4 * IO engine using the xNVMe C API.
6 * See: http://xnvme.io/
8 * SPDX-License-Identifier: Apache-2.0
13 #include <libxnvme_libconf.h>
14 #include <libxnvme_nvm.h>
15 #include <libxnvme_znd.h>
16 #include <libxnvme_spec_fs.h>
18 #include "zbd_types.h"
21 static pthread_mutex_t g_serialize = PTHREAD_MUTEX_INITIALIZER;
23 struct xnvme_fioe_fwrap {
24 /* fio file representation */
25 struct fio_file *fio_file;
27 /* xNVMe device handle */
28 struct xnvme_dev *dev;
29 /* xNVMe device geometry */
30 const struct xnvme_geo *geo;
32 struct xnvme_queue *queue;
39 XNVME_STATIC_ASSERT(sizeof(struct xnvme_fioe_fwrap) == 64, "Incorrect size")
41 struct xnvme_fioe_data {
42 /* I/O completion queue */
45 /* # of iocq entries; incremented via getevents()/cb_pool() */
49 * # of errors; incremented when observed on completion via
50 * getevents()/cb_pool()
54 /* Controller which device/file to select */
58 /* Number of devices/files for which open() has been called */
60 /* Number of devices/files allocated in files[] */
67 struct xnvme_fioe_fwrap files[];
69 XNVME_STATIC_ASSERT(sizeof(struct xnvme_fioe_data) == 64, "Incorrect size")
71 struct xnvme_fioe_options {
74 unsigned int sqpoll_thread;
75 unsigned int xnvme_dev_nsid;
76 unsigned int xnvme_iovec;
83 static struct fio_option options[] = {
86 .lname = "High Priority",
87 .type = FIO_OPT_STR_SET,
88 .off1 = offsetof(struct xnvme_fioe_options, hipri),
89 .help = "Use polled IO completions",
90 .category = FIO_OPT_C_ENGINE,
91 .group = FIO_OPT_G_XNVME,
94 .name = "sqthread_poll",
95 .lname = "Kernel SQ thread polling",
96 .type = FIO_OPT_STR_SET,
97 .off1 = offsetof(struct xnvme_fioe_options, sqpoll_thread),
98 .help = "Offload submission/completion to kernel thread",
99 .category = FIO_OPT_C_ENGINE,
100 .group = FIO_OPT_G_XNVME,
104 .lname = "xNVMe Backend",
105 .type = FIO_OPT_STR_STORE,
106 .off1 = offsetof(struct xnvme_fioe_options, xnvme_be),
107 .help = "Select xNVMe backend [spdk,linux,fbsd]",
108 .category = FIO_OPT_C_ENGINE,
109 .group = FIO_OPT_G_XNVME,
112 .name = "xnvme_async",
113 .lname = "xNVMe Asynchronous command-interface",
114 .type = FIO_OPT_STR_STORE,
115 .off1 = offsetof(struct xnvme_fioe_options, xnvme_async),
116 .help = "Select xNVMe async. interface: [emu,thrpool,io_uring,libaio,posix,nil]",
117 .category = FIO_OPT_C_ENGINE,
118 .group = FIO_OPT_G_XNVME,
121 .name = "xnvme_sync",
122 .lname = "xNVMe Synchronous. command-interface",
123 .type = FIO_OPT_STR_STORE,
124 .off1 = offsetof(struct xnvme_fioe_options, xnvme_sync),
125 .help = "Select xNVMe sync. interface: [nvme,psync]",
126 .category = FIO_OPT_C_ENGINE,
127 .group = FIO_OPT_G_XNVME,
130 .name = "xnvme_admin",
131 .lname = "xNVMe Admin command-interface",
132 .type = FIO_OPT_STR_STORE,
133 .off1 = offsetof(struct xnvme_fioe_options, xnvme_admin),
134 .help = "Select xNVMe admin. cmd-interface: [nvme,block,file_as_ns]",
135 .category = FIO_OPT_C_ENGINE,
136 .group = FIO_OPT_G_XNVME,
139 .name = "xnvme_dev_nsid",
140 .lname = "xNVMe Namespace-Identifier, for user-space NVMe driver",
142 .off1 = offsetof(struct xnvme_fioe_options, xnvme_dev_nsid),
143 .help = "xNVMe Namespace-Identifier, for user-space NVMe driver",
144 .category = FIO_OPT_C_ENGINE,
145 .group = FIO_OPT_G_XNVME,
148 .name = "xnvme_iovec",
149 .lname = "Vectored IOs",
