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-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
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>
74 #include <asm/firmware.h>
76 #include <asm/firmware.h>
77 #include <scsi/scsi.h>
78 #include <scsi/scsi_cmnd.h>
79 #include <scsi/scsi_host.h>
80 #include <scsi/scsi_device.h>
81 #include <scsi/scsi_transport_srp.h>
84 /* The values below are somewhat arbitrary default values, but
85 * OS/400 will use 3 busses (disks, CDs, tapes, I think.)
86 * Note that there are 3 bits of channel value, 6 bits of id, and
89 static int max_id = 64;
90 static int max_channel = 3;
91 static int init_timeout = 300;
92 static int login_timeout = 60;
93 static int info_timeout = 30;
94 static int abort_timeout = 60;
95 static int reset_timeout = 60;
96 static int max_requests = IBMVSCSI_MAX_REQUESTS_DEFAULT;
97 static int max_events = IBMVSCSI_MAX_REQUESTS_DEFAULT + 2;
98 static int fast_fail = 1;
99 static int client_reserve = 1;
101 static struct scsi_transport_template *ibmvscsi_transport_template;
103 #define IBMVSCSI_VERSION "1.5.8"
105 static struct ibmvscsi_ops *ibmvscsi_ops;
107 MODULE_DESCRIPTION("IBM Virtual SCSI");
108 MODULE_AUTHOR("Dave Boutcher");
109 MODULE_LICENSE("GPL");
110 MODULE_VERSION(IBMVSCSI_VERSION);
112 module_param_named(max_id, max_id, int, S_IRUGO | S_IWUSR);
113 MODULE_PARM_DESC(max_id, "Largest ID value for each channel");
114 module_param_named(max_channel, max_channel, int, S_IRUGO | S_IWUSR);
115 MODULE_PARM_DESC(max_channel, "Largest channel value");
116 module_param_named(init_timeout, init_timeout, int, S_IRUGO | S_IWUSR);
117 MODULE_PARM_DESC(init_timeout, "Initialization timeout in seconds");
118 module_param_named(max_requests, max_requests, int, S_IRUGO);
119 MODULE_PARM_DESC(max_requests, "Maximum requests for this adapter");
120 module_param_named(fast_fail, fast_fail, int, S_IRUGO | S_IWUSR);
121 MODULE_PARM_DESC(fast_fail, "Enable fast fail. [Default=1]");
122 module_param_named(client_reserve, client_reserve, int, S_IRUGO );
123 MODULE_PARM_DESC(client_reserve, "Attempt client managed reserve/release");
125 /* ------------------------------------------------------------
126 * Routines for the event pool and event structs
129 * initialize_event_pool: - Allocates and initializes the event pool for a host
130 * @pool: event_pool to be initialized
131 * @size: Number of events in pool
132 * @hostdata: ibmvscsi_host_data who owns the event pool
134 * Returns zero on success.
136 static int initialize_event_pool(struct event_pool *pool,
137 int size, struct ibmvscsi_host_data *hostdata)
143 pool->events = kcalloc(pool->size, sizeof(*pool->events), GFP_KERNEL);
148 dma_alloc_coherent(hostdata->dev,
149 pool->size * sizeof(*pool->iu_storage),
151 if (!pool->iu_storage) {
156 for (i = 0; i < pool->size; ++i) {
157 struct srp_event_struct *evt = &pool->events[i];
158 memset(&evt->crq, 0x00, sizeof(evt->crq));
159 atomic_set(&evt->free, 1);
160 evt->crq.valid = 0x80;
161 evt->crq.IU_length = sizeof(*evt->xfer_iu);
162 evt->crq.IU_data_ptr = pool->iu_token +
163 sizeof(*evt->xfer_iu) * i;
164 evt->xfer_iu = pool->iu_storage + i;
165 evt->hostdata = hostdata;
166 evt->ext_list = NULL;
167 evt->ext_list_token = 0;
174 * release_event_pool: - Frees memory of an event pool of a host
175 * @pool: event_pool to be released
176 * @hostdata: ibmvscsi_host_data who owns the even pool
178 * Returns zero on success.
180 static void release_event_pool(struct event_pool *pool,
181 struct ibmvscsi_host_data *hostdata)
184 for (i = 0; i < pool->size; ++i) {
185 if (atomic_read(&pool->events[i].free) != 1)
187 if (pool->events[i].ext_list) {
188 dma_free_coherent(hostdata->dev,
189 SG_ALL * sizeof(struct srp_direct_buf),
190 pool->events[i].ext_list,
191 pool->events[i].ext_list_token);
195 dev_warn(hostdata->dev, "releasing event pool with %d "
196 "events still in use?\n", in_use);
198 dma_free_coherent(hostdata->dev,
199 pool->size * sizeof(*pool->iu_storage),
200 pool->iu_storage, pool->iu_token);
204 * valid_event_struct: - Determines if event is valid.
205 * @pool: event_pool that contains the event
206 * @evt: srp_event_struct to be checked for validity
208 * Returns zero if event is invalid, one otherwise.
210 static int valid_event_struct(struct event_pool *pool,
211 struct srp_event_struct *evt)
213 int index = evt - pool->events;
214 if (index < 0 || index >= pool->size) /* outside of bounds */
216 if (evt != pool->events + index) /* unaligned */
222 * ibmvscsi_free-event_struct: - Changes status of event to "free"
223 * @pool: event_pool that contains the event
224 * @evt: srp_event_struct to be modified
227 static void free_event_struct(struct event_pool *pool,
228 struct srp_event_struct *evt)
230 if (!valid_event_struct(pool, evt)) {
231 dev_err(evt->hostdata->dev, "Freeing invalid event_struct %p "
232 "(not in pool %p)\n", evt, pool->events);
235 if (atomic_inc_return(&evt->free) != 1) {
236 dev_err(evt->hostdata->dev, "Freeing event_struct %p "
237 "which is not in use!\n", evt);
243 * get_evt_struct: - Gets the next free event in pool
244 * @pool: event_pool that contains the events to be searched
246 * Returns the next event in "free" state, and NULL if none are free.
247 * Note that no synchronization is done here, we assume the host_lock
248 * will syncrhonze things.
250 static struct srp_event_struct *get_event_struct(struct event_pool *pool)
253 int poolsize = pool->size;
254 int offset = pool->next;
256 for (i = 0; i < poolsize; i++) {
257 offset = (offset + 1) % poolsize;
258 if (!atomic_dec_if_positive(&pool->events[offset].free)) {
260 return &pool->events[offset];
264 printk(KERN_ERR "ibmvscsi: found no event struct in pool!\n");
269 * init_event_struct: Initialize fields in an event struct that are always
272 * @done: Routine to call when the event is responded to
273 * @format: SRP or MAD format
274 * @timeout: timeout value set in the CRQ
276 static void init_event_struct(struct srp_event_struct *evt_struct,
277 void (*done) (struct srp_event_struct *),
281 evt_struct->cmnd = NULL;
282 evt_struct->cmnd_done = NULL;
283 evt_struct->sync_srp = NULL;
284 evt_struct->crq.format = format;
285 evt_struct->crq.timeout = timeout;
286 evt_struct->done = done;
289 /* ------------------------------------------------------------
290 * Routines for receiving SCSI responses from the hosting partition
294 * set_srp_direction: Set the fields in the srp related to data
295 * direction and number of buffers based on the direction in
296 * the scsi_cmnd and the number of buffers
298 static void set_srp_direction(struct scsi_cmnd *cmd,
299 struct srp_cmd *srp_cmd,
308 fmt = SRP_DATA_DESC_DIRECT;
310 fmt = SRP_DATA_DESC_INDIRECT;
311 numbuf = min(numbuf, MAX_INDIRECT_BUFS);
313 if (cmd->sc_data_direction == DMA_TO_DEVICE)
314 srp_cmd->data_out_desc_cnt = numbuf;
316 srp_cmd->data_in_desc_cnt = numbuf;
319 if (cmd->sc_data_direction == DMA_TO_DEVICE)
320 srp_cmd->buf_fmt = fmt << 4;
322 srp_cmd->buf_fmt = fmt;
325 static void unmap_sg_list(int num_entries,
327 struct srp_direct_buf *md)
331 for (i = 0; i < num_entries; ++i)
332 dma_unmap_single(dev, md[i].va, md[i].len, DMA_BIDIRECTIONAL);
336 * unmap_cmd_data: - Unmap data pointed in srp_cmd based on the format
337 * @cmd: srp_cmd whose additional_data member will be unmapped
338 * @dev: device for which the memory is mapped
341 static void unmap_cmd_data(struct srp_cmd *cmd,
342 struct srp_event_struct *evt_struct,
347 out_fmt = cmd->buf_fmt >> 4;
348 in_fmt = cmd->buf_fmt & ((1U << 4) - 1);
350 if (out_fmt == SRP_NO_DATA_DESC && in_fmt == SRP_NO_DATA_DESC)
352 else if (out_fmt == SRP_DATA_DESC_DIRECT ||
353 in_fmt == SRP_DATA_DESC_DIRECT) {
354 struct srp_direct_buf *data =
355 (struct srp_direct_buf *) cmd->add_data;
356 dma_unmap_single(dev, data->va, data->len, DMA_BIDIRECTIONAL);
358 struct srp_indirect_buf *indirect =
359 (struct srp_indirect_buf *) cmd->add_data;
360 int num_mapped = indirect->table_desc.len /
361 sizeof(struct srp_direct_buf);
363 if (num_mapped <= MAX_INDIRECT_BUFS) {
364 unmap_sg_list(num_mapped, dev, &indirect->desc_list[0]);
368 unmap_sg_list(num_mapped, dev, evt_struct->ext_list);
372 static int map_sg_list(struct scsi_cmnd *cmd, int nseg,
373 struct srp_direct_buf *md)
376 struct scatterlist *sg;
377 u64 total_length = 0;
379 scsi_for_each_sg(cmd, sg, nseg, i) {
380 struct srp_direct_buf *descr = md + i;
381 descr->va = sg_dma_address(sg);
382 descr->len = sg_dma_len(sg);
384 total_length += sg_dma_len(sg);
390 * map_sg_data: - Maps dma for a scatterlist and initializes decriptor fields
391 * @cmd: Scsi_Cmnd with the scatterlist
392 * @srp_cmd: srp_cmd that contains the memory descriptor
393 * @dev: device for which to map dma memory
395 * Called by map_data_for_srp_cmd() when building srp cmd from scsi cmd.
