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 #define VDC_MAX_RETRIES 10
  45
  46 static struct workqueue_struct *sunvdc_wq;
  47
  48 struct vdc_req_entry {
  49         struct request          *req;
  50 };
  51
  52 struct vdc_port {
  53         struct vio_driver_state vio;
  54
  55         struct gendisk          *disk;
  56
  57         struct vdc_completion   *cmp;
  58
  59         u64                     req_id;
  60         u64                     seq;
  61         struct vdc_req_entry    rq_arr[VDC_TX_RING_SIZE];
  62
  63         unsigned long           ring_cookies;
  64
  65         u64                     max_xfer_size;
  66         u32                     vdisk_block_size;
  67
  68         u64                     ldc_timeout;
  69         struct timer_list       ldc_reset_timer;
  70         struct work_struct      ldc_reset_work;
  71
  72         /* The server fills these in for us in the disk attribute
  73          * ACK packet.
  74          */
  75         u64                     operations;
  76         u32                     vdisk_size;
  77         u8                      vdisk_type;
  78         u8                      vdisk_mtype;
  79         u32                     vdisk_phys_blksz;
  80
  81         char                    disk_name[32];
  82 };
  83
  84 static void vdc_ldc_reset(struct vdc_port *port);
  85 static void vdc_ldc_reset_work(struct work_struct *work);
  86 static void vdc_ldc_reset_timer(unsigned long _arg);
  87
  88 static inline struct vdc_port *to_vdc_port(struct vio_driver_state *vio)
  89 {
  90         return container_of(vio, struct vdc_port, vio);
  91 }
  92
  93 /* Ordered from largest major to lowest */
  94 static struct vio_version vdc_versions[] = {
  95         { .major = 1, .minor = 2 },
  96         { .major = 1, .minor = 1 },
  97         { .major = 1, .minor = 0 },
  98 };
  99
 100 static inline int vdc_version_supported(struct vdc_port *port,
 101                                         u16 major, u16 minor)
 102 {
 103         return port->vio.ver.major == major && port->vio.ver.minor >= minor;
 104 }
 105
 106 #define VDCBLK_NAME     "vdisk"
 107 static int vdc_major;
 108 #define PARTITION_SHIFT 3
 109
 110 static inline u32 vdc_tx_dring_avail(struct vio_dring_state *dr)
 111 {
 112         return vio_dring_avail(dr, VDC_TX_RING_SIZE);
 113 }
 114
 115 static int vdc_getgeo(struct block_device *bdev, struct hd_geometry *geo)
 116 {
 117         struct gendisk *disk = bdev->bd_disk;
 118         sector_t nsect = get_capacity(disk);
 119         sector_t cylinders = nsect;
 120
 121         geo->heads = 0xff;
 122         geo->sectors = 0x3f;
 123         sector_div(cylinders, geo->heads * geo->sectors);
 124         geo->cylinders = cylinders;
 125         if ((sector_t)(geo->cylinders + 1) * geo->heads * geo->sectors < nsect)
 126                 geo->cylinders = 0xffff;
 127
 128         return 0;
 129 }
 130
 131 /* Add ioctl/CDROM_GET_CAPABILITY to support cdrom_id in udev
 132  * when vdisk_mtype is VD_MEDIA_TYPE_CD or VD_MEDIA_TYPE_DVD.
 133  * Needed to be able to install inside an ldom from an iso image.
 134  */
 135 static int vdc_ioctl(struct block_device *bdev, fmode_t mode,
 136                      unsigned command, unsigned long argument)
 137 {
 138         int i;
 139         struct gendisk *disk;
 140
 141         switch (command) {
 142         case CDROMMULTISESSION:
 143                 pr_debug(PFX "Multisession CDs not supported\n");
 144                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 145                         if (put_user(0, (char __user *)(argument + i)))
 146                                 return -EFAULT;
 147                 return 0;
 148
 149         case CDROM_GET_CAPABILITY:
 150                 disk = bdev->bd_disk;
 151
 152                 if (bdev->bd_disk && (disk->flags & GENHD_FL_CD))
 153                         return 0;
 154                 return -EINVAL;
 155
 156         default:
 157                 pr_debug(PFX "ioctl %08x not supported\n", command);
 158                 return -EINVAL;
 159         }
 160 }
 161
 162 static const struct block_device_operations vdc_fops = {
 163         .owner          = THIS_MODULE,
 164         .getgeo         = vdc_getgeo,
 165         .ioctl          = vdc_ioctl,
 166 };
 167
 168 static void vdc_blk_queue_start(struct vdc_port *port)
 169 {
 170         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 171
 172         /* restart blk queue when ring is half emptied. also called after
 173          * handshake completes, so check for initial handshake before we've
 174          * allocated a disk.
