Documentation/vm/page-types.c

   1 /*
   2  * page-types: Tool for querying page flags
   3  *
   4  * Copyright (C) 2009 Intel corporation
   5  * Copyright (C) 2009 Wu Fengguang <fengguang.wu@intel.com>
   6  */
   7
   8 #include <stdio.h>
   9 #include <stdlib.h>
  10 #include <unistd.h>
  11 #include <stdint.h>
  12 #include <stdarg.h>
  13 #include <string.h>
  14 #include <getopt.h>
  15 #include <limits.h>
  16 #include <sys/types.h>
  17 #include <sys/errno.h>
  18 #include <sys/fcntl.h>
  19
  20
  21 /*
  22  * kernel page flags
  23  */
  24
  25 #define KPF_BYTES               8
  26 #define PROC_KPAGEFLAGS         "/proc/kpageflags"
  27
  28 /* copied from kpageflags_read() */
  29 #define KPF_LOCKED              0
  30 #define KPF_ERROR               1
  31 #define KPF_REFERENCED          2
  32 #define KPF_UPTODATE            3
  33 #define KPF_DIRTY               4
  34 #define KPF_LRU                 5
  35 #define KPF_ACTIVE              6
  36 #define KPF_SLAB                7
  37 #define KPF_WRITEBACK           8
  38 #define KPF_RECLAIM             9
  39 #define KPF_BUDDY               10
  40
  41 /* [11-20] new additions in 2.6.31 */
  42 #define KPF_MMAP                11
  43 #define KPF_ANON                12
  44 #define KPF_SWAPCACHE           13
  45 #define KPF_SWAPBACKED          14
  46 #define KPF_COMPOUND_HEAD       15
  47 #define KPF_COMPOUND_TAIL       16
  48 #define KPF_HUGE                17
  49 #define KPF_UNEVICTABLE         18
  50 #define KPF_NOPAGE              20
  51
  52 /* [32-] kernel hacking assistances */
  53 #define KPF_RESERVED            32
  54 #define KPF_MLOCKED             33
  55 #define KPF_MAPPEDTODISK        34
  56 #define KPF_PRIVATE             35
  57 #define KPF_PRIVATE_2           36
  58 #define KPF_OWNER_PRIVATE       37
  59 #define KPF_ARCH                38
  60 #define KPF_UNCACHED            39
  61
  62 /* [48-] take some arbitrary free slots for expanding overloaded flags
  63  * not part of kernel API
  64  */
  65 #define KPF_READAHEAD           48
  66 #define KPF_SLOB_FREE           49
  67 #define KPF_SLUB_FROZEN         50
  68 #define KPF_SLUB_DEBUG          51
  69
  70 #define KPF_ALL_BITS            ((uint64_t)~0ULL)
  71 #define KPF_HACKERS_BITS        (0xffffULL << 32)
  72 #define KPF_OVERLOADED_BITS     (0xffffULL << 48)
  73 #define BIT(name)               (1ULL << KPF_##name)
  74 #define BITS_COMPOUND           (BIT(COMPOUND_HEAD) | BIT(COMPOUND_TAIL))
  75
  76 static char *page_flag_names[] = {
  77         [KPF_LOCKED]            = "L:locked",
  78         [KPF_ERROR]             = "E:error",
  79         [KPF_REFERENCED]        = "R:referenced",
  80         [KPF_UPTODATE]          = "U:uptodate",
  81         [KPF_DIRTY]             = "D:dirty",
  82         [KPF_LRU]               = "l:lru",
  83         [KPF_ACTIVE]            = "A:active",
  84         [KPF_SLAB]              = "S:slab",
  85         [KPF_WRITEBACK]         = "W:writeback",
