xref: /spdk/lib/nvme/nvme_internal.h (revision ee32a82bfd3ff5b1a10ed775ee06f0eaffce60eb)
1 /*   SPDX-License-Identifier: BSD-3-Clause
2  *   Copyright (C) 2015 Intel Corporation. All rights reserved.
3  *   Copyright (c) 2020, 2021 Mellanox Technologies LTD. All rights reserved.
4  *   Copyright (c) 2021-2023 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
5  */
6 
7 #ifndef __NVME_INTERNAL_H__
8 #define __NVME_INTERNAL_H__
9 
10 #include "spdk/config.h"
11 #include "spdk/likely.h"
12 #include "spdk/stdinc.h"
13 
14 #include "spdk/nvme.h"
15 
16 #if defined(__i386__) || defined(__x86_64__)
17 #include <x86intrin.h>
18 #endif
19 
20 #include "spdk/queue.h"
21 #include "spdk/barrier.h"
22 #include "spdk/bit_array.h"
23 #include "spdk/mmio.h"
24 #include "spdk/pci_ids.h"
25 #include "spdk/util.h"
26 #include "spdk/memory.h"
27 #include "spdk/nvme_intel.h"
28 #include "spdk/nvmf_spec.h"
29 #include "spdk/tree.h"
30 #include "spdk/uuid.h"
31 
32 #include "spdk_internal/assert.h"
33 #include "spdk/log.h"
34 
35 extern pid_t g_spdk_nvme_pid;
36 
37 extern struct spdk_nvme_transport_opts g_spdk_nvme_transport_opts;
38 
39 /*
40  * Some Intel devices support vendor-unique read latency log page even
41  * though the log page directory says otherwise.
42  */
43 #define NVME_INTEL_QUIRK_READ_LATENCY 0x1
44 
45 /*
46  * Some Intel devices support vendor-unique write latency log page even
47  * though the log page directory says otherwise.
48  */
49 #define NVME_INTEL_QUIRK_WRITE_LATENCY 0x2
50 
51 /*
52  * The controller needs a delay before starts checking the device
53  * readiness, which is done by reading the NVME_CSTS_RDY bit.
54  */
55 #define NVME_QUIRK_DELAY_BEFORE_CHK_RDY	0x4
56 
57 /*
58  * The controller performs best when I/O is split on particular
59  * LBA boundaries.
60  */
61 #define NVME_INTEL_QUIRK_STRIPING 0x8
62 
63 /*
64  * The controller needs a delay after allocating an I/O queue pair
65  * before it is ready to accept I/O commands.
66  */
67 #define NVME_QUIRK_DELAY_AFTER_QUEUE_ALLOC 0x10
68 
69 /*
70  * Earlier NVMe devices do not indicate whether unmapped blocks
71  * will read all zeroes or not. This define indicates that the
72  * device does in fact read all zeroes after an unmap event
73  */
74 #define NVME_QUIRK_READ_ZERO_AFTER_DEALLOCATE 0x20
75 
76 /*
77  * The controller doesn't handle Identify value others than 0 or 1 correctly.
78  */
79 #define NVME_QUIRK_IDENTIFY_CNS 0x40
80 
81 /*
82  * The controller supports Open Channel command set if matching additional
83  * condition, like the first byte (value 0x1) in the vendor specific
84  * bits of the namespace identify structure is set.
85  */
86 #define NVME_QUIRK_OCSSD 0x80
87 
88 /*
89  * The controller has an Intel vendor ID but does not support Intel vendor-specific
90  * log pages.  This is primarily for QEMU emulated SSDs which report an Intel vendor
91  * ID but do not support these log pages.
92  */
93 #define NVME_INTEL_QUIRK_NO_LOG_PAGES 0x100
94 
95 /*
96  * The controller does not set SHST_COMPLETE in a reasonable amount of time.  This
97  * is primarily seen in virtual VMWare NVMe SSDs.  This quirk merely adds an additional
98  * error message that on VMWare NVMe SSDs, the shutdown timeout may be expected.
99  */
100 #define NVME_QUIRK_SHST_COMPLETE 0x200
101 
102 /*
103  * The controller requires an extra delay before starting the initialization process
104  * during attach.
105  */
106 #define NVME_QUIRK_DELAY_BEFORE_INIT 0x400
107 
108 /*
109  * Some SSDs exhibit poor performance with the default SPDK NVMe IO queue size.
110  * This quirk will increase the default to 1024 which matches other operating
111  * systems, at the cost of some extra memory usage.  Users can still override
112  * the increased default by changing the spdk_nvme_io_qpair_opts when allocating
113  * a new queue pair.
114  */
115 #define NVME_QUIRK_MINIMUM_IO_QUEUE_SIZE 0x800
116 
117 /**
118  * The maximum access width to PCI memory space is 8 Bytes, don't use AVX2 or
119  * SSE instructions to optimize the memory access(memcpy or memset) larger than
120  * 8 Bytes.
121  */
122 #define NVME_QUIRK_MAXIMUM_PCI_ACCESS_WIDTH 0x1000
123 
124 /**
125  * The SSD does not support OPAL even through it sets the security bit in OACS.
126  */
127 #define NVME_QUIRK_OACS_SECURITY 0x2000
128 
129 /**
130  * Intel P55XX SSDs can't support Dataset Management command with SGL format,
131  * so use PRP with DSM command.
132  */
133 #define NVME_QUIRK_NO_SGL_FOR_DSM 0x4000
134 
135 /**
136  * Maximum Data Transfer Size(MDTS) excludes interleaved metadata.
137  */
138 #define NVME_QUIRK_MDTS_EXCLUDE_MD 0x8000
139 
140 /**
141  * Force not to use SGL even the controller report that it can
142  * support it.
143  */
144 #define NVME_QUIRK_NOT_USE_SGL 0x10000
145 
146 /*
147  * Some SSDs require the admin submission queue size to equate to an even
148  * 4KiB multiple.
149  */
150 #define NVME_QUIRK_MINIMUM_ADMIN_QUEUE_SIZE 0x20000
151 
152 #define NVME_MAX_ASYNC_EVENTS	(8)
153 
154 #define NVME_MAX_ADMIN_TIMEOUT_IN_SECS	(30)
155 
156 /* Maximum log page size to fetch for AERs. */
157 #define NVME_MAX_AER_LOG_SIZE		(4096)
158 
159 /*
160  * NVME_MAX_IO_QUEUES in nvme_spec.h defines the 64K spec-limit, but this
161  *  define specifies the maximum number of queues this driver will actually
162  *  try to configure, if available.
163  */
164 #define DEFAULT_MAX_IO_QUEUES		(1024)
165 #define DEFAULT_ADMIN_QUEUE_SIZE	(32)
166 #define DEFAULT_IO_QUEUE_SIZE		(256)
167 #define DEFAULT_IO_QUEUE_SIZE_FOR_QUIRK	(1024) /* Matches Linux kernel driver */
168 
169 #define DEFAULT_IO_QUEUE_REQUESTS	(512)
170 
171 #define SPDK_NVME_DEFAULT_RETRY_COUNT	(4)
172 
173 #define SPDK_NVME_TRANSPORT_ACK_TIMEOUT_DISABLED	(0)
174 #define SPDK_NVME_DEFAULT_TRANSPORT_ACK_TIMEOUT	SPDK_NVME_TRANSPORT_ACK_TIMEOUT_DISABLED
175 
176 #define SPDK_NVME_TRANSPORT_TOS_DISABLED	(0)
177 
178 #define MIN_KEEP_ALIVE_TIMEOUT_IN_MS	(10000)
179 
180 /* We want to fit submission and completion rings each in a single 2MB
181  * hugepage to ensure physical address contiguity.
182  */
183 #define MAX_IO_QUEUE_ENTRIES		(VALUE_2MB / spdk_max( \
184 						sizeof(struct spdk_nvme_cmd), \
185 						sizeof(struct spdk_nvme_cpl)))
186 
187 /* Default timeout for fabrics connect commands. */
188 #ifdef DEBUG
189 #define NVME_FABRIC_CONNECT_COMMAND_TIMEOUT 0
190 #else
191 /* 500 millisecond timeout. */
192 #define NVME_FABRIC_CONNECT_COMMAND_TIMEOUT 500000
193 #endif
194 
195 /* This value indicates that a read from a PCIe register is invalid. This can happen when a device is no longer present */
196 #define SPDK_NVME_INVALID_REGISTER_VALUE 0xFFFFFFFFu
197 
198 enum nvme_payload_type {
199 	NVME_PAYLOAD_TYPE_INVALID = 0,
200 
201 	/** nvme_request::u.payload.contig_buffer is valid for this request */
202 	NVME_PAYLOAD_TYPE_CONTIG,
203 
204 	/** nvme_request::u.sgl is valid for this request */
205 	NVME_PAYLOAD_TYPE_SGL,
206 };
207 
208 /** Boot partition write states */
209 enum nvme_bp_write_state {
210 	SPDK_NVME_BP_WS_DOWNLOADING	= 0x0,
211 	SPDK_NVME_BP_WS_DOWNLOADED	= 0x1,
212 	SPDK_NVME_BP_WS_REPLACE		= 0x2,
213 	SPDK_NVME_BP_WS_ACTIVATE	= 0x3,
214 };
215 
216 /**
217  * Descriptor for a request data payload.
218  */
219 struct nvme_payload {
220 	/**
221 	 * Functions for retrieving physical addresses for scattered payloads.
222 	 */
223 	spdk_nvme_req_reset_sgl_cb reset_sgl_fn;
224 	spdk_nvme_req_next_sge_cb next_sge_fn;
225 
226 	/**
227 	 * Extended IO options passed by the user
228 	 */
229 	struct spdk_nvme_ns_cmd_ext_io_opts *opts;
230 	/**
231 	 * If reset_sgl_fn == NULL, this is a contig payload, and contig_or_cb_arg contains the
232 	 * virtual memory address of a single virtually contiguous buffer.
233 	 *
234 	 * If reset_sgl_fn != NULL, this is a SGL payload, and contig_or_cb_arg contains the
235 	 * cb_arg that will be passed to the SGL callback functions.
