2 * Copyright (c) 2011-2014, Intel Corporation.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
17 #include <linux/nvme.h>
18 #include <linux/pci.h>
19 #include <linux/kref.h>
20 #include <linux/blk-mq.h>
21 #include <linux/lightnvm.h>
22 #include <linux/sed-opal.h>
26 * Driver internal status code for commands that were cancelled due
27 * to timeouts or controller shutdown. The value is negative so
28 * that it a) doesn't overlap with the unsigned hardware error codes,
29 * and b) can easily be tested for.
31 NVME_SC_CANCELLED = -EINTR,
34 extern unsigned char nvme_io_timeout;
35 #define NVME_IO_TIMEOUT (nvme_io_timeout * HZ)
37 extern unsigned char admin_timeout;
38 #define ADMIN_TIMEOUT (admin_timeout * HZ)
40 extern unsigned char shutdown_timeout;
41 #define SHUTDOWN_TIMEOUT (shutdown_timeout * HZ)
43 #define NVME_DEFAULT_KATO 5
44 #define NVME_KATO_GRACE 10
46 extern unsigned int nvme_max_retries;
54 * List of workarounds for devices that required behavior not specified in
59 * Prefers I/O aligned to a stripe size specified in a vendor
60 * specific Identify field.
62 NVME_QUIRK_STRIPE_SIZE = (1 << 0),
65 * The controller doesn't handle Identify value others than 0 or 1
68 NVME_QUIRK_IDENTIFY_CNS = (1 << 1),
71 * The controller deterministically returns O's on reads to discarded
74 NVME_QUIRK_DISCARD_ZEROES = (1 << 2),
77 * The controller needs a delay before starts checking the device
78 * readiness, which is done by reading the NVME_CSTS_RDY bit.
80 NVME_QUIRK_DELAY_BEFORE_CHK_RDY = (1 << 3),
83 * APST should not be used.
85 NVME_QUIRK_NO_APST = (1 << 4),
88 * The deepest sleep state should not be used.
90 NVME_QUIRK_NO_DEEPEST_PS = (1 << 5),
94 * Common request structure for NVMe passthrough. All drivers must have
95 * this structure as the first member of their request-private data.
98 struct nvme_command *cmd;
99 union nvme_result result;
102 static inline struct nvme_request *nvme_req(struct request *req)
104 return blk_mq_rq_to_pdu(req);
107 /* The below value is the specific amount of delay needed before checking
108 * readiness in case of the PCI_DEVICE(0x1c58, 0x0003), which needs the
109 * NVME_QUIRK_DELAY_BEFORE_CHK_RDY quirk enabled. The value (in ms) was
112 #define NVME_QUIRK_DELAY_AMOUNT 2000
114 enum nvme_ctrl_state {
118 NVME_CTRL_RECONNECTING,
124 enum nvme_ctrl_state state;
127 const struct nvme_ctrl_ops *ops;
128 struct request_queue *admin_q;
129 struct request_queue *connect_q;
133 struct blk_mq_tag_set *tagset;
134 struct list_head namespaces;
135 struct mutex namespaces_mutex;
136 struct device *device; /* char device */
137 struct list_head node;
140 struct opal_dev *opal_dev;
145 char firmware_rev[8];
155 atomic_t abort_limit;
165 unsigned long quirks;
166 struct nvme_id_power_state psd[32];
167 struct work_struct scan_work;
168 struct work_struct async_event_work;
169 struct delayed_work ka_work;
171 /* Power saving configuration */
172 u64 ps_max_latency_us;
180 struct nvmf_ctrl_options *opts;
184 * An NVM Express namespace is equivalent to a SCSI LUN
187 struct list_head list;
189 struct nvme_ctrl *ctrl;
190 struct request_queue *queue;
191 struct gendisk *disk;
192 struct nvm_dev *ndev;
206 #define NVME_NS_REMOVING 0
207 #define NVME_NS_DEAD 1
209 u64 mode_select_num_blocks;
210 u32 mode_select_block_len;
213 struct nvme_ctrl_ops {
215 struct module *module;
217 int (*reg_read32)(struct nvme_ctrl *ctrl, u32 off, u32 *val);
218 int (*reg_write32)(struct nvme_ctrl *ctrl, u32 off, u32 val);
219 int (*reg_read64)(struct nvme_ctrl *ctrl, u32 off, u64 *val);
220 int (*reset_ctrl)(struct nvme_ctrl *ctrl);
221 void (*free_ctrl)(struct nvme_ctrl *ctrl);
222 void (*submit_async_event)(struct nvme_ctrl *ctrl, int aer_idx);
223 int (*delete_ctrl)(struct nvme_ctrl *ctrl);
224 const char *(*get_subsysnqn)(struct nvme_ctrl *ctrl);
225 int (*get_address)(struct nvme_ctrl *ctrl, char *buf, int size);
228 static inline bool nvme_ctrl_ready(struct nvme_ctrl *ctrl)
232 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &val))
234 return val & NVME_CSTS_RDY;
237 static inline int nvme_reset_subsystem(struct nvme_ctrl *ctrl)
239 if (!ctrl->subsystem)
241 return ctrl->ops->reg_write32(ctrl, NVME_REG_NSSR, 0x4E564D65);
244 static inline u64 nvme_block_nr(struct nvme_ns *ns, sector_t sector)
246 return (sector >> (ns->lba_shift - 9));
249 static inline void nvme_cleanup_cmd(struct request *req)
251 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
252 kfree(page_address(req->special_vec.bv_page) +
