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)
199 req = blk_mq_alloc_request(q, nvme_is_write(cmd), flags);
203 req->cmd_type = REQ_TYPE_DRV_PRIV;
204 req->cmd_flags |= REQ_FAILFAST_DRIVER;
206 req->__sector = (sector_t) -1;
207 req->bio = req->biotail = NULL;
209 req->cmd = (unsigned char *)cmd;
210 req->cmd_len = sizeof(struct nvme_command);
214 EXPORT_SYMBOL_GPL(nvme_alloc_request);
216 static inline void nvme_setup_flush(struct nvme_ns *ns,
217 struct nvme_command *cmnd)
219 memset(cmnd, 0, sizeof(*cmnd));
220 cmnd->common.opcode = nvme_cmd_flush;
221 cmnd->common.nsid = cpu_to_le32(ns->ns_id);
224 static inline int nvme_setup_discard(struct nvme_ns *ns, struct request *req,
225 struct nvme_command *cmnd)
227 struct nvme_dsm_range *range;
230 unsigned int nr_bytes = blk_rq_bytes(req);
232 range = kmalloc(sizeof(*range), GFP_ATOMIC);
234 return BLK_MQ_RQ_QUEUE_BUSY;
236 range->cattr = cpu_to_le32(0);
237 range->nlb = cpu_to_le32(nr_bytes >> ns->lba_shift);
238 range->slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
240 memset(cmnd, 0, sizeof(*cmnd));
241 cmnd->dsm.opcode = nvme_cmd_dsm;
242 cmnd->dsm.nsid = cpu_to_le32(ns->ns_id);
244 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
246 req->completion_data = range;
247 page = virt_to_page(range);
248 offset = offset_in_page(range);
249 blk_add_request_payload(req, page, offset, sizeof(*range));
252 * we set __data_len back to the size of the area to be discarded
253 * on disk. This allows us to report completion on the full amount
254 * of blocks described by the request.
256 req->__data_len = nr_bytes;
261 static inline void nvme_setup_rw(struct nvme_ns *ns, struct request *req,
262 struct nvme_command *cmnd)
267 if (req->cmd_flags & REQ_FUA)
268 control |= NVME_RW_FUA;
269 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
270 control |= NVME_RW_LR;
272 if (req->cmd_flags & REQ_RAHEAD)
273 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
275 memset(cmnd, 0, sizeof(*cmnd));
276 cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
277 cmnd->rw.command_id = req->tag;
278 cmnd->rw.nsid = cpu_to_le32(ns->ns_id);
279 cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
280 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
283 switch (ns->pi_type) {
284 case NVME_NS_DPS_PI_TYPE3:
285 control |= NVME_RW_PRINFO_PRCHK_GUARD;
287 case NVME_NS_DPS_PI_TYPE1:
288 case NVME_NS_DPS_PI_TYPE2:
289 control |= NVME_RW_PRINFO_PRCHK_GUARD |
290 NVME_RW_PRINFO_PRCHK_REF;
291 cmnd->rw.reftag = cpu_to_le32(
292 nvme_block_nr(ns, blk_rq_pos(req)));
295 if (!blk_integrity_rq(req))
296 control |= NVME_RW_PRINFO_PRACT;
299 cmnd->rw.control = cpu_to_le16(control);
300 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
303 int nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
304 struct nvme_command *cmd)
308 if (req->cmd_type == REQ_TYPE_DRV_PRIV)
309 memcpy(cmd, req->cmd, sizeof(*cmd));
310 else if (req_op(req) == REQ_OP_FLUSH)
311 nvme_setup_flush(ns, cmd);
312 else if (req_op(req) == REQ_OP_DISCARD)
313 ret = nvme_setup_discard(ns, req, cmd);
315 nvme_setup_rw(ns, req, cmd);
319 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
322 * Returns 0 on success. If the result is negative, it's a Linux error code;
323 * if the result is positive, it's an NVM Express status code
325 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
326 struct nvme_completion *cqe, void *buffer, unsigned bufflen,
332 req = nvme_alloc_request(q, cmd, 0);
336 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
339 if (buffer && bufflen) {
340 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
345 blk_execute_rq(req->q, NULL, req, 0);
348 blk_mq_free_request(req);
352 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
353 void *buffer, unsigned bufflen)
355 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0);
357 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
359 int __nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
360 void __user *ubuffer, unsigned bufflen,
361 void __user *meta_buffer, unsigned meta_len, u32 meta_seed,
362 u32 *result, unsigned timeout)
364 bool write = nvme_is_write(cmd);
365 struct nvme_completion cqe;
366 struct nvme_ns *ns = q->queuedata;
367 struct gendisk *disk = ns ? ns->disk : NULL;
369 struct bio *bio = NULL;
373 req = nvme_alloc_request(q, cmd, 0);
377 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
380 if (ubuffer && bufflen) {
381 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
389 bio->bi_bdev = bdget_disk(disk, 0);
395 if (meta_buffer && meta_len) {
396 struct bio_integrity_payload *bip;
398 meta = kmalloc(meta_len, GFP_KERNEL);
405 if (copy_from_user(meta, meta_buffer,
412 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
418 bip->bip_iter.bi_size = meta_len;
419 bip->bip_iter.bi_sector = meta_seed;
421 ret = bio_integrity_add_page(bio, virt_to_page(meta),
