• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * page-types: Tool for querying page flags
4  *
5  * Copyright (C) 2009 Intel corporation
6  *
7  * Authors: Wu Fengguang <fengguang.wu@intel.com>
8  */
9 
10 #define _FILE_OFFSET_BITS 64
11 #define _GNU_SOURCE
12 #include <stdio.h>
13 #include <stdlib.h>
14 #include <unistd.h>
15 #include <stdint.h>
16 #include <stdarg.h>
17 #include <string.h>
18 #include <getopt.h>
19 #include <limits.h>
20 #include <assert.h>
21 #include <ftw.h>
22 #include <time.h>
23 #include <setjmp.h>
24 #include <signal.h>
25 #include <sys/types.h>
26 #include <sys/errno.h>
27 #include <sys/fcntl.h>
28 #include <sys/mount.h>
29 #include <sys/statfs.h>
30 #include <sys/mman.h>
31 #include "../../include/uapi/linux/magic.h"
32 #include "../../include/uapi/linux/kernel-page-flags.h"
33 #include <api/fs/fs.h>
34 
35 #ifndef MAX_PATH
36 # define MAX_PATH 256
37 #endif
38 
39 #ifndef STR
40 # define _STR(x) #x
41 # define STR(x) _STR(x)
42 #endif
43 
44 /*
45  * pagemap kernel ABI bits
46  */
47 
48 #define PM_ENTRY_BYTES		8
49 #define PM_PFRAME_BITS		55
50 #define PM_PFRAME_MASK		((1LL << PM_PFRAME_BITS) - 1)
51 #define PM_PFRAME(x)		((x) & PM_PFRAME_MASK)
52 #define MAX_SWAPFILES_SHIFT	5
53 #define PM_SWAP_OFFSET(x)	(((x) & PM_PFRAME_MASK) >> MAX_SWAPFILES_SHIFT)
54 #define PM_SOFT_DIRTY		(1ULL << 55)
55 #define PM_MMAP_EXCLUSIVE	(1ULL << 56)
56 #define PM_FILE			(1ULL << 61)
57 #define PM_SWAP			(1ULL << 62)
58 #define PM_PRESENT		(1ULL << 63)
59 
60 /*
61  * kernel page flags
62  */
63 
64 #define KPF_BYTES		8
65 #define PROC_KPAGEFLAGS		"/proc/kpageflags"
66 #define PROC_KPAGECOUNT		"/proc/kpagecount"
67 #define PROC_KPAGECGROUP	"/proc/kpagecgroup"
68 
69 #define SYS_KERNEL_MM_PAGE_IDLE "/sys/kernel/mm/page_idle/bitmap"
70 
71 /* [32-] kernel hacking assistances */
72 #define KPF_RESERVED		32
73 #define KPF_MLOCKED		33
74 #define KPF_MAPPEDTODISK	34
75 #define KPF_PRIVATE		35
76 #define KPF_PRIVATE_2		36
77 #define KPF_OWNER_PRIVATE	37
78 #define KPF_ARCH		38
79 #define KPF_UNCACHED		39
80 #define KPF_SOFTDIRTY		40
81 
82 /* [48-] take some arbitrary free slots for expanding overloaded flags
83  * not part of kernel API
84  */
85 #define KPF_READAHEAD		48
86 #define KPF_SLOB_FREE		49
87 #define KPF_SLUB_FROZEN		50
88 #define KPF_SLUB_DEBUG		51
89 #define KPF_FILE		61
90 #define KPF_SWAP		62
91 #define KPF_MMAP_EXCLUSIVE	63
92 
93 #define KPF_ALL_BITS		((uint64_t)~0ULL)
94 #define KPF_HACKERS_BITS	(0xffffULL << 32)
95 #define KPF_OVERLOADED_BITS	(0xffffULL << 48)
96 #define BIT(name)		(1ULL << KPF_##name)
97 #define BITS_COMPOUND		(BIT(COMPOUND_HEAD) | BIT(COMPOUND_TAIL))
98 
99 static const char * const page_flag_names[] = {
100 	[KPF_LOCKED]		= "L:locked",
101 	[KPF_ERROR]		= "E:error",
102 	[KPF_REFERENCED]	= "R:referenced",
103 	[KPF_UPTODATE]		= "U:uptodate",
104 	[KPF_DIRTY]		= "D:dirty",
105 	[KPF_LRU]		= "l:lru",
106 	[KPF_ACTIVE]		= "A:active",
107 	[KPF_SLAB]		= "S:slab",
108 	[KPF_WRITEBACK]		= "W:writeback",
109 	[KPF_RECLAIM]		= "I:reclaim",
110 	[KPF_BUDDY]		= "B:buddy",
111 
112 	[KPF_MMAP]		= "M:mmap",
113 	[KPF_ANON]		= "a:anonymous",
114 	[KPF_SWAPCACHE]		= "s:swapcache",
115 	[KPF_SWAPBACKED]	= "b:swapbacked",
116 	[KPF_COMPOUND_HEAD]	= "H:compound_head",
117 	[KPF_COMPOUND_TAIL]	= "T:compound_tail",
118 	[KPF_HUGE]		= "G:huge",
119 	[KPF_UNEVICTABLE]	= "u:unevictable",
120 	[KPF_HWPOISON]		= "X:hwpoison",
121 	[KPF_NOPAGE]		= "n:nopage",
122 	[KPF_KSM]		= "x:ksm",
123 	[KPF_THP]		= "t:thp",
124 	[KPF_OFFLINE]		= "o:offline",
125 	[KPF_PGTABLE]		= "g:pgtable",
126 	[KPF_ZERO_PAGE]		= "z:zero_page",
127 	[KPF_IDLE]              = "i:idle_page",
128 
129 	[KPF_RESERVED]		= "r:reserved",
130 	[KPF_MLOCKED]		= "m:mlocked",
131 	[KPF_MAPPEDTODISK]	= "d:mappedtodisk",
132 	[KPF_PRIVATE]		= "P:private",
133 	[KPF_PRIVATE_2]		= "p:private_2",
134 	[KPF_OWNER_PRIVATE]	= "O:owner_private",
135 	[KPF_ARCH]		= "h:arch",
136 	[KPF_UNCACHED]		= "c:uncached",
137 	[KPF_SOFTDIRTY]		= "f:softdirty",
138 
139 	[KPF_READAHEAD]		= "I:readahead",
140 	[KPF_SLOB_FREE]		= "P:slob_free",
141 	[KPF_SLUB_FROZEN]	= "A:slub_frozen",
142 	[KPF_SLUB_DEBUG]	= "E:slub_debug",
143 
144 	[KPF_FILE]		= "F:file",
145 	[KPF_SWAP]		= "w:swap",
146 	[KPF_MMAP_EXCLUSIVE]	= "1:mmap_exclusive",
147 };
148 
149 
150 /*
151  * data structures
152  */
153 
154 static int		opt_raw;	/* for kernel developers */
155 static int		opt_list;	/* list pages (in ranges) */
