1 /*- 2 * Copyright (c) 2015 Netflix, Inc 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer, 10 * without modification, immediately at the beginning of the file. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 18 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 19 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 20 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 24 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 25 * 26 * Derived from ata_da.c: 27 * Copyright (c) 2009 Alexander Motin <mav@FreeBSD.org> 28 */ 29 30 #include <sys/cdefs.h> 31 __FBSDID("$FreeBSD$"); 32 33 #include <sys/param.h> 34 35 #ifdef _KERNEL 36 #include <sys/systm.h> 37 #include <sys/kernel.h> 38 #include <sys/bio.h> 39 #include <sys/sysctl.h> 40 #include <sys/taskqueue.h> 41 #include <sys/lock.h> 42 #include <sys/mutex.h> 43 #include <sys/conf.h> 44 #include <sys/devicestat.h> 45 #include <sys/eventhandler.h> 46 #include <sys/malloc.h> 47 #include <sys/cons.h> 48 #include <sys/proc.h> 49 #include <sys/reboot.h> 50 #include <geom/geom_disk.h> 51 #endif /* _KERNEL */ 52 53 #ifndef _KERNEL 54 #include <stdio.h> 55 #include <string.h> 56 #endif /* _KERNEL */ 57 58 #include <cam/cam.h> 59 #include <cam/cam_ccb.h> 60 #include <cam/cam_periph.h> 61 #include <cam/cam_xpt_periph.h> 62 #include <cam/cam_sim.h> 63 #include <cam/cam_iosched.h> 64 65 #include <cam/nvme/nvme_all.h> 66 67 typedef enum { 68 NDA_STATE_NORMAL 69 } nda_state; 70 71 typedef enum { 72 NDA_FLAG_OPEN = 0x0001, 73 NDA_FLAG_DIRTY = 0x0002, 74 NDA_FLAG_SCTX_INIT = 0x0004, 75 } nda_flags; 76 77 typedef enum { 78 NDA_Q_4K = 0x01, 79 NDA_Q_NONE = 0x00, 80 } nda_quirks; 81 82 #define NDA_Q_BIT_STRING \ 83 "\020" \ 84 "\001Bit 0" 85 86 typedef enum { 87 NDA_CCB_BUFFER_IO = 0x01, 88 NDA_CCB_DUMP = 0x02, 89 NDA_CCB_TRIM = 0x03, 90 NDA_CCB_TYPE_MASK = 0x0F, 91 } nda_ccb_state; 92 93 /* Offsets into our private area for storing information */ 94 #define ccb_state ppriv_field0 95 #define ccb_bp ppriv_ptr1 96 97 struct trim_request { 98 TAILQ_HEAD(, bio) bps; 99 }; 100 struct nda_softc { 101 struct cam_iosched_softc *cam_iosched; 102 int outstanding_cmds; /* Number of active commands */ 103 int refcount; /* Active xpt_action() calls */ 104 nda_state state; 105 nda_flags flags; 106 nda_quirks quirks; 107 int unmappedio; 108 uint32_t nsid; /* Namespace ID for this nda device */ 109 struct disk *disk; 110 struct task sysctl_task; 111 struct sysctl_ctx_list sysctl_ctx; 112 struct sysctl_oid *sysctl_tree; 113 struct trim_request trim_req; 114 #ifdef CAM_IO_STATS 115 struct sysctl_ctx_list sysctl_stats_ctx; 116 struct sysctl_oid *sysctl_stats_tree; 117 u_int timeouts; 118 u_int errors; 119 u_int invalidations; 120 #endif 121 }; 122 123 /* Need quirk table */ 124 125 static disk_strategy_t ndastrategy; 126 static dumper_t ndadump; 127 static periph_init_t ndainit; 128 static void ndaasync(void *callback_arg, u_int32_t code, 129 struct cam_path *path, void *arg); 130 static void ndasysctlinit(void *context, int pending); 131 static periph_ctor_t ndaregister; 132 static periph_dtor_t ndacleanup; 133 static periph_start_t ndastart; 134 static periph_oninv_t ndaoninvalidate; 135 static void ndadone(struct cam_periph *periph, 136 union ccb *done_ccb); 137 static int ndaerror(union ccb *ccb, u_int32_t cam_flags, 138 u_int32_t sense_flags); 139 static void ndashutdown(void *arg, int howto); 140 static void ndasuspend(void *arg); 141 142 #ifndef NDA_DEFAULT_SEND_ORDERED 143 #define NDA_DEFAULT_SEND_ORDERED 1 144 #endif 145 #ifndef NDA_DEFAULT_TIMEOUT 146 #define NDA_DEFAULT_TIMEOUT 30 /* Timeout in seconds */ 147 #endif 148 #ifndef NDA_DEFAULT_RETRY 149 #define NDA_DEFAULT_RETRY 4 150 #endif 151 152 153 //static int nda_retry_count = NDA_DEFAULT_RETRY; 154 static int nda_send_ordered = NDA_DEFAULT_SEND_ORDERED; 155 static int nda_default_timeout = NDA_DEFAULT_TIMEOUT; 156 157 /* 158 * All NVMe media is non-rotational, so all nvme device instances 159 * share this to implement the sysctl. 