1 /* 2 * XenBSD block device driver 3 * 4 * Copyright (c) 2010-2013 Spectra Logic Corporation 5 * Copyright (c) 2009 Scott Long, Yahoo! 6 * Copyright (c) 2009 Frank Suchomel, Citrix 7 * Copyright (c) 2009 Doug F. Rabson, Citrix 8 * Copyright (c) 2005 Kip Macy 9 * Copyright (c) 2003-2004, Keir Fraser & Steve Hand 10 * Modifications by Mark A. Williamson are (c) Intel Research Cambridge 11 * 12 * 13 * Permission is hereby granted, free of charge, to any person obtaining a copy 14 * of this software and associated documentation files (the "Software"), to 15 * deal in the Software without restriction, including without limitation the 16 * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or 17 * sell copies of the Software, and to permit persons to whom the Software is 18 * furnished to do so, subject to the following conditions: 19 * 20 * The above copyright notice and this permission notice shall be included in 21 * all copies or substantial portions of the Software. 22 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 23 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 24 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 25 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 26 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING 27 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER 28 * DEALINGS IN THE SOFTWARE. 29 */ 30 31 #include <sys/cdefs.h> 32 __FBSDID("$FreeBSD$"); 33 34 #include <sys/param.h> 35 #include <sys/systm.h> 36 #include <sys/malloc.h> 37 #include <sys/kernel.h> 38 #include <vm/vm.h> 39 #include <vm/pmap.h> 40 41 #include <sys/bio.h> 42 #include <sys/bus.h> 43 #include <sys/conf.h> 44 #include <sys/module.h> 45 #include <sys/sysctl.h> 46 47 #include <machine/bus.h> 48 #include <sys/rman.h> 49 #include <machine/resource.h> 50 #include <machine/vmparam.h> 51 52 #include <xen/xen-os.h> 53 #include <xen/hypervisor.h> 54 #include <xen/xen_intr.h> 55 #include <xen/gnttab.h> 56 #include <contrib/xen/grant_table.h> 57 #include <contrib/xen/io/protocols.h> 58 #include <xen/xenbus/xenbusvar.h> 59 60 #include <machine/_inttypes.h> 61 62 #include <geom/geom_disk.h> 63 64 #include <dev/xen/blkfront/block.h> 65 66 #include "xenbus_if.h" 67 68 /*--------------------------- Forward Declarations ---------------------------*/ 69 static void xbd_closing(device_t); 70 static void xbd_startio(struct xbd_softc *sc); 71 72 /*---------------------------------- Macros ----------------------------------*/ 73 #if 0 74 #define DPRINTK(fmt, args...) printf("[XEN] %s:%d: " fmt ".\n", __func__, __LINE__, ##args) 75 #else 76 #define DPRINTK(fmt, args...) 77 #endif 78 79 #define XBD_SECTOR_SHFT 9 80 81 /*---------------------------- Global Static Data ----------------------------*/ 82 static MALLOC_DEFINE(M_XENBLOCKFRONT, "xbd", "Xen Block Front driver data"); 83 84 static int xbd_enable_indirect = 1; 85 SYSCTL_NODE(_hw, OID_AUTO, xbd, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 86 "xbd driver parameters"); 87 SYSCTL_INT(_hw_xbd, OID_AUTO, xbd_enable_indirect, CTLFLAG_RDTUN, 88 &xbd_enable_indirect, 0, "Enable xbd indirect segments"); 89 90 /*---------------------------- Command Processing ----------------------------*/ 91 static void 92 xbd_freeze(struct xbd_softc *sc, xbd_flag_t xbd_flag) 93 { 94 if (xbd_flag != XBDF_NONE && (sc->xbd_flags & xbd_flag) != 0) 95 return; 96 97 sc->xbd_flags |= xbd_flag; 98 sc->xbd_qfrozen_cnt++; 99 } 100 101 static void 102 xbd_thaw(struct xbd_softc *sc, xbd_flag_t xbd_flag) 103 { 104 if (xbd_flag != XBDF_NONE && (sc->xbd_flags & xbd_flag) == 0) 105 return; 106 107 if (sc->xbd_qfrozen_cnt == 0) 108 panic("%s: Thaw with flag 0x%x while not frozen.", 109 __func__, xbd_flag); 110 111 sc->xbd_flags &= ~xbd_flag; 112 sc->xbd_qfrozen_cnt--; 113 } 114 115 static void 116 xbd_cm_freeze(struct xbd_softc *sc, struct xbd_command *cm, xbdc_flag_t cm_flag) 117 { 118 if ((cm->cm_flags & XBDCF_FROZEN) != 0) 119 return; 120 121 cm->cm_flags |= XBDCF_FROZEN|cm_flag; 122 xbd_freeze(sc, XBDF_NONE); 123 } 124 125 static void 126 xbd_cm_thaw(struct xbd_softc *sc, struct xbd_command *cm) 127 { 128 if ((cm->cm_flags & XBDCF_FROZEN) == 0) 129 return; 130 131 cm->cm_flags &= ~XBDCF_FROZEN; 132 xbd_thaw(sc, XBDF_NONE); 133 } 134 135 static inline void 136 xbd_flush_requests(struct xbd_softc *sc) 137 { 138 int notify; 139 140 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->xbd_ring, notify); 141 142 if (notify) 143 xen_intr_signal(sc->xen_intr_handle); 144 } 145 146 static void 147 xbd_free_command(struct xbd_command *cm) 148 { 149 150 KASSERT((cm->cm_flags & XBDCF_Q_MASK) == XBD_Q_NONE, 151 ("Freeing command that is still on queue %d.", 152 cm->cm_flags & XBDCF_Q_MASK)); 153 154 cm->cm_flags = XBDCF_INITIALIZER; 155 cm->cm_bp = NULL; 156 cm->cm_complete = NULL; 157 xbd_enqueue_cm(cm, XBD_Q_FREE); 158 xbd_thaw(cm->cm_sc, XBDF_CM_SHORTAGE); 159 } 160 161 static void 162 xbd_mksegarray(bus_dma_segment_t *segs, int nsegs, 163 grant_ref_t * gref_head, int otherend_id, int readonly, 164 grant_ref_t * sg_ref, struct blkif_request_segment *sg) 165 { 166 struct blkif_request_segment *last_block_sg = sg + nsegs; 167 vm_paddr_t buffer_ma; 168 uint64_t fsect, lsect; 169 int ref; 170 171 while (sg < last_block_sg) { 172 KASSERT(segs->ds_addr % (1 << XBD_SECTOR_SHFT) == 0, 173 ("XEN disk driver I/O must be sector aligned")); 174 KASSERT(segs->ds_len % (1 << XBD_SECTOR_SHFT) == 0, 175 ("XEN disk driver I/Os must be a multiple of " 176 "the sector length")); 177 buffer_ma = segs->ds_addr; 178 fsect = (buffer_ma & PAGE_MASK) >> XBD_SECTOR_SHFT; 179 lsect = fsect + (segs->ds_len >> XBD_SECTOR_SHFT) - 1; 180 181 KASSERT(lsect <= 7, ("XEN disk driver data cannot " 182 "cross a page boundary")); 183 184 /* install a grant reference. */ 185 ref = gnttab_claim_grant_reference(gref_head); 186 