156 static int		opt_mark_idle;	/* set accessed bit */
157 static int		opt_no_summary;	/* don't show summary */
158 static pid_t		opt_pid;	/* process to walk */
159 const char		*opt_file;	/* file or directory path */
160 static uint64_t		opt_cgroup;	/* cgroup inode */
161 static int		opt_list_cgroup;/* list page cgroup */
162 static int		opt_list_mapcnt;/* list page map count */
163 static const char	*opt_kpageflags;/* kpageflags file to parse */
164 
165 #define MAX_ADDR_RANGES	1024
166 static int		nr_addr_ranges;
167 static unsigned long	opt_offset[MAX_ADDR_RANGES];
168 static unsigned long	opt_size[MAX_ADDR_RANGES];
169 
170 #define MAX_VMAS	10240
171 static int		nr_vmas;
172 static unsigned long	pg_start[MAX_VMAS];
173 static unsigned long	pg_end[MAX_VMAS];
174 
175 #define MAX_BIT_FILTERS	64
176 static int		nr_bit_filters;
177 static uint64_t		opt_mask[MAX_BIT_FILTERS];
178 static uint64_t		opt_bits[MAX_BIT_FILTERS];
179 
180 static int		page_size;
181 
182 static int		pagemap_fd;
183 static int		kpageflags_fd;
184 static int		kpagecount_fd = -1;
185 static int		kpagecgroup_fd = -1;
186 static int		page_idle_fd = -1;
187 
188 static int		opt_hwpoison;
189 static int		opt_unpoison;
190 
191 static const char	*hwpoison_debug_fs;
192 static int		hwpoison_inject_fd;
193 static int		hwpoison_forget_fd;
194 
195 #define HASH_SHIFT	13
196 #define HASH_SIZE	(1 << HASH_SHIFT)
197 #define HASH_MASK	(HASH_SIZE - 1)
198 #define HASH_KEY(flags)	(flags & HASH_MASK)
199 
200 static unsigned long	total_pages;
201 static unsigned long	nr_pages[HASH_SIZE];
202 static uint64_t		page_flags[HASH_SIZE];
203 
204 
205 /*
206  * helper functions
207  */
208 
209 #define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0]))
210 
211 #define min_t(type, x, y) ({			\
212 	type __min1 = (x);			\
213 	type __min2 = (y);			\
214 	__min1 < __min2 ? __min1 : __min2; })
215 
216 #define max_t(type, x, y) ({			\
217 	type __max1 = (x);			\
218 	type __max2 = (y);			\
219 	__max1 > __max2 ? __max1 : __max2; })
220 
pages2mb(unsigned long pages)221 static unsigned long pages2mb(unsigned long pages)
222 {
223 	return (pages * page_size) >> 20;
224 }
225 
fatal(const char * x,...)226 static void fatal(const char *x, ...)
227 {
228 	va_list ap;
229 
230 	va_start(ap, x);
231 	vfprintf(stderr, x, ap);
232 	va_end(ap);
233 	exit(EXIT_FAILURE);
234 }
235 
checked_open(const char * pathname,int flags)236 static int checked_open(const char *pathname, int flags)
237 {
238 	int fd = open(pathname, flags);
239 
240 	if (fd < 0) {
241 		perror(pathname);
242 		exit(EXIT_FAILURE);
243 	}
244 
245 	return fd;
246 }
247 
248 /*
249  * pagemap/kpageflags routines
250  */
251 
do_u64_read(int fd,const char * name,uint64_t * buf,unsigned long index,unsigned long count)252 static unsigned long do_u64_read(int fd, const char *name,
253 				 uint64_t *buf,
254 				 unsigned long index,
255 				 unsigned long count)
256 {
257 	long bytes;
258 
259 	if (index > ULONG_MAX / 8)
260 		fatal("index overflow: %lu\n", index);
261 
262 	bytes = pread(fd, buf, count * 8, (off_t)index * 8);
263 	if (bytes < 0) {
264 		perror(name);
265 		exit(EXIT_FAILURE);
266 	}
267 	if (bytes % 8)
268 		fatal("partial read: %lu bytes\n", bytes);
269 
270 	return bytes / 8;
271 }
272 
kpageflags_read(uint64_t * buf,unsigned long index,unsigned long pages)273 static unsigned long kpageflags_read(uint64_t *buf,
274 				     unsigned long index,
275 				     unsigned long pages)
276 {
277 	return do_u64_read(kpageflags_fd, opt_kpageflags, buf, index, pages);
278 }
279 
kpagecgroup_read(uint64_t * buf,unsigned long index,unsigned long pages)280 static unsigned long kpagecgroup_read(uint64_t *buf,
281 				      unsigned long index,
282 				      unsigned long pages)
283 {
284 	if (kpagecgroup_fd < 0)
285 		return pages;
286 
287 	return do_u64_read(kpagecgroup_fd, opt_kpageflags, buf, index, pages);
288 }
289 
kpagecount_read(uint64_t * buf,unsigned long index,unsigned long pages)290 static unsigned long kpagecount_read(uint64_t *buf,
291 				     unsigned long index,
292 				     unsigned long pages)
293 {
294 	return kpagecount_fd < 0 ? pages :
295 		do_u64_read(kpagecount_fd, PROC_KPAGECOUNT,
296 			    buf, index, pages);
297 }
298 
pagemap_read(uint64_t * buf,unsigned long index,unsigned long pages)299 static 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 
pagemap_pfn(uint64_t val)306 static 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 