150 .type = FIO_OPT_STR_SET,
151 .off1 = offsetof(struct xnvme_fioe_options, xnvme_iovec),
152 .help = "Send vectored IOs",
153 .category = FIO_OPT_C_ENGINE,
154 .group = FIO_OPT_G_XNVME,
162 static void cb_pool(struct xnvme_cmd_ctx *ctx, void *cb_arg)
164 struct io_u *io_u = cb_arg;
165 struct xnvme_fioe_data *xd = io_u->mmap_data;
167 if (xnvme_cmd_ctx_cpl_status(ctx)) {
168 xnvme_cmd_ctx_pr(ctx, XNVME_PR_DEF);
173 xd->iocq[xd->completed++] = io_u;
174 xnvme_queue_put_cmd_ctx(ctx->async.queue, ctx);
177 static struct xnvme_opts xnvme_opts_from_fioe(struct thread_data *td)
179 struct xnvme_fioe_options *o = td->eo;
180 struct xnvme_opts opts = xnvme_opts_default();
182 opts.nsid = o->xnvme_dev_nsid;
183 opts.be = o->xnvme_be;
184 opts.async = o->xnvme_async;
185 opts.sync = o->xnvme_sync;
186 opts.admin = o->xnvme_admin;
188 opts.poll_io = o->hipri;
189 opts.poll_sq = o->sqpoll_thread;
191 opts.direct = td->o.odirect;
196 static void _dev_close(struct thread_data *td, struct xnvme_fioe_fwrap *fwrap)
199 xnvme_queue_term(fwrap->queue);
201 xnvme_dev_close(fwrap->dev);
203 memset(fwrap, 0, sizeof(*fwrap));
206 static void xnvme_fioe_cleanup(struct thread_data *td)
208 struct xnvme_fioe_data *xd = NULL;
211 if (!td->io_ops_data)
214 xd = td->io_ops_data;
216 err = pthread_mutex_lock(&g_serialize);
218 log_err("ioeng->cleanup(): pthread_mutex_lock(), err(%d)\n", err);
219 /* NOTE: not returning here */
221 for (uint64_t i = 0; i < xd->nallocated; ++i)
222 _dev_close(td, &xd->files[i]);
225 err = pthread_mutex_unlock(&g_serialize);
227 log_err("ioeng->cleanup(): pthread_mutex_unlock(), err(%d)\n", err);
233 td->io_ops_data = NULL;
237 * Helper function setting up device handles as addressed by the naming
238 * convention of the given `fio_file` filename.
240 * Checks thread-options for explicit control of asynchronous implementation via
241 * the ``--xnvme_async={thrpool,emu,posix,io_uring,libaio,nil}``.
243 static int _dev_open(struct thread_data *td, struct fio_file *f)
245 struct xnvme_opts opts = xnvme_opts_from_fioe(td);
246 struct xnvme_fioe_data *xd = td->io_ops_data;
247 struct xnvme_fioe_fwrap *fwrap;
251 if (f->fileno > (int)xd->nallocated) {
252 log_err("ioeng->_dev_open(%s): invalid assumption\n", f->file_name);
256 fwrap = &xd->files[f->fileno];
258 err = pthread_mutex_lock(&g_serialize);
260 log_err("ioeng->_dev_open(%s): pthread_mutex_lock(), err(%d)\n", f->file_name,
265 fwrap->dev = xnvme_dev_open(f->file_name, &opts);
267 log_err("ioeng->_dev_open(%s): xnvme_dev_open(), err(%d)\n", f->file_name, errno);
270 fwrap->geo = xnvme_dev_get_geo(fwrap->dev);
272 if (xnvme_queue_init(fwrap->dev, td->o.iodepth, flags, &(fwrap->queue))) {
273 log_err("ioeng->_dev_open(%s): xnvme_queue_init(), err(?)\n", f->file_name);
276 xnvme_queue_set_cb(fwrap->queue, cb_pool, NULL);
278 fwrap->ssw = xnvme_dev_get_ssw(fwrap->dev);
279 fwrap->lba_nbytes = fwrap->geo->lba_nbytes;
282 fwrap->fio_file->filetype = FIO_TYPE_BLOCK;
283 fwrap->fio_file->real_file_size = fwrap->geo->tbytes;
284 fio_file_set_size_known(fwrap->fio_file);
286 err = pthread_mutex_unlock(&g_serialize);
288 log_err("ioeng->_dev_open(%s): pthread_mutex_unlock(), err(%d)\n", f->file_name,
294 xnvme_queue_term(fwrap->queue);
295 xnvme_dev_close(fwrap->dev);
297 err = pthread_mutex_unlock(&g_serialize);
