sys/dev/xen/blkfront/blkfront.c

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