2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (C) 2012-2016 Intel Corporation
6 * Copyright (C) 2018 Alexander Motin <mav@FreeBSD.org>
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
33 #include <sys/param.h>
35 #include <sys/kernel.h>
36 #include <sys/malloc.h>
37 #include <sys/module.h>
38 #include <sys/queue.h>
39 #include <sys/sysctl.h>
40 #include <sys/systm.h>
41 #include <sys/taskqueue.h>
42 #include <machine/atomic.h>
44 #include <geom/geom.h>
45 #include <geom/geom_disk.h>
47 #include <dev/nvme/nvme.h>
52 struct nvd_controller;
54 static disk_ioctl_t nvd_ioctl;
55 static disk_strategy_t nvd_strategy;
56 static dumper_t nvd_dump;
57 static disk_getattr_t nvd_getattr;
59 static void nvd_done(void *arg, const struct nvme_completion *cpl);
60 static void nvd_gone(struct nvd_disk *ndisk);
62 static void *nvd_new_disk(struct nvme_namespace *ns, void *ctrlr);
64 static void *nvd_new_controller(struct nvme_controller *ctrlr);
65 static void nvd_controller_fail(void *ctrlr);
67 static int nvd_load(void);
68 static void nvd_unload(void);
70 MALLOC_DEFINE(M_NVD, "nvd", "nvd(4) allocations");
72 struct nvme_consumer *consumer_handle;
75 struct nvd_controller *ctrlr;
77 struct bio_queue_head bioq;
83 struct nvme_namespace *ns;
86 #define NVD_ODEPTH (1 << 30)
87 uint32_t ordered_in_flight;
90 TAILQ_ENTRY(nvd_disk) global_tailq;
91 TAILQ_ENTRY(nvd_disk) ctrlr_tailq;
94 struct nvd_controller {
96 TAILQ_ENTRY(nvd_controller) tailq;
97 TAILQ_HEAD(, nvd_disk) disk_head;
100 static struct mtx nvd_lock;
101 static TAILQ_HEAD(, nvd_controller) ctrlr_head;
102 static TAILQ_HEAD(disk_list, nvd_disk) disk_head;
104 static SYSCTL_NODE(_hw, OID_AUTO, nvd, CTLFLAG_RD, 0, "nvd driver parameters");
106 * The NVMe specification does not define a maximum or optimal delete size, so
107 * technically max delete size is min(full size of the namespace, 2^32 - 1
108 * LBAs). A single delete for a multi-TB NVMe namespace though may take much
109 * longer to complete than the nvme(4) I/O timeout period. So choose a sensible
110 * default here that is still suitably large to minimize the number of overall
113 static uint64_t nvd_delete_max = (1024 * 1024 * 1024); /* 1GB */
114 SYSCTL_UQUAD(_hw_nvd, OID_AUTO, delete_max, CTLFLAG_RDTUN, &nvd_delete_max, 0,
115 "nvd maximum BIO_DELETE size in bytes");
117 static int nvd_modevent(module_t mod, int type, void *arg)
135 moduledata_t nvd_mod = {
137 (modeventhand_t)nvd_modevent,
141 DECLARE_MODULE(nvd, nvd_mod, SI_SUB_DRIVERS, SI_ORDER_ANY);
142 MODULE_VERSION(nvd, 1);
143 MODULE_DEPEND(nvd, nvme, 1, 1, 1);
151 mtx_init(&nvd_lock, "nvd_lock", NULL, MTX_DEF);
