2 * page-types: Tool for querying page flags
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms of the GNU General Public License as published by the Free
6 * Software Foundation; version 2.
8 * This program is distributed in the hope that it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * You should find a copy of v2 of the GNU General Public License somewhere on
14 * your Linux system; if not, write to the Free Software Foundation, Inc., 59
15 * Temple Place, Suite 330, Boston, MA 02111-1307 USA.
17 * Copyright (C) 2009 Intel corporation
19 * Authors: Wu Fengguang <fengguang.wu@intel.com>
22 #define _LARGEFILE64_SOURCE
32 #include <sys/types.h>
33 #include <sys/errno.h>
34 #include <sys/fcntl.h>
35 #include <sys/mount.h>
36 #include <sys/statfs.h>
37 #include "../../include/uapi/linux/magic.h"
38 #include "../../include/uapi/linux/kernel-page-flags.h"
39 #include <api/fs/debugfs.h>
47 # define STR(x) _STR(x)
51 * pagemap kernel ABI bits
54 #define PM_ENTRY_BYTES sizeof(uint64_t)
55 #define PM_STATUS_BITS 3
56 #define PM_STATUS_OFFSET (64 - PM_STATUS_BITS)
57 #define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
58 #define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
59 #define PM_PSHIFT_BITS 6
60 #define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
61 #define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
62 #define __PM_PSHIFT(x) (((uint64_t) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
63 #define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
64 #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
66 #define __PM_SOFT_DIRTY (1LL)
67 #define PM_PRESENT PM_STATUS(4LL)
68 #define PM_SWAP PM_STATUS(2LL)
69 #define PM_SOFT_DIRTY __PM_PSHIFT(__PM_SOFT_DIRTY)
77 #define PROC_KPAGEFLAGS "/proc/kpageflags"
79 /* [32-] kernel hacking assistances */
80 #define KPF_RESERVED 32
81 #define KPF_MLOCKED 33
82 #define KPF_MAPPEDTODISK 34
83 #define KPF_PRIVATE 35
84 #define KPF_PRIVATE_2 36
85 #define KPF_OWNER_PRIVATE 37
87 #define KPF_UNCACHED 39
88 #define KPF_SOFTDIRTY 40
90 /* [48-] take some arbitrary free slots for expanding overloaded flags
91 * not part of kernel API
93 #define KPF_READAHEAD 48
94 #define KPF_SLOB_FREE 49
95 #define KPF_SLUB_FROZEN 50
96 #define KPF_SLUB_DEBUG 51
98 #define KPF_ALL_BITS ((uint64_t)~0ULL)
99 #define KPF_HACKERS_BITS (0xffffULL << 32)
100 #define KPF_OVERLOADED_BITS (0xffffULL << 48)
101 #define BIT(name) (1ULL << KPF_##name)
102 #define BITS_COMPOUND (BIT(COMPOUND_HEAD) | BIT(COMPOUND_TAIL))
104 static const char * const page_flag_names
[] = {
105 [KPF_LOCKED
] = "L:locked",
106 [KPF_ERROR
] = "E:error",
107 [KPF_REFERENCED
] = "R:referenced",
108 [KPF_UPTODATE
