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 * $FreeBSD$ 29 */ 30 31#ifndef __NVME_PRIVATE_H__ 32#define __NVME_PRIVATE_H__ 33 34#include <sys/param.h> 35#include <sys/bio.h> 36#include <sys/bus.h> 37#include <sys/kernel.h> 38#include <sys/lock.h> 39#include <sys/malloc.h> 40#include <sys/module.h> 41#include <sys/mutex.h> 42#include <sys/rman.h> 43#include <sys/systm.h> 44#include <sys/taskqueue.h> 45 46#include <vm/uma.h> 47 48#include <machine/bus.h> 49 50#include "nvme.h" 51 52#define DEVICE2SOFTC(dev) ((struct nvme_controller *) device_get_softc(dev)) 53 54MALLOC_DECLARE(M_NVME); 55 56#define IDT32_PCI_ID 0x80d0111d /* 32 channel board */ 57#define IDT8_PCI_ID 0x80d2111d /* 8 channel board */ 58 59#define NVME_ADMIN_TRACKERS (16) 60#define NVME_ADMIN_ENTRIES (128) 61/* min and max are defined in admin queue attributes section of spec */ 62#define NVME_MIN_ADMIN_ENTRIES (2) 63#define NVME_MAX_ADMIN_ENTRIES (4096) 64 65/* 66 * NVME_IO_ENTRIES defines the size of an I/O qpair's submission and completion 67 * queues, while NVME_IO_TRACKERS defines the maximum number of I/O that we 68 * will allow outstanding on an I/O qpair at any time. The only advantage in 69 * having IO_ENTRIES > IO_TRACKERS is for debugging purposes - when dumping 70 * the contents of the submission and completion queues, it will show a longer 71 * history of data. 72 */ 73#define NVME_IO_ENTRIES (256) 74#define NVME_IO_TRACKERS (128) 75#define NVME_MIN_IO_TRACKERS (4) 76#define NVME_MAX_IO_TRACKERS (1024) 77 78/* 79 * NVME_MAX_IO_ENTRIES is not defined, since it is specified in CC.MQES 80 * for each controller. 81 */ 82 83#define NVME_INT_COAL_TIME (0) /* disabled */ 84#define NVME_INT_COAL_THRESHOLD (0) /* 0-based */ 85 86#define NVME_MAX_NAMESPACES (16) 87#define NVME_MAX_CONSUMERS (2) 88#define NVME_MAX_ASYNC_EVENTS (8) 89 90#define NVME_DEFAULT_TIMEOUT_PERIOD (30) /* in seconds */ 91#define NVME_MIN_TIMEOUT_PERIOD (5) 92#define NVME_MAX_TIMEOUT_PERIOD (120) 93 94#define NVME_DEFAULT_RETRY_COUNT (4) 95 96/* Maximum log page size to fetch for AERs. */ 97#define NVME_MAX_AER_LOG_SIZE (4096) 98 99/* 100 * Define CACHE_LINE_SIZE here for older FreeBSD versions that do not define 101 * it. 102 */ 103#ifndef CACHE_LINE_SIZE 104#define CACHE_LINE_SIZE (64) 105#endif 106 107#define NVME_GONE 0xfffffffful 108 109extern int32_t nvme_retry_count; 110extern bool nvme_verbose_cmd_dump; 111 112struct nvme_completion_poll_status { 113 struct nvme_completion cpl; 114 int done; 115}; 116 117extern devclass_t nvme_devclass; 118 119#define NVME_REQUEST_VADDR 1 120#define NVME_REQUEST_NULL 2 /* For requests with no payload. */ 121#define NVME_REQUEST_UIO 3 122#define NVME_REQUEST_BIO 4 123#define NVME_REQUEST_CCB 5 124 125struct nvme_request { 126 struct nvme_command cmd; 127 struct nvme_qpair *qpair; 128 union { 129 void *payload; 130 struct bio *bio; 131 } u; 132 uint32_t type; 133 uint32_t payload_size; 134 bool timeout; 135 nvme_cb_fn_t cb_fn; 136 void *cb_arg; 137 int32_t retries; 138 STAILQ_ENTRY(nvme_request) stailq; 139}; 140 141struct nvme_async_event_request { 142 struct nvme_controller *ctrlr; 143 struct nvme_request *req; 144 struct nvme_completion cpl; 145 uint32_t log_page_id; 146 