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 static int nvme_major;
37 module_param(nvme_major, int, 0);
39 static int nvme_char_major;
40 module_param(nvme_char_major, int, 0);
42 static LIST_HEAD(nvme_ctrl_list);
43 DEFINE_SPINLOCK(dev_list_lock);
45 static struct class *nvme_class;
47 static void nvme_free_ns(struct kref *kref)
49 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
51 if (ns->type == NVME_NS_LIGHTNVM)
52 nvme_nvm_unregister(ns->queue, ns->disk->disk_name);
54 spin_lock(&dev_list_lock);
55 ns->disk->private_data = NULL;
56 spin_unlock(&dev_list_lock);
58 nvme_put_ctrl(ns->ctrl);
63 static void nvme_put_ns(struct nvme_ns *ns)
65 kref_put(&ns->kref, nvme_free_ns);
68 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk)
72 spin_lock(&dev_list_lock);
73 ns = disk->private_data;
74 if (ns && !kref_get_unless_zero(&ns->kref))
76 spin_unlock(&dev_list_lock);
81 void nvme_requeue_req(struct request *req)
85 blk_mq_requeue_request(req);
86 spin_lock_irqsave(req->q->queue_lock, flags);
87 if (!blk_queue_stopped(req->q))
88 blk_mq_kick_requeue_list(req->q);
89 spin_unlock_irqrestore(req->q->queue_lock, flags);
92 struct request *nvme_alloc_request(struct request_queue *q,
93 struct nvme_command *cmd, unsigned int flags)
95 bool write = cmd->common.opcode & 1;
98 req = blk_mq_alloc_request(q, write, flags);
102 req->cmd_type = REQ_TYPE_DRV_PRIV;
103 req->cmd_flags |= REQ_FAILFAST_DRIVER;
105 req->__sector = (sector_t) -1;
106 req->bio = req->biotail = NULL;
108 req->cmd = (unsigned char *)cmd;
109 req->cmd_len = sizeof(struct nvme_command);
110 req->special = (void *)0;
116 * Returns 0 on success. If the result is negative, it's a Linux error code;
117 * if the result is positive, it's an NVM Express status code
119 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
120 void *buffer, unsigned bufflen, u32 *result, unsigned timeout)
125 req = nvme_alloc_request(q, cmd, 0);
129 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
131 if (buffer && bufflen) {
132 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
137 blk_execute_rq(req->q, NULL, req, 0);
139 *result = (u32)(uintptr_t)req->special;
142 blk_mq_free_request(req);
146 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
147 void *buffer, unsigned bufflen)
149 return __nvme_submit_sync_cmd(q, cmd, buffer, bufflen, NULL, 0);
152 int __nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
153 void __user *ubuffer, unsigned bufflen,
154 void __user *meta_buffer, unsigned meta_len, u32 meta_seed,
155 u32 *result, unsigned timeout)
157 bool write = cmd->common.opcode & 1;
158 struct nvme_ns *ns = q->queuedata;
159 struct gendisk *disk = ns ? ns->disk : NULL;
161 struct bio *bio = NULL;
165 req = nvme_alloc_request(q, cmd, 0);
169 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
171 if (ubuffer && bufflen) {
172 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
180 bio->bi_bdev = bdget_disk(disk, 0);
187 struct bio_integrity_payload *bip;
189 meta = kmalloc(meta_len, GFP_KERNEL);
196 if (copy_from_user(meta, meta_buffer,
203 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
209 bip->bip_iter.bi_size = meta_len;
210 bip->bip_iter.bi_sector = meta_seed;
212 ret = bio_integrity_add_page(bio, virt_to_page(meta),
213 meta_len, offset_in_page(meta));
214 if (ret != meta_len) {
221 blk_execute_rq(req->q, disk, req, 0);
224 *result = (u32)(uintptr_t)req->special;
225 if (meta && !ret && !write) {
226 if (copy_to_user(meta_buffer, meta, meta_len))
233 if (disk && bio->bi_bdev)
235 blk_rq_unmap_user(bio);
238 blk_mq_free_request(req);
242 int nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
243 void __user *ubuffer, unsigned bufflen, u32 *result,
246 return __nvme_submit_user_cmd(q, cmd, ubuffer, bufflen, NULL, 0, 0,
250 int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
252 struct nvme_command c = { };
255 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
256 c.identify.opcode = nvme_admin_identify;
