KVM: Move apic timer migration away from critical section
[sfrench/cifs-2.6.git] / drivers / scsi / ibmvscsi / ibmvscsi.c
1 /* ------------------------------------------------------------
2  * ibmvscsi.c
3  * (C) Copyright IBM Corporation 1994, 2004
4  * Authors: Colin DeVilbiss (devilbis@us.ibm.com)
5  *          Santiago Leon (santil@us.ibm.com)
6  *          Dave Boutcher (sleddog@us.ibm.com)
7  *
8  * This program is free software; you can redistribute it and/or modify
9  * it under the terms of the GNU General Public License as published by
10  * the Free Software Foundation; either version 2 of the License, or
11  * (at your option) any later version.
12  *
13  * This program is distributed in the hope that it will be useful,
14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16  * GNU General Public License for more details.
17  *
18  * You should have received a copy of the GNU General Public License
19  * along with this program; if not, write to the Free Software
20  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307
21  * USA
22  *
23  * ------------------------------------------------------------
24  * Emulation of a SCSI host adapter for Virtual I/O devices
25  *
26  * This driver supports the SCSI adapter implemented by the IBM
27  * Power5 firmware.  That SCSI adapter is not a physical adapter,
28  * but allows Linux SCSI peripheral drivers to directly
29  * access devices in another logical partition on the physical system.
30  *
31  * The virtual adapter(s) are present in the open firmware device
32  * tree just like real adapters.
33  *
34  * One of the capabilities provided on these systems is the ability
35  * to DMA between partitions.  The architecture states that for VSCSI,
36  * the server side is allowed to DMA to and from the client.  The client
37  * is never trusted to DMA to or from the server directly.
38  *
39  * Messages are sent between partitions on a "Command/Response Queue" 
40  * (CRQ), which is just a buffer of 16 byte entries in the receiver's 
41  * Senders cannot access the buffer directly, but send messages by
42  * making a hypervisor call and passing in the 16 bytes.  The hypervisor
43  * puts the message in the next 16 byte space in round-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                 /* If send_crq returns H_CLOSED, return SCSI_MLQUEUE_HOST_BUSY.
633                  * Firmware will send a CRQ with a transport event (0xFF) to
634                  * tell this client what has happened to the transport.  This
635                  * will be handled in ibmvscsi_handle_crq()
636                  */
637                 if (rc == H_CLOSED) {
638                         dev_warn(hostdata->dev, "send warning. "
639                                  "Receive queue closed, will retry.\n");
640                         goto send_busy;
641                 }
642                 dev_err(hostdata->dev, "send error %d\n", rc);
643                 atomic_inc(&hostdata->request_limit);
644                 goto send_error;
645         }
646
647         return 0;
648
649  send_busy:
650         unmap_cmd_data(&evt_struct->iu.srp.cmd, evt_struct, hostdata->dev);
651
652         free_event_struct(&hostdata->pool, evt_struct);
653         if (request_status != -1)
654                 atomic_inc(&hostdata->request_limit);
655         return SCSI_MLQUEUE_HOST_BUSY;
656
657  send_error:
658         unmap_cmd_data(&evt_struct->iu.srp.cmd, evt_struct, hostdata->dev);
659
660         if (evt_struct->cmnd != NULL) {
661                 evt_struct->cmnd->result = DID_ERROR << 16;
662                 evt_struct->cmnd_done(evt_struct->cmnd);
663         } else if (evt_struct->done)
664                 evt_struct->done(evt_struct);
665
666         free_event_struct(&hostdata->pool, evt_struct);
667         return 0;
668 }
669
670 /**
671  * handle_cmd_rsp: -  Handle responses from commands
672  * @evt_struct: srp_event_struct to be handled
673  *
674  * Used as a callback by when sending scsi cmds.
675  * Gets called by ibmvscsi_handle_crq()
676 */
677 static void handle_cmd_rsp(struct srp_event_struct *evt_struct)
678 {
679         struct srp_rsp *rsp = &evt_struct->xfer_iu->srp.rsp;
680         struct scsi_cmnd *cmnd = evt_struct->cmnd;
681
682         if (unlikely(rsp->opcode != SRP_RSP)) {
683                 if (printk_ratelimit())
684                         dev_warn(evt_struct->hostdata->dev,
685                                  "bad SRP RSP type %d\n", rsp->opcode);
686         }
687         
688         if (cmnd) {
689                 cmnd->result = rsp->status;
690                 if (((cmnd->result >> 1) & 0x1f) == CHECK_CONDITION)
691                         memcpy(cmnd->sense_buffer,
692                                rsp->data,
693                                rsp->sense_data_len);
694                 unmap_cmd_data(&evt_struct->iu.srp.cmd, 
695                                evt_struct, 
696                                evt_struct->hostdata->dev);
697
698                 if (rsp->flags & SRP_RSP_FLAG_DOOVER)
699                         scsi_set_resid(cmnd, rsp->data_out_res_cnt);
700                 else if (rsp->flags & SRP_RSP_FLAG_DIOVER)
701                         scsi_set_resid(cmnd, rsp->data_in_res_cnt);
702         }
703
704         if (evt_struct->cmnd_done)
705                 evt_struct->cmnd_done(cmnd);
