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 <inttypes.h>
26 #include <sys/types.h>
27 #include <sys/errno.h>
28 #include <sys/fcntl.h>
29 #include <sys/mount.h>
30 #include <sys/statfs.h>
31 #include <sys/mman.h>
32 #include "../../include/uapi/linux/magic.h"
33 #include "../../include/uapi/linux/kernel-page-flags.h"
34 #include <api/fs/fs.h>
35
36 #ifndef MAX_PATH
37 # define MAX_PATH 256
38 #endif
39
40 #ifndef STR
41 # define _STR(x) #x
42 # define STR(x) _STR(x)
43 #endif
44
45 /*
46 * pagemap kernel ABI bits
47 */
48
49 #define PM_ENTRY_BYTES 8
50 #define PM_PFRAME_BITS 55
51 #define PM_PFRAME_MASK ((1LL << PM_PFRAME_BITS) - 1)
52 #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
53 #define MAX_SWAPFILES_SHIFT 5
54 #define PM_SWAP_OFFSET(x) (((x) & PM_PFRAME_MASK) >> MAX_SWAPFILES_SHIFT)
55 #define PM_SOFT_DIRTY (1ULL << 55)
56 #define PM_MMAP_EXCLUSIVE (1ULL << 56)
57 #define PM_FILE (1ULL << 61)
58 #define PM_SWAP (1ULL << 62)
59 #define PM_PRESENT (1ULL << 63)
60
61 /*
62 * kernel page flags
63 */
64
65 #define KPF_BYTES 8
66 #define PROC_KPAGEFLAGS "/proc/kpageflags"
67 #define PROC_KPAGECOUNT "/proc/kpagecount"
68 #define PROC_KPAGECGROUP "/proc/kpagecgroup"
69
70 #define SYS_KERNEL_MM_PAGE_IDLE "/sys/kernel/mm/page_idle/bitmap"
71
72 /* [32-] kernel hacking assistances */
73 #define KPF_RESERVED 32
74 #define KPF_MLOCKED 33
75 #define KPF_OWNER_2 34
76 #define KPF_PRIVATE 35
77 #define KPF_PRIVATE_2 36
78 #define KPF_OWNER_PRIVATE 37
79 #define KPF_ARCH 38
80 #define KPF_UNCACHED 39 /* unused */
81 #define KPF_SOFTDIRTY 40
82 #define KPF_ARCH_2 41
83
84 /* [47-] take some arbitrary free slots for expanding overloaded flags
85 * not part of kernel API
86 */
87 #define KPF_ANON_EXCLUSIVE 47
88 #define KPF_READAHEAD 48
89 #define KPF_SLUB_FROZEN 50
90 #define KPF_SLUB_DEBUG 51
91 #define KPF_FILE 61
92 #define KPF_SWAP 62
93 #define KPF_MMAP_EXCLUSIVE 63
94
95 #define KPF_ALL_BITS ((uint64_t)~0ULL)
96 #define KPF_HACKERS_BITS (0xffffULL << 32)
97 #define KPF_OVERLOADED_BITS (0xffffULL << 48)
98 #define BIT(name) (1ULL << KPF_##name)
99 #define BITS_COMPOUND (BIT(COMPOUND_HEAD) | BIT(COMPOUND_TAIL))
100
101 static const char * const page_flag_names[] = {
102 [KPF_LOCKED] = "L:locked",
103 [KPF_ERROR] = "E:error",
104 [KPF_REFERENCED] = "R:referenced",
105 [KPF_UPTODATE] = "U:uptodate",
106 [KPF_DIRTY] = "D:dirty",
107 [KPF_LRU] = "l:lru",
108 [KPF_ACTIVE] = "A:active",
109 [KPF_SLAB] = "S:slab",
110 [KPF_WRITEBACK] = "W:writeback",
111 [KPF_RECLAIM] = "I:reclaim",
112 [KPF_BUDDY] = "B:buddy",
113
114 [KPF_MMAP] = "M:mmap",
115 [KPF_ANON] = "a:anonymous",
116 [KPF_SWAPCACHE] = "s:swapcache",
117 [KPF_SWAPBACKED] = "b:swapbacked",
118 [KPF_COMPOUND_HEAD] = "H:compound_head",
119 [KPF_COMPOUND_TAIL] = "T:compound_tail",
120 [KPF_HUGE] = "G:huge",
121 [KPF_UNEVICTABLE] = "u:unevictable",
122 [KPF_HWPOISON] = "X:hwpoison",
123 [KPF_NOPAGE] = "n:nopage",
124 [KPF_KSM] = "x:ksm",
125 [KPF_THP] = "t:thp",
126 [KPF_OFFLINE] = "o:offline",
127 [KPF_PGTABLE] = "g:pgtable",
128 [KPF_ZERO_PAGE] = "z:zero_page",
129 [KPF_IDLE] = "i:idle_page",
130
131 [KPF_RESERVED] = "r:reserved",
132 [KPF_MLOCKED] = "m:mlocked",
133 [KPF_OWNER_2] = "d:owner_2",
134 [KPF_PRIVATE] = "P:private",
135 [KPF_PRIVATE_2] = "p:private_2",
136 [KPF_OWNER_PRIVATE] = "O:owner_private",
137 [KPF_ARCH] = "h:arch",
138 [KPF_SOFTDIRTY] = "f:softdirty",
139 [KPF_ARCH_2] = "H:arch_2",
140
141 [KPF_ANON_EXCLUSIVE] = "d:anon_exclusive",
142 [KPF_READAHEAD] = "I:readahead",
143 [KPF_SLUB_FROZEN] = "A:slub_frozen",
144 [KPF_SLUB_DEBUG] = "E:slub_debug",
145
146 [KPF_FILE] = "F:file",
147 [KPF_SWAP] = "w:swap",
148 [KPF_MMAP_EXCLUSIVE] = "1:mmap_exclusive",
149 };
150
151
152 /*
153 * data structures
154 */
155
156 static int opt_raw; /* for kernel developers */
157 static int opt_list; /* list pages (in ranges) */
158 static int opt_mark_idle; /* set accessed bit */
159 static int opt_no_summary; /* don't show summary */
160 static pid_t opt_pid; /* process to walk */
161 const char *opt_file; /* file or directory path */
162 static uint64_t opt_cgroup; /* cgroup inode */
163 static int opt_list_cgroup;/* list page cgroup */
164 static int opt_list_mapcnt;/* list page map count */
165 static const char *opt_kpageflags;/* kpageflags file to parse */
166
167 #define MAX_ADDR_RANGES 1024
168 static int nr_addr_ranges;
169 static unsigned long opt_offset[MAX_ADDR_RANGES];