uint32_t log_page_size; 147 uint8_t log_page_buffer[NVME_MAX_AER_LOG_SIZE]; 148}; 149 150struct nvme_tracker { 151 TAILQ_ENTRY(nvme_tracker) tailq; 152 struct nvme_request *req; 153 struct nvme_qpair *qpair; 154 struct callout timer; 155 bus_dmamap_t payload_dma_map; 156 uint16_t cid; 157 158 uint64_t *prp; 159 bus_addr_t prp_bus_addr; 160}; 161 162struct nvme_qpair { 163 struct nvme_controller *ctrlr; 164 uint32_t id; 165 int domain; 166 int cpu; 167 168 uint16_t vector; 169 int rid; 170 struct resource *res; 171 void *tag; 172 173 uint32_t num_entries; 174 uint32_t num_trackers; 175 uint32_t sq_tdbl_off; 176 uint32_t cq_hdbl_off; 177 178 uint32_t phase; 179 uint32_t sq_head; 180 uint32_t sq_tail; 181 uint32_t cq_head; 182 183 int64_t num_cmds; 184 int64_t num_intr_handler_calls; 185 int64_t num_retries; 186 int64_t num_failures; 187 188 struct nvme_command *cmd; 189 struct nvme_completion *cpl; 190 191 bus_dma_tag_t dma_tag; 192 bus_dma_tag_t dma_tag_payload; 193 194 bus_dmamap_t queuemem_map; 195 uint64_t cmd_bus_addr; 196 uint64_t cpl_bus_addr; 197 198 TAILQ_HEAD(, nvme_tracker) free_tr; 199 TAILQ_HEAD(, nvme_tracker) outstanding_tr; 200 STAILQ_HEAD(, nvme_request) queued_req; 201 202 struct nvme_tracker **act_tr; 203 204 bool is_enabled; 205 206 struct mtx lock __aligned(CACHE_LINE_SIZE); 207 208} __aligned(CACHE_LINE_SIZE); 209 210struct nvme_namespace { 211 struct nvme_controller *ctrlr; 212 struct nvme_namespace_data data; 213 uint32_t id; 214 uint32_t flags; 215 struct cdev *cdev; 216 void *cons_cookie[NVME_MAX_CONSUMERS]; 217 uint32_t boundary; 218 struct mtx lock; 219}; 220 221/* 222 * One of these per allocated PCI device. 223 */ 224struct nvme_controller { 225 device_t dev; 226 227 struct mtx lock; 228 int domain; 229 uint32_t ready_timeout_in_ms; 230 uint32_t quirks; 231#define QUIRK_DELAY_B4_CHK_RDY 1 /* Can't touch MMIO on disable */ 232#define QUIRK_DISABLE_TIMEOUT 2 /* Disable broken completion timeout feature */ 233 234 bus_space_tag_t bus_tag; 235 bus_space_handle_t bus_handle; 236 int resource_id; 237 struct resource *resource; 238 239 /* 240 * The NVMe spec allows for the MSI-X table to be placed in BAR 4/5, 241 * separate from the control registers which are in BAR 0/1. These 242 * members track the mapping of BAR 4/5 for that reason. 243 */ 244 int bar4_resource_id; 245 struct resource *bar4_resource; 246 247 uint32_t msix_enabled; 248 uint32_t enable_aborts; 249 250 uint32_t num_io_queues; 251 uint32_t max_hw_pend_io; 252 253 /* Fields for tracking progress during controller initialization. */ 254 struct intr_config_hook config_hook; 255 uint32_t ns_identified; 256 uint32_t queues_created; 257 258 struct task reset_task; 259 struct task fail_req_task; 260 struct taskqueue *taskqueue; 261 262 /* For shared legacy interrupt. */ 263 int rid; 264 struct resource *res; 265 void *tag; 266 267 /** maximum i/o size in bytes */ 268 uint32_t max_xfer_size; 269 270 /** minimum page size supported by this controller in bytes */ 271 uint32_t min_page_size; 272 273 /** interrupt coalescing time period (in microseconds) */ 274 uint32_t int_coal_time; 275 276 /** interrupt coalescing threshold */ 277 uint32_t int_coal_threshold; 278 279 /** timeout period