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-robin 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 * TODO: This is currently pretty tied to the IBM pSeries hypervisor
  59 * interfaces.  It would be really nice to abstract this above an RDMA
  60 * layer.
  61 */
  62
  63#include <linux/module.h>
  64#include <linux/moduleparam.h>
  65#include <linux/dma-mapping.h>
  66#include <linux/delay.h>
  67#include <linux/slab.h>
  68#include <linux/of.h>
  69#include <linux/pm.h>
  70#include <linux/kthread.h>
  71#include <asm/firmware.h>
  72#include <asm/vio.h>
  73#include <scsi/scsi.h>
  74#include <scsi/scsi_cmnd.h>
  75#include <scsi/scsi_host.h>
  76#include <scsi/scsi_device.h>
  77#include <scsi/scsi_transport_srp.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 */
  85static int max_id = 64;
  86static int max_channel = 3;
  87static int init_timeout = 300;
  88static int login_timeout = 60;
  89static int info_timeout = 30;
  90static int abort_timeout = 60;
  91static int reset_timeout = 60;
  92static int max_requests = IBMVSCSI_MAX_REQUESTS_DEFAULT;
  93static int max_events = IBMVSCSI_MAX_REQUESTS_DEFAULT + 2;
  94static int fast_fail = 1;
  95static int client_reserve = 1;
  96static char partition_name[97] = "UNKNOWN";
  97static unsigned int partition_number = -1;
  98static LIST_HEAD(ibmvscsi_head);
  99
 100static struct scsi_transport_template *ibmvscsi_transport_template;
 101
 102#define IBMVSCSI_VERSION "1.5.9"
 103
 104MODULE_DESCRIPTION("IBM Virtual SCSI");
 105MODULE_AUTHOR("Dave Boutcher");
 106MODULE_LICENSE("GPL");
 107MODULE_VERSION(IBMVSCSI_VERSION);
 108
 109module_param_named(max_id, max_id, int, S_IRUGO | S_IWUSR);
 110MODULE_PARM_DESC(max_id, "Largest ID value for each channel [Default=64]");
 111module_param_named(max_channel, max_channel, int, S_IRUGO | S_IWUSR);
 112MODULE_PARM_DESC(max_channel, "Largest channel value [Default=3]");
 113module_param_named(init_timeout, init_timeout, int, S_IRUGO | S_IWUSR);
 114MODULE_PARM_DESC(init_timeout, "Initialization timeout in seconds");
 115module_param_named(max_requests, max_requests, int, S_IRUGO);
 116MODULE_PARM_DESC(max_requests, "Maximum requests for this adapter");
 117module_param_named(fast_fail, fast_fail, int, S_IRUGO | S_IWUSR);
 118MODULE_PARM_DESC(fast_fail, "Enable fast fail. [Default=1]");
 119module_param_named(client_reserve, client_reserve, int, S_IRUGO );
 120MODULE_PARM_DESC(client_reserve, "Attempt client managed reserve/release");
 121
 122static void ibmvscsi_handle_crq(struct viosrp_crq *crq,
 123				struct ibmvscsi_host_data *hostdata);
 124
 125/* ------------------------------------------------------------
 126 * Routines for managing the command/response queue
 127 */
 128/**
 129 * ibmvscsi_handle_event: - Interrupt handler for crq events
 130 * @irq:	number of irq to handle, not used
 131 * @dev_instance: ibmvscsi_host_data of host that received interrupt
 132 *
 133 * Disables interrupts and schedules srp_task
 134 * Always returns IRQ_HANDLED
 135 */
 136static irqreturn_t ibmvscsi_handle_event(int irq, void *dev_instance)
 137{
 138	struct ibmvscsi_host_data *hostdata =
 139	    (struct ibmvscsi_host_data *)dev_instance;
 140	vio_disable_interrupts(to_vio_dev(hostdata->dev));
 141	tasklet_schedule(&hostdata->srp_task);
 142	return IRQ_HANDLED;
 143}
 144
 145/**
 146 * release_crq_queue: - Deallocates data and unregisters CRQ
 147 * @queue:	crq_queue to initialize and register
 148 * @host_data:	ibmvscsi_host_data of host
 149 *
 150 * Frees irq, deallocates a page for messages, unmaps dma, and unregisters
 151 * the crq with the hypervisor.
 152 */
 153static void ibmvscsi_release_crq_queue(struct crq_queue *queue,
 154				       struct ibmvscsi_host_data *hostdata,
 155				       int max_requests)
 156{
 157	long rc = 0;
 158	struct vio_dev *vdev = to_vio_dev(hostdata->dev);
 159	free_irq(vdev->irq, (void *)hostdata);
 160	tasklet_kill(&hostdata->srp_task);
 161	do {
 162		if (rc)
 163			msleep(100);
 164		rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
 165	} while ((rc == H_BUSY) || (H_IS_LONG_BUSY(rc)));
 166	dma_unmap_single(hostdata->dev,
 167			 queue->msg_token,
 168			 queue->size * sizeof(*queue->msgs), DMA_BIDIRECTIONAL);
 169	free_page((unsigned long)queue->msgs);
 170}
 171
 172/**
 173 * crq_queue_next_crq: - Returns the next entry in message queue
 174 * @queue:	crq_queue to use
 175 *
 176 * Returns pointer to next entry in queue, or NULL if there are no new
 177 * entried in the CRQ.
 178 */
 179static struct viosrp_crq *crq_queue_next_crq(struct crq_queue *queue)
 180{
 181	struct viosrp_crq *crq;
 182	unsigned long flags;
 183
 184	spin_lock_irqsave(&queue->lock, flags);
 185	crq = &queue->msgs[queue->cur];
 186	if (crq->valid != VIOSRP_CRQ_FREE) {
 187		if (++queue->cur == queue->size)
 188			queue->cur = 0;
 189
 190		/* Ensure the read of the valid bit occurs before reading any
 191		 * other bits of the CRQ entry
 192		 */
 193		rmb();
 194	} else
 195		crq = NULL;
 196	spin_unlock_irqrestore(&queue->lock, flags);
 197
 198	return crq;
 199}
 200
 201/**
 202 * ibmvscsi_send_crq: - Send a CRQ
 203 * @hostdata:	the adapter
 204 * @word1:	the first 64 bits of the data
 205 * @word2:	the second 64 bits of the data
 206 */
 207static int ibmvscsi_send_crq(struct ibmvscsi_host_data *hostdata,
 208			     u64 word1, u64 word2)
 209{
 210	struct vio_dev *vdev = to_vio_dev(hostdata->dev);
 211
 212	/*
 213	 * Ensure the command buffer is flushed to memory before handing it
 214	 * over to the VIOS to prevent it from fetching any stale data.
 215	 */
 216	mb();
 217	return plpar_hcall_norets(H_SEND_CRQ, vdev->unit_address, word1, word2);
 218}
 219
 220/**
 221 * ibmvscsi_task: - Process srps asynchronously
 222 * @data:	ibmvscsi_host_data of host
 223 */
 224static void ibmvscsi_task(void *data)
 225{
 226	struct ibmvscsi_host_data *hostdata = (struct ibmvscsi_host_data *)data;
 227	struct vio_dev *vdev = to_vio_dev(hostdata->dev);
 228	struct viosrp_crq *crq;
 229	int done = 0;
 230
 231	while (!done) {
 232		/* Pull all the valid messages off the CRQ */
 233		while ((crq = crq_queue_next_crq(&hostdata->queue)) != NULL) {
 234			ibmvscsi_handle_crq(crq, hostdata);
 235			crq->valid = VIOSRP_CRQ_FREE;
 236			wmb();
 237		}
 238
 239		vio_enable_interrupts(vdev);
 240		crq = crq_queue_next_crq(&hostdata->queue);
 241		if (crq != NULL) {
 242			vio_disable_interrupts(vdev);
 243			ibmvscsi_handle_crq(crq, hostdata);
 244			crq->valid = VIOSRP_CRQ_FREE;
 245			wmb();
 246		} else {
 247			done = 1;
 248		}
 249	}
 250}
 251
 252static void gather_partition_info(void)
 253{
 
 
 254	const char *ppartition_name;
 255	const __be32 *p_number_ptr;
 256
 257	/* Retrieve information about this partition */
 258	if (!of_root)
 
 259		return;
 
 260
 261	of_node_get(of_root);
 262
 263	ppartition_name = of_get_property(of_root, "ibm,partition-name", NULL);
 264	if (ppartition_name)
 265		strncpy(partition_name, ppartition_name,
 266				sizeof(partition_name));
 267	p_number_ptr = of_get_property(of_root, "ibm,partition-no", NULL);
 268	if (p_number_ptr)
 269		partition_number = of_read_number(p_number_ptr, 1);
 270	of_node_put(of_root);
 271}
 272
 273static void set_adapter_info(struct ibmvscsi_host_data *hostdata)
 274{
 275	memset(&hostdata->madapter_info, 0x00,
 276			sizeof(hostdata->madapter_info));
 277
 278	dev_info(hostdata->dev, "SRP_VERSION: %s\n", SRP_VERSION);
 279	strcpy(hostdata->madapter_info.srp_version, SRP_VERSION);
 280
 281	strncpy(hostdata->madapter_info.partition_name, partition_name,
 282			sizeof(hostdata->madapter_info.partition_name));
 283
 284	hostdata->madapter_info.partition_number =
 285					cpu_to_be32(partition_number);
 286
 287	hostdata->madapter_info.mad_version = cpu_to_be32(SRP_MAD_VERSION_1);
 288	hostdata->madapter_info.os_type = cpu_to_be32(SRP_MAD_OS_LINUX);
 289}
 290
 291/**
 292 * reset_crq_queue: - resets a crq after a failure
 293 * @queue:	crq_queue to initialize and register
 294 * @hostdata:	ibmvscsi_host_data of host
 295 *
 296 */
 297static int ibmvscsi_reset_crq_queue(struct crq_queue *queue,
 298				    struct ibmvscsi_host_data *hostdata)
 299{
 300	int rc = 0;
 301	struct vio_dev *vdev = to_vio_dev(hostdata->dev);
 302
 303	/* Close the CRQ */
 304	do {
 305		if (rc)
 306			msleep(100);
 307		rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
 308	} while ((rc == H_BUSY) || (H_IS_LONG_BUSY(rc)));
 309
 310	/* Clean out the queue */
 311	memset(queue->msgs, 0x00, PAGE_SIZE);
 312	queue->cur = 0;
 313
 314	set_adapter_info(hostdata);
 315
 316	/* And re-open it again */
 317	rc = plpar_hcall_norets(H_REG_CRQ,
 318				vdev->unit_address,
 319				queue->msg_token, PAGE_SIZE);
 320	if (rc == H_CLOSED) {
 321		/* Adapter is good, but other end is not ready */
 322		dev_warn(hostdata->dev, "Partner adapter not ready\n");
 323	} else if (rc != 0) {
 324		dev_warn(hostdata->dev, "couldn't register crq--rc 0x%x\n", rc);
 325	}
 326	return rc;
 327}
 328
 329/**
 330 * initialize_crq_queue: - Initializes and registers CRQ with hypervisor
 331 * @queue:	crq_queue to initialize and register
 332 * @hostdata:	ibmvscsi_host_data of host
 333 *
 334 * Allocates a page for messages, maps it for dma, and registers
 335 * the crq with the hypervisor.
 336 * Returns zero on success.
