backend: clear IO_U_F_FLIGHT flag in zero byte read path
[fio.git] / os / os-linux.h
CommitLineData
ebac4655
JA
1#ifndef FIO_OS_LINUX_H
2#define FIO_OS_LINUX_H
3
910f378a
BVA
4#ifdef __ANDROID__
5#define FIO_OS os_android
6#else
cca84643 7#define FIO_OS os_linux
910f378a 8#endif
cca84643 9
ebac4655 10#include <sys/ioctl.h>
8756e4d4 11#include <sys/uio.h>
3c39a379 12#include <sys/syscall.h>
250e878a 13#include <sys/sysmacros.h>
2e3bd4c2 14#include <sys/vfs.h>
39b93568 15#include <sys/mman.h>
3c39a379 16#include <unistd.h>
97fbdfa1 17#include <fcntl.h>
3c2d93ed 18#include <errno.h>
d09913c2 19#include <sched.h>
ea421790 20#include <linux/unistd.h>
07e5b264 21#include <linux/major.h>
1894d171
TK
22#include <linux/fs.h>
23#include <scsi/sg.h>
e69d8385 24#include <asm/byteorder.h>
910f378a
BVA
25#ifdef __ANDROID__
26#include "os-ashmem.h"
27#define FIO_NO_HAVE_SHM_H
28#endif
ebac4655 29
58828d07
SW
30#ifdef ARCH_HAVE_CRC_CRYPTO
31#include <sys/auxv.h>
3721c7fe
SW
32#ifndef HWCAP_PMULL
33#define HWCAP_PMULL (1 << 4)
34#endif /* HWCAP_PMULL */
58828d07
SW
35#ifndef HWCAP_CRC32
36#define HWCAP_CRC32 (1 << 7)
37#endif /* HWCAP_CRC32 */
38#endif /* ARCH_HAVE_CRC_CRYPTO */
39
1c764dbe 40#include "./os-linux-syscall.h"
e2e58886 41#include "../file.h"
f3de88a7 42
7922a7b7
OM
43#ifndef __has_builtin // Optional of course.
44 #define __has_builtin(x) 0 // Compatibility with non-clang compilers.
45#endif
46
ebac4655
JA
47#define FIO_HAVE_CPU_AFFINITY
48#define FIO_HAVE_DISK_UTIL
49#define FIO_HAVE_SGIO
ba4f8923 50#define FIO_HAVE_IOPRIO
32ef447a 51#define FIO_HAVE_IOPRIO_CLASS
22f78b32 52#define FIO_HAVE_IOSCHED_SWITCH
2c0ecd28 53#define FIO_HAVE_ODIRECT
74b025b0 54#define FIO_HAVE_HUGETLB
5e62c22a 55#define FIO_HAVE_BLKTRACE
eb7ccf38 56#define FIO_HAVE_CL_SIZE
a696fa2a 57#define FIO_HAVE_CGROUPS
2e3bd4c2 58#define FIO_HAVE_FS_STAT
a5f3027c 59#define FIO_HAVE_TRIM
47f767c1 60#define FIO_HAVE_GETTID
93bcfd20 61#define FIO_USE_GENERIC_INIT_RANDOM_STATE
e69d8385 62#define FIO_HAVE_BYTEORDER_FUNCS
6562685f 63#define FIO_HAVE_PWRITEV2
97900ebf 64#define FIO_HAVE_SHM_ATTACH_REMOVED
b6959b55 65
39b93568
JA
66#ifdef MAP_HUGETLB
67#define FIO_HAVE_MMAP_HUGE
68#endif
69
dc873b6f 70#define OS_MAP_ANON MAP_ANONYMOUS
ebac4655 71
4db9ac71 72#define FIO_EXT_ENG_DIR "/usr/local/lib/fio"
5a8a6a03 73
ebac4655 74typedef cpu_set_t os_cpu_mask_t;
4e78e405 75
67bf9823 76#ifdef CONFIG_3ARG_AFFINITY
e8462bd8
JA
77#define fio_setaffinity(pid, cpumask) \
78 sched_setaffinity((pid), sizeof(cpumask), &(cpumask))
ebac4655
JA
79#define fio_getaffinity(pid, ptr) \
80 sched_getaffinity((pid), sizeof(cpu_set_t), (ptr))
67bf9823 81#elif defined(CONFIG_2ARG_AFFINITY)
e8462bd8
JA
82#define fio_setaffinity(pid, cpumask) \
83 sched_setaffinity((pid), &(cpumask))
c8f025f8
JA
84#define fio_getaffinity(pid, ptr) \
85 sched_getaffinity((pid), (ptr))
86#endif
be4ecfdf 87
e9d2a04d
TK
88#ifdef CONFIG_PTHREAD_GETAFFINITY
89#define FIO_HAVE_GET_THREAD_AFFINITY
874d55e5
JA
90#define fio_get_thread_affinity(mask) \
