aboutsummaryrefslogtreecommitdiffstats
path: root/Documentation/vm
diff options
context:
space:
mode:
Diffstat (limited to 'Documentation/vm')
-rw-r--r--Documentation/vm/Makefile8
-rw-r--r--Documentation/vm/hugepage-mmap.c91
-rw-r--r--Documentation/vm/hugepage-shm.c98
-rw-r--r--Documentation/vm/map_hugetlb.c77
-rw-r--r--Documentation/vm/page-types.c1100
5 files changed, 1374 insertions, 0 deletions
diff --git a/Documentation/vm/Makefile b/Documentation/vm/Makefile
new file mode 100644
index 00000000000..3fa4d066886
--- /dev/null
+++ b/Documentation/vm/Makefile
@@ -0,0 +1,8 @@
1# kbuild trick to avoid linker error. Can be omitted if a module is built.
2obj- := dummy.o
3
4# List of programs to build
5hostprogs-y := page-types hugepage-mmap hugepage-shm map_hugetlb
6
7# Tell kbuild to always build the programs
8always := $(hostprogs-y)
diff --git a/Documentation/vm/hugepage-mmap.c b/Documentation/vm/hugepage-mmap.c
new file mode 100644
index 00000000000..db0dd9a33d5
--- /dev/null
+++ b/Documentation/vm/hugepage-mmap.c
@@ -0,0 +1,91 @@
1/*
2 * hugepage-mmap:
3 *
4 * Example of using huge page memory in a user application using the mmap
5 * system call. Before running this application, make sure that the
6 * administrator has mounted the hugetlbfs filesystem (on some directory
7 * like /mnt) using the command mount -t hugetlbfs nodev /mnt. In this
8 * example, the app is requesting memory of size 256MB that is backed by
9 * huge pages.
10 *
11 * For the ia64 architecture, the Linux kernel reserves Region number 4 for
12 * huge pages. That means that if one requires a fixed address, a huge page
13 * aligned address starting with 0x800000... will be required. If a fixed
14 * address is not required, the kernel will select an address in the proper
15 * range.
16 * Other architectures, such as ppc64, i386 or x86_64 are not so constrained.
17 */
18
19#include <stdlib.h>
20#include <stdio.h>
21#include <unistd.h>
22#include <sys/mman.h>
23#include <fcntl.h>
24
25#define FILE_NAME "/mnt/hugepagefile"
26#define LENGTH (256UL*1024*1024)
27#define PROTECTION (PROT_READ | PROT_WRITE)
28
29/* Only ia64 requires this */
30#ifdef __ia64__
31#define ADDR (void *)(0x8000000000000000UL)
32#define FLAGS (MAP_SHARED | MAP_FIXED)
33#else
34#define ADDR (void *)(0x0UL)
35#define FLAGS (MAP_SHARED)
36#endif
37
38static void check_bytes(char *addr)
39{
40 printf("First hex is %x\n", *((unsigned int *)addr));
41}
42
43static void write_bytes(char *addr)
44{
45 unsigned long i;
46
47 for (i = 0; i < LENGTH; i++)
48 *(addr + i) = (char)i;
49}
50
51static void read_bytes(char *addr)
52{
53 unsigned long i;
54
55 check_bytes(addr);
56 for (i = 0; i < LENGTH; i++)
57 if (*(addr + i) != (char)i) {
58 printf("Mismatch at %lu\n", i);
59 break;
60 }
61}
62
63int main(void)
64{
65 void *addr;
66 int fd;
67
68 fd = open(FILE_NAME, O_CREAT | O_RDWR, 0755);
69 if (fd < 0) {
70 perror("Open failed");
71 exit(1);
72 }
73
74 addr = mmap(ADDR, LENGTH, PROTECTION, FLAGS, fd, 0);
75 if (addr == MAP_FAILED) {
76 perror("mmap");
77 unlink(FILE_NAME);
78 exit(1);
79 }
80
81 printf("Returned address is %p\n", addr);
82 check_bytes(addr);
83 write_bytes(addr);
84 read_bytes(addr);
85
86 munmap(addr, LENGTH);
87 close(fd);
88 unlink(FILE_NAME);
89
90 return 0;
91}
diff --git a/Documentation/vm/hugepage-shm.c b/Documentation/vm/hugepage-shm.c
new file mode 100644
index 00000000000..07956d8592c
--- /dev/null
+++ b/Documentation/vm/hugepage-shm.c
@@ -0,0 +1,98 @@
1/*
2 * hugepage-shm:
3 *
4 * Example of using huge page memory in a user application using Sys V shared
5 * memory system calls. In this example the app is requesting 256MB of
6 * memory that is backed by huge pages. The application uses the flag
7 * SHM_HUGETLB in the shmget system call to inform the kernel that it is
8 * requesting huge pages.
9 *
10 * For the ia64 architecture, the Linux kernel reserves Region number 4 for
11 * huge pages. That means that if one requires a fixed address, a huge page
12 * aligned address starting with 0x800000... will be required. If a fixed
13 * address is not required, the kernel will select an address in the proper
14 * range.
15 * Other architectures, such as ppc64, i386 or x86_64 are not so constrained.
16 *
17 * Note: The default shared memory limit is quite low on many kernels,
18 * you may need to increase it via:
19 *
20 * echo 268435456 > /proc/sys/kernel/shmmax
21 *
22 * This will increase the maximum size per shared memory segment to 256MB.
