xref: /spdk/lib/nvme/nvme_ctrlr.c (revision 56e12b00711b40d1e2ff45d4147193f45fdd9af0)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright (c) Intel Corporation.
5  *   All rights reserved.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
11  *     * Redistributions of source code must retain the above copyright
12  *       notice, this list of conditions and the following disclaimer.
13  *     * Redistributions in binary form must reproduce the above copyright
14  *       notice, this list of conditions and the following disclaimer in
15  *       the documentation and/or other materials provided with the
16  *       distribution.
17  *     * Neither the name of Intel Corporation nor the names of its
18  *       contributors may be used to endorse or promote products derived
19  *       from this software without specific prior written permission.
20  *
21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include "spdk/stdinc.h"
35 
36 #include "nvme_internal.h"
37 
38 #include "spdk/env.h"
39 #include "spdk/string.h"
40 
41 static int nvme_ctrlr_construct_and_submit_aer(struct spdk_nvme_ctrlr *ctrlr,
42 		struct nvme_async_event_request *aer);
43 static int nvme_ctrlr_identify_ns_async(struct spdk_nvme_ns *ns);
44 static int nvme_ctrlr_identify_id_desc_async(struct spdk_nvme_ns *ns);
45 
46 static int
47 nvme_ctrlr_get_cc(struct spdk_nvme_ctrlr *ctrlr, union spdk_nvme_cc_register *cc)
48 {
49 	return nvme_transport_ctrlr_get_reg_4(ctrlr, offsetof(struct spdk_nvme_registers, cc.raw),
50 					      &cc->raw);
51 }
52 
53 static int
54 nvme_ctrlr_get_csts(struct spdk_nvme_ctrlr *ctrlr, union spdk_nvme_csts_register *csts)
55 {
56 	return nvme_transport_ctrlr_get_reg_4(ctrlr, offsetof(struct spdk_nvme_registers, csts.raw),
57 					      &csts->raw);
58 }
59 
60 int
61 nvme_ctrlr_get_cap(struct spdk_nvme_ctrlr *ctrlr, union spdk_nvme_cap_register *cap)
62 {
63 	return nvme_transport_ctrlr_get_reg_8(ctrlr, offsetof(struct spdk_nvme_registers, cap.raw),
64 					      &cap->raw);
65 }
66 
67 int
68 nvme_ctrlr_get_vs(struct spdk_nvme_ctrlr *ctrlr, union spdk_nvme_vs_register *vs)
69 {
70 	return nvme_transport_ctrlr_get_reg_4(ctrlr, offsetof(struct spdk_nvme_registers, vs.raw),
71 					      &vs->raw);
72 }
73 
74 static int
75 nvme_ctrlr_set_cc(struct spdk_nvme_ctrlr *ctrlr, const union spdk_nvme_cc_register *cc)
76 {
77 	return nvme_transport_ctrlr_set_reg_4(ctrlr, offsetof(struct spdk_nvme_registers, cc.raw),
78 					      cc->raw);
79 }
80 
81 int
82 nvme_ctrlr_get_cmbsz(struct spdk_nvme_ctrlr *ctrlr, union spdk_nvme_cmbsz_register *cmbsz)
83 {
84 	return nvme_transport_ctrlr_get_reg_4(ctrlr, offsetof(struct spdk_nvme_registers, cmbsz.raw),
85 					      &cmbsz->raw);
86 }
87 
88 void
89 spdk_nvme_ctrlr_get_default_ctrlr_opts(struct spdk_nvme_ctrlr_opts *opts, size_t opts_size)
90 {
91 	char host_id_str[SPDK_UUID_STRING_LEN];
92 
93 	assert(opts);
94 
95 	memset(opts, 0, opts_size);
96 
97 #define FIELD_OK(field) \
98 	offsetof(struct spdk_nvme_ctrlr_opts, field) + sizeof(opts->field) <= opts_size
99 
100 	if (FIELD_OK(num_io_queues)) {
101 		opts->num_io_queues = DEFAULT_MAX_IO_QUEUES;
102 	}
103 
104 	if (FIELD_OK(use_cmb_sqs)) {
105 		opts->use_cmb_sqs = true;
106 	}
107 
108 	if (FIELD_OK(arb_mechanism)) {
109 		opts->arb_mechanism = SPDK_NVME_CC_AMS_RR;
110 	}
111 
112 	if (FIELD_OK(keep_alive_timeout_ms)) {
113 		opts->keep_alive_timeout_ms = MIN_KEEP_ALIVE_TIMEOUT_IN_MS;
114 	}
115 
116 	if (FIELD_OK(io_queue_size)) {
117 		opts->io_queue_size = DEFAULT_IO_QUEUE_SIZE;
118 	}
119 
120 	if (FIELD_OK(io_queue_requests)) {
121 		opts->io_queue_requests = DEFAULT_IO_QUEUE_REQUESTS;
122 	}
123 
124 	if (FIELD_OK(host_id)) {
125 		memset(opts->host_id, 0, sizeof(opts->host_id));
126 	}
127 
128 	if (nvme_driver_init() == 0) {
129 		if (FIELD_OK(extended_host_id)) {
130 			memcpy(opts->extended_host_id, &g_spdk_nvme_driver->default_extended_host_id,
131 			       sizeof(opts->extended_host_id));
132 		}
133 
134 		if (FIELD_OK(hostnqn)) {
135 			spdk_uuid_fmt_lower(host_id_str, sizeof(host_id_str),
136 					    &g_spdk_nvme_driver->default_extended_host_id);
137 			snprintf(opts->hostnqn, sizeof(opts->hostnqn), "2014-08.org.nvmexpress:uuid:%s", host_id_str);
138 		}
139 	}
140 
141 	if (FIELD_OK(src_addr)) {
142 		memset(opts->src_addr, 0, sizeof(opts->src_addr));
143 	}
144 
145 	if (FIELD_OK(src_svcid)) {
146 		memset(opts->src_svcid, 0, sizeof(opts->src_svcid));
147 	}
148 
149 	if (FIELD_OK(command_set)) {
150 		opts->command_set = SPDK_NVME_CC_CSS_NVM;
151 	}
152 
153 	if (FIELD_OK(admin_timeout_ms)) {
154 		opts->admin_timeout_ms = NVME_MAX_TIMEOUT_PERIOD * 1000;
155 	}
156 
157 	if (FIELD_OK(header_digest)) {
158 		opts->header_digest = false;
159 	}
160 
161 	if (FIELD_OK(data_digest)) {
162 		opts->data_digest = false;
163 	}
164 #undef FIELD_OK
165 }
166 
167 /**
168  * This function will be called when the process allocates the IO qpair.
169  * Note: the ctrlr_lock must be held when calling this function.
170  */
171 static void
172 nvme_ctrlr_proc_add_io_qpair(struct spdk_nvme_qpair *qpair)
173 {
174 	struct spdk_nvme_ctrlr_process	*active_proc;
175 	struct spdk_nvme_ctrlr		*ctrlr = qpair->ctrlr;
176 
177 	active_proc = spdk_nvme_ctrlr_get_current_process(ctrlr);
178 	if (active_proc) {
179 		TAILQ_INSERT_TAIL(&active_proc->allocated_io_qpairs, qpair, per_process_tailq);
180 		qpair->active_proc = active_proc;
181 	}
182 }
183 
184 /**
185  * This function will be called when the process frees the IO qpair.
186  * Note: the ctrlr_lock must be held when calling this function.
187  */
188 static void
189 nvme_ctrlr_proc_remove_io_qpair(struct spdk_nvme_qpair *qpair)
190 {
191 	struct spdk_nvme_ctrlr_process	*active_proc;
192 	struct spdk_nvme_ctrlr		*ctrlr = qpair->ctrlr;
193 	struct spdk_nvme_qpair          *active_qpair, *tmp_qpair;
194 
195 	active_proc = spdk_nvme_ctrlr_get_current_process(ctrlr);
196 	if (!active_proc) {
197 		return;
198 	}
199 
200 	TAILQ_FOREACH_SAFE(active_qpair, &active_proc->allocated_io_qpairs,
201 			   per_process_tailq, tmp_qpair) {
202 		if (active_qpair == qpair) {
203 			TAILQ_REMOVE(&active_proc->allocated_io_qpairs,
204 				     active_qpair, per_process_tailq);
205 
206 			break;
207 		}
208 	}
209 }
210 
211 void
212 spdk_nvme_ctrlr_get_default_io_qpair_opts(struct spdk_nvme_ctrlr *ctrlr,
213 		struct spdk_nvme_io_qpair_opts *opts,
214 		size_t opts_size)
215 {
216 	assert(ctrlr);
217 
218 	assert(opts);
219 
220 	memset(opts, 0, opts_size);
221 
222 #define FIELD_OK(field) \
223 	offsetof(struct spdk_nvme_io_qpair_opts, field) + sizeof(opts->field) <= opts_size
224 
225 	if (FIELD_OK(qprio)) {
226 		opts->qprio = SPDK_NVME_QPRIO_URGENT;
227 	}
228 
229 	if (FIELD_OK(io_queue_size)) {
230 		opts->io_queue_size = ctrlr->opts.io_queue_size;
231 	}
232 
233 	if (FIELD_OK(io_queue_requests)) {
234 		opts->io_queue_requests = ctrlr->opts.io_queue_requests;
235 	}
236 
237 #undef FIELD_OK
238 }
239 
240 struct spdk_nvme_qpair *
241 spdk_nvme_ctrlr_alloc_io_qpair(struct spdk_nvme_ctrlr *ctrlr,
242 			       const struct spdk_nvme_io_qpair_opts *user_opts,
243 			       size_t opts_size)
244 {
245 	uint32_t				qid;
246 	struct spdk_nvme_qpair			*qpair;
247 	union spdk_nvme_cc_register		cc;
248 	struct spdk_nvme_io_qpair_opts		opts;
249 
250 	if (!ctrlr) {
251 		return NULL;
252 	}
253 
254 	/*
255 	 * Get the default options, then overwrite them with the user-provided options
256 	 * up to opts_size.
257 	 *
258 	 * This allows for extensions of the opts structure without breaking
259 	 * ABI compatibility.
260 	 */
261 	spdk_nvme_ctrlr_get_default_io_qpair_opts(ctrlr, &opts, sizeof(opts));
262 	if (user_opts) {
263 		memcpy(&opts, user_opts, spdk_min(sizeof(opts), opts_size));
264 	}
265 
266 	nvme_robust_mutex_lock(&ctrlr->ctrlr_lock);
267 	if (nvme_ctrlr_get_cc(ctrlr, &cc)) {
268 		SPDK_ERRLOG("get_cc failed\n");
269 		nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
270 		return NULL;
271 	}
272 
273 	/* Only the low 2 bits (values 0, 1, 2, 3) of QPRIO are valid. */
274 	if ((opts.qprio & 3) != opts.qprio) {
275 		nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
276 		return NULL;
277 	}
278 
279 	/*
280 	 * Only value SPDK_NVME_QPRIO_URGENT(0) is valid for the
281 	 * default round robin arbitration method.
282 	 */
283 	if ((cc.bits.ams == SPDK_NVME_CC_AMS_RR) && (opts.qprio != SPDK_NVME_QPRIO_URGENT)) {
284 		SPDK_ERRLOG("invalid queue priority for default round robin arbitration method\n");
285 		nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
286 		return NULL;
287 	}
288 
289 	/*
290 	 * Get the first available I/O queue ID.
291 	 */
292 	qid = spdk_bit_array_find_first_set(ctrlr->free_io_qids, 1);
293 	if (qid > ctrlr->opts.num_io_queues) {
294 		SPDK_ERRLOG("No free I/O queue IDs\n");
295 		nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
296 		return NULL;
297 	}
298 
299 	qpair = nvme_transport_ctrlr_create_io_qpair(ctrlr, qid, &opts);
300 	if (qpair == NULL) {
301 		SPDK_ERRLOG("nvme_transport_ctrlr_create_io_qpair() failed\n");
302 		nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
303 		return NULL;
304 	}
305 	spdk_bit_array_clear(ctrlr->free_io_qids, qid);
306 	TAILQ_INSERT_TAIL(&ctrlr->active_io_qpairs, qpair, tailq);
307 
308 	nvme_ctrlr_proc_add_io_qpair(qpair);
309 
310 	nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
311 
312 	if (ctrlr->quirks & NVME_QUIRK_DELAY_AFTER_QUEUE_ALLOC) {
313 		spdk_delay_us(100);
314 	}
315 
316 	return qpair;
317 }
318 
319 int
320 spdk_nvme_ctrlr_free_io_qpair(struct spdk_nvme_qpair *qpair)
321 {
322 	struct spdk_nvme_ctrlr *ctrlr;
323 
324 	if (qpair == NULL) {
325 		return 0;
326 	}
327 
328 	ctrlr = qpair->ctrlr;
329 
330 	if (qpair->in_completion_context) {
331 		/*
332 		 * There are many cases where it is convenient to delete an io qpair in the context
333 		 *  of that qpair's completion routine.  To handle this properly, set a flag here
334 		 *  so that the completion routine will perform an actual delete after the context
335 		 *  unwinds.