236 	 */
237 	void *contig_or_cb_arg;
238 
239 	/** Virtual memory address of a single virtually contiguous metadata buffer */
240 	void *md;
241 };
242 
243 #define NVME_PAYLOAD_CONTIG(contig_, md_) \
244 	(struct nvme_payload) { \
245 		.reset_sgl_fn = NULL, \
246 		.next_sge_fn = NULL, \
247 		.contig_or_cb_arg = (contig_), \
248 		.md = (md_), \
249 	}
250 
251 #define NVME_PAYLOAD_SGL(reset_sgl_fn_, next_sge_fn_, cb_arg_, md_) \
252 	(struct nvme_payload) { \
253 		.reset_sgl_fn = (reset_sgl_fn_), \
254 		.next_sge_fn = (next_sge_fn_), \
255 		.contig_or_cb_arg = (cb_arg_), \
256 		.md = (md_), \
257 	}
258 
259 static inline enum nvme_payload_type
260 nvme_payload_type(const struct nvme_payload *payload) {
261 	return payload->reset_sgl_fn ? NVME_PAYLOAD_TYPE_SGL : NVME_PAYLOAD_TYPE_CONTIG;
262 }
263 
264 struct nvme_error_cmd {
265 	bool				do_not_submit;
266 	uint64_t			timeout_tsc;
267 	uint32_t			err_count;
268 	uint8_t				opc;
269 	struct spdk_nvme_status		status;
270 	TAILQ_ENTRY(nvme_error_cmd)	link;
271 };
272 
273 struct nvme_request {
274 	struct spdk_nvme_cmd		cmd;
275 
276 	uint8_t				retries;
277 
278 	uint8_t				timed_out : 1;
279 
280 	/**
281 	 * True if the request is in the queued_req list.
282 	 */
283 	uint8_t				queued : 1;
284 	uint8_t				reserved : 6;
285 
286 	/**
287 	 * Number of children requests still outstanding for this
288 	 *  request which was split into multiple child requests.
289 	 */
290 	uint16_t			num_children;
291 
292 	/**
293 	 * Offset in bytes from the beginning of payload for this request.
294 	 * This is used for I/O commands that are split into multiple requests.
295 	 */
296 	uint32_t			payload_offset;
297 	uint32_t			md_offset;
298 
299 	uint32_t			payload_size;
300 
301 	/**
302 	 * Timeout ticks for error injection requests, can be extended in future
303 	 * to support per-request timeout feature.
304 	 */
305 	uint64_t			timeout_tsc;
306 
307 	/**
308 	 * Data payload for this request's command.
309 	 */
310 	struct nvme_payload		payload;
311 
312 	spdk_nvme_cmd_cb		cb_fn;
313 	void				*cb_arg;
314 	STAILQ_ENTRY(nvme_request)	stailq;
315 
316 	struct spdk_nvme_qpair		*qpair;
317 
318 	/*
319 	 * The value of spdk_get_ticks() when the request was submitted to the hardware.
320 	 * Only set if ctrlr->timeout_enabled is true.
321 	 */
322 	uint64_t			submit_tick;
323 
324 	/**
325 	 * The active admin request can be moved to a per process pending
326 	 *  list based on the saved pid to tell which process it belongs
327 	 *  to. The cpl saves the original completion information which
328 	 *  is used in the completion callback.
329 	 * NOTE: these below two fields are only used for admin request.
330 	 */
331 	pid_t				pid;
332 	struct spdk_nvme_cpl		cpl;
333 
334 	uint32_t			md_size;
335 
336 	/**
337 	 * The following members should not be reordered with members
338 	 *  above.  These members are only needed when splitting
339 	 *  requests which is done rarely, and the driver is careful
340 	 *  to not touch the following fields until a split operation is
341 	 *  needed, to avoid touching an extra cacheline.
342 	 */
343 
344 	/**
345 	 * Points to the outstanding child requests for a parent request.
346 	 *  Only valid if a request was split into multiple children
347 	 *  requests, and is not initialized for non-split requests.
348 	 */
349 	TAILQ_HEAD(, nvme_request)	children;
350 
351 	/**
352 	 * Linked-list pointers for a child request in its parent's list.
353 	 */
354 	TAILQ_ENTRY(nvme_request)	child_tailq;
355 
356 	/**
357 	 * Points to a parent request if part of a split request,
358 	 *   NULL otherwise.
359 	 */
360 	struct nvme_request		*parent;
361 
362 	/**
363 	 * Completion status for a parent request.  Initialized to all 0's
364 	 *  (SUCCESS) before child requests are submitted.  If a child
365 	 *  request completes with error, the error status is copied here,
366 	 *  to ensure that the parent request is also completed with error
367 	 *  status once all child requests are completed.
368 	 */
369 	struct spdk_nvme_cpl		parent_status;
370 
371 	/**
372 	 * The user_cb_fn and user_cb_arg fields are used for holding the original
373 	 * callback data when using nvme_allocate_request_user_copy.
374 	 */
375 	spdk_nvme_cmd_cb		user_cb_fn;
376 	void				*user_cb_arg;
377 	void				*user_buffer;
378 
379 	/** Sequence of accel operations associated with this request */
380 	void				*accel_sequence;
381 };
382 
383 struct nvme_completion_poll_status {
384 	struct spdk_nvme_cpl	cpl;
385 	uint64_t		timeout_tsc;
386 	/**
387 	 * DMA buffer retained throughout the duration of the command.  It'll be released
388 	 * automatically if the command times out, otherwise the user is responsible for freeing it.
389 	 */
390 	void			*dma_data;
391 	bool			done;
392 	/* This flag indicates that the request has been timed out and the memory
393 	   must be freed in a completion callback */
394 	bool			timed_out;
395 };
396 
397 struct nvme_async_event_request {
398 	struct spdk_nvme_ctrlr		*ctrlr;
399 	struct nvme_request		*req;
400 	struct spdk_nvme_cpl		cpl;
401 };
402 
403 enum nvme_qpair_state {
404 	NVME_QPAIR_DISCONNECTED,
405 	NVME_QPAIR_DISCONNECTING,
406 	NVME_QPAIR_CONNECTING,
407 	NVME_QPAIR_CONNECTED,
408 	NVME_QPAIR_ENABLING,
409 	NVME_QPAIR_ENABLED,
410 	NVME_QPAIR_DESTROYING,
411 };
412 
413 enum nvme_qpair_auth_state {
414 	NVME_QPAIR_AUTH_STATE_NEGOTIATE,
415 	NVME_QPAIR_AUTH_STATE_AWAIT_NEGOTIATE,
416 	NVME_QPAIR_AUTH_STATE_AWAIT_CHALLENGE,
417 	NVME_QPAIR_AUTH_STATE_AWAIT_REPLY,
418 	NVME_QPAIR_AUTH_STATE_AWAIT_SUCCESS1,
419 	NVME_QPAIR_AUTH_STATE_AWAIT_SUCCESS2,
420 	NVME_QPAIR_AUTH_STATE_AWAIT_FAILURE2,
421 	NVME_QPAIR_AUTH_STATE_DONE,
422 };
423 
424 /* Authentication transaction required (authreq.atr) */
425 #define NVME_QPAIR_AUTH_FLAG_ATR	(1 << 0)
426 /* Authentication and secure channel required (authreq.ascr) */
427 #define NVME_QPAIR_AUTH_FLAG_ASCR	(1 << 1)
428 
429 /* Maximum size of a digest */
430 #define NVME_AUTH_DIGEST_MAX_SIZE	64
431 
432 struct nvme_auth {
433 	/* State of the authentication */
434 	enum nvme_qpair_auth_state	state;
435 	/* Status of the authentication */
436 	int				status;
437 	/* Transaction ID */
438 	uint16_t			tid;
439 	/* Flags */
440 	uint32_t			flags;
441 	/* Selected hash function */
442 	uint8_t				hash;
443 	/* Buffer used for controller challenge */
444 	uint8_t				challenge[NVME_AUTH_DIGEST_MAX_SIZE];
445 	/* User's auth cb fn/ctx */
446 	spdk_nvme_authenticate_cb	cb_fn;
447 	void				*cb_ctx;
448 };
449 
450 struct spdk_nvme_qpair {
451 	struct spdk_nvme_ctrlr			*ctrlr;
452 
453 	uint16_t				id;
454 
455 	uint8_t					qprio: 2;
456 
457 	uint8_t					state: 3;
458 
459 	uint8_t					async: 1;
460 
461 	uint8_t					is_new_qpair: 1;
462 
463 	uint8_t					abort_dnr: 1;
464 	/*
465 	 * Members for handling IO qpair deletion inside of a completion context.
466 	 * These are specifically defined as single bits, so that they do not
467 	 *  push this data structure out to another cacheline.
468 	 */
469 	uint8_t					in_completion_context: 1;
470 	uint8_t					delete_after_completion_context: 1;
471 
472 	/*
473 	 * Set when no deletion notification is needed. For example, the process
474 	 * which allocated this qpair exited unexpectedly.
