Linux Audio

Check our new training course

Loading...
v4.10.11
   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 _FILE_OFFSET_BITS 64
  23#define _GNU_SOURCE
  24#include <stdio.h>
  25#include <stdlib.h>
  26#include <unistd.h>
  27#include <stdint.h>
  28#include <stdarg.h>
  29#include <string.h>
  30#include <getopt.h>
  31#include <limits.h>
  32#include <assert.h>
  33#include <ftw.h>
  34#include <time.h>
  35#include <setjmp.h>
  36#include <signal.h>
  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>
  42#include <sys/mman.h>
  43#include "../../include/uapi/linux/magic.h"
  44#include "../../include/uapi/linux/kernel-page-flags.h"
  45#include <api/fs/fs.h>
  46
  47#ifndef MAX_PATH
  48# define MAX_PATH 256
  49#endif
  50
  51#ifndef STR
  52# define _STR(x) #x
  53# define STR(x) _STR(x)
  54#endif
  55
  56/*
  57 * pagemap kernel ABI bits
  58 */
  59
  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)
 
 
 
 
 
 
  71
  72/*
  73 * kernel page flags
  74 */
  75
  76#define KPF_BYTES		8
  77#define PROC_KPAGEFLAGS		"/proc/kpageflags"
  78#define PROC_KPAGECGROUP	"/proc/kpagecgroup"
  79
  80/* [32-] kernel hacking assistances */
  81#define KPF_RESERVED		32
  82#define KPF_MLOCKED		33
  83#define KPF_MAPPEDTODISK	34
  84#define KPF_PRIVATE		35
  85#define KPF_PRIVATE_2		36
  86#define KPF_OWNER_PRIVATE	37
  87#define KPF_ARCH		38
  88#define KPF_UNCACHED		39
  89#define KPF_SOFTDIRTY		40
  90
  91/* [48-] take some arbitrary free slots for expanding overloaded flags
  92 * not part of kernel API
  93 */
  94#define KPF_READAHEAD		48
  95#define KPF_SLOB_FREE		49
  96#define KPF_SLUB_FROZEN		50
  97#define KPF_SLUB_DEBUG		51
  98#define KPF_FILE		61
  99#define KPF_SWAP		62
 100#define KPF_MMAP_EXCLUSIVE	63
 101
 102#define KPF_ALL_BITS		((uint64_t)~0ULL)
 103#define KPF_HACKERS_BITS	(0xffffULL << 32)
 104#define KPF_OVERLOADED_BITS	(0xffffULL << 48)
 105#define BIT(name)		(1ULL << KPF_##name)
 106#define BITS_COMPOUND		(BIT(COMPOUND_HEAD) | BIT(COMPOUND_TAIL))
 107
 108static const char * const page_flag_names[] = {
 109	[KPF_LOCKED]		= "L:locked",
 110	[KPF_ERROR]		= "E:error",
 111	[KPF_REFERENCED]	= "R:referenced",
 112	[KPF_UPTODATE]		= "U:uptodate",
 113	[KPF_DIRTY]		= "D:dirty",
 114	[KPF_LRU]		= "l:lru",
 115	[KPF_ACTIVE]		= "A:active",
 116	[KPF_SLAB]		= "S:slab",
 117	[KPF_WRITEBACK]		= "W:writeback",
 118	[KPF_RECLAIM]		= "I:reclaim",
 119	[KPF_BUDDY]		= "B:buddy",
 120
 121	[KPF_MMAP]		= "M:mmap",
 122	[KPF_ANON]		= "a:anonymous",
 123	[KPF_SWAPCACHE]		= "s:swapcache",
 124	[KPF_SWAPBACKED]	= "b:swapbacked",
 125	[KPF_COMPOUND_HEAD]	= "H:compound_head",
 126	[KPF_COMPOUND_TAIL]	= "T:compound_tail",
 127	[KPF_HUGE]		= "G:huge",
 128	[KPF_UNEVICTABLE]	= "u:unevictable",
 129	[KPF_HWPOISON]		= "X:hwpoison",
 130	[KPF_NOPAGE]		= "n:nopage",
 131	[KPF_KSM]		= "x:ksm",
 132	[KPF_THP]		= "t:thp",
 133	[KPF_BALLOON]		= "o:balloon",
 134	[KPF_ZERO_PAGE]		= "z:zero_page",
 135	[KPF_IDLE]              = "i:idle_page",
 136
 137	[KPF_RESERVED]		= "r:reserved",
 138	[KPF_MLOCKED]		= "m:mlocked",
 139	[KPF_MAPPEDTODISK]	= "d:mappedtodisk",
 140	[KPF_PRIVATE]		= "P:private",
 141	[KPF_PRIVATE_2]		= "p:private_2",
 142	[KPF_OWNER_PRIVATE]	= "O:owner_private",
 143	[KPF_ARCH]		= "h:arch",
 144	[KPF_UNCACHED]		= "c:uncached",
 145	[KPF_SOFTDIRTY]		= "f:softdirty",
 146
 147	[KPF_READAHEAD]		= "I:readahead",
 148	[KPF_SLOB_FREE]		= "P:slob_free",
 149	[KPF_SLUB_FROZEN]	= "A:slub_frozen",
 150	[KPF_SLUB_DEBUG]	= "E:slub_debug",
 151
 152	[KPF_FILE]		= "F:file",
 153	[KPF_SWAP]		= "w:swap",
 154	[KPF_MMAP_EXCLUSIVE]	= "1:mmap_exclusive",
 155};
 156
 157
 158static const char * const debugfs_known_mountpoints[] = {
 159	"/sys/kernel/debug",
 160	"/debug",
 161	0,
 162};
 163
 164/*
 165 * data structures
 166 */
 167
 168static int		opt_raw;	/* for kernel developers */
 169static int		opt_list;	/* list pages (in ranges) */
 170static int		opt_no_summary;	/* don't show summary */
 171static pid_t		opt_pid;	/* process to walk */
 172const char *		opt_file;	/* file or directory path */
 173static uint64_t		opt_cgroup;	/* cgroup inode */
 174static int		opt_list_cgroup;/* list page cgroup */
 175
 176#define MAX_ADDR_RANGES	1024
 177static int		nr_addr_ranges;
 178static unsigned long	opt_offset[MAX_ADDR_RANGES];
 179static unsigned long	opt_size[MAX_ADDR_RANGES];
 180
 181#define MAX_VMAS	10240
 182static int		nr_vmas;
 183static unsigned long	pg_start[MAX_VMAS];
 184static unsigned long	pg_end[MAX_VMAS];
 185
 186#define MAX_BIT_FILTERS	64
 187static int		nr_bit_filters;
 188static uint64_t		opt_mask[MAX_BIT_FILTERS];
 189static uint64_t		opt_bits[MAX_BIT_FILTERS];
 190
 191static int		page_size;
 192
 193static int		pagemap_fd;
 194static int		kpageflags_fd;
 195static int		kpagecgroup_fd = -1;
 196
 197static int		opt_hwpoison;
 198static int		opt_unpoison;
 199
 200static const char	*hwpoison_debug_fs;
 201static int		hwpoison_inject_fd;
 202static int		hwpoison_forget_fd;
 203
 204#define HASH_SHIFT	13
 205#define HASH_SIZE	(1 << HASH_SHIFT)
 206#define HASH_MASK	(HASH_SIZE - 1)
 207#define HASH_KEY(flags)	(flags & HASH_MASK)
 208
 209static unsigned long	total_pages;
 210static unsigned long	nr_pages[HASH_SIZE];
 211static uint64_t		page_flags[HASH_SIZE];
 212
 213
 214/*
 215 * helper functions
 216 */
 217
 218#define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0]))
 219
 220#define min_t(type, x, y) ({			\
 221	type __min1 = (x);			\
 222	type __min2 = (y);			\
 223	__min1 < __min2 ? __min1 : __min2; })
 224
 225#define max_t(type, x, y) ({			\
 226	type __max1 = (x);			\
 227	type __max2 = (y);			\
 228	__max1 > __max2 ? __max1 : __max2; })
 229
 230static unsigned long pages2mb(unsigned long pages)
 231{
 232	return (pages * page_size) >> 20;
 233}
 234
 235static void fatal(const char *x, ...)
 236{
 237	va_list ap;
 238
 239	va_start(ap, x);
 240	vfprintf(stderr, x, ap);
 241	va_end(ap);
 242	exit(EXIT_FAILURE);
 243}
 244
 245static int checked_open(const char *pathname, int flags)
 246{
 247	int fd = open(pathname, flags);
 248
 249	if (fd < 0) {
 250		perror(pathname);
 251		exit(EXIT_FAILURE);
 252	}
 253
 254	return fd;
 255}
 256
 257/*
 258 * pagemap/kpageflags routines
 259 */
 260
 261static unsigned long do_u64_read(int fd, char *name,
 262				 uint64_t *buf,
 263				 unsigned long index,
 264				 unsigned long count)
 265{
 266	long bytes;
 267
 268	if (index > ULONG_MAX / 8)
 269		fatal("index overflow: %lu\n", index);
 270
 271	bytes = pread(fd, buf, count * 8, (off_t)index * 8);
 272	if (bytes < 0) {
 273		perror(name);
 274		exit(EXIT_FAILURE);
 275	}
 276	if (bytes % 8)
 277		fatal("partial read: %lu bytes\n", bytes);
 278
 279	return bytes / 8;
 280}
 281
 282static unsigned long kpageflags_read(uint64_t *buf,
 283				     unsigned long index,
 284				     unsigned long pages)
 285{
 286	return do_u64_read(kpageflags_fd, PROC_KPAGEFLAGS, buf, index, pages);
 287}
 288
 289static unsigned long kpagecgroup_read(uint64_t *buf,
 290				      unsigned long index,
 291				      unsigned long pages)
 292{
 293	if (kpagecgroup_fd < 0)
 294		return pages;
 295
 296	return do_u64_read(kpagecgroup_fd, PROC_KPAGEFLAGS, buf, index, pages);
 297}
 298
 299static unsigned long pagemap_read(uint64_t *buf,
 300				  unsigned long index,
 301				  unsigned long pages)
 302{
 303	return do_u64_read(pagemap_fd, "/proc/pid/pagemap", buf, index, pages);
 304}
 305
 306static unsigned long pagemap_pfn(uint64_t val)
 307{
 308	unsigned long pfn;
 309
 310	if (val & PM_PRESENT)
 311		pfn = PM_PFRAME(val);
 312	else
 313		pfn = 0;
 314
 315	return pfn;
 316}
 317
 318static unsigned long pagemap_swap_offset(uint64_t val)
 319{
 320	return val & PM_SWAP ? PM_SWAP_OFFSET(val) : 0;
 321}
 322
 323/*
 324 * page flag names
 325 */
 326
 327static char *page_flag_name(uint64_t flags)
 328{
 329	static char buf[65];
 330	int present;
 331	size_t i, j;
 332
 333	for (i = 0, j = 0; i < ARRAY_SIZE(page_flag_names); i++) {
 334		present = (flags >> i) & 1;
 335		if (!page_flag_names[i]) {
 336			if (present)
 337				fatal("unknown flag bit %d\n", i);
 338			continue;
 339		}
 340		buf[j++] = present ? page_flag_names[i][0] : '_';
 341	}
 342
 343	return buf;
 344}
 345
 346static char *page_flag_longname(uint64_t flags)
 347{
 348	static char buf[1024];
 349	size_t i, n;
 350
 351	for (i = 0, n = 0; i < ARRAY_SIZE(page_flag_names); i++) {
 352		if (!page_flag_names[i])
 353			continue;
 354		if ((flags >> i) & 1)
 355			n += snprintf(buf + n, sizeof(buf) - n, "%s,",
 356					page_flag_names[i] + 2);
 357	}
 358	if (n)
 359		n--;
 360	buf[n] = '\0';
 361
 362	return buf;
 363}
 364
 365
 366/*
 367 * page list and summary
 368 */
 369
 370static void show_page_range(unsigned long voffset, unsigned long offset,
 371			    unsigned long size, uint64_t flags, uint64_t cgroup)
 372{
 373	static uint64_t      flags0;
 374	static uint64_t	     cgroup0;
 375	static unsigned long voff;
 376	static unsigned long index;
 377	static unsigned long count;
 378
 379	if (flags == flags0 && cgroup == cgroup0 && offset == index + count &&
 380	    size && voffset == voff + count) {
 381		count += size;
 382		return;
 383	}
 384
 385	if (count) {
 386		if (opt_pid)
 387			printf("%lx\t", voff);
 388		if (opt_file)
 389			printf("%lu\t", voff);
 390		if (opt_list_cgroup)
 391			printf("@%llu\t", (unsigned long long)cgroup0);
 392		printf("%lx\t%lx\t%s\n",
 393				index, count, page_flag_name(flags0));
 394	}
 395
 396	flags0 = flags;
 397	cgroup0= cgroup;
 398	index  = offset;
 399	voff   = voffset;
 400	count  = size;
 401}
 402
 403static void flush_page_range(void)
 404{
 405	show_page_range(0, 0, 0, 0, 0);
 406}
 407
 408static void show_page(unsigned long voffset, unsigned long offset,
 409		      uint64_t flags, uint64_t cgroup)
 410{
 411	if (opt_pid)
 412		printf("%lx\t", voffset);
 413	if (opt_file)
 414		printf("%lu\t", voffset);
 415	if (opt_list_cgroup)
 416		printf("@%llu\t", (unsigned long long)cgroup);
 417	printf("%lx\t%s\n", offset, page_flag_name(flags));
 418}
 419
 420static void show_summary(void)
 421{
 422	size_t i;
 423
 424	printf("             flags\tpage-count       MB"
 425		"  symbolic-flags\t\t\tlong-symbolic-flags\n");
 426
 427	for (i = 0; i < ARRAY_SIZE(nr_pages); i++) {
 428		if (nr_pages[i])
 429			printf("0x%016llx\t%10lu %8lu  %s\t%s\n",
 430				(unsigned long long)page_flags[i],
 431				nr_pages[i],
 432				pages2mb(nr_pages[i]),
 433				page_flag_name(page_flags[i]),
 434				page_flag_longname(page_flags[i]));
 435	}
 436
 437	printf("             total\t%10lu %8lu\n",
 438			total_pages, pages2mb(total_pages));
 439}
 440
 441
 442/*
 443 * page flag filters
 444 */
 445
 446static int bit_mask_ok(uint64_t flags)
 447{
 448	int i;
 449
 450	for (i = 0; i < nr_bit_filters; i++) {
 451		if (opt_bits[i] == KPF_ALL_BITS) {
 452			if ((flags & opt_mask[i]) == 0)
 453				return 0;
 454		} else {
 455			if ((flags & opt_mask[i]) != opt_bits[i])
 456				return 0;
 457		}
 458	}
 459
 460	return 1;
 461}
 462
 463static uint64_t expand_overloaded_flags(uint64_t flags, uint64_t pme)
 464{
 465	/* SLOB/SLUB overload several page flags */
 466	if (flags & BIT(SLAB)) {
 467		if (flags & BIT(PRIVATE))
 468			flags ^= BIT(PRIVATE) | BIT(SLOB_FREE);
 469		if (flags & BIT(ACTIVE))
 470			flags ^= BIT(ACTIVE) | BIT(SLUB_FROZEN);
 471		if (flags & BIT(ERROR))
 472			flags ^= BIT(ERROR) | BIT(SLUB_DEBUG);
 473	}
 474
 475	/* PG_reclaim is overloaded as PG_readahead in the read path */
 476	if ((flags & (BIT(RECLAIM) | BIT(WRITEBACK))) == BIT(RECLAIM))
 477		flags ^= BIT(RECLAIM) | BIT(READAHEAD);
 478
 479	if (pme & PM_SOFT_DIRTY)
 480		flags |= BIT(SOFTDIRTY);
 481	if (pme & PM_FILE)
 482		flags |= BIT(FILE);
 483	if (pme & PM_SWAP)
 484		flags |= BIT(SWAP);
 485	if (pme & PM_MMAP_EXCLUSIVE)
 486		flags |= BIT(MMAP_EXCLUSIVE);
 487
 488	return flags;
 489}
 490
 491static uint64_t well_known_flags(uint64_t flags)
 492{
