xref: /spdk/lib/nvme/nvme_ctrlr.c (revision 179ed697b3c461d100e675915d074be717b7b9cc)
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 
40 static int nvme_ctrlr_construct_and_submit_aer(struct spdk_nvme_ctrlr *ctrlr,
41 		struct nvme_async_event_request *aer);
42 
43 static int
44 nvme_ctrlr_get_cc(struct spdk_nvme_ctrlr *ctrlr, union spdk_nvme_cc_register *cc)
45 {
46 	return nvme_transport_ctrlr_get_reg_4(ctrlr, offsetof(struct spdk_nvme_registers, cc.raw),
47 					      &cc->raw);
48 }
49 
50 static int
51 nvme_ctrlr_get_csts(struct spdk_nvme_ctrlr *ctrlr, union spdk_nvme_csts_register *csts)
52 {
53 	return nvme_transport_ctrlr_get_reg_4(ctrlr, offsetof(struct spdk_nvme_registers, csts.raw),
54 					      &csts->raw);
55 }
56 
57 int
58 nvme_ctrlr_get_cap(struct spdk_nvme_ctrlr *ctrlr, union spdk_nvme_cap_register *cap)
59 {
60 	return nvme_transport_ctrlr_get_reg_8(ctrlr, offsetof(struct spdk_nvme_registers, cap.raw),
61 					      &cap->raw);
62 }
63 
64 static int
65 nvme_ctrlr_get_vs(struct spdk_nvme_ctrlr *ctrlr, union spdk_nvme_vs_register *vs)
66 {
67 	return nvme_transport_ctrlr_get_reg_4(ctrlr, offsetof(struct spdk_nvme_registers, vs.raw),
68 					      &vs->raw);
69 }
70 
71 static int
72 nvme_ctrlr_set_cc(struct spdk_nvme_ctrlr *ctrlr, const union spdk_nvme_cc_register *cc)
73 {
74 	return nvme_transport_ctrlr_set_reg_4(ctrlr, offsetof(struct spdk_nvme_registers, cc.raw),
75 					      cc->raw);
76 }
77 
78 void
79 spdk_nvme_ctrlr_opts_set_defaults(struct spdk_nvme_ctrlr_opts *opts)
80 {
81 	opts->num_io_queues = DEFAULT_MAX_IO_QUEUES;
82 	opts->use_cmb_sqs = true;
83 	opts->arb_mechanism = SPDK_NVME_CC_AMS_RR;
84 	opts->keep_alive_timeout_ms = 10 * 1000;
85 	opts->io_queue_size = DEFAULT_IO_QUEUE_SIZE;
86 	strncpy(opts->hostnqn, DEFAULT_HOSTNQN, sizeof(opts->hostnqn));
87 	opts->io_queue_requests = DEFAULT_IO_QUEUE_REQUESTS;
88 }
89 
90 /**
91  * This function will be called when the process allocates the IO qpair.
92  * Note: the ctrlr_lock must be held when calling this function.
93  */
94 static void
95 nvme_ctrlr_proc_add_io_qpair(struct spdk_nvme_qpair *qpair)
96 {
97 	struct spdk_nvme_ctrlr_process	*active_proc;
98 	struct spdk_nvme_ctrlr		*ctrlr = qpair->ctrlr;
99 	pid_t				pid = getpid();
100 
101 	TAILQ_FOREACH(active_proc, &ctrlr->active_procs, tailq) {
102 		if (active_proc->pid == pid) {
103 			TAILQ_INSERT_TAIL(&active_proc->allocated_io_qpairs, qpair,
104 					  per_process_tailq);
105 			break;
106 		}
107 	}
108 }
109 
110 /**
111  * This function will be called when the process frees the IO qpair.
112  * Note: the ctrlr_lock must be held when calling this function.
113  */
114 static void
115 nvme_ctrlr_proc_remove_io_qpair(struct spdk_nvme_qpair *qpair)
116 {
117 	struct spdk_nvme_ctrlr_process	*active_proc;
118 	struct spdk_nvme_ctrlr		*ctrlr = qpair->ctrlr;
119 	struct spdk_nvme_qpair          *active_qpair, *tmp_qpair;
120 	pid_t				pid = getpid();
121 	bool				proc_found = false;
122 
123 	TAILQ_FOREACH(active_proc, &ctrlr->active_procs, tailq) {
124 		if (active_proc->pid == pid) {
125 			proc_found = true;
126 			break;
127 		}
128 	}
129 
130 	if (proc_found == false) {
131 		return;
132 	}
133 
134 	TAILQ_FOREACH_SAFE(active_qpair, &active_proc->allocated_io_qpairs,
135 			   per_process_tailq, tmp_qpair) {
136 		if (active_qpair == qpair) {
137 			TAILQ_REMOVE(&active_proc->allocated_io_qpairs,
138 				     active_qpair, per_process_tailq);
139 
140 			break;
141 		}
142 	}
143 }
144 
145 void
146 spdk_nvme_ctrlr_get_default_io_qpair_opts(struct spdk_nvme_ctrlr *ctrlr,
147 		struct spdk_nvme_io_qpair_opts *opts,
148 		size_t opts_size)
149 {
150 	if (!ctrlr || !opts) {
151 		return;
152 	}
153 
154 	memset(opts, 0, opts_size);
155 
156 #define FIELD_OK(field) \
157 	offsetof(struct spdk_nvme_io_qpair_opts, field) + sizeof(opts->field) <= opts_size
158 
159 	if (FIELD_OK(qprio)) {
160 		opts->qprio = SPDK_NVME_QPRIO_URGENT;
161 	}
162 
163 	if (FIELD_OK(io_queue_size)) {
164 		opts->io_queue_size = ctrlr->opts.io_queue_size;
165 	}
166 
167 	if (FIELD_OK(io_queue_requests)) {
168 		opts->io_queue_requests = ctrlr->opts.io_queue_requests;
169 	}
170 
171 #undef FIELD_OK
172 }
173 
174 struct spdk_nvme_qpair *
175 spdk_nvme_ctrlr_alloc_io_qpair(struct spdk_nvme_ctrlr *ctrlr,
176 			       const struct spdk_nvme_io_qpair_opts *user_opts,
177 			       size_t opts_size)
178 {
179 	uint32_t				qid;
180 	struct spdk_nvme_qpair			*qpair;
181 	union spdk_nvme_cc_register		cc;
182 	struct spdk_nvme_io_qpair_opts		opts;
183 
184 	if (!ctrlr) {
185 		return NULL;
186 	}
187 
188 	/*
189 	 * Get the default options, then overwrite them with the user-provided options
190 	 * up to opts_size.
191 	 *
192 	 * This allows for extensions of the opts structure without breaking
193 	 * ABI compatibility.
194 	 */
195 	spdk_nvme_ctrlr_get_default_io_qpair_opts(ctrlr, &opts, sizeof(opts));
196 	if (user_opts) {
197 		memcpy(&opts, user_opts, spdk_min(sizeof(opts), opts_size));
198 	}
199 
200 	if (nvme_ctrlr_get_cc(ctrlr, &cc)) {
201 		SPDK_ERRLOG("get_cc failed\n");
202 		return NULL;
203 	}
204 
205 	/* Only the low 2 bits (values 0, 1, 2, 3) of QPRIO are valid. */
206 	if ((opts.qprio & 3) != opts.qprio) {
207 		return NULL;
208 	}
209 
210 	/*
211 	 * Only value SPDK_NVME_QPRIO_URGENT(0) is valid for the
212 	 * default round robin arbitration method.
213 	 */
214 	if ((cc.bits.ams == SPDK_NVME_CC_AMS_RR) && (opts.qprio != SPDK_NVME_QPRIO_URGENT)) {
215 		SPDK_ERRLOG("invalid queue priority for default round robin arbitration method\n");
216 		return NULL;
217 	}
218 
219 	nvme_robust_mutex_lock(&ctrlr->ctrlr_lock);
220 
221 	/*
222 	 * Get the first available I/O queue ID.
223 	 */
224 	qid = spdk_bit_array_find_first_set(ctrlr->free_io_qids, 1);
225 	if (qid > ctrlr->opts.num_io_queues) {
226 		SPDK_ERRLOG("No free I/O queue IDs\n");
227 		nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
228 		return NULL;
229 	}
230 
231 	qpair = nvme_transport_ctrlr_create_io_qpair(ctrlr, qid, &opts);
232 	if (qpair == NULL) {
233 		SPDK_ERRLOG("nvme_transport_ctrlr_create_io_qpair() failed\n");
234 		nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
235 		return NULL;
236 	}
237 	spdk_bit_array_clear(ctrlr->free_io_qids, qid);
238 	TAILQ_INSERT_TAIL(&ctrlr->active_io_qpairs, qpair, tailq);
239 
240 	nvme_ctrlr_proc_add_io_qpair(qpair);
241 
242 	nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
243 
244 	if (ctrlr->quirks & NVME_QUIRK_DELAY_AFTER_QUEUE_ALLOC) {
245 		spdk_delay_us(100);
246 	}
247 
248 	return qpair;
249 }
250 
251 int
252 spdk_nvme_ctrlr_free_io_qpair(struct spdk_nvme_qpair *qpair)
253 {
254 	struct spdk_nvme_ctrlr *ctrlr;
255 	void *req_buf;
256 
257 	if (qpair == NULL) {
258 		return 0;
259 	}
260 
261 	ctrlr = qpair->ctrlr;
262 
263 	if (qpair->in_completion_context) {
264 		/*
265 		 * There are many cases where it is convenient to delete an io qpair in the context
266 		 *  of that qpair's completion routine.  To handle this properly, set a flag here
267 		 *  so that the completion routine will perform an actual delete after the context
268 		 *  unwinds.
