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 KASSERT(softc->outstanding_cmds == 0, 340 ("nda %d outstanding commands", softc->outstanding_cmds)); 341 cam_periph_unlock(periph); 342 cam_periph_release(periph); 343 return (0); 344 } 345 346 static void 347 ndaschedule(struct cam_periph *periph) 348 { 349 struct nda_softc *softc = (struct nda_softc *)periph->softc; 350 351 if (softc->state != NDA_STATE_NORMAL) 352 return; 353 354 cam_iosched_schedule(softc->cam_iosched, periph); 355 } 356 357 /* 358 * Actually translate the requested transfer into one the physical driver 359 * can understand. The transfer is described by a buf and will include 360 * only one physical transfer. 361 */ 362 static void 363 ndastrategy(struct bio *bp) 364 { 365 struct cam_periph *periph; 366 struct nda_softc *softc; 367 368 periph = (struct cam_periph *)bp->bio_disk->d_drv1; 369 softc = (struct nda_softc *)periph->softc; 370 371 cam_periph_lock(periph); 372 373 CAM_DEBUG(periph->path, CAM_DEBUG_TRACE, ("ndastrategy(%p)\n", bp)); 374 375 /* 376 * If the device has been made invalid, error out 377 */ 378 if ((periph->flags & CAM_PERIPH_INVALID) != 0) { 379 cam_periph_unlock(periph); 380 biofinish(bp, NULL, ENXIO); 381 return; 382 } 383 384 if (bp->bio_cmd == BIO_DELETE) 385 softc->deletes++; 386 387 /* 388 * Place it in the queue of disk activities for this disk 389 */ 390 cam_iosched_queue_work(softc->cam_iosched, bp); 391 392 /* 393 * Schedule ourselves for performing the work. 394 */ 395 ndaschedule(periph); 396 cam_periph_unlock(periph); 397 398 return; 399 } 400 401 static int 402 ndadump(void *arg, void *virtual, vm_offset_t physical, off_t offset, size_t length) 403 { 404 struct cam_periph *periph; 405 struct nda_softc *softc; 406 u_int secsize; 407 struct ccb_nvmeio nvmeio; 408 struct disk *dp; 409 uint64_t lba; 410 uint32_t count; 411 int error = 0; 412 413 dp = arg; 414 periph = dp->d_drv1; 415 softc = (struct nda_softc *)periph->softc; 416 secsize = softc->disk->d_sectorsize; 417 lba = offset / secsize; 418 count = length / secsize; 419 420 if ((periph->flags & CAM_PERIPH_INVALID) != 0) 421 return (ENXIO); 422 423 /* xpt_get_ccb returns a zero'd allocation for the ccb, mimic that here */ 424 memset(&nvmeio, 0, sizeof(nvmeio)); 425 if (length > 0) { 426 xpt_setup_ccb(&nvmeio.ccb_h, periph->path, CAM_PRIORITY_NORMAL); 427 nvmeio.ccb_state = NDA_CCB_DUMP; 428 nda_nvme_write(softc, &nvmeio, virtual, lba, length, count); 429 error = cam_periph_runccb((union ccb *)&nvmeio, cam_periph_error, 430 0, SF_NO_RECOVERY | SF_NO_RETRY, NULL); 431 if (error != 0) 432 printf("Aborting dump due to I/O error %d.\n", error); 433 434 return (error); 435 } 436 437 /* Flush */ 438 xpt_setup_ccb(&nvmeio.ccb_h, periph->path, CAM_PRIORITY_NORMAL); 439 440 nvmeio.ccb_state = NDA_CCB_DUMP; 441 nda_nvme_flush(softc, &nvmeio); 442 error = cam_periph_runccb((union ccb *)&nvmeio, cam_periph_error, 443 0, SF_NO_RECOVERY | SF_NO_RETRY, NULL); 444 if (error != 0) 445 xpt_print(periph->path, "flush cmd failed\n"); 446 return (error); 447 } 448 449 static void 450 ndainit(void) 451 { 452 cam_status status; 453 454 /* 455 * Install a global async callback. This callback will 456 * receive async callbacks like "new device found". 