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 | ||
f53eaac0 | 471 | if (cqe->res != 0) { |
855dc4d4 | 472 | io_u->error = -cqe->res; |
f53eaac0 AK |
473 | return io_u; |
474 | } else { | |
855dc4d4 | 475 | io_u->error = 0; |
f53eaac0 | 476 | } |
855dc4d4 | 477 | |
5163f35e AK |
478 | if (o->cmd_type == FIO_URING_CMD_NVME) { |
479 | data = FILE_ENG_DATA(io_u->file); | |
480 | if (data->pi_type && (io_u->ddir == DDIR_READ) && !o->pi_act) { | |
481 | ret = fio_nvme_pi_verify(data, io_u); | |
482 | if (ret) | |
483 | io_u->error = ret; | |
484 | } | |
485 | } | |
486 | ||
855dc4d4 AG |
487 | return io_u; |
488 | } | |
489 | ||
bffad86f | 490 | static int fio_ioring_cqring_reap(struct thread_data *td, unsigned int events, |
52885fa2 JA |
491 | unsigned int max) |
492 | { | |
bffad86f JA |
493 | struct ioring_data *ld = td->io_ops_data; |
494 | struct io_cq_ring *ring = &ld->cq_ring; | |
e2239016 | 495 | unsigned head, reaped = 0; |
52885fa2 | 496 | |
9a2d78b3 | 497 | head = *ring->head; |
52885fa2 | 498 | do { |
9e26aff9 | 499 | if (head == atomic_load_acquire(ring->tail)) |
52885fa2 JA |
500 | break; |
501 | reaped++; | |
502 | head++; | |
52885fa2 JA |
503 | } while (reaped + events < max); |
504 | ||
76ce63dd AB |
505 | if (reaped) |
506 | atomic_store_release(ring->head, head); | |
507 | ||
52885fa2 JA |
508 | return reaped; |
509 | } | |
510 | ||
bffad86f JA |
511 | static int fio_ioring_getevents(struct thread_data *td, unsigned int min, |
512 | unsigned int max, const struct timespec *t) | |
52885fa2 | 513 | { |
bffad86f | 514 | struct ioring_data *ld = td->io_ops_data; |
52885fa2 | 515 | unsigned actual_min = td->o.iodepth_batch_complete_min == 0 ? 0 : min; |
bffad86f JA |
516 | struct ioring_options *o = td->eo; |
517 | struct io_cq_ring *ring = &ld->cq_ring; | |
b87aa01a JA |
518 | unsigned events = 0; |
519 | int r; | |
52885fa2 | 520 | |
9a2d78b3 | 521 | ld->cq_ring_off = *ring->head; |
52885fa2 | 522 | do { |
bffad86f | 523 | r = fio_ioring_cqring_reap(td, events, max); |
52885fa2 JA |
524 | if (r) { |
525 | events += r; | |
ae8646a1 | 526 | max -= r; |
f7cbbbf8 ST |
527 | if (actual_min != 0) |
528 | actual_min -= r; | |
52885fa2 JA |
529 | continue; |
530 | } | |
531 | ||
3d7d00a3 | 532 | if (!o->sqpoll_thread) { |
9a2d78b3 JA |
533 | r = io_uring_enter(ld, 0, actual_min, |
534 | IORING_ENTER_GETEVENTS); | |
771c9901 | 535 | if (r < 0) { |
f6abd731 | 536 | if (errno == EAGAIN || errno == EINTR) |
771c9901 | 537 | continue; |
1816895b | 538 | r = -errno; |
9a2d78b3 | 539 | td_verror(td, errno, "io_uring_enter"); |
771c9901 JA |
540 | break; |
541 | } | |
52885fa2 JA |
542 | } |
543 | } while (events < min); | |
544 | ||
545 | return r < 0 ? r : events; | |
546 | } | |
547 | ||
3ee8311a AK |
548 | static inline void fio_ioring_cmd_nvme_pi(struct thread_data *td, |
549 | struct io_u *io_u) | |
550 | { | |
551 | struct ioring_data *ld = td->io_ops_data; | |
552 | struct ioring_options *o = td->eo; | |
553 | struct nvme_uring_cmd *cmd; | |
554 | struct io_uring_sqe *sqe; | |
555 | struct nvme_cmd_ext_io_opts ext_opts = {0}; | |
556 | struct nvme_data *data = FILE_ENG_DATA(io_u->file); | |
557 | ||
558 | if (io_u->ddir == DDIR_TRIM) | |
559 | return; | |
560 | ||
561 | sqe = &ld->sqes[(io_u->index) << 1]; | |
562 | cmd = (struct nvme_uring_cmd *)sqe->cmd; | |
563 | ||
564 | if (data->pi_type) { | |
565 | if (o->pi_act) | |
566 | ext_opts.io_flags |= NVME_IO_PRINFO_PRACT; | |
567 | ext_opts.io_flags |= o->prchk; | |
568 | ext_opts.apptag = o->apptag; | |
569 | ext_opts.apptag_mask = o->apptag_mask; | |
570 | } | |
571 | ||
572 | fio_nvme_pi_fill(cmd, io_u, &ext_opts); | |
573 | } | |
574 | ||
127715b6 NC |
575 | static inline void fio_ioring_cmdprio_prep(struct thread_data *td, |
576 | struct io_u *io_u) | |
b2a432bf | 577 | { |
b2a432bf | 578 | struct ioring_data *ld = td->io_ops_data; |
d6cbeab4 | 579 | struct cmdprio *cmdprio = &ld->cmdprio; |
127715b6 NC |
580 | |
581 | if (fio_cmdprio_set_ioprio(td, cmdprio, io_u)) | |
582 | ld->sqes[io_u->index].ioprio = io_u->ioprio; | |
b2a432bf PC |
583 | } |
584 | ||
bffad86f JA |
585 | static enum fio_q_status fio_ioring_queue(struct thread_data *td, |
586 | struct io_u *io_u) | |
52885fa2 | 587 | { |
bffad86f | 588 | struct ioring_data *ld = td->io_ops_data; |
3ee8311a | 589 | struct ioring_options *o = td->eo; |
bffad86f | 590 | struct io_sq_ring *ring = &ld->sq_ring; |
52885fa2 JA |
591 | unsigned tail, next_tail; |
592 | ||
593 | fio_ro_check(td, io_u); | |
594 | ||
b87aa01a | 595 | if (ld->queued == ld->iodepth) |
52885fa2 JA |
596 | return FIO_Q_BUSY; |
597 | ||
4885a6eb | 598 | if (io_u->ddir == DDIR_TRIM && td->io_ops->flags & FIO_ASYNCIO_SYNC_TRIM) { |
52885fa2 JA |