269 		 */
270 		qpair->delete_after_completion_context = 1;
271 		return 0;
272 	}
273 
274 	nvme_robust_mutex_lock(&ctrlr->ctrlr_lock);
275 
276 	nvme_ctrlr_proc_remove_io_qpair(qpair);
277 
278 	TAILQ_REMOVE(&ctrlr->active_io_qpairs, qpair, tailq);
279 	spdk_bit_array_set(ctrlr->free_io_qids, qpair->id);
280 
281 	req_buf = qpair->req_buf;
282 
283 	if (nvme_transport_ctrlr_delete_io_qpair(ctrlr, qpair)) {
284 		nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
285 		return -1;
286 	}
287 
288 	spdk_dma_free(req_buf);
289 
290 	nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
291 	return 0;
292 }
293 
294 static void
295 nvme_ctrlr_construct_intel_support_log_page_list(struct spdk_nvme_ctrlr *ctrlr,
296 		struct spdk_nvme_intel_log_page_directory *log_page_directory)
297 {
298 	if (log_page_directory == NULL) {
299 		return;
300 	}
301 
302 	if (ctrlr->cdata.vid != SPDK_PCI_VID_INTEL) {
303 		return;
304 	}
305 
306 	ctrlr->log_page_supported[SPDK_NVME_INTEL_LOG_PAGE_DIRECTORY] = true;
307 
308 	if (log_page_directory->read_latency_log_len ||
309 	    (ctrlr->quirks & NVME_INTEL_QUIRK_READ_LATENCY)) {
310 		ctrlr->log_page_supported[SPDK_NVME_INTEL_LOG_READ_CMD_LATENCY] = true;
311 	}
312 	if (log_page_directory->write_latency_log_len ||
313 	    (ctrlr->quirks & NVME_INTEL_QUIRK_WRITE_LATENCY)) {
314 		ctrlr->log_page_supported[SPDK_NVME_INTEL_LOG_WRITE_CMD_LATENCY] = true;
315 	}
316 	if (log_page_directory->temperature_statistics_log_len) {
317 		ctrlr->log_page_supported[SPDK_NVME_INTEL_LOG_TEMPERATURE] = true;
318 	}
319 	if (log_page_directory->smart_log_len) {
320 		ctrlr->log_page_supported[SPDK_NVME_INTEL_LOG_SMART] = true;
321 	}
322 	if (log_page_directory->marketing_description_log_len) {
323 		ctrlr->log_page_supported[SPDK_NVME_INTEL_MARKETING_DESCRIPTION] = true;
324 	}
325 }
326 
327 static int nvme_ctrlr_set_intel_support_log_pages(struct spdk_nvme_ctrlr *ctrlr)
328 {
329 	uint64_t phys_addr = 0;
330 	struct nvme_completion_poll_status	status;
331 	struct spdk_nvme_intel_log_page_directory *log_page_directory;
332 
333 	log_page_directory = spdk_dma_zmalloc(sizeof(struct spdk_nvme_intel_log_page_directory),
334 					      64, &phys_addr);
335 	if (log_page_directory == NULL) {
336 		SPDK_ERRLOG("could not allocate log_page_directory\n");
337 		return -ENXIO;
338 	}
339 
340 	status.done = false;
341 	spdk_nvme_ctrlr_cmd_get_log_page(ctrlr, SPDK_NVME_INTEL_LOG_PAGE_DIRECTORY, SPDK_NVME_GLOBAL_NS_TAG,
342 					 log_page_directory, sizeof(struct spdk_nvme_intel_log_page_directory), 0,
343 					 nvme_completion_poll_cb,
344 					 &status);
345 	while (status.done == false) {
346 		spdk_nvme_qpair_process_completions(ctrlr->adminq, 0);
347 	}
348 	if (spdk_nvme_cpl_is_error(&status.cpl)) {
349 		spdk_dma_free(log_page_directory);
350 		SPDK_ERRLOG("nvme_ctrlr_cmd_get_log_page failed!\n");
351 		return -ENXIO;
352 	}
353 
354 	nvme_ctrlr_construct_intel_support_log_page_list(ctrlr, log_page_directory);
355 	spdk_dma_free(log_page_directory);
356 	return 0;
357 }
358 
359 static void
360 nvme_ctrlr_set_supported_log_pages(struct spdk_nvme_ctrlr *ctrlr)
361 {
362 	memset(ctrlr->log_page_supported, 0, sizeof(ctrlr->log_page_supported));
363 	/* Mandatory pages */
364 	ctrlr->log_page_supported[SPDK_NVME_LOG_ERROR] = true;
365 	ctrlr->log_page_supported[SPDK_NVME_LOG_HEALTH_INFORMATION] = true;
366 	ctrlr->log_page_supported[SPDK_NVME_LOG_FIRMWARE_SLOT] = true;
367 	if (ctrlr->cdata.lpa.celp) {
368 		ctrlr->log_page_supported[SPDK_NVME_LOG_COMMAND_EFFECTS_LOG] = true;
369 	}
370 	if (ctrlr->cdata.vid == SPDK_PCI_VID_INTEL) {
371 		nvme_ctrlr_set_intel_support_log_pages(ctrlr);
372 	}
373 }
374 
375 static void
376 nvme_ctrlr_set_intel_supported_features(struct spdk_nvme_ctrlr *ctrlr)
377 {
378 	ctrlr->feature_supported[SPDK_NVME_INTEL_FEAT_MAX_LBA] = true;
379 	ctrlr->feature_supported[SPDK_NVME_INTEL_FEAT_NATIVE_MAX_LBA] = true;
380 	ctrlr->feature_supported[SPDK_NVME_INTEL_FEAT_POWER_GOVERNOR_SETTING] = true;
381 	ctrlr->feature_supported[SPDK_NVME_INTEL_FEAT_SMBUS_ADDRESS] = true;
382 	ctrlr->feature_supported[SPDK_NVME_INTEL_FEAT_LED_PATTERN] = true;
383 	ctrlr->feature_supported[SPDK_NVME_INTEL_FEAT_RESET_TIMED_WORKLOAD_COUNTERS] = true;
384 	ctrlr->feature_supported[SPDK_NVME_INTEL_FEAT_LATENCY_TRACKING] = true;
385 }
386 
387 static void
388 nvme_ctrlr_set_supported_features(struct spdk_nvme_ctrlr *ctrlr)
389 {
390 	memset(ctrlr->feature_supported, 0, sizeof(ctrlr->feature_supported));
391 	/* Mandatory features */
392 	ctrlr->feature_supported[SPDK_NVME_FEAT_ARBITRATION] = true;
393 	ctrlr->feature_supported[SPDK_NVME_FEAT_POWER_MANAGEMENT] = true;
394 	ctrlr->feature_supported[SPDK_NVME_FEAT_TEMPERATURE_THRESHOLD] = true;
395 	ctrlr->feature_supported[SPDK_NVME_FEAT_ERROR_RECOVERY] = true;
396 	ctrlr->feature_supported[SPDK_NVME_FEAT_NUMBER_OF_QUEUES] = true;
397 	ctrlr->feature_supported[SPDK_NVME_FEAT_INTERRUPT_COALESCING] = true;
398 	ctrlr->feature_supported[SPDK_NVME_FEAT_INTERRUPT_VECTOR_CONFIGURATION] = true;
399 	ctrlr->feature_supported[SPDK_NVME_FEAT_WRITE_ATOMICITY] = true;
400 	ctrlr->feature_supported[SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION] = true;
401 	/* Optional features */
402 	if (ctrlr->cdata.vwc.present) {
403 		ctrlr->feature_supported[SPDK_NVME_FEAT_VOLATILE_WRITE_CACHE] = true;
404 	}
405 	if (ctrlr->cdata.apsta.supported) {
406 		ctrlr->feature_supported[SPDK_NVME_FEAT_AUTONOMOUS_POWER_STATE_TRANSITION] = true;
407 	}
408 	if (ctrlr->cdata.hmpre) {
409 		ctrlr->feature_supported[SPDK_NVME_FEAT_HOST_MEM_BUFFER] = true;
410 	}
411 	if (ctrlr->cdata.vid == SPDK_PCI_VID_INTEL) {
412 		nvme_ctrlr_set_intel_supported_features(ctrlr);
413 	}
414 }
415 
416 void
417 nvme_ctrlr_fail(struct spdk_nvme_ctrlr *ctrlr, bool hot_remove)
418 {
419 	/*
420 	 * Set the flag here and leave the work failure of qpairs to
421 	 * spdk_nvme_qpair_process_completions().
422 	 */
423 	if (hot_remove) {
424 		ctrlr->is_removed = true;
425 	}
426 	ctrlr->is_failed = true;
427 }
428 
429 static void
430 nvme_ctrlr_shutdown(struct spdk_nvme_ctrlr *ctrlr)
431 {
432 	union spdk_nvme_cc_register	cc;
433 	union spdk_nvme_csts_register	csts;
434 	int				ms_waited = 0;
435 
436 	if (ctrlr->is_removed) {
437 		return;
438 	}
439 
440 	if (nvme_ctrlr_get_cc(ctrlr, &cc)) {
441 		SPDK_ERRLOG("get_cc() failed\n");
442 		return;
443 	}
444 
445 	cc.bits.shn = SPDK_NVME_SHN_NORMAL;
446 
447 	if (nvme_ctrlr_set_cc(ctrlr, &cc)) {
448 		SPDK_ERRLOG("set_cc() failed\n");
449 		return;
450 	}
451 
452 	/*
453 	 * The NVMe spec does not define a timeout period
454 	 *  for shutdown notification, so we just pick
455 	 *  5 seconds as a reasonable amount of time to
456 	 *  wait before proceeding.
