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>
51 #include <geom/geom_disk.h>
60 #include <cam/cam_ccb.h>
61 #include <cam/cam_periph.h>
62 #include <cam/cam_xpt_periph.h>
63 #include <cam/cam_sim.h>
64 #include <cam/cam_iosched.h>
66 #include <cam/nvme/nvme_all.h>
73 NDA_FLAG_OPEN = 0x0001,
74 NDA_FLAG_DIRTY = 0x0002,
75 NDA_FLAG_SCTX_INIT = 0x0004,
83 #define NDA_Q_BIT_STRING \
88 NDA_CCB_BUFFER_IO = 0x01,
91 NDA_CCB_TYPE_MASK = 0x0F,
94 /* Offsets into our private area for storing information */
95 #define ccb_state ccb_h.ppriv_field0
96 #define ccb_bp ccb_h.ppriv_ptr1 /* For NDA_CCB_BUFFER_IO */
97 #define ccb_trim ccb_h.ppriv_ptr1 /* For NDA_CCB_TRIM */
100 struct cam_iosched_softc *cam_iosched;
101 int outstanding_cmds; /* Number of active commands */
102 int refcount; /* Active xpt_action() calls */
108 uint32_t nsid; /* Namespace ID for this nda device */
110 struct task sysctl_task;
111 struct sysctl_ctx_list sysctl_ctx;
112 struct sysctl_oid *sysctl_tree;
114 uint64_t trim_ranges;
116 #ifdef CAM_TEST_FAILURE
117 int force_read_error;
118 int force_write_error;
119 int periodic_read_error;
120 int periodic_read_count;
123 struct sysctl_ctx_list sysctl_stats_ctx;
124 struct sysctl_oid *sysctl_stats_tree;
131 struct nda_trim_request {
132 struct nvme_dsm_range dsm[NVME_MAX_DSM_TRIM / sizeof(struct nvme_dsm_range)];
133 TAILQ_HEAD(, bio) bps;
135 _Static_assert(NVME_MAX_DSM_TRIM % sizeof(struct nvme_dsm_range) == 0,
136 "NVME_MAX_DSM_TRIM must be an integral number of ranges");
138 /* Need quirk table */
140 static disk_strategy_t ndastrategy;
141 static dumper_t ndadump;
142 static periph_init_t ndainit;
143 static void ndaasync(void *callback_arg, u_int32_t code,
144 struct cam_path *path, void *arg);
145 static void ndasysctlinit(void *context, int pending);
146 static periph_ctor_t ndaregister;
147 static periph_dtor_t ndacleanup;
148 static periph_start_t ndastart;
149 static periph_oninv_t ndaoninvalidate;
150 static void ndadone(struct cam_periph *periph,
151 union ccb *done_ccb);
152 static int ndaerror(union ccb *ccb, u_int32_t cam_flags,
153 u_int32_t sense_flags);
154 static void ndashutdown(void *arg, int howto);
155 static void ndasuspend(void *arg);
157 #ifndef NDA_DEFAULT_SEND_ORDERED
158 #define NDA_DEFAULT_SEND_ORDERED 1
160 #ifndef NDA_DEFAULT_TIMEOUT
161 #define NDA_DEFAULT_TIMEOUT 30 /* Timeout in seconds */
163 #ifndef NDA_DEFAULT_RETRY
164 #define NDA_DEFAULT_RETRY 4
166 #ifndef NDA_MAX_TRIM_ENTRIES
167 #define NDA_MAX_TRIM_ENTRIES (NVME_MAX_DSM_TRIM / sizeof(struct nvme_dsm_range))/* Number of DSM trims to use, max 256 */
170 static SYSCTL_NODE(_kern_cam, OID_AUTO, nda, CTLFLAG_RD, 0,
171 "CAM Direct Access Disk driver");
173 //static int nda_retry_count = NDA_DEFAULT_RETRY;
174 static int nda_send_ordered = NDA_DEFAULT_SEND_ORDERED;
175 static int nda_default_timeout = NDA_DEFAULT_TIMEOUT;
176 static int nda_max_trim_entries = NDA_MAX_TRIM_ENTRIES;
177 SYSCTL_INT(_kern_cam_nda, OID_AUTO, max_trim, CTLFLAG_RDTUN,
178 &nda_max_trim_entries, NDA_MAX_TRIM_ENTRIES,
179 "Maximum number of BIO_DELETE to send down as a DSM TRIM.");
182 * All NVMe media is non-rotational, so all nvme device instances
183 * share this to implement the sysctl.
185 static int nda_rotating_media = 0;
187 static struct periph_driver ndadriver =
190 TAILQ_HEAD_INITIALIZER(ndadriver.units), /* generation */ 0
193 PERIPHDRIVER_DECLARE(nda, ndadriver);
195 static MALLOC_DEFINE(M_NVMEDA, "nvme_da", "nvme_da buffers");
198 * nice wrappers. Maybe these belong in nvme_all.c instead of
199 * here, but this is the only place that uses these. Should
200 * we ever grow another NVME periph, we should move them
201 * all there wholesale.