 493	/* hide flags intended only for kernel hacker */
 494	flags &= ~KPF_HACKERS_BITS;
 495
 496	/* hide non-hugeTLB compound pages */
 497	if ((flags & BITS_COMPOUND) && !(flags & BIT(HUGE)))
 498		flags &= ~BITS_COMPOUND;
 499
 500	return flags;
 501}
 502
 503static uint64_t kpageflags_flags(uint64_t flags, uint64_t pme)
 504{
 505	if (opt_raw)
 506		flags = expand_overloaded_flags(flags, pme);
 507	else
 508		flags = well_known_flags(flags);
 509
 510	return flags;
 511}
 512
 513/*
 514 * page actions
 515 */
 516
 517static void prepare_hwpoison_fd(void)
 518{
 519	char buf[MAX_PATH + 1];
 520
 521	hwpoison_debug_fs = debugfs__mount();
 522	if (!hwpoison_debug_fs) {
 523		perror("mount debugfs");
 524		exit(EXIT_FAILURE);
 525	}
 526
 527	if (opt_hwpoison && !hwpoison_inject_fd) {
 528		snprintf(buf, MAX_PATH, "%s/hwpoison/corrupt-pfn",
 529			hwpoison_debug_fs);
 530		hwpoison_inject_fd = checked_open(buf, O_WRONLY);
 531	}
 532
 533	if (opt_unpoison && !hwpoison_forget_fd) {
 534		snprintf(buf, MAX_PATH, "%s/hwpoison/unpoison-pfn",
 535			hwpoison_debug_fs);
 536		hwpoison_forget_fd = checked_open(buf, O_WRONLY);
 537	}
 538}
 539
 540static int hwpoison_page(unsigned long offset)
 541{
 542	char buf[100];
 543	int len;
 544
 545	len = sprintf(buf, "0x%lx\n", offset);
 546	len = write(hwpoison_inject_fd, buf, len);
 547	if (len < 0) {
 548		perror("hwpoison inject");
 549		return len;
 550	}
 551	return 0;
 552}
 553
 554static int unpoison_page(unsigned long offset)
 555{
 556	char buf[100];
 557	int len;
 558
 559	len = sprintf(buf, "0x%lx\n", offset);
 560	len = write(hwpoison_forget_fd, buf, len);
 561	if (len < 0) {
 562		perror("hwpoison forget");
 563		return len;
 564	}
 565	return 0;
 566}
 567
 568/*
 569 * page frame walker
 570 */
 571
 572static size_t hash_slot(uint64_t flags)
 573{
 574	size_t k = HASH_KEY(flags);
 575	size_t i;
 576
 577	/* Explicitly reserve slot 0 for flags 0: the following logic
 578	 * cannot distinguish an unoccupied slot from slot (flags==0).
 579	 */
 580	if (flags == 0)
 581		return 0;
 582
 583	/* search through the remaining (HASH_SIZE-1) slots */
 584	for (i = 1; i < ARRAY_SIZE(page_flags); i++, k++) {
 585		if (!k || k >= ARRAY_SIZE(page_flags))
 586			k = 1;
 587		if (page_flags[k] == 0) {
 588			page_flags[k] = flags;
 589			return k;
 590		}
 591		if (page_flags[k] == flags)
 592			return k;
 593	}
 594
 595	fatal("hash table full: bump up HASH_SHIFT?\n");
 596	exit(EXIT_FAILURE);
 597}
 598
 599static void add_page(unsigned long voffset, unsigned long offset,
 600		     uint64_t flags, uint64_t cgroup, uint64_t pme)
 601{
 602	flags = kpageflags_flags(flags, pme);
 603
 604	if (!bit_mask_ok(flags))
 605		return;
 606
 607	if (opt_cgroup && cgroup != (uint64_t)opt_cgroup)
 608		return;
 609
 610	if (opt_hwpoison)
 611		hwpoison_page(offset);
 612	if (opt_unpoison)
 613		unpoison_page(offset);
 614
 615	if (opt_list == 1)
 616		show_page_range(voffset, offset, 1, flags, cgroup);
 617	else if (opt_list == 2)
 618		show_page(voffset, offset, flags, cgroup);
 619
 620	nr_pages[hash_slot(flags)]++;
 621	total_pages++;
 622}
 623
 624#define KPAGEFLAGS_BATCH	(64 << 10)	/* 64k pages */
 625static void walk_pfn(unsigned long voffset,
 626		     unsigned long index,
 627		     unsigned long count,
 628		     uint64_t pme)
 629{
 630	uint64_t buf[KPAGEFLAGS_BATCH];
 631	uint64_t cgi[KPAGEFLAGS_BATCH];
 632	unsigned long batch;
 633	unsigned long pages;
 634	unsigned long i;
 635
 636	/*
 637	 * kpagecgroup_read() reads only if kpagecgroup were opened, but
 638	 * /proc/kpagecgroup might even not exist, so it's better to fill
 639	 * them with zeros here.
 640	 */
 641	if (count == 1)
 642		cgi[0] = 0;
 643	else
 644		memset(cgi, 0, sizeof cgi);
 645
 646	while (count) {
 647		batch = min_t(unsigned long, count, KPAGEFLAGS_BATCH);
 648		pages = kpageflags_read(buf, index, batch);
 649		if (pages == 0)
 650			break;
 651
 652		if (kpagecgroup_read(cgi, index, pages) != pages)
 653			fatal("kpagecgroup returned fewer pages than expected");
 654
 655		for (i = 0; i < pages; i++)
 656			add_page(voffset + i, index + i, buf[i], cgi[i], pme);
 657
 658		index += pages;
 659		count -= pages;
 660	}
 661}
 662
 663static void walk_swap(unsigned long voffset, uint64_t pme)
 664{
 665	uint64_t flags = kpageflags_flags(0, pme);
 666
 667	if (!bit_mask_ok(flags))
 668		return;
 669
 670	if (opt_cgroup)
 671		return;
 672
 673	if (opt_list == 1)
 674		show_page_range(voffset, pagemap_swap_offset(pme), 1, flags, 0);
 675	else if (opt_list == 2)
 676		show_page(voffset, pagemap_swap_offset(pme), flags, 0);
 677
 678	nr_pages[hash_slot(flags)]++;
 679	total_pages++;
 680}
 681
 682#define PAGEMAP_BATCH	(64 << 10)
 683static void walk_vma(unsigned long index, unsigned long count)
 684{
 685	uint64_t buf[PAGEMAP_BATCH];
 686	unsigned long batch;
 687	unsigned long pages;
 688	unsigned long pfn;
 689	unsigned long i;
 690
 691	while (count) {
 692		batch = min_t(unsigned long, count, PAGEMAP_BATCH);
 693		pages = pagemap_read(buf, index, batch);
 694		if (pages == 0)
 695			break;
 696
 697		for (i = 0; i < pages; i++) {
 698			pfn = pagemap_pfn(buf[i]);
 699			if (pfn)
 700				walk_pfn(index + i, pfn, 1, buf[i]);
 701			if (buf[i] & PM_SWAP)
 702				walk_swap(index + i, buf[i]);
 703		}
 704
 705		index += pages;
 706		count -= pages;
 707	}
 708}
 709
 710static void walk_task(unsigned long index, unsigned long count)
 711{
 712	const unsigned long end = index + count;
 713	unsigned long start;
 714	int i = 0;
 715
 716	while (index < end) {
 717
 718		while (pg_end[i] <= index)
 719			if (++i >= nr_vmas)
 720				return;
 721		if (pg_start[i] >= end)
 722			return;
 723
 724		start = max_t(unsigned long, pg_start[i], index);
 725		index = min_t(unsigned long, pg_end[i], end);
 726
 727		assert(start < index);
 728		walk_vma(start, index - start);
 729	}
 730}
 731
 732static void add_addr_range(unsigned long offset, unsigned long size)
 733{
 734	if (nr_addr_ranges >= MAX_ADDR_RANGES)
 735		fatal("too many addr ranges\n");
 736
 737	opt_offset[nr_addr_ranges] = offset;
 738	opt_size[nr_addr_ranges] = min_t(unsigned long, size, ULONG_MAX-offset);
 739	nr_addr_ranges++;
 740}
 741
 742static void walk_addr_ranges(void)
 743{
 744	int i;
 745
 746	kpageflags_fd = checked_open(PROC_KPAGEFLAGS, O_RDONLY);
 747
 748	if (!nr_addr_ranges)
 749		add_addr_range(0, ULONG_MAX);
 750
 751	for (i = 0; i < nr_addr_ranges; i++)
 752		if (!opt_pid)
 753			walk_pfn(opt_offset[i], opt_offset[i], opt_size[i], 0);
 754		else
 755			walk_task(opt_offset[i], opt_size[i]);
 756
 757	close(kpageflags_fd);
 758}
 759
 760
 761/*
 762 * user interface
 763 */
 764
 765static const char *page_flag_type(uint64_t flag)
 766{
 767	if (flag & KPF_HACKERS_BITS)
 768		return "(r)";
 769	if (flag & KPF_OVERLOADED_BITS)
 770		return "(o)";
 771	return "   ";
 772}
 773
 774static void usage(void)
 775{
 776	size_t i, j;
 777
 778	printf(
 779"page-types [options]\n"
 780"            -r|--raw                   Raw mode, for kernel developers\n"
 781"            -d|--describe flags        Describe flags\n"
 782"            -a|--addr    addr-spec     Walk a range of pages\n"
 783"            -b|--bits    bits-spec     Walk pages with specified bits\n"
 784"            -c|--cgroup  path|@inode   Walk pages within memory cgroup\n"
 785"            -p|--pid     pid           Walk process address space\n"
 786"            -f|--file    filename      Walk file address space\n"
 787"            -l|--list                  Show page details in ranges\n"
 788"            -L|--list-each             Show page details one by one\n"
 789"            -C|--list-cgroup           Show cgroup inode for pages\n"
 790"            -N|--no-summary            Don't show summary info\n"
 791"            -X|--hwpoison              hwpoison pages\n"
 792"            -x|--unpoison              unpoison pages\n"
 793"            -h|--help                  Show this usage message\n"
 794"flags:\n"
 795"            0x10                       bitfield format, e.g.\n"
 796"            anon                       bit-name, e.g.\n"
 797"            0x10,anon                  comma-separated list, e.g.\n"
 798"addr-spec:\n"
 799"            N                          one page at offset N (unit: pages)\n"
 800"            N+M                        pages range from N to N+M-1\n"
 801"            N,M                        pages range from N to M-1\n"
 802"            N,                         pages range from N to end\n"
 803"            ,M                         pages range from 0 to M-1\n"
 804"bits-spec:\n"
 805"            bit1,bit2                  (flags & (bit1|bit2)) != 0\n"
 806"            bit1,bit2=bit1             (flags & (bit1|bit2)) == bit1\n"
 807"            bit1,~bit2                 (flags & (bit1|bit2)) == bit1\n"
 808"            =bit1,bit2                 flags == (bit1|bit2)\n"
 809"bit-names:\n"
 810	);
 811
 812	for (i = 0, j = 0; i < ARRAY_SIZE(page_flag_names); i++) {
 813		if (!page_flag_names[i])
 814			continue;
 815		printf("%16s%s", page_flag_names[i] + 2,
 816				 page_flag_type(1ULL << i));
 817		if (++j > 3) {
 818			j = 0;
 819			putchar('\n');
 820		}
 821	}
 822	printf("\n                                   "
 823		"(r) raw mode bits  (o) overloaded bits\n");
 824}
 825
 826static unsigned long long parse_number(const char *str)
 827{
 828	unsigned long long n;
 829
 830	n = strtoll(str, NULL, 0);
 831
 832	if (n == 0 && str[0] != '0')
 833		fatal("invalid name or number: %s\n", str);
 834
 835	return n;
 836}
 837
 838static void parse_pid(const char *str)
 839{
 840	FILE *file;
 841	char buf[5000];
 842
 843	opt_pid = parse_number(str);
 844
 845	sprintf(buf, "/proc/%d/pagemap", opt_pid);
 846	pagemap_fd = checked_open(buf, O_RDONLY);
 847
 848	sprintf(buf, "/proc/%d/maps", opt_pid);
 849	file = fopen(buf, "r");
 850	if (!file) {
 851		perror(buf);
 852		exit(EXIT_FAILURE);
 853	}
 854
 855	while (fgets(buf, sizeof(buf), file) != NULL) {
 856		unsigned long vm_start;
 857		unsigned long vm_end;
 858		unsigned long long pgoff;
 859		int major, minor;
 860		char r, w, x, s;
 861		unsigned long ino;
 862		int n;
 863
 864		n = sscanf(buf, "%lx-%lx %c%c%c%c %llx %x:%x %lu",
 865			   &vm_start,
 866			   &vm_end,
 867			   &r, &w, &x, &s,
 868			   &pgoff,
 869			   &major, &minor,
 870			   &ino);
 871		if (n < 10) {
 872			fprintf(stderr, "unexpected line: %s\n", buf);
 873			continue;
 874		}
 875		pg_start[nr_vmas] = vm_start / page_size;
 876		pg_end[nr_vmas] = vm_end / page_size;
 877		if (++nr_vmas >= MAX_VMAS) {
 878			fprintf(stderr, "too many VMAs\n");
 879			break;
 880		}
 881	}
 882	fclose(file);
 883}
 884
 885static void show_file(const char *name, const struct stat *st)
 886{
 887	unsigned long long size = st->st_size;
 888	char atime[64], mtime[64];
 889	long now = time(NULL);
 890
 891	printf("%s\tInode: %u\tSize: %llu (%llu pages)\n",
 892			name, (unsigned)st->st_ino,
 893			size, (size + page_size - 1) / page_size);
 894
 895	strftime(atime, sizeof(atime), "%c", localtime(&st->st_atime));
 896	strftime(mtime, sizeof(mtime), "%c", localtime(&st->st_mtime));
 897
 898	printf("Modify: %s (%ld seconds ago)\nAccess: %s (%ld seconds ago)\n",
 899			mtime, now - st->st_mtime,
 900			atime, now - st->st_atime);
 901}
 902
 903static sigjmp_buf sigbus_jmp;
 904
 905static void * volatile sigbus_addr;
 906
 907static void sigbus_handler(int sig, siginfo_t *info, void *ucontex)
 908{
 909	(void)sig;
 910	(void)ucontex;
 911	sigbus_addr = info ? info->si_addr : NULL;
 912	siglongjmp(sigbus_jmp, 1);
 913}
 914
 915static struct sigaction sigbus_action = {
 916	.sa_sigaction = sigbus_handler,
 917	.sa_flags = SA_SIGINFO,
 918};
 919
 920static void walk_file(const char *name, const struct stat *st)
 921{
 922	uint8_t vec[PAGEMAP_BATCH];
 923	uint64_t buf[PAGEMAP_BATCH], flags;
 924	uint64_t cgroup = 0;
 925	unsigned long nr_pages, pfn, i;
 926	off_t off, end = st->st_size;
 927	int fd;
 