457 	 */
458 	do {
459 		if (nvme_ctrlr_get_csts(ctrlr, &csts)) {
460 			SPDK_ERRLOG("get_csts() failed\n");
461 			return;
462 		}
463 
464 		if (csts.bits.shst == SPDK_NVME_SHST_COMPLETE) {
465 			SPDK_TRACELOG(SPDK_TRACE_NVME, "shutdown complete\n");
466 			return;
467 		}
468 
469 		nvme_delay(1000);
470 		ms_waited++;
471 	} while (ms_waited < 5000);
472 
473 	SPDK_ERRLOG("did not shutdown within 5 seconds\n");
474 }
475 
476 static int
477 nvme_ctrlr_enable(struct spdk_nvme_ctrlr *ctrlr)
478 {
479 	union spdk_nvme_cc_register	cc;
480 	int				rc;
481 
482 	rc = nvme_transport_ctrlr_enable(ctrlr);
483 	if (rc != 0) {
484 		SPDK_ERRLOG("transport ctrlr_enable failed\n");
485 		return rc;
486 	}
487 
488 	if (nvme_ctrlr_get_cc(ctrlr, &cc)) {
489 		SPDK_ERRLOG("get_cc() failed\n");
490 		return -EIO;
491 	}
492 
493 	if (cc.bits.en != 0) {
494 		SPDK_ERRLOG("%s called with CC.EN = 1\n", __func__);
495 		return -EINVAL;
496 	}
497 
498 	cc.bits.en = 1;
499 	cc.bits.css = 0;
500 	cc.bits.shn = 0;
501 	cc.bits.iosqes = 6; /* SQ entry size == 64 == 2^6 */
502 	cc.bits.iocqes = 4; /* CQ entry size == 16 == 2^4 */
503 
504 	/* Page size is 2 ^ (12 + mps). */
505 	cc.bits.mps = spdk_u32log2(ctrlr->page_size) - 12;
506 
507 	switch (ctrlr->opts.arb_mechanism) {
508 	case SPDK_NVME_CC_AMS_RR:
509 		break;
510 	case SPDK_NVME_CC_AMS_WRR:
511 		if (SPDK_NVME_CAP_AMS_WRR & ctrlr->cap.bits.ams) {
512 			break;
513 		}
514 		return -EINVAL;
515 	case SPDK_NVME_CC_AMS_VS:
516 		if (SPDK_NVME_CAP_AMS_VS & ctrlr->cap.bits.ams) {
517 			break;
518 		}
519 		return -EINVAL;
520 	default:
521 		return -EINVAL;
522 	}
523 
524 	cc.bits.ams = ctrlr->opts.arb_mechanism;
525 
526 	if (nvme_ctrlr_set_cc(ctrlr, &cc)) {
527 		SPDK_ERRLOG("set_cc() failed\n");
528 		return -EIO;
529 	}
530 
531 	return 0;
532 }
533 
534 #ifdef DEBUG
535 static const char *
536 nvme_ctrlr_state_string(enum nvme_ctrlr_state state)
537 {
538 	switch (state) {
539 	case NVME_CTRLR_STATE_INIT:
540 		return "init";
541 	case NVME_CTRLR_STATE_DISABLE_WAIT_FOR_READY_1:
542 		return "disable and wait for CSTS.RDY = 1";
543 	case NVME_CTRLR_STATE_DISABLE_WAIT_FOR_READY_0:
544 		return "disable and wait for CSTS.RDY = 0";
545 	case NVME_CTRLR_STATE_ENABLE:
546 		return "enable controller by writing CC.EN = 1";
547 	case NVME_CTRLR_STATE_ENABLE_WAIT_FOR_READY_1:
548 		return "wait for CSTS.RDY = 1";
549 	case NVME_CTRLR_STATE_READY:
550 		return "ready";
551 	}
552 	return "unknown";
553 };
554 #endif /* DEBUG */
555 
556 static void
557 nvme_ctrlr_set_state(struct spdk_nvme_ctrlr *ctrlr, enum nvme_ctrlr_state state,
558 		     uint64_t timeout_in_ms)
559 {
560 	ctrlr->state = state;
561 	if (timeout_in_ms == NVME_TIMEOUT_INFINITE) {
562 		SPDK_TRACELOG(SPDK_TRACE_NVME, "setting state to %s (no timeout)\n",
563 			      nvme_ctrlr_state_string(ctrlr->state));
564 		ctrlr->state_timeout_tsc = NVME_TIMEOUT_INFINITE;
565 	} else {
566 		SPDK_TRACELOG(SPDK_TRACE_NVME, "setting state to %s (timeout %" PRIu64 " ms)\n",
567 			      nvme_ctrlr_state_string(ctrlr->state), timeout_in_ms);
568 		ctrlr->state_timeout_tsc = spdk_get_ticks() + (timeout_in_ms * spdk_get_ticks_hz()) / 1000;
569 	}
570 }
571 
572 int
573 spdk_nvme_ctrlr_reset(struct spdk_nvme_ctrlr *ctrlr)
574 {
575 	int rc = 0;
576 	struct spdk_nvme_qpair	*qpair;
577 	struct nvme_request	*req, *tmp;
578 
579 	nvme_robust_mutex_lock(&ctrlr->ctrlr_lock);
580 
581 	if (ctrlr->is_resetting || ctrlr->is_failed) {
582 		/*
583 		 * Controller is already resetting or has failed.  Return
584 		 *  immediately since there is no need to kick off another
585 		 *  reset in these cases.
586 		 */
587 		nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
588 		return 0;
589 	}
590 
591 	ctrlr->is_resetting = true;
592 
593 	SPDK_NOTICELOG("resetting controller\n");
594 
595 	/* Free all of the queued abort requests */
596 	STAILQ_FOREACH_SAFE(req, &ctrlr->queued_aborts, stailq, tmp) {
597 		STAILQ_REMOVE_HEAD(&ctrlr->queued_aborts, stailq);
598 		nvme_free_request(req);
599 		ctrlr->outstanding_aborts--;
600 	}
601 
602 	/* Disable all queues before disabling the controller hardware. */
603 	nvme_qpair_disable(ctrlr->adminq);
604 	TAILQ_FOREACH(qpair, &ctrlr->active_io_qpairs, tailq) {
605 		nvme_qpair_disable(qpair);
606 	}
607 
608 	/* Set the state back to INIT to cause a full hardware reset. */
609 	nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_INIT, NVME_TIMEOUT_INFINITE);
610 
611 	while (ctrlr->state != NVME_CTRLR_STATE_READY) {
612 		if (nvme_ctrlr_process_init(ctrlr) != 0) {
613 			SPDK_ERRLOG("%s: controller reinitialization failed\n", __func__);
614 			nvme_ctrlr_fail(ctrlr, false);
615 			rc = -1;
616 			break;
617 		}
618 	}
619 
620 	if (!ctrlr->is_failed) {
621 		/* Reinitialize qpairs */
622 		TAILQ_FOREACH(qpair, &ctrlr->active_io_qpairs, tailq) {
623 			if (nvme_transport_ctrlr_reinit_io_qpair(ctrlr, qpair) != 0) {
624 				nvme_ctrlr_fail(ctrlr, false);
625 				rc = -1;
626 			}
627 		}
628 	}
629 
630 	ctrlr->is_resetting = false;
631 
632 	nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
633 
634 	return rc;
635 }
636 
637 static int
638 nvme_ctrlr_identify(struct spdk_nvme_ctrlr *ctrlr)
639 {
640 	struct nvme_completion_poll_status	status;
641 	int					rc;
642 
643 	status.done = false;
644 	rc = nvme_ctrlr_cmd_identify_controller(ctrlr, &ctrlr->cdata,
645 						nvme_completion_poll_cb, &status);
646 	if (rc != 0) {
647 		return rc;
648 	}
649 
650 	while (status.done == false) {
651 		spdk_nvme_qpair_process_completions(ctrlr->adminq, 0);
652 	}
653 	if (spdk_nvme_cpl_is_error(&status.cpl)) {
654 		SPDK_ERRLOG("nvme_identify_controller failed!\n");
655 		return -ENXIO;
656 	}
657 
658 	/*
659 	 * Use MDTS to ensure our default max_xfer_size doesn't exceed what the
660 	 *  controller supports.
661 	 */
662 	ctrlr->max_xfer_size = nvme_transport_ctrlr_get_max_xfer_size(ctrlr);
663 	SPDK_TRACELOG(SPDK_TRACE_NVME, "transport max_xfer_size %u\n", ctrlr->max_xfer_size);
664 	if (ctrlr->cdata.mdts > 0) {
665 		ctrlr->max_xfer_size = spdk_min(ctrlr->max_xfer_size,
666 						ctrlr->min_page_size * (1 << (ctrlr->cdata.mdts)));
667 		SPDK_TRACELOG(SPDK_TRACE_NVME, "MDTS max_xfer_size %u\n", ctrlr->max_xfer_size);
668 	}
669 
670 	return 0;
671 }
672 
673 static int
674 nvme_ctrlr_set_num_qpairs(struct spdk_nvme_ctrlr *ctrlr)
675 {
676 	struct nvme_completion_poll_status	status;
677 	int					cq_allocated, sq_allocated;
678 	int					rc;
679 	uint32_t				i;
680 
681 	status.done = false;
682 
683 	if (ctrlr->opts.num_io_queues > SPDK_NVME_MAX_IO_QUEUES) {
684 		SPDK_NOTICELOG("Limiting requested num_io_queues %u to max %d\n",
685 			       ctrlr->opts.num_io_queues, SPDK_NVME_MAX_IO_QUEUES);
686 		ctrlr->opts.num_io_queues = SPDK_NVME_MAX_IO_QUEUES;
687 	} else if (ctrlr->opts.num_io_queues < 1) {
688 		SPDK_NOTICELOG("Requested num_io_queues 0, increasing to 1\n");
689 		ctrlr->opts.num_io_queues = 1;
690 	}
691 
692 	rc = nvme_ctrlr_cmd_set_num_queues(ctrlr, ctrlr->opts.num_io_queues,
693 					   nvme_completion_poll_cb, &status);
694 	if (rc != 0) {
695 		return rc;
696 	}
697 
698 	while (status.done == false) {
699 		spdk_nvme_qpair_process_completions(ctrlr->adminq, 0);
700 	}
701 	if (spdk_nvme_cpl_is_error(&status.cpl)) {
702 		SPDK_ERRLOG("nvme_set_num_queues failed!\n");
703 		return -ENXIO;
704 	}
705 
706 	/*
707 	 * Data in cdw0 is 0-based.