422 meta_len, offset_in_page(meta));
423 if (ret != meta_len) {
430 blk_execute_rq(req->q, disk, req, 0);
433 *result = le32_to_cpu(cqe.result);
434 if (meta && !ret && !write) {
435 if (copy_to_user(meta_buffer, meta, meta_len))
442 if (disk && bio->bi_bdev)
444 blk_rq_unmap_user(bio);
447 blk_mq_free_request(req);
451 int nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
452 void __user *ubuffer, unsigned bufflen, u32 *result,
455 return __nvme_submit_user_cmd(q, cmd, ubuffer, bufflen, NULL, 0, 0,
459 int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
461 struct nvme_command c = { };
464 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
465 c.identify.opcode = nvme_admin_identify;
466 c.identify.cns = cpu_to_le32(1);
468 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
472 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
473 sizeof(struct nvme_id_ctrl));
479 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
481 struct nvme_command c = { };
483 c.identify.opcode = nvme_admin_identify;
484 c.identify.cns = cpu_to_le32(2);
485 c.identify.nsid = cpu_to_le32(nsid);
486 return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000);
489 int nvme_identify_ns(struct nvme_ctrl *dev, unsigned nsid,
490 struct nvme_id_ns **id)
492 struct nvme_command c = { };
495 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
496 c.identify.opcode = nvme_admin_identify,
497 c.identify.nsid = cpu_to_le32(nsid),
499 *id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL);
503 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
504 sizeof(struct nvme_id_ns));
510 int nvme_get_features(struct nvme_ctrl *dev, unsigned fid, unsigned nsid,
511 dma_addr_t dma_addr, u32 *result)
513 struct nvme_command c;
514 struct nvme_completion cqe;
517 memset(&c, 0, sizeof(c));
518 c.features.opcode = nvme_admin_get_features;
519 c.features.nsid = cpu_to_le32(nsid);
520 c.features.prp1 = cpu_to_le64(dma_addr);
521 c.features.fid = cpu_to_le32(fid);
523 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &cqe, NULL, 0, 0);
525 *result = le32_to_cpu(cqe.result);
529 int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
530 dma_addr_t dma_addr, u32 *result)
532 struct nvme_command c;
533 struct nvme_completion cqe;
536 memset(&c, 0, sizeof(c));
537 c.features.opcode = nvme_admin_set_features;
538 c.features.prp1 = cpu_to_le64(dma_addr);
539 c.features.fid = cpu_to_le32(fid);
540 c.features.dword11 = cpu_to_le32(dword11);
542 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &cqe, NULL, 0, 0);
544 *result = le32_to_cpu(cqe.result);
548 int nvme_get_log_page(struct nvme_ctrl *dev, struct nvme_smart_log **log)
550 struct nvme_command c = { };
553 c.common.opcode = nvme_admin_get_log_page,
554 c.common.nsid = cpu_to_le32(0xFFFFFFFF),
555 c.common.cdw10[0] = cpu_to_le32(
556 (((sizeof(struct nvme_smart_log) / 4) - 1) << 16) |
559 *log = kmalloc(sizeof(struct nvme_smart_log), GFP_KERNEL);
563 error = nvme_submit_sync_cmd(dev->admin_q, &c, *log,
564 sizeof(struct nvme_smart_log));
570 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
572 u32 q_count = (*count - 1) | ((*count - 1) << 16);
574 int status, nr_io_queues;
576 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, 0,
582 * Degraded controllers might return an error when setting the queue
583 * count. We still want to be able to bring them online and offer
584 * access to the admin queue, as that might be only way to fix them up.
587 dev_err(ctrl->dev, "Could not set queue count (%d)\n", status);
590 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
591 *count = min(*count, nr_io_queues);
596 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
598 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
600 struct nvme_user_io io;
601 struct nvme_command c;
602 unsigned length, meta_len;
603 void __user *metadata;
605 if (copy_from_user(&io, uio, sizeof(io)))
613 case nvme_cmd_compare:
619 length = (io.nblocks + 1) << ns->lba_shift;
620 meta_len = (io.nblocks + 1) * ns->ms;
621 metadata = (void __user *)(uintptr_t)io.metadata;
626 } else if (meta_len) {
627 if ((io.metadata & 3) || !io.metadata)
631 memset(&c, 0, sizeof(c));
632 c.rw.opcode = io.opcode;
633 c.rw.flags = io.flags;
634 c.rw.nsid = cpu_to_le32(ns->ns_id);
635 c.rw.slba = cpu_to_le64(io.slba);
636 c.rw.length = cpu_to_le16(io.nblocks);
637 c.rw.control = cpu_to_le16(io.control);
638 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
639 c.rw.reftag = cpu_to_le32(io.reftag);
640 c.rw.apptag = cpu_to_le16(io.apptag);
641 c.rw.appmask = cpu_to_le16(io.appmask);
643 return __nvme_submit_user_cmd(ns->queue, &c,
644 (void __user *)(uintptr_t)io.addr, length,
645 metadata, meta_len, io.slba, NULL, 0);
648 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
649 struct nvme_passthru_cmd __user *ucmd)
651 struct nvme_passthru_cmd cmd;
