Commit | Line | Data |
---|---|---|
a3ff873e AK |
1 | /* |
2 | * fio xNVMe IO Engine | |
3 | * | |
4 | * IO engine using the xNVMe C API. | |
5 | * | |
6 | * See: http://xnvme.io/ | |
7 | * | |
8 | * SPDX-License-Identifier: Apache-2.0 | |
9 | */ | |
10 | #include <stdlib.h> | |
11 | #include <assert.h> | |
12 | #include <libxnvme.h> | |
a3ff873e AK |
13 | #include "fio.h" |
14 | #include "zbd_types.h" | |
e5f3b613 | 15 | #include "fdp.h" |
a3ff873e AK |
16 | #include "optgroup.h" |
17 | ||
18 | static pthread_mutex_t g_serialize = PTHREAD_MUTEX_INITIALIZER; | |
19 | ||
20 | struct xnvme_fioe_fwrap { | |
21 | /* fio file representation */ | |
22 | struct fio_file *fio_file; | |
23 | ||
24 | /* xNVMe device handle */ | |
25 | struct xnvme_dev *dev; | |
26 | /* xNVMe device geometry */ | |
27 | const struct xnvme_geo *geo; | |
28 | ||
29 | struct xnvme_queue *queue; | |
30 | ||
31 | uint32_t ssw; | |
32 | uint32_t lba_nbytes; | |
be5514e3 AK |
33 | uint32_t md_nbytes; |
34 | uint32_t lba_pow2; | |
a3ff873e | 35 | |
be5514e3 | 36 | uint8_t _pad[16]; |
a3ff873e AK |
37 | }; |
38 | XNVME_STATIC_ASSERT(sizeof(struct xnvme_fioe_fwrap) == 64, "Incorrect size") | |
39 | ||
40 | struct xnvme_fioe_data { | |
41 | /* I/O completion queue */ | |
42 | struct io_u **iocq; | |
43 | ||
44 | /* # of iocq entries; incremented via getevents()/cb_pool() */ | |
45 | uint64_t completed; | |
46 | ||
47 | /* | |
48 | * # of errors; incremented when observed on completion via | |
49 | * getevents()/cb_pool() | |
50 | */ | |
51 | uint64_t ecount; | |
52 | ||
53 | /* Controller which device/file to select */ | |
54 | int32_t prev; | |
55 | int32_t cur; | |
56 | ||
57 | /* Number of devices/files for which open() has been called */ | |
58 | int64_t nopen; | |
59 | /* Number of devices/files allocated in files[] */ | |
60 | uint64_t nallocated; | |
61 | ||
62 | struct iovec *iovec; | |
be5514e3 | 63 | struct iovec *md_iovec; |
a3ff873e AK |
64 | |
65 | struct xnvme_fioe_fwrap files[]; | |
66 | }; | |
67 | XNVME_STATIC_ASSERT(sizeof(struct xnvme_fioe_data) == 64, "Incorrect size") | |
68 | ||
be5514e3 AK |
69 | struct xnvme_fioe_request { |
70 | /* Separate metadata buffer pointer */ | |
71 | void *md_buf; | |
72 | }; | |
73 | ||
a3ff873e AK |
74 | struct xnvme_fioe_options { |
75 | void *padding; | |
76 | unsigned int hipri; | |
77 | unsigned int sqpoll_thread; | |
78 | unsigned int xnvme_dev_nsid; | |
79 | unsigned int xnvme_iovec; | |
be5514e3 | 80 | unsigned int md_per_io_size; |
a3ff873e | 81 | char *xnvme_be; |
c945074c | 82 | char *xnvme_mem; |
a3ff873e AK |
83 | char *xnvme_async; |
84 | char *xnvme_sync; | |
85 | char *xnvme_admin; | |
efbafe2a | 86 | char *xnvme_dev_subnqn; |
a3ff873e AK |
87 | }; |
88 | ||
89 | static struct fio_option options[] = { | |
90 | { | |
91 | .name = "hipri", | |
92 | .lname = "High Priority", | |
93 | .type = FIO_OPT_STR_SET, | |
94 | .off1 = offsetof(struct xnvme_fioe_options, hipri), | |
95 | .help = "Use polled IO completions", | |
96 | .category = FIO_OPT_C_ENGINE, | |
97 | .group = FIO_OPT_G_XNVME, | |
98 | }, | |
99 | { | |
100 | .name = "sqthread_poll", | |
101 | .lname = "Kernel SQ thread polling", | |
102 | .type = FIO_OPT_STR_SET, | |
103 | .off1 = offsetof(struct xnvme_fioe_options, sqpoll_thread), | |
104 | .help = "Offload submission/completion to kernel thread", | |
105 | .category = FIO_OPT_C_ENGINE, | |
106 | .group = FIO_OPT_G_XNVME, | |
107 | }, | |
108 | { | |
109 | .name = "xnvme_be", | |
110 | .lname = "xNVMe Backend", | |
111 | .type = FIO_OPT_STR_STORE, | |
112 | .off1 = offsetof(struct xnvme_fioe_options, xnvme_be), | |
113 | .help = "Select xNVMe backend [spdk,linux,fbsd]", | |
114 | .category = FIO_OPT_C_ENGINE, | |
115 | .group = FIO_OPT_G_XNVME, | |
116 | }, | |
c945074c AK |
117 | { |
118 | .name = "xnvme_mem", | |
119 | .lname = "xNVMe Memory Backend", | |
120 | .type = FIO_OPT_STR_STORE, | |
121 | .off1 = offsetof(struct xnvme_fioe_options, xnvme_mem), | |
122 | .help = "Select xNVMe memory backend", | |
123 | .category = FIO_OPT_C_ENGINE, | |
124 | .group = FIO_OPT_G_XNVME, | |
125 | }, | |
a3ff873e AK |
126 | { |
127 | .name = "xnvme_async", | |
128 | .lname = "xNVMe Asynchronous command-interface", | |
129 | .type = FIO_OPT_STR_STORE, | |
130 | .off1 = offsetof(struct xnvme_fioe_options, xnvme_async), | |
203a4c7c AK |
131 | .help = "Select xNVMe async. interface: " |
132 | "[emu,thrpool,io_uring,io_uring_cmd,libaio,posix,vfio,nil]", | |
a3ff873e AK |
133 | .category = FIO_OPT_C_ENGINE, |
134 | .group = FIO_OPT_G_XNVME, | |
135 | }, | |
136 | { | |
137 | .name = "xnvme_sync", | |
138 | .lname = "xNVMe Synchronous. command-interface", | |
139 | .type = FIO_OPT_STR_STORE, | |
140 | .off1 = offsetof(struct xnvme_fioe_options, xnvme_sync), | |
203a4c7c | 141 | .help = "Select xNVMe sync. interface: [nvme,psync,block]", |
a3ff873e AK |
142 | .category = FIO_OPT_C_ENGINE, |
143 | .group = FIO_OPT_G_XNVME, | |
144 | }, | |
145 | { | |
146 | .name = "xnvme_admin", | |
147 | .lname = "xNVMe Admin command-interface", | |
148 | .type = FIO_OPT_STR_STORE, | |
149 | .off1 = offsetof(struct xnvme_fioe_options, xnvme_admin), | |
