projects / fio.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
engines/io_uring: add multi range dsm support
[fio.git] / engines / nvme.c
CommitLineData
ca59c41c 1// SPDX-License-Identifier: GPL-2.0
b3d5e3fd
AK		 2/*
3 * nvme structure declarations and helper functions for the
4 * io_uring_cmd engine.
5 */
6
7#include "nvme.h"
5163f35e 8#include "../crc/crc-t10dif.h"
08371767 9#include "../crc/crc64.h"
b3d5e3fd 10
0d610785 11static inline __u64 get_slba(struct nvme_data *data, __u64 offset)
4885a6eb
VF		 12{
13 if (data->lba_ext)
0d610785
AK		 14 return offset / data->lba_ext;
15
16 return offset >> data->lba_shift;
4885a6eb
VF		 17}
18
0d610785 19static inline __u32 get_nlb(struct nvme_data *data, __u64 len)
4885a6eb
VF		 20{
21 if (data->lba_ext)
0d610785
AK		 22 return len / data->lba_ext - 1;
23
24 return (len >> data->lba_shift) - 1;
4885a6eb
VF		 25}
26
5163f35e
AK		 27static void fio_nvme_generate_pi_16b_guard(struct nvme_data *data,
28 struct io_u *io_u,
29 struct nvme_cmd_ext_io_opts *opts)
30{
31 struct nvme_pi_data *pi_data = io_u->engine_data;
32 struct nvme_16b_guard_pif *pi;
33 unsigned char *buf = io_u->xfer_buf;
34 unsigned char *md_buf = io_u->mmap_data;
0d610785
AK		 35 __u64 slba = get_slba(data, io_u->offset);
36 __u32 nlb = get_nlb(data, io_u->xfer_buflen) + 1;
5163f35e
AK		 37 __u32 lba_num = 0;
38 __u16 guard = 0;
39
40 if (data->pi_loc) {
41 if (data->lba_ext)
42 pi_data->interval = data->lba_ext - data->ms;
43 else
44 pi_data->interval = 0;
45 } else {
46 if (data->lba_ext)
47 pi_data->interval = data->lba_ext - sizeof(struct nvme_16b_guard_pif);
48 else
49 pi_data->interval = data->ms - sizeof(struct nvme_16b_guard_pif);
50 }
51
52 if (io_u->ddir != DDIR_WRITE)
53 return;
54
55 while (lba_num < nlb) {
56 if (data->lba_ext)
57 pi = (struct nvme_16b_guard_pif *)(buf + pi_data->interval);
58 else
59 pi = (struct nvme_16b_guard_pif *)(md_buf + pi_data->interval);
60
61 if (opts->io_flags & NVME_IO_PRINFO_PRCHK_GUARD) {
62 if (data->lba_ext) {
63 guard = fio_crc_t10dif(0, buf, pi_data->interval);
64 } else {
65 guard = fio_crc_t10dif(0, buf, data->lba_size);
66 guard = fio_crc_t10dif(guard, md_buf, pi_data->interval);
67 }
68 pi->guard = cpu_to_be16(guard);
69 }
70
71 if (opts->io_flags & NVME_IO_PRINFO_PRCHK_APP)
72 pi->apptag = cpu_to_be16(pi_data->apptag);
73
74 if (opts->io_flags & NVME_IO_PRINFO_PRCHK_REF) {
75 switch (data->pi_type) {
76 case NVME_NS_DPS_PI_TYPE1:
77 case NVME_NS_DPS_PI_TYPE2:
78 pi->srtag = cpu_to_be32((__u32)slba + lba_num);
79 break;
80 case NVME_NS_DPS_PI_TYPE3:
81 break;
82 }
83 }
84 if (data->lba_ext) {
85 buf += data->lba_ext;
86 } else {
87 buf += data->lba_size;
88 md_buf += data->ms;
89 }
90 lba_num++;
91 }
92}
93
94static int fio_nvme_verify_pi_16b_guard(struct nvme_data *data,
95 struct io_u *io_u)
96{
97 struct nvme_pi_data *pi_data = io_u->engine_data;
98 struct nvme_16b_guard_pif *pi;
99 struct fio_file *f = io_u->file;
100 unsigned char *buf = io_u->xfer_buf;
101 unsigned char *md_buf = io_u->mmap_data;
0d610785
AK		 102 __u64 slba = get_slba(data, io_u->offset);
103 __u32 nlb = get_nlb(data, io_u->xfer_buflen) + 1;
5163f35e
AK		 104 __u32 lba_num = 0;
105 __u16 unmask_app, unmask_app_exp, guard = 0;
106
107 while (lba_num < nlb) {
108 if (data->lba_ext)
109 pi = (struct nvme_16b_guard_pif *)(buf + pi_data->interval);
110 else
