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 <xen/interface/grant_table.h> 57 #include <xen/interface/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, vm_offset_t physical, off_t offset, 597 size_t length) 598 { 599 struct disk *dp = arg; 600 struct xbd_softc *sc = dp->d_drv1; 601 struct xbd_command *cm; 602 size_t chunk; 603 int sbp; 604 int rc = 0; 605 606 if (length == 0) 607 return (0); 608 609 xbd_quiesce(sc); /* All quiet on the western front. */ 610 611 /* 612 * If this lock is held, then this module is failing, and a 613 * successful kernel dump is highly unlikely anyway. 614 */ 615 mtx_lock(&sc->xbd_io_lock); 616 617 /* Split the 64KB block as needed */ 618 for (sbp=0; length > 0; sbp++) { 619 cm = xbd_dequeue_cm(sc, XBD_Q_FREE); 620 if (cm == NULL) { 621 mtx_unlock(&sc->xbd_io_lock); 622 device_printf(sc->xbd_dev, "dump: no more commands?\n"); 623 return (EBUSY); 624 } 625 626 if (gnttab_alloc_grant_references(sc->xbd_max_request_segments, 627 &cm->cm_gref_head) != 0) { 628 xbd_free_command(cm); 629 mtx_unlock(&sc->xbd_io_lock); 630 device_printf(sc->xbd_dev, "no more grant allocs?\n"); 631 return (EBUSY); 632 } 633 634 chunk = length > sc->xbd_max_request_size ? 635 sc->xbd_max_request_size : length; 636 cm->cm_data = virtual; 637 cm->cm_datalen = chunk; 638 cm->cm_operation = BLKIF_OP_WRITE; 639 cm->cm_sector_number = offset / dp->d_sectorsize; 640 cm->cm_complete = xbd_dump_complete; 641 642 xbd_enqueue_cm(cm, XBD_Q_READY); 643 644 length -= chunk; 645 offset += chunk; 646 virtual = (char *) virtual + chunk; 647 } 648 649 /* Tell DOM0 to do the I/O */ 650 xbd_startio(sc); 651 mtx_unlock(&sc->xbd_io_lock); 652 653 /* Poll for the completion. */ 654 xbd_quiesce(sc); /* All quite on the eastern front */ 655 656 /* If there were any errors, bail out... */ 657 while ((cm = xbd_dequeue_cm(sc, XBD_Q_COMPLETE)) != NULL) { 658 if (cm->cm_status != BLKIF_RSP_OKAY) { 659 device_printf(sc->xbd_dev, 660 "Dump I/O failed at sector %jd\n", 661 cm->cm_sector_number); 662 rc = EIO; 663 } 664 xbd_free_command(cm); 665 } 666 667 return (rc); 668 } 669 670 /*----------------------------- Disk Entrypoints -----------------------------*/ 671 static int 672 xbd_open(struct disk *dp) 673 { 674 struct xbd_softc *sc = dp->d_drv1; 675 676 if (sc == NULL) { 677 printf("xbd%d: not found", dp->d_unit); 678 return (ENXIO); 679 } 680 681 sc->xbd_flags |= XBDF_OPEN; 682 sc->xbd_users++; 683 return (0); 684 } 685 686 static int 687 xbd_close(struct disk *dp) 688 { 689 struct xbd_softc *sc = dp->d_drv1; 690 691 if (sc == NULL) 692 return (ENXIO); 693 sc->xbd_flags &= ~XBDF_OPEN; 694 if (--(sc->xbd_users) == 0) { 695 /* 696 * Check whether we have been instructed to close. We will 697 * have ignored this request initially, as the device was 698 * still mounted. 699 */ 700 if (xenbus_get_otherend_state(sc->xbd_dev) == 701 XenbusStateClosing) 702 xbd_closing(sc->xbd_dev); 703 } 704 return (0); 705 } 706 707 static int 708 xbd_ioctl(struct disk *dp, u_long cmd, void *addr, int flag, struct thread *td) 709 { 710 struct xbd_softc *sc = dp->d_drv1; 711 712 if (sc == NULL) 713 return (ENXIO); 714 715 return (ENOTTY); 716 } 717 718 /* 719 * Read/write routine for a buffer. Finds the proper unit, place it on 720 * the sortq and kick the controller. 