Fix compile for older Linux without mmap MAP_HUGETLB
[fio.git] / os / os-linux.h
CommitLineData
ebac4655
JA
1#ifndef FIO_OS_LINUX_H
2#define FIO_OS_LINUX_H
3
cca84643
JA
4#define FIO_OS os_linux
5
ebac4655 6#include <sys/ioctl.h>
8756e4d4 7#include <sys/uio.h>
3c39a379 8#include <sys/syscall.h>
2e3bd4c2 9#include <sys/vfs.h>
39b93568 10#include <sys/mman.h>
3c39a379 11#include <unistd.h>
97fbdfa1 12#include <fcntl.h>
3c2d93ed 13#include <errno.h>
d09913c2 14#include <sched.h>
ea421790 15#include <linux/unistd.h>
07e5b264
JA
16#include <linux/raw.h>
17#include <linux/major.h>
610a730c 18#include <endian.h>
ebac4655 19
f3de88a7 20#include "indirect.h"
79a43187 21#include "binject.h"
e2e58886 22#include "../file.h"
f3de88a7 23
ebac4655
JA
24#define FIO_HAVE_LIBAIO
25#define FIO_HAVE_POSIXAIO
26#define FIO_HAVE_FADVISE
27#define FIO_HAVE_CPU_AFFINITY
28#define FIO_HAVE_DISK_UTIL
29#define FIO_HAVE_SGIO
ba4f8923 30#define FIO_HAVE_IOPRIO
8756e4d4 31#define FIO_HAVE_SPLICE
22f78b32 32#define FIO_HAVE_IOSCHED_SWITCH
2c0ecd28 33#define FIO_HAVE_ODIRECT
74b025b0 34#define FIO_HAVE_HUGETLB
07e5b264 35#define FIO_HAVE_RAWBIND
5e62c22a 36#define FIO_HAVE_BLKTRACE
5921e80c 37#define FIO_HAVE_STRSEP
207cb0f0 38#define FIO_HAVE_POSIXAIO_FSYNC
f356d01d 39#define FIO_HAVE_PSHARED_MUTEX
eb7ccf38 40#define FIO_HAVE_CL_SIZE
a696fa2a 41#define FIO_HAVE_CGROUPS
c36d16f5 42#define FIO_HAVE_FDATASYNC
2e3bd4c2 43#define FIO_HAVE_FS_STAT
a5f3027c 44#define FIO_HAVE_TRIM
79a43187 45#define FIO_HAVE_BINJECT
ecc314ba 46#define FIO_HAVE_CLOCK_MONOTONIC
47f767c1 47#define FIO_HAVE_GETTID
93bcfd20 48#define FIO_USE_GENERIC_INIT_RANDOM_STATE
1ecc95ca 49#define FIO_HAVE_E4_ENG
b6959b55 50
39b93568
JA
51#ifdef MAP_HUGETLB
52#define FIO_HAVE_MMAP_HUGE
53#endif
54
804bb9a1
JA
55/*
56 * Can only enable this for newer glibcs, or the header and defines are
57 * missing
58 */
52ad4b94
JA
59#if __GLIBC__ >= 2 && __GLIBC_MINOR__ >= 6
60#define FIO_HAVE_FALLOCATE
61#endif
78b269e3 62#if __GLIBC__ >= 2 && __GLIBC_MINOR__ >= 8
804bb9a1
JA
63#define FIO_HAVE_LINUX_FALLOCATE
64#endif
65
1ecc95ca
JA
66#ifdef FIO_HAVE_LINUX_FALLOCATE
67#define FIO_HAVE_FALLOC_ENG
68#endif
69
093729f7
JA
70#ifdef SYNC_FILE_RANGE_WAIT_BEFORE
71#define FIO_HAVE_SYNC_FILE_RANGE
72#endif
73
dc873b6f 74#define OS_MAP_ANON MAP_ANONYMOUS
ebac4655 75
4d8947de
SL
76#ifndef CLOCK_MONOTONIC
77#define CLOCK_MONOTONIC 1
78#endif
79
ebac4655 80typedef cpu_set_t os_cpu_mask_t;
4e78e405 81
6dfd46b9 82typedef struct drand48_data os_random_state_t;
ebac4655
JA
83
84/*
85 * we want fadvise64 really, but it's so tangled... later
86 */
4d8947de 87#ifdef FIO_HAVE_FADVISE
ebac4655
JA
88#define fadvise(fd, off, len, advice) \
89 posix_fadvise((fd), (off_t)(off), (len), (advice))
4d8947de 90#endif
ebac4655 91
c8f025f8
JA
92/*
93 * If you are on an ancient glibc (2.3.2), then define GLIBC_2_3_2 if you want
94 * the affinity helpers to work.