257 c.identify.cns = cpu_to_le32(1);
259 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
263 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
264 sizeof(struct nvme_id_ctrl));
270 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
272 struct nvme_command c = { };
274 c.identify.opcode = nvme_admin_identify;
275 c.identify.cns = cpu_to_le32(2);
276 c.identify.nsid = cpu_to_le32(nsid);
277 return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000);
280 int nvme_identify_ns(struct nvme_ctrl *dev, unsigned nsid,
281 struct nvme_id_ns **id)
283 struct nvme_command c = { };
286 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
287 c.identify.opcode = nvme_admin_identify,
288 c.identify.nsid = cpu_to_le32(nsid),
290 *id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL);
294 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
295 sizeof(struct nvme_id_ns));
301 int nvme_get_features(struct nvme_ctrl *dev, unsigned fid, unsigned nsid,
302 dma_addr_t dma_addr, u32 *result)
304 struct nvme_command c;
306 memset(&c, 0, sizeof(c));
307 c.features.opcode = nvme_admin_get_features;
308 c.features.nsid = cpu_to_le32(nsid);
309 c.features.prp1 = cpu_to_le64(dma_addr);
310 c.features.fid = cpu_to_le32(fid);
312 return __nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0, result, 0);
315 int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
316 dma_addr_t dma_addr, u32 *result)
318 struct nvme_command c;
320 memset(&c, 0, sizeof(c));
321 c.features.opcode = nvme_admin_set_features;
322 c.features.prp1 = cpu_to_le64(dma_addr);
323 c.features.fid = cpu_to_le32(fid);
324 c.features.dword11 = cpu_to_le32(dword11);
326 return __nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0, result, 0);
329 int nvme_get_log_page(struct nvme_ctrl *dev, struct nvme_smart_log **log)
331 struct nvme_command c = { };
334 c.common.opcode = nvme_admin_get_log_page,
335 c.common.nsid = cpu_to_le32(0xFFFFFFFF),
336 c.common.cdw10[0] = cpu_to_le32(
337 (((sizeof(struct nvme_smart_log) / 4) - 1) << 16) |
340 *log = kmalloc(sizeof(struct nvme_smart_log), GFP_KERNEL);
344 error = nvme_submit_sync_cmd(dev->admin_q, &c, *log,
345 sizeof(struct nvme_smart_log));
351 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
353 u32 q_count = (*count - 1) | ((*count - 1) << 16);
355 int status, nr_io_queues;
357 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, 0,
362 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
363 *count = min(*count, nr_io_queues);
367 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
369 struct nvme_user_io io;
370 struct nvme_command c;
371 unsigned length, meta_len;
372 void __user *metadata;
374 if (copy_from_user(&io, uio, sizeof(io)))
380 case nvme_cmd_compare:
386 length = (io.nblocks + 1) << ns->lba_shift;
387 meta_len = (io.nblocks + 1) * ns->ms;
388 metadata = (void __user *)(uintptr_t)io.metadata;
393 } else if (meta_len) {
394 if ((io.metadata & 3) || !io.metadata)
398 memset(&c, 0, sizeof(c));
399 c.rw.opcode = io.opcode;
400 c.rw.flags = io.flags;
401 c.rw.nsid = cpu_to_le32(ns->ns_id);
402 c.rw.slba = cpu_to_le64(io.slba);
403 c.rw.length = cpu_to_le16(io.nblocks);
404 c.rw.control = cpu_to_le16(io.control);
405 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
406 c.rw.reftag = cpu_to_le32(io.reftag);
407 c.rw.apptag = cpu_to_le16(io.apptag);
408 c.rw.appmask = cpu_to_le16(io.appmask);
410 return __nvme_submit_user_cmd(ns->queue, &c,
411 (void __user *)(uintptr_t)io.addr, length,
412 metadata, meta_len, io.slba, NULL, 0);
415 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
416 struct nvme_passthru_cmd __user *ucmd)
418 struct nvme_passthru_cmd cmd;
419 struct nvme_command c;
420 unsigned timeout = 0;
423 if (!capable(CAP_SYS_ADMIN))
425 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
428 memset(&c, 0, sizeof(c));
429 c.common.opcode = cmd.opcode;
430 c.common.flags = cmd.flags;
431 c.common.nsid = cpu_to_le32(cmd.nsid);
