2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (C) 2012-2014 Intel Corporation
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
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.
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
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
32 #include <sys/param.h>
35 #include <sys/domainset.h>
38 #include <dev/pci/pcivar.h>
40 #include "nvme_private.h"
42 typedef enum error_print { ERROR_PRINT_NONE, ERROR_PRINT_NO_RETRY, ERROR_PRINT_ALL } error_print_t;
43 #define DO_NOT_RETRY 1
45 static void _nvme_qpair_submit_request(struct nvme_qpair *qpair,
46 struct nvme_request *req);
47 static void nvme_qpair_destroy(struct nvme_qpair *qpair);
49 struct nvme_opcode_string {
54 static struct nvme_opcode_string admin_opcode[] = {
55 { NVME_OPC_DELETE_IO_SQ, "DELETE IO SQ" },
56 { NVME_OPC_CREATE_IO_SQ, "CREATE IO SQ" },
57 { NVME_OPC_GET_LOG_PAGE, "GET LOG PAGE" },
58 { NVME_OPC_DELETE_IO_CQ, "DELETE IO CQ" },
59 { NVME_OPC_CREATE_IO_CQ, "CREATE IO CQ" },
60 { NVME_OPC_IDENTIFY, "IDENTIFY" },
61 { NVME_OPC_ABORT, "ABORT" },
62 { NVME_OPC_SET_FEATURES, "SET FEATURES" },
63 { NVME_OPC_GET_FEATURES, "GET FEATURES" },
64 { NVME_OPC_ASYNC_EVENT_REQUEST, "ASYNC EVENT REQUEST" },
65 { NVME_OPC_FIRMWARE_ACTIVATE, "FIRMWARE ACTIVATE" },
66 { NVME_OPC_FIRMWARE_IMAGE_DOWNLOAD, "FIRMWARE IMAGE DOWNLOAD" },
67 { NVME_OPC_DEVICE_SELF_TEST, "DEVICE SELF-TEST" },
68 { NVME_OPC_NAMESPACE_ATTACHMENT, "NAMESPACE ATTACHMENT" },
69 { NVME_OPC_KEEP_ALIVE, "KEEP ALIVE" },
70 { NVME_OPC_DIRECTIVE_SEND, "DIRECTIVE SEND" },
71 { NVME_OPC_DIRECTIVE_RECEIVE, "DIRECTIVE RECEIVE" },
72 { NVME_OPC_VIRTUALIZATION_MANAGEMENT, "VIRTUALIZATION MANAGEMENT" },
73 { NVME_OPC_NVME_MI_SEND, "NVME-MI SEND" },
74 { NVME_OPC_NVME_MI_RECEIVE, "NVME-MI RECEIVE" },
75 { NVME_OPC_DOORBELL_BUFFER_CONFIG, "DOORBELL BUFFER CONFIG" },
76 { NVME_OPC_FORMAT_NVM, "FORMAT NVM" },
77 { NVME_OPC_SECURITY_SEND, "SECURITY SEND" },
78 { NVME_OPC_SECURITY_RECEIVE, "SECURITY RECEIVE" },
79 { NVME_OPC_SANITIZE, "SANITIZE" },
80 { NVME_OPC_GET_LBA_STATUS, "GET LBA STATUS" },
81 { 0xFFFF, "ADMIN COMMAND" }
84 static struct nvme_opcode_string io_opcode[] = {
85 { NVME_OPC_FLUSH, "FLUSH" },
86 { NVME_OPC_WRITE, "WRITE" },
87 { NVME_OPC_READ, "READ" },
88 { NVME_OPC_WRITE_UNCORRECTABLE, "WRITE UNCORRECTABLE" },
89 { NVME_OPC_COMPARE, "COMPARE" },
90 { NVME_OPC_WRITE_ZEROES, "WRITE ZEROES" },
91 { NVME_OPC_DATASET_MANAGEMENT, "DATASET MANAGEMENT" },
92 { NVME_OPC_VERIFY, "VERIFY" },
93 { NVME_OPC_RESERVATION_REGISTER, "RESERVATION REGISTER" },
94 { NVME_OPC_RESERVATION_REPORT, "RESERVATION REPORT" },
95 { NVME_OPC_RESERVATION_ACQUIRE, "RESERVATION ACQUIRE" },
96 { NVME_OPC_RESERVATION_RELEASE, "RESERVATION RELEASE" },
97 { 0xFFFF, "IO COMMAND" }
101 get_admin_opcode_string(uint16_t opc)
103 struct nvme_opcode_string *entry;
105 entry = admin_opcode;
107 while (entry->opc != 0xFFFF) {
108 if (entry->opc == opc)
116 get_io_opcode_string(uint16_t opc)
118 struct nvme_opcode_string *entry;
122 while (entry->opc != 0xFFFF) {
123 if (entry->opc == opc)
131 nvme_admin_qpair_print_command(struct nvme_qpair *qpair,
132 struct nvme_command *cmd)
135 nvme_printf(qpair->ctrlr, "%s (%02x) sqid:%d cid:%d nsid:%x "
136 "cdw10:%08x cdw11:%08x\n",
137 get_admin_opcode_string(cmd->opc), cmd->opc, qpair->id, cmd->cid,
138 le32toh(cmd->nsid), le32toh(cmd->cdw10), le32toh(cmd->cdw11));
142 nvme_io_qpair_print_command(struct nvme_qpair *qpair,
143 struct nvme_command *cmd)
149 case NVME_OPC_WRITE_UNCORRECTABLE:
150 case NVME_OPC_COMPARE:
151 case NVME_OPC_WRITE_ZEROES:
152 case NVME_OPC_VERIFY:
153 nvme_printf(qpair->ctrlr, "%s sqid:%d cid:%d nsid:%d "
155 get_io_opcode_string(cmd->opc), qpair->id, cmd->cid, le32toh(cmd->nsid),
156 ((unsigned long long)le32toh(cmd->cdw11) << 32) + le32toh(cmd->cdw10),
157 (le32toh(cmd->cdw12) & 0xFFFF) + 1);
160 case NVME_OPC_DATASET_MANAGEMENT:
161 case NVME_OPC_RESERVATION_REGISTER:
162 case NVME_OPC_RESERVATION_REPORT:
163 case NVME_OPC_RESERVATION_ACQUIRE:
164 case NVME_OPC_RESERVATION_RELEASE:
165 nvme_printf(qpair->ctrlr, "%s sqid:%d cid:%d nsid:%d\n",
166 get_io_opcode_string(cmd->opc), qpair->id, cmd->cid, le32toh(cmd->nsid));
169 nvme_printf(qpair->ctrlr, "%s (%02x) sqid:%d cid:%d nsid:%d\n",
170 get_io_opcode_string(cmd->opc), cmd->opc, qpair->id,