187 /* 188 * GNTTAB_LIST_END == 0xffffffff, but it is private 189 * to gnttab.c. 190 */ 191 KASSERT(ref != ~0, ("grant_reference failed")); 192 193 gnttab_grant_foreign_access_ref( 194 ref, 195 otherend_id, 196 buffer_ma >> PAGE_SHIFT, 197 readonly); 198 199 *sg_ref = ref; 200 *sg = (struct blkif_request_segment) { 201 .gref = ref, 202 .first_sect = fsect, 203 .last_sect = lsect 204 }; 205 sg++; 206 sg_ref++; 207 segs++; 208 } 209 } 210 211 static void 212 xbd_queue_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error) 213 { 214 struct xbd_softc *sc; 215 struct xbd_command *cm; 216 int op; 217 218 cm = arg; 219 sc = cm->cm_sc; 220 221 if (error) { 222 cm->cm_bp->bio_error = EIO; 223 biodone(cm->cm_bp); 224 xbd_free_command(cm); 225 return; 226 } 227 228 KASSERT(nsegs <= sc->xbd_max_request_segments, 229 ("Too many segments in a blkfront I/O")); 230 231 if (nsegs <= BLKIF_MAX_SEGMENTS_PER_REQUEST) { 232 blkif_request_t *ring_req; 233 234 /* Fill out a blkif_request_t structure. */ 235 ring_req = (blkif_request_t *) 236 RING_GET_REQUEST(&sc->xbd_ring, sc->xbd_ring.req_prod_pvt); 237 sc->xbd_ring.req_prod_pvt++; 238 ring_req->id = cm->cm_id; 239 ring_req->operation = cm->cm_operation; 240 ring_req->sector_number = cm->cm_sector_number; 241 ring_req->handle = (blkif_vdev_t)(uintptr_t)sc->xbd_disk; 242 ring_req->nr_segments = nsegs; 243 cm->cm_nseg = nsegs; 244 xbd_mksegarray(segs, nsegs, &cm->cm_gref_head, 245 xenbus_get_otherend_id(sc->xbd_dev), 246 cm->cm_operation == BLKIF_OP_WRITE, 247 cm->cm_sg_refs, ring_req->seg); 248 } else { 249 blkif_request_indirect_t *ring_req; 250 251 /* Fill out a blkif_request_indirect_t structure. */ 252 ring_req = (blkif_request_indirect_t *) 253 RING_GET_REQUEST(&sc->xbd_ring, sc->xbd_ring.req_prod_pvt); 254 sc->xbd_ring.req_prod_pvt++; 255 ring_req->id = cm->cm_id; 256 ring_req->operation = BLKIF_OP_INDIRECT; 257 ring_req->indirect_op = cm->cm_operation; 258 ring_req->sector_number = cm->cm_sector_number; 259 ring_req->handle = (blkif_vdev_t)(uintptr_t)sc->xbd_disk; 260 ring_req->nr_segments = nsegs; 261 cm->cm_nseg = nsegs; 262 xbd_mksegarray(segs, nsegs, &cm->cm_gref_head, 263 xenbus_get_otherend_id(sc->xbd_dev), 264 cm->cm_operation == BLKIF_OP_WRITE, 265 cm->cm_sg_refs, cm->cm_indirectionpages); 266 memcpy(ring_req->indirect_grefs, &cm->cm_indirectionrefs, 267 sizeof(grant_ref_t) * sc->xbd_max_request_indirectpages); 268 } 269 270 if (cm->cm_operation == BLKIF_OP_READ) 271 op = BUS_DMASYNC_PREREAD; 272 else if (cm->cm_operation == BLKIF_OP_WRITE) 273 op = BUS_DMASYNC_PREWRITE; 274 else 275 op = 0; 276 bus_dmamap_sync(sc->xbd_io_dmat, cm->cm_map, op); 277 278 gnttab_free_grant_references(cm->cm_gref_head); 279 280 xbd_enqueue_cm(cm, XBD_Q_BUSY); 281 282 /* 283 * If bus dma had to asynchronously call us back to dispatch 284 * this command, we are no longer executing in the context of 285 * xbd_startio(). Thus we cannot rely on xbd_startio()'s call to 286 * xbd_flush_requests() to publish this command to the backend 287 * along with any other commands that it could batch. 288 */ 289 if ((cm->cm_flags & XBDCF_ASYNC_MAPPING) != 0) 290 xbd_flush_requests(sc); 291 292 return; 293 } 294 295 static int 296 xbd_queue_request(struct xbd_softc *sc, struct xbd_command *cm) 297 { 298 int error; 299 300 if (cm->cm_bp != NULL) 301 error = bus_dmamap_load_bio(sc->xbd_io_dmat, cm->cm_map, 302 cm->cm_bp, xbd_queue_cb, cm, 0); 303 else 304 error = bus_dmamap_load(sc->xbd_io_dmat, cm->cm_map, 305 cm->cm_data, cm->cm_datalen, xbd_queue_cb, cm, 0); 306 if (error == EINPROGRESS) { 307 /* 308 * Maintain queuing order by freezing the queue. The next 309 * command may not require as many resources as the command 310 * we just attempted to map, so we can't rely on bus dma 311 * blocking for it too. 312 */ 313 xbd_cm_freeze(sc, cm, XBDCF_ASYNC_MAPPING); 314 return (0); 315 } 316 317 return (error); 318 } 319 320 static void 321 xbd_restart_queue_callback(void *arg) 322 { 323 struct xbd_softc *sc = arg; 324 325 mtx_lock(&sc->xbd_io_lock); 326 327 xbd_thaw(sc, XBDF_GNT_SHORTAGE); 328 329 xbd_startio(sc); 330 331 mtx_unlock(&sc->xbd_io_lock); 332 } 333 334 static struct xbd_command * 335 xbd_bio_command(struct xbd_softc *sc) 336 { 337 struct xbd_command *cm; 338 struct bio *bp; 339 340 if (__predict_false(sc->xbd_state != XBD_STATE_CONNECTED)) 341 return (NULL); 342 343 bp = xbd_dequeue_bio(sc); 344 if (bp == NULL) 345 return (NULL); 346 347 if ((cm = xbd_dequeue_cm(sc, XBD_Q_FREE)) == NULL) { 348 xbd_freeze(sc, XBDF_CM_SHORTAGE); 349 xbd_requeue_bio(sc, bp); 350 return (NULL); 351 } 352 353 if (gnttab_alloc_grant_references(sc->xbd_max_request_segments, 354 &cm->cm_gref_head) != 0) { 355 gnttab_request_free_callback(&sc->xbd_callback, 356 xbd_restart_queue_callback, sc, 357 sc->xbd_max_request_segments); 358 xbd_freeze(sc, XBDF_GNT_SHORTAGE); 359 xbd_requeue_bio(sc, bp); 360 xbd_enqueue_cm(cm, XBD_Q_FREE); 361 return (NULL); 362 } 363 364 cm->cm_bp = bp; 365 cm->cm_sector_number = (blkif_sector_t)bp->bio_pblkno; 366 367 switch (bp->bio_cmd) { 368 case BIO_READ: 369 cm->cm_operation = BLKIF_OP_READ; 370 break; 371 case BIO_WRITE: 372 cm->cm_operation = BLKIF_OP_WRITE; 373 if ((bp->bio_flags & BIO_ORDERED) != 0) { 374 if ((sc->xbd_flags & XBDF_BARRIER) != 0) { 375 cm->cm_operation = BLKIF_OP_WRITE_BARRIER; 376 } else { 377 /* 378 * Single step this command. 379 */ 380 cm->cm_flags |= XBDCF_Q_FREEZE; 381 if (xbd_queue_length(sc, XBD_Q_BUSY) != 0) { 382 /* 383 * Wait for in-flight requests to 384 * finish. 