nvd.c revision 344033 
1/*-
2 * Copyright (C) 2012-2016 Intel Corporation
3 * All rights reserved.
4 * Copyright (C) 2018 Alexander Motin <mav@FreeBSD.org>
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 * 1. Redistributions of source code must retain the above copyright
10 *    notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 *    notice, this list of conditions and the following disclaimer in the
13 *    documentation and/or other materials provided with the distribution.
14 *
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25 * SUCH DAMAGE.
26 */
27
28#include <sys/cdefs.h>
29__FBSDID("$FreeBSD: stable/11/sys/dev/nvd/nvd.c 344033 2019-02-12 00:53:43Z mav $");
30
31#include <sys/param.h>
32#include <sys/bio.h>
33#include <sys/kernel.h>
34#include <sys/malloc.h>
35#include <sys/module.h>
36#include <sys/queue.h>
37#include <sys/sysctl.h>
38#include <sys/systm.h>
39#include <sys/taskqueue.h>
40#include <machine/atomic.h>
41
42#include <geom/geom.h>
43#include <geom/geom_disk.h>
44
45#include <dev/nvme/nvme.h>
46
47#define NVD_STR		"nvd"
48
49struct nvd_disk;
50struct nvd_controller;
51
52static disk_ioctl_t nvd_ioctl;
53static disk_strategy_t nvd_strategy;
54static dumper_t nvd_dump;
55
56static void nvd_done(void *arg, const struct nvme_completion *cpl);
57static void nvd_gone(struct nvd_disk *ndisk);
58
59static void *nvd_new_disk(struct nvme_namespace *ns, void *ctrlr);
60
61static void *nvd_new_controller(struct nvme_controller *ctrlr);
62static void nvd_controller_fail(void *ctrlr);
63
64static int nvd_load(void);
65static void nvd_unload(void);
66
67MALLOC_DEFINE(M_NVD, "nvd", "nvd(4) allocations");
68
69struct nvme_consumer *consumer_handle;
70
71struct nvd_disk {
72	struct nvd_controller	*ctrlr;
73
74	struct bio_queue_head	bioq;
75	struct task		bioqtask;
76	struct mtx		bioqlock;
77
78	struct disk		*disk;
79	struct taskqueue	*tq;
80	struct nvme_namespace	*ns;
81
82	uint32_t		cur_depth;
83#define	NVD_ODEPTH	(1 << 30)
84	uint32_t		ordered_in_flight;
85	u_int			unit;
86
87	TAILQ_ENTRY(nvd_disk)	global_tailq;
88	TAILQ_ENTRY(nvd_disk)	ctrlr_tailq;
89};
90
91struct nvd_controller {
92
93	TAILQ_ENTRY(nvd_controller)	tailq;
94	TAILQ_HEAD(, nvd_disk)		disk_head;
95};
96
97static struct mtx			nvd_lock;
98static TAILQ_HEAD(, nvd_controller)	ctrlr_head;
99static TAILQ_HEAD(disk_list, nvd_disk)	disk_head;
100
101static SYSCTL_NODE(_hw, OID_AUTO, nvd, CTLFLAG_RD, 0, "nvd driver parameters");
102/*
103 * The NVMe specification does not define a maximum or optimal delete size, so
104 *  technically max delete size is min(full size of the namespace, 2^32 - 1
105 *  LBAs).  A single delete for a multi-TB NVMe namespace though may take much
106 *  longer to complete than the nvme(4) I/O timeout period.  So choose a sensible
107 *  default here that is still suitably large to minimize the number of overall
108 *  delete operations.
