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 #define VDC_DEFAULT_BLK_SIZE    512
  38
  39 #define MAX_XFER_BLKS           (128 * 1024)
  40 #define MAX_XFER_SIZE           (MAX_XFER_BLKS / VDC_DEFAULT_BLK_SIZE)
  41 #define MAX_RING_COOKIES        ((MAX_XFER_BLKS / PAGE_SIZE) + 2)
  42
  43 #define WAITING_FOR_LINK_UP     0x01
  44 #define WAITING_FOR_TX_SPACE    0x02
  45 #define WAITING_FOR_GEN_CMD     0x04
  46 #define WAITING_FOR_ANY         -1
  47
  48 static struct workqueue_struct *sunvdc_wq;
  49
  50 struct vdc_req_entry {
  51         struct request          *req;
  52 };
  53
  54 struct vdc_port {
  55         struct vio_driver_state vio;
  56
  57         struct gendisk          *disk;
  58
  59         struct vdc_completion   *cmp;
  60
  61         u64                     req_id;
  62         u64                     seq;
  63         struct vdc_req_entry    rq_arr[VDC_TX_RING_SIZE];
  64
  65         unsigned long           ring_cookies;
  66
  67         u64                     max_xfer_size;
  68         u32                     vdisk_block_size;
  69
  70         u64                     ldc_timeout;
  71         struct timer_list       ldc_reset_timer;
  72         struct work_struct      ldc_reset_work;
  73
  74         /* The server fills these in for us in the disk attribute
  75          * ACK packet.
  76          */
  77         u64                     operations;
  78         u32                     vdisk_size;
  79         u8                      vdisk_type;
  80         u8                      vdisk_mtype;
  81         u32                     vdisk_phys_blksz;
  82
  83         char                    disk_name[32];
  84 };
  85
  86 static void vdc_ldc_reset(struct vdc_port *port);
  87 static void vdc_ldc_reset_work(struct work_struct *work);
  88 static void vdc_ldc_reset_timer(struct timer_list *t);
  89
  90 static inline struct vdc_port *to_vdc_port(struct vio_driver_state *vio)
  91 {
  92         return container_of(vio, struct vdc_port, vio);
  93 }
  94
  95 /* Ordered from largest major to lowest */
  96 static struct vio_version vdc_versions[] = {
  97         { .major = 1, .minor = 2 },
  98         { .major = 1, .minor = 1 },
  99         { .major = 1, .minor = 0 },
 100 };
 101
 102 static inline int vdc_version_supported(struct vdc_port *port,
 103                                         u16 major, u16 minor)
 104 {
 105         return port->vio.ver.major == major && port->vio.ver.minor >= minor;
 106 }
 107
 108 #define VDCBLK_NAME     "vdisk"
 109 static int vdc_major;
 110 #define PARTITION_SHIFT 3
 111
 112 static inline u32 vdc_tx_dring_avail(struct vio_dring_state *dr)
 113 {
 114         return vio_dring_avail(dr, VDC_TX_RING_SIZE);
 115 }
 116
 117 static int vdc_getgeo(struct block_device *bdev, struct hd_geometry *geo)
 118 {
 119         struct gendisk *disk = bdev->bd_disk;
 120         sector_t nsect = get_capacity(disk);
 121         sector_t cylinders = nsect;
 122
 123         geo->heads = 0xff;
 124         geo->sectors = 0x3f;
 125         sector_div(cylinders, geo->heads * geo->sectors);
 126         geo->cylinders = cylinders;
 127         if ((sector_t)(geo->cylinders + 1) * geo->heads * geo->sectors < nsect)
 128                 geo->cylinders = 0xffff;
 129
 130         return 0;
 131 }
 132
 133 /* Add ioctl/CDROM_GET_CAPABILITY to support cdrom_id in udev
 134  * when vdisk_mtype is VD_MEDIA_TYPE_CD or VD_MEDIA_TYPE_DVD.
 135  * Needed to be able to install inside an ldom from an iso image.
 136  */
 137 static int vdc_ioctl(struct block_device *bdev, fmode_t mode,
 138                      unsigned command, unsigned long argument)
 139 {
 140         int i;
 141         struct gendisk *disk;
 142
 143         switch (command) {
 144         case CDROMMULTISESSION:
 145                 pr_debug(PFX "Multisession CDs not supported\n");
 146                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 147                         if (put_user(0, (char __user *)(argument + i)))
 148                                 return -EFAULT;
 149                 return 0;
 150
 151         case CDROM_GET_CAPABILITY:
 152                 disk = bdev->bd_disk;
 153
 154                 if (bdev->bd_disk && (disk->flags & GENHD_FL_CD))
 155                         return 0;
 156                 return -EINVAL;
 157
 158         default:
 159                 pr_debug(PFX "ioctl %08x not supported\n", command);
 160                 return -EINVAL;
 161         }
 162 }
 163
 164 static const struct block_device_operations vdc_fops = {
 165         .owner          = THIS_MODULE,
 166         .getgeo         = vdc_getgeo,
 167         .ioctl          = vdc_ioctl,
 168 };
 169
 170 static void vdc_blk_queue_start(struct vdc_port *port)
 171 {
 172         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 173
 174         /* restart blk queue when ring is half emptied. also called after
 175          * handshake completes, so check for initial handshake before we've
 176          * allocated a disk.
