Commit | Line | Data |
---|---|---|
ebac4655 JA |
1 | #ifndef FIO_OS_LINUX_H |
2 | #define FIO_OS_LINUX_H | |
3 | ||
4 | #include <sys/ioctl.h> | |
8756e4d4 | 5 | #include <sys/uio.h> |
3c39a379 JA |
6 | #include <sys/syscall.h> |
7 | #include <unistd.h> | |
97fbdfa1 | 8 | #include <fcntl.h> |
3c2d93ed | 9 | #include <errno.h> |
d09913c2 | 10 | #include <sched.h> |
ea421790 | 11 | #include <linux/unistd.h> |
07e5b264 JA |
12 | #include <linux/raw.h> |
13 | #include <linux/major.h> | |
ebac4655 | 14 | |
f3de88a7 JA |
15 | #include "indirect.h" |
16 | ||
ebac4655 JA |
17 | #define FIO_HAVE_LIBAIO |
18 | #define FIO_HAVE_POSIXAIO | |
19 | #define FIO_HAVE_FADVISE | |
20 | #define FIO_HAVE_CPU_AFFINITY | |
21 | #define FIO_HAVE_DISK_UTIL | |
22 | #define FIO_HAVE_SGIO | |
ba4f8923 | 23 | #define FIO_HAVE_IOPRIO |
8756e4d4 | 24 | #define FIO_HAVE_SPLICE |
22f78b32 | 25 | #define FIO_HAVE_IOSCHED_SWITCH |
2c0ecd28 | 26 | #define FIO_HAVE_ODIRECT |
74b025b0 | 27 | #define FIO_HAVE_HUGETLB |
07e5b264 | 28 | #define FIO_HAVE_RAWBIND |
5e62c22a | 29 | #define FIO_HAVE_BLKTRACE |
5921e80c | 30 | #define FIO_HAVE_STRSEP |
fffca02d | 31 | #define FIO_HAVE_FALLOCATE |
207cb0f0 | 32 | #define FIO_HAVE_POSIXAIO_FSYNC |
f356d01d | 33 | #define FIO_HAVE_PSHARED_MUTEX |
eb7ccf38 | 34 | #define FIO_HAVE_CL_SIZE |
a696fa2a | 35 | #define FIO_HAVE_CGROUPS |
c36d16f5 | 36 | #define FIO_HAVE_FDATASYNC |
44f29692 | 37 | #define FIO_HAVE_SYNC_FILE_RANGE |
b6959b55 | 38 | |
dc873b6f | 39 | #define OS_MAP_ANON MAP_ANONYMOUS |
ebac4655 | 40 | |
4d8947de SL |
41 | #ifndef CLOCK_MONOTONIC |
42 | #define CLOCK_MONOTONIC 1 | |
43 | #endif | |
44 | ||
ebac4655 | 45 | typedef cpu_set_t os_cpu_mask_t; |
4e78e405 | 46 | |
6dfd46b9 | 47 | typedef struct drand48_data os_random_state_t; |
ebac4655 JA |
48 | |
49 | /* | |
50 | * we want fadvise64 really, but it's so tangled... later | |
51 | */ | |
4d8947de | 52 | #ifdef FIO_HAVE_FADVISE |
ebac4655 JA |
53 | #define fadvise(fd, off, len, advice) \ |
54 | posix_fadvise((fd), (off_t)(off), (len), (advice)) | |
4d8947de | 55 | #endif |
ebac4655 | 56 | |
c8f025f8 JA |
57 | /* |
58 | * If you are on an ancient glibc (2.3.2), then define GLIBC_2_3_2 if you want | |
59 | * the affinity helpers to work. | |
60 | */ | |
61 | #ifndef GLIBC_2_3_2 | |
e8462bd8 JA |
62 | #define fio_setaffinity(pid, cpumask) \ |
63 | sched_setaffinity((pid), sizeof(cpumask), &(cpumask)) | |
ebac4655 JA |
64 | #define fio_getaffinity(pid, ptr) \ |
65 | sched_getaffinity((pid), sizeof(cpu_set_t), (ptr)) | |
c8f025f8 | 66 | #else |
e8462bd8 JA |
67 | #define fio_setaffinity(pid, cpumask) \ |
68 | sched_setaffinity((pid), &(cpumask)) | |
c8f025f8 JA |
69 | #define fio_getaffinity(pid, ptr) \ |
70 | sched_getaffinity((pid), (ptr)) | |
71 | #endif | |
be4ecfdf JA |
72 | |
73 | #define fio_cpu_clear(mask, cpu) CPU_CLR((cpu), (mask)) | |
6d459ee7 | 74 | #define fio_cpu_set(mask, cpu) CPU_SET((cpu), (mask)) |
d2ce18b5 JA |
75 | |
76 | static inline int fio_cpuset_init(os_cpu_mask_t *mask) | |
77 | { | |
78 | CPU_ZERO(mask); | |
79 | return 0; | |
80 | } | |
81 | ||
82 | static inline int fio_cpuset_exit(os_cpu_mask_t *mask) | |
83 | { | |
84 | return 0; | |
85 | } | |
6d459ee7 JA |
86 | |
