2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (C) 2012-2016 Intel Corporation
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
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 AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
32 #include <sys/param.h>
34 #include <sys/kernel.h>
35 #include <sys/malloc.h>
36 #include <sys/module.h>
37 #include <sys/sysctl.h>
38 #include <sys/systm.h>
39 #include <sys/taskqueue.h>
41 #include <geom/geom.h>
42 #include <geom/geom_disk.h>
44 #include <dev/nvme/nvme.h>
50 static disk_ioctl_t nvd_ioctl;
51 static disk_strategy_t nvd_strategy;
52 static dumper_t nvd_dump;
54 static void nvd_done(void *arg, const struct nvme_completion *cpl);
56 static void *nvd_new_disk(struct nvme_namespace *ns, void *ctrlr);
57 static void destroy_geom_disk(struct nvd_disk *ndisk);
59 static void *nvd_new_controller(struct nvme_controller *ctrlr);
60 static void nvd_controller_fail(void *ctrlr);
62 static int nvd_load(void);
63 static void nvd_unload(void);
65 MALLOC_DEFINE(M_NVD, "nvd", "nvd(4) allocations");
67 struct nvme_consumer *consumer_handle;
71 struct bio_queue_head bioq;
77 struct nvme_namespace *ns;
80 uint32_t ordered_in_flight;
82 TAILQ_ENTRY(nvd_disk) global_tailq;
83 TAILQ_ENTRY(nvd_disk) ctrlr_tailq;
86 struct nvd_controller {
88 TAILQ_ENTRY(nvd_controller) tailq;
89 TAILQ_HEAD(, nvd_disk) disk_head;
92 static TAILQ_HEAD(, nvd_controller) ctrlr_head;
93 static TAILQ_HEAD(disk_list, nvd_disk) disk_head;
95 static SYSCTL_NODE(_hw, OID_AUTO, nvd, CTLFLAG_RD, 0, "nvd driver parameters");
97 * The NVMe specification does not define a maximum or optimal delete size, so
98 * technically max delete size is min(full size of the namespace, 2^32 - 1
99 * LBAs). A single delete for a multi-TB NVMe namespace though may take much
100 * longer to complete than the nvme(4) I/O timeout period. So choose a sensible
101 * default here that is still suitably large to minimize the number of overall
104 static uint64_t nvd_delete_max = (1024 * 1024 * 1024); /* 1GB */
105 SYSCTL_UQUAD(_hw_nvd, OID_AUTO, delete_max, CTLFLAG_RDTUN, &nvd_delete_max, 0,
106 "nvd maximum BIO_DELETE size in bytes");
108 static int nvd_modevent(module_t mod, int type, void *arg)
126 moduledata_t nvd_mod = {
128 (modeventhand_t)nvd_modevent,
132 DECLARE_MODULE(nvd, nvd_mod, SI_SUB_DRIVERS, SI_ORDER_ANY);
133 MODULE_VERSION(nvd, 1);
134 MODULE_DEPEND(nvd, nvme, 1, 1, 1);
142 TAILQ_INIT(&ctrlr_head);
143 TAILQ_INIT(&disk_head);
145 consumer_handle = nvme_register_consumer(nvd_new_disk,
146 nvd_new_controller, NULL, nvd_controller_fail);
148 return (consumer_handle != NULL ? 0 : -1);
154 struct nvd_controller *ctrlr;
155 struct nvd_disk *disk;
160 while (!TAILQ_EMPTY(&ctrlr_head)) {
161 ctrlr = TAILQ_FIRST(&ctrlr_head);
162 TAILQ_REMOVE(&ctrlr_head, ctrlr, tailq);
166 while (!TAILQ_EMPTY(&disk_head)) {
167 disk = TAILQ_FIRST(&disk_head);
168 TAILQ_REMOVE(&disk_head, disk, global_tailq);
169 destroy_geom_disk(disk);
173 nvme_unregister_consumer(consumer_handle);
177 nvd_bio_submit(struct nvd_disk *ndisk, struct bio *bp)
181 bp->bio_driver1 = NULL;
182 atomic_add_int(&ndisk->cur_depth, 1);
183 err = nvme_ns_bio_process(ndisk->ns, bp, nvd_done);
185 atomic_add_int(&ndisk->cur_depth, -1);
186 if (__predict_false(bp->bio_flags & BIO_ORDERED))
187 atomic_add_int(&ndisk->ordered_in_flight, -1);
189 bp->bio_flags |= BIO_ERROR;
190 bp->bio_resid = bp->bio_bcount;
199 nvd_strategy(struct bio *bp)
201 struct nvd_disk *ndisk;
203 ndisk = (struct nvd_disk *)bp->bio_disk->d_drv1;
205 if (__predict_false(bp->bio_flags & BIO_ORDERED))
206 atomic_add_int(&ndisk->ordered_in_flight, 1);
208 if (__predict_true(ndisk->ordered_in_flight == 0)) {
209 nvd_bio_submit(ndisk, bp);
214 * There are ordered bios in flight, so we need to submit
215 * bios through the task queue to enforce ordering.