 175          */
 176         if (port->disk && blk_queue_stopped(port->disk->queue) &&
 177             vdc_tx_dring_avail(dr) * 100 / VDC_TX_RING_SIZE >= 50) {
 178                 blk_start_queue(port->disk->queue);
 179         }
 180
 181 }
 182
 183 static void vdc_finish(struct vio_driver_state *vio, int err, int waiting_for)
 184 {
 185         if (vio->cmp &&
 186             (waiting_for == -1 ||
 187              vio->cmp->waiting_for == waiting_for)) {
 188                 vio->cmp->err = err;
 189                 complete(&vio->cmp->com);
 190                 vio->cmp = NULL;
 191         }
 192 }
 193
 194 static void vdc_handshake_complete(struct vio_driver_state *vio)
 195 {
 196         struct vdc_port *port = to_vdc_port(vio);
 197
 198         del_timer(&port->ldc_reset_timer);
 199         vdc_finish(vio, 0, WAITING_FOR_LINK_UP);
 200         vdc_blk_queue_start(port);
 201 }
 202
 203 static int vdc_handle_unknown(struct vdc_port *port, void *arg)
 204 {
 205         struct vio_msg_tag *pkt = arg;
 206
 207         printk(KERN_ERR PFX "Received unknown msg [%02x:%02x:%04x:%08x]\n",
 208                pkt->type, pkt->stype, pkt->stype_env, pkt->sid);
 209         printk(KERN_ERR PFX "Resetting connection.\n");
 210
 211         ldc_disconnect(port->vio.lp);
 212
 213         return -ECONNRESET;
 214 }
 215
 216 static int vdc_send_attr(struct vio_driver_state *vio)
 217 {
 218         struct vdc_port *port = to_vdc_port(vio);
 219         struct vio_disk_attr_info pkt;
 220
 221         memset(&pkt, 0, sizeof(pkt));
 222
 223         pkt.tag.type = VIO_TYPE_CTRL;
 224         pkt.tag.stype = VIO_SUBTYPE_INFO;
 225         pkt.tag.stype_env = VIO_ATTR_INFO;
 226         pkt.tag.sid = vio_send_sid(vio);
 227
 228         pkt.xfer_mode = VIO_DRING_MODE;
 229         pkt.vdisk_block_size = port->vdisk_block_size;
 230         pkt.max_xfer_size = port->max_xfer_size;
 231
 232         viodbg(HS, "SEND ATTR xfer_mode[0x%x] blksz[%u] max_xfer[%llu]\n",
 233                pkt.xfer_mode, pkt.vdisk_block_size, pkt.max_xfer_size);
 234
 235         return vio_ldc_send(&port->vio, &pkt, sizeof(pkt));
 236 }
 237
 238 static int vdc_handle_attr(struct vio_driver_state *vio, void *arg)
 239 {
 240         struct vdc_port *port = to_vdc_port(vio);
 241         struct vio_disk_attr_info *pkt = arg;
 242
 243         viodbg(HS, "GOT ATTR stype[0x%x] ops[%llx] disk_size[%llu] disk_type[%x] "
 244                "mtype[0x%x] xfer_mode[0x%x] blksz[%u] max_xfer[%llu]\n",
 245                pkt->tag.stype, pkt->operations,
 246                pkt->vdisk_size, pkt->vdisk_type, pkt->vdisk_mtype,
 247                pkt->xfer_mode, pkt->vdisk_block_size,
 248                pkt->max_xfer_size);
 249
 250         if (pkt->tag.stype == VIO_SUBTYPE_ACK) {
 251                 switch (pkt->vdisk_type) {
 252                 case VD_DISK_TYPE_DISK:
 253                 case VD_DISK_TYPE_SLICE:
 254                         break;
 255
 256                 default:
 257                         printk(KERN_ERR PFX "%s: Bogus vdisk_type 0x%x\n",
 258                                vio->name, pkt->vdisk_type);
 259                         return -ECONNRESET;
 260                 }
 261
 262                 if (pkt->vdisk_block_size > port->vdisk_block_size) {
 263                         printk(KERN_ERR PFX "%s: BLOCK size increased "
 264                                "%u --> %u\n",
 265                                vio->name,
 266                                port->vdisk_block_size, pkt->vdisk_block_size);
 267                         return -ECONNRESET;
 268                 }
 269
 270                 port->operations = pkt->operations;
 271                 port->vdisk_type = pkt->vdisk_type;
 272                 if (vdc_version_supported(port, 1, 1)) {
 273                         port->vdisk_size = pkt->vdisk_size;
 274                         port->vdisk_mtype = pkt->vdisk_mtype;
 275                 }
 276                 if (pkt->max_xfer_size < port->max_xfer_size)
 277                         port->max_xfer_size = pkt->max_xfer_size;
 278                 port->vdisk_block_size = pkt->vdisk_block_size;
 279
 280                 port->vdisk_phys_blksz = VDC_DEFAULT_BLK_SIZE;
 281                 if (vdc_version_supported(port, 1, 2))
 282                         port->vdisk_phys_blksz = pkt->phys_block_size;
 283
 284                 return 0;
 285         } else {
 286                 printk(KERN_ERR PFX "%s: Attribute NACK\n", vio->name);
 287
 288                 return -ECONNRESET;
 289         }
 290 }
 291
 292 static void vdc_end_special(struct vdc_port *port, struct vio_disk_desc *desc)
 293 {
 294         int err = desc->status;
 295
 296         vdc_finish(&port->vio, -err, WAITING_FOR_GEN_CMD);
 297 }
 298
 299 static void vdc_end_one(struct vdc_port *port, struct vio_dring_state *dr,
 300                         unsigned int index)
 301 {
 302         struct vio_disk_desc *desc = vio_dring_entry(dr, index);
 303         struct vdc_req_entry *rqe = &port->rq_arr[index];
 304         struct request *req;
 305
 306         if (unlikely(desc->hdr.state != VIO_DESC_DONE))
 307                 return;
 308
 309         ldc_unmap(port->vio.lp, desc->cookies, desc->ncookies);