  86         [KPF_RECLAIM]           = "I:reclaim",
  87         [KPF_BUDDY]             = "B:buddy",
  88
  89         [KPF_MMAP]              = "M:mmap",
  90         [KPF_ANON]              = "a:anonymous",
  91         [KPF_SWAPCACHE]         = "s:swapcache",
  92         [KPF_SWAPBACKED]        = "b:swapbacked",
  93         [KPF_COMPOUND_HEAD]     = "H:compound_head",
  94         [KPF_COMPOUND_TAIL]     = "T:compound_tail",
  95         [KPF_HUGE]              = "G:huge",
  96         [KPF_UNEVICTABLE]       = "u:unevictable",
  97         [KPF_NOPAGE]            = "n:nopage",
  98
  99         [KPF_RESERVED]          = "r:reserved",
 100         [KPF_MLOCKED]           = "m:mlocked",
 101         [KPF_MAPPEDTODISK]      = "d:mappedtodisk",
 102         [KPF_PRIVATE]           = "P:private",
 103         [KPF_PRIVATE_2]         = "p:private_2",
 104         [KPF_OWNER_PRIVATE]     = "O:owner_private",
 105         [KPF_ARCH]              = "h:arch",
 106         [KPF_UNCACHED]          = "c:uncached",
 107
 108         [KPF_READAHEAD]         = "I:readahead",
 109         [KPF_SLOB_FREE]         = "P:slob_free",
 110         [KPF_SLUB_FROZEN]       = "A:slub_frozen",
 111         [KPF_SLUB_DEBUG]        = "E:slub_debug",
 112 };
 113
 114
 115 /*
 116  * data structures
 117  */
 118
 119 static int              opt_raw;        /* for kernel developers */
 120 static int              opt_list;       /* list pages (in ranges) */
 121 static int              opt_no_summary; /* don't show summary */
 122 static pid_t            opt_pid;        /* process to walk */
 123
 124 #define MAX_ADDR_RANGES 1024
 125 static int              nr_addr_ranges;
 126 static unsigned long    opt_offset[MAX_ADDR_RANGES];
 127 static unsigned long    opt_size[MAX_ADDR_RANGES];
 128
 129 #define MAX_BIT_FILTERS 64
 130 static int              nr_bit_filters;
 131 static uint64_t         opt_mask[MAX_BIT_FILTERS];
 132 static uint64_t         opt_bits[MAX_BIT_FILTERS];
 133
 134 static int              page_size;
 135
 136 #define PAGES_BATCH     (64 << 10)      /* 64k pages */
 137 static int              kpageflags_fd;
 138 static uint64_t         kpageflags_buf[KPF_BYTES * PAGES_BATCH];
 139
 140 #define HASH_SHIFT      13
 141 #define HASH_SIZE       (1 << HASH_SHIFT)
 142 #define HASH_MASK       (HASH_SIZE - 1)
 143 #define HASH_KEY(flags) (flags & HASH_MASK)
 144
 145 static unsigned long    total_pages;
 146 static unsigned long    nr_pages[HASH_SIZE];
 147 static uint64_t         page_flags[HASH_SIZE];
 148
 149
 150 /*
 151  * helper functions
 152  */
 153
 154 #define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0]))
 155
 156 #define min_t(type, x, y) ({                    \
 157         type __min1 = (x);                      \
 158         type __min2 = (y);                      \
 159         __min1 < __min2 ? __min1 : __min2; })
 160
 161 unsigned long pages2mb(unsigned long pages)
 162 {
 163         return (pages * page_size) >> 20;
 164 }
 165
 166 void fatal(const char *x, ...)
