drivers/block/viodasd.c

   1 /* -*- linux-c -*-
   2  * viodasd.c
   3  *  Authors: Dave Boutcher <boutcher@us.ibm.com>
   4  *           Ryan Arnold <ryanarn@us.ibm.com>
   5  *           Colin Devilbiss <devilbis@us.ibm.com>
   6  *           Stephen Rothwell <sfr@au1.ibm.com>
   7  *
   8  * (C) Copyright 2000-2004 IBM Corporation
   9  *
  10  * This program is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU General Public License as
  12  * published by the Free Software Foundation; either version 2 of the
  13  * License, or (at your option) any later version.
  14  *
  15  * This program is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  18  * GNU General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU General Public License
  21  * along with this program; if not, write to the Free Software
  22  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  23  *
  24  * This routine provides access to disk space (termed "DASD" in historical
  25  * IBM terms) owned and managed by an OS/400 partition running on the
  26  * same box as this Linux partition.
  27  *
  28  * All disk operations are performed by sending messages back and forth to
  29  * the OS/400 partition.
  30  */
  31 #include <linux/major.h>
  32 #include <linux/fs.h>
  33 #include <linux/module.h>
  34 #include <linux/kernel.h>
  35 #include <linux/blkdev.h>
  36 #include <linux/genhd.h>
  37 #include <linux/hdreg.h>
  38 #include <linux/errno.h>
  39 #include <linux/init.h>
  40 #include <linux/string.h>
  41 #include <linux/dma-mapping.h>
  42 #include <linux/completion.h>
  43 #include <linux/device.h>
  44 #include <linux/kernel.h>
  45
  46 #include <asm/uaccess.h>
  47 #include <asm/vio.h>
  48 #include <asm/iSeries/HvTypes.h>
  49 #include <asm/iSeries/HvLpEvent.h>
  50 #include <asm/iSeries/HvLpConfig.h>
  51 #include <asm/iSeries/vio.h>
  52
  53 MODULE_DESCRIPTION("iSeries Virtual DASD");
  54 MODULE_AUTHOR("Dave Boutcher");
  55 MODULE_LICENSE("GPL");
  56
  57 /*
  58  * We only support 7 partitions per physical disk....so with minor
  59  * numbers 0-255 we get a maximum of 32 disks.
  60  */
  61 #define VIOD_GENHD_NAME         "iseries/vd"
  62 #define VIOD_GENHD_DEVFS_NAME   "iseries/disc"
  63
  64 #define VIOD_VERS               "1.64"
  65
  66 #define VIOD_KERN_WARNING       KERN_WARNING "viod: "
  67 #define VIOD_KERN_INFO          KERN_INFO "viod: "
  68
  69 enum {
  70         PARTITION_SHIFT = 3,
  71         MAX_DISKNO = HVMAXARCHITECTEDVIRTUALDISKS,
  72         MAX_DISK_NAME = sizeof(((struct gendisk *)0)->disk_name)
  73 };
  74
  75 static spinlock_t       viodasd_spinlock = SPIN_LOCK_UNLOCKED;
  76
  77 #define VIOMAXREQ               16
  78 #define VIOMAXBLOCKDMA          12
  79
  80 #define DEVICE_NO(cell) ((struct viodasd_device *)(cell) - &viodasd_devices[0])
  81
  82 struct open_data {
  83         u64     disk_size;
  84         u16     max_disk;
  85         u16     cylinders;
  86         u16     tracks;
  87         u16     sectors;
  88         u16     bytes_per_sector;
  89 };
  90
  91 struct rw_data {
  92         u64     offset;
  93         struct {
  94                 u32     token;
  95                 u32     reserved;
  96                 u64     len;
  97         } dma_info[VIOMAXBLOCKDMA];
  98 };
  99
 100 struct vioblocklpevent {
 101         struct HvLpEvent        event;
 102         u32                     reserved;
 103         u16                     version;
 104         u16                     sub_result;
 105         u16                     disk;
 106         u16                     flags;
 107         union {
 108                 struct open_data        open_data;
 109                 struct rw_data          rw_data;
 110                 u64                     changed;
 111         } u;
 112 };
 113
 114 #define vioblockflags_ro   0x0001
 115
 116 enum vioblocksubtype {
 117         vioblockopen = 0x0001,