652 struct nvme_command c;
653 unsigned timeout = 0;
656 if (!capable(CAP_SYS_ADMIN))
658 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
663 memset(&c, 0, sizeof(c));
664 c.common.opcode = cmd.opcode;
665 c.common.flags = cmd.flags;
666 c.common.nsid = cpu_to_le32(cmd.nsid);
667 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
668 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
669 c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
670 c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
671 c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
672 c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
673 c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
674 c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
677 timeout = msecs_to_jiffies(cmd.timeout_ms);
679 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
680 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
681 &cmd.result, timeout);
683 if (put_user(cmd.result, &ucmd->result))
690 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
691 unsigned int cmd, unsigned long arg)
693 struct nvme_ns *ns = bdev->bd_disk->private_data;
697 force_successful_syscall_return();
699 case NVME_IOCTL_ADMIN_CMD:
700 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
701 case NVME_IOCTL_IO_CMD:
702 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
703 case NVME_IOCTL_SUBMIT_IO:
704 return nvme_submit_io(ns, (void __user *)arg);
705 #ifdef CONFIG_BLK_DEV_NVME_SCSI
706 case SG_GET_VERSION_NUM:
707 return nvme_sg_get_version_num((void __user *)arg);
709 return nvme_sg_io(ns, (void __user *)arg);
717 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
718 unsigned int cmd, unsigned long arg)
724 return nvme_ioctl(bdev, mode, cmd, arg);
727 #define nvme_compat_ioctl NULL
730 static int nvme_open(struct block_device *bdev, fmode_t mode)
732 return nvme_get_ns_from_disk(bdev->bd_disk) ? 0 : -ENXIO;
735 static void nvme_release(struct gendisk *disk, fmode_t mode)
737 struct nvme_ns *ns = disk->private_data;
739 module_put(ns->ctrl->ops->module);
743 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
745 /* some standard values */
747 geo->sectors = 1 << 5;
748 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
752 #ifdef CONFIG_BLK_DEV_INTEGRITY
753 static void nvme_init_integrity(struct nvme_ns *ns)
755 struct blk_integrity integrity;
757 switch (ns->pi_type) {
758 case NVME_NS_DPS_PI_TYPE3:
759 integrity.profile = &t10_pi_type3_crc;
760 integrity.tag_size = sizeof(u16) + sizeof(u32);
761 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
763 case NVME_NS_DPS_PI_TYPE1:
764 case NVME_NS_DPS_PI_TYPE2:
765 integrity.profile = &t10_pi_type1_crc;
766 integrity.tag_size = sizeof(u16);
767 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
770 integrity.profile = NULL;
773 integrity.tuple_size = ns->ms;
774 blk_integrity_register(ns->disk, &integrity);
775 blk_queue_max_integrity_segments(ns->queue, 1);
778 static void nvme_init_integrity(struct nvme_ns *ns)
781 #endif /* CONFIG_BLK_DEV_INTEGRITY */
783 static void nvme_config_discard(struct nvme_ns *ns)
785 struct nvme_ctrl *ctrl = ns->ctrl;
786 u32 logical_block_size = queue_logical_block_size(ns->queue);
788 if (ctrl->quirks & NVME_QUIRK_DISCARD_ZEROES)
789 ns->queue->limits.discard_zeroes_data = 1;
791 ns->queue->limits.discard_zeroes_data = 0;
793 ns->queue->limits.discard_alignment = logical_block_size;
794 ns->queue->limits.discard_granularity = logical_block_size;
795 blk_queue_max_discard_sectors(ns->queue, UINT_MAX);
796 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue);
799 static int nvme_revalidate_disk(struct gendisk *disk)
801 struct nvme_ns *ns = disk->private_data;
802 struct nvme_id_ns *id;
807 if (test_bit(NVME_NS_DEAD, &ns->flags)) {
808 set_capacity(disk, 0);
811 if (nvme_identify_ns(ns->ctrl, ns->ns_id, &id)) {
812 dev_warn(disk_to_dev(ns->disk), "%s: Identify failure\n",
821 if (nvme_nvm_ns_supported(ns, id) && ns->type != NVME_NS_LIGHTNVM) {
822 if (nvme_nvm_register(ns->queue, disk->disk_name)) {
823 dev_warn(disk_to_dev(ns->disk),
824 "%s: LightNVM init failure\n", __func__);
828 ns->type = NVME_NS_LIGHTNVM;
831 if (ns->ctrl->vs >= NVME_VS(1, 1))
832 memcpy(ns->eui, id->eui64, sizeof(ns->eui));
833 if (ns->ctrl->vs >= NVME_VS(1, 2))
834 memcpy(ns->uuid, id->nguid, sizeof(ns->uuid));
837 lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
838 ns->lba_shift = id->lbaf[lbaf].ds;
839 ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
840 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
843 * If identify namespace failed, use default 512 byte block size so
844 * block layer can use before failing read/write for 0 capacity.
846 if (ns->lba_shift == 0)
848 bs = 1 << ns->lba_shift;
849 /* XXX: PI implementation requires metadata equal t10 pi tuple size */
850 pi_type = ns->ms == sizeof(struct t10_pi_tuple) ?
