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 static void nvme_free_ns(struct kref *kref)
63 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
65 if (ns->type == NVME_NS_LIGHTNVM)
66 nvme_nvm_unregister(ns->queue, ns->disk->disk_name);
68 spin_lock(&dev_list_lock);
69 ns->disk->private_data = NULL;
70 spin_unlock(&dev_list_lock);
72 nvme_put_ctrl(ns->ctrl);
77 static void nvme_put_ns(struct nvme_ns *ns)
79 kref_put(&ns->kref, nvme_free_ns);
82 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk)
86 spin_lock(&dev_list_lock);
87 ns = disk->private_data;
89 if (!kref_get_unless_zero(&ns->kref))
91 if (!try_module_get(ns->ctrl->ops->module))
94 spin_unlock(&dev_list_lock);
99 kref_put(&ns->kref, nvme_free_ns);
101 spin_unlock(&dev_list_lock);
105 void nvme_requeue_req(struct request *req)
109 blk_mq_requeue_request(req);
110 spin_lock_irqsave(req->q->queue_lock, flags);
111 if (!blk_queue_stopped(req->q))
112 blk_mq_kick_requeue_list(req->q);
113 spin_unlock_irqrestore(req->q->queue_lock, flags);
115 EXPORT_SYMBOL_GPL(nvme_requeue_req);
117 struct request *nvme_alloc_request(struct request_queue *q,
118 struct nvme_command *cmd, unsigned int flags)
120 bool write = cmd->common.opcode & 1;
123 req = blk_mq_alloc_request(q, write, flags);
127 req->cmd_type = REQ_TYPE_DRV_PRIV;
128 req->cmd_flags |= REQ_FAILFAST_DRIVER;
130 req->__sector = (sector_t) -1;
131 req->bio = req->biotail = NULL;
133 req->cmd = (unsigned char *)cmd;
134 req->cmd_len = sizeof(struct nvme_command);
135 req->special = (void *)0;
139 EXPORT_SYMBOL_GPL(nvme_alloc_request);
142 * Returns 0 on success. If the result is negative, it's a Linux error code;
143 * if the result is positive, it's an NVM Express status code
145 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
146 void *buffer, unsigned bufflen, u32 *result, unsigned timeout)
151 req = nvme_alloc_request(q, cmd, 0);
155 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
157 if (buffer && bufflen) {
158 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
163 blk_execute_rq(req->q, NULL, req, 0);
165 *result = (u32)(uintptr_t)req->special;
168 blk_mq_free_request(req);
172 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
173 void *buffer, unsigned bufflen)
175 return __nvme_submit_sync_cmd(q, cmd, buffer, bufflen, NULL, 0);
177 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
179 int __nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
180 void __user *ubuffer, unsigned bufflen,
181 void __user *meta_buffer, unsigned meta_len, u32 meta_seed,
182 u32 *result, unsigned timeout)
184 bool write = cmd->common.opcode & 1;
185 struct nvme_ns *ns = q->queuedata;
186 struct gendisk *disk = ns ? ns->disk : NULL;
188 struct bio *bio = NULL;
192 req = nvme_alloc_request(q, cmd, 0);
196 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
198 if (ubuffer && bufflen) {
199 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
207 bio->bi_bdev = bdget_disk(disk, 0);
214 struct bio_integrity_payload *bip;
216 meta = kmalloc(meta_len, GFP_KERNEL);
223 if (copy_from_user(meta, meta_buffer,
230 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
236 bip->bip_iter.bi_size = meta_len;
237 bip->bip_iter.bi_sector = meta_seed;
239 ret = bio_integrity_add_page(bio, virt_to_page(meta),
240 meta_len, offset_in_page(meta));
241 if (ret != meta_len) {
248 blk_execute_rq(req->q, disk, req, 0);
251 *result = (u32)(uintptr_t)req->special;
252 if (meta && !ret && !write) {
253 if (copy_to_user(meta_buffer, meta, meta_len))
260 if (disk && bio->bi_bdev)
262 blk_rq_unmap_user(bio);
265 blk_mq_free_request(req);
269 int nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
270 void __user *ubuffer, unsigned bufflen, u32 *result,
273 return __nvme_submit_user_cmd(q, cmd, ubuffer, bufflen, NULL, 0, 0,
277 int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
279 struct nvme_command c = { };
282 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
283 c.identify.opcode = nvme_admin_identify;