 118         vioblockclose = 0x0002,
 119         vioblockread = 0x0003,
 120         vioblockwrite = 0x0004,
 121         vioblockflush = 0x0005,
 122         vioblockcheck = 0x0007
 123 };
 124
 125 struct viodasd_waitevent {
 126         struct completion       com;
 127         int                     rc;
 128         u16                     sub_result;
 129         int                     max_disk;       /* open */
 130 };
 131
 132 static const struct vio_error_entry viodasd_err_table[] = {
 133         { 0x0201, EINVAL, "Invalid Range" },
 134         { 0x0202, EINVAL, "Invalid Token" },
 135         { 0x0203, EIO, "DMA Error" },
 136         { 0x0204, EIO, "Use Error" },
 137         { 0x0205, EIO, "Release Error" },
 138         { 0x0206, EINVAL, "Invalid Disk" },
 139         { 0x0207, EBUSY, "Cant Lock" },
 140         { 0x0208, EIO, "Already Locked" },
 141         { 0x0209, EIO, "Already Unlocked" },
 142         { 0x020A, EIO, "Invalid Arg" },
 143         { 0x020B, EIO, "Bad IFS File" },
 144         { 0x020C, EROFS, "Read Only Device" },
 145         { 0x02FF, EIO, "Internal Error" },
 146         { 0x0000, 0, NULL },
 147 };
 148
 149 /*
 150  * Figure out the biggest I/O request (in sectors) we can accept
 151  */
 152 #define VIODASD_MAXSECTORS (4096 / 512 * VIOMAXBLOCKDMA)
 153
 154 /*
 155  * Number of disk I/O requests we've sent to OS/400
 156  */
 157 static int num_req_outstanding;
 158
 159 /*
 160  * This is our internal structure for keeping track of disk devices
 161  */
 162 struct viodasd_device {
 163         u16             cylinders;
 164         u16             tracks;
 165         u16             sectors;
 166         u16             bytes_per_sector;
 167         u64             size;
 168         int             read_only;
 169         spinlock_t      q_lock;
 170         struct gendisk  *disk;
 171         struct device   *dev;
 172 } viodasd_devices[MAX_DISKNO];
 173
 174 /*
 175  * External open entry point.
 176  */
 177 static int viodasd_open(struct inode *ino, struct file *fil)
 178 {
 179         struct viodasd_device *d = ino->i_bdev->bd_disk->private_data;
 180         HvLpEvent_Rc hvrc;
 181         struct viodasd_waitevent we;
 182         u16 flags = 0;
 183
 184         if (d->read_only) {
 185                 if ((fil != NULL) && (fil->f_mode & FMODE_WRITE))
 186                         return -EROFS;
 187                 flags = vioblockflags_ro;
 188         }
 189
 190         init_completion(&we.com);
 191
 192         /* Send the open event to OS/400 */
 193         hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp,
 194                         HvLpEvent_Type_VirtualIo,
 195                         viomajorsubtype_blockio | vioblockopen,
 196                         HvLpEvent_AckInd_DoAck, HvLpEvent_AckType_ImmediateAck,
 197                         viopath_sourceinst(viopath_hostLp),
 198                         viopath_targetinst(viopath_hostLp),
 199                         (u64)(unsigned long)&we, VIOVERSION << 16,
 200                         ((u64)DEVICE_NO(d) << 48) | ((u64)flags << 32),
 201                         0, 0, 0);
 202         if (hvrc != 0) {
 203                 printk(VIOD_KERN_WARNING "HV open failed %d\n", (int)hvrc);
 204                 return -EIO;
 205         }
 206
 207         wait_for_completion(&we.com);
 208
 209         /* Check the return code */
 210         if (we.rc != 0) {
 211                 const struct vio_error_entry *err =
 212                         vio_lookup_rc(viodasd_err_table, we.sub_result);
 213
 214                 printk(VIOD_KERN_WARNING
 215                                 "bad rc opening disk: %d:0x%04x (%s)\n",
 216                                 (int)we.rc, we.sub_result, err->msg);
 217                 return -EIO;
 218         }
 219
 220         return 0;
 221 }
 222
 223 /*
 224  * External release entry point.
