sys/dev/nvme/nvme_qpair.c

   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. Requests failed with ABORTED_BY_REQUEST
 335          *  set the DNR bit correctly since the driver controls that.
 336          */
 337         switch (sct) {
 338         case NVME_SCT_GENERIC:
 339                 switch (sc) {
 340                 case NVME_SC_ABORTED_BY_REQUEST:
 341                 case NVME_SC_NAMESPACE_NOT_READY:
 342                         if (dnr)
 343                                 return (0);
 344                         else
 345                                 return (1);
 346                 case NVME_SC_INVALID_OPCODE:
 347                 case NVME_SC_INVALID_FIELD:
 348                 case NVME_SC_COMMAND_ID_CONFLICT:
 349                 case NVME_SC_DATA_TRANSFER_ERROR:
 350                 case NVME_SC_ABORTED_POWER_LOSS:
 351                 case NVME_SC_INTERNAL_DEVICE_ERROR:
 352                 case NVME_SC_ABORTED_SQ_DELETION:
 353                 case NVME_SC_ABORTED_FAILED_FUSED:
 354                 case NVME_SC_ABORTED_MISSING_FUSED:
 355                 case NVME_SC_INVALID_NAMESPACE_OR_FORMAT:
 356                 case NVME_SC_COMMAND_SEQUENCE_ERROR:
 357                 case NVME_SC_LBA_OUT_OF_RANGE:
 358                 case NVME_SC_CAPACITY_EXCEEDED:
 359                 default:
 360                         return (0);
 361                 }
 362         case NVME_SCT_COMMAND_SPECIFIC:
 363         case NVME_SCT_MEDIA_ERROR:
 364         case NVME_SCT_VENDOR_SPECIFIC:
 365         default:
 366                 return (0);
 367         }
 368 }
 369
 370 static void
 371 nvme_qpair_complete_tracker(struct nvme_qpair *qpair, struct nvme_tracker *tr,
 372     struct nvme_completion *cpl, boolean_t print_on_error)
 373 {
 374         struct nvme_request     *req;
 375         boolean_t               retry, error;
 376
 377         req = tr->req;
 378         error = nvme_completion_is_error(cpl);
 379         retry = error && nvme_completion_is_retry(cpl) &&
 380            req->retries < nvme_retry_count;
 381
 382         if (error && print_on_error) {
 383                 nvme_qpair_print_command(qpair, &req->cmd);
 384                 nvme_qpair_print_completion(qpair, cpl);
 385         }
 386
 387         qpair->act_tr[cpl->cid] = NULL;
 388
 389         KASSERT(cpl->cid == req->cmd.cid, ("cpl cid does not match cmd cid\n"));
 390
 391         if (req->cb_fn && !retry)
 392                 req->cb_fn(req->cb_arg, cpl);
 393
 394         mtx_lock(&qpair->lock);
 395         callout_stop(&tr->timer);
 396
 397         if (retry) {
 398                 req->retries++;
 399                 nvme_qpair_submit_tracker(qpair, tr);
 400         } else {
 401                 if (req->type != NVME_REQUEST_NULL) {
 402                         bus_dmamap_sync(qpair->dma_tag_payload,
 403                             tr->payload_dma_map,
 404                             BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
 405                         bus_dmamap_unload(qpair->dma_tag_payload,
 406                             tr->payload_dma_map);
 407                 }
 408
 409                 nvme_free_request(req);
 410                 tr->req = NULL;
 411
 412                 TAILQ_REMOVE(&qpair->outstanding_tr, tr, tailq);
 413                 TAILQ_INSERT_HEAD(&qpair->free_tr, tr, tailq);
 414
 415                 /*
 416                  * If the controller is in the middle of resetting, don't
 417                  *  try to submit queued requests here - let the reset logic
 418                  *  handle that instead.
