2 * Copyright (C) 2012-2014 Intel Corporation
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
30 #include <sys/param.h>
33 #include <dev/pci/pcivar.h>
35 #include "nvme_private.h"
37 static void _nvme_qpair_submit_request(struct nvme_qpair *qpair,
38 struct nvme_request *req);
40 struct nvme_opcode_string {
46 static struct nvme_opcode_string admin_opcode[] = {
47 { NVME_OPC_DELETE_IO_SQ, "DELETE IO SQ" },
48 { NVME_OPC_CREATE_IO_SQ, "CREATE IO SQ" },
49 { NVME_OPC_GET_LOG_PAGE, "GET LOG PAGE" },
50 { NVME_OPC_DELETE_IO_CQ, "DELETE IO CQ" },
51 { NVME_OPC_CREATE_IO_CQ, "CREATE IO CQ" },
52 { NVME_OPC_IDENTIFY, "IDENTIFY" },
53 { NVME_OPC_ABORT, "ABORT" },
54 { NVME_OPC_SET_FEATURES, "SET FEATURES" },
55 { NVME_OPC_GET_FEATURES, "GET FEATURES" },
56 { NVME_OPC_ASYNC_EVENT_REQUEST, "ASYNC EVENT REQUEST" },
57 { NVME_OPC_FIRMWARE_ACTIVATE, "FIRMWARE ACTIVATE" },
58 { NVME_OPC_FIRMWARE_IMAGE_DOWNLOAD, "FIRMWARE IMAGE DOWNLOAD" },
59 { NVME_OPC_FORMAT_NVM, "FORMAT NVM" },
60 { NVME_OPC_SECURITY_SEND, "SECURITY SEND" },
61 { NVME_OPC_SECURITY_RECEIVE, "SECURITY RECEIVE" },
62 { 0xFFFF, "ADMIN COMMAND" }
65 static struct nvme_opcode_string io_opcode[] = {
66 { NVME_OPC_FLUSH, "FLUSH" },
67 { NVME_OPC_WRITE, "WRITE" },
68 { NVME_OPC_READ, "READ" },
69 { NVME_OPC_WRITE_UNCORRECTABLE, "WRITE UNCORRECTABLE" },
70 { NVME_OPC_COMPARE, "COMPARE" },
71 { NVME_OPC_DATASET_MANAGEMENT, "DATASET MANAGEMENT" },
72 { 0xFFFF, "IO COMMAND" }
76 get_admin_opcode_string(uint16_t opc)
78 struct nvme_opcode_string *entry;
82 while (entry->opc != 0xFFFF) {
83 if (entry->opc == opc)
91 get_io_opcode_string(uint16_t opc)
93 struct nvme_opcode_string *entry;
97 while (entry->opc != 0xFFFF) {
98 if (entry->opc == opc)
107 nvme_admin_qpair_print_command(struct nvme_qpair *qpair,
108 struct nvme_command *cmd)
111 nvme_printf(qpair->ctrlr, "%s (%02x) sqid:%d cid:%d nsid:%x "
112 "cdw10:%08x cdw11:%08x\n",
113 get_admin_opcode_string(cmd->opc), cmd->opc, qpair->id, cmd->cid,
114 cmd->nsid, cmd->cdw10, cmd->cdw11);
118 nvme_io_qpair_print_command(struct nvme_qpair *qpair,
119 struct nvme_command *cmd)
125 case NVME_OPC_WRITE_UNCORRECTABLE:
126 case NVME_OPC_COMPARE:
127 nvme_printf(qpair->ctrlr, "%s sqid:%d cid:%d nsid:%d "
129 get_io_opcode_string(cmd->opc), qpair->id, cmd->cid,
131 ((unsigned long long)cmd->cdw11 << 32) + cmd->cdw10,
132 (cmd->cdw12 & 0xFFFF) + 1);
135 case NVME_OPC_DATASET_MANAGEMENT:
136 nvme_printf(qpair->ctrlr, "%s sqid:%d cid:%d nsid:%d\n",
137 get_io_opcode_string(cmd->opc), qpair->id, cmd->cid,
141 nvme_printf(qpair->ctrlr, "%s (%02x) sqid:%d cid:%d nsid:%d\n",
142 get_io_opcode_string(cmd->opc), cmd->opc, qpair->id,
143 cmd->cid, cmd->nsid);
149 nvme_qpair_print_command(struct nvme_qpair *qpair, struct nvme_command *cmd)
152 nvme_admin_qpair_print_command(qpair, cmd);
154 nvme_io_qpair_print_command(qpair, cmd);
157 struct nvme_status_string {