170 static unsigned long opt_size[MAX_ADDR_RANGES];
171
172 #define MAX_VMAS 10240
173 static int nr_vmas;
174 static unsigned long pg_start[MAX_VMAS];
175 static unsigned long pg_end[MAX_VMAS];
176
177 #define MAX_BIT_FILTERS 64
178 static int nr_bit_filters;
179 static uint64_t opt_mask[MAX_BIT_FILTERS];
180 static uint64_t opt_bits[MAX_BIT_FILTERS];
181
182 static int page_size;
183
184 static int pagemap_fd;
185 static int kpageflags_fd;
186 static int kpagecount_fd = -1;
187 static int kpagecgroup_fd = -1;
188 static int page_idle_fd = -1;
189
190 static int opt_hwpoison;
191 static int opt_unpoison;
192
193 static const char *hwpoison_debug_fs;
194 static int hwpoison_inject_fd;
195 static int hwpoison_forget_fd;
196
197 #define HASH_SHIFT 13
198 #define HASH_SIZE (1 << HASH_SHIFT)
199 #define HASH_MASK (HASH_SIZE - 1)
200 #define HASH_KEY(flags) (flags & HASH_MASK)
201
202 static unsigned long total_pages;
203 static unsigned long nr_pages[HASH_SIZE];
204 static uint64_t page_flags[HASH_SIZE];
205
206
207 /*
208 * helper functions
209 */
210
211 #define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0]))
212
213 #define min_t(type, x, y) ({ \
214 type __min1 = (x); \
215 type __min2 = (y); \
216 __min1 < __min2 ? __min1 : __min2; })
217
218 #define max_t(type, x, y) ({ \
219 type __max1 = (x); \
220 type __max2 = (y); \
221 __max1 > __max2 ? __max1 : __max2; })
222
pages2mb(unsigned long pages)223 static unsigned long pages2mb(unsigned long pages)
224 {
225 return (pages * page_size) >> 20;
226 }
227
fatal(const char * x,...)228 static void fatal(const char *x, ...)
229 {
230 va_list ap;
231
232 va_start(ap, x);
233 vfprintf(stderr, x, ap);
234 va_end(ap);
235 exit(EXIT_FAILURE);
236 }
237
checked_open(const char * pathname,int flags)238 static int checked_open(const char *pathname, int flags)
239 {
240 int fd = open(pathname, flags);
241
242 if (fd < 0) {
243 perror(pathname);
244 exit(EXIT_FAILURE);
245 }
246
247 return fd;
248 }
249
250 /*
251 * pagemap/kpageflags routines
252 */
253
do_u64_read(int fd,const char * name,uint64_t * buf,unsigned long index,unsigned long count)254 static unsigned long do_u64_read(int fd, const char *name,
255 uint64_t *buf,
256 unsigned long index,
257 unsigned long count)
258 {
259 long bytes;
260
261 if (index > ULONG_MAX / 8)
262 fatal("index overflow: %lu\n", index);
263
264 bytes = pread(fd, buf, count * 8, (off_t)index * 8);
265 if (bytes < 0) {
266 perror(name);
267 exit(EXIT_FAILURE);
268 }
269 if (bytes % 8)
270 fatal("partial read: %lu bytes\n", bytes);
271
272 return bytes / 8;
273 }
274
kpageflags_read(uint64_t * buf,unsigned long index,unsigned long pages)275 static unsigned long kpageflags_read(uint64_t *buf,
276 unsigned long index,
277 unsigned long pages)
278 {
279 return do_u64_read(kpageflags_fd, opt_kpageflags, buf, index, pages);
280 }
281
kpagecgroup_read(uint64_t * buf,unsigned long index,unsigned long pages)282 static unsigned long kpagecgroup_read(uint64_t *buf,
283 unsigned long index,
284 unsigned long pages)
285 {
286 if (kpagecgroup_fd < 0)
287 return pages;
288
289 return do_u64_read(kpagecgroup_fd, opt_kpageflags, buf, index, pages);
290 }
291
kpagecount_read(uint64_t * buf,unsigned long index,unsigned long pages)292 static unsigned long kpagecount_read(uint64_t *buf,
293 unsigned long index,
294 unsigned long pages)
295 {
296 return kpagecount_fd < 0 ? pages :
297 do_u64_read(kpagecount_fd, PROC_KPAGECOUNT,
298 buf, index, pages);
299 }
300
pagemap_read(uint64_t * buf,unsigned long index,unsigned long pages)301 static unsigned long pagemap_read(uint64_t *buf,
302 unsigned long index,
303 unsigned long pages)
304 {
305 return do_u64_read(pagemap_fd, "/proc/pid/pagemap", buf, index, pages);
306 }
307
pagemap_pfn(uint64_t val)308 static unsigned long pagemap_pfn(uint64_t val)
309 {
310 unsigned long pfn;
311
312 if (val & PM_PRESENT)
313 pfn = PM_PFRAME(val);
314 else
315 pfn = 0;
316
317 return pfn;
318 }
319
pagemap_swap_offset(uint64_t val)320 static unsigned long pagemap_swap_offset(uint64_t val)
321 {
322 return val & PM_SWAP ? PM_SWAP_OFFSET(val) : 0;
323 }
324
325 /*
326 * page flag names
327 */
328
page_flag_name(uint64_t flags)329 static char *page_flag_name(uint64_t flags)
330 {
331 static char buf[65];
332 int present;
333 size_t i, j;
334
335 for (i = 0, j = 0; i < ARRAY_SIZE(page_flag_names); i++) {
336 present = (flags >> i) & 1;
