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 sbp; 603 int rc = 0; 604 605 if (length == 0) 606 return (0); 607 608 xbd_quiesce(sc); /* All quiet on the western front. */ 609 610 /* 611 * If this lock is held, then this module is failing, and a 612 * successful kernel dump is highly unlikely anyway. 613 */ 614 mtx_lock(&sc->xbd_io_lock); 615 616 /* Split the 64KB block as needed */ 617 for (sbp=0; length > 0; sbp++) { 618 cm = xbd_dequeue_cm(sc, XBD_Q_FREE); 619 if (cm == NULL) { 620 mtx_unlock(&sc->xbd_io_lock); 621 device_printf(sc->xbd_dev, "dump: no more commands?\n"); 622 return (EBUSY); 623 } 624 625 if (gnttab_alloc_grant_references(sc->xbd_max_request_segments, 626 &cm->cm_gref_head) != 0) { 627 xbd_free_command(cm); 628 mtx_unlock(&sc->xbd_io_lock); 629 device_printf(sc->xbd_dev, "no more grant allocs?\n"); 630 return (EBUSY); 631 } 632 633 chunk = length > sc->xbd_max_request_size ? 634 sc->xbd_max_request_size : length; 635 cm->cm_data = virtual; 636 cm->cm_datalen = chunk; 637 cm->cm_operation = BLKIF_OP_WRITE; 638 cm->cm_sector_number = offset / dp->d_sectorsize; 639 cm->cm_complete = xbd_dump_complete; 640 641 xbd_enqueue_cm(cm, XBD_Q_READY); 642 643 length -= chunk; 644 offset += chunk; 645 virtual = (char *) virtual + chunk; 646 } 647 648 /* Tell DOM0 to do the I/O */ 649 xbd_startio(sc); 650 mtx_unlock(&sc->xbd_io_lock); 651 652 /* Poll for the completion. */ 653 xbd_quiesce(sc); /* All quite on the eastern front */ 654 655 /* If there were any errors, bail out... */ 656 while ((cm = xbd_dequeue_cm(sc, XBD_Q_COMPLETE)) != NULL) { 657 if (cm->cm_status != BLKIF_RSP_OKAY) { 658 device_printf(sc->xbd_dev, 659 "Dump I/O failed at sector %jd\n", 660 cm->cm_sector_number); 661 rc = EIO; 662 } 663 xbd_free_command(cm); 664 } 665 666 return (rc); 667 } 668 669 /*----------------------------- Disk Entrypoints -----------------------------*/ 670 static int 671 xbd_open(struct disk *dp) 672 { 673 struct xbd_softc *sc = dp->d_drv1; 674 675 if (sc == NULL) { 676 printf("xbd%d: not found", dp->d_unit); 677 return (ENXIO); 678 } 679 680 sc->xbd_flags |= XBDF_OPEN; 681 sc->xbd_users++; 682 return (0); 683 } 684 685 static int 686 xbd_close(struct disk *dp) 687 { 688 struct xbd_softc *sc = dp->d_drv1; 689 690 if (sc == NULL) 691 return (ENXIO); 692 sc->xbd_flags &= ~XBDF_OPEN; 693 if (--(sc->xbd_users) == 0) { 694 /* 695 * Check whether we have been instructed to close. We will 696 * have ignored this request initially, as the device was 697 * still mounted. 698 */ 699 if (xenbus_get_otherend_state(sc->xbd_dev) == 700 XenbusStateClosing) 701 xbd_closing(sc->xbd_dev); 702 } 703 return (0); 704 } 705 706 static int 707 xbd_ioctl(struct disk *dp, u_long cmd, void *addr, int flag, struct thread *td) 708 { 709 struct xbd_softc *sc = dp->d_drv1; 710 711 if (sc == NULL) 712 return (ENXIO); 713 714 return (ENOTTY); 715 } 716 717 /* 718 * Read/write routine for a buffer. Finds the proper unit, place it on 719 * the sortq and kick the controller. 720 */ 721 static void 722 xbd_strategy(struct bio *bp) 723 { 724 struct xbd_softc *sc = bp->bio_disk->d_drv1; 725 726 /* bogus disk? */ 727 if (sc == NULL) { 728 bp->bio_error = EINVAL; 729 bp->bio_flags |= BIO_ERROR; 730 bp->bio_resid = bp->bio_bcount; 731 biodone(bp); 732 return; 733 } 734 735 /* 736 * Place it in the queue of disk activities for this disk 737 */ 738 mtx_lock(&sc->xbd_io_lock); 739 740 xbd_enqueue_bio(sc, bp); 741 xbd_startio(sc); 742 743 mtx_unlock(&sc->xbd_io_lock); 744 return; 745 } 746 747 /*------------------------------ Ring Management -----------------------------*/ 748 static int 749 xbd_alloc_ring(struct xbd_softc *sc) 750 { 751 blkif_sring_t *sring; 752 uintptr_t sring_page_addr; 753 int error; 754 int i; 755 756 sring = malloc(sc->xbd_ring_pages * PAGE_SIZE, M_XENBLOCKFRONT, 757 M_NOWAIT|M_ZERO); 758 if (sring == NULL) { 759 xenbus_dev_fatal(sc->xbd_dev, ENOMEM, "allocating shared ring"); 760 return (ENOMEM); 761 } 762 SHARED_RING_INIT(sring); 763 FRONT_RING_INIT(&sc->xbd_ring, sring, sc->xbd_ring_pages * PAGE_SIZE); 764 765 for (i = 0, sring_page_addr = (uintptr_t)sring; 766 i < sc->xbd_ring_pages; 767 i++, sring_page_addr += PAGE_SIZE) { 768 error = xenbus_grant_ring(sc->xbd_dev, 769 (vtophys(sring_page_addr) >> PAGE_SHIFT), 770 &sc->xbd_ring_ref[i]); 771 if (error) { 772 xenbus_dev_fatal(sc->xbd_dev, error, 773 "granting ring_ref(%d)", i); 774 return (error); 775 } 776 } 777 if (sc->xbd_ring_pages == 1) { 778 error = xs_printf(XST_NIL, xenbus_get_node(sc->xbd_dev), 779 "ring-ref", "%u", sc->xbd_ring_ref[0]); 780 if (error) { 781 xenbus_dev_fatal(sc->xbd_dev, error, 782 "writing %s/ring-ref", 783 xenbus_get_node(sc->xbd_dev)); 784 return (error); 785 } 786 } else { 787 for (i = 0; i < sc->xbd_ring_pages; i++) { 788 char ring_ref_name[]= "ring_refXX"; 789 790 snprintf(ring_ref_name, sizeof(ring_ref_name), 791 "ring-ref%u", i); 792 error = xs_printf(XST_NIL, xenbus_get_node(sc->xbd_dev), 793 ring_ref_name, "%u", sc->xbd_ring_ref[i]); 794 if (error) { 795 xenbus_dev_fatal(sc->xbd_dev, error, 796 "writing %s/%s", 797 xenbus_get_node(sc->xbd_dev), 798 ring_ref_name); 799 return (error); 800 } 801 } 802 } 803 804 error = xen_intr_alloc_and_bind_local_port(sc->xbd_dev, 805 xenbus_get_otherend_id(sc->xbd_dev), NULL, xbd_int, sc, 806 INTR_TYPE_BIO | INTR_MPSAFE, &sc->xen_intr_handle); 807 if (error) { 808 xenbus_dev_fatal(sc->xbd_dev, error, 809 "xen_intr_alloc_and_bind_local_port failed"); 810 return (error); 811 } 812 813 return (0); 814 } 815 816 static void 817 xbd_free_ring(struct xbd_softc *sc) 818 { 819 int i; 820 821 if (sc->xbd_ring.sring == NULL) 822 return; 823 824 for (i = 0; i < sc->xbd_ring_pages; i++) { 825 if (sc->xbd_ring_ref[i] != GRANT_REF_INVALID) { 826 gnttab_end_foreign_access_ref(sc->xbd_ring_ref[i]); 827 sc->xbd_ring_ref[i] = GRANT_REF_INVALID; 828 } 829 } 830 free(sc->xbd_ring.sring, M_XENBLOCKFRONT); 831 sc->xbd_ring.sring = NULL; 832 } 833 834 /*-------------------------- Initialization/Teardown -------------------------*/ 835 static int 836 xbd_feature_string(struct xbd_softc *sc, char *features, size_t len) 837 { 838 struct sbuf sb; 839 int feature_cnt; 840 841 sbuf_new(&sb, features, len, SBUF_FIXEDLEN); 842 843 feature_cnt = 0; 844 if ((sc->xbd_flags & XBDF_FLUSH) != 0) { 845 sbuf_printf(&sb, "flush"); 846 feature_cnt++; 847 } 848 849 if ((sc->xbd_flags & XBDF_BARRIER) != 0) { 850 if (feature_cnt != 0) 851 sbuf_printf(&sb, ", "); 852 sbuf_printf(&sb, "write_barrier"); 853 feature_cnt++; 854 } 855 856 if ((sc->xbd_flags & XBDF_DISCARD) != 0) { 857 if (feature_cnt != 0) 858 sbuf_printf(&sb, ", "); 859 sbuf_printf(&sb, "discard"); 860 feature_cnt++; 861 } 862 863 if ((sc->xbd_flags & XBDF_PERSISTENT) != 0) { 864 if (feature_cnt != 0) 865 sbuf_printf(&sb, ", "); 866 sbuf_printf(&sb, "persistent_grants"); 867 feature_cnt++; 868 } 869 870 (void) sbuf_finish(&sb); 871 return (sbuf_len(&sb)); 872 } 873 874 static int 875 xbd_sysctl_features(SYSCTL_HANDLER_ARGS) 876 { 877 char features[80]; 878 struct xbd_softc *sc = arg1; 879 int error; 880 int len; 881 882 error = sysctl_wire_old_buffer(req, 0); 883 if (error != 0) 884 return (error); 885 886 len = xbd_feature_string(sc, features, sizeof(features)); 887 888 /* len is -1 on error, which will make the SYSCTL_OUT a no-op. */ 889 return (SYSCTL_OUT(req, features, len + 1/*NUL*/)); 890 } 891 892 static void 893 xbd_setup_sysctl(struct xbd_softc *xbd) 894 { 895 struct sysctl_ctx_list *sysctl_ctx = NULL; 896 struct sysctl_oid *sysctl_tree = NULL; 897 struct sysctl_oid_list *children; 898 899 sysctl_ctx = device_get_sysctl_ctx(xbd->xbd_dev); 900 if (sysctl_ctx == NULL) 901 return; 902 903 sysctl_tree = device_get_sysctl_tree(xbd->xbd_dev); 904 if (sysctl_tree == NULL) 905 return; 906 907 children = SYSCTL_CHILDREN(sysctl_tree); 908 SYSCTL_ADD_UINT(sysctl_ctx, children, OID_AUTO, 909 "max_requests", CTLFLAG_RD, &xbd->xbd_max_requests, -1, 910 "maximum outstanding requests (negotiated)"); 911 912 SYSCTL_ADD_UINT(sysctl_ctx, children, OID_AUTO, 913 "max_request_segments", CTLFLAG_RD, 914 &xbd->xbd_max_request_segments, 0, 915 "maximum number of pages per requests (negotiated)"); 916 917 SYSCTL_ADD_UINT(sysctl_ctx, children, OID_AUTO, 918 "max_request_size", CTLFLAG_RD, &xbd->xbd_max_request_size, 0, 919 "maximum size in bytes of a request (negotiated)"); 920 921 SYSCTL_ADD_UINT(sysctl_ctx, children, OID_AUTO, 922 "ring_pages", CTLFLAG_RD, &xbd->xbd_ring_pages, 0, 923 "communication channel pages (negotiated)"); 924 925 SYSCTL_ADD_PROC(sysctl_ctx, children, OID_AUTO, 926 "features", CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, xbd, 927 0, xbd_sysctl_features, "A", "protocol features (negotiated)"); 928 } 929 930 /* 931 * Translate Linux major/minor to an appropriate name and unit 932 * number. For HVM guests, this allows us to use the same drive names 933 * with blkfront as the emulated drives, easing transition slightly. 934 */ 935 static void 936 xbd_vdevice_to_unit(uint32_t vdevice, int *unit, const char **name) 937 { 938 static struct vdev_info { 939 int major; 940 int shift; 941 int base; 942 const char *name; 943 } info[] = { 944 {3, 6, 0, "ada"}, /* ide0 */ 945 {22, 6, 2, "ada"}, /* ide1 */ 946 {33, 6, 4, "ada"}, /* ide2 */ 947 {34, 6, 6, "ada"}, /* ide3 */ 948 {56, 6, 8, "ada"}, /* ide4 */ 949 {57, 6, 10, "ada"}, /* ide5 */ 950 {88, 6, 12, "ada"}, /* ide6 */ 951 {89, 6, 14, "ada"}, /* ide7 */ 952 {90, 6, 16, "ada"}, /* ide8 */ 953 {91, 6, 18, "ada"}, /* ide9 */ 954 955 {8, 4, 0, "da"}, /* scsi disk0 */ 956 {65, 4, 16, "da"}, /* scsi disk1 */ 957 {66, 4, 32, "da"}, /* scsi disk2 */ 958 {67, 4, 48, "da"}, /* scsi disk3 */ 959 {68, 4, 64, "da"}, /* scsi disk4 */ 960 {69, 4, 80, "da"}, /* scsi disk5 */ 961 {70, 4, 96, "da"}, /* scsi disk6 */ 962 {71, 4, 112, "da"}, /* scsi disk7 */ 963 {128, 4, 128, "da"}, /* scsi disk8 */ 964 {129, 4, 144, "da"}, /* scsi disk9 */ 965 {130, 4, 160, "da"}, /* scsi disk10 */ 966 {131, 4, 176, "da"}, /* scsi disk11 */ 967 {132, 4, 192, "da"}, /* scsi disk12 */ 968 {133, 4, 208, "da"}, /* scsi disk13 */ 969 {134, 4, 224, "da"}, /* scsi disk14 */ 970 {135, 4, 240, "da"}, /* scsi disk15 */ 971 972 {202, 4, 0, "xbd"}, /* xbd */ 973 974 {0, 0, 0, NULL}, 975 }; 976 int major = vdevice >> 8; 977 int minor = vdevice & 0xff; 978 int i; 979 980 if (vdevice & (1 << 28)) { 981 *unit = (vdevice & ((1 << 28) - 1)) >> 8; 982 *name = "xbd"; 983 return; 984 } 985 986 for (i = 0; info[i].major; i++) { 987 if (info[i].major == major) { 988 *unit = info[i].base + (minor >> info[i].shift); 989 *name = info[i].name; 990 return; 991 } 992 } 993 994 *unit = minor >> 4; 995 *name = "xbd"; 996 } 997 998 int 999 xbd_instance_create(struct xbd_softc *sc, blkif_sector_t sectors, 1000 int vdevice, uint16_t vdisk_info, unsigned long sector_size, 1001 unsigned long phys_sector_size) 1002 { 1003 char features[80]; 1004 int unit, error = 0; 1005 const char *name; 1006 1007 xbd_vdevice_to_unit(vdevice, &unit, &name); 1008 1009 sc->xbd_unit = unit; 1010 1011 if (strcmp(name, "xbd") != 0) 1012 device_printf(sc->xbd_dev, "attaching as %s%d\n", name, unit); 1013 1014 if (xbd_feature_string(sc, features, sizeof(features)) > 0) { 1015 device_printf(sc->xbd_dev, "features: %s\n", 1016 features); 1017 } 1018 1019 sc->xbd_disk = disk_alloc(); 1020 sc->xbd_disk->d_unit = sc->xbd_unit; 1021 sc->xbd_disk->d_open = xbd_open; 1022 sc->xbd_disk->d_close = xbd_close; 1023 sc->xbd_disk->d_ioctl = xbd_ioctl; 1024 sc->xbd_disk->d_strategy = xbd_strategy; 1025 sc->xbd_disk->d_dump = xbd_dump; 1026 sc->xbd_disk->d_name = name; 1027 sc->xbd_disk->d_drv1 = sc; 1028 sc->xbd_disk->d_sectorsize = sector_size; 1029 sc->xbd_disk->d_stripesize = phys_sector_size; 1030 sc->xbd_disk->d_stripeoffset = 0; 1031 1032 sc->xbd_disk->d_mediasize = sectors * sector_size; 1033 sc->xbd_disk->d_maxsize = sc->xbd_max_request_size; 1034 sc->xbd_disk->d_flags = DISKFLAG_UNMAPPED_BIO; 1035 if ((sc->xbd_flags & (XBDF_FLUSH|XBDF_BARRIER)) != 0) { 1036 sc->xbd_disk->d_flags |= DISKFLAG_CANFLUSHCACHE; 1037 device_printf(sc->xbd_dev, 1038 "synchronize cache commands enabled.\n"); 1039 } 1040 disk_create(sc->xbd_disk, DISK_VERSION); 1041 1042 return error; 1043 } 1044 1045 static void 1046 xbd_free(struct xbd_softc *sc) 1047 { 1048 int i; 1049 1050 /* Prevent new requests being issued until we fix things up. */ 1051 mtx_lock(&sc->xbd_io_lock); 1052 sc->xbd_state = XBD_STATE_DISCONNECTED; 1053 mtx_unlock(&sc->xbd_io_lock); 1054 1055 /* Free resources associated with old device channel. */ 1056 xbd_free_ring(sc); 1057 if (sc->xbd_shadow) { 1058 for (i = 0; i < sc->xbd_max_requests; i++) { 1059 struct xbd_command *cm; 1060 1061 cm = &sc->xbd_shadow[i]; 1062 if (cm->cm_sg_refs != NULL) { 1063 free(cm->cm_sg_refs, M_XENBLOCKFRONT); 1064 cm->cm_sg_refs = NULL; 1065 } 1066 1067 if (cm->cm_indirectionpages != NULL) { 1068 gnttab_end_foreign_access_references( 1069 sc->xbd_max_request_indirectpages, 1070 &cm->cm_indirectionrefs[0]); 1071 contigfree(cm->cm_indirectionpages, PAGE_SIZE * 1072 sc->xbd_max_request_indirectpages, 1073 M_XENBLOCKFRONT); 1074 cm->cm_indirectionpages = NULL; 1075 } 1076 1077 bus_dmamap_destroy(sc->xbd_io_dmat, cm->cm_map); 1078 } 1079 free(sc->xbd_shadow, M_XENBLOCKFRONT); 1080 sc->xbd_shadow = NULL; 1081 1082 bus_dma_tag_destroy(sc->xbd_io_dmat); 1083 1084 xbd_initq_cm(sc, XBD_Q_FREE); 1085 xbd_initq_cm(sc, XBD_Q_READY); 1086 xbd_initq_cm(sc, XBD_Q_COMPLETE); 1087 } 1088 1089 xen_intr_unbind(&sc->xen_intr_handle); 1090 1091 } 1092 1093 /*--------------------------- State Change Handlers --------------------------*/ 1094 static void 1095 xbd_initialize(struct xbd_softc *sc) 1096 { 1097 const char *otherend_path; 1098 const char *node_path; 1099 uint32_t max_ring_page_order; 1100 int error; 1101 1102 if (xenbus_get_state(sc->xbd_dev) != XenbusStateInitialising) { 1103 /* Initialization has already been performed. */ 1104 return; 1105 } 1106 1107 /* 1108 * Protocol defaults valid even if negotiation for a 1109 * setting fails. 1110 */ 1111 max_ring_page_order = 0; 1112 sc->xbd_ring_pages = 1; 1113 1114 /* 1115 * Protocol negotiation. 1116 * 1117 * \note xs_gather() returns on the first encountered error, so 1118 * we must use independent calls in order to guarantee 1119 * we don't miss information in a sparsly populated back-end 1120 * tree. 1121 * 1122 * \note xs_scanf() does not update variables for unmatched 1123 * fields. 1124 */ 1125 otherend_path = xenbus_get_otherend_path(sc->xbd_dev); 1126 node_path = xenbus_get_node(sc->xbd_dev); 1127 1128 /* Support both backend schemes for relaying ring page limits. */ 1129 (void)xs_scanf(XST_NIL, otherend_path, 1130 "max-ring-page-order", NULL, "%" PRIu32, 1131 &max_ring_page_order); 1132 sc->xbd_ring_pages = 1 << max_ring_page_order; 1133 (void)xs_scanf(XST_NIL, otherend_path, 1134 "max-ring-pages", NULL, "%" PRIu32, 1135 &sc->xbd_ring_pages); 1136 if (sc->xbd_ring_pages < 1) 1137 sc->xbd_ring_pages = 1; 1138 1139 if (sc->xbd_ring_pages > XBD_MAX_RING_PAGES) { 1140 device_printf(sc->xbd_dev, 1141 "Back-end specified ring-pages of %u " 1142 "limited to front-end limit of %u.