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" |
ba342e58 | 21 | #include "../verify.h" |
52885fa2 | 22 | |
bffad86f | 23 | #ifdef ARCH_HAVE_IOURING |
52885fa2 | 24 | |
57fa61f0 | 25 | #include "../lib/types.h" |
f3e769a4 | 26 | #include "../os/linux/io_uring.h" |
e9f6567a | 27 | #include "cmdprio.h" |
16be6037 | 28 | #include "zbd.h" |
855dc4d4 AG |
29 | #include "nvme.h" |
30 | ||
31 | #include <sys/stat.h> | |
32 | ||
33 | enum uring_cmd_type { | |
34 | FIO_URING_CMD_NVME = 1, | |
35 | }; | |
9a2d78b3 | 36 | |
bffad86f | 37 | struct io_sq_ring { |
e2239016 JA |
38 | unsigned *head; |
39 | unsigned *tail; | |
40 | unsigned *ring_mask; | |
41 | unsigned *ring_entries; | |
42 | unsigned *flags; | |
43 | unsigned *array; | |
52885fa2 JA |
44 | }; |
45 | ||
bffad86f | 46 | struct io_cq_ring { |
e2239016 JA |
47 | unsigned *head; |
48 | unsigned *tail; | |
49 | unsigned *ring_mask; | |
50 | unsigned *ring_entries; | |
f0403f94 | 51 | struct io_uring_cqe *cqes; |
9a2d78b3 JA |
52 | }; |
53 | ||
bffad86f | 54 | struct ioring_mmap { |
9a2d78b3 JA |
55 | void *ptr; |
56 | size_t len; | |
52885fa2 JA |
57 | }; |
58 | ||
bffad86f | 59 | struct ioring_data { |
9a2d78b3 JA |
60 | int ring_fd; |
61 | ||
52885fa2 | 62 | struct io_u **io_u_index; |
2d6451c9 | 63 | char *md_buf; |
52885fa2 | 64 | |
5ffd5626 JA |
65 | int *fds; |
66 | ||
bffad86f | 67 | struct io_sq_ring sq_ring; |
f0403f94 | 68 | struct io_uring_sqe *sqes; |
9a2d78b3 | 69 | struct iovec *iovecs; |
b87aa01a | 70 | unsigned sq_ring_mask; |
52885fa2 | 71 | |
bffad86f | 72 | struct io_cq_ring cq_ring; |
b87aa01a | 73 | unsigned cq_ring_mask; |
52885fa2 JA |
74 | |
75 | int queued; | |
76 | int cq_ring_off; | |
b87aa01a | 77 | unsigned iodepth; |
5a59a81d | 78 | int prepped; |
96563db9 | 79 | |
bffad86f | 80 | struct ioring_mmap mmap[3]; |
d6cbeab4 NC |
81 | |
82 | struct cmdprio cmdprio; | |
4885a6eb VF |
83 | |
84 | struct nvme_dsm_range *dsm; | |
52885fa2 JA |
85 | }; |
86 | ||
bffad86f | 87 | struct ioring_options { |
a48f0cc7 | 88 | struct thread_data *td; |
52885fa2 | 89 | unsigned int hipri; |
d6cbeab4 | 90 | struct cmdprio_options cmdprio_options; |
52885fa2 | 91 | unsigned int fixedbufs; |
5ffd5626 | 92 | unsigned int registerfiles; |
3d7d00a3 | 93 | unsigned int sqpoll_thread; |
2ea53ca3 JA |
94 | unsigned int sqpoll_set; |
95 | unsigned int sqpoll_cpu; | |
b10b1e70 | 96 | unsigned int nonvectored; |
4a87b584 | 97 | unsigned int uncached; |
7d42e66e | 98 | unsigned int nowait; |
5a59a81d | 99 | unsigned int force_async; |
2d6451c9 | 100 | unsigned int md_per_io_size; |
3ee8311a AK |
101 | unsigned int pi_act; |
102 | unsigned int apptag; | |
103 | unsigned int apptag_mask; | |
104 | unsigned int prchk; | |
105 | char *pi_chk; | |
855dc4d4 | 106 | enum uring_cmd_type cmd_type; |
52885fa2 JA |
107 | }; |
108 | ||
b10b1e70 JA |
109 | static const int ddir_to_op[2][2] = { |
110 | { IORING_OP_READV, IORING_OP_READ }, | |
111 | { IORING_OP_WRITEV, IORING_OP_WRITE } | |
112 | }; | |
113 | ||
3f1e3af7 KB |
114 | static const int fixed_ddir_to_op[2] = { |
115 | IORING_OP_READ_FIXED, | |
116 | IORING_OP_WRITE_FIXED | |
117 | }; | |
118 | ||
2ea53ca3 | 119 | static int fio_ioring_sqpoll_cb(void *data, unsigned long long *val) |
a90cd050 | 120 | { |
bffad86f | 121 | struct ioring_options *o = data; |
a90cd050 | 122 | |
2ea53ca3 JA |
123 | o->sqpoll_cpu = *val; |
124 | o->sqpoll_set = 1; | |
a90cd050 JA |
125 | return 0; |
126 | } | |
127 | ||
52885fa2 JA |
128 | static struct fio_option options[] = { |
129 | { | |
130 | .name = "hipri", | |
131 | .lname = "High Priority", | |
132 | .type = FIO_OPT_STR_SET, | |
bffad86f | 133 | .off1 = offsetof(struct ioring_options, hipri), |
52885fa2 JA |
134 | .help = "Use polled IO completions", |
135 | .category = FIO_OPT_C_ENGINE, | |
27f436d9 | 136 | .group = FIO_OPT_G_IOURING, |
52885fa2 JA |
137 | }, |
138 | { | |
139 | .name = "fixedbufs", | |
140 | .lname = "Fixed (pre-mapped) IO buffers", | |
141 | .type = FIO_OPT_STR_SET, | |
bffad86f | 142 | .off1 = offsetof(struct ioring_options, fixedbufs), |
52885fa2 JA |
143 | .help = "Pre map IO buffers", |
144 | .category = FIO_OPT_C_ENGINE, | |
27f436d9 | 145 | .group = FIO_OPT_G_IOURING, |
52885fa2 | 146 | }, |
5ffd5626 JA |
147 | { |
148 | .name = "registerfiles", | |
149 | .lname = "Register file set", | |
150 | .type = FIO_OPT_STR_SET, | |
151 | .off1 = offsetof(struct ioring_options, registerfiles), | |
152 | .help = "Pre-open/register files", | |
153 | .category = FIO_OPT_C_ENGINE, | |
27f436d9 | 154 | .group = FIO_OPT_G_IOURING, |
5ffd5626 | 155 | }, |
771c9901 JA |
156 | { |
157 | .name = "sqthread_poll", | |
3d7d00a3 | 158 | .lname = "Kernel SQ thread polling", |
d6f936d1 | 159 | .type = FIO_OPT_STR_SET, |
3d7d00a3 JA |
160 | .off1 = offsetof(struct ioring_options, sqpoll_thread), |
161 | .help = "Offload submission/completion to kernel thread", | |
162 | .category = FIO_OPT_C_ENGINE, | |
27f436d9 | 163 | .group = FIO_OPT_G_IOURING, |
3d7d00a3 JA |
164 | }, |
165 | { | |
166 | .name = "sqthread_poll_cpu", | |
167 | .lname = "SQ Thread Poll CPU", | |
2ea53ca3 JA |
168 | .type = FIO_OPT_INT, |
169 | .cb = fio_ioring_sqpoll_cb, | |
3d7d00a3 | 170 | .help = "What CPU to run SQ thread polling on", |
a90cd050 | 171 | .category = FIO_OPT_C_ENGINE, |
27f436d9 | 172 | .group = FIO_OPT_G_IOURING, |
a90cd050 | 173 | }, |
b10b1e70 JA |
174 | { |
175 | .name = "nonvectored", | |
176 | .lname = "Non-vectored", | |
177 | .type = FIO_OPT_INT, | |
178 | .off1 = offsetof(struct ioring_options, nonvectored), | |
556d8415 | 179 | .def = "-1", |
b10b1e70 JA |
180 | .help = "Use non-vectored read/write commands", |
181 | .category = FIO_OPT_C_ENGINE, | |
182 | .group = FIO_OPT_G_IOURING, | |
183 | }, | |
4a87b584 JA |
184 | { |
185 | .name = "uncached", | |
186 | .lname = "Uncached", | |
187 | .type = FIO_OPT_INT, | |
188 | .off1 = offsetof(struct ioring_options, uncached), | |
189 | .help = "Use RWF_UNCACHED for buffered read/writes", | |
190 | .category = FIO_OPT_C_ENGINE, | |
191 | .group = FIO_OPT_G_IOURING, | |
192 | }, | |
7d42e66e KK |
193 | { |
194 | .name = "nowait", | |
195 | .lname = "RWF_NOWAIT", | |
196 | .type = FIO_OPT_BOOL, | |
197 | .off1 = offsetof(struct ioring_options, nowait), | |
198 | .help = "Use RWF_NOWAIT for reads/writes", | |
199 | .category = FIO_OPT_C_ENGINE, | |
200 | .group = FIO_OPT_G_IOURING, | |
201 | }, | |
5a59a81d JA |
202 | { |
203 | .name = "force_async", | |
204 | .lname = "Force async", | |
205 | .type = FIO_OPT_INT, | |
206 | .off1 = offsetof(struct ioring_options, force_async), | |
207 | .help = "Set IOSQE_ASYNC every N requests", | |
208 | .category = FIO_OPT_C_ENGINE, | |
209 | .group = FIO_OPT_G_IOURING, | |
210 | }, | |
855dc4d4 AG |
211 | { |