 167 {
 168         va_list ap;
 169
 170         va_start(ap, x);
 171         vfprintf(stderr, x, ap);
 172         va_end(ap);
 173         exit(EXIT_FAILURE);
 174 }
 175
 176
 177 /*
 178  * page flag names
 179  */
 180
 181 char *page_flag_name(uint64_t flags)
 182 {
 183         static char buf[65];
 184         int present;
 185         int i, j;
 186
 187         for (i = 0, j = 0; i < ARRAY_SIZE(page_flag_names); i++) {
 188                 present = (flags >> i) & 1;
 189                 if (!page_flag_names[i]) {
 190                         if (present)
 191                                 fatal("unkown flag bit %d\n", i);
 192                         continue;
 193                 }
 194                 buf[j++] = present ? page_flag_names[i][0] : '_';
 195         }
 196
 197         return buf;
 198 }
 199
 200 char *page_flag_longname(uint64_t flags)
 201 {
 202         static char buf[1024];
 203         int i, n;
 204
 205         for (i = 0, n = 0; i < ARRAY_SIZE(page_flag_names); i++) {
 206                 if (!page_flag_names[i])
 207                         continue;
 208                 if ((flags >> i) & 1)
 209                         n += snprintf(buf + n, sizeof(buf) - n, "%s,",
 210                                         page_flag_names[i] + 2);
 211         }
 212         if (n)
 213                 n--;
 214         buf[n] = '\0';
 215
 216         return buf;
 217 }
 218
 219
 220 /*
 221  * page list and summary
 222  */
 223
 224 void show_page_range(unsigned long offset, uint64_t flags)
 225 {
 226         static uint64_t      flags0;
 227         static unsigned long index;
 228         static unsigned long count;
 229
 230         if (flags == flags0 && offset == index + count) {
 231                 count++;
 232                 return;
 233         }
 234
 235         if (count)
 236                 printf("%lu\t%lu\t%s\n",
 237                                 index, count, page_flag_name(flags0));
 238
 239         flags0 = flags;
 240         index  = offset;
 241         count  = 1;
 242 }
 243
 244 void show_page(unsigned long offset, uint64_t flags)
 245 {
 246         printf("%lu\t%s\n", offset, page_flag_name(flags));
 247 }
 248
 249 void show_summary(void)
 250 {
 251         int i;
 252
 253         printf("             flags\tpage-count       MB"
 254                 "  symbolic-flags\t\t\tlong-symbolic-flags\n");
 255
 256         for (i = 0; i < ARRAY_SIZE(nr_pages); i++) {
 257                 if (nr_pages[i])
 258                         printf("0x%016llx\t%10lu %8lu  %s\t%s\n",
 259                                 (unsigned long long)page_flags[i],
 260                                 nr_pages[i],
 261                                 pages2mb(nr_pages[i]),
 262                                 page_flag_name(page_flags[i]),
 263                                 page_flag_longname(page_flags[i]));
 264         }
 265
 266         printf("             total\t%10lu %8lu\n",
 267                         total_pages, pages2mb(total_pages));
 268 }
 269
 270
 271 /*
 272  * page flag filters
 273  */
 274
 275 int bit_mask_ok(uint64_t flags)
 276 {
 277         int i;
 278
 279         for (i = 0; i < nr_bit_filters; i++) {
 280                 if (opt_bits[i] == KPF_ALL_BITS) {
 281                         if ((flags & opt_mask[i]) == 0)
 282                                 return 0;
 283                 } else {
 284                         if ((flags & opt_mask[i]) != opt_bits[i])
 285                                 return 0;
 286                 }
 287         }
 288
 289         return 1;
 290 }
 291
 292 uint64_t expand_overloaded_flags(uint64_t flags)
 293 {
 294         /* SLOB/SLUB overload several page flags */
 295         if (flags & BIT(SLAB)) {
 296                 if (flags & BIT(PRIVATE))
 297                         flags ^= BIT(PRIVATE) | BIT(SLOB_FREE);
 298                 if (flags & BIT(ACTIVE))
 299                         flags ^= BIT(ACTIVE) | BIT(SLUB_FROZEN);
 300                 if (flags & BIT(ERROR))
 301                         flags ^= BIT(ERROR) | BIT(SLUB_DEBUG);
 302         }
 303
 304         /* PG_reclaim is overloaded as PG_readahead in the read path */
 305         if ((flags & (BIT(RECLAIM) | BIT(WRITEBACK))) == BIT(RECLAIM))
 306                 flags ^= BIT(RECLAIM) | BIT(READAHEAD);
 307
 308         return flags;
 309 }
 310
 311 uint64_t well_known_flags(uint64_t flags)
 312 {
 313         /* hide flags intended only for kernel hacker */
 314         flags &= ~KPF_HACKERS_BITS;
 315
 316         /* hide non-hugeTLB compound pages */
 317         if ((flags & BITS_COMPOUND) && !(flags & BIT(HUGE)))
 318                 flags &= ~BITS_COMPOUND;
 319
 320         return flags;
 321 }
 322
 323
 324 /*
 325  * page frame walker
 326  */
 327
 328 int hash_slot(uint64_t flags)
 329 {
 330         int k = HASH_KEY(flags);
 331         int i;
 332
 333         /* Explicitly reserve slot 0 for flags 0: the following logic
 334          * cannot distinguish an unoccupied slot from slot (flags==0).