299 log_err("ioeng->_dev_open(%s): pthread_mutex_unlock(), err(%d)\n", f->file_name,
305 static int xnvme_fioe_init(struct thread_data *td)
307 struct xnvme_fioe_data *xd = NULL;
311 if (!td->o.use_thread) {
312 log_err("ioeng->init(): --thread=1 is required\n");
316 /* Allocate xd and iocq */
317 xd = calloc(1, sizeof(*xd) + sizeof(*xd->files) * td->o.nr_files);
319 log_err("ioeng->init(): !calloc(), err(%d)\n", errno);
323 xd->iocq = calloc(td->o.iodepth, sizeof(struct io_u *));
326 log_err("ioeng->init(): !calloc(xd->iocq), err(%d)\n", errno);
330 xd->iovec = calloc(td->o.iodepth, sizeof(*xd->iovec));
334 log_err("ioeng->init(): !calloc(xd->iovec), err(%d)\n", errno);
339 td->io_ops_data = xd;
341 for_each_file(td, f, i)
343 if (_dev_open(td, f)) {
345 * Note: We are not freeing xd, iocq and iovec. This
346 * will be done as part of cleanup routine.
348 log_err("ioeng->init(): failed; _dev_open(%s)\n", f->file_name);
355 if (xd->nallocated != td->o.nr_files) {
356 log_err("ioeng->init(): failed; nallocated != td->o.nr_files\n");
363 /* NOTE: using the first device for buffer-allocators) */
364 static int xnvme_fioe_iomem_alloc(struct thread_data *td, size_t total_mem)
366 struct xnvme_fioe_data *xd = td->io_ops_data;
367 struct xnvme_fioe_fwrap *fwrap = &xd->files[0];
370 log_err("ioeng->iomem_alloc(): failed; no dev-handle\n");
374 td->orig_buffer = xnvme_buf_alloc(fwrap->dev, total_mem);
376 return td->orig_buffer == NULL;
379 /* NOTE: using the first device for buffer-allocators) */
380 static void xnvme_fioe_iomem_free(struct thread_data *td)
382 struct xnvme_fioe_data *xd = NULL;
383 struct xnvme_fioe_fwrap *fwrap = NULL;
385 if (!td->io_ops_data)
388 xd = td->io_ops_data;
389 fwrap = &xd->files[0];
392 log_err("ioeng->iomem_free(): failed no dev-handle\n");
396 xnvme_buf_free(fwrap->dev, td->orig_buffer);
399 static int xnvme_fioe_io_u_init(struct thread_data *td, struct io_u *io_u)
401 io_u->mmap_data = td->io_ops_data;
406 static void xnvme_fioe_io_u_free(struct thread_data *td, struct io_u *io_u)
408 io_u->mmap_data = NULL;
411 static struct io_u *xnvme_fioe_event(struct thread_data *td, int event)
413 struct xnvme_fioe_data *xd = td->io_ops_data;
416 assert((unsigned)event < xd->completed);
418 return xd->iocq[event];
421 static int xnvme_fioe_getevents(struct thread_data *td, unsigned int min, unsigned int max,
422 const struct timespec *t)
424 struct xnvme_fioe_data *xd = td->io_ops_data;
425 struct xnvme_fioe_fwrap *fwrap = NULL;
426 int nfiles = xd->nallocated;
429 if (xd->prev != -1 && ++xd->prev < nfiles) {
430 fwrap = &xd->files[xd->prev];
436 if (fwrap == NULL || xd->cur == nfiles) {
437 fwrap = &xd->files[0];
441 while (fwrap != NULL && xd->cur < nfiles && err >= 0) {
442 err = xnvme_queue_poke(fwrap->queue, max - xd->completed);
451 log_err("ioeng->getevents(): unhandled IO error\n");
456 if (xd->completed >= min) {
458 return xd->completed;
461 fwrap = &xd->files[xd->cur];
476 return xd->completed;
479 static enum fio_q_status xnvme_fioe_queue(struct thread_data *td, struct io_u *io_u)
481 struct xnvme_fioe_data *xd = td->io_ops_data;
482 struct xnvme_fioe_fwrap *fwrap;
483 struct xnvme_cmd_ctx *ctx;
488 bool vectored_io = ((struct xnvme_fioe_options *)td->eo)->xnvme_iovec;
490 fio_ro_check(td, io_u);
492 fwrap = &xd->files[io_u->file->fileno];