599 | if (ld->queued) |
600 | return FIO_Q_BUSY; | |
601 | ||
7f57e30f | 602 | do_io_u_trim(td, io_u); |
16be6037 | 603 | |
52885fa2 JA |
604 | io_u_mark_submit(td, 1); |
605 | io_u_mark_complete(td, 1); | |
606 | return FIO_Q_COMPLETED; | |
607 | } | |
608 | ||
9a2d78b3 | 609 | tail = *ring->tail; |
52885fa2 | 610 | next_tail = tail + 1; |
5c15a911 | 611 | if (next_tail == atomic_load_relaxed(ring->head)) |
52885fa2 JA |
612 | return FIO_Q_BUSY; |
613 | ||
d6cbeab4 | 614 | if (ld->cmdprio.mode != CMDPRIO_MODE_NONE) |
ff00f247 NC |
615 | fio_ioring_cmdprio_prep(td, io_u); |
616 | ||
3ee8311a AK |
617 | if (!strcmp(td->io_ops->name, "io_uring_cmd") && |
618 | o->cmd_type == FIO_URING_CMD_NVME) | |
619 | fio_ioring_cmd_nvme_pi(td, io_u); | |
620 | ||
b87aa01a | 621 | ring->array[tail & ld->sq_ring_mask] = io_u->index; |
9e26aff9 | 622 | atomic_store_release(ring->tail, next_tail); |
52885fa2 JA |
623 | |
624 | ld->queued++; | |
625 | return FIO_Q_QUEUED; | |
626 | } | |
627 | ||
bffad86f | 628 | static void fio_ioring_queued(struct thread_data *td, int start, int nr) |
52885fa2 | 629 | { |
bffad86f | 630 | struct ioring_data *ld = td->io_ops_data; |
52885fa2 JA |
631 | struct timespec now; |
632 | ||
633 | if (!fio_fill_issue_time(td)) | |
634 | return; | |
635 | ||
636 | fio_gettime(&now, NULL); | |
637 | ||
638 | while (nr--) { | |
bffad86f | 639 | struct io_sq_ring *ring = &ld->sq_ring; |
9a2d78b3 | 640 | int index = ring->array[start & ld->sq_ring_mask]; |
f8289afc | 641 | struct io_u *io_u = ld->io_u_index[index]; |
52885fa2 JA |
642 | |
643 | memcpy(&io_u->issue_time, &now, sizeof(now)); | |
644 | io_u_queued(td, io_u); | |
645 | ||
646 | start++; | |
52885fa2 | 647 | } |
39f56400 VF |
648 | |
649 | /* | |
650 | * only used for iolog | |
651 | */ | |
652 | if (td->o.read_iolog_file) | |
653 | memcpy(&td->last_issue, &now, sizeof(now)); | |
52885fa2 JA |
654 | } |
655 | ||
bffad86f | 656 | static int fio_ioring_commit(struct thread_data *td) |
52885fa2 | 657 | { |
bffad86f JA |
658 | struct ioring_data *ld = td->io_ops_data; |
659 | struct ioring_options *o = td->eo; | |
52885fa2 JA |
660 | int ret; |
661 | ||
662 | if (!ld->queued) | |
663 | return 0; | |
664 | ||
3d7d00a3 JA |
665 | /* |
666 | * Kernel side does submission. just need to check if the ring is | |
667 | * flagged as needing a kick, if so, call io_uring_enter(). This | |
668 | * only happens if we've been idle too long. | |
669 | */ | |
670 | if (o->sqpoll_thread) { | |
bffad86f | 671 | struct io_sq_ring *ring = &ld->sq_ring; |
10bad6b9 | 672 | unsigned start = *ld->sq_ring.tail - ld->queued; |
2dd96cc4 | 673 | unsigned flags; |
4cdbc048 | 674 | |
5c15a911 | 675 | flags = atomic_load_relaxed(ring->flags); |
2dd96cc4 | 676 | if (flags & IORING_SQ_NEED_WAKEUP) |
b532dd6d JA |
677 | io_uring_enter(ld, ld->queued, 0, |
678 | IORING_ENTER_SQ_WAKEUP); | |
c011bf12 AK |
679 | fio_ioring_queued(td, start, ld->queued); |
680 | io_u_mark_submit(td, ld->queued); | |
681 | ||
771c9901 JA |
682 | ld->queued = 0; |
683 | return 0; | |
684 | } | |
685 | ||
52885fa2 | 686 | do { |
9a2d78b3 | 687 | unsigned start = *ld->sq_ring.head; |
52885fa2 JA |
688 | long nr = ld->queued; |
689 | ||
9a2d78b3 | 690 | ret = io_uring_enter(ld, nr, 0, IORING_ENTER_GETEVENTS); |
52885fa2 | 691 | if (ret > 0) { |
bffad86f | 692 | fio_ioring_queued(td, start, ret); |
52885fa2 JA |
693 | io_u_mark_submit(td, ret); |
694 | ||
695 | ld->queued -= ret; | |
696 | ret = 0; | |
a90cd050 JA |
697 | } else if (!ret) { |
698 | io_u_mark_submit(td, ret); | |
52885fa2 | 699 | continue; |
a90cd050 | 700 | } else { |
f6abd731 | 701 | if (errno == EAGAIN || errno == EINTR) { |
bffad86f | 702 | ret = fio_ioring_cqring_reap(td, 0, ld->queued); |
a90cd050 JA |
703 | if (ret) |
704 | continue; | |
705 | /* Shouldn't happen */ | |
706 | usleep(1); | |
707 | continue; | |
52885fa2 | 708 | } |
1816895b | 709 | ret = -errno; |
9a2d78b3 | 710 | td_verror(td, errno, "io_uring_enter submit"); |
52885fa2 | 711 | break; |
a90cd050 | 712 | } |
52885fa2 JA |
713 | } while (ld->queued); |
714 | ||
715 | return ret; | |
716 | } | |
717 | ||
bffad86f | 718 | static void fio_ioring_unmap(struct ioring_data *ld) |
52885fa2 | 719 | { |
9a2d78b3 | 720 | int i; |
52885fa2 | 721 | |
59f94d26 | 722 | for (i = 0; i < FIO_ARRAY_SIZE(ld->mmap); i++) |
9a2d78b3 JA |
723 | munmap(ld->mmap[i].ptr, ld->mmap[i].len); |
724 | close(ld->ring_fd); | |
b87aa01a JA |
725 | } |
726 | ||
bffad86f | 727 | static void fio_ioring_cleanup(struct thread_data *td) |
52885fa2 | 728 | { |
bffad86f | 729 | struct ioring_data *ld = td->io_ops_data; |
52885fa2 JA |
730 | |
731 | if (ld) { | |
52885fa2 | 732 | if (!(td->flags & TD_F_CHILD)) |
bffad86f | 733 | fio_ioring_unmap(ld); |