171 cmd->cid, le32toh(cmd->nsid));
177 nvme_qpair_print_command(struct nvme_qpair *qpair, struct nvme_command *cmd)
180 nvme_admin_qpair_print_command(qpair, cmd);
182 nvme_io_qpair_print_command(qpair, cmd);
183 if (nvme_verbose_cmd_dump) {
184 nvme_printf(qpair->ctrlr,
185 "nsid:%#x rsvd2:%#x rsvd3:%#x mptr:%#jx prp1:%#jx prp2:%#jx\n",
186 cmd->nsid, cmd->rsvd2, cmd->rsvd3, (uintmax_t)cmd->mptr,
187 (uintmax_t)cmd->prp1, (uintmax_t)cmd->prp2);
188 nvme_printf(qpair->ctrlr,
189 "cdw10: %#x cdw11:%#x cdw12:%#x cdw13:%#x cdw14:%#x cdw15:%#x\n",
190 cmd->cdw10, cmd->cdw11, cmd->cdw12, cmd->cdw13, cmd->cdw14,
195 struct nvme_status_string {
200 static struct nvme_status_string generic_status[] = {
201 { NVME_SC_SUCCESS, "SUCCESS" },
202 { NVME_SC_INVALID_OPCODE, "INVALID OPCODE" },
203 { NVME_SC_INVALID_FIELD, "INVALID_FIELD" },
204 { NVME_SC_COMMAND_ID_CONFLICT, "COMMAND ID CONFLICT" },
205 { NVME_SC_DATA_TRANSFER_ERROR, "DATA TRANSFER ERROR" },
206 { NVME_SC_ABORTED_POWER_LOSS, "ABORTED - POWER LOSS" },
207 { NVME_SC_INTERNAL_DEVICE_ERROR, "INTERNAL DEVICE ERROR" },
208 { NVME_SC_ABORTED_BY_REQUEST, "ABORTED - BY REQUEST" },
209 { NVME_SC_ABORTED_SQ_DELETION, "ABORTED - SQ DELETION" },
210 { NVME_SC_ABORTED_FAILED_FUSED, "ABORTED - FAILED FUSED" },
211 { NVME_SC_ABORTED_MISSING_FUSED, "ABORTED - MISSING FUSED" },
212 { NVME_SC_INVALID_NAMESPACE_OR_FORMAT, "INVALID NAMESPACE OR FORMAT" },
213 { NVME_SC_COMMAND_SEQUENCE_ERROR, "COMMAND SEQUENCE ERROR" },
214 { NVME_SC_INVALID_SGL_SEGMENT_DESCR, "INVALID SGL SEGMENT DESCRIPTOR" },
215 { NVME_SC_INVALID_NUMBER_OF_SGL_DESCR, "INVALID NUMBER OF SGL DESCRIPTORS" },
216 { NVME_SC_DATA_SGL_LENGTH_INVALID, "DATA SGL LENGTH INVALID" },
217 { NVME_SC_METADATA_SGL_LENGTH_INVALID, "METADATA SGL LENGTH INVALID" },
218 { NVME_SC_SGL_DESCRIPTOR_TYPE_INVALID, "SGL DESCRIPTOR TYPE INVALID" },
219 { NVME_SC_INVALID_USE_OF_CMB, "INVALID USE OF CONTROLLER MEMORY BUFFER" },
220 { NVME_SC_PRP_OFFET_INVALID, "PRP OFFET INVALID" },
221 { NVME_SC_ATOMIC_WRITE_UNIT_EXCEEDED, "ATOMIC WRITE UNIT EXCEEDED" },
222 { NVME_SC_OPERATION_DENIED, "OPERATION DENIED" },
223 { NVME_SC_SGL_OFFSET_INVALID, "SGL OFFSET INVALID" },
224 { NVME_SC_HOST_ID_INCONSISTENT_FORMAT, "HOST IDENTIFIER INCONSISTENT FORMAT" },
225 { NVME_SC_KEEP_ALIVE_TIMEOUT_EXPIRED, "KEEP ALIVE TIMEOUT EXPIRED" },
226 { NVME_SC_KEEP_ALIVE_TIMEOUT_INVALID, "KEEP ALIVE TIMEOUT INVALID" },
227 { NVME_SC_ABORTED_DUE_TO_PREEMPT, "COMMAND ABORTED DUE TO PREEMPT AND ABORT" },
228 { NVME_SC_SANITIZE_FAILED, "SANITIZE FAILED" },
229 { NVME_SC_SANITIZE_IN_PROGRESS, "SANITIZE IN PROGRESS" },
230 { NVME_SC_SGL_DATA_BLOCK_GRAN_INVALID, "SGL_DATA_BLOCK_GRANULARITY_INVALID" },
231 { NVME_SC_NOT_SUPPORTED_IN_CMB, "COMMAND NOT SUPPORTED FOR QUEUE IN CMB" },
232 { NVME_SC_NAMESPACE_IS_WRITE_PROTECTED, "NAMESPACE IS WRITE PROTECTED" },
233 { NVME_SC_COMMAND_INTERRUPTED, "COMMAND INTERRUPTED" },
234 { NVME_SC_TRANSIENT_TRANSPORT_ERROR, "TRANSIENT TRANSPORT ERROR" },
236 { NVME_SC_LBA_OUT_OF_RANGE, "LBA OUT OF RANGE" },
237 { NVME_SC_CAPACITY_EXCEEDED, "CAPACITY EXCEEDED" },
238 { NVME_SC_NAMESPACE_NOT_READY, "NAMESPACE NOT READY" },
239 { NVME_SC_RESERVATION_CONFLICT, "RESERVATION CONFLICT" },
240 { NVME_SC_FORMAT_IN_PROGRESS, "FORMAT IN PROGRESS" },
241 { 0xFFFF, "GENERIC" }
244 static struct nvme_status_string command_specific_status[] = {
245 { NVME_SC_COMPLETION_QUEUE_INVALID, "INVALID COMPLETION QUEUE" },
246 { NVME_SC_INVALID_QUEUE_IDENTIFIER, "INVALID QUEUE IDENTIFIER" },
247 { NVME_SC_MAXIMUM_QUEUE_SIZE_EXCEEDED, "MAX QUEUE SIZE EXCEEDED" },
248 { NVME_SC_ABORT_COMMAND_LIMIT_EXCEEDED, "ABORT CMD LIMIT EXCEEDED" },
249 { NVME_SC_ASYNC_EVENT_REQUEST_LIMIT_EXCEEDED, "ASYNC LIMIT EXCEEDED" },
250 { NVME_SC_INVALID_FIRMWARE_SLOT, "INVALID FIRMWARE SLOT" },
251 { NVME_SC_INVALID_FIRMWARE_IMAGE, "INVALID FIRMWARE IMAGE" },
252 { NVME_SC_INVALID_INTERRUPT_VECTOR, "INVALID INTERRUPT VECTOR" },
253 { NVME_SC_INVALID_LOG_PAGE, "INVALID LOG PAGE" },
254 { NVME_SC_INVALID_FORMAT, "INVALID FORMAT" },
255 { NVME_SC_FIRMWARE_REQUIRES_RESET, "FIRMWARE REQUIRES RESET" },
256 { NVME_SC_INVALID_QUEUE_DELETION, "INVALID QUEUE DELETION" },
257 { NVME_SC_FEATURE_NOT_SAVEABLE, "FEATURE IDENTIFIER NOT SAVEABLE" },
258 { NVME_SC_FEATURE_NOT_CHANGEABLE, "FEATURE NOT CHANGEABLE" },
