]> bbs.cooldavid.org Git - net-next-2.6.git/blob - drivers/scsi/ibmvscsi/ibmvscsi.c
[SCSI] ibmvscsi: Fix error path deadlock
[net-next-2.6.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  * Note that some of the underlying infrastructure is different between
59  * machines conforming to the "RS/6000 Platform Architecture" (RPA) and
60  * the older iSeries hypervisor models.  To support both, some low level
61  * routines have been broken out into rpa_vscsi.c and iseries_vscsi.c.
62  * The Makefile should pick one, not two, not zero, of these.
63  *
64  * TODO: This is currently pretty tied to the IBM i/pSeries hypervisor
65  * interfaces.  It would be really nice to abstract this above an RDMA
66  * layer.
67  */
68
69 #include <linux/module.h>
70 #include <linux/moduleparam.h>
71 #include <linux/dma-mapping.h>
72 #include <linux/delay.h>
73 #include <linux/slab.h>
74 #include <linux/of.h>
75 #include <linux/pm.h>
76 #include <linux/kthread.h>
77 #include <asm/firmware.h>
78 #include <asm/vio.h>
79 #include <scsi/scsi.h>
80 #include <scsi/scsi_cmnd.h>
81 #include <scsi/scsi_host.h>
82 #include <scsi/scsi_device.h>
83 #include <scsi/scsi_transport_srp.h>
84 #include "ibmvscsi.h"
85
86 /* The values below are somewhat arbitrary default values, but 
87  * OS/400 will use 3 busses (disks, CDs, tapes, I think.)
88  * Note that there are 3 bits of channel value, 6 bits of id, and
89  * 5 bits of LUN.
90  */
91 static int max_id = 64;
92 static int max_channel = 3;
93 static int init_timeout = 300;
94 static int login_timeout = 60;
95 static int info_timeout = 30;
96 static int abort_timeout = 60;
97 static int reset_timeout = 60;
98 static int max_requests = IBMVSCSI_MAX_REQUESTS_DEFAULT;
99 static int max_events = IBMVSCSI_MAX_REQUESTS_DEFAULT + 2;
100 static int fast_fail = 1;
101 static int client_reserve = 1;
102
103 static struct scsi_transport_template *ibmvscsi_transport_template;
104
105 #define IBMVSCSI_VERSION "1.5.8"
106
107 static struct ibmvscsi_ops *ibmvscsi_ops;
108
109 MODULE_DESCRIPTION("IBM Virtual SCSI");
110 MODULE_AUTHOR("Dave Boutcher");
111 MODULE_LICENSE("GPL");
112 MODULE_VERSION(IBMVSCSI_VERSION);
113
114 module_param_named(max_id, max_id, int, S_IRUGO | S_IWUSR);
115 MODULE_PARM_DESC(max_id, "Largest ID value for each channel");
116 module_param_named(max_channel, max_channel, int, S_IRUGO | S_IWUSR);
117 MODULE_PARM_DESC(max_channel, "Largest channel value");
118 module_param_named(init_timeout, init_timeout, int, S_IRUGO | S_IWUSR);
119 MODULE_PARM_DESC(init_timeout, "Initialization timeout in seconds");
120 module_param_named(max_requests, max_requests, int, S_IRUGO);
121 MODULE_PARM_DESC(max_requests, "Maximum requests for this adapter");
122 module_param_named(fast_fail, fast_fail, int, S_IRUGO | S_IWUSR);
123 MODULE_PARM_DESC(fast_fail, "Enable fast fail. [Default=1]");
124 module_param_named(client_reserve, client_reserve, int, S_IRUGO );
125 MODULE_PARM_DESC(client_reserve, "Attempt client managed reserve/release");
126
127 /* ------------------------------------------------------------
128  * Routines for the event pool and event structs
129  */
130 /**
131  * initialize_event_pool: - Allocates and initializes the event pool for a host
132  * @pool:       event_pool to be initialized
133  * @size:       Number of events in pool
134  * @hostdata:   ibmvscsi_host_data who owns the event pool
135  *
136  * Returns zero on success.
137 */
138 static int initialize_event_pool(struct event_pool *pool,
139                                  int size, struct ibmvscsi_host_data *hostdata)
140 {
141         int i;
142
143         pool->size = size;
144         pool->next = 0;
145         pool->events = kcalloc(pool->size, sizeof(*pool->events), GFP_KERNEL);
146         if (!pool->events)
147                 return -ENOMEM;
148
149         pool->iu_storage =
150             dma_alloc_coherent(hostdata->dev,
151                                pool->size * sizeof(*pool->iu_storage),
152                                &pool->iu_token, 0);
153         if (!pool->iu_storage) {
154                 kfree(pool->events);
155                 return -ENOMEM;
156         }
157
158         for (i = 0; i < pool->size; ++i) {
159                 struct srp_event_struct *evt = &pool->events[i];
160                 memset(&evt->crq, 0x00, sizeof(evt->crq));
161                 atomic_set(&evt->free, 1);
162                 evt->crq.valid = 0x80;
163                 evt->crq.IU_length = sizeof(*evt->xfer_iu);
164                 evt->crq.IU_data_ptr = pool->iu_token + 
165                         sizeof(*evt->xfer_iu) * i;
166                 evt->xfer_iu = pool->iu_storage + i;
167                 evt->hostdata = hostdata;
168                 evt->ext_list = NULL;
169                 evt->ext_list_token = 0;
170         }
171
172         return 0;
173 }
174
175 /**
176  * release_event_pool: - Frees memory of an event pool of a host
177  * @pool:       event_pool to be released
178  * @hostdata:   ibmvscsi_host_data who owns the even pool
179  *
180  * Returns zero on success.
181 */
182 static void release_event_pool(struct event_pool *pool,
183                                struct ibmvscsi_host_data *hostdata)
184 {
185         int i, in_use = 0;
186         for (i = 0; i < pool->size; ++i) {
187                 if (atomic_read(&pool->events[i].free) != 1)
188                         ++in_use;
189                 if (pool->events[i].ext_list) {
190                         dma_free_coherent(hostdata->dev,
191                                   SG_ALL * sizeof(struct srp_direct_buf),
192                                   pool->events[i].ext_list,
193                                   pool->events[i].ext_list_token);
194                 }
195         }
196         if (in_use)
197                 dev_warn(hostdata->dev, "releasing event pool with %d "
198                          "events still in use?\n", in_use);
199         kfree(pool->events);
200         dma_free_coherent(hostdata->dev,
201                           pool->size * sizeof(*pool->iu_storage),
202                           pool->iu_storage, pool->iu_token);
203 }
204
205 /**
206  * valid_event_struct: - Determines if event is valid.
207  * @pool:       event_pool that contains the event
208  * @evt:        srp_event_struct to be checked for validity
209  *
210  * Returns zero if event is invalid, one otherwise.
211 */
212 static int valid_event_struct(struct event_pool *pool,
213                                 struct srp_event_struct *evt)
214 {
215         int index = evt - pool->events;
216         if (index < 0 || index >= pool->size)   /* outside of bounds */
217                 return 0;
218         if (evt != pool->events + index)        /* unaligned */
219                 return 0;
220         return 1;
221 }
222
223 /**
224  * ibmvscsi_free-event_struct: - Changes status of event to "free"
225  * @pool:       event_pool that contains the event
226  * @evt:        srp_event_struct to be modified
227  *
228 */
229 static void free_event_struct(struct event_pool *pool,
230                                        struct srp_event_struct *evt)
231 {
232         if (!valid_event_struct(pool, evt)) {
233                 dev_err(evt->hostdata->dev, "Freeing invalid event_struct %p "
234                         "(not in pool %p)\n", evt, pool->events);
235                 return;
236         }
237         if (atomic_inc_return(&evt->free) != 1) {
238                 dev_err(evt->hostdata->dev, "Freeing event_struct %p "
239                         "which is not in use!\n", evt);
240                 return;
241         }
242 }
243
244 /**
245  * get_evt_struct: - Gets the next free event in pool
246  * @pool:       event_pool that contains the events to be searched
247  *
248  * Returns the next event in "free" state, and NULL if none are free.
249  * Note that no synchronization is done here, we assume the host_lock
250  * will syncrhonze things.
251 */
252 static struct srp_event_struct *get_event_struct(struct event_pool *pool)
253 {
254         int i;
255         int poolsize = pool->size;
256         int offset = pool->next;
257
258         for (i = 0; i < poolsize; i++) {
259                 offset = (offset + 1) % poolsize;
260                 if (!atomic_dec_if_positive(&pool->events[offset].free)) {
261                         pool->next = offset;
262                         return &pool->events[offset];
263                 }
264         }
265
266         printk(KERN_ERR "ibmvscsi: found no event struct in pool!\n");
267         return NULL;
268 }
269
270 /**
271  * init_event_struct: Initialize fields in an event struct that are always 
272  *                    required.