396 * Returns 1 on success.
398 static int map_sg_data(struct scsi_cmnd *cmd,
399 struct srp_event_struct *evt_struct,
400 struct srp_cmd *srp_cmd, struct device *dev)
404 u64 total_length = 0;
405 struct srp_direct_buf *data =
406 (struct srp_direct_buf *) srp_cmd->add_data;
407 struct srp_indirect_buf *indirect =
408 (struct srp_indirect_buf *) data;
410 sg_mapped = scsi_dma_map(cmd);
413 else if (sg_mapped < 0)
416 set_srp_direction(cmd, srp_cmd, sg_mapped);
418 /* special case; we can use a single direct descriptor */
419 if (sg_mapped == 1) {
420 map_sg_list(cmd, sg_mapped, data);
424 indirect->table_desc.va = 0;
425 indirect->table_desc.len = sg_mapped * sizeof(struct srp_direct_buf);
426 indirect->table_desc.key = 0;
428 if (sg_mapped <= MAX_INDIRECT_BUFS) {
429 total_length = map_sg_list(cmd, sg_mapped,
430 &indirect->desc_list[0]);
431 indirect->len = total_length;
435 /* get indirect table */
436 if (!evt_struct->ext_list) {
437 evt_struct->ext_list = (struct srp_direct_buf *)
438 dma_alloc_coherent(dev,
439 SG_ALL * sizeof(struct srp_direct_buf),
440 &evt_struct->ext_list_token, 0);
441 if (!evt_struct->ext_list) {
442 if (!firmware_has_feature(FW_FEATURE_CMO))
443 sdev_printk(KERN_ERR, cmd->device,
444 "Can't allocate memory "
445 "for indirect table\n");
451 total_length = map_sg_list(cmd, sg_mapped, evt_struct->ext_list);
453 indirect->len = total_length;
454 indirect->table_desc.va = evt_struct->ext_list_token;
455 indirect->table_desc.len = sg_mapped * sizeof(indirect->desc_list[0]);
456 memcpy(indirect->desc_list, evt_struct->ext_list,
457 MAX_INDIRECT_BUFS * sizeof(struct srp_direct_buf));
462 * map_data_for_srp_cmd: - Calls functions to map data for srp cmds
463 * @cmd: struct scsi_cmnd with the memory to be mapped
464 * @srp_cmd: srp_cmd that contains the memory descriptor
465 * @dev: dma device for which to map dma memory
467 * Called by scsi_cmd_to_srp_cmd() when converting scsi cmds to srp cmds
468 * Returns 1 on success.
470 static int map_data_for_srp_cmd(struct scsi_cmnd *cmd,
471 struct srp_event_struct *evt_struct,
472 struct srp_cmd *srp_cmd, struct device *dev)
474 switch (cmd->sc_data_direction) {
475 case DMA_FROM_DEVICE:
480 case DMA_BIDIRECTIONAL:
481 sdev_printk(KERN_ERR, cmd->device,
482 "Can't map DMA_BIDIRECTIONAL to read/write\n");
485 sdev_printk(KERN_ERR, cmd->device,
486 "Unknown data direction 0x%02x; can't map!\n",
487 cmd->sc_data_direction);
491 return map_sg_data(cmd, evt_struct, srp_cmd, dev);
495 * purge_requests: Our virtual adapter just shut down. purge any sent requests
496 * @hostdata: the adapter
498 static void purge_requests(struct ibmvscsi_host_data *hostdata, int error_code)
500 struct srp_event_struct *tmp_evt, *pos;
503 spin_lock_irqsave(hostdata->host->host_lock, flags);
504 list_for_each_entry_safe(tmp_evt, pos, &hostdata->sent, list) {
505 list_del(&tmp_evt->list);
506 del_timer(&tmp_evt->timer);
508 tmp_evt->cmnd->result = (error_code << 16);
509 unmap_cmd_data(&tmp_evt->iu.srp.cmd,
511 tmp_evt->hostdata->dev);
512 if (tmp_evt->cmnd_done)
513 tmp_evt->cmnd_done(tmp_evt->cmnd);
514 } else if (tmp_evt->done)
515 tmp_evt->done(tmp_evt);
516 free_event_struct(&tmp_evt->hostdata->pool, tmp_evt);
518 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
522 * ibmvscsi_reset_host - Reset the connection to the server
523 * @hostdata: struct ibmvscsi_host_data to reset
525 static void ibmvscsi_reset_host(struct ibmvscsi_host_data *hostdata)
527 scsi_block_requests(hostdata->host);
528 atomic_set(&hostdata->request_limit, 0);
530 purge_requests(hostdata, DID_ERROR);
531 if ((ibmvscsi_ops->reset_crq_queue(&hostdata->queue, hostdata)) ||
532 (ibmvscsi_ops->send_crq(hostdata, 0xC001000000000000LL, 0)) ||
533 (vio_enable_interrupts(to_vio_dev(hostdata->dev)))) {
534 atomic_set(&hostdata->request_limit, -1);
535 dev_err(hostdata->dev, "error after reset\n");
538 scsi_unblock_requests(hostdata->host);
542 * ibmvscsi_timeout - Internal command timeout handler
543 * @evt_struct: struct srp_event_struct that timed out
545 * Called when an internally generated command times out
547 static void ibmvscsi_timeout(struct srp_event_struct *evt_struct)
549 struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
551 dev_err(hostdata->dev, "Command timed out (%x). Resetting connection\n",
552 evt_struct->iu.srp.cmd.opcode);
554 ibmvscsi_reset_host(hostdata);
558 /* ------------------------------------------------------------
559 * Routines for sending and receiving SRPs
562 * ibmvscsi_send_srp_event: - Transforms event to u64 array and calls send_crq()
563 * @evt_struct: evt_struct to be sent
564 * @hostdata: ibmvscsi_host_data of host
565 * @timeout: timeout in seconds - 0 means do not time command
567 * Returns the value returned from ibmvscsi_send_crq(). (Zero for success)
568 * Note that this routine assumes that host_lock is held for synchronization
570 static int ibmvscsi_send_srp_event(struct srp_event_struct *evt_struct,
571 struct ibmvscsi_host_data *hostdata,
572 unsigned long timeout)
574 u64 *crq_as_u64 = (u64 *) &evt_struct->crq;
575 int request_status = 0;
578 /* If we have exhausted our request limit, just fail this request,
579 * unless it is for a reset or abort.
