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 _FILE_OFFSET_BITS 64
37 #include <sys/types.h>
38 #include <sys/errno.h>
39 #include <sys/fcntl.h>
40 #include <sys/mount.h>
41 #include <sys/statfs.h>
43 #include "../../include/uapi/linux/magic.h"
44 #include "../../include/uapi/linux/kernel-page-flags.h"
45 #include <api/fs/fs.h>
53 # define STR(x) _STR(x)
57 * pagemap kernel ABI bits
60 #define PM_ENTRY_BYTES 8
61 #define PM_PFRAME_BITS 55
62 #define PM_PFRAME_MASK ((1LL << PM_PFRAME_BITS) - 1)
63 #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
64 #define MAX_SWAPFILES_SHIFT 5
65 #define PM_SWAP_OFFSET(x) (((x) & PM_PFRAME_MASK) >> MAX_SWAPFILES_SHIFT)
66 #define PM_SOFT_DIRTY (1ULL << 55)
67 #define PM_MMAP_EXCLUSIVE (1ULL << 56)
68 #define PM_FILE (1ULL << 61)
69 #define PM_SWAP (1ULL << 62)
70 #define PM_PRESENT (1ULL << 63)
77 #define PROC_KPAGEFLAGS "/proc/kpageflags"
78 #define PROC_KPAGECOUNT "/proc/kpagecount"
79 #define PROC_KPAGECGROUP "/proc/kpagecgroup"
81 #define SYS_KERNEL_MM_PAGE_IDLE "/sys/kernel/mm/page_idle/bitmap"
83 /* [32-] kernel hacking assistances */
84 #define KPF_RESERVED 32
85 #define KPF_MLOCKED 33
86 #define KPF_MAPPEDTODISK 34
87 #define KPF_PRIVATE 35
88 #define KPF_PRIVATE_2 36
89 #define KPF_OWNER_PRIVATE 37
91 #define KPF_UNCACHED 39
92 #define KPF_SOFTDIRTY 40
94 /* [48-] take some arbitrary free slots for expanding overloaded flags
95 * not part of kernel API
97 #define KPF_READAHEAD 48
98 #define KPF_SLOB_FREE 49
99 #define KPF_SLUB_FROZEN 50
100 #define KPF_SLUB_DEBUG 51
103 #define KPF_MMAP_EXCLUSIVE 63
105 #define KPF_ALL_BITS ((uint64_t)~0ULL)
106 #define KPF_HACKERS_BITS (0xffffULL << 32)
107 #define KPF_OVERLOADED_BITS (0xffffULL << 48)
108 #define BIT(name) (1ULL << KPF_##name)
109 #define BITS_COMPOUND (BIT(COMPOUND_HEAD) | BIT(COMPOUND_TAIL))
111 static const char * const page_flag_names[] = {
112 [KPF_LOCKED] = "L:locked",
113 [KPF_ERROR] = "E:error",
114 [KPF_REFERENCED] = "R:referenced",
115 [KPF_UPTODATE] = "U:uptodate",
116 [KPF_DIRTY] = "D:dirty",
118 [KPF_ACTIVE] = "A:active",
119 [KPF_SLAB] = "S:slab",
120 [KPF_WRITEBACK] = "W:writeback",
121 [KPF_RECLAIM] = "I:reclaim",
122 [KPF_BUDDY] = "B:buddy",
124 [KPF_MMAP] = "M:mmap",
125 [KPF_ANON] = "a:anonymous",
126 [KPF_SWAPCACHE] = "s:swapcache",
127 [KPF_SWAPBACKED] = "b:swapbacked",
128 [KPF_COMPOUND_HEAD] = "H:compound_head",
129 [KPF_COMPOUND_TAIL] = "T:compound_tail",
130 [KPF_HUGE] = "G:huge",
131 [KPF_UNEVICTABLE] = "u:unevictable",
132 [KPF_HWPOISON] = "X:hwpoison",
133 [KPF_NOPAGE] = "n:nopage",
136 [KPF_BALLOON] = "o:balloon",
137 [KPF_PGTABLE] = "g:pgtable",
138 [KPF_ZERO_PAGE] = "z:zero_page",