9a2d78b3 | 734 | |
d6cbeab4 | 735 | fio_cmdprio_cleanup(&ld->cmdprio); |
52885fa2 | 736 | free(ld->io_u_index); |
2d6451c9 | 737 | free(ld->md_buf); |
9a2d78b3 | 738 | free(ld->iovecs); |
5ffd5626 | 739 | free(ld->fds); |
4885a6eb | 740 | free(ld->dsm); |
52885fa2 JA |
741 | free(ld); |
742 | } | |
743 | } | |
744 | ||
bffad86f | 745 | static int fio_ioring_mmap(struct ioring_data *ld, struct io_uring_params *p) |
9a2d78b3 | 746 | { |
bffad86f JA |
747 | struct io_sq_ring *sring = &ld->sq_ring; |
748 | struct io_cq_ring *cring = &ld->cq_ring; | |
9a2d78b3 JA |
749 | void *ptr; |
750 | ||
e2239016 | 751 | ld->mmap[0].len = p->sq_off.array + p->sq_entries * sizeof(__u32); |
9a2d78b3 JA |
752 | ptr = mmap(0, ld->mmap[0].len, PROT_READ | PROT_WRITE, |
753 | MAP_SHARED | MAP_POPULATE, ld->ring_fd, | |
754 | IORING_OFF_SQ_RING); | |
755 | ld->mmap[0].ptr = ptr; | |
756 | sring->head = ptr + p->sq_off.head; | |
757 | sring->tail = ptr + p->sq_off.tail; | |
758 | sring->ring_mask = ptr + p->sq_off.ring_mask; | |
759 | sring->ring_entries = ptr + p->sq_off.ring_entries; | |
760 | sring->flags = ptr + p->sq_off.flags; | |
ac122fea | 761 | sring->array = ptr + p->sq_off.array; |
9a2d78b3 JA |
762 | ld->sq_ring_mask = *sring->ring_mask; |
763 | ||
855dc4d4 AG |
764 | if (p->flags & IORING_SETUP_SQE128) |
765 | ld->mmap[1].len = 2 * p->sq_entries * sizeof(struct io_uring_sqe); | |
766 | else | |
767 | ld->mmap[1].len = p->sq_entries * sizeof(struct io_uring_sqe); | |
f0403f94 | 768 | ld->sqes = mmap(0, ld->mmap[1].len, PROT_READ | PROT_WRITE, |
9a2d78b3 | 769 | MAP_SHARED | MAP_POPULATE, ld->ring_fd, |
f0403f94 JA |
770 | IORING_OFF_SQES); |
771 | ld->mmap[1].ptr = ld->sqes; | |
9a2d78b3 | 772 | |
855dc4d4 AG |
773 | if (p->flags & IORING_SETUP_CQE32) { |
774 | ld->mmap[2].len = p->cq_off.cqes + | |
775 | 2 * p->cq_entries * sizeof(struct io_uring_cqe); | |
776 | } else { | |
777 | ld->mmap[2].len = p->cq_off.cqes + | |
778 | p->cq_entries * sizeof(struct io_uring_cqe); | |
779 | } | |
9a2d78b3 JA |
780 | ptr = mmap(0, ld->mmap[2].len, PROT_READ | PROT_WRITE, |
781 | MAP_SHARED | MAP_POPULATE, ld->ring_fd, | |
782 | IORING_OFF_CQ_RING); | |
783 | ld->mmap[2].ptr = ptr; | |
784 | cring->head = ptr + p->cq_off.head; | |
785 | cring->tail = ptr + p->cq_off.tail; | |
786 | cring->ring_mask = ptr + p->cq_off.ring_mask; | |
787 | cring->ring_entries = ptr + p->cq_off.ring_entries; | |
f0403f94 | 788 | cring->cqes = ptr + p->cq_off.cqes; |
9a2d78b3 JA |
789 | ld->cq_ring_mask = *cring->ring_mask; |
790 | return 0; | |
791 | } | |
792 | ||
556d8415 JA |
793 | static void fio_ioring_probe(struct thread_data *td) |
794 | { | |
795 | struct ioring_data *ld = td->io_ops_data; | |
796 | struct ioring_options *o = td->eo; | |
797 | struct io_uring_probe *p; | |
798 | int ret; | |
799 | ||
800 | /* already set by user, don't touch */ | |
801 | if (o->nonvectored != -1) | |
802 | return; | |
803 | ||
804 | /* default to off, as that's always safe */ | |
805 | o->nonvectored = 0; | |
806 | ||
223decdd | 807 | p = calloc(1, sizeof(*p) + 256 * sizeof(struct io_uring_probe_op)); |
556d8415 JA |
808 | if (!p) |
809 | return; | |
810 | ||
556d8415 JA |
811 | ret = syscall(__NR_io_uring_register, ld->ring_fd, |
812 | IORING_REGISTER_PROBE, p, 256); | |
813 | if (ret < 0) | |
814 | goto out; | |
815 | ||
816 | if (IORING_OP_WRITE > p->ops_len) | |
817 | goto out; | |
818 | ||
819 | if ((p->ops[IORING_OP_READ].flags & IO_URING_OP_SUPPORTED) && | |
820 | (p->ops[IORING_OP_WRITE].flags & IO_URING_OP_SUPPORTED)) | |
821 | o->nonvectored = 1; | |
822 | out: | |
823 | free(p); | |
824 | } | |
825 | ||
bffad86f | 826 | static int fio_ioring_queue_init(struct thread_data *td) |
52885fa2 | 827 | { |
bffad86f JA |
828 | struct ioring_data *ld = td->io_ops_data; |
829 | struct ioring_options *o = td->eo; | |
52885fa2 | 830 | int depth = td->o.iodepth; |
bffad86f | 831 | struct io_uring_params p; |
9a2d78b3 JA |
832 | int ret; |
833 | ||
834 | memset(&p, 0, sizeof(p)); | |
52885fa2 JA |
835 | |
836 | if (o->hipri) | |
bffad86f | 837 | p.flags |= IORING_SETUP_IOPOLL; |
3d7d00a3 JA |
838 | if (o->sqpoll_thread) { |
839 | p.flags |= IORING_SETUP_SQPOLL; | |
840 | if (o->sqpoll_set) { | |
841 | p.flags |= IORING_SETUP_SQ_AFF; | |
842 | p.sq_thread_cpu = o->sqpoll_cpu; | |
843 | } | |
c011bf12 AK |
844 | |
845 | /* | |
846 | * Submission latency for sqpoll_thread is just the time it | |
847 | * takes to fill in the SQ ring entries, and any syscall if | |
848 | * IORING_SQ_NEED_WAKEUP is set, we don't need to log that time | |
849 | * separately. | |
850 | */ | |
851 | td->o.disable_slat = 1; | |
f635f1fb | 852 | } |
a90cd050 | 853 | |
1db268db JA |
854 | /* |