284 c.identify.cns = cpu_to_le32(1);
286 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
290 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
291 sizeof(struct nvme_id_ctrl));
297 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
299 struct nvme_command c = { };
301 c.identify.opcode = nvme_admin_identify;
302 c.identify.cns = cpu_to_le32(2);
303 c.identify.nsid = cpu_to_le32(nsid);
304 return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000);
307 int nvme_identify_ns(struct nvme_ctrl *dev, unsigned nsid,
308 struct nvme_id_ns **id)
310 struct nvme_command c = { };
313 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
314 c.identify.opcode = nvme_admin_identify,
315 c.identify.nsid = cpu_to_le32(nsid),
317 *id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL);
321 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
322 sizeof(struct nvme_id_ns));
328 int nvme_get_features(struct nvme_ctrl *dev, unsigned fid, unsigned nsid,
329 dma_addr_t dma_addr, u32 *result)
331 struct nvme_command c;
333 memset(&c, 0, sizeof(c));
334 c.features.opcode = nvme_admin_get_features;
335 c.features.nsid = cpu_to_le32(nsid);
336 c.features.prp1 = cpu_to_le64(dma_addr);
337 c.features.fid = cpu_to_le32(fid);
339 return __nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0, result, 0);
342 int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
343 dma_addr_t dma_addr, u32 *result)
345 struct nvme_command c;
347 memset(&c, 0, sizeof(c));
348 c.features.opcode = nvme_admin_set_features;
349 c.features.prp1 = cpu_to_le64(dma_addr);
350 c.features.fid = cpu_to_le32(fid);
351 c.features.dword11 = cpu_to_le32(dword11);
353 return __nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0, result, 0);
356 int nvme_get_log_page(struct nvme_ctrl *dev, struct nvme_smart_log **log)
358 struct nvme_command c = { };
361 c.common.opcode = nvme_admin_get_log_page,
362 c.common.nsid = cpu_to_le32(0xFFFFFFFF),
363 c.common.cdw10[0] = cpu_to_le32(
364 (((sizeof(struct nvme_smart_log) / 4) - 1) << 16) |
367 *log = kmalloc(sizeof(struct nvme_smart_log), GFP_KERNEL);
371 error = nvme_submit_sync_cmd(dev->admin_q, &c, *log,
372 sizeof(struct nvme_smart_log));
378 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
380 u32 q_count = (*count - 1) | ((*count - 1) << 16);
382 int status, nr_io_queues;
384 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, 0,
389 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
390 *count = min(*count, nr_io_queues);
393 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
395 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
397 struct nvme_user_io io;
398 struct nvme_command c;
399 unsigned length, meta_len;
400 void __user *metadata;
402 if (copy_from_user(&io, uio, sizeof(io)))
408 case nvme_cmd_compare:
414 length = (io.nblocks + 1) << ns->lba_shift;
415 meta_len = (io.nblocks + 1) * ns->ms;
416 metadata = (void __user *)(uintptr_t)io.metadata;
421 } else if (meta_len) {
422 if ((io.metadata & 3) || !io.metadata)
426 memset(&c, 0, sizeof(c));
427 c.rw.opcode = io.opcode;
428 c.rw.flags = io.flags;
429 c.rw.nsid = cpu_to_le32(ns->ns_id);
430 c.rw.slba = cpu_to_le64(io.slba);
431 c.rw.length = cpu_to_le16(io.nblocks);
432 c.rw.control = cpu_to_le16(io.control);
433 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
434 c.rw.reftag = cpu_to_le32(io.reftag);
435 c.rw.apptag = cpu_to_le16(io.apptag);
436 c.rw.appmask = cpu_to_le16(io.appmask);
438 return __nvme_submit_user_cmd(ns->queue, &c,
439 (void __user *)(uintptr_t)io.addr, length,
440 metadata, meta_len, io.slba, NULL, 0);
443 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
444 struct nvme_passthru_cmd __user *ucmd)
446 struct nvme_passthru_cmd cmd;
447 struct nvme_command c;
448 unsigned timeout = 0;
451 if (!capable(CAP_SYS_ADMIN))
453 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
456 memset(&c, 0, sizeof(c));
457 c.common.opcode = cmd.opcode;
458 c.common.flags = cmd.flags;
459 c.common.nsid = cpu_to_le32(cmd.nsid);
