linux/tools/vm/page-types.c
<<
>>
Prefs
   1/*
   2 * page-types: Tool for querying page flags
   3 *
   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.
   7 *
   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
  11 * more details.
  12 *
  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.
  16 *
  17 * Copyright (C) 2009 Intel corporation
  18 *
  19 * Authors: Wu Fengguang <fengguang.wu@intel.com>
  20 */
  21
  22#define _LARGEFILE64_SOURCE
  23#include <stdio.h>
  24#include <stdlib.h>
  25#include <unistd.h>
  26#include <stdint.h>
  27#include <stdarg.h>
  28#include <string.h>
  29#include <getopt.h>
  30#include <limits.h>
  31#include <assert.h>
  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 <lk/debugfs.h>
  40
  41#ifndef MAX_PATH
  42# define MAX_PATH 256
  43#endif
  44
  45#ifndef STR
  46# define _STR(x) #x
  47# define STR(x) _STR(x)
  48#endif
  49
  50/*
  51 * pagemap kernel ABI bits
  52 */
  53
  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)        (((u64) (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)
  65
  66#define PM_PRESENT          PM_STATUS(4LL)
  67#define PM_SWAP             PM_STATUS(2LL)
  68
  69
  70/*
  71 * kernel page flags
  72 */
  73
  74#define KPF_BYTES               8
  75#define PROC_KPAGEFLAGS         "/proc/kpageflags"
  76
  77/* [32-] kernel hacking assistances */
  78#define KPF_RESERVED            32
  79#define KPF_MLOCKED             33
  80#define KPF_MAPPEDTODISK        34
  81#define KPF_PRIVATE             35
  82#define KPF_PRIVATE_2           36
  83#define KPF_OWNER_PRIVATE       37
  84#define KPF_ARCH                38
  85#define KPF_UNCACHED            39
  86
  87/* [48-] take some arbitrary free slots for expanding overloaded flags
  88 * not part of kernel API
  89 */
  90#define KPF_READAHEAD           48
  91#define KPF_SLOB_FREE           49
  92#define KPF_SLUB_FROZEN         50
  93#define KPF_SLUB_DEBUG          51
  94
  95#define KPF_ALL_BITS            ((uint64_t)~0ULL)
  96#define KPF_HACKERS_BITS        (0xffffULL << 32)
  97#define KPF_OVERLOADED_BITS     (0xffffULL << 48)
  98#define BIT(name)               (1ULL << KPF_##name)
  99#define BITS_COMPOUND           (BIT(COMPOUND_HEAD) | BIT(COMPOUND_TAIL))
 100
 101static const char * const page_flag_names[] = {
 102        [KPF_LOCKED]            = "L:locked",
 103        [KPF_ERROR]             = "E:error",
 104        [KPF_REFERENCED]        = "R:referenced",
 105        [KPF_UPTODATE]          = "U:uptodate",
 106        [KPF_DIRTY]             = "D:dirty",
 107        [KPF_LRU]               = "l:lru",
 108        [KPF_ACTIVE]            = "A:active",
 109        [KPF_SLAB]              = "S:slab",
 110        [KPF_WRITEBACK]         = "W:writeback",
 111        [KPF_RECLAIM]           = "I:reclaim",
 112        [KPF_BUDDY]             = "B:buddy",
 113
 114        [KPF_MMAP]              = "M:mmap",
 115        [KPF_ANON]              = "a:anonymous",
 116        [KPF_SWAPCACHE]         = "s:swapcache",
 117        [KPF_SWAPBACKED]        = "b:swapbacked",
 118        [KPF_COMPOUND_HEAD]     = "H:compound_head",
 119        [KPF_COMPOUND_TAIL]     = "T:compound_tail",
 120        [KPF_HUGE]              = "G:huge",
 121        [KPF_UNEVICTABLE]       = "u:unevictable",