336 		 */
337 		qpair->delete_after_completion_context = 1;
338 		return 0;
339 	}
340 
341 	nvme_robust_mutex_lock(&ctrlr->ctrlr_lock);
342 
343 	nvme_ctrlr_proc_remove_io_qpair(qpair);
344 
345 	TAILQ_REMOVE(&ctrlr->active_io_qpairs, qpair, tailq);
346 	spdk_bit_array_set(ctrlr->free_io_qids, qpair->id);
347 
348 	if (nvme_transport_ctrlr_delete_io_qpair(ctrlr, qpair)) {
349 		nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
350 		return -1;
351 	}
352 
353 	nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
354 	return 0;
355 }
356 
357 static void
358 nvme_ctrlr_construct_intel_support_log_page_list(struct spdk_nvme_ctrlr *ctrlr,
359 		struct spdk_nvme_intel_log_page_directory *log_page_directory)
360 {
361 	if (log_page_directory == NULL) {
362 		return;
363 	}
364 
365 	if (ctrlr->cdata.vid != SPDK_PCI_VID_INTEL) {
366 		return;
367 	}
368 
369 	ctrlr->log_page_supported[SPDK_NVME_INTEL_LOG_PAGE_DIRECTORY] = true;
370 
371 	if (log_page_directory->read_latency_log_len ||
372 	    (ctrlr->quirks & NVME_INTEL_QUIRK_READ_LATENCY)) {
373 		ctrlr->log_page_supported[SPDK_NVME_INTEL_LOG_READ_CMD_LATENCY] = true;
374 	}
375 	if (log_page_directory->write_latency_log_len ||
376 	    (ctrlr->quirks & NVME_INTEL_QUIRK_WRITE_LATENCY)) {
377 		ctrlr->log_page_supported[SPDK_NVME_INTEL_LOG_WRITE_CMD_LATENCY] = true;
378 	}
379 	if (log_page_directory->temperature_statistics_log_len) {
380 		ctrlr->log_page_supported[SPDK_NVME_INTEL_LOG_TEMPERATURE] = true;
381 	}
382 	if (log_page_directory->smart_log_len) {
383 		ctrlr->log_page_supported[SPDK_NVME_INTEL_LOG_SMART] = true;
384 	}
385 	if (log_page_directory->marketing_description_log_len) {
386 		ctrlr->log_page_supported[SPDK_NVME_INTEL_MARKETING_DESCRIPTION] = true;
387 	}
388 }
389 
390 static int nvme_ctrlr_set_intel_support_log_pages(struct spdk_nvme_ctrlr *ctrlr)
391 {
392 	int rc = 0;
393 	uint64_t phys_addr = 0;
394 	struct nvme_completion_poll_status	status;
395 	struct spdk_nvme_intel_log_page_directory *log_page_directory;
396 
397 	log_page_directory = spdk_zmalloc(sizeof(struct spdk_nvme_intel_log_page_directory),
398 					  64, &phys_addr, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
399 	if (log_page_directory == NULL) {
400 		SPDK_ERRLOG("could not allocate log_page_directory\n");
401 		return -ENXIO;
402 	}
403 
404 	rc = spdk_nvme_ctrlr_cmd_get_log_page(ctrlr, SPDK_NVME_INTEL_LOG_PAGE_DIRECTORY,
405 					      SPDK_NVME_GLOBAL_NS_TAG, log_page_directory,
406 					      sizeof(struct spdk_nvme_intel_log_page_directory),
407 					      0, nvme_completion_poll_cb, &status);
408 	if (rc != 0) {
409 		spdk_free(log_page_directory);
410 		return rc;
411 	}
412 
413 	if (spdk_nvme_wait_for_completion_timeout(ctrlr->adminq, &status,
414 			ctrlr->opts.admin_timeout_ms / 1000)) {
415 		spdk_free(log_page_directory);
416 		SPDK_WARNLOG("Intel log pages not supported on Intel drive!\n");
417 		return 0;
418 	}
419 
420 	nvme_ctrlr_construct_intel_support_log_page_list(ctrlr, log_page_directory);
421 	spdk_free(log_page_directory);
422 	return 0;
423 }
424 
425 static int
426 nvme_ctrlr_set_supported_log_pages(struct spdk_nvme_ctrlr *ctrlr)
427 {
428 	int	rc = 0;
429 
430 	memset(ctrlr->log_page_supported, 0, sizeof(ctrlr->log_page_supported));
431 	/* Mandatory pages */
432 	ctrlr->log_page_supported[SPDK_NVME_LOG_ERROR] = true;
433 	ctrlr->log_page_supported[SPDK_NVME_LOG_HEALTH_INFORMATION] = true;
434 	ctrlr->log_page_supported[SPDK_NVME_LOG_FIRMWARE_SLOT] = true;
435 	if (ctrlr->cdata.lpa.celp) {
436 		ctrlr->log_page_supported[SPDK_NVME_LOG_COMMAND_EFFECTS_LOG] = true;
437 	}
438 	if (ctrlr->cdata.vid == SPDK_PCI_VID_INTEL && !(ctrlr->quirks & NVME_INTEL_QUIRK_NO_LOG_PAGES)) {
439 		rc = nvme_ctrlr_set_intel_support_log_pages(ctrlr);
440 	}
441 
442 	return rc;
443 }
444 
445 static void
446 nvme_ctrlr_set_intel_supported_features(struct spdk_nvme_ctrlr *ctrlr)
447 {
448 	ctrlr->feature_supported[SPDK_NVME_INTEL_FEAT_MAX_LBA] = true;
449 	ctrlr->feature_supported[SPDK_NVME_INTEL_FEAT_NATIVE_MAX_LBA] = true;
450 	ctrlr->feature_supported[SPDK_NVME_INTEL_FEAT_POWER_GOVERNOR_SETTING] = true;
451 	ctrlr->feature_supported[SPDK_NVME_INTEL_FEAT_SMBUS_ADDRESS] = true;
452 	ctrlr->feature_supported[SPDK_NVME_INTEL_FEAT_LED_PATTERN] = true;
453 	ctrlr->feature_supported[SPDK_NVME_INTEL_FEAT_RESET_TIMED_WORKLOAD_COUNTERS] = true;
454 	ctrlr->feature_supported[SPDK_NVME_INTEL_FEAT_LATENCY_TRACKING] = true;
455 }
456 
457 static void
458 nvme_ctrlr_set_supported_features(struct spdk_nvme_ctrlr *ctrlr)
459 {
460 	memset(ctrlr->feature_supported, 0, sizeof(ctrlr->feature_supported));
461 	/* Mandatory features */
462 	ctrlr->feature_supported[SPDK_NVME_FEAT_ARBITRATION] = true;
463 	ctrlr->feature_supported[SPDK_NVME_FEAT_POWER_MANAGEMENT] = true;
464 	ctrlr->feature_supported[SPDK_NVME_FEAT_TEMPERATURE_THRESHOLD] = true;
465 	ctrlr->feature_supported[SPDK_NVME_FEAT_ERROR_RECOVERY] = true;
466 	ctrlr->feature_supported[SPDK_NVME_FEAT_NUMBER_OF_QUEUES] = true;
467 	ctrlr->feature_supported[SPDK_NVME_FEAT_INTERRUPT_COALESCING] = true;
468 	ctrlr->feature_supported[SPDK_NVME_FEAT_INTERRUPT_VECTOR_CONFIGURATION] = true;
469 	ctrlr->feature_supported[SPDK_NVME_FEAT_WRITE_ATOMICITY] = true;
470 	ctrlr->feature_supported[SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION] = true;
471 	/* Optional features */
472 	if (ctrlr->cdata.vwc.present) {
473 		ctrlr->feature_supported[SPDK_NVME_FEAT_VOLATILE_WRITE_CACHE] = true;
474 	}
475 	if (ctrlr->cdata.apsta.supported) {
476 		ctrlr->feature_supported[SPDK_NVME_FEAT_AUTONOMOUS_POWER_STATE_TRANSITION] = true;
477 	}
478 	if (ctrlr->cdata.hmpre) {
479 		ctrlr->feature_supported[SPDK_NVME_FEAT_HOST_MEM_BUFFER] = true;
480 	}
481 	if (ctrlr->cdata.vid == SPDK_PCI_VID_INTEL) {
482 		nvme_ctrlr_set_intel_supported_features(ctrlr);
483 	}
484 }
485 
486 void
487 nvme_ctrlr_fail(struct spdk_nvme_ctrlr *ctrlr, bool hot_remove)
488 {
489 	/*
490 	 * Set the flag here and leave the work failure of qpairs to
491 	 * spdk_nvme_qpair_process_completions().
492 	 */
493 	if (hot_remove) {
494 		ctrlr->is_removed = true;
495 	}
496 	ctrlr->is_failed = true;
497 	SPDK_ERRLOG("ctrlr %s in failed state.\n", ctrlr->trid.traddr);
498 }
499 
500 static void
501 nvme_ctrlr_shutdown(struct spdk_nvme_ctrlr *ctrlr)
502 {
503 	union spdk_nvme_cc_register	cc;
504 	union spdk_nvme_csts_register	csts;
505 	uint32_t			ms_waited = 0;
506 	uint32_t			shutdown_timeout_ms;
507 
508 	if (ctrlr->is_removed) {
509 		return;
510 	}
511 
512 	if (nvme_ctrlr_get_cc(ctrlr, &cc)) {
513 		SPDK_ERRLOG("get_cc() failed\n");
514 		return;
515 	}
516 
517 	cc.bits.shn = SPDK_NVME_SHN_NORMAL;
518 
519 	if (nvme_ctrlr_set_cc(ctrlr, &cc)) {
520 		SPDK_ERRLOG("set_cc() failed\n");
521 		return;
522 	}
523 
524 	/*
525 	 * The NVMe specification defines RTD3E to be the time between
526 	 *  setting SHN = 1 until the controller will set SHST = 10b.
527 	 * If the device doesn't report RTD3 entry latency, or if it
528 	 *  reports RTD3 entry latency less than 10 seconds, pick
529 	 *  10 seconds as a reasonable amount of time to
530 	 *  wait before proceeding.
531 	 */
532 	SPDK_DEBUGLOG(SPDK_LOG_NVME, "RTD3E = %" PRIu32 " us\n", ctrlr->cdata.rtd3e);
533 	shutdown_timeout_ms = (ctrlr->cdata.rtd3e + 999) / 1000;
534 	shutdown_timeout_ms = spdk_max(shutdown_timeout_ms, 10000);
535 	SPDK_DEBUGLOG(SPDK_LOG_NVME, "shutdown timeout = %" PRIu32 " ms\n", shutdown_timeout_ms);
536 
537 	do {
538 		if (nvme_ctrlr_get_csts(ctrlr, &csts)) {
539 			SPDK_ERRLOG("get_csts() failed\n");
540 			return;
541 		}
542 
543 		if (csts.bits.shst == SPDK_NVME_SHST_COMPLETE) {
544 			SPDK_DEBUGLOG(SPDK_LOG_NVME, "shutdown complete in %u milliseconds\n",
545 				      ms_waited);
546 			ctrlr->is_shutdown = true;
547 			return;
548 		}
549 
550 		nvme_delay(1000);
551 		ms_waited++;
552 	} while (ms_waited < shutdown_timeout_ms);
553 
554 	SPDK_ERRLOG("did not shutdown within %u milliseconds\n", shutdown_timeout_ms);
555 }
556 
557 static int
558 nvme_ctrlr_enable(struct spdk_nvme_ctrlr *ctrlr)
559 {
560 	union spdk_nvme_cc_register	cc;
561 	int				rc;
562 
563 	rc = nvme_transport_ctrlr_enable(ctrlr);
564 	if (rc != 0) {
565 		SPDK_ERRLOG("transport ctrlr_enable failed\n");
566 		return rc;
567 	}
568 
569 	if (nvme_ctrlr_get_cc(ctrlr, &cc)) {
570 		SPDK_ERRLOG("get_cc() failed\n");
571 		return -EIO;
572 	}
573 
574 	if (cc.bits.en != 0) {
575 		SPDK_ERRLOG("%s called with CC.EN = 1\n", __func__);
576 		return -EINVAL;
577 	}
578 
579 	cc.bits.en = 1;
580 	cc.bits.css = 0;
581 	cc.bits.shn = 0;
582 	cc.bits.iosqes = 6; /* SQ entry size == 64 == 2^6 */
583 	cc.bits.iocqes = 4; /* CQ entry size == 16 == 2^4 */
584 
585 	/* Page size is 2 ^ (12 + mps). */
586 	cc.bits.mps = spdk_u32log2(ctrlr->page_size) - 12;
587 
588 	if (ctrlr->cap.bits.css == 0) {
589 		SPDK_INFOLOG(SPDK_LOG_NVME,
590 			     "Drive reports no command sets supported. Assuming NVM is supported.\n");
591 		ctrlr->cap.bits.css = SPDK_NVME_CAP_CSS_NVM;
592 	}
593 
594 	if (!(ctrlr->cap.bits.css & (1u << ctrlr->opts.command_set))) {
595 		SPDK_DEBUGLOG(SPDK_LOG_NVME, "Requested I/O command set %u but supported mask is 0x%x\n",
596 			      ctrlr->opts.command_set, ctrlr->cap.bits.css);
597 		return -EINVAL;
598 	}
599 
600 	cc.bits.css = ctrlr->opts.command_set;
601 
602 	switch (ctrlr->opts.arb_mechanism) {
603 	case SPDK_NVME_CC_AMS_RR:
604 		break;
605 	case SPDK_NVME_CC_AMS_WRR:
606 		if (SPDK_NVME_CAP_AMS_WRR & ctrlr->cap.bits.ams) {
607 			break;
608 		}
609 		return -EINVAL;
610 	case SPDK_NVME_CC_AMS_VS:
611 		if (SPDK_NVME_CAP_AMS_VS & ctrlr->cap.bits.ams) {
612 			break;
613 		}
614 		return -EINVAL;
615 	default:
616 		return -EINVAL;
617 	}
618 
619 	cc.bits.ams = ctrlr->opts.arb_mechanism;
620 
621 	if (nvme_ctrlr_set_cc(ctrlr, &cc)) {
622 		SPDK_ERRLOG("set_cc() failed\n");
623 		return -EIO;
624 	}
625 
626 	return 0;
627 }
628 
629 #ifdef DEBUG
630 static const char *
631 nvme_ctrlr_state_string(enum nvme_ctrlr_state state)
632 {
633 	switch (state) {
634 	case NVME_CTRLR_STATE_INIT_DELAY:
635 		return "delay init";
636 	case NVME_CTRLR_STATE_INIT:
637 		return "init";
638 	case NVME_CTRLR_STATE_DISABLE_WAIT_FOR_READY_1:
639 		return "disable and wait for CSTS.RDY = 1";
640 	case NVME_CTRLR_STATE_DISABLE_WAIT_FOR_READY_0:
641 		return "disable and wait for CSTS.RDY = 0";
642 	case NVME_CTRLR_STATE_ENABLE:
643 		return "enable controller by writing CC.EN = 1";
644 	case NVME_CTRLR_STATE_ENABLE_WAIT_FOR_READY_1:
645 		return "wait for CSTS.RDY = 1";
646 	case NVME_CTRLR_STATE_ENABLE_ADMIN_QUEUE:
647 		return "enable admin queue";
648 	case NVME_CTRLR_STATE_IDENTIFY:
649 		return "identify controller";
650 	case NVME_CTRLR_STATE_WAIT_FOR_IDENTIFY:
651 		return "wait for identify controller";
652 	case NVME_CTRLR_STATE_SET_NUM_QUEUES:
653 		return "set number of queues";
654 	case NVME_CTRLR_STATE_WAIT_FOR_SET_NUM_QUEUES:
655 		return "wait for set number of queues";
656 	case NVME_CTRLR_STATE_GET_NUM_QUEUES:
657 		return "get number of queues";
658 	case NVME_CTRLR_STATE_WAIT_FOR_GET_NUM_QUEUES:
659 		return "wait for get number of queues";
660 	case NVME_CTRLR_STATE_CONSTRUCT_NS:
661 		return "construct namespaces";
662 	case NVME_CTRLR_STATE_IDENTIFY_ACTIVE_NS:
663 		return "identify active ns";
664 	case NVME_CTRLR_STATE_IDENTIFY_NS:
665 		return "identify ns";
666 	case NVME_CTRLR_STATE_WAIT_FOR_IDENTIFY_NS:
667 		return "wait for identify ns";
668 	case NVME_CTRLR_STATE_IDENTIFY_ID_DESCS:
669 		return "identify namespace id descriptors";
670 	case NVME_CTRLR_STATE_WAIT_FOR_IDENTIFY_ID_DESCS:
671 		return "wait for identify namespace id descriptors";
672 	case NVME_CTRLR_STATE_CONFIGURE_AER:
673 		return "configure AER";
674 	case NVME_CTRLR_STATE_WAIT_FOR_CONFIGURE_AER:
675 		return "wait for configure aer";
676 	case NVME_CTRLR_STATE_SET_SUPPORTED_LOG_PAGES:
677 		return "set supported log pages";
678 	case NVME_CTRLR_STATE_SET_SUPPORTED_FEATURES:
679 		return "set supported features";
680 	case NVME_CTRLR_STATE_SET_DB_BUF_CFG:
681 		return "set doorbell buffer config";
682 	case NVME_CTRLR_STATE_WAIT_FOR_DB_BUF_CFG:
683 		return "wait for doorbell buffer config";
684 	case NVME_CTRLR_STATE_SET_KEEP_ALIVE_TIMEOUT:
685 		return "set keep alive timeout";
686 	case NVME_CTRLR_STATE_WAIT_FOR_KEEP_ALIVE_TIMEOUT:
687 		return "wait for set keep alive timeout";
688 	case NVME_CTRLR_STATE_SET_HOST_ID:
689 		return "set host ID";
690 	case NVME_CTRLR_STATE_WAIT_FOR_HOST_ID:
691 		return "wait for set host ID";
692 	case NVME_CTRLR_STATE_READY:
693 		return "ready";
694 	case NVME_CTRLR_STATE_ERROR:
695 		return "error";
696 	}
697 	return "unknown";
698 };
699 #endif /* DEBUG */
700 
701 static void
702 nvme_ctrlr_set_state(struct spdk_nvme_ctrlr *ctrlr, enum nvme_ctrlr_state state,
703 		     uint64_t timeout_in_ms)
704 {
705 	ctrlr->state = state;
706 	if (timeout_in_ms == 0) {
707 		SPDK_DEBUGLOG(SPDK_LOG_NVME, "setting state to %s (no timeout)\n",
708 			      nvme_ctrlr_state_string(ctrlr->state));
709 		ctrlr->state_timeout_tsc = NVME_TIMEOUT_INFINITE;
710 	} else {
711 		SPDK_DEBUGLOG(SPDK_LOG_NVME, "setting state to %s (timeout %" PRIu64 " ms)\n",
712 			      nvme_ctrlr_state_string(ctrlr->state), timeout_in_ms);
713 		ctrlr->state_timeout_tsc = spdk_get_ticks() + (timeout_in_ms * spdk_get_ticks_hz()) / 1000;
714 	}
715 }
716 
717 static void
718 nvme_ctrlr_free_doorbell_buffer(struct spdk_nvme_ctrlr *ctrlr)
719 {
720 	if (ctrlr->shadow_doorbell) {
721 		spdk_dma_free(ctrlr->shadow_doorbell);
722 		ctrlr->shadow_doorbell = NULL;
723 	}
724 
725 	if (ctrlr->eventidx) {
726 		spdk_dma_free(ctrlr->eventidx);
727 		ctrlr->eventidx = NULL;
728 	}
729 }
730 
731 static void
732 nvme_ctrlr_set_doorbell_buffer_config_done(void *arg, const struct spdk_nvme_cpl *cpl)
733 {
734 	struct spdk_nvme_ctrlr *ctrlr = (struct spdk_nvme_ctrlr *)arg;
735 
736 	if (spdk_nvme_cpl_is_error(cpl)) {
737 		SPDK_WARNLOG("Doorbell buffer config failed\n");
738 	} else {
739 		SPDK_INFOLOG(SPDK_LOG_NVME, "NVMe controller: %s doorbell buffer config enabled\n",
740 			     ctrlr->trid.traddr);
741 	}
742 	nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_SET_KEEP_ALIVE_TIMEOUT,
743 			     ctrlr->opts.admin_timeout_ms);
744 }
745 
746 static int
747 nvme_ctrlr_set_doorbell_buffer_config(struct spdk_nvme_ctrlr *ctrlr)
748 {
749 	int rc = 0;
750 	uint64_t prp1, prp2;
751 
752 	if (!ctrlr->cdata.oacs.doorbell_buffer_config) {
753 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_SET_KEEP_ALIVE_TIMEOUT,
754 				     ctrlr->opts.admin_timeout_ms);
755 		return 0;
756 	}
757 
758 	if (ctrlr->trid.trtype != SPDK_NVME_TRANSPORT_PCIE) {
759 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_SET_KEEP_ALIVE_TIMEOUT,
760 				     ctrlr->opts.admin_timeout_ms);
761 		return 0;
762 	}
763 
764 	/* only 1 page size for doorbell buffer */
765 	ctrlr->shadow_doorbell = spdk_dma_zmalloc(ctrlr->page_size, ctrlr->page_size,
766 				 &prp1);
767 	if (ctrlr->shadow_doorbell == NULL) {
768 		rc = -ENOMEM;
769 		goto error;
770 	}
771 
772 	ctrlr->eventidx = spdk_dma_zmalloc(ctrlr->page_size, ctrlr->page_size, &prp2);
773 	if (ctrlr->eventidx == NULL) {
774 		rc = -ENOMEM;
775 		goto error;
776 	}
777 
778 	nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_WAIT_FOR_DB_BUF_CFG,
779 			     ctrlr->opts.admin_timeout_ms);
780 
781 	rc = nvme_ctrlr_cmd_doorbell_buffer_config(ctrlr, prp1, prp2,
782 			nvme_ctrlr_set_doorbell_buffer_config_done, ctrlr);
783 	if (rc != 0) {
784 		goto error;
785 	}
786 
787 	return 0;
788 
789 error:
790 	nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_ERROR, NVME_TIMEOUT_INFINITE);
791 	nvme_ctrlr_free_doorbell_buffer(ctrlr);
792 	return rc;
793 }
794 
795 int
796 spdk_nvme_ctrlr_reset(struct spdk_nvme_ctrlr *ctrlr)
797 {
798 	int rc = 0;
799 	struct spdk_nvme_qpair	*qpair;
800 	struct nvme_request	*req, *tmp;
801 
802 	nvme_robust_mutex_lock(&ctrlr->ctrlr_lock);
803 
804 	if (ctrlr->is_resetting || ctrlr->is_failed) {
805 		/*
806 		 * Controller is already resetting or has failed.  Return
807 		 *  immediately since there is no need to kick off another
808 		 *  reset in these cases.
