Commit | Line | Data |
---|---|---|
52885fa2 | 1 | /* |
bffad86f | 2 | * io_uring engine |
52885fa2 | 3 | * |
bffad86f | 4 | * IO engine using the new native Linux aio io_uring interface. See: |
a90cd050 | 5 | * |
bffad86f | 6 | * http://git.kernel.dk/cgit/linux-block/log/?h=io_uring |
52885fa2 JA |
7 | * |
8 | */ | |
9 | #include <stdlib.h> | |
10 | #include <unistd.h> | |
11 | #include <errno.h> | |
52885fa2 JA |
12 | #include <sys/time.h> |
13 | #include <sys/resource.h> | |
14 | ||
15 | #include "../fio.h" | |
16 | #include "../lib/pow2.h" | |
17 | #include "../optgroup.h" | |
18 | #include "../lib/memalign.h" | |
b87aa01a | 19 | #include "../lib/fls.h" |
6d975f2c | 20 | #include "../lib/roundup.h" |
52885fa2 | 21 | |
bffad86f | 22 | #ifdef ARCH_HAVE_IOURING |
52885fa2 | 23 | |
57fa61f0 | 24 | #include "../lib/types.h" |
f3e769a4 | 25 | #include "../os/linux/io_uring.h" |
e9f6567a | 26 | #include "cmdprio.h" |
16be6037 | 27 | #include "zbd.h" |
855dc4d4 AG |
28 | #include "nvme.h" |
29 | ||
30 | #include <sys/stat.h> | |
31 | ||
32 | enum uring_cmd_type { | |
33 | FIO_URING_CMD_NVME = 1, | |
34 | }; | |
9a2d78b3 | 35 | |
bffad86f | 36 | struct io_sq_ring { |
e2239016 JA |
37 | unsigned *head; |
38 | unsigned *tail; | |
39 | unsigned *ring_mask; | |
40 | unsigned *ring_entries; | |
41 | unsigned *flags; | |
42 | unsigned *array; | |
52885fa2 JA |
43 | }; |
44 | ||
bffad86f | 45 | struct io_cq_ring { |
e2239016 JA |
46 | unsigned *head; |
47 | unsigned *tail; | |
48 | unsigned *ring_mask; | |
49 | unsigned *ring_entries; | |
f0403f94 | 50 | struct io_uring_cqe *cqes; |
9a2d78b3 JA |
51 | }; |
52 | ||
bffad86f | 53 | struct ioring_mmap { |
9a2d78b3 JA |
54 | void *ptr; |
55 | size_t len; | |
52885fa2 JA |
56 | }; |
57 | ||
bffad86f | 58 | struct ioring_data { |
9a2d78b3 JA |
59 | int ring_fd; |
60 | ||
52885fa2 | 61 | struct io_u **io_u_index; |
2d6451c9 | 62 | char *md_buf; |
52885fa2 | 63 | |
5ffd5626 JA |
64 | int *fds; |
65 | ||
bffad86f | 66 | struct io_sq_ring sq_ring; |
f0403f94 | 67 | struct io_uring_sqe *sqes; |
9a2d78b3 | 68 | struct iovec *iovecs; |
b87aa01a | 69 | unsigned sq_ring_mask; |
52885fa2 | 70 | |
bffad86f | 71 | struct io_cq_ring cq_ring; |
b87aa01a | 72 | unsigned cq_ring_mask; |
52885fa2 JA |
73 | |
74 | int queued; | |
75 | int cq_ring_off; | |
b87aa01a | 76 | unsigned iodepth; |
5a59a81d | 77 | int prepped; |
96563db9 | 78 | |
bffad86f | 79 | struct ioring_mmap mmap[3]; |
d6cbeab4 NC |
80 | |
81 | struct cmdprio cmdprio; | |
4885a6eb VF |
82 | |
83 | struct nvme_dsm_range *dsm; | |
52885fa2 JA |
84 | }; |
85 | ||
bffad86f | 86 | struct ioring_options { |
a48f0cc7 | 87 | struct thread_data *td; |
52885fa2 | 88 | unsigned int hipri; |
d6cbeab4 | 89 | struct cmdprio_options cmdprio_options; |
52885fa2 | 90 | unsigned int fixedbufs; |
5ffd5626 | 91 | unsigned int registerfiles; |
3d7d00a3 | 92 | unsigned int sqpoll_thread; |
2ea53ca3 JA |
93 | unsigned int sqpoll_set; |
94 | unsigned int sqpoll_cpu; | |
b10b1e70 | 95 | unsigned int nonvectored; |
4a87b584 | 96 | unsigned int uncached; |
7d42e66e | 97 | unsigned int nowait; |
5a59a81d | 98 | unsigned int force_async; |
2d6451c9 | 99 | unsigned int md_per_io_size; |
3ee8311a AK |
100 | unsigned int pi_act; |
101 | unsigned int apptag; | |
102 | unsigned int apptag_mask; | |
103 | unsigned int prchk; | |
104 | char *pi_chk; | |
855dc4d4 | 105 | enum uring_cmd_type cmd_type; |
52885fa2 JA |
106 | }; |
107 | ||
b10b1e70 JA |
108 | static const int ddir_to_op[2][2] = { |
109 | { IORING_OP_READV, IORING_OP_READ }, | |
110 | { IORING_OP_WRITEV, IORING_OP_WRITE } | |
111 | }; | |
112 | ||
3f1e3af7 KB |
113 | static const int fixed_ddir_to_op[2] = { |
114 | IORING_OP_READ_FIXED, | |
115 | IORING_OP_WRITE_FIXED | |
116 | }; | |
117 | ||
2ea53ca3 | 118 | static int fio_ioring_sqpoll_cb(void *data, unsigned long long *val) |
a90cd050 | 119 | { |
bffad86f | 120 | struct ioring_options *o = data; |
a90cd050 | 121 | |
2ea53ca3 JA |
122 | o->sqpoll_cpu = *val; |
123 | o->sqpoll_set = 1; | |
a90cd050 JA |
124 | return 0; |
125 | } | |
126 | ||
52885fa2 JA |
127 | static struct fio_option options[] = { |
128 | { | |
129 | .name = "hipri", | |
130 | .lname = "High Priority", | |
131 | .type = FIO_OPT_STR_SET, | |
bffad86f | 132 | .off1 = offsetof(struct ioring_options, hipri), |
52885fa2 JA |
133 | .help = "Use polled IO completions", |
134 | .category = FIO_OPT_C_ENGINE, | |
27f436d9 | 135 | .group = FIO_OPT_G_IOURING, |
52885fa2 JA |
136 | }, |
137 | { | |
138 | .name = "fixedbufs", | |
139 | .lname = "Fixed (pre-mapped) IO buffers", | |
140 | .type = FIO_OPT_STR_SET, | |
bffad86f | 141 | .off1 = offsetof(struct ioring_options, fixedbufs), |
52885fa2 JA |
142 | .help = "Pre map IO buffers", |
143 | .category = FIO_OPT_C_ENGINE, | |
27f436d9 | 144 | .group = FIO_OPT_G_IOURING, |
52885fa2 | 145 | }, |
5ffd5626 JA |
146 | { |
147 | .name = "registerfiles", | |
148 | .lname = "Register file set", | |
149 | .type = FIO_OPT_STR_SET, | |
150 | .off1 = offsetof(struct ioring_options, registerfiles), | |
151 | .help = "Pre-open/register files", | |
152 | .category = FIO_OPT_C_ENGINE, | |
27f436d9 | 153 | .group = FIO_OPT_G_IOURING, |
5ffd5626 | 154 | }, |
771c9901 JA |
155 | { |
156 | .name = "sqthread_poll", | |
3d7d00a3 | 157 | .lname = "Kernel SQ thread polling", |
d6f936d1 | 158 | .type = FIO_OPT_STR_SET, |
3d7d00a3 JA |
159 | .off1 = offsetof(struct ioring_options, sqpoll_thread), |
160 | .help = "Offload submission/completion to kernel thread", | |
161 | .category = FIO_OPT_C_ENGINE, | |
27f436d9 | 162 | .group = FIO_OPT_G_IOURING, |
3d7d00a3 JA |
163 | }, |
164 | { | |
165 | .name = "sqthread_poll_cpu", | |
166 | .lname = "SQ Thread Poll CPU", | |
2ea53ca3 JA |
167 | .type = FIO_OPT_INT, |
168 | .cb = fio_ioring_sqpoll_cb, | |
3d7d00a3 | 169 | .help = "What CPU to run SQ thread polling on", |
a90cd050 | 170 | .category = FIO_OPT_C_ENGINE, |
27f436d9 | 171 | .group = FIO_OPT_G_IOURING, |
a90cd050 | 172 | }, |
b10b1e70 JA |
173 | { |
174 | .name = "nonvectored", | |
175 | .lname = "Non-vectored", | |
176 | .type = FIO_OPT_INT, | |
177 | .off1 = offsetof(struct ioring_options, nonvectored), | |
556d8415 | 178 | .def = "-1", |
b10b1e70 JA |
179 | .help = "Use non-vectored read/write commands", |
180 | .category = FIO_OPT_C_ENGINE, | |
181 | .group = FIO_OPT_G_IOURING, | |
182 | }, | |
4a87b584 JA |
183 | { |
184 | .name = "uncached", | |
185 | .lname = "Uncached", | |
186 | .type = FIO_OPT_INT, | |
187 | .off1 = offsetof(struct ioring_options, uncached), | |
188 | .help = "Use RWF_UNCACHED for buffered read/writes", | |
189 | .category = FIO_OPT_C_ENGINE, | |
190 | .group = FIO_OPT_G_IOURING, | |
191 | }, | |
7d42e66e KK |
192 | { |
193 | .name = "nowait", | |
194 | .lname = "RWF_NOWAIT", | |
