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 static int nda_nvd_compat = 1;
189 SYSCTL_INT(_kern_cam_nda, OID_AUTO, max_trim, CTLFLAG_RDTUN,
190 &nda_max_trim_entries, NDA_MAX_TRIM_ENTRIES,
191 "Maximum number of BIO_DELETE to send down as a DSM TRIM.");
192 SYSCTL_INT(_kern_cam_nda, OID_AUTO, enable_biospeedup, CTLFLAG_RDTUN,
193 &nda_enable_biospeedup, 0, "Enable BIO_SPEEDUP processing.");
194 SYSCTL_INT(_kern_cam_nda, OID_AUTO, nvd_compat, CTLFLAG_RDTUN,
195 &nda_nvd_compat, 1, "Enable creation of nvd aliases.");
198 * All NVMe media is non-rotational, so all nvme device instances
199 * share this to implement the sysctl.
201 static int nda_rotating_media = 0;
203 static struct periph_driver ndadriver =
206 TAILQ_HEAD_INITIALIZER(ndadriver.units), /* generation */ 0
209 PERIPHDRIVER_DECLARE(nda, ndadriver);
211 static MALLOC_DEFINE(M_NVMEDA, "nvme_da", "nvme_da buffers");
214 * nice wrappers. Maybe these belong in nvme_all.c instead of
215 * here, but this is the only place that uses these. Should
216 * we ever grow another NVME periph, we should move them
217 * all there wholesale.
221 nda_nvme_flush(struct nda_softc *softc, struct ccb_nvmeio *nvmeio)
223 cam_fill_nvmeio(nvmeio,
225 ndadone, /* cbfcnp */
226 CAM_DIR_NONE, /* flags */
229 nda_default_timeout * 1000); /* timeout 30s */
230 nvme_ns_flush_cmd(&nvmeio->cmd, softc->nsid);
234 nda_nvme_trim(struct nda_softc *softc, struct ccb_nvmeio *nvmeio,
235 void *payload, uint32_t num_ranges)
237 cam_fill_nvmeio(nvmeio,
239 ndadone, /* cbfcnp */
240 CAM_DIR_OUT, /* flags */
241 payload, /* data_ptr */
242 num_ranges * sizeof(struct nvme_dsm_range), /* dxfer_len */
243 nda_default_timeout * 1000); /* timeout 30s */
244 nvme_ns_trim_cmd(&nvmeio->cmd, softc->nsid, num_ranges);
248 nda_nvme_write(struct nda_softc *softc, struct ccb_nvmeio *nvmeio,
249 void *payload, uint64_t lba, uint32_t len, uint32_t count)
251 cam_fill_nvmeio(nvmeio,
253 ndadone, /* cbfcnp */
254 CAM_DIR_OUT, /* flags */
255 payload, /* data_ptr */
257 nda_default_timeout * 1000); /* timeout 30s */
258 nvme_ns_write_cmd(&nvmeio->cmd, softc->nsid, lba, count);
262 nda_nvme_rw_bio(struct nda_softc *softc, struct ccb_nvmeio *nvmeio,
263 struct bio *bp, uint32_t rwcmd)
265 int flags = rwcmd == NVME_OPC_READ ? CAM_DIR_IN : CAM_DIR_OUT;
270 if (bp->bio_flags & BIO_UNMAPPED) {
271 flags |= CAM_DATA_BIO;
274 payload = bp->bio_data;
277 lba = bp->bio_pblkno;
278 count = bp->bio_bcount / softc->disk->d_sectorsize;
280 cam_fill_nvmeio(nvmeio,
282 ndadone, /* cbfcnp */
284 payload, /* data_ptr */
285 bp->bio_bcount, /* dxfer_len */
286 nda_default_timeout * 1000); /* timeout 30s */
287 nvme_ns_rw_cmd(&nvmeio->cmd, rwcmd, softc->nsid, lba, count);
291 ndaopen(struct disk *dp)
293 struct cam_periph *periph;
294 struct nda_softc *softc;
297 periph = (struct cam_periph *)dp->d_drv1;
298 if (cam_periph_acquire(periph) != 0) {
302 cam_periph_lock(periph);
303 if ((error = cam_periph_hold(periph, PRIBIO|PCATCH)) != 0) {
304 cam_periph_unlock(periph);
305 cam_periph_release(periph);
309 CAM_DEBUG(periph->path, CAM_DEBUG_TRACE | CAM_DEBUG_PERIPH,
312 softc = (struct nda_softc *)periph->softc;
313 softc->flags |= NDA_FLAG_OPEN;
315 cam_periph_unhold(periph);
316 cam_periph_unlock(periph);
321 ndaclose(struct disk *dp)
323 struct cam_periph *periph;
324 struct nda_softc *softc;
328 periph = (struct cam_periph *)dp->d_drv1;
329 softc = (struct nda_softc *)periph->softc;
330 cam_periph_lock(periph);
332 CAM_DEBUG(periph->path, CAM_DEBUG_TRACE | CAM_DEBUG_PERIPH,
