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;
58 static void nvd_done(void *arg, const struct nvme_completion *cpl);
59 static void nvd_gone(struct nvd_disk *ndisk);
61 static void *nvd_new_disk(struct nvme_namespace *ns, void *ctrlr);
63 static void *nvd_new_controller(struct nvme_controller *ctrlr);
64 static void nvd_controller_fail(void *ctrlr);
66 static int nvd_load(void);
67 static void nvd_unload(void);
69 MALLOC_DEFINE(M_NVD, "nvd", "nvd(4) allocations");
71 struct nvme_consumer *consumer_handle;
74 struct nvd_controller *ctrlr;
76 struct bio_queue_head bioq;
82 struct nvme_namespace *ns;
85 uint32_t ordered_in_flight;
88 TAILQ_ENTRY(nvd_disk) global_tailq;
89 TAILQ_ENTRY(nvd_disk) ctrlr_tailq;
92 struct nvd_controller {
94 TAILQ_ENTRY(nvd_controller) tailq;
95 TAILQ_HEAD(, nvd_disk) disk_head;
98 static struct mtx nvd_lock;
99 static TAILQ_HEAD(, nvd_controller) ctrlr_head;
100 static TAILQ_HEAD(disk_list, nvd_disk) disk_head;
102 static SYSCTL_NODE(_hw, OID_AUTO, nvd, CTLFLAG_RD, 0, "nvd driver parameters");
104 * The NVMe specification does not define a maximum or optimal delete size, so
105 * technically max delete size is min(full size of the namespace, 2^32 - 1
106 * LBAs). A single delete for a multi-TB NVMe namespace though may take much
107 * longer to complete than the nvme(4) I/O timeout period. So choose a sensible
108 * default here that is still suitably large to minimize the number of overall
111 static uint64_t nvd_delete_max = (1024 * 1024 * 1024); /* 1GB */
112 SYSCTL_UQUAD(_hw_nvd, OID_AUTO, delete_max, CTLFLAG_RDTUN, &nvd_delete_max, 0,
113 "nvd maximum BIO_DELETE size in bytes");
115 static int nvd_modevent(module_t mod, int type, void *arg)
133 moduledata_t nvd_mod = {
135 (modeventhand_t)nvd_modevent,
139 DECLARE_MODULE(nvd, nvd_mod, SI_SUB_DRIVERS, SI_ORDER_ANY);
140 MODULE_VERSION(nvd, 1);
141 MODULE_DEPEND(nvd, nvme, 1, 1, 1);
149 mtx_init(&nvd_lock, "nvd_lock", NULL, MTX_DEF);
150 TAILQ_INIT(&ctrlr_head);
151 TAILQ_INIT(&disk_head);
153 consumer_handle = nvme_register_consumer(nvd_new_disk,
154 nvd_new_controller, NULL, nvd_controller_fail);
156 return (consumer_handle != NULL ? 0 : -1);
162 struct nvd_controller *ctrlr;
163 struct nvd_disk *ndisk;
169 while ((ctrlr = TAILQ_FIRST(&ctrlr_head)) != NULL) {
170 TAILQ_REMOVE(&ctrlr_head, ctrlr, tailq);
171 TAILQ_FOREACH(ndisk, &ctrlr->disk_head, ctrlr_tailq)
173 while (!TAILQ_EMPTY(&ctrlr->disk_head))
174 msleep(&ctrlr->disk_head, &nvd_lock, 0, "nvd_unload",0);
177 mtx_unlock(&nvd_lock);
179 nvme_unregister_consumer(consumer_handle);
181 mtx_destroy(&nvd_lock);
185 nvd_bio_submit(struct nvd_disk *ndisk, struct bio *bp)
189 bp->bio_driver1 = NULL;
190 atomic_add_int(&ndisk->cur_depth, 1);
191 err = nvme_ns_bio_process(ndisk->ns, bp, nvd_done);
193 atomic_add_int(&ndisk->cur_depth, -1);
194 if (__predict_false(bp->bio_flags & BIO_ORDERED))
195 atomic_add_int(&ndisk->ordered_in_flight, -1);
197 bp->bio_flags |= BIO_ERROR;
198 bp->bio_resid = bp->bio_bcount;
207 nvd_strategy(struct bio *bp)
209 struct nvd_disk *ndisk;
211 ndisk = (struct nvd_disk *)bp->bio_disk->d_drv1;
213 if (__predict_false(bp->bio_flags & BIO_ORDERED))
214 atomic_add_int(&ndisk->ordered_in_flight, 1);
216 if (__predict_true(ndisk->ordered_in_flight == 0)) {
217 nvd_bio_submit(ndisk, bp);
222 * There are ordered bios in flight, so we need to submit
223 * bios through the task queue to enforce ordering.