217 mtx_lock(&ndisk->bioqlock);
218 bioq_insert_tail(&ndisk->bioq, bp);
219 mtx_unlock(&ndisk->bioqlock);
220 taskqueue_enqueue(ndisk->tq, &ndisk->bioqtask);
224 nvd_ioctl(struct disk *ndisk, u_long cmd, void *data, int fflag,
238 nvd_dump(void *arg, void *virt, vm_offset_t phys, off_t offset, size_t len)
240 struct nvd_disk *ndisk;
246 return (nvme_ns_dump(ndisk->ns, virt, offset, len));
250 nvd_done(void *arg, const struct nvme_completion *cpl)
253 struct nvd_disk *ndisk;
255 bp = (struct bio *)arg;
257 ndisk = bp->bio_disk->d_drv1;
259 atomic_add_int(&ndisk->cur_depth, -1);
260 if (__predict_false(bp->bio_flags & BIO_ORDERED))
261 atomic_add_int(&ndisk->ordered_in_flight, -1);
267 nvd_bioq_process(void *arg, int pending)
269 struct nvd_disk *ndisk = arg;
273 mtx_lock(&ndisk->bioqlock);
274 bp = bioq_takefirst(&ndisk->bioq);
275 mtx_unlock(&ndisk->bioqlock);
279 if (nvd_bio_submit(ndisk, bp) != 0) {
285 * BIO_ORDERED flag dictates that the bio with BIO_ORDERED
286 * flag set must be completed before proceeding with
289 if (bp->bio_flags & BIO_ORDERED) {
290 while (ndisk->cur_depth > 0) {
291 pause("nvd flush", 1);
299 nvd_new_controller(struct nvme_controller *ctrlr)
301 struct nvd_controller *nvd_ctrlr;
303 nvd_ctrlr = malloc(sizeof(struct nvd_controller), M_NVD,
306 TAILQ_INIT(&nvd_ctrlr->disk_head);
307 TAILQ_INSERT_TAIL(&ctrlr_head, nvd_ctrlr, tailq);
313 nvd_new_disk(struct nvme_namespace *ns, void *ctrlr_arg)
315 uint8_t descr[NVME_MODEL_NUMBER_LENGTH+1];
316 struct nvd_disk *ndisk;
318 struct nvd_controller *ctrlr = ctrlr_arg;
320 ndisk = malloc(sizeof(struct nvd_disk), M_NVD, M_ZERO | M_WAITOK);
323 disk->d_strategy = nvd_strategy;
324 disk->d_ioctl = nvd_ioctl;
325 disk->d_dump = nvd_dump;
326 disk->d_name = NVD_STR;
327 disk->d_drv1 = ndisk;
329 disk->d_maxsize = nvme_ns_get_max_io_xfer_size(ns);
330 disk->d_sectorsize = nvme_ns_get_sector_size(ns);
331 disk->d_mediasize = (off_t)nvme_ns_get_size(ns);
332 disk->d_delmaxsize = (off_t)nvme_ns_get_size(ns);
333 if (disk->d_delmaxsize > nvd_delete_max)
334 disk->d_delmaxsize = nvd_delete_max;
335 disk->d_stripesize = nvme_ns_get_stripesize(ns);
337 if (TAILQ_EMPTY(&disk_head))
341 TAILQ_LAST(&disk_head, disk_list)->disk->d_unit + 1;
343 disk->d_flags = DISKFLAG_DIRECT_COMPLETION;
345 if (nvme_ns_get_flags(ns) & NVME_NS_DEALLOCATE_SUPPORTED)
346 disk->d_flags |= DISKFLAG_CANDELETE;
348 if (nvme_ns_get_flags(ns) & NVME_NS_FLUSH_SUPPORTED)
349 disk->d_flags |= DISKFLAG_CANFLUSHCACHE;
351 /* ifdef used here to ease porting to stable branches at a later point. */
352 #ifdef DISKFLAG_UNMAPPED_BIO
353 disk->d_flags |= DISKFLAG_UNMAPPED_BIO;
357 * d_ident and d_descr are both far bigger than the length of either
358 * the serial or model number strings.
