drivers/scsi/ibmvscsi/ibmvscsi.c

   1 /* ------------------------------------------------------------
   2  * ibmvscsi.c
   3  * (C) Copyright IBM Corporation 1994, 2004
   4  * Authors: Colin DeVilbiss (devilbis@us.ibm.com)
   5  *          Santiago Leon (santil@us.ibm.com)
   6  *          Dave Boutcher (sleddog@us.ibm.com)
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License as published by
  10  * the Free Software Foundation; either version 2 of the License, or
  11  * (at your option) any later version.
  12  *
  13  * This program is distributed in the hope that it will be useful,
  14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  16  * GNU General Public License for more details.
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * along with this program; if not, write to the Free Software
  20  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307
  21  * USA
  22  *
  23  * ------------------------------------------------------------
  24  * Emulation of a SCSI host adapter for Virtual I/O devices
  25  *
  26  * This driver supports the SCSI adapter implemented by the IBM
  27  * Power5 firmware.  That SCSI adapter is not a physical adapter,
  28  * but allows Linux SCSI peripheral drivers to directly
  29  * access devices in another logical partition on the physical system.
  30  *
  31  * The virtual adapter(s) are present in the open firmware device
  32  * tree just like real adapters.
  33  *
  34  * One of the capabilities provided on these systems is the ability
  35  * to DMA between partitions.  The architecture states that for VSCSI,
  36  * the server side is allowed to DMA to and from the client.  The client
  37  * is never trusted to DMA to or from the server directly.
  38  *
  39  * Messages are sent between partitions on a "Command/Response Queue"
  40  * (CRQ), which is just a buffer of 16 byte entries in the receiver's
  41  * Senders cannot access the buffer directly, but send messages by
  42  * making a hypervisor call and passing in the 16 bytes.  The hypervisor
  43  * puts the message in the next 16 byte space in round-robbin fashion,
  44  * turns on the high order bit of the message (the valid bit), and
  45  * generates an interrupt to the receiver (if interrupts are turned on.)
  46  * The receiver just turns off the valid bit when they have copied out
  47  * the message.
  48  *
  49  * The VSCSI client builds a SCSI Remote Protocol (SRP) Information Unit
  50  * (IU) (as defined in the T10 standard available at www.t10.org), gets
  51  * a DMA address for the message, and sends it to the server as the
  52  * payload of a CRQ message.  The server DMAs the SRP IU and processes it,
  53  * including doing any additional data transfers.  When it is done, it
  54  * DMAs the SRP response back to the same address as the request came from,
  55  * and sends a CRQ message back to inform the client that the request has
  56  * completed.
  57  *
  58  * Note that some of the underlying infrastructure is different between
  59  * machines conforming to the "RS/6000 Platform Architecture" (RPA) and
  60  * the older iSeries hypervisor models.  To support both, some low level
  61  * routines have been broken out into rpa_vscsi.c and iseries_vscsi.c.
  62  * The Makefile should pick one, not two, not zero, of these.
  63  *
  64  * TODO: This is currently pretty tied to the IBM i/pSeries hypervisor
  65  * interfaces.  It would be really nice to abstract this above an RDMA
  66  * layer.
  67  */
  68
  69 #include <linux/module.h>
  70 #include <linux/moduleparam.h>
  71 #include <linux/dma-mapping.h>
  72 #include <linux/delay.h>
  73 #include <asm/vio.h>
  74 #include <scsi/scsi.h>
  75 #include <scsi/scsi_cmnd.h>
  76 #include <scsi/scsi_host.h>
  77 #include <scsi/scsi_device.h>
  78 #include "ibmvscsi.h"
  79
  80 /* The values below are somewhat arbitrary default values, but
  81  * OS/400 will use 3 busses (disks, CDs, tapes, I think.)
  82  * Note that there are 3 bits of channel value, 6 bits of id, and
  83  * 5 bits of LUN.
  84  */
  85 static int max_id = 64;
  86 static int max_channel = 3;
  87 static int init_timeout = 5;
  88 static int max_requests = IBMVSCSI_MAX_REQUESTS_DEFAULT;
  89
  90 #define IBMVSCSI_VERSION "1.5.8"
  91
  92 MODULE_DESCRIPTION("IBM Virtual SCSI");
  93 MODULE_AUTHOR("Dave Boutcher");
  94 MODULE_LICENSE("GPL");
  95 MODULE_VERSION(IBMVSCSI_VERSION);
  96
  97 module_param_named(max_id, max_id, int, S_IRUGO | S_IWUSR);
  98 MODULE_PARM_DESC(max_id, "Largest ID value for each channel");
  99 module_param_named(max_channel, max_channel, int, S_IRUGO | S_IWUSR);
 100 MODULE_PARM_DESC(max_channel, "Largest channel value");
 101 module_param_named(init_timeout, init_timeout, int, S_IRUGO | S_IWUSR);
 102 MODULE_PARM_DESC(init_timeout, "Initialization timeout in seconds");
 103 module_param_named(max_requests, max_requests, int, S_IRUGO | S_IWUSR);
 104 MODULE_PARM_DESC(max_requests, "Maximum requests for this adapter");
 105
 106 /* ------------------------------------------------------------
 107  * Routines for the event pool and event structs
 108  */
 109 /**
 110  * initialize_event_pool: - Allocates and initializes the event pool for a host
 111  * @pool:       event_pool to be initialized
 112  * @size:       Number of events in pool
 113  * @hostdata:   ibmvscsi_host_data who owns the event pool
 114  *
 115  * Returns zero on success.
 116 */
 117 static int initialize_event_pool(struct event_pool *pool,
 118                                  int size, struct ibmvscsi_host_data *hostdata)
 119 {
 120         int i;
 121
 122         pool->size = size;
 123         pool->next = 0;
 124         pool->events = kcalloc(pool->size, sizeof(*pool->events), GFP_KERNEL);
 125         if (!pool->events)
 126                 return -ENOMEM;
 127
 128         pool->iu_storage =
 129             dma_alloc_coherent(hostdata->dev,
 130                                pool->size * sizeof(*pool->iu_storage),
 131                                &pool->iu_token, 0);
 132         if (!pool->iu_storage) {
 133                 kfree(pool->events);
 134                 return -ENOMEM;
 135         }
 136
 137         for (i = 0; i < pool->size; ++i) {
 138                 struct srp_event_struct *evt = &pool->events[i];
 139                 memset(&evt->crq, 0x00, sizeof(evt->crq));
 140                 atomic_set(&evt->free, 1);
 141                 evt->crq.valid = 0x80;
 142                 evt->crq.IU_length = sizeof(*evt->xfer_iu);
 143                 evt->crq.IU_data_ptr = pool->iu_token +
 144                         sizeof(*evt->xfer_iu) * i;
 145                 evt->xfer_iu = pool->iu_storage + i;
 146                 evt->hostdata = hostdata;
 147                 evt->ext_list = NULL;
 148                 evt->ext_list_token = 0;
 149         }
 150
 151         return 0;
 152 }
 153
 154 /**
 155  * release_event_pool: - Frees memory of an event pool of a host
 156  * @pool:       event_pool to be released
 157  * @hostdata:   ibmvscsi_host_data who owns the even pool
 158  *
 159  * Returns zero on success.
 160 */
 161 static void release_event_pool(struct event_pool *pool,
 162                                struct ibmvscsi_host_data *hostdata)
 163 {
 164         int i, in_use = 0;
 165         for (i = 0; i < pool->size; ++i) {
 166                 if (atomic_read(&pool->events[i].free) != 1)
 167                         ++in_use;
 168                 if (pool->events[i].ext_list) {
 169                         dma_free_coherent(hostdata->dev,
 170                                   SG_ALL * sizeof(struct srp_direct_buf),
 171                                   pool->events[i].ext_list,
 172                                   pool->events[i].ext_list_token);
 173                 }
 174         }
 175         if (in_use)
 176                 dev_warn(hostdata->dev, "releasing event pool with %d "
 177                          "events still in use?\n", in_use);
 178         kfree(pool->events);
 179         dma_free_coherent(hostdata->dev,
 180                           pool->size * sizeof(*pool->iu_storage),
 181                           pool->iu_storage, pool->iu_token);
 182 }
 183
 184 /**
 185  * valid_event_struct: - Determines if event is valid.
 186  * @pool:       event_pool that contains the event
 187  * @evt:        srp_event_struct to be checked for validity
 188  *
 189  * Returns zero if event is invalid, one otherwise.