 122        [KPF_HWPOISON]          = "X:hwpoison",
 123        [KPF_NOPAGE]            = "n:nopage",
 124        [KPF_KSM]               = "x:ksm",
 125        [KPF_THP]               = "t:thp",
 126
 127        [KPF_RESERVED]          = "r:reserved",
 128        [KPF_MLOCKED]           = "m:mlocked",
 129        [KPF_MAPPEDTODISK]      = "d:mappedtodisk",
 130        [KPF_PRIVATE]           = "P:private",
 131        [KPF_PRIVATE_2]         = "p:private_2",
 132        [KPF_OWNER_PRIVATE]     = "O:owner_private",
 133        [KPF_ARCH]              = "h:arch",
 134        [KPF_UNCACHED]          = "c:uncached",
 135
 136        [KPF_READAHEAD]         = "I:readahead",
 137        [KPF_SLOB_FREE]         = "P:slob_free",
 138        [KPF_SLUB_FROZEN]       = "A:slub_frozen",
 139        [KPF_SLUB_DEBUG]        = "E:slub_debug",
 140};
 141
 142
 143static const char * const debugfs_known_mountpoints[] = {
 144        "/sys/kernel/debug",
 145        "/debug",
 146        0,
 147};
 148
 149/*
 150 * data structures
 151 */
 152
 153static int              opt_raw;        /* for kernel developers */
 154static int              opt_list;       /* list pages (in ranges) */
 155static int              opt_no_summary; /* don't show summary */
 156static pid_t            opt_pid;        /* process to walk */
 157
 158#define MAX_ADDR_RANGES 1024
 159static int              nr_addr_ranges;
 160static unsigned long    opt_offset[MAX_ADDR_RANGES];
 161static unsigned long    opt_size[MAX_ADDR_RANGES];
 162
 163#define MAX_VMAS        10240
 164static int              nr_vmas;
 165static unsigned long    pg_start[MAX_VMAS];
 166static unsigned long    pg_end[MAX_VMAS];
 167
 168#define MAX_BIT_FILTERS 64
 169static int              nr_bit_filters;
 170static uint64_t         opt_mask[MAX_BIT_FILTERS];
 171static uint64_t         opt_bits[MAX_BIT_FILTERS];
 172
 173static int              page_size;
 174
 175static int              pagemap_fd;
 176static int              kpageflags_fd;
 177
 178static int              opt_hwpoison;
 179static int              opt_unpoison;
 180
 181static char             *hwpoison_debug_fs;
 182static int              hwpoison_inject_fd;
 183static int              hwpoison_forget_fd;
 184
 185#define HASH_SHIFT      13
 186#define HASH_SIZE       (1 << HASH_SHIFT)
 187#define HASH_MASK       (HASH_SIZE - 1)
 188#define HASH_KEY(flags) (flags & HASH_MASK)
 189
 190static unsigned long    total_pages;
 191static unsigned long    nr_pages[HASH_SIZE];
 192static uint64_t         page_flags[HASH_SIZE];
 193
 194
 195/*
 196 * helper functions
 197 */
 198
 199#define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0]))
 200
 201#define min_t(type, x, y) ({                    \
 202        type __min1 = (x);                      \
 203        type __min2 = (y);                      \
 204        __min1 < __min2 ? __min1 : __min2; })
 205
 206#define max_t(type, x, y) ({                    \
 207        type __max1 = (x);                      \
 208        type __max2 = (y);                      \
 209        __max1 > __max2 ? __max1 : __max2; })
 210
 211static unsigned long pages2mb(unsigned long pages)
 212{
 213        return (pages * page_size) >> 20;
 214}
 215
 216static void fatal(const char *x, ...)