580 * Note that there are rare cases involving driver generated requests
581 * (such as task management requests) that the mid layer may think we
582 * can handle more requests (can_queue) when we actually can't
584 if (evt_struct->crq.format == VIOSRP_SRP_FORMAT) {
586 atomic_dec_if_positive(&hostdata->request_limit);
587 /* If request limit was -1 when we started, it is now even
590 if (request_status < -1)
592 /* Otherwise, we may have run out of requests. */
593 /* If request limit was 0 when we started the adapter is in the
594 * process of performing a login with the server adapter, or
595 * we may have run out of requests.
597 else if (request_status == -1 &&
598 evt_struct->iu.srp.login_req.opcode != SRP_LOGIN_REQ)
600 /* Abort and reset calls should make it through.
601 * Nothing except abort and reset should use the last two
602 * slots unless we had two or less to begin with.
604 else if (request_status < 2 &&
605 evt_struct->iu.srp.cmd.opcode != SRP_TSK_MGMT) {
606 /* In the case that we have less than two requests
607 * available, check the server limit as a combination
608 * of the request limit and the number of requests
609 * in-flight (the size of the send list). If the
610 * server limit is greater than 2, return busy so
611 * that the last two are reserved for reset and abort.
613 int server_limit = request_status;
614 struct srp_event_struct *tmp_evt;
616 list_for_each_entry(tmp_evt, &hostdata->sent, list) {
620 if (server_limit > 2)
625 /* Copy the IU into the transfer area */
626 *evt_struct->xfer_iu = evt_struct->iu;
627 evt_struct->xfer_iu->srp.rsp.tag = (u64)evt_struct;
629 /* Add this to the sent list. We need to do this
630 * before we actually send
631 * in case it comes back REALLY fast
633 list_add_tail(&evt_struct->list, &hostdata->sent);
635 init_timer(&evt_struct->timer);
637 evt_struct->timer.data = (unsigned long) evt_struct;
638 evt_struct->timer.expires = jiffies + (timeout * HZ);
639 evt_struct->timer.function = (void (*)(unsigned long))ibmvscsi_timeout;
640 add_timer(&evt_struct->timer);
644 ibmvscsi_ops->send_crq(hostdata, crq_as_u64[0], crq_as_u64[1])) != 0) {
645 list_del(&evt_struct->list);
646 del_timer(&evt_struct->timer);
648 /* If send_crq returns H_CLOSED, return SCSI_MLQUEUE_HOST_BUSY.
649 * Firmware will send a CRQ with a transport event (0xFF) to
650 * tell this client what has happened to the transport. This
651 * will be handled in ibmvscsi_handle_crq()
653 if (rc == H_CLOSED) {
654 dev_warn(hostdata->dev, "send warning. "
655 "Receive queue closed, will retry.\n");
658 dev_err(hostdata->dev, "send error %d\n", rc);
659 atomic_inc(&hostdata->request_limit);
666 unmap_cmd_data(&evt_struct->iu.srp.cmd, evt_struct, hostdata->dev);
668 free_event_struct(&hostdata->pool, evt_struct);
669 if (request_status != -1)
670 atomic_inc(&hostdata->request_limit);
671 return SCSI_MLQUEUE_HOST_BUSY;
674 unmap_cmd_data(&evt_struct->iu.srp.cmd, evt_struct, hostdata->dev);
676 if (evt_struct->cmnd != NULL) {
677 evt_struct->cmnd->result = DID_ERROR << 16;
678 evt_struct->cmnd_done(evt_struct->cmnd);
679 } else if (evt_struct->done)
680 evt_struct->done(evt_struct);
682 free_event_struct(&hostdata->pool, evt_struct);
687 * handle_cmd_rsp: - Handle responses from commands
688 * @evt_struct: srp_event_struct to be handled
690 * Used as a callback by when sending scsi cmds.
691 * Gets called by ibmvscsi_handle_crq()
693 static void handle_cmd_rsp(struct srp_event_struct *evt_struct)
695 struct srp_rsp *rsp = &evt_struct->xfer_iu->srp.rsp;
696 struct scsi_cmnd *cmnd = evt_struct->cmnd;
698 if (unlikely(rsp->opcode != SRP_RSP)) {
699 if (printk_ratelimit())
700 dev_warn(evt_struct->hostdata->dev,
701 "bad SRP RSP type %d\n", rsp->opcode);
705 cmnd->result |= rsp->status;
706 if (((cmnd->result >> 1) & 0x1f) == CHECK_CONDITION)
707 memcpy(cmnd->sense_buffer,
709 rsp->sense_data_len);
710 unmap_cmd_data(&evt_struct->iu.srp.cmd,
712 evt_struct->hostdata->dev);
714 if (rsp->flags & SRP_RSP_FLAG_DOOVER)
715 scsi_set_resid(cmnd, rsp->data_out_res_cnt);
716 else if (rsp->flags & SRP_RSP_FLAG_DIOVER)
717 scsi_set_resid(cmnd, rsp->data_in_res_cnt);
720 if (evt_struct->cmnd_done)
721 evt_struct->cmnd_done(cmnd);
725 * lun_from_dev: - Returns the lun of the scsi device
726 * @dev: struct scsi_device
729 static inline u16 lun_from_dev(struct scsi_device *dev)
731 return (0x2 << 14) | (dev->id << 8) | (dev->channel << 5) | dev->lun;
735 * ibmvscsi_queue: - The queuecommand function of the scsi template
736 * @cmd: struct scsi_cmnd to be executed
737 * @done: Callback function to be called when cmd is completed
739 static int ibmvscsi_queuecommand(struct scsi_cmnd *cmnd,
740 void (*done) (struct scsi_cmnd *))
742 struct srp_cmd *srp_cmd;
743 struct srp_event_struct *evt_struct;
744 struct srp_indirect_buf *indirect;
745 struct ibmvscsi_host_data *hostdata = shost_priv(cmnd->device->host);
746 u16 lun = lun_from_dev(cmnd->device);
749 cmnd->result = (DID_OK << 16);
750 evt_struct = get_event_struct(&hostdata->pool);
752 return SCSI_MLQUEUE_HOST_BUSY;
754 /* Set up the actual SRP IU */
755 srp_cmd = &evt_struct->iu.srp.cmd;
756 memset(srp_cmd, 0x00, SRP_MAX_IU_LEN);
757 srp_cmd->opcode = SRP_CMD;
758 memcpy(srp_cmd->cdb, cmnd->cmnd, sizeof(srp_cmd->cdb));
759 srp_cmd->lun = ((u64) lun) << 48;
761 if (!map_data_for_srp_cmd(cmnd, evt_struct, srp_cmd, hostdata->dev)) {
762 if (!firmware_has_feature(FW_FEATURE_CMO))
763 sdev_printk(KERN_ERR, cmnd->device,
764 "couldn't convert cmd to srp_cmd\n");
765 free_event_struct(&hostdata->pool, evt_struct);
766 return SCSI_MLQUEUE_HOST_BUSY;
769 init_event_struct(evt_struct,
772 cmnd->request->timeout/HZ);
774 evt_struct->cmnd = cmnd;
775 evt_struct->cmnd_done = done;
777 /* Fix up dma address of the buffer itself */
778 indirect = (struct srp_indirect_buf *) srp_cmd->add_data;
779 out_fmt = srp_cmd->buf_fmt >> 4;
780 in_fmt = srp_cmd->buf_fmt & ((1U << 4) - 1);
781 if ((in_fmt == SRP_DATA_DESC_INDIRECT ||
782 out_fmt == SRP_DATA_DESC_INDIRECT) &&
783 indirect->table_desc.va == 0) {
784 indirect->table_desc.va = evt_struct->crq.IU_data_ptr +
785 offsetof(struct srp_cmd, add_data) +
786 offsetof(struct srp_indirect_buf, desc_list);
789 return ibmvscsi_send_srp_event(evt_struct, hostdata, 0);
792 /* ------------------------------------------------------------
793 * Routines for driver initialization
797 * map_persist_bufs: - Pre-map persistent data for adapter logins
798 * @hostdata: ibmvscsi_host_data of host
800 * Map the capabilities and adapter info DMA buffers to avoid runtime failures.
801 * Return 1 on error, 0 on success.
