2 * page-types: Tool for querying page flags
4 * Copyright (C) 2009 Intel corporation
6 * Authors: Wu Fengguang <fengguang.wu@intel.com>
8 * Released under the General Public License (GPL).
11 #define _LARGEFILE64_SOURCE
21 #include <sys/types.h>
22 #include <sys/errno.h>
23 #include <sys/fcntl.h>
27 * pagemap kernel ABI bits
30 #define PM_ENTRY_BYTES sizeof(uint64_t)
31 #define PM_STATUS_BITS 3
32 #define PM_STATUS_OFFSET (64 - PM_STATUS_BITS)
33 #define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
34 #define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
35 #define PM_PSHIFT_BITS 6
36 #define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
37 #define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
38 #define PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
39 #define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
40 #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
42 #define PM_PRESENT PM_STATUS(4LL)
43 #define PM_SWAP PM_STATUS(2LL)
51 #define PROC_KPAGEFLAGS "/proc/kpageflags"
53 /* copied from kpageflags_read() */
56 #define KPF_REFERENCED 2
57 #define KPF_UPTODATE 3
62 #define KPF_WRITEBACK 8
66 /* [11-20] new additions in 2.6.31 */
69 #define KPF_SWAPCACHE 13
70 #define KPF_SWAPBACKED 14
71 #define KPF_COMPOUND_HEAD 15
72 #define KPF_COMPOUND_TAIL 16
74 #define KPF_UNEVICTABLE 18
75 #define KPF_HWPOISON 19
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
89 /* [48-] take some arbitrary free slots for expanding overloaded flags
90 * not part of kernel API
92 #define KPF_READAHEAD 48
93 #define KPF_SLOB_FREE 49
94 #define KPF_SLUB_FROZEN 50
95 #define KPF_SLUB_DEBUG 51
97 #define KPF_ALL_BITS ((uint64_t)~0ULL)
98 #define KPF_HACKERS_BITS (0xffffULL << 32)
99 #define KPF_OVERLOADED_BITS (0xffffULL << 48)
100 #define BIT(name) (1ULL << KPF_##name)
101 #define BITS_COMPOUND (BIT(COMPOUND_HEAD) | BIT(COMPOUND_TAIL))
103 static char *page_flag_names[] = {
104 [KPF_LOCKED] = "L:locked",
105 [KPF_ERROR] = "E:error",
106 [KPF_REFERENCED] = "R:referenced",
107 [KPF_UPTODATE] = "U:uptodate",
108 [KPF_DIRTY] = "D:dirty",
110 [KPF_ACTIVE] = "A:active",
111 [KPF_SLAB] = "S:slab",
112 [KPF_WRITEBACK] = "W:writeback",
113 [KPF_RECLAIM] = "I:reclaim",
114 [KPF_BUDDY] = "B:buddy",
116 [KPF_MMAP] = "M:mmap",
117 [KPF_ANON] = "a:anonymous",
118 [KPF_SWAPCACHE] = "s:swapcache",
119 [KPF_SWAPBACKED] = "b:swapbacked",
120 [KPF_COMPOUND_HEAD] = "H:compound_head",
121 [KPF_COMPOUND_TAIL] = "T:compound_tail",
122 [KPF_HUGE] = "G:huge",
123 [KPF_UNEVICTABLE] = "u:unevictable",
124 [KPF_HWPOISON] = "X:hwpoison",
125 [KPF_NOPAGE] = "n:nopage",
128 [KPF_RESERVED] = "r:reserved",
129 [KPF_MLOCKED] = "m:mlocked",
130 [KPF_MAPPEDTODISK] = "d:mappedtodisk",
131 [KPF_PRIVATE] = "P:private",
132 [KPF_PRIVATE_2] = "p:private_2",
133 [KPF_OWNER_PRIVATE] = "O:owner_private",
134 [KPF_ARCH] = "h:arch",
135 [KPF_UNCACHED] = "c:uncached",
137 [KPF_READAHEAD] = "I:readahead",
138 [KPF_SLOB_FREE] = "P:slob_free",
139 [KPF_SLUB_FROZEN] = "A:slub_frozen",
140 [KPF_SLUB_DEBUG] = "E:slub_debug",
148 static int opt_raw; /* for kernel developers */
149 static int opt_list; /* list pages (in ranges) */
150 static int opt_no_summary; /* don't show summary */
151 static pid_t opt_pid; /* process to walk */
153 #define MAX_ADDR_RANGES 1024
154 static int nr_addr_ranges;
155 static unsigned long opt_offset[MAX_ADDR_RANGES];
156 static unsigned long opt_size[MAX_ADDR_RANGES];
158 #define MAX_VMAS 10240
160 static unsigned long pg_start[MAX_VMAS];
161 static unsigned long pg_end[MAX_VMAS];
162 static unsigned long voffset;
164 static int pagemap_fd;
166 #define MAX_BIT_FILTERS 64
167 static int nr_bit_filters;
168 static uint64_t opt_mask[MAX_BIT_FILTERS];
169 static uint64_t opt_bits[MAX_BIT_FILTERS];
171 static int page_size;
173 #define PAGES_BATCH (64 << 10) /* 64k pages */
174 static int kpageflags_fd;
176 #define HASH_SHIFT 13
177 #define HASH_SIZE (1 << HASH_SHIFT)
178 #define HASH_MASK (HASH_SIZE - 1)
179 #define HASH_KEY(flags) (flags & HASH_MASK)
181 static unsigned long total_pages;
182 static unsigned long nr_pages[HASH_SIZE];
183 static uint64_t page_flags[HASH_SIZE];
190 #define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0]))
192 #define min_t(type, x, y) ({ \
195 __min1 < __min2 ? __min1 : __min2; })
197 #define max_t(type, x, y) ({ \
200 __max1 > __max2 ? __max1 : __max2; })
202 static unsigned long pages2mb(unsigned long pages)
204 return (pages * page_size) >> 20;
207 static void fatal(const char *x, ...)
