immedieately. 1121 * If shrink==true, for avoiding to free too much, this returns immedieately.
856 */ 1122 */
857static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem, 1123static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem,
858 gfp_t gfp_mask, bool noswap, bool shrink) 1124 struct zone *zone,
1125 gfp_t gfp_mask,
1126 unsigned long reclaim_options)
859{ 1127{
860 struct mem_cgroup *victim; 1128 struct mem_cgroup *victim;
861 int ret, total = 0; 1129 int ret, total = 0;
862 int loop = 0; 1130 int loop = 0;
1131 bool noswap = reclaim_options & MEM_CGROUP_RECLAIM_NOSWAP;
1132 bool shrink = reclaim_options & MEM_CGROUP_RECLAIM_SHRINK;
1133 bool check_soft = reclaim_options & MEM_CGROUP_RECLAIM_SOFT;
1134 unsigned long excess = mem_cgroup_get_excess(root_mem);
863 1135
864 /* If memsw_is_minimum==1, swap-out is of-no-use. */ 1136 /* If memsw_is_minimum==1, swap-out is of-no-use. */
865 if (root_mem->memsw_is_minimum) 1137 if (root_mem->memsw_is_minimum)
866 noswap = true; 1138 noswap = true;
867 1139
868 while (loop < 2) { 1140 while (1) {
869 victim = mem_cgroup_select_victim(root_mem); 1141 victim = mem_cgroup_select_victim(root_mem);
870 if (victim == root_mem) 1142 if (victim == root_mem) {
871 loop++; 1143 loop++;
1144 if (loop >= 2) {
1145 /*
1146 * If we have not been able to reclaim
1147 * anything, it might because there are
1148 * no reclaimable pages under this hierarchy
1149 */
1150 if (!check_soft || !total) {
1151 css_put(&victim->css);
1152 break;
1153 }
1154 /*
1155 * We want to do more targetted reclaim.
1156 * excess >> 2 is not to excessive so as to
1157 * reclaim too much, nor too less that we keep
1158 * coming back to reclaim from this cgroup
1159 */
1160 if (total >= (excess >> 2) ||
1161 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) {
1162 css_put(&victim->css);
1163 break;
1164 }
1165 }
1166 }
872 if (!mem_cgroup_local_usage(&victim->stat)) { 1167 if (!mem_cgroup_local_usage(&victim->stat)) {
873 /* this cgroup's local usage == 0 */ 1168 /* this cgroup's local usage == 0 */
874 css_put(&victim->css); 1169 css_put(&victim->css);
875 continue; 1170 continue;
876 } 1171 }
877 /* we use swappiness of local cgroup */ 1172 /* we use swappiness of local cgroup */
878 ret = try_to_free_mem_cgroup_pages(victim, gfp_mask, noswap, 1173 if (check_soft)
879 get_swappiness(victim)); 1174 ret = mem_cgroup_shrink_node_zone(victim, gfp_mask,
1175 noswap, get_swappiness(victim), zone,
1176 zone->zone_pgdat->node_id);
1177 else
1178 ret = try_to_free_mem_cgroup_pages(victim, gfp_mask,
1179 noswap, get_swappiness(victim));
880 css_put(&victim->css); 1180 css_put(&victim->css);
881 /* 1181 /*
882 * At shrinking usage, we can't check we should stop here or 1182 * At shrinking usage, we can't check we should stop here or
@@ -886,7 +1186,10 @@ static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem,
886 if (shrink) 1186 if (shrink)
887 return ret; 1187 return ret;
888 total += ret; 1188 total += ret;
889 if (mem_cgroup_check_under_limit(root_mem)) 1189 if (check_soft) {
1190 if (res_counter_check_under_soft_limit(&root_mem->res))
1191 return total;
1192 } else if (mem_cgroup_check_under_limit(root_mem))
890 return 1 + total; 1193 return 1 + total;
891 } 1194 }
892 return total; 1195 return total;
@@ -965,11 +1268,11 @@ done:
965 */ 1268 */
966static int __mem_cgroup_try_charge(struct mm_struct *mm, 1269static int __mem_cgroup_try_charge(struct mm_struct *mm,
967 gfp_t gfp_mask, struct mem_cgroup **memcg, 1270 gfp_t gfp_mask, struct mem_cgroup **memcg,
968 bool oom) 1271 bool oom, struct page *page)
969{ 1272{
970 struct mem_cgroup *mem, *mem_over_limit; 1273 struct mem_cgroup *mem, *mem_over_limit, *mem_over_soft_limit;
971 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; 1274 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
972 struct res_counter *fail_res; 1275 struct res_counter *fail_res, *soft_fail_res = NULL;
973 1276
974 if (unlikely(test_thread_flag(TIF_MEMDIE))) { 1277 if (unlikely(test_thread_flag(TIF_MEMDIE))) {
975 /* Don't account this! */ 1278 /* Don't account this! */
@@ -996,20 +1299,23 @@ static int __mem_cgroup_try_charge(struct mm_struct *mm,
996 VM_BUG_ON(css_is_removed(&mem->css)); 1299 VM_BUG_ON(css_is_removed(&mem->css));
997 1300
998 while (1) { 1301 while (1) {
999 int ret; 1302 int ret = 0;
1000 bool noswap = false; 1303 unsigned long flags = 0;
1001 1304
1002 ret = res_counter_charge(&mem->res, PAGE_SIZE, &fail_res); 1305 if (mem_cgroup_is_root(mem))
1306 goto done;
1307 ret = res_counter_charge(&mem->res, PAGE_SIZE, &fail_res,
1308 &soft_fail_res);
1003 if (likely(!ret)) { 1309 if (likely(!ret)) {
1004 if (!do_swap_account) 1310 if (!do_swap_account)
1005 break; 1311 break;
1006 ret = res_counter_charge(&mem->memsw, PAGE_SIZE, 1312 ret = res_counter_charge(&mem->memsw, PAGE_SIZE,
1007 &fail_res); 1313 &fail_res, NULL);
1008 if (likely(!ret)) 1314 if (likely(!ret))
1009 break; 1315 break;
1010 /* mem+swap counter fails */ 1316 /* mem+swap counter fails */
1011 res_counter_uncharge(&mem->res, PAGE_SIZE); 1317 res_counter_uncharge(&mem->res, PAGE_SIZE, NULL);
1012 noswap = true; 1318 flags |= MEM_CGROUP_RECLAIM_NOSWAP;
1013 mem_over_limit = mem_cgroup_from_res_counter(fail_res, 1319 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
1014 memsw); 1320 memsw);
1015 } else 1321 } else
@@ -1020,8 +1326,8 @@ static int __mem_cgroup_try_charge(struct mm_struct *mm,
1020 if (!(gfp_mask & __GFP_WAIT)) 1326 if (!(gfp_mask & __GFP_WAIT))
1021 goto nomem; 1327 goto nomem;
1022 1328
1023 ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, gfp_mask, 1329 ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, NULL,
1024 noswap, false); 1330 gfp_mask, flags);
1025 if (ret) 1331 if (ret)
1026 continue; 1332 continue;
1027 1333
@@ -1046,13 +1352,24 @@ static int __mem_cgroup_try_charge(struct mm_struct *mm,
1046 goto nomem; 1352 goto nomem;
1047 } 1353 }
1048 } 1354 }
1355 /*
1356 * Insert just the ancestor, we should trickle down to the correct
1357 * cgroup for reclaim, since the other nodes will be below their
1358 * soft limit
1359 */
1360 if (soft_fail_res) {
1361 mem_over_soft_limit =
1362 mem_cgroup_from_res_counter(soft_fail_res, res);
1363 if (mem_cgroup_soft_limit_check(mem_over_soft_limit))
1364 mem_cgroup_update_tree(mem_over_soft_limit, page);
1365 }
1366done:
1049 return 0; 1367 return 0;
1050nomem: 1368nomem:
1051 css_put(&mem->css); 1369 css_put(&mem->css);
1052 return -ENOMEM; 1370 return -ENOMEM;
1053} 1371}
1054 1372
1055
1056/* 1373/*
1057 * A helper function to get mem_cgroup from ID. must be called under 1374 * A helper function to get mem_cgroup from ID. must be called under
1058 * rcu_read_lock(). The caller must check css_is_removed() or some if 1375 * rcu_read_lock(). The caller must check css_is_removed() or some if
@@ -1119,15 +1436,38 @@ static void __mem_cgroup_commit_charge(struct mem_cgroup *mem,
1119 lock_page_cgroup(pc); 1436 lock_page_cgroup(pc);
1120 if (unlikely(PageCgroupUsed(pc))) { 1437 if (unlikely(PageCgroupUsed(pc))) {
1121 unlock_page_cgroup(pc); 1438 unlock_page_cgroup(pc);
1122 res_counter_uncharge(&mem->res, PAGE_SIZE); 1439 if (!mem_cgroup_is_root(mem)) {
1123 if (do_swap_account) 1440 res_counter_uncharge(&mem->res, PAGE_SIZE, NULL);
1124 res_counter_uncharge(&mem->memsw, PAGE_SIZE); 1441 if (do_swap_account)
1442 res_counter_uncharge(&mem->memsw, PAGE_SIZE,
1443 NULL);
1444 }
1125 css_put(&mem->css); 1445 css_put(&mem->css);
1126 return; 1446 return;
1127 } 1447 }
1448
1128 pc->mem_cgroup = mem; 1449 pc->mem_cgroup = mem;
1450 /*
1451 * We access a page_cgroup asynchronously without lock_page_cgroup().
1452 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
1453 * is accessed after testing USED bit. To make pc->mem_cgroup visible
1454 * before USED bit, we need memory barrier here.
1455 * See mem_cgroup_add_lru_list(), etc.
1456 */
1129 smp_wmb(); 1457 smp_wmb();
1130 pc->flags = pcg_default_flags[ctype]; 1458 switch (ctype) {
1459 case MEM_CGROUP_CHARGE_TYPE_CACHE:
1460 case MEM_CGROUP_CHARGE_TYPE_SHMEM:
1461 SetPageCgroupCache(pc);
1462 SetPageCgroupUsed(pc);
1463 break;
1464 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
1465 ClearPageCgroupCache(pc);
1466 SetPageCgroupUsed(pc);
1467 break;
1468 default:
1469 break;
1470 }
1131 1471
1132 mem_cgroup_charge_statistics(mem, pc, true); 1472 mem_cgroup_charge_statistics(mem, pc, true);
1133 1473
@@ -1178,7 +1518,8 @@ static int mem_cgroup_move_account(struct page_cgroup *pc,
1178 if (pc->mem_cgroup != from) 1518 if (pc->mem_cgroup != from)
1179 goto out; 1519 goto out;
1180 1520
1181 res_counter_uncharge(&from->res, PAGE_SIZE); 1521 if (!mem_cgroup_is_root(from))
1522 res_counter_uncharge(&from->res, PAGE_SIZE, NULL);
1182 mem_cgroup_charge_statistics(from, pc, false); 1523 mem_cgroup_charge_statistics(from, pc, false);
1183 1524
1184 page = pc->page; 1525 page = pc->page;
@@ -1197,8 +1538,8 @@ static int mem_cgroup_move_account(struct page_cgroup *pc,
1197 1); 1538 1);
1198 } 1539 }
1199 1540
1200 if (do_swap_account) 1541 if (do_swap_account && !mem_cgroup_is_root(from))
1201 res_counter_uncharge(&from->memsw, PAGE_SIZE); 1542 res_counter_uncharge(&from->memsw, PAGE_SIZE, NULL);
1202 css_put(&from->css); 1543 css_put(&from->css);
1203 1544
1204 css_get(&to->css); 1545 css_get(&to->css);
@@ -1238,7 +1579,7 @@ static int mem_cgroup_move_parent(struct page_cgroup *pc,
1238 parent = mem_cgroup_from_cont(pcg); 1579 parent = mem_cgroup_from_cont(pcg);
1239 1580
1240 1581
1241 ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false); 1582 ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false, page);
1242 if (ret || !parent) 1583 if (ret || !parent)
1243 return ret; 1584 return ret;
1244 1585
@@ -1268,9 +1609,11 @@ uncharge:
1268 /* drop extra refcnt by try_charge() */ 1609 /* drop extra refcnt by try_charge() */
1269 css_put(&parent->css); 1610 css_put(&parent->css);
1270 /* uncharge if move fails */ 1611 /* uncharge if move fails */
1271 res_counter_uncharge(&parent->res, PAGE_SIZE); 1612 if (!mem_cgroup_is_root(parent)) {
1272 if (do_swap_account) 1613 res_counter_uncharge(&parent->res, PAGE_SIZE, NULL);
1273 res_counter_uncharge(&parent->memsw, PAGE_SIZE); 1614 if (do_swap_account)
1615 res_counter_uncharge(&parent->memsw, PAGE_SIZE, NULL);
1616 }
1274 return ret; 1617 return ret;
1275} 1618}
1276 1619
@@ -1295,7 +1638,7 @@ static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
1295 prefetchw(pc); 1638 prefetchw(pc);
1296 1639
1297 mem = memcg; 1640 mem = memcg;
1298 ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true); 1641 ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true, page);
1299 if (ret || !mem) 1642 if (ret || !mem)
1300 return ret; 1643 return ret;
1301 1644
@@ -1414,14 +1757,14 @@ int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
1414 if (!mem) 1757 if (!mem)
1415 goto charge_cur_mm; 1758 goto charge_cur_mm;
1416 *ptr = mem; 1759 *ptr = mem;
1417 ret = __mem_cgroup_try_charge(NULL, mask, ptr, true); 1760 ret = __mem_cgroup_try_charge(NULL, mask, ptr, true, page);
1418 /* drop extra refcnt from tryget */ 1761 /* drop extra refcnt from tryget */
1419 css_put(&mem->css); 1762 css_put(&mem->css);
1420 return ret; 1763 return ret;
1421charge_cur_mm: 1764charge_cur_mm:
1422 if (unlikely(!mm)) 1765 if (unlikely(!mm))
1423 mm = &init_mm; 1766 mm = &init_mm;
1424 return __mem_cgroup_try_charge(mm, mask, ptr, true); 1767 return __mem_cgroup_try_charge(mm, mask, ptr, true, page);
1425} 1768}
1426 1769
1427static void 1770static void
@@ -1459,7 +1802,10 @@ __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
1459 * This recorded memcg can be obsolete one. So, avoid 1802 * This recorded memcg can be obsolete one. So, avoid
1460 * calling css_tryget 1803 * calling css_tryget
1461 */ 1804 */
1462 res_counter_uncharge(&memcg->memsw, PAGE_SIZE); 1805 if (!mem_cgroup_is_root(memcg))
1806 res_counter_uncharge(&memcg->memsw, PAGE_SIZE,
1807 NULL);
1808 mem_cgroup_swap_statistics(memcg, false);
1463 mem_cgroup_put(memcg); 1809 mem_cgroup_put(memcg);
1464 } 1810 }
1465 rcu_read_unlock(); 1811 rcu_read_unlock();
@@ -1484,9 +1830,11 @@ void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem)
1484 return; 1830 return;
1485 if (!mem) 1831 if (!mem)
1486 return; 1832 return;
1487 res_counter_uncharge(&mem->res, PAGE_SIZE); 1833 if (!mem_cgroup_is_root(mem)) {
1488 if (do_swap_account) 1834 res_counter_uncharge(&mem->res, PAGE_SIZE, NULL);
1489 res_counter_uncharge(&mem->memsw, PAGE_SIZE); 1835 if (do_swap_account)
1836 res_counter_uncharge(&mem->memsw, PAGE_SIZE, NULL);
1837 }
1490 css_put(&mem->css); 1838 css_put(&mem->css);
1491} 1839}
1492 1840
@@ -1500,6 +1848,7 @@ __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
1500 struct page_cgroup *pc; 1848 struct page_cgroup *pc;
1501 struct mem_cgroup *mem = NULL; 1849 struct mem_cgroup *mem = NULL;
1502 struct mem_cgroup_per_zone *mz; 1850 struct mem_cgroup_per_zone *mz;
1851 bool soft_limit_excess = false;
1503 1852
1504 if (mem_cgroup_disabled()) 1853 if (mem_cgroup_disabled())
1505 return NULL; 1854 return NULL;
@@ -1538,9 +1887,14 @@ __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
1538 break; 1887 break;
1539 } 1888 }
1540 1889
1541 res_counter_uncharge(&mem->res, PAGE_SIZE); 1890 if (!mem_cgroup_is_root(mem)) {
1542 if (do_swap_account && (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT)) 1891 res_counter_uncharge(&mem->res, PAGE_SIZE, &soft_limit_excess);
1543 res_counter_uncharge(&mem->memsw, PAGE_SIZE); 1892 if (do_swap_account &&
1893 (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT))
1894 res_counter_uncharge(&mem->memsw, PAGE_SIZE, NULL);
1895 }
1896 if (ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
1897 mem_cgroup_swap_statistics(mem, true);
1544 mem_cgroup_charge_statistics(mem, pc, false); 1898 mem_cgroup_charge_statistics(mem, pc, false);
1545 1899
1546 ClearPageCgroupUsed(pc); 1900 ClearPageCgroupUsed(pc);
@@ -1554,6 +1908,8 @@ __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
1554 mz = page_cgroup_zoneinfo(pc); 1908 mz = page_cgroup_zoneinfo(pc);
1555 unlock_page_cgroup(pc); 1909 unlock_page_cgroup(pc);
1556 1910
1911 if (soft_limit_excess && mem_cgroup_soft_limit_check(mem))
1912 mem_cgroup_update_tree(mem, page);
1557 /* at swapout, this memcg will be accessed to record to swap */ 1913 /* at swapout, this memcg will be accessed to record to swap */
1558 if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT) 1914 if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
1559 css_put(&mem->css); 1915 css_put(&mem->css);
@@ -1629,7 +1985,9 @@ void mem_cgroup_uncharge_swap(swp_entry_t ent)
1629 * We uncharge this because swap is freed. 1985 * We uncharge this because swap is freed.
