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