335 if ((softc->flags & NDA_FLAG_DIRTY) != 0 &&
336 (periph->flags & CAM_PERIPH_INVALID) == 0 &&
337 cam_periph_hold(periph, PRIBIO) == 0) {
339 ccb = cam_periph_getccb(periph, CAM_PRIORITY_NORMAL);
340 nda_nvme_flush(softc, &ccb->nvmeio);
341 error = cam_periph_runccb(ccb, ndaerror, /*cam_flags*/0,
342 /*sense_flags*/0, softc->disk->d_devstat);
345 xpt_print(periph->path, "Synchronize cache failed\n");
347 softc->flags &= ~NDA_FLAG_DIRTY;
348 xpt_release_ccb(ccb);
349 cam_periph_unhold(periph);
352 softc->flags &= ~NDA_FLAG_OPEN;
354 while (softc->refcount != 0)
355 cam_periph_sleep(periph, &softc->refcount, PRIBIO, "ndaclose", 1);
356 KASSERT(softc->outstanding_cmds == 0,
357 ("nda %d outstanding commands", softc->outstanding_cmds));
358 cam_periph_unlock(periph);
359 cam_periph_release(periph);
364 ndaschedule(struct cam_periph *periph)
366 struct nda_softc *softc = (struct nda_softc *)periph->softc;
368 if (softc->state != NDA_STATE_NORMAL)
371 cam_iosched_schedule(softc->cam_iosched, periph);
375 ndaioctl(struct disk *dp, u_long cmd, void *data, int fflag,
378 struct cam_periph *periph;
379 struct nda_softc *softc;
381 periph = (struct cam_periph *)dp->d_drv1;
382 softc = (struct nda_softc *)periph->softc;
388 * These don't map well to the underlying CCBs, so
389 * they are usupported via CAM.
394 struct nvme_get_nsid *gnsid = (struct nvme_get_nsid *)data;
395 struct ccb_pathinq cpi;
397 xpt_path_inq(&cpi, periph->path);
398 strncpy(gnsid->cdev, cpi.xport_specific.nvme.dev_name,
399 sizeof(gnsid->cdev));
400 gnsid->nsid = cpi.xport_specific.nvme.nsid;
403 case NVME_PASSTHROUGH_CMD:
405 struct nvme_pt_command *pt;
407 struct cam_periph_map_info mapinfo;
408 u_int maxmap = dp->d_maxsize;
412 * Create a NVME_IO CCB to do the passthrough command.
414 pt = (struct nvme_pt_command *)data;
415 ccb = xpt_alloc_ccb();
416 xpt_setup_ccb(&ccb->ccb_h, periph->path, CAM_PRIORITY_NORMAL);
417 ccb->ccb_state = NDA_CCB_PASS;
418 cam_fill_nvmeio(&ccb->nvmeio,
421 (pt->is_read ? CAM_DIR_IN : CAM_DIR_OUT) | CAM_DATA_VADDR,
424 nda_default_timeout * 1000);
425 memcpy(&ccb->nvmeio.cmd, &pt->cmd, sizeof(pt->cmd));
428 * Wire the user memory in this request for the I/O
430 memset(&mapinfo, 0, sizeof(mapinfo));
431 error = cam_periph_mapmem(ccb, &mapinfo, maxmap);
436 * Lock the periph and run the command.
438 cam_periph_lock(periph);
439 cam_periph_runccb(ccb, NULL, CAM_RETRY_SELTO,
440 SF_RETRY_UA | SF_NO_PRINT, NULL);
443 * Tear down mapping and return status.
445 cam_periph_unlock(periph);
446 cam_periph_unmapmem(ccb, &mapinfo);
447 error = (ccb->ccb_h.status == CAM_REQ_CMP) ? 0 : EIO;
449 cam_periph_lock(periph);
450 xpt_release_ccb(ccb);
451 cam_periph_unlock(periph);
461 * Actually translate the requested transfer into one the physical driver
462 * can understand. The transfer is described by a buf and will include
463 * only one physical transfer.
466 ndastrategy(struct bio *bp)
468 struct cam_periph *periph;
469 struct nda_softc *softc;
471 periph = (struct cam_periph *)bp->bio_disk->d_drv1;
472 softc = (struct nda_softc *)periph->softc;
474 cam_periph_lock(periph);
476 CAM_DEBUG(periph->path, CAM_DEBUG_TRACE, ("ndastrategy(%p)\n", bp));
479 * If the device has been made invalid, error out
481 if ((periph->flags & CAM_PERIPH_INVALID) != 0) {
482 cam_periph_unlock(periph);
483 biofinish(bp, NULL, ENXIO);
487 if (bp->bio_cmd == BIO_DELETE)
491 * Place it in the queue of disk activities for this disk
493 cam_iosched_queue_work(softc->cam_iosched, bp);
496 * Schedule ourselves for performing the work.