111 pi = (struct nvme_16b_guard_pif *)(md_buf + pi_data->interval);
112
113 if (data->pi_type == NVME_NS_DPS_PI_TYPE3) {
114 if (pi->apptag == NVME_PI_APP_DISABLE &&
115 pi->srtag == NVME_PI_REF_DISABLE)
116 goto next;
117 } else if (data->pi_type == NVME_NS_DPS_PI_TYPE1 ||
118 data->pi_type == NVME_NS_DPS_PI_TYPE2) {
119 if (pi->apptag == NVME_PI_APP_DISABLE)
120 goto next;
121 }
122
123 if (pi_data->io_flags & NVME_IO_PRINFO_PRCHK_GUARD) {
124 if (data->lba_ext) {
125 guard = fio_crc_t10dif(0, buf, pi_data->interval);
126 } else {
127 guard = fio_crc_t10dif(0, buf, data->lba_size);
128 guard = fio_crc_t10dif(guard, md_buf, pi_data->interval);
129 }
130 if (be16_to_cpu(pi->guard) != guard) {
131 log_err("%s: Guard compare error: LBA: %llu Expected=%x, Actual=%x\n",
132 f->file_name, (unsigned long long)slba,
133 guard, be16_to_cpu(pi->guard));
134 return -EIO;
135 }
136 }
137
138 if (pi_data->io_flags & NVME_IO_PRINFO_PRCHK_APP) {
139 unmask_app = be16_to_cpu(pi->apptag) & pi_data->apptag_mask;
140 unmask_app_exp = pi_data->apptag & pi_data->apptag_mask;
141 if (unmask_app != unmask_app_exp) {
142 log_err("%s: APPTAG compare error: LBA: %llu Expected=%x, Actual=%x\n",
143 f->file_name, (unsigned long long)slba,
144 unmask_app_exp, unmask_app);
145 return -EIO;
146 }
147 }
148
149 if (pi_data->io_flags & NVME_IO_PRINFO_PRCHK_REF) {
150 switch (data->pi_type) {
151 case NVME_NS_DPS_PI_TYPE1:
152 case NVME_NS_DPS_PI_TYPE2:
153 if (be32_to_cpu(pi->srtag) !=
154 ((__u32)slba + lba_num)) {
155 log_err("%s: REFTAG compare error: LBA: %llu Expected=%x, Actual=%x\n",
156 f->file_name, (unsigned long long)slba,
157 (__u32)slba + lba_num,
158 be32_to_cpu(pi->srtag));
159 return -EIO;
160 }
161 break;
162 case NVME_NS_DPS_PI_TYPE3:
163 break;
164 }
165 }
166next:
167 if (data->lba_ext) {
168 buf += data->lba_ext;
169 } else {
170 buf += data->lba_size;
171 md_buf += data->ms;
172 }
173 lba_num++;
174 }
175
176 return 0;
177}
178
08371767
AK		 179static void fio_nvme_generate_pi_64b_guard(struct nvme_data *data,
180 struct io_u *io_u,
181 struct nvme_cmd_ext_io_opts *opts)
182{
183 struct nvme_pi_data *pi_data = io_u->engine_data;
184 struct nvme_64b_guard_pif *pi;
185 unsigned char *buf = io_u->xfer_buf;
186 unsigned char *md_buf = io_u->mmap_data;
187 uint64_t guard = 0;
0d610785
AK		 188 __u64 slba = get_slba(data, io_u->offset);
189 __u32 nlb = get_nlb(data, io_u->xfer_buflen) + 1;
08371767
AK		 190 __u32 lba_num = 0;
191
192 if (data->pi_loc) {
193 if (data->lba_ext)
194 pi_data->interval = data->lba_ext - data->ms;
195 else
196 pi_data->interval = 0;
197 } else {
198 if (data->lba_ext)
199 pi_data->interval = data->lba_ext - sizeof(struct nvme_64b_guard_pif);
200 else
201 pi_data->interval = data->ms - sizeof(struct nvme_64b_guard_pif);
202 }
203
204 if (io_u->ddir != DDIR_WRITE)
205 return;
206
207 while (lba_num < nlb) {
208 if (data->lba_ext)
209 pi = (struct nvme_64b_guard_pif *)(buf + pi_data->interval);
210 else
211 pi = (struct nvme_64b_guard_pif *)(md_buf + pi_data->interval);
212
213 if (opts->io_flags & NVME_IO_PRINFO_PRCHK_GUARD) {
214 if (data->lba_ext) {
215 guard = fio_crc64_nvme(0, buf, pi_data->interval);
216 } else {
217 guard = fio_crc64_nvme(0, buf, data->lba_size);
218 guard = fio_crc64_nvme(guard, md_buf, pi_data->interval);
219 }
220 pi->guard = cpu_to_be64(guard);
221 }