273  * @evt:        The event
274  * @done:       Routine to call when the event is responded to
275  * @format:     SRP or MAD format
276  * @timeout:    timeout value set in the CRQ
277  */
278 static void init_event_struct(struct srp_event_struct *evt_struct,
279                               void (*done) (struct srp_event_struct *),
280                               u8 format,
281                               int timeout)
282 {
283         evt_struct->cmnd = NULL;
284         evt_struct->cmnd_done = NULL;
285         evt_struct->sync_srp = NULL;
286         evt_struct->crq.format = format;
287         evt_struct->crq.timeout = timeout;
288         evt_struct->done = done;
289 }
290
291 /* ------------------------------------------------------------
292  * Routines for receiving SCSI responses from the hosting partition
293  */
294
295 /**
296  * set_srp_direction: Set the fields in the srp related to data
297  *     direction and number of buffers based on the direction in
298  *     the scsi_cmnd and the number of buffers
299  */
300 static void set_srp_direction(struct scsi_cmnd *cmd,
301                               struct srp_cmd *srp_cmd, 
302                               int numbuf)
303 {
304         u8 fmt;
305
306         if (numbuf == 0)
307                 return;
308         
309         if (numbuf == 1)
310                 fmt = SRP_DATA_DESC_DIRECT;
311         else {
312                 fmt = SRP_DATA_DESC_INDIRECT;
313                 numbuf = min(numbuf, MAX_INDIRECT_BUFS);
314
315                 if (cmd->sc_data_direction == DMA_TO_DEVICE)
316                         srp_cmd->data_out_desc_cnt = numbuf;
317                 else
318                         srp_cmd->data_in_desc_cnt = numbuf;
319         }
320
321         if (cmd->sc_data_direction == DMA_TO_DEVICE)
322                 srp_cmd->buf_fmt = fmt << 4;
323         else
324                 srp_cmd->buf_fmt = fmt;
325 }
326
327 /**
328  * unmap_cmd_data: - Unmap data pointed in srp_cmd based on the format
329  * @cmd:        srp_cmd whose additional_data member will be unmapped
330  * @dev:        device for which the memory is mapped
331  *
332 */
333 static void unmap_cmd_data(struct srp_cmd *cmd,
334                            struct srp_event_struct *evt_struct,
335                            struct device *dev)
336 {
337         u8 out_fmt, in_fmt;
338
339         out_fmt = cmd->buf_fmt >> 4;
340         in_fmt = cmd->buf_fmt & ((1U << 4) - 1);
341
342         if (out_fmt == SRP_NO_DATA_DESC && in_fmt == SRP_NO_DATA_DESC)
343                 return;
344
345         if (evt_struct->cmnd)
346                 scsi_dma_unmap(evt_struct->cmnd);
347 }
348
349 static int map_sg_list(struct scsi_cmnd *cmd, int nseg,
350                        struct srp_direct_buf *md)
351 {
352         int i;
353         struct scatterlist *sg;
354         u64 total_length = 0;
355
356         scsi_for_each_sg(cmd, sg, nseg, i) {
357                 struct srp_direct_buf *descr = md + i;
358                 descr->va = sg_dma_address(sg);
359                 descr->len = sg_dma_len(sg);
360                 descr->key = 0;
361                 total_length += sg_dma_len(sg);
362         }
363         return total_length;
364 }
365
366 /**
367  * map_sg_data: - Maps dma for a scatterlist and initializes decriptor fields
368  * @cmd:        Scsi_Cmnd with the scatterlist
369  * @srp_cmd:    srp_cmd that contains the memory descriptor
370  * @dev:        device for which to map dma memory
371  *
372  * Called by map_data_for_srp_cmd() when building srp cmd from scsi cmd.
373  * Returns 1 on success.
374 */
375 static int map_sg_data(struct scsi_cmnd *cmd,
376                        struct srp_event_struct *evt_struct,
377                        struct srp_cmd *srp_cmd, struct device *dev)
378 {
379
380         int sg_mapped;
381         u64 total_length = 0;
382         struct srp_direct_buf *data =
383                 (struct srp_direct_buf *) srp_cmd->add_data;
384         struct srp_indirect_buf *indirect =
385                 (struct srp_indirect_buf *) data;
386
387         sg_mapped = scsi_dma_map(cmd);
388         if (!sg_mapped)
389                 return 1;
390         else if (sg_mapped < 0)
391                 return 0;
392
393         set_srp_direction(cmd, srp_cmd, sg_mapped);
394
395         /* special case; we can use a single direct descriptor */
396         if (sg_mapped == 1) {
397                 map_sg_list(cmd, sg_mapped, data);
398                 return 1;
399         }
400
401         indirect->table_desc.va = 0;
402         indirect->table_desc.len = sg_mapped * sizeof(struct srp_direct_buf);
403         indirect->table_desc.key = 0;
404
405         if (sg_mapped <= MAX_INDIRECT_BUFS) {
406                 total_length = map_sg_list(cmd, sg_mapped,
407                                            &indirect->desc_list[0]);
408                 indirect->len = total_length;
409                 return 1;
410         }
411
412         /* get indirect table */
413         if (!evt_struct->ext_list) {
414                 evt_struct->ext_list = (struct srp_direct_buf *)
415                         dma_alloc_coherent(dev,
416                                            SG_ALL * sizeof(struct srp_direct_buf),
417                                            &evt_struct->ext_list_token, 0);
418                 if (!evt_struct->ext_list) {
419                         if (!firmware_has_feature(FW_FEATURE_CMO))
420                                 sdev_printk(KERN_ERR, cmd->device,
421                                             "Can't allocate memory "
422                                             "for indirect table\n");
423                         scsi_dma_unmap(cmd);
424                         return 0;
425                 }
426         }
427
428         total_length = map_sg_list(cmd, sg_mapped, evt_struct->ext_list);
429
430         indirect->len = total_length;
431         indirect->table_desc.va = evt_struct->ext_list_token;
432         indirect->table_desc.len = sg_mapped * sizeof(indirect->desc_list[0]);
433         memcpy(indirect->desc_list, evt_struct->ext_list,
434                MAX_INDIRECT_BUFS * sizeof(struct srp_direct_buf));
435         return 1;
436 }
437
438 /**
439  * map_data_for_srp_cmd: - Calls functions to map data for srp cmds
440  * @cmd:        struct scsi_cmnd with the memory to be mapped
441  * @srp_cmd:    srp_cmd that contains the memory descriptor
442  * @dev:        dma device for which to map dma memory
443  *
444  * Called by scsi_cmd_to_srp_cmd() when converting scsi cmds to srp cmds 
445  * Returns 1 on success.
446 */
447 static int map_data_for_srp_cmd(struct scsi_cmnd *cmd,
448                                 struct srp_event_struct *evt_struct,
449                                 struct srp_cmd *srp_cmd, struct device *dev)
450 {
451         switch (cmd->sc_data_direction) {
452         case DMA_FROM_DEVICE:
453         case DMA_TO_DEVICE:
454                 break;
455         case DMA_NONE:
456                 return 1;
457         case DMA_BIDIRECTIONAL:
458                 sdev_printk(KERN_ERR, cmd->device,
459                             "Can't map DMA_BIDIRECTIONAL to read/write\n");
460                 return 0;
461         default:
462                 sdev_printk(KERN_ERR, cmd->device,
463                             "Unknown data direction 0x%02x; can't map!\n",
464                             cmd->sc_data_direction);
465                 return 0;
466         }
467
468         return map_sg_data(cmd, evt_struct, srp_cmd, dev);
469 }
470
471 /**
472  * purge_requests: Our virtual adapter just shut down.  purge any sent requests
473  * @hostdata:    the adapter
474  */
475 static void purge_requests(struct ibmvscsi_host_data *hostdata, int error_code)
476 {
477         struct srp_event_struct *evt;
478         unsigned long flags;
479
480         spin_lock_irqsave(hostdata->host->host_lock, flags);
481         while (!list_empty(&hostdata->sent)) {
482                 evt = list_first_entry(&hostdata->sent, struct srp_event_struct, list);
483                 list_del(&evt->list);
484                 del_timer(&evt->timer);
485
486                 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
487                 if (evt->cmnd) {
488                         evt->cmnd->result = (error_code << 16);
489                         unmap_cmd_data(&evt->iu.srp.cmd, evt,
490                                        evt->hostdata->dev);
491                         if (evt->cmnd_done)
492                                 evt->cmnd_done(evt->cmnd);
493                 } else if (evt->done)
494                         evt->done(evt);
495                 free_event_struct(&evt->hostdata->pool, evt);
496                 spin_lock_irqsave(hostdata->host->host_lock, flags);
497         }
498         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
499 }
500
501 /**
502  * ibmvscsi_reset_host - Reset the connection to the server
503  * @hostdata:   struct ibmvscsi_host_data to reset
504 */
505 static void ibmvscsi_reset_host(struct ibmvscsi_host_data *hostdata)
506 {
507         scsi_block_requests(hostdata->host);
508         atomic_set(&hostdata->request_limit, 0);
509
510         purge_requests(hostdata, DID_ERROR);
511         hostdata->reset_crq = 1;
512         wake_up(&hostdata->work_wait_q);
513 }
514
515 /**
516  * ibmvscsi_timeout - Internal command timeout handler
517  * @evt_struct: struct srp_event_struct that timed out
518  *
519  * Called when an internally generated command times out
520 */
521 static void ibmvscsi_timeout(struct srp_event_struct *evt_struct)
522 {
523         struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
524
525         dev_err(hostdata->dev, "Command timed out (%x). Resetting connection\n",
526                 evt_struct->iu.srp.cmd.opcode);
527
528         ibmvscsi_reset_host(hostdata);
529 }
530
531
532 /* ------------------------------------------------------------
533  * Routines for sending and receiving SRPs
534  */
535 /**
536  * ibmvscsi_send_srp_event: - Transforms event to u64 array and calls send_crq()
537  * @evt_struct: evt_struct to be sent
538  * @hostdata:   ibmvscsi_host_data of host
539  * @timeout:    timeout in seconds - 0 means do not time command
540  *
541  * Returns the value returned from ibmvscsi_send_crq(). (Zero for success)
542  * Note that this routine assumes that host_lock is held for synchronization
543 */
544 static int ibmvscsi_send_srp_event(struct srp_event_struct *evt_struct,
545                                    struct ibmvscsi_host_data *hostdata,
546                                    unsigned long timeout)
547 {
548         u64 *crq_as_u64 = (u64 *) &evt_struct->crq;
549         int request_status = 0;
550         int rc;
551
552         /* If we have exhausted our request limit, just fail this request,
553          * unless it is for a reset or abort.
554          * Note that there are rare cases involving driver generated requests 
555          * (such as task management requests) that the mid layer may think we
556          * can handle more requests (can_queue) when we actually can't
557          */
558         if (evt_struct->crq.format == VIOSRP_SRP_FORMAT) {
559                 request_status =
560                         atomic_dec_if_positive(&hostdata->request_limit);
561                 /* If request limit was -1 when we started, it is now even
562                  * less than that
563                  */
564                 if (request_status < -1)
565                         goto send_error;
566                 /* Otherwise, we may have run out of requests. */
567                 /* If request limit was 0 when we started the adapter is in the
568                  * process of performing a login with the server adapter, or
569                  * we may have run out of requests.
570                  */
571                 else if (request_status == -1 &&
572                          evt_struct->iu.srp.login_req.opcode != SRP_LOGIN_REQ)
573                         goto send_busy;
574                 /* Abort and reset calls should make it through.
575                  * Nothing except abort and reset should use the last two
576                  * slots unless we had two or less to begin with.
577                  */
578                 else if (request_status < 2 &&
579                          evt_struct->iu.srp.cmd.opcode != SRP_TSK_MGMT) {
580                         /* In the case that we have less than two requests
581                          * available, check the server limit as a combination
582                          * of the request limit and the number of requests
583                          * in-flight (the size of the send list).  If the
584                          * server limit is greater than 2, return busy so
585                          * that the last two are reserved for reset and abort.
