2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (C) 2012-2013 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>
37 #include <sys/fcntl.h>
38 #include <sys/ioccom.h>
39 #include <sys/malloc.h>
40 #include <sys/module.h>
42 #include <sys/systm.h>
44 #include <dev/pci/pcivar.h>
46 #include <geom/geom.h>
48 #include "nvme_private.h"
50 static void nvme_bio_child_inbed(struct bio *parent, int bio_error);
51 static void nvme_bio_child_done(void *arg,
52 const struct nvme_completion *cpl);
53 static uint32_t nvme_get_num_segments(uint64_t addr, uint64_t size,
55 static void nvme_free_child_bios(int num_bios,
56 struct bio **child_bios);
57 static struct bio ** nvme_allocate_child_bios(int num_bios);
58 static struct bio ** nvme_construct_child_bios(struct bio *bp,
61 static int nvme_ns_split_bio(struct nvme_namespace *ns,
66 nvme_ns_ioctl(struct cdev *cdev, u_long cmd, caddr_t arg, int flag,
69 struct nvme_namespace *ns;
70 struct nvme_controller *ctrlr;
71 struct nvme_pt_command *pt;
79 nvme_ns_test(ns, cmd, arg);
81 case NVME_PASSTHROUGH_CMD:
82 pt = (struct nvme_pt_command *)arg;
83 return (nvme_ctrlr_passthrough_cmd(ctrlr, pt, ns->id,
84 1 /* is_user_buffer */, 0 /* is_admin_cmd */));
86 *(off_t *)arg = (off_t)nvme_ns_get_size(ns);
89 *(u_int *)arg = nvme_ns_get_sector_size(ns);
99 nvme_ns_open(struct cdev *dev __unused, int flags, int fmt __unused,
105 error = securelevel_gt(td->td_ucred, 0);
111 nvme_ns_close(struct cdev *dev __unused, int flags, int fmt __unused,
119 nvme_ns_strategy_done(void *arg, const struct nvme_completion *cpl)
121 struct bio *bp = arg;
124 * TODO: add more extensive translation of NVMe status codes
125 * to different bio error codes (i.e. EIO, EINVAL, etc.)
127 if (nvme_completion_is_error(cpl)) {
129 bp->bio_flags |= BIO_ERROR;
130 bp->bio_resid = bp->bio_bcount;
138 nvme_ns_strategy(struct bio *bp)
140 struct nvme_namespace *ns;
143 ns = bp->bio_dev->si_drv1;
144 err = nvme_ns_bio_process(ns, bp, nvme_ns_strategy_done);
148 bp->bio_flags |= BIO_ERROR;
149 bp->bio_resid = bp->bio_bcount;
155 static struct cdevsw nvme_ns_cdevsw = {
156 .d_version = D_VERSION,
159 .d_write = physwrite,
160 .d_open = nvme_ns_open,
161 .d_close = nvme_ns_close,
162 .d_strategy = nvme_ns_strategy,
163 .d_ioctl = nvme_ns_ioctl
167 nvme_ns_get_max_io_xfer_size(struct nvme_namespace *ns)
169 return ns->ctrlr->max_xfer_size;
173 nvme_ns_get_sector_size(struct nvme_namespace *ns)
175 uint8_t flbas_fmt, lbads;
177 flbas_fmt = (ns->data.flbas >> NVME_NS_DATA_FLBAS_FORMAT_SHIFT) &
178 NVME_NS_DATA_FLBAS_FORMAT_MASK;
179 lbads = (ns->data.lbaf[flbas_fmt] >> NVME_NS_DATA_LBAF_LBADS_SHIFT) &
180 NVME_NS_DATA_LBAF_LBADS_MASK;
186 nvme_ns_get_num_sectors(struct nvme_namespace *ns)
188 return (ns->data.nsze);
192 nvme_ns_get_size(struct nvme_namespace *ns)
194 return (nvme_ns_get_num_sectors(ns) * nvme_ns_get_sector_size(ns));
198 nvme_ns_get_flags(struct nvme_namespace *ns)
204 nvme_ns_get_serial_number(struct nvme_namespace *ns)
206 return ((const char *)ns->ctrlr->cdata.sn);
210 nvme_ns_get_model_number(struct nvme_namespace *ns)
212 return ((const char *)ns->ctrlr->cdata.mn);
215 const struct nvme_namespace_data *
216 nvme_ns_get_data(struct nvme_namespace *ns)
223 nvme_ns_get_stripesize(struct nvme_namespace *ns)
226 return (ns->stripesize);
230 nvme_ns_bio_done(void *arg, const struct nvme_completion *status)
232 struct bio *bp = arg;
233 nvme_cb_fn_t bp_cb_fn;
235 bp_cb_fn = bp->bio_driver1;
238 free(bp->bio_driver2, M_NVME);
240 if (nvme_completion_is_error(status)) {
241 bp->bio_flags |= BIO_ERROR;
242 if (bp->bio_error == 0)
246 if ((bp->bio_flags & BIO_ERROR) == 0)
249 bp->bio_resid = bp->bio_bcount;
251 bp_cb_fn(bp, status);
255 nvme_bio_child_inbed(struct bio *parent, int bio_error)
257 struct nvme_completion parent_cpl;
260 if (bio_error != 0) {
261 parent->bio_flags |= BIO_ERROR;
262 parent->bio_error = bio_error;
266 * atomic_fetchadd will return value before adding 1, so we still
267 * must add 1 to get the updated inbed number. Save bio_children
268 * before incrementing to guard against race conditions when
269 * two children bios complete on different queues.