pagemap_swap_offset(uint64_t val)318 static 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 
page_flag_name(uint64_t flags)327 static 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 
page_flag_longname(uint64_t flags)346 static 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 
show_page_range(unsigned long voffset,unsigned long offset,unsigned long size,uint64_t flags,uint64_t cgroup,uint64_t mapcnt)370 static void show_page_range(unsigned long voffset, unsigned long offset,
371 			    unsigned long size, uint64_t flags,
372 			    uint64_t cgroup, uint64_t mapcnt)
373 {
374 	static uint64_t      flags0;
375 	static uint64_t	     cgroup0;
376 	static uint64_t      mapcnt0;
377 	static unsigned long voff;
378 	static unsigned long index;
379 	static unsigned long count;
380 
381 	if (flags == flags0 && cgroup == cgroup0 && mapcnt == mapcnt0 &&
382 	    offset == index + count && size && voffset == voff + count) {
383 		count += size;
384 		return;
385 	}
386 
387 	if (count) {
388 		if (opt_pid)
389 			printf("%lx\t", voff);
390 		if (opt_file)
391 			printf("%lu\t", voff);
392 		if (opt_list_cgroup)
393 			printf("@%llu\t", (unsigned long long)cgroup0);
394 		if (opt_list_mapcnt)
395 			printf("%lu\t", mapcnt0);
396 		printf("%lx\t%lx\t%s\n",
397 				index, count, page_flag_name(flags0));
398 	}
399 
400 	flags0 = flags;
401 	cgroup0 = cgroup;
402 	mapcnt0 = mapcnt;
403 	index  = offset;
404 	voff   = voffset;
405 	count  = size;
406 }
407 
flush_page_range(void)408 static void flush_page_range(void)
409 {
410 	show_page_range(0, 0, 0, 0, 0, 0);
411 }
412 
show_page(unsigned long voffset,unsigned long offset,uint64_t flags,uint64_t cgroup,uint64_t mapcnt)413 static void show_page(unsigned long voffset, unsigned long offset,
414 		      uint64_t flags, uint64_t cgroup, uint64_t mapcnt)
415 {
416 	if (opt_pid)
417 		printf("%lx\t", voffset);
418 	if (opt_file)
419 		printf("%lu\t", voffset);
420 	if (opt_list_cgroup)
421 		printf("@%llu\t", (unsigned long long)cgroup);
422 	if (opt_list_mapcnt)
423 		printf("%lu\t", mapcnt);
424 
425 	printf("%lx\t%s\n", offset, page_flag_name(flags));
426 }
427 
show_summary(void)428 static void show_summary(void)
429 {
430 	size_t i;
431 
432 	printf("             flags\tpage-count       MB"
433 		"  symbolic-flags\t\t\tlong-symbolic-flags\n");
434 
435 	for (i = 0; i < ARRAY_SIZE(nr_pages); i++) {
436 		if (nr_pages[i])
437 			printf("0x%016llx\t%10lu %8lu  %s\t%s\n",
438 				(unsigned long long)page_flags[i],
439 				nr_pages[i],
440 				pages2mb(nr_pages[i]),
441 				page_flag_name(page_flags[i]),
442 				page_flag_longname(page_flags[i]));
443 	}
444 
445 	printf("             total\t%10lu %8lu\n",
446 			total_pages, pages2mb(total_pages));
447 }
448 
449 
450 /*
451  * page flag filters
452  */
453 
bit_mask_ok(uint64_t flags)454 static int bit_mask_ok(uint64_t flags)
455 {
456 	int i;
457 
458 	for (i = 0; i < nr_bit_filters; i++) {
459 		if (opt_bits[i] == KPF_ALL_BITS) {
460 			if ((flags & opt_mask[i]) == 0)
461 				return 0;
462 		} else {
463 			if ((flags & opt_mask[i]) != opt_bits[i])
464 				return 0;
465 		}
466 	}
467 
468 	return 1;
469 }
470 
expand_overloaded_flags(uint64_t flags,uint64_t pme)471 static uint64_t expand_overloaded_flags(uint64_t flags, uint64_t pme)
472 {
473 	/* SLOB/SLUB overload several page flags */
474 	if (flags & BIT(SLAB)) {
475 		if (flags & BIT(PRIVATE))
476 			flags ^= BIT(PRIVATE) | BIT(SLOB_FREE);
477 		if (flags & BIT(ACTIVE))
478 			flags ^= BIT(ACTIVE) | BIT(SLUB_FROZEN);
479 		if (flags & BIT(ERROR))
480 			flags ^= BIT(ERROR) | BIT(SLUB_DEBUG);
481 	}
482 
483 	/* PG_reclaim is overloaded as PG_readahead in the read path */
484 	if ((flags & (BIT(RECLAIM) | BIT(WRITEBACK))) == BIT(RECLAIM))
485 		flags ^= BIT(RECLAIM) | BIT(READAHEAD);
486 
487 	if (pme & PM_SOFT_DIRTY)
488 		flags |= BIT(SOFTDIRTY);
489 	if (pme & PM_FILE)
490 		flags |= BIT(FILE);
491 	if (pme & PM_SWAP)
492 		flags |= BIT(SWAP);
493 	if (pme & PM_MMAP_EXCLUSIVE)
494 		flags |= BIT(MMAP_EXCLUSIVE);
495 
496 	return flags;
497 }
498 
well_known_flags(uint64_t flags)499 static uint64_t well_known_flags(uint64_t flags)
500 {
501 	/* hide flags intended only for kernel hacker */
502 	flags &= ~KPF_HACKERS_BITS;
503 
504 	/* hide non-hugeTLB compound pages */
505 	if ((flags & BITS_COMPOUND) && !(flags & BIT(HUGE)))
506 		flags &= ~BITS_COMPOUND;
507 
508 	return flags;