708 	 * Lower 16-bits indicate number of submission queues allocated.
709 	 * Upper 16-bits indicate number of completion queues allocated.
710 	 */
711 	sq_allocated = (status.cpl.cdw0 & 0xFFFF) + 1;
712 	cq_allocated = (status.cpl.cdw0 >> 16) + 1;
713 
714 	ctrlr->opts.num_io_queues = spdk_min(sq_allocated, cq_allocated);
715 
716 	ctrlr->free_io_qids = spdk_bit_array_create(ctrlr->opts.num_io_queues + 1);
717 	if (ctrlr->free_io_qids == NULL) {
718 		return -ENOMEM;
719 	}
720 
721 	/* Initialize list of free I/O queue IDs. QID 0 is the admin queue. */
722 	spdk_bit_array_clear(ctrlr->free_io_qids, 0);
723 	for (i = 1; i <= ctrlr->opts.num_io_queues; i++) {
724 		spdk_bit_array_set(ctrlr->free_io_qids, i);
725 	}
726 
727 	return 0;
728 }
729 
730 static int
731 nvme_ctrlr_set_keep_alive_timeout(struct spdk_nvme_ctrlr *ctrlr)
732 {
733 	struct nvme_completion_poll_status status;
734 	uint32_t keep_alive_interval_ms;
735 	int rc;
736 
737 	if (ctrlr->opts.keep_alive_timeout_ms == 0) {
738 		return 0;
739 	}
740 
741 	if (ctrlr->cdata.kas == 0) {
742 		SPDK_TRACELOG(SPDK_TRACE_NVME, "Controller KAS is 0 - not enabling Keep Alive\n");
743 		ctrlr->opts.keep_alive_timeout_ms = 0;
744 		return 0;
745 	}
746 
747 	/* Retrieve actual keep alive timeout, since the controller may have adjusted it. */
748 	status.done = false;
749 	rc = spdk_nvme_ctrlr_cmd_get_feature(ctrlr, SPDK_NVME_FEAT_KEEP_ALIVE_TIMER, 0, NULL, 0,
750 					     nvme_completion_poll_cb, &status);
751 	if (rc != 0) {
752 		SPDK_ERRLOG("Keep alive timeout Get Feature failed: %d\n", rc);
753 		ctrlr->opts.keep_alive_timeout_ms = 0;
754 		return rc;
755 	}
756 
757 	while (status.done == false) {
758 		spdk_nvme_qpair_process_completions(ctrlr->adminq, 0);
759 	}
760 	if (spdk_nvme_cpl_is_error(&status.cpl)) {
761 		SPDK_ERRLOG("Keep alive timeout Get Feature failed: SC %x SCT %x\n",
762 			    status.cpl.status.sc, status.cpl.status.sct);
763 		ctrlr->opts.keep_alive_timeout_ms = 0;
764 		return -ENXIO;
765 	}
766 
767 	if (ctrlr->opts.keep_alive_timeout_ms != status.cpl.cdw0) {
768 		SPDK_TRACELOG(SPDK_TRACE_NVME, "Controller adjusted keep alive timeout to %u ms\n",
769 			      status.cpl.cdw0);
770 	}
771 
772 	ctrlr->opts.keep_alive_timeout_ms = status.cpl.cdw0;
773 
774 	keep_alive_interval_ms = ctrlr->opts.keep_alive_timeout_ms / 2;
775 	if (keep_alive_interval_ms == 0) {
776 		keep_alive_interval_ms = 1;
777 	}
778 	SPDK_TRACELOG(SPDK_TRACE_NVME, "Sending keep alive every %u ms\n", keep_alive_interval_ms);
779 
780 	ctrlr->keep_alive_interval_ticks = (keep_alive_interval_ms * spdk_get_ticks_hz()) / UINT64_C(1000);
781 
782 	/* Schedule the first Keep Alive to be sent as soon as possible. */
783 	ctrlr->next_keep_alive_tick = spdk_get_ticks();
784 
785 	return 0;
786 }
787 
788 static void
789 nvme_ctrlr_destruct_namespaces(struct spdk_nvme_ctrlr *ctrlr)
790 {
791 	if (ctrlr->ns) {
792 		uint32_t i, num_ns = ctrlr->num_ns;
793 
794 		for (i = 0; i < num_ns; i++) {
795 			nvme_ns_destruct(&ctrlr->ns[i]);
796 		}
797 
798 		spdk_dma_free(ctrlr->ns);
799 		ctrlr->ns = NULL;
800 		ctrlr->num_ns = 0;
801 	}
802 
803 	if (ctrlr->nsdata) {
804 		spdk_dma_free(ctrlr->nsdata);
805 		ctrlr->nsdata = NULL;
806 	}
807 }
808 
809 static int
810 nvme_ctrlr_construct_namespaces(struct spdk_nvme_ctrlr *ctrlr)
811 {
812 	uint32_t i, nn = ctrlr->cdata.nn;
813 	uint64_t phys_addr = 0;
814 
815 	if (nn == 0) {
816 		SPDK_ERRLOG("controller has 0 namespaces\n");
817 		return -1;
818 	}
819 
820 	/* ctrlr->num_ns may be 0 (startup) or a different number of namespaces (reset),
821 	 * so check if we need to reallocate.
822 	 */
823 	if (nn != ctrlr->num_ns) {
824 		nvme_ctrlr_destruct_namespaces(ctrlr);
825 
826 		ctrlr->ns = spdk_dma_zmalloc(nn * sizeof(struct spdk_nvme_ns), 64,
827 					     &phys_addr);
828 		if (ctrlr->ns == NULL) {
829 			goto fail;
830 		}
831 
832 		ctrlr->nsdata = spdk_dma_zmalloc(nn * sizeof(struct spdk_nvme_ns_data), 64,
833 						 &phys_addr);
834 		if (ctrlr->nsdata == NULL) {
835 			goto fail;
836 		}
837 
838 		ctrlr->num_ns = nn;
839 	}
840 
841 	for (i = 0; i < nn; i++) {
842 		struct spdk_nvme_ns	*ns = &ctrlr->ns[i];
843 		uint32_t 		nsid = i + 1;
844 
845 		if (nvme_ns_construct(ns, nsid, ctrlr) != 0) {
846 			goto fail;
847 		}
848 	}
849 
850 	return 0;
851 
852 fail:
853 	nvme_ctrlr_destruct_namespaces(ctrlr);
854 	return -1;
855 }
856 
857 static void
858 nvme_ctrlr_async_event_cb(void *arg, const struct spdk_nvme_cpl *cpl)
859 {
860 	struct nvme_async_event_request	*aer = arg;
861 	struct spdk_nvme_ctrlr		*ctrlr = aer->ctrlr;
862 
863 	if (cpl->status.sc == SPDK_NVME_SC_ABORTED_SQ_DELETION) {
864 		/*
865 		 *  This is simulated when controller is being shut down, to
866 		 *  effectively abort outstanding asynchronous event requests
867 		 *  and make sure all memory is freed.  Do not repost the
868 		 *  request in this case.
869 		 */
870 		return;
871 	}
872 
873 	if (ctrlr->aer_cb_fn != NULL) {
874 		ctrlr->aer_cb_fn(ctrlr->aer_cb_arg, cpl);
875 	}
876 
877 	/*
878 	 * Repost another asynchronous event request to replace the one
879 	 *  that just completed.
880 	 */
881 	if (nvme_ctrlr_construct_and_submit_aer(ctrlr, aer)) {
882 		/*
883 		 * We can't do anything to recover from a failure here,
884 		 * so just print a warning message and leave the AER unsubmitted.
885 		 */
886 		SPDK_ERRLOG("resubmitting AER failed!\n");
887 	}
888 }
889 
890 static int
891 nvme_ctrlr_construct_and_submit_aer(struct spdk_nvme_ctrlr *ctrlr,
892 				    struct nvme_async_event_request *aer)
893 {
894 	struct nvme_request *req;
895 
896 	aer->ctrlr = ctrlr;
897 	req = nvme_allocate_request_null(ctrlr->adminq, nvme_ctrlr_async_event_cb, aer);
898 	aer->req = req;
899 	if (req == NULL) {
900 		return -1;
901 	}
902 
903 	req->cmd.opc = SPDK_NVME_OPC_ASYNC_EVENT_REQUEST;
904 	return nvme_ctrlr_submit_admin_request(ctrlr, req);
905 }
906 
907 static int
908 nvme_ctrlr_configure_aer(struct spdk_nvme_ctrlr *ctrlr)
909 {
910 	union spdk_nvme_critical_warning_state	state;
911 	struct nvme_async_event_request		*aer;
912 	uint32_t				i;
913 	struct nvme_completion_poll_status	status;
914 	int					rc;
915 
916 	status.done = false;
917 
918 	state.raw = 0xFF;
919 	state.bits.reserved = 0;
920 	rc = nvme_ctrlr_cmd_set_async_event_config(ctrlr, state, nvme_completion_poll_cb, &status);
921 	if (rc != 0) {
922 		return rc;
923 	}
924 
925 	while (status.done == false) {
926 		spdk_nvme_qpair_process_completions(ctrlr->adminq, 0);
927 	}
928 	if (spdk_nvme_cpl_is_error(&status.cpl)) {
929 		SPDK_ERRLOG("nvme_ctrlr_cmd_set_async_event_config failed!\n");
930 		return 0;
931 	}
932 
933 	/* aerl is a zero-based value, so we need to add 1 here. */
934 	ctrlr->num_aers = spdk_min(NVME_MAX_ASYNC_EVENTS, (ctrlr->cdata.aerl + 1));
935 
936 	for (i = 0; i < ctrlr->num_aers; i++) {
937 		aer = &ctrlr->aer[i];
938 		if (nvme_ctrlr_construct_and_submit_aer(ctrlr, aer)) {
939 			SPDK_ERRLOG("nvme_ctrlr_construct_and_submit_aer failed!\n");
940 			return -1;
941 		}
942 	}
943 
944 	return 0;
945 }
946 
947 /**
948  * This function will be called when a process is using the controller.