 337 */
 338static int ibmvscsi_init_crq_queue(struct crq_queue *queue,
 339				   struct ibmvscsi_host_data *hostdata,
 340				   int max_requests)
 341{
 342	int rc;
 343	int retrc;
 344	struct vio_dev *vdev = to_vio_dev(hostdata->dev);
 345
 346	queue->msgs = (struct viosrp_crq *)get_zeroed_page(GFP_KERNEL);
 347
 348	if (!queue->msgs)
 349		goto malloc_failed;
 350	queue->size = PAGE_SIZE / sizeof(*queue->msgs);
 351
 352	queue->msg_token = dma_map_single(hostdata->dev, queue->msgs,
 353					  queue->size * sizeof(*queue->msgs),
 354					  DMA_BIDIRECTIONAL);
 355
 356	if (dma_mapping_error(hostdata->dev, queue->msg_token))
 357		goto map_failed;
 358
 359	gather_partition_info();
 360	set_adapter_info(hostdata);
 361
 362	retrc = rc = plpar_hcall_norets(H_REG_CRQ,
 363				vdev->unit_address,
 364				queue->msg_token, PAGE_SIZE);
 365	if (rc == H_RESOURCE)
 366		/* maybe kexecing and resource is busy. try a reset */
 367		rc = ibmvscsi_reset_crq_queue(queue,
 368					      hostdata);
 369
 370	if (rc == H_CLOSED) {
 371		/* Adapter is good, but other end is not ready */
 372		dev_warn(hostdata->dev, "Partner adapter not ready\n");
 373		retrc = 0;
 374	} else if (rc != 0) {
 375		dev_warn(hostdata->dev, "Error %d opening adapter\n", rc);
 376		goto reg_crq_failed;
 377	}
 378
 379	queue->cur = 0;
 380	spin_lock_init(&queue->lock);
 381
 382	tasklet_init(&hostdata->srp_task, (void *)ibmvscsi_task,
 383		     (unsigned long)hostdata);
 384
 385	if (request_irq(vdev->irq,
 386			ibmvscsi_handle_event,
 387			0, "ibmvscsi", (void *)hostdata) != 0) {
 388		dev_err(hostdata->dev, "couldn't register irq 0x%x\n",
 389			vdev->irq);
 390		goto req_irq_failed;
 391	}
 392
 393	rc = vio_enable_interrupts(vdev);
 394	if (rc != 0) {
 395		dev_err(hostdata->dev, "Error %d enabling interrupts!!!\n", rc);
 396		goto req_irq_failed;
 397	}
 398
 399	return retrc;
 400
 401      req_irq_failed:
 402	tasklet_kill(&hostdata->srp_task);
 403	rc = 0;
 404	do {
 405		if (rc)
 406			msleep(100);
 407		rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
 408	} while ((rc == H_BUSY) || (H_IS_LONG_BUSY(rc)));
 409      reg_crq_failed:
 410	dma_unmap_single(hostdata->dev,
 411			 queue->msg_token,
 412			 queue->size * sizeof(*queue->msgs), DMA_BIDIRECTIONAL);
 413      map_failed:
 414	free_page((unsigned long)queue->msgs);
 415      malloc_failed:
 416	return -1;
 417}
 418
 419/**
 420 * reenable_crq_queue: - reenables a crq after
 421 * @queue:	crq_queue to initialize and register
 422 * @hostdata:	ibmvscsi_host_data of host
 423 *
 424 */
 425static int ibmvscsi_reenable_crq_queue(struct crq_queue *queue,
 426				       struct ibmvscsi_host_data *hostdata)
 427{
 428	int rc = 0;
 429	struct vio_dev *vdev = to_vio_dev(hostdata->dev);
 430
 431	/* Re-enable the CRQ */
 432	do {
 433		if (rc)
 434			msleep(100);
 435		rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address);
 436	} while ((rc == H_IN_PROGRESS) || (rc == H_BUSY) || (H_IS_LONG_BUSY(rc)));
 437
 438	if (rc)
 439		dev_err(hostdata->dev, "Error %d enabling adapter\n", rc);
 440	return rc;
 441}
 442
 443/* ------------------------------------------------------------
 444 * Routines for the event pool and event structs
 445 */
 446/**
 447 * initialize_event_pool: - Allocates and initializes the event pool for a host
 448 * @pool:	event_pool to be initialized
 449 * @size:	Number of events in pool
 450 * @hostdata:	ibmvscsi_host_data who owns the event pool
 451 *
 452 * Returns zero on success.
 453*/
 454static int initialize_event_pool(struct event_pool *pool,
 455				 int size, struct ibmvscsi_host_data *hostdata)
 456{
 457	int i;
 458
 459	pool->size = size;
 460	pool->next = 0;
 461	pool->events = kcalloc(pool->size, sizeof(*pool->events), GFP_KERNEL);
 462	if (!pool->events)
 463		return -ENOMEM;
 464
 465	pool->iu_storage =
 466	    dma_alloc_coherent(hostdata->dev,
 467			       pool->size * sizeof(*pool->iu_storage),
 468			       &pool->iu_token, 0);
 469	if (!pool->iu_storage) {
 470		kfree(pool->events);
 471		return -ENOMEM;
 472	}
 473
 474	for (i = 0; i < pool->size; ++i) {
 475		struct srp_event_struct *evt = &pool->events[i];
 476		memset(&evt->crq, 0x00, sizeof(evt->crq));
 477		atomic_set(&evt->free, 1);
 478		evt->crq.valid = VIOSRP_CRQ_CMD_RSP;
 479		evt->crq.IU_length = cpu_to_be16(sizeof(*evt->xfer_iu));
 480		evt->crq.IU_data_ptr = cpu_to_be64(pool->iu_token +
 481			sizeof(*evt->xfer_iu) * i);
 482		evt->xfer_iu = pool->iu_storage + i;
 483		evt->hostdata = hostdata;
 484		evt->ext_list = NULL;
 485		evt->ext_list_token = 0;
 486	}
 487
 488	return 0;
 489}
 490
 491/**
 492 * release_event_pool: - Frees memory of an event pool of a host
 493 * @pool:	event_pool to be released
 494 * @hostdata:	ibmvscsi_host_data who owns the even pool
 495 *
 496 * Returns zero on success.
 497*/
 498static void release_event_pool(struct event_pool *pool,
 499			       struct ibmvscsi_host_data *hostdata)
 500{
 501	int i, in_use = 0;
 502	for (i = 0; i < pool->size; ++i) {
 503		if (atomic_read(&pool->events[i].free) != 1)
 504			++in_use;
 505		if (pool->events[i].ext_list) {
 506			dma_free_coherent(hostdata->dev,
 507				  SG_ALL * sizeof(struct srp_direct_buf),
 508				  pool->events[i].ext_list,
 509				  pool->events[i].ext_list_token);
 510		}
 511	}
 512	if (in_use)
 513		dev_warn(hostdata->dev, "releasing event pool with %d "
 514			 "events still in use?\n", in_use);
 515	kfree(pool->events);
 516	dma_free_coherent(hostdata->dev,
 517			  pool->size * sizeof(*pool->iu_storage),
 518			  pool->iu_storage, pool->iu_token);
 519}
 520
 521/**
 522 * valid_event_struct: - Determines if event is valid.
 523 * @pool:	event_pool that contains the event
 524 * @evt:	srp_event_struct to be checked for validity
 525 *
 526 * Returns zero if event is invalid, one otherwise.
 527*/
 528static int valid_event_struct(struct event_pool *pool,
 529				struct srp_event_struct *evt)
 530{
 531	int index = evt - pool->events;
 532	if (index < 0 || index >= pool->size)	/* outside of bounds */
 533		return 0;
 534	if (evt != pool->events + index)	/* unaligned */
 535		return 0;
 536	return 1;
 537}
 538
 539/**
 540 * ibmvscsi_free-event_struct: - Changes status of event to "free"
 541 * @pool:	event_pool that contains the event
 542 * @evt:	srp_event_struct to be modified
 543 *
 544*/
 545static void free_event_struct(struct event_pool *pool,
 546				       struct srp_event_struct *evt)
 547{
 548	if (!valid_event_struct(pool, evt)) {
 549		dev_err(evt->hostdata->dev, "Freeing invalid event_struct %p "
 550			"(not in pool %p)\n", evt, pool->events);
 551		return;
 552	}
 553	if (atomic_inc_return(&evt->free) != 1) {
 554		dev_err(evt->hostdata->dev, "Freeing event_struct %p "
 555			"which is not in use!\n", evt);
 556		return;
 557	}
 558}
 559
 560/**
 561 * get_evt_struct: - Gets the next free event in pool
 562 * @pool:	event_pool that contains the events to be searched
 563 *
 564 * Returns the next event in "free" state, and NULL if none are free.
 565 * Note that no synchronization is done here, we assume the host_lock
 566 * will syncrhonze things.
 567*/
 568static struct srp_event_struct *get_event_struct(struct event_pool *pool)
 569{
 570	int i;
 571	int poolsize = pool->size;
 572	int offset = pool->next;
 573
 574	for (i = 0; i < poolsize; i++) {
 575		offset = (offset + 1) % poolsize;
 576		if (!atomic_dec_if_positive(&pool->events[offset].free)) {
 577			pool->next = offset;
 578			return &pool->events[offset];
 579		}
 580	}
 581
 582	printk(KERN_ERR "ibmvscsi: found no event struct in pool!\n");
 583	return NULL;
 584}
 585
 586/**
 587 * init_event_struct: Initialize fields in an event struct that are always 
 588 *                    required.
 589 * @evt:        The event
 590 * @done:       Routine to call when the event is responded to
 591 * @format:     SRP or MAD format
 592 * @timeout:    timeout value set in the CRQ
 593 */
 594static void init_event_struct(struct srp_event_struct *evt_struct,
 595			      void (*done) (struct srp_event_struct *),
 596			      u8 format,
 597			      int timeout)
 598{
 599	evt_struct->cmnd = NULL;
 600	evt_struct->cmnd_done = NULL;
 601	evt_struct->sync_srp = NULL;
 602	evt_struct->crq.format = format;
 603	evt_struct->crq.timeout = cpu_to_be16(timeout);
 604	evt_struct->done = done;
 605}
 606
 607/* ------------------------------------------------------------
 608 * Routines for receiving SCSI responses from the hosting partition
 609 */
 610
 611/**
 612 * set_srp_direction: Set the fields in the srp related to data
 613 *     direction and number of buffers based on the direction in
 614 *     the scsi_cmnd and the number of buffers
 615 */
 616static void set_srp_direction(struct scsi_cmnd *cmd,
 617			      struct srp_cmd *srp_cmd, 
 618			      int numbuf)
 619{
 620	u8 fmt;
 621
 622	if (numbuf == 0)
 623		return;
 624	
 625	if (numbuf == 1)
 626		fmt = SRP_DATA_DESC_DIRECT;
 627	else {
 628		fmt = SRP_DATA_DESC_INDIRECT;
 629		numbuf = min(numbuf, MAX_INDIRECT_BUFS);
 630
 631		if (cmd->sc_data_direction == DMA_TO_DEVICE)
 632			srp_cmd->data_out_desc_cnt = numbuf;
 633		else
 634			srp_cmd->data_in_desc_cnt = numbuf;
 635	}
 636
 637	if (cmd->sc_data_direction == DMA_TO_DEVICE)
 638		srp_cmd->buf_fmt = fmt << 4;
 639	else
 640		srp_cmd->buf_fmt = fmt;
 641}
 642
 643/**
 644 * unmap_cmd_data: - Unmap data pointed in srp_cmd based on the format
 645 * @cmd:	srp_cmd whose additional_data member will be unmapped
 646 * @dev:	device for which the memory is mapped
 647 *
 648*/
 649static void unmap_cmd_data(struct srp_cmd *cmd,
 650			   struct srp_event_struct *evt_struct,
 651			   struct device *dev)
 652{
 653	u8 out_fmt, in_fmt;
 654
 655	out_fmt = cmd->buf_fmt >> 4;
 656	in_fmt = cmd->buf_fmt & ((1U << 4) - 1);
 657
 658	if (out_fmt == SRP_NO_DATA_DESC && in_fmt == SRP_NO_DATA_DESC)
 659		return;
 660
 661	if (evt_struct->cmnd)
 662		scsi_dma_unmap(evt_struct->cmnd);
 663}
 664
 665static int map_sg_list(struct scsi_cmnd *cmd, int nseg,
 666		       struct srp_direct_buf *md)
 667{
 668	int i;
 669	struct scatterlist *sg;
 670	u64 total_length = 0;
 671
 672	scsi_for_each_sg(cmd, sg, nseg, i) {
 673		struct srp_direct_buf *descr = md + i;
 674		descr->va = cpu_to_be64(sg_dma_address(sg));
 675		descr->len = cpu_to_be32(sg_dma_len(sg));
 676		descr->key = 0;
 677		total_length += sg_dma_len(sg);
 678 	}
 679	return total_length;
 680}
 681
 682/**
 683 * map_sg_data: - Maps dma for a scatterlist and initializes decriptor fields
 684 * @cmd:	Scsi_Cmnd with the scatterlist
 685 * @srp_cmd:	srp_cmd that contains the memory descriptor
 686 * @dev:	device for which to map dma memory
 687 *
 688 * Called by map_data_for_srp_cmd() when building srp cmd from scsi cmd.