509 }
510 
kpageflags_flags(uint64_t flags,uint64_t pme)511 static uint64_t kpageflags_flags(uint64_t flags, uint64_t pme)
512 {
513 	if (opt_raw)
514 		flags = expand_overloaded_flags(flags, pme);
515 	else
516 		flags = well_known_flags(flags);
517 
518 	return flags;
519 }
520 
521 /*
522  * page actions
523  */
524 
prepare_hwpoison_fd(void)525 static void prepare_hwpoison_fd(void)
526 {
527 	char buf[MAX_PATH + 1];
528 
529 	hwpoison_debug_fs = debugfs__mount();
530 	if (!hwpoison_debug_fs) {
531 		perror("mount debugfs");
532 		exit(EXIT_FAILURE);
533 	}
534 
535 	if (opt_hwpoison && !hwpoison_inject_fd) {
536 		snprintf(buf, MAX_PATH, "%s/hwpoison/corrupt-pfn",
537 			hwpoison_debug_fs);
538 		hwpoison_inject_fd = checked_open(buf, O_WRONLY);
539 	}
540 
541 	if (opt_unpoison && !hwpoison_forget_fd) {
542 		snprintf(buf, MAX_PATH, "%s/hwpoison/unpoison-pfn",
543 			hwpoison_debug_fs);
544 		hwpoison_forget_fd = checked_open(buf, O_WRONLY);
545 	}
546 }
547 
hwpoison_page(unsigned long offset)548 static int hwpoison_page(unsigned long offset)
549 {
550 	char buf[100];
551 	int len;
552 
553 	len = sprintf(buf, "0x%lx\n", offset);
554 	len = write(hwpoison_inject_fd, buf, len);
555 	if (len < 0) {
556 		perror("hwpoison inject");
557 		return len;
558 	}
559 	return 0;
560 }
561 
unpoison_page(unsigned long offset)562 static int unpoison_page(unsigned long offset)
563 {
564 	char buf[100];
565 	int len;
566 
567 	len = sprintf(buf, "0x%lx\n", offset);
568 	len = write(hwpoison_forget_fd, buf, len);
569 	if (len < 0) {
570 		perror("hwpoison forget");
571 		return len;
572 	}
573 	return 0;
574 }
575 
mark_page_idle(unsigned long offset)576 static int mark_page_idle(unsigned long offset)
577 {
578 	static unsigned long off;
579 	static uint64_t buf;
580 	int len;
581 
582 	if ((offset / 64 == off / 64) || buf == 0) {
583 		buf |= 1UL << (offset % 64);
584 		off = offset;
585 		return 0;
586 	}
587 
588 	len = pwrite(page_idle_fd, &buf, 8, 8 * (off / 64));
589 	if (len < 0) {
590 		perror("mark page idle");
591 		return len;
592 	}
593 
594 	buf = 1UL << (offset % 64);
595 	off = offset;
596 
597 	return 0;
598 }
599 
600 /*
601  * page frame walker
602  */
603 
hash_slot(uint64_t flags)604 static size_t hash_slot(uint64_t flags)
605 {
606 	size_t k = HASH_KEY(flags);
607 	size_t i;
608 
609 	/* Explicitly reserve slot 0 for flags 0: the following logic
610 	 * cannot distinguish an unoccupied slot from slot (flags==0).
611 	 */
612 	if (flags == 0)
613 		return 0;
614 
615 	/* search through the remaining (HASH_SIZE-1) slots */
616 	for (i = 1; i < ARRAY_SIZE(page_flags); i++, k++) {
617 		if (!k || k >= ARRAY_SIZE(page_flags))
618 			k = 1;
619 		if (page_flags[k] == 0) {
620 			page_flags[k] = flags;
621 			return k;
622 		}
623 		if (page_flags[k] == flags)
624 			return k;
625 	}
626 
627 	fatal("hash table full: bump up HASH_SHIFT?\n");
628 	exit(EXIT_FAILURE);
629 }
630 
add_page(unsigned long voffset,unsigned long offset,uint64_t flags,uint64_t cgroup,uint64_t mapcnt,uint64_t pme)631 static void add_page(unsigned long voffset, unsigned long offset,
632 		     uint64_t flags, uint64_t cgroup, uint64_t mapcnt,
633 		     uint64_t pme)
634 {
635 	flags = kpageflags_flags(flags, pme);
636 
637 	if (!bit_mask_ok(flags))
638 		return;
639 
640 	if (opt_cgroup && cgroup != (uint64_t)opt_cgroup)
641 		return;
642 
643 	if (opt_hwpoison)
644 		hwpoison_page(offset);
645 	if (opt_unpoison)
646 		unpoison_page(offset);
647 
648 	if (opt_mark_idle)
649 		mark_page_idle(offset);
650 
651 	if (opt_list == 1)
652 		show_page_range(voffset, offset, 1, flags, cgroup, mapcnt);
653 	else if (opt_list == 2)
654 		show_page(voffset, offset, flags, cgroup, mapcnt);
655 
656 	nr_pages[hash_slot(flags)]++;
657 	total_pages++;
658 }
659 
660 #define KPAGEFLAGS_BATCH	(64 << 10)	/* 64k pages */
walk_pfn(unsigned long voffset,unsigned long index,unsigned long count,uint64_t pme)661 static void walk_pfn(unsigned long voffset,
662 		     unsigned long index,
663 		     unsigned long count,
664 		     uint64_t pme)
665 {
666 	uint64_t buf[KPAGEFLAGS_BATCH];
667 	uint64_t cgi[KPAGEFLAGS_BATCH];
668 	uint64_t cnt[KPAGEFLAGS_BATCH];
669 	unsigned long batch;
670 	unsigned long pages;
671 	unsigned long i;
672 
673 	/*
674 	 * kpagecgroup_read() reads only if kpagecgroup were opened, but
675 	 * /proc/kpagecgroup might even not exist, so it's better to fill
676 	 * them with zeros here.