253 req->special_vec.bv_offset);
257 static inline int nvme_error_status(u16 status)
259 switch (status & 0x7ff) {
260 case NVME_SC_SUCCESS:
262 case NVME_SC_CAP_EXCEEDED:
269 static inline bool nvme_req_needs_retry(struct request *req, u16 status)
271 return !(status & NVME_SC_DNR || blk_noretry_request(req)) &&
272 (jiffies - req->start_time) < req->timeout &&
273 req->retries < nvme_max_retries;
276 void nvme_cancel_request(struct request *req, void *data, bool reserved);
277 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
278 enum nvme_ctrl_state new_state);
279 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap);
280 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap);
281 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl);
282 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
283 const struct nvme_ctrl_ops *ops, unsigned long quirks);
284 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl);
285 void nvme_put_ctrl(struct nvme_ctrl *ctrl);
286 int nvme_init_identify(struct nvme_ctrl *ctrl);
288 void nvme_queue_scan(struct nvme_ctrl *ctrl);
289 void nvme_remove_namespaces(struct nvme_ctrl *ctrl);
291 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
294 #define NVME_NR_AERS 1
295 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
296 union nvme_result *res);
297 void nvme_queue_async_events(struct nvme_ctrl *ctrl);
299 void nvme_stop_queues(struct nvme_ctrl *ctrl);
300 void nvme_start_queues(struct nvme_ctrl *ctrl);
301 void nvme_kill_queues(struct nvme_ctrl *ctrl);
302 void nvme_unfreeze(struct nvme_ctrl *ctrl);
303 void nvme_wait_freeze(struct nvme_ctrl *ctrl);
304 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout);
305 void nvme_start_freeze(struct nvme_ctrl *ctrl);
307 #define NVME_QID_ANY -1
308 struct request *nvme_alloc_request(struct request_queue *q,
309 struct nvme_command *cmd, unsigned int flags, int qid);
310 void nvme_requeue_req(struct request *req);
311 int nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
312 struct nvme_command *cmd);
313 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
314 void *buf, unsigned bufflen);
315 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
316 union nvme_result *result, void *buffer, unsigned bufflen,
317 unsigned timeout, int qid, int at_head, int flags);
318 int nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
319 void __user *ubuffer, unsigned bufflen, u32 *result,
321 int __nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
322 void __user *ubuffer, unsigned bufflen,
323 void __user *meta_buffer, unsigned meta_len, u32 meta_seed,
324 u32 *result, unsigned timeout);
325 int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id);
326 int nvme_identify_ns(struct nvme_ctrl *dev, unsigned nsid,
327 struct nvme_id_ns **id);
328 int nvme_get_log_page(struct nvme_ctrl *dev, struct nvme_smart_log **log);
329 int nvme_get_features(struct nvme_ctrl *dev, unsigned fid, unsigned nsid,
330 void *buffer, size_t buflen, u32 *result);
331 int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
332 void *buffer, size_t buflen, u32 *result);
333 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count);
334 void nvme_start_keep_alive(struct nvme_ctrl *ctrl);
335 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl);
339 int nvme_sg_io(struct nvme_ns *ns, struct sg_io_hdr __user *u_hdr);
340 int nvme_sg_io32(struct nvme_ns *ns, unsigned long arg);
341 int nvme_sg_get_version_num(int __user *ip);
344 int nvme_nvm_ns_supported(struct nvme_ns *ns, struct nvme_id_ns *id);
345 int nvme_nvm_register(struct nvme_ns *ns, char *disk_name, int node);
346 void nvme_nvm_unregister(struct nvme_ns *ns);
347 int nvme_nvm_register_sysfs(struct nvme_ns *ns);
348 void nvme_nvm_unregister_sysfs(struct nvme_ns *ns);
349 int nvme_nvm_ioctl(struct nvme_ns *ns, unsigned int cmd, unsigned long arg);
351 static inline int nvme_nvm_register(struct nvme_ns *ns, char *disk_name,
357 static inline void nvme_nvm_unregister(struct nvme_ns *ns) {};
358 static inline int nvme_nvm_register_sysfs(struct nvme_ns *ns)
362 static inline void nvme_nvm_unregister_sysfs(struct nvme_ns *ns) {};
363 static inline int nvme_nvm_ns_supported(struct nvme_ns *ns, struct nvme_id_ns *id)
367 static inline int nvme_nvm_ioctl(struct nvme_ns *ns, unsigned int cmd,
372 #endif /* CONFIG_NVM */
374 static inline struct nvme_ns *nvme_get_ns_from_dev(struct device *dev)
376 return dev_to_disk(dev)->private_data;
379 int __init nvme_core_init(void);
380 void nvme_core_exit(void);