259 { NVME_SC_FEATURE_NOT_NS_SPECIFIC, "FEATURE NOT NAMESPACE SPECIFIC" },
260 { NVME_SC_FW_ACT_REQUIRES_NVMS_RESET, "FIRMWARE ACTIVATION REQUIRES NVM SUBSYSTEM RESET" },
261 { NVME_SC_FW_ACT_REQUIRES_RESET, "FIRMWARE ACTIVATION REQUIRES RESET" },
262 { NVME_SC_FW_ACT_REQUIRES_TIME, "FIRMWARE ACTIVATION REQUIRES MAXIMUM TIME VIOLATION" },
263 { NVME_SC_FW_ACT_PROHIBITED, "FIRMWARE ACTIVATION PROHIBITED" },
264 { NVME_SC_OVERLAPPING_RANGE, "OVERLAPPING RANGE" },
265 { NVME_SC_NS_INSUFFICIENT_CAPACITY, "NAMESPACE INSUFFICIENT CAPACITY" },
266 { NVME_SC_NS_ID_UNAVAILABLE, "NAMESPACE IDENTIFIER UNAVAILABLE" },
267 { NVME_SC_NS_ALREADY_ATTACHED, "NAMESPACE ALREADY ATTACHED" },
268 { NVME_SC_NS_IS_PRIVATE, "NAMESPACE IS PRIVATE" },
269 { NVME_SC_NS_NOT_ATTACHED, "NS NOT ATTACHED" },
270 { NVME_SC_THIN_PROV_NOT_SUPPORTED, "THIN PROVISIONING NOT SUPPORTED" },
271 { NVME_SC_CTRLR_LIST_INVALID, "CONTROLLER LIST INVALID" },
272 { NVME_SC_SELF_TEST_IN_PROGRESS, "DEVICE SELF-TEST IN PROGRESS" },
273 { NVME_SC_BOOT_PART_WRITE_PROHIB, "BOOT PARTITION WRITE PROHIBITED" },
274 { NVME_SC_INVALID_CTRLR_ID, "INVALID CONTROLLER IDENTIFIER" },
275 { NVME_SC_INVALID_SEC_CTRLR_STATE, "INVALID SECONDARY CONTROLLER STATE" },
276 { NVME_SC_INVALID_NUM_OF_CTRLR_RESRC, "INVALID NUMBER OF CONTROLLER RESOURCES" },
277 { NVME_SC_INVALID_RESOURCE_ID, "INVALID RESOURCE IDENTIFIER" },
278 { NVME_SC_SANITIZE_PROHIBITED_WPMRE, "SANITIZE PROHIBITED WRITE PERSISTENT MEMORY REGION ENABLED" },
279 { NVME_SC_ANA_GROUP_ID_INVALID, "ANA GROUP IDENTIFIED INVALID" },
280 { NVME_SC_ANA_ATTACH_FAILED, "ANA ATTACH FAILED" },
282 { NVME_SC_CONFLICTING_ATTRIBUTES, "CONFLICTING ATTRIBUTES" },
283 { NVME_SC_INVALID_PROTECTION_INFO, "INVALID PROTECTION INFO" },
284 { NVME_SC_ATTEMPTED_WRITE_TO_RO_PAGE, "WRITE TO RO PAGE" },
285 { 0xFFFF, "COMMAND SPECIFIC" }
288 static struct nvme_status_string media_error_status[] = {
289 { NVME_SC_WRITE_FAULTS, "WRITE FAULTS" },
290 { NVME_SC_UNRECOVERED_READ_ERROR, "UNRECOVERED READ ERROR" },
291 { NVME_SC_GUARD_CHECK_ERROR, "GUARD CHECK ERROR" },
292 { NVME_SC_APPLICATION_TAG_CHECK_ERROR, "APPLICATION TAG CHECK ERROR" },
293 { NVME_SC_REFERENCE_TAG_CHECK_ERROR, "REFERENCE TAG CHECK ERROR" },
294 { NVME_SC_COMPARE_FAILURE, "COMPARE FAILURE" },
295 { NVME_SC_ACCESS_DENIED, "ACCESS DENIED" },
296 { NVME_SC_DEALLOCATED_OR_UNWRITTEN, "DEALLOCATED OR UNWRITTEN LOGICAL BLOCK" },
297 { 0xFFFF, "MEDIA ERROR" }
300 static struct nvme_status_string path_related_status[] = {
301 { NVME_SC_INTERNAL_PATH_ERROR, "INTERNAL PATH ERROR" },
302 { NVME_SC_ASYMMETRIC_ACCESS_PERSISTENT_LOSS, "ASYMMETRIC ACCESS PERSISTENT LOSS" },
303 { NVME_SC_ASYMMETRIC_ACCESS_INACCESSIBLE, "ASYMMETRIC ACCESS INACCESSIBLE" },
304 { NVME_SC_ASYMMETRIC_ACCESS_TRANSITION, "ASYMMETRIC ACCESS TRANSITION" },
305 { NVME_SC_CONTROLLER_PATHING_ERROR, "CONTROLLER PATHING ERROR" },
306 { NVME_SC_HOST_PATHING_ERROR, "HOST PATHING ERROR" },
307 { NVME_SC_COMMAND_ABOTHED_BY_HOST, "COMMAND ABOTHED BY HOST" },
308 { 0xFFFF, "PATH RELATED" },
312 get_status_string(uint16_t sct, uint16_t sc)
314 struct nvme_status_string *entry;
317 case NVME_SCT_GENERIC:
318 entry = generic_status;
320 case NVME_SCT_COMMAND_SPECIFIC:
321 entry = command_specific_status;
323 case NVME_SCT_MEDIA_ERROR:
324 entry = media_error_status;
326 case NVME_SCT_PATH_RELATED:
327 entry = path_related_status;
329 case NVME_SCT_VENDOR_SPECIFIC:
330 return ("VENDOR SPECIFIC");
335 while (entry->sc != 0xFFFF) {
344 nvme_qpair_print_completion(struct nvme_qpair *qpair,
345 struct nvme_completion *cpl)
349 sct = NVME_STATUS_GET_SCT(cpl->status);
350 sc = NVME_STATUS_GET_SC(cpl->status);
352 nvme_printf(qpair->ctrlr, "%s (%02x/%02x) sqid:%d cid:%d cdw0:%x\n",
353 get_status_string(sct, sc), sct, sc, cpl->sqid, cpl->cid,
358 nvme_completion_is_retry(const struct nvme_completion *cpl)
360 uint8_t sct, sc, dnr;
362 sct = NVME_STATUS_GET_SCT(cpl->status);
363 sc = NVME_STATUS_GET_SC(cpl->status);
364 dnr = NVME_STATUS_GET_DNR(cpl->status); /* Do Not Retry Bit */
367 * TODO: spec is not clear how commands that are aborted due
368 * to TLER will be marked. So for now, it seems
369 * NAMESPACE_NOT_READY is the only case where we should
370 * look at the DNR bit. Requests failed with ABORTED_BY_REQUEST
371 * set the DNR bit correctly since the driver controls that.