203a4c7c | 150 | .help = "Select xNVMe admin. cmd-interface: [nvme,block]", |
a3ff873e AK |
151 | .category = FIO_OPT_C_ENGINE, |
152 | .group = FIO_OPT_G_XNVME, | |
153 | }, | |
154 | { | |
155 | .name = "xnvme_dev_nsid", | |
156 | .lname = "xNVMe Namespace-Identifier, for user-space NVMe driver", | |
157 | .type = FIO_OPT_INT, | |
158 | .off1 = offsetof(struct xnvme_fioe_options, xnvme_dev_nsid), | |
159 | .help = "xNVMe Namespace-Identifier, for user-space NVMe driver", | |
160 | .category = FIO_OPT_C_ENGINE, | |
161 | .group = FIO_OPT_G_XNVME, | |
162 | }, | |
efbafe2a AK |
163 | { |
164 | .name = "xnvme_dev_subnqn", | |
165 | .lname = "Subsystem nqn for Fabrics", | |
166 | .type = FIO_OPT_STR_STORE, | |
167 | .off1 = offsetof(struct xnvme_fioe_options, xnvme_dev_subnqn), | |
168 | .help = "Subsystem NQN for Fabrics", | |
169 | .category = FIO_OPT_C_ENGINE, | |
170 | .group = FIO_OPT_G_XNVME, | |
171 | }, | |
a3ff873e AK |
172 | { |
173 | .name = "xnvme_iovec", | |
174 | .lname = "Vectored IOs", | |
175 | .type = FIO_OPT_STR_SET, | |
176 | .off1 = offsetof(struct xnvme_fioe_options, xnvme_iovec), | |
177 | .help = "Send vectored IOs", | |
178 | .category = FIO_OPT_C_ENGINE, | |
179 | .group = FIO_OPT_G_XNVME, | |
180 | }, | |
be5514e3 AK |
181 | { |
182 | .name = "md_per_io_size", | |
183 | .lname = "Separate Metadata Buffer Size per I/O", | |
184 | .type = FIO_OPT_INT, | |
185 | .off1 = offsetof(struct xnvme_fioe_options, md_per_io_size), | |
186 | .def = "0", | |
187 | .help = "Size of separate metadata buffer per I/O (Default: 0)", | |
188 | .category = FIO_OPT_C_ENGINE, | |
189 | .group = FIO_OPT_G_XNVME, | |
190 | }, | |
a3ff873e AK |
191 | |
192 | { | |
193 | .name = NULL, | |
194 | }, | |
195 | }; | |
196 | ||
197 | static void cb_pool(struct xnvme_cmd_ctx *ctx, void *cb_arg) | |
198 | { | |
199 | struct io_u *io_u = cb_arg; | |
200 | struct xnvme_fioe_data *xd = io_u->mmap_data; | |
201 | ||
202 | if (xnvme_cmd_ctx_cpl_status(ctx)) { | |
203 | xnvme_cmd_ctx_pr(ctx, XNVME_PR_DEF); | |
204 | xd->ecount += 1; | |
205 | io_u->error = EIO; | |
206 | } | |
207 | ||
208 | xd->iocq[xd->completed++] = io_u; | |
209 | xnvme_queue_put_cmd_ctx(ctx->async.queue, ctx); | |
210 | } | |
211 | ||
212 | static struct xnvme_opts xnvme_opts_from_fioe(struct thread_data *td) | |
213 | { | |
214 | struct xnvme_fioe_options *o = td->eo; | |
215 | struct xnvme_opts opts = xnvme_opts_default(); | |
216 | ||
217 | opts.nsid = o->xnvme_dev_nsid; | |
efbafe2a | 218 | opts.subnqn = o->xnvme_dev_subnqn; |
a3ff873e | 219 | opts.be = o->xnvme_be; |
c945074c | 220 | opts.mem = o->xnvme_mem; |
a3ff873e AK |
221 | opts.async = o->xnvme_async; |
222 | opts.sync = o->xnvme_sync; | |
223 | opts.admin = o->xnvme_admin; | |
224 | ||
225 | opts.poll_io = o->hipri; | |
226 | opts.poll_sq = o->sqpoll_thread; | |
227 | ||
228 | opts.direct = td->o.odirect; | |
229 | ||
230 | return opts; | |
231 | } | |
232 | ||
233 | static void _dev_close(struct thread_data *td, struct xnvme_fioe_fwrap *fwrap) | |
234 | { | |
235 | if (fwrap->dev) | |
236 | xnvme_queue_term(fwrap->queue); | |
237 | ||
238 | xnvme_dev_close(fwrap->dev); | |
239 | ||
240 | memset(fwrap, 0, sizeof(*fwrap)); | |
241 | } | |
242 | ||
243 | static void xnvme_fioe_cleanup(struct thread_data *td) | |
244 | { | |
fdac9c68 | 245 | struct xnvme_fioe_data *xd = NULL; |
a3ff873e AK |
246 | int err; |
247 | ||
fdac9c68 AK |
248 | if (!td->io_ops_data) |
249 | return; | |
250 | ||
251 | xd = td->io_ops_data; | |
252 | ||
a3ff873e AK |
253 | err = pthread_mutex_lock(&g_serialize); |
254 | if (err) | |
255 | log_err("ioeng->cleanup(): pthread_mutex_lock(), err(%d)\n", err); | |
256 | /* NOTE: not returning here */ | |
257 | ||
258 | for (uint64_t i = 0; i < xd->nallocated; ++i) | |
259 | _dev_close(td, &xd->files[i]); | |
260 | ||
261 | if (!err) { | |
262 | err = pthread_mutex_unlock(&g_serialize); | |
263 | if (err) | |
264 | log_err("ioeng->cleanup(): pthread_mutex_unlock(), err(%d)\n", err); | |
265 | } | |
266 | ||
267 | free(xd->iocq); | |
268 | free(xd->iovec); | |
be5514e3 | 269 | free(xd->md_iovec); |
a3ff873e AK |
270 | free(xd); |
271 | td->io_ops_data = NULL; | |
272 | } | |
273 | ||
274 | /** | |
275 | * Helper function setting up device handles as addressed by the naming | |
276 | * convention of the given `fio_file` filename. | |
277 | * | |
278 | * Checks thread-options for explicit control of asynchronous implementation via | |
279 | * the ``--xnvme_async={thrpool,emu,posix,io_uring,libaio,nil}``. | |
280 | */ | |
281 | static int _dev_open(struct thread_data *td, struct fio_file *f) | |
282 | { | |
283 | struct xnvme_opts opts = xnvme_opts_from_fioe(td); | |
284 | struct xnvme_fioe_data *xd = td->io_ops_data; | |
285 | struct xnvme_fioe_fwrap *fwrap; | |
286 | int flags = 0; | |
287 | int err; | |
288 | ||
289 | if (f->fileno > (int)xd->nallocated) { | |
290 | log_err("ioeng->_dev_open(%s): invalid assumption\n", f->file_name); | |
291 | return 1; | |
292 | } | |
293 | ||
294 | fwrap = &xd->files[f->fileno]; | |
295 | ||
296 | err = pthread_mutex_lock(&g_serialize); | |
297 | if (err) { | |
298 | log_err("ioeng->_dev_open(%s): pthread_mutex_lock(), err(%d)\n", f->file_name, | |
