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 54 MALLOC_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 108 109 extern int32_t nvme_retry_count; 110 extern bool nvme_verbose_cmd_dump; 111 112 struct nvme_completion_poll_status { 113 struct nvme_completion cpl; 114 int done; 115 }; 116 117 #define NVME_REQUEST_VADDR 1 118 #define NVME_REQUEST_NULL 2 /* For requests with no payload. */ 119 #define NVME_REQUEST_UIO 3 120 #define NVME_REQUEST_BIO 4 121 #define NVME_REQUEST_CCB 5 122 123 struct nvme_request { 124 struct nvme_command cmd; 125 struct nvme_qpair *qpair; 126 union { 127 void *payload; 128 struct bio *bio; 129 } u; 130 uint32_t type; 131 uint32_t payload_size; 132 bool timeout; 133 nvme_cb_fn_t cb_fn; 134 void *cb_arg; 135 int32_t retries; 136 STAILQ_ENTRY(nvme_request) stailq; 137 }; 138 139 struct nvme_async_event_request { 140 struct nvme_controller *ctrlr; 141 struct nvme_request *req; 142 struct nvme_completion cpl; 143 uint32_t log_page_id; 144 uint32_t log_page_size; 145 uint8_t log_page_buffer[NVME_MAX_AER_LOG_SIZE]; 146 }; 147 148 struct nvme_tracker { 149 TAILQ_ENTRY(nvme_tracker) tailq; 150 struct nvme_request *req; 151 struct nvme_qpair *qpair; 152 sbintime_t deadline; 153 bus_dmamap_t payload_dma_map; 154 uint16_t cid; 155 156 uint64_t *prp; 157 bus_addr_t prp_bus_addr; 158 }; 159 160 enum nvme_recovery { 161 RECOVERY_NONE = 0, /* Normal operations */ 162 RECOVERY_START, /* Deadline has passed, start recovering */ 163 RECOVERY_RESET, /* This pass, initiate reset of controller */ 164 RECOVERY_WAITING, /* waiting for the reset to complete */ 165 }; 166 struct nvme_qpair { 167 struct nvme_controller *ctrlr; 168 uint32_t id; 169 int domain; 170 int cpu; 171 172 uint16_t vector; 173 int rid; 174 struct resource *res; 175 void *tag; 176 177 struct callout timer; 178 sbintime_t deadline; 179 bool timer_armed; 180 enum nvme_recovery recovery_state; 181 182 uint32_t num_entries; 183 uint32_t num_trackers; 184 uint32_t sq_tdbl_off; 185 uint32_t cq_hdbl_off; 186 187 uint32_t phase; 188 uint32_t sq_head; 189 uint32_t sq_tail; 190 uint32_t cq_head; 191 192 int64_t num_cmds; 193 int64_t num_intr_handler_calls; 194 int64_t num_retries; 195 int64_t num_failures; 196 int64_t num_ignored; 197 198 struct nvme_command *cmd; 199 struct nvme_completion *cpl; 200 201 bus_dma_tag_t dma_tag; 202 bus_dma_tag_t dma_tag_payload; 203 204 bus_dmamap_t queuemem_map; 205 uint64_t cmd_bus_addr; 206 uint64_t cpl_bus_addr; 207 208 TAILQ_HEAD(, nvme_tracker) free_tr; 209 TAILQ_HEAD(, nvme_tracker) outstanding_tr; 210 STAILQ_HEAD(, nvme_request) queued_req; 211 212 struct nvme_tracker **act_tr; 213 214 struct mtx lock __aligned(CACHE_LINE_SIZE); 215 216 } __aligned(CACHE_LINE_SIZE); 217 218 struct nvme_namespace { 219 struct nvme_controller *ctrlr; 220 struct nvme_namespace_data data; 221 uint32_t id; 222 uint32_t flags; 223 struct cdev *cdev; 224 void *cons_cookie[NVME_MAX_CONSUMERS]; 225 uint32_t boundary; 226 struct mtx lock; 227 }; 228 229 /* 230 * One of these per allocated PCI device. 