] = "U:uptodate",
109 [KPF_DIRTY
] = "D:dirty",
111 [KPF_ACTIVE
] = "A:active",
112 [KPF_SLAB
] = "S:slab",
113 [KPF_WRITEBACK
] = "W:writeback",
114 [KPF_RECLAIM
] = "I:reclaim",
115 [KPF_BUDDY
] = "B:buddy",
117 [KPF_MMAP
] = "M:mmap",
118 [KPF_ANON
] = "a:anonymous",
119 [KPF_SWAPCACHE
] = "s:swapcache",
120 [KPF_SWAPBACKED
] = "b:swapbacked",
121 [KPF_COMPOUND_HEAD
] = "H:compound_head",
122 [KPF_COMPOUND_TAIL
] = "T:compound_tail",
123 [KPF_HUGE
] = "G:huge",
124 [KPF_UNEVICTABLE
] = "u:unevictable",
125 [KPF_HWPOISON
] = "X:hwpoison",
126 [KPF_NOPAGE
] = "n:nopage",
130 [KPF_RESERVED
] = "r:reserved",
131 [KPF_MLOCKED
] = "m:mlocked",
132 [KPF_MAPPEDTODISK
] = "d:mappedtodisk",
133 [KPF_PRIVATE
] = "P:private",
134 [KPF_PRIVATE_2
] = "p:private_2",
135 [KPF_OWNER_PRIVATE
] = "O:owner_private",
136 [KPF_ARCH
] = "h:arch",
137 [KPF_UNCACHED
] = "c:uncached",
138 [KPF_SOFTDIRTY
] = "f:softdirty",
140 [KPF_READAHEAD
] = "I:readahead",
141 [KPF_SLOB_FREE
] = "P:slob_free",
142 [KPF_SLUB_FROZEN
] = "A:slub_frozen",
143 [KPF_SLUB_DEBUG
] = "E:slub_debug",
147 static const char * const debugfs_known_mountpoints
[] = {
157 static int opt_raw
; /* for kernel developers */
158 static int opt_list
; /* list pages (in ranges) */
159 static int opt_no_summary
; /* don't show summary */
160 static pid_t opt_pid
; /* process to walk */
162 #define MAX_ADDR_RANGES 1024
163 static int nr_addr_ranges
;
164 static unsigned long opt_offset
[MAX_ADDR_RANGES
];
165 static unsigned long opt_size
[MAX_ADDR_RANGES
];
167 #define MAX_VMAS 10240
169 static unsigned long pg_start
[MAX_VMAS
];
170 static unsigned long pg_end
[MAX_VMAS
];
172 #define MAX_BIT_FILTERS 64
173 static int nr_bit_filters
;
174 static uint64_t opt_mask
[MAX_BIT_FILTERS
];
175 static uint64_t opt_bits
[MAX_BIT_FILTERS
];
177 static int page_size
;
179 static int pagemap_fd
;
180 static int kpageflags_fd
;
182 static int opt_hwpoison
;
183 static int opt_unpoison
;
185 static char *hwpoison_debug_fs
;
186 static int hwpoison_inject_fd
;
187 static int hwpoison_forget_fd
;
189 #define HASH_SHIFT 13
190 #define HASH_SIZE (1 << HASH_SHIFT)
191 #define HASH_MASK (HASH_SIZE - 1)
192 #define HASH_KEY(flags) (flags & HASH_MASK)
194 static unsigned long total_pages
;
195 static unsigned long nr_pages
[HASH_SIZE
];
196 static uint64_t page_flags
[HASH_SIZE
];
203 #define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0]))
205 #define min_t(type, x, y) ({ \
208 __min1 < __min2 ? __min1 : __min2; })
210 #define max_t(type, x, y) ({ \
213 __max1 > __max2 ? __max1 : __max2; })
215 static unsigned long pages2mb(unsigned long pages
)
217 return (pages
* page_size
) >> 20;
220 static void fatal(const char *x
, ...)