803 static int map_persist_bufs(struct ibmvscsi_host_data *hostdata)
806 hostdata->caps_addr = dma_map_single(hostdata->dev, &hostdata->caps,
807 sizeof(hostdata->caps), DMA_BIDIRECTIONAL);
809 if (dma_mapping_error(hostdata->dev, hostdata->caps_addr)) {
810 dev_err(hostdata->dev, "Unable to map capabilities buffer!\n");
814 hostdata->adapter_info_addr = dma_map_single(hostdata->dev,
815 &hostdata->madapter_info,
816 sizeof(hostdata->madapter_info),
818 if (dma_mapping_error(hostdata->dev, hostdata->adapter_info_addr)) {
819 dev_err(hostdata->dev, "Unable to map adapter info buffer!\n");
820 dma_unmap_single(hostdata->dev, hostdata->caps_addr,
821 sizeof(hostdata->caps), DMA_BIDIRECTIONAL);
829 * unmap_persist_bufs: - Unmap persistent data needed for adapter logins
830 * @hostdata: ibmvscsi_host_data of host
832 * Unmap the capabilities and adapter info DMA buffers
834 static void unmap_persist_bufs(struct ibmvscsi_host_data *hostdata)
836 dma_unmap_single(hostdata->dev, hostdata->caps_addr,
837 sizeof(hostdata->caps), DMA_BIDIRECTIONAL);
839 dma_unmap_single(hostdata->dev, hostdata->adapter_info_addr,
840 sizeof(hostdata->madapter_info), DMA_BIDIRECTIONAL);
844 * login_rsp: - Handle response to SRP login request
845 * @evt_struct: srp_event_struct with the response
847 * Used as a "done" callback by when sending srp_login. Gets called
848 * by ibmvscsi_handle_crq()
850 static void login_rsp(struct srp_event_struct *evt_struct)
852 struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
853 switch (evt_struct->xfer_iu->srp.login_rsp.opcode) {
854 case SRP_LOGIN_RSP: /* it worked! */
856 case SRP_LOGIN_REJ: /* refused! */
857 dev_info(hostdata->dev, "SRP_LOGIN_REJ reason %u\n",
858 evt_struct->xfer_iu->srp.login_rej.reason);
860 atomic_set(&hostdata->request_limit, -1);
863 dev_err(hostdata->dev, "Invalid login response typecode 0x%02x!\n",
864 evt_struct->xfer_iu->srp.login_rsp.opcode);
866 atomic_set(&hostdata->request_limit, -1);
870 dev_info(hostdata->dev, "SRP_LOGIN succeeded\n");
871 hostdata->client_migrated = 0;
873 /* Now we know what the real request-limit is.
874 * This value is set rather than added to request_limit because
875 * request_limit could have been set to -1 by this client.
877 atomic_set(&hostdata->request_limit,
878 evt_struct->xfer_iu->srp.login_rsp.req_lim_delta);
880 /* If we had any pending I/Os, kick them */
881 scsi_unblock_requests(hostdata->host);
885 * send_srp_login: - Sends the srp login
886 * @hostdata: ibmvscsi_host_data of host
888 * Returns zero if successful.
890 static int send_srp_login(struct ibmvscsi_host_data *hostdata)
894 struct srp_login_req *login;
895 struct srp_event_struct *evt_struct = get_event_struct(&hostdata->pool);
898 init_event_struct(evt_struct, login_rsp,
899 VIOSRP_SRP_FORMAT, login_timeout);
901 login = &evt_struct->iu.srp.login_req;
902 memset(login, 0, sizeof(*login));
903 login->opcode = SRP_LOGIN_REQ;
904 login->req_it_iu_len = sizeof(union srp_iu);
905 login->req_buf_fmt = SRP_BUF_FORMAT_DIRECT | SRP_BUF_FORMAT_INDIRECT;
907 spin_lock_irqsave(hostdata->host->host_lock, flags);
908 /* Start out with a request limit of 0, since this is negotiated in
909 * the login request we are just sending and login requests always
910 * get sent by the driver regardless of request_limit.
912 atomic_set(&hostdata->request_limit, 0);
914 rc = ibmvscsi_send_srp_event(evt_struct, hostdata, login_timeout * 2);
915 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
916 dev_info(hostdata->dev, "sent SRP login\n");
921 * capabilities_rsp: - Handle response to MAD adapter capabilities request
922 * @evt_struct: srp_event_struct with the response
924 * Used as a "done" callback by when sending adapter_info.
926 static void capabilities_rsp(struct srp_event_struct *evt_struct)
928 struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
930 if (evt_struct->xfer_iu->mad.capabilities.common.status) {
931 dev_err(hostdata->dev, "error 0x%X getting capabilities info\n",
932 evt_struct->xfer_iu->mad.capabilities.common.status);
934 if (hostdata->caps.migration.common.server_support != SERVER_SUPPORTS_CAP)
935 dev_info(hostdata->dev, "Partition migration not supported\n");
937 if (client_reserve) {
938 if (hostdata->caps.reserve.common.server_support ==
940 dev_info(hostdata->dev, "Client reserve enabled\n");
942 dev_info(hostdata->dev, "Client reserve not supported\n");
946 send_srp_login(hostdata);
950 * send_mad_capabilities: - Sends the mad capabilities request
951 * and stores the result so it can be retrieved with
952 * @hostdata: ibmvscsi_host_data of host
954 static void send_mad_capabilities(struct ibmvscsi_host_data *hostdata)
956 struct viosrp_capabilities *req;
957 struct srp_event_struct *evt_struct;
959 struct device_node *of_node = hostdata->dev->archdata.of_node;
960 const char *location;
962 evt_struct = get_event_struct(&hostdata->pool);
965 init_event_struct(evt_struct, capabilities_rsp,
966 VIOSRP_MAD_FORMAT, info_timeout);
968 req = &evt_struct->iu.mad.capabilities;
969 memset(req, 0, sizeof(*req));
971 hostdata->caps.flags = CAP_LIST_SUPPORTED;
972 if (hostdata->client_migrated)
973 hostdata->caps.flags |= CLIENT_MIGRATED;
975 strncpy(hostdata->caps.name, dev_name(&hostdata->host->shost_gendev),
976 sizeof(hostdata->caps.name));
977 hostdata->caps.name[sizeof(hostdata->caps.name) - 1] = '\0';
979 location = of_get_property(of_node, "ibm,loc-code", NULL);
980 location = location ? location : dev_name(hostdata->dev);
981 strncpy(hostdata->caps.loc, location, sizeof(hostdata->caps.loc));
982 hostdata->caps.loc[sizeof(hostdata->caps.loc) - 1] = '\0';
984 req->common.type = VIOSRP_CAPABILITIES_TYPE;
985 req->buffer = hostdata->caps_addr;
987 hostdata->caps.migration.common.cap_type = MIGRATION_CAPABILITIES;
988 hostdata->caps.migration.common.length = sizeof(hostdata->caps.migration);
989 hostdata->caps.migration.common.server_support = SERVER_SUPPORTS_CAP;
990 hostdata->caps.migration.ecl = 1;
992 if (client_reserve) {
993 hostdata->caps.reserve.common.cap_type = RESERVATION_CAPABILITIES;
994 hostdata->caps.reserve.common.length = sizeof(hostdata->caps.reserve);
995 hostdata->caps.reserve.common.server_support = SERVER_SUPPORTS_CAP;
996 hostdata->caps.reserve.type = CLIENT_RESERVE_SCSI_2;
997 req->common.length = sizeof(hostdata->caps);
999 req->common.length = sizeof(hostdata->caps) - sizeof(hostdata->caps.reserve);
1001 spin_lock_irqsave(hostdata->host->host_lock, flags);
1002 if (ibmvscsi_send_srp_event(evt_struct, hostdata, info_timeout * 2))
1003 dev_err(hostdata->dev, "couldn't send CAPABILITIES_REQ!\n");
1004 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1008 * fast_fail_rsp: - Handle response to MAD enable fast fail
1009 * @evt_struct: srp_event_struct with the response
1011 * Used as a "done" callback by when sending enable fast fail. Gets called
1012 * by ibmvscsi_handle_crq()
1014 static void fast_fail_rsp(struct srp_event_struct *evt_struct)
1016 struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
1017 u8 status = evt_struct->xfer_iu->mad.fast_fail.common.status;
1019 if (status == VIOSRP_MAD_NOT_SUPPORTED)
1020 dev_err(hostdata->dev, "fast_fail not supported in server\n");
1021 else if (status == VIOSRP_MAD_FAILED)
1022 dev_err(hostdata->dev, "fast_fail request failed\n");
1023 else if (status != VIOSRP_MAD_SUCCESS)
1024 dev_err(hostdata->dev, "error 0x%X enabling fast_fail\n", status);
1026 send_mad_capabilities(hostdata);
1030 * init_host - Start host initialization
1031 * @hostdata: ibmvscsi_host_data of host
1033 * Returns zero if successful.
