1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 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 * 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 AND CONTRIBUTORS ``AS IS'' AND 16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 25 * 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 #include <sys/param.h> 33 34 #ifdef _KERNEL 35 #include <sys/systm.h> 36 #include <sys/kernel.h> 37 #include <sys/bio.h> 38 #include <sys/sysctl.h> 39 #include <sys/taskqueue.h> 40 #include <sys/lock.h> 41 #include <sys/mutex.h> 42 #include <sys/conf.h> 43 #include <sys/devicestat.h> 44 #include <sys/eventhandler.h> 45 #include <sys/malloc.h> 46 #include <sys/cons.h> 47 #include <sys/proc.h> 48 #include <sys/reboot.h> 49 #include <sys/sbuf.h> 50 #include <geom/geom.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 #define NDA_FLAG_STRING \ 78 "\020" \ 79 "\001OPEN" \ 80 "\002DIRTY" \ 81 "\003SCTX_INIT" 82 83 typedef enum { 84 NDA_Q_4K = 0x01, 85 NDA_Q_NONE = 0x00, 86 } nda_quirks; 87 88 #define NDA_Q_BIT_STRING \ 89 "\020" \ 90 "\001Bit 0" 91 92 typedef enum { 93 NDA_CCB_BUFFER_IO = 0x01, 94 NDA_CCB_DUMP = 0x02, 95 NDA_CCB_TRIM = 0x03, 96 NDA_CCB_PASS = 0x04, 97 NDA_CCB_TYPE_MASK = 0x0F, 98 } nda_ccb_state; 99 100 /* Offsets into our private area for storing information */ 101 #define ccb_state ccb_h.ppriv_field0 102 #define ccb_bp ccb_h.ppriv_ptr1 /* For NDA_CCB_BUFFER_IO */ 103 #define ccb_trim ccb_h.ppriv_ptr1 /* For NDA_CCB_TRIM */ 104 105 struct nda_softc { 106 struct cam_iosched_softc *cam_iosched; 107 int outstanding_cmds; /* Number of active commands */ 108 int refcount; /* Active xpt_action() calls */ 109 nda_state state; 110 nda_flags flags; 111 nda_quirks quirks; 112 int unmappedio; 113 quad_t deletes; 114 uint32_t nsid; /* Namespace ID for this nda device */ 115 struct disk *disk; 116 struct task sysctl_task; 117 struct sysctl_ctx_list sysctl_ctx; 118 struct sysctl_oid *sysctl_tree; 119 uint64_t trim_count; 120 uint64_t trim_ranges; 121 uint64_t trim_lbas; 122 #ifdef CAM_TEST_FAILURE 123 int force_read_error; 124 int force_write_error; 125 int periodic_read_error; 126 int periodic_read_count; 127 #endif 128 #ifdef CAM_IO_STATS 129 struct sysctl_ctx_list sysctl_stats_ctx; 130 struct sysctl_oid *sysctl_stats_tree; 131 u_int timeouts; 132 u_int errors; 133 u_int invalidations; 134 #endif 135 }; 136 137 struct nda_trim_request { 138 struct nvme_dsm_range dsm[NVME_MAX_DSM_TRIM / sizeof(struct nvme_dsm_range)]; 139 TAILQ_HEAD(, bio) bps; 140 }; 141 _Static_assert(NVME_MAX_DSM_TRIM % sizeof(struct nvme_dsm_range) == 0, 142 "NVME_MAX_DSM_TRIM must be an integral number of ranges"); 143 144 /* Need quirk table */ 145 146 static disk_ioctl_t ndaioctl; 147 static disk_strategy_t ndastrategy; 148 static dumper_t ndadump; 149 static periph_init_t ndainit; 150 static void ndaasync(void *callback_arg, uint32_t code, 151 struct cam_path *path, void *arg); 152 static void ndasysctlinit(void *context, int pending); 153 static int ndaflagssysctl(SYSCTL_HANDLER_ARGS); 154 static periph_ctor_t ndaregister; 155 static periph_dtor_t ndacleanup; 156 static periph_start_t ndastart; 157 static periph_oninv_t ndaoninvalidate; 158 static void ndadone(struct cam_periph *periph, 159 union ccb *done_ccb); 160 static int ndaerror(union ccb *ccb, uint32_t cam_flags, 161 uint32_t sense_flags); 162 static void ndashutdown(void *arg, int howto); 163 static void ndasuspend(void *arg); 164 165 #ifndef NDA_DEFAULT_SEND_ORDERED 166 #define NDA_DEFAULT_SEND_ORDERED 1 167 #endif 168 #ifndef NDA_DEFAULT_TIMEOUT 169 #define NDA_DEFAULT_TIMEOUT 30 /* Timeout in seconds */ 170 #endif 171 #ifndef NDA_DEFAULT_RETRY 172 #define NDA_DEFAULT_RETRY 4 173 #endif 174 #ifndef NDA_MAX_TRIM_ENTRIES 175 #define NDA_MAX_TRIM_ENTRIES (NVME_MAX_DSM_TRIM / sizeof(struct nvme_dsm_range))/* Number of DSM trims to use, max 256 */ 176 #endif 177 178 static SYSCTL_NODE(_kern_cam, OID_AUTO, nda, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 179 "CAM Direct Access Disk driver"); 180 181 //static int nda_retry_count = NDA_DEFAULT_RETRY; 182 static int nda_send_ordered = NDA_DEFAULT_SEND_ORDERED; 183 static int nda_default_timeout = NDA_DEFAULT_TIMEOUT; 184 static int nda_max_trim_entries = NDA_MAX_TRIM_ENTRIES; 185 static int nda_enable_biospeedup = 1; 186 static int nda_nvd_compat = 1; 187 SYSCTL_INT(_kern_cam_nda, OID_AUTO, max_trim, CTLFLAG_RDTUN, 188 &nda_max_trim_entries, NDA_MAX_TRIM_ENTRIES, 189 "Maximum number of BIO_DELETE to send down as a DSM TRIM."); 190 SYSCTL_INT(_kern_cam_nda, OID_AUTO, enable_biospeedup, CTLFLAG_RDTUN, 191 &nda_enable_biospeedup, 0, "Enable BIO_SPEEDUP processing."); 192 SYSCTL_INT(_kern_cam_nda, OID_AUTO, nvd_compat, CTLFLAG_RDTUN, 193 &nda_nvd_compat, 1, "Enable creation of nvd aliases."); 194 195 /* 196 * All NVMe media is non-rotational, so all nvme device instances 197 * share this to implement the sysctl. 