1 /* ------------------------------------------------------------
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)
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.
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.
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
23 * ------------------------------------------------------------
24 * Emulation of a SCSI host adapter for Virtual I/O devices
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.
31 * The virtual adapter(s) are present in the open firmware device
32 * tree just like real adapters.
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.
39 * Messages are sent between partitions on a "Command/Response Queue"
40 * (CRQ), which is just a buffer of 16 byte entries in the receiver's
41 * Senders cannot access the buffer directly, but send messages by
42 * making a hypervisor call and passing in the 16 bytes. The hypervisor
43 * puts the message in the next 16 byte space in round-robin fashion,
44 * turns on the high order bit of the message (the valid bit), and
45 * generates an interrupt to the receiver (if interrupts are turned on.)
46 * The receiver just turns off the valid bit when they have copied out
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
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.
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
69 #include <linux/module.h>
70 #include <linux/moduleparam.h>
71 #include <linux/dma-mapping.h>
72 #include <linux/delay.h>
73 #include <linux/slab.h>
76 #include <asm/firmware.h>
78 #include <scsi/scsi.h>
79 #include <scsi/scsi_cmnd.h>
80 #include <scsi/scsi_host.h>
81 #include <scsi/scsi_device.h>
82 #include <scsi/scsi_transport_srp.h>
85 /* The values below are somewhat arbitrary default values, but
86 * OS/400 will use 3 busses (disks, CDs, tapes, I think.)
87 * Note that there are 3 bits of channel value, 6 bits of id, and
90 static int max_id = 64;
91 static int max_channel = 3;
92 static int init_timeout = 300;
93 static int login_timeout = 60;
94 static int info_timeout = 30;
95 static int abort_timeout = 60;
96 static int reset_timeout = 60;
97 static int max_requests = IBMVSCSI_MAX_REQUESTS_DEFAULT;
98 static int max_events = IBMVSCSI_MAX_REQUESTS_DEFAULT + 2;
99 static int fast_fail = 1;
100 static int client_reserve = 1;
102 static struct scsi_transport_template *ibmvscsi_transport_template;
104 #define IBMVSCSI_VERSION "1.5.8"
106 static struct ibmvscsi_ops *ibmvscsi_ops;
108 MODULE_DESCRIPTION("IBM Virtual SCSI");
109 MODULE_AUTHOR("Dave Boutcher");
110 MODULE_LICENSE("GPL");
111 MODULE_VERSION(IBMVSCSI_VERSION);
113 module_param_named(max_id, max_id, int, S_IRUGO | S_IWUSR);
114 MODULE_PARM_DESC(max_id, "Largest ID value for each channel");
115 module_param_named(max_channel, max_channel, int, S_IRUGO | S_IWUSR);
116 MODULE_PARM_DESC(max_channel, "Largest channel value");
117 module_param_named(init_timeout, init_timeout, int, S_IRUGO | S_IWUSR);
118 MODULE_PARM_DESC(init_timeout, "Initialization timeout in seconds");
119 module_param_named(max_requests, max_requests, int, S_IRUGO);
120 MODULE_PARM_DESC(max_requests, "Maximum requests for this adapter");
121 module_param_named(fast_fail, fast_fail, int, S_IRUGO | S_IWUSR);
122 MODULE_PARM_DESC(fast_fail, "Enable fast fail. [Default=1]");
123 module_param_named(client_reserve, client_reserve, int, S_IRUGO );
124 MODULE_PARM_DESC(client_reserve, "Attempt client managed reserve/release");
126 /* ------------------------------------------------------------
127 * Routines for the event pool and event structs
130 * initialize_event_pool: - Allocates and initializes the event pool for a host
131 * @pool: event_pool to be initialized
132 * @size: Number of events in pool
133 * @hostdata: ibmvscsi_host_data who owns the event pool
135 * Returns zero on success.
137 static int initialize_event_pool(struct event_pool *pool,
138 int size, struct ibmvscsi_host_data *hostdata)
144 pool->events = kcalloc(pool->size, sizeof(*pool->events), GFP_KERNEL);
149 dma_alloc_coherent(hostdata->dev,
150 pool->size * sizeof(*pool->iu_storage),
152 if (!pool->iu_storage) {
157 for (i = 0; i < pool->size; ++i) {
158 struct srp_event_struct *evt = &pool->events[i];
159 memset(&evt->crq, 0x00, sizeof(evt->crq));
160 atomic_set(&evt->free, 1);
161 evt->crq.valid = 0x80;
162 evt->crq.IU_length = sizeof(*evt->xfer_iu);
163 evt->crq.IU_data_ptr = pool->iu_token +
164 sizeof(*evt->xfer_iu) * i;
165 evt->xfer_iu = pool->iu_storage + i;
166 evt->hostdata = hostdata;
167 evt->ext_list = NULL;
168 evt->ext_list_token = 0;
175 * release_event_pool: - Frees memory of an event pool of a host
176 * @pool: event_pool to be released
177 * @hostdata: ibmvscsi_host_data who owns the even pool
179 * Returns zero on success.
181 static void release_event_pool(struct event_pool *pool,
182 struct ibmvscsi_host_data *hostdata)
185 for (i = 0; i < pool->size; ++i) {
186 if (atomic_read(&pool->events[i].free) != 1)
188 if (pool->events[i].ext_list) {
189 dma_free_coherent(hostdata->dev,
190 SG_ALL * sizeof(struct srp_direct_buf),
191 pool->events[i].ext_list,
192 pool->events[i].ext_list_token);
196 dev_warn(hostdata->dev, "releasing event pool with %d "
197 "events still in use?\n", in_use);
199 dma_free_coherent(hostdata->dev,
200 pool->size * sizeof(*pool->iu_storage),
201 pool->iu_storage, pool->iu_token);
205 * valid_event_struct: - Determines if event is valid.
206 * @pool: event_pool that contains the event
207 * @evt: srp_event_struct to be checked for validity
209 * Returns zero if event is invalid, one otherwise.
211 static int valid_event_struct(struct event_pool *pool,
212 struct srp_event_struct *evt)
214 int index = evt - pool->events;
215 if (index < 0 || index >= pool->size) /* outside of bounds */
217 if (evt != pool->events + index) /* unaligned */
223 * ibmvscsi_free-event_struct: - Changes status of event to "free"
224 * @pool: event_pool that contains the event
225 * @evt: srp_event_struct to be modified
228 static void free_event_struct(struct event_pool *pool,
229 struct srp_event_struct *evt)
231 if (!valid_event_struct(pool, evt)) {
232 dev_err(evt->hostdata->dev, "Freeing invalid event_struct %p "
233 "(not in pool %p)\n", evt, pool->events);
236 if (atomic_inc_return(&evt->free) != 1) {
237 dev_err(evt->hostdata->dev, "Freeing event_struct %p "
238 "which is not in use!\n", evt);
244 * get_evt_struct: - Gets the next free event in pool
245 * @pool: event_pool that contains the events to be searched
247 * Returns the next event in "free" state, and NULL if none are free.
248 * Note that no synchronization is done here, we assume the host_lock
249 * will syncrhonze things.