949  *  1. For the primary process, it is called when constructing the controller.
950  *  2. For the secondary process, it is called at probing the controller.
951  * Note: will check whether the process is already added for the same process.
952  */
953 int
954 nvme_ctrlr_add_process(struct spdk_nvme_ctrlr *ctrlr, void *devhandle)
955 {
956 	struct spdk_nvme_ctrlr_process	*ctrlr_proc, *active_proc;
957 	pid_t				pid = getpid();
958 
959 	/* Check whether the process is already added or not */
960 	TAILQ_FOREACH(active_proc, &ctrlr->active_procs, tailq) {
961 		if (active_proc->pid == pid) {
962 			return 0;
963 		}
964 	}
965 
966 	/* Initialize the per process properties for this ctrlr */
967 	ctrlr_proc = spdk_dma_zmalloc(sizeof(struct spdk_nvme_ctrlr_process), 64, NULL);
968 	if (ctrlr_proc == NULL) {
969 		SPDK_ERRLOG("failed to allocate memory to track the process props\n");
970 
971 		return -1;
972 	}
973 
974 	ctrlr_proc->is_primary = spdk_process_is_primary();
975 	ctrlr_proc->pid = pid;
976 	STAILQ_INIT(&ctrlr_proc->active_reqs);
977 	ctrlr_proc->devhandle = devhandle;
978 	ctrlr_proc->ref = 0;
979 	TAILQ_INIT(&ctrlr_proc->allocated_io_qpairs);
980 
981 	TAILQ_INSERT_TAIL(&ctrlr->active_procs, ctrlr_proc, tailq);
982 
983 	return 0;
984 }
985 
986 /**
987  * This function will be called when the process detaches the controller.
988  * Note: the ctrlr_lock must be held when calling this function.
989  */
990 static void
991 nvme_ctrlr_remove_process(struct spdk_nvme_ctrlr *ctrlr,
992 			  struct spdk_nvme_ctrlr_process *proc)
993 {
994 	struct spdk_nvme_qpair	*qpair, *tmp_qpair;
995 
996 	assert(STAILQ_EMPTY(&proc->active_reqs));
997 
998 	TAILQ_FOREACH_SAFE(qpair, &proc->allocated_io_qpairs, per_process_tailq, tmp_qpair) {
999 		spdk_nvme_ctrlr_free_io_qpair(qpair);
1000 	}
1001 
1002 	TAILQ_REMOVE(&ctrlr->active_procs, proc, tailq);
1003 
1004 	spdk_dma_free(proc);
1005 }
1006 
1007 /**
1008  * This function will be called when the process exited unexpectedly
1009  *  in order to free any incomplete nvme request, allocated IO qpairs
1010  *  and allocated memory.
1011  * Note: the ctrlr_lock must be held when calling this function.
1012  */
1013 static void
1014 nvme_ctrlr_cleanup_process(struct spdk_nvme_ctrlr_process *proc)
1015 {
1016 	struct nvme_request	*req, *tmp_req;
1017 	struct spdk_nvme_qpair	*qpair, *tmp_qpair;
1018 
1019 	STAILQ_FOREACH_SAFE(req, &proc->active_reqs, stailq, tmp_req) {
1020 		STAILQ_REMOVE(&proc->active_reqs, req, nvme_request, stailq);
1021 
1022 		assert(req->pid == proc->pid);
1023 
1024 		nvme_free_request(req);
1025 	}
1026 
1027 	TAILQ_FOREACH_SAFE(qpair, &proc->allocated_io_qpairs, per_process_tailq, tmp_qpair) {
1028 		TAILQ_REMOVE(&proc->allocated_io_qpairs, qpair, per_process_tailq);
1029 
1030 		/*
1031 		 * The process may have been killed while some qpairs were in their
1032 		 *  completion context.  Clear that flag here to allow these IO
1033 		 *  qpairs to be deleted.
1034 		 */
1035 		qpair->in_completion_context = 0;
1036 
1037 		qpair->no_deletion_notification_needed = 1;
1038 
1039 		spdk_nvme_ctrlr_free_io_qpair(qpair);
1040 	}
1041 
1042 	spdk_dma_free(proc);
1043 }
1044 
1045 /**
1046  * This function will be called when destructing the controller.
1047  *  1. There is no more admin request on this controller.
1048  *  2. Clean up any left resource allocation when its associated process is gone.
1049  */
1050 void
1051 nvme_ctrlr_free_processes(struct spdk_nvme_ctrlr *ctrlr)
1052 {
1053 	struct spdk_nvme_ctrlr_process	*active_proc, *tmp;
1054 
1055 	/* Free all the processes' properties and make sure no pending admin IOs */
1056 	TAILQ_FOREACH_SAFE(active_proc, &ctrlr->active_procs, tailq, tmp) {
1057 		TAILQ_REMOVE(&ctrlr->active_procs, active_proc, tailq);
1058 
1059 		assert(STAILQ_EMPTY(&active_proc->active_reqs));
1060 
1061 		spdk_dma_free(active_proc);
1062 	}
1063 }
1064 
1065 /**
1066  * This function will be called when any other process attaches or
1067  *  detaches the controller in order to cleanup those unexpectedly
1068  *  terminated processes.
1069  * Note: the ctrlr_lock must be held when calling this function.
1070  */
1071 static int
1072 nvme_ctrlr_remove_inactive_proc(struct spdk_nvme_ctrlr *ctrlr)
1073 {
1074 	struct spdk_nvme_ctrlr_process	*active_proc, *tmp;
1075 	int				active_proc_count = 0;
1076 
1077 	TAILQ_FOREACH_SAFE(active_proc, &ctrlr->active_procs, tailq, tmp) {
1078 		if ((kill(active_proc->pid, 0) == -1) && (errno == ESRCH)) {
1079 			SPDK_ERRLOG("process %d terminated unexpected\n", active_proc->pid);
1080 
1081 			TAILQ_REMOVE(&ctrlr->active_procs, active_proc, tailq);
1082 
1083 			nvme_ctrlr_cleanup_process(active_proc);
1084 		} else {
1085 			active_proc_count++;
1086 		}
1087 	}
1088 
1089 	return active_proc_count;
1090 }
1091 
1092 void
1093 nvme_ctrlr_proc_get_ref(struct spdk_nvme_ctrlr *ctrlr)
1094 {
1095 	struct spdk_nvme_ctrlr_process	*active_proc;
1096 	pid_t				pid = getpid();
1097 
1098 	nvme_robust_mutex_lock(&ctrlr->ctrlr_lock);
1099 
1100 	nvme_ctrlr_remove_inactive_proc(ctrlr);
1101 
1102 	TAILQ_FOREACH(active_proc, &ctrlr->active_procs, tailq) {
1103 		if (active_proc->pid == pid) {
1104 			active_proc->ref++;
1105 			break;
1106 		}
1107 	}
1108 
1109 	nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
1110 }
1111 
1112 void
1113 nvme_ctrlr_proc_put_ref(struct spdk_nvme_ctrlr *ctrlr)
1114 {
1115 	struct spdk_nvme_ctrlr_process	*active_proc, *tmp;
1116 	pid_t				pid = getpid();
1117 	int				proc_count;
1118 
1119 	nvme_robust_mutex_lock(&ctrlr->ctrlr_lock);
1120 
1121 	proc_count = nvme_ctrlr_remove_inactive_proc(ctrlr);
1122 
1123 	TAILQ_FOREACH_SAFE(active_proc, &ctrlr->active_procs, tailq, tmp) {
1124 		if (active_proc->pid == pid) {
1125 			active_proc->ref--;
1126 			assert(active_proc->ref >= 0);
1127 
1128 			/*
1129 			 * The last active process will be removed at the end of
1130 			 * the destruction of the controller.
1131 			 */
1132 			if (active_proc->ref == 0 && proc_count != 1) {
1133 				nvme_ctrlr_remove_process(ctrlr, active_proc);
1134 			}
1135 
1136 			break;
1137 		}
1138 	}
1139 
1140 	nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
1141 }
1142 
1143 int
1144 nvme_ctrlr_get_ref_count(struct spdk_nvme_ctrlr *ctrlr)
1145 {
1146 	struct spdk_nvme_ctrlr_process	*active_proc;
1147 	int				ref = 0;
1148 
1149 	nvme_robust_mutex_lock(&ctrlr->ctrlr_lock);
1150 
1151 	nvme_ctrlr_remove_inactive_proc(ctrlr);
1152 
1153 	TAILQ_FOREACH(active_proc, &ctrlr->active_procs, tailq) {
1154 		ref += active_proc->ref;
1155 	}
1156 
1157 	nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
1158 
1159 	return ref;
1160 }
1161 
1162 /**
1163  *  Get the PCI device handle which is only visible to its associated process.
1164  */
1165 struct spdk_pci_device *
1166 nvme_ctrlr_proc_get_devhandle(struct spdk_nvme_ctrlr *ctrlr)
1167 {
1168 	struct spdk_nvme_ctrlr_process	*active_proc;
1169 	pid_t				pid = getpid();
1170 	struct spdk_pci_device		*devhandle = NULL;
1171 
1172 	nvme_robust_mutex_lock(&ctrlr->ctrlr_lock);
1173 
1174 	TAILQ_FOREACH(active_proc, &ctrlr->active_procs, tailq) {
1175 		if (active_proc->pid == pid) {
1176 			devhandle = active_proc->devhandle;
1177 			break;
1178 		}
1179 	}
1180 
1181 	nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
1182 
1183 	return devhandle;
1184 }
1185 
1186 /**
1187  * This function will be called repeatedly during initialization until the controller is ready.