432 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
433 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
434 c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
435 c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
436 c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
437 c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
438 c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
439 c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
442 timeout = msecs_to_jiffies(cmd.timeout_ms);
444 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
445 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
446 &cmd.result, timeout);
448 if (put_user(cmd.result, &ucmd->result))
455 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
456 unsigned int cmd, unsigned long arg)
458 struct nvme_ns *ns = bdev->bd_disk->private_data;
462 force_successful_syscall_return();
464 case NVME_IOCTL_ADMIN_CMD:
465 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
466 case NVME_IOCTL_IO_CMD:
467 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
468 case NVME_IOCTL_SUBMIT_IO:
469 return nvme_submit_io(ns, (void __user *)arg);
470 #ifdef CONFIG_BLK_DEV_NVME_SCSI
471 case SG_GET_VERSION_NUM:
472 return nvme_sg_get_version_num((void __user *)arg);
474 return nvme_sg_io(ns, (void __user *)arg);
482 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
483 unsigned int cmd, unsigned long arg)
489 return nvme_ioctl(bdev, mode, cmd, arg);
492 #define nvme_compat_ioctl NULL
495 static int nvme_open(struct block_device *bdev, fmode_t mode)
497 return nvme_get_ns_from_disk(bdev->bd_disk) ? 0 : -ENXIO;
500 static void nvme_release(struct gendisk *disk, fmode_t mode)
502 nvme_put_ns(disk->private_data);
505 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
507 /* some standard values */
509 geo->sectors = 1 << 5;
510 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
514 #ifdef CONFIG_BLK_DEV_INTEGRITY
515 static void nvme_init_integrity(struct nvme_ns *ns)
517 struct blk_integrity integrity;
519 switch (ns->pi_type) {
520 case NVME_NS_DPS_PI_TYPE3:
521 integrity.profile = &t10_pi_type3_crc;
523 case NVME_NS_DPS_PI_TYPE1:
524 case NVME_NS_DPS_PI_TYPE2:
525 integrity.profile = &t10_pi_type1_crc;
528 integrity.profile = NULL;
531 integrity.tuple_size = ns->ms;
532 blk_integrity_register(ns->disk, &integrity);
533 blk_queue_max_integrity_segments(ns->queue, 1);
536 static void nvme_init_integrity(struct nvme_ns *ns)
539 #endif /* CONFIG_BLK_DEV_INTEGRITY */
541 static void nvme_config_discard(struct nvme_ns *ns)
543 u32 logical_block_size = queue_logical_block_size(ns->queue);
544 ns->queue->limits.discard_zeroes_data = 0;
545 ns->queue->limits.discard_alignment = logical_block_size;
546 ns->queue->limits.discard_granularity = logical_block_size;
547 blk_queue_max_discard_sectors(ns->queue, 0xffffffff);
548 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue);
551 static int nvme_revalidate_disk(struct gendisk *disk)
553 struct nvme_ns *ns = disk->private_data;
554 struct nvme_id_ns *id;
559 if (nvme_identify_ns(ns->ctrl, ns->ns_id, &id)) {
560 dev_warn(ns->ctrl->dev, "%s: Identify failure nvme%dn%d\n",
561 __func__, ns->ctrl->instance, ns->ns_id);
569 if (nvme_nvm_ns_supported(ns, id) && ns->type != NVME_NS_LIGHTNVM) {
570 if (nvme_nvm_register(ns->queue, disk->disk_name)) {
571 dev_warn(ns->ctrl->dev,
572 "%s: LightNVM init failure\n", __func__);
576 ns->type = NVME_NS_LIGHTNVM;
579 if (ns->ctrl->vs >= NVME_VS(1, 1))
580 memcpy(ns->eui, id->eui64, sizeof(ns->eui));
581 if (ns->ctrl->vs >= NVME_VS(1, 2))
582 memcpy(ns->uuid, id->nguid, sizeof(ns->uuid));
585 lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
586 ns->lba_shift = id->lbaf[lbaf].ds;
587 ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
588 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
591 * If identify namespace failed, use default 512 byte block size so
592 * block layer can use before failing read/write for 0 capacity.