195 | .type = FIO_OPT_BOOL, | |
196 | .off1 = offsetof(struct ioring_options, nowait), | |
197 | .help = "Use RWF_NOWAIT for reads/writes", | |
198 | .category = FIO_OPT_C_ENGINE, | |
199 | .group = FIO_OPT_G_IOURING, | |
200 | }, | |
5a59a81d JA |
201 | { |
202 | .name = "force_async", | |
203 | .lname = "Force async", | |
204 | .type = FIO_OPT_INT, | |
205 | .off1 = offsetof(struct ioring_options, force_async), | |
206 | .help = "Set IOSQE_ASYNC every N requests", | |
207 | .category = FIO_OPT_C_ENGINE, | |
208 | .group = FIO_OPT_G_IOURING, | |
209 | }, | |
855dc4d4 AG |
210 | { |
211 | .name = "cmd_type", | |
212 | .lname = "Uring cmd type", | |
213 | .type = FIO_OPT_STR, | |
214 | .off1 = offsetof(struct ioring_options, cmd_type), | |
215 | .help = "Specify uring-cmd type", | |
216 | .def = "nvme", | |
217 | .posval = { | |
218 | { .ival = "nvme", | |
219 | .oval = FIO_URING_CMD_NVME, | |
220 | .help = "Issue nvme-uring-cmd", | |
221 | }, | |
222 | }, | |
223 | .category = FIO_OPT_C_ENGINE, | |
224 | .group = FIO_OPT_G_IOURING, | |
225 | }, | |
2838f77a | 226 | CMDPRIO_OPTIONS(struct ioring_options, FIO_OPT_G_IOURING), |
2d6451c9 AK |
227 | { |
228 | .name = "md_per_io_size", | |
229 | .lname = "Separate Metadata Buffer Size per I/O", | |
230 | .type = FIO_OPT_INT, | |
231 | .off1 = offsetof(struct ioring_options, md_per_io_size), | |
232 | .def = "0", | |
233 | .help = "Size of separate metadata buffer per I/O (Default: 0)", | |
234 | .category = FIO_OPT_C_ENGINE, | |
235 | .group = FIO_OPT_G_IOURING, | |
236 | }, | |
3ee8311a AK |
237 | { |
238 | .name = "pi_act", | |
239 | .lname = "Protection Information Action", | |
240 | .type = FIO_OPT_BOOL, | |
241 | .off1 = offsetof(struct ioring_options, pi_act), | |
242 | .def = "1", | |
243 | .help = "Protection Information Action bit (pi_act=1 or pi_act=0)", | |
244 | .category = FIO_OPT_C_ENGINE, | |
245 | .group = FIO_OPT_G_IOURING, | |
246 | }, | |
247 | { | |
248 | .name = "pi_chk", | |
249 | .lname = "Protection Information Check", | |
250 | .type = FIO_OPT_STR_STORE, | |
251 | .off1 = offsetof(struct ioring_options, pi_chk), | |
252 | .def = NULL, | |
253 | .help = "Control of Protection Information Checking (pi_chk=GUARD,REFTAG,APPTAG)", | |
254 | .category = FIO_OPT_C_ENGINE, | |
255 | .group = FIO_OPT_G_IOURING, | |
256 | }, | |
257 | { | |
258 | .name = "apptag", | |
259 | .lname = "Application Tag used in Protection Information", | |
260 | .type = FIO_OPT_INT, | |
261 | .off1 = offsetof(struct ioring_options, apptag), | |
262 | .def = "0x1234", | |
263 | .help = "Application Tag used in Protection Information field (Default: 0x1234)", | |
264 | .category = FIO_OPT_C_ENGINE, | |
265 | .group = FIO_OPT_G_IOURING, | |
266 | }, | |
267 | { | |
268 | .name = "apptag_mask", | |
269 | .lname = "Application Tag Mask", | |
270 | .type = FIO_OPT_INT, | |
271 | .off1 = offsetof(struct ioring_options, apptag_mask), | |
272 | .def = "0xffff", | |
273 | .help = "Application Tag Mask used with Application Tag (Default: 0xffff)", | |
274 | .category = FIO_OPT_C_ENGINE, | |
275 | .group = FIO_OPT_G_IOURING, | |
276 | }, | |
52885fa2 JA |
277 | { |
278 | .name = NULL, | |
279 | }, | |
280 | }; | |
281 | ||
bffad86f | 282 | static int io_uring_enter(struct ioring_data *ld, unsigned int to_submit, |
52885fa2 JA |
283 | unsigned int min_complete, unsigned int flags) |
284 | { | |
c377f4f8 JA |
285 | #ifdef FIO_ARCH_HAS_SYSCALL |
286 | return __do_syscall6(__NR_io_uring_enter, ld->ring_fd, to_submit, | |
287 | min_complete, flags, NULL, 0); | |
288 | #else | |
bfed648c | 289 | return syscall(__NR_io_uring_enter, ld->ring_fd, to_submit, |
521164fa | 290 | min_complete, flags, NULL, 0); |
c377f4f8 | 291 | #endif |
52885fa2 JA |
292 | } |
293 | ||
bffad86f | 294 | static int fio_ioring_prep(struct thread_data *td, struct io_u *io_u) |
52885fa2 | 295 | { |
bffad86f | 296 | struct ioring_data *ld = td->io_ops_data; |
cfcc8564 | 297 | struct ioring_options *o = td->eo; |
52885fa2 | 298 | struct fio_file *f = io_u->file; |
f0403f94 | 299 | struct io_uring_sqe *sqe; |
52885fa2 | 300 | |
f0403f94 | 301 | sqe = &ld->sqes[io_u->index]; |
34d6090e | 302 | |
5ffd5626 JA |
303 | if (o->registerfiles) { |
304 | sqe->fd = f->engine_pos; | |
305 | sqe->flags = IOSQE_FIXED_FILE; | |
306 | } else { | |
307 | sqe->fd = f->fd; | |
87b69ef2 | 308 | sqe->flags = 0; |
5ffd5626 | 309 | } |
52885fa2 | 310 | |
e3970057 | 311 | if (io_u->ddir == DDIR_READ || io_u->ddir == DDIR_WRITE) { |
f0403f94 | 312 | if (o->fixedbufs) { |
3f1e3af7 | 313 | sqe->opcode = fixed_ddir_to_op[io_u->ddir]; |
919850d2 | 314 | sqe->addr = (unsigned long) io_u->xfer_buf; |
f0403f94 | 315 | sqe->len = io_u->xfer_buflen; |
2ea53ca3 | 316 | sqe->buf_index = io_u->index; |
cfcc8564 | 317 | } else { |
832faaaf JA |
318 | struct iovec *iov = &ld->iovecs[io_u->index]; |
319 | ||
320 | /* | |
321 | * Update based on actual io_u, requeue could have | |
322 | * adjusted these | |
323 | */ | |
324 | iov->iov_base = io_u->xfer_buf; | |
325 | iov->iov_len = io_u->xfer_buflen; | |
326 | ||
3f1e3af7 | 327 | sqe->opcode = ddir_to_op[io_u->ddir][!!o->nonvectored]; |
b10b1e70 | 328 | if (o->nonvectored) { |
832faaaf JA |
329 | sqe->addr = (unsigned long) iov->iov_base; |
330 | sqe->len = iov->iov_len; | |
b10b1e70 | 331 | } else { |
832faaaf | 332 | sqe->addr = (unsigned long) iov; |
b10b1e70 JA |
333 | sqe->len = 1; |
334 | } | |
cfcc8564 | 335 | } |
fd70e361 | 336 | sqe->rw_flags = 0; |
4a87b584 | 337 | if (!td->o.odirect && o->uncached) |
fd70e361 | 338 | sqe->rw_flags |= RWF_UNCACHED; |
7d42e66e KK |
339 | if (o->nowait) |
340 | sqe->rw_flags |= RWF_NOWAIT; | |
8ff6b289 NC |
341 | |
342 | /* | |
343 | * Since io_uring can have a submission context (sqthread_poll) | |
344 | * that is different from the process context, we cannot rely on | |
79012fec DLM |
345 | * the IO priority set by ioprio_set() (options prio, prioclass, |
346 | * and priohint) to be inherited. | |
8ff6b289 NC |
347 | * td->ioprio will have the value of the "default prio", so set |
348 | * this unconditionally. This value might get overridden by | |
ff00f247 | 349 | * fio_ioring_cmdprio_prep() if the option cmdprio_percentage or |
8ff6b289 NC |
350 | * cmdprio_bssplit is used. |
351 | */ | |
352 | sqe->ioprio = td->ioprio; | |
f0403f94 | 353 | sqe->off = io_u->offset; |
48e698fa | 354 | } else if (ddir_sync(io_u->ddir)) { |
7c70f506 | 355 | sqe->ioprio = 0; |
01387bfe AF |
356 | if (io_u->ddir == DDIR_SYNC_FILE_RANGE) { |
357 | sqe->off = f->first_write; | |
358 | sqe->len = f->last_write - f->first_write; | |
359 | sqe->sync_range_flags = td->o.sync_file_range; | |
360 | sqe->opcode = IORING_OP_SYNC_FILE_RANGE; | |
361 | } else { | |
7c70f506 JA |
362 | sqe->off = 0; |