499 cam_periph_unlock(periph);
505 ndadump(void *arg, void *virtual, vm_offset_t physical, off_t offset, size_t length)
507 struct cam_periph *periph;
508 struct nda_softc *softc;
510 struct ccb_nvmeio nvmeio;
518 softc = (struct nda_softc *)periph->softc;
519 secsize = softc->disk->d_sectorsize;
520 lba = offset / secsize;
521 count = length / secsize;
523 if ((periph->flags & CAM_PERIPH_INVALID) != 0)
526 /* xpt_get_ccb returns a zero'd allocation for the ccb, mimic that here */
527 memset(&nvmeio, 0, sizeof(nvmeio));
529 xpt_setup_ccb(&nvmeio.ccb_h, periph->path, CAM_PRIORITY_NORMAL);
530 nvmeio.ccb_state = NDA_CCB_DUMP;
531 nda_nvme_write(softc, &nvmeio, virtual, lba, length, count);
532 error = cam_periph_runccb((union ccb *)&nvmeio, cam_periph_error,
533 0, SF_NO_RECOVERY | SF_NO_RETRY, NULL);
535 printf("Aborting dump due to I/O error %d.\n", error);
541 xpt_setup_ccb(&nvmeio.ccb_h, periph->path, CAM_PRIORITY_NORMAL);
543 nvmeio.ccb_state = NDA_CCB_DUMP;
544 nda_nvme_flush(softc, &nvmeio);
545 error = cam_periph_runccb((union ccb *)&nvmeio, cam_periph_error,
546 0, SF_NO_RECOVERY | SF_NO_RETRY, NULL);
548 xpt_print(periph->path, "flush cmd failed\n");
558 * Install a global async callback. This callback will
559 * receive async callbacks like "new device found".
561 status = xpt_register_async(AC_FOUND_DEVICE, ndaasync, NULL, NULL);
563 if (status != CAM_REQ_CMP) {
564 printf("nda: Failed to attach master async callback "
565 "due to status 0x%x!\n", status);
566 } else if (nda_send_ordered) {
568 /* Register our event handlers */
569 if ((EVENTHANDLER_REGISTER(power_suspend, ndasuspend,
570 NULL, EVENTHANDLER_PRI_LAST)) == NULL)
571 printf("ndainit: power event registration failed!\n");
572 if ((EVENTHANDLER_REGISTER(shutdown_post_sync, ndashutdown,
573 NULL, SHUTDOWN_PRI_DEFAULT)) == NULL)
574 printf("ndainit: shutdown event registration failed!\n");
579 * Callback from GEOM, called when it has finished cleaning up its
583 ndadiskgonecb(struct disk *dp)
585 struct cam_periph *periph;
587 periph = (struct cam_periph *)dp->d_drv1;
589 cam_periph_release(periph);
593 ndaoninvalidate(struct cam_periph *periph)
595 struct nda_softc *softc;
597 softc = (struct nda_softc *)periph->softc;
600 * De-register any async callbacks.
602 xpt_register_async(0, ndaasync, periph, periph->path);
604 softc->invalidations++;
608 * Return all queued I/O with ENXIO.
609 * XXX Handle any transactions queued to the card
610 * with XPT_ABORT_CCB.
612 cam_iosched_flush(softc->cam_iosched, NULL, ENXIO);
614 disk_gone(softc->disk);
618 ndacleanup(struct cam_periph *periph)
620 struct nda_softc *softc;
622 softc = (struct nda_softc *)periph->softc;
624 cam_periph_unlock(periph);
626 cam_iosched_fini(softc->cam_iosched);
629 * If we can't free the sysctl tree, oh well...
