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