2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2015 Netflix, Inc.
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer,
11 * without modification, immediately at the beginning of the file.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
17 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
18 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
19 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
20 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
21 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
22 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
23 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
24 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
25 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27 * Derived from ata_da.c:
28 * Copyright (c) 2009 Alexander Motin <mav@FreeBSD.org>
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
34 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/kernel.h>
40 #include <sys/sysctl.h>
41 #include <sys/taskqueue.h>
43 #include <sys/mutex.h>
45 #include <sys/devicestat.h>
46 #include <sys/eventhandler.h>
47 #include <sys/malloc.h>
50 #include <sys/reboot.h>
52 #include <geom/geom.h>
53 #include <geom/geom_disk.h>
62 #include <cam/cam_ccb.h>
63 #include <cam/cam_periph.h>
64 #include <cam/cam_xpt_periph.h>
65 #include <cam/cam_sim.h>
66 #include <cam/cam_iosched.h>
68 #include <cam/nvme/nvme_all.h>
75 NDA_FLAG_OPEN = 0x0001,
76 NDA_FLAG_DIRTY = 0x0002,
77 NDA_FLAG_SCTX_INIT = 0x0004,
79 #define NDA_FLAG_STRING \
90 #define NDA_Q_BIT_STRING \
95 NDA_CCB_BUFFER_IO = 0x01,
99 NDA_CCB_TYPE_MASK = 0x0F,
102 /* Offsets into our private area for storing information */
103 #define ccb_state ccb_h.ppriv_field0
104 #define ccb_bp ccb_h.ppriv_ptr1 /* For NDA_CCB_BUFFER_IO */
105 #define ccb_trim ccb_h.ppriv_ptr1 /* For NDA_CCB_TRIM */
108 struct cam_iosched_softc *cam_iosched;
109 int outstanding_cmds; /* Number of active commands */
110 int refcount; /* Active xpt_action() calls */
116 uint32_t nsid; /* Namespace ID for this nda device */
118 struct task sysctl_task;
119 struct sysctl_ctx_list sysctl_ctx;
120 struct sysctl_oid *sysctl_tree;
122 uint64_t trim_ranges;
124 #ifdef CAM_TEST_FAILURE
125 int force_read_error;
126 int force_write_error;
127 int periodic_read_error;
128 int periodic_read_count;
131 struct sysctl_ctx_list sysctl_stats_ctx;
132 struct sysctl_oid *sysctl_stats_tree;
139 struct nda_trim_request {
140 struct nvme_dsm_range dsm[NVME_MAX_DSM_TRIM / sizeof(struct nvme_dsm_range)];
141 TAILQ_HEAD(, bio) bps;
143 _Static_assert(NVME_MAX_DSM_TRIM % sizeof(struct nvme_dsm_range) == 0,
144 "NVME_MAX_DSM_TRIM must be an integral number of ranges");
146 /* Need quirk table */
148 static disk_ioctl_t ndaioctl;
149 static disk_strategy_t ndastrategy;
150 static dumper_t ndadump;
151 static periph_init_t ndainit;
152 static void ndaasync(void *callback_arg, u_int32_t code,
153 struct cam_path *path, void *arg);
154 static void ndasysctlinit(void *context, int pending);
155 static int ndaflagssysctl(SYSCTL_HANDLER_ARGS);
156 static periph_ctor_t ndaregister;
157 static periph_dtor_t ndacleanup;
158 static periph_start_t ndastart;
159 static periph_oninv_t ndaoninvalidate;
160 static void ndadone(struct cam_periph *periph,
161 union ccb *done_ccb);
162 static int ndaerror(union ccb *ccb, u_int32_t cam_flags,
163 u_int32_t sense_flags);
164 static void ndashutdown(void *arg, int howto);
165 static void ndasuspend(void *arg);
167 #ifndef NDA_DEFAULT_SEND_ORDERED
168 #define NDA_DEFAULT_SEND_ORDERED 1
170 #ifndef NDA_DEFAULT_TIMEOUT
171 #define NDA_DEFAULT_TIMEOUT 30 /* Timeout in seconds */
173 #ifndef NDA_DEFAULT_RETRY
174 #define NDA_DEFAULT_RETRY 4
176 #ifndef NDA_MAX_TRIM_ENTRIES
177 #define NDA_MAX_TRIM_ENTRIES (NVME_MAX_DSM_TRIM / sizeof(struct nvme_dsm_range))/* Number of DSM trims to use, max 256 */
180 static SYSCTL_NODE(_kern_cam, OID_AUTO, nda, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
181 "CAM Direct Access Disk driver");
183 //static int nda_retry_count = NDA_DEFAULT_RETRY;
184 static int nda_send_ordered = NDA_DEFAULT_SEND_ORDERED;
185 static int nda_default_timeout = NDA_DEFAULT_TIMEOUT;
186 static int nda_max_trim_entries = NDA_MAX_TRIM_ENTRIES;
187 static int nda_enable_biospeedup = 1;
188 SYSCTL_INT(_kern_cam_nda, OID_AUTO, max_trim, CTLFLAG_RDTUN,
189 &nda_max_trim_entries, NDA_MAX_TRIM_ENTRIES,
190 "Maximum number of BIO_DELETE to send down as a DSM TRIM.");
191 SYSCTL_INT(_kern_cam_nda, OID_AUTO, enable_biospeedup, CTLFLAG_RDTUN,
192 &nda_enable_biospeedup, 0, "Enable BIO_SPEEDUP processing");
195 * All NVMe media is non-rotational, so all nvme device instances
196 * share this to implement the sysctl.