1188  */
1189 int
1190 nvme_ctrlr_process_init(struct spdk_nvme_ctrlr *ctrlr)
1191 {
1192 	union spdk_nvme_cc_register cc;
1193 	union spdk_nvme_csts_register csts;
1194 	uint32_t ready_timeout_in_ms;
1195 	int rc;
1196 
1197 	/*
1198 	 * May need to avoid accessing any register on the target controller
1199 	 * for a while. Return early without touching the FSM.
1200 	 * Check sleep_timeout_tsc > 0 for unit test.
1201 	 */
1202 	if ((ctrlr->sleep_timeout_tsc > 0) &&
1203 	    (spdk_get_ticks() <= ctrlr->sleep_timeout_tsc)) {
1204 		return 0;
1205 	}
1206 	ctrlr->sleep_timeout_tsc = 0;
1207 
1208 	if (nvme_ctrlr_get_cc(ctrlr, &cc) ||
1209 	    nvme_ctrlr_get_csts(ctrlr, &csts)) {
1210 		if (ctrlr->state_timeout_tsc != NVME_TIMEOUT_INFINITE) {
1211 			/* While a device is resetting, it may be unable to service MMIO reads
1212 			 * temporarily. Allow for this case.
1213 			 */
1214 			SPDK_TRACELOG(SPDK_TRACE_NVME, "Get registers failed while waiting for CSTS.RDY == 0\n");
1215 			goto init_timeout;
1216 		}
1217 		SPDK_ERRLOG("Failed to read CC and CSTS in state %d\n", ctrlr->state);
1218 		nvme_ctrlr_fail(ctrlr, false);
1219 		return -EIO;
1220 	}
1221 
1222 	ready_timeout_in_ms = 500 * ctrlr->cap.bits.to;
1223 
1224 	/*
1225 	 * Check if the current initialization step is done or has timed out.
1226 	 */
1227 	switch (ctrlr->state) {
1228 	case NVME_CTRLR_STATE_INIT:
1229 		/* Begin the hardware initialization by making sure the controller is disabled. */
1230 		if (cc.bits.en) {
1231 			SPDK_TRACELOG(SPDK_TRACE_NVME, "CC.EN = 1\n");
1232 			/*
1233 			 * Controller is currently enabled. We need to disable it to cause a reset.
1234 			 *
1235 			 * If CC.EN = 1 && CSTS.RDY = 0, the controller is in the process of becoming ready.
1236 			 *  Wait for the ready bit to be 1 before disabling the controller.
1237 			 */
1238 			if (csts.bits.rdy == 0) {
1239 				SPDK_TRACELOG(SPDK_TRACE_NVME, "CC.EN = 1 && CSTS.RDY = 0 - waiting for reset to complete\n");
1240 				nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_DISABLE_WAIT_FOR_READY_1, ready_timeout_in_ms);
1241 				return 0;
1242 			}
1243 
1244 			/* CC.EN = 1 && CSTS.RDY == 1, so we can immediately disable the controller. */
1245 			SPDK_TRACELOG(SPDK_TRACE_NVME, "Setting CC.EN = 0\n");
1246 			cc.bits.en = 0;
1247 			if (nvme_ctrlr_set_cc(ctrlr, &cc)) {
1248 				SPDK_ERRLOG("set_cc() failed\n");
1249 				nvme_ctrlr_fail(ctrlr, false);
1250 				return -EIO;
1251 			}
1252 			nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_DISABLE_WAIT_FOR_READY_0, ready_timeout_in_ms);
1253 
1254 			/*
1255 			 * Wait 2 secsonds before accessing PCI registers.
1256 			 * Not using sleep() to avoid blocking other controller's initialization.
1257 			 */
1258 			if (ctrlr->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY) {
1259 				SPDK_TRACELOG(SPDK_TRACE_NVME, "Applying quirk: delay 2 seconds before reading registers\n");
1260 				ctrlr->sleep_timeout_tsc = spdk_get_ticks() + 2 * spdk_get_ticks_hz();
1261 			}
1262 			return 0;
1263 		} else {
1264 			if (csts.bits.rdy == 1) {
1265 				SPDK_TRACELOG(SPDK_TRACE_NVME, "CC.EN = 0 && CSTS.RDY = 1 - waiting for shutdown to complete\n");
1266 			}
1267 
1268 			nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_DISABLE_WAIT_FOR_READY_0, ready_timeout_in_ms);
1269 			return 0;
1270 		}
1271 		break;
1272 
1273 	case NVME_CTRLR_STATE_DISABLE_WAIT_FOR_READY_1:
1274 		if (csts.bits.rdy == 1) {
1275 			SPDK_TRACELOG(SPDK_TRACE_NVME, "CC.EN = 1 && CSTS.RDY = 1 - disabling controller\n");
1276 			/* CC.EN = 1 && CSTS.RDY = 1, so we can set CC.EN = 0 now. */
1277 			SPDK_TRACELOG(SPDK_TRACE_NVME, "Setting CC.EN = 0\n");
1278 			cc.bits.en = 0;
1279 			if (nvme_ctrlr_set_cc(ctrlr, &cc)) {
1280 				SPDK_ERRLOG("set_cc() failed\n");
1281 				nvme_ctrlr_fail(ctrlr, false);
1282 				return -EIO;
1283 			}
1284 			nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_DISABLE_WAIT_FOR_READY_0, ready_timeout_in_ms);
1285 			return 0;
1286 		}
1287 		break;
1288 
1289 	case NVME_CTRLR_STATE_DISABLE_WAIT_FOR_READY_0:
1290 		if (csts.bits.rdy == 0) {
1291 			SPDK_TRACELOG(SPDK_TRACE_NVME, "CC.EN = 0 && CSTS.RDY = 0\n");
1292 			nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_ENABLE, ready_timeout_in_ms);
1293 			/*
1294 			 * Delay 100us before setting CC.EN = 1.  Some NVMe SSDs miss CC.EN getting
1295 			 *  set to 1 if it is too soon after CSTS.RDY is reported as 0.
1296 			 */
1297 			spdk_delay_us(100);
1298 			return 0;
1299 		}
1300 		break;
1301 
1302 	case NVME_CTRLR_STATE_ENABLE:
1303 		SPDK_TRACELOG(SPDK_TRACE_NVME, "Setting CC.EN = 1\n");
1304 		rc = nvme_ctrlr_enable(ctrlr);
1305 		nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_ENABLE_WAIT_FOR_READY_1, ready_timeout_in_ms);
1306 		return rc;
1307 
1308 	case NVME_CTRLR_STATE_ENABLE_WAIT_FOR_READY_1:
1309 		if (csts.bits.rdy == 1) {
1310 			SPDK_TRACELOG(SPDK_TRACE_NVME, "CC.EN = 1 && CSTS.RDY = 1 - controller is ready\n");
1311 			/*
1312 			 * The controller has been enabled.
1313 			 *  Perform the rest of initialization in nvme_ctrlr_start() serially.
