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 <dev/pci/pcivar.h>
37 #include "nvme_private.h"
39 typedef enum error_print { ERROR_PRINT_NONE, ERROR_PRINT_NO_RETRY, ERROR_PRINT_ALL } error_print_t;
40 #define DO_NOT_RETRY 1
42 static void _nvme_qpair_submit_request(struct nvme_qpair *qpair,
43 struct nvme_request *req);
44 static void nvme_qpair_destroy(struct nvme_qpair *qpair);
46 struct nvme_opcode_string {
52 static struct nvme_opcode_string admin_opcode[] = {
53 { NVME_OPC_DELETE_IO_SQ, "DELETE IO SQ" },
54 { NVME_OPC_CREATE_IO_SQ, "CREATE IO SQ" },
55 { NVME_OPC_GET_LOG_PAGE, "GET LOG PAGE" },
56 { NVME_OPC_DELETE_IO_CQ, "DELETE IO CQ" },
57 { NVME_OPC_CREATE_IO_CQ, "CREATE IO CQ" },
58 { NVME_OPC_IDENTIFY, "IDENTIFY" },
59 { NVME_OPC_ABORT, "ABORT" },
60 { NVME_OPC_SET_FEATURES, "SET FEATURES" },
61 { NVME_OPC_GET_FEATURES, "GET FEATURES" },
62 { NVME_OPC_ASYNC_EVENT_REQUEST, "ASYNC EVENT REQUEST" },
63 { NVME_OPC_FIRMWARE_ACTIVATE, "FIRMWARE ACTIVATE" },
64 { NVME_OPC_FIRMWARE_IMAGE_DOWNLOAD, "FIRMWARE IMAGE DOWNLOAD" },
65 { NVME_OPC_DEVICE_SELF_TEST, "DEVICE SELF-TEST" },
66 { NVME_OPC_NAMESPACE_ATTACHMENT, "NAMESPACE ATTACHMENT" },
67 { NVME_OPC_KEEP_ALIVE, "KEEP ALIVE" },
68 { NVME_OPC_DIRECTIVE_SEND, "DIRECTIVE SEND" },
69 { NVME_OPC_DIRECTIVE_RECEIVE, "DIRECTIVE RECEIVE" },
70 { NVME_OPC_VIRTUALIZATION_MANAGEMENT, "VIRTUALIZATION MANAGEMENT" },
71 { NVME_OPC_NVME_MI_SEND, "NVME-MI SEND" },
72 { NVME_OPC_NVME_MI_RECEIVE, "NVME-MI RECEIVE" },
73 { NVME_OPC_DOORBELL_BUFFER_CONFIG, "DOORBELL BUFFER CONFIG" },
74 { NVME_OPC_FORMAT_NVM, "FORMAT NVM" },
75 { NVME_OPC_SECURITY_SEND, "SECURITY SEND" },
76 { NVME_OPC_SECURITY_RECEIVE, "SECURITY RECEIVE" },
77 { NVME_OPC_SANITIZE, "SANITIZE" },
78 { 0xFFFF, "ADMIN COMMAND" }
81 static struct nvme_opcode_string io_opcode[] = {
82 { NVME_OPC_FLUSH, "FLUSH" },
83 { NVME_OPC_WRITE, "WRITE" },
84 { NVME_OPC_READ, "READ" },
85 { NVME_OPC_WRITE_UNCORRECTABLE, "WRITE UNCORRECTABLE" },
86 { NVME_OPC_COMPARE, "COMPARE" },
87 { NVME_OPC_WRITE_ZEROES, "WRITE ZEROES" },
88 { NVME_OPC_DATASET_MANAGEMENT, "DATASET MANAGEMENT" },
89 { NVME_OPC_RESERVATION_REGISTER, "RESERVATION REGISTER" },
90 { NVME_OPC_RESERVATION_REPORT, "RESERVATION REPORT" },
91 { NVME_OPC_RESERVATION_ACQUIRE, "RESERVATION ACQUIRE" },
92 { NVME_OPC_RESERVATION_RELEASE, "RESERVATION RELEASE" },
93 { 0xFFFF, "IO COMMAND" }
97 get_admin_opcode_string(uint16_t opc)
99 struct nvme_opcode_string *entry;
101 entry = admin_opcode;
103 while (entry->opc != 0xFFFF) {
104 if (entry->opc == opc)
112 get_io_opcode_string(uint16_t opc)
114 struct nvme_opcode_string *entry;
118 while (entry->opc != 0xFFFF) {
119 if (entry->opc == opc)
128 nvme_admin_qpair_print_command(struct nvme_qpair *qpair,
129 struct nvme_command *cmd)
132 nvme_printf(qpair->ctrlr, "%s (%02x) sqid:%d cid:%d nsid:%x "
133 "cdw10:%08x cdw11:%08x\n",
134 get_admin_opcode_string(cmd->opc), cmd->opc, qpair->id, cmd->cid,
135 le32toh(cmd->nsid), le32toh(cmd->cdw10), le32toh(cmd->cdw11));
139 nvme_io_qpair_print_command(struct nvme_qpair *qpair,
140 struct nvme_command *cmd)
146 case NVME_OPC_WRITE_UNCORRECTABLE:
147 case NVME_OPC_COMPARE:
148 case NVME_OPC_WRITE_ZEROES:
149 nvme_printf(qpair->ctrlr, "%s sqid:%d cid:%d nsid:%d "
151 get_io_opcode_string(cmd->opc), qpair->id, cmd->cid, le32toh(cmd->nsid),
152 ((unsigned long long)le32toh(cmd->cdw11) << 32) + le32toh(cmd->cdw10),
153 (le32toh(cmd->cdw12) & 0xFFFF) + 1);
156 case NVME_OPC_DATASET_MANAGEMENT:
157 case NVME_OPC_RESERVATION_REGISTER:
158 case NVME_OPC_RESERVATION_REPORT:
159 case NVME_OPC_RESERVATION_ACQUIRE:
160 case NVME_OPC_RESERVATION_RELEASE:
161 nvme_printf(qpair->ctrlr, "%s sqid:%d cid:%d nsid:%d\n",