251 static struct srp_event_struct *get_event_struct(struct event_pool *pool)
254 int poolsize = pool->size;
255 int offset = pool->next;
257 for (i = 0; i < poolsize; i++) {
258 offset = (offset + 1) % poolsize;
259 if (!atomic_dec_if_positive(&pool->events[offset].free)) {
261 return &pool->events[offset];
265 printk(KERN_ERR "ibmvscsi: found no event struct in pool!\n");
270 * init_event_struct: Initialize fields in an event struct that are always
273 * @done: Routine to call when the event is responded to
274 * @format: SRP or MAD format
275 * @timeout: timeout value set in the CRQ
277 static void init_event_struct(struct srp_event_struct *evt_struct,
278 void (*done) (struct srp_event_struct *),
282 evt_struct->cmnd = NULL;
283 evt_struct->cmnd_done = NULL;
284 evt_struct->sync_srp = NULL;
285 evt_struct->crq.format = format;
286 evt_struct->crq.timeout = timeout;
287 evt_struct->done = done;
290 /* ------------------------------------------------------------
291 * Routines for receiving SCSI responses from the hosting partition
295 * set_srp_direction: Set the fields in the srp related to data
296 * direction and number of buffers based on the direction in
297 * the scsi_cmnd and the number of buffers
299 static void set_srp_direction(struct scsi_cmnd *cmd,
300 struct srp_cmd *srp_cmd,
309 fmt = SRP_DATA_DESC_DIRECT;
311 fmt = SRP_DATA_DESC_INDIRECT;
312 numbuf = min(numbuf, MAX_INDIRECT_BUFS);
314 if (cmd->sc_data_direction == DMA_TO_DEVICE)
315 srp_cmd->data_out_desc_cnt = numbuf;
317 srp_cmd->data_in_desc_cnt = numbuf;
320 if (cmd->sc_data_direction == DMA_TO_DEVICE)
321 srp_cmd->buf_fmt = fmt << 4;
323 srp_cmd->buf_fmt = fmt;
326 static void unmap_sg_list(int num_entries,
328 struct srp_direct_buf *md)
332 for (i = 0; i < num_entries; ++i)
333 dma_unmap_single(dev, md[i].va, md[i].len, DMA_BIDIRECTIONAL);
337 * unmap_cmd_data: - Unmap data pointed in srp_cmd based on the format
338 * @cmd: srp_cmd whose additional_data member will be unmapped
339 * @dev: device for which the memory is mapped
342 static void unmap_cmd_data(struct srp_cmd *cmd,
343 struct srp_event_struct *evt_struct,
348 out_fmt = cmd->buf_fmt >> 4;
349 in_fmt = cmd->buf_fmt & ((1U << 4) - 1);
351 if (out_fmt == SRP_NO_DATA_DESC && in_fmt == SRP_NO_DATA_DESC)
353 else if (out_fmt == SRP_DATA_DESC_DIRECT ||
354 in_fmt == SRP_DATA_DESC_DIRECT) {
355 struct srp_direct_buf *data =
356 (struct srp_direct_buf *) cmd->add_data;
357 dma_unmap_single(dev, data->va, data->len, DMA_BIDIRECTIONAL);
359 struct srp_indirect_buf *indirect =
360 (struct srp_indirect_buf *) cmd->add_data;
361 int num_mapped = indirect->table_desc.len /
362 sizeof(struct srp_direct_buf);
364 if (num_mapped <= MAX_INDIRECT_BUFS) {
365 unmap_sg_list(num_mapped, dev, &indirect->desc_list[0]);
369 unmap_sg_list(num_mapped, dev, evt_struct->ext_list);
373 static int map_sg_list(struct scsi_cmnd *cmd, int nseg,
374 struct srp_direct_buf *md)
377 struct scatterlist *sg;
378 u64 total_length = 0;
380 scsi_for_each_sg(cmd, sg, nseg, i) {
381 struct srp_direct_buf *descr = md + i;
382 descr->va = sg_dma_address(sg);
383 descr->len = sg_dma_len(sg);
385 total_length += sg_dma_len(sg);
391 * map_sg_data: - Maps dma for a scatterlist and initializes decriptor fields
392 * @cmd: Scsi_Cmnd with the scatterlist
393 * @srp_cmd: srp_cmd that contains the memory descriptor
394 * @dev: device for which to map dma memory
396 * Called by map_data_for_srp_cmd() when building srp cmd from scsi cmd.
397 * Returns 1 on success.
399 static int map_sg_data(struct scsi_cmnd *cmd,
400 struct srp_event_struct *evt_struct,
401 struct srp_cmd *srp_cmd, struct device *dev)
405 u64 total_length = 0;
406 struct srp_direct_buf *data =
407 (struct srp_direct_buf *) srp_cmd->add_data;
408 struct srp_indirect_buf *indirect =
409 (struct srp_indirect_buf *) data;
411 sg_mapped = scsi_dma_map(cmd);
414 else if (sg_mapped < 0)
417 set_srp_direction(cmd, srp_cmd, sg_mapped);
419 /* special case; we can use a single direct descriptor */
420 if (sg_mapped == 1) {
421 map_sg_list(cmd, sg_mapped, data);
425 indirect->table_desc.va = 0;
426 indirect->table_desc.len = sg_mapped * sizeof(struct srp_direct_buf);
427 indirect->table_desc.key = 0;
429 if (sg_mapped <= MAX_INDIRECT_BUFS) {
430 total_length = map_sg_list(cmd, sg_mapped,
431 &indirect->desc_list[0]);
432 indirect->len = total_length;
436 /* get indirect table */
437 if (!evt_struct->ext_list) {
438 evt_struct->ext_list = (struct srp_direct_buf *)
439 dma_alloc_coherent(dev,
440 SG_ALL * sizeof(struct srp_direct_buf),
441 &evt_struct->ext_list_token, 0);
442 if (!evt_struct->ext_list) {
443 if (!firmware_has_feature(FW_FEATURE_CMO))
444 sdev_printk(KERN_ERR, cmd->device,
445 "Can't allocate memory "
446 "for indirect table\n");
452 total_length = map_sg_list(cmd, sg_mapped, evt_struct->ext_list);
454 indirect->len = total_length;
455 indirect->table_desc.va = evt_struct->ext_list_token;
456 indirect->table_desc.len = sg_mapped * sizeof(indirect->desc_list[0]);
457 memcpy(indirect->desc_list, evt_struct->ext_list,
458 MAX_INDIRECT_BUFS * sizeof(struct srp_direct_buf));
463 * map_data_for_srp_cmd: - Calls functions to map data for srp cmds
464 * @cmd: struct scsi_cmnd with the memory to be mapped
465 * @srp_cmd: srp_cmd that contains the memory descriptor
466 * @dev: dma device for which to map dma memory
468 * Called by scsi_cmd_to_srp_cmd() when converting scsi cmds to srp cmds
469 * Returns 1 on success.
471 static int map_data_for_srp_cmd(struct scsi_cmnd *cmd,
472 struct srp_event_struct *evt_struct,
473 struct srp_cmd *srp_cmd, struct device *dev)
475 switch (cmd->sc_data_direction) {
476 case DMA_FROM_DEVICE:
481 case DMA_BIDIRECTIONAL:
482 sdev_printk(KERN_ERR, cmd->device,
483 "Can't map DMA_BIDIRECTIONAL to read/write\n");
486 sdev_printk(KERN_ERR, cmd->device,
487 "Unknown data direction 0x%02x; can't map!\n",
488 cmd->sc_data_direction);
492 return map_sg_data(cmd, evt_struct, srp_cmd, dev);
496 * purge_requests: Our virtual adapter just shut down. purge any sent requests
497 * @hostdata: the adapter
499 static void purge_requests(struct ibmvscsi_host_data *hostdata, int error_code)
501 struct srp_event_struct *tmp_evt, *pos;
504 spin_lock_irqsave(hostdata->host->host_lock, flags);
505 list_for_each_entry_safe(tmp_evt, pos, &hostdata->sent, list) {
506 list_del(&tmp_evt->list);
507 del_timer(&tmp_evt->timer);
509 tmp_evt->cmnd->result = (error_code << 16);
510 unmap_cmd_data(&tmp_evt->iu.srp.cmd,
512 tmp_evt->hostdata->dev);
513 if (tmp_evt->cmnd_done)
514 tmp_evt->cmnd_done(tmp_evt->cmnd);
515 } else if (tmp_evt->done)
516 tmp_evt->done(tmp_evt);
517 free_event_struct(&tmp_evt->hostdata->pool, tmp_evt);
519 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
523 * ibmvscsi_reset_host - Reset the connection to the server
524 * @hostdata: struct ibmvscsi_host_data to reset
526 static void ibmvscsi_reset_host(struct ibmvscsi_host_data *hostdata)
528 scsi_block_requests(hostdata->host);
529 atomic_set(&hostdata->request_limit, 0);
531 purge_requests(hostdata, DID_ERROR);
532 if ((ibmvscsi_ops->reset_crq_queue(&hostdata->queue, hostdata)) ||
533 (ibmvscsi_ops->send_crq(hostdata, 0xC001000000000000LL, 0)) ||
534 (vio_enable_interrupts(to_vio_dev(hostdata->dev)))) {
535 atomic_set(&hostdata->request_limit, -1);
536 dev_err(hostdata->dev, "error after reset\n");
539 scsi_unblock_requests(hostdata->host);
543 * ibmvscsi_timeout - Internal command timeout handler
544 * @evt_struct: struct srp_event_struct that timed out
546 * Called when an internally generated command times out
548 static void ibmvscsi_timeout(struct srp_event_struct *evt_struct)
550 struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
552 dev_err(hostdata->dev, "Command timed out (%x). Resetting connection\n",
553 evt_struct->iu.srp.cmd.opcode);
555 ibmvscsi_reset_host(hostdata);
559 /* ------------------------------------------------------------
560 * Routines for sending and receiving SRPs
563 * ibmvscsi_send_srp_event: - Transforms event to u64 array and calls send_crq()
564 * @evt_struct: evt_struct to be sent
565 * @hostdata: ibmvscsi_host_data of host
566 * @timeout: timeout in seconds - 0 means do not time command
568 * Returns the value returned from ibmvscsi_send_crq(). (Zero for success)
569 * Note that this routine assumes that host_lock is held for synchronization
571 static int ibmvscsi_send_srp_event(struct srp_event_struct *evt_struct,
572 struct ibmvscsi_host_data *hostdata,
573 unsigned long timeout)
575 u64 *crq_as_u64 = (u64 *) &evt_struct->crq;
576 int request_status = 0;
579 /* If we have exhausted our request limit, just fail this request,
580 * unless it is for a reset or abort.