 177          */
 178         if (port->disk && blk_queue_stopped(port->disk->queue) &&
 179             vdc_tx_dring_avail(dr) * 100 / VDC_TX_RING_SIZE >= 50) {
 180                 blk_start_queue(port->disk->queue);
 181         }
 182
 183 }
 184
 185 static void vdc_finish(struct vio_driver_state *vio, int err, int waiting_for)
 186 {
 187         if (vio->cmp &&
 188             (waiting_for == -1 ||
 189              vio->cmp->waiting_for == waiting_for)) {
 190                 vio->cmp->err = err;
 191                 complete(&vio->cmp->com);
 192                 vio->cmp = NULL;
 193         }
 194 }
 195
 196 static void vdc_handshake_complete(struct vio_driver_state *vio)
 197 {
 198         struct vdc_port *port = to_vdc_port(vio);
 199
 200         del_timer(&port->ldc_reset_timer);
 201         vdc_finish(vio, 0, WAITING_FOR_LINK_UP);
 202         vdc_blk_queue_start(port);
 203 }
 204
 205 static int vdc_handle_unknown(struct vdc_port *port, void *arg)
 206 {
 207         struct vio_msg_tag *pkt = arg;
 208
 209         printk(KERN_ERR PFX "Received unknown msg [%02x:%02x:%04x:%08x]\n",
 210                pkt->type, pkt->stype, pkt->stype_env, pkt->sid);
 211         printk(KERN_ERR PFX "Resetting connection.\n");
 212
 213         ldc_disconnect(port->vio.lp);
 214
 215         return -ECONNRESET;
 216 }
 217
 218 static int vdc_send_attr(struct vio_driver_state *vio)
 219 {
 220         struct vdc_port *port = to_vdc_port(vio);
 221         struct vio_disk_attr_info pkt;
 222
 223         memset(&pkt, 0, sizeof(pkt));
 224
 225         pkt.tag.type = VIO_TYPE_CTRL;
 226         pkt.tag.stype = VIO_SUBTYPE_INFO;
 227         pkt.tag.stype_env = VIO_ATTR_INFO;
 228         pkt.tag.sid = vio_send_sid(vio);
 229
 230         pkt.xfer_mode = VIO_DRING_MODE;
 231         pkt.vdisk_block_size = port->vdisk_block_size;
 232         pkt.max_xfer_size = port->max_xfer_size;
 233
 234         viodbg(HS, "SEND ATTR xfer_mode[0x%x] blksz[%u] max_xfer[%llu]\n",
 235                pkt.xfer_mode, pkt.vdisk_block_size, pkt.max_xfer_size);
 236
 237         return vio_ldc_send(&port->vio, &pkt, sizeof(pkt));
 238 }
 239
 240 static int vdc_handle_attr(struct vio_driver_state *vio, void *arg)
 241 {
 242         struct vdc_port *port = to_vdc_port(vio);
 243         struct vio_disk_attr_info *pkt = arg;
 244
 245         viodbg(HS, "GOT ATTR stype[0x%x] ops[%llx] disk_size[%llu] disk_type[%x] "
 246                "mtype[0x%x] xfer_mode[0x%x] blksz[%u] max_xfer[%llu]\n",
 247                pkt->tag.stype, pkt->operations,
 248                pkt->vdisk_size, pkt->vdisk_type, pkt->vdisk_mtype,
 249                pkt->xfer_mode, pkt->vdisk_block_size,
 250                pkt->max_xfer_size);
 251
 252         if (pkt->tag.stype == VIO_SUBTYPE_ACK) {
 253                 switch (pkt->vdisk_type) {
 254                 case VD_DISK_TYPE_DISK:
 255                 case VD_DISK_TYPE_SLICE:
 256                         break;
 257
 258                 default:
 259                         printk(KERN_ERR PFX "%s: Bogus vdisk_type 0x%x\n",
 260                                vio->name, pkt->vdisk_type);
 261                         return -ECONNRESET;
 262                 }
 263
 264                 if (pkt->vdisk_block_size > port->vdisk_block_size) {
 265                         printk(KERN_ERR PFX "%s: BLOCK size increased "
 266                                "%u --> %u\n",
 267                                vio->name,
 268                                port->vdisk_block_size, pkt->vdisk_block_size);
 269                         return -ECONNRESET;
 270                 }
 271
 272                 port->operations = pkt->operations;
 273                 port->vdisk_type = pkt->vdisk_type;
 274                 if (vdc_version_supported(port, 1, 1)) {
 275                         port->vdisk_size = pkt->vdisk_size;
 276                         port->vdisk_mtype = pkt->vdisk_mtype;
 277                 }
 278                 if (pkt->max_xfer_size < port->max_xfer_size)
 279                         port->max_xfer_size = pkt->max_xfer_size;
 280                 port->vdisk_block_size = pkt->vdisk_block_size;
 281
 282                 port->vdisk_phys_blksz = VDC_DEFAULT_BLK_SIZE;
 283                 if (vdc_version_supported(port, 1, 2))
 284                         port->vdisk_phys_blksz = pkt->phys_block_size;
 285
 286                 return 0;
 287         } else {
 288                 printk(KERN_ERR PFX "%s: Attribute NACK\n", vio->name);
 289
 290                 return -ECONNRESET;
 291         }
 292 }
 293
 294 static void vdc_end_special(struct vdc_port *port, struct vio_disk_desc *desc)
 295 {
 296         int err = desc->status;
 297
 298         vdc_finish(&port->vio, -err, WAITING_FOR_GEN_CMD);
 299 }
 300
 301 static void vdc_end_one(struct vdc_port *port, struct vio_dring_state *dr,
 302                         unsigned int index)
 303 {
 304         struct vio_disk_desc *desc = vio_dring_entry(dr, index);
 305         struct vdc_req_entry *rqe = &port->rq_arr[index];
 306         struct request *req;
 307
 308         if (unlikely(desc->hdr.state != VIO_DESC_DONE))
 309                 return;
 310
 311         ldc_unmap(port->vio.lp, desc->cookies, desc->ncookies);
 312         desc->hdr.state = VIO_DESC_FREE;
 313         dr->cons = vio_dring_next(dr, index);
 314
 315         req = rqe->req;
 316         if (req == NULL) {
 317                 vdc_end_special(port, desc);
 318                 return;
 319         }
 320
 321         rqe->req = NULL;