457 */ 458 status = xpt_register_async(AC_FOUND_DEVICE, ndaasync, NULL, NULL); 459 460 if (status != CAM_REQ_CMP) { 461 printf("nda: Failed to attach master async callback " 462 "due to status 0x%x!\n", status); 463 } else if (nda_send_ordered) { 464 465 /* Register our event handlers */ 466 if ((EVENTHANDLER_REGISTER(power_suspend, ndasuspend, 467 NULL, EVENTHANDLER_PRI_LAST)) == NULL) 468 printf("ndainit: power event registration failed!\n"); 469 if ((EVENTHANDLER_REGISTER(shutdown_post_sync, ndashutdown, 470 NULL, SHUTDOWN_PRI_DEFAULT)) == NULL) 471 printf("ndainit: shutdown event registration failed!\n"); 472 } 473 } 474 475 /* 476 * Callback from GEOM, called when it has finished cleaning up its 477 * resources. 478 */ 479 static void 480 ndadiskgonecb(struct disk *dp) 481 { 482 struct cam_periph *periph; 483 484 periph = (struct cam_periph *)dp->d_drv1; 485 486 cam_periph_release(periph); 487 } 488 489 static void 490 ndaoninvalidate(struct cam_periph *periph) 491 { 492 struct nda_softc *softc; 493 494 softc = (struct nda_softc *)periph->softc; 495 496 /* 497 * De-register any async callbacks. 498 */ 499 xpt_register_async(0, ndaasync, periph, periph->path); 500 #ifdef CAM_IO_STATS 501 softc->invalidations++; 502 #endif 503 504 /* 505 * Return all queued I/O with ENXIO. 506 * XXX Handle any transactions queued to the card 507 * with XPT_ABORT_CCB. 508 */ 509 cam_iosched_flush(softc->cam_iosched, NULL, ENXIO); 510 511 disk_gone(softc->disk); 512 } 513 514 static void 515 ndacleanup(struct cam_periph *periph) 516 { 517 struct nda_softc *softc; 518 519 softc = (struct nda_softc *)periph->softc; 520 521 cam_periph_unlock(periph); 522 523 cam_iosched_fini(softc->cam_iosched); 524 525 /* 526 * If we can't free the sysctl tree, oh well... 527 */ 528 if ((softc->flags & NDA_FLAG_SCTX_INIT) != 0) { 529 #ifdef CAM_IO_STATS 530 if (sysctl_ctx_free(&softc->sysctl_stats_ctx) != 0) 531 xpt_print(periph->path, 532 "can't remove sysctl stats context\n"); 533 #endif 534 if (sysctl_ctx_free(&softc->sysctl_ctx) != 0) 535 xpt_print(periph->path, 536 "can't remove sysctl context\n"); 537 } 538 539 disk_destroy(softc->disk); 540 free(softc, M_DEVBUF); 541 cam_periph_lock(periph); 542 } 543 544 static void 545 ndaasync(void *callback_arg, u_int32_t code, 546 struct cam_path *path, void *arg) 547 { 548 struct cam_periph *periph; 549 550 periph = (struct cam_periph *)callback_arg; 551 switch (code) { 552 case AC_FOUND_DEVICE: 553 { 554 struct ccb_getdev *cgd; 555 cam_status status; 556 557 cgd = (struct ccb_getdev *)arg; 558 if (cgd == NULL) 559 break; 560 561 if (cgd->protocol != PROTO_NVME) 562 break; 563 564 /* 565 * Allocate a peripheral instance for 566 * this device and start the probe 567 * process. 