225 mtx_lock(&ndisk->bioqlock);
226 bioq_insert_tail(&ndisk->bioq, bp);
227 mtx_unlock(&ndisk->bioqlock);
228 taskqueue_enqueue(ndisk->tq, &ndisk->bioqtask);
232 nvd_gone(struct nvd_disk *ndisk)
236 printf(NVD_STR"%u: detached\n", ndisk->unit);
237 mtx_lock(&ndisk->bioqlock);
238 disk_gone(ndisk->disk);
239 while ((bp = bioq_takefirst(&ndisk->bioq)) != NULL) {
240 if (__predict_false(bp->bio_flags & BIO_ORDERED))
241 atomic_add_int(&ndisk->ordered_in_flight, -1);
242 bp->bio_error = ENXIO;
243 bp->bio_flags |= BIO_ERROR;
244 bp->bio_resid = bp->bio_bcount;
247 mtx_unlock(&ndisk->bioqlock);
251 nvd_gonecb(struct disk *dp)
253 struct nvd_disk *ndisk = (struct nvd_disk *)dp->d_drv1;
255 disk_destroy(ndisk->disk);
257 TAILQ_REMOVE(&disk_head, ndisk, global_tailq);
258 TAILQ_REMOVE(&ndisk->ctrlr->disk_head, ndisk, ctrlr_tailq);
259 if (TAILQ_EMPTY(&ndisk->ctrlr->disk_head))
260 wakeup(&ndisk->ctrlr->disk_head);
261 mtx_unlock(&nvd_lock);
262 taskqueue_free(ndisk->tq);
263 mtx_destroy(&ndisk->bioqlock);
268 nvd_ioctl(struct disk *ndisk, u_long cmd, void *data, int fflag,
282 nvd_dump(void *arg, void *virt, vm_offset_t phys, off_t offset, size_t len)
284 struct nvd_disk *ndisk;
290 return (nvme_ns_dump(ndisk->ns, virt, offset, len));
294 nvd_done(void *arg, const struct nvme_completion *cpl)
297 struct nvd_disk *ndisk;
299 bp = (struct bio *)arg;
301 ndisk = bp->bio_disk->d_drv1;
303 atomic_add_int(&ndisk->cur_depth, -1);
304 if (__predict_false(bp->bio_flags & BIO_ORDERED))
305 atomic_add_int(&ndisk->ordered_in_flight, -1);
311 nvd_bioq_process(void *arg, int pending)
313 struct nvd_disk *ndisk = arg;
317 mtx_lock(&ndisk->bioqlock);
318 bp = bioq_takefirst(&ndisk->bioq);
319 mtx_unlock(&ndisk->bioqlock);
323 if (nvd_bio_submit(ndisk, bp) != 0) {
329 * BIO_ORDERED flag dictates that the bio with BIO_ORDERED
330 * flag set must be completed before proceeding with
333 if (bp->bio_flags & BIO_ORDERED) {
334 while (ndisk->cur_depth > 0) {
335 pause("nvd flush", 1);
343 nvd_new_controller(struct nvme_controller *ctrlr)
345 struct nvd_controller *nvd_ctrlr;
347 nvd_ctrlr = malloc(sizeof(struct nvd_controller), M_NVD,
350 TAILQ_INIT(&nvd_ctrlr->disk_head);
352 TAILQ_INSERT_TAIL(&ctrlr_head, nvd_ctrlr, tailq);
353 mtx_unlock(&nvd_lock);
359 nvd_new_disk(struct nvme_namespace *ns, void *ctrlr_arg)
361 uint8_t descr[NVME_MODEL_NUMBER_LENGTH+1];
362 struct nvd_disk *ndisk, *tnd;
364 struct nvd_controller *ctrlr = ctrlr_arg;
367 ndisk = malloc(sizeof(struct nvd_disk), M_NVD, M_ZERO | M_WAITOK);
368 ndisk->ctrlr = ctrlr;
370 ndisk->cur_depth = 0;