299 | err); | |
300 | return -err; | |
301 | } | |
302 | ||
303 | fwrap->dev = xnvme_dev_open(f->file_name, &opts); | |
304 | if (!fwrap->dev) { | |
305 | log_err("ioeng->_dev_open(%s): xnvme_dev_open(), err(%d)\n", f->file_name, errno); | |
306 | goto failure; | |
307 | } | |
308 | fwrap->geo = xnvme_dev_get_geo(fwrap->dev); | |
309 | ||
310 | if (xnvme_queue_init(fwrap->dev, td->o.iodepth, flags, &(fwrap->queue))) { | |
311 | log_err("ioeng->_dev_open(%s): xnvme_queue_init(), err(?)\n", f->file_name); | |
312 | goto failure; | |
313 | } | |
314 | xnvme_queue_set_cb(fwrap->queue, cb_pool, NULL); | |
315 | ||
316 | fwrap->ssw = xnvme_dev_get_ssw(fwrap->dev); | |
317 | fwrap->lba_nbytes = fwrap->geo->lba_nbytes; | |
be5514e3 AK |
318 | fwrap->md_nbytes = fwrap->geo->nbytes_oob; |
319 | ||
320 | if (fwrap->geo->lba_extended) | |
321 | fwrap->lba_pow2 = 0; | |
322 | else | |
323 | fwrap->lba_pow2 = 1; | |
a3ff873e AK |
324 | |
325 | fwrap->fio_file = f; | |
326 | fwrap->fio_file->filetype = FIO_TYPE_BLOCK; | |
327 | fwrap->fio_file->real_file_size = fwrap->geo->tbytes; | |
328 | fio_file_set_size_known(fwrap->fio_file); | |
329 | ||
330 | err = pthread_mutex_unlock(&g_serialize); | |
331 | if (err) | |
332 | log_err("ioeng->_dev_open(%s): pthread_mutex_unlock(), err(%d)\n", f->file_name, | |
333 | err); | |
334 | ||
335 | return 0; | |
336 | ||
337 | failure: | |
338 | xnvme_queue_term(fwrap->queue); | |
339 | xnvme_dev_close(fwrap->dev); | |
340 | ||
341 | err = pthread_mutex_unlock(&g_serialize); | |
342 | if (err) | |
343 | log_err("ioeng->_dev_open(%s): pthread_mutex_unlock(), err(%d)\n", f->file_name, | |
344 | err); | |
345 | ||
346 | return 1; | |
347 | } | |
348 | ||
349 | static int xnvme_fioe_init(struct thread_data *td) | |
350 | { | |
351 | struct xnvme_fioe_data *xd = NULL; | |
7f6a3869 | 352 | struct xnvme_fioe_options *o = td->eo; |
a3ff873e AK |
353 | struct fio_file *f; |
354 | unsigned int i; | |
355 | ||
356 | if (!td->o.use_thread) { | |
357 | log_err("ioeng->init(): --thread=1 is required\n"); | |
358 | return 1; | |
359 | } | |
360 | ||
361 | /* Allocate xd and iocq */ | |
362 | xd = calloc(1, sizeof(*xd) + sizeof(*xd->files) * td->o.nr_files); | |
363 | if (!xd) { | |
364 | log_err("ioeng->init(): !calloc(), err(%d)\n", errno); | |
365 | return 1; | |
366 | } | |
367 | ||
368 | xd->iocq = calloc(td->o.iodepth, sizeof(struct io_u *)); | |
369 | if (!xd->iocq) { | |
eb3570b5 AK |
370 | free(xd); |
371 | log_err("ioeng->init(): !calloc(xd->iocq), err(%d)\n", errno); | |
a3ff873e AK |
372 | return 1; |
373 | } | |
374 | ||
7f6a3869 AK |
375 | if (o->xnvme_iovec) { |
376 | xd->iovec = calloc(td->o.iodepth, sizeof(*xd->iovec)); | |
377 | if (!xd->iovec) { | |
378 | free(xd->iocq); | |
379 | free(xd); | |
380 | log_err("ioeng->init(): !calloc(xd->iovec), err(%d)\n", errno); | |
381 | return 1; | |
382 | } | |
a3ff873e AK |
383 | } |
384 | ||
be5514e3 AK |
385 | if (o->xnvme_iovec && o->md_per_io_size) { |
386 | xd->md_iovec = calloc(td->o.iodepth, sizeof(*xd->md_iovec)); | |
387 | if (!xd->md_iovec) { | |
388 | free(xd->iocq); | |
389 | free(xd->iovec); | |
390 | free(xd); | |
391 | log_err("ioeng->init(): !calloc(xd->md_iovec), err(%d)\n", errno); | |
392 | return 1; | |
393 | } | |
394 | } | |
395 | ||
a3ff873e AK |
396 | xd->prev = -1; |
397 | td->io_ops_data = xd; | |
398 | ||
399 | for_each_file(td, f, i) | |
400 | { | |
401 | if (_dev_open(td, f)) { | |
eb3570b5 | 402 | /* |
be5514e3 AK |
403 | * Note: We are not freeing xd, iocq, iovec and md_iovec. |
404 | * This will be done as part of cleanup routine. | |
eb3570b5 | 405 | */ |
a3ff873e AK |
406 | log_err("ioeng->init(): failed; _dev_open(%s)\n", f->file_name); |
407 | return 1; | |
408 | } | |
409 | ||
410 | ++(xd->nallocated); | |
411 | } | |
412 | ||
413 | if (xd->nallocated != td->o.nr_files) { | |
414 | log_err("ioeng->init(): failed; nallocated != td->o.nr_files\n"); | |
415 | return 1; | |
416 | } | |
417 | ||
418 | return 0; | |
419 | } | |
420 | ||
421 | /* NOTE: using the first device for buffer-allocators) */ | |
422 | static int xnvme_fioe_iomem_alloc(struct thread_data *td, size_t total_mem) | |
423 | { | |
424 | struct xnvme_fioe_data *xd = td->io_ops_data; | |
425 | struct xnvme_fioe_fwrap *fwrap = &xd->files[0]; | |
426 | ||
427 | if (!fwrap->dev) { | |
428 | log_err("ioeng->iomem_alloc(): failed; no dev-handle\n"); | |
429 | return 1; | |
430 | } | |
431 | ||
432 | td->orig_buffer = xnvme_buf_alloc(fwrap->dev, total_mem); | |
433 | ||
434 | return td->orig_buffer == NULL; | |
435 | } | |
436 | ||
437 | /* NOTE: using the first device for buffer-allocators) */ | |
438 | static void xnvme_fioe_iomem_free(struct thread_data *td) | |
439 | { | |
fdac9c68 AK |
440 | struct xnvme_fioe_data *xd = NULL; |
441 | struct xnvme_fioe_fwrap *fwrap = NULL; | |
442 | ||
443 | if (!td->io_ops_data) | |
444 | return; | |
445 | ||
446 | xd = td->io_ops_data; | |
447 | fwrap = &xd->files[0]; | |
a3ff873e AK |
448 | |
449 | if (!fwrap->dev) { | |
450 | log_err("ioeng->iomem_free(): failed no dev-handle\n"); | |
451 | return; | |
452 | } | |
453 | ||
454 | xnvme_buf_free(fwrap->dev, td->orig_buffer); | |
