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 struct request *nvme_alloc_request(struct request_queue *q,
82 struct nvme_command *cmd, unsigned int flags)
84 bool write = cmd->common.opcode & 1;
87 req = blk_mq_alloc_request(q, write, flags);
91 req->cmd_type = REQ_TYPE_DRV_PRIV;
92 req->cmd_flags |= REQ_FAILFAST_DRIVER;
94 req->__sector = (sector_t) -1;
95 req->bio = req->biotail = NULL;
97 req->cmd = (unsigned char *)cmd;
98 req->cmd_len = sizeof(struct nvme_command);
99 req->special = (void *)0;
105 * Returns 0 on success. If the result is negative, it's a Linux error code;
106 * if the result is positive, it's an NVM Express status code
108 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
109 void *buffer, unsigned bufflen, u32 *result, unsigned timeout)
114 req = nvme_alloc_request(q, cmd, 0);
118 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
120 if (buffer && bufflen) {
121 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
126 blk_execute_rq(req->q, NULL, req, 0);
128 *result = (u32)(uintptr_t)req->special;
131 blk_mq_free_request(req);
135 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
136 void *buffer, unsigned bufflen)
138 return __nvme_submit_sync_cmd(q, cmd, buffer, bufflen, NULL, 0);
141 int __nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
142 void __user *ubuffer, unsigned bufflen,
143 void __user *meta_buffer, unsigned meta_len, u32 meta_seed,
144 u32 *result, unsigned timeout)
146 bool write = cmd->common.opcode & 1;
147 struct nvme_ns *ns = q->queuedata;
148 struct gendisk *disk = ns ? ns->disk : NULL;
150 struct bio *bio = NULL;
154 req = nvme_alloc_request(q, cmd, 0);
158 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
160 if (ubuffer && bufflen) {
161 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
169 bio->bi_bdev = bdget_disk(disk, 0);
176 struct bio_integrity_payload *bip;
178 meta = kmalloc(meta_len, GFP_KERNEL);
185 if (copy_from_user(meta, meta_buffer,
192 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
198 bip->bip_iter.bi_size = meta_len;
199 bip->bip_iter.bi_sector = meta_seed;
201 ret = bio_integrity_add_page(bio, virt_to_page(meta),
202 meta_len, offset_in_page(meta));
203 if (ret != meta_len) {
210 blk_execute_rq(req->q, disk, req, 0);
213 *result = (u32)(uintptr_t)req->special;
214 if (meta && !ret && !write) {
215 if (copy_to_user(meta_buffer, meta, meta_len))
222 if (disk && bio->bi_bdev)
224 blk_rq_unmap_user(bio);
227 blk_mq_free_request(req);
231 int nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
232 void __user *ubuffer, unsigned bufflen, u32 *result,
235 return __nvme_submit_user_cmd(q, cmd, ubuffer, bufflen, NULL, 0, 0,
239 int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
241 struct nvme_command c = { };
244 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
245 c.identify.opcode = nvme_admin_identify;
246 c.identify.cns = cpu_to_le32(1);
248 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
252 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
253 sizeof(struct nvme_id_ctrl));
259 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
261 struct nvme_command c = { };
263 c.identify.opcode = nvme_admin_identify;
264 c.identify.cns = cpu_to_le32(2);
265 c.identify.nsid = cpu_to_le32(nsid);
266 return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000);
269 int nvme_identify_ns(struct nvme_ctrl *dev, unsigned nsid,
270 struct nvme_id_ns **id)
272 struct nvme_command c = { };
275 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
276 c.identify.opcode = nvme_admin_identify,
277 c.identify.nsid = cpu_to_le32(nsid),