460 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
461 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
462 c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
463 c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
464 c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
465 c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
466 c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
467 c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
470 timeout = msecs_to_jiffies(cmd.timeout_ms);
472 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
473 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
474 &cmd.result, timeout);
476 if (put_user(cmd.result, &ucmd->result))
483 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
484 unsigned int cmd, unsigned long arg)
486 struct nvme_ns *ns = bdev->bd_disk->private_data;
490 force_successful_syscall_return();
492 case NVME_IOCTL_ADMIN_CMD:
493 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
494 case NVME_IOCTL_IO_CMD:
495 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
496 case NVME_IOCTL_SUBMIT_IO:
497 return nvme_submit_io(ns, (void __user *)arg);
498 #ifdef CONFIG_BLK_DEV_NVME_SCSI
499 case SG_GET_VERSION_NUM:
500 return nvme_sg_get_version_num((void __user *)arg);
502 return nvme_sg_io(ns, (void __user *)arg);
510 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
511 unsigned int cmd, unsigned long arg)
517 return nvme_ioctl(bdev, mode, cmd, arg);
520 #define nvme_compat_ioctl NULL
523 static int nvme_open(struct block_device *bdev, fmode_t mode)
525 return nvme_get_ns_from_disk(bdev->bd_disk) ? 0 : -ENXIO;
528 static void nvme_release(struct gendisk *disk, fmode_t mode)
530 struct nvme_ns *ns = disk->private_data;
532 module_put(ns->ctrl->ops->module);
536 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
538 /* some standard values */
540 geo->sectors = 1 << 5;
541 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
545 #ifdef CONFIG_BLK_DEV_INTEGRITY
546 static void nvme_init_integrity(struct nvme_ns *ns)
548 struct blk_integrity integrity;
550 switch (ns->pi_type) {
551 case NVME_NS_DPS_PI_TYPE3:
552 integrity.profile = &t10_pi_type3_crc;
554 case NVME_NS_DPS_PI_TYPE1:
555 case NVME_NS_DPS_PI_TYPE2:
556 integrity.profile = &t10_pi_type1_crc;
559 integrity.profile = NULL;
562 integrity.tuple_size = ns->ms;
563 blk_integrity_register(ns->disk, &integrity);
564 blk_queue_max_integrity_segments(ns->queue, 1);
567 static void nvme_init_integrity(struct nvme_ns *ns)
570 #endif /* CONFIG_BLK_DEV_INTEGRITY */
572 static void nvme_config_discard(struct nvme_ns *ns)
574 u32 logical_block_size = queue_logical_block_size(ns->queue);
575 ns->queue->limits.discard_zeroes_data = 0;
576 ns->queue->limits.discard_alignment = logical_block_size;
577 ns->queue->limits.discard_granularity = logical_block_size;
578 blk_queue_max_discard_sectors(ns->queue, 0xffffffff);
579 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue);
582 static int nvme_revalidate_disk(struct gendisk *disk)
584 struct nvme_ns *ns = disk->private_data;
585 struct nvme_id_ns *id;
590 if (nvme_identify_ns(ns->ctrl, ns->ns_id, &id)) {
591 dev_warn(disk_to_dev(ns->disk), "%s: Identify failure\n",
600 if (nvme_nvm_ns_supported(ns, id) && ns->type != NVME_NS_LIGHTNVM) {
601 if (nvme_nvm_register(ns->queue, disk->disk_name)) {
602 dev_warn(disk_to_dev(ns->disk),
603 "%s: LightNVM init failure\n", __func__);
607 ns->type = NVME_NS_LIGHTNVM;
610 if (ns->ctrl->vs >= NVME_VS(1, 1))
611 memcpy(ns->eui, id->eui64, sizeof(ns->eui));
612 if (ns->ctrl->vs >= NVME_VS(1, 2))
613 memcpy(ns->uuid, id->nguid, sizeof(ns->uuid));
616 lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
617 ns->lba_shift = id->lbaf[lbaf].ds;
618 ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
619 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
622 * If identify namespace failed, use default 512 byte block size so
623 * block layer can use before failing read/write for 0 capacity.
