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 #define VDC_MAX_RETRIES 10
  49
  50 static struct workqueue_struct *sunvdc_wq;
  51
  52 struct vdc_req_entry {
  53         struct request          *req;
  54 };
  55
  56 struct vdc_port {
  57         struct vio_driver_state vio;
  58
  59         struct gendisk          *disk;
  60
  61         struct vdc_completion   *cmp;
  62
  63         u64                     req_id;
  64         u64                     seq;
  65         struct vdc_req_entry    rq_arr[VDC_TX_RING_SIZE];
  66
  67         unsigned long           ring_cookies;
  68
  69         u64                     max_xfer_size;
  70         u32                     vdisk_block_size;
  71
  72         u64                     ldc_timeout;
  73         struct timer_list       ldc_reset_timer;
  74         struct work_struct      ldc_reset_work;
  75
  76         /* The server fills these in for us in the disk attribute
  77          * ACK packet.
  78          */
  79         u64                     operations;
  80         u32                     vdisk_size;
  81         u8                      vdisk_type;
  82         u8                      vdisk_mtype;
  83         u32                     vdisk_phys_blksz;
  84
  85         char                    disk_name[32];
  86 };
  87
  88 static void vdc_ldc_reset(struct vdc_port *port);
  89 static void vdc_ldc_reset_work(struct work_struct *work);
  90 static void vdc_ldc_reset_timer(struct timer_list *t);
  91
  92 static inline struct vdc_port *to_vdc_port(struct vio_driver_state *vio)
  93 {
  94         return container_of(vio, struct vdc_port, vio);
  95 }
  96
  97 /* Ordered from largest major to lowest */
  98 static struct vio_version vdc_versions[] = {
  99         { .major = 1, .minor = 2 },
 100         { .major = 1, .minor = 1 },
 101         { .major = 1, .minor = 0 },
 102 };
 103
 104 static inline int vdc_version_supported(struct vdc_port *port,
 105                                         u16 major, u16 minor)
 106 {
 107         return port->vio.ver.major == major && port->vio.ver.minor >= minor;
 108 }
 109
 110 #define VDCBLK_NAME     "vdisk"
 111 static int vdc_major;
 112 #define PARTITION_SHIFT 3
 113
 114 static inline u32 vdc_tx_dring_avail(struct vio_dring_state *dr)
 115 {
 116         return vio_dring_avail(dr, VDC_TX_RING_SIZE);
 117 }
 118
 119 static int vdc_getgeo(struct block_device *bdev, struct hd_geometry *geo)
 120 {
 121         struct gendisk *disk = bdev->bd_disk;
 122         sector_t nsect = get_capacity(disk);
 123         sector_t cylinders = nsect;
 124
 125         geo->heads = 0xff;
 126         geo->sectors = 0x3f;
 127         sector_div(cylinders, geo->heads * geo->sectors);
 128         geo->cylinders = cylinders;
 129         if ((sector_t)(geo->cylinders + 1) * geo->heads * geo->sectors < nsect)
 130                 geo->cylinders = 0xffff;
 131
 132         return 0;
 133 }
 134
 135 /* Add ioctl/CDROM_GET_CAPABILITY to support cdrom_id in udev
 136  * when vdisk_mtype is VD_MEDIA_TYPE_CD or VD_MEDIA_TYPE_DVD.
 137  * Needed to be able to install inside an ldom from an iso image.
 138  */
 139 static int vdc_ioctl(struct block_device *bdev, fmode_t mode,
 140                      unsigned command, unsigned long argument)
 141 {
 142         int i;
 143         struct gendisk *disk;
 144
 145         switch (command) {
 146         case CDROMMULTISESSION:
 147                 pr_debug(PFX "Multisession CDs not supported\n");
 148                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 149                         if (put_user(0, (char __user *)(argument + i)))
 150                                 return -EFAULT;
 151                 return 0;
 152
 153         case CDROM_GET_CAPABILITY:
 154                 disk = bdev->bd_disk;
 155
 156                 if (bdev->bd_disk && (disk->flags & GENHD_FL_CD))
 157                         return 0;
 158                 return -EINVAL;
 159
 160         default:
 161                 pr_debug(PFX "ioctl %08x not supported\n", command);
 162                 return -EINVAL;
 163         }
 164 }
 165
 166 static const struct block_device_operations vdc_fops = {
 167         .owner          = THIS_MODULE,
 168         .getgeo         = vdc_getgeo,
 169         .ioctl          = vdc_ioctl,
 170 };
 171
 172 static void vdc_blk_queue_start(struct vdc_port *port)
 173 {
 174         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 175
 176         /* restart blk queue when ring is half emptied. also called after
 177          * handshake completes, so check for initial handshake before we've
 178          * allocated a disk.
