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