198 */ 199 static int nda_rotating_media = 0; 200 201 static struct periph_driver ndadriver = 202 { 203 ndainit, "nda", 204 TAILQ_HEAD_INITIALIZER(ndadriver.units), /* generation */ 0 205 }; 206 207 PERIPHDRIVER_DECLARE(nda, ndadriver); 208 209 static MALLOC_DEFINE(M_NVMEDA, "nvme_da", "nvme_da buffers"); 210 211 /* 212 * nice wrappers. Maybe these belong in nvme_all.c instead of 213 * here, but this is the only place that uses these. Should 214 * we ever grow another NVME periph, we should move them 215 * all there wholesale. 216 */ 217 218 static void 219 nda_nvme_flush(struct nda_softc *softc, struct ccb_nvmeio *nvmeio) 220 { 221 cam_fill_nvmeio(nvmeio, 222 0, /* retries */ 223 ndadone, /* cbfcnp */ 224 CAM_DIR_NONE, /* flags */ 225 NULL, /* data_ptr */ 226 0, /* dxfer_len */ 227 nda_default_timeout * 1000); /* timeout 30s */ 228 nvme_ns_flush_cmd(&nvmeio->cmd, softc->nsid); 229 } 230 231 static void 232 nda_nvme_trim(struct nda_softc *softc, struct ccb_nvmeio *nvmeio, 233 void *payload, uint32_t num_ranges) 234 { 235 cam_fill_nvmeio(nvmeio, 236 0, /* retries */ 237 ndadone, /* cbfcnp */ 238 CAM_DIR_OUT, /* flags */ 239 payload, /* data_ptr */ 240 num_ranges * sizeof(struct nvme_dsm_range), /* dxfer_len */ 241 nda_default_timeout * 1000); /* timeout 30s */ 242 nvme_ns_trim_cmd(&nvmeio->cmd, softc->nsid, num_ranges); 243 } 244 245 static void 246 nda_nvme_write(struct nda_softc *softc, struct ccb_nvmeio *nvmeio, 247 void *payload, uint64_t lba, uint32_t len, uint32_t count) 248 { 249 cam_fill_nvmeio(nvmeio, 250 0, /* retries */ 251 ndadone, /* cbfcnp */ 252 CAM_DIR_OUT, /* flags */ 253 payload, /* data_ptr */ 254 len, /* dxfer_len */ 255 nda_default_timeout * 1000); /* timeout 30s */ 256 nvme_ns_write_cmd(&nvmeio->cmd, softc->nsid, lba, count); 257 } 258 259 static void 260 nda_nvme_rw_bio(struct nda_softc *softc, struct ccb_nvmeio *nvmeio, 261 struct bio *bp, uint32_t rwcmd) 262 { 263 int flags = rwcmd == NVME_OPC_READ ? CAM_DIR_IN : CAM_DIR_OUT; 264 void *payload; 265 uint64_t lba; 266 uint32_t count; 267 268 if (bp->bio_flags & BIO_UNMAPPED) { 269 flags |= CAM_DATA_BIO; 270 payload = bp; 271 } else { 272 payload = bp->bio_data; 273 } 274 275 lba = bp->bio_pblkno; 276 count = bp->bio_bcount / softc->disk->d_sectorsize; 277 278 cam_fill_nvmeio(nvmeio, 279 0, /* retries */ 280 ndadone, /* cbfcnp */ 281 flags, /* flags */ 282 payload, /* data_ptr */ 283 bp->bio_bcount, /* dxfer_len */ 284 nda_default_timeout * 1000); /* timeout 30s */ 285 nvme_ns_rw_cmd(&nvmeio->cmd, rwcmd, softc->nsid, lba, count); 286 } 287 288 static int 289 ndaopen(struct disk *dp) 290 { 291 struct cam_periph *periph; 292 struct nda_softc *softc; 293 int error; 294 295 periph = (struct cam_periph *)dp->d_drv1; 296 if (cam_periph_acquire(periph) != 0) { 297 return(ENXIO); 298 } 299 300 cam_periph_lock(periph); 301 if ((error = cam_periph_hold(periph, PRIBIO|PCATCH)) != 0) { 302 cam_periph_unlock(periph); 303 cam_periph_release(periph); 304 return (error); 305 } 306 307 CAM_DEBUG(periph->path, CAM_DEBUG_TRACE | CAM_DEBUG_PERIPH, 308 ("ndaopen\n")); 309 310 softc = (struct nda_softc *)periph->softc; 311 softc->flags |= NDA_FLAG_OPEN; 312 313 cam_periph_unhold(periph); 314 cam_periph_unlock(periph); 315 return (0); 316 } 317 318 static int 319 ndaclose(struct disk *dp) 320 { 321 struct cam_periph *periph; 322 struct nda_softc *softc; 323 union ccb *ccb; 324 int error; 325 326 periph = (struct cam_periph *)dp->d_drv1; 327 softc = (struct nda_softc *)periph->softc; 328 cam_periph_lock(periph); 329 330 CAM_DEBUG(periph->path, CAM_DEBUG_TRACE | CAM_DEBUG_PERIPH, 331 ("ndaclose\n")); 332 333 if ((softc->flags & NDA_FLAG_DIRTY) != 0 && 334 (periph->flags & CAM_PERIPH_INVALID) == 0 && 335 cam_periph_hold(periph, PRIBIO) == 0) { 336 ccb = cam_periph_getccb(periph, CAM_PRIORITY_NORMAL); 337 nda_nvme_flush(softc, &ccb->nvmeio); 338 error = cam_periph_runccb(ccb, ndaerror, /*cam_flags*/0, 339 /*sense_flags*/0, softc->disk->d_devstat); 340 341 if (error != 0) 342 xpt_print(periph->path, "Synchronize cache failed\n"); 343 else 344 softc->flags &= ~NDA_FLAG_DIRTY; 345 xpt_release_ccb(ccb); 346 cam_periph_unhold(periph); 347 } 348 349 softc->flags &= ~NDA_FLAG_OPEN; 350 351 while (softc->refcount != 0) 352 cam_periph_sleep(periph, &softc->refcount, PRIBIO, "ndaclose", 1); 353 KASSERT(softc->outstanding_cmds == 0, 354 ("nda %d outstanding commands", softc->outstanding_cmds)); 355 cam_periph_unlock(periph); 356 cam_periph_release(periph); 357 return (0); 358 } 359 360 static void 361 ndaschedule(struct cam_periph *periph) 362 { 363 struct nda_softc *softc = (struct nda_softc *)periph->softc; 364 365 if (softc->state != NDA_STATE_NORMAL) 366 return; 367 368 