212 | .name = "cmd_type", | |
213 | .lname = "Uring cmd type", | |
214 | .type = FIO_OPT_STR, | |
215 | .off1 = offsetof(struct ioring_options, cmd_type), | |
216 | .help = "Specify uring-cmd type", | |
217 | .def = "nvme", | |
218 | .posval = { | |
219 | { .ival = "nvme", | |
220 | .oval = FIO_URING_CMD_NVME, | |
221 | .help = "Issue nvme-uring-cmd", | |
222 | }, | |
223 | }, | |
224 | .category = FIO_OPT_C_ENGINE, | |
225 | .group = FIO_OPT_G_IOURING, | |
226 | }, | |
2838f77a | 227 | CMDPRIO_OPTIONS(struct ioring_options, FIO_OPT_G_IOURING), |
2d6451c9 AK |
228 | { |
229 | .name = "md_per_io_size", | |
230 | .lname = "Separate Metadata Buffer Size per I/O", | |
231 | .type = FIO_OPT_INT, | |
232 | .off1 = offsetof(struct ioring_options, md_per_io_size), | |
233 | .def = "0", | |
234 | .help = "Size of separate metadata buffer per I/O (Default: 0)", | |
235 | .category = FIO_OPT_C_ENGINE, | |
236 | .group = FIO_OPT_G_IOURING, | |
237 | }, | |
3ee8311a AK |
238 | { |
239 | .name = "pi_act", | |
240 | .lname = "Protection Information Action", | |
241 | .type = FIO_OPT_BOOL, | |
242 | .off1 = offsetof(struct ioring_options, pi_act), | |
243 | .def = "1", | |
244 | .help = "Protection Information Action bit (pi_act=1 or pi_act=0)", | |
245 | .category = FIO_OPT_C_ENGINE, | |
246 | .group = FIO_OPT_G_IOURING, | |
247 | }, | |
248 | { | |
249 | .name = "pi_chk", | |
250 | .lname = "Protection Information Check", | |
251 | .type = FIO_OPT_STR_STORE, | |
252 | .off1 = offsetof(struct ioring_options, pi_chk), | |
253 | .def = NULL, | |
254 | .help = "Control of Protection Information Checking (pi_chk=GUARD,REFTAG,APPTAG)", | |
255 | .category = FIO_OPT_C_ENGINE, | |
256 | .group = FIO_OPT_G_IOURING, | |
257 | }, | |
258 | { | |
259 | .name = "apptag", | |
260 | .lname = "Application Tag used in Protection Information", | |
261 | .type = FIO_OPT_INT, | |
262 | .off1 = offsetof(struct ioring_options, apptag), | |
263 | .def = "0x1234", | |
264 | .help = "Application Tag used in Protection Information field (Default: 0x1234)", | |
265 | .category = FIO_OPT_C_ENGINE, | |
266 | .group = FIO_OPT_G_IOURING, | |
267 | }, | |
268 | { | |
269 | .name = "apptag_mask", | |
270 | .lname = "Application Tag Mask", | |
271 | .type = FIO_OPT_INT, | |
272 | .off1 = offsetof(struct ioring_options, apptag_mask), | |
273 | .def = "0xffff", | |
274 | .help = "Application Tag Mask used with Application Tag (Default: 0xffff)", | |
275 | .category = FIO_OPT_C_ENGINE, | |
276 | .group = FIO_OPT_G_IOURING, | |
277 | }, | |
52885fa2 JA |
278 | { |
279 | .name = NULL, | |
280 | }, | |
281 | }; | |
282 | ||
bffad86f | 283 | static int io_uring_enter(struct ioring_data *ld, unsigned int to_submit, |
52885fa2 JA |
284 | unsigned int min_complete, unsigned int flags) |
285 | { | |
c377f4f8 JA |
286 | #ifdef FIO_ARCH_HAS_SYSCALL |
287 | return __do_syscall6(__NR_io_uring_enter, ld->ring_fd, to_submit, | |
288 | min_complete, flags, NULL, 0); | |
289 | #else | |
bfed648c | 290 | return syscall(__NR_io_uring_enter, ld->ring_fd, to_submit, |
521164fa | 291 | min_complete, flags, NULL, 0); |
c377f4f8 | 292 | #endif |
52885fa2 JA |
293 | } |
294 | ||
bffad86f | 295 | static int fio_ioring_prep(struct thread_data *td, struct io_u *io_u) |
52885fa2 | 296 | { |
bffad86f | 297 | struct ioring_data *ld = td->io_ops_data; |
cfcc8564 | 298 | struct ioring_options *o = td->eo; |
52885fa2 | 299 | struct fio_file *f = io_u->file; |
f0403f94 | 300 | struct io_uring_sqe *sqe; |
52885fa2 | 301 | |
f0403f94 | 302 | sqe = &ld->sqes[io_u->index]; |
34d6090e | 303 | |
5ffd5626 JA |
304 | if (o->registerfiles) { |
305 | sqe->fd = f->engine_pos; | |
306 | sqe->flags = IOSQE_FIXED_FILE; | |
307 | } else { | |
308 | sqe->fd = f->fd; | |
87b69ef2 | 309 | sqe->flags = 0; |
5ffd5626 | 310 | } |
52885fa2 | 311 | |
e3970057 | 312 | if (io_u->ddir == DDIR_READ || io_u->ddir == DDIR_WRITE) { |
f0403f94 | 313 | if (o->fixedbufs) { |
3f1e3af7 | 314 | sqe->opcode = fixed_ddir_to_op[io_u->ddir]; |
919850d2 | 315 | sqe->addr = (unsigned long) io_u->xfer_buf; |
f0403f94 | 316 | sqe->len = io_u->xfer_buflen; |
2ea53ca3 | 317 | sqe->buf_index = io_u->index; |
cfcc8564 | 318 | } else { |
832faaaf JA |
319 | struct iovec *iov = &ld->iovecs[io_u->index]; |
320 | ||
321 | /* | |
322 | * Update based on actual io_u, requeue could have | |
323 | * adjusted these | |
324 | */ | |
325 | iov->iov_base = io_u->xfer_buf; | |
326 | iov->iov_len = io_u->xfer_buflen; | |
327 | ||
3f1e3af7 | 328 | sqe->opcode = ddir_to_op[io_u->ddir][!!o->nonvectored]; |
b10b1e70 | 329 | if (o->nonvectored) { |
832faaaf JA |
330 | sqe->addr = (unsigned long) iov->iov_base; |
331 | sqe->len = iov->iov_len; | |
b10b1e70 | 332 | } else { |
832faaaf | 333 | sqe->addr = (unsigned long) iov; |
b10b1e70 JA |
334 | sqe->len = 1; |
335 | } | |
cfcc8564 | 336 | } |
fd70e361 | 337 | sqe->rw_flags = 0; |
4a87b584 | 338 | if (!td->o.odirect && o->uncached) |
fd70e361 | 339 | sqe->rw_flags |= RWF_UNCACHED; |
7d42e66e KK |
340 | if (o->nowait) |
341 | sqe->rw_flags |= RWF_NOWAIT; | |
8ff6b289 NC |
342 | |
343 | /* | |
344 | * Since io_uring can have a submission context (sqthread_poll) | |
345 | * that is different from the process context, we cannot rely on | |
79012fec DLM |
346 | * the IO priority set by ioprio_set() (options prio, prioclass, |
347 | * and priohint) to be inherited. | |
8ff6b289 NC |
348 | * td->ioprio will have the value of the "default prio", so set |
349 | * this unconditionally. This value might get overridden by | |
ff00f247 | 350 | * fio_ioring_cmdprio_prep() if the option cmdprio_percentage or |
8ff6b289 NC |
351 | * cmdprio_bssplit is used. |
352 | */ | |
353 | sqe->ioprio = td->ioprio; | |
f0403f94 | 354 | sqe->off = io_u->offset; |
48e698fa | 355 | } else if (ddir_sync(io_u->ddir)) { |
7c70f506 | 356 | sqe->ioprio = 0; |
01387bfe AF |
357 | if (io_u->ddir == DDIR_SYNC_FILE_RANGE) { |
358 | sqe->off = f->first_write; | |
359 | sqe->len = f->last_write - f->first_write; | |
360 | sqe->sync_range_flags = td->o.sync_file_range; | |
361 | sqe->opcode = IORING_OP_SYNC_FILE_RANGE; | |
362 | } else { | |
7c70f506 JA |
363 | sqe->off = 0; |
364 | sqe->addr = 0; | |
365 | sqe->len = 0; | |
01387bfe AF |
366 | if (io_u->ddir == DDIR_DATASYNC) |
367 | sqe->fsync_flags |= IORING_FSYNC_DATASYNC; | |
368 | sqe->opcode = IORING_OP_FSYNC; | |
369 | } | |
48e698fa | 370 | } |
52885fa2 | 371 | |
5a59a81d JA |
372 | if (o->force_async && ++ld->prepped == o->force_async) { |
373 | ld->prepped = 0; | |
374 | sqe->flags |= IOSQE_ASYNC; | |
375 | } | |
376 | ||
48e698fa | 377 | sqe->user_data = (unsigned long) io_u; |
52885fa2 JA |
378 | return 0; |
379 | } | |
380 | ||
855dc4d4 AG |
381 | static int fio_ioring_cmd_prep(struct thread_data *td, struct io_u *io_u) |
