drivers/block/sunvdc.c

   1 /* sunvdc.c: Sun LDOM Virtual Disk Client.
   2  *
   3  * Copyright (C) 2007, 2008 David S. Miller <davem@davemloft.net>
   4  */
   5
   6 #include <linux/module.h>
   7 #include <linux/kernel.h>
   8 #include <linux/types.h>
   9 #include <linux/blkdev.h>
  10 #include <linux/hdreg.h>
  11 #include <linux/genhd.h>
  12 #include <linux/cdrom.h>
  13 #include <linux/slab.h>
  14 #include <linux/spinlock.h>
  15 #include <linux/completion.h>
  16 #include <linux/delay.h>
  17 #include <linux/init.h>
  18 #include <linux/list.h>
  19 #include <linux/scatterlist.h>
  20
  21 #include <asm/vio.h>
  22 #include <asm/ldc.h>
  23
  24 #define DRV_MODULE_NAME         "sunvdc"
  25 #define PFX DRV_MODULE_NAME     ": "
  26 #define DRV_MODULE_VERSION      "1.2"
  27 #define DRV_MODULE_RELDATE      "November 24, 2014"
  28
  29 static char version[] =
  30         DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
  31 MODULE_AUTHOR("David S. Miller (davem@davemloft.net)");
  32 MODULE_DESCRIPTION("Sun LDOM virtual disk client driver");
  33 MODULE_LICENSE("GPL");
  34 MODULE_VERSION(DRV_MODULE_VERSION);
  35
  36 #define VDC_TX_RING_SIZE        512
  37
  38 #define WAITING_FOR_LINK_UP     0x01
  39 #define WAITING_FOR_TX_SPACE    0x02
  40 #define WAITING_FOR_GEN_CMD     0x04
  41 #define WAITING_FOR_ANY         -1
  42
  43 #define VDC_MAX_RETRIES 10
  44
  45 static struct workqueue_struct *sunvdc_wq;
  46
  47 struct vdc_req_entry {
  48         struct request          *req;
  49 };
  50
  51 struct vdc_port {
  52         struct vio_driver_state vio;
  53
  54         struct gendisk          *disk;
  55
  56         struct vdc_completion   *cmp;
  57
  58         u64                     req_id;
  59         u64                     seq;
  60         struct vdc_req_entry    rq_arr[VDC_TX_RING_SIZE];
  61
  62         unsigned long           ring_cookies;
  63
  64         u64                     max_xfer_size;
  65         u32                     vdisk_block_size;
  66
  67         u64                     ldc_timeout;
  68         struct timer_list       ldc_reset_timer;
  69         struct work_struct      ldc_reset_work;
  70
  71         /* The server fills these in for us in the disk attribute
  72          * ACK packet.
  73          */
  74         u64                     operations;
  75         u32                     vdisk_size;
  76         u8                      vdisk_type;
  77         u8                      vdisk_mtype;
  78
  79         char                    disk_name[32];
  80 };
  81
  82 static void vdc_ldc_reset(struct vdc_port *port);
  83 static void vdc_ldc_reset_work(struct work_struct *work);
  84 static void vdc_ldc_reset_timer(unsigned long _arg);
  85
  86 static inline struct vdc_port *to_vdc_port(struct vio_driver_state *vio)
  87 {
  88         return container_of(vio, struct vdc_port, vio);
  89 }
  90
  91 /* Ordered from largest major to lowest */
  92 static struct vio_version vdc_versions[] = {
  93         { .major = 1, .minor = 1 },
  94         { .major = 1, .minor = 0 },
  95 };
  96
  97 static inline int vdc_version_supported(struct vdc_port *port,
  98                                         u16 major, u16 minor)
  99 {
 100         return port->vio.ver.major == major && port->vio.ver.minor >= minor;
 101 }
 102
 103 #define VDCBLK_NAME     "vdisk"
 104 static int vdc_major;
 105 #define PARTITION_SHIFT 3
 106
 107 static inline u32 vdc_tx_dring_avail(struct vio_dring_state *dr)
 108 {
 109         return vio_dring_avail(dr, VDC_TX_RING_SIZE);
 110 }
 111
 112 static int vdc_getgeo(struct block_device *bdev, struct hd_geometry *geo)
 113 {
 114         struct gendisk *disk = bdev->bd_disk;
 115         sector_t nsect = get_capacity(disk);
 116         sector_t cylinders = nsect;
 117
 118         geo->heads = 0xff;
 119         geo->sectors = 0x3f;
 120         sector_div(cylinders, geo->heads * geo->sectors);
 121         geo->cylinders = cylinders;
 122         if ((sector_t)(geo->cylinders + 1) * geo->heads * geo->sectors < nsect)
 123                 geo->cylinders = 0xffff;
 124
 125         return 0;
 126 }
 127
 128 /* Add ioctl/CDROM_GET_CAPABILITY to support cdrom_id in udev
 129  * when vdisk_mtype is VD_MEDIA_TYPE_CD or VD_MEDIA_TYPE_DVD.
 130  * Needed to be able to install inside an ldom from an iso image.