706 }
707
708 /**
709  * lun_from_dev: - Returns the lun of the scsi device
710  * @dev:        struct scsi_device
711  *
712 */
713 static inline u16 lun_from_dev(struct scsi_device *dev)
714 {
715         return (0x2 << 14) | (dev->id << 8) | (dev->channel << 5) | dev->lun;
716 }
717
718 /**
719  * ibmvscsi_queue: - The queuecommand function of the scsi template 
720  * @cmd:        struct scsi_cmnd to be executed
721  * @done:       Callback function to be called when cmd is completed
722 */
723 static int ibmvscsi_queuecommand(struct scsi_cmnd *cmnd,
724                                  void (*done) (struct scsi_cmnd *))
725 {
726         struct srp_cmd *srp_cmd;
727         struct srp_event_struct *evt_struct;
728         struct srp_indirect_buf *indirect;
729         struct ibmvscsi_host_data *hostdata = shost_priv(cmnd->device->host);
730         u16 lun = lun_from_dev(cmnd->device);
731         u8 out_fmt, in_fmt;
732
733         evt_struct = get_event_struct(&hostdata->pool);
734         if (!evt_struct)
735                 return SCSI_MLQUEUE_HOST_BUSY;
736
737         /* Set up the actual SRP IU */
738         srp_cmd = &evt_struct->iu.srp.cmd;
739         memset(srp_cmd, 0x00, SRP_MAX_IU_LEN);
740         srp_cmd->opcode = SRP_CMD;
741         memcpy(srp_cmd->cdb, cmnd->cmnd, sizeof(cmnd->cmnd));
742         srp_cmd->lun = ((u64) lun) << 48;
743
744         if (!map_data_for_srp_cmd(cmnd, evt_struct, srp_cmd, hostdata->dev)) {
745                 sdev_printk(KERN_ERR, cmnd->device, "couldn't convert cmd to srp_cmd\n");
746                 free_event_struct(&hostdata->pool, evt_struct);
747                 return SCSI_MLQUEUE_HOST_BUSY;
748         }
749
750         init_event_struct(evt_struct,
751                           handle_cmd_rsp,
752                           VIOSRP_SRP_FORMAT,
753                           cmnd->timeout_per_command/HZ);
754
755         evt_struct->cmnd = cmnd;
756         evt_struct->cmnd_done = done;
757
758         /* Fix up dma address of the buffer itself */
759         indirect = (struct srp_indirect_buf *) srp_cmd->add_data;
760         out_fmt = srp_cmd->buf_fmt >> 4;
761         in_fmt = srp_cmd->buf_fmt & ((1U << 4) - 1);
762         if ((in_fmt == SRP_DATA_DESC_INDIRECT ||
763              out_fmt == SRP_DATA_DESC_INDIRECT) &&
764             indirect->table_desc.va == 0) {
765                 indirect->table_desc.va = evt_struct->crq.IU_data_ptr +
766                         offsetof(struct srp_cmd, add_data) +
767                         offsetof(struct srp_indirect_buf, desc_list);
768         }
769
770         return ibmvscsi_send_srp_event(evt_struct, hostdata, 0);
771 }
772
773 /* ------------------------------------------------------------
774  * Routines for driver initialization
775  */
776 /**
777  * adapter_info_rsp: - Handle response to MAD adapter info request
778  * @evt_struct: srp_event_struct with the response
779  *
780  * Used as a "done" callback by when sending adapter_info. Gets called
781  * by ibmvscsi_handle_crq()
782 */
783 static void adapter_info_rsp(struct srp_event_struct *evt_struct)
784 {
785         struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
786         dma_unmap_single(hostdata->dev,
787                          evt_struct->iu.mad.adapter_info.buffer,
788                          evt_struct->iu.mad.adapter_info.common.length,
789                          DMA_BIDIRECTIONAL);
790
791         if (evt_struct->xfer_iu->mad.adapter_info.common.status) {
792                 dev_err(hostdata->dev, "error %d getting adapter info\n",
793                         evt_struct->xfer_iu->mad.adapter_info.common.status);
794         } else {
795                 dev_info(hostdata->dev, "host srp version: %s, "
796                          "host partition %s (%d), OS %d, max io %u\n",
797                          hostdata->madapter_info.srp_version,
798                          hostdata->madapter_info.partition_name,
799                          hostdata->madapter_info.partition_number,
800                          hostdata->madapter_info.os_type,
801                          hostdata->madapter_info.port_max_txu[0]);
802                 
803                 if (hostdata->madapter_info.port_max_txu[0]) 
804                         hostdata->host->max_sectors = 
805                                 hostdata->madapter_info.port_max_txu[0] >> 9;
806                 
807                 if (hostdata->madapter_info.os_type == 3 &&
808                     strcmp(hostdata->madapter_info.srp_version, "1.6a") <= 0) {
809                         dev_err(hostdata->dev, "host (Ver. %s) doesn't support large transfers\n",
810                                 hostdata->madapter_info.srp_version);
811                         dev_err(hostdata->dev, "limiting scatterlists to %d\n",
812                                 MAX_INDIRECT_BUFS);
813                         hostdata->host->sg_tablesize = MAX_INDIRECT_BUFS;
814                 }
815         }
816 }
817
818 /**
819  * send_mad_adapter_info: - Sends the mad adapter info request
820  *      and stores the result so it can be retrieved with
821  *      sysfs.  We COULD consider causing a failure if the
822  *      returned SRP version doesn't match ours.
823  * @hostdata:   ibmvscsi_host_data of host
824  * 
825  * Returns zero if successful.