162 get_io_opcode_string(cmd->opc), qpair->id, cmd->cid, le32toh(cmd->nsid));
165 nvme_printf(qpair->ctrlr, "%s (%02x) sqid:%d cid:%d nsid:%d\n",
166 get_io_opcode_string(cmd->opc), cmd->opc, qpair->id,
167 cmd->cid, le32toh(cmd->nsid));
173 nvme_qpair_print_command(struct nvme_qpair *qpair, struct nvme_command *cmd)
176 nvme_admin_qpair_print_command(qpair, cmd);
178 nvme_io_qpair_print_command(qpair, cmd);
181 struct nvme_status_string {
187 static struct nvme_status_string generic_status[] = {
188 { NVME_SC_SUCCESS, "SUCCESS" },
189 { NVME_SC_INVALID_OPCODE, "INVALID OPCODE" },
190 { NVME_SC_INVALID_FIELD, "INVALID_FIELD" },
191 { NVME_SC_COMMAND_ID_CONFLICT, "COMMAND ID CONFLICT" },
192 { NVME_SC_DATA_TRANSFER_ERROR, "DATA TRANSFER ERROR" },
193 { NVME_SC_ABORTED_POWER_LOSS, "ABORTED - POWER LOSS" },
194 { NVME_SC_INTERNAL_DEVICE_ERROR, "INTERNAL DEVICE ERROR" },
195 { NVME_SC_ABORTED_BY_REQUEST, "ABORTED - BY REQUEST" },
196 { NVME_SC_ABORTED_SQ_DELETION, "ABORTED - SQ DELETION" },
197 { NVME_SC_ABORTED_FAILED_FUSED, "ABORTED - FAILED FUSED" },
198 { NVME_SC_ABORTED_MISSING_FUSED, "ABORTED - MISSING FUSED" },
199 { NVME_SC_INVALID_NAMESPACE_OR_FORMAT, "INVALID NAMESPACE OR FORMAT" },
200 { NVME_SC_COMMAND_SEQUENCE_ERROR, "COMMAND SEQUENCE ERROR" },
201 { NVME_SC_INVALID_SGL_SEGMENT_DESCR, "INVALID SGL SEGMENT DESCRIPTOR" },
202 { NVME_SC_INVALID_NUMBER_OF_SGL_DESCR, "INVALID NUMBER OF SGL DESCRIPTORS" },
203 { NVME_SC_DATA_SGL_LENGTH_INVALID, "DATA SGL LENGTH INVALID" },
204 { NVME_SC_METADATA_SGL_LENGTH_INVALID, "METADATA SGL LENGTH INVALID" },
205 { NVME_SC_SGL_DESCRIPTOR_TYPE_INVALID, "SGL DESCRIPTOR TYPE INVALID" },
206 { NVME_SC_INVALID_USE_OF_CMB, "INVALID USE OF CONTROLLER MEMORY BUFFER" },
207 { NVME_SC_PRP_OFFET_INVALID, "PRP OFFET INVALID" },
208 { NVME_SC_ATOMIC_WRITE_UNIT_EXCEEDED, "ATOMIC WRITE UNIT EXCEEDED" },
209 { NVME_SC_OPERATION_DENIED, "OPERATION DENIED" },
210 { NVME_SC_SGL_OFFSET_INVALID, "SGL OFFSET INVALID" },
211 { NVME_SC_HOST_ID_INCONSISTENT_FORMAT, "HOST IDENTIFIER INCONSISTENT FORMAT" },
212 { NVME_SC_KEEP_ALIVE_TIMEOUT_EXPIRED, "KEEP ALIVE TIMEOUT EXPIRED" },
213 { NVME_SC_KEEP_ALIVE_TIMEOUT_INVALID, "KEEP ALIVE TIMEOUT INVALID" },
214 { NVME_SC_ABORTED_DUE_TO_PREEMPT, "COMMAND ABORTED DUE TO PREEMPT AND ABORT" },
215 { NVME_SC_SANITIZE_FAILED, "SANITIZE FAILED" },
216 { NVME_SC_SANITIZE_IN_PROGRESS, "SANITIZE IN PROGRESS" },
217 { NVME_SC_SGL_DATA_BLOCK_GRAN_INVALID, "SGL_DATA_BLOCK_GRANULARITY_INVALID" },
218 { NVME_SC_NOT_SUPPORTED_IN_CMB, "COMMAND NOT SUPPORTED FOR QUEUE IN CMB" },
220 { NVME_SC_LBA_OUT_OF_RANGE, "LBA OUT OF RANGE" },
221 { NVME_SC_CAPACITY_EXCEEDED, "CAPACITY EXCEEDED" },
222 { NVME_SC_NAMESPACE_NOT_READY, "NAMESPACE NOT READY" },
223 { NVME_SC_RESERVATION_CONFLICT, "RESERVATION CONFLICT" },
224 { NVME_SC_FORMAT_IN_PROGRESS, "FORMAT IN PROGRESS" },
225 { 0xFFFF, "GENERIC" }
228 static struct nvme_status_string command_specific_status[] = {
229 { NVME_SC_COMPLETION_QUEUE_INVALID, "INVALID COMPLETION QUEUE" },
230 { NVME_SC_INVALID_QUEUE_IDENTIFIER, "INVALID QUEUE IDENTIFIER" },
231 { NVME_SC_MAXIMUM_QUEUE_SIZE_EXCEEDED, "MAX QUEUE SIZE EXCEEDED" },
232 { NVME_SC_ABORT_COMMAND_LIMIT_EXCEEDED, "ABORT CMD LIMIT EXCEEDED" },
233 { NVME_SC_ASYNC_EVENT_REQUEST_LIMIT_EXCEEDED, "ASYNC LIMIT EXCEEDED" },
234 { NVME_SC_INVALID_FIRMWARE_SLOT, "INVALID FIRMWARE SLOT" },
235 { NVME_SC_INVALID_FIRMWARE_IMAGE, "INVALID FIRMWARE IMAGE" },
236 { NVME_SC_INVALID_INTERRUPT_VECTOR, "INVALID INTERRUPT VECTOR" },
237 { NVME_SC_INVALID_LOG_PAGE, "INVALID LOG PAGE" },
238 { NVME_SC_INVALID_FORMAT, "INVALID FORMAT" },
