1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2015 Netflix, Inc 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer, 11 * without modification, immediately at the beginning of the file. 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 ``AS IS'' AND ANY EXPRESS OR 17 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 18 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 19 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 20 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 21 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 22 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 23 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 24 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 25 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 26 * 27 * Derived from ata_da.c: 28 * Copyright (c) 2009 Alexander Motin <mav@FreeBSD.org> 29 */ 30 31 #include <sys/cdefs.h> 32 __FBSDID("$FreeBSD$"); 33 34 #include <sys/param.h> 35 36 #ifdef _KERNEL 37 #include <sys/systm.h> 38 #include <sys/kernel.h> 39 #include <sys/bio.h> 40 #include <sys/sysctl.h> 41 #include <sys/taskqueue.h> 42 #include <sys/lock.h> 43 #include <sys/mutex.h> 44 #include <sys/conf.h> 45 #include <sys/devicestat.h> 46 #include <sys/eventhandler.h> 47 #include <sys/malloc.h> 48 #include <sys/cons.h> 49 #include <sys/proc.h> 50 #include <sys/reboot.h> 51 #include <geom/geom_disk.h> 52 #endif /* _KERNEL */ 53 54 #ifndef _KERNEL 55 #include <stdio.h> 56 #include <string.h> 57 #endif /* _KERNEL */ 58 59 #include <cam/cam.h> 60 #include <cam/cam_ccb.h> 61 #include <cam/cam_periph.h> 62 #include <cam/cam_xpt_periph.h> 63 #include <cam/cam_sim.h> 64 #include <cam/cam_iosched.h> 65 66 #include <cam/nvme/nvme_all.h> 67 68 typedef enum { 69 NDA_STATE_NORMAL 70 } nda_state; 71 72 typedef enum { 73 NDA_FLAG_OPEN = 0x0001, 74 NDA_FLAG_DIRTY = 0x0002, 75 NDA_FLAG_SCTX_INIT = 0x0004, 76 } nda_flags; 77 78 typedef enum { 79 NDA_Q_4K = 0x01, 80 NDA_Q_NONE = 0x00, 81 } nda_quirks; 82 83 #define NDA_Q_BIT_STRING \ 84 "\020" \ 85 "\001Bit 0" 86 87 typedef enum { 88 NDA_CCB_BUFFER_IO = 0x01, 89 NDA_CCB_DUMP = 0x02, 90 NDA_CCB_TRIM = 0x03, 91 NDA_CCB_TYPE_MASK = 0x0F, 92 } nda_ccb_state; 93 94 /* Offsets into our private area for storing information */ 95 #define ccb_state ccb_h.ppriv_field0 96 #define ccb_bp ccb_h.ppriv_ptr1 /* For NDA_CCB_BUFFER_IO */ 97 #define ccb_trim ccb_h.ppriv_ptr1 /* For NDA_CCB_TRIM */ 98 99 struct nda_softc { 100 struct cam_iosched_softc *cam_iosched; 101 int outstanding_cmds; /* Number of active commands */ 102 int refcount; /* Active xpt_action() calls */ 103 nda_state state; 104 nda_flags flags; 105 nda_quirks quirks; 106 int unmappedio; 107 quad_t deletes; 108 quad_t dsm_req; 109 uint32_t nsid; /* Namespace ID for this nda device */ 110 struct disk *disk; 111 struct task sysctl_task; 112 struct sysctl_ctx_list sysctl_ctx; 113 struct sysctl_oid *sysctl_tree; 114 #ifdef CAM_TEST_FAILURE 115 int force_read_error; 116 int force_write_error; 117 int periodic_read_error; 118 int periodic_read_count; 119 #endif 120 #ifdef CAM_IO_STATS 121 struct sysctl_ctx_list sysctl_stats_ctx; 122 struct sysctl_oid *sysctl_stats_tree; 123 u_int timeouts; 124 u_int errors; 125 u_int invalidations; 126 #endif 127 }; 128 129 struct nda_trim_request { 130 union { 131 struct nvme_dsm_range dsm; 132 uint8_t data[NVME_MAX_DSM_TRIM]; 133 }; 134 TAILQ_HEAD(, bio) bps; 135 }; 136 137 /* Need quirk table */ 138 139 static disk_strategy_t ndastrategy; 140 static dumper_t ndadump; 141 static periph_init_t ndainit; 142 static void ndaasync(void *callback_arg, u_int32_t code, 143 struct cam_path *path, void *arg); 144 static void ndasysctlinit(void *context, int pending); 145 static periph_ctor_t ndaregister; 146 static periph_dtor_t ndacleanup; 147 static periph_start_t ndastart; 148 static periph_oninv_t ndaoninvalidate; 149 static void ndadone(struct cam_periph *periph, 150 union ccb *done_ccb); 151 static int ndaerror(union ccb *ccb, u_int32_t cam_flags, 152 u_int32_t sense_flags); 153 static void ndashutdown(void *arg, int howto); 154 static void ndasuspend(void *arg); 155 156 #ifndef NDA_DEFAULT_SEND_ORDERED 157 #define NDA_DEFAULT_SEND_ORDERED 1 158 #endif 159 #ifndef NDA_DEFAULT_TIMEOUT 160 #define NDA_DEFAULT_TIMEOUT 30 /* Timeout in seconds */ 161 #endif 162 #ifndef