677 	 */
678 	if (count == 1)
679 		cgi[0] = 0;
680 	else
681 		memset(cgi, 0, sizeof cgi);
682 
683 	while (count) {
684 		batch = min_t(unsigned long, count, KPAGEFLAGS_BATCH);
685 		pages = kpageflags_read(buf, index, batch);
686 		if (pages == 0)
687 			break;
688 
689 		if (kpagecgroup_read(cgi, index, pages) != pages)
690 			fatal("kpagecgroup returned fewer pages than expected");
691 
692 		if (kpagecount_read(cnt, index, pages) != pages)
693 			fatal("kpagecount returned fewer pages than expected");
694 
695 		for (i = 0; i < pages; i++)
696 			add_page(voffset + i, index + i,
697 				 buf[i], cgi[i], cnt[i], pme);
698 
699 		index += pages;
700 		count -= pages;
701 	}
702 }
703 
walk_swap(unsigned long voffset,uint64_t pme)704 static void walk_swap(unsigned long voffset, uint64_t pme)
705 {
706 	uint64_t flags = kpageflags_flags(0, pme);
707 
708 	if (!bit_mask_ok(flags))
709 		return;
710 
711 	if (opt_cgroup)
712 		return;
713 
714 	if (opt_list == 1)
715 		show_page_range(voffset, pagemap_swap_offset(pme),
716 				1, flags, 0, 0);
717 	else if (opt_list == 2)
718 		show_page(voffset, pagemap_swap_offset(pme), flags, 0, 0);
719 
720 	nr_pages[hash_slot(flags)]++;
721 	total_pages++;
722 }
723 
724 #define PAGEMAP_BATCH	(64 << 10)
walk_vma(unsigned long index,unsigned long count)725 static void walk_vma(unsigned long index, unsigned long count)
726 {
727 	uint64_t buf[PAGEMAP_BATCH];
728 	unsigned long batch;
729 	unsigned long pages;
730 	unsigned long pfn;
731 	unsigned long i;
732 
733 	while (count) {
734 		batch = min_t(unsigned long, count, PAGEMAP_BATCH);
735 		pages = pagemap_read(buf, index, batch);
736 		if (pages == 0)
737 			break;
738 
739 		for (i = 0; i < pages; i++) {
740 			pfn = pagemap_pfn(buf[i]);
741 			if (pfn)
742 				walk_pfn(index + i, pfn, 1, buf[i]);
743 			if (buf[i] & PM_SWAP)
744 				walk_swap(index + i, buf[i]);
745 		}
746 
747 		index += pages;
748 		count -= pages;
749 	}
750 }
751 
walk_task(unsigned long index,unsigned long count)752 static void walk_task(unsigned long index, unsigned long count)
753 {
754 	const unsigned long end = index + count;
755 	unsigned long start;
756 	int i = 0;
757 
758 	while (index < end) {
759 
760 		while (pg_end[i] <= index)
761 			if (++i >= nr_vmas)
762 				return;
763 		if (pg_start[i] >= end)
764 			return;
765 
766 		start = max_t(unsigned long, pg_start[i], index);
767 		index = min_t(unsigned long, pg_end[i], end);
768 
769 		assert(start < index);
770 		walk_vma(start, index - start);
771 	}
772 }
773 
add_addr_range(unsigned long offset,unsigned long size)774 static void add_addr_range(unsigned long offset, unsigned long size)
775 {
776 	if (nr_addr_ranges >= MAX_ADDR_RANGES)
777 		fatal("too many addr ranges\n");
778 
779 	opt_offset[nr_addr_ranges] = offset;
780 	opt_size[nr_addr_ranges] = min_t(unsigned long, size, ULONG_MAX-offset);
781 	nr_addr_ranges++;
782 }
783 
walk_addr_ranges(void)784 static void walk_addr_ranges(void)
785 {
786 	int i;
787 
788 	kpageflags_fd = checked_open(opt_kpageflags, O_RDONLY);
789 
790 	if (!nr_addr_ranges)
791 		add_addr_range(0, ULONG_MAX);
792 
793 	for (i = 0; i < nr_addr_ranges; i++)
794 		if (!opt_pid)
795 			walk_pfn(opt_offset[i], opt_offset[i], opt_size[i], 0);
796 		else
797 			walk_task(opt_offset[i], opt_size[i]);
798 
799 	if (opt_mark_idle)
800 		mark_page_idle(0);
801 
802 	close(kpageflags_fd);
803 }
804 
805 
806 /*
807  * user interface
808  */
809 
page_flag_type(uint64_t flag)810 static const char *page_flag_type(uint64_t flag)
811 {
812 	if (flag & KPF_HACKERS_BITS)
813 		return "(r)";
814 	if (flag & KPF_OVERLOADED_BITS)
815 		return "(o)";
816 	return "   ";
817 }
818 
usage(void)819 static void usage(void)
820 {
821 	size_t i, j;
822 
823 	printf(
824 "page-types [options]\n"
825 "            -r|--raw                   Raw mode, for kernel developers\n"
826 "            -d|--describe flags        Describe flags\n"
827 "            -a|--addr    addr-spec     Walk a range of pages\n"
828 "            -b|--bits    bits-spec     Walk pages with specified bits\n"
829 "            -c|--cgroup  path|@inode   Walk pages within memory cgroup\n"
830 "            -p|--pid     pid           Walk process address space\n"
831 "            -f|--file    filename      Walk file address space\n"
832 "            -i|--mark-idle             Mark pages idle\n"
833 "            -l|--list                  Show page details in ranges\n"
834 "            -L|--list-each             Show page details one by one\n"
835 "            -C|--list-cgroup           Show cgroup inode for pages\n"
836 "            -M|--list-mapcnt           Show page map count\n"
837 "            -N|--no-summary            Don't show summary info\n"
838 "            -X|--hwpoison              hwpoison pages\n"
839 "            -x|--unpoison              unpoison pages\n"
840 "            -F|--kpageflags filename   kpageflags file to parse\n"