225 vfprintf(stderr
, x
, ap
);
230 static int checked_open(const char *pathname
, int flags
)
232 int fd
= open(pathname
, flags
);
243 * pagemap/kpageflags routines
246 static unsigned long do_u64_read(int fd
, char *name
,
253 if (index
> ULONG_MAX
/ 8)
254 fatal("index overflow: %lu\n", index
);
256 if (lseek(fd
, index
* 8, SEEK_SET
) < 0) {
261 bytes
= read(fd
, buf
, count
* 8);
267 fatal("partial read: %lu bytes\n", bytes
);
272 static unsigned long kpageflags_read(uint64_t *buf
,
276 return do_u64_read(kpageflags_fd
, PROC_KPAGEFLAGS
, buf
, index
, pages
);
279 static unsigned long pagemap_read(uint64_t *buf
,
283 return do_u64_read(pagemap_fd
, "/proc/pid/pagemap", buf
, index
, pages
);
286 static unsigned long pagemap_pfn(uint64_t val
)
290 if (val
& PM_PRESENT
)
291 pfn
= PM_PFRAME(val
);
303 static char *page_flag_name(uint64_t flags
)
309 for (i
= 0, j
= 0; i
< ARRAY_SIZE(page_flag_names
); i
++) {
310 present
= (flags
>> i
) & 1;
311 if (!page_flag_names
[i
]) {
313 fatal("unknown flag bit %d\n", i
);
316 buf
[j
++] = present
? page_flag_names
[i
][0] : '_';
322 static char *page_flag_longname(uint64_t flags
)
324 static char buf
[1024];
327 for (i
= 0, n
= 0; i
< ARRAY_SIZE(page_flag_names
); i
++) {
328 if (!page_flag_names
[i
])
330 if ((flags
>> i
) & 1)
331 n
+= snprintf(buf
+ n
, sizeof(buf
) - n
, "%s,",
332 page_flag_names
[i
] + 2);
343 * page list and summary
346 static void show_page_range(unsigned long voffset
,
347 unsigned long offset
, uint64_t flags
)
349 static uint64_t flags0
;
350 static unsigned long voff
;
351 static unsigned long index
;
352 static unsigned long count
;
354 if (flags
== flags0
&& offset
== index
+ count
&&
355 (!opt_pid
|| voffset
== voff
+ count
)) {
362 printf("%lx\t", voff
);
363 printf("%lx\t%lx\t%s\n",
364 index
, count
, page_flag_name(flags0
));
373 static void show_page(unsigned long voffset
,
374 unsigned long offset
, uint64_t flags
)
377 printf("%lx\t", voffset
);
378 printf("%lx\t%s\n", offset
, page_flag_name(flags
));
381 static void show_summary(void)
385 printf(" flags\tpage-count MB"
386 " symbolic-flags\t\t\tlong-symbolic-flags\n");
388 for (i
= 0; i
< ARRAY_SIZE(nr_pages
); i
++) {
390 printf("0x%016llx\t%10lu %8lu %s\t%s\n",
391 (unsigned long long)page_flags
[i
],
393 pages2mb(nr_pages
[i
]),
394 page_flag_name(page_flags
[i
]),
395 page_flag_longname(page_flags
[i
]));
398 printf(" total\t%10lu %8lu\n",
399 total_pages
, pages2mb(total_pages
));
407 static int bit_mask_ok(uint64_t flags
)
411 for (i
= 0; i
< nr_bit_filters
; i
++) {
412 if (opt_bits
[i
] == KPF_ALL_BITS
) {
413 if ((flags
& opt_mask
[i
]) == 0)
416 if ((flags
& opt_mask
[i
]) != opt_bits
[i
])
424 static uint64_t expand_overloaded_flags(uint64_t flags
, uint64_t pme
)
426 /* SLOB/SLUB overload several page flags */
427 if (flags
& BIT(SLAB
)) {
428 if (flags
& BIT(PRIVATE
))
429 flags
^= BIT(PRIVATE
) | BIT(SLOB_FREE
);
430 if (flags
& BIT(ACTIVE
))
431 flags
^= BIT(ACTIVE
) | BIT(SLUB_FROZEN
);
432 if (flags
& BIT(ERROR
))
433 flags
^= BIT(ERROR
) | BIT(SLUB_DEBUG
);