382 | { | |
383 | struct ioring_data *ld = td->io_ops_data; | |
384 | struct ioring_options *o = td->eo; | |
385 | struct fio_file *f = io_u->file; | |
3ce6a3de | 386 | struct nvme_uring_cmd *cmd; |
855dc4d4 | 387 | struct io_uring_sqe *sqe; |
855dc4d4 | 388 | |
3ce6a3de JA |
389 | /* only supports nvme_uring_cmd */ |
390 | if (o->cmd_type != FIO_URING_CMD_NVME) | |
391 | return -EINVAL; | |
855dc4d4 | 392 | |
4885a6eb | 393 | if (io_u->ddir == DDIR_TRIM && td->io_ops->flags & FIO_ASYNCIO_SYNC_TRIM) |
16be6037 AK |
394 | return 0; |
395 | ||
3ce6a3de | 396 | sqe = &ld->sqes[(io_u->index) << 1]; |
855dc4d4 | 397 | |
3ce6a3de JA |
398 | if (o->registerfiles) { |
399 | sqe->fd = f->engine_pos; | |
400 | sqe->flags = IOSQE_FIXED_FILE; | |
401 | } else { | |
402 | sqe->fd = f->fd; | |
403 | } | |
404 | sqe->rw_flags = 0; | |
405 | if (!td->o.odirect && o->uncached) | |
406 | sqe->rw_flags |= RWF_UNCACHED; | |
407 | if (o->nowait) | |
408 | sqe->rw_flags |= RWF_NOWAIT; | |
855dc4d4 | 409 | |
3ce6a3de JA |
410 | sqe->opcode = IORING_OP_URING_CMD; |
411 | sqe->user_data = (unsigned long) io_u; | |
412 | if (o->nonvectored) | |
413 | sqe->cmd_op = NVME_URING_CMD_IO; | |
414 | else | |
415 | sqe->cmd_op = NVME_URING_CMD_IO_VEC; | |
416 | if (o->force_async && ++ld->prepped == o->force_async) { | |
417 | ld->prepped = 0; | |
418 | sqe->flags |= IOSQE_ASYNC; | |
855dc4d4 | 419 | } |
0ebd3bf6 AG |
420 | if (o->fixedbufs) { |
421 | sqe->uring_cmd_flags = IORING_URING_CMD_FIXED; | |
422 | sqe->buf_index = io_u->index; | |
423 | } | |
3ce6a3de JA |
424 | |
425 | cmd = (struct nvme_uring_cmd *)sqe->cmd; | |
426 | return fio_nvme_uring_cmd_prep(cmd, io_u, | |
4885a6eb VF |
427 | o->nonvectored ? NULL : &ld->iovecs[io_u->index], |
428 | &ld->dsm[io_u->index]); | |
855dc4d4 AG |
429 | } |
430 | ||
bffad86f | 431 | static struct io_u *fio_ioring_event(struct thread_data *td, int event) |
52885fa2 | 432 | { |
bffad86f | 433 | struct ioring_data *ld = td->io_ops_data; |
f0403f94 | 434 | struct io_uring_cqe *cqe; |
52885fa2 | 435 | struct io_u *io_u; |
b87aa01a | 436 | unsigned index; |
52885fa2 | 437 | |
b87aa01a | 438 | index = (event + ld->cq_ring_off) & ld->cq_ring_mask; |
52885fa2 | 439 | |
f0403f94 | 440 | cqe = &ld->cq_ring.cqes[index]; |
e3466352 | 441 | io_u = (struct io_u *) (uintptr_t) cqe->user_data; |
52885fa2 | 442 | |
f0403f94 JA |
443 | if (cqe->res != io_u->xfer_buflen) { |
444 | if (cqe->res > io_u->xfer_buflen) | |
445 | io_u->error = -cqe->res; | |
52885fa2 | 446 | else |
f0403f94 | 447 | io_u->resid = io_u->xfer_buflen - cqe->res; |
52885fa2 JA |
448 | } else |
449 | io_u->error = 0; | |
450 | ||
451 | return io_u; | |
452 | } | |
453 | ||
855dc4d4 AG |
454 | static struct io_u *fio_ioring_cmd_event(struct thread_data *td, int event) |
455 | { | |
456 | struct ioring_data *ld = td->io_ops_data; | |
457 | struct ioring_options *o = td->eo; | |
458 | struct io_uring_cqe *cqe; | |
459 | struct io_u *io_u; | |
5163f35e | 460 | struct nvme_data *data; |
855dc4d4 | 461 | unsigned index; |
5163f35e | 462 | int ret; |
855dc4d4 AG |
463 | |
464 | index = (event + ld->cq_ring_off) & ld->cq_ring_mask; | |
465 | if (o->cmd_type == FIO_URING_CMD_NVME) | |
466 | index <<= 1; | |
467 | ||
468 | cqe = &ld->cq_ring.cqes[index]; | |
469 | io_u = (struct io_u *) (uintptr_t) cqe->user_data; | |
470 | ||
471 | if (cqe->res != 0) | |
472 | io_u->error = -cqe->res; | |
473 | else | |
474 | io_u->error = 0; | |
475 | ||
5163f35e AK |
476 | if (o->cmd_type == FIO_URING_CMD_NVME) { |
477 | data = FILE_ENG_DATA(io_u->file); | |
478 | if (data->pi_type && (io_u->ddir == DDIR_READ) && !o->pi_act) { | |
479 | ret = fio_nvme_pi_verify(data, io_u); | |
480 | if (ret) | |
481 | io_u->error = ret; | |
482 | } | |
483 | } | |
484 | ||
855dc4d4 AG |
485 | return io_u; |
486 | } | |
487 | ||
bffad86f | 488 | static int fio_ioring_cqring_reap(struct thread_data *td, unsigned int events, |
52885fa2 JA |
489 | unsigned int max) |
490 | { | |
bffad86f JA |
491 | struct ioring_data *ld = td->io_ops_data; |
492 | struct io_cq_ring *ring = &ld->cq_ring; | |
e2239016 | 493 | unsigned head, reaped = 0; |
52885fa2 | 494 | |
9a2d78b3 | 495 | head = *ring->head; |
52885fa2 | 496 | do { |
9e26aff9 | 497 | if (head == atomic_load_acquire(ring->tail)) |
52885fa2 JA |
498 | break; |
499 | reaped++; | |
500 | head++; | |
52885fa2 JA |
501 | } while (reaped + events < max); |
502 | ||
76ce63dd AB |
503 | if (reaped) |
504 | atomic_store_release(ring->head, head); | |
505 | ||
52885fa2 JA |
506 | return reaped; |
507 | } | |
508 | ||
bffad86f JA |
509 | static int fio_ioring_getevents(struct thread_data *td, unsigned int min, |
510 | unsigned int max, const struct timespec *t) | |
52885fa2 | 511 | { |
bffad86f | 512 | struct ioring_data *ld = td->io_ops_data; |
52885fa2 | 513 | unsigned actual_min = td->o.iodepth_batch_complete_min == 0 ? 0 : min; |
bffad86f JA |
514 | struct ioring_options *o = td->eo; |
515 | struct io_cq_ring *ring = &ld->cq_ring; | |
b87aa01a JA |
516 | unsigned events = 0; |
517 | int r; | |
52885fa2 | 518 | |
9a2d78b3 | 519 | ld->cq_ring_off = *ring->head; |
52885fa2 | 520 | do { |
bffad86f | 521 | r = fio_ioring_cqring_reap(td, events, max); |
52885fa2 JA |
522 | if (r) { |
523 | events += r; | |
ae8646a1 | 524 | max -= r; |
f7cbbbf8 ST |
525 | if (actual_min != 0) |
526 | actual_min -= r; | |
52885fa2 JA |
527 | continue; |
528 | } | |
529 | ||
3d7d00a3 | 530 | if (!o->sqpoll_thread) { |
9a2d78b3 JA |
531 | r = io_uring_enter(ld, 0, actual_min, |
532 | IORING_ENTER_GETEVENTS); | |
771c9901 | 533 | if (r < 0) { |
f6abd731 | 534 | if (errno == EAGAIN || errno == EINTR) |
771c9901 | 535 | continue; |
1816895b | 536 | r = -errno; |
9a2d78b3 | 537 | td_verror(td, errno, "io_uring_enter"); |
771c9901 JA |
538 | break; |
539 | } | |
52885fa2 JA |
540 | } |
541 | } while (events < min); | |
542 | ||
543 | return r < 0 ? r : events; | |
544 | } | |
545 | ||
3ee8311a AK |
546 | static inline void fio_ioring_cmd_nvme_pi(struct thread_data *td, |
547 | struct io_u *io_u) | |
548 | { | |
549 | struct ioring_data *ld = td->io_ops_data; | |
550 | struct ioring_options *o = td->eo; | |
551 | struct nvme_uring_cmd *cmd; | |
552 | struct io_uring_sqe *sqe; | |
553 | struct nvme_cmd_ext_io_opts ext_opts = {0}; | |
554 | struct nvme_data *data = FILE_ENG_DATA(io_u->file); | |
555 | ||
556 | if (io_u->ddir == DDIR_TRIM) | |
557 | return; | |
558 | ||
559 | sqe = &ld->sqes[(io_u->index) << 1]; | |
560 | cmd = (struct nvme_uring_cmd *)sqe->cmd; | |
561 | ||
562 | if (data->pi_type) { | |
563 | if (o->pi_act) | |
564 | ext_opts.io_flags |= NVME_IO_PRINFO_PRACT; | |
565 | ext_opts.io_flags |= o->prchk; | |
566 | ext_opts.apptag = o->apptag; | |
567 | ext_opts.apptag_mask = o->apptag_mask; | |