363 | sqe->addr = 0; | |
364 | sqe->len = 0; | |
01387bfe AF |
365 | if (io_u->ddir == DDIR_DATASYNC) |
366 | sqe->fsync_flags |= IORING_FSYNC_DATASYNC; | |
367 | sqe->opcode = IORING_OP_FSYNC; | |
368 | } | |
48e698fa | 369 | } |
52885fa2 | 370 | |
5a59a81d JA |
371 | if (o->force_async && ++ld->prepped == o->force_async) { |
372 | ld->prepped = 0; | |
373 | sqe->flags |= IOSQE_ASYNC; | |
374 | } | |
375 | ||
48e698fa | 376 | sqe->user_data = (unsigned long) io_u; |
52885fa2 JA |
377 | return 0; |
378 | } | |
379 | ||
855dc4d4 AG |
380 | static int fio_ioring_cmd_prep(struct thread_data *td, struct io_u *io_u) |
381 | { | |
382 | struct ioring_data *ld = td->io_ops_data; | |
383 | struct ioring_options *o = td->eo; | |
384 | struct fio_file *f = io_u->file; | |
3ce6a3de | 385 | struct nvme_uring_cmd *cmd; |
855dc4d4 | 386 | struct io_uring_sqe *sqe; |
855dc4d4 | 387 | |
3ce6a3de JA |
388 | /* only supports nvme_uring_cmd */ |
389 | if (o->cmd_type != FIO_URING_CMD_NVME) | |
390 | return -EINVAL; | |
855dc4d4 | 391 | |
4885a6eb | 392 | if (io_u->ddir == DDIR_TRIM && td->io_ops->flags & FIO_ASYNCIO_SYNC_TRIM) |
16be6037 AK |
393 | return 0; |
394 | ||
3ce6a3de | 395 | sqe = &ld->sqes[(io_u->index) << 1]; |
855dc4d4 | 396 | |
3ce6a3de JA |
397 | if (o->registerfiles) { |
398 | sqe->fd = f->engine_pos; | |
399 | sqe->flags = IOSQE_FIXED_FILE; | |
400 | } else { | |
401 | sqe->fd = f->fd; | |
402 | } | |
403 | sqe->rw_flags = 0; | |
404 | if (!td->o.odirect && o->uncached) | |
405 | sqe->rw_flags |= RWF_UNCACHED; | |
406 | if (o->nowait) | |
407 | sqe->rw_flags |= RWF_NOWAIT; | |
855dc4d4 | 408 | |
3ce6a3de JA |
409 | sqe->opcode = IORING_OP_URING_CMD; |
410 | sqe->user_data = (unsigned long) io_u; | |
411 | if (o->nonvectored) | |
412 | sqe->cmd_op = NVME_URING_CMD_IO; | |
413 | else | |
414 | sqe->cmd_op = NVME_URING_CMD_IO_VEC; | |
415 | if (o->force_async && ++ld->prepped == o->force_async) { | |
416 | ld->prepped = 0; | |
417 | sqe->flags |= IOSQE_ASYNC; | |
855dc4d4 | 418 | } |
0ebd3bf6 AG |
419 | if (o->fixedbufs) { |
420 | sqe->uring_cmd_flags = IORING_URING_CMD_FIXED; | |
421 | sqe->buf_index = io_u->index; | |
422 | } | |
3ce6a3de JA |
423 | |
424 | cmd = (struct nvme_uring_cmd *)sqe->cmd; | |
425 | return fio_nvme_uring_cmd_prep(cmd, io_u, | |
4885a6eb VF |
426 | o->nonvectored ? NULL : &ld->iovecs[io_u->index], |
427 | &ld->dsm[io_u->index]); | |
855dc4d4 AG |
428 | } |
429 | ||
bffad86f | 430 | static struct io_u *fio_ioring_event(struct thread_data *td, int event) |
52885fa2 | 431 | { |
bffad86f | 432 | struct ioring_data *ld = td->io_ops_data; |
f0403f94 | 433 | struct io_uring_cqe *cqe; |
52885fa2 | 434 | struct io_u *io_u; |
b87aa01a | 435 | unsigned index; |
52885fa2 | 436 | |
b87aa01a | 437 | index = (event + ld->cq_ring_off) & ld->cq_ring_mask; |
52885fa2 | 438 | |
f0403f94 | 439 | cqe = &ld->cq_ring.cqes[index]; |
e3466352 | 440 | io_u = (struct io_u *) (uintptr_t) cqe->user_data; |
52885fa2 | 441 | |
f0403f94 JA |
442 | if (cqe->res != io_u->xfer_buflen) { |
443 | if (cqe->res > io_u->xfer_buflen) | |
444 | io_u->error = -cqe->res; | |
52885fa2 | 445 | else |
f0403f94 | 446 | io_u->resid = io_u->xfer_buflen - cqe->res; |
52885fa2 JA |
447 | } else |
448 | io_u->error = 0; | |
449 | ||
450 | return io_u; | |
451 | } | |
452 | ||
855dc4d4 AG |
453 | static struct io_u *fio_ioring_cmd_event(struct thread_data *td, int event) |
454 | { | |
455 | struct ioring_data *ld = td->io_ops_data; | |
456 | struct ioring_options *o = td->eo; | |
457 | struct io_uring_cqe *cqe; | |
458 | struct io_u *io_u; | |
5163f35e | 459 | struct nvme_data *data; |
855dc4d4 | 460 | unsigned index; |
5163f35e | 461 | int ret; |
855dc4d4 AG |
462 | |
463 | index = (event + ld->cq_ring_off) & ld->cq_ring_mask; | |
464 | if (o->cmd_type == FIO_URING_CMD_NVME) | |
465 | index <<= 1; | |
466 | ||
467 | cqe = &ld->cq_ring.cqes[index]; | |
468 | io_u = (struct io_u *) (uintptr_t) cqe->user_data; | |
469 | ||
470 | if (cqe->res != 0) | |
471 | io_u->error = -cqe->res; | |
472 | else | |
473 | io_u->error = 0; | |
474 | ||
5163f35e AK |
475 | if (o->cmd_type == FIO_URING_CMD_NVME) { |
476 | data = FILE_ENG_DATA(io_u->file); | |
477 | if (data->pi_type && (io_u->ddir == DDIR_READ) && !o->pi_act) { | |
478 | ret = fio_nvme_pi_verify(data, io_u); | |
479 | if (ret) | |
480 | io_u->error = ret; | |
481 | } | |
482 | } | |
483 | ||
855dc4d4 AG |
484 | return io_u; |
485 | } | |
486 | ||
bffad86f | 487 | static int fio_ioring_cqring_reap(struct thread_data *td, unsigned int events, |
52885fa2 JA |
488 | unsigned int max) |
489 | { | |
bffad86f JA |
490 | struct ioring_data *ld = td->io_ops_data; |
491 | struct io_cq_ring *ring = &ld->cq_ring; | |
e2239016 | 492 | unsigned head, reaped = 0; |
52885fa2 | 493 | |
9a2d78b3 | 494 | head = *ring->head; |
52885fa2 | 495 | do { |
9e26aff9 | 496 | if (head == atomic_load_acquire(ring->tail)) |
52885fa2 JA |
497 | break; |
498 | reaped++; | |
499 | head++; | |
52885fa2 JA |
500 | } while (reaped + events < max); |
501 | ||
76ce63dd AB |
502 | if (reaped) |
503 | atomic_store_release(ring->head, head); | |
504 | ||
52885fa2 JA |
505 | return reaped; |
506 | } | |
507 | ||
bffad86f JA |
508 | static int fio_ioring_getevents(struct thread_data *td, unsigned int min, |
509 | unsigned int max, const struct timespec *t) | |
52885fa2 | 510 | { |
bffad86f | 511 | struct ioring_data *ld = td->io_ops_data; |
52885fa2 | 512 | unsigned actual_min = td->o.iodepth_batch_complete_min == 0 ? 0 : min; |
bffad86f JA |
513 | struct ioring_options *o = td->eo; |
514 | struct io_cq_ring *ring = &ld->cq_ring; | |
b87aa01a JA |
515 | unsigned events = 0; |
516 | int r; | |
52885fa2 | 517 | |
9a2d78b3 | 518 | ld->cq_ring_off = *ring->head; |
52885fa2 | 519 | do { |
bffad86f | 520 | r = fio_ioring_cqring_reap(td, events, max); |
52885fa2 JA |
521 | if (r) { |
522 | events += r; | |
ae8646a1 | 523 | max -= r; |
f7cbbbf8 ST |
524 | if (actual_min != 0) |
525 | actual_min -= r; | |
52885fa2 JA |
526 | continue; |
527 | } | |
528 | ||
3d7d00a3 | 529 | if (!o->sqpoll_thread) { |
9a2d78b3 JA |
530 | r = io_uring_enter(ld, 0, actual_min, |
531 | IORING_ENTER_GETEVENTS); | |
771c9901 | 532 | if (r < 0) { |
f6abd731 | 533 | if (errno == EAGAIN || errno == EINTR) |
771c9901 | 534 | continue; |
1816895b | 535 | r = -errno; |
9a2d78b3 | 536 | td_verror(td, errno, "io_uring_enter"); |
771c9901 JA |
537 | break; |
538 | } | |
52885fa2 JA |
539 | } |
540 | } while (events < min); | |
541 | ||
542 | return r < 0 ? r : events; | |
543 | } | |
544 | ||
3ee8311a AK |
545 | static inline void fio_ioring_cmd_nvme_pi(struct thread_data *td, |
546 | struct io_u *io_u) | |
547 | { | |
548 | struct ioring_data *ld = td->io_ops_data; | |
549 | struct ioring_options *o = td->eo; | |