475 	 */
476 	uint8_t					no_deletion_notification_needed: 1;
477 
478 	uint8_t					last_fuse: 2;
479 
480 	uint8_t					transport_failure_reason: 3;
481 	uint8_t					last_transport_failure_reason: 3;
482 
483 	/* The user is destroying qpair */
484 	uint8_t					destroy_in_progress: 1;
485 
486 	/* Number of IO outstanding at transport level */
487 	uint16_t				queue_depth;
488 
489 	enum spdk_nvme_transport_type		trtype;
490 
491 	uint32_t				num_outstanding_reqs;
492 
493 	/* request object used only for this qpair's FABRICS/CONNECT command (if needed) */
494 	struct nvme_request			*reserved_req;
495 
496 	STAILQ_HEAD(, nvme_request)		free_req;
497 	STAILQ_HEAD(, nvme_request)		queued_req;
498 
499 	/* List entry for spdk_nvme_transport_poll_group::qpairs */
500 	STAILQ_ENTRY(spdk_nvme_qpair)		poll_group_stailq;
501 
502 	/** Commands opcode in this list will return error */
503 	TAILQ_HEAD(, nvme_error_cmd)		err_cmd_head;
504 	/** Requests in this list will return error */
505 	STAILQ_HEAD(, nvme_request)		err_req_head;
506 
507 	struct spdk_nvme_ctrlr_process		*active_proc;
508 
509 	struct spdk_nvme_transport_poll_group	*poll_group;
510 
511 	void					*poll_group_tailq_head;
512 
513 	const struct spdk_nvme_transport	*transport;
514 
515 	/* Entries below here are not touched in the main I/O path. */
516 
517 	struct nvme_completion_poll_status	*poll_status;
518 
519 	/* List entry for spdk_nvme_ctrlr::active_io_qpairs */
520 	TAILQ_ENTRY(spdk_nvme_qpair)		tailq;
521 
522 	/* List entry for spdk_nvme_ctrlr_process::allocated_io_qpairs */
523 	TAILQ_ENTRY(spdk_nvme_qpair)		per_process_tailq;
524 
525 	STAILQ_HEAD(, nvme_request)		aborting_queued_req;
526 
527 	void					*req_buf;
528 
529 	/* In-band authentication state */
530 	struct nvme_auth			auth;
531 };
532 
533 struct spdk_nvme_poll_group {
534 	void						*ctx;
535 	struct spdk_nvme_accel_fn_table			accel_fn_table;
536 	STAILQ_HEAD(, spdk_nvme_transport_poll_group)	tgroups;
537 	bool						in_process_completions;
538 };
539 
540 struct spdk_nvme_transport_poll_group {
541 	struct spdk_nvme_poll_group			*group;
542 	const struct spdk_nvme_transport		*transport;
543 	STAILQ_HEAD(, spdk_nvme_qpair)			connected_qpairs;
544 	STAILQ_HEAD(, spdk_nvme_qpair)			disconnected_qpairs;
545 	STAILQ_ENTRY(spdk_nvme_transport_poll_group)	link;
546 	uint32_t					num_connected_qpairs;
547 };
548 
549 struct spdk_nvme_ns {
550 	struct spdk_nvme_ctrlr		*ctrlr;
551 	uint32_t			sector_size;
552 
553 	/*
554 	 * Size of data transferred as part of each block,
555 	 * including metadata if FLBAS indicates the metadata is transferred
556 	 * as part of the data buffer at the end of each LBA.
557 	 */
558 	uint32_t			extended_lba_size;
559 
560 	uint32_t			md_size;
561 	uint32_t			pi_type;
562 	uint32_t			pi_format;
563 	uint32_t			sectors_per_max_io;
564 	uint32_t			sectors_per_max_io_no_md;
565 	uint32_t			sectors_per_stripe;
566 	uint32_t			id;
567 	uint16_t			flags;
568 	bool				active;
569 
570 	/* Command Set Identifier */
571 	enum spdk_nvme_csi		csi;
572 
573 	/* Namespace Identification Descriptor List (CNS = 03h) */
574 	uint8_t				id_desc_list[4096];
575 
576 	uint32_t			ana_group_id;
577 	enum spdk_nvme_ana_state	ana_state;
578 
579 	/* Identify Namespace data. */
580 	struct spdk_nvme_ns_data	nsdata;
581 
582 	/* Zoned Namespace Command Set Specific Identify Namespace data. */
583 	struct spdk_nvme_zns_ns_data	*nsdata_zns;
584 
585 	struct spdk_nvme_nvm_ns_data	*nsdata_nvm;
586 
587 	RB_ENTRY(spdk_nvme_ns)		node;
588 };
589 
590 /**
591  * State of struct spdk_nvme_ctrlr (in particular, during initialization).
592  */
593 enum nvme_ctrlr_state {
594 	/**
595 	 * Wait before initializing the controller.
596 	 */
597 	NVME_CTRLR_STATE_INIT_DELAY,
598 
599 	/**
600 	 * Connect the admin queue.
601 	 */
602 	NVME_CTRLR_STATE_CONNECT_ADMINQ,
603 
604 	/**
605 	 * Controller has not started initialized yet.
606 	 */
607 	NVME_CTRLR_STATE_INIT = NVME_CTRLR_STATE_CONNECT_ADMINQ,
608 
609 	/**
610 	 * Waiting for admin queue to connect.
611 	 */
612 	NVME_CTRLR_STATE_WAIT_FOR_CONNECT_ADMINQ,
613 
614 	/**
615 	 * Read Version (VS) register.
616 	 */
617 	NVME_CTRLR_STATE_READ_VS,
618 
619 	/**
620 	 * Waiting for Version (VS) register to be read.
621 	 */
622 	NVME_CTRLR_STATE_READ_VS_WAIT_FOR_VS,
623 
624 	/**
625 	 * Read Capabilities (CAP) register.
626 	 */
627 	NVME_CTRLR_STATE_READ_CAP,
628 
629 	/**
630 	 * Waiting for Capabilities (CAP) register to be read.
631 	 */
632 	NVME_CTRLR_STATE_READ_CAP_WAIT_FOR_CAP,
633 
634 	/**
635 	 * Check EN to prepare for controller initialization.
636 	 */
637 	NVME_CTRLR_STATE_CHECK_EN,
638 
639 	/**
640 	 * Waiting for CC to be read as part of EN check.
641 	 */
642 	NVME_CTRLR_STATE_CHECK_EN_WAIT_FOR_CC,
643 
644 	/**
645 	 * Waiting for CSTS.RDY to transition from 0 to 1 so that CC.EN may be set to 0.
646 	 */
647 	NVME_CTRLR_STATE_DISABLE_WAIT_FOR_READY_1,
648 
649 	/**
650 	 * Waiting for CSTS register to be read as part of waiting for CSTS.RDY = 1.
651 	 */
652 	NVME_CTRLR_STATE_DISABLE_WAIT_FOR_READY_1_WAIT_FOR_CSTS,
653 
654 	/**
655 	 * Disabling the controller by setting CC.EN to 0.
656 	 */
657 	NVME_CTRLR_STATE_SET_EN_0,
658 
659 	/**
660 	 * Waiting for the CC register to be read as part of disabling the controller.
661 	 */
662 	NVME_CTRLR_STATE_SET_EN_0_WAIT_FOR_CC,
663 
664 	/**
665 	 * Waiting for CSTS.RDY to transition from 1 to 0 so that CC.EN may be set to 1.
666 	 */
667 	NVME_CTRLR_STATE_DISABLE_WAIT_FOR_READY_0,
668 
669 	/**
670 	 * Waiting for CSTS register to be read as part of waiting for CSTS.RDY = 0.
671 	 */
672 	NVME_CTRLR_STATE_DISABLE_WAIT_FOR_READY_0_WAIT_FOR_CSTS,
673 
674 	/**
675 	 * The controller is disabled. (CC.EN and CSTS.RDY are 0.)
676 	 */
677 	NVME_CTRLR_STATE_DISABLED,
678 
679 	/**
680 	 * Enable the controller by writing CC.EN to 1
681 	 */
682 	NVME_CTRLR_STATE_ENABLE,
683 
684 	/**
685 	 * Waiting for CC register to be written as part of enabling the controller.
686 	 */
687 	NVME_CTRLR_STATE_ENABLE_WAIT_FOR_CC,
688 
689 	/**
690 	 * Waiting for CSTS.RDY to transition from 0 to 1 after enabling the controller.
691 	 */
692 	NVME_CTRLR_STATE_ENABLE_WAIT_FOR_READY_1,
693 
694 	/**
695 	 * Waiting for CSTS register to be read as part of waiting for CSTS.RDY = 1.
696 	 */
697 	NVME_CTRLR_STATE_ENABLE_WAIT_FOR_READY_1_WAIT_FOR_CSTS,
698 
699 	/**
700 	 * Reset the Admin queue of the controller.
701 	 */
702 	NVME_CTRLR_STATE_RESET_ADMIN_QUEUE,
703 
704 	/**
705 	 * Identify Controller command will be sent to then controller.
706 	 */
707 	NVME_CTRLR_STATE_IDENTIFY,
708 
709 	/**
710 	 * Waiting for Identify Controller command be completed.
711 	 */
712 	NVME_CTRLR_STATE_WAIT_FOR_IDENTIFY,
713 
714 	/**
715 	 * Configure AER of the controller.
716 	 */
717 	NVME_CTRLR_STATE_CONFIGURE_AER,
718 
719 	/**
720 	 * Waiting for the Configure AER to be completed.
721 	 */
722 	NVME_CTRLR_STATE_WAIT_FOR_CONFIGURE_AER,
723 
724 	/**
725 	 * Set Keep Alive Timeout of the controller.
726 	 */
727 	NVME_CTRLR_STATE_SET_KEEP_ALIVE_TIMEOUT,
728 
729 	/**
730 	 * Waiting for Set Keep Alive Timeout to be completed.
731 	 */
732 	NVME_CTRLR_STATE_WAIT_FOR_KEEP_ALIVE_TIMEOUT,
733 
734 	/**
735 	 * Get Identify I/O Command Set Specific Controller data structure.
736 	 */
737 	NVME_CTRLR_STATE_IDENTIFY_IOCS_SPECIFIC,
738 
739 	/**
740 	 * Waiting for Identify I/O Command Set Specific Controller command to be completed.
741 	 */
742 	NVME_CTRLR_STATE_WAIT_FOR_IDENTIFY_IOCS_SPECIFIC,
743 
744 	/**
745 	 * Get Commands Supported and Effects log page for the Zoned Namespace Command Set.
746 	 */
747 	NVME_CTRLR_STATE_GET_ZNS_CMD_EFFECTS_LOG,
748 
749 	/**
750 	 * Waiting for the Get Log Page command to be completed.
751 	 */
752 	NVME_CTRLR_STATE_WAIT_FOR_GET_ZNS_CMD_EFFECTS_LOG,
753 
754 	/**
755 	 * Set Number of Queues of the controller.
756 	 */
757 	NVME_CTRLR_STATE_SET_NUM_QUEUES,
758 
759 	/**
760 	 * Waiting for Set Num of Queues command to be completed.
761 	 */
762 	NVME_CTRLR_STATE_WAIT_FOR_SET_NUM_QUEUES,
763 
764 	/**
765 	 * Get active Namespace list of the controller.
766 	 */
767 	NVME_CTRLR_STATE_IDENTIFY_ACTIVE_NS,
768 
769 	/**
770 	 * Waiting for the Identify Active Namespace commands to be completed.
