2 * NVM Express device driver
3 * Copyright (c) 2011-2014, Intel Corporation.
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 #include <linux/blkdev.h>
16 #include <linux/blk-mq.h>
17 #include <linux/delay.h>
18 #include <linux/errno.h>
19 #include <linux/hdreg.h>
20 #include <linux/kernel.h>
21 #include <linux/module.h>
22 #include <linux/list_sort.h>
23 #include <linux/slab.h>
24 #include <linux/types.h>
26 #include <linux/ptrace.h>
27 #include <linux/nvme_ioctl.h>
28 #include <linux/t10-pi.h>
30 #include <asm/unaligned.h>
34 #define NVME_MINORS (1U << MINORBITS)
36 unsigned char admin_timeout = 60;
37 module_param(admin_timeout, byte, 0644);
38 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
39 EXPORT_SYMBOL_GPL(admin_timeout);
41 unsigned char nvme_io_timeout = 30;
42 module_param_named(io_timeout, nvme_io_timeout, byte, 0644);
43 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
44 EXPORT_SYMBOL_GPL(nvme_io_timeout);
46 unsigned char shutdown_timeout = 5;
47 module_param(shutdown_timeout, byte, 0644);
48 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
50 static int nvme_major;
51 module_param(nvme_major, int, 0);
53 static int nvme_char_major;
54 module_param(nvme_char_major, int, 0);
56 static LIST_HEAD(nvme_ctrl_list);
57 static DEFINE_SPINLOCK(dev_list_lock);
59 static struct class *nvme_class;
61 void nvme_cancel_request(struct request *req, void *data, bool reserved)
65 if (!blk_mq_request_started(req))
68 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
69 "Cancelling I/O %d", req->tag);
71 status = NVME_SC_ABORT_REQ;
72 if (blk_queue_dying(req->q))
73 status |= NVME_SC_DNR;
74 blk_mq_complete_request(req, status);
76 EXPORT_SYMBOL_GPL(nvme_cancel_request);
78 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
79 enum nvme_ctrl_state new_state)
81 enum nvme_ctrl_state old_state = ctrl->state;
84 spin_lock_irq(&ctrl->lock);
89 case NVME_CTRL_RESETTING:
96 case NVME_CTRL_RESETTING:
106 case NVME_CTRL_DELETING:
109 case NVME_CTRL_RESETTING:
118 case NVME_CTRL_DELETING:
128 spin_unlock_irq(&ctrl->lock);
131 ctrl->state = new_state;
135 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
137 static void nvme_free_ns(struct kref *kref)
139 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
141 if (ns->type == NVME_NS_LIGHTNVM)
142 nvme_nvm_unregister(ns->queue, ns->disk->disk_name);
144 spin_lock(&dev_list_lock);
145 ns->disk->private_data = NULL;
146 spin_unlock(&dev_list_lock);
149 ida_simple_remove(&ns->ctrl->ns_ida, ns->instance);
150 nvme_put_ctrl(ns->ctrl);
154 static void nvme_put_ns(struct nvme_ns *ns)
156 kref_put(&ns->kref, nvme_free_ns);
159 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk)
163 spin_lock(&dev_list_lock);
164 ns = disk->private_data;
166 if (!kref_get_unless_zero(&ns->kref))
168 if (!try_module_get(ns->ctrl->ops->module))
171 spin_unlock(&dev_list_lock);
176 kref_put(&ns->kref, nvme_free_ns);
178 spin_unlock(&dev_list_lock);
182 void nvme_requeue_req(struct request *req)
186 blk_mq_requeue_request(req);
187 spin_lock_irqsave(req->q->queue_lock, flags);
188 if (!blk_queue_stopped(req->q))
189 blk_mq_kick_requeue_list(req->q);
190 spin_unlock_irqrestore(req->q->queue_lock, flags);
192 EXPORT_SYMBOL_GPL(nvme_requeue_req);
194 struct request *nvme_alloc_request(struct request_queue *q,
195 struct nvme_command *cmd, unsigned int flags, int qid)
199 if (qid == NVME_QID_ANY) {
200 req = blk_mq_alloc_request(q, nvme_is_write(cmd), flags);
202 req = blk_mq_alloc_request_hctx(q, nvme_is_write(cmd), flags,
208 req->cmd_type = REQ_TYPE_DRV_PRIV;
209 req->cmd_flags |= REQ_FAILFAST_DRIVER;
211 req->__sector = (sector_t) -1;
212 req->bio = req->biotail = NULL;
214 req->cmd = (unsigned char *)cmd;
215 req->cmd_len = sizeof(struct nvme_command);
219 EXPORT_SYMBOL_GPL(nvme_alloc_request);
221 static inline void nvme_setup_flush(struct nvme_ns *ns,
222 struct nvme_command *cmnd)
224 memset(cmnd, 0, sizeof(*cmnd));
225 cmnd->common.opcode = nvme_cmd_flush;
226 cmnd->common.nsid = cpu_to_le32(ns->ns_id);
229 static inline int nvme_setup_discard(struct nvme_ns *ns, struct request *req,
230 struct nvme_command *cmnd)
232 struct nvme_dsm_range *range;
235 unsigned int nr_bytes = blk_rq_bytes(req);
237 range = kmalloc(sizeof(*range), GFP_ATOMIC);
239 return BLK_MQ_RQ_QUEUE_BUSY;
241 range->cattr = cpu_to_le32(0);
242 range->nlb = cpu_to_le32(nr_bytes >> ns->lba_shift);
243 range->slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
245 memset(cmnd, 0, sizeof(*cmnd));
246 cmnd->dsm.opcode = nvme_cmd_dsm;
247 cmnd->dsm.nsid = cpu_to_le32(ns->ns_id);
249 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
251 req->completion_data = range;
252 page = virt_to_page(range);
253 offset = offset_in_page(range);
254 blk_add_request_payload(req, page, offset, sizeof(*range));
257 * we set __data_len back to the size of the area to be discarded
258 * on disk. This allows us to report completion on the full amount
259 * of blocks described by the request.