455 | } | |
456 | ||
457 | static int xnvme_fioe_io_u_init(struct thread_data *td, struct io_u *io_u) | |
458 | { | |
be5514e3 AK |
459 | struct xnvme_fioe_request *fio_req; |
460 | struct xnvme_fioe_options *o = td->eo; | |
461 | struct xnvme_fioe_data *xd = td->io_ops_data; | |
462 | struct xnvme_fioe_fwrap *fwrap = &xd->files[0]; | |
463 | ||
464 | if (!fwrap->dev) { | |
465 | log_err("ioeng->io_u_init(): failed; no dev-handle\n"); | |
466 | return 1; | |
467 | } | |
468 | ||
a3ff873e | 469 | io_u->mmap_data = td->io_ops_data; |
be5514e3 AK |
470 | io_u->engine_data = NULL; |
471 | ||
472 | fio_req = calloc(1, sizeof(*fio_req)); | |
473 | if (!fio_req) { | |
474 | log_err("ioeng->io_u_init(): !calloc(fio_req), err(%d)\n", errno); | |
475 | return 1; | |
476 | } | |
477 | ||
478 | if (o->md_per_io_size) { | |
479 | fio_req->md_buf = xnvme_buf_alloc(fwrap->dev, o->md_per_io_size); | |
480 | if (!fio_req->md_buf) { | |
481 | free(fio_req); | |
482 | return 1; | |
483 | } | |
484 | } | |
485 | ||
486 | io_u->engine_data = fio_req; | |
a3ff873e AK |
487 | |
488 | return 0; | |
489 | } | |
490 | ||
491 | static void xnvme_fioe_io_u_free(struct thread_data *td, struct io_u *io_u) | |
492 | { | |
be5514e3 AK |
493 | struct xnvme_fioe_data *xd = NULL; |
494 | struct xnvme_fioe_fwrap *fwrap = NULL; | |
495 | struct xnvme_fioe_request *fio_req = NULL; | |
496 | ||
497 | if (!td->io_ops_data) | |
498 | return; | |
499 | ||
500 | xd = td->io_ops_data; | |
501 | fwrap = &xd->files[0]; | |
502 | ||
503 | if (!fwrap->dev) { | |
504 | log_err("ioeng->io_u_free(): failed no dev-handle\n"); | |
505 | return; | |
506 | } | |
507 | ||
508 | fio_req = io_u->engine_data; | |
509 | if (fio_req->md_buf) | |
510 | xnvme_buf_free(fwrap->dev, fio_req->md_buf); | |
511 | ||
512 | free(fio_req); | |
513 | ||
a3ff873e AK |
514 | io_u->mmap_data = NULL; |
515 | } | |
516 | ||
517 | static struct io_u *xnvme_fioe_event(struct thread_data *td, int event) | |
518 | { | |
519 | struct xnvme_fioe_data *xd = td->io_ops_data; | |
520 | ||
521 | assert(event >= 0); | |
522 | assert((unsigned)event < xd->completed); | |
523 | ||
524 | return xd->iocq[event]; | |
525 | } | |
526 | ||
527 | static int xnvme_fioe_getevents(struct thread_data *td, unsigned int min, unsigned int max, | |
528 | const struct timespec *t) | |
529 | { | |
530 | struct xnvme_fioe_data *xd = td->io_ops_data; | |
531 | struct xnvme_fioe_fwrap *fwrap = NULL; | |
532 | int nfiles = xd->nallocated; | |
533 | int err = 0; | |
534 | ||
535 | if (xd->prev != -1 && ++xd->prev < nfiles) { | |
536 | fwrap = &xd->files[xd->prev]; | |
537 | xd->cur = xd->prev; | |
538 | } | |
539 | ||
540 | xd->completed = 0; | |
541 | for (;;) { | |
542 | if (fwrap == NULL || xd->cur == nfiles) { | |
543 | fwrap = &xd->files[0]; | |
544 | xd->cur = 0; | |
545 | } | |
546 | ||
547 | while (fwrap != NULL && xd->cur < nfiles && err >= 0) { | |
548 | err = xnvme_queue_poke(fwrap->queue, max - xd->completed); | |
549 | if (err < 0) { | |
550 | switch (err) { | |
551 | case -EBUSY: | |
552 | case -EAGAIN: | |
553 | usleep(1); | |
554 | break; | |
555 | ||
556 | default: | |
557 | log_err("ioeng->getevents(): unhandled IO error\n"); | |
558 | assert(false); | |
559 | return 0; | |
560 | } | |
561 | } | |
562 | if (xd->completed >= min) { | |
563 | xd->prev = xd->cur; | |
564 | return xd->completed; | |
565 | } | |
566 | xd->cur++; | |
567 | fwrap = &xd->files[xd->cur]; | |
568 | ||
569 | if (err < 0) { | |
570 | switch (err) { | |
571 | case -EBUSY: | |
572 | case -EAGAIN: | |
573 | usleep(1); | |
574 | break; | |
575 | } | |
576 | } | |
577 | } | |
578 | } | |
579 | ||
580 | xd->cur = 0; | |
581 | ||
582 | return xd->completed; | |
583 | } | |
584 | ||
585 | static enum fio_q_status xnvme_fioe_queue(struct thread_data *td, struct io_u *io_u) | |
586 | { | |
587 | struct xnvme_fioe_data *xd = td->io_ops_data; | |
588 | struct xnvme_fioe_fwrap *fwrap; | |
589 | struct xnvme_cmd_ctx *ctx; | |
be5514e3 | 590 | struct xnvme_fioe_request *fio_req = io_u->engine_data; |
a3ff873e AK |
591 | uint32_t nsid; |
592 | uint64_t slba; | |
593 | uint16_t nlb; | |
594 | int err; | |
595 | bool vectored_io = ((struct xnvme_fioe_options *)td->eo)->xnvme_iovec; | |
e5f3b613 | 596 | uint32_t dir = io_u->dtype; |
a3ff873e AK |
597 | |
598 | fio_ro_check(td, io_u); | |
599 | ||
600 | fwrap = &xd->files[io_u->file->fileno]; | |
601 | nsid = xnvme_dev_get_nsid(fwrap->dev); | |
602 | ||
be5514e3 AK |
603 | if (fwrap->lba_pow2) { |
604 | slba = io_u->offset >> fwrap->ssw; | |
605 | nlb = (io_u->xfer_buflen >> fwrap->ssw) - 1; | |
606 | } else { | |
607 | slba = io_u->offset / fwrap->lba_nbytes; | |
608 | nlb = (io_u->xfer_buflen / fwrap->lba_nbytes) - 1; | |
609 | } | |
a3ff873e AK |
610 | |
611 | ctx = xnvme_queue_get_cmd_ctx(fwrap->queue); | |
612 | ctx->async.cb_arg = io_u; | |
613 | ||
614 | ctx->cmd.common.nsid = nsid; | |
615 | ctx->cmd.nvm.slba = slba; | |
616 | ctx->cmd.nvm.nlb = nlb; | |
e5f3b613 AK |
617 | if (dir) { |
618 | ctx->cmd.nvm.dtype = io_u->dtype; | |
619 | ctx->cmd.nvm.cdw13.dspec = io_u->dspec; | |
620 | } | |
a3ff873e AK |
621 | |
622 | switch (io_u->ddir) { | |
623 | case DDIR_READ: | |