239 { NVME_SC_FIRMWARE_REQUIRES_RESET, "FIRMWARE REQUIRES RESET" },
240 { NVME_SC_INVALID_QUEUE_DELETION, "INVALID QUEUE DELETION" },
241 { NVME_SC_FEATURE_NOT_SAVEABLE, "FEATURE IDENTIFIER NOT SAVEABLE" },
242 { NVME_SC_FEATURE_NOT_CHANGEABLE, "FEATURE NOT CHANGEABLE" },
243 { NVME_SC_FEATURE_NOT_NS_SPECIFIC, "FEATURE NOT NAMESPACE SPECIFIC" },
244 { NVME_SC_FW_ACT_REQUIRES_NVMS_RESET, "FIRMWARE ACTIVATION REQUIRES NVM SUBSYSTEM RESET" },
245 { NVME_SC_FW_ACT_REQUIRES_RESET, "FIRMWARE ACTIVATION REQUIRES RESET" },
246 { NVME_SC_FW_ACT_REQUIRES_TIME, "FIRMWARE ACTIVATION REQUIRES MAXIMUM TIME VIOLATION" },
247 { NVME_SC_FW_ACT_PROHIBITED, "FIRMWARE ACTIVATION PROHIBITED" },
248 { NVME_SC_OVERLAPPING_RANGE, "OVERLAPPING RANGE" },
249 { NVME_SC_NS_INSUFFICIENT_CAPACITY, "NAMESPACE INSUFFICIENT CAPACITY" },
250 { NVME_SC_NS_ID_UNAVAILABLE, "NAMESPACE IDENTIFIER UNAVAILABLE" },
251 { NVME_SC_NS_ALREADY_ATTACHED, "NAMESPACE ALREADY ATTACHED" },
252 { NVME_SC_NS_IS_PRIVATE, "NAMESPACE IS PRIVATE" },
253 { NVME_SC_NS_NOT_ATTACHED, "NS NOT ATTACHED" },
254 { NVME_SC_THIN_PROV_NOT_SUPPORTED, "THIN PROVISIONING NOT SUPPORTED" },
255 { NVME_SC_CTRLR_LIST_INVALID, "CONTROLLER LIST INVALID" },
256 { NVME_SC_SELT_TEST_IN_PROGRESS, "DEVICE SELT-TEST IN PROGRESS" },
257 { NVME_SC_BOOT_PART_WRITE_PROHIB, "BOOT PARTITION WRITE PROHIBITED" },
258 { NVME_SC_INVALID_CTRLR_ID, "INVALID CONTROLLER IDENTIFIER" },
259 { NVME_SC_INVALID_SEC_CTRLR_STATE, "INVALID SECONDARY CONTROLLER STATE" },
260 { NVME_SC_INVALID_NUM_OF_CTRLR_RESRC, "INVALID NUMBER OF CONTROLLER RESOURCES" },
261 { NVME_SC_INVALID_RESOURCE_ID, "INVALID RESOURCE IDENTIFIER" },
263 { NVME_SC_CONFLICTING_ATTRIBUTES, "CONFLICTING ATTRIBUTES" },
264 { NVME_SC_INVALID_PROTECTION_INFO, "INVALID PROTECTION INFO" },
265 { NVME_SC_ATTEMPTED_WRITE_TO_RO_PAGE, "WRITE TO RO PAGE" },
266 { 0xFFFF, "COMMAND SPECIFIC" }
269 static struct nvme_status_string media_error_status[] = {
270 { NVME_SC_WRITE_FAULTS, "WRITE FAULTS" },
271 { NVME_SC_UNRECOVERED_READ_ERROR, "UNRECOVERED READ ERROR" },
272 { NVME_SC_GUARD_CHECK_ERROR, "GUARD CHECK ERROR" },
273 { NVME_SC_APPLICATION_TAG_CHECK_ERROR, "APPLICATION TAG CHECK ERROR" },
274 { NVME_SC_REFERENCE_TAG_CHECK_ERROR, "REFERENCE TAG CHECK ERROR" },
275 { NVME_SC_COMPARE_FAILURE, "COMPARE FAILURE" },
276 { NVME_SC_ACCESS_DENIED, "ACCESS DENIED" },
277 { NVME_SC_DEALLOCATED_OR_UNWRITTEN, "DEALLOCATED OR UNWRITTEN LOGICAL BLOCK" },
278 { 0xFFFF, "MEDIA ERROR" }
282 get_status_string(uint16_t sct, uint16_t sc)
284 struct nvme_status_string *entry;
287 case NVME_SCT_GENERIC:
288 entry = generic_status;
290 case NVME_SCT_COMMAND_SPECIFIC:
291 entry = command_specific_status;
293 case NVME_SCT_MEDIA_ERROR:
294 entry = media_error_status;
296 case NVME_SCT_VENDOR_SPECIFIC:
297 return ("VENDOR SPECIFIC");
302 while (entry->sc != 0xFFFF) {
311 nvme_qpair_print_completion(struct nvme_qpair *qpair,
312 struct nvme_completion *cpl)
316 sct = NVME_STATUS_GET_SCT(cpl->status);
317 sc = NVME_STATUS_GET_SC(cpl->status);
319 nvme_printf(qpair->ctrlr, "%s (%02x/%02x) sqid:%d cid:%d cdw0:%x\n",
320 get_status_string(sct, sc), sct, sc, cpl->sqid, cpl->cid,
325 nvme_completion_is_retry(const struct nvme_completion *cpl)
327 uint8_t sct, sc, dnr;
329 sct = NVME_STATUS_GET_SCT(cpl->status);
330 sc = NVME_STATUS_GET_SC(cpl->status);
331 dnr = NVME_STATUS_GET_DNR(cpl->status); /* Do Not Retry Bit */
334 * TODO: spec is not clear how commands that are aborted due
335 * to TLER will be marked. So for now, it seems
336 * NAMESPACE_NOT_READY is the only case where we should
337 * look at the DNR bit. Requests failed with ABORTED_BY_REQUEST
338 * set the DNR bit correctly since the driver controls that.