261 req->__data_len = nr_bytes;
266 static inline void nvme_setup_rw(struct nvme_ns *ns, struct request *req,
267 struct nvme_command *cmnd)
272 if (req->cmd_flags & REQ_FUA)
273 control |= NVME_RW_FUA;
274 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
275 control |= NVME_RW_LR;
277 if (req->cmd_flags & REQ_RAHEAD)
278 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
280 memset(cmnd, 0, sizeof(*cmnd));
281 cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
282 cmnd->rw.command_id = req->tag;
283 cmnd->rw.nsid = cpu_to_le32(ns->ns_id);
284 cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
285 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
288 switch (ns->pi_type) {
289 case NVME_NS_DPS_PI_TYPE3:
290 control |= NVME_RW_PRINFO_PRCHK_GUARD;
292 case NVME_NS_DPS_PI_TYPE1:
293 case NVME_NS_DPS_PI_TYPE2:
294 control |= NVME_RW_PRINFO_PRCHK_GUARD |
295 NVME_RW_PRINFO_PRCHK_REF;
296 cmnd->rw.reftag = cpu_to_le32(
297 nvme_block_nr(ns, blk_rq_pos(req)));
300 if (!blk_integrity_rq(req))
301 control |= NVME_RW_PRINFO_PRACT;
304 cmnd->rw.control = cpu_to_le16(control);
305 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
308 int nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
309 struct nvme_command *cmd)
313 if (req->cmd_type == REQ_TYPE_DRV_PRIV)
314 memcpy(cmd, req->cmd, sizeof(*cmd));
315 else if (req_op(req) == REQ_OP_FLUSH)
316 nvme_setup_flush(ns, cmd);
317 else if (req_op(req) == REQ_OP_DISCARD)
318 ret = nvme_setup_discard(ns, req, cmd);
320 nvme_setup_rw(ns, req, cmd);
324 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
327 * Returns 0 on success. If the result is negative, it's a Linux error code;
328 * if the result is positive, it's an NVM Express status code
330 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
331 struct nvme_completion *cqe, void *buffer, unsigned bufflen,
332 unsigned timeout, int qid, int at_head, int flags)
337 req = nvme_alloc_request(q, cmd, flags, qid);
341 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
344 if (buffer && bufflen) {
345 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
350 blk_execute_rq(req->q, NULL, req, at_head);
353 blk_mq_free_request(req);
356 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
358 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
359 void *buffer, unsigned bufflen)
361 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
364 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
366 int __nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
367 void __user *ubuffer, unsigned bufflen,
368 void __user *meta_buffer, unsigned meta_len, u32 meta_seed,
369 u32 *result, unsigned timeout)
371 bool write = nvme_is_write(cmd);
372 struct nvme_completion cqe;
373 struct nvme_ns *ns = q->queuedata;
374 struct gendisk *disk = ns ? ns->disk : NULL;
376 struct bio *bio = NULL;
380 req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
384 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
387 if (ubuffer && bufflen) {
388 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
396 bio->bi_bdev = bdget_disk(disk, 0);
402 if (meta_buffer && meta_len) {
403 struct bio_integrity_payload *bip;
405 meta = kmalloc(meta_len, GFP_KERNEL);
412 if (copy_from_user(meta, meta_buffer,
419 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
425 bip->bip_iter.bi_size = meta_len;
426 bip->bip_iter.bi_sector = meta_seed;
428 ret = bio_integrity_add_page(bio, virt_to_page(meta),
429 meta_len, offset_in_page(meta));
430 if (ret != meta_len) {
437 blk_execute_rq(req->q, disk, req, 0);
440 *result = le32_to_cpu(cqe.result);
441 if (meta && !ret && !write) {
442 if (copy_to_user(meta_buffer, meta, meta_len))
449 if (disk && bio->bi_bdev)
451 blk_rq_unmap_user(bio);
454 blk_mq_free_request(req);
458 int nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
459 void __user *ubuffer, unsigned bufflen, u32 *result,
462 return __nvme_submit_user_cmd(q, cmd, ubuffer, bufflen, NULL, 0, 0,
466 int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
468 struct nvme_command c = { };
471 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
472 c.identify.opcode = nvme_admin_identify;
473 c.identify.cns = cpu_to_le32(1);
475 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
479 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
480 sizeof(struct nvme_id_ctrl));
486 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
488 struct nvme_command c = { };
490 c.identify.opcode = nvme_admin_identify;
491 c.identify.cns = cpu_to_le32(2);
492 c.identify.nsid = cpu_to_le32(nsid);
493 return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000);
496 int nvme_identify_ns(struct nvme_ctrl *dev, unsigned nsid,
497 struct nvme_id_ns **id)
499 struct nvme_command c = { };
502 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
503 c.identify.opcode = nvme_admin_identify,
504 c.identify.nsid = cpu_to_le32(nsid),
506 *id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL);
510 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
511 sizeof(struct nvme_id_ns));
517 int nvme_get_features(struct nvme_ctrl *dev, unsigned fid, unsigned nsid,
518 dma_addr_t dma_addr, u32 *result)
520 struct nvme_command c;
521 struct nvme_completion cqe;
524 memset(&c, 0, sizeof(c));
525 c.features.opcode = nvme_admin_get_features;
526 c.features.nsid = cpu_to_le32(nsid);
527 c.features.dptr.prp1 = cpu_to_le64(dma_addr);
528 c.features.fid = cpu_to_le32(fid);
530 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &cqe, NULL, 0, 0,
533 *result = le32_to_cpu(cqe.result);
537 int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
538 dma_addr_t dma_addr, u32 *result)
540 struct nvme_command c;
541 struct nvme_completion cqe;
544 memset(&c, 0, sizeof(c));
545 c.features.opcode = nvme_admin_set_features;
546 c.features.dptr.prp1 = cpu_to_le64(dma_addr);
547 c.features.fid = cpu_to_le32(fid);
548 c.features.dword11 = cpu_to_le32(dword11);
550 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &cqe, NULL, 0, 0,
553 *result = le32_to_cpu(cqe.result);
557 int nvme_get_log_page(struct nvme_ctrl *dev, struct nvme_smart_log **log)
559 struct nvme_command c = { };
562 c.common.opcode = nvme_admin_get_log_page,
563 c.common.nsid = cpu_to_le32(0xFFFFFFFF),
564 c.common.cdw10[0] = cpu_to_le32(
565 (((sizeof(struct nvme_smart_log) / 4) - 1) << 16) |
568 *log = kmalloc(sizeof(struct nvme_smart_log), GFP_KERNEL);
572 error = nvme_submit_sync_cmd(dev->admin_q, &c, *log,
573 sizeof(struct nvme_smart_log));
579 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
581 u32 q_count = (*count - 1) | ((*count - 1) << 16);
583 int status, nr_io_queues;
585 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, 0,
591 * Degraded controllers might return an error when setting the queue
592 * count. We still want to be able to bring them online and offer
593 * access to the admin queue, as that might be only way to fix them up.