198 static int nda_rotating_media = 0;
200 static struct periph_driver ndadriver =
203 TAILQ_HEAD_INITIALIZER(ndadriver.units), /* generation */ 0
206 PERIPHDRIVER_DECLARE(nda, ndadriver);
208 static MALLOC_DEFINE(M_NVMEDA, "nvme_da", "nvme_da buffers");
211 * nice wrappers. Maybe these belong in nvme_all.c instead of
212 * here, but this is the only place that uses these. Should
213 * we ever grow another NVME periph, we should move them
214 * all there wholesale.
218 nda_nvme_flush(struct nda_softc *softc, struct ccb_nvmeio *nvmeio)
220 cam_fill_nvmeio(nvmeio,
222 ndadone, /* cbfcnp */
223 CAM_DIR_NONE, /* flags */
226 nda_default_timeout * 1000); /* timeout 30s */
227 nvme_ns_flush_cmd(&nvmeio->cmd, softc->nsid);
231 nda_nvme_trim(struct nda_softc *softc, struct ccb_nvmeio *nvmeio,
232 void *payload, uint32_t num_ranges)
234 cam_fill_nvmeio(nvmeio,
236 ndadone, /* cbfcnp */
237 CAM_DIR_OUT, /* flags */
238 payload, /* data_ptr */
239 num_ranges * sizeof(struct nvme_dsm_range), /* dxfer_len */
240 nda_default_timeout * 1000); /* timeout 30s */
241 nvme_ns_trim_cmd(&nvmeio->cmd, softc->nsid, num_ranges);
245 nda_nvme_write(struct nda_softc *softc, struct ccb_nvmeio *nvmeio,
246 void *payload, uint64_t lba, uint32_t len, uint32_t count)
248 cam_fill_nvmeio(nvmeio,
250 ndadone, /* cbfcnp */
251 CAM_DIR_OUT, /* flags */
252 payload, /* data_ptr */
254 nda_default_timeout * 1000); /* timeout 30s */
255 nvme_ns_write_cmd(&nvmeio->cmd, softc->nsid, lba, count);
259 nda_nvme_rw_bio(struct nda_softc *softc, struct ccb_nvmeio *nvmeio,
260 struct bio *bp, uint32_t rwcmd)
262 int flags = rwcmd == NVME_OPC_READ ? CAM_DIR_IN : CAM_DIR_OUT;
267 if (bp->bio_flags & BIO_UNMAPPED) {
268 flags |= CAM_DATA_BIO;
271 payload = bp->bio_data;
274 lba = bp->bio_pblkno;
275 count = bp->bio_bcount / softc->disk->d_sectorsize;
277 cam_fill_nvmeio(nvmeio,
279 ndadone, /* cbfcnp */
281 payload, /* data_ptr */
282 bp->bio_bcount, /* dxfer_len */
283 nda_default_timeout * 1000); /* timeout 30s */
284 nvme_ns_rw_cmd(&nvmeio->cmd, rwcmd, softc->nsid, lba, count);
288 ndaopen(struct disk *dp)
290 struct cam_periph *periph;
291 struct nda_softc *softc;
294 periph = (struct cam_periph *)dp->d_drv1;
295 if (cam_periph_acquire(periph) != 0) {
299 cam_periph_lock(periph);
300 if ((error = cam_periph_hold(periph, PRIBIO|PCATCH)) != 0) {
301 cam_periph_unlock(periph);
302 cam_periph_release(periph);
306 CAM_DEBUG(periph->path, CAM_DEBUG_TRACE | CAM_DEBUG_PERIPH,
309 softc = (struct nda_softc *)periph->softc;
310 softc->flags |= NDA_FLAG_OPEN;
312 cam_periph_unhold(periph);
313 cam_periph_unlock(periph);
318 ndaclose(struct disk *dp)
320 struct cam_periph *periph;
321 struct nda_softc *softc;
325 periph = (struct cam_periph *)dp->d_drv1;
326 softc = (struct nda_softc *)periph->softc;
327 cam_periph_lock(periph);
329 CAM_DEBUG(periph->path, CAM_DEBUG_TRACE | CAM_DEBUG_PERIPH,
332 if ((softc->flags & NDA_FLAG_DIRTY) != 0 &&
333 (periph->flags & CAM_PERIPH_INVALID) == 0 &&
334 cam_periph_hold(periph, PRIBIO) == 0) {
336 ccb = cam_periph_getccb(periph, CAM_PRIORITY_NORMAL);
337 nda_nvme_flush(softc, &ccb->nvmeio);
338 error = cam_periph_runccb(ccb, ndaerror, /*cam_flags*/0,
339 /*sense_flags*/0, softc->disk->d_devstat);
342 xpt_print(periph->path, "Synchronize cache failed\n");
344 softc->flags &= ~NDA_FLAG_DIRTY;
345 xpt_release_ccb(ccb);
346 cam_periph_unhold(periph);
349 softc->flags &= ~NDA_FLAG_OPEN;
351 while (softc->refcount != 0)
352 cam_periph_sleep(periph, &softc->refcount, PRIBIO, "ndaclose", 1);
353 KASSERT(softc->outstanding_cmds == 0,
354 ("nda %d outstanding commands", softc->outstanding_cmds));
355 cam_periph_unlock(periph);
356 cam_periph_release(periph);
361 ndaschedule(struct cam_periph *periph)
363 struct nda_softc *softc = (struct nda_softc *)periph->softc;
365 if (softc->state != NDA_STATE_NORMAL)
368 cam_iosched_schedule(softc->cam_iosched, periph);
372 ndaioctl(struct disk *dp, u_long cmd, void *data, int fflag,
375 struct cam_periph *periph;
376 struct nda_softc *softc;
378 periph = (struct cam_periph *)dp->d_drv1;
379 softc = (struct nda_softc *)periph->softc;
385 * These don't map well to the underlying CCBs, so
386 * they are usupported via CAM.