1630 * This memcg can be obsolete one. We avoid calling css_tryget 1986 * This memcg can be obsolete one. We avoid calling css_tryget
1631 */ 1987 */
1632 res_counter_uncharge(&memcg->memsw, PAGE_SIZE); 1988 if (!mem_cgroup_is_root(memcg))
1989 res_counter_uncharge(&memcg->memsw, PAGE_SIZE, NULL);
1990 mem_cgroup_swap_statistics(memcg, false);
1633 mem_cgroup_put(memcg); 1991 mem_cgroup_put(memcg);
1634 } 1992 }
1635 rcu_read_unlock(); 1993 rcu_read_unlock();
@@ -1658,7 +2016,8 @@ int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr)
1658 unlock_page_cgroup(pc); 2016 unlock_page_cgroup(pc);
1659 2017
1660 if (mem) { 2018 if (mem) {
1661 ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false); 2019 ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false,
2020 page);
1662 css_put(&mem->css); 2021 css_put(&mem->css);
1663 } 2022 }
1664 *ptr = mem; 2023 *ptr = mem;
@@ -1798,8 +2157,9 @@ static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
1798 if (!ret) 2157 if (!ret)
1799 break; 2158 break;
1800 2159
1801 progress = mem_cgroup_hierarchical_reclaim(memcg, GFP_KERNEL, 2160 progress = mem_cgroup_hierarchical_reclaim(memcg, NULL,
1802 false, true); 2161 GFP_KERNEL,
2162 MEM_CGROUP_RECLAIM_SHRINK);
1803 curusage = res_counter_read_u64(&memcg->res, RES_USAGE); 2163 curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
1804 /* Usage is reduced ? */ 2164 /* Usage is reduced ? */
1805 if (curusage >= oldusage) 2165 if (curusage >= oldusage)
@@ -1851,7 +2211,9 @@ static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
1851 if (!ret) 2211 if (!ret)
1852 break; 2212 break;
1853 2213
1854 mem_cgroup_hierarchical_reclaim(memcg, GFP_KERNEL, true, true); 2214 mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL,
2215 MEM_CGROUP_RECLAIM_NOSWAP |
2216 MEM_CGROUP_RECLAIM_SHRINK);
1855 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); 2217 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
1856 /* Usage is reduced ? */ 2218 /* Usage is reduced ? */
1857 if (curusage >= oldusage) 2219 if (curusage >= oldusage)
@@ -1862,6 +2224,97 @@ static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
1862 return ret; 2224 return ret;
1863} 2225}
1864 2226
2227unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
2228 gfp_t gfp_mask, int nid,
2229 int zid)
2230{
2231 unsigned long nr_reclaimed = 0;
2232 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
2233 unsigned long reclaimed;
2234 int loop = 0;
2235 struct mem_cgroup_tree_per_zone *mctz;
2236
2237 if (order > 0)
2238 return 0;
2239
2240 mctz = soft_limit_tree_node_zone(nid, zid);
2241 /*
2242 * This loop can run a while, specially if mem_cgroup's continuously
2243 * keep exceeding their soft limit and putting the system under
2244 * pressure
2245 */
2246 do {
2247 if (next_mz)
2248 mz = next_mz;
2249 else
2250 mz = mem_cgroup_largest_soft_limit_node(mctz);
2251 if (!mz)
2252 break;
2253
2254 reclaimed = mem_cgroup_hierarchical_reclaim(mz->mem, zone,
2255 gfp_mask,
2256 MEM_CGROUP_RECLAIM_SOFT);
2257 nr_reclaimed += reclaimed;
2258 spin_lock(&mctz->lock);
2259
2260 /*
2261 * If we failed to reclaim anything from this memory cgroup
2262 * it is time to move on to the next cgroup
2263 */
2264 next_mz = NULL;
2265 if (!reclaimed) {
2266 do {
2267 /*
2268 * Loop until we find yet another one.
2269 *
2270 * By the time we get the soft_limit lock
2271 * again, someone might have aded the
2272 * group back on the RB tree. Iterate to
2273 * make sure we get a different mem.
2274 * mem_cgroup_largest_soft_limit_node returns
2275 * NULL if no other cgroup is present on
2276 * the tree
2277 */
2278 next_mz =
2279 __mem_cgroup_largest_soft_limit_node(mctz);
2280 if (next_mz == mz) {
2281 css_put(&next_mz->mem->css);
2282 next_mz = NULL;
2283 } else /* next_mz == NULL or other memcg */
2284 break;
2285 } while (1);
2286 }
2287 mz->usage_in_excess =
2288 res_counter_soft_limit_excess(&mz->mem->res);
2289 __mem_cgroup_remove_exceeded(mz->mem, mz, mctz);
2290 /*
2291 * One school of thought says that we should not add
2292 * back the node to the tree if reclaim returns 0.
2293 * But our reclaim could return 0, simply because due
2294 * to priority we are exposing a smaller subset of
2295 * memory to reclaim from. Consider this as a longer
2296 * term TODO.
2297 */
2298 if (mz->usage_in_excess)
2299 __mem_cgroup_insert_exceeded(mz->mem, mz, mctz);
2300 spin_unlock(&mctz->lock);
2301 css_put(&mz->mem->css);
2302 loop++;
2303 /*
2304 * Could not reclaim anything and there are no more
2305 * mem cgroups to try or we seem to be looping without
2306 * reclaiming anything.
2307 */
2308 if (!nr_reclaimed &&
2309 (next_mz == NULL ||
2310 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
2311 break;
2312 } while (!nr_reclaimed);
2313 if (next_mz)
2314 css_put(&next_mz->mem->css);
2315 return nr_reclaimed;
2316}
2317
1865/* 2318/*
1866 * This routine traverse page_cgroup in given list and drop them all. 2319 * This routine traverse page_cgroup in given list and drop them all.
1867 * *And* this routine doesn't reclaim page itself, just removes page_cgroup. 2320 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
@@ -2046,20 +2499,64 @@ static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
2046 return retval; 2499 return retval;
2047} 2500}
2048 2501
2502struct mem_cgroup_idx_data {
2503 s64 val;
2504 enum mem_cgroup_stat_index idx;
2505};
2506
2507static int
2508mem_cgroup_get_idx_stat(struct mem_cgroup *mem, void *data)
2509{
2510 struct mem_cgroup_idx_data *d = data;
2511 d->val += mem_cgroup_read_stat(&mem->stat, d->idx);
2512 return 0;
2513}
2514
2515static void
2516mem_cgroup_get_recursive_idx_stat(struct mem_cgroup *mem,
2517 enum mem_cgroup_stat_index idx, s64 *val)
2518{
2519 struct mem_cgroup_idx_data d;
2520 d.idx = idx;
2521 d.val = 0;
2522 mem_cgroup_walk_tree(mem, &d, mem_cgroup_get_idx_stat);
2523 *val = d.val;
2524}
2525
2049static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft) 2526static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
2050{ 2527{
2051 struct mem_cgroup *mem = mem_cgroup_from_cont(cont); 2528 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
2052 u64 val = 0; 2529 u64 idx_val, val;
2053 int type, name; 2530 int type, name;
2054 2531
2055 type = MEMFILE_TYPE(cft->private); 2532 type = MEMFILE_TYPE(cft->private);
2056 name = MEMFILE_ATTR(cft->private); 2533 name = MEMFILE_ATTR(cft->private);
2057 switch (type) { 2534 switch (type) {
2058 case _MEM: 2535 case _MEM:
2059 val = res_counter_read_u64(&mem->res, name); 2536 if (name == RES_USAGE && mem_cgroup_is_root(mem)) {
2537 mem_cgroup_get_recursive_idx_stat(mem,
2538 MEM_CGROUP_STAT_CACHE, &idx_val);
2539 val = idx_val;
2540 mem_cgroup_get_recursive_idx_stat(mem,
2541 MEM_CGROUP_STAT_RSS, &idx_val);
2542 val += idx_val;
2543 val <<= PAGE_SHIFT;
2544 } else
2545 val = res_counter_read_u64(&mem->res, name);
2060 break; 2546 break;
2061 case _MEMSWAP: 2547 case _MEMSWAP:
2062 val = res_counter_read_u64(&mem->memsw, name); 2548 if (name == RES_USAGE && mem_cgroup_is_root(mem)) {
2549 mem_cgroup_get_recursive_idx_stat(mem,
2550 MEM_CGROUP_STAT_CACHE, &idx_val);
2551 val = idx_val;
2552 mem_cgroup_get_recursive_idx_stat(mem,
2553 MEM_CGROUP_STAT_RSS, &idx_val);
2554 val += idx_val;
2555 mem_cgroup_get_recursive_idx_stat(mem,
2556 MEM_CGROUP_STAT_SWAPOUT, &idx_val);
2557 val <<= PAGE_SHIFT;
2558 } else
2559 val = res_counter_read_u64(&mem->memsw, name);
2063 break; 2560 break;
2064 default: 2561 default:
2065 BUG(); 2562 BUG();
@@ -2083,6 +2580,10 @@ static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
2083 name = MEMFILE_ATTR(cft->private); 2580 name = MEMFILE_ATTR(cft->private);
2084 switch (name) { 2581 switch (name) {
2085 case RES_LIMIT: 2582 case RES_LIMIT:
2583 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
2584 ret = -EINVAL;
2585 break;
2586 }
2086 /* This function does all necessary parse...reuse it */ 2587 /* This function does all necessary parse...reuse it */
2087 ret = res_counter_memparse_write_strategy(buffer, &val); 2588 ret = res_counter_memparse_write_strategy(buffer, &val);
2088 if (ret) 2589 if (ret)
@@ -2092,6 +2593,20 @@ static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
2092 else 2593 else
2093 ret = mem_cgroup_resize_memsw_limit(memcg, val); 2594 ret = mem_cgroup_resize_memsw_limit(memcg, val);
2094 break; 2595 break;
2596 case RES_SOFT_LIMIT:
2597 ret = res_counter_memparse_write_strategy(buffer, &val);
2598 if (ret)
2599 break;
2600 /*
2601 * For memsw, soft limits are hard to implement in terms
2602 * of semantics, for now, we support soft limits for
2603 * control without swap
2604 */
2605 if (type == _MEM)
2606 ret = res_counter_set_soft_limit(&memcg->res, val);
2607 else
2608 ret = -EINVAL;
2609 break;
2095 default: 2610 default:
2096 ret = -EINVAL; /* should be BUG() ? */ 2611 ret = -EINVAL; /* should be BUG() ? */
2097 break; 2612 break;
@@ -2149,6 +2664,7 @@ static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
2149 res_counter_reset_failcnt(&mem->memsw); 2664 res_counter_reset_failcnt(&mem->memsw);
2150 break; 2665 break;
2151 } 2666 }
2667
2152 return 0; 2668 return 0;
2153} 2669}
2154 2670
@@ -2160,6 +2676,7 @@ enum {
2160 MCS_MAPPED_FILE, 2676 MCS_MAPPED_FILE,
2161 MCS_PGPGIN, 2677 MCS_PGPGIN,
2162 MCS_PGPGOUT, 2678 MCS_PGPGOUT,
2679 MCS_SWAP,
2163 MCS_INACTIVE_ANON, 2680 MCS_INACTIVE_ANON,
2164 MCS_ACTIVE_ANON, 2681 MCS_ACTIVE_ANON,
2165 MCS_INACTIVE_FILE, 2682 MCS_INACTIVE_FILE,
@@ -2181,6 +2698,7 @@ struct {
2181 {"mapped_file", "total_mapped_file"}, 2698 {"mapped_file", "total_mapped_file"},
2182 {"pgpgin", "total_pgpgin"}, 2699 {"pgpgin", "total_pgpgin"},
2183 {"pgpgout", "total_pgpgout"}, 2700 {"pgpgout", "total_pgpgout"},
2701 {"swap", "total_swap"},
2184 {"inactive_anon", "total_inactive_anon"}, 2702 {"inactive_anon", "total_inactive_anon"},
2185 {"active_anon", "total_active_anon"}, 2703 {"active_anon", "total_active_anon"},
2186 {"inactive_file", "total_inactive_file"}, 2704 {"inactive_file", "total_inactive_file"},
@@ -2205,6 +2723,10 @@ static int mem_cgroup_get_local_stat(struct mem_cgroup *mem, void *data)
2205 s->stat[MCS_PGPGIN] += val; 2723 s->stat[MCS_PGPGIN] += val;
2206 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGOUT_COUNT); 2724 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGOUT_COUNT);
2207 s->stat[MCS_PGPGOUT] += val; 2725 s->stat[MCS_PGPGOUT] += val;
2726 if (do_swap_account) {
2727 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_SWAPOUT);
2728 s->stat[MCS_SWAP] += val * PAGE_SIZE;
2729 }
2208 2730
2209 /* per zone stat */ 2731 /* per zone stat */
2210 val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON); 2732 val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON);
@@ -2236,8 +2758,11 @@ static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
2236 memset(&mystat, 0, sizeof(mystat)); 2758 memset(&mystat, 0, sizeof(mystat));
2237 mem_cgroup_get_local_stat(mem_cont, &mystat); 2759 mem_cgroup_get_local_stat(mem_cont, &mystat);
2238 2760
2239 for (i = 0; i < NR_MCS_STAT; i++) 2761 for (i = 0; i < NR_MCS_STAT; i++) {
2762 if (i == MCS_SWAP && !do_swap_account)
2763 continue;
2240 cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]); 2764 cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]);
2765 }
2241 2766
2242 /* Hierarchical information */ 2767 /* Hierarchical information */
2243 { 2768 {
@@ -2250,9 +2775,11 @@ static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
2250 2775
2251 memset(&mystat, 0, sizeof(mystat)); 2776 memset(&mystat, 0, sizeof(mystat));
2252 mem_cgroup_get_total_stat(mem_cont, &mystat); 2777 mem_cgroup_get_total_stat(mem_cont, &mystat);
2253 for (i = 0; i < NR_MCS_STAT; i++) 2778 for (i = 0; i < NR_MCS_STAT; i++) {
2779 if (i == MCS_SWAP && !do_swap_account)
2780 continue;
2254 cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]); 2781 cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]);
2255 2782 }
2256 2783
2257#ifdef CONFIG_DEBUG_VM 2784#ifdef CONFIG_DEBUG_VM
2258 cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL)); 2785 cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL));
@@ -2345,6 +2872,12 @@ static struct cftype mem_cgroup_files[] = {
2345 .read_u64 = mem_cgroup_read, 2872 .read_u64 = mem_cgroup_read,
2346 }, 2873 },
2347 { 2874 {
2875 .name = "soft_limit_in_bytes",
2876 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
2877 .write_string = mem_cgroup_write,
2878 .read_u64 = mem_cgroup_read,
2879 },
2880 {
2348 .name = "failcnt", 2881 .name = "failcnt",
2349 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), 2882 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
2350 .trigger = mem_cgroup_reset, 2883 .trigger = mem_cgroup_reset,
@@ -2438,6 +2971,9 @@ static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
2438 mz = &pn->zoneinfo[zone]; 2971 mz = &pn->zoneinfo[zone];
2439 for_each_lru(l) 2972 for_each_lru(l)
2440 INIT_LIST_HEAD(&mz->lists[l]); 2973 INIT_LIST_HEAD(&mz->lists[l]);
2974 mz->usage_in_excess = 0;
2975 mz->on_tree = false;
2976 mz->mem = mem;
2441 } 2977 }
2442 return 0; 2978 return 0;