436 /* PG_reclaim is overloaded as PG_readahead in the read path */
437 if ((flags
& (BIT(RECLAIM
) | BIT(WRITEBACK
))) == BIT(RECLAIM
))
438 flags
^= BIT(RECLAIM
) | BIT(READAHEAD
);
440 if (pme
& PM_SOFT_DIRTY
)
441 flags
|= BIT(SOFTDIRTY
);
446 static uint64_t well_known_flags(uint64_t flags
)
448 /* hide flags intended only for kernel hacker */
449 flags
&= ~KPF_HACKERS_BITS
;
451 /* hide non-hugeTLB compound pages */
452 if ((flags
& BITS_COMPOUND
) && !(flags
& BIT(HUGE
)))
453 flags
&= ~BITS_COMPOUND
;
458 static uint64_t kpageflags_flags(uint64_t flags
, uint64_t pme
)
461 flags
= expand_overloaded_flags(flags
, pme
);
463 flags
= well_known_flags(flags
);
472 static void prepare_hwpoison_fd(void)
474 char buf
[MAX_PATH
+ 1];
476 hwpoison_debug_fs
= debugfs_mount(NULL
);
477 if (!hwpoison_debug_fs
) {
478 perror("mount debugfs");
482 if (opt_hwpoison
&& !hwpoison_inject_fd
) {
483 snprintf(buf
, MAX_PATH
, "%s/hwpoison/corrupt-pfn",
485 hwpoison_inject_fd
= checked_open(buf
, O_WRONLY
);
488 if (opt_unpoison
&& !hwpoison_forget_fd
) {
489 snprintf(buf
, MAX_PATH
, "%s/hwpoison/unpoison-pfn",
491 hwpoison_forget_fd
= checked_open(buf
, O_WRONLY
);
495 static int hwpoison_page(unsigned long offset
)
500 len
= sprintf(buf
, "0x%lx\n", offset
);
501 len
= write(hwpoison_inject_fd
, buf
, len
);
503 perror("hwpoison inject");
509 static int unpoison_page(unsigned long offset
)
514 len
= sprintf(buf
, "0x%lx\n", offset
);
515 len
= write(hwpoison_forget_fd
, buf
, len
);
517 perror("hwpoison forget");
527 static size_t hash_slot(uint64_t flags
)
529 size_t k
= HASH_KEY(flags
);
532 /* Explicitly reserve slot 0 for flags 0: the following logic
533 * cannot distinguish an unoccupied slot from slot (flags==0).
538 /* search through the remaining (HASH_SIZE-1) slots */
539 for (i
= 1; i
< ARRAY_SIZE(page_flags
); i
++, k
++) {
540 if (!k
|| k
>= ARRAY_SIZE(page_flags
))
542 if (page_flags
[k
] == 0) {
543 page_flags
[k
] = flags
;
546 if (page_flags
[k
] == flags
)
550 fatal("hash table full: bump up HASH_SHIFT?\n");
554 static void add_page(unsigned long voffset
,
555 unsigned long offset
, uint64_t flags
, uint64_t pme
)
557 flags
= kpageflags_flags(flags
, pme
);
559 if (!bit_mask_ok(flags
))
563 hwpoison_page(offset
);
565 unpoison_page(offset
);
568 show_page_range(voffset
, offset
, flags
);
569 else if (opt_list
== 2)
570 show_page(voffset
, offset
, flags
);
572 nr_pages
[hash_slot(flags
)]++;
576 #define KPAGEFLAGS_BATCH (64 << 10) /* 64k pages */
577 static void walk_pfn(unsigned long voffset
,
582 uint64_t buf
[KPAGEFLAGS_BATCH
];
588 batch
= min_t(unsigned long, count
, KPAGEFLAGS_BATCH
);
589 pages
= kpageflags_read(buf
, index
, batch
);
593 for (i
= 0; i
< pages
; i
++)
594 add_page(voffset
+ i
, index
+ i
, buf
[i
], pme
);
601 #define PAGEMAP_BATCH (64 << 10)
602 static void walk_vma(unsigned long index
, unsigned long count
)
604 uint64_t buf
[PAGEMAP_BATCH
];
611 batch
= min_t(unsigned long, count
, PAGEMAP_BATCH
);
612 pages
= pagemap_read(buf
, index
, batch
);
616 for (i
= 0; i
< pages
; i
++) {
617 pfn