855 | * Clamp CQ ring size at our SQ ring size, we don't need more entries | |
856 | * than that. | |
857 | */ | |
858 | p.flags |= IORING_SETUP_CQSIZE; | |
859 | p.cq_entries = depth; | |
860 | ||
4d22c103 JA |
861 | /* |
862 | * Setup COOP_TASKRUN as we don't need to get IPI interrupted for | |
863 | * completing IO operations. | |
864 | */ | |
865 | p.flags |= IORING_SETUP_COOP_TASKRUN; | |
866 | ||
e453f369 JA |
867 | /* |
868 | * io_uring is always a single issuer, and we can defer task_work | |
869 | * runs until we reap events. | |
870 | */ | |
871 | p.flags |= IORING_SETUP_SINGLE_ISSUER | IORING_SETUP_DEFER_TASKRUN; | |
872 | ||
b5e99df6 | 873 | retry: |
bfed648c | 874 | ret = syscall(__NR_io_uring_setup, depth, &p); |
b5e99df6 | 875 | if (ret < 0) { |
e453f369 JA |
876 | if (errno == EINVAL && p.flags & IORING_SETUP_DEFER_TASKRUN) { |
877 | p.flags &= ~IORING_SETUP_DEFER_TASKRUN; | |
878 | p.flags &= ~IORING_SETUP_SINGLE_ISSUER; | |
879 | goto retry; | |
880 | } | |
4d22c103 JA |
881 | if (errno == EINVAL && p.flags & IORING_SETUP_COOP_TASKRUN) { |
882 | p.flags &= ~IORING_SETUP_COOP_TASKRUN; | |
883 | goto retry; | |
884 | } | |
b5e99df6 JA |
885 | if (errno == EINVAL && p.flags & IORING_SETUP_CQSIZE) { |
886 | p.flags &= ~IORING_SETUP_CQSIZE; | |
887 | goto retry; | |
888 | } | |
9a2d78b3 | 889 | return ret; |
b5e99df6 | 890 | } |
9a2d78b3 JA |
891 | |
892 | ld->ring_fd = ret; | |
2ea53ca3 | 893 | |
556d8415 JA |
894 | fio_ioring_probe(td); |
895 | ||
2ea53ca3 | 896 | if (o->fixedbufs) { |
bfed648c | 897 | ret = syscall(__NR_io_uring_register, ld->ring_fd, |
919850d2 | 898 | IORING_REGISTER_BUFFERS, ld->iovecs, depth); |
2ea53ca3 JA |
899 | if (ret < 0) |
900 | return ret; | |
901 | } | |
902 | ||
bffad86f | 903 | return fio_ioring_mmap(ld, &p); |
52885fa2 JA |
904 | } |
905 | ||
855dc4d4 AG |
906 | static int fio_ioring_cmd_queue_init(struct thread_data *td) |
907 | { | |
908 | struct ioring_data *ld = td->io_ops_data; | |
909 | struct ioring_options *o = td->eo; | |
910 | int depth = td->o.iodepth; | |
911 | struct io_uring_params p; | |
912 | int ret; | |
913 | ||
914 | memset(&p, 0, sizeof(p)); | |
915 | ||
916 | if (o->hipri) | |
917 | p.flags |= IORING_SETUP_IOPOLL; | |
918 | if (o->sqpoll_thread) { | |
919 | p.flags |= IORING_SETUP_SQPOLL; | |
920 | if (o->sqpoll_set) { | |
921 | p.flags |= IORING_SETUP_SQ_AFF; | |
922 | p.sq_thread_cpu = o->sqpoll_cpu; | |
923 | } | |
c011bf12 AK |
924 | |
925 | /* | |
926 | * Submission latency for sqpoll_thread is just the time it | |
927 | * takes to fill in the SQ ring entries, and any syscall if | |
928 | * IORING_SQ_NEED_WAKEUP is set, we don't need to log that time | |
929 | * separately. | |
930 | */ | |
931 | td->o.disable_slat = 1; | |
855dc4d4 AG |
932 | } |
933 | if (o->cmd_type == FIO_URING_CMD_NVME) { | |
934 | p.flags |= IORING_SETUP_SQE128; | |
935 | p.flags |= IORING_SETUP_CQE32; | |
936 | } | |
937 | ||
938 | /* | |
939 | * Clamp CQ ring size at our SQ ring size, we don't need more entries | |
940 | * than that. | |
941 | */ | |
942 | p.flags |= IORING_SETUP_CQSIZE; | |
943 | p.cq_entries = depth; | |
944 | ||
07f78c37 AK |
945 | /* |
946 | * Setup COOP_TASKRUN as we don't need to get IPI interrupted for | |
947 | * completing IO operations. | |
948 | */ | |
949 | p.flags |= IORING_SETUP_COOP_TASKRUN; | |
950 | ||
951 | /* | |
952 | * io_uring is always a single issuer, and we can defer task_work | |
953 | * runs until we reap events. | |
954 | */ | |
955 | p.flags |= IORING_SETUP_SINGLE_ISSUER | IORING_SETUP_DEFER_TASKRUN; | |
956 | ||
855dc4d4 AG |
957 | retry: |
958 | ret = syscall(__NR_io_uring_setup, depth, &p); | |
959 | if (ret < 0) { | |
07f78c37 AK |
960 | if (errno == EINVAL && p.flags & IORING_SETUP_DEFER_TASKRUN) { |
961 | p.flags &= ~IORING_SETUP_DEFER_TASKRUN; | |
962 | p.flags &= ~IORING_SETUP_SINGLE_ISSUER; | |
963 | goto retry; | |
964 | } | |
965 | if (errno == EINVAL && p.flags & IORING_SETUP_COOP_TASKRUN) { | |
966 | p.flags &= ~IORING_SETUP_COOP_TASKRUN; | |
967 | goto retry; | |
968 | } | |
855dc4d4 AG |
969 | if (errno == EINVAL && p.flags & IORING_SETUP_CQSIZE) { |
970 | p.flags &= ~IORING_SETUP_CQSIZE; | |
971 | goto retry; | |
972 | } | |
973 | return ret; | |
974 | } | |
975 | ||
976 | ld->ring_fd = ret; | |
977 | ||
978 | fio_ioring_probe(td); | |
979 | ||
980 | if (o->fixedbufs) { | |
981 | ret = syscall(__NR_io_uring_register, ld->ring_fd, | |
982 | IORING_REGISTER_BUFFERS, ld->iovecs, depth); | |
983 | if (ret < 0) | |
984 | return ret; | |
985 | } | |
986 | ||
987 | return fio_ioring_mmap(ld, &p); | |
988 | } | |
989 | ||
5ffd5626 JA |
990 | static int fio_ioring_register_files(struct thread_data *td) |
991 | { | |
992 | struct ioring_data *ld = td->io_ops_data; | |