 179          */
 180         if (port->disk && blk_queue_stopped(port->disk->queue) &&
 181             vdc_tx_dring_avail(dr) * 100 / VDC_TX_RING_SIZE >= 50) {
 182                 blk_start_queue(port->disk->queue);
 183         }
 184
 185 }
 186
 187 static void vdc_finish(struct vio_driver_state *vio, int err, int waiting_for)
 188 {
 189         if (vio->cmp &&
 190             (waiting_for == -1 ||
 191              vio->cmp->waiting_for == waiting_for)) {
 192                 vio->cmp->err = err;
 193                 complete(&vio->cmp->com);
 194                 vio->cmp = NULL;
 195         }
 196 }
 197
 198 static void vdc_handshake_complete(struct vio_driver_state *vio)
 199 {
 200         struct vdc_port *port = to_vdc_port(vio);
 201
 202         del_timer(&port->ldc_reset_timer);
 203         vdc_finish(vio, 0, WAITING_FOR_LINK_UP);
 204         vdc_blk_queue_start(port);
 205 }
 206
 207 static int vdc_handle_unknown(struct vdc_port *port, void *arg)
 208 {
 209         struct vio_msg_tag *pkt = arg;
 210
 211         printk(KERN_ERR PFX "Received unknown msg [%02x:%02x:%04x:%08x]\n",
 212                pkt->type, pkt->stype, pkt->stype_env, pkt->sid);
 213         printk(KERN_ERR PFX "Resetting connection.\n");
 214
 215         ldc_disconnect(port->vio.lp);
 216
 217         return -ECONNRESET;
 218 }
 219
 220 static int vdc_send_attr(struct vio_driver_state *vio)
 221 {
 222         struct vdc_port *port = to_vdc_port(vio);
 223         struct vio_disk_attr_info pkt;
 224
 225         memset(&pkt, 0, sizeof(pkt));
 226
 227         pkt.tag.type = VIO_TYPE_CTRL;
 228         pkt.tag.stype = VIO_SUBTYPE_INFO;
 229         pkt.tag.stype_env = VIO_ATTR_INFO;
 230         pkt.tag.sid = vio_send_sid(vio);
 231
 232         pkt.xfer_mode = VIO_DRING_MODE;
 233         pkt.vdisk_block_size = port->vdisk_block_size;
 234         pkt.max_xfer_size = port->max_xfer_size;
 235
 236         viodbg(HS, "SEND ATTR xfer_mode[0x%x] blksz[%u] max_xfer[%llu]\n",
 237                pkt.xfer_mode, pkt.vdisk_block_size, pkt.max_xfer_size);
 238
 239         return vio_ldc_send(&port->vio, &pkt, sizeof(pkt));
 240 }
 241
 242 static int vdc_handle_attr(struct vio_driver_state *vio, void *arg)
 243 {
 244         struct vdc_port *port = to_vdc_port(vio);
 245         struct vio_disk_attr_info *pkt = arg;
 246
 247         viodbg(HS, "GOT ATTR stype[0x%x] ops[%llx] disk_size[%llu] disk_type[%x] "
 248                "mtype[0x%x] xfer_mode[0x%x] blksz[%u] max_xfer[%llu]\n",
 249                pkt->tag.stype, pkt->operations,
 250                pkt->vdisk_size, pkt->vdisk_type, pkt->vdisk_mtype,
 251                pkt->xfer_mode, pkt->vdisk_block_size,
 252                pkt->max_xfer_size);
 253
 254         if (pkt->tag.stype == VIO_SUBTYPE_ACK) {
 255                 switch (pkt->vdisk_type) {
 256                 case VD_DISK_TYPE_DISK:
 257                 case VD_DISK_TYPE_SLICE:
 258                         break;
 259
 260                 default:
 261                         printk(KERN_ERR PFX "%s: Bogus vdisk_type 0x%x\n",
 262                                vio->name, pkt->vdisk_type);
 263                         return -ECONNRESET;
 264                 }
 265
 266                 if (pkt->vdisk_block_size > port->vdisk_block_size) {
 267                         printk(KERN_ERR PFX "%s: BLOCK size increased "
 268                                "%u --> %u\n",
 269                                vio->name,
 270                                port->vdisk_block_size, pkt->vdisk_block_size);
 271                         return -ECONNRESET;
 272                 }
 273
 274                 port->operations = pkt->operations;
 275                 port->vdisk_type = pkt->vdisk_type;
 276                 if (vdc_version_supported(port, 1, 1)) {
 277                         port->vdisk_size = pkt->vdisk_size;
 278                         port->vdisk_mtype = pkt->vdisk_mtype;
 279                 }
 280                 if (pkt->max_xfer_size < port->max_xfer_size)
 281                         port->max_xfer_size = pkt->max_xfer_size;
 282                 port->vdisk_block_size = pkt->vdisk_block_size;
 283
 284                 port->vdisk_phys_blksz = VDC_DEFAULT_BLK_SIZE;
 285                 if (vdc_version_supported(port, 1, 2))
 286                         port->vdisk_phys_blksz = pkt->phys_block_size;
 287
 288                 return 0;
 289         } else {
 290                 printk(KERN_ERR PFX "%s: Attribute NACK\n", vio->name);
 291
 292                 return -ECONNRESET;
 293         }
 294 }
 295
 296 static void vdc_end_special(struct vdc_port *port, struct vio_disk_desc *desc)
 297 {
 298         int err = desc->status;
 299
 300         vdc_finish(&port->vio, -err, WAITING_FOR_GEN_CMD);
 301 }
 302
 303 static void vdc_end_one(struct vdc_port *port, struct vio_dring_state *dr,
 304                         unsigned int index)
 305 {
 306         struct vio_disk_desc *desc = vio_dring_entry(dr, index);
 307         struct vdc_req_entry *rqe = &port->rq_arr[index];
 308         struct request *req;
 309
 310         if (unlikely(desc->hdr.state != VIO_DESC_DONE))
 311                 return;
 312
 313         ldc_unmap(port->vio.lp, desc->cookies, desc->ncookies);
 314         desc->hdr.state = VIO_DESC_FREE;
 315         dr->cons = vio_dring_next(dr, index);
 316
 317         req = rqe->req;
 318         if (req == NULL) {
 319                 vdc_end_special(port, desc);
 320                 return;
 321         }
 322
 323         rqe->req = NULL;
 324