337 if (!page_flag_names[i]) {
338 if (present)
339 fatal("unknown flag bit %d\n", i);
340 continue;
341 }
342 buf[j++] = present ? page_flag_names[i][0] : '_';
343 }
344
345 return buf;
346 }
347
page_flag_longname(uint64_t flags)348 static char *page_flag_longname(uint64_t flags)
349 {
350 static char buf[1024];
351 size_t i, n;
352
353 for (i = 0, n = 0; i < ARRAY_SIZE(page_flag_names); i++) {
354 if (!page_flag_names[i])
355 continue;
356 if ((flags >> i) & 1)
357 n += snprintf(buf + n, sizeof(buf) - n, "%s,",
358 page_flag_names[i] + 2);
359 }
360 if (n)
361 n--;
362 buf[n] = '\0';
363
364 return buf;
365 }
366
367
368 /*
369 * page list and summary
370 */
371
show_page_range(unsigned long voffset,unsigned long offset,unsigned long size,uint64_t flags,uint64_t cgroup,uint64_t mapcnt)372 static void show_page_range(unsigned long voffset, unsigned long offset,
373 unsigned long size, uint64_t flags,
374 uint64_t cgroup, uint64_t mapcnt)
375 {
376 static uint64_t flags0;
377 static uint64_t cgroup0;
378 static uint64_t mapcnt0;
379 static unsigned long voff;
380 static unsigned long index;
381 static unsigned long count;
382
383 if (flags == flags0 && cgroup == cgroup0 && mapcnt == mapcnt0 &&
384 offset == index + count && size && voffset == voff + count) {
385 count += size;
386 return;
387 }
388
389 if (count) {
390 if (opt_pid)
391 printf("%lx\t", voff);
392 if (opt_file)
393 printf("%lx\t", voff);
394 if (opt_list_cgroup)
395 printf("@%" PRIu64 "\t", cgroup0);
396 if (opt_list_mapcnt)
397 printf("%" PRIu64 "\t", mapcnt0);
398 printf("%lx\t%lx\t%s\n",
399 index, count, page_flag_name(flags0));
400 }
401
402 flags0 = flags;
403 cgroup0 = cgroup;
404 mapcnt0 = mapcnt;
405 index = offset;
406 voff = voffset;
407 count = size;
408 }
409
flush_page_range(void)410 static void flush_page_range(void)
411 {
412 show_page_range(0, 0, 0, 0, 0, 0);
413 }
414
show_page(unsigned long voffset,unsigned long offset,uint64_t flags,uint64_t cgroup,uint64_t mapcnt)415 static void show_page(unsigned long voffset, unsigned long offset,
416 uint64_t flags, uint64_t cgroup, uint64_t mapcnt)
417 {
418 if (opt_pid)
419 printf("%lx\t", voffset);
420 if (opt_file)
421 printf("%lx\t", voffset);
422 if (opt_list_cgroup)
423 printf("@%" PRIu64 "\t", cgroup)
424 if (opt_list_mapcnt)
425 printf("%" PRIu64 "\t", mapcnt);
426
427 printf("%lx\t%s\n", offset, page_flag_name(flags));
428 }
429
show_summary(void)430 static void show_summary(void)
431 {
432 size_t i;
433
434 printf(" flags\tpage-count MB"
435 " symbolic-flags\t\t\tlong-symbolic-flags\n");
436
437 for (i = 0; i < ARRAY_SIZE(nr_pages); i++) {
438 if (nr_pages[i])
439 printf("0x%016llx\t%10lu %8lu %s\t%s\n",
440 (unsigned long long)page_flags[i],
441 nr_pages[i],
442 pages2mb(nr_pages[i]),
443 page_flag_name(page_flags[i]),
444 page_flag_longname(page_flags[i]));
445 }
446
447 printf(" total\t%10lu %8lu\n",
448 total_pages, pages2mb(total_pages));
449 }
450
451
452 /*
453 * page flag filters
454 */
455
bit_mask_ok(uint64_t flags)456 static int bit_mask_ok(uint64_t flags)
457 {
458 int i;
459
460 for (i = 0; i < nr_bit_filters; i++) {
461 if (opt_bits[i] == KPF_ALL_BITS) {
462 if ((flags & opt_mask[i]) == 0)
463 return 0;
464 } else {
465 if ((flags & opt_mask[i]) != opt_bits[i])
466 return 0;
467 }
468 }
469
470 return 1;
471 }
472
expand_overloaded_flags(uint64_t flags,uint64_t pme)473 static uint64_t expand_overloaded_flags(uint64_t flags, uint64_t pme)
474 {
475 /* Anonymous pages use PG_owner_2 for anon_exclusive */
476 if ((flags & BIT(ANON)) && (flags & BIT(OWNER_2)))
477 flags ^= BIT(OWNER_2) | BIT(ANON_EXCLUSIVE);
478
479 /* SLUB overloads several page flags */
480 if (flags & BIT(SLAB)) {
481 if (flags & BIT(ACTIVE))
482 flags ^= BIT(ACTIVE) | BIT(SLUB_FROZEN);
483 if (flags & BIT(ERROR))
484 flags ^= BIT(ERROR) | BIT(SLUB_DEBUG);
485 }
486
487 /* PG_reclaim is overloaded as PG_readahead in the read path */
488 if ((flags & (BIT(RECLAIM) | BIT(WRITEBACK))) == BIT(RECLAIM))
489 flags ^= BIT(RECLAIM) | BIT(READAHEAD);
490
491 if (pme & PM_SOFT_DIRTY)
492 flags |= BIT(SOFTDIRTY);
493 if (pme & PM_FILE)
494 flags |= BIT(FILE);
495 if (pme & PM_SWAP)
496 flags |= BIT(SWAP);
497 if (pme & PM_MMAP_EXCLUSIVE)
498 flags |= BIT(MMAP_EXCLUSIVE);
499
500 return flags;
501 }
502
well_known_flags(uint64_t flags)503 static uint64_t well_known_flags(uint64_t flags)
504 {
505 /* hide flags intended only for kernel hacker */