993 | struct fio_file *f; | |
994 | unsigned int i; | |
995 | int ret; | |
996 | ||
997 | ld->fds = calloc(td->o.nr_files, sizeof(int)); | |
998 | ||
999 | for_each_file(td, f, i) { | |
1000 | ret = generic_open_file(td, f); | |
1001 | if (ret) | |
1002 | goto err; | |
1003 | ld->fds[i] = f->fd; | |
1004 | f->engine_pos = i; | |
1005 | } | |
1006 | ||
bfed648c | 1007 | ret = syscall(__NR_io_uring_register, ld->ring_fd, |
5ffd5626 JA |
1008 | IORING_REGISTER_FILES, ld->fds, td->o.nr_files); |
1009 | if (ret) { | |
1010 | err: | |
1011 | free(ld->fds); | |
1012 | ld->fds = NULL; | |
1013 | } | |
1014 | ||
1015 | /* | |
1016 | * Pretend the file is closed again, and really close it if we hit | |
1017 | * an error. | |
1018 | */ | |
1019 | for_each_file(td, f, i) { | |
1020 | if (ret) { | |
1021 | int fio_unused ret2; | |
1022 | ret2 = generic_close_file(td, f); | |
1023 | } else | |
1024 | f->fd = -1; | |
1025 | } | |
1026 | ||
1027 | return ret; | |
1028 | } | |
1029 | ||
bffad86f | 1030 | static int fio_ioring_post_init(struct thread_data *td) |
52885fa2 | 1031 | { |
bffad86f | 1032 | struct ioring_data *ld = td->io_ops_data; |
5ffd5626 | 1033 | struct ioring_options *o = td->eo; |
52885fa2 | 1034 | struct io_u *io_u; |
650346e1 | 1035 | int err, i; |
52885fa2 | 1036 | |
650346e1 JA |
1037 | for (i = 0; i < td->o.iodepth; i++) { |
1038 | struct iovec *iov = &ld->iovecs[i]; | |
9a2d78b3 | 1039 | |
650346e1 JA |
1040 | io_u = ld->io_u_index[i]; |
1041 | iov->iov_base = io_u->buf; | |
1042 | iov->iov_len = td_max_bs(td); | |
52885fa2 JA |
1043 | } |
1044 | ||
bffad86f | 1045 | err = fio_ioring_queue_init(td); |
52885fa2 | 1046 | if (err) { |
0442b53f | 1047 | int init_err = errno; |
c4f5c92f | 1048 | |
0442b53f | 1049 | if (init_err == ENOSYS) |
c4f5c92f | 1050 | log_err("fio: your kernel doesn't support io_uring\n"); |
0442b53f | 1051 | td_verror(td, init_err, "io_queue_init"); |
52885fa2 JA |
1052 | return 1; |
1053 | } | |
1054 | ||
7c70f506 JA |
1055 | for (i = 0; i < td->o.iodepth; i++) { |
1056 | struct io_uring_sqe *sqe; | |
1057 | ||
1058 | sqe = &ld->sqes[i]; | |
1059 | memset(sqe, 0, sizeof(*sqe)); | |
1060 | } | |
1061 | ||
5ffd5626 JA |
1062 | if (o->registerfiles) { |
1063 | err = fio_ioring_register_files(td); | |
1064 | if (err) { | |
1065 | td_verror(td, errno, "ioring_register_files"); | |
1066 | return 1; | |
1067 | } | |
1068 | } | |
1069 | ||
52885fa2 JA |
1070 | return 0; |
1071 | } | |
1072 | ||
855dc4d4 AG |
1073 | static int fio_ioring_cmd_post_init(struct thread_data *td) |
1074 | { | |
1075 | struct ioring_data *ld = td->io_ops_data; | |
1076 | struct ioring_options *o = td->eo; | |
1077 | struct io_u *io_u; | |
1078 | int err, i; | |
1079 | ||
1080 | for (i = 0; i < td->o.iodepth; i++) { | |
1081 | struct iovec *iov = &ld->iovecs[i]; | |
1082 | ||
1083 | io_u = ld->io_u_index[i]; | |
1084 | iov->iov_base = io_u->buf; | |
1085 | iov->iov_len = td_max_bs(td); | |
1086 | } | |
1087 | ||
1088 | err = fio_ioring_cmd_queue_init(td); | |
1089 | if (err) { | |
1090 | int init_err = errno; | |
1091 | ||
1092 | td_verror(td, init_err, "io_queue_init"); | |
1093 | return 1; | |
1094 | } | |
1095 | ||
1096 | for (i = 0; i < td->o.iodepth; i++) { | |
1097 | struct io_uring_sqe *sqe; | |
1098 | ||
1099 | if (o->cmd_type == FIO_URING_CMD_NVME) { | |
1100 | sqe = &ld->sqes[i << 1]; | |
1101 | memset(sqe, 0, 2 * sizeof(*sqe)); | |
1102 | } else { | |
1103 | sqe = &ld->sqes[i]; | |
1104 | memset(sqe, 0, sizeof(*sqe)); | |
1105 | } | |
1106 | } | |
1107 | ||
1108 | if (o->registerfiles) { | |
1109 | err = fio_ioring_register_files(td); | |
1110 | if (err) { | |
1111 | td_verror(td, errno, "ioring_register_files"); | |
1112 | return 1; | |
1113 | } | |
1114 | } | |
1115 | ||
1116 | return 0; | |
1117 | } | |
1118 | ||
3ee8311a AK |
1119 | static void parse_prchk_flags(struct ioring_options *o) |
1120 | { | |
1121 | if (!o->pi_chk) | |
1122 | return; | |
1123 | ||
1124 | if (strstr(o->pi_chk, "GUARD") != NULL) | |
1125 | o->prchk = NVME_IO_PRINFO_PRCHK_GUARD; | |
1126 | if (strstr(o->pi_chk, "REFTAG") != NULL) | |
1127 | o->prchk |= NVME_IO_PRINFO_PRCHK_REF; | |
1128 | if (strstr(o->pi_chk, "APPTAG") != NULL) | |
1129 | o->prchk |= NVME_IO_PRINFO_PRCHK_APP; | |
1130 | } | |
1131 | ||
bffad86f | 1132 | static int fio_ioring_init(struct thread_data *td) |
52885fa2 | 1133 | { |
5ffd5626 | 1134 | struct ioring_options *o = td->eo; |
bffad86f | 1135 | struct ioring_data *ld; |
2d6451c9 | 1136 | unsigned long long md_size; |
e9f6567a | 1137 | int ret; |
52885fa2 | 1138 | |
5ffd5626 JA |
1139 | /* sqthread submission requires registered files */ |
1140 | if (o->sqpoll_thread) | |
1141 | o->registerfiles = 1; | |
1142 | ||