550 | struct nvme_uring_cmd *cmd; | |
551 | struct io_uring_sqe *sqe; | |
552 | struct nvme_cmd_ext_io_opts ext_opts = {0}; | |
553 | struct nvme_data *data = FILE_ENG_DATA(io_u->file); | |
554 | ||
555 | if (io_u->ddir == DDIR_TRIM) | |
556 | return; | |
557 | ||
558 | sqe = &ld->sqes[(io_u->index) << 1]; | |
559 | cmd = (struct nvme_uring_cmd *)sqe->cmd; | |
560 | ||
561 | if (data->pi_type) { | |
562 | if (o->pi_act) | |
563 | ext_opts.io_flags |= NVME_IO_PRINFO_PRACT; | |
564 | ext_opts.io_flags |= o->prchk; | |
565 | ext_opts.apptag = o->apptag; | |
566 | ext_opts.apptag_mask = o->apptag_mask; | |
567 | } | |
568 | ||
569 | fio_nvme_pi_fill(cmd, io_u, &ext_opts); | |
570 | } | |
571 | ||
127715b6 NC |
572 | static inline void fio_ioring_cmdprio_prep(struct thread_data *td, |
573 | struct io_u *io_u) | |
b2a432bf | 574 | { |
b2a432bf | 575 | struct ioring_data *ld = td->io_ops_data; |
d6cbeab4 | 576 | struct cmdprio *cmdprio = &ld->cmdprio; |
127715b6 NC |
577 | |
578 | if (fio_cmdprio_set_ioprio(td, cmdprio, io_u)) | |
579 | ld->sqes[io_u->index].ioprio = io_u->ioprio; | |
b2a432bf PC |
580 | } |
581 | ||
bffad86f JA |
582 | static enum fio_q_status fio_ioring_queue(struct thread_data *td, |
583 | struct io_u *io_u) | |
52885fa2 | 584 | { |
bffad86f | 585 | struct ioring_data *ld = td->io_ops_data; |
3ee8311a | 586 | struct ioring_options *o = td->eo; |
bffad86f | 587 | struct io_sq_ring *ring = &ld->sq_ring; |
52885fa2 JA |
588 | unsigned tail, next_tail; |
589 | ||
590 | fio_ro_check(td, io_u); | |
591 | ||
b87aa01a | 592 | if (ld->queued == ld->iodepth) |
52885fa2 JA |
593 | return FIO_Q_BUSY; |
594 | ||
4885a6eb | 595 | if (io_u->ddir == DDIR_TRIM && td->io_ops->flags & FIO_ASYNCIO_SYNC_TRIM) { |
52885fa2 JA |
596 | if (ld->queued) |
597 | return FIO_Q_BUSY; | |
598 | ||
7f57e30f | 599 | do_io_u_trim(td, io_u); |
16be6037 | 600 | |
52885fa2 JA |
601 | io_u_mark_submit(td, 1); |
602 | io_u_mark_complete(td, 1); | |
603 | return FIO_Q_COMPLETED; | |
604 | } | |
605 | ||
9a2d78b3 | 606 | tail = *ring->tail; |
52885fa2 | 607 | next_tail = tail + 1; |
5c15a911 | 608 | if (next_tail == atomic_load_relaxed(ring->head)) |
52885fa2 JA |
609 | return FIO_Q_BUSY; |
610 | ||
d6cbeab4 | 611 | if (ld->cmdprio.mode != CMDPRIO_MODE_NONE) |
ff00f247 NC |
612 | fio_ioring_cmdprio_prep(td, io_u); |
613 | ||
3ee8311a AK |
614 | if (!strcmp(td->io_ops->name, "io_uring_cmd") && |
615 | o->cmd_type == FIO_URING_CMD_NVME) | |
616 | fio_ioring_cmd_nvme_pi(td, io_u); | |
617 | ||
b87aa01a | 618 | ring->array[tail & ld->sq_ring_mask] = io_u->index; |
9e26aff9 | 619 | atomic_store_release(ring->tail, next_tail); |
52885fa2 JA |
620 | |
621 | ld->queued++; | |
622 | return FIO_Q_QUEUED; | |
623 | } | |
624 | ||
bffad86f | 625 | static void fio_ioring_queued(struct thread_data *td, int start, int nr) |
52885fa2 | 626 | { |
bffad86f | 627 | struct ioring_data *ld = td->io_ops_data; |
52885fa2 JA |
628 | struct timespec now; |
629 | ||
630 | if (!fio_fill_issue_time(td)) | |
631 | return; | |
632 | ||
633 | fio_gettime(&now, NULL); | |
634 | ||
635 | while (nr--) { | |
bffad86f | 636 | struct io_sq_ring *ring = &ld->sq_ring; |
9a2d78b3 | 637 | int index = ring->array[start & ld->sq_ring_mask]; |
f8289afc | 638 | struct io_u *io_u = ld->io_u_index[index]; |
52885fa2 JA |
639 | |
640 | memcpy(&io_u->issue_time, &now, sizeof(now)); | |
641 | io_u_queued(td, io_u); | |
642 | ||
643 | start++; | |
52885fa2 | 644 | } |
39f56400 VF |
645 | |
646 | /* | |
647 | * only used for iolog | |
648 | */ | |
649 | if (td->o.read_iolog_file) | |
650 | memcpy(&td->last_issue, &now, sizeof(now)); | |
52885fa2 JA |
651 | } |
652 | ||
bffad86f | 653 | static int fio_ioring_commit(struct thread_data *td) |
52885fa2 | 654 | { |
bffad86f JA |
655 | struct ioring_data *ld = td->io_ops_data; |
656 | struct ioring_options *o = td->eo; | |
52885fa2 JA |
657 | int ret; |
658 | ||
659 | if (!ld->queued) | |
660 | return 0; | |
661 | ||
3d7d00a3 JA |
662 | /* |
663 | * Kernel side does submission. just need to check if the ring is | |
664 | * flagged as needing a kick, if so, call io_uring_enter(). This | |
665 | * only happens if we've been idle too long. | |
666 | */ | |
667 | if (o->sqpoll_thread) { | |
bffad86f | 668 | struct io_sq_ring *ring = &ld->sq_ring; |
10bad6b9 | 669 | unsigned start = *ld->sq_ring.tail - ld->queued; |
2dd96cc4 | 670 | unsigned flags; |
4cdbc048 | 671 | |
5c15a911 | 672 | flags = atomic_load_relaxed(ring->flags); |
2dd96cc4 | 673 | if (flags & IORING_SQ_NEED_WAKEUP) |
b532dd6d JA |
674 | io_uring_enter(ld, ld->queued, 0, |
675 | IORING_ENTER_SQ_WAKEUP); | |
c011bf12 AK |
676 | fio_ioring_queued(td, start, ld->queued); |
677 | io_u_mark_submit(td, ld->queued); | |
678 | ||
771c9901 JA |
679 | ld->queued = 0; |
680 | return 0; | |
681 | } | |
682 | ||
52885fa2 | 683 | do { |
9a2d78b3 | 684 | unsigned start = *ld->sq_ring.head; |
52885fa2 JA |
685 | long nr = ld->queued; |
686 | ||
9a2d78b3 | 687 | ret = io_uring_enter(ld, nr, 0, IORING_ENTER_GETEVENTS); |
52885fa2 | 688 | if (ret > 0) { |
bffad86f | 689 | fio_ioring_queued(td, start, ret); |
52885fa2 JA |
690 | io_u_mark_submit(td, ret); |
691 | ||
692 | ld->queued -= ret; | |
693 | ret = 0; | |
a90cd050 JA |
694 | } else if (!ret) { |
695 | io_u_mark_submit(td, ret); | |
52885fa2 | 696 | continue; |
a90cd050 | 697 | } else { |
f6abd731 | 698 | if (errno == EAGAIN || errno == EINTR) { |
bffad86f | 699 | ret = fio_ioring_cqring_reap(td, 0, ld->queued); |
a90cd050 JA |
700 | if (ret) |
701 | continue; | |
702 | /* Shouldn't happen */ | |
703 | usleep(1); | |
704 | continue; | |
52885fa2 | 705 | } |
1816895b | 706 | ret = -errno; |
9a2d78b3 | 707 | td_verror(td, errno, "io_uring_enter submit"); |
52885fa2 | 708 | break; |
a90cd050 | 709 | } |
52885fa2 JA |
710 | } while (ld->queued); |
711 | ||
712 | return ret; | |
713 | } | |
714 | ||
bffad86f | 715 | static void fio_ioring_unmap(struct ioring_data *ld) |
52885fa2 | 716 | { |
9a2d78b3 | 717 | int i; |
52885fa2 | 718 | |
59f94d26 | 719 | for (i = 0; i < FIO_ARRAY_SIZE(ld->mmap); i++) |
9a2d78b3 JA |
720 | munmap(ld->mmap[i].ptr, ld->mmap[i].len); |
721 | close(ld->ring_fd); | |
b87aa01a JA |
722 | } |
723 | ||
bffad86f | 724 | static void fio_ioring_cleanup(struct thread_data *td) |
52885fa2 | 725 | { |
bffad86f | 726 | struct ioring_data *ld = td->io_ops_data; |
52885fa2 JA |
727 | |
728 | if (ld) { | |
52885fa2 | 729 | if (!(td->flags & TD_F_CHILD)) |
bffad86f | 730 | fio_ioring_unmap(ld); |
9a2d78b3 | 731 | |
d6cbeab4 | 732 | fio_cmdprio_cleanup(&ld->cmdprio); |
52885fa2 | 733 | free(ld->io_u_index); |
2d6451c9 | 734 | free(ld->md_buf); |
9a2d78b3 | 735 | free(ld->iovecs); |
5ffd5626 | 736 | free(ld->fds); |
4885a6eb | 737 | free(ld->dsm); |
52885fa2 JA |