109 */
110static uint64_t nvd_delete_max = (1024 * 1024 * 1024);  /* 1GB */
111SYSCTL_UQUAD(_hw_nvd, OID_AUTO, delete_max, CTLFLAG_RDTUN, &nvd_delete_max, 0,
112	     "nvd maximum BIO_DELETE size in bytes");
113
114static int nvd_modevent(module_t mod, int type, void *arg)
115{
116	int error = 0;
117
118	switch (type) {
119	case MOD_LOAD:
120		error = nvd_load();
121		break;
122	case MOD_UNLOAD:
123		nvd_unload();
124		break;
125	default:
126		break;
127	}
128
129	return (error);
130}
131
132moduledata_t nvd_mod = {
133	NVD_STR,
134	(modeventhand_t)nvd_modevent,
135	0
136};
137
138DECLARE_MODULE(nvd, nvd_mod, SI_SUB_DRIVERS, SI_ORDER_ANY);
139MODULE_VERSION(nvd, 1);
140MODULE_DEPEND(nvd, nvme, 1, 1, 1);
141
142static int
143nvd_load()
144{
145	if (!nvme_use_nvd)
146		return 0;
147
148	mtx_init(&nvd_lock, "nvd_lock", NULL, MTX_DEF);
149	TAILQ_INIT(&ctrlr_head);
150	TAILQ_INIT(&disk_head);
151
152	consumer_handle = nvme_register_consumer(nvd_new_disk,
153	    nvd_new_controller, NULL, nvd_controller_fail);
154
155	return (consumer_handle != NULL ? 0 : -1);
156}
157
158static void
159nvd_unload()
160{
161	struct nvd_controller	*ctrlr;
162	struct nvd_disk		*ndisk;
163
164	if (!nvme_use_nvd)
165		return;
166
167	mtx_lock(&nvd_lock);
168	while ((ctrlr = TAILQ_FIRST(&ctrlr_head)) != NULL) {
169		TAILQ_REMOVE(&ctrlr_head, ctrlr, tailq);
170		TAILQ_FOREACH(ndisk, &ctrlr->disk_head, ctrlr_tailq)
171			nvd_gone(ndisk);
172		while (!TAILQ_EMPTY(&ctrlr->disk_head))
173			msleep(&ctrlr->disk_head, &nvd_lock, 0, "nvd_unload",0);
174		free(ctrlr, M_NVD);
175	}
176	mtx_unlock(&nvd_lock);
177
178	nvme_unregister_consumer(consumer_handle);
179
180	mtx_destroy(&nvd_lock);
181}
182
183static void
184nvd_bio_submit(struct nvd_disk *ndisk, struct bio *bp)
185{
186	int err;
187
188	bp->bio_driver1 = NULL;
189	if (__predict_false(bp->bio_flags & BIO_ORDERED))
190		atomic_add_int(&ndisk->cur_depth, NVD_ODEPTH);
191	else
192		atomic_add_int(&ndisk->cur_depth, 1);
193	err = nvme_ns_bio_process(ndisk->ns, bp, nvd_done);
194	if (err) {
195		if (__predict_false(bp->bio_flags & BIO_ORDERED)) {
196			atomic_add_int(&ndisk->cur_depth, -NVD_ODEPTH);
197			atomic_add_int(&ndisk->ordered_in_flight, -1);
198			wakeup(&ndisk->cur_depth);
199		} else {
200			if (atomic_fetchadd_int(&ndisk->cur_depth, -1) == 1 &&
201			    __predict_false(ndisk->ordered_in_flight != 0))
202				wakeup(&ndisk->cur_depth);
203		}
204		bp->bio_error = err;
205		bp->bio_flags |= BIO_ERROR;
206		bp->bio_resid = bp->bio_bcount;
207		biodone(bp);
208	}
209}
210
211static void
212nvd_strategy(struct bio *bp)
213{
214	struct nvd_disk *ndisk = (struct nvd_disk *)bp->bio_disk->d_drv1;
215
216	/*
217	 * bio with BIO_ORDERED flag must be executed after all previous
218	 * bios in the queue, and before any successive bios.
219	 */
220	if (__predict_false(bp->bio_flags & BIO_ORDERED)) {
221		if (atomic_fetchadd_int(&ndisk->ordered_in_flight, 1) == 0 &&
222		    ndisk->cur_depth == 0 && bioq_first(&ndisk->bioq) == NULL) {
223			nvd_bio_submit(ndisk, bp);
224			return;
225		}
226	} else if (__predict_true(ndisk->ordered_in_flight == 0)) {
227		nvd_bio_submit(ndisk, bp);
228		return;
229	}
230
231	/*
232	 * There are ordered bios in flight, so we need to submit
233	 *  bios through the task queue to enforce ordering.