95 */
96#ifndef GLIBC_2_3_2
e8462bd8
JA
97#define fio_setaffinity(pid, cpumask) \
98 sched_setaffinity((pid), sizeof(cpumask), &(cpumask))
ebac4655
JA
99#define fio_getaffinity(pid, ptr) \
100 sched_getaffinity((pid), sizeof(cpu_set_t), (ptr))
c8f025f8 101#else
e8462bd8
JA
102#define fio_setaffinity(pid, cpumask) \
103 sched_setaffinity((pid), &(cpumask))
c8f025f8
JA
104#define fio_getaffinity(pid, ptr) \
105 sched_getaffinity((pid), (ptr))
106#endif
be4ecfdf 107
1f4c0a4f
JA
108#define fio_cpu_clear(mask, cpu) (void) CPU_CLR((cpu), (mask))
109#define fio_cpu_set(mask, cpu) (void) CPU_SET((cpu), (mask))
d2ce18b5
JA
110
111static inline int fio_cpuset_init(os_cpu_mask_t *mask)
112{
113 CPU_ZERO(mask);
114 return 0;
115}
116
117static inline int fio_cpuset_exit(os_cpu_mask_t *mask)
118{
119 return 0;
120}
6d459ee7
JA
121
122#define FIO_MAX_CPUS CPU_SETSIZE
4d8947de 123
ebac4655
JA
124static inline int ioprio_set(int which, int who, int ioprio)
125{
126 return syscall(__NR_ioprio_set, which, who, ioprio);
127}
128
47f767c1
JA
129static inline int gettid(void)
130{
131 return syscall(__NR_gettid);
132}
133
97fbdfa1
JA
134/*
135 * Just check for SPLICE_F_MOVE, if that isn't there, assume the others
136 * aren't either.
137 */
138#ifndef SPLICE_F_MOVE
139#define SPLICE_F_MOVE (0x01) /* move pages instead of copying */
140#define SPLICE_F_NONBLOCK (0x02) /* don't block on the pipe splicing (but */
141 /* we may still block on the fd we splice */
142 /* from/to, of course */
143#define SPLICE_F_MORE (0x04) /* expect more data */
144#define SPLICE_F_GIFT (0x08) /* pages passed in are a gift */
8756e4d4
JA
145
146static inline int splice(int fdin, loff_t *off_in, int fdout, loff_t *off_out,
495ee9b6 147 size_t len, unsigned int flags)
8756e4d4 148{
97fbdfa1 149 return syscall(__NR_sys_splice, fdin, off_in, fdout, off_out, len, flags);
8756e4d4
JA
150}
151
152static inline int tee(int fdin, int fdout, size_t len, unsigned int flags)
153{
97fbdfa1 154 return syscall(__NR_sys_tee, fdin, fdout, len, flags);
8756e4d4
JA
155}
156
157static inline int vmsplice(int fd, const struct iovec *iov,
158 unsigned long nr_segs, unsigned int flags)
159{
97fbdfa1 160 return syscall(__NR_sys_vmsplice, fd, iov, nr_segs, flags);
8756e4d4 161}
97fbdfa1 162#endif
8756e4d4 163
3feedc60
JA
164#define SPLICE_DEF_SIZE (64*1024)
165
b6959b55 166#ifdef FIO_HAVE_SYSLET
bf0dc8fa
IM
167
168struct syslet_uatom;
169struct async_head_user;
170
a4f4fdd7
JA
171/*
172 * syslet stuff
173 */
bf0dc8fa
IM
174static inline struct syslet_uatom *
175async_exec(struct syslet_uatom *atom, struct async_head_user *ahu)
a4f4fdd7 176{
2b52511f 177 return (struct syslet_uatom *) syscall(__NR_async_exec, atom, ahu);
a4f4fdd7
JA
178}
179
bf0dc8fa
IM
180static inline long
181async_wait(unsigned long min_wait_events, unsigned long user_ring_idx,
182 struct async_head_user *ahu)
a4f4fdd7 183{
bf0dc8fa
IM
184 return syscall(__NR_async_wait, min_wait_events,
185 user_ring_idx, ahu);
a4f4fdd7
JA
186}
187
7756b0d0 188static inline long async_thread(void *event, struct async_head_user *ahu)
a4f4fdd7 189{
7756b0d0 190 return syscall(__NR_async_thread, event, ahu);
a4f4fdd7
JA
191}
192
193static inline long umem_add(unsigned long *uptr, unsigned long inc)
194{
195 return syscall(__NR_umem_add, uptr, inc);
196}
b6959b55 197#endif /* FIO_HAVE_SYSLET */
a4f4fdd7 198
6cefbe33
JA
199enum {
200 IOPRIO_CLASS_NONE,
201 IOPRIO_CLASS_RT,
202 IOPRIO_CLASS_BE,
203 IOPRIO_CLASS_IDLE,
204};
205
ebac4655
JA
206enum {
207 IOPRIO_WHO_PROCESS = 1,
208 IOPRIO_WHO_PGRP,
209 IOPRIO_WHO_USER,
210};
211