2443} 2979}
@@ -2483,6 +3019,7 @@ static void __mem_cgroup_free(struct mem_cgroup *mem)
2483{ 3019{
2484 int node; 3020 int node;
2485 3021
3022 mem_cgroup_remove_from_trees(mem);
2486 free_css_id(&mem_cgroup_subsys, &mem->css); 3023 free_css_id(&mem_cgroup_subsys, &mem->css);
2487 3024
2488 for_each_node_state(node, N_POSSIBLE) 3025 for_each_node_state(node, N_POSSIBLE)
@@ -2531,6 +3068,31 @@ static void __init enable_swap_cgroup(void)
2531} 3068}
2532#endif 3069#endif
2533 3070
3071static int mem_cgroup_soft_limit_tree_init(void)
3072{
3073 struct mem_cgroup_tree_per_node *rtpn;
3074 struct mem_cgroup_tree_per_zone *rtpz;
3075 int tmp, node, zone;
3076
3077 for_each_node_state(node, N_POSSIBLE) {
3078 tmp = node;
3079 if (!node_state(node, N_NORMAL_MEMORY))
3080 tmp = -1;
3081 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp);
3082 if (!rtpn)
3083 return 1;
3084
3085 soft_limit_tree.rb_tree_per_node[node] = rtpn;
3086
3087 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
3088 rtpz = &rtpn->rb_tree_per_zone[zone];
3089 rtpz->rb_root = RB_ROOT;
3090 spin_lock_init(&rtpz->lock);
3091 }
3092 }
3093 return 0;
3094}
3095
2534static struct cgroup_subsys_state * __ref 3096static struct cgroup_subsys_state * __ref
2535mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) 3097mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
2536{ 3098{
@@ -2545,10 +3107,15 @@ mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
2545 for_each_node_state(node, N_POSSIBLE) 3107 for_each_node_state(node, N_POSSIBLE)
2546 if (alloc_mem_cgroup_per_zone_info(mem, node)) 3108 if (alloc_mem_cgroup_per_zone_info(mem, node))
2547 goto free_out; 3109 goto free_out;
3110
2548 /* root ? */ 3111 /* root ? */
2549 if (cont->parent == NULL) { 3112 if (cont->parent == NULL) {
2550 enable_swap_cgroup(); 3113 enable_swap_cgroup();
2551 parent = NULL; 3114 parent = NULL;
3115 root_mem_cgroup = mem;
3116 if (mem_cgroup_soft_limit_tree_init())
3117 goto free_out;
3118
2552 } else { 3119 } else {
2553 parent = mem_cgroup_from_cont(cont->parent); 3120 parent = mem_cgroup_from_cont(cont->parent);
2554 mem->use_hierarchy = parent->use_hierarchy; 3121 mem->use_hierarchy = parent->use_hierarchy;
@@ -2577,6 +3144,7 @@ mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
2577 return &mem->css; 3144 return &mem->css;
2578free_out: 3145free_out:
2579 __mem_cgroup_free(mem); 3146 __mem_cgroup_free(mem);
3147 root_mem_cgroup = NULL;
2580 return ERR_PTR(error); 3148 return ERR_PTR(error);
2581} 3149}
2582 3150
@@ -2612,7 +3180,8 @@ static int mem_cgroup_populate(struct cgroup_subsys *ss,
2612static void mem_cgroup_move_task(struct cgroup_subsys *ss, 3180static void mem_cgroup_move_task(struct cgroup_subsys *ss,
2613 struct cgroup *cont, 3181 struct cgroup *cont,
2614 struct cgroup *old_cont, 3182 struct cgroup *old_cont,
2615 struct task_struct *p) 3183 struct task_struct *p,
3184 bool threadgroup)
2616{ 3185{
2617 mutex_lock(&memcg_tasklist); 3186 mutex_lock(&memcg_tasklist);
2618 /* 3187 /*
diff --git a/mm/memory-failure.c b/mm/memory-failure.c
new file mode 100644
index 000000000000..729d4b15b645
--- /dev/null
+++ b/mm/memory-failure.c
@@ -0,0 +1,832 @@
1/*
2 * Copyright (C) 2008, 2009 Intel Corporation
3 * Authors: Andi Kleen, Fengguang Wu
4 *
5 * This software may be redistributed and/or modified under the terms of
6 * the GNU General Public License ("GPL") version 2 only as published by the
7 * Free Software Foundation.
8 *
9 * High level machine check handler. Handles pages reported by the
10 * hardware as being corrupted usually due to a 2bit ECC memory or cache
11 * failure.
12 *
13 * Handles page cache pages in various states. The tricky part
14 * here is that we can access any page asynchronous to other VM
15 * users, because memory failures could happen anytime and anywhere,
16 * possibly violating some of their assumptions. This is why this code
17 * has to be extremely careful. Generally it tries to use normal locking
18 * rules, as in get the standard locks, even if that means the
19 * error handling takes potentially a long time.
20 *
21 * The operation to map back from RMAP chains to processes has to walk
22 * the complete process list and has non linear complexity with the number
23 * mappings. In short it can be quite slow. But since memory corruptions
24 * are rare we hope to get away with this.
25 */
26
27/*
28 * Notebook:
29 * - hugetlb needs more code
30 * - kcore/oldmem/vmcore/mem/kmem check for hwpoison pages
31 * - pass bad pages to kdump next kernel
32 */
33#define DEBUG 1 /* remove me in 2.6.34 */
34#include <linux/kernel.h>
35#include <linux/mm.h>
36#include <linux/page-flags.h>
37#include <linux/sched.h>
38#include <linux/rmap.h>
39#include <linux/pagemap.h>
40#include <linux/swap.h>
41#include <linux/backing-dev.h>
42#include "internal.h"
43
44int sysctl_memory_failure_early_kill __read_mostly = 0;
45
46int sysctl_memory_failure_recovery __read_mostly = 1;
47
48atomic_long_t mce_bad_pages __read_mostly = ATOMIC_LONG_INIT(0);
49
50/*
51 * Send all the processes who have the page mapped an ``action optional''
52 * signal.
53 */
54static int kill_proc_ao(struct task_struct *t, unsigned long addr, int trapno,
55 unsigned long pfn)
56{
57 struct siginfo si;
58 int ret;
59
60 printk(KERN_ERR
61 "MCE %#lx: Killing %s:%d early due to hardware memory corruption\n",
62 pfn, t->comm, t->pid);
63 si.si_signo = SIGBUS;
64 si.si_errno = 0;
65 si.si_code = BUS_MCEERR_AO;
66 si.si_addr = (void *)addr;
67#ifdef __ARCH_SI_TRAPNO
68 si.si_trapno = trapno;
69#endif
70 si.si_addr_lsb = PAGE_SHIFT;
71 /*
72 * Don't use force here, it's convenient if the signal
73 * can be temporarily blocked.
74 * This could cause a loop when the user sets SIGBUS
75 * to SIG_IGN, but hopefully noone will do that?
76 */
77 ret = send_sig_info(SIGBUS, &si, t); /* synchronous? */
78 if (ret < 0)
79 printk(KERN_INFO "MCE: Error sending signal to %s:%d: %d\n",
80 t->comm, t->pid, ret);
81 return ret;
82}
83
84/*
85 * Kill all processes that have a poisoned page mapped and then isolate
86 * the page.
87 *
88 * General strategy:
89 * Find all processes having the page mapped and kill them.
90 * But we keep a page reference around so that the page is not
91 * actually freed yet.
92 * Then stash the page away
93 *
94 * There's no convenient way to get back to mapped processes
95 * from the VMAs. So do a brute-force search over all
96 * running processes.
97 *
98 * Remember that machine checks are not common (or rather
99 * if they are common you have other problems), so this shouldn't
100 * be a performance issue.
101 *
102 * Also there are some races possible while we get from the
103 * error detection to actually handle it.
104 */
105
106struct to_kill {
107 struct list_head nd;
108 struct task_struct *tsk;
109 unsigned long addr;
110 unsigned addr_valid:1;
111};
112
113/*
114 * Failure handling: if we can't find or can't kill a process there's
115 * not much we can do. We just print a message and ignore otherwise.
116 */
117
118/*
119 * Schedule a process for later kill.
120 * Uses GFP_ATOMIC allocations to avoid potential recursions in the VM.
121 * TBD would GFP_NOIO be enough?
122 */
123static void add_to_kill(struct task_struct *tsk, struct page *p,
124 struct vm_area_struct *vma,
125 struct list_head *to_kill,
126 struct to_kill **tkc)
127{
128 struct to_kill *tk;
129
130 if (*tkc) {
131 tk = *tkc;
132 *tkc = NULL;
133 } else {
134 tk = kmalloc(sizeof(struct to_kill), GFP_ATOMIC);
135 if (!tk) {
136 printk(KERN_ERR
137 "MCE: Out of memory while machine check handling\n");
138 return;
139 }
140 }
141 tk->addr = page_address_in_vma(p, vma);
142 tk->addr_valid = 1;
143
144 /*
145 * In theory we don't have to kill when the page was
146 * munmaped. But it could be also a mremap. Since that's
147 * likely very rare kill anyways just out of paranoia, but use
148 * a SIGKILL because the error is not contained anymore.
149 */
150 if (tk->addr == -EFAULT) {
151 pr_debug("MCE: Unable to find user space address %lx in %s\n",
152 page_to_pfn(p), tsk->comm);
153 tk->addr_valid = 0;
154 }
155 get_task_struct(tsk);
156 tk->tsk = tsk;
157 list_add_tail(&tk->nd, to_kill);
158}
159
160/*
161 * Kill the processes that have been collected earlier.
162 *
163 * Only do anything when DOIT is set, otherwise just free the list
164 * (this is used for clean pages which do not need killing)
165 * Also when FAIL is set do a force kill because something went
166 * wrong earlier.
167 */
168static void kill_procs_ao(struct list_head *to_kill, int doit, int trapno,
169 int fail, unsigned long pfn)
170{
171 struct to_kill *tk, *next;
172
173 list_for_each_entry_safe (tk, next, to_kill, nd) {
174 if (doit) {
175 /*
176 * In case something went wrong with munmaping
177 * make sure the process doesn't catch the
178 * signal and then access the memory. Just kill it.
179 * the signal handlers
180 */
181 if (fail || tk->addr_valid == 0) {
182 printk(KERN_ERR
183 "MCE %#lx: forcibly killing %s:%d because of failure to unmap corrupted page\n",
184 pfn, tk->tsk->comm, tk->tsk->pid);
185 force_sig(SIGKILL, tk->tsk);
186 }
187
188 /*
189 * In theory the process could have mapped
190 * something else on the address in-between. We could
191 * check for that, but we need to tell the
192 * process anyways.
193 */
194 else if (kill_proc_ao(tk->tsk, tk->addr, trapno,
195 pfn) < 0)
196 printk(KERN_ERR
197 "MCE %#lx: Cannot send advisory machine check signal to %s:%d\n",
198 pfn, tk->tsk->comm, tk->tsk->pid);
199 }
200 put_task_struct(tk->tsk);
201 kfree(tk);
202 }
203}
204
205static int task_early_kill(struct task_struct *tsk)
206{
207 if (!tsk->mm)
208 return 0;
209 if (tsk->flags & PF_MCE_PROCESS)
210 return !!(tsk->flags & PF_MCE_EARLY);
211 return sysctl_memory_failure_early_kill;
212}
213
214/*
215 * Collect processes when the error hit an anonymous page.
216 */
217static void collect_procs_anon(struct page *page, struct list_head *to_kill,
218 struct to_kill **tkc)
219{
220 struct vm_area_struct *vma;
221 struct task_struct *tsk;
222 struct anon_vma *av;
223
224 read_lock(&tasklist_lock);
225 av = page_lock_anon_vma(page);
226 if (av == NULL) /* Not actually mapped anymore */
227 goto out;
228 for_each_process (tsk) {
229 if (!task_early_kill(tsk))
230 continue;
231 list_for_each_entry (vma, &av->head, anon_vma_node) {
232 if (!page_mapped_in_vma(page, vma))
233 continue;
234 if (vma->vm_mm == tsk->mm)
235 add_to_kill(tsk, page, vma, to_kill, tkc);
236 }
237 }
238 page_unlock_anon_vma(av);
239out:
240 read_unlock(&tasklist_lock);
241}
242
243/*
244 * Collect processes when the error hit a file mapped page.
245 */
246static void collect_procs_file(struct page *page, struct list_head *to_kill,
247 struct to_kill **tkc)
248{
249 struct vm_area_struct *vma;
250 struct task_struct *tsk;
251 struct prio_tree_iter iter;
252 struct address_space *mapping = page->mapping;
253
254 /*
255 * A note on the locking order between the two locks.
256 * We don't rely on this particular order.
257 * If you have some other code that needs a different order
258 * feel free to switch them around. Or add a reverse link
259 * from mm_struct to task_struct, then this could be all
260 * done without taking tasklist_lock and looping over all tasks.
261 */
262
263 read_lock(&tasklist_lock);
264 spin_lock(&mapping->i_mmap_lock);
265 for_each_process(tsk) {
266 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
267
268 if (!task_early_kill(tsk))
269 continue;
270
271 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff,
272 pgoff) {
273 /*
274 * Send early kill signal to tasks where a vma covers
275 * the page but the corrupted page is not necessarily
276 * mapped it in its pte.
277 * Assume applications who requested early kill want
278 * to be informed of all such data corruptions.
279 */
280 if (vma->vm_mm == tsk->mm)
281 add_to_kill(tsk, page, vma, to_kill, tkc);
282 }
283 }
284 spin_unlock(&mapping->i_mmap_lock);
285 read_unlock(&tasklist_lock);
286}
287
288/*
289 * Collect the processes who have the corrupted page mapped to kill.
290 * This is done in two steps for locking reasons.
291 * First preallocate one tokill structure outside the spin locks,
292 * so that we can kill at least one process reasonably reliable.
293 */
294static void collect_procs(struct page *page, struct list_head *tokill)
295{
296 struct to_kill *tk;
297
298 if (!page->mapping)
299 return;
300
301 tk = kmalloc(sizeof(struct to_kill), GFP_NOIO);
302 if (!tk)
303 return;
304 if (PageAnon(page))
305 collect_procs_anon(page, tokill, &tk);
306 else
307 collect_procs_file(page, tokill, &tk);
308 kfree(tk);
309}
310
311/*
312 * Error handlers for various types of pages.
313 */
314
315enum outcome {
316 FAILED, /* Error handling failed */
317 DELAYED, /* Will be handled later */
318 IGNORED, /* Error safely ignored */
319 RECOVERED, /* Successfully recovered */
320};
321
322static const char *action_name[] = {
323 [FAILED] = "Failed",
324 [DELAYED] = "Delayed",
325 [IGNORED] = "Ignored",
326 [RECOVERED] = "Recovered",
327};
328
329/*
330 * Error hit kernel page.