 325         __blk_end_request(req, (desc->status ? BLK_STS_IOERR : 0), desc->size);
 326
 327         vdc_blk_queue_start(port);
 328 }
 329
 330 static int vdc_ack(struct vdc_port *port, void *msgbuf)
 331 {
 332         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 333         struct vio_dring_data *pkt = msgbuf;
 334
 335         if (unlikely(pkt->dring_ident != dr->ident ||
 336                      pkt->start_idx != pkt->end_idx ||
 337                      pkt->start_idx >= VDC_TX_RING_SIZE))
 338                 return 0;
 339
 340         vdc_end_one(port, dr, pkt->start_idx);
 341
 342         return 0;
 343 }
 344
 345 static int vdc_nack(struct vdc_port *port, void *msgbuf)
 346 {
 347         /* XXX Implement me XXX */
 348         return 0;
 349 }
 350
 351 static void vdc_event(void *arg, int event)
 352 {
 353         struct vdc_port *port = arg;
 354         struct vio_driver_state *vio = &port->vio;
 355         unsigned long flags;
 356         int err;
 357
 358         spin_lock_irqsave(&vio->lock, flags);
 359
 360         if (unlikely(event == LDC_EVENT_RESET)) {
 361                 vio_link_state_change(vio, event);
 362                 queue_work(sunvdc_wq, &port->ldc_reset_work);
 363                 goto out;
 364         }
 365
 366         if (unlikely(event == LDC_EVENT_UP)) {
 367                 vio_link_state_change(vio, event);
 368                 goto out;
 369         }
 370
 371         if (unlikely(event != LDC_EVENT_DATA_READY)) {
 372                 pr_warn(PFX "Unexpected LDC event %d\n", event);
 373                 goto out;
 374         }
 375
 376         err = 0;
 377         while (1) {
 378                 union {
 379                         struct vio_msg_tag tag;
 380                         u64 raw[8];
 381                 } msgbuf;
 382
 383                 err = ldc_read(vio->lp, &msgbuf, sizeof(msgbuf));
 384                 if (unlikely(err < 0)) {
 385                         if (err == -ECONNRESET)
 386                                 vio_conn_reset(vio);
 387                         break;
 388                 }
 389                 if (err == 0)
 390                         break;
 391                 viodbg(DATA, "TAG [%02x:%02x:%04x:%08x]\n",
 392                        msgbuf.tag.type,
 393                        msgbuf.tag.stype,
 394                        msgbuf.tag.stype_env,
 395                        msgbuf.tag.sid);
 396                 err = vio_validate_sid(vio, &msgbuf.tag);
 397                 if (err < 0)
 398                         break;
 399
 400                 if (likely(msgbuf.tag.type == VIO_TYPE_DATA)) {
 401                         if (msgbuf.tag.stype == VIO_SUBTYPE_ACK)
 402                                 err = vdc_ack(port, &msgbuf);
 403                         else if (msgbuf.tag.stype == VIO_SUBTYPE_NACK)
 404                                 err = vdc_nack(port, &msgbuf);
 405                         else
 406                                 err = vdc_handle_unknown(port, &msgbuf);
 407                 } else if (msgbuf.tag.type == VIO_TYPE_CTRL) {
 408                         err = vio_control_pkt_engine(vio, &msgbuf);
 409                 } else {
 410                         err = vdc_handle_unknown(port, &msgbuf);
 411                 }
 412                 if (err < 0)
 413                         break;
 414         }
 415         if (err < 0)
 416                 vdc_finish(&port->vio, err, WAITING_FOR_ANY);
 417 out:
 418         spin_unlock_irqrestore(&vio->lock, flags);
 419 }
 420
 421 static int __vdc_tx_trigger(struct vdc_port *port)
 422 {
 423         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 424         struct vio_dring_data hdr = {
 425                 .tag = {
 426                         .type           = VIO_TYPE_DATA,
 427                         .stype          = VIO_SUBTYPE_INFO,
 428                         .stype_env      = VIO_DRING_DATA,
 429                         .sid            = vio_send_sid(&port->vio),
 430                 },
 431                 .dring_ident            = dr->ident,
 432                 .start_idx              = dr->prod,
 433                 .end_idx                = dr->prod,
 434         };
 435         int err, delay;
 436         int retries = 0;
 437
 438         hdr.seq = dr->snd_nxt;
 439         delay = 1;
 440         do {
 441                 err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
 442                 if (err > 0) {
 443                         dr->snd_nxt++;
 444                         break;
 445                 }
 446                 udelay(delay);
 447                 if ((delay <<= 1) > 128)
 448                         delay = 128;
 449                 if (retries++ > VDC_MAX_RETRIES)
 450                         break;
 451         } while (err == -EAGAIN);
 452
 453         if (err == -ENOTCONN)
 454                 vdc_ldc_reset(port);
 455         return err;
 456 }
 457
 458 static int __send_request(struct request *req)
 459 {
 460         struct vdc_port *port = req->rq_disk->private_data;
 461         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 462         struct scatterlist sg[MAX_RING_COOKIES];
 463         struct vdc_req_entry *rqe;
 464         struct vio_disk_desc *desc;
 465         unsigned int map_perm;
 466         int nsg, err, i;
 467         u64 len;
 468         u8 op;
 469
 470         if (WARN_ON(port->ring_cookies > MAX_RING_COOKIES))
 471                 return -EINVAL;
 472
 473         map_perm = LDC_MAP_SHADOW | LDC_MAP_DIRECT | LDC_MAP_IO;
 474
 475         if (rq_data_dir(req) == READ) {