568 | } | |
569 | ||
570 | fio_nvme_pi_fill(cmd, io_u, &ext_opts); | |
571 | } | |
572 | ||
127715b6 NC |
573 | static inline void fio_ioring_cmdprio_prep(struct thread_data *td, |
574 | struct io_u *io_u) | |
b2a432bf | 575 | { |
b2a432bf | 576 | struct ioring_data *ld = td->io_ops_data; |
d6cbeab4 | 577 | struct cmdprio *cmdprio = &ld->cmdprio; |
127715b6 NC |
578 | |
579 | if (fio_cmdprio_set_ioprio(td, cmdprio, io_u)) | |
580 | ld->sqes[io_u->index].ioprio = io_u->ioprio; | |
b2a432bf PC |
581 | } |
582 | ||
bffad86f JA |
583 | static enum fio_q_status fio_ioring_queue(struct thread_data *td, |
584 | struct io_u *io_u) | |
52885fa2 | 585 | { |
bffad86f | 586 | struct ioring_data *ld = td->io_ops_data; |
3ee8311a | 587 | struct ioring_options *o = td->eo; |
bffad86f | 588 | struct io_sq_ring *ring = &ld->sq_ring; |
52885fa2 JA |
589 | unsigned tail, next_tail; |
590 | ||
591 | fio_ro_check(td, io_u); | |
592 | ||
b87aa01a | 593 | if (ld->queued == ld->iodepth) |
52885fa2 JA |
594 | return FIO_Q_BUSY; |
595 | ||
4885a6eb | 596 | if (io_u->ddir == DDIR_TRIM && td->io_ops->flags & FIO_ASYNCIO_SYNC_TRIM) { |
52885fa2 JA |
597 | if (ld->queued) |
598 | return FIO_Q_BUSY; | |
599 | ||
7f57e30f | 600 | do_io_u_trim(td, io_u); |
16be6037 | 601 | |
52885fa2 JA |
602 | io_u_mark_submit(td, 1); |
603 | io_u_mark_complete(td, 1); | |
604 | return FIO_Q_COMPLETED; | |
605 | } | |
606 | ||
9a2d78b3 | 607 | tail = *ring->tail; |
52885fa2 | 608 | next_tail = tail + 1; |
5c15a911 | 609 | if (next_tail == atomic_load_relaxed(ring->head)) |
52885fa2 JA |
610 | return FIO_Q_BUSY; |
611 | ||
d6cbeab4 | 612 | if (ld->cmdprio.mode != CMDPRIO_MODE_NONE) |
ff00f247 NC |
613 | fio_ioring_cmdprio_prep(td, io_u); |
614 | ||
3ee8311a AK |
615 | if (!strcmp(td->io_ops->name, "io_uring_cmd") && |
616 | o->cmd_type == FIO_URING_CMD_NVME) | |
617 | fio_ioring_cmd_nvme_pi(td, io_u); | |
618 | ||
b87aa01a | 619 | ring->array[tail & ld->sq_ring_mask] = io_u->index; |
9e26aff9 | 620 | atomic_store_release(ring->tail, next_tail); |
52885fa2 JA |
621 | |
622 | ld->queued++; | |
623 | return FIO_Q_QUEUED; | |
624 | } | |
625 | ||
bffad86f | 626 | static void fio_ioring_queued(struct thread_data *td, int start, int nr) |
52885fa2 | 627 | { |
bffad86f | 628 | struct ioring_data *ld = td->io_ops_data; |
52885fa2 JA |
629 | struct timespec now; |
630 | ||
631 | if (!fio_fill_issue_time(td)) | |
632 | return; | |
633 | ||
634 | fio_gettime(&now, NULL); | |
635 | ||
636 | while (nr--) { | |
bffad86f | 637 | struct io_sq_ring *ring = &ld->sq_ring; |
9a2d78b3 | 638 | int index = ring->array[start & ld->sq_ring_mask]; |
f8289afc | 639 | struct io_u *io_u = ld->io_u_index[index]; |
52885fa2 JA |
640 | |
641 | memcpy(&io_u->issue_time, &now, sizeof(now)); | |
642 | io_u_queued(td, io_u); | |
643 | ||
644 | start++; | |
52885fa2 | 645 | } |
39f56400 VF |
646 | |
647 | /* | |
648 | * only used for iolog | |
649 | */ | |
650 | if (td->o.read_iolog_file) | |
651 | memcpy(&td->last_issue, &now, sizeof(now)); | |
52885fa2 JA |
652 | } |
653 | ||
bffad86f | 654 | static int fio_ioring_commit(struct thread_data *td) |
52885fa2 | 655 | { |
bffad86f JA |
656 | struct ioring_data *ld = td->io_ops_data; |
657 | struct ioring_options *o = td->eo; | |
52885fa2 JA |
658 | int ret; |
659 | ||
660 | if (!ld->queued) | |
661 | return 0; | |
662 | ||
3d7d00a3 JA |
663 | /* |
664 | * Kernel side does submission. just need to check if the ring is | |
665 | * flagged as needing a kick, if so, call io_uring_enter(). This | |
666 | * only happens if we've been idle too long. | |
667 | */ | |
668 | if (o->sqpoll_thread) { | |
bffad86f | 669 | struct io_sq_ring *ring = &ld->sq_ring; |
10bad6b9 | 670 | unsigned start = *ld->sq_ring.tail - ld->queued; |
2dd96cc4 | 671 | unsigned flags; |
4cdbc048 | 672 | |
5c15a911 | 673 | flags = atomic_load_relaxed(ring->flags); |
2dd96cc4 | 674 | if (flags & IORING_SQ_NEED_WAKEUP) |
b532dd6d JA |
675 | io_uring_enter(ld, ld->queued, 0, |
676 | IORING_ENTER_SQ_WAKEUP); | |
c011bf12 AK |
677 | fio_ioring_queued(td, start, ld->queued); |
678 | io_u_mark_submit(td, ld->queued); | |
679 | ||
771c9901 JA |
680 | ld->queued = 0; |
681 | return 0; | |
682 | } | |
683 | ||
52885fa2 | 684 | do { |
9a2d78b3 | 685 | unsigned start = *ld->sq_ring.head; |
52885fa2 JA |
686 | long nr = ld->queued; |
687 | ||
9a2d78b3 | 688 | ret = io_uring_enter(ld, nr, 0, IORING_ENTER_GETEVENTS); |
52885fa2 | 689 | if (ret > 0) { |
bffad86f | 690 | fio_ioring_queued(td, start, ret); |
52885fa2 JA |
691 | io_u_mark_submit(td, ret); |
692 | ||
693 | ld->queued -= ret; | |
694 | ret = 0; | |
a90cd050 JA |
695 | } else if (!ret) { |
696 | io_u_mark_submit(td, ret); | |
52885fa2 | 697 | continue; |
a90cd050 | 698 | } else { |
f6abd731 | 699 | if (errno == EAGAIN || errno == EINTR) { |
bffad86f | 700 | ret = fio_ioring_cqring_reap(td, 0, ld->queued); |
a90cd050 JA |
701 | if (ret) |
702 | continue; | |
703 | /* Shouldn't happen */ | |
704 | usleep(1); | |
705 | continue; | |
52885fa2 | 706 | } |
1816895b | 707 | ret = -errno; |
9a2d78b3 | 708 | td_verror(td, errno, "io_uring_enter submit"); |
52885fa2 | 709 | break; |
a90cd050 | 710 | } |
52885fa2 JA |
711 | } while (ld->queued); |
712 | ||
713 | return ret; | |
714 | } | |
715 | ||
bffad86f | 716 | static void fio_ioring_unmap(struct ioring_data *ld) |
52885fa2 | 717 | { |
9a2d78b3 | 718 | int i; |
52885fa2 | 719 | |
59f94d26 | 720 | for (i = 0; i < FIO_ARRAY_SIZE(ld->mmap); i++) |
9a2d78b3 JA |
721 | munmap(ld->mmap[i].ptr, ld->mmap[i].len); |
722 | close(ld->ring_fd); | |
b87aa01a JA |
723 | } |
724 | ||
bffad86f | 725 | static void fio_ioring_cleanup(struct thread_data *td) |
52885fa2 | 726 | { |
bffad86f | 727 | struct ioring_data *ld = td->io_ops_data; |
52885fa2 JA |
728 | |
729 | if (ld) { | |
52885fa2 | 730 | if (!(td->flags & TD_F_CHILD)) |
bffad86f | 731 | fio_ioring_unmap(ld); |
9a2d78b3 | 732 | |
d6cbeab4 | 733 | fio_cmdprio_cleanup(&ld->cmdprio); |
52885fa2 | 734 | free(ld->io_u_index); |
2d6451c9 | 735 | free(ld->md_buf); |
9a2d78b3 | 736 | free(ld->iovecs); |
5ffd5626 | 737 | free(ld->fds); |
4885a6eb | 738 | free(ld->dsm); |
52885fa2 JA |
739 | free(ld); |
740 | } | |
741 | } | |
742 | ||
bffad86f | 743 | static int fio_ioring_mmap(struct ioring_data *ld, struct io_uring_params *p) |
9a2d78b3 | 744 | { |
bffad86f JA |
745 | struct io_sq_ring *sring = &ld->sq_ring; |
746 | struct io_cq_ring *cring = &ld->cq_ring; | |
9a2d78b3 JA |
747 | void *ptr; |
748 | ||
e2239016 | 749 | ld->mmap[0].len = p->sq_off.array + p->sq_entries * sizeof(__u32); |
9a2d78b3 JA |
750 | ptr = mmap(0, ld->mmap[0].len, PROT_READ | PROT_WRITE, |
751 | MAP_SHARED | MAP_POPULATE, ld->ring_fd, | |
752 | IORING_OFF_SQ_RING); | |
753 | ld->mmap[0].ptr = ptr; | |