594 if (ns->lba_shift == 0)
596 bs = 1 << ns->lba_shift;
597 /* XXX: PI implementation requires metadata equal t10 pi tuple size */
598 pi_type = ns->ms == sizeof(struct t10_pi_tuple) ?
599 id->dps & NVME_NS_DPS_PI_MASK : 0;
601 blk_mq_freeze_queue(disk->queue);
602 if (blk_get_integrity(disk) && (ns->pi_type != pi_type ||
604 bs != queue_logical_block_size(disk->queue) ||
605 (ns->ms && ns->ext)))
606 blk_integrity_unregister(disk);
608 ns->pi_type = pi_type;
609 blk_queue_logical_block_size(ns->queue, bs);
611 if (ns->ms && !blk_get_integrity(disk) && !ns->ext)
612 nvme_init_integrity(ns);
613 if (ns->ms && !(ns->ms == 8 && ns->pi_type) && !blk_get_integrity(disk))
614 set_capacity(disk, 0);
616 set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
618 if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
619 nvme_config_discard(ns);
620 blk_mq_unfreeze_queue(disk->queue);
626 static char nvme_pr_type(enum pr_type type)
629 case PR_WRITE_EXCLUSIVE:
631 case PR_EXCLUSIVE_ACCESS:
633 case PR_WRITE_EXCLUSIVE_REG_ONLY:
635 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
637 case PR_WRITE_EXCLUSIVE_ALL_REGS:
639 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
646 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
647 u64 key, u64 sa_key, u8 op)
649 struct nvme_ns *ns = bdev->bd_disk->private_data;
650 struct nvme_command c;
651 u8 data[16] = { 0, };
653 put_unaligned_le64(key, &data[0]);
654 put_unaligned_le64(sa_key, &data[8]);
656 memset(&c, 0, sizeof(c));
657 c.common.opcode = op;
658 c.common.nsid = cpu_to_le32(ns->ns_id);
659 c.common.cdw10[0] = cpu_to_le32(cdw10);
661 return nvme_submit_sync_cmd(ns->queue, &c, data, 16);
664 static int nvme_pr_register(struct block_device *bdev, u64 old,
665 u64 new, unsigned flags)
669 if (flags & ~PR_FL_IGNORE_KEY)
673 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
674 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
675 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
678 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
679 enum pr_type type, unsigned flags)
683 if (flags & ~PR_FL_IGNORE_KEY)
686 cdw10 = nvme_pr_type(type) << 8;
687 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
688 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
691 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
692 enum pr_type type, bool abort)
694 u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
695 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
698 static int nvme_pr_clear(struct block_device *bdev, u64 key)
700 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
701 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
704 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
706 u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
707 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
710 static const struct pr_ops nvme_pr_ops = {
711 .pr_register = nvme_pr_register,
712 .pr_reserve = nvme_pr_reserve,
713 .pr_release = nvme_pr_release,
714 .pr_preempt = nvme_pr_preempt,
715 .pr_clear = nvme_pr_clear,
718 static const struct block_device_operations nvme_fops = {
719 .owner = THIS_MODULE,
721 .compat_ioctl = nvme_compat_ioctl,
723 .release = nvme_release,
724 .getgeo = nvme_getgeo,
725 .revalidate_disk= nvme_revalidate_disk,
726 .pr_ops = &nvme_pr_ops,
729 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
731 unsigned long timeout =
732 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
733 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
736 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
737 if ((csts & NVME_CSTS_RDY) == bit)
741 if (fatal_signal_pending(current))
743 if (time_after(jiffies, timeout)) {
745 "Device not ready; aborting %s\n", enabled ?
746 "initialisation" : "reset");
755 * If the device has been passed off to us in an enabled state, just clear
756 * the enabled bit. The spec says we should set the 'shutdown notification
757 * bits', but doing so may cause the device to complete commands to the
758 * admin queue ... and we don't know what memory that might be pointing at!
760 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
764 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
765 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
767 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
770 return nvme_wait_ready(ctrl, cap, false);
773 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
776 * Default to a 4K page size, with the intention to update this
777 * path in the future to accomodate architectures with differing
778 * kernel and IO page sizes.