624 | ctx->cmd.common.opcode = XNVME_SPEC_NVM_OPC_READ; | |
625 | break; | |
626 | ||
627 | case DDIR_WRITE: | |
628 | ctx->cmd.common.opcode = XNVME_SPEC_NVM_OPC_WRITE; | |
629 | break; | |
630 | ||
631 | default: | |
632 | log_err("ioeng->queue(): ENOSYS: %u\n", io_u->ddir); | |
eb3570b5 AK |
633 | xnvme_queue_put_cmd_ctx(ctx->async.queue, ctx); |
634 | ||
635 | io_u->error = ENOSYS; | |
a3ff873e | 636 | assert(false); |
eb3570b5 | 637 | return FIO_Q_COMPLETED; |
a3ff873e AK |
638 | } |
639 | ||
640 | if (vectored_io) { | |
641 | xd->iovec[io_u->index].iov_base = io_u->xfer_buf; | |
642 | xd->iovec[io_u->index].iov_len = io_u->xfer_buflen; | |
be5514e3 AK |
643 | if (fwrap->md_nbytes && fwrap->lba_pow2) { |
644 | xd->md_iovec[io_u->index].iov_base = fio_req->md_buf; | |
645 | xd->md_iovec[io_u->index].iov_len = fwrap->md_nbytes * (nlb + 1); | |
646 | err = xnvme_cmd_passv(ctx, &xd->iovec[io_u->index], 1, io_u->xfer_buflen, | |
647 | &xd->md_iovec[io_u->index], 1, | |
648 | fwrap->md_nbytes * (nlb + 1)); | |
649 | } else { | |
650 | err = xnvme_cmd_passv(ctx, &xd->iovec[io_u->index], 1, io_u->xfer_buflen, | |
651 | NULL, 0, 0); | |
652 | } | |
a3ff873e | 653 | } else { |
be5514e3 AK |
654 | if (fwrap->md_nbytes && fwrap->lba_pow2) |
655 | err = xnvme_cmd_pass(ctx, io_u->xfer_buf, io_u->xfer_buflen, | |
656 | fio_req->md_buf, fwrap->md_nbytes * (nlb + 1)); | |
657 | else | |
658 | err = xnvme_cmd_pass(ctx, io_u->xfer_buf, io_u->xfer_buflen, NULL, 0); | |
a3ff873e AK |
659 | } |
660 | switch (err) { | |
661 | case 0: | |
662 | return FIO_Q_QUEUED; | |
663 | ||
664 | case -EBUSY: | |
665 | case -EAGAIN: | |
666 | xnvme_queue_put_cmd_ctx(ctx->async.queue, ctx); | |
667 | return FIO_Q_BUSY; | |
668 | ||
669 | default: | |
670 | log_err("ioeng->queue(): err: '%d'\n", err); | |
671 | ||
672 | xnvme_queue_put_cmd_ctx(ctx->async.queue, ctx); | |
673 | ||
674 | io_u->error = abs(err); | |
675 | assert(false); | |
676 | return FIO_Q_COMPLETED; | |
677 | } | |
678 | } | |
679 | ||
680 | static int xnvme_fioe_close(struct thread_data *td, struct fio_file *f) | |
681 | { | |
682 | struct xnvme_fioe_data *xd = td->io_ops_data; | |
683 | ||
684 | dprint(FD_FILE, "xnvme close %s -- nopen: %ld\n", f->file_name, xd->nopen); | |
685 | ||
686 | --(xd->nopen); | |
687 | ||
688 | return 0; | |
689 | } | |
690 | ||
691 | static int xnvme_fioe_open(struct thread_data *td, struct fio_file *f) | |
692 | { | |
693 | struct xnvme_fioe_data *xd = td->io_ops_data; | |
694 | ||
695 | dprint(FD_FILE, "xnvme open %s -- nopen: %ld\n", f->file_name, xd->nopen); | |
696 | ||
697 | if (f->fileno > (int)xd->nallocated) { | |
698 | log_err("ioeng->open(): f->fileno > xd->nallocated; invalid assumption\n"); | |
699 | return 1; | |
700 | } | |
701 | if (xd->files[f->fileno].fio_file != f) { | |
702 | log_err("ioeng->open(): fio_file != f; invalid assumption\n"); | |
703 | return 1; | |
704 | } | |
705 | ||
706 | ++(xd->nopen); | |
707 | ||
708 | return 0; | |
709 | } | |
710 | ||
711 | static int xnvme_fioe_invalidate(struct thread_data *td, struct fio_file *f) | |
712 | { | |
713 | /* Consider only doing this with be:spdk */ | |
714 | return 0; | |
715 | } | |
716 | ||
717 | static int xnvme_fioe_get_max_open_zones(struct thread_data *td, struct fio_file *f, | |
718 | unsigned int *max_open_zones) | |
719 | { | |
720 | struct xnvme_opts opts = xnvme_opts_from_fioe(td); | |
721 | struct xnvme_dev *dev; | |
722 | const struct xnvme_spec_znd_idfy_ns *zns; | |
723 | int err = 0, err_lock; | |
724 | ||
725 | if (f->filetype != FIO_TYPE_FILE && f->filetype != FIO_TYPE_BLOCK && | |
726 | f->filetype != FIO_TYPE_CHAR) { | |
727 | log_info("ioeng->get_max_open_zoned(): ignoring filetype: %d\n", f->filetype); | |
728 | return 0; | |
729 | } | |
730 | err_lock = pthread_mutex_lock(&g_serialize); | |
731 | if (err_lock) { | |
732 | log_err("ioeng->get_max_open_zones(): pthread_mutex_lock(), err(%d)\n", err_lock); | |
733 | return -err_lock; | |
734 | } | |
735 | ||
736 | dev = xnvme_dev_open(f->file_name, &opts); | |
737 | if (!dev) { | |
738 | log_err("ioeng->get_max_open_zones(): xnvme_dev_open(), err(%d)\n", err_lock); | |
739 | err = -errno; | |
740 | goto exit; | |
741 | } | |
742 | if (xnvme_dev_get_geo(dev)->type != XNVME_GEO_ZONED) { | |
743 | errno = EINVAL; | |
744 | err = -errno; | |
745 | goto exit; | |
746 | } | |
747 | ||
748 | zns = (void *)xnvme_dev_get_ns_css(dev); | |
749 | if (!zns) { | |
750 | log_err("ioeng->get_max_open_zones(): xnvme_dev_get_ns_css(), err(%d)\n", errno); | |
751 | err = -errno; | |
752 | goto exit; | |
753 | } | |
754 | ||
755 | /* | |
756 | * intentional overflow as the value is zero-based and NVMe | |
757 | * defines 0xFFFFFFFF as unlimited thus overflowing to 0 which | |
758 | * is how fio indicates unlimited and otherwise just converting | |
759 | * to one-based. | |
760 | */ | |
761 | *max_open_zones = zns->mor + 1; | |
762 | ||
763 | exit: | |
764 | xnvme_dev_close(dev); | |
765 | err_lock = pthread_mutex_unlock(&g_serialize); | |
766 | if (err_lock) | |
767 | log_err("ioeng->get_max_open_zones(): pthread_mutex_unlock(), err(%d)\n", | |
768 | err_lock); | |
769 | ||
770 | return err; | |