1035 static int enable_fast_fail(struct ibmvscsi_host_data *hostdata)
1038 unsigned long flags;
1039 struct viosrp_fast_fail *fast_fail_mad;
1040 struct srp_event_struct *evt_struct;
1043 send_mad_capabilities(hostdata);
1047 evt_struct = get_event_struct(&hostdata->pool);
1048 BUG_ON(!evt_struct);
1050 init_event_struct(evt_struct, fast_fail_rsp, VIOSRP_MAD_FORMAT, info_timeout);
1052 fast_fail_mad = &evt_struct->iu.mad.fast_fail;
1053 memset(fast_fail_mad, 0, sizeof(*fast_fail_mad));
1054 fast_fail_mad->common.type = VIOSRP_ENABLE_FAST_FAIL;
1055 fast_fail_mad->common.length = sizeof(*fast_fail_mad);
1057 spin_lock_irqsave(hostdata->host->host_lock, flags);
1058 rc = ibmvscsi_send_srp_event(evt_struct, hostdata, info_timeout * 2);
1059 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1064 * adapter_info_rsp: - Handle response to MAD adapter info request
1065 * @evt_struct: srp_event_struct with the response
1067 * Used as a "done" callback by when sending adapter_info. Gets called
1068 * by ibmvscsi_handle_crq()
1070 static void adapter_info_rsp(struct srp_event_struct *evt_struct)
1072 struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
1074 if (evt_struct->xfer_iu->mad.adapter_info.common.status) {
1075 dev_err(hostdata->dev, "error %d getting adapter info\n",
1076 evt_struct->xfer_iu->mad.adapter_info.common.status);
1078 dev_info(hostdata->dev, "host srp version: %s, "
1079 "host partition %s (%d), OS %d, max io %u\n",
1080 hostdata->madapter_info.srp_version,
1081 hostdata->madapter_info.partition_name,
1082 hostdata->madapter_info.partition_number,
1083 hostdata->madapter_info.os_type,
1084 hostdata->madapter_info.port_max_txu[0]);
1086 if (hostdata->madapter_info.port_max_txu[0])
1087 hostdata->host->max_sectors =
1088 hostdata->madapter_info.port_max_txu[0] >> 9;
1090 if (hostdata->madapter_info.os_type == 3 &&
1091 strcmp(hostdata->madapter_info.srp_version, "1.6a") <= 0) {
1092 dev_err(hostdata->dev, "host (Ver. %s) doesn't support large transfers\n",
1093 hostdata->madapter_info.srp_version);
1094 dev_err(hostdata->dev, "limiting scatterlists to %d\n",
1096 hostdata->host->sg_tablesize = MAX_INDIRECT_BUFS;
1100 enable_fast_fail(hostdata);
1104 * send_mad_adapter_info: - Sends the mad adapter info request
1105 * and stores the result so it can be retrieved with
1106 * sysfs. We COULD consider causing a failure if the
1107 * returned SRP version doesn't match ours.
1108 * @hostdata: ibmvscsi_host_data of host
1110 * Returns zero if successful.
1112 static void send_mad_adapter_info(struct ibmvscsi_host_data *hostdata)
1114 struct viosrp_adapter_info *req;
1115 struct srp_event_struct *evt_struct;
1116 unsigned long flags;
1118 evt_struct = get_event_struct(&hostdata->pool);
1119 BUG_ON(!evt_struct);
1121 init_event_struct(evt_struct,
1126 req = &evt_struct->iu.mad.adapter_info;
1127 memset(req, 0x00, sizeof(*req));
1129 req->common.type = VIOSRP_ADAPTER_INFO_TYPE;
1130 req->common.length = sizeof(hostdata->madapter_info);
1131 req->buffer = hostdata->adapter_info_addr;
1133 spin_lock_irqsave(hostdata->host->host_lock, flags);
1134 if (ibmvscsi_send_srp_event(evt_struct, hostdata, info_timeout * 2))
1135 dev_err(hostdata->dev, "couldn't send ADAPTER_INFO_REQ!\n");
1136 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1140 * init_adapter: Start virtual adapter initialization sequence
1143 static void init_adapter(struct ibmvscsi_host_data *hostdata)
1145 send_mad_adapter_info(hostdata);
1149 * sync_completion: Signal that a synchronous command has completed
1150 * Note that after returning from this call, the evt_struct is freed.
1151 * the caller waiting on this completion shouldn't touch the evt_struct
1154 static void sync_completion(struct srp_event_struct *evt_struct)
1156 /* copy the response back */
1157 if (evt_struct->sync_srp)
1158 *evt_struct->sync_srp = *evt_struct->xfer_iu;
1160 complete(&evt_struct->comp);
1164 * ibmvscsi_abort: Abort a command...from scsi host template
1165 * send this over to the server and wait synchronously for the response
1167 static int ibmvscsi_eh_abort_handler(struct scsi_cmnd *cmd)
1169 struct ibmvscsi_host_data *hostdata = shost_priv(cmd->device->host);
1170 struct srp_tsk_mgmt *tsk_mgmt;
1171 struct srp_event_struct *evt;
1172 struct srp_event_struct *tmp_evt, *found_evt;
1173 union viosrp_iu srp_rsp;
1175 unsigned long flags;
1176 u16 lun = lun_from_dev(cmd->device);
1177 unsigned long wait_switch = 0;
1179 /* First, find this command in our sent list so we can figure
1180 * out the correct tag
1182 spin_lock_irqsave(hostdata->host->host_lock, flags);
1183 wait_switch = jiffies + (init_timeout * HZ);
1186 list_for_each_entry(tmp_evt, &hostdata->sent, list) {
1187 if (tmp_evt->cmnd == cmd) {
1188 found_evt = tmp_evt;
1194 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1198 evt = get_event_struct(&hostdata->pool);
1200 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1201 sdev_printk(KERN_ERR, cmd->device,
1202 "failed to allocate abort event\n");
1206 init_event_struct(evt,
1211 tsk_mgmt = &evt->iu.srp.tsk_mgmt;
1213 /* Set up an abort SRP command */
1214 memset(tsk_mgmt, 0x00, sizeof(*tsk_mgmt));
1215 tsk_mgmt->opcode = SRP_TSK_MGMT;
1216 tsk_mgmt->lun = ((u64) lun) << 48;
1217 tsk_mgmt->tsk_mgmt_func = SRP_TSK_ABORT_TASK;
1218 tsk_mgmt->task_tag = (u64) found_evt;
1220 evt->sync_srp = &srp_rsp;
1222 init_completion(&evt->comp);
1223 rsp_rc = ibmvscsi_send_srp_event(evt, hostdata, abort_timeout * 2);
1225 if (rsp_rc != SCSI_MLQUEUE_HOST_BUSY)
1228 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1230 spin_lock_irqsave(hostdata->host->host_lock, flags);
1231 } while (time_before(jiffies, wait_switch));
1233 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1236 sdev_printk(KERN_ERR, cmd->device,
1237 "failed to send abort() event. rc=%d\n", rsp_rc);
1241 sdev_printk(KERN_INFO, cmd->device,
1242 "aborting command. lun 0x%llx, tag 0x%llx\n",
1243 (((u64) lun) << 48), (u64) found_evt);
1245 wait_for_completion(&evt->comp);
1247 /* make sure we got a good response */
1248 if (unlikely(srp_rsp.srp.rsp.opcode != SRP_RSP)) {
1249 if (printk_ratelimit())
1250 sdev_printk(KERN_WARNING, cmd->device, "abort bad SRP RSP type %d\n",
1251 srp_rsp.srp.rsp.opcode);
1255 if (srp_rsp.srp.rsp.flags & SRP_RSP_FLAG_RSPVALID)
1256 rsp_rc = *((int *)srp_rsp.srp.rsp.data);
1258 rsp_rc = srp_rsp.srp.rsp.status;
1261 if (printk_ratelimit())
1262 sdev_printk(KERN_WARNING, cmd->device,
1263 "abort code %d for task tag 0x%llx\n",
1264 rsp_rc, tsk_mgmt->task_tag);
1268 /* Because we dropped the spinlock above, it's possible
1269 * The event is no longer in our list. Make sure it didn't
1270 * complete while we were aborting
1272 spin_lock_irqsave(hostdata->host->host_lock, flags);
1274 list_for_each_entry(tmp_evt, &hostdata->sent, list) {
1275 if (tmp_evt->cmnd == cmd) {
1276 found_evt = tmp_evt;
1281 if (found_evt == NULL) {
1282 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1283 sdev_printk(KERN_INFO, cmd->device, "aborted task tag 0x%llx completed\n",
1284 tsk_mgmt->task_tag);
1288 sdev_printk(KERN_INFO, cmd->device, "successfully aborted task tag 0x%llx\n",
1289 tsk_mgmt->task_tag);
1291 cmd->result = (DID_ABORT << 16);
1292 list_del(&found_evt->list);
1293 unmap_cmd_data(&found_evt->iu.srp.cmd, found_evt,