341 case NVME_SCT_GENERIC:
343 case NVME_SC_ABORTED_BY_REQUEST:
344 case NVME_SC_NAMESPACE_NOT_READY:
349 case NVME_SC_INVALID_OPCODE:
350 case NVME_SC_INVALID_FIELD:
351 case NVME_SC_COMMAND_ID_CONFLICT:
352 case NVME_SC_DATA_TRANSFER_ERROR:
353 case NVME_SC_ABORTED_POWER_LOSS:
354 case NVME_SC_INTERNAL_DEVICE_ERROR:
355 case NVME_SC_ABORTED_SQ_DELETION:
356 case NVME_SC_ABORTED_FAILED_FUSED:
357 case NVME_SC_ABORTED_MISSING_FUSED:
358 case NVME_SC_INVALID_NAMESPACE_OR_FORMAT:
359 case NVME_SC_COMMAND_SEQUENCE_ERROR:
360 case NVME_SC_LBA_OUT_OF_RANGE:
361 case NVME_SC_CAPACITY_EXCEEDED:
365 case NVME_SCT_COMMAND_SPECIFIC:
366 case NVME_SCT_MEDIA_ERROR:
367 case NVME_SCT_VENDOR_SPECIFIC:
374 nvme_qpair_complete_tracker(struct nvme_qpair *qpair, struct nvme_tracker *tr,
375 struct nvme_completion *cpl, error_print_t print_on_error)
377 struct nvme_request *req;
378 boolean_t retry, error;
381 error = nvme_completion_is_error(cpl);
382 retry = error && nvme_completion_is_retry(cpl) &&
383 req->retries < nvme_retry_count;
385 if (error && (print_on_error == ERROR_PRINT_ALL ||
386 (!retry && print_on_error == ERROR_PRINT_NO_RETRY))) {
387 nvme_qpair_print_command(qpair, &req->cmd);
388 nvme_qpair_print_completion(qpair, cpl);
391 qpair->act_tr[cpl->cid] = NULL;
393 KASSERT(cpl->cid == req->cmd.cid, ("cpl cid does not match cmd cid\n"));
395 if (req->cb_fn && !retry)
396 req->cb_fn(req->cb_arg, cpl);
398 mtx_lock(&qpair->lock);
399 callout_stop(&tr->timer);
403 nvme_qpair_submit_tracker(qpair, tr);
405 if (req->type != NVME_REQUEST_NULL) {
406 bus_dmamap_sync(qpair->dma_tag_payload,
408 BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
409 bus_dmamap_unload(qpair->dma_tag_payload,
410 tr->payload_dma_map);
413 nvme_free_request(req);
416 TAILQ_REMOVE(&qpair->outstanding_tr, tr, tailq);
417 TAILQ_INSERT_HEAD(&qpair->free_tr, tr, tailq);
420 * If the controller is in the middle of resetting, don't
421 * try to submit queued requests here - let the reset logic
422 * handle that instead.
424 if (!STAILQ_EMPTY(&qpair->queued_req) &&
425 !qpair->ctrlr->is_resetting) {
426 req = STAILQ_FIRST(&qpair->queued_req);
427 STAILQ_REMOVE_HEAD(&qpair->queued_req, stailq);
428 _nvme_qpair_submit_request(qpair, req);
432 mtx_unlock(&qpair->lock);
436 nvme_qpair_manual_complete_tracker(struct nvme_qpair *qpair,
437 struct nvme_tracker *tr, uint32_t sct, uint32_t sc, uint32_t dnr,
438 error_print_t print_on_error)
440 struct nvme_completion cpl;
442 memset(&cpl, 0, sizeof(cpl));
443 cpl.sqid = qpair->id;
445 cpl.status |= (sct & NVME_STATUS_SCT_MASK) << NVME_STATUS_SCT_SHIFT;
446 cpl.status |= (sc & NVME_STATUS_SC_MASK) << NVME_STATUS_SC_SHIFT;
447 cpl.status |= (dnr & NVME_STATUS_DNR_MASK) << NVME_STATUS_DNR_SHIFT;
448 nvme_qpair_complete_tracker(qpair, tr, &cpl, print_on_error);
452 nvme_qpair_manual_complete_request(struct nvme_qpair *qpair,
453 struct nvme_request *req, uint32_t sct, uint32_t sc)
455 struct nvme_completion cpl;
458 memset(&cpl, 0, sizeof(cpl));
459 cpl.sqid = qpair->id;
460 cpl.status |= (sct & NVME_STATUS_SCT_MASK) << NVME_STATUS_SCT_SHIFT;
461 cpl.status |= (sc & NVME_STATUS_SC_MASK) << NVME_STATUS_SC_SHIFT;
463 error = nvme_completion_is_error(&cpl);
466 nvme_qpair_print_command(qpair, &req->cmd);
467 nvme_qpair_print_completion(qpair, &cpl);
471 req->cb_fn(req->cb_arg, &cpl);
473 nvme_free_request(req);
477 nvme_qpair_process_completions(struct nvme_qpair *qpair)
479 struct nvme_tracker *tr;
480 struct nvme_completion cpl;
483 qpair->num_intr_handler_calls++;
485 if (!qpair->is_enabled)
487 * qpair is not enabled, likely because a controller reset is
488 * is in progress. Ignore the interrupt - any I/O that was
489 * associated with this interrupt will get retried when the
494 bus_dmamap_sync(qpair->dma_tag, qpair->queuemem_map,
495 BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
497 cpl = qpair->cpl[qpair->cq_head];
499 /* Convert to host endian */
500 nvme_completion_swapbytes(&cpl);