139 [KPF_IDLE] = "i:idle_page",
141 [KPF_RESERVED] = "r:reserved",
142 [KPF_MLOCKED] = "m:mlocked",
143 [KPF_MAPPEDTODISK] = "d:mappedtodisk",
144 [KPF_PRIVATE] = "P:private",
145 [KPF_PRIVATE_2] = "p:private_2",
146 [KPF_OWNER_PRIVATE] = "O:owner_private",
147 [KPF_ARCH] = "h:arch",
148 [KPF_UNCACHED] = "c:uncached",
149 [KPF_SOFTDIRTY] = "f:softdirty",
151 [KPF_READAHEAD] = "I:readahead",
152 [KPF_SLOB_FREE] = "P:slob_free",
153 [KPF_SLUB_FROZEN] = "A:slub_frozen",
154 [KPF_SLUB_DEBUG] = "E:slub_debug",
156 [KPF_FILE] = "F:file",
157 [KPF_SWAP] = "w:swap",
158 [KPF_MMAP_EXCLUSIVE] = "1:mmap_exclusive",
162 static const char * const debugfs_known_mountpoints[] = {
172 static int opt_raw; /* for kernel developers */
173 static int opt_list; /* list pages (in ranges) */
174 static int opt_mark_idle; /* set accessed bit */
175 static int opt_no_summary; /* don't show summary */
176 static pid_t opt_pid; /* process to walk */
177 const char *opt_file; /* file or directory path */
178 static uint64_t opt_cgroup; /* cgroup inode */
179 static int opt_list_cgroup;/* list page cgroup */
180 static int opt_list_mapcnt;/* list page map count */
181 static const char *opt_kpageflags;/* kpageflags file to parse */
183 #define MAX_ADDR_RANGES 1024
184 static int nr_addr_ranges;
185 static unsigned long opt_offset[MAX_ADDR_RANGES];
186 static unsigned long opt_size[MAX_ADDR_RANGES];
188 #define MAX_VMAS 10240
190 static unsigned long pg_start[MAX_VMAS];
191 static unsigned long pg_end[MAX_VMAS];
193 #define MAX_BIT_FILTERS 64
194 static int nr_bit_filters;
195 static uint64_t opt_mask[MAX_BIT_FILTERS];
196 static uint64_t opt_bits[MAX_BIT_FILTERS];
198 static int page_size;
200 static int pagemap_fd;
201 static int kpageflags_fd;
202 static int kpagecount_fd = -1;
203 static int kpagecgroup_fd = -1;
204 static int page_idle_fd = -1;
206 static int opt_hwpoison;
207 static int opt_unpoison;
209 static const char *hwpoison_debug_fs;
210 static int hwpoison_inject_fd;
211 static int hwpoison_forget_fd;
213 #define HASH_SHIFT 13
214 #define HASH_SIZE (1 << HASH_SHIFT)
215 #define HASH_MASK (HASH_SIZE - 1)
216 #define HASH_KEY(flags) (flags & HASH_MASK)
218 static unsigned long total_pages;
219 static unsigned long nr_pages[HASH_SIZE];
220 static uint64_t page_flags[HASH_SIZE];
227 #define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0]))
229 #define min_t(type, x, y) ({ \
232 __min1 < __min2 ? __min1 : __min2; })
234 #define max_t(type, x, y) ({ \
237 __max1 > __max2 ? __max1 : __max2; })
239 static unsigned long pages2mb(unsigned long pages)
241 return (pages * page_size) >> 20;
244 static void fatal(const char *x, ...)