331 * Do nothing, try to be lucky and not touch this instead. For a few cases we
332 * could be more sophisticated.
333 */
334static int me_kernel(struct page *p, unsigned long pfn)
335{
336 return DELAYED;
337}
338
339/*
340 * Already poisoned page.
341 */
342static int me_ignore(struct page *p, unsigned long pfn)
343{
344 return IGNORED;
345}
346
347/*
348 * Page in unknown state. Do nothing.
349 */
350static int me_unknown(struct page *p, unsigned long pfn)
351{
352 printk(KERN_ERR "MCE %#lx: Unknown page state\n", pfn);
353 return FAILED;
354}
355
356/*
357 * Free memory
358 */
359static int me_free(struct page *p, unsigned long pfn)
360{
361 return DELAYED;
362}
363
364/*
365 * Clean (or cleaned) page cache page.
366 */
367static int me_pagecache_clean(struct page *p, unsigned long pfn)
368{
369 int err;
370 int ret = FAILED;
371 struct address_space *mapping;
372
373 if (!isolate_lru_page(p))
374 page_cache_release(p);
375
376 /*
377 * For anonymous pages we're done the only reference left
378 * should be the one m_f() holds.
379 */
380 if (PageAnon(p))
381 return RECOVERED;
382
383 /*
384 * Now truncate the page in the page cache. This is really
385 * more like a "temporary hole punch"
386 * Don't do this for block devices when someone else
387 * has a reference, because it could be file system metadata
388 * and that's not safe to truncate.
389 */
390 mapping = page_mapping(p);
391 if (!mapping) {
392 /*
393 * Page has been teared down in the meanwhile
394 */
395 return FAILED;
396 }
397
398 /*
399 * Truncation is a bit tricky. Enable it per file system for now.
400 *
401 * Open: to take i_mutex or not for this? Right now we don't.
402 */
403 if (mapping->a_ops->error_remove_page) {
404 err = mapping->a_ops->error_remove_page(mapping, p);
405 if (err != 0) {
406 printk(KERN_INFO "MCE %#lx: Failed to punch page: %d\n",
407 pfn, err);
408 } else if (page_has_private(p) &&
409 !try_to_release_page(p, GFP_NOIO)) {
410 pr_debug("MCE %#lx: failed to release buffers\n", pfn);
411 } else {
412 ret = RECOVERED;
413 }
414 } else {
415 /*
416 * If the file system doesn't support it just invalidate
417 * This fails on dirty or anything with private pages
418 */
419 if (invalidate_inode_page(p))
420 ret = RECOVERED;
421 else
422 printk(KERN_INFO "MCE %#lx: Failed to invalidate\n",
423 pfn);
424 }
425 return ret;
426}
427
428/*
429 * Dirty cache page page
430 * Issues: when the error hit a hole page the error is not properly
431 * propagated.
432 */
433static int me_pagecache_dirty(struct page *p, unsigned long pfn)
434{
435 struct address_space *mapping = page_mapping(p);
436
437 SetPageError(p);
438 /* TBD: print more information about the file. */
439 if (mapping) {
440 /*
441 * IO error will be reported by write(), fsync(), etc.
442 * who check the mapping.
443 * This way the application knows that something went
444 * wrong with its dirty file data.
445 *
446 * There's one open issue:
447 *
448 * The EIO will be only reported on the next IO
449 * operation and then cleared through the IO map.
450 * Normally Linux has two mechanisms to pass IO error
451 * first through the AS_EIO flag in the address space
452 * and then through the PageError flag in the page.
453 * Since we drop pages on memory failure handling the
454 * only mechanism open to use is through AS_AIO.
455 *
456 * This has the disadvantage that it gets cleared on
457 * the first operation that returns an error, while
458 * the PageError bit is more sticky and only cleared
459 * when the page is reread or dropped. If an
460 * application assumes it will always get error on
461 * fsync, but does other operations on the fd before
462 * and the page is dropped inbetween then the error
463 * will not be properly reported.
464 *
465 * This can already happen even without hwpoisoned
466 * pages: first on metadata IO errors (which only
467 * report through AS_EIO) or when the page is dropped
468 * at the wrong time.
469 *
470 * So right now we assume that the application DTRT on
471 * the first EIO, but we're not worse than other parts
472 * of the kernel.
473 */
474 mapping_set_error(mapping, EIO);
475 }
476
477 return me_pagecache_clean(p, pfn);
478}
479
480/*
481 * Clean and dirty swap cache.
482 *
483 * Dirty swap cache page is tricky to handle. The page could live both in page
484 * cache and swap cache(ie. page is freshly swapped in). So it could be
485 * referenced concurrently by 2 types of PTEs:
486 * normal PTEs and swap PTEs. We try to handle them consistently by calling
487 * try_to_unmap(TTU_IGNORE_HWPOISON) to convert the normal PTEs to swap PTEs,
488 * and then
489 * - clear dirty bit to prevent IO
490 * - remove from LRU
491 * - but keep in the swap cache, so that when we return to it on
492 * a later page fault, we know the application is accessing
493 * corrupted data and shall be killed (we installed simple
494 * interception code in do_swap_page to catch it).
495 *
496 * Clean swap cache pages can be directly isolated. A later page fault will
497 * bring in the known good data from disk.
498 */
499static int me_swapcache_dirty(struct page *p, unsigned long pfn)
500{
501 int ret = FAILED;
502
503 ClearPageDirty(p);
504 /* Trigger EIO in shmem: */
505 ClearPageUptodate(p);
506
507 if (!isolate_lru_page(p)) {
508 page_cache_release(p);
509 ret = DELAYED;
510 }
511
512 return ret;
513}
514
515static int me_swapcache_clean(struct page *p, unsigned long pfn)
516{
517 int ret = FAILED;
518
519 if (!isolate_lru_page(p)) {
520 page_cache_release(p);
521 ret = RECOVERED;
522 }
523 delete_from_swap_cache(p);
524 return ret;
525}
526
527/*
528 * Huge pages. Needs work.
529 * Issues:
530 * No rmap support so we cannot find the original mapper. In theory could walk
531 * all MMs and look for the mappings, but that would be non atomic and racy.
532 * Need rmap for hugepages for this. Alternatively we could employ a heuristic,
533 * like just walking the current process and hoping it has it mapped (that
534 * should be usually true for the common "shared database cache" case)
535 * Should handle free huge pages and dequeue them too, but this needs to
536 * handle huge page accounting correctly.
537 */
538static int me_huge_page(struct page *p, unsigned long pfn)
539{
540 return FAILED;
541}
542
543/*
544 * Various page states we can handle.
545 *
546 * A page state is defined by its current page->flags bits.
547 * The table matches them in order and calls the right handler.
548 *
549 * This is quite tricky because we can access page at any time
550 * in its live cycle, so all accesses have to be extremly careful.
551 *
552 * This is not complete. More states could be added.
553 * For any missing state don't attempt recovery.
554 */
555
556#define dirty (1UL << PG_dirty)
557#define sc (1UL << PG_swapcache)
558#define unevict (1UL << PG_unevictable)
559#define mlock (1UL << PG_mlocked)
560#define writeback (1UL << PG_writeback)
561#define lru (1UL << PG_lru)
562#define swapbacked (1UL << PG_swapbacked)
563#define head (1UL << PG_head)
564#define tail (1UL << PG_tail)
565#define compound (1UL << PG_compound)
566#define slab (1UL << PG_slab)
567#define buddy (1UL << PG_buddy)
568#define reserved (1UL << PG_reserved)
569
570static struct page_state {
571 unsigned long mask;
572 unsigned long res;
573 char *msg;
574 int (*action)(struct page *p, unsigned long pfn);
575} error_states[] = {
576 { reserved, reserved, "reserved kernel", me_ignore },
577 { buddy, buddy, "free kernel", me_free },
578
579 /*
580 * Could in theory check if slab page is free or if we can drop
581 * currently unused objects without touching them. But just
582 * treat it as standard kernel for now.
583 */
584 { slab, slab, "kernel slab", me_kernel },
585
586#ifdef CONFIG_PAGEFLAGS_EXTENDED
587 { head, head, "huge", me_huge_page },
588 { tail, tail, "huge", me_huge_page },
589#else
590 { compound, compound, "huge", me_huge_page },
591#endif
592
593 { sc|dirty, sc|dirty, "swapcache", me_swapcache_dirty },
594 { sc|dirty, sc, "swapcache", me_swapcache_clean },
595
596 { unevict|dirty, unevict|dirty, "unevictable LRU", me_pagecache_dirty},
597 { unevict, unevict, "unevictable LRU", me_pagecache_clean},
598
599#ifdef CONFIG_HAVE_MLOCKED_PAGE_BIT
600 { mlock|dirty, mlock|dirty, "mlocked LRU", me_pagecache_dirty },
601 { mlock, mlock, "mlocked LRU", me_pagecache_clean },
602#endif
603
604 { lru|dirty, lru|dirty, "LRU", me_pagecache_dirty },
605 { lru|dirty, lru, "clean LRU", me_pagecache_clean },
606 { swapbacked, swapbacked, "anonymous", me_pagecache_clean },
607
608 /*
609 * Catchall entry: must be at end.
610 */
611 { 0, 0, "unknown page state", me_unknown },
612};
613
614#undef lru
615
616static void action_result(unsigned long pfn, char *msg, int result)
617{
618 struct page *page = NULL;
619 if (pfn_valid(pfn))
620 page = pfn_to_page(pfn);
621
622 printk(KERN_ERR "MCE %#lx: %s%s page recovery: %s\n",
623 pfn,
624 page && PageDirty(page) ? "dirty " : "",
625 msg, action_name[result]);
626}
627
628static int page_action(struct page_state *ps, struct page *p,
629 unsigned long pfn, int ref)
630{
631 int result;
632
633 result = ps->action(p, pfn);
634 action_result(pfn, ps->msg, result);
635 if (page_count(p) != 1 + ref)
636 printk(KERN_ERR
637 "MCE %#lx: %s page still referenced by %d users\n",
638 pfn, ps->msg, page_count(p) - 1);
639
640 /* Could do more checks here if page looks ok */
641 /*
642 * Could adjust zone counters here to correct for the missing page.
643 */
644
645 return result == RECOVERED ? 0 : -EBUSY;
646}
647
648#define N_UNMAP_TRIES 5
649
650/*
651 * Do all that is necessary to remove user space mappings. Unmap
652 * the pages and send SIGBUS to the processes if the data was dirty.
653 */
654static void hwpoison_user_mappings(struct page *p, unsigned long pfn,
655 int trapno)
656{
657 enum ttu_flags ttu = TTU_UNMAP | TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS;
658 struct address_space *mapping;
659 LIST_HEAD(tokill);
660 int ret;
661 int i;
662 int kill = 1;
663
664 if (PageReserved(p) || PageCompound(p) || PageSlab(p))
665 return;
666
667 if (!PageLRU(p))
668 lru_add_drain_all();
669
670 /*
671 * This check implies we don't kill processes if their pages
672 * are in the swap cache early. Those are always late kills.
673 */
674 if (!page_mapped(p))
675 return;
676
677 if (PageSwapCache(p)) {
678 printk(KERN_ERR
679 "MCE %#lx: keeping poisoned page in swap cache\n", pfn);
680 ttu |= TTU_IGNORE_HWPOISON;
681 }
682
683 /*
684 * Propagate the dirty bit from PTEs to struct page first, because we
685 * need this to decide if we should kill or just drop the page.
686 */
687 mapping = page_mapping(p);
688 if (!PageDirty(p) && mapping && mapping_cap_writeback_dirty(mapping)) {
689 if (page_mkclean(p)) {
690 SetPageDirty(p);
691 } else {
692 kill = 0;
693 ttu |= TTU_IGNORE_HWPOISON;
694 printk(KERN_INFO
695 "MCE %#lx: corrupted page was clean: dropped without side effects\n",
696 pfn);
697 }
698 }
699
700 /*
701 * First collect all the processes that have the page
702 * mapped in dirty form. This has to be done before try_to_unmap,
703 * because ttu takes the rmap data structures down.
704 *
705 * Error handling: We ignore errors here because
706 * there's nothing that can be done.
707 */
708 if (kill)
709 collect_procs(p, &tokill);
710
711 /*
712 * try_to_unmap can fail temporarily due to races.
713 * Try a few times (RED-PEN better strategy?)
714 */
715 for (i = 0; i < N_UNMAP_TRIES; i++) {
716 ret = try_to_unmap(p, ttu);
717 if (ret == SWAP_SUCCESS)
718 break;
719 pr_debug("MCE %#lx: try_to_unmap retry needed %d\n", pfn, ret);
720 }
721
722 if (ret != SWAP_SUCCESS)
723 printk(KERN_ERR "MCE %#lx: failed to unmap page (mapcount=%d)\n",
724 pfn, page_mapcount(p));
725
726 /*
727 * Now that the dirty bit has been propagated to the
728 * struct page and all unmaps done we can decide if
729 * killing is needed or not. Only kill when the page
730 * was dirty, otherwise the tokill list is merely
731 * freed. When there was a problem unmapping earlier
732 * use a more force-full uncatchable kill to prevent
733 * any accesses to the poisoned memory.
734 */
735 kill_procs_ao(&tokill, !!PageDirty(p), trapno,
736 ret != SWAP_SUCCESS, pfn);
737}
738
739int __memory_failure(unsigned long pfn, int trapno, int ref)
740{
741 struct page_state *ps;
742 struct page *p;
743 int res;
744
745 if (!sysctl_memory_failure_recovery)
746 panic("Memory failure from trap %d on page %lx", trapno, pfn);
747
748 if (!pfn_valid(pfn)) {
749 action_result(pfn, "memory outside kernel control", IGNORED);
750 return -EIO;
751 }
752
753 p = pfn_to_page(pfn);
754 if (TestSetPageHWPoison(p)) {
755 action_result(pfn, "already hardware poisoned", IGNORED);
756 return 0;
757 }
758
759 atomic_long_add(1, &mce_bad_pages);
760
761 /*
762 * We need/can do nothing about count=0 pages.
763 * 1) it's a free page, and therefore in safe hand:
764 * prep_new_page() will be the gate keeper.
765 * 2) it's part of a non-compound high order page.
766 * Implies some kernel user: cannot stop them from
767 * R/W the page; let's pray that the page has been
768 * used and will be freed some time later.
769 * In fact it's dangerous to directly bump up page count from 0,
770 * that may make page_freeze_refs()/page_unfreeze_refs() mismatch.
771 */
772 if (!get_page_unless_zero(compound_head(p))) {
773 action_result(pfn, "free or high order kernel", IGNORED);
774 return PageBuddy(compound_head(p)) ? 0 : -EBUSY;
775 }
776
777 /*
778 * Lock the page and wait for writeback to finish.
779 * It's very difficult to mess with pages currently under IO
780 * and in many cases impossible, so we just avoid it here.
781 */
782 lock_page_nosync(p);
783 wait_on_page_writeback(p);
784
785 /*
786 * Now take care of user space mappings.
787 */
788 hwpoison_user_mappings(p, pfn, trapno);
789
790 /*
791 * Torn down by someone else?
792 */
793 if (PageLRU(p) && !PageSwapCache(p) && p->mapping == NULL) {
794 action_result(pfn, "already truncated LRU", IGNORED);
795 res = 0;
796 goto out;
797 }
798
799 res = -EBUSY;
800 for (ps = error_states;; ps++) {
801 if ((p->flags & ps->mask) == ps->res) {
802 res = page_action(ps, p, pfn, ref);
803 break;
804 }
805 }
806out:
807 unlock_page(p);
808 return res;
809}
810EXPORT_SYMBOL_GPL(__memory_failure);
811
812/**
813 * memory_failure - Handle memory failure of a page.
814 * @pfn: Page Number of the corrupted page
815 * @trapno: Trap number reported in the signal to user space.
816 *
817 * This function is called by the low level machine check code
818 * of an architecture when it detects hardware memory corruption
819 * of a page. It tries its best to recover, which includes
820 * dropping pages, killing processes etc.