205 nda_nvme_flush(struct nda_softc *softc, struct ccb_nvmeio *nvmeio)
207 cam_fill_nvmeio(nvmeio,
209 ndadone, /* cbfcnp */
210 CAM_DIR_NONE, /* flags */
213 nda_default_timeout * 1000); /* timeout 30s */
214 nvme_ns_flush_cmd(&nvmeio->cmd, softc->nsid);
218 nda_nvme_trim(struct nda_softc *softc, struct ccb_nvmeio *nvmeio,
219 void *payload, uint32_t num_ranges)
221 cam_fill_nvmeio(nvmeio,
223 ndadone, /* cbfcnp */
224 CAM_DIR_OUT, /* flags */
225 payload, /* data_ptr */
226 num_ranges * sizeof(struct nvme_dsm_range), /* dxfer_len */
227 nda_default_timeout * 1000); /* timeout 30s */
228 nvme_ns_trim_cmd(&nvmeio->cmd, softc->nsid, num_ranges);
232 nda_nvme_write(struct nda_softc *softc, struct ccb_nvmeio *nvmeio,
233 void *payload, uint64_t lba, uint32_t len, uint32_t count)
235 cam_fill_nvmeio(nvmeio,
237 ndadone, /* cbfcnp */
238 CAM_DIR_OUT, /* flags */
239 payload, /* data_ptr */
241 nda_default_timeout * 1000); /* timeout 30s */
242 nvme_ns_write_cmd(&nvmeio->cmd, softc->nsid, lba, count);
246 nda_nvme_rw_bio(struct nda_softc *softc, struct ccb_nvmeio *nvmeio,
247 struct bio *bp, uint32_t rwcmd)
249 int flags = rwcmd == NVME_OPC_READ ? CAM_DIR_IN : CAM_DIR_OUT;
254 if (bp->bio_flags & BIO_UNMAPPED) {
255 flags |= CAM_DATA_BIO;
258 payload = bp->bio_data;
261 lba = bp->bio_pblkno;
262 count = bp->bio_bcount / softc->disk->d_sectorsize;
264 cam_fill_nvmeio(nvmeio,
266 ndadone, /* cbfcnp */
268 payload, /* data_ptr */
269 bp->bio_bcount, /* dxfer_len */
270 nda_default_timeout * 1000); /* timeout 30s */
271 nvme_ns_rw_cmd(&nvmeio->cmd, rwcmd, softc->nsid, lba, count);
275 ndaopen(struct disk *dp)
277 struct cam_periph *periph;
278 struct nda_softc *softc;
281 periph = (struct cam_periph *)dp->d_drv1;
282 if (cam_periph_acquire(periph) != 0) {
286 cam_periph_lock(periph);
287 if ((error = cam_periph_hold(periph, PRIBIO|PCATCH)) != 0) {
288 cam_periph_unlock(periph);
289 cam_periph_release(periph);
293 CAM_DEBUG(periph->path, CAM_DEBUG_TRACE | CAM_DEBUG_PERIPH,
296 softc = (struct nda_softc *)periph->softc;
297 softc->flags |= NDA_FLAG_OPEN;
299 cam_periph_unhold(periph);
300 cam_periph_unlock(periph);
305 ndaclose(struct disk *dp)
307 struct cam_periph *periph;
308 struct nda_softc *softc;
312 periph = (struct cam_periph *)dp->d_drv1;
313 softc = (struct nda_softc *)periph->softc;
314 cam_periph_lock(periph);
316 CAM_DEBUG(periph->path, CAM_DEBUG_TRACE | CAM_DEBUG_PERIPH,
319 if ((softc->flags & NDA_FLAG_DIRTY) != 0 &&
320 (periph->flags & CAM_PERIPH_INVALID) == 0 &&
321 cam_periph_hold(periph, PRIBIO) == 0) {
323 ccb = cam_periph_getccb(periph, CAM_PRIORITY_NORMAL);
324 nda_nvme_flush(softc, &ccb->nvmeio);
325 error = cam_periph_runccb(ccb, ndaerror, /*cam_flags*/0,
326 /*sense_flags*/0, softc->disk->d_devstat);
329 xpt_print(periph->path, "Synchronize cache failed\n");
331 softc->flags &= ~NDA_FLAG_DIRTY;
332 xpt_release_ccb(ccb);
333 cam_periph_unhold(periph);
336 softc->flags &= ~NDA_FLAG_OPEN;
338 while (softc->refcount != 0)
339 cam_periph_sleep(periph, &softc->refcount, PRIBIO, "ndaclose", 1);
340 KASSERT(softc->outstanding_cmds == 0,
341 ("nda %d outstanding commands", softc->outstanding_cmds));
342 cam_periph_unlock(periph);
343 cam_periph_release(periph);
348 ndaschedule(struct cam_periph *periph)
350 struct nda_softc *softc = (struct nda_softc *)periph->softc;
352 if (softc->state != NDA_STATE_NORMAL)
355 cam_iosched_schedule(softc->cam_iosched, periph);
359 * Actually translate the requested transfer into one the physical driver
360 * can understand. The transfer is described by a buf and will include
361 * only one physical transfer.