851 id->dps & NVME_NS_DPS_PI_MASK : 0;
853 blk_mq_freeze_queue(disk->queue);
854 if (blk_get_integrity(disk) && (ns->pi_type != pi_type ||
856 bs != queue_logical_block_size(disk->queue) ||
857 (ns->ms && ns->ext)))
858 blk_integrity_unregister(disk);
860 ns->pi_type = pi_type;
861 blk_queue_logical_block_size(ns->queue, bs);
863 if (ns->ms && !blk_get_integrity(disk) && !ns->ext)
864 nvme_init_integrity(ns);
865 if (ns->ms && !(ns->ms == 8 && ns->pi_type) && !blk_get_integrity(disk))
866 set_capacity(disk, 0);
868 set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
870 if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
871 nvme_config_discard(ns);
872 blk_mq_unfreeze_queue(disk->queue);
878 static char nvme_pr_type(enum pr_type type)
881 case PR_WRITE_EXCLUSIVE:
883 case PR_EXCLUSIVE_ACCESS:
885 case PR_WRITE_EXCLUSIVE_REG_ONLY:
887 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
889 case PR_WRITE_EXCLUSIVE_ALL_REGS:
891 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
898 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
899 u64 key, u64 sa_key, u8 op)
901 struct nvme_ns *ns = bdev->bd_disk->private_data;
902 struct nvme_command c;
903 u8 data[16] = { 0, };
905 put_unaligned_le64(key, &data[0]);
906 put_unaligned_le64(sa_key, &data[8]);
908 memset(&c, 0, sizeof(c));
909 c.common.opcode = op;
910 c.common.nsid = cpu_to_le32(ns->ns_id);
911 c.common.cdw10[0] = cpu_to_le32(cdw10);
913 return nvme_submit_sync_cmd(ns->queue, &c, data, 16);
916 static int nvme_pr_register(struct block_device *bdev, u64 old,
917 u64 new, unsigned flags)
921 if (flags & ~PR_FL_IGNORE_KEY)
925 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
926 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
927 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
930 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
931 enum pr_type type, unsigned flags)
935 if (flags & ~PR_FL_IGNORE_KEY)
938 cdw10 = nvme_pr_type(type) << 8;
939 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
940 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
943 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
944 enum pr_type type, bool abort)
946 u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
947 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
950 static int nvme_pr_clear(struct block_device *bdev, u64 key)
952 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
953 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
956 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
958 u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
959 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
962 static const struct pr_ops nvme_pr_ops = {
963 .pr_register = nvme_pr_register,
964 .pr_reserve = nvme_pr_reserve,
965 .pr_release = nvme_pr_release,
966 .pr_preempt = nvme_pr_preempt,
967 .pr_clear = nvme_pr_clear,
970 static const struct block_device_operations nvme_fops = {
971 .owner = THIS_MODULE,
973 .compat_ioctl = nvme_compat_ioctl,
975 .release = nvme_release,
976 .getgeo = nvme_getgeo,
977 .revalidate_disk= nvme_revalidate_disk,
978 .pr_ops = &nvme_pr_ops,
981 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
983 unsigned long timeout =
984 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
985 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
988 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
989 if ((csts & NVME_CSTS_RDY) == bit)
993 if (fatal_signal_pending(current))
995 if (time_after(jiffies, timeout)) {
996 dev_err(ctrl->device,
997 "Device not ready; aborting %s\n", enabled ?
998 "initialisation" : "reset");
1007 * If the device has been passed off to us in an enabled state, just clear
1008 * the enabled bit. The spec says we should set the 'shutdown notification
1009 * bits', but doing so may cause the device to complete commands to the
1010 * admin queue ... and we don't know what memory that might be pointing at!
1012 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1016 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1017 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1019 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1022 return nvme_wait_ready(ctrl, cap, false);
1024 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1026 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1029 * Default to a 4K page size, with the intention to update this
1030 * path in the future to accomodate architectures with differing
1031 * kernel and IO page sizes.
1033 unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1036 if (page_shift < dev_page_min) {
1037 dev_err(ctrl->device,
1038 "Minimum device page size %u too large for host (%u)\n",
1039 1 << dev_page_min, 1 << page_shift);
1043 ctrl->page_size = 1 << page_shift;
1045 ctrl->ctrl_config = NVME_CC_CSS_NVM;
1046 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1047 ctrl->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE;
1048 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1049 ctrl->ctrl_config |= NVME_CC_ENABLE;
1051 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1054 return nvme_wait_ready(ctrl, cap, true);
1056 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1058 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1060 unsigned long timeout = SHUTDOWN_TIMEOUT + jiffies;
1064 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1065 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1067 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1071 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1072 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1076 if (fatal_signal_pending(current))
1078 if (time_after(jiffies, timeout)) {
1079 dev_err(ctrl->device,
1080 "Device shutdown incomplete; abort shutdown\n");
1087 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1089 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1090 struct request_queue *q)
1094 if (ctrl->max_hw_sectors) {
1096 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1098 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1099 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1101 if (ctrl->stripe_size)
1102 blk_queue_chunk_sectors(q, ctrl->stripe_size >> 9);
1103 blk_queue_virt_boundary(q, ctrl->page_size - 1);
1104 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1106 blk_queue_write_cache(q, vwc, vwc);
1110 * Initialize the cached copies of the Identify data and various controller
1111 * register in our nvme_ctrl structure. This should be called as soon as
1112 * the admin queue is fully up and running.