249 vfprintf(stderr, x, ap);
254 static int checked_open(const char *pathname, int flags)
256 int fd = open(pathname, flags);
267 * pagemap/kpageflags routines
270 static unsigned long do_u64_read(int fd, const char *name,
277 if (index > ULONG_MAX / 8)
278 fatal("index overflow: %lu\n", index);
280 bytes = pread(fd, buf, count * 8, (off_t)index * 8);
286 fatal("partial read: %lu bytes\n", bytes);
291 static unsigned long kpageflags_read(uint64_t *buf,
295 return do_u64_read(kpageflags_fd, opt_kpageflags, buf, index, pages);
298 static unsigned long kpagecgroup_read(uint64_t *buf,
302 if (kpagecgroup_fd < 0)
305 return do_u64_read(kpagecgroup_fd, opt_kpageflags, buf, index, pages);
308 static unsigned long kpagecount_read(uint64_t *buf,
312 return kpagecount_fd < 0 ? pages :
313 do_u64_read(kpagecount_fd, PROC_KPAGECOUNT,
317 static unsigned long pagemap_read(uint64_t *buf,
321 return do_u64_read(pagemap_fd, "/proc/pid/pagemap", buf, index, pages);
324 static unsigned long pagemap_pfn(uint64_t val)
328 if (val & PM_PRESENT)
329 pfn = PM_PFRAME(val);
336 static unsigned long pagemap_swap_offset(uint64_t val)
338 return val & PM_SWAP ? PM_SWAP_OFFSET(val) : 0;
345 static char *page_flag_name(uint64_t flags)
351 for (i = 0, j = 0; i < ARRAY_SIZE(page_flag_names); i++) {
352 present = (flags >> i) & 1;
353 if (!page_flag_names[i]) {
355 fatal("unknown flag bit %d\n", i);
358 buf[j++] = present ? page_flag_names[i][0] : '_';
364 static char *page_flag_longname(uint64_t flags)
366 static char buf[1024];
369 for (i = 0, n = 0; i < ARRAY_SIZE(page_flag_names); i++) {
370 if (!page_flag_names[i])
372 if ((flags >> i) & 1)
373 n += snprintf(buf + n, sizeof(buf) - n, "%s,",
374 page_flag_names[i] + 2);
385 * page list and summary
388 static void show_page_range(unsigned long voffset, unsigned long offset,
389 unsigned long size, uint64_t flags,
390 uint64_t cgroup, uint64_t mapcnt)
392 static uint64_t flags0;
393 static uint64_t cgroup0;
394 static uint64_t mapcnt0;
395 static unsigned long voff;
396 static unsigned long index;
397 static unsigned long count;
399 if (flags == flags0 && cgroup == cgroup0 && mapcnt == mapcnt0 &&
400 offset == index + count && size && voffset == voff + count) {
407 printf("%lx\t", voff);
409 printf("%lu\t", voff);
411 printf("@%llu\t", (unsigned long long)cgroup0);
413 printf("%lu\t", mapcnt0);
414 printf("%lx\t%lx\t%s\n",
415 index, count, page_flag_name(flags0));
426 static void flush_page_range(void)
428 show_page_range(0, 0, 0, 0, 0, 0);
431 static void show_page(unsigned long voffset, unsigned long offset,
432 uint64_t flags, uint64_t cgroup, uint64_t mapcnt)
435 printf("%lx\t", voffset);
437 printf("%lu\t", voffset);
439 printf("@%llu\t", (unsigned long long)cgroup);
441 printf("%lu\t", mapcnt);
443 printf("%lx\t%s\n", offset, page_flag_name(flags));
446 static void show_summary(void)
450 printf(" flags\tpage-count MB"
451 " symbolic-flags\t\t\tlong-symbolic-flags\n");
453 for (i = 0; i < ARRAY_SIZE(nr_pages); i++) {
455 printf("0x%016llx\t%10lu %8lu %s\t%s\n",
456 (unsigned long long)page_flags[i],
458 pages2mb(nr_pages[i]),
459 page_flag_name(page_flags[i]),
460 page_flag_longname(page_flags[i]));
463 printf(" total\t%10lu %8lu\n",
464 total_pages, pages2mb(total_pages));
472 static int bit_mask_ok(uint64_t flags)
476 for (i = 0; i < nr_bit_filters; i++) {
477 if (opt_bits[i] == KPF_ALL_BITS) {
478 if ((flags & opt_mask[i]) == 0)
481 if ((flags & opt_mask[i]) != opt_bits[i])
489 static uint64_t expand_overloaded_flags(uint64_t flags, uint64_t pme)