NDA_DEFAULT_RETRY 163 #define NDA_DEFAULT_RETRY 4 164 #endif 165 #ifndef NDA_MAX_TRIM_ENTRIES 166 #define NDA_MAX_TRIM_ENTRIES (NVME_MAX_DSM_TRIM / sizeof(struct nvme_dsm_range))/* Number of DSM trims to use, max 256 */ 167 #endif 168 169 static SYSCTL_NODE(_kern_cam, OID_AUTO, nda, CTLFLAG_RD, 0, 170 "CAM Direct Access Disk driver"); 171 172 //static int nda_retry_count = NDA_DEFAULT_RETRY; 173 static int nda_send_ordered = NDA_DEFAULT_SEND_ORDERED; 174 static int nda_default_timeout = NDA_DEFAULT_TIMEOUT; 175 static int nda_max_trim_entries = NDA_MAX_TRIM_ENTRIES; 176 SYSCTL_INT(_kern_cam_nda, OID_AUTO, max_trim, CTLFLAG_RDTUN, 177 &nda_max_trim_entries, NDA_MAX_TRIM_ENTRIES, 178 "Maximum number of BIO_DELETE to send down as a DSM TRIM."); 179 180 /* 181 * All NVMe media is non-rotational, so all nvme device instances 182 * share this to implement the sysctl. 183 */ 184 static int nda_rotating_media = 0; 185 186 static struct periph_driver ndadriver = 187 { 188 ndainit, "nda", 189 TAILQ_HEAD_INITIALIZER(ndadriver.units), /* generation */ 0 190 }; 191 192 PERIPHDRIVER_DECLARE(nda, ndadriver); 193 194 static MALLOC_DEFINE(M_NVMEDA, "nvme_da", "nvme_da buffers"); 195 196 /* 197 * nice wrappers. Maybe these belong in nvme_all.c instead of 198 * here, but this is the only place that uses these. Should 199 * we ever grow another NVME periph, we should move them 200 * all there wholesale. 201 */ 202 203 static void 204 nda_nvme_flush(struct nda_softc *softc, struct ccb_nvmeio *nvmeio) 205 { 206 cam_fill_nvmeio(nvmeio, 207 0, /* retries */ 208 ndadone, /* cbfcnp */ 209 CAM_DIR_NONE, /* flags */ 210 NULL, /* data_ptr */ 211 0, /* dxfer_len */ 212 nda_default_timeout * 1000); /* timeout 30s */ 213 nvme_ns_flush_cmd(&nvmeio->cmd, softc->nsid); 214 } 215 216 static void 217 nda_nvme_trim(struct nda_softc *softc, struct ccb_nvmeio *nvmeio, 218 void *payload, uint32_t num_ranges) 219 { 220 cam_fill_nvmeio(nvmeio, 221 0, /* retries */ 222 ndadone, /* cbfcnp */ 223 CAM_DIR_OUT, /* flags */ 224 payload, /* data_ptr */ 225 num_ranges * sizeof(struct nvme_dsm_range), /* dxfer_len */ 226 nda_default_timeout * 1000); /* timeout 30s */ 227 nvme_ns_trim_cmd(&nvmeio->cmd, softc->nsid, num_ranges); 228 } 229 230 static void 231 nda_nvme_write(struct nda_softc *softc, struct ccb_nvmeio *nvmeio, 232 void *payload, uint64_t lba, uint32_t len, uint32_t count) 233 { 234 cam_fill_nvmeio(nvmeio, 235 0, /* retries */ 236 ndadone, /* cbfcnp */ 237 CAM_DIR_OUT, /* flags */ 238 payload, /* data_ptr */ 239 len, /* dxfer_len */ 240 nda_default_timeout * 1000); /* timeout 30s */ 241 nvme_ns_write_cmd(&nvmeio->cmd, softc->nsid, lba, count); 242 } 243 244 static void 245 nda_nvme_rw_bio(struct nda_softc *softc, struct ccb_nvmeio *nvmeio, 246 struct bio *bp, uint32_t rwcmd) 247 { 248 int flags = rwcmd == NVME_OPC_READ ? CAM_DIR_IN : CAM_DIR_OUT; 249 void *payload; 250 uint64_t lba; 251 uint32_t count; 252 253 if (bp->bio_flags & BIO_UNMAPPED) { 254 flags |= CAM_DATA_BIO; 255 payload = bp; 256 } else { 257 payload = bp->bio_data; 258 } 259 260 lba = bp->bio_pblkno; 261 count = bp->bio_bcount / softc->disk->d_sectorsize; 262 263 cam_fill_nvmeio(nvmeio, 264 0, /* retries */ 265 ndadone, /* cbfcnp */ 266 flags, /* flags */ 267 payload, /* data_ptr */ 268 bp->bio_bcount, /* dxfer_len */ 269 nda_default_timeout * 1000); /* timeout 30s */ 270 nvme_ns_rw_cmd(&nvmeio->cmd, rwcmd, softc->nsid, lba, count); 271 } 272 273 static int 274 ndaopen(struct disk *dp) 275 { 276 struct cam_periph *periph; 277 struct nda_softc *softc; 278 int error; 279 280 periph = (struct cam_periph *)dp->d_drv1; 281 if (cam_periph_acquire(periph) != 0) { 282 return(ENXIO); 283 } 284 285 cam_periph_lock(periph); 286 if ((error = cam_periph_hold(periph, PRIBIO|PCATCH)) != 0) { 287 cam_periph_unlock(periph); 288 cam_periph_release(periph); 289 return (error); 290 } 291 292 CAM_DEBUG(periph->path, CAM_DEBUG_TRACE | CAM_DEBUG_PERIPH, 293 ("ndaopen\n")); 294 295 softc = (struct nda_softc *)periph->softc; 296 softc->flags |= NDA_FLAG_OPEN; 297 298 cam_periph_unhold(periph); 299 cam_periph_unlock(periph); 300 return (0); 301 } 302 303 static int 304 ndaclose(struct disk *dp) 305 { 306 struct cam_periph *periph; 307 struct nda_softc *softc; 308 union ccb *ccb; 309 int error; 310 311 periph = (struct cam_periph *)dp->d_drv1; 312 softc = (struct