826 */
827 static void send_mad_adapter_info(struct ibmvscsi_host_data *hostdata)
828 {
829         struct viosrp_adapter_info *req;
830         struct srp_event_struct *evt_struct;
831         unsigned long flags;
832         dma_addr_t addr;
833
834         evt_struct = get_event_struct(&hostdata->pool);
835         if (!evt_struct) {
836                 dev_err(hostdata->dev,
837                         "couldn't allocate an event for ADAPTER_INFO_REQ!\n");
838                 return;
839         }
840
841         init_event_struct(evt_struct,
842                           adapter_info_rsp,
843                           VIOSRP_MAD_FORMAT,
844                           init_timeout);
845         
846         req = &evt_struct->iu.mad.adapter_info;
847         memset(req, 0x00, sizeof(*req));
848         
849         req->common.type = VIOSRP_ADAPTER_INFO_TYPE;
850         req->common.length = sizeof(hostdata->madapter_info);
851         req->buffer = addr = dma_map_single(hostdata->dev,
852                                             &hostdata->madapter_info,
853                                             sizeof(hostdata->madapter_info),
854                                             DMA_BIDIRECTIONAL);
855
856         if (dma_mapping_error(req->buffer)) {
857                 dev_err(hostdata->dev, "Unable to map request_buffer for adapter_info!\n");
858                 free_event_struct(&hostdata->pool, evt_struct);
859                 return;
860         }
861         
862         spin_lock_irqsave(hostdata->host->host_lock, flags);
863         if (ibmvscsi_send_srp_event(evt_struct, hostdata, init_timeout * 2)) {
864                 dev_err(hostdata->dev, "couldn't send ADAPTER_INFO_REQ!\n");
865                 dma_unmap_single(hostdata->dev,
866                                  addr,
867                                  sizeof(hostdata->madapter_info),
868                                  DMA_BIDIRECTIONAL);
869         }
870         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
871 };
872
873 /**
874  * login_rsp: - Handle response to SRP login request
875  * @evt_struct: srp_event_struct with the response
876  *
877  * Used as a "done" callback by when sending srp_login. Gets called
878  * by ibmvscsi_handle_crq()
879 */
880 static void login_rsp(struct srp_event_struct *evt_struct)
881 {
882         struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
883         switch (evt_struct->xfer_iu->srp.login_rsp.opcode) {
884         case SRP_LOGIN_RSP:     /* it worked! */
885                 break;
886         case SRP_LOGIN_REJ:     /* refused! */
887                 dev_info(hostdata->dev, "SRP_LOGIN_REJ reason %u\n",
888                          evt_struct->xfer_iu->srp.login_rej.reason);
889                 /* Login failed.  */
890                 atomic_set(&hostdata->request_limit, -1);
891                 return;
892         default:
893                 dev_err(hostdata->dev, "Invalid login response typecode 0x%02x!\n",
894                         evt_struct->xfer_iu->srp.login_rsp.opcode);
895                 /* Login failed.  */
896                 atomic_set(&hostdata->request_limit, -1);
897                 return;
898         }
899
900         dev_info(hostdata->dev, "SRP_LOGIN succeeded\n");
901
902         if (evt_struct->xfer_iu->srp.login_rsp.req_lim_delta < 0)
903                 dev_err(hostdata->dev, "Invalid request_limit.\n");
904
905         /* Now we know what the real request-limit is.
906          * This value is set rather than added to request_limit because
907          * request_limit could have been set to -1 by this client.
908          */
909         atomic_set(&hostdata->request_limit,
910                    evt_struct->xfer_iu->srp.login_rsp.req_lim_delta);
911
912         /* If we had any pending I/Os, kick them */
913         scsi_unblock_requests(hostdata->host);
914
915         send_mad_adapter_info(hostdata);
916         return;
917 }
918
919 /**
920  * send_srp_login: - Sends the srp login
921  * @hostdata:   ibmvscsi_host_data of host
922  * 
923  * Returns zero if successful.
924 */
925 static int send_srp_login(struct ibmvscsi_host_data *hostdata)
926 {
927         int rc;
928         unsigned long flags;
929         struct srp_login_req *login;
930         struct srp_event_struct *evt_struct = get_event_struct(&hostdata->pool);
931         if (!evt_struct) {
932                 dev_err(hostdata->dev, "couldn't allocate an event for login req!\n");
933                 return FAILED;
934         }
935
936         init_event_struct(evt_struct,
937                           login_rsp,
938                           VIOSRP_SRP_FORMAT,
939                           init_timeout);
940
941         login = &evt_struct->iu.srp.login_req;
942         memset(login, 0x00, sizeof(struct srp_login_req));
943         login->opcode = SRP_LOGIN_REQ;
944         login->req_it_iu_len = sizeof(union srp_iu);
945         login->req_buf_fmt = SRP_BUF_FORMAT_DIRECT | SRP_BUF_FORMAT_INDIRECT;
946         
947         spin_lock_irqsave(hostdata->host->host_lock, flags);
948         /* Start out with a request limit of 0, since this is negotiated in
949          * the login request we are just sending and login requests always
950          * get sent by the driver regardless of request_limit.
951          */
952         atomic_set(&hostdata->request_limit, 0);
953
954         rc = ibmvscsi_send_srp_event(evt_struct, hostdata, init_timeout * 2);
955         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
956         dev_info(hostdata->dev, "sent SRP login\n");
957         return rc;
958 };
959
960 /**
961  * sync_completion: Signal that a synchronous command has completed
962  * Note that after returning from this call, the evt_struct is freed.
963  * the caller waiting on this completion shouldn't touch the evt_struct
964  * again.
965  */
966 static void sync_completion(struct srp_event_struct *evt_struct)
967 {
968         /* copy the response back */
969         if (evt_struct->sync_srp)
970                 *evt_struct->sync_srp = *evt_struct->xfer_iu;
971         
972         complete(&evt_struct->comp);
973 }
974
975 /**
976  * ibmvscsi_abort: Abort a command...from scsi host template
977  * send this over to the server and wait synchronously for the response
978  */
979 static int ibmvscsi_eh_abort_handler(struct scsi_cmnd *cmd)
980 {
981         struct ibmvscsi_host_data *hostdata = shost_priv(cmd->device->host);
982         struct srp_tsk_mgmt *tsk_mgmt;
983         struct srp_event_struct *evt;
