Pull thinkpad-2.6.24 into release branch
[linux-2.6] / drivers / scsi / ibmvscsi / ibmvscsi.c
1 /* ------------------------------------------------------------
2  * ibmvscsi.c
3  * (C) Copyright IBM Corporation 1994, 2004
4  * Authors: Colin DeVilbiss (devilbis@us.ibm.com)
5  *          Santiago Leon (santil@us.ibm.com)
6  *          Dave Boutcher (sleddog@us.ibm.com)
7  *
8  * This program is free software; you can redistribute it and/or modify
9  * it under the terms of the GNU General Public License as published by
10  * the Free Software Foundation; either version 2 of the License, or
11  * (at your option) any later version.
12  *
13  * This program is distributed in the hope that it will be useful,
14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16  * GNU General Public License for more details.
17  *
18  * You should have received a copy of the GNU General Public License
19  * along with this program; if not, write to the Free Software
20  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307
21  * USA
22  *
23  * ------------------------------------------------------------
24  * Emulation of a SCSI host adapter for Virtual I/O devices
25  *
26  * This driver supports the SCSI adapter implemented by the IBM
27  * Power5 firmware.  That SCSI adapter is not a physical adapter,
28  * but allows Linux SCSI peripheral drivers to directly
29  * access devices in another logical partition on the physical system.
30  *
31  * The virtual adapter(s) are present in the open firmware device
32  * tree just like real adapters.
33  *
34  * One of the capabilities provided on these systems is the ability
35  * to DMA between partitions.  The architecture states that for VSCSI,
36  * the server side is allowed to DMA to and from the client.  The client
37  * is never trusted to DMA to or from the server directly.
38  *
39  * Messages are sent between partitions on a "Command/Response Queue" 
40  * (CRQ), which is just a buffer of 16 byte entries in the receiver's 
41  * Senders cannot access the buffer directly, but send messages by
42  * making a hypervisor call and passing in the 16 bytes.  The hypervisor
43  * puts the message in the next 16 byte space in round-robbin fashion,
44  * turns on the high order bit of the message (the valid bit), and 
45  * generates an interrupt to the receiver (if interrupts are turned on.) 
46  * The receiver just turns off the valid bit when they have copied out
47  * the message.
48  *
49  * The VSCSI client builds a SCSI Remote Protocol (SRP) Information Unit
50  * (IU) (as defined in the T10 standard available at www.t10.org), gets 
51  * a DMA address for the message, and sends it to the server as the
52  * payload of a CRQ message.  The server DMAs the SRP IU and processes it,
53  * including doing any additional data transfers.  When it is done, it
54  * DMAs the SRP response back to the same address as the request came from,
55  * and sends a CRQ message back to inform the client that the request has
56  * completed.
57  *
58  * Note that some of the underlying infrastructure is different between
59  * machines conforming to the "RS/6000 Platform Architecture" (RPA) and
60  * the older iSeries hypervisor models.  To support both, some low level
61  * routines have been broken out into rpa_vscsi.c and iseries_vscsi.c.
62  * The Makefile should pick one, not two, not zero, of these.
63  *
64  * TODO: This is currently pretty tied to the IBM i/pSeries hypervisor
65  * interfaces.  It would be really nice to abstract this above an RDMA
66  * layer.
67  */
68
69 #include <linux/module.h>
70 #include <linux/moduleparam.h>
71 #include <linux/dma-mapping.h>
72 #include <linux/delay.h>
73 #include <asm/firmware.h>
74 #include <asm/vio.h>
75 #include <scsi/scsi.h>
76 #include <scsi/scsi_cmnd.h>
77 #include <scsi/scsi_host.h>
78 #include <scsi/scsi_device.h>
79 #include <scsi/scsi_transport_srp.h>
80 #include "ibmvscsi.h"
81
82 /* The values below are somewhat arbitrary default values, but 
83  * OS/400 will use 3 busses (disks, CDs, tapes, I think.)
84  * Note that there are 3 bits of channel value, 6 bits of id, and
85  * 5 bits of LUN.
86  */
87 static int max_id = 64;
88 static int max_channel = 3;
89 static int init_timeout = 5;
90 static int max_requests = IBMVSCSI_MAX_REQUESTS_DEFAULT;
91
92 static struct scsi_transport_template *ibmvscsi_transport_template;
93
94 #define IBMVSCSI_VERSION "1.5.8"
95
96 static struct ibmvscsi_ops *ibmvscsi_ops;
97
98 MODULE_DESCRIPTION("IBM Virtual SCSI");
99 MODULE_AUTHOR("Dave Boutcher");
100 MODULE_LICENSE("GPL");
101 MODULE_VERSION(IBMVSCSI_VERSION);
102
103 module_param_named(max_id, max_id, int, S_IRUGO | S_IWUSR);
104 MODULE_PARM_DESC(max_id, "Largest ID value for each channel");
105 module_param_named(max_channel, max_channel, int, S_IRUGO | S_IWUSR);
106 MODULE_PARM_DESC(max_channel, "Largest channel value");
107 module_param_named(init_timeout, init_timeout, int, S_IRUGO | S_IWUSR);
108 MODULE_PARM_DESC(init_timeout, "Initialization timeout in seconds");
109 module_param_named(max_requests, max_requests, int, S_IRUGO | S_IWUSR);
110 MODULE_PARM_DESC(max_requests, "Maximum requests for this adapter");
111
112 /* ------------------------------------------------------------
113  * Routines for the event pool and event structs
114  */
115 /**
116  * initialize_event_pool: - Allocates and initializes the event pool for a host
117  * @pool:       event_pool to be initialized
118  * @size:       Number of events in pool
119  * @hostdata:   ibmvscsi_host_data who owns the event pool
120  *
121  * Returns zero on success.
122 */
123 static int initialize_event_pool(struct event_pool *pool,
124                                  int size, struct ibmvscsi_host_data *hostdata)
125 {
126         int i;
127
128         pool->size = size;
129         pool->next = 0;
130         pool->events = kcalloc(pool->size, sizeof(*pool->events), GFP_KERNEL);
131         if (!pool->events)
132                 return -ENOMEM;
133
134         pool->iu_storage =
135             dma_alloc_coherent(hostdata->dev,
136                                pool->size * sizeof(*pool->iu_storage),
137                                &pool->iu_token, 0);
138         if (!pool->iu_storage) {
139                 kfree(pool->events);
140                 return -ENOMEM;
141         }
142
143         for (i = 0; i < pool->size; ++i) {
144                 struct srp_event_struct *evt = &pool->events[i];
145                 memset(&evt->crq, 0x00, sizeof(evt->crq));
146                 atomic_set(&evt->free, 1);
147                 evt->crq.valid = 0x80;
148                 evt->crq.IU_length = sizeof(*evt->xfer_iu);
149                 evt->crq.IU_data_ptr = pool->iu_token + 
150                         sizeof(*evt->xfer_iu) * i;
151                 evt->xfer_iu = pool->iu_storage + i;
152                 evt->hostdata = hostdata;
153                 evt->ext_list = NULL;
154                 evt->ext_list_token = 0;
155         }
156
157         return 0;
158 }
159
160 /**
161  * release_event_pool: - Frees memory of an event pool of a host
162  * @pool:       event_pool to be released
163  * @hostdata:   ibmvscsi_host_data who owns the even pool
164  *
165  * Returns zero on success.
166 */
167 static void release_event_pool(struct event_pool *pool,
168                                struct ibmvscsi_host_data *hostdata)
169 {
170         int i, in_use = 0;
171         for (i = 0; i < pool->size; ++i) {
172                 if (atomic_read(&pool->events[i].free) != 1)
173                         ++in_use;
174                 if (pool->events[i].ext_list) {
175                         dma_free_coherent(hostdata->dev,
176                                   SG_ALL * sizeof(struct srp_direct_buf),
177                                   pool->events[i].ext_list,
178                                   pool->events[i].ext_list_token);
179                 }
180         }
181         if (in_use)
182                 dev_warn(hostdata->dev, "releasing event pool with %d "
183                          "events still in use?\n", in_use);
184         kfree(pool->events);
185         dma_free_coherent(hostdata->dev,
186                           pool->size * sizeof(*pool->iu_storage),
187                           pool->iu_storage, pool->iu_token);
188 }
189
190 /**
191  * valid_event_struct: - Determines if event is valid.