152 TAILQ_INIT(&ctrlr_head);
153 TAILQ_INIT(&disk_head);
155 consumer_handle = nvme_register_consumer(nvd_new_disk,
156 nvd_new_controller, NULL, nvd_controller_fail);
158 return (consumer_handle != NULL ? 0 : -1);
164 struct nvd_controller *ctrlr;
165 struct nvd_disk *ndisk;
171 while ((ctrlr = TAILQ_FIRST(&ctrlr_head)) != NULL) {
172 TAILQ_REMOVE(&ctrlr_head, ctrlr, tailq);
173 TAILQ_FOREACH(ndisk, &ctrlr->disk_head, ctrlr_tailq)
175 while (!TAILQ_EMPTY(&ctrlr->disk_head))
176 msleep(&ctrlr->disk_head, &nvd_lock, 0, "nvd_unload",0);
179 mtx_unlock(&nvd_lock);
181 nvme_unregister_consumer(consumer_handle);
183 mtx_destroy(&nvd_lock);
187 nvd_bio_submit(struct nvd_disk *ndisk, struct bio *bp)
191 bp->bio_driver1 = NULL;
192 if (__predict_false(bp->bio_flags & BIO_ORDERED))
193 atomic_add_int(&ndisk->cur_depth, NVD_ODEPTH);
195 atomic_add_int(&ndisk->cur_depth, 1);
196 err = nvme_ns_bio_process(ndisk->ns, bp, nvd_done);
198 if (__predict_false(bp->bio_flags & BIO_ORDERED)) {
199 atomic_add_int(&ndisk->cur_depth, -NVD_ODEPTH);
200 atomic_add_int(&ndisk->ordered_in_flight, -1);
201 wakeup(&ndisk->cur_depth);
203 if (atomic_fetchadd_int(&ndisk->cur_depth, -1) == 1 &&
204 __predict_false(ndisk->ordered_in_flight != 0))
205 wakeup(&ndisk->cur_depth);
208 bp->bio_flags |= BIO_ERROR;
209 bp->bio_resid = bp->bio_bcount;
215 nvd_strategy(struct bio *bp)
217 struct nvd_disk *ndisk = (struct nvd_disk *)bp->bio_disk->d_drv1;
220 * bio with BIO_ORDERED flag must be executed after all previous
221 * bios in the queue, and before any successive bios.
223 if (__predict_false(bp->bio_flags & BIO_ORDERED)) {
224 if (atomic_fetchadd_int(&ndisk->ordered_in_flight, 1) == 0 &&
225 ndisk->cur_depth == 0 && bioq_first(&ndisk->bioq) == NULL) {
226 nvd_bio_submit(ndisk, bp);
229 } else if (__predict_true(ndisk->ordered_in_flight == 0)) {
230 nvd_bio_submit(ndisk, bp);
235 * There are ordered bios in flight, so we need to submit
236 * bios through the task queue to enforce ordering.
238 mtx_lock(&ndisk->bioqlock);
239 bioq_insert_tail(&ndisk->bioq, bp);
240 mtx_unlock(&ndisk->bioqlock);
241 taskqueue_enqueue(ndisk->tq, &ndisk->bioqtask);
245 nvd_gone(struct nvd_disk *ndisk)
249 printf(NVD_STR"%u: detached\n", ndisk->unit);
250 mtx_lock(&ndisk->bioqlock);
251 disk_gone(ndisk->disk);
252 while ((bp = bioq_takefirst(&ndisk->bioq)) != NULL) {
253 if (__predict_false(bp->bio_flags & BIO_ORDERED))
254 atomic_add_int(&ndisk->ordered_in_flight, -1);
255 bp->bio_error = ENXIO;
256 bp->bio_flags |= BIO_ERROR;
257 bp->bio_resid = bp->bio_bcount;
260 mtx_unlock(&ndisk->bioqlock);
264 nvd_gonecb(struct disk *dp)
266 struct nvd_disk *ndisk = (struct nvd_disk *)dp->d_drv1;