\n", 1143 sc->xbd_ring_pages, XBD_MAX_RING_PAGES); 1144 sc->xbd_ring_pages = XBD_MAX_RING_PAGES; 1145 } 1146 1147 if (powerof2(sc->xbd_ring_pages) == 0) { 1148 uint32_t new_page_limit; 1149 1150 new_page_limit = 0x01 << (fls(sc->xbd_ring_pages) - 1); 1151 device_printf(sc->xbd_dev, 1152 "Back-end specified ring-pages of %u " 1153 "is not a power of 2. Limited to %u.\n", 1154 sc->xbd_ring_pages, new_page_limit); 1155 sc->xbd_ring_pages = new_page_limit; 1156 } 1157 1158 sc->xbd_max_requests = 1159 BLKIF_MAX_RING_REQUESTS(sc->xbd_ring_pages * PAGE_SIZE); 1160 if (sc->xbd_max_requests > XBD_MAX_REQUESTS) { 1161 device_printf(sc->xbd_dev, 1162 "Back-end specified max_requests of %u " 1163 "limited to front-end limit of %zu.\n", 1164 sc->xbd_max_requests, XBD_MAX_REQUESTS); 1165 sc->xbd_max_requests = XBD_MAX_REQUESTS; 1166 } 1167 1168 if (xbd_alloc_ring(sc) != 0) 1169 return; 1170 1171 /* Support both backend schemes for relaying ring page limits. */ 1172 if (sc->xbd_ring_pages > 1) { 1173 error = xs_printf(XST_NIL, node_path, 1174 "num-ring-pages","%u", 1175 sc->xbd_ring_pages); 1176 if (error) { 1177 xenbus_dev_fatal(sc->xbd_dev, error, 1178 "writing %s/num-ring-pages", 1179 node_path); 1180 return; 1181 } 1182 1183 error = xs_printf(XST_NIL, node_path, 1184 "ring-page-order", "%u", 1185 fls(sc->xbd_ring_pages) - 1); 1186 if (error) { 1187 xenbus_dev_fatal(sc->xbd_dev, error, 1188 "writing %s/ring-page-order", 1189 node_path); 1190 return; 1191 } 1192 } 1193 1194 error = xs_printf(XST_NIL, node_path, "event-channel", 1195 "%u", xen_intr_port(sc->xen_intr_handle)); 1196 if (error) { 1197 xenbus_dev_fatal(sc->xbd_dev, error, 1198 "writing %s/event-channel", 1199 node_path); 1200 return; 1201 } 1202 1203 error = xs_printf(XST_NIL, node_path, "protocol", 1204 "%s", XEN_IO_PROTO_ABI_NATIVE); 1205 if (error) { 1206 xenbus_dev_fatal(sc->xbd_dev, error, 1207 "writing %s/protocol", 1208 node_path); 1209 return; 1210 } 1211 1212 xenbus_set_state(sc->xbd_dev, XenbusStateInitialised); 1213 } 1214 1215 /* 1216 * Invoked when the backend is finally 'ready' (and has published 1217 * the details about the physical device - #sectors, size, etc). 1218 */ 1219 static void 1220 xbd_connect(struct xbd_softc *sc) 1221 { 1222 device_t dev = sc->xbd_dev; 1223 blkif_sector_t sectors; 1224 unsigned long sector_size, phys_sector_size; 1225 unsigned int binfo; 1226 int err, feature_barrier, feature_flush; 1227 int i, j; 1228 1229 DPRINTK("blkfront.c:connect:%s.\n", xenbus_get_otherend_path(dev)); 1230 1231 if (sc->xbd_state == XBD_STATE_SUSPENDED) { 1232 return; 1233 } 1234 1235 if (sc->xbd_state == XBD_STATE_CONNECTED) { 1236 struct disk *disk; 1237 1238 disk = sc->xbd_disk; 1239 if (disk == NULL) { 1240 return; 1241 } 1242 err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev), 1243 "sectors", "%"PRIu64, §ors, NULL); 1244 if (err != 0) { 1245 xenbus_dev_error(dev, err, 1246 "reading sectors at %s", 1247 xenbus_get_otherend_path(dev)); 1248 return; 1249 } 1250 disk->d_mediasize = disk->d_sectorsize * sectors; 1251 err = disk_resize(disk, M_NOWAIT); 1252 if (err) { 1253 xenbus_dev_error(dev, err, 1254 "unable to resize disk %s%u", 1255 disk->d_name, disk->d_unit); 1256 return; 1257 } 1258 device_printf(sc->xbd_dev, 1259 "changed capacity to %jd\n", 1260 (intmax_t)disk->d_mediasize); 1261 return; 1262 } 1263 1264 err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev), 1265 "sectors", "%"PRIu64, §ors, 1266 "info", "%u", &binfo, 1267 "sector-size", "%lu", §or_size, 1268 NULL); 1269 if (err) { 1270 xenbus_dev_fatal(dev, err, 1271 "reading backend fields at %s", 1272 xenbus_get_otherend_path(dev)); 1273 return; 1274 } 1275 if ((sectors == 0) || (sector_size == 0)) { 1276 xenbus_dev_fatal(dev, 0, 1277 "invalid parameters from %s:" 1278 " sectors = %"PRIu64", sector_size = %lu", 1279 xenbus_get_otherend_path(dev), 1280 sectors, sector_size); 1281 return; 1282 } 1283 err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev), 1284 "physical-sector-size", "%lu", &phys_sector_size, 1285 NULL); 1286 if (err || phys_sector_size <= sector_size) 1287 phys_sector_size = 0; 1288 err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev), 1289 "feature-barrier", "%d", &feature_barrier, 1290 NULL); 1291 if (err == 0 && feature_barrier != 0) 1292 sc->xbd_flags |= XBDF_BARRIER; 1293 1294 err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev), 1295 "feature-flush-cache", "%d", &feature_flush, 1296 NULL); 1297 if (err == 0 && feature_flush != 0) 1298 sc->xbd_flags |= XBDF_FLUSH; 1299 1300 err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev), 1301 "feature-max-indirect-segments", "%" PRIu32, 1302 &sc->xbd_max_request_segments, NULL); 1303 if ((err != 0) || (xbd_enable_indirect == 0)) 1304 sc->xbd_max_request_segments = 0; 1305 if (sc->xbd_max_request_segments > XBD_MAX_INDIRECT_SEGMENTS) 1306 sc->xbd_max_request_segments = XBD_MAX_INDIRECT_SEGMENTS; 1307 if (sc->xbd_max_request_segments > XBD_SIZE_TO_SEGS(maxphys)) 1308 sc->xbd_max_request_segments = XBD_SIZE_TO_SEGS(maxphys); 1309 sc->xbd_max_request_indirectpages = 1310 XBD_INDIRECT_SEGS_TO_PAGES(sc->xbd_max_request_segments); 1311 if (sc->xbd_max_request_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST) 1312 sc->xbd_max_request_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST; 1313 sc->xbd_max_request_size = 1314 XBD_SEGS_TO_SIZE(sc->xbd_max_request_segments); 1315 1316 /* Allocate datastructures based on negotiated values. */ 1317 err = bus_dma_tag_create( 1318 bus_get_dma_tag(sc->xbd_dev), /* parent */ 1319 512, PAGE_SIZE, /* algnmnt, boundary */ 1320 BUS_SPACE_MAXADDR, /* lowaddr */ 1321 BUS_SPACE_MAXADDR, /* highaddr */ 1322 NULL, NULL, /* filter, filterarg */ 1323 sc->xbd_max_request_size, 1324 sc->xbd_max_request_segments, 1325 PAGE_SIZE, /* maxsegsize */ 1326 BUS_DMA_ALLOCNOW, /* flags */ 1327 busdma_lock_mutex, /* lockfunc */ 1328 &sc->xbd_io_lock, /* lockarg */ 1329 &sc->xbd_io_dmat); 1330 if (err != 0) { 1331 xenbus_dev_fatal(sc->xbd_dev, err, 1332 "Cannot allocate parent DMA tag\n"); 1333 return; 1334 } 1335 1336 /* Per-transaction data allocation. */ 1337 sc->xbd_shadow = malloc(sizeof(*sc->xbd_shadow) * sc->xbd_max_requests, 1338 M_XENBLOCKFRONT, M_NOWAIT|M_ZERO); 1339 if (sc->xbd_shadow == NULL) { 1340 bus_dma_tag_destroy(sc->xbd_io_dmat); 1341 xenbus_dev_fatal(sc->xbd_dev, ENOMEM, 1342 "Cannot allocate request structures\n"); 1343 return; 1344 } 1345 1346 for (i = 0; i < sc->xbd_max_requests; i++) { 1347 struct xbd_command *cm; 1348 void * indirectpages; 1349 1350 cm = &sc->xbd_shadow[i]; 1351 cm->cm_sg_refs = malloc( 1352 sizeof(grant_ref_t) * sc->xbd_max_request_segments, 1353 M_XENBLOCKFRONT, M_NOWAIT); 1354 if (cm->cm_sg_refs == NULL) 1355 break; 1356 cm->cm_id = i; 1357 cm->cm_flags = XBDCF_INITIALIZER; 1358 cm->cm_sc = sc; 1359 if (bus_dmamap_create(sc->xbd_io_dmat, 0, &cm->cm_map) != 0) 1360 break; 1361 if (sc->xbd_max_request_indirectpages > 0) { 1362 indirectpages = contigmalloc( 1363 PAGE_SIZE * sc->xbd_max_request_indirectpages, 1364 M_XENBLOCKFRONT, M_ZERO | M_NOWAIT, 0, ~0, 1365 PAGE_SIZE, 0); 1366 if (indirectpages == NULL) 1367 sc->xbd_max_request_indirectpages = 0; 1368 } else { 1369 indirectpages = NULL; 1370 } 1371 for (j = 0; j < sc->xbd_max_request_indirectpages; j++) { 1372 if (gnttab_grant_foreign_access( 1373 xenbus_get_otherend_id(sc->xbd_dev), 1374 (vtophys(indirectpages) >> PAGE_SHIFT) + j, 1375 1 /* grant read-only