1114 int nvme_init_identify(struct nvme_ctrl *ctrl)
1116 struct nvme_id_ctrl *id;
1118 int ret, page_shift;
1121 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
1123 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
1127 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
1129 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
1132 page_shift = NVME_CAP_MPSMIN(cap) + 12;
1134 if (ctrl->vs >= NVME_VS(1, 1))
1135 ctrl->subsystem = NVME_CAP_NSSRC(cap);
1137 ret = nvme_identify_ctrl(ctrl, &id);
1139 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
1143 ctrl->vid = le16_to_cpu(id->vid);
1144 ctrl->oncs = le16_to_cpup(&id->oncs);
1145 atomic_set(&ctrl->abort_limit, id->acl + 1);
1146 ctrl->vwc = id->vwc;
1147 ctrl->cntlid = le16_to_cpup(&id->cntlid);
1148 memcpy(ctrl->serial, id->sn, sizeof(id->sn));
1149 memcpy(ctrl->model, id->mn, sizeof(id->mn));
1150 memcpy(ctrl->firmware_rev, id->fr, sizeof(id->fr));
1152 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
1154 max_hw_sectors = UINT_MAX;
1155 ctrl->max_hw_sectors =
1156 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
1158 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) && id->vs[3]) {
1159 unsigned int max_hw_sectors;
1161 ctrl->stripe_size = 1 << (id->vs[3] + page_shift);
1162 max_hw_sectors = ctrl->stripe_size >> (page_shift - 9);
1163 if (ctrl->max_hw_sectors) {
1164 ctrl->max_hw_sectors = min(max_hw_sectors,
1165 ctrl->max_hw_sectors);
1167 ctrl->max_hw_sectors = max_hw_sectors;
1171 nvme_set_queue_limits(ctrl, ctrl->admin_q);
1176 EXPORT_SYMBOL_GPL(nvme_init_identify);
1178 static int nvme_dev_open(struct inode *inode, struct file *file)
1180 struct nvme_ctrl *ctrl;
1181 int instance = iminor(inode);
1184 spin_lock(&dev_list_lock);
1185 list_for_each_entry(ctrl, &nvme_ctrl_list, node) {
1186 if (ctrl->instance != instance)
1189 if (!ctrl->admin_q) {
1193 if (!kref_get_unless_zero(&ctrl->kref))
1195 file->private_data = ctrl;
1199 spin_unlock(&dev_list_lock);
1204 static int nvme_dev_release(struct inode *inode, struct file *file)
1206 nvme_put_ctrl(file->private_data);
1210 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
1215 mutex_lock(&ctrl->namespaces_mutex);
1216 if (list_empty(&ctrl->namespaces)) {
1221 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
1222 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
1223 dev_warn(ctrl->device,
1224 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
1229 dev_warn(ctrl->device,
1230 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
1231 kref_get(&ns->kref);
1232 mutex_unlock(&ctrl->namespaces_mutex);
1234 ret = nvme_user_cmd(ctrl, ns, argp);
1239 mutex_unlock(&ctrl->namespaces_mutex);
1243 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
1246 struct nvme_ctrl *ctrl = file->private_data;
1247 void __user *argp = (void __user *)arg;
1250 case NVME_IOCTL_ADMIN_CMD:
1251 return nvme_user_cmd(ctrl, NULL, argp);
1252 case NVME_IOCTL_IO_CMD:
1253 return nvme_dev_user_cmd(ctrl, argp);
1254 case NVME_IOCTL_RESET:
1255 dev_warn(ctrl->device, "resetting controller\n");
1256 return ctrl->ops->reset_ctrl(ctrl);
1257 case NVME_IOCTL_SUBSYS_RESET:
1258 return nvme_reset_subsystem(ctrl);
1259 case NVME_IOCTL_RESCAN:
1260 nvme_queue_scan(ctrl);
1267 static const struct file_operations nvme_dev_fops = {
1268 .owner = THIS_MODULE,
1269 .open = nvme_dev_open,
1270 .release = nvme_dev_release,
1271 .unlocked_ioctl = nvme_dev_ioctl,
1272 .compat_ioctl = nvme_dev_ioctl,
1275 static ssize_t nvme_sysfs_reset(struct device *dev,
1276 struct device_attribute *attr, const char *buf,
1279 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1282 ret = ctrl->ops->reset_ctrl(ctrl);
1287 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
1289 static ssize_t nvme_sysfs_rescan(struct device *dev,
1290 struct device_attribute *attr, const char *buf,
1293 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1295 nvme_queue_scan(ctrl);
1298 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
1300 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
1303 struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1304 struct nvme_ctrl *ctrl = ns->ctrl;
1305 int serial_len = sizeof(ctrl->serial);
1306 int model_len = sizeof(ctrl->model);
1308 if (memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1309 return sprintf(buf, "eui.%16phN\n", ns->uuid);
1311 if (memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1312 return sprintf(buf, "eui.%8phN\n", ns->eui);
1314 while (ctrl->serial[serial_len - 1] == ' ')
1316 while (ctrl->model[model_len - 1] == ' ')
1319 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", ctrl->vid,