 928	ssize_t len;
 929	void *ptr;
 930	int first = 1;
 931
 932	fd = checked_open(name, O_RDONLY|O_NOATIME|O_NOFOLLOW);
 933
 934	for (off = 0; off < end; off += len) {
 935		nr_pages = (end - off + page_size - 1) / page_size;
 936		if (nr_pages > PAGEMAP_BATCH)
 937			nr_pages = PAGEMAP_BATCH;
 938		len = nr_pages * page_size;
 939
 940		ptr = mmap(NULL, len, PROT_READ, MAP_SHARED, fd, off);
 941		if (ptr == MAP_FAILED)
 942			fatal("mmap failed: %s", name);
 943
 944		/* determine cached pages */
 945		if (mincore(ptr, len, vec))
 946			fatal("mincore failed: %s", name);
 947
 948		/* turn off readahead */
 949		if (madvise(ptr, len, MADV_RANDOM))
 950			fatal("madvice failed: %s", name);
 951
 952		if (sigsetjmp(sigbus_jmp, 1)) {
 953			end = off + sigbus_addr ? sigbus_addr - ptr : 0;
 954			fprintf(stderr, "got sigbus at offset %lld: %s\n",
 955					(long long)end, name);
 956			goto got_sigbus;
 957		}
 958
 959		/* populate ptes */
 960		for (i = 0; i < nr_pages ; i++) {
 961			if (vec[i] & 1)
 962				(void)*(volatile int *)(ptr + i * page_size);
 963		}
 964got_sigbus:
 965
 966		/* turn off harvesting reference bits */
 967		if (madvise(ptr, len, MADV_SEQUENTIAL))
 968			fatal("madvice failed: %s", name);
 969
 970		if (pagemap_read(buf, (unsigned long)ptr / page_size,
 971					nr_pages) != nr_pages)
 972			fatal("cannot read pagemap");
 973
 974		munmap(ptr, len);
 975
 976		for (i = 0; i < nr_pages; i++) {
 977			pfn = pagemap_pfn(buf[i]);
 978			if (!pfn)
 979				continue;
 980			if (!kpageflags_read(&flags, pfn, 1))
 981				continue;
 982			if (!kpagecgroup_read(&cgroup, pfn, 1))
 983				fatal("kpagecgroup_read failed");
 984			if (first && opt_list) {
 985				first = 0;
 986				flush_page_range();
 987				show_file(name, st);
 988			}
 989			add_page(off / page_size + i, pfn,
 990				 flags, cgroup, buf[i]);
 991		}
 992	}
 993
 994	close(fd);
 995}
 996
 997int walk_tree(const char *name, const struct stat *st, int type, struct FTW *f)
 998{
 999	(void)f;
1000	switch (type) {
1001	case FTW_F:
1002		if (S_ISREG(st->st_mode))
1003			walk_file(name, st);
1004		break;
1005	case FTW_DNR:
1006		fprintf(stderr, "cannot read dir: %s\n", name);
1007		break;
1008	}
1009	return 0;
1010}
1011
1012static void walk_page_cache(void)
1013{
1014	struct stat st;
1015
1016	kpageflags_fd = checked_open(PROC_KPAGEFLAGS, O_RDONLY);
1017	pagemap_fd = checked_open("/proc/self/pagemap", O_RDONLY);
1018	sigaction(SIGBUS, &sigbus_action, NULL);
1019
1020	if (stat(opt_file, &st))
1021		fatal("stat failed: %s\n", opt_file);
1022
1023	if (S_ISREG(st.st_mode)) {
1024		walk_file(opt_file, &st);
1025	} else if (S_ISDIR(st.st_mode)) {
1026		/* do not follow symlinks and mountpoints */
1027		if (nftw(opt_file, walk_tree, 64, FTW_MOUNT | FTW_PHYS) < 0)
1028			fatal("nftw failed: %s\n", opt_file);
1029	} else
1030		fatal("unhandled file type: %s\n", opt_file);
1031
1032	close(kpageflags_fd);
1033	close(pagemap_fd);
1034	signal(SIGBUS, SIG_DFL);
1035}
1036
1037static void parse_file(const char *name)
1038{
1039	opt_file = name;
1040}
1041
1042static void parse_cgroup(const char *path)
1043{
1044	if (path[0] == '@') {
1045		opt_cgroup = parse_number(path + 1);
1046		return;
1047	}
1048
1049	struct stat st;
1050
1051	if (stat(path, &st))
1052		fatal("stat failed: %s: %m\n", path);
1053
1054	if (!S_ISDIR(st.st_mode))
1055		fatal("cgroup supposed to be a directory: %s\n", path);
1056
1057	opt_cgroup = st.st_ino;
1058}
1059
1060static void parse_addr_range(const char *optarg)
1061{
1062	unsigned long offset;
1063	unsigned long size;
1064	char *p;
1065
1066	p = strchr(optarg, ',');
1067	if (!p)
1068		p = strchr(optarg, '+');
1069
1070	if (p == optarg) {
1071		offset = 0;
1072		size   = parse_number(p + 1);
1073	} else if (p) {
1074		offset = parse_number(optarg);
1075		if (p[1] == '\0')
1076			size = ULONG_MAX;
1077		else {
1078			size = parse_number(p + 1);
1079			if (*p == ',') {
1080				if (size < offset)
1081					fatal("invalid range: %lu,%lu\n",
1082							offset, size);
1083				size -= offset;
1084			}
1085		}
1086	} else {
1087		offset = parse_number(optarg);
1088		size   = 1;
1089	}
1090
1091	add_addr_range(offset, size);
1092}
1093
1094static void add_bits_filter(uint64_t mask, uint64_t bits)
1095{
1096	if (nr_bit_filters >= MAX_BIT_FILTERS)
1097		fatal("too much bit filters\n");
1098
1099	opt_mask[nr_bit_filters] = mask;
1100	opt_bits[nr_bit_filters] = bits;
1101	nr_bit_filters++;
1102}
1103
1104static uint64_t parse_flag_name(const char *str, int len)
1105{
1106	size_t i;
1107
1108	if (!*str || !len)
1109		return 0;
1110
1111	if (len <= 8 && !strncmp(str, "compound", len))
1112		return BITS_COMPOUND;
1113
1114	for (i = 0; i < ARRAY_SIZE(page_flag_names); i++) {
1115		if (!page_flag_names[i])
1116			continue;
1117		if (!strncmp(str, page_flag_names[i] + 2, len))
1118			return 1ULL << i;
1119	}
1120
1121	return parse_number(str);
1122}
1123
1124static uint64_t parse_flag_names(const char *str, int all)
1125{
1126	const char *p    = str;
1127	uint64_t   flags = 0;
1128
1129	while (1) {
1130		if (*p == ',' || *p == '=' || *p == '\0') {
1131			if ((*str != '~') || (*str == '~' && all && *++str))
1132				flags |= parse_flag_name(str, p - str);
1133			if (*p != ',')
1134				break;
1135			str = p + 1;
1136		}
1137		p++;
1138	}
1139
1140	return flags;
1141}
1142
1143static void parse_bits_mask(const char *optarg)
1144{
1145	uint64_t mask;
1146	uint64_t bits;
1147	const char *p;
1148
1149	p = strchr(optarg, '=');
1150	if (p == optarg) {
1151		mask = KPF_ALL_BITS;
1152		bits = parse_flag_names(p + 1, 0);
1153	} else if (p) {
1154		mask = parse_flag_names(optarg, 0);
1155		bits = parse_flag_names(p + 1, 0);
1156	} else if (strchr(optarg, '~')) {
1157		mask = parse_flag_names(optarg, 1);
1158		bits = parse_flag_names(optarg, 0);
1159	} else {
1160		mask = parse_flag_names(optarg, 0);
1161		bits = KPF_ALL_BITS;
1162	}
1163
1164	add_bits_filter(mask, bits);
1165}
1166
1167static void describe_flags(const char *optarg)
1168{
1169	uint64_t flags = parse_flag_names(optarg, 0);
1170
1171	printf("0x%016llx\t%s\t%s\n",
1172		(unsigned long long)flags,
1173		page_flag_name(flags),
1174		page_flag_longname(flags));
1175}
1176
1177static const struct option opts[] = {
1178	{ "raw"       , 0, NULL, 'r' },
1179	{ "pid"       , 1, NULL, 'p' },
1180	{ "file"      , 1, NULL, 'f' },
1181	{ "addr"      , 1, NULL, 'a' },
1182	{ "bits"      , 1, NULL, 'b' },
1183	{ "cgroup"    , 1, NULL, 'c' },
1184	{ "describe"  , 1, NULL, 'd' },
1185	{ "list"      , 0, NULL, 'l' },
1186	{ "list-each" , 0, NULL, 'L' },
1187	{ "list-cgroup", 0, NULL, 'C' },
1188	{ "no-summary", 0, NULL, 'N' },
1189	{ "hwpoison"  , 0, NULL, 'X' },
1190	{ "unpoison"  , 0, NULL, 'x' },
1191	{ "help"      , 0, NULL, 'h' },
1192	{ NULL        , 0, NULL, 0 }
1193};
1194
1195int main(int argc, char *argv[])
1196{
1197	int c;
1198
1199	page_size = getpagesize();
1200
1201	while ((c = getopt_long(argc, argv,
1202				"rp:f:a:b:d:c:ClLNXxh", opts, NULL)) != -1) {
1203		switch (c) {
1204		case 'r':
1205			opt_raw = 1;
1206			break;
1207		case 'p':
1208			parse_pid(optarg);
1209			break;
1210		case 'f':
1211			parse_file(optarg);
1212			break;
1213		case 'a':
1214			parse_addr_range(optarg);
1215			break;
1216		case 'b':
1217			parse_bits_mask(optarg);
1218			break;
1219		case 'c':
1220			parse_cgroup(optarg);
1221			break;
1222		case 'C':
1223			opt_list_cgroup = 1;
1224			break;
1225		case 'd':
1226			describe_flags(optarg);
1227			exit(0);
1228		case 'l':
1229			opt_list = 1;
1230			break;
1231		case 'L':
1232			opt_list = 2;
1233			break;
1234		case 'N':
1235			opt_no_summary = 1;
1236			break;
1237		case 'X':
1238			opt_hwpoison = 1;
1239			prepare_hwpoison_fd();
1240			break;
1241		case 'x':
1242			opt_unpoison = 1;
1243			prepare_hwpoison_fd();
1244			break;
1245		case 'h':
1246			usage();
1247			exit(0);
1248		default:
1249			usage();
1250			exit(1);
1251		}
1252	}
1253
1254	if (opt_cgroup || opt_list_cgroup)
1255		kpagecgroup_fd = checked_open(PROC_KPAGECGROUP, O_RDONLY);
1256
1257	if (opt_list && opt_pid)
1258		printf("voffset\t");
1259	if (opt_list && opt_file)
1260		printf("foffset\t");
1261	if (opt_list && opt_list_cgroup)
1262		printf("cgroup\t");
1263	if (opt_list == 1)
1264		printf("offset\tlen\tflags\n");
1265	if (opt_list == 2)
1266		printf("offset\tflags\n");
1267
1268	if (opt_file)
1269		walk_page_cache();
1270	else
1271		walk_addr_ranges();
1272
1273	if (opt_list == 1)
1274		flush_page_range();
1275
1276	if (opt_no_summary)
1277		return 0;
1278
1279	if (opt_list)
1280		printf("\n\n");
1281
1282	show_summary();
1283
1284	return 0;
1285}
v3.15
   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 _FILE_OFFSET_BITS 64
  23#define _GNU_SOURCE
  24#include <stdio.h>
  25#include <stdlib.h>
  26#include <unistd.h>
  27#include <stdint.h>
  28#include <stdarg.h>
  29#include <string.h>
  30#include <getopt.h>
  31#include <limits.h>
  32#include <assert.h>
  33#include <ftw.h>
  34#include <time.h>
 