 190 */
 191 static int valid_event_struct(struct event_pool *pool,
 192                                 struct srp_event_struct *evt)
 193 {
 194         int index = evt - pool->events;
 195         if (index < 0 || index >= pool->size)   /* outside of bounds */
 196                 return 0;
 197         if (evt != pool->events + index)        /* unaligned */
 198                 return 0;
 199         return 1;
 200 }
 201
 202 /**
 203  * ibmvscsi_free-event_struct: - Changes status of event to "free"
 204  * @pool:       event_pool that contains the event
 205  * @evt:        srp_event_struct to be modified
 206  *
 207 */
 208 static void free_event_struct(struct event_pool *pool,
 209                                        struct srp_event_struct *evt)
 210 {
 211         if (!valid_event_struct(pool, evt)) {
 212                 dev_err(evt->hostdata->dev, "Freeing invalid event_struct %p "
 213                         "(not in pool %p)\n", evt, pool->events);
 214                 return;
 215         }
 216         if (atomic_inc_return(&evt->free) != 1) {
 217                 dev_err(evt->hostdata->dev, "Freeing event_struct %p "
 218                         "which is not in use!\n", evt);
 219                 return;
 220         }
 221 }
 222
 223 /**
 224  * get_evt_struct: - Gets the next free event in pool
 225  * @pool:       event_pool that contains the events to be searched
 226  *
 227  * Returns the next event in "free" state, and NULL if none are free.
 228  * Note that no synchronization is done here, we assume the host_lock
 229  * will syncrhonze things.
 230 */
 231 static struct srp_event_struct *get_event_struct(struct event_pool *pool)
 232 {
 233         int i;
 234         int poolsize = pool->size;
 235         int offset = pool->next;
 236
 237         for (i = 0; i < poolsize; i++) {
 238                 offset = (offset + 1) % poolsize;
 239                 if (!atomic_dec_if_positive(&pool->events[offset].free)) {
 240                         pool->next = offset;
 241                         return &pool->events[offset];
 242                 }
 243         }
 244
 245         printk(KERN_ERR "ibmvscsi: found no event struct in pool!\n");
 246         return NULL;
 247 }
 248
 249 /**
 250  * init_event_struct: Initialize fields in an event struct that are always
 251  *                    required.
 252  * @evt:        The event
 253  * @done:       Routine to call when the event is responded to
 254  * @format:     SRP or MAD format
 255  * @timeout:    timeout value set in the CRQ
 256  */
 257 static void init_event_struct(struct srp_event_struct *evt_struct,
 258                               void (*done) (struct srp_event_struct *),
 259                               u8 format,
 260                               int timeout)
 261 {
 262         evt_struct->cmnd = NULL;
 263         evt_struct->cmnd_done = NULL;
 264         evt_struct->sync_srp = NULL;
 265         evt_struct->crq.format = format;
 266         evt_struct->crq.timeout = timeout;
 267         evt_struct->done = done;
 268 }
 269
 270 /* ------------------------------------------------------------
 271  * Routines for receiving SCSI responses from the hosting partition
 272  */
 273
 274 /**
 275  * set_srp_direction: Set the fields in the srp related to data
 276  *     direction and number of buffers based on the direction in
 277  *     the scsi_cmnd and the number of buffers
 278  */
 279 static void set_srp_direction(struct scsi_cmnd *cmd,
 280                               struct srp_cmd *srp_cmd,
 281                               int numbuf)
 282 {
 283         u8 fmt;
 284
 285         if (numbuf == 0)
 286                 return;
 287
 288         if (numbuf == 1)
 289                 fmt = SRP_DATA_DESC_DIRECT;
 290         else {
 291                 fmt = SRP_DATA_DESC_INDIRECT;
 292                 numbuf = min(numbuf, MAX_INDIRECT_BUFS);
 293
 294                 if (cmd->sc_data_direction == DMA_TO_DEVICE)
 295                         srp_cmd->data_out_desc_cnt = numbuf;
 296                 else
 297                         srp_cmd->data_in_desc_cnt = numbuf;
 298         }
 299
 300         if (cmd->sc_data_direction == DMA_TO_DEVICE)
 301                 srp_cmd->buf_fmt = fmt << 4;
 302         else
 303                 srp_cmd->buf_fmt = fmt;
 304 }
 305
 306 static void unmap_sg_list(int num_entries,
 307                 struct device *dev,
 308                 struct srp_direct_buf *md)
 309 {
 310         int i;
 311
 312         for (i = 0; i < num_entries; ++i)
 313                 dma_unmap_single(dev, md[i].va, md[i].len, DMA_BIDIRECTIONAL);
 314 }
 315
 316 /**
 317  * unmap_cmd_data: - Unmap data pointed in srp_cmd based on the format
 318  * @cmd:        srp_cmd whose additional_data member will be unmapped
 319  * @dev:        device for which the memory is mapped
 320  *
 321 */
 322 static void unmap_cmd_data(struct srp_cmd *cmd,
 323                            struct srp_event_struct *evt_struct,
 324                            struct device *dev)
 325 {
 326         u8 out_fmt, in_fmt;
 327
 328         out_fmt = cmd->buf_fmt >> 4;
 329         in_fmt = cmd->buf_fmt & ((1U << 4) - 1);
 330
 331         if (out_fmt == SRP_NO_DATA_DESC && in_fmt == SRP_NO_DATA_DESC)
 332                 return;
 333         else if (out_fmt == SRP_DATA_DESC_DIRECT ||
 334                  in_fmt == SRP_DATA_DESC_DIRECT) {
 335                 struct srp_direct_buf *data =
 336                         (struct srp_direct_buf *) cmd->add_data;
 337                 dma_unmap_single(dev, data->va, data->len, DMA_BIDIRECTIONAL);
 338         } else {
 339                 struct srp_indirect_buf *indirect =
 340                         (struct srp_indirect_buf *) cmd->add_data;
 341                 int num_mapped = indirect->table_desc.len /
 342                         sizeof(struct srp_direct_buf);
 343
 344                 if (num_mapped <= MAX_INDIRECT_BUFS) {
 345                         unmap_sg_list(num_mapped, dev, &indirect->desc_list[0]);
 346                         return;
 347                 }
 348
 349                 unmap_sg_list(num_mapped, dev, evt_struct->ext_list);
 350         }
 351 }
 352
 353 static int map_sg_list(struct scsi_cmnd *cmd, int nseg,
 354                        struct srp_direct_buf *md)
 355 {
 356         int i;
 357         struct scatterlist *sg;
 358         u64 total_length = 0;
 359
 360         scsi_for_each_sg(cmd, sg, nseg, i) {
 361                 struct srp_direct_buf *descr = md + i;
 362                 descr->va = sg_dma_address(sg);
 363                 descr->len = sg_dma_len(sg);
 364                 descr->key = 0;
 365                 total_length += sg_dma_len(sg);
 366         }
 367         return total_length;
 368 }
 369
 370 /**
 371  * map_sg_data: - Maps dma for a scatterlist and initializes decriptor fields
 372  * @cmd:        Scsi_Cmnd with the scatterlist
 373  * @srp_cmd:    srp_cmd that contains the memory descriptor
 374  * @dev:        device for which to map dma memory
 375  *
 376  * Called by map_data_for_srp_cmd() when building srp cmd from scsi cmd.
 377  * Returns 1 on success.
 378 */
 379 static int map_sg_data(struct scsi_cmnd *cmd,
 380                        struct srp_event_struct *evt_struct,
 381                        struct srp_cmd *srp_cmd, struct device *dev)
 382 {
 383
 384         int sg_mapped;
 385         u64 total_length = 0;
 386         struct srp_direct_buf *data =
 387                 (struct srp_direct_buf *) srp_cmd->add_data;
 388         struct srp_indirect_buf *indirect =
 389                 (struct srp_indirect_buf *) data;
 390
 391         sg_mapped = scsi_dma_map(cmd);
 392         if (!sg_mapped)
 393                 return 1;
 394         else if (sg_mapped < 0)
 395                 return 0;
 396
 397         set_srp_direction(cmd, srp_cmd, sg_mapped);
 398
 399         /* special case; we can use a single direct descriptor */
 400         if (sg_mapped == 1) {
 401                 map_sg_list(cmd, sg_mapped, data);
 402                 return 1;
 403         }
 404
 405         indirect->table_desc.va = 0;
 406         indirect->table_desc.len = sg_mapped * sizeof(struct srp_direct_buf);
 407         indirect->table_desc.key = 0;
 408
 409         if (sg_mapped <= MAX_INDIRECT_BUFS) {
 410                 total_length = map_sg_list(cmd, sg_mapped,
 411                                            &indirect->desc_list[0]);
 412                 indirect->len = total_length;
 413                 return 1;
 414         }
 415
 416         /* get indirect table */
 417         if (!evt_struct->ext_list) {
 418                 evt_struct->ext_list = (struct srp_direct_buf *)
 419                         dma_alloc_coherent(dev,
 420                                            SG_ALL * sizeof(struct srp_direct_buf),
 421                                            &evt_struct->ext_list_token, 0);
 422                 if (!evt_struct->ext_list) {
 423                         sdev_printk(KERN_ERR, cmd->device,
 424                                     "Can't allocate memory for indirect table\n");
 425                         return 0;
 426                 }
 427         }
 428
 429         total_length = map_sg_list(cmd, sg_mapped, evt_struct->ext_list);
 430
 431         indirect->len = total_length;
 432         indirect->table_desc.va = evt_struct->ext_list_token;
 433         indirect->table_desc.len = sg_mapped * sizeof(indirect->desc_list[0]);
 434         memcpy(indirect->desc_list, evt_struct->ext_list,
 435                MAX_INDIRECT_BUFS * sizeof(struct srp_direct_buf));
 436         return 1;
 437 }
 438
 439 /**
 440  * map_data_for_srp_cmd: - Calls functions to map data for srp cmds
 441  * @cmd:        struct scsi_cmnd with the memory to be mapped
 442  * @srp_cmd:    srp_cmd that contains the memory descriptor
 443  * @dev:        dma device for which to map dma memory
 444  *
 445  * Called by scsi_cmd_to_srp_cmd() when converting scsi cmds to srp cmds
 446  * Returns 1 on success.