160 */ 161 static int nda_rotating_media = 0; 162 163 static SYSCTL_NODE(_kern_cam, OID_AUTO, nda, CTLFLAG_RD, 0, 164 "CAM Direct Access Disk driver"); 165 166 static struct periph_driver ndadriver = 167 { 168 ndainit, "nda", 169 TAILQ_HEAD_INITIALIZER(ndadriver.units), /* generation */ 0 170 }; 171 172 PERIPHDRIVER_DECLARE(nda, ndadriver); 173 174 static MALLOC_DEFINE(M_NVMEDA, "nvme_da", "nvme_da buffers"); 175 176 /* 177 * nice wrappers. Maybe these belong in nvme_all.c instead of 178 * here, but this is the only place that uses these. Should 179 * we ever grow another NVME periph, we should move them 180 * all there wholesale. 181 */ 182 183 static void 184 nda_nvme_flush(struct nda_softc *softc, struct ccb_nvmeio *nvmeio) 185 { 186 cam_fill_nvmeio(nvmeio, 187 0, /* retries */ 188 ndadone, /* cbfcnp */ 189 CAM_DIR_NONE, /* flags */ 190 NULL, /* data_ptr */ 191 0, /* dxfer_len */ 192 nda_default_timeout * 1000); /* timeout 5s */ 193 nvme_ns_flush_cmd(&nvmeio->cmd, softc->nsid); 194 } 195 196 static void 197 nda_nvme_trim(struct nda_softc *softc, struct ccb_nvmeio *nvmeio, 198 void *payload, uint32_t num_ranges) 199 { 200 cam_fill_nvmeio(nvmeio, 201 0, /* retries */ 202 ndadone, /* cbfcnp */ 203 CAM_DIR_OUT, /* flags */ 204 payload, /* data_ptr */ 205 num_ranges * sizeof(struct nvme_dsm_range), /* dxfer_len */ 206 nda_default_timeout * 1000); /* timeout 5s */ 207 nvme_ns_trim_cmd(&nvmeio->cmd, softc->nsid, num_ranges); 208 } 209 210 static void 211 nda_nvme_write(struct nda_softc *softc, struct ccb_nvmeio *nvmeio, 212 void *payload, uint64_t lba, uint32_t len, uint32_t count) 213 { 214 cam_fill_nvmeio(nvmeio, 215 0, /* retries */ 216 ndadone, /* cbfcnp */ 217 CAM_DIR_OUT, /* flags */ 218 payload, /* data_ptr */ 219 len, /* dxfer_len */ 220 nda_default_timeout * 1000); /* timeout 5s */ 221 nvme_ns_write_cmd(&nvmeio->cmd, softc->nsid, lba, count); 222 } 223 224 static void 225 nda_nvme_rw_bio(struct nda_softc *softc, struct ccb_nvmeio *nvmeio, 226 struct bio *bp, uint32_t rwcmd) 227 { 228 int flags = rwcmd == NVME_OPC_READ ? CAM_DIR_IN : CAM_DIR_OUT; 229 void *payload; 230 uint64_t lba; 231 uint32_t count; 232 233 if (bp->bio_flags & BIO_UNMAPPED) { 234 flags |= CAM_DATA_BIO; 235 payload = bp; 236 } else { 237 payload = bp->bio_data; 238 } 239 240 lba = bp->bio_pblkno; 241 count = bp->bio_bcount / softc->disk->d_sectorsize; 242 243 cam_fill_nvmeio(nvmeio, 244 0, /* retries */ 245 ndadone, /* cbfcnp */ 246 flags, /* flags */ 247 payload, /* data_ptr */ 248 bp->bio_bcount, /* dxfer_len */ 249 nda_default_timeout * 1000); /* timeout 5s */ 250 nvme_ns_rw_cmd(&nvmeio->cmd, rwcmd, softc->nsid, lba, count); 251 } 252 253 static int 254 ndaopen(struct disk *dp) 255 { 256 struct cam_periph *periph; 257 struct nda_softc *softc; 258 int error; 259 260 periph = (struct cam_periph *)dp->d_drv1; 261 if (cam_periph_acquire(periph) != CAM_REQ_CMP) { 262 return(ENXIO); 263 } 264 265 cam_periph_lock(periph); 266 if ((error = cam_periph_hold(periph, PRIBIO|PCATCH)) != 0) { 267 cam_periph_unlock(periph); 268 cam_periph_release(periph); 269 return (error); 270 } 271 272 CAM_DEBUG(periph->path, CAM_DEBUG_TRACE | CAM_DEBUG_PERIPH, 273 ("ndaopen\n")); 274 275 softc = (struct nda_softc *)periph->softc; 276 softc->flags |= NDA_FLAG_OPEN; 277 278 cam_periph_unhold(periph); 279 cam_periph_unlock(periph); 280 return (0); 281 } 282 283 static int 284 ndaclose(struct disk *dp) 285 { 286 struct cam_periph *periph; 287 struct nda_softc *softc; 288 union ccb *ccb; 289 int error; 290 291 periph = (struct cam_periph *)dp->d_drv1; 292 softc = (struct nda_softc *)periph->softc; 293 cam_periph_lock(periph); 294 295 CAM_DEBUG(periph->path, CAM_DEBUG_TRACE | CAM_DEBUG_PERIPH, 296 ("ndaclose\n")); 297 298 if ((softc->flags & NDA_FLAG_DIRTY) != 0 && 299 (periph->flags & CAM_PERIPH_INVALID) == 0 && 300 cam_periph_hold(periph, PRIBIO) == 0) { 301 302 ccb = cam_periph_getccb(periph, CAM_PRIORITY_NORMAL); 303 nda_nvme_flush(softc, &ccb->nvmeio); 304 error = cam_periph_runccb(ccb, ndaerror, /*cam_flags*/0, 305 /*sense_flags*/0, softc->disk->d_devstat); 306 307 if (error != 0) 308 xpt_print(periph->path, "Synchronize cache failed\n"); 309 else 310 softc->flags &= ~NDA_FLAG_DIRTY; 311 xpt_release_ccb(ccb); 312 cam_periph_unhold(periph); 313 } 314 315 softc->flags &= ~NDA_FLAG_OPEN; 316 317 while (softc->refcount != 0) 318 cam_periph_sleep(periph, &softc->refcount, PRIBIO, "ndaclose", 1); 319 cam_periph_unlock(periph); 320 cam_periph_release(periph); 321 return (0); 322 } 323 324 static void 325 ndaschedule(struct cam_periph *periph) 326 { 327 struct nda_softc *softc = (struct nda_softc *)periph->softc; 328 329 if (softc->state != NDA_STATE_NORMAL) 330 return; 331 332 cam_iosched_schedule(softc->cam_iosched, periph); 333 } 334 335 /* 336 * Actually translate the requested transfer into one the physical driver 337 * can understand. The transfer is described by a buf and will include 338 * only one physical transfer. 339 */ 340 static void 341 ndastrategy(struct bio *bp) 342 { 343 struct cam_periph *periph; 344 struct nda_softc *softc; 345 346 periph = (struct cam_periph *)bp->bio_disk->d_drv1; 347 softc = (struct nda_softc *)periph->softc; 348 349 cam_periph_lock(periph); 350 351 CAM_DEBUG(periph->path, CAM_DEBUG_TRACE, ("ndastrategy(%p)\n", bp)); 352 353 /* 354 * If the device has been made invalid, error out 355 */ 356 if ((periph->flags & CAM_PERIPH_INVALID) != 0) { 357 cam_periph_unlock(periph); 358 biofinish(bp, NULL, ENXIO); 359 return; 360 } 361 362 /* 363 * Place it in the queue of disk activities for this disk 364 */ 365 cam_iosched_queue_work(softc->cam_iosched, bp); 366 367 /* 368 * Schedule ourselves for performing the work. 369 */ 370 ndaschedule(periph); 371 cam_periph_unlock(periph); 372 373 return; 374 } 375 376 static int 377 ndadump(void *arg, void *virtual, vm_offset_t physical, off_t offset, size_t length) 378 { 379 struct cam_periph *periph; 380 struct nda_softc *softc; 381 u_int secsize; 382 union ccb ccb; 383 struct disk *dp; 384 uint64_t lba; 385 uint32_t count; 386 int error = 0; 387 388 dp = arg; 389 periph = dp->d_drv1; 390 softc = (struct nda_softc *)periph->softc; 391 cam_periph_lock(periph); 392 secsize = softc->disk->d_sectorsize; 393 lba = offset / secsize; 394 count = length / secsize; 395 396 if ((periph->flags & CAM_PERIPH_INVALID) != 0) { 397 cam_periph_unlock(periph); 398 return (ENXIO); 399 } 400 401 if (length > 0) { 402 xpt_setup_ccb(&ccb.ccb_h, periph->path, CAM_PRIORITY_NORMAL); 403 ccb.ccb_h.ccb_state = NDA_CCB_DUMP; 404 nda_nvme_write(softc, &ccb.nvmeio, virtual, lba, length, count); 405 xpt_polled_action(&ccb); 406 407 error = cam_periph_error(&ccb, 408 0, SF_NO_RECOVERY | SF_NO_RETRY, NULL); 409 if ((ccb.ccb_h.status & CAM_DEV_QFRZN) != 0) 410 cam_release_devq(ccb.ccb_h.path, /*relsim_flags*/0, 411 /*reduction*/0, /*timeout*/0, /*getcount_only*/0); 412 if (error != 0) 413 printf("Aborting dump due to I/O error.\n"); 414 415 cam_periph_unlock(periph); 416 return (error); 417 } 418 419 /* Flush */ 420 xpt_setup_ccb(&ccb.ccb_h, periph->path, CAM_PRIORITY_NORMAL); 421 422 ccb.ccb_h.ccb_state = NDA_CCB_DUMP; 423 nda_nvme_flush(softc, &ccb.nvmeio); 424 xpt_polled_action(&ccb); 425 426 error = cam_periph_error(&ccb, 427 0, SF_NO_RECOVERY | SF_NO_RETRY, NULL); 428 if ((ccb.ccb_h.status & CAM_DEV_QFRZN) != 0) 429 cam_release_devq(ccb.ccb_h.path, /*relsim_flags*/0, 430 /*reduction*/0, /*timeout*/0, /*getcount_only*/0); 431 if (error != 0) 432 xpt_print(periph->path, "flush cmd failed\n"); 433 cam_periph_unlock(periph); 434 return (error); 435 } 436 437 static void 438 ndainit(void) 439 { 440 cam_status status; 441 442 /* 443 * Install a global async callback. This callback will 444 * receive async callbacks like "new device found". 