754 | sring->head = ptr + p->sq_off.head; | |
755 | sring->tail = ptr + p->sq_off.tail; | |
756 | sring->ring_mask = ptr + p->sq_off.ring_mask; | |
757 | sring->ring_entries = ptr + p->sq_off.ring_entries; | |
758 | sring->flags = ptr + p->sq_off.flags; | |
ac122fea | 759 | sring->array = ptr + p->sq_off.array; |
9a2d78b3 JA |
760 | ld->sq_ring_mask = *sring->ring_mask; |
761 | ||
855dc4d4 AG |
762 | if (p->flags & IORING_SETUP_SQE128) |
763 | ld->mmap[1].len = 2 * p->sq_entries * sizeof(struct io_uring_sqe); | |
764 | else | |
765 | ld->mmap[1].len = p->sq_entries * sizeof(struct io_uring_sqe); | |
f0403f94 | 766 | ld->sqes = mmap(0, ld->mmap[1].len, PROT_READ | PROT_WRITE, |
9a2d78b3 | 767 | MAP_SHARED | MAP_POPULATE, ld->ring_fd, |
f0403f94 JA |
768 | IORING_OFF_SQES); |
769 | ld->mmap[1].ptr = ld->sqes; | |
9a2d78b3 | 770 | |
855dc4d4 AG |
771 | if (p->flags & IORING_SETUP_CQE32) { |
772 | ld->mmap[2].len = p->cq_off.cqes + | |
773 | 2 * p->cq_entries * sizeof(struct io_uring_cqe); | |
774 | } else { | |
775 | ld->mmap[2].len = p->cq_off.cqes + | |
776 | p->cq_entries * sizeof(struct io_uring_cqe); | |
777 | } | |
9a2d78b3 JA |
778 | ptr = mmap(0, ld->mmap[2].len, PROT_READ | PROT_WRITE, |
779 | MAP_SHARED | MAP_POPULATE, ld->ring_fd, | |
780 | IORING_OFF_CQ_RING); | |
781 | ld->mmap[2].ptr = ptr; | |
782 | cring->head = ptr + p->cq_off.head; | |
783 | cring->tail = ptr + p->cq_off.tail; | |
784 | cring->ring_mask = ptr + p->cq_off.ring_mask; | |
785 | cring->ring_entries = ptr + p->cq_off.ring_entries; | |
f0403f94 | 786 | cring->cqes = ptr + p->cq_off.cqes; |
9a2d78b3 JA |
787 | ld->cq_ring_mask = *cring->ring_mask; |
788 | return 0; | |
789 | } | |
790 | ||
556d8415 JA |
791 | static void fio_ioring_probe(struct thread_data *td) |
792 | { | |
793 | struct ioring_data *ld = td->io_ops_data; | |
794 | struct ioring_options *o = td->eo; | |
795 | struct io_uring_probe *p; | |
796 | int ret; | |
797 | ||
798 | /* already set by user, don't touch */ | |
799 | if (o->nonvectored != -1) | |
800 | return; | |
801 | ||
802 | /* default to off, as that's always safe */ | |
803 | o->nonvectored = 0; | |
804 | ||
223decdd | 805 | p = calloc(1, sizeof(*p) + 256 * sizeof(struct io_uring_probe_op)); |
556d8415 JA |
806 | if (!p) |
807 | return; | |
808 | ||
556d8415 JA |
809 | ret = syscall(__NR_io_uring_register, ld->ring_fd, |
810 | IORING_REGISTER_PROBE, p, 256); | |
811 | if (ret < 0) | |
812 | goto out; | |
813 | ||
814 | if (IORING_OP_WRITE > p->ops_len) | |
815 | goto out; | |
816 | ||
817 | if ((p->ops[IORING_OP_READ].flags & IO_URING_OP_SUPPORTED) && | |
818 | (p->ops[IORING_OP_WRITE].flags & IO_URING_OP_SUPPORTED)) | |
819 | o->nonvectored = 1; | |
820 | out: | |
821 | free(p); | |
822 | } | |
823 | ||
bffad86f | 824 | static int fio_ioring_queue_init(struct thread_data *td) |
52885fa2 | 825 | { |
bffad86f JA |
826 | struct ioring_data *ld = td->io_ops_data; |
827 | struct ioring_options *o = td->eo; | |
52885fa2 | 828 | int depth = td->o.iodepth; |
bffad86f | 829 | struct io_uring_params p; |
9a2d78b3 JA |
830 | int ret; |
831 | ||
832 | memset(&p, 0, sizeof(p)); | |
52885fa2 JA |
833 | |
834 | if (o->hipri) | |
bffad86f | 835 | p.flags |= IORING_SETUP_IOPOLL; |
3d7d00a3 JA |
836 | if (o->sqpoll_thread) { |
837 | p.flags |= IORING_SETUP_SQPOLL; | |
838 | if (o->sqpoll_set) { | |
839 | p.flags |= IORING_SETUP_SQ_AFF; | |
840 | p.sq_thread_cpu = o->sqpoll_cpu; | |
841 | } | |
c011bf12 AK |
842 | |
843 | /* | |
844 | * Submission latency for sqpoll_thread is just the time it | |
845 | * takes to fill in the SQ ring entries, and any syscall if | |
846 | * IORING_SQ_NEED_WAKEUP is set, we don't need to log that time | |
847 | * separately. | |
848 | */ | |
849 | td->o.disable_slat = 1; | |
f635f1fb | 850 | } |
a90cd050 | 851 | |
1db268db JA |
852 | /* |
853 | * Clamp CQ ring size at our SQ ring size, we don't need more entries | |
854 | * than that. | |
855 | */ | |
856 | p.flags |= IORING_SETUP_CQSIZE; | |
857 | p.cq_entries = depth; | |
858 | ||
4d22c103 JA |
859 | /* |
860 | * Setup COOP_TASKRUN as we don't need to get IPI interrupted for | |
861 | * completing IO operations. | |
862 | */ | |
863 | p.flags |= IORING_SETUP_COOP_TASKRUN; | |
864 | ||
e453f369 JA |
865 | /* |
866 | * io_uring is always a single issuer, and we can defer task_work | |
867 | * runs until we reap events. | |
868 | */ | |
869 | p.flags |= IORING_SETUP_SINGLE_ISSUER | IORING_SETUP_DEFER_TASKRUN; | |
870 | ||
b5e99df6 | 871 | retry: |
bfed648c | 872 | ret = syscall(__NR_io_uring_setup, depth, &p); |
b5e99df6 | 873 | if (ret < 0) { |
e453f369 JA |
874 | if (errno == EINVAL && p.flags & IORING_SETUP_DEFER_TASKRUN) { |
875 | p.flags &= ~IORING_SETUP_DEFER_TASKRUN; | |
876 | p.flags &= ~IORING_SETUP_SINGLE_ISSUER; | |
877 | goto retry; | |
878 | } | |
4d22c103 JA |
879 | if (errno == EINVAL && p.flags & IORING_SETUP_COOP_TASKRUN) { |
880 | p.flags &= ~IORING_SETUP_COOP_TASKRUN; | |
881 | goto retry; | |
882 | } | |
b5e99df6 JA |
883 | if (errno == EINVAL && p.flags & IORING_SETUP_CQSIZE) { |
884 | p.flags &= ~IORING_SETUP_CQSIZE; | |
885 | goto retry; | |
886 | } | |
9a2d78b3 | 887 | return ret; |
b5e99df6 | 888 | } |
9a2d78b3 JA |
889 | |
890 | ld->ring_fd = ret; | |
2ea53ca3 | 891 | |
556d8415 JA |
892 | fio_ioring_probe(td); |
893 | ||
2ea53ca3 | 894 | if (o->fixedbufs) { |
bfed648c | 895 | ret = syscall(__NR_io_uring_register, ld->ring_fd, |
919850d2 | 896 | IORING_REGISTER_BUFFERS, ld->iovecs, depth); |
2ea53ca3 JA |
897 | if (ret < 0) |
898 | return ret; | |
899 | } | |
900 | ||
bffad86f | 901 | return fio_ioring_mmap(ld, &p); |
52885fa2 JA |
902 | } |
903 | ||
855dc4d4 AG |
904 | static int fio_ioring_cmd_queue_init(struct thread_data *td) |
905 | { | |
906 | struct ioring_data *ld = td->io_ops_data; | |
907 | struct ioring_options *o = td->eo; | |
908 | int depth = td->o.iodepth; | |
909 | struct io_uring_params p; | |
910 | int ret; | |
911 | ||
912 | memset(&p, 0, sizeof(p)); | |
913 | ||
914 | if (o->hipri) | |
915 | p.flags |= IORING_SETUP_IOPOLL; | |
916 | if (o->sqpoll_thread) { | |
917 | p.flags |= IORING_SETUP_SQPOLL; | |
918 | if (o->sqpoll_set) { | |
919 | p.flags |= IORING_SETUP_SQ_AFF; | |
920 | p.sq_thread_cpu = o->sqpoll_cpu; | |
921 | } | |
c011bf12 AK |
922 | |
923 | /* | |
924 | * Submission latency for sqpoll_thread is just the time it | |
925 | * takes to fill in the SQ ring entries, and any syscall if | |
926 | * IORING_SQ_NEED_WAKEUP is set, we don't need to log that time | |
927 | * separately. | |
928 | */ | |
929 | td->o.disable_slat = 1; | |
855dc4d4 AG |
930 | } |
931 | if (o->cmd_type == FIO_URING_CMD_NVME) { | |
932 | p.flags |= IORING_SETUP_SQE128; | |
933 | p.flags |= IORING_SETUP_CQE32; | |
934 | } | |
935 | ||