809 		 */
810 		nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
811 		return 0;
812 	}
813 
814 	ctrlr->is_resetting = true;
815 
816 	SPDK_NOTICELOG("resetting controller\n");
817 
818 	/* Free all of the queued abort requests */
819 	STAILQ_FOREACH_SAFE(req, &ctrlr->queued_aborts, stailq, tmp) {
820 		STAILQ_REMOVE_HEAD(&ctrlr->queued_aborts, stailq);
821 		nvme_free_request(req);
822 		ctrlr->outstanding_aborts--;
823 	}
824 
825 	/* Disable all queues before disabling the controller hardware. */
826 	nvme_qpair_disable(ctrlr->adminq);
827 	TAILQ_FOREACH(qpair, &ctrlr->active_io_qpairs, tailq) {
828 		nvme_qpair_disable(qpair);
829 	}
830 
831 	/* Doorbell buffer config is invalid during reset */
832 	nvme_ctrlr_free_doorbell_buffer(ctrlr);
833 
834 	/* Set the state back to INIT to cause a full hardware reset. */
835 	nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_INIT, NVME_TIMEOUT_INFINITE);
836 
837 	while (ctrlr->state != NVME_CTRLR_STATE_READY) {
838 		if (nvme_ctrlr_process_init(ctrlr) != 0) {
839 			SPDK_ERRLOG("%s: controller reinitialization failed\n", __func__);
840 			nvme_ctrlr_fail(ctrlr, false);
841 			rc = -1;
842 			break;
843 		}
844 	}
845 
846 	if (!ctrlr->is_failed) {
847 		/* Reinitialize qpairs */
848 		TAILQ_FOREACH(qpair, &ctrlr->active_io_qpairs, tailq) {
849 			if (nvme_transport_ctrlr_reinit_io_qpair(ctrlr, qpair) != 0) {
850 				nvme_ctrlr_fail(ctrlr, false);
851 				rc = -1;
852 			}
853 		}
854 	}
855 
856 	ctrlr->is_resetting = false;
857 
858 	nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
859 
860 	return rc;
861 }
862 
863 static void
864 nvme_ctrlr_identify_done(void *arg, const struct spdk_nvme_cpl *cpl)
865 {
866 	struct spdk_nvme_ctrlr *ctrlr = (struct spdk_nvme_ctrlr *)arg;
867 
868 	if (spdk_nvme_cpl_is_error(cpl)) {
869 		SPDK_ERRLOG("nvme_identify_controller failed!\n");
870 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_ERROR, NVME_TIMEOUT_INFINITE);
871 		return;
872 	}
873 
874 	/*
875 	 * Use MDTS to ensure our default max_xfer_size doesn't exceed what the
876 	 *  controller supports.
877 	 */
878 	ctrlr->max_xfer_size = nvme_transport_ctrlr_get_max_xfer_size(ctrlr);
879 	SPDK_DEBUGLOG(SPDK_LOG_NVME, "transport max_xfer_size %u\n", ctrlr->max_xfer_size);
880 	if (ctrlr->cdata.mdts > 0) {
881 		ctrlr->max_xfer_size = spdk_min(ctrlr->max_xfer_size,
882 						ctrlr->min_page_size * (1 << (ctrlr->cdata.mdts)));
883 		SPDK_DEBUGLOG(SPDK_LOG_NVME, "MDTS max_xfer_size %u\n", ctrlr->max_xfer_size);
884 	}
885 
886 	SPDK_DEBUGLOG(SPDK_LOG_NVME, "CNTLID 0x%04" PRIx16 "\n", ctrlr->cdata.cntlid);
887 	if (ctrlr->trid.trtype == SPDK_NVME_TRANSPORT_PCIE) {
888 		ctrlr->cntlid = ctrlr->cdata.cntlid;
889 	} else {
890 		/*
891 		 * Fabrics controllers should already have CNTLID from the Connect command.
892 		 *
893 		 * If CNTLID from Connect doesn't match CNTLID in the Identify Controller data,
894 		 * trust the one from Connect.
895 		 */
896 		if (ctrlr->cntlid != ctrlr->cdata.cntlid) {
897 			SPDK_DEBUGLOG(SPDK_LOG_NVME,
898 				      "Identify CNTLID 0x%04" PRIx16 " != Connect CNTLID 0x%04" PRIx16 "\n",
899 				      ctrlr->cdata.cntlid, ctrlr->cntlid);
900 		}
901 	}
902 
903 	if (ctrlr->cdata.sgls.supported) {
904 		ctrlr->flags |= SPDK_NVME_CTRLR_SGL_SUPPORTED;
905 		ctrlr->max_sges = nvme_transport_ctrlr_get_max_sges(ctrlr);
906 	}
907 
908 	nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_SET_NUM_QUEUES,
909 			     ctrlr->opts.admin_timeout_ms);
910 }
911 
912 static int
913 nvme_ctrlr_identify(struct spdk_nvme_ctrlr *ctrlr)
914 {
915 	int	rc;
916 
917 	nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_WAIT_FOR_IDENTIFY,
918 			     ctrlr->opts.admin_timeout_ms);
919 
920 	rc = nvme_ctrlr_cmd_identify(ctrlr, SPDK_NVME_IDENTIFY_CTRLR, 0, 0,
921 				     &ctrlr->cdata, sizeof(ctrlr->cdata),
922 				     nvme_ctrlr_identify_done, ctrlr);
923 	if (rc != 0) {
924 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_ERROR, NVME_TIMEOUT_INFINITE);
925 		return rc;
926 	}
927 
928 	return 0;
929 }
930 
931 int
932 nvme_ctrlr_identify_active_ns(struct spdk_nvme_ctrlr *ctrlr)
933 {
934 	struct nvme_completion_poll_status	status;
935 	int					rc;
936 	uint32_t				i;
937 	uint32_t				num_pages;
938 	uint32_t				next_nsid = 0;
939 	uint32_t				*new_ns_list = NULL;
940 
941 	if (ctrlr->num_ns == 0) {
942 		spdk_dma_free(ctrlr->active_ns_list);
943 		ctrlr->active_ns_list = NULL;
944 
945 		return 0;
946 	}
947 
948 	/*
949 	 * The allocated size must be a multiple of sizeof(struct spdk_nvme_ns_list)
950 	 */
951 	num_pages = (ctrlr->num_ns * sizeof(new_ns_list[0]) - 1) / sizeof(struct spdk_nvme_ns_list) + 1;
952 	new_ns_list = spdk_dma_zmalloc(num_pages * sizeof(struct spdk_nvme_ns_list), ctrlr->page_size,
953 				       NULL);
954 	if (!new_ns_list) {
955 		SPDK_ERRLOG("Failed to allocate active_ns_list!\n");
956 		return -ENOMEM;
957 	}
958 
959 	if (ctrlr->vs.raw >= SPDK_NVME_VERSION(1, 1, 0) && !(ctrlr->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
960 		/*
961 		 * Iterate through the pages and fetch each chunk of 1024 namespaces until
962 		 * there are no more active namespaces
963 		 */
964 		for (i = 0; i < num_pages; i++) {
965 			rc = nvme_ctrlr_cmd_identify(ctrlr, SPDK_NVME_IDENTIFY_ACTIVE_NS_LIST, 0, next_nsid,
966 						     &new_ns_list[1024 * i], sizeof(struct spdk_nvme_ns_list),
967 						     nvme_completion_poll_cb, &status);
968 			if (rc != 0) {
969 				goto fail;
970 			}
971 			if (spdk_nvme_wait_for_completion(ctrlr->adminq, &status)) {
972 				SPDK_ERRLOG("nvme_ctrlr_cmd_identify_active_ns_list failed!\n");
973 				rc = -ENXIO;
974 				goto fail;
975 			}
976 			next_nsid = new_ns_list[1024 * i + 1023];
977 			if (next_nsid == 0) {
978 				/*
979 				 * No more active namespaces found, no need to fetch additional chunks
980 				 */
981 				break;
982 			}
983 		}
984 
985 	} else {
986 		/*
987 		 * Controller doesn't support active ns list CNS 0x02 so dummy up
988 		 * an active ns list
989 		 */
990 		for (i = 0; i < ctrlr->num_ns; i++) {
991 			new_ns_list[i] = i + 1;
992 		}
993 	}
994 
995 	/*
996 	 * Now that that the list is properly setup, we can swap it in to the ctrlr and
997 	 * free up the previous one.