cam_iosched_schedule(softc->cam_iosched, periph); 369 } 370 371 static int 372 ndaioctl(struct disk *dp, u_long cmd, void *data, int fflag, 373 struct thread *td) 374 { 375 struct cam_periph *periph; 376 377 periph = (struct cam_periph *)dp->d_drv1; 378 379 switch (cmd) { 380 case NVME_IO_TEST: 381 case NVME_BIO_TEST: 382 /* 383 * These don't map well to the underlying CCBs, so 384 * they are usupported via CAM. 385 */ 386 return (ENOTTY); 387 case NVME_GET_NSID: 388 { 389 struct nvme_get_nsid *gnsid = (struct nvme_get_nsid *)data; 390 struct ccb_pathinq cpi; 391 392 xpt_path_inq(&cpi, periph->path); 393 strncpy(gnsid->cdev, cpi.xport_specific.nvme.dev_name, 394 sizeof(gnsid->cdev)); 395 gnsid->nsid = cpi.xport_specific.nvme.nsid; 396 return (0); 397 } 398 case NVME_PASSTHROUGH_CMD: 399 { 400 struct nvme_pt_command *pt; 401 union ccb *ccb; 402 struct cam_periph_map_info mapinfo; 403 u_int maxmap = dp->d_maxsize; 404 int error; 405 406 /* 407 * Create a NVME_IO CCB to do the passthrough command. 408 */ 409 pt = (struct nvme_pt_command *)data; 410 ccb = xpt_alloc_ccb(); 411 xpt_setup_ccb(&ccb->ccb_h, periph->path, CAM_PRIORITY_NORMAL); 412 ccb->ccb_state = NDA_CCB_PASS; 413 cam_fill_nvmeio(&ccb->nvmeio, 414 0, /* Retries */ 415 ndadone, 416 (pt->is_read ? CAM_DIR_IN : CAM_DIR_OUT) | CAM_DATA_VADDR, 417 pt->buf, 418 pt->len, 419 nda_default_timeout * 1000); 420 memcpy(&ccb->nvmeio.cmd, &pt->cmd, sizeof(pt->cmd)); 421 422 /* 423 * Wire the user memory in this request for the I/O 424 */ 425 memset(&mapinfo, 0, sizeof(mapinfo)); 426 error = cam_periph_mapmem(ccb, &mapinfo, maxmap); 427 if (error) 428 goto out; 429 430 /* 431 * Lock the periph and run the command. 432 */ 433 cam_periph_lock(periph); 434 cam_periph_runccb(ccb, NULL, CAM_RETRY_SELTO, 435 SF_RETRY_UA | SF_NO_PRINT, NULL); 436 437 /* 438 * Tear down mapping and return status. 439 */ 440 cam_periph_unlock(periph); 441 cam_periph_unmapmem(ccb, &mapinfo); 442 error = cam_ccb_success(ccb) ? 0 : EIO; 443 out: 444 cam_periph_lock(periph); 445 xpt_release_ccb(ccb); 446 cam_periph_unlock(periph); 447 return (error); 448 } 449 default: 450 break; 451 } 452 return (ENOTTY); 453 } 454 455 /* 456 * Actually translate the requested transfer into one the physical driver 457 * can understand. The transfer is described by a buf and will include 458 * only one physical transfer. 459 */ 460 static void 461 ndastrategy(struct bio *bp) 462 { 463 struct cam_periph *periph; 464 struct nda_softc *softc; 465 466 periph = (struct cam_periph *)bp->bio_disk->d_drv1; 467 softc = (struct nda_softc *)periph->softc; 468 469 cam_periph_lock(periph); 470 471 CAM_DEBUG(periph->path, CAM_DEBUG_TRACE, ("ndastrategy(%p)\n", bp)); 472 473 /* 474 * If the device has been made invalid, error out 475 */ 476 if ((periph->flags & CAM_PERIPH_INVALID) != 0) { 477 cam_periph_unlock(periph); 478 biofinish(bp, NULL, ENXIO); 479 return; 480 } 481 482 if (bp->bio_cmd == BIO_DELETE) 483 softc->deletes++; 484 485 /* 486 * Place it in the queue of disk activities for this disk 487 */ 488 cam_iosched_queue_work(softc->cam_iosched, bp); 489 490 /* 491 * Schedule ourselves for performing the work. 492 */ 493 ndaschedule(periph); 494 cam_periph_unlock(periph); 495 496 return; 497 } 498 499 static int 500 ndadump(void *arg, void *virtual, off_t offset, size_t length) 501 { 502 struct cam_periph *periph; 503 struct nda_softc *softc; 504 u_int secsize; 505 struct ccb_nvmeio nvmeio; 506 struct disk *dp; 507 uint64_t lba; 508 uint32_t count; 509 int error = 0; 510 511 dp = arg; 512 periph = dp->d_drv1; 513 softc = (struct nda_softc *)periph->softc; 514 secsize = softc->disk->d_sectorsize; 515 lba = offset / secsize; 516 count = length / secsize; 517 518 if ((periph->flags & CAM_PERIPH_INVALID) != 0) 519 return (ENXIO); 520 521 /* xpt_get_ccb returns a zero'd allocation for the ccb, mimic that here */ 522 memset(&nvmeio, 0, sizeof(nvmeio)); 523 if (length > 0) { 524 xpt_setup_ccb(&nvmeio.ccb_h, periph->path, CAM_PRIORITY_NORMAL); 525 nvmeio.ccb_state = NDA_CCB_DUMP; 526 nda_nvme_write(softc, &nvmeio, virtual, lba, length, count); 527 error = cam_periph_runccb((union ccb *)&nvmeio, cam_periph_error, 528 0, SF_NO_RECOVERY | SF_NO_RETRY, NULL); 529 if (error != 0) 530 printf("Aborting dump due to I/O error %d.\n", error); 531 532 return (error); 533 } 534 535 /* Flush */ 536 xpt_setup_ccb(&nvmeio.ccb_h, periph->path, CAM_PRIORITY_NORMAL); 537 538 nvmeio.ccb_state = NDA_CCB_DUMP; 539 nda_nvme_flush(softc, &nvmeio); 540 error = cam_periph_runccb((union ccb *)&nvmeio, cam_periph_error, 541 0, SF_NO_RECOVERY | SF_NO_RETRY, NULL); 542 if (error != 0) 543 xpt_print(periph->path, "flush cmd failed\n"); 544 return (error); 545 } 546 547 static void 548 ndainit(void) 549 { 550 cam_status status; 551 552 /* 553 * Install a global async callback. This callback will 554 * receive async callbacks like "new device found". 