 322
 323         __blk_end_request(req, (desc->status ? BLK_STS_IOERR : 0), desc->size);
 324
 325         vdc_blk_queue_start(port);
 326 }
 327
 328 static int vdc_ack(struct vdc_port *port, void *msgbuf)
 329 {
 330         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 331         struct vio_dring_data *pkt = msgbuf;
 332
 333         if (unlikely(pkt->dring_ident != dr->ident ||
 334                      pkt->start_idx != pkt->end_idx ||
 335                      pkt->start_idx >= VDC_TX_RING_SIZE))
 336                 return 0;
 337
 338         vdc_end_one(port, dr, pkt->start_idx);
 339
 340         return 0;
 341 }
 342
 343 static int vdc_nack(struct vdc_port *port, void *msgbuf)
 344 {
 345         /* XXX Implement me XXX */
 346         return 0;
 347 }
 348
 349 static void vdc_event(void *arg, int event)
 350 {
 351         struct vdc_port *port = arg;
 352         struct vio_driver_state *vio = &port->vio;
 353         unsigned long flags;
 354         int err;
 355
 356         spin_lock_irqsave(&vio->lock, flags);
 357
 358         if (unlikely(event == LDC_EVENT_RESET)) {
 359                 vio_link_state_change(vio, event);
 360                 queue_work(sunvdc_wq, &port->ldc_reset_work);
 361                 goto out;
 362         }
 363
 364         if (unlikely(event == LDC_EVENT_UP)) {
 365                 vio_link_state_change(vio, event);
 366                 goto out;
 367         }
 368
 369         if (unlikely(event != LDC_EVENT_DATA_READY)) {
 370                 pr_warn(PFX "Unexpected LDC event %d\n", event);
 371                 goto out;
 372         }
 373
 374         err = 0;
 375         while (1) {
 376                 union {
 377                         struct vio_msg_tag tag;
 378                         u64 raw[8];
 379                 } msgbuf;
 380
 381                 err = ldc_read(vio->lp, &msgbuf, sizeof(msgbuf));
 382                 if (unlikely(err < 0)) {
 383                         if (err == -ECONNRESET)
 384                                 vio_conn_reset(vio);
 385                         break;
 386                 }
 387                 if (err == 0)
 388                         break;
 389                 viodbg(DATA, "TAG [%02x:%02x:%04x:%08x]\n",
 390                        msgbuf.tag.type,
 391                        msgbuf.tag.stype,
 392                        msgbuf.tag.stype_env,
 393                        msgbuf.tag.sid);
 394                 err = vio_validate_sid(vio, &msgbuf.tag);
 395                 if (err < 0)
 396                         break;
 397
 398                 if (likely(msgbuf.tag.type == VIO_TYPE_DATA)) {
 399                         if (msgbuf.tag.stype == VIO_SUBTYPE_ACK)
 400                                 err = vdc_ack(port, &msgbuf);
 401                         else if (msgbuf.tag.stype == VIO_SUBTYPE_NACK)
 402                                 err = vdc_nack(port, &msgbuf);
 403                         else
 404                                 err = vdc_handle_unknown(port, &msgbuf);
 405                 } else if (msgbuf.tag.type == VIO_TYPE_CTRL) {
 406                         err = vio_control_pkt_engine(vio, &msgbuf);
 407                 } else {
 408                         err = vdc_handle_unknown(port, &msgbuf);
 409                 }
 410                 if (err < 0)
 411                         break;
 412         }
 413         if (err < 0)
 414                 vdc_finish(&port->vio, err, WAITING_FOR_ANY);
 415 out:
 416         spin_unlock_irqrestore(&vio->lock, flags);
 417 }
 418
 419 static int __vdc_tx_trigger(struct vdc_port *port)
 420 {
 421         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 422         struct vio_dring_data hdr = {
 423                 .tag = {
 424                         .type           = VIO_TYPE_DATA,
 425                         .stype          = VIO_SUBTYPE_INFO,
 426                         .stype_env      = VIO_DRING_DATA,
 427                         .sid            = vio_send_sid(&port->vio),
 428                 },
 429                 .dring_ident            = dr->ident,
 430                 .start_idx              = dr->prod,
 431                 .end_idx                = dr->prod,
 432         };
 433         int err, delay;
 434
 435         hdr.seq = dr->snd_nxt;
 436         delay = 1;
 437         do {
 438                 err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
 439                 if (err > 0) {
 440                         dr->snd_nxt++;
 441                         break;
 442                 }
 443                 udelay(delay);
 444                 if ((delay <<= 1) > 128)
 445                         delay = 128;
 446         } while (err == -EAGAIN);
 447
 448         if (err == -ENOTCONN)
 449                 vdc_ldc_reset(port);
 450         return err;
 451 }
 452
 453 static int __send_request(struct request *req)
 454 {
 455         struct vdc_port *port = req->rq_disk->private_data;
 456         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 457         struct scatterlist sg[MAX_RING_COOKIES];
 458         struct vdc_req_entry *rqe;
 459         struct vio_disk_desc *desc;
 460         unsigned int map_perm;
 461         int nsg, err, i;
 462         u64 len;
 463         u8 op;
 464
 465         if (WARN_ON(port->ring_cookies > MAX_RING_COOKIES))
 466                 return -EINVAL;
 467
 468         map_perm = LDC_MAP_SHADOW | LDC_MAP_DIRECT | LDC_MAP_IO;
 469
 470         if (rq_data_dir(req) == READ) {
 471                 map_perm |= LDC_MAP_W;