nda_softc *)periph->softc; 313 cam_periph_lock(periph); 314 315 CAM_DEBUG(periph->path, CAM_DEBUG_TRACE | CAM_DEBUG_PERIPH, 316 ("ndaclose\n")); 317 318 if ((softc->flags & NDA_FLAG_DIRTY) != 0 && 319 (periph->flags & CAM_PERIPH_INVALID) == 0 && 320 cam_periph_hold(periph, PRIBIO) == 0) { 321 322 ccb = cam_periph_getccb(periph, CAM_PRIORITY_NORMAL); 323 nda_nvme_flush(softc, &ccb->nvmeio); 324 error = cam_periph_runccb(ccb, ndaerror, /*cam_flags*/0, 325 /*sense_flags*/0, softc->disk->d_devstat); 326 327 if (error != 0) 328 xpt_print(periph->path, "Synchronize cache failed\n"); 329 else 330 softc->flags &= ~NDA_FLAG_DIRTY; 331 xpt_release_ccb(ccb); 332 cam_periph_unhold(periph); 333 } 334 335 softc->flags &= ~NDA_FLAG_OPEN; 336 337 while (softc->refcount != 0) 338 cam_periph_sleep(periph, &softc->refcount, PRIBIO, "ndaclose", 1); 339 cam_periph_unlock(periph); 340 cam_periph_release(periph); 341 return (0); 342 } 343 344 static void 345 ndaschedule(struct cam_periph *periph) 346 { 347 struct nda_softc *softc = (struct nda_softc *)periph->softc; 348 349 if (softc->state != NDA_STATE_NORMAL) 350 return; 351 352 cam_iosched_schedule(softc->cam_iosched, periph); 353 } 354 355 /* 356 * Actually translate the requested transfer into one the physical driver 357 * can understand. The transfer is described by a buf and will include 358 * only one physical transfer. 359 */ 360 static void 361 ndastrategy(struct bio *bp) 362 { 363 struct cam_periph *periph; 364 struct nda_softc *softc; 365 366 periph = (struct cam_periph *)bp->bio_disk->d_drv1; 367 softc = (struct nda_softc *)periph->softc; 368 369 cam_periph_lock(periph); 370 371 CAM_DEBUG(periph->path, CAM_DEBUG_TRACE, ("ndastrategy(%p)\n", bp)); 372 373 /* 374 * If the device has been made invalid, error out 375 */ 376 if ((periph->flags & CAM_PERIPH_INVALID) != 0) { 377 cam_periph_unlock(periph); 378 biofinish(bp, NULL, ENXIO); 379 return; 380 } 381 382 if (bp->bio_cmd == BIO_DELETE) 383 softc->deletes++; 384 385 /* 386 * Place it in the queue of disk activities for this disk 387 */ 388 cam_iosched_queue_work(softc->cam_iosched, bp); 389 390 /* 391 * Schedule ourselves for performing the work. 392 */ 393 ndaschedule(periph); 394 cam_periph_unlock(periph); 395 396 return; 397 } 398 399 static int 400 ndadump(void *arg, void *virtual, vm_offset_t physical, off_t offset, size_t length) 401 { 402 struct cam_periph *periph; 403 struct nda_softc *softc; 404 u_int secsize; 405 struct ccb_nvmeio nvmeio; 406 struct disk *dp; 407 uint64_t lba; 408 uint32_t count; 409 int error = 0; 410 411 dp = arg; 412 periph = dp->d_drv1; 413 softc = (struct nda_softc *)periph->softc; 414 secsize = softc->disk->d_sectorsize; 415 lba = offset / secsize; 416 count = length / secsize; 417 418 if ((periph->flags & CAM_PERIPH_INVALID) != 0) 419 return (ENXIO); 420 421 /* xpt_get_ccb returns a zero'd allocation for the ccb, mimic that here */ 422 memset(&nvmeio, 0, sizeof(nvmeio)); 423 if (length > 0) { 424 xpt_setup_ccb(&nvmeio.ccb_h, periph->path, CAM_PRIORITY_NORMAL); 425 nvmeio.ccb_state = NDA_CCB_DUMP; 426 nda_nvme_write(softc, &nvmeio, virtual, lba, length, count); 427 error = cam_periph_runccb((union ccb *)&nvmeio, cam_periph_error, 428 0, SF_NO_RECOVERY | SF_NO_RETRY, NULL); 429 if (error != 0) 430 printf("Aborting dump due to I/O error %d.\n", error); 431 432 return (error); 433 } 434 435 /* Flush */ 436 xpt_setup_ccb(&nvmeio.ccb_h, periph->path, CAM_PRIORITY_NORMAL); 437 438 nvmeio.ccb_state = NDA_CCB_DUMP; 439 nda_nvme_flush(softc, &nvmeio); 440 error = cam_periph_runccb((union ccb *)&nvmeio, cam_periph_error, 441 0, SF_NO_RECOVERY | SF_NO_RETRY, NULL); 442 if (error != 0) 443 xpt_print(periph->path, "flush cmd failed\n"); 444 return (error); 445 } 446 447 static void 448 ndainit(void) 449 { 450 cam_status status; 451 452 /* 453 * Install a global async callback. This callback will 454 * receive async callbacks like "new device found". 455 */ 456 status = xpt_register_async(AC_FOUND_DEVICE, ndaasync, NULL, NULL); 457 458 if (status != CAM_REQ_CMP) { 459 printf("nda: Failed to attach master async callback " 460 "due to status 0x%x!\n", status); 461 } else if (nda_send_ordered) { 462 463 /* Register our event handlers */ 464 if ((EVENTHANDLER_REGISTER(power_suspend, ndasuspend, 465 NULL, EVENTHANDLER_PRI_LAST)) == NULL) 466 printf("ndainit: power event registration failed!