771 | } | |
772 | ||
773 | /** | |
774 | * Currently, this function is called before of I/O engine initialization, so, | |
775 | * we cannot consult the file-wrapping done when 'fioe' initializes. | |
776 | * Instead we just open based on the given filename. | |
777 | * | |
778 | * TODO: unify the different setup methods, consider keeping the handle around, | |
779 | * and consider how to support the --be option in this usecase | |
780 | */ | |
781 | static int xnvme_fioe_get_zoned_model(struct thread_data *td, struct fio_file *f, | |
782 | enum zbd_zoned_model *model) | |
783 | { | |
784 | struct xnvme_opts opts = xnvme_opts_from_fioe(td); | |
785 | struct xnvme_dev *dev; | |
786 | int err = 0, err_lock; | |
787 | ||
788 | if (f->filetype != FIO_TYPE_FILE && f->filetype != FIO_TYPE_BLOCK && | |
789 | f->filetype != FIO_TYPE_CHAR) { | |
790 | log_info("ioeng->get_zoned_model(): ignoring filetype: %d\n", f->filetype); | |
791 | return -EINVAL; | |
792 | } | |
793 | ||
794 | err = pthread_mutex_lock(&g_serialize); | |
795 | if (err) { | |
796 | log_err("ioeng->get_zoned_model(): pthread_mutex_lock(), err(%d)\n", err); | |
797 | return -err; | |
798 | } | |
799 | ||
800 | dev = xnvme_dev_open(f->file_name, &opts); | |
801 | if (!dev) { | |
802 | log_err("ioeng->get_zoned_model(): xnvme_dev_open(%s) failed, errno: %d\n", | |
803 | f->file_name, errno); | |
804 | err = -errno; | |
805 | goto exit; | |
806 | } | |
807 | ||
808 | switch (xnvme_dev_get_geo(dev)->type) { | |
809 | case XNVME_GEO_UNKNOWN: | |
810 | dprint(FD_ZBD, "%s: got 'unknown', assigning ZBD_NONE\n", f->file_name); | |
811 | *model = ZBD_NONE; | |
812 | break; | |
813 | ||
814 | case XNVME_GEO_CONVENTIONAL: | |
815 | dprint(FD_ZBD, "%s: got 'conventional', assigning ZBD_NONE\n", f->file_name); | |
816 | *model = ZBD_NONE; | |
817 | break; | |
818 | ||
819 | case XNVME_GEO_ZONED: | |
820 | dprint(FD_ZBD, "%s: got 'zoned', assigning ZBD_HOST_MANAGED\n", f->file_name); | |
821 | *model = ZBD_HOST_MANAGED; | |
822 | break; | |
823 | ||
824 | default: | |
825 | dprint(FD_ZBD, "%s: hit-default, assigning ZBD_NONE\n", f->file_name); | |
826 | *model = ZBD_NONE; | |
827 | errno = EINVAL; | |
828 | err = -errno; | |
829 | break; | |
830 | } | |
831 | ||
832 | exit: | |
833 | xnvme_dev_close(dev); | |
834 | ||
835 | err_lock = pthread_mutex_unlock(&g_serialize); | |
836 | if (err_lock) | |
837 | log_err("ioeng->get_zoned_model(): pthread_mutex_unlock(), err(%d)\n", err_lock); | |
838 | ||
839 | return err; | |
840 | } | |
841 | ||
842 | /** | |
843 | * Fills the given ``zbdz`` with at most ``nr_zones`` zone-descriptors. | |
844 | * | |
845 | * The implementation converts the NVMe Zoned Command Set log-pages for Zone | |
846 | * descriptors into the Linux Kernel Zoned Block Report format. | |
847 | * | |
848 | * NOTE: This function is called before I/O engine initialization, that is, | |
849 | * before ``_dev_open`` has been called and file-wrapping is setup. Thus is has | |
850 | * to do the ``_dev_open`` itself, and shut it down again once it is done | |
851 | * retrieving the log-pages and converting them to the report format. | |
852 | * | |
853 | * TODO: unify the different setup methods, consider keeping the handle around, | |
854 | * and consider how to support the --async option in this usecase | |
855 | */ | |
856 | static int xnvme_fioe_report_zones(struct thread_data *td, struct fio_file *f, uint64_t offset, | |
857 | struct zbd_zone *zbdz, unsigned int nr_zones) | |
858 | { | |
859 | struct xnvme_opts opts = xnvme_opts_from_fioe(td); | |
860 | const struct xnvme_spec_znd_idfy_lbafe *lbafe = NULL; | |
861 | struct xnvme_dev *dev = NULL; | |
862 | const struct xnvme_geo *geo = NULL; | |
863 | struct xnvme_znd_report *rprt = NULL; | |
864 | uint32_t ssw; | |
865 | uint64_t slba; | |
866 | unsigned int limit = 0; | |
867 | int err = 0, err_lock; | |
868 | ||
869 | dprint(FD_ZBD, "%s: report_zones() offset: %zu, nr_zones: %u\n", f->file_name, offset, | |
870 | nr_zones); | |
871 | ||
872 | err = pthread_mutex_lock(&g_serialize); | |
873 | if (err) { | |
874 | log_err("ioeng->report_zones(%s): pthread_mutex_lock(), err(%d)\n", f->file_name, | |
875 | err); | |
876 | return -err; | |
877 | } | |
878 | ||
879 | dev = xnvme_dev_open(f->file_name, &opts); | |
880 | if (!dev) { | |
881 | log_err("ioeng->report_zones(%s): xnvme_dev_open(), err(%d)\n", f->file_name, | |
882 | errno); | |
883 | goto exit; | |
884 | } | |
885 | ||
886 | geo = xnvme_dev_get_geo(dev); | |
887 | ssw = xnvme_dev_get_ssw(dev); | |
888 | lbafe = xnvme_znd_dev_get_lbafe(dev); | |
889 | ||
890 | limit = nr_zones > geo->nzone ? geo->nzone : nr_zones; | |
891 | ||
892 | dprint(FD_ZBD, "%s: limit: %u\n", f->file_name, limit); | |
893 | ||
894 | slba = ((offset >> ssw) / geo->nsect) * geo->nsect; | |
895 | ||
896 | rprt = xnvme_znd_report_from_dev(dev, slba, limit, 0); | |
897 | if (!rprt) { | |
898 | log_err("ioeng->report_zones(%s): xnvme_znd_report_from_dev(), err(%d)\n", | |
899 | f->file_name, errno); | |
900 | err = -errno; | |
901 | goto exit; | |
902 | } | |
903 | if (rprt->nentries != limit) { | |
904 | log_err("ioeng->report_zones(%s): nentries != nr_zones\n", f->file_name); | |