364 ndastrategy(struct bio *bp)
366 struct cam_periph *periph;
367 struct nda_softc *softc;
369 periph = (struct cam_periph *)bp->bio_disk->d_drv1;
370 softc = (struct nda_softc *)periph->softc;
372 cam_periph_lock(periph);
374 CAM_DEBUG(periph->path, CAM_DEBUG_TRACE, ("ndastrategy(%p)\n", bp));
377 * If the device has been made invalid, error out
379 if ((periph->flags & CAM_PERIPH_INVALID) != 0) {
380 cam_periph_unlock(periph);
381 biofinish(bp, NULL, ENXIO);
385 if (bp->bio_cmd == BIO_DELETE)
389 * Place it in the queue of disk activities for this disk
391 cam_iosched_queue_work(softc->cam_iosched, bp);
394 * Schedule ourselves for performing the work.
397 cam_periph_unlock(periph);
403 ndadump(void *arg, void *virtual, vm_offset_t physical, off_t offset, size_t length)
405 struct cam_periph *periph;
406 struct nda_softc *softc;
408 struct ccb_nvmeio nvmeio;
416 softc = (struct nda_softc *)periph->softc;
417 secsize = softc->disk->d_sectorsize;
418 lba = offset / secsize;
419 count = length / secsize;
421 if ((periph->flags & CAM_PERIPH_INVALID) != 0)
424 /* xpt_get_ccb returns a zero'd allocation for the ccb, mimic that here */
425 memset(&nvmeio, 0, sizeof(nvmeio));
427 xpt_setup_ccb(&nvmeio.ccb_h, periph->path, CAM_PRIORITY_NORMAL);
428 nvmeio.ccb_state = NDA_CCB_DUMP;
429 nda_nvme_write(softc, &nvmeio, virtual, lba, length, count);
430 error = cam_periph_runccb((union ccb *)&nvmeio, cam_periph_error,
431 0, SF_NO_RECOVERY | SF_NO_RETRY, NULL);
433 printf("Aborting dump due to I/O error %d.\n", error);
439 xpt_setup_ccb(&nvmeio.ccb_h, periph->path, CAM_PRIORITY_NORMAL);
441 nvmeio.ccb_state = NDA_CCB_DUMP;
442 nda_nvme_flush(softc, &nvmeio);
443 error = cam_periph_runccb((union ccb *)&nvmeio, cam_periph_error,
444 0, SF_NO_RECOVERY | SF_NO_RETRY, NULL);
446 xpt_print(periph->path, "flush cmd failed\n");
456 * Install a global async callback. This callback will
457 * receive async callbacks like "new device found".
459 status = xpt_register_async(AC_FOUND_DEVICE, ndaasync, NULL, NULL);
461 if (status != CAM_REQ_CMP) {
462 printf("nda: Failed to attach master async callback "
463 "due to status 0x%x!\n", status);
464 } else if (nda_send_ordered) {
466 /* Register our event handlers */
467 if ((EVENTHANDLER_REGISTER(power_suspend, ndasuspend,
468 NULL, EVENTHANDLER_PRI_LAST)) == NULL)
469 printf("ndainit: power event registration failed!\n");
470 if ((EVENTHANDLER_REGISTER(shutdown_post_sync, ndashutdown,
471 NULL, SHUTDOWN_PRI_DEFAULT)) == NULL)
472 printf("ndainit: shutdown event registration failed!\n");
477 * Callback from GEOM, called when it has finished cleaning up its
481 ndadiskgonecb(struct disk *dp)
483 struct cam_periph *periph;
485 periph = (struct cam_periph *)dp->d_drv1;
487 cam_periph_release(periph);
491 ndaoninvalidate(struct cam_periph *periph)
493 struct nda_softc *softc;
495 softc = (struct nda_softc *)periph->softc;
498 * De-register any async callbacks.
500 xpt_register_async(0, ndaasync, periph, periph->path);
502 softc->invalidations++;
506 * Return all queued I/O with ENXIO.
507 * XXX Handle any transactions queued to the card
508 * with XPT_ABORT_CCB.