738 | free(ld); |
739 | } | |
740 | } | |
741 | ||
bffad86f | 742 | static int fio_ioring_mmap(struct ioring_data *ld, struct io_uring_params *p) |
9a2d78b3 | 743 | { |
bffad86f JA |
744 | struct io_sq_ring *sring = &ld->sq_ring; |
745 | struct io_cq_ring *cring = &ld->cq_ring; | |
9a2d78b3 JA |
746 | void *ptr; |
747 | ||
e2239016 | 748 | ld->mmap[0].len = p->sq_off.array + p->sq_entries * sizeof(__u32); |
9a2d78b3 JA |
749 | ptr = mmap(0, ld->mmap[0].len, PROT_READ | PROT_WRITE, |
750 | MAP_SHARED | MAP_POPULATE, ld->ring_fd, | |
751 | IORING_OFF_SQ_RING); | |
752 | ld->mmap[0].ptr = ptr; | |
753 | sring->head = ptr + p->sq_off.head; | |
754 | sring->tail = ptr + p->sq_off.tail; | |
755 | sring->ring_mask = ptr + p->sq_off.ring_mask; | |
756 | sring->ring_entries = ptr + p->sq_off.ring_entries; | |
757 | sring->flags = ptr + p->sq_off.flags; | |
ac122fea | 758 | sring->array = ptr + p->sq_off.array; |
9a2d78b3 JA |
759 | ld->sq_ring_mask = *sring->ring_mask; |
760 | ||
855dc4d4 AG |
761 | if (p->flags & IORING_SETUP_SQE128) |
762 | ld->mmap[1].len = 2 * p->sq_entries * sizeof(struct io_uring_sqe); | |
763 | else | |
764 | ld->mmap[1].len = p->sq_entries * sizeof(struct io_uring_sqe); | |
f0403f94 | 765 | ld->sqes = mmap(0, ld->mmap[1].len, PROT_READ | PROT_WRITE, |
9a2d78b3 | 766 | MAP_SHARED | MAP_POPULATE, ld->ring_fd, |
f0403f94 JA |
767 | IORING_OFF_SQES); |
768 | ld->mmap[1].ptr = ld->sqes; | |
9a2d78b3 | 769 | |
855dc4d4 AG |
770 | if (p->flags & IORING_SETUP_CQE32) { |
771 | ld->mmap[2].len = p->cq_off.cqes + | |
772 | 2 * p->cq_entries * sizeof(struct io_uring_cqe); | |
773 | } else { | |
774 | ld->mmap[2].len = p->cq_off.cqes + | |
775 | p->cq_entries * sizeof(struct io_uring_cqe); | |
776 | } | |
9a2d78b3 JA |
777 | ptr = mmap(0, ld->mmap[2].len, PROT_READ | PROT_WRITE, |
778 | MAP_SHARED | MAP_POPULATE, ld->ring_fd, | |
779 | IORING_OFF_CQ_RING); | |
780 | ld->mmap[2].ptr = ptr; | |
781 | cring->head = ptr + p->cq_off.head; | |
782 | cring->tail = ptr + p->cq_off.tail; | |
783 | cring->ring_mask = ptr + p->cq_off.ring_mask; | |
784 | cring->ring_entries = ptr + p->cq_off.ring_entries; | |
f0403f94 | 785 | cring->cqes = ptr + p->cq_off.cqes; |
9a2d78b3 JA |
786 | ld->cq_ring_mask = *cring->ring_mask; |
787 | return 0; | |
788 | } | |
789 | ||
556d8415 JA |
790 | static void fio_ioring_probe(struct thread_data *td) |
791 | { | |
792 | struct ioring_data *ld = td->io_ops_data; | |
793 | struct ioring_options *o = td->eo; | |
794 | struct io_uring_probe *p; | |
795 | int ret; | |
796 | ||
797 | /* already set by user, don't touch */ | |
798 | if (o->nonvectored != -1) | |
799 | return; | |
800 | ||
801 | /* default to off, as that's always safe */ | |
802 | o->nonvectored = 0; | |
803 | ||
223decdd | 804 | p = calloc(1, sizeof(*p) + 256 * sizeof(struct io_uring_probe_op)); |
556d8415 JA |
805 | if (!p) |
806 | return; | |
807 | ||
556d8415 JA |
808 | ret = syscall(__NR_io_uring_register, ld->ring_fd, |
809 | IORING_REGISTER_PROBE, p, 256); | |
810 | if (ret < 0) | |
811 | goto out; | |
812 | ||
813 | if (IORING_OP_WRITE > p->ops_len) | |
814 | goto out; | |
815 | ||
816 | if ((p->ops[IORING_OP_READ].flags & IO_URING_OP_SUPPORTED) && | |
817 | (p->ops[IORING_OP_WRITE].flags & IO_URING_OP_SUPPORTED)) | |
818 | o->nonvectored = 1; | |
819 | out: | |
820 | free(p); | |
821 | } | |
822 | ||
bffad86f | 823 | static int fio_ioring_queue_init(struct thread_data *td) |
52885fa2 | 824 | { |
bffad86f JA |
825 | struct ioring_data *ld = td->io_ops_data; |
826 | struct ioring_options *o = td->eo; | |
52885fa2 | 827 | int depth = td->o.iodepth; |
bffad86f | 828 | struct io_uring_params p; |
9a2d78b3 JA |
829 | int ret; |
830 | ||
831 | memset(&p, 0, sizeof(p)); | |
52885fa2 JA |
832 | |
833 | if (o->hipri) | |
bffad86f | 834 | p.flags |= IORING_SETUP_IOPOLL; |
3d7d00a3 JA |
835 | if (o->sqpoll_thread) { |
836 | p.flags |= IORING_SETUP_SQPOLL; | |
837 | if (o->sqpoll_set) { | |
838 | p.flags |= IORING_SETUP_SQ_AFF; | |
839 | p.sq_thread_cpu = o->sqpoll_cpu; | |
840 | } | |
c011bf12 AK |
841 | |
842 | /* | |
843 | * Submission latency for sqpoll_thread is just the time it | |
844 | * takes to fill in the SQ ring entries, and any syscall if | |
845 | * IORING_SQ_NEED_WAKEUP is set, we don't need to log that time | |
846 | * separately. | |
847 | */ | |
848 | td->o.disable_slat = 1; | |
f635f1fb | 849 | } |
a90cd050 | 850 | |
1db268db JA |
851 | /* |
852 | * Clamp CQ ring size at our SQ ring size, we don't need more entries | |
853 | * than that. | |
854 | */ | |
855 | p.flags |= IORING_SETUP_CQSIZE; | |
856 | p.cq_entries = depth; | |
857 | ||
4d22c103 JA |
858 | /* |
859 | * Setup COOP_TASKRUN as we don't need to get IPI interrupted for | |
860 | * completing IO operations. | |
861 | */ | |
862 | p.flags |= IORING_SETUP_COOP_TASKRUN; | |
863 | ||
e453f369 JA |
864 | /* |
865 | * io_uring is always a single issuer, and we can defer task_work | |
866 | * runs until we reap events. | |
867 | */ | |
868 | p.flags |= IORING_SETUP_SINGLE_ISSUER | IORING_SETUP_DEFER_TASKRUN; | |
869 | ||
b5e99df6 | 870 | retry: |
bfed648c | 871 | ret = syscall(__NR_io_uring_setup, depth, &p); |
b5e99df6 | 872 | if (ret < 0) { |
e453f369 JA |
873 | if (errno == EINVAL && p.flags & IORING_SETUP_DEFER_TASKRUN) { |
874 | p.flags &= ~IORING_SETUP_DEFER_TASKRUN; | |
875 | p.flags &= ~IORING_SETUP_SINGLE_ISSUER; | |
876 | goto retry; | |
877 | } | |
4d22c103 JA |
878 | if (errno == EINVAL && p.flags & IORING_SETUP_COOP_TASKRUN) { |
879 | p.flags &= ~IORING_SETUP_COOP_TASKRUN; | |
880 | goto retry; | |
881 | } | |
b5e99df6 JA |
882 | if (errno == EINVAL && p.flags & IORING_SETUP_CQSIZE) { |
883 | p.flags &= ~IORING_SETUP_CQSIZE; | |
884 | goto retry; | |
885 | } | |
9a2d78b3 | 886 | return ret; |
b5e99df6 | 887 | } |
9a2d78b3 JA |
888 | |
889 | ld->ring_fd = ret; | |
2ea53ca3 | 890 | |
556d8415 JA |
891 | fio_ioring_probe(td); |
892 | ||
2ea53ca3 | 893 | if (o->fixedbufs) { |
bfed648c | 894 | ret = syscall(__NR_io_uring_register, ld->ring_fd, |
919850d2 | 895 | IORING_REGISTER_BUFFERS, ld->iovecs, depth); |
2ea53ca3 JA |
896 | if (ret < 0) |
897 | return ret; | |
898 | } | |
899 | ||
bffad86f | 900 | return fio_ioring_mmap(ld, &p); |
52885fa2 JA |
901 | } |
902 | ||
855dc4d4 AG |
903 | static int fio_ioring_cmd_queue_init(struct thread_data *td) |
904 | { | |
905 | struct ioring_data *ld = td->io_ops_data; | |
906 | struct ioring_options *o = td->eo; | |
907 | int depth = td->o.iodepth; | |
908 | struct io_uring_params p; | |
909 | int ret; | |
910 | ||
911 | memset(&p, 0, sizeof(p)); | |
912 | ||
913 | if (o->hipri) | |
914 | p.flags |= IORING_SETUP_IOPOLL; | |
915 | if (o->sqpoll_thread) { | |
916 | p.flags |= IORING_SETUP_SQPOLL; | |
917 | if (o->sqpoll_set) { | |