385 */ 386 xbd_freeze(sc, XBDF_WAIT_IDLE); 387 xbd_requeue_cm(cm, XBD_Q_READY); 388 return (NULL); 389 } 390 } 391 } 392 break; 393 case BIO_FLUSH: 394 if ((sc->xbd_flags & XBDF_FLUSH) != 0) 395 cm->cm_operation = BLKIF_OP_FLUSH_DISKCACHE; 396 else if ((sc->xbd_flags & XBDF_BARRIER) != 0) 397 cm->cm_operation = BLKIF_OP_WRITE_BARRIER; 398 else 399 panic("flush request, but no flush support available"); 400 break; 401 default: 402 biofinish(bp, NULL, EOPNOTSUPP); 403 xbd_enqueue_cm(cm, XBD_Q_FREE); 404 return (NULL); 405 } 406 407 return (cm); 408 } 409 410 /* 411 * Dequeue buffers and place them in the shared communication ring. 412 * Return when no more requests can be accepted or all buffers have 413 * been queued. 414 * 415 * Signal XEN once the ring has been filled out. 416 */ 417 static void 418 xbd_startio(struct xbd_softc *sc) 419 { 420 struct xbd_command *cm; 421 int error, queued = 0; 422 423 mtx_assert(&sc->xbd_io_lock, MA_OWNED); 424 425 if (sc->xbd_state != XBD_STATE_CONNECTED) 426 return; 427 428 while (!RING_FULL(&sc->xbd_ring)) { 429 if (sc->xbd_qfrozen_cnt != 0) 430 break; 431 432 cm = xbd_dequeue_cm(sc, XBD_Q_READY); 433 434 if (cm == NULL) 435 cm = xbd_bio_command(sc); 436 437 if (cm == NULL) 438 break; 439 440 if ((cm->cm_flags & XBDCF_Q_FREEZE) != 0) { 441 /* 442 * Single step command. Future work is 443 * held off until this command completes. 444 */ 445 xbd_cm_freeze(sc, cm, XBDCF_Q_FREEZE); 446 } 447 448 if ((error = xbd_queue_request(sc, cm)) != 0) { 449 printf("xbd_queue_request returned %d\n", error); 450 break; 451 } 452 queued++; 453 } 454 455 if (queued != 0) 456 xbd_flush_requests(sc); 457 } 458 459 static void 460 xbd_bio_complete(struct xbd_softc *sc, struct xbd_command *cm) 461 { 462 struct bio *bp; 463 464 bp = cm->cm_bp; 465 466 if (__predict_false(cm->cm_status != BLKIF_RSP_OKAY)) { 467 disk_err(bp, "disk error" , -1, 0); 468 printf(" status: %x\n", cm->cm_status); 469 bp->bio_flags |= BIO_ERROR; 470 } 471 472 if (bp->bio_flags & BIO_ERROR) 473 bp->bio_error = EIO; 474 else 475 bp->bio_resid = 0; 476 477 xbd_free_command(cm); 478 biodone(bp); 479 } 480 481 static void 482 xbd_int(void *xsc) 483 { 484 struct xbd_softc *sc = xsc; 485 struct xbd_command *cm; 486 blkif_response_t *bret; 487 RING_IDX i, rp; 488 int op; 489 490 mtx_lock(&sc->xbd_io_lock); 491 492 if (__predict_false(sc->xbd_state == XBD_STATE_DISCONNECTED)) { 493 mtx_unlock(&sc->xbd_io_lock); 494 return; 495 } 496 497 again: 498 rp = sc->xbd_ring.sring->rsp_prod; 499 rmb(); /* Ensure we see queued responses up to 'rp'. */ 500 501 for (i = sc->xbd_ring.rsp_cons; i != rp;) { 502 bret = RING_GET_RESPONSE(&sc->xbd_ring, i); 503 cm = &sc->xbd_shadow[bret->id]; 504 505 xbd_remove_cm(cm, XBD_Q_BUSY); 506 gnttab_end_foreign_access_references(cm->cm_nseg, 507 cm->cm_sg_refs); 508 i++; 509 510 if (cm->cm_operation == BLKIF_OP_READ) 511 op = BUS_DMASYNC_POSTREAD; 512 else if (cm->cm_operation == BLKIF_OP_WRITE || 513 cm->cm_operation == BLKIF_OP_WRITE_BARRIER) 514 op = BUS_DMASYNC_POSTWRITE; 515 else 516 op = 0; 517 bus_dmamap_sync(sc->xbd_io_dmat, cm->cm_map, op); 518 bus_dmamap_unload(sc->xbd_io_dmat, cm->cm_map); 519 520 /* 521 * Release any hold this command has on future command 522 * dispatch. 523 */ 524 xbd_cm_thaw(sc, cm); 525 526 /* 527 * Directly call the i/o complete routine to save an 528 * an indirection in the common case. 529 */ 530 cm->cm_status = bret->status; 531 if (cm->cm_bp) 532 xbd_bio_complete(sc, cm); 533 else if (cm->cm_complete != NULL) 534 cm->cm_complete(cm); 535 else 536 xbd_free_command(cm); 537 } 538 539 sc->xbd_ring.rsp_cons = i; 540 541 if (i != sc->xbd_ring.req_prod_pvt) { 542 int more_to_do; 543 RING_FINAL_CHECK_FOR_RESPONSES(&sc->xbd_ring, more_to_do); 544 if (more_to_do) 545 goto again; 546 } else { 547 sc->xbd_ring.sring->rsp_event = i + 1; 548 } 549 550 if (xbd_queue_length(sc, XBD_Q_BUSY) == 0) 551 xbd_thaw(sc, XBDF_WAIT_IDLE); 552 553 xbd_startio(sc); 554 555 if (__predict_false(sc->xbd_state == XBD_STATE_SUSPENDED)) 556 wakeup(&sc->xbd_cm_q[XBD_Q_BUSY]); 557 558 mtx_unlock(&sc->xbd_io_lock); 559 } 560 561 /*------------------------------- Dump Support -------------------------------*/ 562 /** 563 * Quiesce the disk writes for a dump file before allowing the next buffer. 564 */ 565 static void 566 xbd_quiesce(struct xbd_softc *sc) 567 { 568 int mtd; 569 570 // While there are outstanding requests 571 while (xbd_queue_length(sc, XBD_Q_BUSY) != 0) { 572 RING_FINAL_CHECK_FOR_RESPONSES(&sc->xbd_ring, mtd); 573 if (mtd) { 574 /* Received request completions, update queue. */ 575 xbd_int(sc); 576 } 577 if (xbd_queue_length(sc, XBD_Q_BUSY) != 0) { 578 /* 579 * Still pending requests, wait for the disk i/o 580 * to complete. 581 */ 582 HYPERVISOR_yield(); 583 } 584 } 585 } 586 587 /* Kernel dump function for a paravirtualized disk device */ 588 static void 589 xbd_dump_complete(struct xbd_command *cm) 590 { 591 592 xbd_enqueue_cm(cm, XBD_Q_COMPLETE); 593 } 594 595 static int 596 xbd_dump(void *arg, void *virtual, off_t offset, size_t length) 597 { 598 struct disk *dp = arg; 599 struct xbd_softc *sc = dp->d_drv1; 600 struct xbd_command *cm; 601 size_t chunk; 602 int rc = 0; 603 604 if (length == 0) 605 return (0); 606 607 xbd_quiesce(sc); /* All quiet on the western front. */ 608 609 /* 610 * If this lock is held, then this module is failing, and a 611 * successful kernel dump is highly unlikely anyway. 612 */ 613 mtx_lock(&sc->xbd_io_lock); 614 615 /* Split the 64KB block as needed */ 616 while (length > 0) { 617 cm = xbd_dequeue_cm(sc, XBD_Q_FREE); 618 if (cm == NULL) { 619 mtx_unlock(&sc->xbd_io_lock); 620 device_printf(sc->xbd_dev, "dump: no more commands?\n"); 621 return (EBUSY); 622 } 623 624 if (gnttab_alloc_grant_references(sc->xbd_max_request_segments, 625 &cm->cm_gref_head) != 0) { 626 xbd_free_command(cm); 627 mtx_unlock(&sc->xbd_io_lock); 628 device_printf(sc->xbd_dev, "no more grant allocs?\n"); 629 return (EBUSY); 630 } 631 632 chunk = length > sc->xbd_max_request_size ? 