510 cam_iosched_flush(softc->cam_iosched, NULL, ENXIO);
512 disk_gone(softc->disk);
516 ndacleanup(struct cam_periph *periph)
518 struct nda_softc *softc;
520 softc = (struct nda_softc *)periph->softc;
522 cam_periph_unlock(periph);
524 cam_iosched_fini(softc->cam_iosched);
527 * If we can't free the sysctl tree, oh well...
529 if ((softc->flags & NDA_FLAG_SCTX_INIT) != 0) {
531 if (sysctl_ctx_free(&softc->sysctl_stats_ctx) != 0)
532 xpt_print(periph->path,
533 "can't remove sysctl stats context\n");
535 if (sysctl_ctx_free(&softc->sysctl_ctx) != 0)
536 xpt_print(periph->path,
537 "can't remove sysctl context\n");
540 disk_destroy(softc->disk);
541 free(softc, M_DEVBUF);
542 cam_periph_lock(periph);
546 ndaasync(void *callback_arg, u_int32_t code,
547 struct cam_path *path, void *arg)
549 struct cam_periph *periph;
551 periph = (struct cam_periph *)callback_arg;
553 case AC_FOUND_DEVICE:
555 struct ccb_getdev *cgd;
558 cgd = (struct ccb_getdev *)arg;
562 if (cgd->protocol != PROTO_NVME)
566 * Allocate a peripheral instance for
567 * this device and start the probe
570 status = cam_periph_alloc(ndaregister, ndaoninvalidate,
571 ndacleanup, ndastart,
572 "nda", CAM_PERIPH_BIO,
574 AC_FOUND_DEVICE, cgd);
576 if (status != CAM_REQ_CMP
577 && status != CAM_REQ_INPROG)
578 printf("ndaasync: Unable to attach to new device "
579 "due to status 0x%x\n", status);
582 case AC_ADVINFO_CHANGED:
586 buftype = (uintptr_t)arg;
587 if (buftype == CDAI_TYPE_PHYS_PATH) {
588 struct nda_softc *softc;
590 softc = periph->softc;
591 disk_attr_changed(softc->disk, "GEOM::physpath",
598 cam_periph_async(periph, code, path, arg);
604 ndasysctlinit(void *context, int pending)
606 struct cam_periph *periph;
607 struct nda_softc *softc;
608 char tmpstr[32], tmpstr2[16];
610 periph = (struct cam_periph *)context;
612 /* periph was held for us when this task was enqueued */
613 if ((periph->flags & CAM_PERIPH_INVALID) != 0) {
614 cam_periph_release(periph);
618 softc = (struct nda_softc *)periph->softc;
619 snprintf(tmpstr, sizeof(tmpstr), "CAM NDA unit %d", periph->unit_number);
620 snprintf(tmpstr2, sizeof(tmpstr2), "%d", periph->unit_number);
622 sysctl_ctx_init(&softc->sysctl_ctx);
623 softc->flags |= NDA_FLAG_SCTX_INIT;
624 softc->sysctl_tree = SYSCTL_ADD_NODE_WITH_LABEL(&softc->sysctl_ctx,
625 SYSCTL_STATIC_CHILDREN(_kern_cam_nda), OID_AUTO, tmpstr2,
626 CTLFLAG_RD, 0, tmpstr, "device_index");
627 if (softc->sysctl_tree == NULL) {
628 printf("ndasysctlinit: unable to allocate sysctl tree\n");
629 cam_periph_release(periph);
633 SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
634 OID_AUTO, "unmapped_io", CTLFLAG_RD,
635 &softc->unmappedio, 0, "Unmapped I/O leaf");
637 SYSCTL_ADD_QUAD(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
638 OID_AUTO, "deletes", CTLFLAG_RD,
639 &softc->deletes, "Number of BIO_DELETE requests");
641 SYSCTL_ADD_UQUAD(&softc->sysctl_ctx,
642 SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
643 "trim_count", CTLFLAG_RD, &softc->trim_count,
644 "Total number of unmap/dsm commands sent");
645 SYSCTL_ADD_UQUAD(&softc->sysctl_ctx,
646 SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
647 "trim_ranges", CTLFLAG_RD, &softc->trim_ranges,
648 "Total number of ranges in unmap/dsm commands");
649 SYSCTL_ADD_UQUAD(&softc->sysctl_ctx,
650 SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
651 "trim_lbas", CTLFLAG_RD, &softc->trim_lbas,
652 "Total lbas in the unmap/dsm commands sent");
654 SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
655 OID_AUTO, "rotating", CTLFLAG_RD, &nda_rotating_media, 1,
659 softc->sysctl_stats_tree = SYSCTL_ADD_NODE(&softc->sysctl_stats_ctx,
660 SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO, "stats",
661 CTLFLAG_RD, 0, "Statistics");
662 if (softc->sysctl_stats_tree == NULL) {
663 printf("ndasysctlinit: unable to allocate sysctl tree for stats\n");
664 cam_periph_release(periph);
667 SYSCTL_ADD_INT(&softc->sysctl_stats_ctx,
668 SYSCTL_CHILDREN(softc->sysctl_stats_tree),