212 vfprintf(stderr, x, ap);
217 int checked_open(const char *pathname, int flags)
219 int fd = open(pathname, flags);
234 static char *page_flag_name(uint64_t flags)
240 for (i = 0, j = 0; i < ARRAY_SIZE(page_flag_names); i++) {
241 present = (flags >> i) & 1;
242 if (!page_flag_names[i]) {
244 fatal("unkown flag bit %d\n", i);
247 buf[j++] = present ? page_flag_names[i][0] : '_';
253 static char *page_flag_longname(uint64_t flags)
255 static char buf[1024];
258 for (i = 0, n = 0; i < ARRAY_SIZE(page_flag_names); i++) {
259 if (!page_flag_names[i])
261 if ((flags >> i) & 1)
262 n += snprintf(buf + n, sizeof(buf) - n, "%s,",
263 page_flag_names[i] + 2);
274 * page list and summary
277 static void show_page_range(unsigned long offset, uint64_t flags)
279 static uint64_t flags0;
280 static unsigned long voff;
281 static unsigned long index;
282 static unsigned long count;
284 if (flags == flags0 && offset == index + count &&
285 (!opt_pid || voffset == voff + count)) {
292 printf("%lx\t", voff);
293 printf("%lx\t%lx\t%s\n",
294 index, count, page_flag_name(flags0));
303 static void show_page(unsigned long offset, uint64_t flags)
306 printf("%lx\t", voffset);
307 printf("%lx\t%s\n", offset, page_flag_name(flags));
310 static void show_summary(void)
314 printf(" flags\tpage-count MB"
315 " symbolic-flags\t\t\tlong-symbolic-flags\n");
317 for (i = 0; i < ARRAY_SIZE(nr_pages); i++) {
319 printf("0x%016llx\t%10lu %8lu %s\t%s\n",
320 (unsigned long long)page_flags[i],
322 pages2mb(nr_pages[i]),
323 page_flag_name(page_flags[i]),
324 page_flag_longname(page_flags[i]));
327 printf(" total\t%10lu %8lu\n",
328 total_pages, pages2mb(total_pages));
336 static int bit_mask_ok(uint64_t flags)
340 for (i = 0; i < nr_bit_filters; i++) {
341 if (opt_bits[i] == KPF_ALL_BITS) {
342 if ((flags & opt_mask[i]) == 0)
345 if ((flags & opt_mask[i]) != opt_bits[i])
353 static uint64_t expand_overloaded_flags(uint64_t flags)
355 /* SLOB/SLUB overload several page flags */
356 if (flags & BIT(SLAB)) {
357 if (flags & BIT(PRIVATE))
358 flags ^= BIT(PRIVATE) | BIT(SLOB_FREE);
359 if (flags & BIT(ACTIVE))
360 flags ^= BIT(ACTIVE) | BIT(SLUB_FROZEN);
361 if (flags & BIT(ERROR))
362 flags ^= BIT(ERROR) | BIT(SLUB_DEBUG);
365 /* PG_reclaim is overloaded as PG_readahead in the read path */
366 if ((flags & (BIT(RECLAIM) | BIT(WRITEBACK))) == BIT(RECLAIM))
367 flags ^= BIT(RECLAIM) | BIT(READAHEAD);
372 static uint64_t well_known_flags(uint64_t flags)
374 /* hide flags intended only for kernel hacker */
375 flags &= ~KPF_HACKERS_BITS;
377 /* hide non-hugeTLB compound pages */
378 if ((flags & BITS_COMPOUND) && !(flags & BIT(HUGE)))
379 flags &= ~BITS_COMPOUND;
389 static int hash_slot(uint64_t flags)
391 int k = HASH_KEY(flags);
394 /* Explicitly reserve slot 0 for flags 0: the following logic
395 * cannot distinguish an unoccupied slot from slot (flags==0).