586                          */
587                         int server_limit = request_status;
588                         struct srp_event_struct *tmp_evt;
589
590                         list_for_each_entry(tmp_evt, &hostdata->sent, list) {
591                                 server_limit++;
592                         }
593
594                         if (server_limit > 2)
595                                 goto send_busy;
596                 }
597         }
598
599         /* Copy the IU into the transfer area */
600         *evt_struct->xfer_iu = evt_struct->iu;
601         evt_struct->xfer_iu->srp.rsp.tag = (u64)evt_struct;
602
603         /* Add this to the sent list.  We need to do this 
604          * before we actually send 
605          * in case it comes back REALLY fast
606          */
607         list_add_tail(&evt_struct->list, &hostdata->sent);
608
609         init_timer(&evt_struct->timer);
610         if (timeout) {
611                 evt_struct->timer.data = (unsigned long) evt_struct;
612                 evt_struct->timer.expires = jiffies + (timeout * HZ);
613                 evt_struct->timer.function = (void (*)(unsigned long))ibmvscsi_timeout;
614                 add_timer(&evt_struct->timer);
615         }
616
617         if ((rc =
618              ibmvscsi_ops->send_crq(hostdata, crq_as_u64[0], crq_as_u64[1])) != 0) {
619                 list_del(&evt_struct->list);
620                 del_timer(&evt_struct->timer);
621
622                 /* If send_crq returns H_CLOSED, return SCSI_MLQUEUE_HOST_BUSY.
623                  * Firmware will send a CRQ with a transport event (0xFF) to
624                  * tell this client what has happened to the transport.  This
625                  * will be handled in ibmvscsi_handle_crq()
626                  */
627                 if (rc == H_CLOSED) {
628                         dev_warn(hostdata->dev, "send warning. "
629                                  "Receive queue closed, will retry.\n");
630                         goto send_busy;
631                 }
632                 dev_err(hostdata->dev, "send error %d\n", rc);
633                 atomic_inc(&hostdata->request_limit);
634                 goto send_error;
635         }
636
637         return 0;
638
639  send_busy:
640         unmap_cmd_data(&evt_struct->iu.srp.cmd, evt_struct, hostdata->dev);
641
642         free_event_struct(&hostdata->pool, evt_struct);
643         if (request_status != -1)
644                 atomic_inc(&hostdata->request_limit);
645         return SCSI_MLQUEUE_HOST_BUSY;
646
647  send_error:
648         unmap_cmd_data(&evt_struct->iu.srp.cmd, evt_struct, hostdata->dev);
649
650         if (evt_struct->cmnd != NULL) {
651                 evt_struct->cmnd->result = DID_ERROR << 16;
652                 evt_struct->cmnd_done(evt_struct->cmnd);
653         } else if (evt_struct->done)
654                 evt_struct->done(evt_struct);
655
656         free_event_struct(&hostdata->pool, evt_struct);
657         return 0;
658 }
659
660 /**
661  * handle_cmd_rsp: -  Handle responses from commands
662  * @evt_struct: srp_event_struct to be handled
663  *
664  * Used as a callback by when sending scsi cmds.
665  * Gets called by ibmvscsi_handle_crq()
666 */
667 static void handle_cmd_rsp(struct srp_event_struct *evt_struct)
668 {
669         struct srp_rsp *rsp = &evt_struct->xfer_iu->srp.rsp;
670         struct scsi_cmnd *cmnd = evt_struct->cmnd;
671
672         if (unlikely(rsp->opcode != SRP_RSP)) {
673                 if (printk_ratelimit())
674                         dev_warn(evt_struct->hostdata->dev,
675                                  "bad SRP RSP type %d\n", rsp->opcode);
676         }
677         
678         if (cmnd) {
679                 cmnd->result |= rsp->status;
680                 if (((cmnd->result >> 1) & 0x1f) == CHECK_CONDITION)
681                         memcpy(cmnd->sense_buffer,
682                                rsp->data,
683                                rsp->sense_data_len);
684                 unmap_cmd_data(&evt_struct->iu.srp.cmd, 
685                                evt_struct, 
686                                evt_struct->hostdata->dev);
687
688                 if (rsp->flags & SRP_RSP_FLAG_DOOVER)
689                         scsi_set_resid(cmnd, rsp->data_out_res_cnt);
690                 else if (rsp->flags & SRP_RSP_FLAG_DIOVER)
691                         scsi_set_resid(cmnd, rsp->data_in_res_cnt);
692         }
693
694         if (evt_struct->cmnd_done)
695                 evt_struct->cmnd_done(cmnd);
696 }
697
698 /**
699  * lun_from_dev: - Returns the lun of the scsi device
700  * @dev:        struct scsi_device
701  *
702 */
703 static inline u16 lun_from_dev(struct scsi_device *dev)
704 {
705         return (0x2 << 14) | (dev->id << 8) | (dev->channel << 5) | dev->lun;
706 }
707
708 /**
709  * ibmvscsi_queue: - The queuecommand function of the scsi template 
710  * @cmd:        struct scsi_cmnd to be executed
711  * @done:       Callback function to be called when cmd is completed
712 */
713 static int ibmvscsi_queuecommand(struct scsi_cmnd *cmnd,
714                                  void (*done) (struct scsi_cmnd *))
715 {
716         struct srp_cmd *srp_cmd;
717         struct srp_event_struct *evt_struct;
718         struct srp_indirect_buf *indirect;
719         struct ibmvscsi_host_data *hostdata = shost_priv(cmnd->device->host);
720         u16 lun = lun_from_dev(cmnd->device);
721         u8 out_fmt, in_fmt;
722
723         cmnd->result = (DID_OK << 16);
724         evt_struct = get_event_struct(&hostdata->pool);
725         if (!evt_struct)
726                 return SCSI_MLQUEUE_HOST_BUSY;
727
728         /* Set up the actual SRP IU */
729         srp_cmd = &evt_struct->iu.srp.cmd;
730         memset(srp_cmd, 0x00, SRP_MAX_IU_LEN);
731         srp_cmd->opcode = SRP_CMD;
732         memcpy(srp_cmd->cdb, cmnd->cmnd, sizeof(srp_cmd->cdb));
733         srp_cmd->lun = ((u64) lun) << 48;
734
735         if (!map_data_for_srp_cmd(cmnd, evt_struct, srp_cmd, hostdata->dev)) {
736                 if (!firmware_has_feature(FW_FEATURE_CMO))
737                         sdev_printk(KERN_ERR, cmnd->device,
738                                     "couldn't convert cmd to srp_cmd\n");
739                 free_event_struct(&hostdata->pool, evt_struct);
740                 return SCSI_MLQUEUE_HOST_BUSY;
741         }
742
743         init_event_struct(evt_struct,
744                           handle_cmd_rsp,
745                           VIOSRP_SRP_FORMAT,
746                           cmnd->request->timeout/HZ);
747
748         evt_struct->cmnd = cmnd;
749         evt_struct->cmnd_done = done;
750
751         /* Fix up dma address of the buffer itself */
752         indirect = (struct srp_indirect_buf *) srp_cmd->add_data;
753         out_fmt = srp_cmd->buf_fmt >> 4;
754         in_fmt = srp_cmd->buf_fmt & ((1U << 4) - 1);
755         if ((in_fmt == SRP_DATA_DESC_INDIRECT ||
756              out_fmt == SRP_DATA_DESC_INDIRECT) &&
757             indirect->table_desc.va == 0) {
758                 indirect->table_desc.va = evt_struct->crq.IU_data_ptr +
759                         offsetof(struct srp_cmd, add_data) +
760                         offsetof(struct srp_indirect_buf, desc_list);
761         }
762
763         return ibmvscsi_send_srp_event(evt_struct, hostdata, 0);
764 }
765
766 /* ------------------------------------------------------------
767  * Routines for driver initialization
768  */
769
770 /**
771  * map_persist_bufs: - Pre-map persistent data for adapter logins
772  * @hostdata:   ibmvscsi_host_data of host
773  *
774  * Map the capabilities and adapter info DMA buffers to avoid runtime failures.
775  * Return 1 on error, 0 on success.
776  */
777 static int map_persist_bufs(struct ibmvscsi_host_data *hostdata)
778 {
779
780         hostdata->caps_addr = dma_map_single(hostdata->dev, &hostdata->caps,
781                                              sizeof(hostdata->caps), DMA_BIDIRECTIONAL);
782
783         if (dma_mapping_error(hostdata->dev, hostdata->caps_addr)) {
784                 dev_err(hostdata->dev, "Unable to map capabilities buffer!\n");
785                 return 1;
786         }
787
788         hostdata->adapter_info_addr = dma_map_single(hostdata->dev,
789                                                      &hostdata->madapter_info,
790                                                      sizeof(hostdata->madapter_info),
791                                                      DMA_BIDIRECTIONAL);
792         if (dma_mapping_error(hostdata->dev, hostdata->adapter_info_addr)) {
793                 dev_err(hostdata->dev, "Unable to map adapter info buffer!\n");
794                 dma_unmap_single(hostdata->dev, hostdata->caps_addr,
795                                  sizeof(hostdata->caps), DMA_BIDIRECTIONAL);
796                 return 1;
797         }
798
799         return 0;
800 }
801
802 /**
803  * unmap_persist_bufs: - Unmap persistent data needed for adapter logins
804  * @hostdata:   ibmvscsi_host_data of host
805  *
806  * Unmap the capabilities and adapter info DMA buffers
807  */
808 static void unmap_persist_bufs(struct ibmvscsi_host_data *hostdata)
809 {
810         dma_unmap_single(hostdata->dev, hostdata->caps_addr,
811                          sizeof(hostdata->caps), DMA_BIDIRECTIONAL);
812
813         dma_unmap_single(hostdata->dev, hostdata->adapter_info_addr,
814                          sizeof(hostdata->madapter_info), DMA_BIDIRECTIONAL);
815 }
816
817 /**
818  * login_rsp: - Handle response to SRP login request
819  * @evt_struct: srp_event_struct with the response
820  *
821  * Used as a "done" callback by when sending srp_login. Gets called
822  * by ibmvscsi_handle_crq()
823 */
824 static void login_rsp(struct srp_event_struct *evt_struct)
825 {
826         struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
827         switch (evt_struct->xfer_iu->srp.login_rsp.opcode) {
828         case SRP_LOGIN_RSP:     /* it worked! */
829                 break;
830         case SRP_LOGIN_REJ:     /* refused! */
831                 dev_info(hostdata->dev, "SRP_LOGIN_REJ reason %u\n",
832                          evt_struct->xfer_iu->srp.login_rej.reason);
833                 /* Login failed.  */
834                 atomic_set(&hostdata->request_limit, -1);
835                 return;
836         default:
837                 dev_err(hostdata->dev, "Invalid login response typecode 0x%02x!\n",
838                         evt_struct->xfer_iu->srp.login_rsp.opcode);
839                 /* Login failed.  */
840                 atomic_set(&hostdata->request_limit, -1);
841                 return;
842         }
843
844         dev_info(hostdata->dev, "SRP_LOGIN succeeded\n");
845         hostdata->client_migrated = 0;
846
847         /* Now we know what the real request-limit is.