 689 * Returns 1 on success.
 690*/
 691static int map_sg_data(struct scsi_cmnd *cmd,
 692		       struct srp_event_struct *evt_struct,
 693		       struct srp_cmd *srp_cmd, struct device *dev)
 694{
 695
 696	int sg_mapped;
 697	u64 total_length = 0;
 698	struct srp_direct_buf *data =
 699		(struct srp_direct_buf *) srp_cmd->add_data;
 700	struct srp_indirect_buf *indirect =
 701		(struct srp_indirect_buf *) data;
 702
 703	sg_mapped = scsi_dma_map(cmd);
 704	if (!sg_mapped)
 705		return 1;
 706	else if (sg_mapped < 0)
 707		return 0;
 708
 709	set_srp_direction(cmd, srp_cmd, sg_mapped);
 710
 711	/* special case; we can use a single direct descriptor */
 712	if (sg_mapped == 1) {
 713		map_sg_list(cmd, sg_mapped, data);
 714		return 1;
 715	}
 716
 717	indirect->table_desc.va = 0;
 718	indirect->table_desc.len = cpu_to_be32(sg_mapped *
 719					       sizeof(struct srp_direct_buf));
 720	indirect->table_desc.key = 0;
 721
 722	if (sg_mapped <= MAX_INDIRECT_BUFS) {
 723		total_length = map_sg_list(cmd, sg_mapped,
 724					   &indirect->desc_list[0]);
 725		indirect->len = cpu_to_be32(total_length);
 726		return 1;
 727	}
 728
 729	/* get indirect table */
 730	if (!evt_struct->ext_list) {
 731		evt_struct->ext_list = (struct srp_direct_buf *)
 732			dma_alloc_coherent(dev,
 733					   SG_ALL * sizeof(struct srp_direct_buf),
 734					   &evt_struct->ext_list_token, 0);
 735		if (!evt_struct->ext_list) {
 736			if (!firmware_has_feature(FW_FEATURE_CMO))
 737				sdev_printk(KERN_ERR, cmd->device,
 738				            "Can't allocate memory "
 739				            "for indirect table\n");
 740			scsi_dma_unmap(cmd);
 741			return 0;
 742		}
 743	}
 744
 745	total_length = map_sg_list(cmd, sg_mapped, evt_struct->ext_list);
 746
 747	indirect->len = cpu_to_be32(total_length);
 748	indirect->table_desc.va = cpu_to_be64(evt_struct->ext_list_token);
 749	indirect->table_desc.len = cpu_to_be32(sg_mapped *
 750					       sizeof(indirect->desc_list[0]));
 751	memcpy(indirect->desc_list, evt_struct->ext_list,
 752	       MAX_INDIRECT_BUFS * sizeof(struct srp_direct_buf));
 753 	return 1;
 754}
 755
 756/**
 757 * map_data_for_srp_cmd: - Calls functions to map data for srp cmds
 758 * @cmd:	struct scsi_cmnd with the memory to be mapped
 759 * @srp_cmd:	srp_cmd that contains the memory descriptor
 760 * @dev:	dma device for which to map dma memory
 761 *
 762 * Called by scsi_cmd_to_srp_cmd() when converting scsi cmds to srp cmds 
 763 * Returns 1 on success.
 764*/
 765static int map_data_for_srp_cmd(struct scsi_cmnd *cmd,
 766				struct srp_event_struct *evt_struct,
 767				struct srp_cmd *srp_cmd, struct device *dev)
 768{
 769	switch (cmd->sc_data_direction) {
 770	case DMA_FROM_DEVICE:
 771	case DMA_TO_DEVICE:
 772		break;
 773	case DMA_NONE:
 774		return 1;
 775	case DMA_BIDIRECTIONAL:
 776		sdev_printk(KERN_ERR, cmd->device,
 777			    "Can't map DMA_BIDIRECTIONAL to read/write\n");
 778		return 0;
 779	default:
 780		sdev_printk(KERN_ERR, cmd->device,
 781			    "Unknown data direction 0x%02x; can't map!\n",
 782			    cmd->sc_data_direction);
 783		return 0;
 784	}
 785
 786	return map_sg_data(cmd, evt_struct, srp_cmd, dev);
 787}
 788
 789/**
 790 * purge_requests: Our virtual adapter just shut down.  purge any sent requests
 791 * @hostdata:    the adapter
 792 */
 793static void purge_requests(struct ibmvscsi_host_data *hostdata, int error_code)
 794{
 795	struct srp_event_struct *evt;
 796	unsigned long flags;
 797
 798	spin_lock_irqsave(hostdata->host->host_lock, flags);
 799	while (!list_empty(&hostdata->sent)) {
 800		evt = list_first_entry(&hostdata->sent, struct srp_event_struct, list);
 801		list_del(&evt->list);
 802		del_timer(&evt->timer);
 803
 804		spin_unlock_irqrestore(hostdata->host->host_lock, flags);
 805		if (evt->cmnd) {
 806			evt->cmnd->result = (error_code << 16);
 807			unmap_cmd_data(&evt->iu.srp.cmd, evt,
 808				       evt->hostdata->dev);
 809			if (evt->cmnd_done)
 810				evt->cmnd_done(evt->cmnd);
 811		} else if (evt->done && evt->crq.format != VIOSRP_MAD_FORMAT &&
 812			   evt->iu.srp.login_req.opcode != SRP_LOGIN_REQ)
 813			evt->done(evt);
 814		free_event_struct(&evt->hostdata->pool, evt);
 815		spin_lock_irqsave(hostdata->host->host_lock, flags);
 816	}
 817	spin_unlock_irqrestore(hostdata->host->host_lock, flags);
 818}
 819
 820/**
 821 * ibmvscsi_reset_host - Reset the connection to the server
 822 * @hostdata:	struct ibmvscsi_host_data to reset
 823*/
 824static void ibmvscsi_reset_host(struct ibmvscsi_host_data *hostdata)
 825{
 826	scsi_block_requests(hostdata->host);
 827	atomic_set(&hostdata->request_limit, 0);
 828
 829	purge_requests(hostdata, DID_ERROR);
 830	hostdata->reset_crq = 1;
 831	wake_up(&hostdata->work_wait_q);
 832}
 833
 834/**
 835 * ibmvscsi_timeout - Internal command timeout handler
 836 * @evt_struct:	struct srp_event_struct that timed out
 837 *
 838 * Called when an internally generated command times out
 839*/
 840static void ibmvscsi_timeout(struct srp_event_struct *evt_struct)
 841{
 842	struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
 843
 844	dev_err(hostdata->dev, "Command timed out (%x). Resetting connection\n",
 845		evt_struct->iu.srp.cmd.opcode);
 846
 847	ibmvscsi_reset_host(hostdata);
 848}
 849
 850
 851/* ------------------------------------------------------------
 852 * Routines for sending and receiving SRPs
 853 */
 854/**
 855 * ibmvscsi_send_srp_event: - Transforms event to u64 array and calls send_crq()
 856 * @evt_struct:	evt_struct to be sent
 857 * @hostdata:	ibmvscsi_host_data of host
 858 * @timeout:	timeout in seconds - 0 means do not time command
 859 *
 860 * Returns the value returned from ibmvscsi_send_crq(). (Zero for success)
 861 * Note that this routine assumes that host_lock is held for synchronization
 862*/
 863static int ibmvscsi_send_srp_event(struct srp_event_struct *evt_struct,
 864				   struct ibmvscsi_host_data *hostdata,
 865				   unsigned long timeout)
 866{
 867	__be64 *crq_as_u64 = (__be64 *)&evt_struct->crq;
 868	int request_status = 0;
 869	int rc;
 870	int srp_req = 0;
 871
 872	/* If we have exhausted our request limit, just fail this request,
 873	 * unless it is for a reset or abort.
 874	 * Note that there are rare cases involving driver generated requests 
 875	 * (such as task management requests) that the mid layer may think we
 876	 * can handle more requests (can_queue) when we actually can't
 877	 */
 878	if (evt_struct->crq.format == VIOSRP_SRP_FORMAT) {
 879		srp_req = 1;
 880		request_status =
 881			atomic_dec_if_positive(&hostdata->request_limit);
 882		/* If request limit was -1 when we started, it is now even
 883		 * less than that
 884		 */
 885		if (request_status < -1)
 886			goto send_error;
 887		/* Otherwise, we may have run out of requests. */
 888		/* If request limit was 0 when we started the adapter is in the
 889		 * process of performing a login with the server adapter, or
 890		 * we may have run out of requests.
 891		 */
 892		else if (request_status == -1 &&
 893		         evt_struct->iu.srp.login_req.opcode != SRP_LOGIN_REQ)
 894			goto send_busy;
 895		/* Abort and reset calls should make it through.
 896		 * Nothing except abort and reset should use the last two
 897		 * slots unless we had two or less to begin with.
 898		 */
 899		else if (request_status < 2 &&
 900		         evt_struct->iu.srp.cmd.opcode != SRP_TSK_MGMT) {
 901			/* In the case that we have less than two requests
 902			 * available, check the server limit as a combination
 903			 * of the request limit and the number of requests
 904			 * in-flight (the size of the send list).  If the
 905			 * server limit is greater than 2, return busy so
 906			 * that the last two are reserved for reset and abort.
 907			 */
 908			int server_limit = request_status;
 909			struct srp_event_struct *tmp_evt;
 910
 911			list_for_each_entry(tmp_evt, &hostdata->sent, list) {
 912				server_limit++;
 913			}
 914
 915			if (server_limit > 2)
 916				goto send_busy;
 917		}
 918	}
 919
 920	/* Copy the IU into the transfer area */
 921	*evt_struct->xfer_iu = evt_struct->iu;
 922	evt_struct->xfer_iu->srp.rsp.tag = (u64)evt_struct;
 923
 924	/* Add this to the sent list.  We need to do this 
 925	 * before we actually send 
 926	 * in case it comes back REALLY fast
 927	 */
 928	list_add_tail(&evt_struct->list, &hostdata->sent);
 929
 930	init_timer(&evt_struct->timer);
 931	if (timeout) {
 932		evt_struct->timer.data = (unsigned long) evt_struct;
 933		evt_struct->timer.expires = jiffies + (timeout * HZ);
 934		evt_struct->timer.function = (void (*)(unsigned long))ibmvscsi_timeout;
 935		add_timer(&evt_struct->timer);
 936	}
 937
 938	rc = ibmvscsi_send_crq(hostdata, be64_to_cpu(crq_as_u64[0]),
 939			       be64_to_cpu(crq_as_u64[1]));
 940	if (rc != 0) {
 941		list_del(&evt_struct->list);
 942		del_timer(&evt_struct->timer);
 943
 944		/* If send_crq returns H_CLOSED, return SCSI_MLQUEUE_HOST_BUSY.