506 flags &= ~KPF_HACKERS_BITS;
507
508 /* hide non-hugeTLB compound pages */
509 if ((flags & BITS_COMPOUND) && !(flags & BIT(HUGE)))
510 flags &= ~BITS_COMPOUND;
511
512 return flags;
513 }
514
kpageflags_flags(uint64_t flags,uint64_t pme)515 static uint64_t kpageflags_flags(uint64_t flags, uint64_t pme)
516 {
517 if (opt_raw)
518 flags = expand_overloaded_flags(flags, pme);
519 else
520 flags = well_known_flags(flags);
521
522 return flags;
523 }
524
525 /*
526 * page actions
527 */
528
prepare_hwpoison_fd(void)529 static void prepare_hwpoison_fd(void)
530 {
531 char buf[MAX_PATH + 1];
532
533 hwpoison_debug_fs = debugfs__mount();
534 if (!hwpoison_debug_fs) {
535 perror("mount debugfs");
536 exit(EXIT_FAILURE);
537 }
538
539 if (opt_hwpoison && !hwpoison_inject_fd) {
540 snprintf(buf, MAX_PATH, "%s/hwpoison/corrupt-pfn",
541 hwpoison_debug_fs);
542 hwpoison_inject_fd = checked_open(buf, O_WRONLY);
543 }
544
545 if (opt_unpoison && !hwpoison_forget_fd) {
546 snprintf(buf, MAX_PATH, "%s/hwpoison/unpoison-pfn",
547 hwpoison_debug_fs);
548 hwpoison_forget_fd = checked_open(buf, O_WRONLY);
549 }
550 }
551
hwpoison_page(unsigned long offset)552 static int hwpoison_page(unsigned long offset)
553 {
554 char buf[100];
555 int len;
556
557 len = sprintf(buf, "0x%lx\n", offset);
558 len = write(hwpoison_inject_fd, buf, len);
559 if (len < 0) {
560 perror("hwpoison inject");
561 return len;
562 }
563 return 0;
564 }
565
unpoison_page(unsigned long offset)566 static int unpoison_page(unsigned long offset)
567 {
568 char buf[100];
569 int len;
570
571 len = sprintf(buf, "0x%lx\n", offset);
572 len = write(hwpoison_forget_fd, buf, len);
573 if (len < 0) {
574 perror("hwpoison forget");
575 return len;
576 }
577 return 0;
578 }
579
mark_page_idle(unsigned long offset)580 static int mark_page_idle(unsigned long offset)
581 {
582 static unsigned long off;
583 static uint64_t buf;
584 int len;
585
586 if ((offset / 64 == off / 64) || buf == 0) {
587 buf |= 1UL << (offset % 64);
588 off = offset;
589 return 0;
590 }
591
592 len = pwrite(page_idle_fd, &buf, 8, 8 * (off / 64));
593 if (len < 0) {
594 perror("mark page idle");
595 return len;
596 }
597
598 buf = 1UL << (offset % 64);
599 off = offset;
600
601 return 0;
602 }
603
604 /*
605 * page frame walker
606 */
607
hash_slot(uint64_t flags)608 static size_t hash_slot(uint64_t flags)
609 {
610 size_t k = HASH_KEY(flags);
611 size_t i;
612
613 /* Explicitly reserve slot 0 for flags 0: the following logic
614 * cannot distinguish an unoccupied slot from slot (flags==0).
615 */
616 if (flags == 0)
617 return 0;
618
619 /* search through the remaining (HASH_SIZE-1) slots */
620 for (i = 1; i < ARRAY_SIZE(page_flags); i++, k++) {
621 if (!k || k >= ARRAY_SIZE(page_flags))
622 k = 1;
623 if (page_flags[k] == 0) {
624 page_flags[k] = flags;
625 return k;
626 }
627 if (page_flags[k] == flags)
628 return k;
629 }
630
631 fatal("hash table full: bump up HASH_SHIFT?\n");
632 exit(EXIT_FAILURE);
633 }
634
add_page(unsigned long voffset,unsigned long offset,uint64_t flags,uint64_t cgroup,uint64_t mapcnt,uint64_t pme)635 static void add_page(unsigned long voffset, unsigned long offset,
636 uint64_t flags, uint64_t cgroup, uint64_t mapcnt,
637 uint64_t pme)
638 {
639 flags = kpageflags_flags(flags, pme);
640
641 if (!bit_mask_ok(flags))
642 return;
643
644 if (opt_cgroup && cgroup != (uint64_t)opt_cgroup)
645 return;
646
647 if (opt_hwpoison)
648 hwpoison_page(offset);
649 if (opt_unpoison)
650 unpoison_page(offset);
651
652 if (opt_mark_idle)
653 mark_page_idle(offset);
654
655 if (opt_list == 1)
656 show_page_range(voffset, offset, 1, flags, cgroup, mapcnt);
657 else if (opt_list == 2)
658 show_page(voffset, offset, flags, cgroup, mapcnt);
659
660 nr_pages[hash_slot(flags)]++;
661 total_pages++;
662 }
663
664 #define KPAGEFLAGS_BATCH (64 << 10) /* 64k pages */
walk_pfn(unsigned long voffset,unsigned long index,unsigned long count,uint64_t pme)665 static void walk_pfn(unsigned long voffset,
666 unsigned long index,
667 unsigned long count,
668 uint64_t pme)
669 {
670 uint64_t buf[KPAGEFLAGS_BATCH];
671 uint64_t cgi[KPAGEFLAGS_BATCH];
672 uint64_t cnt[KPAGEFLAGS_BATCH];
673 unsigned long batch;
674 unsigned long pages;
675 unsigned long i;
676
677 /*
678 * kpagecgroup_read() reads only if kpagecgroup were opened, but
679 * /proc/kpagecgroup might even not exist, so it's better to fill
680 * them with zeros here.