 310         desc->hdr.state = VIO_DESC_FREE;
 311         dr->cons = vio_dring_next(dr, index);
 312
 313         req = rqe->req;
 314         if (req == NULL) {
 315                 vdc_end_special(port, desc);
 316                 return;
 317         }
 318
 319         rqe->req = NULL;
 320
 321         __blk_end_request(req, (desc->status ? BLK_STS_IOERR : 0), desc->size);
 322
 323         vdc_blk_queue_start(port);
 324 }
 325
 326 static int vdc_ack(struct vdc_port *port, void *msgbuf)
 327 {
 328         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 329         struct vio_dring_data *pkt = msgbuf;
 330
 331         if (unlikely(pkt->dring_ident != dr->ident ||
 332                      pkt->start_idx != pkt->end_idx ||
 333                      pkt->start_idx >= VDC_TX_RING_SIZE))
 334                 return 0;
 335
 336         vdc_end_one(port, dr, pkt->start_idx);
 337
 338         return 0;
 339 }
 340
 341 static int vdc_nack(struct vdc_port *port, void *msgbuf)
 342 {
 343         /* XXX Implement me XXX */
 344         return 0;
 345 }
 346
 347 static void vdc_event(void *arg, int event)
 348 {
 349         struct vdc_port *port = arg;
 350         struct vio_driver_state *vio = &port->vio;
 351         unsigned long flags;
 352         int err;
 353
 354         spin_lock_irqsave(&vio->lock, flags);
 355
 356         if (unlikely(event == LDC_EVENT_RESET)) {
 357                 vio_link_state_change(vio, event);
 358                 queue_work(sunvdc_wq, &port->ldc_reset_work);
 359                 goto out;
 360         }
 361
 362         if (unlikely(event == LDC_EVENT_UP)) {
 363                 vio_link_state_change(vio, event);
 364                 goto out;
 365         }
 366
 367         if (unlikely(event != LDC_EVENT_DATA_READY)) {
 368                 pr_warn(PFX "Unexpected LDC event %d\n", event);
 369                 goto out;
 370         }
 371
 372         err = 0;
 373         while (1) {
 374                 union {
 375                         struct vio_msg_tag tag;
 376                         u64 raw[8];
 377                 } msgbuf;
 378
 379                 err = ldc_read(vio->lp, &msgbuf, sizeof(msgbuf));
 380                 if (unlikely(err < 0)) {
 381                         if (err == -ECONNRESET)
 382                                 vio_conn_reset(vio);
 383                         break;
 384                 }
 385                 if (err == 0)
 386                         break;
 387                 viodbg(DATA, "TAG [%02x:%02x:%04x:%08x]\n",
 388                        msgbuf.tag.type,
 389                        msgbuf.tag.stype,
 390                        msgbuf.tag.stype_env,
 391                        msgbuf.tag.sid);
 392                 err = vio_validate_sid(vio, &msgbuf.tag);
 393                 if (err < 0)
 394                         break;
 395
 396                 if (likely(msgbuf.tag.type == VIO_TYPE_DATA)) {
 397                         if (msgbuf.tag.stype == VIO_SUBTYPE_ACK)
 398                                 err = vdc_ack(port, &msgbuf);
 399                         else if (msgbuf.tag.stype == VIO_SUBTYPE_NACK)
 400                                 err = vdc_nack(port, &msgbuf);
 401                         else
 402                                 err = vdc_handle_unknown(port, &msgbuf);
 403                 } else if (msgbuf.tag.type == VIO_TYPE_CTRL) {
 404                         err = vio_control_pkt_engine(vio, &msgbuf);
 405                 } else {
 406                         err = vdc_handle_unknown(port, &msgbuf);
 407                 }
 408                 if (err < 0)
 409                         break;
 410         }
 411         if (err < 0)
 412                 vdc_finish(&port->vio, err, WAITING_FOR_ANY);
 413 out:
 414         spin_unlock_irqrestore(&vio->lock, flags);
 415 }
 416
 417 static int __vdc_tx_trigger(struct vdc_port *port)
 418 {
 419         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 420         struct vio_dring_data hdr = {
 421                 .tag = {
 422                         .type           = VIO_TYPE_DATA,
 423                         .stype          = VIO_SUBTYPE_INFO,
 424                         .stype_env      = VIO_DRING_DATA,
 425                         .sid            = vio_send_sid(&port->vio),
 426                 },
 427                 .dring_ident            = dr->ident,
 428                 .start_idx              = dr->prod,
 429                 .end_idx                = dr->prod,
 430         };
 431         int err, delay;
 432         int retries = 0;
 433
 434         hdr.seq = dr->snd_nxt;
 435         delay = 1;
 436         do {
 437                 err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
 438                 if (err > 0) {
 439                         dr->snd_nxt++;
 440                         break;
 441                 }
 442                 udelay(delay);
 443                 if ((delay <<= 1) > 128)
 444                         delay = 128;
 445                 if (retries++ > VDC_MAX_RETRIES)
 446                         break;
 447         } while (err == -EAGAIN);
 448
 449         if (err == -ENOTCONN)
 450                 vdc_ldc_reset(port);
 451         return err;
 452 }
 453
 454 static int __send_request(struct request *req)
 455 {