371 ndisk->ordered_in_flight = 0;
372 mtx_init(&ndisk->bioqlock, "nvd bioq lock", NULL, MTX_DEF);
373 bioq_init(&ndisk->bioq);
374 TASK_INIT(&ndisk->bioqtask, 0, nvd_bioq_process, ndisk);
378 TAILQ_FOREACH(tnd, &disk_head, global_tailq) {
379 if (tnd->unit > unit)
381 unit = tnd->unit + 1;
385 TAILQ_INSERT_BEFORE(tnd, ndisk, global_tailq);
387 TAILQ_INSERT_TAIL(&disk_head, ndisk, global_tailq);
388 TAILQ_INSERT_TAIL(&ctrlr->disk_head, ndisk, ctrlr_tailq);
389 mtx_unlock(&nvd_lock);
391 ndisk->tq = taskqueue_create("nvd_taskq", M_WAITOK,
392 taskqueue_thread_enqueue, &ndisk->tq);
393 taskqueue_start_threads(&ndisk->tq, 1, PI_DISK, "nvd taskq");
395 disk = ndisk->disk = disk_alloc();
396 disk->d_strategy = nvd_strategy;
397 disk->d_ioctl = nvd_ioctl;
398 disk->d_dump = nvd_dump;
399 disk->d_gone = nvd_gonecb;
400 disk->d_name = NVD_STR;
401 disk->d_unit = ndisk->unit;
402 disk->d_drv1 = ndisk;
404 disk->d_sectorsize = nvme_ns_get_sector_size(ns);
405 disk->d_mediasize = (off_t)nvme_ns_get_size(ns);
406 disk->d_maxsize = nvme_ns_get_max_io_xfer_size(ns);
407 disk->d_delmaxsize = (off_t)nvme_ns_get_size(ns);
408 if (disk->d_delmaxsize > nvd_delete_max)
409 disk->d_delmaxsize = nvd_delete_max;
410 disk->d_stripesize = nvme_ns_get_stripesize(ns);
411 disk->d_flags = DISKFLAG_UNMAPPED_BIO | DISKFLAG_DIRECT_COMPLETION;
412 if (nvme_ns_get_flags(ns) & NVME_NS_DEALLOCATE_SUPPORTED)
413 disk->d_flags |= DISKFLAG_CANDELETE;
414 if (nvme_ns_get_flags(ns) & NVME_NS_FLUSH_SUPPORTED)
415 disk->d_flags |= DISKFLAG_CANFLUSHCACHE;
418 * d_ident and d_descr are both far bigger than the length of either
419 * the serial or model number strings.
421 nvme_strvis(disk->d_ident, nvme_ns_get_serial_number(ns),
422 sizeof(disk->d_ident), NVME_SERIAL_NUMBER_LENGTH);
423 nvme_strvis(descr, nvme_ns_get_model_number(ns), sizeof(descr),
424 NVME_MODEL_NUMBER_LENGTH);
425 strlcpy(disk->d_descr, descr, sizeof(descr));
427 disk->d_rotation_rate = DISK_RR_NON_ROTATING;
429 disk_create(disk, DISK_VERSION);
431 printf(NVD_STR"%u: <%s> NVMe namespace\n", disk->d_unit, descr);
432 printf(NVD_STR"%u: %juMB (%ju %u byte sectors)\n", disk->d_unit,
433 (uintmax_t)disk->d_mediasize / (1024*1024),
434 (uintmax_t)disk->d_mediasize / disk->d_sectorsize,
441 nvd_controller_fail(void *ctrlr_arg)
443 struct nvd_controller *ctrlr = ctrlr_arg;
444 struct nvd_disk *ndisk;
447 TAILQ_REMOVE(&ctrlr_head, ctrlr, tailq);
448 TAILQ_FOREACH(ndisk, &ctrlr->disk_head, ctrlr_tailq)
450 while (!TAILQ_EMPTY(&ctrlr->disk_head))
451 msleep(&ctrlr->disk_head, &nvd_lock, 0, "nvd_fail", 0);
452 mtx_unlock(&nvd_lock);