231 */ 232 struct nvme_controller { 233 device_t dev; 234 235 struct mtx lock; 236 int domain; 237 uint32_t ready_timeout_in_ms; 238 uint32_t quirks; 239 #define QUIRK_DELAY_B4_CHK_RDY 1 /* Can't touch MMIO on disable */ 240 #define QUIRK_DISABLE_TIMEOUT 2 /* Disable broken completion timeout feature */ 241 #define QUIRK_INTEL_ALIGNMENT 4 /* Pre NVMe 1.3 performance alignment */ 242 #define QUIRK_AHCI 8 /* Attached via AHCI redirect */ 243 244 bus_space_tag_t bus_tag; 245 bus_space_handle_t bus_handle; 246 int resource_id; 247 struct resource *resource; 248 249 /* 250 * The NVMe spec allows for the MSI-X table to be placed in BAR 4/5, 251 * separate from the control registers which are in BAR 0/1. These 252 * members track the mapping of BAR 4/5 for that reason. 253 */ 254 int bar4_resource_id; 255 struct resource *bar4_resource; 256 257 int msi_count; 258 uint32_t enable_aborts; 259 260 uint32_t num_io_queues; 261 uint32_t max_hw_pend_io; 262 263 /* Fields for tracking progress during controller initialization. */ 264 struct intr_config_hook config_hook; 265 uint32_t ns_identified; 266 uint32_t queues_created; 267 268 struct task reset_task; 269 struct task fail_req_task; 270 struct taskqueue *taskqueue; 271 272 /* For shared legacy interrupt. */ 273 int rid; 274 struct resource *res; 275 void *tag; 276 277 /** maximum i/o size in bytes */ 278 uint32_t max_xfer_size; 279 280 /** LO and HI capacity mask */ 281 uint32_t cap_lo; 282 uint32_t cap_hi; 283 284 /** Page size and log2(page_size) - 12 that we're currently using */ 285 uint32_t page_size; 286 uint32_t mps; 287 288 /** interrupt coalescing time period (in microseconds) */ 289 uint32_t int_coal_time; 290 291 /** interrupt coalescing threshold */ 292 uint32_t int_coal_threshold; 293 294 /** timeout period in seconds */ 295 uint32_t timeout_period; 296 297 /** doorbell stride */ 298 uint32_t dstrd; 299 300 struct nvme_qpair adminq; 301 struct nvme_qpair *ioq; 302 303 struct nvme_registers *regs; 304 305 struct nvme_controller_data cdata; 306 struct nvme_namespace ns[NVME_MAX_NAMESPACES]; 307 308 struct cdev *cdev; 309 310 /** bit mask of event types currently enabled for async events */ 311 uint32_t async_event_config; 312 313 uint32_t num_aers; 314 struct nvme_async_event_request aer[NVME_MAX_ASYNC_EVENTS]; 315 316 void *cons_cookie[NVME_MAX_CONSUMERS]; 317 318 uint32_t is_resetting; 319 uint32_t is_initialized; 320 uint32_t notification_sent; 321 322 bool is_failed; 323 bool is_dying; 324 STAILQ_HEAD(, nvme_request) fail_req; 325 326 /* Host Memory Buffer */ 327 int hmb_nchunks; 328 size_t hmb_chunk; 329 bus_dma_tag_t hmb_tag; 330 struct nvme_hmb_chunk { 331 bus_dmamap_t hmbc_map; 332 void *hmbc_vaddr; 333 uint64_t hmbc_paddr; 334 } *hmb_chunks; 335 bus_dma_tag_t hmb_desc_tag; 336 bus_dmamap_t hmb_desc_map; 337 struct nvme_hmb_desc *hmb_desc_vaddr; 338 uint64_t hmb_desc_paddr; 339 }; 340 341 #define nvme_mmio_offsetof(reg) \ 342 offsetof(struct nvme_registers, reg) 