 476                 map_perm |= LDC_MAP_W;
 477                 op = VD_OP_BREAD;
 478         } else {
 479                 map_perm |= LDC_MAP_R;
 480                 op = VD_OP_BWRITE;
 481         }
 482
 483         sg_init_table(sg, port->ring_cookies);
 484         nsg = blk_rq_map_sg(req->q, req, sg);
 485
 486         len = 0;
 487         for (i = 0; i < nsg; i++)
 488                 len += sg[i].length;
 489
 490         desc = vio_dring_cur(dr);
 491
 492         err = ldc_map_sg(port->vio.lp, sg, nsg,
 493                          desc->cookies, port->ring_cookies,
 494                          map_perm);
 495         if (err < 0) {
 496                 printk(KERN_ERR PFX "ldc_map_sg() failure, err=%d.\n", err);
 497                 return err;
 498         }
 499
 500         rqe = &port->rq_arr[dr->prod];
 501         rqe->req = req;
 502
 503         desc->hdr.ack = VIO_ACK_ENABLE;
 504         desc->req_id = port->req_id;
 505         desc->operation = op;
 506         if (port->vdisk_type == VD_DISK_TYPE_DISK) {
 507                 desc->slice = 0xff;
 508         } else {
 509                 desc->slice = 0;
 510         }
 511         desc->status = ~0;
 512         desc->offset = (blk_rq_pos(req) << 9) / port->vdisk_block_size;
 513         desc->size = len;
 514         desc->ncookies = err;
 515
 516         /* This has to be a non-SMP write barrier because we are writing
 517          * to memory which is shared with the peer LDOM.
 518          */
 519         wmb();
 520         desc->hdr.state = VIO_DESC_READY;
 521
 522         err = __vdc_tx_trigger(port);
 523         if (err < 0) {
 524                 printk(KERN_ERR PFX "vdc_tx_trigger() failure, err=%d\n", err);
 525         } else {
 526                 port->req_id++;
 527                 dr->prod = vio_dring_next(dr, dr->prod);
 528         }
 529
 530         return err;
 531 }
 532
 533 static void do_vdc_request(struct request_queue *rq)
 534 {
 535         struct request *req;
 536
 537         while ((req = blk_peek_request(rq)) != NULL) {
 538                 struct vdc_port *port;
 539                 struct vio_dring_state *dr;
 540
 541                 port = req->rq_disk->private_data;
 542                 dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 543                 if (unlikely(vdc_tx_dring_avail(dr) < 1))
 544                         goto wait;
 545
 546                 blk_start_request(req);
 547
 548                 if (__send_request(req) < 0) {
 549                         blk_requeue_request(rq, req);
 550 wait:
 551                         /* Avoid pointless unplugs. */
 552                         blk_stop_queue(rq);
 553                         break;
 554                 }
 555         }
 556 }
 557
 558 static int generic_request(struct vdc_port *port, u8 op, void *buf, int len)
 559 {
 560         struct vio_dring_state *dr;
 561         struct vio_completion comp;
 562         struct vio_disk_desc *desc;
 563         unsigned int map_perm;
 564         unsigned long flags;
 565         int op_len, err;
 566         void *req_buf;
 567
 568         if (!(((u64)1 << (u64)op) & port->operations))
 569                 return -EOPNOTSUPP;
 570
 571         switch (op) {
 572         case VD_OP_BREAD:
 573         case VD_OP_BWRITE:
 574         default:
 575                 return -EINVAL;
 576
 577         case VD_OP_FLUSH:
 578                 op_len = 0;
 579                 map_perm = 0;
 580                 break;
 581
 582         case VD_OP_GET_WCE:
 583                 op_len = sizeof(u32);
 584                 map_perm = LDC_MAP_W;
 585                 break;
 586
 587         case VD_OP_SET_WCE:
 588                 op_len = sizeof(u32);
 589                 map_perm = LDC_MAP_R;
 590                 break;
 591
 592         case VD_OP_GET_VTOC:
 593                 op_len = sizeof(struct vio_disk_vtoc);
 594                 map_perm = LDC_MAP_W;
 595                 break;
 596
 597         case VD_OP_SET_VTOC:
 598                 op_len = sizeof(struct vio_disk_vtoc);
 599                 map_perm = LDC_MAP_R;
 600                 break;
 601
 602         case VD_OP_GET_DISKGEOM:
 603                 op_len = sizeof(struct vio_disk_geom);
 604                 map_perm = LDC_MAP_W;
 605                 break;
 606
 607         case VD_OP_SET_DISKGEOM:
 608                 op_len = sizeof(struct vio_disk_geom);
 609                 map_perm = LDC_MAP_R;
 610                 break;
 611
 612         case VD_OP_SCSICMD:
 613                 op_len = 16;
 614                 map_perm = LDC_MAP_RW;
 615                 break;
 616
 617         case VD_OP_GET_DEVID:
 618                 op_len = sizeof(struct vio_disk_devid);
 619                 map_perm = LDC_MAP_W;
 620                 break;
 621
 622         case VD_OP_GET_EFI:
 623         case VD_OP_SET_EFI:
 624                 return -EOPNOTSUPP;
 625                 break;
 626         };
 627
 628         map_perm |= LDC_MAP_SHADOW | LDC_MAP_DIRECT | LDC_MAP_IO;
 629
 630         op_len = (op_len + 7) & ~7;
 631         req_buf = kzalloc(op_len, GFP_KERNEL);
 632         if (!req_buf)
 633                 return -ENOMEM;
 634
 635         if (len > op_len)
 636                 len = op_len;
 637
 638         if (map_perm & LDC_MAP_R)
 639                 memcpy(req_buf, buf, len);
 640
 641         spin_lock_irqsave(&port->vio.lock, flags);
 642
 643         dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 644
 645         /* XXX If we want to use this code generically we have to
 646          * XXX handle TX ring exhaustion etc.