374 case NVME_SCT_GENERIC:
376 case NVME_SC_ABORTED_BY_REQUEST:
377 case NVME_SC_NAMESPACE_NOT_READY:
382 case NVME_SC_INVALID_OPCODE:
383 case NVME_SC_INVALID_FIELD:
384 case NVME_SC_COMMAND_ID_CONFLICT:
385 case NVME_SC_DATA_TRANSFER_ERROR:
386 case NVME_SC_ABORTED_POWER_LOSS:
387 case NVME_SC_INTERNAL_DEVICE_ERROR:
388 case NVME_SC_ABORTED_SQ_DELETION:
389 case NVME_SC_ABORTED_FAILED_FUSED:
390 case NVME_SC_ABORTED_MISSING_FUSED:
391 case NVME_SC_INVALID_NAMESPACE_OR_FORMAT:
392 case NVME_SC_COMMAND_SEQUENCE_ERROR:
393 case NVME_SC_LBA_OUT_OF_RANGE:
394 case NVME_SC_CAPACITY_EXCEEDED:
398 case NVME_SCT_COMMAND_SPECIFIC:
399 case NVME_SCT_MEDIA_ERROR:
401 case NVME_SCT_PATH_RELATED:
403 case NVME_SC_INTERNAL_PATH_ERROR:
411 case NVME_SCT_VENDOR_SPECIFIC:
418 nvme_qpair_complete_tracker(struct nvme_tracker *tr,
419 struct nvme_completion *cpl, error_print_t print_on_error)
421 struct nvme_qpair * qpair = tr->qpair;
422 struct nvme_request *req;
423 bool retry, error, retriable;
426 error = nvme_completion_is_error(cpl);
427 retriable = nvme_completion_is_retry(cpl);
428 retry = error && retriable && req->retries < nvme_retry_count;
430 qpair->num_retries++;
431 if (error && req->retries >= nvme_retry_count && retriable)
432 qpair->num_failures++;
434 if (error && (print_on_error == ERROR_PRINT_ALL ||
435 (!retry && print_on_error == ERROR_PRINT_NO_RETRY))) {
436 nvme_qpair_print_command(qpair, &req->cmd);
437 nvme_qpair_print_completion(qpair, cpl);
440 qpair->act_tr[cpl->cid] = NULL;
442 KASSERT(cpl->cid == req->cmd.cid, ("cpl cid does not match cmd cid\n"));
445 if (req->type != NVME_REQUEST_NULL) {
446 bus_dmamap_sync(qpair->dma_tag_payload,
448 BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
451 req->cb_fn(req->cb_arg, cpl);
454 mtx_lock(&qpair->lock);
455 callout_stop(&tr->timer);
459 nvme_qpair_submit_tracker(qpair, tr);
461 if (req->type != NVME_REQUEST_NULL) {
462 bus_dmamap_unload(qpair->dma_tag_payload,
463 tr->payload_dma_map);
466 nvme_free_request(req);
469 TAILQ_REMOVE(&qpair->outstanding_tr, tr, tailq);
470 TAILQ_INSERT_HEAD(&qpair->free_tr, tr, tailq);
473 * If the controller is in the middle of resetting, don't
474 * try to submit queued requests here - let the reset logic
475 * handle that instead.
477 if (!STAILQ_EMPTY(&qpair->queued_req) &&
478 !qpair->ctrlr->is_resetting) {
479 req = STAILQ_FIRST(&qpair->queued_req);
480 STAILQ_REMOVE_HEAD(&qpair->queued_req, stailq);
481 _nvme_qpair_submit_request(qpair, req);
485 mtx_unlock(&qpair->lock);
489 nvme_qpair_manual_complete_tracker(
490 struct nvme_tracker *tr, uint32_t sct, uint32_t sc, uint32_t dnr,
491 error_print_t print_on_error)
493 struct nvme_completion cpl;
495 memset(&cpl, 0, sizeof(cpl));
497 struct nvme_qpair * qpair = tr->qpair;
499 cpl.sqid = qpair->id;
501 cpl.status |= (sct & NVME_STATUS_SCT_MASK) << NVME_STATUS_SCT_SHIFT;
502 cpl.status |= (sc & NVME_STATUS_SC_MASK) << NVME_STATUS_SC_SHIFT;
503 cpl.status |= (dnr & NVME_STATUS_DNR_MASK) << NVME_STATUS_DNR_SHIFT;
504 nvme_qpair_complete_tracker(tr, &cpl, print_on_error);
508 nvme_qpair_manual_complete_request(struct nvme_qpair *qpair,
509 struct nvme_request *req, uint32_t sct, uint32_t sc)
511 struct nvme_completion cpl;
514 memset(&cpl, 0, sizeof(cpl));
515 cpl.sqid = qpair->id;
516 cpl.status |= (sct & NVME_STATUS_SCT_MASK) << NVME_STATUS_SCT_SHIFT;
517 cpl.status |= (sc & NVME_STATUS_SC_MASK) << NVME_STATUS_SC_SHIFT;
519 error = nvme_completion_is_error(&cpl);
522 nvme_qpair_print_command(qpair, &req->cmd);
523 nvme_qpair_print_completion(qpair, &cpl);
527 req->cb_fn(req->cb_arg, &cpl);
529 nvme_free_request(req);
533 nvme_qpair_process_completions(struct nvme_qpair *qpair)
535 struct nvme_tracker *tr;
536 struct nvme_completion cpl;
538 bool in_panic = dumping || SCHEDULER_STOPPED();
540 qpair->num_intr_handler_calls++;
543 * qpair is not enabled, likely because a controller reset is is in
544 * progress. Ignore the interrupt - any I/O that was associated with
545 * this interrupt will get retried when the reset is complete.