343 344 #define nvme_mmio_read_4(sc, reg) \ 345 bus_space_read_4((sc)->bus_tag, (sc)->bus_handle, \ 346 nvme_mmio_offsetof(reg)) 347 348 #define nvme_mmio_write_4(sc, reg, val) \ 349 bus_space_write_4((sc)->bus_tag, (sc)->bus_handle, \ 350 nvme_mmio_offsetof(reg), val) 351 352 #define nvme_mmio_write_8(sc, reg, val) \ 353 do { \ 354 bus_space_write_4((sc)->bus_tag, (sc)->bus_handle, \ 355 nvme_mmio_offsetof(reg), val & 0xFFFFFFFF); \ 356 bus_space_write_4((sc)->bus_tag, (sc)->bus_handle, \ 357 nvme_mmio_offsetof(reg)+4, \ 358 (val & 0xFFFFFFFF00000000ULL) >> 32); \ 359 } while (0); 360 361 #define nvme_printf(ctrlr, fmt, args...) \ 362 device_printf(ctrlr->dev, fmt, ##args) 363 364 void nvme_ns_test(struct nvme_namespace *ns, u_long cmd, caddr_t arg); 365 366 void nvme_ctrlr_cmd_identify_controller(struct nvme_controller *ctrlr, 367 void *payload, 368 nvme_cb_fn_t cb_fn, void *cb_arg); 369 void nvme_ctrlr_cmd_identify_namespace(struct nvme_controller *ctrlr, 370 uint32_t nsid, void *payload, 371 nvme_cb_fn_t cb_fn, void *cb_arg); 372 void nvme_ctrlr_cmd_set_interrupt_coalescing(struct nvme_controller *ctrlr, 373 uint32_t microseconds, 374 uint32_t threshold, 375 nvme_cb_fn_t cb_fn, 376 void *cb_arg); 377 void nvme_ctrlr_cmd_get_error_page(struct nvme_controller *ctrlr, 378 struct nvme_error_information_entry *payload, 379 uint32_t num_entries, /* 0 = max */ 380 nvme_cb_fn_t cb_fn, 381 void *cb_arg); 382 void nvme_ctrlr_cmd_get_health_information_page(struct nvme_controller *ctrlr, 383 uint32_t nsid, 384 struct nvme_health_information_page *payload, 385 nvme_cb_fn_t cb_fn, 386 void *cb_arg); 387 void nvme_ctrlr_cmd_get_firmware_page(struct nvme_controller *ctrlr, 388 struct nvme_firmware_page *payload, 389 nvme_cb_fn_t cb_fn, 390 void *cb_arg); 391 void nvme_ctrlr_cmd_create_io_cq(struct nvme_controller *ctrlr, 392 struct nvme_qpair *io_que, 393 nvme_cb_fn_t cb_fn, void *cb_arg); 394 void nvme_ctrlr_cmd_create_io_sq(struct nvme_controller *ctrlr, 395 struct nvme_qpair *io_que, 396 nvme_cb_fn_t cb_fn, void *cb_arg); 397 void nvme_ctrlr_cmd_delete_io_cq(struct nvme_controller *ctrlr, 398 struct nvme_qpair *io_que, 399 nvme_cb_fn_t cb_fn, void *cb_arg); 400 void nvme_ctrlr_cmd_delete_io_sq(struct nvme_controller *ctrlr, 401 struct nvme_qpair *io_que, 402 nvme_cb_fn_t cb_fn, void *cb_arg); 403 void nvme_ctrlr_cmd_set_num_queues(struct nvme_controller *ctrlr, 404 uint32_t num_queues, nvme_cb_fn_t cb_fn, 405 void *cb_arg); 406 void nvme_ctrlr_cmd_set_async_event_config(struct nvme_controller *ctrlr, 407 uint32_t state, 408 nvme_cb_fn_t cb_fn, void *cb_arg); 409 void nvme_ctrlr_cmd_abort(struct nvme_controller *ctrlr, uint16_t cid, 410 uint16_t sqid, nvme_cb_fn_t cb_fn, void *cb_arg); 411 412 void nvme_completion_poll_cb(void *arg, const struct nvme_completion *cpl); 413 414 int nvme_ctrlr_construct(struct nvme_controller *ctrlr, device_t dev); 415 void nvme_ctrlr_destruct(struct nvme_controller *ctrlr, device_t dev); 416 void nvme_ctrlr_shutdown(struct