 419                  */
 420                 if (!STAILQ_EMPTY(&qpair->queued_req) &&
 421                     !qpair->ctrlr->is_resetting) {
 422                         req = STAILQ_FIRST(&qpair->queued_req);
 423                         STAILQ_REMOVE_HEAD(&qpair->queued_req, stailq);
 424                         _nvme_qpair_submit_request(qpair, req);
 425                 }
 426         }
 427
 428         mtx_unlock(&qpair->lock);
 429 }
 430
 431 static void
 432 nvme_qpair_manual_complete_tracker(struct nvme_qpair *qpair,
 433     struct nvme_tracker *tr, uint32_t sct, uint32_t sc, uint32_t dnr,
 434     boolean_t print_on_error)
 435 {
 436         struct nvme_completion  cpl;
 437
 438         memset(&cpl, 0, sizeof(cpl));
 439         cpl.sqid = qpair->id;
 440         cpl.cid = tr->cid;
 441         cpl.status |= (sct & NVME_STATUS_SCT_MASK) << NVME_STATUS_SCT_SHIFT;
 442         cpl.status |= (sc & NVME_STATUS_SC_MASK) << NVME_STATUS_SC_SHIFT;
 443         cpl.status |= (dnr & NVME_STATUS_DNR_MASK) << NVME_STATUS_DNR_SHIFT;
 444         nvme_qpair_complete_tracker(qpair, tr, &cpl, print_on_error);
 445 }
 446
 447 void
 448 nvme_qpair_manual_complete_request(struct nvme_qpair *qpair,
 449     struct nvme_request *req, uint32_t sct, uint32_t sc,
 450     boolean_t print_on_error)
 451 {
 452         struct nvme_completion  cpl;
 453         boolean_t               error;
 454
 455         memset(&cpl, 0, sizeof(cpl));
 456         cpl.sqid = qpair->id;
 457         cpl.status |= (sct & NVME_STATUS_SCT_MASK) << NVME_STATUS_SCT_SHIFT;
 458         cpl.status |= (sc & NVME_STATUS_SC_MASK) << NVME_STATUS_SC_SHIFT;
 459
 460         error = nvme_completion_is_error(&cpl);
 461
 462         if (error && print_on_error) {
 463                 nvme_qpair_print_command(qpair, &req->cmd);
 464                 nvme_qpair_print_completion(qpair, &cpl);
 465         }
 466
 467         if (req->cb_fn)
 468                 req->cb_fn(req->cb_arg, &cpl);
 469
 470         nvme_free_request(req);
 471 }
 472
 473 bool
 474 nvme_qpair_process_completions(struct nvme_qpair *qpair)
 475 {
 476         struct nvme_tracker     *tr;
 477         struct nvme_completion  cpl;
 478         int done = 0;
 479
 480         qpair->num_intr_handler_calls++;
 481
 482         if (!qpair->is_enabled)
 483                 /*
 484                  * qpair is not enabled, likely because a controller reset is
 485                  *  is in progress.  Ignore the interrupt - any I/O that was
 486                  *  associated with this interrupt will get retried when the
 487                  *  reset is complete.
 488                  */
 489                 return (false);
 490
 491         bus_dmamap_sync(qpair->dma_tag, qpair->queuemem_map,
 492             BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
 493         while (1) {
 494                 cpl = qpair->cpl[qpair->cq_head];
 495
 496                 /* Convert to host endian */
 497                 nvme_completion_swapbytes(&cpl);
 498
 499                 if (NVME_STATUS_GET_P(cpl.status) != qpair->phase)
 500                         break;
 501
 502                 tr = qpair->act_tr[cpl.cid];
 503
 504                 if (tr != NULL) {
 505                         nvme_qpair_complete_tracker(qpair, tr, &cpl, TRUE);
 506                         qpair->sq_head = cpl.sqhd;
 507                         done++;
 508                 } else {
 509                         nvme_printf(qpair->ctrlr,
 510                             "cpl does not map to outstanding cmd\n");
 511                         /* nvme_dump_completion expects device endianess */
 512                         nvme_dump_completion(&qpair->cpl[qpair->cq_head]);
 513                         KASSERT(0, ("received completion for unknown cmd\n"));
 514                 }
 515
 516                 if (++qpair->cq_head == qpair->num_entries) {
 517                         qpair->cq_head = 0;
 518                         qpair->phase = !qpair->phase;
 519                 }
 520
 521                 nvme_mmio_write_4(qpair->ctrlr, doorbell[qpair->id].cq_hdbl,
 522                     qpair->cq_head);
 523         }
 524         return (done != 0);
 525 }
 526
 527 static void
 528 nvme_qpair_msix_handler(void *arg)
 529 {
 530         struct nvme_qpair *qpair = arg;
 531
 532         nvme_qpair_process_completions(qpair);
 533 }
 534
 535 int
 536 nvme_qpair_construct(struct nvme_qpair *qpair, uint32_t id,
 537     uint16_t vector, uint32_t num_entries, uint32_t num_trackers,
 538     struct nvme_controller *ctrlr)
 539 {
 540         struct nvme_tracker     *tr;
 541         size_t                  cmdsz, cplsz, prpsz, allocsz, prpmemsz;
 542         uint64_t                queuemem_phys, prpmem_phys, list_phys;
 543         uint8_t                 *queuemem, *prpmem, *prp_list;
 544         int                     i, err;
 545
 546         qpair->id = id;
 547         qpair->vector = vector;
 548         qpair->num_entries = num_entries;
 549         qpair->num_trackers = num_trackers;
 550         qpair->ctrlr = ctrlr;
 551
 552         if (ctrlr->msix_enabled) {
 553
 554                 /*
 555                  * MSI-X vector resource IDs start at 1, so we add one to
 556                  *  the queue's vector to get the corresponding rid to use.