 472                 op = VD_OP_BREAD;
 473         } else {
 474                 map_perm |= LDC_MAP_R;
 475                 op = VD_OP_BWRITE;
 476         }
 477
 478         sg_init_table(sg, port->ring_cookies);
 479         nsg = blk_rq_map_sg(req->q, req, sg);
 480
 481         len = 0;
 482         for (i = 0; i < nsg; i++)
 483                 len += sg[i].length;
 484
 485         desc = vio_dring_cur(dr);
 486
 487         err = ldc_map_sg(port->vio.lp, sg, nsg,
 488                          desc->cookies, port->ring_cookies,
 489                          map_perm);
 490         if (err < 0) {
 491                 printk(KERN_ERR PFX "ldc_map_sg() failure, err=%d.\n", err);
 492                 return err;
 493         }
 494
 495         rqe = &port->rq_arr[dr->prod];
 496         rqe->req = req;
 497
 498         desc->hdr.ack = VIO_ACK_ENABLE;
 499         desc->req_id = port->req_id;
 500         desc->operation = op;
 501         if (port->vdisk_type == VD_DISK_TYPE_DISK) {
 502                 desc->slice = 0xff;
 503         } else {
 504                 desc->slice = 0;
 505         }
 506         desc->status = ~0;
 507         desc->offset = (blk_rq_pos(req) << 9) / port->vdisk_block_size;
 508         desc->size = len;
 509         desc->ncookies = err;
 510
 511         /* This has to be a non-SMP write barrier because we are writing
 512          * to memory which is shared with the peer LDOM.
 513          */
 514         wmb();
 515         desc->hdr.state = VIO_DESC_READY;
 516
 517         err = __vdc_tx_trigger(port);
 518         if (err < 0) {
 519                 printk(KERN_ERR PFX "vdc_tx_trigger() failure, err=%d\n", err);
 520         } else {
 521                 port->req_id++;
 522                 dr->prod = vio_dring_next(dr, dr->prod);
 523         }
 524
 525         return err;
 526 }
 527
 528 static void do_vdc_request(struct request_queue *rq)
 529 {
 530         struct request *req;
 531
 532         while ((req = blk_peek_request(rq)) != NULL) {
 533                 struct vdc_port *port;
 534                 struct vio_dring_state *dr;
 535
 536                 port = req->rq_disk->private_data;
 537                 dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 538                 if (unlikely(vdc_tx_dring_avail(dr) < 1))
 539                         goto wait;
 540
 541                 blk_start_request(req);
 542
 543                 if (__send_request(req) < 0) {
 544                         blk_requeue_request(rq, req);
 545 wait:
 546                         /* Avoid pointless unplugs. */
 547                         blk_stop_queue(rq);
 548                         break;
 549                 }
 550         }
 551 }
 552
 553 static int generic_request(struct vdc_port *port, u8 op, void *buf, int len)
 554 {
 555         struct vio_dring_state *dr;
 556         struct vio_completion comp;
 557         struct vio_disk_desc *desc;
 558         unsigned int map_perm;
 559         unsigned long flags;
 560         int op_len, err;
 561         void *req_buf;
 562
 563         if (!(((u64)1 << (u64)op) & port->operations))
 564                 return -EOPNOTSUPP;
 565
 566         switch (op) {
 567         case VD_OP_BREAD:
 568         case VD_OP_BWRITE:
 569         default:
 570                 return -EINVAL;
 571
 572         case VD_OP_FLUSH:
 573                 op_len = 0;
 574                 map_perm = 0;
 575                 break;
 576
 577         case VD_OP_GET_WCE:
 578                 op_len = sizeof(u32);
 579                 map_perm = LDC_MAP_W;
 580                 break;
 581
 582         case VD_OP_SET_WCE:
 583                 op_len = sizeof(u32);
 584                 map_perm = LDC_MAP_R;
 585                 break;
 586
 587         case VD_OP_GET_VTOC:
 588                 op_len = sizeof(struct vio_disk_vtoc);
 589                 map_perm = LDC_MAP_W;
 590                 break;
 591
 592         case VD_OP_SET_VTOC:
 593                 op_len = sizeof(struct vio_disk_vtoc);
 594                 map_perm = LDC_MAP_R;
 595                 break;
 596
 597         case VD_OP_GET_DISKGEOM:
 598                 op_len = sizeof(struct vio_disk_geom);
 599                 map_perm = LDC_MAP_W;
 600                 break;
 601
 602         case VD_OP_SET_DISKGEOM:
 603                 op_len = sizeof(struct vio_disk_geom);
 604                 map_perm = LDC_MAP_R;
 605                 break;
 606
 607         case VD_OP_SCSICMD:
 608                 op_len = 16;
 609                 map_perm = LDC_MAP_RW;
 610                 break;
 611
 612         case VD_OP_GET_DEVID:
 613                 op_len = sizeof(struct vio_disk_devid);
 614                 map_perm = LDC_MAP_W;
 615                 break;
 616
 617         case VD_OP_GET_EFI:
 618         case VD_OP_SET_EFI:
 619                 return -EOPNOTSUPP;
 620                 break;
 621         };
 622
 623         map_perm |= LDC_MAP_SHADOW | LDC_MAP_DIRECT | LDC_MAP_IO;
 624
 625         op_len = (op_len + 7) & ~7;
 626         req_buf = kzalloc(op_len, GFP_KERNEL);
 627         if (!req_buf)
 628                 return -ENOMEM;
 629
 630         if (len > op_len)
 631                 len = op_len;
 632
 633         if (map_perm & LDC_MAP_R)
 634                 memcpy(req_buf, buf, len);
 635
 636         spin_lock_irqsave(&port->vio.lock, flags);
 637
 638         dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 639
 640         /* XXX If we want to use this code generically we have to
 641          * XXX handle TX ring exhaustion etc.