163 static struct nvme_status_string generic_status[] = {
164 { NVME_SC_SUCCESS, "SUCCESS" },
165 { NVME_SC_INVALID_OPCODE, "INVALID OPCODE" },
166 { NVME_SC_INVALID_FIELD, "INVALID_FIELD" },
167 { NVME_SC_COMMAND_ID_CONFLICT, "COMMAND ID CONFLICT" },
168 { NVME_SC_DATA_TRANSFER_ERROR, "DATA TRANSFER ERROR" },
169 { NVME_SC_ABORTED_POWER_LOSS, "ABORTED - POWER LOSS" },
170 { NVME_SC_INTERNAL_DEVICE_ERROR, "INTERNAL DEVICE ERROR" },
171 { NVME_SC_ABORTED_BY_REQUEST, "ABORTED - BY REQUEST" },
172 { NVME_SC_ABORTED_SQ_DELETION, "ABORTED - SQ DELETION" },
173 { NVME_SC_ABORTED_FAILED_FUSED, "ABORTED - FAILED FUSED" },
174 { NVME_SC_ABORTED_MISSING_FUSED, "ABORTED - MISSING FUSED" },
175 { NVME_SC_INVALID_NAMESPACE_OR_FORMAT, "INVALID NAMESPACE OR FORMAT" },
176 { NVME_SC_COMMAND_SEQUENCE_ERROR, "COMMAND SEQUENCE ERROR" },
177 { NVME_SC_LBA_OUT_OF_RANGE, "LBA OUT OF RANGE" },
178 { NVME_SC_CAPACITY_EXCEEDED, "CAPACITY EXCEEDED" },
179 { NVME_SC_NAMESPACE_NOT_READY, "NAMESPACE NOT READY" },
180 { 0xFFFF, "GENERIC" }
183 static struct nvme_status_string command_specific_status[] = {
184 { NVME_SC_COMPLETION_QUEUE_INVALID, "INVALID COMPLETION QUEUE" },
185 { NVME_SC_INVALID_QUEUE_IDENTIFIER, "INVALID QUEUE IDENTIFIER" },
186 { NVME_SC_MAXIMUM_QUEUE_SIZE_EXCEEDED, "MAX QUEUE SIZE EXCEEDED" },
187 { NVME_SC_ABORT_COMMAND_LIMIT_EXCEEDED, "ABORT CMD LIMIT EXCEEDED" },
188 { NVME_SC_ASYNC_EVENT_REQUEST_LIMIT_EXCEEDED, "ASYNC LIMIT EXCEEDED" },
189 { NVME_SC_INVALID_FIRMWARE_SLOT, "INVALID FIRMWARE SLOT" },
190 { NVME_SC_INVALID_FIRMWARE_IMAGE, "INVALID FIRMWARE IMAGE" },
191 { NVME_SC_INVALID_INTERRUPT_VECTOR, "INVALID INTERRUPT VECTOR" },
192 { NVME_SC_INVALID_LOG_PAGE, "INVALID LOG PAGE" },
193 { NVME_SC_INVALID_FORMAT, "INVALID FORMAT" },
194 { NVME_SC_FIRMWARE_REQUIRES_RESET, "FIRMWARE REQUIRES RESET" },
195 { NVME_SC_CONFLICTING_ATTRIBUTES, "CONFLICTING ATTRIBUTES" },
196 { NVME_SC_INVALID_PROTECTION_INFO, "INVALID PROTECTION INFO" },
197 { NVME_SC_ATTEMPTED_WRITE_TO_RO_PAGE, "WRITE TO RO PAGE" },
198 { 0xFFFF, "COMMAND SPECIFIC" }
201 static struct nvme_status_string media_error_status[] = {
202 { NVME_SC_WRITE_FAULTS, "WRITE FAULTS" },
203 { NVME_SC_UNRECOVERED_READ_ERROR, "UNRECOVERED READ ERROR" },
204 { NVME_SC_GUARD_CHECK_ERROR, "GUARD CHECK ERROR" },
205 { NVME_SC_APPLICATION_TAG_CHECK_ERROR, "APPLICATION TAG CHECK ERROR" },
206 { NVME_SC_REFERENCE_TAG_CHECK_ERROR, "REFERENCE TAG CHECK ERROR" },
207 { NVME_SC_COMPARE_FAILURE, "COMPARE FAILURE" },
208 { NVME_SC_ACCESS_DENIED, "ACCESS DENIED" },
209 { 0xFFFF, "MEDIA ERROR" }
213 get_status_string(uint16_t sct, uint16_t sc)
215 struct nvme_status_string *entry;
218 case NVME_SCT_GENERIC:
219 entry = generic_status;
221 case NVME_SCT_COMMAND_SPECIFIC:
222 entry = command_specific_status;
224 case NVME_SCT_MEDIA_ERROR:
225 entry = media_error_status;
227 case NVME_SCT_VENDOR_SPECIFIC:
228 return ("VENDOR SPECIFIC");
233 while (entry->sc != 0xFFFF) {
242 nvme_qpair_print_completion(struct nvme_qpair *qpair,
243 struct nvme_completion *cpl)
245 nvme_printf(qpair->ctrlr, "%s (%02x/%02x) sqid:%d cid:%d cdw0:%x\n",
246 get_status_string(cpl->status.sct, cpl->status.sc),
247 cpl->status.sct, cpl->status.sc, cpl->sqid, cpl->cid, cpl->cdw0);
251 nvme_completion_is_retry(const struct nvme_completion *cpl)
254 * TODO: spec is not clear how commands that are aborted due
255 * to TLER will be marked. So for now, it seems
256 * NAMESPACE_NOT_READY is the only case where we should
257 * look at the DNR bit.
259 switch (cpl->status.sct) {
260 case NVME_SCT_GENERIC:
261 switch (cpl->status.sc) {
262 case NVME_SC_ABORTED_BY_REQUEST:
263 case NVME_SC_NAMESPACE_NOT_READY:
268 case NVME_SC_INVALID_OPCODE:
269 case NVME_SC_INVALID_FIELD:
270 case NVME_SC_COMMAND_ID_CONFLICT:
271 case NVME_SC_DATA_TRANSFER_ERROR:
272 case NVME_SC_ABORTED_POWER_LOSS:
273 case NVME_SC_INTERNAL_DEVICE_ERROR:
274 case NVME_SC_ABORTED_SQ_DELETION:
275 case NVME_SC_ABORTED_FAILED_FUSED:
276 case NVME_SC_ABORTED_MISSING_FUSED:
277 case NVME_SC_INVALID_NAMESPACE_OR_FORMAT:
278 case NVME_SC_COMMAND_SEQUENCE_ERROR:
279 case NVME_SC_LBA_OUT_OF_RANGE:
280 case NVME_SC_CAPACITY_EXCEEDED:
284 case NVME_SCT_COMMAND_SPECIFIC:
285 case NVME_SCT_MEDIA_ERROR:
286 case NVME_SCT_VENDOR_SPECIFIC:
293 nvme_qpair_construct_tracker(struct nvme_qpair *qpair, struct nvme_tracker *tr,
297 bus_dmamap_create(qpair->dma_tag_payload, 0, &tr->payload_dma_map);
298 bus_dmamap_create(qpair->dma_tag, 0, &tr->prp_dma_map);
300 bus_dmamap_load(qpair->dma_tag, tr->prp_dma_map, tr->prp,
301 sizeof(tr->prp), nvme_single_map, &tr->prp_bus_addr, 0);
303 callout_init(&tr->timer, 1);
309 nvme_qpair_complete_tracker(struct nvme_qpair *qpair, struct nvme_tracker *tr,
310 struct nvme_completion *cpl, boolean_t print_on_error)
312 struct nvme_request *req;
313 boolean_t retry, error;
316 error = nvme_completion_is_error(cpl);
317 retry = error && nvme_completion_is_retry(cpl) &&
318 req->retries < nvme_retry_count;
320 if (error && print_on_error) {
321 nvme_qpair_print_command(qpair, &req->cmd);
322 nvme_qpair_print_completion(qpair, cpl);
325 qpair->act_tr[cpl->cid] = NULL;
327 KASSERT(cpl->cid == req->cmd.cid, ("cpl cid does not match cmd cid\n"));
329 if (req->cb_fn && !retry)
330 req->cb_fn(req->cb_arg, cpl);
332 mtx_lock(&qpair->lock);
333 callout_stop(&tr->timer);
337 nvme_qpair_submit_tracker(qpair, tr);
339 if (req->type != NVME_REQUEST_NULL)
340 bus_dmamap_unload(qpair->dma_tag_payload,
341 tr->payload_dma_map);
343 nvme_free_request(req);
346 TAILQ_REMOVE(&qpair->outstanding_tr, tr, tailq);
347 TAILQ_INSERT_HEAD(&qpair->free_tr, tr, tailq);
350 * If the controller is in the middle of resetting, don't
351 * try to submit queued requests here - let the reset logic
352 * handle that instead.
