xref: /freebsd/sys/dev/nvme/nvme_qpair.c (revision d3d381b2b194b4d24853e92eecef55f262688d1a)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (C) 2012-2014 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  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include <sys/param.h>
33 #include <sys/bus.h>
34 
35 #include <dev/pci/pcivar.h>
36 
37 #include "nvme_private.h"
38 
39 static void	_nvme_qpair_submit_request(struct nvme_qpair *qpair,
40 					   struct nvme_request *req);
41 static void	nvme_qpair_destroy(struct nvme_qpair *qpair);
42 
43 struct nvme_opcode_string {
44 
45 	uint16_t	opc;
46 	const char *	str;
47 };
48 
49 static struct nvme_opcode_string admin_opcode[] = {
50 	{ NVME_OPC_DELETE_IO_SQ, "DELETE IO SQ" },
51 	{ NVME_OPC_CREATE_IO_SQ, "CREATE IO SQ" },
52 	{ NVME_OPC_GET_LOG_PAGE, "GET LOG PAGE" },
53 	{ NVME_OPC_DELETE_IO_CQ, "DELETE IO CQ" },
54 	{ NVME_OPC_CREATE_IO_CQ, "CREATE IO CQ" },
55 	{ NVME_OPC_IDENTIFY, "IDENTIFY" },
56 	{ NVME_OPC_ABORT, "ABORT" },
57 	{ NVME_OPC_SET_FEATURES, "SET FEATURES" },
58 	{ NVME_OPC_GET_FEATURES, "GET FEATURES" },
59 	{ NVME_OPC_ASYNC_EVENT_REQUEST, "ASYNC EVENT REQUEST" },
60 	{ NVME_OPC_FIRMWARE_ACTIVATE, "FIRMWARE ACTIVATE" },
61 	{ NVME_OPC_FIRMWARE_IMAGE_DOWNLOAD, "FIRMWARE IMAGE DOWNLOAD" },
62 	{ NVME_OPC_DEVICE_SELF_TEST, "DEVICE SELF-TEST" },
63 	{ NVME_OPC_NAMESPACE_ATTACHMENT, "NAMESPACE ATTACHMENT" },
64 	{ NVME_OPC_KEEP_ALIVE, "KEEP ALIVE" },
65 	{ NVME_OPC_DIRECTIVE_SEND, "DIRECTIVE SEND" },
66 	{ NVME_OPC_DIRECTIVE_RECEIVE, "DIRECTIVE RECEIVE" },
67 	{ NVME_OPC_VIRTUALIZATION_MANAGEMENT, "VIRTUALIZATION MANAGEMENT" },
68 	{ NVME_OPC_NVME_MI_SEND, "NVME-MI SEND" },
69 	{ NVME_OPC_NVME_MI_RECEIVE, "NVME-MI RECEIVE" },
70 	{ NVME_OPC_DOORBELL_BUFFER_CONFIG, "DOORBELL BUFFER CONFIG" },
71 	{ NVME_OPC_FORMAT_NVM, "FORMAT NVM" },
72 	{ NVME_OPC_SECURITY_SEND, "SECURITY SEND" },
73 	{ NVME_OPC_SECURITY_RECEIVE, "SECURITY RECEIVE" },
74 	{ NVME_OPC_SANITIZE, "SANITIZE" },
75 	{ 0xFFFF, "ADMIN COMMAND" }
76 };
77 
78 static struct nvme_opcode_string io_opcode[] = {
79 	{ NVME_OPC_FLUSH, "FLUSH" },
80 	{ NVME_OPC_WRITE, "WRITE" },
81 	{ NVME_OPC_READ, "READ" },
82 	{ NVME_OPC_WRITE_UNCORRECTABLE, "WRITE UNCORRECTABLE" },
83 	{ NVME_OPC_COMPARE, "COMPARE" },
84 	{ NVME_OPC_WRITE_ZEROES, "WRITE ZEROES" },
85 	{ NVME_OPC_DATASET_MANAGEMENT, "DATASET MANAGEMENT" },
86 	{ NVME_OPC_RESERVATION_REGISTER, "RESERVATION REGISTER" },
87 	{ NVME_OPC_RESERVATION_REPORT, "RESERVATION REPORT" },
88 	{ NVME_OPC_RESERVATION_ACQUIRE, "RESERVATION ACQUIRE" },
89 	{ NVME_OPC_RESERVATION_RELEASE, "RESERVATION RELEASE" },
90 	{ 0xFFFF, "IO COMMAND" }
91 };
92 
93 static const char *
94 get_admin_opcode_string(uint16_t opc)
95 {
96 	struct nvme_opcode_string *entry;
97 
98 	entry = admin_opcode;
99 
100 	while (entry->opc != 0xFFFF) {
101 		if (entry->opc == opc)
102 			return (entry->str);
103 		entry++;
104 	}
105 	return (entry->str);
106 }
107 
108 static const char *
109 get_io_opcode_string(uint16_t opc)
110 {
111 	struct nvme_opcode_string *entry;
112 
113 	entry = io_opcode;
114 
115 	while (entry->opc != 0xFFFF) {
116 		if (entry->opc == opc)
117 			return (entry->str);
118 		entry++;
119 	}
120 	return (entry->str);
121 }
122 
123 
124 static void
125 nvme_admin_qpair_print_command(struct nvme_qpair *qpair,
126     struct nvme_command *cmd)
127 {
128 
129 	nvme_printf(qpair->ctrlr, "%s (%02x) sqid:%d cid:%d nsid:%x "
130 	    "cdw10:%08x cdw11:%08x\n",
131 	    get_admin_opcode_string(cmd->opc), cmd->opc, qpair->id, cmd->cid,
132 	    le32toh(cmd->nsid), le32toh(cmd->cdw10), le32toh(cmd->cdw11));
133 }
134 
135 static void
136 nvme_io_qpair_print_command(struct nvme_qpair *qpair,
137     struct nvme_command *cmd)
138 {
139 
140 	switch (cmd->opc) {
141 	case NVME_OPC_WRITE:
142 	case NVME_OPC_READ:
143 	case NVME_OPC_WRITE_UNCORRECTABLE:
144 	case NVME_OPC_COMPARE:
145 	case NVME_OPC_WRITE_ZEROES:
146 		nvme_printf(qpair->ctrlr, "%s sqid:%d cid:%d nsid:%d "
147 		    "lba:%llu len:%d\n",
148 		    get_io_opcode_string(cmd->opc), qpair->id, cmd->cid, le32toh(cmd->nsid),
149 		    ((unsigned long long)le32toh(cmd->cdw11) << 32) + le32toh(cmd->cdw10),