nvme_controller *ctrlr); 417 void nvme_ctrlr_reset(struct nvme_controller *ctrlr); 418 /* ctrlr defined as void * to allow use with config_intrhook. */ 419 void nvme_ctrlr_start_config_hook(void *ctrlr_arg); 420 void nvme_ctrlr_submit_admin_request(struct nvme_controller *ctrlr, 421 struct nvme_request *req); 422 void nvme_ctrlr_submit_io_request(struct nvme_controller *ctrlr, 423 struct nvme_request *req); 424 void nvme_ctrlr_post_failed_request(struct nvme_controller *ctrlr, 425 struct nvme_request *req); 426 427 int nvme_qpair_construct(struct nvme_qpair *qpair, 428 uint32_t num_entries, uint32_t num_trackers, 429 struct nvme_controller *ctrlr); 430 void nvme_qpair_submit_tracker(struct nvme_qpair *qpair, 431 struct nvme_tracker *tr); 432 bool nvme_qpair_process_completions(struct nvme_qpair *qpair); 433 void nvme_qpair_submit_request(struct nvme_qpair *qpair, 434 struct nvme_request *req); 435 void nvme_qpair_reset(struct nvme_qpair *qpair); 436 void nvme_qpair_fail(struct nvme_qpair *qpair); 437 void nvme_qpair_manual_complete_request(struct nvme_qpair *qpair, 438 struct nvme_request *req, 439 uint32_t sct, uint32_t sc); 440 441 void nvme_admin_qpair_enable(struct nvme_qpair *qpair); 442 void nvme_admin_qpair_disable(struct nvme_qpair *qpair); 443 void nvme_admin_qpair_destroy(struct nvme_qpair *qpair); 444 445 void nvme_io_qpair_enable(struct nvme_qpair *qpair); 446 void nvme_io_qpair_disable(struct nvme_qpair *qpair); 447 void nvme_io_qpair_destroy(struct nvme_qpair *qpair); 448 449 int nvme_ns_construct(struct nvme_namespace *ns, uint32_t id, 450 struct nvme_controller *ctrlr); 451 void nvme_ns_destruct(struct nvme_namespace *ns); 452 453 void nvme_sysctl_initialize_ctrlr(struct nvme_controller *ctrlr); 454 455 void nvme_dump_command(struct nvme_command *cmd); 456 void nvme_dump_completion(struct nvme_completion *cpl); 457 458 int nvme_attach(device_t dev); 459 int nvme_shutdown(device_t dev); 460 int nvme_detach(device_t dev); 461 462 /* 463 * Wait for a command to complete using the nvme_completion_poll_cb. Used in 464 * limited contexts where the caller knows it's OK to block briefly while the 465 * command runs. The ISR will run the callback which will set status->done to 466 * true, usually within microseconds. If not, then after one second timeout 467 * handler should reset the controller and abort all outstanding requests 468 * including this polled one. If still not after ten seconds, then something is 469 * wrong with the driver, and panic is the only way to recover. 470 * 471 * Most commands using this interface aren't actual I/O to the drive's media so 472 * complete within a few microseconds. Adaptively spin for one tick to catch the 473 * vast majority of these without waiting for a tick plus scheduling delays. Since 474 * these are on startup, this drastically reduces startup time. 475 */ 476 static __inline 477 void 478 nvme_completion_poll(struct nvme_completion_poll_status *status) 479 { 480 int timeout = ticks + 10 * hz; 481 