354 if (!STAILQ_EMPTY(&qpair->queued_req) &&
355 !qpair->ctrlr->is_resetting) {
356 req = STAILQ_FIRST(&qpair->queued_req);
357 STAILQ_REMOVE_HEAD(&qpair->queued_req, stailq);
358 _nvme_qpair_submit_request(qpair, req);
362 mtx_unlock(&qpair->lock);
366 nvme_qpair_manual_complete_tracker(struct nvme_qpair *qpair,
367 struct nvme_tracker *tr, uint32_t sct, uint32_t sc, uint32_t dnr,
368 boolean_t print_on_error)
370 struct nvme_completion cpl;
372 memset(&cpl, 0, sizeof(cpl));
373 cpl.sqid = qpair->id;
375 cpl.status.sct = sct;
377 cpl.status.dnr = dnr;
378 nvme_qpair_complete_tracker(qpair, tr, &cpl, print_on_error);
382 nvme_qpair_manual_complete_request(struct nvme_qpair *qpair,
383 struct nvme_request *req, uint32_t sct, uint32_t sc,
384 boolean_t print_on_error)
386 struct nvme_completion cpl;
389 memset(&cpl, 0, sizeof(cpl));
390 cpl.sqid = qpair->id;
391 cpl.status.sct = sct;
394 error = nvme_completion_is_error(&cpl);
396 if (error && print_on_error) {
397 nvme_qpair_print_command(qpair, &req->cmd);
398 nvme_qpair_print_completion(qpair, &cpl);
402 req->cb_fn(req->cb_arg, &cpl);
404 nvme_free_request(req);
408 nvme_qpair_process_completions(struct nvme_qpair *qpair)
410 struct nvme_tracker *tr;
411 struct nvme_completion *cpl;
413 qpair->num_intr_handler_calls++;
415 if (!qpair->is_enabled)
417 * qpair is not enabled, likely because a controller reset is
418 * is in progress. Ignore the interrupt - any I/O that was
419 * associated with this interrupt will get retried when the
425 cpl = &qpair->cpl[qpair->cq_head];
427 if (cpl->status.p != qpair->phase)
430 tr = qpair->act_tr[cpl->cid];
433 nvme_qpair_complete_tracker(qpair, tr, cpl, TRUE);
434 qpair->sq_head = cpl->sqhd;
436 nvme_printf(qpair->ctrlr,
437 "cpl does not map to outstanding cmd\n");
438 nvme_dump_completion(cpl);
439 KASSERT(0, ("received completion for unknown cmd\n"));
442 if (++qpair->cq_head == qpair->num_entries) {
444 qpair->phase = !qpair->phase;
447 nvme_mmio_write_4(qpair->ctrlr, doorbell[qpair->id].cq_hdbl,
453 nvme_qpair_msix_handler(void *arg)
455 struct nvme_qpair *qpair = arg;
457 nvme_qpair_process_completions(qpair);
461 nvme_qpair_construct(struct nvme_qpair *qpair, uint32_t id,
462 uint16_t vector, uint32_t num_entries, uint32_t num_trackers,
463 struct nvme_controller *ctrlr)
465 struct nvme_tracker *tr;
470 qpair->vector = vector;
471 qpair->num_entries = num_entries;
474 * Chatham prototype board starts having issues at higher queue
475 * depths. So use a conservative estimate here of no more than 64
476 * outstanding I/O per queue at any one point.
478 if (pci_get_devid(ctrlr->dev) == CHATHAM_PCI_ID)
479 num_trackers = min(num_trackers, 64);
481 qpair->num_trackers = num_trackers;
482 qpair->ctrlr = ctrlr;
484 if (ctrlr->msix_enabled) {
487 * MSI-X vector resource IDs start at 1, so we add one to
488 * the queue's vector to get the corresponding rid to use.