222
223 if (opts->io_flags & NVME_IO_PRINFO_PRCHK_APP)
224 pi->apptag = cpu_to_be16(pi_data->apptag);
225
226 if (opts->io_flags & NVME_IO_PRINFO_PRCHK_REF) {
227 switch (data->pi_type) {
228 case NVME_NS_DPS_PI_TYPE1:
229 case NVME_NS_DPS_PI_TYPE2:
230 put_unaligned_be48(slba + lba_num, pi->srtag);
231 break;
232 case NVME_NS_DPS_PI_TYPE3:
233 break;
234 }
235 }
236 if (data->lba_ext) {
237 buf += data->lba_ext;
238 } else {
239 buf += data->lba_size;
240 md_buf += data->ms;
241 }
242 lba_num++;
243 }
244}
245
246static int fio_nvme_verify_pi_64b_guard(struct nvme_data *data,
247 struct io_u *io_u)
248{
249 struct nvme_pi_data *pi_data = io_u->engine_data;
250 struct nvme_64b_guard_pif *pi;
251 struct fio_file *f = io_u->file;
252 unsigned char *buf = io_u->xfer_buf;
253 unsigned char *md_buf = io_u->mmap_data;
0d610785 254 __u64 slba = get_slba(data, io_u->offset);
08371767 255 __u64 ref, ref_exp, guard = 0;
0d610785 256 __u32 nlb = get_nlb(data, io_u->xfer_buflen) + 1;
08371767
AK		 257 __u32 lba_num = 0;
258 __u16 unmask_app, unmask_app_exp;
259
260 while (lba_num < nlb) {
261 if (data->lba_ext)
262 pi = (struct nvme_64b_guard_pif *)(buf + pi_data->interval);
263 else
264 pi = (struct nvme_64b_guard_pif *)(md_buf + pi_data->interval);
265
266 if (data->pi_type == NVME_NS_DPS_PI_TYPE3) {
267 if (pi->apptag == NVME_PI_APP_DISABLE &&
268 fio_nvme_pi_ref_escape(pi->srtag))
269 goto next;
270 } else if (data->pi_type == NVME_NS_DPS_PI_TYPE1 ||
271 data->pi_type == NVME_NS_DPS_PI_TYPE2) {
272 if (pi->apptag == NVME_PI_APP_DISABLE)
273 goto next;
274 }
275
276 if (pi_data->io_flags & NVME_IO_PRINFO_PRCHK_GUARD) {
277 if (data->lba_ext) {
278 guard = fio_crc64_nvme(0, buf, pi_data->interval);
279 } else {
280 guard = fio_crc64_nvme(0, buf, data->lba_size);
281 guard = fio_crc64_nvme(guard, md_buf, pi_data->interval);
282 }
283 if (be64_to_cpu((uint64_t)pi->guard) != guard) {
284 log_err("%s: Guard compare error: LBA: %llu Expected=%llx, Actual=%llx\n",
285 f->file_name, (unsigned long long)slba,
286 guard, be64_to_cpu((uint64_t)pi->guard));
287 return -EIO;
288 }
289 }
290
291 if (pi_data->io_flags & NVME_IO_PRINFO_PRCHK_APP) {
292 unmask_app = be16_to_cpu(pi->apptag) & pi_data->apptag_mask;
293 unmask_app_exp = pi_data->apptag & pi_data->apptag_mask;
294 if (unmask_app != unmask_app_exp) {
295 log_err("%s: APPTAG compare error: LBA: %llu Expected=%x, Actual=%x\n",
296 f->file_name, (unsigned long long)slba,
297 unmask_app_exp, unmask_app);
298 return -EIO;
299 }
300 }
301
302 if (pi_data->io_flags & NVME_IO_PRINFO_PRCHK_REF) {
303 switch (data->pi_type) {
304 case NVME_NS_DPS_PI_TYPE1:
305 case NVME_NS_DPS_PI_TYPE2:
306 ref = get_unaligned_be48(pi->srtag);
307 ref_exp = (slba + lba_num) & ((1ULL << 48) - 1);
308 if (ref != ref_exp) {
309 log_err("%s: REFTAG compare error: LBA: %llu Expected=%llx, Actual=%llx\n",
310 f->file_name, (unsigned long long)slba,
311 ref_exp, ref);
312 return -EIO;
313 }
314 break;
315 case NVME_NS_DPS_PI_TYPE3:
316 break;
317 }
318 }
319next:
320 if (data->lba_ext) {
321 buf += data->lba_ext;
322 } else {
323 buf += data->lba_size;
324 md_buf += data->ms;
325 }
326 lba_num++;
327 }
328
329 return 0;
330}
4885a6eb 331void fio_nvme_uring_cmd_trim_prep(struct nvme_uring_cmd *cmd, struct io_u *io_u,
5d4ee0de 332 struct nvme_dsm *dsm)
4885a6eb
VF		 333{