 225  */
 226 static int viodasd_release(struct inode *ino, struct file *fil)
 227 {
 228         struct viodasd_device *d = ino->i_bdev->bd_disk->private_data;
 229         HvLpEvent_Rc hvrc;
 230
 231         /* Send the event to OS/400.  We DON'T expect a response */
 232         hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp,
 233                         HvLpEvent_Type_VirtualIo,
 234                         viomajorsubtype_blockio | vioblockclose,
 235                         HvLpEvent_AckInd_NoAck, HvLpEvent_AckType_ImmediateAck,
 236                         viopath_sourceinst(viopath_hostLp),
 237                         viopath_targetinst(viopath_hostLp),
 238                         0, VIOVERSION << 16,
 239                         ((u64)DEVICE_NO(d) << 48) /* | ((u64)flags << 32) */,
 240                         0, 0, 0);
 241         if (hvrc != 0)
 242                 printk(VIOD_KERN_WARNING "HV close call failed %d\n",
 243                                 (int)hvrc);
 244         return 0;
 245 }
 246
 247
 248 /* External ioctl entry point.
 249  */
 250 static int viodasd_ioctl(struct inode *ino, struct file *fil,
 251                          unsigned int cmd, unsigned long arg)
 252 {
 253         int err;
 254         unsigned char sectors;
 255         unsigned char heads;
 256         unsigned short cylinders;
 257         struct hd_geometry *geo;
 258         struct gendisk *gendisk;
 259         struct viodasd_device *d;
 260
 261         switch (cmd) {
 262         case HDIO_GETGEO:
 263                 geo = (struct hd_geometry *)arg;
 264                 if (geo == NULL)
 265                         return -EINVAL;
 266                 err = verify_area(VERIFY_WRITE, geo, sizeof(*geo));
 267                 if (err)
 268                         return err;
 269                 gendisk = ino->i_bdev->bd_disk;
 270                 d = gendisk->private_data;
 271                 sectors = d->sectors;
 272                 if (sectors == 0)
 273                         sectors = 32;
 274                 heads = d->tracks;
 275                 if (heads == 0)
 276                         heads = 64;
 277                 cylinders = d->cylinders;
 278                 if (cylinders == 0)
 279                         cylinders = get_capacity(gendisk) / (sectors * heads);
 280                 if (__put_user(sectors, &geo->sectors) ||
 281                     __put_user(heads, &geo->heads) ||
 282                     __put_user(cylinders, &geo->cylinders) ||
 283                     __put_user(get_start_sect(ino->i_bdev), &geo->start))
 284                         return -EFAULT;
 285                 return 0;
 286         }
 287
 288         return -EINVAL;
 289 }
 290
 291 /*
 292  * Our file operations table
 293  */
 294 static struct block_device_operations viodasd_fops = {
 295         .owner = THIS_MODULE,
 296         .open = viodasd_open,
 297         .release = viodasd_release,
 298         .ioctl = viodasd_ioctl,
 299 };
 300
 301 /*
 302  * End a request
 303  */
 304 static void viodasd_end_request(struct request *req, int uptodate,
 305                 int num_sectors)
 306 {
 307         if (end_that_request_first(req, uptodate, num_sectors))
 308                 return;
 309         add_disk_randomness(req->rq_disk);
 310         end_that_request_last(req);
 311 }
 312
 313 /*
 314  * Send an actual I/O request to OS/400
 315  */
 316 static int send_request(struct request *req)
 317 {
 318         u64 start;
 319         int direction;
 320         int nsg;
 321         u16 viocmd;
 322         HvLpEvent_Rc hvrc;
 323         struct vioblocklpevent *bevent;
 324         struct scatterlist sg[VIOMAXBLOCKDMA];
 325         int sgindex;
 326         int statindex;
 327         struct viodasd_device *d;
 328         unsigned long flags;
 329
 330         start = (u64)req->sector << 9;
 331
 332         if (rq_data_dir(req) == READ) {
 333                 direction = DMA_FROM_DEVICE;
 334                 viocmd = viomajorsubtype_blockio | vioblockread;
 335                 statindex = 0;
 336         } else {
 337                 direction = DMA_TO_DEVICE;
 338                 viocmd = viomajorsubtype_blockio | vioblockwrite;
 339                 statindex = 1;
 340         }
 341
 342         d = req->rq_disk->private_data;
 343
 344         /* Now build the scatter-gather list */
 345         nsg = blk_rq_map_sg(req->q, req, sg);
 346         nsg = dma_map_sg(d->dev, sg, nsg, direction);
 347