445 */ 446 status = xpt_register_async(AC_FOUND_DEVICE, ndaasync, NULL, NULL); 447 448 if (status != CAM_REQ_CMP) { 449 printf("nda: Failed to attach master async callback " 450 "due to status 0x%x!\n", status); 451 } else if (nda_send_ordered) { 452 453 /* Register our event handlers */ 454 if ((EVENTHANDLER_REGISTER(power_suspend, ndasuspend, 455 NULL, EVENTHANDLER_PRI_LAST)) == NULL) 456 printf("ndainit: power event registration failed!\n"); 457 if ((EVENTHANDLER_REGISTER(shutdown_post_sync, ndashutdown, 458 NULL, SHUTDOWN_PRI_DEFAULT)) == NULL) 459 printf("ndainit: shutdown event registration failed!\n"); 460 } 461 } 462 463 /* 464 * Callback from GEOM, called when it has finished cleaning up its 465 * resources. 466 */ 467 static void 468 ndadiskgonecb(struct disk *dp) 469 { 470 struct cam_periph *periph; 471 472 periph = (struct cam_periph *)dp->d_drv1; 473 474 cam_periph_release(periph); 475 } 476 477 static void 478 ndaoninvalidate(struct cam_periph *periph) 479 { 480 struct nda_softc *softc; 481 482 softc = (struct nda_softc *)periph->softc; 483 484 /* 485 * De-register any async callbacks. 486 */ 487 xpt_register_async(0, ndaasync, periph, periph->path); 488 #ifdef CAM_IO_STATS 489 softc->invalidations++; 490 #endif 491 492 /* 493 * Return all queued I/O with ENXIO. 494 * XXX Handle any transactions queued to the card 495 * with XPT_ABORT_CCB. 496 */ 497 cam_iosched_flush(softc->cam_iosched, NULL, ENXIO); 498 499 disk_gone(softc->disk); 500 } 501 502 static void 503 ndacleanup(struct cam_periph *periph) 504 { 505 struct nda_softc *softc; 506 507 softc = (struct nda_softc *)periph->softc; 508 509 cam_periph_unlock(periph); 510 511 cam_iosched_fini(softc->cam_iosched); 512 513 /* 514 * If we can't free the sysctl tree, oh well... 515 */ 516 if ((softc->flags & NDA_FLAG_SCTX_INIT) != 0) { 517 #ifdef CAM_IO_STATS 518 if (sysctl_ctx_free(&softc->sysctl_stats_ctx) != 0) 519 xpt_print(periph->path, 520 "can't remove sysctl stats context\n"); 521 #endif 522 if (sysctl_ctx_free(&softc->sysctl_ctx) != 0) 523 xpt_print(periph->path, 524 "can't remove sysctl context\n"); 525 } 526 527 disk_destroy(softc->disk); 528 free(softc, M_DEVBUF); 529 cam_periph_lock(periph); 530 } 531 532 static void 533 ndaasync(void *callback_arg, u_int32_t code, 534 struct cam_path *path, void *arg) 535 { 536 struct cam_periph *periph; 537 538 periph = (struct cam_periph *)callback_arg; 539 switch (code) { 540 case AC_FOUND_DEVICE: 541 { 542 struct ccb_getdev *cgd; 543 cam_status status; 544 545 cgd = (struct ccb_getdev *)arg; 546 if (cgd == NULL) 547 break; 548 549 if (cgd->protocol != PROTO_NVME) 550 break; 551 552 /* 553 * Allocate a peripheral instance for 554 * this device and start the probe 555 * process. 556 */ 557 status = cam_periph_alloc(ndaregister, ndaoninvalidate, 558 ndacleanup, ndastart, 559 "nda", CAM_PERIPH_BIO, 560 path, ndaasync, 561 AC_FOUND_DEVICE, cgd); 562 563 if (status != CAM_REQ_CMP 564 && status != CAM_REQ_INPROG) 565 printf("ndaasync: Unable to attach to new device " 566 "due to status 0x%x\n", status); 567 break; 568 } 569 case AC_ADVINFO_CHANGED: 570 { 571 uintptr_t buftype; 572 573 buftype = (uintptr_t)arg; 574 if (buftype == CDAI_TYPE_PHYS_PATH) { 575 struct nda_softc *softc; 576 577 softc = periph->softc; 578 disk_attr_changed(softc->disk, "GEOM::physpath", 579 M_NOWAIT); 580 } 581 break; 582 } 583 case AC_LOST_DEVICE: 584 default: 585 