 456         struct vdc_port *port = req->rq_disk->private_data;
 457         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 458         struct scatterlist sg[port->ring_cookies];
 459         struct vdc_req_entry *rqe;
 460         struct vio_disk_desc *desc;
 461         unsigned int map_perm;
 462         int nsg, err, i;
 463         u64 len;
 464         u8 op;
 465
 466         map_perm = LDC_MAP_SHADOW | LDC_MAP_DIRECT | LDC_MAP_IO;
 467
 468         if (rq_data_dir(req) == READ) {
 469                 map_perm |= LDC_MAP_W;
 470                 op = VD_OP_BREAD;
 471         } else {
 472                 map_perm |= LDC_MAP_R;
 473                 op = VD_OP_BWRITE;
 474         }
 475
 476         sg_init_table(sg, port->ring_cookies);
 477         nsg = blk_rq_map_sg(req->q, req, sg);
 478
 479         len = 0;
 480         for (i = 0; i < nsg; i++)
 481                 len += sg[i].length;
 482
 483         desc = vio_dring_cur(dr);
 484
 485         err = ldc_map_sg(port->vio.lp, sg, nsg,
 486                          desc->cookies, port->ring_cookies,
 487                          map_perm);
 488         if (err < 0) {
 489                 printk(KERN_ERR PFX "ldc_map_sg() failure, err=%d.\n", err);
 490                 return err;
 491         }
 492
 493         rqe = &port->rq_arr[dr->prod];
 494         rqe->req = req;
 495
 496         desc->hdr.ack = VIO_ACK_ENABLE;
 497         desc->req_id = port->req_id;
 498         desc->operation = op;
 499         if (port->vdisk_type == VD_DISK_TYPE_DISK) {
 500                 desc->slice = 0xff;
 501         } else {
 502                 desc->slice = 0;
 503         }
 504         desc->status = ~0;
 505         desc->offset = (blk_rq_pos(req) << 9) / port->vdisk_block_size;
 506         desc->size = len;
 507         desc->ncookies = err;
 508
 509         /* This has to be a non-SMP write barrier because we are writing
 510          * to memory which is shared with the peer LDOM.
 511          */
 512         wmb();
 513         desc->hdr.state = VIO_DESC_READY;
 514
 515         err = __vdc_tx_trigger(port);
 516         if (err < 0) {
 517                 printk(KERN_ERR PFX "vdc_tx_trigger() failure, err=%d\n", err);
 518         } else {
 519                 port->req_id++;
 520                 dr->prod = vio_dring_next(dr, dr->prod);
 521         }
 522
 523         return err;
 524 }
 525
 526 static void do_vdc_request(struct request_queue *rq)
 527 {
 528         struct request *req;
 529
 530         while ((req = blk_peek_request(rq)) != NULL) {
 531                 struct vdc_port *port;
 532                 struct vio_dring_state *dr;
 533
 534                 port = req->rq_disk->private_data;
 535                 dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 536                 if (unlikely(vdc_tx_dring_avail(dr) < 1))
 537                         goto wait;
 538
 539                 blk_start_request(req);
 540
 541                 if (__send_request(req) < 0) {
 542                         blk_requeue_request(rq, req);
 543 wait:
 544                         /* Avoid pointless unplugs. */
 545                         blk_stop_queue(rq);
 546                         break;
 547                 }
 548         }
 549 }
 550
 551 static int generic_request(struct vdc_port *port, u8 op, void *buf, int len)
 552 {
 553         struct vio_dring_state *dr;
 554         struct vio_completion comp;
 555         struct vio_disk_desc *desc;
 556         unsigned int map_perm;
 557         unsigned long flags;
 558         int op_len, err;
 559         void *req_buf;
 560
 561         if (!(((u64)1 << (u64)op) & port->operations))
 562                 return -EOPNOTSUPP;
 563
 564         switch (op) {
 565         case VD_OP_BREAD:
 566         case VD_OP_BWRITE:
 567         default:
 568                 return -EINVAL;
 569
 570         case VD_OP_FLUSH:
 571                 op_len = 0;
 572                 map_perm = 0;
 573                 break;
 574
 575         case VD_OP_GET_WCE:
 576                 op_len = sizeof(u32);
 577                 map_perm = LDC_MAP_W;
 578                 break;
 579
 580         case VD_OP_SET_WCE:
 581                 op_len = sizeof(u32);
 582                 map_perm = LDC_MAP_R;
 583                 break;
 584
 585         case VD_OP_GET_VTOC:
 586                 op_len = sizeof(struct vio_disk_vtoc);
 587                 map_perm = LDC_MAP_W;
 588                 break;
 589
 590         case VD_OP_SET_VTOC:
 591                 op_len = sizeof(struct vio_disk_vtoc);
 592                 map_perm = LDC_MAP_R;
 593                 break;
 594
 595         case VD_OP_GET_DISKGEOM:
 596                 op_len = sizeof(struct vio_disk_geom);
 597                 map_perm = LDC_MAP_W;
 598                 break;
 599
 600         case VD_OP_SET_DISKGEOM:
 601                 op_len = sizeof(struct vio_disk_geom);
 602                 map_perm = LDC_MAP_R;
 603                 break;
 604
 605         case VD_OP_SCSICMD:
 606                 op_len = 16;
 607                 map_perm = LDC_MAP_RW;
 608                 break;
 609
 610         case VD_OP_GET_DEVID:
 611                 op_len = sizeof(struct vio_disk_devid);
 612                 map_perm = LDC_MAP_W;