279 *id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL);
283 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
284 sizeof(struct nvme_id_ns));
290 int nvme_get_features(struct nvme_ctrl *dev, unsigned fid, unsigned nsid,
291 dma_addr_t dma_addr, u32 *result)
293 struct nvme_command c;
295 memset(&c, 0, sizeof(c));
296 c.features.opcode = nvme_admin_get_features;
297 c.features.nsid = cpu_to_le32(nsid);
298 c.features.prp1 = cpu_to_le64(dma_addr);
299 c.features.fid = cpu_to_le32(fid);
301 return __nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0, result, 0);
304 int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
305 dma_addr_t dma_addr, u32 *result)
307 struct nvme_command c;
309 memset(&c, 0, sizeof(c));
310 c.features.opcode = nvme_admin_set_features;
311 c.features.prp1 = cpu_to_le64(dma_addr);
312 c.features.fid = cpu_to_le32(fid);
313 c.features.dword11 = cpu_to_le32(dword11);
315 return __nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0, result, 0);
318 int nvme_get_log_page(struct nvme_ctrl *dev, struct nvme_smart_log **log)
320 struct nvme_command c = { };
323 c.common.opcode = nvme_admin_get_log_page,
324 c.common.nsid = cpu_to_le32(0xFFFFFFFF),
325 c.common.cdw10[0] = cpu_to_le32(
326 (((sizeof(struct nvme_smart_log) / 4) - 1) << 16) |
329 *log = kmalloc(sizeof(struct nvme_smart_log), GFP_KERNEL);
333 error = nvme_submit_sync_cmd(dev->admin_q, &c, *log,
334 sizeof(struct nvme_smart_log));
340 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
342 u32 q_count = (*count - 1) | ((*count - 1) << 16);
344 int status, nr_io_queues;
346 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, 0,
351 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
352 *count = min(*count, nr_io_queues);
356 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
358 struct nvme_user_io io;
359 struct nvme_command c;
360 unsigned length, meta_len;
361 void __user *metadata;
363 if (copy_from_user(&io, uio, sizeof(io)))
369 case nvme_cmd_compare:
375 length = (io.nblocks + 1) << ns->lba_shift;
376 meta_len = (io.nblocks + 1) * ns->ms;
377 metadata = (void __user *)(uintptr_t)io.metadata;
382 } else if (meta_len) {
383 if ((io.metadata & 3) || !io.metadata)
387 memset(&c, 0, sizeof(c));
388 c.rw.opcode = io.opcode;
389 c.rw.flags = io.flags;
390 c.rw.nsid = cpu_to_le32(ns->ns_id);
391 c.rw.slba = cpu_to_le64(io.slba);
392 c.rw.length = cpu_to_le16(io.nblocks);
393 c.rw.control = cpu_to_le16(io.control);
394 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
395 c.rw.reftag = cpu_to_le32(io.reftag);
396 c.rw.apptag = cpu_to_le16(io.apptag);
397 c.rw.appmask = cpu_to_le16(io.appmask);
399 return __nvme_submit_user_cmd(ns->queue, &c,
400 (void __user *)(uintptr_t)io.addr, length,
401 metadata, meta_len, io.slba, NULL, 0);
404 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
405 struct nvme_passthru_cmd __user *ucmd)
407 struct nvme_passthru_cmd cmd;
408 struct nvme_command c;
409 unsigned timeout = 0;
412 if (!capable(CAP_SYS_ADMIN))
414 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
417 memset(&c, 0, sizeof(c));
418 c.common.opcode = cmd.opcode;
419 c.common.flags = cmd.flags;
420 c.common.nsid = cpu_to_le32(cmd.nsid);
421 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
422 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
423 c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
424 c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
425 c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
426 c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