493 nsid = xnvme_dev_get_nsid(fwrap->dev);
495 slba = io_u->offset >> fwrap->ssw;
496 nlb = (io_u->xfer_buflen >> fwrap->ssw) - 1;
498 ctx = xnvme_queue_get_cmd_ctx(fwrap->queue);
499 ctx->async.cb_arg = io_u;
501 ctx->cmd.common.nsid = nsid;
502 ctx->cmd.nvm.slba = slba;
503 ctx->cmd.nvm.nlb = nlb;
505 switch (io_u->ddir) {
507 ctx->cmd.common.opcode = XNVME_SPEC_NVM_OPC_READ;
511 ctx->cmd.common.opcode = XNVME_SPEC_NVM_OPC_WRITE;
515 log_err("ioeng->queue(): ENOSYS: %u\n", io_u->ddir);
516 xnvme_queue_put_cmd_ctx(ctx->async.queue, ctx);
518 io_u->error = ENOSYS;
520 return FIO_Q_COMPLETED;
524 xd->iovec[io_u->index].iov_base = io_u->xfer_buf;
525 xd->iovec[io_u->index].iov_len = io_u->xfer_buflen;
527 err = xnvme_cmd_passv(ctx, &xd->iovec[io_u->index], 1, io_u->xfer_buflen, NULL, 0,
530 err = xnvme_cmd_pass(ctx, io_u->xfer_buf, io_u->xfer_buflen, NULL, 0);
538 xnvme_queue_put_cmd_ctx(ctx->async.queue, ctx);
542 log_err("ioeng->queue(): err: '%d'\n", err);
544 xnvme_queue_put_cmd_ctx(ctx->async.queue, ctx);
546 io_u->error = abs(err);
548 return FIO_Q_COMPLETED;
552 static int xnvme_fioe_close(struct thread_data *td, struct fio_file *f)
554 struct xnvme_fioe_data *xd = td->io_ops_data;
556 dprint(FD_FILE, "xnvme close %s -- nopen: %ld\n", f->file_name, xd->nopen);
563 static int xnvme_fioe_open(struct thread_data *td, struct fio_file *f)
565 struct xnvme_fioe_data *xd = td->io_ops_data;
567 dprint(FD_FILE, "xnvme open %s -- nopen: %ld\n", f->file_name, xd->nopen);
569 if (f->fileno > (int)xd->nallocated) {
570 log_err("ioeng->open(): f->fileno > xd->nallocated; invalid assumption\n");
573 if (xd->files[f->fileno].fio_file != f) {
574 log_err("ioeng->open(): fio_file != f; invalid assumption\n");
583 static int xnvme_fioe_invalidate(struct thread_data *td, struct fio_file *f)
585 /* Consider only doing this with be:spdk */
589 static int xnvme_fioe_get_max_open_zones(struct thread_data *td, struct fio_file *f,
590 unsigned int *max_open_zones)
592 struct xnvme_opts opts = xnvme_opts_from_fioe(td);
593 struct xnvme_dev *dev;
594 const struct xnvme_spec_znd_idfy_ns *zns;
595 int err = 0, err_lock;
597 if (f->filetype != FIO_TYPE_FILE && f->filetype != FIO_TYPE_BLOCK &&
598 f->filetype != FIO_TYPE_CHAR) {
599 log_info("ioeng->get_max_open_zoned(): ignoring filetype: %d\n", f->filetype);
602 err_lock = pthread_mutex_lock(&g_serialize);
604 log_err("ioeng->get_max_open_zones(): pthread_mutex_lock(), err(%d)\n", err_lock);
608 dev = xnvme_dev_open(f->file_name, &opts);
610 log_err("ioeng->get_max_open_zones(): xnvme_dev_open(), err(%d)\n", err_lock);
614 if (xnvme_dev_get_geo(dev)->type != XNVME_GEO_ZONED) {
620 zns = (void *)xnvme_dev_get_ns_css(dev);
622 log_err("ioeng->get_max_open_zones(): xnvme_dev_get_ns_css(), err(%d)\n", errno);
628 * intentional overflow as the value is zero-based and NVMe
629 * defines 0xFFFFFFFF as unlimited thus overflowing to 0 which
630 * is how fio indicates unlimited and otherwise just converting
633 *max_open_zones = zns->mor + 1;
636 xnvme_dev_close(dev);
637 err_lock = pthread_mutex_unlock(&g_serialize);
639 log_err("ioeng->get_max_open_zones(): pthread_mutex_unlock(), err(%d)\n",
646 * Currently, this function is called before of I/O engine initialization, so,
647 * we cannot consult the file-wrapping done when 'fioe' initializes.