334 struct nvme_data *data = FILE_ENG_DATA(io_u->file);
5d4ee0de
AK		 335 struct trim_range *range;
336 uint8_t *buf_point;
337 int i;
4885a6eb
VF		 338
339 cmd->opcode = nvme_cmd_dsm;
340 cmd->nsid = data->nsid;
4885a6eb 341 cmd->cdw11 = NVME_ATTRIBUTE_DEALLOCATE;
5d4ee0de
AK		 342 cmd->addr = (__u64) (uintptr_t) (&dsm->range[0]);
343
344 if (dsm->nr_ranges == 1) {
345 dsm->range[0].slba = get_slba(data, io_u->offset);
346 /* nlb is a 1-based value for deallocate */
347 dsm->range[0].nlb = get_nlb(data, io_u->xfer_buflen) + 1;
348 cmd->cdw10 = 0;
349 cmd->data_len = sizeof(struct nvme_dsm_range);
350 } else {
351 buf_point = io_u->xfer_buf;
352 for (i = 0; i < io_u->number_trim; i++) {
353 range = (struct trim_range *)buf_point;
354 dsm->range[i].slba = get_slba(data, range->start);
355 /* nlb is a 1-based value for deallocate */
356 dsm->range[i].nlb = get_nlb(data, range->len) + 1;
357 buf_point += sizeof(struct trim_range);
358 }
359 cmd->cdw10 = io_u->number_trim - 1;
360 cmd->data_len = io_u->number_trim * sizeof(struct nvme_dsm_range);
361 }
4885a6eb
VF		 362}
363
b3d5e3fd 364int fio_nvme_uring_cmd_prep(struct nvme_uring_cmd *cmd, struct io_u *io_u,
5d4ee0de 365 struct iovec *iov, struct nvme_dsm *dsm)
b3d5e3fd
AK		 366{
367 struct nvme_data *data = FILE_ENG_DATA(io_u->file);
368 __u64 slba;
369 __u32 nlb;
370
371 memset(cmd, 0, sizeof(struct nvme_uring_cmd));
372
4885a6eb
VF		 373 switch (io_u->ddir) {
374 case DDIR_READ:
b3d5e3fd 375 cmd->opcode = nvme_cmd_read;
4885a6eb
VF		 376 break;
377 case DDIR_WRITE:
b3d5e3fd 378 cmd->opcode = nvme_cmd_write;
4885a6eb
VF		 379 break;
380 case DDIR_TRIM:
381 fio_nvme_uring_cmd_trim_prep(cmd, io_u, dsm);
382 return 0;
383 default:
b3d5e3fd 384 return -ENOTSUP;
345fa8fd 385 }
b3d5e3fd 386
0d610785
AK		 387 slba = get_slba(data, io_u->offset);
388 nlb = get_nlb(data, io_u->xfer_buflen);
4885a6eb 389
b3d5e3fd
AK		 390 /* cdw10 and cdw11 represent starting lba */
391 cmd->cdw10 = slba & 0xffffffff;
392 cmd->cdw11 = slba >> 32;
393 /* cdw12 represent number of lba's for read/write */
a7e8aae0
KB		 394 cmd->cdw12 = nlb | (io_u->dtype << 20);
395 cmd->cdw13 = io_u->dspec << 16;
b3d5e3fd
AK		 396 if (iov) {
397 iov->iov_base = io_u->xfer_buf;
398 iov->iov_len = io_u->xfer_buflen;
399 cmd->addr = (__u64)(uintptr_t)iov;
400 cmd->data_len = 1;
401 } else {
402 cmd->addr = (__u64)(uintptr_t)io_u->xfer_buf;
403 cmd->data_len = io_u->xfer_buflen;
404 }
2d6451c9
AK		 405 if (data->lba_shift && data->ms) {
406 cmd->metadata = (__u64)(uintptr_t)io_u->mmap_data;
407 cmd->metadata_len = (nlb + 1) * data->ms;
408 }
b3d5e3fd
AK		 409 cmd->nsid = data->nsid;
410 return 0;
411}
412
3ee8311a
AK		 413void fio_nvme_pi_fill(struct nvme_uring_cmd *cmd, struct io_u *io_u,
414 struct nvme_cmd_ext_io_opts *opts)
415{
416 struct nvme_data *data = FILE_ENG_DATA(io_u->file);
417 __u64 slba;
418
0d610785 419 slba = get_slba(data, io_u->offset);
3ee8311a
AK		 420 cmd->cdw12 |= opts->io_flags;
421
5163f35e
AK		 422 if (data->pi_type && !(opts->io_flags & NVME_IO_PRINFO_PRACT)) {
423 if (data->guard_type == NVME_NVM_NS_16B_GUARD)
424 fio_nvme_generate_pi_16b_guard(data, io_u, opts);
08371767
AK		 425 else if (data->guard_type == NVME_NVM_NS_64B_GUARD)
426 fio_nvme_generate_pi_64b_guard(data, io_u, opts);
5163f35e
AK		 427 }
428
3ee8311a
AK		 429 switch (data->pi_type) {
430 case NVME_NS_DPS_PI_TYPE1:
431 case NVME_NS_DPS_PI_TYPE2:
432 switch (data->guard_type) {
433 case NVME_NVM_NS_16B_GUARD:
e4a9812d
AK		 434 if (opts->io_flags & NVME_IO_PRINFO_PRCHK_REF)
435 cmd->cdw14 = (__u32)slba;
3ee8311a
AK		 436 break;
437 case NVME_NVM_NS_64B_GUARD:
e4a9812d
AK		 438 if (opts->io_flags & NVME_IO_PRINFO_PRCHK_REF) {
439 cmd->cdw14 = (__u32)slba;
440 cmd->cdw3 = ((slba >> 32) & 0xffff);
441 }
3ee8311a
AK		 442 break;
443 default:
444 break;
445 }
e4a9812d
AK		 446 if (opts->io_flags & NVME_IO_PRINFO_PRCHK_APP)
447 cmd->cdw15 = (opts->apptag_mask << 16 | opts->apptag);
3ee8311a
AK		 448 break;
449 case NVME_NS_DPS_PI_TYPE3:
e4a9812d
AK		 450 if (opts->io_flags & NVME_IO_PRINFO_PRCHK_APP)
451 cmd->cdw15 = (opts->apptag_mask << 16 | opts->apptag);
3ee8311a
AK		 452 break;
453 case NVME_NS_DPS_PI_NONE:
454 break;
455 }
456}
457
5163f35e
AK		 458int fio_nvme_pi_verify(struct nvme_data *data, struct io_u *io_u)
459{
460 int ret = 0;
461
462 switch (data->guard_type) {
463 case NVME_NVM_NS_16B_GUARD:
464 ret = fio_nvme_verify_pi_16b_guard(data, io_u);
465 break;
08371767
AK		 466 case NVME_NVM_NS_64B_GUARD:
467 ret = fio_nvme_verify_pi_64b_guard(data, io_u);
468 break;
5163f35e
AK		 469 default:
470 break;
471 }
472
473 return ret;
474}
475
b3d5e3fd
AK		 476static int nvme_identify(int fd, __u32 nsid, enum nvme_identify_cns cns,
477 enum nvme_csi csi, void *data)
478{
479 struct nvme_passthru_cmd cmd = {
480 .opcode = nvme_admin_identify,
481 .nsid = nsid,
482 .addr = (__u64)(uintptr_t)data,
483 .data_len = NVME_IDENTIFY_DATA_SIZE,
484 .cdw10 = cns,
485 .cdw11 = csi << NVME_IDENTIFY_CSI_SHIFT,
486 .timeout_ms = NVME_DEFAULT_IOCTL_TIMEOUT,
487 };
488
489 return ioctl(fd, NVME_IOCTL_ADMIN_CMD, &cmd);
490}
491
3ee8311a
AK		 492int fio_nvme_get_info(struct fio_file *f, __u64 *nlba, __u32 pi_act,
493 struct nvme_data *data)
b3d5e3fd
AK		 494{
495 struct nvme_id_ns ns;
3ee8311a
AK		 496 struct nvme_id_ctrl ctrl;
497 struct nvme_nvm_id_ns nvm_ns;
37a0881f 498 int namespace_id;
b3d5e3fd 499 int fd, err;
3ee8311a 500 __u32 format_idx, elbaf;
b3d5e3fd
AK		 501
502 if (f->filetype != FIO_TYPE_CHAR) {
503 log_err("ioengine io_uring_cmd only works with nvme ns "
504 "generic char devices (/dev/ngXnY)\n");
505 return 1;
506 }
507
508 fd = open(f->file_name, O_RDONLY);
509 if (fd < 0)
510 return -errno;
511
512 namespace_id = ioctl(fd, NVME_IOCTL_ID);
513 if (namespace_id < 0) {
af10f514 514 err = -errno;
345fa8fd
AK		 515 log_err("%s: failed to fetch namespace-id\n", f->file_name);
516 goto out;
b3d5e3fd
AK		 517 }
518
3ee8311a
AK		 519 err = nvme_identify(fd, 0, NVME_IDENTIFY_CNS_CTRL, NVME_CSI_NVM, &ctrl);
520 if (err) {
521 log_err("%s: failed to fetch identify ctrl\n", f->file_name);
522 goto out;
523 }
524
b3d5e3fd
AK		 525 /*
526 * Identify namespace to get namespace-id, namespace size in LBA's
527 * and LBA data size.
528 */
529 err = nvme_identify(fd, namespace_id, NVME_IDENTIFY_CNS_NS,
530 NVME_CSI_NVM, &ns);
531 if (err) {
345fa8fd
AK		 532 log_err("%s: failed to fetch identify namespace\n",
533 f->file_name);
3ee8311a 534 goto out;
b3d5e3fd
AK		 535 }
536
e7e5023b 537 data->nsid = namespace_id;
01a7d384
AK		 538
539 /*
540 * 16 or 64 as maximum number of supported LBA formats.
541 * From flbas bit 0-3 indicates lsb and bit 5-6 indicates msb
542 * of the format index used to format the namespace.