150 		    (le32toh(cmd->cdw12) & 0xFFFF) + 1);
151 		break;
152 	case NVME_OPC_FLUSH:
153 	case NVME_OPC_DATASET_MANAGEMENT:
154 	case NVME_OPC_RESERVATION_REGISTER:
155 	case NVME_OPC_RESERVATION_REPORT:
156 	case NVME_OPC_RESERVATION_ACQUIRE:
157 	case NVME_OPC_RESERVATION_RELEASE:
158 		nvme_printf(qpair->ctrlr, "%s sqid:%d cid:%d nsid:%d\n",
159 		    get_io_opcode_string(cmd->opc), qpair->id, cmd->cid, le32toh(cmd->nsid));
160 		break;
161 	default:
162 		nvme_printf(qpair->ctrlr, "%s (%02x) sqid:%d cid:%d nsid:%d\n",
163 		    get_io_opcode_string(cmd->opc), cmd->opc, qpair->id,
164 		    cmd->cid, le32toh(cmd->nsid));
165 		break;
166 	}
167 }
168 
169 static void
170 nvme_qpair_print_command(struct nvme_qpair *qpair, struct nvme_command *cmd)
171 {
172 	if (qpair->id == 0)
173 		nvme_admin_qpair_print_command(qpair, cmd);
174 	else
175 		nvme_io_qpair_print_command(qpair, cmd);
176 }
177 
178 struct nvme_status_string {
179 
180 	uint16_t	sc;
181 	const char *	str;
182 };
183 
184 static struct nvme_status_string generic_status[] = {
185 	{ NVME_SC_SUCCESS, "SUCCESS" },
186 	{ NVME_SC_INVALID_OPCODE, "INVALID OPCODE" },
187 	{ NVME_SC_INVALID_FIELD, "INVALID_FIELD" },
188 	{ NVME_SC_COMMAND_ID_CONFLICT, "COMMAND ID CONFLICT" },
189 	{ NVME_SC_DATA_TRANSFER_ERROR, "DATA TRANSFER ERROR" },
190 	{ NVME_SC_ABORTED_POWER_LOSS, "ABORTED - POWER LOSS" },
191 	{ NVME_SC_INTERNAL_DEVICE_ERROR, "INTERNAL DEVICE ERROR" },
192 	{ NVME_SC_ABORTED_BY_REQUEST, "ABORTED - BY REQUEST" },
193 	{ NVME_SC_ABORTED_SQ_DELETION, "ABORTED - SQ DELETION" },
194 	{ NVME_SC_ABORTED_FAILED_FUSED, "ABORTED - FAILED FUSED" },
195 	{ NVME_SC_ABORTED_MISSING_FUSED, "ABORTED - MISSING FUSED" },
196 	{ NVME_SC_INVALID_NAMESPACE_OR_FORMAT, "INVALID NAMESPACE OR FORMAT" },
197 	{ NVME_SC_COMMAND_SEQUENCE_ERROR, "COMMAND SEQUENCE ERROR" },
198 	{ NVME_SC_INVALID_SGL_SEGMENT_DESCR, "INVALID SGL SEGMENT DESCRIPTOR" },
199 	{ NVME_SC_INVALID_NUMBER_OF_SGL_DESCR, "INVALID NUMBER OF SGL DESCRIPTORS" },
200 	{ NVME_SC_DATA_SGL_LENGTH_INVALID, "DATA SGL LENGTH INVALID" },
201 	{ NVME_SC_METADATA_SGL_LENGTH_INVALID, "METADATA SGL LENGTH INVALID" },
202 	{ NVME_SC_SGL_DESCRIPTOR_TYPE_INVALID, "SGL DESCRIPTOR TYPE INVALID" },
203 	{ NVME_SC_INVALID_USE_OF_CMB, "INVALID USE OF CONTROLLER MEMORY BUFFER" },
204 	{ NVME_SC_PRP_OFFET_INVALID, "PRP OFFET INVALID" },
205 	{ NVME_SC_ATOMIC_WRITE_UNIT_EXCEEDED, "ATOMIC WRITE UNIT EXCEEDED" },
206 	{ NVME_SC_OPERATION_DENIED, "OPERATION DENIED" },
207 	{ NVME_SC_SGL_OFFSET_INVALID, "SGL OFFSET INVALID" },
208 	{ NVME_SC_HOST_ID_INCONSISTENT_FORMAT, "HOST IDENTIFIER INCONSISTENT FORMAT" },
209 	{ NVME_SC_KEEP_ALIVE_TIMEOUT_EXPIRED, "KEEP ALIVE TIMEOUT EXPIRED" },
210 	{ NVME_SC_KEEP_ALIVE_TIMEOUT_INVALID, "KEEP ALIVE TIMEOUT INVALID" },
211 	{ NVME_SC_ABORTED_DUE_TO_PREEMPT, "COMMAND ABORTED DUE TO PREEMPT AND ABORT" },
212 	{ NVME_SC_SANITIZE_FAILED, "SANITIZE FAILED" },
213 	{ NVME_SC_SANITIZE_IN_PROGRESS, "SANITIZE IN PROGRESS" },
214 	{ NVME_SC_SGL_DATA_BLOCK_GRAN_INVALID, "SGL_DATA_BLOCK_GRANULARITY_INVALID" },
215 	{ NVME_SC_NOT_SUPPORTED_IN_CMB, "COMMAND NOT SUPPORTED FOR QUEUE IN CMB" },
216 
217 	{ NVME_SC_LBA_OUT_OF_RANGE, "LBA OUT OF RANGE" },
218 	{ NVME_SC_CAPACITY_EXCEEDED, "CAPACITY EXCEEDED" },
219 	{ NVME_SC_NAMESPACE_NOT_READY, "NAMESPACE NOT READY" },
220 	{ NVME_SC_RESERVATION_CONFLICT, "RESERVATION CONFLICT" },
221 	{ NVME_SC_FORMAT_IN_PROGRESS, "FORMAT IN PROGRESS" },
222 	{ 0xFFFF, "GENERIC" }
223 };
224 
225 static struct nvme_status_string command_specific_status[] = {
226 	{ NVME_SC_COMPLETION_QUEUE_INVALID, "INVALID COMPLETION QUEUE" },
227 	{ NVME_SC_INVALID_QUEUE_IDENTIFIER, "INVALID QUEUE IDENTIFIER" },
228 	{ NVME_SC_MAXIMUM_QUEUE_SIZE_EXCEEDED, "MAX QUEUE SIZE EXCEEDED" },
229 	{ NVME_SC_ABORT_COMMAND_LIMIT_EXCEEDED, "ABORT CMD LIMIT EXCEEDED" },
230 	{ NVME_SC_ASYNC_EVENT_REQUEST_LIMIT_EXCEEDED, "ASYNC LIMIT EXCEEDED" },
231 	{ NVME_SC_INVALID_FIRMWARE_SLOT, "INVALID FIRMWARE SLOT" },
232 	{ NVME_SC_INVALID_FIRMWARE_IMAGE, "INVALID FIRMWARE IMAGE" },
233 	{ NVME_SC_INVALID_INTERRUPT_VECTOR, "INVALID INTERRUPT VECTOR" },