 335          */
 336         if (flags == 0)
 337                 return 0;
 338
 339         /* search through the remaining (HASH_SIZE-1) slots */
 340         for (i = 1; i < ARRAY_SIZE(page_flags); i++, k++) {
 341                 if (!k || k >= ARRAY_SIZE(page_flags))
 342                         k = 1;
 343                 if (page_flags[k] == 0) {
 344                         page_flags[k] = flags;
 345                         return k;
 346                 }
 347                 if (page_flags[k] == flags)
 348                         return k;
 349         }
 350
 351         fatal("hash table full: bump up HASH_SHIFT?\n");
 352         exit(EXIT_FAILURE);
 353 }
 354
 355 void add_page(unsigned long offset, uint64_t flags)
 356 {
 357         flags = expand_overloaded_flags(flags);
 358
 359         if (!opt_raw)
 360                 flags = well_known_flags(flags);
 361
 362         if (!bit_mask_ok(flags))
 363                 return;
 364
 365         if (opt_list == 1)
 366                 show_page_range(offset, flags);
 367         else if (opt_list == 2)
 368                 show_page(offset, flags);
 369
 370         nr_pages[hash_slot(flags)]++;
 371         total_pages++;
 372 }
 373
 374 void walk_pfn(unsigned long index, unsigned long count)
 375 {
 376         unsigned long batch;
 377         unsigned long n;
 378         unsigned long i;
 379
 380         if (index > ULONG_MAX / KPF_BYTES)
 381                 fatal("index overflow: %lu\n", index);
 382
 383         lseek(kpageflags_fd, index * KPF_BYTES, SEEK_SET);
 384
 385         while (count) {
 386                 batch = min_t(unsigned long, count, PAGES_BATCH);
 387                 n = read(kpageflags_fd, kpageflags_buf, batch * KPF_BYTES);
 388                 if (n == 0)
 389                         break;
 390                 if (n < 0) {
 391                         perror(PROC_KPAGEFLAGS);
 392                         exit(EXIT_FAILURE);
 393                 }
 394
 395                 if (n % KPF_BYTES != 0)
 396                         fatal("partial read: %lu bytes\n", n);
 397                 n = n / KPF_BYTES;
 398
 399                 for (i = 0; i < n; i++)
 400                         add_page(index + i, kpageflags_buf[i]);
 401
 402                 index += batch;
 403                 count -= batch;
 404         }
 405 }
 406
 407 void walk_addr_ranges(void)
 408 {
 409         int i;
 410
 411         kpageflags_fd = open(PROC_KPAGEFLAGS, O_RDONLY);
 412         if (kpageflags_fd < 0) {
 413                 perror(PROC_KPAGEFLAGS);
 414                 exit(EXIT_FAILURE);
 415         }
 416
 417         if (!nr_addr_ranges)
 418                 walk_pfn(0, ULONG_MAX);
 419
 420         for (i = 0; i < nr_addr_ranges; i++)
 421                 walk_pfn(opt_offset[i], opt_size[i]);
 422
 423         close(kpageflags_fd);
 424 }
 425
 426
 427 /*
 428  * user interface
 429  */
 430
 431 const char *page_flag_type(uint64_t flag)
 432 {
 433         if (flag & KPF_HACKERS_BITS)
 434                 return "(r)";
 435         if (flag & KPF_OVERLOADED_BITS)
 436                 return "(o)";
 437         return "   ";
 438 }
 439
 440 void usage(void)
 441 {
 442         int i, j;
 443
 444         printf(
 445 "page-types [options]\n"
 446 "            -r|--raw                  Raw mode, for kernel developers\n"
 447 "            -a|--addr    addr-spec    Walk a range of pages\n"
 448 "            -b|--bits    bits-spec    Walk pages with specified bits\n"
 449 #if 0 /* planned features */
 450 "            -p|--pid     pid          Walk process address space\n"
 451 "            -f|--file    filename     Walk file address space\n"
 452 #endif
 453 "            -l|--list                 Show page details in ranges\n"
 454 "            -L|--list-each            Show page details one by one\n"
 455 "            -N|--no-summary           Don't show summay info\n"
 456 "            -h|--help                 Show this usage message\n"
 457 "addr-spec:\n"