1294 found_evt->hostdata->dev);
1295 free_event_struct(&found_evt->hostdata->pool, found_evt);
1296 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1297 atomic_inc(&hostdata->request_limit);
1302 * ibmvscsi_eh_device_reset_handler: Reset a single LUN...from scsi host
1303 * template send this over to the server and wait synchronously for the
1306 static int ibmvscsi_eh_device_reset_handler(struct scsi_cmnd *cmd)
1308 struct ibmvscsi_host_data *hostdata = shost_priv(cmd->device->host);
1309 struct srp_tsk_mgmt *tsk_mgmt;
1310 struct srp_event_struct *evt;
1311 struct srp_event_struct *tmp_evt, *pos;
1312 union viosrp_iu srp_rsp;
1314 unsigned long flags;
1315 u16 lun = lun_from_dev(cmd->device);
1316 unsigned long wait_switch = 0;
1318 spin_lock_irqsave(hostdata->host->host_lock, flags);
1319 wait_switch = jiffies + (init_timeout * HZ);
1321 evt = get_event_struct(&hostdata->pool);
1323 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1324 sdev_printk(KERN_ERR, cmd->device,
1325 "failed to allocate reset event\n");
1329 init_event_struct(evt,
1334 tsk_mgmt = &evt->iu.srp.tsk_mgmt;
1336 /* Set up a lun reset SRP command */
1337 memset(tsk_mgmt, 0x00, sizeof(*tsk_mgmt));
1338 tsk_mgmt->opcode = SRP_TSK_MGMT;
1339 tsk_mgmt->lun = ((u64) lun) << 48;
1340 tsk_mgmt->tsk_mgmt_func = SRP_TSK_LUN_RESET;
1342 evt->sync_srp = &srp_rsp;
1344 init_completion(&evt->comp);
1345 rsp_rc = ibmvscsi_send_srp_event(evt, hostdata, reset_timeout * 2);
1347 if (rsp_rc != SCSI_MLQUEUE_HOST_BUSY)
1350 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1352 spin_lock_irqsave(hostdata->host->host_lock, flags);
1353 } while (time_before(jiffies, wait_switch));
1355 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1358 sdev_printk(KERN_ERR, cmd->device,
1359 "failed to send reset event. rc=%d\n", rsp_rc);
1363 sdev_printk(KERN_INFO, cmd->device, "resetting device. lun 0x%llx\n",
1364 (((u64) lun) << 48));
1366 wait_for_completion(&evt->comp);
1368 /* make sure we got a good response */
1369 if (unlikely(srp_rsp.srp.rsp.opcode != SRP_RSP)) {
1370 if (printk_ratelimit())
1371 sdev_printk(KERN_WARNING, cmd->device, "reset bad SRP RSP type %d\n",
1372 srp_rsp.srp.rsp.opcode);
1376 if (srp_rsp.srp.rsp.flags & SRP_RSP_FLAG_RSPVALID)
1377 rsp_rc = *((int *)srp_rsp.srp.rsp.data);
1379 rsp_rc = srp_rsp.srp.rsp.status;
1382 if (printk_ratelimit())
1383 sdev_printk(KERN_WARNING, cmd->device,
1384 "reset code %d for task tag 0x%llx\n",
1385 rsp_rc, tsk_mgmt->task_tag);
1389 /* We need to find all commands for this LUN that have not yet been
1390 * responded to, and fail them with DID_RESET
1392 spin_lock_irqsave(hostdata->host->host_lock, flags);
1393 list_for_each_entry_safe(tmp_evt, pos, &hostdata->sent, list) {
1394 if ((tmp_evt->cmnd) && (tmp_evt->cmnd->device == cmd->device)) {
1396 tmp_evt->cmnd->result = (DID_RESET << 16);
1397 list_del(&tmp_evt->list);
1398 unmap_cmd_data(&tmp_evt->iu.srp.cmd, tmp_evt,
1399 tmp_evt->hostdata->dev);
1400 free_event_struct(&tmp_evt->hostdata->pool,
1402 atomic_inc(&hostdata->request_limit);
1403 if (tmp_evt->cmnd_done)
1404 tmp_evt->cmnd_done(tmp_evt->cmnd);
1405 else if (tmp_evt->done)
1406 tmp_evt->done(tmp_evt);
1409 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1414 * ibmvscsi_eh_host_reset_handler - Reset the connection to the server
1415 * @cmd: struct scsi_cmnd having problems
1417 static int ibmvscsi_eh_host_reset_handler(struct scsi_cmnd *cmd)
1419 unsigned long wait_switch = 0;
1420 struct ibmvscsi_host_data *hostdata = shost_priv(cmd->device->host);
1422 dev_err(hostdata->dev, "Resetting connection due to error recovery\n");
1424 ibmvscsi_reset_host(hostdata);
1426 for (wait_switch = jiffies + (init_timeout * HZ);
1427 time_before(jiffies, wait_switch) &&
1428 atomic_read(&hostdata->request_limit) < 2;) {
1433 if (atomic_read(&hostdata->request_limit) <= 0)
1440 * ibmvscsi_handle_crq: - Handles and frees received events in the CRQ
1441 * @crq: Command/Response queue
1442 * @hostdata: ibmvscsi_host_data of host
1445 void ibmvscsi_handle_crq(struct viosrp_crq *crq,
1446 struct ibmvscsi_host_data *hostdata)
1449 unsigned long flags;
1450 struct srp_event_struct *evt_struct =
1451 (struct srp_event_struct *)crq->IU_data_ptr;
1452 switch (crq->valid) {
1453 case 0xC0: /* initialization */
1454 switch (crq->format) {
1455 case 0x01: /* Initialization message */
1456 dev_info(hostdata->dev, "partner initialized\n");
1457 /* Send back a response */
1458 if ((rc = ibmvscsi_ops->send_crq(hostdata,
1459 0xC002000000000000LL, 0)) == 0) {
1461 init_adapter(hostdata);
1463 dev_err(hostdata->dev, "Unable to send init rsp. rc=%ld\n", rc);
1467 case 0x02: /* Initialization response */
1468 dev_info(hostdata->dev, "partner initialization complete\n");
1471 init_adapter(hostdata);
1474 dev_err(hostdata->dev, "unknown crq message type: %d\n", crq->format);
1477 case 0xFF: /* Hypervisor telling us the connection is closed */
1478 scsi_block_requests(hostdata->host);
1479 atomic_set(&hostdata->request_limit, 0);
1480 if (crq->format == 0x06) {
1481 /* We need to re-setup the interpartition connection */
1482 dev_info(hostdata->dev, "Re-enabling adapter!\n");
1483 hostdata->client_migrated = 1;
1484 purge_requests(hostdata, DID_REQUEUE);
1485 if ((ibmvscsi_ops->reenable_crq_queue(&hostdata->queue,
1487 (ibmvscsi_ops->send_crq(hostdata,
1488 0xC001000000000000LL, 0))) {
1489 atomic_set(&hostdata->request_limit,
1491 dev_err(hostdata->dev, "error after enable\n");
1494 dev_err(hostdata->dev, "Virtual adapter failed rc %d!\n",
1497 purge_requests(hostdata, DID_ERROR);
1498 if ((ibmvscsi_ops->reset_crq_queue(&hostdata->queue,
1500 (ibmvscsi_ops->send_crq(hostdata,
1501 0xC001000000000000LL, 0))) {
1502 atomic_set(&hostdata->request_limit,
1504 dev_err(hostdata->dev, "error after reset\n");
1507 scsi_unblock_requests(hostdata->host);
1509 case 0x80: /* real payload */
1512 dev_err(hostdata->dev, "got an invalid message type 0x%02x\n",
1517 /* The only kind of payload CRQs we should get are responses to
1518 * things we send. Make sure this response is to something we
1521 if (!valid_event_struct(&hostdata->pool, evt_struct)) {
1522 dev_err(hostdata->dev, "returned correlation_token 0x%p is invalid!\n",
1523 (void *)crq->IU_data_ptr);
1527 if (atomic_read(&evt_struct->free)) {
1528 dev_err(hostdata->dev, "received duplicate correlation_token 0x%p!\n",
1529 (void *)crq->IU_data_ptr);
1533 if (crq->format == VIOSRP_SRP_FORMAT)
1534 atomic_add(evt_struct->xfer_iu->srp.rsp.req_lim_delta,
1535 &hostdata->request_limit);
1537 del_timer(&evt_struct->timer);
1539 if ((crq->status != VIOSRP_OK && crq->status != VIOSRP_OK2) && evt_struct->cmnd)
1540 evt_struct->cmnd->result = DID_ERROR << 16;
1541 if (evt_struct->done)
1542 evt_struct->done(evt_struct);
1544 dev_err(hostdata->dev, "returned done() is NULL; not running it!\n");
1547 * Lock the host_lock before messing with these structures, since we
1548 * are running in a task context
1550 spin_lock_irqsave(evt_struct->hostdata->host->host_lock, flags);
1551 list_del(&evt_struct->list);
1552 free_event_struct(&evt_struct->hostdata->pool, evt_struct);
1553 spin_unlock_irqrestore(evt_struct->hostdata->host->host_lock, flags);
1557 * ibmvscsi_get_host_config: Send the command to the server to get host
1558 * configuration data. The data is opaque to us.