 613                 break;
 614
 615         case VD_OP_GET_EFI:
 616         case VD_OP_SET_EFI:
 617                 return -EOPNOTSUPP;
 618                 break;
 619         };
 620
 621         map_perm |= LDC_MAP_SHADOW | LDC_MAP_DIRECT | LDC_MAP_IO;
 622
 623         op_len = (op_len + 7) & ~7;
 624         req_buf = kzalloc(op_len, GFP_KERNEL);
 625         if (!req_buf)
 626                 return -ENOMEM;
 627
 628         if (len > op_len)
 629                 len = op_len;
 630
 631         if (map_perm & LDC_MAP_R)
 632                 memcpy(req_buf, buf, len);
 633
 634         spin_lock_irqsave(&port->vio.lock, flags);
 635
 636         dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 637
 638         /* XXX If we want to use this code generically we have to
 639          * XXX handle TX ring exhaustion etc.
 640          */
 641         desc = vio_dring_cur(dr);
 642
 643         err = ldc_map_single(port->vio.lp, req_buf, op_len,
 644                              desc->cookies, port->ring_cookies,
 645                              map_perm);
 646         if (err < 0) {
 647                 spin_unlock_irqrestore(&port->vio.lock, flags);
 648                 kfree(req_buf);
 649                 return err;
 650         }
 651
 652         init_completion(&comp.com);
 653         comp.waiting_for = WAITING_FOR_GEN_CMD;
 654         port->vio.cmp = &comp;
 655
 656         desc->hdr.ack = VIO_ACK_ENABLE;
 657         desc->req_id = port->req_id;
 658         desc->operation = op;
 659         desc->slice = 0;
 660         desc->status = ~0;
 661         desc->offset = 0;
 662         desc->size = op_len;
 663         desc->ncookies = err;
 664
 665         /* This has to be a non-SMP write barrier because we are writing
 666          * to memory which is shared with the peer LDOM.
 667          */
 668         wmb();
 669         desc->hdr.state = VIO_DESC_READY;
 670
 671         err = __vdc_tx_trigger(port);
 672         if (err >= 0) {
 673                 port->req_id++;
 674                 dr->prod = vio_dring_next(dr, dr->prod);
 675                 spin_unlock_irqrestore(&port->vio.lock, flags);
 676
 677                 wait_for_completion(&comp.com);
 678                 err = comp.err;
 679         } else {
 680                 port->vio.cmp = NULL;
 681                 spin_unlock_irqrestore(&port->vio.lock, flags);
 682         }
 683
 684         if (map_perm & LDC_MAP_W)
 685                 memcpy(buf, req_buf, len);
 686
 687         kfree(req_buf);
 688
 689         return err;
 690 }
 691
 692 static int vdc_alloc_tx_ring(struct vdc_port *port)
 693 {
 694         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 695         unsigned long len, entry_size;
 696         int ncookies;
 697         void *dring;
 698
 699         entry_size = sizeof(struct vio_disk_desc) +
 700                 (sizeof(struct ldc_trans_cookie) * port->ring_cookies);
 701         len = (VDC_TX_RING_SIZE * entry_size);
 702
 703         ncookies = VIO_MAX_RING_COOKIES;
 704         dring = ldc_alloc_exp_dring(port->vio.lp, len,
 705                                     dr->cookies, &ncookies,
 706                                     (LDC_MAP_SHADOW |
 707                                      LDC_MAP_DIRECT |
 708                                      LDC_MAP_RW));
 709         if (IS_ERR(dring))
 710                 return PTR_ERR(dring);
 711
 712         dr->base = dring;
 713         dr->entry_size = entry_size;
 714         dr->num_entries = VDC_TX_RING_SIZE;
 715         dr->prod = dr->cons = 0;
 716         dr->pending = VDC_TX_RING_SIZE;
 717         dr->ncookies = ncookies;
 718
 719         return 0;
 720 }
 721
 722 static void vdc_free_tx_ring(struct vdc_port *port)
 723 {
 724         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 725
 726         if (dr->base) {
 727                 ldc_free_exp_dring(port->vio.lp, dr->base,
 728                                    (dr->entry_size * dr->num_entries),
 729                                    dr->cookies, dr->ncookies);
 730                 dr->base = NULL;
 731                 dr->entry_size = 0;
 732                 dr->num_entries = 0;
 733                 dr->pending = 0;
 734                 dr->ncookies = 0;
 735         }
 736 }
 737
 738 static int vdc_port_up(struct vdc_port *port)
 739 {
 740         struct vio_completion comp;
 741
 742         init_completion(&comp.com);
 743         comp.err = 0;
 744         comp.waiting_for = WAITING_FOR_LINK_UP;
 745         port->vio.cmp = &comp;
 746
 747         vio_port_up(&port->vio);
 748         wait_for_completion(&comp.com);
 749         return comp.err;
 750 }
 751
 752 static void vdc_port_down(struct vdc_port *port)
 753 {
 754         ldc_disconnect(port->vio.lp);
 755         ldc_unbind(port->vio.lp);
 756         vdc_free_tx_ring(port);
 757         vio_ldc_free(&port->vio);
 758 }
 759
 760 static int probe_disk(struct vdc_port *port)
 761 {
 762         struct request_queue *q;
 763         struct gendisk *g;
 764         int err;
 765
 766         err = vdc_port_up(port);
 767         if (err)
 768                 return err;
 769
 770         /* Using version 1.2 means vdisk_phys_blksz should be set unless the
 771          * disk is reserved by another system.