848          * This value is set rather than added to request_limit because
849          * request_limit could have been set to -1 by this client.
850          */
851         atomic_set(&hostdata->request_limit,
852                    evt_struct->xfer_iu->srp.login_rsp.req_lim_delta);
853
854         /* If we had any pending I/Os, kick them */
855         scsi_unblock_requests(hostdata->host);
856 }
857
858 /**
859  * send_srp_login: - Sends the srp login
860  * @hostdata:   ibmvscsi_host_data of host
861  *
862  * Returns zero if successful.
863 */
864 static int send_srp_login(struct ibmvscsi_host_data *hostdata)
865 {
866         int rc;
867         unsigned long flags;
868         struct srp_login_req *login;
869         struct srp_event_struct *evt_struct = get_event_struct(&hostdata->pool);
870
871         BUG_ON(!evt_struct);
872         init_event_struct(evt_struct, login_rsp,
873                           VIOSRP_SRP_FORMAT, login_timeout);
874
875         login = &evt_struct->iu.srp.login_req;
876         memset(login, 0, sizeof(*login));
877         login->opcode = SRP_LOGIN_REQ;
878         login->req_it_iu_len = sizeof(union srp_iu);
879         login->req_buf_fmt = SRP_BUF_FORMAT_DIRECT | SRP_BUF_FORMAT_INDIRECT;
880
881         spin_lock_irqsave(hostdata->host->host_lock, flags);
882         /* Start out with a request limit of 0, since this is negotiated in
883          * the login request we are just sending and login requests always
884          * get sent by the driver regardless of request_limit.
885          */
886         atomic_set(&hostdata->request_limit, 0);
887
888         rc = ibmvscsi_send_srp_event(evt_struct, hostdata, login_timeout * 2);
889         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
890         dev_info(hostdata->dev, "sent SRP login\n");
891         return rc;
892 };
893
894 /**
895  * capabilities_rsp: - Handle response to MAD adapter capabilities request
896  * @evt_struct: srp_event_struct with the response
897  *
898  * Used as a "done" callback by when sending adapter_info.
899  */
900 static void capabilities_rsp(struct srp_event_struct *evt_struct)
901 {
902         struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
903
904         if (evt_struct->xfer_iu->mad.capabilities.common.status) {
905                 dev_err(hostdata->dev, "error 0x%X getting capabilities info\n",
906                         evt_struct->xfer_iu->mad.capabilities.common.status);
907         } else {
908                 if (hostdata->caps.migration.common.server_support != SERVER_SUPPORTS_CAP)
909                         dev_info(hostdata->dev, "Partition migration not supported\n");
910
911                 if (client_reserve) {
912                         if (hostdata->caps.reserve.common.server_support ==
913                             SERVER_SUPPORTS_CAP)
914                                 dev_info(hostdata->dev, "Client reserve enabled\n");
915                         else
916                                 dev_info(hostdata->dev, "Client reserve not supported\n");
917                 }
918         }
919
920         send_srp_login(hostdata);
921 }
922
923 /**
924  * send_mad_capabilities: - Sends the mad capabilities request
925  *      and stores the result so it can be retrieved with
926  * @hostdata:   ibmvscsi_host_data of host
927  */
928 static void send_mad_capabilities(struct ibmvscsi_host_data *hostdata)
929 {
930         struct viosrp_capabilities *req;
931         struct srp_event_struct *evt_struct;
932         unsigned long flags;
933         struct device_node *of_node = hostdata->dev->of_node;
934         const char *location;
935
936         evt_struct = get_event_struct(&hostdata->pool);
937         BUG_ON(!evt_struct);
938
939         init_event_struct(evt_struct, capabilities_rsp,
940                           VIOSRP_MAD_FORMAT, info_timeout);
941
942         req = &evt_struct->iu.mad.capabilities;
943         memset(req, 0, sizeof(*req));
944
945         hostdata->caps.flags = CAP_LIST_SUPPORTED;
946         if (hostdata->client_migrated)
947                 hostdata->caps.flags |= CLIENT_MIGRATED;
948
949         strncpy(hostdata->caps.name, dev_name(&hostdata->host->shost_gendev),
950                 sizeof(hostdata->caps.name));
951         hostdata->caps.name[sizeof(hostdata->caps.name) - 1] = '\0';
952
953         location = of_get_property(of_node, "ibm,loc-code", NULL);
954         location = location ? location : dev_name(hostdata->dev);
955         strncpy(hostdata->caps.loc, location, sizeof(hostdata->caps.loc));
956         hostdata->caps.loc[sizeof(hostdata->caps.loc) - 1] = '\0';
957
958         req->common.type = VIOSRP_CAPABILITIES_TYPE;
959         req->buffer = hostdata->caps_addr;
960
961         hostdata->caps.migration.common.cap_type = MIGRATION_CAPABILITIES;
962         hostdata->caps.migration.common.length = sizeof(hostdata->caps.migration);
963         hostdata->caps.migration.common.server_support = SERVER_SUPPORTS_CAP;
964         hostdata->caps.migration.ecl = 1;
965
966         if (client_reserve) {
967                 hostdata->caps.reserve.common.cap_type = RESERVATION_CAPABILITIES;
968                 hostdata->caps.reserve.common.length = sizeof(hostdata->caps.reserve);
969                 hostdata->caps.reserve.common.server_support = SERVER_SUPPORTS_CAP;
970                 hostdata->caps.reserve.type = CLIENT_RESERVE_SCSI_2;
971                 req->common.length = sizeof(hostdata->caps);
972         } else
973                 req->common.length = sizeof(hostdata->caps) - sizeof(hostdata->caps.reserve);
974
975         spin_lock_irqsave(hostdata->host->host_lock, flags);
976         if (ibmvscsi_send_srp_event(evt_struct, hostdata, info_timeout * 2))
977                 dev_err(hostdata->dev, "couldn't send CAPABILITIES_REQ!\n");
978         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
979 };
980
981 /**
982  * fast_fail_rsp: - Handle response to MAD enable fast fail
983  * @evt_struct: srp_event_struct with the response
984  *
985  * Used as a "done" callback by when sending enable fast fail. Gets called
986  * by ibmvscsi_handle_crq()
987  */
988 static void fast_fail_rsp(struct srp_event_struct *evt_struct)
989 {
990         struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
991         u8 status = evt_struct->xfer_iu->mad.fast_fail.common.status;
992
993         if (status == VIOSRP_MAD_NOT_SUPPORTED)
994                 dev_err(hostdata->dev, "fast_fail not supported in server\n");
995         else if (status == VIOSRP_MAD_FAILED)
996                 dev_err(hostdata->dev, "fast_fail request failed\n");
997         else if (status != VIOSRP_MAD_SUCCESS)
998                 dev_err(hostdata->dev, "error 0x%X enabling fast_fail\n", status);
999
1000         send_mad_capabilities(hostdata);
1001 }
1002
1003 /**
1004  * init_host - Start host initialization
1005  * @hostdata:   ibmvscsi_host_data of host
1006  *
1007  * Returns zero if successful.
1008  */
1009 static int enable_fast_fail(struct ibmvscsi_host_data *hostdata)
1010 {
1011         int rc;
1012         unsigned long flags;
1013         struct viosrp_fast_fail *fast_fail_mad;
1014         struct srp_event_struct *evt_struct;
1015
1016         if (!fast_fail) {
1017                 send_mad_capabilities(hostdata);
1018                 return 0;
1019         }
1020
1021         evt_struct = get_event_struct(&hostdata->pool);
1022         BUG_ON(!evt_struct);
1023
1024         init_event_struct(evt_struct, fast_fail_rsp, VIOSRP_MAD_FORMAT, info_timeout);
1025
1026         fast_fail_mad = &evt_struct->iu.mad.fast_fail;
1027         memset(fast_fail_mad, 0, sizeof(*fast_fail_mad));
1028         fast_fail_mad->common.type = VIOSRP_ENABLE_FAST_FAIL;
1029         fast_fail_mad->common.length = sizeof(*fast_fail_mad);
1030
1031         spin_lock_irqsave(hostdata->host->host_lock, flags);
1032         rc = ibmvscsi_send_srp_event(evt_struct, hostdata, info_timeout * 2);
1033         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1034         return rc;
1035 }
1036
1037 /**
1038  * adapter_info_rsp: - Handle response to MAD adapter info request
1039  * @evt_struct: srp_event_struct with the response
1040  *
1041  * Used as a "done" callback by when sending adapter_info. Gets called
1042  * by ibmvscsi_handle_crq()
1043 */
1044 static void adapter_info_rsp(struct srp_event_struct *evt_struct)
1045 {
1046         struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
1047
1048         if (evt_struct->xfer_iu->mad.adapter_info.common.status) {
1049                 dev_err(hostdata->dev, "error %d getting adapter info\n",
1050                         evt_struct->xfer_iu->mad.adapter_info.common.status);
1051         } else {
1052                 dev_info(hostdata->dev, "host srp version: %s, "
1053                          "host partition %s (%d), OS %d, max io %u\n",
1054                          hostdata->madapter_info.srp_version,
1055                          hostdata->madapter_info.partition_name,
1056                          hostdata->madapter_info.partition_number,
1057                          hostdata->madapter_info.os_type,
1058                          hostdata->madapter_info.port_max_txu[0]);
1059                 
1060                 if (hostdata->madapter_info.port_max_txu[0]) 
1061                         hostdata->host->max_sectors = 
1062                                 hostdata->madapter_info.port_max_txu[0] >> 9;
1063                 
1064                 if (hostdata->madapter_info.os_type == 3 &&
1065                     strcmp(hostdata->madapter_info.srp_version, "1.6a") <= 0) {
1066                         dev_err(hostdata->dev, "host (Ver. %s) doesn't support large transfers\n",
1067                                 hostdata->madapter_info.srp_version);
1068                         dev_err(hostdata->dev, "limiting scatterlists to %d\n",
1069                                 MAX_INDIRECT_BUFS);
1070                         hostdata->host->sg_tablesize = MAX_INDIRECT_BUFS;
1071                 }
1072
1073                 if (hostdata->madapter_info.os_type == 3) {
1074                         enable_fast_fail(hostdata);
1075                         return;
1076                 }
1077         }
1078
1079         send_srp_login(hostdata);
1080 }
1081
1082 /**
1083  * send_mad_adapter_info: - Sends the mad adapter info request
1084  *      and stores the result so it can be retrieved with
1085  *      sysfs.  We COULD consider causing a failure if the
1086  *      returned SRP version doesn't match ours.