192  * @pool:       event_pool that contains the event
193  * @evt:        srp_event_struct to be checked for validity
194  *
195  * Returns zero if event is invalid, one otherwise.
196 */
197 static int valid_event_struct(struct event_pool *pool,
198                                 struct srp_event_struct *evt)
199 {
200         int index = evt - pool->events;
201         if (index < 0 || index >= pool->size)   /* outside of bounds */
202                 return 0;
203         if (evt != pool->events + index)        /* unaligned */
204                 return 0;
205         return 1;
206 }
207
208 /**
209  * ibmvscsi_free-event_struct: - Changes status of event to "free"
210  * @pool:       event_pool that contains the event
211  * @evt:        srp_event_struct to be modified
212  *
213 */
214 static void free_event_struct(struct event_pool *pool,
215                                        struct srp_event_struct *evt)
216 {
217         if (!valid_event_struct(pool, evt)) {
218                 dev_err(evt->hostdata->dev, "Freeing invalid event_struct %p "
219                         "(not in pool %p)\n", evt, pool->events);
220                 return;
221         }
222         if (atomic_inc_return(&evt->free) != 1) {
223                 dev_err(evt->hostdata->dev, "Freeing event_struct %p "
224                         "which is not in use!\n", evt);
225                 return;
226         }
227 }
228
229 /**
230  * get_evt_struct: - Gets the next free event in pool
231  * @pool:       event_pool that contains the events to be searched
232  *
233  * Returns the next event in "free" state, and NULL if none are free.
234  * Note that no synchronization is done here, we assume the host_lock
235  * will syncrhonze things.
236 */
237 static struct srp_event_struct *get_event_struct(struct event_pool *pool)
238 {
239         int i;
240         int poolsize = pool->size;
241         int offset = pool->next;
242
243         for (i = 0; i < poolsize; i++) {
244                 offset = (offset + 1) % poolsize;
245                 if (!atomic_dec_if_positive(&pool->events[offset].free)) {
246                         pool->next = offset;
247                         return &pool->events[offset];
248                 }
249         }
250
251         printk(KERN_ERR "ibmvscsi: found no event struct in pool!\n");
252         return NULL;
253 }
254
255 /**
256  * init_event_struct: Initialize fields in an event struct that are always 
257  *                    required.
258  * @evt:        The event
259  * @done:       Routine to call when the event is responded to
260  * @format:     SRP or MAD format
261  * @timeout:    timeout value set in the CRQ
262  */
263 static void init_event_struct(struct srp_event_struct *evt_struct,
264                               void (*done) (struct srp_event_struct *),
265                               u8 format,
266                               int timeout)
267 {
268         evt_struct->cmnd = NULL;
269         evt_struct->cmnd_done = NULL;
270         evt_struct->sync_srp = NULL;
271         evt_struct->crq.format = format;
272         evt_struct->crq.timeout = timeout;
273         evt_struct->done = done;
274 }
275
276 /* ------------------------------------------------------------
277  * Routines for receiving SCSI responses from the hosting partition
278  */
279
280 /**
281  * set_srp_direction: Set the fields in the srp related to data
282  *     direction and number of buffers based on the direction in
283  *     the scsi_cmnd and the number of buffers
284  */
285 static void set_srp_direction(struct scsi_cmnd *cmd,
286                               struct srp_cmd *srp_cmd, 
287                               int numbuf)
288 {
289         u8 fmt;
290
291         if (numbuf == 0)
292                 return;
293         
294         if (numbuf == 1)
295                 fmt = SRP_DATA_DESC_DIRECT;
296         else {
297                 fmt = SRP_DATA_DESC_INDIRECT;
298                 numbuf = min(numbuf, MAX_INDIRECT_BUFS);
299
300                 if (cmd->sc_data_direction == DMA_TO_DEVICE)
301                         srp_cmd->data_out_desc_cnt = numbuf;
302                 else
303                         srp_cmd->data_in_desc_cnt = numbuf;
304         }
305
306         if (cmd->sc_data_direction == DMA_TO_DEVICE)
307                 srp_cmd->buf_fmt = fmt << 4;
308         else
309                 srp_cmd->buf_fmt = fmt;
310 }
311
312 static void unmap_sg_list(int num_entries,
313                 struct device *dev,
314                 struct srp_direct_buf *md)
315 {
316         int i;
317
318         for (i = 0; i < num_entries; ++i)
319                 dma_unmap_single(dev, md[i].va, md[i].len, DMA_BIDIRECTIONAL);
320 }
321
322 /**
323  * unmap_cmd_data: - Unmap data pointed in srp_cmd based on the format
324  * @cmd:        srp_cmd whose additional_data member will be unmapped
325  * @dev:        device for which the memory is mapped
326  *
327 */
328 static void unmap_cmd_data(struct srp_cmd *cmd,
329                            struct srp_event_struct *evt_struct,
330                            struct device *dev)
331 {
332         u8 out_fmt, in_fmt;
333
334         out_fmt = cmd->buf_fmt >> 4;
335         in_fmt = cmd->buf_fmt & ((1U << 4) - 1);
336
337         if (out_fmt == SRP_NO_DATA_DESC && in_fmt == SRP_NO_DATA_DESC)
338                 return;
339         else if (out_fmt == SRP_DATA_DESC_DIRECT ||
340                  in_fmt == SRP_DATA_DESC_DIRECT) {
341                 struct srp_direct_buf *data =
342                         (struct srp_direct_buf *) cmd->add_data;
343                 dma_unmap_single(dev, data->va, data->len, DMA_BIDIRECTIONAL);
344         } else {
345                 struct srp_indirect_buf *indirect =
346                         (struct srp_indirect_buf *) cmd->add_data;
347                 int num_mapped = indirect->table_desc.len /
348                         sizeof(struct srp_direct_buf);
349
350                 if (num_mapped <= MAX_INDIRECT_BUFS) {
351                         unmap_sg_list(num_mapped, dev, &indirect->desc_list[0]);
352                         return;
353                 }
354
355                 unmap_sg_list(num_mapped, dev, evt_struct->ext_list);
356         }
357 }
358
359 static int map_sg_list(struct scsi_cmnd *cmd, int nseg,
360                        struct srp_direct_buf *md)
361 {
362         int i;
363         struct scatterlist *sg;
364         u64 total_length = 0;
365
366         scsi_for_each_sg(cmd, sg, nseg, i) {
367                 struct srp_direct_buf *descr = md + i;
368                 descr->va = sg_dma_address(sg);
369                 descr->len = sg_dma_len(sg);
370                 descr->key = 0;
371                 total_length += sg_dma_len(sg);
372         }
373         return total_length;
374 }
375
376 /**
377  * map_sg_data: - Maps dma for a scatterlist and initializes decriptor fields
378  * @cmd:        Scsi_Cmnd with the scatterlist
379  * @srp_cmd:    srp_cmd that contains the memory descriptor
380  * @dev:        device for which to map dma memory
381  *
382  * Called by map_data_for_srp_cmd() when building srp cmd from scsi cmd.
383  * Returns 1 on success.