 
  35#include <sys/types.h>
  36#include <sys/errno.h>
  37#include <sys/fcntl.h>
  38#include <sys/mount.h>
  39#include <sys/statfs.h>
  40#include <sys/mman.h>
  41#include "../../include/uapi/linux/magic.h"
  42#include "../../include/uapi/linux/kernel-page-flags.h"
  43#include <api/fs/debugfs.h>
  44
  45#ifndef MAX_PATH
  46# define MAX_PATH 256
  47#endif
  48
  49#ifndef STR
  50# define _STR(x) #x
  51# define STR(x) _STR(x)
  52#endif
  53
  54/*
  55 * pagemap kernel ABI bits
  56 */
  57
  58#define PM_ENTRY_BYTES      sizeof(uint64_t)
  59#define PM_STATUS_BITS      3
  60#define PM_STATUS_OFFSET    (64 - PM_STATUS_BITS)
  61#define PM_STATUS_MASK      (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
  62#define PM_STATUS(nr)       (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
  63#define PM_PSHIFT_BITS      6
  64#define PM_PSHIFT_OFFSET    (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
  65#define PM_PSHIFT_MASK      (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
  66#define __PM_PSHIFT(x)      (((uint64_t) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
  67#define PM_PFRAME_MASK      ((1LL << PM_PSHIFT_OFFSET) - 1)
  68#define PM_PFRAME(x)        ((x) & PM_PFRAME_MASK)
  69
  70#define __PM_SOFT_DIRTY      (1LL)
  71#define PM_PRESENT          PM_STATUS(4LL)
  72#define PM_SWAP             PM_STATUS(2LL)
  73#define PM_SOFT_DIRTY       __PM_PSHIFT(__PM_SOFT_DIRTY)
  74
  75
  76/*
  77 * kernel page flags
  78 */
  79
  80#define KPF_BYTES		8
  81#define PROC_KPAGEFLAGS		"/proc/kpageflags"
 