581 * Note that there are rare cases involving driver generated requests
582 * (such as task management requests) that the mid layer may think we
583 * can handle more requests (can_queue) when we actually can't
585 if (evt_struct->crq.format == VIOSRP_SRP_FORMAT) {
587 atomic_dec_if_positive(&hostdata->request_limit);
588 /* If request limit was -1 when we started, it is now even
591 if (request_status < -1)
593 /* Otherwise, we may have run out of requests. */
594 /* If request limit was 0 when we started the adapter is in the
595 * process of performing a login with the server adapter, or
596 * we may have run out of requests.
598 else if (request_status == -1 &&
599 evt_struct->iu.srp.login_req.opcode != SRP_LOGIN_REQ)
601 /* Abort and reset calls should make it through.
602 * Nothing except abort and reset should use the last two
603 * slots unless we had two or less to begin with.
605 else if (request_status < 2 &&
606 evt_struct->iu.srp.cmd.opcode != SRP_TSK_MGMT) {
607 /* In the case that we have less than two requests
608 * available, check the server limit as a combination
609 * of the request limit and the number of requests
610 * in-flight (the size of the send list). If the
611 * server limit is greater than 2, return busy so
612 * that the last two are reserved for reset and abort.
614 int server_limit = request_status;
615 struct srp_event_struct *tmp_evt;
617 list_for_each_entry(tmp_evt, &hostdata->sent, list) {
621 if (server_limit > 2)
626 /* Copy the IU into the transfer area */
627 *evt_struct->xfer_iu = evt_struct->iu;
628 evt_struct->xfer_iu->srp.rsp.tag = (u64)evt_struct;
630 /* Add this to the sent list. We need to do this
631 * before we actually send
632 * in case it comes back REALLY fast
634 list_add_tail(&evt_struct->list, &hostdata->sent);
636 init_timer(&evt_struct->timer);
638 evt_struct->timer.data = (unsigned long) evt_struct;
639 evt_struct->timer.expires = jiffies + (timeout * HZ);
640 evt_struct->timer.function = (void (*)(unsigned long))ibmvscsi_timeout;
641 add_timer(&evt_struct->timer);
645 ibmvscsi_ops->send_crq(hostdata, crq_as_u64[0], crq_as_u64[1])) != 0) {
646 list_del(&evt_struct->list);
647 del_timer(&evt_struct->timer);
649 /* If send_crq returns H_CLOSED, return SCSI_MLQUEUE_HOST_BUSY.
650 * Firmware will send a CRQ with a transport event (0xFF) to
651 * tell this client what has happened to the transport. This
652 * will be handled in ibmvscsi_handle_crq()
654 if (rc == H_CLOSED) {
655 dev_warn(hostdata->dev, "send warning. "
656 "Receive queue closed, will retry.\n");
659 dev_err(hostdata->dev, "send error %d\n", rc);
660 atomic_inc(&hostdata->request_limit);
667 unmap_cmd_data(&evt_struct->iu.srp.cmd, evt_struct, hostdata->dev);
669 free_event_struct(&hostdata->pool, evt_struct);
670 if (request_status != -1)
671 atomic_inc(&hostdata->request_limit);
672 return SCSI_MLQUEUE_HOST_BUSY;
675 unmap_cmd_data(&evt_struct->iu.srp.cmd, evt_struct, hostdata->dev);
677 if (evt_struct->cmnd != NULL) {
678 evt_struct->cmnd->result = DID_ERROR << 16;
679 evt_struct->cmnd_done(evt_struct->cmnd);
680 } else if (evt_struct->done)
681 evt_struct->done(evt_struct);
683 free_event_struct(&hostdata->pool, evt_struct);
688 * handle_cmd_rsp: - Handle responses from commands
689 * @evt_struct: srp_event_struct to be handled
691 * Used as a callback by when sending scsi cmds.
692 * Gets called by ibmvscsi_handle_crq()
694 static void handle_cmd_rsp(struct srp_event_struct *evt_struct)
696 struct srp_rsp *rsp = &evt_struct->xfer_iu->srp.rsp;
697 struct scsi_cmnd *cmnd = evt_struct->cmnd;
699 if (unlikely(rsp->opcode != SRP_RSP)) {
700 if (printk_ratelimit())
701 dev_warn(evt_struct->hostdata->dev,
702 "bad SRP RSP type %d\n", rsp->opcode);
706 cmnd->result |= rsp->status;
707 if (((cmnd->result >> 1) & 0x1f) == CHECK_CONDITION)
708 memcpy(cmnd->sense_buffer,
710 rsp->sense_data_len);
711 unmap_cmd_data(&evt_struct->iu.srp.cmd,
713 evt_struct->hostdata->dev);
715 if (rsp->flags & SRP_RSP_FLAG_DOOVER)
716 scsi_set_resid(cmnd, rsp->data_out_res_cnt);
717 else if (rsp->flags & SRP_RSP_FLAG_DIOVER)
718 scsi_set_resid(cmnd, rsp->data_in_res_cnt);
721 if (evt_struct->cmnd_done)
722 evt_struct->cmnd_done(cmnd);
726 * lun_from_dev: - Returns the lun of the scsi device
727 * @dev: struct scsi_device
730 static inline u16 lun_from_dev(struct scsi_device *dev)
732 return (0x2 << 14) | (dev->id << 8) | (dev->channel << 5) | dev->lun;
736 * ibmvscsi_queue: - The queuecommand function of the scsi template
737 * @cmd: struct scsi_cmnd to be executed
738 * @done: Callback function to be called when cmd is completed
740 static int ibmvscsi_queuecommand(struct scsi_cmnd *cmnd,
741 void (*done) (struct scsi_cmnd *))
743 struct srp_cmd *srp_cmd;
744 struct srp_event_struct *evt_struct;
745 struct srp_indirect_buf *indirect;
746 struct ibmvscsi_host_data *hostdata = shost_priv(cmnd->device->host);
747 u16 lun = lun_from_dev(cmnd->device);
750 cmnd->result = (DID_OK << 16);
751 evt_struct = get_event_struct(&hostdata->pool);
753 return SCSI_MLQUEUE_HOST_BUSY;
755 /* Set up the actual SRP IU */
756 srp_cmd = &evt_struct->iu.srp.cmd;
757 memset(srp_cmd, 0x00, SRP_MAX_IU_LEN);
758 srp_cmd->opcode = SRP_CMD;
759 memcpy(srp_cmd->cdb, cmnd->cmnd, sizeof(srp_cmd->cdb));
760 srp_cmd->lun = ((u64) lun) << 48;
762 if (!map_data_for_srp_cmd(cmnd, evt_struct, srp_cmd, hostdata->dev)) {
763 if (!firmware_has_feature(FW_FEATURE_CMO))
764 sdev_printk(KERN_ERR, cmnd->device,
765 "couldn't convert cmd to srp_cmd\n");
766 free_event_struct(&hostdata->pool, evt_struct);
767 return SCSI_MLQUEUE_HOST_BUSY;
770 init_event_struct(evt_struct,
773 cmnd->request->timeout/HZ);
775 evt_struct->cmnd = cmnd;
776 evt_struct->cmnd_done = done;
778 /* Fix up dma address of the buffer itself */
779 indirect = (struct srp_indirect_buf *) srp_cmd->add_data;
780 out_fmt = srp_cmd->buf_fmt >> 4;
781 in_fmt = srp_cmd->buf_fmt & ((1U << 4) - 1);
782 if ((in_fmt == SRP_DATA_DESC_INDIRECT ||
783 out_fmt == SRP_DATA_DESC_INDIRECT) &&
784 indirect->table_desc.va == 0) {
785 indirect->table_desc.va = evt_struct->crq.IU_data_ptr +
786 offsetof(struct srp_cmd, add_data) +
787 offsetof(struct srp_indirect_buf, desc_list);
790 return ibmvscsi_send_srp_event(evt_struct, hostdata, 0);
793 /* ------------------------------------------------------------
794 * Routines for driver initialization
798 * map_persist_bufs: - Pre-map persistent data for adapter logins
799 * @hostdata: ibmvscsi_host_data of host
801 * Map the capabilities and adapter info DMA buffers to avoid runtime failures.