 217{
 218        va_list ap;
 219
 220        va_start(ap, x);
 221        vfprintf(stderr, x, ap);
 222        va_end(ap);
 223        exit(EXIT_FAILURE);
 224}
 225
 226static int checked_open(const char *pathname, int flags)
 227{
 228        int fd = open(pathname, flags);
 229
 230        if (fd < 0) {
 231                perror(pathname);
 232                exit(EXIT_FAILURE);
 233        }
 234
 235        return fd;
 236}
 237
 238/*
 239 * pagemap/kpageflags routines
 240 */
 241
 242static unsigned long do_u64_read(int fd, char *name,
 243                                 uint64_t *buf,
 244                                 unsigned long index,
 245                                 unsigned long count)
 246{
 247        long bytes;
 248
 249        if (index > ULONG_MAX / 8)
 250                fatal("index overflow: %lu\n", index);
 251
 252        if (lseek(fd, index * 8, SEEK_SET) < 0) {
 253                perror(name);
 254                exit(EXIT_FAILURE);
 255        }
 256
 257        bytes = read(fd, buf, count * 8);
 258        if (bytes < 0) {
 259                perror(name);
 260                exit(EXIT_FAILURE);
 261        }
 262        if (bytes % 8)
 263                fatal("partial read: %lu bytes\n", bytes);
 264
 265        return bytes / 8;
 266}
 267
 268static unsigned long kpageflags_read(uint64_t *buf,
 269                                     unsigned long index,
 270                                     unsigned long pages)
 271{
 272        return do_u64_read(kpageflags_fd, PROC_KPAGEFLAGS, buf, index, pages);
 273}
 274
 275static unsigned long pagemap_read(uint64_t *buf,
 276                                  unsigned long index,
 277                                  unsigned long pages)
 278{
 279        return do_u64_read(pagemap_fd, "/proc/pid/pagemap", buf, index, pages);
 280}
 281
 282static unsigned long pagemap_pfn(uint64_t val)
 283{
 284        unsigned long pfn;
 285
 286        if (val & PM_PRESENT)
 287                pfn = PM_PFRAME(val);
 288        else
 289                pfn = 0;
 290
 291        return pfn;
 292}
 293
 294
 295/*
 296 * page flag names
 297 */
 298
 299static char *page_flag_name(uint64_t flags)
 300{
 301        static char buf[65];
 302        int present;
 303        size_t i, j;
 304
 305        for (i = 0, j = 0; i < ARRAY_SIZE(page_flag_names); i++) {
 306                present = (flags >> i) & 1;
 307                if (!page_flag_names[i]) {
 308                        if (present)
 309                                fatal("unknown flag bit %d\n", i);
 310                        continue;
 311                }
 312                buf[j++] = present ? page_flag_names[i][0] : '_';
 313        }
 314
 315        return buf;
 316}
 317
 318static char *page_flag_longname(uint64_t flags)
 319{
 320        static char buf[1024];
 321        size_t i, n;
 322
 323        for (i = 0, n = 0; i < ARRAY_SIZE(page_flag_names); i++) {
 324                if (!page_flag_names[i])
 325                        continue;
 326                if ((flags >> i) & 1)
 327                        n += snprintf(buf + n, sizeof(buf) - n, "%s,",
 328                                        page_flag_names[i] + 2);
 329        }
 330        if (n)
 331                n--;
 332        buf[n] = '\0';
 333
 334        return buf;
 335}
 336
 337
 338/*
 339 * page list and summary
 340 */
 341
 342static void show_page_range(unsigned long voffset,
 343                            unsigned long offset, uint64_t flags)
 344{
 345        static uint64_t      flags0;
 346        static unsigned long voff;
 347        static unsigned long index;
 348        static unsigned long count;
 349
 350        if (flags == flags0 && offset == index + count &&
 351            (!opt_pid || voffset == voff + count)) {
 352                count++;
 353                return;
 354        }
 355
 356        if (count) {
 357                if (opt_pid)
 358                        printf("%lx\t", voff);
 359                printf("%lx\t%lx\t%s\n",
 360                                index, count, page_flag_name(flags0));
 361        }
 362
 363        flags0 = flags;
 364        index  = offset;
 365        voff   = voffset;
 366        count  = 1;
 367}
 368
 369static void show_page(unsigned long voffset,
 370                      unsigned long offset, uint64_t flags)
 371{
 372        if (opt_pid)
 373                printf("%lx\t", voffset);
 374        printf("%lx\t%s\n", offset, page_flag_name(flags));
 375}
 376
 377static void show_summary(void)
 378{
 379        size_t i;
 380