1143 | if (o->registerfiles && td->o.nr_files != td->o.open_files) { | |
1144 | log_err("fio: io_uring registered files require nr_files to " | |
1145 | "be identical to open_files\n"); | |
1146 | return 1; | |
1147 | } | |
1148 | ||
52885fa2 JA |
1149 | ld = calloc(1, sizeof(*ld)); |
1150 | ||
b87aa01a JA |
1151 | /* ring depth must be a power-of-2 */ |
1152 | ld->iodepth = td->o.iodepth; | |
1153 | td->o.iodepth = roundup_pow2(td->o.iodepth); | |
1154 | ||
52885fa2 JA |
1155 | /* io_u index */ |
1156 | ld->io_u_index = calloc(td->o.iodepth, sizeof(struct io_u *)); | |
2d6451c9 AK |
1157 | |
1158 | /* | |
1159 | * metadata buffer for nvme command. | |
1160 | * We are only supporting iomem=malloc / mem=malloc as of now. | |
1161 | */ | |
1162 | if (!strcmp(td->io_ops->name, "io_uring_cmd") && | |
1163 | (o->cmd_type == FIO_URING_CMD_NVME) && o->md_per_io_size) { | |
1164 | md_size = (unsigned long long) o->md_per_io_size | |
1165 | * (unsigned long long) td->o.iodepth; | |
1166 | md_size += page_mask + td->o.mem_align; | |
1167 | if (td->o.mem_align && td->o.mem_align > page_size) | |
1168 | md_size += td->o.mem_align - page_size; | |
1169 | if (td->o.mem_type == MEM_MALLOC) { | |
1170 | ld->md_buf = malloc(md_size); | |
6795954b JA |
1171 | if (!ld->md_buf) { |
1172 | free(ld); | |
2d6451c9 | 1173 | return 1; |
6795954b | 1174 | } |
2d6451c9 AK |
1175 | } else { |
1176 | log_err("fio: Only iomem=malloc or mem=malloc is supported\n"); | |
6795954b | 1177 | free(ld); |
2d6451c9 AK |
1178 | return 1; |
1179 | } | |
1180 | } | |
3ee8311a | 1181 | parse_prchk_flags(o); |
2d6451c9 | 1182 | |
650346e1 | 1183 | ld->iovecs = calloc(td->o.iodepth, sizeof(struct iovec)); |
52885fa2 JA |
1184 | |
1185 | td->io_ops_data = ld; | |
b2a432bf | 1186 | |
d6cbeab4 | 1187 | ret = fio_cmdprio_init(td, &ld->cmdprio, &o->cmdprio_options); |
e9f6567a DLM |
1188 | if (ret) { |
1189 | td_verror(td, EINVAL, "fio_ioring_init"); | |
b2a432bf PC |
1190 | return 1; |
1191 | } | |
1af44196 | 1192 | |
4885a6eb VF |
1193 | /* |
1194 | * For io_uring_cmd, trims are async operations unless we are operating | |
1195 | * in zbd mode where trim means zone reset. | |
1196 | */ | |
1197 | if (!strcmp(td->io_ops->name, "io_uring_cmd") && td_trim(td) && | |
1198 | td->o.zone_mode == ZONE_MODE_ZBD) | |
1199 | td->io_ops->flags |= FIO_ASYNCIO_SYNC_TRIM; | |
1200 | else | |
ecc734c6 | 1201 | ld->dsm = calloc(td->o.iodepth, sizeof(*ld->dsm)); |
4885a6eb | 1202 | |
52885fa2 JA |
1203 | return 0; |
1204 | } | |
1205 | ||
bffad86f | 1206 | static int fio_ioring_io_u_init(struct thread_data *td, struct io_u *io_u) |
52885fa2 | 1207 | { |
bffad86f | 1208 | struct ioring_data *ld = td->io_ops_data; |
2d6451c9 | 1209 | struct ioring_options *o = td->eo; |
5163f35e | 1210 | struct nvme_pi_data *pi_data; |
2d6451c9 | 1211 | char *p; |
52885fa2 JA |
1212 | |
1213 | ld->io_u_index[io_u->index] = io_u; | |
2d6451c9 AK |
1214 | |
1215 | if (!strcmp(td->io_ops->name, "io_uring_cmd")) { | |
1216 | p = PTR_ALIGN(ld->md_buf, page_mask) + td->o.mem_align; | |
1217 | p += o->md_per_io_size * io_u->index; | |
1218 | io_u->mmap_data = p; | |
5163f35e AK |
1219 | |
1220 | if (!o->pi_act) { | |
1221 | pi_data = calloc(1, sizeof(*pi_data)); | |
1222 | pi_data->io_flags |= o->prchk; | |
1223 | pi_data->apptag_mask = o->apptag_mask; | |
1224 | pi_data->apptag = o->apptag; | |
1225 | io_u->engine_data = pi_data; | |
1226 | } | |
2d6451c9 AK |
1227 | } |
1228 | ||
52885fa2 JA |
1229 | return 0; |
1230 | } | |
1231 | ||
5163f35e AK |
1232 | static void fio_ioring_io_u_free(struct thread_data *td, struct io_u *io_u) |
1233 | { | |
1234 | struct ioring_options *o = td->eo; | |
1235 | struct nvme_pi *pi; | |
1236 | ||
1237 | if (!strcmp(td->io_ops->name, "io_uring_cmd") && | |
1238 | (o->cmd_type == FIO_URING_CMD_NVME)) { | |
1239 | pi = io_u->engine_data; | |
1240 | free(pi); | |
1241 | io_u->engine_data = NULL; | |
1242 | } | |
1243 | } | |
1244 | ||
5ffd5626 JA |
1245 | static int fio_ioring_open_file(struct thread_data *td, struct fio_file *f) |
1246 | { | |
1247 | struct ioring_data *ld = td->io_ops_data; | |
1248 | struct ioring_options *o = td->eo; | |
1249 | ||
17318cf6 | 1250 | if (!ld || !o->registerfiles) |
5ffd5626 JA |
1251 | return generic_open_file(td, f); |
1252 | ||
1253 | f->fd = ld->fds[f->engine_pos]; | |
1254 | return 0; | |
1255 | } | |
1256 | ||
855dc4d4 AG |
1257 | static int fio_ioring_cmd_open_file(struct thread_data *td, struct fio_file *f) |
1258 | { | |
1259 | struct ioring_data *ld = td->io_ops_data; | |
1260 | struct ioring_options *o = td->eo; | |
1261 | ||
1262 | if (o->cmd_type == FIO_URING_CMD_NVME) { | |
1263 | struct nvme_data *data = NULL; | |
e7e5023b | 1264 | unsigned int lba_size = 0; |