648 * Instead we just open based on the given filename.
650 * TODO: unify the different setup methods, consider keeping the handle around,
651 * and consider how to support the --be option in this usecase
653 static int xnvme_fioe_get_zoned_model(struct thread_data *td, struct fio_file *f,
654 enum zbd_zoned_model *model)
656 struct xnvme_opts opts = xnvme_opts_from_fioe(td);
657 struct xnvme_dev *dev;
658 int err = 0, err_lock;
660 if (f->filetype != FIO_TYPE_FILE && f->filetype != FIO_TYPE_BLOCK &&
661 f->filetype != FIO_TYPE_CHAR) {
662 log_info("ioeng->get_zoned_model(): ignoring filetype: %d\n", f->filetype);
666 err = pthread_mutex_lock(&g_serialize);
668 log_err("ioeng->get_zoned_model(): pthread_mutex_lock(), err(%d)\n", err);
672 dev = xnvme_dev_open(f->file_name, &opts);
674 log_err("ioeng->get_zoned_model(): xnvme_dev_open(%s) failed, errno: %d\n",
675 f->file_name, errno);
680 switch (xnvme_dev_get_geo(dev)->type) {
681 case XNVME_GEO_UNKNOWN:
682 dprint(FD_ZBD, "%s: got 'unknown', assigning ZBD_NONE\n", f->file_name);
686 case XNVME_GEO_CONVENTIONAL:
687 dprint(FD_ZBD, "%s: got 'conventional', assigning ZBD_NONE\n", f->file_name);
691 case XNVME_GEO_ZONED:
692 dprint(FD_ZBD, "%s: got 'zoned', assigning ZBD_HOST_MANAGED\n", f->file_name);
693 *model = ZBD_HOST_MANAGED;
697 dprint(FD_ZBD, "%s: hit-default, assigning ZBD_NONE\n", f->file_name);
705 xnvme_dev_close(dev);
707 err_lock = pthread_mutex_unlock(&g_serialize);
709 log_err("ioeng->get_zoned_model(): pthread_mutex_unlock(), err(%d)\n", err_lock);
715 * Fills the given ``zbdz`` with at most ``nr_zones`` zone-descriptors.
717 * The implementation converts the NVMe Zoned Command Set log-pages for Zone
718 * descriptors into the Linux Kernel Zoned Block Report format.
720 * NOTE: This function is called before I/O engine initialization, that is,
721 * before ``_dev_open`` has been called and file-wrapping is setup. Thus is has
722 * to do the ``_dev_open`` itself, and shut it down again once it is done
723 * retrieving the log-pages and converting them to the report format.