 348         spin_lock_irqsave(&viodasd_spinlock, flags);
 349         num_req_outstanding++;
 350
 351         /* This optimization handles a single DMA block */
 352         if (nsg == 1)
 353                 hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp,
 354                                 HvLpEvent_Type_VirtualIo, viocmd,
 355                                 HvLpEvent_AckInd_DoAck,
 356                                 HvLpEvent_AckType_ImmediateAck,
 357                                 viopath_sourceinst(viopath_hostLp),
 358                                 viopath_targetinst(viopath_hostLp),
 359                                 (u64)(unsigned long)req, VIOVERSION << 16,
 360                                 ((u64)DEVICE_NO(d) << 48), start,
 361                                 ((u64)sg_dma_address(&sg[0])) << 32,
 362                                 sg_dma_len(&sg[0]));
 363         else {
 364                 bevent = (struct vioblocklpevent *)
 365                         vio_get_event_buffer(viomajorsubtype_blockio);
 366                 if (bevent == NULL) {
 367                         printk(VIOD_KERN_WARNING
 368                                "error allocating disk event buffer\n");
 369                         goto error_ret;
 370                 }
 371
 372                 /*
 373                  * Now build up the actual request.  Note that we store
 374                  * the pointer to the request in the correlation
 375                  * token so we can match the response up later
 376                  */
 377                 memset(bevent, 0, sizeof(struct vioblocklpevent));
 378                 bevent->event.xFlags.xValid = 1;
 379                 bevent->event.xFlags.xFunction = HvLpEvent_Function_Int;
 380                 bevent->event.xFlags.xAckInd = HvLpEvent_AckInd_DoAck;
 381                 bevent->event.xFlags.xAckType = HvLpEvent_AckType_ImmediateAck;
 382                 bevent->event.xType = HvLpEvent_Type_VirtualIo;
 383                 bevent->event.xSubtype = viocmd;
 384                 bevent->event.xSourceLp = HvLpConfig_getLpIndex();
 385                 bevent->event.xTargetLp = viopath_hostLp;
 386                 bevent->event.xSizeMinus1 =
 387                         offsetof(struct vioblocklpevent, u.rw_data.dma_info) +
 388                         (sizeof(bevent->u.rw_data.dma_info[0]) * nsg) - 1;
 389                 bevent->event.xSourceInstanceId =
 390                         viopath_sourceinst(viopath_hostLp);
 391                 bevent->event.xTargetInstanceId =
 392                         viopath_targetinst(viopath_hostLp);
 393                 bevent->event.xCorrelationToken = (u64)req;
 394                 bevent->version = VIOVERSION;
 395                 bevent->disk = DEVICE_NO(d);
 396                 bevent->u.rw_data.offset = start;
 397
 398                 /*
 399                  * Copy just the dma information from the sg list
 400                  * into the request
 401                  */
 402                 for (sgindex = 0; sgindex < nsg; sgindex++) {
 403                         bevent->u.rw_data.dma_info[sgindex].token =
 404                                 sg_dma_address(&sg[sgindex]);
 405                         bevent->u.rw_data.dma_info[sgindex].len =
 406                                 sg_dma_len(&sg[sgindex]);
 407                 }
 408
 409                 /* Send the request */
 410                 hvrc = HvCallEvent_signalLpEvent(&bevent->event);
 411                 vio_free_event_buffer(viomajorsubtype_blockio, bevent);
 412         }
 413
 414         if (hvrc != HvLpEvent_Rc_Good) {
 415                 printk(VIOD_KERN_WARNING
 416                        "error sending disk event to OS/400 (rc %d)\n",
 417                        (int)hvrc);
 418                 goto error_ret;
 419         }
 420         spin_unlock_irqrestore(&viodasd_spinlock, flags);
 421         return 0;
 422
 423 error_ret:
 424         num_req_outstanding--;
 425         spin_unlock_irqrestore(&viodasd_spinlock, flags);
 426         dma_unmap_sg(d->dev, sg, nsg, direction);
 427         return -1;
 428 }
 429
 430 /*
 431  * This is the external request processing routine
 432  */
 433 static void do_viodasd_request(request_queue_t *q)
 434 {
 435         struct request *req;
 436
 437         /*
 438          * If we already have the maximum number of requests
 439          * outstanding to OS/400 just bail out. We'll come
 440          * back later.