821 *
822 * The function is primarily of use for corruptions that
823 * happen outside the current execution context (e.g. when
824 * detected by a background scrubber)
825 *
826 * Must run in process context (e.g. a work queue) with interrupts
827 * enabled and no spinlocks hold.
828 */
829void memory_failure(unsigned long pfn, int trapno)
830{
831 __memory_failure(pfn, trapno, 0);
832}
diff --git a/mm/memory.c b/mm/memory.c
index b1443ac07c00..7e91b5f9f690 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -297,7 +297,8 @@ void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma,
297 unsigned long addr = vma->vm_start; 297 unsigned long addr = vma->vm_start;
298 298
299 /* 299 /*
300 * Hide vma from rmap and vmtruncate before freeing pgtables 300 * Hide vma from rmap and truncate_pagecache before freeing
301 * pgtables
301 */ 302 */
302 anon_vma_unlink(vma); 303 anon_vma_unlink(vma);
303 unlink_file_vma(vma); 304 unlink_file_vma(vma);
@@ -1325,7 +1326,8 @@ int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
1325 if (ret & VM_FAULT_ERROR) { 1326 if (ret & VM_FAULT_ERROR) {
1326 if (ret & VM_FAULT_OOM) 1327 if (ret & VM_FAULT_OOM)
1327 return i ? i : -ENOMEM; 1328 return i ? i : -ENOMEM;
1328 else if (ret & VM_FAULT_SIGBUS) 1329 if (ret &
1330 (VM_FAULT_HWPOISON|VM_FAULT_SIGBUS))
1329 return i ? i : -EFAULT; 1331 return i ? i : -EFAULT;
1330 BUG(); 1332 BUG();
1331 } 1333 }
@@ -2407,7 +2409,7 @@ restart:
2407 * @mapping: the address space containing mmaps to be unmapped. 2409 * @mapping: the address space containing mmaps to be unmapped.
2408 * @holebegin: byte in first page to unmap, relative to the start of 2410 * @holebegin: byte in first page to unmap, relative to the start of
2409 * the underlying file. This will be rounded down to a PAGE_SIZE 2411 * the underlying file. This will be rounded down to a PAGE_SIZE
2410 * boundary. Note that this is different from vmtruncate(), which 2412 * boundary. Note that this is different from truncate_pagecache(), which
2411 * must keep the partial page. In contrast, we must get rid of 2413 * must keep the partial page. In contrast, we must get rid of
2412 * partial pages. 2414 * partial pages.
2413 * @holelen: size of prospective hole in bytes. This will be rounded 2415 * @holelen: size of prospective hole in bytes. This will be rounded
@@ -2458,63 +2460,6 @@ void unmap_mapping_range(struct address_space *mapping,
2458} 2460}
2459EXPORT_SYMBOL(unmap_mapping_range); 2461EXPORT_SYMBOL(unmap_mapping_range);
2460 2462
2461/**
2462 * vmtruncate - unmap mappings "freed" by truncate() syscall
2463 * @inode: inode of the file used
2464 * @offset: file offset to start truncating
2465 *
2466 * NOTE! We have to be ready to update the memory sharing
2467 * between the file and the memory map for a potential last
2468 * incomplete page. Ugly, but necessary.
2469 */
2470int vmtruncate(struct inode * inode, loff_t offset)
2471{
2472 if (inode->i_size < offset) {
2473 unsigned long limit;
2474
2475 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
2476 if (limit != RLIM_INFINITY && offset > limit)
2477 goto out_sig;
2478 if (offset > inode->i_sb->s_maxbytes)
2479 goto out_big;
2480 i_size_write(inode, offset);
2481 } else {
2482 struct address_space *mapping = inode->i_mapping;
2483
2484 /*
2485 * truncation of in-use swapfiles is disallowed - it would
2486 * cause subsequent swapout to scribble on the now-freed
2487 * blocks.
2488 */
2489 if (IS_SWAPFILE(inode))
2490 return -ETXTBSY;
2491 i_size_write(inode, offset);
2492
2493 /*
2494 * unmap_mapping_range is called twice, first simply for
2495 * efficiency so that truncate_inode_pages does fewer
2496 * single-page unmaps. However after this first call, and
2497 * before truncate_inode_pages finishes, it is possible for
2498 * private pages to be COWed, which remain after
2499 * truncate_inode_pages finishes, hence the second
2500 * unmap_mapping_range call must be made for correctness.
2501 */
2502 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
2503 truncate_inode_pages(mapping, offset);
2504 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
2505 }
2506
2507 if (inode->i_op->truncate)
2508 inode->i_op->truncate(inode);
2509 return 0;
2510
2511out_sig:
2512 send_sig(SIGXFSZ, current, 0);
2513out_big:
2514 return -EFBIG;
2515}
2516EXPORT_SYMBOL(vmtruncate);
2517
2518int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end) 2463int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end)
2519{ 2464{
2520 struct address_space *mapping = inode->i_mapping; 2465 struct address_space *mapping = inode->i_mapping;
@@ -2559,8 +2504,15 @@ static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
2559 goto out; 2504 goto out;
2560 2505
2561 entry = pte_to_swp_entry(orig_pte); 2506 entry = pte_to_swp_entry(orig_pte);
2562 if (is_migration_entry(entry)) { 2507 if (unlikely(non_swap_entry(entry))) {
2563 migration_entry_wait(mm, pmd, address); 2508 if (is_migration_entry(entry)) {
2509 migration_entry_wait(mm, pmd, address);
2510 } else if (is_hwpoison_entry(entry)) {
2511 ret = VM_FAULT_HWPOISON;
2512 } else {
2513 print_bad_pte(vma, address, orig_pte, NULL);
2514 ret = VM_FAULT_OOM;
2515 }
2564 goto out; 2516 goto out;
2565 } 2517 }
2566 delayacct_set_flag(DELAYACCT_PF_SWAPIN); 2518 delayacct_set_flag(DELAYACCT_PF_SWAPIN);
@@ -2584,6 +2536,10 @@ static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
2584 /* Had to read the page from swap area: Major fault */ 2536 /* Had to read the page from swap area: Major fault */
2585 ret = VM_FAULT_MAJOR; 2537 ret = VM_FAULT_MAJOR;
2586 count_vm_event(PGMAJFAULT); 2538 count_vm_event(PGMAJFAULT);
2539 } else if (PageHWPoison(page)) {
2540 ret = VM_FAULT_HWPOISON;
2541 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
2542 goto out;
2587 } 2543 }
2588 2544
2589 lock_page(page); 2545 lock_page(page);
@@ -2760,6 +2716,12 @@ static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2760 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) 2716 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))
2761 return ret; 2717 return ret;
2762 2718
2719 if (unlikely(PageHWPoison(vmf.page))) {
2720 if (ret & VM_FAULT_LOCKED)
2721 unlock_page(vmf.page);
2722 return VM_FAULT_HWPOISON;
2723 }
2724
2763 /* 2725 /*
2764 * For consistency in subsequent calls, make the faulted page always 2726 * For consistency in subsequent calls, make the faulted page always
2765 * locked. 2727 * locked.
diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
index efe3e0ec2e61..821dee596377 100644
--- a/mm/memory_hotplug.c
+++ b/mm/memory_hotplug.c
@@ -413,7 +413,7 @@ int online_pages(unsigned long pfn, unsigned long nr_pages)
413 if (!populated_zone(zone)) 413 if (!populated_zone(zone))
414 need_zonelists_rebuild = 1; 414 need_zonelists_rebuild = 1;
415 415
416 ret = walk_memory_resource(pfn, nr_pages, &onlined_pages, 416 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
417 online_pages_range); 417 online_pages_range);
418 if (ret) { 418 if (ret) {
419 printk(KERN_DEBUG "online_pages %lx at %lx failed\n", 419 printk(KERN_DEBUG "online_pages %lx at %lx failed\n",
@@ -705,7 +705,7 @@ offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
705static void 705static void
706offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) 706offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
707{ 707{
708 walk_memory_resource(start_pfn, end_pfn - start_pfn, NULL, 708 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
709 offline_isolated_pages_cb); 709 offline_isolated_pages_cb);
710} 710}
711 711
@@ -731,7 +731,7 @@ check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
731 long offlined = 0; 731 long offlined = 0;
732 int ret; 732 int ret;
733 733
734 ret = walk_memory_resource(start_pfn, end_pfn - start_pfn, &offlined, 734 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
735 check_pages_isolated_cb); 735 check_pages_isolated_cb);
736 if (ret < 0) 736 if (ret < 0)
737 offlined = (long)ret; 737 offlined = (long)ret;
diff --git a/mm/migrate.c b/mm/migrate.c
index 16052e80aaac..1a4bf4813780 100644
--- a/mm/migrate.c
+++ b/mm/migrate.c
@@ -675,7 +675,7 @@ static int unmap_and_move(new_page_t get_new_page, unsigned long private,
675 } 675 }
676 676
677 /* Establish migration ptes or remove ptes */ 677 /* Establish migration ptes or remove ptes */
678 try_to_unmap(page, 1); 678 try_to_unmap(page, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS);
679 679
680skip_unmap: 680skip_unmap:
681 if (!page_mapped(page)) 681 if (!page_mapped(page))
diff --git a/mm/mremap.c b/mm/mremap.c
index 20a07dba6be0..97bff2547719 100644
--- a/mm/mremap.c
+++ b/mm/mremap.c
@@ -86,8 +86,8 @@ static void move_ptes(struct vm_area_struct *vma, pmd_t *old_pmd,
86 if (vma->vm_file) { 86 if (vma->vm_file) {
87 /* 87 /*
88 * Subtle point from Rajesh Venkatasubramanian: before 88 * Subtle point from Rajesh Venkatasubramanian: before
89 * moving file-based ptes, we must lock vmtruncate out, 89 * moving file-based ptes, we must lock truncate_pagecache
90 * since it might clean the dst vma before the src vma, 90 * out, since it might clean the dst vma before the src vma,
91 * and we propagate stale pages into the dst afterward. 91 * and we propagate stale pages into the dst afterward.
92 */ 92 */
93 mapping = vma->vm_file->f_mapping; 93 mapping = vma->vm_file->f_mapping;
diff --git a/mm/nommu.c b/mm/nommu.c
index 1a4473faac48..c73aa4753d79 100644
--- a/mm/nommu.c
+++ b/mm/nommu.c
@@ -61,6 +61,7 @@ void *high_memory;
61struct page *mem_map; 61struct page *mem_map;
62unsigned long max_mapnr; 62unsigned long max_mapnr;
63unsigned long num_physpages; 63unsigned long num_physpages;
64unsigned long highest_memmap_pfn;
64struct percpu_counter vm_committed_as; 65struct percpu_counter vm_committed_as;
65int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */ 66int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
66int sysctl_overcommit_ratio = 50; /* default is 50% */ 67int sysctl_overcommit_ratio = 50; /* default is 50% */
@@ -82,46 +83,6 @@ struct vm_operations_struct generic_file_vm_ops = {
82}; 83};
83 84
84/* 85/*
85 * Handle all mappings that got truncated by a "truncate()"
86 * system call.
87 *
88 * NOTE! We have to be ready to update the memory sharing
89 * between the file and the memory map for a potential last
90 * incomplete page. Ugly, but necessary.
91 */
92int vmtruncate(struct inode *inode, loff_t offset)
93{
94 struct address_space *mapping = inode->i_mapping;
95 unsigned long limit;
96
97 if (inode->i_size < offset)
98 goto do_expand;
99 i_size_write(inode, offset);
100
101 truncate_inode_pages(mapping, offset);
102 goto out_truncate;
103
104do_expand:
105 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
106 if (limit != RLIM_INFINITY && offset > limit)
107 goto out_sig;
108 if (offset > inode->i_sb->s_maxbytes)
109 goto out;
110 i_size_write(inode, offset);
111
112out_truncate:
113 if (inode->i_op->truncate)
114 inode->i_op->truncate(inode);
115 return 0;
116out_sig:
117 send_sig(SIGXFSZ, current, 0);
118out:
119 return -EFBIG;
120}
121
122EXPORT_SYMBOL(vmtruncate);
123
124/*
125 * Return the total memory allocated for this pointer, not 86 * Return the total memory allocated for this pointer, not
126 * just what the caller asked for. 87 * just what the caller asked for.