936 | /* | |
937 | * Clamp CQ ring size at our SQ ring size, we don't need more entries | |
938 | * than that. | |
939 | */ | |
940 | p.flags |= IORING_SETUP_CQSIZE; | |
941 | p.cq_entries = depth; | |
942 | ||
07f78c37 AK |
943 | /* |
944 | * Setup COOP_TASKRUN as we don't need to get IPI interrupted for | |
945 | * completing IO operations. | |
946 | */ | |
947 | p.flags |= IORING_SETUP_COOP_TASKRUN; | |
948 | ||
949 | /* | |
950 | * io_uring is always a single issuer, and we can defer task_work | |
951 | * runs until we reap events. | |
952 | */ | |
953 | p.flags |= IORING_SETUP_SINGLE_ISSUER | IORING_SETUP_DEFER_TASKRUN; | |
954 | ||
855dc4d4 AG |
955 | retry: |
956 | ret = syscall(__NR_io_uring_setup, depth, &p); | |
957 | if (ret < 0) { | |
07f78c37 AK |
958 | if (errno == EINVAL && p.flags & IORING_SETUP_DEFER_TASKRUN) { |
959 | p.flags &= ~IORING_SETUP_DEFER_TASKRUN; | |
960 | p.flags &= ~IORING_SETUP_SINGLE_ISSUER; | |
961 | goto retry; | |
962 | } | |
963 | if (errno == EINVAL && p.flags & IORING_SETUP_COOP_TASKRUN) { | |
964 | p.flags &= ~IORING_SETUP_COOP_TASKRUN; | |
965 | goto retry; | |
966 | } | |
855dc4d4 AG |
967 | if (errno == EINVAL && p.flags & IORING_SETUP_CQSIZE) { |
968 | p.flags &= ~IORING_SETUP_CQSIZE; | |
969 | goto retry; | |
970 | } | |
971 | return ret; | |
972 | } | |
973 | ||
974 | ld->ring_fd = ret; | |
975 | ||
976 | fio_ioring_probe(td); | |
977 | ||
978 | if (o->fixedbufs) { | |
979 | ret = syscall(__NR_io_uring_register, ld->ring_fd, | |
980 | IORING_REGISTER_BUFFERS, ld->iovecs, depth); | |
981 | if (ret < 0) | |
982 | return ret; | |
983 | } | |
984 | ||
985 | return fio_ioring_mmap(ld, &p); | |
986 | } | |
987 | ||
5ffd5626 JA |
988 | static int fio_ioring_register_files(struct thread_data *td) |
989 | { | |
990 | struct ioring_data *ld = td->io_ops_data; | |
991 | struct fio_file *f; | |
992 | unsigned int i; | |
993 | int ret; | |
994 | ||
995 | ld->fds = calloc(td->o.nr_files, sizeof(int)); | |
996 | ||
997 | for_each_file(td, f, i) { | |
998 | ret = generic_open_file(td, f); | |
999 | if (ret) | |
1000 | goto err; | |
1001 | ld->fds[i] = f->fd; | |
1002 | f->engine_pos = i; | |
1003 | } | |
1004 | ||
bfed648c | 1005 | ret = syscall(__NR_io_uring_register, ld->ring_fd, |
5ffd5626 JA |
1006 | IORING_REGISTER_FILES, ld->fds, td->o.nr_files); |
1007 | if (ret) { | |
1008 | err: | |
1009 | free(ld->fds); | |
1010 | ld->fds = NULL; | |
1011 | } | |
1012 | ||
1013 | /* | |
1014 | * Pretend the file is closed again, and really close it if we hit | |
1015 | * an error. | |
1016 | */ | |
1017 | for_each_file(td, f, i) { | |
1018 | if (ret) { | |
1019 | int fio_unused ret2; | |
1020 | ret2 = generic_close_file(td, f); | |
1021 | } else | |
1022 | f->fd = -1; | |
1023 | } | |
1024 | ||
1025 | return ret; | |
1026 | } | |
1027 | ||
bffad86f | 1028 | static int fio_ioring_post_init(struct thread_data *td) |
52885fa2 | 1029 | { |
bffad86f | 1030 | struct ioring_data *ld = td->io_ops_data; |
5ffd5626 | 1031 | struct ioring_options *o = td->eo; |
52885fa2 | 1032 | struct io_u *io_u; |
650346e1 | 1033 | int err, i; |
52885fa2 | 1034 | |
650346e1 JA |
1035 | for (i = 0; i < td->o.iodepth; i++) { |
1036 | struct iovec *iov = &ld->iovecs[i]; | |
9a2d78b3 | 1037 | |
650346e1 JA |
1038 | io_u = ld->io_u_index[i]; |
1039 | iov->iov_base = io_u->buf; | |
1040 | iov->iov_len = td_max_bs(td); | |
52885fa2 JA |
1041 | } |
1042 | ||
bffad86f | 1043 | err = fio_ioring_queue_init(td); |
52885fa2 | 1044 | if (err) { |
0442b53f | 1045 | int init_err = errno; |
c4f5c92f | 1046 | |
0442b53f | 1047 | if (init_err == ENOSYS) |
c4f5c92f | 1048 | log_err("fio: your kernel doesn't support io_uring\n"); |
0442b53f | 1049 | td_verror(td, init_err, "io_queue_init"); |
52885fa2 JA |
1050 | return 1; |
1051 | } | |
1052 | ||
7c70f506 JA |
1053 | for (i = 0; i < td->o.iodepth; i++) { |
1054 | struct io_uring_sqe *sqe; | |
1055 | ||
1056 | sqe = &ld->sqes[i]; | |
1057 | memset(sqe, 0, sizeof(*sqe)); | |
1058 | } | |
1059 | ||
5ffd5626 JA |
1060 | if (o->registerfiles) { |
1061 | err = fio_ioring_register_files(td); | |
1062 | if (err) { | |
1063 | td_verror(td, errno, "ioring_register_files"); | |
1064 | return 1; | |
1065 | } | |
1066 | } | |
1067 | ||
52885fa2 JA |
1068 | return 0; |
1069 | } | |
1070 | ||
855dc4d4 AG |
1071 | static int fio_ioring_cmd_post_init(struct thread_data *td) |
1072 | { | |
1073 | struct ioring_data *ld = td->io_ops_data; | |
1074 | struct ioring_options *o = td->eo; | |
1075 | struct io_u *io_u; | |
1076 | int err, i; | |
1077 | ||
1078 | for (i = 0; i < td->o.iodepth; i++) { | |
1079 | struct iovec *iov = &ld->iovecs[i]; | |
1080 | ||
1081 | io_u = ld->io_u_index[i]; | |
1082 | iov->iov_base = io_u->buf; | |
1083 | iov->iov_len = td_max_bs(td); | |
1084 | } | |
1085 | ||
1086 | err = fio_ioring_cmd_queue_init(td); | |
1087 | if (err) { | |
1088 | int init_err = errno; | |
1089 | ||
1090 | td_verror(td, init_err, "io_queue_init"); | |
1091 | return 1; | |
1092 | } | |
1093 | ||
1094 | for (i = 0; i < td->o.iodepth; i++) { | |
1095 | struct io_uring_sqe *sqe; | |
1096 | ||
1097 | if (o->cmd_type == FIO_URING_CMD_NVME) { | |
1098 | sqe = &ld->sqes[i << 1]; | |
1099 | memset(sqe, 0, 2 * sizeof(*sqe)); | |
1100 | } else { | |
1101 | sqe = &ld->sqes[i]; | |
1102 | memset(sqe, 0, sizeof(*sqe)); | |
1103 | } | |
1104 | } | |
1105 | ||
1106 | if (o->registerfiles) { | |
1107 | err = fio_ioring_register_files(td); | |
1108 | if (err) { | |
1109 | td_verror(td, errno, "ioring_register_files"); | |
1110 | return 1; | |
1111 | } | |
1112 | } | |
1113 | ||
1114 | return 0; | |
1115 | } | |
1116 | ||
3ee8311a AK |
1117 | static void parse_prchk_flags(struct ioring_options *o) |
1118 | { | |
1119 | if (!o->pi_chk) | |
1120 | return; | |
1121 | ||
1122 | if (strstr(o->pi_chk, "GUARD") != NULL) | |
1123 | o->prchk = NVME_IO_PRINFO_PRCHK_GUARD; | |
1124 | if (strstr(o->pi_chk, "REFTAG") != NULL) | |
1125 | o->prchk |= NVME_IO_PRINFO_PRCHK_REF; | |
1126 | if (strstr(o->pi_chk, "APPTAG") != NULL) | |
1127 | o->prchk |= NVME_IO_PRINFO_PRCHK_APP; | |
1128 | } | |
1129 | ||
bffad86f | 1130 | static int fio_ioring_init(struct thread_data *td) |
52885fa2 | 1131 | { |
5ffd5626 | 1132 | struct ioring_options *o = td->eo; |
bffad86f | 1133 | struct ioring_data *ld; |
2d6451c9 | 1134 | unsigned long long md_size; |
e9f6567a | 1135 | int ret; |
52885fa2 | 1136 | |
5ffd5626 JA |
1137 | /* sqthread submission requires registered files */ |
1138 | if (o->sqpoll_thread) | |
1139 | o->registerfiles = 1; | |
1140 | ||
1141 | if (o->registerfiles && td->o.nr_files != td->o.open_files) { | |
1142 | log_err("fio: io_uring registered files require nr_files to " | |
1143 | "be identical to open_files\n"); | |
1144 | return 1; | |
1145 | } | |