= pagemap_pfn(buf
[i
]);
619 walk_pfn(index
+ i
, pfn
, 1, buf
[i
]);
627 static void walk_task(unsigned long index
, unsigned long count
)
629 const unsigned long end
= index
+ count
;
633 while (index
< end
) {
635 while (pg_end
[i
] <= index
)
638 if (pg_start
[i
] >= end
)
641 start
= max_t(unsigned long, pg_start
[i
], index
);
642 index
= min_t(unsigned long, pg_end
[i
], end
);
644 assert(start
< index
);
645 walk_vma(start
, index
- start
);
649 static void add_addr_range(unsigned long offset
, unsigned long size
)
651 if (nr_addr_ranges
>= MAX_ADDR_RANGES
)
652 fatal("too many addr ranges\n");
654 opt_offset
[nr_addr_ranges
] = offset
;
655 opt_size
[nr_addr_ranges
] = min_t(unsigned long, size
, ULONG_MAX
-offset
);
659 static void walk_addr_ranges(void)
663 kpageflags_fd
= checked_open(PROC_KPAGEFLAGS
, O_RDONLY
);
666 add_addr_range(0, ULONG_MAX
);
668 for (i
= 0; i
< nr_addr_ranges
; i
++)
670 walk_pfn(0, opt_offset
[i
], opt_size
[i
], 0);
672 walk_task(opt_offset
[i
], opt_size
[i
]);
674 close(kpageflags_fd
);
682 static const char *page_flag_type(uint64_t flag
)
684 if (flag
& KPF_HACKERS_BITS
)
686 if (flag
& KPF_OVERLOADED_BITS
)
691 static void usage(void)
696 "page-types [options]\n"
697 " -r|--raw Raw mode, for kernel developers\n"
698 " -d|--describe flags Describe flags\n"
699 " -a|--addr addr-spec Walk a range of pages\n"
700 " -b|--bits bits-spec Walk pages with specified bits\n"
701 " -p|--pid pid Walk process address space\n"
702 #if 0 /* planned features */
703 " -f|--file filename Walk file address space\n"
705 " -l|--list Show page details in ranges\n"
706 " -L|--list-each Show page details one by one\n"
707 " -N|--no-summary Don't show summary info\n"
708 " -X|--hwpoison hwpoison pages\n"
709 " -x|--unpoison unpoison pages\n"
710 " -h|--help Show this usage message\n"
712 " 0x10 bitfield format, e.g.\n"
713 " anon bit-name, e.g.\n"
714 " 0x10,anon comma-separated list, e.g.\n"
716 " N one page at offset N (unit: pages)\n"
717 " N+M pages range from N to N+M-1\n"
718 " N,M pages range from N to M-1\n"
719 " N, pages range from N to end\n"
720 " ,M pages range from 0 to M-1\n"
722 " bit1,bit2 (flags & (bit1|bit2)) != 0\n"
723 " bit1,bit2=bit1 (flags & (bit1|bit2)) == bit1\n"
724 " bit1,~bit2 (flags & (bit1|bit2)) == bit1\n"
725 " =bit1,bit2 flags == (bit1|bit2)\n"
729 for (i
= 0, j
= 0; i
< ARRAY_SIZE(page_flag_names
); i
++) {
730 if (!page_flag_names
[i
])
732 printf("%16s%s", page_flag_names
[i
] + 2,
733 page_flag_type(1ULL << i
));
740 "(r) raw mode bits (o) overloaded bits\n");
743 static unsigned long long parse_number(const char *str
)
745 unsigned long long n
;
747 n
= strtoll(str
, NULL
, 0);
749 if (n
== 0 && str
[0] != '0')
750 fatal("invalid name or number: %s\n", str
);
755 static void parse_pid(const char *str
)
760 opt_pid
= parse_number(str
);
762 sprintf(buf
, "/proc/%d/pagemap", opt_pid
);
763 pagemap_fd
= checked_open(buf
, O_RDONLY
);
765 sprintf(buf
, "/proc/%d/maps", opt_pid
);
766 file
= fopen(buf
, "r");
772 while (fgets(buf
, sizeof(buf
), file
) != NULL
) {