568 */ 569 status = cam_periph_alloc(ndaregister, ndaoninvalidate, 570 ndacleanup, ndastart, 571 "nda", CAM_PERIPH_BIO, 572 path, ndaasync, 573 AC_FOUND_DEVICE, cgd); 574 575 if (status != CAM_REQ_CMP 576 && status != CAM_REQ_INPROG) 577 printf("ndaasync: Unable to attach to new device " 578 "due to status 0x%x\n", status); 579 break; 580 } 581 case AC_ADVINFO_CHANGED: 582 { 583 uintptr_t buftype; 584 585 buftype = (uintptr_t)arg; 586 if (buftype == CDAI_TYPE_PHYS_PATH) { 587 struct nda_softc *softc; 588 589 softc = periph->softc; 590 disk_attr_changed(softc->disk, "GEOM::physpath", 591 M_NOWAIT); 592 } 593 break; 594 } 595 case AC_LOST_DEVICE: 596 default: 597 cam_periph_async(periph, code, path, arg); 598 break; 599 } 600 } 601 602 static void 603 ndasysctlinit(void *context, int pending) 604 { 605 struct cam_periph *periph; 606 struct nda_softc *softc; 607 char tmpstr[32], tmpstr2[16]; 608 609 periph = (struct cam_periph *)context; 610 611 /* periph was held for us when this task was enqueued */ 612 if ((periph->flags & CAM_PERIPH_INVALID) != 0) { 613 cam_periph_release(periph); 614 return; 615 } 616 617 softc = (struct nda_softc *)periph->softc; 618 snprintf(tmpstr, sizeof(tmpstr), "CAM NDA unit %d", periph->unit_number); 619 snprintf(tmpstr2, sizeof(tmpstr2), "%d", periph->unit_number); 620 621 sysctl_ctx_init(&softc->sysctl_ctx); 622 softc->flags |= NDA_FLAG_SCTX_INIT; 623 softc->sysctl_tree = SYSCTL_ADD_NODE_WITH_LABEL(&softc->sysctl_ctx, 624 SYSCTL_STATIC_CHILDREN(_kern_cam_nda), OID_AUTO, tmpstr2, 625 CTLFLAG_RD, 0, tmpstr, "device_index"); 626 if (softc->sysctl_tree == NULL) { 627 printf("ndasysctlinit: unable to allocate sysctl tree\n"); 628 cam_periph_release(periph); 629 return; 630 } 631 632 SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree), 633 OID_AUTO, "unmapped_io", CTLFLAG_RD, 634 &softc->unmappedio, 0, "Unmapped I/O leaf"); 635 636 SYSCTL_ADD_QUAD(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree), 637 OID_AUTO, "deletes", CTLFLAG_RD, 638 &softc->deletes, "Number of BIO_DELETE requests"); 639 640 SYSCTL_ADD_QUAD(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree), 641 OID_AUTO, "dsm_req", CTLFLAG_RD, 642 &softc->dsm_req, "Number of DSM requests sent to SIM"); 643 644 SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree), 645 OID_AUTO, "rotating", CTLFLAG_RD, &nda_rotating_media, 1, 646 "Rotating media"); 647 648 #ifdef CAM_IO_STATS 649 softc->sysctl_stats_tree = SYSCTL_ADD_NODE(&softc->sysctl_stats_ctx, 650 SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO, "stats", 651 CTLFLAG_RD, 0, "Statistics"); 652 if (softc->sysctl_stats_tree == NULL) { 653 printf("ndasysctlinit: unable to allocate sysctl tree for stats\n"); 654 cam_periph_release(periph); 655 return; 656 } 657 SYSCTL_ADD_INT(&softc->sysctl_stats_ctx, 658 SYSCTL_CHILDREN(softc->sysctl_stats_tree), 659 OID_AUTO, "timeouts", CTLFLAG_RD, 660 &softc->timeouts, 0, 661 "Device timeouts reported by the SIM"); 662 SYSCTL_ADD_INT(&softc->sysctl_stats_ctx, 663 SYSCTL_CHILDREN(softc->sysctl_stats_tree), 664 OID_AUTO, "errors", CTLFLAG_RD, 665 &softc->errors, 0, 666 "Transport errors reported by the SIM."); 667 SYSCTL_ADD_INT(&softc->sysctl_stats_ctx, 668 SYSCTL_CHILDREN(softc->sysctl_stats_tree), 669 OID_AUTO, "pack_invalidations", CTLFLAG_RD, 670 &softc->invalidations, 0, 671 "Device pack invalidations."); 672 #endif 673 674 #ifdef CAM_TEST_FAILURE 675 SYSCTL_ADD_PROC(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree), 