234 	{ NVME_SC_INVALID_LOG_PAGE, "INVALID LOG PAGE" },
235 	{ NVME_SC_INVALID_FORMAT, "INVALID FORMAT" },
236 	{ NVME_SC_FIRMWARE_REQUIRES_RESET, "FIRMWARE REQUIRES RESET" },
237 	{ NVME_SC_INVALID_QUEUE_DELETION, "INVALID QUEUE DELETION" },
238 	{ NVME_SC_FEATURE_NOT_SAVEABLE, "FEATURE IDENTIFIER NOT SAVEABLE" },
239 	{ NVME_SC_FEATURE_NOT_CHANGEABLE, "FEATURE NOT CHANGEABLE" },
240 	{ NVME_SC_FEATURE_NOT_NS_SPECIFIC, "FEATURE NOT NAMESPACE SPECIFIC" },
241 	{ NVME_SC_FW_ACT_REQUIRES_NVMS_RESET, "FIRMWARE ACTIVATION REQUIRES NVM SUBSYSTEM RESET" },
242 	{ NVME_SC_FW_ACT_REQUIRES_RESET, "FIRMWARE ACTIVATION REQUIRES RESET" },
243 	{ NVME_SC_FW_ACT_REQUIRES_TIME, "FIRMWARE ACTIVATION REQUIRES MAXIMUM TIME VIOLATION" },
244 	{ NVME_SC_FW_ACT_PROHIBITED, "FIRMWARE ACTIVATION PROHIBITED" },
245 	{ NVME_SC_OVERLAPPING_RANGE, "OVERLAPPING RANGE" },
246 	{ NVME_SC_NS_INSUFFICIENT_CAPACITY, "NAMESPACE INSUFFICIENT CAPACITY" },
247 	{ NVME_SC_NS_ID_UNAVAILABLE, "NAMESPACE IDENTIFIER UNAVAILABLE" },
248 	{ NVME_SC_NS_ALREADY_ATTACHED, "NAMESPACE ALREADY ATTACHED" },
249 	{ NVME_SC_NS_IS_PRIVATE, "NAMESPACE IS PRIVATE" },
250 	{ NVME_SC_NS_NOT_ATTACHED, "NS NOT ATTACHED" },
251 	{ NVME_SC_THIN_PROV_NOT_SUPPORTED, "THIN PROVISIONING NOT SUPPORTED" },
252 	{ NVME_SC_CTRLR_LIST_INVALID, "CONTROLLER LIST INVALID" },
253 	{ NVME_SC_SELT_TEST_IN_PROGRESS, "DEVICE SELT-TEST IN PROGRESS" },
254 	{ NVME_SC_BOOT_PART_WRITE_PROHIB, "BOOT PARTITION WRITE PROHIBITED" },
255 	{ NVME_SC_INVALID_CTRLR_ID, "INVALID CONTROLLER IDENTIFIER" },
256 	{ NVME_SC_INVALID_SEC_CTRLR_STATE, "INVALID SECONDARY CONTROLLER STATE" },
257 	{ NVME_SC_INVALID_NUM_OF_CTRLR_RESRC, "INVALID NUMBER OF CONTROLLER RESOURCES" },
258 	{ NVME_SC_INVALID_RESOURCE_ID, "INVALID RESOURCE IDENTIFIER" },
259 
260 	{ NVME_SC_CONFLICTING_ATTRIBUTES, "CONFLICTING ATTRIBUTES" },
261 	{ NVME_SC_INVALID_PROTECTION_INFO, "INVALID PROTECTION INFO" },
262 	{ NVME_SC_ATTEMPTED_WRITE_TO_RO_PAGE, "WRITE TO RO PAGE" },
263 	{ 0xFFFF, "COMMAND SPECIFIC" }
264 };
265 
266 static struct nvme_status_string media_error_status[] = {
267 	{ NVME_SC_WRITE_FAULTS, "WRITE FAULTS" },
268 	{ NVME_SC_UNRECOVERED_READ_ERROR, "UNRECOVERED READ ERROR" },
269 	{ NVME_SC_GUARD_CHECK_ERROR, "GUARD CHECK ERROR" },
270 	{ NVME_SC_APPLICATION_TAG_CHECK_ERROR, "APPLICATION TAG CHECK ERROR" },
271 	{ NVME_SC_REFERENCE_TAG_CHECK_ERROR, "REFERENCE TAG CHECK ERROR" },
272 	{ NVME_SC_COMPARE_FAILURE, "COMPARE FAILURE" },
273 	{ NVME_SC_ACCESS_DENIED, "ACCESS DENIED" },
274 	{ NVME_SC_DEALLOCATED_OR_UNWRITTEN, "DEALLOCATED OR UNWRITTEN LOGICAL BLOCK" },
275 	{ 0xFFFF, "MEDIA ERROR" }
276 };
277 
278 static const char *
279 get_status_string(uint16_t sct, uint16_t sc)
280 {
281 	struct nvme_status_string *entry;
282 
283 	switch (sct) {
284 	case NVME_SCT_GENERIC:
285 		entry = generic_status;
286 		break;
287 	case NVME_SCT_COMMAND_SPECIFIC:
288 		entry = command_specific_status;
289 		break;
290 	case NVME_SCT_MEDIA_ERROR:
291 		entry = media_error_status;
292 		break;
293 	case NVME_SCT_VENDOR_SPECIFIC:
294 		return ("VENDOR SPECIFIC");
295 	default:
296 		return ("RESERVED");
297 	}
298 
299 	while (entry->sc != 0xFFFF) {
300 		if (entry->sc == sc)
301 			return (entry->str);
302 		entry++;
303 	}
304 	return (entry->str);
305 }
306 
307 static void
308 nvme_qpair_print_completion(struct nvme_qpair *qpair,
309     struct nvme_completion *cpl)
310 {
311 	uint16_t sct, sc;
312 
313 	sct = NVME_STATUS_GET_SCT(cpl->status);
314 	sc = NVME_STATUS_GET_SC(cpl->status);
315 
316 	nvme_printf(qpair->ctrlr, "%s (%02x/%02x) sqid:%d cid:%d cdw0:%x\n",
317 	    get_status_string(sct, sc), sct, sc, cpl->sqid, cpl->cid,
318 	    cpl->cdw0);
319 }
320 
321 static boolean_t
322 nvme_completion_is_retry(const struct nvme_completion *cpl)
323 {
324 	uint8_t sct, sc, dnr;
325 
326 	sct = NVME_STATUS_GET_SCT(cpl->status);
327 	sc = NVME_STATUS_GET_SC(cpl->status);
328 	dnr = NVME_STATUS_GET_DNR(cpl->status);
329 
330 	/*
331 	 * TODO: spec is not clear how commands that are aborted due
332 	 *  to TLER will be marked.  So for now, it seems
333 	 *  NAMESPACE_NOT_READY is the only case where we should
334 	 *  look at the DNR bit.