cam_periph_async(periph, code, path, arg); 586 break; 587 } 588 } 589 590 static void 591 ndasysctlinit(void *context, int pending) 592 { 593 struct cam_periph *periph; 594 struct nda_softc *softc; 595 char tmpstr[80], tmpstr2[80]; 596 597 periph = (struct cam_periph *)context; 598 599 /* periph was held for us when this task was enqueued */ 600 if ((periph->flags & CAM_PERIPH_INVALID) != 0) { 601 cam_periph_release(periph); 602 return; 603 } 604 605 softc = (struct nda_softc *)periph->softc; 606 snprintf(tmpstr, sizeof(tmpstr), "CAM NDA unit %d", periph->unit_number); 607 snprintf(tmpstr2, sizeof(tmpstr2), "%d", periph->unit_number); 608 609 sysctl_ctx_init(&softc->sysctl_ctx); 610 softc->flags |= NDA_FLAG_SCTX_INIT; 611 softc->sysctl_tree = SYSCTL_ADD_NODE_WITH_LABEL(&softc->sysctl_ctx, 612 SYSCTL_STATIC_CHILDREN(_kern_cam_nda), OID_AUTO, tmpstr2, 613 CTLFLAG_RD, 0, tmpstr, "device_index"); 614 if (softc->sysctl_tree == NULL) { 615 printf("ndasysctlinit: unable to allocate sysctl tree\n"); 616 cam_periph_release(periph); 617 return; 618 } 619 620 SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree), 621 OID_AUTO, "unmapped_io", CTLFLAG_RD | CTLFLAG_MPSAFE, 622 &softc->unmappedio, 0, "Unmapped I/O leaf"); 623 624 SYSCTL_ADD_INT(&softc->sysctl_ctx, 625 SYSCTL_CHILDREN(softc->sysctl_tree), 626 OID_AUTO, 627 "rotating", 628 CTLFLAG_RD | CTLFLAG_MPSAFE, 629 &nda_rotating_media, 630 0, 631 "Rotating media"); 632 633 #ifdef CAM_IO_STATS 634 softc->sysctl_stats_tree = SYSCTL_ADD_NODE(&softc->sysctl_stats_ctx, 635 SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO, "stats", 636 CTLFLAG_RD, 0, "Statistics"); 637 if (softc->sysctl_stats_tree == NULL) { 638 printf("ndasysctlinit: unable to allocate sysctl tree for stats\n"); 639 cam_periph_release(periph); 640 return; 641 } 642 SYSCTL_ADD_INT(&softc->sysctl_stats_ctx, 643 SYSCTL_CHILDREN(softc->sysctl_stats_tree), 644 OID_AUTO, "timeouts", CTLFLAG_RD | CTLFLAG_MPSAFE, 645 &softc->timeouts, 0, 646 "Device timeouts reported by the SIM"); 647 SYSCTL_ADD_INT(&softc->sysctl_stats_ctx, 648 SYSCTL_CHILDREN(softc->sysctl_stats_tree), 649 OID_AUTO, "errors", CTLFLAG_RD | CTLFLAG_MPSAFE, 650 &softc->errors, 0, 651 "Transport errors reported by the SIM."); 652 SYSCTL_ADD_INT(&softc->sysctl_stats_ctx, 653 SYSCTL_CHILDREN(softc->sysctl_stats_tree), 654 OID_AUTO, "pack_invalidations", CTLFLAG_RD | CTLFLAG_MPSAFE, 655 &softc->invalidations, 0, 656 "Device pack invalidations."); 657 #endif 658 659 cam_iosched_sysctl_init(softc->cam_iosched, &softc->sysctl_ctx, 660 softc->sysctl_tree); 661 662 cam_periph_release(periph); 663 } 664 665 static int 666 ndagetattr(struct bio *bp) 667 { 668 int ret; 669 struct cam_periph *periph; 670 671 periph = (struct cam_periph *)bp->bio_disk->d_drv1; 672 cam_periph_lock(periph); 673 ret = xpt_getattr(bp->bio_data, bp->bio_length, bp->bio_attribute, 674 periph->path); 675 cam_periph_unlock(periph); 676 if (ret == 0) 677 bp->bio_completed = bp->bio_length; 678 return ret; 679 } 680 681 static cam_status 682 ndaregister(struct cam_periph *periph, void *arg) 683 { 684 struct nda_softc *softc; 685 struct disk *disk; 686 struct ccb_pathinq cpi; 687 struct ccb_getdev *cgd; 688 const struct nvme_namespace_data *nsd; 689 const struct nvme_controller_data *cd; 690 char announce_buf[80]; 691 // caddr_t match; 692 u_int maxio; 693 int quirks; 694 695 cgd = (struct ccb_getdev *)arg; 696 if (cgd == NULL) { 697 printf("ndaregister: no getdev CCB, can't register device\n"); 698 return(CAM_REQ_CMP_ERR); 699 } 700 nsd = cgd->nvme_data; 701 cd = cgd->nvme_cdata; 702 703 softc = (struct nda_softc *)malloc(sizeof(*softc), M_DEVBUF, 704 M_NOWAIT | M_ZERO); 705 706 if (softc == NULL) { 707 printf("ndaregister: Unable to probe new device. " 708 "Unable to allocate softc\n"); 709 return(CAM_REQ_CMP_ERR); 710 } 711 712 if (cam_iosched_init(&softc->cam_iosched, periph) != 0) { 713 printf("ndaregister: Unable to probe new device. " 714 "Unable to allocate iosched memory\n"); 715 return(CAM_REQ_CMP_ERR); 716 } 717 718 /* ident_data parsing */ 719 720 periph->softc = softc; 721 722 #if 0 723 /* 724 * See if this device has any quirks. 