 557                  */
 558                 qpair->rid = vector + 1;
 559
 560                 qpair->res = bus_alloc_resource_any(ctrlr->dev, SYS_RES_IRQ,
 561                     &qpair->rid, RF_ACTIVE);
 562                 bus_setup_intr(ctrlr->dev, qpair->res,
 563                     INTR_TYPE_MISC | INTR_MPSAFE, NULL,
 564                     nvme_qpair_msix_handler, qpair, &qpair->tag);
 565                 if (id == 0) {
 566                         bus_describe_intr(ctrlr->dev, qpair->res, qpair->tag,
 567                             "admin");
 568                 } else {
 569                         bus_describe_intr(ctrlr->dev, qpair->res, qpair->tag,
 570                             "io%d", id - 1);
 571                 }
 572         }
 573
 574         mtx_init(&qpair->lock, "nvme qpair lock", NULL, MTX_DEF);
 575
 576         /* Note: NVMe PRP format is restricted to 4-byte alignment. */
 577         err = bus_dma_tag_create(bus_get_dma_tag(ctrlr->dev),
 578             4, PAGE_SIZE, BUS_SPACE_MAXADDR,
 579             BUS_SPACE_MAXADDR, NULL, NULL, NVME_MAX_XFER_SIZE,
 580             (NVME_MAX_XFER_SIZE/PAGE_SIZE)+1, PAGE_SIZE, 0,
 581             NULL, NULL, &qpair->dma_tag_payload);
 582         if (err != 0) {
 583                 nvme_printf(ctrlr, "payload tag create failed %d\n", err);
 584                 goto out;
 585         }
 586
 587         /*
 588          * Each component must be page aligned, and individual PRP lists
 589          * cannot cross a page boundary.
 590          */
 591         cmdsz = qpair->num_entries * sizeof(struct nvme_command);
 592         cmdsz = roundup2(cmdsz, PAGE_SIZE);
 593         cplsz = qpair->num_entries * sizeof(struct nvme_completion);
 594         cplsz = roundup2(cplsz, PAGE_SIZE);
 595         prpsz = sizeof(uint64_t) * NVME_MAX_PRP_LIST_ENTRIES;;
 596         prpmemsz = qpair->num_trackers * prpsz;
 597         allocsz = cmdsz + cplsz + prpmemsz;
 598
 599         err = bus_dma_tag_create(bus_get_dma_tag(ctrlr->dev),
 600             PAGE_SIZE, 0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
 601             allocsz, 1, allocsz, 0, NULL, NULL, &qpair->dma_tag);
 602         if (err != 0) {
 603                 nvme_printf(ctrlr, "tag create failed %d\n", err);
 604                 goto out;
 605         }
 606
 607         if (bus_dmamem_alloc(qpair->dma_tag, (void **)&queuemem,
 608             BUS_DMA_NOWAIT, &qpair->queuemem_map)) {
 609                 nvme_printf(ctrlr, "failed to alloc qpair memory\n");
 610                 goto out;
 611         }
 612
 613         if (bus_dmamap_load(qpair->dma_tag, qpair->queuemem_map,
 614             queuemem, allocsz, nvme_single_map, &queuemem_phys, 0) != 0) {
 615                 nvme_printf(ctrlr, "failed to load qpair memory\n");
 616                 goto out;
 617         }
 618
 619         qpair->num_cmds = 0;
 620         qpair->num_intr_handler_calls = 0;
 621         qpair->cmd = (struct nvme_command *)queuemem;
 622         qpair->cpl = (struct nvme_completion *)(queuemem + cmdsz);
 623         prpmem = (uint8_t *)(queuemem + cmdsz + cplsz);
 624         qpair->cmd_bus_addr = queuemem_phys;
 625         qpair->cpl_bus_addr = queuemem_phys + cmdsz;
 626         prpmem_phys = queuemem_phys + cmdsz + cplsz;
 627
 628         qpair->sq_tdbl_off = nvme_mmio_offsetof(doorbell[id].sq_tdbl);
 629         qpair->cq_hdbl_off = nvme_mmio_offsetof(doorbell[id].cq_hdbl);
 630
 631         TAILQ_INIT(&qpair->free_tr);
 632         TAILQ_INIT(&qpair->outstanding_tr);
 633         STAILQ_INIT(&qpair->queued_req);
 634
 635         list_phys = prpmem_phys;
 636         prp_list = prpmem;
 637         for (i = 0; i < qpair->num_trackers; i++) {
 638
 639                 if (list_phys + prpsz > prpmem_phys + prpmemsz) {
 640                         qpair->num_trackers = i;
 641                         break;
 642                 }
 643
 644                 /*
 645                  * Make sure that the PRP list for this tracker doesn't
 646                  * overflow to another page.