671aa9f5 | 1265 | __u64 nlba = 0; |
855dc4d4 AG |
1266 | int ret; |
1267 | ||
1268 | /* Store the namespace-id and lba size. */ | |
1269 | data = FILE_ENG_DATA(f); | |
1270 | if (data == NULL) { | |
855dc4d4 | 1271 | data = calloc(1, sizeof(struct nvme_data)); |
3ee8311a | 1272 | ret = fio_nvme_get_info(f, &nlba, o->pi_act, data); |
e7e5023b AK |
1273 | if (ret) { |
1274 | free(data); | |
1275 | return ret; | |
1276 | } | |
855dc4d4 AG |
1277 | |
1278 | FILE_SET_ENG_DATA(f, data); | |
1279 | } | |
345fa8fd | 1280 | |
e7e5023b | 1281 | lba_size = data->lba_ext ? data->lba_ext : data->lba_size; |
345fa8fd AK |
1282 | |
1283 | for_each_rw_ddir(ddir) { | |
3cb50530 VF |
1284 | if (td->o.min_bs[ddir] % lba_size || td->o.max_bs[ddir] % lba_size) { |
1285 | if (data->lba_ext) { | |
1286 | log_err("%s: block size must be a multiple of %u " | |
1287 | "(LBA data size + Metadata size)\n", f->file_name, lba_size); | |
1288 | if (td->o.min_bs[ddir] == td->o.max_bs[ddir] && | |
1289 | !(td->o.min_bs[ddir] % data->lba_size)) { | |
1290 | /* fixed block size is actually a multiple of LBA data size */ | |
1291 | unsigned long long suggestion = lba_size * | |
1292 | (td->o.min_bs[ddir] / data->lba_size); | |
1293 | log_err("Did you mean to use a block size of %llu?\n", suggestion); | |
1294 | } | |
1295 | } else { | |
a904d182 AK |
1296 | log_err("%s: block size must be a multiple of LBA data size\n", |
1297 | f->file_name); | |
3cb50530 | 1298 | } |
a904d182 | 1299 | td_verror(td, EINVAL, "fio_ioring_cmd_open_file"); |
345fa8fd AK |
1300 | return 1; |
1301 | } | |
2d6451c9 AK |
1302 | if (data->ms && !data->lba_ext && ddir != DDIR_TRIM && |
1303 | (o->md_per_io_size < ((td->o.max_bs[ddir] / data->lba_size) * | |
1304 | data->ms))) { | |
1305 | log_err("%s: md_per_io_size should be at least %llu bytes\n", | |
1306 | f->file_name, | |
1307 | ((td->o.max_bs[ddir] / data->lba_size) * data->ms)); | |
1308 | td_verror(td, EINVAL, "fio_ioring_cmd_open_file"); | |
1309 | return 1; | |
1310 | } | |
345fa8fd | 1311 | } |
ba342e58 AK |
1312 | |
1313 | /* | |
1314 | * For extended logical block sizes we cannot use verify when | |
1315 | * end to end data protection checks are enabled, as the PI | |
1316 | * section of data buffer conflicts with verify. | |
1317 | */ | |
1318 | if (data->ms && data->pi_type && data->lba_ext && | |
1319 | td->o.verify != VERIFY_NONE) { | |
1320 | log_err("%s: for extended LBA, verify cannot be used when E2E data protection is enabled\n", | |
1321 | f->file_name); | |
1322 | td_verror(td, EINVAL, "fio_ioring_cmd_open_file"); | |
1323 | return 1; | |
1324 | } | |
855dc4d4 AG |
1325 | } |
1326 | if (!ld || !o->registerfiles) | |
1327 | return generic_open_file(td, f); | |
1328 | ||
1329 | f->fd = ld->fds[f->engine_pos]; | |
1330 | return 0; | |
1331 | } | |
1332 | ||
5ffd5626 JA |
1333 | static int fio_ioring_close_file(struct thread_data *td, struct fio_file *f) |
1334 | { | |
17318cf6 | 1335 | struct ioring_data *ld = td->io_ops_data; |
5ffd5626 JA |
1336 | struct ioring_options *o = td->eo; |
1337 | ||
17318cf6 | 1338 | if (!ld || !o->registerfiles) |
5ffd5626 JA |
1339 | return generic_close_file(td, f); |
1340 | ||
1341 | f->fd = -1; | |
1342 | return 0; | |
1343 | } | |
1344 | ||
855dc4d4 AG |
1345 | static int fio_ioring_cmd_close_file(struct thread_data *td, |
1346 | struct fio_file *f) | |
1347 | { | |
1348 | struct ioring_data *ld = td->io_ops_data; | |
1349 | struct ioring_options *o = td->eo; | |
1350 | ||
1351 | if (o->cmd_type == FIO_URING_CMD_NVME) { | |
1352 | struct nvme_data *data = FILE_ENG_DATA(f); | |
1353 | ||
1354 | FILE_SET_ENG_DATA(f, NULL); | |
1355 | free(data); | |
1356 | } | |
1357 | if (!ld || !o->registerfiles) | |
1358 | return generic_close_file(td, f); | |
1359 | ||
1360 | f->fd = -1; | |
1361 | return 0; | |
1362 | } | |
1363 | ||
1364 | static int fio_ioring_cmd_get_file_size(struct thread_data *td, | |
1365 | struct fio_file *f) | |
1366 | { | |
1367 | struct ioring_options *o = td->eo; | |
1368 | ||
1369 | if (fio_file_size_known(f)) | |
1370 | return 0; | |
1371 | ||
1372 | if (o->cmd_type == FIO_URING_CMD_NVME) { | |
1373 | struct nvme_data *data = NULL; | |
671aa9f5 | 1374 | __u64 nlba = 0; |
855dc4d4 AG |
1375 | int ret; |
1376 | ||
855dc4d4 | 1377 | data = calloc(1, sizeof(struct nvme_data)); |
3ee8311a | 1378 | ret = fio_nvme_get_info(f, &nlba, o->pi_act, data); |
e7e5023b AK |
1379 | if (ret) { |
1380 | free(data); | |
1381 | return ret; | |
1382 | } | |
855dc4d4 | 1383 | |
e7e5023b | 1384 | f->real_file_size = data->lba_size * nlba; |
855dc4d4 AG |
1385 | fio_file_set_size_known(f); |
1386 | ||
1387 | FILE_SET_ENG_DATA(f, data); | |
1388 | return 0; | |
1389 | } | |