998 	 */
999 	spdk_dma_free(ctrlr->active_ns_list);
1000 	ctrlr->active_ns_list = new_ns_list;
1001 
1002 	return 0;
1003 fail:
1004 	spdk_dma_free(new_ns_list);
1005 	return rc;
1006 }
1007 
1008 static void
1009 nvme_ctrlr_identify_ns_async_done(void *arg, const struct spdk_nvme_cpl *cpl)
1010 {
1011 	struct spdk_nvme_ns *ns = (struct spdk_nvme_ns *)arg;
1012 	struct spdk_nvme_ctrlr *ctrlr = ns->ctrlr;
1013 	uint32_t nsid;
1014 	int rc;
1015 
1016 	if (spdk_nvme_cpl_is_error(cpl)) {
1017 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_ERROR, NVME_TIMEOUT_INFINITE);
1018 		return;
1019 	} else {
1020 		nvme_ns_set_identify_data(ns);
1021 	}
1022 
1023 	/* move on to the next active NS */
1024 	nsid = spdk_nvme_ctrlr_get_next_active_ns(ctrlr, ns->id);
1025 	ns = spdk_nvme_ctrlr_get_ns(ctrlr, nsid);
1026 	if (ns == NULL) {
1027 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_IDENTIFY_ID_DESCS,
1028 				     ctrlr->opts.admin_timeout_ms);
1029 		return;
1030 	}
1031 	ns->ctrlr = ctrlr;
1032 	ns->id = nsid;
1033 
1034 	rc = nvme_ctrlr_identify_ns_async(ns);
1035 	if (rc) {
1036 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_ERROR, NVME_TIMEOUT_INFINITE);
1037 	}
1038 }
1039 
1040 static int
1041 nvme_ctrlr_identify_ns_async(struct spdk_nvme_ns *ns)
1042 {
1043 	struct spdk_nvme_ctrlr *ctrlr = ns->ctrlr;
1044 	struct spdk_nvme_ns_data *nsdata;
1045 
1046 	nsdata = &ctrlr->nsdata[ns->id - 1];
1047 
1048 	nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_WAIT_FOR_IDENTIFY_NS,
1049 			     ctrlr->opts.admin_timeout_ms);
1050 	return nvme_ctrlr_cmd_identify(ns->ctrlr, SPDK_NVME_IDENTIFY_NS, 0, ns->id,
1051 				       nsdata, sizeof(*nsdata),
1052 				       nvme_ctrlr_identify_ns_async_done, ns);
1053 }
1054 
1055 static int
1056 nvme_ctrlr_identify_namespaces(struct spdk_nvme_ctrlr *ctrlr)
1057 {
1058 	uint32_t nsid;
1059 	struct spdk_nvme_ns *ns;
1060 	int rc;
1061 
1062 	nsid = spdk_nvme_ctrlr_get_first_active_ns(ctrlr);
1063 	ns = spdk_nvme_ctrlr_get_ns(ctrlr, nsid);
1064 	if (ns == NULL) {
1065 		/* No active NS, move on to the next state */
1066 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_CONFIGURE_AER,
1067 				     ctrlr->opts.admin_timeout_ms);
1068 		return 0;
1069 	}
1070 
1071 	ns->ctrlr = ctrlr;
1072 	ns->id = nsid;
1073 
1074 	rc = nvme_ctrlr_identify_ns_async(ns);
1075 	if (rc) {
1076 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_ERROR, NVME_TIMEOUT_INFINITE);
1077 	}
1078 
1079 	return rc;
1080 }
1081 
1082 static void
1083 nvme_ctrlr_identify_id_desc_async_done(void *arg, const struct spdk_nvme_cpl *cpl)
1084 {
1085 	struct spdk_nvme_ns *ns = (struct spdk_nvme_ns *)arg;
1086 	struct spdk_nvme_ctrlr *ctrlr = ns->ctrlr;
1087 	uint32_t nsid;
1088 	int rc;
1089 
1090 	if (spdk_nvme_cpl_is_error(cpl)) {
1091 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_CONFIGURE_AER,
1092 				     ctrlr->opts.admin_timeout_ms);
1093 		return;
1094 	}
1095 
1096 	/* move on to the next active NS */
1097 	nsid = spdk_nvme_ctrlr_get_next_active_ns(ctrlr, ns->id);
1098 	ns = spdk_nvme_ctrlr_get_ns(ctrlr, nsid);
1099 	if (ns == NULL) {
1100 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_CONFIGURE_AER,
1101 				     ctrlr->opts.admin_timeout_ms);
1102 		return;
1103 	}
1104 
1105 	rc = nvme_ctrlr_identify_id_desc_async(ns);
1106 	if (rc) {
1107 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_ERROR, NVME_TIMEOUT_INFINITE);
1108 	}
1109 }
1110 
1111 static int
1112 nvme_ctrlr_identify_id_desc_async(struct spdk_nvme_ns *ns)
1113 {
1114 	struct spdk_nvme_ctrlr *ctrlr = ns->ctrlr;
1115 
1116 	memset(ns->id_desc_list, 0, sizeof(ns->id_desc_list));
1117 
1118 	nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_WAIT_FOR_IDENTIFY_ID_DESCS,
1119 			     ctrlr->opts.admin_timeout_ms);
1120 	return nvme_ctrlr_cmd_identify(ns->ctrlr, SPDK_NVME_IDENTIFY_NS_ID_DESCRIPTOR_LIST,
1121 				       0, ns->id, ns->id_desc_list, sizeof(ns->id_desc_list),
1122 				       nvme_ctrlr_identify_id_desc_async_done, ns);
1123 }
1124 
1125 static int
1126 nvme_ctrlr_identify_id_desc_namespaces(struct spdk_nvme_ctrlr *ctrlr)
1127 {
1128 	uint32_t nsid;
1129 	struct spdk_nvme_ns *ns;
1130 	int rc;
1131 
1132 	if (ctrlr->vs.raw < SPDK_NVME_VERSION(1, 3, 0) ||
1133 	    (ctrlr->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
1134 		SPDK_DEBUGLOG(SPDK_LOG_NVME, "Version < 1.3; not attempting to retrieve NS ID Descriptor List\n");
1135 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_CONFIGURE_AER,
1136 				     ctrlr->opts.admin_timeout_ms);
1137 		return 0;
1138 	}
1139 
1140 	nsid = spdk_nvme_ctrlr_get_first_active_ns(ctrlr);
1141 	ns = spdk_nvme_ctrlr_get_ns(ctrlr, nsid);
1142 	if (ns == NULL) {
1143 		/* No active NS, move on to the next state */
1144 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_CONFIGURE_AER,
1145 				     ctrlr->opts.admin_timeout_ms);
1146 		return 0;
1147 	}
1148 
1149 	rc = nvme_ctrlr_identify_id_desc_async(ns);
1150 	if (rc) {
1151 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_ERROR, NVME_TIMEOUT_INFINITE);
1152 	}
1153 
1154 	return rc;
1155 }
1156 
1157 static void
1158 nvme_ctrlr_set_num_queues_done(void *arg, const struct spdk_nvme_cpl *cpl)
1159 {
1160 	struct spdk_nvme_ctrlr *ctrlr = (struct spdk_nvme_ctrlr *)arg;
1161 
1162 	if (spdk_nvme_cpl_is_error(cpl)) {
1163 		SPDK_ERRLOG("Set Features - Number of Queues failed!\n");
1164 	}
1165 	nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_GET_NUM_QUEUES,
1166 			     ctrlr->opts.admin_timeout_ms);
1167 }
1168 
1169 static int
1170 nvme_ctrlr_set_num_queues(struct spdk_nvme_ctrlr *ctrlr)
1171 {
1172 	int rc;
1173 
1174 	if (ctrlr->opts.num_io_queues > SPDK_NVME_MAX_IO_QUEUES) {
1175 		SPDK_NOTICELOG("Limiting requested num_io_queues %u to max %d\n",
1176 			       ctrlr->opts.num_io_queues, SPDK_NVME_MAX_IO_QUEUES);
1177 		ctrlr->opts.num_io_queues = SPDK_NVME_MAX_IO_QUEUES;
1178 	} else if (ctrlr->opts.num_io_queues < 1) {
1179 		SPDK_NOTICELOG("Requested num_io_queues 0, increasing to 1\n");
1180 		ctrlr->opts.num_io_queues = 1;
1181 	}
1182 
1183 	nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_WAIT_FOR_SET_NUM_QUEUES,
1184 			     ctrlr->opts.admin_timeout_ms);
1185 
1186 	rc = nvme_ctrlr_cmd_set_num_queues(ctrlr, ctrlr->opts.num_io_queues,
1187 					   nvme_ctrlr_set_num_queues_done, ctrlr);
1188 	if (rc != 0) {
1189 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_ERROR, NVME_TIMEOUT_INFINITE);
1190 		return rc;
1191 	}
1192 
1193 	return 0;
1194 }
1195 
1196 static void
1197 nvme_ctrlr_get_num_queues_done(void *arg, const struct spdk_nvme_cpl *cpl)
1198 {
1199 	uint32_t cq_allocated, sq_allocated, min_allocated, i;
1200 	struct spdk_nvme_ctrlr *ctrlr = (struct spdk_nvme_ctrlr *)arg;
1201 
1202 	if (spdk_nvme_cpl_is_error(cpl)) {
1203 		SPDK_ERRLOG("Get Features - Number of Queues failed!\n");
1204 		ctrlr->opts.num_io_queues = 0;
1205 	} else {
1206 		/*
1207 		 * Data in cdw0 is 0-based.
1208 		 * Lower 16-bits indicate number of submission queues allocated.
1209 		 * Upper 16-bits indicate number of completion queues allocated.
1210 		 */
1211 		sq_allocated = (cpl->cdw0 & 0xFFFF) + 1;
1212 		cq_allocated = (cpl->cdw0 >> 16) + 1;
1213 
1214 		/*
1215 		 * For 1:1 queue mapping, set number of allocated queues to be minimum of
1216 		 * submission and completion queues.
1217 		 */
1218 		min_allocated = spdk_min(sq_allocated, cq_allocated);
1219 
1220 		/* Set number of queues to be minimum of requested and actually allocated. */
1221 		ctrlr->opts.num_io_queues = spdk_min(min_allocated, ctrlr->opts.num_io_queues);
1222 	}
1223 
1224 	ctrlr->free_io_qids = spdk_bit_array_create(ctrlr->opts.num_io_queues + 1);
1225 	if (ctrlr->free_io_qids == NULL) {
1226 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_ERROR, NVME_TIMEOUT_INFINITE);
1227 		return;
1228 	}
1229 
1230 	/* Initialize list of free I/O queue IDs. QID 0 is the admin queue. */
1231 	spdk_bit_array_clear(ctrlr->free_io_qids, 0);
1232 	for (i = 1; i <= ctrlr->opts.num_io_queues; i++) {
1233 		spdk_bit_array_set(ctrlr->free_io_qids, i);
1234 	}
1235 	nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_CONSTRUCT_NS,
1236 			     ctrlr->opts.admin_timeout_ms);
1237 }
1238 
1239 static int
1240 nvme_ctrlr_get_num_queues(struct spdk_nvme_ctrlr *ctrlr)
1241 {
1242 	int rc;
1243 
1244 	nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_WAIT_FOR_GET_NUM_QUEUES,
1245 			     ctrlr->opts.admin_timeout_ms);
1246 
1247 	/* Obtain the number of queues allocated using Get Features. */
1248 	rc = nvme_ctrlr_cmd_get_num_queues(ctrlr, nvme_ctrlr_get_num_queues_done, ctrlr);
1249 	if (rc != 0) {
1250 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_ERROR, NVME_TIMEOUT_INFINITE);
1251 		return rc;
1252 	}
1253 
1254 	return 0;
1255 }
1256 
1257 static void
1258 nvme_ctrlr_set_keep_alive_timeout_done(void *arg, const struct spdk_nvme_cpl *cpl)
1259 {
1260 	uint32_t keep_alive_interval_ms;
1261 	struct spdk_nvme_ctrlr *ctrlr = (struct spdk_nvme_ctrlr *)arg;
1262 
1263 	if (spdk_nvme_cpl_is_error(cpl)) {
1264 		SPDK_ERRLOG("Keep alive timeout Get Feature failed: SC %x SCT %x\n",
1265 			    cpl->status.sc, cpl->status.sct);
1266 		ctrlr->opts.keep_alive_timeout_ms = 0;
1267 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_ERROR, NVME_TIMEOUT_INFINITE);
1268 		return;
1269 	}
1270 
1271 	if (ctrlr->opts.keep_alive_timeout_ms != cpl->cdw0) {
1272 		SPDK_DEBUGLOG(SPDK_LOG_NVME, "Controller adjusted keep alive timeout to %u ms\n",
1273 			      cpl->cdw0);
1274 	}
1275 
1276 	ctrlr->opts.keep_alive_timeout_ms = cpl->cdw0;
1277 
1278 	keep_alive_interval_ms = ctrlr->opts.keep_alive_timeout_ms / 2;
1279 	if (keep_alive_interval_ms == 0) {
1280 		keep_alive_interval_ms = 1;
1281 	}
1282 	SPDK_DEBUGLOG(SPDK_LOG_NVME, "Sending keep alive every %u ms\n", keep_alive_interval_ms);
1283 
1284 	ctrlr->keep_alive_interval_ticks = (keep_alive_interval_ms * spdk_get_ticks_hz()) / UINT64_C(1000);
1285 
1286 	/* Schedule the first Keep Alive to be sent as soon as possible. */
1287 	ctrlr->next_keep_alive_tick = spdk_get_ticks();
1288 	nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_SET_HOST_ID,
1289 			     ctrlr->opts.admin_timeout_ms);
1290 }
1291 
1292 static int
1293 nvme_ctrlr_set_keep_alive_timeout(struct spdk_nvme_ctrlr *ctrlr)
1294 {
1295 	int rc;
1296 
1297 	if (ctrlr->opts.keep_alive_timeout_ms == 0) {
1298 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_SET_HOST_ID,
1299 				     ctrlr->opts.admin_timeout_ms);
1300 		return 0;
1301 	}
1302 
1303 	if (ctrlr->cdata.kas == 0) {
1304 		SPDK_DEBUGLOG(SPDK_LOG_NVME, "Controller KAS is 0 - not enabling Keep Alive\n");
1305 		ctrlr->opts.keep_alive_timeout_ms = 0;
1306 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_SET_HOST_ID,
1307 				     ctrlr->opts.admin_timeout_ms);
1308 		return 0;
1309 	}
1310 
1311 	nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_WAIT_FOR_KEEP_ALIVE_TIMEOUT,
1312 			     ctrlr->opts.admin_timeout_ms);
1313 
1314 	/* Retrieve actual keep alive timeout, since the controller may have adjusted it. */
1315 	rc = spdk_nvme_ctrlr_cmd_get_feature(ctrlr, SPDK_NVME_FEAT_KEEP_ALIVE_TIMER, 0, NULL, 0,
1316 					     nvme_ctrlr_set_keep_alive_timeout_done, ctrlr);
1317 	if (rc != 0) {
1318 		SPDK_ERRLOG("Keep alive timeout Get Feature failed: %d\n", rc);
1319 		ctrlr->opts.keep_alive_timeout_ms = 0;
1320 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_ERROR, NVME_TIMEOUT_INFINITE);
1321 		return rc;
1322 	}
1323 
1324 	return 0;
1325 }
1326 
1327 static void
1328 nvme_ctrlr_set_host_id_done(void *arg, const struct spdk_nvme_cpl *cpl)
1329 {
1330 	struct spdk_nvme_ctrlr *ctrlr = (struct spdk_nvme_ctrlr *)arg;
1331 
1332 	if (spdk_nvme_cpl_is_error(cpl)) {
1333 		/*
1334 		 * Treat Set Features - Host ID failure as non-fatal, since the Host ID feature
1335 		 * is optional.
1336 		 */
1337 		SPDK_WARNLOG("Set Features - Host ID failed: SC 0x%x SCT 0x%x\n",
1338 			     cpl->status.sc, cpl->status.sct);
1339 	} else {
1340 		SPDK_DEBUGLOG(SPDK_LOG_NVME, "Set Features - Host ID was successful\n");
1341 	}
1342 
1343 	nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_READY, NVME_TIMEOUT_INFINITE);
1344 }
1345 
1346 static int
1347 nvme_ctrlr_set_host_id(struct spdk_nvme_ctrlr *ctrlr)
1348 {
1349 	uint8_t *host_id;
1350 	uint32_t host_id_size;
1351 	int rc;
1352 
1353 	if (ctrlr->trid.trtype != SPDK_NVME_TRANSPORT_PCIE) {
1354 		/*
1355 		 * NVMe-oF sends the host ID during Connect and doesn't allow
1356 		 * Set Features - Host Identifier after Connect, so we don't need to do anything here.