 647                  */
 648                 if (trunc_page(list_phys) !=
 649                     trunc_page(list_phys + prpsz - 1)) {
 650                         list_phys = roundup2(list_phys, PAGE_SIZE);
 651                         prp_list =
 652                             (uint8_t *)roundup2((uintptr_t)prp_list, PAGE_SIZE);
 653                 }
 654
 655                 tr = malloc(sizeof(*tr), M_NVME, M_ZERO | M_WAITOK);
 656                 bus_dmamap_create(qpair->dma_tag_payload, 0,
 657                     &tr->payload_dma_map);
 658                 callout_init(&tr->timer, 1);
 659                 tr->cid = i;
 660                 tr->qpair = qpair;
 661                 tr->prp = (uint64_t *)prp_list;
 662                 tr->prp_bus_addr = list_phys;
 663                 TAILQ_INSERT_HEAD(&qpair->free_tr, tr, tailq);
 664                 list_phys += prpsz;
 665                 prp_list += prpsz;
 666         }
 667
 668         if (qpair->num_trackers == 0) {
 669                 nvme_printf(ctrlr, "failed to allocate enough trackers\n");
 670                 goto out;
 671         }
 672
 673         qpair->act_tr = malloc(sizeof(struct nvme_tracker *) *
 674             qpair->num_entries, M_NVME, M_ZERO | M_WAITOK);
 675         return (0);
 676
 677 out:
 678         nvme_qpair_destroy(qpair);
 679         return (ENOMEM);
 680 }
 681
 682 static void
 683 nvme_qpair_destroy(struct nvme_qpair *qpair)
 684 {
 685         struct nvme_tracker     *tr;
 686
 687         if (qpair->tag)
 688                 bus_teardown_intr(qpair->ctrlr->dev, qpair->res, qpair->tag);
 689
 690         if (mtx_initialized(&qpair->lock))
 691                 mtx_destroy(&qpair->lock);
 692
 693         if (qpair->res)
 694                 bus_release_resource(qpair->ctrlr->dev, SYS_RES_IRQ,
 695                     rman_get_rid(qpair->res), qpair->res);
 696
 697         if (qpair->cmd != NULL) {
 698                 bus_dmamap_unload(qpair->dma_tag, qpair->queuemem_map);
 699                 bus_dmamem_free(qpair->dma_tag, qpair->cmd,
 700                     qpair->queuemem_map);
 701         }
 702
 703         if (qpair->act_tr)
 704                 free(qpair->act_tr, M_NVME);
 705
 706         while (!TAILQ_EMPTY(&qpair->free_tr)) {
 707                 tr = TAILQ_FIRST(&qpair->free_tr);
 708                 TAILQ_REMOVE(&qpair->free_tr, tr, tailq);
 709                 bus_dmamap_destroy(qpair->dma_tag_payload,
 710                     tr->payload_dma_map);
 711                 free(tr, M_NVME);
 712         }
 713
 714         if (qpair->dma_tag)
 715                 bus_dma_tag_destroy(qpair->dma_tag);
 716
 717         if (qpair->dma_tag_payload)
 718                 bus_dma_tag_destroy(qpair->dma_tag_payload);
 719 }
 720
 721 static void
 722 nvme_admin_qpair_abort_aers(struct nvme_qpair *qpair)
 723 {
 724         struct nvme_tracker     *tr;
 725
 726         tr = TAILQ_FIRST(&qpair->outstanding_tr);
 727         while (tr != NULL) {
 728                 if (tr->req->cmd.opc == NVME_OPC_ASYNC_EVENT_REQUEST) {
 729                         nvme_qpair_manual_complete_tracker(qpair, tr,
 730                             NVME_SCT_GENERIC, NVME_SC_ABORTED_SQ_DELETION, 0,
 731                             FALSE);
 732                         tr = TAILQ_FIRST(&qpair->outstanding_tr);
 733                 } else {
 734                         tr = TAILQ_NEXT(tr, tailq);
 735                 }
 736         }
 737 }
 738
 739 void
 740 nvme_admin_qpair_destroy(struct nvme_qpair *qpair)
 741 {
 742
 743         nvme_admin_qpair_abort_aers(qpair);
 744         nvme_qpair_destroy(qpair);
 745 }
 746
 747 void
 748 nvme_io_qpair_destroy(struct nvme_qpair *qpair)
 749 {
 750
 751         nvme_qpair_destroy(qpair);
 752 }
 753
 754 static void
 755 nvme_abort_complete(void *arg, const struct nvme_completion *status)
 756 {
 757         struct nvme_tracker     *tr = arg;
 758
 759         /*
 760          * If cdw0 == 1, the controller was not able to abort the command
 761          *  we requested.  We still need to check the active tracker array,
 762          *  to cover race where I/O timed out at same time controller was
 763          *  completing the I/O.