490 qpair->rid = vector + 1;
491 qpair->res = ctrlr->msi_res[vector];
493 bus_setup_intr(ctrlr->dev, qpair->res,
494 INTR_TYPE_MISC | INTR_MPSAFE, NULL,
495 nvme_qpair_msix_handler, qpair, &qpair->tag);
498 mtx_init(&qpair->lock, "nvme qpair lock", NULL, MTX_DEF);
500 /* Note: NVMe PRP format is restricted to 4-byte alignment. */
501 err = bus_dma_tag_create(bus_get_dma_tag(ctrlr->dev),
502 4, PAGE_SIZE, BUS_SPACE_MAXADDR,
503 BUS_SPACE_MAXADDR, NULL, NULL, NVME_MAX_XFER_SIZE,
504 (NVME_MAX_XFER_SIZE/PAGE_SIZE)+1, PAGE_SIZE, 0,
505 NULL, NULL, &qpair->dma_tag_payload);
507 nvme_printf(ctrlr, "payload tag create failed %d\n", err);
509 err = bus_dma_tag_create(bus_get_dma_tag(ctrlr->dev),
510 4, 0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
511 BUS_SPACE_MAXSIZE, 1, BUS_SPACE_MAXSIZE, 0,
512 NULL, NULL, &qpair->dma_tag);
514 nvme_printf(ctrlr, "tag create failed %d\n", err);
517 qpair->num_intr_handler_calls = 0;
519 qpair->cmd = contigmalloc(qpair->num_entries *
520 sizeof(struct nvme_command), M_NVME, M_ZERO,
521 0, BUS_SPACE_MAXADDR, PAGE_SIZE, 0);
522 qpair->cpl = contigmalloc(qpair->num_entries *
523 sizeof(struct nvme_completion), M_NVME, M_ZERO,
524 0, BUS_SPACE_MAXADDR, PAGE_SIZE, 0);
526 err = bus_dmamap_create(qpair->dma_tag, 0, &qpair->cmd_dma_map);
528 nvme_printf(ctrlr, "cmd_dma_map create failed %d\n", err);
530 err = bus_dmamap_create(qpair->dma_tag, 0, &qpair->cpl_dma_map);
532 nvme_printf(ctrlr, "cpl_dma_map create failed %d\n", err);
534 bus_dmamap_load(qpair->dma_tag, qpair->cmd_dma_map,
535 qpair->cmd, qpair->num_entries * sizeof(struct nvme_command),
536 nvme_single_map, &qpair->cmd_bus_addr, 0);
537 bus_dmamap_load(qpair->dma_tag, qpair->cpl_dma_map,
538 qpair->cpl, qpair->num_entries * sizeof(struct nvme_completion),
539 nvme_single_map, &qpair->cpl_bus_addr, 0);
541 qpair->sq_tdbl_off = nvme_mmio_offsetof(doorbell[id].sq_tdbl);
542 qpair->cq_hdbl_off = nvme_mmio_offsetof(doorbell[id].cq_hdbl);
544 TAILQ_INIT(&qpair->free_tr);
545 TAILQ_INIT(&qpair->outstanding_tr);
546 STAILQ_INIT(&qpair->queued_req);
548 for (i = 0; i < qpair->num_trackers; i++) {
549 tr = malloc(sizeof(*tr), M_NVME, M_ZERO | M_WAITOK);
550 nvme_qpair_construct_tracker(qpair, tr, i);
551 TAILQ_INSERT_HEAD(&qpair->free_tr, tr, tailq);
554 qpair->act_tr = malloc(sizeof(struct nvme_tracker *) * qpair->num_entries,
555 M_NVME, M_ZERO | M_WAITOK);
559 nvme_qpair_destroy(struct nvme_qpair *qpair)
561 struct nvme_tracker *tr;
564 bus_teardown_intr(qpair->ctrlr->dev, qpair->res, qpair->tag);
567 bus_release_resource(qpair->ctrlr->dev, SYS_RES_IRQ,
568 rman_get_rid(qpair->res), qpair->res);
571 bus_dmamap_unload(qpair->dma_tag, qpair->cmd_dma_map);
572 bus_dmamap_destroy(qpair->dma_tag, qpair->cmd_dma_map);
573 contigfree(qpair->cmd,
574 qpair->num_entries * sizeof(struct nvme_command), M_NVME);
578 bus_dmamap_unload(qpair->dma_tag, qpair->cpl_dma_map);
579 bus_dmamap_destroy(qpair->dma_tag, qpair->cpl_dma_map);
580 contigfree(qpair->cpl,
581 qpair->num_entries * sizeof(struct nvme_completion),
586 bus_dma_tag_destroy(qpair->dma_tag);
588 if (qpair->dma_tag_payload)
589 bus_dma_tag_destroy(qpair->dma_tag_payload);
592 free(qpair->act_tr, M_NVME);
594 while (!TAILQ_EMPTY(&qpair->free_tr)) {
595 tr = TAILQ_FIRST(&qpair->free_tr);