721 */ 722 static void 723 xbd_strategy(struct bio *bp) 724 { 725 struct xbd_softc *sc = bp->bio_disk->d_drv1; 726 727 /* bogus disk? */ 728 if (sc == NULL) { 729 bp->bio_error = EINVAL; 730 bp->bio_flags |= BIO_ERROR; 731 bp->bio_resid = bp->bio_bcount; 732 biodone(bp); 733 return; 734 } 735 736 /* 737 * Place it in the queue of disk activities for this disk 738 */ 739 mtx_lock(&sc->xbd_io_lock); 740 741 xbd_enqueue_bio(sc, bp); 742 xbd_startio(sc); 743 744 mtx_unlock(&sc->xbd_io_lock); 745 return; 746 } 747 748 /*------------------------------ Ring Management -----------------------------*/ 749 static int 750 xbd_alloc_ring(struct xbd_softc *sc) 751 { 752 blkif_sring_t *sring; 753 uintptr_t sring_page_addr; 754 int error; 755 int i; 756 757 sring = malloc(sc->xbd_ring_pages * PAGE_SIZE, M_XENBLOCKFRONT, 758 M_NOWAIT|M_ZERO); 759 if (sring == NULL) { 760 xenbus_dev_fatal(sc->xbd_dev, ENOMEM, "allocating shared ring"); 761 return (ENOMEM); 762 } 763 SHARED_RING_INIT(sring); 764 FRONT_RING_INIT(&sc->xbd_ring, sring, sc->xbd_ring_pages * PAGE_SIZE); 765 766 for (i = 0, sring_page_addr = (uintptr_t)sring; 767 i < sc->xbd_ring_pages; 768 i++, sring_page_addr += PAGE_SIZE) { 769 error = xenbus_grant_ring(sc->xbd_dev, 770 (vtophys(sring_page_addr) >> PAGE_SHIFT), 771 &sc->xbd_ring_ref[i]); 772 if (error) { 773 xenbus_dev_fatal(sc->xbd_dev, error, 774 "granting ring_ref(%d)", i); 775 return (error); 776 } 777 } 778 if (sc->xbd_ring_pages == 1) { 779 error = xs_printf(XST_NIL, xenbus_get_node(sc->xbd_dev), 780 "ring-ref", "%u", sc->xbd_ring_ref[0]); 781 if (error) { 782 xenbus_dev_fatal(sc->xbd_dev, error, 783 "writing %s/ring-ref", 784 xenbus_get_node(sc->xbd_dev)); 785 return (error); 786 } 787 } else { 788 for (i = 0; i < sc->xbd_ring_pages; i++) { 789 char ring_ref_name[]= "ring_refXX"; 790 791 snprintf(ring_ref_name, sizeof(ring_ref_name), 792 "ring-ref%u", i); 793 error = xs_printf(XST_NIL, xenbus_get_node(sc->xbd_dev), 794 ring_ref_name, "%u", sc->xbd_ring_ref[i]); 795 if (error) { 796 xenbus_dev_fatal(sc->xbd_dev, error, 797 "writing %s/%s", 798 xenbus_get_node(sc->xbd_dev), 799 ring_ref_name); 800 return (error); 801 } 802 } 803 } 804 805 error = xen_intr_alloc_and_bind_local_port(sc->xbd_dev, 806 xenbus_get_otherend_id(sc->xbd_dev), NULL, xbd_int, sc, 807 INTR_TYPE_BIO | INTR_MPSAFE, &sc->xen_intr_handle); 808 if (error) { 809 xenbus_dev_fatal(sc->xbd_dev, error, 810 "xen_intr_alloc_and_bind_local_port failed"); 811 return (error); 812 } 813 814 return (0); 815 } 816 817 static void 818 xbd_free_ring(struct xbd_softc *sc) 819 { 820 int i; 821 822 if (sc->xbd_ring.sring == NULL) 823 return; 824 825 for (i = 0; i < sc->xbd_ring_pages; i++) { 826 if (sc->xbd_ring_ref[i] != GRANT_REF_INVALID) { 827 gnttab_end_foreign_access_ref(sc->xbd_ring_ref[i]); 828 sc->xbd_ring_ref[i] = GRANT_REF_INVALID; 829 } 830 } 831 free(sc->xbd_ring.sring, M_XENBLOCKFRONT); 832 sc->xbd_ring.sring = NULL; 833 } 834 835 /*-------------------------- Initialization/Teardown -------------------------*/ 836 static int 837 xbd_feature_string(struct xbd_softc *sc, char *features, size_t len) 838 { 839 struct sbuf sb; 840 int feature_cnt; 841 842 sbuf_new(&sb, features, len, SBUF_FIXEDLEN); 843 844 feature_cnt = 0; 845 if ((sc->xbd_flags & XBDF_FLUSH) != 0) { 846 sbuf_printf(&sb, "flush"); 847 feature_cnt++; 848 } 849 850 if ((sc->xbd_flags & XBDF_BARRIER) != 0) { 851 if (feature_cnt != 0) 852 sbuf_printf(&sb, ", "); 853 sbuf_printf(&sb, "write_barrier"); 854 feature_cnt++; 855 } 856 857 if ((sc->xbd_flags & XBDF_DISCARD) != 0) { 858 if (feature_cnt != 0) 859 sbuf_printf(&sb, ", "); 860 sbuf_printf(&sb, "discard"); 861 feature_cnt++; 862 } 863 864 if ((sc->xbd_flags & XBDF_PERSISTENT) != 0) { 865 if (feature_cnt != 0) 866 sbuf_printf(&sb, ", "); 867 sbuf_printf(&sb, "persistent_grants"); 868 feature_cnt++; 869 } 870 871 (void) sbuf_finish(&sb); 872 return (sbuf_len(&sb)); 873 } 874 875 static int 876 xbd_sysctl_features(SYSCTL_HANDLER_ARGS) 877 { 878 char features[80]; 879 struct xbd_softc *sc = arg1; 880 int error; 881 int len; 882 883 error = sysctl_wire_old_buffer(req, 0); 884 if (error != 0) 885 return (error); 886 887 len = xbd_feature_string(sc, features, sizeof(features)); 888 889 /* len is -1 on error, which will make the SYSCTL_OUT a no-op. */ 890 return (SYSCTL_OUT(req, features, len + 1/*NUL*/)); 891 } 892 893 static void 894 xbd_setup_sysctl(struct xbd_softc *xbd) 895 { 896 struct sysctl_ctx_list *sysctl_ctx = NULL; 897 struct sysctl_oid *sysctl_tree = NULL; 898 struct sysctl_oid_list *children; 899 900 sysctl_ctx = device_get_sysctl_ctx(xbd->xbd_dev); 901 if (sysctl_ctx == NULL) 902 return; 903 904 sysctl_tree = device_get_sysctl_tree(xbd->xbd_dev); 905 if (sysctl_tree == NULL) 906 return; 907 908 children = SYSCTL_CHILDREN(sysctl_tree); 909 SYSCTL_ADD_UINT(sysctl_ctx, children, OID_AUTO, 910 "max_requests", CTLFLAG_RD, &xbd->xbd_max_requests, -1, 911 "maximum outstanding requests (negotiated)"); 912 913 SYSCTL_ADD_UINT(sysctl_ctx, children, OID_AUTO, 914 "max_request_segments", CTLFLAG_RD, 915 &xbd->xbd_max_request_segments, 0, 916 "maximum number of pages per requests (negotiated)"); 917 918 SYSCTL_ADD_UINT(sysctl_ctx, children, OID_AUTO, 919 "max_request_size", CTLFLAG_RD, &xbd->xbd_max_request_size, 0, 920 "maximum size in bytes of a request (negotiated)"); 921 922 SYSCTL_ADD_UINT(sysctl_ctx, children, OID_AUTO, 923 "ring_pages", CTLFLAG_RD, &xbd->xbd_ring_pages, 0, 924 "communication channel pages (negotiated)"); 925 926 SYSCTL_ADD_PROC(sysctl_ctx, children, OID_AUTO, 927 "features", CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, xbd, 928 0, xbd_sysctl_features, "A", "protocol features (negotiated)"); 929 } 930 931 /* 932 * Translate Linux major/minor to an appropriate name and unit 933 * number. For HVM guests, this allows us to use the same drive names 934 * with blkfront as the emulated drives, easing transition slightly. 935 */ 936 static void 937 xbd_vdevice_to_unit(uint32_t vdevice, int *unit, const char **name) 938 { 939 static struct vdev_info { 940 int major; 941 int shift; 942 int base; 943 const char *name; 944 } info[] = { 945 {3, 6, 0, "ada"}, /* ide0 */ 946 {22, 6, 2, "ada"}, /* ide1 */ 947 {33, 6, 4, "ada"}, /* ide2 */ 948 {34, 6, 6, "ada"}, /* ide3 */ 949 {56, 6, 8, "ada"}, /* ide4 */ 950 {57, 6, 10, "ada"}, /* ide5 */ 951 {88, 6, 12, "ada"}, /* ide6 */ 952 {89, 6, 14, "ada"}, /* ide7 */ 953 {90, 6, 16, "ada"}, /* ide8 */ 954 {91, 6, 18, "ada"}, /* ide9 */ 955 956 {8, 4, 0, "da"}, /* scsi disk0 */ 957 {65, 4, 16, "da"}, /* scsi disk1 */ 958 {66, 4, 32, "da"}, /* scsi disk2 */ 959 {67, 4, 48, "da"}, /* scsi disk3 */ 960 {68, 4, 64, "da"}, /* scsi disk4 */ 961 {69, 4, 80, "da"}, /* scsi disk5 */ 962 {70, 4, 96, "da"}, /* scsi disk6 */ 963 {71, 4, 112, "da"}, /* scsi disk7 */ 964 {128, 4, 128, "da"}, /* scsi disk8 */ 965 {129, 4, 144, "da"}, /* scsi disk9 */ 966 {130, 4, 160, "da"}, /* scsi disk10 */ 967 {131, 4, 176, "da"}, /* scsi disk11 */ 968 {132, 4, 192, "da"}, /* scsi disk12 */ 969 {133, 4, 208, "da"}, /* scsi disk13 */ 970 {134, 4, 224, "da"}, /* scsi disk14 */ 971 {135, 4, 240, "da"}, /* scsi disk15 */ 972 973 {202, 4, 0, "xbd"}, /* xbd */ 974 975 {0, 0, 0, NULL}, 976 }; 977 int major = vdevice >> 8; 978 int minor = vdevice & 0xff; 979 int i; 980 981 if (vdevice & (1 << 28)) { 982 *unit = (vdevice & ((1 << 28) - 1)) >> 8; 983 *name = "xbd"; 984 return; 985 } 986 987 for (i = 0; info[i].major; i++) { 988 if (info[i].major == major) { 989 *unit = info[i].base + (minor >> info[i].shift); 990 *name = info[i].name; 991 return; 992 } 993 } 994 995 *unit = minor >> 4; 996 *name = "xbd"; 997 } 998 999 int 1000 xbd_instance_create(struct xbd_softc *sc, blkif_sector_t sectors, 1001 int vdevice, uint16_t vdisk_info, unsigned long sector_size, 1002 unsigned long phys_sector_size) 1003 { 1004 char features[80]; 1005 int unit, error = 0; 1006 const char *name; 1007 1008 xbd_vdevice_to_unit(vdevice, &unit, &name); 1009 1010 sc->xbd_unit = unit; 1011 1012 if (strcmp(name, "xbd") != 0) 1013 device_printf(sc->xbd_dev, "attaching as %s%d\n", name, unit); 1014 