1087  * @hostdata:   ibmvscsi_host_data of host
1088  * 
1089  * Returns zero if successful.
1090 */
1091 static void send_mad_adapter_info(struct ibmvscsi_host_data *hostdata)
1092 {
1093         struct viosrp_adapter_info *req;
1094         struct srp_event_struct *evt_struct;
1095         unsigned long flags;
1096
1097         evt_struct = get_event_struct(&hostdata->pool);
1098         BUG_ON(!evt_struct);
1099
1100         init_event_struct(evt_struct,
1101                           adapter_info_rsp,
1102                           VIOSRP_MAD_FORMAT,
1103                           info_timeout);
1104         
1105         req = &evt_struct->iu.mad.adapter_info;
1106         memset(req, 0x00, sizeof(*req));
1107         
1108         req->common.type = VIOSRP_ADAPTER_INFO_TYPE;
1109         req->common.length = sizeof(hostdata->madapter_info);
1110         req->buffer = hostdata->adapter_info_addr;
1111
1112         spin_lock_irqsave(hostdata->host->host_lock, flags);
1113         if (ibmvscsi_send_srp_event(evt_struct, hostdata, info_timeout * 2))
1114                 dev_err(hostdata->dev, "couldn't send ADAPTER_INFO_REQ!\n");
1115         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1116 };
1117
1118 /**
1119  * init_adapter: Start virtual adapter initialization sequence
1120  *
1121  */
1122 static void init_adapter(struct ibmvscsi_host_data *hostdata)
1123 {
1124         send_mad_adapter_info(hostdata);
1125 }
1126
1127 /**
1128  * sync_completion: Signal that a synchronous command has completed
1129  * Note that after returning from this call, the evt_struct is freed.
1130  * the caller waiting on this completion shouldn't touch the evt_struct
1131  * again.
1132  */
1133 static void sync_completion(struct srp_event_struct *evt_struct)
1134 {
1135         /* copy the response back */
1136         if (evt_struct->sync_srp)
1137                 *evt_struct->sync_srp = *evt_struct->xfer_iu;
1138         
1139         complete(&evt_struct->comp);
1140 }
1141
1142 /**
1143  * ibmvscsi_abort: Abort a command...from scsi host template
1144  * send this over to the server and wait synchronously for the response
1145  */
1146 static int ibmvscsi_eh_abort_handler(struct scsi_cmnd *cmd)
1147 {
1148         struct ibmvscsi_host_data *hostdata = shost_priv(cmd->device->host);
1149         struct srp_tsk_mgmt *tsk_mgmt;
1150         struct srp_event_struct *evt;
1151         struct srp_event_struct *tmp_evt, *found_evt;
1152         union viosrp_iu srp_rsp;
1153         int rsp_rc;
1154         unsigned long flags;
1155         u16 lun = lun_from_dev(cmd->device);
1156         unsigned long wait_switch = 0;
1157
1158         /* First, find this command in our sent list so we can figure
1159          * out the correct tag
1160          */
1161         spin_lock_irqsave(hostdata->host->host_lock, flags);
1162         wait_switch = jiffies + (init_timeout * HZ);
1163         do {
1164                 found_evt = NULL;
1165                 list_for_each_entry(tmp_evt, &hostdata->sent, list) {
1166                         if (tmp_evt->cmnd == cmd) {
1167                                 found_evt = tmp_evt;
1168                                 break;
1169                         }
1170                 }
1171
1172                 if (!found_evt) {
1173                         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1174                         return SUCCESS;
1175                 }
1176
1177                 evt = get_event_struct(&hostdata->pool);
1178                 if (evt == NULL) {
1179                         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1180                         sdev_printk(KERN_ERR, cmd->device,
1181                                 "failed to allocate abort event\n");
1182                         return FAILED;
1183                 }
1184         
1185                 init_event_struct(evt,
1186                                   sync_completion,
1187                                   VIOSRP_SRP_FORMAT,
1188                                   abort_timeout);
1189
1190                 tsk_mgmt = &evt->iu.srp.tsk_mgmt;
1191         
1192                 /* Set up an abort SRP command */
1193                 memset(tsk_mgmt, 0x00, sizeof(*tsk_mgmt));
1194                 tsk_mgmt->opcode = SRP_TSK_MGMT;
1195                 tsk_mgmt->lun = ((u64) lun) << 48;
1196                 tsk_mgmt->tsk_mgmt_func = SRP_TSK_ABORT_TASK;
1197                 tsk_mgmt->task_tag = (u64) found_evt;
1198
1199                 evt->sync_srp = &srp_rsp;
1200
1201                 init_completion(&evt->comp);
1202                 rsp_rc = ibmvscsi_send_srp_event(evt, hostdata, abort_timeout * 2);
1203
1204                 if (rsp_rc != SCSI_MLQUEUE_HOST_BUSY)
1205                         break;
1206
1207                 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1208                 msleep(10);
1209                 spin_lock_irqsave(hostdata->host->host_lock, flags);
1210         } while (time_before(jiffies, wait_switch));
1211
1212         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1213
1214         if (rsp_rc != 0) {
1215                 sdev_printk(KERN_ERR, cmd->device,
1216                             "failed to send abort() event. rc=%d\n", rsp_rc);
1217                 return FAILED;
1218         }
1219
1220         sdev_printk(KERN_INFO, cmd->device,
1221                     "aborting command. lun 0x%llx, tag 0x%llx\n",
1222                     (((u64) lun) << 48), (u64) found_evt);
1223
1224         wait_for_completion(&evt->comp);
1225
1226         /* make sure we got a good response */
1227         if (unlikely(srp_rsp.srp.rsp.opcode != SRP_RSP)) {
1228                 if (printk_ratelimit())
1229                         sdev_printk(KERN_WARNING, cmd->device, "abort bad SRP RSP type %d\n",
1230                                     srp_rsp.srp.rsp.opcode);
1231                 return FAILED;
1232         }
1233
1234         if (srp_rsp.srp.rsp.flags & SRP_RSP_FLAG_RSPVALID)
1235                 rsp_rc = *((int *)srp_rsp.srp.rsp.data);
1236         else
1237                 rsp_rc = srp_rsp.srp.rsp.status;
1238
1239         if (rsp_rc) {
1240                 if (printk_ratelimit())
1241                         sdev_printk(KERN_WARNING, cmd->device,
1242                                     "abort code %d for task tag 0x%llx\n",
1243                                     rsp_rc, tsk_mgmt->task_tag);
1244                 return FAILED;
1245         }
1246
1247         /* Because we dropped the spinlock above, it's possible
1248          * The event is no longer in our list.  Make sure it didn't
1249          * complete while we were aborting
1250          */
1251         spin_lock_irqsave(hostdata->host->host_lock, flags);
1252         found_evt = NULL;
1253         list_for_each_entry(tmp_evt, &hostdata->sent, list) {
1254                 if (tmp_evt->cmnd == cmd) {
1255                         found_evt = tmp_evt;
1256                         break;
1257                 }
1258         }
1259
1260         if (found_evt == NULL) {
1261                 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1262                 sdev_printk(KERN_INFO, cmd->device, "aborted task tag 0x%llx completed\n",
1263                             tsk_mgmt->task_tag);
1264                 return SUCCESS;
1265         }
1266
1267         sdev_printk(KERN_INFO, cmd->device, "successfully aborted task tag 0x%llx\n",
1268                     tsk_mgmt->task_tag);
1269
1270         cmd->result = (DID_ABORT << 16);
1271         list_del(&found_evt->list);
1272         unmap_cmd_data(&found_evt->iu.srp.cmd, found_evt,
1273                        found_evt->hostdata->dev);
1274         free_event_struct(&found_evt->hostdata->pool, found_evt);
1275         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1276         atomic_inc(&hostdata->request_limit);
1277         return SUCCESS;
1278 }
1279
1280 /**
1281  * ibmvscsi_eh_device_reset_handler: Reset a single LUN...from scsi host 
1282  * template send this over to the server and wait synchronously for the 
1283  * response
1284  */
1285 static int ibmvscsi_eh_device_reset_handler(struct scsi_cmnd *cmd)
1286 {
1287         struct ibmvscsi_host_data *hostdata = shost_priv(cmd->device->host);
1288         struct srp_tsk_mgmt *tsk_mgmt;
1289         struct srp_event_struct *evt;
1290         struct srp_event_struct *tmp_evt, *pos;
1291         union viosrp_iu srp_rsp;
1292         int rsp_rc;
1293         unsigned long flags;
1294         u16 lun = lun_from_dev(cmd->device);
1295         unsigned long wait_switch = 0;
1296
1297         spin_lock_irqsave(hostdata->host->host_lock, flags);
1298         wait_switch = jiffies + (init_timeout * HZ);
1299         do {
1300                 evt = get_event_struct(&hostdata->pool);
1301                 if (evt == NULL) {
1302                         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1303                         sdev_printk(KERN_ERR, cmd->device,
1304                                 "failed to allocate reset event\n");
1305                         return FAILED;
1306                 }
1307         
1308                 init_event_struct(evt,
1309                                   sync_completion,
1310                                   VIOSRP_SRP_FORMAT,
1311                                   reset_timeout);
1312
1313                 tsk_mgmt = &evt->iu.srp.tsk_mgmt;
1314
1315                 /* Set up a lun reset SRP command */
1316                 memset(tsk_mgmt, 0x00, sizeof(*tsk_mgmt));
1317                 tsk_mgmt->opcode = SRP_TSK_MGMT;
1318                 tsk_mgmt->lun = ((u64) lun) << 48;
1319                 tsk_mgmt->tsk_mgmt_func = SRP_TSK_LUN_RESET;
1320
1321                 evt->sync_srp = &srp_rsp;
1322
1323                 init_completion(&evt->comp);
1324                 rsp_rc = ibmvscsi_send_srp_event(evt, hostdata, reset_timeout * 2);
1325
1326                 if (rsp_rc != SCSI_MLQUEUE_HOST_BUSY)
1327                         break;
1328
1329                 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1330                 msleep(10);
1331                 spin_lock_irqsave(hostdata->host->host_lock, flags);
1332         } while (time_before(jiffies, wait_switch));
1333
1334         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1335
1336         if (rsp_rc != 0) {
1337                 sdev_printk(KERN_ERR, cmd->device,
1338                             "failed to send reset event. rc=%d\n", rsp_rc);