in seconds */ 280 uint32_t timeout_period; 281 282 /** doorbell stride */ 283 uint32_t dstrd; 284 285 struct nvme_qpair adminq; 286 struct nvme_qpair *ioq; 287 288 struct nvme_registers *regs; 289 290 struct nvme_controller_data cdata; 291 struct nvme_namespace ns[NVME_MAX_NAMESPACES]; 292 293 struct cdev *cdev; 294 295 /** bit mask of event types currently enabled for async events */ 296 uint32_t async_event_config; 297 298 uint32_t num_aers; 299 struct nvme_async_event_request aer[NVME_MAX_ASYNC_EVENTS]; 300 301 void *cons_cookie[NVME_MAX_CONSUMERS]; 302 303 uint32_t is_resetting; 304 uint32_t is_initialized; 305 uint32_t notification_sent; 306 307 bool is_failed; 308 STAILQ_HEAD(, nvme_request) fail_req; 309 310 /* Host Memory Buffer */ 311 int hmb_nchunks; 312 size_t hmb_chunk; 313 bus_dma_tag_t hmb_tag; 314 struct nvme_hmb_chunk { 315 bus_dmamap_t hmbc_map; 316 void *hmbc_vaddr; 317 uint64_t hmbc_paddr; 318 } *hmb_chunks; 319 bus_dma_tag_t hmb_desc_tag; 320 bus_dmamap_t hmb_desc_map; 321 struct nvme_hmb_desc *hmb_desc_vaddr; 322 uint64_t hmb_desc_paddr; 323}; 324 325#define nvme_mmio_offsetof(reg) \ 326 offsetof(struct nvme_registers, reg) 327 328#define nvme_mmio_read_4(sc, reg) \ 329 bus_space_read_4((sc)->bus_tag, (sc)->bus_handle, \ 330 nvme_mmio_offsetof(reg)) 331 332#define nvme_mmio_write_4(sc, reg, val) \ 333 bus_space_write_4((sc)->bus_tag, (sc)->bus_handle, \ 334 nvme_mmio_offsetof(reg), val) 335 336#define nvme_mmio_write_8(sc, reg, val) \ 337 do { \ 338 bus_space_write_4((sc)->bus_tag, (sc)->bus_handle, \ 339 nvme_mmio_offsetof(reg), val & 0xFFFFFFFF); \ 340 bus_space_write_4((sc)->bus_tag, (sc)->bus_handle, \ 341 nvme_mmio_offsetof(reg)+4, \ 342 (val & 0xFFFFFFFF00000000ULL) >> 32); \ 343 } while (0); 344 345#define nvme_printf(ctrlr, fmt, args...) \ 346 device_printf(ctrlr->dev, fmt, ##args) 347 348void nvme_ns_test(struct nvme_namespace *ns, u_long cmd, caddr_t arg); 349 350void nvme_ctrlr_cmd_identify_controller(struct nvme_controller *ctrlr, 351 void *payload, 352 nvme_cb_fn_t cb_fn, void *cb_arg); 353void nvme_ctrlr_cmd_identify_namespace(struct nvme_controller *ctrlr, 354 uint32_t nsid, void *payload, 355 nvme_cb_fn_t cb_fn, void *cb_arg); 356void nvme_ctrlr_cmd_set_interrupt_coalescing(struct nvme_controller *ctrlr, 357 uint32_t microseconds, 358 uint32_t threshold, 359 nvme_cb_fn_t cb_fn, 360 void *cb_arg); 361void nvme_ctrlr_cmd_get_error_page(struct nvme_controller *ctrlr, 362 struct nvme_error_information_entry *payload, 363 uint32_t num_entries, /* 0 = max */ 364 nvme_cb_fn_t cb_fn, 365 void *cb_arg); 366void nvme_ctrlr_cmd_get_health_information_page(struct nvme_controller *ctrlr, 367 uint32_t nsid, 368 struct nvme_health_information_page *payload, 369 nvme_cb_fn_t cb_fn, 370 void *cb_arg); 371void nvme_ctrlr_cmd_get_firmware_page(struct nvme_controller *ctrlr, 372 struct nvme_firmware_page *payload, 373 nvme_cb_fn_t cb_fn, 374 void *cb_arg); 375void nvme_ctrlr_cmd_create_io_cq(struct nvme_controller *ctrlr, 376 struct nvme_qpair *io_que, 377 nvme_cb_fn_t cb_fn, void *cb_arg); 378void nvme_ctrlr_cmd_create_io_sq(struct nvme_controller *ctrlr, 379 struct nvme_qpair *io_que, 380 nvme_cb_fn_t cb_fn, void *cb_arg); 381void