 131  */
 132 static int vdc_ioctl(struct block_device *bdev, fmode_t mode,
 133                      unsigned command, unsigned long argument)
 134 {
 135         int i;
 136         struct gendisk *disk;
 137
 138         switch (command) {
 139         case CDROMMULTISESSION:
 140                 pr_debug(PFX "Multisession CDs not supported\n");
 141                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 142                         if (put_user(0, (char __user *)(argument + i)))
 143                                 return -EFAULT;
 144                 return 0;
 145
 146         case CDROM_GET_CAPABILITY:
 147                 disk = bdev->bd_disk;
 148
 149                 if (bdev->bd_disk && (disk->flags & GENHD_FL_CD))
 150                         return 0;
 151                 return -EINVAL;
 152
 153         default:
 154                 pr_debug(PFX "ioctl %08x not supported\n", command);
 155                 return -EINVAL;
 156         }
 157 }
 158
 159 static const struct block_device_operations vdc_fops = {
 160         .owner          = THIS_MODULE,
 161         .getgeo         = vdc_getgeo,
 162         .ioctl          = vdc_ioctl,
 163 };
 164
 165 static void vdc_blk_queue_start(struct vdc_port *port)
 166 {
 167         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 168
 169         /* restart blk queue when ring is half emptied. also called after
 170          * handshake completes, so check for initial handshake before we've
 171          * allocated a disk.
 172          */
 173         if (port->disk && blk_queue_stopped(port->disk->queue) &&
 174             vdc_tx_dring_avail(dr) * 100 / VDC_TX_RING_SIZE >= 50) {
 175                 blk_start_queue(port->disk->queue);
 176         }
 177
 178 }
 179
 180 static void vdc_finish(struct vio_driver_state *vio, int err, int waiting_for)
 181 {
 182         if (vio->cmp &&
 183             (waiting_for == -1 ||
 184              vio->cmp->waiting_for == waiting_for)) {
 185                 vio->cmp->err = err;
 186                 complete(&vio->cmp->com);
 187                 vio->cmp = NULL;
 188         }
 189 }
 190
 191 static void vdc_handshake_complete(struct vio_driver_state *vio)
 192 {
 193         struct vdc_port *port = to_vdc_port(vio);
 194
 195         del_timer(&port->ldc_reset_timer);
 196         vdc_finish(vio, 0, WAITING_FOR_LINK_UP);
 197         vdc_blk_queue_start(port);
 198 }
 199
 200 static int vdc_handle_unknown(struct vdc_port *port, void *arg)
 201 {
 202         struct vio_msg_tag *pkt = arg;
 203
 204         printk(KERN_ERR PFX "Received unknown msg [%02x:%02x:%04x:%08x]\n",
 205                pkt->type, pkt->stype, pkt->stype_env, pkt->sid);
 206         printk(KERN_ERR PFX "Resetting connection.\n");
 207
 208         ldc_disconnect(port->vio.lp);
 209
 210         return -ECONNRESET;
 211 }
 212
 213 static int vdc_send_attr(struct vio_driver_state *vio)
 214 {
 215         struct vdc_port *port = to_vdc_port(vio);
 216         struct vio_disk_attr_info pkt;
 217
 218         memset(&pkt, 0, sizeof(pkt));
 219
 220         pkt.tag.type = VIO_TYPE_CTRL;
 221         pkt.tag.stype = VIO_SUBTYPE_INFO;
 222         pkt.tag.stype_env = VIO_ATTR_INFO;
 223         pkt.tag.sid = vio_send_sid(vio);
 224
 225         pkt.xfer_mode = VIO_DRING_MODE;
 226         pkt.vdisk_block_size = port->vdisk_block_size;
 227         pkt.max_xfer_size = port->max_xfer_size;
 228
 229         viodbg(HS, "SEND ATTR xfer_mode[0x%x] blksz[%u] max_xfer[%llu]\n",
 230                pkt.xfer_mode, pkt.vdisk_block_size, pkt.max_xfer_size);
 231
 232         return vio_ldc_send(&port->vio, &pkt, sizeof(pkt));
 233 }
 234
 235 static int vdc_handle_attr(struct vio_driver_state *vio, void *arg)
 236 {
 237         struct vdc_port *port = to_vdc_port(vio);
 238         struct vio_disk_attr_info *pkt = arg;
 239
 240         viodbg(HS, "GOT ATTR stype[0x%x] ops[%llx] disk_size[%llu] disk_type[%x] "
 241                "mtype[0x%x] xfer_mode[0x%x] blksz[%u] max_xfer[%llu]\n",
 242                pkt->tag.stype, pkt->operations,
 243                pkt->vdisk_size, pkt->vdisk_type, pkt->vdisk_mtype,
 244                pkt->xfer_mode, pkt->vdisk_block_size,
 245                pkt->max_xfer_size);
 246
 247         if (pkt->tag.stype == VIO_SUBTYPE_ACK) {
 248                 switch (pkt->vdisk_type) {
 249                 case VD_DISK_TYPE_DISK:
 250                 case VD_DISK_TYPE_SLICE:
 251                         break;
 252
 253                 default:
 254                         printk(KERN_ERR PFX "%s: Bogus vdisk_type 0x%x\n",
 255                                vio->name, pkt->vdisk_type);
 256                         return -ECONNRESET;
 257                 }
 258
 259                 if (pkt->vdisk_block_size > port->vdisk_block_size) {
 260                         printk(KERN_ERR PFX "%s: BLOCK size increased "
 261                                "%u --> %u\n",
 262                                vio->name,
 263                                port->vdisk_block_size, pkt->vdisk_block_size);