427 c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
428 c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
431 timeout = msecs_to_jiffies(cmd.timeout_ms);
433 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
434 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
435 &cmd.result, timeout);
437 if (put_user(cmd.result, &ucmd->result))
444 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
445 unsigned int cmd, unsigned long arg)
447 struct nvme_ns *ns = bdev->bd_disk->private_data;
451 force_successful_syscall_return();
453 case NVME_IOCTL_ADMIN_CMD:
454 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
455 case NVME_IOCTL_IO_CMD:
456 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
457 case NVME_IOCTL_SUBMIT_IO:
458 return nvme_submit_io(ns, (void __user *)arg);
459 case SG_GET_VERSION_NUM:
460 return nvme_sg_get_version_num((void __user *)arg);
462 return nvme_sg_io(ns, (void __user *)arg);
469 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
470 unsigned int cmd, unsigned long arg)
476 return nvme_ioctl(bdev, mode, cmd, arg);
479 #define nvme_compat_ioctl NULL
482 static int nvme_open(struct block_device *bdev, fmode_t mode)
484 return nvme_get_ns_from_disk(bdev->bd_disk) ? 0 : -ENXIO;
487 static void nvme_release(struct gendisk *disk, fmode_t mode)
489 nvme_put_ns(disk->private_data);
492 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
494 /* some standard values */
496 geo->sectors = 1 << 5;
497 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
501 #ifdef CONFIG_BLK_DEV_INTEGRITY
502 static void nvme_init_integrity(struct nvme_ns *ns)
504 struct blk_integrity integrity;
506 switch (ns->pi_type) {
507 case NVME_NS_DPS_PI_TYPE3:
508 integrity.profile = &t10_pi_type3_crc;
510 case NVME_NS_DPS_PI_TYPE1:
511 case NVME_NS_DPS_PI_TYPE2:
512 integrity.profile = &t10_pi_type1_crc;
515 integrity.profile = NULL;
518 integrity.tuple_size = ns->ms;
519 blk_integrity_register(ns->disk, &integrity);
520 blk_queue_max_integrity_segments(ns->queue, 1);
523 static void nvme_init_integrity(struct nvme_ns *ns)
526 #endif /* CONFIG_BLK_DEV_INTEGRITY */
528 static void nvme_config_discard(struct nvme_ns *ns)
530 u32 logical_block_size = queue_logical_block_size(ns->queue);
531 ns->queue->limits.discard_zeroes_data = 0;
532 ns->queue->limits.discard_alignment = logical_block_size;
533 ns->queue->limits.discard_granularity = logical_block_size;
534 blk_queue_max_discard_sectors(ns->queue, 0xffffffff);
535 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue);
538 static int nvme_revalidate_disk(struct gendisk *disk)
540 struct nvme_ns *ns = disk->private_data;
541 struct nvme_id_ns *id;
546 if (nvme_identify_ns(ns->ctrl, ns->ns_id, &id)) {
547 dev_warn(ns->ctrl->dev, "%s: Identify failure nvme%dn%d\n",
548 __func__, ns->ctrl->instance, ns->ns_id);
556 if (nvme_nvm_ns_supported(ns, id) && ns->type != NVME_NS_LIGHTNVM) {
557 if (nvme_nvm_register(ns->queue, disk->disk_name)) {
558 dev_warn(ns->ctrl->dev,
559 "%s: LightNVM init failure\n", __func__);
563 ns->type = NVME_NS_LIGHTNVM;
567 lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
568 ns->lba_shift = id->lbaf[lbaf].ds;
569 ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
570 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
573 * If identify namespace failed, use default 512 byte block size so
574 * block layer can use before failing read/write for 0 capacity.
576 if (ns->lba_shift == 0)
578 bs = 1 << ns->lba_shift;
579 /* XXX: PI implementation requires metadata equal t10 pi tuple size */
580 pi_type = ns->ms == sizeof(struct t10_pi_tuple) ?