nvme_ctrlr_cmd_delete_io_cq(struct nvme_controller *ctrlr, 382 struct nvme_qpair *io_que, 383 nvme_cb_fn_t cb_fn, void *cb_arg); 384void nvme_ctrlr_cmd_delete_io_sq(struct nvme_controller *ctrlr, 385 struct nvme_qpair *io_que, 386 nvme_cb_fn_t cb_fn, void *cb_arg); 387void nvme_ctrlr_cmd_set_num_queues(struct nvme_controller *ctrlr, 388 uint32_t num_queues, nvme_cb_fn_t cb_fn, 389 void *cb_arg); 390void nvme_ctrlr_cmd_set_async_event_config(struct nvme_controller *ctrlr, 391 uint32_t state, 392 nvme_cb_fn_t cb_fn, void *cb_arg); 393void nvme_ctrlr_cmd_abort(struct nvme_controller *ctrlr, uint16_t cid, 394 uint16_t sqid, nvme_cb_fn_t cb_fn, void *cb_arg); 395 396void nvme_completion_poll_cb(void *arg, const struct nvme_completion *cpl); 397 398int nvme_ctrlr_construct(struct nvme_controller *ctrlr, device_t dev); 399void nvme_ctrlr_destruct(struct nvme_controller *ctrlr, device_t dev); 400void nvme_ctrlr_shutdown(struct nvme_controller *ctrlr); 401void nvme_ctrlr_reset(struct nvme_controller *ctrlr); 402/* ctrlr defined as void * to allow use with config_intrhook. */ 403void nvme_ctrlr_start_config_hook(void *ctrlr_arg); 404void nvme_ctrlr_submit_admin_request(struct nvme_controller *ctrlr, 405 struct nvme_request *req); 406void nvme_ctrlr_submit_io_request(struct nvme_controller *ctrlr, 407 struct nvme_request *req); 408void nvme_ctrlr_post_failed_request(struct nvme_controller *ctrlr, 409 struct nvme_request *req); 410 411int nvme_qpair_construct(struct nvme_qpair *qpair, 412 uint32_t num_entries, uint32_t num_trackers, 413 struct nvme_controller *ctrlr); 414void nvme_qpair_submit_tracker(struct nvme_qpair *qpair, 415 struct nvme_tracker *tr); 416bool nvme_qpair_process_completions(struct nvme_qpair *qpair); 417void nvme_qpair_submit_request(struct nvme_qpair *qpair, 418 struct nvme_request *req); 419void nvme_qpair_reset(struct nvme_qpair *qpair); 420void nvme_qpair_fail(struct nvme_qpair *qpair); 421void nvme_qpair_manual_complete_request(struct nvme_qpair *qpair, 422 struct nvme_request *req, 423 uint32_t sct, uint32_t sc); 424 425void nvme_admin_qpair_enable(struct nvme_qpair *qpair); 426void nvme_admin_qpair_disable(struct nvme_qpair *qpair); 427void nvme_admin_qpair_destroy(struct nvme_qpair *qpair); 428 429void nvme_io_qpair_enable(struct nvme_qpair *qpair); 430void nvme_io_qpair_disable(struct nvme_qpair *qpair); 431void nvme_io_qpair_destroy(struct nvme_qpair *qpair); 432 433int nvme_ns_construct(struct nvme_namespace *ns, uint32_t id, 434 struct nvme_controller *ctrlr); 435void nvme_ns_destruct(struct nvme_namespace *ns); 436 437void nvme_sysctl_initialize_ctrlr(struct nvme_controller *ctrlr); 438 439void nvme_dump_command(struct nvme_command *cmd); 440void nvme_dump_completion(struct nvme_completion *cpl); 441 442int nvme_attach(device_t dev); 443int nvme_shutdown(device_t dev); 444int nvme_detach(device_t dev); 445 446/* 447 * Wait for a command to complete using the nvme_completion_poll_cb. 448 * Used in limited contexts where the caller knows it's OK to block 449 * briefly while the command runs. The ISR will run the callback which 450 * will set status->done to true, usually within microseconds. If not, 451 * then after one second timeout handler should reset the controller 452 * and abort all outstanding requests including this polled one. If 453 * still not after ten seconds, then something is wrong with the driver, 454 * and panic is the only way to recover. 