725 */ 726 match = cam_quirkmatch((caddr_t)&cgd->ident_data, 727 (caddr_t)nda_quirk_table, 728 sizeof(nda_quirk_table)/sizeof(*nda_quirk_table), 729 sizeof(*nda_quirk_table), ata_identify_match); 730 if (match != NULL) 731 softc->quirks = ((struct nda_quirk_entry *)match)->quirks; 732 else 733 #endif 734 softc->quirks = NDA_Q_NONE; 735 736 bzero(&cpi, sizeof(cpi)); 737 xpt_setup_ccb(&cpi.ccb_h, periph->path, CAM_PRIORITY_NONE); 738 cpi.ccb_h.func_code = XPT_PATH_INQ; 739 xpt_action((union ccb *)&cpi); 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 disk->d_sectorsize = 1 << nsd->lbaf[nsd->flbas.format].lbads; 781 disk->d_mediasize = (off_t)(disk->d_sectorsize * nsd->nsze); 782 disk->d_delmaxsize = disk->d_mediasize; 783 disk->d_flags = DISKFLAG_DIRECT_COMPLETION; 784 // if (cd->oncs.dsm) // XXX broken? 785 disk->d_flags |= DISKFLAG_CANDELETE; 786 if (cd->vwc.present) 787 disk->d_flags |= DISKFLAG_CANFLUSHCACHE; 788 if ((cpi.hba_misc & PIM_UNMAPPED) != 0) { 789 disk->d_flags |= DISKFLAG_UNMAPPED_BIO; 790 softc->unmappedio = 1; 791 } 792 /* 793 * d_ident and d_descr are both far bigger than the length of either 794 * the serial or model number strings. 795 */ 796 nvme_strvis(disk->d_descr, cd->mn, 797 sizeof(disk->d_descr), NVME_MODEL_NUMBER_LENGTH); 798 nvme_strvis(disk->d_ident, cd->sn, 799 sizeof(disk->d_ident), NVME_SERIAL_NUMBER_LENGTH); 800 disk->d_hba_vendor = cpi.hba_vendor; 801 disk->d_hba_device = cpi.hba_device; 802 disk->d_hba_subvendor = cpi.hba_subvendor; 803 disk->d_hba_subdevice = cpi.hba_subdevice; 804 disk->d_stripesize = disk->d_sectorsize; 805 disk->d_stripeoffset = 0; 806 disk->d_devstat = devstat_new_entry(periph->periph_name, 807 periph->unit_number, disk->d_sectorsize, 808 DEVSTAT_ALL_SUPPORTED, 809 DEVSTAT_TYPE_DIRECT | XPORT_DEVSTAT_TYPE(cpi.transport), 810 DEVSTAT_PRIORITY_DISK); 811 812 /* 813 * Acquire a reference to the periph before we register with GEOM. 814 * We'll release this reference once GEOM calls us back (via 815 * ndadiskgonecb()) telling us that our provider has been freed. 816 */ 817 if (cam_periph_acquire(periph) != CAM_REQ_CMP) { 818 xpt_print(periph->path, "%s: lost periph during " 819 "registration!\n", __func__); 820 cam_periph_lock(periph); 821 return (CAM_REQ_CMP_ERR); 822 } 823 disk_create(softc->disk, DISK_VERSION); 824 cam_periph_lock(periph); 825 cam_periph_unhold(periph); 826 827 snprintf(announce_buf, sizeof(announce_buf), 828 "%juMB (%ju %u byte sectors)", 829 (uintmax_t)((uintmax_t)disk->d_mediasize / (1024*1024)), 830 (uintmax_t)disk->d_mediasize / disk->d_sectorsize, 831 disk->d_sectorsize); 832 xpt_announce_periph(periph, announce_buf); 833 xpt_announce_quirks(periph, softc->quirks, NDA_Q_BIT_STRING); 834 835 /* 836 * Create our sysctl variables, now that we know 837 * we have successfully attached. 838 */ 839 if (cam_periph_acquire(periph) == CAM_REQ_CMP) 840 taskqueue_enqueue(taskqueue_thread, &softc->sysctl_task); 841 842 /* 843 * Register for device going away and info about the drive 844 * changing (though with NVMe, it can't) 845 */ 846 xpt_register_async(AC_LOST_DEVICE | AC_ADVINFO_CHANGED, 847 ndaasync, periph, periph->path); 848 849 softc->state = NDA_STATE_NORMAL; 850 return(CAM_REQ_CMP); 851 } 852 853 static void 854 ndastart(struct cam_periph *periph, union ccb *start_ccb) 855 { 856 struct nda_softc *softc = (struct nda_softc *)periph->softc; 857 struct ccb_nvmeio *nvmeio = &start_ccb->nvmeio; 858 859 