 447 */
 448 static int map_data_for_srp_cmd(struct scsi_cmnd *cmd,
 449                                 struct srp_event_struct *evt_struct,
 450                                 struct srp_cmd *srp_cmd, struct device *dev)
 451 {
 452         switch (cmd->sc_data_direction) {
 453         case DMA_FROM_DEVICE:
 454         case DMA_TO_DEVICE:
 455                 break;
 456         case DMA_NONE:
 457                 return 1;
 458         case DMA_BIDIRECTIONAL:
 459                 sdev_printk(KERN_ERR, cmd->device,
 460                             "Can't map DMA_BIDIRECTIONAL to read/write\n");
 461                 return 0;
 462         default:
 463                 sdev_printk(KERN_ERR, cmd->device,
 464                             "Unknown data direction 0x%02x; can't map!\n",
 465                             cmd->sc_data_direction);
 466                 return 0;
 467         }
 468
 469         return map_sg_data(cmd, evt_struct, srp_cmd, dev);
 470 }
 471
 472 /**
 473  * purge_requests: Our virtual adapter just shut down.  purge any sent requests
 474  * @hostdata:    the adapter
 475  */
 476 static void purge_requests(struct ibmvscsi_host_data *hostdata, int error_code)
 477 {
 478         struct srp_event_struct *tmp_evt, *pos;
 479         unsigned long flags;
 480
 481         spin_lock_irqsave(hostdata->host->host_lock, flags);
 482         list_for_each_entry_safe(tmp_evt, pos, &hostdata->sent, list) {
 483                 list_del(&tmp_evt->list);
 484                 del_timer(&tmp_evt->timer);
 485                 if (tmp_evt->cmnd) {
 486                         tmp_evt->cmnd->result = (error_code << 16);
 487                         unmap_cmd_data(&tmp_evt->iu.srp.cmd,
 488                                        tmp_evt,
 489                                        tmp_evt->hostdata->dev);
 490                         if (tmp_evt->cmnd_done)
 491                                 tmp_evt->cmnd_done(tmp_evt->cmnd);
 492                 } else if (tmp_evt->done)
 493                         tmp_evt->done(tmp_evt);
 494                 free_event_struct(&tmp_evt->hostdata->pool, tmp_evt);
 495         }
 496         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
 497 }
 498
 499 /**
 500  * ibmvscsi_reset_host - Reset the connection to the server
 501  * @hostdata:   struct ibmvscsi_host_data to reset
 502 */
 503 static void ibmvscsi_reset_host(struct ibmvscsi_host_data *hostdata)
 504 {
 505         scsi_block_requests(hostdata->host);
 506         atomic_set(&hostdata->request_limit, 0);
 507
 508         purge_requests(hostdata, DID_ERROR);
 509         if ((ibmvscsi_reset_crq_queue(&hostdata->queue, hostdata)) ||
 510             (ibmvscsi_send_crq(hostdata, 0xC001000000000000LL, 0)) ||
 511             (vio_enable_interrupts(to_vio_dev(hostdata->dev)))) {
 512                 atomic_set(&hostdata->request_limit, -1);
 513                 dev_err(hostdata->dev, "error after reset\n");
 514         }
 515
 516         scsi_unblock_requests(hostdata->host);
 517 }
 518
 519 /**
 520  * ibmvscsi_timeout - Internal command timeout handler
 521  * @evt_struct: struct srp_event_struct that timed out
 522  *
 523  * Called when an internally generated command times out
 524 */
 525 static void ibmvscsi_timeout(struct srp_event_struct *evt_struct)
 526 {
 527         struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
 528
 529         dev_err(hostdata->dev, "Command timed out (%x). Resetting connection\n",
 530                 evt_struct->iu.srp.cmd.opcode);
 531
 532         ibmvscsi_reset_host(hostdata);
 533 }
 534
 535
 536 /* ------------------------------------------------------------
 537  * Routines for sending and receiving SRPs
 538  */
 539 /**
 540  * ibmvscsi_send_srp_event: - Transforms event to u64 array and calls send_crq()
 541  * @evt_struct: evt_struct to be sent
 542  * @hostdata:   ibmvscsi_host_data of host
 543  * @timeout:    timeout in seconds - 0 means do not time command
 544  *
 545  * Returns the value returned from ibmvscsi_send_crq(). (Zero for success)
 546  * Note that this routine assumes that host_lock is held for synchronization
 547 */
 548 static int ibmvscsi_send_srp_event(struct srp_event_struct *evt_struct,
 549                                    struct ibmvscsi_host_data *hostdata,
 550                                    unsigned long timeout)
 551 {
 552         u64 *crq_as_u64 = (u64 *) &evt_struct->crq;
 553         int request_status;
 554         int rc;
 555
 556         /* If we have exhausted our request limit, just fail this request,
 557          * unless it is for a reset or abort.
 558          * Note that there are rare cases involving driver generated requests
 559          * (such as task management requests) that the mid layer may think we
 560          * can handle more requests (can_queue) when we actually can't
 561          */
 562         if (evt_struct->crq.format == VIOSRP_SRP_FORMAT) {
 563                 request_status =
 564                         atomic_dec_if_positive(&hostdata->request_limit);
 565                 /* If request limit was -1 when we started, it is now even
 566                  * less than that
 567                  */
 568                 if (request_status < -1)
 569                         goto send_error;
 570                 /* Otherwise, we may have run out of requests. */
 571                 /* Abort and reset calls should make it through.
 572                  * Nothing except abort and reset should use the last two
 573                  * slots unless we had two or less to begin with.
 574                  */
 575                 else if (request_status < 2 &&
 576                          evt_struct->iu.srp.cmd.opcode != SRP_TSK_MGMT) {
 577                         /* In the case that we have less than two requests
 578                          * available, check the server limit as a combination
 579                          * of the request limit and the number of requests
 580                          * in-flight (the size of the send list).  If the
 581                          * server limit is greater than 2, return busy so
 582                          * that the last two are reserved for reset and abort.
 583                          */
 584                         int server_limit = request_status;
 585                         struct srp_event_struct *tmp_evt;
 586
 587                         list_for_each_entry(tmp_evt, &hostdata->sent, list) {
 588                                 server_limit++;
 589                         }
 590
 591                         if (server_limit > 2)
 592                                 goto send_busy;
 593                 }
 594         }
 595
 596         /* Copy the IU into the transfer area */
 597         *evt_struct->xfer_iu = evt_struct->iu;
 598         evt_struct->xfer_iu->srp.rsp.tag = (u64)evt_struct;
 599
 600         /* Add this to the sent list.  We need to do this
 601          * before we actually send
 602          * in case it comes back REALLY fast
 603          */
 604         list_add_tail(&evt_struct->list, &hostdata->sent);
 605
 606         init_timer(&evt_struct->timer);
 607         if (timeout) {
 608                 evt_struct->timer.data = (unsigned long) evt_struct;
 609                 evt_struct->timer.expires = jiffies + (timeout * HZ);
 610                 evt_struct->timer.function = (void (*)(unsigned long))ibmvscsi_timeout;
 611                 add_timer(&evt_struct->timer);
 612         }
 613
 614         if ((rc =
 615              ibmvscsi_send_crq(hostdata, crq_as_u64[0], crq_as_u64[1])) != 0) {
 616                 list_del(&evt_struct->list);
 617                 del_timer(&evt_struct->timer);
 618
 619                 dev_err(hostdata->dev, "send error %d\n", rc);
 620                 atomic_inc(&hostdata->request_limit);
 621                 goto send_error;
 622         }
 623
 624         return 0;
 625
 626  send_busy:
 627         unmap_cmd_data(&evt_struct->iu.srp.cmd, evt_struct, hostdata->dev);
 628
 629         free_event_struct(&hostdata->pool, evt_struct);
 630         atomic_inc(&hostdata->request_limit);
 631         return SCSI_MLQUEUE_HOST_BUSY;
 632
 633  send_error:
 634         unmap_cmd_data(&evt_struct->iu.srp.cmd, evt_struct, hostdata->dev);
 635
 636         if (evt_struct->cmnd != NULL) {
 637                 evt_struct->cmnd->result = DID_ERROR << 16;
 638                 evt_struct->cmnd_done(evt_struct->cmnd);
 639         } else if (evt_struct->done)
 640                 evt_struct->done(evt_struct);
 641
 642         free_event_struct(&hostdata->pool, evt_struct);
 643         return 0;
 644 }
 645
 646 /**
 647  * handle_cmd_rsp: -  Handle responses from commands
 648  * @evt_struct: srp_event_struct to be handled
 649  *
 650  * Used as a callback by when sending scsi cmds.