87 | #define FIO_MAX_CPUS CPU_SETSIZE | |
4d8947de | 88 | |
ebac4655 JA |
89 | static inline int ioprio_set(int which, int who, int ioprio) |
90 | { | |
91 | return syscall(__NR_ioprio_set, which, who, ioprio); | |
92 | } | |
93 | ||
97fbdfa1 JA |
94 | /* |
95 | * Just check for SPLICE_F_MOVE, if that isn't there, assume the others | |
96 | * aren't either. | |
97 | */ | |
98 | #ifndef SPLICE_F_MOVE | |
99 | #define SPLICE_F_MOVE (0x01) /* move pages instead of copying */ | |
100 | #define SPLICE_F_NONBLOCK (0x02) /* don't block on the pipe splicing (but */ | |
101 | /* we may still block on the fd we splice */ | |
102 | /* from/to, of course */ | |
103 | #define SPLICE_F_MORE (0x04) /* expect more data */ | |
104 | #define SPLICE_F_GIFT (0x08) /* pages passed in are a gift */ | |
8756e4d4 JA |
105 | |
106 | static inline int splice(int fdin, loff_t *off_in, int fdout, loff_t *off_out, | |
495ee9b6 | 107 | size_t len, unsigned int flags) |
8756e4d4 | 108 | { |
97fbdfa1 | 109 | return syscall(__NR_sys_splice, fdin, off_in, fdout, off_out, len, flags); |
8756e4d4 JA |
110 | } |
111 | ||
112 | static inline int tee(int fdin, int fdout, size_t len, unsigned int flags) | |
113 | { | |
97fbdfa1 | 114 | return syscall(__NR_sys_tee, fdin, fdout, len, flags); |
8756e4d4 JA |
115 | } |
116 | ||
117 | static inline int vmsplice(int fd, const struct iovec *iov, | |
118 | unsigned long nr_segs, unsigned int flags) | |
119 | { | |
97fbdfa1 | 120 | return syscall(__NR_sys_vmsplice, fd, iov, nr_segs, flags); |
8756e4d4 | 121 | } |
97fbdfa1 | 122 | #endif |
8756e4d4 | 123 | |
3feedc60 JA |
124 | #define SPLICE_DEF_SIZE (64*1024) |
125 | ||
b6959b55 | 126 | #ifdef FIO_HAVE_SYSLET |
bf0dc8fa IM |
127 | |
128 | struct syslet_uatom; | |
129 | struct async_head_user; | |
130 | ||
a4f4fdd7 JA |
131 | /* |
132 | * syslet stuff | |
133 | */ | |
bf0dc8fa IM |
134 | static inline struct syslet_uatom * |
135 | async_exec(struct syslet_uatom *atom, struct async_head_user *ahu) | |
a4f4fdd7 | 136 | { |
2b52511f | 137 | return (struct syslet_uatom *) syscall(__NR_async_exec, atom, ahu); |
a4f4fdd7 JA |
138 | } |
139 | ||
bf0dc8fa IM |
140 | static inline long |
141 | async_wait(unsigned long min_wait_events, unsigned long user_ring_idx, | |
142 | struct async_head_user *ahu) | |
a4f4fdd7 | 143 | { |
bf0dc8fa IM |
144 | return syscall(__NR_async_wait, min_wait_events, |
145 | user_ring_idx, ahu); | |
a4f4fdd7 JA |
146 | } |
147 | ||
7756b0d0 | 148 | static inline long async_thread(void *event, struct async_head_user *ahu) |
a4f4fdd7 | 149 | { |
7756b0d0 | 150 | return syscall(__NR_async_thread, event, ahu); |
a4f4fdd7 JA |
151 | } |
152 | ||
153 | static inline long umem_add(unsigned long *uptr, unsigned long inc) | |
154 | { | |
155 | return syscall(__NR_umem_add, uptr, inc); | |
156 | } | |
b6959b55 | 157 | #endif /* FIO_HAVE_SYSLET */ |
a4f4fdd7 | 158 | |
6cefbe33 JA |
159 | enum { |
160 | IOPRIO_CLASS_NONE, | |
161 | IOPRIO_CLASS_RT, | |
162 | IOPRIO_CLASS_BE, | |
163 | IOPRIO_CLASS_IDLE, | |
164 | }; | |
165 | ||
ebac4655 JA |
166 | enum { |
167 | IOPRIO_WHO_PROCESS = 1, | |
168 | IOPRIO_WHO_PGRP, | |
169 | IOPRIO_WHO_USER, | |
170 | }; | |
171 | ||
6cefbe33 | 172 | #define IOPRIO_BITS 16 |