984         struct srp_event_struct *tmp_evt, *found_evt;
985         union viosrp_iu srp_rsp;
986         int rsp_rc;
987         unsigned long flags;
988         u16 lun = lun_from_dev(cmd->device);
989         unsigned long wait_switch = 0;
990
991         /* First, find this command in our sent list so we can figure
992          * out the correct tag
993          */
994         spin_lock_irqsave(hostdata->host->host_lock, flags);
995         wait_switch = jiffies + (init_timeout * HZ);
996         do {
997                 found_evt = NULL;
998                 list_for_each_entry(tmp_evt, &hostdata->sent, list) {
999                         if (tmp_evt->cmnd == cmd) {
1000                                 found_evt = tmp_evt;
1001                                 break;
1002                         }
1003                 }
1004
1005                 if (!found_evt) {
1006                         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1007                         return SUCCESS;
1008                 }
1009
1010                 evt = get_event_struct(&hostdata->pool);
1011                 if (evt == NULL) {
1012                         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1013                         sdev_printk(KERN_ERR, cmd->device,
1014                                 "failed to allocate abort event\n");
1015                         return FAILED;
1016                 }
1017         
1018                 init_event_struct(evt,
1019                                   sync_completion,
1020                                   VIOSRP_SRP_FORMAT,
1021                                   init_timeout);
1022
1023                 tsk_mgmt = &evt->iu.srp.tsk_mgmt;
1024         
1025                 /* Set up an abort SRP command */
1026                 memset(tsk_mgmt, 0x00, sizeof(*tsk_mgmt));
1027                 tsk_mgmt->opcode = SRP_TSK_MGMT;
1028                 tsk_mgmt->lun = ((u64) lun) << 48;
1029                 tsk_mgmt->tsk_mgmt_func = SRP_TSK_ABORT_TASK;
1030                 tsk_mgmt->task_tag = (u64) found_evt;
1031
1032                 evt->sync_srp = &srp_rsp;
1033
1034                 init_completion(&evt->comp);
1035                 rsp_rc = ibmvscsi_send_srp_event(evt, hostdata, init_timeout * 2);
1036
1037                 if (rsp_rc != SCSI_MLQUEUE_HOST_BUSY)
1038                         break;
1039
1040                 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1041                 msleep(10);
1042                 spin_lock_irqsave(hostdata->host->host_lock, flags);
1043         } while (time_before(jiffies, wait_switch));
1044
1045         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1046
1047         if (rsp_rc != 0) {
1048                 sdev_printk(KERN_ERR, cmd->device,
1049                             "failed to send abort() event. rc=%d\n", rsp_rc);
1050                 return FAILED;
1051         }
1052
1053         sdev_printk(KERN_INFO, cmd->device,
1054                     "aborting command. lun 0x%lx, tag 0x%lx\n",
1055                     (((u64) lun) << 48), (u64) found_evt);
1056
1057         wait_for_completion(&evt->comp);
1058
1059         /* make sure we got a good response */
1060         if (unlikely(srp_rsp.srp.rsp.opcode != SRP_RSP)) {
1061                 if (printk_ratelimit())
1062                         sdev_printk(KERN_WARNING, cmd->device, "abort bad SRP RSP type %d\n",
1063                                     srp_rsp.srp.rsp.opcode);
1064                 return FAILED;
1065         }
1066
1067         if (srp_rsp.srp.rsp.flags & SRP_RSP_FLAG_RSPVALID)
1068                 rsp_rc = *((int *)srp_rsp.srp.rsp.data);
1069         else
1070                 rsp_rc = srp_rsp.srp.rsp.status;
1071
1072         if (rsp_rc) {
1073                 if (printk_ratelimit())
1074                         sdev_printk(KERN_WARNING, cmd->device,
1075                                     "abort code %d for task tag 0x%lx\n",
1076                                     rsp_rc, tsk_mgmt->task_tag);
1077                 return FAILED;
1078         }
1079
1080         /* Because we dropped the spinlock above, it's possible
1081          * The event is no longer in our list.  Make sure it didn't
1082          * complete while we were aborting
1083          */
1084         spin_lock_irqsave(hostdata->host->host_lock, flags);
1085         found_evt = NULL;
1086         list_for_each_entry(tmp_evt, &hostdata->sent, list) {
1087                 if (tmp_evt->cmnd == cmd) {
1088                         found_evt = tmp_evt;
1089                         break;
1090                 }
1091         }
1092
1093         if (found_evt == NULL) {
1094                 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1095                 sdev_printk(KERN_INFO, cmd->device, "aborted task tag 0x%lx completed\n",
1096                             tsk_mgmt->task_tag);
1097                 return SUCCESS;
1098         }
1099
1100         sdev_printk(KERN_INFO, cmd->device, "successfully aborted task tag 0x%lx\n",
1101                     tsk_mgmt->task_tag);
1102
1103         cmd->result = (DID_ABORT << 16);
1104         list_del(&found_evt->list);
1105         unmap_cmd_data(&found_evt->iu.srp.cmd, found_evt,
1106                        found_evt->hostdata->dev);
1107         free_event_struct(&found_evt->hostdata->pool, found_evt);
1108         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1109         atomic_inc(&hostdata->request_limit);
1110         return SUCCESS;
1111 }
1112
1113 /**
1114  * ibmvscsi_eh_device_reset_handler: Reset a single LUN...from scsi host 
1115  * template send this over to the server and wait synchronously for the 
1116  * response
1117  */
1118 static int ibmvscsi_eh_device_reset_handler(struct scsi_cmnd *cmd)
1119 {
1120         struct ibmvscsi_host_data *hostdata = shost_priv(cmd->device->host);
1121         struct srp_tsk_mgmt *tsk_mgmt;
1122         struct srp_event_struct *evt;
1123         struct srp_event_struct *tmp_evt, *pos;
1124         union viosrp_iu srp_rsp;
1125         int rsp_rc;
1126         unsigned long flags;
1127         u16 lun = lun_from_dev(cmd->device);
1128         unsigned long wait_switch = 0;
1129
1130         spin_lock_irqsave(hostdata->host->host_lock, flags);
1131         wait_switch = jiffies + (init_timeout * HZ);
1132         do {
1133                 evt = get_event_struct(&hostdata->pool);
1134                 if (evt == NULL) {
1135                         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1136                         sdev_printk(KERN_ERR, cmd->device,