335 	 */
336 	switch (sct) {
337 	case NVME_SCT_GENERIC:
338 		switch (sc) {
339 		case NVME_SC_ABORTED_BY_REQUEST:
340 		case NVME_SC_NAMESPACE_NOT_READY:
341 			if (dnr)
342 				return (0);
343 			else
344 				return (1);
345 		case NVME_SC_INVALID_OPCODE:
346 		case NVME_SC_INVALID_FIELD:
347 		case NVME_SC_COMMAND_ID_CONFLICT:
348 		case NVME_SC_DATA_TRANSFER_ERROR:
349 		case NVME_SC_ABORTED_POWER_LOSS:
350 		case NVME_SC_INTERNAL_DEVICE_ERROR:
351 		case NVME_SC_ABORTED_SQ_DELETION:
352 		case NVME_SC_ABORTED_FAILED_FUSED:
353 		case NVME_SC_ABORTED_MISSING_FUSED:
354 		case NVME_SC_INVALID_NAMESPACE_OR_FORMAT:
355 		case NVME_SC_COMMAND_SEQUENCE_ERROR:
356 		case NVME_SC_LBA_OUT_OF_RANGE:
357 		case NVME_SC_CAPACITY_EXCEEDED:
358 		default:
359 			return (0);
360 		}
361 	case NVME_SCT_COMMAND_SPECIFIC:
362 	case NVME_SCT_MEDIA_ERROR:
363 	case NVME_SCT_VENDOR_SPECIFIC:
364 	default:
365 		return (0);
366 	}
367 }
368 
369 static void
370 nvme_qpair_complete_tracker(struct nvme_qpair *qpair, struct nvme_tracker *tr,
371     struct nvme_completion *cpl, boolean_t print_on_error)
372 {
373 	struct nvme_request	*req;
374 	boolean_t		retry, error;
375 
376 	req = tr->req;
377 	error = nvme_completion_is_error(cpl);
378 	retry = error && nvme_completion_is_retry(cpl) &&
379 	   req->retries < nvme_retry_count;
380 
381 	if (error && print_on_error) {
382 		nvme_qpair_print_command(qpair, &req->cmd);
383 		nvme_qpair_print_completion(qpair, cpl);
384 	}
385 
386 	qpair->act_tr[cpl->cid] = NULL;
387 
388 	KASSERT(cpl->cid == req->cmd.cid, ("cpl cid does not match cmd cid\n"));
389 
390 	if (req->cb_fn && !retry)
391 		req->cb_fn(req->cb_arg, cpl);
392 
393 	mtx_lock(&qpair->lock);
394 	callout_stop(&tr->timer);
395 
396 	if (retry) {
397 		req->retries++;
398 		nvme_qpair_submit_tracker(qpair, tr);
399 	} else {
400 		if (req->type != NVME_REQUEST_NULL) {
401 			bus_dmamap_sync(qpair->dma_tag_payload,
402 			    tr->payload_dma_map,
403 			    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
404 			bus_dmamap_unload(qpair->dma_tag_payload,
405 			    tr->payload_dma_map);
406 		}
407 
408 		nvme_free_request(req);
409 		tr->req = NULL;
410 
411 		TAILQ_REMOVE(&qpair->outstanding_tr, tr, tailq);
412 		TAILQ_INSERT_HEAD(&qpair->free_tr, tr, tailq);
413 
414 		/*
415 		 * If the controller is in the middle of resetting, don't
416 		 *  try to submit queued requests here - let the reset logic
417 		 *  handle that instead.
418 		 */
419 		if (!STAILQ_EMPTY(&qpair->queued_req) &&
420 		    !qpair->ctrlr->is_resetting) {
421 			req = STAILQ_FIRST(&qpair->queued_req);
422 			STAILQ_REMOVE_HEAD(&qpair->queued_req, stailq);
423 			_nvme_qpair_submit_request(qpair, req);
424 		}
425 	}
426 
427 	mtx_unlock(&qpair->lock);
428 }
429 
430 static void
431 nvme_qpair_manual_complete_tracker(struct nvme_qpair *qpair,
432     struct nvme_tracker *tr, uint32_t sct, uint32_t sc, uint32_t dnr,
433     boolean_t print_on_error)
434 {
435 	struct nvme_completion	cpl;
436 
437 	memset(&cpl, 0, sizeof(cpl));
438 	cpl.sqid = qpair->id;
439 	cpl.cid = tr->cid;
440 	cpl.status |= (sct & NVME_STATUS_SCT_MASK) << NVME_STATUS_SCT_SHIFT;
441 	cpl.status |= (sc & NVME_STATUS_SC_MASK) << NVME_STATUS_SC_SHIFT;
442 	cpl.status |= (dnr & NVME_STATUS_DNR_MASK) << NVME_STATUS_DNR_SHIFT;
443 	nvme_qpair_complete_tracker(qpair, tr, &cpl, print_on_error);
444 }
445 
446 void
447 nvme_qpair_manual_complete_request(struct nvme_qpair *qpair,
448     struct nvme_request *req, uint32_t sct, uint32_t sc,
449     boolean_t print_on_error)
450 {
451 	struct nvme_completion	cpl;
452 	boolean_t		error;
453 
454 	memset(&cpl, 0, sizeof(cpl));
455 	cpl.sqid = qpair->id;
456 	cpl.status |= (sct & NVME_STATUS_SCT_MASK) << NVME_STATUS_SCT_SHIFT;
457 	cpl.status |= (sc & NVME_STATUS_SC_MASK) << NVME_STATUS_SC_SHIFT;
458 
459 	error = nvme_completion_is_error(&cpl);
460 
461 	if (error && print_on_error) {
462 		nvme_qpair_print_command(qpair, &req->cmd);
463 		nvme_qpair_print_completion(qpair, &cpl);
464 	}
465 
466 	if (req->cb_fn)
467 		req->cb_fn(req->cb_arg, &cpl);
468 
469 	nvme_free_request(req);
470 }
471 
472 bool
473 nvme_qpair_process_completions(struct nvme_qpair *qpair)
474 {
475 	struct nvme_tracker	*tr;
476 	struct nvme_completion	cpl;
477 	int done = 0;
478 
479 	qpair->num_intr_handler_calls++;
480 
481 	if (!qpair->is_enabled)
482 		/*
483 		 * qpair is not enabled, likely because a controller reset is
484 		 *  is in progress.  Ignore the interrupt - any I/O that was
485 		 *  associated with this interrupt will get retried when the
486 		 *  reset is complete.