234	 */
235	mtx_lock(&ndisk->bioqlock);
236	bioq_insert_tail(&ndisk->bioq, bp);
237	mtx_unlock(&ndisk->bioqlock);
238	taskqueue_enqueue(ndisk->tq, &ndisk->bioqtask);
239}
240
241static void
242nvd_gone(struct nvd_disk *ndisk)
243{
244	struct bio	*bp;
245
246	printf(NVD_STR"%u: detached\n", ndisk->unit);
247	mtx_lock(&ndisk->bioqlock);
248	disk_gone(ndisk->disk);
249	while ((bp = bioq_takefirst(&ndisk->bioq)) != NULL) {
250		if (__predict_false(bp->bio_flags & BIO_ORDERED))
251			atomic_add_int(&ndisk->ordered_in_flight, -1);
252		bp->bio_error = ENXIO;
253		bp->bio_flags |= BIO_ERROR;
254		bp->bio_resid = bp->bio_bcount;
255		biodone(bp);
256	}
257	mtx_unlock(&ndisk->bioqlock);
258}
259
260static void
261nvd_gonecb(struct disk *dp)
262{
263	struct nvd_disk *ndisk = (struct nvd_disk *)dp->d_drv1;
264
265	disk_destroy(ndisk->disk);
266	mtx_lock(&nvd_lock);
267	TAILQ_REMOVE(&disk_head, ndisk, global_tailq);
268	TAILQ_REMOVE(&ndisk->ctrlr->disk_head, ndisk, ctrlr_tailq);
269	if (TAILQ_EMPTY(&ndisk->ctrlr->disk_head))
270		wakeup(&ndisk->ctrlr->disk_head);
271	mtx_unlock(&nvd_lock);
272	taskqueue_free(ndisk->tq);
273	mtx_destroy(&ndisk->bioqlock);
274	free(ndisk, M_NVD);
275}
276
277static int
278nvd_ioctl(struct disk *ndisk, u_long cmd, void *data, int fflag,
279    struct thread *td)
280{
281	int ret = 0;
282
283	switch (cmd) {
284	default:
285		ret = EIO;
286	}
287
288	return (ret);
289}
290
291static int
292nvd_dump(void *arg, void *virt, vm_offset_t phys, off_t offset, size_t len)
293{
294	struct disk *dp = arg;
295	struct nvd_disk *ndisk = dp->d_drv1;
296
297	return (nvme_ns_dump(ndisk->ns, virt, offset, len));
298}
299
300static void
301nvd_done(void *arg, const struct nvme_completion *cpl)
302{
303	struct bio *bp = (struct bio *)arg;
304	struct nvd_disk *ndisk = bp->bio_disk->d_drv1;
305
306	if (__predict_false(bp->bio_flags & BIO_ORDERED)) {
307		atomic_add_int(&ndisk->cur_depth, -NVD_ODEPTH);
308		atomic_add_int(&ndisk->ordered_in_flight, -1);
309		wakeup(&ndisk->cur_depth);
310	} else {
311		if (atomic_fetchadd_int(&ndisk->cur_depth, -1) == 1 &&
312		    __predict_false(ndisk->ordered_in_flight != 0))
313			wakeup(&ndisk->cur_depth);
314	}
315
316	biodone(bp);
317}
318
319static void
320nvd_bioq_process(void *arg, int pending)
321{
322	struct nvd_disk *ndisk = arg;
323	struct bio *bp;
324
325	for (;;) {
326		mtx_lock(&ndisk->bioqlock);
327		bp = bioq_takefirst(&ndisk->bioq);
328		mtx_unlock(&ndisk->bioqlock);
329		if (bp == NULL)
330			break;
331
332		if (__predict_false(bp->bio_flags & BIO_ORDERED)) {
333			/*
334			 * bio with BIO_ORDERED flag set must be executed
335			 * after all previous bios.
336			 */
337			while (ndisk->cur_depth > 0)
338				tsleep(&ndisk->cur_depth, 0, "nvdorb", 1);
339		} else {
340			/*
341			 * bio with BIO_ORDERED flag set must be completed
342			 * before proceeding with additional bios.
343			 */
344			while (ndisk->cur_depth >= NVD_ODEPTH)
345				tsleep(&ndisk->cur_depth, 0, "nvdora", 1);
346		}
347
348		nvd_bio_submit(ndisk, bp);
349	}
350}
351
352static void *
353nvd_new_controller(struct nvme_controller *ctrlr)
354{
355	struct nvd_controller	*nvd_ctrlr;
356
357	nvd_ctrlr = malloc(sizeof(struct nvd_controller), M_NVD,
358	    M_ZERO | M_WAITOK);
359
360	TAILQ_INIT(&nvd_ctrlr->disk_head);
361	mtx_lock(&nvd_lock);
362	TAILQ_INSERT_TAIL(&ctrlr_head, nvd_ctrlr, tailq);
363	mtx_unlock(&nvd_lock);