1137                                 "failed to allocate reset event\n");
1138                         return FAILED;
1139                 }
1140         
1141                 init_event_struct(evt,
1142                                   sync_completion,
1143                                   VIOSRP_SRP_FORMAT,
1144                                   init_timeout);
1145
1146                 tsk_mgmt = &evt->iu.srp.tsk_mgmt;
1147
1148                 /* Set up a lun reset SRP command */
1149                 memset(tsk_mgmt, 0x00, sizeof(*tsk_mgmt));
1150                 tsk_mgmt->opcode = SRP_TSK_MGMT;
1151                 tsk_mgmt->lun = ((u64) lun) << 48;
1152                 tsk_mgmt->tsk_mgmt_func = SRP_TSK_LUN_RESET;
1153
1154                 evt->sync_srp = &srp_rsp;
1155
1156                 init_completion(&evt->comp);
1157                 rsp_rc = ibmvscsi_send_srp_event(evt, hostdata, init_timeout * 2);
1158
1159                 if (rsp_rc != SCSI_MLQUEUE_HOST_BUSY)
1160                         break;
1161
1162                 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1163                 msleep(10);
1164                 spin_lock_irqsave(hostdata->host->host_lock, flags);
1165         } while (time_before(jiffies, wait_switch));
1166
1167         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1168
1169         if (rsp_rc != 0) {
1170                 sdev_printk(KERN_ERR, cmd->device,
1171                             "failed to send reset event. rc=%d\n", rsp_rc);
1172                 return FAILED;
1173         }
1174
1175         sdev_printk(KERN_INFO, cmd->device, "resetting device. lun 0x%lx\n",
1176                     (((u64) lun) << 48));
1177
1178         wait_for_completion(&evt->comp);
1179
1180         /* make sure we got a good response */
1181         if (unlikely(srp_rsp.srp.rsp.opcode != SRP_RSP)) {
1182                 if (printk_ratelimit())
1183                         sdev_printk(KERN_WARNING, cmd->device, "reset bad SRP RSP type %d\n",
1184                                     srp_rsp.srp.rsp.opcode);
1185                 return FAILED;
1186         }
1187
1188         if (srp_rsp.srp.rsp.flags & SRP_RSP_FLAG_RSPVALID)
1189                 rsp_rc = *((int *)srp_rsp.srp.rsp.data);
1190         else
1191                 rsp_rc = srp_rsp.srp.rsp.status;
1192
1193         if (rsp_rc) {
1194                 if (printk_ratelimit())
1195                         sdev_printk(KERN_WARNING, cmd->device,
1196                                     "reset code %d for task tag 0x%lx\n",
1197                                     rsp_rc, tsk_mgmt->task_tag);
1198                 return FAILED;
1199         }
1200
1201         /* We need to find all commands for this LUN that have not yet been
1202          * responded to, and fail them with DID_RESET
1203          */
1204         spin_lock_irqsave(hostdata->host->host_lock, flags);
1205         list_for_each_entry_safe(tmp_evt, pos, &hostdata->sent, list) {
1206                 if ((tmp_evt->cmnd) && (tmp_evt->cmnd->device == cmd->device)) {
1207                         if (tmp_evt->cmnd)
1208                                 tmp_evt->cmnd->result = (DID_RESET << 16);
1209                         list_del(&tmp_evt->list);
1210                         unmap_cmd_data(&tmp_evt->iu.srp.cmd, tmp_evt,
1211                                        tmp_evt->hostdata->dev);
1212                         free_event_struct(&tmp_evt->hostdata->pool,
1213                                                    tmp_evt);
1214                         atomic_inc(&hostdata->request_limit);
1215                         if (tmp_evt->cmnd_done)
1216                                 tmp_evt->cmnd_done(tmp_evt->cmnd);
1217                         else if (tmp_evt->done)
1218                                 tmp_evt->done(tmp_evt);
1219                 }
1220         }
1221         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1222         return SUCCESS;
1223 }
1224
1225 /**
1226  * ibmvscsi_eh_host_reset_handler - Reset the connection to the server
1227  * @cmd:        struct scsi_cmnd having problems
1228 */
1229 static int ibmvscsi_eh_host_reset_handler(struct scsi_cmnd *cmd)
1230 {
1231         unsigned long wait_switch = 0;
1232         struct ibmvscsi_host_data *hostdata = shost_priv(cmd->device->host);
1233
1234         dev_err(hostdata->dev, "Resetting connection due to error recovery\n");
1235
1236         ibmvscsi_reset_host(hostdata);
1237
1238         for (wait_switch = jiffies + (init_timeout * HZ);
1239              time_before(jiffies, wait_switch) &&
1240                      atomic_read(&hostdata->request_limit) < 2;) {
1241
1242                 msleep(10);
1243         }
1244
1245         if (atomic_read(&hostdata->request_limit) <= 0)
1246                 return FAILED;
1247
1248         return SUCCESS;
1249 }
1250
1251 /**
1252  * ibmvscsi_handle_crq: - Handles and frees received events in the CRQ
1253  * @crq:        Command/Response queue
1254  * @hostdata:   ibmvscsi_host_data of host
1255  *
1256 */
1257 void ibmvscsi_handle_crq(struct viosrp_crq *crq,
1258                          struct ibmvscsi_host_data *hostdata)
1259 {
1260         long rc;
1261         unsigned long flags;
1262         struct srp_event_struct *evt_struct =
1263             (struct srp_event_struct *)crq->IU_data_ptr;
1264         switch (crq->valid) {
1265         case 0xC0:              /* initialization */
1266                 switch (crq->format) {
1267                 case 0x01:      /* Initialization message */
1268                         dev_info(hostdata->dev, "partner initialized\n");
1269                         /* Send back a response */
1270                         if ((rc = ibmvscsi_ops->send_crq(hostdata,
1271                                                          0xC002000000000000LL, 0)) == 0) {
1272                                 /* Now login */
1273                                 send_srp_login(hostdata);
1274                         } else {
1275                                 dev_err(hostdata->dev, "Unable to send init rsp. rc=%ld\n", rc);
1276                         }
1277
1278                         break;
1279                 case 0x02:      /* Initialization response */
1280                         dev_info(hostdata->dev, "partner initialization complete\n");
1281
1282                         /* Now login */
1283                         send_srp_login(hostdata);
1284                         break;
1285                 default:
1286                         dev_err(hostdata->dev, "unknown crq message type: %d\n", crq->format);
1287                 }
1288                 return;
1289         case 0xFF:      /* Hypervisor telling us the connection is closed */
1290                 scsi_block_requests(hostdata->host);
1291                 atomic_set(&hostdata->request_limit, 0);
1292                 if (crq->format == 0x06) {
1293                         /* We need to re-setup the interpartition connection */
1294                         dev_info(hostdata->dev, "Re-enabling adapter!\n");
1295                         purge_requests(hostdata, DID_REQUEUE);
1296                         if ((ibmvscsi_ops->reenable_crq_queue(&hostdata->queue,
1297                                                               hostdata)) ||
1298                             (ibmvscsi_ops->send_crq(hostdata,
1299                                                     0xC001000000000000LL, 0))) {
1300                                         atomic_set(&hostdata->request_limit,
1301                                                    -1);
1302                                         dev_err(hostdata->dev, "error after enable\n");
1303                         }
1304                 } else {
1305                         dev_err(hostdata->dev, "Virtual adapter failed rc %d!\n",
1306                                 crq->format);
1307
1308                         purge_requests(hostdata, DID_ERROR);
1309                         if ((ibmvscsi_ops->reset_crq_queue(&hostdata->queue,
1310                                                            hostdata)) ||
1311                             (ibmvscsi_ops->send_crq(hostdata,
1312                                                     0xC001000000000000LL, 0))) {
1313                                         atomic_set(&hostdata->request_limit,
1314                                                    -1);
1315                                         dev_err(hostdata->dev, "error after reset\n");
1316                         }
1317                 }
1318                 scsi_unblock_requests(hostdata->host);
1319                 return;
1320         case 0x80:              /* real payload */
1321                 break;
1322         default:
1323                 dev_err(hostdata->dev, "got an invalid message type 0x%02x\n",
1324                         crq->valid);
1325                 return;
1326         }
1327
1328         /* The only kind of payload CRQs we should get are responses to
1329          * things we send. Make sure this response is to something we
1330          * actually sent
1331          */
1332         if (!valid_event_struct(&hostdata->pool, evt_struct)) {
1333                 dev_err(hostdata->dev, "returned correlation_token 0x%p is invalid!\n",
1334                        (void *)crq->IU_data_ptr);
1335                 return;
1336         }
1337
1338         if (atomic_read(&evt_struct->free)) {
1339                 dev_err(hostdata->dev, "received duplicate correlation_token 0x%p!\n",
1340                         (void *)crq->IU_data_ptr);
1341                 return;
1342         }
1343
1344         if (crq->format == VIOSRP_SRP_FORMAT)
1345                 atomic_add(evt_struct->xfer_iu->srp.rsp.req_lim_delta,
1346                            &hostdata->request_limit);
1347
1348         del_timer(&evt_struct->timer);
1349
1350         if (evt_struct->done)
1351                 evt_struct->done(evt_struct);
1352         else
1353                 dev_err(hostdata->dev, "returned done() is NULL; not running it!\n");
1354
1355         /*
1356          * Lock the host_lock before messing with these structures, since we
1357          * are running in a task context
1358          */
1359         spin_lock_irqsave(evt_struct->hostdata->host->host_lock, flags);
1360         list_del(&evt_struct->list);
1361         free_event_struct(&evt_struct->hostdata->pool, evt_struct);
1362         spin_unlock_irqrestore(evt_struct->hostdata->host->host_lock, flags);
1363 }
1364
1365 /**
1366  * ibmvscsi_get_host_config: Send the command to the server to get host
1367  * configuration data.  The data is opaque to us.
1368  */
1369 static int ibmvscsi_do_host_config(struct ibmvscsi_host_data *hostdata,
1370                                    unsigned char *buffer, int length)
1371 {
1372         struct viosrp_host_config *host_config;
1373         struct srp_event_struct *evt_struct;
1374         unsigned long flags;
1375         dma_addr_t addr;
1376         int rc;
1377
1378         evt_struct = get_event_struct(&hostdata->pool);
1379         if (!evt_struct) {
1380                 dev_err(hostdata->dev, "couldn't allocate event for HOST_CONFIG!\n");
1381                 return -1;
1382         }
1383
1384         init_event_struct(evt_struct,
1385                           sync_completion,
1386                           VIOSRP_MAD_FORMAT,
1387                           init_timeout);
1388
1389         host_config = &evt_struct->iu.mad.host_config;
1390
1391         /* Set up a lun reset SRP command */
1392         memset(host_config, 0x00, sizeof(*host_config));
1393         host_config->common.type = VIOSRP_HOST_CONFIG_TYPE;
1394         host_config->common.length = length;
1395         host_config->buffer = addr = dma_map_single(hostdata->dev, buffer,
1396                                                     length,
1397                                                     DMA_BIDIRECTIONAL);
1398
1399         if (dma_mapping_error(host_config->buffer)) {
1400                 dev_err(hostdata->dev, "dma_mapping error getting host config\n");
1401                 free_event_struct(&hostdata->pool, evt_struct);
1402                 return -1;
1403         }
1404
1405         init_completion(&evt_struct->comp);
1406         spin_lock_irqsave(hostdata->host->host_lock, flags);
1407         rc = ibmvscsi_send_srp_event(evt_struct, hostdata, init_timeout * 2);
1408         spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1409         if (rc == 0)
1410                 wait_for_completion(&evt_struct->comp);
1411         dma_unmap_single(hostdata->dev, addr, length, DMA_BIDIRECTIONAL);
1412
1413         return rc;
1414 }
1415
1416 /**
1417  * ibmvscsi_slave_configure: Set the "allow_restart" flag for each disk.
1418  * @sdev:       struct scsi_device device to configure
1419  *
1420  * Enable allow_restart for a device if it is a disk.  Adjust the
1421  * queue_depth here also as is required by the documentation for
1422  * struct scsi_host_template.