 651  * Gets called by ibmvscsi_handle_crq()
 652 */
 653 static void handle_cmd_rsp(struct srp_event_struct *evt_struct)
 654 {
 655         struct srp_rsp *rsp = &evt_struct->xfer_iu->srp.rsp;
 656         struct scsi_cmnd *cmnd = evt_struct->cmnd;
 657
 658         if (unlikely(rsp->opcode != SRP_RSP)) {
 659                 if (printk_ratelimit())
 660                         dev_warn(evt_struct->hostdata->dev,
 661                                  "bad SRP RSP type %d\n", rsp->opcode);
 662         }
 663
 664         if (cmnd) {
 665                 cmnd->result = rsp->status;
 666                 if (((cmnd->result >> 1) & 0x1f) == CHECK_CONDITION)
 667                         memcpy(cmnd->sense_buffer,
 668                                rsp->data,
 669                                rsp->sense_data_len);
 670                 unmap_cmd_data(&evt_struct->iu.srp.cmd,
 671                                evt_struct,
 672                                evt_struct->hostdata->dev);
 673
 674                 if (rsp->flags & SRP_RSP_FLAG_DOOVER)
 675                         scsi_set_resid(cmnd, rsp->data_out_res_cnt);
 676                 else if (rsp->flags & SRP_RSP_FLAG_DIOVER)
 677                         scsi_set_resid(cmnd, rsp->data_in_res_cnt);
 678         }
 679
 680         if (evt_struct->cmnd_done)
 681                 evt_struct->cmnd_done(cmnd);
 682 }
 683
 684 /**
 685  * lun_from_dev: - Returns the lun of the scsi device
 686  * @dev:        struct scsi_device
 687  *
 688 */
 689 static inline u16 lun_from_dev(struct scsi_device *dev)
 690 {
 691         return (0x2 << 14) | (dev->id << 8) | (dev->channel << 5) | dev->lun;
 692 }
 693
 694 /**
 695  * ibmvscsi_queue: - The queuecommand function of the scsi template
 696  * @cmd:        struct scsi_cmnd to be executed
 697  * @done:       Callback function to be called when cmd is completed
 698 */
 699 static int ibmvscsi_queuecommand(struct scsi_cmnd *cmnd,
 700                                  void (*done) (struct scsi_cmnd *))
 701 {
 702         struct srp_cmd *srp_cmd;
 703         struct srp_event_struct *evt_struct;
 704         struct srp_indirect_buf *indirect;
 705         struct ibmvscsi_host_data *hostdata = shost_priv(cmnd->device->host);
 706         u16 lun = lun_from_dev(cmnd->device);
 707         u8 out_fmt, in_fmt;
 708
 709         evt_struct = get_event_struct(&hostdata->pool);
 710         if (!evt_struct)
 711                 return SCSI_MLQUEUE_HOST_BUSY;
 712
 713         /* Set up the actual SRP IU */
 714         srp_cmd = &evt_struct->iu.srp.cmd;
 715         memset(srp_cmd, 0x00, SRP_MAX_IU_LEN);
 716         srp_cmd->opcode = SRP_CMD;
 717         memcpy(srp_cmd->cdb, cmnd->cmnd, sizeof(cmnd->cmnd));
 718         srp_cmd->lun = ((u64) lun) << 48;
 719
 720         if (!map_data_for_srp_cmd(cmnd, evt_struct, srp_cmd, hostdata->dev)) {
 721                 sdev_printk(KERN_ERR, cmnd->device, "couldn't convert cmd to srp_cmd\n");
 722                 free_event_struct(&hostdata->pool, evt_struct);
 723                 return SCSI_MLQUEUE_HOST_BUSY;
 724         }
 725
 726         init_event_struct(evt_struct,
 727                           handle_cmd_rsp,
 728                           VIOSRP_SRP_FORMAT,
 729                           cmnd->timeout_per_command/HZ);
 730
 731         evt_struct->cmnd = cmnd;
 732         evt_struct->cmnd_done = done;
 733
 734         /* Fix up dma address of the buffer itself */
 735         indirect = (struct srp_indirect_buf *) srp_cmd->add_data;
 736         out_fmt = srp_cmd->buf_fmt >> 4;
 737         in_fmt = srp_cmd->buf_fmt & ((1U << 4) - 1);
 738         if ((in_fmt == SRP_DATA_DESC_INDIRECT ||
 739              out_fmt == SRP_DATA_DESC_INDIRECT) &&
 740             indirect->table_desc.va == 0) {
 741                 indirect->table_desc.va = evt_struct->crq.IU_data_ptr +
 742                         offsetof(struct srp_cmd, add_data) +
 743                         offsetof(struct srp_indirect_buf, desc_list);
 744         }
 745
 746         return ibmvscsi_send_srp_event(evt_struct, hostdata, 0);
 747 }
 748
 749 /* ------------------------------------------------------------
 750  * Routines for driver initialization
 751  */
 752 /**
 753  * adapter_info_rsp: - Handle response to MAD adapter info request
 754  * @evt_struct: srp_event_struct with the response
 755  *
 756  * Used as a "done" callback by when sending adapter_info. Gets called
 757  * by ibmvscsi_handle_crq()
 758 */
 759 static void adapter_info_rsp(struct srp_event_struct *evt_struct)
 760 {
 761         struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
 762         dma_unmap_single(hostdata->dev,
 763                          evt_struct->iu.mad.adapter_info.buffer,
 764                          evt_struct->iu.mad.adapter_info.common.length,
 765                          DMA_BIDIRECTIONAL);
 766
 767         if (evt_struct->xfer_iu->mad.adapter_info.common.status) {
 768                 dev_err(hostdata->dev, "error %d getting adapter info\n",
 769                         evt_struct->xfer_iu->mad.adapter_info.common.status);
 770         } else {
 771                 dev_info(hostdata->dev, "host srp version: %s, "
 772                          "host partition %s (%d), OS %d, max io %u\n",
 773                          hostdata->madapter_info.srp_version,
 774                          hostdata->madapter_info.partition_name,
 775                          hostdata->madapter_info.partition_number,
 776                          hostdata->madapter_info.os_type,
 777                          hostdata->madapter_info.port_max_txu[0]);
 778
 779                 if (hostdata->madapter_info.port_max_txu[0])
 780                         hostdata->host->max_sectors =
 781                                 hostdata->madapter_info.port_max_txu[0] >> 9;
 782
 783                 if (hostdata->madapter_info.os_type == 3 &&
 784                     strcmp(hostdata->madapter_info.srp_version, "1.6a") <= 0) {
 785                         dev_err(hostdata->dev, "host (Ver. %s) doesn't support large transfers\n",
 786                                 hostdata->madapter_info.srp_version);
 787                         dev_err(hostdata->dev, "limiting scatterlists to %d\n",
 788                                 MAX_INDIRECT_BUFS);
 789                         hostdata->host->sg_tablesize = MAX_INDIRECT_BUFS;
 790                 }
 791         }
 792 }
 793
 794 /**
 795  * send_mad_adapter_info: - Sends the mad adapter info request
 796  *      and stores the result so it can be retrieved with
 797  *      sysfs.  We COULD consider causing a failure if the
 798  *      returned SRP version doesn't match ours.
 799  * @hostdata:   ibmvscsi_host_data of host
 800  *
 801  * Returns zero if successful.