676 OID_AUTO, "invalidate", CTLTYPE_U64 | CTLFLAG_RW | CTLFLAG_MPSAFE, 677 periph, 0, cam_periph_invalidate_sysctl, "I", 678 "Write 1 to invalidate the drive immediately"); 679 #endif 680 681 cam_iosched_sysctl_init(softc->cam_iosched, &softc->sysctl_ctx, 682 softc->sysctl_tree); 683 684 cam_periph_release(periph); 685 } 686 687 static int 688 ndagetattr(struct bio *bp) 689 { 690 int ret; 691 struct cam_periph *periph; 692 693 periph = (struct cam_periph *)bp->bio_disk->d_drv1; 694 cam_periph_lock(periph); 695 ret = xpt_getattr(bp->bio_data, bp->bio_length, bp->bio_attribute, 696 periph->path); 697 cam_periph_unlock(periph); 698 if (ret == 0) 699 bp->bio_completed = bp->bio_length; 700 return ret; 701 } 702 703 static cam_status 704 ndaregister(struct cam_periph *periph, void *arg) 705 { 706 struct nda_softc *softc; 707 struct disk *disk; 708 struct ccb_pathinq cpi; 709 const struct nvme_namespace_data *nsd; 710 const struct nvme_controller_data *cd; 711 char announce_buf[80]; 712 uint8_t flbas_fmt, lbads, vwc_present; 713 u_int maxio; 714 int quirks; 715 716 nsd = nvme_get_identify_ns(periph); 717 cd = nvme_get_identify_cntrl(periph); 718 719 softc = (struct nda_softc *)malloc(sizeof(*softc), M_DEVBUF, 720 M_NOWAIT | M_ZERO); 721 722 if (softc == NULL) { 723 printf("ndaregister: Unable to probe new device. " 724 "Unable to allocate softc\n"); 725 return(CAM_REQ_CMP_ERR); 726 } 727 728 if (cam_iosched_init(&softc->cam_iosched, periph) != 0) { 729 printf("ndaregister: Unable to probe new device. " 730 "Unable to allocate iosched memory\n"); 731 free(softc, M_DEVBUF); 732 return(CAM_REQ_CMP_ERR); 733 } 734 735 /* ident_data parsing */ 736 737 periph->softc = softc; 738 739 softc->quirks = NDA_Q_NONE; 740 741 xpt_path_inq(&cpi, periph->path); 742 743 TASK_INIT(&softc->sysctl_task, 0, ndasysctlinit, periph); 744 745 /* 746 * The name space ID is the lun, save it for later I/O 747 */ 748 softc->nsid = (uint32_t)xpt_path_lun_id(periph->path); 749 750 /* 751 * Register this media as a disk 752 */ 753 (void)cam_periph_hold(periph, PRIBIO); 754 cam_periph_unlock(periph); 755 snprintf(announce_buf, sizeof(announce_buf), 756 "kern.cam.nda.%d.quirks", periph->unit_number); 757 quirks = softc->quirks; 758 TUNABLE_INT_FETCH(announce_buf, &quirks); 759 softc->quirks = quirks; 760 cam_iosched_set_sort_queue(softc->cam_iosched, 0); 761 softc->disk = disk = disk_alloc(); 762 strlcpy(softc->disk->d_descr, cd->mn, 763 MIN(sizeof(softc->disk->d_descr), sizeof(cd->mn))); 764 strlcpy(softc->disk->d_ident, cd->sn, 765 MIN(sizeof(softc->disk->d_ident), sizeof(cd->sn))); 766 disk->d_rotation_rate = DISK_RR_NON_ROTATING; 767 disk->d_open = ndaopen; 768 disk->d_close = ndaclose; 769 disk->d_strategy = ndastrategy; 770 disk->d_getattr = ndagetattr; 771 disk->d_dump = ndadump; 772 disk->d_gone = ndadiskgonecb; 773 disk->d_name = "nda"; 774 disk->d_drv1 = periph; 775 disk->d_unit = periph->unit_number; 776 maxio = cpi.maxio; /* Honor max I/O size of SIM */ 777 if (maxio == 0) 778 maxio = DFLTPHYS; /* traditional default */ 779 else if (maxio > MAXPHYS) 780 maxio = MAXPHYS; /* for safety */ 781 disk->d_maxsize = maxio; 782 flbas_fmt = (nsd->flbas >> NVME_NS_DATA_FLBAS_FORMAT_SHIFT) & 783 NVME_NS_DATA_FLBAS_FORMAT_MASK; 784 lbads = (nsd->lbaf[flbas_fmt] >> NVME_NS_DATA_LBAF_LBADS_SHIFT) & 785 NVME_NS_DATA_LBAF_LBADS_MASK; 786 disk->d_sectorsize = 1 << lbads; 787 disk->d_mediasize = (off_t)(disk->d_sectorsize * nsd->nsze); 788 disk->d_delmaxsize = disk->d_mediasize; 789 disk->d_flags = DISKFLAG_DIRECT_COMPLETION; 790 // if (cd->oncs.dsm) // XXX broken? 