581 id->dps & NVME_NS_DPS_PI_MASK : 0;
583 blk_mq_freeze_queue(disk->queue);
584 if (blk_get_integrity(disk) && (ns->pi_type != pi_type ||
586 bs != queue_logical_block_size(disk->queue) ||
587 (ns->ms && ns->ext)))
588 blk_integrity_unregister(disk);
590 ns->pi_type = pi_type;
591 blk_queue_logical_block_size(ns->queue, bs);
593 if (ns->ms && !blk_get_integrity(disk) && !ns->ext)
594 nvme_init_integrity(ns);
595 if (ns->ms && !(ns->ms == 8 && ns->pi_type) && !blk_get_integrity(disk))
596 set_capacity(disk, 0);
598 set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
600 if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
601 nvme_config_discard(ns);
602 blk_mq_unfreeze_queue(disk->queue);
608 static char nvme_pr_type(enum pr_type type)
611 case PR_WRITE_EXCLUSIVE:
613 case PR_EXCLUSIVE_ACCESS:
615 case PR_WRITE_EXCLUSIVE_REG_ONLY:
617 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
619 case PR_WRITE_EXCLUSIVE_ALL_REGS:
621 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
628 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
629 u64 key, u64 sa_key, u8 op)
631 struct nvme_ns *ns = bdev->bd_disk->private_data;
632 struct nvme_command c;
633 u8 data[16] = { 0, };
635 put_unaligned_le64(key, &data[0]);
636 put_unaligned_le64(sa_key, &data[8]);
638 memset(&c, 0, sizeof(c));
639 c.common.opcode = op;
640 c.common.nsid = cpu_to_le32(ns->ns_id);
641 c.common.cdw10[0] = cpu_to_le32(cdw10);
643 return nvme_submit_sync_cmd(ns->queue, &c, data, 16);
646 static int nvme_pr_register(struct block_device *bdev, u64 old,
647 u64 new, unsigned flags)
651 if (flags & ~PR_FL_IGNORE_KEY)
655 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
656 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
657 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
660 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
661 enum pr_type type, unsigned flags)
665 if (flags & ~PR_FL_IGNORE_KEY)
668 cdw10 = nvme_pr_type(type) << 8;
669 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
670 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
673 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
674 enum pr_type type, bool abort)
676 u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
677 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
680 static int nvme_pr_clear(struct block_device *bdev, u64 key)
682 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
683 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
686 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
688 u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
689 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
692 static const struct pr_ops nvme_pr_ops = {
693 .pr_register = nvme_pr_register,
694 .pr_reserve = nvme_pr_reserve,
695 .pr_release = nvme_pr_release,
696 .pr_preempt = nvme_pr_preempt,
697 .pr_clear = nvme_pr_clear,
700 static const struct block_device_operations nvme_fops = {
701 .owner = THIS_MODULE,
703 .compat_ioctl = nvme_compat_ioctl,
705 .release = nvme_release,
706 .getgeo = nvme_getgeo,
707 .revalidate_disk= nvme_revalidate_disk,
708 .pr_ops = &nvme_pr_ops,
711 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
713 unsigned long timeout =
714 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
715 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
718 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
719 if ((csts & NVME_CSTS_RDY) == bit)
723 if (fatal_signal_pending(current))
725 if (time_after(jiffies, timeout)) {
727 "Device not ready; aborting %s\n", enabled ?
728 "initialisation" : "reset");
737 * If the device has been passed off to us in an enabled state, just clear
738 * the enabled bit. The spec says we should set the 'shutdown notification
739 * bits', but doing so may cause the device to complete commands to the
740 * admin queue ... and we don't know what memory that might be pointing at!
742 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
746 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
747 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
749 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
752 return nvme_wait_ready(ctrl, cap, false);
755 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
758 * Default to a 4K page size, with the intention to update this
759 * path in the future to accomodate architectures with differing
760 * kernel and IO page sizes.
762 unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
765 if (page_shift < dev_page_min) {
767 "Minimum device page size %u too large for host (%u)\n",
768 1 << dev_page_min, 1 << page_shift);
772 ctrl->page_size = 1 << page_shift;
774 ctrl->ctrl_config = NVME_CC_CSS_NVM;
775 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
776 ctrl->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE;
777 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
778 ctrl->ctrl_config |= NVME_CC_ENABLE;
780 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
783 return nvme_wait_ready(ctrl, cap, true);
786 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
788 unsigned long timeout = SHUTDOWN_TIMEOUT + jiffies;
792 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
793 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
795 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
799 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
800 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
804 if (fatal_signal_pending(current))
806 if (time_after(jiffies, timeout)) {
808 "Device shutdown incomplete; abort shutdown\n");
817 * Initialize the cached copies of the Identify data and various controller
818 * register in our nvme_ctrl structure. This should be called as soon as
819 * the admin queue is fully up and running.