555 */ 556 status = xpt_register_async(AC_FOUND_DEVICE, ndaasync, NULL, NULL); 557 558 if (status != CAM_REQ_CMP) { 559 printf("nda: Failed to attach master async callback " 560 "due to status 0x%x!\n", status); 561 } else if (nda_send_ordered) { 562 /* Register our event handlers */ 563 if ((EVENTHANDLER_REGISTER(power_suspend, ndasuspend, 564 NULL, EVENTHANDLER_PRI_LAST)) == NULL) 565 printf("ndainit: power event registration failed!\n"); 566 if ((EVENTHANDLER_REGISTER(shutdown_post_sync, ndashutdown, 567 NULL, SHUTDOWN_PRI_DEFAULT)) == NULL) 568 printf("ndainit: shutdown event registration failed!\n"); 569 } 570 } 571 572 /* 573 * Callback from GEOM, called when it has finished cleaning up its 574 * resources. 575 */ 576 static void 577 ndadiskgonecb(struct disk *dp) 578 { 579 struct cam_periph *periph; 580 581 periph = (struct cam_periph *)dp->d_drv1; 582 583 cam_periph_release(periph); 584 } 585 586 static void 587 ndaoninvalidate(struct cam_periph *periph) 588 { 589 struct nda_softc *softc; 590 591 softc = (struct nda_softc *)periph->softc; 592 593 /* 594 * De-register any async callbacks. 595 */ 596 xpt_register_async(0, ndaasync, periph, periph->path); 597 #ifdef CAM_IO_STATS 598 softc->invalidations++; 599 #endif 600 601 /* 602 * Return all queued I/O with ENXIO. Transactions may be queued up here 603 * for retry (since we are called while there's other transactions 604 * pending). Any requests in the hardware will drain before ndacleanup 605 * is called. 606 */ 607 cam_iosched_flush(softc->cam_iosched, NULL, ENXIO); 608 609 /* 610 * Tell GEOM that we've gone away, we'll get a callback when it is 611 * done cleaning up its resources. 612 */ 613 disk_gone(softc->disk); 614 } 615 616 static void 617 ndacleanup(struct cam_periph *periph) 618 { 619 struct nda_softc *softc; 620 621 softc = (struct nda_softc *)periph->softc; 622 623 cam_periph_unlock(periph); 624 625 cam_iosched_fini(softc->cam_iosched); 626 627 /* 628 * If we can't free the sysctl tree, oh well... 629 */ 630 if ((softc->flags & NDA_FLAG_SCTX_INIT) != 0) { 631 #ifdef CAM_IO_STATS 632 if (sysctl_ctx_free(&softc->sysctl_stats_ctx) != 0) 633 xpt_print(periph->path, 634 "can't remove sysctl stats context\n"); 635 #endif 636 if (sysctl_ctx_free(&softc->sysctl_ctx) != 0) 637 xpt_print(periph->path, 638 "can't remove sysctl context\n"); 639 } 640 641 disk_destroy(softc->disk); 642 free(softc, M_DEVBUF); 643 cam_periph_lock(periph); 644 } 645 646 static void 647 ndaasync(void *callback_arg, uint32_t code, 648 struct cam_path *path, void *arg) 649 { 650 struct cam_periph *periph; 651 652 periph = (struct cam_periph *)callback_arg; 653 switch (code) { 654 case AC_FOUND_DEVICE: 655 { 656 struct ccb_getdev *cgd; 657 cam_status status; 658 659 cgd = (struct ccb_getdev *)arg; 660 if (cgd == NULL) 661 break; 662 663 if (cgd->protocol != PROTO_NVME) 664 break; 665 666 /* 667 * Allocate a peripheral instance for 668 * this device and start the probe 669 * process. 670 */ 671 status = cam_periph_alloc(ndaregister, ndaoninvalidate, 672 ndacleanup, ndastart, 673 "nda", CAM_PERIPH_BIO, 674 path, ndaasync, 675 AC_FOUND_DEVICE, cgd); 676 677 if (status != CAM_REQ_CMP 678 && status != CAM_REQ_INPROG) 679 printf("ndaasync: Unable to attach to new device " 680 "due to status 0x%x\n", status); 681 break; 682 } 683 case AC_ADVINFO_CHANGED: 684 { 685 uintptr_t buftype; 686 687 buftype = (uintptr_t)arg; 688 if (buftype == CDAI_TYPE_PHYS_PATH) { 689 struct nda_softc *softc; 690 691 softc = periph->softc; 692 disk_attr_changed(softc->disk, "GEOM::physpath", 693 M_NOWAIT); 694 } 695 break; 696 } 697 case AC_LOST_DEVICE: 698 default: 699 break; 700 } 701 cam_periph_async(periph, code, path, arg); 702 } 703 704 static void 705 ndasysctlinit(void *context, int pending) 706 { 707 struct cam_periph *periph; 708 struct nda_softc *softc; 709 char tmpstr[32], tmpstr2[16]; 710 711 periph = (struct cam_periph *)context; 712 713 /* periph was held for us when this task was enqueued */ 714 if ((periph->flags & CAM_PERIPH_INVALID) != 0) { 715 cam_periph_release(periph); 716 return; 717 } 718 719 softc = (struct nda_softc *)periph->softc; 720 snprintf(tmpstr, sizeof(tmpstr), "CAM NDA unit %d", periph->unit_number); 721 snprintf(tmpstr2, sizeof(tmpstr2), "%d", periph->unit_number); 722 723 sysctl_ctx_init(&softc->sysctl_ctx); 724 softc->flags |= NDA_FLAG_SCTX_INIT; 725 softc->sysctl_tree = SYSCTL_ADD_NODE_WITH_LABEL(&softc->sysctl_ctx, 726 SYSCTL_STATIC_CHILDREN(_kern_cam_nda), OID_AUTO, tmpstr2, 727 CTLFLAG_RD | CTLFLAG_MPSAFE, 0, tmpstr, "device_index"); 728 if (softc->sysctl_tree == NULL) { 729 printf("ndasysctlinit: unable to allocate sysctl tree\n"); 730 cam_periph_release(periph); 731 return; 732 } 733 734 SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree), 735 OID_AUTO, "unmapped_io", CTLFLAG_RD, 736 &softc->unmappedio, 0, "Unmapped I/O leaf"); 737 738 SYSCTL_ADD_QUAD(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree), 739 OID_AUTO, "deletes", CTLFLAG_RD, 740 &softc->deletes, "Number of BIO_DELETE requests"); 741 742 SYSCTL_ADD_UQUAD(&softc->sysctl_ctx, 743 SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO, 744 "trim_count", CTLFLAG_RD, &softc->trim_count, 745 "Total number of unmap/dsm commands sent"); 746 SYSCTL_ADD_UQUAD(&softc->sysctl_ctx, 747 SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO, 748 "trim_ranges", CTLFLAG_RD, &softc->trim_ranges, 749 "Total number of ranges in unmap/dsm commands"); 750 SYSCTL_ADD_UQUAD(&softc->sysctl_ctx, 751 SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO, 752 "trim_lbas", CTLFLAG_RD, &softc->trim_lbas, 753 "Total lbas in the unmap/dsm commands sent"); 754 755 SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree), 756 OID_AUTO, "rotating", CTLFLAG_RD, &nda_rotating_media, 1, 757 "Rotating media"); 758 759 SYSCTL_ADD_PROC(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree), 760 OID_AUTO, "flags", CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, 761 softc, 0, ndaflagssysctl, "A", 762 "Flags for drive"); 763 764 #ifdef CAM_IO_STATS 765 softc->sysctl_stats_tree = SYSCTL_ADD_NODE(&softc->sysctl_stats_ctx, 766 SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO, "stats", 767 CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "Statistics"); 768 if (softc->sysctl_stats_tree == NULL) { 769 printf("ndasysctlinit: unable to allocate sysctl tree for stats\n"); 770 cam_periph_release(periph); 771 return; 772 } 773 SYSCTL_ADD_INT(&softc->sysctl_stats_ctx, 774 SYSCTL_CHILDREN(softc->sysctl_stats_tree), 775 OID_AUTO, "timeouts", CTLFLAG_RD, 776 &softc->timeouts, 0, 777 "Device timeouts reported by the SIM"); 778 SYSCTL_ADD_INT(&softc->sysctl_stats_ctx, 779 SYSCTL_CHILDREN(softc->sysctl_stats_tree), 780 OID_AUTO, "errors", CTLFLAG_RD, 781 &softc->errors, 0, 782 "Transport errors reported by the SIM."); 783 SYSCTL_ADD_INT(&softc->sysctl_stats_ctx, 784 SYSCTL_CHILDREN(softc->sysctl_stats_tree), 785 OID_AUTO, "pack_invalidations", CTLFLAG_RD, 786 &softc->invalidations, 0, 787 "Device pack invalidations."); 788 #endif 789 790 #ifdef CAM_TEST_FAILURE 791 SYSCTL_ADD_PROC(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree), 792 OID_AUTO, "invalidate", CTLTYPE_U64 | CTLFLAG_RW | CTLFLAG_MPSAFE, 793 periph, 0, cam_periph_invalidate_sysctl, "I", 794 "Write 1 to invalidate the drive immediately"); 795 #endif 796 797 cam_iosched_sysctl_init(softc->cam_iosched, &softc->sysctl_ctx, 798 softc->sysctl_tree); 799 800 cam_periph_release(periph); 801 } 802 803 static int 804 ndaflagssysctl(SYSCTL_HANDLER_ARGS) 805 { 806 struct sbuf sbuf; 807 struct nda_softc *softc = arg1; 808 int error; 809 810 sbuf_new_for_sysctl(&sbuf, NULL, 0, req); 811 if (softc->flags != 0) 812 sbuf_printf(&sbuf, "0x%b", (unsigned)softc->flags, NDA_FLAG_STRING); 813 else 814 sbuf_printf(&sbuf, "0"); 815 error = sbuf_finish(&sbuf); 816 sbuf_delete(&sbuf); 817 818 return (error); 819 } 820 821 static int 822 ndagetattr(struct bio *bp) 823 { 824 int ret; 825 struct cam_periph *periph; 826 827 if (g_handleattr_int(bp, "GEOM::canspeedup", nda_enable_biospeedup)) 828 return (EJUSTRETURN); 829 830 periph = (struct cam_periph *)bp->bio_disk->d_drv1; 831 cam_periph_lock(periph); 832 ret = xpt_getattr(bp->bio_data, bp->bio_length, bp->bio_attribute, 833 periph->path); 834 cam_periph_unlock(periph); 835 if (ret == 0) 836 bp->bio_completed = bp->bio_length; 837 return ret; 838 } 839 840 static cam_status 841 ndaregister(struct cam_periph *periph, void *arg) 842 { 843 struct nda_softc *softc; 844 struct disk *disk; 845 struct ccb_pathinq cpi; 846 const struct nvme_namespace_data *nsd; 847 const struct nvme_controller_data *cd; 848 char announce_buf[80]; 849 uint8_t flbas_fmt, lbads, vwc_present; 850 u_int maxio; 851 int quirks; 852 853 nsd = nvme_get_identify_ns(periph); 854 cd = nvme_get_identify_cntrl(periph); 855 856 softc = (struct nda_softc *)malloc(sizeof(*softc), M_DEVBUF, 857 M_NOWAIT | M_ZERO); 858 859 if (softc == NULL) { 860 printf("ndaregister: Unable to probe new device. " 861 "Unable to allocate softc\n"); 862 return(CAM_REQ_CMP_ERR); 863 } 864 865 if (cam_iosched_init(&softc->cam_iosched, periph) != 0) { 866 printf("ndaregister: Unable to probe new device. " 867 "Unable to allocate iosched memory\n"); 868 free(softc, M_DEVBUF); 869 return(CAM_REQ_CMP_ERR); 870 } 871 872 /* ident_data parsing */ 873 874 periph->softc = softc; 875 softc->quirks = NDA_Q_NONE; 876 xpt_path_inq(&cpi, periph->path); 877 TASK_INIT(&softc->sysctl_task, 0, ndasysctlinit, periph); 878 879 /* 880 * The name space ID is the lun, save it for later I/O 881 */ 882 softc->nsid = (uint32_t)xpt_path_lun_id(periph->path); 883 884 /* 885 * Register this media as a disk 886 */ 887 (void)cam_periph_acquire(periph); 888 cam_periph_unlock(periph); 889 snprintf(announce_buf, sizeof(announce_buf), 890 "kern.cam.nda.