596 dev_err(ctrl->dev, "Could not set queue count (%d)\n", status);
599 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
600 *count = min(*count, nr_io_queues);
605 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
607 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
609 struct nvme_user_io io;
610 struct nvme_command c;
611 unsigned length, meta_len;
612 void __user *metadata;
614 if (copy_from_user(&io, uio, sizeof(io)))
622 case nvme_cmd_compare:
628 length = (io.nblocks + 1) << ns->lba_shift;
629 meta_len = (io.nblocks + 1) * ns->ms;
630 metadata = (void __user *)(uintptr_t)io.metadata;
635 } else if (meta_len) {
636 if ((io.metadata & 3) || !io.metadata)
640 memset(&c, 0, sizeof(c));
641 c.rw.opcode = io.opcode;
642 c.rw.flags = io.flags;
643 c.rw.nsid = cpu_to_le32(ns->ns_id);
644 c.rw.slba = cpu_to_le64(io.slba);
645 c.rw.length = cpu_to_le16(io.nblocks);
646 c.rw.control = cpu_to_le16(io.control);
647 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
648 c.rw.reftag = cpu_to_le32(io.reftag);
649 c.rw.apptag = cpu_to_le16(io.apptag);
650 c.rw.appmask = cpu_to_le16(io.appmask);
652 return __nvme_submit_user_cmd(ns->queue, &c,
653 (void __user *)(uintptr_t)io.addr, length,
654 metadata, meta_len, io.slba, NULL, 0);
657 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
658 struct nvme_passthru_cmd __user *ucmd)
660 struct nvme_passthru_cmd cmd;
661 struct nvme_command c;
662 unsigned timeout = 0;
665 if (!capable(CAP_SYS_ADMIN))
667 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
672 memset(&c, 0, sizeof(c));
673 c.common.opcode = cmd.opcode;
674 c.common.flags = cmd.flags;
675 c.common.nsid = cpu_to_le32(cmd.nsid);
676 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
677 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
678 c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
679 c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
680 c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
681 c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
682 c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
683 c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
686 timeout = msecs_to_jiffies(cmd.timeout_ms);
688 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
689 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
690 &cmd.result, timeout);
692 if (put_user(cmd.result, &ucmd->result))
699 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
700 unsigned int cmd, unsigned long arg)
702 struct nvme_ns *ns = bdev->bd_disk->private_data;
706 force_successful_syscall_return();
708 case NVME_IOCTL_ADMIN_CMD:
709 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
710 case NVME_IOCTL_IO_CMD:
711 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
712 case NVME_IOCTL_SUBMIT_IO:
713 return nvme_submit_io(ns, (void __user *)arg);
714 #ifdef CONFIG_BLK_DEV_NVME_SCSI
715 case SG_GET_VERSION_NUM:
716 return nvme_sg_get_version_num((void __user *)arg);
718 return nvme_sg_io(ns, (void __user *)arg);
726 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
727 unsigned int cmd, unsigned long arg)
733 return nvme_ioctl(bdev, mode, cmd, arg);
736 #define nvme_compat_ioctl NULL
739 static int nvme_open(struct block_device *bdev, fmode_t mode)
741 return nvme_get_ns_from_disk(bdev->bd_disk) ? 0 : -ENXIO;
744 static void nvme_release(struct gendisk *disk, fmode_t mode)
746 struct nvme_ns *ns = disk->private_data;
748 module_put(ns->ctrl->ops->module);
752 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
754 /* some standard values */
756 geo->sectors = 1 << 5;
757 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
761 #ifdef CONFIG_BLK_DEV_INTEGRITY
762 static void nvme_init_integrity(struct nvme_ns *ns)
764 struct blk_integrity integrity;
766 switch (ns->pi_type) {
767 case NVME_NS_DPS_PI_TYPE3:
768 integrity.profile = &t10_pi_type3_crc;
769 integrity.tag_size = sizeof(u16) + sizeof(u32);
770 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
772 case NVME_NS_DPS_PI_TYPE1:
773 case NVME_NS_DPS_PI_TYPE2:
774 integrity.profile = &t10_pi_type1_crc;
775 integrity.tag_size = sizeof(u16);
776 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
779 integrity.profile = NULL;
782 integrity.tuple_size = ns->ms;
783 blk_integrity_register(ns->disk, &integrity);
784 blk_queue_max_integrity_segments(ns->queue, 1);
787 static void nvme_init_integrity(struct nvme_ns *ns)
790 #endif /* CONFIG_BLK_DEV_INTEGRITY */
792 static void nvme_config_discard(struct nvme_ns *ns)
794 struct nvme_ctrl *ctrl = ns->ctrl;
795 u32 logical_block_size = queue_logical_block_size(ns->queue);
797 if (ctrl->quirks & NVME_QUIRK_DISCARD_ZEROES)
798 ns->queue->limits.discard_zeroes_data = 1;
800 ns->queue->limits.discard_zeroes_data = 0;
802 ns->queue->limits.discard_alignment = logical_block_size;
803 ns->queue->limits.discard_granularity = logical_block_size;
804 blk_queue_max_discard_sectors(ns->queue, UINT_MAX);
805 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue);
808 static int nvme_revalidate_disk(struct gendisk *disk)
810 struct nvme_ns *ns = disk->private_data;
811 struct nvme_id_ns *id;
816 if (test_bit(NVME_NS_DEAD, &ns->flags)) {
817 set_capacity(disk, 0);
820 if (nvme_identify_ns(ns->ctrl, ns->ns_id, &id)) {
821 dev_warn(disk_to_dev(ns->disk), "%s: Identify failure\n",
830 if (nvme_nvm_ns_supported(ns, id) && ns->type != NVME_NS_LIGHTNVM) {
831 if (nvme_nvm_register(ns->queue, disk->disk_name)) {
832 dev_warn(disk_to_dev(ns->disk),
833 "%s: LightNVM init failure\n", __func__);
837 ns->type = NVME_NS_LIGHTNVM;
840 if (ns->ctrl->vs >= NVME_VS(1, 1))
841 memcpy(ns->eui, id->eui64, sizeof(ns->eui));
842 if (ns->ctrl->vs >= NVME_VS(1, 2))
843 memcpy(ns->uuid, id->nguid, sizeof(ns->uuid));
846 lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
847 ns->lba_shift = id->lbaf[lbaf].ds;
848 ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
849 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
852 * If identify namespace failed, use default 512 byte block size so
853 * block layer can use before failing read/write for 0 capacity.