23 * The other limit that you will hit eventually is shmall which is the
24 * total amount of shared memory in pages. To set it to 16GB on a system
25 * with a 4kB pagesize do:
26 *
27 * echo 4194304 > /proc/sys/kernel/shmall
28 */
29
30#include <stdlib.h>
31#include <stdio.h>
32#include <sys/types.h>
33#include <sys/ipc.h>
34#include <sys/shm.h>
35#include <sys/mman.h>
36
37#ifndef SHM_HUGETLB
38#define SHM_HUGETLB 04000
39#endif
40
41#define LENGTH (256UL*1024*1024)
42
43#define dprintf(x) printf(x)
44
45/* Only ia64 requires this */
46#ifdef __ia64__
47#define ADDR (void *)(0x8000000000000000UL)
48#define SHMAT_FLAGS (SHM_RND)
49#else
50#define ADDR (void *)(0x0UL)
51#define SHMAT_FLAGS (0)
52#endif
53
54int main(void)
55{
56 int shmid;
57 unsigned long i;
58 char *shmaddr;
59
60 if ((shmid = shmget(2, LENGTH,
61 SHM_HUGETLB | IPC_CREAT | SHM_R | SHM_W)) < 0) {
62 perror("shmget");
63 exit(1);
64 }
65 printf("shmid: 0x%x\n", shmid);
66
67 shmaddr = shmat(shmid, ADDR, SHMAT_FLAGS);
68 if (shmaddr == (char *)-1) {
69 perror("Shared memory attach failure");
70 shmctl(shmid, IPC_RMID, NULL);
71 exit(2);
72 }
73 printf("shmaddr: %p\n", shmaddr);
74
75 dprintf("Starting the writes:\n");
76 for (i = 0; i < LENGTH; i++) {
77 shmaddr[i] = (char)(i);
78 if (!(i % (1024 * 1024)))
79 dprintf(".");
80 }
81 dprintf("\n");
82
83 dprintf("Starting the Check...");
84 for (i = 0; i < LENGTH; i++)
85 if (shmaddr[i] != (char)i)
86 printf("\nIndex %lu mismatched\n", i);
87 dprintf("Done.\n");
88
89 if (shmdt((const void *)shmaddr) != 0) {
90 perror("Detach failure");
91 shmctl(shmid, IPC_RMID, NULL);
92 exit(3);
93 }
94
95 shmctl(shmid, IPC_RMID, NULL);
96
97 return 0;
98}
diff --git a/Documentation/vm/map_hugetlb.c b/Documentation/vm/map_hugetlb.c
new file mode 100644
index 00000000000..eda1a6d3578
--- /dev/null
+++ b/Documentation/vm/map_hugetlb.c
@@ -0,0 +1,77 @@
1/*
2 * Example of using hugepage memory in a user application using the mmap
3 * system call with MAP_HUGETLB flag. Before running this program make
4 * sure the administrator has allocated enough default sized huge pages
5 * to cover the 256 MB allocation.
6 *
7 * For ia64 architecture, Linux kernel reserves Region number 4 for hugepages.
8 * That means the addresses starting with 0x800000... will need to be
9 * specified. Specifying a fixed address is not required on ppc64, i386
10 * or x86_64.
11 */
12#include <stdlib.h>
13#include <stdio.h>
14#include <unistd.h>
15#include <sys/mman.h>
16#include <fcntl.h>
17
18#define LENGTH (256UL*1024*1024)
19#define PROTECTION (PROT_READ | PROT_WRITE)
20
21#ifndef MAP_HUGETLB
22#define MAP_HUGETLB 0x40000 /* arch specific */
23#endif
24
25/* Only ia64 requires this */
26#ifdef __ia64__
27#define ADDR (void *)(0x8000000000000000UL)
28#define FLAGS (MAP_PRIVATE | MAP_ANONYMOUS | MAP_HUGETLB | MAP_FIXED)
29#else
30#define ADDR (void *)(0x0UL)
31#define FLAGS (MAP_PRIVATE | MAP_ANONYMOUS | MAP_HUGETLB)
32#endif
33
34static void check_bytes(char *addr)
35{
36 printf("First hex is %x\n", *((unsigned int *)addr));
37}
38
39static void write_bytes(char *addr)
40{
41 unsigned long i;
42
43 for (i = 0; i < LENGTH; i++)
44 *(addr + i) = (char)i;
45}
46
47static void read_bytes(char *addr)
48{
49 unsigned long i;
50
51 check_bytes(addr);
52 for (i = 0; i < LENGTH; i++)
53 if (*(addr + i) != (char)i) {
54 printf("Mismatch at %lu\n", i);
55 break;
56 }
57}
58
59int main(void)
60{
61 void *addr;
62
63 addr = mmap(ADDR, LENGTH, PROTECTION, FLAGS, 0, 0);
64 if (addr == MAP_FAILED) {
65 perror("mmap");
66 exit(1);
67 }
68
69 printf("Returned address is %p\n", addr);
70 check_bytes(addr);
71 write_bytes(addr);
72 read_bytes(addr);
73
74 munmap(addr, LENGTH);
75
76 return 0;
77}
diff --git a/Documentation/vm/page-types.c b/Documentation/vm/page-types.c
new file mode 100644
index 00000000000..7445caa26d0
--- /dev/null
+++ b/Documentation/vm/page-types.c
@@ -0,0 +1,1100 @@
1/*
2 * page-types: Tool for querying page flags
3 *
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms of the GNU General Public License as published by the Free
6 * Software Foundation; version 2.