821 int nvme_init_identify(struct nvme_ctrl *ctrl)
823 struct nvme_id_ctrl *id;
827 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
829 dev_err(ctrl->dev, "Reading VS failed (%d)\n", ret);
833 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
835 dev_err(ctrl->dev, "Reading CAP failed (%d)\n", ret);
838 page_shift = NVME_CAP_MPSMIN(cap) + 12;
840 if (ctrl->vs >= NVME_VS(1, 1))
841 ctrl->subsystem = NVME_CAP_NSSRC(cap);
843 ret = nvme_identify_ctrl(ctrl, &id);
845 dev_err(ctrl->dev, "Identify Controller failed (%d)\n", ret);
849 ctrl->oncs = le16_to_cpup(&id->oncs);
850 atomic_set(&ctrl->abort_limit, id->acl + 1);
852 memcpy(ctrl->serial, id->sn, sizeof(id->sn));
853 memcpy(ctrl->model, id->mn, sizeof(id->mn));
854 memcpy(ctrl->firmware_rev, id->fr, sizeof(id->fr));
856 ctrl->max_hw_sectors = 1 << (id->mdts + page_shift - 9);
858 ctrl->max_hw_sectors = UINT_MAX;
860 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) && id->vs[3]) {
861 unsigned int max_hw_sectors;
863 ctrl->stripe_size = 1 << (id->vs[3] + page_shift);
864 max_hw_sectors = ctrl->stripe_size >> (page_shift - 9);
865 if (ctrl->max_hw_sectors) {
866 ctrl->max_hw_sectors = min(max_hw_sectors,
867 ctrl->max_hw_sectors);
869 ctrl->max_hw_sectors = max_hw_sectors;
877 static int nvme_dev_open(struct inode *inode, struct file *file)
879 struct nvme_ctrl *ctrl;
880 int instance = iminor(inode);
883 spin_lock(&dev_list_lock);
884 list_for_each_entry(ctrl, &nvme_ctrl_list, node) {
885 if (ctrl->instance != instance)
888 if (!ctrl->admin_q) {
892 if (!kref_get_unless_zero(&ctrl->kref))
894 file->private_data = ctrl;
898 spin_unlock(&dev_list_lock);
903 static int nvme_dev_release(struct inode *inode, struct file *file)
905 nvme_put_ctrl(file->private_data);
909 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
912 struct nvme_ctrl *ctrl = file->private_data;
913 void __user *argp = (void __user *)arg;
917 case NVME_IOCTL_ADMIN_CMD:
918 return nvme_user_cmd(ctrl, NULL, argp);
919 case NVME_IOCTL_IO_CMD:
920 if (list_empty(&ctrl->namespaces))
922 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
923 return nvme_user_cmd(ctrl, ns, argp);
924 case NVME_IOCTL_RESET:
925 dev_warn(ctrl->dev, "resetting controller\n");
926 return ctrl->ops->reset_ctrl(ctrl);
927 case NVME_IOCTL_SUBSYS_RESET:
928 return nvme_reset_subsystem(ctrl);
934 static const struct file_operations nvme_dev_fops = {
935 .owner = THIS_MODULE,
936 .open = nvme_dev_open,
937 .release = nvme_dev_release,
938 .unlocked_ioctl = nvme_dev_ioctl,
939 .compat_ioctl = nvme_dev_ioctl,
942 static ssize_t nvme_sysfs_reset(struct device *dev,
943 struct device_attribute *attr, const char *buf,
946 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
949 ret = ctrl->ops->reset_ctrl(ctrl);
954 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
956 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
958 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
959 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
961 return nsa->ns_id - nsb->ns_id;
964 static struct nvme_ns *nvme_find_ns(struct nvme_ctrl *ctrl, unsigned nsid)
968 list_for_each_entry(ns, &ctrl->namespaces, list) {
969 if (ns->ns_id == nsid)
971 if (ns->ns_id > nsid)