455 */ 456static __inline 457void 458nvme_completion_poll(struct nvme_completion_poll_status *status) 459{ 460 int sanity = hz * 10; 461 462 while (!atomic_load_acq_int(&status->done) && --sanity > 0) 463 pause("nvme", 1); 464 if (sanity <= 0) 465 panic("NVME polled command failed to complete within 10s."); 466} 467 468static __inline void 469nvme_single_map(void *arg, bus_dma_segment_t *seg, int nseg, int error) 470{ 471 uint64_t *bus_addr = (uint64_t *)arg; 472 473 KASSERT(nseg == 1, ("number of segments (%d) is not 1", nseg)); 474 if (error != 0) 475 printf("nvme_single_map err %d\n", error); 476 *bus_addr = seg[0].ds_addr; 477} 478 479static __inline struct nvme_request * 480_nvme_allocate_request(nvme_cb_fn_t cb_fn, void *cb_arg) 481{ 482 struct nvme_request *req; 483 484 req = malloc(sizeof(*req), M_NVME, M_NOWAIT | M_ZERO); 485 if (req != NULL) { 486 req->cb_fn = cb_fn; 487 req->cb_arg = cb_arg; 488 req->timeout = true; 489 } 490 return (req); 491} 492 493static __inline struct nvme_request * 494nvme_allocate_request_vaddr(void *payload, uint32_t payload_size, 495 nvme_cb_fn_t cb_fn, void *cb_arg) 496{ 497 struct nvme_request *req; 498 499 req = _nvme_allocate_request(cb_fn, cb_arg); 500 if (req != NULL) { 501 req->type = NVME_REQUEST_VADDR; 502 req->u.payload = payload; 503 req->payload_size = payload_size; 504 } 505 return (req); 506} 507 508static __inline struct nvme_request * 509nvme_allocate_request_null(nvme_cb_fn_t cb_fn, void *cb_arg) 510{ 511 struct nvme_request *req; 512 513 req = _nvme_allocate_request(cb_fn, cb_arg); 514 if (req != NULL) 515 req->type = NVME_REQUEST_NULL; 516 return (req); 517} 518 519static __inline struct nvme_request * 520nvme_allocate_request_bio(struct bio *bio, nvme_cb_fn_t cb_fn, void *cb_arg) 521{ 522 struct nvme_request *req; 523 524 req = _nvme_allocate_request(cb_fn, cb_arg); 525 if (req != NULL) { 526 req->type = NVME_REQUEST_BIO; 527 req->u.bio = bio; 528 } 529 return (req); 530} 531 532static __inline struct nvme_request * 533nvme_allocate_request_ccb(union ccb *ccb, nvme_cb_fn_t cb_fn, void *cb_arg) 534{ 535 struct nvme_request *req; 536 537 req = _nvme_allocate_request(cb_fn, cb_arg); 538 if (req != NULL) { 539 req->type = NVME_REQUEST_CCB; 540 req->u.payload = ccb; 541 } 542 543 return (req); 544} 545 546#define nvme_free_request(req) free(req, M_NVME) 547 548void nvme_notify_async_consumers(struct nvme_controller *ctrlr, 549 const struct nvme_completion *async_cpl, 550 uint32_t log_page_id, void *log_page_buffer, 551 uint32_t log_page_size); 552void nvme_notify_fail_consumers(struct nvme_controller *ctrlr); 553void nvme_notify_new_controller(struct nvme_controller *ctrlr); 554void nvme_notify_ns(struct nvme_controller *ctrlr, int nsid); 555 556void nvme_ctrlr_intx_handler(void *arg); 557void nvme_ctrlr_poll(struct nvme_controller *ctrlr); 558 559int nvme_ctrlr_suspend(struct nvme_controller *ctrlr); 560int nvme_ctrlr_resume(struct nvme_controller *ctrlr); 561 562#endif /* __NVME_PRIVATE_H__ */ 563