 441          */
 442         while (num_req_outstanding < VIOMAXREQ) {
 443                 req = elv_next_request(q);
 444                 if (req == NULL)
 445                         return;
 446                 /* dequeue the current request from the queue */
 447                 blkdev_dequeue_request(req);
 448                 /* check that request contains a valid command */
 449                 if (!blk_fs_request(req)) {
 450                         viodasd_end_request(req, 0, req->hard_nr_sectors);
 451                         continue;
 452                 }
 453                 /* Try sending the request */
 454                 if (send_request(req) != 0)
 455                         viodasd_end_request(req, 0, req->hard_nr_sectors);
 456         }
 457 }
 458
 459 /*
 460  * Probe a single disk and fill in the viodasd_device structure
 461  * for it.
 462  */
 463 static void probe_disk(struct viodasd_device *d)
 464 {
 465         HvLpEvent_Rc hvrc;
 466         struct viodasd_waitevent we;
 467         int dev_no = DEVICE_NO(d);
 468         struct gendisk *g;
 469         struct request_queue *q;
 470         u16 flags = 0;
 471
 472 retry:
 473         init_completion(&we.com);
 474
 475         /* Send the open event to OS/400 */
 476         hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp,
 477                         HvLpEvent_Type_VirtualIo,
 478                         viomajorsubtype_blockio | vioblockopen,
 479                         HvLpEvent_AckInd_DoAck, HvLpEvent_AckType_ImmediateAck,
 480                         viopath_sourceinst(viopath_hostLp),
 481                         viopath_targetinst(viopath_hostLp),
 482                         (u64)(unsigned long)&we, VIOVERSION << 16,
 483                         ((u64)dev_no << 48) | ((u64)flags<< 32),
 484                         0, 0, 0);
 485         if (hvrc != 0) {
 486                 printk(VIOD_KERN_WARNING "bad rc on HV open %d\n", (int)hvrc);
 487                 return;
 488         }
 489
 490         wait_for_completion(&we.com);
 491
 492         if (we.rc != 0) {
 493                 if (flags != 0)
 494                         return;
 495                 /* try again with read only flag set */
 496                 flags = vioblockflags_ro;
 497                 goto retry;
 498         }
 499         if (we.max_disk > (MAX_DISKNO - 1)) {
 500                 static int warned;
 501
 502                 if (warned == 0) {
 503                         warned++;
 504                         printk(VIOD_KERN_INFO
 505                                 "Only examining the first %d "
 506                                 "of %d disks connected\n",
 507                                 MAX_DISKNO, we.max_disk + 1);
 508                 }
 509         }
 510
 511         /* Send the close event to OS/400.  We DON'T expect a response */
 512         hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp,
 513                         HvLpEvent_Type_VirtualIo,
 514                         viomajorsubtype_blockio | vioblockclose,
 515                         HvLpEvent_AckInd_NoAck, HvLpEvent_AckType_ImmediateAck,
 516                         viopath_sourceinst(viopath_hostLp),
 517                         viopath_targetinst(viopath_hostLp),
 518                         0, VIOVERSION << 16,
 519                         ((u64)dev_no << 48) | ((u64)flags << 32),
 520                         0, 0, 0);
 521         if (hvrc != 0) {
 522                 printk(VIOD_KERN_WARNING
 523                        "bad rc sending event to OS/400 %d\n", (int)hvrc);
 524                 return;
 525         }
 526         /* create the request queue for the disk */
 527         spin_lock_init(&d->q_lock);
 528         q = blk_init_queue(do_viodasd_request, &d->q_lock);
 529         if (q == NULL) {
 530                 printk(VIOD_KERN_WARNING "cannot allocate queue for disk %d\n",