 764          */
 765         if (status->cdw0 == 1 && tr->qpair->act_tr[tr->cid] != NULL) {
 766                 /*
 767                  * An I/O has timed out, and the controller was unable to
 768                  *  abort it for some reason.  Construct a fake completion
 769                  *  status, and then complete the I/O's tracker manually.
 770                  */
 771                 nvme_printf(tr->qpair->ctrlr,
 772                     "abort command failed, aborting command manually\n");
 773                 nvme_qpair_manual_complete_tracker(tr->qpair, tr,
 774                     NVME_SCT_GENERIC, NVME_SC_ABORTED_BY_REQUEST, 0, TRUE);
 775         }
 776 }
 777
 778 static void
 779 nvme_timeout(void *arg)
 780 {
 781         struct nvme_tracker     *tr = arg;
 782         struct nvme_qpair       *qpair = tr->qpair;
 783         struct nvme_controller  *ctrlr = qpair->ctrlr;
 784         uint32_t                csts;
 785         uint8_t                 cfs;
 786
 787         /*
 788          * Read csts to get value of cfs - controller fatal status.
 789          * If no fatal status, try to call the completion routine, and
 790          * if completes transactions, report a missed interrupt and
 791          * return (this may need to be rate limited). Otherwise, if
 792          * aborts are enabled and the controller is not reporting
 793          * fatal status, abort the command. Otherwise, just reset the
 794          * controller and hope for the best.
 795          */
 796         csts = nvme_mmio_read_4(ctrlr, csts);
 797         cfs = (csts >> NVME_CSTS_REG_CFS_SHIFT) & NVME_CSTS_REG_CFS_MASK;
 798         if (cfs == 0 && nvme_qpair_process_completions(qpair)) {
 799                 nvme_printf(ctrlr, "Missing interrupt\n");
 800                 return;
 801         }
 802         if (ctrlr->enable_aborts && cfs == 0) {
 803                 nvme_printf(ctrlr, "Aborting command due to a timeout.\n");
 804                 nvme_ctrlr_cmd_abort(ctrlr, tr->cid, qpair->id,
 805                     nvme_abort_complete, tr);
 806         } else {
 807                 nvme_printf(ctrlr, "Resetting controller due to a timeout%s.\n",
 808                     cfs ? " and fatal error status" : "");
 809                 nvme_ctrlr_reset(ctrlr);
 810         }
 811 }
 812
 813 void
 814 nvme_qpair_submit_tracker(struct nvme_qpair *qpair, struct nvme_tracker *tr)
 815 {
 816         struct nvme_request     *req;
 817         struct nvme_controller  *ctrlr;
 818
 819         mtx_assert(&qpair->lock, MA_OWNED);
 820
 821         req = tr->req;
 822         req->cmd.cid = tr->cid;
 823         qpair->act_tr[tr->cid] = tr;
 824         ctrlr = qpair->ctrlr;
 825
 826         if (req->timeout)
 827                 callout_reset_curcpu(&tr->timer, ctrlr->timeout_period * hz,
 828                     nvme_timeout, tr);
 829
 830         /* Copy the command from the tracker to the submission queue. */
 831         memcpy(&qpair->cmd[qpair->sq_tail], &req->cmd, sizeof(req->cmd));
 832
 833         if (++qpair->sq_tail == qpair->num_entries)
 834                 qpair->sq_tail = 0;
 835
 836         bus_dmamap_sync(qpair->dma_tag, qpair->queuemem_map,
 837             BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
 838 #ifndef __powerpc__
 839         /*
 840          * powerpc's bus_dmamap_sync() already includes a heavyweight sync, but
 841          * no other archs do.