 802 */
 803 static void send_mad_adapter_info(struct ibmvscsi_host_data *hostdata)
 804 {
 805         struct viosrp_adapter_info *req;
 806         struct srp_event_struct *evt_struct;
 807         unsigned long flags;
 808         dma_addr_t addr;
 809
 810         evt_struct = get_event_struct(&hostdata->pool);
 811         if (!evt_struct) {
 812                 dev_err(hostdata->dev,
 813                         "couldn't allocate an event for ADAPTER_INFO_REQ!\n");
 814                 return;
 815         }
 816
 817         init_event_struct(evt_struct,
 818                           adapter_info_rsp,
 819                           VIOSRP_MAD_FORMAT,
 820                           init_timeout);
 821
 822         req = &evt_struct->iu.mad.adapter_info;
 823         memset(req, 0x00, sizeof(*req));
 824
 825         req->common.type = VIOSRP_ADAPTER_INFO_TYPE;
 826         req->common.length = sizeof(hostdata->madapter_info);
 827         req->buffer = addr = dma_map_single(hostdata->dev,
 828                                             &hostdata->madapter_info,
 829                                             sizeof(hostdata->madapter_info),
 830                                             DMA_BIDIRECTIONAL);
 831
 832         if (dma_mapping_error(req->buffer)) {
 833                 dev_err(hostdata->dev, "Unable to map request_buffer for adapter_info!\n");
 834                 free_event_struct(&hostdata->pool, evt_struct);
 835                 return;
 836         }
 837
 838         spin_lock_irqsave(hostdata->host->host_lock, flags);
 839         if (ibmvscsi_send_srp_event(evt_struct, hostdata, init_timeout * 2)) {
 840                 dev_err(hostdata->dev, "couldn't send ADAPTER_INFO_REQ!\n");
 841                 dma_unmap_single(hostdata->dev,
 842                                  addr,
 843                                  sizeof(hostdata->madapter_info),
 844                                  DMA_BIDIRECTIONAL);
 845         }
 846         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
 847 };
 848
 849 /**
 850  * login_rsp: - Handle response to SRP login request
 851  * @evt_struct: srp_event_struct with the response
 852  *
 853  * Used as a "done" callback by when sending srp_login. Gets called
 854  * by ibmvscsi_handle_crq()
 855 */
 856 static void login_rsp(struct srp_event_struct *evt_struct)
 857 {
 858         struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
 859         switch (evt_struct->xfer_iu->srp.login_rsp.opcode) {
 860         case SRP_LOGIN_RSP:     /* it worked! */
 861                 break;
 862         case SRP_LOGIN_REJ:     /* refused! */
 863                 dev_info(hostdata->dev, "SRP_LOGIN_REJ reason %u\n",
 864                          evt_struct->xfer_iu->srp.login_rej.reason);
 865                 /* Login failed.  */
 866                 atomic_set(&hostdata->request_limit, -1);
 867                 return;
 868         default:
 869                 dev_err(hostdata->dev, "Invalid login response typecode 0x%02x!\n",
 870                         evt_struct->xfer_iu->srp.login_rsp.opcode);
 871                 /* Login failed.  */
 872                 atomic_set(&hostdata->request_limit, -1);
 873                 return;
 874         }
 875
 876         dev_info(hostdata->dev, "SRP_LOGIN succeeded\n");
 877
 878         if (evt_struct->xfer_iu->srp.login_rsp.req_lim_delta < 0)
 879                 dev_err(hostdata->dev, "Invalid request_limit.\n");
 880
 881         /* Now we know what the real request-limit is.
 882          * This value is set rather than added to request_limit because
 883          * request_limit could have been set to -1 by this client.
 884          */
 885         atomic_set(&hostdata->request_limit,
 886                    evt_struct->xfer_iu->srp.login_rsp.req_lim_delta);
 887
 888         /* If we had any pending I/Os, kick them */
 889         scsi_unblock_requests(hostdata->host);
 890
 891         send_mad_adapter_info(hostdata);
 892         return;
 893 }
 894
 895 /**
 896  * send_srp_login: - Sends the srp login
 897  * @hostdata:   ibmvscsi_host_data of host
 898  *
 899  * Returns zero if successful.
 900 */
 901 static int send_srp_login(struct ibmvscsi_host_data *hostdata)
 902 {
 903         int rc;
 904         unsigned long flags;
 905         struct srp_login_req *login;
 906         struct srp_event_struct *evt_struct = get_event_struct(&hostdata->pool);
 907         if (!evt_struct) {
 908                 dev_err(hostdata->dev, "couldn't allocate an event for login req!\n");
 909                 return FAILED;
 910         }
 911
 912         init_event_struct(evt_struct,
 913                           login_rsp,
 914                           VIOSRP_SRP_FORMAT,
 915                           init_timeout);
 916
 917         login = &evt_struct->iu.srp.login_req;
 918         memset(login, 0x00, sizeof(struct srp_login_req));
 919         login->opcode = SRP_LOGIN_REQ;
 920         login->req_it_iu_len = sizeof(union srp_iu);
 921         login->req_buf_fmt = SRP_BUF_FORMAT_DIRECT | SRP_BUF_FORMAT_INDIRECT;
 922
 923         spin_lock_irqsave(hostdata->host->host_lock, flags);
 924         /* Start out with a request limit of 1, since this is negotiated in
 925          * the login request we are just sending
 926          */
 927         atomic_set(&hostdata->request_limit, 1);
 928
 929         rc = ibmvscsi_send_srp_event(evt_struct, hostdata, init_timeout * 2);
 930         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
 931         dev_info(hostdata->dev, "sent SRP login\n");
 932         return rc;
 933 };
 934
 935 /**
 936  * sync_completion: Signal that a synchronous command has completed
 937  * Note that after returning from this call, the evt_struct is freed.
 938  * the caller waiting on this completion shouldn't touch the evt_struct
 939  * again.
 940  */
 941 static void sync_completion(struct srp_event_struct *evt_struct)
 942 {
 943         /* copy the response back */
 944         if (evt_struct->sync_srp)
 945                 *evt_struct->sync_srp = *evt_struct->xfer_iu;
 946
 947         complete(&evt_struct->comp);
 948 }
 949
 950 /**
 951  * ibmvscsi_abort: Abort a command...from scsi host template
 952  * send this over to the server and wait synchronously for the response
 953  */
 954 static int ibmvscsi_eh_abort_handler(struct scsi_cmnd *cmd)
 955 {
 956         struct ibmvscsi_host_data *hostdata = shost_priv(cmd->device->host);
 957         struct srp_tsk_mgmt *tsk_mgmt;
 958         struct srp_event_struct *evt;
 959         struct srp_event_struct *tmp_evt, *found_evt;
 960         union viosrp_iu srp_rsp;
 961         int rsp_rc;
 962         unsigned long flags;
 963         u16 lun = lun_from_dev(cmd->device);
 964
 965         /* First, find this command in our sent list so we can figure
 966          * out the correct tag
 967          */
 968         spin_lock_irqsave(hostdata->host->host_lock, flags);
 969         found_evt = NULL;
 970         list_for_each_entry(tmp_evt, &hostdata->sent, list) {
 971                 if (tmp_evt->cmnd == cmd) {
 972                         found_evt = tmp_evt;
 973                         break;
 974                 }
 975         }
 976
 977         if (!found_evt) {
 978                 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
 979                 return SUCCESS;
 980         }
 981
 982         evt = get_event_struct(&hostdata->pool);
 983         if (evt == NULL) {
 984                 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
 985                 sdev_printk(KERN_ERR, cmd->device, "failed to allocate abort event\n");
 986                 return FAILED;
 987         }
 988
 989         init_event_struct(evt,
 990                           sync_completion,
 991                           VIOSRP_SRP_FORMAT,
 992                           init_timeout);
 993
 994         tsk_mgmt = &evt->iu.srp.tsk_mgmt;
 995
 996         /* Set up an abort SRP command */
 997         memset(tsk_mgmt, 0x00, sizeof(*tsk_mgmt));
 998         tsk_mgmt->opcode = SRP_TSK_MGMT;
 999         tsk_mgmt->lun = ((u64) lun) << 48;
1000         tsk_mgmt->tsk_mgmt_func = SRP_TSK_ABORT_TASK;
1001         tsk_mgmt->task_tag = (u64) found_evt;
1002
1003         sdev_printk(KERN_INFO, cmd->device, "aborting command. lun 0x%lx, tag 0x%lx\n",
1004                     tsk_mgmt->lun, tsk_mgmt->task_tag);
1005
1006         evt->sync_srp = &srp_rsp;
1007         init_completion(&evt->comp);
1008         rsp_rc = ibmvscsi_send_srp_event(evt, hostdata, init_timeout * 2);
1009         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1010         if (rsp_rc != 0) {
1011                 sdev_printk(KERN_ERR, cmd->device,
1012                             "failed to send abort() event. rc=%d\n", rsp_rc);
1013                 return FAILED;
1014         }
1015
1016         wait_for_completion(&evt->comp);
1017
1018         /* make sure we got a good response */
