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 * TODO: This is currently pretty tied to the IBM pSeries hypervisor
59 * interfaces. It would be really nice to abstract this above an RDMA
63 #include <linux/module.h>
64 #include <linux/moduleparam.h>
65 #include <linux/dma-mapping.h>
66 #include <linux/delay.h>
67 #include <linux/slab.h>
70 #include <linux/kthread.h>
71 #include <asm/firmware.h>
73 #include <scsi/scsi.h>
74 #include <scsi/scsi_cmnd.h>
75 #include <scsi/scsi_host.h>
76 #include <scsi/scsi_device.h>
77 #include <scsi/scsi_transport_srp.h>
80 /* The values below are somewhat arbitrary default values, but
81 * OS/400 will use 3 busses (disks, CDs, tapes, I think.)
82 * Note that there are 3 bits of channel value, 6 bits of id, and
85 static int max_id = 64;
86 static int max_channel = 3;
87 static int init_timeout = 300;
88 static int login_timeout = 60;
89 static int info_timeout = 30;
90 static int abort_timeout = 60;
91 static int reset_timeout = 60;
92 static int max_requests = IBMVSCSI_MAX_REQUESTS_DEFAULT;
93 static int max_events = IBMVSCSI_MAX_REQUESTS_DEFAULT + 2;
94 static int fast_fail = 1;
95 static int client_reserve = 1;
97 static struct scsi_transport_template *ibmvscsi_transport_template;
99 #define IBMVSCSI_VERSION "1.5.9"
101 static struct ibmvscsi_ops *ibmvscsi_ops;
103 MODULE_DESCRIPTION("IBM Virtual SCSI");
104 MODULE_AUTHOR("Dave Boutcher");
105 MODULE_LICENSE("GPL");
106 MODULE_VERSION(IBMVSCSI_VERSION);
108 module_param_named(max_id, max_id, int, S_IRUGO | S_IWUSR);
109 MODULE_PARM_DESC(max_id, "Largest ID value for each channel");
110 module_param_named(max_channel, max_channel, int, S_IRUGO | S_IWUSR);
111 MODULE_PARM_DESC(max_channel, "Largest channel value");
112 module_param_named(init_timeout, init_timeout, int, S_IRUGO | S_IWUSR);
113 MODULE_PARM_DESC(init_timeout, "Initialization timeout in seconds");
114 module_param_named(max_requests, max_requests, int, S_IRUGO);
115 MODULE_PARM_DESC(max_requests, "Maximum requests for this adapter");
116 module_param_named(fast_fail, fast_fail, int, S_IRUGO | S_IWUSR);
117 MODULE_PARM_DESC(fast_fail, "Enable fast fail. [Default=1]");
118 module_param_named(client_reserve, client_reserve, int, S_IRUGO );
119 MODULE_PARM_DESC(client_reserve, "Attempt client managed reserve/release");
121 /* ------------------------------------------------------------
122 * Routines for the event pool and event structs
125 * initialize_event_pool: - Allocates and initializes the event pool for a host
126 * @pool: event_pool to be initialized
127 * @size: Number of events in pool
128 * @hostdata: ibmvscsi_host_data who owns the event pool
130 * Returns zero on success.
132 static int initialize_event_pool(struct event_pool *pool,
133 int size, struct ibmvscsi_host_data *hostdata)
139 pool->events = kcalloc(pool->size, sizeof(*pool->events), GFP_KERNEL);
144 dma_alloc_coherent(hostdata->dev,
145 pool->size * sizeof(*pool->iu_storage),
147 if (!pool->iu_storage) {
152 for (i = 0; i < pool->size; ++i) {
153 struct srp_event_struct *evt = &pool->events[i];
154 memset(&evt->crq, 0x00, sizeof(evt->crq));
155 atomic_set(&evt->free, 1);
156 evt->crq.valid = 0x80;
157 evt->crq.IU_length = sizeof(*evt->xfer_iu);
158 evt->crq.IU_data_ptr = pool->iu_token +
159 sizeof(*evt->xfer_iu) * i;
160 evt->xfer_iu = pool->iu_storage + i;
161 evt->hostdata = hostdata;
162 evt->ext_list = NULL;
163 evt->ext_list_token = 0;
170 * release_event_pool: - Frees memory of an event pool of a host
171 * @pool: event_pool to be released
172 * @hostdata: ibmvscsi_host_data who owns the even pool
174 * Returns zero on success.
176 static void release_event_pool(struct event_pool *pool,
177 struct ibmvscsi_host_data *hostdata)
180 for (i = 0; i < pool->size; ++i) {
181 if (atomic_read(&pool->events[i].free) != 1)
183 if (pool->events[i].ext_list) {
184 dma_free_coherent(hostdata->dev,
185 SG_ALL * sizeof(struct srp_direct_buf),
186 pool->events[i].ext_list,
187 pool->events[i].ext_list_token);
191 dev_warn(hostdata->dev, "releasing event pool with %d "
192 "events still in use?\n", in_use);
194 dma_free_coherent(hostdata->dev,
195 pool->size * sizeof(*pool->iu_storage),
196 pool->iu_storage, pool->iu_token);
200 * valid_event_struct: - Determines if event is valid.
201 * @pool: event_pool that contains the event
202 * @evt: srp_event_struct to be checked for validity
204 * Returns zero if event is invalid, one otherwise.
206 static int valid_event_struct(struct event_pool *pool,
207 struct srp_event_struct *evt)
209 int index = evt - pool->events;
210 if (index < 0 || index >= pool->size) /* outside of bounds */
212 if (evt != pool->events + index) /* unaligned */
218 * ibmvscsi_free-event_struct: - Changes status of event to "free"
219 * @pool: event_pool that contains the event
220 * @evt: srp_event_struct to be modified
223 static void free_event_struct(struct event_pool *pool,
224 struct srp_event_struct *evt)
226 if (!valid_event_struct(pool, evt)) {
227 dev_err(evt->hostdata->dev, "Freeing invalid event_struct %p "
228 "(not in pool %p)\n", evt, pool->events);
231 if (atomic_inc_return(&evt->free) != 1) {
232 dev_err(evt->hostdata->dev, "Freeing event_struct %p "
233 "which is not in use!\n", evt);
239 * get_evt_struct: - Gets the next free event in pool
240 * @pool: event_pool that contains the events to be searched
242 * Returns the next event in "free" state, and NULL if none are free.
243 * Note that no synchronization is done here, we assume the host_lock
244 * will syncrhonze things.