802 * Return 1 on error, 0 on success.
804 static int map_persist_bufs(struct ibmvscsi_host_data *hostdata)
807 hostdata->caps_addr = dma_map_single(hostdata->dev, &hostdata->caps,
808 sizeof(hostdata->caps), DMA_BIDIRECTIONAL);
810 if (dma_mapping_error(hostdata->dev, hostdata->caps_addr)) {
811 dev_err(hostdata->dev, "Unable to map capabilities buffer!\n");
815 hostdata->adapter_info_addr = dma_map_single(hostdata->dev,
816 &hostdata->madapter_info,
817 sizeof(hostdata->madapter_info),
819 if (dma_mapping_error(hostdata->dev, hostdata->adapter_info_addr)) {
820 dev_err(hostdata->dev, "Unable to map adapter info buffer!\n");
821 dma_unmap_single(hostdata->dev, hostdata->caps_addr,
822 sizeof(hostdata->caps), DMA_BIDIRECTIONAL);
830 * unmap_persist_bufs: - Unmap persistent data needed for adapter logins
831 * @hostdata: ibmvscsi_host_data of host
833 * Unmap the capabilities and adapter info DMA buffers
835 static void unmap_persist_bufs(struct ibmvscsi_host_data *hostdata)
837 dma_unmap_single(hostdata->dev, hostdata->caps_addr,
838 sizeof(hostdata->caps), DMA_BIDIRECTIONAL);
840 dma_unmap_single(hostdata->dev, hostdata->adapter_info_addr,
841 sizeof(hostdata->madapter_info), DMA_BIDIRECTIONAL);
845 * login_rsp: - Handle response to SRP login request
846 * @evt_struct: srp_event_struct with the response
848 * Used as a "done" callback by when sending srp_login. Gets called
849 * by ibmvscsi_handle_crq()
851 static void login_rsp(struct srp_event_struct *evt_struct)
853 struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
854 switch (evt_struct->xfer_iu->srp.login_rsp.opcode) {
855 case SRP_LOGIN_RSP: /* it worked! */
857 case SRP_LOGIN_REJ: /* refused! */
858 dev_info(hostdata->dev, "SRP_LOGIN_REJ reason %u\n",
859 evt_struct->xfer_iu->srp.login_rej.reason);
861 atomic_set(&hostdata->request_limit, -1);
864 dev_err(hostdata->dev, "Invalid login response typecode 0x%02x!\n",
865 evt_struct->xfer_iu->srp.login_rsp.opcode);
867 atomic_set(&hostdata->request_limit, -1);
871 dev_info(hostdata->dev, "SRP_LOGIN succeeded\n");
872 hostdata->client_migrated = 0;
874 /* Now we know what the real request-limit is.
875 * This value is set rather than added to request_limit because
876 * request_limit could have been set to -1 by this client.
878 atomic_set(&hostdata->request_limit,
879 evt_struct->xfer_iu->srp.login_rsp.req_lim_delta);
881 /* If we had any pending I/Os, kick them */
882 scsi_unblock_requests(hostdata->host);
886 * send_srp_login: - Sends the srp login
887 * @hostdata: ibmvscsi_host_data of host
889 * Returns zero if successful.
891 static int send_srp_login(struct ibmvscsi_host_data *hostdata)
895 struct srp_login_req *login;
896 struct srp_event_struct *evt_struct = get_event_struct(&hostdata->pool);
899 init_event_struct(evt_struct, login_rsp,
900 VIOSRP_SRP_FORMAT, login_timeout);
902 login = &evt_struct->iu.srp.login_req;
903 memset(login, 0, sizeof(*login));
904 login->opcode = SRP_LOGIN_REQ;
905 login->req_it_iu_len = sizeof(union srp_iu);
906 login->req_buf_fmt = SRP_BUF_FORMAT_DIRECT | SRP_BUF_FORMAT_INDIRECT;
908 spin_lock_irqsave(hostdata->host->host_lock, flags);
909 /* Start out with a request limit of 0, since this is negotiated in
910 * the login request we are just sending and login requests always
911 * get sent by the driver regardless of request_limit.
913 atomic_set(&hostdata->request_limit, 0);
915 rc = ibmvscsi_send_srp_event(evt_struct, hostdata, login_timeout * 2);
916 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
917 dev_info(hostdata->dev, "sent SRP login\n");
922 * capabilities_rsp: - Handle response to MAD adapter capabilities request
923 * @evt_struct: srp_event_struct with the response
925 * Used as a "done" callback by when sending adapter_info.
927 static void capabilities_rsp(struct srp_event_struct *evt_struct)
929 struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
931 if (evt_struct->xfer_iu->mad.capabilities.common.status) {
932 dev_err(hostdata->dev, "error 0x%X getting capabilities info\n",
933 evt_struct->xfer_iu->mad.capabilities.common.status);
935 if (hostdata->caps.migration.common.server_support != SERVER_SUPPORTS_CAP)
936 dev_info(hostdata->dev, "Partition migration not supported\n");
938 if (client_reserve) {
939 if (hostdata->caps.reserve.common.server_support ==
941 dev_info(hostdata->dev, "Client reserve enabled\n");
943 dev_info(hostdata->dev, "Client reserve not supported\n");
947 send_srp_login(hostdata);
951 * send_mad_capabilities: - Sends the mad capabilities request
952 * and stores the result so it can be retrieved with
953 * @hostdata: ibmvscsi_host_data of host
955 static void send_mad_capabilities(struct ibmvscsi_host_data *hostdata)
957 struct viosrp_capabilities *req;
958 struct srp_event_struct *evt_struct;
960 struct device_node *of_node = hostdata->dev->archdata.of_node;
961 const char *location;
963 evt_struct = get_event_struct(&hostdata->pool);
966 init_event_struct(evt_struct, capabilities_rsp,
967 VIOSRP_MAD_FORMAT, info_timeout);
969 req = &evt_struct->iu.mad.capabilities;
970 memset(req, 0, sizeof(*req));
972 hostdata->caps.flags = CAP_LIST_SUPPORTED;
973 if (hostdata->client_migrated)
974 hostdata->caps.flags |= CLIENT_MIGRATED;
976 strncpy(hostdata->caps.name, dev_name(&hostdata->host->shost_gendev),
977 sizeof(hostdata->caps.name));
978 hostdata->caps.name[sizeof(hostdata->caps.name) - 1] = '\0';
980 location = of_get_property(of_node, "ibm,loc-code", NULL);
981 location = location ? location : dev_name(hostdata->dev);
982 strncpy(hostdata->caps.loc, location, sizeof(hostdata->caps.loc));
983 hostdata->caps.loc[sizeof(hostdata->caps.loc) - 1] = '\0';
985 req->common.type = VIOSRP_CAPABILITIES_TYPE;
986 req->buffer = hostdata->caps_addr;
988 hostdata->caps.migration.common.cap_type = MIGRATION_CAPABILITIES;
989 hostdata->caps.migration.common.length = sizeof(hostdata->caps.migration);
990 hostdata->caps.migration.common.server_support = SERVER_SUPPORTS_CAP;
991 hostdata->caps.migration.ecl = 1;
993 if (client_reserve) {
994 hostdata->caps.reserve.common.cap_type = RESERVATION_CAPABILITIES;
995 hostdata->caps.reserve.common.length = sizeof(hostdata->caps.reserve);
996 hostdata->caps.reserve.common.server_support = SERVER_SUPPORTS_CAP;
997 hostdata->caps.reserve.type = CLIENT_RESERVE_SCSI_2;
998 req->common.length = sizeof(hostdata->caps);
1000 req->common.length = sizeof(hostdata->caps) - sizeof(hostdata->caps.reserve);
1002 spin_lock_irqsave(hostdata->host->host_lock, flags);
1003 if (ibmvscsi_send_srp_event(evt_struct, hostdata, info_timeout * 2))
1004 dev_err(hostdata->dev, "couldn't send CAPABILITIES_REQ!\n");
1005 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1009 * fast_fail_rsp: - Handle response to MAD enable fast fail
1010 * @evt_struct: srp_event_struct with the response
1012 * Used as a "done" callback by when sending enable fast fail. Gets called
1013 * by ibmvscsi_handle_crq()
1015 static void fast_fail_rsp(struct srp_event_struct *evt_struct)
1017 struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
1018 u8 status = evt_struct->xfer_iu->mad.fast_fail.common.status;
1020 if (status == VIOSRP_MAD_NOT_SUPPORTED)
1021 dev_err(hostdata->dev, "fast_fail not supported in server\n");
1022 else if (status == VIOSRP_MAD_FAILED)
1023 dev_err(hostdata->dev, "fast_fail request failed\n");
1024 else if (status != VIOSRP_MAD_SUCCESS)
1025 dev_err(hostdata->dev, "error 0x%X enabling fast_fail\n", status);
1027 send_mad_capabilities(hostdata);
1031 * init_host - Start host initialization
1032 * @hostdata: ibmvscsi_host_data of host
1034 * Returns zero if successful.