791 disk->d_flags |= DISKFLAG_CANDELETE; 792 vwc_present = (cd->vwc >> NVME_CTRLR_DATA_VWC_PRESENT_SHIFT) & 793 NVME_CTRLR_DATA_VWC_PRESENT_MASK; 794 if (vwc_present) 795 disk->d_flags |= DISKFLAG_CANFLUSHCACHE; 796 if ((cpi.hba_misc & PIM_UNMAPPED) != 0) { 797 disk->d_flags |= DISKFLAG_UNMAPPED_BIO; 798 softc->unmappedio = 1; 799 } 800 /* 801 * d_ident and d_descr are both far bigger than the length of either 802 * the serial or model number strings. 803 */ 804 nvme_strvis(disk->d_descr, cd->mn, 805 sizeof(disk->d_descr), NVME_MODEL_NUMBER_LENGTH); 806 nvme_strvis(disk->d_ident, cd->sn, 807 sizeof(disk->d_ident), NVME_SERIAL_NUMBER_LENGTH); 808 disk->d_hba_vendor = cpi.hba_vendor; 809 disk->d_hba_device = cpi.hba_device; 810 disk->d_hba_subvendor = cpi.hba_subvendor; 811 disk->d_hba_subdevice = cpi.hba_subdevice; 812 disk->d_stripesize = disk->d_sectorsize; 813 disk->d_stripeoffset = 0; 814 disk->d_devstat = devstat_new_entry(periph->periph_name, 815 periph->unit_number, disk->d_sectorsize, 816 DEVSTAT_ALL_SUPPORTED, 817 DEVSTAT_TYPE_DIRECT | XPORT_DEVSTAT_TYPE(cpi.transport), 818 DEVSTAT_PRIORITY_DISK); 819 /* 820 * Add alias for older nvd drives to ease transition. 821 */ 822 /* disk_add_alias(disk, "nvd"); Have reports of this causing problems */ 823 824 /* 825 * Acquire a reference to the periph before we register with GEOM. 826 * We'll release this reference once GEOM calls us back (via 827 * ndadiskgonecb()) telling us that our provider has been freed. 828 */ 829 if (cam_periph_acquire(periph) != 0) { 830 xpt_print(periph->path, "%s: lost periph during " 831 "registration!\n", __func__); 832 cam_periph_lock(periph); 833 return (CAM_REQ_CMP_ERR); 834 } 835 disk_create(softc->disk, DISK_VERSION); 836 cam_periph_lock(periph); 837 cam_periph_unhold(periph); 838 839 snprintf(announce_buf, sizeof(announce_buf), 840 "%juMB (%ju %u byte sectors)", 841 (uintmax_t)((uintmax_t)disk->d_mediasize / (1024*1024)), 842 (uintmax_t)disk->d_mediasize / disk->d_sectorsize, 843 disk->d_sectorsize); 844 xpt_announce_periph(periph, announce_buf); 845 xpt_announce_quirks(periph, softc->quirks, NDA_Q_BIT_STRING); 846 847 /* 848 * Create our sysctl variables, now that we know 849 * we have successfully attached. 850 */ 851 if (cam_periph_acquire(periph) == 0) 852 taskqueue_enqueue(taskqueue_thread, &softc->sysctl_task); 853 854 /* 855 * Register for device going away and info about the drive 856 * changing (though with NVMe, it can't) 857 */ 858 xpt_register_async(AC_LOST_DEVICE | AC_ADVINFO_CHANGED, 859 ndaasync, periph, periph->path); 860 861 softc->state = NDA_STATE_NORMAL; 862 return(CAM_REQ_CMP); 863 } 864 865 static void 866 ndastart(struct cam_periph *periph, union ccb *start_ccb) 