631 if ((softc->flags & NDA_FLAG_SCTX_INIT) != 0) {
633 if (sysctl_ctx_free(&softc->sysctl_stats_ctx) != 0)
634 xpt_print(periph->path,
635 "can't remove sysctl stats context\n");
637 if (sysctl_ctx_free(&softc->sysctl_ctx) != 0)
638 xpt_print(periph->path,
639 "can't remove sysctl context\n");
642 disk_destroy(softc->disk);
643 free(softc, M_DEVBUF);
644 cam_periph_lock(periph);
648 ndaasync(void *callback_arg, u_int32_t code,
649 struct cam_path *path, void *arg)
651 struct cam_periph *periph;
653 periph = (struct cam_periph *)callback_arg;
655 case AC_FOUND_DEVICE:
657 struct ccb_getdev *cgd;
660 cgd = (struct ccb_getdev *)arg;
664 if (cgd->protocol != PROTO_NVME)
668 * Allocate a peripheral instance for
669 * this device and start the probe
672 status = cam_periph_alloc(ndaregister, ndaoninvalidate,
673 ndacleanup, ndastart,
674 "nda", CAM_PERIPH_BIO,
676 AC_FOUND_DEVICE, cgd);
678 if (status != CAM_REQ_CMP
679 && status != CAM_REQ_INPROG)
680 printf("ndaasync: Unable to attach to new device "
681 "due to status 0x%x\n", status);
684 case AC_ADVINFO_CHANGED:
688 buftype = (uintptr_t)arg;
689 if (buftype == CDAI_TYPE_PHYS_PATH) {
690 struct nda_softc *softc;
692 softc = periph->softc;
693 disk_attr_changed(softc->disk, "GEOM::physpath",
700 cam_periph_async(periph, code, path, arg);
706 ndasysctlinit(void *context, int pending)
708 struct cam_periph *periph;
709 struct nda_softc *softc;
710 char tmpstr[32], tmpstr2[16];
712 periph = (struct cam_periph *)context;
714 /* periph was held for us when this task was enqueued */
715 if ((periph->flags & CAM_PERIPH_INVALID) != 0) {
716 cam_periph_release(periph);
720 softc = (struct nda_softc *)periph->softc;
721 snprintf(tmpstr, sizeof(tmpstr), "CAM NDA unit %d", periph->unit_number);
722 snprintf(tmpstr2, sizeof(tmpstr2), "%d", periph->unit_number);
724 sysctl_ctx_init(&softc->sysctl_ctx);
725 softc->flags |= NDA_FLAG_SCTX_INIT;
726 softc->sysctl_tree = SYSCTL_ADD_NODE_WITH_LABEL(&softc->sysctl_ctx,
727 SYSCTL_STATIC_CHILDREN(_kern_cam_nda), OID_AUTO, tmpstr2,
728 CTLFLAG_RD | CTLFLAG_MPSAFE, 0, tmpstr, "device_index");
729 if (softc->sysctl_tree == NULL) {
730 printf("ndasysctlinit: unable to allocate sysctl tree\n");
731 cam_periph_release(periph);
735 SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
736 OID_AUTO, "unmapped_io", CTLFLAG_RD,
737 &softc->unmappedio, 0, "Unmapped I/O leaf");
739 SYSCTL_ADD_QUAD(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
740 OID_AUTO, "deletes", CTLFLAG_RD,
741 &softc->deletes, "Number of BIO_DELETE requests");
743 SYSCTL_ADD_UQUAD(&softc->sysctl_ctx,
744 SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
745 "trim_count", CTLFLAG_RD, &softc->trim_count,
746 "Total number of unmap/dsm commands sent");
747 SYSCTL_ADD_UQUAD(&softc->sysctl_ctx,
748 SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
749 "trim_ranges", CTLFLAG_RD, &softc->trim_ranges,
750 "Total number of ranges in unmap/dsm commands");
751 SYSCTL_ADD_UQUAD(&softc->sysctl_ctx,
752 SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
753 "trim_lbas", CTLFLAG_RD, &softc->trim_lbas,
754 "Total lbas in the unmap/dsm commands sent");
756 SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
757 OID_AUTO, "rotating", CTLFLAG_RD, &nda_rotating_media, 1,
760 SYSCTL_ADD_PROC(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
761 OID_AUTO, "flags", CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE,
762 softc, 0, ndaflagssysctl, "A",
766 softc->sysctl_stats_tree = SYSCTL_ADD_NODE(&softc->sysctl_stats_ctx,
767 SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO, "stats",
768 CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "Statistics");
769 if (softc->sysctl_stats_tree == NULL) {
770 printf("ndasysctlinit: unable to allocate sysctl tree for stats\n");
771 cam_periph_release(periph);
774 SYSCTL_ADD_INT(&softc->sysctl_stats_ctx,
775 SYSCTL_CHILDREN(softc->sysctl_stats_tree),
776 OID_AUTO, "timeouts", CTLFLAG_RD,
778 "Device timeouts reported by the SIM");
779 SYSCTL_ADD_INT(&softc->sysctl_stats_ctx,
780 SYSCTL_CHILDREN(softc->sysctl_stats_tree),
781 OID_AUTO, "errors", CTLFLAG_RD,
783 "Transport errors reported by the SIM.");