487 		 */
488 		return (false);
489 
490 	bus_dmamap_sync(qpair->dma_tag, qpair->queuemem_map,
491 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
492 	while (1) {
493 		cpl = qpair->cpl[qpair->cq_head];
494 
495 		/* Convert to host endian */
496 		nvme_completion_swapbytes(&cpl);
497 
498 		if (NVME_STATUS_GET_P(cpl.status) != qpair->phase)
499 			break;
500 
501 		tr = qpair->act_tr[cpl.cid];
502 
503 		if (tr != NULL) {
504 			nvme_qpair_complete_tracker(qpair, tr, &cpl, TRUE);
505 			qpair->sq_head = cpl.sqhd;
506 			done++;
507 		} else {
508 			nvme_printf(qpair->ctrlr,
509 			    "cpl does not map to outstanding cmd\n");
510 			/* nvme_dump_completion expects device endianess */
511 			nvme_dump_completion(&qpair->cpl[qpair->cq_head]);
512 			KASSERT(0, ("received completion for unknown cmd\n"));
513 		}
514 
515 		if (++qpair->cq_head == qpair->num_entries) {
516 			qpair->cq_head = 0;
517 			qpair->phase = !qpair->phase;
518 		}
519 
520 		nvme_mmio_write_4(qpair->ctrlr, doorbell[qpair->id].cq_hdbl,
521 		    qpair->cq_head);
522 	}
523 	return (done != 0);
524 }
525 
526 static void
527 nvme_qpair_msix_handler(void *arg)
528 {
529 	struct nvme_qpair *qpair = arg;
530 
531 	nvme_qpair_process_completions(qpair);
532 }
533 
534 int
535 nvme_qpair_construct(struct nvme_qpair *qpair, uint32_t id,
536     uint16_t vector, uint32_t num_entries, uint32_t num_trackers,
537     struct nvme_controller *ctrlr)
538 {
539 	struct nvme_tracker	*tr;
540 	size_t			cmdsz, cplsz, prpsz, allocsz, prpmemsz;
541 	uint64_t		queuemem_phys, prpmem_phys, list_phys;
542 	uint8_t			*queuemem, *prpmem, *prp_list;
543 	int			i, err;
544 
545 	qpair->id = id;
546 	qpair->vector = vector;
547 	qpair->num_entries = num_entries;
548 	qpair->num_trackers = num_trackers;
549 	qpair->ctrlr = ctrlr;
550 
551 	if (ctrlr->msix_enabled) {
552 
553 		/*
554 		 * MSI-X vector resource IDs start at 1, so we add one to
555 		 *  the queue's vector to get the corresponding rid to use.
556 		 */
557 		qpair->rid = vector + 1;
558 
559 		qpair->res = bus_alloc_resource_any(ctrlr->dev, SYS_RES_IRQ,
560 		    &qpair->rid, RF_ACTIVE);
561 		bus_setup_intr(ctrlr->dev, qpair->res,
562 		    INTR_TYPE_MISC | INTR_MPSAFE, NULL,
563 		    nvme_qpair_msix_handler, qpair, &qpair->tag);
564 	}
565 
566 	mtx_init(&qpair->lock, "nvme qpair lock", NULL, MTX_DEF);
567 
568 	/* Note: NVMe PRP format is restricted to 4-byte alignment. */
569 	err = bus_dma_tag_create(bus_get_dma_tag(ctrlr->dev),
570 	    4, PAGE_SIZE, BUS_SPACE_MAXADDR,
571 	    BUS_SPACE_MAXADDR, NULL, NULL, NVME_MAX_XFER_SIZE,
572 	    (NVME_MAX_XFER_SIZE/PAGE_SIZE)+1, PAGE_SIZE, 0,
573 	    NULL, NULL, &qpair->dma_tag_payload);
574 	if (err != 0) {
575 		nvme_printf(ctrlr, "payload tag create failed %d\n", err);
576 		goto out;
577 	}
578 
579 	/*
580 	 * Each component must be page aligned, and individual PRP lists
581 	 * cannot cross a page boundary.
582 	 */
583 	cmdsz = qpair->num_entries * sizeof(struct nvme_command);
584 	cmdsz = roundup2(cmdsz, PAGE_SIZE);
585 	cplsz = qpair->num_entries * sizeof(struct nvme_completion);
586 	cplsz = roundup2(cplsz, PAGE_SIZE);
587 	prpsz = sizeof(uint64_t) * NVME_MAX_PRP_LIST_ENTRIES;;
588 	prpmemsz = qpair->num_trackers * prpsz;
589 	allocsz = cmdsz + cplsz + prpmemsz;
590 
591 	err = bus_dma_tag_create(bus_get_dma_tag(ctrlr->dev),
592 	    PAGE_SIZE, 0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
593 	    allocsz, 1, allocsz, 0, NULL, NULL, &qpair->dma_tag);
594 	if (err != 0) {
595 		nvme_printf(ctrlr, "tag create failed %d\n", err);
596 		goto out;
597 	}
598 
599 	if (bus_dmamem_alloc(qpair->dma_tag, (void **)&queuemem,
600 	    BUS_DMA_NOWAIT, &qpair->queuemem_map)) {
601 		nvme_printf(ctrlr, "failed to alloc qpair memory\n");
602 		goto out;
603 	}
604 
605 	if (bus_dmamap_load(qpair->dma_tag, qpair->queuemem_map,
606 	    queuemem, allocsz, nvme_single_map, &queuemem_phys, 0) != 0) {
607 		nvme_printf(ctrlr, "failed to load qpair memory\n");
608 		goto out;
609 	}
610 
611 	qpair->num_cmds = 0;
612 	qpair->num_intr_handler_calls = 0;
613 	qpair->cmd = (struct nvme_command *)queuemem;
614 	qpair->cpl = (struct nvme_completion *)(queuemem + cmdsz);
615 	prpmem = (uint8_t *)(queuemem + cmdsz + cplsz);
616 	qpair->cmd_bus_addr = queuemem_phys;
617 	qpair->cpl_bus_addr = queuemem_phys + cmdsz;
618 	prpmem_phys = queuemem_phys + cmdsz + cplsz;
619 
620 	qpair->sq_tdbl_off = nvme_mmio_offsetof(doorbell[id].sq_tdbl);
621 	qpair->cq_hdbl_off = nvme_mmio_offsetof(doorbell[id].cq_hdbl);
622 
623 	TAILQ_INIT(&qpair->free_tr);
624 	TAILQ_INIT(&qpair->outstanding_tr);
625 	STAILQ_INIT(&qpair->queued_req);
626 
627 	list_phys = prpmem_phys;
628 	prp_list = prpmem;
629 	for (i = 0; i < qpair->num_trackers; i++) {
630 
631 		if (list_phys + prpsz > prpmem_phys + prpmemsz) {
632 			qpair->num_trackers = i;
633 			break;
634 		}
635 
636 		/*
637 		 * Make sure that the PRP list for this tracker doesn't
638 		 * overflow to another page.