491 /* SLOB/SLUB overload several page flags */
492 if (flags & BIT(SLAB)) {
493 if (flags & BIT(PRIVATE))
494 flags ^= BIT(PRIVATE) | BIT(SLOB_FREE);
495 if (flags & BIT(ACTIVE))
496 flags ^= BIT(ACTIVE) | BIT(SLUB_FROZEN);
497 if (flags & BIT(ERROR))
498 flags ^= BIT(ERROR) | BIT(SLUB_DEBUG);
501 /* PG_reclaim is overloaded as PG_readahead in the read path */
502 if ((flags & (BIT(RECLAIM) | BIT(WRITEBACK))) == BIT(RECLAIM))
503 flags ^= BIT(RECLAIM) | BIT(READAHEAD);
505 if (pme & PM_SOFT_DIRTY)
506 flags |= BIT(SOFTDIRTY);
511 if (pme & PM_MMAP_EXCLUSIVE)
512 flags |= BIT(MMAP_EXCLUSIVE);
517 static uint64_t well_known_flags(uint64_t flags)
519 /* hide flags intended only for kernel hacker */
520 flags &= ~KPF_HACKERS_BITS;
522 /* hide non-hugeTLB compound pages */
523 if ((flags & BITS_COMPOUND) && !(flags & BIT(HUGE)))
524 flags &= ~BITS_COMPOUND;
529 static uint64_t kpageflags_flags(uint64_t flags, uint64_t pme)
532 flags = expand_overloaded_flags(flags, pme);
534 flags = well_known_flags(flags);
543 static void prepare_hwpoison_fd(void)
545 char buf[MAX_PATH + 1];
547 hwpoison_debug_fs = debugfs__mount();
548 if (!hwpoison_debug_fs) {
549 perror("mount debugfs");
553 if (opt_hwpoison && !hwpoison_inject_fd) {
554 snprintf(buf, MAX_PATH, "%s/hwpoison/corrupt-pfn",
556 hwpoison_inject_fd = checked_open(buf, O_WRONLY);
559 if (opt_unpoison && !hwpoison_forget_fd) {
560 snprintf(buf, MAX_PATH, "%s/hwpoison/unpoison-pfn",
562 hwpoison_forget_fd = checked_open(buf, O_WRONLY);
566 static int hwpoison_page(unsigned long offset)
571 len = sprintf(buf, "0x%lx\n", offset);
572 len = write(hwpoison_inject_fd, buf, len);
574 perror("hwpoison inject");
580 static int unpoison_page(unsigned long offset)
585 len = sprintf(buf, "0x%lx\n", offset);
586 len = write(hwpoison_forget_fd, buf, len);
588 perror("hwpoison forget");
594 static int mark_page_idle(unsigned long offset)
596 static unsigned long off;
600 if ((offset / 64 == off / 64) || buf == 0) {
601 buf |= 1UL << (offset % 64);
606 len = pwrite(page_idle_fd, &buf, 8, 8 * (off / 64));
608 perror("mark page idle");
612 buf = 1UL << (offset % 64);
622 static size_t hash_slot(uint64_t flags)
624 size_t k = HASH_KEY(flags);
627 /* Explicitly reserve slot 0 for flags 0: the following logic
628 * cannot distinguish an unoccupied slot from slot (flags==0).
633 /* search through the remaining (HASH_SIZE-1) slots */
634 for (i = 1; i < ARRAY_SIZE(page_flags); i++, k++) {
635 if (!k || k >= ARRAY_SIZE(page_flags))
637 if (page_flags[k] == 0) {
638 page_flags[k] = flags;
641 if (page_flags[k] == flags)
645 fatal("hash table full: bump up HASH_SHIFT?\n");
649 static void add_page(unsigned long voffset, unsigned long offset,
650 uint64_t flags, uint64_t cgroup, uint64_t mapcnt,
653 flags = kpageflags_flags(flags, pme);
655 if (!bit_mask_ok(flags))
658 if (opt_cgroup && cgroup != (uint64_t)opt_cgroup)
662 hwpoison_page(offset);
664 unpoison_page(offset);
667 mark_page_idle(offset);
670 show_page_range(voffset, offset, 1, flags, cgroup, mapcnt);
671 else if (opt_list == 2)
672 show_page(voffset, offset, flags, cgroup, mapcnt);
674 nr_pages[hash_slot(flags)]++;
678 #define KPAGEFLAGS_BATCH (64 << 10) /* 64k pages */
679 static void walk_pfn(unsigned long voffset,
684 uint64_t buf[KPAGEFLAGS_BATCH];
685 uint64_t cgi[KPAGEFLAGS_BATCH];
686 uint64_t cnt[KPAGEFLAGS_BATCH];
692 * kpagecgroup_read() reads only if kpagecgroup were opened, but
693 * /proc/kpagecgroup might even not exist, so it's better to fill
694 * them with zeros here.