 381        printf("             flags\tpage-count       MB"
 382                "  symbolic-flags\t\t\tlong-symbolic-flags\n");
 383
 384        for (i = 0; i < ARRAY_SIZE(nr_pages); i++) {
 385                if (nr_pages[i])
 386                        printf("0x%016llx\t%10lu %8lu  %s\t%s\n",
 387                                (unsigned long long)page_flags[i],
 388                                nr_pages[i],
 389                                pages2mb(nr_pages[i]),
 390                                page_flag_name(page_flags[i]),
 391                                page_flag_longname(page_flags[i]));
 392        }
 393
 394        printf("             total\t%10lu %8lu\n",
 395                        total_pages, pages2mb(total_pages));
 396}
 397
 398
 399/*
 400 * page flag filters
 401 */
 402
 403static int bit_mask_ok(uint64_t flags)
 404{
 405        int i;
 406
 407        for (i = 0; i < nr_bit_filters; i++) {
 408                if (opt_bits[i] == KPF_ALL_BITS) {
 409                        if ((flags & opt_mask[i]) == 0)
 410                                return 0;
 411                } else {
 412                        if ((flags & opt_mask[i]) != opt_bits[i])
 413                                return 0;
 414                }
 415        }
 416
 417        return 1;
 418}
 419
 420static uint64_t expand_overloaded_flags(uint64_t flags)
 421{
 422        /* SLOB/SLUB overload several page flags */
 423        if (flags & BIT(SLAB)) {
 424                if (flags & BIT(PRIVATE))
 425                        flags ^= BIT(PRIVATE) | BIT(SLOB_FREE);
 426                if (flags & BIT(ACTIVE))
 427                        flags ^= BIT(ACTIVE) | BIT(SLUB_FROZEN);
 428                if (flags & BIT(ERROR))
 429                        flags ^= BIT(ERROR) | BIT(SLUB_DEBUG);
 430        }
 431
 432        /* PG_reclaim is overloaded as PG_readahead in the read path */
 433        if ((flags & (BIT(RECLAIM) | BIT(WRITEBACK))) == BIT(RECLAIM))
 434                flags ^= BIT(RECLAIM) | BIT(READAHEAD);
 435
 436        return flags;
 437}
 438
 439static uint64_t well_known_flags(uint64_t flags)
 440{
 441        /* hide flags intended only for kernel hacker */
 442        flags &= ~KPF_HACKERS_BITS;
 443
 444        /* hide non-hugeTLB compound pages */
 445        if ((flags & BITS_COMPOUND) && !(flags & BIT(HUGE)))
 446                flags &= ~BITS_COMPOUND;
 447
 448        return flags;
 449}
 450
 451static uint64_t kpageflags_flags(uint64_t flags)
 452{
 453        flags = expand_overloaded_flags(flags);
 454
 455        if (!opt_raw)
 456                flags = well_known_flags(flags);
 457
 458        return flags;
 459}
 460
 461/*
 462 * page actions
 463 */
 464
 465static void prepare_hwpoison_fd(void)
 466{
 467        char buf[MAX_PATH + 1];
 468
 469        hwpoison_debug_fs = debugfs_mount(NULL);
 470        if (!hwpoison_debug_fs) {
 471                perror("mount debugfs");
 472                exit(EXIT_FAILURE);
 473        }
 474
 475        if (opt_hwpoison && !hwpoison_inject_fd) {
 476                snprintf(buf, MAX_PATH, "%s/hwpoison/corrupt-pfn",
 477                        hwpoison_debug_fs);
 478                hwpoison_inject_fd = checked_open(buf, O_WRONLY);
 479        }
 480
 481        if (opt_unpoison && !hwpoison_forget_fd) {
 482                snprintf(buf, MAX_PATH, "%s/hwpoison/unpoison-pfn",
 483                        hwpoison_debug_fs);
 484                hwpoison_forget_fd = checked_open(buf, O_WRONLY);
 485        }
 486}
 487
 488static int hwpoison_page(unsigned long offset)
 489{
 490        char buf[100];
 491        int len;
 492
 493        len = sprintf(buf, "0x%lx\n", offset);
 494        len = write(hwpoison_inject_fd, buf, len);
 495        if (len < 0) {
 496                perror("hwpoison inject");
 497                return len;
 498        }
 499        return 0;
 500}
 501
 502static int unpoison_page(unsigned long offset)
 503{
 504        char buf[100];
 505        int len;
 506
 507        len = sprintf(buf, "0x%lx\n", offset);
 508        len = write(hwpoison_forget_fd, buf, len);
 509        if (len < 0) {
 510                perror("hwpoison forget");
 511                return len;
 512        }
 513        return 0;
 514}
 515
 516/*
 517 * page frame walker
 518 */
 519
 520static size_t hash_slot(uint64_t flags)
 521{
 522        size_t k = HASH_KEY(flags);
 523        size_t i;
 524
 525        /* Explicitly reserve slot 0 for flags 0: the following logic
 526         * cannot distinguish an unoccupied slot from slot (flags==0).