access */, 1376 &cm->cm_indirectionrefs[j])) 1377 break; 1378 } 1379 if (j < sc->xbd_max_request_indirectpages) { 1380 contigfree(indirectpages, 1381 PAGE_SIZE * sc->xbd_max_request_indirectpages, 1382 M_XENBLOCKFRONT); 1383 break; 1384 } 1385 cm->cm_indirectionpages = indirectpages; 1386 xbd_free_command(cm); 1387 } 1388 1389 if (sc->xbd_disk == NULL) { 1390 device_printf(dev, "%juMB <%s> at %s", 1391 (uintmax_t) sectors / (1048576 / sector_size), 1392 device_get_desc(dev), 1393 xenbus_get_node(dev)); 1394 bus_print_child_footer(device_get_parent(dev), dev); 1395 1396 xbd_instance_create(sc, sectors, sc->xbd_vdevice, binfo, 1397 sector_size, phys_sector_size); 1398 } 1399 1400 (void)xenbus_set_state(dev, XenbusStateConnected); 1401 1402 /* Kick pending requests. */ 1403 mtx_lock(&sc->xbd_io_lock); 1404 sc->xbd_state = XBD_STATE_CONNECTED; 1405 xbd_startio(sc); 1406 sc->xbd_flags |= XBDF_READY; 1407 mtx_unlock(&sc->xbd_io_lock); 1408 } 1409 1410 /** 1411 * Handle the change of state of the backend to Closing. We must delete our 1412 * device-layer structures now, to ensure that writes are flushed through to 1413 * the backend. Once this is done, we can switch to Closed in 1414 * acknowledgement. 1415 */ 1416 static void 1417 xbd_closing(device_t dev) 1418 { 1419 struct xbd_softc *sc = device_get_softc(dev); 1420 1421 xenbus_set_state(dev, XenbusStateClosing); 1422 1423 DPRINTK("xbd_closing: %s removed\n", xenbus_get_node(dev)); 1424 1425 if (sc->xbd_disk != NULL) { 1426 disk_destroy(sc->xbd_disk); 1427 sc->xbd_disk = NULL; 1428 } 1429 1430 xenbus_set_state(dev, XenbusStateClosed); 1431 } 1432 1433 /*---------------------------- NewBus Entrypoints ----------------------------*/ 1434 static int 1435 xbd_probe(device_t dev) 1436 { 1437 if (strcmp(xenbus_get_type(dev), "vbd") != 0) 1438 return (ENXIO); 1439 1440 if (xen_pv_disks_disabled()) 1441 return (ENXIO); 1442 1443 if (xen_hvm_domain()) { 1444 int error; 1445 char *type; 1446 1447 /* 1448 * When running in an HVM domain, IDE disk emulation is 1449 * disabled early in boot so that native drivers will 1450 * not see emulated hardware. However, CDROM device 1451 * emulation cannot be disabled. 1452 * 1453 * Through use of FreeBSD's vm_guest and xen_hvm_domain() 1454 * APIs, we could modify the native CDROM driver to fail its 1455 * probe when running under Xen. Unfortunatlely, the PV 1456 * CDROM support in XenServer (up through at least version 1457 * 6.2) isn't functional, so we instead rely on the emulated 1458 * CDROM instance, and fail to attach the PV one here in 1459 * the blkfront driver. 1460 */ 1461 error = xs_read(XST_NIL, xenbus_get_node(dev), 1462 "device-type", NULL, (void **) &type); 1463 if (error) 1464 return (ENXIO); 1465 1466 if (strncmp(type, "cdrom", 5) == 0) { 1467 free(type, M_XENSTORE); 1468 return (ENXIO); 1469 } 1470 free(type, M_XENSTORE); 1471 } 1472 1473 device_set_desc(dev, "Virtual Block Device"); 1474 device_quiet(dev); 1475 return (0); 1476 } 1477 1478 /* 1479 * Setup supplies the backend dir, virtual device. We place an event 1480 * channel and shared frame entries. We watch backend to wait if it's 1481 * ok. 1482 */ 1483 static int 1484 xbd_attach(device_t dev) 1485 { 1486 struct xbd_softc *sc; 1487 const char *name; 1488 uint32_t vdevice; 1489 int error; 1490 int i; 1491 int unit; 1492 1493 /* FIXME: Use dynamic device id if this is not set. */ 1494 error = xs_scanf(XST_NIL, xenbus_get_node(dev), 1495 "virtual-device", NULL, "%" PRIu32, &vdevice); 1496 if (error) 1497 error = xs_scanf(XST_NIL, xenbus_get_node(dev), 1498 "virtual-device-ext", NULL, "%" PRIu32, &vdevice); 1499 if (error) { 1500 xenbus_dev_fatal(dev, error, "reading virtual-device"); 1501 device_printf(dev, "Couldn't determine virtual device.