1423  */
1424 static int ibmvscsi_slave_configure(struct scsi_device *sdev)
1425 {
1426         struct Scsi_Host *shost = sdev->host;
1427         unsigned long lock_flags = 0;
1428
1429         spin_lock_irqsave(shost->host_lock, lock_flags);
1430         if (sdev->type == TYPE_DISK) {
1431                 sdev->allow_restart = 1;
1432                 sdev->timeout = 60 * HZ;
1433         }
1434         scsi_adjust_queue_depth(sdev, 0, shost->cmd_per_lun);
1435         spin_unlock_irqrestore(shost->host_lock, lock_flags);
1436         return 0;
1437 }
1438
1439 /**
1440  * ibmvscsi_change_queue_depth - Change the device's queue depth
1441  * @sdev:       scsi device struct
1442  * @qdepth:     depth to set
1443  *
1444  * Return value:
1445  *      actual depth set
1446  **/
1447 static int ibmvscsi_change_queue_depth(struct scsi_device *sdev, int qdepth)
1448 {
1449         if (qdepth > IBMVSCSI_MAX_CMDS_PER_LUN)
1450                 qdepth = IBMVSCSI_MAX_CMDS_PER_LUN;
1451
1452         scsi_adjust_queue_depth(sdev, 0, qdepth);
1453         return sdev->queue_depth;
1454 }
1455
1456 /* ------------------------------------------------------------
1457  * sysfs attributes
1458  */
1459 static ssize_t show_host_srp_version(struct class_device *class_dev, char *buf)
1460 {
1461         struct Scsi_Host *shost = class_to_shost(class_dev);
1462         struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1463         int len;
1464
1465         len = snprintf(buf, PAGE_SIZE, "%s\n",
1466                        hostdata->madapter_info.srp_version);
1467         return len;
1468 }
1469
1470 static struct class_device_attribute ibmvscsi_host_srp_version = {
1471         .attr = {
1472                  .name = "srp_version",
1473                  .mode = S_IRUGO,
1474                  },
1475         .show = show_host_srp_version,
1476 };
1477
1478 static ssize_t show_host_partition_name(struct class_device *class_dev,
1479                                         char *buf)
1480 {
1481         struct Scsi_Host *shost = class_to_shost(class_dev);
1482         struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1483         int len;
1484
1485         len = snprintf(buf, PAGE_SIZE, "%s\n",
1486                        hostdata->madapter_info.partition_name);
1487         return len;
1488 }
1489
1490 static struct class_device_attribute ibmvscsi_host_partition_name = {
1491         .attr = {
1492                  .name = "partition_name",
1493                  .mode = S_IRUGO,
1494                  },
1495         .show = show_host_partition_name,
1496 };
1497
1498 static ssize_t show_host_partition_number(struct class_device *class_dev,
1499                                           char *buf)
1500 {
1501         struct Scsi_Host *shost = class_to_shost(class_dev);
1502         struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1503         int len;
1504
1505         len = snprintf(buf, PAGE_SIZE, "%d\n",
1506                        hostdata->madapter_info.partition_number);
1507         return len;
1508 }
1509
1510 static struct class_device_attribute ibmvscsi_host_partition_number = {
1511         .attr = {
1512                  .name = "partition_number",
1513                  .mode = S_IRUGO,
1514                  },
1515         .show = show_host_partition_number,
1516 };
1517
1518 static ssize_t show_host_mad_version(struct class_device *class_dev, char *buf)
1519 {
1520         struct Scsi_Host *shost = class_to_shost(class_dev);
1521         struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1522         int len;
1523
1524         len = snprintf(buf, PAGE_SIZE, "%d\n",
1525                        hostdata->madapter_info.mad_version);
1526         return len;
1527 }
1528
1529 static struct class_device_attribute ibmvscsi_host_mad_version = {
1530         .attr = {
1531                  .name = "mad_version",
1532                  .mode = S_IRUGO,
1533                  },
1534         .show = show_host_mad_version,
1535 };
1536
1537 static ssize_t show_host_os_type(struct class_device *class_dev, char *buf)
1538 {
1539         struct Scsi_Host *shost = class_to_shost(class_dev);
1540         struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1541         int len;
1542
1543         len = snprintf(buf, PAGE_SIZE, "%d\n", hostdata->madapter_info.os_type);
1544         return len;
1545 }
1546
1547 static struct class_device_attribute ibmvscsi_host_os_type = {
1548         .attr = {
1549                  .name = "os_type",
1550                  .mode = S_IRUGO,
1551                  },
1552         .show = show_host_os_type,
1553 };
1554
1555 static ssize_t show_host_config(struct class_device *class_dev, char *buf)
1556 {
1557         struct Scsi_Host *shost = class_to_shost(class_dev);
1558         struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1559
1560         /* returns null-terminated host config data */
1561         if (ibmvscsi_do_host_config(hostdata, buf, PAGE_SIZE) == 0)
1562                 return strlen(buf);
1563         else
1564                 return 0;
1565 }
1566
1567 static struct class_device_attribute ibmvscsi_host_config = {
1568         .attr = {
1569                  .name = "config",
1570                  .mode = S_IRUGO,
1571                  },
1572         .show = show_host_config,
1573 };
1574
1575 static struct class_device_attribute *ibmvscsi_attrs[] = {
1576         &ibmvscsi_host_srp_version,
1577         &ibmvscsi_host_partition_name,
1578         &ibmvscsi_host_partition_number,
1579         &ibmvscsi_host_mad_version,
1580         &ibmvscsi_host_os_type,
1581         &ibmvscsi_host_config,
1582         NULL
1583 };
1584
1585 /* ------------------------------------------------------------
1586  * SCSI driver registration
1587  */
1588 static struct scsi_host_template driver_template = {
1589         .module = THIS_MODULE,
1590         .name = "IBM POWER Virtual SCSI Adapter " IBMVSCSI_VERSION,
1591         .proc_name = "ibmvscsi",
1592         .queuecommand = ibmvscsi_queuecommand,
1593         .eh_abort_handler = ibmvscsi_eh_abort_handler,
1594         .eh_device_reset_handler = ibmvscsi_eh_device_reset_handler,
1595         .eh_host_reset_handler = ibmvscsi_eh_host_reset_handler,
1596         .slave_configure = ibmvscsi_slave_configure,
1597         .change_queue_depth = ibmvscsi_change_queue_depth,
1598         .cmd_per_lun = 16,
1599         .can_queue = IBMVSCSI_MAX_REQUESTS_DEFAULT,