1390 | return generic_get_file_size(td, f); | |
1391 | } | |
1392 | ||
3d05e0ff AK |
1393 | static int fio_ioring_cmd_get_zoned_model(struct thread_data *td, |
1394 | struct fio_file *f, | |
1395 | enum zbd_zoned_model *model) | |
1396 | { | |
1397 | return fio_nvme_get_zoned_model(td, f, model); | |
1398 | } | |
1399 | ||
1400 | static int fio_ioring_cmd_report_zones(struct thread_data *td, | |
1401 | struct fio_file *f, uint64_t offset, | |
1402 | struct zbd_zone *zbdz, | |
1403 | unsigned int nr_zones) | |
1404 | { | |
1405 | return fio_nvme_report_zones(td, f, offset, zbdz, nr_zones); | |
1406 | } | |
1407 | ||
1408 | static int fio_ioring_cmd_reset_wp(struct thread_data *td, struct fio_file *f, | |
1409 | uint64_t offset, uint64_t length) | |
1410 | { | |
1411 | return fio_nvme_reset_wp(td, f, offset, length); | |
1412 | } | |
1413 | ||
1414 | static int fio_ioring_cmd_get_max_open_zones(struct thread_data *td, | |
1415 | struct fio_file *f, | |
1416 | unsigned int *max_open_zones) | |
1417 | { | |
1418 | return fio_nvme_get_max_open_zones(td, f, max_open_zones); | |
1419 | } | |
1420 | ||
a7e8aae0 KB |
1421 | static int fio_ioring_cmd_fetch_ruhs(struct thread_data *td, struct fio_file *f, |
1422 | struct fio_ruhs_info *fruhs_info) | |
1423 | { | |
1424 | struct nvme_fdp_ruh_status *ruhs; | |
1425 | int bytes, ret, i; | |
1426 | ||
98cd3c0e | 1427 | bytes = sizeof(*ruhs) + FDP_MAX_RUHS * sizeof(struct nvme_fdp_ruh_status_desc); |
a7e8aae0 KB |
1428 | ruhs = scalloc(1, bytes); |
1429 | if (!ruhs) | |
1430 | return -ENOMEM; | |
1431 | ||
1432 | ret = fio_nvme_iomgmt_ruhs(td, f, ruhs, bytes); | |
1433 | if (ret) | |
1434 | goto free; | |
1435 | ||
1436 | fruhs_info->nr_ruhs = le16_to_cpu(ruhs->nruhsd); | |
1437 | for (i = 0; i < fruhs_info->nr_ruhs; i++) | |
1438 | fruhs_info->plis[i] = le16_to_cpu(ruhs->ruhss[i].pid); | |
1439 | free: | |
1440 | sfree(ruhs); | |
1441 | return ret; | |
1442 | } | |
1443 | ||
855dc4d4 | 1444 | static struct ioengine_ops ioengine_uring = { |
bffad86f | 1445 | .name = "io_uring", |
52885fa2 | 1446 | .version = FIO_IOOPS_VERSION, |
4e7e7898 VF |
1447 | .flags = FIO_ASYNCIO_SYNC_TRIM | FIO_NO_OFFLOAD | |
1448 | FIO_ASYNCIO_SETS_ISSUE_TIME, | |
bffad86f JA |
1449 | .init = fio_ioring_init, |
1450 | .post_init = fio_ioring_post_init, | |
1451 | .io_u_init = fio_ioring_io_u_init, | |
1452 | .prep = fio_ioring_prep, | |
1453 | .queue = fio_ioring_queue, | |
1454 | .commit = fio_ioring_commit, | |
1455 | .getevents = fio_ioring_getevents, | |
1456 | .event = fio_ioring_event, | |
1457 | .cleanup = fio_ioring_cleanup, | |
5ffd5626 JA |
1458 | .open_file = fio_ioring_open_file, |
1459 | .close_file = fio_ioring_close_file, | |
52885fa2 JA |
1460 | .get_file_size = generic_get_file_size, |
1461 | .options = options, | |
bffad86f | 1462 | .option_struct_size = sizeof(struct ioring_options), |
52885fa2 JA |
1463 | }; |
1464 | ||
855dc4d4 AG |
1465 | static struct ioengine_ops ioengine_uring_cmd = { |
1466 | .name = "io_uring_cmd", | |
1467 | .version = FIO_IOOPS_VERSION, | |
4885a6eb | 1468 | .flags = FIO_NO_OFFLOAD | FIO_MEMALIGN | FIO_RAWIO | |
4e7e7898 | 1469 | FIO_ASYNCIO_SETS_ISSUE_TIME, |
855dc4d4 AG |
1470 | .init = fio_ioring_init, |
1471 | .post_init = fio_ioring_cmd_post_init, | |
1472 | .io_u_init = fio_ioring_io_u_init, | |
5163f35e | 1473 | .io_u_free = fio_ioring_io_u_free, |
855dc4d4 AG |
1474 | .prep = fio_ioring_cmd_prep, |
1475 | .queue = fio_ioring_queue, | |
1476 | .commit = fio_ioring_commit, | |
1477 | .getevents = fio_ioring_getevents, | |
1478 | .event = fio_ioring_cmd_event, | |
1479 | .cleanup = fio_ioring_cleanup, | |
1480 | .open_file = fio_ioring_cmd_open_file, | |
1481 | .close_file = fio_ioring_cmd_close_file, | |
1482 | .get_file_size = fio_ioring_cmd_get_file_size, | |
3d05e0ff AK |
1483 | .get_zoned_model = fio_ioring_cmd_get_zoned_model, |
1484 | .report_zones = fio_ioring_cmd_report_zones, | |
1485 | .reset_wp = fio_ioring_cmd_reset_wp, | |
1486 | .get_max_open_zones = fio_ioring_cmd_get_max_open_zones, | |
855dc4d4 AG |
1487 | .options = options, |
1488 | .option_struct_size = sizeof(struct ioring_options), | |
a7e8aae0 | 1489 | .fdp_fetch_ruhs = fio_ioring_cmd_fetch_ruhs, |
855dc4d4 AG |
1490 | }; |
1491 | ||
bffad86f | 1492 | static void fio_init fio_ioring_register(void) |
52885fa2 | 1493 | { |
855dc4d4 AG |
1494 | register_ioengine(&ioengine_uring); |
1495 | register_ioengine(&ioengine_uring_cmd); | |
52885fa2 JA |
1496 | } |
1497 | ||
bffad86f | 1498 | static void fio_exit fio_ioring_unregister(void) |
52885fa2 | 1499 | { |
855dc4d4 AG |
1500 | unregister_ioengine(&ioengine_uring); |
1501 | unregister_ioengine(&ioengine_uring_cmd); | |
52885fa2 | 1502 | } |
1f90e9bb | 1503 | #endif |