 945		 * Firmware will send a CRQ with a transport event (0xFF) to
 946		 * tell this client what has happened to the transport.  This
 947		 * will be handled in ibmvscsi_handle_crq()
 948		 */
 949		if (rc == H_CLOSED) {
 950			dev_warn(hostdata->dev, "send warning. "
 951			         "Receive queue closed, will retry.\n");
 952			goto send_busy;
 953		}
 954		dev_err(hostdata->dev, "send error %d\n", rc);
 955		if (srp_req)
 956			atomic_inc(&hostdata->request_limit);
 957		goto send_error;
 958	}
 959
 960	return 0;
 961
 962 send_busy:
 963	unmap_cmd_data(&evt_struct->iu.srp.cmd, evt_struct, hostdata->dev);
 964
 965	free_event_struct(&hostdata->pool, evt_struct);
 966	if (srp_req && request_status != -1)
 967		atomic_inc(&hostdata->request_limit);
 968	return SCSI_MLQUEUE_HOST_BUSY;
 969
 970 send_error:
 971	unmap_cmd_data(&evt_struct->iu.srp.cmd, evt_struct, hostdata->dev);
 972
 973	if (evt_struct->cmnd != NULL) {
 974		evt_struct->cmnd->result = DID_ERROR << 16;
 975		evt_struct->cmnd_done(evt_struct->cmnd);
 976	} else if (evt_struct->done)
 977		evt_struct->done(evt_struct);
 978
 979	free_event_struct(&hostdata->pool, evt_struct);
 980	return 0;
 981}
 982
 983/**
 984 * handle_cmd_rsp: -  Handle responses from commands
 985 * @evt_struct:	srp_event_struct to be handled
 986 *
 987 * Used as a callback by when sending scsi cmds.
 988 * Gets called by ibmvscsi_handle_crq()
 989*/
 990static void handle_cmd_rsp(struct srp_event_struct *evt_struct)
 991{
 992	struct srp_rsp *rsp = &evt_struct->xfer_iu->srp.rsp;
 993	struct scsi_cmnd *cmnd = evt_struct->cmnd;
 994
 995	if (unlikely(rsp->opcode != SRP_RSP)) {
 996		if (printk_ratelimit())
 997			dev_warn(evt_struct->hostdata->dev,
 998				 "bad SRP RSP type %#02x\n", rsp->opcode);
 999	}
1000	
1001	if (cmnd) {
1002		cmnd->result |= rsp->status;
1003		if (((cmnd->result >> 1) & 0x1f) == CHECK_CONDITION)
1004			memcpy(cmnd->sense_buffer,
1005			       rsp->data,
1006			       be32_to_cpu(rsp->sense_data_len));
1007		unmap_cmd_data(&evt_struct->iu.srp.cmd, 
1008			       evt_struct, 
1009			       evt_struct->hostdata->dev);
1010
1011		if (rsp->flags & SRP_RSP_FLAG_DOOVER)
1012			scsi_set_resid(cmnd,
1013				       be32_to_cpu(rsp->data_out_res_cnt));
1014		else if (rsp->flags & SRP_RSP_FLAG_DIOVER)
1015			scsi_set_resid(cmnd, be32_to_cpu(rsp->data_in_res_cnt));
1016	}
1017
1018	if (evt_struct->cmnd_done)
1019		evt_struct->cmnd_done(cmnd);
1020}
1021
1022/**
1023 * lun_from_dev: - Returns the lun of the scsi device
1024 * @dev:	struct scsi_device
1025 *
1026*/
1027static inline u16 lun_from_dev(struct scsi_device *dev)
1028{
1029	return (0x2 << 14) | (dev->id << 8) | (dev->channel << 5) | dev->lun;
1030}
1031
1032/**
1033 * ibmvscsi_queue: - The queuecommand function of the scsi template 
1034 * @cmd:	struct scsi_cmnd to be executed
1035 * @done:	Callback function to be called when cmd is completed
1036*/
1037static int ibmvscsi_queuecommand_lck(struct scsi_cmnd *cmnd,
1038				 void (*done) (struct scsi_cmnd *))
1039{
1040	struct srp_cmd *srp_cmd;
1041	struct srp_event_struct *evt_struct;
1042	struct srp_indirect_buf *indirect;
1043	struct ibmvscsi_host_data *hostdata = shost_priv(cmnd->device->host);
1044	u16 lun = lun_from_dev(cmnd->device);
1045	u8 out_fmt, in_fmt;
1046
1047	cmnd->result = (DID_OK << 16);
1048	evt_struct = get_event_struct(&hostdata->pool);
1049	if (!evt_struct)
1050		return SCSI_MLQUEUE_HOST_BUSY;
1051
1052	/* Set up the actual SRP IU */
1053	srp_cmd = &evt_struct->iu.srp.cmd;
1054	memset(srp_cmd, 0x00, SRP_MAX_IU_LEN);
1055	srp_cmd->opcode = SRP_CMD;
1056	memcpy(srp_cmd->cdb, cmnd->cmnd, sizeof(srp_cmd->cdb));
1057	int_to_scsilun(lun, &srp_cmd->lun);
1058
1059	if (!map_data_for_srp_cmd(cmnd, evt_struct, srp_cmd, hostdata->dev)) {
1060		if (!firmware_has_feature(FW_FEATURE_CMO))
1061			sdev_printk(KERN_ERR, cmnd->device,
1062			            "couldn't convert cmd to srp_cmd\n");
1063		free_event_struct(&hostdata->pool, evt_struct);
1064		return SCSI_MLQUEUE_HOST_BUSY;
1065	}
1066
1067	init_event_struct(evt_struct,
1068			  handle_cmd_rsp,
1069			  VIOSRP_SRP_FORMAT,
1070			  cmnd->request->timeout/HZ);
1071
1072	evt_struct->cmnd = cmnd;
1073	evt_struct->cmnd_done = done;
1074
1075	/* Fix up dma address of the buffer itself */
1076	indirect = (struct srp_indirect_buf *) srp_cmd->add_data;
1077	out_fmt = srp_cmd->buf_fmt >> 4;
1078	in_fmt = srp_cmd->buf_fmt & ((1U << 4) - 1);
1079	if ((in_fmt == SRP_DATA_DESC_INDIRECT ||
1080	     out_fmt == SRP_DATA_DESC_INDIRECT) &&
1081	    indirect->table_desc.va == 0) {
1082		indirect->table_desc.va =
1083			cpu_to_be64(be64_to_cpu(evt_struct->crq.IU_data_ptr) +
1084			offsetof(struct srp_cmd, add_data) +
1085			offsetof(struct srp_indirect_buf, desc_list));
1086	}
1087
1088	return ibmvscsi_send_srp_event(evt_struct, hostdata, 0);
1089}
1090
1091static DEF_SCSI_QCMD(ibmvscsi_queuecommand)
1092
1093/* ------------------------------------------------------------
1094 * Routines for driver initialization
1095 */
1096
1097/**
1098 * map_persist_bufs: - Pre-map persistent data for adapter logins
1099 * @hostdata:   ibmvscsi_host_data of host
1100 *
1101 * Map the capabilities and adapter info DMA buffers to avoid runtime failures.
1102 * Return 1 on error, 0 on success.
1103 */
1104static int map_persist_bufs(struct ibmvscsi_host_data *hostdata)
1105{
1106
1107	hostdata->caps_addr = dma_map_single(hostdata->dev, &hostdata->caps,
1108					     sizeof(hostdata->caps), DMA_BIDIRECTIONAL);
1109
1110	if (dma_mapping_error(hostdata->dev, hostdata->caps_addr)) {
1111		dev_err(hostdata->dev, "Unable to map capabilities buffer!\n");
1112		return 1;
1113	}
1114
1115	hostdata->adapter_info_addr = dma_map_single(hostdata->dev,
1116						     &hostdata->madapter_info,
1117						     sizeof(hostdata->madapter_info),
1118						     DMA_BIDIRECTIONAL);
1119	if (dma_mapping_error(hostdata->dev, hostdata->adapter_info_addr)) {
1120		dev_err(hostdata->dev, "Unable to map adapter info buffer!\n");
1121		dma_unmap_single(hostdata->dev, hostdata->caps_addr,
1122				 sizeof(hostdata->caps), DMA_BIDIRECTIONAL);
1123		return 1;
1124	}
1125
1126	return 0;
1127}
1128
1129/**
1130 * unmap_persist_bufs: - Unmap persistent data needed for adapter logins
1131 * @hostdata:   ibmvscsi_host_data of host
1132 *
1133 * Unmap the capabilities and adapter info DMA buffers
1134 */
1135static void unmap_persist_bufs(struct ibmvscsi_host_data *hostdata)
1136{
1137	dma_unmap_single(hostdata->dev, hostdata->caps_addr,
1138			 sizeof(hostdata->caps), DMA_BIDIRECTIONAL);
1139
1140	dma_unmap_single(hostdata->dev, hostdata->adapter_info_addr,
1141			 sizeof(hostdata->madapter_info), DMA_BIDIRECTIONAL);
1142}
1143
1144/**
1145 * login_rsp: - Handle response to SRP login request
1146 * @evt_struct:	srp_event_struct with the response
1147 *
1148 * Used as a "done" callback by when sending srp_login. Gets called
1149 * by ibmvscsi_handle_crq()
1150*/
1151static void login_rsp(struct srp_event_struct *evt_struct)
1152{
1153	struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
1154	switch (evt_struct->xfer_iu->srp.login_rsp.opcode) {
1155	case SRP_LOGIN_RSP:	/* it worked! */
1156		break;
1157	case SRP_LOGIN_REJ:	/* refused! */
1158		dev_info(hostdata->dev, "SRP_LOGIN_REJ reason %u\n",
1159			 evt_struct->xfer_iu->srp.login_rej.reason);
1160		/* Login failed.  */
1161		atomic_set(&hostdata->request_limit, -1);
1162		return;
1163	default:
1164		dev_err(hostdata->dev, "Invalid login response typecode 0x%02x!\n",
1165			evt_struct->xfer_iu->srp.login_rsp.opcode);
1166		/* Login failed.  */
1167		atomic_set(&hostdata->request_limit, -1);
1168		return;
1169	}
1170
1171	dev_info(hostdata->dev, "SRP_LOGIN succeeded\n");
1172	hostdata->client_migrated = 0;
1173
1174	/* Now we know what the real request-limit is.
1175	 * This value is set rather than added to request_limit because
1176	 * request_limit could have been set to -1 by this client.
1177	 */
1178	atomic_set(&hostdata->request_limit,
1179		   be32_to_cpu(evt_struct->xfer_iu->srp.login_rsp.req_lim_delta));
1180
1181	/* If we had any pending I/Os, kick them */
1182	scsi_unblock_requests(hostdata->host);
1183}
1184
1185/**
1186 * send_srp_login: - Sends the srp login
1187 * @hostdata:	ibmvscsi_host_data of host
1188 *
1189 * Returns zero if successful.
1190*/
1191static int send_srp_login(struct ibmvscsi_host_data *hostdata)
1192{
1193	int rc;
1194	unsigned long flags;
1195	struct srp_login_req *login;
1196	struct srp_event_struct *evt_struct = get_event_struct(&hostdata->pool);
1197
1198	BUG_ON(!evt_struct);
1199	init_event_struct(evt_struct, login_rsp,
1200			  VIOSRP_SRP_FORMAT, login_timeout);
1201
1202	login = &evt_struct->iu.srp.login_req;
1203	memset(login, 0, sizeof(*login));
1204	login->opcode = SRP_LOGIN_REQ;
1205	login->req_it_iu_len = cpu_to_be32(sizeof(union srp_iu));
1206	login->req_buf_fmt = cpu_to_be16(SRP_BUF_FORMAT_DIRECT |
1207					 SRP_BUF_FORMAT_INDIRECT);
1208
1209	spin_lock_irqsave(hostdata->host->host_lock, flags);
1210	/* Start out with a request limit of 0, since this is negotiated in
1211	 * the login request we are just sending and login requests always
1212	 * get sent by the driver regardless of request_limit.