271 children = atomic_load_acq_int(&parent->bio_children);
272 inbed = atomic_fetchadd_int(&parent->bio_inbed, 1) + 1;
273 if (inbed == children) {
274 bzero(&parent_cpl, sizeof(parent_cpl));
275 if (parent->bio_flags & BIO_ERROR) {
276 parent_cpl.status &= ~(NVME_STATUS_SC_MASK << NVME_STATUS_SC_SHIFT);
277 parent_cpl.status |= (NVME_SC_DATA_TRANSFER_ERROR) << NVME_STATUS_SC_SHIFT;
279 nvme_ns_bio_done(parent, &parent_cpl);
284 nvme_bio_child_done(void *arg, const struct nvme_completion *cpl)
286 struct bio *child = arg;
290 parent = child->bio_parent;
291 g_destroy_bio(child);
292 bio_error = nvme_completion_is_error(cpl) ? EIO : 0;
293 nvme_bio_child_inbed(parent, bio_error);
297 nvme_get_num_segments(uint64_t addr, uint64_t size, uint32_t align)
299 uint32_t num_segs, offset, remainder;
304 KASSERT((align & (align - 1)) == 0, ("alignment not power of 2\n"));
306 num_segs = size / align;
307 remainder = size & (align - 1);
308 offset = addr & (align - 1);
309 if (remainder > 0 || offset > 0)
310 num_segs += 1 + (remainder + offset - 1) / align;
315 nvme_free_child_bios(int num_bios, struct bio **child_bios)
319 for (i = 0; i < num_bios; i++) {
320 if (child_bios[i] != NULL)
321 g_destroy_bio(child_bios[i]);
324 free(child_bios, M_NVME);
328 nvme_allocate_child_bios(int num_bios)
330 struct bio **child_bios;
333 child_bios = malloc(num_bios * sizeof(struct bio *), M_NVME, M_NOWAIT);
334 if (child_bios == NULL)
337 for (i = 0; i < num_bios; i++) {
338 child_bios[i] = g_new_bio();
339 if (child_bios[i] == NULL)
344 nvme_free_child_bios(num_bios, child_bios);
352 nvme_construct_child_bios(struct bio *bp, uint32_t alignment, int *num_bios)
354 struct bio **child_bios;
360 #ifdef NVME_UNMAPPED_BIO_SUPPORT
365 *num_bios = nvme_get_num_segments(bp->bio_offset, bp->bio_bcount,
367 child_bios = nvme_allocate_child_bios(*num_bios);
368 if (child_bios == NULL)
371 bp->bio_children = *num_bios;
373 cur_offset = bp->bio_offset;
374 rem_bcount = bp->bio_bcount;
376 #ifdef NVME_UNMAPPED_BIO_SUPPORT
377 ma_offset = bp->bio_ma_offset;
381 for (i = 0; i < *num_bios; i++) {
382 child = child_bios[i];
383 child->bio_parent = bp;
384 child->bio_cmd = bp->bio_cmd;
385 child->bio_offset = cur_offset;
386 child->bio_bcount = min(rem_bcount,
387 alignment - (cur_offset & (alignment - 1)));
388 child->bio_flags = bp->bio_flags;
389 #ifdef NVME_UNMAPPED_BIO_SUPPORT
390 if (bp->bio_flags & BIO_UNMAPPED) {
391 child->bio_ma_offset = ma_offset;
394 nvme_get_num_segments(child->bio_ma_offset,
395 child->bio_bcount, PAGE_SIZE);
396 ma_offset = (ma_offset + child->bio_bcount) &
398 ma += child->bio_ma_n;
404 child->bio_data = data;
405 data += child->bio_bcount;
407 cur_offset += child->bio_bcount;
408 rem_bcount -= child->bio_bcount;
415 nvme_ns_split_bio(struct nvme_namespace *ns, struct bio *bp,
419 struct bio **child_bios;
420 int err, i, num_bios;
422 child_bios = nvme_construct_child_bios(bp, alignment, &num_bios);
423 if (child_bios == NULL)
426 for (i = 0; i < num_bios; i++) {
427 child = child_bios[i];
428 err = nvme_ns_bio_process(ns, child, nvme_bio_child_done);
430 nvme_bio_child_inbed(bp, err);
431 g_destroy_bio(child);
435 free(child_bios, M_NVME);
440 nvme_ns_bio_process(struct nvme_namespace *ns, struct bio *bp,
443 struct nvme_dsm_range *dsm_range;
447 bp->bio_driver1 = cb_fn;
449 if (ns->stripesize > 0 &&
450 (bp->bio_cmd == BIO_READ || bp->bio_cmd == BIO_WRITE)) {
451 num_bios = nvme_get_num_segments(bp->bio_offset,