  82
  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
  90#define KPF_ARCH		38
  91#define KPF_UNCACHED		39
  92#define KPF_SOFTDIRTY		40
  93
  94/* [48-] take some arbitrary free slots for expanding overloaded flags
  95 * not part of kernel API
  96 */
  97#define KPF_READAHEAD		48
  98#define KPF_SLOB_FREE		49
  99#define KPF_SLUB_FROZEN		50
 100#define KPF_SLUB_DEBUG		51
 
 
 
 101
 102#define KPF_ALL_BITS		((uint64_t)~0ULL)
 103#define KPF_HACKERS_BITS	(0xffffULL << 32)
 104#define KPF_OVERLOADED_BITS	(0xffffULL << 48)
 105#define BIT(name)		(1ULL << KPF_##name)
 106#define BITS_COMPOUND		(BIT(COMPOUND_HEAD) | BIT(COMPOUND_TAIL))
 107
 108static const char * const page_flag_names[] = {
 109	[KPF_LOCKED]		= "L:locked",
 110	[KPF_ERROR]		= "E:error",
 111	[KPF_REFERENCED]	= "R:referenced",
 112	[KPF_UPTODATE]		= "U:uptodate",
 113	[KPF_DIRTY]		= "D:dirty",
 114	[KPF_LRU]		= "l:lru",
 115	[KPF_ACTIVE]		= "A:active",
 116	[KPF_SLAB]		= "S:slab",
 117	[KPF_WRITEBACK]		= "W:writeback",
 118	[KPF_RECLAIM]		= "I:reclaim",
 119	[KPF_BUDDY]		= "B:buddy",
 120
 121	[KPF_MMAP]		= "M:mmap",
 122	[KPF_ANON]		= "a:anonymous",
 123	[KPF_SWAPCACHE]		= "s:swapcache",
 124	[KPF_SWAPBACKED]	= "b:swapbacked",
 125	[KPF_COMPOUND_HEAD]	= "H:compound_head",
 126	[KPF_COMPOUND_TAIL]	= "T:compound_tail",
 127	[KPF_HUGE]		= "G:huge",
 128	[KPF_UNEVICTABLE]	= "u:unevictable",
 129	[KPF_HWPOISON]		= "X:hwpoison",
 130	[KPF_NOPAGE]		= "n:nopage",
 131	[KPF_KSM]		= "x:ksm",
 132	[KPF_THP]		= "t:thp",
 
 
 
 133
 134	[KPF_RESERVED]		= "r:reserved",
 135	[KPF_MLOCKED]		= "m:mlocked",
 136	[KPF_MAPPEDTODISK]	= "d:mappedtodisk",
 137	[KPF_PRIVATE]		= "P:private",
 138	[KPF_PRIVATE_2]		= "p:private_2",
 139	[KPF_OWNER_PRIVATE]	= "O:owner_private",
 140	[KPF_ARCH]		= "h:arch",
 141	[KPF_UNCACHED]		= "c:uncached",
 142	[KPF_SOFTDIRTY]		= "f:softdirty",
 143
 144	[KPF_READAHEAD]		= "I:readahead",
 145	[KPF_SLOB_FREE]		= "P:slob_free",
 146	[KPF_SLUB_FROZEN]	= "A:slub_frozen",
 147	[KPF_SLUB_DEBUG]	= "E:slub_debug",
 
 
 
 
 148};
 149
 150
 151static const char * const debugfs_known_mountpoints[] = {
 152	"/sys/kernel/debug",
 153	"/debug",
 154	0,
 155};
 156
 157/*
 158 * data structures
 159 */
 160
 161static int		opt_raw;	/* for kernel developers */
 162static int		opt_list;	/* list pages (in ranges) */
 163static int		opt_no_summary;	/* don't show summary */
 164static pid_t		opt_pid;	/* process to walk */
 165const char *		opt_file;
 
 
 166
 167#define MAX_ADDR_RANGES	1024
 168static int		nr_addr_ranges;
 169static unsigned long	opt_offset[MAX_ADDR_RANGES];
 170static unsigned long	opt_size[MAX_ADDR_RANGES];
 171
 172#define MAX_VMAS	10240
 173static int		nr_vmas;
 174static unsigned long	pg_start[MAX_VMAS];
 175static unsigned long	pg_end[MAX_VMAS];
 176
 177#define MAX_BIT_FILTERS	64
 178static int		nr_bit_filters;
 179static uint64_t		opt_mask[MAX_BIT_FILTERS];
 180static uint64_t		opt_bits[MAX_BIT_FILTERS];
 181
 182static int		page_size;
 183
 184static int		pagemap_fd;
 185static int		kpageflags_fd;
 