 647          */
 648         desc = vio_dring_cur(dr);
 649
 650         err = ldc_map_single(port->vio.lp, req_buf, op_len,
 651                              desc->cookies, port->ring_cookies,
 652                              map_perm);
 653         if (err < 0) {
 654                 spin_unlock_irqrestore(&port->vio.lock, flags);
 655                 kfree(req_buf);
 656                 return err;
 657         }
 658
 659         init_completion(&comp.com);
 660         comp.waiting_for = WAITING_FOR_GEN_CMD;
 661         port->vio.cmp = &comp;
 662
 663         desc->hdr.ack = VIO_ACK_ENABLE;
 664         desc->req_id = port->req_id;
 665         desc->operation = op;
 666         desc->slice = 0;
 667         desc->status = ~0;
 668         desc->offset = 0;
 669         desc->size = op_len;
 670         desc->ncookies = err;
 671
 672         /* This has to be a non-SMP write barrier because we are writing
 673          * to memory which is shared with the peer LDOM.
 674          */
 675         wmb();
 676         desc->hdr.state = VIO_DESC_READY;
 677
 678         err = __vdc_tx_trigger(port);
 679         if (err >= 0) {
 680                 port->req_id++;
 681                 dr->prod = vio_dring_next(dr, dr->prod);
 682                 spin_unlock_irqrestore(&port->vio.lock, flags);
 683
 684                 wait_for_completion(&comp.com);
 685                 err = comp.err;
 686         } else {
 687                 port->vio.cmp = NULL;
 688                 spin_unlock_irqrestore(&port->vio.lock, flags);
 689         }
 690
 691         if (map_perm & LDC_MAP_W)
 692                 memcpy(buf, req_buf, len);
 693
 694         kfree(req_buf);
 695
 696         return err;
 697 }
 698
 699 static int vdc_alloc_tx_ring(struct vdc_port *port)
 700 {
 701         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 702         unsigned long len, entry_size;
 703         int ncookies;
 704         void *dring;
 705
 706         entry_size = sizeof(struct vio_disk_desc) +
 707                 (sizeof(struct ldc_trans_cookie) * port->ring_cookies);
 708         len = (VDC_TX_RING_SIZE * entry_size);
 709
 710         ncookies = VIO_MAX_RING_COOKIES;
 711         dring = ldc_alloc_exp_dring(port->vio.lp, len,
 712                                     dr->cookies, &ncookies,
 713                                     (LDC_MAP_SHADOW |
 714                                      LDC_MAP_DIRECT |
 715                                      LDC_MAP_RW));
 716         if (IS_ERR(dring))
 717                 return PTR_ERR(dring);
 718
 719         dr->base = dring;
 720         dr->entry_size = entry_size;
 721         dr->num_entries = VDC_TX_RING_SIZE;
 722         dr->prod = dr->cons = 0;
 723         dr->pending = VDC_TX_RING_SIZE;
 724         dr->ncookies = ncookies;
 725
 726         return 0;
 727 }
 728
 729 static void vdc_free_tx_ring(struct vdc_port *port)
 730 {
 731         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 732
 733         if (dr->base) {
 734                 ldc_free_exp_dring(port->vio.lp, dr->base,
 735                                    (dr->entry_size * dr->num_entries),
 736                                    dr->cookies, dr->ncookies);
 737                 dr->base = NULL;
 738                 dr->entry_size = 0;
 739                 dr->num_entries = 0;
 740                 dr->pending = 0;
 741                 dr->ncookies = 0;
 742         }
 743 }
 744
 745 static int vdc_port_up(struct vdc_port *port)
 746 {
 747         struct vio_completion comp;
 748
 749         init_completion(&comp.com);
 750         comp.err = 0;
 751         comp.waiting_for = WAITING_FOR_LINK_UP;
 752         port->vio.cmp = &comp;
 753
 754         vio_port_up(&port->vio);
 755         wait_for_completion(&comp.com);
 756         return comp.err;
 757 }
 758
 759 static void vdc_port_down(struct vdc_port *port)
 760 {
 761         ldc_disconnect(port->vio.lp);
 762         ldc_unbind(port->vio.lp);
 763         vdc_free_tx_ring(port);
 764         vio_ldc_free(&port->vio);
 765 }
 766
 767 static int probe_disk(struct vdc_port *port)
 768 {
 769         struct request_queue *q;
 770         struct gendisk *g;
 771         int err;
 772
 773         err = vdc_port_up(port);
 774         if (err)
 775                 return err;
 776
 777         /* Using version 1.2 means vdisk_phys_blksz should be set unless the
 778          * disk is reserved by another system.