%d.quirks", periph->unit_number); 891 quirks = softc->quirks; 892 TUNABLE_INT_FETCH(announce_buf, &quirks); 893 softc->quirks = quirks; 894 cam_iosched_set_sort_queue(softc->cam_iosched, 0); 895 softc->disk = disk = disk_alloc(); 896 disk->d_rotation_rate = DISK_RR_NON_ROTATING; 897 disk->d_open = ndaopen; 898 disk->d_close = ndaclose; 899 disk->d_strategy = ndastrategy; 900 disk->d_ioctl = ndaioctl; 901 disk->d_getattr = ndagetattr; 902 if (cam_sim_pollable(periph->sim)) 903 disk->d_dump = ndadump; 904 disk->d_gone = ndadiskgonecb; 905 disk->d_name = "nda"; 906 disk->d_drv1 = periph; 907 disk->d_unit = periph->unit_number; 908 maxio = cpi.maxio; /* Honor max I/O size of SIM */ 909 if (maxio == 0) 910 maxio = DFLTPHYS; /* traditional default */ 911 else if (maxio > maxphys) 912 maxio = maxphys; /* for safety */ 913 disk->d_maxsize = maxio; 914 flbas_fmt = (nsd->flbas >> NVME_NS_DATA_FLBAS_FORMAT_SHIFT) & 915 NVME_NS_DATA_FLBAS_FORMAT_MASK; 916 lbads = (nsd->lbaf[flbas_fmt] >> NVME_NS_DATA_LBAF_LBADS_SHIFT) & 917 NVME_NS_DATA_LBAF_LBADS_MASK; 918 disk->d_sectorsize = 1 << lbads; 919 disk->d_mediasize = (off_t)(disk->d_sectorsize * nsd->nsze); 920 disk->d_delmaxsize = disk->d_mediasize; 921 disk->d_flags = DISKFLAG_DIRECT_COMPLETION; 922 if (nvme_ctrlr_has_dataset_mgmt(cd)) 923 disk->d_flags |= DISKFLAG_CANDELETE; 924 vwc_present = (cd->vwc >> NVME_CTRLR_DATA_VWC_PRESENT_SHIFT) & 925 NVME_CTRLR_DATA_VWC_PRESENT_MASK; 926 if (vwc_present) 927 disk->d_flags |= DISKFLAG_CANFLUSHCACHE; 928 if ((cpi.hba_misc & PIM_UNMAPPED) != 0) { 929 disk->d_flags |= DISKFLAG_UNMAPPED_BIO; 930 softc->unmappedio = 1; 931 } 932 /* 933 * d_ident and d_descr are both far bigger than the length of either 934 * the serial or model number strings. 935 */ 936 cam_strvis_flag(disk->d_descr, cd->mn, NVME_MODEL_NUMBER_LENGTH, 937 sizeof(disk->d_descr), CAM_STRVIS_FLAG_NONASCII_SPC); 938 939 cam_strvis_flag(disk->d_ident, cd->sn, NVME_SERIAL_NUMBER_LENGTH, 940 sizeof(disk->d_ident), CAM_STRVIS_FLAG_NONASCII_SPC); 941 942 disk->d_hba_vendor = cpi.hba_vendor; 943 disk->d_hba_device = cpi.hba_device; 944 disk->d_hba_subvendor = cpi.hba_subvendor; 945 disk->d_hba_subdevice = cpi.hba_subdevice; 946 snprintf(disk->d_attachment, sizeof(disk->d_attachment), 947 "%s%d", cpi.dev_name, cpi.unit_number); 948 if (((nsd->nsfeat >> NVME_NS_DATA_NSFEAT_NPVALID_SHIFT) & 949 NVME_NS_DATA_NSFEAT_NPVALID_MASK) != 0 && nsd->npwg != 0) 950 disk->d_stripesize = ((nsd->npwg + 1) * disk->d_sectorsize); 951 else 952 disk->d_stripesize = nsd->noiob * disk->d_sectorsize; 953 disk->d_stripeoffset = 0; 954 disk->d_devstat = devstat_new_entry(periph->periph_name, 955 periph->unit_number, disk->d_sectorsize, 956 DEVSTAT_ALL_SUPPORTED, 957 DEVSTAT_TYPE_DIRECT | XPORT_DEVSTAT_TYPE(cpi.transport), 958 DEVSTAT_PRIORITY_DISK); 959 /* 960 * Add alias for older nvd drives to ease transition. 961 */ 962 if (nda_nvd_compat) 963 disk_add_alias(disk, "nvd"); 964 965 cam_periph_lock(periph); 966 967 snprintf(announce_buf, sizeof(announce_buf), 968 "%juMB (%ju %u byte sectors)", 969 (uintmax_t)((uintmax_t)disk->d_mediasize / (1024*1024)), 970 (uintmax_t)disk->d_mediasize / disk->d_sectorsize, 971 disk->d_sectorsize); 972 xpt_announce_periph(periph, announce_buf); 973 xpt_announce_quirks(periph, softc->quirks, NDA_Q_BIT_STRING); 974 975 /* 976 * Create our sysctl variables, now that we know 977 * we have successfully attached. 978 */ 979 if (cam_periph_acquire(periph) == 0) 980 taskqueue_enqueue(taskqueue_thread, &softc->sysctl_task); 981 982 /* 983 * Register for device going away and info about the drive 984 * changing (though with NVMe, it can't) 985 */ 986 xpt_register_async(AC_LOST_DEVICE | AC_ADVINFO_CHANGED, 987 ndaasync, periph, periph->path); 988 989 softc->state = NDA_STATE_NORMAL; 990 991 /* 992 * We'll release this reference once GEOM calls us back via 993 * ndadiskgonecb(), telling us that our provider has been freed. 