1357 		 */
1358 		SPDK_DEBUGLOG(SPDK_LOG_NVME, "NVMe-oF transport - not sending Set Features - Host ID\n");
1359 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_READY, NVME_TIMEOUT_INFINITE);
1360 		return 0;
1361 	}
1362 
1363 	if (ctrlr->cdata.ctratt.host_id_exhid_supported) {
1364 		SPDK_DEBUGLOG(SPDK_LOG_NVME, "Using 128-bit extended host identifier\n");
1365 		host_id = ctrlr->opts.extended_host_id;
1366 		host_id_size = sizeof(ctrlr->opts.extended_host_id);
1367 	} else {
1368 		SPDK_DEBUGLOG(SPDK_LOG_NVME, "Using 64-bit host identifier\n");
1369 		host_id = ctrlr->opts.host_id;
1370 		host_id_size = sizeof(ctrlr->opts.host_id);
1371 	}
1372 
1373 	/* If the user specified an all-zeroes host identifier, don't send the command. */
1374 	if (spdk_mem_all_zero(host_id, host_id_size)) {
1375 		SPDK_DEBUGLOG(SPDK_LOG_NVME,
1376 			      "User did not specify host ID - not sending Set Features - Host ID\n");
1377 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_READY, NVME_TIMEOUT_INFINITE);
1378 		return 0;
1379 	}
1380 
1381 	SPDK_LOGDUMP(SPDK_LOG_NVME, "host_id", host_id, host_id_size);
1382 
1383 	nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_WAIT_FOR_HOST_ID,
1384 			     ctrlr->opts.admin_timeout_ms);
1385 
1386 	rc = nvme_ctrlr_cmd_set_host_id(ctrlr, host_id, host_id_size, nvme_ctrlr_set_host_id_done, ctrlr);
1387 	if (rc != 0) {
1388 		SPDK_ERRLOG("Set Features - Host ID failed: %d\n", rc);
1389 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_ERROR, NVME_TIMEOUT_INFINITE);
1390 		return rc;
1391 	}
1392 
1393 	return 0;
1394 }
1395 
1396 static void
1397 nvme_ctrlr_destruct_namespaces(struct spdk_nvme_ctrlr *ctrlr)
1398 {
1399 	if (ctrlr->ns) {
1400 		uint32_t i, num_ns = ctrlr->num_ns;
1401 
1402 		for (i = 0; i < num_ns; i++) {
1403 			nvme_ns_destruct(&ctrlr->ns[i]);
1404 		}
1405 
1406 		spdk_free(ctrlr->ns);
1407 		ctrlr->ns = NULL;
1408 		ctrlr->num_ns = 0;
1409 	}
1410 
1411 	if (ctrlr->nsdata) {
1412 		spdk_free(ctrlr->nsdata);
1413 		ctrlr->nsdata = NULL;
1414 	}
1415 
1416 	spdk_dma_free(ctrlr->active_ns_list);
1417 	ctrlr->active_ns_list = NULL;
1418 }
1419 
1420 static void
1421 nvme_ctrlr_update_namespaces(struct spdk_nvme_ctrlr *ctrlr)
1422 {
1423 	uint32_t i, nn = ctrlr->cdata.nn;
1424 	struct spdk_nvme_ns_data *nsdata;
1425 
1426 	for (i = 0; i < nn; i++) {
1427 		struct spdk_nvme_ns	*ns = &ctrlr->ns[i];
1428 		uint32_t		nsid = i + 1;
1429 		nsdata			= &ctrlr->nsdata[nsid - 1];
1430 
1431 		if ((nsdata->ncap == 0) && spdk_nvme_ctrlr_is_active_ns(ctrlr, nsid)) {
1432 			if (nvme_ns_construct(ns, nsid, ctrlr) != 0) {
1433 				continue;
1434 			}
1435 		}
1436 
1437 		if (nsdata->ncap && !spdk_nvme_ctrlr_is_active_ns(ctrlr, nsid)) {
1438 			nvme_ns_destruct(ns);
1439 		}
1440 	}
1441 }
1442 
1443 static int
1444 nvme_ctrlr_construct_namespaces(struct spdk_nvme_ctrlr *ctrlr)
1445 {
1446 	int rc = 0;
1447 	uint32_t nn = ctrlr->cdata.nn;
1448 	uint64_t phys_addr = 0;
1449 
1450 	/* ctrlr->num_ns may be 0 (startup) or a different number of namespaces (reset),
1451 	 * so check if we need to reallocate.
1452 	 */
1453 	if (nn != ctrlr->num_ns) {
1454 		nvme_ctrlr_destruct_namespaces(ctrlr);
1455 
1456 		if (nn == 0) {
1457 			SPDK_WARNLOG("controller has 0 namespaces\n");
1458 			return 0;
1459 		}
1460 
1461 		ctrlr->ns = spdk_zmalloc(nn * sizeof(struct spdk_nvme_ns), 64,
1462 					 &phys_addr, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_SHARE);
1463 		if (ctrlr->ns == NULL) {
1464 			rc = -ENOMEM;
1465 			goto fail;
1466 		}
1467 
1468 		ctrlr->nsdata = spdk_zmalloc(nn * sizeof(struct spdk_nvme_ns_data), 64,
1469 					     &phys_addr, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_SHARE | SPDK_MALLOC_DMA);
1470 		if (ctrlr->nsdata == NULL) {
1471 			rc = -ENOMEM;
1472 			goto fail;
1473 		}
1474 
1475 		ctrlr->num_ns = nn;
1476 	}
1477 
1478 	return 0;
1479 
1480 fail:
1481 	nvme_ctrlr_destruct_namespaces(ctrlr);
1482 	return rc;
1483 }
1484 
1485 static void
1486 nvme_ctrlr_async_event_cb(void *arg, const struct spdk_nvme_cpl *cpl)
1487 {
1488 	struct nvme_async_event_request	*aer = arg;
1489 	struct spdk_nvme_ctrlr		*ctrlr = aer->ctrlr;
1490 	struct spdk_nvme_ctrlr_process	*active_proc;
1491 	union spdk_nvme_async_event_completion	event;
1492 	int					rc;
1493 
1494 	if (cpl->status.sct == SPDK_NVME_SCT_GENERIC &&
1495 	    cpl->status.sc == SPDK_NVME_SC_ABORTED_SQ_DELETION) {
1496 		/*
1497 		 *  This is simulated when controller is being shut down, to
1498 		 *  effectively abort outstanding asynchronous event requests
1499 		 *  and make sure all memory is freed.  Do not repost the
1500 		 *  request in this case.
1501 		 */
1502 		return;
1503 	}
1504 
1505 	if (cpl->status.sct == SPDK_NVME_SCT_COMMAND_SPECIFIC &&
1506 	    cpl->status.sc == SPDK_NVME_SC_ASYNC_EVENT_REQUEST_LIMIT_EXCEEDED) {
1507 		/*
1508 		 *  SPDK will only send as many AERs as the device says it supports,
1509 		 *  so this status code indicates an out-of-spec device.  Do not repost
1510 		 *  the request in this case.
1511 		 */
1512 		SPDK_ERRLOG("Controller appears out-of-spec for asynchronous event request\n"
1513 			    "handling.  Do not repost this AER.\n");
1514 		return;
1515 	}
1516 
1517 	event.raw = cpl->cdw0;
1518 	if ((event.bits.async_event_type == SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE) &&
1519 	    (event.bits.async_event_info == SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGED)) {
1520 		rc = nvme_ctrlr_identify_active_ns(ctrlr);
1521 		if (rc) {
1522 			return;
1523 		}
1524 		nvme_ctrlr_update_namespaces(ctrlr);
1525 	}
1526 
1527 	active_proc = spdk_nvme_ctrlr_get_current_process(ctrlr);
1528 	if (active_proc && active_proc->aer_cb_fn) {
1529 		active_proc->aer_cb_fn(active_proc->aer_cb_arg, cpl);
1530 	}
1531 
1532 	/* If the ctrlr is already shutdown, we should not send aer again */
1533 	if (ctrlr->is_shutdown) {
1534 		return;
1535 	}
1536 
1537 	/*
1538 	 * Repost another asynchronous event request to replace the one
1539 	 *  that just completed.
1540 	 */
1541 	if (nvme_ctrlr_construct_and_submit_aer(ctrlr, aer)) {
1542 		/*
1543 		 * We can't do anything to recover from a failure here,
1544 		 * so just print a warning message and leave the AER unsubmitted.
1545 		 */
1546 		SPDK_ERRLOG("resubmitting AER failed!\n");
1547 	}
1548 }
1549 
1550 static int
1551 nvme_ctrlr_construct_and_submit_aer(struct spdk_nvme_ctrlr *ctrlr,
1552 				    struct nvme_async_event_request *aer)
1553 {
1554 	struct nvme_request *req;
1555 
1556 	aer->ctrlr = ctrlr;
1557 	req = nvme_allocate_request_null(ctrlr->adminq, nvme_ctrlr_async_event_cb, aer);
1558 	aer->req = req;
1559 	if (req == NULL) {
1560 		return -1;
1561 	}
1562 
1563 	req->cmd.opc = SPDK_NVME_OPC_ASYNC_EVENT_REQUEST;
1564 	return nvme_ctrlr_submit_admin_request(ctrlr, req);
1565 }
1566 
1567 static void
1568 nvme_ctrlr_configure_aer_done(void *arg, const struct spdk_nvme_cpl *cpl)
1569 {
1570 	struct nvme_async_event_request		*aer;
1571 	int					rc;
1572 	uint32_t				i;
1573 	struct spdk_nvme_ctrlr *ctrlr =	(struct spdk_nvme_ctrlr *)arg;
1574 
1575 	if (spdk_nvme_cpl_is_error(cpl)) {
1576 		SPDK_NOTICELOG("nvme_ctrlr_configure_aer failed!\n");
1577 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_SET_SUPPORTED_LOG_PAGES,
1578 				     ctrlr->opts.admin_timeout_ms);
1579 		return;
1580 	}
1581 
1582 	/* aerl is a zero-based value, so we need to add 1 here. */
1583 	ctrlr->num_aers = spdk_min(NVME_MAX_ASYNC_EVENTS, (ctrlr->cdata.aerl + 1));
1584 
1585 	for (i = 0; i < ctrlr->num_aers; i++) {
1586 		aer = &ctrlr->aer[i];
1587 		rc = nvme_ctrlr_construct_and_submit_aer(ctrlr, aer);
1588 		if (rc) {
1589 			SPDK_ERRLOG("nvme_ctrlr_construct_and_submit_aer failed!\n");
1590 			nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_ERROR, NVME_TIMEOUT_INFINITE);
1591 			return;
1592 		}
1593 	}
1594 	nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_SET_SUPPORTED_LOG_PAGES,
1595 			     ctrlr->opts.admin_timeout_ms);
1596 }
1597 
1598 static int
1599 nvme_ctrlr_configure_aer(struct spdk_nvme_ctrlr *ctrlr)
1600 {
1601 	union spdk_nvme_feat_async_event_configuration	config;
1602 	int						rc;
1603 
1604 	config.raw = 0;
1605 	config.bits.crit_warn.bits.available_spare = 1;
1606 	config.bits.crit_warn.bits.temperature = 1;
1607 	config.bits.crit_warn.bits.device_reliability = 1;
1608 	config.bits.crit_warn.bits.read_only = 1;
1609 	config.bits.crit_warn.bits.volatile_memory_backup = 1;
1610 
1611 	if (ctrlr->vs.raw >= SPDK_NVME_VERSION(1, 2, 0)) {
1612 		if (ctrlr->cdata.oaes.ns_attribute_notices) {
1613 			config.bits.ns_attr_notice = 1;
1614 		}
1615 		if (ctrlr->cdata.oaes.fw_activation_notices) {
1616 			config.bits.fw_activation_notice = 1;
1617 		}
1618 	}
1619 	if (ctrlr->vs.raw >= SPDK_NVME_VERSION(1, 3, 0) && ctrlr->cdata.lpa.telemetry) {
1620 		config.bits.telemetry_log_notice = 1;
1621 	}
1622 
1623 	nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_WAIT_FOR_CONFIGURE_AER,
1624 			     ctrlr->opts.admin_timeout_ms);
1625 
1626 	rc = nvme_ctrlr_cmd_set_async_event_config(ctrlr, config,
1627 			nvme_ctrlr_configure_aer_done,
1628 			ctrlr);
1629 	if (rc != 0) {
1630 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_ERROR, NVME_TIMEOUT_INFINITE);
1631 		return rc;
1632 	}
1633 
1634 	return 0;
1635 }
1636 
1637 struct spdk_nvme_ctrlr_process *
1638 spdk_nvme_ctrlr_get_process(struct spdk_nvme_ctrlr *ctrlr, pid_t pid)
1639 {
1640 	struct spdk_nvme_ctrlr_process	*active_proc;
1641 
1642 	TAILQ_FOREACH(active_proc, &ctrlr->active_procs, tailq) {
1643 		if (active_proc->pid == pid) {
1644 			return active_proc;
1645 		}
1646 	}
1647 
1648 	return NULL;
1649 }
1650 
1651 struct spdk_nvme_ctrlr_process *
1652 spdk_nvme_ctrlr_get_current_process(struct spdk_nvme_ctrlr *ctrlr)
1653 {
1654 	return spdk_nvme_ctrlr_get_process(ctrlr, getpid());
1655 }
1656 
1657 /**
1658  * This function will be called when a process is using the controller.
1659  *  1. For the primary process, it is called when constructing the controller.
1660  *  2. For the secondary process, it is called at probing the controller.
1661  * Note: will check whether the process is already added for the same process.
1662  */
1663 int
1664 nvme_ctrlr_add_process(struct spdk_nvme_ctrlr *ctrlr, void *devhandle)
1665 {
1666 	struct spdk_nvme_ctrlr_process	*ctrlr_proc;
1667 	pid_t				pid = getpid();
1668 
1669 	/* Check whether the process is already added or not */
1670 	if (spdk_nvme_ctrlr_get_process(ctrlr, pid)) {
1671 		return 0;
1672 	}
1673 
1674 	/* Initialize the per process properties for this ctrlr */
1675 	ctrlr_proc = spdk_zmalloc(sizeof(struct spdk_nvme_ctrlr_process),
1676 				  64, NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_SHARE);
1677 	if (ctrlr_proc == NULL) {
1678 		SPDK_ERRLOG("failed to allocate memory to track the process props\n");
1679 
1680 		return -1;
1681 	}
1682 
1683 	ctrlr_proc->is_primary = spdk_process_is_primary();
1684 	ctrlr_proc->pid = pid;
1685 	STAILQ_INIT(&ctrlr_proc->active_reqs);
1686 	ctrlr_proc->devhandle = devhandle;
1687 	ctrlr_proc->ref = 0;
1688 	TAILQ_INIT(&ctrlr_proc->allocated_io_qpairs);
1689 
1690 	TAILQ_INSERT_TAIL(&ctrlr->active_procs, ctrlr_proc, tailq);
1691 
1692 	return 0;
1693 }
1694 
1695 /**
1696  * This function will be called when the process detaches the controller.
1697  * Note: the ctrlr_lock must be held when calling this function.
1698  */
1699 static void
1700 nvme_ctrlr_remove_process(struct spdk_nvme_ctrlr *ctrlr,
1701 			  struct spdk_nvme_ctrlr_process *proc)
1702 {
1703 	struct spdk_nvme_qpair	*qpair, *tmp_qpair;
1704 
1705 	assert(STAILQ_EMPTY(&proc->active_reqs));
1706 
1707 	TAILQ_FOREACH_SAFE(qpair, &proc->allocated_io_qpairs, per_process_tailq, tmp_qpair) {
1708 		spdk_nvme_ctrlr_free_io_qpair(qpair);
1709 	}
1710 
1711 	TAILQ_REMOVE(&ctrlr->active_procs, proc, tailq);
1712 
1713 	if (ctrlr->trid.trtype == SPDK_NVME_TRANSPORT_PCIE) {
1714 		spdk_pci_device_detach(proc->devhandle);
1715 	}
1716 
1717 	spdk_dma_free(proc);
1718 }
1719 
1720 /**
1721  * This function will be called when the process exited unexpectedly
1722  *  in order to free any incomplete nvme request, allocated IO qpairs
1723  *  and allocated memory.
1724  * Note: the ctrlr_lock must be held when calling this function.
1725  */
1726 static void
1727 nvme_ctrlr_cleanup_process(struct spdk_nvme_ctrlr_process *proc)
1728 {
1729 	struct nvme_request	*req, *tmp_req;
1730 	struct spdk_nvme_qpair	*qpair, *tmp_qpair;
1731 
1732 	STAILQ_FOREACH_SAFE(req, &proc->active_reqs, stailq, tmp_req) {
1733 		STAILQ_REMOVE(&proc->active_reqs, req, nvme_request, stailq);
1734 
1735 		assert(req->pid == proc->pid);
1736 
1737 		nvme_free_request(req);
1738 	}
1739 
1740 	TAILQ_FOREACH_SAFE(qpair, &proc->allocated_io_qpairs, per_process_tailq, tmp_qpair) {
1741 		TAILQ_REMOVE(&proc->allocated_io_qpairs, qpair, per_process_tailq);
1742 
1743 		/*
1744 		 * The process may have been killed while some qpairs were in their
1745 		 *  completion context.  Clear that flag here to allow these IO
1746 		 *  qpairs to be deleted.