1015 if (xbd_feature_string(sc, features, sizeof(features)) > 0) { 1016 device_printf(sc->xbd_dev, "features: %s\n", 1017 features); 1018 } 1019 1020 sc->xbd_disk = disk_alloc(); 1021 sc->xbd_disk->d_unit = sc->xbd_unit; 1022 sc->xbd_disk->d_open = xbd_open; 1023 sc->xbd_disk->d_close = xbd_close; 1024 sc->xbd_disk->d_ioctl = xbd_ioctl; 1025 sc->xbd_disk->d_strategy = xbd_strategy; 1026 sc->xbd_disk->d_dump = xbd_dump; 1027 sc->xbd_disk->d_name = name; 1028 sc->xbd_disk->d_drv1 = sc; 1029 sc->xbd_disk->d_sectorsize = sector_size; 1030 sc->xbd_disk->d_stripesize = phys_sector_size; 1031 sc->xbd_disk->d_stripeoffset = 0; 1032 1033 sc->xbd_disk->d_mediasize = sectors * sector_size; 1034 sc->xbd_disk->d_maxsize = sc->xbd_max_request_size; 1035 sc->xbd_disk->d_flags = DISKFLAG_UNMAPPED_BIO; 1036 if ((sc->xbd_flags & (XBDF_FLUSH|XBDF_BARRIER)) != 0) { 1037 sc->xbd_disk->d_flags |= DISKFLAG_CANFLUSHCACHE; 1038 device_printf(sc->xbd_dev, 1039 "synchronize cache commands enabled.\n"); 1040 } 1041 disk_create(sc->xbd_disk, DISK_VERSION); 1042 1043 return error; 1044 } 1045 1046 static void 1047 xbd_free(struct xbd_softc *sc) 1048 { 1049 int i; 1050 1051 /* Prevent new requests being issued until we fix things up. */ 1052 mtx_lock(&sc->xbd_io_lock); 1053 sc->xbd_state = XBD_STATE_DISCONNECTED; 1054 mtx_unlock(&sc->xbd_io_lock); 1055 1056 /* Free resources associated with old device channel. */ 1057 xbd_free_ring(sc); 1058 if (sc->xbd_shadow) { 1059 for (i = 0; i < sc->xbd_max_requests; i++) { 1060 struct xbd_command *cm; 1061 1062 cm = &sc->xbd_shadow[i]; 1063 if (cm->cm_sg_refs != NULL) { 1064 free(cm->cm_sg_refs, M_XENBLOCKFRONT); 1065 cm->cm_sg_refs = NULL; 1066 } 1067 1068 if (cm->cm_indirectionpages != NULL) { 1069 gnttab_end_foreign_access_references( 1070 sc->xbd_max_request_indirectpages, 1071 &cm->cm_indirectionrefs[0]); 1072 contigfree(cm->cm_indirectionpages, PAGE_SIZE * 1073 sc->xbd_max_request_indirectpages, 1074 M_XENBLOCKFRONT); 1075 cm->cm_indirectionpages = NULL; 1076 } 1077 1078 bus_dmamap_destroy(sc->xbd_io_dmat, cm->cm_map); 1079 } 1080 free(sc->xbd_shadow, M_XENBLOCKFRONT); 1081 sc->xbd_shadow = NULL; 1082 1083 bus_dma_tag_destroy(sc->xbd_io_dmat); 1084 1085 xbd_initq_cm(sc, XBD_Q_FREE); 1086 xbd_initq_cm(sc, XBD_Q_READY); 1087 xbd_initq_cm(sc, XBD_Q_COMPLETE); 1088 } 1089 1090 xen_intr_unbind(&sc->xen_intr_handle); 1091 1092 } 1093 1094 /*--------------------------- State Change Handlers --------------------------*/ 1095 static void 1096 xbd_initialize(struct xbd_softc *sc) 1097 { 1098 const char *otherend_path; 1099 const char *node_path; 1100 uint32_t max_ring_page_order; 1101 int error; 1102 1103 if (xenbus_get_state(sc->xbd_dev) != XenbusStateInitialising) { 1104 /* Initialization has already been performed. */ 1105 return; 1106 } 1107 1108 /* 1109 * Protocol defaults valid even if negotiation for a 1110 * setting fails. 1111 */ 1112 max_ring_page_order = 0; 1113 sc->xbd_ring_pages = 1; 1114 1115 /* 1116 * Protocol negotiation. 1117 * 1118 * \note xs_gather() returns on the first encountered error, so 1119 * we must use independent calls in order to guarantee 1120 * we don't miss information in a sparsly populated back-end 1121 * tree. 1122 * 1123 * \note xs_scanf() does not update variables for unmatched 1124 * fields. 1125 */ 1126 otherend_path = xenbus_get_otherend_path(sc->xbd_dev); 1127 node_path = xenbus_get_node(sc->xbd_dev); 1128 1129 /* Support both backend schemes for relaying ring page limits. */ 1130 (void)xs_scanf(XST_NIL, otherend_path, 1131 "max-ring-page-order", NULL, "%" PRIu32, 1132 &max_ring_page_order); 1133 sc->xbd_ring_pages = 1 << max_ring_page_order; 1134 (void)xs_scanf(XST_NIL, otherend_path, 1135 "max-ring-pages", NULL, "%" PRIu32, 1136 &sc->xbd_ring_pages); 1137 if (sc->xbd_ring_pages < 1) 1138 sc->xbd_ring_pages = 1; 1139 1140 if (sc->xbd_ring_pages > XBD_MAX_RING_PAGES) { 1141 device_printf(sc->xbd_dev, 1142 "Back-end specified ring-pages of %u " 1143 "limited to front-end limit of %u.