\n"); 467 if ((EVENTHANDLER_REGISTER(shutdown_post_sync, ndashutdown, 468 NULL, SHUTDOWN_PRI_DEFAULT)) == NULL) 469 printf("ndainit: shutdown event registration failed!\n"); 470 } 471 } 472 473 /* 474 * Callback from GEOM, called when it has finished cleaning up its 475 * resources. 476 */ 477 static void 478 ndadiskgonecb(struct disk *dp) 479 { 480 struct cam_periph *periph; 481 482 periph = (struct cam_periph *)dp->d_drv1; 483 484 cam_periph_release(periph); 485 } 486 487 static void 488 ndaoninvalidate(struct cam_periph *periph) 489 { 490 struct nda_softc *softc; 491 492 softc = (struct nda_softc *)periph->softc; 493 494 /* 495 * De-register any async callbacks. 496 */ 497 xpt_register_async(0, ndaasync, periph, periph->path); 498 #ifdef CAM_IO_STATS 499 softc->invalidations++; 500 #endif 501 502 /* 503 * Return all queued I/O with ENXIO. 504 * XXX Handle any transactions queued to the card 505 * with XPT_ABORT_CCB. 506 */ 507 cam_iosched_flush(softc->cam_iosched, NULL, ENXIO); 508 509 disk_gone(softc->disk); 510 } 511 512 static void 513 ndacleanup(struct cam_periph *periph) 514 { 515 struct nda_softc *softc; 516 517 softc = (struct nda_softc *)periph->softc; 518 519 cam_periph_unlock(periph); 520 521 cam_iosched_fini(softc->cam_iosched); 522 523 /* 524 * If we can't free the sysctl tree, oh well... 525 */ 526 if ((softc->flags & NDA_FLAG_SCTX_INIT) != 0) { 527 #ifdef CAM_IO_STATS 528 if (sysctl_ctx_free(&softc->sysctl_stats_ctx) != 0) 529 xpt_print(periph->path, 530 "can't remove sysctl stats context\n"); 531 #endif 532 if (sysctl_ctx_free(&softc->sysctl_ctx) != 0) 533 xpt_print(periph->path, 534 "can't remove sysctl context\n"); 535 } 536 537 disk_destroy(softc->disk); 538 free(softc, M_DEVBUF); 539 cam_periph_lock(periph); 540 } 541 542 static void 543 ndaasync(void *callback_arg, u_int32_t code, 544 struct cam_path *path, void *arg) 545 { 546 struct cam_periph *periph; 547 548 periph = (struct cam_periph *)callback_arg; 549 switch (code) { 550 case AC_FOUND_DEVICE: 551 { 552 struct ccb_getdev *cgd; 553 cam_status status; 554 555 cgd = (struct ccb_getdev *)arg; 556 if (cgd == NULL) 557 break; 558 559 if (cgd->protocol != PROTO_NVME) 560 break; 561 562 /* 563 * Allocate a peripheral instance for 564 * this device and start the probe 565 * process. 566 */ 567 status = cam_periph_alloc(ndaregister, ndaoninvalidate, 568 ndacleanup, ndastart, 569 "nda", CAM_PERIPH_BIO, 570 path, ndaasync, 571 AC_FOUND_DEVICE, cgd); 572 573 if (status != CAM_REQ_CMP 574 && status != CAM_REQ_INPROG) 575 printf("ndaasync: Unable to attach to new device " 576 "due to status 0x%x\n", status); 577 break; 578 } 579 case AC_ADVINFO_CHANGED: 580 { 581 uintptr_t buftype; 582 583 buftype = (uintptr_t)arg; 584 if (buftype == CDAI_TYPE_PHYS_PATH) { 585 struct nda_softc *softc; 586 587 softc = periph->softc; 588 disk_attr_changed(softc->disk, "GEOM::physpath", 589 M_NOWAIT); 590 } 591 break; 592 } 593 case AC_LOST_DEVICE: 594 default: 595 cam_periph_async(periph, code, path, arg); 596 break; 597 } 598 } 599 600 static void 601 ndasysctlinit(void *context, int pending) 602 { 603 struct cam_periph *periph; 604 struct nda_softc *softc; 605 char tmpstr[32], tmpstr2[16]; 606 607 periph = (struct cam_periph *)context; 608 609 /* periph was held for us when this task was enqueued */ 610 if ((periph->flags & CAM_PERIPH_INVALID) != 0) { 611 cam_periph_release(periph); 612 return; 613 } 614 615 softc = (struct nda_softc *)periph->softc; 616 snprintf(tmpstr, sizeof(tmpstr), "CAM NDA unit %d", periph->unit_number); 617 snprintf(tmpstr2, sizeof(tmpstr2), "%d", periph->unit_number); 618 619 sysctl_ctx_init(&softc->sysctl_ctx); 620 softc->flags |= NDA_FLAG_SCTX_INIT; 621 softc->sysctl_tree = SYSCTL_ADD_NODE_WITH_LABEL(&softc->sysctl_ctx, 622 SYSCTL_STATIC_CHILDREN(_kern_cam_nda), OID_AUTO, tmpstr2, 623 CTLFLAG_RD, 0, tmpstr, "device_index"); 624 if (softc->sysctl_tree == NULL) { 625 printf("ndasysctlinit: unable to allocate sysctl tree\n"); 626 cam_periph_release(periph); 627 return; 628 } 629 630 SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree), 631 OID_AUTO, "unmapped_io", CTLFLAG_RD, 632 &softc->unmappedio, 0, "Unmapped I/O leaf"); 633 634 SYSCTL_ADD_QUAD(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree), 635 OID_AUTO, "deletes", CTLFLAG_RD, 