699 memset(cgi, 0, sizeof cgi);
702 batch = min_t(unsigned long, count, KPAGEFLAGS_BATCH);
703 pages = kpageflags_read(buf, index, batch);
707 if (kpagecgroup_read(cgi, index, pages) != pages)
708 fatal("kpagecgroup returned fewer pages than expected");
710 if (kpagecount_read(cnt, index, batch) != pages)
711 fatal("kpagecount returned fewer pages than expected");
713 for (i = 0; i < pages; i++)
714 add_page(voffset + i, index + i,
715 buf[i], cgi[i], cnt[i], pme);
722 static void walk_swap(unsigned long voffset, uint64_t pme)
724 uint64_t flags = kpageflags_flags(0, pme);
726 if (!bit_mask_ok(flags))
733 show_page_range(voffset, pagemap_swap_offset(pme),
735 else if (opt_list == 2)
736 show_page(voffset, pagemap_swap_offset(pme), flags, 0, 0);
738 nr_pages[hash_slot(flags)]++;
742 #define PAGEMAP_BATCH (64 << 10)
743 static void walk_vma(unsigned long index, unsigned long count)
745 uint64_t buf[PAGEMAP_BATCH];
752 batch = min_t(unsigned long, count, PAGEMAP_BATCH);
753 pages = pagemap_read(buf, index, batch);
757 for (i = 0; i < pages; i++) {
758 pfn = pagemap_pfn(buf[i]);
760 walk_pfn(index + i, pfn, 1, buf[i]);
761 if (buf[i] & PM_SWAP)
762 walk_swap(index + i, buf[i]);
770 static void walk_task(unsigned long index, unsigned long count)
772 const unsigned long end = index + count;
776 while (index < end) {
778 while (pg_end[i] <= index)
781 if (pg_start[i] >= end)
784 start = max_t(unsigned long, pg_start[i], index);
785 index = min_t(unsigned long, pg_end[i], end);
787 assert(start < index);
788 walk_vma(start, index - start);
792 static void add_addr_range(unsigned long offset, unsigned long size)
794 if (nr_addr_ranges >= MAX_ADDR_RANGES)
795 fatal("too many addr ranges\n");
797 opt_offset[nr_addr_ranges] = offset;
798 opt_size[nr_addr_ranges] = min_t(unsigned long, size, ULONG_MAX-offset);
802 static void walk_addr_ranges(void)
806 kpageflags_fd = checked_open(opt_kpageflags, O_RDONLY);
809 add_addr_range(0, ULONG_MAX);
811 for (i = 0; i < nr_addr_ranges; i++)
813 walk_pfn(opt_offset[i], opt_offset[i], opt_size[i], 0);
815 walk_task(opt_offset[i], opt_size[i]);
820 close(kpageflags_fd);
828 static const char *page_flag_type(uint64_t flag)
830 if (flag & KPF_HACKERS_BITS)
832 if (flag & KPF_OVERLOADED_BITS)
837 static void usage(void)
842 "page-types [options]\n"
843 " -r|--raw Raw mode, for kernel developers\n"
844 " -d|--describe flags Describe flags\n"
845 " -a|--addr addr-spec Walk a range of pages\n"
846 " -b|--bits bits-spec Walk pages with specified bits\n"
847 " -c|--cgroup path|@inode Walk pages within memory cgroup\n"
848 " -p|--pid pid Walk process address space\n"
849 " -f|--file filename Walk file address space\n"
850 " -i|--mark-idle Mark pages idle\n"
851 " -l|--list Show page details in ranges\n"
852 " -L|--list-each Show page details one by one\n"
853 " -C|--list-cgroup Show cgroup inode for pages\n"
854 " -M|--list-mapcnt Show page map count\n"
855 " -N|--no-summary Don't show summary info\n"
856 " -X|--hwpoison hwpoison pages\n"
857 " -x|--unpoison unpoison pages\n"