 642          */
 643         desc = vio_dring_cur(dr);
 644
 645         err = ldc_map_single(port->vio.lp, req_buf, op_len,
 646                              desc->cookies, port->ring_cookies,
 647                              map_perm);
 648         if (err < 0) {
 649                 spin_unlock_irqrestore(&port->vio.lock, flags);
 650                 kfree(req_buf);
 651                 return err;
 652         }
 653
 654         init_completion(&comp.com);
 655         comp.waiting_for = WAITING_FOR_GEN_CMD;
 656         port->vio.cmp = &comp;
 657
 658         desc->hdr.ack = VIO_ACK_ENABLE;
 659         desc->req_id = port->req_id;
 660         desc->operation = op;
 661         desc->slice = 0;
 662         desc->status = ~0;
 663         desc->offset = 0;
 664         desc->size = op_len;
 665         desc->ncookies = err;
 666
 667         /* This has to be a non-SMP write barrier because we are writing
 668          * to memory which is shared with the peer LDOM.
 669          */
 670         wmb();
 671         desc->hdr.state = VIO_DESC_READY;
 672
 673         err = __vdc_tx_trigger(port);
 674         if (err >= 0) {
 675                 port->req_id++;
 676                 dr->prod = vio_dring_next(dr, dr->prod);
 677                 spin_unlock_irqrestore(&port->vio.lock, flags);
 678
 679                 wait_for_completion(&comp.com);
 680                 err = comp.err;
 681         } else {
 682                 port->vio.cmp = NULL;
 683                 spin_unlock_irqrestore(&port->vio.lock, flags);
 684         }
 685
 686         if (map_perm & LDC_MAP_W)
 687                 memcpy(buf, req_buf, len);
 688
 689         kfree(req_buf);
 690
 691         return err;
 692 }
 693
 694 static int vdc_alloc_tx_ring(struct vdc_port *port)
 695 {
 696         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 697         unsigned long len, entry_size;
 698         int ncookies;
 699         void *dring;
 700
 701         entry_size = sizeof(struct vio_disk_desc) +
 702                 (sizeof(struct ldc_trans_cookie) * port->ring_cookies);
 703         len = (VDC_TX_RING_SIZE * entry_size);
 704
 705         ncookies = VIO_MAX_RING_COOKIES;
 706         dring = ldc_alloc_exp_dring(port->vio.lp, len,
 707                                     dr->cookies, &ncookies,
 708                                     (LDC_MAP_SHADOW |
 709                                      LDC_MAP_DIRECT |
 710                                      LDC_MAP_RW));
 711         if (IS_ERR(dring))
 712                 return PTR_ERR(dring);
 713
 714         dr->base = dring;
 715         dr->entry_size = entry_size;
 716         dr->num_entries = VDC_TX_RING_SIZE;
 717         dr->prod = dr->cons = 0;
 718         dr->pending = VDC_TX_RING_SIZE;
 719         dr->ncookies = ncookies;
 720
 721         return 0;
 722 }
 723
 724 static void vdc_free_tx_ring(struct vdc_port *port)
 725 {
 726         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 727
 728         if (dr->base) {
 729                 ldc_free_exp_dring(port->vio.lp, dr->base,
 730                                    (dr->entry_size * dr->num_entries),
 731                                    dr->cookies, dr->ncookies);
 732                 dr->base = NULL;
 733                 dr->entry_size = 0;
 734                 dr->num_entries = 0;
 735                 dr->pending = 0;
 736                 dr->ncookies = 0;
 737         }
 738 }
 739
 740 static int vdc_port_up(struct vdc_port *port)
 741 {
 742         struct vio_completion comp;
 743
 744         init_completion(&comp.com);
 745         comp.err = 0;
 746         comp.waiting_for = WAITING_FOR_LINK_UP;
 747         port->vio.cmp = &comp;
 748
 749         vio_port_up(&port->vio);
 750         wait_for_completion(&comp.com);
 751         return comp.err;
 752 }
 753
 754 static void vdc_port_down(struct vdc_port *port)
 755 {
 756         ldc_disconnect(port->vio.lp);
 757         ldc_unbind(port->vio.lp);
 758         vdc_free_tx_ring(port);
 759         vio_ldc_free(&port->vio);
 760 }
 761
 762 static int probe_disk(struct vdc_port *port)
 763 {
 764         struct request_queue *q;
 765         struct gendisk *g;
 766         int err;
 767
 768         err = vdc_port_up(port);
 769         if (err)
 770                 return err;
 771
 772         /* Using version 1.2 means vdisk_phys_blksz should be set unless the
 773          * disk is reserved by another system.