 779          */
 780         if (vdc_version_supported(port, 1, 2) && !port->vdisk_phys_blksz)
 781                 return -ENODEV;
 782
 783         if (vdc_version_supported(port, 1, 1)) {
 784                 /* vdisk_size should be set during the handshake, if it wasn't
 785                  * then the underlying disk is reserved by another system
 786                  */
 787                 if (port->vdisk_size == -1)
 788                         return -ENODEV;
 789         } else {
 790                 struct vio_disk_geom geom;
 791
 792                 err = generic_request(port, VD_OP_GET_DISKGEOM,
 793                                       &geom, sizeof(geom));
 794                 if (err < 0) {
 795                         printk(KERN_ERR PFX "VD_OP_GET_DISKGEOM returns "
 796                                "error %d\n", err);
 797                         return err;
 798                 }
 799                 port->vdisk_size = ((u64)geom.num_cyl *
 800                                     (u64)geom.num_hd *
 801                                     (u64)geom.num_sec);
 802         }
 803
 804         q = blk_init_queue(do_vdc_request, &port->vio.lock);
 805         if (!q) {
 806                 printk(KERN_ERR PFX "%s: Could not allocate queue.\n",
 807                        port->vio.name);
 808                 return -ENOMEM;
 809         }
 810         g = alloc_disk(1 << PARTITION_SHIFT);
 811         if (!g) {
 812                 printk(KERN_ERR PFX "%s: Could not allocate gendisk.\n",
 813                        port->vio.name);
 814                 blk_cleanup_queue(q);
 815                 return -ENOMEM;
 816         }
 817
 818         port->disk = g;
 819
 820         /* Each segment in a request is up to an aligned page in size. */
 821         blk_queue_segment_boundary(q, PAGE_SIZE - 1);
 822         blk_queue_max_segment_size(q, PAGE_SIZE);
 823
 824         blk_queue_max_segments(q, port->ring_cookies);
 825         blk_queue_max_hw_sectors(q, port->max_xfer_size);
 826         g->major = vdc_major;
 827         g->first_minor = port->vio.vdev->dev_no << PARTITION_SHIFT;
 828         strcpy(g->disk_name, port->disk_name);
 829
 830         g->fops = &vdc_fops;
 831         g->queue = q;
 832         g->private_data = port;
 833
 834         set_capacity(g, port->vdisk_size);
 835
 836         if (vdc_version_supported(port, 1, 1)) {
 837                 switch (port->vdisk_mtype) {
 838                 case VD_MEDIA_TYPE_CD:
 839                         pr_info(PFX "Virtual CDROM %s\n", port->disk_name);
 840                         g->flags |= GENHD_FL_CD;
 841                         g->flags |= GENHD_FL_REMOVABLE;
 842                         set_disk_ro(g, 1);
 843                         break;
 844
 845                 case VD_MEDIA_TYPE_DVD:
 846                         pr_info(PFX "Virtual DVD %s\n", port->disk_name);
 847                         g->flags |= GENHD_FL_CD;
 848                         g->flags |= GENHD_FL_REMOVABLE;
 849                         set_disk_ro(g, 1);
 850                         break;
 851
 852                 case VD_MEDIA_TYPE_FIXED:
 853                         pr_info(PFX "Virtual Hard disk %s\n", port->disk_name);
 854                         break;
 855                 }
 856         }
 857
 858         blk_queue_physical_block_size(q, port->vdisk_phys_blksz);
 859
 860         pr_info(PFX "%s: %u sectors (%u MB) protocol %d.%d\n",
 861                g->disk_name,
 862                port->vdisk_size, (port->vdisk_size >> (20 - 9)),
 863                port->vio.ver.major, port->vio.ver.minor);
 864
 865         device_add_disk(&port->vio.vdev->dev, g, NULL);
 866
 867         return 0;
 868 }
 869
 870 static struct ldc_channel_config vdc_ldc_cfg = {
 871         .event          = vdc_event,
 872         .mtu            = 64,
 873         .mode           = LDC_MODE_UNRELIABLE,
 874 };
 875
 876 static struct vio_driver_ops vdc_vio_ops = {
 877         .send_attr              = vdc_send_attr,
 878         .handle_attr            = vdc_handle_attr,
 879         .handshake_complete     = vdc_handshake_complete,
 880 };
 881
 882 static void print_version(void)
 883 {
 884         static int version_printed;
 885
 886         if (version_printed++ == 0)
 887                 printk(KERN_INFO "%s", version);
 888 }
 889
 890 struct vdc_check_port_data {
 891         int     dev_no;
 892         char    *type;
 893 };
 894
 895 static int vdc_device_probed(struct device *dev, void *arg)
 896 {
 897         struct vio_dev *vdev = to_vio_dev(dev);
 898         struct vdc_check_port_data *port_data;
 899
 900         port_data = (struct vdc_check_port_data *)arg;
 901
 902         if ((vdev->dev_no == port_data->dev_no) &&
 903             (!(strcmp((char *)&vdev->type, port_data->type))) &&
 904                 dev_get_drvdata(dev)) {
 905                 /* This device has already been configured
 906                  * by vdc_port_probe()
 907                  */
 908                 return 1;
 909         } else {
 910                 return 0;
 911         }
 912 }
 913
 914 /* Determine whether the VIO device is part of an mpgroup
 915  * by locating all the virtual-device-port nodes associated
 916  * with the parent virtual-device node for the VIO device
 917  * and checking whether any of these nodes are vdc-ports
 918  * which have already been configured.
 919  *
 920  * Returns true if this device is part of an mpgroup and has
 921  * already been probed.