918 | p.flags |= IORING_SETUP_SQ_AFF; | |
919 | p.sq_thread_cpu = o->sqpoll_cpu; | |
920 | } | |
c011bf12 AK |
921 | |
922 | /* | |
923 | * Submission latency for sqpoll_thread is just the time it | |
924 | * takes to fill in the SQ ring entries, and any syscall if | |
925 | * IORING_SQ_NEED_WAKEUP is set, we don't need to log that time | |
926 | * separately. | |
927 | */ | |
928 | td->o.disable_slat = 1; | |
855dc4d4 AG |
929 | } |
930 | if (o->cmd_type == FIO_URING_CMD_NVME) { | |
931 | p.flags |= IORING_SETUP_SQE128; | |
932 | p.flags |= IORING_SETUP_CQE32; | |
933 | } | |
934 | ||
935 | /* | |
936 | * Clamp CQ ring size at our SQ ring size, we don't need more entries | |
937 | * than that. | |
938 | */ | |
939 | p.flags |= IORING_SETUP_CQSIZE; | |
940 | p.cq_entries = depth; | |
941 | ||
07f78c37 AK |
942 | /* |
943 | * Setup COOP_TASKRUN as we don't need to get IPI interrupted for | |
944 | * completing IO operations. | |
945 | */ | |
946 | p.flags |= IORING_SETUP_COOP_TASKRUN; | |
947 | ||
948 | /* | |
949 | * io_uring is always a single issuer, and we can defer task_work | |
950 | * runs until we reap events. | |
951 | */ | |
952 | p.flags |= IORING_SETUP_SINGLE_ISSUER | IORING_SETUP_DEFER_TASKRUN; | |
953 | ||
855dc4d4 AG |
954 | retry: |
955 | ret = syscall(__NR_io_uring_setup, depth, &p); | |
956 | if (ret < 0) { | |
07f78c37 AK |
957 | if (errno == EINVAL && p.flags & IORING_SETUP_DEFER_TASKRUN) { |
958 | p.flags &= ~IORING_SETUP_DEFER_TASKRUN; | |
959 | p.flags &= ~IORING_SETUP_SINGLE_ISSUER; | |
960 | goto retry; | |
961 | } | |
962 | if (errno == EINVAL && p.flags & IORING_SETUP_COOP_TASKRUN) { | |
963 | p.flags &= ~IORING_SETUP_COOP_TASKRUN; | |
964 | goto retry; | |
965 | } | |
855dc4d4 AG |
966 | if (errno == EINVAL && p.flags & IORING_SETUP_CQSIZE) { |
967 | p.flags &= ~IORING_SETUP_CQSIZE; | |
968 | goto retry; | |
969 | } | |
970 | return ret; | |
971 | } | |
972 | ||
973 | ld->ring_fd = ret; | |
974 | ||
975 | fio_ioring_probe(td); | |
976 | ||
977 | if (o->fixedbufs) { | |
978 | ret = syscall(__NR_io_uring_register, ld->ring_fd, | |
979 | IORING_REGISTER_BUFFERS, ld->iovecs, depth); | |
980 | if (ret < 0) | |
981 | return ret; | |
982 | } | |
983 | ||
984 | return fio_ioring_mmap(ld, &p); | |
985 | } | |
986 | ||
5ffd5626 JA |
987 | static int fio_ioring_register_files(struct thread_data *td) |
988 | { | |
989 | struct ioring_data *ld = td->io_ops_data; | |
990 | struct fio_file *f; | |
991 | unsigned int i; | |
992 | int ret; | |
993 | ||
994 | ld->fds = calloc(td->o.nr_files, sizeof(int)); | |
995 | ||
996 | for_each_file(td, f, i) { | |
997 | ret = generic_open_file(td, f); | |
998 | if (ret) | |
999 | goto err; | |
1000 | ld->fds[i] = f->fd; | |
1001 | f->engine_pos = i; | |
1002 | } | |
1003 | ||
bfed648c | 1004 | ret = syscall(__NR_io_uring_register, ld->ring_fd, |
5ffd5626 JA |
1005 | IORING_REGISTER_FILES, ld->fds, td->o.nr_files); |
1006 | if (ret) { | |
1007 | err: | |
1008 | free(ld->fds); | |
1009 | ld->fds = NULL; | |
1010 | } | |
1011 | ||
1012 | /* | |
1013 | * Pretend the file is closed again, and really close it if we hit | |
1014 | * an error. | |
1015 | */ | |
1016 | for_each_file(td, f, i) { | |
1017 | if (ret) { | |
1018 | int fio_unused ret2; | |
1019 | ret2 = generic_close_file(td, f); | |
1020 | } else | |
1021 | f->fd = -1; | |
1022 | } | |
1023 | ||
1024 | return ret; | |
1025 | } | |
1026 | ||
bffad86f | 1027 | static int fio_ioring_post_init(struct thread_data *td) |
52885fa2 | 1028 | { |
bffad86f | 1029 | struct ioring_data *ld = td->io_ops_data; |
5ffd5626 | 1030 | struct ioring_options *o = td->eo; |
52885fa2 | 1031 | struct io_u *io_u; |
650346e1 | 1032 | int err, i; |
52885fa2 | 1033 | |
650346e1 JA |
1034 | for (i = 0; i < td->o.iodepth; i++) { |
1035 | struct iovec *iov = &ld->iovecs[i]; | |
9a2d78b3 | 1036 | |
650346e1 JA |
1037 | io_u = ld->io_u_index[i]; |
1038 | iov->iov_base = io_u->buf; | |
1039 | iov->iov_len = td_max_bs(td); | |
52885fa2 JA |
1040 | } |
1041 | ||
bffad86f | 1042 | err = fio_ioring_queue_init(td); |
52885fa2 | 1043 | if (err) { |
0442b53f | 1044 | int init_err = errno; |
c4f5c92f | 1045 | |
0442b53f | 1046 | if (init_err == ENOSYS) |
c4f5c92f | 1047 | log_err("fio: your kernel doesn't support io_uring\n"); |
0442b53f | 1048 | td_verror(td, init_err, "io_queue_init"); |
52885fa2 JA |
1049 | return 1; |
1050 | } | |
1051 | ||
7c70f506 JA |
1052 | for (i = 0; i < td->o.iodepth; i++) { |
1053 | struct io_uring_sqe *sqe; | |
1054 | ||
1055 | sqe = &ld->sqes[i]; | |
1056 | memset(sqe, 0, sizeof(*sqe)); | |
1057 | } | |
1058 | ||
5ffd5626 JA |
1059 | if (o->registerfiles) { |
1060 | err = fio_ioring_register_files(td); | |
1061 | if (err) { | |
1062 | td_verror(td, errno, "ioring_register_files"); | |
1063 | return 1; | |
1064 | } | |
1065 | } | |
1066 | ||
52885fa2 JA |
1067 | return 0; |
1068 | } | |
1069 | ||
855dc4d4 AG |
1070 | static int fio_ioring_cmd_post_init(struct thread_data *td) |
1071 | { | |
1072 | struct ioring_data *ld = td->io_ops_data; | |
1073 | struct ioring_options *o = td->eo; | |
1074 | struct io_u *io_u; | |
1075 | int err, i; | |
1076 | ||
1077 | for (i = 0; i < td->o.iodepth; i++) { | |
1078 | struct iovec *iov = &ld->iovecs[i]; | |
1079 | ||
1080 | io_u = ld->io_u_index[i]; | |
1081 | iov->iov_base = io_u->buf; | |
1082 | iov->iov_len = td_max_bs(td); | |
1083 | } | |
1084 | ||
1085 | err = fio_ioring_cmd_queue_init(td); | |
1086 | if (err) { | |
1087 | int init_err = errno; | |
1088 | ||
1089 | td_verror(td, init_err, "io_queue_init"); | |
1090 | return 1; | |
1091 | } | |
1092 | ||
1093 | for (i = 0; i < td->o.iodepth; i++) { | |
1094 | struct io_uring_sqe *sqe; | |
1095 | ||
1096 | if (o->cmd_type == FIO_URING_CMD_NVME) { | |
1097 | sqe = &ld->sqes[i << 1]; | |
1098 | memset(sqe, 0, 2 * sizeof(*sqe)); | |
1099 | } else { | |
1100 | sqe = &ld->sqes[i]; | |
1101 | memset(sqe, 0, sizeof(*sqe)); | |
1102 | } | |
1103 | } | |
1104 | ||
1105 | if (o->registerfiles) { | |
1106 | err = fio_ioring_register_files(td); | |
1107 | if (err) { | |
1108 | td_verror(td, errno, "ioring_register_files"); | |
1109 | return 1; | |
1110 | } | |
1111 | } | |
1112 | ||
1113 | return 0; | |
1114 | } | |
1115 | ||
3ee8311a AK |
1116 | static void parse_prchk_flags(struct ioring_options *o) |
1117 | { | |
1118 | if (!o->pi_chk) | |
1119 | return; | |
1120 | ||
1121 | if (strstr(o->pi_chk, "GUARD") != NULL) | |
1122 | o->prchk = NVME_IO_PRINFO_PRCHK_GUARD; | |
1123 | if (strstr(o->pi_chk, "REFTAG") != NULL) | |
1124 | o->prchk |= NVME_IO_PRINFO_PRCHK_REF; | |
1125 | if (strstr(o->pi_chk, "APPTAG") != NULL) | |
1126 | o->prchk |= NVME_IO_PRINFO_PRCHK_APP; | |
1127 | } | |
1128 | ||
bffad86f | 1129 | static int fio_ioring_init(struct thread_data *td) |