sbintime_t delta_t = SBT_1US; 482 483 while (!atomic_load_acq_int(&status->done)) { 484 if (timeout - ticks < 0) 485 panic("NVME polled command failed to complete within 10s."); 486 pause_sbt("nvme", delta_t, 0, C_PREL(1)); 487 delta_t = min(SBT_1MS, delta_t * 3 / 2); 488 } 489 } 490 491 static __inline void 492 nvme_single_map(void *arg, bus_dma_segment_t *seg, int nseg, int error) 493 { 494 uint64_t *bus_addr = (uint64_t *)arg; 495 496 KASSERT(nseg == 1, ("number of segments (%d) is not 1", nseg)); 497 if (error != 0) 498 printf("nvme_single_map err %d\n", error); 499 *bus_addr = seg[0].ds_addr; 500 } 501 502 static __inline struct nvme_request * 503 _nvme_allocate_request(nvme_cb_fn_t cb_fn, void *cb_arg) 504 { 505 struct nvme_request *req; 506 507 req = malloc(sizeof(*req), M_NVME, M_NOWAIT | M_ZERO); 508 if (req != NULL) { 509 req->cb_fn = cb_fn; 510 req->cb_arg = cb_arg; 511 req->timeout = true; 512 } 513 return (req); 514 } 515 516 static __inline struct nvme_request * 517 nvme_allocate_request_vaddr(void *payload, uint32_t payload_size, 518 nvme_cb_fn_t cb_fn, void *cb_arg) 519 { 520 struct nvme_request *req; 521 522 req = _nvme_allocate_request(cb_fn, cb_arg); 523 if (req != NULL) { 524 req->type = NVME_REQUEST_VADDR; 525 req->u.payload = payload; 526 req->payload_size = payload_size; 527 } 528 return (req); 529 } 530 531 static __inline struct nvme_request * 532 nvme_allocate_request_null(nvme_cb_fn_t cb_fn, void *cb_arg) 533 { 534 struct nvme_request *req; 535 536 req = _nvme_allocate_request(cb_fn, cb_arg); 537 if (req != NULL) 538 req->type = NVME_REQUEST_NULL; 539 return (req); 540 } 541 542 static __inline struct nvme_request * 543 nvme_allocate_request_bio(struct bio *bio, nvme_cb_fn_t cb_fn, void *cb_arg) 544 { 545 struct nvme_request *req; 546 547 req = _nvme_allocate_request(cb_fn, cb_arg); 548 if (req != NULL) { 549 req->type = NVME_REQUEST_BIO; 550 req->u.bio = bio; 551 } 552 return (req); 553 } 554 555 static __inline struct nvme_request * 556 nvme_allocate_request_ccb(union ccb *ccb, nvme_cb_fn_t cb_fn, void *cb_arg) 557 { 558 struct nvme_request *req; 559 560 req = _nvme_allocate_request(cb_fn, cb_arg); 561 if (req != NULL) { 562 req->type = NVME_REQUEST_CCB; 563 req->u.payload = ccb; 564 } 565 566 return (req); 567 } 568 569 #define nvme_free_request(req) free(req, M_NVME) 570 571 void nvme_notify_async_consumers(struct nvme_controller *ctrlr, 572 const struct nvme_completion *async_cpl, 573 uint32_t log_page_id, void *log_page_buffer, 574 uint32_t log_page_size); 575 void nvme_notify_fail_consumers(struct nvme_controller *ctrlr); 576 void nvme_notify_new_controller(struct nvme_controller *ctrlr); 577 void nvme_notify_ns(struct nvme_controller *ctrlr, int nsid); 578 579 void nvme_ctrlr_shared_handler(void *arg); 580 void nvme_ctrlr_poll(struct nvme_controller *ctrlr); 581 582 int nvme_ctrlr_suspend(struct nvme_controller *ctrlr); 583 int nvme_ctrlr_resume(struct nvme_controller *ctrlr); 584 585 #endif /* __NVME_PRIVATE_H__ */ 586