858 " -F|--kpageflags filename kpageflags file to parse\n"
859 " -h|--help Show this usage message\n"
861 " 0x10 bitfield format, e.g.\n"
862 " anon bit-name, e.g.\n"
863 " 0x10,anon comma-separated list, e.g.\n"
865 " N one page at offset N (unit: pages)\n"
866 " N+M pages range from N to N+M-1\n"
867 " N,M pages range from N to M-1\n"
868 " N, pages range from N to end\n"
869 " ,M pages range from 0 to M-1\n"
871 " bit1,bit2 (flags & (bit1|bit2)) != 0\n"
872 " bit1,bit2=bit1 (flags & (bit1|bit2)) == bit1\n"
873 " bit1,~bit2 (flags & (bit1|bit2)) == bit1\n"
874 " =bit1,bit2 flags == (bit1|bit2)\n"
878 for (i = 0, j = 0; i < ARRAY_SIZE(page_flag_names); i++) {
879 if (!page_flag_names[i])
881 printf("%16s%s", page_flag_names[i] + 2,
882 page_flag_type(1ULL << i));
889 "(r) raw mode bits (o) overloaded bits\n");
892 static unsigned long long parse_number(const char *str)
894 unsigned long long n;
896 n = strtoll(str, NULL, 0);
898 if (n == 0 && str[0] != '0')
899 fatal("invalid name or number: %s\n", str);
904 static void parse_pid(const char *str)
909 opt_pid = parse_number(str);
911 sprintf(buf, "/proc/%d/pagemap", opt_pid);
912 pagemap_fd = checked_open(buf, O_RDONLY);
914 sprintf(buf, "/proc/%d/maps", opt_pid);
915 file = fopen(buf, "r");
921 while (fgets(buf, sizeof(buf), file) != NULL) {
922 unsigned long vm_start;
923 unsigned long vm_end;
924 unsigned long long pgoff;
930 n = sscanf(buf, "%lx-%lx %c%c%c%c %llx %x:%x %lu",
938 fprintf(stderr, "unexpected line: %s\n", buf);
941 pg_start[nr_vmas] = vm_start / page_size;
942 pg_end[nr_vmas] = vm_end / page_size;
943 if (++nr_vmas >= MAX_VMAS) {
944 fprintf(stderr, "too many VMAs\n");
951 static void show_file(const char *name, const struct stat *st)
953 unsigned long long size = st->st_size;
954 char atime[64], mtime[64];
955 long now = time(NULL);
957 printf("%s\tInode: %u\tSize: %llu (%llu pages)\n",
958 name, (unsigned)st->st_ino,
959 size, (size + page_size - 1) / page_size);
961 strftime(atime, sizeof(atime), "%c", localtime(&st->st_atime));
962 strftime(mtime, sizeof(mtime), "%c", localtime(&st->st_mtime));
964 printf("Modify: %s (%ld seconds ago)\nAccess: %s (%ld seconds ago)\n",
965 mtime, now - st->st_mtime,
966 atime, now - st->st_atime);
969 static sigjmp_buf sigbus_jmp;
971 static void * volatile sigbus_addr;
973 static void sigbus_handler(int sig, siginfo_t *info, void *ucontex)
977 sigbus_addr = info ? info->si_addr : NULL;
978 siglongjmp(sigbus_jmp, 1);
981 static struct sigaction sigbus_action = {
982 .sa_sigaction = sigbus_handler,
983 .sa_flags = SA_SIGINFO,
986 static void walk_file(const char *name, const struct stat *st)
988 uint8_t vec[PAGEMAP_BATCH];
989 uint64_t buf[PAGEMAP_BATCH], flags;
992 unsigned long nr_pages, pfn, i;
993 off_t off, end = st->st_size;
999 fd = checked_open(name, O_RDONLY|O_NOATIME|O_NOFOLLOW);
1001 for (off = 0; off < end; off += len) {
1002 nr_pages = (end - off + page_size - 1) / page_size;
1003 if (nr_pages > PAGEMAP_BATCH)
1004 nr_pages = PAGEMAP_BATCH;
1005 len = nr_pages * page_size;
1007 ptr = mmap(NULL, len, PROT_READ, MAP_SHARED, fd, off);