771 	 */
772 	NVME_CTRLR_STATE_WAIT_FOR_IDENTIFY_ACTIVE_NS,
773 
774 	/**
775 	 * Get Identify Namespace Data structure for each NS.
776 	 */
777 	NVME_CTRLR_STATE_IDENTIFY_NS,
778 
779 	/**
780 	 * Waiting for the Identify Namespace commands to be completed.
781 	 */
782 	NVME_CTRLR_STATE_WAIT_FOR_IDENTIFY_NS,
783 
784 	/**
785 	 * Get Identify Namespace Identification Descriptors.
786 	 */
787 	NVME_CTRLR_STATE_IDENTIFY_ID_DESCS,
788 
789 	/**
790 	 * Get Identify I/O Command Set Specific Namespace data structure for each NS.
791 	 */
792 	NVME_CTRLR_STATE_IDENTIFY_NS_IOCS_SPECIFIC,
793 
794 	/**
795 	 * Waiting for the Identify I/O Command Set Specific Namespace commands to be completed.
796 	 */
797 	NVME_CTRLR_STATE_WAIT_FOR_IDENTIFY_NS_IOCS_SPECIFIC,
798 
799 	/**
800 	 * Waiting for the Identify Namespace Identification
801 	 * Descriptors to be completed.
802 	 */
803 	NVME_CTRLR_STATE_WAIT_FOR_IDENTIFY_ID_DESCS,
804 
805 	/**
806 	 * Set supported log pages of the controller.
807 	 */
808 	NVME_CTRLR_STATE_SET_SUPPORTED_LOG_PAGES,
809 
810 	/**
811 	 * Set supported log pages of INTEL controller.
812 	 */
813 	NVME_CTRLR_STATE_SET_SUPPORTED_INTEL_LOG_PAGES,
814 
815 	/**
816 	 * Waiting for supported log pages of INTEL controller.
817 	 */
818 	NVME_CTRLR_STATE_WAIT_FOR_SUPPORTED_INTEL_LOG_PAGES,
819 
820 	/**
821 	 * Set supported features of the controller.
822 	 */
823 	NVME_CTRLR_STATE_SET_SUPPORTED_FEATURES,
824 
825 	/**
826 	 * Set the Host Behavior Support feature of the controller.
827 	 */
828 	NVME_CTRLR_STATE_SET_HOST_FEATURE,
829 
830 	/**
831 	 * Waiting for the Host Behavior Support feature of the controller.
832 	 */
833 	NVME_CTRLR_STATE_WAIT_FOR_SET_HOST_FEATURE,
834 
835 	/**
836 	 * Set Doorbell Buffer Config of the controller.
837 	 */
838 	NVME_CTRLR_STATE_SET_DB_BUF_CFG,
839 
840 	/**
841 	 * Waiting for Doorbell Buffer Config to be completed.
842 	 */
843 	NVME_CTRLR_STATE_WAIT_FOR_DB_BUF_CFG,
844 
845 	/**
846 	 * Set Host ID of the controller.
847 	 */
848 	NVME_CTRLR_STATE_SET_HOST_ID,
849 
850 	/**
851 	 * Waiting for Set Host ID to be completed.
852 	 */
853 	NVME_CTRLR_STATE_WAIT_FOR_HOST_ID,
854 
855 	/**
856 	 * Let transport layer do its part of initialization.
857 	 */
858 	NVME_CTRLR_STATE_TRANSPORT_READY,
859 
860 	/**
861 	 * Controller initialization has completed and the controller is ready.
862 	 */
863 	NVME_CTRLR_STATE_READY,
864 
865 	/**
866 	 * Controller initialization has an error.
867 	 */
868 	NVME_CTRLR_STATE_ERROR,
869 
870 	/**
871 	 * Admin qpair was disconnected, controller needs to be re-initialized
872 	 */
873 	NVME_CTRLR_STATE_DISCONNECTED,
874 };
875 
876 #define NVME_TIMEOUT_INFINITE		0
877 #define NVME_TIMEOUT_KEEP_EXISTING	UINT64_MAX
878 
879 struct spdk_nvme_ctrlr_aer_completion_list {
880 	struct spdk_nvme_cpl	cpl;
881 	STAILQ_ENTRY(spdk_nvme_ctrlr_aer_completion_list) link;
882 };
883 
884 /*
885  * Used to track properties for all processes accessing the controller.
886  */
887 struct spdk_nvme_ctrlr_process {
888 	/** Whether it is the primary process  */
889 	bool						is_primary;
890 
891 	/** Process ID */
892 	pid_t						pid;
893 
894 	/** Active admin requests to be completed */
895 	STAILQ_HEAD(, nvme_request)			active_reqs;
896 
897 	TAILQ_ENTRY(spdk_nvme_ctrlr_process)		tailq;
898 
899 	/** Per process PCI device handle */
900 	struct spdk_pci_device				*devhandle;
901 
902 	/** Reference to track the number of attachment to this controller. */
903 	int						ref;
904 
905 	/** Allocated IO qpairs */
906 	TAILQ_HEAD(, spdk_nvme_qpair)			allocated_io_qpairs;
907 
908 	spdk_nvme_aer_cb				aer_cb_fn;
909 	void						*aer_cb_arg;
910 
911 	/**
912 	 * A function pointer to timeout callback function
913 	 */
914 	spdk_nvme_timeout_cb		timeout_cb_fn;
915 	void				*timeout_cb_arg;
916 	/** separate timeout values for io vs. admin reqs */
917 	uint64_t			timeout_io_ticks;
918 	uint64_t			timeout_admin_ticks;
919 
920 	/** List to publish AENs to all procs in multiprocess setup */
921 	STAILQ_HEAD(, spdk_nvme_ctrlr_aer_completion_list)      async_events;
922 };
923 
924 struct nvme_register_completion {
925 	struct spdk_nvme_cpl			cpl;
926 	uint64_t				value;
927 	spdk_nvme_reg_cb			cb_fn;
928 	void					*cb_ctx;
929 	STAILQ_ENTRY(nvme_register_completion)	stailq;
930 	pid_t					pid;
931 };
932 
933 struct spdk_nvme_ctrlr {
934 	/* Hot data (accessed in I/O path) starts here. */
935 
936 	/* Tree of namespaces */
937 	RB_HEAD(nvme_ns_tree, spdk_nvme_ns)	ns;
938 
939 	/* The number of active namespaces */
940 	uint32_t			active_ns_count;
941 
942 	bool				is_removed;
943 
944 	bool				is_resetting;
945 
946 	bool				is_failed;
947 
948 	bool				is_destructed;
949 
950 	bool				timeout_enabled;
951 
952 	/* The application is preparing to reset the controller.  Transports
953 	 * can use this to skip unnecessary parts of the qpair deletion process
954 	 * for example, like the DELETE_SQ/CQ commands.
955 	 */
956 	bool				prepare_for_reset;
957 
958 	bool				is_disconnecting;
959 
960 	bool				needs_io_msg_update;
961 
962 	uint16_t			max_sges;
963 
964 	uint16_t			cntlid;
965 
966 	/** Controller support flags */
967 	uint64_t			flags;
968 
969 	/** NVMEoF in-capsule data size in bytes */
970 	uint32_t			ioccsz_bytes;
971 
972 	/** NVMEoF in-capsule data offset in 16 byte units */
973 	uint16_t			icdoff;
974 
975 	/* Cold data (not accessed in normal I/O path) is after this point. */
976 
977 	struct spdk_nvme_transport_id	trid;
978 
979 	struct {
980 		/** Is numa.id valid? Ensures numa.id == 0 is interpreted correctly. */
981 		uint32_t		id_valid : 1;
982 		int32_t			id : 31;
983 	} numa;
984 
985 	union spdk_nvme_cap_register	cap;
986 	union spdk_nvme_vs_register	vs;
987 
988 	int				state;
989 	uint64_t			state_timeout_tsc;
990 
991 	uint64_t			next_keep_alive_tick;
992 	uint64_t			keep_alive_interval_ticks;
993 
994 	TAILQ_ENTRY(spdk_nvme_ctrlr)	tailq;
995 
996 	/** All the log pages supported */
997 	bool				log_page_supported[256];
998 
999 	/** All the features supported */
1000 	bool				feature_supported[256];
1001 
1002 	/** maximum i/o size in bytes */
1003 	uint32_t			max_xfer_size;
1004 
1005 	/** minimum page size supported by this controller in bytes */
1006 	uint32_t			min_page_size;
1007 
1008 	/** selected memory page size for this controller in bytes */
1009 	uint32_t			page_size;
1010 
1011 	uint32_t			num_aers;
1012 	struct nvme_async_event_request	aer[NVME_MAX_ASYNC_EVENTS];
1013 
1014 	/** guards access to the controller itself, including admin queues */
1015 	pthread_mutex_t			ctrlr_lock;
1016 
1017 	struct spdk_nvme_qpair		*adminq;
1018 
1019 	/** shadow doorbell buffer */
1020 	uint32_t			*shadow_doorbell;
1021 	/** eventidx buffer */
1022 	uint32_t			*eventidx;
1023 
1024 	/**
1025 	 * Identify Controller data.
1026 	 */
1027 	struct spdk_nvme_ctrlr_data	cdata;
1028 
1029 	/**
1030 	 * Zoned Namespace Command Set Specific Identify Controller data.