784 SYSCTL_ADD_INT(&softc->sysctl_stats_ctx,
785 SYSCTL_CHILDREN(softc->sysctl_stats_tree),
786 OID_AUTO, "pack_invalidations", CTLFLAG_RD,
787 &softc->invalidations, 0,
788 "Device pack invalidations.");
791 #ifdef CAM_TEST_FAILURE
792 SYSCTL_ADD_PROC(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
793 OID_AUTO, "invalidate", CTLTYPE_U64 | CTLFLAG_RW | CTLFLAG_MPSAFE,
794 periph, 0, cam_periph_invalidate_sysctl, "I",
795 "Write 1 to invalidate the drive immediately");
798 cam_iosched_sysctl_init(softc->cam_iosched, &softc->sysctl_ctx,
801 cam_periph_release(periph);
805 ndaflagssysctl(SYSCTL_HANDLER_ARGS)
808 struct nda_softc *softc = arg1;
811 sbuf_new_for_sysctl(&sbuf, NULL, 0, req);
812 if (softc->flags != 0)
813 sbuf_printf(&sbuf, "0x%b", (unsigned)softc->flags, NDA_FLAG_STRING);
815 sbuf_printf(&sbuf, "0");
816 error = sbuf_finish(&sbuf);
823 ndagetattr(struct bio *bp)
826 struct cam_periph *periph;
828 if (g_handleattr_int(bp, "GEOM::canspeedup", nda_enable_biospeedup))
829 return (EJUSTRETURN);
831 periph = (struct cam_periph *)bp->bio_disk->d_drv1;
832 cam_periph_lock(periph);
833 ret = xpt_getattr(bp->bio_data, bp->bio_length, bp->bio_attribute,
835 cam_periph_unlock(periph);
837 bp->bio_completed = bp->bio_length;
842 ndaregister(struct cam_periph *periph, void *arg)
844 struct nda_softc *softc;
846 struct ccb_pathinq cpi;
847 const struct nvme_namespace_data *nsd;
848 const struct nvme_controller_data *cd;
849 char announce_buf[80];
850 uint8_t flbas_fmt, lbads, vwc_present;
854 nsd = nvme_get_identify_ns(periph);
855 cd = nvme_get_identify_cntrl(periph);
857 softc = (struct nda_softc *)malloc(sizeof(*softc), M_DEVBUF,
861 printf("ndaregister: Unable to probe new device. "
862 "Unable to allocate softc\n");
863 return(CAM_REQ_CMP_ERR);
866 if (cam_iosched_init(&softc->cam_iosched, periph) != 0) {
867 printf("ndaregister: Unable to probe new device. "
868 "Unable to allocate iosched memory\n");
869 free(softc, M_DEVBUF);
870 return(CAM_REQ_CMP_ERR);
873 /* ident_data parsing */
875 periph->softc = softc;
876 softc->quirks = NDA_Q_NONE;
877 xpt_path_inq(&cpi, periph->path);
878 TASK_INIT(&softc->sysctl_task, 0, ndasysctlinit, periph);
881 * The name space ID is the lun, save it for later I/O
883 softc->nsid = (uint32_t)xpt_path_lun_id(periph->path);
886 * Register this media as a disk
888 (void)cam_periph_hold(periph, PRIBIO);
889 cam_periph_unlock(periph);
890 snprintf(announce_buf, sizeof(announce_buf),
891 "kern.cam.nda.%d.quirks", periph->unit_number);
892 quirks = softc->quirks;
893 TUNABLE_INT_FETCH(announce_buf, &quirks);
894 softc->quirks = quirks;
895 cam_iosched_set_sort_queue(softc->cam_iosched, 0);
896 softc->disk = disk = disk_alloc();
897 disk->d_rotation_rate = DISK_RR_NON_ROTATING;
898 disk->d_open = ndaopen;
899 disk->d_close = ndaclose;
900 disk->d_strategy = ndastrategy;
901 disk->d_ioctl = ndaioctl;
902 disk->d_getattr = ndagetattr;
903 disk->d_dump = ndadump;
904 disk->d_gone = ndadiskgonecb;
905 disk->d_name = "nda";
906 disk->d_drv1 = periph;
907 disk->d_unit = periph->unit_number;
908 maxio = cpi.maxio; /* Honor max I/O size of SIM */
910 maxio = DFLTPHYS; /* traditional default */
911 else if (maxio > MAXPHYS)
912 maxio = MAXPHYS; /* for safety */
913 disk->d_maxsize = maxio;
914 flbas_fmt = (nsd->flbas >> NVME_NS_DATA_FLBAS_FORMAT_SHIFT) &
915 NVME_NS_DATA_FLBAS_FORMAT_MASK;
916 lbads = (nsd->lbaf[flbas_fmt] >> NVME_NS_DATA_LBAF_LBADS_SHIFT) &
917 NVME_NS_DATA_LBAF_LBADS_MASK;
918 disk->d_sectorsize = 1 << lbads;
919 disk->d_mediasize = (off_t)(disk->d_sectorsize * nsd->nsze);
920 disk->d_delmaxsize = disk->d_mediasize;
921 disk->d_flags = DISKFLAG_DIRECT_COMPLETION;
922 if (nvme_ctrlr_has_dataset_mgmt(cd))
923 disk->d_flags |= DISKFLAG_CANDELETE;
924 vwc_present = (cd->vwc >> NVME_CTRLR_DATA_VWC_PRESENT_SHIFT) &
925 NVME_CTRLR_DATA_VWC_PRESENT_MASK;
927 disk->d_flags |= DISKFLAG_CANFLUSHCACHE;
928 if ((cpi.hba_misc & PIM_UNMAPPED) != 0) {
929 disk->d_flags |= DISKFLAG_UNMAPPED_BIO;
930 softc->unmappedio = 1;
933 * d_ident and d_descr are both far bigger than the length of either
934 * the serial or model number strings.