1008 if (ptr == MAP_FAILED)
1009 fatal("mmap failed: %s", name);
1011 /* determine cached pages */
1012 if (mincore(ptr, len, vec))
1013 fatal("mincore failed: %s", name);
1015 /* turn off readahead */
1016 if (madvise(ptr, len, MADV_RANDOM))
1017 fatal("madvice failed: %s", name);
1019 if (sigsetjmp(sigbus_jmp, 1)) {
1020 end = off + sigbus_addr ? sigbus_addr - ptr : 0;
1021 fprintf(stderr, "got sigbus at offset %lld: %s\n",
1022 (long long)end, name);
1027 for (i = 0; i < nr_pages ; i++) {
1029 (void)*(volatile int *)(ptr + i * page_size);
1033 /* turn off harvesting reference bits */
1034 if (madvise(ptr, len, MADV_SEQUENTIAL))
1035 fatal("madvice failed: %s", name);
1037 if (pagemap_read(buf, (unsigned long)ptr / page_size,
1038 nr_pages) != nr_pages)
1039 fatal("cannot read pagemap");
1043 for (i = 0; i < nr_pages; i++) {
1044 pfn = pagemap_pfn(buf[i]);
1047 if (!kpageflags_read(&flags, pfn, 1))
1049 if (!kpagecgroup_read(&cgroup, pfn, 1))
1050 fatal("kpagecgroup_read failed");
1051 if (!kpagecount_read(&mapcnt, pfn, 1))
1052 fatal("kpagecount_read failed");
1053 if (first && opt_list) {
1056 show_file(name, st);
1058 add_page(off / page_size + i, pfn,
1059 flags, cgroup, mapcnt, buf[i]);
1066 int walk_tree(const char *name, const struct stat *st, int type, struct FTW *f)
1071 if (S_ISREG(st->st_mode))
1072 walk_file(name, st);
1075 fprintf(stderr, "cannot read dir: %s\n", name);
1081 static void walk_page_cache(void)
1085 kpageflags_fd = checked_open(opt_kpageflags, O_RDONLY);
1086 pagemap_fd = checked_open("/proc/self/pagemap", O_RDONLY);
1087 sigaction(SIGBUS, &sigbus_action, NULL);
1089 if (stat(opt_file, &st))
1090 fatal("stat failed: %s\n", opt_file);
1092 if (S_ISREG(st.st_mode)) {
1093 walk_file(opt_file, &st);
1094 } else if (S_ISDIR(st.st_mode)) {
1095 /* do not follow symlinks and mountpoints */
1096 if (nftw(opt_file, walk_tree, 64, FTW_MOUNT | FTW_PHYS) < 0)
1097 fatal("nftw failed: %s\n", opt_file);
1099 fatal("unhandled file type: %s\n", opt_file);
1101 close(kpageflags_fd);
1103 signal(SIGBUS, SIG_DFL);
1106 static void parse_file(const char *name)
1111 static void parse_cgroup(const char *path)
1113 if (path[0] == '@') {
1114 opt_cgroup = parse_number(path + 1);
1120 if (stat(path, &st))
1121 fatal("stat failed: %s: %m\n", path);
1123 if (!S_ISDIR(st.st_mode))
1124 fatal("cgroup supposed to be a directory: %s\n", path);
1126 opt_cgroup = st.st_ino;
1129 static void parse_addr_range(const char *optarg)
1131 unsigned long offset;
1135 p = strchr(optarg, ',');
1137 p = strchr(optarg, '+');
1141 size = parse_number(p + 1);
1143 offset = parse_number(optarg);
1147 size = parse_number(p + 1);
1150 fatal("invalid range: %lu,%lu\n",
1156 offset = parse_number(optarg);
1160 add_addr_range(offset, size);
1163 static void add_bits_filter(uint64_t mask, uint64_t bits)
1165 if (nr_bit_filters >= MAX_BIT_FILTERS)
1166 fatal("too much bit filters\n");
1168 opt_mask[nr_bit_filters] = mask;
1169 opt_bits[nr_bit_filters] = bits;
1173 static uint64_t parse_flag_name(const char *str, int len)
1180 if (len <= 8 && !strncmp(str, "compound", len))