636 &softc->deletes, "Number of BIO_DELETE requests"); 637 638 SYSCTL_ADD_QUAD(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree), 639 OID_AUTO, "dsm_req", CTLFLAG_RD, 640 &softc->dsm_req, "Number of DSM requests sent to SIM"); 641 642 SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree), 643 OID_AUTO, "rotating", CTLFLAG_RD, &nda_rotating_media, 1, 644 "Rotating media"); 645 646 #ifdef CAM_IO_STATS 647 softc->sysctl_stats_tree = SYSCTL_ADD_NODE(&softc->sysctl_stats_ctx, 648 SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO, "stats", 649 CTLFLAG_RD, 0, "Statistics"); 650 if (softc->sysctl_stats_tree == NULL) { 651 printf("ndasysctlinit: unable to allocate sysctl tree for stats\n"); 652 cam_periph_release(periph); 653 return; 654 } 655 SYSCTL_ADD_INT(&softc->sysctl_stats_ctx, 656 SYSCTL_CHILDREN(softc->sysctl_stats_tree), 657 OID_AUTO, "timeouts", CTLFLAG_RD, 658 &softc->timeouts, 0, 659 "Device timeouts reported by the SIM"); 660 SYSCTL_ADD_INT(&softc->sysctl_stats_ctx, 661 SYSCTL_CHILDREN(softc->sysctl_stats_tree), 662 OID_AUTO, "errors", CTLFLAG_RD, 663 &softc->errors, 0, 664 "Transport errors reported by the SIM."); 665 SYSCTL_ADD_INT(&softc->sysctl_stats_ctx, 666 SYSCTL_CHILDREN(softc->sysctl_stats_tree), 667 OID_AUTO, "pack_invalidations", CTLFLAG_RD, 668 &softc->invalidations, 0, 669 "Device pack invalidations."); 670 #endif 671 672 #ifdef CAM_TEST_FAILURE 673 SYSCTL_ADD_PROC(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree), 674 OID_AUTO, "invalidate", CTLTYPE_U64 | CTLFLAG_RW | CTLFLAG_MPSAFE, 675 periph, 0, cam_periph_invalidate_sysctl, "I", 676 "Write 1 to invalidate the drive immediately"); 677 #endif 678 679 cam_iosched_sysctl_init(softc->cam_iosched, &softc->sysctl_ctx, 680 softc->sysctl_tree); 681 682 cam_periph_release(periph); 683 } 684 685 static int 686 ndagetattr(struct bio *bp) 687 { 688 int ret; 689 struct cam_periph *periph; 690 691 periph = (struct cam_periph *)bp->bio_disk->d_drv1; 692 cam_periph_lock(periph); 693 ret = xpt_getattr(bp->bio_data, bp->bio_length, bp->bio_attribute, 694 periph->path); 695 cam_periph_unlock(periph); 696 if (ret == 0) 697 bp->bio_completed = bp->bio_length; 698 return ret; 699 } 700 701 static cam_status 702 ndaregister(struct cam_periph *periph, void *arg) 703 { 704 struct nda_softc *softc; 705 struct disk *disk; 706 struct ccb_pathinq cpi; 707 const struct nvme_namespace_data *nsd; 708 const struct nvme_controller_data *cd; 709 char announce_buf[80]; 710 uint8_t flbas_fmt, lbads, vwc_present; 711 u_int maxio; 712 int quirks; 713 714 nsd = nvme_get_identify_ns(periph); 715 cd = nvme_get_identify_cntrl(periph); 716 717 softc = (struct nda_softc *)malloc(sizeof(*softc), M_DEVBUF, 718 M_NOWAIT | M_ZERO); 719 720 if (softc == NULL) { 721 printf("ndaregister: Unable to probe new device. " 722 "Unable to allocate softc\n"); 723 return(CAM_REQ_CMP_ERR); 724 } 725 726 if (cam_iosched_init(&softc->cam_iosched, periph) != 0) { 727 printf("ndaregister: Unable to probe new device. " 728 "Unable to allocate iosched memory\n"); 729 free(softc, M_DEVBUF); 730 return(CAM_REQ_CMP_ERR); 731 } 732 733 /* ident_data parsing */ 734 735 periph->softc = softc; 736 737 softc->quirks = NDA_Q_NONE; 738 739 xpt_path_inq(&cpi, periph->path); 740 741 TASK_INIT(&softc->sysctl_task, 0, ndasysctlinit, periph); 742 743 /* 744 * The name space ID is the lun, save it for later I/O 745 */ 746 softc->nsid = (uint32_t)xpt_path_lun_id(periph->path); 747 748 /* 749 * Register this media as a disk 750 */ 751 (void)cam_periph_hold(periph, PRIBIO); 752 cam_periph_unlock(periph); 753 snprintf(announce_buf, sizeof(announce_buf), 754 "kern.cam.nda.%d.quirks", periph->unit_number); 755 quirks = softc->quirks; 756 TUNABLE_INT_FETCH(announce_buf, &quirks); 757 softc->quirks = quirks; 758 cam_iosched_set_sort_queue(softc->cam_iosched, 0); 759 softc->disk = disk = disk_alloc(); 760 strlcpy(softc->disk->d_descr, cd->mn, 761 MIN(sizeof(softc->disk->d_descr), sizeof(cd->mn))); 762 strlcpy(softc->disk->d_ident, cd->sn, 763 MIN(sizeof(softc->disk->d_ident), sizeof(cd->sn))); 764 disk->d_rotation_rate = DISK_RR_NON_ROTATING; 765 disk->d_open = ndaopen; 766 disk->d_close = ndaclose; 767 disk->d_strategy = ndastrategy; 768 disk->d_getattr = ndagetattr; 769 disk->d_dump = ndadump; 770 disk->d_gone = ndadiskgonecb; 771 disk->d_name = "nda"; 