CAM_DEBUG(periph->path, CAM_DEBUG_TRACE, ("ndastart\n")); 860 861 switch (softc->state) { 862 case NDA_STATE_NORMAL: 863 { 864 struct bio *bp; 865 866 bp = cam_iosched_next_bio(softc->cam_iosched); 867 CAM_DEBUG(periph->path, CAM_DEBUG_TRACE, ("ndastart: bio %p\n", bp)); 868 if (bp == NULL) { 869 xpt_release_ccb(start_ccb); 870 break; 871 } 872 873 switch (bp->bio_cmd) { 874 case BIO_WRITE: 875 softc->flags |= NDA_FLAG_DIRTY; 876 /* FALLTHROUGH */ 877 case BIO_READ: 878 { 879 #ifdef NDA_TEST_FAILURE 880 int fail = 0; 881 882 /* 883 * Support the failure ioctls. If the command is a 884 * read, and there are pending forced read errors, or 885 * if a write and pending write errors, then fail this 886 * operation with EIO. This is useful for testing 887 * purposes. Also, support having every Nth read fail. 888 * 889 * This is a rather blunt tool. 890 */ 891 if (bp->bio_cmd == BIO_READ) { 892 if (softc->force_read_error) { 893 softc->force_read_error--; 894 fail = 1; 895 } 896 if (softc->periodic_read_error > 0) { 897 if (++softc->periodic_read_count >= 898 softc->periodic_read_error) { 899 softc->periodic_read_count = 0; 900 fail = 1; 901 } 902 } 903 } else { 904 if (softc->force_write_error) { 905 softc->force_write_error--; 906 fail = 1; 907 } 908 } 909 if (fail) { 910 biofinish(bp, NULL, EIO); 911 xpt_release_ccb(start_ccb); 912 ndaschedule(periph); 913 return; 914 } 915 #endif 916 KASSERT((bp->bio_flags & BIO_UNMAPPED) == 0 || 917 round_page(bp->bio_bcount + bp->bio_ma_offset) / 918 PAGE_SIZE == bp->bio_ma_n, 919 ("Short bio %p", bp)); 920 nda_nvme_rw_bio(softc, &start_ccb->nvmeio, bp, bp->bio_cmd == BIO_READ ? 921 NVME_OPC_READ : NVME_OPC_WRITE); 922 break; 923 } 924 case BIO_DELETE: 925 { 926 struct nvme_dsm_range *dsm_range; 927 928 dsm_range = 929 malloc(sizeof(*dsm_range), M_NVMEDA, M_ZERO | M_WAITOK); 930 dsm_range->length = 931 bp->bio_bcount / softc->disk->d_sectorsize; 932 dsm_range->starting_lba = 933 bp->bio_offset / softc->disk->d_sectorsize; 934 bp->bio_driver2 = dsm_range; 935 nda_nvme_trim(softc, &start_ccb->nvmeio, dsm_range, 1); 936 start_ccb->ccb_h.ccb_state = NDA_CCB_TRIM; 937 start_ccb->ccb_h.flags |= CAM_UNLOCKED; 938 cam_iosched_submit_trim(softc->cam_iosched); /* XXX */ 939 goto out; 940 } 941 case BIO_FLUSH: 942 nda_nvme_flush(softc, nvmeio); 943 break; 944 } 945 start_ccb->ccb_h.ccb_state = NDA_CCB_BUFFER_IO; 946 start_ccb->ccb_h.flags |= CAM_UNLOCKED; 947 out: 948 start_ccb->ccb_h.ccb_bp = bp; 949 softc->outstanding_cmds++; 950 softc->refcount++; 951 cam_periph_unlock(periph); 952 xpt_action(start_ccb); 953 cam_periph_lock(periph); 954 softc->refcount--; 955 956 /* May have more work to do, so ensure we stay scheduled */ 957 ndaschedule(periph); 958 break; 959 } 960 } 961 } 962 963 static void 964 ndadone(struct cam_periph *periph, union ccb *done_ccb) 965 { 966 struct nda_softc *softc; 967 struct ccb_nvmeio *nvmeio = &done_ccb->nvmeio; 968 struct cam_path *path; 969 int state; 970 971 softc = (struct nda_softc *)periph->softc; 972 path = done_ccb->ccb_h.path; 973 974 CAM_DEBUG(path, CAM_DEBUG_TRACE, ("ndadone\n")); 975 976 state = nvmeio->ccb_h.ccb_state & NDA_CCB_TYPE_MASK; 977 switch (state) { 978 case NDA_CCB_BUFFER_IO: 979 case NDA_CCB_TRIM: 980 { 981 struct bio *bp; 982 int error; 983 984 cam_periph_lock(periph); 985 bp = (struct bio *)done_ccb->ccb_h.ccb_bp; 986 if ((done_ccb->ccb_h.status & CAM_STATUS_MASK) != CAM_REQ_CMP) { 987 error = ndaerror(done_ccb, 0, 0); 988 if (error == ERESTART) { 989 /* A retry was scheduled, so just return. */ 990 cam_periph_unlock(periph); 991 return; 992 } 993 if ((done_ccb->ccb_h.status & CAM_DEV_QFRZN) != 0) 994 cam_release_devq(path, 995 /*relsim_flags*/0, 996 /*reduction*/0, 997 /*timeout*/0, 998 /*getcount_only*/0); 999 } else { 1000 if ((done_ccb->ccb_h.status & CAM_DEV_QFRZN) != 