639 		 */
640 		if (trunc_page(list_phys) !=
641 		    trunc_page(list_phys + prpsz - 1)) {
642 			list_phys = roundup2(list_phys, PAGE_SIZE);
643 			prp_list =
644 			    (uint8_t *)roundup2((uintptr_t)prp_list, PAGE_SIZE);
645 		}
646 
647 		tr = malloc(sizeof(*tr), M_NVME, M_ZERO | M_WAITOK);
648 		bus_dmamap_create(qpair->dma_tag_payload, 0,
649 		    &tr->payload_dma_map);
650 		callout_init(&tr->timer, 1);
651 		tr->cid = i;
652 		tr->qpair = qpair;
653 		tr->prp = (uint64_t *)prp_list;
654 		tr->prp_bus_addr = list_phys;
655 		TAILQ_INSERT_HEAD(&qpair->free_tr, tr, tailq);
656 		list_phys += prpsz;
657 		prp_list += prpsz;
658 	}
659 
660 	if (qpair->num_trackers == 0) {
661 		nvme_printf(ctrlr, "failed to allocate enough trackers\n");
662 		goto out;
663 	}
664 
665 	qpair->act_tr = malloc(sizeof(struct nvme_tracker *) *
666 	    qpair->num_entries, M_NVME, M_ZERO | M_WAITOK);
667 	return (0);
668 
669 out:
670 	nvme_qpair_destroy(qpair);
671 	return (ENOMEM);
672 }
673 
674 static void
675 nvme_qpair_destroy(struct nvme_qpair *qpair)
676 {
677 	struct nvme_tracker	*tr;
678 
679 	if (qpair->tag)
680 		bus_teardown_intr(qpair->ctrlr->dev, qpair->res, qpair->tag);
681 
682 	if (mtx_initialized(&qpair->lock))
683 		mtx_destroy(&qpair->lock);
684 
685 	if (qpair->res)
686 		bus_release_resource(qpair->ctrlr->dev, SYS_RES_IRQ,
687 		    rman_get_rid(qpair->res), qpair->res);
688 
689 	if (qpair->cmd != NULL) {
690 		bus_dmamap_unload(qpair->dma_tag, qpair->queuemem_map);
691 		bus_dmamem_free(qpair->dma_tag, qpair->cmd,
692 		    qpair->queuemem_map);
693 	}
694 
695 	if (qpair->act_tr)
696 		free(qpair->act_tr, M_NVME);
697 
698 	while (!TAILQ_EMPTY(&qpair->free_tr)) {
699 		tr = TAILQ_FIRST(&qpair->free_tr);
700 		TAILQ_REMOVE(&qpair->free_tr, tr, tailq);
701 		bus_dmamap_destroy(qpair->dma_tag_payload,
702 		    tr->payload_dma_map);
703 		free(tr, M_NVME);
704 	}
705 
706 	if (qpair->dma_tag)
707 		bus_dma_tag_destroy(qpair->dma_tag);
708 
709 	if (qpair->dma_tag_payload)
710 		bus_dma_tag_destroy(qpair->dma_tag_payload);
711 }
712 
713 static void
714 nvme_admin_qpair_abort_aers(struct nvme_qpair *qpair)
715 {
716 	struct nvme_tracker	*tr;
717 
718 	tr = TAILQ_FIRST(&qpair->outstanding_tr);
719 	while (tr != NULL) {
720 		if (tr->req->cmd.opc == NVME_OPC_ASYNC_EVENT_REQUEST) {
721 			nvme_qpair_manual_complete_tracker(qpair, tr,
722 			    NVME_SCT_GENERIC, NVME_SC_ABORTED_SQ_DELETION, 0,
723 			    FALSE);
724 			tr = TAILQ_FIRST(&qpair->outstanding_tr);
725 		} else {
726 			tr = TAILQ_NEXT(tr, tailq);
727 		}
728 	}
729 }
730 
731 void
732 nvme_admin_qpair_destroy(struct nvme_qpair *qpair)
733 {
734 
735 	nvme_admin_qpair_abort_aers(qpair);
736 	nvme_qpair_destroy(qpair);
737 }
738 
739 void
740 nvme_io_qpair_destroy(struct nvme_qpair *qpair)
741 {
742 
743 	nvme_qpair_destroy(qpair);
744 }
745 
746 static void
747 nvme_abort_complete(void *arg, const struct nvme_completion *status)
748 {
749 	struct nvme_tracker	*tr = arg;
750 
751 	/*
752 	 * If cdw0 == 1, the controller was not able to abort the command
753 	 *  we requested.  We still need to check the active tracker array,
754 	 *  to cover race where I/O timed out at same time controller was
755 	 *  completing the I/O.
756 	 */
757 	if (status->cdw0 == 1 && tr->qpair->act_tr[tr->cid] != NULL) {
758 		/*
759 		 * An I/O has timed out, and the controller was unable to
760 		 *  abort it for some reason.  Construct a fake completion
761 		 *  status, and then complete the I/O's tracker manually.
762 		 */
763 		nvme_printf(tr->qpair->ctrlr,
764 		    "abort command failed, aborting command manually\n");
765 		nvme_qpair_manual_complete_tracker(tr->qpair, tr,
766 		    NVME_SCT_GENERIC, NVME_SC_ABORTED_BY_REQUEST, 0, TRUE);
767 	}
768 }
769 
770 static void
771 nvme_timeout(void *arg)
772 {
773 	struct nvme_tracker	*tr = arg;
774 	struct nvme_qpair	*qpair = tr->qpair;
775 	struct nvme_controller	*ctrlr = qpair->ctrlr;
776 	uint32_t		csts;
777 	uint8_t			cfs;
778 
779 	/*
780 	 * Read csts to get value of cfs - controller fatal status.
781 	 * If no fatal status, try to call the completion routine, and
782 	 * if completes transactions, report a missed interrupt and
783 	 * return (this may need to be rate limited). Otherwise, if
784 	 * aborts are enabled and the controller is not reporting
785 	 * fatal status, abort the command. Otherwise, just reset the
786 	 * controller and hope for the best.
787 	 */
788 	csts = nvme_mmio_read_4(ctrlr, csts);
789 	cfs = (csts >> NVME_CSTS_REG_CFS_SHIFT) & NVME_CSTS_REG_CFS_MASK;
790 	if (cfs == 0 && nvme_qpair_process_completions(qpair)) {
791 		nvme_printf(ctrlr, "Missing interrupt\n");
792 		return;
793 	}
794 	if (ctrlr->enable_aborts && cfs == 0) {
795 		nvme_printf(ctrlr, "Aborting command due to a timeout.\n");
796 		nvme_ctrlr_cmd_abort(ctrlr, tr->cid, qpair->id,
797 		    nvme_abort_complete, tr);
798 	} else {
799 		nvme_printf(ctrlr, "Resetting controller due to a timeout%s.\n",
800 		    cfs ? " and fatal error status" : "");
801 		nvme_ctrlr_reset(ctrlr);
802 	}
803 }
804 
805 void
806 nvme_qpair_submit_tracker(struct nvme_qpair *qpair, struct nvme_tracker *tr)
807 {
808 	struct nvme_request	*req;
809 	struct nvme_controller	*ctrlr;
810 
811 	mtx_assert(&qpair->lock, MA_OWNED);
812 
813 	req = tr->req;
814 	req->cmd.cid = tr->cid;
815 	qpair->act_tr[tr->cid] = tr;
816 	ctrlr = qpair->ctrlr;
817 
818 	if (req->timeout)
819 #if __FreeBSD_version >= 800030
820 		callout_reset_curcpu(&tr->timer, ctrlr->timeout_period * hz,
821 		    nvme_timeout, tr);
822 #else
823 		callout_reset(&tr->timer, ctrlr->timeout_period * hz,
824 		    nvme_timeout, tr);
825 #endif
826 
827 	/* Copy the command from the tracker to the submission queue. */
828 	memcpy(&qpair->cmd[qpair->sq_tail], &req->cmd, sizeof(req->cmd));
829 
830 	if (++qpair->sq_tail == qpair->num_entries)
831 		qpair->sq_tail = 0;
832 
833 	bus_dmamap_sync(qpair->dma_tag, qpair->queuemem_map,
834 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
835 #ifndef __powerpc__
836 	/*
837 	 * powerpc's bus_dmamap_sync() already includes a heavyweight sync, but
838 	 * no other archs do.