 264                         return -ECONNRESET;
 265                 }
 266
 267                 port->operations = pkt->operations;
 268                 port->vdisk_type = pkt->vdisk_type;
 269                 if (vdc_version_supported(port, 1, 1)) {
 270                         port->vdisk_size = pkt->vdisk_size;
 271                         port->vdisk_mtype = pkt->vdisk_mtype;
 272                 }
 273                 if (pkt->max_xfer_size < port->max_xfer_size)
 274                         port->max_xfer_size = pkt->max_xfer_size;
 275                 port->vdisk_block_size = pkt->vdisk_block_size;
 276                 return 0;
 277         } else {
 278                 printk(KERN_ERR PFX "%s: Attribute NACK\n", vio->name);
 279
 280                 return -ECONNRESET;
 281         }
 282 }
 283
 284 static void vdc_end_special(struct vdc_port *port, struct vio_disk_desc *desc)
 285 {
 286         int err = desc->status;
 287
 288         vdc_finish(&port->vio, -err, WAITING_FOR_GEN_CMD);
 289 }
 290
 291 static void vdc_end_one(struct vdc_port *port, struct vio_dring_state *dr,
 292                         unsigned int index)
 293 {
 294         struct vio_disk_desc *desc = vio_dring_entry(dr, index);
 295         struct vdc_req_entry *rqe = &port->rq_arr[index];
 296         struct request *req;
 297
 298         if (unlikely(desc->hdr.state != VIO_DESC_DONE))
 299                 return;
 300
 301         ldc_unmap(port->vio.lp, desc->cookies, desc->ncookies);
 302         desc->hdr.state = VIO_DESC_FREE;
 303         dr->cons = vio_dring_next(dr, index);
 304
 305         req = rqe->req;
 306         if (req == NULL) {
 307                 vdc_end_special(port, desc);
 308                 return;
 309         }
 310
 311         rqe->req = NULL;
 312
 313         __blk_end_request(req, (desc->status ? -EIO : 0), desc->size);
 314
 315         vdc_blk_queue_start(port);
 316 }
 317
 318 static int vdc_ack(struct vdc_port *port, void *msgbuf)
 319 {
 320         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 321         struct vio_dring_data *pkt = msgbuf;
 322
 323         if (unlikely(pkt->dring_ident != dr->ident ||
 324                      pkt->start_idx != pkt->end_idx ||
 325                      pkt->start_idx >= VDC_TX_RING_SIZE))
 326                 return 0;
 327
 328         vdc_end_one(port, dr, pkt->start_idx);
 329
 330         return 0;
 331 }
 332
 333 static int vdc_nack(struct vdc_port *port, void *msgbuf)
 334 {
 335         /* XXX Implement me XXX */
 336         return 0;
 337 }
 338
 339 static void vdc_event(void *arg, int event)
 340 {
 341         struct vdc_port *port = arg;
 342         struct vio_driver_state *vio = &port->vio;
 343         unsigned long flags;
 344         int err;
 345
 346         spin_lock_irqsave(&vio->lock, flags);
 347
 348         if (unlikely(event == LDC_EVENT_RESET)) {
 349                 vio_link_state_change(vio, event);
 350                 queue_work(sunvdc_wq, &port->ldc_reset_work);
 351                 goto out;
 352         }
 353
 354         if (unlikely(event == LDC_EVENT_UP)) {
 355                 vio_link_state_change(vio, event);
 356                 goto out;
 357         }
 358
 359         if (unlikely(event != LDC_EVENT_DATA_READY)) {
 360                 pr_warn(PFX "Unexpected LDC event %d\n", event);
 361                 goto out;
 362         }
 363
 364         err = 0;
 365         while (1) {
 366                 union {
 367                         struct vio_msg_tag tag;
 368                         u64 raw[8];
 369                 } msgbuf;
 370
 371                 err = ldc_read(vio->lp, &msgbuf, sizeof(msgbuf));
 372                 if (unlikely(err < 0)) {
 373                         if (err == -ECONNRESET)
 374                                 vio_conn_reset(vio);
 375                         break;
 376                 }
 377                 if (err == 0)
 378                         break;
 379                 viodbg(DATA, "TAG [%02x:%02x:%04x:%08x]\n",
 380                        msgbuf.tag.type,
 381                        msgbuf.tag.stype,
 382                        msgbuf.tag.stype_env,
 383                        msgbuf.tag.sid);
 384                 err = vio_validate_sid(vio, &msgbuf.tag);
 385                 if (err < 0)
 386                         break;
 387
 388                 if (likely(msgbuf.tag.type == VIO_TYPE_DATA)) {
 389                         if (msgbuf.tag.stype == VIO_SUBTYPE_ACK)
 390                                 err = vdc_ack(port, &msgbuf);
 391                         else if (msgbuf.tag.stype == VIO_SUBTYPE_NACK)
 392                                 err = vdc_nack(port, &msgbuf);
 393                         else
 394                                 err = vdc_handle_unknown(port, &msgbuf);
 395                 } else if (msgbuf.tag.type == VIO_TYPE_CTRL) {
 396                         err = vio_control_pkt_engine(vio, &msgbuf);
 397                 } else {
 398                         err = vdc_handle_unknown(port, &msgbuf);
 399                 }