 531                                 dev_no);
 532                 return;
 533         }
 534         g = alloc_disk(1 << PARTITION_SHIFT);
 535         if (g == NULL) {
 536                 printk(VIOD_KERN_WARNING
 537                                 "cannot allocate disk structure for disk %d\n",
 538                                 dev_no);
 539                 blk_cleanup_queue(q);
 540                 return;
 541         }
 542
 543         d->disk = g;
 544         blk_queue_max_hw_segments(q, VIOMAXBLOCKDMA);
 545         blk_queue_max_phys_segments(q, VIOMAXBLOCKDMA);
 546         blk_queue_max_sectors(q, VIODASD_MAXSECTORS);
 547         g->major = VIODASD_MAJOR;
 548         g->first_minor = dev_no << PARTITION_SHIFT;
 549         if (dev_no >= 26)
 550                 snprintf(g->disk_name, sizeof(g->disk_name),
 551                                 VIOD_GENHD_NAME "%c%c",
 552                                 'a' + (dev_no / 26) - 1, 'a' + (dev_no % 26));
 553         else
 554                 snprintf(g->disk_name, sizeof(g->disk_name),
 555                                 VIOD_GENHD_NAME "%c", 'a' + (dev_no % 26));
 556         snprintf(g->devfs_name, sizeof(g->devfs_name),
 557                         "%s%d", VIOD_GENHD_DEVFS_NAME, dev_no);
 558         g->fops = &viodasd_fops;
 559         g->queue = q;
 560         g->private_data = d;
 561         g->driverfs_dev = d->dev;
 562         set_capacity(g, d->size >> 9);
 563
 564         printk(VIOD_KERN_INFO "disk %d: %lu sectors (%lu MB) "
 565                         "CHS=%d/%d/%d sector size %d%s\n",
 566                         dev_no, (unsigned long)(d->size >> 9),
 567                         (unsigned long)(d->size >> 20),
 568                         (int)d->cylinders, (int)d->tracks,
 569                         (int)d->sectors, (int)d->bytes_per_sector,
 570                         d->read_only ? " (RO)" : "");
 571
 572         /* register us in the global list */
 573         add_disk(g);
 574 }
 575
 576 /* returns the total number of scatterlist elements converted */
 577 static int block_event_to_scatterlist(const struct vioblocklpevent *bevent,
 578                 struct scatterlist *sg, int *total_len)
 579 {
 580         int i, numsg;
 581         const struct rw_data *rw_data = &bevent->u.rw_data;
 582         static const int offset =
 583                 offsetof(struct vioblocklpevent, u.rw_data.dma_info);
 584         static const int element_size = sizeof(rw_data->dma_info[0]);
 585
 586         numsg = ((bevent->event.xSizeMinus1 + 1) - offset) / element_size;
 587         if (numsg > VIOMAXBLOCKDMA)
 588                 numsg = VIOMAXBLOCKDMA;
 589
 590         *total_len = 0;
 591         memset(sg, 0, sizeof(sg[0]) * VIOMAXBLOCKDMA);
 592
 593         for (i = 0; (i < numsg) && (rw_data->dma_info[i].len > 0); ++i) {
 594                 sg_dma_address(&sg[i]) = rw_data->dma_info[i].token;
 595                 sg_dma_len(&sg[i]) = rw_data->dma_info[i].len;
 596                 *total_len += rw_data->dma_info[i].len;
 597         }
 598         return i;
 599 }
 600
 601 /*
 602  * Restart all queues, starting with the one _after_ the disk given,
 603  * thus reducing the chance of starvation of higher numbered disks.
 604  */
 605 static void viodasd_restart_all_queues_starting_from(int first_index)
 606 {
 607         int i;
 608
 609         for (i = first_index + 1; i < MAX_DISKNO; ++i)
 610                 if (viodasd_devices[i].disk)
 611                         blk_run_queue(viodasd_devices[i].disk->queue);
 612         for (i = 0; i <= first_index; ++i)
 613                 if (viodasd_devices[i].disk)
 614                         blk_run_queue(viodasd_devices[i].disk->queue);
 615 }
 616
 617 /*
 618  * For read and write requests, decrement the number of outstanding requests,
 619  * Free the DMA buffers we allocated.