 774          */
 775         if (vdc_version_supported(port, 1, 2) && !port->vdisk_phys_blksz)
 776                 return -ENODEV;
 777
 778         if (vdc_version_supported(port, 1, 1)) {
 779                 /* vdisk_size should be set during the handshake, if it wasn't
 780                  * then the underlying disk is reserved by another system
 781                  */
 782                 if (port->vdisk_size == -1)
 783                         return -ENODEV;
 784         } else {
 785                 struct vio_disk_geom geom;
 786
 787                 err = generic_request(port, VD_OP_GET_DISKGEOM,
 788                                       &geom, sizeof(geom));
 789                 if (err < 0) {
 790                         printk(KERN_ERR PFX "VD_OP_GET_DISKGEOM returns "
 791                                "error %d\n", err);
 792                         return err;
 793                 }
 794                 port->vdisk_size = ((u64)geom.num_cyl *
 795                                     (u64)geom.num_hd *
 796                                     (u64)geom.num_sec);
 797         }
 798
 799         q = blk_init_queue(do_vdc_request, &port->vio.lock);
 800         if (!q) {
 801                 printk(KERN_ERR PFX "%s: Could not allocate queue.\n",
 802                        port->vio.name);
 803                 return -ENOMEM;
 804         }
 805         g = alloc_disk(1 << PARTITION_SHIFT);
 806         if (!g) {
 807                 printk(KERN_ERR PFX "%s: Could not allocate gendisk.\n",
 808                        port->vio.name);
 809                 blk_cleanup_queue(q);
 810                 return -ENOMEM;
 811         }
 812
 813         port->disk = g;
 814
 815         /* Each segment in a request is up to an aligned page in size. */
 816         blk_queue_segment_boundary(q, PAGE_SIZE - 1);
 817         blk_queue_max_segment_size(q, PAGE_SIZE);
 818
 819         blk_queue_max_segments(q, port->ring_cookies);
 820         blk_queue_max_hw_sectors(q, port->max_xfer_size);
 821         g->major = vdc_major;
 822         g->first_minor = port->vio.vdev->dev_no << PARTITION_SHIFT;
 823         strcpy(g->disk_name, port->disk_name);
 824
 825         g->fops = &vdc_fops;
 826         g->queue = q;
 827         g->private_data = port;
 828
 829         set_capacity(g, port->vdisk_size);
 830
 831         if (vdc_version_supported(port, 1, 1)) {
 832                 switch (port->vdisk_mtype) {
 833                 case VD_MEDIA_TYPE_CD:
 834                         pr_info(PFX "Virtual CDROM %s\n", port->disk_name);
 835                         g->flags |= GENHD_FL_CD;
 836                         g->flags |= GENHD_FL_REMOVABLE;
 837                         set_disk_ro(g, 1);
 838                         break;
 839
 840                 case VD_MEDIA_TYPE_DVD:
 841                         pr_info(PFX "Virtual DVD %s\n", port->disk_name);
 842                         g->flags |= GENHD_FL_CD;
 843                         g->flags |= GENHD_FL_REMOVABLE;
 844                         set_disk_ro(g, 1);
 845                         break;
 846
 847                 case VD_MEDIA_TYPE_FIXED:
 848                         pr_info(PFX "Virtual Hard disk %s\n", port->disk_name);
 849                         break;
 850                 }
 851         }
 852
 853         blk_queue_physical_block_size(q, port->vdisk_phys_blksz);
 854
 855         pr_info(PFX "%s: %u sectors (%u MB) protocol %d.%d\n",
 856                g->disk_name,
 857                port->vdisk_size, (port->vdisk_size >> (20 - 9)),
 858                port->vio.ver.major, port->vio.ver.minor);
 859
 860         device_add_disk(&port->vio.vdev->dev, g, NULL);
 861
 862         return 0;
 863 }
 864
 865 static struct ldc_channel_config vdc_ldc_cfg = {
 866         .event          = vdc_event,
 867         .mtu            = 64,
 868         .mode           = LDC_MODE_UNRELIABLE,
 869 };
 870
 871 static struct vio_driver_ops vdc_vio_ops = {
 872         .send_attr              = vdc_send_attr,
 873         .handle_attr            = vdc_handle_attr,
 874         .handshake_complete     = vdc_handshake_complete,
 875 };
 876
 877 static void print_version(void)
 878 {
 879         static int version_printed;
 880
 881         if (version_printed++ == 0)
 882                 printk(KERN_INFO "%s", version);
 883 }
 884
 885 struct vdc_check_port_data {
 886         int     dev_no;
 887         char    *type;
 888 };
 889
 890 static int vdc_device_probed(struct device *dev, void *arg)
 891 {
 892         struct vio_dev *vdev = to_vio_dev(dev);
 893         struct vdc_check_port_data *port_data;
 894
 895         port_data = (struct vdc_check_port_data *)arg;
 896
 897         if ((vdev->dev_no == port_data->dev_no) &&
 898             (!(strcmp((char *)&vdev->type, port_data->type))) &&
 899                 dev_get_drvdata(dev)) {
 900                 /* This device has already been configured
 901                  * by vdc_port_probe()
 902                  */
 903                 return 1;
 904         } else {
 905                 return 0;
 906         }
 907 }
 908
 909 /* Determine whether the VIO device is part of an mpgroup
 910  * by locating all the virtual-device-port nodes associated
 911  * with the parent virtual-device node for the VIO device
 912  * and checking whether any of these nodes are vdc-ports
 913  * which have already been configured.