 400                 if (err < 0)
 401                         break;
 402         }
 403         if (err < 0)
 404                 vdc_finish(&port->vio, err, WAITING_FOR_ANY);
 405 out:
 406         spin_unlock_irqrestore(&vio->lock, flags);
 407 }
 408
 409 static int __vdc_tx_trigger(struct vdc_port *port)
 410 {
 411         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 412         struct vio_dring_data hdr = {
 413                 .tag = {
 414                         .type           = VIO_TYPE_DATA,
 415                         .stype          = VIO_SUBTYPE_INFO,
 416                         .stype_env      = VIO_DRING_DATA,
 417                         .sid            = vio_send_sid(&port->vio),
 418                 },
 419                 .dring_ident            = dr->ident,
 420                 .start_idx              = dr->prod,
 421                 .end_idx                = dr->prod,
 422         };
 423         int err, delay;
 424         int retries = 0;
 425
 426         hdr.seq = dr->snd_nxt;
 427         delay = 1;
 428         do {
 429                 err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
 430                 if (err > 0) {
 431                         dr->snd_nxt++;
 432                         break;
 433                 }
 434                 udelay(delay);
 435                 if ((delay <<= 1) > 128)
 436                         delay = 128;
 437                 if (retries++ > VDC_MAX_RETRIES)
 438                         break;
 439         } while (err == -EAGAIN);
 440
 441         if (err == -ENOTCONN)
 442                 vdc_ldc_reset(port);
 443         return err;
 444 }
 445
 446 static int __send_request(struct request *req)
 447 {
 448         struct vdc_port *port = req->rq_disk->private_data;
 449         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 450         struct scatterlist sg[port->ring_cookies];
 451         struct vdc_req_entry *rqe;
 452         struct vio_disk_desc *desc;
 453         unsigned int map_perm;
 454         int nsg, err, i;
 455         u64 len;
 456         u8 op;
 457
 458         map_perm = LDC_MAP_SHADOW | LDC_MAP_DIRECT | LDC_MAP_IO;
 459
 460         if (rq_data_dir(req) == READ) {
 461                 map_perm |= LDC_MAP_W;
 462                 op = VD_OP_BREAD;
 463         } else {
 464                 map_perm |= LDC_MAP_R;
 465                 op = VD_OP_BWRITE;
 466         }
 467
 468         sg_init_table(sg, port->ring_cookies);
 469         nsg = blk_rq_map_sg(req->q, req, sg);
 470
 471         len = 0;
 472         for (i = 0; i < nsg; i++)
 473                 len += sg[i].length;
 474
 475         desc = vio_dring_cur(dr);
 476
 477         err = ldc_map_sg(port->vio.lp, sg, nsg,
 478                          desc->cookies, port->ring_cookies,
 479                          map_perm);
 480         if (err < 0) {
 481                 printk(KERN_ERR PFX "ldc_map_sg() failure, err=%d.\n", err);
 482                 return err;
 483         }
 484
 485         rqe = &port->rq_arr[dr->prod];
 486         rqe->req = req;
 487
 488         desc->hdr.ack = VIO_ACK_ENABLE;
 489         desc->req_id = port->req_id;
 490         desc->operation = op;
 491         if (port->vdisk_type == VD_DISK_TYPE_DISK) {
 492                 desc->slice = 0xff;
 493         } else {
 494                 desc->slice = 0;
 495         }
 496         desc->status = ~0;
 497         desc->offset = (blk_rq_pos(req) << 9) / port->vdisk_block_size;
 498         desc->size = len;
 499         desc->ncookies = err;
 500
 501         /* This has to be a non-SMP write barrier because we are writing
 502          * to memory which is shared with the peer LDOM.
 503          */
 504         wmb();
 505         desc->hdr.state = VIO_DESC_READY;
 506
 507         err = __vdc_tx_trigger(port);
 508         if (err < 0) {
 509                 printk(KERN_ERR PFX "vdc_tx_trigger() failure, err=%d\n", err);
 510         } else {
 511                 port->req_id++;
 512                 dr->prod = vio_dring_next(dr, dr->prod);
 513         }
 514
 515         return err;
 516 }
 517
 518 static void do_vdc_request(struct request_queue *rq)
 519 {
 520         struct request *req;
 521
 522         while ((req = blk_peek_request(rq)) != NULL) {
 523                 struct vdc_port *port;
 524                 struct vio_dring_state *dr;
 525
 526                 port = req->rq_disk->private_data;
 527                 dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 528                 if (unlikely(vdc_tx_dring_avail(dr) < 1))
 529                         goto wait;
 530
 531                 blk_start_request(req);
 532
 533                 if (__send_request(req) < 0) {
 534                         blk_requeue_request(rq, req);
 535 wait:
 536                         /* Avoid pointless unplugs. */
 537                         blk_stop_queue(rq);
 538                         break;
 539                 }
 540         }
 541 }
 542
 543 static int generic_request(struct vdc_port *port, u8 op, void *buf, int len)
 544 {