936 cam_strvis(disk->d_descr, cd->mn,
937 NVME_MODEL_NUMBER_LENGTH, sizeof(disk->d_descr));
938 cam_strvis(disk->d_ident, cd->sn,
939 NVME_SERIAL_NUMBER_LENGTH, sizeof(disk->d_ident));
940 disk->d_hba_vendor = cpi.hba_vendor;
941 disk->d_hba_device = cpi.hba_device;
942 disk->d_hba_subvendor = cpi.hba_subvendor;
943 disk->d_hba_subdevice = cpi.hba_subdevice;
944 snprintf(disk->d_attachment, sizeof(disk->d_attachment),
945 "%s%d", cpi.dev_name, cpi.unit_number);
946 disk->d_stripesize = disk->d_sectorsize;
947 disk->d_stripeoffset = 0;
948 disk->d_devstat = devstat_new_entry(periph->periph_name,
949 periph->unit_number, disk->d_sectorsize,
950 DEVSTAT_ALL_SUPPORTED,
951 DEVSTAT_TYPE_DIRECT | XPORT_DEVSTAT_TYPE(cpi.transport),
952 DEVSTAT_PRIORITY_DISK);
954 * Add alias for older nvd drives to ease transition.
957 disk_add_alias(disk, "nvd");
960 * Acquire a reference to the periph before we register with GEOM.
961 * We'll release this reference once GEOM calls us back (via
962 * ndadiskgonecb()) telling us that our provider has been freed.
964 if (cam_periph_acquire(periph) != 0) {
965 xpt_print(periph->path, "%s: lost periph during "
966 "registration!\n", __func__);
967 cam_periph_lock(periph);
968 return (CAM_REQ_CMP_ERR);
970 disk_create(softc->disk, DISK_VERSION);
971 cam_periph_lock(periph);
972 cam_periph_unhold(periph);
974 snprintf(announce_buf, sizeof(announce_buf),
975 "%juMB (%ju %u byte sectors)",
976 (uintmax_t)((uintmax_t)disk->d_mediasize / (1024*1024)),
977 (uintmax_t)disk->d_mediasize / disk->d_sectorsize,
979 xpt_announce_periph(periph, announce_buf);
980 xpt_announce_quirks(periph, softc->quirks, NDA_Q_BIT_STRING);
983 * Create our sysctl variables, now that we know
984 * we have successfully attached.
986 if (cam_periph_acquire(periph) == 0)
987 taskqueue_enqueue(taskqueue_thread, &softc->sysctl_task);
990 * Register for device going away and info about the drive
991 * changing (though with NVMe, it can't)
993 xpt_register_async(AC_LOST_DEVICE | AC_ADVINFO_CHANGED,
994 ndaasync, periph, periph->path);
996 softc->state = NDA_STATE_NORMAL;
1001 ndastart(struct cam_periph *periph, union ccb *start_ccb)
1003 struct nda_softc *softc = (struct nda_softc *)periph->softc;
1004 struct ccb_nvmeio *nvmeio = &start_ccb->nvmeio;
1006 CAM_DEBUG(periph->path, CAM_DEBUG_TRACE, ("ndastart\n"));
1008 switch (softc->state) {
1009 case NDA_STATE_NORMAL:
1013 bp = cam_iosched_next_bio(softc->cam_iosched);
1014 CAM_DEBUG(periph->path, CAM_DEBUG_TRACE, ("ndastart: bio %p\n", bp));
1016 xpt_release_ccb(start_ccb);
1020 switch (bp->bio_cmd) {
1022 softc->flags |= NDA_FLAG_DIRTY;
1026 #ifdef CAM_TEST_FAILURE
1030 * Support the failure ioctls. If the command is a
1031 * read, and there are pending forced read errors, or
1032 * if a write and pending write errors, then fail this
1033 * operation with EIO. This is useful for testing
1034 * purposes. Also, support having every Nth read fail.
1036 * This is a rather blunt tool.