384 */
385 static int map_sg_data(struct scsi_cmnd *cmd,
386                        struct srp_event_struct *evt_struct,
387                        struct srp_cmd *srp_cmd, struct device *dev)
388 {
389
390         int sg_mapped;
391         u64 total_length = 0;
392         struct srp_direct_buf *data =
393                 (struct srp_direct_buf *) srp_cmd->add_data;
394         struct srp_indirect_buf *indirect =
395                 (struct srp_indirect_buf *) data;
396
397         sg_mapped = scsi_dma_map(cmd);
398         if (!sg_mapped)
399                 return 1;
400         else if (sg_mapped < 0)
401                 return 0;
402
403         set_srp_direction(cmd, srp_cmd, sg_mapped);
404
405         /* special case; we can use a single direct descriptor */
406         if (sg_mapped == 1) {
407                 map_sg_list(cmd, sg_mapped, data);
408                 return 1;
409         }
410
411         indirect->table_desc.va = 0;
412         indirect->table_desc.len = sg_mapped * sizeof(struct srp_direct_buf);
413         indirect->table_desc.key = 0;
414
415         if (sg_mapped <= MAX_INDIRECT_BUFS) {
416                 total_length = map_sg_list(cmd, sg_mapped,
417                                            &indirect->desc_list[0]);
418                 indirect->len = total_length;
419                 return 1;
420         }
421
422         /* get indirect table */
423         if (!evt_struct->ext_list) {
424                 evt_struct->ext_list = (struct srp_direct_buf *)
425                         dma_alloc_coherent(dev,
426                                            SG_ALL * sizeof(struct srp_direct_buf),
427                                            &evt_struct->ext_list_token, 0);
428                 if (!evt_struct->ext_list) {
429                         sdev_printk(KERN_ERR, cmd->device,
430                                     "Can't allocate memory for indirect table\n");
431                         return 0;
432                 }
433         }
434
435         total_length = map_sg_list(cmd, sg_mapped, evt_struct->ext_list);
436
437         indirect->len = total_length;
438         indirect->table_desc.va = evt_struct->ext_list_token;
439         indirect->table_desc.len = sg_mapped * sizeof(indirect->desc_list[0]);
440         memcpy(indirect->desc_list, evt_struct->ext_list,
441                MAX_INDIRECT_BUFS * sizeof(struct srp_direct_buf));
442         return 1;
443 }
444
445 /**
446  * map_data_for_srp_cmd: - Calls functions to map data for srp cmds
447  * @cmd:        struct scsi_cmnd with the memory to be mapped
448  * @srp_cmd:    srp_cmd that contains the memory descriptor
449  * @dev:        dma device for which to map dma memory
450  *
451  * Called by scsi_cmd_to_srp_cmd() when converting scsi cmds to srp cmds 
452  * Returns 1 on success.
453 */
454 static int map_data_for_srp_cmd(struct scsi_cmnd *cmd,
455                                 struct srp_event_struct *evt_struct,
456                                 struct srp_cmd *srp_cmd, struct device *dev)
457 {
458         switch (cmd->sc_data_direction) {
459         case DMA_FROM_DEVICE:
460         case DMA_TO_DEVICE:
461                 break;
462         case DMA_NONE:
463                 return 1;
464         case DMA_BIDIRECTIONAL:
465                 sdev_printk(KERN_ERR, cmd->device,
466                             "Can't map DMA_BIDIRECTIONAL to read/write\n");
467                 return 0;
468         default:
469                 sdev_printk(KERN_ERR, cmd->device,
470                             "Unknown data direction 0x%02x; can't map!\n",
471                             cmd->sc_data_direction);
472                 return 0;
473         }
474
475         return map_sg_data(cmd, evt_struct, srp_cmd, dev);
476 }
477
478 /**
479  * purge_requests: Our virtual adapter just shut down.  purge any sent requests
480  * @hostdata:    the adapter
481  */
482 static void purge_requests(struct ibmvscsi_host_data *hostdata, int error_code)
483 {
484         struct srp_event_struct *tmp_evt, *pos;
485         unsigned long flags;
486
487         spin_lock_irqsave(hostdata->host->host_lock, flags);
488         list_for_each_entry_safe(tmp_evt, pos, &hostdata->sent, list) {
489                 list_del(&tmp_evt->list);
490                 del_timer(&tmp_evt->timer);
491                 if (tmp_evt->cmnd) {
492                         tmp_evt->cmnd->result = (error_code << 16);
493                         unmap_cmd_data(&tmp_evt->iu.srp.cmd,
494                                        tmp_evt,
495                                        tmp_evt->hostdata->dev);
496                         if (tmp_evt->cmnd_done)
497                                 tmp_evt->cmnd_done(tmp_evt->cmnd);
498                 } else if (tmp_evt->done)
499                         tmp_evt->done(tmp_evt);
500                 free_event_struct(&tmp_evt->hostdata->pool, tmp_evt);
501         }
502         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
503 }
504
505 /**
506  * ibmvscsi_reset_host - Reset the connection to the server
507  * @hostdata:   struct ibmvscsi_host_data to reset
508 */
509 static void ibmvscsi_reset_host(struct ibmvscsi_host_data *hostdata)
510 {
511         scsi_block_requests(hostdata->host);
512         atomic_set(&hostdata->request_limit, 0);
513
514         purge_requests(hostdata, DID_ERROR);
515         if ((ibmvscsi_ops->reset_crq_queue(&hostdata->queue, hostdata)) ||
516             (ibmvscsi_ops->send_crq(hostdata, 0xC001000000000000LL, 0)) ||
517             (vio_enable_interrupts(to_vio_dev(hostdata->dev)))) {
518                 atomic_set(&hostdata->request_limit, -1);
519                 dev_err(hostdata->dev, "error after reset\n");
520         }
521
522         scsi_unblock_requests(hostdata->host);
523 }
524
525 /**
526  * ibmvscsi_timeout - Internal command timeout handler
527  * @evt_struct: struct srp_event_struct that timed out
528  *
529  * Called when an internally generated command times out
530 */
531 static void ibmvscsi_timeout(struct srp_event_struct *evt_struct)
532 {
533         struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
534
535         dev_err(hostdata->dev, "Command timed out (%x). Resetting connection\n",
536                 evt_struct->iu.srp.cmd.opcode);
537
538         ibmvscsi_reset_host(hostdata);
539 }
540
541
542 /* ------------------------------------------------------------
543  * Routines for sending and receiving SRPs
544  */
545 /**
546  * ibmvscsi_send_srp_event: - Transforms event to u64 array and calls send_crq()
547  * @evt_struct: evt_struct to be sent
548  * @hostdata:   ibmvscsi_host_data of host
549  * @timeout:    timeout in seconds - 0 means do not time command
550  *
551  * Returns the value returned from ibmvscsi_send_crq(). (Zero for success)
552  * Note that this routine assumes that host_lock is held for synchronization
553 */
554 static int ibmvscsi_send_srp_event(struct srp_event_struct *evt_struct,
555                                    struct ibmvscsi_host_data *hostdata,
556                                    unsigned long timeout)
557 {
558         u64 *crq_as_u64 = (u64 *) &evt_struct->crq;
559         int request_status = 0;
560         int rc;
561
562         /* If we have exhausted our request limit, just fail this request,
563          * unless it is for a reset or abort.
564          * Note that there are rare cases involving driver generated requests 
565          * (such as task management requests) that the mid layer may think we
566          * can handle more requests (can_queue) when we actually can't
567          */
568         if (evt_struct->crq.format == VIOSRP_SRP_FORMAT) {
569                 request_status =
570                         atomic_dec_if_positive(&hostdata->request_limit);
571                 /* If request limit was -1 when we started, it is now even
572                  * less than that
573                  */
574                 if (request_status < -1)
575                         goto send_error;
576                 /* Otherwise, we may have run out of requests. */
577                 /* If request limit was 0 when we started the adapter is in the
578                  * process of performing a login with the server adapter, or
579                  * we may have run out of requests.
580                  */
581                 else if (request_status == -1 &&
582                          evt_struct->iu.srp.login_req.opcode != SRP_LOGIN_REQ)
583                         goto send_busy;
584                 /* Abort and reset calls should make it through.
585                  * Nothing except abort and reset should use the last two
586                  * slots unless we had two or less to begin with.
587                  */
588                 else if (request_status < 2 &&
589                          evt_struct->iu.srp.cmd.opcode != SRP_TSK_MGMT) {
590                         /* In the case that we have less than two requests
591                          * available, check the server limit as a combination
592                          * of the request limit and the number of requests
593                          * in-flight (the size of the send list).  If the
594                          * server limit is greater than 2, return busy so
595                          * that the last two are reserved for reset and abort.