 545         struct vio_dring_state *dr;
 546         struct vio_completion comp;
 547         struct vio_disk_desc *desc;
 548         unsigned int map_perm;
 549         unsigned long flags;
 550         int op_len, err;
 551         void *req_buf;
 552
 553         if (!(((u64)1 << (u64)op) & port->operations))
 554                 return -EOPNOTSUPP;
 555
 556         switch (op) {
 557         case VD_OP_BREAD:
 558         case VD_OP_BWRITE:
 559         default:
 560                 return -EINVAL;
 561
 562         case VD_OP_FLUSH:
 563                 op_len = 0;
 564                 map_perm = 0;
 565                 break;
 566
 567         case VD_OP_GET_WCE:
 568                 op_len = sizeof(u32);
 569                 map_perm = LDC_MAP_W;
 570                 break;
 571
 572         case VD_OP_SET_WCE:
 573                 op_len = sizeof(u32);
 574                 map_perm = LDC_MAP_R;
 575                 break;
 576
 577         case VD_OP_GET_VTOC:
 578                 op_len = sizeof(struct vio_disk_vtoc);
 579                 map_perm = LDC_MAP_W;
 580                 break;
 581
 582         case VD_OP_SET_VTOC:
 583                 op_len = sizeof(struct vio_disk_vtoc);
 584                 map_perm = LDC_MAP_R;
 585                 break;
 586
 587         case VD_OP_GET_DISKGEOM:
 588                 op_len = sizeof(struct vio_disk_geom);
 589                 map_perm = LDC_MAP_W;
 590                 break;
 591
 592         case VD_OP_SET_DISKGEOM:
 593                 op_len = sizeof(struct vio_disk_geom);
 594                 map_perm = LDC_MAP_R;
 595                 break;
 596
 597         case VD_OP_SCSICMD:
 598                 op_len = 16;
 599                 map_perm = LDC_MAP_RW;
 600                 break;
 601
 602         case VD_OP_GET_DEVID:
 603                 op_len = sizeof(struct vio_disk_devid);
 604                 map_perm = LDC_MAP_W;
 605                 break;
 606
 607         case VD_OP_GET_EFI:
 608         case VD_OP_SET_EFI:
 609                 return -EOPNOTSUPP;
 610                 break;
 611         };
 612
 613         map_perm |= LDC_MAP_SHADOW | LDC_MAP_DIRECT | LDC_MAP_IO;
 614
 615         op_len = (op_len + 7) & ~7;
 616         req_buf = kzalloc(op_len, GFP_KERNEL);
 617         if (!req_buf)
 618                 return -ENOMEM;
 619
 620         if (len > op_len)
 621                 len = op_len;
 622
 623         if (map_perm & LDC_MAP_R)
 624                 memcpy(req_buf, buf, len);
 625
 626         spin_lock_irqsave(&port->vio.lock, flags);
 627
 628         dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 629
 630         /* XXX If we want to use this code generically we have to
 631          * XXX handle TX ring exhaustion etc.
 632          */
 633         desc = vio_dring_cur(dr);
 634
 635         err = ldc_map_single(port->vio.lp, req_buf, op_len,
 636                              desc->cookies, port->ring_cookies,
 637                              map_perm);
 638         if (err < 0) {
 639                 spin_unlock_irqrestore(&port->vio.lock, flags);
 640                 kfree(req_buf);
 641                 return err;
 642         }
 643
 644         init_completion(&comp.com);
 645         comp.waiting_for = WAITING_FOR_GEN_CMD;
 646         port->vio.cmp = &comp;
 647
 648         desc->hdr.ack = VIO_ACK_ENABLE;
 649         desc->req_id = port->req_id;
 650         desc->operation = op;
 651         desc->slice = 0;
 652         desc->status = ~0;
 653         desc->offset = 0;
 654         desc->size = op_len;
 655         desc->ncookies = err;
 656
 657         /* This has to be a non-SMP write barrier because we are writing
 658          * to memory which is shared with the peer LDOM.
 659          */
 660         wmb();
 661         desc->hdr.state = VIO_DESC_READY;
 662
 663         err = __vdc_tx_trigger(port);
 664         if (err >= 0) {
 665                 port->req_id++;
 666                 dr->prod = vio_dring_next(dr, dr->prod);
 667                 spin_unlock_irqrestore(&port->vio.lock, flags);
 668
 669                 wait_for_completion(&comp.com);
 670                 err = comp.err;
 671         } else {
 672                 port->vio.cmp = NULL;
 673                 spin_unlock_irqrestore(&port->vio.lock, flags);
 674         }
 675
 676         if (map_perm & LDC_MAP_W)
 677                 memcpy(buf, req_buf, len);
 678
 679         kfree(req_buf);
 680
 681         return err;
 682 }
 683
 684 static int vdc_alloc_tx_ring(struct vdc_port *port)
 685 {
 686         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 687         unsigned long len, entry_size;
 688         int ncookies;
 689         void *dring;
 690
 691         entry_size = sizeof(struct vio_disk_desc) +
 692                 (sizeof(struct ldc_trans_cookie) * port->ring_cookies);
 693         len = (VDC_TX_RING_SIZE * entry_size);
 694
 695         ncookies = VIO_MAX_RING_COOKIES;
 696         dring = ldc_alloc_exp_dring(port->vio.lp, len,
 697                                     dr->cookies, &ncookies,