681 */
682 if (count == 1)
683 cgi[0] = 0;
684 else
685 memset(cgi, 0, sizeof cgi);
686
687 while (count) {
688 batch = min_t(unsigned long, count, KPAGEFLAGS_BATCH);
689 pages = kpageflags_read(buf, index, batch);
690 if (pages == 0)
691 break;
692
693 if (kpagecgroup_read(cgi, index, pages) != pages)
694 fatal("kpagecgroup returned fewer pages than expected");
695
696 if (kpagecount_read(cnt, index, pages) != pages)
697 fatal("kpagecount returned fewer pages than expected");
698
699 for (i = 0; i < pages; i++)
700 add_page(voffset + i, index + i,
701 buf[i], cgi[i], cnt[i], pme);
702
703 index += pages;
704 count -= pages;
705 }
706 }
707
walk_swap(unsigned long voffset,uint64_t pme)708 static void walk_swap(unsigned long voffset, uint64_t pme)
709 {
710 uint64_t flags = kpageflags_flags(0, pme);
711
712 if (!bit_mask_ok(flags))
713 return;
714
715 if (opt_cgroup)
716 return;
717
718 if (opt_list == 1)
719 show_page_range(voffset, pagemap_swap_offset(pme),
720 1, flags, 0, 0);
721 else if (opt_list == 2)
722 show_page(voffset, pagemap_swap_offset(pme), flags, 0, 0);
723
724 nr_pages[hash_slot(flags)]++;
725 total_pages++;
726 }
727
728 #define PAGEMAP_BATCH (64 << 10)
walk_vma(unsigned long index,unsigned long count)729 static void walk_vma(unsigned long index, unsigned long count)
730 {
731 uint64_t buf[PAGEMAP_BATCH];
732 unsigned long batch;
733 unsigned long pages;
734 unsigned long pfn;
735 unsigned long i;
736
737 while (count) {
738 batch = min_t(unsigned long, count, PAGEMAP_BATCH);
739 pages = pagemap_read(buf, index, batch);
740 if (pages == 0)
741 break;
742
743 for (i = 0; i < pages; i++) {
744 pfn = pagemap_pfn(buf[i]);
745 if (pfn)
746 walk_pfn(index + i, pfn, 1, buf[i]);
747 if (buf[i] & PM_SWAP)
748 walk_swap(index + i, buf[i]);
749 }
750
751 index += pages;
752 count -= pages;
753 }
754 }
755
walk_task(unsigned long index,unsigned long count)756 static void walk_task(unsigned long index, unsigned long count)
757 {
758 const unsigned long end = index + count;
759 unsigned long start;
760 int i = 0;
761
762 while (index < end) {
763
764 while (pg_end[i] <= index)
765 if (++i >= nr_vmas)
766 return;
767 if (pg_start[i] >= end)
768 return;
769
770 start = max_t(unsigned long, pg_start[i], index);
771 index = min_t(unsigned long, pg_end[i], end);
772
773 assert(start < index);
774 walk_vma(start, index - start);
775 }
776 }
777
add_addr_range(unsigned long offset,unsigned long size)778 static void add_addr_range(unsigned long offset, unsigned long size)
779 {
780 if (nr_addr_ranges >= MAX_ADDR_RANGES)
781 fatal("too many addr ranges\n");
782
783 opt_offset[nr_addr_ranges] = offset;
784 opt_size[nr_addr_ranges] = min_t(unsigned long, size, ULONG_MAX-offset);
785 nr_addr_ranges++;
786 }
787
walk_addr_ranges(void)788 static void walk_addr_ranges(void)
789 {
790 int i;
791
792 kpageflags_fd = checked_open(opt_kpageflags, O_RDONLY);
793
794 if (!nr_addr_ranges)
795 add_addr_range(0, ULONG_MAX);
796
797 for (i = 0; i < nr_addr_ranges; i++)
798 if (!opt_pid)
799 walk_pfn(opt_offset[i], opt_offset[i], opt_size[i], 0);
800 else
801 walk_task(opt_offset[i], opt_size[i]);
802
803 if (opt_mark_idle)
804 mark_page_idle(0);
805
806 close(kpageflags_fd);
807 }
808
809
810 /*
811 * user interface
812 */
813
page_flag_type(uint64_t flag)814 static const char *page_flag_type(uint64_t flag)
815 {
816 if (flag & KPF_HACKERS_BITS)
817 return "(r)";
818 if (flag & KPF_OVERLOADED_BITS)
819 return "(o)";
820 return " ";
821 }
822
usage(void)823 static void usage(void)
824 {
825 size_t i, j;
826
827 printf(
828 "page-types [options]\n"
829 " -r|--raw Raw mode, for kernel developers\n"
830 " -d|--describe flags Describe flags\n"
831 " -a|--addr addr-spec Walk a range of pages\n"
832 " -b|--bits bits-spec Walk pages with specified bits\n"
833 " -c|--cgroup path|@inode Walk pages within memory cgroup\n"
834 " -p|--pid pid Walk process address space\n"
835 " -f|--file filename Walk file address space\n"
836 " -i|--mark-idle Mark pages idle\n"
837 " -l|--list Show page details in ranges\n"
838 " -L|--list-each Show page details one by one\n"
839 " -C|--list-cgroup Show cgroup inode for pages\n"
840 " -M|--list-mapcnt Show page map count\n"
841 " -N|--no-summary Don't show summary info\n"
842 " -X|--hwpoison hwpoison pages\n"
843 " -x|--unpoison unpoison pages\n"
844 " -F|--kpageflags filename kpageflags file to parse\n"
845 " -h|--help Show this usage message\n"
846 "flags:\n"
847 " 0x10 bitfield format, e.g.\n"
848 " anon bit-name, e.g.\n"
849 " 0x10,anon comma-separated list, e.g.\n"
850 "addr-spec:\n"