543 */
544 if (ns.nlbaf < 16)
545 format_idx = ns.flbas & 0xf;
546 else
547 format_idx = (ns.flbas & 0xf) + (((ns.flbas >> 5) & 0x3) << 4);
548
e7e5023b 549 data->lba_size = 1 << ns.lbaf[format_idx].ds;
2d6451c9 550 data->ms = le16_to_cpu(ns.lbaf[format_idx].ms);
345fa8fd 551
3ee8311a
AK		 552 /* Check for end to end data protection support */
553 if (data->ms && (ns.dps & NVME_NS_DPS_PI_MASK))
554 data->pi_type = (ns.dps & NVME_NS_DPS_PI_MASK);
555
556 if (!data->pi_type)
557 goto check_elba;
558
559 if (ctrl.ctratt & NVME_CTRL_CTRATT_ELBAS) {
560 err = nvme_identify(fd, namespace_id, NVME_IDENTIFY_CNS_CSI_NS,
561 NVME_CSI_NVM, &nvm_ns);
562 if (err) {
563 log_err("%s: failed to fetch identify nvm namespace\n",
564 f->file_name);
565 goto out;
566 }
567
568 elbaf = le32_to_cpu(nvm_ns.elbaf[format_idx]);
569
570 /* Currently we don't support storage tags */
571 if (elbaf & NVME_ID_NS_NVM_STS_MASK) {
572 log_err("%s: Storage tag not supported\n",
573 f->file_name);
574 err = -ENOTSUP;
575 goto out;
576 }
577
578 data->guard_type = (elbaf >> NVME_ID_NS_NVM_GUARD_SHIFT) &
579 NVME_ID_NS_NVM_GUARD_MASK;
580
581 /* No 32 bit guard, as storage tag is mandatory for it */
582 switch (data->guard_type) {
583 case NVME_NVM_NS_16B_GUARD:
584 data->pi_size = sizeof(struct nvme_16b_guard_pif);
585 break;
586 case NVME_NVM_NS_64B_GUARD:
587 data->pi_size = sizeof(struct nvme_64b_guard_pif);
588 break;
589 default:
590 break;
591 }
592 } else {
593 data->guard_type = NVME_NVM_NS_16B_GUARD;
594 data->pi_size = sizeof(struct nvme_16b_guard_pif);
595 }
596
597 /*
598 * when PRACT bit is set to 1, and metadata size is equal to protection
599 * information size, controller inserts and removes PI for write and
600 * read commands respectively.
601 */
602 if (pi_act && data->ms == data->pi_size)
603 data->ms = 0;
604
605 data->pi_loc = (ns.dps & NVME_NS_DPS_PI_FIRST);
606
607check_elba:
345fa8fd 608 /*
345fa8fd
AK		 609 * Bit 4 for flbas indicates if metadata is transferred at the end of
610 * logical block creating an extended LBA.
611 */
2d6451c9 612 if (data->ms && ((ns.flbas >> 4) & 0x1))
e7e5023b
AK		 613 data->lba_ext = data->lba_size + data->ms;
614 else
615 data->lba_shift = ilog2(data->lba_size);
616
b3d5e3fd
AK		 617 *nlba = ns.nsze;
618
345fa8fd 619out:
b3d5e3fd 620 close(fd);
345fa8fd 621 return err;
b3d5e3fd 622}
3d05e0ff
AK		 623
624int fio_nvme_get_zoned_model(struct thread_data *td, struct fio_file *f,
625 enum zbd_zoned_model *model)
626{
627 struct nvme_data *data = FILE_ENG_DATA(f);
628 struct nvme_id_ns ns;
629 struct nvme_passthru_cmd cmd;
630 int fd, ret = 0;
631
632 if (f->filetype != FIO_TYPE_CHAR)
633 return -EINVAL;
634
635 /* File is not yet opened */
636 fd = open(f->file_name, O_RDONLY | O_LARGEFILE);
637 if (fd < 0)
638 return -errno;
639
640 /* Using nvme_id_ns for data as sizes are same */
641 ret = nvme_identify(fd, data->nsid, NVME_IDENTIFY_CNS_CSI_CTRL,
642 NVME_CSI_ZNS, &ns);
643 if (ret) {
644 *model = ZBD_NONE;
645 goto out;
646 }
647
648 memset(&cmd, 0, sizeof(struct nvme_passthru_cmd));
649
650 /* Using nvme_id_ns for data as sizes are same */
651 ret = nvme_identify(fd, data->nsid, NVME_IDENTIFY_CNS_CSI_NS,
652 NVME_CSI_ZNS, &ns);
653 if (ret) {
654 *model = ZBD_NONE;
655 goto out;
656 }
657
658 *model = ZBD_HOST_MANAGED;
659out:
660 close(fd);
661 return 0;
662}
663
