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