780 unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
783 if (page_shift < dev_page_min) {
785 "Minimum device page size %u too large for host (%u)\n",
786 1 << dev_page_min, 1 << page_shift);
790 ctrl->page_size = 1 << page_shift;
792 ctrl->ctrl_config = NVME_CC_CSS_NVM;
793 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
794 ctrl->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE;
795 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
796 ctrl->ctrl_config |= NVME_CC_ENABLE;
798 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
801 return nvme_wait_ready(ctrl, cap, true);
804 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
806 unsigned long timeout = SHUTDOWN_TIMEOUT + jiffies;
810 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
811 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
813 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
817 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
818 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
822 if (fatal_signal_pending(current))
824 if (time_after(jiffies, timeout)) {
826 "Device shutdown incomplete; abort shutdown\n");
835 * Initialize the cached copies of the Identify data and various controller
836 * register in our nvme_ctrl structure. This should be called as soon as
837 * the admin queue is fully up and running.
839 int nvme_init_identify(struct nvme_ctrl *ctrl)
841 struct nvme_id_ctrl *id;
845 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
847 dev_err(ctrl->dev, "Reading VS failed (%d)\n", ret);
851 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
853 dev_err(ctrl->dev, "Reading CAP failed (%d)\n", ret);
856 page_shift = NVME_CAP_MPSMIN(cap) + 12;
858 if (ctrl->vs >= NVME_VS(1, 1))
859 ctrl->subsystem = NVME_CAP_NSSRC(cap);
861 ret = nvme_identify_ctrl(ctrl, &id);
863 dev_err(ctrl->dev, "Identify Controller failed (%d)\n", ret);
867 ctrl->oncs = le16_to_cpup(&id->oncs);
868 atomic_set(&ctrl->abort_limit, id->acl + 1);
870 memcpy(ctrl->serial, id->sn, sizeof(id->sn));
871 memcpy(ctrl->model, id->mn, sizeof(id->mn));
872 memcpy(ctrl->firmware_rev, id->fr, sizeof(id->fr));
874 ctrl->max_hw_sectors = 1 << (id->mdts + page_shift - 9);
876 ctrl->max_hw_sectors = UINT_MAX;
878 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) && id->vs[3]) {
879 unsigned int max_hw_sectors;
881 ctrl->stripe_size = 1 << (id->vs[3] + page_shift);
882 max_hw_sectors = ctrl->stripe_size >> (page_shift - 9);
883 if (ctrl->max_hw_sectors) {
884 ctrl->max_hw_sectors = min(max_hw_sectors,
885 ctrl->max_hw_sectors);
887 ctrl->max_hw_sectors = max_hw_sectors;
895 static int nvme_dev_open(struct inode *inode, struct file *file)
897 struct nvme_ctrl *ctrl;
898 int instance = iminor(inode);
901 spin_lock(&dev_list_lock);
902 list_for_each_entry(ctrl, &nvme_ctrl_list, node) {
903 if (ctrl->instance != instance)
906 if (!ctrl->admin_q) {
910 if (!kref_get_unless_zero(&ctrl->kref))
912 file->private_data = ctrl;
916 spin_unlock(&dev_list_lock);
921 static int nvme_dev_release(struct inode *inode, struct file *file)
923 nvme_put_ctrl(file->private_data);
927 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
932 mutex_lock(&ctrl->namespaces_mutex);
933 if (list_empty(&ctrl->namespaces)) {
938 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
939 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
941 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
947 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
949 mutex_unlock(&ctrl->namespaces_mutex);
951 ret = nvme_user_cmd(ctrl, ns, argp);
956 mutex_unlock(&ctrl->namespaces_mutex);
960 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
963 struct nvme_ctrl *ctrl = file->private_data;
964 void __user *argp = (void __user *)arg;
967 case NVME_IOCTL_ADMIN_CMD:
968 return nvme_user_cmd(ctrl, NULL, argp);
969 case NVME_IOCTL_IO_CMD:
970 return nvme_dev_user_cmd(ctrl, argp);
971 case NVME_IOCTL_RESET:
972 dev_warn(ctrl->dev, "resetting controller\n");
973 return ctrl->ops->reset_ctrl(ctrl);
974 case NVME_IOCTL_SUBSYS_RESET:
975 return nvme_reset_subsystem(ctrl);
981 static const struct file_operations nvme_dev_fops = {
982 .owner = THIS_MODULE,
983 .open = nvme_dev_open,
984 .release = nvme_dev_release,
985 .unlocked_ioctl = nvme_dev_ioctl,
986 .compat_ioctl = nvme_dev_ioctl,
989 static ssize_t nvme_sysfs_reset(struct device *dev,
990 struct device_attribute *attr, const char *buf,
993 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
996 ret = ctrl->ops->reset_ctrl(ctrl);
1001 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
1003 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
1006 struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1007 return sprintf(buf, "%pU\n", ns->uuid);
1009 static DEVICE_ATTR(uuid, S_IRUGO, uuid_show, NULL);
1011 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
1014 struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1015 return sprintf(buf, "%8phd\n", ns->eui);
1017 static DEVICE_ATTR(eui, S_IRUGO, eui_show, NULL);
1019 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
1022 struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1023 return sprintf(buf, "%d\n", ns->ns_id);
1025 static DEVICE_ATTR(nsid, S_IRUGO, nsid_show, NULL);
1027 static struct attribute *nvme_ns_attrs[] = {
1028 &dev_attr_uuid.attr,
1030 &dev_attr_nsid.attr,
1034 static umode_t nvme_attrs_are_visible(struct kobject *kobj,
1035 struct attribute *a, int n)
1037 struct device *dev = container_of(kobj, struct device, kobj);
1038 struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1040 if (a == &dev_attr_uuid.attr) {
1041 if (!memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1044 if (a == &dev_attr_eui.attr) {
1045 if (!memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1051 static const struct attribute_group nvme_ns_attr_group = {
1052 .attrs = nvme_ns_attrs,
1053 .is_visible = nvme_attrs_are_visible,
1056 #define nvme_show_function(field) \
1057 static ssize_t field##_show(struct device *dev, \
1058 struct device_attribute *attr, char *buf) \
1060 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
1061 return sprintf(buf, "%.*s\n", (int)sizeof(ctrl->field), ctrl->field); \
1063 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1065 nvme_show_function(model);
1066 nvme_show_function(serial);
1067 nvme_show_function(firmware_rev);
1069 static struct attribute *nvme_dev_attrs[] = {
1070 &dev_attr_reset_controller.attr,
1071 &dev_attr_model.attr,
1072 &dev_attr_serial.attr,
1073 &dev_attr_firmware_rev.attr,
1077 static struct attribute_group nvme_dev_attrs_group = {
1078 .attrs = nvme_dev_attrs,
1081 static const struct attribute_group *nvme_dev_attr_groups[] = {
1082 &nvme_dev_attrs_group,
1086 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
1088 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
1089 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
1091 return nsa->ns_id - nsb->ns_id;
1094 static struct nvme_ns *nvme_find_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1098 lockdep_assert_held(&ctrl->namespaces_mutex);
1100 list_for_each_entry(ns, &ctrl->namespaces, list) {
1101 if (ns->ns_id == nsid)
1103 if (ns->ns_id > nsid)
1109 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1112 struct gendisk *disk;
1113 int node = dev_to_node(ctrl->dev);
1115 lockdep_assert_held(&ctrl->namespaces_mutex);
1117 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
1121 ns->queue = blk_mq_init_queue(ctrl->tagset);
1122 if (IS_ERR(ns->queue))
1124 queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES, ns->queue);