1747 		 */
1748 		qpair->in_completion_context = 0;
1749 
1750 		qpair->no_deletion_notification_needed = 1;
1751 
1752 		spdk_nvme_ctrlr_free_io_qpair(qpair);
1753 	}
1754 
1755 	spdk_dma_free(proc);
1756 }
1757 
1758 /**
1759  * This function will be called when destructing the controller.
1760  *  1. There is no more admin request on this controller.
1761  *  2. Clean up any left resource allocation when its associated process is gone.
1762  */
1763 void
1764 nvme_ctrlr_free_processes(struct spdk_nvme_ctrlr *ctrlr)
1765 {
1766 	struct spdk_nvme_ctrlr_process	*active_proc, *tmp;
1767 
1768 	/* Free all the processes' properties and make sure no pending admin IOs */
1769 	TAILQ_FOREACH_SAFE(active_proc, &ctrlr->active_procs, tailq, tmp) {
1770 		TAILQ_REMOVE(&ctrlr->active_procs, active_proc, tailq);
1771 
1772 		assert(STAILQ_EMPTY(&active_proc->active_reqs));
1773 
1774 		spdk_free(active_proc);
1775 	}
1776 }
1777 
1778 /**
1779  * This function will be called when any other process attaches or
1780  *  detaches the controller in order to cleanup those unexpectedly
1781  *  terminated processes.
1782  * Note: the ctrlr_lock must be held when calling this function.
1783  */
1784 static int
1785 nvme_ctrlr_remove_inactive_proc(struct spdk_nvme_ctrlr *ctrlr)
1786 {
1787 	struct spdk_nvme_ctrlr_process	*active_proc, *tmp;
1788 	int				active_proc_count = 0;
1789 
1790 	TAILQ_FOREACH_SAFE(active_proc, &ctrlr->active_procs, tailq, tmp) {
1791 		if ((kill(active_proc->pid, 0) == -1) && (errno == ESRCH)) {
1792 			SPDK_ERRLOG("process %d terminated unexpected\n", active_proc->pid);
1793 
1794 			TAILQ_REMOVE(&ctrlr->active_procs, active_proc, tailq);
1795 
1796 			nvme_ctrlr_cleanup_process(active_proc);
1797 		} else {
1798 			active_proc_count++;
1799 		}
1800 	}
1801 
1802 	return active_proc_count;
1803 }
1804 
1805 void
1806 nvme_ctrlr_proc_get_ref(struct spdk_nvme_ctrlr *ctrlr)
1807 {
1808 	struct spdk_nvme_ctrlr_process	*active_proc;
1809 
1810 	nvme_robust_mutex_lock(&ctrlr->ctrlr_lock);
1811 
1812 	nvme_ctrlr_remove_inactive_proc(ctrlr);
1813 
1814 	active_proc = spdk_nvme_ctrlr_get_current_process(ctrlr);
1815 	if (active_proc) {
1816 		active_proc->ref++;
1817 	}
1818 
1819 	nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
1820 }
1821 
1822 void
1823 nvme_ctrlr_proc_put_ref(struct spdk_nvme_ctrlr *ctrlr)
1824 {
1825 	struct spdk_nvme_ctrlr_process	*active_proc;
1826 	int				proc_count;
1827 
1828 	nvme_robust_mutex_lock(&ctrlr->ctrlr_lock);
1829 
1830 	proc_count = nvme_ctrlr_remove_inactive_proc(ctrlr);
1831 
1832 	active_proc = spdk_nvme_ctrlr_get_current_process(ctrlr);
1833 	if (active_proc) {
1834 		active_proc->ref--;
1835 		assert(active_proc->ref >= 0);
1836 
1837 		/*
1838 		 * The last active process will be removed at the end of
1839 		 * the destruction of the controller.
1840 		 */
1841 		if (active_proc->ref == 0 && proc_count != 1) {
1842 			nvme_ctrlr_remove_process(ctrlr, active_proc);
1843 		}
1844 	}
1845 
1846 	nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
1847 }
1848 
1849 int
1850 nvme_ctrlr_get_ref_count(struct spdk_nvme_ctrlr *ctrlr)
1851 {
1852 	struct spdk_nvme_ctrlr_process	*active_proc;
1853 	int				ref = 0;
1854 
1855 	nvme_robust_mutex_lock(&ctrlr->ctrlr_lock);
1856 
1857 	nvme_ctrlr_remove_inactive_proc(ctrlr);
1858 
1859 	TAILQ_FOREACH(active_proc, &ctrlr->active_procs, tailq) {
1860 		ref += active_proc->ref;
1861 	}
1862 
1863 	nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
1864 
1865 	return ref;
1866 }
1867 
1868 /**
1869  *  Get the PCI device handle which is only visible to its associated process.
1870  */
1871 struct spdk_pci_device *
1872 nvme_ctrlr_proc_get_devhandle(struct spdk_nvme_ctrlr *ctrlr)
1873 {
1874 	struct spdk_nvme_ctrlr_process	*active_proc;
1875 	struct spdk_pci_device		*devhandle = NULL;
1876 
1877 	nvme_robust_mutex_lock(&ctrlr->ctrlr_lock);
1878 
1879 	active_proc = spdk_nvme_ctrlr_get_current_process(ctrlr);
1880 	if (active_proc) {
1881 		devhandle = active_proc->devhandle;
1882 	}
1883 
1884 	nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
1885 
1886 	return devhandle;
1887 }
1888 
1889 static void
1890 nvme_ctrlr_enable_admin_queue(struct spdk_nvme_ctrlr *ctrlr)
1891 {
1892 	nvme_transport_qpair_reset(ctrlr->adminq);
1893 	nvme_qpair_enable(ctrlr->adminq);
1894 }
1895 
1896 /**
1897  * This function will be called repeatedly during initialization until the controller is ready.
1898  */
1899 int
1900 nvme_ctrlr_process_init(struct spdk_nvme_ctrlr *ctrlr)
1901 {
1902 	union spdk_nvme_cc_register cc;
1903 	union spdk_nvme_csts_register csts;
1904 	uint32_t ready_timeout_in_ms;
1905 	int rc = 0;
1906 
1907 	/*
1908 	 * May need to avoid accessing any register on the target controller
1909 	 * for a while. Return early without touching the FSM.
1910 	 * Check sleep_timeout_tsc > 0 for unit test.
1911 	 */
1912 	if ((ctrlr->sleep_timeout_tsc > 0) &&
1913 	    (spdk_get_ticks() <= ctrlr->sleep_timeout_tsc)) {
1914 		return 0;
1915 	}
1916 	ctrlr->sleep_timeout_tsc = 0;
1917 
1918 	if (nvme_ctrlr_get_cc(ctrlr, &cc) ||
1919 	    nvme_ctrlr_get_csts(ctrlr, &csts)) {
1920 		if (ctrlr->state_timeout_tsc != NVME_TIMEOUT_INFINITE) {
1921 			/* While a device is resetting, it may be unable to service MMIO reads
1922 			 * temporarily. Allow for this case.
1923 			 */
1924 			SPDK_ERRLOG("Get registers failed while waiting for CSTS.RDY == 0\n");
1925 			goto init_timeout;
1926 		}
1927 		SPDK_ERRLOG("Failed to read CC and CSTS in state %d\n", ctrlr->state);
1928 		nvme_ctrlr_fail(ctrlr, false);
1929 		return -EIO;
1930 	}
1931 
1932 	ready_timeout_in_ms = 500 * ctrlr->cap.bits.to;
1933 
1934 	/*
1935 	 * Check if the current initialization step is done or has timed out.
1936 	 */
1937 	switch (ctrlr->state) {
1938 	case NVME_CTRLR_STATE_INIT_DELAY:
1939 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_INIT, ready_timeout_in_ms);
1940 		/*
1941 		 * Controller may need some delay before it's enabled.
1942 		 *
1943 		 * This is a workaround for an issue where the PCIe-attached NVMe controller
1944 		 * is not ready after VFIO reset. We delay the initialization rather than the
1945 		 * enabling itself, because this is required only for the very first enabling
1946 		 * - directly after a VFIO reset.
1947 		 *
1948 		 * TODO: Figure out what is actually going wrong.
1949 		 */
1950 		SPDK_DEBUGLOG(SPDK_LOG_NVME, "Adding 2 second delay before initializing the controller\n");
1951 		ctrlr->sleep_timeout_tsc = spdk_get_ticks() + (2000 * spdk_get_ticks_hz() / 1000);
1952 		break;
1953 
1954 	case NVME_CTRLR_STATE_INIT:
1955 		/* Begin the hardware initialization by making sure the controller is disabled. */
1956 		if (cc.bits.en) {
1957 			SPDK_DEBUGLOG(SPDK_LOG_NVME, "CC.EN = 1\n");
1958 			/*
1959 			 * Controller is currently enabled. We need to disable it to cause a reset.
1960 			 *
1961 			 * If CC.EN = 1 && CSTS.RDY = 0, the controller is in the process of becoming ready.
1962 			 *  Wait for the ready bit to be 1 before disabling the controller.
1963 			 */
1964 			if (csts.bits.rdy == 0) {
1965 				SPDK_DEBUGLOG(SPDK_LOG_NVME, "CC.EN = 1 && CSTS.RDY = 0 - waiting for reset to complete\n");
1966 				nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_DISABLE_WAIT_FOR_READY_1, ready_timeout_in_ms);
1967 				return 0;
1968 			}
1969 
1970 			/* CC.EN = 1 && CSTS.RDY == 1, so we can immediately disable the controller. */
1971 			SPDK_DEBUGLOG(SPDK_LOG_NVME, "Setting CC.EN = 0\n");
1972 			cc.bits.en = 0;
1973 			if (nvme_ctrlr_set_cc(ctrlr, &cc)) {
1974 				SPDK_ERRLOG("set_cc() failed\n");
1975 				nvme_ctrlr_fail(ctrlr, false);
1976 				return -EIO;
1977 			}
1978 			nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_DISABLE_WAIT_FOR_READY_0, ready_timeout_in_ms);
1979 
1980 			/*
1981 			 * Wait 2.5 seconds before accessing PCI registers.
1982 			 * Not using sleep() to avoid blocking other controller's initialization.
1983 			 */
1984 			if (ctrlr->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY) {
1985 				SPDK_DEBUGLOG(SPDK_LOG_NVME, "Applying quirk: delay 2.5 seconds before reading registers\n");
1986 				ctrlr->sleep_timeout_tsc = spdk_get_ticks() + (2500 * spdk_get_ticks_hz() / 1000);
1987 			}
1988 			return 0;
1989 		} else {
1990 			if (csts.bits.rdy == 1) {
1991 				SPDK_DEBUGLOG(SPDK_LOG_NVME, "CC.EN = 0 && CSTS.RDY = 1 - waiting for shutdown to complete\n");
1992 			}
1993 
1994 			nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_DISABLE_WAIT_FOR_READY_0, ready_timeout_in_ms);
1995 			return 0;
1996 		}
1997 		break;
1998 
1999 	case NVME_CTRLR_STATE_DISABLE_WAIT_FOR_READY_1:
2000 		if (csts.bits.rdy == 1) {
2001 			SPDK_DEBUGLOG(SPDK_LOG_NVME, "CC.EN = 1 && CSTS.RDY = 1 - disabling controller\n");
2002 			/* CC.EN = 1 && CSTS.RDY = 1, so we can set CC.EN = 0 now. */
2003 			SPDK_DEBUGLOG(SPDK_LOG_NVME, "Setting CC.EN = 0\n");
2004 			cc.bits.en = 0;
2005 			if (nvme_ctrlr_set_cc(ctrlr, &cc)) {
2006 				SPDK_ERRLOG("set_cc() failed\n");
2007 				nvme_ctrlr_fail(ctrlr, false);
2008 				return -EIO;
2009 			}
2010 			nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_DISABLE_WAIT_FOR_READY_0, ready_timeout_in_ms);
2011 			return 0;
2012 		}
2013 		break;
2014 
2015 	case NVME_CTRLR_STATE_DISABLE_WAIT_FOR_READY_0:
2016 		if (csts.bits.rdy == 0) {
2017 			SPDK_DEBUGLOG(SPDK_LOG_NVME, "CC.EN = 0 && CSTS.RDY = 0\n");
2018 			nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_ENABLE, ready_timeout_in_ms);
2019 			/*
2020 			 * Delay 100us before setting CC.EN = 1.  Some NVMe SSDs miss CC.EN getting
2021 			 *  set to 1 if it is too soon after CSTS.RDY is reported as 0.
2022 			 */
2023 			spdk_delay_us(100);
2024 			return 0;
2025 		}
2026 		break;
2027 
2028 	case NVME_CTRLR_STATE_ENABLE:
2029 		SPDK_DEBUGLOG(SPDK_LOG_NVME, "Setting CC.EN = 1\n");
2030 		rc = nvme_ctrlr_enable(ctrlr);
2031 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_ENABLE_WAIT_FOR_READY_1, ready_timeout_in_ms);
2032 		return rc;
2033 
2034 	case NVME_CTRLR_STATE_ENABLE_WAIT_FOR_READY_1:
2035 		if (csts.bits.rdy == 1) {
2036 			SPDK_DEBUGLOG(SPDK_LOG_NVME, "CC.EN = 1 && CSTS.RDY = 1 - controller is ready\n");
2037 			/*
2038 			 * The controller has been enabled.
2039 			 *  Perform the rest of initialization serially.