1031 	 */
1032 	struct spdk_nvme_zns_ctrlr_data	*cdata_zns;
1033 
1034 	struct spdk_bit_array		*free_io_qids;
1035 	TAILQ_HEAD(, spdk_nvme_qpair)	active_io_qpairs;
1036 
1037 	struct spdk_nvme_ctrlr_opts	opts;
1038 
1039 	uint64_t			quirks;
1040 
1041 	/* Extra sleep time during controller initialization */
1042 	uint64_t			sleep_timeout_tsc;
1043 
1044 	/** Track all the processes manage this controller */
1045 	TAILQ_HEAD(, spdk_nvme_ctrlr_process)	active_procs;
1046 
1047 
1048 	STAILQ_HEAD(, nvme_request)	queued_aborts;
1049 	uint32_t			outstanding_aborts;
1050 
1051 	uint32_t			lock_depth;
1052 
1053 	/* CB to notify the user when the ctrlr is removed/failed. */
1054 	spdk_nvme_remove_cb			remove_cb;
1055 	void					*cb_ctx;
1056 
1057 	struct spdk_nvme_qpair		*external_io_msgs_qpair;
1058 	pthread_mutex_t			external_io_msgs_lock;
1059 	struct spdk_ring		*external_io_msgs;
1060 
1061 	STAILQ_HEAD(, nvme_io_msg_producer) io_producers;
1062 
1063 	struct spdk_nvme_ana_page		*ana_log_page;
1064 	struct spdk_nvme_ana_group_descriptor	*copied_ana_desc;
1065 	uint32_t				ana_log_page_size;
1066 
1067 	/* scratchpad pointer that can be used to send data between two NVME_CTRLR_STATEs */
1068 	void				*tmp_ptr;
1069 
1070 	/* maximum zone append size in bytes */
1071 	uint32_t			max_zone_append_size;
1072 
1073 	/* PMR size in bytes */
1074 	uint64_t			pmr_size;
1075 
1076 	/* Boot Partition Info */
1077 	enum nvme_bp_write_state	bp_ws;
1078 	uint32_t			bpid;
1079 	spdk_nvme_cmd_cb		bp_write_cb_fn;
1080 	void				*bp_write_cb_arg;
1081 
1082 	/* Firmware Download */
1083 	void				*fw_payload;
1084 	unsigned int			fw_size_remaining;
1085 	unsigned int			fw_offset;
1086 	unsigned int			fw_transfer_size;
1087 
1088 	/* Completed register operations */
1089 	STAILQ_HEAD(, nvme_register_completion)	register_operations;
1090 
1091 	union spdk_nvme_cc_register		process_init_cc;
1092 
1093 	/* Authentication transaction ID */
1094 	uint16_t				auth_tid;
1095 	/* Authentication sequence number */
1096 	uint32_t				auth_seqnum;
1097 };
1098 
1099 struct spdk_nvme_probe_ctx {
1100 	struct spdk_nvme_transport_id		trid;
1101 	const struct spdk_nvme_ctrlr_opts	*opts;
1102 	void					*cb_ctx;
1103 	spdk_nvme_probe_cb			probe_cb;
1104 	spdk_nvme_attach_cb			attach_cb;
1105 	spdk_nvme_attach_fail_cb		attach_fail_cb;
1106 	spdk_nvme_remove_cb			remove_cb;
1107 	TAILQ_HEAD(, spdk_nvme_ctrlr)		init_ctrlrs;
1108 };
1109 
1110 typedef void (*nvme_ctrlr_detach_cb)(struct spdk_nvme_ctrlr *ctrlr);
1111 
1112 enum nvme_ctrlr_detach_state {
1113 	NVME_CTRLR_DETACH_SET_CC,
1114 	NVME_CTRLR_DETACH_CHECK_CSTS,
1115 	NVME_CTRLR_DETACH_GET_CSTS,
1116 	NVME_CTRLR_DETACH_GET_CSTS_DONE,
1117 };
1118 
1119 struct nvme_ctrlr_detach_ctx {
1120 	struct spdk_nvme_ctrlr			*ctrlr;
1121 	nvme_ctrlr_detach_cb			cb_fn;
1122 	uint64_t				shutdown_start_tsc;
1123 	uint32_t				shutdown_timeout_ms;
1124 	bool					shutdown_complete;
1125 	enum nvme_ctrlr_detach_state		state;
1126 	union spdk_nvme_csts_register		csts;
1127 	TAILQ_ENTRY(nvme_ctrlr_detach_ctx)	link;
1128 };
1129 
1130 struct spdk_nvme_detach_ctx {
1131 	TAILQ_HEAD(, nvme_ctrlr_detach_ctx)	head;
1132 };
1133 
1134 struct nvme_driver {
1135 	pthread_mutex_t			lock;
1136 
1137 	/** Multi-process shared attached controller list */
1138 	TAILQ_HEAD(, spdk_nvme_ctrlr)	shared_attached_ctrlrs;
1139 
1140 	bool				initialized;
1141 	struct spdk_uuid		default_extended_host_id;
1142 
1143 	/** netlink socket fd for hotplug messages */
1144 	int				hotplug_fd;
1145 };
1146 
1147 #define nvme_ns_cmd_get_ext_io_opt(opts, field, defval) \
1148        ((opts) != NULL && offsetof(struct spdk_nvme_ns_cmd_ext_io_opts, field) + \
1149         sizeof((opts)->field) <= (opts)->size ? (opts)->field : (defval))
1150 
1151 extern struct nvme_driver *g_spdk_nvme_driver;
1152 
1153 int nvme_driver_init(void);
1154 
1155 #define nvme_delay		usleep
1156 
1157 static inline bool
1158 nvme_qpair_is_admin_queue(struct spdk_nvme_qpair *qpair)
1159 {
1160 	return qpair->id == 0;
1161 }
1162 
1163 static inline bool
1164 nvme_qpair_is_io_queue(struct spdk_nvme_qpair *qpair)
1165 {
1166 	return qpair->id != 0;
1167 }
1168 
1169 static inline int
1170 nvme_robust_mutex_lock(pthread_mutex_t *mtx)
1171 {
1172 	int rc = pthread_mutex_lock(mtx);
1173 
1174 #ifndef __FreeBSD__
1175 	if (rc == EOWNERDEAD) {
1176 		rc = pthread_mutex_consistent(mtx);
1177 	}
1178 #endif
1179 
1180 	return rc;
1181 }
1182 
1183 static inline int
1184 nvme_ctrlr_lock(struct spdk_nvme_ctrlr *ctrlr)
1185 {
1186 	int rc;
1187 
1188 	rc = nvme_robust_mutex_lock(&ctrlr->ctrlr_lock);
1189 	ctrlr->lock_depth++;
1190 	return rc;
1191 }
1192 
1193 static inline int
1194 nvme_robust_mutex_unlock(pthread_mutex_t *mtx)
1195 {
1196 	return pthread_mutex_unlock(mtx);
1197 }
1198 
1199 static inline int
1200 nvme_ctrlr_unlock(struct spdk_nvme_ctrlr *ctrlr)
1201 {
1202 	ctrlr->lock_depth--;
1203 	return nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
1204 }
1205 
1206 /* Poll group management functions. */
1207 int nvme_poll_group_connect_qpair(struct spdk_nvme_qpair *qpair);
1208 int nvme_poll_group_disconnect_qpair(struct spdk_nvme_qpair *qpair);
1209 
1210 /* Admin functions */
1211 int	nvme_ctrlr_cmd_identify(struct spdk_nvme_ctrlr *ctrlr,
1212 				uint8_t cns, uint16_t cntid, uint32_t nsid,
1213 				uint8_t csi, void *payload, size_t payload_size,
1214 				spdk_nvme_cmd_cb cb_fn, void *cb_arg);
1215 int	nvme_ctrlr_cmd_set_num_queues(struct spdk_nvme_ctrlr *ctrlr,
1216 				      uint32_t num_queues, spdk_nvme_cmd_cb cb_fn,
1217 				      void *cb_arg);
1218 int	nvme_ctrlr_cmd_get_num_queues(struct spdk_nvme_ctrlr *ctrlr,
1219 				      spdk_nvme_cmd_cb cb_fn, void *cb_arg);
1220 int	nvme_ctrlr_cmd_set_async_event_config(struct spdk_nvme_ctrlr *ctrlr,
1221 		union spdk_nvme_feat_async_event_configuration config,
1222 		spdk_nvme_cmd_cb cb_fn, void *cb_arg);
1223 int	nvme_ctrlr_cmd_set_host_id(struct spdk_nvme_ctrlr *ctrlr, void *host_id, uint32_t host_id_size,
1224 				   spdk_nvme_cmd_cb cb_fn, void *cb_arg);
1225 int	nvme_ctrlr_cmd_attach_ns(struct spdk_nvme_ctrlr *ctrlr, uint32_t nsid,
1226 				 struct spdk_nvme_ctrlr_list *payload, spdk_nvme_cmd_cb cb_fn, void *cb_arg);
1227 int	nvme_ctrlr_cmd_detach_ns(struct spdk_nvme_ctrlr *ctrlr, uint32_t nsid,
1228 				 struct spdk_nvme_ctrlr_list *payload, spdk_nvme_cmd_cb cb_fn, void *cb_arg);
1229 int	nvme_ctrlr_cmd_create_ns(struct spdk_nvme_ctrlr *ctrlr, struct spdk_nvme_ns_data *payload,
1230 				 spdk_nvme_cmd_cb cb_fn, void *cb_arg);
1231 int	nvme_ctrlr_cmd_doorbell_buffer_config(struct spdk_nvme_ctrlr *ctrlr,
1232 		uint64_t prp1, uint64_t prp2,
1233 		spdk_nvme_cmd_cb cb_fn, void *cb_arg);
1234 int	nvme_ctrlr_cmd_delete_ns(struct spdk_nvme_ctrlr *ctrlr, uint32_t nsid, spdk_nvme_cmd_cb cb_fn,
1235 				 void *cb_arg);
1236 int	nvme_ctrlr_cmd_format(struct spdk_nvme_ctrlr *ctrlr, uint32_t nsid,
1237 			      struct spdk_nvme_format *format, spdk_nvme_cmd_cb cb_fn, void *cb_arg);
1238 int	nvme_ctrlr_cmd_fw_commit(struct spdk_nvme_ctrlr *ctrlr,
1239 				 const struct spdk_nvme_fw_commit *fw_commit,
1240 				 spdk_nvme_cmd_cb cb_fn, void *cb_arg);
1241 int	nvme_ctrlr_cmd_fw_image_download(struct spdk_nvme_ctrlr *ctrlr,
1242 		uint32_t size, uint32_t offset, void *payload,
1243 		spdk_nvme_cmd_cb cb_fn, void *cb_arg);
1244 int	nvme_ctrlr_cmd_sanitize(struct spdk_nvme_ctrlr *ctrlr, uint32_t nsid,
1245 				struct spdk_nvme_sanitize *sanitize, uint32_t cdw11,
1246 				spdk_nvme_cmd_cb cb_fn, void *cb_arg);
1247 void	nvme_completion_poll_cb(void *arg, const struct spdk_nvme_cpl *cpl);
1248 int	nvme_wait_for_completion(struct spdk_nvme_qpair *qpair,
1249 				 struct nvme_completion_poll_status *status);