625 if (ns->lba_shift == 0)
627 bs = 1 << ns->lba_shift;
628 /* XXX: PI implementation requires metadata equal t10 pi tuple size */
629 pi_type = ns->ms == sizeof(struct t10_pi_tuple) ?
630 id->dps & NVME_NS_DPS_PI_MASK : 0;
632 blk_mq_freeze_queue(disk->queue);
633 if (blk_get_integrity(disk) && (ns->pi_type != pi_type ||
635 bs != queue_logical_block_size(disk->queue) ||
636 (ns->ms && ns->ext)))
637 blk_integrity_unregister(disk);
639 ns->pi_type = pi_type;
640 blk_queue_logical_block_size(ns->queue, bs);
642 if (ns->ms && !blk_get_integrity(disk) && !ns->ext)
643 nvme_init_integrity(ns);
644 if (ns->ms && !(ns->ms == 8 && ns->pi_type) && !blk_get_integrity(disk))
645 set_capacity(disk, 0);
647 set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
649 if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
650 nvme_config_discard(ns);
651 blk_mq_unfreeze_queue(disk->queue);
657 static char nvme_pr_type(enum pr_type type)
660 case PR_WRITE_EXCLUSIVE:
662 case PR_EXCLUSIVE_ACCESS:
664 case PR_WRITE_EXCLUSIVE_REG_ONLY:
666 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
668 case PR_WRITE_EXCLUSIVE_ALL_REGS:
670 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
677 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
678 u64 key, u64 sa_key, u8 op)
680 struct nvme_ns *ns = bdev->bd_disk->private_data;
681 struct nvme_command c;
682 u8 data[16] = { 0, };
684 put_unaligned_le64(key, &data[0]);
685 put_unaligned_le64(sa_key, &data[8]);
687 memset(&c, 0, sizeof(c));
688 c.common.opcode = op;
689 c.common.nsid = cpu_to_le32(ns->ns_id);
690 c.common.cdw10[0] = cpu_to_le32(cdw10);
692 return nvme_submit_sync_cmd(ns->queue, &c, data, 16);
695 static int nvme_pr_register(struct block_device *bdev, u64 old,
696 u64 new, unsigned flags)
700 if (flags & ~PR_FL_IGNORE_KEY)
704 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
705 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
706 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
709 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
710 enum pr_type type, unsigned flags)
714 if (flags & ~PR_FL_IGNORE_KEY)
717 cdw10 = nvme_pr_type(type) << 8;
718 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
719 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
722 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
723 enum pr_type type, bool abort)
725 u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
726 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
729 static int nvme_pr_clear(struct block_device *bdev, u64 key)
731 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
732 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
735 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
737 u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
738 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
741 static const struct pr_ops nvme_pr_ops = {
742 .pr_register = nvme_pr_register,
743 .pr_reserve = nvme_pr_reserve,
744 .pr_release = nvme_pr_release,
745 .pr_preempt = nvme_pr_preempt,
746 .pr_clear = nvme_pr_clear,
749 static const struct block_device_operations nvme_fops = {
750 .owner = THIS_MODULE,
752 .compat_ioctl = nvme_compat_ioctl,
754 .release = nvme_release,
755 .getgeo = nvme_getgeo,
756 .revalidate_disk= nvme_revalidate_disk,
757 .pr_ops = &nvme_pr_ops,
760 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
762 unsigned long timeout =
763 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
764 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
767 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
768 if ((csts & NVME_CSTS_RDY) == bit)
772 if (fatal_signal_pending(current))
774 if (time_after(jiffies, timeout)) {
775 dev_err(ctrl->device,
776 "Device not ready; aborting %s\n", enabled ?
777 "initialisation" : "reset");
786 * If the device has been passed off to us in an enabled state, just clear
787 * the enabled bit. The spec says we should set the 'shutdown notification
788 * bits', but doing so may cause the device to complete commands to the
789 * admin queue ... and we don't know what memory that might be pointing at!
791 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
795 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
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, false);
803 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
805 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
808 * Default to a 4K page size, with the intention to update this
809 * path in the future to accomodate architectures with differing
810 * kernel and IO page sizes.