2040 			 */
2041 			nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_ENABLE_ADMIN_QUEUE,
2042 					     ctrlr->opts.admin_timeout_ms);
2043 			return 0;
2044 		}
2045 		break;
2046 
2047 	case NVME_CTRLR_STATE_ENABLE_ADMIN_QUEUE:
2048 		nvme_ctrlr_enable_admin_queue(ctrlr);
2049 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_IDENTIFY,
2050 				     ctrlr->opts.admin_timeout_ms);
2051 		break;
2052 
2053 	case NVME_CTRLR_STATE_IDENTIFY:
2054 		rc = nvme_ctrlr_identify(ctrlr);
2055 		break;
2056 
2057 	case NVME_CTRLR_STATE_WAIT_FOR_IDENTIFY:
2058 		spdk_nvme_qpair_process_completions(ctrlr->adminq, 0);
2059 		break;
2060 
2061 	case NVME_CTRLR_STATE_SET_NUM_QUEUES:
2062 		rc = nvme_ctrlr_set_num_queues(ctrlr);
2063 		break;
2064 
2065 	case NVME_CTRLR_STATE_WAIT_FOR_SET_NUM_QUEUES:
2066 		spdk_nvme_qpair_process_completions(ctrlr->adminq, 0);
2067 		break;
2068 
2069 	case NVME_CTRLR_STATE_GET_NUM_QUEUES:
2070 		rc = nvme_ctrlr_get_num_queues(ctrlr);
2071 		break;
2072 
2073 	case NVME_CTRLR_STATE_WAIT_FOR_GET_NUM_QUEUES:
2074 		spdk_nvme_qpair_process_completions(ctrlr->adminq, 0);
2075 		break;
2076 
2077 	case NVME_CTRLR_STATE_CONSTRUCT_NS:
2078 		rc = nvme_ctrlr_construct_namespaces(ctrlr);
2079 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_IDENTIFY_ACTIVE_NS,
2080 				     ctrlr->opts.admin_timeout_ms);
2081 		break;
2082 
2083 	case NVME_CTRLR_STATE_IDENTIFY_ACTIVE_NS:
2084 		rc = nvme_ctrlr_identify_active_ns(ctrlr);
2085 		if (rc < 0) {
2086 			nvme_ctrlr_destruct_namespaces(ctrlr);
2087 		}
2088 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_IDENTIFY_NS,
2089 				     ctrlr->opts.admin_timeout_ms);
2090 		break;
2091 
2092 	case NVME_CTRLR_STATE_IDENTIFY_NS:
2093 		rc = nvme_ctrlr_identify_namespaces(ctrlr);
2094 		break;
2095 
2096 	case NVME_CTRLR_STATE_WAIT_FOR_IDENTIFY_NS:
2097 		spdk_nvme_qpair_process_completions(ctrlr->adminq, 0);
2098 		break;
2099 
2100 	case NVME_CTRLR_STATE_IDENTIFY_ID_DESCS:
2101 		rc = nvme_ctrlr_identify_id_desc_namespaces(ctrlr);
2102 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_CONFIGURE_AER,
2103 				     ctrlr->opts.admin_timeout_ms);
2104 		break;
2105 
2106 	case NVME_CTRLR_STATE_WAIT_FOR_IDENTIFY_ID_DESCS:
2107 		spdk_nvme_qpair_process_completions(ctrlr->adminq, 0);
2108 		break;
2109 
2110 	case NVME_CTRLR_STATE_CONFIGURE_AER:
2111 		rc = nvme_ctrlr_configure_aer(ctrlr);
2112 		break;
2113 
2114 	case NVME_CTRLR_STATE_WAIT_FOR_CONFIGURE_AER:
2115 		spdk_nvme_qpair_process_completions(ctrlr->adminq, 0);
2116 		break;
2117 
2118 	case NVME_CTRLR_STATE_SET_SUPPORTED_LOG_PAGES:
2119 		rc = nvme_ctrlr_set_supported_log_pages(ctrlr);
2120 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_SET_SUPPORTED_FEATURES,
2121 				     ctrlr->opts.admin_timeout_ms);
2122 		break;
2123 
2124 	case NVME_CTRLR_STATE_SET_SUPPORTED_FEATURES:
2125 		nvme_ctrlr_set_supported_features(ctrlr);
2126 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_SET_DB_BUF_CFG,
2127 				     ctrlr->opts.admin_timeout_ms);
2128 		break;
2129 
2130 	case NVME_CTRLR_STATE_SET_DB_BUF_CFG:
2131 		rc = nvme_ctrlr_set_doorbell_buffer_config(ctrlr);
2132 		break;
2133 
2134 	case NVME_CTRLR_STATE_WAIT_FOR_DB_BUF_CFG:
2135 		spdk_nvme_qpair_process_completions(ctrlr->adminq, 0);
2136 		break;
2137 
2138 	case NVME_CTRLR_STATE_SET_KEEP_ALIVE_TIMEOUT:
2139 		rc = nvme_ctrlr_set_keep_alive_timeout(ctrlr);
2140 		break;
2141 
2142 	case NVME_CTRLR_STATE_WAIT_FOR_KEEP_ALIVE_TIMEOUT:
2143 		spdk_nvme_qpair_process_completions(ctrlr->adminq, 0);
2144 		break;
2145 
2146 	case NVME_CTRLR_STATE_SET_HOST_ID:
2147 		rc = nvme_ctrlr_set_host_id(ctrlr);
2148 		break;
2149 
2150 	case NVME_CTRLR_STATE_WAIT_FOR_HOST_ID:
2151 		spdk_nvme_qpair_process_completions(ctrlr->adminq, 0);
2152 		break;
2153 
2154 	case NVME_CTRLR_STATE_READY:
2155 		SPDK_DEBUGLOG(SPDK_LOG_NVME, "Ctrlr already in ready state\n");
2156 		return 0;
2157 
2158 	case NVME_CTRLR_STATE_ERROR:
2159 		SPDK_ERRLOG("Ctrlr %s is in error state\n", ctrlr->trid.traddr);
2160 		return -1;
2161 
2162 	default:
2163 		assert(0);
2164 		nvme_ctrlr_fail(ctrlr, false);
2165 		return -1;
2166 	}
2167 
2168 init_timeout:
2169 	if (ctrlr->state_timeout_tsc != NVME_TIMEOUT_INFINITE &&
2170 	    spdk_get_ticks() > ctrlr->state_timeout_tsc) {
2171 		SPDK_ERRLOG("Initialization timed out in state %d\n", ctrlr->state);
2172 		nvme_ctrlr_fail(ctrlr, false);
2173 		return -1;
2174 	}
2175 
2176 	return rc;
2177 }
2178 
2179 int
2180 nvme_robust_mutex_init_recursive_shared(pthread_mutex_t *mtx)
2181 {
2182 	pthread_mutexattr_t attr;
2183 	int rc = 0;
2184 
2185 	if (pthread_mutexattr_init(&attr)) {
2186 		return -1;
2187 	}
2188 	if (pthread_mutexattr_settype(&attr, PTHREAD_MUTEX_RECURSIVE) ||
2189 #ifndef __FreeBSD__
2190 	    pthread_mutexattr_setrobust(&attr, PTHREAD_MUTEX_ROBUST) ||
2191 	    pthread_mutexattr_setpshared(&attr, PTHREAD_PROCESS_SHARED) ||
2192 #endif
2193 	    pthread_mutex_init(mtx, &attr)) {
2194 		rc = -1;
2195 	}
2196 	pthread_mutexattr_destroy(&attr);
2197 	return rc;
2198 }
2199 
2200 int
2201 nvme_ctrlr_construct(struct spdk_nvme_ctrlr *ctrlr)
2202 {
2203 	int rc;
2204 
2205 	if (ctrlr->trid.trtype == SPDK_NVME_TRANSPORT_PCIE) {
2206 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_INIT_DELAY, NVME_TIMEOUT_INFINITE);
2207 	} else {
2208 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_INIT, NVME_TIMEOUT_INFINITE);
2209 	}
2210 
2211 	ctrlr->flags = 0;
2212 	ctrlr->free_io_qids = NULL;
2213 	ctrlr->is_resetting = false;
2214 	ctrlr->is_failed = false;
2215 	ctrlr->is_shutdown = false;
2216 
2217 	TAILQ_INIT(&ctrlr->active_io_qpairs);
2218 	STAILQ_INIT(&ctrlr->queued_aborts);
2219 	ctrlr->outstanding_aborts = 0;
2220 
2221 	rc = nvme_robust_mutex_init_recursive_shared(&ctrlr->ctrlr_lock);
2222 	if (rc != 0) {
2223 		return rc;
2224 	}
2225 
2226 	TAILQ_INIT(&ctrlr->active_procs);
2227 
2228 	return rc;
2229 }
2230 
2231 /* This function should be called once at ctrlr initialization to set up constant properties. */
2232 void
2233 nvme_ctrlr_init_cap(struct spdk_nvme_ctrlr *ctrlr, const union spdk_nvme_cap_register *cap,
2234 		    const union spdk_nvme_vs_register *vs)
2235 {
2236 	ctrlr->cap = *cap;
2237 	ctrlr->vs = *vs;
2238 
2239 	ctrlr->min_page_size = 1u << (12 + ctrlr->cap.bits.mpsmin);
2240 
2241 	/* For now, always select page_size == min_page_size. */
2242 	ctrlr->page_size = ctrlr->min_page_size;
2243 
2244 	ctrlr->opts.io_queue_size = spdk_max(ctrlr->opts.io_queue_size, SPDK_NVME_IO_QUEUE_MIN_ENTRIES);
2245 	ctrlr->opts.io_queue_size = spdk_min(ctrlr->opts.io_queue_size, MAX_IO_QUEUE_ENTRIES);
2246 	ctrlr->opts.io_queue_size = spdk_min(ctrlr->opts.io_queue_size, ctrlr->cap.bits.mqes + 1u);
2247 
2248 	ctrlr->opts.io_queue_requests = spdk_max(ctrlr->opts.io_queue_requests, ctrlr->opts.io_queue_size);
2249 }
2250 
2251 void
2252 nvme_ctrlr_destruct_finish(struct spdk_nvme_ctrlr *ctrlr)
2253 {
2254 	pthread_mutex_destroy(&ctrlr->ctrlr_lock);
2255 }
2256 
2257 void
2258 nvme_ctrlr_destruct(struct spdk_nvme_ctrlr *ctrlr)
2259 {
2260 	struct spdk_nvme_qpair *qpair, *tmp;
2261 
2262 	SPDK_DEBUGLOG(SPDK_LOG_NVME, "Prepare to destruct SSD: %s\n", ctrlr->trid.traddr);
2263 	TAILQ_FOREACH_SAFE(qpair, &ctrlr->active_io_qpairs, tailq, tmp) {
2264 		spdk_nvme_ctrlr_free_io_qpair(qpair);
2265 	}
2266 
2267 	nvme_ctrlr_free_doorbell_buffer(ctrlr);
2268 
2269 	nvme_ctrlr_shutdown(ctrlr);
2270 
2271 	nvme_ctrlr_destruct_namespaces(ctrlr);
2272 
2273 	spdk_bit_array_free(&ctrlr->free_io_qids);
2274 
2275 	nvme_transport_ctrlr_destruct(ctrlr);
2276 }
2277 
2278 int
2279 nvme_ctrlr_submit_admin_request(struct spdk_nvme_ctrlr *ctrlr,
2280 				struct nvme_request *req)
2281 {
2282 	return nvme_qpair_submit_request(ctrlr->adminq, req);
2283 }
2284 
2285 static void
2286 nvme_keep_alive_completion(void *cb_ctx, const struct spdk_nvme_cpl *cpl)
2287 {
2288 	/* Do nothing */
2289 }
2290 
2291 /*
2292  * Check if we need to send a Keep Alive command.
2293  * Caller must hold ctrlr->ctrlr_lock.
2294  */
2295 static void
2296 nvme_ctrlr_keep_alive(struct spdk_nvme_ctrlr *ctrlr)
2297 {
2298 	uint64_t now;
2299 	struct nvme_request *req;
2300 	struct spdk_nvme_cmd *cmd;
2301 	int rc;
2302 
2303 	now = spdk_get_ticks();
2304 	if (now < ctrlr->next_keep_alive_tick) {
2305 		return;
2306 	}
2307 
2308 	req = nvme_allocate_request_null(ctrlr->adminq, nvme_keep_alive_completion, NULL);
2309 	if (req == NULL) {
2310 		return;
2311 	}
2312 
2313 	cmd = &req->cmd;
2314 	cmd->opc = SPDK_NVME_OPC_KEEP_ALIVE;
2315 
2316 	rc = nvme_ctrlr_submit_admin_request(ctrlr, req);
2317 	if (rc != 0) {
2318 		SPDK_ERRLOG("Submitting Keep Alive failed\n");
2319 	}
2320 
2321 	ctrlr->next_keep_alive_tick = now + ctrlr->keep_alive_interval_ticks;
2322 }
2323 
2324 int32_t
2325 spdk_nvme_ctrlr_process_admin_completions(struct spdk_nvme_ctrlr *ctrlr)
2326 {
2327 	int32_t num_completions;
2328 
2329 	nvme_robust_mutex_lock(&ctrlr->ctrlr_lock);
2330 	if (ctrlr->keep_alive_interval_ticks) {
2331 		nvme_ctrlr_keep_alive(ctrlr);
2332 	}
2333 	num_completions = spdk_nvme_qpair_process_completions(ctrlr->adminq, 0);
2334 	nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
2335 
2336 	return num_completions;
2337 }
2338 
2339 const struct spdk_nvme_ctrlr_data *
2340 spdk_nvme_ctrlr_get_data(struct spdk_nvme_ctrlr *ctrlr)
2341 {
2342 	return &ctrlr->cdata;
2343 }
2344 
2345 union spdk_nvme_csts_register spdk_nvme_ctrlr_get_regs_csts(struct spdk_nvme_ctrlr *ctrlr)
2346 {
2347 	union spdk_nvme_csts_register csts;
2348 
2349 	if (nvme_ctrlr_get_csts(ctrlr, &csts)) {
2350 		csts.raw = 0xFFFFFFFFu;
2351 	}
2352 	return csts;
2353 }
2354 
2355 union spdk_nvme_cap_register spdk_nvme_ctrlr_get_regs_cap(struct spdk_nvme_ctrlr *ctrlr)
2356 {
2357 	return ctrlr->cap;
2358 }
2359 
2360 union spdk_nvme_vs_register spdk_nvme_ctrlr_get_regs_vs(struct spdk_nvme_ctrlr *ctrlr)
2361 {
2362 	return ctrlr->vs;
2363 }
2364 
2365 union spdk_nvme_cmbsz_register spdk_nvme_ctrlr_get_regs_cmbsz(struct spdk_nvme_ctrlr *ctrlr)
2366 {
2367 	union spdk_nvme_cmbsz_register cmbsz;
2368 
2369 	if (nvme_ctrlr_get_cmbsz(ctrlr, &cmbsz)) {
2370 		cmbsz.raw = 0;
2371 	}
2372 
2373 	return cmbsz;
2374 }
2375 
2376 uint32_t
2377 spdk_nvme_ctrlr_get_num_ns(struct spdk_nvme_ctrlr *ctrlr)
2378 {
2379 	return ctrlr->num_ns;
2380 }
2381 
2382 static int32_t
2383 spdk_nvme_ctrlr_active_ns_idx(struct spdk_nvme_ctrlr *ctrlr, uint32_t nsid)
2384 {
2385 	int32_t result = -1;
2386 
2387 	if (ctrlr->active_ns_list == NULL || nsid == 0 || nsid > ctrlr->num_ns) {
2388 		return result;
2389 	}
2390 
2391 	int32_t lower = 0;
2392 	int32_t upper = ctrlr->num_ns - 1;
2393 	int32_t mid;
2394 
2395 	while (lower <= upper) {
2396 		mid = lower + (upper - lower) / 2;
2397 		if (ctrlr->active_ns_list[mid] == nsid) {
2398 			result = mid;
2399 			break;
2400 		} else {
2401 			if (ctrlr->active_ns_list[mid] != 0 && ctrlr->active_ns_list[mid] < nsid) {
2402 				lower = mid + 1;
2403 			} else {
2404 				upper = mid - 1;
2405 			}
2406 
2407 		}
2408 	}
2409 
2410 	return result;
2411 }
2412 
2413 bool
2414 spdk_nvme_ctrlr_is_active_ns(struct spdk_nvme_ctrlr *ctrlr, uint32_t nsid)
2415 {
2416 	return spdk_nvme_ctrlr_active_ns_idx(ctrlr, nsid) != -1;
2417 }
2418 
2419 uint32_t
2420 spdk_nvme_ctrlr_get_first_active_ns(struct spdk_nvme_ctrlr *ctrlr)
2421 {
2422 	return ctrlr->active_ns_list ? ctrlr->active_ns_list[0] : 0;
2423 }
2424 
2425 uint32_t
2426 spdk_nvme_ctrlr_get_next_active_ns(struct spdk_nvme_ctrlr *ctrlr, uint32_t prev_nsid)
2427 {
2428 	int32_t nsid_idx = spdk_nvme_ctrlr_active_ns_idx(ctrlr, prev_nsid);
2429 	if (ctrlr->active_ns_list && nsid_idx >= 0 && (uint32_t)nsid_idx < ctrlr->num_ns - 1) {