1036 static int enable_fast_fail(struct ibmvscsi_host_data *hostdata)
1039 unsigned long flags;
1040 struct viosrp_fast_fail *fast_fail_mad;
1041 struct srp_event_struct *evt_struct;
1044 send_mad_capabilities(hostdata);
1048 evt_struct = get_event_struct(&hostdata->pool);
1049 BUG_ON(!evt_struct);
1051 init_event_struct(evt_struct, fast_fail_rsp, VIOSRP_MAD_FORMAT, info_timeout);
1053 fast_fail_mad = &evt_struct->iu.mad.fast_fail;
1054 memset(fast_fail_mad, 0, sizeof(*fast_fail_mad));
1055 fast_fail_mad->common.type = VIOSRP_ENABLE_FAST_FAIL;
1056 fast_fail_mad->common.length = sizeof(*fast_fail_mad);
1058 spin_lock_irqsave(hostdata->host->host_lock, flags);
1059 rc = ibmvscsi_send_srp_event(evt_struct, hostdata, info_timeout * 2);
1060 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1065 * adapter_info_rsp: - Handle response to MAD adapter info request
1066 * @evt_struct: srp_event_struct with the response
1068 * Used as a "done" callback by when sending adapter_info. Gets called
1069 * by ibmvscsi_handle_crq()
1071 static void adapter_info_rsp(struct srp_event_struct *evt_struct)
1073 struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
1075 if (evt_struct->xfer_iu->mad.adapter_info.common.status) {
1076 dev_err(hostdata->dev, "error %d getting adapter info\n",
1077 evt_struct->xfer_iu->mad.adapter_info.common.status);
1079 dev_info(hostdata->dev, "host srp version: %s, "
1080 "host partition %s (%d), OS %d, max io %u\n",
1081 hostdata->madapter_info.srp_version,
1082 hostdata->madapter_info.partition_name,
1083 hostdata->madapter_info.partition_number,
1084 hostdata->madapter_info.os_type,
1085 hostdata->madapter_info.port_max_txu[0]);
1087 if (hostdata->madapter_info.port_max_txu[0])
1088 hostdata->host->max_sectors =
1089 hostdata->madapter_info.port_max_txu[0] >> 9;
1091 if (hostdata->madapter_info.os_type == 3 &&
1092 strcmp(hostdata->madapter_info.srp_version, "1.6a") <= 0) {
1093 dev_err(hostdata->dev, "host (Ver. %s) doesn't support large transfers\n",
1094 hostdata->madapter_info.srp_version);
1095 dev_err(hostdata->dev, "limiting scatterlists to %d\n",
1097 hostdata->host->sg_tablesize = MAX_INDIRECT_BUFS;
1100 if (hostdata->madapter_info.os_type == 3) {
1101 enable_fast_fail(hostdata);
1106 send_srp_login(hostdata);
1110 * send_mad_adapter_info: - Sends the mad adapter info request
1111 * and stores the result so it can be retrieved with
1112 * sysfs. We COULD consider causing a failure if the
1113 * returned SRP version doesn't match ours.
1114 * @hostdata: ibmvscsi_host_data of host
1116 * Returns zero if successful.
1118 static void send_mad_adapter_info(struct ibmvscsi_host_data *hostdata)
1120 struct viosrp_adapter_info *req;
1121 struct srp_event_struct *evt_struct;
1122 unsigned long flags;
1124 evt_struct = get_event_struct(&hostdata->pool);
1125 BUG_ON(!evt_struct);
1127 init_event_struct(evt_struct,
1132 req = &evt_struct->iu.mad.adapter_info;
1133 memset(req, 0x00, sizeof(*req));
1135 req->common.type = VIOSRP_ADAPTER_INFO_TYPE;
1136 req->common.length = sizeof(hostdata->madapter_info);
1137 req->buffer = hostdata->adapter_info_addr;
1139 spin_lock_irqsave(hostdata->host->host_lock, flags);
1140 if (ibmvscsi_send_srp_event(evt_struct, hostdata, info_timeout * 2))
1141 dev_err(hostdata->dev, "couldn't send ADAPTER_INFO_REQ!\n");
1142 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1146 * init_adapter: Start virtual adapter initialization sequence
1149 static void init_adapter(struct ibmvscsi_host_data *hostdata)
1151 send_mad_adapter_info(hostdata);
1155 * sync_completion: Signal that a synchronous command has completed
1156 * Note that after returning from this call, the evt_struct is freed.
1157 * the caller waiting on this completion shouldn't touch the evt_struct
1160 static void sync_completion(struct srp_event_struct *evt_struct)
1162 /* copy the response back */
1163 if (evt_struct->sync_srp)
1164 *evt_struct->sync_srp = *evt_struct->xfer_iu;
1166 complete(&evt_struct->comp);
1170 * ibmvscsi_abort: Abort a command...from scsi host template
1171 * send this over to the server and wait synchronously for the response
1173 static int ibmvscsi_eh_abort_handler(struct scsi_cmnd *cmd)
1175 struct ibmvscsi_host_data *hostdata = shost_priv(cmd->device->host);
1176 struct srp_tsk_mgmt *tsk_mgmt;
1177 struct srp_event_struct *evt;
1178 struct srp_event_struct *tmp_evt, *found_evt;
1179 union viosrp_iu srp_rsp;
1181 unsigned long flags;
1182 u16 lun = lun_from_dev(cmd->device);
1183 unsigned long wait_switch = 0;
1185 /* First, find this command in our sent list so we can figure
1186 * out the correct tag
1188 spin_lock_irqsave(hostdata->host->host_lock, flags);
1189 wait_switch = jiffies + (init_timeout * HZ);
1192 list_for_each_entry(tmp_evt, &hostdata->sent, list) {
1193 if (tmp_evt->cmnd == cmd) {
1194 found_evt = tmp_evt;
1200 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1204 evt = get_event_struct(&hostdata->pool);
1206 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1207 sdev_printk(KERN_ERR, cmd->device,
1208 "failed to allocate abort event\n");
1212 init_event_struct(evt,
1217 tsk_mgmt = &evt->iu.srp.tsk_mgmt;
1219 /* Set up an abort SRP command */
1220 memset(tsk_mgmt, 0x00, sizeof(*tsk_mgmt));
1221 tsk_mgmt->opcode = SRP_TSK_MGMT;
1222 tsk_mgmt->lun = ((u64) lun) << 48;
1223 tsk_mgmt->tsk_mgmt_func = SRP_TSK_ABORT_TASK;
1224 tsk_mgmt->task_tag = (u64) found_evt;
1226 evt->sync_srp = &srp_rsp;
1228 init_completion(&evt->comp);
1229 rsp_rc = ibmvscsi_send_srp_event(evt, hostdata, abort_timeout * 2);
1231 if (rsp_rc != SCSI_MLQUEUE_HOST_BUSY)
1234 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1236 spin_lock_irqsave(hostdata->host->host_lock, flags);
1237 } while (time_before(jiffies, wait_switch));
1239 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1242 sdev_printk(KERN_ERR, cmd->device,
1243 "failed to send abort() event. rc=%d\n", rsp_rc);
1247 sdev_printk(KERN_INFO, cmd->device,
1248 "aborting command. lun 0x%llx, tag 0x%llx\n",
1249 (((u64) lun) << 48), (u64) found_evt);
1251 wait_for_completion(&evt->comp);
1253 /* make sure we got a good response */
1254 if (unlikely(srp_rsp.srp.rsp.opcode != SRP_RSP)) {
1255 if (printk_ratelimit())
1256 sdev_printk(KERN_WARNING, cmd->device, "abort bad SRP RSP type %d\n",
1257 srp_rsp.srp.rsp.opcode);
1261 if (srp_rsp.srp.rsp.flags & SRP_RSP_FLAG_RSPVALID)
1262 rsp_rc = *((int *)srp_rsp.srp.rsp.data);
1264 rsp_rc = srp_rsp.srp.rsp.status;
1267 if (printk_ratelimit())
1268 sdev_printk(KERN_WARNING, cmd->device,
1269 "abort code %d for task tag 0x%llx\n",
1270 rsp_rc, tsk_mgmt->task_tag);
1274 /* Because we dropped the spinlock above, it's possible
1275 * The event is no longer in our list. Make sure it didn't
1276 * complete while we were aborting
1278 spin_lock_irqsave(hostdata->host->host_lock, flags);
1280 list_for_each_entry(tmp_evt, &hostdata->sent, list) {
1281 if (tmp_evt->cmnd == cmd) {
1282 found_evt = tmp_evt;
1287 if (found_evt == NULL) {
1288 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1289 sdev_printk(KERN_INFO, cmd->device, "aborted task tag 0x%llx completed\n",
1290 tsk_mgmt->task_tag);
1294 sdev_printk(KERN_INFO, cmd->device, "successfully aborted task tag 0x%llx\n",
1295 tsk_mgmt->task_tag);
1297 cmd->result = (DID_ABORT << 16);
1298 list_del(&found_evt->list);