1560 static int ibmvscsi_do_host_config(struct ibmvscsi_host_data *hostdata,
1561 unsigned char *buffer, int length)
1563 struct viosrp_host_config *host_config;
1564 struct srp_event_struct *evt_struct;
1565 unsigned long flags;
1569 evt_struct = get_event_struct(&hostdata->pool);
1571 dev_err(hostdata->dev, "couldn't allocate event for HOST_CONFIG!\n");
1575 init_event_struct(evt_struct,
1580 host_config = &evt_struct->iu.mad.host_config;
1582 /* Set up a lun reset SRP command */
1583 memset(host_config, 0x00, sizeof(*host_config));
1584 host_config->common.type = VIOSRP_HOST_CONFIG_TYPE;
1585 host_config->common.length = length;
1586 host_config->buffer = addr = dma_map_single(hostdata->dev, buffer,
1590 if (dma_mapping_error(hostdata->dev, host_config->buffer)) {
1591 if (!firmware_has_feature(FW_FEATURE_CMO))
1592 dev_err(hostdata->dev,
1593 "dma_mapping error getting host config\n");
1594 free_event_struct(&hostdata->pool, evt_struct);
1598 init_completion(&evt_struct->comp);
1599 spin_lock_irqsave(hostdata->host->host_lock, flags);
1600 rc = ibmvscsi_send_srp_event(evt_struct, hostdata, info_timeout * 2);
1601 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1603 wait_for_completion(&evt_struct->comp);
1604 dma_unmap_single(hostdata->dev, addr, length, DMA_BIDIRECTIONAL);
1610 * ibmvscsi_slave_configure: Set the "allow_restart" flag for each disk.
1611 * @sdev: struct scsi_device device to configure
1613 * Enable allow_restart for a device if it is a disk. Adjust the
1614 * queue_depth here also as is required by the documentation for
1615 * struct scsi_host_template.
1617 static int ibmvscsi_slave_configure(struct scsi_device *sdev)
1619 struct Scsi_Host *shost = sdev->host;
1620 unsigned long lock_flags = 0;
1622 spin_lock_irqsave(shost->host_lock, lock_flags);
1623 if (sdev->type == TYPE_DISK) {
1624 sdev->allow_restart = 1;
1625 blk_queue_rq_timeout(sdev->request_queue, 120 * HZ);
1627 scsi_adjust_queue_depth(sdev, 0, shost->cmd_per_lun);
1628 spin_unlock_irqrestore(shost->host_lock, lock_flags);
1633 * ibmvscsi_change_queue_depth - Change the device's queue depth
1634 * @sdev: scsi device struct
1635 * @qdepth: depth to set
1640 static int ibmvscsi_change_queue_depth(struct scsi_device *sdev, int qdepth)
1642 if (qdepth > IBMVSCSI_MAX_CMDS_PER_LUN)
1643 qdepth = IBMVSCSI_MAX_CMDS_PER_LUN;
1645 scsi_adjust_queue_depth(sdev, 0, qdepth);
1646 return sdev->queue_depth;
1649 /* ------------------------------------------------------------
1652 static ssize_t show_host_vhost_loc(struct device *dev,
1653 struct device_attribute *attr, char *buf)
1655 struct Scsi_Host *shost = class_to_shost(dev);
1656 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1659 len = snprintf(buf, sizeof(hostdata->caps.loc), "%s\n",
1660 hostdata->caps.loc);
1664 static struct device_attribute ibmvscsi_host_vhost_loc = {
1666 .name = "vhost_loc",
1669 .show = show_host_vhost_loc,
1672 static ssize_t show_host_vhost_name(struct device *dev,
1673 struct device_attribute *attr, char *buf)
1675 struct Scsi_Host *shost = class_to_shost(dev);
1676 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1679 len = snprintf(buf, sizeof(hostdata->caps.name), "%s\n",
1680 hostdata->caps.name);
1684 static struct device_attribute ibmvscsi_host_vhost_name = {
1686 .name = "vhost_name",
1689 .show = show_host_vhost_name,
1692 static ssize_t show_host_srp_version(struct device *dev,
1693 struct device_attribute *attr, char *buf)
1695 struct Scsi_Host *shost = class_to_shost(dev);
1696 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1699 len = snprintf(buf, PAGE_SIZE, "%s\n",
1700 hostdata->madapter_info.srp_version);
1704 static struct device_attribute ibmvscsi_host_srp_version = {
1706 .name = "srp_version",
1709 .show = show_host_srp_version,
1712 static ssize_t show_host_partition_name(struct device *dev,
1713 struct device_attribute *attr,
1716 struct Scsi_Host *shost = class_to_shost(dev);
1717 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1720 len = snprintf(buf, PAGE_SIZE, "%s\n",
1721 hostdata->madapter_info.partition_name);
1725 static struct device_attribute ibmvscsi_host_partition_name = {
1727 .name = "partition_name",
1730 .show = show_host_partition_name,
1733 static ssize_t show_host_partition_number(struct device *dev,
1734 struct device_attribute *attr,
1737 struct Scsi_Host *shost = class_to_shost(dev);
1738 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1741 len = snprintf(buf, PAGE_SIZE, "%d\n",
1742 hostdata->madapter_info.partition_number);
1746 static struct device_attribute ibmvscsi_host_partition_number = {
1748 .name = "partition_number",
1751 .show = show_host_partition_number,
1754 static ssize_t show_host_mad_version(struct device *dev,
1755 struct device_attribute *attr, char *buf)
1757 struct Scsi_Host *shost = class_to_shost(dev);
1758 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1761 len = snprintf(buf, PAGE_SIZE, "%d\n",
1762 hostdata->madapter_info.mad_version);
1766 static struct device_attribute ibmvscsi_host_mad_version = {
1768 .name = "mad_version",
1771 .show = show_host_mad_version,
1774 static ssize_t show_host_os_type(struct device *dev,
1775 struct device_attribute *attr, char *buf)
1777 struct Scsi_Host *shost = class_to_shost(dev);
1778 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1781 len = snprintf(buf, PAGE_SIZE, "%d\n", hostdata->madapter_info.os_type);
1785 static struct device_attribute ibmvscsi_host_os_type = {
1790 .show = show_host_os_type,
1793 static ssize_t show_host_config(struct device *dev,
1794 struct device_attribute *attr, char *buf)
1796 struct Scsi_Host *shost = class_to_shost(dev);
1797 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1799 /* returns null-terminated host config data */
1800 if (ibmvscsi_do_host_config(hostdata, buf, PAGE_SIZE) == 0)
1806 static struct device_attribute ibmvscsi_host_config = {
1811 .show = show_host_config,
1814 static struct device_attribute *ibmvscsi_attrs[] = {
1815 &ibmvscsi_host_vhost_loc,
1816 &ibmvscsi_host_vhost_name,
1817 &ibmvscsi_host_srp_version,
1818 &ibmvscsi_host_partition_name,
1819 &ibmvscsi_host_partition_number,
1820 &ibmvscsi_host_mad_version,
1821 &ibmvscsi_host_os_type,
1822 &ibmvscsi_host_config,
1826 /* ------------------------------------------------------------
1827 * SCSI driver registration
1829 static struct scsi_host_template driver_template = {
1830 .module = THIS_MODULE,
1831 .name = "IBM POWER Virtual SCSI Adapter " IBMVSCSI_VERSION,
1832 .proc_name = "ibmvscsi",
1833 .queuecommand = ibmvscsi_queuecommand,
1834 .eh_abort_handler = ibmvscsi_eh_abort_handler,
1835 .eh_device_reset_handler = ibmvscsi_eh_device_reset_handler,
1836 .eh_host_reset_handler = ibmvscsi_eh_host_reset_handler,
1837 .slave_configure = ibmvscsi_slave_configure,
1838 .change_queue_depth = ibmvscsi_change_queue_depth,
1839 .cmd_per_lun = IBMVSCSI_CMDS_PER_LUN_DEFAULT,
1840 .can_queue = IBMVSCSI_MAX_REQUESTS_DEFAULT,
1842 .sg_tablesize = SG_ALL,
1843 .use_clustering = ENABLE_CLUSTERING,
1844 .shost_attrs = ibmvscsi_attrs,
1848 * ibmvscsi_get_desired_dma - Calculate IO memory desired by the driver
1850 * @vdev: struct vio_dev for the device whose desired IO mem is to be returned
1853 * Number of bytes of IO data the driver will need to perform well.
