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