1320 serial_len, ctrl->serial, model_len, ctrl->model, ns->ns_id);
1322 static DEVICE_ATTR(wwid, S_IRUGO, wwid_show, NULL);
1324 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
1327 struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1328 return sprintf(buf, "%pU\n", ns->uuid);
1330 static DEVICE_ATTR(uuid, S_IRUGO, uuid_show, NULL);
1332 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
1335 struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1336 return sprintf(buf, "%8phd\n", ns->eui);
1338 static DEVICE_ATTR(eui, S_IRUGO, eui_show, NULL);
1340 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
1343 struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1344 return sprintf(buf, "%d\n", ns->ns_id);
1346 static DEVICE_ATTR(nsid, S_IRUGO, nsid_show, NULL);
1348 static struct attribute *nvme_ns_attrs[] = {
1349 &dev_attr_wwid.attr,
1350 &dev_attr_uuid.attr,
1352 &dev_attr_nsid.attr,
1356 static umode_t nvme_attrs_are_visible(struct kobject *kobj,
1357 struct attribute *a, int n)
1359 struct device *dev = container_of(kobj, struct device, kobj);
1360 struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1362 if (a == &dev_attr_uuid.attr) {
1363 if (!memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1366 if (a == &dev_attr_eui.attr) {
1367 if (!memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1373 static const struct attribute_group nvme_ns_attr_group = {
1374 .attrs = nvme_ns_attrs,
1375 .is_visible = nvme_attrs_are_visible,
1378 #define nvme_show_str_function(field) \
1379 static ssize_t field##_show(struct device *dev, \
1380 struct device_attribute *attr, char *buf) \
1382 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
1383 return sprintf(buf, "%.*s\n", (int)sizeof(ctrl->field), ctrl->field); \
1385 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1387 #define nvme_show_int_function(field) \
1388 static ssize_t field##_show(struct device *dev, \
1389 struct device_attribute *attr, char *buf) \
1391 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
1392 return sprintf(buf, "%d\n", ctrl->field); \
1394 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1396 nvme_show_str_function(model);
1397 nvme_show_str_function(serial);
1398 nvme_show_str_function(firmware_rev);
1399 nvme_show_int_function(cntlid);
1401 static struct attribute *nvme_dev_attrs[] = {
1402 &dev_attr_reset_controller.attr,
1403 &dev_attr_rescan_controller.attr,
1404 &dev_attr_model.attr,
1405 &dev_attr_serial.attr,
1406 &dev_attr_firmware_rev.attr,
1407 &dev_attr_cntlid.attr,
1411 static struct attribute_group nvme_dev_attrs_group = {
1412 .attrs = nvme_dev_attrs,
1415 static const struct attribute_group *nvme_dev_attr_groups[] = {
1416 &nvme_dev_attrs_group,
1420 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
1422 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
1423 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
1425 return nsa->ns_id - nsb->ns_id;
1428 static struct nvme_ns *nvme_find_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1432 lockdep_assert_held(&ctrl->namespaces_mutex);
1434 list_for_each_entry(ns, &ctrl->namespaces, list) {
1435 if (ns->ns_id == nsid)
1437 if (ns->ns_id > nsid)
1443 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1446 struct gendisk *disk;
1447 int node = dev_to_node(ctrl->dev);
1449 lockdep_assert_held(&ctrl->namespaces_mutex);
1451 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
1455 ns->instance = ida_simple_get(&ctrl->ns_ida, 1, 0, GFP_KERNEL);
1456 if (ns->instance < 0)
1459 ns->queue = blk_mq_init_queue(ctrl->tagset);
1460 if (IS_ERR(ns->queue))
1461 goto out_release_instance;
1462 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
1463 ns->queue->queuedata = ns;
1466 disk = alloc_disk_node(0, node);
1468 goto out_free_queue;
1470 kref_init(&ns->kref);
1473 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
1476 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
1477 nvme_set_queue_limits(ctrl, ns->queue);
1479 disk->major = nvme_major;
1480 disk->first_minor = 0;
1481 disk->fops = &nvme_fops;
1482 disk->private_data = ns;
1483 disk->queue = ns->queue;
1484 disk->driverfs_dev = ctrl->device;
1485 disk->flags = GENHD_FL_EXT_DEVT;
1486 sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance, ns->instance);
1488 if (nvme_revalidate_disk(ns->disk))
1491 list_add_tail_rcu(&ns->list, &ctrl->namespaces);
1492 kref_get(&ctrl->kref);
1493 if (ns->type == NVME_NS_LIGHTNVM)
1497 if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
1498 &nvme_ns_attr_group))
1499 pr_warn("%s: failed to create sysfs group for identification\n",
1500 ns->disk->disk_name);
1505 blk_cleanup_queue(ns->queue);