7 *
8 * This program is distributed in the hope that it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
11 * more details.
12 *
13 * You should find a copy of v2 of the GNU General Public License somewhere on
14 * your Linux system; if not, write to the Free Software Foundation, Inc., 59
15 * Temple Place, Suite 330, Boston, MA 02111-1307 USA.
16 *
17 * Copyright (C) 2009 Intel corporation
18 *
19 * Authors: Wu Fengguang <fengguang.wu@intel.com>
20 */
21
22#define _LARGEFILE64_SOURCE
23#include <stdio.h>
24#include <stdlib.h>
25#include <unistd.h>
26#include <stdint.h>
27#include <stdarg.h>
28#include <string.h>
29#include <getopt.h>
30#include <limits.h>
31#include <assert.h>
32#include <sys/types.h>
33#include <sys/errno.h>
34#include <sys/fcntl.h>
35#include <sys/mount.h>
36#include <sys/statfs.h>
37#include "../../include/linux/magic.h"
38
39
40#ifndef MAX_PATH
41# define MAX_PATH 256
42#endif
43
44#ifndef STR
45# define _STR(x) #x
46# define STR(x) _STR(x)
47#endif
48
49/*
50 * pagemap kernel ABI bits
51 */
52
53#define PM_ENTRY_BYTES sizeof(uint64_t)
54#define PM_STATUS_BITS 3
55#define PM_STATUS_OFFSET (64 - PM_STATUS_BITS)
56#define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
57#define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
58#define PM_PSHIFT_BITS 6
59#define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
60#define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
61#define PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
62#define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
63#define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
64
65#define PM_PRESENT PM_STATUS(4LL)
66#define PM_SWAP PM_STATUS(2LL)
67
68
69/*
70 * kernel page flags
71 */
72
73#define KPF_BYTES 8
74#define PROC_KPAGEFLAGS "/proc/kpageflags"
75
76/* copied from kpageflags_read() */
77#define KPF_LOCKED 0
78#define KPF_ERROR 1
79#define KPF_REFERENCED 2
80#define KPF_UPTODATE 3
81#define KPF_DIRTY 4
82#define KPF_LRU 5
83#define KPF_ACTIVE 6
84#define KPF_SLAB 7
85#define KPF_WRITEBACK 8
86#define KPF_RECLAIM 9
87#define KPF_BUDDY 10
88
89/* [11-20] new additions in 2.6.31 */
90#define KPF_MMAP 11
91#define KPF_ANON 12
92#define KPF_SWAPCACHE 13
93#define KPF_SWAPBACKED 14
94#define KPF_COMPOUND_HEAD 15
95#define KPF_COMPOUND_TAIL 16
96#define KPF_HUGE 17
97#define KPF_UNEVICTABLE 18
98#define KPF_HWPOISON 19
99#define KPF_NOPAGE 20
100#define KPF_KSM 21
101
102/* [32-] kernel hacking assistances */
103#define KPF_RESERVED 32
104#define KPF_MLOCKED 33
105#define KPF_MAPPEDTODISK 34
106#define KPF_PRIVATE 35
107#define KPF_PRIVATE_2 36
108#define KPF_OWNER_PRIVATE 37
109#define KPF_ARCH 38
110#define KPF_UNCACHED 39
111
112/* [48-] take some arbitrary free slots for expanding overloaded flags
113 * not part of kernel API
114 */
115#define KPF_READAHEAD 48
116#define KPF_SLOB_FREE 49
117#define KPF_SLUB_FROZEN 50
118#define KPF_SLUB_DEBUG 51
119
120#define KPF_ALL_BITS ((uint64_t)~0ULL)
121#define KPF_HACKERS_BITS (0xffffULL << 32)
122#define KPF_OVERLOADED_BITS (0xffffULL << 48)
123#define BIT(name) (1ULL << KPF_##name)
124#define BITS_COMPOUND (BIT(COMPOUND_HEAD) | BIT(COMPOUND_TAIL))
125
126static const char *page_flag_names[] = {
127 [KPF_LOCKED] = "L:locked",
128 [KPF_ERROR] = "E:error",
129 [KPF_REFERENCED] = "R:referenced",
130 [KPF_UPTODATE] = "U:uptodate",
131 [KPF_DIRTY] = "D:dirty",
132 [KPF_LRU] = "l:lru",
133 [KPF_ACTIVE] = "A:active",
134 [KPF_SLAB] = "S:slab",
135 [KPF_WRITEBACK] = "W:writeback",
136 [KPF_RECLAIM] = "I:reclaim",
137 [KPF_BUDDY] = "B:buddy",
138
139 [KPF_MMAP] = "M:mmap",
140 [KPF_ANON] = "a:anonymous",
141 [KPF_SWAPCACHE] = "s:swapcache",
142 [KPF_SWAPBACKED] = "b:swapbacked",
143 [KPF_COMPOUND_HEAD] = "H:compound_head",
144 [KPF_COMPOUND_TAIL] = "T:compound_tail",
145 [KPF_HUGE] = "G:huge",
146 [KPF_UNEVICTABLE] = "u:unevictable",
147 [KPF_HWPOISON] = "X:hwpoison",
148 [KPF_NOPAGE] = "n:nopage",
149 [KPF_KSM] = "x:ksm",
150
151 [KPF_RESERVED] = "r:reserved",
152 [KPF_MLOCKED] = "m:mlocked",
153 [KPF_MAPPEDTODISK] = "d:mappedtodisk",
154 [KPF_PRIVATE] = "P:private",
155 [KPF_PRIVATE_2] = "p:private_2",