1125 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
1126 ns->queue->queuedata = ns;
1129 disk = alloc_disk_node(0, node);
1131 goto out_free_queue;
1133 kref_init(&ns->kref);
1136 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
1138 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
1139 if (ctrl->max_hw_sectors) {
1140 blk_queue_max_hw_sectors(ns->queue, ctrl->max_hw_sectors);
1141 blk_queue_max_segments(ns->queue,
1142 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1);
1144 if (ctrl->stripe_size)
1145 blk_queue_chunk_sectors(ns->queue, ctrl->stripe_size >> 9);
1146 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1147 blk_queue_flush(ns->queue, REQ_FLUSH | REQ_FUA);
1148 blk_queue_virt_boundary(ns->queue, ctrl->page_size - 1);
1150 disk->major = nvme_major;
1151 disk->first_minor = 0;
1152 disk->fops = &nvme_fops;
1153 disk->private_data = ns;
1154 disk->queue = ns->queue;
1155 disk->driverfs_dev = ctrl->device;
1156 disk->flags = GENHD_FL_EXT_DEVT;
1157 sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance, nsid);
1159 if (nvme_revalidate_disk(ns->disk))
1162 list_add_tail(&ns->list, &ctrl->namespaces);
1163 kref_get(&ctrl->kref);
1164 if (ns->type == NVME_NS_LIGHTNVM)
1168 if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
1169 &nvme_ns_attr_group))
1170 pr_warn("%s: failed to create sysfs group for identification\n",
1171 ns->disk->disk_name);
1176 blk_cleanup_queue(ns->queue);
1181 static void nvme_ns_remove(struct nvme_ns *ns)
1183 bool kill = nvme_io_incapable(ns->ctrl) &&
1184 !blk_queue_dying(ns->queue);
1186 lockdep_assert_held(&ns->ctrl->namespaces_mutex);
1189 blk_set_queue_dying(ns->queue);
1192 * The controller was shutdown first if we got here through
1193 * device removal. The shutdown may requeue outstanding
1194 * requests. These need to be aborted immediately so
1195 * del_gendisk doesn't block indefinitely for their completion.
1197 blk_mq_abort_requeue_list(ns->queue);
1199 if (ns->disk->flags & GENHD_FL_UP) {
1200 if (blk_get_integrity(ns->disk))
1201 blk_integrity_unregister(ns->disk);
1202 sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
1203 &nvme_ns_attr_group);
1204 del_gendisk(ns->disk);
1206 if (kill || !blk_queue_dying(ns->queue)) {
1207 blk_mq_abort_requeue_list(ns->queue);
1208 blk_cleanup_queue(ns->queue);
1210 list_del_init(&ns->list);
1214 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1218 ns = nvme_find_ns(ctrl, nsid);
1220 if (revalidate_disk(ns->disk))
1223 nvme_alloc_ns(ctrl, nsid);
1226 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
1230 unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
1233 ns_list = kzalloc(0x1000, GFP_KERNEL);
1237 for (i = 0; i < num_lists; i++) {
1238 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
1242 for (j = 0; j < min(nn, 1024U); j++) {
1243 nsid = le32_to_cpu(ns_list[j]);
1247 nvme_validate_ns(ctrl, nsid);
1249 while (++prev < nsid) {
1250 ns = nvme_find_ns(ctrl, prev);
1262 static void __nvme_scan_namespaces(struct nvme_ctrl *ctrl, unsigned nn)
1264 struct nvme_ns *ns, *next;
1267 lockdep_assert_held(&ctrl->namespaces_mutex);
1269 for (i = 1; i <= nn; i++)
1270 nvme_validate_ns(ctrl, i);
1272 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
1278 void nvme_scan_namespaces(struct nvme_ctrl *ctrl)
1280 struct nvme_id_ctrl *id;
1283 if (nvme_identify_ctrl(ctrl, &id))
1286 mutex_lock(&ctrl->namespaces_mutex);
1287 nn = le32_to_cpu(id->nn);
1288 if (ctrl->vs >= NVME_VS(1, 1) &&
1289 !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
1290 if (!nvme_scan_ns_list(ctrl, nn))
1293 __nvme_scan_namespaces(ctrl, le32_to_cpup(&id->nn));
1295 list_sort(NULL, &ctrl->namespaces, ns_cmp);
1296 mutex_unlock(&ctrl->namespaces_mutex);
1300 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