1339                 return FAILED;
1340         }
1341
1342         sdev_printk(KERN_INFO, cmd->device, "resetting device. lun 0x%llx\n",
1343                     (((u64) lun) << 48));
1344
1345         wait_for_completion(&evt->comp);
1346
1347         /* make sure we got a good response */
1348         if (unlikely(srp_rsp.srp.rsp.opcode != SRP_RSP)) {
1349                 if (printk_ratelimit())
1350                         sdev_printk(KERN_WARNING, cmd->device, "reset bad SRP RSP type %d\n",
1351                                     srp_rsp.srp.rsp.opcode);
1352                 return FAILED;
1353         }
1354
1355         if (srp_rsp.srp.rsp.flags & SRP_RSP_FLAG_RSPVALID)
1356                 rsp_rc = *((int *)srp_rsp.srp.rsp.data);
1357         else
1358                 rsp_rc = srp_rsp.srp.rsp.status;
1359
1360         if (rsp_rc) {
1361                 if (printk_ratelimit())
1362                         sdev_printk(KERN_WARNING, cmd->device,
1363                                     "reset code %d for task tag 0x%llx\n",
1364                                     rsp_rc, tsk_mgmt->task_tag);
1365                 return FAILED;
1366         }
1367
1368         /* We need to find all commands for this LUN that have not yet been
1369          * responded to, and fail them with DID_RESET
1370          */
1371         spin_lock_irqsave(hostdata->host->host_lock, flags);
1372         list_for_each_entry_safe(tmp_evt, pos, &hostdata->sent, list) {
1373                 if ((tmp_evt->cmnd) && (tmp_evt->cmnd->device == cmd->device)) {
1374                         if (tmp_evt->cmnd)
1375                                 tmp_evt->cmnd->result = (DID_RESET << 16);
1376                         list_del(&tmp_evt->list);
1377                         unmap_cmd_data(&tmp_evt->iu.srp.cmd, tmp_evt,
1378                                        tmp_evt->hostdata->dev);
1379                         free_event_struct(&tmp_evt->hostdata->pool,
1380                                                    tmp_evt);
1381                         atomic_inc(&hostdata->request_limit);
1382                         if (tmp_evt->cmnd_done)
1383                                 tmp_evt->cmnd_done(tmp_evt->cmnd);
1384                         else if (tmp_evt->done)
1385                                 tmp_evt->done(tmp_evt);
1386                 }
1387         }
1388         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1389         return SUCCESS;
1390 }
1391
1392 /**
1393  * ibmvscsi_eh_host_reset_handler - Reset the connection to the server
1394  * @cmd:        struct scsi_cmnd having problems
1395 */
1396 static int ibmvscsi_eh_host_reset_handler(struct scsi_cmnd *cmd)
1397 {
1398         unsigned long wait_switch = 0;
1399         struct ibmvscsi_host_data *hostdata = shost_priv(cmd->device->host);
1400
1401         dev_err(hostdata->dev, "Resetting connection due to error recovery\n");
1402
1403         ibmvscsi_reset_host(hostdata);
1404
1405         for (wait_switch = jiffies + (init_timeout * HZ);
1406              time_before(jiffies, wait_switch) &&
1407                      atomic_read(&hostdata->request_limit) < 2;) {
1408
1409                 msleep(10);
1410         }
1411
1412         if (atomic_read(&hostdata->request_limit) <= 0)
1413                 return FAILED;
1414
1415         return SUCCESS;
1416 }
1417
1418 /**
1419  * ibmvscsi_handle_crq: - Handles and frees received events in the CRQ
1420  * @crq:        Command/Response queue
1421  * @hostdata:   ibmvscsi_host_data of host
1422  *
1423 */
1424 void ibmvscsi_handle_crq(struct viosrp_crq *crq,
1425                          struct ibmvscsi_host_data *hostdata)
1426 {
1427         long rc;
1428         unsigned long flags;
1429         struct srp_event_struct *evt_struct =
1430             (struct srp_event_struct *)crq->IU_data_ptr;
1431         switch (crq->valid) {
1432         case 0xC0:              /* initialization */
1433                 switch (crq->format) {
1434                 case 0x01:      /* Initialization message */
1435                         dev_info(hostdata->dev, "partner initialized\n");
1436                         /* Send back a response */
1437                         if ((rc = ibmvscsi_ops->send_crq(hostdata,
1438                                                          0xC002000000000000LL, 0)) == 0) {
1439                                 /* Now login */
1440                                 init_adapter(hostdata);
1441                         } else {
1442                                 dev_err(hostdata->dev, "Unable to send init rsp. rc=%ld\n", rc);
1443                         }
1444
1445                         break;
1446                 case 0x02:      /* Initialization response */
1447                         dev_info(hostdata->dev, "partner initialization complete\n");
1448
1449                         /* Now login */
1450                         init_adapter(hostdata);
1451                         break;
1452                 default:
1453                         dev_err(hostdata->dev, "unknown crq message type: %d\n", crq->format);
1454                 }
1455                 return;
1456         case 0xFF:      /* Hypervisor telling us the connection is closed */
1457                 scsi_block_requests(hostdata->host);
1458                 atomic_set(&hostdata->request_limit, 0);
1459                 if (crq->format == 0x06) {
1460                         /* We need to re-setup the interpartition connection */
1461                         dev_info(hostdata->dev, "Re-enabling adapter!\n");
1462                         hostdata->client_migrated = 1;
1463                         hostdata->reenable_crq = 1;
1464                         purge_requests(hostdata, DID_REQUEUE);
1465                         wake_up(&hostdata->work_wait_q);
1466                 } else {
1467                         dev_err(hostdata->dev, "Virtual adapter failed rc %d!\n",
1468                                 crq->format);
1469                         ibmvscsi_reset_host(hostdata);
1470                 }
1471                 return;
1472         case 0x80:              /* real payload */
1473                 break;
1474         default:
1475                 dev_err(hostdata->dev, "got an invalid message type 0x%02x\n",
1476                         crq->valid);
1477                 return;
1478         }
1479
1480         /* The only kind of payload CRQs we should get are responses to
1481          * things we send. Make sure this response is to something we
1482          * actually sent
1483          */
1484         if (!valid_event_struct(&hostdata->pool, evt_struct)) {
1485                 dev_err(hostdata->dev, "returned correlation_token 0x%p is invalid!\n",
1486                        (void *)crq->IU_data_ptr);
1487                 return;
1488         }
1489
1490         if (atomic_read(&evt_struct->free)) {
1491                 dev_err(hostdata->dev, "received duplicate correlation_token 0x%p!\n",
1492                         (void *)crq->IU_data_ptr);
1493                 return;
1494         }
1495
1496         if (crq->format == VIOSRP_SRP_FORMAT)
1497                 atomic_add(evt_struct->xfer_iu->srp.rsp.req_lim_delta,
1498                            &hostdata->request_limit);
1499
1500         del_timer(&evt_struct->timer);
1501
1502         if ((crq->status != VIOSRP_OK && crq->status != VIOSRP_OK2) && evt_struct->cmnd)
1503                 evt_struct->cmnd->result = DID_ERROR << 16;
1504         if (evt_struct->done)
1505                 evt_struct->done(evt_struct);
1506         else
1507                 dev_err(hostdata->dev, "returned done() is NULL; not running it!\n");
1508
1509         /*
1510          * Lock the host_lock before messing with these structures, since we
1511          * are running in a task context
1512          */
1513         spin_lock_irqsave(evt_struct->hostdata->host->host_lock, flags);
1514         list_del(&evt_struct->list);
1515         free_event_struct(&evt_struct->hostdata->pool, evt_struct);
1516         spin_unlock_irqrestore(evt_struct->hostdata->host->host_lock, flags);
1517 }
1518
1519 /**
1520  * ibmvscsi_get_host_config: Send the command to the server to get host
1521  * configuration data.  The data is opaque to us.
1522  */
1523 static int ibmvscsi_do_host_config(struct ibmvscsi_host_data *hostdata,
1524                                    unsigned char *buffer, int length)
1525 {
1526         struct viosrp_host_config *host_config;
1527         struct srp_event_struct *evt_struct;
1528         unsigned long flags;
1529         dma_addr_t addr;
1530         int rc;
1531
1532         evt_struct = get_event_struct(&hostdata->pool);
1533         if (!evt_struct) {
1534                 dev_err(hostdata->dev, "couldn't allocate event for HOST_CONFIG!\n");
1535                 return -1;
1536         }
1537
1538         init_event_struct(evt_struct,
1539                           sync_completion,
1540                           VIOSRP_MAD_FORMAT,
1541                           info_timeout);
1542
1543         host_config = &evt_struct->iu.mad.host_config;
1544
1545         /* Set up a lun reset SRP command */
1546         memset(host_config, 0x00, sizeof(*host_config));
1547         host_config->common.type = VIOSRP_HOST_CONFIG_TYPE;
1548         host_config->common.length = length;
1549         host_config->buffer = addr = dma_map_single(hostdata->dev, buffer,
1550                                                     length,
1551                                                     DMA_BIDIRECTIONAL);
1552
1553         if (dma_mapping_error(hostdata->dev, host_config->buffer)) {
1554                 if (!firmware_has_feature(FW_FEATURE_CMO))
1555                         dev_err(hostdata->dev,
1556                                 "dma_mapping error getting host config\n");
1557                 free_event_struct(&hostdata->pool, evt_struct);
1558                 return -1;
1559         }
1560
1561         init_completion(&evt_struct->comp);
1562         spin_lock_irqsave(hostdata->host->host_lock, flags);
1563         rc = ibmvscsi_send_srp_event(evt_struct, hostdata, info_timeout * 2);
1564         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1565         if (rc == 0)
1566                 wait_for_completion(&evt_struct->comp);
1567         dma_unmap_single(hostdata->dev, addr, length, DMA_BIDIRECTIONAL);
1568
1569         return rc;
1570 }
1571
1572 /**
1573  * ibmvscsi_slave_configure: Set the "allow_restart" flag for each disk.