977 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
980 struct gendisk *disk;
981 int node = dev_to_node(ctrl->dev);
983 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
987 ns->queue = blk_mq_init_queue(ctrl->tagset);
988 if (IS_ERR(ns->queue))
990 queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES, ns->queue);
991 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
992 ns->queue->queuedata = ns;
995 disk = alloc_disk_node(0, node);
999 kref_init(&ns->kref);
1002 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
1004 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
1005 if (ctrl->max_hw_sectors) {
1006 blk_queue_max_hw_sectors(ns->queue, ctrl->max_hw_sectors);
1007 blk_queue_max_segments(ns->queue,
1008 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1);
1010 if (ctrl->stripe_size)
1011 blk_queue_chunk_sectors(ns->queue, ctrl->stripe_size >> 9);
1012 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1013 blk_queue_flush(ns->queue, REQ_FLUSH | REQ_FUA);
1014 blk_queue_virt_boundary(ns->queue, ctrl->page_size - 1);
1016 disk->major = nvme_major;
1017 disk->first_minor = 0;
1018 disk->fops = &nvme_fops;
1019 disk->private_data = ns;
1020 disk->queue = ns->queue;
1021 disk->driverfs_dev = ctrl->device;
1022 disk->flags = GENHD_FL_EXT_DEVT;
1023 sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance, nsid);
1025 if (nvme_revalidate_disk(ns->disk))
1028 list_add_tail(&ns->list, &ctrl->namespaces);
1029 kref_get(&ctrl->kref);
1030 if (ns->type != NVME_NS_LIGHTNVM)
1037 blk_cleanup_queue(ns->queue);
1042 static void nvme_ns_remove(struct nvme_ns *ns)
1044 bool kill = nvme_io_incapable(ns->ctrl) &&
1045 !blk_queue_dying(ns->queue);
1048 blk_set_queue_dying(ns->queue);
1049 if (ns->disk->flags & GENHD_FL_UP) {
1050 if (blk_get_integrity(ns->disk))
1051 blk_integrity_unregister(ns->disk);
1052 del_gendisk(ns->disk);
1054 if (kill || !blk_queue_dying(ns->queue)) {
1055 blk_mq_abort_requeue_list(ns->queue);
1056 blk_cleanup_queue(ns->queue);
1058 list_del_init(&ns->list);
1062 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1066 ns = nvme_find_ns(ctrl, nsid);
1068 if (revalidate_disk(ns->disk))
1071 nvme_alloc_ns(ctrl, nsid);
1074 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
1078 unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
1081 ns_list = kzalloc(0x1000, GFP_KERNEL);
1085 for (i = 0; i < num_lists; i++) {
1086 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
1090 for (j = 0; j < min(nn, 1024U); j++) {
1091 nsid = le32_to_cpu(ns_list[j]);
1095 nvme_validate_ns(ctrl, nsid);
1097 while (++prev < nsid) {
1098 ns = nvme_find_ns(ctrl, prev);
1110 static void __nvme_scan_namespaces(struct nvme_ctrl *ctrl, unsigned nn)
1112 struct nvme_ns *ns, *next;
1115 for (i = 1; i <= nn; i++)
1116 nvme_validate_ns(ctrl, i);
1118 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
1124 void nvme_scan_namespaces(struct nvme_ctrl *ctrl)
1126 struct nvme_id_ctrl *id;
1129 if (nvme_identify_ctrl(ctrl, &id))
1132 nn = le32_to_cpu(id->nn);
1133 if (ctrl->vs >= NVME_VS(1, 1) &&
1134 !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
1135 if (!nvme_scan_ns_list(ctrl, nn))
1138 __nvme_scan_namespaces(ctrl, le32_to_cpup(&id->nn));