1299 unmap_cmd_data(&found_evt->iu.srp.cmd, found_evt,
1300 found_evt->hostdata->dev);
1301 free_event_struct(&found_evt->hostdata->pool, found_evt);
1302 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1303 atomic_inc(&hostdata->request_limit);
1308 * ibmvscsi_eh_device_reset_handler: Reset a single LUN...from scsi host
1309 * template send this over to the server and wait synchronously for the
1312 static int ibmvscsi_eh_device_reset_handler(struct scsi_cmnd *cmd)
1314 struct ibmvscsi_host_data *hostdata = shost_priv(cmd->device->host);
1315 struct srp_tsk_mgmt *tsk_mgmt;
1316 struct srp_event_struct *evt;
1317 struct srp_event_struct *tmp_evt, *pos;
1318 union viosrp_iu srp_rsp;
1320 unsigned long flags;
1321 u16 lun = lun_from_dev(cmd->device);
1322 unsigned long wait_switch = 0;
1324 spin_lock_irqsave(hostdata->host->host_lock, flags);
1325 wait_switch = jiffies + (init_timeout * HZ);
1327 evt = get_event_struct(&hostdata->pool);
1329 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1330 sdev_printk(KERN_ERR, cmd->device,
1331 "failed to allocate reset event\n");
1335 init_event_struct(evt,
1340 tsk_mgmt = &evt->iu.srp.tsk_mgmt;
1342 /* Set up a lun reset SRP command */
1343 memset(tsk_mgmt, 0x00, sizeof(*tsk_mgmt));
1344 tsk_mgmt->opcode = SRP_TSK_MGMT;
1345 tsk_mgmt->lun = ((u64) lun) << 48;
1346 tsk_mgmt->tsk_mgmt_func = SRP_TSK_LUN_RESET;
1348 evt->sync_srp = &srp_rsp;
1350 init_completion(&evt->comp);
1351 rsp_rc = ibmvscsi_send_srp_event(evt, hostdata, reset_timeout * 2);
1353 if (rsp_rc != SCSI_MLQUEUE_HOST_BUSY)
1356 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1358 spin_lock_irqsave(hostdata->host->host_lock, flags);
1359 } while (time_before(jiffies, wait_switch));
1361 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1364 sdev_printk(KERN_ERR, cmd->device,
1365 "failed to send reset event. rc=%d\n", rsp_rc);
1369 sdev_printk(KERN_INFO, cmd->device, "resetting device. lun 0x%llx\n",
1370 (((u64) lun) << 48));
1372 wait_for_completion(&evt->comp);
1374 /* make sure we got a good response */
1375 if (unlikely(srp_rsp.srp.rsp.opcode != SRP_RSP)) {
1376 if (printk_ratelimit())
1377 sdev_printk(KERN_WARNING, cmd->device, "reset bad SRP RSP type %d\n",
1378 srp_rsp.srp.rsp.opcode);
1382 if (srp_rsp.srp.rsp.flags & SRP_RSP_FLAG_RSPVALID)
1383 rsp_rc = *((int *)srp_rsp.srp.rsp.data);
1385 rsp_rc = srp_rsp.srp.rsp.status;
1388 if (printk_ratelimit())
1389 sdev_printk(KERN_WARNING, cmd->device,
1390 "reset code %d for task tag 0x%llx\n",
1391 rsp_rc, tsk_mgmt->task_tag);
1395 /* We need to find all commands for this LUN that have not yet been
1396 * responded to, and fail them with DID_RESET
1398 spin_lock_irqsave(hostdata->host->host_lock, flags);
1399 list_for_each_entry_safe(tmp_evt, pos, &hostdata->sent, list) {
1400 if ((tmp_evt->cmnd) && (tmp_evt->cmnd->device == cmd->device)) {
1402 tmp_evt->cmnd->result = (DID_RESET << 16);
1403 list_del(&tmp_evt->list);
1404 unmap_cmd_data(&tmp_evt->iu.srp.cmd, tmp_evt,
1405 tmp_evt->hostdata->dev);
1406 free_event_struct(&tmp_evt->hostdata->pool,
1408 atomic_inc(&hostdata->request_limit);
1409 if (tmp_evt->cmnd_done)
1410 tmp_evt->cmnd_done(tmp_evt->cmnd);
1411 else if (tmp_evt->done)
1412 tmp_evt->done(tmp_evt);
1415 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1420 * ibmvscsi_eh_host_reset_handler - Reset the connection to the server
1421 * @cmd: struct scsi_cmnd having problems
1423 static int ibmvscsi_eh_host_reset_handler(struct scsi_cmnd *cmd)
1425 unsigned long wait_switch = 0;
1426 struct ibmvscsi_host_data *hostdata = shost_priv(cmd->device->host);
1428 dev_err(hostdata->dev, "Resetting connection due to error recovery\n");
1430 ibmvscsi_reset_host(hostdata);
1432 for (wait_switch = jiffies + (init_timeout * HZ);
1433 time_before(jiffies, wait_switch) &&
1434 atomic_read(&hostdata->request_limit) < 2;) {
1439 if (atomic_read(&hostdata->request_limit) <= 0)
1446 * ibmvscsi_handle_crq: - Handles and frees received events in the CRQ
1447 * @crq: Command/Response queue
1448 * @hostdata: ibmvscsi_host_data of host
1451 void ibmvscsi_handle_crq(struct viosrp_crq *crq,
1452 struct ibmvscsi_host_data *hostdata)
1455 unsigned long flags;
1456 struct srp_event_struct *evt_struct =
1457 (struct srp_event_struct *)crq->IU_data_ptr;
1458 switch (crq->valid) {
1459 case 0xC0: /* initialization */
1460 switch (crq->format) {
1461 case 0x01: /* Initialization message */
1462 dev_info(hostdata->dev, "partner initialized\n");
1463 /* Send back a response */
1464 if ((rc = ibmvscsi_ops->send_crq(hostdata,
1465 0xC002000000000000LL, 0)) == 0) {
1467 init_adapter(hostdata);
1469 dev_err(hostdata->dev, "Unable to send init rsp. rc=%ld\n", rc);
1473 case 0x02: /* Initialization response */
1474 dev_info(hostdata->dev, "partner initialization complete\n");
1477 init_adapter(hostdata);
1480 dev_err(hostdata->dev, "unknown crq message type: %d\n", crq->format);
1483 case 0xFF: /* Hypervisor telling us the connection is closed */
1484 scsi_block_requests(hostdata->host);
1485 atomic_set(&hostdata->request_limit, 0);
1486 if (crq->format == 0x06) {
1487 /* We need to re-setup the interpartition connection */
1488 dev_info(hostdata->dev, "Re-enabling adapter!\n");
1489 hostdata->client_migrated = 1;
1490 purge_requests(hostdata, DID_REQUEUE);
1491 if ((ibmvscsi_ops->reenable_crq_queue(&hostdata->queue,
1493 (ibmvscsi_ops->send_crq(hostdata,
1494 0xC001000000000000LL, 0))) {
1495 atomic_set(&hostdata->request_limit,
1497 dev_err(hostdata->dev, "error after enable\n");
1500 dev_err(hostdata->dev, "Virtual adapter failed rc %d!\n",
1503 purge_requests(hostdata, DID_ERROR);
1504 if ((ibmvscsi_ops->reset_crq_queue(&hostdata->queue,
1506 (ibmvscsi_ops->send_crq(hostdata,
1507 0xC001000000000000LL, 0))) {
1508 atomic_set(&hostdata->request_limit,
1510 dev_err(hostdata->dev, "error after reset\n");
1513 scsi_unblock_requests(hostdata->host);
1515 case 0x80: /* real payload */
1518 dev_err(hostdata->dev, "got an invalid message type 0x%02x\n",
1523 /* The only kind of payload CRQs we should get are responses to
1524 * things we send. Make sure this response is to something we
1527 if (!valid_event_struct(&hostdata->pool, evt_struct)) {
1528 dev_err(hostdata->dev, "returned correlation_token 0x%p is invalid!\n",
1529 (void *)crq->IU_data_ptr);
1533 if (atomic_read(&evt_struct->free)) {
1534 dev_err(hostdata->dev, "received duplicate correlation_token 0x%p!\n",
1535 (void *)crq->IU_data_ptr);
1539 if (crq->format == VIOSRP_SRP_FORMAT)
1540 atomic_add(evt_struct->xfer_iu->srp.rsp.req_lim_delta,
1541 &hostdata->request_limit);
1543 del_timer(&evt_struct->timer);
1545 if ((crq->status != VIOSRP_OK && crq->status != VIOSRP_OK2) && evt_struct->cmnd)
1546 evt_struct->cmnd->result = DID_ERROR << 16;
1547 if (evt_struct->done)
1548 evt_struct->done(evt_struct);
1550 dev_err(hostdata->dev, "returned done() is NULL; not running it!\n");
1553 * Lock the host_lock before messing with these structures, since we
1554 * are running in a task context
1556 spin_lock_irqsave(evt_struct->hostdata->host->host_lock, flags);
1557 list_del(&evt_struct->list);
1558 free_event_struct(&evt_struct->hostdata->pool, evt_struct);
1559 spin_unlock_irqrestore(evt_struct->hostdata->host->host_lock, flags);
1563 * ibmvscsi_get_host_config: Send the command to the server to get host
1564 * configuration data. The data is opaque to us.
