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