91 pthread_getaffinity_np(pthread_self(), sizeof(mask), &(mask))
e9d2a04d 92#endif
874d55e5 93
6e0cefdc
BVA
94#define fio_cpu_clear(mask, cpu) CPU_CLR((cpu), (mask))
95#define fio_cpu_set(mask, cpu) CPU_SET((cpu), (mask))
52fd65f4 96#define fio_cpu_isset(mask, cpu) (CPU_ISSET((cpu), (mask)) != 0)
d004a209 97#define fio_cpu_count(mask) CPU_COUNT((mask))
d2ce18b5
JA
98
99static inline int fio_cpuset_init(os_cpu_mask_t *mask)
100{
101 CPU_ZERO(mask);
102 return 0;
103}
104
105static inline int fio_cpuset_exit(os_cpu_mask_t *mask)
106{
107 return 0;
108}
6d459ee7
JA
109
110#define FIO_MAX_CPUS CPU_SETSIZE
4d8947de 111
28727df7
JA
112enum {
113 IOPRIO_CLASS_NONE,
114 IOPRIO_CLASS_RT,
115 IOPRIO_CLASS_BE,
116 IOPRIO_CLASS_IDLE,
117};
118
119enum {
120 IOPRIO_WHO_PROCESS = 1,
121 IOPRIO_WHO_PGRP,
122 IOPRIO_WHO_USER,
123};
124
125#define IOPRIO_BITS 16
126#define IOPRIO_CLASS_SHIFT 13
127
1767bd34
TK
128#define IOPRIO_MIN_PRIO 0 /* highest priority */
129#define IOPRIO_MAX_PRIO 7 /* lowest priority */
130
131#define IOPRIO_MIN_PRIO_CLASS 0
132#define IOPRIO_MAX_PRIO_CLASS 3
133
8d6e8d99 134static inline int ioprio_value(int ioprio_class, int ioprio)
ebac4655 135{
28727df7
JA
136 /*
137 * If no class is set, assume BE
138 */
8d6e8d99
DLM
139 if (!ioprio_class)
140 ioprio_class = IOPRIO_CLASS_BE;
141
142 return (ioprio_class << IOPRIO_CLASS_SHIFT) | ioprio;
143}
28727df7 144
03ec570f
DLM
145static inline bool ioprio_value_is_class_rt(unsigned int priority)
146{
147 return (priority >> IOPRIO_CLASS_SHIFT) == IOPRIO_CLASS_RT;
148}
149
8d6e8d99
DLM
150static inline int ioprio_set(int which, int who, int ioprio_class, int ioprio)
151{
152 return syscall(__NR_ioprio_set, which, who,
153 ioprio_value(ioprio_class, ioprio));
ebac4655
JA
154}
155
a967e54d
DLM
156#define ioprio_class(ioprio) ((ioprio) >> IOPRIO_CLASS_SHIFT)
157#define ioprio(ioprio) ((ioprio) & 7)
158
de5ed0e4 159#ifndef CONFIG_HAVE_GETTID
47f767c1
JA
160static inline int gettid(void)
161{
162 return syscall(__NR_gettid);
163}
de5ed0e4 164#endif
47f767c1 165
3feedc60
JA
166#define SPLICE_DEF_SIZE (64*1024)
167
ebac4655
JA
168#ifndef BLKGETSIZE64
169#define BLKGETSIZE64 _IOR(0x12,114,size_t)
170#endif
171
e5b401d4
JA
172#ifndef BLKFLSBUF
173#define BLKFLSBUF _IO(0x12,97)
174#endif
175
a5f3027c
JA
176#ifndef BLKDISCARD
177#define BLKDISCARD _IO(0x12,119)
178#endif
179
e2e58886 180static inline int blockdev_invalidate_cache(struct fio_file *f)
e5b401d4 181{
ecc314ba 182 return ioctl(f->fd, BLKFLSBUF);
e5b401d4
JA
183}
184
ecc314ba 185static inline int blockdev_size(struct fio_file *f, unsigned long long *bytes)
ebac4655 186{
ecc314ba 187 if (!ioctl(f->fd, BLKGETSIZE64, bytes))
ebac4655
JA
188 return 0;
189
190 return errno;
191}
192
32cd46a0
JA
193static inline unsigned long long os_phys_mem(void)
194{
195 long pagesize, pages;
196
197 pagesize = sysconf(_SC_PAGESIZE);
198 pages = sysconf(_SC_PHYS_PAGES);
199 if (pages == -1 || pagesize == -1)
200 return 0;
201
202 return (unsigned long long) pages * (unsigned long long) pagesize;