\n", 1144 sc->xbd_ring_pages, XBD_MAX_RING_PAGES); 1145 sc->xbd_ring_pages = XBD_MAX_RING_PAGES; 1146 } 1147 1148 if (powerof2(sc->xbd_ring_pages) == 0) { 1149 uint32_t new_page_limit; 1150 1151 new_page_limit = 0x01 << (fls(sc->xbd_ring_pages) - 1); 1152 device_printf(sc->xbd_dev, 1153 "Back-end specified ring-pages of %u " 1154 "is not a power of 2. Limited to %u.\n", 1155 sc->xbd_ring_pages, new_page_limit); 1156 sc->xbd_ring_pages = new_page_limit; 1157 } 1158 1159 sc->xbd_max_requests = 1160 BLKIF_MAX_RING_REQUESTS(sc->xbd_ring_pages * PAGE_SIZE); 1161 if (sc->xbd_max_requests > XBD_MAX_REQUESTS) { 1162 device_printf(sc->xbd_dev, 1163 "Back-end specified max_requests of %u " 1164 "limited to front-end limit of %zu.\n", 1165 sc->xbd_max_requests, XBD_MAX_REQUESTS); 1166 sc->xbd_max_requests = XBD_MAX_REQUESTS; 1167 } 1168 1169 if (xbd_alloc_ring(sc) != 0) 1170 return; 1171 1172 /* Support both backend schemes for relaying ring page limits. */ 1173 if (sc->xbd_ring_pages > 1) { 1174 error = xs_printf(XST_NIL, node_path, 1175 "num-ring-pages","%u", 1176 sc->xbd_ring_pages); 1177 if (error) { 1178 xenbus_dev_fatal(sc->xbd_dev, error, 1179 "writing %s/num-ring-pages", 1180 node_path); 1181 return; 1182 } 1183 1184 error = xs_printf(XST_NIL, node_path, 1185 "ring-page-order", "%u", 1186 fls(sc->xbd_ring_pages) - 1); 1187 if (error) { 1188 xenbus_dev_fatal(sc->xbd_dev, error, 1189 "writing %s/ring-page-order", 1190 node_path); 1191 return; 1192 } 1193 } 1194 1195 error = xs_printf(XST_NIL, node_path, "event-channel", 1196 "%u", xen_intr_port(sc->xen_intr_handle)); 1197 if (error) { 1198 xenbus_dev_fatal(sc->xbd_dev, error, 1199 "writing %s/event-channel", 1200 node_path); 1201 return; 1202 } 1203 1204 error = xs_printf(XST_NIL, node_path, "protocol", 1205 "%s", XEN_IO_PROTO_ABI_NATIVE); 1206 if (error) { 1207 xenbus_dev_fatal(sc->xbd_dev, error, 1208 "writing %s/protocol", 1209 node_path); 1210 return; 1211 } 1212 1213 xenbus_set_state(sc->xbd_dev, XenbusStateInitialised); 1214 } 1215 1216 /* 1217 * Invoked when the backend is finally 'ready' (and has published 1218 * the details about the physical device - #sectors, size, etc). 1219 */ 1220 static void 1221 xbd_connect(struct xbd_softc *sc) 1222 { 1223 device_t dev = sc->xbd_dev; 1224 blkif_sector_t sectors; 1225 unsigned long sector_size, phys_sector_size; 1226 unsigned int binfo; 1227 int err, feature_barrier, feature_flush; 1228 int i, j; 1229 1230 DPRINTK("blkfront.c:connect:%s.\n", xenbus_get_otherend_path(dev)); 1231 1232 if (sc->xbd_state == XBD_STATE_SUSPENDED) { 1233 return; 1234 } 1235 1236 if (sc->xbd_state == XBD_STATE_CONNECTED) { 1237 struct disk *disk; 1238 1239 disk = sc->xbd_disk; 1240 if (disk == NULL) { 1241 return; 1242 } 1243 err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev), 1244 "sectors", "%"PRIu64, §ors, NULL); 1245 if (err != 0) { 1246 xenbus_dev_error(dev, err, 1247 "reading sectors at %s", 1248 xenbus_get_otherend_path(dev)); 1249 return; 1250 } 1251 disk->d_mediasize = disk->d_sectorsize * sectors; 1252 err = disk_resize(disk, M_NOWAIT); 1253 if (err) { 1254 xenbus_dev_error(dev, err, 1255 "unable to resize disk %s%u", 1256 disk->d_name, disk->d_unit); 1257 return; 1258 } 1259 device_printf(sc->xbd_dev, 1260 "changed capacity to %jd\n", 1261 (intmax_t)disk->d_mediasize); 1262 return; 1263 } 1264 1265 err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev), 1266 "sectors", "%"PRIu64, §ors, 1267 "info", "%u", &binfo, 1268 "sector-size", "%lu", §or_size, 1269 NULL); 1270 if (err) { 1271 xenbus_dev_fatal(dev, err, 1272 "reading backend fields