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