851 " N one page at offset N (unit: pages)\n"
852 " N+M pages range from N to N+M-1\n"
853 " N,M pages range from N to M-1\n"
854 " N, pages range from N to end\n"
855 " ,M pages range from 0 to M-1\n"
856 "bits-spec:\n"
857 " bit1,bit2 (flags & (bit1|bit2)) != 0\n"
858 " bit1,bit2=bit1 (flags & (bit1|bit2)) == bit1\n"
859 " bit1,~bit2 (flags & (bit1|bit2)) == bit1\n"
860 " =bit1,bit2 flags == (bit1|bit2)\n"
861 "bit-names:\n"
862 );
863
864 for (i = 0, j = 0; i < ARRAY_SIZE(page_flag_names); i++) {
865 if (!page_flag_names[i])
866 continue;
867 printf("%16s%s", page_flag_names[i] + 2,
868 page_flag_type(1ULL << i));
869 if (++j > 3) {
870 j = 0;
871 putchar('\n');
872 }
873 }
874 printf("\n "
875 "(r) raw mode bits (o) overloaded bits\n");
876 }
877
parse_number(const char * str)878 static unsigned long long parse_number(const char *str)
879 {
880 unsigned long long n;
881
882 n = strtoll(str, NULL, 0);
883
884 if (n == 0 && str[0] != '0')
885 fatal("invalid name or number: %s\n", str);
886
887 return n;
888 }
889
parse_pid(const char * str)890 static void parse_pid(const char *str)
891 {
892 FILE *file;
893 char buf[5000];
894
895 opt_pid = parse_number(str);
896
897 sprintf(buf, "/proc/%d/pagemap", opt_pid);
898 pagemap_fd = checked_open(buf, O_RDONLY);
899
900 sprintf(buf, "/proc/%d/maps", opt_pid);
901 file = fopen(buf, "r");
902 if (!file) {
903 perror(buf);
904 exit(EXIT_FAILURE);
905 }
906
907 while (fgets(buf, sizeof(buf), file) != NULL) {
908 unsigned long vm_start;
909 unsigned long vm_end;
910 unsigned long long pgoff;
911 int major, minor;
912 char r, w, x, s;
913 unsigned long ino;
914 int n;
915
916 n = sscanf(buf, "%lx-%lx %c%c%c%c %llx %x:%x %lu",
917 &vm_start,
918 &vm_end,
919 &r, &w, &x, &s,
920 &pgoff,
921 &major, &minor,
922 &ino);
923 if (n < 10) {
924 fprintf(stderr, "unexpected line: %s\n", buf);
925 continue;
926 }
927 pg_start[nr_vmas] = vm_start / page_size;
928 pg_end[nr_vmas] = vm_end / page_size;
929 if (++nr_vmas >= MAX_VMAS) {
930 fprintf(stderr, "too many VMAs\n");
931 break;
932 }
933 }
934 fclose(file);
935 }
936
show_file(const char * name,const struct stat * st)937 static void show_file(const char *name, const struct stat *st)
938 {
939 unsigned long long size = st->st_size;
940 char atime[64], mtime[64];
941 long now = time(NULL);
942
943 printf("%s\tInode: %u\tSize: %llu (%llu pages)\n",
944 name, (unsigned)st->st_ino,
945 size, (size + page_size - 1) / page_size);
946
947 strftime(atime, sizeof(atime), "%c", localtime(&st->st_atime));
948 strftime(mtime, sizeof(mtime), "%c", localtime(&st->st_mtime));
949
950 printf("Modify: %s (%ld seconds ago)\nAccess: %s (%ld seconds ago)\n",
951 mtime, now - st->st_mtime,
952 atime, now - st->st_atime);
953 }
954
955 static sigjmp_buf sigbus_jmp;
956
957 static void * volatile sigbus_addr;
958
sigbus_handler(int sig,siginfo_t * info,void * ucontex)959 static void sigbus_handler(int sig, siginfo_t *info, void *ucontex)
960 {
961 (void)sig;
962 (void)ucontex;
963 sigbus_addr = info ? info->si_addr : NULL;
964 siglongjmp(sigbus_jmp, 1);
965 }
966
967 static struct sigaction sigbus_action = {
968 .sa_sigaction = sigbus_handler,
969 .sa_flags = SA_SIGINFO,
970 };
971
walk_file_range(const char * name,int fd,unsigned long off,unsigned long end)972 static void walk_file_range(const char *name, int fd,
973 unsigned long off, unsigned long end)
974 {
975 uint8_t vec[PAGEMAP_BATCH];
976 uint64_t buf[PAGEMAP_BATCH], flags;
977 uint64_t cgroup = 0;
978 uint64_t mapcnt = 0;
979 unsigned long nr_pages, pfn, i;
980 ssize_t len;
981 void *ptr;
982 int first = 1;
983
984 for (; off < end; off += len) {
985 nr_pages = (end - off + page_size - 1) / page_size;
986 if (nr_pages > PAGEMAP_BATCH)
987 nr_pages = PAGEMAP_BATCH;
988 len = nr_pages * page_size;
989
990 ptr = mmap(NULL, len, PROT_READ, MAP_SHARED, fd, off);
991 if (ptr == MAP_FAILED)
992 fatal("mmap failed: %s", name);
993
994 /* determine cached pages */
995 if (mincore(ptr, len, vec))
996 fatal("mincore failed: %s", name);
997
998 /* turn off readahead */
999 if (madvise(ptr, len, MADV_RANDOM))
1000 fatal("madvice failed: %s", name);
1001
1002 if (sigsetjmp(sigbus_jmp, 1)) {
1003 end = off + sigbus_addr ? sigbus_addr - ptr : 0;
1004 fprintf(stderr, "got sigbus at offset %lld: %s\n",
1005 (long long)end, name);
1006 goto got_sigbus;
1007 }
1008
1009 /* populate ptes */
1010 for (i = 0; i < nr_pages ; i++) {
1011 if (vec[i] & 1)
1012 (void)*(volatile int *)(ptr + i * page_size);
1013 }
1014 got_sigbus:
1015
1016 /* turn off harvesting reference bits */