812 unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
815 if (page_shift < dev_page_min) {
816 dev_err(ctrl->device,
817 "Minimum device page size %u too large for host (%u)\n",
818 1 << dev_page_min, 1 << page_shift);
822 ctrl->page_size = 1 << page_shift;
824 ctrl->ctrl_config = NVME_CC_CSS_NVM;
825 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
826 ctrl->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE;
827 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
828 ctrl->ctrl_config |= NVME_CC_ENABLE;
830 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
833 return nvme_wait_ready(ctrl, cap, true);
835 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
837 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
839 unsigned long timeout = SHUTDOWN_TIMEOUT + jiffies;
843 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
844 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
846 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
850 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
851 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
855 if (fatal_signal_pending(current))
857 if (time_after(jiffies, timeout)) {
858 dev_err(ctrl->device,
859 "Device shutdown incomplete; abort shutdown\n");
866 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
869 * Initialize the cached copies of the Identify data and various controller
870 * register in our nvme_ctrl structure. This should be called as soon as
871 * the admin queue is fully up and running.
873 int nvme_init_identify(struct nvme_ctrl *ctrl)
875 struct nvme_id_ctrl *id;
879 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
881 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
885 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
887 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
890 page_shift = NVME_CAP_MPSMIN(cap) + 12;
892 if (ctrl->vs >= NVME_VS(1, 1))
893 ctrl->subsystem = NVME_CAP_NSSRC(cap);
895 ret = nvme_identify_ctrl(ctrl, &id);
897 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
901 ctrl->oncs = le16_to_cpup(&id->oncs);
902 atomic_set(&ctrl->abort_limit, id->acl + 1);
904 memcpy(ctrl->serial, id->sn, sizeof(id->sn));
905 memcpy(ctrl->model, id->mn, sizeof(id->mn));
906 memcpy(ctrl->firmware_rev, id->fr, sizeof(id->fr));
908 ctrl->max_hw_sectors = 1 << (id->mdts + page_shift - 9);
910 ctrl->max_hw_sectors = UINT_MAX;
912 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) && id->vs[3]) {
913 unsigned int max_hw_sectors;
915 ctrl->stripe_size = 1 << (id->vs[3] + page_shift);
916 max_hw_sectors = ctrl->stripe_size >> (page_shift - 9);
917 if (ctrl->max_hw_sectors) {
918 ctrl->max_hw_sectors = min(max_hw_sectors,
919 ctrl->max_hw_sectors);
921 ctrl->max_hw_sectors = max_hw_sectors;
928 EXPORT_SYMBOL_GPL(nvme_init_identify);
930 static int nvme_dev_open(struct inode *inode, struct file *file)
932 struct nvme_ctrl *ctrl;
933 int instance = iminor(inode);
936 spin_lock(&dev_list_lock);
937 list_for_each_entry(ctrl, &nvme_ctrl_list, node) {
938 if (ctrl->instance != instance)
941 if (!ctrl->admin_q) {
945 if (!kref_get_unless_zero(&ctrl->kref))
947 file->private_data = ctrl;
951 spin_unlock(&dev_list_lock);
956 static int nvme_dev_release(struct inode *inode, struct file *file)
958 nvme_put_ctrl(file->private_data);
962 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
967 mutex_lock(&ctrl->namespaces_mutex);
968 if (list_empty(&ctrl->namespaces)) {
973 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
974 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
975 dev_warn(ctrl->device,
976 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
981 dev_warn(ctrl->device,
982 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
984 mutex_unlock(&ctrl->namespaces_mutex);
986 ret = nvme_user_cmd(ctrl, ns, argp);
991 mutex_unlock(&ctrl->namespaces_mutex);
995 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
998 struct nvme_ctrl *ctrl = file->private_data;
999 void __user *argp = (void __user *)arg;
1002 case NVME_IOCTL_ADMIN_CMD:
1003 return nvme_user_cmd(ctrl, NULL, argp);
1004 case NVME_IOCTL_IO_CMD:
1005 return nvme_dev_user_cmd(ctrl, argp);
1006 case NVME_IOCTL_RESET:
1007 dev_warn(ctrl->device, "resetting controller\n");
1008 return ctrl->ops->reset_ctrl(ctrl);
1009 case NVME_IOCTL_SUBSYS_RESET:
1010 return nvme_reset_subsystem(ctrl);
1016 static const struct file_operations nvme_dev_fops = {
1017 .owner = THIS_MODULE,
1018 .open = nvme_dev_open,
1019 .release = nvme_dev_release,
1020 .unlocked_ioctl = nvme_dev_ioctl,
1021 .compat_ioctl = nvme_dev_ioctl,
1024 static ssize_t nvme_sysfs_reset(struct device *dev,
1025 struct device_attribute *attr, const char *buf,
1028 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1031 ret = ctrl->ops->reset_ctrl(ctrl);
1036 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
1038 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