391 struct nvme_get_nsid *gnsid = (struct nvme_get_nsid *)data;
392 struct ccb_pathinq cpi;
394 xpt_path_inq(&cpi, periph->path);
395 strncpy(gnsid->cdev, cpi.xport_specific.nvme.dev_name,
396 sizeof(gnsid->cdev));
397 gnsid->nsid = cpi.xport_specific.nvme.nsid;
400 case NVME_PASSTHROUGH_CMD:
402 struct nvme_pt_command *pt;
404 struct cam_periph_map_info mapinfo;
405 u_int maxmap = dp->d_maxsize;
409 * Create a NVME_IO CCB to do the passthrough command.
411 pt = (struct nvme_pt_command *)data;
412 ccb = xpt_alloc_ccb();
413 xpt_setup_ccb(&ccb->ccb_h, periph->path, CAM_PRIORITY_NORMAL);
414 ccb->ccb_state = NDA_CCB_PASS;
415 cam_fill_nvmeio(&ccb->nvmeio,
418 (pt->is_read ? CAM_DIR_IN : CAM_DIR_OUT) | CAM_DATA_VADDR,
421 nda_default_timeout * 1000);
422 memcpy(&ccb->nvmeio.cmd, &pt->cmd, sizeof(pt->cmd));
425 * Wire the user memory in this request for the I/O
427 memset(&mapinfo, 0, sizeof(mapinfo));
428 error = cam_periph_mapmem(ccb, &mapinfo, maxmap);
433 * Lock the periph and run the command.
435 cam_periph_lock(periph);
436 cam_periph_runccb(ccb, NULL, CAM_RETRY_SELTO,
437 SF_RETRY_UA | SF_NO_PRINT, NULL);
440 * Tear down mapping and return status.
442 cam_periph_unlock(periph);
443 cam_periph_unmapmem(ccb, &mapinfo);
444 error = (ccb->ccb_h.status == CAM_REQ_CMP) ? 0 : EIO;
446 cam_periph_lock(periph);
447 xpt_release_ccb(ccb);
448 cam_periph_unlock(periph);
458 * Actually translate the requested transfer into one the physical driver
459 * can understand. The transfer is described by a buf and will include
460 * only one physical transfer.
463 ndastrategy(struct bio *bp)
465 struct cam_periph *periph;
466 struct nda_softc *softc;
468 periph = (struct cam_periph *)bp->bio_disk->d_drv1;
469 softc = (struct nda_softc *)periph->softc;
471 cam_periph_lock(periph);
473 CAM_DEBUG(periph->path, CAM_DEBUG_TRACE, ("ndastrategy(%p)\n", bp));
476 * If the device has been made invalid, error out
478 if ((periph->flags & CAM_PERIPH_INVALID) != 0) {
479 cam_periph_unlock(periph);
480 biofinish(bp, NULL, ENXIO);
484 if (bp->bio_cmd == BIO_DELETE)
488 * Place it in the queue of disk activities for this disk
490 cam_iosched_queue_work(softc->cam_iosched, bp);
493 * Schedule ourselves for performing the work.
496 cam_periph_unlock(periph);
502 ndadump(void *arg, void *virtual, vm_offset_t physical, off_t offset, size_t length)
504 struct cam_periph *periph;
505 struct nda_softc *softc;
507 struct ccb_nvmeio nvmeio;
515 softc = (struct nda_softc *)periph->softc;
516 secsize = softc->disk->d_sectorsize;
517 lba = offset / secsize;
518 count = length / secsize;
520 if ((periph->flags & CAM_PERIPH_INVALID) != 0)
523 /* xpt_get_ccb returns a zero'd allocation for the ccb, mimic that here */
524 memset(&nvmeio, 0, sizeof(nvmeio));
526 xpt_setup_ccb(&nvmeio.ccb_h, periph->path, CAM_PRIORITY_NORMAL);
527 nvmeio.ccb_state = NDA_CCB_DUMP;
528 nda_nvme_write(softc, &nvmeio, virtual, lba, length, count);
529 error = cam_periph_runccb((union ccb *)&nvmeio, cam_periph_error,
530 0, SF_NO_RECOVERY | SF_NO_RETRY, NULL);
532 printf("Aborting dump due to I/O error %d.\n", error);
538 xpt_setup_ccb(&nvmeio.ccb_h, periph->path, CAM_PRIORITY_NORMAL);
540 nvmeio.ccb_state = NDA_CCB_DUMP;
541 nda_nvme_flush(softc, &nvmeio);
542 error = cam_periph_runccb((union ccb *)&nvmeio, cam_periph_error,
543 0, SF_NO_RECOVERY | SF_NO_RETRY, NULL);
545 xpt_print(periph->path, "flush cmd failed\n");
555 * Install a global async callback. This callback will
556 * receive async callbacks like "new device found".