\n"); 1502 return (error); 1503 } 1504 1505 xbd_vdevice_to_unit(vdevice, &unit, &name); 1506 if (!strcmp(name, "xbd")) 1507 device_set_unit(dev, unit); 1508 1509 sc = device_get_softc(dev); 1510 mtx_init(&sc->xbd_io_lock, "blkfront i/o lock", NULL, MTX_DEF); 1511 xbd_initqs(sc); 1512 for (i = 0; i < XBD_MAX_RING_PAGES; i++) 1513 sc->xbd_ring_ref[i] = GRANT_REF_INVALID; 1514 1515 sc->xbd_dev = dev; 1516 sc->xbd_vdevice = vdevice; 1517 sc->xbd_state = XBD_STATE_DISCONNECTED; 1518 1519 xbd_setup_sysctl(sc); 1520 1521 /* Wait for backend device to publish its protocol capabilities. */ 1522 xenbus_set_state(dev, XenbusStateInitialising); 1523 1524 return (0); 1525 } 1526 1527 static int 1528 xbd_detach(device_t dev) 1529 { 1530 struct xbd_softc *sc = device_get_softc(dev); 1531 1532 DPRINTK("%s: %s removed\n", __func__, xenbus_get_node(dev)); 1533 1534 xbd_free(sc); 1535 mtx_destroy(&sc->xbd_io_lock); 1536 1537 return 0; 1538 } 1539 1540 static int 1541 xbd_suspend(device_t dev) 1542 { 1543 struct xbd_softc *sc = device_get_softc(dev); 1544 int retval; 1545 int saved_state; 1546 1547 /* Prevent new requests being issued until we fix things up. */ 1548 mtx_lock(&sc->xbd_io_lock); 1549 saved_state = sc->xbd_state; 1550 sc->xbd_state = XBD_STATE_SUSPENDED; 1551 1552 /* Wait for outstanding I/O to drain. */ 1553 retval = 0; 1554 while (xbd_queue_length(sc, XBD_Q_BUSY) != 0) { 1555 if (msleep(&sc->xbd_cm_q[XBD_Q_BUSY], &sc->xbd_io_lock, 1556 PRIBIO, "blkf_susp", 30 * hz) == EWOULDBLOCK) { 1557 retval = EBUSY; 1558 break; 1559 } 1560 } 1561 mtx_unlock(&sc->xbd_io_lock); 1562 1563 if (retval != 0) 1564 sc->xbd_state = saved_state; 1565 1566 return (retval); 1567 } 1568 1569 static int 1570 xbd_resume(device_t dev) 1571 { 1572 struct xbd_softc *sc = device_get_softc(dev); 1573 1574 if (xen_suspend_cancelled) { 1575 sc->xbd_state = XBD_STATE_CONNECTED; 1576 return (0); 1577 } 1578 1579 DPRINTK("xbd_resume: %s\n", xenbus_get_node(dev)); 1580 1581 xbd_free(sc); 1582 xbd_initialize(sc); 1583 return (0); 1584 } 1585 1586 /** 1587 * Callback received when the backend's state changes. 1588 */ 1589 static void 1590 xbd_backend_changed(device_t dev, XenbusState backend_state) 1591 { 1592 struct xbd_softc *sc = device_get_softc(dev); 1593 1594 DPRINTK("backend_state=%d\n", backend_state); 1595 1596 switch (backend_state) { 1597 case XenbusStateUnknown: 1598 case XenbusStateInitialising: 1599 case XenbusStateReconfigured: 1600 case XenbusStateReconfiguring: 1601 case XenbusStateClosed: 1602 break; 1603 1604 case XenbusStateInitWait: 1605 case XenbusStateInitialised: 1606 xbd_initialize(sc); 1607 break; 1608 1609 case XenbusStateConnected: 1610 xbd_initialize(sc); 1611 xbd_connect(sc); 1612 break; 1613 1614 case XenbusStateClosing: 1615 if (sc->xbd_users > 0) { 1616 device_printf(dev, "detaching with pending users\n"); 1617 KASSERT(sc->xbd_disk != NULL, 1618 ("NULL disk with pending users\n")); 1619 disk_gone(sc->xbd_disk); 1620 } else { 1621 xbd_closing(dev); 1622 } 1623 break; 1624 } 1625 } 1626 1627 /*---------------------------- NewBus Registration ---------------------------*/ 1628 static device_method_t xbd_methods[] = { 1629 /* Device interface */ 1630 DEVMETHOD(device_probe, xbd_probe), 1631 DEVMETHOD(device_attach, xbd_attach), 1632 DEVMETHOD(device_detach, xbd_detach), 1633 DEVMETHOD(device_shutdown, bus_generic_shutdown), 1634 DEVMETHOD(device_suspend, xbd_suspend), 1635 DEVMETHOD(device_resume, xbd_resume), 1636 1637 /* Xenbus interface */ 1638 DEVMETHOD(xenbus_otherend_changed, xbd_backend_changed), 1639 1640 { 0, 0 } 1641 }; 1642 1643 static driver_t xbd_driver = { 1644 "xbd", 1645 xbd_methods, 1646 sizeof(struct xbd_softc), 1647 }; 1648 1649 DRIVER_MODULE(xbd, xenbusb_front, xbd_driver, 0, 0); 1650