596                          */
597                         int server_limit = request_status;
598                         struct srp_event_struct *tmp_evt;
599
600                         list_for_each_entry(tmp_evt, &hostdata->sent, list) {
601                                 server_limit++;
602                         }
603
604                         if (server_limit > 2)
605                                 goto send_busy;
606                 }
607         }
608
609         /* Copy the IU into the transfer area */
610         *evt_struct->xfer_iu = evt_struct->iu;
611         evt_struct->xfer_iu->srp.rsp.tag = (u64)evt_struct;
612
613         /* Add this to the sent list.  We need to do this 
614          * before we actually send 
615          * in case it comes back REALLY fast
616          */
617         list_add_tail(&evt_struct->list, &hostdata->sent);
618
619         init_timer(&evt_struct->timer);
620         if (timeout) {
621                 evt_struct->timer.data = (unsigned long) evt_struct;
622                 evt_struct->timer.expires = jiffies + (timeout * HZ);
623                 evt_struct->timer.function = (void (*)(unsigned long))ibmvscsi_timeout;
624                 add_timer(&evt_struct->timer);
625         }
626
627         if ((rc =
628              ibmvscsi_ops->send_crq(hostdata, crq_as_u64[0], crq_as_u64[1])) != 0) {
629                 list_del(&evt_struct->list);
630                 del_timer(&evt_struct->timer);
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         evt_struct = get_event_struct(&hostdata->pool);
724         if (!evt_struct)
725                 return SCSI_MLQUEUE_HOST_BUSY;
726
727         /* Set up the actual SRP IU */
728         srp_cmd = &evt_struct->iu.srp.cmd;
729         memset(srp_cmd, 0x00, SRP_MAX_IU_LEN);
730         srp_cmd->opcode = SRP_CMD;
731         memcpy(srp_cmd->cdb, cmnd->cmnd, sizeof(cmnd->cmnd));
732         srp_cmd->lun = ((u64) lun) << 48;
733
734         if (!map_data_for_srp_cmd(cmnd, evt_struct, srp_cmd, hostdata->dev)) {
735                 sdev_printk(KERN_ERR, cmnd->device, "couldn't convert cmd to srp_cmd\n");
736                 free_event_struct(&hostdata->pool, evt_struct);
737                 return SCSI_MLQUEUE_HOST_BUSY;
738         }
739
740         init_event_struct(evt_struct,
741                           handle_cmd_rsp,
742                           VIOSRP_SRP_FORMAT,
743                           cmnd->timeout_per_command/HZ);
744
745         evt_struct->cmnd = cmnd;
746         evt_struct->cmnd_done = done;
747
748         /* Fix up dma address of the buffer itself */
749         indirect = (struct srp_indirect_buf *) srp_cmd->add_data;
750         out_fmt = srp_cmd->buf_fmt >> 4;
751         in_fmt = srp_cmd->buf_fmt & ((1U << 4) - 1);
752         if ((in_fmt == SRP_DATA_DESC_INDIRECT ||
753              out_fmt == SRP_DATA_DESC_INDIRECT) &&
754             indirect->table_desc.va == 0) {
755                 indirect->table_desc.va = evt_struct->crq.IU_data_ptr +
756                         offsetof(struct srp_cmd, add_data) +
757                         offsetof(struct srp_indirect_buf, desc_list);
758         }
759
760         return ibmvscsi_send_srp_event(evt_struct, hostdata, 0);
761 }
762
763 /* ------------------------------------------------------------
764  * Routines for driver initialization
765  */
766 /**
767  * adapter_info_rsp: - Handle response to MAD adapter info request
768  * @evt_struct: srp_event_struct with the response
769  *
770  * Used as a "done" callback by when sending adapter_info. Gets called
771  * by ibmvscsi_handle_crq()
772 */
773 static void adapter_info_rsp(struct srp_event_struct *evt_struct)
774 {
775         struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
776         dma_unmap_single(hostdata->dev,
777                          evt_struct->iu.mad.adapter_info.buffer,
778                          evt_struct->iu.mad.adapter_info.common.length,
779                          DMA_BIDIRECTIONAL);
780
781         if (evt_struct->xfer_iu->mad.adapter_info.common.status) {
782                 dev_err(hostdata->dev, "error %d getting adapter info\n",
783                         evt_struct->xfer_iu->mad.adapter_info.common.status);
784         } else {
785                 dev_info(hostdata->dev, "host srp version: %s, "
786                          "host partition %s (%d), OS %d, max io %u\n",
787                          hostdata->madapter_info.srp_version,
788                          hostdata->madapter_info.partition_name,
789                          hostdata->madapter_info.partition_number,
790                          hostdata->madapter_info.os_type,
791                          hostdata->madapter_info.port_max_txu[0]);
792                 
793                 if (hostdata->madapter_info.port_max_txu[0]) 
794                         hostdata->host->max_sectors = 
795                                 hostdata->madapter_info.port_max_txu[0] >> 9;
796                 
797                 if (hostdata->madapter_info.os_type == 3 &&
798                     strcmp(hostdata->madapter_info.srp_version, "1.6a") <= 0) {
799                         dev_err(hostdata->dev, "host (Ver. %s) doesn't support large transfers\n",
800                                 hostdata->madapter_info.srp_version);
801                         dev_err(hostdata->dev, "limiting scatterlists to %d\n",
802                                 MAX_INDIRECT_BUFS);
803                         hostdata->host->sg_tablesize = MAX_INDIRECT_BUFS;
804                 }
805         }
806 }
807
808 /**
809  * send_mad_adapter_info: - Sends the mad adapter info request
810  *      and stores the result so it can be retrieved with
811  *      sysfs.  We COULD consider causing a failure if the
812  *      returned SRP version doesn't match ours.
813  * @hostdata:   ibmvscsi_host_data of host
814  * 
815  * Returns zero if successful.
816 */
817 static void send_mad_adapter_info(struct ibmvscsi_host_data *hostdata)
818 {
819         struct viosrp_adapter_info *req;
820         struct srp_event_struct *evt_struct;
821         unsigned long flags;
822         dma_addr_t addr;
823
824         evt_struct = get_event_struct(&hostdata->pool);
825         if (!evt_struct) {
826                 dev_err(hostdata->dev,
827                         "couldn't allocate an event for ADAPTER_INFO_REQ!\n");
828                 return;
829         }
830
831         init_event_struct(evt_struct,
832                           adapter_info_rsp,
833                           VIOSRP_MAD_FORMAT,
834                           init_timeout);
835         
836         req = &evt_struct->iu.mad.adapter_info;
837         memset(req, 0x00, sizeof(*req));
838         
839         req->common.type = VIOSRP_ADAPTER_INFO_TYPE;
840         req->common.length = sizeof(hostdata->madapter_info);
841         req->buffer = addr = dma_map_single(hostdata->dev,
842                                             &hostdata->madapter_info,
843                                             sizeof(hostdata->madapter_info),
844                                             DMA_BIDIRECTIONAL);
845
846         if (dma_mapping_error(req->buffer)) {
847                 dev_err(hostdata->dev, "Unable to map request_buffer for adapter_info!\n");
848                 free_event_struct(&hostdata->pool, evt_struct);
849                 return;
850         }
851         
852         spin_lock_irqsave(hostdata->host->host_lock, flags);
853         if (ibmvscsi_send_srp_event(evt_struct, hostdata, init_timeout * 2)) {
854                 dev_err(hostdata->dev, "couldn't send ADAPTER_INFO_REQ!\n");
855                 dma_unmap_single(hostdata->dev,
856                                  addr,
857                                  sizeof(hostdata->madapter_info),
858                                  DMA_BIDIRECTIONAL);
859         }
860         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
861 };
862
863 /**
864  * login_rsp: - Handle response to SRP login request
865  * @evt_struct: srp_event_struct with the response
866  *
867  * Used as a "done" callback by when sending srp_login. Gets called
868  * by ibmvscsi_handle_crq()
869 */
870 static void login_rsp(struct srp_event_struct *evt_struct)
871 {
872         struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
873         switch (evt_struct->xfer_iu->srp.login_rsp.opcode) {
874         case SRP_LOGIN_RSP:     /* it worked! */
875                 break;
876         case SRP_LOGIN_REJ:     /* refused! */
877                 dev_info(hostdata->dev, "SRP_LOGIN_REJ reason %u\n",
878                          evt_struct->xfer_iu->srp.login_rej.reason);
879                 /* Login failed.  */
880                 atomic_set(&hostdata->request_limit, -1);
881                 return;
882         default:
883                 dev_err(hostdata->dev, "Invalid login response typecode 0x%02x!\n",
884                         evt_struct->xfer_iu->srp.login_rsp.opcode);
885                 /* Login failed.  */
886                 atomic_set(&hostdata->request_limit, -1);
887                 return;
888         }
889
890         dev_info(hostdata->dev, "SRP_LOGIN succeeded\n");
891
892         if (evt_struct->xfer_iu->srp.login_rsp.req_lim_delta < 0)
893                 dev_err(hostdata->dev, "Invalid request_limit.\n");
894
895         /* Now we know what the real request-limit is.
896          * This value is set rather than added to request_limit because
897          * request_limit could have been set to -1 by this client.