 842          */
 843         wmb();
 844 #endif
 845
 846         nvme_mmio_write_4(qpair->ctrlr, doorbell[qpair->id].sq_tdbl,
 847             qpair->sq_tail);
 848
 849         qpair->num_cmds++;
 850 }
 851
 852 static void
 853 nvme_payload_map(void *arg, bus_dma_segment_t *seg, int nseg, int error)
 854 {
 855         struct nvme_tracker     *tr = arg;
 856         uint32_t                cur_nseg;
 857
 858         /*
 859          * If the mapping operation failed, return immediately.  The caller
 860          *  is responsible for detecting the error status and failing the
 861          *  tracker manually.
 862          */
 863         if (error != 0) {
 864                 nvme_printf(tr->qpair->ctrlr,
 865                     "nvme_payload_map err %d\n", error);
 866                 return;
 867         }
 868
 869         /*
 870          * Note that we specified PAGE_SIZE for alignment and max
 871          *  segment size when creating the bus dma tags.  So here
 872          *  we can safely just transfer each segment to its
 873          *  associated PRP entry.
 874          */
 875         tr->req->cmd.prp1 = htole64(seg[0].ds_addr);
 876
 877         if (nseg == 2) {
 878                 tr->req->cmd.prp2 = htole64(seg[1].ds_addr);
 879         } else if (nseg > 2) {
 880                 cur_nseg = 1;
 881                 tr->req->cmd.prp2 = htole64((uint64_t)tr->prp_bus_addr);
 882                 while (cur_nseg < nseg) {
 883                         tr->prp[cur_nseg-1] =
 884                             htole64((uint64_t)seg[cur_nseg].ds_addr);
 885                         cur_nseg++;
 886                 }
 887         } else {
 888                 /*
 889                  * prp2 should not be used by the controller
 890                  *  since there is only one segment, but set
 891                  *  to 0 just to be safe.
 892                  */
 893                 tr->req->cmd.prp2 = 0;
 894         }
 895
 896         bus_dmamap_sync(tr->qpair->dma_tag_payload, tr->payload_dma_map,
 897             BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
 898         nvme_qpair_submit_tracker(tr->qpair, tr);
 899 }
 900
 901 static void
 902 _nvme_qpair_submit_request(struct nvme_qpair *qpair, struct nvme_request *req)
 903 {
 904         struct nvme_tracker     *tr;
 905         int                     err = 0;
 906
 907         mtx_assert(&qpair->lock, MA_OWNED);
 908
 909         tr = TAILQ_FIRST(&qpair->free_tr);
 910         req->qpair = qpair;
 911
 912         if (tr == NULL || !qpair->is_enabled) {
 913                 /*
 914                  * No tracker is available, or the qpair is disabled due to
 915                  *  an in-progress controller-level reset or controller
 916                  *  failure.
 917                  */
 918
 919                 if (qpair->ctrlr->is_failed) {
 920                         /*
 921                          * The controller has failed.  Post the request to a
 922                          *  task where it will be aborted, so that we do not
 923                          *  invoke the request's callback in the context
 924                          *  of the submission.
 925                          */
 926                         nvme_ctrlr_post_failed_request(qpair->ctrlr, req);
 927                 } else {
 928                         /*
 929                          * Put the request on the qpair's request queue to be
 930                          *  processed when a tracker frees up via a command
 931                          *  completion or when the controller reset is
 932                          *  completed.