1213	 */
1214	atomic_set(&hostdata->request_limit, 0);
1215
1216	rc = ibmvscsi_send_srp_event(evt_struct, hostdata, login_timeout * 2);
1217	spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1218	dev_info(hostdata->dev, "sent SRP login\n");
1219	return rc;
1220};
1221
1222/**
1223 * capabilities_rsp: - Handle response to MAD adapter capabilities request
1224 * @evt_struct:	srp_event_struct with the response
1225 *
1226 * Used as a "done" callback by when sending adapter_info.
1227 */
1228static void capabilities_rsp(struct srp_event_struct *evt_struct)
1229{
1230	struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
1231
1232	if (evt_struct->xfer_iu->mad.capabilities.common.status) {
1233		dev_err(hostdata->dev, "error 0x%X getting capabilities info\n",
1234			evt_struct->xfer_iu->mad.capabilities.common.status);
1235	} else {
1236		if (hostdata->caps.migration.common.server_support !=
1237		    cpu_to_be16(SERVER_SUPPORTS_CAP))
1238			dev_info(hostdata->dev, "Partition migration not supported\n");
1239
1240		if (client_reserve) {
1241			if (hostdata->caps.reserve.common.server_support ==
1242			    cpu_to_be16(SERVER_SUPPORTS_CAP))
1243				dev_info(hostdata->dev, "Client reserve enabled\n");
1244			else
1245				dev_info(hostdata->dev, "Client reserve not supported\n");
1246		}
1247	}
1248
1249	send_srp_login(hostdata);
1250}
1251
1252/**
1253 * send_mad_capabilities: - Sends the mad capabilities request
1254 *      and stores the result so it can be retrieved with
1255 * @hostdata:	ibmvscsi_host_data of host
1256 */
1257static void send_mad_capabilities(struct ibmvscsi_host_data *hostdata)
1258{
1259	struct viosrp_capabilities *req;
1260	struct srp_event_struct *evt_struct;
1261	unsigned long flags;
1262	struct device_node *of_node = hostdata->dev->of_node;
1263	const char *location;
1264
1265	evt_struct = get_event_struct(&hostdata->pool);
1266	BUG_ON(!evt_struct);
1267
1268	init_event_struct(evt_struct, capabilities_rsp,
1269			  VIOSRP_MAD_FORMAT, info_timeout);
1270
1271	req = &evt_struct->iu.mad.capabilities;
1272	memset(req, 0, sizeof(*req));
1273
1274	hostdata->caps.flags = cpu_to_be32(CAP_LIST_SUPPORTED);
1275	if (hostdata->client_migrated)
1276		hostdata->caps.flags |= cpu_to_be32(CLIENT_MIGRATED);
1277
1278	strncpy(hostdata->caps.name, dev_name(&hostdata->host->shost_gendev),
1279		sizeof(hostdata->caps.name));
1280	hostdata->caps.name[sizeof(hostdata->caps.name) - 1] = '\0';
1281
1282	location = of_get_property(of_node, "ibm,loc-code", NULL);
1283	location = location ? location : dev_name(hostdata->dev);
1284	strncpy(hostdata->caps.loc, location, sizeof(hostdata->caps.loc));
1285	hostdata->caps.loc[sizeof(hostdata->caps.loc) - 1] = '\0';
1286
1287	req->common.type = cpu_to_be32(VIOSRP_CAPABILITIES_TYPE);
1288	req->buffer = cpu_to_be64(hostdata->caps_addr);
1289
1290	hostdata->caps.migration.common.cap_type =
1291				cpu_to_be32(MIGRATION_CAPABILITIES);
1292	hostdata->caps.migration.common.length =
1293				cpu_to_be16(sizeof(hostdata->caps.migration));
1294	hostdata->caps.migration.common.server_support =
1295				cpu_to_be16(SERVER_SUPPORTS_CAP);
1296	hostdata->caps.migration.ecl = cpu_to_be32(1);
1297
1298	if (client_reserve) {
1299		hostdata->caps.reserve.common.cap_type =
1300					cpu_to_be32(RESERVATION_CAPABILITIES);
1301		hostdata->caps.reserve.common.length =
1302				cpu_to_be16(sizeof(hostdata->caps.reserve));
1303		hostdata->caps.reserve.common.server_support =
1304				cpu_to_be16(SERVER_SUPPORTS_CAP);
1305		hostdata->caps.reserve.type =
1306				cpu_to_be32(CLIENT_RESERVE_SCSI_2);
1307		req->common.length =
1308				cpu_to_be16(sizeof(hostdata->caps));
1309	} else
1310		req->common.length = cpu_to_be16(sizeof(hostdata->caps) -
1311						sizeof(hostdata->caps.reserve));
1312
1313	spin_lock_irqsave(hostdata->host->host_lock, flags);
1314	if (ibmvscsi_send_srp_event(evt_struct, hostdata, info_timeout * 2))
1315		dev_err(hostdata->dev, "couldn't send CAPABILITIES_REQ!\n");
1316	spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1317};
1318
1319/**
1320 * fast_fail_rsp: - Handle response to MAD enable fast fail
1321 * @evt_struct:	srp_event_struct with the response
1322 *
1323 * Used as a "done" callback by when sending enable fast fail. Gets called
1324 * by ibmvscsi_handle_crq()
1325 */
1326static void fast_fail_rsp(struct srp_event_struct *evt_struct)
1327{
1328	struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
1329	u16 status = be16_to_cpu(evt_struct->xfer_iu->mad.fast_fail.common.status);
1330
1331	if (status == VIOSRP_MAD_NOT_SUPPORTED)
1332		dev_err(hostdata->dev, "fast_fail not supported in server\n");
1333	else if (status == VIOSRP_MAD_FAILED)
1334		dev_err(hostdata->dev, "fast_fail request failed\n");
1335	else if (status != VIOSRP_MAD_SUCCESS)
1336		dev_err(hostdata->dev, "error 0x%X enabling fast_fail\n", status);
1337
1338	send_mad_capabilities(hostdata);
1339}
1340
1341/**
1342 * init_host - Start host initialization
1343 * @hostdata:	ibmvscsi_host_data of host
1344 *
1345 * Returns zero if successful.
1346 */
1347static int enable_fast_fail(struct ibmvscsi_host_data *hostdata)
1348{
1349	int rc;
1350	unsigned long flags;
1351	struct viosrp_fast_fail *fast_fail_mad;
1352	struct srp_event_struct *evt_struct;
1353
1354	if (!fast_fail) {
1355		send_mad_capabilities(hostdata);
1356		return 0;
1357	}
1358
1359	evt_struct = get_event_struct(&hostdata->pool);
1360	BUG_ON(!evt_struct);
1361
1362	init_event_struct(evt_struct, fast_fail_rsp, VIOSRP_MAD_FORMAT, info_timeout);
1363
1364	fast_fail_mad = &evt_struct->iu.mad.fast_fail;
1365	memset(fast_fail_mad, 0, sizeof(*fast_fail_mad));
1366	fast_fail_mad->common.type = cpu_to_be32(VIOSRP_ENABLE_FAST_FAIL);
1367	fast_fail_mad->common.length = cpu_to_be16(sizeof(*fast_fail_mad));
1368
1369	spin_lock_irqsave(hostdata->host->host_lock, flags);
1370	rc = ibmvscsi_send_srp_event(evt_struct, hostdata, info_timeout * 2);
1371	spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1372	return rc;
1373}
1374
1375/**
1376 * adapter_info_rsp: - Handle response to MAD adapter info request
1377 * @evt_struct:	srp_event_struct with the response
1378 *
1379 * Used as a "done" callback by when sending adapter_info. Gets called
1380 * by ibmvscsi_handle_crq()
1381*/
1382static void adapter_info_rsp(struct srp_event_struct *evt_struct)
1383{
1384	struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
1385
1386	if (evt_struct->xfer_iu->mad.adapter_info.common.status) {
1387		dev_err(hostdata->dev, "error %d getting adapter info\n",
1388			evt_struct->xfer_iu->mad.adapter_info.common.status);
1389	} else {
1390		dev_info(hostdata->dev, "host srp version: %s, "
1391			 "host partition %s (%d), OS %d, max io %u\n",
1392			 hostdata->madapter_info.srp_version,
1393			 hostdata->madapter_info.partition_name,
1394			 be32_to_cpu(hostdata->madapter_info.partition_number),
1395			 be32_to_cpu(hostdata->madapter_info.os_type),
1396			 be32_to_cpu(hostdata->madapter_info.port_max_txu[0]));
1397		
1398		if (hostdata->madapter_info.port_max_txu[0]) 
1399			hostdata->host->max_sectors = 
1400				be32_to_cpu(hostdata->madapter_info.port_max_txu[0]) >> 9;
1401		
1402		if (be32_to_cpu(hostdata->madapter_info.os_type) == SRP_MAD_OS_AIX &&
1403		    strcmp(hostdata->madapter_info.srp_version, "1.6a") <= 0) {
1404			dev_err(hostdata->dev, "host (Ver. %s) doesn't support large transfers\n",
1405				hostdata->madapter_info.srp_version);
1406			dev_err(hostdata->dev, "limiting scatterlists to %d\n",
1407				MAX_INDIRECT_BUFS);
1408			hostdata->host->sg_tablesize = MAX_INDIRECT_BUFS;
1409		}
1410
1411		if (be32_to_cpu(hostdata->madapter_info.os_type) == SRP_MAD_OS_AIX) {
1412			enable_fast_fail(hostdata);
1413			return;
1414		}
1415	}
1416
1417	send_srp_login(hostdata);
1418}
1419
1420/**
1421 * send_mad_adapter_info: - Sends the mad adapter info request
1422 *      and stores the result so it can be retrieved with
1423 *      sysfs.  We COULD consider causing a failure if the
1424 *      returned SRP version doesn't match ours.
1425 * @hostdata:	ibmvscsi_host_data of host
1426 * 
1427 * Returns zero if successful.
1428*/
1429static void send_mad_adapter_info(struct ibmvscsi_host_data *hostdata)
1430{
1431	struct viosrp_adapter_info *req;
1432	struct srp_event_struct *evt_struct;
1433	unsigned long flags;
1434
1435	evt_struct = get_event_struct(&hostdata->pool);
1436	BUG_ON(!evt_struct);
1437
1438	init_event_struct(evt_struct,
1439			  adapter_info_rsp,
1440			  VIOSRP_MAD_FORMAT,
1441			  info_timeout);
1442	
1443	req = &evt_struct->iu.mad.adapter_info;
1444	memset(req, 0x00, sizeof(*req));
1445	
1446	req->common.type = cpu_to_be32(VIOSRP_ADAPTER_INFO_TYPE);
1447	req->common.length = cpu_to_be16(sizeof(hostdata->madapter_info));
1448	req->buffer = cpu_to_be64(hostdata->adapter_info_addr);
1449
1450	spin_lock_irqsave(hostdata->host->host_lock, flags);
1451	if (ibmvscsi_send_srp_event(evt_struct, hostdata, info_timeout * 2))
1452		dev_err(hostdata->dev, "couldn't send ADAPTER_INFO_REQ!\n");
1453	spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1454};
1455
1456/**
1457 * init_adapter: Start virtual adapter initialization sequence
1458 *
1459 */
1460static void init_adapter(struct ibmvscsi_host_data *hostdata)
1461{
1462	send_mad_adapter_info(hostdata);
1463}
1464
1465/**
1466 * sync_completion: Signal that a synchronous command has completed
1467 * Note that after returning from this call, the evt_struct is freed.
1468 * the caller waiting on this completion shouldn't touch the evt_struct
1469 * again.