841 "            -h|--help                  Show this usage message\n"
842 "flags:\n"
843 "            0x10                       bitfield format, e.g.\n"
844 "            anon                       bit-name, e.g.\n"
845 "            0x10,anon                  comma-separated list, e.g.\n"
846 "addr-spec:\n"
847 "            N                          one page at offset N (unit: pages)\n"
848 "            N+M                        pages range from N to N+M-1\n"
849 "            N,M                        pages range from N to M-1\n"
850 "            N,                         pages range from N to end\n"
851 "            ,M                         pages range from 0 to M-1\n"
852 "bits-spec:\n"
853 "            bit1,bit2                  (flags & (bit1|bit2)) != 0\n"
854 "            bit1,bit2=bit1             (flags & (bit1|bit2)) == bit1\n"
855 "            bit1,~bit2                 (flags & (bit1|bit2)) == bit1\n"
856 "            =bit1,bit2                 flags == (bit1|bit2)\n"
857 "bit-names:\n"
858 	);
859 
860 	for (i = 0, j = 0; i < ARRAY_SIZE(page_flag_names); i++) {
861 		if (!page_flag_names[i])
862 			continue;
863 		printf("%16s%s", page_flag_names[i] + 2,
864 				 page_flag_type(1ULL << i));
865 		if (++j > 3) {
866 			j = 0;
867 			putchar('\n');
868 		}
869 	}
870 	printf("\n                                   "
871 		"(r) raw mode bits  (o) overloaded bits\n");
872 }
873 
parse_number(const char * str)874 static unsigned long long parse_number(const char *str)
875 {
876 	unsigned long long n;
877 
878 	n = strtoll(str, NULL, 0);
879 
880 	if (n == 0 && str[0] != '0')
881 		fatal("invalid name or number: %s\n", str);
882 
883 	return n;
884 }
885 
parse_pid(const char * str)886 static void parse_pid(const char *str)
887 {
888 	FILE *file;
889 	char buf[5000];
890 
891 	opt_pid = parse_number(str);
892 
893 	sprintf(buf, "/proc/%d/pagemap", opt_pid);
894 	pagemap_fd = checked_open(buf, O_RDONLY);
895 
896 	sprintf(buf, "/proc/%d/maps", opt_pid);
897 	file = fopen(buf, "r");
898 	if (!file) {
899 		perror(buf);
900 		exit(EXIT_FAILURE);
901 	}
902 
903 	while (fgets(buf, sizeof(buf), file) != NULL) {
904 		unsigned long vm_start;
905 		unsigned long vm_end;
906 		unsigned long long pgoff;
907 		int major, minor;
908 		char r, w, x, s;
909 		unsigned long ino;
910 		int n;
911 
912 		n = sscanf(buf, "%lx-%lx %c%c%c%c %llx %x:%x %lu",
913 			   &vm_start,
914 			   &vm_end,
915 			   &r, &w, &x, &s,
916 			   &pgoff,
917 			   &major, &minor,
918 			   &ino);
919 		if (n < 10) {
920 			fprintf(stderr, "unexpected line: %s\n", buf);
921 			continue;
922 		}
923 		pg_start[nr_vmas] = vm_start / page_size;
924 		pg_end[nr_vmas] = vm_end / page_size;
925 		if (++nr_vmas >= MAX_VMAS) {
926 			fprintf(stderr, "too many VMAs\n");
927 			break;
928 		}
929 	}
930 	fclose(file);
931 }
932 
show_file(const char * name,const struct stat * st)933 static void show_file(const char *name, const struct stat *st)
934 {
935 	unsigned long long size = st->st_size;
936 	char atime[64], mtime[64];
937 	long now = time(NULL);
938 
939 	printf("%s\tInode: %u\tSize: %llu (%llu pages)\n",
940 			name, (unsigned)st->st_ino,
941 			size, (size + page_size - 1) / page_size);
942 
943 	strftime(atime, sizeof(atime), "%c", localtime(&st->st_atime));
944 	strftime(mtime, sizeof(mtime), "%c", localtime(&st->st_mtime));
945 
946 	printf("Modify: %s (%ld seconds ago)\nAccess: %s (%ld seconds ago)\n",
947 			mtime, now - st->st_mtime,
948 			atime, now - st->st_atime);
949 }
950 
951 static sigjmp_buf sigbus_jmp;
952 
953 static void * volatile sigbus_addr;
954 
sigbus_handler(int sig,siginfo_t * info,void * ucontex)955 static void sigbus_handler(int sig, siginfo_t *info, void *ucontex)
956 {
957 	(void)sig;
958 	(void)ucontex;
959 	sigbus_addr = info ? info->si_addr : NULL;
960 	siglongjmp(sigbus_jmp, 1);
961 }
962 
963 static struct sigaction sigbus_action = {
964 	.sa_sigaction = sigbus_handler,
965 	.sa_flags = SA_SIGINFO,
966 };
967 
walk_file(const char * name,const struct stat * st)968 static void walk_file(const char *name, const struct stat *st)
969 {
970 	uint8_t vec[PAGEMAP_BATCH];
971 	uint64_t buf[PAGEMAP_BATCH], flags;
972 	uint64_t cgroup = 0;
973 	uint64_t mapcnt = 0;
974 	unsigned long nr_pages, pfn, i;
975 	off_t off, end = st->st_size;
976 	int fd;
977 	ssize_t len;
978 	void *ptr;
979 	int first = 1;
980 
981 	fd = checked_open(name, O_RDONLY|O_NOATIME|O_NOFOLLOW);
982 
983 	for (off = 0; off < end; off += len) {
984 		nr_pages = (end - off + page_size - 1) / page_size;
985 		if (nr_pages > PAGEMAP_BATCH)
986 			nr_pages = PAGEMAP_BATCH;
987 		len = nr_pages * page_size;
988 
989 		ptr = mmap(NULL, len, PROT_READ, MAP_SHARED, fd, off);
990 		if (ptr == MAP_FAILED)
991 			fatal("mmap failed: %s", name);
992 
993 		/* determine cached pages */
994 		if (mincore(ptr, len, vec))
995 			fatal("mincore failed: %s", name);
996 