905 | err = 1; | |
906 | goto exit; | |
907 | } | |
908 | if (offset > geo->tbytes) { | |
909 | log_err("ioeng->report_zones(%s): out-of-bounds\n", f->file_name); | |
910 | goto exit; | |
911 | } | |
912 | ||
913 | /* Transform the zone-report */ | |
914 | for (uint32_t idx = 0; idx < rprt->nentries; ++idx) { | |
915 | struct xnvme_spec_znd_descr *descr = XNVME_ZND_REPORT_DESCR(rprt, idx); | |
916 | ||
917 | zbdz[idx].start = descr->zslba << ssw; | |
918 | zbdz[idx].len = lbafe->zsze << ssw; | |
919 | zbdz[idx].capacity = descr->zcap << ssw; | |
920 | zbdz[idx].wp = descr->wp << ssw; | |
921 | ||
922 | switch (descr->zt) { | |
923 | case XNVME_SPEC_ZND_TYPE_SEQWR: | |
924 | zbdz[idx].type = ZBD_ZONE_TYPE_SWR; | |
925 | break; | |
926 | ||
927 | default: | |
928 | log_err("ioeng->report_zones(%s): invalid type for zone at offset(%zu)\n", | |
929 | f->file_name, zbdz[idx].start); | |
930 | err = -EIO; | |
931 | goto exit; | |
932 | } | |
933 | ||
934 | switch (descr->zs) { | |
935 | case XNVME_SPEC_ZND_STATE_EMPTY: | |
936 | zbdz[idx].cond = ZBD_ZONE_COND_EMPTY; | |
937 | break; | |
938 | case XNVME_SPEC_ZND_STATE_IOPEN: | |
939 | zbdz[idx].cond = ZBD_ZONE_COND_IMP_OPEN; | |
940 | break; | |
941 | case XNVME_SPEC_ZND_STATE_EOPEN: | |
942 | zbdz[idx].cond = ZBD_ZONE_COND_EXP_OPEN; | |
943 | break; | |
944 | case XNVME_SPEC_ZND_STATE_CLOSED: | |
945 | zbdz[idx].cond = ZBD_ZONE_COND_CLOSED; | |
946 | break; | |
947 | case XNVME_SPEC_ZND_STATE_FULL: | |
948 | zbdz[idx].cond = ZBD_ZONE_COND_FULL; | |
949 | break; | |
950 | ||
951 | case XNVME_SPEC_ZND_STATE_RONLY: | |
952 | case XNVME_SPEC_ZND_STATE_OFFLINE: | |
953 | default: | |
954 | zbdz[idx].cond = ZBD_ZONE_COND_OFFLINE; | |
955 | break; | |
956 | } | |
957 | } | |
958 | ||
959 | exit: | |
960 | xnvme_buf_virt_free(rprt); | |
961 | ||
962 | xnvme_dev_close(dev); | |
963 | ||
964 | err_lock = pthread_mutex_unlock(&g_serialize); | |
965 | if (err_lock) | |
966 | log_err("ioeng->report_zones(): pthread_mutex_unlock(), err: %d\n", err_lock); | |
967 | ||
968 | dprint(FD_ZBD, "err: %d, nr_zones: %d\n", err, (int)nr_zones); | |
969 | ||
970 | return err ? err : (int)limit; | |
971 | } | |
972 | ||
973 | /** | |
974 | * NOTE: This function may get called before I/O engine initialization, that is, | |
975 | * before ``_dev_open`` has been called and file-wrapping is setup. In such | |
976 | * case it has to do ``_dev_open`` itself, and shut it down again once it is | |
977 | * done resetting write pointer of zones. | |
978 | */ | |
979 | static int xnvme_fioe_reset_wp(struct thread_data *td, struct fio_file *f, uint64_t offset, | |
980 | uint64_t length) | |
981 | { | |
982 | struct xnvme_opts opts = xnvme_opts_from_fioe(td); | |
983 | struct xnvme_fioe_data *xd = NULL; | |
984 | struct xnvme_fioe_fwrap *fwrap = NULL; | |
985 | struct xnvme_dev *dev = NULL; | |
986 | const struct xnvme_geo *geo = NULL; | |
987 | uint64_t first, last; | |
988 | uint32_t ssw; | |
989 | uint32_t nsid; | |
990 | int err = 0, err_lock; | |
991 | ||
992 | if (td->io_ops_data) { | |
993 | xd = td->io_ops_data; | |
994 | fwrap = &xd->files[f->fileno]; | |
995 | ||
996 | assert(fwrap->dev); | |
997 | assert(fwrap->geo); | |
998 | ||
999 | dev = fwrap->dev; | |
1000 | geo = fwrap->geo; | |
1001 | ssw = fwrap->ssw; | |
1002 | } else { | |
1003 | err = pthread_mutex_lock(&g_serialize); | |
1004 | if (err) { | |
1005 | log_err("ioeng->reset_wp(): pthread_mutex_lock(), err(%d)\n", err); | |
1006 | return -err; | |
1007 | } | |
1008 | ||
1009 | dev = xnvme_dev_open(f->file_name, &opts); | |
1010 | if (!dev) { | |
1011 | log_err("ioeng->reset_wp(): xnvme_dev_open(%s) failed, errno(%d)\n", | |
1012 | f->file_name, errno); | |
1013 | goto exit; | |
1014 | } | |
1015 | geo = xnvme_dev_get_geo(dev); | |
1016 | ssw = xnvme_dev_get_ssw(dev); | |
1017 | } | |
1018 | ||
1019 | nsid = xnvme_dev_get_nsid(dev); | |
1020 | ||
1021 | first = ((offset >> ssw) / geo->nsect) * geo->nsect; | |
1022 | last = (((offset + length) >> ssw) / geo->nsect) * geo->nsect; | |
1023 | dprint(FD_ZBD, "first: 0x%lx, last: 0x%lx\n", first, last); | |
1024 | ||
1025 | for (uint64_t zslba = first; zslba < last; zslba += geo->nsect) { | |
1026 | struct xnvme_cmd_ctx ctx = xnvme_cmd_ctx_from_dev(dev); | |
1027 | ||
1028 | if (zslba >= (geo->nsect * geo->nzone)) { | |
1029 | log_err("ioeng->reset_wp(): out-of-bounds\n"); | |
1030 | err = 0; | |
1031 | break; | |
1032 | } | |
1033 | ||
1034 | err = xnvme_znd_mgmt_send(&ctx, nsid, zslba, false, | |
1035 | XNVME_SPEC_ZND_CMD_MGMT_SEND_RESET, 0x0, NULL); | |
1036 | if (err || xnvme_cmd_ctx_cpl_status(&ctx)) { | |
1037 | err = err ? err : -EIO; | |
1038 | log_err("ioeng->reset_wp(): err(%d), sc(%d)", err, ctx.cpl.status.sc); | |
1039 | goto exit; | |
1040 | } | |
1041 | } | |
1042 | ||
1043 | exit: | |
1044 | if (!td->io_ops_data) { | |
1045 | xnvme_dev_close(dev); | |
1046 | ||
1047 | err_lock = pthread_mutex_unlock(&g_serialize); | |
1048 | if (err_lock) | |
1049 | log_err("ioeng->reset_wp(): pthread_mutex_unlock(), err(%d)\n", err_lock); | |
1050 | } | |
1051 | ||
1052 | return err; | |
1053 | } | |
1054 | ||
e5f3b613 AK |