0) 1001 panic("REQ_CMP with QFRZN"); 1002 error = 0; 1003 } 1004 bp->bio_error = error; 1005 if (error != 0) { 1006 bp->bio_resid = bp->bio_bcount; 1007 bp->bio_flags |= BIO_ERROR; 1008 } else { 1009 if (state == NDA_CCB_TRIM) 1010 bp->bio_resid = 0; 1011 else 1012 bp->bio_resid = nvmeio->resid; 1013 if (bp->bio_resid > 0) 1014 bp->bio_flags |= BIO_ERROR; 1015 } 1016 if (state == NDA_CCB_TRIM) 1017 free(bp->bio_driver2, M_NVMEDA); 1018 softc->outstanding_cmds--; 1019 1020 cam_iosched_bio_complete(softc->cam_iosched, bp, done_ccb); 1021 xpt_release_ccb(done_ccb); 1022 if (state == NDA_CCB_TRIM) { 1023 #ifdef notyet 1024 TAILQ_HEAD(, bio) queue; 1025 struct bio *bp1; 1026 1027 TAILQ_INIT(&queue); 1028 TAILQ_CONCAT(&queue, &softc->trim_req.bps, bio_queue); 1029 #endif 1030 cam_iosched_trim_done(softc->cam_iosched); 1031 ndaschedule(periph); 1032 cam_periph_unlock(periph); 1033 #ifdef notyet 1034 /* Not yet collapsing several BIO_DELETE requests into one TRIM */ 1035 while ((bp1 = TAILQ_FIRST(&queue)) != NULL) { 1036 TAILQ_REMOVE(&queue, bp1, bio_queue); 1037 bp1->bio_error = error; 1038 if (error != 0) { 1039 bp1->bio_flags |= BIO_ERROR; 1040 bp1->bio_resid = bp1->bio_bcount; 1041 } else 1042 bp1->bio_resid = 0; 1043 biodone(bp1); 1044 } 1045 #else 1046 biodone(bp); 1047 #endif 1048 } else { 1049 ndaschedule(periph); 1050 cam_periph_unlock(periph); 1051 biodone(bp); 1052 } 1053 return; 1054 } 1055 case NDA_CCB_DUMP: 1056 /* No-op. We're polling */ 1057 return; 1058 default: 1059 break; 1060 } 1061 xpt_release_ccb(done_ccb); 1062 } 1063 1064 static int 1065 ndaerror(union ccb *ccb, u_int32_t cam_flags, u_int32_t sense_flags) 1066 { 1067 struct nda_softc *softc; 1068 struct cam_periph *periph; 1069 1070 periph = xpt_path_periph(ccb->ccb_h.path); 1071 softc = (struct nda_softc *)periph->softc; 1072 1073 switch (ccb->ccb_h.status & CAM_STATUS_MASK) { 1074 case CAM_CMD_TIMEOUT: 1075 #ifdef CAM_IO_STATS 1076 softc->timeouts++; 1077 #endif 1078 break; 1079 case CAM_REQ_ABORTED: 1080 case CAM_REQ_CMP_ERR: 1081 case CAM_REQ_TERMIO: 1082 case CAM_UNREC_HBA_ERROR: 1083 case CAM_DATA_RUN_ERR: 1084 case CAM_ATA_STATUS_ERROR: 1085 #ifdef CAM_IO_STATS 1086 softc->errors++; 1087 #endif 1088 break; 1089 default: 1090 break; 1091 } 1092 1093 return(cam_periph_error(ccb, cam_flags, sense_flags, NULL)); 1094 } 1095 1096 /* 1097 * Step through all NDA peripheral drivers, and if the device is still open, 1098 * sync the disk cache to physical media. 1099 */ 1100 static void 1101 ndaflush(void) 1102 { 1103 struct cam_periph *periph; 1104 struct nda_softc *softc; 1105 union ccb *ccb; 1106 int error; 1107 1108 CAM_PERIPH_FOREACH(periph, &ndadriver) { 1109 softc = (struct nda_softc *)periph->softc; 1110 if (SCHEDULER_STOPPED()) { 1111 /* If we paniced with the lock held, do not recurse. */ 1112 if (!cam_periph_owned(periph) && 1113 (softc->flags & NDA_FLAG_OPEN)) { 1114 ndadump(softc->disk, NULL, 0, 0, 0); 1115 } 1116 continue; 1117 } 1118 cam_periph_lock(periph); 1119 /* 1120 * We only sync the cache if the drive is still open, and 1121 * if the drive is capable of it.. 1122 */ 1123 if ((softc->flags & NDA_FLAG_OPEN) == 0) { 1124 cam_periph_unlock(periph); 1125 continue; 1126 } 1127 1128 ccb = cam_periph_getccb(periph, CAM_PRIORITY_NORMAL); 1129 nda_nvme_flush(softc, &ccb->nvmeio); 1130 error = cam_periph_runccb(ccb, ndaerror, /*cam_flags*/0, 1131 /*sense_flags*/ SF_NO_RECOVERY | SF_NO_RETRY, 1132 softc->disk->d_devstat); 1133 if (error != 0) 1134 xpt_print(periph->path, "Synchronize cache failed\n"); 1135 xpt_release_ccb(ccb); 1136 cam_periph_unlock(periph); 1137 } 1138 } 1139 1140 static void 1141 ndashutdown(void *arg, int howto) 1142 { 1143 1144 ndaflush(); 1145 } 1146 1147 static void 1148 ndasuspend(void *arg) 1149 { 1150 1151 ndaflush(); 1152 } 1153