725 * TODO: unify the different setup methods, consider keeping the handle around,
726 * and consider how to support the --async option in this usecase
728 static int xnvme_fioe_report_zones(struct thread_data *td, struct fio_file *f, uint64_t offset,
729 struct zbd_zone *zbdz, unsigned int nr_zones)
731 struct xnvme_opts opts = xnvme_opts_from_fioe(td);
732 const struct xnvme_spec_znd_idfy_lbafe *lbafe = NULL;
733 struct xnvme_dev *dev = NULL;
734 const struct xnvme_geo *geo = NULL;
735 struct xnvme_znd_report *rprt = NULL;
738 unsigned int limit = 0;
739 int err = 0, err_lock;
741 dprint(FD_ZBD, "%s: report_zones() offset: %zu, nr_zones: %u\n", f->file_name, offset,
744 err = pthread_mutex_lock(&g_serialize);
746 log_err("ioeng->report_zones(%s): pthread_mutex_lock(), err(%d)\n", f->file_name,
751 dev = xnvme_dev_open(f->file_name, &opts);
753 log_err("ioeng->report_zones(%s): xnvme_dev_open(), err(%d)\n", f->file_name,
758 geo = xnvme_dev_get_geo(dev);
759 ssw = xnvme_dev_get_ssw(dev);
760 lbafe = xnvme_znd_dev_get_lbafe(dev);
762 limit = nr_zones > geo->nzone ? geo->nzone : nr_zones;
764 dprint(FD_ZBD, "%s: limit: %u\n", f->file_name, limit);
766 slba = ((offset >> ssw) / geo->nsect) * geo->nsect;
768 rprt = xnvme_znd_report_from_dev(dev, slba, limit, 0);
770 log_err("ioeng->report_zones(%s): xnvme_znd_report_from_dev(), err(%d)\n",
771 f->file_name, errno);
775 if (rprt->nentries != limit) {
776 log_err("ioeng->report_zones(%s): nentries != nr_zones\n", f->file_name);
780 if (offset > geo->tbytes) {
781 log_err("ioeng->report_zones(%s): out-of-bounds\n", f->file_name);
785 /* Transform the zone-report */
786 for (uint32_t idx = 0; idx < rprt->nentries; ++idx) {
787 struct xnvme_spec_znd_descr *descr = XNVME_ZND_REPORT_DESCR(rprt, idx);
789 zbdz[idx].start = descr->zslba << ssw;
790 zbdz[idx].len = lbafe->zsze << ssw;
791 zbdz[idx].capacity = descr->zcap << ssw;
792 zbdz[idx].wp = descr->wp << ssw;
795 case XNVME_SPEC_ZND_TYPE_SEQWR:
796 zbdz[idx].type = ZBD_ZONE_TYPE_SWR;
800 log_err("ioeng->report_zones(%s): invalid type for zone at offset(%zu)\n",
801 f->file_name, zbdz[idx].start);
807 case XNVME_SPEC_ZND_STATE_EMPTY:
808 zbdz[idx].cond = ZBD_ZONE_COND_EMPTY;
810 case XNVME_SPEC_ZND_STATE_IOPEN:
811 zbdz[idx].cond = ZBD_ZONE_COND_IMP_OPEN;
813 case XNVME_SPEC_ZND_STATE_EOPEN:
814 zbdz[idx].cond = ZBD_ZONE_COND_EXP_OPEN;
816 case XNVME_SPEC_ZND_STATE_CLOSED:
817 zbdz[idx].cond = ZBD_ZONE_COND_CLOSED;
819 case XNVME_SPEC_ZND_STATE_FULL:
820 zbdz[idx].cond = ZBD_ZONE_COND_FULL;
823 case XNVME_SPEC_ZND_STATE_RONLY:
824 case XNVME_SPEC_ZND_STATE_OFFLINE:
826 zbdz[idx].cond = ZBD_ZONE_COND_OFFLINE;
832 xnvme_buf_virt_free(rprt);
834 xnvme_dev_close(dev);
836 err_lock = pthread_mutex_unlock(&g_serialize);
838 log_err("ioeng->report_zones(): pthread_mutex_unlock(), err: %d\n", err_lock);
840 dprint(FD_ZBD, "err: %d, nr_zones: %d\n", err, (int)nr_zones);
842 return err ? err : (int)limit;
846 * NOTE: This function may get called before I/O engine initialization, that is,
847 * before ``_dev_open`` has been called and file-wrapping is setup. In such
848 * case it has to do ``_dev_open`` itself, and shut it down again once it is
849 * done resetting write pointer of zones.