596 TAILQ_REMOVE(&qpair->free_tr, tr, tailq);
597 bus_dmamap_destroy(qpair->dma_tag, tr->payload_dma_map);
598 bus_dmamap_destroy(qpair->dma_tag, tr->prp_dma_map);
604 nvme_admin_qpair_abort_aers(struct nvme_qpair *qpair)
606 struct nvme_tracker *tr;
608 tr = TAILQ_FIRST(&qpair->outstanding_tr);
610 if (tr->req->cmd.opc == NVME_OPC_ASYNC_EVENT_REQUEST) {
611 nvme_qpair_manual_complete_tracker(qpair, tr,
612 NVME_SCT_GENERIC, NVME_SC_ABORTED_SQ_DELETION, 0,
614 tr = TAILQ_FIRST(&qpair->outstanding_tr);
616 tr = TAILQ_NEXT(tr, tailq);
622 nvme_admin_qpair_destroy(struct nvme_qpair *qpair)
625 nvme_admin_qpair_abort_aers(qpair);
626 nvme_qpair_destroy(qpair);
630 nvme_io_qpair_destroy(struct nvme_qpair *qpair)
633 nvme_qpair_destroy(qpair);
637 nvme_abort_complete(void *arg, const struct nvme_completion *status)
639 struct nvme_tracker *tr = arg;
642 * If cdw0 == 1, the controller was not able to abort the command
643 * we requested. We still need to check the active tracker array,
644 * to cover race where I/O timed out at same time controller was
645 * completing the I/O.
647 if (status->cdw0 == 1 && tr->qpair->act_tr[tr->cid] != NULL) {
649 * An I/O has timed out, and the controller was unable to
650 * abort it for some reason. Construct a fake completion
651 * status, and then complete the I/O's tracker manually.
653 nvme_printf(tr->qpair->ctrlr,
654 "abort command failed, aborting command manually\n");
655 nvme_qpair_manual_complete_tracker(tr->qpair, tr,
656 NVME_SCT_GENERIC, NVME_SC_ABORTED_BY_REQUEST, 0, TRUE);
661 nvme_timeout(void *arg)
663 struct nvme_tracker *tr = arg;
664 struct nvme_qpair *qpair = tr->qpair;
665 struct nvme_controller *ctrlr = qpair->ctrlr;
666 union csts_register csts;
668 /* Read csts to get value of cfs - controller fatal status. */
669 csts.raw = nvme_mmio_read_4(ctrlr, csts);
671 if (ctrlr->enable_aborts && csts.bits.cfs == 0) {
673 * If aborts are enabled, only use them if the controller is
674 * not reporting fatal status.
676 nvme_ctrlr_cmd_abort(ctrlr, tr->cid, qpair->id,
677 nvme_abort_complete, tr);
679 nvme_ctrlr_reset(ctrlr);
683 nvme_qpair_submit_tracker(struct nvme_qpair *qpair, struct nvme_tracker *tr)
685 struct nvme_request *req;
686 struct nvme_controller *ctrlr;
688 mtx_assert(&qpair->lock, MA_OWNED);
691 req->cmd.cid = tr->cid;
692 qpair->act_tr[tr->cid] = tr;
693 ctrlr = qpair->ctrlr;
696 #if __FreeBSD_version >= 800030
697 callout_reset_curcpu(&tr->timer, ctrlr->timeout_period * hz,
700 callout_reset(&tr->timer, ctrlr->timeout_period * hz,
704 /* Copy the command from the tracker to the submission queue. */
705 memcpy(&qpair->cmd[qpair->sq_tail], &req->cmd, sizeof(req->cmd));
707 if (++qpair->sq_tail == qpair->num_entries)
711 nvme_mmio_write_4(qpair->ctrlr, doorbell[qpair->id].sq_tdbl,
718 nvme_payload_map(void *arg, bus_dma_segment_t *seg, int nseg, int error)
720 struct nvme_tracker *tr = arg;
724 * If the mapping operation failed, return immediately. The caller
725 * is responsible for detecting the error status and failing the
729 nvme_printf(tr->qpair->ctrlr,
730 "nvme_payload_map err %d\n", error);
735 * Note that we specified PAGE_SIZE for alignment and max
736 * segment size when creating the bus dma tags. So here
737 * we can safely just transfer each segment to its
738 * associated PRP entry.