127 * 88 *
@@ -169,7 +130,7 @@ unsigned int kobjsize(const void *objp)
169} 130}
170 131
171int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm, 132int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
172 unsigned long start, int nr_pages, int foll_flags, 133 unsigned long start, int nr_pages, unsigned int foll_flags,
173 struct page **pages, struct vm_area_struct **vmas) 134 struct page **pages, struct vm_area_struct **vmas)
174{ 135{
175 struct vm_area_struct *vma; 136 struct vm_area_struct *vma;
@@ -865,7 +826,7 @@ static int validate_mmap_request(struct file *file,
865 int ret; 826 int ret;
866 827
867 /* do the simple checks first */ 828 /* do the simple checks first */
868 if (flags & MAP_FIXED || addr) { 829 if (flags & MAP_FIXED) {
869 printk(KERN_DEBUG 830 printk(KERN_DEBUG
870 "%d: Can't do fixed-address/overlay mmap of RAM\n", 831 "%d: Can't do fixed-address/overlay mmap of RAM\n",
871 current->pid); 832 current->pid);
@@ -1073,7 +1034,7 @@ static int do_mmap_shared_file(struct vm_area_struct *vma)
1073 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); 1034 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1074 if (ret == 0) { 1035 if (ret == 0) {
1075 vma->vm_region->vm_top = vma->vm_region->vm_end; 1036 vma->vm_region->vm_top = vma->vm_region->vm_end;
1076 return ret; 1037 return 0;
1077 } 1038 }
1078 if (ret != -ENOSYS) 1039 if (ret != -ENOSYS)
1079 return ret; 1040 return ret;
@@ -1090,7 +1051,8 @@ static int do_mmap_shared_file(struct vm_area_struct *vma)
1090 */ 1051 */
1091static int do_mmap_private(struct vm_area_struct *vma, 1052static int do_mmap_private(struct vm_area_struct *vma,
1092 struct vm_region *region, 1053 struct vm_region *region,
1093 unsigned long len) 1054 unsigned long len,
1055 unsigned long capabilities)
1094{ 1056{
1095 struct page *pages; 1057 struct page *pages;
1096 unsigned long total, point, n, rlen; 1058 unsigned long total, point, n, rlen;
@@ -1101,13 +1063,13 @@ static int do_mmap_private(struct vm_area_struct *vma,
1101 * shared mappings on devices or memory 1063 * shared mappings on devices or memory
1102 * - VM_MAYSHARE will be set if it may attempt to share 1064 * - VM_MAYSHARE will be set if it may attempt to share
1103 */ 1065 */
1104 if (vma->vm_file) { 1066 if (capabilities & BDI_CAP_MAP_DIRECT) {
1105 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); 1067 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1106 if (ret == 0) { 1068 if (ret == 0) {
1107 /* shouldn't return success if we're not sharing */ 1069 /* shouldn't return success if we're not sharing */
1108 BUG_ON(!(vma->vm_flags & VM_MAYSHARE)); 1070 BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1109 vma->vm_region->vm_top = vma->vm_region->vm_end; 1071 vma->vm_region->vm_top = vma->vm_region->vm_end;
1110 return ret; 1072 return 0;
1111 } 1073 }
1112 if (ret != -ENOSYS) 1074 if (ret != -ENOSYS)
1113 return ret; 1075 return ret;
@@ -1220,9 +1182,6 @@ unsigned long do_mmap_pgoff(struct file *file,
1220 1182
1221 kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff); 1183 kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff);
1222 1184
1223 if (!(flags & MAP_FIXED))
1224 addr = round_hint_to_min(addr);
1225
1226 /* decide whether we should attempt the mapping, and if so what sort of 1185 /* decide whether we should attempt the mapping, and if so what sort of
1227 * mapping */ 1186 * mapping */
1228 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff, 1187 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
@@ -1232,6 +1191,9 @@ unsigned long do_mmap_pgoff(struct file *file,
1232 return ret; 1191 return ret;
1233 } 1192 }
1234 1193
1194 /* we ignore the address hint */
1195 addr = 0;
1196
1235 /* we've determined that we can make the mapping, now translate what we 1197 /* we've determined that we can make the mapping, now translate what we
1236 * now know into VMA flags */ 1198 * now know into VMA flags */
1237 vm_flags = determine_vm_flags(file, prot, flags, capabilities); 1199 vm_flags = determine_vm_flags(file, prot, flags, capabilities);
@@ -1345,7 +1307,7 @@ unsigned long do_mmap_pgoff(struct file *file,
1345 * - this is the hook for quasi-memory character devices to 1307 * - this is the hook for quasi-memory character devices to
1346 * tell us the location of a shared mapping 1308 * tell us the location of a shared mapping
1347 */ 1309 */
1348 if (file && file->f_op->get_unmapped_area) { 1310 if (capabilities & BDI_CAP_MAP_DIRECT) {
1349 addr = file->f_op->get_unmapped_area(file, addr, len, 1311 addr = file->f_op->get_unmapped_area(file, addr, len,
1350 pgoff, flags); 1312 pgoff, flags);
1351 if (IS_ERR((void *) addr)) { 1313 if (IS_ERR((void *) addr)) {
@@ -1369,15 +1331,17 @@ unsigned long do_mmap_pgoff(struct file *file,
1369 } 1331 }
1370 1332
1371 vma->vm_region = region; 1333 vma->vm_region = region;
1372 add_nommu_region(region);
1373 1334
1374 /* set up the mapping */ 1335 /* set up the mapping
1336 * - the region is filled in if BDI_CAP_MAP_DIRECT is still set
1337 */
1375 if (file && vma->vm_flags & VM_SHARED) 1338 if (file && vma->vm_flags & VM_SHARED)
1376 ret = do_mmap_shared_file(vma); 1339 ret = do_mmap_shared_file(vma);
1377 else 1340 else
1378 ret = do_mmap_private(vma, region, len); 1341 ret = do_mmap_private(vma, region, len, capabilities);
1379 if (ret < 0) 1342 if (ret < 0)
1380 goto error_put_region; 1343 goto error_just_free;
1344 add_nommu_region(region);
1381 1345
1382 /* okay... we have a mapping; now we have to register it */ 1346 /* okay... we have a mapping; now we have to register it */
1383 result = vma->vm_start; 1347 result = vma->vm_start;
@@ -1395,19 +1359,6 @@ share:
1395 kleave(" = %lx", result); 1359 kleave(" = %lx", result);
1396 return result; 1360 return result;
1397 1361
1398error_put_region:
1399 __put_nommu_region(region);
1400 if (vma) {
1401 if (vma->vm_file) {
1402 fput(vma->vm_file);
1403 if (vma->vm_flags & VM_EXECUTABLE)
1404 removed_exe_file_vma(vma->vm_mm);
1405 }
1406 kmem_cache_free(vm_area_cachep, vma);
1407 }
1408 kleave(" = %d [pr]", ret);
1409 return ret;
1410
1411error_just_free: 1362error_just_free:
1412 up_write(&nommu_region_sem); 1363 up_write(&nommu_region_sem);
1413error: 1364error:
diff --git a/mm/page-writeback.c b/mm/page-writeback.c
index 8bef063125b1..69b5fbabc8bd 100644
--- a/mm/page-writeback.c
+++ b/mm/page-writeback.c
@@ -158,37 +158,37 @@ static void update_completion_period(void)
158} 158}
159 159
160int dirty_background_ratio_handler(struct ctl_table *table, int write, 160int dirty_background_ratio_handler(struct ctl_table *table, int write,
161 struct file *filp, void __user *buffer, size_t *lenp, 161 void __user *buffer, size_t *lenp,
162 loff_t *ppos) 162 loff_t *ppos)
163{ 163{
164 int ret; 164 int ret;
165 165
166 ret = proc_dointvec_minmax(table, write, filp, buffer, lenp, ppos); 166 ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
167 if (ret == 0 && write) 167 if (ret == 0 && write)
168 dirty_background_bytes = 0; 168 dirty_background_bytes = 0;
169 return ret; 169 return ret;
170} 170}
171 171
172int dirty_background_bytes_handler(struct ctl_table *table, int write, 172int dirty_background_bytes_handler(struct ctl_table *table, int write,
173 struct file *filp, void __user *buffer, size_t *lenp, 173 void __user *buffer, size_t *lenp,
174 loff_t *ppos) 174 loff_t *ppos)
175{ 175{
176 int ret; 176 int ret;
177 177
178 ret = proc_doulongvec_minmax(table, write, filp, buffer, lenp, ppos); 178 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
179 if (ret == 0 && write) 179 if (ret == 0 && write)
180 dirty_background_ratio = 0; 180 dirty_background_ratio = 0;
181 return ret; 181 return ret;
182} 182}
183 183
184int dirty_ratio_handler(struct ctl_table *table, int write, 184int dirty_ratio_handler(struct ctl_table *table, int write,
185 struct file *filp, void __user *buffer, size_t *lenp, 185 void __user *buffer, size_t *lenp,
186 loff_t *ppos) 186 loff_t *ppos)
187{ 187{
188 int old_ratio = vm_dirty_ratio; 188 int old_ratio = vm_dirty_ratio;
189 int ret; 189 int ret;
190 190
191 ret = proc_dointvec_minmax(table, write, filp, buffer, lenp, ppos); 191 ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
192 if (ret == 0 && write && vm_dirty_ratio != old_ratio) { 192 if (ret == 0 && write && vm_dirty_ratio != old_ratio) {
193 update_completion_period(); 193 update_completion_period();
194 vm_dirty_bytes = 0; 194 vm_dirty_bytes = 0;
@@ -198,13 +198,13 @@ int dirty_ratio_handler(struct ctl_table *table, int write,
198 198
199 199
200int dirty_bytes_handler(struct ctl_table *table, int write, 200int dirty_bytes_handler(struct ctl_table *table, int write,
201 struct file *filp, void __user *buffer, size_t *lenp, 201 void __user *buffer, size_t *lenp,
202 loff_t *ppos) 202 loff_t *ppos)
203{ 203{
204 unsigned long old_bytes = vm_dirty_bytes; 204 unsigned long old_bytes = vm_dirty_bytes;
205 int ret; 205 int ret;
206 206
207 ret = proc_doulongvec_minmax(table, write, filp, buffer, lenp, ppos); 207 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
208 if (ret == 0 && write && vm_dirty_bytes != old_bytes) { 208 if (ret == 0 && write && vm_dirty_bytes != old_bytes) {
209 update_completion_period(); 209 update_completion_period();
210 vm_dirty_ratio = 0; 210 vm_dirty_ratio = 0;
@@ -690,9 +690,9 @@ static DEFINE_TIMER(laptop_mode_wb_timer, laptop_timer_fn, 0, 0);
690 * sysctl handler for /proc/sys/vm/dirty_writeback_centisecs 690 * sysctl handler for /proc/sys/vm/dirty_writeback_centisecs
691 */ 691 */
692int dirty_writeback_centisecs_handler(ctl_table *table, int write, 692int dirty_writeback_centisecs_handler(ctl_table *table, int write,
693 struct file *file, void __user *buffer, size_t *length, loff_t *ppos) 693 void __user *buffer, size_t *length, loff_t *ppos)
694{ 694{
695 proc_dointvec(table, write, file, buffer, length, ppos); 695 proc_dointvec(table, write, buffer, length, ppos);
696 return 0; 696 return 0;
697} 697}
698 698
@@ -1153,6 +1153,13 @@ int redirty_page_for_writepage(struct writeback_control *wbc, struct page *page)
1153EXPORT_SYMBOL(redirty_page_for_writepage); 1153EXPORT_SYMBOL(redirty_page_for_writepage);
1154 1154
1155/* 1155/*
1156 * Dirty a page.
1157 *
1158 * For pages with a mapping this should be done under the page lock
1159 * for the benefit of asynchronous memory errors who prefer a consistent
1160 * dirty state. This rule can be broken in some special cases,
1161 * but should be better not to.
1162 *
1156 * If the mapping doesn't provide a set_page_dirty a_op, then 1163 * If the mapping doesn't provide a set_page_dirty a_op, then
1157 * just fall through and assume that it wants buffer_heads. 1164 * just fall through and assume that it wants buffer_heads.
1158 */ 1165 */
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 5717f27a0704..bf720550b44d 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -234,6 +234,12 @@ static void bad_page(struct page *page)
234 static unsigned long nr_shown; 234 static unsigned long nr_shown;
235 static unsigned long nr_unshown; 235 static unsigned long nr_unshown;
236 236
237 /* Don't complain about poisoned pages */
238 if (PageHWPoison(page)) {
239 __ClearPageBuddy(page);
240 return;
241 }
242
237 /* 243 /*
238 * Allow a burst of 60 reports, then keep quiet for that minute; 244 * Allow a burst of 60 reports, then keep quiet for that minute;
239 * or allow a steady drip of one report per second. 245 * or allow a steady drip of one report per second.
@@ -666,7 +672,7 @@ static inline void expand(struct zone *zone, struct page *page,
666/* 672/*
667 * This page is about to be returned from the page allocator 673 * This page is about to be returned from the page allocator
668 */ 674 */
669static int prep_new_page(struct page *page, int order, gfp_t gfp_flags) 675static inline int check_new_page(struct page *page)
670{ 676{
671 if (unlikely(page_mapcount(page) | 677 if (unlikely(page_mapcount(page) |
672 (page->mapping != NULL) | 678 (page->mapping != NULL) |
@@ -675,6 +681,18 @@ static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
675 bad_page(page); 681 bad_page(page);
676 return 1; 682 return 1;
677 } 683 }
684 return 0;
685}
686
687static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
688{
689 int i;
690
691 for (i = 0; i < (1 << order); i++) {
692 struct page *p = page + i;
693 if (unlikely(check_new_page(p)))
694 return 1;
695 }
678 696
679 set_page_private(page, 0); 697 set_page_private(page, 0);
680 set_page_refcounted(page); 698 set_page_refcounted(page);
@@ -2373,7 +2391,7 @@ early_param("numa_zonelist_order", setup_numa_zonelist_order);
2373 * sysctl handler for numa_zonelist_order 2391 * sysctl handler for numa_zonelist_order
2374 */ 2392 */
2375int numa_zonelist_order_handler(ctl_table *table, int write, 2393int numa_zonelist_order_handler(ctl_table *table, int write,
2376 struct file *file, void __user *buffer, size_t *length, 2394 void __user *buffer, size_t *length,
2377 loff_t *ppos) 2395 loff_t *ppos)
2378{ 2396{
2379 char saved_string[NUMA_ZONELIST_ORDER_LEN]; 2397 char saved_string[NUMA_ZONELIST_ORDER_LEN];
@@ -2382,7 +2400,7 @@ int numa_zonelist_order_handler(ctl_table *table, int write,
2382 if (write) 2400 if (write)
2383 strncpy(saved_string, (char*)table->data, 2401 strncpy(saved_string, (char*)table->data,
2384 NUMA_ZONELIST_ORDER_LEN); 2402 NUMA_ZONELIST_ORDER_LEN);
2385 ret = proc_dostring(table, write, file, buffer, length, ppos); 2403 ret = proc_dostring(table, write, buffer, length, ppos);
2386 if (ret) 2404 if (ret)
2387 return ret; 2405 return ret;
2388 if (write) { 2406 if (write) {
@@ -4706,9 +4724,9 @@ module_init(init_per_zone_wmark_min)
4706 * changes. 4724 * changes.
4707 */ 4725 */
4708int min_free_kbytes_sysctl_handler(ctl_table *table, int write, 4726int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
4709 struct file *file, void __user *buffer, size_t *length, loff_t *ppos) 4727 void __user *buffer, size_t *length, loff_t *ppos)
4710{ 4728{
4711 proc_dointvec(table, write, file, buffer, length, ppos); 4729 proc_dointvec(table, write, buffer, length, ppos);
4712 if (write) 4730 if (write)
4713 setup_per_zone_wmarks(); 4731 setup_per_zone_wmarks();
4714 return 0; 4732 return 0;
@@ -4716,12 +4734,12 @@ int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
4716 4734
4717#ifdef CONFIG_NUMA 4735#ifdef CONFIG_NUMA
4718int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write, 4736int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
4719 struct file *file, void __user *buffer, size_t *length, loff_t *ppos) 4737 void __user *buffer, size_t *length, loff_t *ppos)
4720{ 4738{
4721 struct zone *zone; 4739 struct zone *zone;
4722 int rc; 4740 int rc;
4723 4741
4724 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos); 4742 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
4725 if (rc) 4743 if (rc)
4726 return rc; 4744 return rc;
4727 4745
@@ -4732,12 +4750,12 @@ int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
4732} 4750}
4733 4751
4734int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write, 4752int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
4735 struct file *file, void __user *buffer, size_t *length, loff_t *ppos) 4753 void __user *buffer, size_t *length, loff_t *ppos)
4736{ 4754{
4737 struct zone *zone; 4755 struct zone *zone;
4738 int rc; 4756 int rc;
4739 4757
4740 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos); 4758 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
4741 if (rc) 4759 if (rc)
4742 return rc; 4760 return rc;
4743 4761
@@ -4758,9 +4776,9 @@ int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
4758 * if in function of the boot time zone sizes. 4776 * if in function of the boot time zone sizes.
4759 */ 4777 */
4760int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write, 4778int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
4761 struct file *file, void __user *buffer, size_t *length, loff_t *ppos) 4779 void __user *buffer, size_t *length, loff_t *ppos)
4762{ 4780{
4763 proc_dointvec_minmax(table, write, file, buffer, length, ppos); 4781 proc_dointvec_minmax(table, write, buffer, length, ppos);
4764 setup_per_zone_lowmem_reserve(); 4782 setup_per_zone_lowmem_reserve();
4765 return 0; 4783 return 0;
4766} 4784}
@@ -4772,13 +4790,13 @@ int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
4772 */ 4790 */
4773 4791
4774int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write, 4792int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
4775 struct file *file, void __user *buffer, size_t *length, loff_t *ppos) 4793 void __user *buffer, size_t *length, loff_t *ppos)
4776{ 4794{
4777 struct zone *zone; 4795 struct zone *zone;
4778 unsigned int cpu; 4796 unsigned int cpu;
4779 int ret; 4797 int ret;
4780 4798
4781 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos); 4799 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
4782 if (!write || (ret == -EINVAL)) 4800 if (!write || (ret == -EINVAL))
4783 return ret; 4801 return ret;
4784 for_each_populated_zone(zone) { 4802 for_each_populated_zone(zone) {
diff --git a/mm/quicklist.c b/mm/quicklist.c
index 6eedf7e473d1..6633965bb27b 100644
--- a/mm/quicklist.c
+++ b/mm/quicklist.c
@@ -29,7 +29,6 @@ static unsigned long max_pages(unsigned long min_pages)
29 int node = numa_node_id(); 29 int node = numa_node_id();
30 struct zone *zones = NODE_DATA(node)->node_zones; 30 struct zone *zones = NODE_DATA(node)->node_zones;
31 int num_cpus_on_node; 31 int num_cpus_on_node;
32 const struct cpumask *cpumask_on_node = cpumask_of_node(node);
33 32
34 node_free_pages = 33 node_free_pages =
35#ifdef CONFIG_ZONE_DMA 34#ifdef CONFIG_ZONE_DMA
@@ -42,7 +41,7 @@ static unsigned long max_pages(unsigned long min_pages)
42 41
43 max = node_free_pages / FRACTION_OF_NODE_MEM; 42 max = node_free_pages / FRACTION_OF_NODE_MEM;
44 43
45 num_cpus_on_node = cpus_weight_nr(*cpumask_on_node); 44 num_cpus_on_node = cpumask_weight(cpumask_of_node(node));
46 max /= num_cpus_on_node; 45 max /= num_cpus_on_node;
47 46
48 return max(max, min_pages); 47 return max(max, min_pages);
diff --git a/mm/rmap.c b/mm/rmap.c
index 720fc03a7bc4..28aafe2b5306 100644
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -36,6 +36,11 @@
36 * mapping->tree_lock (widely used, in set_page_dirty, 36 * mapping->tree_lock (widely used, in set_page_dirty,
37 * in arch-dependent flush_dcache_mmap_lock, 37 * in arch-dependent flush_dcache_mmap_lock,
38 * within inode_lock in __sync_single_inode) 38 * within inode_lock in __sync_single_inode)
39 *
40 * (code doesn't rely on that order so it could be switched around)
41 * ->tasklist_lock
42 * anon_vma->lock (memory_failure, collect_procs_anon)
43 * pte map lock
39 */ 44 */
40 45
41#include <linux/mm.h> 46#include <linux/mm.h>
@@ -191,7 +196,7 @@ void __init anon_vma_init(void)
191 * Getting a lock on a stable anon_vma from a page off the LRU is 196 * Getting a lock on a stable anon_vma from a page off the LRU is
192 * tricky: page_lock_anon_vma rely on RCU to guard against the races. 197 * tricky: page_lock_anon_vma rely on RCU to guard against the races.