1019         if (unlikely(srp_rsp.srp.rsp.opcode != SRP_RSP)) {
1020                 if (printk_ratelimit())
1021                         sdev_printk(KERN_WARNING, cmd->device, "abort bad SRP RSP type %d\n",
1022                                     srp_rsp.srp.rsp.opcode);
1023                 return FAILED;
1024         }
1025
1026         if (srp_rsp.srp.rsp.flags & SRP_RSP_FLAG_RSPVALID)
1027                 rsp_rc = *((int *)srp_rsp.srp.rsp.data);
1028         else
1029                 rsp_rc = srp_rsp.srp.rsp.status;
1030
1031         if (rsp_rc) {
1032                 if (printk_ratelimit())
1033                         sdev_printk(KERN_WARNING, cmd->device,
1034                                     "abort code %d for task tag 0x%lx\n",
1035                                     rsp_rc, tsk_mgmt->task_tag);
1036                 return FAILED;
1037         }
1038
1039         /* Because we dropped the spinlock above, it's possible
1040          * The event is no longer in our list.  Make sure it didn't
1041          * complete while we were aborting
1042          */
1043         spin_lock_irqsave(hostdata->host->host_lock, flags);
1044         found_evt = NULL;
1045         list_for_each_entry(tmp_evt, &hostdata->sent, list) {
1046                 if (tmp_evt->cmnd == cmd) {
1047                         found_evt = tmp_evt;
1048                         break;
1049                 }
1050         }
1051
1052         if (found_evt == NULL) {
1053                 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1054                 sdev_printk(KERN_INFO, cmd->device, "aborted task tag 0x%lx completed\n",
1055                             tsk_mgmt->task_tag);
1056                 return SUCCESS;
1057         }
1058
1059         sdev_printk(KERN_INFO, cmd->device, "successfully aborted task tag 0x%lx\n",
1060                     tsk_mgmt->task_tag);
1061
1062         cmd->result = (DID_ABORT << 16);
1063         list_del(&found_evt->list);
1064         unmap_cmd_data(&found_evt->iu.srp.cmd, found_evt,
1065                        found_evt->hostdata->dev);
1066         free_event_struct(&found_evt->hostdata->pool, found_evt);
1067         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1068         atomic_inc(&hostdata->request_limit);
1069         return SUCCESS;
1070 }
1071
1072 /**
1073  * ibmvscsi_eh_device_reset_handler: Reset a single LUN...from scsi host
1074  * template send this over to the server and wait synchronously for the
1075  * response
1076  */
1077 static int ibmvscsi_eh_device_reset_handler(struct scsi_cmnd *cmd)
1078 {
1079         struct ibmvscsi_host_data *hostdata = shost_priv(cmd->device->host);
1080         struct srp_tsk_mgmt *tsk_mgmt;
1081         struct srp_event_struct *evt;
1082         struct srp_event_struct *tmp_evt, *pos;
1083         union viosrp_iu srp_rsp;
1084         int rsp_rc;
1085         unsigned long flags;
1086         u16 lun = lun_from_dev(cmd->device);
1087
1088         spin_lock_irqsave(hostdata->host->host_lock, flags);
1089         evt = get_event_struct(&hostdata->pool);
1090         if (evt == NULL) {
1091                 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1092                 sdev_printk(KERN_ERR, cmd->device, "failed to allocate reset event\n");
1093                 return FAILED;
1094         }
1095
1096         init_event_struct(evt,
1097                           sync_completion,
1098                           VIOSRP_SRP_FORMAT,
1099                           init_timeout);
1100
1101         tsk_mgmt = &evt->iu.srp.tsk_mgmt;
1102
1103         /* Set up a lun reset SRP command */
1104         memset(tsk_mgmt, 0x00, sizeof(*tsk_mgmt));
1105         tsk_mgmt->opcode = SRP_TSK_MGMT;
1106         tsk_mgmt->lun = ((u64) lun) << 48;
1107         tsk_mgmt->tsk_mgmt_func = SRP_TSK_LUN_RESET;
1108
1109         sdev_printk(KERN_INFO, cmd->device, "resetting device. lun 0x%lx\n",
1110                     tsk_mgmt->lun);
1111
1112         evt->sync_srp = &srp_rsp;
1113         init_completion(&evt->comp);
1114         rsp_rc = ibmvscsi_send_srp_event(evt, hostdata, init_timeout * 2);
1115         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1116         if (rsp_rc != 0) {
1117                 sdev_printk(KERN_ERR, cmd->device,
1118                             "failed to send reset event. rc=%d\n", rsp_rc);
1119                 return FAILED;
1120         }
1121
1122         wait_for_completion(&evt->comp);
1123
1124         /* make sure we got a good response */
1125         if (unlikely(srp_rsp.srp.rsp.opcode != SRP_RSP)) {
1126                 if (printk_ratelimit())
1127                         sdev_printk(KERN_WARNING, cmd->device, "reset bad SRP RSP type %d\n",
1128                                     srp_rsp.srp.rsp.opcode);
1129                 return FAILED;
1130         }
1131
1132         if (srp_rsp.srp.rsp.flags & SRP_RSP_FLAG_RSPVALID)
1133                 rsp_rc = *((int *)srp_rsp.srp.rsp.data);
1134         else
1135                 rsp_rc = srp_rsp.srp.rsp.status;
1136
1137         if (rsp_rc) {
1138                 if (printk_ratelimit())
1139                         sdev_printk(KERN_WARNING, cmd->device,
1140                                     "reset code %d for task tag 0x%lx\n",
1141                                     rsp_rc, tsk_mgmt->task_tag);
1142                 return FAILED;
1143         }
1144
1145         /* We need to find all commands for this LUN that have not yet been
1146          * responded to, and fail them with DID_RESET
1147          */
1148         spin_lock_irqsave(hostdata->host->host_lock, flags);
1149         list_for_each_entry_safe(tmp_evt, pos, &hostdata->sent, list) {
1150                 if ((tmp_evt->cmnd) && (tmp_evt->cmnd->device == cmd->device)) {
1151                         if (tmp_evt->cmnd)
1152                                 tmp_evt->cmnd->result = (DID_RESET << 16);
1153                         list_del(&tmp_evt->list);
1154                         unmap_cmd_data(&tmp_evt->iu.srp.cmd, tmp_evt,
1155                                        tmp_evt->hostdata->dev);
1156                         free_event_struct(&tmp_evt->hostdata->pool,
1157                                                    tmp_evt);
1158                         atomic_inc(&hostdata->request_limit);
1159                         if (tmp_evt->cmnd_done)
1160                                 tmp_evt->cmnd_done(tmp_evt->cmnd);
1161                         else if (tmp_evt->done)
1162                                 tmp_evt->done(tmp_evt);
1163                 }
1164         }
1165         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1166         return SUCCESS;
1167 }
1168
1169 /**
1170  * ibmvscsi_eh_host_reset_handler - Reset the connection to the server
1171  * @cmd:        struct scsi_cmnd having problems
1172 */
1173 static int ibmvscsi_eh_host_reset_handler(struct scsi_cmnd *cmd)
1174 {
1175         unsigned long wait_switch = 0;
1176         struct ibmvscsi_host_data *hostdata = shost_priv(cmd->device->host);
1177
1178         dev_err(hostdata->dev, "Resetting connection due to error recovery\n");
1179
1180         ibmvscsi_reset_host(hostdata);
1181
1182         for (wait_switch = jiffies + (init_timeout * HZ);
1183              time_before(jiffies, wait_switch) &&
1184                      atomic_read(&hostdata->request_limit) < 2;) {
1185
1186                 msleep(10);
1187         }
1188
1189         if (atomic_read(&hostdata->request_limit) <= 0)
1190                 return FAILED;
1191
1192         return SUCCESS;
1193 }
1194
1195 /**
1196  * ibmvscsi_handle_crq: - Handles and frees received events in the CRQ
1197  * @crq:        Command/Response queue
1198  * @hostdata:   ibmvscsi_host_data of host
1199  *
1200 */
1201 void ibmvscsi_handle_crq(struct viosrp_crq *crq,
1202                          struct ibmvscsi_host_data *hostdata)
1203 {
1204         long rc;
1205         unsigned long flags;
1206         struct srp_event_struct *evt_struct =
1207             (struct srp_event_struct *)crq->IU_data_ptr;
1208         switch (crq->valid) {
1209         case 0xC0:              /* initialization */
1210                 switch (crq->format) {
1211                 case 0x01:      /* Initialization message */
1212                         dev_info(hostdata->dev, "partner initialized\n");
1213                         /* Send back a response */
1214                         if ((rc = ibmvscsi_send_crq(hostdata,
1215                                                     0xC002000000000000LL, 0)) == 0) {
1216                                 /* Now login */
1217                                 send_srp_login(hostdata);
1218                         } else {
1219                                 dev_err(hostdata->dev, "Unable to send init rsp. rc=%ld\n", rc);
1220                         }
1221
1222                         break;
1223                 case 0x02:      /* Initialization response */
1224                         dev_info(hostdata->dev, "partner initialization complete\n");