1250 int	nvme_wait_for_completion_robust_lock(struct spdk_nvme_qpair *qpair,
1251 		struct nvme_completion_poll_status *status,
1252 		pthread_mutex_t *robust_mutex);
1253 int	nvme_wait_for_completion_timeout(struct spdk_nvme_qpair *qpair,
1254 		struct nvme_completion_poll_status *status,
1255 		uint64_t timeout_in_usecs);
1256 int	nvme_wait_for_completion_robust_lock_timeout(struct spdk_nvme_qpair *qpair,
1257 		struct nvme_completion_poll_status *status,
1258 		pthread_mutex_t *robust_mutex,
1259 		uint64_t timeout_in_usecs);
1260 int	nvme_wait_for_completion_robust_lock_timeout_poll(struct spdk_nvme_qpair *qpair,
1261 		struct nvme_completion_poll_status *status,
1262 		pthread_mutex_t *robust_mutex);
1263 
1264 struct spdk_nvme_ctrlr_process *nvme_ctrlr_get_process(struct spdk_nvme_ctrlr *ctrlr,
1265 		pid_t pid);
1266 struct spdk_nvme_ctrlr_process *nvme_ctrlr_get_current_process(struct spdk_nvme_ctrlr *ctrlr);
1267 int	nvme_ctrlr_add_process(struct spdk_nvme_ctrlr *ctrlr, void *devhandle);
1268 void	nvme_ctrlr_free_processes(struct spdk_nvme_ctrlr *ctrlr);
1269 struct spdk_pci_device *nvme_ctrlr_proc_get_devhandle(struct spdk_nvme_ctrlr *ctrlr);
1270 
1271 int	nvme_ctrlr_probe(const struct spdk_nvme_transport_id *trid,
1272 			 struct spdk_nvme_probe_ctx *probe_ctx, void *devhandle);
1273 
1274 int	nvme_ctrlr_construct(struct spdk_nvme_ctrlr *ctrlr);
1275 void	nvme_ctrlr_destruct_finish(struct spdk_nvme_ctrlr *ctrlr);
1276 void	nvme_ctrlr_destruct(struct spdk_nvme_ctrlr *ctrlr);
1277 void	nvme_ctrlr_destruct_async(struct spdk_nvme_ctrlr *ctrlr,
1278 				  struct nvme_ctrlr_detach_ctx *ctx);
1279 int	nvme_ctrlr_destruct_poll_async(struct spdk_nvme_ctrlr *ctrlr,
1280 				       struct nvme_ctrlr_detach_ctx *ctx);
1281 void	nvme_ctrlr_fail(struct spdk_nvme_ctrlr *ctrlr, bool hot_remove);
1282 int	nvme_ctrlr_process_init(struct spdk_nvme_ctrlr *ctrlr);
1283 void	nvme_ctrlr_disable(struct spdk_nvme_ctrlr *ctrlr);
1284 int	nvme_ctrlr_disable_poll(struct spdk_nvme_ctrlr *ctrlr);
1285 void	nvme_ctrlr_connected(struct spdk_nvme_probe_ctx *probe_ctx,
1286 			     struct spdk_nvme_ctrlr *ctrlr);
1287 
1288 int	nvme_ctrlr_submit_admin_request(struct spdk_nvme_ctrlr *ctrlr,
1289 					struct nvme_request *req);
1290 int	nvme_ctrlr_get_cap(struct spdk_nvme_ctrlr *ctrlr, union spdk_nvme_cap_register *cap);
1291 int	nvme_ctrlr_get_vs(struct spdk_nvme_ctrlr *ctrlr, union spdk_nvme_vs_register *vs);
1292 int	nvme_ctrlr_get_cmbsz(struct spdk_nvme_ctrlr *ctrlr, union spdk_nvme_cmbsz_register *cmbsz);
1293 int	nvme_ctrlr_get_pmrcap(struct spdk_nvme_ctrlr *ctrlr, union spdk_nvme_pmrcap_register *pmrcap);
1294 int	nvme_ctrlr_get_bpinfo(struct spdk_nvme_ctrlr *ctrlr, union spdk_nvme_bpinfo_register *bpinfo);
1295 int	nvme_ctrlr_set_bprsel(struct spdk_nvme_ctrlr *ctrlr, union spdk_nvme_bprsel_register *bprsel);
1296 int	nvme_ctrlr_set_bpmbl(struct spdk_nvme_ctrlr *ctrlr, uint64_t bpmbl_value);
1297 bool	nvme_ctrlr_multi_iocs_enabled(struct spdk_nvme_ctrlr *ctrlr);
1298 void nvme_ctrlr_disconnect_qpair(struct spdk_nvme_qpair *qpair);
1299 void nvme_ctrlr_abort_queued_aborts(struct spdk_nvme_ctrlr *ctrlr);
1300 int nvme_qpair_init(struct spdk_nvme_qpair *qpair, uint16_t id,
1301 		    struct spdk_nvme_ctrlr *ctrlr,
1302 		    enum spdk_nvme_qprio qprio,
1303 		    uint32_t num_requests, bool async);
1304 void	nvme_qpair_deinit(struct spdk_nvme_qpair *qpair);
1305 void	nvme_qpair_complete_error_reqs(struct spdk_nvme_qpair *qpair);
1306 int	nvme_qpair_submit_request(struct spdk_nvme_qpair *qpair,
1307 				  struct nvme_request *req);
1308 void	nvme_qpair_abort_all_queued_reqs(struct spdk_nvme_qpair *qpair);
1309 uint32_t nvme_qpair_abort_queued_reqs_with_cbarg(struct spdk_nvme_qpair *qpair, void *cmd_cb_arg);
1310 void	nvme_qpair_abort_queued_reqs(struct spdk_nvme_qpair *qpair);
1311 void	nvme_qpair_resubmit_requests(struct spdk_nvme_qpair *qpair, uint32_t num_requests);
1312 int	nvme_ctrlr_identify_active_ns(struct spdk_nvme_ctrlr *ctrlr);
1313 void	nvme_ns_set_identify_data(struct spdk_nvme_ns *ns);
1314 void	nvme_ns_set_id_desc_list_data(struct spdk_nvme_ns *ns);
1315 void	nvme_ns_free_zns_specific_data(struct spdk_nvme_ns *ns);
1316 void	nvme_ns_free_nvm_specific_data(struct spdk_nvme_ns *ns);
1317 void	nvme_ns_free_iocs_specific_data(struct spdk_nvme_ns *ns);
1318 bool	nvme_ns_has_supported_iocs_specific_data(struct spdk_nvme_ns *ns);
1319 int	nvme_ns_construct(struct spdk_nvme_ns *ns, uint32_t id,
1320 			  struct spdk_nvme_ctrlr *ctrlr);
1321 void	nvme_ns_destruct(struct spdk_nvme_ns *ns);
1322 int	nvme_ns_cmd_zone_append_with_md(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair,
1323 					void *buffer, void *metadata, uint64_t zslba,
1324 					uint32_t lba_count, spdk_nvme_cmd_cb cb_fn, void *cb_arg,
1325 					uint32_t io_flags, uint16_t apptag_mask, uint16_t apptag);
1326 int nvme_ns_cmd_zone_appendv_with_md(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair,
1327 				     uint64_t zslba, uint32_t lba_count,
1328 				     spdk_nvme_cmd_cb cb_fn, void *cb_arg, uint32_t io_flags,
1329 				     spdk_nvme_req_reset_sgl_cb reset_sgl_fn,
1330 				     spdk_nvme_req_next_sge_cb next_sge_fn, void *metadata,
1331 				     uint16_t apptag_mask, uint16_t apptag);
1332 
1333 int	nvme_fabric_ctrlr_set_reg_4(struct spdk_nvme_ctrlr *ctrlr, uint32_t offset, uint32_t value);
1334 int	nvme_fabric_ctrlr_set_reg_8(struct spdk_nvme_ctrlr *ctrlr, uint32_t offset, uint64_t value);
1335 int	nvme_fabric_ctrlr_get_reg_4(struct spdk_nvme_ctrlr *ctrlr, uint32_t offset, uint32_t *value);
1336 int	nvme_fabric_ctrlr_get_reg_8(struct spdk_nvme_ctrlr *ctrlr, uint32_t offset, uint64_t *value);
1337 int	nvme_fabric_ctrlr_set_reg_4_async(struct spdk_nvme_ctrlr *ctrlr, uint32_t offset,
1338 		uint32_t value, spdk_nvme_reg_cb cb_fn, void *cb_arg);
1339 int	nvme_fabric_ctrlr_set_reg_8_async(struct spdk_nvme_ctrlr *ctrlr, uint32_t offset,
1340 		uint64_t value, spdk_nvme_reg_cb cb_fn, void *cb_arg);
1341 int	nvme_fabric_ctrlr_get_reg_4_async(struct spdk_nvme_ctrlr *ctrlr, uint32_t offset,
1342 		spdk_nvme_reg_cb cb_fn, void *cb_arg);
1343 int	nvme_fabric_ctrlr_get_reg_8_async(struct spdk_nvme_ctrlr *ctrlr, uint32_t offset,
1344 		spdk_nvme_reg_cb cb_fn, void *cb_arg);
1345 int	nvme_fabric_ctrlr_scan(struct spdk_nvme_probe_ctx *probe_ctx, bool direct_connect);
1346 int	nvme_fabric_ctrlr_discover(struct spdk_nvme_ctrlr *ctrlr,
1347 				   struct spdk_nvme_probe_ctx *probe_ctx);
1348 int	nvme_fabric_qpair_connect(struct spdk_nvme_qpair *qpair, uint32_t num_entries);
1349 int	nvme_fabric_qpair_connect_async(struct spdk_nvme_qpair *qpair, uint32_t num_entries);
1350 int	nvme_fabric_qpair_connect_poll(struct spdk_nvme_qpair *qpair);
1351 bool	nvme_fabric_qpair_auth_required(struct spdk_nvme_qpair *qpair);
1352 int	nvme_fabric_qpair_authenticate_async(struct spdk_nvme_qpair *qpair);
1353 int	nvme_fabric_qpair_authenticate_poll(struct spdk_nvme_qpair *qpair);
1354 
1355 typedef int (*spdk_nvme_parse_ana_log_page_cb)(
1356 	const struct spdk_nvme_ana_group_descriptor *desc, void *cb_arg);
1357 int	nvme_ctrlr_parse_ana_log_page(struct spdk_nvme_ctrlr *ctrlr,
1358 				      spdk_nvme_parse_ana_log_page_cb cb_fn, void *cb_arg);
1359 
1360 static inline void
1361 nvme_request_clear(struct nvme_request *req)
1362 {
1363 	/*
1364 	 * Only memset/zero fields that need it.  All other fields
1365 	 *  will be initialized appropriately either later in this
1366 	 *  function, or before they are needed later in the
1367 	 *  submission patch.  For example, the children
1368 	 *  TAILQ_ENTRY and following members are
1369 	 *  only used as part of I/O splitting so we avoid
1370 	 *  memsetting them until it is actually needed.