502 if (NVME_STATUS_GET_P(cpl.status) != qpair->phase)
505 tr = qpair->act_tr[cpl.cid];
508 nvme_qpair_complete_tracker(qpair, tr, &cpl, ERROR_PRINT_ALL);
509 qpair->sq_head = cpl.sqhd;
512 nvme_printf(qpair->ctrlr,
513 "cpl does not map to outstanding cmd\n");
514 /* nvme_dump_completion expects device endianess */
515 nvme_dump_completion(&qpair->cpl[qpair->cq_head]);
516 KASSERT(0, ("received completion for unknown cmd\n"));
519 if (++qpair->cq_head == qpair->num_entries) {
521 qpair->phase = !qpair->phase;
524 nvme_mmio_write_4(qpair->ctrlr, doorbell[qpair->id].cq_hdbl,
531 nvme_qpair_msix_handler(void *arg)
533 struct nvme_qpair *qpair = arg;
535 nvme_qpair_process_completions(qpair);
539 nvme_qpair_construct(struct nvme_qpair *qpair, uint32_t id,
540 uint16_t vector, uint32_t num_entries, uint32_t num_trackers,
541 struct nvme_controller *ctrlr)
543 struct nvme_tracker *tr;
544 size_t cmdsz, cplsz, prpsz, allocsz, prpmemsz;
545 uint64_t queuemem_phys, prpmem_phys, list_phys;
546 uint8_t *queuemem, *prpmem, *prp_list;
550 qpair->vector = vector;
551 qpair->num_entries = num_entries;
552 qpair->num_trackers = num_trackers;
553 qpair->ctrlr = ctrlr;
555 if (ctrlr->msix_enabled) {
558 * MSI-X vector resource IDs start at 1, so we add one to
559 * the queue's vector to get the corresponding rid to use.
561 qpair->rid = vector + 1;
563 qpair->res = bus_alloc_resource_any(ctrlr->dev, SYS_RES_IRQ,
564 &qpair->rid, RF_ACTIVE);
565 bus_setup_intr(ctrlr->dev, qpair->res,
566 INTR_TYPE_MISC | INTR_MPSAFE, NULL,
567 nvme_qpair_msix_handler, qpair, &qpair->tag);
569 bus_describe_intr(ctrlr->dev, qpair->res, qpair->tag,
572 bus_describe_intr(ctrlr->dev, qpair->res, qpair->tag,
577 mtx_init(&qpair->lock, "nvme qpair lock", NULL, MTX_DEF);
579 /* Note: NVMe PRP format is restricted to 4-byte alignment. */
580 err = bus_dma_tag_create(bus_get_dma_tag(ctrlr->dev),
581 4, PAGE_SIZE, BUS_SPACE_MAXADDR,
582 BUS_SPACE_MAXADDR, NULL, NULL, NVME_MAX_XFER_SIZE,
583 (NVME_MAX_XFER_SIZE/PAGE_SIZE)+1, PAGE_SIZE, 0,
584 NULL, NULL, &qpair->dma_tag_payload);
586 nvme_printf(ctrlr, "payload tag create failed %d\n", err);
591 * Each component must be page aligned, and individual PRP lists
592 * cannot cross a page boundary.
594 cmdsz = qpair->num_entries * sizeof(struct nvme_command);
595 cmdsz = roundup2(cmdsz, PAGE_SIZE);
596 cplsz = qpair->num_entries * sizeof(struct nvme_completion);
597 cplsz = roundup2(cplsz, PAGE_SIZE);
598 prpsz = sizeof(uint64_t) * NVME_MAX_PRP_LIST_ENTRIES;;
599 prpmemsz = qpair->num_trackers * prpsz;
600 allocsz = cmdsz + cplsz + prpmemsz;
602 err = bus_dma_tag_create(bus_get_dma_tag(ctrlr->dev),
603 PAGE_SIZE, 0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
604 allocsz, 1, allocsz, 0, NULL, NULL, &qpair->dma_tag);
606 nvme_printf(ctrlr, "tag create failed %d\n", err);
610 if (bus_dmamem_alloc(qpair->dma_tag, (void **)&queuemem,
611 BUS_DMA_NOWAIT, &qpair->queuemem_map)) {
612 nvme_printf(ctrlr, "failed to alloc qpair memory\n");
616 if (bus_dmamap_load(qpair->dma_tag, qpair->queuemem_map,
617 queuemem, allocsz, nvme_single_map, &queuemem_phys, 0) != 0) {
618 nvme_printf(ctrlr, "failed to load qpair memory\n");
623 qpair->num_intr_handler_calls = 0;
624 qpair->cmd = (struct nvme_command *)queuemem;
625 qpair->cpl = (struct nvme_completion *)(queuemem + cmdsz);
626 prpmem = (uint8_t *)(queuemem + cmdsz + cplsz);
627 qpair->cmd_bus_addr = queuemem_phys;
628 qpair->cpl_bus_addr = queuemem_phys + cmdsz;
629 prpmem_phys = queuemem_phys + cmdsz + cplsz;
631 qpair->sq_tdbl_off = nvme_mmio_offsetof(doorbell[id].sq_tdbl);
632 qpair->cq_hdbl_off = nvme_mmio_offsetof(doorbell[id].cq_hdbl);
634 TAILQ_INIT(&qpair->free_tr);
635 TAILQ_INIT(&qpair->outstanding_tr);
636 STAILQ_INIT(&qpair->queued_req);
638 list_phys = prpmem_phys;
640 for (i = 0; i < qpair->num_trackers; i++) {
642 if (list_phys + prpsz > prpmem_phys + prpmemsz) {
643 qpair->num_trackers = i;
648 * Make sure that the PRP list for this tracker doesn't
649 * overflow to another page.