452 bp->bio_bcount, ns->stripesize);
454 return (nvme_ns_split_bio(ns, bp, ns->stripesize));
457 switch (bp->bio_cmd) {
459 err = nvme_ns_cmd_read_bio(ns, bp, nvme_ns_bio_done, bp);
462 err = nvme_ns_cmd_write_bio(ns, bp, nvme_ns_bio_done, bp);
465 err = nvme_ns_cmd_flush(ns, nvme_ns_bio_done, bp);
469 malloc(sizeof(struct nvme_dsm_range), M_NVME,
476 htole32(bp->bio_bcount/nvme_ns_get_sector_size(ns));
477 dsm_range->starting_lba =
478 htole64(bp->bio_offset/nvme_ns_get_sector_size(ns));
479 bp->bio_driver2 = dsm_range;
480 err = nvme_ns_cmd_deallocate(ns, dsm_range, 1,
481 nvme_ns_bio_done, bp);
483 free(dsm_range, M_NVME);
494 nvme_ns_construct(struct nvme_namespace *ns, uint32_t id,
495 struct nvme_controller *ctrlr)
497 struct nvme_completion_poll_status status;
509 * Older Intel devices advertise in vendor specific space an alignment
510 * that improves performance. If present use for the stripe size. NVMe
511 * 1.3 standardized this as NOIOB, and newer Intel drives use that.
513 switch (pci_get_devid(ctrlr->dev)) {
514 case 0x09538086: /* Intel DC PC3500 */
515 case 0x0a538086: /* Intel DC PC3520 */
516 case 0x0a548086: /* Intel DC PC4500 */
517 if (ctrlr->cdata.vs[3] != 0)
519 (1 << ctrlr->cdata.vs[3]) * ctrlr->min_page_size;
526 * Namespaces are reconstructed after a controller reset, so check
527 * to make sure we only call mtx_init once on each mtx.
529 * TODO: Move this somewhere where it gets called at controller
530 * construction time, which is not invoked as part of each
533 if (!mtx_initialized(&ns->lock))
534 mtx_init(&ns->lock, "nvme ns lock", NULL, MTX_DEF);
537 nvme_ctrlr_cmd_identify_namespace(ctrlr, id, &ns->data,
538 nvme_completion_poll_cb, &status);
539 while (status.done == FALSE)
541 if (nvme_completion_is_error(&status.cpl)) {
542 nvme_printf(ctrlr, "nvme_identify_namespace failed\n");
546 /* Convert data to host endian */
547 nvme_namespace_data_swapbytes(&ns->data);
550 * If the size of is zero, chances are this isn't a valid
551 * namespace (eg one that's not been configured yet). The
552 * standard says the entire id will be zeros, so this is a
553 * cheap way to test for that.
555 if (ns->data.nsze == 0)
558 flbas_fmt = (ns->data.flbas >> NVME_NS_DATA_FLBAS_FORMAT_SHIFT) &
559 NVME_NS_DATA_FLBAS_FORMAT_MASK;
561 * Note: format is a 0-based value, so > is appropriate here,
564 if (flbas_fmt > ns->data.nlbaf) {
565 printf("lba format %d exceeds number supported (%d)\n",
566 flbas_fmt, ns->data.nlbaf + 1);
570 oncs = ctrlr->cdata.oncs;
571 dsm = (oncs >> NVME_CTRLR_DATA_ONCS_DSM_SHIFT) & NVME_CTRLR_DATA_ONCS_DSM_MASK;
573 ns->flags |= NVME_NS_DEALLOCATE_SUPPORTED;
575 vwc_present = (ctrlr->cdata.vwc >> NVME_CTRLR_DATA_VWC_PRESENT_SHIFT) &
576 NVME_CTRLR_DATA_VWC_PRESENT_MASK;
578 ns->flags |= NVME_NS_FLUSH_SUPPORTED;
581 * cdev may have already been created, if we are reconstructing the
582 * namespace after a controller-level reset.
584 if (ns->cdev != NULL)
588 * Namespace IDs start at 1, so we need to subtract 1 to create a
589 * correct unit number.
591 unit = device_get_unit(ctrlr->dev) * NVME_MAX_NAMESPACES + ns->id - 1;
593 ns->cdev = make_dev_credf(0, &nvme_ns_cdevsw, unit,
594 NULL, UID_ROOT, GID_WHEEL, 0600, "nvme%dns%d",
595 device_get_unit(ctrlr->dev), ns->id);
596 #ifdef NVME_UNMAPPED_BIO_SUPPORT
597 ns->cdev->si_flags |= SI_UNMAPPED;
600 if (ns->cdev != NULL)
601 ns->cdev->si_drv1 = ns;
606 void nvme_ns_destruct(struct nvme_namespace *ns)
609 if (ns->cdev != NULL)
610 destroy_dev(ns->cdev);