 922  */
 923 static bool vdc_port_mpgroup_check(struct vio_dev *vdev)
 924 {
 925         struct vdc_check_port_data port_data;
 926         struct device *dev;
 927
 928         port_data.dev_no = vdev->dev_no;
 929         port_data.type = (char *)&vdev->type;
 930
 931         dev = device_find_child(vdev->dev.parent, &port_data,
 932                                 vdc_device_probed);
 933
 934         if (dev)
 935                 return true;
 936
 937         return false;
 938 }
 939
 940 static int vdc_port_probe(struct vio_dev *vdev, const struct vio_device_id *id)
 941 {
 942         struct mdesc_handle *hp;
 943         struct vdc_port *port;
 944         int err;
 945         const u64 *ldc_timeout;
 946
 947         print_version();
 948
 949         hp = mdesc_grab();
 950
 951         err = -ENODEV;
 952         if ((vdev->dev_no << PARTITION_SHIFT) & ~(u64)MINORMASK) {
 953                 printk(KERN_ERR PFX "Port id [%llu] too large.\n",
 954                        vdev->dev_no);
 955                 goto err_out_release_mdesc;
 956         }
 957
 958         /* Check if this device is part of an mpgroup */
 959         if (vdc_port_mpgroup_check(vdev)) {
 960                 printk(KERN_WARNING
 961                         "VIO: Ignoring extra vdisk port %s",
 962                         dev_name(&vdev->dev));
 963                 goto err_out_release_mdesc;
 964         }
 965
 966         port = kzalloc(sizeof(*port), GFP_KERNEL);
 967         err = -ENOMEM;
 968         if (!port) {
 969                 printk(KERN_ERR PFX "Cannot allocate vdc_port.\n");
 970                 goto err_out_release_mdesc;
 971         }
 972
 973         if (vdev->dev_no >= 26)
 974                 snprintf(port->disk_name, sizeof(port->disk_name),
 975                          VDCBLK_NAME "%c%c",
 976                          'a' + ((int)vdev->dev_no / 26) - 1,
 977                          'a' + ((int)vdev->dev_no % 26));
 978         else
 979                 snprintf(port->disk_name, sizeof(port->disk_name),
 980                          VDCBLK_NAME "%c", 'a' + ((int)vdev->dev_no % 26));
 981         port->vdisk_size = -1;
 982
 983         /* Actual wall time may be double due to do_generic_file_read() doing
 984          * a readahead I/O first, and once that fails it will try to read a
 985          * single page.
 986          */
 987         ldc_timeout = mdesc_get_property(hp, vdev->mp, "vdc-timeout", NULL);
 988         port->ldc_timeout = ldc_timeout ? *ldc_timeout : 0;
 989         timer_setup(&port->ldc_reset_timer, vdc_ldc_reset_timer, 0);
 990         INIT_WORK(&port->ldc_reset_work, vdc_ldc_reset_work);
 991
 992         err = vio_driver_init(&port->vio, vdev, VDEV_DISK,
 993                               vdc_versions, ARRAY_SIZE(vdc_versions),
 994                               &vdc_vio_ops, port->disk_name);
 995         if (err)
 996                 goto err_out_free_port;
 997
 998         port->vdisk_block_size = VDC_DEFAULT_BLK_SIZE;
 999         port->max_xfer_size = MAX_XFER_SIZE;
1000         port->ring_cookies = MAX_RING_COOKIES;
1001
1002         err = vio_ldc_alloc(&port->vio, &vdc_ldc_cfg, port);
1003         if (err)
1004                 goto err_out_free_port;
1005
1006         err = vdc_alloc_tx_ring(port);
1007         if (err)
1008                 goto err_out_free_ldc;
1009
1010         err = probe_disk(port);
1011         if (err)
1012                 goto err_out_free_tx_ring;
1013
1014         /* Note that the device driver_data is used to determine
1015          * whether the port has been probed.