547 if (!qpair->is_enabled)
550 bus_dmamap_sync(qpair->dma_tag, qpair->queuemem_map,
551 BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
553 * A panic can stop the CPU this routine is running on at any point. If
554 * we're called during a panic, complete the sq_head wrap protocol for
555 * the case where we are interrupted just after the increment at 1
556 * below, but before we can reset cq_head to zero at 2. Also cope with
557 * the case where we do the zero at 2, but may or may not have done the
558 * phase adjustment at step 3. The panic machinery flushes all pending
559 * memory writes, so we can make these strong ordering assumptions
560 * that would otherwise be unwise if we were racing in real time.
562 if (__predict_false(in_panic)) {
563 if (qpair->cq_head == qpair->num_entries) {
565 * Here we know that we need to zero cq_head and then negate
566 * the phase, which hasn't been assigned if cq_head isn't
567 * zero due to the atomic_store_rel.
570 qpair->phase = !qpair->phase;
571 } else if (qpair->cq_head == 0) {
573 * In this case, we know that the assignment at 2
574 * happened below, but we don't know if it 3 happened or
575 * not. To do this, we look at the last completion
576 * entry and set the phase to the opposite phase
577 * that it has. This gets us back in sync
579 cpl = qpair->cpl[qpair->num_entries - 1];
580 nvme_completion_swapbytes(&cpl);
581 qpair->phase = !NVME_STATUS_GET_P(cpl.status);
586 cpl = qpair->cpl[qpair->cq_head];
588 /* Convert to host endian */
589 nvme_completion_swapbytes(&cpl);
591 if (NVME_STATUS_GET_P(cpl.status) != qpair->phase)
594 tr = qpair->act_tr[cpl.cid];
597 nvme_qpair_complete_tracker(tr, &cpl, ERROR_PRINT_ALL);
598 qpair->sq_head = cpl.sqhd;
600 } else if (!in_panic) {
602 * A missing tracker is normally an error. However, a
603 * panic can stop the CPU this routine is running on
604 * after completing an I/O but before updating
605 * qpair->cq_head at 1 below. Later, we re-enter this
606 * routine to poll I/O associated with the kernel
607 * dump. We find that the tr has been set to null before
608 * calling the completion routine. If it hasn't
609 * completed (or it triggers a panic), then '1' below
610 * won't have updated cq_head. Rather than panic again,
611 * ignore this condition because it's not unexpected.
613 nvme_printf(qpair->ctrlr,
614 "cpl does not map to outstanding cmd\n");
615 /* nvme_dump_completion expects device endianess */
616 nvme_dump_completion(&qpair->cpl[qpair->cq_head]);
617 KASSERT(0, ("received completion for unknown cmd"));
621 * There's a number of races with the following (see above) when
622 * the system panics. We compensate for each one of them by
623 * using the atomic store to force strong ordering (at least when
624 * viewed in the aftermath of a panic).
626 if (++qpair->cq_head == qpair->num_entries) { /* 1 */
627 atomic_store_rel_int(&qpair->cq_head, 0); /* 2 */
628 qpair->phase = !qpair->phase; /* 3 */
631 bus_space_write_4(qpair->ctrlr->bus_tag, qpair->ctrlr->bus_handle,
632 qpair->cq_hdbl_off, qpair->cq_head);
638 nvme_qpair_msix_handler(void *arg)
640 struct nvme_qpair *qpair = arg;
642 nvme_qpair_process_completions(qpair);
646 nvme_qpair_construct(struct nvme_qpair *qpair,
647 uint32_t num_entries, uint32_t num_trackers,
648 struct nvme_controller *ctrlr)
650 struct nvme_tracker *tr;
651 size_t cmdsz, cplsz, prpsz, allocsz, prpmemsz;
652 uint64_t queuemem_phys, prpmem_phys, list_phys;
653 uint8_t *queuemem, *prpmem, *prp_list;
656 qpair->vector = ctrlr->msix_enabled ? qpair->id : 0;
657 qpair->num_entries = num_entries;
658 qpair->num_trackers = num_trackers;
659 qpair->ctrlr = ctrlr;
661 if (ctrlr->msix_enabled) {
663 * MSI-X vector resource IDs start at 1, so we add one to
664 * the queue's vector to get the corresponding rid to use.
666 qpair->rid = qpair->vector + 1;
668 qpair->res = bus_alloc_resource_any(ctrlr->dev, SYS_RES_IRQ,
669 &qpair->rid, RF_ACTIVE);
670 if (bus_setup_intr(ctrlr->dev, qpair->res,
671 INTR_TYPE_MISC | INTR_MPSAFE, NULL,
672 nvme_qpair_msix_handler, qpair, &qpair->tag) != 0) {
673 nvme_printf(ctrlr, "unable to setup intx handler\n");
676 if (qpair->id == 0) {
677 bus_describe_intr(ctrlr->dev, qpair->res, qpair->tag,
680 bus_describe_intr(ctrlr->dev, qpair->res, qpair->tag,
681 "io%d", qpair->id - 1);
685 mtx_init(&qpair->lock, "nvme qpair lock", NULL, MTX_DEF);
687 /* Note: NVMe PRP format is restricted to 4-byte alignment. */
688 err = bus_dma_tag_create(bus_get_dma_tag(ctrlr->dev),
689 4, PAGE_SIZE, BUS_SPACE_MAXADDR,
690 BUS_SPACE_MAXADDR, NULL, NULL, ctrlr->max_xfer_size,
691 btoc(ctrlr->max_xfer_size) + 1, PAGE_SIZE, 0,
692 NULL, NULL, &qpair->dma_tag_payload);
694 nvme_printf(ctrlr, "payload tag create failed %d\n", err);
699 * Each component must be page aligned, and individual PRP lists
700 * cannot cross a page boundary.
702 cmdsz = qpair->num_entries * sizeof(struct nvme_command);
703 cmdsz = roundup2(cmdsz, PAGE_SIZE);
704 cplsz = qpair->num_entries * sizeof(struct nvme_completion);
705 cplsz = roundup2(cplsz, PAGE_SIZE);
707 * For commands requiring more than 2 PRP entries, one PRP will be
708 * embedded in the command (prp1), and the rest of the PRP entries
709 * will be in a list pointed to by the command (prp2).