651 if (trunc_page(list_phys) !=
652 trunc_page(list_phys + prpsz - 1)) {
653 list_phys = roundup2(list_phys, PAGE_SIZE);
655 (uint8_t *)roundup2((uintptr_t)prp_list, PAGE_SIZE);
658 tr = malloc(sizeof(*tr), M_NVME, M_ZERO | M_WAITOK);
659 bus_dmamap_create(qpair->dma_tag_payload, 0,
660 &tr->payload_dma_map);
661 callout_init(&tr->timer, 1);
664 tr->prp = (uint64_t *)prp_list;
665 tr->prp_bus_addr = list_phys;
666 TAILQ_INSERT_HEAD(&qpair->free_tr, tr, tailq);
671 if (qpair->num_trackers == 0) {
672 nvme_printf(ctrlr, "failed to allocate enough trackers\n");
676 qpair->act_tr = malloc(sizeof(struct nvme_tracker *) *
677 qpair->num_entries, M_NVME, M_ZERO | M_WAITOK);
681 nvme_qpair_destroy(qpair);
686 nvme_qpair_destroy(struct nvme_qpair *qpair)
688 struct nvme_tracker *tr;
691 bus_teardown_intr(qpair->ctrlr->dev, qpair->res, qpair->tag);
693 if (mtx_initialized(&qpair->lock))
694 mtx_destroy(&qpair->lock);
697 bus_release_resource(qpair->ctrlr->dev, SYS_RES_IRQ,
698 rman_get_rid(qpair->res), qpair->res);
700 if (qpair->cmd != NULL) {
701 bus_dmamap_unload(qpair->dma_tag, qpair->queuemem_map);
702 bus_dmamem_free(qpair->dma_tag, qpair->cmd,
703 qpair->queuemem_map);
707 free(qpair->act_tr, M_NVME);
709 while (!TAILQ_EMPTY(&qpair->free_tr)) {
710 tr = TAILQ_FIRST(&qpair->free_tr);
711 TAILQ_REMOVE(&qpair->free_tr, tr, tailq);
712 bus_dmamap_destroy(qpair->dma_tag_payload,
713 tr->payload_dma_map);
718 bus_dma_tag_destroy(qpair->dma_tag);
720 if (qpair->dma_tag_payload)
721 bus_dma_tag_destroy(qpair->dma_tag_payload);
725 nvme_admin_qpair_abort_aers(struct nvme_qpair *qpair)
727 struct nvme_tracker *tr;
729 tr = TAILQ_FIRST(&qpair->outstanding_tr);
731 if (tr->req->cmd.opc == NVME_OPC_ASYNC_EVENT_REQUEST) {
732 nvme_qpair_manual_complete_tracker(qpair, tr,
733 NVME_SCT_GENERIC, NVME_SC_ABORTED_SQ_DELETION, 0,
735 tr = TAILQ_FIRST(&qpair->outstanding_tr);
737 tr = TAILQ_NEXT(tr, tailq);
743 nvme_admin_qpair_destroy(struct nvme_qpair *qpair)
746 nvme_admin_qpair_abort_aers(qpair);
747 nvme_qpair_destroy(qpair);
751 nvme_io_qpair_destroy(struct nvme_qpair *qpair)
754 nvme_qpair_destroy(qpair);
758 nvme_abort_complete(void *arg, const struct nvme_completion *status)
760 struct nvme_tracker *tr = arg;
763 * If cdw0 == 1, the controller was not able to abort the command
764 * we requested. We still need to check the active tracker array,
765 * to cover race where I/O timed out at same time controller was
766 * completing the I/O.
768 if (status->cdw0 == 1 && tr->qpair->act_tr[tr->cid] != NULL) {
770 * An I/O has timed out, and the controller was unable to
771 * abort it for some reason. Construct a fake completion
772 * status, and then complete the I/O's tracker manually.
774 nvme_printf(tr->qpair->ctrlr,
775 "abort command failed, aborting command manually\n");
776 nvme_qpair_manual_complete_tracker(tr->qpair, tr,
777 NVME_SCT_GENERIC, NVME_SC_ABORTED_BY_REQUEST, 0, ERROR_PRINT_ALL);
782 nvme_timeout(void *arg)
784 struct nvme_tracker *tr = arg;
785 struct nvme_qpair *qpair = tr->qpair;
786 struct nvme_controller *ctrlr = qpair->ctrlr;
791 * Read csts to get value of cfs - controller fatal status.
792 * If no fatal status, try to call the completion routine, and
793 * if completes transactions, report a missed interrupt and
794 * return (this may need to be rate limited). Otherwise, if
795 * aborts are enabled and the controller is not reporting
796 * fatal status, abort the command. Otherwise, just reset the
797 * controller and hope for the best.