156 [KPF_OWNER_PRIVATE] = "O:owner_private",
157 [KPF_ARCH] = "h:arch",
158 [KPF_UNCACHED] = "c:uncached",
159
160 [KPF_READAHEAD] = "I:readahead",
161 [KPF_SLOB_FREE] = "P:slob_free",
162 [KPF_SLUB_FROZEN] = "A:slub_frozen",
163 [KPF_SLUB_DEBUG] = "E:slub_debug",
164};
165
166
167static const char *debugfs_known_mountpoints[] = {
168 "/sys/kernel/debug",
169 "/debug",
170 0,
171};
172
173/*
174 * data structures
175 */
176
177static int opt_raw; /* for kernel developers */
178static int opt_list; /* list pages (in ranges) */
179static int opt_no_summary; /* don't show summary */
180static pid_t opt_pid; /* process to walk */
181
182#define MAX_ADDR_RANGES 1024
183static int nr_addr_ranges;
184static unsigned long opt_offset[MAX_ADDR_RANGES];
185static unsigned long opt_size[MAX_ADDR_RANGES];
186
187#define MAX_VMAS 10240
188static int nr_vmas;
189static unsigned long pg_start[MAX_VMAS];
190static unsigned long pg_end[MAX_VMAS];
191
192#define MAX_BIT_FILTERS 64
193static int nr_bit_filters;
194static uint64_t opt_mask[MAX_BIT_FILTERS];
195static uint64_t opt_bits[MAX_BIT_FILTERS];
196
197static int page_size;
198
199static int pagemap_fd;
200static int kpageflags_fd;
201
202static int opt_hwpoison;
203static int opt_unpoison;
204
205static char hwpoison_debug_fs[MAX_PATH+1];
206static int hwpoison_inject_fd;
207static int hwpoison_forget_fd;
208
209#define HASH_SHIFT 13
210#define HASH_SIZE (1 << HASH_SHIFT)
211#define HASH_MASK (HASH_SIZE - 1)
212#define HASH_KEY(flags) (flags & HASH_MASK)
213
214static unsigned long total_pages;
215static unsigned long nr_pages[HASH_SIZE];
216static uint64_t page_flags[HASH_SIZE];
217
218
219/*
220 * helper functions
221 */
222
223#define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0]))
224
225#define min_t(type, x, y) ({ \
226 type __min1 = (x); \
227 type __min2 = (y); \
228 __min1 < __min2 ? __min1 : __min2; })
229
230#define max_t(type, x, y) ({ \
231 type __max1 = (x); \
232 type __max2 = (y); \
233 __max1 > __max2 ? __max1 : __max2; })
234
235static unsigned long pages2mb(unsigned long pages)
236{
237 return (pages * page_size) >> 20;
238}
239
240static void fatal(const char *x, ...)
241{
242 va_list ap;
243
244 va_start(ap, x);
245 vfprintf(stderr, x, ap);
246 va_end(ap);
247 exit(EXIT_FAILURE);
248}
249
250static int checked_open(const char *pathname, int flags)
251{
252 int fd = open(pathname, flags);
253
254 if (fd < 0) {
255 perror(pathname);
256 exit(EXIT_FAILURE);
257 }
258
259 return fd;
260}
261
262/*
263 * pagemap/kpageflags routines
264 */
265
266static unsigned long do_u64_read(int fd, char *name,
267 uint64_t *buf,
268 unsigned long index,
269 unsigned long count)
270{
271 long bytes;
272
273 if (index > ULONG_MAX / 8)
274 fatal("index overflow: %lu\n", index);
275
276 if (lseek(fd, index * 8, SEEK_SET) < 0) {
277 perror(name);
278 exit(EXIT_FAILURE);
279 }
280
281 bytes = read(fd, buf, count * 8);
282 if (bytes < 0) {
283 perror(name);
284 exit(EXIT_FAILURE);
285 }
286 if (bytes % 8)
287 fatal("partial read: %lu bytes\n", bytes);
288
289 return bytes / 8;
290}
291
292static unsigned long kpageflags_read(uint64_t *buf,
293 unsigned long index,
294 unsigned long pages)
295{
296 return do_u64_read(kpageflags_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
318
319/*
320 * page flag names
321 */
322
323static char *page_flag_name(uint64_t flags)
324{
325 static char buf[65];
326 int present;
327 int i, j;
328
329 for (i = 0, j = 0; i < ARRAY_SIZE(page_flag_names); i++) {
330 present = (flags >> i) & 1;
331 if (!page_flag_names[i]) {
332 if (present)
333 fatal("unknown flag bit %d\n", i);
334 continue;
335 }
336 buf[j++] = present ? page_flag_names[i][0] : '_';
337 }
338
339 return buf;
340}
341
342static char *page_flag_longname(uint64_t flags)
343{
344 static char buf[1024];
345 int i, n;
346
347 for (i = 0, n = 0; i < ARRAY_SIZE(page_flag_names); i++) {
348 if (!page_flag_names[i])
349 continue;
350 if ((flags >> i) & 1)
351 n += snprintf(buf + n, sizeof(buf) - n, "%s,",
352 page_flag_names[i] + 2);
353 }
354 if (n)
355 n--;
356 buf[n] = '\0';
357
358 return buf;
359}
360
361
362/*