1041 struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1042 return sprintf(buf, "%pU\n", ns->uuid);
1044 static DEVICE_ATTR(uuid, S_IRUGO, uuid_show, NULL);
1046 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
1049 struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1050 return sprintf(buf, "%8phd\n", ns->eui);
1052 static DEVICE_ATTR(eui, S_IRUGO, eui_show, NULL);
1054 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
1057 struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1058 return sprintf(buf, "%d\n", ns->ns_id);
1060 static DEVICE_ATTR(nsid, S_IRUGO, nsid_show, NULL);
1062 static struct attribute *nvme_ns_attrs[] = {
1063 &dev_attr_uuid.attr,
1065 &dev_attr_nsid.attr,
1069 static umode_t nvme_attrs_are_visible(struct kobject *kobj,
1070 struct attribute *a, int n)
1072 struct device *dev = container_of(kobj, struct device, kobj);
1073 struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1075 if (a == &dev_attr_uuid.attr) {
1076 if (!memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1079 if (a == &dev_attr_eui.attr) {
1080 if (!memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1086 static const struct attribute_group nvme_ns_attr_group = {
1087 .attrs = nvme_ns_attrs,
1088 .is_visible = nvme_attrs_are_visible,
1091 #define nvme_show_function(field) \
1092 static ssize_t field##_show(struct device *dev, \
1093 struct device_attribute *attr, char *buf) \
1095 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
1096 return sprintf(buf, "%.*s\n", (int)sizeof(ctrl->field), ctrl->field); \
1098 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1100 nvme_show_function(model);
1101 nvme_show_function(serial);
1102 nvme_show_function(firmware_rev);
1104 static struct attribute *nvme_dev_attrs[] = {
1105 &dev_attr_reset_controller.attr,
1106 &dev_attr_model.attr,
1107 &dev_attr_serial.attr,
1108 &dev_attr_firmware_rev.attr,
1112 static struct attribute_group nvme_dev_attrs_group = {
1113 .attrs = nvme_dev_attrs,
1116 static const struct attribute_group *nvme_dev_attr_groups[] = {
1117 &nvme_dev_attrs_group,
1121 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
1123 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
1124 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
1126 return nsa->ns_id - nsb->ns_id;
1129 static struct nvme_ns *nvme_find_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1133 lockdep_assert_held(&ctrl->namespaces_mutex);
1135 list_for_each_entry(ns, &ctrl->namespaces, list) {
1136 if (ns->ns_id == nsid)
1138 if (ns->ns_id > nsid)
1144 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1147 struct gendisk *disk;
1148 int node = dev_to_node(ctrl->dev);
1150 lockdep_assert_held(&ctrl->namespaces_mutex);
1152 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
1156 ns->queue = blk_mq_init_queue(ctrl->tagset);
1157 if (IS_ERR(ns->queue))
1159 queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES, ns->queue);
1160 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
1161 ns->queue->queuedata = ns;
1164 disk = alloc_disk_node(0, node);
1166 goto out_free_queue;
1168 kref_init(&ns->kref);
1171 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
1173 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
1174 if (ctrl->max_hw_sectors) {
1175 blk_queue_max_hw_sectors(ns->queue, ctrl->max_hw_sectors);
1176 blk_queue_max_segments(ns->queue,
1177 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1);
1179 if (ctrl->stripe_size)
1180 blk_queue_chunk_sectors(ns->queue, ctrl->stripe_size >> 9);
1181 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1182 blk_queue_flush(ns->queue, REQ_FLUSH | REQ_FUA);
1183 blk_queue_virt_boundary(ns->queue, ctrl->page_size - 1);
1185 disk->major = nvme_major;
1186 disk->first_minor = 0;
1187 disk->fops = &nvme_fops;
1188 disk->private_data = ns;
1189 disk->queue = ns->queue;
1190 disk->driverfs_dev = ctrl->device;
1191 disk->flags = GENHD_FL_EXT_DEVT;
1192 sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance, nsid);
1194 if (nvme_revalidate_disk(ns->disk))
1197 list_add_tail(&ns->list, &ctrl->namespaces);
1198 kref_get(&ctrl->kref);
1199 if (ns->type == NVME_NS_LIGHTNVM)
1203 if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
1204 &nvme_ns_attr_group))
1205 pr_warn("%s: failed to create sysfs group for identification\n",
1206 ns->disk->disk_name);
1211 blk_cleanup_queue(ns->queue);
1216 static void nvme_ns_remove(struct nvme_ns *ns)
1218 bool kill = nvme_io_incapable(ns->ctrl) &&
1219 !blk_queue_dying(ns->queue);
1221 lockdep_assert_held(&ns->ctrl->namespaces_mutex);
1224 blk_set_queue_dying(ns->queue);
1227 * The controller was shutdown first if we got here through
1228 * device removal. The shutdown may requeue outstanding
1229 * requests. These need to be aborted immediately so
1230 * del_gendisk doesn't block indefinitely for their completion.