867 { 868 struct nda_softc *softc = (struct nda_softc *)periph->softc; 869 struct ccb_nvmeio *nvmeio = &start_ccb->nvmeio; 870 871 CAM_DEBUG(periph->path, CAM_DEBUG_TRACE, ("ndastart\n")); 872 873 switch (softc->state) { 874 case NDA_STATE_NORMAL: 875 { 876 struct bio *bp; 877 878 bp = cam_iosched_next_bio(softc->cam_iosched); 879 CAM_DEBUG(periph->path, CAM_DEBUG_TRACE, ("ndastart: bio %p\n", bp)); 880 if (bp == NULL) { 881 xpt_release_ccb(start_ccb); 882 break; 883 } 884 885 switch (bp->bio_cmd) { 886 case BIO_WRITE: 887 softc->flags |= NDA_FLAG_DIRTY; 888 /* FALLTHROUGH */ 889 case BIO_READ: 890 { 891 #ifdef CAM_TEST_FAILURE 892 int fail = 0; 893 894 /* 895 * Support the failure ioctls. If the command is a 896 * read, and there are pending forced read errors, or 897 * if a write and pending write errors, then fail this 898 * operation with EIO. This is useful for testing 899 * purposes. Also, support having every Nth read fail. 900 * 901 * This is a rather blunt tool. 902 */ 903 if (bp->bio_cmd == BIO_READ) { 904 if (softc->force_read_error) { 905 softc->force_read_error--; 906 fail = 1; 907 } 908 if (softc->periodic_read_error > 0) { 909 if (++softc->periodic_read_count >= 910 softc->periodic_read_error) { 911 softc->periodic_read_count = 0; 912 fail = 1; 913 } 914 } 915 } else { 916 if (softc->force_write_error) { 917 softc->force_write_error--; 918 fail = 1; 919 } 920 } 921 if (fail) { 922 biofinish(bp, NULL, EIO); 923 xpt_release_ccb(start_ccb); 924 ndaschedule(periph); 925 return; 926 } 927 #endif 928 KASSERT((bp->bio_flags & BIO_UNMAPPED) == 0 || 929 round_page(bp->bio_bcount + bp->bio_ma_offset) / 930 PAGE_SIZE == bp->bio_ma_n, 931 ("Short bio %p", bp)); 932 nda_nvme_rw_bio(softc, &start_ccb->nvmeio, bp, bp->bio_cmd == BIO_READ ? 933 NVME_OPC_READ : NVME_OPC_WRITE); 934 break; 935 } 936 case BIO_DELETE: 937 { 938 struct nvme_dsm_range *dsm_range, *dsm_end; 939 struct nda_trim_request *trim; 940 struct bio *bp1; 941 int ents; 942 943 trim = malloc(sizeof(*trim), M_NVMEDA, M_ZERO | M_NOWAIT); 944 if (trim == NULL) { 945 biofinish(bp, NULL, ENOMEM); 946 xpt_release_ccb(start_ccb); 947 ndaschedule(periph); 948 return; 949 } 950 TAILQ_INIT(&trim->bps); 951 bp1 = bp; 952 ents = sizeof(trim->data) / sizeof(struct nvme_dsm_range); 953 ents = min(ents, nda_max_trim_entries); 954 dsm_range = &trim->dsm; 955 dsm_end = dsm_range + ents; 956 do { 957 TAILQ_INSERT_TAIL(&trim->bps, bp1, bio_queue); 958 dsm_range->length = 959 htole32(bp1->bio_bcount / softc->disk->d_sectorsize); 960 dsm_range->starting_lba = 961 htole64(bp1->bio_offset / softc->disk->d_sectorsize); 962 dsm_range++; 963 if (dsm_range >= dsm_end) 964 break; 965 bp1 = cam_iosched_next_trim(softc->cam_iosched); 966 /* XXX -- Could collapse adjacent ranges, but we don't for now */ 967 /* XXX -- Could limit based on total payload size */ 968 } while (bp1 != NULL); 969 start_ccb->ccb_trim = trim; 970 softc->dsm_req++; 971 nda_nvme_trim(softc, &start_ccb->nvmeio, &trim->dsm, 972 dsm_range - &trim->dsm); 973 start_ccb->ccb_state = NDA_CCB_TRIM; 974 /* 975 * Note: We can have multiple TRIMs in flight, so we don't call 976 * cam_iosched_submit_trim(softc->cam_iosched); 977 * since that forces the I/O scheduler to only schedule one at a time. 978 * On NVMe drives, this is a performance disaster. 