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