 914  *
 915  * Returns true if this device is part of an mpgroup and has
 916  * already been probed.
 917  */
 918 static bool vdc_port_mpgroup_check(struct vio_dev *vdev)
 919 {
 920         struct vdc_check_port_data port_data;
 921         struct device *dev;
 922
 923         port_data.dev_no = vdev->dev_no;
 924         port_data.type = (char *)&vdev->type;
 925
 926         dev = device_find_child(vdev->dev.parent, &port_data,
 927                                 vdc_device_probed);
 928
 929         if (dev)
 930                 return true;
 931
 932         return false;
 933 }
 934
 935 static int vdc_port_probe(struct vio_dev *vdev, const struct vio_device_id *id)
 936 {
 937         struct mdesc_handle *hp;
 938         struct vdc_port *port;
 939         int err;
 940         const u64 *ldc_timeout;
 941
 942         print_version();
 943
 944         hp = mdesc_grab();
 945
 946         err = -ENODEV;
 947         if ((vdev->dev_no << PARTITION_SHIFT) & ~(u64)MINORMASK) {
 948                 printk(KERN_ERR PFX "Port id [%llu] too large.\n",
 949                        vdev->dev_no);
 950                 goto err_out_release_mdesc;
 951         }
 952
 953         /* Check if this device is part of an mpgroup */
 954         if (vdc_port_mpgroup_check(vdev)) {
 955                 printk(KERN_WARNING
 956                         "VIO: Ignoring extra vdisk port %s",
 957                         dev_name(&vdev->dev));
 958                 goto err_out_release_mdesc;
 959         }
 960
 961         port = kzalloc(sizeof(*port), GFP_KERNEL);
 962         err = -ENOMEM;
 963         if (!port) {
 964                 printk(KERN_ERR PFX "Cannot allocate vdc_port.\n");
 965                 goto err_out_release_mdesc;
 966         }
 967
 968         if (vdev->dev_no >= 26)
 969                 snprintf(port->disk_name, sizeof(port->disk_name),
 970                          VDCBLK_NAME "%c%c",
 971                          'a' + ((int)vdev->dev_no / 26) - 1,
 972                          'a' + ((int)vdev->dev_no % 26));
 973         else
 974                 snprintf(port->disk_name, sizeof(port->disk_name),
 975                          VDCBLK_NAME "%c", 'a' + ((int)vdev->dev_no % 26));
 976         port->vdisk_size = -1;
 977
 978         /* Actual wall time may be double due to do_generic_file_read() doing
 979          * a readahead I/O first, and once that fails it will try to read a
 980          * single page.
 981          */
 982         ldc_timeout = mdesc_get_property(hp, vdev->mp, "vdc-timeout", NULL);
 983         port->ldc_timeout = ldc_timeout ? *ldc_timeout : 0;
 984         timer_setup(&port->ldc_reset_timer, vdc_ldc_reset_timer, 0);
 985         INIT_WORK(&port->ldc_reset_work, vdc_ldc_reset_work);
 986
 987         err = vio_driver_init(&port->vio, vdev, VDEV_DISK,
 988                               vdc_versions, ARRAY_SIZE(vdc_versions),
 989                               &vdc_vio_ops, port->disk_name);
 990         if (err)
 991                 goto err_out_free_port;
 992
 993         port->vdisk_block_size = VDC_DEFAULT_BLK_SIZE;
 994         port->max_xfer_size = MAX_XFER_SIZE;
 995         port->ring_cookies = MAX_RING_COOKIES;
 996
 997         err = vio_ldc_alloc(&port->vio, &vdc_ldc_cfg, port);
 998         if (err)
 999                 goto err_out_free_port;
1000
1001         err = vdc_alloc_tx_ring(port);
1002         if (err)
1003                 goto err_out_free_ldc;
1004
1005         err = probe_disk(port);
1006         if (err)
1007                 goto err_out_free_tx_ring;
1008
1009         /* Note that the device driver_data is used to determine
1010          * whether the port has been probed.