 186
 187static int		opt_hwpoison;
 188static int		opt_unpoison;
 189
 190static char		*hwpoison_debug_fs;
 191static int		hwpoison_inject_fd;
 192static int		hwpoison_forget_fd;
 193
 194#define HASH_SHIFT	13
 195#define HASH_SIZE	(1 << HASH_SHIFT)
 196#define HASH_MASK	(HASH_SIZE - 1)
 197#define HASH_KEY(flags)	(flags & HASH_MASK)
 198
 199static unsigned long	total_pages;
 200static unsigned long	nr_pages[HASH_SIZE];
 201static uint64_t		page_flags[HASH_SIZE];
 202
 203
 204/*
 205 * helper functions
 206 */
 207
 208#define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0]))
 209
 210#define min_t(type, x, y) ({			\
 211	type __min1 = (x);			\
 212	type __min2 = (y);			\
 213	__min1 < __min2 ? __min1 : __min2; })
 214
 215#define max_t(type, x, y) ({			\
 216	type __max1 = (x);			\
 217	type __max2 = (y);			\
 218	__max1 > __max2 ? __max1 : __max2; })
 219
 220static unsigned long pages2mb(unsigned long pages)
 221{
 222	return (pages * page_size) >> 20;
 223}
 224
 225static void fatal(const char *x, ...)
 226{
 227	va_list ap;
 228
 229	va_start(ap, x);
 230	vfprintf(stderr, x, ap);
 231	va_end(ap);
 232	exit(EXIT_FAILURE);
 233}
 234
 235static int checked_open(const char *pathname, int flags)
 236{
 237	int fd = open(pathname, flags);
 238
 239	if (fd < 0) {
 240		perror(pathname);
 241		exit(EXIT_FAILURE);
 242	}
 243
 244	return fd;
 245}
 246
 247/*
 248 * pagemap/kpageflags routines
 249 */
 250
 251static unsigned long do_u64_read(int fd, char *name,
 252				 uint64_t *buf,
 253				 unsigned long index,
 254				 unsigned long count)
 255{
 256	long bytes;
 257
 258	if (index > ULONG_MAX / 8)
 259		fatal("index overflow: %lu\n", index);
 260
 261	bytes = pread(fd, buf, count * 8, (off_t)index * 8);
 262	if (bytes < 0) {
 263		perror(name);
 264		exit(EXIT_FAILURE);
 265	}
 266	if (bytes % 8)
 267		fatal("partial read: %lu bytes\n", bytes);
 268
 269	return bytes / 8;
 270}
 271
 272static unsigned long kpageflags_read(uint64_t *buf,
 273				     unsigned long index,
 274				     unsigned long pages)
 275{
 276	return do_u64_read(kpageflags_fd, PROC_KPAGEFLAGS, buf, index, pages);
 277}
 278
 
 
 
 
 
 
 
 
 
 
 279static unsigned long pagemap_read(uint64_t *buf,
 280				  unsigned long index,
 281				  unsigned long pages)
 282{
 283	return do_u64_read(pagemap_fd, "/proc/pid/pagemap", buf, index, pages);
 284}
 285
 286static unsigned long pagemap_pfn(uint64_t val)
 287{
 288	unsigned long pfn;
 289
 290	if (val & PM_PRESENT)
 291		pfn = PM_PFRAME(val);
 292	else
 293		pfn = 0;
 294
 295	return pfn;
 296}
 297
 
 
 
 
 298
 299/*
 300 * page flag names
 301 */
 302
 303static char *page_flag_name(uint64_t flags)
 304{
 305	static char buf[65];
 306	int present;
 307	size_t i, j;
 308
 309	for (i = 0, j = 0; i < ARRAY_SIZE(page_flag_names); i++) {
 310		present = (flags >> i) & 1;
 311		if (!page_flag_names[i]) {
 312			if (present)
 313				fatal("unknown flag bit %d\n", i);
 314			continue;
 315		}
 316		buf[j++] = present ? page_flag_names[i][0] : '_';
 317	}
 318
 319	return buf;
 320}
 321
 322static char *page_flag_longname(uint64_t flags)
 323{
 324	static char buf[1024];
 325	size_t i, n;
 326
 327	for (i = 0, n = 0; i < ARRAY_SIZE(page_flag_names); i++) {
 328		if (!page_flag_names[i])
 329			continue;
 330		if ((flags >> i) & 1)
 331			n += snprintf(buf + n, sizeof(buf) - n, "%s,",
 332					page_flag_names[i] + 2);
 333	}
 334	if (n)
 335		n--;
 336	buf[n] = '\0';
 337
 338	return buf;
 339}
 340
 341
 342/*
 343 * page list and summary
 344 */
 345
 346static void show_page_range(unsigned long voffset, unsigned long offset,
 347			    unsigned long size, uint64_t flags)
 348{
 349	static uint64_t      flags0;
 
 350	static unsigned long voff;
 351	static unsigned long index;
 352	static unsigned long count;
 353
 354	if (flags == flags0 && offset == index + count &&
 355	    size && voffset == voff + count) {
 356		count += size;
 357		return;
 358	}
 359
 360	if (count) {
 361		if (opt_pid)
 362			printf("%lx\t", voff);
 363		if (opt_file)
 364			printf("%lu\t", voff);
 
 
 365		printf("%lx\t%lx\t%s\n",
 366				index, count, page_flag_name(flags0));
 367	}
 368
 369	flags0 = flags;
 
 370	index  = offset;
 371	voff   = voffset;
 372	count  = size;
 373}
 374
 375static void flush_page_range(void)
 376{
 377	show_page_range(0, 0, 0, 0);
 378}
 379
 380static void show_page(unsigned long voffset,
 381		      unsigned long offset, uint64_t flags)
 382{
 383	if (opt_pid)
 384		printf("%lx\t", voffset);
 385	if (opt_file)
 386		printf("%lu\t", voffset);
 
 
 387	printf("%lx\t%s\n", offset, page_flag_name(flags));
 388}
 389
 390static void show_summary(void)
 391{
 392	size_t i;
 393
 394	printf("             flags\tpage-count       MB"
 395		"  symbolic-flags\t\t\tlong-symbolic-flags\n");
 396
 397	for (i = 0; i < ARRAY_SIZE(nr_pages); i++) {
 398		if (nr_pages[i])
 399			printf("0x%016llx\t%10lu %8lu  %s\t%s\n",
 400				(unsigned long long)page_flags[i],
 401				nr_pages[i],
 402				pages2mb(nr_pages[i]),
 403				page_flag_name(page_flags[i]),
 404				page_flag_longname(page_flags[i]));
 405	}
 406
 407	printf("             total\t%10lu %8lu\n",
 408			total_pages, pages2mb(total_pages));
 409}
 410
 411
 412/*
 413 * page flag filters
 414 */
 415
 416static int bit_mask_ok(uint64_t flags)
 417{
 418	int i;
 419
 420	for (i = 0; i < nr_bit_filters; i++) {
 421		if (opt_bits[i] == KPF_ALL_BITS) {
 422			if ((flags & opt_mask[i]) == 0)
 423				return 0;
 424		} else {
 425			if ((flags & opt_mask[i]) != opt_bits[i])
 426				return 0;
 427		}
 428	}
 429
 430	return 1;
 431}
 432
 433static uint64_t expand_overloaded_flags(uint64_t flags, uint64_t pme)
 434{
 435	/* SLOB/SLUB overload several page flags */
 436	if (flags & BIT(SLAB)) {
 437		if (flags & BIT(PRIVATE))
 438			flags ^= BIT(PRIVATE) | BIT(SLOB_FREE);
 439		if (flags & BIT(ACTIVE))
 440			flags ^= BIT(ACTIVE) | BIT(SLUB_FROZEN);
 441		if (flags & BIT(ERROR))
 442			flags ^= BIT(ERROR) | BIT(SLUB_DEBUG);
 443	}
 444
 445	/* PG_reclaim is overloaded as PG_readahead in the read path */
 446	if ((flags & (BIT(RECLAIM) | BIT(WRITEBACK))) == BIT(RECLAIM))
 447		flags ^= BIT(RECLAIM) | BIT(READAHEAD);
 448
 449	if (pme & PM_SOFT_DIRTY)
 450		flags |= BIT(SOFTDIRTY);
 
 
 
 
 
 
 451
 452	return flags;
 453}
 454
 455static uint64_t well_known_flags(uint64_t flags)
 456{
 457	/* hide flags intended only for kernel hacker */
 458	flags &= ~KPF_HACKERS_BITS;
 459
 460	/* hide non-hugeTLB compound pages */
 461	if ((flags & BITS_COMPOUND) && !(flags & BIT(HUGE)))
 462		flags &= ~BITS_COMPOUND;
 463
 464	return flags;
 465}
 466
 467static uint64_t kpageflags_flags(uint64_t flags, uint64_t pme)
 468{
 469	if (opt_raw)
 470		flags = expand_overloaded_flags(flags, pme);
 471	else
 472		flags = well_known_flags(flags);
 473
 474	return flags;
 475}
 476
 477/*
 478 * page actions
 479 */
 480
 481static void prepare_hwpoison_fd(void)
 482{
 483	char buf[MAX_PATH + 1];
 484
 485	hwpoison_debug_fs = debugfs_mount(NULL);
 486	if (!hwpoison_debug_fs) {
 487		perror("mount debugfs");
 488		exit(EXIT_FAILURE);
 489	}
 490
 491	if (opt_hwpoison && !hwpoison_inject_fd) {
 492		snprintf(buf, MAX_PATH, "%s/hwpoison/corrupt-pfn",
 493			hwpoison_debug_fs);
 494		hwpoison_inject_fd = checked_open(buf, O_WRONLY);
 495	}
 496
 497	if (opt_unpoison && !hwpoison_forget_fd) {
 498		snprintf(buf, MAX_PATH, "%s/hwpoison/unpoison-pfn",
 499			hwpoison_debug_fs);
 500		hwpoison_forget_fd = checked_open(buf, O_WRONLY);
 501	}
 502}
 503
 504static int hwpoison_page(unsigned long offset)
 505{
 506	char buf[100];
 507	int len;
 508
 509	len = sprintf(buf, "0x%lx\n", offset);
 510	len = write(hwpoison_inject_fd, buf, len);
 511	if (len < 0) {
 512		perror("hwpoison inject");
 513		return len;
 514	}
 515	return 0;
 516}
 517
 518static int unpoison_page(unsigned long offset)
 519{
 520	char buf[100];
 521	int len;
 522
 523	len = sprintf(buf, "0x%lx\n", offset);
 524	len = write(hwpoison_forget_fd, buf, len);
 525	if (len < 0) {
 526		perror("hwpoison forget");
 527		return len;
 528	}
 529	return 0;
 530}
 531
 532/*
 533 * page frame walker
 534 */
 535
 536static size_t hash_slot(uint64_t flags)
 537{
 538	size_t k = HASH_KEY(flags);
 539	size_t i;
 540
 541	/* Explicitly reserve slot 0 for flags 0: the following logic
 542	 * cannot distinguish an unoccupied slot from slot (flags==0).
 543	 */
 544	if (flags == 0)
 545		return 0;
 546
 547	/* search through the remaining (HASH_SIZE-1) slots */
 548	for (i = 1; i < ARRAY_SIZE(page_flags); i++, k++) {
 549		if (!k || k >= ARRAY_SIZE(page_flags))
 550			k = 1;
 551		if (page_flags[k] == 0) {
 552			page_flags[k] = flags;
 553			return k;
 554		}
 555		if (page_flags[k] == flags)
 556			return k;
 557	}
 558
 559	fatal("hash table full: bump up HASH_SHIFT?\n");
 560	exit(EXIT_FAILURE);
 561}
 562
 563static void add_page(unsigned long voffset,
 564		     unsigned long offset, uint64_t flags, uint64_t pme)
 565{
 566	flags = kpageflags_flags(flags, pme);
 567
 568	if (!bit_mask_ok(flags))
 569		return;
 570
 