633 sc->xbd_max_request_size : length; 634 cm->cm_data = virtual; 635 cm->cm_datalen = chunk; 636 cm->cm_operation = BLKIF_OP_WRITE; 637 cm->cm_sector_number = offset / dp->d_sectorsize; 638 cm->cm_complete = xbd_dump_complete; 639 640 xbd_enqueue_cm(cm, XBD_Q_READY); 641 642 length -= chunk; 643 offset += chunk; 644 virtual = (char *) virtual + chunk; 645 } 646 647 /* Tell DOM0 to do the I/O */ 648 xbd_startio(sc); 649 mtx_unlock(&sc->xbd_io_lock); 650 651 /* Poll for the completion. */ 652 xbd_quiesce(sc); /* All quite on the eastern front */ 653 654 /* If there were any errors, bail out... */ 655 while ((cm = xbd_dequeue_cm(sc, XBD_Q_COMPLETE)) != NULL) { 656 if (cm->cm_status != BLKIF_RSP_OKAY) { 657 device_printf(sc->xbd_dev, 658 "Dump I/O failed at sector %jd\n", 659 cm->cm_sector_number); 660 rc = EIO; 661 } 662 xbd_free_command(cm); 663 } 664 665 return (rc); 666 } 667 668 /*----------------------------- Disk Entrypoints -----------------------------*/ 669 static int 670 xbd_open(struct disk *dp) 671 { 672 struct xbd_softc *sc = dp->d_drv1; 673 674 if (sc == NULL) { 675 printf("xbd%d: not found", dp->d_unit); 676 return (ENXIO); 677 } 678 679 sc->xbd_flags |= XBDF_OPEN; 680 sc->xbd_users++; 681 return (0); 682 } 683 684 static int 685 xbd_close(struct disk *dp) 686 { 687 struct xbd_softc *sc = dp->d_drv1; 688 689 if (sc == NULL) 690 return (ENXIO); 691 sc->xbd_flags &= ~XBDF_OPEN; 692 if (--(sc->xbd_users) == 0) { 693 /* 694 * Check whether we have been instructed to close. We will 695 * have ignored this request initially, as the device was 696 * still mounted. 697 */ 698 if (xenbus_get_otherend_state(sc->xbd_dev) == 699 XenbusStateClosing) 700 xbd_closing(sc->xbd_dev); 701 } 702 return (0); 703 } 704 705 static int 706 xbd_ioctl(struct disk *dp, u_long cmd, void *addr, int flag, struct thread *td) 707 { 708 struct xbd_softc *sc = dp->d_drv1; 709 710 if (sc == NULL) 711 return (ENXIO); 712 713 return (ENOTTY); 714 } 715 716 /* 717 * Read/write routine for a buffer. Finds the proper unit, place it on 718 * the sortq and kick the controller. 719 */ 720 static void 721 xbd_strategy(struct bio *bp) 722 { 723 struct xbd_softc *sc = bp->bio_disk->d_drv1; 724 725 /* bogus disk? */ 726 if (sc == NULL) { 727 bp->bio_error = EINVAL; 728 bp->bio_flags |= BIO_ERROR; 729 bp->bio_resid = bp->bio_bcount; 730 biodone(bp); 731 return; 732 } 733 734 /* 735 * Place it in the queue of disk activities for this disk 736 */ 737 mtx_lock(&sc->xbd_io_lock); 738 739 xbd_enqueue_bio(sc, bp); 740 xbd_startio(sc); 741 742 mtx_unlock(&sc->xbd_io_lock); 743 return; 744 } 745 746 /*------------------------------ Ring Management -----------------------------*/ 747 static int 748 xbd_alloc_ring(struct xbd_softc *sc) 749 { 750 blkif_sring_t *sring; 751 uintptr_t sring_page_addr; 752 int error; 753 int i; 754 755 sring = malloc(sc->xbd_ring_pages * PAGE_SIZE, M_XENBLOCKFRONT, 756 M_NOWAIT|M_ZERO); 757 if (sring == NULL) { 758 xenbus_dev_fatal(sc->xbd_dev, ENOMEM, "allocating shared ring"); 759 return (ENOMEM); 760 } 761 SHARED_RING_INIT(sring); 762 FRONT_RING_INIT(&sc->xbd_ring, sring, sc->xbd_ring_pages * PAGE_SIZE); 763 764 for (i = 0, sring_page_addr = (uintptr_t)sring; 765 i < sc->xbd_ring_pages; 766 i++, sring_page_addr += PAGE_SIZE) { 767 error = xenbus_grant_ring(sc->xbd_dev, 768 (vtophys(sring_page_addr) >> PAGE_SHIFT), 769 &sc->xbd_ring_ref[i]); 770 if (error) { 771 xenbus_dev_fatal(sc->xbd_dev, error, 772 "granting ring_ref(%d)", i); 773 return (error); 774 } 775 } 776 if (sc->xbd_ring_pages == 1) { 777 error = xs_printf(XST_NIL, xenbus_get_node(sc->xbd_dev), 778 "ring-ref", "%u", sc->xbd_ring_ref[0]); 779 if (error) { 780 xenbus_dev_fatal(sc->xbd_dev, error, 781 "writing %s/ring-ref", 782 xenbus_get_node(sc->xbd_dev)); 783 return (error); 784 } 785 } else { 786 for (i = 0; i < sc->xbd_ring_pages; i++) { 787 char ring_ref_name[]= "ring_refXX"; 788 789 snprintf(ring_ref_name, sizeof(ring_ref_name), 790 "ring-ref%u", i); 791 error = xs_printf(XST_NIL, xenbus_get_node(sc->xbd_dev), 792 ring_ref_name, "%u", sc->xbd_ring_ref[i]); 793 if (error) { 794 xenbus_dev_fatal(sc->xbd_dev, error, 795 "writing %s/%s", 796 xenbus_get_node(sc->xbd_dev), 797 ring_ref_name); 798 return (error); 799 } 800 } 801 } 802 803 error = xen_intr_alloc_and_bind_local_port(sc->xbd_dev, 804 xenbus_get_otherend_id(sc->xbd_dev), NULL, xbd_int, sc, 805 INTR_TYPE_BIO | INTR_MPSAFE, &sc->xen_intr_handle); 806 if (error) { 807 xenbus_dev_fatal(sc->xbd_dev, error, 808 "xen_intr_alloc_and_bind_local_port failed"); 809 return (error); 810 } 811 812 return (0); 813 } 814 815 static void 816 xbd_free_ring(struct xbd_softc *sc) 817 { 818 int i; 819 820 if (sc->xbd_ring.sring == NULL) 821 return; 822 823 for (i = 0; i < sc->xbd_ring_pages; i++) { 824 if (sc->xbd_ring_ref[i] != GRANT_REF_INVALID) { 825 gnttab_end_foreign_access_ref(sc->xbd_ring_ref[i]); 826 sc->xbd_ring_ref[i] = GRANT_REF_INVALID; 827 } 828 } 829 free(sc->xbd_ring.sring, M_XENBLOCKFRONT); 830 sc->xbd_ring.sring = NULL; 831 } 832 833 /*-------------------------- Initialization/Teardown -------------------------*/ 834 static int 835 xbd_feature_string(struct xbd_softc *sc, char *features, size_t len) 836 { 837 struct sbuf sb; 838 int feature_cnt; 839 840 sbuf_new(&sb, features, len, SBUF_FIXEDLEN); 841 842 feature_cnt = 0; 843 if ((sc->xbd_flags & XBDF_FLUSH) != 0) { 844 sbuf_printf(&sb, "flush"); 845 feature_cnt++; 846 } 847 848 if ((sc->xbd_flags & XBDF_BARRIER) != 0) { 849 if (feature_cnt != 0) 850 sbuf_printf(&sb, ", "); 851 sbuf_printf(&sb, "write_barrier"); 852 feature_cnt++; 853 } 854 855 if ((sc->xbd_flags & XBDF_DISCARD) != 0) { 856 if (feature_cnt != 0) 857 sbuf_printf(&sb, ", "); 858 sbuf_printf(&sb, "discard"); 859 feature_cnt++; 860 } 861 862 if ((sc->xbd_flags & XBDF_PERSISTENT) != 0) { 863 if (feature_cnt != 0) 864 sbuf_printf(&sb, ", "); 865 sbuf_printf(&sb, "persistent_grants"); 