1470 */
1471static void sync_completion(struct srp_event_struct *evt_struct)
1472{
1473	/* copy the response back */
1474	if (evt_struct->sync_srp)
1475		*evt_struct->sync_srp = *evt_struct->xfer_iu;
1476	
1477	complete(&evt_struct->comp);
1478}
1479
1480/**
1481 * ibmvscsi_abort: Abort a command...from scsi host template
1482 * send this over to the server and wait synchronously for the response
1483 */
1484static int ibmvscsi_eh_abort_handler(struct scsi_cmnd *cmd)
1485{
1486	struct ibmvscsi_host_data *hostdata = shost_priv(cmd->device->host);
1487	struct srp_tsk_mgmt *tsk_mgmt;
1488	struct srp_event_struct *evt;
1489	struct srp_event_struct *tmp_evt, *found_evt;
1490	union viosrp_iu srp_rsp;
1491	int rsp_rc;
1492	unsigned long flags;
1493	u16 lun = lun_from_dev(cmd->device);
1494	unsigned long wait_switch = 0;
1495
1496	/* First, find this command in our sent list so we can figure
1497	 * out the correct tag
1498	 */
1499	spin_lock_irqsave(hostdata->host->host_lock, flags);
1500	wait_switch = jiffies + (init_timeout * HZ);
1501	do {
1502		found_evt = NULL;
1503		list_for_each_entry(tmp_evt, &hostdata->sent, list) {
1504			if (tmp_evt->cmnd == cmd) {
1505				found_evt = tmp_evt;
1506				break;
1507			}
1508		}
1509
1510		if (!found_evt) {
1511			spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1512			return SUCCESS;
1513		}
1514
1515		evt = get_event_struct(&hostdata->pool);
1516		if (evt == NULL) {
1517			spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1518			sdev_printk(KERN_ERR, cmd->device,
1519				"failed to allocate abort event\n");
1520			return FAILED;
1521		}
1522	
1523		init_event_struct(evt,
1524				  sync_completion,
1525				  VIOSRP_SRP_FORMAT,
1526				  abort_timeout);
1527
1528		tsk_mgmt = &evt->iu.srp.tsk_mgmt;
1529	
1530		/* Set up an abort SRP command */
1531		memset(tsk_mgmt, 0x00, sizeof(*tsk_mgmt));
1532		tsk_mgmt->opcode = SRP_TSK_MGMT;
1533		int_to_scsilun(lun, &tsk_mgmt->lun);
1534		tsk_mgmt->tsk_mgmt_func = SRP_TSK_ABORT_TASK;
1535		tsk_mgmt->task_tag = (u64) found_evt;
1536
1537		evt->sync_srp = &srp_rsp;
1538
1539		init_completion(&evt->comp);
1540		rsp_rc = ibmvscsi_send_srp_event(evt, hostdata, abort_timeout * 2);
1541
1542		if (rsp_rc != SCSI_MLQUEUE_HOST_BUSY)
1543			break;
1544
1545		spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1546		msleep(10);
1547		spin_lock_irqsave(hostdata->host->host_lock, flags);
1548	} while (time_before(jiffies, wait_switch));
1549
1550	spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1551
1552	if (rsp_rc != 0) {
1553		sdev_printk(KERN_ERR, cmd->device,
1554			    "failed to send abort() event. rc=%d\n", rsp_rc);
1555		return FAILED;
1556	}
1557
1558	sdev_printk(KERN_INFO, cmd->device,
1559                    "aborting command. lun 0x%llx, tag 0x%llx\n",
1560		    (((u64) lun) << 48), (u64) found_evt);
1561
1562	wait_for_completion(&evt->comp);
1563
1564	/* make sure we got a good response */
1565	if (unlikely(srp_rsp.srp.rsp.opcode != SRP_RSP)) {
1566		if (printk_ratelimit())
1567			sdev_printk(KERN_WARNING, cmd->device, "abort bad SRP RSP type %d\n",
1568				    srp_rsp.srp.rsp.opcode);
1569		return FAILED;
1570	}
1571
1572	if (srp_rsp.srp.rsp.flags & SRP_RSP_FLAG_RSPVALID)
1573		rsp_rc = *((int *)srp_rsp.srp.rsp.data);
1574	else
1575		rsp_rc = srp_rsp.srp.rsp.status;
1576
1577	if (rsp_rc) {
1578		if (printk_ratelimit())
1579			sdev_printk(KERN_WARNING, cmd->device,
1580				    "abort code %d for task tag 0x%llx\n",
1581				    rsp_rc, tsk_mgmt->task_tag);
1582		return FAILED;
1583	}
1584
1585	/* Because we dropped the spinlock above, it's possible
1586	 * The event is no longer in our list.  Make sure it didn't
1587	 * complete while we were aborting
1588	 */
1589	spin_lock_irqsave(hostdata->host->host_lock, flags);
1590	found_evt = NULL;
1591	list_for_each_entry(tmp_evt, &hostdata->sent, list) {
1592		if (tmp_evt->cmnd == cmd) {
1593			found_evt = tmp_evt;
1594			break;
1595		}
1596	}
1597
1598	if (found_evt == NULL) {
1599		spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1600		sdev_printk(KERN_INFO, cmd->device, "aborted task tag 0x%llx completed\n",
1601			    tsk_mgmt->task_tag);
1602		return SUCCESS;
1603	}
1604
1605	sdev_printk(KERN_INFO, cmd->device, "successfully aborted task tag 0x%llx\n",
1606		    tsk_mgmt->task_tag);
1607
1608	cmd->result = (DID_ABORT << 16);
1609	list_del(&found_evt->list);
1610	unmap_cmd_data(&found_evt->iu.srp.cmd, found_evt,
1611		       found_evt->hostdata->dev);
1612	free_event_struct(&found_evt->hostdata->pool, found_evt);
1613	spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1614	atomic_inc(&hostdata->request_limit);
1615	return SUCCESS;
1616}
1617
1618/**
1619 * ibmvscsi_eh_device_reset_handler: Reset a single LUN...from scsi host 
1620 * template send this over to the server and wait synchronously for the 
1621 * response
1622 */
1623static int ibmvscsi_eh_device_reset_handler(struct scsi_cmnd *cmd)
1624{
1625	struct ibmvscsi_host_data *hostdata = shost_priv(cmd->device->host);
1626	struct srp_tsk_mgmt *tsk_mgmt;
1627	struct srp_event_struct *evt;
1628	struct srp_event_struct *tmp_evt, *pos;
1629	union viosrp_iu srp_rsp;
1630	int rsp_rc;
1631	unsigned long flags;
1632	u16 lun = lun_from_dev(cmd->device);
1633	unsigned long wait_switch = 0;
1634
1635	spin_lock_irqsave(hostdata->host->host_lock, flags);
1636	wait_switch = jiffies + (init_timeout * HZ);
1637	do {
1638		evt = get_event_struct(&hostdata->pool);
1639		if (evt == NULL) {
1640			spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1641			sdev_printk(KERN_ERR, cmd->device,
1642				"failed to allocate reset event\n");
1643			return FAILED;
1644		}
1645	
1646		init_event_struct(evt,
1647				  sync_completion,
1648				  VIOSRP_SRP_FORMAT,
1649				  reset_timeout);
1650
1651		tsk_mgmt = &evt->iu.srp.tsk_mgmt;
1652
1653		/* Set up a lun reset SRP command */
1654		memset(tsk_mgmt, 0x00, sizeof(*tsk_mgmt));
1655		tsk_mgmt->opcode = SRP_TSK_MGMT;
1656		int_to_scsilun(lun, &tsk_mgmt->lun);
1657		tsk_mgmt->tsk_mgmt_func = SRP_TSK_LUN_RESET;
1658
1659		evt->sync_srp = &srp_rsp;
1660
1661		init_completion(&evt->comp);
1662		rsp_rc = ibmvscsi_send_srp_event(evt, hostdata, reset_timeout * 2);
1663
1664		if (rsp_rc != SCSI_MLQUEUE_HOST_BUSY)
1665			break;
1666
1667		spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1668		msleep(10);
1669		spin_lock_irqsave(hostdata->host->host_lock, flags);
1670	} while (time_before(jiffies, wait_switch));
1671
1672	spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1673
1674	if (rsp_rc != 0) {
1675		sdev_printk(KERN_ERR, cmd->device,
1676			    "failed to send reset event. rc=%d\n", rsp_rc);
1677		return FAILED;
1678	}
1679
1680	sdev_printk(KERN_INFO, cmd->device, "resetting device. lun 0x%llx\n",
1681		    (((u64) lun) << 48));
1682
1683	wait_for_completion(&evt->comp);
1684
1685	/* make sure we got a good response */
1686	if (unlikely(srp_rsp.srp.rsp.opcode != SRP_RSP)) {
1687		if (printk_ratelimit())
1688			sdev_printk(KERN_WARNING, cmd->device, "reset bad SRP RSP type %d\n",
1689				    srp_rsp.srp.rsp.opcode);
1690		return FAILED;
1691	}
1692
1693	if (srp_rsp.srp.rsp.flags & SRP_RSP_FLAG_RSPVALID)
1694		rsp_rc = *((int *)srp_rsp.srp.rsp.data);
1695	else
1696		rsp_rc = srp_rsp.srp.rsp.status;
1697
1698	if (rsp_rc) {
1699		if (printk_ratelimit())
1700			sdev_printk(KERN_WARNING, cmd->device,
1701				    "reset code %d for task tag 0x%llx\n",
1702				    rsp_rc, tsk_mgmt->task_tag);
1703		return FAILED;
1704	}
1705
1706	/* We need to find all commands for this LUN that have not yet been
1707	 * responded to, and fail them with DID_RESET
1708	 */
1709	spin_lock_irqsave(hostdata->host->host_lock, flags);
1710	list_for_each_entry_safe(tmp_evt, pos, &hostdata->sent, list) {
1711		if ((tmp_evt->cmnd) && (tmp_evt->cmnd->device == cmd->device)) {
1712			if (tmp_evt->cmnd)
1713				tmp_evt->cmnd->result = (DID_RESET << 16);
1714			list_del(&tmp_evt->list);
1715			unmap_cmd_data(&tmp_evt->iu.srp.cmd, tmp_evt,
1716				       tmp_evt->hostdata->dev);
1717			free_event_struct(&tmp_evt->hostdata->pool,
1718						   tmp_evt);
1719			atomic_inc(&hostdata->request_limit);
1720			if (tmp_evt->cmnd_done)
1721				tmp_evt->cmnd_done(tmp_evt->cmnd);
1722			else if (tmp_evt->done)
1723				tmp_evt->done(tmp_evt);
1724		}
1725	}
1726	spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1727	return SUCCESS;
1728}
1729
1730/**
1731 * ibmvscsi_eh_host_reset_handler - Reset the connection to the server
1732 * @cmd:	struct scsi_cmnd having problems
1733*/
1734static int ibmvscsi_eh_host_reset_handler(struct scsi_cmnd *cmd)
1735{
1736	unsigned long wait_switch = 0;
1737	struct ibmvscsi_host_data *hostdata = shost_priv(cmd->device->host);
1738
1739	dev_err(hostdata->dev, "Resetting connection due to error recovery\n");
1740
1741	ibmvscsi_reset_host(hostdata);
1742
1743	for (wait_switch = jiffies + (init_timeout * HZ);
1744	     time_before(jiffies, wait_switch) &&
1745		     atomic_read(&hostdata->request_limit) < 2;) {
1746
1747		msleep(10);
1748	}
1749
1750	if (atomic_read(&hostdata->request_limit) <= 0)
1751		return FAILED;
1752
1753	return SUCCESS;
1754}
1755
1756/**
1757 * ibmvscsi_handle_crq: - Handles and frees received events in the CRQ
1758 * @crq:	Command/Response queue
1759 * @hostdata:	ibmvscsi_host_data of host
1760 *
1761*/
1762static void ibmvscsi_handle_crq(struct viosrp_crq *crq,
1763				struct ibmvscsi_host_data *hostdata)
1764{
1765	long rc;
1766	unsigned long flags;
1767	/* The hypervisor copies our tag value here so no byteswapping */
1768	struct srp_event_struct *evt_struct =
1769			(__force struct srp_event_struct *)crq->IU_data_ptr;
1770	switch (crq->valid) {
1771	case VIOSRP_CRQ_INIT_RSP:		/* initialization */
1772		switch (crq->format) {
1773		case VIOSRP_CRQ_INIT:	/* Initialization message */
1774			dev_info(hostdata->dev, "partner initialized\n");
1775			/* Send back a response */
1776			rc = ibmvscsi_send_crq(hostdata, 0xC002000000000000LL, 0);
1777			if (rc == 0) {