979 */ 980 goto out; 981 } 982 case BIO_FLUSH: 983 nda_nvme_flush(softc, nvmeio); 984 break; 985 } 986 start_ccb->ccb_state = NDA_CCB_BUFFER_IO; 987 start_ccb->ccb_bp = bp; 988 out: 989 start_ccb->ccb_h.flags |= CAM_UNLOCKED; 990 softc->outstanding_cmds++; 991 softc->refcount++; /* For submission only */ 992 cam_periph_unlock(periph); 993 xpt_action(start_ccb); 994 cam_periph_lock(periph); 995 softc->refcount--; /* Submission done */ 996 997 /* May have more work to do, so ensure we stay scheduled */ 998 ndaschedule(periph); 999 break; 1000 } 1001 } 1002 } 1003 1004 static void 1005 ndadone(struct cam_periph *periph, union ccb *done_ccb) 1006 { 1007 struct nda_softc *softc; 1008 struct ccb_nvmeio *nvmeio = &done_ccb->nvmeio; 1009 struct cam_path *path; 1010 int state; 1011 1012 softc = (struct nda_softc *)periph->softc; 1013 path = done_ccb->ccb_h.path; 1014 1015 CAM_DEBUG(path, CAM_DEBUG_TRACE, ("ndadone\n")); 1016 1017 state = nvmeio->ccb_state & NDA_CCB_TYPE_MASK; 1018 switch (state) { 1019 case NDA_CCB_BUFFER_IO: 1020 case NDA_CCB_TRIM: 1021 { 1022 int error; 1023 1024 cam_periph_lock(periph); 1025 if ((done_ccb->ccb_h.status & CAM_STATUS_MASK) != CAM_REQ_CMP) { 1026 error = ndaerror(done_ccb, 0, 0); 1027 if (error == ERESTART) { 1028 /* A retry was scheduled, so just return. */ 1029 cam_periph_unlock(periph); 1030 return; 1031 } 1032 if ((done_ccb->ccb_h.status & CAM_DEV_QFRZN) != 0) 1033 cam_release_devq(path, 1034 /*relsim_flags*/0, 1035 /*reduction*/0, 1036 /*timeout*/0, 1037 /*getcount_only*/0); 1038 } else { 1039 if ((done_ccb->ccb_h.status & CAM_DEV_QFRZN) != 0) 1040 panic("REQ_CMP with QFRZN"); 1041 error = 0; 1042 } 1043 if (state == NDA_CCB_BUFFER_IO) { 1044 struct bio *bp; 1045 1046 bp = (struct bio *)done_ccb->ccb_bp; 1047 bp->bio_error = error; 1048 if (error != 0) { 1049 bp->bio_resid = bp->bio_bcount; 1050 bp->bio_flags |= BIO_ERROR; 1051 } else { 1052 bp->bio_resid = 0; 1053 } 1054 softc->outstanding_cmds--; 1055 1056 /* 1057 * We need to call cam_iosched before we call biodone so that we 1058 * don't measure any activity that happens in the completion 1059 * routine, which in the case of sendfile can be quite 1060 * extensive. 1061 */ 1062 cam_iosched_bio_complete(softc->cam_iosched, bp, done_ccb); 1063 xpt_release_ccb(done_ccb); 1064 ndaschedule(periph); 1065 cam_periph_unlock(periph); 1066 biodone(bp); 1067 } else { /* state == NDA_CCB_TRIM */ 1068 struct nda_trim_request *trim; 1069 struct bio *bp1, *bp2; 1070 TAILQ_HEAD(, bio) queue; 1071 1072 trim = nvmeio->ccb_trim; 1073 TAILQ_INIT(&queue); 1074 TAILQ_CONCAT(&queue, &trim->bps, bio_queue); 1075 free(trim, M_NVMEDA); 1076 1077 /* 1078 * Since we can have multiple trims in flight, we don't 1079 * need to call this here. 1080 * cam_iosched_trim_done(softc->cam_iosched); 1081 */ 1082 /* 1083 * The the I/O scheduler that we're finishing the I/O 1084 * so we can keep book. The first one we pass in the CCB 1085 * which has the timing information. The rest we pass in NULL 1086 * so we can keep proper counts. 