997 		/* turn off readahead */
998 		if (madvise(ptr, len, MADV_RANDOM))
999 			fatal("madvice failed: %s", name);
1000 
1001 		if (sigsetjmp(sigbus_jmp, 1)) {
1002 			end = off + sigbus_addr ? sigbus_addr - ptr : 0;
1003 			fprintf(stderr, "got sigbus at offset %lld: %s\n",
1004 					(long long)end, name);
1005 			goto got_sigbus;
1006 		}
1007 
1008 		/* populate ptes */
1009 		for (i = 0; i < nr_pages ; i++) {
1010 			if (vec[i] & 1)
1011 				(void)*(volatile int *)(ptr + i * page_size);
1012 		}
1013 got_sigbus:
1014 
1015 		/* turn off harvesting reference bits */
1016 		if (madvise(ptr, len, MADV_SEQUENTIAL))
1017 			fatal("madvice failed: %s", name);
1018 
1019 		if (pagemap_read(buf, (unsigned long)ptr / page_size,
1020 					nr_pages) != nr_pages)
1021 			fatal("cannot read pagemap");
1022 
1023 		munmap(ptr, len);
1024 
1025 		for (i = 0; i < nr_pages; i++) {
1026 			pfn = pagemap_pfn(buf[i]);
1027 			if (!pfn)
1028 				continue;
1029 			if (!kpageflags_read(&flags, pfn, 1))
1030 				continue;
1031 			if (!kpagecgroup_read(&cgroup, pfn, 1))
1032 				fatal("kpagecgroup_read failed");
1033 			if (!kpagecount_read(&mapcnt, pfn, 1))
1034 				fatal("kpagecount_read failed");
1035 			if (first && opt_list) {
1036 				first = 0;
1037 				flush_page_range();
1038 				show_file(name, st);
1039 			}
1040 			add_page(off / page_size + i, pfn,
1041 				 flags, cgroup, mapcnt, buf[i]);
1042 		}
1043 	}
1044 
1045 	close(fd);
1046 }
1047 
walk_tree(const char * name,const struct stat * st,int type,struct FTW * f)1048 int walk_tree(const char *name, const struct stat *st, int type, struct FTW *f)
1049 {
1050 	(void)f;
1051 	switch (type) {
1052 	case FTW_F:
1053 		if (S_ISREG(st->st_mode))
1054 			walk_file(name, st);
1055 		break;
1056 	case FTW_DNR:
1057 		fprintf(stderr, "cannot read dir: %s\n", name);
1058 		break;
1059 	}
1060 	return 0;
1061 }
1062 
walk_page_cache(void)1063 static void walk_page_cache(void)
1064 {
1065 	struct stat st;
1066 
1067 	kpageflags_fd = checked_open(opt_kpageflags, O_RDONLY);
1068 	pagemap_fd = checked_open("/proc/self/pagemap", O_RDONLY);
1069 	sigaction(SIGBUS, &sigbus_action, NULL);
1070 
1071 	if (stat(opt_file, &st))
1072 		fatal("stat failed: %s\n", opt_file);
1073 
1074 	if (S_ISREG(st.st_mode)) {
1075 		walk_file(opt_file, &st);
1076 	} else if (S_ISDIR(st.st_mode)) {
1077 		/* do not follow symlinks and mountpoints */
1078 		if (nftw(opt_file, walk_tree, 64, FTW_MOUNT | FTW_PHYS) < 0)
1079 			fatal("nftw failed: %s\n", opt_file);
1080 	} else
1081 		fatal("unhandled file type: %s\n", opt_file);
1082 
1083 	close(kpageflags_fd);
1084 	close(pagemap_fd);
1085 	signal(SIGBUS, SIG_DFL);
1086 }
1087 
parse_file(const char * name)1088 static void parse_file(const char *name)
1089 {
1090 	opt_file = name;
1091 }
1092 
parse_cgroup(const char * path)1093 static void parse_cgroup(const char *path)
1094 {
1095 	if (path[0] == '@') {
1096 		opt_cgroup = parse_number(path + 1);
1097 		return;
1098 	}
1099 
1100 	struct stat st;
1101 
1102 	if (stat(path, &st))
1103 		fatal("stat failed: %s: %m\n", path);
1104 
1105 	if (!S_ISDIR(st.st_mode))
1106 		fatal("cgroup supposed to be a directory: %s\n", path);
1107 
1108 	opt_cgroup = st.st_ino;
1109 }
1110 
parse_addr_range(const char * optarg)1111 static void parse_addr_range(const char *optarg)
1112 {
1113 	unsigned long offset;
1114 	unsigned long size;
1115 	char *p;
1116 
1117 	p = strchr(optarg, ',');
1118 	if (!p)
1119 		p = strchr(optarg, '+');
1120 
1121 	if (p == optarg) {
1122 		offset = 0;
1123 		size   = parse_number(p + 1);
1124 	} else if (p) {
1125 		offset = parse_number(optarg);
1126 		if (p[1] == '\0')
1127 			size = ULONG_MAX;
1128 		else {
1129 			size = parse_number(p + 1);
1130 			if (*p == ',') {
1131 				if (size < offset)
1132 					fatal("invalid range: %lu,%lu\n",
1133 							offset, size);
1134 				size -= offset;
1135 			}
1136 		}
1137 	} else {
1138 		offset = parse_number(optarg);
1139 		size   = 1;
1140 	}
1141 
1142 	add_addr_range(offset, size);
1143 }
1144 
add_bits_filter(uint64_t mask,uint64_t bits)1145 static void add_bits_filter(uint64_t mask, uint64_t bits)
1146 {
1147 	if (nr_bit_filters >= MAX_BIT_FILTERS)
1148 		fatal("too much bit filters\n");
1149 
1150 	opt_mask[nr_bit_filters] = mask;
1151 	opt_bits[nr_bit_filters] = bits;
1152 	nr_bit_filters++;
1153 }
1154 
parse_flag_name(const char * str,int len)1155 static uint64_t parse_flag_name(const char *str, int len)
1156 {
1157 	size_t i;
1158 
1159 	if (!*str || !len)
1160 		return 0;
1161 
1162 	if (len <= 8 && !strncmp(str, "compound", len))
1163 		return BITS_COMPOUND;
1164 
1165 	for (i = 0; i < ARRAY_SIZE(page_flag_names); i++) {
1166 		if (!page_flag_names[i])
1167 			continue;
1168 		if (!strncmp(str, page_flag_names[i] + 2, len))
1169 			return 1ULL << i;
1170 	}
1171 
1172 	return parse_number(str);
1173 }
1174 