at %s", 1273 xenbus_get_otherend_path(dev)); 1274 return; 1275 } 1276 if ((sectors == 0) || (sector_size == 0)) { 1277 xenbus_dev_fatal(dev, 0, 1278 "invalid parameters from %s:" 1279 " sectors = %"PRIu64", sector_size = %lu", 1280 xenbus_get_otherend_path(dev), 1281 sectors, sector_size); 1282 return; 1283 } 1284 err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev), 1285 "physical-sector-size", "%lu", &phys_sector_size, 1286 NULL); 1287 if (err || phys_sector_size <= sector_size) 1288 phys_sector_size = 0; 1289 err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev), 1290 "feature-barrier", "%d", &feature_barrier, 1291 NULL); 1292 if (err == 0 && feature_barrier != 0) 1293 sc->xbd_flags |= XBDF_BARRIER; 1294 1295 err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev), 1296 "feature-flush-cache", "%d", &feature_flush, 1297 NULL); 1298 if (err == 0 && feature_flush != 0) 1299 sc->xbd_flags |= XBDF_FLUSH; 1300 1301 err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev), 1302 "feature-max-indirect-segments", "%" PRIu32, 1303 &sc->xbd_max_request_segments, NULL); 1304 if ((err != 0) || (xbd_enable_indirect == 0)) 1305 sc->xbd_max_request_segments = 0; 1306 if (sc->xbd_max_request_segments > XBD_MAX_INDIRECT_SEGMENTS) 1307 sc->xbd_max_request_segments = XBD_MAX_INDIRECT_SEGMENTS; 1308 if (sc->xbd_max_request_segments > XBD_SIZE_TO_SEGS(maxphys)) 1309 sc->xbd_max_request_segments = XBD_SIZE_TO_SEGS(maxphys); 1310 sc->xbd_max_request_indirectpages = 1311 XBD_INDIRECT_SEGS_TO_PAGES(sc->xbd_max_request_segments); 1312 if (sc->xbd_max_request_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST) 1313 sc->xbd_max_request_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST; 1314 sc->xbd_max_request_size = 1315 XBD_SEGS_TO_SIZE(sc->xbd_max_request_segments); 1316 1317 /* Allocate datastructures based on negotiated values. */ 1318 err = bus_dma_tag_create( 1319 bus_get_dma_tag(sc->xbd_dev), /* parent */ 1320 512, PAGE_SIZE, /* algnmnt, boundary */ 1321 BUS_SPACE_MAXADDR, /* lowaddr */ 1322 BUS_SPACE_MAXADDR, /* highaddr */ 1323 NULL, NULL, /* filter, filterarg */ 1324 sc->xbd_max_request_size, 1325 sc->xbd_max_request_segments, 1326 PAGE_SIZE, /* maxsegsize */ 1327 BUS_DMA_ALLOCNOW, /* flags */ 1328 busdma_lock_mutex, /* lockfunc */ 1329 &sc->xbd_io_lock, /* lockarg */ 1330 &sc->xbd_io_dmat); 1331 if (err != 0) { 1332 xenbus_dev_fatal(sc->xbd_dev, err, 1333 "Cannot allocate parent DMA tag\n"); 1334 return; 1335 } 1336 1337 /* Per-transaction data allocation. */ 1338 sc->xbd_shadow = malloc(sizeof(*sc->xbd_shadow) * sc->xbd_max_requests, 1339 M_XENBLOCKFRONT, M_NOWAIT|M_ZERO); 1340 if (sc->xbd_shadow == NULL) { 1341 bus_dma_tag_destroy(sc->xbd_io_dmat); 1342 xenbus_dev_fatal(sc->xbd_dev, ENOMEM, 1343 "Cannot allocate request structures\n"); 1344 return; 1345 } 1346 1347 for (i = 0; i < sc->xbd_max_requests; i++) { 1348 struct xbd_command *cm; 1349 void * indirectpages; 1350 1351 cm = &sc->xbd_shadow[i]; 1352 cm->cm_sg_refs = malloc( 1353 sizeof(grant_ref_t) * sc->xbd_max_request_segments, 1354 M_XENBLOCKFRONT, M_NOWAIT); 1355 if (cm->cm_sg_refs == NULL) 1356 break; 1357 cm->cm_id = i; 1358 cm->cm_flags = XBDCF_INITIALIZER; 1359 cm->cm_sc = sc; 1360 if (bus_dmamap_create(sc->xbd_io_dmat, 0, &cm->cm_map) != 0) 1361 break; 1362 if (sc->xbd_max_request_indirectpages > 0) { 1363 indirectpages = contigmalloc( 1364 PAGE_SIZE * sc->xbd_max_request_indirectpages, 1365 M_XENBLOCKFRONT, M_ZERO | M_NOWAIT, 0, ~0, 1366 PAGE_SIZE, 0); 1367 if (indirectpages == NULL) 1368 sc->xbd_max_request_indirectpages = 0; 1369 } else { 1370 indirectpages = NULL; 1371 } 1372 for (j = 0; j < sc->xbd_max_request_indirectpages; j++) { 1373 if (gnttab_grant_foreign_access( 1374 xenbus_get_otherend_id(sc->xbd_dev), 1375 (vtophys(indirectpages) >> PAGE_SHIFT) + j, 