558 status = xpt_register_async(AC_FOUND_DEVICE, ndaasync, NULL, NULL);
560 if (status != CAM_REQ_CMP) {
561 printf("nda: Failed to attach master async callback "
562 "due to status 0x%x!\n", status);
563 } else if (nda_send_ordered) {
565 /* Register our event handlers */
566 if ((EVENTHANDLER_REGISTER(power_suspend, ndasuspend,
567 NULL, EVENTHANDLER_PRI_LAST)) == NULL)
568 printf("ndainit: power event registration failed!\n");
569 if ((EVENTHANDLER_REGISTER(shutdown_post_sync, ndashutdown,
570 NULL, SHUTDOWN_PRI_DEFAULT)) == NULL)
571 printf("ndainit: shutdown event registration failed!\n");
576 * Callback from GEOM, called when it has finished cleaning up its
580 ndadiskgonecb(struct disk *dp)
582 struct cam_periph *periph;
584 periph = (struct cam_periph *)dp->d_drv1;
586 cam_periph_release(periph);
590 ndaoninvalidate(struct cam_periph *periph)
592 struct nda_softc *softc;
594 softc = (struct nda_softc *)periph->softc;
597 * De-register any async callbacks.
599 xpt_register_async(0, ndaasync, periph, periph->path);
601 softc->invalidations++;
605 * Return all queued I/O with ENXIO.
606 * XXX Handle any transactions queued to the card
607 * with XPT_ABORT_CCB.
609 cam_iosched_flush(softc->cam_iosched, NULL, ENXIO);
611 disk_gone(softc->disk);
615 ndacleanup(struct cam_periph *periph)
617 struct nda_softc *softc;
619 softc = (struct nda_softc *)periph->softc;
621 cam_periph_unlock(periph);
623 cam_iosched_fini(softc->cam_iosched);
626 * If we can't free the sysctl tree, oh well...
628 if ((softc->flags & NDA_FLAG_SCTX_INIT) != 0) {
630 if (sysctl_ctx_free(&softc->sysctl_stats_ctx) != 0)
631 xpt_print(periph->path,
632 "can't remove sysctl stats context\n");
634 if (sysctl_ctx_free(&softc->sysctl_ctx) != 0)
635 xpt_print(periph->path,
636 "can't remove sysctl context\n");
639 disk_destroy(softc->disk);
640 free(softc, M_DEVBUF);
641 cam_periph_lock(periph);
645 ndaasync(void *callback_arg, u_int32_t code,
646 struct cam_path *path, void *arg)
648 struct cam_periph *periph;
650 periph = (struct cam_periph *)callback_arg;
652 case AC_FOUND_DEVICE:
654 struct ccb_getdev *cgd;
657 cgd = (struct ccb_getdev *)arg;
661 if (cgd->protocol != PROTO_NVME)
665 * Allocate a peripheral instance for
666 * this device and start the probe
669 status = cam_periph_alloc(ndaregister, ndaoninvalidate,
670 ndacleanup, ndastart,
671 "nda", CAM_PERIPH_BIO,
673 AC_FOUND_DEVICE, cgd);
675 if (status != CAM_REQ_CMP
676 && status != CAM_REQ_INPROG)
677 printf("ndaasync: Unable to attach to new device "
678 "due to status 0x%x\n", status);
681 case AC_ADVINFO_CHANGED:
685 buftype = (uintptr_t)arg;
686 if (buftype == CDAI_TYPE_PHYS_PATH) {
687 struct nda_softc *softc;
689 softc = periph->softc;
690 disk_attr_changed(softc->disk, "GEOM::physpath",
697 cam_periph_async(periph, code, path, arg);
703 ndasysctlinit(void *context, int pending)
705 struct cam_periph *periph;
706 struct nda_softc *softc;
707 char tmpstr[32], tmpstr2[16];
709 periph = (struct cam_periph *)context;
711 /* periph was held for us when this task was enqueued */
712 if ((periph->flags & CAM_PERIPH_INVALID) != 0) {
713 cam_periph_release(periph);
717 softc = (struct nda_softc *)periph->softc;
718 snprintf(tmpstr, sizeof(tmpstr), "CAM NDA unit %d", periph->unit_number);
719 snprintf(tmpstr2, sizeof(tmpstr2), "%d", periph->unit_number);
721 sysctl_ctx_init(&softc->sysctl_ctx);
722 softc->flags |= NDA_FLAG_SCTX_INIT;
723 softc->sysctl_tree = SYSCTL_ADD_NODE_WITH_LABEL(&softc->sysctl_ctx,
724 SYSCTL_STATIC_CHILDREN(_kern_cam_nda), OID_AUTO, tmpstr2,
725 CTLFLAG_RD | CTLFLAG_MPSAFE, 0, tmpstr, "device_index");
726 if (softc->sysctl_tree == NULL) {
727 printf("ndasysctlinit: unable to allocate sysctl tree\n");
728 cam_periph_release(periph);
732 SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
733 OID_AUTO, "unmapped_io", CTLFLAG_RD,