246 static struct srp_event_struct *get_event_struct(struct event_pool *pool)
249 int poolsize = pool->size;
250 int offset = pool->next;
252 for (i = 0; i < poolsize; i++) {
253 offset = (offset + 1) % poolsize;
254 if (!atomic_dec_if_positive(&pool->events[offset].free)) {
256 return &pool->events[offset];
260 printk(KERN_ERR "ibmvscsi: found no event struct in pool!\n");
265 * init_event_struct: Initialize fields in an event struct that are always
268 * @done: Routine to call when the event is responded to
269 * @format: SRP or MAD format
270 * @timeout: timeout value set in the CRQ
272 static void init_event_struct(struct srp_event_struct *evt_struct,
273 void (*done) (struct srp_event_struct *),
277 evt_struct->cmnd = NULL;
278 evt_struct->cmnd_done = NULL;
279 evt_struct->sync_srp = NULL;
280 evt_struct->crq.format = format;
281 evt_struct->crq.timeout = timeout;
282 evt_struct->done = done;
285 /* ------------------------------------------------------------
286 * Routines for receiving SCSI responses from the hosting partition
290 * set_srp_direction: Set the fields in the srp related to data
291 * direction and number of buffers based on the direction in
292 * the scsi_cmnd and the number of buffers
294 static void set_srp_direction(struct scsi_cmnd *cmd,
295 struct srp_cmd *srp_cmd,
304 fmt = SRP_DATA_DESC_DIRECT;
306 fmt = SRP_DATA_DESC_INDIRECT;
307 numbuf = min(numbuf, MAX_INDIRECT_BUFS);
309 if (cmd->sc_data_direction == DMA_TO_DEVICE)
310 srp_cmd->data_out_desc_cnt = numbuf;
312 srp_cmd->data_in_desc_cnt = numbuf;
315 if (cmd->sc_data_direction == DMA_TO_DEVICE)
316 srp_cmd->buf_fmt = fmt << 4;
318 srp_cmd->buf_fmt = fmt;
322 * unmap_cmd_data: - Unmap data pointed in srp_cmd based on the format
323 * @cmd: srp_cmd whose additional_data member will be unmapped
324 * @dev: device for which the memory is mapped
327 static void unmap_cmd_data(struct srp_cmd *cmd,
328 struct srp_event_struct *evt_struct,
333 out_fmt = cmd->buf_fmt >> 4;
334 in_fmt = cmd->buf_fmt & ((1U << 4) - 1);
336 if (out_fmt == SRP_NO_DATA_DESC && in_fmt == SRP_NO_DATA_DESC)
339 if (evt_struct->cmnd)
340 scsi_dma_unmap(evt_struct->cmnd);
343 static int map_sg_list(struct scsi_cmnd *cmd, int nseg,
344 struct srp_direct_buf *md)
347 struct scatterlist *sg;
348 u64 total_length = 0;
350 scsi_for_each_sg(cmd, sg, nseg, i) {
351 struct srp_direct_buf *descr = md + i;
352 descr->va = sg_dma_address(sg);
353 descr->len = sg_dma_len(sg);
355 total_length += sg_dma_len(sg);
361 * map_sg_data: - Maps dma for a scatterlist and initializes decriptor fields
362 * @cmd: Scsi_Cmnd with the scatterlist
363 * @srp_cmd: srp_cmd that contains the memory descriptor
364 * @dev: device for which to map dma memory
366 * Called by map_data_for_srp_cmd() when building srp cmd from scsi cmd.
367 * Returns 1 on success.
369 static int map_sg_data(struct scsi_cmnd *cmd,
370 struct srp_event_struct *evt_struct,
371 struct srp_cmd *srp_cmd, struct device *dev)
375 u64 total_length = 0;
376 struct srp_direct_buf *data =
377 (struct srp_direct_buf *) srp_cmd->add_data;
378 struct srp_indirect_buf *indirect =
379 (struct srp_indirect_buf *) data;
381 sg_mapped = scsi_dma_map(cmd);
384 else if (sg_mapped < 0)
387 set_srp_direction(cmd, srp_cmd, sg_mapped);
389 /* special case; we can use a single direct descriptor */
390 if (sg_mapped == 1) {
391 map_sg_list(cmd, sg_mapped, data);
395 indirect->table_desc.va = 0;
396 indirect->table_desc.len = sg_mapped * sizeof(struct srp_direct_buf);
397 indirect->table_desc.key = 0;
399 if (sg_mapped <= MAX_INDIRECT_BUFS) {
400 total_length = map_sg_list(cmd, sg_mapped,
401 &indirect->desc_list[0]);
402 indirect->len = total_length;
406 /* get indirect table */
407 if (!evt_struct->ext_list) {
408 evt_struct->ext_list = (struct srp_direct_buf *)
409 dma_alloc_coherent(dev,
410 SG_ALL * sizeof(struct srp_direct_buf),
411 &evt_struct->ext_list_token, 0);
412 if (!evt_struct->ext_list) {
413 if (!firmware_has_feature(FW_FEATURE_CMO))
414 sdev_printk(KERN_ERR, cmd->device,
415 "Can't allocate memory "
416 "for indirect table\n");
422 total_length = map_sg_list(cmd, sg_mapped, evt_struct->ext_list);
424 indirect->len = total_length;
425 indirect->table_desc.va = evt_struct->ext_list_token;
426 indirect->table_desc.len = sg_mapped * sizeof(indirect->desc_list[0]);
427 memcpy(indirect->desc_list, evt_struct->ext_list,
428 MAX_INDIRECT_BUFS * sizeof(struct srp_direct_buf));
433 * map_data_for_srp_cmd: - Calls functions to map data for srp cmds
434 * @cmd: struct scsi_cmnd with the memory to be mapped
435 * @srp_cmd: srp_cmd that contains the memory descriptor
436 * @dev: dma device for which to map dma memory
438 * Called by scsi_cmd_to_srp_cmd() when converting scsi cmds to srp cmds
439 * Returns 1 on success.