1302 struct nvme_ns *ns, *next;
1304 mutex_lock(&ctrl->namespaces_mutex);
1305 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
1307 mutex_unlock(&ctrl->namespaces_mutex);
1310 static DEFINE_IDA(nvme_instance_ida);
1312 static int nvme_set_instance(struct nvme_ctrl *ctrl)
1314 int instance, error;
1317 if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL))
1320 spin_lock(&dev_list_lock);
1321 error = ida_get_new(&nvme_instance_ida, &instance);
1322 spin_unlock(&dev_list_lock);
1323 } while (error == -EAGAIN);
1328 ctrl->instance = instance;
1332 static void nvme_release_instance(struct nvme_ctrl *ctrl)
1334 spin_lock(&dev_list_lock);
1335 ida_remove(&nvme_instance_ida, ctrl->instance);
1336 spin_unlock(&dev_list_lock);
1339 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
1341 device_destroy(nvme_class, MKDEV(nvme_char_major, ctrl->instance));
1343 spin_lock(&dev_list_lock);
1344 list_del(&ctrl->node);
1345 spin_unlock(&dev_list_lock);
1348 static void nvme_free_ctrl(struct kref *kref)
1350 struct nvme_ctrl *ctrl = container_of(kref, struct nvme_ctrl, kref);
1352 put_device(ctrl->device);
1353 nvme_release_instance(ctrl);
1355 ctrl->ops->free_ctrl(ctrl);
1358 void nvme_put_ctrl(struct nvme_ctrl *ctrl)
1360 kref_put(&ctrl->kref, nvme_free_ctrl);
1364 * Initialize a NVMe controller structures. This needs to be called during
1365 * earliest initialization so that we have the initialized structured around
1368 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
1369 const struct nvme_ctrl_ops *ops, unsigned long quirks)
1373 INIT_LIST_HEAD(&ctrl->namespaces);
1374 mutex_init(&ctrl->namespaces_mutex);
1375 kref_init(&ctrl->kref);
1378 ctrl->quirks = quirks;
1380 ret = nvme_set_instance(ctrl);
1384 ctrl->device = device_create_with_groups(nvme_class, ctrl->dev,
1385 MKDEV(nvme_char_major, ctrl->instance),
1386 dev, nvme_dev_attr_groups,
1387 "nvme%d", ctrl->instance);
1388 if (IS_ERR(ctrl->device)) {
1389 ret = PTR_ERR(ctrl->device);
1390 goto out_release_instance;
1392 get_device(ctrl->device);
1393 dev_set_drvdata(ctrl->device, ctrl);
1395 spin_lock(&dev_list_lock);
1396 list_add_tail(&ctrl->node, &nvme_ctrl_list);
1397 spin_unlock(&dev_list_lock);
1400 out_release_instance:
1401 nvme_release_instance(ctrl);
1406 void nvme_stop_queues(struct nvme_ctrl *ctrl)
1410 mutex_lock(&ctrl->namespaces_mutex);
1411 list_for_each_entry(ns, &ctrl->namespaces, list) {
1412 spin_lock_irq(ns->queue->queue_lock);
1413 queue_flag_set(QUEUE_FLAG_STOPPED, ns->queue);
1414 spin_unlock_irq(ns->queue->queue_lock);
1416 blk_mq_cancel_requeue_work(ns->queue);
1417 blk_mq_stop_hw_queues(ns->queue);
1419 mutex_unlock(&ctrl->namespaces_mutex);
1422 void nvme_start_queues(struct nvme_ctrl *ctrl)
1426 mutex_lock(&ctrl->namespaces_mutex);
1427 list_for_each_entry(ns, &ctrl->namespaces, list) {
1428 queue_flag_clear_unlocked(QUEUE_FLAG_STOPPED, ns->queue);
1429 blk_mq_start_stopped_hw_queues(ns->queue, true);
1430 blk_mq_kick_requeue_list(ns->queue);
1432 mutex_unlock(&ctrl->namespaces_mutex);
1435 int __init nvme_core_init(void)
1439 result = register_blkdev(nvme_major, "nvme");
1442 else if (result > 0)
1443 nvme_major = result;
1445 result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme",
1448 goto unregister_blkdev;
1449 else if (result > 0)
1450 nvme_char_major = result;
1452 nvme_class = class_create(THIS_MODULE, "nvme");
1453 if (IS_ERR(nvme_class)) {
1454 result = PTR_ERR(nvme_class);
1455 goto unregister_chrdev;
1461 __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
1463 unregister_blkdev(nvme_major, "nvme");
1467 void nvme_core_exit(void)
1469 unregister_blkdev(nvme_major, "nvme");
1470 class_destroy(nvme_class);
1471 __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");