1376 1 /* grant read-only access */, 1377 &cm->cm_indirectionrefs[j])) 1378 break; 1379 } 1380 if (j < sc->xbd_max_request_indirectpages) { 1381 contigfree(indirectpages, 1382 PAGE_SIZE * sc->xbd_max_request_indirectpages, 1383 M_XENBLOCKFRONT); 1384 break; 1385 } 1386 cm->cm_indirectionpages = indirectpages; 1387 xbd_free_command(cm); 1388 } 1389 1390 if (sc->xbd_disk == NULL) { 1391 device_printf(dev, "%juMB <%s> at %s", 1392 (uintmax_t) sectors / (1048576 / sector_size), 1393 device_get_desc(dev), 1394 xenbus_get_node(dev)); 1395 bus_print_child_footer(device_get_parent(dev), dev); 1396 1397 xbd_instance_create(sc, sectors, sc->xbd_vdevice, binfo, 1398 sector_size, phys_sector_size); 1399 } 1400 1401 (void)xenbus_set_state(dev, XenbusStateConnected); 1402 1403 /* Kick pending requests. */ 1404 mtx_lock(&sc->xbd_io_lock); 1405 sc->xbd_state = XBD_STATE_CONNECTED; 1406 xbd_startio(sc); 1407 sc->xbd_flags |= XBDF_READY; 1408 mtx_unlock(&sc->xbd_io_lock); 1409 } 1410 1411 /** 1412 * Handle the change of state of the backend to Closing. We must delete our 1413 * device-layer structures now, to ensure that writes are flushed through to 1414 * the backend. Once this is done, we can switch to Closed in 1415 * acknowledgement. 1416 */ 1417 static void 1418 xbd_closing(device_t dev) 1419 { 1420 struct xbd_softc *sc = device_get_softc(dev); 1421 1422 xenbus_set_state(dev, XenbusStateClosing); 1423 1424 DPRINTK("xbd_closing: %s removed\n", xenbus_get_node(dev)); 1425 1426 if (sc->xbd_disk != NULL) { 1427 disk_destroy(sc->xbd_disk); 1428 sc->xbd_disk = NULL; 1429 } 1430 1431 xenbus_set_state(dev, XenbusStateClosed); 1432 } 1433 1434 /*---------------------------- NewBus Entrypoints ----------------------------*/ 1435 static int 1436 xbd_probe(device_t dev) 1437 { 1438 if (strcmp(xenbus_get_type(dev), "vbd") != 0) 1439 return (ENXIO); 1440 1441 if (xen_pv_disks_disabled()) 1442 return (ENXIO); 1443 1444 if (xen_hvm_domain()) { 1445 int error; 1446 char *type; 1447 1448 /* 1449 * When running in an HVM domain, IDE disk emulation is 1450 * disabled early in boot so that native drivers will 1451 * not see emulated hardware. However, CDROM device 1452 * emulation cannot be disabled. 1453 * 1454 * Through use of FreeBSD's vm_guest and xen_hvm_domain() 1455 * APIs, we could modify the native CDROM driver to fail its 1456 * probe when running under Xen. Unfortunatlely, the PV 1457 * CDROM support in XenServer (up through at least version 1458 * 6.2) isn't functional, so we instead rely on the emulated 1459 * CDROM instance, and fail to attach the PV one here in 1460 * the blkfront driver. 1461 */ 1462 error = xs_read(XST_NIL, xenbus_get_node(dev), 1463 "device-type", NULL, (void **) &type); 1464 if (error) 1465 return (ENXIO); 1466 1467 if (strncmp(type, "cdrom", 5) == 0) { 1468 free(type, M_XENSTORE); 1469 return (ENXIO); 1470 } 1471 free(type, M_XENSTORE); 1472 } 1473 1474 device_set_desc(dev, "Virtual Block Device"); 1475 device_quiet(dev); 1476 return (0); 1477 } 1478 1479 /* 1480 * Setup supplies the backend dir, virtual device. We place an event 1481 * channel and shared frame entries. We watch backend to wait if it's 1482 * ok. 1483 */ 1484 static int 1485 xbd_attach(device_t dev) 1486 { 1487 struct xbd_softc *sc; 1488 const char *name; 1489 uint32_t vdevice; 1490 int error; 1491 int i; 1492 int unit; 1493 1494 /* FIXME: Use dynamic device id if this is not set. */ 1495 error = xs_scanf(XST_NIL, xenbus_get_node(dev), 1496 "virtual-device", NULL, "%" PRIu32, &vdevice); 1497 if (error) 1498 error = xs_scanf(XST_NIL, xenbus_get_node(dev), 1499 "virtual-device-ext", NULL, "%" PRIu32, &vdevice); 1500 if (error) { 1501 xenbus_dev_fatal(dev, error, "reading virtual-device"); 1502 device_printf(dev, "Couldn't determine virtual device.