994 */ 995 if (cam_periph_acquire(periph) == 0) 996 disk_create(softc->disk, DISK_VERSION); 997 998 cam_periph_release_locked(periph); 999 return(CAM_REQ_CMP); 1000 } 1001 1002 static void 1003 ndastart(struct cam_periph *periph, union ccb *start_ccb) 1004 { 1005 struct nda_softc *softc = (struct nda_softc *)periph->softc; 1006 struct ccb_nvmeio *nvmeio = &start_ccb->nvmeio; 1007 1008 CAM_DEBUG(periph->path, CAM_DEBUG_TRACE, ("ndastart\n")); 1009 1010 switch (softc->state) { 1011 case NDA_STATE_NORMAL: 1012 { 1013 struct bio *bp; 1014 1015 bp = cam_iosched_next_bio(softc->cam_iosched); 1016 CAM_DEBUG(periph->path, CAM_DEBUG_TRACE, ("ndastart: bio %p\n", bp)); 1017 if (bp == NULL) { 1018 xpt_release_ccb(start_ccb); 1019 break; 1020 } 1021 1022 switch (bp->bio_cmd) { 1023 case BIO_WRITE: 1024 softc->flags |= NDA_FLAG_DIRTY; 1025 /* FALLTHROUGH */ 1026 case BIO_READ: 1027 { 1028 #ifdef CAM_TEST_FAILURE 1029 int fail = 0; 1030 1031 /* 1032 * Support the failure ioctls. If the command is a 1033 * read, and there are pending forced read errors, or 1034 * if a write and pending write errors, then fail this 1035 * operation with EIO. This is useful for testing 1036 * purposes. Also, support having every Nth read fail. 1037 * 1038 * This is a rather blunt tool. 1039 */ 1040 if (bp->bio_cmd == BIO_READ) { 1041 if (softc->force_read_error) { 1042 softc->force_read_error--; 1043 fail = 1; 1044 } 1045 if (softc->periodic_read_error > 0) { 1046 if (++softc->periodic_read_count >= 1047 softc->periodic_read_error) { 1048 softc->periodic_read_count = 0; 1049 fail = 1; 1050 } 1051 } 1052 } else { 1053 if (softc->force_write_error) { 1054 softc->force_write_error--; 1055 fail = 1; 1056 } 1057 } 1058 if (fail) { 1059 biofinish(bp, NULL, EIO); 1060 xpt_release_ccb(start_ccb); 1061 ndaschedule(periph); 1062 return; 1063 } 1064 #endif 1065 KASSERT((bp->bio_flags & BIO_UNMAPPED) == 0 || 1066 round_page(bp->bio_bcount + bp->bio_ma_offset) / 1067 PAGE_SIZE == bp->bio_ma_n, 1068 ("Short bio %p", bp)); 1069 nda_nvme_rw_bio(softc, &start_ccb->nvmeio, bp, bp->bio_cmd == BIO_READ ? 1070 NVME_OPC_READ : NVME_OPC_WRITE); 1071 break; 1072 } 1073 case BIO_DELETE: 1074 { 1075 struct nvme_dsm_range *dsm_range, *dsm_end; 1076 struct nda_trim_request *trim; 1077 struct bio *bp1; 1078 int ents; 1079 uint32_t totalcount = 0, ranges = 0; 1080 1081 trim = malloc(sizeof(*trim), M_NVMEDA, M_ZERO | M_NOWAIT); 1082 if (trim == NULL) { 1083 biofinish(bp, NULL, ENOMEM); 1084 xpt_release_ccb(start_ccb); 1085 ndaschedule(periph); 1086 return; 1087 } 1088 TAILQ_INIT(&trim->bps); 1089 bp1 = bp; 1090 ents = min(nitems(trim->dsm), nda_max_trim_entries); 1091 ents = max(ents, 1); 1092 dsm_range = trim->dsm; 1093 dsm_end = dsm_range + ents; 1094 do { 1095 TAILQ_INSERT_TAIL(&trim->bps, bp1, bio_queue); 1096 dsm_range->length = 1097 htole32(bp1->bio_bcount / softc->disk->d_sectorsize); 1098 dsm_range->starting_lba = 1099 htole64(bp1->bio_offset / softc->disk->d_sectorsize); 1100 ranges++; 1101 totalcount += dsm_range->length; 1102 dsm_range++; 1103 if (dsm_range >= dsm_end) 1104 break; 1105 bp1 = cam_iosched_next_trim(softc->cam_iosched); 1106 /* XXX -- Could collapse adjacent ranges, but we don't for now */ 1107 /* XXX -- Could limit based on total payload size */ 1108 } while (bp1 != NULL); 1109 start_ccb->ccb_trim = trim; 1110 nda_nvme_trim(softc, &start_ccb->nvmeio, trim->dsm, 1111 dsm_range - trim->dsm); 1112 start_ccb->ccb_state = NDA_CCB_TRIM; 1113 softc->trim_count++; 1114 softc->trim_ranges += ranges; 1115 softc->trim_lbas += totalcount; 1116 /* 1117 * Note: We can have multiple TRIMs in flight, so we don't call 1118 * cam_iosched_submit_trim(softc->cam_iosched); 1119 * since that forces the I/O scheduler to only schedule one at a time. 1120 * On NVMe drives, this is a performance disaster. 1121 */ 1122 goto out; 1123 } 1124 case BIO_FLUSH: 1125 nda_nvme_flush(softc, nvmeio); 1126 break; 1127 default: 1128 biofinish(bp, NULL, EOPNOTSUPP); 1129 xpt_release_ccb(start_ccb); 1130 ndaschedule(periph); 1131 return; 1132 } 1133 start_ccb->ccb_state = NDA_CCB_BUFFER_IO; 1134 start_ccb->ccb_bp = bp; 1135 out: 1136 start_ccb->ccb_h.flags |= CAM_UNLOCKED; 1137 softc->outstanding_cmds++; 1138 softc->refcount++; /* For submission only */ 1139 cam_periph_unlock(periph); 1140 xpt_action(start_ccb); 1141 cam_periph_lock(periph); 1142 softc->refcount--; /* Submission done */ 1143 1144 /* May have more work to do, so ensure we stay scheduled */ 1145 ndaschedule(periph); 1146 break; 1147 } 1148 } 1149 } 1150 1151 static void 1152 ndadone(struct cam_periph *periph, union ccb *done_ccb) 1153 { 1154 struct nda_softc *softc; 1155 struct ccb_nvmeio *nvmeio = &done_ccb->nvmeio; 1156 struct cam_path *path; 1157 int state; 1158 1159 softc = (struct nda_softc *)periph->softc; 1160 path = done_ccb->ccb_h.path; 1161 1162 CAM_DEBUG(path, CAM_DEBUG_TRACE, ("ndadone\n")); 1163 1164 state = nvmeio->ccb_state & NDA_CCB_TYPE_MASK; 1165 switch (state) { 1166 case NDA_CCB_BUFFER_IO: 1167 case NDA_CCB_TRIM: 1168 { 1169 int error; 1170 1171 cam_periph_lock(periph); 1172 if ((done_ccb->ccb_h.status & CAM_STATUS_MASK) != CAM_REQ_CMP) { 1173 error = ndaerror(done_ccb, 0, 0); 1174 if (error == ERESTART) { 1175 /* A retry was scheduled, so just return. */ 1176 cam_periph_unlock(periph); 1177 return; 1178 } 1179 if ((done_ccb->ccb_h.status & CAM_DEV_QFRZN) != 0) 1180 cam_release_devq(path, 