 933                          */
 934                         STAILQ_INSERT_TAIL(&qpair->queued_req, req, stailq);
 935                 }
 936                 return;
 937         }
 938
 939         TAILQ_REMOVE(&qpair->free_tr, tr, tailq);
 940         TAILQ_INSERT_TAIL(&qpair->outstanding_tr, tr, tailq);
 941         tr->req = req;
 942
 943         switch (req->type) {
 944         case NVME_REQUEST_VADDR:
 945                 KASSERT(req->payload_size <= qpair->ctrlr->max_xfer_size,
 946                     ("payload_size (%d) exceeds max_xfer_size (%d)\n",
 947                     req->payload_size, qpair->ctrlr->max_xfer_size));
 948                 err = bus_dmamap_load(tr->qpair->dma_tag_payload,
 949                     tr->payload_dma_map, req->u.payload, req->payload_size,
 950                     nvme_payload_map, tr, 0);
 951                 if (err != 0)
 952                         nvme_printf(qpair->ctrlr,
 953                             "bus_dmamap_load returned 0x%x!\n", err);
 954                 break;
 955         case NVME_REQUEST_NULL:
 956                 nvme_qpair_submit_tracker(tr->qpair, tr);
 957                 break;
 958         case NVME_REQUEST_BIO:
 959                 KASSERT(req->u.bio->bio_bcount <= qpair->ctrlr->max_xfer_size,
 960                     ("bio->bio_bcount (%jd) exceeds max_xfer_size (%d)\n",
 961                     (intmax_t)req->u.bio->bio_bcount,
 962                     qpair->ctrlr->max_xfer_size));
 963                 err = bus_dmamap_load_bio(tr->qpair->dma_tag_payload,
 964                     tr->payload_dma_map, req->u.bio, nvme_payload_map, tr, 0);
 965                 if (err != 0)
 966                         nvme_printf(qpair->ctrlr,
 967                             "bus_dmamap_load_bio returned 0x%x!\n", err);
 968                 break;
 969         case NVME_REQUEST_CCB:
 970                 err = bus_dmamap_load_ccb(tr->qpair->dma_tag_payload,
 971                     tr->payload_dma_map, req->u.payload,
 972                     nvme_payload_map, tr, 0);
 973                 if (err != 0)
 974                         nvme_printf(qpair->ctrlr,
 975                             "bus_dmamap_load_ccb returned 0x%x!\n", err);
 976                 break;
 977         default:
 978                 panic("unknown nvme request type 0x%x\n", req->type);
 979                 break;
 980         }
 981
 982         if (err != 0) {
 983                 /*
 984                  * The dmamap operation failed, so we manually fail the
 985                  *  tracker here with DATA_TRANSFER_ERROR status.
 986                  *
 987                  * nvme_qpair_manual_complete_tracker must not be called
 988                  *  with the qpair lock held.
 989                  */
 990                 mtx_unlock(&qpair->lock);
 991                 nvme_qpair_manual_complete_tracker(qpair, tr, NVME_SCT_GENERIC,
 992                     NVME_SC_DATA_TRANSFER_ERROR, 1 /* do not retry */, TRUE);
 993                 mtx_lock(&qpair->lock);
 994         }
 995 }
 996
 997 void
 998 nvme_qpair_submit_request(struct nvme_qpair *qpair, struct nvme_request *req)
 999 {
1000
1001         mtx_lock(&qpair->lock);
1002         _nvme_qpair_submit_request(qpair, req);
1003         mtx_unlock(&qpair->lock);
1004 }
1005
1006 static void
1007 nvme_qpair_enable(struct nvme_qpair *qpair)
1008 {
1009
1010         qpair->is_enabled = TRUE;
1011 }
1012
1013 void
1014 nvme_qpair_reset(struct nvme_qpair *qpair)
1015 {
1016
1017         qpair->sq_head = qpair->sq_tail = qpair->cq_head = 0;
1018
1019         /*
1020          * First time through the completion queue, HW will set phase
1021          *  bit on completions to 1.  So set this to 1 here, indicating
1022          *  we're looking for a 1 to know which entries have completed.
1023          *  we'll toggle the bit each time when the completion queue
1024          *  rolls over.
1025          */
1026         qpair->phase = 1;
1027
1028         memset(qpair->cmd, 0,
1029             qpair->num_entries * sizeof(struct nvme_command));
1030         memset(qpair->cpl, 0,
1031             qpair->num_entries * sizeof(struct nvme_completion));
1032 }
1033
1034 void
1035 nvme_admin_qpair_enable(struct nvme_qpair *qpair)
1036 {
1037         struct nvme_tracker             *tr;
1038         struct nvme_tracker             *tr_temp;
1039
1040         /*
1041          * Manually abort each outstanding admin command.  Do not retry
1042          *  admin commands found here, since they will be left over from
1043          *  a controller reset and its likely the context in which the
1044          *  command was issued no longer applies.