1016          */
1017         dev_set_drvdata(&vdev->dev, port);
1018
1019         mdesc_release(hp);
1020
1021         return 0;
1022
1023 err_out_free_tx_ring:
1024         vdc_free_tx_ring(port);
1025
1026 err_out_free_ldc:
1027         vio_ldc_free(&port->vio);
1028
1029 err_out_free_port:
1030         kfree(port);
1031
1032 err_out_release_mdesc:
1033         mdesc_release(hp);
1034         return err;
1035 }
1036
1037 static int vdc_port_remove(struct vio_dev *vdev)
1038 {
1039         struct vdc_port *port = dev_get_drvdata(&vdev->dev);
1040
1041         if (port) {
1042                 unsigned long flags;
1043
1044                 spin_lock_irqsave(&port->vio.lock, flags);
1045                 blk_stop_queue(port->disk->queue);
1046                 spin_unlock_irqrestore(&port->vio.lock, flags);
1047
1048                 flush_work(&port->ldc_reset_work);
1049                 del_timer_sync(&port->ldc_reset_timer);
1050                 del_timer_sync(&port->vio.timer);
1051
1052                 del_gendisk(port->disk);
1053                 blk_cleanup_queue(port->disk->queue);
1054                 put_disk(port->disk);
1055                 port->disk = NULL;
1056
1057                 vdc_free_tx_ring(port);
1058                 vio_ldc_free(&port->vio);
1059
1060                 dev_set_drvdata(&vdev->dev, NULL);
1061
1062                 kfree(port);
1063         }
1064         return 0;
1065 }
1066
1067 static void vdc_requeue_inflight(struct vdc_port *port)
1068 {
1069         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1070         u32 idx;
1071
1072         for (idx = dr->cons; idx != dr->prod; idx = vio_dring_next(dr, idx)) {
1073                 struct vio_disk_desc *desc = vio_dring_entry(dr, idx);
1074                 struct vdc_req_entry *rqe = &port->rq_arr[idx];
1075                 struct request *req;
1076
1077                 ldc_unmap(port->vio.lp, desc->cookies, desc->ncookies);
1078                 desc->hdr.state = VIO_DESC_FREE;
1079                 dr->cons = vio_dring_next(dr, idx);
1080
1081                 req = rqe->req;
1082                 if (req == NULL) {
1083                         vdc_end_special(port, desc);
1084                         continue;
1085                 }
1086
1087                 rqe->req = NULL;
1088                 blk_requeue_request(port->disk->queue, req);
1089         }
1090 }
1091
1092 static void vdc_queue_drain(struct vdc_port *port)
1093 {
1094         struct request *req;
1095
1096         while ((req = blk_fetch_request(port->disk->queue)) != NULL)
1097                 __blk_end_request_all(req, BLK_STS_IOERR);
1098 }
1099
1100 static void vdc_ldc_reset_timer(struct timer_list *t)
1101 {
1102         struct vdc_port *port = from_timer(port, t, ldc_reset_timer);
1103         struct vio_driver_state *vio = &port->vio;
1104         unsigned long flags;
1105
1106         spin_lock_irqsave(&vio->lock, flags);
1107         if (!(port->vio.hs_state & VIO_HS_COMPLETE)) {
1108                 pr_warn(PFX "%s ldc down %llu seconds, draining queue\n",
1109                         port->disk_name, port->ldc_timeout);
1110                 vdc_queue_drain(port);
1111                 vdc_blk_queue_start(port);
1112         }
1113         spin_unlock_irqrestore(&vio->lock, flags);
1114 }
1115
1116 static void vdc_ldc_reset_work(struct work_struct *work)
1117 {
1118         struct vdc_port *port;
1119         struct vio_driver_state *vio;
1120         unsigned long flags;
1121
1122         port = container_of(work, struct vdc_port, ldc_reset_work);
1123         vio = &port->vio;
1124
1125         spin_lock_irqsave(&vio->lock, flags);
1126         vdc_ldc_reset(port);
1127         spin_unlock_irqrestore(&vio->lock, flags);
1128 }
1129
1130 static void vdc_ldc_reset(struct vdc_port *port)
1131 {
1132         int err;
1133
1134         assert_spin_locked(&port->vio.lock);
1135
1136         pr_warn(PFX "%s ldc link reset\n", port->disk_name);
1137         blk_stop_queue(port->disk->queue);
1138         vdc_requeue_inflight(port);
1139         vdc_port_down(port);
1140
1141         err = vio_ldc_alloc(&port->vio, &vdc_ldc_cfg, port);
1142         if (err) {
1143                 pr_err(PFX "%s vio_ldc_alloc:%d\n", port->disk_name, err);
1144                 return;
1145         }
1146
1147         err = vdc_alloc_tx_ring(port);
1148         if (err) {
1149                 pr_err(PFX "%s vio_alloc_tx_ring:%d\n", port->disk_name, err);
1150                 goto err_free_ldc;
1151         }
1152
1153         if (port->ldc_timeout)
1154                 mod_timer(&port->ldc_reset_timer,
1155                           round_jiffies(jiffies + HZ * port->ldc_timeout));
1156         mod_timer(&port->vio.timer, round_jiffies(jiffies + HZ));
1157         return;
1158
1159 err_free_ldc:
1160         vio_ldc_free(&port->vio);
1161 }
1162
1163 static const struct vio_device_id vdc_port_match[] = {
1164         {
1165                 .type = "vdc-port",
1166         },
1167         {},
1168 };
1169 MODULE_DEVICE_TABLE(vio, vdc_port_match);
1170
1171 static struct vio_driver vdc_port_driver = {
1172         .id_table       = vdc_port_match,
1173         .probe          = vdc_port_probe,
1174         .remove         = vdc_port_remove,
1175         .name           = "vdc_port",
1176 };
1177
1178 static int __init vdc_init(void)
1179 {
1180         int err;
1181
1182         sunvdc_wq = alloc_workqueue("sunvdc", 0, 0);
1183         if (!sunvdc_wq)
1184                 return -ENOMEM;
1185
1186         err = register_blkdev(0, VDCBLK_NAME);
1187         if (err < 0)
1188                 goto out_free_wq;
1189
1190         vdc_major = err;
1191
1192         err = vio_register_driver(&vdc_port_driver);
1193         if (err)
1194                 goto out_unregister_blkdev;
1195
1196         return 0;
1197
1198 out_unregister_blkdev:
1199         unregister_blkdev(vdc_major, VDCBLK_NAME);
1200         vdc_major = 0;
1201
1202 out_free_wq:
1203         destroy_workqueue(sunvdc_wq);
1204         return err;
1205 }
1206
1207 static void __exit vdc_exit(void)
1208 {
1209         vio_unregister_driver(&vdc_port_driver);
1210         unregister_blkdev(vdc_major, VDCBLK_NAME);
1211         destroy_workqueue(sunvdc_wq);
1212 }
1213
1214 module_init(vdc_init);
1215 module_exit(vdc_exit);