400 /* search through the remaining (HASH_SIZE-1) slots */
401 for (i = 1; i < ARRAY_SIZE(page_flags); i++, k++) {
402 if (!k || k >= ARRAY_SIZE(page_flags))
404 if (page_flags[k] == 0) {
405 page_flags[k] = flags;
408 if (page_flags[k] == flags)
412 fatal("hash table full: bump up HASH_SHIFT?\n");
416 static void add_page(unsigned long offset, uint64_t flags)
418 flags = expand_overloaded_flags(flags);
421 flags = well_known_flags(flags);
423 if (!bit_mask_ok(flags))
427 show_page_range(offset, flags);
428 else if (opt_list == 2)
429 show_page(offset, flags);
431 nr_pages[hash_slot(flags)]++;
435 static void walk_pfn(unsigned long index, unsigned long count)
441 if (index > ULONG_MAX / KPF_BYTES)
442 fatal("index overflow: %lu\n", index);
444 lseek(kpageflags_fd, index * KPF_BYTES, SEEK_SET);
447 uint64_t kpageflags_buf[KPF_BYTES * PAGES_BATCH];
449 batch = min_t(unsigned long, count, PAGES_BATCH);
450 n = read(kpageflags_fd, kpageflags_buf, batch * KPF_BYTES);
454 perror(PROC_KPAGEFLAGS);
458 if (n % KPF_BYTES != 0)
459 fatal("partial read: %lu bytes\n", n);
462 for (i = 0; i < n; i++)
463 add_page(index + i, kpageflags_buf[i]);
471 #define PAGEMAP_BATCH 4096
472 static unsigned long task_pfn(unsigned long pgoff)
474 static uint64_t buf[PAGEMAP_BATCH];
475 static unsigned long start;
479 if (pgoff < start || pgoff >= start + count) {
480 if (lseek64(pagemap_fd,
481 (uint64_t)pgoff * PM_ENTRY_BYTES,
483 perror("pagemap seek");
486 count = read(pagemap_fd, buf, sizeof(buf));
490 perror("pagemap read");
493 if (count % PM_ENTRY_BYTES) {
494 fatal("pagemap read not aligned.\n");
497 count /= PM_ENTRY_BYTES;
501 pfn = buf[pgoff - start];
502 if (pfn & PM_PRESENT)
503 pfn = PM_PFRAME(pfn);
510 static void walk_task(unsigned long index, unsigned long count)
513 const unsigned long end = index + count;
515 while (index < end) {
517 while (pg_end[i] <= index)
520 if (pg_start[i] >= end)
523 voffset = max_t(unsigned long, pg_start[i], index);
524 index = min_t(unsigned long, pg_end[i], end);
526 assert(voffset < index);
527 for (; voffset < index; voffset++) {
528 unsigned long pfn = task_pfn(voffset);
535 static void add_addr_range(unsigned long offset, unsigned long size)
537 if (nr_addr_ranges >= MAX_ADDR_RANGES)
538 fatal("too many addr ranges\n");
540 opt_offset[nr_addr_ranges] = offset;
541 opt_size[nr_addr_ranges] = min_t(unsigned long, size, ULONG_MAX-offset);
545 static void walk_addr_ranges(void)
549 kpageflags_fd = checked_open(PROC_KPAGEFLAGS, O_RDONLY);
552 add_addr_range(0, ULONG_MAX);
554 for (i = 0; i < nr_addr_ranges; i++)
556 walk_pfn(opt_offset[i], opt_size[i]);
558 walk_task(opt_offset[i], opt_size[i]);
560 close(kpageflags_fd);
568 static const char *page_flag_type(uint64_t flag)
570 if (flag & KPF_HACKERS_BITS)
572 if (flag & KPF_OVERLOADED_BITS)
577 static void usage(void)
582 "page-types [options]\n"
583 " -r|--raw Raw mode, for kernel developers\n"
584 " -a|--addr addr-spec Walk a range of pages\n"
585 " -b|--bits bits-spec Walk pages with specified bits\n"
586 " -p|--pid pid Walk process address space\n"
587 #if 0 /* planned features */
588 " -f|--file filename Walk file address space\n"
590 " -l|--list Show page details in ranges\n"
591 " -L|--list-each Show page details one by one\n"
592 " -N|--no-summary Don't show summay info\n"
593 " -h|--help Show this usage message\n"
595 " N one page at offset N (unit: pages)\n"
596 " N+M pages range from N to N+M-1\n"
597 " N,M pages range from N to M-1\n"
598 " N, pages range from N to end\n"