1017 if (madvise(ptr, len, MADV_SEQUENTIAL))
1018 fatal("madvice failed: %s", name);
1019
1020 if (pagemap_read(buf, (unsigned long)ptr / page_size,
1021 nr_pages) != nr_pages)
1022 fatal("cannot read pagemap");
1023
1024 munmap(ptr, len);
1025
1026 for (i = 0; i < nr_pages; i++) {
1027 pfn = pagemap_pfn(buf[i]);
1028 if (!pfn)
1029 continue;
1030 if (!kpageflags_read(&flags, pfn, 1))
1031 continue;
1032 if (!kpagecgroup_read(&cgroup, pfn, 1))
1033 fatal("kpagecgroup_read failed");
1034 if (!kpagecount_read(&mapcnt, pfn, 1))
1035 fatal("kpagecount_read failed");
1036 if (first && opt_list) {
1037 first = 0;
1038 flush_page_range();
1039 }
1040 add_page(off / page_size + i, pfn,
1041 flags, cgroup, mapcnt, buf[i]);
1042 }
1043 }
1044 }
1045
walk_file(const char * name,const struct stat * st)1046 static void walk_file(const char *name, const struct stat *st)
1047 {
1048 int i;
1049 int fd;
1050
1051 fd = checked_open(name, O_RDONLY|O_NOATIME|O_NOFOLLOW);
1052
1053 if (!nr_addr_ranges)
1054 add_addr_range(0, st->st_size / page_size);
1055
1056 for (i = 0; i < nr_addr_ranges; i++)
1057 walk_file_range(name, fd, opt_offset[i] * page_size,
1058 (opt_offset[i] + opt_size[i]) * page_size);
1059
1060 close(fd);
1061 }
1062
walk_tree(const char * name,const struct stat * st,int type,struct FTW * f)1063 int walk_tree(const char *name, const struct stat *st, int type, struct FTW *f)
1064 {
1065 (void)f;
1066 switch (type) {
1067 case FTW_F:
1068 if (S_ISREG(st->st_mode))
1069 walk_file(name, st);
1070 break;
1071 case FTW_DNR:
1072 fprintf(stderr, "cannot read dir: %s\n", name);
1073 break;
1074 }
1075 return 0;
1076 }
1077
1078 struct stat st;
1079
walk_page_cache(void)1080 static void walk_page_cache(void)
1081 {
1082 kpageflags_fd = checked_open(opt_kpageflags, O_RDONLY);
1083 pagemap_fd = checked_open("/proc/self/pagemap", O_RDONLY);
1084 sigaction(SIGBUS, &sigbus_action, NULL);
1085
1086 if (stat(opt_file, &st))
1087 fatal("stat failed: %s\n", opt_file);
1088
1089 if (S_ISREG(st.st_mode)) {
1090 walk_file(opt_file, &st);
1091 } else if (S_ISDIR(st.st_mode)) {
1092 /* do not follow symlinks and mountpoints */
1093 if (nftw(opt_file, walk_tree, 64, FTW_MOUNT | FTW_PHYS) < 0)
1094 fatal("nftw failed: %s\n", opt_file);
1095 } else
1096 fatal("unhandled file type: %s\n", opt_file);
1097
1098 close(kpageflags_fd);
1099 close(pagemap_fd);
1100 signal(SIGBUS, SIG_DFL);
1101 }
1102
parse_file(const char * name)1103 static void parse_file(const char *name)
1104 {
1105 opt_file = name;
1106 }
1107
parse_cgroup(const char * path)1108 static void parse_cgroup(const char *path)
1109 {
1110 if (path[0] == '@') {
1111 opt_cgroup = parse_number(path + 1);
1112 return;
1113 }
1114
1115 struct stat st;
1116
1117 if (stat(path, &st))
1118 fatal("stat failed: %s: %m\n", path);
1119
1120 if (!S_ISDIR(st.st_mode))
1121 fatal("cgroup supposed to be a directory: %s\n", path);
1122
1123 opt_cgroup = st.st_ino;
1124 }
1125
parse_addr_range(const char * optarg)1126 static void parse_addr_range(const char *optarg)
1127 {
1128 unsigned long offset;
1129 unsigned long size;
1130 char *p;
1131
1132 p = strchr(optarg, ',');
1133 if (!p)
1134 p = strchr(optarg, '+');
1135
1136 if (p == optarg) {
1137 offset = 0;
1138 size = parse_number(p + 1);
1139 } else if (p) {
1140 offset = parse_number(optarg);
1141 if (p[1] == '\0')
1142 size = ULONG_MAX;
1143 else {
1144 size = parse_number(p + 1);
1145 if (*p == ',') {
1146 if (size < offset)
1147 fatal("invalid range: %lu,%lu\n",
1148 offset, size);
1149 size -= offset;
1150 }
1151 }
1152 } else {
1153 offset = parse_number(optarg);
1154 size = 1;
1155 }
1156
1157 add_addr_range(offset, size);
1158 }
1159
add_bits_filter(uint64_t mask,uint64_t bits)1160 static void add_bits_filter(uint64_t mask, uint64_t bits)
1161 {
1162 if (nr_bit_filters >= MAX_BIT_FILTERS)
1163 fatal("too much bit filters\n");
1164
1165 opt_mask[nr_bit_filters] = mask;
1166 opt_bits[nr_bit_filters] = bits;
1167 nr_bit_filters++;
1168 }
1169
parse_flag_name(const char * str,int len)1170 static uint64_t parse_flag_name(const char *str, int len)
1171 {
1172 size_t i;
1173
1174 if (!*str || !len)
1175 return 0;
1176
1177 if (len <= 8 && !strncmp(str, "compound", len))
1178 return BITS_COMPOUND;
1179
1180 for (i = 0; i < ARRAY_SIZE(page_flag_names); i++) {
1181 if (!page_flag_names[i])
1182 continue;
1183 if (!strncmp(str, page_flag_names[i] + 2, len))
1184 return 1ULL << i;
1185 }
1186
1187 return parse_number(str);
1188 }
1189
parse_flag_names(const char * str,int all)1190 static uint64_t parse_flag_names(const char *str, int all)