 698                                     (LDC_MAP_SHADOW |
 699                                      LDC_MAP_DIRECT |
 700                                      LDC_MAP_RW));
 701         if (IS_ERR(dring))
 702                 return PTR_ERR(dring);
 703
 704         dr->base = dring;
 705         dr->entry_size = entry_size;
 706         dr->num_entries = VDC_TX_RING_SIZE;
 707         dr->prod = dr->cons = 0;
 708         dr->pending = VDC_TX_RING_SIZE;
 709         dr->ncookies = ncookies;
 710
 711         return 0;
 712 }
 713
 714 static void vdc_free_tx_ring(struct vdc_port *port)
 715 {
 716         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 717
 718         if (dr->base) {
 719                 ldc_free_exp_dring(port->vio.lp, dr->base,
 720                                    (dr->entry_size * dr->num_entries),
 721                                    dr->cookies, dr->ncookies);
 722                 dr->base = NULL;
 723                 dr->entry_size = 0;
 724                 dr->num_entries = 0;
 725                 dr->pending = 0;
 726                 dr->ncookies = 0;
 727         }
 728 }
 729
 730 static int vdc_port_up(struct vdc_port *port)
 731 {
 732         struct vio_completion comp;
 733
 734         init_completion(&comp.com);
 735         comp.err = 0;
 736         comp.waiting_for = WAITING_FOR_LINK_UP;
 737         port->vio.cmp = &comp;
 738
 739         vio_port_up(&port->vio);
 740         wait_for_completion(&comp.com);
 741         return comp.err;
 742 }
 743
 744 static void vdc_port_down(struct vdc_port *port)
 745 {
 746         ldc_disconnect(port->vio.lp);
 747         ldc_unbind(port->vio.lp);
 748         vdc_free_tx_ring(port);
 749         vio_ldc_free(&port->vio);
 750 }
 751
 752 static int probe_disk(struct vdc_port *port)
 753 {
 754         struct request_queue *q;
 755         struct gendisk *g;
 756         int err;
 757
 758         err = vdc_port_up(port);
 759         if (err)
 760                 return err;
 761
 762         if (vdc_version_supported(port, 1, 1)) {
 763                 /* vdisk_size should be set during the handshake, if it wasn't
 764                  * then the underlying disk is reserved by another system
 765                  */
 766                 if (port->vdisk_size == -1)
 767                         return -ENODEV;
 768         } else {
 769                 struct vio_disk_geom geom;
 770
 771                 err = generic_request(port, VD_OP_GET_DISKGEOM,
 772                                       &geom, sizeof(geom));
 773                 if (err < 0) {
 774                         printk(KERN_ERR PFX "VD_OP_GET_DISKGEOM returns "
 775                                "error %d\n", err);
 776                         return err;
 777                 }
 778                 port->vdisk_size = ((u64)geom.num_cyl *
 779                                     (u64)geom.num_hd *
 780                                     (u64)geom.num_sec);
 781         }
 782
 783         q = blk_init_queue(do_vdc_request, &port->vio.lock);
 784         if (!q) {
 785                 printk(KERN_ERR PFX "%s: Could not allocate queue.\n",
 786                        port->vio.name);
 787                 return -ENOMEM;
 788         }
 789         g = alloc_disk(1 << PARTITION_SHIFT);
 790         if (!g) {
 791                 printk(KERN_ERR PFX "%s: Could not allocate gendisk.\n",
 792                        port->vio.name);
 793                 blk_cleanup_queue(q);
 794                 return -ENOMEM;
 795         }
 796
 797         port->disk = g;
 798
 799         /* Each segment in a request is up to an aligned page in size. */
 800         blk_queue_segment_boundary(q, PAGE_SIZE - 1);
 801         blk_queue_max_segment_size(q, PAGE_SIZE);
 802
 803         blk_queue_max_segments(q, port->ring_cookies);
 804         blk_queue_max_hw_sectors(q, port->max_xfer_size);
 805         g->major = vdc_major;
 806         g->first_minor = port->vio.vdev->dev_no << PARTITION_SHIFT;
 807         strcpy(g->disk_name, port->disk_name);
 808
 809         g->fops = &vdc_fops;
 810         g->queue = q;
 811         g->private_data = port;
 812         g->driverfs_dev = &port->vio.vdev->dev;
 813
 814         set_capacity(g, port->vdisk_size);
 815
 816         if (vdc_version_supported(port, 1, 1)) {
 817                 switch (port->vdisk_mtype) {
 818                 case VD_MEDIA_TYPE_CD:
 819                         pr_info(PFX "Virtual CDROM %s\n", port->disk_name);
 820                         g->flags |= GENHD_FL_CD;
 821                         g->flags |= GENHD_FL_REMOVABLE;
 822                         set_disk_ro(g, 1);
 823                         break;
 824
 825                 case VD_MEDIA_TYPE_DVD:
 826                         pr_info(PFX "Virtual DVD %s\n", port->disk_name);
 827                         g->flags |= GENHD_FL_CD;
 828                         g->flags |= GENHD_FL_REMOVABLE;
 829                         set_disk_ro(g, 1);
 830                         break;
 831
 832                 case VD_MEDIA_TYPE_FIXED:
 833                         pr_info(PFX "Virtual Hard disk %s\n", port->disk_name);
 834                         break;
 835                 }
 836         }
 837
 838         pr_info(PFX "%s: %u sectors (%u MB) protocol %d.%d\n",
 839                g->disk_name,
 840                port->vdisk_size, (port->vdisk_size >> (20 - 9)),
 841                port->vio.ver.major, port->vio.ver.minor);
 842
 843         add_disk(g);
 844
 845         return 0;
 846 }
 847
 848 static struct ldc_channel_config vdc_ldc_cfg = {
 849         .event          = vdc_event,
 850         .mtu            = 64,
 851         .mode           = LDC_MODE_UNRELIABLE,
 852 };
 853
 854 static struct vio_driver_ops vdc_vio_ops = {
 855         .send_attr              = vdc_send_attr,
 856         .handle_attr            = vdc_handle_attr,
 857         .handshake_complete     = vdc_handshake_complete,
 858 };
 859
 860 static void print_version(void)
 861 {
 862         static int version_printed;
 863
 864         if (version_printed++ == 0)
 865                 printk(KERN_INFO "%s", version);
 866 }
 867
 868 static int vdc_port_probe(struct vio_dev *vdev, const struct vio_device_id *id)
 869 {
 870         struct mdesc_handle *hp;
 871         struct vdc_port *port;
 872         int err;
 873         const u64 *ldc_timeout;
 874
 875         print_version();
 876
 877         hp = mdesc_grab();
 878
 879         err = -ENODEV;
 880         if ((vdev->dev_no << PARTITION_SHIFT) & ~(u64)MINORMASK) {
 881                 printk(KERN_ERR PFX "Port id [%llu] too large.\n",
 882                        vdev->dev_no);
 883                 goto err_out_release_mdesc;
 884         }
 885
 886         port = kzalloc(sizeof(*port), GFP_KERNEL);
 887         err = -ENOMEM;
 888         if (!port) {
 889                 printk(KERN_ERR PFX "Cannot allocate vdc_port.\n");
 890                 goto err_out_release_mdesc;
 891         }
 892
 893         if (vdev->dev_no >= 26)
 894                 snprintf(port->disk_name, sizeof(port->disk_name),
 895                          VDCBLK_NAME "%c%c",
 896                          'a' + ((int)vdev->dev_no / 26) - 1,
 897                          'a' + ((int)vdev->dev_no % 26));
 898         else
 899                 snprintf(port->disk_name, sizeof(port->disk_name),
 900                          VDCBLK_NAME "%c", 'a' + ((int)vdev->dev_no % 26));
 901         port->vdisk_size = -1;
 902
 903         /* Actual wall time may be double due to do_generic_file_read() doing
 904          * a readahead I/O first, and once that fails it will try to read a
 905          * single page.
 906          */
 907         ldc_timeout = mdesc_get_property(hp, vdev->mp, "vdc-timeout", NULL);
 908         port->ldc_timeout = ldc_timeout ? *ldc_timeout : 0;
 909         setup_timer(&port->ldc_reset_timer, vdc_ldc_reset_timer,
 910                     (unsigned long)port);
 911         INIT_WORK(&port->ldc_reset_work, vdc_ldc_reset_work);
 912
 913         err = vio_driver_init(&port->vio, vdev, VDEV_DISK,
 914                               vdc_versions, ARRAY_SIZE(vdc_versions),
 915                               &vdc_vio_ops, port->disk_name);
 916         if (err)
 917                 goto err_out_free_port;
 918
 919         port->vdisk_block_size = 512;
 920         port->max_xfer_size = ((128 * 1024) / port->vdisk_block_size);
 921         port->ring_cookies = ((port->max_xfer_size *
 922                                port->vdisk_block_size) / PAGE_SIZE) + 2;
 923
 924         err = vio_ldc_alloc(&port->vio, &vdc_ldc_cfg, port);
 925         if (err)
 926                 goto err_out_free_port;
 927
 928         err = vdc_alloc_tx_ring(port);
 929         if (err)
 930                 goto err_out_free_ldc;
 931
 932         err = probe_disk(port);
 933         if (err)
 934                 goto err_out_free_tx_ring;
 935
 936         dev_set_drvdata(&vdev->dev, port);
 937
 938         mdesc_release(hp);
 939
 940         return 0;
 941
 942 err_out_free_tx_ring:
 943         vdc_free_tx_ring(port);
 944
 945 err_out_free_ldc:
 946         vio_ldc_free(&port->vio);
 947
 948 err_out_free_port:
 949         kfree(port);
 950
 951 err_out_release_mdesc:
 952         mdesc_release(hp);
 953         return err;
 954 }
 955
 956 static int vdc_port_remove(struct vio_dev *vdev)
 957 {
 958         struct vdc_port *port = dev_get_drvdata(&vdev->dev);
 959
 960         if (port) {
 961                 unsigned long flags;
 962
 963                 spin_lock_irqsave(&port->vio.lock, flags);
 964                 blk_stop_queue(port->disk->queue);
 965                 spin_unlock_irqrestore(&port->vio.lock, flags);
 966
 967                 flush_work(&port->ldc_reset_work);
 968                 del_timer_sync(&port->ldc_reset_timer);
 969                 del_timer_sync(&port->vio.timer);
 970
 971                 del_gendisk(port->disk);
 972                 blk_cleanup_queue(port->disk->queue);
 973                 put_disk(port->disk);
 974                 port->disk = NULL;
 975
 976                 vdc_free_tx_ring(port);