 772          */
 773         if (vdc_version_supported(port, 1, 2) && !port->vdisk_phys_blksz)
 774                 return -ENODEV;
 775
 776         if (vdc_version_supported(port, 1, 1)) {
 777                 /* vdisk_size should be set during the handshake, if it wasn't
 778                  * then the underlying disk is reserved by another system
 779                  */
 780                 if (port->vdisk_size == -1)
 781                         return -ENODEV;
 782         } else {
 783                 struct vio_disk_geom geom;
 784
 785                 err = generic_request(port, VD_OP_GET_DISKGEOM,
 786                                       &geom, sizeof(geom));
 787                 if (err < 0) {
 788                         printk(KERN_ERR PFX "VD_OP_GET_DISKGEOM returns "
 789                                "error %d\n", err);
 790                         return err;
 791                 }
 792                 port->vdisk_size = ((u64)geom.num_cyl *
 793                                     (u64)geom.num_hd *
 794                                     (u64)geom.num_sec);
 795         }
 796
 797         q = blk_init_queue(do_vdc_request, &port->vio.lock);
 798         if (!q) {
 799                 printk(KERN_ERR PFX "%s: Could not allocate queue.\n",
 800                        port->vio.name);
 801                 return -ENOMEM;
 802         }
 803         g = alloc_disk(1 << PARTITION_SHIFT);
 804         if (!g) {
 805                 printk(KERN_ERR PFX "%s: Could not allocate gendisk.\n",
 806                        port->vio.name);
 807                 blk_cleanup_queue(q);
 808                 return -ENOMEM;
 809         }
 810
 811         port->disk = g;
 812
 813         /* Each segment in a request is up to an aligned page in size. */
 814         blk_queue_segment_boundary(q, PAGE_SIZE - 1);
 815         blk_queue_max_segment_size(q, PAGE_SIZE);
 816
 817         blk_queue_max_segments(q, port->ring_cookies);
 818         blk_queue_max_hw_sectors(q, port->max_xfer_size);
 819         g->major = vdc_major;
 820         g->first_minor = port->vio.vdev->dev_no << PARTITION_SHIFT;
 821         strcpy(g->disk_name, port->disk_name);
 822
 823         g->fops = &vdc_fops;
 824         g->queue = q;
 825         g->private_data = port;
 826
 827         set_capacity(g, port->vdisk_size);
 828
 829         if (vdc_version_supported(port, 1, 1)) {
 830                 switch (port->vdisk_mtype) {
 831                 case VD_MEDIA_TYPE_CD:
 832                         pr_info(PFX "Virtual CDROM %s\n", port->disk_name);
 833                         g->flags |= GENHD_FL_CD;
 834                         g->flags |= GENHD_FL_REMOVABLE;
 835                         set_disk_ro(g, 1);
 836                         break;
 837
 838                 case VD_MEDIA_TYPE_DVD:
 839                         pr_info(PFX "Virtual DVD %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_FIXED:
 846                         pr_info(PFX "Virtual Hard disk %s\n", port->disk_name);
 847                         break;
 848                 }
 849         }
 850
 851         blk_queue_physical_block_size(q, port->vdisk_phys_blksz);
 852
 853         pr_info(PFX "%s: %u sectors (%u MB) protocol %d.%d\n",
 854                g->disk_name,
 855                port->vdisk_size, (port->vdisk_size >> (20 - 9)),
 856                port->vio.ver.major, port->vio.ver.minor);
 857
 858         device_add_disk(&port->vio.vdev->dev, g);
 859
 860         return 0;
 861 }
 862
 863 static struct ldc_channel_config vdc_ldc_cfg = {
 864         .event          = vdc_event,
 865         .mtu            = 64,
 866         .mode           = LDC_MODE_UNRELIABLE,
 867 };
 868
 869 static struct vio_driver_ops vdc_vio_ops = {
 870         .send_attr              = vdc_send_attr,
 871         .handle_attr            = vdc_handle_attr,
 872         .handshake_complete     = vdc_handshake_complete,
 873 };
 874
 875 static void print_version(void)
 876 {
 877         static int version_printed;
 878
 879         if (version_printed++ == 0)
 880                 printk(KERN_INFO "%s", version);
 881 }
 882
 883 struct vdc_check_port_data {
 884         int     dev_no;
 885         char    *type;
 886 };
 887
 888 static int vdc_device_probed(struct device *dev, void *arg)
 889 {
 890         struct vio_dev *vdev = to_vio_dev(dev);
 891         struct vdc_check_port_data *port_data;
 892
 893         port_data = (struct vdc_check_port_data *)arg;
 894
 895         if ((vdev->dev_no == port_data->dev_no) &&
 896             (!(strcmp((char *)&vdev->type, port_data->type))) &&
 897                 dev_get_drvdata(dev)) {
 898                 /* This device has already been configured
 899                  * by vdc_port_probe()
 900                  */
 901                 return 1;
 902         } else {
 903                 return 0;
 904         }
 905 }
 906
 907 /* Determine whether the VIO device is part of an mpgroup
 908  * by locating all the virtual-device-port nodes associated
 909  * with the parent virtual-device node for the VIO device
 910  * and checking whether any of these nodes are vdc-ports
 911  * which have already been configured.