799 csts = nvme_mmio_read_4(ctrlr, csts);
800 cfs = (csts >> NVME_CSTS_REG_CFS_SHIFT) & NVME_CSTS_REG_CFS_MASK;
801 if (cfs == 0 && nvme_qpair_process_completions(qpair)) {
802 nvme_printf(ctrlr, "Missing interrupt\n");
805 if (ctrlr->enable_aborts && cfs == 0) {
806 nvme_printf(ctrlr, "Aborting command due to a timeout.\n");
807 nvme_ctrlr_cmd_abort(ctrlr, tr->cid, qpair->id,
808 nvme_abort_complete, tr);
810 nvme_printf(ctrlr, "Resetting controller due to a timeout%s.\n",
811 cfs ? " and fatal error status" : "");
812 nvme_ctrlr_reset(ctrlr);
817 nvme_qpair_submit_tracker(struct nvme_qpair *qpair, struct nvme_tracker *tr)
819 struct nvme_request *req;
820 struct nvme_controller *ctrlr;
822 mtx_assert(&qpair->lock, MA_OWNED);
825 req->cmd.cid = tr->cid;
826 qpair->act_tr[tr->cid] = tr;
827 ctrlr = qpair->ctrlr;
830 callout_reset_curcpu(&tr->timer, ctrlr->timeout_period * hz,
833 /* Copy the command from the tracker to the submission queue. */
834 memcpy(&qpair->cmd[qpair->sq_tail], &req->cmd, sizeof(req->cmd));
836 if (++qpair->sq_tail == qpair->num_entries)
839 bus_dmamap_sync(qpair->dma_tag, qpair->queuemem_map,
840 BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
843 * powerpc's bus_dmamap_sync() already includes a heavyweight sync, but
849 nvme_mmio_write_4(qpair->ctrlr, doorbell[qpair->id].sq_tdbl,
856 nvme_payload_map(void *arg, bus_dma_segment_t *seg, int nseg, int error)
858 struct nvme_tracker *tr = arg;
862 * If the mapping operation failed, return immediately. The caller
863 * is responsible for detecting the error status and failing the
867 nvme_printf(tr->qpair->ctrlr,
868 "nvme_payload_map err %d\n", error);
873 * Note that we specified PAGE_SIZE for alignment and max
874 * segment size when creating the bus dma tags. So here
875 * we can safely just transfer each segment to its
876 * associated PRP entry.
878 tr->req->cmd.prp1 = htole64(seg[0].ds_addr);
881 tr->req->cmd.prp2 = htole64(seg[1].ds_addr);
882 } else if (nseg > 2) {
884 tr->req->cmd.prp2 = htole64((uint64_t)tr->prp_bus_addr);
885 while (cur_nseg < nseg) {
886 tr->prp[cur_nseg-1] =
887 htole64((uint64_t)seg[cur_nseg].ds_addr);
892 * prp2 should not be used by the controller
893 * since there is only one segment, but set
894 * to 0 just to be safe.
896 tr->req->cmd.prp2 = 0;
899 bus_dmamap_sync(tr->qpair->dma_tag_payload, tr->payload_dma_map,
900 BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
901 nvme_qpair_submit_tracker(tr->qpair, tr);
905 _nvme_qpair_submit_request(struct nvme_qpair *qpair, struct nvme_request *req)
907 struct nvme_tracker *tr;
910 mtx_assert(&qpair->lock, MA_OWNED);
912 tr = TAILQ_FIRST(&qpair->free_tr);
915 if (tr == NULL || !qpair->is_enabled) {
917 * No tracker is available, or the qpair is disabled due to
918 * an in-progress controller-level reset or controller
922 if (qpair->ctrlr->is_failed) {
924 * The controller has failed. Post the request to a
925 * task where it will be aborted, so that we do not
926 * invoke the request's callback in the context
929 nvme_ctrlr_post_failed_request(qpair->ctrlr, req);
932 * Put the request on the qpair's request queue to be
933 * processed when a tracker frees up via a command
934 * completion or when the controller reset is
937 STAILQ_INSERT_TAIL(&qpair->queued_req, req, stailq);
942 TAILQ_REMOVE(&qpair->free_tr, tr, tailq);
943 TAILQ_INSERT_TAIL(&qpair->outstanding_tr, tr, tailq);
947 case NVME_REQUEST_VADDR:
948 KASSERT(req->payload_size <= qpair->ctrlr->max_xfer_size,
949 ("payload_size (%d) exceeds max_xfer_size (%d)\n",
950 req->payload_size, qpair->ctrlr->max_xfer_size));
951 err = bus_dmamap_load(tr->qpair->dma_tag_payload,
952 tr->payload_dma_map, req->u.payload, req->payload_size,
953 nvme_payload_map, tr, 0);
955 nvme_printf(qpair->ctrlr,
956 "bus_dmamap_load returned 0x%x!\n", err);
958 case NVME_REQUEST_NULL:
959 nvme_qpair_submit_tracker(tr->qpair, tr);
961 case NVME_REQUEST_BIO:
962 KASSERT(req->u.bio->bio_bcount <= qpair->ctrlr->max_xfer_size,
963 ("bio->bio_bcount (%jd) exceeds max_xfer_size (%d)\n",
964 (intmax_t)req->u.bio->bio_bcount,
965 qpair->ctrlr->max_xfer_size));
966 err = bus_dmamap_load_bio(tr->qpair->dma_tag_payload,
967 tr->payload_dma_map, req->u.bio, nvme_payload_map, tr, 0);
969 nvme_printf(qpair->ctrlr,
970 "bus_dmamap_load_bio returned 0x%x!\n", err);
972 case NVME_REQUEST_CCB:
973 err = bus_dmamap_load_ccb(tr->qpair->dma_tag_payload,
974 tr->payload_dma_map, req->u.payload,
975 nvme_payload_map, tr, 0);
977 nvme_printf(qpair->ctrlr,
978 "bus_dmamap_load_ccb returned 0x%x!\n", err);
981 panic("unknown nvme request type 0x%x\n", req->type);
987 * The dmamap operation failed, so we manually fail the
988 * tracker here with DATA_TRANSFER_ERROR status.
990 * nvme_qpair_manual_complete_tracker must not be called
991 * with the qpair lock held.