 
 
 571	if (opt_hwpoison)
 572		hwpoison_page(offset);
 573	if (opt_unpoison)
 574		unpoison_page(offset);
 575
 576	if (opt_list == 1)
 577		show_page_range(voffset, offset, 1, flags);
 578	else if (opt_list == 2)
 579		show_page(voffset, offset, flags);
 580
 581	nr_pages[hash_slot(flags)]++;
 582	total_pages++;
 583}
 584
 585#define KPAGEFLAGS_BATCH	(64 << 10)	/* 64k pages */
 586static void walk_pfn(unsigned long voffset,
 587		     unsigned long index,
 588		     unsigned long count,
 589		     uint64_t pme)
 590{
 591	uint64_t buf[KPAGEFLAGS_BATCH];
 
 592	unsigned long batch;
 593	unsigned long pages;
 594	unsigned long i;
 595
 
 
 
 
 
 
 
 
 
 
 596	while (count) {
 597		batch = min_t(unsigned long, count, KPAGEFLAGS_BATCH);
 598		pages = kpageflags_read(buf, index, batch);
 599		if (pages == 0)
 600			break;
 601
 
 
 
 602		for (i = 0; i < pages; i++)
 603			add_page(voffset + i, index + i, buf[i], pme);
 604
 605		index += pages;
 606		count -= pages;
 607	}
 608}
 609
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 610#define PAGEMAP_BATCH	(64 << 10)
 611static void walk_vma(unsigned long index, unsigned long count)
 612{
 613	uint64_t buf[PAGEMAP_BATCH];
 614	unsigned long batch;
 615	unsigned long pages;
 616	unsigned long pfn;
 617	unsigned long i;
 618
 619	while (count) {
 620		batch = min_t(unsigned long, count, PAGEMAP_BATCH);
 621		pages = pagemap_read(buf, index, batch);
 622		if (pages == 0)
 623			break;
 624
 625		for (i = 0; i < pages; i++) {
 626			pfn = pagemap_pfn(buf[i]);
 627			if (pfn)
 628				walk_pfn(index + i, pfn, 1, buf[i]);
 
 
 629		}
 630
 631		index += pages;
 632		count -= pages;
 633	}
 634}
 635
 636static void walk_task(unsigned long index, unsigned long count)
 637{
 638	const unsigned long end = index + count;
 639	unsigned long start;
 640	int i = 0;
 641
 642	while (index < end) {
 643
 644		while (pg_end[i] <= index)
 645			if (++i >= nr_vmas)
 646				return;
 647		if (pg_start[i] >= end)
 648			return;
 649
 650		start = max_t(unsigned long, pg_start[i], index);
 651		index = min_t(unsigned long, pg_end[i], end);
 652
 653		assert(start < index);
 654		walk_vma(start, index - start);
 655	}
 656}
 657
 658static void add_addr_range(unsigned long offset, unsigned long size)
 659{
 660	if (nr_addr_ranges >= MAX_ADDR_RANGES)
 661		fatal("too many addr ranges\n");
 662
 663	opt_offset[nr_addr_ranges] = offset;
 664	opt_size[nr_addr_ranges] = min_t(unsigned long, size, ULONG_MAX-offset);
 665	nr_addr_ranges++;
 666}
 667
 668static void walk_addr_ranges(void)
 669{
 670	int i;
 671
 672	kpageflags_fd = checked_open(PROC_KPAGEFLAGS, O_RDONLY);
 673
 674	if (!nr_addr_ranges)
 675		add_addr_range(0, ULONG_MAX);
 676
 677	for (i = 0; i < nr_addr_ranges; i++)
 678		if (!opt_pid)
 679			walk_pfn(opt_offset[i], opt_offset[i], opt_size[i], 0);
 680		else
 681			walk_task(opt_offset[i], opt_size[i]);
 682
 683	close(kpageflags_fd);
 684}
 685
 686
 687/*
 688 * user interface
 689 */
 690
 691static const char *page_flag_type(uint64_t flag)
 692{
 693	if (flag & KPF_HACKERS_BITS)
 694		return "(r)";
 695	if (flag & KPF_OVERLOADED_BITS)
 696		return "(o)";
 697	return "   ";
 698}
 699
 700static void usage(void)
 701{
 702	size_t i, j;
 703
 704	printf(
 705"page-types [options]\n"
 706"            -r|--raw                   Raw mode, for kernel developers\n"
 707"            -d|--describe flags        Describe flags\n"
 708"            -a|--addr    addr-spec     Walk a range of pages\n"
 709"            -b|--bits    bits-spec     Walk pages with specified bits\n"
 
 710"            -p|--pid     pid           Walk process address space\n"
 711"            -f|--file    filename      Walk file address space\n"
 712"            -l|--list                  Show page details in ranges\n"
 713"            -L|--list-each             Show page details one by one\n"
 
 714"            -N|--no-summary            Don't show summary info\n"
 715"            -X|--hwpoison              hwpoison pages\n"
 716"            -x|--unpoison              unpoison pages\n"
 717"            -h|--help                  Show this usage message\n"
 718"flags:\n"
 719"            0x10                       bitfield format, e.g.\n"
 720"            anon                       bit-name, e.g.\n"
 721"            0x10,anon                  comma-separated list, e.g.\n"
 722"addr-spec:\n"
 723"            N                          one page at offset N (unit: pages)\n"
 724"            N+M                        pages range from N to N+M-1\n"
 725"            N,M                        pages range from N to M-1\n"
 726"            N,                         pages range from N to end\n"
 727"            ,M                         pages range from 0 to M-1\n"
 728"bits-spec:\n"
 729"            bit1,bit2                  (flags & (bit1|bit2)) != 0\n"
 730"            bit1,bit2=bit1             (flags & (bit1|bit2)) == bit1\n"
 731"            bit1,~bit2                 (flags & (bit1|bit2)) == bit1\n"
 732"            =bit1,bit2                 flags == (bit1|bit2)\n"
 733"bit-names:\n"
 734	);
 735
 736	for (i = 0, j = 0; i < ARRAY_SIZE(page_flag_names); i++) {
 737		if (!page_flag_names[i])
 738			continue;
 739		printf("%16s%s", page_flag_names[i] + 2,
 740				 page_flag_type(1ULL << i));
 741		if (++j > 3) {
 742			j = 0;
 743			putchar('\n');
 744		}
 745	}
 746	printf("\n                                   "
 747		"(r) raw mode bits  (o) overloaded bits\n");
 748}
 749
 750static unsigned long long parse_number(const char *str)
 751{
 752	unsigned long long n;
 753
 754	n = strtoll(str, NULL, 0);
 755
 756	if (n == 0 && str[0] != '0')
 757		fatal("invalid name or number: %s\n", str);
 758
 759	return n;
 760}
 761
 762static void parse_pid(const char *str)
 763{
 764	FILE *file;
 765	char buf[5000];
 766
 767	opt_pid = parse_number(str);
 768
 769	sprintf(buf, "/proc/%d/pagemap", opt_pid);
 770	pagemap_fd = checked_open(buf, O_RDONLY);
 771
 772	sprintf(buf, "/proc/%d/maps", opt_pid);
 773	file = fopen(buf, "r");
 774	if (!file) {
 775		perror(buf);
 776		exit(EXIT_FAILURE);
 777	}
 778
 779	while (fgets(buf, sizeof(buf), file) != NULL) {
 780		unsigned long vm_start;
 781		unsigned long vm_end;
 782		unsigned long long pgoff;
 783		int major, minor;
 784		char r, w, x, s;
 785		unsigned long ino;
 786		int n;
 787
 788		n = sscanf(buf, "%lx-%lx %c%c%c%c %llx %x:%x %lu",
 789			   &vm_start,
 790			   &vm_end,
 791			   &r, &w, &x, &s,
 792			   &pgoff,
 793			   &major, &minor,
 794			   &ino);
 795		if (n < 10) {
 796			fprintf(stderr, "unexpected line: %s\n", buf);
 797			continue;
 798		}
 799		pg_start[nr_vmas] = vm_start / page_size;
 800		pg_end[nr_vmas] = vm_end / page_size;
 801		if (++nr_vmas >= MAX_VMAS) {
 802			fprintf(stderr, "too many VMAs\n");
 803			break;
 804		}
 805	}
 806	fclose(file);
 807}
 808
 809static void show_file(const char *name, const struct stat *st)
 810{
 811	unsigned long long size = st->st_size;
 812	char atime[64], mtime[64];
 813	long now = time(NULL);
 814
 815	printf("%s\tInode: %u\tSize: %llu (%llu pages)\n",
 816			name, (unsigned)st->st_ino,
 817			size, (size + page_size - 1) / page_size);
 818
 819	strftime(atime, sizeof(atime), "%c", localtime(&st->st_atime));
 820	strftime(mtime, sizeof(mtime), "%c", localtime(&st->st_mtime));
 821
 822	printf("Modify: %s (%ld seconds ago)\nAccess: %s (%ld seconds ago)\n",
 823			mtime, now - st->st_mtime,
 824			atime, now - st->st_atime);
 825}
 826
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 827static void walk_file(const char *name, const struct stat *st)
 828{
 829	uint8_t vec[PAGEMAP_BATCH];
 830	uint64_t buf[PAGEMAP_BATCH], flags;
 