363 * page list and summary
364 */
365
366static void show_page_range(unsigned long voffset,
367 unsigned long offset, uint64_t flags)
368{
369 static uint64_t flags0;
370 static unsigned long voff;
371 static unsigned long index;
372 static unsigned long count;
373
374 if (flags == flags0 && offset == index + count &&
375 (!opt_pid || voffset == voff + count)) {
376 count++;
377 return;
378 }
379
380 if (count) {
381 if (opt_pid)
382 printf("%lx\t", voff);
383 printf("%lx\t%lx\t%s\n",
384 index, count, page_flag_name(flags0));
385 }
386
387 flags0 = flags;
388 index = offset;
389 voff = voffset;
390 count = 1;
391}
392
393static void show_page(unsigned long voffset,
394 unsigned long offset, uint64_t flags)
395{
396 if (opt_pid)
397 printf("%lx\t", voffset);
398 printf("%lx\t%s\n", offset, page_flag_name(flags));
399}
400
401static void show_summary(void)
402{
403 int i;
404
405 printf(" flags\tpage-count MB"
406 " symbolic-flags\t\t\tlong-symbolic-flags\n");
407
408 for (i = 0; i < ARRAY_SIZE(nr_pages); i++) {
409 if (nr_pages[i])
410 printf("0x%016llx\t%10lu %8lu %s\t%s\n",
411 (unsigned long long)page_flags[i],
412 nr_pages[i],
413 pages2mb(nr_pages[i]),
414 page_flag_name(page_flags[i]),
415 page_flag_longname(page_flags[i]));
416 }
417
418 printf(" total\t%10lu %8lu\n",
419 total_pages, pages2mb(total_pages));
420}
421
422
423/*
424 * page flag filters
425 */
426
427static int bit_mask_ok(uint64_t flags)
428{
429 int i;
430
431 for (i = 0; i < nr_bit_filters; i++) {
432 if (opt_bits[i] == KPF_ALL_BITS) {
433 if ((flags & opt_mask[i]) == 0)
434 return 0;
435 } else {
436 if ((flags & opt_mask[i]) != opt_bits[i])
437 return 0;
438 }
439 }
440
441 return 1;
442}
443
444static uint64_t expand_overloaded_flags(uint64_t flags)
445{
446 /* SLOB/SLUB overload several page flags */
447 if (flags & BIT(SLAB)) {
448 if (flags & BIT(PRIVATE))
449 flags ^= BIT(PRIVATE) | BIT(SLOB_FREE);
450 if (flags & BIT(ACTIVE))
451 flags ^= BIT(ACTIVE) | BIT(SLUB_FROZEN);
452 if (flags & BIT(ERROR))
453 flags ^= BIT(ERROR) | BIT(SLUB_DEBUG);
454 }
455
456 /* PG_reclaim is overloaded as PG_readahead in the read path */
457 if ((flags & (BIT(RECLAIM) | BIT(WRITEBACK))) == BIT(RECLAIM))
458 flags ^= BIT(RECLAIM) | BIT(READAHEAD);
459
460 return flags;
461}
462
463static uint64_t well_known_flags(uint64_t flags)
464{
465 /* hide flags intended only for kernel hacker */
466 flags &= ~KPF_HACKERS_BITS;
467
468 /* hide non-hugeTLB compound pages */
469 if ((flags & BITS_COMPOUND) && !(flags & BIT(HUGE)))
470 flags &= ~BITS_COMPOUND;
471
472 return flags;
473}
474
475static uint64_t kpageflags_flags(uint64_t flags)
476{
477 flags = expand_overloaded_flags(flags);
478
479 if (!opt_raw)
480 flags = well_known_flags(flags);
481
482 return flags;
483}
484
485/* verify that a mountpoint is actually a debugfs instance */
486static int debugfs_valid_mountpoint(const char *debugfs)
487{
488 struct statfs st_fs;
489
490 if (statfs(debugfs, &st_fs) < 0)
491 return -ENOENT;
492 else if (st_fs.f_type != (long) DEBUGFS_MAGIC)
493 return -ENOENT;
494
495 return 0;
496}
497
498/* find the path to the mounted debugfs */
499static const char *debugfs_find_mountpoint(void)
500{
501 const char **ptr;
502 char type[100];
503 FILE *fp;
504
505 ptr = debugfs_known_mountpoints;
506 while (*ptr) {
507 if (debugfs_valid_mountpoint(*ptr) == 0) {
508 strcpy(hwpoison_debug_fs, *ptr);
509 return hwpoison_debug_fs;
510 }
511 ptr++;
512 }
513
514 /* give up and parse /proc/mounts */
515 fp = fopen("/proc/mounts", "r");
516 if (fp == NULL)
517 perror("Can't open /proc/mounts for read");
518
519 while (fscanf(fp, "%*s %"
520 STR(MAX_PATH)
521 "s %99s %*s %*d %*d\n",
522 hwpoison_debug_fs, type) == 2) {
523 if (strcmp(type, "debugfs") == 0)
524 break;
525 }
526 fclose(fp);
527
528 if (strcmp(type, "debugfs") != 0)
529 return NULL;
530
531 return hwpoison_debug_fs;
532}
533
534/* mount the debugfs somewhere if it's not mounted */
535
536static void debugfs_mount(void)
537{
538 const char **ptr;
539
540 /* see if it's already mounted */
541 if (debugfs_find_mountpoint())
542 return;
543
544 ptr = debugfs_known_mountpoints;
545 while (*ptr) {