772 disk->d_drv1 = periph; 773 disk->d_unit = periph->unit_number; 774 maxio = cpi.maxio; /* Honor max I/O size of SIM */ 775 if (maxio == 0) 776 maxio = DFLTPHYS; /* traditional default */ 777 else if (maxio > MAXPHYS) 778 maxio = MAXPHYS; /* for safety */ 779 disk->d_maxsize = maxio; 780 flbas_fmt = (nsd->flbas >> NVME_NS_DATA_FLBAS_FORMAT_SHIFT) & 781 NVME_NS_DATA_FLBAS_FORMAT_MASK; 782 lbads = (nsd->lbaf[flbas_fmt] >> NVME_NS_DATA_LBAF_LBADS_SHIFT) & 783 NVME_NS_DATA_LBAF_LBADS_MASK; 784 disk->d_sectorsize = 1 << lbads; 785 disk->d_mediasize = (off_t)(disk->d_sectorsize * nsd->nsze); 786 disk->d_delmaxsize = disk->d_mediasize; 787 disk->d_flags = DISKFLAG_DIRECT_COMPLETION; 788 // if (cd->oncs.dsm) // XXX broken? 789 disk->d_flags |= DISKFLAG_CANDELETE; 790 vwc_present = (cd->vwc >> NVME_CTRLR_DATA_VWC_PRESENT_SHIFT) & 791 NVME_CTRLR_DATA_VWC_PRESENT_MASK; 792 if (vwc_present) 793 disk->d_flags |= DISKFLAG_CANFLUSHCACHE; 794 if ((cpi.hba_misc & PIM_UNMAPPED) != 0) { 795 disk->d_flags |= DISKFLAG_UNMAPPED_BIO; 796 softc->unmappedio = 1; 797 } 798 /* 799 * d_ident and d_descr are both far bigger than the length of either 800 * the serial or model number strings. 801 */ 802 nvme_strvis(disk->d_descr, cd->mn, 803 sizeof(disk->d_descr), NVME_MODEL_NUMBER_LENGTH); 804 nvme_strvis(disk->d_ident, cd->sn, 805 sizeof(disk->d_ident), NVME_SERIAL_NUMBER_LENGTH); 806 disk->d_hba_vendor = cpi.hba_vendor; 807 disk->d_hba_device = cpi.hba_device; 808 disk->d_hba_subvendor = cpi.hba_subvendor; 809 disk->d_hba_subdevice = cpi.hba_subdevice; 810 disk->d_stripesize = disk->d_sectorsize; 811 disk->d_stripeoffset = 0; 812 disk->d_devstat = devstat_new_entry(periph->periph_name, 813 periph->unit_number, disk->d_sectorsize, 814 DEVSTAT_ALL_SUPPORTED, 815 DEVSTAT_TYPE_DIRECT | XPORT_DEVSTAT_TYPE(cpi.transport), 816 DEVSTAT_PRIORITY_DISK); 817 /* 818 * Add alias for older nvd drives to ease transition. 819 */ 820 /* disk_add_alias(disk, "nvd"); Have reports of this causing problems */ 821 822 /* 823 * Acquire a reference to the periph before we register with GEOM. 824 * We'll release this reference once GEOM calls us back (via 825 * ndadiskgonecb()) telling us that our provider has been freed. 826 */ 827 if (cam_periph_acquire(periph) != 0) { 828 xpt_print(periph->path, "%s: lost periph during " 829 "registration!\n", __func__); 830 cam_periph_lock(periph); 831 return (CAM_REQ_CMP_ERR); 832 } 833 disk_create(softc->disk, DISK_VERSION); 834 cam_periph_lock(periph); 835 cam_periph_unhold(periph); 836 837 snprintf(announce_buf, sizeof(announce_buf), 838 "%juMB (%ju %u byte sectors)", 839 (uintmax_t)((uintmax_t)disk->d_mediasize / (1024*1024)), 840 (uintmax_t)disk->d_mediasize / disk->d_sectorsize, 841 disk->d_sectorsize); 842 xpt_announce_periph(periph, announce_buf); 843 xpt_announce_quirks(periph, softc->quirks, NDA_Q_BIT_STRING); 844 845 /* 846 * Create our sysctl variables, now that we know 847 * we have successfully attached. 848 */ 849 if (cam_periph_acquire(periph) == 0) 850 taskqueue_enqueue(taskqueue_thread, &softc->sysctl_task); 851 852 /* 853 * Register for device going away and info about the drive 854 * changing (though with NVMe, it can't) 855 */ 856 xpt_register_async(AC_LOST_DEVICE | AC_ADVINFO_CHANGED, 857 ndaasync, periph, periph->path); 858 859 softc->state = NDA_STATE_NORMAL; 860 return(CAM_REQ_CMP); 861 } 862 863 static void 864 ndastart(struct cam_periph *periph, union ccb *start_ccb) 865 { 866 struct nda_softc *softc = (struct nda_softc *)periph->softc; 867 struct ccb_nvmeio *nvmeio = &start_ccb->nvmeio; 868 869 CAM_DEBUG(periph->path, CAM_DEBUG_TRACE, ("ndastart\n")); 870 871 switch (softc->state) { 872 case NDA_STATE_NORMAL: 873 { 874 struct bio *bp; 875 876 bp = cam_iosched_next_bio(softc->cam_iosched); 877 CAM_DEBUG(periph->path, CAM_DEBUG_TRACE, ("ndastart: bio %p\n", bp)); 878 if (bp == NULL) { 879 xpt_release_ccb(start_ccb); 880 break; 881 } 882 883 switch (bp->bio_cmd) { 884 case BIO_WRITE: 885 softc->flags |= NDA_FLAG_DIRTY; 886 /* FALLTHROUGH */ 887 case BIO_READ: 888 { 889 #ifdef CAM_TEST_FAILURE 890 int fail = 0; 891 892 /* 893 * Support the failure ioctls. If the command is a 894 * read, and there are pending forced read errors, or 895 * if a write and pending write errors, then fail this 896 * operation with EIO. This is useful for testing 897 * purposes. Also, support having every Nth read fail. 898 * 899 * This is a rather blunt tool. 900 */ 901 if (bp->bio_cmd == BIO_READ) { 902 if (softc->force_read_error) { 903 softc->force_read_error--; 904 fail = 1; 905 } 906 if (softc->periodic_read_error > 0) { 907 if (++softc->periodic_read_count >= 908 softc->periodic_read_error) { 909 softc->periodic_read_count = 0; 910 fail = 1; 911 } 912 } 913 } else { 914 if (softc->force_write_error) { 915 softc->force_write_error--; 916 fail = 1; 917 } 918 } 919 if (fail) { 920 biofinish(bp, NULL, EIO); 921 xpt_release_ccb(start_ccb); 922 ndaschedule(periph); 923 return; 924 } 925 #endif 926 KASSERT((bp->bio_flags & BIO_UNMAPPED) == 0 || 927 round_page(bp->bio_bcount + bp->bio_ma_offset) / 928 PAGE_SIZE == bp->bio_ma_n, 929 ("Short bio %p", bp)); 930 nda_nvme_rw_bio(softc, &start_ccb->nvmeio, bp, bp->bio_cmd == BIO_READ ? 931 NVME_OPC_READ : NVME_OPC_WRITE); 932 break; 933 } 934 case BIO_DELETE: 935 { 936 struct nvme_dsm_range *dsm_range, *dsm_end; 937 struct nda_trim_request *trim; 938 struct bio *bp1; 939 int ents; 940 941 trim = malloc(sizeof(*trim), M_NVMEDA, M_ZERO | M_NOWAIT); 942 if (trim == NULL) { 943 biofinish(bp, NULL, ENOMEM); 944 xpt_release_ccb(start_ccb); 945 ndaschedule(periph); 946 return; 947 } 948 TAILQ_INIT(&trim->bps); 949 bp1 = bp; 950 ents = sizeof(trim->data) / sizeof(struct nvme_dsm_range); 951 ents = min(ents, nda_max_trim_entries); 952 dsm_range = &trim->dsm; 953 dsm_end = dsm_range + ents; 954 do { 955 TAILQ_INSERT_TAIL(&trim->bps, bp1, bio_queue); 956 dsm_range->length = 957 htole32(bp1->bio_bcount / softc->disk->d_sectorsize); 958 dsm_range->starting_lba = 959 htole64(bp1->bio_offset / softc->disk->d_sectorsize); 960 dsm_range++; 961 if (dsm_range >= dsm_end) 962 break; 963 bp1 = cam_iosched_next_trim(softc->cam_iosched); 964 /* XXX -- Could collapse adjacent ranges, but we don't for now */ 965 /* XXX -- Could limit based on total payload size */ 966 } while (bp1 != NULL); 967 start_ccb->ccb_trim = trim; 968 softc->dsm_req++; 969 nda_nvme_trim(softc, &start_ccb->nvmeio, &trim->dsm, 970 dsm_range - &trim->dsm); 971 start_ccb->ccb_state = NDA_CCB_TRIM; 972 /* 973 * Note: We can have multiple TRIMs in flight, so we don't call 974 * cam_iosched_submit_trim(softc->cam_iosched); 975 * since that forces the I/O scheduler to only schedule one at a time. 976 * On NVMe drives, this is a performance disaster. 977 */ 978 goto out; 979 } 980 case BIO_FLUSH: 981 nda_nvme_flush(softc, nvmeio); 982 break; 983 } 984 start_ccb->ccb_state = NDA_CCB_BUFFER_IO; 985 start_ccb->ccb_bp = bp; 986 out: 987 start_ccb->ccb_h.flags |= CAM_UNLOCKED; 988 softc->outstanding_cmds++; 989 softc->refcount++; 990 cam_periph_unlock(periph); 991 xpt_action(start_ccb); 992 cam_periph_lock(periph); 993 994 /* May have more work to do, so ensure we stay scheduled */ 995 ndaschedule(periph); 996 break; 997 } 998 } 999 } 1000 1001 static void 1002 ndadone(struct cam_periph *periph, union ccb *done_ccb) 1003 { 1004 struct nda_softc *softc; 1005 struct ccb_nvmeio *nvmeio = &done_ccb->nvmeio; 1006 struct cam_path *path; 1007 int state; 1008 1009 softc = (struct nda_softc *)periph->softc; 1010 path = done_ccb->ccb_h.path; 1011 1012 CAM_DEBUG(path, CAM_DEBUG_TRACE, ("ndadone\n")); 1013 1014 state = nvmeio->ccb_state & NDA_CCB_TYPE_MASK; 1015 switch (state) { 1016 case NDA_CCB_BUFFER_IO: 1017 case NDA_CCB_TRIM: 1018 { 1019 int error; 1020 1021 cam_periph_lock(periph); 1022 if ((done_ccb->ccb_h.status & CAM_STATUS_MASK) != CAM_REQ_CMP) { 1023 error = ndaerror(done_ccb, 0, 0); 1024 if (error == ERESTART) { 1025 /* A retry was scheduled, so just return. */ 1026 cam_periph_unlock(periph); 1027 return; 1028 } 1029 if ((done_ccb->ccb_h.status & CAM_DEV_QFRZN) != 0) 1030 cam_release_devq(path, 1031 /*relsim_flags*/0, 1032 /*reduction*/0, 1033 /*timeout*/0, 1034 /*getcount_only*/0); 1035 } else { 1036 if ((done_ccb->ccb_h.status & CAM_DEV_QFRZN) != 0) 1037 panic("REQ_CMP with QFRZN"); 1038 error = 0; 1039 } 1040 if (state == NDA_CCB_BUFFER_IO) { 1041 struct bio *bp; 1042 1043 bp = (struct bio *)done_ccb->ccb_bp; 1044 bp->bio_error = error; 1045 if (error != 0) { 1046 bp->bio_resid = bp->bio_bcount; 1047 bp->bio_flags |= BIO_ERROR; 1048 } else { 1049 bp->bio_resid = 0; 1050 } 1051 softc->outstanding_cmds--; 1052 1053 /* 1054 * We need to call cam_iosched before we call biodone so that we 1055 * don't measure any activity that happens in the completion 1056 * routine, which in the case of sendfile can be quite 1057 * extensive. 