669 OID_AUTO, "timeouts", CTLFLAG_RD,
671 "Device timeouts reported by the SIM");
672 SYSCTL_ADD_INT(&softc->sysctl_stats_ctx,
673 SYSCTL_CHILDREN(softc->sysctl_stats_tree),
674 OID_AUTO, "errors", CTLFLAG_RD,
676 "Transport errors reported by the SIM.");
677 SYSCTL_ADD_INT(&softc->sysctl_stats_ctx,
678 SYSCTL_CHILDREN(softc->sysctl_stats_tree),
679 OID_AUTO, "pack_invalidations", CTLFLAG_RD,
680 &softc->invalidations, 0,
681 "Device pack invalidations.");
684 #ifdef CAM_TEST_FAILURE
685 SYSCTL_ADD_PROC(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
686 OID_AUTO, "invalidate", CTLTYPE_U64 | CTLFLAG_RW | CTLFLAG_MPSAFE,
687 periph, 0, cam_periph_invalidate_sysctl, "I",
688 "Write 1 to invalidate the drive immediately");
691 cam_iosched_sysctl_init(softc->cam_iosched, &softc->sysctl_ctx,
694 cam_periph_release(periph);
698 ndagetattr(struct bio *bp)
701 struct cam_periph *periph;
703 periph = (struct cam_periph *)bp->bio_disk->d_drv1;
704 cam_periph_lock(periph);
705 ret = xpt_getattr(bp->bio_data, bp->bio_length, bp->bio_attribute,
707 cam_periph_unlock(periph);
709 bp->bio_completed = bp->bio_length;
714 ndaregister(struct cam_periph *periph, void *arg)
716 struct nda_softc *softc;
718 struct ccb_pathinq cpi;
719 const struct nvme_namespace_data *nsd;
720 const struct nvme_controller_data *cd;
721 char announce_buf[80];
722 uint8_t flbas_fmt, lbads, vwc_present;
726 nsd = nvme_get_identify_ns(periph);
727 cd = nvme_get_identify_cntrl(periph);
729 softc = (struct nda_softc *)malloc(sizeof(*softc), M_DEVBUF,
733 printf("ndaregister: Unable to probe new device. "
734 "Unable to allocate softc\n");
735 return(CAM_REQ_CMP_ERR);
738 if (cam_iosched_init(&softc->cam_iosched, periph) != 0) {
739 printf("ndaregister: Unable to probe new device. "
740 "Unable to allocate iosched memory\n");
741 free(softc, M_DEVBUF);
742 return(CAM_REQ_CMP_ERR);
745 /* ident_data parsing */
747 periph->softc = softc;
749 softc->quirks = NDA_Q_NONE;
751 xpt_path_inq(&cpi, periph->path);
753 TASK_INIT(&softc->sysctl_task, 0, ndasysctlinit, periph);
756 * The name space ID is the lun, save it for later I/O
758 softc->nsid = (uint32_t)xpt_path_lun_id(periph->path);
761 * Register this media as a disk
763 (void)cam_periph_hold(periph, PRIBIO);
764 cam_periph_unlock(periph);
765 snprintf(announce_buf, sizeof(announce_buf),
766 "kern.cam.nda.%d.quirks", periph->unit_number);
767 quirks = softc->quirks;
768 TUNABLE_INT_FETCH(announce_buf, &quirks);
769 softc->quirks = quirks;
770 cam_iosched_set_sort_queue(softc->cam_iosched, 0);
771 softc->disk = disk = disk_alloc();
772 disk->d_rotation_rate = DISK_RR_NON_ROTATING;
773 disk->d_open = ndaopen;
774 disk->d_close = ndaclose;
775 disk->d_strategy = ndastrategy;
776 disk->d_getattr = ndagetattr;
777 disk->d_dump = ndadump;
778 disk->d_gone = ndadiskgonecb;
779 disk->d_name = "nda";
780 disk->d_drv1 = periph;
781 disk->d_unit = periph->unit_number;
782 maxio = cpi.maxio; /* Honor max I/O size of SIM */
784 maxio = DFLTPHYS; /* traditional default */
785 else if (maxio > MAXPHYS)
786 maxio = MAXPHYS; /* for safety */
787 disk->d_maxsize = maxio;
788 flbas_fmt = (nsd->flbas >> NVME_NS_DATA_FLBAS_FORMAT_SHIFT) &
789 NVME_NS_DATA_FLBAS_FORMAT_MASK;
790 lbads = (nsd->lbaf[flbas_fmt] >> NVME_NS_DATA_LBAF_LBADS_SHIFT) &
791 NVME_NS_DATA_LBAF_LBADS_MASK;
792 disk->d_sectorsize = 1 << lbads;
793 disk->d_mediasize = (off_t)(disk->d_sectorsize * nsd->nsze);
794 disk->d_delmaxsize = disk->d_mediasize;
795 disk->d_flags = DISKFLAG_DIRECT_COMPLETION;
796 if (nvme_ctrlr_has_dataset_mgmt(cd))
797 disk->d_flags |= DISKFLAG_CANDELETE;
798 vwc_present = (cd->vwc >> NVME_CTRLR_DATA_VWC_PRESENT_SHIFT) &
799 NVME_CTRLR_DATA_VWC_PRESENT_MASK;
801 disk->d_flags |= DISKFLAG_CANFLUSHCACHE;
802 if ((cpi.hba_misc & PIM_UNMAPPED) != 0) {
803 disk->d_flags |= DISKFLAG_UNMAPPED_BIO;
804 softc->unmappedio = 1;
807 * d_ident and d_descr are both far bigger than the length of either
808 * the serial or model number strings.