898          */
899         atomic_set(&hostdata->request_limit,
900                    evt_struct->xfer_iu->srp.login_rsp.req_lim_delta);
901
902         /* If we had any pending I/Os, kick them */
903         scsi_unblock_requests(hostdata->host);
904
905         send_mad_adapter_info(hostdata);
906         return;
907 }
908
909 /**
910  * send_srp_login: - Sends the srp login
911  * @hostdata:   ibmvscsi_host_data of host
912  * 
913  * Returns zero if successful.
914 */
915 static int send_srp_login(struct ibmvscsi_host_data *hostdata)
916 {
917         int rc;
918         unsigned long flags;
919         struct srp_login_req *login;
920         struct srp_event_struct *evt_struct = get_event_struct(&hostdata->pool);
921         if (!evt_struct) {
922                 dev_err(hostdata->dev, "couldn't allocate an event for login req!\n");
923                 return FAILED;
924         }
925
926         init_event_struct(evt_struct,
927                           login_rsp,
928                           VIOSRP_SRP_FORMAT,
929                           init_timeout);
930
931         login = &evt_struct->iu.srp.login_req;
932         memset(login, 0x00, sizeof(struct srp_login_req));
933         login->opcode = SRP_LOGIN_REQ;
934         login->req_it_iu_len = sizeof(union srp_iu);
935         login->req_buf_fmt = SRP_BUF_FORMAT_DIRECT | SRP_BUF_FORMAT_INDIRECT;
936         
937         spin_lock_irqsave(hostdata->host->host_lock, flags);
938         /* Start out with a request limit of 0, since this is negotiated in
939          * the login request we are just sending and login requests always
940          * get sent by the driver regardless of request_limit.
941          */
942         atomic_set(&hostdata->request_limit, 0);
943
944         rc = ibmvscsi_send_srp_event(evt_struct, hostdata, init_timeout * 2);
945         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
946         dev_info(hostdata->dev, "sent SRP login\n");
947         return rc;
948 };
949
950 /**
951  * sync_completion: Signal that a synchronous command has completed
952  * Note that after returning from this call, the evt_struct is freed.
953  * the caller waiting on this completion shouldn't touch the evt_struct
954  * again.
955  */
956 static void sync_completion(struct srp_event_struct *evt_struct)
957 {
958         /* copy the response back */
959         if (evt_struct->sync_srp)
960                 *evt_struct->sync_srp = *evt_struct->xfer_iu;
961         
962         complete(&evt_struct->comp);
963 }
964
965 /**
966  * ibmvscsi_abort: Abort a command...from scsi host template
967  * send this over to the server and wait synchronously for the response
968  */
969 static int ibmvscsi_eh_abort_handler(struct scsi_cmnd *cmd)
970 {
971         struct ibmvscsi_host_data *hostdata = shost_priv(cmd->device->host);
972         struct srp_tsk_mgmt *tsk_mgmt;
973         struct srp_event_struct *evt;
974         struct srp_event_struct *tmp_evt, *found_evt;
975         union viosrp_iu srp_rsp;
976         int rsp_rc;
977         unsigned long flags;
978         u16 lun = lun_from_dev(cmd->device);
979
980         /* First, find this command in our sent list so we can figure
981          * out the correct tag
982          */
983         spin_lock_irqsave(hostdata->host->host_lock, flags);
984         found_evt = NULL;
985         list_for_each_entry(tmp_evt, &hostdata->sent, list) {
986                 if (tmp_evt->cmnd == cmd) {
987                         found_evt = tmp_evt;
988                         break;
989                 }
990         }
991
992         if (!found_evt) {
993                 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
994                 return SUCCESS;
995         }
996
997         evt = get_event_struct(&hostdata->pool);
998         if (evt == NULL) {
999                 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1000                 sdev_printk(KERN_ERR, cmd->device, "failed to allocate abort event\n");
1001                 return FAILED;
1002         }
1003         
1004         init_event_struct(evt,
1005                           sync_completion,
1006                           VIOSRP_SRP_FORMAT,
1007                           init_timeout);
1008
1009         tsk_mgmt = &evt->iu.srp.tsk_mgmt;
1010         
1011         /* Set up an abort SRP command */
1012         memset(tsk_mgmt, 0x00, sizeof(*tsk_mgmt));
1013         tsk_mgmt->opcode = SRP_TSK_MGMT;
1014         tsk_mgmt->lun = ((u64) lun) << 48;
1015         tsk_mgmt->tsk_mgmt_func = SRP_TSK_ABORT_TASK;
1016         tsk_mgmt->task_tag = (u64) found_evt;
1017
1018         sdev_printk(KERN_INFO, cmd->device, "aborting command. lun 0x%lx, tag 0x%lx\n",
1019                     tsk_mgmt->lun, tsk_mgmt->task_tag);
1020
1021         evt->sync_srp = &srp_rsp;
1022         init_completion(&evt->comp);
1023         rsp_rc = ibmvscsi_send_srp_event(evt, hostdata, init_timeout * 2);
1024         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1025         if (rsp_rc != 0) {
1026                 sdev_printk(KERN_ERR, cmd->device,
1027                             "failed to send abort() event. rc=%d\n", rsp_rc);
1028                 return FAILED;
1029         }
1030
1031         wait_for_completion(&evt->comp);
1032
1033         /* make sure we got a good response */
1034         if (unlikely(srp_rsp.srp.rsp.opcode != SRP_RSP)) {
1035                 if (printk_ratelimit())
1036                         sdev_printk(KERN_WARNING, cmd->device, "abort bad SRP RSP type %d\n",
1037                                     srp_rsp.srp.rsp.opcode);
1038                 return FAILED;
1039         }
1040
1041         if (srp_rsp.srp.rsp.flags & SRP_RSP_FLAG_RSPVALID)
1042                 rsp_rc = *((int *)srp_rsp.srp.rsp.data);
1043         else
1044                 rsp_rc = srp_rsp.srp.rsp.status;
1045
1046         if (rsp_rc) {
1047                 if (printk_ratelimit())
1048                         sdev_printk(KERN_WARNING, cmd->device,
1049                                     "abort code %d for task tag 0x%lx\n",
1050                                     rsp_rc, tsk_mgmt->task_tag);
1051                 return FAILED;
1052         }
1053
1054         /* Because we dropped the spinlock above, it's possible
1055          * The event is no longer in our list.  Make sure it didn't
1056          * complete while we were aborting
1057          */
1058         spin_lock_irqsave(hostdata->host->host_lock, flags);
1059         found_evt = NULL;
1060         list_for_each_entry(tmp_evt, &hostdata->sent, list) {
1061                 if (tmp_evt->cmnd == cmd) {
1062                         found_evt = tmp_evt;
1063                         break;
1064                 }
1065         }
1066
1067         if (found_evt == NULL) {
1068                 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1069                 sdev_printk(KERN_INFO, cmd->device, "aborted task tag 0x%lx completed\n",
1070                             tsk_mgmt->task_tag);
1071                 return SUCCESS;
1072         }
1073
1074         sdev_printk(KERN_INFO, cmd->device, "successfully aborted task tag 0x%lx\n",
1075                     tsk_mgmt->task_tag);
1076
1077         cmd->result = (DID_ABORT << 16);
1078         list_del(&found_evt->list);
1079         unmap_cmd_data(&found_evt->iu.srp.cmd, found_evt,
1080                        found_evt->hostdata->dev);
1081         free_event_struct(&found_evt->hostdata->pool, found_evt);
1082         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1083         atomic_inc(&hostdata->request_limit);
1084         return SUCCESS;
1085 }
1086
1087 /**
1088  * ibmvscsi_eh_device_reset_handler: Reset a single LUN...from scsi host 
1089  * template send this over to the server and wait synchronously for the 
1090  * response
1091  */
1092 static int ibmvscsi_eh_device_reset_handler(struct scsi_cmnd *cmd)
1093 {
1094         struct ibmvscsi_host_data *hostdata = shost_priv(cmd->device->host);
1095         struct srp_tsk_mgmt *tsk_mgmt;
1096         struct srp_event_struct *evt;
1097         struct srp_event_struct *tmp_evt, *pos;
1098         union viosrp_iu srp_rsp;
1099         int rsp_rc;
1100         unsigned long flags;
1101         u16 lun = lun_from_dev(cmd->device);
1102
1103         spin_lock_irqsave(hostdata->host->host_lock, flags);
1104         evt = get_event_struct(&hostdata->pool);
1105         if (evt == NULL) {
1106                 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1107                 sdev_printk(KERN_ERR, cmd->device, "failed to allocate reset event\n");
1108                 return FAILED;
1109         }
1110         
1111         init_event_struct(evt,
1112                           sync_completion,
1113                           VIOSRP_SRP_FORMAT,
1114                           init_timeout);
1115
1116         tsk_mgmt = &evt->iu.srp.tsk_mgmt;
1117
1118         /* Set up a lun reset SRP command */
1119         memset(tsk_mgmt, 0x00, sizeof(*tsk_mgmt));
1120         tsk_mgmt->opcode = SRP_TSK_MGMT;
1121         tsk_mgmt->lun = ((u64) lun) << 48;
1122         tsk_mgmt->tsk_mgmt_func = SRP_TSK_LUN_RESET;
1123
1124         sdev_printk(KERN_INFO, cmd->device, "resetting device. lun 0x%lx\n",
1125                     tsk_mgmt->lun);
1126
1127         evt->sync_srp = &srp_rsp;
1128         init_completion(&evt->comp);