 527         */
 528        if (flags == 0)
 529                return 0;
 530
 531        /* search through the remaining (HASH_SIZE-1) slots */
 532        for (i = 1; i < ARRAY_SIZE(page_flags); i++, k++) {
 533                if (!k || k >= ARRAY_SIZE(page_flags))
 534                        k = 1;
 535                if (page_flags[k] == 0) {
 536                        page_flags[k] = flags;
 537                        return k;
 538                }
 539                if (page_flags[k] == flags)
 540                        return k;
 541        }
 542
 543        fatal("hash table full: bump up HASH_SHIFT?\n");
 544        exit(EXIT_FAILURE);
 545}
 546
 547static void add_page(unsigned long voffset,
 548                     unsigned long offset, uint64_t flags)
 549{
 550        flags = kpageflags_flags(flags);
 551
 552        if (!bit_mask_ok(flags))
 553                return;
 554
 555        if (opt_hwpoison)
 556                hwpoison_page(offset);
 557        if (opt_unpoison)
 558                unpoison_page(offset);
 559
 560        if (opt_list == 1)
 561                show_page_range(voffset, offset, flags);
 562        else if (opt_list == 2)
 563                show_page(voffset, offset, flags);
 564
 565        nr_pages[hash_slot(flags)]++;
 566        total_pages++;
 567}
 568
 569#define KPAGEFLAGS_BATCH        (64 << 10)      /* 64k pages */
 570static void walk_pfn(unsigned long voffset,
 571                     unsigned long index,
 572                     unsigned long count)
 573{
 574        uint64_t buf[KPAGEFLAGS_BATCH];
 575        unsigned long batch;
 576        unsigned long pages;
 577        unsigned long i;
 578
 579        while (count) {
 580                batch = min_t(unsigned long, count, KPAGEFLAGS_BATCH);
 581                pages = kpageflags_read(buf, index, batch);
 582                if (pages == 0)
 583                        break;
 584
 585                for (i = 0; i < pages; i++)
 586                        add_page(voffset + i, index + i, buf[i]);
 587
 588                index += pages;
 589                count -= pages;
 590        }
 591}
 592
 593#define PAGEMAP_BATCH   (64 << 10)
 594static void walk_vma(unsigned long index, unsigned long count)
 595{
 596        uint64_t buf[PAGEMAP_BATCH];
 597        unsigned long batch;
 598        unsigned long pages;
 599        unsigned long pfn;
 600        unsigned long i;
 601
 602        while (count) {
 603                batch = min_t(unsigned long, count, PAGEMAP_BATCH);
 604                pages = pagemap_read(buf, index, batch);
 605                if (pages == 0)
 606                        break;
 607
 608                for (i = 0; i < pages; i++) {
 609                        pfn = pagemap_pfn(buf[i]);
 610                        if (pfn)
 611                                walk_pfn(index + i, pfn, 1);
 612                }
 613
 614                index += pages;
 615                count -= pages;
 616        }
 617}
 618
 619static void walk_task(unsigned long index, unsigned long count)
 620{
 621        const unsigned long end = index + count;
 622        unsigned long start;
 623        int i = 0;
 624
 625        while (index < end) {
 626
 627                while (pg_end[i] <= index)
 628                        if (++i >= nr_vmas)
 629                                return;
 630                if (pg_start[i] >= end)
 631                        return;
 632
 633                start = max_t(unsigned long, pg_start[i], index);
 634                index = min_t(unsigned long, pg_end[i], end);
 635
 636                assert(start < index);
 637                walk_vma(start, index - start);
 638        }
 639}
 640
 641static void add_addr_range(unsigned long offset, unsigned long size)
 642{
 643        if (nr_addr_ranges >= MAX_ADDR_RANGES)
 644                fatal("too many addr ranges\n");
 645
 646        opt_offset[nr_addr_ranges] = offset;
 647        opt_size[nr_addr_ranges] = min_t(unsigned long, size, ULONG_MAX-offset);
 648        nr_addr_ranges++;
 649}