193 */ 198 */
194static struct anon_vma *page_lock_anon_vma(struct page *page) 199struct anon_vma *page_lock_anon_vma(struct page *page)
195{ 200{
196 struct anon_vma *anon_vma; 201 struct anon_vma *anon_vma;
197 unsigned long anon_mapping; 202 unsigned long anon_mapping;
@@ -211,7 +216,7 @@ out:
211 return NULL; 216 return NULL;
212} 217}
213 218
214static void page_unlock_anon_vma(struct anon_vma *anon_vma) 219void page_unlock_anon_vma(struct anon_vma *anon_vma)
215{ 220{
216 spin_unlock(&anon_vma->lock); 221 spin_unlock(&anon_vma->lock);
217 rcu_read_unlock(); 222 rcu_read_unlock();
@@ -311,7 +316,7 @@ pte_t *page_check_address(struct page *page, struct mm_struct *mm,
311 * if the page is not mapped into the page tables of this VMA. Only 316 * if the page is not mapped into the page tables of this VMA. Only
312 * valid for normal file or anonymous VMAs. 317 * valid for normal file or anonymous VMAs.
313 */ 318 */
314static int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma) 319int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma)
315{ 320{
316 unsigned long address; 321 unsigned long address;
317 pte_t *pte; 322 pte_t *pte;
@@ -756,7 +761,7 @@ void page_remove_rmap(struct page *page)
756 * repeatedly from either try_to_unmap_anon or try_to_unmap_file. 761 * repeatedly from either try_to_unmap_anon or try_to_unmap_file.
757 */ 762 */
758static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma, 763static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
759 int migration) 764 enum ttu_flags flags)
760{ 765{
761 struct mm_struct *mm = vma->vm_mm; 766 struct mm_struct *mm = vma->vm_mm;
762 unsigned long address; 767 unsigned long address;
@@ -778,11 +783,13 @@ static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
778 * If it's recently referenced (perhaps page_referenced 783 * If it's recently referenced (perhaps page_referenced
779 * skipped over this mm) then we should reactivate it. 784 * skipped over this mm) then we should reactivate it.
780 */ 785 */
781 if (!migration) { 786 if (!(flags & TTU_IGNORE_MLOCK)) {
782 if (vma->vm_flags & VM_LOCKED) { 787 if (vma->vm_flags & VM_LOCKED) {
783 ret = SWAP_MLOCK; 788 ret = SWAP_MLOCK;
784 goto out_unmap; 789 goto out_unmap;
785 } 790 }
791 }
792 if (!(flags & TTU_IGNORE_ACCESS)) {
786 if (ptep_clear_flush_young_notify(vma, address, pte)) { 793 if (ptep_clear_flush_young_notify(vma, address, pte)) {
787 ret = SWAP_FAIL; 794 ret = SWAP_FAIL;
788 goto out_unmap; 795 goto out_unmap;
@@ -800,7 +807,14 @@ static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
800 /* Update high watermark before we lower rss */ 807 /* Update high watermark before we lower rss */
801 update_hiwater_rss(mm); 808 update_hiwater_rss(mm);
802 809
803 if (PageAnon(page)) { 810 if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) {
811 if (PageAnon(page))
812 dec_mm_counter(mm, anon_rss);
813 else
814 dec_mm_counter(mm, file_rss);
815 set_pte_at(mm, address, pte,
816 swp_entry_to_pte(make_hwpoison_entry(page)));
817 } else if (PageAnon(page)) {
804 swp_entry_t entry = { .val = page_private(page) }; 818 swp_entry_t entry = { .val = page_private(page) };
805 819
806 if (PageSwapCache(page)) { 820 if (PageSwapCache(page)) {
@@ -822,12 +836,12 @@ static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
822 * pte. do_swap_page() will wait until the migration 836 * pte. do_swap_page() will wait until the migration
823 * pte is removed and then restart fault handling. 837 * pte is removed and then restart fault handling.
824 */ 838 */
825 BUG_ON(!migration); 839 BUG_ON(TTU_ACTION(flags) != TTU_MIGRATION);
826 entry = make_migration_entry(page, pte_write(pteval)); 840 entry = make_migration_entry(page, pte_write(pteval));
827 } 841 }
828 set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); 842 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
829 BUG_ON(pte_file(*pte)); 843 BUG_ON(pte_file(*pte));
830 } else if (PAGE_MIGRATION && migration) { 844 } else if (PAGE_MIGRATION && (TTU_ACTION(flags) == TTU_MIGRATION)) {
831 /* Establish migration entry for a file page */ 845 /* Establish migration entry for a file page */
832 swp_entry_t entry; 846 swp_entry_t entry;
833 entry = make_migration_entry(page, pte_write(pteval)); 847 entry = make_migration_entry(page, pte_write(pteval));
@@ -996,12 +1010,13 @@ static int try_to_mlock_page(struct page *page, struct vm_area_struct *vma)
996 * vm_flags for that VMA. That should be OK, because that vma shouldn't be 1010 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
997 * 'LOCKED. 1011 * 'LOCKED.
998 */ 1012 */
999static int try_to_unmap_anon(struct page *page, int unlock, int migration) 1013static int try_to_unmap_anon(struct page *page, enum ttu_flags flags)
1000{ 1014{
1001 struct anon_vma *anon_vma; 1015 struct anon_vma *anon_vma;
1002 struct vm_area_struct *vma; 1016 struct vm_area_struct *vma;
1003 unsigned int mlocked = 0; 1017 unsigned int mlocked = 0;
1004 int ret = SWAP_AGAIN; 1018 int ret = SWAP_AGAIN;
1019 int unlock = TTU_ACTION(flags) == TTU_MUNLOCK;
1005 1020
1006 if (MLOCK_PAGES && unlikely(unlock)) 1021 if (MLOCK_PAGES && unlikely(unlock))
1007 ret = SWAP_SUCCESS; /* default for try_to_munlock() */ 1022 ret = SWAP_SUCCESS; /* default for try_to_munlock() */
@@ -1017,7 +1032,7 @@ static int try_to_unmap_anon(struct page *page, int unlock, int migration)
1017 continue; /* must visit all unlocked vmas */ 1032 continue; /* must visit all unlocked vmas */
1018 ret = SWAP_MLOCK; /* saw at least one mlocked vma */ 1033 ret = SWAP_MLOCK; /* saw at least one mlocked vma */
1019 } else { 1034 } else {
1020 ret = try_to_unmap_one(page, vma, migration); 1035 ret = try_to_unmap_one(page, vma, flags);
1021 if (ret == SWAP_FAIL || !page_mapped(page)) 1036 if (ret == SWAP_FAIL || !page_mapped(page))
1022 break; 1037 break;
1023 } 1038 }
@@ -1041,8 +1056,7 @@ static int try_to_unmap_anon(struct page *page, int unlock, int migration)
1041/** 1056/**
1042 * try_to_unmap_file - unmap/unlock file page using the object-based rmap method 1057 * try_to_unmap_file - unmap/unlock file page using the object-based rmap method
1043 * @page: the page to unmap/unlock 1058 * @page: the page to unmap/unlock
1044 * @unlock: request for unlock rather than unmap [unlikely] 1059 * @flags: action and flags
1045 * @migration: unmapping for migration - ignored if @unlock
1046 * 1060 *
1047 * Find all the mappings of a page using the mapping pointer and the vma chains 1061 * Find all the mappings of a page using the mapping pointer and the vma chains
1048 * contained in the address_space struct it points to. 1062 * contained in the address_space struct it points to.
@@ -1054,7 +1068,7 @@ static int try_to_unmap_anon(struct page *page, int unlock, int migration)
1054 * vm_flags for that VMA. That should be OK, because that vma shouldn't be 1068 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1055 * 'LOCKED. 1069 * 'LOCKED.
1056 */ 1070 */
1057static int try_to_unmap_file(struct page *page, int unlock, int migration) 1071static int try_to_unmap_file(struct page *page, enum ttu_flags flags)
1058{ 1072{
1059 struct address_space *mapping = page->mapping; 1073 struct address_space *mapping = page->mapping;
1060 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); 1074 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
@@ -1066,6 +1080,7 @@ static int try_to_unmap_file(struct page *page, int unlock, int migration)
1066 unsigned long max_nl_size = 0; 1080 unsigned long max_nl_size = 0;
1067 unsigned int mapcount; 1081 unsigned int mapcount;
1068 unsigned int mlocked = 0; 1082 unsigned int mlocked = 0;
1083 int unlock = TTU_ACTION(flags) == TTU_MUNLOCK;
1069 1084
1070 if (MLOCK_PAGES && unlikely(unlock)) 1085 if (MLOCK_PAGES && unlikely(unlock))
1071 ret = SWAP_SUCCESS; /* default for try_to_munlock() */ 1086 ret = SWAP_SUCCESS; /* default for try_to_munlock() */
@@ -1078,7 +1093,7 @@ static int try_to_unmap_file(struct page *page, int unlock, int migration)
1078 continue; /* must visit all vmas */ 1093 continue; /* must visit all vmas */
1079 ret = SWAP_MLOCK; 1094 ret = SWAP_MLOCK;
1080 } else { 1095 } else {
1081 ret = try_to_unmap_one(page, vma, migration); 1096 ret = try_to_unmap_one(page, vma, flags);
1082 if (ret == SWAP_FAIL || !page_mapped(page)) 1097 if (ret == SWAP_FAIL || !page_mapped(page))
1083 goto out; 1098 goto out;
1084 } 1099 }
@@ -1103,7 +1118,8 @@ static int try_to_unmap_file(struct page *page, int unlock, int migration)
1103 ret = SWAP_MLOCK; /* leave mlocked == 0 */ 1118 ret = SWAP_MLOCK; /* leave mlocked == 0 */
1104 goto out; /* no need to look further */ 1119 goto out; /* no need to look further */
1105 } 1120 }
1106 if (!MLOCK_PAGES && !migration && (vma->vm_flags & VM_LOCKED)) 1121 if (!MLOCK_PAGES && !(flags & TTU_IGNORE_MLOCK) &&
1122 (vma->vm_flags & VM_LOCKED))
1107 continue; 1123 continue;
1108 cursor = (unsigned long) vma->vm_private_data; 1124 cursor = (unsigned long) vma->vm_private_data;
1109 if (cursor > max_nl_cursor) 1125 if (cursor > max_nl_cursor)
@@ -1137,7 +1153,7 @@ static int try_to_unmap_file(struct page *page, int unlock, int migration)
1137 do { 1153 do {
1138 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, 1154 list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
1139 shared.vm_set.list) { 1155 shared.vm_set.list) {
1140 if (!MLOCK_PAGES && !migration && 1156 if (!MLOCK_PAGES && !(flags & TTU_IGNORE_MLOCK) &&
1141 (vma->vm_flags & VM_LOCKED)) 1157 (vma->vm_flags & VM_LOCKED))
1142 continue; 1158 continue;
1143 cursor = (unsigned long) vma->vm_private_data; 1159 cursor = (unsigned long) vma->vm_private_data;
@@ -1177,7 +1193,7 @@ out:
1177/** 1193/**
1178 * try_to_unmap - try to remove all page table mappings to a page 1194 * try_to_unmap - try to remove all page table mappings to a page
1179 * @page: the page to get unmapped 1195 * @page: the page to get unmapped
1180 * @migration: migration flag 1196 * @flags: action and flags
1181 * 1197 *
1182 * Tries to remove all the page table entries which are mapping this 1198 * Tries to remove all the page table entries which are mapping this
1183 * page, used in the pageout path. Caller must hold the page lock. 1199 * page, used in the pageout path. Caller must hold the page lock.
@@ -1188,16 +1204,16 @@ out:
1188 * SWAP_FAIL - the page is unswappable 1204 * SWAP_FAIL - the page is unswappable
1189 * SWAP_MLOCK - page is mlocked. 1205 * SWAP_MLOCK - page is mlocked.