1129         rsp_rc = ibmvscsi_send_srp_event(evt, hostdata, init_timeout * 2);
1130         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1131         if (rsp_rc != 0) {
1132                 sdev_printk(KERN_ERR, cmd->device,
1133                             "failed to send reset event. rc=%d\n", rsp_rc);
1134                 return FAILED;
1135         }
1136
1137         wait_for_completion(&evt->comp);
1138
1139         /* make sure we got a good response */
1140         if (unlikely(srp_rsp.srp.rsp.opcode != SRP_RSP)) {
1141                 if (printk_ratelimit())
1142                         sdev_printk(KERN_WARNING, cmd->device, "reset bad SRP RSP type %d\n",
1143                                     srp_rsp.srp.rsp.opcode);
1144                 return FAILED;
1145         }
1146
1147         if (srp_rsp.srp.rsp.flags & SRP_RSP_FLAG_RSPVALID)
1148                 rsp_rc = *((int *)srp_rsp.srp.rsp.data);
1149         else
1150                 rsp_rc = srp_rsp.srp.rsp.status;
1151
1152         if (rsp_rc) {
1153                 if (printk_ratelimit())
1154                         sdev_printk(KERN_WARNING, cmd->device,
1155                                     "reset code %d for task tag 0x%lx\n",
1156                                     rsp_rc, tsk_mgmt->task_tag);
1157                 return FAILED;
1158         }
1159
1160         /* We need to find all commands for this LUN that have not yet been
1161          * responded to, and fail them with DID_RESET
1162          */
1163         spin_lock_irqsave(hostdata->host->host_lock, flags);
1164         list_for_each_entry_safe(tmp_evt, pos, &hostdata->sent, list) {
1165                 if ((tmp_evt->cmnd) && (tmp_evt->cmnd->device == cmd->device)) {
1166                         if (tmp_evt->cmnd)
1167                                 tmp_evt->cmnd->result = (DID_RESET << 16);
1168                         list_del(&tmp_evt->list);
1169                         unmap_cmd_data(&tmp_evt->iu.srp.cmd, tmp_evt,
1170                                        tmp_evt->hostdata->dev);
1171                         free_event_struct(&tmp_evt->hostdata->pool,
1172                                                    tmp_evt);
1173                         atomic_inc(&hostdata->request_limit);
1174                         if (tmp_evt->cmnd_done)
1175                                 tmp_evt->cmnd_done(tmp_evt->cmnd);
1176                         else if (tmp_evt->done)
1177                                 tmp_evt->done(tmp_evt);
1178                 }
1179         }
1180         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1181         return SUCCESS;
1182 }
1183
1184 /**
1185  * ibmvscsi_eh_host_reset_handler - Reset the connection to the server
1186  * @cmd:        struct scsi_cmnd having problems
1187 */
1188 static int ibmvscsi_eh_host_reset_handler(struct scsi_cmnd *cmd)
1189 {
1190         unsigned long wait_switch = 0;
1191         struct ibmvscsi_host_data *hostdata = shost_priv(cmd->device->host);
1192
1193         dev_err(hostdata->dev, "Resetting connection due to error recovery\n");
1194
1195         ibmvscsi_reset_host(hostdata);
1196
1197         for (wait_switch = jiffies + (init_timeout * HZ);
1198              time_before(jiffies, wait_switch) &&
1199                      atomic_read(&hostdata->request_limit) < 2;) {
1200
1201                 msleep(10);
1202         }
1203
1204         if (atomic_read(&hostdata->request_limit) <= 0)
1205                 return FAILED;
1206
1207         return SUCCESS;
1208 }
1209
1210 /**
1211  * ibmvscsi_handle_crq: - Handles and frees received events in the CRQ
1212  * @crq:        Command/Response queue
1213  * @hostdata:   ibmvscsi_host_data of host
1214  *
1215 */
1216 void ibmvscsi_handle_crq(struct viosrp_crq *crq,
1217                          struct ibmvscsi_host_data *hostdata)
1218 {
1219         long rc;
1220         unsigned long flags;
1221         struct srp_event_struct *evt_struct =
1222             (struct srp_event_struct *)crq->IU_data_ptr;
1223         switch (crq->valid) {
1224         case 0xC0:              /* initialization */
1225                 switch (crq->format) {
1226                 case 0x01:      /* Initialization message */
1227                         dev_info(hostdata->dev, "partner initialized\n");
1228                         /* Send back a response */
1229                         if ((rc = ibmvscsi_ops->send_crq(hostdata,
1230                                                          0xC002000000000000LL, 0)) == 0) {
1231                                 /* Now login */
1232                                 send_srp_login(hostdata);
1233                         } else {
1234                                 dev_err(hostdata->dev, "Unable to send init rsp. rc=%ld\n", rc);
1235                         }
1236
1237                         break;
1238                 case 0x02:      /* Initialization response */
1239                         dev_info(hostdata->dev, "partner initialization complete\n");
1240
1241                         /* Now login */
1242                         send_srp_login(hostdata);
1243                         break;
1244                 default:
1245                         dev_err(hostdata->dev, "unknown crq message type: %d\n", crq->format);
1246                 }
1247                 return;
1248         case 0xFF:      /* Hypervisor telling us the connection is closed */
1249                 scsi_block_requests(hostdata->host);
1250                 atomic_set(&hostdata->request_limit, 0);
1251                 if (crq->format == 0x06) {
1252                         /* We need to re-setup the interpartition connection */
1253                         dev_info(hostdata->dev, "Re-enabling adapter!\n");
1254                         purge_requests(hostdata, DID_REQUEUE);
1255                         if ((ibmvscsi_ops->reenable_crq_queue(&hostdata->queue,
1256                                                               hostdata)) ||
1257                             (ibmvscsi_ops->send_crq(hostdata,
1258                                                     0xC001000000000000LL, 0))) {
1259                                         atomic_set(&hostdata->request_limit,
1260                                                    -1);
1261                                         dev_err(hostdata->dev, "error after enable\n");
1262                         }
1263                 } else {
1264                         dev_err(hostdata->dev, "Virtual adapter failed rc %d!\n",
1265                                 crq->format);
1266
1267                         purge_requests(hostdata, DID_ERROR);
1268                         if ((ibmvscsi_ops->reset_crq_queue(&hostdata->queue,
1269                                                            hostdata)) ||
1270                             (ibmvscsi_ops->send_crq(hostdata,
1271                                                     0xC001000000000000LL, 0))) {
1272                                         atomic_set(&hostdata->request_limit,
1273                                                    -1);
1274                                         dev_err(hostdata->dev, "error after reset\n");
1275                         }
1276                 }
1277                 scsi_unblock_requests(hostdata->host);
1278                 return;
1279         case 0x80:              /* real payload */
1280                 break;
1281         default:
1282                 dev_err(hostdata->dev, "got an invalid message type 0x%02x\n",
1283                         crq->valid);
1284                 return;
1285         }
1286
1287         /* The only kind of payload CRQs we should get are responses to
1288          * things we send. Make sure this response is to something we
1289          * actually sent
1290          */
1291         if (!valid_event_struct(&hostdata->pool, evt_struct)) {
1292                 dev_err(hostdata->dev, "returned correlation_token 0x%p is invalid!\n",
1293                        (void *)crq->IU_data_ptr);
1294                 return;
1295         }
1296
1297         if (atomic_read(&evt_struct->free)) {
1298                 dev_err(hostdata->dev, "received duplicate correlation_token 0x%p!\n",
1299                         (void *)crq->IU_data_ptr);
1300                 return;
1301         }
1302
1303         if (crq->format == VIOSRP_SRP_FORMAT)
1304                 atomic_add(evt_struct->xfer_iu->srp.rsp.req_lim_delta,
1305                            &hostdata->request_limit);
1306
1307         del_timer(&evt_struct->timer);
1308
1309         if (evt_struct->done)
1310                 evt_struct->done(evt_struct);
1311         else
1312                 dev_err(hostdata->dev, "returned done() is NULL; not running it!\n");
1313
1314         /*
1315          * Lock the host_lock before messing with these structures, since we
1316          * are running in a task context
1317          */
1318         spin_lock_irqsave(evt_struct->hostdata->host->host_lock, flags);
1319         list_del(&evt_struct->list);
1320         free_event_struct(&evt_struct->hostdata->pool, evt_struct);
1321         spin_unlock_irqrestore(evt_struct->hostdata->host->host_lock, flags);
1322 }
1323
1324 /**
1325  * ibmvscsi_get_host_config: Send the command to the server to get host
1326  * configuration data.  The data is opaque to us.