364
365	return (nvd_ctrlr);
366}
367
368static void *
369nvd_new_disk(struct nvme_namespace *ns, void *ctrlr_arg)
370{
371	uint8_t			descr[NVME_MODEL_NUMBER_LENGTH+1];
372	struct nvd_disk		*ndisk, *tnd;
373	struct disk		*disk;
374	struct nvd_controller	*ctrlr = ctrlr_arg;
375	int unit;
376
377	ndisk = malloc(sizeof(struct nvd_disk), M_NVD, M_ZERO | M_WAITOK);
378	ndisk->ctrlr = ctrlr;
379	ndisk->ns = ns;
380	ndisk->cur_depth = 0;
381	ndisk->ordered_in_flight = 0;
382	mtx_init(&ndisk->bioqlock, "nvd bioq lock", NULL, MTX_DEF);
383	bioq_init(&ndisk->bioq);
384	TASK_INIT(&ndisk->bioqtask, 0, nvd_bioq_process, ndisk);
385
386	mtx_lock(&nvd_lock);
387	unit = 0;
388	TAILQ_FOREACH(tnd, &disk_head, global_tailq) {
389		if (tnd->unit > unit)
390			break;
391		unit = tnd->unit + 1;
392	}
393	ndisk->unit = unit;
394	if (tnd != NULL)
395		TAILQ_INSERT_BEFORE(tnd, ndisk, global_tailq);
396	else
397		TAILQ_INSERT_TAIL(&disk_head, ndisk, global_tailq);
398	TAILQ_INSERT_TAIL(&ctrlr->disk_head, ndisk, ctrlr_tailq);
399	mtx_unlock(&nvd_lock);
400
401	ndisk->tq = taskqueue_create("nvd_taskq", M_WAITOK,
402	    taskqueue_thread_enqueue, &ndisk->tq);
403	taskqueue_start_threads(&ndisk->tq, 1, PI_DISK, "nvd taskq");
404
405	disk = ndisk->disk = disk_alloc();
406	disk->d_strategy = nvd_strategy;
407	disk->d_ioctl = nvd_ioctl;
408	disk->d_dump = nvd_dump;
409	disk->d_gone = nvd_gonecb;
410	disk->d_name = NVD_STR;
411	disk->d_unit = ndisk->unit;
412	disk->d_drv1 = ndisk;
413
414	disk->d_sectorsize = nvme_ns_get_sector_size(ns);
415	disk->d_mediasize = (off_t)nvme_ns_get_size(ns);
416	disk->d_maxsize = nvme_ns_get_max_io_xfer_size(ns);
417	disk->d_delmaxsize = (off_t)nvme_ns_get_size(ns);
418	if (disk->d_delmaxsize > nvd_delete_max)
419		disk->d_delmaxsize = nvd_delete_max;
420	disk->d_stripesize = nvme_ns_get_stripesize(ns);
421	disk->d_flags = DISKFLAG_UNMAPPED_BIO | DISKFLAG_DIRECT_COMPLETION;
422	if (nvme_ns_get_flags(ns) & NVME_NS_DEALLOCATE_SUPPORTED)
423		disk->d_flags |= DISKFLAG_CANDELETE;
424	if (nvme_ns_get_flags(ns) & NVME_NS_FLUSH_SUPPORTED)
425		disk->d_flags |= DISKFLAG_CANFLUSHCACHE;
426
427	/*
428	 * d_ident and d_descr are both far bigger than the length of either
429	 *  the serial or model number strings.
430	 */
431	nvme_strvis(disk->d_ident, nvme_ns_get_serial_number(ns),
432	    sizeof(disk->d_ident), NVME_SERIAL_NUMBER_LENGTH);
433	nvme_strvis(descr, nvme_ns_get_model_number(ns), sizeof(descr),
434	    NVME_MODEL_NUMBER_LENGTH);
435	strlcpy(disk->d_descr, descr, sizeof(descr));
436
437	disk->d_rotation_rate = DISK_RR_NON_ROTATING;
438
439	disk_create(disk, DISK_VERSION);
440
441	printf(NVD_STR"%u: <%s> NVMe namespace\n", disk->d_unit, descr);
442	printf(NVD_STR"%u: %juMB (%ju %u byte sectors)\n", disk->d_unit,
443		(uintmax_t)disk->d_mediasize / (1024*1024),
444		(uintmax_t)disk->d_mediasize / disk->d_sectorsize,
445		disk->d_sectorsize);
446
447	return (ndisk);
448}
449
450static void
451nvd_controller_fail(void *ctrlr_arg)
452{
453	struct nvd_controller	*ctrlr = ctrlr_arg;
454	struct nvd_disk		*ndisk;
455
456	mtx_lock(&nvd_lock);
457	TAILQ_REMOVE(&ctrlr_head, ctrlr, tailq);
458	TAILQ_FOREACH(ndisk, &ctrlr->disk_head, ctrlr_tailq)
459		nvd_gone(ndisk);
460	while (!TAILQ_EMPTY(&ctrlr->disk_head))
461		msleep(&ctrlr->disk_head, &nvd_lock, 0, "nvd_fail", 0);
462	mtx_unlock(&nvd_lock);
463	free(ctrlr, M_NVD);
464}
465
466