1087 */ 1088 bp1 = TAILQ_FIRST(&queue); 1089 cam_iosched_bio_complete(softc->cam_iosched, bp1, done_ccb); 1090 xpt_release_ccb(done_ccb); 1091 softc->outstanding_cmds--; 1092 ndaschedule(periph); 1093 cam_periph_unlock(periph); 1094 while ((bp2 = TAILQ_FIRST(&queue)) != NULL) { 1095 TAILQ_REMOVE(&queue, bp2, bio_queue); 1096 bp2->bio_error = error; 1097 if (error != 0) { 1098 bp2->bio_flags |= BIO_ERROR; 1099 bp2->bio_resid = bp1->bio_bcount; 1100 } else 1101 bp2->bio_resid = 0; 1102 if (bp1 != bp2) 1103 cam_iosched_bio_complete(softc->cam_iosched, bp2, NULL); 1104 biodone(bp2); 1105 } 1106 } 1107 return; 1108 } 1109 case NDA_CCB_DUMP: 1110 /* No-op. We're polling */ 1111 return; 1112 default: 1113 break; 1114 } 1115 xpt_release_ccb(done_ccb); 1116 } 1117 1118 static int 1119 ndaerror(union ccb *ccb, u_int32_t cam_flags, u_int32_t sense_flags) 1120 { 1121 struct nda_softc *softc; 1122 struct cam_periph *periph; 1123 1124 periph = xpt_path_periph(ccb->ccb_h.path); 1125 softc = (struct nda_softc *)periph->softc; 1126 1127 switch (ccb->ccb_h.status & CAM_STATUS_MASK) { 1128 case CAM_CMD_TIMEOUT: 1129 #ifdef CAM_IO_STATS 1130 softc->timeouts++; 1131 #endif 1132 break; 1133 case CAM_REQ_ABORTED: 1134 case CAM_REQ_CMP_ERR: 1135 case CAM_REQ_TERMIO: 1136 case CAM_UNREC_HBA_ERROR: 1137 case CAM_DATA_RUN_ERR: 1138 case CAM_ATA_STATUS_ERROR: 1139 #ifdef CAM_IO_STATS 1140 softc->errors++; 1141 #endif 1142 break; 1143 default: 1144 break; 1145 } 1146 1147 return(cam_periph_error(ccb, cam_flags, sense_flags)); 1148 } 1149 1150 /* 1151 * Step through all NDA peripheral drivers, and if the device is still open, 1152 * sync the disk cache to physical media. 1153 */ 1154 static void 1155 ndaflush(void) 1156 { 1157 struct cam_periph *periph; 1158 struct nda_softc *softc; 1159 union ccb *ccb; 1160 int error; 1161 1162 CAM_PERIPH_FOREACH(periph, &ndadriver) { 1163 softc = (struct nda_softc *)periph->softc; 1164 1165 if (SCHEDULER_STOPPED()) { 1166 /* 1167 * If we paniced with the lock held or the periph is not 1168 * open, do not recurse. Otherwise, call ndadump since 1169 * that avoids the sleeping cam_periph_getccb does if no 1170 * CCBs are available. 1171 */ 1172 if (!cam_periph_owned(periph) && 1173 (softc->flags & NDA_FLAG_OPEN)) { 1174 ndadump(softc->disk, NULL, 0, 0, 0); 1175 } 1176 continue; 1177 } 1178 1179 /* 1180 * We only sync the cache if the drive is still open 1181 */ 1182 cam_periph_lock(periph); 1183 if ((softc->flags & NDA_FLAG_OPEN) == 0) { 1184 cam_periph_unlock(periph); 1185 continue; 1186 } 1187 1188 ccb = cam_periph_getccb(periph, CAM_PRIORITY_NORMAL); 1189 nda_nvme_flush(softc, &ccb->nvmeio); 1190 error = cam_periph_runccb(ccb, ndaerror, /*cam_flags*/0, 1191 /*sense_flags*/ SF_NO_RECOVERY | SF_NO_RETRY, 1192 softc->disk->d_devstat); 1193 if (error != 0) 1194 xpt_print(periph->path, "Synchronize cache failed\n"); 1195 xpt_release_ccb(ccb); 1196 cam_periph_unlock(periph); 1197 } 1198 } 1199 1200 static void 1201 ndashutdown(void *arg, int howto) 1202 { 1203 1204 ndaflush(); 1205 } 1206 1207 static void 1208 ndasuspend(void *arg) 1209 { 1210 1211 ndaflush(); 1212 } 1213