203}
204
5b6f5c66 205#ifdef O_NOATIME
5921e80c 206#define FIO_O_NOATIME O_NOATIME
5b6f5c66
JA
207#else
208#define FIO_O_NOATIME 0
209#endif
5921e80c 210
a1c58075
JA
211#ifdef MADV_REMOVE
212#define FIO_MADV_FREE MADV_REMOVE
213#endif
214
7922a7b7
OM
215/* Check for GCC or Clang byte swap intrinsics */
216#if (__has_builtin(__builtin_bswap16) && __has_builtin(__builtin_bswap32) \
217 && __has_builtin(__builtin_bswap64)) || (__GNUC__ > 4 \
218 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 8)) /* fio_swapN */
df73aab1 219#define fio_swap16(x) __builtin_bswap16(x)
df73aab1 220#define fio_swap32(x) __builtin_bswap32(x)
df73aab1
JA
221#define fio_swap64(x) __builtin_bswap64(x)
222#else
7922a7b7
OM
223#include <byteswap.h>
224#define fio_swap16(x) bswap_16(x)
225#define fio_swap32(x) bswap_32(x)
226#define fio_swap64(x) bswap_64(x)
227#endif /* fio_swapN */
ff245192 228
eb7ccf38
JA
229#define CACHE_LINE_FILE \
230 "/sys/devices/system/cpu/cpu0/cache/index0/coherency_line_size"
231
232static inline int arch_cache_line_size(void)
233{
234 char size[32];
235 int fd, ret;
236
237 fd = open(CACHE_LINE_FILE, O_RDONLY);
238 if (fd < 0)
239 return -1;
240
241 ret = read(fd, size, sizeof(size));
ec76f15a
JA
242
243 close(fd);
244
eb7ccf38
JA
245 if (ret <= 0)
246 return -1;
ec76f15a
JA
247 else
248 return atoi(size);
eb7ccf38
JA
249}
250
c08ad04c 251static inline unsigned long long get_fs_free_size(const char *path)
2e3bd4c2
JA
252{
253 unsigned long long ret;
254 struct statfs s;
255
256 if (statfs(path, &s) < 0)
257 return -1ULL;
258
259 ret = s.f_bsize;
260 ret *= (unsigned long long) s.f_bfree;
261 return ret;
262}
263
496b1f9e 264static inline int os_trim(struct fio_file *f, unsigned long long start,
a5f3027c
JA
265 unsigned long long len)
266{
267 uint64_t range[2];
268
269 range[0] = start;
270 range[1] = len;
271
496b1f9e 272 if (!ioctl(f->fd, BLKDISCARD, range))
a5f3027c
JA
273 return 0;
274
275 return errno;
276}
277
7e09a9f1 278#ifdef CONFIG_SCHED_IDLE
f2a2ce0e
HL
279static inline int fio_set_sched_idle(void)
280{
281 struct sched_param p = { .sched_priority = 0, };
282 return sched_setscheduler(gettid(), SCHED_IDLE, &p);
283}
7e09a9f1 284#endif
f2a2ce0e 285
ae8e559e
JA
286#ifndef F_GET_RW_HINT
287#ifndef F_LINUX_SPECIFIC_BASE
288#define F_LINUX_SPECIFIC_BASE 1024
289#endif
290#define F_GET_RW_HINT (F_LINUX_SPECIFIC_BASE + 11)
291#define F_SET_RW_HINT (F_LINUX_SPECIFIC_BASE + 12)
bd553af6
JA
292#define F_GET_FILE_RW_HINT (F_LINUX_SPECIFIC_BASE + 13)
293#define F_SET_FILE_RW_HINT (F_LINUX_SPECIFIC_BASE + 14)
ae8e559e
JA
294#endif
295
296#ifndef RWH_WRITE_LIFE_NONE
bd553af6
JA
297#define RWH_WRITE_LIFE_NOT_SET 0
298#define RWH_WRITE_LIFE_NONE 1
299#define RWH_WRITE_LIFE_SHORT 2
300#define RWH_WRITE_LIFE_MEDIUM 3
301#define RWH_WRITE_LIFE_LONG 4
302#define RWH_WRITE_LIFE_EXTREME 5
37659335
JA
303#endif
304
ae8e559e 305#define FIO_HAVE_WRITE_HINT
37659335 306
6562685f
JA
307#ifndef RWF_HIPRI
308#define RWF_HIPRI 0x00000001
309#endif