851 static int xnvme_fioe_reset_wp(struct thread_data *td, struct fio_file *f, uint64_t offset,
854 struct xnvme_opts opts = xnvme_opts_from_fioe(td);
855 struct xnvme_fioe_data *xd = NULL;
856 struct xnvme_fioe_fwrap *fwrap = NULL;
857 struct xnvme_dev *dev = NULL;
858 const struct xnvme_geo *geo = NULL;
859 uint64_t first, last;
862 int err = 0, err_lock;
864 if (td->io_ops_data) {
865 xd = td->io_ops_data;
866 fwrap = &xd->files[f->fileno];
875 err = pthread_mutex_lock(&g_serialize);
877 log_err("ioeng->reset_wp(): pthread_mutex_lock(), err(%d)\n", err);
881 dev = xnvme_dev_open(f->file_name, &opts);
883 log_err("ioeng->reset_wp(): xnvme_dev_open(%s) failed, errno(%d)\n",
884 f->file_name, errno);
887 geo = xnvme_dev_get_geo(dev);
888 ssw = xnvme_dev_get_ssw(dev);
891 nsid = xnvme_dev_get_nsid(dev);
893 first = ((offset >> ssw) / geo->nsect) * geo->nsect;
894 last = (((offset + length) >> ssw) / geo->nsect) * geo->nsect;
895 dprint(FD_ZBD, "first: 0x%lx, last: 0x%lx\n", first, last);
897 for (uint64_t zslba = first; zslba < last; zslba += geo->nsect) {
898 struct xnvme_cmd_ctx ctx = xnvme_cmd_ctx_from_dev(dev);
900 if (zslba >= (geo->nsect * geo->nzone)) {
901 log_err("ioeng->reset_wp(): out-of-bounds\n");
906 err = xnvme_znd_mgmt_send(&ctx, nsid, zslba, false,
907 XNVME_SPEC_ZND_CMD_MGMT_SEND_RESET, 0x0, NULL);
908 if (err || xnvme_cmd_ctx_cpl_status(&ctx)) {
909 err = err ? err : -EIO;
910 log_err("ioeng->reset_wp(): err(%d), sc(%d)", err, ctx.cpl.status.sc);
916 if (!td->io_ops_data) {
917 xnvme_dev_close(dev);
919 err_lock = pthread_mutex_unlock(&g_serialize);
921 log_err("ioeng->reset_wp(): pthread_mutex_unlock(), err(%d)\n", err_lock);
927 static int xnvme_fioe_get_file_size(struct thread_data *td, struct fio_file *f)
929 struct xnvme_opts opts = xnvme_opts_from_fioe(td);
930 struct xnvme_dev *dev;
933 if (fio_file_size_known(f))
936 ret = pthread_mutex_lock(&g_serialize);
938 log_err("ioeng->reset_wp(): pthread_mutex_lock(), err(%d)\n", ret);
942 dev = xnvme_dev_open(f->file_name, &opts);
944 log_err("%s: failed retrieving device handle, errno: %d\n", f->file_name, errno);
949 f->real_file_size = xnvme_dev_get_geo(dev)->tbytes;
950 fio_file_set_size_known(f);
951 f->filetype = FIO_TYPE_BLOCK;
954 xnvme_dev_close(dev);
955 err = pthread_mutex_unlock(&g_serialize);
957 log_err("ioeng->reset_wp(): pthread_mutex_unlock(), err(%d)\n", err);
962 FIO_STATIC struct ioengine_ops ioengine = {
964 .version = FIO_IOOPS_VERSION,
966 .option_struct_size = sizeof(struct xnvme_fioe_options),
967 .flags = FIO_DISKLESSIO | FIO_NODISKUTIL | FIO_NOEXTEND | FIO_MEMALIGN | FIO_RAWIO,
969 .cleanup = xnvme_fioe_cleanup,
970 .init = xnvme_fioe_init,
972 .iomem_free = xnvme_fioe_iomem_free,
973 .iomem_alloc = xnvme_fioe_iomem_alloc,
975 .io_u_free = xnvme_fioe_io_u_free,
976 .io_u_init = xnvme_fioe_io_u_init,
978 .event = xnvme_fioe_event,
979 .getevents = xnvme_fioe_getevents,
980 .queue = xnvme_fioe_queue,
982 .close_file = xnvme_fioe_close,
983 .open_file = xnvme_fioe_open,
984 .get_file_size = xnvme_fioe_get_file_size,
986 .invalidate = xnvme_fioe_invalidate,
987 .get_max_open_zones = xnvme_fioe_get_max_open_zones,
988 .get_zoned_model = xnvme_fioe_get_zoned_model,
989 .report_zones = xnvme_fioe_report_zones,
990 .reset_wp = xnvme_fioe_reset_wp,
993 static void fio_init fio_xnvme_register(void)
995 register_ioengine(&ioengine);
998 static void fio_exit fio_xnvme_unregister(void)
1000 unregister_ioengine(&ioengine);