740 tr->req->cmd.prp1 = seg[0].ds_addr;
743 tr->req->cmd.prp2 = seg[1].ds_addr;
744 } else if (nseg > 2) {
746 tr->req->cmd.prp2 = (uint64_t)tr->prp_bus_addr;
747 while (cur_nseg < nseg) {
748 tr->prp[cur_nseg-1] =
749 (uint64_t)seg[cur_nseg].ds_addr;
754 * prp2 should not be used by the controller
755 * since there is only one segment, but set
756 * to 0 just to be safe.
758 tr->req->cmd.prp2 = 0;
761 nvme_qpair_submit_tracker(tr->qpair, tr);
765 _nvme_qpair_submit_request(struct nvme_qpair *qpair, struct nvme_request *req)
767 struct nvme_tracker *tr;
770 mtx_assert(&qpair->lock, MA_OWNED);
772 tr = TAILQ_FIRST(&qpair->free_tr);
775 if (tr == NULL || !qpair->is_enabled) {
777 * No tracker is available, or the qpair is disabled due to
778 * an in-progress controller-level reset or controller
782 if (qpair->ctrlr->is_failed) {
784 * The controller has failed. Post the request to a
785 * task where it will be aborted, so that we do not
786 * invoke the request's callback in the context
789 nvme_ctrlr_post_failed_request(qpair->ctrlr, req);
792 * Put the request on the qpair's request queue to be
793 * processed when a tracker frees up via a command
794 * completion or when the controller reset is
797 STAILQ_INSERT_TAIL(&qpair->queued_req, req, stailq);
802 TAILQ_REMOVE(&qpair->free_tr, tr, tailq);
803 TAILQ_INSERT_TAIL(&qpair->outstanding_tr, tr, tailq);
807 case NVME_REQUEST_VADDR:
808 KASSERT(req->payload_size <= qpair->ctrlr->max_xfer_size,
809 ("payload_size (%d) exceeds max_xfer_size (%d)\n",
810 req->payload_size, qpair->ctrlr->max_xfer_size));
811 err = bus_dmamap_load(tr->qpair->dma_tag_payload,
812 tr->payload_dma_map, req->u.payload, req->payload_size,
813 nvme_payload_map, tr, 0);
815 nvme_printf(qpair->ctrlr,
816 "bus_dmamap_load returned 0x%x!\n", err);
818 case NVME_REQUEST_NULL:
819 nvme_qpair_submit_tracker(tr->qpair, tr);
821 #ifdef NVME_UNMAPPED_BIO_SUPPORT
822 case NVME_REQUEST_BIO:
823 KASSERT(req->u.bio->bio_bcount <= qpair->ctrlr->max_xfer_size,
824 ("bio->bio_bcount (%jd) exceeds max_xfer_size (%d)\n",
825 (intmax_t)req->u.bio->bio_bcount,
826 qpair->ctrlr->max_xfer_size));
827 err = bus_dmamap_load_bio(tr->qpair->dma_tag_payload,
828 tr->payload_dma_map, req->u.bio, nvme_payload_map, tr, 0);
830 nvme_printf(qpair->ctrlr,
831 "bus_dmamap_load_bio returned 0x%x!\n", err);
835 panic("unknown nvme request type 0x%x\n", req->type);
841 * The dmamap operation failed, so we manually fail the
842 * tracker here with DATA_TRANSFER_ERROR status.
844 * nvme_qpair_manual_complete_tracker must not be called
845 * with the qpair lock held.
847 mtx_unlock(&qpair->lock);
848 nvme_qpair_manual_complete_tracker(qpair, tr, NVME_SCT_GENERIC,
849 NVME_SC_DATA_TRANSFER_ERROR, 1 /* do not retry */, TRUE);
850 mtx_lock(&qpair->lock);
855 nvme_qpair_submit_request(struct nvme_qpair *qpair, struct nvme_request *req)
858 mtx_lock(&qpair->lock);
859 _nvme_qpair_submit_request(qpair, req);
860 mtx_unlock(&qpair->lock);
864 nvme_qpair_enable(struct nvme_qpair *qpair)
867 qpair->is_enabled = TRUE;
871 nvme_qpair_reset(struct nvme_qpair *qpair)
874 qpair->sq_head = qpair->sq_tail = qpair->cq_head = 0;
877 * First time through the completion queue, HW will set phase
878 * bit on completions to 1. So set this to 1 here, indicating
879 * we're looking for a 1 to know which entries have completed.