993 mtx_unlock(&qpair->lock);
994 nvme_qpair_manual_complete_tracker(qpair, tr, NVME_SCT_GENERIC,
995 NVME_SC_DATA_TRANSFER_ERROR, DO_NOT_RETRY, ERROR_PRINT_ALL);
996 mtx_lock(&qpair->lock);
1001 nvme_qpair_submit_request(struct nvme_qpair *qpair, struct nvme_request *req)
1004 mtx_lock(&qpair->lock);
1005 _nvme_qpair_submit_request(qpair, req);
1006 mtx_unlock(&qpair->lock);
1010 nvme_qpair_enable(struct nvme_qpair *qpair)
1013 qpair->is_enabled = TRUE;
1017 nvme_qpair_reset(struct nvme_qpair *qpair)
1020 qpair->sq_head = qpair->sq_tail = qpair->cq_head = 0;
1023 * First time through the completion queue, HW will set phase
1024 * bit on completions to 1. So set this to 1 here, indicating
1025 * we're looking for a 1 to know which entries have completed.
1026 * we'll toggle the bit each time when the completion queue
1031 memset(qpair->cmd, 0,
1032 qpair->num_entries * sizeof(struct nvme_command));
1033 memset(qpair->cpl, 0,
1034 qpair->num_entries * sizeof(struct nvme_completion));
1038 nvme_admin_qpair_enable(struct nvme_qpair *qpair)
1040 struct nvme_tracker *tr;
1041 struct nvme_tracker *tr_temp;
1044 * Manually abort each outstanding admin command. Do not retry
1045 * admin commands found here, since they will be left over from
1046 * a controller reset and its likely the context in which the
1047 * command was issued no longer applies.
1049 TAILQ_FOREACH_SAFE(tr, &qpair->outstanding_tr, tailq, tr_temp) {
1050 nvme_printf(qpair->ctrlr,
1051 "aborting outstanding admin command\n");
1052 nvme_qpair_manual_complete_tracker(qpair, tr, NVME_SCT_GENERIC,
1053 NVME_SC_ABORTED_BY_REQUEST, DO_NOT_RETRY, ERROR_PRINT_ALL);
1056 nvme_qpair_enable(qpair);
1060 nvme_io_qpair_enable(struct nvme_qpair *qpair)
1062 STAILQ_HEAD(, nvme_request) temp;
1063 struct nvme_tracker *tr;
1064 struct nvme_tracker *tr_temp;
1065 struct nvme_request *req;
1068 * Manually abort each outstanding I/O. This normally results in a
1069 * retry, unless the retry count on the associated request has
1070 * reached its limit.
1072 TAILQ_FOREACH_SAFE(tr, &qpair->outstanding_tr, tailq, tr_temp) {
1073 nvme_printf(qpair->ctrlr, "aborting outstanding i/o\n");
1074 nvme_qpair_manual_complete_tracker(qpair, tr, NVME_SCT_GENERIC,
1075 NVME_SC_ABORTED_BY_REQUEST, 0, ERROR_PRINT_NO_RETRY);
1078 mtx_lock(&qpair->lock);
1080 nvme_qpair_enable(qpair);
1083 STAILQ_SWAP(&qpair->queued_req, &temp, nvme_request);
1085 while (!STAILQ_EMPTY(&temp)) {
1086 req = STAILQ_FIRST(&temp);
1087 STAILQ_REMOVE_HEAD(&temp, stailq);
1088 nvme_printf(qpair->ctrlr, "resubmitting queued i/o\n");
1089 nvme_qpair_print_command(qpair, &req->cmd);
1090 _nvme_qpair_submit_request(qpair, req);
1093 mtx_unlock(&qpair->lock);
1097 nvme_qpair_disable(struct nvme_qpair *qpair)
1099 struct nvme_tracker *tr;
1101 qpair->is_enabled = FALSE;
1102 mtx_lock(&qpair->lock);
1103 TAILQ_FOREACH(tr, &qpair->outstanding_tr, tailq)
1104 callout_stop(&tr->timer);
1105 mtx_unlock(&qpair->lock);
1109 nvme_admin_qpair_disable(struct nvme_qpair *qpair)
1112 nvme_qpair_disable(qpair);
1113 nvme_admin_qpair_abort_aers(qpair);
1117 nvme_io_qpair_disable(struct nvme_qpair *qpair)
1120 nvme_qpair_disable(qpair);
1124 nvme_qpair_fail(struct nvme_qpair *qpair)
1126 struct nvme_tracker *tr;
1127 struct nvme_request *req;
1129 if (!mtx_initialized(&qpair->lock))
1132 mtx_lock(&qpair->lock);
1134 while (!STAILQ_EMPTY(&qpair->queued_req)) {
1135 req = STAILQ_FIRST(&qpair->queued_req);
1136 STAILQ_REMOVE_HEAD(&qpair->queued_req, stailq);
1137 nvme_printf(qpair->ctrlr, "failing queued i/o\n");
1138 mtx_unlock(&qpair->lock);
1139 nvme_qpair_manual_complete_request(qpair, req, NVME_SCT_GENERIC,
1140 NVME_SC_ABORTED_BY_REQUEST);
1141 mtx_lock(&qpair->lock);
1144 /* Manually abort each outstanding I/O. */
1145 while (!TAILQ_EMPTY(&qpair->outstanding_tr)) {
1146 tr = TAILQ_FIRST(&qpair->outstanding_tr);
1148 * Do not remove the tracker. The abort_tracker path will
1151 nvme_printf(qpair->ctrlr, "failing outstanding i/o\n");
1152 mtx_unlock(&qpair->lock);
1153 nvme_qpair_manual_complete_tracker(qpair, tr, NVME_SCT_GENERIC,
1154 NVME_SC_ABORTED_BY_REQUEST, DO_NOT_RETRY, ERROR_PRINT_ALL);
1155 mtx_lock(&qpair->lock);
1158 mtx_unlock(&qpair->lock);