52885fa2 | 1130 | { |
5ffd5626 | 1131 | struct ioring_options *o = td->eo; |
bffad86f | 1132 | struct ioring_data *ld; |
2d6451c9 | 1133 | unsigned long long md_size; |
e9f6567a | 1134 | int ret; |
52885fa2 | 1135 | |
5ffd5626 JA |
1136 | /* sqthread submission requires registered files */ |
1137 | if (o->sqpoll_thread) | |
1138 | o->registerfiles = 1; | |
1139 | ||
1140 | if (o->registerfiles && td->o.nr_files != td->o.open_files) { | |
1141 | log_err("fio: io_uring registered files require nr_files to " | |
1142 | "be identical to open_files\n"); | |
1143 | return 1; | |
1144 | } | |
1145 | ||
52885fa2 JA |
1146 | ld = calloc(1, sizeof(*ld)); |
1147 | ||
b87aa01a JA |
1148 | /* ring depth must be a power-of-2 */ |
1149 | ld->iodepth = td->o.iodepth; | |
1150 | td->o.iodepth = roundup_pow2(td->o.iodepth); | |
1151 | ||
52885fa2 JA |
1152 | /* io_u index */ |
1153 | ld->io_u_index = calloc(td->o.iodepth, sizeof(struct io_u *)); | |
2d6451c9 AK |
1154 | |
1155 | /* | |
1156 | * metadata buffer for nvme command. | |
1157 | * We are only supporting iomem=malloc / mem=malloc as of now. | |
1158 | */ | |
1159 | if (!strcmp(td->io_ops->name, "io_uring_cmd") && | |
1160 | (o->cmd_type == FIO_URING_CMD_NVME) && o->md_per_io_size) { | |
1161 | md_size = (unsigned long long) o->md_per_io_size | |
1162 | * (unsigned long long) td->o.iodepth; | |
1163 | md_size += page_mask + td->o.mem_align; | |
1164 | if (td->o.mem_align && td->o.mem_align > page_size) | |
1165 | md_size += td->o.mem_align - page_size; | |
1166 | if (td->o.mem_type == MEM_MALLOC) { | |
1167 | ld->md_buf = malloc(md_size); | |
6795954b JA |
1168 | if (!ld->md_buf) { |
1169 | free(ld); | |
2d6451c9 | 1170 | return 1; |
6795954b | 1171 | } |
2d6451c9 AK |
1172 | } else { |
1173 | log_err("fio: Only iomem=malloc or mem=malloc is supported\n"); | |
6795954b | 1174 | free(ld); |
2d6451c9 AK |
1175 | return 1; |
1176 | } | |
1177 | } | |
3ee8311a | 1178 | parse_prchk_flags(o); |
2d6451c9 | 1179 | |
650346e1 | 1180 | ld->iovecs = calloc(td->o.iodepth, sizeof(struct iovec)); |
52885fa2 JA |
1181 | |
1182 | td->io_ops_data = ld; | |
b2a432bf | 1183 | |
d6cbeab4 | 1184 | ret = fio_cmdprio_init(td, &ld->cmdprio, &o->cmdprio_options); |
e9f6567a DLM |
1185 | if (ret) { |
1186 | td_verror(td, EINVAL, "fio_ioring_init"); | |
b2a432bf PC |
1187 | return 1; |
1188 | } | |
1af44196 | 1189 | |
4885a6eb VF |
1190 | /* |
1191 | * For io_uring_cmd, trims are async operations unless we are operating | |
1192 | * in zbd mode where trim means zone reset. | |
1193 | */ | |
1194 | if (!strcmp(td->io_ops->name, "io_uring_cmd") && td_trim(td) && | |
1195 | td->o.zone_mode == ZONE_MODE_ZBD) | |
1196 | td->io_ops->flags |= FIO_ASYNCIO_SYNC_TRIM; | |
1197 | else | |
1198 | ld->dsm = calloc(ld->iodepth, sizeof(*ld->dsm)); | |
1199 | ||
52885fa2 JA |
1200 | return 0; |
1201 | } | |
1202 | ||
bffad86f | 1203 | static int fio_ioring_io_u_init(struct thread_data *td, struct io_u *io_u) |
52885fa2 | 1204 | { |
bffad86f | 1205 | struct ioring_data *ld = td->io_ops_data; |
2d6451c9 | 1206 | struct ioring_options *o = td->eo; |
5163f35e | 1207 | struct nvme_pi_data *pi_data; |
2d6451c9 | 1208 | char *p; |
52885fa2 JA |
1209 | |
1210 | ld->io_u_index[io_u->index] = io_u; | |
2d6451c9 AK |
1211 | |
1212 | if (!strcmp(td->io_ops->name, "io_uring_cmd")) { | |
1213 | p = PTR_ALIGN(ld->md_buf, page_mask) + td->o.mem_align; | |
1214 | p += o->md_per_io_size * io_u->index; | |
1215 | io_u->mmap_data = p; | |
5163f35e AK |
1216 | |
1217 | if (!o->pi_act) { | |
1218 | pi_data = calloc(1, sizeof(*pi_data)); | |
1219 | pi_data->io_flags |= o->prchk; | |
1220 | pi_data->apptag_mask = o->apptag_mask; | |
1221 | pi_data->apptag = o->apptag; | |
1222 | io_u->engine_data = pi_data; | |
1223 | } | |
2d6451c9 AK |
1224 | } |
1225 | ||
52885fa2 JA |
1226 | return 0; |
1227 | } | |
1228 | ||
5163f35e AK |
1229 | static void fio_ioring_io_u_free(struct thread_data *td, struct io_u *io_u) |
1230 | { | |
1231 | struct ioring_options *o = td->eo; | |
1232 | struct nvme_pi *pi; | |
1233 | ||
1234 | if (!strcmp(td->io_ops->name, "io_uring_cmd") && | |
1235 | (o->cmd_type == FIO_URING_CMD_NVME)) { | |
1236 | pi = io_u->engine_data; | |
1237 | free(pi); | |
1238 | io_u->engine_data = NULL; | |
1239 | } | |
1240 | } | |
1241 | ||
5ffd5626 JA |
1242 | static int fio_ioring_open_file(struct thread_data *td, struct fio_file *f) |
1243 | { | |
1244 | struct ioring_data *ld = td->io_ops_data; | |
1245 | struct ioring_options *o = td->eo; | |
1246 | ||
17318cf6 | 1247 | if (!ld || !o->registerfiles) |
5ffd5626 JA |
1248 | return generic_open_file(td, f); |
1249 | ||
1250 | f->fd = ld->fds[f->engine_pos]; | |
1251 | return 0; | |
1252 | } | |
1253 | ||
855dc4d4 AG |
1254 | static int fio_ioring_cmd_open_file(struct thread_data *td, struct fio_file *f) |
1255 | { | |
1256 | struct ioring_data *ld = td->io_ops_data; | |
1257 | struct ioring_options *o = td->eo; | |
1258 | ||
1259 | if (o->cmd_type == FIO_URING_CMD_NVME) { | |
1260 | struct nvme_data *data = NULL; | |
e7e5023b | 1261 | unsigned int lba_size = 0; |
671aa9f5 | 1262 | __u64 nlba = 0; |
855dc4d4 AG |
1263 | int ret; |
1264 | ||
1265 | /* Store the namespace-id and lba size. */ | |
1266 | data = FILE_ENG_DATA(f); | |
1267 | if (data == NULL) { | |
855dc4d4 | 1268 | data = calloc(1, sizeof(struct nvme_data)); |
3ee8311a | 1269 | ret = fio_nvme_get_info(f, &nlba, o->pi_act, data); |
e7e5023b AK |
1270 | if (ret) { |
1271 | free(data); | |
1272 | return ret; | |
1273 | } | |
855dc4d4 AG |
1274 | |
1275 | FILE_SET_ENG_DATA(f, data); | |
1276 | } | |
345fa8fd | 1277 | |
e7e5023b | 1278 | lba_size = data->lba_ext ? data->lba_ext : data->lba_size; |
345fa8fd AK |
1279 | |
1280 | for_each_rw_ddir(ddir) { | |
1281 | if (td->o.min_bs[ddir] % lba_size || | |
1282 | td->o.max_bs[ddir] % lba_size) { | |
1283 | if (data->lba_ext) | |
a904d182 AK |
1284 | log_err("%s: block size must be a multiple of (LBA data size + Metadata size)\n", |
1285 | f->file_name); | |
345fa8fd | 1286 | else |
a904d182 AK |
1287 | log_err("%s: block size must be a multiple of LBA data size\n", |
1288 | f->file_name); | |
1289 | td_verror(td, EINVAL, "fio_ioring_cmd_open_file"); | |
345fa8fd AK |
1290 | return 1; |
1291 | } | |
2d6451c9 AK |
1292 | if (data->ms && !data->lba_ext && ddir != DDIR_TRIM && |
1293 | (o->md_per_io_size < ((td->o.max_bs[ddir] / data->lba_size) * | |
1294 | data->ms))) { | |
1295 | log_err("%s: md_per_io_size should be at least %llu bytes\n", | |
1296 | f->file_name, | |
1297 | ((td->o.max_bs[ddir] / data->lba_size) * data->ms)); | |
1298 | td_verror(td, EINVAL, "fio_ioring_cmd_open_file"); | |
1299 | return 1; | |
1300 | } | |
345fa8fd | 1301 | } |
855dc4d4 AG |
1302 | } |
1303 | if (!ld || !o->registerfiles) | |