1045          */
1046         TAILQ_FOREACH_SAFE(tr, &qpair->outstanding_tr, tailq, tr_temp) {
1047                 nvme_printf(qpair->ctrlr,
1048                     "aborting outstanding admin command\n");
1049                 nvme_qpair_manual_complete_tracker(qpair, tr, NVME_SCT_GENERIC,
1050                     NVME_SC_ABORTED_BY_REQUEST, 1 /* do not retry */, TRUE);
1051         }
1052
1053         nvme_qpair_enable(qpair);
1054 }
1055
1056 void
1057 nvme_io_qpair_enable(struct nvme_qpair *qpair)
1058 {
1059         STAILQ_HEAD(, nvme_request)     temp;
1060         struct nvme_tracker             *tr;
1061         struct nvme_tracker             *tr_temp;
1062         struct nvme_request             *req;
1063
1064         /*
1065          * Manually abort each outstanding I/O.  This normally results in a
1066          *  retry, unless the retry count on the associated request has
1067          *  reached its limit.
1068          */
1069         TAILQ_FOREACH_SAFE(tr, &qpair->outstanding_tr, tailq, tr_temp) {
1070                 nvme_printf(qpair->ctrlr, "aborting outstanding i/o\n");
1071                 nvme_qpair_manual_complete_tracker(qpair, tr, NVME_SCT_GENERIC,
1072                     NVME_SC_ABORTED_BY_REQUEST, 0, TRUE);
1073         }
1074
1075         mtx_lock(&qpair->lock);
1076
1077         nvme_qpair_enable(qpair);
1078
1079         STAILQ_INIT(&temp);
1080         STAILQ_SWAP(&qpair->queued_req, &temp, nvme_request);
1081
1082         while (!STAILQ_EMPTY(&temp)) {
1083                 req = STAILQ_FIRST(&temp);
1084                 STAILQ_REMOVE_HEAD(&temp, stailq);
1085                 nvme_printf(qpair->ctrlr, "resubmitting queued i/o\n");
1086                 nvme_qpair_print_command(qpair, &req->cmd);
1087                 _nvme_qpair_submit_request(qpair, req);
1088         }
1089
1090         mtx_unlock(&qpair->lock);
1091 }
1092
1093 static void
1094 nvme_qpair_disable(struct nvme_qpair *qpair)
1095 {
1096         struct nvme_tracker *tr;
1097
1098         qpair->is_enabled = FALSE;
1099         mtx_lock(&qpair->lock);
1100         TAILQ_FOREACH(tr, &qpair->outstanding_tr, tailq)
1101                 callout_stop(&tr->timer);
1102         mtx_unlock(&qpair->lock);
1103 }
1104
1105 void
1106 nvme_admin_qpair_disable(struct nvme_qpair *qpair)
1107 {
1108
1109         nvme_qpair_disable(qpair);
1110         nvme_admin_qpair_abort_aers(qpair);
1111 }
1112
1113 void
1114 nvme_io_qpair_disable(struct nvme_qpair *qpair)
1115 {
1116
1117         nvme_qpair_disable(qpair);
1118 }
1119
1120 void
1121 nvme_qpair_fail(struct nvme_qpair *qpair)
1122 {
1123         struct nvme_tracker             *tr;
1124         struct nvme_request             *req;
1125
1126         if (!mtx_initialized(&qpair->lock))
1127                 return;
1128
1129         mtx_lock(&qpair->lock);
1130
1131         while (!STAILQ_EMPTY(&qpair->queued_req)) {
1132                 req = STAILQ_FIRST(&qpair->queued_req);
1133                 STAILQ_REMOVE_HEAD(&qpair->queued_req, stailq);
1134                 nvme_printf(qpair->ctrlr, "failing queued i/o\n");
1135                 mtx_unlock(&qpair->lock);
1136                 nvme_qpair_manual_complete_request(qpair, req, NVME_SCT_GENERIC,
1137                     NVME_SC_ABORTED_BY_REQUEST, TRUE);
1138                 mtx_lock(&qpair->lock);
1139         }
1140
1141         /* Manually abort each outstanding I/O. */
1142         while (!TAILQ_EMPTY(&qpair->outstanding_tr)) {
1143                 tr = TAILQ_FIRST(&qpair->outstanding_tr);
1144                 /*
1145                  * Do not remove the tracker.  The abort_tracker path will
1146                  *  do that for us.
1147                  */
1148                 nvme_printf(qpair->ctrlr, "failing outstanding i/o\n");
1149                 mtx_unlock(&qpair->lock);
1150                 nvme_qpair_manual_complete_tracker(qpair, tr, NVME_SCT_GENERIC,
1151                     NVME_SC_ABORTED_BY_REQUEST, 1 /* do not retry */, TRUE);
1152                 mtx_lock(&qpair->lock);
1153         }
1154
1155         mtx_unlock(&qpair->lock);
1156 }
1157