734 &softc->unmappedio, 0, "Unmapped I/O leaf");
736 SYSCTL_ADD_QUAD(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
737 OID_AUTO, "deletes", CTLFLAG_RD,
738 &softc->deletes, "Number of BIO_DELETE requests");
740 SYSCTL_ADD_UQUAD(&softc->sysctl_ctx,
741 SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
742 "trim_count", CTLFLAG_RD, &softc->trim_count,
743 "Total number of unmap/dsm commands sent");
744 SYSCTL_ADD_UQUAD(&softc->sysctl_ctx,
745 SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
746 "trim_ranges", CTLFLAG_RD, &softc->trim_ranges,
747 "Total number of ranges in unmap/dsm commands");
748 SYSCTL_ADD_UQUAD(&softc->sysctl_ctx,
749 SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
750 "trim_lbas", CTLFLAG_RD, &softc->trim_lbas,
751 "Total lbas in the unmap/dsm commands sent");
753 SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
754 OID_AUTO, "rotating", CTLFLAG_RD, &nda_rotating_media, 1,
757 SYSCTL_ADD_PROC(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
758 OID_AUTO, "flags", CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE,
759 softc, 0, ndaflagssysctl, "A",
763 softc->sysctl_stats_tree = SYSCTL_ADD_NODE(&softc->sysctl_stats_ctx,
764 SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO, "stats",
765 CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "Statistics");
766 if (softc->sysctl_stats_tree == NULL) {
767 printf("ndasysctlinit: unable to allocate sysctl tree for stats\n");
768 cam_periph_release(periph);
771 SYSCTL_ADD_INT(&softc->sysctl_stats_ctx,
772 SYSCTL_CHILDREN(softc->sysctl_stats_tree),
773 OID_AUTO, "timeouts", CTLFLAG_RD,
775 "Device timeouts reported by the SIM");
776 SYSCTL_ADD_INT(&softc->sysctl_stats_ctx,
777 SYSCTL_CHILDREN(softc->sysctl_stats_tree),
778 OID_AUTO, "errors", CTLFLAG_RD,
780 "Transport errors reported by the SIM.");
781 SYSCTL_ADD_INT(&softc->sysctl_stats_ctx,
782 SYSCTL_CHILDREN(softc->sysctl_stats_tree),
783 OID_AUTO, "pack_invalidations", CTLFLAG_RD,
784 &softc->invalidations, 0,
785 "Device pack invalidations.");
788 #ifdef CAM_TEST_FAILURE
789 SYSCTL_ADD_PROC(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
790 OID_AUTO, "invalidate", CTLTYPE_U64 | CTLFLAG_RW | CTLFLAG_MPSAFE,
791 periph, 0, cam_periph_invalidate_sysctl, "I",
792 "Write 1 to invalidate the drive immediately");
795 cam_iosched_sysctl_init(softc->cam_iosched, &softc->sysctl_ctx,
798 cam_periph_release(periph);
802 ndaflagssysctl(SYSCTL_HANDLER_ARGS)
805 struct nda_softc *softc = arg1;
808 sbuf_new_for_sysctl(&sbuf, NULL, 0, req);
809 if (softc->flags != 0)
810 sbuf_printf(&sbuf, "0x%b", (unsigned)softc->flags, NDA_FLAG_STRING);
812 sbuf_printf(&sbuf, "0");
813 error = sbuf_finish(&sbuf);
820 ndagetattr(struct bio *bp)
823 struct cam_periph *periph;
825 if (g_handleattr_int(bp, "GEOM::canspeedup", nda_enable_biospeedup))
826 return (EJUSTRETURN);
828 periph = (struct cam_periph *)bp->bio_disk->d_drv1;
829 cam_periph_lock(periph);
830 ret = xpt_getattr(bp->bio_data, bp->bio_length, bp->bio_attribute,
832 cam_periph_unlock(periph);
834 bp->bio_completed = bp->bio_length;
839 ndaregister(struct cam_periph *periph, void *arg)
841 struct nda_softc *softc;
843 struct ccb_pathinq cpi;
844 const struct nvme_namespace_data *nsd;
845 const struct nvme_controller_data *cd;
846 char announce_buf[80];
847 uint8_t flbas_fmt, lbads, vwc_present;
851 nsd = nvme_get_identify_ns(periph);
852 cd = nvme_get_identify_cntrl(periph);
854 softc = (struct nda_softc *)malloc(sizeof(*softc), M_DEVBUF,
858 printf("ndaregister: Unable to probe new device. "
859 "Unable to allocate softc\n");
860 return(CAM_REQ_CMP_ERR);
863 if (cam_iosched_init(&softc->cam_iosched, periph) != 0) {
864 printf("ndaregister: Unable to probe new device. "
865 "Unable to allocate iosched memory\n");
866 free(softc, M_DEVBUF);
867 return(CAM_REQ_CMP_ERR);
870 /* ident_data parsing */