855 if (ns->lba_shift == 0)
857 bs = 1 << ns->lba_shift;
858 /* XXX: PI implementation requires metadata equal t10 pi tuple size */
859 pi_type = ns->ms == sizeof(struct t10_pi_tuple) ?
860 id->dps & NVME_NS_DPS_PI_MASK : 0;
862 blk_mq_freeze_queue(disk->queue);
863 if (blk_get_integrity(disk) && (ns->pi_type != pi_type ||
865 bs != queue_logical_block_size(disk->queue) ||
866 (ns->ms && ns->ext)))
867 blk_integrity_unregister(disk);
869 ns->pi_type = pi_type;
870 blk_queue_logical_block_size(ns->queue, bs);
872 if (ns->ms && !blk_get_integrity(disk) && !ns->ext)
873 nvme_init_integrity(ns);
874 if (ns->ms && !(ns->ms == 8 && ns->pi_type) && !blk_get_integrity(disk))
875 set_capacity(disk, 0);
877 set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
879 if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
880 nvme_config_discard(ns);
881 blk_mq_unfreeze_queue(disk->queue);
887 static char nvme_pr_type(enum pr_type type)
890 case PR_WRITE_EXCLUSIVE:
892 case PR_EXCLUSIVE_ACCESS:
894 case PR_WRITE_EXCLUSIVE_REG_ONLY:
896 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
898 case PR_WRITE_EXCLUSIVE_ALL_REGS:
900 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
907 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
908 u64 key, u64 sa_key, u8 op)
910 struct nvme_ns *ns = bdev->bd_disk->private_data;
911 struct nvme_command c;
912 u8 data[16] = { 0, };
914 put_unaligned_le64(key, &data[0]);
915 put_unaligned_le64(sa_key, &data[8]);
917 memset(&c, 0, sizeof(c));
918 c.common.opcode = op;
919 c.common.nsid = cpu_to_le32(ns->ns_id);
920 c.common.cdw10[0] = cpu_to_le32(cdw10);
922 return nvme_submit_sync_cmd(ns->queue, &c, data, 16);
925 static int nvme_pr_register(struct block_device *bdev, u64 old,
926 u64 new, unsigned flags)
930 if (flags & ~PR_FL_IGNORE_KEY)
934 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
935 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
936 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
939 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
940 enum pr_type type, unsigned flags)
944 if (flags & ~PR_FL_IGNORE_KEY)
947 cdw10 = nvme_pr_type(type) << 8;
948 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
949 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
952 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
953 enum pr_type type, bool abort)
955 u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
956 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
959 static int nvme_pr_clear(struct block_device *bdev, u64 key)
961 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
962 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
965 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
967 u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
968 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
971 static const struct pr_ops nvme_pr_ops = {
972 .pr_register = nvme_pr_register,
973 .pr_reserve = nvme_pr_reserve,
974 .pr_release = nvme_pr_release,
975 .pr_preempt = nvme_pr_preempt,
976 .pr_clear = nvme_pr_clear,
979 static const struct block_device_operations nvme_fops = {
980 .owner = THIS_MODULE,
982 .compat_ioctl = nvme_compat_ioctl,
984 .release = nvme_release,
985 .getgeo = nvme_getgeo,
986 .revalidate_disk= nvme_revalidate_disk,
987 .pr_ops = &nvme_pr_ops,
990 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
992 unsigned long timeout =
993 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
994 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
997 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
998 if ((csts & NVME_CSTS_RDY) == bit)
1002 if (fatal_signal_pending(current))
1004 if (time_after(jiffies, timeout)) {
1005 dev_err(ctrl->device,
1006 "Device not ready; aborting %s\n", enabled ?
1007 "initialisation" : "reset");
1016 * If the device has been passed off to us in an enabled state, just clear
1017 * the enabled bit. The spec says we should set the 'shutdown notification
1018 * bits', but doing so may cause the device to complete commands to the
1019 * admin queue ... and we don't know what memory that might be pointing at!
1021 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1025 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1026 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1028 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1031 return nvme_wait_ready(ctrl, cap, false);
1033 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1035 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1038 * Default to a 4K page size, with the intention to update this
1039 * path in the future to accomodate architectures with differing
1040 * kernel and IO page sizes.