1011          */
1012         dev_set_drvdata(&vdev->dev, port);
1013
1014         mdesc_release(hp);
1015
1016         return 0;
1017
1018 err_out_free_tx_ring:
1019         vdc_free_tx_ring(port);
1020
1021 err_out_free_ldc:
1022         vio_ldc_free(&port->vio);
1023
1024 err_out_free_port:
1025         kfree(port);
1026
1027 err_out_release_mdesc:
1028         mdesc_release(hp);
1029         return err;
1030 }
1031
1032 static int vdc_port_remove(struct vio_dev *vdev)
1033 {
1034         struct vdc_port *port = dev_get_drvdata(&vdev->dev);
1035
1036         if (port) {
1037                 unsigned long flags;
1038
1039                 spin_lock_irqsave(&port->vio.lock, flags);
1040                 blk_stop_queue(port->disk->queue);
1041                 spin_unlock_irqrestore(&port->vio.lock, flags);
1042
1043                 flush_work(&port->ldc_reset_work);
1044                 del_timer_sync(&port->ldc_reset_timer);
1045                 del_timer_sync(&port->vio.timer);
1046
1047                 del_gendisk(port->disk);
1048                 blk_cleanup_queue(port->disk->queue);
1049                 put_disk(port->disk);
1050                 port->disk = NULL;
1051
1052                 vdc_free_tx_ring(port);
1053                 vio_ldc_free(&port->vio);
1054
1055                 dev_set_drvdata(&vdev->dev, NULL);
1056
1057                 kfree(port);
1058         }
1059         return 0;
1060 }
1061
1062 static void vdc_requeue_inflight(struct vdc_port *port)
1063 {
1064         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1065         u32 idx;
1066
1067         for (idx = dr->cons; idx != dr->prod; idx = vio_dring_next(dr, idx)) {
1068                 struct vio_disk_desc *desc = vio_dring_entry(dr, idx);
1069                 struct vdc_req_entry *rqe = &port->rq_arr[idx];
1070                 struct request *req;
1071
1072                 ldc_unmap(port->vio.lp, desc->cookies, desc->ncookies);
1073                 desc->hdr.state = VIO_DESC_FREE;
1074                 dr->cons = vio_dring_next(dr, idx);
1075
1076                 req = rqe->req;
1077                 if (req == NULL) {
1078                         vdc_end_special(port, desc);
1079                         continue;
1080                 }
1081
1082                 rqe->req = NULL;
1083                 blk_requeue_request(port->disk->queue, req);
1084         }
1085 }
1086
1087 static void vdc_queue_drain(struct vdc_port *port)
1088 {
1089         struct request *req;
1090
1091         while ((req = blk_fetch_request(port->disk->queue)) != NULL)
1092                 __blk_end_request_all(req, BLK_STS_IOERR);
1093 }
1094
1095 static void vdc_ldc_reset_timer(struct timer_list *t)
1096 {
1097         struct vdc_port *port = from_timer(port, t, ldc_reset_timer);
1098         struct vio_driver_state *vio = &port->vio;
1099         unsigned long flags;
1100
1101         spin_lock_irqsave(&vio->lock, flags);
1102         if (!(port->vio.hs_state & VIO_HS_COMPLETE)) {
1103                 pr_warn(PFX "%s ldc down %llu seconds, draining queue\n",
1104                         port->disk_name, port->ldc_timeout);
1105                 vdc_queue_drain(port);
1106                 vdc_blk_queue_start(port);
1107         }
1108         spin_unlock_irqrestore(&vio->lock, flags);
1109 }
1110
1111 static void vdc_ldc_reset_work(struct work_struct *work)
1112 {
1113         struct vdc_port *port;
1114         struct vio_driver_state *vio;
1115         unsigned long flags;
1116
1117         port = container_of(work, struct vdc_port, ldc_reset_work);
1118         vio = &port->vio;
1119
1120         spin_lock_irqsave(&vio->lock, flags);
1121         vdc_ldc_reset(port);
1122         spin_unlock_irqrestore(&vio->lock, flags);
1123 }
1124
1125 static void vdc_ldc_reset(struct vdc_port *port)
1126 {
1127         int err;
1128
1129         assert_spin_locked(&port->vio.lock);
1130
1131         pr_warn(PFX "%s ldc link reset\n", port->disk_name);
1132         blk_stop_queue(port->disk->queue);
1133         vdc_requeue_inflight(port);
1134         vdc_port_down(port);
1135
1136         err = vio_ldc_alloc(&port->vio, &vdc_ldc_cfg, port);
1137         if (err) {
1138                 pr_err(PFX "%s vio_ldc_alloc:%d\n", port->disk_name, err);
1139                 return;
1140         }
1141
1142         err = vdc_alloc_tx_ring(port);
1143         if (err) {
1144                 pr_err(PFX "%s vio_alloc_tx_ring:%d\n", port->disk_name, err);
1145                 goto err_free_ldc;
1146         }
1147
1148         if (port->ldc_timeout)
1149                 mod_timer(&port->ldc_reset_timer,
1150                           round_jiffies(jiffies + HZ * port->ldc_timeout));
1151         mod_timer(&port->vio.timer, round_jiffies(jiffies + HZ));
1152         return;
1153
1154 err_free_ldc:
1155         vio_ldc_free(&port->vio);
1156 }
1157
1158 static const struct vio_device_id vdc_port_match[] = {
1159         {
1160                 .type = "vdc-port",
1161         },
1162         {},
1163 };
1164 MODULE_DEVICE_TABLE(vio, vdc_port_match);
1165
1166 static struct vio_driver vdc_port_driver = {
1167         .id_table       = vdc_port_match,
1168         .probe          = vdc_port_probe,
1169         .remove         = vdc_port_remove,
1170         .name           = "vdc_port",
1171 };
1172
1173 static int __init vdc_init(void)
1174 {
1175         int err;
1176
1177         sunvdc_wq = alloc_workqueue("sunvdc", 0, 0);
1178         if (!sunvdc_wq)
1179                 return -ENOMEM;
1180
1181         err = register_blkdev(0, VDCBLK_NAME);
1182         if (err < 0)
1183                 goto out_free_wq;
1184
1185         vdc_major = err;
1186
1187         err = vio_register_driver(&vdc_port_driver);
1188         if (err)
1189                 goto out_unregister_blkdev;
1190
1191         return 0;
1192
1193 out_unregister_blkdev:
1194         unregister_blkdev(vdc_major, VDCBLK_NAME);
1195         vdc_major = 0;
1196
1197 out_free_wq:
1198         destroy_workqueue(sunvdc_wq);
1199         return err;
1200 }
1201
1202 static void __exit vdc_exit(void)
1203 {
1204         vio_unregister_driver(&vdc_port_driver);
1205         unregister_blkdev(vdc_major, VDCBLK_NAME);
1206         destroy_workqueue(sunvdc_wq);
1207 }
1208
1209 module_init(vdc_init);
1210 module_exit(vdc_exit);