 977                 vio_ldc_free(&port->vio);
 978
 979                 dev_set_drvdata(&vdev->dev, NULL);
 980
 981                 kfree(port);
 982         }
 983         return 0;
 984 }
 985
 986 static void vdc_requeue_inflight(struct vdc_port *port)
 987 {
 988         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 989         u32 idx;
 990
 991         for (idx = dr->cons; idx != dr->prod; idx = vio_dring_next(dr, idx)) {
 992                 struct vio_disk_desc *desc = vio_dring_entry(dr, idx);
 993                 struct vdc_req_entry *rqe = &port->rq_arr[idx];
 994                 struct request *req;
 995
 996                 ldc_unmap(port->vio.lp, desc->cookies, desc->ncookies);
 997                 desc->hdr.state = VIO_DESC_FREE;
 998                 dr->cons = vio_dring_next(dr, idx);
 999
1000                 req = rqe->req;
1001                 if (req == NULL) {
1002                         vdc_end_special(port, desc);
1003                         continue;
1004                 }
1005
1006                 rqe->req = NULL;
1007                 blk_requeue_request(port->disk->queue, req);
1008         }
1009 }
1010
1011 static void vdc_queue_drain(struct vdc_port *port)
1012 {
1013         struct request *req;
1014
1015         while ((req = blk_fetch_request(port->disk->queue)) != NULL)
1016                 __blk_end_request_all(req, -EIO);
1017 }
1018
1019 static void vdc_ldc_reset_timer(unsigned long _arg)
1020 {
1021         struct vdc_port *port = (struct vdc_port *) _arg;
1022         struct vio_driver_state *vio = &port->vio;
1023         unsigned long flags;
1024
1025         spin_lock_irqsave(&vio->lock, flags);
1026         if (!(port->vio.hs_state & VIO_HS_COMPLETE)) {
1027                 pr_warn(PFX "%s ldc down %llu seconds, draining queue\n",
1028                         port->disk_name, port->ldc_timeout);
1029                 vdc_queue_drain(port);
1030                 vdc_blk_queue_start(port);
1031         }
1032         spin_unlock_irqrestore(&vio->lock, flags);
1033 }
1034
1035 static void vdc_ldc_reset_work(struct work_struct *work)
1036 {
1037         struct vdc_port *port;
1038         struct vio_driver_state *vio;
1039         unsigned long flags;
1040
1041         port = container_of(work, struct vdc_port, ldc_reset_work);
1042         vio = &port->vio;
1043
1044         spin_lock_irqsave(&vio->lock, flags);
1045         vdc_ldc_reset(port);
1046         spin_unlock_irqrestore(&vio->lock, flags);
1047 }
1048
1049 static void vdc_ldc_reset(struct vdc_port *port)
1050 {
1051         int err;
1052
1053         assert_spin_locked(&port->vio.lock);
1054
1055         pr_warn(PFX "%s ldc link reset\n", port->disk_name);
1056         blk_stop_queue(port->disk->queue);
1057         vdc_requeue_inflight(port);
1058         vdc_port_down(port);
1059
1060         err = vio_ldc_alloc(&port->vio, &vdc_ldc_cfg, port);
1061         if (err) {
1062                 pr_err(PFX "%s vio_ldc_alloc:%d\n", port->disk_name, err);
1063                 return;
1064         }
1065
1066         err = vdc_alloc_tx_ring(port);
1067         if (err) {
1068                 pr_err(PFX "%s vio_alloc_tx_ring:%d\n", port->disk_name, err);
1069                 goto err_free_ldc;
1070         }
1071
1072         if (port->ldc_timeout)
1073                 mod_timer(&port->ldc_reset_timer,
1074                           round_jiffies(jiffies + HZ * port->ldc_timeout));
1075         mod_timer(&port->vio.timer, round_jiffies(jiffies + HZ));
1076         return;
1077
1078 err_free_ldc:
1079         vio_ldc_free(&port->vio);
1080 }
1081
1082 static const struct vio_device_id vdc_port_match[] = {
1083         {
1084                 .type = "vdc-port",
1085         },
1086         {},
1087 };
1088 MODULE_DEVICE_TABLE(vio, vdc_port_match);
1089
1090 static struct vio_driver vdc_port_driver = {
1091         .id_table       = vdc_port_match,
1092         .probe          = vdc_port_probe,
1093         .remove         = vdc_port_remove,
1094         .name           = "vdc_port",
1095 };
1096
1097 static int __init vdc_init(void)
1098 {
1099         int err;
1100
1101         sunvdc_wq = alloc_workqueue("sunvdc", 0, 0);
1102         if (!sunvdc_wq)
1103                 return -ENOMEM;
1104
1105         err = register_blkdev(0, VDCBLK_NAME);
1106         if (err < 0)
1107                 goto out_free_wq;
1108
1109         vdc_major = err;
1110
1111         err = vio_register_driver(&vdc_port_driver);
1112         if (err)
1113                 goto out_unregister_blkdev;
1114
1115         return 0;
1116
1117 out_unregister_blkdev:
1118         unregister_blkdev(vdc_major, VDCBLK_NAME);
1119         vdc_major = 0;
1120
1121 out_free_wq:
1122         destroy_workqueue(sunvdc_wq);
1123         return err;
1124 }
1125
1126 static void __exit vdc_exit(void)
1127 {
1128         vio_unregister_driver(&vdc_port_driver);
1129         unregister_blkdev(vdc_major, VDCBLK_NAME);
1130         destroy_workqueue(sunvdc_wq);
1131 }
1132
1133 module_init(vdc_init);
1134 module_exit(vdc_exit);