441 static int map_data_for_srp_cmd(struct scsi_cmnd *cmd,
442 struct srp_event_struct *evt_struct,
443 struct srp_cmd *srp_cmd, struct device *dev)
445 switch (cmd->sc_data_direction) {
446 case DMA_FROM_DEVICE:
451 case DMA_BIDIRECTIONAL:
452 sdev_printk(KERN_ERR, cmd->device,
453 "Can't map DMA_BIDIRECTIONAL to read/write\n");
456 sdev_printk(KERN_ERR, cmd->device,
457 "Unknown data direction 0x%02x; can't map!\n",
458 cmd->sc_data_direction);
462 return map_sg_data(cmd, evt_struct, srp_cmd, dev);
466 * purge_requests: Our virtual adapter just shut down. purge any sent requests
467 * @hostdata: the adapter
469 static void purge_requests(struct ibmvscsi_host_data *hostdata, int error_code)
471 struct srp_event_struct *evt;
474 spin_lock_irqsave(hostdata->host->host_lock, flags);
475 while (!list_empty(&hostdata->sent)) {
476 evt = list_first_entry(&hostdata->sent, struct srp_event_struct, list);
477 list_del(&evt->list);
478 del_timer(&evt->timer);
480 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
482 evt->cmnd->result = (error_code << 16);
483 unmap_cmd_data(&evt->iu.srp.cmd, evt,
486 evt->cmnd_done(evt->cmnd);
487 } else if (evt->done)
489 free_event_struct(&evt->hostdata->pool, evt);
490 spin_lock_irqsave(hostdata->host->host_lock, flags);
492 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
496 * ibmvscsi_reset_host - Reset the connection to the server
497 * @hostdata: struct ibmvscsi_host_data to reset
499 static void ibmvscsi_reset_host(struct ibmvscsi_host_data *hostdata)
501 scsi_block_requests(hostdata->host);
502 atomic_set(&hostdata->request_limit, 0);
504 purge_requests(hostdata, DID_ERROR);
505 hostdata->reset_crq = 1;
506 wake_up(&hostdata->work_wait_q);
510 * ibmvscsi_timeout - Internal command timeout handler
511 * @evt_struct: struct srp_event_struct that timed out
513 * Called when an internally generated command times out
515 static void ibmvscsi_timeout(struct srp_event_struct *evt_struct)
517 struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
519 dev_err(hostdata->dev, "Command timed out (%x). Resetting connection\n",
520 evt_struct->iu.srp.cmd.opcode);
522 ibmvscsi_reset_host(hostdata);
526 /* ------------------------------------------------------------
527 * Routines for sending and receiving SRPs
530 * ibmvscsi_send_srp_event: - Transforms event to u64 array and calls send_crq()
531 * @evt_struct: evt_struct to be sent
532 * @hostdata: ibmvscsi_host_data of host
533 * @timeout: timeout in seconds - 0 means do not time command
535 * Returns the value returned from ibmvscsi_send_crq(). (Zero for success)
536 * Note that this routine assumes that host_lock is held for synchronization
538 static int ibmvscsi_send_srp_event(struct srp_event_struct *evt_struct,
539 struct ibmvscsi_host_data *hostdata,
540 unsigned long timeout)
542 u64 *crq_as_u64 = (u64 *) &evt_struct->crq;
543 int request_status = 0;
547 /* If we have exhausted our request limit, just fail this request,
548 * unless it is for a reset or abort.
549 * Note that there are rare cases involving driver generated requests
550 * (such as task management requests) that the mid layer may think we
551 * can handle more requests (can_queue) when we actually can't
553 if (evt_struct->crq.format == VIOSRP_SRP_FORMAT) {
556 atomic_dec_if_positive(&hostdata->request_limit);
557 /* If request limit was -1 when we started, it is now even
560 if (request_status < -1)
562 /* Otherwise, we may have run out of requests. */
563 /* If request limit was 0 when we started the adapter is in the
564 * process of performing a login with the server adapter, or
565 * we may have run out of requests.
567 else if (request_status == -1 &&
568 evt_struct->iu.srp.login_req.opcode != SRP_LOGIN_REQ)
570 /* Abort and reset calls should make it through.
571 * Nothing except abort and reset should use the last two
572 * slots unless we had two or less to begin with.
574 else if (request_status < 2 &&
575 evt_struct->iu.srp.cmd.opcode != SRP_TSK_MGMT) {
576 /* In the case that we have less than two requests
577 * available, check the server limit as a combination
578 * of the request limit and the number of requests
579 * in-flight (the size of the send list). If the
580 * server limit is greater than 2, return busy so
581 * that the last two are reserved for reset and abort.
583 int server_limit = request_status;
584 struct srp_event_struct *tmp_evt;
586 list_for_each_entry(tmp_evt, &hostdata->sent, list) {
590 if (server_limit > 2)
595 /* Copy the IU into the transfer area */
596 *evt_struct->xfer_iu = evt_struct->iu;
597 evt_struct->xfer_iu->srp.rsp.tag = (u64)evt_struct;
599 /* Add this to the sent list. We need to do this
600 * before we actually send
601 * in case it comes back REALLY fast
603 list_add_tail(&evt_struct->list, &hostdata->sent);
605 init_timer(&evt_struct->timer);
607 evt_struct->timer.data = (unsigned long) evt_struct;
608 evt_struct->timer.expires = jiffies + (timeout * HZ);
609 evt_struct->timer.function = (void (*)(unsigned long))ibmvscsi_timeout;
610 add_timer(&evt_struct->timer);
614 ibmvscsi_ops->send_crq(hostdata, crq_as_u64[0], crq_as_u64[1])) != 0) {
615 list_del(&evt_struct->list);
616 del_timer(&evt_struct->timer);
618 /* If send_crq returns H_CLOSED, return SCSI_MLQUEUE_HOST_BUSY.
619 * Firmware will send a CRQ with a transport event (0xFF) to
620 * tell this client what has happened to the transport. This
621 * will be handled in ibmvscsi_handle_crq()
623 if (rc == H_CLOSED) {
624 dev_warn(hostdata->dev, "send warning. "
625 "Receive queue closed, will retry.\n");
628 dev_err(hostdata->dev, "send error %d\n", rc);
630 atomic_inc(&hostdata->request_limit);
637 unmap_cmd_data(&evt_struct->iu.srp.cmd, evt_struct, hostdata->dev);
639 free_event_struct(&hostdata->pool, evt_struct);
640 if (srp_req && request_status != -1)
641 atomic_inc(&hostdata->request_limit);
642 return SCSI_MLQUEUE_HOST_BUSY;
645 unmap_cmd_data(&evt_struct->iu.srp.cmd, evt_struct, hostdata->dev);
647 if (evt_struct->cmnd != NULL) {
648 evt_struct->cmnd->result = DID_ERROR << 16;
649 evt_struct->cmnd_done(evt_struct->cmnd);
650 } else if (evt_struct->done)
651 evt_struct->done(evt_struct);
653 free_event_struct(&hostdata->pool, evt_struct);
658 * handle_cmd_rsp: - Handle responses from commands
659 * @evt_struct: srp_event_struct to be handled
661 * Used as a callback by when sending scsi cmds.