1181 /*relsim_flags*/0, 1182 /*reduction*/0, 1183 /*timeout*/0, 1184 /*getcount_only*/0); 1185 } else { 1186 if ((done_ccb->ccb_h.status & CAM_DEV_QFRZN) != 0) 1187 panic("REQ_CMP with QFRZN"); 1188 error = 0; 1189 } 1190 if (state == NDA_CCB_BUFFER_IO) { 1191 struct bio *bp; 1192 1193 bp = (struct bio *)done_ccb->ccb_bp; 1194 bp->bio_error = error; 1195 if (error != 0) { 1196 bp->bio_resid = bp->bio_bcount; 1197 bp->bio_flags |= BIO_ERROR; 1198 } else { 1199 bp->bio_resid = 0; 1200 } 1201 softc->outstanding_cmds--; 1202 1203 /* 1204 * We need to call cam_iosched before we call biodone so that we 1205 * don't measure any activity that happens in the completion 1206 * routine, which in the case of sendfile can be quite 1207 * extensive. 1208 */ 1209 cam_iosched_bio_complete(softc->cam_iosched, bp, done_ccb); 1210 xpt_release_ccb(done_ccb); 1211 ndaschedule(periph); 1212 cam_periph_unlock(periph); 1213 biodone(bp); 1214 } else { /* state == NDA_CCB_TRIM */ 1215 struct nda_trim_request *trim; 1216 struct bio *bp1, *bp2; 1217 TAILQ_HEAD(, bio) queue; 1218 1219 trim = nvmeio->ccb_trim; 1220 TAILQ_INIT(&queue); 1221 TAILQ_CONCAT(&queue, &trim->bps, bio_queue); 1222 free(trim, M_NVMEDA); 1223 1224 /* 1225 * Since we can have multiple trims in flight, we don't 1226 * need to call this here. 1227 * cam_iosched_trim_done(softc->cam_iosched); 1228 */ 1229 /* 1230 * The the I/O scheduler that we're finishing the I/O 1231 * so we can keep book. The first one we pass in the CCB 1232 * which has the timing information. The rest we pass in NULL 1233 * so we can keep proper counts. 1234 */ 1235 bp1 = TAILQ_FIRST(&queue); 1236 cam_iosched_bio_complete(softc->cam_iosched, bp1, done_ccb); 1237 xpt_release_ccb(done_ccb); 1238 softc->outstanding_cmds--; 1239 ndaschedule(periph); 1240 cam_periph_unlock(periph); 1241 while ((bp2 = TAILQ_FIRST(&queue)) != NULL) { 1242 TAILQ_REMOVE(&queue, bp2, bio_queue); 1243 bp2->bio_error = error; 1244 if (error != 0) { 1245 bp2->bio_flags |= BIO_ERROR; 1246 bp2->bio_resid = bp1->bio_bcount; 1247 } else 1248 bp2->bio_resid = 0; 1249 if (bp1 != bp2) 1250 cam_iosched_bio_complete(softc->cam_iosched, bp2, NULL); 1251 biodone(bp2); 1252 } 1253 } 1254 return; 1255 } 1256 case NDA_CCB_DUMP: 1257 /* No-op. We're polling */ 1258 return; 1259 case NDA_CCB_PASS: 1260 /* NVME_PASSTHROUGH_CMD runs this CCB and releases it */ 1261 return; 1262 default: 1263 break; 1264 } 1265 xpt_release_ccb(done_ccb); 1266 } 1267 1268 static int 1269 ndaerror(union ccb *ccb, uint32_t cam_flags, uint32_t sense_flags) 1270 { 1271 #ifdef CAM_IO_STATS 1272 struct nda_softc *softc; 1273 struct cam_periph *periph; 1274 1275 periph = xpt_path_periph(ccb->ccb_h.path); 1276 softc = (struct nda_softc *)periph->softc; 1277 #endif 1278 1279 switch (ccb->ccb_h.status & CAM_STATUS_MASK) { 1280 case CAM_CMD_TIMEOUT: 1281 #ifdef CAM_IO_STATS 1282 softc->timeouts++; 1283 #endif 1284 break; 1285 case CAM_REQ_CMP_ERR: 1286 case CAM_NVME_STATUS_ERROR: 1287 #ifdef CAM_IO_STATS 1288 softc->errors++; 1289 #endif 1290 break; 1291 default: 1292 break; 1293 } 1294 1295 return(cam_periph_error(ccb, cam_flags, sense_flags)); 1296 } 1297 1298 /* 1299 * Step through all NDA peripheral drivers, and if the device is still open, 1300 * sync the disk cache to physical media. 1301 */ 1302 static void 1303 ndaflush(void) 1304 { 1305 struct cam_periph *periph; 1306 struct nda_softc *softc; 1307 union ccb *ccb; 1308 int error; 1309 1310 CAM_PERIPH_FOREACH(periph, &ndadriver) { 1311 softc = (struct nda_softc *)periph->softc; 1312 1313 if (SCHEDULER_STOPPED()) { 1314 /* 1315 * If we panicked with the lock held or the periph is not 1316 * open, do not recurse. Otherwise, call ndadump since 1317 * that avoids the sleeping cam_periph_getccb does if no 1318 * CCBs are available. 1319 */ 1320 if (!cam_periph_owned(periph) && 1321 (softc->flags & NDA_FLAG_OPEN)) { 1322 ndadump(softc->disk, NULL, 0, 0); 1323 } 1324 continue; 1325 } 1326 1327 /* 1328 * We only sync the cache if the drive is still open 1329 */ 1330 cam_periph_lock(periph); 1331 if ((softc->flags & NDA_FLAG_OPEN) == 0) { 1332 cam_periph_unlock(periph); 1333 continue; 1334 } 1335 1336 ccb = cam_periph_getccb(periph, CAM_PRIORITY_NORMAL); 1337 nda_nvme_flush(softc, &ccb->nvmeio); 1338 error = cam_periph_runccb(ccb, ndaerror, /*cam_flags*/0, 1339 /*sense_flags*/ SF_NO_RECOVERY | SF_NO_RETRY, 1340 softc->disk->d_devstat); 1341 if (error != 0) 1342 xpt_print(periph->path, "Synchronize cache failed\n"); 1343 xpt_release_ccb(ccb); 1344 cam_periph_unlock(periph); 1345 } 1346 } 1347 1348 static void 1349 ndashutdown(void *arg, int howto) 1350 { 1351 1352 if ((howto & RB_NOSYNC) != 0) 1353 return; 1354 1355 ndaflush(); 1356 } 1357 1358 static void 1359 ndasuspend(void *arg) 1360 { 1361 1362 ndaflush(); 1363 } 1364