880 * we'll toggle the bit each time when the completion queue
885 memset(qpair->cmd, 0,
886 qpair->num_entries * sizeof(struct nvme_command));
887 memset(qpair->cpl, 0,
888 qpair->num_entries * sizeof(struct nvme_completion));
892 nvme_admin_qpair_enable(struct nvme_qpair *qpair)
894 struct nvme_tracker *tr;
895 struct nvme_tracker *tr_temp;
898 * Manually abort each outstanding admin command. Do not retry
899 * admin commands found here, since they will be left over from
900 * a controller reset and its likely the context in which the
901 * command was issued no longer applies.
903 TAILQ_FOREACH_SAFE(tr, &qpair->outstanding_tr, tailq, tr_temp) {
904 nvme_printf(qpair->ctrlr,
905 "aborting outstanding admin command\n");
906 nvme_qpair_manual_complete_tracker(qpair, tr, NVME_SCT_GENERIC,
907 NVME_SC_ABORTED_BY_REQUEST, 1 /* do not retry */, TRUE);
910 nvme_qpair_enable(qpair);
914 nvme_io_qpair_enable(struct nvme_qpair *qpair)
916 STAILQ_HEAD(, nvme_request) temp;
917 struct nvme_tracker *tr;
918 struct nvme_tracker *tr_temp;
919 struct nvme_request *req;
922 * Manually abort each outstanding I/O. This normally results in a
923 * retry, unless the retry count on the associated request has
926 TAILQ_FOREACH_SAFE(tr, &qpair->outstanding_tr, tailq, tr_temp) {
927 nvme_printf(qpair->ctrlr, "aborting outstanding i/o\n");
928 nvme_qpair_manual_complete_tracker(qpair, tr, NVME_SCT_GENERIC,
929 NVME_SC_ABORTED_BY_REQUEST, 0, TRUE);
932 mtx_lock(&qpair->lock);
934 nvme_qpair_enable(qpair);
937 STAILQ_SWAP(&qpair->queued_req, &temp, nvme_request);
939 while (!STAILQ_EMPTY(&temp)) {
940 req = STAILQ_FIRST(&temp);
941 STAILQ_REMOVE_HEAD(&temp, stailq);
942 nvme_printf(qpair->ctrlr, "resubmitting queued i/o\n");
943 nvme_qpair_print_command(qpair, &req->cmd);
944 _nvme_qpair_submit_request(qpair, req);
947 mtx_unlock(&qpair->lock);
951 nvme_qpair_disable(struct nvme_qpair *qpair)
953 struct nvme_tracker *tr;
955 qpair->is_enabled = FALSE;
956 mtx_lock(&qpair->lock);
957 TAILQ_FOREACH(tr, &qpair->outstanding_tr, tailq)
958 callout_stop(&tr->timer);
959 mtx_unlock(&qpair->lock);
963 nvme_admin_qpair_disable(struct nvme_qpair *qpair)
966 nvme_qpair_disable(qpair);
967 nvme_admin_qpair_abort_aers(qpair);
971 nvme_io_qpair_disable(struct nvme_qpair *qpair)
974 nvme_qpair_disable(qpair);
978 nvme_qpair_fail(struct nvme_qpair *qpair)
980 struct nvme_tracker *tr;
981 struct nvme_request *req;
983 mtx_lock(&qpair->lock);
985 while (!STAILQ_EMPTY(&qpair->queued_req)) {
986 req = STAILQ_FIRST(&qpair->queued_req);
987 STAILQ_REMOVE_HEAD(&qpair->queued_req, stailq);
988 nvme_printf(qpair->ctrlr, "failing queued i/o\n");
989 mtx_unlock(&qpair->lock);
990 nvme_qpair_manual_complete_request(qpair, req, NVME_SCT_GENERIC,
991 NVME_SC_ABORTED_BY_REQUEST, TRUE);
992 mtx_lock(&qpair->lock);
995 /* Manually abort each outstanding I/O. */
996 while (!TAILQ_EMPTY(&qpair->outstanding_tr)) {
997 tr = TAILQ_FIRST(&qpair->outstanding_tr);
999 * Do not remove the tracker. The abort_tracker path will
1002 nvme_printf(qpair->ctrlr, "failing outstanding i/o\n");
1003 mtx_unlock(&qpair->lock);
1004 nvme_qpair_manual_complete_tracker(qpair, tr, NVME_SCT_GENERIC,
1005 NVME_SC_ABORTED_BY_REQUEST, 1 /* do not retry */, TRUE);
1006 mtx_lock(&qpair->lock);
1009 mtx_unlock(&qpair->lock);