664static int nvme_report_zones(int fd, __u32 nsid, __u64 slba, __u32 zras_feat,
665 __u32 data_len, void *data)
666{
667 struct nvme_passthru_cmd cmd = {
668 .opcode = nvme_zns_cmd_mgmt_recv,
669 .nsid = nsid,
670 .addr = (__u64)(uintptr_t)data,
671 .data_len = data_len,
672 .cdw10 = slba & 0xffffffff,
673 .cdw11 = slba >> 32,
674 .cdw12 = (data_len >> 2) - 1,
675 .cdw13 = NVME_ZNS_ZRA_REPORT_ZONES | zras_feat,
676 .timeout_ms = NVME_DEFAULT_IOCTL_TIMEOUT,
677 };
678
679 return ioctl(fd, NVME_IOCTL_IO_CMD, &cmd);
680}
681
682int fio_nvme_report_zones(struct thread_data *td, struct fio_file *f,
683 uint64_t offset, struct zbd_zone *zbdz,
684 unsigned int nr_zones)
685{
686 struct nvme_data *data = FILE_ENG_DATA(f);
687 struct nvme_zone_report *zr;
688 struct nvme_zns_id_ns zns_ns;
689 struct nvme_id_ns ns;
690 unsigned int i = 0, j, zones_fetched = 0;
691 unsigned int max_zones, zones_chunks = 1024;
692 int fd, ret = 0;
693 __u32 zr_len;
694 __u64 zlen;
695
696 /* File is not yet opened */
697 fd = open(f->file_name, O_RDONLY | O_LARGEFILE);
698 if (fd < 0)
699 return -errno;
700
701 zones_fetched = 0;
702 zr_len = sizeof(*zr) + (zones_chunks * sizeof(struct nvme_zns_desc));
703 zr = calloc(1, zr_len);
3efcb23f
JA		 704 if (!zr) {
705 close(fd);
3d05e0ff 706 return -ENOMEM;
3efcb23f 707 }
3d05e0ff
AK		 708
709 ret = nvme_identify(fd, data->nsid, NVME_IDENTIFY_CNS_NS,
710 NVME_CSI_NVM, &ns);
711 if (ret) {
712 log_err("%s: nvme_identify_ns failed, err=%d\n", f->file_name,
713 ret);
714 goto out;
715 }
716
717 ret = nvme_identify(fd, data->nsid, NVME_IDENTIFY_CNS_CSI_NS,
718 NVME_CSI_ZNS, &zns_ns);
719 if (ret) {
720 log_err("%s: nvme_zns_identify_ns failed, err=%d\n",
721 f->file_name, ret);
722 goto out;
723 }
724 zlen = zns_ns.lbafe[ns.flbas & 0x0f].zsze << data->lba_shift;
725
726 max_zones = (f->real_file_size - offset) / zlen;
727 if (max_zones < nr_zones)
728 nr_zones = max_zones;
729
730 if (nr_zones < zones_chunks)
731 zones_chunks = nr_zones;
732
733 while (zones_fetched < nr_zones) {
734 if (zones_fetched + zones_chunks >= nr_zones) {
735 zones_chunks = nr_zones - zones_fetched;
736 zr_len = sizeof(*zr) + (zones_chunks * sizeof(struct nvme_zns_desc));
737 }
738 ret = nvme_report_zones(fd, data->nsid, offset >> data->lba_shift,
739 NVME_ZNS_ZRAS_FEAT_ERZ, zr_len, (void *)zr);
740 if (ret) {
741 log_err("%s: nvme_zns_report_zones failed, err=%d\n",
742 f->file_name, ret);
743 goto out;
744 }
745
746 /* Transform the zone-report */
747 for (j = 0; j < zr->nr_zones; j++, i++) {
748 struct nvme_zns_desc *desc = (struct nvme_zns_desc *)&(zr->entries[j]);
749
750 zbdz[i].start = desc->zslba << data->lba_shift;
751 zbdz[i].len = zlen;
752 zbdz[i].wp = desc->wp << data->lba_shift;
753 zbdz[i].capacity = desc->zcap << data->lba_shift;
754
755 /* Zone Type is stored in first 4 bits. */
756 switch (desc->zt & 0x0f) {
757 case NVME_ZONE_TYPE_SEQWRITE_REQ:
758 zbdz[i].type = ZBD_ZONE_TYPE_SWR;
759 break;
760 default:
761 log_err("%s: invalid type for zone at offset %llu.\n",
2fa0ab21 762 f->file_name, (unsigned long long) desc->zslba);
3d05e0ff
AK		 763 ret = -EIO;
764 goto out;
765 }
766
767 /* Zone State is stored in last 4 bits. */
768 switch (desc->zs >> 4) {
769 case NVME_ZNS_ZS_EMPTY:
770 zbdz[i].cond = ZBD_ZONE_COND_EMPTY;
771 break;
772 case NVME_ZNS_ZS_IMPL_OPEN:
773 zbdz[i].cond = ZBD_ZONE_COND_IMP_OPEN;