546 if (mount(NULL, *ptr, "debugfs", 0, NULL) == 0) {
547 /* save the mountpoint */
548 strcpy(hwpoison_debug_fs, *ptr);
549 break;
550 }
551 ptr++;
552 }
553
554 if (*ptr == NULL) {
555 perror("mount debugfs");
556 exit(EXIT_FAILURE);
557 }
558}
559
560/*
561 * page actions
562 */
563
564static void prepare_hwpoison_fd(void)
565{
566 char buf[MAX_PATH + 1];
567
568 debugfs_mount();
569
570 if (opt_hwpoison && !hwpoison_inject_fd) {
571 snprintf(buf, MAX_PATH, "%s/hwpoison/corrupt-pfn",
572 hwpoison_debug_fs);
573 hwpoison_inject_fd = checked_open(buf, O_WRONLY);
574 }
575
576 if (opt_unpoison && !hwpoison_forget_fd) {
577 snprintf(buf, MAX_PATH, "%s/hwpoison/unpoison-pfn",
578 hwpoison_debug_fs);
579 hwpoison_forget_fd = checked_open(buf, O_WRONLY);
580 }
581}
582
583static int hwpoison_page(unsigned long offset)
584{
585 char buf[100];
586 int len;
587
588 len = sprintf(buf, "0x%lx\n", offset);
589 len = write(hwpoison_inject_fd, buf, len);
590 if (len < 0) {
591 perror("hwpoison inject");
592 return len;
593 }
594 return 0;
595}
596
597static int unpoison_page(unsigned long offset)
598{
599 char buf[100];
600 int len;
601
602 len = sprintf(buf, "0x%lx\n", offset);
603 len = write(hwpoison_forget_fd, buf, len);
604 if (len < 0) {
605 perror("hwpoison forget");
606 return len;
607 }
608 return 0;
609}
610
611/*
612 * page frame walker
613 */
614
615static int hash_slot(uint64_t flags)
616{
617 int k = HASH_KEY(flags);
618 int i;
619
620 /* Explicitly reserve slot 0 for flags 0: the following logic
621 * cannot distinguish an unoccupied slot from slot (flags==0).
622 */
623 if (flags == 0)
624 return 0;
625
626 /* search through the remaining (HASH_SIZE-1) slots */
627 for (i = 1; i < ARRAY_SIZE(page_flags); i++, k++) {
628 if (!k || k >= ARRAY_SIZE(page_flags))
629 k = 1;
630 if (page_flags[k] == 0) {
631 page_flags[k] = flags;
632 return k;
633 }
634 if (page_flags[k] == flags)
635 return k;
636 }
637
638 fatal("hash table full: bump up HASH_SHIFT?\n");
639 exit(EXIT_FAILURE);
640}
641
642static void add_page(unsigned long voffset,
643 unsigned long offset, uint64_t flags)
644{
645 flags = kpageflags_flags(flags);
646
647 if (!bit_mask_ok(flags))
648 return;
649
650 if (opt_hwpoison)
651 hwpoison_page(offset);
652 if (opt_unpoison)
653 unpoison_page(offset);
654
655 if (opt_list == 1)
656 show_page_range(voffset, offset, flags);
657 else if (opt_list == 2)
658 show_page(voffset, offset, flags);
659
660 nr_pages[hash_slot(flags)]++;
661 total_pages++;
662}
663
664#define KPAGEFLAGS_BATCH (64 << 10) /* 64k pages */
665static void walk_pfn(unsigned long voffset,
666 unsigned long index,
667 unsigned long count)
668{
669 uint64_t buf[KPAGEFLAGS_BATCH];
670 unsigned long batch;
671 long pages;
672 unsigned long i;
673
674 while (count) {
675 batch = min_t(unsigned long, count, KPAGEFLAGS_BATCH);
676 pages = kpageflags_read(buf, index, batch);
677 if (pages == 0)
678 break;
679
680 for (i = 0; i < pages; i++)
681 add_page(voffset + i, index + i, buf[i]);
682
683 index += pages;
684 count -= pages;
685 }
686}
687
688#define PAGEMAP_BATCH (64 << 10)
689static void walk_vma(unsigned long index, unsigned long count)
690{
691 uint64_t buf[PAGEMAP_BATCH];
692 unsigned long batch;
693 unsigned long pages;
694 unsigned long pfn;
695 unsigned long i;
696
697 while (count) {
698 batch = min_t(unsigned long, count, PAGEMAP_BATCH);
699 pages = pagemap_read(buf, index, batch);
700 if (pages == 0)
701 break;
702
703 for (i = 0; i < pages; i++) {
704 pfn = pagemap_pfn(buf[i]);
705 if (pfn)
706 walk_pfn(index + i, pfn, 1);
707 }
708
709 index += pages;
710 count -= pages;
711 }
712}
713
714static void walk_task(unsigned long index, unsigned long count)
715{
716 const unsigned long end = index + count;
717 unsigned long start;
718 int i = 0;
719
720 while (index < end) {
721
722 while (pg_end[i] <= index)
723 if (++i >= nr_vmas)
724 return;
725 if (pg_start[i] >= end)
726 return;
727
728 start = max_t(unsigned long, pg_start[i], index);
729 index = min_t(unsigned long, pg_end[i], end);
730
731 assert(start < index);
732 walk_vma(start, index - start);
733 }
734}
735