866 feature_cnt++; 867 } 868 869 (void) sbuf_finish(&sb); 870 return (sbuf_len(&sb)); 871 } 872 873 static int 874 xbd_sysctl_features(SYSCTL_HANDLER_ARGS) 875 { 876 char features[80]; 877 struct xbd_softc *sc = arg1; 878 int error; 879 int len; 880 881 error = sysctl_wire_old_buffer(req, 0); 882 if (error != 0) 883 return (error); 884 885 len = xbd_feature_string(sc, features, sizeof(features)); 886 887 /* len is -1 on error, which will make the SYSCTL_OUT a no-op. */ 888 return (SYSCTL_OUT(req, features, len + 1/*NUL*/)); 889 } 890 891 static void 892 xbd_setup_sysctl(struct xbd_softc *xbd) 893 { 894 struct sysctl_ctx_list *sysctl_ctx = NULL; 895 struct sysctl_oid *sysctl_tree = NULL; 896 struct sysctl_oid_list *children; 897 898 sysctl_ctx = device_get_sysctl_ctx(xbd->xbd_dev); 899 if (sysctl_ctx == NULL) 900 return; 901 902 sysctl_tree = device_get_sysctl_tree(xbd->xbd_dev); 903 if (sysctl_tree == NULL) 904 return; 905 906 children = SYSCTL_CHILDREN(sysctl_tree); 907 SYSCTL_ADD_UINT(sysctl_ctx, children, OID_AUTO, 908 "max_requests", CTLFLAG_RD, &xbd->xbd_max_requests, -1, 909 "maximum outstanding requests (negotiated)"); 910 911 SYSCTL_ADD_UINT(sysctl_ctx, children, OID_AUTO, 912 "max_request_segments", CTLFLAG_RD, 913 &xbd->xbd_max_request_segments, 0, 914 "maximum number of pages per requests (negotiated)"); 915 916 SYSCTL_ADD_UINT(sysctl_ctx, children, OID_AUTO, 917 "max_request_size", CTLFLAG_RD, &xbd->xbd_max_request_size, 0, 918 "maximum size in bytes of a request (negotiated)"); 919 920 SYSCTL_ADD_UINT(sysctl_ctx, children, OID_AUTO, 921 "ring_pages", CTLFLAG_RD, &xbd->xbd_ring_pages, 0, 922 "communication channel pages (negotiated)"); 923 924 SYSCTL_ADD_PROC(sysctl_ctx, children, OID_AUTO, 925 "features", CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, xbd, 926 0, xbd_sysctl_features, "A", "protocol features (negotiated)"); 927 } 928 929 /* 930 * Translate Linux major/minor to an appropriate name and unit 931 * number. For HVM guests, this allows us to use the same drive names 932 * with blkfront as the emulated drives, easing transition slightly. 933 */ 934 static void 935 xbd_vdevice_to_unit(uint32_t vdevice, int *unit, const char **name) 936 { 937 static struct vdev_info { 938 int major; 939 int shift; 940 int base; 941 const char *name; 942 } info[] = { 943 {3, 6, 0, "ada"}, /* ide0 */ 944 {22, 6, 2, "ada"}, /* ide1 */ 945 {33, 6, 4, "ada"}, /* ide2 */ 946 {34, 6, 6, "ada"}, /* ide3 */ 947 {56, 6, 8, "ada"}, /* ide4 */ 948 {57, 6, 10, "ada"}, /* ide5 */ 949 {88, 6, 12, "ada"}, /* ide6 */ 950 {89, 6, 14, "ada"}, /* ide7 */ 951 {90, 6, 16, "ada"}, /* ide8 */ 952 {91, 6, 18, "ada"}, /* ide9 */ 953 954 {8, 4, 0, "da"}, /* scsi disk0 */ 955 {65, 4, 16, "da"}, /* scsi disk1 */ 956 {66, 4, 32, "da"}, /* scsi disk2 */ 957 {67, 4, 48, "da"}, /* scsi disk3 */ 958 {68, 4, 64, "da"}, /* scsi disk4 */ 959 {69, 4, 80, "da"}, /* scsi disk5 */ 960 {70, 4, 96, "da"}, /* scsi disk6 */ 961 {71, 4, 112, "da"}, /* scsi disk7 */ 962 {128, 4, 128, "da"}, /* scsi disk8 */ 963 {129, 4, 144, "da"}, /* scsi disk9 */ 964 {130, 4, 160, "da"}, /* scsi disk10 */ 965 {131, 4, 176, "da"}, /* scsi disk11 */ 966 {132, 4, 192, "da"}, /* scsi disk12 */ 967 {133, 4, 208, "da"}, /* scsi disk13 */ 968 {134, 4, 224, "da"}, /* scsi disk14 */ 969 {135, 4, 240, "da"}, /* scsi disk15 */ 970 971 {202, 4, 0, "xbd"}, /* xbd */ 972 973 {0, 0, 0, NULL}, 974 }; 975 int major = vdevice >> 8; 976 int minor = vdevice & 0xff; 977 int i; 978 979 if (vdevice & (1 << 28)) { 980 *unit = (vdevice & ((1 << 28) - 1)) >> 8; 981 *name = "xbd"; 982 return; 983 } 984 985 for (i = 0; info[i].major; i++) { 986 if (info[i].major == major) { 987 *unit = info[i].base + (minor >> info[i].shift); 988 *name = info[i].name; 989 return; 990 } 991 } 992 993 *unit = minor >> 4; 994 *name = "xbd"; 995 } 996 997 int 998 xbd_instance_create(struct xbd_softc *sc, blkif_sector_t sectors, 999 int vdevice, uint16_t vdisk_info, unsigned long sector_size, 1000 unsigned long phys_sector_size) 1001 { 1002 char features[80]; 1003 int unit, error = 0; 1004 const char *name; 1005 1006 xbd_vdevice_to_unit(vdevice, &unit, &name); 1007 1008 sc->xbd_unit = unit; 1009 1010 if (strcmp(name, "xbd") != 0) 1011 device_printf(sc->xbd_dev, "attaching as %s%d\n", name, unit); 1012 1013 if (xbd_feature_string(sc, features, sizeof(features)) > 0) { 1014 device_printf(sc->xbd_dev, "features: %s\n", 1015 features); 1016 } 1017 1018 sc->xbd_disk = disk_alloc(); 1019 sc->xbd_disk->d_unit = sc->xbd_unit; 1020 sc->xbd_disk->d_open = xbd_open; 1021 sc->xbd_disk->d_close = xbd_close; 1022 sc->xbd_disk->d_ioctl = xbd_ioctl; 1023 sc->xbd_disk->d_strategy = xbd_strategy; 1024 sc->xbd_disk->d_dump = xbd_dump; 1025 sc->xbd_disk->d_name = name; 1026 sc->xbd_disk->d_drv1 = sc; 1027 sc->xbd_disk->d_sectorsize = sector_size; 1028 sc->xbd_disk->d_stripesize = phys_sector_size; 1029 sc->xbd_disk->d_stripeoffset = 0; 1030 1031 sc->xbd_disk->d_mediasize = sectors * sector_size; 1032 sc->xbd_disk->d_maxsize = sc->xbd_max_request_size; 1033 sc->xbd_disk->d_flags = DISKFLAG_UNMAPPED_BIO; 1034 if ((sc->xbd_flags & (XBDF_FLUSH|XBDF_BARRIER)) != 0) { 1035 sc->xbd_disk->d_flags |= DISKFLAG_CANFLUSHCACHE; 1036 device_printf(sc->xbd_dev, 1037 "synchronize cache commands enabled.