1146 | ||
52885fa2 JA |
1147 | ld = calloc(1, sizeof(*ld)); |
1148 | ||
b87aa01a JA |
1149 | /* ring depth must be a power-of-2 */ |
1150 | ld->iodepth = td->o.iodepth; | |
1151 | td->o.iodepth = roundup_pow2(td->o.iodepth); | |
1152 | ||
52885fa2 JA |
1153 | /* io_u index */ |
1154 | ld->io_u_index = calloc(td->o.iodepth, sizeof(struct io_u *)); | |
2d6451c9 AK |
1155 | |
1156 | /* | |
1157 | * metadata buffer for nvme command. | |
1158 | * We are only supporting iomem=malloc / mem=malloc as of now. | |
1159 | */ | |
1160 | if (!strcmp(td->io_ops->name, "io_uring_cmd") && | |
1161 | (o->cmd_type == FIO_URING_CMD_NVME) && o->md_per_io_size) { | |
1162 | md_size = (unsigned long long) o->md_per_io_size | |
1163 | * (unsigned long long) td->o.iodepth; | |
1164 | md_size += page_mask + td->o.mem_align; | |
1165 | if (td->o.mem_align && td->o.mem_align > page_size) | |
1166 | md_size += td->o.mem_align - page_size; | |
1167 | if (td->o.mem_type == MEM_MALLOC) { | |
1168 | ld->md_buf = malloc(md_size); | |
6795954b JA |
1169 | if (!ld->md_buf) { |
1170 | free(ld); | |
2d6451c9 | 1171 | return 1; |
6795954b | 1172 | } |
2d6451c9 AK |
1173 | } else { |
1174 | log_err("fio: Only iomem=malloc or mem=malloc is supported\n"); | |
6795954b | 1175 | free(ld); |
2d6451c9 AK |
1176 | return 1; |
1177 | } | |
1178 | } | |
3ee8311a | 1179 | parse_prchk_flags(o); |
2d6451c9 | 1180 | |
650346e1 | 1181 | ld->iovecs = calloc(td->o.iodepth, sizeof(struct iovec)); |
52885fa2 JA |
1182 | |
1183 | td->io_ops_data = ld; | |
b2a432bf | 1184 | |
d6cbeab4 | 1185 | ret = fio_cmdprio_init(td, &ld->cmdprio, &o->cmdprio_options); |
e9f6567a DLM |
1186 | if (ret) { |
1187 | td_verror(td, EINVAL, "fio_ioring_init"); | |
b2a432bf PC |
1188 | return 1; |
1189 | } | |
1af44196 | 1190 | |
4885a6eb VF |
1191 | /* |
1192 | * For io_uring_cmd, trims are async operations unless we are operating | |
1193 | * in zbd mode where trim means zone reset. | |
1194 | */ | |
1195 | if (!strcmp(td->io_ops->name, "io_uring_cmd") && td_trim(td) && | |
1196 | td->o.zone_mode == ZONE_MODE_ZBD) | |
1197 | td->io_ops->flags |= FIO_ASYNCIO_SYNC_TRIM; | |
1198 | else | |
1199 | ld->dsm = calloc(ld->iodepth, sizeof(*ld->dsm)); | |
1200 | ||
52885fa2 JA |
1201 | return 0; |
1202 | } | |
1203 | ||
bffad86f | 1204 | static int fio_ioring_io_u_init(struct thread_data *td, struct io_u *io_u) |
52885fa2 | 1205 | { |
bffad86f | 1206 | struct ioring_data *ld = td->io_ops_data; |
2d6451c9 | 1207 | struct ioring_options *o = td->eo; |
5163f35e | 1208 | struct nvme_pi_data *pi_data; |
2d6451c9 | 1209 | char *p; |
52885fa2 JA |
1210 | |
1211 | ld->io_u_index[io_u->index] = io_u; | |
2d6451c9 AK |
1212 | |
1213 | if (!strcmp(td->io_ops->name, "io_uring_cmd")) { | |
1214 | p = PTR_ALIGN(ld->md_buf, page_mask) + td->o.mem_align; | |
1215 | p += o->md_per_io_size * io_u->index; | |
1216 | io_u->mmap_data = p; | |
5163f35e AK |
1217 | |
1218 | if (!o->pi_act) { | |
1219 | pi_data = calloc(1, sizeof(*pi_data)); | |
1220 | pi_data->io_flags |= o->prchk; | |
1221 | pi_data->apptag_mask = o->apptag_mask; | |
1222 | pi_data->apptag = o->apptag; | |
1223 | io_u->engine_data = pi_data; | |
1224 | } | |
2d6451c9 AK |
1225 | } |
1226 | ||
52885fa2 JA |
1227 | return 0; |
1228 | } | |
1229 | ||
5163f35e AK |
1230 | static void fio_ioring_io_u_free(struct thread_data *td, struct io_u *io_u) |
1231 | { | |
1232 | struct ioring_options *o = td->eo; | |
1233 | struct nvme_pi *pi; | |
1234 | ||
1235 | if (!strcmp(td->io_ops->name, "io_uring_cmd") && | |
1236 | (o->cmd_type == FIO_URING_CMD_NVME)) { | |
1237 | pi = io_u->engine_data; | |
1238 | free(pi); | |
1239 | io_u->engine_data = NULL; | |
1240 | } | |
1241 | } | |
1242 | ||
5ffd5626 JA |
1243 | static int fio_ioring_open_file(struct thread_data *td, struct fio_file *f) |
1244 | { | |
1245 | struct ioring_data *ld = td->io_ops_data; | |
1246 | struct ioring_options *o = td->eo; | |
1247 | ||
17318cf6 | 1248 | if (!ld || !o->registerfiles) |
5ffd5626 JA |
1249 | return generic_open_file(td, f); |
1250 | ||
1251 | f->fd = ld->fds[f->engine_pos]; | |
1252 | return 0; | |
1253 | } | |
1254 | ||
855dc4d4 AG |
1255 | static int fio_ioring_cmd_open_file(struct thread_data *td, struct fio_file *f) |
1256 | { | |
1257 | struct ioring_data *ld = td->io_ops_data; | |
1258 | struct ioring_options *o = td->eo; | |
1259 | ||
1260 | if (o->cmd_type == FIO_URING_CMD_NVME) { | |
1261 | struct nvme_data *data = NULL; | |
e7e5023b | 1262 | unsigned int lba_size = 0; |
671aa9f5 | 1263 | __u64 nlba = 0; |
855dc4d4 AG |
1264 | int ret; |
1265 | ||
1266 | /* Store the namespace-id and lba size. */ | |
1267 | data = FILE_ENG_DATA(f); | |
1268 | if (data == NULL) { | |
855dc4d4 | 1269 | data = calloc(1, sizeof(struct nvme_data)); |
3ee8311a | 1270 | ret = fio_nvme_get_info(f, &nlba, o->pi_act, data); |
e7e5023b AK |
1271 | if (ret) { |
1272 | free(data); | |
1273 | return ret; | |
1274 | } | |
855dc4d4 AG |
1275 | |
1276 | FILE_SET_ENG_DATA(f, data); | |
1277 | } | |
345fa8fd | 1278 | |
e7e5023b | 1279 | lba_size = data->lba_ext ? data->lba_ext : data->lba_size; |
345fa8fd AK |
1280 | |
1281 | for_each_rw_ddir(ddir) { | |
1282 | if (td->o.min_bs[ddir] % lba_size || | |
1283 | td->o.max_bs[ddir] % lba_size) { | |
1284 | if (data->lba_ext) | |
a904d182 AK |
1285 | log_err("%s: block size must be a multiple of (LBA data size + Metadata size)\n", |
1286 | f->file_name); | |
345fa8fd | 1287 | else |
a904d182 AK |
1288 | log_err("%s: block size must be a multiple of LBA data size\n", |
1289 | f->file_name); | |
1290 | td_verror(td, EINVAL, "fio_ioring_cmd_open_file"); | |
345fa8fd AK |
1291 | return 1; |
1292 | } | |
2d6451c9 AK |
1293 | if (data->ms && !data->lba_ext && ddir != DDIR_TRIM && |
1294 | (o->md_per_io_size < ((td->o.max_bs[ddir] / data->lba_size) * | |
1295 | data->ms))) { | |
1296 | log_err("%s: md_per_io_size should be at least %llu bytes\n", | |
1297 | f->file_name, | |
1298 | ((td->o.max_bs[ddir] / data->lba_size) * data->ms)); | |
1299 | td_verror(td, EINVAL, "fio_ioring_cmd_open_file"); | |
1300 | return 1; | |
1301 | } | |
345fa8fd | 1302 | } |
ba342e58 AK |
1303 | |
1304 | /* | |
1305 | * For extended logical block sizes we cannot use verify when | |
1306 | * end to end data protection checks are enabled, as the PI | |
1307 | * section of data buffer conflicts with verify. | |
1308 | */ | |
1309 | if (data->ms && data->pi_type && data->lba_ext && | |
1310 | td->o.verify != VERIFY_NONE) { | |
1311 | log_err("%s: for extended LBA, verify cannot be used when E2E data protection is enabled\n", | |
1312 | f->file_name); | |
1313 | td_verror(td, EINVAL, "fio_ioring_cmd_open_file"); | |
1314 | return 1; | |
1315 | } | |
855dc4d4 AG |
1316 | } |
1317 | if (!ld || !o->registerfiles) | |