1778				/* Now login */
1779				init_adapter(hostdata);
1780			} else {
1781				dev_err(hostdata->dev, "Unable to send init rsp. rc=%ld\n", rc);
1782			}
1783
1784			break;
1785		case VIOSRP_CRQ_INIT_COMPLETE:	/* Initialization response */
1786			dev_info(hostdata->dev, "partner initialization complete\n");
1787
1788			/* Now login */
1789			init_adapter(hostdata);
1790			break;
1791		default:
1792			dev_err(hostdata->dev, "unknown crq message type: %d\n", crq->format);
1793		}
1794		return;
1795	case VIOSRP_CRQ_XPORT_EVENT:	/* Hypervisor telling us the connection is closed */
1796		scsi_block_requests(hostdata->host);
1797		atomic_set(&hostdata->request_limit, 0);
1798		if (crq->format == 0x06) {
1799			/* We need to re-setup the interpartition connection */
1800			dev_info(hostdata->dev, "Re-enabling adapter!\n");
1801			hostdata->client_migrated = 1;
1802			hostdata->reenable_crq = 1;
1803			purge_requests(hostdata, DID_REQUEUE);
1804			wake_up(&hostdata->work_wait_q);
1805		} else {
1806			dev_err(hostdata->dev, "Virtual adapter failed rc %d!\n",
1807				crq->format);
1808			ibmvscsi_reset_host(hostdata);
1809		}
1810		return;
1811	case VIOSRP_CRQ_CMD_RSP:		/* real payload */
1812		break;
1813	default:
1814		dev_err(hostdata->dev, "got an invalid message type 0x%02x\n",
1815			crq->valid);
1816		return;
1817	}
1818
1819	/* The only kind of payload CRQs we should get are responses to
1820	 * things we send. Make sure this response is to something we
1821	 * actually sent
1822	 */
1823	if (!valid_event_struct(&hostdata->pool, evt_struct)) {
1824		dev_err(hostdata->dev, "returned correlation_token 0x%p is invalid!\n",
1825		       evt_struct);
1826		return;
1827	}
1828
1829	if (atomic_read(&evt_struct->free)) {
1830		dev_err(hostdata->dev, "received duplicate correlation_token 0x%p!\n",
1831			evt_struct);
1832		return;
1833	}
1834
1835	if (crq->format == VIOSRP_SRP_FORMAT)
1836		atomic_add(be32_to_cpu(evt_struct->xfer_iu->srp.rsp.req_lim_delta),
1837			   &hostdata->request_limit);
1838
1839	del_timer(&evt_struct->timer);
1840
1841	if ((crq->status != VIOSRP_OK && crq->status != VIOSRP_OK2) && evt_struct->cmnd)
1842		evt_struct->cmnd->result = DID_ERROR << 16;
1843	if (evt_struct->done)
1844		evt_struct->done(evt_struct);
1845	else
1846		dev_err(hostdata->dev, "returned done() is NULL; not running it!\n");
1847
1848	/*
1849	 * Lock the host_lock before messing with these structures, since we
1850	 * are running in a task context
1851	 */
1852	spin_lock_irqsave(evt_struct->hostdata->host->host_lock, flags);
1853	list_del(&evt_struct->list);
1854	free_event_struct(&evt_struct->hostdata->pool, evt_struct);
1855	spin_unlock_irqrestore(evt_struct->hostdata->host->host_lock, flags);
1856}
1857
1858/**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1859 * ibmvscsi_slave_configure: Set the "allow_restart" flag for each disk.
1860 * @sdev:	struct scsi_device device to configure
1861 *
1862 * Enable allow_restart for a device if it is a disk.  Adjust the
1863 * queue_depth here also as is required by the documentation for
1864 * struct scsi_host_template.
1865 */
1866static int ibmvscsi_slave_configure(struct scsi_device *sdev)
1867{
1868	struct Scsi_Host *shost = sdev->host;
1869	unsigned long lock_flags = 0;
1870
1871	spin_lock_irqsave(shost->host_lock, lock_flags);
1872	if (sdev->type == TYPE_DISK) {
1873		sdev->allow_restart = 1;
1874		blk_queue_rq_timeout(sdev->request_queue, 120 * HZ);
1875	}
1876	spin_unlock_irqrestore(shost->host_lock, lock_flags);
 
1877	return 0;
1878}
1879
1880/**
1881 * ibmvscsi_change_queue_depth - Change the device's queue depth
1882 * @sdev:	scsi device struct
1883 * @qdepth:	depth to set
1884 * @reason:	calling context
1885 *
1886 * Return value:
1887 * 	actual depth set
1888 **/
1889static int ibmvscsi_change_queue_depth(struct scsi_device *sdev, int qdepth)
 
1890{
 
 
 
1891	if (qdepth > IBMVSCSI_MAX_CMDS_PER_LUN)
1892		qdepth = IBMVSCSI_MAX_CMDS_PER_LUN;
1893	return scsi_change_queue_depth(sdev, qdepth);
 
 
1894}
1895
1896/* ------------------------------------------------------------
1897 * sysfs attributes
1898 */
1899static ssize_t show_host_vhost_loc(struct device *dev,
1900				   struct device_attribute *attr, char *buf)
1901{
1902	struct Scsi_Host *shost = class_to_shost(dev);
1903	struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1904	int len;
1905
1906	len = snprintf(buf, sizeof(hostdata->caps.loc), "%s\n",
1907		       hostdata->caps.loc);
1908	return len;
1909}
1910
1911static struct device_attribute ibmvscsi_host_vhost_loc = {
1912	.attr = {
1913		 .name = "vhost_loc",
1914		 .mode = S_IRUGO,
1915		 },
1916	.show = show_host_vhost_loc,
1917};
1918
1919static ssize_t show_host_vhost_name(struct device *dev,
1920				    struct device_attribute *attr, char *buf)
1921{
1922	struct Scsi_Host *shost = class_to_shost(dev);
1923	struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1924	int len;
1925
1926	len = snprintf(buf, sizeof(hostdata->caps.name), "%s\n",
1927		       hostdata->caps.name);
1928	return len;
1929}
1930
1931static struct device_attribute ibmvscsi_host_vhost_name = {
1932	.attr = {
1933		 .name = "vhost_name",
1934		 .mode = S_IRUGO,
1935		 },
1936	.show = show_host_vhost_name,
1937};
1938
1939static ssize_t show_host_srp_version(struct device *dev,
1940				     struct device_attribute *attr, char *buf)
1941{
1942	struct Scsi_Host *shost = class_to_shost(dev);
1943	struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1944	int len;
1945
1946	len = snprintf(buf, PAGE_SIZE, "%s\n",
1947		       hostdata->madapter_info.srp_version);
1948	return len;
1949}
1950
1951static struct device_attribute ibmvscsi_host_srp_version = {
1952	.attr = {
1953		 .name = "srp_version",
1954		 .mode = S_IRUGO,
1955		 },
1956	.show = show_host_srp_version,
1957};
1958
1959static ssize_t show_host_partition_name(struct device *dev,
1960					struct device_attribute *attr,
1961					char *buf)
1962{
1963	struct Scsi_Host *shost = class_to_shost(dev);
1964	struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1965	int len;
1966
1967	len = snprintf(buf, PAGE_SIZE, "%s\n",
1968		       hostdata->madapter_info.partition_name);
1969	return len;
1970}
1971
1972static struct device_attribute ibmvscsi_host_partition_name = {
1973	.attr = {
1974		 .name = "partition_name",
1975		 .mode = S_IRUGO,
1976		 },
1977	.show = show_host_partition_name,
1978};
1979
1980static ssize_t show_host_partition_number(struct device *dev,
1981					  struct device_attribute *attr,
1982					  char *buf)
1983{
1984	struct Scsi_Host *shost = class_to_shost(dev);
1985	struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1986	int len;
1987
1988	len = snprintf(buf, PAGE_SIZE, "%d\n",
1989		       be32_to_cpu(hostdata->madapter_info.partition_number));
1990	return len;
1991}
1992
1993static struct device_attribute ibmvscsi_host_partition_number = {
1994	.attr = {
1995		 .name = "partition_number",
1996		 .mode = S_IRUGO,
1997		 },
1998	.show = show_host_partition_number,
1999};
2000
2001static ssize_t show_host_mad_version(struct device *dev,
2002				     struct device_attribute *attr, char *buf)
2003{
2004	struct Scsi_Host *shost = class_to_shost(dev);
2005	struct ibmvscsi_host_data *hostdata = shost_priv(shost);
2006	int len;
2007
2008	len = snprintf(buf, PAGE_SIZE, "%d\n",
2009		       be32_to_cpu(hostdata->madapter_info.mad_version));
2010	return len;
2011}
2012
2013static struct device_attribute ibmvscsi_host_mad_version = {
2014	.attr = {
2015		 .name = "mad_version",
2016		 .mode = S_IRUGO,
2017		 },
2018	.show = show_host_mad_version,
2019};
2020
2021static ssize_t show_host_os_type(struct device *dev,
2022				 struct device_attribute *attr, char *buf)
2023{
2024	struct Scsi_Host *shost = class_to_shost(dev);
2025	struct ibmvscsi_host_data *hostdata = shost_priv(shost);
2026	int len;
2027
2028	len = snprintf(buf, PAGE_SIZE, "%d\n",
2029		       be32_to_cpu(hostdata->madapter_info.os_type));
2030	return len;
2031}
2032
2033static struct device_attribute ibmvscsi_host_os_type = {
2034	.attr = {
2035		 .name = "os_type",
2036		 .mode = S_IRUGO,
2037		 },
2038	.show = show_host_os_type,
2039};
2040
2041static ssize_t show_host_config(struct device *dev,
2042				struct device_attribute *attr, char *buf)
2043{
2044	return 0;
 
 
 
 
 
 
 
2045}
2046
2047static struct device_attribute ibmvscsi_host_config = {
2048	.attr = {
2049		.name = "config",
2050		.mode = S_IRUGO,
2051		},
2052	.show = show_host_config,
2053};
2054
2055static struct device_attribute *ibmvscsi_attrs[] = {
2056	&ibmvscsi_host_vhost_loc,
2057	&ibmvscsi_host_vhost_name,
2058	&ibmvscsi_host_srp_version,
2059	&ibmvscsi_host_partition_name,
2060	&ibmvscsi_host_partition_number,
2061	&ibmvscsi_host_mad_version,
2062	&ibmvscsi_host_os_type,
2063	&ibmvscsi_host_config,
2064	NULL
2065};
2066
2067/* ------------------------------------------------------------
2068 * SCSI driver registration
2069 */
2070static struct scsi_host_template driver_template = {
2071	.module = THIS_MODULE,
2072	.name = "IBM POWER Virtual SCSI Adapter " IBMVSCSI_VERSION,
2073	.proc_name = "ibmvscsi",
2074	.queuecommand = ibmvscsi_queuecommand,
2075	.eh_abort_handler = ibmvscsi_eh_abort_handler,
2076	.eh_device_reset_handler = ibmvscsi_eh_device_reset_handler,
2077	.eh_host_reset_handler = ibmvscsi_eh_host_reset_handler,
2078	.slave_configure = ibmvscsi_slave_configure,
2079	.change_queue_depth = ibmvscsi_change_queue_depth,
2080	.cmd_per_lun = IBMVSCSI_CMDS_PER_LUN_DEFAULT,
2081	.can_queue = IBMVSCSI_MAX_REQUESTS_DEFAULT,
2082	.this_id = -1,
2083	.sg_tablesize = SG_ALL,
2084	.use_clustering = ENABLE_CLUSTERING,
2085	.shost_attrs = ibmvscsi_attrs,
2086};
2087
2088/**
2089 * ibmvscsi_get_desired_dma - Calculate IO memory desired by the driver
2090 *
2091 * @vdev: struct vio_dev for the device whose desired IO mem is to be returned
2092 *
2093 * Return value:
2094 *	Number of bytes of IO data the driver will need to perform well.