1574  * @sdev:       struct scsi_device device to configure
1575  *
1576  * Enable allow_restart for a device if it is a disk.  Adjust the
1577  * queue_depth here also as is required by the documentation for
1578  * struct scsi_host_template.
1579  */
1580 static int ibmvscsi_slave_configure(struct scsi_device *sdev)
1581 {
1582         struct Scsi_Host *shost = sdev->host;
1583         unsigned long lock_flags = 0;
1584
1585         spin_lock_irqsave(shost->host_lock, lock_flags);
1586         if (sdev->type == TYPE_DISK) {
1587                 sdev->allow_restart = 1;
1588                 blk_queue_rq_timeout(sdev->request_queue, 120 * HZ);
1589         }
1590         scsi_adjust_queue_depth(sdev, 0, shost->cmd_per_lun);
1591         spin_unlock_irqrestore(shost->host_lock, lock_flags);
1592         return 0;
1593 }
1594
1595 /**
1596  * ibmvscsi_change_queue_depth - Change the device's queue depth
1597  * @sdev:       scsi device struct
1598  * @qdepth:     depth to set
1599  * @reason:     calling context
1600  *
1601  * Return value:
1602  *      actual depth set
1603  **/
1604 static int ibmvscsi_change_queue_depth(struct scsi_device *sdev, int qdepth,
1605                                        int reason)
1606 {
1607         if (reason != SCSI_QDEPTH_DEFAULT)
1608                 return -EOPNOTSUPP;
1609
1610         if (qdepth > IBMVSCSI_MAX_CMDS_PER_LUN)
1611                 qdepth = IBMVSCSI_MAX_CMDS_PER_LUN;
1612
1613         scsi_adjust_queue_depth(sdev, 0, qdepth);
1614         return sdev->queue_depth;
1615 }
1616
1617 /* ------------------------------------------------------------
1618  * sysfs attributes
1619  */
1620 static ssize_t show_host_vhost_loc(struct device *dev,
1621                                    struct device_attribute *attr, char *buf)
1622 {
1623         struct Scsi_Host *shost = class_to_shost(dev);
1624         struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1625         int len;
1626
1627         len = snprintf(buf, sizeof(hostdata->caps.loc), "%s\n",
1628                        hostdata->caps.loc);
1629         return len;
1630 }
1631
1632 static struct device_attribute ibmvscsi_host_vhost_loc = {
1633         .attr = {
1634                  .name = "vhost_loc",
1635                  .mode = S_IRUGO,
1636                  },
1637         .show = show_host_vhost_loc,
1638 };
1639
1640 static ssize_t show_host_vhost_name(struct device *dev,
1641                                     struct device_attribute *attr, char *buf)
1642 {
1643         struct Scsi_Host *shost = class_to_shost(dev);
1644         struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1645         int len;
1646
1647         len = snprintf(buf, sizeof(hostdata->caps.name), "%s\n",
1648                        hostdata->caps.name);
1649         return len;
1650 }
1651
1652 static struct device_attribute ibmvscsi_host_vhost_name = {
1653         .attr = {
1654                  .name = "vhost_name",
1655                  .mode = S_IRUGO,
1656                  },
1657         .show = show_host_vhost_name,
1658 };
1659
1660 static ssize_t show_host_srp_version(struct device *dev,
1661                                      struct device_attribute *attr, char *buf)
1662 {
1663         struct Scsi_Host *shost = class_to_shost(dev);
1664         struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1665         int len;
1666
1667         len = snprintf(buf, PAGE_SIZE, "%s\n",
1668                        hostdata->madapter_info.srp_version);
1669         return len;
1670 }
1671
1672 static struct device_attribute ibmvscsi_host_srp_version = {
1673         .attr = {
1674                  .name = "srp_version",
1675                  .mode = S_IRUGO,
1676                  },
1677         .show = show_host_srp_version,
1678 };
1679
1680 static ssize_t show_host_partition_name(struct device *dev,
1681                                         struct device_attribute *attr,
1682                                         char *buf)
1683 {
1684         struct Scsi_Host *shost = class_to_shost(dev);
1685         struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1686         int len;
1687
1688         len = snprintf(buf, PAGE_SIZE, "%s\n",
1689                        hostdata->madapter_info.partition_name);
1690         return len;
1691 }
1692
1693 static struct device_attribute ibmvscsi_host_partition_name = {
1694         .attr = {
1695                  .name = "partition_name",
1696                  .mode = S_IRUGO,
1697                  },
1698         .show = show_host_partition_name,
1699 };
1700
1701 static ssize_t show_host_partition_number(struct device *dev,
1702                                           struct device_attribute *attr,
1703                                           char *buf)
1704 {
1705         struct Scsi_Host *shost = class_to_shost(dev);
1706         struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1707         int len;
1708
1709         len = snprintf(buf, PAGE_SIZE, "%d\n",
1710                        hostdata->madapter_info.partition_number);
1711         return len;
1712 }
1713
1714 static struct device_attribute ibmvscsi_host_partition_number = {
1715         .attr = {
1716                  .name = "partition_number",
1717                  .mode = S_IRUGO,
1718                  },
1719         .show = show_host_partition_number,
1720 };
1721
1722 static ssize_t show_host_mad_version(struct device *dev,
1723                                      struct device_attribute *attr, char *buf)
1724 {
1725         struct Scsi_Host *shost = class_to_shost(dev);
1726         struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1727         int len;
1728
1729         len = snprintf(buf, PAGE_SIZE, "%d\n",
1730                        hostdata->madapter_info.mad_version);
1731         return len;
1732 }
1733
1734 static struct device_attribute ibmvscsi_host_mad_version = {
1735         .attr = {
1736                  .name = "mad_version",
1737                  .mode = S_IRUGO,
1738                  },
1739         .show = show_host_mad_version,
1740 };
1741
1742 static ssize_t show_host_os_type(struct device *dev,
1743                                  struct device_attribute *attr, char *buf)
1744 {
1745         struct Scsi_Host *shost = class_to_shost(dev);
1746         struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1747         int len;
1748
1749         len = snprintf(buf, PAGE_SIZE, "%d\n", hostdata->madapter_info.os_type);
1750         return len;
1751 }
1752
1753 static struct device_attribute ibmvscsi_host_os_type = {
1754         .attr = {
1755                  .name = "os_type",
1756                  .mode = S_IRUGO,
1757                  },
1758         .show = show_host_os_type,
1759 };
1760
1761 static ssize_t show_host_config(struct device *dev,
1762                                 struct device_attribute *attr, char *buf)
1763 {
1764         struct Scsi_Host *shost = class_to_shost(dev);
1765         struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1766
1767         /* returns null-terminated host config data */
1768         if (ibmvscsi_do_host_config(hostdata, buf, PAGE_SIZE) == 0)
1769                 return strlen(buf);
1770         else
1771                 return 0;
1772 }
1773
1774 static struct device_attribute ibmvscsi_host_config = {
1775         .attr = {
1776                  .name = "config",
1777                  .mode = S_IRUGO,
1778                  },
1779         .show = show_host_config,
1780 };
1781
1782 static struct device_attribute *ibmvscsi_attrs[] = {
1783         &ibmvscsi_host_vhost_loc,
1784         &ibmvscsi_host_vhost_name,
1785         &ibmvscsi_host_srp_version,
1786         &ibmvscsi_host_partition_name,
1787         &ibmvscsi_host_partition_number,
1788         &ibmvscsi_host_mad_version,
1789         &ibmvscsi_host_os_type,
1790         &ibmvscsi_host_config,
1791         NULL
1792 };
1793
1794 /* ------------------------------------------------------------
1795  * SCSI driver registration
1796  */
1797 static struct scsi_host_template driver_template = {
1798         .module = THIS_MODULE,
1799         .name = "IBM POWER Virtual SCSI Adapter " IBMVSCSI_VERSION,
1800         .proc_name = "ibmvscsi",
1801         .queuecommand = ibmvscsi_queuecommand,
1802         .eh_abort_handler = ibmvscsi_eh_abort_handler,
1803         .eh_device_reset_handler = ibmvscsi_eh_device_reset_handler,
1804         .eh_host_reset_handler = ibmvscsi_eh_host_reset_handler,
1805         .slave_configure = ibmvscsi_slave_configure,
1806         .change_queue_depth = ibmvscsi_change_queue_depth,
1807         .cmd_per_lun = IBMVSCSI_CMDS_PER_LUN_DEFAULT,
1808         .can_queue = IBMVSCSI_MAX_REQUESTS_DEFAULT,
1809         .this_id = -1,
1810         .sg_tablesize = SG_ALL,
1811         .use_clustering = ENABLE_CLUSTERING,
1812         .shost_attrs = ibmvscsi_attrs,
1813 };
1814
1815 /**
1816  * ibmvscsi_get_desired_dma - Calculate IO memory desired by the driver
1817  *
1818  * @vdev: struct vio_dev for the device whose desired IO mem is to be returned
1819  *
1820  * Return value:
1821  *      Number of bytes of IO data the driver will need to perform well.