872 periph->softc = softc;
873 softc->quirks = NDA_Q_NONE;
874 xpt_path_inq(&cpi, periph->path);
875 TASK_INIT(&softc->sysctl_task, 0, ndasysctlinit, periph);
878 * The name space ID is the lun, save it for later I/O
880 softc->nsid = (uint32_t)xpt_path_lun_id(periph->path);
883 * Register this media as a disk
885 (void)cam_periph_hold(periph, PRIBIO);
886 cam_periph_unlock(periph);
887 snprintf(announce_buf, sizeof(announce_buf),
888 "kern.cam.nda.%d.quirks", periph->unit_number);
889 quirks = softc->quirks;
890 TUNABLE_INT_FETCH(announce_buf, &quirks);
891 softc->quirks = quirks;
892 cam_iosched_set_sort_queue(softc->cam_iosched, 0);
893 softc->disk = disk = disk_alloc();
894 disk->d_rotation_rate = DISK_RR_NON_ROTATING;
895 disk->d_open = ndaopen;
896 disk->d_close = ndaclose;
897 disk->d_strategy = ndastrategy;
898 disk->d_ioctl = ndaioctl;
899 disk->d_getattr = ndagetattr;
900 disk->d_dump = ndadump;
901 disk->d_gone = ndadiskgonecb;
902 disk->d_name = "nda";
903 disk->d_drv1 = periph;
904 disk->d_unit = periph->unit_number;
905 maxio = cpi.maxio; /* Honor max I/O size of SIM */
907 maxio = DFLTPHYS; /* traditional default */
908 else if (maxio > MAXPHYS)
909 maxio = MAXPHYS; /* for safety */
910 disk->d_maxsize = maxio;
911 flbas_fmt = (nsd->flbas >> NVME_NS_DATA_FLBAS_FORMAT_SHIFT) &
912 NVME_NS_DATA_FLBAS_FORMAT_MASK;
913 lbads = (nsd->lbaf[flbas_fmt] >> NVME_NS_DATA_LBAF_LBADS_SHIFT) &
914 NVME_NS_DATA_LBAF_LBADS_MASK;
915 disk->d_sectorsize = 1 << lbads;
916 disk->d_mediasize = (off_t)(disk->d_sectorsize * nsd->nsze);
917 disk->d_delmaxsize = disk->d_mediasize;
918 disk->d_flags = DISKFLAG_DIRECT_COMPLETION;
919 if (nvme_ctrlr_has_dataset_mgmt(cd))
920 disk->d_flags |= DISKFLAG_CANDELETE;
921 vwc_present = (cd->vwc >> NVME_CTRLR_DATA_VWC_PRESENT_SHIFT) &
922 NVME_CTRLR_DATA_VWC_PRESENT_MASK;
924 disk->d_flags |= DISKFLAG_CANFLUSHCACHE;
925 if ((cpi.hba_misc & PIM_UNMAPPED) != 0) {
926 disk->d_flags |= DISKFLAG_UNMAPPED_BIO;
927 softc->unmappedio = 1;
930 * d_ident and d_descr are both far bigger than the length of either
931 * the serial or model number strings.
933 cam_strvis(disk->d_descr, cd->mn,
934 NVME_MODEL_NUMBER_LENGTH, sizeof(disk->d_descr));
935 cam_strvis(disk->d_ident, cd->sn,
936 NVME_SERIAL_NUMBER_LENGTH, sizeof(disk->d_ident));
937 disk->d_hba_vendor = cpi.hba_vendor;
938 disk->d_hba_device = cpi.hba_device;
939 disk->d_hba_subvendor = cpi.hba_subvendor;
940 disk->d_hba_subdevice = cpi.hba_subdevice;
941 snprintf(disk->d_attachment, sizeof(disk->d_attachment),
942 "%s%d", cpi.dev_name, cpi.unit_number);
943 disk->d_stripesize = disk->d_sectorsize;
944 disk->d_stripeoffset = 0;
945 disk->d_devstat = devstat_new_entry(periph->periph_name,
946 periph->unit_number, disk->d_sectorsize,
947 DEVSTAT_ALL_SUPPORTED,
948 DEVSTAT_TYPE_DIRECT | XPORT_DEVSTAT_TYPE(cpi.transport),
949 DEVSTAT_PRIORITY_DISK);
951 * Add alias for older nvd drives to ease transition.
953 disk_add_alias(disk, "nvd");
956 * Acquire a reference to the periph before we register with GEOM.
957 * We'll release this reference once GEOM calls us back (via
958 * ndadiskgonecb()) telling us that our provider has been freed.
960 if (cam_periph_acquire(periph) != 0) {
961 xpt_print(periph->path, "%s: lost periph during "
962 "registration!\n", __func__);
963 cam_periph_lock(periph);
964 return (CAM_REQ_CMP_ERR);
966 disk_create(softc->disk, DISK_VERSION);
967 cam_periph_lock(periph);
968 cam_periph_unhold(periph);
970 snprintf(announce_buf, sizeof(announce_buf),
971 "%juMB (%ju %u byte sectors)",
972 (uintmax_t)((uintmax_t)disk->d_mediasize / (1024*1024)),
973 (uintmax_t)disk->d_mediasize / disk->d_sectorsize,
975 xpt_announce_periph(periph, announce_buf);
976 xpt_announce_quirks(periph, softc->quirks, NDA_Q_BIT_STRING);
979 * Create our sysctl variables, now that we know
980 * we have successfully attached.