1225
1226                         /* Now login */
1227                         send_srp_login(hostdata);
1228                         break;
1229                 default:
1230                         dev_err(hostdata->dev, "unknown crq message type: %d\n", crq->format);
1231                 }
1232                 return;
1233         case 0xFF:      /* Hypervisor telling us the connection is closed */
1234                 scsi_block_requests(hostdata->host);
1235                 atomic_set(&hostdata->request_limit, 0);
1236                 if (crq->format == 0x06) {
1237                         /* We need to re-setup the interpartition connection */
1238                         dev_info(hostdata->dev, "Re-enabling adapter!\n");
1239                         purge_requests(hostdata, DID_REQUEUE);
1240                         if ((ibmvscsi_reenable_crq_queue(&hostdata->queue,
1241                                                         hostdata)) ||
1242                             (ibmvscsi_send_crq(hostdata,
1243                                                0xC001000000000000LL, 0))) {
1244                                         atomic_set(&hostdata->request_limit,
1245                                                    -1);
1246                                         dev_err(hostdata->dev, "error after enable\n");
1247                         }
1248                 } else {
1249                         dev_err(hostdata->dev, "Virtual adapter failed rc %d!\n",
1250                                 crq->format);
1251
1252                         purge_requests(hostdata, DID_ERROR);
1253                         if ((ibmvscsi_reset_crq_queue(&hostdata->queue,
1254                                                         hostdata)) ||
1255                             (ibmvscsi_send_crq(hostdata,
1256                                                0xC001000000000000LL, 0))) {
1257                                         atomic_set(&hostdata->request_limit,
1258                                                    -1);
1259                                         dev_err(hostdata->dev, "error after reset\n");
1260                         }
1261                 }
1262                 scsi_unblock_requests(hostdata->host);
1263                 return;
1264         case 0x80:              /* real payload */
1265                 break;
1266         default:
1267                 dev_err(hostdata->dev, "got an invalid message type 0x%02x\n",
1268                         crq->valid);
1269                 return;
1270         }
1271
1272         /* The only kind of payload CRQs we should get are responses to
1273          * things we send. Make sure this response is to something we
1274          * actually sent
1275          */
1276         if (!valid_event_struct(&hostdata->pool, evt_struct)) {
1277                 dev_err(hostdata->dev, "returned correlation_token 0x%p is invalid!\n",
1278                        (void *)crq->IU_data_ptr);
1279                 return;
1280         }
1281
1282         if (atomic_read(&evt_struct->free)) {
1283                 dev_err(hostdata->dev, "received duplicate correlation_token 0x%p!\n",
1284                         (void *)crq->IU_data_ptr);
1285                 return;
1286         }
1287
1288         if (crq->format == VIOSRP_SRP_FORMAT)
1289                 atomic_add(evt_struct->xfer_iu->srp.rsp.req_lim_delta,
1290                            &hostdata->request_limit);
1291
1292         del_timer(&evt_struct->timer);
1293
1294         if (evt_struct->done)
1295                 evt_struct->done(evt_struct);
1296         else
1297                 dev_err(hostdata->dev, "returned done() is NULL; not running it!\n");
1298
1299         /*
1300          * Lock the host_lock before messing with these structures, since we
1301          * are running in a task context
1302          */
1303         spin_lock_irqsave(evt_struct->hostdata->host->host_lock, flags);
1304         list_del(&evt_struct->list);
1305         free_event_struct(&evt_struct->hostdata->pool, evt_struct);
1306         spin_unlock_irqrestore(evt_struct->hostdata->host->host_lock, flags);
1307 }
1308
1309 /**
1310  * ibmvscsi_get_host_config: Send the command to the server to get host
1311  * configuration data.  The data is opaque to us.
1312  */
1313 static int ibmvscsi_do_host_config(struct ibmvscsi_host_data *hostdata,
1314                                    unsigned char *buffer, int length)
1315 {
1316         struct viosrp_host_config *host_config;
1317         struct srp_event_struct *evt_struct;
1318         unsigned long flags;
1319         dma_addr_t addr;
1320         int rc;
1321
1322         evt_struct = get_event_struct(&hostdata->pool);
1323         if (!evt_struct) {
1324                 dev_err(hostdata->dev, "couldn't allocate event for HOST_CONFIG!\n");
1325                 return -1;
1326         }
1327
1328         init_event_struct(evt_struct,
1329                           sync_completion,
1330                           VIOSRP_MAD_FORMAT,
1331                           init_timeout);
1332
1333         host_config = &evt_struct->iu.mad.host_config;
1334
1335         /* Set up a lun reset SRP command */
1336         memset(host_config, 0x00, sizeof(*host_config));
1337         host_config->common.type = VIOSRP_HOST_CONFIG_TYPE;
1338         host_config->common.length = length;
1339         host_config->buffer = addr = dma_map_single(hostdata->dev, buffer,
1340                                                     length,
1341                                                     DMA_BIDIRECTIONAL);
1342
1343         if (dma_mapping_error(host_config->buffer)) {
1344                 dev_err(hostdata->dev, "dma_mapping error getting host config\n");
1345                 free_event_struct(&hostdata->pool, evt_struct);
1346                 return -1;
1347         }
1348
1349         init_completion(&evt_struct->comp);
1350         spin_lock_irqsave(hostdata->host->host_lock, flags);
1351         rc = ibmvscsi_send_srp_event(evt_struct, hostdata, init_timeout * 2);
1352         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1353         if (rc == 0)
1354                 wait_for_completion(&evt_struct->comp);
1355         dma_unmap_single(hostdata->dev, addr, length, DMA_BIDIRECTIONAL);
1356
1357         return rc;
1358 }
1359
1360 /**
1361  * ibmvscsi_slave_configure: Set the "allow_restart" flag for each disk.
1362  * @sdev:       struct scsi_device device to configure
1363  *
1364  * Enable allow_restart for a device if it is a disk.  Adjust the
1365  * queue_depth here also as is required by the documentation for
1366  * struct scsi_host_template.
1367  */
1368 static int ibmvscsi_slave_configure(struct scsi_device *sdev)
1369 {
1370         struct Scsi_Host *shost = sdev->host;
1371         unsigned long lock_flags = 0;
1372
1373         spin_lock_irqsave(shost->host_lock, lock_flags);
1374         if (sdev->type == TYPE_DISK)
1375                 sdev->allow_restart = 1;
1376         scsi_adjust_queue_depth(sdev, 0, shost->cmd_per_lun);
1377         spin_unlock_irqrestore(shost->host_lock, lock_flags);
1378         return 0;
1379 }
1380
1381 /**
1382  * ibmvscsi_change_queue_depth - Change the device's queue depth
1383  * @sdev:       scsi device struct
1384  * @qdepth:     depth to set
1385  *
1386  * Return value:
1387  *      actual depth set
1388  **/
1389 static int ibmvscsi_change_queue_depth(struct scsi_device *sdev, int qdepth)
1390 {
1391         if (qdepth > IBMVSCSI_MAX_CMDS_PER_LUN)
1392                 qdepth = IBMVSCSI_MAX_CMDS_PER_LUN;
1393
1394         scsi_adjust_queue_depth(sdev, 0, qdepth);
1395         return sdev->queue_depth;
1396 }
1397
1398 /* ------------------------------------------------------------
1399  * sysfs attributes
1400  */
1401 static ssize_t show_host_srp_version(struct class_device *class_dev, char *buf)
1402 {
1403         struct Scsi_Host *shost = class_to_shost(class_dev);
1404         struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1405         int len;
1406
1407         len = snprintf(buf, PAGE_SIZE, "%s\n",
1408                        hostdata->madapter_info.srp_version);
1409         return len;
1410 }
1411
1412 static struct class_device_attribute ibmvscsi_host_srp_version = {
1413         .attr = {
1414                  .name = "srp_version",
1415                  .mode = S_IRUGO,
1416                  },
1417         .show = show_host_srp_version,
1418 };
1419
1420 static ssize_t show_host_partition_name(struct class_device *class_dev,
1421                                         char *buf)
1422 {
1423         struct Scsi_Host *shost = class_to_shost(class_dev);
1424         struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1425         int len;
1426
1427         len = snprintf(buf, PAGE_SIZE, "%s\n",
1428                        hostdata->madapter_info.partition_name);
1429         return len;
1430 }
1431
1432 static struct class_device_attribute ibmvscsi_host_partition_name = {