1304 | return generic_open_file(td, f); | |
1305 | ||
1306 | f->fd = ld->fds[f->engine_pos]; | |
1307 | return 0; | |
1308 | } | |
1309 | ||
5ffd5626 JA |
1310 | static int fio_ioring_close_file(struct thread_data *td, struct fio_file *f) |
1311 | { | |
17318cf6 | 1312 | struct ioring_data *ld = td->io_ops_data; |
5ffd5626 JA |
1313 | struct ioring_options *o = td->eo; |
1314 | ||
17318cf6 | 1315 | if (!ld || !o->registerfiles) |
5ffd5626 JA |
1316 | return generic_close_file(td, f); |
1317 | ||
1318 | f->fd = -1; | |
1319 | return 0; | |
1320 | } | |
1321 | ||
855dc4d4 AG |
1322 | static int fio_ioring_cmd_close_file(struct thread_data *td, |
1323 | struct fio_file *f) | |
1324 | { | |
1325 | struct ioring_data *ld = td->io_ops_data; | |
1326 | struct ioring_options *o = td->eo; | |
1327 | ||
1328 | if (o->cmd_type == FIO_URING_CMD_NVME) { | |
1329 | struct nvme_data *data = FILE_ENG_DATA(f); | |
1330 | ||
1331 | FILE_SET_ENG_DATA(f, NULL); | |
1332 | free(data); | |
1333 | } | |
1334 | if (!ld || !o->registerfiles) | |
1335 | return generic_close_file(td, f); | |
1336 | ||
1337 | f->fd = -1; | |
1338 | return 0; | |
1339 | } | |
1340 | ||
1341 | static int fio_ioring_cmd_get_file_size(struct thread_data *td, | |
1342 | struct fio_file *f) | |
1343 | { | |
1344 | struct ioring_options *o = td->eo; | |
1345 | ||
1346 | if (fio_file_size_known(f)) | |
1347 | return 0; | |
1348 | ||
1349 | if (o->cmd_type == FIO_URING_CMD_NVME) { | |
1350 | struct nvme_data *data = NULL; | |
671aa9f5 | 1351 | __u64 nlba = 0; |
855dc4d4 AG |
1352 | int ret; |
1353 | ||
855dc4d4 | 1354 | data = calloc(1, sizeof(struct nvme_data)); |
3ee8311a | 1355 | ret = fio_nvme_get_info(f, &nlba, o->pi_act, data); |
e7e5023b AK |
1356 | if (ret) { |
1357 | free(data); | |
1358 | return ret; | |
1359 | } | |
855dc4d4 | 1360 | |
e7e5023b | 1361 | f->real_file_size = data->lba_size * nlba; |
855dc4d4 AG |
1362 | fio_file_set_size_known(f); |
1363 | ||
1364 | FILE_SET_ENG_DATA(f, data); | |
1365 | return 0; | |
1366 | } | |
1367 | return generic_get_file_size(td, f); | |
1368 | } | |
1369 | ||
3d05e0ff AK |
1370 | static int fio_ioring_cmd_get_zoned_model(struct thread_data *td, |
1371 | struct fio_file *f, | |
1372 | enum zbd_zoned_model *model) | |
1373 | { | |
1374 | return fio_nvme_get_zoned_model(td, f, model); | |
1375 | } | |
1376 | ||
1377 | static int fio_ioring_cmd_report_zones(struct thread_data *td, | |
1378 | struct fio_file *f, uint64_t offset, | |
1379 | struct zbd_zone *zbdz, | |
1380 | unsigned int nr_zones) | |
1381 | { | |
1382 | return fio_nvme_report_zones(td, f, offset, zbdz, nr_zones); | |
1383 | } | |
1384 | ||
1385 | static int fio_ioring_cmd_reset_wp(struct thread_data *td, struct fio_file *f, | |
1386 | uint64_t offset, uint64_t length) | |
1387 | { | |
1388 | return fio_nvme_reset_wp(td, f, offset, length); | |
1389 | } | |
1390 | ||
1391 | static int fio_ioring_cmd_get_max_open_zones(struct thread_data *td, | |
1392 | struct fio_file *f, | |
1393 | unsigned int *max_open_zones) | |
1394 | { | |
1395 | return fio_nvme_get_max_open_zones(td, f, max_open_zones); | |
1396 | } | |
1397 | ||
a7e8aae0 KB |
1398 | static int fio_ioring_cmd_fetch_ruhs(struct thread_data *td, struct fio_file *f, |
1399 | struct fio_ruhs_info *fruhs_info) | |
1400 | { | |
1401 | struct nvme_fdp_ruh_status *ruhs; | |
1402 | int bytes, ret, i; | |
1403 | ||
98cd3c0e | 1404 | bytes = sizeof(*ruhs) + FDP_MAX_RUHS * sizeof(struct nvme_fdp_ruh_status_desc); |
a7e8aae0 KB |
1405 | ruhs = scalloc(1, bytes); |
1406 | if (!ruhs) | |
1407 | return -ENOMEM; | |
1408 | ||
1409 | ret = fio_nvme_iomgmt_ruhs(td, f, ruhs, bytes); | |
1410 | if (ret) | |
1411 | goto free; | |
1412 | ||
1413 | fruhs_info->nr_ruhs = le16_to_cpu(ruhs->nruhsd); | |
1414 | for (i = 0; i < fruhs_info->nr_ruhs; i++) | |
1415 | fruhs_info->plis[i] = le16_to_cpu(ruhs->ruhss[i].pid); | |
1416 | free: | |
1417 | sfree(ruhs); | |
1418 | return ret; | |
1419 | } | |
1420 | ||
855dc4d4 | 1421 | static struct ioengine_ops ioengine_uring = { |
bffad86f | 1422 | .name = "io_uring", |
52885fa2 | 1423 | .version = FIO_IOOPS_VERSION, |
4e7e7898 VF |
1424 | .flags = FIO_ASYNCIO_SYNC_TRIM | FIO_NO_OFFLOAD | |
1425 | FIO_ASYNCIO_SETS_ISSUE_TIME, | |
bffad86f JA |
1426 | .init = fio_ioring_init, |
1427 | .post_init = fio_ioring_post_init, | |
1428 | .io_u_init = fio_ioring_io_u_init, | |
1429 | .prep = fio_ioring_prep, | |
1430 | .queue = fio_ioring_queue, | |
1431 | .commit = fio_ioring_commit, | |
1432 | .getevents = fio_ioring_getevents, | |
1433 | .event = fio_ioring_event, | |
1434 | .cleanup = fio_ioring_cleanup, | |
5ffd5626 JA |
1435 | .open_file = fio_ioring_open_file, |
1436 | .close_file = fio_ioring_close_file, | |
52885fa2 JA |
1437 | .get_file_size = generic_get_file_size, |
1438 | .options = options, | |
bffad86f | 1439 | .option_struct_size = sizeof(struct ioring_options), |
52885fa2 JA |
1440 | }; |
1441 | ||
855dc4d4 AG |
1442 | static struct ioengine_ops ioengine_uring_cmd = { |
1443 | .name = "io_uring_cmd", | |
1444 | .version = FIO_IOOPS_VERSION, | |
4885a6eb | 1445 | .flags = FIO_NO_OFFLOAD | FIO_MEMALIGN | FIO_RAWIO | |
4e7e7898 | 1446 | FIO_ASYNCIO_SETS_ISSUE_TIME, |
855dc4d4 AG |
1447 | .init = fio_ioring_init, |
1448 | .post_init = fio_ioring_cmd_post_init, | |
1449 | .io_u_init = fio_ioring_io_u_init, | |
5163f35e | 1450 | .io_u_free = fio_ioring_io_u_free, |
855dc4d4 AG |
1451 | .prep = fio_ioring_cmd_prep, |
1452 | .queue = fio_ioring_queue, | |
1453 | .commit = fio_ioring_commit, | |
1454 | .getevents = fio_ioring_getevents, | |
1455 | .event = fio_ioring_cmd_event, | |
1456 | .cleanup = fio_ioring_cleanup, | |
1457 | .open_file = fio_ioring_cmd_open_file, | |
1458 | .close_file = fio_ioring_cmd_close_file, | |
1459 | .get_file_size = fio_ioring_cmd_get_file_size, | |
3d05e0ff AK |
1460 | .get_zoned_model = fio_ioring_cmd_get_zoned_model, |
1461 | .report_zones = fio_ioring_cmd_report_zones, | |
1462 | .reset_wp = fio_ioring_cmd_reset_wp, | |
1463 | .get_max_open_zones = fio_ioring_cmd_get_max_open_zones, | |
855dc4d4 AG |
1464 | .options = options, |
1465 | .option_struct_size = sizeof(struct ioring_options), | |
a7e8aae0 | 1466 | .fdp_fetch_ruhs = fio_ioring_cmd_fetch_ruhs, |
855dc4d4 AG |
1467 | }; |
1468 | ||
bffad86f | 1469 | static void fio_init fio_ioring_register(void) |
52885fa2 | 1470 | { |
855dc4d4 AG |
1471 | register_ioengine(&ioengine_uring); |
1472 | register_ioengine(&ioengine_uring_cmd); | |
52885fa2 JA |
1473 | } |
1474 | ||
bffad86f | 1475 | static void fio_exit fio_ioring_unregister(void) |
52885fa2 | 1476 | { |
855dc4d4 AG |
1477 | unregister_ioengine(&ioengine_uring); |
1478 | unregister_ioengine(&ioengine_uring_cmd); | |
52885fa2 | 1479 | } |
1f90e9bb | 1480 | #endif |