1371 	 *  They will be initialized in nvme_request_add_child()
1372 	 *  if the request is split.
1373 	 */
1374 	memset(req, 0, offsetof(struct nvme_request, payload_size));
1375 }
1376 
1377 #define NVME_INIT_REQUEST(req, _cb_fn, _cb_arg, _payload, _payload_size, _md_size)	\
1378 	do {						\
1379 		nvme_request_clear(req);		\
1380 		req->cb_fn = _cb_fn;			\
1381 		req->cb_arg = _cb_arg;			\
1382 		req->payload = _payload;		\
1383 		req->payload_size = _payload_size;	\
1384 		req->md_size = _md_size;		\
1385 		req->pid = g_spdk_nvme_pid;		\
1386 		req->submit_tick = 0;			\
1387 		req->accel_sequence = NULL;		\
1388 	} while (0);
1389 
1390 static inline struct nvme_request *
1391 nvme_allocate_request(struct spdk_nvme_qpair *qpair,
1392 		      const struct nvme_payload *payload, uint32_t payload_size, uint32_t md_size,
1393 		      spdk_nvme_cmd_cb cb_fn, void *cb_arg)
1394 {
1395 	struct nvme_request *req;
1396 
1397 	req = STAILQ_FIRST(&qpair->free_req);
1398 	if (req == NULL) {
1399 		return req;
1400 	}
1401 
1402 	STAILQ_REMOVE_HEAD(&qpair->free_req, stailq);
1403 	qpair->num_outstanding_reqs++;
1404 
1405 	NVME_INIT_REQUEST(req, cb_fn, cb_arg, *payload, payload_size, md_size);
1406 
1407 	return req;
1408 }
1409 
1410 static inline struct nvme_request *
1411 nvme_allocate_request_contig(struct spdk_nvme_qpair *qpair,
1412 			     void *buffer, uint32_t payload_size,
1413 			     spdk_nvme_cmd_cb cb_fn, void *cb_arg)
1414 {
1415 	struct nvme_payload payload;
1416 
1417 	payload = NVME_PAYLOAD_CONTIG(buffer, NULL);
1418 
1419 	return nvme_allocate_request(qpair, &payload, payload_size, 0, cb_fn, cb_arg);
1420 }
1421 
1422 static inline struct nvme_request *
1423 nvme_allocate_request_null(struct spdk_nvme_qpair *qpair, spdk_nvme_cmd_cb cb_fn, void *cb_arg)
1424 {
1425 	return nvme_allocate_request_contig(qpair, NULL, 0, cb_fn, cb_arg);
1426 }
1427 
1428 struct nvme_request *nvme_allocate_request_user_copy(struct spdk_nvme_qpair *qpair,
1429 		void *buffer, uint32_t payload_size,
1430 		spdk_nvme_cmd_cb cb_fn, void *cb_arg, bool host_to_controller);
1431 
1432 static inline void
1433 _nvme_free_request(struct nvme_request *req, struct spdk_nvme_qpair *qpair)
1434 {
1435 	assert(req != NULL);
1436 	assert(req->num_children == 0);
1437 	assert(qpair != NULL);
1438 
1439 	/* The reserved_req does not go in the free_req STAILQ - it is
1440 	 * saved only for use with a FABRICS/CONNECT command.
1441 	 */
1442 	if (spdk_likely(qpair->reserved_req != req)) {
1443 		STAILQ_INSERT_HEAD(&qpair->free_req, req, stailq);
1444 
1445 		assert(qpair->num_outstanding_reqs > 0);
1446 		qpair->num_outstanding_reqs--;
1447 	}
1448 }
1449 
1450 static inline void
1451 nvme_free_request(struct nvme_request *req)
1452 {
1453 	_nvme_free_request(req, req->qpair);
1454 }
1455 
1456 static inline void
1457 nvme_complete_request(spdk_nvme_cmd_cb cb_fn, void *cb_arg, struct spdk_nvme_qpair *qpair,
1458 		      struct nvme_request *req, struct spdk_nvme_cpl *cpl)
1459 {
1460 	struct spdk_nvme_cpl            err_cpl;
1461 	struct nvme_error_cmd           *cmd;
1462 
1463 	if (spdk_unlikely(req->accel_sequence != NULL)) {
1464 		struct spdk_nvme_poll_group *pg = qpair->poll_group->group;
1465 
1466 		/* Transports are required to execute the sequence and clear req->accel_sequence.
1467 		 * If it's left non-NULL it must mean the request is failed. */
1468 		assert(spdk_nvme_cpl_is_error(cpl));
1469 		pg->accel_fn_table.abort_sequence(req->accel_sequence);
1470 		req->accel_sequence = NULL;
1471 	}
1472 
1473 	/* error injection at completion path,
1474 	 * only inject for successful completed commands
1475 	 */
1476 	if (spdk_unlikely(!TAILQ_EMPTY(&qpair->err_cmd_head) &&
1477 			  !spdk_nvme_cpl_is_error(cpl))) {
1478 		TAILQ_FOREACH(cmd, &qpair->err_cmd_head, link) {
1479 
1480 			if (cmd->do_not_submit) {
1481 				continue;
1482 			}
1483 
1484 			if ((cmd->opc == req->cmd.opc) && cmd->err_count) {
1485 
1486 				err_cpl = *cpl;
1487 				err_cpl.status.sct = cmd->status.sct;
1488 				err_cpl.status.sc = cmd->status.sc;
1489 
1490 				cpl = &err_cpl;
1491 				cmd->err_count--;
1492 				break;
1493 			}
1494 		}
1495 	}
1496 
1497 	/* For PCIe completions, we want to avoid touching the req itself to avoid
1498 	 * dependencies on loading those cachelines. So call the internal helper
1499 	 * function instead using the qpair that was passed by the caller, instead
1500 	 * of getting it from the req.
1501 	 */
1502 	_nvme_free_request(req, qpair);
1503 
1504 	if (spdk_likely(cb_fn)) {
1505 		cb_fn(cb_arg, cpl);
1506 	}
1507 }
1508 
1509 static inline void
1510 nvme_cleanup_user_req(struct nvme_request *req)
1511 {
1512 	if (req->user_buffer && req->payload_size) {
1513 		spdk_free(req->payload.contig_or_cb_arg);
1514 		req->user_buffer = NULL;
1515 	}
1516 
1517 	req->user_cb_arg = NULL;
1518 	req->user_cb_fn = NULL;
1519 }
1520 
1521 static inline bool
1522 nvme_request_abort_match(struct nvme_request *req, void *cmd_cb_arg)
1523 {
1524 	return req->cb_arg == cmd_cb_arg ||
1525 	       req->user_cb_arg == cmd_cb_arg ||
1526 	       (req->parent != NULL && req->parent->cb_arg == cmd_cb_arg);
1527 }
1528 
1529 static inline void
1530 nvme_qpair_set_state(struct spdk_nvme_qpair *qpair, enum nvme_qpair_state state)
1531 {
1532 	qpair->state = state;
1533 	if (state == NVME_QPAIR_ENABLED) {
1534 		qpair->is_new_qpair = false;
1535 	}
1536 }
1537 
1538 static inline enum nvme_qpair_state
1539 nvme_qpair_get_state(struct spdk_nvme_qpair *qpair) {
1540 	return qpair->state;
1541 }
1542 
1543 static inline void
1544 nvme_request_remove_child(struct nvme_request *parent, struct nvme_request *child)
1545 {
1546 	assert(parent != NULL);
1547 	assert(child != NULL);
1548 	assert(child->parent == parent);
1549 	assert(parent->num_children != 0);
1550 
1551 	parent->num_children--;
1552 	child->parent = NULL;
1553 	TAILQ_REMOVE(&parent->children, child, child_tailq);
1554 }
1555 
1556 static inline void
1557 nvme_cb_complete_child(void *child_arg, const struct spdk_nvme_cpl *cpl)
1558 {
1559 	struct nvme_request *child = child_arg;
1560 	struct nvme_request *parent = child->parent;
1561 
1562 	nvme_request_remove_child(parent, child);
1563 
1564 	if (spdk_nvme_cpl_is_error(cpl)) {
1565 		memcpy(&parent->parent_status, cpl, sizeof(*cpl));
1566 	}
1567 
1568 	if (parent->num_children == 0) {
1569 		nvme_complete_request(parent->cb_fn, parent->cb_arg, parent->qpair,
1570 				      parent, &parent->parent_status);
1571 	}
1572 }
1573 
1574 static inline void
1575 nvme_request_add_child(struct nvme_request *parent, struct nvme_request *child)
1576 {
1577 	assert(parent->num_children != UINT16_MAX);
1578 
1579 	if (parent->num_children == 0) {
1580 		/*
1581 		 * Defer initialization of the children TAILQ since it falls
1582 		 *  on a separate cacheline.  This ensures we do not touch this
1583 		 *  cacheline except on request splitting cases, which are
1584 		 *  relatively rare.