\n"); 1038 } 1039 disk_create(sc->xbd_disk, DISK_VERSION); 1040 1041 return error; 1042 } 1043 1044 static void 1045 xbd_free(struct xbd_softc *sc) 1046 { 1047 int i; 1048 1049 /* Prevent new requests being issued until we fix things up. */ 1050 mtx_lock(&sc->xbd_io_lock); 1051 sc->xbd_state = XBD_STATE_DISCONNECTED; 1052 mtx_unlock(&sc->xbd_io_lock); 1053 1054 /* Free resources associated with old device channel. */ 1055 xbd_free_ring(sc); 1056 if (sc->xbd_shadow) { 1057 for (i = 0; i < sc->xbd_max_requests; i++) { 1058 struct xbd_command *cm; 1059 1060 cm = &sc->xbd_shadow[i]; 1061 if (cm->cm_sg_refs != NULL) { 1062 free(cm->cm_sg_refs, M_XENBLOCKFRONT); 1063 cm->cm_sg_refs = NULL; 1064 } 1065 1066 if (cm->cm_indirectionpages != NULL) { 1067 gnttab_end_foreign_access_references( 1068 sc->xbd_max_request_indirectpages, 1069 &cm->cm_indirectionrefs[0]); 1070 contigfree(cm->cm_indirectionpages, PAGE_SIZE * 1071 sc->xbd_max_request_indirectpages, 1072 M_XENBLOCKFRONT); 1073 cm->cm_indirectionpages = NULL; 1074 } 1075 1076 bus_dmamap_destroy(sc->xbd_io_dmat, cm->cm_map); 1077 } 1078 free(sc->xbd_shadow, M_XENBLOCKFRONT); 1079 sc->xbd_shadow = NULL; 1080 1081 bus_dma_tag_destroy(sc->xbd_io_dmat); 1082 1083 xbd_initq_cm(sc, XBD_Q_FREE); 1084 xbd_initq_cm(sc, XBD_Q_READY); 1085 xbd_initq_cm(sc, XBD_Q_COMPLETE); 1086 } 1087 1088 xen_intr_unbind(&sc->xen_intr_handle); 1089 1090 } 1091 1092 /*--------------------------- State Change Handlers --------------------------*/ 1093 static void 1094 xbd_initialize(struct xbd_softc *sc) 1095 { 1096 const char *otherend_path; 1097 const char *node_path; 1098 uint32_t max_ring_page_order; 1099 int error; 1100 1101 if (xenbus_get_state(sc->xbd_dev) != XenbusStateInitialising) { 1102 /* Initialization has already been performed. */ 1103 return; 1104 } 1105 1106 /* 1107 * Protocol defaults valid even if negotiation for a 1108 * setting fails. 1109 */ 1110 max_ring_page_order = 0; 1111 sc->xbd_ring_pages = 1; 1112 1113 /* 1114 * Protocol negotiation. 1115 * 1116 * \note xs_gather() returns on the first encountered error, so 1117 * we must use independent calls in order to guarantee 1118 * we don't miss information in a sparsly populated back-end 1119 * tree. 1120 * 1121 * \note xs_scanf() does not update variables for unmatched 1122 * fields. 1123 */ 1124 otherend_path = xenbus_get_otherend_path(sc->xbd_dev); 1125 node_path = xenbus_get_node(sc->xbd_dev); 1126 1127 /* Support both backend schemes for relaying ring page limits. */ 1128 (void)xs_scanf(XST_NIL, otherend_path, 1129 "max-ring-page-order", NULL, "%" PRIu32, 1130 &max_ring_page_order); 1131 sc->xbd_ring_pages = 1 << max_ring_page_order; 1132 (void)xs_scanf(XST_NIL, otherend_path, 1133 "max-ring-pages", NULL, "%" PRIu32, 1134 &sc->xbd_ring_pages); 1135 if (sc->xbd_ring_pages < 1) 1136 sc->xbd_ring_pages = 1; 1137 1138 if (sc->xbd_ring_pages > XBD_MAX_RING_PAGES) { 1139 device_printf(sc->xbd_dev, 1140 "Back-end specified ring-pages of %u " 1141 "limited to front-end limit of %u.\n", 1142 sc->xbd_ring_pages, XBD_MAX_RING_PAGES); 1143 sc->xbd_ring_pages = XBD_MAX_RING_PAGES; 1144 } 1145 1146 if (powerof2(sc->xbd_ring_pages) == 0) { 1147 uint32_t new_page_limit; 1148 1149 new_page_limit = 0x01 << (fls(sc->xbd_ring_pages) - 1); 1150 device_printf(sc->xbd_dev, 1151 "Back-end specified ring-pages of %u " 1152 "is not a power of 2. Limited to %u.\n", 1153 sc->xbd_ring_pages, new_page_limit); 1154 sc->xbd_ring_pages = new_page_limit; 1155 } 1156 1157 sc->xbd_max_requests = 1158 BLKIF_MAX_RING_REQUESTS(sc->xbd_ring_pages * PAGE_SIZE); 1159 if (sc->xbd_max_requests > XBD_MAX_REQUESTS) { 1160 device_printf(sc->xbd_dev, 1161 "Back-end specified max_requests of %u " 1162 "limited to front-end limit of %zu.\n", 1163 sc->xbd_max_requests, XBD_MAX_REQUESTS); 1164 sc->xbd_max_requests = XBD_MAX_REQUESTS; 1165 } 1166 1167 if (xbd_alloc_ring(sc) != 0) 1168 return; 1169 1170 /* Support both backend schemes for relaying ring page limits. */ 1171 if (sc->xbd_ring_pages > 1) { 1172 error = xs_printf(XST_NIL, node_path, 1173 "num-ring-pages","%u", 1174 sc->xbd_ring_pages); 1175 if (error) { 1176 xenbus_dev_fatal(sc->xbd_dev, error, 1177 "writing %s/num-ring-pages", 1178 node_path); 1179 return; 1180 } 1181 1182 error = xs_printf(XST_NIL, node_path, 1183 "ring-page-order", "%u", 1184 fls(sc->xbd_ring_pages) - 1); 1185 if (error) { 1186 xenbus_dev_fatal(sc->xbd_dev, error, 1187 "writing %s/ring-page-order", 1188 node_path); 1189 return; 1190 } 1191 } 1192 1193 error = xs_printf(XST_NIL, node_path, "event-channel", 1194 "%u", xen_intr_port(sc->xen_intr_handle)); 1195 if (error) { 1196 xenbus_dev_fatal(sc->xbd_dev, error, 1197 "writing %s/event-channel", 1198 node_path); 1199 return; 1200 } 1201 1202 error = xs_printf(XST_NIL, node_path, "protocol", 1203 "%s", XEN_IO_PROTO_ABI_NATIVE); 1204 if (error) { 1205 xenbus_dev_fatal(sc->xbd_dev, error, 1206 "writing %s/protocol", 1207 node_path); 1208 return; 1209 } 1210 1211 xenbus_set_state(sc->xbd_dev, XenbusStateInitialised); 1212 } 1213 1214 /* 1215 * Invoked when the backend is finally 'ready' (and has published 1216 * the details about the physical device - #sectors, size, etc). 1217 */ 1218 static void 1219 xbd_connect(struct xbd_softc *sc) 1220 { 1221 device_t dev = sc->xbd_dev; 1222 blkif_sector_t sectors; 1223 unsigned long sector_size, phys_sector_size; 1224 unsigned int binfo; 1225 int err, feature_barrier, feature_flush; 1226 int i, j; 1227 1228 DPRINTK("blkfront.c:connect:%s.