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