1232 blk_mq_abort_requeue_list(ns->queue);
1234 if (ns->disk->flags & GENHD_FL_UP) {
1235 if (blk_get_integrity(ns->disk))
1236 blk_integrity_unregister(ns->disk);
1237 sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
1238 &nvme_ns_attr_group);
1239 del_gendisk(ns->disk);
1241 if (kill || !blk_queue_dying(ns->queue)) {
1242 blk_mq_abort_requeue_list(ns->queue);
1243 blk_cleanup_queue(ns->queue);
1245 list_del_init(&ns->list);
1249 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1253 ns = nvme_find_ns(ctrl, nsid);
1255 if (revalidate_disk(ns->disk))
1258 nvme_alloc_ns(ctrl, nsid);
1261 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
1265 unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
1268 ns_list = kzalloc(0x1000, GFP_KERNEL);
1272 for (i = 0; i < num_lists; i++) {
1273 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
1277 for (j = 0; j < min(nn, 1024U); j++) {
1278 nsid = le32_to_cpu(ns_list[j]);
1282 nvme_validate_ns(ctrl, nsid);
1284 while (++prev < nsid) {
1285 ns = nvme_find_ns(ctrl, prev);
1297 static void __nvme_scan_namespaces(struct nvme_ctrl *ctrl, unsigned nn)
1299 struct nvme_ns *ns, *next;
1302 lockdep_assert_held(&ctrl->namespaces_mutex);
1304 for (i = 1; i <= nn; i++)
1305 nvme_validate_ns(ctrl, i);
1307 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
1313 void nvme_scan_namespaces(struct nvme_ctrl *ctrl)
1315 struct nvme_id_ctrl *id;
1318 if (nvme_identify_ctrl(ctrl, &id))
1321 mutex_lock(&ctrl->namespaces_mutex);
1322 nn = le32_to_cpu(id->nn);
1323 if (ctrl->vs >= NVME_VS(1, 1) &&
1324 !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
1325 if (!nvme_scan_ns_list(ctrl, nn))
1328 __nvme_scan_namespaces(ctrl, le32_to_cpup(&id->nn));
1330 list_sort(NULL, &ctrl->namespaces, ns_cmp);
1331 mutex_unlock(&ctrl->namespaces_mutex);
1334 EXPORT_SYMBOL_GPL(nvme_scan_namespaces);
1336 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
1338 struct nvme_ns *ns, *next;
1340 mutex_lock(&ctrl->namespaces_mutex);
1341 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
1343 mutex_unlock(&ctrl->namespaces_mutex);
1345 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
1347 static DEFINE_IDA(nvme_instance_ida);
1349 static int nvme_set_instance(struct nvme_ctrl *ctrl)
1351 int instance, error;
1354 if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL))
1357 spin_lock(&dev_list_lock);
1358 error = ida_get_new(&nvme_instance_ida, &instance);
1359 spin_unlock(&dev_list_lock);
1360 } while (error == -EAGAIN);
1365 ctrl->instance = instance;
1369 static void nvme_release_instance(struct nvme_ctrl *ctrl)
1371 spin_lock(&dev_list_lock);
1372 ida_remove(&nvme_instance_ida, ctrl->instance);
1373 spin_unlock(&dev_list_lock);
1376 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
1378 device_destroy(nvme_class, MKDEV(nvme_char_major, ctrl->instance));
1380 spin_lock(&dev_list_lock);
1381 list_del(&ctrl->node);
1382 spin_unlock(&dev_list_lock);
1384 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