\n", xenbus_get_otherend_path(dev)); 1229 1230 if (sc->xbd_state == XBD_STATE_SUSPENDED) { 1231 return; 1232 } 1233 1234 if (sc->xbd_state == XBD_STATE_CONNECTED) { 1235 struct disk *disk; 1236 1237 disk = sc->xbd_disk; 1238 if (disk == NULL) { 1239 return; 1240 } 1241 err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev), 1242 "sectors", "%"PRIu64, §ors, NULL); 1243 if (err != 0) { 1244 xenbus_dev_error(dev, err, 1245 "reading sectors at %s", 1246 xenbus_get_otherend_path(dev)); 1247 return; 1248 } 1249 disk->d_mediasize = disk->d_sectorsize * sectors; 1250 err = disk_resize(disk, M_NOWAIT); 1251 if (err) { 1252 xenbus_dev_error(dev, err, 1253 "unable to resize disk %s%u", 1254 disk->d_name, disk->d_unit); 1255 return; 1256 } 1257 device_printf(sc->xbd_dev, 1258 "changed capacity to %jd\n", 1259 (intmax_t)disk->d_mediasize); 1260 return; 1261 } 1262 1263 err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev), 1264 "sectors", "%"PRIu64, §ors, 1265 "info", "%u", &binfo, 1266 "sector-size", "%lu", §or_size, 1267 NULL); 1268 if (err) { 1269 xenbus_dev_fatal(dev, err, 1270 "reading backend fields at %s", 1271 xenbus_get_otherend_path(dev)); 1272 return; 1273 } 1274 if ((sectors == 0) || (sector_size == 0)) { 1275 xenbus_dev_fatal(dev, 0, 1276 "invalid parameters from %s:" 1277 " sectors = %"PRIu64", sector_size = %lu", 1278 xenbus_get_otherend_path(dev), 1279 sectors, sector_size); 1280 return; 1281 } 1282 err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev), 1283 "physical-sector-size", "%lu", &phys_sector_size, 1284 NULL); 1285 if (err || phys_sector_size <= sector_size) 1286 phys_sector_size = 0; 1287 err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev), 1288 "feature-barrier", "%d", &feature_barrier, 1289 NULL); 1290 if (err == 0 && feature_barrier != 0) 1291 sc->xbd_flags |= XBDF_BARRIER; 1292 1293 err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev), 1294 "feature-flush-cache", "%d", &feature_flush, 1295 NULL); 1296 if (err == 0 && feature_flush != 0) 1297 sc->xbd_flags |= XBDF_FLUSH; 1298 1299 err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev), 1300 "feature-max-indirect-segments", "%" PRIu32, 1301 &sc->xbd_max_request_segments, NULL); 1302 if ((err != 0) || (xbd_enable_indirect == 0)) 1303 sc->xbd_max_request_segments = 0; 1304 if (sc->xbd_max_request_segments > XBD_MAX_INDIRECT_SEGMENTS) 1305 sc->xbd_max_request_segments = XBD_MAX_INDIRECT_SEGMENTS; 1306 if (sc->xbd_max_request_segments > XBD_SIZE_TO_SEGS(maxphys)) 1307 sc->xbd_max_request_segments = XBD_SIZE_TO_SEGS(maxphys); 1308 sc->xbd_max_request_indirectpages = 1309 XBD_INDIRECT_SEGS_TO_PAGES(sc->xbd_max_request_segments); 1310 if (sc->xbd_max_request_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST) 1311 sc->xbd_max_request_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST; 1312 sc->xbd_max_request_size = 1313 XBD_SEGS_TO_SIZE(sc->xbd_max_request_segments); 1314 1315 /* Allocate datastructures based on negotiated values. */ 1316 err = bus_dma_tag_create( 1317 bus_get_dma_tag(sc->xbd_dev), /* parent */ 1318 512, PAGE_SIZE, /* algnmnt, boundary */ 1319 BUS_SPACE_MAXADDR, /* lowaddr */ 1320 BUS_SPACE_MAXADDR, /* highaddr */ 1321 NULL, NULL, /* filter, filterarg */ 1322 sc->xbd_max_request_size, 1323 sc->xbd_max_request_segments, 1324 PAGE_SIZE, /* maxsegsize */ 1325 BUS_DMA_ALLOCNOW, /* flags */ 1326 busdma_lock_mutex, /* lockfunc */ 1327 &sc->xbd_io_lock, /* lockarg */ 1328 &sc->xbd_io_dmat); 1329 if (err != 0) { 1330 xenbus_dev_fatal(sc->xbd_dev, err, 1331 "Cannot allocate parent DMA tag\n"); 1332 return; 1333 } 1334 1335 /* Per-transaction data allocation. */ 1336 sc->xbd_shadow = malloc(sizeof(*sc->xbd_shadow) * sc->xbd_max_requests, 1337 M_XENBLOCKFRONT, M_NOWAIT|M_ZERO); 1338 if (sc->xbd_shadow == NULL) { 1339 bus_dma_tag_destroy(sc->xbd_io_dmat); 1340 xenbus_dev_fatal(sc->xbd_dev, ENOMEM, 1341 "Cannot allocate request structures\n"); 1342 return; 1343 } 1344 1345 for (i = 0; i < sc->xbd_max_requests; i++) { 1346 struct xbd_command *cm; 1347 void * indirectpages; 1348 1349 cm = &sc->xbd_shadow[i]; 1350 cm->cm_sg_refs = malloc( 1351 sizeof(grant_ref_t) * sc->xbd_max_request_segments, 1352 M_XENBLOCKFRONT, M_NOWAIT); 1353 if (cm->cm_sg_refs == NULL) 1354 break; 1355 cm->cm_id = i; 1356 cm->cm_flags = XBDCF_INITIALIZER; 1357 cm->cm_sc = sc; 1358 if (bus_dmamap_create(sc->xbd_io_dmat, 0, &cm->cm_map) != 0) 1359 break; 1360 if (sc->xbd_max_request_indirectpages > 0) { 1361 indirectpages = contigmalloc( 1362 PAGE_SIZE * sc->xbd_max_request_indirectpages, 1363 M_XENBLOCKFRONT, M_ZERO | M_NOWAIT, 0, ~0, 1364 PAGE_SIZE, 0); 1365 if (indirectpages == NULL) 1366 sc->xbd_max_request_indirectpages = 0; 1367 } else { 1368 indirectpages = NULL; 1369 } 1370 for (j = 0; j < sc->xbd_max_request_indirectpages; j++) { 1371 if (gnttab_grant_foreign_access( 1372 xenbus_get_otherend_id(sc->xbd_dev), 1373 (vtophys(indirectpages) >> PAGE_SHIFT) + j, 1374 1 /* grant read-only access */, 1375 &cm->cm_indirectionrefs[j])) 1376 break; 1377 } 1378 if (j < sc->xbd_max_request_indirectpages) { 1379 contigfree(indirectpages, 1380 PAGE_SIZE * sc->xbd_max_request_indirectpages, 1381 M_XENBLOCKFRONT); 1382 break; 1383 } 1384 cm->cm_indirectionpages = indirectpages; 1385 xbd_free_command(cm); 1386 } 1387 1388 if (sc->xbd_disk == NULL) { 1389 device_printf(dev, "%juMB <%s> at %s", 1390 (uintmax_t) sectors / (1048576 / sector_size), 1391 device_get_desc(dev), 1392 xenbus_get_node(dev)); 1393 bus_print_child_footer(device_get_parent(dev), dev); 1394 1395 xbd_instance_create(sc, sectors, sc->xbd_vdevice, binfo, 1396 sector_size, phys_sector_size); 1397 } 1398 1399 (void)xenbus_set_state(dev, XenbusStateConnected); 1400 1401 /* Kick pending requests. */ 1402 mtx_lock(&sc->xbd_io_lock); 1403 sc->xbd_state = XBD_STATE_CONNECTED; 1404 xbd_startio(sc); 1405 sc->xbd_flags |= XBDF_READY; 1406 mtx_unlock(&sc->xbd_io_lock); 1407 } 1408 1409 /** 1410 * Handle the change of state of the backend to Closing. We must delete our 1411 * device-layer structures now, to ensure that writes are flushed through to 1412 * the backend. Once this is done, we can switch to Closed in 1413 * acknowledgement. 1414 */ 1415 static void 1416 xbd_closing(device_t dev) 1417 { 1418 struct xbd_softc *sc = device_get_softc(dev); 1419 1420 xenbus_set_state(dev, XenbusStateClosing); 1421 1422 DPRINTK("xbd_closing: %s removed\n", xenbus_get_node(dev)); 1423 1424 if (sc->xbd_disk != NULL) { 1425 disk_destroy(sc->xbd_disk); 1426 sc->xbd_disk = NULL; 1427 } 1428 1429 xenbus_set_state(dev, XenbusStateClosed); 1430 } 1431 1432 /*---------------------------- NewBus Entrypoints ----------------------------*/ 1433 static int 1434 xbd_probe(device_t dev) 1435 { 1436 if (strcmp(xenbus_get_type(dev), "vbd") != 0) 1437 return (ENXIO); 1438 1439 if (xen_pv_disks_disabled()) 1440 return (ENXIO); 1441 1442 if (xen_hvm_domain()) { 1443 int error; 1444 char *type; 1445 1446 /* 1447 * When running in an HVM domain, IDE disk emulation is 1448 * disabled early in boot so that native drivers will 1449 * not see emulated hardware. However, CDROM device 1450 * emulation cannot be disabled. 