736static void add_addr_range(unsigned long offset, unsigned long size)
737{
738 if (nr_addr_ranges >= MAX_ADDR_RANGES)
739 fatal("too many addr ranges\n");
740
741 opt_offset[nr_addr_ranges] = offset;
742 opt_size[nr_addr_ranges] = min_t(unsigned long, size, ULONG_MAX-offset);
743 nr_addr_ranges++;
744}
745
746static void walk_addr_ranges(void)
747{
748 int i;
749
750 kpageflags_fd = checked_open(PROC_KPAGEFLAGS, O_RDONLY);
751
752 if (!nr_addr_ranges)
753 add_addr_range(0, ULONG_MAX);
754
755 for (i = 0; i < nr_addr_ranges; i++)
756 if (!opt_pid)
757 walk_pfn(0, opt_offset[i], opt_size[i]);
758 else
759 walk_task(opt_offset[i], opt_size[i]);
760
761 close(kpageflags_fd);
762}
763
764
765/*
766 * user interface
767 */
768
769static const char *page_flag_type(uint64_t flag)
770{
771 if (flag & KPF_HACKERS_BITS)
772 return "(r)";
773 if (flag & KPF_OVERLOADED_BITS)
774 return "(o)";
775 return " ";
776}
777
778static void usage(void)
779{
780 int i, j;
781
782 printf(
783"page-types [options]\n"
784" -r|--raw Raw mode, for kernel developers\n"
785" -d|--describe flags Describe flags\n"
786" -a|--addr addr-spec Walk a range of pages\n"
787" -b|--bits bits-spec Walk pages with specified bits\n"
788" -p|--pid pid Walk process address space\n"
789#if 0 /* planned features */
790" -f|--file filename Walk file address space\n"
791#endif
792" -l|--list Show page details in ranges\n"
793" -L|--list-each Show page details one by one\n"
794" -N|--no-summary Don't show summary info\n"
795" -X|--hwpoison hwpoison pages\n"
796" -x|--unpoison unpoison pages\n"
797" -h|--help Show this usage message\n"
798"flags:\n"
799" 0x10 bitfield format, e.g.\n"
800" anon bit-name, e.g.\n"
801" 0x10,anon comma-separated list, e.g.\n"
802"addr-spec:\n"
803" N one page at offset N (unit: pages)\n"
804" N+M pages range from N to N+M-1\n"
805" N,M pages range from N to M-1\n"
806" N, pages range from N to end\n"
807" ,M pages range from 0 to M-1\n"
808"bits-spec:\n"
809" bit1,bit2 (flags & (bit1|bit2)) != 0\n"
810" bit1,bit2=bit1 (flags & (bit1|bit2)) == bit1\n"
811" bit1,~bit2 (flags & (bit1|bit2)) == bit1\n"
812" =bit1,bit2 flags == (bit1|bit2)\n"
813"bit-names:\n"
814 );
815
816 for (i = 0, j = 0; i < ARRAY_SIZE(page_flag_names); i++) {
817 if (!page_flag_names[i])
818 continue;
819 printf("%16s%s", page_flag_names[i] + 2,
820 page_flag_type(1ULL << i));
821 if (++j > 3) {
822 j = 0;
823 putchar('\n');
824 }
825 }
826 printf("\n "
827 "(r) raw mode bits (o) overloaded bits\n");
828}
829
830static unsigned long long parse_number(const char *str)
831{
832 unsigned long long n;
833
834 n = strtoll(str, NULL, 0);
835
836 if (n == 0 && str[0] != '0')
837 fatal("invalid name or number: %s\n", str);
838
839 return n;
840}
841
842static void parse_pid(const char *str)
843{
844 FILE *file;
845 char buf[5000];
846
847 opt_pid = parse_number(str);
848
849 sprintf(buf, "/proc/%d/pagemap", opt_pid);
850 pagemap_fd = checked_open(buf, O_RDONLY);
851
852 sprintf(buf, "/proc/%d/maps", opt_pid);
853 file = fopen(buf, "r");
854 if (!file) {
855 perror(buf);
856 exit(EXIT_FAILURE);
857 }
858
859 while (fgets(buf, sizeof(buf), file) != NULL) {
860 unsigned long vm_start;
861 unsigned long vm_end;
862 unsigned long long pgoff;
863 int major, minor;
864 char r, w, x, s;
865 unsigned long ino;
866 int n;
867
868 n = sscanf(buf, "%lx-%lx %c%c%c%c %llx %x:%x %lu",
869 &vm_start,
870 &vm_end,
871 &r, &w, &x, &s,
872 &pgoff,
873 &major, &minor,
874 &ino);
875 if (n < 10) {
876 fprintf(stderr, "unexpected line: %s\n", buf);
877 continue;
878 }
879 pg_start[nr_vmas] = vm_start / page_size;
880 pg_end[nr_vmas] = vm_end / page_size;
881 if (++nr_vmas >= MAX_VMAS) {
882 fprintf(stderr, "too many VMAs\n");
883 break;
884 }
885 }
886 fclose(file);
887}
888
889static void parse_file(const char *name)
890{
891}
892
893static void parse_addr_range(const char *optarg)
894{
895 unsigned long offset;
896 unsigned long size;
897 char *p;
898
899 p = strchr(optarg, ',');
900 if (!p)
901 p = strchr(optarg, '+');
902
903 if (p == optarg) {
904 offset = 0;
905 size = parse_number(p + 1);
906 } else if (p) {
907 offset = parse_number(optarg);