1566 static int ibmvscsi_do_host_config(struct ibmvscsi_host_data *hostdata,
1567 unsigned char *buffer, int length)
1569 struct viosrp_host_config *host_config;
1570 struct srp_event_struct *evt_struct;
1571 unsigned long flags;
1575 evt_struct = get_event_struct(&hostdata->pool);
1577 dev_err(hostdata->dev, "couldn't allocate event for HOST_CONFIG!\n");
1581 init_event_struct(evt_struct,
1586 host_config = &evt_struct->iu.mad.host_config;
1588 /* Set up a lun reset SRP command */
1589 memset(host_config, 0x00, sizeof(*host_config));
1590 host_config->common.type = VIOSRP_HOST_CONFIG_TYPE;
1591 host_config->common.length = length;
1592 host_config->buffer = addr = dma_map_single(hostdata->dev, buffer,
1596 if (dma_mapping_error(hostdata->dev, host_config->buffer)) {
1597 if (!firmware_has_feature(FW_FEATURE_CMO))
1598 dev_err(hostdata->dev,
1599 "dma_mapping error getting host config\n");
1600 free_event_struct(&hostdata->pool, evt_struct);
1604 init_completion(&evt_struct->comp);
1605 spin_lock_irqsave(hostdata->host->host_lock, flags);
1606 rc = ibmvscsi_send_srp_event(evt_struct, hostdata, info_timeout * 2);
1607 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1609 wait_for_completion(&evt_struct->comp);
1610 dma_unmap_single(hostdata->dev, addr, length, DMA_BIDIRECTIONAL);
1616 * ibmvscsi_slave_configure: Set the "allow_restart" flag for each disk.
1617 * @sdev: struct scsi_device device to configure
1619 * Enable allow_restart for a device if it is a disk. Adjust the
1620 * queue_depth here also as is required by the documentation for
1621 * struct scsi_host_template.
1623 static int ibmvscsi_slave_configure(struct scsi_device *sdev)
1625 struct Scsi_Host *shost = sdev->host;
1626 unsigned long lock_flags = 0;
1628 spin_lock_irqsave(shost->host_lock, lock_flags);
1629 if (sdev->type == TYPE_DISK) {
1630 sdev->allow_restart = 1;
1631 blk_queue_rq_timeout(sdev->request_queue, 120 * HZ);
1633 scsi_adjust_queue_depth(sdev, 0, shost->cmd_per_lun);
1634 spin_unlock_irqrestore(shost->host_lock, lock_flags);
1639 * ibmvscsi_change_queue_depth - Change the device's queue depth
1640 * @sdev: scsi device struct
1641 * @qdepth: depth to set
1642 * @reason: calling context
1647 static int ibmvscsi_change_queue_depth(struct scsi_device *sdev, int qdepth,
1650 if (reason != SCSI_QDEPTH_DEFAULT)
1653 if (qdepth > IBMVSCSI_MAX_CMDS_PER_LUN)
1654 qdepth = IBMVSCSI_MAX_CMDS_PER_LUN;
1656 scsi_adjust_queue_depth(sdev, 0, qdepth);
1657 return sdev->queue_depth;
1660 /* ------------------------------------------------------------
1663 static ssize_t show_host_vhost_loc(struct device *dev,
1664 struct device_attribute *attr, char *buf)
1666 struct Scsi_Host *shost = class_to_shost(dev);
1667 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1670 len = snprintf(buf, sizeof(hostdata->caps.loc), "%s\n",
1671 hostdata->caps.loc);
1675 static struct device_attribute ibmvscsi_host_vhost_loc = {
1677 .name = "vhost_loc",
1680 .show = show_host_vhost_loc,
1683 static ssize_t show_host_vhost_name(struct device *dev,
1684 struct device_attribute *attr, char *buf)
1686 struct Scsi_Host *shost = class_to_shost(dev);
1687 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1690 len = snprintf(buf, sizeof(hostdata->caps.name), "%s\n",
1691 hostdata->caps.name);
1695 static struct device_attribute ibmvscsi_host_vhost_name = {
1697 .name = "vhost_name",
1700 .show = show_host_vhost_name,
1703 static ssize_t show_host_srp_version(struct device *dev,
1704 struct device_attribute *attr, char *buf)
1706 struct Scsi_Host *shost = class_to_shost(dev);
1707 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1710 len = snprintf(buf, PAGE_SIZE, "%s\n",
1711 hostdata->madapter_info.srp_version);
1715 static struct device_attribute ibmvscsi_host_srp_version = {
1717 .name = "srp_version",
1720 .show = show_host_srp_version,
1723 static ssize_t show_host_partition_name(struct device *dev,
1724 struct device_attribute *attr,
1727 struct Scsi_Host *shost = class_to_shost(dev);
1728 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1731 len = snprintf(buf, PAGE_SIZE, "%s\n",
1732 hostdata->madapter_info.partition_name);
1736 static struct device_attribute ibmvscsi_host_partition_name = {
1738 .name = "partition_name",
1741 .show = show_host_partition_name,
1744 static ssize_t show_host_partition_number(struct device *dev,
1745 struct device_attribute *attr,
1748 struct Scsi_Host *shost = class_to_shost(dev);
1749 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1752 len = snprintf(buf, PAGE_SIZE, "%d\n",
1753 hostdata->madapter_info.partition_number);
1757 static struct device_attribute ibmvscsi_host_partition_number = {
1759 .name = "partition_number",
1762 .show = show_host_partition_number,
1765 static ssize_t show_host_mad_version(struct device *dev,
1766 struct device_attribute *attr, char *buf)
1768 struct Scsi_Host *shost = class_to_shost(dev);
1769 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1772 len = snprintf(buf, PAGE_SIZE, "%d\n",
1773 hostdata->madapter_info.mad_version);
1777 static struct device_attribute ibmvscsi_host_mad_version = {
1779 .name = "mad_version",
1782 .show = show_host_mad_version,
1785 static ssize_t show_host_os_type(struct device *dev,
1786 struct device_attribute *attr, char *buf)
1788 struct Scsi_Host *shost = class_to_shost(dev);
1789 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1792 len = snprintf(buf, PAGE_SIZE, "%d\n", hostdata->madapter_info.os_type);
1796 static struct device_attribute ibmvscsi_host_os_type = {
1801 .show = show_host_os_type,
1804 static ssize_t show_host_config(struct device *dev,
1805 struct device_attribute *attr, char *buf)
1807 struct Scsi_Host *shost = class_to_shost(dev);
1808 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1810 /* returns null-terminated host config data */
1811 if (ibmvscsi_do_host_config(hostdata, buf, PAGE_SIZE) == 0)
1817 static struct device_attribute ibmvscsi_host_config = {
1822 .show = show_host_config,
1825 static struct device_attribute *ibmvscsi_attrs[] = {
1826 &ibmvscsi_host_vhost_loc,
1827 &ibmvscsi_host_vhost_name,
1828 &ibmvscsi_host_srp_version,
1829 &ibmvscsi_host_partition_name,
1830 &ibmvscsi_host_partition_number,
1831 &ibmvscsi_host_mad_version,
1832 &ibmvscsi_host_os_type,
1833 &ibmvscsi_host_config,
1837 /* ------------------------------------------------------------
1838 * SCSI driver registration
1840 static struct scsi_host_template driver_template = {
1841 .module = THIS_MODULE,
1842 .name = "IBM POWER Virtual SCSI Adapter " IBMVSCSI_VERSION,
1843 .proc_name = "ibmvscsi",
1844 .queuecommand = ibmvscsi_queuecommand,
1845 .eh_abort_handler = ibmvscsi_eh_abort_handler,
1846 .eh_device_reset_handler = ibmvscsi_eh_device_reset_handler,
1847 .eh_host_reset_handler = ibmvscsi_eh_host_reset_handler,
1848 .slave_configure = ibmvscsi_slave_configure,
1849 .change_queue_depth = ibmvscsi_change_queue_depth,
1850 .cmd_per_lun = IBMVSCSI_CMDS_PER_LUN_DEFAULT,
1851 .can_queue = IBMVSCSI_MAX_REQUESTS_DEFAULT,
1853 .sg_tablesize = SG_ALL,
1854 .use_clustering = ENABLE_CLUSTERING,
1855 .shost_attrs = ibmvscsi_attrs,
1859 * ibmvscsi_get_desired_dma - Calculate IO memory desired by the driver
1861 * @vdev: struct vio_dev for the device whose desired IO mem is to be returned
1864 * Number of bytes of IO data the driver will need to perform well.