662 * Gets called by ibmvscsi_handle_crq()
664 static void handle_cmd_rsp(struct srp_event_struct *evt_struct)
666 struct srp_rsp *rsp = &evt_struct->xfer_iu->srp.rsp;
667 struct scsi_cmnd *cmnd = evt_struct->cmnd;
669 if (unlikely(rsp->opcode != SRP_RSP)) {
670 if (printk_ratelimit())
671 dev_warn(evt_struct->hostdata->dev,
672 "bad SRP RSP type %d\n", rsp->opcode);
676 cmnd->result |= rsp->status;
677 if (((cmnd->result >> 1) & 0x1f) == CHECK_CONDITION)
678 memcpy(cmnd->sense_buffer,
680 rsp->sense_data_len);
681 unmap_cmd_data(&evt_struct->iu.srp.cmd,
683 evt_struct->hostdata->dev);
685 if (rsp->flags & SRP_RSP_FLAG_DOOVER)
686 scsi_set_resid(cmnd, rsp->data_out_res_cnt);
687 else if (rsp->flags & SRP_RSP_FLAG_DIOVER)
688 scsi_set_resid(cmnd, rsp->data_in_res_cnt);
691 if (evt_struct->cmnd_done)
692 evt_struct->cmnd_done(cmnd);
696 * lun_from_dev: - Returns the lun of the scsi device
697 * @dev: struct scsi_device
700 static inline u16 lun_from_dev(struct scsi_device *dev)
702 return (0x2 << 14) | (dev->id << 8) | (dev->channel << 5) | dev->lun;
706 * ibmvscsi_queue: - The queuecommand function of the scsi template
707 * @cmd: struct scsi_cmnd to be executed
708 * @done: Callback function to be called when cmd is completed
710 static int ibmvscsi_queuecommand_lck(struct scsi_cmnd *cmnd,
711 void (*done) (struct scsi_cmnd *))
713 struct srp_cmd *srp_cmd;
714 struct srp_event_struct *evt_struct;
715 struct srp_indirect_buf *indirect;
716 struct ibmvscsi_host_data *hostdata = shost_priv(cmnd->device->host);
717 u16 lun = lun_from_dev(cmnd->device);
720 cmnd->result = (DID_OK << 16);
721 evt_struct = get_event_struct(&hostdata->pool);
723 return SCSI_MLQUEUE_HOST_BUSY;
725 /* Set up the actual SRP IU */
726 srp_cmd = &evt_struct->iu.srp.cmd;
727 memset(srp_cmd, 0x00, SRP_MAX_IU_LEN);
728 srp_cmd->opcode = SRP_CMD;
729 memcpy(srp_cmd->cdb, cmnd->cmnd, sizeof(srp_cmd->cdb));
730 srp_cmd->lun = ((u64) lun) << 48;
732 if (!map_data_for_srp_cmd(cmnd, evt_struct, srp_cmd, hostdata->dev)) {
733 if (!firmware_has_feature(FW_FEATURE_CMO))
734 sdev_printk(KERN_ERR, cmnd->device,
735 "couldn't convert cmd to srp_cmd\n");
736 free_event_struct(&hostdata->pool, evt_struct);
737 return SCSI_MLQUEUE_HOST_BUSY;
740 init_event_struct(evt_struct,
743 cmnd->request->timeout/HZ);
745 evt_struct->cmnd = cmnd;
746 evt_struct->cmnd_done = done;
748 /* Fix up dma address of the buffer itself */
749 indirect = (struct srp_indirect_buf *) srp_cmd->add_data;
750 out_fmt = srp_cmd->buf_fmt >> 4;
751 in_fmt = srp_cmd->buf_fmt & ((1U << 4) - 1);
752 if ((in_fmt == SRP_DATA_DESC_INDIRECT ||
753 out_fmt == SRP_DATA_DESC_INDIRECT) &&
754 indirect->table_desc.va == 0) {
755 indirect->table_desc.va = evt_struct->crq.IU_data_ptr +
756 offsetof(struct srp_cmd, add_data) +
757 offsetof(struct srp_indirect_buf, desc_list);
760 return ibmvscsi_send_srp_event(evt_struct, hostdata, 0);
763 static DEF_SCSI_QCMD(ibmvscsi_queuecommand)
765 /* ------------------------------------------------------------
766 * Routines for driver initialization
770 * map_persist_bufs: - Pre-map persistent data for adapter logins
771 * @hostdata: ibmvscsi_host_data of host
773 * Map the capabilities and adapter info DMA buffers to avoid runtime failures.
774 * Return 1 on error, 0 on success.
776 static int map_persist_bufs(struct ibmvscsi_host_data *hostdata)
779 hostdata->caps_addr = dma_map_single(hostdata->dev, &hostdata->caps,
780 sizeof(hostdata->caps), DMA_BIDIRECTIONAL);
782 if (dma_mapping_error(hostdata->dev, hostdata->caps_addr)) {
783 dev_err(hostdata->dev, "Unable to map capabilities buffer!\n");
787 hostdata->adapter_info_addr = dma_map_single(hostdata->dev,
788 &hostdata->madapter_info,
789 sizeof(hostdata->madapter_info),
791 if (dma_mapping_error(hostdata->dev, hostdata->adapter_info_addr)) {
792 dev_err(hostdata->dev, "Unable to map adapter info buffer!\n");
793 dma_unmap_single(hostdata->dev, hostdata->caps_addr,
794 sizeof(hostdata->caps), DMA_BIDIRECTIONAL);
802 * unmap_persist_bufs: - Unmap persistent data needed for adapter logins
803 * @hostdata: ibmvscsi_host_data of host
805 * Unmap the capabilities and adapter info DMA buffers
807 static void unmap_persist_bufs(struct ibmvscsi_host_data *hostdata)
809 dma_unmap_single(hostdata->dev, hostdata->caps_addr,
810 sizeof(hostdata->caps), DMA_BIDIRECTIONAL);
812 dma_unmap_single(hostdata->dev, hostdata->adapter_info_addr,
813 sizeof(hostdata->madapter_info), DMA_BIDIRECTIONAL);
817 * login_rsp: - Handle response to SRP login request
818 * @evt_struct: srp_event_struct with the response
820 * Used as a "done" callback by when sending srp_login. Gets called
821 * by ibmvscsi_handle_crq()
823 static void login_rsp(struct srp_event_struct *evt_struct)
825 struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
826 switch (evt_struct->xfer_iu->srp.login_rsp.opcode) {
827 case SRP_LOGIN_RSP: /* it worked! */
829 case SRP_LOGIN_REJ: /* refused! */
830 dev_info(hostdata->dev, "SRP_LOGIN_REJ reason %u\n",
831 evt_struct->xfer_iu->srp.login_rej.reason);
833 atomic_set(&hostdata->request_limit, -1);
836 dev_err(hostdata->dev, "Invalid login response typecode 0x%02x!\n",
837 evt_struct->xfer_iu->srp.login_rsp.opcode);
839 atomic_set(&hostdata->request_limit, -1);
843 dev_info(hostdata->dev, "SRP_LOGIN succeeded\n");
844 hostdata->client_migrated = 0;
846 /* Now we know what the real request-limit is.