599 " ,M pages range from 0 to M-1\n"
601 " bit1,bit2 (flags & (bit1|bit2)) != 0\n"
602 " bit1,bit2=bit1 (flags & (bit1|bit2)) == bit1\n"
603 " bit1,~bit2 (flags & (bit1|bit2)) == bit1\n"
604 " =bit1,bit2 flags == (bit1|bit2)\n"
608 for (i = 0, j = 0; i < ARRAY_SIZE(page_flag_names); i++) {
609 if (!page_flag_names[i])
611 printf("%16s%s", page_flag_names[i] + 2,
612 page_flag_type(1ULL << i));
619 "(r) raw mode bits (o) overloaded bits\n");
622 static unsigned long long parse_number(const char *str)
624 unsigned long long n;
626 n = strtoll(str, NULL, 0);
628 if (n == 0 && str[0] != '0')
629 fatal("invalid name or number: %s\n", str);
634 static void parse_pid(const char *str)
639 opt_pid = parse_number(str);
641 sprintf(buf, "/proc/%d/pagemap", opt_pid);
642 pagemap_fd = checked_open(buf, O_RDONLY);
644 sprintf(buf, "/proc/%d/maps", opt_pid);
645 file = fopen(buf, "r");
651 while (fgets(buf, sizeof(buf), file) != NULL) {
652 unsigned long vm_start;
653 unsigned long vm_end;
654 unsigned long long pgoff;
660 n = sscanf(buf, "%lx-%lx %c%c%c%c %llx %x:%x %lu",
668 fprintf(stderr, "unexpected line: %s\n", buf);
671 pg_start[nr_vmas] = vm_start / page_size;
672 pg_end[nr_vmas] = vm_end / page_size;
673 if (++nr_vmas >= MAX_VMAS) {
674 fprintf(stderr, "too many VMAs\n");
681 static void parse_file(const char *name)
685 static void parse_addr_range(const char *optarg)
687 unsigned long offset;
691 p = strchr(optarg, ',');
693 p = strchr(optarg, '+');
697 size = parse_number(p + 1);
699 offset = parse_number(optarg);
703 size = parse_number(p + 1);
706 fatal("invalid range: %lu,%lu\n",
712 offset = parse_number(optarg);
716 add_addr_range(offset, size);
719 static void add_bits_filter(uint64_t mask, uint64_t bits)
721 if (nr_bit_filters >= MAX_BIT_FILTERS)
722 fatal("too much bit filters\n");
724 opt_mask[nr_bit_filters] = mask;
725 opt_bits[nr_bit_filters] = bits;
729 static uint64_t parse_flag_name(const char *str, int len)
736 if (len <= 8 && !strncmp(str, "compound", len))
737 return BITS_COMPOUND;
739 for (i = 0; i < ARRAY_SIZE(page_flag_names); i++) {
740 if (!page_flag_names[i])
742 if (!strncmp(str, page_flag_names[i] + 2, len))
746 return parse_number(str);
749 static uint64_t parse_flag_names(const char *str, int all)
755 if (*p == ',' || *p == '=' || *p == '\0') {
756 if ((*str != '~') || (*str == '~' && all && *++str))
757 flags |= parse_flag_name(str, p - str);
768 static void parse_bits_mask(const char *optarg)
774 p = strchr(optarg, '=');
777 bits = parse_flag_names(p + 1, 0);
779 mask = parse_flag_names(optarg, 0);
780 bits = parse_flag_names(p + 1, 0);
781 } else if (strchr(optarg, '~')) {
782 mask = parse_flag_names(optarg, 1);
783 bits = parse_flag_names(optarg, 0);
785 mask = parse_flag_names(optarg, 0);
789 add_bits_filter(mask, bits);
793 static struct option opts[] = {
794 { "raw" , 0, NULL, 'r' },
795 { "pid" , 1, NULL, 'p' },
796 { "file" , 1, NULL, 'f' },
797 { "addr" , 1, NULL, 'a' },
798 { "bits" , 1, NULL, 'b' },
799 { "list" , 0, NULL, 'l' },
800 { "list-each" , 0, NULL, 'L' },
801 { "no-summary", 0, NULL, 'N' },
802 { "help" , 0, NULL, 'h' },
803 { NULL , 0, NULL, 0 }
806 int main(int argc, char *argv[])
810 page_size = getpagesize();
812 while ((c = getopt_long(argc, argv,
813 "rp:f:a:b:lLNh", opts, NULL)) != -1) {
825 parse_addr_range(optarg);
828 parse_bits_mask(optarg);
848 if (opt_list && opt_pid)
851 printf("offset\tlen\tflags\n");
853 printf("offset\tflags\n");
858 show_page_range(0, 0); /* drain the buffer */