773 unsigned long vm_start
;
774 unsigned long vm_end
;
775 unsigned long long pgoff
;
781 n
= sscanf(buf
, "%lx-%lx %c%c%c%c %llx %x:%x %lu",
789 fprintf(stderr
, "unexpected line: %s\n", buf
);
792 pg_start
[nr_vmas
] = vm_start
/ page_size
;
793 pg_end
[nr_vmas
] = vm_end
/ page_size
;
794 if (++nr_vmas
>= MAX_VMAS
) {
795 fprintf(stderr
, "too many VMAs\n");
802 static void parse_file(const char *name
)
806 static void parse_addr_range(const char *optarg
)
808 unsigned long offset
;
812 p
= strchr(optarg
, ',');
814 p
= strchr(optarg
, '+');
818 size
= parse_number(p
+ 1);
820 offset
= parse_number(optarg
);
824 size
= parse_number(p
+ 1);
827 fatal("invalid range: %lu,%lu\n",
833 offset
= parse_number(optarg
);
837 add_addr_range(offset
, size
);
840 static void add_bits_filter(uint64_t mask
, uint64_t bits
)
842 if (nr_bit_filters
>= MAX_BIT_FILTERS
)
843 fatal("too much bit filters\n");
845 opt_mask
[nr_bit_filters
] = mask
;
846 opt_bits
[nr_bit_filters
] = bits
;
850 static uint64_t parse_flag_name(const char *str
, int len
)
857 if (len
<= 8 && !strncmp(str
, "compound", len
))
858 return BITS_COMPOUND
;
860 for (i
= 0; i
< ARRAY_SIZE(page_flag_names
); i
++) {
861 if (!page_flag_names
[i
])
863 if (!strncmp(str
, page_flag_names
[i
] + 2, len
))
867 return parse_number(str
);
870 static uint64_t parse_flag_names(const char *str
, int all
)
876 if (*p
== ',' || *p
== '=' || *p
== '\0') {
877 if ((*str
!= '~') || (*str
== '~' && all
&& *++str
))
878 flags
|= parse_flag_name(str
, p
- str
);
889 static void parse_bits_mask(const char *optarg
)
895 p
= strchr(optarg
, '=');
898 bits
= parse_flag_names(p
+ 1, 0);
900 mask
= parse_flag_names(optarg
, 0);
901 bits
= parse_flag_names(p
+ 1, 0);
902 } else if (strchr(optarg
, '~')) {
903 mask
= parse_flag_names(optarg
, 1);
904 bits
= parse_flag_names(optarg
, 0);
906 mask
= parse_flag_names(optarg
, 0);
910 add_bits_filter(mask
, bits
);
913 static void describe_flags(const char *optarg
)
915 uint64_t flags
= parse_flag_names(optarg
, 0);
917 printf("0x%016llx\t%s\t%s\n",
918 (unsigned long long)flags
,
919 page_flag_name(flags
),
920 page_flag_longname(flags
));
923 static const struct option opts
[] = {
924 { "raw" , 0, NULL
, 'r' },
925 { "pid" , 1, NULL
, 'p' },
926 { "file" , 1, NULL
, 'f' },
927 { "addr" , 1, NULL
, 'a' },
928 { "bits" , 1, NULL
, 'b' },
929 { "describe" , 1, NULL
, 'd' },
930 { "list" , 0, NULL
, 'l' },
931 { "list-each" , 0, NULL
, 'L' },
932 { "no-summary", 0, NULL
, 'N' },
933 { "hwpoison" , 0, NULL
, 'X' },
934 { "unpoison" , 0, NULL
, 'x' },
935 { "help" , 0, NULL
, 'h' },
936 { NULL
, 0, NULL
, 0 }
939 int main(int argc
, char *argv
[])
943 page_size
= getpagesize();
945 while ((c
= getopt_long(argc
, argv
,
946 "rp:f:a:b:d:lLNXxh", opts
, NULL
)) != -1) {
958 parse_addr_range(optarg
);
961 parse_bits_mask(optarg
);
964 describe_flags(optarg
);
977 prepare_hwpoison_fd();
981 prepare_hwpoison_fd();
992 if (opt_list
&& opt_pid
)
995 printf("offset\tlen\tflags\n");
997 printf("offset\tflags\n");
1002 show_page_range(0, 0, 0); /* drain the buffer */