839 	 */
840 	wmb();
841 #endif
842 
843 	nvme_mmio_write_4(qpair->ctrlr, doorbell[qpair->id].sq_tdbl,
844 	    qpair->sq_tail);
845 
846 	qpair->num_cmds++;
847 }
848 
849 static void
850 nvme_payload_map(void *arg, bus_dma_segment_t *seg, int nseg, int error)
851 {
852 	struct nvme_tracker 	*tr = arg;
853 	uint32_t		cur_nseg;
854 
855 	/*
856 	 * If the mapping operation failed, return immediately.  The caller
857 	 *  is responsible for detecting the error status and failing the
858 	 *  tracker manually.
859 	 */
860 	if (error != 0) {
861 		nvme_printf(tr->qpair->ctrlr,
862 		    "nvme_payload_map err %d\n", error);
863 		return;
864 	}
865 
866 	/*
867 	 * Note that we specified PAGE_SIZE for alignment and max
868 	 *  segment size when creating the bus dma tags.  So here
869 	 *  we can safely just transfer each segment to its
870 	 *  associated PRP entry.
871 	 */
872 	tr->req->cmd.prp1 = htole64(seg[0].ds_addr);
873 
874 	if (nseg == 2) {
875 		tr->req->cmd.prp2 = htole64(seg[1].ds_addr);
876 	} else if (nseg > 2) {
877 		cur_nseg = 1;
878 		tr->req->cmd.prp2 = htole64((uint64_t)tr->prp_bus_addr);
879 		while (cur_nseg < nseg) {
880 			tr->prp[cur_nseg-1] =
881 			    htole64((uint64_t)seg[cur_nseg].ds_addr);
882 			cur_nseg++;
883 		}
884 	} else {
885 		/*
886 		 * prp2 should not be used by the controller
887 		 *  since there is only one segment, but set
888 		 *  to 0 just to be safe.
889 		 */
890 		tr->req->cmd.prp2 = 0;
891 	}
892 
893 	bus_dmamap_sync(tr->qpair->dma_tag_payload, tr->payload_dma_map,
894 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
895 	nvme_qpair_submit_tracker(tr->qpair, tr);
896 }
897 
898 static void
899 _nvme_qpair_submit_request(struct nvme_qpair *qpair, struct nvme_request *req)
900 {
901 	struct nvme_tracker	*tr;
902 	int			err = 0;
903 
904 	mtx_assert(&qpair->lock, MA_OWNED);
905 
906 	tr = TAILQ_FIRST(&qpair->free_tr);
907 	req->qpair = qpair;
908 
909 	if (tr == NULL || !qpair->is_enabled) {
910 		/*
911 		 * No tracker is available, or the qpair is disabled due to
912 		 *  an in-progress controller-level reset or controller
913 		 *  failure.
914 		 */
915 
916 		if (qpair->ctrlr->is_failed) {
917 			/*
918 			 * The controller has failed.  Post the request to a
919 			 *  task where it will be aborted, so that we do not
920 			 *  invoke the request's callback in the context
921 			 *  of the submission.
922 			 */
923 			nvme_ctrlr_post_failed_request(qpair->ctrlr, req);
924 		} else {
925 			/*
926 			 * Put the request on the qpair's request queue to be
927 			 *  processed when a tracker frees up via a command
928 			 *  completion or when the controller reset is
929 			 *  completed.
930 			 */
931 			STAILQ_INSERT_TAIL(&qpair->queued_req, req, stailq);
932 		}
933 		return;
934 	}
935 
936 	TAILQ_REMOVE(&qpair->free_tr, tr, tailq);
937 	TAILQ_INSERT_TAIL(&qpair->outstanding_tr, tr, tailq);
938 	tr->req = req;
939 
940 	switch (req->type) {
941 	case NVME_REQUEST_VADDR:
942 		KASSERT(req->payload_size <= qpair->ctrlr->max_xfer_size,
943 		    ("payload_size (%d) exceeds max_xfer_size (%d)\n",
944 		    req->payload_size, qpair->ctrlr->max_xfer_size));
945 		err = bus_dmamap_load(tr->qpair->dma_tag_payload,
946 		    tr->payload_dma_map, req->u.payload, req->payload_size,
947 		    nvme_payload_map, tr, 0);
948 		if (err != 0)
949 			nvme_printf(qpair->ctrlr,
950 			    "bus_dmamap_load returned 0x%x!\n", err);
951 		break;
952 	case NVME_REQUEST_NULL:
953 		nvme_qpair_submit_tracker(tr->qpair, tr);
954 		break;
955 #ifdef NVME_UNMAPPED_BIO_SUPPORT
956 	case NVME_REQUEST_BIO:
957 		KASSERT(req->u.bio->bio_bcount <= qpair->ctrlr->max_xfer_size,
958 		    ("bio->bio_bcount (%jd) exceeds max_xfer_size (%d)\n",
959 		    (intmax_t)req->u.bio->bio_bcount,
960 		    qpair->ctrlr->max_xfer_size));
961 		err = bus_dmamap_load_bio(tr->qpair->dma_tag_payload,
962 		    tr->payload_dma_map, req->u.bio, nvme_payload_map, tr, 0);
963 		if (err != 0)
964 			nvme_printf(qpair->ctrlr,
965 			    "bus_dmamap_load_bio returned 0x%x!\n", err);
966 		break;
967 #endif
968 	case NVME_REQUEST_CCB:
969 		err = bus_dmamap_load_ccb(tr->qpair->dma_tag_payload,
970 		    tr->payload_dma_map, req->u.payload,
971 		    nvme_payload_map, tr, 0);
972 		if (err != 0)
973 			nvme_printf(qpair->ctrlr,
974 			    "bus_dmamap_load_ccb returned 0x%x!\n", err);
975 		break;
976 	default:
977 		panic("unknown nvme request type 0x%x\n", req->type);
978 		break;
979 	}
980 
981 	if (err != 0) {
982 		/*
983 		 * The dmamap operation failed, so we manually fail the
984 		 *  tracker here with DATA_TRANSFER_ERROR status.
985 		 *
986 		 * nvme_qpair_manual_complete_tracker must not be called
987 		 *  with the qpair lock held.