847 * This value is set rather than added to request_limit because
848 * request_limit could have been set to -1 by this client.
850 atomic_set(&hostdata->request_limit,
851 evt_struct->xfer_iu->srp.login_rsp.req_lim_delta);
853 /* If we had any pending I/Os, kick them */
854 scsi_unblock_requests(hostdata->host);
858 * send_srp_login: - Sends the srp login
859 * @hostdata: ibmvscsi_host_data of host
861 * Returns zero if successful.
863 static int send_srp_login(struct ibmvscsi_host_data *hostdata)
867 struct srp_login_req *login;
868 struct srp_event_struct *evt_struct = get_event_struct(&hostdata->pool);
871 init_event_struct(evt_struct, login_rsp,
872 VIOSRP_SRP_FORMAT, login_timeout);
874 login = &evt_struct->iu.srp.login_req;
875 memset(login, 0, sizeof(*login));
876 login->opcode = SRP_LOGIN_REQ;
877 login->req_it_iu_len = sizeof(union srp_iu);
878 login->req_buf_fmt = SRP_BUF_FORMAT_DIRECT | SRP_BUF_FORMAT_INDIRECT;
880 spin_lock_irqsave(hostdata->host->host_lock, flags);
881 /* Start out with a request limit of 0, since this is negotiated in
882 * the login request we are just sending and login requests always
883 * get sent by the driver regardless of request_limit.
885 atomic_set(&hostdata->request_limit, 0);
887 rc = ibmvscsi_send_srp_event(evt_struct, hostdata, login_timeout * 2);
888 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
889 dev_info(hostdata->dev, "sent SRP login\n");
894 * capabilities_rsp: - Handle response to MAD adapter capabilities request
895 * @evt_struct: srp_event_struct with the response
897 * Used as a "done" callback by when sending adapter_info.
899 static void capabilities_rsp(struct srp_event_struct *evt_struct)
901 struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
903 if (evt_struct->xfer_iu->mad.capabilities.common.status) {
904 dev_err(hostdata->dev, "error 0x%X getting capabilities info\n",
905 evt_struct->xfer_iu->mad.capabilities.common.status);
907 if (hostdata->caps.migration.common.server_support != SERVER_SUPPORTS_CAP)
908 dev_info(hostdata->dev, "Partition migration not supported\n");
910 if (client_reserve) {
911 if (hostdata->caps.reserve.common.server_support ==
913 dev_info(hostdata->dev, "Client reserve enabled\n");
915 dev_info(hostdata->dev, "Client reserve not supported\n");
919 send_srp_login(hostdata);
923 * send_mad_capabilities: - Sends the mad capabilities request
924 * and stores the result so it can be retrieved with
925 * @hostdata: ibmvscsi_host_data of host
927 static void send_mad_capabilities(struct ibmvscsi_host_data *hostdata)
929 struct viosrp_capabilities *req;
930 struct srp_event_struct *evt_struct;
932 struct device_node *of_node = hostdata->dev->of_node;
933 const char *location;
935 evt_struct = get_event_struct(&hostdata->pool);
938 init_event_struct(evt_struct, capabilities_rsp,
939 VIOSRP_MAD_FORMAT, info_timeout);
941 req = &evt_struct->iu.mad.capabilities;
942 memset(req, 0, sizeof(*req));
944 hostdata->caps.flags = CAP_LIST_SUPPORTED;
945 if (hostdata->client_migrated)
946 hostdata->caps.flags |= CLIENT_MIGRATED;
948 strncpy(hostdata->caps.name, dev_name(&hostdata->host->shost_gendev),
949 sizeof(hostdata->caps.name));
950 hostdata->caps.name[sizeof(hostdata->caps.name) - 1] = '\0';
952 location = of_get_property(of_node, "ibm,loc-code", NULL);
953 location = location ? location : dev_name(hostdata->dev);
954 strncpy(hostdata->caps.loc, location, sizeof(hostdata->caps.loc));
955 hostdata->caps.loc[sizeof(hostdata->caps.loc) - 1] = '\0';
957 req->common.type = VIOSRP_CAPABILITIES_TYPE;
958 req->buffer = hostdata->caps_addr;
960 hostdata->caps.migration.common.cap_type = MIGRATION_CAPABILITIES;
961 hostdata->caps.migration.common.length = sizeof(hostdata->caps.migration);
962 hostdata->caps.migration.common.server_support = SERVER_SUPPORTS_CAP;
963 hostdata->caps.migration.ecl = 1;
965 if (client_reserve) {
966 hostdata->caps.reserve.common.cap_type = RESERVATION_CAPABILITIES;
967 hostdata->caps.reserve.common.length = sizeof(hostdata->caps.reserve);
968 hostdata->caps.reserve.common.server_support = SERVER_SUPPORTS_CAP;
969 hostdata->caps.reserve.type = CLIENT_RESERVE_SCSI_2;
970 req->common.length = sizeof(hostdata->caps);
972 req->common.length = sizeof(hostdata->caps) - sizeof(hostdata->caps.reserve);
974 spin_lock_irqsave(hostdata->host->host_lock, flags);
975 if (ibmvscsi_send_srp_event(evt_struct, hostdata, info_timeout * 2))
976 dev_err(hostdata->dev, "couldn't send CAPABILITIES_REQ!\n");
977 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
981 * fast_fail_rsp: - Handle response to MAD enable fast fail
982 * @evt_struct: srp_event_struct with the response
984 * Used as a "done" callback by when sending enable fast fail. Gets called
985 * by ibmvscsi_handle_crq()
987 static void fast_fail_rsp(struct srp_event_struct *evt_struct)
989 struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
990 u8 status = evt_struct->xfer_iu->mad.fast_fail.common.status;
992 if (status == VIOSRP_MAD_NOT_SUPPORTED)
993 dev_err(hostdata->dev, "fast_fail not supported in server\n");
994 else if (status == VIOSRP_MAD_FAILED)
995 dev_err(hostdata->dev, "fast_fail request failed\n");
996 else if (status != VIOSRP_MAD_SUCCESS)
997 dev_err(hostdata->dev, "error 0x%X enabling fast_fail\n", status);
999 send_mad_capabilities(hostdata);
1003 * init_host - Start host initialization
1004 * @hostdata: ibmvscsi_host_data of host
1006 * Returns zero if successful.