1055 | static int xnvme_fioe_fetch_ruhs(struct thread_data *td, struct fio_file *f, |
1056 | struct fio_ruhs_info *fruhs_info) | |
1057 | { | |
1058 | struct xnvme_opts opts = xnvme_opts_from_fioe(td); | |
1059 | struct xnvme_dev *dev; | |
1060 | struct xnvme_spec_ruhs *ruhs; | |
1061 | struct xnvme_cmd_ctx ctx; | |
1062 | uint32_t ruhs_nbytes; | |
1063 | uint32_t nsid; | |
1064 | int err = 0, err_lock; | |
1065 | ||
48cf0c63 | 1066 | if (f->filetype != FIO_TYPE_CHAR && f->filetype != FIO_TYPE_FILE) { |
e5f3b613 AK |
1067 | log_err("ioeng->fdp_ruhs(): ignoring filetype: %d\n", f->filetype); |
1068 | return -EINVAL; | |
1069 | } | |
1070 | ||
1071 | err = pthread_mutex_lock(&g_serialize); | |
1072 | if (err) { | |
1073 | log_err("ioeng->fdp_ruhs(): pthread_mutex_lock(), err(%d)\n", err); | |
1074 | return -err; | |
1075 | } | |
1076 | ||
1077 | dev = xnvme_dev_open(f->file_name, &opts); | |
1078 | if (!dev) { | |
1079 | log_err("ioeng->fdp_ruhs(): xnvme_dev_open(%s) failed, errno: %d\n", | |
1080 | f->file_name, errno); | |
1081 | err = -errno; | |
1082 | goto exit; | |
1083 | } | |
1084 | ||
1085 | ruhs_nbytes = sizeof(*ruhs) + (FDP_MAX_RUHS * sizeof(struct xnvme_spec_ruhs_desc)); | |
1086 | ruhs = xnvme_buf_alloc(dev, ruhs_nbytes); | |
1087 | if (!ruhs) { | |
1088 | err = -errno; | |
1089 | goto exit; | |
1090 | } | |
1091 | memset(ruhs, 0, ruhs_nbytes); | |
1092 | ||
1093 | ctx = xnvme_cmd_ctx_from_dev(dev); | |
1094 | nsid = xnvme_dev_get_nsid(dev); | |
1095 | ||
1096 | err = xnvme_nvm_mgmt_recv(&ctx, nsid, XNVME_SPEC_IO_MGMT_RECV_RUHS, 0, ruhs, ruhs_nbytes); | |
1097 | ||
1098 | if (err || xnvme_cmd_ctx_cpl_status(&ctx)) { | |
1099 | err = err ? err : -EIO; | |
1100 | log_err("ioeng->fdp_ruhs(): err(%d), sc(%d)", err, ctx.cpl.status.sc); | |
1101 | goto free_buffer; | |
1102 | } | |
1103 | ||
1104 | fruhs_info->nr_ruhs = ruhs->nruhsd; | |
1105 | for (uint32_t idx = 0; idx < fruhs_info->nr_ruhs; ++idx) { | |
1106 | fruhs_info->plis[idx] = le16_to_cpu(ruhs->desc[idx].pi); | |
1107 | } | |
1108 | ||
1109 | free_buffer: | |
1110 | xnvme_buf_free(dev, ruhs); | |
1111 | exit: | |
1112 | xnvme_dev_close(dev); | |
1113 | ||
1114 | err_lock = pthread_mutex_unlock(&g_serialize); | |
1115 | if (err_lock) | |
1116 | log_err("ioeng->fdp_ruhs(): pthread_mutex_unlock(), err(%d)\n", err_lock); | |
1117 | ||
1118 | return err; | |
1119 | } | |
1120 | ||
a3ff873e AK |
1121 | static int xnvme_fioe_get_file_size(struct thread_data *td, struct fio_file *f) |
1122 | { | |
1123 | struct xnvme_opts opts = xnvme_opts_from_fioe(td); | |
1124 | struct xnvme_dev *dev; | |
1125 | int ret = 0, err; | |
1126 | ||
1127 | if (fio_file_size_known(f)) | |
1128 | return 0; | |
1129 | ||
1130 | ret = pthread_mutex_lock(&g_serialize); | |
1131 | if (ret) { | |
1132 | log_err("ioeng->reset_wp(): pthread_mutex_lock(), err(%d)\n", ret); | |
1133 | return -ret; | |
1134 | } | |
1135 | ||
1136 | dev = xnvme_dev_open(f->file_name, &opts); | |
1137 | if (!dev) { | |
1138 | log_err("%s: failed retrieving device handle, errno: %d\n", f->file_name, errno); | |
1139 | ret = -errno; | |
1140 | goto exit; | |
1141 | } | |
1142 | ||
1143 | f->real_file_size = xnvme_dev_get_geo(dev)->tbytes; | |
1144 | fio_file_set_size_known(f); | |
e5f3b613 AK |
1145 | |
1146 | if (td->o.zone_mode == ZONE_MODE_ZBD) | |
1147 | f->filetype = FIO_TYPE_BLOCK; | |
a3ff873e AK |
1148 | |
1149 | exit: | |
1150 | xnvme_dev_close(dev); | |
1151 | err = pthread_mutex_unlock(&g_serialize); | |
1152 | if (err) | |
1153 | log_err("ioeng->reset_wp(): pthread_mutex_unlock(), err(%d)\n", err); | |
1154 | ||
1155 | return ret; | |
1156 | } | |
1157 | ||
1158 | FIO_STATIC struct ioengine_ops ioengine = { | |
1159 | .name = "xnvme", | |
1160 | .version = FIO_IOOPS_VERSION, | |
1161 | .options = options, | |
1162 | .option_struct_size = sizeof(struct xnvme_fioe_options), | |
1163 | .flags = FIO_DISKLESSIO | FIO_NODISKUTIL | FIO_NOEXTEND | FIO_MEMALIGN | FIO_RAWIO, | |
1164 | ||
1165 | .cleanup = xnvme_fioe_cleanup, | |
1166 | .init = xnvme_fioe_init, | |
1167 | ||
1168 | .iomem_free = xnvme_fioe_iomem_free, | |
1169 | .iomem_alloc = xnvme_fioe_iomem_alloc, | |
1170 | ||
1171 | .io_u_free = xnvme_fioe_io_u_free, | |
1172 | .io_u_init = xnvme_fioe_io_u_init, | |
1173 | ||
1174 | .event = xnvme_fioe_event, | |
1175 | .getevents = xnvme_fioe_getevents, | |
1176 | .queue = xnvme_fioe_queue, | |
1177 | ||
1178 | .close_file = xnvme_fioe_close, | |
1179 | .open_file = xnvme_fioe_open, | |
1180 | .get_file_size = xnvme_fioe_get_file_size, | |
1181 | ||
1182 | .invalidate = xnvme_fioe_invalidate, | |
1183 | .get_max_open_zones = xnvme_fioe_get_max_open_zones, | |
1184 | .get_zoned_model = xnvme_fioe_get_zoned_model, | |
1185 | .report_zones = xnvme_fioe_report_zones, | |
1186 | .reset_wp = xnvme_fioe_reset_wp, | |
e5f3b613 AK |
1187 | |
1188 | .fdp_fetch_ruhs = xnvme_fioe_fetch_ruhs, | |
a3ff873e AK |
1189 | }; |
1190 | ||
1191 | static void fio_init fio_xnvme_register(void) | |
1192 | { | |
1193 | register_ioengine(&ioengine); | |
1194 | } | |
1195 | ||
1196 | static void fio_exit fio_xnvme_unregister(void) | |
1197 | { | |
1198 | unregister_ioengine(&ioengine); | |
1199 | } |