 620  */
 621 static int viodasd_handle_read_write(struct vioblocklpevent *bevent)
 622 {
 623         int num_sg, num_sect, pci_direction, total_len;
 624         struct request *req;
 625         struct scatterlist sg[VIOMAXBLOCKDMA];
 626         struct HvLpEvent *event = &bevent->event;
 627         unsigned long irq_flags;
 628         struct viodasd_device *d;
 629         int error;
 630         spinlock_t *qlock;
 631
 632         num_sg = block_event_to_scatterlist(bevent, sg, &total_len);
 633         num_sect = total_len >> 9;
 634         if (event->xSubtype == (viomajorsubtype_blockio | vioblockread))
 635                 pci_direction = DMA_FROM_DEVICE;
 636         else
 637                 pci_direction = DMA_TO_DEVICE;
 638         req = (struct request *)bevent->event.xCorrelationToken;
 639         d = req->rq_disk->private_data;
 640
 641         dma_unmap_sg(d->dev, sg, num_sg, pci_direction);
 642
 643         /*
 644          * Since this is running in interrupt mode, we need to make sure
 645          * we're not stepping on any global I/O operations
 646          */
 647         spin_lock_irqsave(&viodasd_spinlock, irq_flags);
 648         num_req_outstanding--;
 649         spin_unlock_irqrestore(&viodasd_spinlock, irq_flags);
 650
 651         error = event->xRc != HvLpEvent_Rc_Good;
 652         if (error) {
 653                 const struct vio_error_entry *err;
 654                 err = vio_lookup_rc(viodasd_err_table, bevent->sub_result);
 655                 printk(VIOD_KERN_WARNING "read/write error %d:0x%04x (%s)\n",
 656                                 event->xRc, bevent->sub_result, err->msg);
 657                 num_sect = req->hard_nr_sectors;
 658         }
 659         qlock = req->q->queue_lock;
 660         spin_lock_irqsave(qlock, irq_flags);
 661         viodasd_end_request(req, !error, num_sect);
 662         spin_unlock_irqrestore(qlock, irq_flags);
 663
 664         /* Finally, try to get more requests off of this device's queue */
 665         viodasd_restart_all_queues_starting_from(DEVICE_NO(d));
 666
 667         return 0;
 668 }
 669
 670 /* This routine handles incoming block LP events */
 671 static void handle_block_event(struct HvLpEvent *event)
 672 {
 673         struct vioblocklpevent *bevent = (struct vioblocklpevent *)event;
 674         struct viodasd_waitevent *pwe;
 675
 676         if (event == NULL)
 677                 /* Notification that a partition went away! */
 678                 return;
 679         /* First, we should NEVER get an int here...only acks */
 680         if (event->xFlags.xFunction == HvLpEvent_Function_Int) {
 681                 printk(VIOD_KERN_WARNING
 682                        "Yikes! got an int in viodasd event handler!\n");
 683                 if (event->xFlags.xAckInd == HvLpEvent_AckInd_DoAck) {
 684                         event->xRc = HvLpEvent_Rc_InvalidSubtype;
 685                         HvCallEvent_ackLpEvent(event);
 686                 }
 687         }
 688
 689         switch (event->xSubtype & VIOMINOR_SUBTYPE_MASK) {
 690         case vioblockopen:
 691                 /*
 692                  * Handle a response to an open request.  We get all the
 693                  * disk information in the response, so update it.  The
 694                  * correlation token contains a pointer to a waitevent
 695                  * structure that has a completion in it.  update the
 696                  * return code in the waitevent structure and post the
 697                  * completion to wake up the guy who sent the request
 698                  */
 699                 pwe = (struct viodasd_waitevent *)event->xCorrelationToken;
 700                 pwe->rc = event->xRc;
 701                 pwe->sub_result = bevent->sub_result;
 702                 if (event->xRc == HvLpEvent_Rc_Good) {
 703                         const struct open_data *data = &bevent->u.open_data;
 704                         struct viodasd_device *device =
 705                                 &viodasd_devices[bevent->disk];
 706                         device->read_only =
 707                                 bevent->flags & vioblockflags_ro;
 708                         device->size = data->disk_size;
 709                         device->cylinders = data->cylinders;
 710                         device->tracks = data->tracks;
 711                         device->sectors = data->sectors;
 712                         device->bytes_per_sector = data->bytes_per_sector;
 713                         pwe->max_disk = data->max_disk;
 714                 }
 715                 complete(&pwe->com);
 716                 break;
 717         case vioblockclose:
 718                 break;
 719         case vioblockread:
 720         case vioblockwrite:
 721                 viodasd_handle_read_write(bevent);
 722                 break;
 723
 724         default:
 725                 printk(VIOD_KERN_WARNING "invalid subtype!");
 726                 if (event->xFlags.xAckInd == HvLpEvent_AckInd_DoAck) {
 727                         event->xRc = HvLpEvent_Rc_InvalidSubtype;
 728                         HvCallEvent_ackLpEvent(event);
 729                 }
 730         }
 731 }
 732
 733 /*
 734  * Get the driver to reprobe for more disks.