6cefbe33 212#define IOPRIO_BITS 16
ebac4655
JA
213#define IOPRIO_CLASS_SHIFT 13
214
215#ifndef BLKGETSIZE64
216#define BLKGETSIZE64 _IOR(0x12,114,size_t)
217#endif
218
e5b401d4
JA
219#ifndef BLKFLSBUF
220#define BLKFLSBUF _IO(0x12,97)
221#endif
222
a5f3027c
JA
223#ifndef BLKDISCARD
224#define BLKDISCARD _IO(0x12,119)
225#endif
226
e2e58886 227static inline int blockdev_invalidate_cache(struct fio_file *f)
e5b401d4 228{
ecc314ba 229 return ioctl(f->fd, BLKFLSBUF);
e5b401d4
JA
230}
231
ecc314ba 232static inline int blockdev_size(struct fio_file *f, unsigned long long *bytes)
ebac4655 233{
ecc314ba 234 if (!ioctl(f->fd, BLKGETSIZE64, bytes))
ebac4655
JA
235 return 0;
236
237 return errno;
238}
239
32cd46a0
JA
240static inline unsigned long long os_phys_mem(void)
241{
242 long pagesize, pages;
243
244 pagesize = sysconf(_SC_PAGESIZE);
245 pages = sysconf(_SC_PHYS_PAGES);
246 if (pages == -1 || pagesize == -1)
247 return 0;
248
249 return (unsigned long long) pages * (unsigned long long) pagesize;
250}
251
6dfd46b9
JA
252static inline void os_random_seed(unsigned long seed, os_random_state_t *rs)
253{
254 srand48_r(seed, rs);
255}
256
257static inline long os_random_long(os_random_state_t *rs)
258{
259 long val;
260
261 lrand48_r(rs, &val);
262 return val;
263}
264
8cc7afa9 265static inline int fio_lookup_raw(dev_t dev, int *majdev, int *mindev)
07e5b264
JA
266{
267 struct raw_config_request rq;
268 int fd;
269
270 if (major(dev) != RAW_MAJOR)
8cc7afa9 271 return 1;
07e5b264
JA
272
273 /*
274 * we should be able to find /dev/rawctl or /dev/raw/rawctl
275 */
276 fd = open("/dev/rawctl", O_RDONLY);
277 if (fd < 0) {
278 fd = open("/dev/raw/rawctl", O_RDONLY);
279 if (fd < 0)
8cc7afa9 280 return 1;
07e5b264
JA
281 }
282
283 rq.raw_minor = minor(dev);
284 if (ioctl(fd, RAW_GETBIND, &rq) < 0) {
285 close(fd);
8cc7afa9 286 return 1;
07e5b264
JA
287 }
288
289 close(fd);
290 *majdev = rq.block_major;
291 *mindev = rq.block_minor;
8cc7afa9 292 return 0;
07e5b264
JA
293}
294
5b6f5c66 295#ifdef O_NOATIME
5921e80c 296#define FIO_O_NOATIME O_NOATIME
5b6f5c66
JA
297#else
298#define FIO_O_NOATIME 0
299#endif
5921e80c 300
a1c58075
JA
301#ifdef MADV_REMOVE
302#define FIO_MADV_FREE MADV_REMOVE
303#endif
304
610a730c
JA
305#if __BYTE_ORDER == __LITTLE_ENDIAN
306#define FIO_LITTLE_ENDIAN
307#elif __BYTE_ORDER == __BIG_ENDIAN
308#define FIO_BIG_ENDIAN
309#else
310#error "Unknown endianness"
311#endif
312
ff245192
JA
313#define fio_swap16(x) __bswap_16(x)
314#define fio_swap32(x) __bswap_32(x)
315#define fio_swap64(x) __bswap_64(x)
316
eb7ccf38
JA
317#define CACHE_LINE_FILE \
318 "/sys/devices/system/cpu/cpu0/cache/index0/coherency_line_size"
319
320static inline int arch_cache_line_size(void)
321{
322 char size[32];
323 int fd, ret;
324
325 fd = open(CACHE_LINE_FILE, O_RDONLY);
326 if (fd < 0)
327 return -1;
328
329 ret = read(fd, size, sizeof(size));
ec76f15a
JA
330
331 close(fd);
332
eb7ccf38
JA
333 if (ret <= 0)
334 return -1;
ec76f15a
JA
335 else
336 return atoi(size);
eb7ccf38
JA
337}
338
2e3bd4c2
JA
339static inline unsigned long long get_fs_size(const char *path)
340{
341 unsigned long long ret;
342 struct statfs s;
343
344 if (statfs(path, &s) < 0)
345 return -1ULL;
346
347 ret = s.f_bsize;
348 ret *= (unsigned long long) s.f_bfree;
349 return ret;
350}
351
a5f3027c
JA
352static inline int os_trim(int fd, unsigned long long start,
353 unsigned long long len)
354{
355 uint64_t range[2];
356
357 range[0] = start;
358 range[1] = len;
359
360 if (!ioctl(fd, BLKDISCARD, range))
361 return 0;
362
363 return errno;
364}
365
ebac4655 366#endif