982 if (cam_periph_acquire(periph) == 0)
983 taskqueue_enqueue(taskqueue_thread, &softc->sysctl_task);
986 * Register for device going away and info about the drive
987 * changing (though with NVMe, it can't)
989 xpt_register_async(AC_LOST_DEVICE | AC_ADVINFO_CHANGED,
990 ndaasync, periph, periph->path);
992 softc->state = NDA_STATE_NORMAL;
997 ndastart(struct cam_periph *periph, union ccb *start_ccb)
999 struct nda_softc *softc = (struct nda_softc *)periph->softc;
1000 struct ccb_nvmeio *nvmeio = &start_ccb->nvmeio;
1002 CAM_DEBUG(periph->path, CAM_DEBUG_TRACE, ("ndastart\n"));
1004 switch (softc->state) {
1005 case NDA_STATE_NORMAL:
1009 bp = cam_iosched_next_bio(softc->cam_iosched);
1010 CAM_DEBUG(periph->path, CAM_DEBUG_TRACE, ("ndastart: bio %p\n", bp));
1012 xpt_release_ccb(start_ccb);
1016 switch (bp->bio_cmd) {
1018 softc->flags |= NDA_FLAG_DIRTY;
1022 #ifdef CAM_TEST_FAILURE
1026 * Support the failure ioctls. If the command is a
1027 * read, and there are pending forced read errors, or
1028 * if a write and pending write errors, then fail this
1029 * operation with EIO. This is useful for testing
1030 * purposes. Also, support having every Nth read fail.
1032 * This is a rather blunt tool.
1034 if (bp->bio_cmd == BIO_READ) {
1035 if (softc->force_read_error) {
1036 softc->force_read_error--;
1039 if (softc->periodic_read_error > 0) {
1040 if (++softc->periodic_read_count >=
1041 softc->periodic_read_error) {
1042 softc->periodic_read_count = 0;
1047 if (softc->force_write_error) {
1048 softc->force_write_error--;
1053 biofinish(bp, NULL, EIO);
1054 xpt_release_ccb(start_ccb);
1055 ndaschedule(periph);
1059 KASSERT((bp->bio_flags & BIO_UNMAPPED) == 0 ||
1060 round_page(bp->bio_bcount + bp->bio_ma_offset) /
1061 PAGE_SIZE == bp->bio_ma_n,
1062 ("Short bio %p", bp));
1063 nda_nvme_rw_bio(softc, &start_ccb->nvmeio, bp, bp->bio_cmd == BIO_READ ?
1064 NVME_OPC_READ : NVME_OPC_WRITE);
1069 struct nvme_dsm_range *dsm_range, *dsm_end;
1070 struct nda_trim_request *trim;
1073 uint32_t totalcount = 0, ranges = 0;
1075 trim = malloc(sizeof(*trim), M_NVMEDA, M_ZERO | M_NOWAIT);
1077 biofinish(bp, NULL, ENOMEM);
1078 xpt_release_ccb(start_ccb);
1079 ndaschedule(periph);
1082 TAILQ_INIT(&trim->bps);
1084 ents = min(nitems(trim->dsm), nda_max_trim_entries);
1085 dsm_range = trim->dsm;
1086 dsm_end = dsm_range + ents;
1088 TAILQ_INSERT_TAIL(&trim->bps, bp1, bio_queue);
1090 htole32(bp1->bio_bcount / softc->disk->d_sectorsize);
1091 dsm_range->starting_lba =
1092 htole64(bp1->bio_offset / softc->disk->d_sectorsize);
1094 totalcount += dsm_range->length;
1096 if (dsm_range >= dsm_end)
1098 bp1 = cam_iosched_next_trim(softc->cam_iosched);
1099 /* XXX -- Could collapse adjacent ranges, but we don't for now */
1100 /* XXX -- Could limit based on total payload size */
1101 } while (bp1 != NULL);
1102 start_ccb->ccb_trim = trim;
1103 nda_nvme_trim(softc, &start_ccb->nvmeio, trim->dsm,
1104 dsm_range - trim->dsm);
1105 start_ccb->ccb_state = NDA_CCB_TRIM;
1106 softc->trim_count++;
1107 softc->trim_ranges += ranges;
1108 softc->trim_lbas += totalcount;
1110 * Note: We can have multiple TRIMs in flight, so we don't call
1111 * cam_iosched_submit_trim(softc->cam_iosched);
1112 * since that forces the I/O scheduler to only schedule one at a time.
1113 * On NVMe drives, this is a performance disaster.