 735  */
 736 static ssize_t probe_disks(struct device_driver *drv, const char *buf,
 737                 size_t count)
 738 {
 739         struct viodasd_device *d;
 740
 741         for (d = viodasd_devices; d < &viodasd_devices[MAX_DISKNO]; d++) {
 742                 if (d->disk == NULL)
 743                         probe_disk(d);
 744         }
 745         return count;
 746 }
 747 static DRIVER_ATTR(probe, S_IWUSR, NULL, probe_disks);
 748
 749 static int viodasd_probe(struct vio_dev *vdev, const struct vio_device_id *id)
 750 {
 751         struct viodasd_device *d = &viodasd_devices[vdev->unit_address];
 752
 753         d->dev = &vdev->dev;
 754         probe_disk(d);
 755         if (d->disk == NULL)
 756                 return -ENODEV;
 757         return 0;
 758 }
 759
 760 static int viodasd_remove(struct vio_dev *vdev)
 761 {
 762         struct viodasd_device *d;
 763
 764         d = &viodasd_devices[vdev->unit_address];
 765         if (d->disk) {
 766                 del_gendisk(d->disk);
 767                 put_disk(d->disk);
 768                 blk_cleanup_queue(d->disk->queue);
 769                 d->disk = NULL;
 770         }
 771         d->dev = NULL;
 772         return 0;
 773 }
 774
 775 /**
 776  * viodasd_device_table: Used by vio.c to match devices that we
 777  * support.
 778  */
 779 static struct vio_device_id viodasd_device_table[] __devinitdata = {
 780         { "viodasd", "" },
 781         { 0, }
 782 };
 783
 784 MODULE_DEVICE_TABLE(vio, viodasd_device_table);
 785 static struct vio_driver viodasd_driver = {
 786         .name = "viodasd",
 787         .id_table = viodasd_device_table,
 788         .probe = viodasd_probe,
 789         .remove = viodasd_remove
 790 };
 791
 792 /*
 793  * Initialize the whole device driver.  Handle module and non-module
 794  * versions
 795  */
 796 static int __init viodasd_init(void)
 797 {
 798         int rc;
 799
 800         /* Try to open to our host lp */
 801         if (viopath_hostLp == HvLpIndexInvalid)
 802                 vio_set_hostlp();
 803
 804         if (viopath_hostLp == HvLpIndexInvalid) {
 805                 printk(VIOD_KERN_WARNING "invalid hosting partition\n");
 806                 return -EIO;
 807         }
 808
 809         printk(VIOD_KERN_INFO "vers " VIOD_VERS ", hosting partition %d\n",
 810                         viopath_hostLp);
 811
 812         /* register the block device */
 813         if (register_blkdev(VIODASD_MAJOR, VIOD_GENHD_NAME)) {
 814                 printk(VIOD_KERN_WARNING
 815                                 "Unable to get major number %d for %s\n",
 816                                 VIODASD_MAJOR, VIOD_GENHD_NAME);
 817                 return -EIO;
 818         }
 819         /* Actually open the path to the hosting partition */
 820         if (viopath_open(viopath_hostLp, viomajorsubtype_blockio,
 821                                 VIOMAXREQ + 2)) {
 822                 printk(VIOD_KERN_WARNING
 823                        "error opening path to host partition %d\n",
 824                        viopath_hostLp);
 825                 unregister_blkdev(VIODASD_MAJOR, VIOD_GENHD_NAME);
 826                 return -EIO;
 827         }
 828
 829         /* Initialize our request handler */
 830         vio_setHandler(viomajorsubtype_blockio, handle_block_event);
 831
 832         rc = vio_register_driver(&viodasd_driver);
 833         if (rc == 0)
 834                 driver_create_file(&viodasd_driver.driver, &driver_attr_probe);
 835         return rc;
 836 }
 837 module_init(viodasd_init);
 838
 839 void viodasd_exit(void)
 840 {
 841         driver_remove_file(&viodasd_driver.driver, &driver_attr_probe);
 842         vio_unregister_driver(&viodasd_driver);
 843         vio_clearHandler(viomajorsubtype_blockio);
 844         unregister_blkdev(VIODASD_MAJOR, VIOD_GENHD_NAME);
 845         viopath_close(viopath_hostLp, viomajorsubtype_blockio, VIOMAXREQ + 2);
 846 }
 847
 848 module_exit(viodasd_exit);