711 prpsz = sizeof(uint64_t) * btoc(ctrlr->max_xfer_size);
712 prpmemsz = qpair->num_trackers * prpsz;
713 allocsz = cmdsz + cplsz + prpmemsz;
715 err = bus_dma_tag_create(bus_get_dma_tag(ctrlr->dev),
716 PAGE_SIZE, 0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
717 allocsz, 1, allocsz, 0, NULL, NULL, &qpair->dma_tag);
719 nvme_printf(ctrlr, "tag create failed %d\n", err);
722 bus_dma_tag_set_domain(qpair->dma_tag, qpair->domain);
724 if (bus_dmamem_alloc(qpair->dma_tag, (void **)&queuemem,
725 BUS_DMA_COHERENT | BUS_DMA_NOWAIT, &qpair->queuemem_map)) {
726 nvme_printf(ctrlr, "failed to alloc qpair memory\n");
730 if (bus_dmamap_load(qpair->dma_tag, qpair->queuemem_map,
731 queuemem, allocsz, nvme_single_map, &queuemem_phys, 0) != 0) {
732 nvme_printf(ctrlr, "failed to load qpair memory\n");
733 bus_dmamem_free(qpair->dma_tag, qpair->cmd,
734 qpair->queuemem_map);
739 qpair->num_intr_handler_calls = 0;
740 qpair->num_retries = 0;
741 qpair->num_failures = 0;
742 qpair->cmd = (struct nvme_command *)queuemem;
743 qpair->cpl = (struct nvme_completion *)(queuemem + cmdsz);
744 prpmem = (uint8_t *)(queuemem + cmdsz + cplsz);
745 qpair->cmd_bus_addr = queuemem_phys;
746 qpair->cpl_bus_addr = queuemem_phys + cmdsz;
747 prpmem_phys = queuemem_phys + cmdsz + cplsz;
750 * Calcuate the stride of the doorbell register. Many emulators set this
751 * value to correspond to a cache line. However, some hardware has set
752 * it to various small values.
754 qpair->sq_tdbl_off = nvme_mmio_offsetof(doorbell[0]) +
755 (qpair->id << (ctrlr->dstrd + 1));
756 qpair->cq_hdbl_off = nvme_mmio_offsetof(doorbell[0]) +
757 (qpair->id << (ctrlr->dstrd + 1)) + (1 << ctrlr->dstrd);
759 TAILQ_INIT(&qpair->free_tr);
760 TAILQ_INIT(&qpair->outstanding_tr);
761 STAILQ_INIT(&qpair->queued_req);
763 list_phys = prpmem_phys;
765 for (i = 0; i < qpair->num_trackers; i++) {
766 if (list_phys + prpsz > prpmem_phys + prpmemsz) {
767 qpair->num_trackers = i;
772 * Make sure that the PRP list for this tracker doesn't
773 * overflow to another page.
775 if (trunc_page(list_phys) !=
776 trunc_page(list_phys + prpsz - 1)) {
777 list_phys = roundup2(list_phys, PAGE_SIZE);
779 (uint8_t *)roundup2((uintptr_t)prp_list, PAGE_SIZE);
782 tr = malloc_domainset(sizeof(*tr), M_NVME,
783 DOMAINSET_PREF(qpair->domain), M_ZERO | M_WAITOK);
784 bus_dmamap_create(qpair->dma_tag_payload, 0,
785 &tr->payload_dma_map);
786 callout_init(&tr->timer, 1);
789 tr->prp = (uint64_t *)prp_list;
790 tr->prp_bus_addr = list_phys;
791 TAILQ_INSERT_HEAD(&qpair->free_tr, tr, tailq);
796 if (qpair->num_trackers == 0) {
797 nvme_printf(ctrlr, "failed to allocate enough trackers\n");
801 qpair->act_tr = malloc_domainset(sizeof(struct nvme_tracker *) *
802 qpair->num_entries, M_NVME, DOMAINSET_PREF(qpair->domain),
807 nvme_qpair_destroy(qpair);
812 nvme_qpair_destroy(struct nvme_qpair *qpair)
814 struct nvme_tracker *tr;
817 bus_teardown_intr(qpair->ctrlr->dev, qpair->res, qpair->tag);
822 free(qpair->act_tr, M_NVME);
823 qpair->act_tr = NULL;
826 while (!TAILQ_EMPTY(&qpair->free_tr)) {
827 tr = TAILQ_FIRST(&qpair->free_tr);
828 TAILQ_REMOVE(&qpair->free_tr, tr, tailq);
829 bus_dmamap_destroy(qpair->dma_tag_payload,
830 tr->payload_dma_map);
834 if (qpair->cmd != NULL) {
835 bus_dmamap_unload(qpair->dma_tag, qpair->queuemem_map);
836 bus_dmamem_free(qpair->dma_tag, qpair->cmd,
837 qpair->queuemem_map);
841 if (qpair->dma_tag) {
842 bus_dma_tag_destroy(qpair->dma_tag);
843 qpair->dma_tag = NULL;
846 if (qpair->dma_tag_payload) {
847 bus_dma_tag_destroy(qpair->dma_tag_payload);
848 qpair->dma_tag_payload = NULL;
851 if (mtx_initialized(&qpair->lock))
852 mtx_destroy(&qpair->lock);
855 bus_release_resource(qpair->ctrlr->dev, SYS_RES_IRQ,
856 rman_get_rid(qpair->res), qpair->res);
862 nvme_admin_qpair_abort_aers(struct nvme_qpair *qpair)
864 struct nvme_tracker *tr;
866 tr = TAILQ_FIRST(&qpair->outstanding_tr);
868 if (tr->req->cmd.opc == NVME_OPC_ASYNC_EVENT_REQUEST) {
869 nvme_qpair_manual_complete_tracker(tr,
870 NVME_SCT_GENERIC, NVME_SC_ABORTED_SQ_DELETION, 0,
872 tr = TAILQ_FIRST(&qpair->outstanding_tr);
874 tr = TAILQ_NEXT(tr, tailq);
880 nvme_admin_qpair_destroy(struct nvme_qpair *qpair)
883 nvme_admin_qpair_abort_aers(qpair);
884 nvme_qpair_destroy(qpair);
888 nvme_io_qpair_destroy(struct nvme_qpair *qpair)
891 nvme_qpair_destroy(qpair);
895 nvme_abort_complete(void *arg, const struct nvme_completion *status)
897 struct nvme_tracker *tr = arg;
900 * If cdw0 == 1, the controller was not able to abort the command
901 * we requested. We still need to check the active tracker array,
902 * to cover race where I/O timed out at same time controller was
903 * completing the I/O.
905 if (status->cdw0 == 1 && tr->qpair->act_tr[tr->cid] != NULL) {
907 * An I/O has timed out, and the controller was unable to
908 * abort it for some reason. Construct a fake completion
909 * status, and then complete the I/O's tracker manually.