1433         .attr = {
1434                  .name = "partition_name",
1435                  .mode = S_IRUGO,
1436                  },
1437         .show = show_host_partition_name,
1438 };
1439
1440 static ssize_t show_host_partition_number(struct class_device *class_dev,
1441                                           char *buf)
1442 {
1443         struct Scsi_Host *shost = class_to_shost(class_dev);
1444         struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1445         int len;
1446
1447         len = snprintf(buf, PAGE_SIZE, "%d\n",
1448                        hostdata->madapter_info.partition_number);
1449         return len;
1450 }
1451
1452 static struct class_device_attribute ibmvscsi_host_partition_number = {
1453         .attr = {
1454                  .name = "partition_number",
1455                  .mode = S_IRUGO,
1456                  },
1457         .show = show_host_partition_number,
1458 };
1459
1460 static ssize_t show_host_mad_version(struct class_device *class_dev, char *buf)
1461 {
1462         struct Scsi_Host *shost = class_to_shost(class_dev);
1463         struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1464         int len;
1465
1466         len = snprintf(buf, PAGE_SIZE, "%d\n",
1467                        hostdata->madapter_info.mad_version);
1468         return len;
1469 }
1470
1471 static struct class_device_attribute ibmvscsi_host_mad_version = {
1472         .attr = {
1473                  .name = "mad_version",
1474                  .mode = S_IRUGO,
1475                  },
1476         .show = show_host_mad_version,
1477 };
1478
1479 static ssize_t show_host_os_type(struct class_device *class_dev, char *buf)
1480 {
1481         struct Scsi_Host *shost = class_to_shost(class_dev);
1482         struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1483         int len;
1484
1485         len = snprintf(buf, PAGE_SIZE, "%d\n", hostdata->madapter_info.os_type);
1486         return len;
1487 }
1488
1489 static struct class_device_attribute ibmvscsi_host_os_type = {
1490         .attr = {
1491                  .name = "os_type",
1492                  .mode = S_IRUGO,
1493                  },
1494         .show = show_host_os_type,
1495 };
1496
1497 static ssize_t show_host_config(struct class_device *class_dev, char *buf)
1498 {
1499         struct Scsi_Host *shost = class_to_shost(class_dev);
1500         struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1501
1502         /* returns null-terminated host config data */
1503         if (ibmvscsi_do_host_config(hostdata, buf, PAGE_SIZE) == 0)
1504                 return strlen(buf);
1505         else
1506                 return 0;
1507 }
1508
1509 static struct class_device_attribute ibmvscsi_host_config = {
1510         .attr = {
1511                  .name = "config",
1512                  .mode = S_IRUGO,
1513                  },
1514         .show = show_host_config,
1515 };
1516
1517 static struct class_device_attribute *ibmvscsi_attrs[] = {
1518         &ibmvscsi_host_srp_version,
1519         &ibmvscsi_host_partition_name,
1520         &ibmvscsi_host_partition_number,
1521         &ibmvscsi_host_mad_version,
1522         &ibmvscsi_host_os_type,
1523         &ibmvscsi_host_config,
1524         NULL
1525 };
1526
1527 /* ------------------------------------------------------------
1528  * SCSI driver registration
1529  */
1530 static struct scsi_host_template driver_template = {
1531         .module = THIS_MODULE,
1532         .name = "IBM POWER Virtual SCSI Adapter " IBMVSCSI_VERSION,
1533         .proc_name = "ibmvscsi",
1534         .queuecommand = ibmvscsi_queuecommand,
1535         .eh_abort_handler = ibmvscsi_eh_abort_handler,
1536         .eh_device_reset_handler = ibmvscsi_eh_device_reset_handler,
1537         .eh_host_reset_handler = ibmvscsi_eh_host_reset_handler,
1538         .slave_configure = ibmvscsi_slave_configure,
1539         .change_queue_depth = ibmvscsi_change_queue_depth,
1540         .cmd_per_lun = 16,
1541         .can_queue = IBMVSCSI_MAX_REQUESTS_DEFAULT,
1542         .this_id = -1,
1543         .sg_tablesize = SG_ALL,
1544         .use_clustering = ENABLE_CLUSTERING,
1545         .shost_attrs = ibmvscsi_attrs,
1546 };
1547
1548 /**
1549  * Called by bus code for each adapter
1550  */
1551 static int ibmvscsi_probe(struct vio_dev *vdev, const struct vio_device_id *id)
1552 {
1553         struct ibmvscsi_host_data *hostdata;
1554         struct Scsi_Host *host;
1555         struct device *dev = &vdev->dev;
1556         unsigned long wait_switch = 0;
1557         int rc;
1558
1559         vdev->dev.driver_data = NULL;
1560
1561         driver_template.can_queue = max_requests;
1562         host = scsi_host_alloc(&driver_template, sizeof(*hostdata));
1563         if (!host) {
1564                 dev_err(&vdev->dev, "couldn't allocate host data\n");
1565                 goto scsi_host_alloc_failed;
1566         }
1567
1568         hostdata = shost_priv(host);
1569         memset(hostdata, 0x00, sizeof(*hostdata));
1570         INIT_LIST_HEAD(&hostdata->sent);
1571         hostdata->host = host;
1572         hostdata->dev = dev;
1573         atomic_set(&hostdata->request_limit, -1);
1574         hostdata->host->max_sectors = 32 * 8; /* default max I/O 32 pages */
1575
1576         rc = ibmvscsi_init_crq_queue(&hostdata->queue, hostdata, max_requests);
1577         if (rc != 0 && rc != H_RESOURCE) {
1578                 dev_err(&vdev->dev, "couldn't initialize crq. rc=%d\n", rc);
1579                 goto init_crq_failed;
1580         }
1581         if (initialize_event_pool(&hostdata->pool, max_requests, hostdata) != 0) {
1582                 dev_err(&vdev->dev, "couldn't initialize event pool\n");
1583                 goto init_pool_failed;
1584         }
1585
1586         host->max_lun = 8;
1587         host->max_id = max_id;
1588         host->max_channel = max_channel;
1589
1590         if (scsi_add_host(hostdata->host, hostdata->dev))
1591                 goto add_host_failed;
1592
1593         /* Try to send an initialization message.  Note that this is allowed
1594          * to fail if the other end is not acive.  In that case we don't
1595          * want to scan
1596          */
1597         if (ibmvscsi_send_crq(hostdata, 0xC001000000000000LL, 0) == 0
1598             || rc == H_RESOURCE) {
1599                 /*
1600                  * Wait around max init_timeout secs for the adapter to finish
1601                  * initializing. When we are done initializing, we will have a
1602                  * valid request_limit.  We don't want Linux scanning before
1603                  * we are ready.
1604                  */
1605                 for (wait_switch = jiffies + (init_timeout * HZ);
1606                      time_before(jiffies, wait_switch) &&
1607                      atomic_read(&hostdata->request_limit) < 2;) {
1608
1609                         msleep(10);
1610                 }
1611
1612                 /* if we now have a valid request_limit, initiate a scan */
1613                 if (atomic_read(&hostdata->request_limit) > 0)
1614                         scsi_scan_host(host);
1615         }
1616
1617         vdev->dev.driver_data = hostdata;
1618         return 0;
1619
1620       add_host_failed:
1621         release_event_pool(&hostdata->pool, hostdata);
1622       init_pool_failed:
1623         ibmvscsi_release_crq_queue(&hostdata->queue, hostdata, max_requests);
1624       init_crq_failed:
1625         scsi_host_put(host);
1626       scsi_host_alloc_failed:
1627         return -1;
1628 }
1629
1630 static int ibmvscsi_remove(struct vio_dev *vdev)
1631 {
1632         struct ibmvscsi_host_data *hostdata = vdev->dev.driver_data;
1633         release_event_pool(&hostdata->pool, hostdata);
1634         ibmvscsi_release_crq_queue(&hostdata->queue, hostdata,
1635                                    max_requests);
1636
1637         scsi_remove_host(hostdata->host);
1638         scsi_host_put(hostdata->host);
1639
1640         return 0;
1641 }
1642
1643 /**
1644  * ibmvscsi_device_table: Used by vio.c to match devices in the device tree we
1645  * support.
1646  */
1647 static struct vio_device_id ibmvscsi_device_table[] __devinitdata = {
1648         {"vscsi", "IBM,v-scsi"},
1649         { "", "" }
1650 };
1651 MODULE_DEVICE_TABLE(vio, ibmvscsi_device_table);
1652
1653 static struct vio_driver ibmvscsi_driver = {
1654         .id_table = ibmvscsi_device_table,
1655         .probe = ibmvscsi_probe,
1656         .remove = ibmvscsi_remove,
1657         .driver = {
1658                 .name = "ibmvscsi",
1659                 .owner = THIS_MODULE,
1660         }
1661 };
1662
1663 int __init ibmvscsi_module_init(void)
1664 {
1665         return vio_register_driver(&ibmvscsi_driver);
1666 }
1667
1668 void __exit ibmvscsi_module_exit(void)
1669 {
1670         vio_unregister_driver(&ibmvscsi_driver);
1671 }
1672
1673 module_init(ibmvscsi_module_init);
1674 module_exit(ibmvscsi_module_exit);