1191 {
1192 const char *p = str;
1193 uint64_t flags = 0;
1194
1195 while (1) {
1196 if (*p == ',' || *p == '=' || *p == '\0') {
1197 if ((*str != '~') || (*str == '~' && all && *++str))
1198 flags |= parse_flag_name(str, p - str);
1199 if (*p != ',')
1200 break;
1201 str = p + 1;
1202 }
1203 p++;
1204 }
1205
1206 return flags;
1207 }
1208
parse_bits_mask(const char * optarg)1209 static void parse_bits_mask(const char *optarg)
1210 {
1211 uint64_t mask;
1212 uint64_t bits;
1213 const char *p;
1214
1215 p = strchr(optarg, '=');
1216 if (p == optarg) {
1217 mask = KPF_ALL_BITS;
1218 bits = parse_flag_names(p + 1, 0);
1219 } else if (p) {
1220 mask = parse_flag_names(optarg, 0);
1221 bits = parse_flag_names(p + 1, 0);
1222 } else if (strchr(optarg, '~')) {
1223 mask = parse_flag_names(optarg, 1);
1224 bits = parse_flag_names(optarg, 0);
1225 } else {
1226 mask = parse_flag_names(optarg, 0);
1227 bits = KPF_ALL_BITS;
1228 }
1229
1230 add_bits_filter(mask, bits);
1231 }
1232
parse_kpageflags(const char * name)1233 static void parse_kpageflags(const char *name)
1234 {
1235 opt_kpageflags = name;
1236 }
1237
describe_flags(const char * optarg)1238 static void describe_flags(const char *optarg)
1239 {
1240 uint64_t flags = parse_flag_names(optarg, 0);
1241
1242 printf("0x%016llx\t%s\t%s\n",
1243 (unsigned long long)flags,
1244 page_flag_name(flags),
1245 page_flag_longname(flags));
1246 }
1247
1248 static const struct option opts[] = {
1249 { "raw" , 0, NULL, 'r' },
1250 { "pid" , 1, NULL, 'p' },
1251 { "file" , 1, NULL, 'f' },
1252 { "addr" , 1, NULL, 'a' },
1253 { "bits" , 1, NULL, 'b' },
1254 { "cgroup" , 1, NULL, 'c' },
1255 { "describe" , 1, NULL, 'd' },
1256 { "mark-idle" , 0, NULL, 'i' },
1257 { "list" , 0, NULL, 'l' },
1258 { "list-each" , 0, NULL, 'L' },
1259 { "list-cgroup", 0, NULL, 'C' },
1260 { "list-mapcnt", 0, NULL, 'M' },
1261 { "no-summary", 0, NULL, 'N' },
1262 { "hwpoison" , 0, NULL, 'X' },
1263 { "unpoison" , 0, NULL, 'x' },
1264 { "kpageflags", 0, NULL, 'F' },
1265 { "help" , 0, NULL, 'h' },
1266 { NULL , 0, NULL, 0 }
1267 };
1268
main(int argc,char * argv[])1269 int main(int argc, char *argv[])
1270 {
1271 int c;
1272
1273 page_size = getpagesize();
1274
1275 while ((c = getopt_long(argc, argv,
1276 "rp:f:a:b:d:c:CilLMNXxF:h",
1277 opts, NULL)) != -1) {
1278 switch (c) {
1279 case 'r':
1280 opt_raw = 1;
1281 break;
1282 case 'p':
1283 parse_pid(optarg);
1284 break;
1285 case 'f':
1286 parse_file(optarg);
1287 break;
1288 case 'a':
1289 parse_addr_range(optarg);
1290 break;
1291 case 'b':
1292 parse_bits_mask(optarg);
1293 break;
1294 case 'c':
1295 parse_cgroup(optarg);
1296 break;
1297 case 'C':
1298 opt_list_cgroup = 1;
1299 break;
1300 case 'd':
1301 describe_flags(optarg);
1302 exit(0);
1303 case 'i':
1304 opt_mark_idle = 1;
1305 break;
1306 case 'l':
1307 opt_list = 1;
1308 break;
1309 case 'L':
1310 opt_list = 2;
1311 break;
1312 case 'M':
1313 opt_list_mapcnt = 1;
1314 break;
1315 case 'N':
1316 opt_no_summary = 1;
1317 break;
1318 case 'X':
1319 opt_hwpoison = 1;
1320 prepare_hwpoison_fd();
1321 break;
1322 case 'x':
1323 opt_unpoison = 1;
1324 prepare_hwpoison_fd();
1325 break;
1326 case 'F':
1327 parse_kpageflags(optarg);
1328 break;
1329 case 'h':
1330 usage();
1331 exit(0);
1332 default:
1333 usage();
1334 exit(1);
1335 }
1336 }
1337
1338 if (!opt_kpageflags)
1339 opt_kpageflags = PROC_KPAGEFLAGS;
1340
1341 if (opt_cgroup || opt_list_cgroup)
1342 kpagecgroup_fd = checked_open(PROC_KPAGECGROUP, O_RDONLY);
1343
1344 if (opt_list && opt_list_mapcnt)
1345 kpagecount_fd = checked_open(PROC_KPAGECOUNT, O_RDONLY);
1346
1347 if (opt_mark_idle)
1348 page_idle_fd = checked_open(SYS_KERNEL_MM_PAGE_IDLE, O_RDWR);
1349
1350 if (opt_list && opt_pid)
1351 printf("voffset\t");
1352 if (opt_list && opt_file)
1353 printf("foffset\t");
1354 if (opt_list && opt_list_cgroup)
1355 printf("cgroup\t");
1356 if (opt_list && opt_list_mapcnt)
1357 printf("map-cnt\t");
1358
1359 if (opt_list == 1)
1360 printf("offset\tlen\tflags\n");
1361 if (opt_list == 2)
1362 printf("offset\tflags\n");
1363
1364 if (opt_file)
1365 walk_page_cache();
1366 else
1367 walk_addr_ranges();
1368
1369 if (opt_list == 1)
1370 flush_page_range();
1371
1372 if (opt_no_summary)
1373 return 0;
1374
1375 if (opt_list)
1376 printf("\n\n");
1377
1378 if (opt_file) {
1379 show_file(opt_file, &st);
1380 printf("\n");
1381 }
1382
1383 show_summary();
1384
1385 if (opt_list_mapcnt)
1386 close(kpagecount_fd);
1387
1388 if (page_idle_fd >= 0)
1389 close(page_idle_fd);
1390
1391 return 0;
1392 }
1393