1042 unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1045 if (page_shift < dev_page_min) {
1046 dev_err(ctrl->device,
1047 "Minimum device page size %u too large for host (%u)\n",
1048 1 << dev_page_min, 1 << page_shift);
1052 ctrl->page_size = 1 << page_shift;
1054 ctrl->ctrl_config = NVME_CC_CSS_NVM;
1055 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1056 ctrl->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE;
1057 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1058 ctrl->ctrl_config |= NVME_CC_ENABLE;
1060 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1063 return nvme_wait_ready(ctrl, cap, true);
1065 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1067 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1069 unsigned long timeout = SHUTDOWN_TIMEOUT + jiffies;
1073 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1074 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1076 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1080 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1081 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1085 if (fatal_signal_pending(current))
1087 if (time_after(jiffies, timeout)) {
1088 dev_err(ctrl->device,
1089 "Device shutdown incomplete; abort shutdown\n");
1096 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1098 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1099 struct request_queue *q)
1103 if (ctrl->max_hw_sectors) {
1105 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1107 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1108 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1110 if (ctrl->stripe_size)
1111 blk_queue_chunk_sectors(q, ctrl->stripe_size >> 9);
1112 blk_queue_virt_boundary(q, ctrl->page_size - 1);
1113 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1115 blk_queue_write_cache(q, vwc, vwc);
1119 * Initialize the cached copies of the Identify data and various controller
1120 * register in our nvme_ctrl structure. This should be called as soon as
1121 * the admin queue is fully up and running.
1123 int nvme_init_identify(struct nvme_ctrl *ctrl)
1125 struct nvme_id_ctrl *id;
1127 int ret, page_shift;
1130 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
1132 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
1136 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
1138 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
1141 page_shift = NVME_CAP_MPSMIN(cap) + 12;
1143 if (ctrl->vs >= NVME_VS(1, 1))
1144 ctrl->subsystem = NVME_CAP_NSSRC(cap);
1146 ret = nvme_identify_ctrl(ctrl, &id);
1148 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
1152 ctrl->vid = le16_to_cpu(id->vid);
1153 ctrl->oncs = le16_to_cpup(&id->oncs);
1154 atomic_set(&ctrl->abort_limit, id->acl + 1);
1155 ctrl->vwc = id->vwc;
1156 ctrl->cntlid = le16_to_cpup(&id->cntlid);
1157 memcpy(ctrl->serial, id->sn, sizeof(id->sn));
1158 memcpy(ctrl->model, id->mn, sizeof(id->mn));
1159 memcpy(ctrl->firmware_rev, id->fr, sizeof(id->fr));
1161 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
1163 max_hw_sectors = UINT_MAX;
1164 ctrl->max_hw_sectors =
1165 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
1167 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) && id->vs[3]) {
1168 unsigned int max_hw_sectors;
1170 ctrl->stripe_size = 1 << (id->vs[3] + page_shift);
1171 max_hw_sectors = ctrl->stripe_size >> (page_shift - 9);
1172 if (ctrl->max_hw_sectors) {
1173 ctrl->max_hw_sectors = min(max_hw_sectors,
1174 ctrl->max_hw_sectors);
1176 ctrl->max_hw_sectors = max_hw_sectors;
1180 nvme_set_queue_limits(ctrl, ctrl->admin_q);
1181 ctrl->sgls = le32_to_cpu(id->sgls);
1183 if (ctrl->ops->is_fabrics) {
1184 ctrl->icdoff = le16_to_cpu(id->icdoff);
1185 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
1186 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
1187 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
1190 * In fabrics we need to verify the cntlid matches the
1193 if (ctrl->cntlid != le16_to_cpu(id->cntlid))
1196 ctrl->cntlid = le16_to_cpu(id->cntlid);
1202 EXPORT_SYMBOL_GPL(nvme_init_identify);
1204 static int nvme_dev_open(struct inode *inode, struct file *file)
1206 struct nvme_ctrl *ctrl;
1207 int instance = iminor(inode);
1210 spin_lock(&dev_list_lock);
1211 list_for_each_entry(ctrl, &nvme_ctrl_list, node) {
1212 if (ctrl->instance != instance)
1215 if (!ctrl->admin_q) {
1219 if (!kref_get_unless_zero(&ctrl->kref))
1221 file->private_data = ctrl;
1225 spin_unlock(&dev_list_lock);
1230 static int nvme_dev_release(struct inode *inode, struct file *file)
1232 nvme_put_ctrl(file->private_data);
1236 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
1241 mutex_lock(&ctrl->namespaces_mutex);
1242 if (list_empty(&ctrl->namespaces)) {
1247 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
1248 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
1249 dev_warn(ctrl->device,
1250 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
1255 dev_warn(ctrl->device,
1256 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
1257 kref_get(&ns->kref);
1258 mutex_unlock(&ctrl->namespaces_mutex);
1260 ret = nvme_user_cmd(ctrl, ns, argp);
1265 mutex_unlock(&ctrl->namespaces_mutex);
1269 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
1272 struct nvme_ctrl *ctrl = file->private_data;
1273 void __user *argp = (void __user *)arg;
1276 case NVME_IOCTL_ADMIN_CMD:
1277 return nvme_user_cmd(ctrl, NULL, argp);
1278 case NVME_IOCTL_IO_CMD:
1279 return nvme_dev_user_cmd(ctrl, argp);
1280 case NVME_IOCTL_RESET:
1281 dev_warn(ctrl->device, "resetting controller\n");
1282 return ctrl->ops->reset_ctrl(ctrl);
1283 case NVME_IOCTL_SUBSYS_RESET:
1284 return nvme_reset_subsystem(ctrl);
1285 case NVME_IOCTL_RESCAN:
1286 nvme_queue_scan(ctrl);
1293 static const struct file_operations nvme_dev_fops = {
1294 .owner = THIS_MODULE,
1295 .open = nvme_dev_open,
1296 .release = nvme_dev_release,
1297 .unlocked_ioctl = nvme_dev_ioctl,
1298 .compat_ioctl = nvme_dev_ioctl,
1301 static ssize_t nvme_sysfs_reset(struct device *dev,
1302 struct device_attribute *attr, const char *buf,
1305 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1308 ret = ctrl->ops->reset_ctrl(ctrl);
1313 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
1315 static ssize_t nvme_sysfs_rescan(struct device *dev,
1316 struct device_attribute *attr, const char *buf,
1319 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1321 nvme_queue_scan(ctrl);
1324 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
1326 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
1329 struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1330 struct nvme_ctrl *ctrl = ns->ctrl;
1331 int serial_len = sizeof(ctrl->serial);
1332 int model_len = sizeof(ctrl->model);
1334 if (memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1335 return sprintf(buf, "eui.%16phN\n", ns->uuid);
1337 if (memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1338 return sprintf(buf, "eui.%8phN\n", ns->eui);
1340 while (ctrl->serial[serial_len - 1] == ' ')
1342 while (ctrl->model[model_len - 1] == ' ')
1345 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", ctrl->vid,
1346 serial_len, ctrl->serial, model_len, ctrl->model, ns->ns_id);
1348 static DEVICE_ATTR(wwid, S_IRUGO, wwid_show, NULL);
1350 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
1353 struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1354 return sprintf(buf, "%pU\n", ns->uuid);
1356 static DEVICE_ATTR(uuid, S_IRUGO, uuid_show, NULL);
1358 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
1361 struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1362 return sprintf(buf, "%8phd\n", ns->eui);
1364 static DEVICE_ATTR(eui, S_IRUGO, eui_show, NULL);
1366 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
1369 struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1370 return sprintf(buf, "%d\n", ns->ns_id);
1372 static DEVICE_ATTR(nsid, S_IRUGO, nsid_show, NULL);
1374 static struct attribute *nvme_ns_attrs[] = {
1375 &dev_attr_wwid.attr,
1376 &dev_attr_uuid.attr,
1378 &dev_attr_nsid.attr,
1382 static umode_t nvme_ns_attrs_are_visible(struct kobject *kobj,
1383 struct attribute *a, int n)
1385 struct device *dev = container_of(kobj, struct device, kobj);
1386 struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1388 if (a == &dev_attr_uuid.attr) {
1389 if (!memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1392 if (a == &dev_attr_eui.attr) {
1393 if (!memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1399 static const struct attribute_group nvme_ns_attr_group = {
1400 .attrs = nvme_ns_attrs,
1401 .is_visible = nvme_ns_attrs_are_visible,
1404 #define nvme_show_str_function(field) \
1405 static ssize_t field##_show(struct device *dev, \
1406 struct device_attribute *attr, char *buf) \
1408 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
1409 return sprintf(buf, "%.*s\n", (int)sizeof(ctrl->field), ctrl->field); \
1411 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1413 #define nvme_show_int_function(field) \
1414 static ssize_t field##_show(struct device *dev, \
1415 struct device_attribute *attr, char *buf) \
1417 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
1418 return sprintf(buf, "%d\n", ctrl->field); \
1420 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1422 nvme_show_str_function(model);
1423 nvme_show_str_function(serial);
1424 nvme_show_str_function(firmware_rev);
1425 nvme_show_int_function(cntlid);
1427 static ssize_t nvme_sysfs_delete(struct device *dev,
1428 struct device_attribute *attr, const char *buf,
1431 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1433 if (device_remove_file_self(dev, attr))
1434 ctrl->ops->delete_ctrl(ctrl);
1437 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
1439 static ssize_t nvme_sysfs_show_transport(struct device *dev,
1440 struct device_attribute *attr,
1443 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1445 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
1447 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
1449 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
1450 struct device_attribute *attr,
1453 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1455 return snprintf(buf, PAGE_SIZE, "%s\n",
1456 ctrl->ops->get_subsysnqn(ctrl));
1458 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
1460 static ssize_t nvme_sysfs_show_address(struct device *dev,
1461 struct device_attribute *attr,
1464 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1466 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
1468 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
1470 static struct attribute *nvme_dev_attrs[] = {
1471 &dev_attr_reset_controller.attr,
1472 &dev_attr_rescan_controller.attr,
1473 &dev_attr_model.attr,
1474 &dev_attr_serial.attr,
1475 &dev_attr_firmware_rev.attr,
1476 &dev_attr_cntlid.attr,
1477 &dev_attr_delete_controller.attr,
1478 &dev_attr_transport.attr,
1479 &dev_attr_subsysnqn.attr,
1480 &dev_attr_address.attr,
1484 #define CHECK_ATTR(ctrl, a, name) \
1485 if ((a) == &dev_attr_##name.attr && \
1486 !(ctrl)->ops->get_##name) \
1489 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
1490 struct attribute *a, int n)
1492 struct device *dev = container_of(kobj, struct device, kobj);
1493 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1495 if (a == &dev_attr_delete_controller.attr) {
1496 if (!ctrl->ops->delete_ctrl)
1500 CHECK_ATTR(ctrl, a, subsysnqn);
1501 CHECK_ATTR(ctrl, a, address);