 650
 651static void walk_addr_ranges(void)
 652{
 653        int i;
 654
 655        kpageflags_fd = checked_open(PROC_KPAGEFLAGS, O_RDONLY);
 656
 657        if (!nr_addr_ranges)
 658                add_addr_range(0, ULONG_MAX);
 659
 660        for (i = 0; i < nr_addr_ranges; i++)
 661                if (!opt_pid)
 662                        walk_pfn(0, opt_offset[i], opt_size[i]);
 663                else
 664                        walk_task(opt_offset[i], opt_size[i]);
 665
 666        close(kpageflags_fd);
 667}
 668
 669
 670/*
 671 * user interface
 672 */
 673
 674static const char *page_flag_type(uint64_t flag)
 675{
 676        if (flag & KPF_HACKERS_BITS)
 677                return "(r)";
 678        if (flag & KPF_OVERLOADED_BITS)
 679                return "(o)";
 680        return "   ";
 681}
 682
 683static void usage(void)
 684{
 685        size_t i, j;
 686
 687        printf(
 688"page-types [options]\n"
 689"            -r|--raw                   Raw mode, for kernel developers\n"
 690"            -d|--describe flags        Describe flags\n"
 691"            -a|--addr    addr-spec     Walk a range of pages\n"
 692"            -b|--bits    bits-spec     Walk pages with specified bits\n"
 693"            -p|--pid     pid           Walk process address space\n"
 694#if 0 /* planned features */
 695"            -f|--file    filename      Walk file address space\n"
 696#endif
 697"            -l|--list                  Show page details in ranges\n"
 698"            -L|--list-each             Show page details one by one\n"
 699"            -N|--no-summary            Don't show summary info\n"
 700"            -X|--hwpoison              hwpoison pages\n"
 701"            -x|--unpoison              unpoison pages\n"
 702"            -h|--help                  Show this usage message\n"
 703"flags:\n"
 704"            0x10                       bitfield format, e.g.\n"
 705"            anon                       bit-name, e.g.\n"
 706"            0x10,anon                  comma-separated list, e.g.\n"
 707"addr-spec:\n"
 708"            N                          one page at offset N (unit: pages)\n"
 709"            N+M                        pages range from N to N+M-1\n"
 710"            N,M                        pages range from N to M-1\n"
 711"            N,                         pages range from N to end\n"
 712"            ,M                         pages range from 0 to M-1\n"
 713"bits-spec:\n"
 714"            bit1,bit2                  (flags & (bit1|bit2)) != 0\n"
 715"            bit1,bit2=bit1             (flags & (bit1|bit2)) == bit1\n"
 716"            bit1,~bit2                 (flags & (bit1|bit2)) == bit1\n"
 717"            =bit1,bit2                 flags == (bit1|bit2)\n"
 718"bit-names:\n"
 719        );
 720
 721        for (i = 0, j = 0; i < ARRAY_SIZE(page_flag_names); i++) {
 722                if (!page_flag_names[i])
 723                        continue;
 724                printf("%16s%s", page_flag_names[i] + 2,
 725                                 page_flag_type(1ULL << i));
 726                if (++j > 3) {
 727                        j = 0;
 728                        putchar('\n');
 729                }
 730        }
 731        printf("\n                                   "
 732                "(r) raw mode bits  (o) overloaded bits\n");
 733}
 734
 735static unsigned long long parse_number(const char *str)
 736{
 737        unsigned long long n;
 738
 739        n = strtoll(str, NULL, 0);
 740
 741        if (n == 0 && str[0] != '0')
 742                fatal("invalid name or number: %s\n", str);
 743
 744        return n;
 745}
 746
 747static void parse_pid(const char *str)
 748{
 749        FILE *file;
 750        char buf[5000];
 751
 752        opt_pid = parse_number(str);
 753
 754        sprintf(buf, "/proc/%d/pagemap", opt_pid);
 755        pagemap_fd = checked_open(buf, O_RDONLY);
 756
 757        sprintf(buf, "/proc/%d/maps", opt_pid);
 758        file = fopen(buf, "r");
 759        if (!file) {
 760                perror(buf);