\n"); 1503 return (error); 1504 } 1505 1506 xbd_vdevice_to_unit(vdevice, &unit, &name); 1507 if (!strcmp(name, "xbd")) 1508 device_set_unit(dev, unit); 1509 1510 sc = device_get_softc(dev); 1511 mtx_init(&sc->xbd_io_lock, "blkfront i/o lock", NULL, MTX_DEF); 1512 xbd_initqs(sc); 1513 for (i = 0; i < XBD_MAX_RING_PAGES; i++) 1514 sc->xbd_ring_ref[i] = GRANT_REF_INVALID; 1515 1516 sc->xbd_dev = dev; 1517 sc->xbd_vdevice = vdevice; 1518 sc->xbd_state = XBD_STATE_DISCONNECTED; 1519 1520 xbd_setup_sysctl(sc); 1521 1522 /* Wait for backend device to publish its protocol capabilities. */ 1523 xenbus_set_state(dev, XenbusStateInitialising); 1524 1525 return (0); 1526 } 1527 1528 static int 1529 xbd_detach(device_t dev) 1530 { 1531 struct xbd_softc *sc = device_get_softc(dev); 1532 1533 DPRINTK("%s: %s removed\n", __func__, xenbus_get_node(dev)); 1534 1535 xbd_free(sc); 1536 mtx_destroy(&sc->xbd_io_lock); 1537 1538 return 0; 1539 } 1540 1541 static int 1542 xbd_suspend(device_t dev) 1543 { 1544 struct xbd_softc *sc = device_get_softc(dev); 1545 int retval; 1546 int saved_state; 1547 1548 /* Prevent new requests being issued until we fix things up. */ 1549 mtx_lock(&sc->xbd_io_lock); 1550 saved_state = sc->xbd_state; 1551 sc->xbd_state = XBD_STATE_SUSPENDED; 1552 1553 /* Wait for outstanding I/O to drain. */ 1554 retval = 0; 1555 while (xbd_queue_length(sc, XBD_Q_BUSY) != 0) { 1556 if (msleep(&sc->xbd_cm_q[XBD_Q_BUSY], &sc->xbd_io_lock, 1557 PRIBIO, "blkf_susp", 30 * hz) == EWOULDBLOCK) { 1558 retval = EBUSY; 1559 break; 1560 } 1561 } 1562 mtx_unlock(&sc->xbd_io_lock); 1563 1564 if (retval != 0) 1565 sc->xbd_state = saved_state; 1566 1567 return (retval); 1568 } 1569 1570 static int 1571 xbd_resume(device_t dev) 1572 { 1573 struct xbd_softc *sc = device_get_softc(dev); 1574 1575 if (xen_suspend_cancelled) { 1576 sc->xbd_state = XBD_STATE_CONNECTED; 1577 return (0); 1578 } 1579 1580 DPRINTK("xbd_resume: %s\n", xenbus_get_node(dev)); 1581 1582 xbd_free(sc); 1583 xbd_initialize(sc); 1584 return (0); 1585 } 1586 1587 /** 1588 * Callback received when the backend's state changes. 1589 */ 1590 static void 1591 xbd_backend_changed(device_t dev, XenbusState backend_state) 1592 { 1593 struct xbd_softc *sc = device_get_softc(dev); 1594 1595 DPRINTK("backend_state=%d\n", backend_state); 1596 1597 switch (backend_state) { 1598 case XenbusStateUnknown: 1599 case XenbusStateInitialising: 1600 case XenbusStateReconfigured: 1601 case XenbusStateReconfiguring: 1602 case XenbusStateClosed: 1603 break; 1604 1605 case XenbusStateInitWait: 1606 case XenbusStateInitialised: 1607 xbd_initialize(sc); 1608 break; 1609 1610 case XenbusStateConnected: 1611 xbd_initialize(sc); 1612 xbd_connect(sc); 1613 break; 1614 1615 case XenbusStateClosing: 1616 if (sc->xbd_users > 0) { 1617 device_printf(dev, "detaching with pending users\n"); 1618 KASSERT(sc->xbd_disk != NULL, 1619 ("NULL disk with pending users\n")); 1620 disk_gone(sc->xbd_disk); 1621 } else { 1622 xbd_closing(dev); 1623 } 1624 break; 1625 } 1626 } 1627 1628 /*---------------------------- NewBus Registration ---------------------------*/ 1629 static device_method_t xbd_methods[] = { 1630 /* Device interface */ 1631 DEVMETHOD(device_probe, xbd_probe), 1632 DEVMETHOD(device_attach, xbd_attach), 1633 DEVMETHOD(device_detach, xbd_detach), 1634 DEVMETHOD(device_shutdown, bus_generic_shutdown), 1635 DEVMETHOD(device_suspend, xbd_suspend), 1636 DEVMETHOD(device_resume, xbd_resume), 1637 1638 /* Xenbus interface */ 1639 DEVMETHOD(xenbus_otherend_changed, xbd_backend_changed), 1640 1641 { 0, 0 } 1642 }; 1643 1644 static driver_t xbd_driver = { 1645 "xbd", 1646 xbd_methods, 1647 sizeof(struct xbd_softc), 1648 }; 1649 devclass_t xbd_devclass; 1650 1651 DRIVER_MODULE(xbd, xenbusb_front, xbd_driver, xbd_devclass, 0, 0); 1652