310#ifndef RWF_DSYNC
311#define RWF_DSYNC 0x00000002
312#endif
313#ifndef RWF_SYNC
314#define RWF_SYNC 0x00000004
315#endif
7d42e66e
KK
316#ifndef RWF_NOWAIT
317#define RWF_NOWAIT 0x00000008
318#endif
6562685f 319
4a87b584 320#ifndef RWF_UNCACHED
8beee62d 321#define RWF_UNCACHED 0x00000040
4a87b584
JA
322#endif
323
72cd4146
JA
324#ifndef RWF_WRITE_LIFE_SHIFT
325#define RWF_WRITE_LIFE_SHIFT 4
326#define RWF_WRITE_LIFE_SHORT (1 << RWF_WRITE_LIFE_SHIFT)
327#define RWF_WRITE_LIFE_MEDIUM (2 << RWF_WRITE_LIFE_SHIFT)
328#define RWF_WRITE_LIFE_LONG (3 << RWF_WRITE_LIFE_SHIFT)
329#define RWF_WRITE_LIFE_EXTREME (4 << RWF_WRITE_LIFE_SHIFT)
330#endif
331
6562685f
JA
332#ifndef CONFIG_PWRITEV2
333#ifdef __NR_preadv2
334static inline void make_pos_h_l(unsigned long *pos_h, unsigned long *pos_l,
335 off_t offset)
336{
7b5c648f 337#if BITS_PER_LONG == 64
7b5c648f
JA
338 *pos_l = offset;
339 *pos_h = 0;
340#else
6562685f
JA
341 *pos_l = offset & 0xffffffff;
342 *pos_h = ((uint64_t) offset) >> 32;
7b5c648f 343#endif
6562685f
JA
344}
345static inline ssize_t preadv2(int fd, const struct iovec *iov, int iovcnt,
346 off_t offset, unsigned int flags)
347{
348 unsigned long pos_l, pos_h;
349
350 make_pos_h_l(&pos_h, &pos_l, offset);
351 return syscall(__NR_preadv2, fd, iov, iovcnt, pos_l, pos_h, flags);
352}
353static inline ssize_t pwritev2(int fd, const struct iovec *iov, int iovcnt,
354 off_t offset, unsigned int flags)
355{
356 unsigned long pos_l, pos_h;
357
358 make_pos_h_l(&pos_h, &pos_l, offset);
359 return syscall(__NR_pwritev2, fd, iov, iovcnt, pos_l, pos_h, flags);
360}
361#else
362static inline ssize_t preadv2(int fd, const struct iovec *iov, int iovcnt,
363 off_t offset, unsigned int flags)
364{
365 errno = ENOSYS;
366 return -1;
367}
368static inline ssize_t pwritev2(int fd, const struct iovec *iov, int iovcnt,
369 off_t offset, unsigned int flags)
370{
371 errno = ENOSYS;
372 return -1;
373}
374#endif /* __NR_preadv2 */
375#endif /* CONFIG_PWRITEV2 */
376
0cfe2089
TK
377static inline int shm_attach_to_open_removed(void)
378{
379 return 1;
380}
381
2c3e17be
SW
382#ifdef CONFIG_LINUX_FALLOCATE
383#define FIO_HAVE_NATIVE_FALLOCATE
384static inline bool fio_fallocate(struct fio_file *f, uint64_t offset,
385 uint64_t len)
386{
387 int ret;
e69dfc2b 388 ret = fallocate(f->fd, 0, offset, len);
2c3e17be
SW
389 if (ret == 0)
390 return true;
391
392 /* Work around buggy old glibc versions... */
393 if (ret > 0)
394 errno = ret;
395
396 return false;
397}
398#endif
399
58828d07
SW
400#define FIO_HAVE_CPU_HAS
401static inline bool os_cpu_has(cpu_features feature)
402{
403 bool have_feature;
404 unsigned long fio_unused hwcap;
405
406 switch (feature) {
407#ifdef ARCH_HAVE_CRC_CRYPTO
408 case CPU_ARM64_CRC32C:
409 hwcap = getauxval(AT_HWCAP);
3721c7fe
SW
410 have_feature = (hwcap & (HWCAP_PMULL | HWCAP_CRC32)) ==
411 (HWCAP_PMULL | HWCAP_CRC32);
58828d07
SW
412 break;
413#endif
414 default:
415 have_feature = false;
416 }
417
418 return have_feature;
419}
420
ebac4655 421#endif