1140 list_sort(NULL, &ctrl->namespaces, ns_cmp);
1144 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
1146 struct nvme_ns *ns, *next;
1148 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
1152 static DEFINE_IDA(nvme_instance_ida);
1154 static int nvme_set_instance(struct nvme_ctrl *ctrl)
1156 int instance, error;
1159 if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL))
1162 spin_lock(&dev_list_lock);
1163 error = ida_get_new(&nvme_instance_ida, &instance);
1164 spin_unlock(&dev_list_lock);
1165 } while (error == -EAGAIN);
1170 ctrl->instance = instance;
1174 static void nvme_release_instance(struct nvme_ctrl *ctrl)
1176 spin_lock(&dev_list_lock);
1177 ida_remove(&nvme_instance_ida, ctrl->instance);
1178 spin_unlock(&dev_list_lock);
1181 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
1183 device_remove_file(ctrl->device, &dev_attr_reset_controller);
1184 device_destroy(nvme_class, MKDEV(nvme_char_major, ctrl->instance));
1186 spin_lock(&dev_list_lock);
1187 list_del(&ctrl->node);
1188 spin_unlock(&dev_list_lock);
1191 static void nvme_free_ctrl(struct kref *kref)
1193 struct nvme_ctrl *ctrl = container_of(kref, struct nvme_ctrl, kref);
1195 put_device(ctrl->device);
1196 nvme_release_instance(ctrl);
1198 ctrl->ops->free_ctrl(ctrl);
1201 void nvme_put_ctrl(struct nvme_ctrl *ctrl)
1203 kref_put(&ctrl->kref, nvme_free_ctrl);
1207 * Initialize a NVMe controller structures. This needs to be called during
1208 * earliest initialization so that we have the initialized structured around
1211 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
1212 const struct nvme_ctrl_ops *ops, unsigned long quirks)
1216 INIT_LIST_HEAD(&ctrl->namespaces);
1217 kref_init(&ctrl->kref);
1220 ctrl->quirks = quirks;
1222 ret = nvme_set_instance(ctrl);
1226 ctrl->device = device_create(nvme_class, ctrl->dev,
1227 MKDEV(nvme_char_major, ctrl->instance),
1228 dev, "nvme%d", ctrl->instance);
1229 if (IS_ERR(ctrl->device)) {
1230 ret = PTR_ERR(ctrl->device);
1231 goto out_release_instance;
1233 get_device(ctrl->device);
1234 dev_set_drvdata(ctrl->device, ctrl);
1236 ret = device_create_file(ctrl->device, &dev_attr_reset_controller);
1238 goto out_put_device;
1240 spin_lock(&dev_list_lock);
1241 list_add_tail(&ctrl->node, &nvme_ctrl_list);
1242 spin_unlock(&dev_list_lock);
1247 put_device(ctrl->device);
1248 device_destroy(nvme_class, MKDEV(nvme_char_major, ctrl->instance));
1249 out_release_instance:
1250 nvme_release_instance(ctrl);
1255 int __init nvme_core_init(void)
1259 result = register_blkdev(nvme_major, "nvme");
1262 else if (result > 0)
1263 nvme_major = result;
1265 result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme",
1268 goto unregister_blkdev;
1269 else if (result > 0)
1270 nvme_char_major = result;
1272 nvme_class = class_create(THIS_MODULE, "nvme");
1273 if (IS_ERR(nvme_class)) {
1274 result = PTR_ERR(nvme_class);
1275 goto unregister_chrdev;
1281 __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
1283 unregister_blkdev(nvme_major, "nvme");
1287 void nvme_core_exit(void)
1289 unregister_blkdev(nvme_major, "nvme");
1290 class_destroy(nvme_class);
1291 __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");