1181 return BITS_COMPOUND;
1183 for (i = 0; i < ARRAY_SIZE(page_flag_names); i++) {
1184 if (!page_flag_names[i])
1186 if (!strncmp(str, page_flag_names[i] + 2, len))
1190 return parse_number(str);
1193 static uint64_t parse_flag_names(const char *str, int all)
1195 const char *p = str;
1199 if (*p == ',' || *p == '=' || *p == '\0') {
1200 if ((*str != '~') || (*str == '~' && all && *++str))
1201 flags |= parse_flag_name(str, p - str);
1212 static void parse_bits_mask(const char *optarg)
1218 p = strchr(optarg, '=');
1220 mask = KPF_ALL_BITS;
1221 bits = parse_flag_names(p + 1, 0);
1223 mask = parse_flag_names(optarg, 0);
1224 bits = parse_flag_names(p + 1, 0);
1225 } else if (strchr(optarg, '~')) {
1226 mask = parse_flag_names(optarg, 1);
1227 bits = parse_flag_names(optarg, 0);
1229 mask = parse_flag_names(optarg, 0);
1230 bits = KPF_ALL_BITS;
1233 add_bits_filter(mask, bits);
1236 static void parse_kpageflags(const char *name)
1238 opt_kpageflags = name;
1241 static void describe_flags(const char *optarg)
1243 uint64_t flags = parse_flag_names(optarg, 0);
1245 printf("0x%016llx\t%s\t%s\n",
1246 (unsigned long long)flags,
1247 page_flag_name(flags),
1248 page_flag_longname(flags));
1251 static const struct option opts[] = {
1252 { "raw" , 0, NULL, 'r' },
1253 { "pid" , 1, NULL, 'p' },
1254 { "file" , 1, NULL, 'f' },
1255 { "addr" , 1, NULL, 'a' },
1256 { "bits" , 1, NULL, 'b' },
1257 { "cgroup" , 1, NULL, 'c' },
1258 { "describe" , 1, NULL, 'd' },
1259 { "mark-idle" , 0, NULL, 'i' },
1260 { "list" , 0, NULL, 'l' },
1261 { "list-each" , 0, NULL, 'L' },
1262 { "list-cgroup", 0, NULL, 'C' },
1263 { "list-mapcnt", 0, NULL, 'M' },
1264 { "no-summary", 0, NULL, 'N' },
1265 { "hwpoison" , 0, NULL, 'X' },
1266 { "unpoison" , 0, NULL, 'x' },
1267 { "kpageflags", 0, NULL, 'F' },
1268 { "help" , 0, NULL, 'h' },
1269 { NULL , 0, NULL, 0 }
1272 int main(int argc, char *argv[])
1276 page_size = getpagesize();
1278 while ((c = getopt_long(argc, argv,
1279 "rp:f:a:b:d:c:CilLMNXxF:h",
1280 opts, NULL)) != -1) {
1292 parse_addr_range(optarg);
1295 parse_bits_mask(optarg);
1298 parse_cgroup(optarg);
1301 opt_list_cgroup = 1;
1304 describe_flags(optarg);
1316 opt_list_mapcnt = 1;
1323 prepare_hwpoison_fd();
1327 prepare_hwpoison_fd();
1330 parse_kpageflags(optarg);
1341 if (!opt_kpageflags)
1342 opt_kpageflags = PROC_KPAGEFLAGS;
1344 if (opt_cgroup || opt_list_cgroup)
1345 kpagecgroup_fd = checked_open(PROC_KPAGECGROUP, O_RDONLY);
1347 if (opt_list && opt_list_mapcnt)
1348 kpagecount_fd = checked_open(PROC_KPAGECOUNT, O_RDONLY);
1350 if (opt_mark_idle && opt_file)
1351 page_idle_fd = checked_open(SYS_KERNEL_MM_PAGE_IDLE, O_RDWR);
1353 if (opt_list && opt_pid)
1354 printf("voffset\t");
1355 if (opt_list && opt_file)
1356 printf("foffset\t");
1357 if (opt_list && opt_list_cgroup)
1359 if (opt_list && opt_list_mapcnt)
1360 printf("map-cnt\t");
1363 printf("offset\tlen\tflags\n");
1365 printf("offset\tflags\n");
1383 if (opt_list_mapcnt)
1384 close(kpagecount_fd);
1386 if (page_idle_fd >= 0)
1387 close(page_idle_fd);