1506 static struct attribute_group nvme_dev_attrs_group = {
1507 .attrs = nvme_dev_attrs,
1508 .is_visible = nvme_dev_attrs_are_visible,
1511 static const struct attribute_group *nvme_dev_attr_groups[] = {
1512 &nvme_dev_attrs_group,
1516 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
1518 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
1519 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
1521 return nsa->ns_id - nsb->ns_id;
1524 static struct nvme_ns *nvme_find_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1528 lockdep_assert_held(&ctrl->namespaces_mutex);
1530 list_for_each_entry(ns, &ctrl->namespaces, list) {
1531 if (ns->ns_id == nsid)
1533 if (ns->ns_id > nsid)
1539 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1542 struct gendisk *disk;
1543 int node = dev_to_node(ctrl->dev);
1545 lockdep_assert_held(&ctrl->namespaces_mutex);
1547 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
1551 ns->instance = ida_simple_get(&ctrl->ns_ida, 1, 0, GFP_KERNEL);
1552 if (ns->instance < 0)
1555 ns->queue = blk_mq_init_queue(ctrl->tagset);
1556 if (IS_ERR(ns->queue))
1557 goto out_release_instance;
1558 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
1559 ns->queue->queuedata = ns;
1562 disk = alloc_disk_node(0, node);
1564 goto out_free_queue;
1566 kref_init(&ns->kref);
1569 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
1572 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
1573 nvme_set_queue_limits(ctrl, ns->queue);
1575 disk->major = nvme_major;
1576 disk->first_minor = 0;
1577 disk->fops = &nvme_fops;
1578 disk->private_data = ns;
1579 disk->queue = ns->queue;
1580 disk->driverfs_dev = ctrl->device;
1581 disk->flags = GENHD_FL_EXT_DEVT;
1582 sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance, ns->instance);
1584 if (nvme_revalidate_disk(ns->disk))
1587 list_add_tail_rcu(&ns->list, &ctrl->namespaces);
1588 kref_get(&ctrl->kref);
1589 if (ns->type == NVME_NS_LIGHTNVM)
1593 if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
1594 &nvme_ns_attr_group))
1595 pr_warn("%s: failed to create sysfs group for identification\n",
1596 ns->disk->disk_name);
1601 blk_cleanup_queue(ns->queue);
1602 out_release_instance:
1603 ida_simple_remove(&ctrl->ns_ida, ns->instance);
1608 static void nvme_ns_remove(struct nvme_ns *ns)
1610 lockdep_assert_held(&ns->ctrl->namespaces_mutex);
1612 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
1615 if (ns->disk->flags & GENHD_FL_UP) {
1616 if (blk_get_integrity(ns->disk))
1617 blk_integrity_unregister(ns->disk);
1618 sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
1619 &nvme_ns_attr_group);
1620 del_gendisk(ns->disk);
1621 blk_mq_abort_requeue_list(ns->queue);
1622 blk_cleanup_queue(ns->queue);
1624 list_del_init(&ns->list);
1629 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1633 ns = nvme_find_ns(ctrl, nsid);
1635 if (revalidate_disk(ns->disk))
1638 nvme_alloc_ns(ctrl, nsid);
1641 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
1644 struct nvme_ns *ns, *next;
1646 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
1647 if (ns->ns_id > nsid)
1652 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
1656 unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
1659 ns_list = kzalloc(0x1000, GFP_KERNEL);
1663 for (i = 0; i < num_lists; i++) {
1664 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
1668 for (j = 0; j < min(nn, 1024U); j++) {
1669 nsid = le32_to_cpu(ns_list[j]);
1673 nvme_validate_ns(ctrl, nsid);
1675 while (++prev < nsid) {
1676 ns = nvme_find_ns(ctrl, prev);
1684 nvme_remove_invalid_namespaces(ctrl, prev);
1690 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
1694 lockdep_assert_held(&ctrl->namespaces_mutex);
1696 for (i = 1; i <= nn; i++)
1697 nvme_validate_ns(ctrl, i);
1699 nvme_remove_invalid_namespaces(ctrl, nn);
1702 static void nvme_scan_work(struct work_struct *work)
1704 struct nvme_ctrl *ctrl =
1705 container_of(work, struct nvme_ctrl, scan_work);
1706 struct nvme_id_ctrl *id;
1709 if (ctrl->state != NVME_CTRL_LIVE)
1712 if (nvme_identify_ctrl(ctrl, &id))
1715 mutex_lock(&ctrl->namespaces_mutex);
1716 nn = le32_to_cpu(id->nn);
1717 if (ctrl->vs >= NVME_VS(1, 1) &&
1718 !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
1719 if (!nvme_scan_ns_list(ctrl, nn))
1722 nvme_scan_ns_sequential(ctrl, nn);
1724 list_sort(NULL, &ctrl->namespaces, ns_cmp);
1725 mutex_unlock(&ctrl->namespaces_mutex);
1728 if (ctrl->ops->post_scan)
1729 ctrl->ops->post_scan(ctrl);
1732 void nvme_queue_scan(struct nvme_ctrl *ctrl)
1735 * Do not queue new scan work when a controller is reset during
1738 if (ctrl->state == NVME_CTRL_LIVE)
1739 schedule_work(&ctrl->scan_work);
1741 EXPORT_SYMBOL_GPL(nvme_queue_scan);
1743 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
1745 struct nvme_ns *ns, *next;
1748 * The dead states indicates the controller was not gracefully
1749 * disconnected. In that case, we won't be able to flush any data while
1750 * removing the namespaces' disks; fail all the queues now to avoid
1751 * potentially having to clean up the failed sync later.
1753 if (ctrl->state == NVME_CTRL_DEAD)
1754 nvme_kill_queues(ctrl);
1756 mutex_lock(&ctrl->namespaces_mutex);
1757 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
1759 mutex_unlock(&ctrl->namespaces_mutex);
1761 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
1763 static void nvme_async_event_work(struct work_struct *work)
1765 struct nvme_ctrl *ctrl =
1766 container_of(work, struct nvme_ctrl, async_event_work);
1768 spin_lock_irq(&ctrl->lock);
1769 while (ctrl->event_limit > 0) {
1770 int aer_idx = --ctrl->event_limit;
1772 spin_unlock_irq(&ctrl->lock);
1773 ctrl->ops->submit_async_event(ctrl, aer_idx);
1774 spin_lock_irq(&ctrl->lock);