1506 out_release_instance:
1507 ida_simple_remove(&ctrl->ns_ida, ns->instance);
1512 static void nvme_ns_remove(struct nvme_ns *ns)
1514 lockdep_assert_held(&ns->ctrl->namespaces_mutex);
1516 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
1519 if (ns->disk->flags & GENHD_FL_UP) {
1520 if (blk_get_integrity(ns->disk))
1521 blk_integrity_unregister(ns->disk);
1522 sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
1523 &nvme_ns_attr_group);
1524 del_gendisk(ns->disk);
1525 blk_mq_abort_requeue_list(ns->queue);
1526 blk_cleanup_queue(ns->queue);
1528 list_del_init(&ns->list);
1533 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1537 ns = nvme_find_ns(ctrl, nsid);
1539 if (revalidate_disk(ns->disk))
1542 nvme_alloc_ns(ctrl, nsid);
1545 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
1548 struct nvme_ns *ns, *next;
1550 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
1551 if (ns->ns_id > nsid)
1556 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
1560 unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
1563 ns_list = kzalloc(0x1000, GFP_KERNEL);
1567 for (i = 0; i < num_lists; i++) {
1568 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
1572 for (j = 0; j < min(nn, 1024U); j++) {
1573 nsid = le32_to_cpu(ns_list[j]);
1577 nvme_validate_ns(ctrl, nsid);
1579 while (++prev < nsid) {
1580 ns = nvme_find_ns(ctrl, prev);
1588 nvme_remove_invalid_namespaces(ctrl, prev);
1594 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
1598 lockdep_assert_held(&ctrl->namespaces_mutex);
1600 for (i = 1; i <= nn; i++)
1601 nvme_validate_ns(ctrl, i);
1603 nvme_remove_invalid_namespaces(ctrl, nn);
1606 static void nvme_scan_work(struct work_struct *work)
1608 struct nvme_ctrl *ctrl =
1609 container_of(work, struct nvme_ctrl, scan_work);
1610 struct nvme_id_ctrl *id;
1613 if (ctrl->state != NVME_CTRL_LIVE)
1616 if (nvme_identify_ctrl(ctrl, &id))
1619 mutex_lock(&ctrl->namespaces_mutex);
1620 nn = le32_to_cpu(id->nn);
1621 if (ctrl->vs >= NVME_VS(1, 1) &&
1622 !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
1623 if (!nvme_scan_ns_list(ctrl, nn))
1626 nvme_scan_ns_sequential(ctrl, nn);
1628 list_sort(NULL, &ctrl->namespaces, ns_cmp);
1629 mutex_unlock(&ctrl->namespaces_mutex);
1632 if (ctrl->ops->post_scan)
1633 ctrl->ops->post_scan(ctrl);
1636 void nvme_queue_scan(struct nvme_ctrl *ctrl)
1639 * Do not queue new scan work when a controller is reset during
1642 if (ctrl->state == NVME_CTRL_LIVE)
1643 schedule_work(&ctrl->scan_work);
1645 EXPORT_SYMBOL_GPL(nvme_queue_scan);
1647 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
1649 struct nvme_ns *ns, *next;
1652 * The dead states indicates the controller was not gracefully
1653 * disconnected. In that case, we won't be able to flush any data while
1654 * removing the namespaces' disks; fail all the queues now to avoid
1655 * potentially having to clean up the failed sync later.
1657 if (ctrl->state == NVME_CTRL_DEAD)
1658 nvme_kill_queues(ctrl);
1660 mutex_lock(&ctrl->namespaces_mutex);
1661 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
1663 mutex_unlock(&ctrl->namespaces_mutex);
1665 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
1667 static void nvme_async_event_work(struct work_struct *work)
1669 struct nvme_ctrl *ctrl =
1670 container_of(work, struct nvme_ctrl, async_event_work);
1672 spin_lock_irq(&ctrl->lock);
1673 while (ctrl->event_limit > 0) {
1674 int aer_idx = --ctrl->event_limit;
1676 spin_unlock_irq(&ctrl->lock);
1677 ctrl->ops->submit_async_event(ctrl, aer_idx);
1678 spin_lock_irq(&ctrl->lock);
1680 spin_unlock_irq(&ctrl->lock);
1683 void nvme_complete_async_event(struct nvme_ctrl *ctrl,
1684 struct nvme_completion *cqe)
1686 u16 status = le16_to_cpu(cqe->status) >> 1;
1687 u32 result = le32_to_cpu(cqe->result);
1689 if (status == NVME_SC_SUCCESS || status == NVME_SC_ABORT_REQ) {
1690 ++ctrl->event_limit;
1691 schedule_work(&ctrl->async_event_work);
1694 if (status != NVME_SC_SUCCESS)
1697 switch (result & 0xff07) {
1698 case NVME_AER_NOTICE_NS_CHANGED:
1699 dev_info(ctrl->device, "rescanning\n");
1700 nvme_queue_scan(ctrl);
1703 dev_warn(ctrl->device, "async event result %08x\n", result);
1706 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
1708 void nvme_queue_async_events(struct nvme_ctrl *ctrl)
1710 ctrl->event_limit = NVME_NR_AERS;
1711 schedule_work(&ctrl->async_event_work);
1713 EXPORT_SYMBOL_GPL(nvme_queue_async_events);
1715 static DEFINE_IDA(nvme_instance_ida);
1717 static int nvme_set_instance(struct nvme_ctrl *ctrl)
1719 int instance, error;
1722 if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL))
1725 spin_lock(&dev_list_lock);