268 disk_destroy(ndisk->disk);
270 TAILQ_REMOVE(&disk_head, ndisk, global_tailq);
271 TAILQ_REMOVE(&ndisk->ctrlr->disk_head, ndisk, ctrlr_tailq);
272 if (TAILQ_EMPTY(&ndisk->ctrlr->disk_head))
273 wakeup(&ndisk->ctrlr->disk_head);
274 mtx_unlock(&nvd_lock);
275 taskqueue_free(ndisk->tq);
276 mtx_destroy(&ndisk->bioqlock);
281 nvd_ioctl(struct disk *dp, u_long cmd, void *data, int fflag,
284 struct nvd_disk *ndisk = dp->d_drv1;
286 return (nvme_ns_ioctl_process(ndisk->ns, cmd, data, fflag, td));
290 nvd_dump(void *arg, void *virt, vm_offset_t phys, off_t offset, size_t len)
292 struct disk *dp = arg;
293 struct nvd_disk *ndisk = dp->d_drv1;
295 return (nvme_ns_dump(ndisk->ns, virt, offset, len));
299 nvd_getattr(struct bio *bp)
301 struct nvd_disk *ndisk = (struct nvd_disk *)bp->bio_disk->d_drv1;
302 const struct nvme_namespace_data *nsdata;
305 if (!strcmp("GEOM::lunid", bp->bio_attribute)) {
306 nsdata = nvme_ns_get_data(ndisk->ns);
308 /* Try to return NGUID as lunid. */
309 for (i = 0; i < sizeof(nsdata->nguid); i++)
310 if (nsdata->nguid[i] != 0)
311 if (i < sizeof(nsdata->nguid)) {
312 if (bp->bio_length < sizeof(nsdata->nguid) * 2 + 1)
314 for (i = 0; i < sizeof(nsdata->nguid); i++) {
315 sprintf(&bp->bio_data[i * 2], "%02x",
318 bp->bio_completed = bp->bio_length;
322 /* Try to return EUI64 as lunid. */
323 for (i = 0; i < sizeof(nsdata->eui64); i++)
324 if (nsdata->eui64[i] != 0)
325 if (i < sizeof(nsdata->eui64)) {
326 if (bp->bio_length < sizeof(nsdata->eui64) * 2 + 1)
328 for (i = 0; i < sizeof(nsdata->eui64); i++) {
329 sprintf(&bp->bio_data[i * 2], "%02x",
332 bp->bio_completed = bp->bio_length;
340 nvd_done(void *arg, const struct nvme_completion *cpl)
342 struct bio *bp = (struct bio *)arg;
343 struct nvd_disk *ndisk = bp->bio_disk->d_drv1;
345 if (__predict_false(bp->bio_flags & BIO_ORDERED)) {
346 atomic_add_int(&ndisk->cur_depth, -NVD_ODEPTH);
347 atomic_add_int(&ndisk->ordered_in_flight, -1);
348 wakeup(&ndisk->cur_depth);
350 if (atomic_fetchadd_int(&ndisk->cur_depth, -1) == 1 &&
351 __predict_false(ndisk->ordered_in_flight != 0))
352 wakeup(&ndisk->cur_depth);
359 nvd_bioq_process(void *arg, int pending)
361 struct nvd_disk *ndisk = arg;
365 mtx_lock(&ndisk->bioqlock);
366 bp = bioq_takefirst(&ndisk->bioq);
367 mtx_unlock(&ndisk->bioqlock);
371 if (__predict_false(bp->bio_flags & BIO_ORDERED)) {
373 * bio with BIO_ORDERED flag set must be executed
374 * after all previous bios.
376 while (ndisk->cur_depth > 0)
377 tsleep(&ndisk->cur_depth, 0, "nvdorb", 1);
380 * bio with BIO_ORDERED flag set must be completed
381 * before proceeding with additional bios.