 458 "            N                         one page at offset N (unit: pages)\n"
 459 "            N+M                       pages range from N to N+M-1\n"
 460 "            N,M                       pages range from N to M-1\n"
 461 "            N,                        pages range from N to end\n"
 462 "            ,M                        pages range from 0 to M\n"
 463 "bits-spec:\n"
 464 "            bit1,bit2                 (flags & (bit1|bit2)) != 0\n"
 465 "            bit1,bit2=bit1            (flags & (bit1|bit2)) == bit1\n"
 466 "            bit1,~bit2                (flags & (bit1|bit2)) == bit1\n"
 467 "            =bit1,bit2                flags == (bit1|bit2)\n"
 468 "bit-names:\n"
 469         );
 470
 471         for (i = 0, j = 0; i < ARRAY_SIZE(page_flag_names); i++) {
 472                 if (!page_flag_names[i])
 473                         continue;
 474                 printf("%16s%s", page_flag_names[i] + 2,
 475                                  page_flag_type(1ULL << i));
 476                 if (++j > 3) {
 477                         j = 0;
 478                         putchar('\n');
 479                 }
 480         }
 481         printf("\n                                   "
 482                 "(r) raw mode bits  (o) overloaded bits\n");
 483 }
 484
 485 unsigned long long parse_number(const char *str)
 486 {
 487         unsigned long long n;
 488
 489         n = strtoll(str, NULL, 0);
 490
 491         if (n == 0 && str[0] != '0')
 492                 fatal("invalid name or number: %s\n", str);
 493
 494         return n;
 495 }
 496
 497 void parse_pid(const char *str)
 498 {
 499         opt_pid = parse_number(str);
 500 }
 501
 502 void parse_file(const char *name)
 503 {
 504 }
 505
 506 void add_addr_range(unsigned long offset, unsigned long size)
 507 {
 508         if (nr_addr_ranges >= MAX_ADDR_RANGES)
 509                 fatal("too much addr ranges\n");
 510
 511         opt_offset[nr_addr_ranges] = offset;
 512         opt_size[nr_addr_ranges] = size;
 513         nr_addr_ranges++;
 514 }
 515
 516 void parse_addr_range(const char *optarg)
 517 {
 518         unsigned long offset;
 519         unsigned long size;
 520         char *p;
 521
 522         p = strchr(optarg, ',');
 523         if (!p)
 524                 p = strchr(optarg, '+');
 525
 526         if (p == optarg) {
 527                 offset = 0;
 528                 size   = parse_number(p + 1);
 529         } else if (p) {
 530                 offset = parse_number(optarg);
 531                 if (p[1] == '\0')
 532                         size = ULONG_MAX;
 533                 else {
 534                         size = parse_number(p + 1);
 535                         if (*p == ',') {
 536                                 if (size < offset)
 537                                         fatal("invalid range: %lu,%lu\n",
 538                                                         offset, size);
 539                                 size -= offset;
 540                         }
 541                 }
 542         } else {
 543                 offset = parse_number(optarg);
 544                 size   = 1;
 545         }
 546
 547         add_addr_range(offset, size);
 548 }
 549
 550 void add_bits_filter(uint64_t mask, uint64_t bits)
 551 {
 552         if (nr_bit_filters >= MAX_BIT_FILTERS)
 553                 fatal("too much bit filters\n");
 554
 555         opt_mask[nr_bit_filters] = mask;
 556         opt_bits[nr_bit_filters] = bits;
 557         nr_bit_filters++;
 558 }
 559
 560 uint64_t parse_flag_name(const char *str, int len)
 561 {
 562         int i;
 563
 564         if (!*str || !len)
 565                 return 0;
 566
 567         if (len <= 8 && !strncmp(str, "compound", len))
 568                 return BITS_COMPOUND;
 569
 570         for (i = 0; i < ARRAY_SIZE(page_flag_names); i++) {
 571                 if (!page_flag_names[i])
 572                         continue;
 573                 if (!strncmp(str, page_flag_names[i] + 2, len))