911 nvme_printf(tr->qpair->ctrlr,
912 "abort command failed, aborting command manually\n");
913 nvme_qpair_manual_complete_tracker(tr,
914 NVME_SCT_GENERIC, NVME_SC_ABORTED_BY_REQUEST, 0, ERROR_PRINT_ALL);
919 nvme_timeout(void *arg)
921 struct nvme_tracker *tr = arg;
922 struct nvme_qpair *qpair = tr->qpair;
923 struct nvme_controller *ctrlr = qpair->ctrlr;
928 * Read csts to get value of cfs - controller fatal status.
929 * If no fatal status, try to call the completion routine, and
930 * if completes transactions, report a missed interrupt and
931 * return (this may need to be rate limited). Otherwise, if
932 * aborts are enabled and the controller is not reporting
933 * fatal status, abort the command. Otherwise, just reset the
934 * controller and hope for the best.
936 csts = nvme_mmio_read_4(ctrlr, csts);
937 cfs = (csts >> NVME_CSTS_REG_CFS_SHIFT) & NVME_CSTS_REG_CFS_MASK;
938 if (cfs == 0 && nvme_qpair_process_completions(qpair)) {
939 nvme_printf(ctrlr, "Missing interrupt\n");
942 if (ctrlr->enable_aborts && cfs == 0) {
943 nvme_printf(ctrlr, "Aborting command due to a timeout.\n");
944 nvme_ctrlr_cmd_abort(ctrlr, tr->cid, qpair->id,
945 nvme_abort_complete, tr);
947 nvme_printf(ctrlr, "Resetting controller due to a timeout%s.\n",
948 (csts == 0xffffffff) ? " and possible hot unplug" :
949 (cfs ? " and fatal error status" : ""));
950 nvme_ctrlr_reset(ctrlr);
955 nvme_qpair_submit_tracker(struct nvme_qpair *qpair, struct nvme_tracker *tr)
957 struct nvme_request *req;
958 struct nvme_controller *ctrlr;
961 mtx_assert(&qpair->lock, MA_OWNED);
964 req->cmd.cid = tr->cid;
965 qpair->act_tr[tr->cid] = tr;
966 ctrlr = qpair->ctrlr;
969 if (req->cb_fn == nvme_completion_poll_cb)
972 timeout = ctrlr->timeout_period * hz;
973 callout_reset_on(&tr->timer, timeout, nvme_timeout, tr,
977 /* Copy the command from the tracker to the submission queue. */
978 memcpy(&qpair->cmd[qpair->sq_tail], &req->cmd, sizeof(req->cmd));
980 if (++qpair->sq_tail == qpair->num_entries)
983 bus_dmamap_sync(qpair->dma_tag, qpair->queuemem_map,
984 BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
985 bus_space_write_4(qpair->ctrlr->bus_tag, qpair->ctrlr->bus_handle,
986 qpair->sq_tdbl_off, qpair->sq_tail);
991 nvme_payload_map(void *arg, bus_dma_segment_t *seg, int nseg, int error)
993 struct nvme_tracker *tr = arg;
997 * If the mapping operation failed, return immediately. The caller
998 * is responsible for detecting the error status and failing the
1002 nvme_printf(tr->qpair->ctrlr,
1003 "nvme_payload_map err %d\n", error);
1008 * Note that we specified PAGE_SIZE for alignment and max
1009 * segment size when creating the bus dma tags. So here
1010 * we can safely just transfer each segment to its
1011 * associated PRP entry.
1013 tr->req->cmd.prp1 = htole64(seg[0].ds_addr);
1016 tr->req->cmd.prp2 = htole64(seg[1].ds_addr);
1017 } else if (nseg > 2) {
1019 tr->req->cmd.prp2 = htole64((uint64_t)tr->prp_bus_addr);
1020 while (cur_nseg < nseg) {
1021 tr->prp[cur_nseg-1] =
1022 htole64((uint64_t)seg[cur_nseg].ds_addr);
1027 * prp2 should not be used by the controller
1028 * since there is only one segment, but set
1029 * to 0 just to be safe.
1031 tr->req->cmd.prp2 = 0;
1034 bus_dmamap_sync(tr->qpair->dma_tag_payload, tr->payload_dma_map,
1035 BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1036 nvme_qpair_submit_tracker(tr->qpair, tr);
1040 _nvme_qpair_submit_request(struct nvme_qpair *qpair, struct nvme_request *req)
1042 struct nvme_tracker *tr;
1045 mtx_assert(&qpair->lock, MA_OWNED);
1047 tr = TAILQ_FIRST(&qpair->free_tr);
1050 if (tr == NULL || !qpair->is_enabled) {
1052 * No tracker is available, or the qpair is disabled due to
1053 * an in-progress controller-level reset or controller
1057 if (qpair->ctrlr->is_failed) {
1059 * The controller has failed. Post the request to a
1060 * task where it will be aborted, so that we do not
1061 * invoke the request's callback in the context
1062 * of the submission.
1064 nvme_ctrlr_post_failed_request(qpair->ctrlr, req);
1067 * Put the request on the qpair's request queue to be
1068 * processed when a tracker frees up via a command
1069 * completion or when the controller reset is
1072 STAILQ_INSERT_TAIL(&qpair->queued_req, req, stailq);
1077 TAILQ_REMOVE(&qpair->free_tr, tr, tailq);
1078 TAILQ_INSERT_TAIL(&qpair->outstanding_tr, tr, tailq);
1081 switch (req->type) {
1082 case NVME_REQUEST_VADDR:
1083 KASSERT(req->payload_size <= qpair->ctrlr->max_xfer_size,
1084 ("payload_size (%d) exceeds max_xfer_size (%d)\n",
1085 req->payload_size, qpair->ctrlr->max_xfer_size));
1086 err = bus_dmamap_load(tr->qpair->dma_tag_payload,
1087 tr->payload_dma_map, req->u.payload, req->payload_size,
1088 nvme_payload_map, tr, 0);
1090 nvme_printf(qpair->ctrlr,
1091 "bus_dmamap_load returned 0x%x!\n", err);
1093 case NVME_REQUEST_NULL:
1094 nvme_qpair_submit_tracker(tr->qpair, tr);
1096 case NVME_REQUEST_BIO:
1097 KASSERT(req->u.bio->bio_bcount <= qpair->ctrlr->max_xfer_size,
1098 ("bio->bio_bcount (%jd) exceeds max_xfer_size (%d)\n",
1099 (intmax_t)req->u.bio->bio_bcount,
1100 qpair->ctrlr->max_xfer_size));
1101 err = bus_dmamap_load_bio(tr->qpair->dma_tag_payload,
1102 tr->payload_dma_map, req->u.bio, nvme_payload_map, tr, 0);
1104 nvme_printf(qpair->ctrlr,
1105 "bus_dmamap_load_bio returned 0x%x!\n", err);
1107 case NVME_REQUEST_CCB:
1108 err = bus_dmamap_load_ccb(tr->qpair->dma_tag_payload,
1109 tr->payload_dma_map, req->u.payload,
1110 nvme_payload_map, tr, 0);
1112 nvme_printf(qpair->ctrlr,
1113 "bus_dmamap_load_ccb returned 0x%x!\n", err);
1116 panic("unknown nvme request type 0x%x\n", req->type);
1122 * The dmamap operation failed, so we manually fail the
1123 * tracker here with DATA_TRANSFER_ERROR status.