1314 			 */
1315 			rc = nvme_ctrlr_start(ctrlr);
1316 			nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_READY, NVME_TIMEOUT_INFINITE);
1317 			return rc;
1318 		}
1319 		break;
1320 
1321 	default:
1322 		assert(0);
1323 		nvme_ctrlr_fail(ctrlr, false);
1324 		return -1;
1325 	}
1326 
1327 init_timeout:
1328 	if (ctrlr->state_timeout_tsc != NVME_TIMEOUT_INFINITE &&
1329 	    spdk_get_ticks() > ctrlr->state_timeout_tsc) {
1330 		SPDK_ERRLOG("Initialization timed out in state %d\n", ctrlr->state);
1331 		nvme_ctrlr_fail(ctrlr, false);
1332 		return -1;
1333 	}
1334 
1335 	return 0;
1336 }
1337 
1338 int
1339 nvme_ctrlr_start(struct spdk_nvme_ctrlr *ctrlr)
1340 {
1341 	nvme_transport_qpair_reset(ctrlr->adminq);
1342 
1343 	nvme_qpair_enable(ctrlr->adminq);
1344 
1345 	if (nvme_ctrlr_identify(ctrlr) != 0) {
1346 		return -1;
1347 	}
1348 
1349 	if (nvme_ctrlr_set_num_qpairs(ctrlr) != 0) {
1350 		return -1;
1351 	}
1352 
1353 	if (nvme_ctrlr_construct_namespaces(ctrlr) != 0) {
1354 		return -1;
1355 	}
1356 
1357 	if (nvme_ctrlr_configure_aer(ctrlr) != 0) {
1358 		return -1;
1359 	}
1360 
1361 	nvme_ctrlr_set_supported_log_pages(ctrlr);
1362 	nvme_ctrlr_set_supported_features(ctrlr);
1363 
1364 	if (ctrlr->cdata.sgls.supported) {
1365 		ctrlr->flags |= SPDK_NVME_CTRLR_SGL_SUPPORTED;
1366 		ctrlr->max_sges = nvme_transport_ctrlr_get_max_sges(ctrlr);
1367 	}
1368 
1369 	if (nvme_ctrlr_set_keep_alive_timeout(ctrlr) != 0) {
1370 		SPDK_ERRLOG("Setting keep alive timeout failed\n");
1371 		return -1;
1372 	}
1373 
1374 	return 0;
1375 }
1376 
1377 int
1378 nvme_robust_mutex_init_recursive_shared(pthread_mutex_t *mtx)
1379 {
1380 	pthread_mutexattr_t attr;
1381 	int rc = 0;
1382 
1383 	if (pthread_mutexattr_init(&attr)) {
1384 		return -1;
1385 	}
1386 	if (pthread_mutexattr_settype(&attr, PTHREAD_MUTEX_RECURSIVE) ||
1387 #ifndef __FreeBSD__
1388 	    pthread_mutexattr_setrobust(&attr, PTHREAD_MUTEX_ROBUST) ||
1389 	    pthread_mutexattr_setpshared(&attr, PTHREAD_PROCESS_SHARED) ||
1390 #endif
1391 	    pthread_mutex_init(mtx, &attr)) {
1392 		rc = -1;
1393 	}
1394 	pthread_mutexattr_destroy(&attr);
1395 	return rc;
1396 }
1397 
1398 int
1399 nvme_ctrlr_construct(struct spdk_nvme_ctrlr *ctrlr)
1400 {
1401 	int rc;
1402 
1403 	nvme_ctrlr_set_state(ctrlr, NVME_CTRLR_STATE_INIT, NVME_TIMEOUT_INFINITE);
1404 	ctrlr->flags = 0;
1405 	ctrlr->free_io_qids = NULL;
1406 	ctrlr->is_resetting = false;
1407 	ctrlr->is_failed = false;
1408 
1409 	TAILQ_INIT(&ctrlr->active_io_qpairs);
1410 	STAILQ_INIT(&ctrlr->queued_aborts);
1411 	ctrlr->outstanding_aborts = 0;
1412 
1413 	rc = nvme_robust_mutex_init_recursive_shared(&ctrlr->ctrlr_lock);
1414 	if (rc != 0) {
1415 		return rc;
1416 	}
1417 
1418 	TAILQ_INIT(&ctrlr->active_procs);
1419 	ctrlr->timeout_cb_fn = NULL;
1420 	ctrlr->timeout_cb_arg = NULL;
1421 	ctrlr->timeout_ticks = 0;
1422 
1423 	return rc;
1424 }
1425 
1426 /* This function should be called once at ctrlr initialization to set up constant properties. */
1427 void
1428 nvme_ctrlr_init_cap(struct spdk_nvme_ctrlr *ctrlr, const union spdk_nvme_cap_register *cap)
1429 {
1430 	uint32_t max_io_queue_size = nvme_transport_ctrlr_get_max_io_queue_size(ctrlr);
1431 
1432 	ctrlr->cap = *cap;
1433 
1434 	ctrlr->min_page_size = 1u << (12 + ctrlr->cap.bits.mpsmin);
1435 
1436 	/* For now, always select page_size == min_page_size. */
1437 	ctrlr->page_size = ctrlr->min_page_size;
1438 
1439 	ctrlr->opts.io_queue_size = spdk_max(ctrlr->opts.io_queue_size, SPDK_NVME_IO_QUEUE_MIN_ENTRIES);
1440 	ctrlr->opts.io_queue_size = spdk_min(ctrlr->opts.io_queue_size, ctrlr->cap.bits.mqes + 1u);
1441 	ctrlr->opts.io_queue_size = spdk_min(ctrlr->opts.io_queue_size, max_io_queue_size);
1442 
1443 	ctrlr->opts.io_queue_requests = spdk_max(ctrlr->opts.io_queue_requests, ctrlr->opts.io_queue_size);
1444 }
1445 
1446 void
1447 nvme_ctrlr_destruct(struct spdk_nvme_ctrlr *ctrlr)
1448 {
1449 	struct spdk_nvme_qpair *qpair, *tmp;
1450 
1451 	TAILQ_FOREACH_SAFE(qpair, &ctrlr->active_io_qpairs, tailq, tmp) {
1452 		spdk_nvme_ctrlr_free_io_qpair(qpair);
1453 	}
1454 
1455 	nvme_ctrlr_shutdown(ctrlr);
1456 
1457 	nvme_ctrlr_destruct_namespaces(ctrlr);
1458 
1459 	spdk_bit_array_free(&ctrlr->free_io_qids);
1460 
1461 	pthread_mutex_destroy(&ctrlr->ctrlr_lock);
1462 
1463 	nvme_transport_ctrlr_destruct(ctrlr);
1464 }
1465 
1466 int
1467 nvme_ctrlr_submit_admin_request(struct spdk_nvme_ctrlr *ctrlr,
1468 				struct nvme_request *req)
1469 {
1470 	return nvme_qpair_submit_request(ctrlr->adminq, req);
1471 }
1472 
1473 static void
1474 nvme_keep_alive_completion(void *cb_ctx, const struct spdk_nvme_cpl *cpl)
1475 {
1476 	/* Do nothing */
1477 }
1478 
1479 /*
1480  * Check if we need to send a Keep Alive command.
1481  * Caller must hold ctrlr->ctrlr_lock.
1482  */
1483 static void
1484 nvme_ctrlr_keep_alive(struct spdk_nvme_ctrlr *ctrlr)
1485 {
1486 	uint64_t now;
1487 	struct nvme_request *req;
1488 	struct spdk_nvme_cmd *cmd;
1489 	int rc;
1490 
1491 	now = spdk_get_ticks();
1492 	if (now < ctrlr->next_keep_alive_tick) {
1493 		return;
1494 	}
1495 
1496 	req = nvme_allocate_request_null(ctrlr->adminq, nvme_keep_alive_completion, NULL);
1497 	if (req == NULL) {
1498 		return;
1499 	}
1500 
1501 	cmd = &req->cmd;
1502 	cmd->opc = SPDK_NVME_OPC_KEEP_ALIVE;
1503 
1504 	rc = nvme_ctrlr_submit_admin_request(ctrlr, req);
1505 	if (rc != 0) {
1506 		SPDK_ERRLOG("Submitting Keep Alive failed\n");
1507 	}
1508 
1509 	ctrlr->next_keep_alive_tick = now + ctrlr->keep_alive_interval_ticks;
1510 }
1511 
1512 int32_t
1513 spdk_nvme_ctrlr_process_admin_completions(struct spdk_nvme_ctrlr *ctrlr)
1514 {
1515 	int32_t num_completions;
1516 
1517 	nvme_robust_mutex_lock(&ctrlr->ctrlr_lock);
1518 	if (ctrlr->keep_alive_interval_ticks) {
1519 		nvme_ctrlr_keep_alive(ctrlr);
1520 	}
1521 	num_completions = spdk_nvme_qpair_process_completions(ctrlr->adminq, 0);
1522 	nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
1523 
1524 	return num_completions;
1525 }
1526 
1527 const struct spdk_nvme_ctrlr_data *
1528 spdk_nvme_ctrlr_get_data(struct spdk_nvme_ctrlr *ctrlr)
1529 {
1530 	return &ctrlr->cdata;
1531 }
1532 
1533 union spdk_nvme_csts_register spdk_nvme_ctrlr_get_regs_csts(struct spdk_nvme_ctrlr *ctrlr)
1534 {
1535 	union spdk_nvme_csts_register csts;
1536 
1537 	if (nvme_ctrlr_get_csts(ctrlr, &csts)) {
1538 		csts.raw = 0;
1539 	}
1540 	return csts;
1541 }
1542 
1543 union spdk_nvme_cap_register spdk_nvme_ctrlr_get_regs_cap(struct spdk_nvme_ctrlr *ctrlr)
1544 {
1545 	return ctrlr->cap;
1546 }
1547 
1548 union spdk_nvme_vs_register spdk_nvme_ctrlr_get_regs_vs(struct spdk_nvme_ctrlr *ctrlr)
1549 {
1550 	union spdk_nvme_vs_register vs;
1551 
1552 	if (nvme_ctrlr_get_vs(ctrlr, &vs)) {
1553 		vs.raw = 0;
1554 	}
1555 	return vs;
1556 }
1557 
1558 uint32_t
1559 spdk_nvme_ctrlr_get_num_ns(struct spdk_nvme_ctrlr *ctrlr)
1560 {
1561 	return ctrlr->num_ns;
1562 }
1563 
1564 struct spdk_nvme_ns *
1565 spdk_nvme_ctrlr_get_ns(struct spdk_nvme_ctrlr *ctrlr, uint32_t ns_id)
1566 {
1567 	if (ns_id < 1 || ns_id > ctrlr->num_ns) {
1568 		return NULL;
1569 	}
1570 
1571 	return &ctrlr->ns[ns_id - 1];
1572 }
1573 
1574 void
1575 spdk_nvme_ctrlr_register_aer_callback(struct spdk_nvme_ctrlr *ctrlr,
1576 				      spdk_nvme_aer_cb aer_cb_fn,
1577 				      void *aer_cb_arg)
1578 {
1579 	ctrlr->aer_cb_fn = aer_cb_fn;
1580 	ctrlr->aer_cb_arg = aer_cb_arg;
1581 }
1582 
1583 void
1584 spdk_nvme_ctrlr_register_timeout_callback(struct spdk_nvme_ctrlr *ctrlr,
1585 		uint32_t nvme_timeout, spdk_nvme_timeout_cb cb_fn, void *cb_arg)
1586 {
1587 	ctrlr->timeout_ticks = nvme_timeout * spdk_get_ticks_hz();
1588 	ctrlr->timeout_cb_fn = cb_fn;
1589 	ctrlr->timeout_cb_arg = cb_arg;
1590 }
1591 
1592 bool
1593 spdk_nvme_ctrlr_is_log_page_supported(struct spdk_nvme_ctrlr *ctrlr, uint8_t log_page)
1594 {