1008 static int enable_fast_fail(struct ibmvscsi_host_data *hostdata)
1011 unsigned long flags;
1012 struct viosrp_fast_fail *fast_fail_mad;
1013 struct srp_event_struct *evt_struct;
1016 send_mad_capabilities(hostdata);
1020 evt_struct = get_event_struct(&hostdata->pool);
1021 BUG_ON(!evt_struct);
1023 init_event_struct(evt_struct, fast_fail_rsp, VIOSRP_MAD_FORMAT, info_timeout);
1025 fast_fail_mad = &evt_struct->iu.mad.fast_fail;
1026 memset(fast_fail_mad, 0, sizeof(*fast_fail_mad));
1027 fast_fail_mad->common.type = VIOSRP_ENABLE_FAST_FAIL;
1028 fast_fail_mad->common.length = sizeof(*fast_fail_mad);
1030 spin_lock_irqsave(hostdata->host->host_lock, flags);
1031 rc = ibmvscsi_send_srp_event(evt_struct, hostdata, info_timeout * 2);
1032 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1037 * adapter_info_rsp: - Handle response to MAD adapter info request
1038 * @evt_struct: srp_event_struct with the response
1040 * Used as a "done" callback by when sending adapter_info. Gets called
1041 * by ibmvscsi_handle_crq()
1043 static void adapter_info_rsp(struct srp_event_struct *evt_struct)
1045 struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
1047 if (evt_struct->xfer_iu->mad.adapter_info.common.status) {
1048 dev_err(hostdata->dev, "error %d getting adapter info\n",
1049 evt_struct->xfer_iu->mad.adapter_info.common.status);
1051 dev_info(hostdata->dev, "host srp version: %s, "
1052 "host partition %s (%d), OS %d, max io %u\n",
1053 hostdata->madapter_info.srp_version,
1054 hostdata->madapter_info.partition_name,
1055 hostdata->madapter_info.partition_number,
1056 hostdata->madapter_info.os_type,
1057 hostdata->madapter_info.port_max_txu[0]);
1059 if (hostdata->madapter_info.port_max_txu[0])
1060 hostdata->host->max_sectors =
1061 hostdata->madapter_info.port_max_txu[0] >> 9;
1063 if (hostdata->madapter_info.os_type == 3 &&
1064 strcmp(hostdata->madapter_info.srp_version, "1.6a") <= 0) {
1065 dev_err(hostdata->dev, "host (Ver. %s) doesn't support large transfers\n",
1066 hostdata->madapter_info.srp_version);
1067 dev_err(hostdata->dev, "limiting scatterlists to %d\n",
1069 hostdata->host->sg_tablesize = MAX_INDIRECT_BUFS;
1072 if (hostdata->madapter_info.os_type == 3) {
1073 enable_fast_fail(hostdata);
1078 send_srp_login(hostdata);
1082 * send_mad_adapter_info: - Sends the mad adapter info request
1083 * and stores the result so it can be retrieved with
1084 * sysfs. We COULD consider causing a failure if the
1085 * returned SRP version doesn't match ours.
1086 * @hostdata: ibmvscsi_host_data of host
1088 * Returns zero if successful.
1090 static void send_mad_adapter_info(struct ibmvscsi_host_data *hostdata)
1092 struct viosrp_adapter_info *req;
1093 struct srp_event_struct *evt_struct;
1094 unsigned long flags;
1096 evt_struct = get_event_struct(&hostdata->pool);
1097 BUG_ON(!evt_struct);
1099 init_event_struct(evt_struct,
1104 req = &evt_struct->iu.mad.adapter_info;
1105 memset(req, 0x00, sizeof(*req));
1107 req->common.type = VIOSRP_ADAPTER_INFO_TYPE;
1108 req->common.length = sizeof(hostdata->madapter_info);
1109 req->buffer = hostdata->adapter_info_addr;
1111 spin_lock_irqsave(hostdata->host->host_lock, flags);
1112 if (ibmvscsi_send_srp_event(evt_struct, hostdata, info_timeout * 2))
1113 dev_err(hostdata->dev, "couldn't send ADAPTER_INFO_REQ!\n");
1114 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1118 * init_adapter: Start virtual adapter initialization sequence
1121 static void init_adapter(struct ibmvscsi_host_data *hostdata)
1123 send_mad_adapter_info(hostdata);
1127 * sync_completion: Signal that a synchronous command has completed
1128 * Note that after returning from this call, the evt_struct is freed.