810 cam_strvis(disk->d_descr, cd->mn,
811 sizeof(disk->d_descr), NVME_MODEL_NUMBER_LENGTH);
812 cam_strvis(disk->d_ident, cd->sn,
813 sizeof(disk->d_ident), NVME_SERIAL_NUMBER_LENGTH);
814 disk->d_hba_vendor = cpi.hba_vendor;
815 disk->d_hba_device = cpi.hba_device;
816 disk->d_hba_subvendor = cpi.hba_subvendor;
817 disk->d_hba_subdevice = cpi.hba_subdevice;
818 snprintf(disk->d_attachment, sizeof(disk->d_attachment),
819 "%s%d", cpi.dev_name, cpi.unit_number);
820 disk->d_stripesize = disk->d_sectorsize;
821 disk->d_stripeoffset = 0;
822 disk->d_devstat = devstat_new_entry(periph->periph_name,
823 periph->unit_number, disk->d_sectorsize,
824 DEVSTAT_ALL_SUPPORTED,
825 DEVSTAT_TYPE_DIRECT | XPORT_DEVSTAT_TYPE(cpi.transport),
826 DEVSTAT_PRIORITY_DISK);
828 * Add alias for older nvd drives to ease transition.
830 /* disk_add_alias(disk, "nvd"); Have reports of this causing problems */
833 * Acquire a reference to the periph before we register with GEOM.
834 * We'll release this reference once GEOM calls us back (via
835 * ndadiskgonecb()) telling us that our provider has been freed.
837 if (cam_periph_acquire(periph) != 0) {
838 xpt_print(periph->path, "%s: lost periph during "
839 "registration!\n", __func__);
840 cam_periph_lock(periph);
841 return (CAM_REQ_CMP_ERR);
843 disk_create(softc->disk, DISK_VERSION);
844 cam_periph_lock(periph);
845 cam_periph_unhold(periph);
847 snprintf(announce_buf, sizeof(announce_buf),
848 "%juMB (%ju %u byte sectors)",
849 (uintmax_t)((uintmax_t)disk->d_mediasize / (1024*1024)),
850 (uintmax_t)disk->d_mediasize / disk->d_sectorsize,
852 xpt_announce_periph(periph, announce_buf);
853 xpt_announce_quirks(periph, softc->quirks, NDA_Q_BIT_STRING);
856 * Create our sysctl variables, now that we know
857 * we have successfully attached.
859 if (cam_periph_acquire(periph) == 0)
860 taskqueue_enqueue(taskqueue_thread, &softc->sysctl_task);
863 * Register for device going away and info about the drive
864 * changing (though with NVMe, it can't)
866 xpt_register_async(AC_LOST_DEVICE | AC_ADVINFO_CHANGED,
867 ndaasync, periph, periph->path);
869 softc->state = NDA_STATE_NORMAL;
874 ndastart(struct cam_periph *periph, union ccb *start_ccb)
876 struct nda_softc *softc = (struct nda_softc *)periph->softc;
877 struct ccb_nvmeio *nvmeio = &start_ccb->nvmeio;
879 CAM_DEBUG(periph->path, CAM_DEBUG_TRACE, ("ndastart\n"));
881 switch (softc->state) {
882 case NDA_STATE_NORMAL:
886 bp = cam_iosched_next_bio(softc->cam_iosched);
887 CAM_DEBUG(periph->path, CAM_DEBUG_TRACE, ("ndastart: bio %p\n", bp));
889 xpt_release_ccb(start_ccb);
893 switch (bp->bio_cmd) {
895 softc->flags |= NDA_FLAG_DIRTY;
899 #ifdef CAM_TEST_FAILURE
903 * Support the failure ioctls. If the command is a
904 * read, and there are pending forced read errors, or
905 * if a write and pending write errors, then fail this
906 * operation with EIO. This is useful for testing
907 * purposes. Also, support having every Nth read fail.