1451 * 1452 * Through use of FreeBSD's vm_guest and xen_hvm_domain() 1453 * APIs, we could modify the native CDROM driver to fail its 1454 * probe when running under Xen. Unfortunatlely, the PV 1455 * CDROM support in XenServer (up through at least version 1456 * 6.2) isn't functional, so we instead rely on the emulated 1457 * CDROM instance, and fail to attach the PV one here in 1458 * the blkfront driver. 1459 */ 1460 error = xs_read(XST_NIL, xenbus_get_node(dev), 1461 "device-type", NULL, (void **) &type); 1462 if (error) 1463 return (ENXIO); 1464 1465 if (strncmp(type, "cdrom", 5) == 0) { 1466 free(type, M_XENSTORE); 1467 return (ENXIO); 1468 } 1469 free(type, M_XENSTORE); 1470 } 1471 1472 device_set_desc(dev, "Virtual Block Device"); 1473 device_quiet(dev); 1474 return (0); 1475 } 1476 1477 /* 1478 * Setup supplies the backend dir, virtual device. We place an event 1479 * channel and shared frame entries. We watch backend to wait if it's 1480 * ok. 1481 */ 1482 static int 1483 xbd_attach(device_t dev) 1484 { 1485 struct xbd_softc *sc; 1486 const char *name; 1487 uint32_t vdevice; 1488 int error; 1489 int i; 1490 int unit; 1491 1492 /* FIXME: Use dynamic device id if this is not set. */ 1493 error = xs_scanf(XST_NIL, xenbus_get_node(dev), 1494 "virtual-device", NULL, "%" PRIu32, &vdevice); 1495 if (error) 1496 error = xs_scanf(XST_NIL, xenbus_get_node(dev), 1497 "virtual-device-ext", NULL, "%" PRIu32, &vdevice); 1498 if (error) { 1499 xenbus_dev_fatal(dev, error, "reading virtual-device"); 1500 device_printf(dev, "Couldn't determine virtual device.\n"); 1501 return (error); 1502 } 1503 1504 xbd_vdevice_to_unit(vdevice, &unit, &name); 1505 if (!strcmp(name, "xbd")) 1506 device_set_unit(dev, unit); 1507 1508 sc = device_get_softc(dev); 1509 mtx_init(&sc->xbd_io_lock, "blkfront i/o lock", NULL, MTX_DEF); 1510 xbd_initqs(sc); 1511 for (i = 0; i < XBD_MAX_RING_PAGES; i++) 1512 sc->xbd_ring_ref[i] = GRANT_REF_INVALID; 1513 1514 sc->xbd_dev = dev; 1515 sc->xbd_vdevice = vdevice; 1516 sc->xbd_state = XBD_STATE_DISCONNECTED; 1517 1518 xbd_setup_sysctl(sc); 1519 1520 /* Wait for backend device to publish its protocol capabilities. */ 1521 xenbus_set_state(dev, XenbusStateInitialising); 1522 1523 return (0); 1524 } 1525 1526 static int 1527 xbd_detach(device_t dev) 1528 { 1529 struct xbd_softc *sc = device_get_softc(dev); 1530 1531 DPRINTK("%s: %s removed\n", __func__, xenbus_get_node(dev)); 1532 1533 xbd_free(sc); 1534 mtx_destroy(&sc->xbd_io_lock); 1535 1536 return 0; 1537 } 1538 1539 static int 1540 xbd_suspend(device_t dev) 1541 { 1542 struct xbd_softc *sc = device_get_softc(dev); 1543 int retval; 1544 int saved_state; 1545 1546 /* Prevent new requests being issued until we fix things up. */ 1547 mtx_lock(&sc->xbd_io_lock); 1548 saved_state = sc->xbd_state; 1549 sc->xbd_state = XBD_STATE_SUSPENDED; 1550 1551 /* Wait for outstanding I/O to drain. */ 1552 retval = 0; 1553 while (xbd_queue_length(sc, XBD_Q_BUSY) != 0) { 1554 if (msleep(&sc->xbd_cm_q[XBD_Q_BUSY], &sc->xbd_io_lock, 1555 PRIBIO, "blkf_susp", 30 * hz) == EWOULDBLOCK) { 1556 retval = EBUSY; 1557 break; 1558 } 1559 } 1560 mtx_unlock(&sc->xbd_io_lock); 1561 1562 if (retval != 0) 1563 sc->xbd_state = saved_state; 1564 1565 return (retval); 1566 } 1567 1568 static int 1569 xbd_resume(device_t dev) 1570 { 1571 struct xbd_softc *sc = device_get_softc(dev); 1572 1573 if (xen_suspend_cancelled) { 1574 sc->xbd_state = XBD_STATE_CONNECTED; 1575 return (0); 1576 } 1577 1578 DPRINTK("xbd_resume: %s\n", xenbus_get_node(dev)); 1579 1580 xbd_free(sc); 1581 xbd_initialize(sc); 1582 return (0); 1583 } 1584 1585 /** 1586 * Callback received when the backend's state changes. 1587 */ 1588 static void 1589 xbd_backend_changed(device_t dev, XenbusState backend_state) 1590 { 1591 struct xbd_softc *sc = device_get_softc(dev); 1592 1593 DPRINTK("backend_state=%d\n", backend_state); 1594 1595 switch (backend_state) { 1596 case XenbusStateUnknown: 1597 case XenbusStateInitialising: 1598 case XenbusStateReconfigured: 1599 case XenbusStateReconfiguring: 1600 case XenbusStateClosed: 1601 break; 1602 1603 case XenbusStateInitWait: 1604 case XenbusStateInitialised: 1605 xbd_initialize(sc); 1606 break; 1607 1608 case XenbusStateConnected: 1609 xbd_initialize(sc); 1610 xbd_connect(sc); 1611 break; 1612 1613 case XenbusStateClosing: 1614 if (sc->xbd_users > 0) { 1615 device_printf(dev, "detaching with pending users\n"); 1616 KASSERT(sc->xbd_disk != NULL, 1617 ("NULL disk with pending users\n")); 1618 disk_gone(sc->xbd_disk); 1619 } else { 1620 xbd_closing(dev); 1621 } 1622 break; 1623 } 1624 } 1625 1626 /*---------------------------- NewBus Registration ---------------------------*/ 1627 static device_method_t xbd_methods[] = { 1628 /* Device interface */ 1629 DEVMETHOD(device_probe, xbd_probe), 1630 DEVMETHOD(device_attach, xbd_attach), 1631 DEVMETHOD(device_detach, xbd_detach), 1632 DEVMETHOD(device_shutdown, bus_generic_shutdown), 1633 DEVMETHOD(device_suspend, xbd_suspend), 1634 DEVMETHOD(device_resume, xbd_resume), 1635 1636 /* Xenbus interface */ 1637 DEVMETHOD(xenbus_otherend_changed, xbd_backend_changed), 1638 1639 { 0, 0 } 1640 }; 1641 1642 static driver_t xbd_driver = { 1643 "xbd", 1644 xbd_methods, 1645 sizeof(struct xbd_softc), 1646 }; 1647 1648 DRIVER_MODULE(xbd, xenbusb_front, xbd_driver, 0, 0); 1649