1129 * the caller waiting on this completion shouldn't touch the evt_struct
1132 static void sync_completion(struct srp_event_struct *evt_struct)
1134 /* copy the response back */
1135 if (evt_struct->sync_srp)
1136 *evt_struct->sync_srp = *evt_struct->xfer_iu;
1138 complete(&evt_struct->comp);
1142 * ibmvscsi_abort: Abort a command...from scsi host template
1143 * send this over to the server and wait synchronously for the response
1145 static int ibmvscsi_eh_abort_handler(struct scsi_cmnd *cmd)
1147 struct ibmvscsi_host_data *hostdata = shost_priv(cmd->device->host);
1148 struct srp_tsk_mgmt *tsk_mgmt;
1149 struct srp_event_struct *evt;
1150 struct srp_event_struct *tmp_evt, *found_evt;
1151 union viosrp_iu srp_rsp;
1153 unsigned long flags;
1154 u16 lun = lun_from_dev(cmd->device);
1155 unsigned long wait_switch = 0;
1157 /* First, find this command in our sent list so we can figure
1158 * out the correct tag
1160 spin_lock_irqsave(hostdata->host->host_lock, flags);
1161 wait_switch = jiffies + (init_timeout * HZ);
1164 list_for_each_entry(tmp_evt, &hostdata->sent, list) {
1165 if (tmp_evt->cmnd == cmd) {
1166 found_evt = tmp_evt;
1172 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1176 evt = get_event_struct(&hostdata->pool);
1178 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1179 sdev_printk(KERN_ERR, cmd->device,
1180 "failed to allocate abort event\n");
1184 init_event_struct(evt,
1189 tsk_mgmt = &evt->iu.srp.tsk_mgmt;
1191 /* Set up an abort SRP command */
1192 memset(tsk_mgmt, 0x00, sizeof(*tsk_mgmt));
1193 tsk_mgmt->opcode = SRP_TSK_MGMT;
1194 tsk_mgmt->lun = ((u64) lun) << 48;
1195 tsk_mgmt->tsk_mgmt_func = SRP_TSK_ABORT_TASK;
1196 tsk_mgmt->task_tag = (u64) found_evt;
1198 evt->sync_srp = &srp_rsp;
1200 init_completion(&evt->comp);
1201 rsp_rc = ibmvscsi_send_srp_event(evt, hostdata, abort_timeout * 2);
1203 if (rsp_rc != SCSI_MLQUEUE_HOST_BUSY)
1206 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1208 spin_lock_irqsave(hostdata->host->host_lock, flags);
1209 } while (time_before(jiffies, wait_switch));
1211 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1214 sdev_printk(KERN_ERR, cmd->device,
1215 "failed to send abort() event. rc=%d\n", rsp_rc);
1219 sdev_printk(KERN_INFO, cmd->device,
1220 "aborting command. lun 0x%llx, tag 0x%llx\n",
1221 (((u64) lun) << 48), (u64) found_evt);
1223 wait_for_completion(&evt->comp);
1225 /* make sure we got a good response */
1226 if (unlikely(srp_rsp.srp.rsp.opcode != SRP_RSP)) {
1227 if (printk_ratelimit())
1228 sdev_printk(KERN_WARNING, cmd->device, "abort bad SRP RSP type %d\n",
1229 srp_rsp.srp.rsp.opcode);
1233 if (srp_rsp.srp.rsp.flags & SRP_RSP_FLAG_RSPVALID)
1234 rsp_rc = *((int *)srp_rsp.srp.rsp.data);
1236 rsp_rc = srp_rsp.srp.rsp.status;
1239 if (printk_ratelimit())
1240 sdev_printk(KERN_WARNING, cmd->device,
1241 "abort code %d for task tag 0x%llx\n",
1242 rsp_rc, tsk_mgmt->task_tag);
1246 /* Because we dropped the spinlock above, it's possible
1247 * The event is no longer in our list. Make sure it didn't
1248 * complete while we were aborting
1250 spin_lock_irqsave(hostdata->host->host_lock, flags);
1252 list_for_each_entry(tmp_evt, &hostdata->sent, list) {
1253 if (tmp_evt->cmnd == cmd) {
1254 found_evt = tmp_evt;
1259 if (found_evt == NULL) {
1260 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1261 sdev_printk(KERN_INFO, cmd->device, "aborted task tag 0x%llx completed\n",
1262 tsk_mgmt->task_tag);
1266 sdev_printk(KERN_INFO, cmd->device, "successfully aborted task tag 0x%llx\n",
1267 tsk_mgmt->task_tag);
1269 cmd->result = (DID_ABORT << 16);
1270 list_del(&found_evt->list);
1271 unmap_cmd_data(&found_evt->iu.srp.cmd, found_evt,
1272 found_evt->hostdata->dev);
1273 free_event_struct(&found_evt->hostdata->pool, found_evt);
1274 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1275 atomic_inc(&hostdata->request_limit);
1280 * ibmvscsi_eh_device_reset_handler: Reset a single LUN...from scsi host
1281 * template send this over to the server and wait synchronously for the
1284 static int ibmvscsi_eh_device_reset_handler(struct scsi_cmnd *cmd)
1286 struct ibmvscsi_host_data *hostdata = shost_priv(cmd->device->host);
1287 struct srp_tsk_mgmt *tsk_mgmt;
1288 struct srp_event_struct *evt;
1289 struct srp_event_struct *tmp_evt, *pos;
1290 union viosrp_iu srp_rsp;
1292 unsigned long flags;
1293 u16 lun = lun_from_dev(cmd->device);
1294 unsigned long wait_switch = 0;
1296 spin_lock_irqsave(hostdata->host->host_lock, flags);
1297 wait_switch = jiffies + (init_timeout * HZ);
1299 evt = get_event_struct(&hostdata->pool);
1301 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1302 sdev_printk(KERN_ERR, cmd->device,
1303 "failed to allocate reset event\n");
1307 init_event_struct(evt,
1312 tsk_mgmt = &evt->iu.srp.tsk_mgmt;
1314 /* Set up a lun reset SRP command */
1315 memset(tsk_mgmt, 0x00, sizeof(*tsk_mgmt));
1316 tsk_mgmt->opcode = SRP_TSK_MGMT;
1317 tsk_mgmt->lun = ((u64) lun) << 48;
1318 tsk_mgmt->tsk_mgmt_func = SRP_TSK_LUN_RESET;
1320 evt->sync_srp = &srp_rsp;
1322 init_completion(&evt->comp);
1323 rsp_rc = ibmvscsi_send_srp_event(evt, hostdata, reset_timeout * 2);
1325 if (rsp_rc != SCSI_MLQUEUE_HOST_BUSY)
1328 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1330 spin_lock_irqsave(hostdata->host->host_lock, flags);
1331 } while (time_before(jiffies, wait_switch));
1333 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1336 sdev_printk(KERN_ERR, cmd->device,
1337 "failed to send reset event. rc=%d\n", rsp_rc);
1341 sdev_printk(KERN_INFO, cmd->device, "resetting device. lun 0x%llx\n",
1342 (((u64) lun) << 48));
1344 wait_for_completion(&evt->comp);
1346 /* make sure we got a good response */
1347 if (unlikely(srp_rsp.srp.rsp.opcode != SRP_RSP)) {
1348 if (printk_ratelimit())
1349 sdev_printk(KERN_WARNING, cmd->device, "reset bad SRP RSP type %d\n",