 912  *
 913  * Returns true if this device is part of an mpgroup and has
 914  * already been probed.
 915  */
 916 static bool vdc_port_mpgroup_check(struct vio_dev *vdev)
 917 {
 918         struct vdc_check_port_data port_data;
 919         struct device *dev;
 920
 921         port_data.dev_no = vdev->dev_no;
 922         port_data.type = (char *)&vdev->type;
 923
 924         dev = device_find_child(vdev->dev.parent, &port_data,
 925                                 vdc_device_probed);
 926
 927         if (dev)
 928                 return true;
 929
 930         return false;
 931 }
 932
 933 static int vdc_port_probe(struct vio_dev *vdev, const struct vio_device_id *id)
 934 {
 935         struct mdesc_handle *hp;
 936         struct vdc_port *port;
 937         int err;
 938         const u64 *ldc_timeout;
 939
 940         print_version();
 941
 942         hp = mdesc_grab();
 943         if (!hp)
 944                 return -ENODEV;
 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         setup_timer(&port->ldc_reset_timer, vdc_ldc_reset_timer,
 985                     (unsigned long)port);
 986         INIT_WORK(&port->ldc_reset_work, vdc_ldc_reset_work);
 987
 988         err = vio_driver_init(&port->vio, vdev, VDEV_DISK,
 989                               vdc_versions, ARRAY_SIZE(vdc_versions),
 990                               &vdc_vio_ops, port->disk_name);
 991         if (err)
 992                 goto err_out_free_port;
 993
 994         port->vdisk_block_size = VDC_DEFAULT_BLK_SIZE;
 995         port->max_xfer_size = ((128 * 1024) / port->vdisk_block_size);
 996         port->ring_cookies = ((port->max_xfer_size *
 997                                port->vdisk_block_size) / PAGE_SIZE) + 2;
 998
 999         err = vio_ldc_alloc(&port->vio, &vdc_ldc_cfg, port);
1000         if (err)
1001                 goto err_out_free_port;
1002
1003         err = vdc_alloc_tx_ring(port);
1004         if (err)
1005                 goto err_out_free_ldc;
1006
1007         err = probe_disk(port);
1008         if (err)
1009                 goto err_out_free_tx_ring;
1010
1011         /* Note that the device driver_data is used to determine
1012          * whether the port has been probed.
1013          */
1014         dev_set_drvdata(&vdev->dev, port);
1015
1016         mdesc_release(hp);
1017
1018         return 0;
1019
1020 err_out_free_tx_ring:
1021         vdc_free_tx_ring(port);
1022
1023 err_out_free_ldc:
1024         vio_ldc_free(&port->vio);
1025
1026 err_out_free_port:
1027         kfree(port);
1028
1029 err_out_release_mdesc:
1030         mdesc_release(hp);
1031         return err;
1032 }
1033
1034 static int vdc_port_remove(struct vio_dev *vdev)
1035 {
1036         struct vdc_port *port = dev_get_drvdata(&vdev->dev);
1037
1038         if (port) {
1039                 unsigned long flags;
1040
1041                 spin_lock_irqsave(&port->vio.lock, flags);
1042                 blk_stop_queue(port->disk->queue);
1043                 spin_unlock_irqrestore(&port->vio.lock, flags);
1044
1045                 flush_work(&port->ldc_reset_work);
1046                 del_timer_sync(&port->ldc_reset_timer);
1047                 del_timer_sync(&port->vio.timer);
1048
1049                 del_gendisk(port->disk);
1050                 blk_cleanup_queue(port->disk->queue);
1051                 put_disk(port->disk);
1052                 port->disk = NULL;
1053
1054                 vdc_free_tx_ring(port);
1055                 vio_ldc_free(&port->vio);
1056
1057                 dev_set_drvdata(&vdev->dev, NULL);
1058
1059                 kfree(port);
1060         }
1061         return 0;
1062 }
1063
1064 static void vdc_requeue_inflight(struct vdc_port *port)
1065 {
1066         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1067         u32 idx;
1068
1069         for (idx = dr->cons; idx != dr->prod; idx = vio_dring_next(dr, idx)) {
1070                 struct vio_disk_desc *desc = vio_dring_entry(dr, idx);
1071                 struct vdc_req_entry *rqe = &port->rq_arr[idx];
1072                 struct request *req;
1073
1074                 ldc_unmap(port->vio.lp, desc->cookies, desc->ncookies);
1075                 desc->hdr.state = VIO_DESC_FREE;
1076                 dr->cons = vio_dring_next(dr, idx);