909 * This is a rather blunt tool.
911 if (bp->bio_cmd == BIO_READ) {
912 if (softc->force_read_error) {
913 softc->force_read_error--;
916 if (softc->periodic_read_error > 0) {
917 if (++softc->periodic_read_count >=
918 softc->periodic_read_error) {
919 softc->periodic_read_count = 0;
924 if (softc->force_write_error) {
925 softc->force_write_error--;
930 biofinish(bp, NULL, EIO);
931 xpt_release_ccb(start_ccb);
936 KASSERT((bp->bio_flags & BIO_UNMAPPED) == 0 ||
937 round_page(bp->bio_bcount + bp->bio_ma_offset) /
938 PAGE_SIZE == bp->bio_ma_n,
939 ("Short bio %p", bp));
940 nda_nvme_rw_bio(softc, &start_ccb->nvmeio, bp, bp->bio_cmd == BIO_READ ?
941 NVME_OPC_READ : NVME_OPC_WRITE);
946 struct nvme_dsm_range *dsm_range, *dsm_end;
947 struct nda_trim_request *trim;
950 uint32_t totalcount = 0, ranges = 0;
952 trim = malloc(sizeof(*trim), M_NVMEDA, M_ZERO | M_NOWAIT);
954 biofinish(bp, NULL, ENOMEM);
955 xpt_release_ccb(start_ccb);
959 TAILQ_INIT(&trim->bps);
961 ents = min(nitems(trim->dsm), nda_max_trim_entries);
962 dsm_range = trim->dsm;
963 dsm_end = dsm_range + ents;
965 TAILQ_INSERT_TAIL(&trim->bps, bp1, bio_queue);
967 htole32(bp1->bio_bcount / softc->disk->d_sectorsize);
968 dsm_range->starting_lba =
969 htole64(bp1->bio_offset / softc->disk->d_sectorsize);
971 totalcount += dsm_range->length;
973 if (dsm_range >= dsm_end)
975 bp1 = cam_iosched_next_trim(softc->cam_iosched);
976 /* XXX -- Could collapse adjacent ranges, but we don't for now */
977 /* XXX -- Could limit based on total payload size */
978 } while (bp1 != NULL);
979 start_ccb->ccb_trim = trim;
980 nda_nvme_trim(softc, &start_ccb->nvmeio, trim->dsm,
981 dsm_range - trim->dsm);
982 start_ccb->ccb_state = NDA_CCB_TRIM;
984 softc->trim_ranges += ranges;
985 softc->trim_lbas += totalcount;
987 * Note: We can have multiple TRIMs in flight, so we don't call
988 * cam_iosched_submit_trim(softc->cam_iosched);
989 * since that forces the I/O scheduler to only schedule one at a time.
990 * On NVMe drives, this is a performance disaster.
995 nda_nvme_flush(softc, nvmeio);
998 start_ccb->ccb_state = NDA_CCB_BUFFER_IO;
999 start_ccb->ccb_bp = bp;
1001 start_ccb->ccb_h.flags |= CAM_UNLOCKED;
1002 softc->outstanding_cmds++;
1003 softc->refcount++; /* For submission only */
1004 cam_periph_unlock(periph);
1005 xpt_action(start_ccb);
1006 cam_periph_lock(periph);
1007 softc->refcount--; /* Submission done */
1009 /* May have more work to do, so ensure we stay scheduled */
1010 ndaschedule(periph);
1017 ndadone(struct cam_periph *periph, union ccb *done_ccb)
1019 struct nda_softc *softc;
1020 struct ccb_nvmeio *nvmeio = &done_ccb->nvmeio;
1021 struct cam_path *path;
1024 softc = (struct nda_softc *)periph->softc;
1025 path = done_ccb->ccb_h.path;
1027 CAM_DEBUG(path, CAM_DEBUG_TRACE, ("ndadone\n"));
1029 state = nvmeio->ccb_state & NDA_CCB_TYPE_MASK;
1031 case NDA_CCB_BUFFER_IO:
1036 cam_periph_lock(periph);
1037 if ((done_ccb->ccb_h.status & CAM_STATUS_MASK) != CAM_REQ_CMP) {
1038 error = ndaerror(done_ccb, 0, 0);
1039 if (error == ERESTART) {
1040 /* A retry was scheduled, so just return. */
1041 cam_periph_unlock(periph);
1044 if ((done_ccb->ccb_h.status & CAM_DEV_QFRZN) != 0)
1045 cam_release_devq(path,
1049 /*getcount_only*/0);
1051 if ((done_ccb->ccb_h.status & CAM_DEV_QFRZN) != 0)
1052 panic("REQ_CMP with QFRZN");
1055 if (state == NDA_CCB_BUFFER_IO) {
1058 bp = (struct bio *)done_ccb->ccb_bp;
1059 bp->bio_error = error;
1061 bp->bio_resid = bp->bio_bcount;
1062 bp->bio_flags |= BIO_ERROR;
1066 softc->outstanding_cmds--;
1069 * We need to call cam_iosched before we call biodone so that we
1070 * don't measure any activity that happens in the completion
1071 * routine, which in the case of sendfile can be quite
1074 cam_iosched_bio_complete(softc->cam_iosched, bp, done_ccb);
1075 xpt_release_ccb(done_ccb);
1076 ndaschedule(periph);
1077 cam_periph_unlock(periph);
1079 } else { /* state == NDA_CCB_TRIM */
1080 struct nda_trim_request *trim;
1081 struct bio *bp1, *bp2;
1082 TAILQ_HEAD(, bio) queue;
1084 trim = nvmeio->ccb_trim;
1086 TAILQ_CONCAT(&queue, &trim->bps, bio_queue);
1087 free(trim, M_NVMEDA);
1090 * Since we can have multiple trims in flight, we don't
1091 * need to call this here.