360 nvme_strvis(disk->d_ident, nvme_ns_get_serial_number(ns),
361 sizeof(disk->d_ident), NVME_SERIAL_NUMBER_LENGTH);
362 nvme_strvis(descr, nvme_ns_get_model_number(ns), sizeof(descr),
363 NVME_MODEL_NUMBER_LENGTH);
364 strlcpy(disk->d_descr, descr, sizeof(descr));
366 disk->d_rotation_rate = DISK_RR_NON_ROTATING;
370 ndisk->cur_depth = 0;
371 ndisk->ordered_in_flight = 0;
373 mtx_init(&ndisk->bioqlock, "NVD bioq lock", NULL, MTX_DEF);
374 bioq_init(&ndisk->bioq);
376 TASK_INIT(&ndisk->bioqtask, 0, nvd_bioq_process, ndisk);
377 ndisk->tq = taskqueue_create("nvd_taskq", M_WAITOK,
378 taskqueue_thread_enqueue, &ndisk->tq);
379 taskqueue_start_threads(&ndisk->tq, 1, PI_DISK, "nvd taskq");
381 TAILQ_INSERT_TAIL(&disk_head, ndisk, global_tailq);
382 TAILQ_INSERT_TAIL(&ctrlr->disk_head, ndisk, ctrlr_tailq);
384 disk_create(disk, DISK_VERSION);
386 printf(NVD_STR"%u: <%s> NVMe namespace\n", disk->d_unit, descr);
387 printf(NVD_STR"%u: %juMB (%ju %u byte sectors)\n", disk->d_unit,
388 (uintmax_t)disk->d_mediasize / (1024*1024),
389 (uintmax_t)disk->d_mediasize / disk->d_sectorsize,
396 destroy_geom_disk(struct nvd_disk *ndisk)
405 taskqueue_free(ndisk->tq);
407 disk_destroy(ndisk->disk);
409 mtx_lock(&ndisk->bioqlock);
411 bp = bioq_takefirst(&ndisk->bioq);
415 bp->bio_flags |= BIO_ERROR;
416 bp->bio_resid = bp->bio_bcount;
421 printf(NVD_STR"%u: lost device - %d outstanding\n", unit, cnt);
422 printf(NVD_STR"%u: removing device entry\n", unit);
424 mtx_unlock(&ndisk->bioqlock);
426 mtx_destroy(&ndisk->bioqlock);
430 nvd_controller_fail(void *ctrlr_arg)
432 struct nvd_controller *ctrlr = ctrlr_arg;
433 struct nvd_disk *disk;
435 while (!TAILQ_EMPTY(&ctrlr->disk_head)) {
436 disk = TAILQ_FIRST(&ctrlr->disk_head);
437 TAILQ_REMOVE(&disk_head, disk, global_tailq);
438 TAILQ_REMOVE(&ctrlr->disk_head, disk, ctrlr_tailq);
439 destroy_geom_disk(disk);
443 TAILQ_REMOVE(&ctrlr_head, ctrlr, tailq);