1726 error = ida_get_new(&nvme_instance_ida, &instance);
1727 spin_unlock(&dev_list_lock);
1728 } while (error == -EAGAIN);
1733 ctrl->instance = instance;
1737 static void nvme_release_instance(struct nvme_ctrl *ctrl)
1739 spin_lock(&dev_list_lock);
1740 ida_remove(&nvme_instance_ida, ctrl->instance);
1741 spin_unlock(&dev_list_lock);
1744 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
1746 flush_work(&ctrl->async_event_work);
1747 flush_work(&ctrl->scan_work);
1748 nvme_remove_namespaces(ctrl);
1750 device_destroy(nvme_class, MKDEV(nvme_char_major, ctrl->instance));
1752 spin_lock(&dev_list_lock);
1753 list_del(&ctrl->node);
1754 spin_unlock(&dev_list_lock);
1756 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
1758 static void nvme_free_ctrl(struct kref *kref)
1760 struct nvme_ctrl *ctrl = container_of(kref, struct nvme_ctrl, kref);
1762 put_device(ctrl->device);
1763 nvme_release_instance(ctrl);
1764 ida_destroy(&ctrl->ns_ida);
1766 ctrl->ops->free_ctrl(ctrl);
1769 void nvme_put_ctrl(struct nvme_ctrl *ctrl)
1771 kref_put(&ctrl->kref, nvme_free_ctrl);
1773 EXPORT_SYMBOL_GPL(nvme_put_ctrl);
1776 * Initialize a NVMe controller structures. This needs to be called during
1777 * earliest initialization so that we have the initialized structured around
1780 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
1781 const struct nvme_ctrl_ops *ops, unsigned long quirks)
1785 ctrl->state = NVME_CTRL_NEW;
1786 spin_lock_init(&ctrl->lock);
1787 INIT_LIST_HEAD(&ctrl->namespaces);
1788 mutex_init(&ctrl->namespaces_mutex);
1789 kref_init(&ctrl->kref);
1792 ctrl->quirks = quirks;
1793 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
1794 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
1796 ret = nvme_set_instance(ctrl);
1800 ctrl->device = device_create_with_groups(nvme_class, ctrl->dev,
1801 MKDEV(nvme_char_major, ctrl->instance),
1802 ctrl, nvme_dev_attr_groups,
1803 "nvme%d", ctrl->instance);
1804 if (IS_ERR(ctrl->device)) {
1805 ret = PTR_ERR(ctrl->device);
1806 goto out_release_instance;
1808 get_device(ctrl->device);
1809 ida_init(&ctrl->ns_ida);
1811 spin_lock(&dev_list_lock);
1812 list_add_tail(&ctrl->node, &nvme_ctrl_list);
1813 spin_unlock(&dev_list_lock);
1816 out_release_instance:
1817 nvme_release_instance(ctrl);
1821 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
1824 * nvme_kill_queues(): Ends all namespace queues
1825 * @ctrl: the dead controller that needs to end
1827 * Call this function when the driver determines it is unable to get the
1828 * controller in a state capable of servicing IO.
1830 void nvme_kill_queues(struct nvme_ctrl *ctrl)
1835 list_for_each_entry_rcu(ns, &ctrl->namespaces, list) {
1836 if (!kref_get_unless_zero(&ns->kref))
1840 * Revalidating a dead namespace sets capacity to 0. This will
1841 * end buffered writers dirtying pages that can't be synced.
1843 if (!test_and_set_bit(NVME_NS_DEAD, &ns->flags))
1844 revalidate_disk(ns->disk);
1846 blk_set_queue_dying(ns->queue);
1847 blk_mq_abort_requeue_list(ns->queue);
1848 blk_mq_start_stopped_hw_queues(ns->queue, true);
1854 EXPORT_SYMBOL_GPL(nvme_kill_queues);
1856 void nvme_stop_queues(struct nvme_ctrl *ctrl)
1861 list_for_each_entry_rcu(ns, &ctrl->namespaces, list) {
1862 spin_lock_irq(ns->queue->queue_lock);
1863 queue_flag_set(QUEUE_FLAG_STOPPED, ns->queue);
1864 spin_unlock_irq(ns->queue->queue_lock);
1866 blk_mq_cancel_requeue_work(ns->queue);
1867 blk_mq_stop_hw_queues(ns->queue);
1871 EXPORT_SYMBOL_GPL(nvme_stop_queues);
1873 void nvme_start_queues(struct nvme_ctrl *ctrl)
1878 list_for_each_entry_rcu(ns, &ctrl->namespaces, list) {
1879 queue_flag_clear_unlocked(QUEUE_FLAG_STOPPED, ns->queue);
1880 blk_mq_start_stopped_hw_queues(ns->queue, true);
1881 blk_mq_kick_requeue_list(ns->queue);
1885 EXPORT_SYMBOL_GPL(nvme_start_queues);
1887 int __init nvme_core_init(void)
1891 result = register_blkdev(nvme_major, "nvme");
1894 else if (result > 0)
1895 nvme_major = result;
1897 result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme",
1900 goto unregister_blkdev;
1901 else if (result > 0)
1902 nvme_char_major = result;
1904 nvme_class = class_create(THIS_MODULE, "nvme");
1905 if (IS_ERR(nvme_class)) {
1906 result = PTR_ERR(nvme_class);
1907 goto unregister_chrdev;
1913 __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
1915 unregister_blkdev(nvme_major, "nvme");
1919 void nvme_core_exit(void)
1921 class_destroy(nvme_class);
1922 __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
1923 unregister_blkdev(nvme_major, "nvme");
1926 MODULE_LICENSE("GPL");
1927 MODULE_VERSION("1.0");
1928 module_init(nvme_core_init);
1929 module_exit(nvme_core_exit);