774 break;
775 case NVME_ZNS_ZS_EXPL_OPEN:
776 zbdz[i].cond = ZBD_ZONE_COND_EXP_OPEN;
777 break;
778 case NVME_ZNS_ZS_CLOSED:
779 zbdz[i].cond = ZBD_ZONE_COND_CLOSED;
780 break;
781 case NVME_ZNS_ZS_FULL:
782 zbdz[i].cond = ZBD_ZONE_COND_FULL;
783 break;
784 case NVME_ZNS_ZS_READ_ONLY:
785 case NVME_ZNS_ZS_OFFLINE:
786 default:
787 /* Treat all these conditions as offline (don't use!) */
788 zbdz[i].cond = ZBD_ZONE_COND_OFFLINE;
789 zbdz[i].wp = zbdz[i].start;
790 }
791 }
792 zones_fetched += zr->nr_zones;
793 offset += zr->nr_zones * zlen;
794 }
795
796 ret = zones_fetched;
797out:
798 free(zr);
799 close(fd);
800
801 return ret;
802}
803
804int fio_nvme_reset_wp(struct thread_data *td, struct fio_file *f,
805 uint64_t offset, uint64_t length)
806{
807 struct nvme_data *data = FILE_ENG_DATA(f);
808 unsigned int nr_zones;
809 unsigned long long zslba;
810 int i, fd, ret = 0;
811
812 /* If the file is not yet opened, open it for this function. */
813 fd = f->fd;
814 if (fd < 0) {
815 fd = open(f->file_name, O_RDWR | O_LARGEFILE);
816 if (fd < 0)
817 return -errno;
818 }
819
820 zslba = offset >> data->lba_shift;
821 nr_zones = (length + td->o.zone_size - 1) / td->o.zone_size;
822
823 for (i = 0; i < nr_zones; i++, zslba += (td->o.zone_size >> data->lba_shift)) {
824 struct nvme_passthru_cmd cmd = {
825 .opcode = nvme_zns_cmd_mgmt_send,
826 .nsid = data->nsid,
827 .cdw10 = zslba & 0xffffffff,
828 .cdw11 = zslba >> 32,
829 .cdw13 = NVME_ZNS_ZSA_RESET,
830 .addr = (__u64)(uintptr_t)NULL,
831 .data_len = 0,
832 .timeout_ms = NVME_DEFAULT_IOCTL_TIMEOUT,
833 };
834
835 ret = ioctl(fd, NVME_IOCTL_IO_CMD, &cmd);
836 }
837
838 if (f->fd < 0)
839 close(fd);
840 return -ret;
841}
842
843int fio_nvme_get_max_open_zones(struct thread_data *td, struct fio_file *f,
844 unsigned int *max_open_zones)
845{
846 struct nvme_data *data = FILE_ENG_DATA(f);
847 struct nvme_zns_id_ns zns_ns;
848 int fd, ret = 0;
849
850 fd = open(f->file_name, O_RDONLY | O_LARGEFILE);
851 if (fd < 0)
852 return -errno;
853
854 ret = nvme_identify(fd, data->nsid, NVME_IDENTIFY_CNS_CSI_NS,
855 NVME_CSI_ZNS, &zns_ns);
856 if (ret) {
857 log_err("%s: nvme_zns_identify_ns failed, err=%d\n",
858 f->file_name, ret);
859 goto out;
860 }
861
862 *max_open_zones = zns_ns.mor + 1;
863out:
864 close(fd);
865 return ret;
866}
a7e8aae0
KB		 867
868static inline int nvme_fdp_reclaim_unit_handle_status(int fd, __u32 nsid,
869 __u32 data_len, void *data)
870{
871 struct nvme_passthru_cmd cmd = {
872 .opcode = nvme_cmd_io_mgmt_recv,
873 .nsid = nsid,
874 .addr = (__u64)(uintptr_t)data,
875 .data_len = data_len,
876 .cdw10 = 1,
877 .cdw11 = (data_len >> 2) - 1,
878 };
879
880 return ioctl(fd, NVME_IOCTL_IO_CMD, &cmd);
881}
882
883int fio_nvme_iomgmt_ruhs(struct thread_data *td, struct fio_file *f,
884 struct nvme_fdp_ruh_status *ruhs, __u32 bytes)
885{
886 struct nvme_data *data = FILE_ENG_DATA(f);
887 int fd, ret;
888
889 fd = open(f->file_name, O_RDONLY | O_LARGEFILE);
890 if (fd < 0)
891 return -errno;
892
893 ret = nvme_fdp_reclaim_unit_handle_status(fd, data->nsid, bytes, ruhs);
894 if (ret) {
895 log_err("%s: nvme_fdp_reclaim_unit_handle_status failed, err=%d\n",
896 f->file_name, ret);
897 errno = ENOTSUP;
898 } else
899 errno = 0;
900
af10f514 901 ret = -errno;
a7e8aae0 902 close(fd);
af10f514 903 return ret;
a7e8aae0 904}
fio - Flexible IO Tester