1318 | return generic_open_file(td, f); | |
1319 | ||
1320 | f->fd = ld->fds[f->engine_pos]; | |
1321 | return 0; | |
1322 | } | |
1323 | ||
5ffd5626 JA |
1324 | static int fio_ioring_close_file(struct thread_data *td, struct fio_file *f) |
1325 | { | |
17318cf6 | 1326 | struct ioring_data *ld = td->io_ops_data; |
5ffd5626 JA |
1327 | struct ioring_options *o = td->eo; |
1328 | ||
17318cf6 | 1329 | if (!ld || !o->registerfiles) |
5ffd5626 JA |
1330 | return generic_close_file(td, f); |
1331 | ||
1332 | f->fd = -1; | |
1333 | return 0; | |
1334 | } | |
1335 | ||
855dc4d4 AG |
1336 | static int fio_ioring_cmd_close_file(struct thread_data *td, |
1337 | struct fio_file *f) | |
1338 | { | |
1339 | struct ioring_data *ld = td->io_ops_data; | |
1340 | struct ioring_options *o = td->eo; | |
1341 | ||
1342 | if (o->cmd_type == FIO_URING_CMD_NVME) { | |
1343 | struct nvme_data *data = FILE_ENG_DATA(f); | |
1344 | ||
1345 | FILE_SET_ENG_DATA(f, NULL); | |
1346 | free(data); | |
1347 | } | |
1348 | if (!ld || !o->registerfiles) | |
1349 | return generic_close_file(td, f); | |
1350 | ||
1351 | f->fd = -1; | |
1352 | return 0; | |
1353 | } | |
1354 | ||
1355 | static int fio_ioring_cmd_get_file_size(struct thread_data *td, | |
1356 | struct fio_file *f) | |
1357 | { | |
1358 | struct ioring_options *o = td->eo; | |
1359 | ||
1360 | if (fio_file_size_known(f)) | |
1361 | return 0; | |
1362 | ||
1363 | if (o->cmd_type == FIO_URING_CMD_NVME) { | |
1364 | struct nvme_data *data = NULL; | |
671aa9f5 | 1365 | __u64 nlba = 0; |
855dc4d4 AG |
1366 | int ret; |
1367 | ||
855dc4d4 | 1368 | data = calloc(1, sizeof(struct nvme_data)); |
3ee8311a | 1369 | ret = fio_nvme_get_info(f, &nlba, o->pi_act, data); |
e7e5023b AK |
1370 | if (ret) { |
1371 | free(data); | |
1372 | return ret; | |
1373 | } | |
855dc4d4 | 1374 | |
e7e5023b | 1375 | f->real_file_size = data->lba_size * nlba; |
855dc4d4 AG |
1376 | fio_file_set_size_known(f); |
1377 | ||
1378 | FILE_SET_ENG_DATA(f, data); | |
1379 | return 0; | |
1380 | } | |
1381 | return generic_get_file_size(td, f); | |
1382 | } | |
1383 | ||
3d05e0ff AK |
1384 | static int fio_ioring_cmd_get_zoned_model(struct thread_data *td, |
1385 | struct fio_file *f, | |
1386 | enum zbd_zoned_model *model) | |
1387 | { | |
1388 | return fio_nvme_get_zoned_model(td, f, model); | |
1389 | } | |
1390 | ||
1391 | static int fio_ioring_cmd_report_zones(struct thread_data *td, | |
1392 | struct fio_file *f, uint64_t offset, | |
1393 | struct zbd_zone *zbdz, | |
1394 | unsigned int nr_zones) | |
1395 | { | |
1396 | return fio_nvme_report_zones(td, f, offset, zbdz, nr_zones); | |
1397 | } | |
1398 | ||
1399 | static int fio_ioring_cmd_reset_wp(struct thread_data *td, struct fio_file *f, | |
1400 | uint64_t offset, uint64_t length) | |
1401 | { | |
1402 | return fio_nvme_reset_wp(td, f, offset, length); | |
1403 | } | |
1404 | ||
1405 | static int fio_ioring_cmd_get_max_open_zones(struct thread_data *td, | |
1406 | struct fio_file *f, | |
1407 | unsigned int *max_open_zones) | |
1408 | { | |
1409 | return fio_nvme_get_max_open_zones(td, f, max_open_zones); | |
1410 | } | |
1411 | ||
a7e8aae0 KB |
1412 | static int fio_ioring_cmd_fetch_ruhs(struct thread_data *td, struct fio_file *f, |
1413 | struct fio_ruhs_info *fruhs_info) | |
1414 | { | |
1415 | struct nvme_fdp_ruh_status *ruhs; | |
1416 | int bytes, ret, i; | |
1417 | ||
98cd3c0e | 1418 | bytes = sizeof(*ruhs) + FDP_MAX_RUHS * sizeof(struct nvme_fdp_ruh_status_desc); |
a7e8aae0 KB |
1419 | ruhs = scalloc(1, bytes); |
1420 | if (!ruhs) | |
1421 | return -ENOMEM; | |
1422 | ||
1423 | ret = fio_nvme_iomgmt_ruhs(td, f, ruhs, bytes); | |
1424 | if (ret) | |
1425 | goto free; | |
1426 | ||
1427 | fruhs_info->nr_ruhs = le16_to_cpu(ruhs->nruhsd); | |
1428 | for (i = 0; i < fruhs_info->nr_ruhs; i++) | |
1429 | fruhs_info->plis[i] = le16_to_cpu(ruhs->ruhss[i].pid); | |
1430 | free: | |
1431 | sfree(ruhs); | |
1432 | return ret; | |
1433 | } | |
1434 | ||
855dc4d4 | 1435 | static struct ioengine_ops ioengine_uring = { |
bffad86f | 1436 | .name = "io_uring", |
52885fa2 | 1437 | .version = FIO_IOOPS_VERSION, |
4e7e7898 VF |
1438 | .flags = FIO_ASYNCIO_SYNC_TRIM | FIO_NO_OFFLOAD | |
1439 | FIO_ASYNCIO_SETS_ISSUE_TIME, | |
bffad86f JA |
1440 | .init = fio_ioring_init, |
1441 | .post_init = fio_ioring_post_init, | |
1442 | .io_u_init = fio_ioring_io_u_init, | |
1443 | .prep = fio_ioring_prep, | |
1444 | .queue = fio_ioring_queue, | |
1445 | .commit = fio_ioring_commit, | |
1446 | .getevents = fio_ioring_getevents, | |
1447 | .event = fio_ioring_event, | |
1448 | .cleanup = fio_ioring_cleanup, | |
5ffd5626 JA |
1449 | .open_file = fio_ioring_open_file, |
1450 | .close_file = fio_ioring_close_file, | |
52885fa2 JA |
1451 | .get_file_size = generic_get_file_size, |
1452 | .options = options, | |
bffad86f | 1453 | .option_struct_size = sizeof(struct ioring_options), |
52885fa2 JA |
1454 | }; |
1455 | ||
855dc4d4 AG |
1456 | static struct ioengine_ops ioengine_uring_cmd = { |
1457 | .name = "io_uring_cmd", | |
1458 | .version = FIO_IOOPS_VERSION, | |
4885a6eb | 1459 | .flags = FIO_NO_OFFLOAD | FIO_MEMALIGN | FIO_RAWIO | |
4e7e7898 | 1460 | FIO_ASYNCIO_SETS_ISSUE_TIME, |
855dc4d4 AG |
1461 | .init = fio_ioring_init, |
1462 | .post_init = fio_ioring_cmd_post_init, | |
1463 | .io_u_init = fio_ioring_io_u_init, | |
5163f35e | 1464 | .io_u_free = fio_ioring_io_u_free, |
855dc4d4 AG |
1465 | .prep = fio_ioring_cmd_prep, |
1466 | .queue = fio_ioring_queue, | |
1467 | .commit = fio_ioring_commit, | |
1468 | .getevents = fio_ioring_getevents, | |
1469 | .event = fio_ioring_cmd_event, | |
1470 | .cleanup = fio_ioring_cleanup, | |
1471 | .open_file = fio_ioring_cmd_open_file, | |
1472 | .close_file = fio_ioring_cmd_close_file, | |
1473 | .get_file_size = fio_ioring_cmd_get_file_size, | |
3d05e0ff AK |
1474 | .get_zoned_model = fio_ioring_cmd_get_zoned_model, |
1475 | .report_zones = fio_ioring_cmd_report_zones, | |
1476 | .reset_wp = fio_ioring_cmd_reset_wp, | |
1477 | .get_max_open_zones = fio_ioring_cmd_get_max_open_zones, | |
855dc4d4 AG |
1478 | .options = options, |
1479 | .option_struct_size = sizeof(struct ioring_options), | |
a7e8aae0 | 1480 | .fdp_fetch_ruhs = fio_ioring_cmd_fetch_ruhs, |
855dc4d4 AG |
1481 | }; |
1482 | ||
bffad86f | 1483 | static void fio_init fio_ioring_register(void) |
52885fa2 | 1484 | { |
855dc4d4 AG |
1485 | register_ioengine(&ioengine_uring); |
1486 | register_ioengine(&ioengine_uring_cmd); | |
52885fa2 JA |
1487 | } |
1488 | ||
bffad86f | 1489 | static void fio_exit fio_ioring_unregister(void) |
52885fa2 | 1490 | { |
855dc4d4 AG |
1491 | unregister_ioengine(&ioengine_uring); |
1492 | unregister_ioengine(&ioengine_uring_cmd); | |
52885fa2 | 1493 | } |
1f90e9bb | 1494 | #endif |