1038 if (bp->bio_cmd == BIO_READ) {
1039 if (softc->force_read_error) {
1040 softc->force_read_error--;
1043 if (softc->periodic_read_error > 0) {
1044 if (++softc->periodic_read_count >=
1045 softc->periodic_read_error) {
1046 softc->periodic_read_count = 0;
1051 if (softc->force_write_error) {
1052 softc->force_write_error--;
1057 biofinish(bp, NULL, EIO);
1058 xpt_release_ccb(start_ccb);
1059 ndaschedule(periph);
1063 KASSERT((bp->bio_flags & BIO_UNMAPPED) == 0 ||
1064 round_page(bp->bio_bcount + bp->bio_ma_offset) /
1065 PAGE_SIZE == bp->bio_ma_n,
1066 ("Short bio %p", bp));
1067 nda_nvme_rw_bio(softc, &start_ccb->nvmeio, bp, bp->bio_cmd == BIO_READ ?
1068 NVME_OPC_READ : NVME_OPC_WRITE);
1073 struct nvme_dsm_range *dsm_range, *dsm_end;
1074 struct nda_trim_request *trim;
1077 uint32_t totalcount = 0, ranges = 0;
1079 trim = malloc(sizeof(*trim), M_NVMEDA, M_ZERO | M_NOWAIT);
1081 biofinish(bp, NULL, ENOMEM);
1082 xpt_release_ccb(start_ccb);
1083 ndaschedule(periph);
1086 TAILQ_INIT(&trim->bps);
1088 ents = min(nitems(trim->dsm), nda_max_trim_entries);
1089 ents = max(ents, 1);
1090 dsm_range = trim->dsm;
1091 dsm_end = dsm_range + ents;
1093 TAILQ_INSERT_TAIL(&trim->bps, bp1, bio_queue);
1095 htole32(bp1->bio_bcount / softc->disk->d_sectorsize);
1096 dsm_range->starting_lba =
1097 htole64(bp1->bio_offset / softc->disk->d_sectorsize);
1099 totalcount += dsm_range->length;
1101 if (dsm_range >= dsm_end)
1103 bp1 = cam_iosched_next_trim(softc->cam_iosched);
1104 /* XXX -- Could collapse adjacent ranges, but we don't for now */
1105 /* XXX -- Could limit based on total payload size */
1106 } while (bp1 != NULL);
1107 start_ccb->ccb_trim = trim;
1108 nda_nvme_trim(softc, &start_ccb->nvmeio, trim->dsm,
1109 dsm_range - trim->dsm);
1110 start_ccb->ccb_state = NDA_CCB_TRIM;
1111 softc->trim_count++;
1112 softc->trim_ranges += ranges;
1113 softc->trim_lbas += totalcount;
1115 * Note: We can have multiple TRIMs in flight, so we don't call
1116 * cam_iosched_submit_trim(softc->cam_iosched);
1117 * since that forces the I/O scheduler to only schedule one at a time.
1118 * On NVMe drives, this is a performance disaster.
1123 nda_nvme_flush(softc, nvmeio);
1126 biofinish(bp, NULL, EOPNOTSUPP);
1127 xpt_release_ccb(start_ccb);
1128 ndaschedule(periph);
1131 start_ccb->ccb_state = NDA_CCB_BUFFER_IO;
1132 start_ccb->ccb_bp = bp;
1134 start_ccb->ccb_h.flags |= CAM_UNLOCKED;
1135 softc->outstanding_cmds++;
1136 softc->refcount++; /* For submission only */
1137 cam_periph_unlock(periph);
1138 xpt_action(start_ccb);
1139 cam_periph_lock(periph);
1140 softc->refcount--; /* Submission done */
1142 /* May have more work to do, so ensure we stay scheduled */
1143 ndaschedule(periph);
1150 ndadone(struct cam_periph *periph, union ccb *done_ccb)
1152 struct nda_softc *softc;
1153 struct ccb_nvmeio *nvmeio = &done_ccb->nvmeio;
1154 struct cam_path *path;
1157 softc = (struct nda_softc *)periph->softc;
1158 path = done_ccb->ccb_h.path;
1160 CAM_DEBUG(path, CAM_DEBUG_TRACE, ("ndadone\n"));
1162 state = nvmeio->ccb_state & NDA_CCB_TYPE_MASK;
1164 case NDA_CCB_BUFFER_IO:
1169 cam_periph_lock(periph);
1170 if ((done_ccb->ccb_h.status & CAM_STATUS_MASK) != CAM_REQ_CMP) {
1171 error = ndaerror(done_ccb, 0, 0);
1172 if (error == ERESTART) {
1173 /* A retry was scheduled, so just return. */
1174 cam_periph_unlock(periph);
1177 if ((done_ccb->ccb_h.status & CAM_DEV_QFRZN) != 0)
1178 cam_release_devq(path,
1182 /*getcount_only*/0);
1184 if ((done_ccb->ccb_h.status & CAM_DEV_QFRZN) != 0)
1185 panic("REQ_CMP with QFRZN");
1188 if (state == NDA_CCB_BUFFER_IO) {
1191 bp = (struct bio *)done_ccb->ccb_bp;
1192 bp->bio_error = error;
1194 bp->bio_resid = bp->bio_bcount;
1195 bp->bio_flags |= BIO_ERROR;
1199 softc->outstanding_cmds--;
1202 * We need to call cam_iosched before we call biodone so that we
1203 * don't measure any activity that happens in the completion
1204 * routine, which in the case of sendfile can be quite
1207 cam_iosched_bio_complete(softc->cam_iosched, bp, done_ccb);
1208 xpt_release_ccb(done_ccb);
1209 ndaschedule(periph);
1210 cam_periph_unlock(periph);
1212 } else { /* state == NDA_CCB_TRIM */
1213 struct nda_trim_request *trim;
1214 struct bio *bp1, *bp2;
1215 TAILQ_HEAD(, bio) queue;
1217 trim = nvmeio->ccb_trim;
1219 TAILQ_CONCAT(&queue, &trim->bps, bio_queue);
1220 free(trim, M_NVMEDA);
1223 * Since we can have multiple trims in flight, we don't
1224 * need to call this here.