1327  */
1328 static int ibmvscsi_do_host_config(struct ibmvscsi_host_data *hostdata,
1329                                    unsigned char *buffer, int length)
1330 {
1331         struct viosrp_host_config *host_config;
1332         struct srp_event_struct *evt_struct;
1333         unsigned long flags;
1334         dma_addr_t addr;
1335         int rc;
1336
1337         evt_struct = get_event_struct(&hostdata->pool);
1338         if (!evt_struct) {
1339                 dev_err(hostdata->dev, "couldn't allocate event for HOST_CONFIG!\n");
1340                 return -1;
1341         }
1342
1343         init_event_struct(evt_struct,
1344                           sync_completion,
1345                           VIOSRP_MAD_FORMAT,
1346                           init_timeout);
1347
1348         host_config = &evt_struct->iu.mad.host_config;
1349
1350         /* Set up a lun reset SRP command */
1351         memset(host_config, 0x00, sizeof(*host_config));
1352         host_config->common.type = VIOSRP_HOST_CONFIG_TYPE;
1353         host_config->common.length = length;
1354         host_config->buffer = addr = dma_map_single(hostdata->dev, buffer,
1355                                                     length,
1356                                                     DMA_BIDIRECTIONAL);
1357
1358         if (dma_mapping_error(host_config->buffer)) {
1359                 dev_err(hostdata->dev, "dma_mapping error getting host config\n");
1360                 free_event_struct(&hostdata->pool, evt_struct);
1361                 return -1;
1362         }
1363
1364         init_completion(&evt_struct->comp);
1365         spin_lock_irqsave(hostdata->host->host_lock, flags);
1366         rc = ibmvscsi_send_srp_event(evt_struct, hostdata, init_timeout * 2);
1367         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1368         if (rc == 0)
1369                 wait_for_completion(&evt_struct->comp);
1370         dma_unmap_single(hostdata->dev, addr, length, DMA_BIDIRECTIONAL);
1371
1372         return rc;
1373 }
1374
1375 /**
1376  * ibmvscsi_slave_configure: Set the "allow_restart" flag for each disk.
1377  * @sdev:       struct scsi_device device to configure
1378  *
1379  * Enable allow_restart for a device if it is a disk.  Adjust the
1380  * queue_depth here also as is required by the documentation for
1381  * struct scsi_host_template.
1382  */
1383 static int ibmvscsi_slave_configure(struct scsi_device *sdev)
1384 {
1385         struct Scsi_Host *shost = sdev->host;
1386         unsigned long lock_flags = 0;
1387
1388         spin_lock_irqsave(shost->host_lock, lock_flags);
1389         if (sdev->type == TYPE_DISK)
1390                 sdev->allow_restart = 1;
1391         scsi_adjust_queue_depth(sdev, 0, shost->cmd_per_lun);
1392         spin_unlock_irqrestore(shost->host_lock, lock_flags);
1393         return 0;
1394 }
1395
1396 /**
1397  * ibmvscsi_change_queue_depth - Change the device's queue depth
1398  * @sdev:       scsi device struct
1399  * @qdepth:     depth to set
1400  *
1401  * Return value:
1402  *      actual depth set
1403  **/
1404 static int ibmvscsi_change_queue_depth(struct scsi_device *sdev, int qdepth)
1405 {
1406         if (qdepth > IBMVSCSI_MAX_CMDS_PER_LUN)
1407                 qdepth = IBMVSCSI_MAX_CMDS_PER_LUN;
1408
1409         scsi_adjust_queue_depth(sdev, 0, qdepth);
1410         return sdev->queue_depth;
1411 }
1412
1413 /* ------------------------------------------------------------
1414  * sysfs attributes
1415  */
1416 static ssize_t show_host_srp_version(struct class_device *class_dev, char *buf)
1417 {
1418         struct Scsi_Host *shost = class_to_shost(class_dev);
1419         struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1420         int len;
1421
1422         len = snprintf(buf, PAGE_SIZE, "%s\n",
1423                        hostdata->madapter_info.srp_version);
1424         return len;
1425 }
1426
1427 static struct class_device_attribute ibmvscsi_host_srp_version = {
1428         .attr = {
1429                  .name = "srp_version",
1430                  .mode = S_IRUGO,
1431                  },
1432         .show = show_host_srp_version,
1433 };
1434
1435 static ssize_t show_host_partition_name(struct class_device *class_dev,
1436                                         char *buf)
1437 {
1438         struct Scsi_Host *shost = class_to_shost(class_dev);
1439         struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1440         int len;
1441
1442         len = snprintf(buf, PAGE_SIZE, "%s\n",
1443                        hostdata->madapter_info.partition_name);
1444         return len;
1445 }
1446
1447 static struct class_device_attribute ibmvscsi_host_partition_name = {
1448         .attr = {
1449                  .name = "partition_name",
1450                  .mode = S_IRUGO,
1451                  },
1452         .show = show_host_partition_name,
1453 };
1454
1455 static ssize_t show_host_partition_number(struct class_device *class_dev,
1456                                           char *buf)
1457 {
1458         struct Scsi_Host *shost = class_to_shost(class_dev);
1459         struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1460         int len;
1461
1462         len = snprintf(buf, PAGE_SIZE, "%d\n",
1463                        hostdata->madapter_info.partition_number);
1464         return len;
1465 }
1466
1467 static struct class_device_attribute ibmvscsi_host_partition_number = {
1468         .attr = {
1469                  .name = "partition_number",
1470                  .mode = S_IRUGO,
1471                  },
1472         .show = show_host_partition_number,
1473 };
1474
1475 static ssize_t show_host_mad_version(struct class_device *class_dev, char *buf)
1476 {
1477         struct Scsi_Host *shost = class_to_shost(class_dev);
1478         struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1479         int len;
1480
1481         len = snprintf(buf, PAGE_SIZE, "%d\n",
1482                        hostdata->madapter_info.mad_version);
1483         return len;
1484 }
1485
1486 static struct class_device_attribute ibmvscsi_host_mad_version = {
1487         .attr = {
1488                  .name = "mad_version",
1489                  .mode = S_IRUGO,
1490                  },
1491         .show = show_host_mad_version,
1492 };
1493
1494 static ssize_t show_host_os_type(struct class_device *class_dev, char *buf)
1495 {
1496         struct Scsi_Host *shost = class_to_shost(class_dev);
1497         struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1498         int len;
1499
1500         len = snprintf(buf, PAGE_SIZE, "%d\n", hostdata->madapter_info.os_type);
1501         return len;
1502 }
1503
1504 static struct class_device_attribute ibmvscsi_host_os_type = {
1505         .attr = {
1506                  .name = "os_type",
1507                  .mode = S_IRUGO,
1508                  },
1509         .show = show_host_os_type,
1510 };
1511
1512 static ssize_t show_host_config(struct class_device *class_dev, char *buf)
1513 {
1514         struct Scsi_Host *shost = class_to_shost(class_dev);
1515         struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1516
1517         /* returns null-terminated host config data */
1518         if (ibmvscsi_do_host_config(hostdata, buf, PAGE_SIZE) == 0)
1519                 return strlen(buf);
1520         else
1521                 return 0;
1522 }
1523
1524 static struct class_device_attribute ibmvscsi_host_config = {
1525         .attr = {
1526                  .name = "config",
1527                  .mode = S_IRUGO,
1528                  },