2095 */
2096static unsigned long ibmvscsi_get_desired_dma(struct vio_dev *vdev)
2097{
2098	/* iu_storage data allocated in initialize_event_pool */
2099	unsigned long desired_io = max_events * sizeof(union viosrp_iu);
2100
2101	/* add io space for sg data */
2102	desired_io += (IBMVSCSI_MAX_SECTORS_DEFAULT * 512 *
2103	                     IBMVSCSI_CMDS_PER_LUN_DEFAULT);
2104
2105	return desired_io;
2106}
2107
2108static void ibmvscsi_do_work(struct ibmvscsi_host_data *hostdata)
2109{
2110	int rc;
2111	char *action = "reset";
2112
2113	if (hostdata->reset_crq) {
2114		smp_rmb();
2115		hostdata->reset_crq = 0;
2116
2117		rc = ibmvscsi_reset_crq_queue(&hostdata->queue, hostdata);
2118		if (!rc)
2119			rc = ibmvscsi_send_crq(hostdata, 0xC001000000000000LL, 0);
2120		vio_enable_interrupts(to_vio_dev(hostdata->dev));
2121	} else if (hostdata->reenable_crq) {
2122		smp_rmb();
2123		action = "enable";
2124		rc = ibmvscsi_reenable_crq_queue(&hostdata->queue, hostdata);
2125		hostdata->reenable_crq = 0;
2126		if (!rc)
2127			rc = ibmvscsi_send_crq(hostdata, 0xC001000000000000LL, 0);
2128	} else
2129		return;
2130
2131	if (rc) {
2132		atomic_set(&hostdata->request_limit, -1);
2133		dev_err(hostdata->dev, "error after %s\n", action);
2134	}
2135
2136	scsi_unblock_requests(hostdata->host);
2137}
2138
2139static int ibmvscsi_work_to_do(struct ibmvscsi_host_data *hostdata)
2140{
2141	if (kthread_should_stop())
2142		return 1;
2143	else if (hostdata->reset_crq) {
2144		smp_rmb();
2145		return 1;
2146	} else if (hostdata->reenable_crq) {
2147		smp_rmb();
2148		return 1;
2149	}
2150
2151	return 0;
2152}
2153
2154static int ibmvscsi_work(void *data)
2155{
2156	struct ibmvscsi_host_data *hostdata = data;
2157	int rc;
2158
2159	set_user_nice(current, MIN_NICE);
2160
2161	while (1) {
2162		rc = wait_event_interruptible(hostdata->work_wait_q,
2163					      ibmvscsi_work_to_do(hostdata));
2164
2165		BUG_ON(rc);
2166
2167		if (kthread_should_stop())
2168			break;
2169
2170		ibmvscsi_do_work(hostdata);
2171	}
2172
2173	return 0;
2174}
2175
2176/**
2177 * Called by bus code for each adapter
2178 */
2179static int ibmvscsi_probe(struct vio_dev *vdev, const struct vio_device_id *id)
2180{
2181	struct ibmvscsi_host_data *hostdata;
2182	struct Scsi_Host *host;
2183	struct device *dev = &vdev->dev;
2184	struct srp_rport_identifiers ids;
2185	struct srp_rport *rport;
2186	unsigned long wait_switch = 0;
2187	int rc;
2188
2189	dev_set_drvdata(&vdev->dev, NULL);
2190
2191	host = scsi_host_alloc(&driver_template, sizeof(*hostdata));
2192	if (!host) {
2193		dev_err(&vdev->dev, "couldn't allocate host data\n");
2194		goto scsi_host_alloc_failed;
2195	}
2196
2197	host->transportt = ibmvscsi_transport_template;
2198	hostdata = shost_priv(host);
2199	memset(hostdata, 0x00, sizeof(*hostdata));
2200	INIT_LIST_HEAD(&hostdata->sent);
2201	init_waitqueue_head(&hostdata->work_wait_q);
2202	hostdata->host = host;
2203	hostdata->dev = dev;
2204	atomic_set(&hostdata->request_limit, -1);
2205	hostdata->host->max_sectors = IBMVSCSI_MAX_SECTORS_DEFAULT;
2206
2207	if (map_persist_bufs(hostdata)) {
2208		dev_err(&vdev->dev, "couldn't map persistent buffers\n");
2209		goto persist_bufs_failed;
2210	}
2211
2212	hostdata->work_thread = kthread_run(ibmvscsi_work, hostdata, "%s_%d",
2213					    "ibmvscsi", host->host_no);
2214
2215	if (IS_ERR(hostdata->work_thread)) {
2216		dev_err(&vdev->dev, "couldn't initialize kthread. rc=%ld\n",
2217			PTR_ERR(hostdata->work_thread));
2218		goto init_crq_failed;
2219	}
2220
2221	rc = ibmvscsi_init_crq_queue(&hostdata->queue, hostdata, max_events);
2222	if (rc != 0 && rc != H_RESOURCE) {
2223		dev_err(&vdev->dev, "couldn't initialize crq. rc=%d\n", rc);
2224		goto kill_kthread;
2225	}
2226	if (initialize_event_pool(&hostdata->pool, max_events, hostdata) != 0) {
2227		dev_err(&vdev->dev, "couldn't initialize event pool\n");
2228		goto init_pool_failed;
2229	}
2230
2231	host->max_lun = IBMVSCSI_MAX_LUN;
2232	host->max_id = max_id;
2233	host->max_channel = max_channel;
2234	host->max_cmd_len = 16;
2235
2236	dev_info(dev,
2237		 "Maximum ID: %d Maximum LUN: %llu Maximum Channel: %d\n",
2238		 host->max_id, host->max_lun, host->max_channel);
2239
2240	if (scsi_add_host(hostdata->host, hostdata->dev))
2241		goto add_host_failed;
2242
2243	/* we don't have a proper target_port_id so let's use the fake one */
2244	memcpy(ids.port_id, hostdata->madapter_info.partition_name,
2245	       sizeof(ids.port_id));
2246	ids.roles = SRP_RPORT_ROLE_TARGET;
2247	rport = srp_rport_add(host, &ids);
2248	if (IS_ERR(rport))
2249		goto add_srp_port_failed;
2250
2251	/* Try to send an initialization message.  Note that this is allowed
2252	 * to fail if the other end is not acive.  In that case we don't
2253	 * want to scan
2254	 */
2255	if (ibmvscsi_send_crq(hostdata, 0xC001000000000000LL, 0) == 0
2256	    || rc == H_RESOURCE) {
2257		/*
2258		 * Wait around max init_timeout secs for the adapter to finish
2259		 * initializing. When we are done initializing, we will have a
2260		 * valid request_limit.  We don't want Linux scanning before
2261		 * we are ready.
2262		 */
2263		for (wait_switch = jiffies + (init_timeout * HZ);
2264		     time_before(jiffies, wait_switch) &&
2265		     atomic_read(&hostdata->request_limit) < 2;) {
2266
2267			msleep(10);
2268		}
2269
2270		/* if we now have a valid request_limit, initiate a scan */
2271		if (atomic_read(&hostdata->request_limit) > 0)
2272			scsi_scan_host(host);
2273	}
2274
2275	dev_set_drvdata(&vdev->dev, hostdata);
2276	list_add_tail(&hostdata->host_list, &ibmvscsi_head);
2277	return 0;
2278
2279      add_srp_port_failed:
2280	scsi_remove_host(hostdata->host);
2281      add_host_failed:
2282	release_event_pool(&hostdata->pool, hostdata);
2283      init_pool_failed:
2284	ibmvscsi_release_crq_queue(&hostdata->queue, hostdata, max_events);
2285      kill_kthread:
2286      kthread_stop(hostdata->work_thread);
2287      init_crq_failed:
2288	unmap_persist_bufs(hostdata);
2289      persist_bufs_failed:
2290	scsi_host_put(host);
2291      scsi_host_alloc_failed:
2292	return -1;
2293}
2294
2295static int ibmvscsi_remove(struct vio_dev *vdev)
2296{
2297	struct ibmvscsi_host_data *hostdata = dev_get_drvdata(&vdev->dev);
2298	list_del(&hostdata->host_list);
2299	unmap_persist_bufs(hostdata);
2300	release_event_pool(&hostdata->pool, hostdata);
2301	ibmvscsi_release_crq_queue(&hostdata->queue, hostdata,
2302					max_events);
2303
2304	kthread_stop(hostdata->work_thread);
2305	srp_remove_host(hostdata->host);
2306	scsi_remove_host(hostdata->host);
2307	scsi_host_put(hostdata->host);
2308
2309	return 0;
2310}
2311
2312/**
2313 * ibmvscsi_resume: Resume from suspend
2314 * @dev:	device struct
2315 *
2316 * We may have lost an interrupt across suspend/resume, so kick the
2317 * interrupt handler
2318 */
2319static int ibmvscsi_resume(struct device *dev)
2320{
2321	struct ibmvscsi_host_data *hostdata = dev_get_drvdata(dev);
2322	vio_disable_interrupts(to_vio_dev(hostdata->dev));
2323	tasklet_schedule(&hostdata->srp_task);
2324
2325	return 0;
2326}
2327
2328/**
2329 * ibmvscsi_device_table: Used by vio.c to match devices in the device tree we 
2330 * support.
2331 */
2332static struct vio_device_id ibmvscsi_device_table[] = {
2333	{"vscsi", "IBM,v-scsi"},
2334	{ "", "" }
2335};
2336MODULE_DEVICE_TABLE(vio, ibmvscsi_device_table);
2337
2338static struct dev_pm_ops ibmvscsi_pm_ops = {
2339	.resume = ibmvscsi_resume
2340};
2341
2342static struct vio_driver ibmvscsi_driver = {
2343	.id_table = ibmvscsi_device_table,
2344	.probe = ibmvscsi_probe,
2345	.remove = ibmvscsi_remove,
2346	.get_desired_dma = ibmvscsi_get_desired_dma,
2347	.name = "ibmvscsi",
2348	.pm = &ibmvscsi_pm_ops,
2349};
2350
2351static struct srp_function_template ibmvscsi_transport_functions = {
2352};
2353
2354int __init ibmvscsi_module_init(void)
2355{
2356	int ret;
2357
2358	/* Ensure we have two requests to do error recovery */
2359	driver_template.can_queue = max_requests;
2360	max_events = max_requests + 2;
2361
2362	if (!firmware_has_feature(FW_FEATURE_VIO))
2363		return -ENODEV;
2364
2365	ibmvscsi_transport_template =
2366		srp_attach_transport(&ibmvscsi_transport_functions);
2367	if (!ibmvscsi_transport_template)
2368		return -ENOMEM;
2369
2370	ret = vio_register_driver(&ibmvscsi_driver);
2371	if (ret)
2372		srp_release_transport(ibmvscsi_transport_template);
2373	return ret;
2374}
2375
2376void __exit ibmvscsi_module_exit(void)
2377{
2378	vio_unregister_driver(&ibmvscsi_driver);
2379	srp_release_transport(ibmvscsi_transport_template);
2380}
2381
2382module_init(ibmvscsi_module_init);
2383module_exit(ibmvscsi_module_exit);