908 if (p[1] == '\0')
909 size = ULONG_MAX;
910 else {
911 size = parse_number(p + 1);
912 if (*p == ',') {
913 if (size < offset)
914 fatal("invalid range: %lu,%lu\n",
915 offset, size);
916 size -= offset;
917 }
918 }
919 } else {
920 offset = parse_number(optarg);
921 size = 1;
922 }
923
924 add_addr_range(offset, size);
925}
926
927static void add_bits_filter(uint64_t mask, uint64_t bits)
928{
929 if (nr_bit_filters >= MAX_BIT_FILTERS)
930 fatal("too much bit filters\n");
931
932 opt_mask[nr_bit_filters] = mask;
933 opt_bits[nr_bit_filters] = bits;
934 nr_bit_filters++;
935}
936
937static uint64_t parse_flag_name(const char *str, int len)
938{
939 int i;
940
941 if (!*str || !len)
942 return 0;
943
944 if (len <= 8 && !strncmp(str, "compound", len))
945 return BITS_COMPOUND;
946
947 for (i = 0; i < ARRAY_SIZE(page_flag_names); i++) {
948 if (!page_flag_names[i])
949 continue;
950 if (!strncmp(str, page_flag_names[i] + 2, len))
951 return 1ULL << i;
952 }
953
954 return parse_number(str);
955}
956
957static uint64_t parse_flag_names(const char *str, int all)
958{
959 const char *p = str;
960 uint64_t flags = 0;
961
962 while (1) {
963 if (*p == ',' || *p == '=' || *p == '\0') {
964 if ((*str != '~') || (*str == '~' && all && *++str))
965 flags |= parse_flag_name(str, p - str);
966 if (*p != ',')
967 break;
968 str = p + 1;
969 }
970 p++;
971 }
972
973 return flags;
974}
975
976static void parse_bits_mask(const char *optarg)
977{
978 uint64_t mask;
979 uint64_t bits;
980 const char *p;
981
982 p = strchr(optarg, '=');
983 if (p == optarg) {
984 mask = KPF_ALL_BITS;
985 bits = parse_flag_names(p + 1, 0);
986 } else if (p) {
987 mask = parse_flag_names(optarg, 0);
988 bits = parse_flag_names(p + 1, 0);
989 } else if (strchr(optarg, '~')) {
990 mask = parse_flag_names(optarg, 1);
991 bits = parse_flag_names(optarg, 0);
992 } else {
993 mask = parse_flag_names(optarg, 0);
994 bits = KPF_ALL_BITS;
995 }
996
997 add_bits_filter(mask, bits);
998}
999
1000static void describe_flags(const char *optarg)
1001{
1002 uint64_t flags = parse_flag_names(optarg, 0);
1003
1004 printf("0x%016llx\t%s\t%s\n",
1005 (unsigned long long)flags,
1006 page_flag_name(flags),
1007 page_flag_longname(flags));
1008}
1009
1010static const struct option opts[] = {
1011 { "raw" , 0, NULL, 'r' },
1012 { "pid" , 1, NULL, 'p' },
1013 { "file" , 1, NULL, 'f' },
1014 { "addr" , 1, NULL, 'a' },
1015 { "bits" , 1, NULL, 'b' },
1016 { "describe" , 1, NULL, 'd' },
1017 { "list" , 0, NULL, 'l' },
1018 { "list-each" , 0, NULL, 'L' },
1019 { "no-summary", 0, NULL, 'N' },
1020 { "hwpoison" , 0, NULL, 'X' },
1021 { "unpoison" , 0, NULL, 'x' },
1022 { "help" , 0, NULL, 'h' },
1023 { NULL , 0, NULL, 0 }
1024};
1025
1026int main(int argc, char *argv[])
1027{
1028 int c;
1029
1030 page_size = getpagesize();
1031
1032 while ((c = getopt_long(argc, argv,
1033 "rp:f:a:b:d:lLNXxh", opts, NULL)) != -1) {
1034 switch (c) {
1035 case 'r':
1036 opt_raw = 1;
1037 break;
1038 case 'p':
1039 parse_pid(optarg);
1040 break;
1041 case 'f':
1042 parse_file(optarg);
1043 break;
1044 case 'a':
1045 parse_addr_range(optarg);
1046 break;
1047 case 'b':
1048 parse_bits_mask(optarg);
1049 break;
1050 case 'd':
1051 describe_flags(optarg);
1052 exit(0);
1053 case 'l':
1054 opt_list = 1;
1055 break;
1056 case 'L':
1057 opt_list = 2;
1058 break;
1059 case 'N':
1060 opt_no_summary = 1;
1061 break;
1062 case 'X':
1063 opt_hwpoison = 1;
1064 prepare_hwpoison_fd();
1065 break;
1066 case 'x':
1067 opt_unpoison = 1;
1068 prepare_hwpoison_fd();
1069 break;
1070 case 'h':
1071 usage();
1072 exit(0);
1073 default:
1074 usage();
1075 exit(1);
1076 }
1077 }
1078
1079 if (opt_list && opt_pid)
1080 printf("voffset\t");
1081 if (opt_list == 1)
1082 printf("offset\tlen\tflags\n");
1083 if (opt_list == 2)
1084 printf("offset\tflags\n");
1085
1086 walk_addr_ranges();
1087
1088 if (opt_list == 1)
1089 show_page_range(0, 0, 0); /* drain the buffer */
1090
1091 if (opt_no_summary)
1092 return 0;
1093
1094 if (opt_list)
1095 printf("\n\n");
1096
1097 show_summary();
1098
1099 return 0;
1100}
generated by cgit v1.2.2 (git 2.25.0) at 2026-01-11 20:34:47 -0500