1822  */
1823 static unsigned long ibmvscsi_get_desired_dma(struct vio_dev *vdev)
1824 {
1825         /* iu_storage data allocated in initialize_event_pool */
1826         unsigned long desired_io = max_events * sizeof(union viosrp_iu);
1827
1828         /* add io space for sg data */
1829         desired_io += (IBMVSCSI_MAX_SECTORS_DEFAULT * 512 *
1830                              IBMVSCSI_CMDS_PER_LUN_DEFAULT);
1831
1832         return desired_io;
1833 }
1834
1835 static void ibmvscsi_do_work(struct ibmvscsi_host_data *hostdata)
1836 {
1837         int rc;
1838         char *action = "reset";
1839
1840         if (hostdata->reset_crq) {
1841                 smp_rmb();
1842                 hostdata->reset_crq = 0;
1843
1844                 rc = ibmvscsi_ops->reset_crq_queue(&hostdata->queue, hostdata);
1845                 if (!rc)
1846                         rc = ibmvscsi_ops->send_crq(hostdata, 0xC001000000000000LL, 0);
1847                 if (!rc)
1848                         rc = vio_enable_interrupts(to_vio_dev(hostdata->dev));
1849         } else if (hostdata->reenable_crq) {
1850                 smp_rmb();
1851                 action = "enable";
1852                 rc = ibmvscsi_ops->reenable_crq_queue(&hostdata->queue, hostdata);
1853                 hostdata->reenable_crq = 0;
1854                 if (!rc)
1855                         rc = ibmvscsi_ops->send_crq(hostdata, 0xC001000000000000LL, 0);
1856         } else
1857                 return;
1858
1859         if (rc) {
1860                 atomic_set(&hostdata->request_limit, -1);
1861                 dev_err(hostdata->dev, "error after %s\n", action);
1862         }
1863
1864         scsi_unblock_requests(hostdata->host);
1865 }
1866
1867 static int ibmvscsi_work_to_do(struct ibmvscsi_host_data *hostdata)
1868 {
1869         if (kthread_should_stop())
1870                 return 1;
1871         else if (hostdata->reset_crq) {
1872                 smp_rmb();
1873                 return 1;
1874         } else if (hostdata->reenable_crq) {
1875                 smp_rmb();
1876                 return 1;
1877         }
1878
1879         return 0;
1880 }
1881
1882 static int ibmvscsi_work(void *data)
1883 {
1884         struct ibmvscsi_host_data *hostdata = data;
1885         int rc;
1886
1887         set_user_nice(current, -20);
1888
1889         while (1) {
1890                 rc = wait_event_interruptible(hostdata->work_wait_q,
1891                                               ibmvscsi_work_to_do(hostdata));
1892
1893                 BUG_ON(rc);
1894
1895                 if (kthread_should_stop())
1896                         break;
1897
1898                 ibmvscsi_do_work(hostdata);
1899         }
1900
1901         return 0;
1902 }
1903
1904 /**
1905  * Called by bus code for each adapter
1906  */
1907 static int ibmvscsi_probe(struct vio_dev *vdev, const struct vio_device_id *id)
1908 {
1909         struct ibmvscsi_host_data *hostdata;
1910         struct Scsi_Host *host;
1911         struct device *dev = &vdev->dev;
1912         struct srp_rport_identifiers ids;
1913         struct srp_rport *rport;
1914         unsigned long wait_switch = 0;
1915         int rc;
1916
1917         dev_set_drvdata(&vdev->dev, NULL);
1918
1919         host = scsi_host_alloc(&driver_template, sizeof(*hostdata));
1920         if (!host) {
1921                 dev_err(&vdev->dev, "couldn't allocate host data\n");
1922                 goto scsi_host_alloc_failed;
1923         }
1924
1925         host->transportt = ibmvscsi_transport_template;
1926         hostdata = shost_priv(host);
1927         memset(hostdata, 0x00, sizeof(*hostdata));
1928         INIT_LIST_HEAD(&hostdata->sent);
1929         init_waitqueue_head(&hostdata->work_wait_q);
1930         hostdata->host = host;
1931         hostdata->dev = dev;
1932         atomic_set(&hostdata->request_limit, -1);
1933         hostdata->host->max_sectors = IBMVSCSI_MAX_SECTORS_DEFAULT;
1934
1935         if (map_persist_bufs(hostdata)) {
1936                 dev_err(&vdev->dev, "couldn't map persistent buffers\n");
1937                 goto persist_bufs_failed;
1938         }
1939
1940         hostdata->work_thread = kthread_run(ibmvscsi_work, hostdata, "%s_%d",
1941                                             "ibmvscsi", host->host_no);
1942
1943         if (IS_ERR(hostdata->work_thread)) {
1944                 dev_err(&vdev->dev, "couldn't initialize kthread. rc=%ld\n",
1945                         PTR_ERR(hostdata->work_thread));
1946                 goto init_crq_failed;
1947         }
1948
1949         rc = ibmvscsi_ops->init_crq_queue(&hostdata->queue, hostdata, max_events);
1950         if (rc != 0 && rc != H_RESOURCE) {
1951                 dev_err(&vdev->dev, "couldn't initialize crq. rc=%d\n", rc);
1952                 goto kill_kthread;
1953         }
1954         if (initialize_event_pool(&hostdata->pool, max_events, hostdata) != 0) {
1955                 dev_err(&vdev->dev, "couldn't initialize event pool\n");
1956                 goto init_pool_failed;
1957         }
1958
1959         host->max_lun = 8;
1960         host->max_id = max_id;
1961         host->max_channel = max_channel;
1962         host->max_cmd_len = 16;
1963
1964         if (scsi_add_host(hostdata->host, hostdata->dev))
1965                 goto add_host_failed;
1966
1967         /* we don't have a proper target_port_id so let's use the fake one */
1968         memcpy(ids.port_id, hostdata->madapter_info.partition_name,
1969                sizeof(ids.port_id));
1970         ids.roles = SRP_RPORT_ROLE_TARGET;
1971         rport = srp_rport_add(host, &ids);
1972         if (IS_ERR(rport))
1973                 goto add_srp_port_failed;
1974
1975         /* Try to send an initialization message.  Note that this is allowed
1976          * to fail if the other end is not acive.  In that case we don't
1977          * want to scan
1978          */
1979         if (ibmvscsi_ops->send_crq(hostdata, 0xC001000000000000LL, 0) == 0
1980             || rc == H_RESOURCE) {
1981                 /*
1982                  * Wait around max init_timeout secs for the adapter to finish
1983                  * initializing. When we are done initializing, we will have a
1984                  * valid request_limit.  We don't want Linux scanning before
1985                  * we are ready.
1986                  */
1987                 for (wait_switch = jiffies + (init_timeout * HZ);
1988                      time_before(jiffies, wait_switch) &&
1989                      atomic_read(&hostdata->request_limit) < 2;) {
1990
1991                         msleep(10);
1992                 }
1993
1994                 /* if we now have a valid request_limit, initiate a scan */
1995                 if (atomic_read(&hostdata->request_limit) > 0)
1996                         scsi_scan_host(host);
1997         }
1998
1999         dev_set_drvdata(&vdev->dev, hostdata);
2000         return 0;
2001
2002       add_srp_port_failed:
2003         scsi_remove_host(hostdata->host);
2004       add_host_failed:
2005         release_event_pool(&hostdata->pool, hostdata);
2006       init_pool_failed:
2007         ibmvscsi_ops->release_crq_queue(&hostdata->queue, hostdata, max_events);
2008       kill_kthread:
2009       kthread_stop(hostdata->work_thread);
2010       init_crq_failed:
2011         unmap_persist_bufs(hostdata);
2012       persist_bufs_failed:
2013         scsi_host_put(host);
2014       scsi_host_alloc_failed:
2015         return -1;
2016 }
2017
2018 static int ibmvscsi_remove(struct vio_dev *vdev)
2019 {
2020         struct ibmvscsi_host_data *hostdata = dev_get_drvdata(&vdev->dev);
2021         unmap_persist_bufs(hostdata);
2022         release_event_pool(&hostdata->pool, hostdata);
2023         ibmvscsi_ops->release_crq_queue(&hostdata->queue, hostdata,
2024                                         max_events);
2025
2026         kthread_stop(hostdata->work_thread);
2027         srp_remove_host(hostdata->host);
2028         scsi_remove_host(hostdata->host);
2029         scsi_host_put(hostdata->host);
2030
2031         return 0;
2032 }
2033
2034 /**
2035  * ibmvscsi_resume: Resume from suspend
2036  * @dev:        device struct
2037  *
2038  * We may have lost an interrupt across suspend/resume, so kick the
2039  * interrupt handler
2040  */
2041 static int ibmvscsi_resume(struct device *dev)
2042 {
2043         struct ibmvscsi_host_data *hostdata = dev_get_drvdata(dev);
2044         return ibmvscsi_ops->resume(hostdata);
2045 }
2046
2047 /**
2048  * ibmvscsi_device_table: Used by vio.c to match devices in the device tree we 
2049  * support.
2050  */
2051 static struct vio_device_id ibmvscsi_device_table[] __devinitdata = {
2052         {"vscsi", "IBM,v-scsi"},
2053         { "", "" }
2054 };
2055 MODULE_DEVICE_TABLE(vio, ibmvscsi_device_table);
2056
2057 static struct dev_pm_ops ibmvscsi_pm_ops = {
2058         .resume = ibmvscsi_resume
2059 };
2060
2061 static struct vio_driver ibmvscsi_driver = {
2062         .id_table = ibmvscsi_device_table,
2063         .probe = ibmvscsi_probe,
2064         .remove = ibmvscsi_remove,
2065         .get_desired_dma = ibmvscsi_get_desired_dma,
2066         .driver = {
2067                 .name = "ibmvscsi",
2068                 .owner = THIS_MODULE,
2069                 .pm = &ibmvscsi_pm_ops,
2070         }
2071 };
2072
2073 static struct srp_function_template ibmvscsi_transport_functions = {
2074 };
2075
2076 int __init ibmvscsi_module_init(void)
2077 {
2078         int ret;
2079
2080         /* Ensure we have two requests to do error recovery */
2081         driver_template.can_queue = max_requests;
2082         max_events = max_requests + 2;
2083
2084         if (firmware_has_feature(FW_FEATURE_ISERIES))
2085                 ibmvscsi_ops = &iseriesvscsi_ops;
2086         else if (firmware_has_feature(FW_FEATURE_VIO))
2087                 ibmvscsi_ops = &rpavscsi_ops;
2088         else
2089                 return -ENODEV;
2090
2091         ibmvscsi_transport_template =
2092                 srp_attach_transport(&ibmvscsi_transport_functions);
2093         if (!ibmvscsi_transport_template)
2094                 return -ENOMEM;
2095
2096         ret = vio_register_driver(&ibmvscsi_driver);
2097         if (ret)
2098                 srp_release_transport(ibmvscsi_transport_template);
2099         return ret;
2100 }
2101
2102 void __exit ibmvscsi_module_exit(void)
2103 {
2104         vio_unregister_driver(&ibmvscsi_driver);
2105         srp_release_transport(ibmvscsi_transport_template);
2106 }
2107
2108 module_init(ibmvscsi_module_init);
2109 module_exit(ibmvscsi_module_exit);