2430 		return ctrlr->active_ns_list[nsid_idx + 1];
2431 	}
2432 	return 0;
2433 }
2434 
2435 struct spdk_nvme_ns *
2436 spdk_nvme_ctrlr_get_ns(struct spdk_nvme_ctrlr *ctrlr, uint32_t nsid)
2437 {
2438 	if (nsid < 1 || nsid > ctrlr->num_ns) {
2439 		return NULL;
2440 	}
2441 
2442 	return &ctrlr->ns[nsid - 1];
2443 }
2444 
2445 struct spdk_pci_device *
2446 spdk_nvme_ctrlr_get_pci_device(struct spdk_nvme_ctrlr *ctrlr)
2447 {
2448 	if (ctrlr == NULL) {
2449 		return NULL;
2450 	}
2451 
2452 	if (ctrlr->trid.trtype != SPDK_NVME_TRANSPORT_PCIE) {
2453 		return NULL;
2454 	}
2455 
2456 	return nvme_ctrlr_proc_get_devhandle(ctrlr);
2457 }
2458 
2459 uint32_t
2460 spdk_nvme_ctrlr_get_max_xfer_size(const struct spdk_nvme_ctrlr *ctrlr)
2461 {
2462 	return ctrlr->max_xfer_size;
2463 }
2464 
2465 void
2466 spdk_nvme_ctrlr_register_aer_callback(struct spdk_nvme_ctrlr *ctrlr,
2467 				      spdk_nvme_aer_cb aer_cb_fn,
2468 				      void *aer_cb_arg)
2469 {
2470 	struct spdk_nvme_ctrlr_process *active_proc;
2471 
2472 	nvme_robust_mutex_lock(&ctrlr->ctrlr_lock);
2473 
2474 	active_proc = spdk_nvme_ctrlr_get_current_process(ctrlr);
2475 	if (active_proc) {
2476 		active_proc->aer_cb_fn = aer_cb_fn;
2477 		active_proc->aer_cb_arg = aer_cb_arg;
2478 	}
2479 
2480 	nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
2481 }
2482 
2483 void
2484 spdk_nvme_ctrlr_register_timeout_callback(struct spdk_nvme_ctrlr *ctrlr,
2485 		uint64_t timeout_us, spdk_nvme_timeout_cb cb_fn, void *cb_arg)
2486 {
2487 	struct spdk_nvme_ctrlr_process	*active_proc;
2488 
2489 	nvme_robust_mutex_lock(&ctrlr->ctrlr_lock);
2490 
2491 	active_proc = spdk_nvme_ctrlr_get_current_process(ctrlr);
2492 	if (active_proc) {
2493 		active_proc->timeout_ticks = timeout_us * spdk_get_ticks_hz() / 1000000ULL;
2494 		active_proc->timeout_cb_fn = cb_fn;
2495 		active_proc->timeout_cb_arg = cb_arg;
2496 	}
2497 
2498 	ctrlr->timeout_enabled = true;
2499 
2500 	nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
2501 }
2502 
2503 bool
2504 spdk_nvme_ctrlr_is_log_page_supported(struct spdk_nvme_ctrlr *ctrlr, uint8_t log_page)
2505 {
2506 	/* No bounds check necessary, since log_page is uint8_t and log_page_supported has 256 entries */
2507 	SPDK_STATIC_ASSERT(sizeof(ctrlr->log_page_supported) == 256, "log_page_supported size mismatch");
2508 	return ctrlr->log_page_supported[log_page];
2509 }
2510 
2511 bool
2512 spdk_nvme_ctrlr_is_feature_supported(struct spdk_nvme_ctrlr *ctrlr, uint8_t feature_code)
2513 {
2514 	/* No bounds check necessary, since feature_code is uint8_t and feature_supported has 256 entries */
2515 	SPDK_STATIC_ASSERT(sizeof(ctrlr->feature_supported) == 256, "feature_supported size mismatch");
2516 	return ctrlr->feature_supported[feature_code];
2517 }
2518 
2519 int
2520 spdk_nvme_ctrlr_attach_ns(struct spdk_nvme_ctrlr *ctrlr, uint32_t nsid,
2521 			  struct spdk_nvme_ctrlr_list *payload)
2522 {
2523 	struct nvme_completion_poll_status	status;
2524 	int					res;
2525 	struct spdk_nvme_ns			*ns;
2526 
2527 	res = nvme_ctrlr_cmd_attach_ns(ctrlr, nsid, payload,
2528 				       nvme_completion_poll_cb, &status);
2529 	if (res) {
2530 		return res;
2531 	}
2532 	if (spdk_nvme_wait_for_completion_robust_lock(ctrlr->adminq, &status, &ctrlr->ctrlr_lock)) {
2533 		SPDK_ERRLOG("spdk_nvme_ctrlr_attach_ns failed!\n");
2534 		return -ENXIO;
2535 	}
2536 
2537 	res = nvme_ctrlr_identify_active_ns(ctrlr);
2538 	if (res) {
2539 		return res;
2540 	}
2541 
2542 	ns = &ctrlr->ns[nsid - 1];
2543 	return nvme_ns_construct(ns, nsid, ctrlr);
2544 }
2545 
2546 int
2547 spdk_nvme_ctrlr_detach_ns(struct spdk_nvme_ctrlr *ctrlr, uint32_t nsid,
2548 			  struct spdk_nvme_ctrlr_list *payload)
2549 {
2550 	struct nvme_completion_poll_status	status;
2551 	int					res;
2552 	struct spdk_nvme_ns			*ns;
2553 
2554 	res = nvme_ctrlr_cmd_detach_ns(ctrlr, nsid, payload,
2555 				       nvme_completion_poll_cb, &status);
2556 	if (res) {
2557 		return res;
2558 	}
2559 	if (spdk_nvme_wait_for_completion_robust_lock(ctrlr->adminq, &status, &ctrlr->ctrlr_lock)) {
2560 		SPDK_ERRLOG("spdk_nvme_ctrlr_detach_ns failed!\n");
2561 		return -ENXIO;
2562 	}
2563 
2564 	res = nvme_ctrlr_identify_active_ns(ctrlr);
2565 	if (res) {
2566 		return res;
2567 	}
2568 
2569 	ns = &ctrlr->ns[nsid - 1];
2570 	/* Inactive NS */
2571 	nvme_ns_destruct(ns);
2572 
2573 	return 0;
2574 }
2575 
2576 uint32_t
2577 spdk_nvme_ctrlr_create_ns(struct spdk_nvme_ctrlr *ctrlr, struct spdk_nvme_ns_data *payload)
2578 {
2579 	struct nvme_completion_poll_status	status;
2580 	int					res;
2581 	uint32_t				nsid;
2582 	struct spdk_nvme_ns			*ns;
2583 
2584 	res = nvme_ctrlr_cmd_create_ns(ctrlr, payload, nvme_completion_poll_cb, &status);
2585 	if (res) {
2586 		return 0;
2587 	}
2588 	if (spdk_nvme_wait_for_completion_robust_lock(ctrlr->adminq, &status, &ctrlr->ctrlr_lock)) {
2589 		SPDK_ERRLOG("spdk_nvme_ctrlr_create_ns failed!\n");
2590 		return 0;
2591 	}
2592 
2593 	nsid = status.cpl.cdw0;
2594 	ns = &ctrlr->ns[nsid - 1];
2595 	/* Inactive NS */
2596 	res = nvme_ns_construct(ns, nsid, ctrlr);
2597 	if (res) {
2598 		return 0;
2599 	}
2600 
2601 	/* Return the namespace ID that was created */
2602 	return nsid;
2603 }
2604 
2605 int
2606 spdk_nvme_ctrlr_delete_ns(struct spdk_nvme_ctrlr *ctrlr, uint32_t nsid)
2607 {
2608 	struct nvme_completion_poll_status	status;
2609 	int					res;
2610 	struct spdk_nvme_ns			*ns;
2611 
2612 	res = nvme_ctrlr_cmd_delete_ns(ctrlr, nsid, nvme_completion_poll_cb, &status);
2613 	if (res) {
2614 		return res;
2615 	}
2616 	if (spdk_nvme_wait_for_completion_robust_lock(ctrlr->adminq, &status, &ctrlr->ctrlr_lock)) {
2617 		SPDK_ERRLOG("spdk_nvme_ctrlr_delete_ns failed!\n");
2618 		return -ENXIO;
2619 	}
2620 
2621 	res = nvme_ctrlr_identify_active_ns(ctrlr);
2622 	if (res) {
2623 		return res;
2624 	}
2625 
2626 	ns = &ctrlr->ns[nsid - 1];
2627 	nvme_ns_destruct(ns);
2628 
2629 	return 0;
2630 }
2631 
2632 int
2633 spdk_nvme_ctrlr_format(struct spdk_nvme_ctrlr *ctrlr, uint32_t nsid,
2634 		       struct spdk_nvme_format *format)
2635 {
2636 	struct nvme_completion_poll_status	status;
2637 	int					res;
2638 
2639 	res = nvme_ctrlr_cmd_format(ctrlr, nsid, format, nvme_completion_poll_cb,
2640 				    &status);
2641 	if (res) {
2642 		return res;
2643 	}
2644 	if (spdk_nvme_wait_for_completion_robust_lock(ctrlr->adminq, &status, &ctrlr->ctrlr_lock)) {
2645 		SPDK_ERRLOG("spdk_nvme_ctrlr_format failed!\n");
2646 		return -ENXIO;
2647 	}
2648 
2649 	return spdk_nvme_ctrlr_reset(ctrlr);
2650 }
2651 
2652 int
2653 spdk_nvme_ctrlr_update_firmware(struct spdk_nvme_ctrlr *ctrlr, void *payload, uint32_t size,
2654 				int slot, enum spdk_nvme_fw_commit_action commit_action, struct spdk_nvme_status *completion_status)
2655 {
2656 	struct spdk_nvme_fw_commit		fw_commit;
2657 	struct nvme_completion_poll_status	status;
2658 	int					res;
2659 	unsigned int				size_remaining;
2660 	unsigned int				offset;
2661 	unsigned int				transfer;
2662 	void					*p;
2663 
2664 	if (!completion_status) {
2665 		return -EINVAL;
2666 	}
2667 	memset(completion_status, 0, sizeof(struct spdk_nvme_status));
2668 	if (size % 4) {
2669 		SPDK_ERRLOG("spdk_nvme_ctrlr_update_firmware invalid size!\n");
2670 		return -1;
2671 	}
2672 
2673 	/* Current support only for SPDK_NVME_FW_COMMIT_REPLACE_IMG
2674 	 * and SPDK_NVME_FW_COMMIT_REPLACE_AND_ENABLE_IMG
2675 	 */
2676 	if ((commit_action != SPDK_NVME_FW_COMMIT_REPLACE_IMG) &&
2677 	    (commit_action != SPDK_NVME_FW_COMMIT_REPLACE_AND_ENABLE_IMG)) {
2678 		SPDK_ERRLOG("spdk_nvme_ctrlr_update_firmware invalid command!\n");
2679 		return -1;
2680 	}
2681 
2682 	/* Firmware download */
2683 	size_remaining = size;
2684 	offset = 0;
2685 	p = payload;
2686 
2687 	while (size_remaining > 0) {
2688 		transfer = spdk_min(size_remaining, ctrlr->min_page_size);
2689 
2690 		res = nvme_ctrlr_cmd_fw_image_download(ctrlr, transfer, offset, p,
2691 						       nvme_completion_poll_cb,
2692 						       &status);
2693 		if (res) {
2694 			return res;
2695 		}
2696 
2697 		if (spdk_nvme_wait_for_completion_robust_lock(ctrlr->adminq, &status, &ctrlr->ctrlr_lock)) {
2698 			SPDK_ERRLOG("spdk_nvme_ctrlr_fw_image_download failed!\n");
2699 			return -ENXIO;
2700 		}
2701 		p += transfer;
2702 		offset += transfer;
2703 		size_remaining -= transfer;
2704 	}
2705 
2706 	/* Firmware commit */
2707 	memset(&fw_commit, 0, sizeof(struct spdk_nvme_fw_commit));
2708 	fw_commit.fs = slot;
2709 	fw_commit.ca = commit_action;
2710 
2711 	res = nvme_ctrlr_cmd_fw_commit(ctrlr, &fw_commit, nvme_completion_poll_cb,
2712 				       &status);
2713 	if (res) {
2714 		return res;
2715 	}
2716 
2717 	res = spdk_nvme_wait_for_completion_robust_lock(ctrlr->adminq, &status, &ctrlr->ctrlr_lock);
2718 
2719 	memcpy(completion_status, &status.cpl.status, sizeof(struct spdk_nvme_status));
2720 
2721 	if (res) {
2722 		if (status.cpl.status.sct != SPDK_NVME_SCT_COMMAND_SPECIFIC ||
2723 		    status.cpl.status.sc != SPDK_NVME_SC_FIRMWARE_REQ_NVM_RESET) {
2724 			if (status.cpl.status.sct == SPDK_NVME_SCT_COMMAND_SPECIFIC  &&
2725 			    status.cpl.status.sc == SPDK_NVME_SC_FIRMWARE_REQ_CONVENTIONAL_RESET) {
2726 				SPDK_NOTICELOG("firmware activation requires conventional reset to be performed. !\n");
2727 			} else {
2728 				SPDK_ERRLOG("nvme_ctrlr_cmd_fw_commit failed!\n");
2729 			}
2730 			return -ENXIO;
2731 		}
2732 	}
2733 
2734 	return spdk_nvme_ctrlr_reset(ctrlr);
2735 }
2736 
2737 void *
2738 spdk_nvme_ctrlr_alloc_cmb_io_buffer(struct spdk_nvme_ctrlr *ctrlr, size_t size)
2739 {
2740 	void *buf;
2741 
2742 	if (size == 0) {
2743 		return NULL;
2744 	}
2745 
2746 	nvme_robust_mutex_lock(&ctrlr->ctrlr_lock);
2747 	buf = nvme_transport_ctrlr_alloc_cmb_io_buffer(ctrlr, size);
2748 	nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
2749 
2750 	return buf;
2751 }
2752 
2753 void
2754 spdk_nvme_ctrlr_free_cmb_io_buffer(struct spdk_nvme_ctrlr *ctrlr, void *buf, size_t size)
2755 {
2756 	if (buf && size) {
2757 		nvme_robust_mutex_lock(&ctrlr->ctrlr_lock);
2758 		nvme_transport_ctrlr_free_cmb_io_buffer(ctrlr, buf, size);
2759 		nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
2760 	}
2761 }
2762 
2763 bool
2764 spdk_nvme_ctrlr_is_discovery(struct spdk_nvme_ctrlr *ctrlr)
2765 {
2766 	assert(ctrlr);
2767 
2768 	return !strncmp(ctrlr->trid.subnqn, SPDK_NVMF_DISCOVERY_NQN,
2769 			strlen(SPDK_NVMF_DISCOVERY_NQN));
2770 }
2771 
2772 int
2773 spdk_nvme_ctrlr_security_receive(struct spdk_nvme_ctrlr *ctrlr, uint8_t secp,
2774 				 uint16_t spsp, uint8_t nssf, void *payload, size_t size)
2775 {
2776 	struct nvme_completion_poll_status	status;
2777 	int					res;
2778 
2779 	res = nvme_ctrlr_cmd_security_receive(ctrlr, secp, spsp, nssf, payload, size,
2780 					      nvme_completion_poll_cb, &status);
2781 	if (res) {
2782 		return res;
2783 	}
2784 	if (spdk_nvme_wait_for_completion_robust_lock(ctrlr->adminq, &status, &ctrlr->ctrlr_lock)) {
2785 		SPDK_ERRLOG("spdk_nvme_ctrlr_security_receive failed!\n");
2786 		return -ENXIO;
2787 	}
2788 
2789 	return 0;
2790 }
2791 
2792 int
2793 spdk_nvme_ctrlr_security_send(struct spdk_nvme_ctrlr *ctrlr, uint8_t secp,
2794 			      uint16_t spsp, uint8_t nssf, void *payload, size_t size)
2795 {
2796 	struct nvme_completion_poll_status	status;
2797 	int					res;
2798 
2799 	res = nvme_ctrlr_cmd_security_send(ctrlr, secp, spsp, nssf, payload, size, nvme_completion_poll_cb,
2800 					   &status);
2801 	if (res) {
2802 		return res;
2803 	}
2804 	if (spdk_nvme_wait_for_completion_robust_lock(ctrlr->adminq, &status, &ctrlr->ctrlr_lock)) {
2805 		SPDK_ERRLOG("spdk_nvme_ctrlr_security_send failed!\n");
2806 		return -ENXIO;
2807 	}
2808 
2809 	return 0;
2810 }
2811