1190 */ 1206 */
1191int try_to_unmap(struct page *page, int migration) 1207int try_to_unmap(struct page *page, enum ttu_flags flags)
1192{ 1208{
1193 int ret; 1209 int ret;
1194 1210
1195 BUG_ON(!PageLocked(page)); 1211 BUG_ON(!PageLocked(page));
1196 1212
1197 if (PageAnon(page)) 1213 if (PageAnon(page))
1198 ret = try_to_unmap_anon(page, 0, migration); 1214 ret = try_to_unmap_anon(page, flags);
1199 else 1215 else
1200 ret = try_to_unmap_file(page, 0, migration); 1216 ret = try_to_unmap_file(page, flags);
1201 if (ret != SWAP_MLOCK && !page_mapped(page)) 1217 if (ret != SWAP_MLOCK && !page_mapped(page))
1202 ret = SWAP_SUCCESS; 1218 ret = SWAP_SUCCESS;
1203 return ret; 1219 return ret;
@@ -1222,8 +1238,8 @@ int try_to_munlock(struct page *page)
1222 VM_BUG_ON(!PageLocked(page) || PageLRU(page)); 1238 VM_BUG_ON(!PageLocked(page) || PageLRU(page));
1223 1239
1224 if (PageAnon(page)) 1240 if (PageAnon(page))
1225 return try_to_unmap_anon(page, 1, 0); 1241 return try_to_unmap_anon(page, TTU_MUNLOCK);
1226 else 1242 else
1227 return try_to_unmap_file(page, 1, 0); 1243 return try_to_unmap_file(page, TTU_MUNLOCK);
1228} 1244}
1229 1245
diff --git a/mm/shmem.c b/mm/shmem.c
index aa9481166aae..ccf446a9faa1 100644
--- a/mm/shmem.c
+++ b/mm/shmem.c
@@ -1634,8 +1634,8 @@ shmem_write_end(struct file *file, struct address_space *mapping,
1634 if (pos + copied > inode->i_size) 1634 if (pos + copied > inode->i_size)
1635 i_size_write(inode, pos + copied); 1635 i_size_write(inode, pos + copied);
1636 1636
1637 unlock_page(page);
1638 set_page_dirty(page); 1637 set_page_dirty(page);
1638 unlock_page(page);
1639 page_cache_release(page); 1639 page_cache_release(page);
1640 1640
1641 return copied; 1641 return copied;
@@ -1972,13 +1972,13 @@ static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *s
1972 iput(inode); 1972 iput(inode);
1973 return error; 1973 return error;
1974 } 1974 }
1975 unlock_page(page);
1976 inode->i_mapping->a_ops = &shmem_aops; 1975 inode->i_mapping->a_ops = &shmem_aops;
1977 inode->i_op = &shmem_symlink_inode_operations; 1976 inode->i_op = &shmem_symlink_inode_operations;
1978 kaddr = kmap_atomic(page, KM_USER0); 1977 kaddr = kmap_atomic(page, KM_USER0);
1979 memcpy(kaddr, symname, len); 1978 memcpy(kaddr, symname, len);
1980 kunmap_atomic(kaddr, KM_USER0); 1979 kunmap_atomic(kaddr, KM_USER0);
1981 set_page_dirty(page); 1980 set_page_dirty(page);
1981 unlock_page(page);
1982 page_cache_release(page); 1982 page_cache_release(page);
1983 } 1983 }
1984 if (dir->i_mode & S_ISGID) 1984 if (dir->i_mode & S_ISGID)
@@ -2421,6 +2421,7 @@ static const struct address_space_operations shmem_aops = {
2421 .write_end = shmem_write_end, 2421 .write_end = shmem_write_end,
2422#endif 2422#endif
2423 .migratepage = migrate_page, 2423 .migratepage = migrate_page,
2424 .error_remove_page = generic_error_remove_page,
2424}; 2425};
2425 2426
2426static const struct file_operations shmem_file_operations = { 2427static const struct file_operations shmem_file_operations = {
diff --git a/mm/swapfile.c b/mm/swapfile.c
index f1bf19daadc6..4de7f02f820b 100644
--- a/mm/swapfile.c
+++ b/mm/swapfile.c
@@ -699,7 +699,7 @@ int free_swap_and_cache(swp_entry_t entry)
699 struct swap_info_struct *p; 699 struct swap_info_struct *p;
700 struct page *page = NULL; 700 struct page *page = NULL;
701 701
702 if (is_migration_entry(entry)) 702 if (non_swap_entry(entry))
703 return 1; 703 return 1;
704 704
705 p = swap_info_get(entry); 705 p = swap_info_get(entry);
@@ -2085,7 +2085,7 @@ static int __swap_duplicate(swp_entry_t entry, bool cache)
2085 int count; 2085 int count;
2086 bool has_cache; 2086 bool has_cache;
2087 2087
2088 if (is_migration_entry(entry)) 2088 if (non_swap_entry(entry))
2089 return -EINVAL; 2089 return -EINVAL;
2090 2090
2091 type = swp_type(entry); 2091 type = swp_type(entry);
diff --git a/mm/truncate.c b/mm/truncate.c
index ccc3ecf7cb98..450cebdabfc0 100644
--- a/mm/truncate.c
+++ b/mm/truncate.c
@@ -93,11 +93,11 @@ EXPORT_SYMBOL(cancel_dirty_page);
93 * its lock, b) when a concurrent invalidate_mapping_pages got there first and 93 * its lock, b) when a concurrent invalidate_mapping_pages got there first and
94 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. 94 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
95 */ 95 */
96static void 96static int
97truncate_complete_page(struct address_space *mapping, struct page *page) 97truncate_complete_page(struct address_space *mapping, struct page *page)
98{ 98{
99 if (page->mapping != mapping) 99 if (page->mapping != mapping)
100 return; 100 return -EIO;
101 101
102 if (page_has_private(page)) 102 if (page_has_private(page))
103 do_invalidatepage(page, 0); 103 do_invalidatepage(page, 0);
@@ -108,6 +108,7 @@ truncate_complete_page(struct address_space *mapping, struct page *page)
108 remove_from_page_cache(page); 108 remove_from_page_cache(page);
109 ClearPageMappedToDisk(page); 109 ClearPageMappedToDisk(page);
110 page_cache_release(page); /* pagecache ref */ 110 page_cache_release(page); /* pagecache ref */
111 return 0;
111} 112}
112 113
113/* 114/*
@@ -135,6 +136,51 @@ invalidate_complete_page(struct address_space *mapping, struct page *page)
135 return ret; 136 return ret;
136} 137}
137 138
139int truncate_inode_page(struct address_space *mapping, struct page *page)
140{
141 if (page_mapped(page)) {
142 unmap_mapping_range(mapping,
143 (loff_t)page->index << PAGE_CACHE_SHIFT,
144 PAGE_CACHE_SIZE, 0);
145 }
146 return truncate_complete_page(mapping, page);
147}
148
149/*
150 * Used to get rid of pages on hardware memory corruption.
151 */
152int generic_error_remove_page(struct address_space *mapping, struct page *page)
153{
154 if (!mapping)
155 return -EINVAL;
156 /*
157 * Only punch for normal data pages for now.
158 * Handling other types like directories would need more auditing.
159 */
160 if (!S_ISREG(mapping->host->i_mode))
161 return -EIO;
162 return truncate_inode_page(mapping, page);
163}
164EXPORT_SYMBOL(generic_error_remove_page);
165
166/*
167 * Safely invalidate one page from its pagecache mapping.
168 * It only drops clean, unused pages. The page must be locked.
169 *
170 * Returns 1 if the page is successfully invalidated, otherwise 0.
171 */
172int invalidate_inode_page(struct page *page)
173{
174 struct address_space *mapping = page_mapping(page);
175 if (!mapping)
176 return 0;
177 if (PageDirty(page) || PageWriteback(page))
178 return 0;
179 if (page_mapped(page))
180 return 0;
181 return invalidate_complete_page(mapping, page);
182}
183
138/** 184/**
139 * truncate_inode_pages - truncate range of pages specified by start & end byte offsets 185 * truncate_inode_pages - truncate range of pages specified by start & end byte offsets
140 * @mapping: mapping to truncate 186 * @mapping: mapping to truncate
@@ -196,12 +242,7 @@ void truncate_inode_pages_range(struct address_space *mapping,
196 unlock_page(page); 242 unlock_page(page);
197 continue; 243 continue;
198 } 244 }
199 if (page_mapped(page)) { 245 truncate_inode_page(mapping, page);
200 unmap_mapping_range(mapping,
201 (loff_t)page_index<<PAGE_CACHE_SHIFT,
202 PAGE_CACHE_SIZE, 0);
203 }
204 truncate_complete_page(mapping, page);
205 unlock_page(page); 246 unlock_page(page);
206 } 247 }
207 pagevec_release(&pvec); 248 pagevec_release(&pvec);
@@ -238,15 +279,10 @@ void truncate_inode_pages_range(struct address_space *mapping,
238 break; 279 break;
239 lock_page(page); 280 lock_page(page);
240 wait_on_page_writeback(page); 281 wait_on_page_writeback(page);
241 if (page_mapped(page)) { 282 truncate_inode_page(mapping, page);
242 unmap_mapping_range(mapping,
243 (loff_t)page->index<<PAGE_CACHE_SHIFT,
244 PAGE_CACHE_SIZE, 0);
245 }
246 if (page->index > next) 283 if (page->index > next)
247 next = page->index; 284 next = page->index;
248 next++; 285 next++;
249 truncate_complete_page(mapping, page);
250 unlock_page(page); 286 unlock_page(page);
251 } 287 }
252 pagevec_release(&pvec); 288 pagevec_release(&pvec);
@@ -311,12 +347,8 @@ unsigned long invalidate_mapping_pages(struct address_space *mapping,
311 if (lock_failed) 347 if (lock_failed)
312 continue; 348 continue;
313 349
314 if (PageDirty(page) || PageWriteback(page)) 350 ret += invalidate_inode_page(page);
315 goto unlock; 351
316 if (page_mapped(page))
317 goto unlock;
318 ret += invalidate_complete_page(mapping, page);
319unlock:
320 unlock_page(page); 352 unlock_page(page);
321 if (next > end) 353 if (next > end)
322 break; 354 break;
@@ -465,3 +497,67 @@ int invalidate_inode_pages2(struct address_space *mapping)
465 return invalidate_inode_pages2_range(mapping, 0, -1); 497 return invalidate_inode_pages2_range(mapping, 0, -1);
466} 498}
467EXPORT_SYMBOL_GPL(invalidate_inode_pages2); 499EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
500
501/**
502 * truncate_pagecache - unmap and remove pagecache that has been truncated
503 * @inode: inode
504 * @old: old file offset
505 * @new: new file offset
506 *
507 * inode's new i_size must already be written before truncate_pagecache
508 * is called.
509 *
510 * This function should typically be called before the filesystem
511 * releases resources associated with the freed range (eg. deallocates
512 * blocks). This way, pagecache will always stay logically coherent
513 * with on-disk format, and the filesystem would not have to deal with
514 * situations such as writepage being called for a page that has already
515 * had its underlying blocks deallocated.
516 */
517void truncate_pagecache(struct inode *inode, loff_t old, loff_t new)
518{
519 if (new < old) {
520 struct address_space *mapping = inode->i_mapping;
521
522 /*
523 * unmap_mapping_range is called twice, first simply for
524 * efficiency so that truncate_inode_pages does fewer
525 * single-page unmaps. However after this first call, and
526 * before truncate_inode_pages finishes, it is possible for
527 * private pages to be COWed, which remain after
528 * truncate_inode_pages finishes, hence the second
529 * unmap_mapping_range call must be made for correctness.
530 */
531 unmap_mapping_range(mapping, new + PAGE_SIZE - 1, 0, 1);
532 truncate_inode_pages(mapping, new);
533 unmap_mapping_range(mapping, new + PAGE_SIZE - 1, 0, 1);
534 }
535}
536EXPORT_SYMBOL(truncate_pagecache);
537
538/**
539 * vmtruncate - unmap mappings "freed" by truncate() syscall
540 * @inode: inode of the file used
541 * @offset: file offset to start truncating
542 *
543 * NOTE! We have to be ready to update the memory sharing
544 * between the file and the memory map for a potential last
545 * incomplete page. Ugly, but necessary.
546 */
547int vmtruncate(struct inode *inode, loff_t offset)
548{
549 loff_t oldsize;
550 int error;
551
552 error = inode_newsize_ok(inode, offset);
553 if (error)
554 return error;
555 oldsize = inode->i_size;
556 i_size_write(inode, offset);
557 truncate_pagecache(inode, oldsize, offset);
558 if (inode->i_op->truncate)
559 inode->i_op->truncate(inode);
560
561 return error;
562}
563EXPORT_SYMBOL(vmtruncate);
diff --git a/mm/vmalloc.c b/mm/vmalloc.c
index 5535da1d6961..69511e663234 100644
--- a/mm/vmalloc.c
+++ b/mm/vmalloc.c
@@ -184,7 +184,7 @@ static int vmap_page_range(unsigned long start, unsigned long end,
184 return ret; 184 return ret;
185} 185}
186 186
187static inline int is_vmalloc_or_module_addr(const void *x) 187int is_vmalloc_or_module_addr(const void *x)
188{ 188{
189 /* 189 /*
190 * ARM, x86-64 and sparc64 put modules in a special place, 190 * ARM, x86-64 and sparc64 put modules in a special place,
diff --git a/mm/vmscan.c b/mm/vmscan.c
index 359c3c57ef85..64e438898832 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -663,7 +663,7 @@ static unsigned long shrink_page_list(struct list_head *page_list,
663 * processes. Try to unmap it here. 663 * processes. Try to unmap it here.
664 */ 664 */
665 if (page_mapped(page) && mapping) { 665 if (page_mapped(page) && mapping) {
666 switch (try_to_unmap(page, 0)) { 666 switch (try_to_unmap(page, TTU_UNMAP)) {
667 case SWAP_FAIL: 667 case SWAP_FAIL:
668 goto activate_locked; 668 goto activate_locked;
669 case SWAP_AGAIN: 669 case SWAP_AGAIN:
@@ -1836,11 +1836,45 @@ unsigned long try_to_free_pages(struct zonelist *zonelist, int order,
1836 1836
1837#ifdef CONFIG_CGROUP_MEM_RES_CTLR 1837#ifdef CONFIG_CGROUP_MEM_RES_CTLR
1838 1838
1839unsigned long mem_cgroup_shrink_node_zone(struct mem_cgroup *mem,
1840 gfp_t gfp_mask, bool noswap,
1841 unsigned int swappiness,
1842 struct zone *zone, int nid)
1843{
1844 struct scan_control sc = {
1845 .may_writepage = !laptop_mode,
1846 .may_unmap = 1,
1847 .may_swap = !noswap,
1848 .swap_cluster_max = SWAP_CLUSTER_MAX,
1849 .swappiness = swappiness,
1850 .order = 0,
1851 .mem_cgroup = mem,
1852 .isolate_pages = mem_cgroup_isolate_pages,
1853 };
1854 nodemask_t nm = nodemask_of_node(nid);
1855
1856 sc.gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) |
1857 (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK);
1858 sc.nodemask = &nm;
1859 sc.nr_reclaimed = 0;
1860 sc.nr_scanned = 0;
1861 /*
1862 * NOTE: Although we can get the priority field, using it
1863 * here is not a good idea, since it limits the pages we can scan.
1864 * if we don't reclaim here, the shrink_zone from balance_pgdat
1865 * will pick up pages from other mem cgroup's as well. We hack
1866 * the priority and make it zero.
1867 */
1868 shrink_zone(0, zone, &sc);
1869 return sc.nr_reclaimed;
1870}
1871
1839unsigned long try_to_free_mem_cgroup_pages(struct mem_cgroup *mem_cont, 1872unsigned long try_to_free_mem_cgroup_pages(struct mem_cgroup *mem_cont,
1840 gfp_t gfp_mask, 1873 gfp_t gfp_mask,
1841 bool noswap, 1874 bool noswap,
1842 unsigned int swappiness) 1875 unsigned int swappiness)
1843{ 1876{
1877 struct zonelist *zonelist;
1844 struct scan_control sc = { 1878 struct scan_control sc = {
1845 .may_writepage = !laptop_mode, 1879 .may_writepage = !laptop_mode,
1846 .may_unmap = 1, 1880 .may_unmap = 1,
@@ -1852,7 +1886,6 @@ unsigned long try_to_free_mem_cgroup_pages(struct mem_cgroup *mem_cont,
1852 .isolate_pages = mem_cgroup_isolate_pages, 1886 .isolate_pages = mem_cgroup_isolate_pages,
1853 .nodemask = NULL, /* we don't care the placement */ 1887 .nodemask = NULL, /* we don't care the placement */
1854 }; 1888 };
1855 struct zonelist *zonelist;
1856 1889
1857 sc.gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) | 1890 sc.gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) |
1858 (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK); 1891 (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK);
@@ -1974,6 +2007,7 @@ loop_again:
1974 for (i = 0; i <= end_zone; i++) { 2007 for (i = 0; i <= end_zone; i++) {
1975 struct zone *zone = pgdat->node_zones + i; 2008 struct zone *zone = pgdat->node_zones + i;
1976 int nr_slab; 2009 int nr_slab;
2010 int nid, zid;
1977 2011
1978 if (!populated_zone(zone)) 2012 if (!populated_zone(zone))
1979 continue; 2013 continue;
@@ -1988,6 +2022,15 @@ loop_again:
1988 temp_priority[i] = priority; 2022 temp_priority[i] = priority;
1989 sc.nr_scanned = 0; 2023 sc.nr_scanned = 0;
1990 note_zone_scanning_priority(zone, priority); 2024 note_zone_scanning_priority(zone, priority);
2025
2026 nid = pgdat->node_id;
2027 zid = zone_idx(zone);
2028 /*
2029 * Call soft limit reclaim before calling shrink_zone.
2030 * For now we ignore the return value
2031 */
2032 mem_cgroup_soft_limit_reclaim(zone, order, sc.gfp_mask,
2033 nid, zid);
1991 /* 2034 /*
1992 * We put equal pressure on every zone, unless one 2035 * We put equal pressure on every zone, unless one
1993 * zone has way too many pages free already. 2036 * zone has way too many pages free already.
@@ -2801,10 +2844,10 @@ static void scan_all_zones_unevictable_pages(void)
2801unsigned long scan_unevictable_pages; 2844unsigned long scan_unevictable_pages;
2802 2845
2803int scan_unevictable_handler(struct ctl_table *table, int write, 2846int scan_unevictable_handler(struct ctl_table *table, int write,
2804 struct file *file, void __user *buffer, 2847 void __user *buffer,
2805 size_t *length, loff_t *ppos) 2848 size_t *length, loff_t *ppos)
2806{ 2849{
2807 proc_doulongvec_minmax(table, write, file, buffer, length, ppos); 2850 proc_doulongvec_minmax(table, write, buffer, length, ppos);
2808 2851
2809 if (write && *(unsigned long *)table->data) 2852 if (write && *(unsigned long *)table->data)
2810 scan_all_zones_unevictable_pages(); 2853 scan_all_zones_unevictable_pages();
generated by cgit v1.2.2 (git 2.25.0) at 2025-01-20 05:25:47 -0500