1125 * nvme_qpair_manual_complete_tracker must not be called
1126 * with the qpair lock held.
1128 mtx_unlock(&qpair->lock);
1129 nvme_qpair_manual_complete_tracker(tr, NVME_SCT_GENERIC,
1130 NVME_SC_DATA_TRANSFER_ERROR, DO_NOT_RETRY, ERROR_PRINT_ALL);
1131 mtx_lock(&qpair->lock);
1136 nvme_qpair_submit_request(struct nvme_qpair *qpair, struct nvme_request *req)
1139 mtx_lock(&qpair->lock);
1140 _nvme_qpair_submit_request(qpair, req);
1141 mtx_unlock(&qpair->lock);
1145 nvme_qpair_enable(struct nvme_qpair *qpair)
1148 qpair->is_enabled = true;
1152 nvme_qpair_reset(struct nvme_qpair *qpair)
1155 qpair->sq_head = qpair->sq_tail = qpair->cq_head = 0;
1158 * First time through the completion queue, HW will set phase
1159 * bit on completions to 1. So set this to 1 here, indicating
1160 * we're looking for a 1 to know which entries have completed.
1161 * we'll toggle the bit each time when the completion queue
1166 memset(qpair->cmd, 0,
1167 qpair->num_entries * sizeof(struct nvme_command));
1168 memset(qpair->cpl, 0,
1169 qpair->num_entries * sizeof(struct nvme_completion));
1173 nvme_admin_qpair_enable(struct nvme_qpair *qpair)
1175 struct nvme_tracker *tr;
1176 struct nvme_tracker *tr_temp;
1179 * Manually abort each outstanding admin command. Do not retry
1180 * admin commands found here, since they will be left over from
1181 * a controller reset and its likely the context in which the
1182 * command was issued no longer applies.
1184 TAILQ_FOREACH_SAFE(tr, &qpair->outstanding_tr, tailq, tr_temp) {
1185 nvme_printf(qpair->ctrlr,
1186 "aborting outstanding admin command\n");
1187 nvme_qpair_manual_complete_tracker(tr, NVME_SCT_GENERIC,
1188 NVME_SC_ABORTED_BY_REQUEST, DO_NOT_RETRY, ERROR_PRINT_ALL);
1191 nvme_qpair_enable(qpair);
1195 nvme_io_qpair_enable(struct nvme_qpair *qpair)
1197 STAILQ_HEAD(, nvme_request) temp;
1198 struct nvme_tracker *tr;
1199 struct nvme_tracker *tr_temp;
1200 struct nvme_request *req;
1203 * Manually abort each outstanding I/O. This normally results in a
1204 * retry, unless the retry count on the associated request has
1205 * reached its limit.
1207 TAILQ_FOREACH_SAFE(tr, &qpair->outstanding_tr, tailq, tr_temp) {
1208 nvme_printf(qpair->ctrlr, "aborting outstanding i/o\n");
1209 nvme_qpair_manual_complete_tracker(tr, NVME_SCT_GENERIC,
1210 NVME_SC_ABORTED_BY_REQUEST, 0, ERROR_PRINT_NO_RETRY);
1213 mtx_lock(&qpair->lock);
1215 nvme_qpair_enable(qpair);
1218 STAILQ_SWAP(&qpair->queued_req, &temp, nvme_request);
1220 while (!STAILQ_EMPTY(&temp)) {
1221 req = STAILQ_FIRST(&temp);
1222 STAILQ_REMOVE_HEAD(&temp, stailq);
1223 nvme_printf(qpair->ctrlr, "resubmitting queued i/o\n");
1224 nvme_qpair_print_command(qpair, &req->cmd);
1225 _nvme_qpair_submit_request(qpair, req);
1228 mtx_unlock(&qpair->lock);
1232 nvme_qpair_disable(struct nvme_qpair *qpair)
1234 struct nvme_tracker *tr;
1236 qpair->is_enabled = false;
1237 mtx_lock(&qpair->lock);
1238 TAILQ_FOREACH(tr, &qpair->outstanding_tr, tailq)
1239 callout_stop(&tr->timer);
1240 mtx_unlock(&qpair->lock);
1244 nvme_admin_qpair_disable(struct nvme_qpair *qpair)
1247 nvme_qpair_disable(qpair);
1248 nvme_admin_qpair_abort_aers(qpair);
1252 nvme_io_qpair_disable(struct nvme_qpair *qpair)
1255 nvme_qpair_disable(qpair);
1259 nvme_qpair_fail(struct nvme_qpair *qpair)
1261 struct nvme_tracker *tr;
1262 struct nvme_request *req;
1264 if (!mtx_initialized(&qpair->lock))
1267 mtx_lock(&qpair->lock);
1269 while (!STAILQ_EMPTY(&qpair->queued_req)) {
1270 req = STAILQ_FIRST(&qpair->queued_req);
1271 STAILQ_REMOVE_HEAD(&qpair->queued_req, stailq);
1272 nvme_printf(qpair->ctrlr, "failing queued i/o\n");
1273 mtx_unlock(&qpair->lock);
1274 nvme_qpair_manual_complete_request(qpair, req, NVME_SCT_GENERIC,
1275 NVME_SC_ABORTED_BY_REQUEST);
1276 mtx_lock(&qpair->lock);
1279 /* Manually abort each outstanding I/O. */
1280 while (!TAILQ_EMPTY(&qpair->outstanding_tr)) {
1281 tr = TAILQ_FIRST(&qpair->outstanding_tr);
1283 * Do not remove the tracker. The abort_tracker path will
1286 nvme_printf(qpair->ctrlr, "failing outstanding i/o\n");
1287 mtx_unlock(&qpair->lock);
1288 nvme_qpair_manual_complete_tracker(tr, NVME_SCT_GENERIC,
1289 NVME_SC_ABORTED_BY_REQUEST, DO_NOT_RETRY, ERROR_PRINT_ALL);
1290 mtx_lock(&qpair->lock);
1293 mtx_unlock(&qpair->lock);