1077
1078                 req = rqe->req;
1079                 if (req == NULL) {
1080                         vdc_end_special(port, desc);
1081                         continue;
1082                 }
1083
1084                 rqe->req = NULL;
1085                 blk_requeue_request(port->disk->queue, req);
1086         }
1087 }
1088
1089 static void vdc_queue_drain(struct vdc_port *port)
1090 {
1091         struct request *req;
1092
1093         while ((req = blk_fetch_request(port->disk->queue)) != NULL)
1094                 __blk_end_request_all(req, BLK_STS_IOERR);
1095 }
1096
1097 static void vdc_ldc_reset_timer(unsigned long _arg)
1098 {
1099         struct vdc_port *port = (struct vdc_port *) _arg;
1100         struct vio_driver_state *vio = &port->vio;
1101         unsigned long flags;
1102
1103         spin_lock_irqsave(&vio->lock, flags);
1104         if (!(port->vio.hs_state & VIO_HS_COMPLETE)) {
1105                 pr_warn(PFX "%s ldc down %llu seconds, draining queue\n",
1106                         port->disk_name, port->ldc_timeout);
1107                 vdc_queue_drain(port);
1108                 vdc_blk_queue_start(port);
1109         }
1110         spin_unlock_irqrestore(&vio->lock, flags);
1111 }
1112
1113 static void vdc_ldc_reset_work(struct work_struct *work)
1114 {
1115         struct vdc_port *port;
1116         struct vio_driver_state *vio;
1117         unsigned long flags;
1118
1119         port = container_of(work, struct vdc_port, ldc_reset_work);
1120         vio = &port->vio;
1121
1122         spin_lock_irqsave(&vio->lock, flags);
1123         vdc_ldc_reset(port);
1124         spin_unlock_irqrestore(&vio->lock, flags);
1125 }
1126
1127 static void vdc_ldc_reset(struct vdc_port *port)
1128 {
1129         int err;
1130
1131         assert_spin_locked(&port->vio.lock);
1132
1133         pr_warn(PFX "%s ldc link reset\n", port->disk_name);
1134         blk_stop_queue(port->disk->queue);
1135         vdc_requeue_inflight(port);
1136         vdc_port_down(port);
1137
1138         err = vio_ldc_alloc(&port->vio, &vdc_ldc_cfg, port);
1139         if (err) {
1140                 pr_err(PFX "%s vio_ldc_alloc:%d\n", port->disk_name, err);
1141                 return;
1142         }
1143
1144         err = vdc_alloc_tx_ring(port);
1145         if (err) {
1146                 pr_err(PFX "%s vio_alloc_tx_ring:%d\n", port->disk_name, err);
1147                 goto err_free_ldc;
1148         }
1149
1150         if (port->ldc_timeout)
1151                 mod_timer(&port->ldc_reset_timer,
1152                           round_jiffies(jiffies + HZ * port->ldc_timeout));
1153         mod_timer(&port->vio.timer, round_jiffies(jiffies + HZ));
1154         return;
1155
1156 err_free_ldc:
1157         vio_ldc_free(&port->vio);
1158 }
1159
1160 static const struct vio_device_id vdc_port_match[] = {
1161         {
1162                 .type = "vdc-port",
1163         },
1164         {},
1165 };
1166 MODULE_DEVICE_TABLE(vio, vdc_port_match);
1167
1168 static struct vio_driver vdc_port_driver = {
1169         .id_table       = vdc_port_match,
1170         .probe          = vdc_port_probe,
1171         .remove         = vdc_port_remove,
1172         .name           = "vdc_port",
1173 };
1174
1175 static int __init vdc_init(void)
1176 {
1177         int err;
1178
1179         sunvdc_wq = alloc_workqueue("sunvdc", 0, 0);
1180         if (!sunvdc_wq)
1181                 return -ENOMEM;
1182
1183         err = register_blkdev(0, VDCBLK_NAME);
1184         if (err < 0)
1185                 goto out_free_wq;
1186
1187         vdc_major = err;
1188
1189         err = vio_register_driver(&vdc_port_driver);
1190         if (err)
1191                 goto out_unregister_blkdev;
1192
1193         return 0;
1194
1195 out_unregister_blkdev:
1196         unregister_blkdev(vdc_major, VDCBLK_NAME);
1197         vdc_major = 0;
1198
1199 out_free_wq:
1200         destroy_workqueue(sunvdc_wq);
1201         return err;
1202 }
1203
1204 static void __exit vdc_exit(void)
1205 {
1206         vio_unregister_driver(&vdc_port_driver);
1207         unregister_blkdev(vdc_major, VDCBLK_NAME);
1208         destroy_workqueue(sunvdc_wq);
1209 }
1210
1211 module_init(vdc_init);
1212 module_exit(vdc_exit);