1058 */ 1059 cam_iosched_bio_complete(softc->cam_iosched, bp, done_ccb); 1060 xpt_release_ccb(done_ccb); 1061 ndaschedule(periph); 1062 cam_periph_unlock(periph); 1063 biodone(bp); 1064 } else { /* state == NDA_CCB_TRIM */ 1065 struct nda_trim_request *trim; 1066 struct bio *bp1, *bp2; 1067 TAILQ_HEAD(, bio) queue; 1068 1069 trim = nvmeio->ccb_trim; 1070 TAILQ_INIT(&queue); 1071 TAILQ_CONCAT(&queue, &trim->bps, bio_queue); 1072 free(trim, M_NVMEDA); 1073 1074 /* 1075 * Since we can have multiple trims in flight, we don't 1076 * need to call this here. 1077 * cam_iosched_trim_done(softc->cam_iosched); 1078 */ 1079 /* 1080 * The the I/O scheduler that we're finishing the I/O 1081 * so we can keep book. The first one we pass in the CCB 1082 * which has the timing information. The rest we pass in NULL 1083 * so we can keep proper counts. 1084 */ 1085 bp1 = TAILQ_FIRST(&queue); 1086 cam_iosched_bio_complete(softc->cam_iosched, bp1, done_ccb); 1087 xpt_release_ccb(done_ccb); 1088 ndaschedule(periph); 1089 cam_periph_unlock(periph); 1090 while ((bp2 = TAILQ_FIRST(&queue)) != NULL) { 1091 TAILQ_REMOVE(&queue, bp2, bio_queue); 1092 bp2->bio_error = error; 1093 if (error != 0) { 1094 bp2->bio_flags |= BIO_ERROR; 1095 bp2->bio_resid = bp1->bio_bcount; 1096 } else 1097 bp2->bio_resid = 0; 1098 if (bp1 != bp2) 1099 cam_iosched_bio_complete(softc->cam_iosched, bp2, NULL); 1100 biodone(bp2); 1101 } 1102 } 1103 /* 1104 * Release the periph refcount taken in mdastart() for each CCB. 1105 */ 1106 KASSERT(softc->refcount >= 1, ("ndadone softc %p refcount %d", softc, softc->refcount)); 1107 softc->refcount--; 1108 return; 1109 } 1110 case NDA_CCB_DUMP: 1111 /* No-op. We're polling */ 1112 return; 1113 default: 1114 break; 1115 } 1116 xpt_release_ccb(done_ccb); 1117 } 1118 1119 static int 1120 ndaerror(union ccb *ccb, u_int32_t cam_flags, u_int32_t sense_flags) 1121 { 1122 struct nda_softc *softc; 1123 struct cam_periph *periph; 1124 1125 periph = xpt_path_periph(ccb->ccb_h.path); 1126 softc = (struct nda_softc *)periph->softc; 1127 1128 switch (ccb->ccb_h.status & CAM_STATUS_MASK) { 1129 case CAM_CMD_TIMEOUT: 1130 #ifdef CAM_IO_STATS 1131 softc->timeouts++; 1132 #endif 1133 break; 1134 case CAM_REQ_ABORTED: 1135 case CAM_REQ_CMP_ERR: 1136 case CAM_REQ_TERMIO: 1137 case CAM_UNREC_HBA_ERROR: 1138 case CAM_DATA_RUN_ERR: 1139 case CAM_ATA_STATUS_ERROR: 1140 #ifdef CAM_IO_STATS 1141 softc->errors++; 1142 #endif 1143 break; 1144 default: 1145 break; 1146 } 1147 1148 return(cam_periph_error(ccb, cam_flags, sense_flags)); 1149 } 1150 1151 /* 1152 * Step through all NDA peripheral drivers, and if the device is still open, 1153 * sync the disk cache to physical media. 1154 */ 1155 static void 1156 ndaflush(void) 1157 { 1158 struct cam_periph *periph; 1159 struct nda_softc *softc; 1160 union ccb *ccb; 1161 int error; 1162 1163 CAM_PERIPH_FOREACH(periph, &ndadriver) { 1164 softc = (struct nda_softc *)periph->softc; 1165 1166 if (SCHEDULER_STOPPED()) { 1167 /* 1168 * If we paniced with the lock held or the periph is not 1169 * open, do not recurse. Otherwise, call ndadump since 1170 * that avoids the sleeping cam_periph_getccb does if no 1171 * CCBs are available. 1172 */ 1173 if (!cam_periph_owned(periph) && 1174 (softc->flags & NDA_FLAG_OPEN)) { 1175 ndadump(softc->disk, NULL, 0, 0, 0); 1176 } 1177 continue; 1178 } 1179 1180 /* 1181 * We only sync the cache if the drive is still open 1182 */ 1183 cam_periph_lock(periph); 1184 if ((softc->flags & NDA_FLAG_OPEN) == 0) { 1185 cam_periph_unlock(periph); 1186 continue; 1187 } 1188 1189 ccb = cam_periph_getccb(periph, CAM_PRIORITY_NORMAL); 1190 nda_nvme_flush(softc, &ccb->nvmeio); 1191 error = cam_periph_runccb(ccb, ndaerror, /*cam_flags*/0, 1192 /*sense_flags*/ SF_NO_RECOVERY | SF_NO_RETRY, 1193 softc->disk->d_devstat); 1194 if (error != 0) 1195 xpt_print(periph->path, "Synchronize cache failed\n"); 1196 xpt_release_ccb(ccb); 1197 cam_periph_unlock(periph); 1198 } 1199 } 1200 1201 static void 1202 ndashutdown(void *arg, int howto) 1203 { 1204 1205 ndaflush(); 1206 } 1207 1208 static void 1209 ndasuspend(void *arg) 1210 { 1211 1212 ndaflush(); 1213 } 1214