1595 	/* No bounds check necessary, since log_page is uint8_t and log_page_supported has 256 entries */
1596 	SPDK_STATIC_ASSERT(sizeof(ctrlr->log_page_supported) == 256, "log_page_supported size mismatch");
1597 	return ctrlr->log_page_supported[log_page];
1598 }
1599 
1600 bool
1601 spdk_nvme_ctrlr_is_feature_supported(struct spdk_nvme_ctrlr *ctrlr, uint8_t feature_code)
1602 {
1603 	/* No bounds check necessary, since feature_code is uint8_t and feature_supported has 256 entries */
1604 	SPDK_STATIC_ASSERT(sizeof(ctrlr->feature_supported) == 256, "feature_supported size mismatch");
1605 	return ctrlr->feature_supported[feature_code];
1606 }
1607 
1608 int
1609 spdk_nvme_ctrlr_attach_ns(struct spdk_nvme_ctrlr *ctrlr, uint32_t nsid,
1610 			  struct spdk_nvme_ctrlr_list *payload)
1611 {
1612 	struct nvme_completion_poll_status	status;
1613 	int					res;
1614 
1615 	status.done = false;
1616 	res = nvme_ctrlr_cmd_attach_ns(ctrlr, nsid, payload,
1617 				       nvme_completion_poll_cb, &status);
1618 	if (res)
1619 		return res;
1620 	while (status.done == false) {
1621 		nvme_robust_mutex_lock(&ctrlr->ctrlr_lock);
1622 		spdk_nvme_qpair_process_completions(ctrlr->adminq, 0);
1623 		nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
1624 	}
1625 	if (spdk_nvme_cpl_is_error(&status.cpl)) {
1626 		SPDK_ERRLOG("spdk_nvme_ctrlr_attach_ns failed!\n");
1627 		return -ENXIO;
1628 	}
1629 
1630 	return spdk_nvme_ctrlr_reset(ctrlr);
1631 }
1632 
1633 int
1634 spdk_nvme_ctrlr_detach_ns(struct spdk_nvme_ctrlr *ctrlr, uint32_t nsid,
1635 			  struct spdk_nvme_ctrlr_list *payload)
1636 {
1637 	struct nvme_completion_poll_status	status;
1638 	int					res;
1639 
1640 	status.done = false;
1641 	res = nvme_ctrlr_cmd_detach_ns(ctrlr, nsid, payload,
1642 				       nvme_completion_poll_cb, &status);
1643 	if (res)
1644 		return res;
1645 	while (status.done == false) {
1646 		nvme_robust_mutex_lock(&ctrlr->ctrlr_lock);
1647 		spdk_nvme_qpair_process_completions(ctrlr->adminq, 0);
1648 		nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
1649 	}
1650 	if (spdk_nvme_cpl_is_error(&status.cpl)) {
1651 		SPDK_ERRLOG("spdk_nvme_ctrlr_detach_ns failed!\n");
1652 		return -ENXIO;
1653 	}
1654 
1655 	return spdk_nvme_ctrlr_reset(ctrlr);
1656 }
1657 
1658 uint32_t
1659 spdk_nvme_ctrlr_create_ns(struct spdk_nvme_ctrlr *ctrlr, struct spdk_nvme_ns_data *payload)
1660 {
1661 	struct nvme_completion_poll_status	status;
1662 	int					res;
1663 
1664 	status.done = false;
1665 	res = nvme_ctrlr_cmd_create_ns(ctrlr, payload, nvme_completion_poll_cb, &status);
1666 	if (res)
1667 		return 0;
1668 	while (status.done == false) {
1669 		nvme_robust_mutex_lock(&ctrlr->ctrlr_lock);
1670 		spdk_nvme_qpair_process_completions(ctrlr->adminq, 0);
1671 		nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
1672 	}
1673 	if (spdk_nvme_cpl_is_error(&status.cpl)) {
1674 		SPDK_ERRLOG("spdk_nvme_ctrlr_create_ns failed!\n");
1675 		return 0;
1676 	}
1677 
1678 	res = spdk_nvme_ctrlr_reset(ctrlr);
1679 	if (res) {
1680 		return 0;
1681 	}
1682 
1683 	/* Return the namespace ID that was created */
1684 	return status.cpl.cdw0;
1685 }
1686 
1687 int
1688 spdk_nvme_ctrlr_delete_ns(struct spdk_nvme_ctrlr *ctrlr, uint32_t nsid)
1689 {
1690 	struct nvme_completion_poll_status	status;
1691 	int					res;
1692 
1693 	status.done = false;
1694 	res = nvme_ctrlr_cmd_delete_ns(ctrlr, nsid, nvme_completion_poll_cb, &status);
1695 	if (res)
1696 		return res;
1697 	while (status.done == false) {
1698 		nvme_robust_mutex_lock(&ctrlr->ctrlr_lock);
1699 		spdk_nvme_qpair_process_completions(ctrlr->adminq, 0);
1700 		nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
1701 	}
1702 	if (spdk_nvme_cpl_is_error(&status.cpl)) {
1703 		SPDK_ERRLOG("spdk_nvme_ctrlr_delete_ns failed!\n");
1704 		return -ENXIO;
1705 	}
1706 
1707 	return spdk_nvme_ctrlr_reset(ctrlr);
1708 }
1709 
1710 int
1711 spdk_nvme_ctrlr_format(struct spdk_nvme_ctrlr *ctrlr, uint32_t nsid,
1712 		       struct spdk_nvme_format *format)
1713 {
1714 	struct nvme_completion_poll_status	status;
1715 	int					res;
1716 
1717 	status.done = false;
1718 	res = nvme_ctrlr_cmd_format(ctrlr, nsid, format, nvme_completion_poll_cb,
1719 				    &status);
1720 	if (res)
1721 		return res;
1722 	while (status.done == false) {
1723 		nvme_robust_mutex_lock(&ctrlr->ctrlr_lock);
1724 		spdk_nvme_qpair_process_completions(ctrlr->adminq, 0);
1725 		nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
1726 	}
1727 	if (spdk_nvme_cpl_is_error(&status.cpl)) {
1728 		SPDK_ERRLOG("spdk_nvme_ctrlr_format failed!\n");
1729 		return -ENXIO;
1730 	}
1731 
1732 	return spdk_nvme_ctrlr_reset(ctrlr);
1733 }
1734 
1735 int
1736 spdk_nvme_ctrlr_update_firmware(struct spdk_nvme_ctrlr *ctrlr, void *payload, uint32_t size,
1737 				int slot, enum spdk_nvme_fw_commit_action commit_action, struct spdk_nvme_status *completion_status)
1738 {
1739 	struct spdk_nvme_fw_commit		fw_commit;
1740 	struct nvme_completion_poll_status	status;
1741 	int					res;
1742 	unsigned int				size_remaining;
1743 	unsigned int				offset;
1744 	unsigned int				transfer;
1745 	void					*p;
1746 
1747 	if (!completion_status) {
1748 		return -EINVAL;
1749 	}
1750 	memset(completion_status, 0, sizeof(struct spdk_nvme_status));
1751 	if (size % 4) {
1752 		SPDK_ERRLOG("spdk_nvme_ctrlr_update_firmware invalid size!\n");
1753 		return -1;
1754 	}
1755 
1756 	/* Current support only for SPDK_NVME_FW_COMMIT_REPLACE_IMG
1757 	 * and SPDK_NVME_FW_COMMIT_REPLACE_AND_ENABLE_IMG
1758 	 */
1759 	if ((commit_action != SPDK_NVME_FW_COMMIT_REPLACE_IMG) &&
1760 	    (commit_action != SPDK_NVME_FW_COMMIT_REPLACE_AND_ENABLE_IMG)) {
1761 		SPDK_ERRLOG("spdk_nvme_ctrlr_update_firmware invalid command!\n");
1762 		return -1;
1763 	}
1764 
1765 	/* Firmware download */
1766 	size_remaining = size;
1767 	offset = 0;
1768 	p = payload;
1769 
1770 	while (size_remaining > 0) {
1771 		transfer = spdk_min(size_remaining, ctrlr->min_page_size);
1772 		status.done = false;
1773 
1774 		res = nvme_ctrlr_cmd_fw_image_download(ctrlr, transfer, offset, p,
1775 						       nvme_completion_poll_cb,
1776 						       &status);
1777 		if (res)
1778 			return res;
1779 
1780 		while (status.done == false) {
1781 			nvme_robust_mutex_lock(&ctrlr->ctrlr_lock);
1782 			spdk_nvme_qpair_process_completions(ctrlr->adminq, 0);
1783 			nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
1784 		}
1785 		if (spdk_nvme_cpl_is_error(&status.cpl)) {
1786 			SPDK_ERRLOG("spdk_nvme_ctrlr_fw_image_download failed!\n");
1787 			return -ENXIO;
1788 		}
1789 		p += transfer;
1790 		offset += transfer;
1791 		size_remaining -= transfer;
1792 	}
1793 
1794 	/* Firmware commit */
1795 	memset(&fw_commit, 0, sizeof(struct spdk_nvme_fw_commit));
1796 	fw_commit.fs = slot;
1797 	fw_commit.ca = commit_action;
1798 
1799 	status.done = false;
1800 
1801 	res = nvme_ctrlr_cmd_fw_commit(ctrlr, &fw_commit, nvme_completion_poll_cb,
1802 				       &status);
1803 	if (res)
1804 		return res;
1805 
1806 	while (status.done == false) {
1807 		nvme_robust_mutex_lock(&ctrlr->ctrlr_lock);
1808 		spdk_nvme_qpair_process_completions(ctrlr->adminq, 0);
1809 		nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
1810 	}
1811 	memcpy(completion_status, &status.cpl.status, sizeof(struct spdk_nvme_status));
1812 	if (spdk_nvme_cpl_is_error(&status.cpl)) {
1813 		if (status.cpl.status.sct != SPDK_NVME_SCT_COMMAND_SPECIFIC ||
1814 		    status.cpl.status.sc != SPDK_NVME_SC_FIRMWARE_REQ_NVM_RESET) {
1815 			if (status.cpl.status.sct == SPDK_NVME_SCT_COMMAND_SPECIFIC  &&
1816 			    status.cpl.status.sc == SPDK_NVME_SC_FIRMWARE_REQ_CONVENTIONAL_RESET) {
1817 				SPDK_NOTICELOG("firmware activation requires conventional reset to be performed. !\n");
1818 			} else {
1819 				SPDK_ERRLOG("nvme_ctrlr_cmd_fw_commit failed!\n");
1820 			}
1821 			return -ENXIO;
1822 		}
1823 	}
1824 
1825 	return spdk_nvme_ctrlr_reset(ctrlr);
1826 }
1827