 761                exit(EXIT_FAILURE);
 762        }
 763
 764        while (fgets(buf, sizeof(buf), file) != NULL) {
 765                unsigned long vm_start;
 766                unsigned long vm_end;
 767                unsigned long long pgoff;
 768                int major, minor;
 769                char r, w, x, s;
 770                unsigned long ino;
 771                int n;
 772
 773                n = sscanf(buf, "%lx-%lx %c%c%c%c %llx %x:%x %lu",
 774                           &vm_start,
 775                           &vm_end,
 776                           &r, &w, &x, &s,
 777                           &pgoff,
 778                           &major, &minor,
 779                           &ino);
 780                if (n < 10) {
 781                        fprintf(stderr, "unexpected line: %s\n", buf);
 782                        continue;
 783                }
 784                pg_start[nr_vmas] = vm_start / page_size;
 785                pg_end[nr_vmas] = vm_end / page_size;
 786                if (++nr_vmas >= MAX_VMAS) {
 787                        fprintf(stderr, "too many VMAs\n");
 788                        break;
 789                }
 790        }
 791        fclose(file);
 792}
 793
 794static void parse_file(const char *name)
 795{
 796}
 797
 798static void parse_addr_range(const char *optarg)
 799{
 800        unsigned long offset;
 801        unsigned long size;
 802        char *p;
 803
 804        p = strchr(optarg, ',');
 805        if (!p)
 806                p = strchr(optarg, '+');
 807
 808        if (p == optarg) {
 809                offset = 0;
 810                size   = parse_number(p + 1);
 811        } else if (p) {
 812                offset = parse_number(optarg);
 813                if (p[1] == '\0')
 814                        size = ULONG_MAX;
 815                else {
 816                        size = parse_number(p + 1);
 817                        if (*p == ',') {
 818                                if (size < offset)
 819                                        fatal("invalid range: %lu,%lu\n",
 820                                                        offset, size);
 821                                size -= offset;
 822                        }
 823                }
 824        } else {
 825                offset = parse_number(optarg);
 826                size   = 1;
 827        }
 828
 829        add_addr_range(offset, size);
 830}
 831
 832static void add_bits_filter(uint64_t mask, uint64_t bits)
 833{
 834        if (nr_bit_filters >= MAX_BIT_FILTERS)
 835                fatal("too much bit filters\n");
 836
 837        opt_mask[nr_bit_filters] = mask;
 838        opt_bits[nr_bit_filters] = bits;
 839        nr_bit_filters++;
 840}
 841
 842static uint64_t parse_flag_name(const char *str, int len)
 843{
 844        size_t i;
 845
 846        if (!*str || !len)
 847                return 0;
 848
 849        if (len <= 8 && !strncmp(str, "compound", len))
 850                return BITS_COMPOUND;
 851
 852        for (i = 0; i < ARRAY_SIZE(page_flag_names); i++) {
 853                if (!page_flag_names[i])
 854                        continue;
 855                if (!strncmp(str, page_flag_names[i] + 2, len))
 856                        return 1ULL << i;
 857        }
 858
 859        return parse_number(str);
 860}
 861
 862static uint64_t parse_flag_names(const char *str, int all)
 863{
 864        const char *p    = str;
 865        uint64_t   flags = 0;
 866
 867        while (1) {
 868                if (*p == ',' || *p == '=' || *p == '\0') {
 869                        if ((*str != '~') || (*str == '~' && all && *++str))
 870                                flags |= parse_flag_name(str, p - str);
 871                        if (*p != ',')
 872                                break;
 873                        str = p + 1;
 874                }
 875                p++;
 876        }
 877
 878        return flags;
 879}
 880
 881static void parse_bits_mask(const char *optarg)
 882{
 883        uint64_t mask;
 884        uint64_t bits;