1866 static unsigned long ibmvscsi_get_desired_dma(struct vio_dev *vdev)
1868 /* iu_storage data allocated in initialize_event_pool */
1869 unsigned long desired_io = max_events * sizeof(union viosrp_iu);
1871 /* add io space for sg data */
1872 desired_io += (IBMVSCSI_MAX_SECTORS_DEFAULT * 512 *
1873 IBMVSCSI_CMDS_PER_LUN_DEFAULT);
1879 * Called by bus code for each adapter
1881 static int ibmvscsi_probe(struct vio_dev *vdev, const struct vio_device_id *id)
1883 struct ibmvscsi_host_data *hostdata;
1884 struct Scsi_Host *host;
1885 struct device *dev = &vdev->dev;
1886 struct srp_rport_identifiers ids;
1887 struct srp_rport *rport;
1888 unsigned long wait_switch = 0;
1891 dev_set_drvdata(&vdev->dev, NULL);
1893 host = scsi_host_alloc(&driver_template, sizeof(*hostdata));
1895 dev_err(&vdev->dev, "couldn't allocate host data\n");
1896 goto scsi_host_alloc_failed;
1899 host->transportt = ibmvscsi_transport_template;
1900 hostdata = shost_priv(host);
1901 memset(hostdata, 0x00, sizeof(*hostdata));
1902 INIT_LIST_HEAD(&hostdata->sent);
1903 hostdata->host = host;
1904 hostdata->dev = dev;
1905 atomic_set(&hostdata->request_limit, -1);
1906 hostdata->host->max_sectors = IBMVSCSI_MAX_SECTORS_DEFAULT;
1908 if (map_persist_bufs(hostdata)) {
1909 dev_err(&vdev->dev, "couldn't map persistent buffers\n");
1910 goto persist_bufs_failed;
1913 rc = ibmvscsi_ops->init_crq_queue(&hostdata->queue, hostdata, max_events);
1914 if (rc != 0 && rc != H_RESOURCE) {
1915 dev_err(&vdev->dev, "couldn't initialize crq. rc=%d\n", rc);
1916 goto init_crq_failed;
1918 if (initialize_event_pool(&hostdata->pool, max_events, hostdata) != 0) {
1919 dev_err(&vdev->dev, "couldn't initialize event pool\n");
1920 goto init_pool_failed;
1924 host->max_id = max_id;
1925 host->max_channel = max_channel;
1926 host->max_cmd_len = 16;
1928 if (scsi_add_host(hostdata->host, hostdata->dev))
1929 goto add_host_failed;
1931 /* we don't have a proper target_port_id so let's use the fake one */
1932 memcpy(ids.port_id, hostdata->madapter_info.partition_name,
1933 sizeof(ids.port_id));
1934 ids.roles = SRP_RPORT_ROLE_TARGET;
1935 rport = srp_rport_add(host, &ids);
1937 goto add_srp_port_failed;
1939 /* Try to send an initialization message. Note that this is allowed
1940 * to fail if the other end is not acive. In that case we don't
1943 if (ibmvscsi_ops->send_crq(hostdata, 0xC001000000000000LL, 0) == 0
1944 || rc == H_RESOURCE) {
1946 * Wait around max init_timeout secs for the adapter to finish
1947 * initializing. When we are done initializing, we will have a
1948 * valid request_limit. We don't want Linux scanning before
1951 for (wait_switch = jiffies + (init_timeout * HZ);
1952 time_before(jiffies, wait_switch) &&
1953 atomic_read(&hostdata->request_limit) < 2;) {
1958 /* if we now have a valid request_limit, initiate a scan */
1959 if (atomic_read(&hostdata->request_limit) > 0)
1960 scsi_scan_host(host);
1963 dev_set_drvdata(&vdev->dev, hostdata);
1966 add_srp_port_failed:
1967 scsi_remove_host(hostdata->host);
1969 release_event_pool(&hostdata->pool, hostdata);
1971 ibmvscsi_ops->release_crq_queue(&hostdata->queue, hostdata, max_events);
1973 unmap_persist_bufs(hostdata);
1974 persist_bufs_failed:
1975 scsi_host_put(host);
1976 scsi_host_alloc_failed:
1980 static int ibmvscsi_remove(struct vio_dev *vdev)
1982 struct ibmvscsi_host_data *hostdata = dev_get_drvdata(&vdev->dev);
1983 unmap_persist_bufs(hostdata);
1984 release_event_pool(&hostdata->pool, hostdata);
1985 ibmvscsi_ops->release_crq_queue(&hostdata->queue, hostdata,
1988 srp_remove_host(hostdata->host);
1989 scsi_remove_host(hostdata->host);
1990 scsi_host_put(hostdata->host);
1996 * ibmvscsi_resume: Resume from suspend
1997 * @dev: device struct
1999 * We may have lost an interrupt across suspend/resume, so kick the
2002 static int ibmvscsi_resume(struct device *dev)
2004 struct ibmvscsi_host_data *hostdata = dev_get_drvdata(dev);
2005 return ibmvscsi_ops->resume(hostdata);
2009 * ibmvscsi_device_table: Used by vio.c to match devices in the device tree we
2012 static struct vio_device_id ibmvscsi_device_table[] __devinitdata = {
2013 {"vscsi", "IBM,v-scsi"},
2016 MODULE_DEVICE_TABLE(vio, ibmvscsi_device_table);
2018 static struct dev_pm_ops ibmvscsi_pm_ops = {
2019 .resume = ibmvscsi_resume
2022 static struct vio_driver ibmvscsi_driver = {
2023 .id_table = ibmvscsi_device_table,
2024 .probe = ibmvscsi_probe,
2025 .remove = ibmvscsi_remove,
2026 .get_desired_dma = ibmvscsi_get_desired_dma,
2029 .owner = THIS_MODULE,
2030 .pm = &ibmvscsi_pm_ops,
2034 static struct srp_function_template ibmvscsi_transport_functions = {
2037 int __init ibmvscsi_module_init(void)
2041 /* Ensure we have two requests to do error recovery */
2042 driver_template.can_queue = max_requests;
2043 max_events = max_requests + 2;
2045 if (firmware_has_feature(FW_FEATURE_ISERIES))
2046 ibmvscsi_ops = &iseriesvscsi_ops;
2047 else if (firmware_has_feature(FW_FEATURE_VIO))
2048 ibmvscsi_ops = &rpavscsi_ops;
2052 ibmvscsi_transport_template =
2053 srp_attach_transport(&ibmvscsi_transport_functions);
2054 if (!ibmvscsi_transport_template)
2057 ret = vio_register_driver(&ibmvscsi_driver);
2059 srp_release_transport(ibmvscsi_transport_template);
2063 void __exit ibmvscsi_module_exit(void)
2065 vio_unregister_driver(&ibmvscsi_driver);
2066 srp_release_transport(ibmvscsi_transport_template);
2069 module_init(ibmvscsi_module_init);
2070 module_exit(ibmvscsi_module_exit);