1600         .this_id = -1,
1601         .sg_tablesize = SG_ALL,
1602         .use_clustering = ENABLE_CLUSTERING,
1603         .use_sg_chaining = ENABLE_SG_CHAINING,
1604         .shost_attrs = ibmvscsi_attrs,
1605 };
1606
1607 /**
1608  * Called by bus code for each adapter
1609  */
1610 static int ibmvscsi_probe(struct vio_dev *vdev, const struct vio_device_id *id)
1611 {
1612         struct ibmvscsi_host_data *hostdata;
1613         struct Scsi_Host *host;
1614         struct device *dev = &vdev->dev;
1615         struct srp_rport_identifiers ids;
1616         struct srp_rport *rport;
1617         unsigned long wait_switch = 0;
1618         int rc;
1619
1620         vdev->dev.driver_data = NULL;
1621
1622         driver_template.can_queue = max_requests;
1623         host = scsi_host_alloc(&driver_template, sizeof(*hostdata));
1624         if (!host) {
1625                 dev_err(&vdev->dev, "couldn't allocate host data\n");
1626                 goto scsi_host_alloc_failed;
1627         }
1628
1629         host->transportt = ibmvscsi_transport_template;
1630         hostdata = shost_priv(host);
1631         memset(hostdata, 0x00, sizeof(*hostdata));
1632         INIT_LIST_HEAD(&hostdata->sent);
1633         hostdata->host = host;
1634         hostdata->dev = dev;
1635         atomic_set(&hostdata->request_limit, -1);
1636         hostdata->host->max_sectors = 32 * 8; /* default max I/O 32 pages */
1637
1638         rc = ibmvscsi_ops->init_crq_queue(&hostdata->queue, hostdata, max_requests);
1639         if (rc != 0 && rc != H_RESOURCE) {
1640                 dev_err(&vdev->dev, "couldn't initialize crq. rc=%d\n", rc);
1641                 goto init_crq_failed;
1642         }
1643         if (initialize_event_pool(&hostdata->pool, max_requests, hostdata) != 0) {
1644                 dev_err(&vdev->dev, "couldn't initialize event pool\n");
1645                 goto init_pool_failed;
1646         }
1647
1648         host->max_lun = 8;
1649         host->max_id = max_id;
1650         host->max_channel = max_channel;
1651
1652         if (scsi_add_host(hostdata->host, hostdata->dev))
1653                 goto add_host_failed;
1654
1655         /* we don't have a proper target_port_id so let's use the fake one */
1656         memcpy(ids.port_id, hostdata->madapter_info.partition_name,
1657                sizeof(ids.port_id));
1658         ids.roles = SRP_RPORT_ROLE_TARGET;
1659         rport = srp_rport_add(host, &ids);
1660         if (IS_ERR(rport))
1661                 goto add_srp_port_failed;
1662
1663         /* Try to send an initialization message.  Note that this is allowed
1664          * to fail if the other end is not acive.  In that case we don't
1665          * want to scan
1666          */
1667         if (ibmvscsi_ops->send_crq(hostdata, 0xC001000000000000LL, 0) == 0
1668             || rc == H_RESOURCE) {
1669                 /*
1670                  * Wait around max init_timeout secs for the adapter to finish
1671                  * initializing. When we are done initializing, we will have a
1672                  * valid request_limit.  We don't want Linux scanning before
1673                  * we are ready.
1674                  */
1675                 for (wait_switch = jiffies + (init_timeout * HZ);
1676                      time_before(jiffies, wait_switch) &&
1677                      atomic_read(&hostdata->request_limit) < 2;) {
1678
1679                         msleep(10);
1680                 }
1681
1682                 /* if we now have a valid request_limit, initiate a scan */
1683                 if (atomic_read(&hostdata->request_limit) > 0)
1684                         scsi_scan_host(host);
1685         }
1686
1687         vdev->dev.driver_data = hostdata;
1688         return 0;
1689
1690       add_srp_port_failed:
1691         scsi_remove_host(hostdata->host);
1692       add_host_failed:
1693         release_event_pool(&hostdata->pool, hostdata);
1694       init_pool_failed:
1695         ibmvscsi_ops->release_crq_queue(&hostdata->queue, hostdata, max_requests);
1696       init_crq_failed:
1697         scsi_host_put(host);
1698       scsi_host_alloc_failed:
1699         return -1;
1700 }
1701
1702 static int ibmvscsi_remove(struct vio_dev *vdev)
1703 {
1704         struct ibmvscsi_host_data *hostdata = vdev->dev.driver_data;
1705         release_event_pool(&hostdata->pool, hostdata);
1706         ibmvscsi_ops->release_crq_queue(&hostdata->queue, hostdata,
1707                                         max_requests);
1708
1709         srp_remove_host(hostdata->host);
1710         scsi_remove_host(hostdata->host);
1711         scsi_host_put(hostdata->host);
1712
1713         return 0;
1714 }
1715
1716 /**
1717  * ibmvscsi_device_table: Used by vio.c to match devices in the device tree we 
1718  * support.
1719  */
1720 static struct vio_device_id ibmvscsi_device_table[] __devinitdata = {
1721         {"vscsi", "IBM,v-scsi"},
1722         { "", "" }
1723 };
1724 MODULE_DEVICE_TABLE(vio, ibmvscsi_device_table);
1725
1726 static struct vio_driver ibmvscsi_driver = {
1727         .id_table = ibmvscsi_device_table,
1728         .probe = ibmvscsi_probe,
1729         .remove = ibmvscsi_remove,
1730         .driver = {
1731                 .name = "ibmvscsi",
1732                 .owner = THIS_MODULE,
1733         }
1734 };
1735
1736 static struct srp_function_template ibmvscsi_transport_functions = {
1737 };
1738
1739 int __init ibmvscsi_module_init(void)
1740 {
1741         int ret;
1742
1743         if (firmware_has_feature(FW_FEATURE_ISERIES))
1744                 ibmvscsi_ops = &iseriesvscsi_ops;
1745         else if (firmware_has_feature(FW_FEATURE_VIO))
1746                 ibmvscsi_ops = &rpavscsi_ops;
1747         else
1748                 return -ENODEV;
1749
1750         ibmvscsi_transport_template =
1751                 srp_attach_transport(&ibmvscsi_transport_functions);
1752         if (!ibmvscsi_transport_template)
1753                 return -ENOMEM;
1754
1755         ret = vio_register_driver(&ibmvscsi_driver);
1756         if (ret)
1757                 srp_release_transport(ibmvscsi_transport_template);
1758         return ret;
1759 }
1760
1761 void __exit ibmvscsi_module_exit(void)
1762 {
1763         vio_unregister_driver(&ibmvscsi_driver);
1764         srp_release_transport(ibmvscsi_transport_template);
1765 }
1766
1767 module_init(ibmvscsi_module_init);
1768 module_exit(ibmvscsi_module_exit);