 831	unsigned long nr_pages, pfn, i;
 
 832	int fd;
 833	off_t off;
 834	ssize_t len;
 835	void *ptr;
 836	int first = 1;
 837
 838	fd = checked_open(name, O_RDONLY|O_NOATIME|O_NOFOLLOW);
 839
 840	for (off = 0; off < st->st_size; off += len) {
 841		nr_pages = (st->st_size - off + page_size - 1) / page_size;
 842		if (nr_pages > PAGEMAP_BATCH)
 843			nr_pages = PAGEMAP_BATCH;
 844		len = nr_pages * page_size;
 845
 846		ptr = mmap(NULL, len, PROT_READ, MAP_SHARED, fd, off);
 847		if (ptr == MAP_FAILED)
 848			fatal("mmap failed: %s", name);
 849
 850		/* determine cached pages */
 851		if (mincore(ptr, len, vec))
 852			fatal("mincore failed: %s", name);
 853
 854		/* turn off readahead */
 855		if (madvise(ptr, len, MADV_RANDOM))
 856			fatal("madvice failed: %s", name);
 857
 
 
 
 
 
 
 
 858		/* populate ptes */
 859		for (i = 0; i < nr_pages ; i++) {
 860			if (vec[i] & 1)
 861				(void)*(volatile int *)(ptr + i * page_size);
 862		}
 
 863
 864		/* turn off harvesting reference bits */
 865		if (madvise(ptr, len, MADV_SEQUENTIAL))
 866			fatal("madvice failed: %s", name);
 867
 868		if (pagemap_read(buf, (unsigned long)ptr / page_size,
 869					nr_pages) != nr_pages)
 870			fatal("cannot read pagemap");
 871
 872		munmap(ptr, len);
 873
 874		for (i = 0; i < nr_pages; i++) {
 875			pfn = pagemap_pfn(buf[i]);
 876			if (!pfn)
 877				continue;
 878			if (!kpageflags_read(&flags, pfn, 1))
 879				continue;
 
 
 880			if (first && opt_list) {
 881				first = 0;
 882				flush_page_range();
 883				show_file(name, st);
 884			}
 885			add_page(off / page_size + i, pfn, flags, buf[i]);
 
 886		}
 887	}
 888
 889	close(fd);
 890}
 891
 892int walk_tree(const char *name, const struct stat *st, int type, struct FTW *f)
 893{
 894	(void)f;
 895	switch (type) {
 896	case FTW_F:
 897		if (S_ISREG(st->st_mode))
 898			walk_file(name, st);
 899		break;
 900	case FTW_DNR:
 901		fprintf(stderr, "cannot read dir: %s\n", name);
 902		break;
 903	}
 904	return 0;
 905}
 906
 907static void walk_page_cache(void)
 908{
 909	struct stat st;
 910
 911	kpageflags_fd = checked_open(PROC_KPAGEFLAGS, O_RDONLY);
 912	pagemap_fd = checked_open("/proc/self/pagemap", O_RDONLY);
 
 913
 914	if (stat(opt_file, &st))
 915		fatal("stat failed: %s\n", opt_file);
 916
 917	if (S_ISREG(st.st_mode)) {
 918		walk_file(opt_file, &st);
 919	} else if (S_ISDIR(st.st_mode)) {
 920		/* do not follow symlinks and mountpoints */
 921		if (nftw(opt_file, walk_tree, 64, FTW_MOUNT | FTW_PHYS) < 0)
 922			fatal("nftw failed: %s\n", opt_file);
 923	} else
 924		fatal("unhandled file type: %s\n", opt_file);
 925
 926	close(kpageflags_fd);
 927	close(pagemap_fd);
 
 928}
 929
 930static void parse_file(const char *name)
 931{
 932	opt_file = name;
 933}
 934
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 935static void parse_addr_range(const char *optarg)
 936{
 937	unsigned long offset;
 938	unsigned long size;
 939	char *p;
 940
 941	p = strchr(optarg, ',');
 942	if (!p)
 943		p = strchr(optarg, '+');
 944
 945	if (p == optarg) {
 946		offset = 0;
 947		size   = parse_number(p + 1);
 948	} else if (p) {
 949		offset = parse_number(optarg);
 950		if (p[1] == '\0')
 951			size = ULONG_MAX;
 952		else {
 953			size = parse_number(p + 1);
 954			if (*p == ',') {
 955				if (size < offset)
 956					fatal("invalid range: %lu,%lu\n",
 957							offset, size);
 958				size -= offset;
 959			}
 960		}
 961	} else {
 962		offset = parse_number(optarg);
 963		size   = 1;
 964	}
 965
 966	add_addr_range(offset, size);
 967}
 968
 969static void add_bits_filter(uint64_t mask, uint64_t bits)
 970{
 971	if (nr_bit_filters >= MAX_BIT_FILTERS)
 972		fatal("too much bit filters\n");
 973
 974	opt_mask[nr_bit_filters] = mask;
 975	opt_bits[nr_bit_filters] = bits;
 976	nr_bit_filters++;
 977}
 978
 979static uint64_t parse_flag_name(const char *str, int len)
 980{
 981	size_t i;
 982
 983	if (!*str || !len)
 984		return 0;
 985
 986	if (len <= 8 && !strncmp(str, "compound", len))
 987		return BITS_COMPOUND;
 988
 989	for (i = 0; i < ARRAY_SIZE(page_flag_names); i++) {
 990		if (!page_flag_names[i])
 991			continue;
 992		if (!strncmp(str, page_flag_names[i] + 2, len))
 993			return 1ULL << i;
 994	}
 995
 996	return parse_number(str);
 997}
 998
 999static uint64_t parse_flag_names(const char *str, int all)
1000{
1001	const char *p    = str;
1002	uint64_t   flags = 0;
1003
1004	while (1) {
1005		if (*p == ',' || *p == '=' || *p == '\0') {
1006			if ((*str != '~') || (*str == '~' && all && *++str))
1007				flags |= parse_flag_name(str, p - str);
1008			if (*p != ',')
1009				break;
1010			str = p + 1;
1011		}
1012		p++;
1013	}
1014
1015	return flags;
1016}
1017
1018static void parse_bits_mask(const char *optarg)
1019{
1020	uint64_t mask;
1021	uint64_t bits;
1022	const char *p;
1023
1024	p = strchr(optarg, '=');
1025	if (p == optarg) {
1026		mask = KPF_ALL_BITS;
1027		bits = parse_flag_names(p + 1, 0);
1028	} else if (p) {
1029		mask = parse_flag_names(optarg, 0);
1030		bits = parse_flag_names(p + 1, 0);
1031	} else if (strchr(optarg, '~')) {
1032		mask = parse_flag_names(optarg, 1);
1033		bits = parse_flag_names(optarg, 0);
1034	} else {
1035		mask = parse_flag_names(optarg, 0);
1036		bits = KPF_ALL_BITS;
1037	}
1038
1039	add_bits_filter(mask, bits);
1040}
1041
1042static void describe_flags(const char *optarg)
1043{
1044	uint64_t flags = parse_flag_names(optarg, 0);
1045
1046	printf("0x%016llx\t%s\t%s\n",
1047		(unsigned long long)flags,
1048		page_flag_name(flags),
1049		page_flag_longname(flags));
1050}
1051
1052static const struct option opts[] = {
1053	{ "raw"       , 0, NULL, 'r' },
1054	{ "pid"       , 1, NULL, 'p' },
1055	{ "file"      , 1, NULL, 'f' },
1056	{ "addr"      , 1, NULL, 'a' },
1057	{ "bits"      , 1, NULL, 'b' },
 
1058	{ "describe"  , 1, NULL, 'd' },
1059	{ "list"      , 0, NULL, 'l' },
1060	{ "list-each" , 0, NULL, 'L' },
 
1061	{ "no-summary", 0, NULL, 'N' },
1062	{ "hwpoison"  , 0, NULL, 'X' },
1063	{ "unpoison"  , 0, NULL, 'x' },
1064	{ "help"      , 0, NULL, 'h' },
1065	{ NULL        , 0, NULL, 0 }
1066};
1067
1068int main(int argc, char *argv[])
1069{
1070	int c;
1071
1072	page_size = getpagesize();
1073
1074	while ((c = getopt_long(argc, argv,
1075				"rp:f:a:b:d:lLNXxh", opts, NULL)) != -1) {
1076		switch (c) {
1077		case 'r':
1078			opt_raw = 1;
1079			break;
1080		case 'p':
1081			parse_pid(optarg);
1082			break;
1083		case 'f':
1084			parse_file(optarg);
1085			break;
1086		case 'a':
1087			parse_addr_range(optarg);
1088			break;
1089		case 'b':
1090			parse_bits_mask(optarg);
1091			break;
 
 
 
 
 
 
1092		case 'd':
1093			describe_flags(optarg);
1094			exit(0);
1095		case 'l':
1096			opt_list = 1;
1097			break;
1098		case 'L':
1099			opt_list = 2;
1100			break;
1101		case 'N':
1102			opt_no_summary = 1;
1103			break;
1104		case 'X':
1105			opt_hwpoison = 1;
1106			prepare_hwpoison_fd();
1107			break;
1108		case 'x':
1109			opt_unpoison = 1;
1110			prepare_hwpoison_fd();
1111			break;
1112		case 'h':
1113			usage();
1114			exit(0);
1115		default:
1116			usage();
1117			exit(1);
1118		}
1119	}
1120
 
 
 
1121	if (opt_list && opt_pid)
1122		printf("voffset\t");
1123	if (opt_list && opt_file)
1124		printf("foffset\t");
 
 
1125	if (opt_list == 1)
1126		printf("offset\tlen\tflags\n");
1127	if (opt_list == 2)
1128		printf("offset\tflags\n");
1129
1130	if (opt_file)
1131		walk_page_cache();
1132	else
1133		walk_addr_ranges();
1134
1135	if (opt_list == 1)
1136		flush_page_range();
1137
1138	if (opt_no_summary)
1139		return 0;
1140
1141	if (opt_list)
1142		printf("\n\n");
1143
1144	show_summary();
1145
1146	return 0;
1147}