ebac4655 JA |
173 | #define IOPRIO_CLASS_SHIFT 13 |
174 | ||
175 | #ifndef BLKGETSIZE64 | |
176 | #define BLKGETSIZE64 _IOR(0x12,114,size_t) | |
177 | #endif | |
178 | ||
e5b401d4 JA |
179 | #ifndef BLKFLSBUF |
180 | #define BLKFLSBUF _IO(0x12,97) | |
181 | #endif | |
182 | ||
183 | static inline int blockdev_invalidate_cache(int fd) | |
184 | { | |
a16211bb | 185 | return ioctl(fd, BLKFLSBUF); |
e5b401d4 JA |
186 | } |
187 | ||
9104f874 | 188 | static inline int blockdev_size(int fd, unsigned long long *bytes) |
ebac4655 JA |
189 | { |
190 | if (!ioctl(fd, BLKGETSIZE64, bytes)) | |
191 | return 0; | |
192 | ||
193 | return errno; | |
194 | } | |
195 | ||
32cd46a0 JA |
196 | static inline unsigned long long os_phys_mem(void) |
197 | { | |
198 | long pagesize, pages; | |
199 | ||
200 | pagesize = sysconf(_SC_PAGESIZE); | |
201 | pages = sysconf(_SC_PHYS_PAGES); | |
202 | if (pages == -1 || pagesize == -1) | |
203 | return 0; | |
204 | ||
205 | return (unsigned long long) pages * (unsigned long long) pagesize; | |
206 | } | |
207 | ||
6dfd46b9 JA |
208 | static inline void os_random_seed(unsigned long seed, os_random_state_t *rs) |
209 | { | |
210 | srand48_r(seed, rs); | |
211 | } | |
212 | ||
213 | static inline long os_random_long(os_random_state_t *rs) | |
214 | { | |
215 | long val; | |
216 | ||
217 | lrand48_r(rs, &val); | |
218 | return val; | |
219 | } | |
220 | ||
8cc7afa9 | 221 | static inline int fio_lookup_raw(dev_t dev, int *majdev, int *mindev) |
07e5b264 JA |
222 | { |
223 | struct raw_config_request rq; | |
224 | int fd; | |
225 | ||
226 | if (major(dev) != RAW_MAJOR) | |
8cc7afa9 | 227 | return 1; |
07e5b264 JA |
228 | |
229 | /* | |
230 | * we should be able to find /dev/rawctl or /dev/raw/rawctl | |
231 | */ | |
232 | fd = open("/dev/rawctl", O_RDONLY); | |
233 | if (fd < 0) { | |
234 | fd = open("/dev/raw/rawctl", O_RDONLY); | |
235 | if (fd < 0) | |
8cc7afa9 | 236 | return 1; |
07e5b264 JA |
237 | } |
238 | ||
239 | rq.raw_minor = minor(dev); | |
240 | if (ioctl(fd, RAW_GETBIND, &rq) < 0) { | |
241 | close(fd); | |
8cc7afa9 | 242 | return 1; |
07e5b264 JA |
243 | } |
244 | ||
245 | close(fd); | |
246 | *majdev = rq.block_major; | |
247 | *mindev = rq.block_minor; | |
8cc7afa9 | 248 | return 0; |
07e5b264 JA |
249 | } |
250 | ||
5b6f5c66 | 251 | #ifdef O_NOATIME |
5921e80c | 252 | #define FIO_O_NOATIME O_NOATIME |
5b6f5c66 JA |
253 | #else |
254 | #define FIO_O_NOATIME 0 | |
255 | #endif | |
5921e80c | 256 | |
a1c58075 JA |
257 | #ifdef MADV_REMOVE |
258 | #define FIO_MADV_FREE MADV_REMOVE | |
259 | #endif | |
260 | ||
eb7ccf38 JA |
261 | #define CACHE_LINE_FILE \ |
262 | "/sys/devices/system/cpu/cpu0/cache/index0/coherency_line_size" | |
263 | ||
264 | static inline int arch_cache_line_size(void) | |
265 | { | |
266 | char size[32]; | |
267 | int fd, ret; | |
268 | ||
269 | fd = open(CACHE_LINE_FILE, O_RDONLY); | |
270 | if (fd < 0) | |
271 | return -1; | |
272 | ||
273 | ret = read(fd, size, sizeof(size)); | |
ec76f15a JA |
274 | |
275 | close(fd); | |
276 | ||
eb7ccf38 JA |
277 | if (ret <= 0) |
278 | return -1; | |
ec76f15a JA |
279 | else |
280 | return atoi(size); | |
eb7ccf38 JA |
281 | } |
282 | ||
ebac4655 | 283 | #endif |