1776 spin_unlock_irq(&ctrl->lock);
1779 void nvme_complete_async_event(struct nvme_ctrl *ctrl,
1780 struct nvme_completion *cqe)
1782 u16 status = le16_to_cpu(cqe->status) >> 1;
1783 u32 result = le32_to_cpu(cqe->result);
1785 if (status == NVME_SC_SUCCESS || status == NVME_SC_ABORT_REQ) {
1786 ++ctrl->event_limit;
1787 schedule_work(&ctrl->async_event_work);
1790 if (status != NVME_SC_SUCCESS)
1793 switch (result & 0xff07) {
1794 case NVME_AER_NOTICE_NS_CHANGED:
1795 dev_info(ctrl->device, "rescanning\n");
1796 nvme_queue_scan(ctrl);
1799 dev_warn(ctrl->device, "async event result %08x\n", result);
1802 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
1804 void nvme_queue_async_events(struct nvme_ctrl *ctrl)
1806 ctrl->event_limit = NVME_NR_AERS;
1807 schedule_work(&ctrl->async_event_work);
1809 EXPORT_SYMBOL_GPL(nvme_queue_async_events);
1811 static DEFINE_IDA(nvme_instance_ida);
1813 static int nvme_set_instance(struct nvme_ctrl *ctrl)
1815 int instance, error;
1818 if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL))
1821 spin_lock(&dev_list_lock);
1822 error = ida_get_new(&nvme_instance_ida, &instance);
1823 spin_unlock(&dev_list_lock);
1824 } while (error == -EAGAIN);
1829 ctrl->instance = instance;
1833 static void nvme_release_instance(struct nvme_ctrl *ctrl)
1835 spin_lock(&dev_list_lock);
1836 ida_remove(&nvme_instance_ida, ctrl->instance);
1837 spin_unlock(&dev_list_lock);
1840 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
1842 flush_work(&ctrl->async_event_work);
1843 flush_work(&ctrl->scan_work);
1844 nvme_remove_namespaces(ctrl);
1846 device_destroy(nvme_class, MKDEV(nvme_char_major, ctrl->instance));
1848 spin_lock(&dev_list_lock);
1849 list_del(&ctrl->node);
1850 spin_unlock(&dev_list_lock);
1852 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
1854 static void nvme_free_ctrl(struct kref *kref)
1856 struct nvme_ctrl *ctrl = container_of(kref, struct nvme_ctrl, kref);
1858 put_device(ctrl->device);
1859 nvme_release_instance(ctrl);
1860 ida_destroy(&ctrl->ns_ida);
1862 ctrl->ops->free_ctrl(ctrl);
1865 void nvme_put_ctrl(struct nvme_ctrl *ctrl)
1867 kref_put(&ctrl->kref, nvme_free_ctrl);
1869 EXPORT_SYMBOL_GPL(nvme_put_ctrl);
1872 * Initialize a NVMe controller structures. This needs to be called during
1873 * earliest initialization so that we have the initialized structured around
1876 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
1877 const struct nvme_ctrl_ops *ops, unsigned long quirks)
1881 ctrl->state = NVME_CTRL_NEW;
1882 spin_lock_init(&ctrl->lock);
1883 INIT_LIST_HEAD(&ctrl->namespaces);
1884 mutex_init(&ctrl->namespaces_mutex);
1885 kref_init(&ctrl->kref);
1888 ctrl->quirks = quirks;
1889 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
1890 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
1892 ret = nvme_set_instance(ctrl);
1896 ctrl->device = device_create_with_groups(nvme_class, ctrl->dev,
1897 MKDEV(nvme_char_major, ctrl->instance),
1898 ctrl, nvme_dev_attr_groups,
1899 "nvme%d", ctrl->instance);
1900 if (IS_ERR(ctrl->device)) {
1901 ret = PTR_ERR(ctrl->device);
1902 goto out_release_instance;
1904 get_device(ctrl->device);
1905 ida_init(&ctrl->ns_ida);
1907 spin_lock(&dev_list_lock);
1908 list_add_tail(&ctrl->node, &nvme_ctrl_list);
1909 spin_unlock(&dev_list_lock);
1912 out_release_instance:
1913 nvme_release_instance(ctrl);
1917 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
1920 * nvme_kill_queues(): Ends all namespace queues
1921 * @ctrl: the dead controller that needs to end
1923 * Call this function when the driver determines it is unable to get the
1924 * controller in a state capable of servicing IO.
1926 void nvme_kill_queues(struct nvme_ctrl *ctrl)
1931 list_for_each_entry_rcu(ns, &ctrl->namespaces, list) {
1932 if (!kref_get_unless_zero(&ns->kref))
1936 * Revalidating a dead namespace sets capacity to 0. This will
1937 * end buffered writers dirtying pages that can't be synced.
1939 if (!test_and_set_bit(NVME_NS_DEAD, &ns->flags))
1940 revalidate_disk(ns->disk);
1942 blk_set_queue_dying(ns->queue);
1943 blk_mq_abort_requeue_list(ns->queue);
1944 blk_mq_start_stopped_hw_queues(ns->queue, true);
1950 EXPORT_SYMBOL_GPL(nvme_kill_queues);
1952 void nvme_stop_queues(struct nvme_ctrl *ctrl)
1957 list_for_each_entry_rcu(ns, &ctrl->namespaces, list) {
1958 spin_lock_irq(ns->queue->queue_lock);
1959 queue_flag_set(QUEUE_FLAG_STOPPED, ns->queue);
1960 spin_unlock_irq(ns->queue->queue_lock);
1962 blk_mq_cancel_requeue_work(ns->queue);
1963 blk_mq_stop_hw_queues(ns->queue);
1967 EXPORT_SYMBOL_GPL(nvme_stop_queues);
1969 void nvme_start_queues(struct nvme_ctrl *ctrl)
1974 list_for_each_entry_rcu(ns, &ctrl->namespaces, list) {
1975 queue_flag_clear_unlocked(QUEUE_FLAG_STOPPED, ns->queue);
1976 blk_mq_start_stopped_hw_queues(ns->queue, true);
1977 blk_mq_kick_requeue_list(ns->queue);
1981 EXPORT_SYMBOL_GPL(nvme_start_queues);
1983 int __init nvme_core_init(void)
1987 result = register_blkdev(nvme_major, "nvme");
1990 else if (result > 0)
1991 nvme_major = result;
1993 result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme",
1996 goto unregister_blkdev;
1997 else if (result > 0)
1998 nvme_char_major = result;
2000 nvme_class = class_create(THIS_MODULE, "nvme");
2001 if (IS_ERR(nvme_class)) {
2002 result = PTR_ERR(nvme_class);
2003 goto unregister_chrdev;
2009 __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2011 unregister_blkdev(nvme_major, "nvme");
2015 void nvme_core_exit(void)
2017 class_destroy(nvme_class);
2018 __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2019 unregister_blkdev(nvme_major, "nvme");
2022 MODULE_LICENSE("GPL");
2023 MODULE_VERSION("1.0");
2024 module_init(nvme_core_init);
2025 module_exit(nvme_core_exit);