1855 static unsigned long ibmvscsi_get_desired_dma(struct vio_dev *vdev)
1857 /* iu_storage data allocated in initialize_event_pool */
1858 unsigned long desired_io = max_events * sizeof(union viosrp_iu);
1860 /* add io space for sg data */
1861 desired_io += (IBMVSCSI_MAX_SECTORS_DEFAULT * 512 *
1862 IBMVSCSI_CMDS_PER_LUN_DEFAULT);
1868 * Called by bus code for each adapter
1870 static int ibmvscsi_probe(struct vio_dev *vdev, const struct vio_device_id *id)
1872 struct ibmvscsi_host_data *hostdata;
1873 struct Scsi_Host *host;
1874 struct device *dev = &vdev->dev;
1875 struct srp_rport_identifiers ids;
1876 struct srp_rport *rport;
1877 unsigned long wait_switch = 0;
1880 dev_set_drvdata(&vdev->dev, NULL);
1882 host = scsi_host_alloc(&driver_template, sizeof(*hostdata));
1884 dev_err(&vdev->dev, "couldn't allocate host data\n");
1885 goto scsi_host_alloc_failed;
1888 host->transportt = ibmvscsi_transport_template;
1889 hostdata = shost_priv(host);
1890 memset(hostdata, 0x00, sizeof(*hostdata));
1891 INIT_LIST_HEAD(&hostdata->sent);
1892 hostdata->host = host;
1893 hostdata->dev = dev;
1894 atomic_set(&hostdata->request_limit, -1);
1895 hostdata->host->max_sectors = IBMVSCSI_MAX_SECTORS_DEFAULT;
1897 if (map_persist_bufs(hostdata)) {
1898 dev_err(&vdev->dev, "couldn't map persistent buffers\n");
1899 goto persist_bufs_failed;
1902 rc = ibmvscsi_ops->init_crq_queue(&hostdata->queue, hostdata, max_events);
1903 if (rc != 0 && rc != H_RESOURCE) {
1904 dev_err(&vdev->dev, "couldn't initialize crq. rc=%d\n", rc);
1905 goto init_crq_failed;
1907 if (initialize_event_pool(&hostdata->pool, max_events, hostdata) != 0) {
1908 dev_err(&vdev->dev, "couldn't initialize event pool\n");
1909 goto init_pool_failed;
1913 host->max_id = max_id;
1914 host->max_channel = max_channel;
1915 host->max_cmd_len = 16;
1917 if (scsi_add_host(hostdata->host, hostdata->dev))
1918 goto add_host_failed;
1920 /* we don't have a proper target_port_id so let's use the fake one */
1921 memcpy(ids.port_id, hostdata->madapter_info.partition_name,
1922 sizeof(ids.port_id));
1923 ids.roles = SRP_RPORT_ROLE_TARGET;
1924 rport = srp_rport_add(host, &ids);
1926 goto add_srp_port_failed;
1928 /* Try to send an initialization message. Note that this is allowed
1929 * to fail if the other end is not acive. In that case we don't
1932 if (ibmvscsi_ops->send_crq(hostdata, 0xC001000000000000LL, 0) == 0
1933 || rc == H_RESOURCE) {
1935 * Wait around max init_timeout secs for the adapter to finish
1936 * initializing. When we are done initializing, we will have a
1937 * valid request_limit. We don't want Linux scanning before
1940 for (wait_switch = jiffies + (init_timeout * HZ);
1941 time_before(jiffies, wait_switch) &&
1942 atomic_read(&hostdata->request_limit) < 2;) {
1947 /* if we now have a valid request_limit, initiate a scan */
1948 if (atomic_read(&hostdata->request_limit) > 0)
1949 scsi_scan_host(host);
1952 dev_set_drvdata(&vdev->dev, hostdata);
1955 add_srp_port_failed:
1956 scsi_remove_host(hostdata->host);
1958 release_event_pool(&hostdata->pool, hostdata);
1960 ibmvscsi_ops->release_crq_queue(&hostdata->queue, hostdata, max_events);
1962 unmap_persist_bufs(hostdata);
1963 persist_bufs_failed:
1964 scsi_host_put(host);
1965 scsi_host_alloc_failed:
1969 static int ibmvscsi_remove(struct vio_dev *vdev)
1971 struct ibmvscsi_host_data *hostdata = dev_get_drvdata(&vdev->dev);
1972 unmap_persist_bufs(hostdata);
1973 release_event_pool(&hostdata->pool, hostdata);
1974 ibmvscsi_ops->release_crq_queue(&hostdata->queue, hostdata,
1977 srp_remove_host(hostdata->host);
1978 scsi_remove_host(hostdata->host);
1979 scsi_host_put(hostdata->host);
1985 * ibmvscsi_device_table: Used by vio.c to match devices in the device tree we
1988 static struct vio_device_id ibmvscsi_device_table[] __devinitdata = {
1989 {"vscsi", "IBM,v-scsi"},
1992 MODULE_DEVICE_TABLE(vio, ibmvscsi_device_table);
1994 static struct vio_driver ibmvscsi_driver = {
1995 .id_table = ibmvscsi_device_table,
1996 .probe = ibmvscsi_probe,
1997 .remove = ibmvscsi_remove,
1998 .get_desired_dma = ibmvscsi_get_desired_dma,
2001 .owner = THIS_MODULE,
2005 static struct srp_function_template ibmvscsi_transport_functions = {
2008 int __init ibmvscsi_module_init(void)
2012 /* Ensure we have two requests to do error recovery */
2013 driver_template.can_queue = max_requests;
2014 max_events = max_requests + 2;
2016 if (firmware_has_feature(FW_FEATURE_ISERIES))
2017 ibmvscsi_ops = &iseriesvscsi_ops;
2018 else if (firmware_has_feature(FW_FEATURE_VIO))
2019 ibmvscsi_ops = &rpavscsi_ops;
2023 ibmvscsi_transport_template =
2024 srp_attach_transport(&ibmvscsi_transport_functions);
2025 if (!ibmvscsi_transport_template)
2028 ret = vio_register_driver(&ibmvscsi_driver);
2030 srp_release_transport(ibmvscsi_transport_template);
2034 void __exit ibmvscsi_module_exit(void)
2036 vio_unregister_driver(&ibmvscsi_driver);
2037 srp_release_transport(ibmvscsi_transport_template);
2040 module_init(ibmvscsi_module_init);
2041 module_exit(ibmvscsi_module_exit);