988 		 */
989 		mtx_unlock(&qpair->lock);
990 		nvme_qpair_manual_complete_tracker(qpair, tr, NVME_SCT_GENERIC,
991 		    NVME_SC_DATA_TRANSFER_ERROR, 1 /* do not retry */, TRUE);
992 		mtx_lock(&qpair->lock);
993 	}
994 }
995 
996 void
997 nvme_qpair_submit_request(struct nvme_qpair *qpair, struct nvme_request *req)
998 {
999 
1000 	mtx_lock(&qpair->lock);
1001 	_nvme_qpair_submit_request(qpair, req);
1002 	mtx_unlock(&qpair->lock);
1003 }
1004 
1005 static void
1006 nvme_qpair_enable(struct nvme_qpair *qpair)
1007 {
1008 
1009 	qpair->is_enabled = TRUE;
1010 }
1011 
1012 void
1013 nvme_qpair_reset(struct nvme_qpair *qpair)
1014 {
1015 
1016 	qpair->sq_head = qpair->sq_tail = qpair->cq_head = 0;
1017 
1018 	/*
1019 	 * First time through the completion queue, HW will set phase
1020 	 *  bit on completions to 1.  So set this to 1 here, indicating
1021 	 *  we're looking for a 1 to know which entries have completed.
1022 	 *  we'll toggle the bit each time when the completion queue
1023 	 *  rolls over.
1024 	 */
1025 	qpair->phase = 1;
1026 
1027 	memset(qpair->cmd, 0,
1028 	    qpair->num_entries * sizeof(struct nvme_command));
1029 	memset(qpair->cpl, 0,
1030 	    qpair->num_entries * sizeof(struct nvme_completion));
1031 }
1032 
1033 void
1034 nvme_admin_qpair_enable(struct nvme_qpair *qpair)
1035 {
1036 	struct nvme_tracker		*tr;
1037 	struct nvme_tracker		*tr_temp;
1038 
1039 	/*
1040 	 * Manually abort each outstanding admin command.  Do not retry
1041 	 *  admin commands found here, since they will be left over from
1042 	 *  a controller reset and its likely the context in which the
1043 	 *  command was issued no longer applies.
1044 	 */
1045 	TAILQ_FOREACH_SAFE(tr, &qpair->outstanding_tr, tailq, tr_temp) {
1046 		nvme_printf(qpair->ctrlr,
1047 		    "aborting outstanding admin command\n");
1048 		nvme_qpair_manual_complete_tracker(qpair, tr, NVME_SCT_GENERIC,
1049 		    NVME_SC_ABORTED_BY_REQUEST, 1 /* do not retry */, TRUE);
1050 	}
1051 
1052 	nvme_qpair_enable(qpair);
1053 }
1054 
1055 void
1056 nvme_io_qpair_enable(struct nvme_qpair *qpair)
1057 {
1058 	STAILQ_HEAD(, nvme_request)	temp;
1059 	struct nvme_tracker		*tr;
1060 	struct nvme_tracker		*tr_temp;
1061 	struct nvme_request		*req;
1062 
1063 	/*
1064 	 * Manually abort each outstanding I/O.  This normally results in a
1065 	 *  retry, unless the retry count on the associated request has
1066 	 *  reached its limit.
1067 	 */
1068 	TAILQ_FOREACH_SAFE(tr, &qpair->outstanding_tr, tailq, tr_temp) {
1069 		nvme_printf(qpair->ctrlr, "aborting outstanding i/o\n");
1070 		nvme_qpair_manual_complete_tracker(qpair, tr, NVME_SCT_GENERIC,
1071 		    NVME_SC_ABORTED_BY_REQUEST, 0, TRUE);
1072 	}
1073 
1074 	mtx_lock(&qpair->lock);
1075 
1076 	nvme_qpair_enable(qpair);
1077 
1078 	STAILQ_INIT(&temp);
1079 	STAILQ_SWAP(&qpair->queued_req, &temp, nvme_request);
1080 
1081 	while (!STAILQ_EMPTY(&temp)) {
1082 		req = STAILQ_FIRST(&temp);
1083 		STAILQ_REMOVE_HEAD(&temp, stailq);
1084 		nvme_printf(qpair->ctrlr, "resubmitting queued i/o\n");
1085 		nvme_qpair_print_command(qpair, &req->cmd);
1086 		_nvme_qpair_submit_request(qpair, req);
1087 	}
1088 
1089 	mtx_unlock(&qpair->lock);
1090 }
1091 
1092 static void
1093 nvme_qpair_disable(struct nvme_qpair *qpair)
1094 {
1095 	struct nvme_tracker *tr;
1096 
1097 	qpair->is_enabled = FALSE;
1098 	mtx_lock(&qpair->lock);
1099 	TAILQ_FOREACH(tr, &qpair->outstanding_tr, tailq)
1100 		callout_stop(&tr->timer);
1101 	mtx_unlock(&qpair->lock);
1102 }
1103 
1104 void
1105 nvme_admin_qpair_disable(struct nvme_qpair *qpair)
1106 {
1107 
1108 	nvme_qpair_disable(qpair);
1109 	nvme_admin_qpair_abort_aers(qpair);
1110 }
1111 
1112 void
1113 nvme_io_qpair_disable(struct nvme_qpair *qpair)
1114 {
1115 
1116 	nvme_qpair_disable(qpair);
1117 }
1118 
1119 void
1120 nvme_qpair_fail(struct nvme_qpair *qpair)
1121 {
1122 	struct nvme_tracker		*tr;
1123 	struct nvme_request		*req;
1124 
1125 	if (!mtx_initialized(&qpair->lock))
1126 		return;
1127 
1128 	mtx_lock(&qpair->lock);
1129 
1130 	while (!STAILQ_EMPTY(&qpair->queued_req)) {
1131 		req = STAILQ_FIRST(&qpair->queued_req);
1132 		STAILQ_REMOVE_HEAD(&qpair->queued_req, stailq);
1133 		nvme_printf(qpair->ctrlr, "failing queued i/o\n");
1134 		mtx_unlock(&qpair->lock);
1135 		nvme_qpair_manual_complete_request(qpair, req, NVME_SCT_GENERIC,
1136 		    NVME_SC_ABORTED_BY_REQUEST, TRUE);
1137 		mtx_lock(&qpair->lock);
1138 	}
1139 
1140 	/* Manually abort each outstanding I/O. */
1141 	while (!TAILQ_EMPTY(&qpair->outstanding_tr)) {
1142 		tr = TAILQ_FIRST(&qpair->outstanding_tr);
1143 		/*
1144 		 * Do not remove the tracker.  The abort_tracker path will
1145 		 *  do that for us.
1146 		 */
1147 		nvme_printf(qpair->ctrlr, "failing outstanding i/o\n");
1148 		mtx_unlock(&qpair->lock);
1149 		nvme_qpair_manual_complete_tracker(qpair, tr, NVME_SCT_GENERIC,
1150 		    NVME_SC_ABORTED_BY_REQUEST, 1 /* do not retry */, TRUE);
1151 		mtx_lock(&qpair->lock);
1152 	}
1153 
1154 	mtx_unlock(&qpair->lock);
1155 }
1156 
1157