parse_flag_names(const char * str,int all)1175 static uint64_t parse_flag_names(const char *str, int all)
1176 {
1177 	const char *p    = str;
1178 	uint64_t   flags = 0;
1179 
1180 	while (1) {
1181 		if (*p == ',' || *p == '=' || *p == '\0') {
1182 			if ((*str != '~') || (*str == '~' && all && *++str))
1183 				flags |= parse_flag_name(str, p - str);
1184 			if (*p != ',')
1185 				break;
1186 			str = p + 1;
1187 		}
1188 		p++;
1189 	}
1190 
1191 	return flags;
1192 }
1193 
parse_bits_mask(const char * optarg)1194 static void parse_bits_mask(const char *optarg)
1195 {
1196 	uint64_t mask;
1197 	uint64_t bits;
1198 	const char *p;
1199 
1200 	p = strchr(optarg, '=');
1201 	if (p == optarg) {
1202 		mask = KPF_ALL_BITS;
1203 		bits = parse_flag_names(p + 1, 0);
1204 	} else if (p) {
1205 		mask = parse_flag_names(optarg, 0);
1206 		bits = parse_flag_names(p + 1, 0);
1207 	} else if (strchr(optarg, '~')) {
1208 		mask = parse_flag_names(optarg, 1);
1209 		bits = parse_flag_names(optarg, 0);
1210 	} else {
1211 		mask = parse_flag_names(optarg, 0);
1212 		bits = KPF_ALL_BITS;
1213 	}
1214 
1215 	add_bits_filter(mask, bits);
1216 }
1217 
parse_kpageflags(const char * name)1218 static void parse_kpageflags(const char *name)
1219 {
1220 	opt_kpageflags = name;
1221 }
1222 
describe_flags(const char * optarg)1223 static void describe_flags(const char *optarg)
1224 {
1225 	uint64_t flags = parse_flag_names(optarg, 0);
1226 
1227 	printf("0x%016llx\t%s\t%s\n",
1228 		(unsigned long long)flags,
1229 		page_flag_name(flags),
1230 		page_flag_longname(flags));
1231 }
1232 
1233 static const struct option opts[] = {
1234 	{ "raw"       , 0, NULL, 'r' },
1235 	{ "pid"       , 1, NULL, 'p' },
1236 	{ "file"      , 1, NULL, 'f' },
1237 	{ "addr"      , 1, NULL, 'a' },
1238 	{ "bits"      , 1, NULL, 'b' },
1239 	{ "cgroup"    , 1, NULL, 'c' },
1240 	{ "describe"  , 1, NULL, 'd' },
1241 	{ "mark-idle" , 0, NULL, 'i' },
1242 	{ "list"      , 0, NULL, 'l' },
1243 	{ "list-each" , 0, NULL, 'L' },
1244 	{ "list-cgroup", 0, NULL, 'C' },
1245 	{ "list-mapcnt", 0, NULL, 'M' },
1246 	{ "no-summary", 0, NULL, 'N' },
1247 	{ "hwpoison"  , 0, NULL, 'X' },
1248 	{ "unpoison"  , 0, NULL, 'x' },
1249 	{ "kpageflags", 0, NULL, 'F' },
1250 	{ "help"      , 0, NULL, 'h' },
1251 	{ NULL        , 0, NULL, 0 }
1252 };
1253 
main(int argc,char * argv[])1254 int main(int argc, char *argv[])
1255 {
1256 	int c;
1257 
1258 	page_size = getpagesize();
1259 
1260 	while ((c = getopt_long(argc, argv,
1261 				"rp:f:a:b:d:c:CilLMNXxF:h",
1262 				opts, NULL)) != -1) {
1263 		switch (c) {
1264 		case 'r':
1265 			opt_raw = 1;
1266 			break;
1267 		case 'p':
1268 			parse_pid(optarg);
1269 			break;
1270 		case 'f':
1271 			parse_file(optarg);
1272 			break;
1273 		case 'a':
1274 			parse_addr_range(optarg);
1275 			break;
1276 		case 'b':
1277 			parse_bits_mask(optarg);
1278 			break;
1279 		case 'c':
1280 			parse_cgroup(optarg);
1281 			break;
1282 		case 'C':
1283 			opt_list_cgroup = 1;
1284 			break;
1285 		case 'd':
1286 			describe_flags(optarg);
1287 			exit(0);
1288 		case 'i':
1289 			opt_mark_idle = 1;
1290 			break;
1291 		case 'l':
1292 			opt_list = 1;
1293 			break;
1294 		case 'L':
1295 			opt_list = 2;
1296 			break;
1297 		case 'M':
1298 			opt_list_mapcnt = 1;
1299 			break;
1300 		case 'N':
1301 			opt_no_summary = 1;
1302 			break;
1303 		case 'X':
1304 			opt_hwpoison = 1;
1305 			prepare_hwpoison_fd();
1306 			break;
1307 		case 'x':
1308 			opt_unpoison = 1;
1309 			prepare_hwpoison_fd();
1310 			break;
1311 		case 'F':
1312 			parse_kpageflags(optarg);
1313 			break;
1314 		case 'h':
1315 			usage();
1316 			exit(0);
1317 		default:
1318 			usage();
1319 			exit(1);
1320 		}
1321 	}
1322 
1323 	if (!opt_kpageflags)
1324 		opt_kpageflags = PROC_KPAGEFLAGS;
1325 
1326 	if (opt_cgroup || opt_list_cgroup)
1327 		kpagecgroup_fd = checked_open(PROC_KPAGECGROUP, O_RDONLY);
1328 
1329 	if (opt_list && opt_list_mapcnt)
1330 		kpagecount_fd = checked_open(PROC_KPAGECOUNT, O_RDONLY);
1331 
1332 	if (opt_mark_idle && opt_file)
1333 		page_idle_fd = checked_open(SYS_KERNEL_MM_PAGE_IDLE, O_RDWR);
1334 
1335 	if (opt_list && opt_pid)
1336 		printf("voffset\t");
1337 	if (opt_list && opt_file)
1338 		printf("foffset\t");
1339 	if (opt_list && opt_list_cgroup)
1340 		printf("cgroup\t");
1341 	if (opt_list && opt_list_mapcnt)
1342 		printf("map-cnt\t");
1343 
1344 	if (opt_list == 1)
1345 		printf("offset\tlen\tflags\n");
1346 	if (opt_list == 2)
1347 		printf("offset\tflags\n");
1348 
1349 	if (opt_file)
1350 		walk_page_cache();
1351 	else
1352 		walk_addr_ranges();
1353 
1354 	if (opt_list == 1)
1355 		flush_page_range();
1356 
1357 	if (opt_no_summary)
1358 		return 0;
1359 
1360 	if (opt_list)
1361 		printf("\n\n");
1362 
1363 	show_summary();
1364 
1365 	if (opt_list_mapcnt)
1366 		close(kpagecount_fd);
1367 
1368 	if (page_idle_fd >= 0)
1369 		close(page_idle_fd);
1370 
1371 	return 0;
1372 }
1373