 885        const char *p;
 886
 887        p = strchr(optarg, '=');
 888        if (p == optarg) {
 889                mask = KPF_ALL_BITS;
 890                bits = parse_flag_names(p + 1, 0);
 891        } else if (p) {
 892                mask = parse_flag_names(optarg, 0);
 893                bits = parse_flag_names(p + 1, 0);
 894        } else if (strchr(optarg, '~')) {
 895                mask = parse_flag_names(optarg, 1);
 896                bits = parse_flag_names(optarg, 0);
 897        } else {
 898                mask = parse_flag_names(optarg, 0);
 899                bits = KPF_ALL_BITS;
 900        }
 901
 902        add_bits_filter(mask, bits);
 903}
 904
 905static void describe_flags(const char *optarg)
 906{
 907        uint64_t flags = parse_flag_names(optarg, 0);
 908
 909        printf("0x%016llx\t%s\t%s\n",
 910                (unsigned long long)flags,
 911                page_flag_name(flags),
 912                page_flag_longname(flags));
 913}
 914
 915static const struct option opts[] = {
 916        { "raw"       , 0, NULL, 'r' },
 917        { "pid"       , 1, NULL, 'p' },
 918        { "file"      , 1, NULL, 'f' },
 919        { "addr"      , 1, NULL, 'a' },
 920        { "bits"      , 1, NULL, 'b' },
 921        { "describe"  , 1, NULL, 'd' },
 922        { "list"      , 0, NULL, 'l' },
 923        { "list-each" , 0, NULL, 'L' },
 924        { "no-summary", 0, NULL, 'N' },
 925        { "hwpoison"  , 0, NULL, 'X' },
 926        { "unpoison"  , 0, NULL, 'x' },
 927        { "help"      , 0, NULL, 'h' },
 928        { NULL        , 0, NULL, 0 }
 929};
 930
 931int main(int argc, char *argv[])
 932{
 933        int c;
 934
 935        page_size = getpagesize();
 936
 937        while ((c = getopt_long(argc, argv,
 938                                "rp:f:a:b:d:lLNXxh", opts, NULL)) != -1) {
 939                switch (c) {
 940                case 'r':
 941                        opt_raw = 1;
 942                        break;
 943                case 'p':
 944                        parse_pid(optarg);
 945                        break;
 946                case 'f':
 947                        parse_file(optarg);
 948                        break;
 949                case 'a':
 950                        parse_addr_range(optarg);
 951                        break;
 952                case 'b':
 953                        parse_bits_mask(optarg);
 954                        break;
 955                case 'd':
 956                        describe_flags(optarg);
 957                        exit(0);
 958                case 'l':
 959                        opt_list = 1;
 960                        break;
 961                case 'L':
 962                        opt_list = 2;
 963                        break;
 964                case 'N':
 965                        opt_no_summary = 1;
 966                        break;
 967                case 'X':
 968                        opt_hwpoison = 1;
 969                        prepare_hwpoison_fd();
 970                        break;
 971                case 'x':
 972                        opt_unpoison = 1;
 973                        prepare_hwpoison_fd();
 974                        break;
 975                case 'h':
 976                        usage();
 977                        exit(0);
 978                default:
 979                        usage();
 980                        exit(1);
 981                }
 982        }
 983
 984        if (opt_list && opt_pid)
 985                printf("voffset\t");
 986        if (opt_list == 1)
 987                printf("offset\tlen\tflags\n");
 988        if (opt_list == 2)
 989                printf("offset\tflags\n");
 990
 991        walk_addr_ranges();
 992
 993        if (opt_list == 1)
 994                show_page_range(0, 0, 0);  /* drain the buffer */
 995
 996        if (opt_no_summary)
 997                return 0;
 998
 999        if (opt_list)
1000                printf("\n\n");
1001
1002        show_summary();
1003
1004        return 0;
1005}
1006