1529         .show = show_host_config,
1530 };
1531
1532 static struct class_device_attribute *ibmvscsi_attrs[] = {
1533         &ibmvscsi_host_srp_version,
1534         &ibmvscsi_host_partition_name,
1535         &ibmvscsi_host_partition_number,
1536         &ibmvscsi_host_mad_version,
1537         &ibmvscsi_host_os_type,
1538         &ibmvscsi_host_config,
1539         NULL
1540 };
1541
1542 /* ------------------------------------------------------------
1543  * SCSI driver registration
1544  */
1545 static struct scsi_host_template driver_template = {
1546         .module = THIS_MODULE,
1547         .name = "IBM POWER Virtual SCSI Adapter " IBMVSCSI_VERSION,
1548         .proc_name = "ibmvscsi",
1549         .queuecommand = ibmvscsi_queuecommand,
1550         .eh_abort_handler = ibmvscsi_eh_abort_handler,
1551         .eh_device_reset_handler = ibmvscsi_eh_device_reset_handler,
1552         .eh_host_reset_handler = ibmvscsi_eh_host_reset_handler,
1553         .slave_configure = ibmvscsi_slave_configure,
1554         .change_queue_depth = ibmvscsi_change_queue_depth,
1555         .cmd_per_lun = 16,
1556         .can_queue = IBMVSCSI_MAX_REQUESTS_DEFAULT,
1557         .this_id = -1,
1558         .sg_tablesize = SG_ALL,
1559         .use_clustering = ENABLE_CLUSTERING,
1560         .use_sg_chaining = ENABLE_SG_CHAINING,
1561         .shost_attrs = ibmvscsi_attrs,
1562 };
1563
1564 /**
1565  * Called by bus code for each adapter
1566  */
1567 static int ibmvscsi_probe(struct vio_dev *vdev, const struct vio_device_id *id)
1568 {
1569         struct ibmvscsi_host_data *hostdata;
1570         struct Scsi_Host *host;
1571         struct device *dev = &vdev->dev;
1572         struct srp_rport_identifiers ids;
1573         struct srp_rport *rport;
1574         unsigned long wait_switch = 0;
1575         int rc;
1576
1577         vdev->dev.driver_data = NULL;
1578
1579         driver_template.can_queue = max_requests;
1580         host = scsi_host_alloc(&driver_template, sizeof(*hostdata));
1581         if (!host) {
1582                 dev_err(&vdev->dev, "couldn't allocate host data\n");
1583                 goto scsi_host_alloc_failed;
1584         }
1585
1586         host->transportt = ibmvscsi_transport_template;
1587         hostdata = shost_priv(host);
1588         memset(hostdata, 0x00, sizeof(*hostdata));
1589         INIT_LIST_HEAD(&hostdata->sent);
1590         hostdata->host = host;
1591         hostdata->dev = dev;
1592         atomic_set(&hostdata->request_limit, -1);
1593         hostdata->host->max_sectors = 32 * 8; /* default max I/O 32 pages */
1594
1595         rc = ibmvscsi_ops->init_crq_queue(&hostdata->queue, hostdata, max_requests);
1596         if (rc != 0 && rc != H_RESOURCE) {
1597                 dev_err(&vdev->dev, "couldn't initialize crq. rc=%d\n", rc);
1598                 goto init_crq_failed;
1599         }
1600         if (initialize_event_pool(&hostdata->pool, max_requests, hostdata) != 0) {
1601                 dev_err(&vdev->dev, "couldn't initialize event pool\n");
1602                 goto init_pool_failed;
1603         }
1604
1605         host->max_lun = 8;
1606         host->max_id = max_id;
1607         host->max_channel = max_channel;
1608
1609         if (scsi_add_host(hostdata->host, hostdata->dev))
1610                 goto add_host_failed;
1611
1612         /* we don't have a proper target_port_id so let's use the fake one */
1613         memcpy(ids.port_id, hostdata->madapter_info.partition_name,
1614                sizeof(ids.port_id));
1615         ids.roles = SRP_RPORT_ROLE_TARGET;
1616         rport = srp_rport_add(host, &ids);
1617         if (IS_ERR(rport))
1618                 goto add_srp_port_failed;
1619
1620         /* Try to send an initialization message.  Note that this is allowed
1621          * to fail if the other end is not acive.  In that case we don't
1622          * want to scan
1623          */
1624         if (ibmvscsi_ops->send_crq(hostdata, 0xC001000000000000LL, 0) == 0
1625             || rc == H_RESOURCE) {
1626                 /*
1627                  * Wait around max init_timeout secs for the adapter to finish
1628                  * initializing. When we are done initializing, we will have a
1629                  * valid request_limit.  We don't want Linux scanning before
1630                  * we are ready.
1631                  */
1632                 for (wait_switch = jiffies + (init_timeout * HZ);
1633                      time_before(jiffies, wait_switch) &&
1634                      atomic_read(&hostdata->request_limit) < 2;) {
1635
1636                         msleep(10);
1637                 }
1638
1639                 /* if we now have a valid request_limit, initiate a scan */
1640                 if (atomic_read(&hostdata->request_limit) > 0)
1641                         scsi_scan_host(host);
1642         }
1643
1644         vdev->dev.driver_data = hostdata;
1645         return 0;
1646
1647       add_srp_port_failed:
1648         scsi_remove_host(hostdata->host);
1649       add_host_failed:
1650         release_event_pool(&hostdata->pool, hostdata);
1651       init_pool_failed:
1652         ibmvscsi_ops->release_crq_queue(&hostdata->queue, hostdata, max_requests);
1653       init_crq_failed:
1654         scsi_host_put(host);
1655       scsi_host_alloc_failed:
1656         return -1;
1657 }
1658
1659 static int ibmvscsi_remove(struct vio_dev *vdev)
1660 {
1661         struct ibmvscsi_host_data *hostdata = vdev->dev.driver_data;
1662         release_event_pool(&hostdata->pool, hostdata);
1663         ibmvscsi_ops->release_crq_queue(&hostdata->queue, hostdata,
1664                                         max_requests);
1665
1666         srp_remove_host(hostdata->host);
1667         scsi_remove_host(hostdata->host);
1668         scsi_host_put(hostdata->host);
1669
1670         return 0;
1671 }
1672
1673 /**
1674  * ibmvscsi_device_table: Used by vio.c to match devices in the device tree we 
1675  * support.
1676  */
1677 static struct vio_device_id ibmvscsi_device_table[] __devinitdata = {
1678         {"vscsi", "IBM,v-scsi"},
1679         { "", "" }
1680 };
1681 MODULE_DEVICE_TABLE(vio, ibmvscsi_device_table);
1682
1683 static struct vio_driver ibmvscsi_driver = {
1684         .id_table = ibmvscsi_device_table,
1685         .probe = ibmvscsi_probe,
1686         .remove = ibmvscsi_remove,
1687         .driver = {
1688                 .name = "ibmvscsi",
1689                 .owner = THIS_MODULE,
1690         }
1691 };
1692
1693 static struct srp_function_template ibmvscsi_transport_functions = {
1694 };
1695
1696 int __init ibmvscsi_module_init(void)
1697 {
1698         int ret;
1699
1700         if (firmware_has_feature(FW_FEATURE_ISERIES))
1701                 ibmvscsi_ops = &iseriesvscsi_ops;
1702         else if (firmware_has_feature(FW_FEATURE_VIO))
1703                 ibmvscsi_ops = &rpavscsi_ops;
1704         else
1705                 return -ENODEV;
1706
1707         ibmvscsi_transport_template =
1708                 srp_attach_transport(&ibmvscsi_transport_functions);
1709         if (!ibmvscsi_transport_template)
1710                 return -ENOMEM;
1711
1712         ret = vio_register_driver(&ibmvscsi_driver);
1713         if (ret)
1714                 srp_release_transport(ibmvscsi_transport_template);
1715         return ret;
1716 }
1717
1718 void __exit ibmvscsi_module_exit(void)
1719 {
1720         vio_unregister_driver(&ibmvscsi_driver);
1721         srp_release_transport(ibmvscsi_transport_template);
1722 }
1723
1724 module_init(ibmvscsi_module_init);
1725 module_exit(ibmvscsi_module_exit);