1118 nda_nvme_flush(softc, nvmeio);
1121 biofinish(bp, NULL, EOPNOTSUPP);
1122 xpt_release_ccb(start_ccb);
1123 ndaschedule(periph);
1126 start_ccb->ccb_state = NDA_CCB_BUFFER_IO;
1127 start_ccb->ccb_bp = bp;
1129 start_ccb->ccb_h.flags |= CAM_UNLOCKED;
1130 softc->outstanding_cmds++;
1131 softc->refcount++; /* For submission only */
1132 cam_periph_unlock(periph);
1133 xpt_action(start_ccb);
1134 cam_periph_lock(periph);
1135 softc->refcount--; /* Submission done */
1137 /* May have more work to do, so ensure we stay scheduled */
1138 ndaschedule(periph);
1145 ndadone(struct cam_periph *periph, union ccb *done_ccb)
1147 struct nda_softc *softc;
1148 struct ccb_nvmeio *nvmeio = &done_ccb->nvmeio;
1149 struct cam_path *path;
1152 softc = (struct nda_softc *)periph->softc;
1153 path = done_ccb->ccb_h.path;
1155 CAM_DEBUG(path, CAM_DEBUG_TRACE, ("ndadone\n"));
1157 state = nvmeio->ccb_state & NDA_CCB_TYPE_MASK;
1159 case NDA_CCB_BUFFER_IO:
1164 cam_periph_lock(periph);
1165 if ((done_ccb->ccb_h.status & CAM_STATUS_MASK) != CAM_REQ_CMP) {
1166 error = ndaerror(done_ccb, 0, 0);
1167 if (error == ERESTART) {
1168 /* A retry was scheduled, so just return. */
1169 cam_periph_unlock(periph);
1172 if ((done_ccb->ccb_h.status & CAM_DEV_QFRZN) != 0)
1173 cam_release_devq(path,
1177 /*getcount_only*/0);
1179 if ((done_ccb->ccb_h.status & CAM_DEV_QFRZN) != 0)
1180 panic("REQ_CMP with QFRZN");
1183 if (state == NDA_CCB_BUFFER_IO) {
1186 bp = (struct bio *)done_ccb->ccb_bp;
1187 bp->bio_error = error;
1189 bp->bio_resid = bp->bio_bcount;
1190 bp->bio_flags |= BIO_ERROR;
1194 softc->outstanding_cmds--;
1197 * We need to call cam_iosched before we call biodone so that we
1198 * don't measure any activity that happens in the completion
1199 * routine, which in the case of sendfile can be quite
1202 cam_iosched_bio_complete(softc->cam_iosched, bp, done_ccb);
1203 xpt_release_ccb(done_ccb);
1204 ndaschedule(periph);
1205 cam_periph_unlock(periph);
1207 } else { /* state == NDA_CCB_TRIM */
1208 struct nda_trim_request *trim;
1209 struct bio *bp1, *bp2;
1210 TAILQ_HEAD(, bio) queue;
1212 trim = nvmeio->ccb_trim;
1214 TAILQ_CONCAT(&queue, &trim->bps, bio_queue);
1215 free(trim, M_NVMEDA);
1218 * Since we can have multiple trims in flight, we don't
1219 * need to call this here.
1220 * cam_iosched_trim_done(softc->cam_iosched);
1223 * The the I/O scheduler that we're finishing the I/O
1224 * so we can keep book. The first one we pass in the CCB
1225 * which has the timing information. The rest we pass in NULL
1226 * so we can keep proper counts.
1228 bp1 = TAILQ_FIRST(&queue);
1229 cam_iosched_bio_complete(softc->cam_iosched, bp1, done_ccb);
1230 xpt_release_ccb(done_ccb);
1231 softc->outstanding_cmds--;
1232 ndaschedule(periph);
1233 cam_periph_unlock(periph);
1234 while ((bp2 = TAILQ_FIRST(&queue)) != NULL) {
1235 TAILQ_REMOVE(&queue, bp2, bio_queue);
1236 bp2->bio_error = error;
1238 bp2->bio_flags |= BIO_ERROR;
1239 bp2->bio_resid = bp1->bio_bcount;
1243 cam_iosched_bio_complete(softc->cam_iosched, bp2, NULL);
1250 /* No-op. We're polling */
1257 xpt_release_ccb(done_ccb);
1261 ndaerror(union ccb *ccb, u_int32_t cam_flags, u_int32_t sense_flags)
1263 struct nda_softc *softc;
1264 struct cam_periph *periph;
1266 periph = xpt_path_periph(ccb->ccb_h.path);
1267 softc = (struct nda_softc *)periph->softc;
1269 switch (ccb->ccb_h.status & CAM_STATUS_MASK) {
1270 case CAM_CMD_TIMEOUT:
1275 case CAM_REQ_ABORTED:
1276 case CAM_REQ_CMP_ERR:
1277 case CAM_REQ_TERMIO:
1278 case CAM_UNREC_HBA_ERROR:
1279 case CAM_DATA_RUN_ERR:
1280 case CAM_ATA_STATUS_ERROR:
1289 return(cam_periph_error(ccb, cam_flags, sense_flags));
1293 * Step through all NDA peripheral drivers, and if the device is still open,
1294 * sync the disk cache to physical media.
1299 struct cam_periph *periph;
1300 struct nda_softc *softc;
1304 CAM_PERIPH_FOREACH(periph, &ndadriver) {
1305 softc = (struct nda_softc *)periph->softc;
1307 if (SCHEDULER_STOPPED()) {
1309 * If we paniced with the lock held or the periph is not
1310 * open, do not recurse. Otherwise, call ndadump since
1311 * that avoids the sleeping cam_periph_getccb does if no
1312 * CCBs are available.
1314 if (!cam_periph_owned(periph) &&
1315 (softc->flags & NDA_FLAG_OPEN)) {
1316 ndadump(softc->disk, NULL, 0, 0, 0);
1322 * We only sync the cache if the drive is still open
1324 cam_periph_lock(periph);
1325 if ((softc->flags & NDA_FLAG_OPEN) == 0) {
1326 cam_periph_unlock(periph);
1330 ccb = cam_periph_getccb(periph, CAM_PRIORITY_NORMAL);
1331 nda_nvme_flush(softc, &ccb->nvmeio);
1332 error = cam_periph_runccb(ccb, ndaerror, /*cam_flags*/0,
1333 /*sense_flags*/ SF_NO_RECOVERY | SF_NO_RETRY,
1334 softc->disk->d_devstat);
1336 xpt_print(periph->path, "Synchronize cache failed\n");
1337 xpt_release_ccb(ccb);
1338 cam_periph_unlock(periph);
1343 ndashutdown(void *arg, int howto)
1350 ndasuspend(void *arg)