383 while (ndisk->cur_depth >= NVD_ODEPTH)
384 tsleep(&ndisk->cur_depth, 0, "nvdora", 1);
387 nvd_bio_submit(ndisk, bp);
392 nvd_new_controller(struct nvme_controller *ctrlr)
394 struct nvd_controller *nvd_ctrlr;
396 nvd_ctrlr = malloc(sizeof(struct nvd_controller), M_NVD,
399 TAILQ_INIT(&nvd_ctrlr->disk_head);
401 TAILQ_INSERT_TAIL(&ctrlr_head, nvd_ctrlr, tailq);
402 mtx_unlock(&nvd_lock);
408 nvd_new_disk(struct nvme_namespace *ns, void *ctrlr_arg)
410 uint8_t descr[NVME_MODEL_NUMBER_LENGTH+1];
411 struct nvd_disk *ndisk, *tnd;
413 struct nvd_controller *ctrlr = ctrlr_arg;
416 ndisk = malloc(sizeof(struct nvd_disk), M_NVD, M_ZERO | M_WAITOK);
417 ndisk->ctrlr = ctrlr;
419 ndisk->cur_depth = 0;
420 ndisk->ordered_in_flight = 0;
421 mtx_init(&ndisk->bioqlock, "nvd bioq lock", NULL, MTX_DEF);
422 bioq_init(&ndisk->bioq);
423 TASK_INIT(&ndisk->bioqtask, 0, nvd_bioq_process, ndisk);
427 TAILQ_FOREACH(tnd, &disk_head, global_tailq) {
428 if (tnd->unit > unit)
430 unit = tnd->unit + 1;
434 TAILQ_INSERT_BEFORE(tnd, ndisk, global_tailq);
436 TAILQ_INSERT_TAIL(&disk_head, ndisk, global_tailq);
437 TAILQ_INSERT_TAIL(&ctrlr->disk_head, ndisk, ctrlr_tailq);
438 mtx_unlock(&nvd_lock);
440 ndisk->tq = taskqueue_create("nvd_taskq", M_WAITOK,
441 taskqueue_thread_enqueue, &ndisk->tq);
442 taskqueue_start_threads(&ndisk->tq, 1, PI_DISK, "nvd taskq");
444 disk = ndisk->disk = disk_alloc();
445 disk->d_strategy = nvd_strategy;
446 disk->d_ioctl = nvd_ioctl;
447 disk->d_dump = nvd_dump;
448 disk->d_getattr = nvd_getattr;
449 disk->d_gone = nvd_gonecb;
450 disk->d_name = NVD_STR;
451 disk->d_unit = ndisk->unit;
452 disk->d_drv1 = ndisk;
454 disk->d_sectorsize = nvme_ns_get_sector_size(ns);
455 disk->d_mediasize = (off_t)nvme_ns_get_size(ns);
456 disk->d_maxsize = nvme_ns_get_max_io_xfer_size(ns);
457 disk->d_delmaxsize = (off_t)nvme_ns_get_size(ns);
458 if (disk->d_delmaxsize > nvd_delete_max)
459 disk->d_delmaxsize = nvd_delete_max;
460 disk->d_stripesize = nvme_ns_get_stripesize(ns);
461 disk->d_flags = DISKFLAG_UNMAPPED_BIO | DISKFLAG_DIRECT_COMPLETION;
462 if (nvme_ns_get_flags(ns) & NVME_NS_DEALLOCATE_SUPPORTED)
463 disk->d_flags |= DISKFLAG_CANDELETE;
464 if (nvme_ns_get_flags(ns) & NVME_NS_FLUSH_SUPPORTED)
465 disk->d_flags |= DISKFLAG_CANFLUSHCACHE;
468 * d_ident and d_descr are both far bigger than the length of either
469 * the serial or model number strings.
471 nvme_strvis(disk->d_ident, nvme_ns_get_serial_number(ns),
472 sizeof(disk->d_ident), NVME_SERIAL_NUMBER_LENGTH);
473 nvme_strvis(descr, nvme_ns_get_model_number(ns), sizeof(descr),
474 NVME_MODEL_NUMBER_LENGTH);
475 strlcpy(disk->d_descr, descr, sizeof(descr));
477 disk->d_rotation_rate = DISK_RR_NON_ROTATING;
479 disk_create(disk, DISK_VERSION);
481 printf(NVD_STR"%u: <%s> NVMe namespace\n", disk->d_unit, descr);
482 printf(NVD_STR"%u: %juMB (%ju %u byte sectors)\n", disk->d_unit,
483 (uintmax_t)disk->d_mediasize / (1024*1024),
484 (uintmax_t)disk->d_mediasize / disk->d_sectorsize,
491 nvd_controller_fail(void *ctrlr_arg)
493 struct nvd_controller *ctrlr = ctrlr_arg;
494 struct nvd_disk *ndisk;
497 TAILQ_REMOVE(&ctrlr_head, ctrlr, tailq);
498 TAILQ_FOREACH(ndisk, &ctrlr->disk_head, ctrlr_tailq)
500 while (!TAILQ_EMPTY(&ctrlr->disk_head))
501 msleep(&ctrlr->disk_head, &nvd_lock, 0, "nvd_fail", 0);
502 mtx_unlock(&nvd_lock);