1585 		 */
1586 		TAILQ_INIT(&parent->children);
1587 		parent->parent = NULL;
1588 		memset(&parent->parent_status, 0, sizeof(struct spdk_nvme_cpl));
1589 	}
1590 
1591 	parent->num_children++;
1592 	TAILQ_INSERT_TAIL(&parent->children, child, child_tailq);
1593 	child->parent = parent;
1594 	child->cb_fn = nvme_cb_complete_child;
1595 	child->cb_arg = child;
1596 }
1597 
1598 static inline void
1599 nvme_request_free_children(struct nvme_request *req)
1600 {
1601 	struct nvme_request *child, *tmp;
1602 
1603 	if (req->num_children == 0) {
1604 		return;
1605 	}
1606 
1607 	/* free all child nvme_request */
1608 	TAILQ_FOREACH_SAFE(child, &req->children, child_tailq, tmp) {
1609 		nvme_request_remove_child(req, child);
1610 		nvme_request_free_children(child);
1611 		nvme_free_request(child);
1612 	}
1613 }
1614 
1615 int	nvme_request_check_timeout(struct nvme_request *req, uint16_t cid,
1616 				   struct spdk_nvme_ctrlr_process *active_proc, uint64_t now_tick);
1617 uint64_t nvme_get_quirks(const struct spdk_pci_id *id);
1618 
1619 int	nvme_robust_mutex_init_shared(pthread_mutex_t *mtx);
1620 int	nvme_robust_mutex_init_recursive_shared(pthread_mutex_t *mtx);
1621 
1622 bool	nvme_completion_is_retry(const struct spdk_nvme_cpl *cpl);
1623 
1624 struct spdk_nvme_ctrlr *nvme_get_ctrlr_by_trid_unsafe(
1625 	const struct spdk_nvme_transport_id *trid, const char *hostnqn);
1626 
1627 const struct spdk_nvme_transport *nvme_get_transport(const char *transport_name);
1628 const struct spdk_nvme_transport *nvme_get_first_transport(void);
1629 const struct spdk_nvme_transport *nvme_get_next_transport(const struct spdk_nvme_transport
1630 		*transport);
1631 void  nvme_ctrlr_update_namespaces(struct spdk_nvme_ctrlr *ctrlr);
1632 
1633 /* Transport specific functions */
1634 struct spdk_nvme_ctrlr *nvme_transport_ctrlr_construct(const struct spdk_nvme_transport_id *trid,
1635 		const struct spdk_nvme_ctrlr_opts *opts,
1636 		void *devhandle);
1637 int nvme_transport_ctrlr_destruct(struct spdk_nvme_ctrlr *ctrlr);
1638 int nvme_transport_ctrlr_scan(struct spdk_nvme_probe_ctx *probe_ctx, bool direct_connect);
1639 int nvme_transport_ctrlr_scan_attached(struct spdk_nvme_probe_ctx *probe_ctx);
1640 int nvme_transport_ctrlr_enable(struct spdk_nvme_ctrlr *ctrlr);
1641 int nvme_transport_ctrlr_ready(struct spdk_nvme_ctrlr *ctrlr);
1642 int nvme_transport_ctrlr_set_reg_4(struct spdk_nvme_ctrlr *ctrlr, uint32_t offset, uint32_t value);
1643 int nvme_transport_ctrlr_set_reg_8(struct spdk_nvme_ctrlr *ctrlr, uint32_t offset, uint64_t value);
1644 int nvme_transport_ctrlr_get_reg_4(struct spdk_nvme_ctrlr *ctrlr, uint32_t offset, uint32_t *value);
1645 int nvme_transport_ctrlr_get_reg_8(struct spdk_nvme_ctrlr *ctrlr, uint32_t offset, uint64_t *value);
1646 int nvme_transport_ctrlr_set_reg_4_async(struct spdk_nvme_ctrlr *ctrlr, uint32_t offset,
1647 		uint32_t value, spdk_nvme_reg_cb cb_fn, void *cb_arg);
1648 int nvme_transport_ctrlr_set_reg_8_async(struct spdk_nvme_ctrlr *ctrlr, uint32_t offset,
1649 		uint64_t value, spdk_nvme_reg_cb cb_fn, void *cb_arg);
1650 int nvme_transport_ctrlr_get_reg_4_async(struct spdk_nvme_ctrlr *ctrlr, uint32_t offset,
1651 		spdk_nvme_reg_cb cb_fn, void *cb_arg);
1652 int nvme_transport_ctrlr_get_reg_8_async(struct spdk_nvme_ctrlr *ctrlr, uint32_t offset,
1653 		spdk_nvme_reg_cb cb_fn, void *cb_arg);
1654 uint32_t nvme_transport_ctrlr_get_max_xfer_size(struct spdk_nvme_ctrlr *ctrlr);
1655 uint16_t nvme_transport_ctrlr_get_max_sges(struct spdk_nvme_ctrlr *ctrlr);
1656 struct spdk_nvme_qpair *nvme_transport_ctrlr_create_io_qpair(struct spdk_nvme_ctrlr *ctrlr,
1657 		uint16_t qid, const struct spdk_nvme_io_qpair_opts *opts);
1658 int nvme_transport_ctrlr_reserve_cmb(struct spdk_nvme_ctrlr *ctrlr);
1659 void *nvme_transport_ctrlr_map_cmb(struct spdk_nvme_ctrlr *ctrlr, size_t *size);
1660 int nvme_transport_ctrlr_unmap_cmb(struct spdk_nvme_ctrlr *ctrlr);
1661 int nvme_transport_ctrlr_enable_pmr(struct spdk_nvme_ctrlr *ctrlr);
1662 int nvme_transport_ctrlr_disable_pmr(struct spdk_nvme_ctrlr *ctrlr);
1663 void *nvme_transport_ctrlr_map_pmr(struct spdk_nvme_ctrlr *ctrlr, size_t *size);
1664 int nvme_transport_ctrlr_unmap_pmr(struct spdk_nvme_ctrlr *ctrlr);
1665 void nvme_transport_ctrlr_delete_io_qpair(struct spdk_nvme_ctrlr *ctrlr,
1666 		struct spdk_nvme_qpair *qpair);
1667 int nvme_transport_ctrlr_connect_qpair(struct spdk_nvme_ctrlr *ctrlr,
1668 				       struct spdk_nvme_qpair *qpair);
1669 void nvme_transport_ctrlr_disconnect_qpair(struct spdk_nvme_ctrlr *ctrlr,
1670 		struct spdk_nvme_qpair *qpair);
1671 void nvme_transport_ctrlr_disconnect_qpair_done(struct spdk_nvme_qpair *qpair);
1672 int nvme_transport_ctrlr_get_memory_domains(const struct spdk_nvme_ctrlr *ctrlr,
1673 		struct spdk_memory_domain **domains, int array_size);
1674 void nvme_transport_qpair_abort_reqs(struct spdk_nvme_qpair *qpair);
1675 int nvme_transport_qpair_reset(struct spdk_nvme_qpair *qpair);
1676 int nvme_transport_qpair_submit_request(struct spdk_nvme_qpair *qpair, struct nvme_request *req);
1677 int32_t nvme_transport_qpair_process_completions(struct spdk_nvme_qpair *qpair,
1678 		uint32_t max_completions);
1679 void nvme_transport_admin_qpair_abort_aers(struct spdk_nvme_qpair *qpair);
1680 int nvme_transport_qpair_iterate_requests(struct spdk_nvme_qpair *qpair,
1681 		int (*iter_fn)(struct nvme_request *req, void *arg),
1682 		void *arg);
1683 int nvme_transport_qpair_authenticate(struct spdk_nvme_qpair *qpair);
1684 
1685 struct spdk_nvme_transport_poll_group *nvme_transport_poll_group_create(
1686 	const struct spdk_nvme_transport *transport);
1687 struct spdk_nvme_transport_poll_group *nvme_transport_qpair_get_optimal_poll_group(
1688 	const struct spdk_nvme_transport *transport,
1689 	struct spdk_nvme_qpair *qpair);
1690 int nvme_transport_poll_group_add(struct spdk_nvme_transport_poll_group *tgroup,
1691 				  struct spdk_nvme_qpair *qpair);
1692 int nvme_transport_poll_group_remove(struct spdk_nvme_transport_poll_group *tgroup,
1693 				     struct spdk_nvme_qpair *qpair);
1694 int nvme_transport_poll_group_disconnect_qpair(struct spdk_nvme_qpair *qpair);
1695 int nvme_transport_poll_group_connect_qpair(struct spdk_nvme_qpair *qpair);
1696 int64_t nvme_transport_poll_group_process_completions(struct spdk_nvme_transport_poll_group *tgroup,
1697 		uint32_t completions_per_qpair, spdk_nvme_disconnected_qpair_cb disconnected_qpair_cb);
1698 int nvme_transport_poll_group_destroy(struct spdk_nvme_transport_poll_group *tgroup);
1699 int nvme_transport_poll_group_get_stats(struct spdk_nvme_transport_poll_group *tgroup,
1700 					struct spdk_nvme_transport_poll_group_stat **stats);
1701 void nvme_transport_poll_group_free_stats(struct spdk_nvme_transport_poll_group *tgroup,
1702 		struct spdk_nvme_transport_poll_group_stat *stats);
1703 enum spdk_nvme_transport_type nvme_transport_get_trtype(const struct spdk_nvme_transport
1704 		*transport);
1705 /*
1706  * Below ref related functions must be called with the global
1707  *  driver lock held for the multi-process condition.
1708  *  Within these functions, the per ctrlr ctrlr_lock is also
1709  *  acquired for the multi-thread condition.
1710  */
1711 void	nvme_ctrlr_proc_get_ref(struct spdk_nvme_ctrlr *ctrlr);
1712 void	nvme_ctrlr_proc_put_ref(struct spdk_nvme_ctrlr *ctrlr);
1713 int	nvme_ctrlr_get_ref_count(struct spdk_nvme_ctrlr *ctrlr);
1714 
1715 int	nvme_ctrlr_reinitialize_io_qpair(struct spdk_nvme_ctrlr *ctrlr, struct spdk_nvme_qpair *qpair);
1716 int	nvme_parse_addr(struct sockaddr_storage *sa, int family,
1717 			const char *addr, const char *service, long int *port);
1718 int	nvme_get_default_hostnqn(char *buf, int len);
1719 
1720 static inline bool
1721 _is_page_aligned(uint64_t address, uint64_t page_size)
1722 {
1723 	return (address & (page_size - 1)) == 0;
1724 }
1725 
1726 #endif /* __NVME_INTERNAL_H__ */
1727