1092 * cam_iosched_trim_done(softc->cam_iosched);
1095 * The the I/O scheduler that we're finishing the I/O
1096 * so we can keep book. The first one we pass in the CCB
1097 * which has the timing information. The rest we pass in NULL
1098 * so we can keep proper counts.
1100 bp1 = TAILQ_FIRST(&queue);
1101 cam_iosched_bio_complete(softc->cam_iosched, bp1, done_ccb);
1102 xpt_release_ccb(done_ccb);
1103 softc->outstanding_cmds--;
1104 ndaschedule(periph);
1105 cam_periph_unlock(periph);
1106 while ((bp2 = TAILQ_FIRST(&queue)) != NULL) {
1107 TAILQ_REMOVE(&queue, bp2, bio_queue);
1108 bp2->bio_error = error;
1110 bp2->bio_flags |= BIO_ERROR;
1111 bp2->bio_resid = bp1->bio_bcount;
1115 cam_iosched_bio_complete(softc->cam_iosched, bp2, NULL);
1122 /* No-op. We're polling */
1127 xpt_release_ccb(done_ccb);
1131 ndaerror(union ccb *ccb, u_int32_t cam_flags, u_int32_t sense_flags)
1133 struct nda_softc *softc;
1134 struct cam_periph *periph;
1136 periph = xpt_path_periph(ccb->ccb_h.path);
1137 softc = (struct nda_softc *)periph->softc;
1139 switch (ccb->ccb_h.status & CAM_STATUS_MASK) {
1140 case CAM_CMD_TIMEOUT:
1145 case CAM_REQ_ABORTED:
1146 case CAM_REQ_CMP_ERR:
1147 case CAM_REQ_TERMIO:
1148 case CAM_UNREC_HBA_ERROR:
1149 case CAM_DATA_RUN_ERR:
1150 case CAM_ATA_STATUS_ERROR:
1159 return(cam_periph_error(ccb, cam_flags, sense_flags));
1163 * Step through all NDA peripheral drivers, and if the device is still open,
1164 * sync the disk cache to physical media.
1169 struct cam_periph *periph;
1170 struct nda_softc *softc;
1174 CAM_PERIPH_FOREACH(periph, &ndadriver) {
1175 softc = (struct nda_softc *)periph->softc;
1177 if (SCHEDULER_STOPPED()) {
1179 * If we paniced with the lock held or the periph is not
1180 * open, do not recurse. Otherwise, call ndadump since
1181 * that avoids the sleeping cam_periph_getccb does if no
1182 * CCBs are available.
1184 if (!cam_periph_owned(periph) &&
1185 (softc->flags & NDA_FLAG_OPEN)) {
1186 ndadump(softc->disk, NULL, 0, 0, 0);
1192 * We only sync the cache if the drive is still open
1194 cam_periph_lock(periph);
1195 if ((softc->flags & NDA_FLAG_OPEN) == 0) {
1196 cam_periph_unlock(periph);
1200 ccb = cam_periph_getccb(periph, CAM_PRIORITY_NORMAL);
1201 nda_nvme_flush(softc, &ccb->nvmeio);
1202 error = cam_periph_runccb(ccb, ndaerror, /*cam_flags*/0,
1203 /*sense_flags*/ SF_NO_RECOVERY | SF_NO_RETRY,
1204 softc->disk->d_devstat);
1206 xpt_print(periph->path, "Synchronize cache failed\n");
1207 xpt_release_ccb(ccb);
1208 cam_periph_unlock(periph);
1213 ndashutdown(void *arg, int howto)
1220 ndasuspend(void *arg)