 574                         return 1ULL << i;
 575         }
 576
 577         return parse_number(str);
 578 }
 579
 580 uint64_t parse_flag_names(const char *str, int all)
 581 {
 582         const char *p    = str;
 583         uint64_t   flags = 0;
 584
 585         while (1) {
 586                 if (*p == ',' || *p == '=' || *p == '\0') {
 587                         if ((*str != '~') || (*str == '~' && all && *++str))
 588                                 flags |= parse_flag_name(str, p - str);
 589                         if (*p != ',')
 590                                 break;
 591                         str = p + 1;
 592                 }
 593                 p++;
 594         }
 595
 596         return flags;
 597 }
 598
 599 void parse_bits_mask(const char *optarg)
 600 {
 601         uint64_t mask;
 602         uint64_t bits;
 603         const char *p;
 604
 605         p = strchr(optarg, '=');
 606         if (p == optarg) {
 607                 mask = KPF_ALL_BITS;
 608                 bits = parse_flag_names(p + 1, 0);
 609         } else if (p) {
 610                 mask = parse_flag_names(optarg, 0);
 611                 bits = parse_flag_names(p + 1, 0);
 612         } else if (strchr(optarg, '~')) {
 613                 mask = parse_flag_names(optarg, 1);
 614                 bits = parse_flag_names(optarg, 0);
 615         } else {
 616                 mask = parse_flag_names(optarg, 0);
 617                 bits = KPF_ALL_BITS;
 618         }
 619
 620         add_bits_filter(mask, bits);
 621 }
 622
 623
 624 struct option opts[] = {
 625         { "raw"       , 0, NULL, 'r' },
 626         { "pid"       , 1, NULL, 'p' },
 627         { "file"      , 1, NULL, 'f' },
 628         { "addr"      , 1, NULL, 'a' },
 629         { "bits"      , 1, NULL, 'b' },
 630         { "list"      , 0, NULL, 'l' },
 631         { "list-each" , 0, NULL, 'L' },
 632         { "no-summary", 0, NULL, 'N' },
 633         { "help"      , 0, NULL, 'h' },
 634         { NULL        , 0, NULL, 0 }
 635 };
 636
 637 int main(int argc, char *argv[])
 638 {
 639         int c;
 640
 641         page_size = getpagesize();
 642
 643         while ((c = getopt_long(argc, argv,
 644                                 "rp:f:a:b:lLNh", opts, NULL)) != -1) {
 645                 switch (c) {
 646                 case 'r':
 647                         opt_raw = 1;
 648                         break;
 649                 case 'p':
 650                         parse_pid(optarg);
 651                         break;
 652                 case 'f':
 653                         parse_file(optarg);
 654                         break;
 655                 case 'a':
 656                         parse_addr_range(optarg);
 657                         break;
 658                 case 'b':
 659                         parse_bits_mask(optarg);
 660                         break;
 661                 case 'l':
 662                         opt_list = 1;
 663                         break;
 664                 case 'L':
 665                         opt_list = 2;
 666                         break;
 667                 case 'N':
 668                         opt_no_summary = 1;
 669                         break;
 670                 case 'h':
 671                         usage();
 672                         exit(0);
 673                 default:
 674                         usage();
 675                         exit(1);
 676                 }
 677         }
 678
 679         if (opt_list == 1)
 680                 printf("offset\tcount\tflags\n");
 681         if (opt_list == 2)
 682                 printf("offset\tflags\n");
 683
 684         walk_addr_ranges();
 685
 686         if (opt_list == 1)
 687                 show_page_range(0, 0);  /* drain the buffer */
 688
 689         if (opt_no_summary)
 690                 return 0;
 691
 692         if (opt_list)
 693                 printf("\n\n");
 694
 695         show_summary();
 696
 697         return 0;
 698 }