1225 * cam_iosched_trim_done(softc->cam_iosched);
1228 * The the I/O scheduler that we're finishing the I/O
1229 * so we can keep book. The first one we pass in the CCB
1230 * which has the timing information. The rest we pass in NULL
1231 * so we can keep proper counts.
1233 bp1 = TAILQ_FIRST(&queue);
1234 cam_iosched_bio_complete(softc->cam_iosched, bp1, done_ccb);
1235 xpt_release_ccb(done_ccb);
1236 softc->outstanding_cmds--;
1237 ndaschedule(periph);
1238 cam_periph_unlock(periph);
1239 while ((bp2 = TAILQ_FIRST(&queue)) != NULL) {
1240 TAILQ_REMOVE(&queue, bp2, bio_queue);
1241 bp2->bio_error = error;
1243 bp2->bio_flags |= BIO_ERROR;
1244 bp2->bio_resid = bp1->bio_bcount;
1248 cam_iosched_bio_complete(softc->cam_iosched, bp2, NULL);
1255 /* No-op. We're polling */
1262 xpt_release_ccb(done_ccb);
1266 ndaerror(union ccb *ccb, u_int32_t cam_flags, u_int32_t sense_flags)
1268 struct nda_softc *softc;
1269 struct cam_periph *periph;
1271 periph = xpt_path_periph(ccb->ccb_h.path);
1272 softc = (struct nda_softc *)periph->softc;
1274 switch (ccb->ccb_h.status & CAM_STATUS_MASK) {
1275 case CAM_CMD_TIMEOUT:
1280 case CAM_REQ_ABORTED:
1281 case CAM_REQ_CMP_ERR:
1282 case CAM_REQ_TERMIO:
1283 case CAM_UNREC_HBA_ERROR:
1284 case CAM_DATA_RUN_ERR:
1285 case CAM_ATA_STATUS_ERROR:
1294 return(cam_periph_error(ccb, cam_flags, sense_flags));
1298 * Step through all NDA peripheral drivers, and if the device is still open,
1299 * sync the disk cache to physical media.
1304 struct cam_periph *periph;
1305 struct nda_softc *softc;
1309 CAM_PERIPH_FOREACH(periph, &ndadriver) {
1310 softc = (struct nda_softc *)periph->softc;
1312 if (SCHEDULER_STOPPED()) {
1314 * If we paniced with the lock held or the periph is not
1315 * open, do not recurse. Otherwise, call ndadump since
1316 * that avoids the sleeping cam_periph_getccb does if no
1317 * CCBs are available.
1319 if (!cam_periph_owned(periph) &&
1320 (softc->flags & NDA_FLAG_OPEN)) {
1321 ndadump(softc->disk, NULL, 0, 0, 0);
1327 * We only sync the cache if the drive is still open
1329 cam_periph_lock(periph);
1330 if ((softc->flags & NDA_FLAG_OPEN) == 0) {
1331 cam_periph_unlock(periph);
1335 ccb = cam_periph_getccb(periph, CAM_PRIORITY_NORMAL);
1336 nda_nvme_flush(softc, &ccb->nvmeio);
1337 error = cam_periph_runccb(ccb, ndaerror, /*cam_flags*/0,
1338 /*sense_flags*/ SF_NO_RECOVERY | SF_NO_RETRY,
1339 softc->disk->d_devstat);
1341 xpt_print(periph->path, "Synchronize cache failed\n");
1342 xpt_release_ccb(ccb);
1343 cam_periph_unlock(periph);
1348 ndashutdown(void *arg, int howto)
1355 ndasuspend(void *arg)