1386 static void nvme_free_ctrl(struct kref *kref)
1388 struct nvme_ctrl *ctrl = container_of(kref, struct nvme_ctrl, kref);
1390 put_device(ctrl->device);
1391 nvme_release_instance(ctrl);
1393 ctrl->ops->free_ctrl(ctrl);
1396 void nvme_put_ctrl(struct nvme_ctrl *ctrl)
1398 kref_put(&ctrl->kref, nvme_free_ctrl);
1400 EXPORT_SYMBOL_GPL(nvme_put_ctrl);
1403 * Initialize a NVMe controller structures. This needs to be called during
1404 * earliest initialization so that we have the initialized structured around
1407 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
1408 const struct nvme_ctrl_ops *ops, unsigned long quirks)
1412 INIT_LIST_HEAD(&ctrl->namespaces);
1413 mutex_init(&ctrl->namespaces_mutex);
1414 kref_init(&ctrl->kref);
1417 ctrl->quirks = quirks;
1419 ret = nvme_set_instance(ctrl);
1423 ctrl->device = device_create_with_groups(nvme_class, ctrl->dev,
1424 MKDEV(nvme_char_major, ctrl->instance),
1425 ctrl, nvme_dev_attr_groups,
1426 "nvme%d", ctrl->instance);
1427 if (IS_ERR(ctrl->device)) {
1428 ret = PTR_ERR(ctrl->device);
1429 goto out_release_instance;
1431 get_device(ctrl->device);
1433 spin_lock(&dev_list_lock);
1434 list_add_tail(&ctrl->node, &nvme_ctrl_list);
1435 spin_unlock(&dev_list_lock);
1438 out_release_instance:
1439 nvme_release_instance(ctrl);
1443 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
1445 void nvme_stop_queues(struct nvme_ctrl *ctrl)
1449 mutex_lock(&ctrl->namespaces_mutex);
1450 list_for_each_entry(ns, &ctrl->namespaces, list) {
1451 spin_lock_irq(ns->queue->queue_lock);
1452 queue_flag_set(QUEUE_FLAG_STOPPED, ns->queue);
1453 spin_unlock_irq(ns->queue->queue_lock);
1455 blk_mq_cancel_requeue_work(ns->queue);
1456 blk_mq_stop_hw_queues(ns->queue);
1458 mutex_unlock(&ctrl->namespaces_mutex);
1460 EXPORT_SYMBOL_GPL(nvme_stop_queues);
1462 void nvme_start_queues(struct nvme_ctrl *ctrl)
1466 mutex_lock(&ctrl->namespaces_mutex);
1467 list_for_each_entry(ns, &ctrl->namespaces, list) {
1468 queue_flag_clear_unlocked(QUEUE_FLAG_STOPPED, ns->queue);
1469 blk_mq_start_stopped_hw_queues(ns->queue, true);
1470 blk_mq_kick_requeue_list(ns->queue);
1472 mutex_unlock(&ctrl->namespaces_mutex);
1474 EXPORT_SYMBOL_GPL(nvme_start_queues);
1476 int __init nvme_core_init(void)
1480 result = register_blkdev(nvme_major, "nvme");
1483 else if (result > 0)
1484 nvme_major = result;
1486 result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme",
1489 goto unregister_blkdev;
1490 else if (result > 0)
1491 nvme_char_major = result;
1493 nvme_class = class_create(THIS_MODULE, "nvme");
1494 if (IS_ERR(nvme_class)) {
1495 result = PTR_ERR(nvme_class);
1496 goto unregister_chrdev;
1502 __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
1504 unregister_blkdev(nvme_major, "nvme");
1508 void nvme_core_exit(void)
1510 unregister_blkdev(nvme_major, "nvme");
1511 class_destroy(nvme_class);
1512 __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
1515 MODULE_LICENSE("GPL");
1516 MODULE_VERSION("1.0");
1517 module_init(nvme_core_init);
1518 module_exit(nvme_core_exit);