1350 srp_rsp.srp.rsp.opcode);
1354 if (srp_rsp.srp.rsp.flags & SRP_RSP_FLAG_RSPVALID)
1355 rsp_rc = *((int *)srp_rsp.srp.rsp.data);
1357 rsp_rc = srp_rsp.srp.rsp.status;
1360 if (printk_ratelimit())
1361 sdev_printk(KERN_WARNING, cmd->device,
1362 "reset code %d for task tag 0x%llx\n",
1363 rsp_rc, tsk_mgmt->task_tag);
1367 /* We need to find all commands for this LUN that have not yet been
1368 * responded to, and fail them with DID_RESET
1370 spin_lock_irqsave(hostdata->host->host_lock, flags);
1371 list_for_each_entry_safe(tmp_evt, pos, &hostdata->sent, list) {
1372 if ((tmp_evt->cmnd) && (tmp_evt->cmnd->device == cmd->device)) {
1374 tmp_evt->cmnd->result = (DID_RESET << 16);
1375 list_del(&tmp_evt->list);
1376 unmap_cmd_data(&tmp_evt->iu.srp.cmd, tmp_evt,
1377 tmp_evt->hostdata->dev);
1378 free_event_struct(&tmp_evt->hostdata->pool,
1380 atomic_inc(&hostdata->request_limit);
1381 if (tmp_evt->cmnd_done)
1382 tmp_evt->cmnd_done(tmp_evt->cmnd);
1383 else if (tmp_evt->done)
1384 tmp_evt->done(tmp_evt);
1387 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1392 * ibmvscsi_eh_host_reset_handler - Reset the connection to the server
1393 * @cmd: struct scsi_cmnd having problems
1395 static int ibmvscsi_eh_host_reset_handler(struct scsi_cmnd *cmd)
1397 unsigned long wait_switch = 0;
1398 struct ibmvscsi_host_data *hostdata = shost_priv(cmd->device->host);
1400 dev_err(hostdata->dev, "Resetting connection due to error recovery\n");
1402 ibmvscsi_reset_host(hostdata);
1404 for (wait_switch = jiffies + (init_timeout * HZ);
1405 time_before(jiffies, wait_switch) &&
1406 atomic_read(&hostdata->request_limit) < 2;) {
1411 if (atomic_read(&hostdata->request_limit) <= 0)
1418 * ibmvscsi_handle_crq: - Handles and frees received events in the CRQ
1419 * @crq: Command/Response queue
1420 * @hostdata: ibmvscsi_host_data of host
1423 void ibmvscsi_handle_crq(struct viosrp_crq *crq,
1424 struct ibmvscsi_host_data *hostdata)
1427 unsigned long flags;
1428 struct srp_event_struct *evt_struct =
1429 (struct srp_event_struct *)crq->IU_data_ptr;
1430 switch (crq->valid) {
1431 case 0xC0: /* initialization */
1432 switch (crq->format) {
1433 case 0x01: /* Initialization message */
1434 dev_info(hostdata->dev, "partner initialized\n");
1435 /* Send back a response */
1436 if ((rc = ibmvscsi_ops->send_crq(hostdata,
1437 0xC002000000000000LL, 0)) == 0) {
1439 init_adapter(hostdata);
1441 dev_err(hostdata->dev, "Unable to send init rsp. rc=%ld\n", rc);
1445 case 0x02: /* Initialization response */
1446 dev_info(hostdata->dev, "partner initialization complete\n");
1449 init_adapter(hostdata);
1452 dev_err(hostdata->dev, "unknown crq message type: %d\n", crq->format);
1455 case 0xFF: /* Hypervisor telling us the connection is closed */
1456 scsi_block_requests(hostdata->host);
1457 atomic_set(&hostdata->request_limit, 0);
1458 if (crq->format == 0x06) {
1459 /* We need to re-setup the interpartition connection */
1460 dev_info(hostdata->dev, "Re-enabling adapter!\n");
1461 hostdata->client_migrated = 1;
1462 hostdata->reenable_crq = 1;
1463 purge_requests(hostdata, DID_REQUEUE);
1464 wake_up(&hostdata->work_wait_q);
1466 dev_err(hostdata->dev, "Virtual adapter failed rc %d!\n",
1468 ibmvscsi_reset_host(hostdata);
1471 case 0x80: /* real payload */
1474 dev_err(hostdata->dev, "got an invalid message type 0x%02x\n",
1479 /* The only kind of payload CRQs we should get are responses to
1480 * things we send. Make sure this response is to something we
1483 if (!valid_event_struct(&hostdata->pool, evt_struct)) {
1484 dev_err(hostdata->dev, "returned correlation_token 0x%p is invalid!\n",
1485 (void *)crq->IU_data_ptr);
1489 if (atomic_read(&evt_struct->free)) {
1490 dev_err(hostdata->dev, "received duplicate correlation_token 0x%p!\n",
1491 (void *)crq->IU_data_ptr);
1495 if (crq->format == VIOSRP_SRP_FORMAT)
1496 atomic_add(evt_struct->xfer_iu->srp.rsp.req_lim_delta,
1497 &hostdata->request_limit);
1499 del_timer(&evt_struct->timer);
1501 if ((crq->status != VIOSRP_OK && crq->status != VIOSRP_OK2) && evt_struct->cmnd)
1502 evt_struct->cmnd->result = DID_ERROR << 16;
1503 if (evt_struct->done)
1504 evt_struct->done(evt_struct);
1506 dev_err(hostdata->dev, "returned done() is NULL; not running it!\n");
1509 * Lock the host_lock before messing with these structures, since we
1510 * are running in a task context
1512 spin_lock_irqsave(evt_struct->hostdata->host->host_lock, flags);
1513 list_del(&evt_struct->list);
1514 free_event_struct(&evt_struct->hostdata->pool, evt_struct);
1515 spin_unlock_irqrestore(evt_struct->hostdata->host->host_lock, flags);
1519 * ibmvscsi_get_host_config: Send the command to the server to get host
1520 * configuration data. The data is opaque to us.
1522 static int ibmvscsi_do_host_config(struct ibmvscsi_host_data *hostdata,
1523 unsigned char *buffer, int length)
1525 struct viosrp_host_config *host_config;
1526 struct srp_event_struct *evt_struct;
1527 unsigned long flags;
1531 evt_struct = get_event_struct(&hostdata->pool);
1533 dev_err(hostdata->dev, "couldn't allocate event for HOST_CONFIG!\n");
1537 init_event_struct(evt_struct,
1542 host_config = &evt_struct->iu.mad.host_config;
1544 /* Set up a lun reset SRP command */
1545 memset(host_config, 0x00, sizeof(*host_config));
1546 host_config->common.type = VIOSRP_HOST_CONFIG_TYPE;
1547 host_config->common.length = length;
1548 host_config->buffer = addr = dma_map_single(hostdata->dev, buffer,
1552 if (dma_mapping_error(hostdata->dev, host_config->buffer)) {
1553 if (!firmware_has_feature(FW_FEATURE_CMO))
1554 dev_err(hostdata->dev,
1555 "dma_mapping error getting host config\n");
1556 free_event_struct(&hostdata->pool, evt_struct);
1560 init_completion(&evt_struct->comp);
1561 spin_lock_irqsave(hostdata->host->host_lock, flags);
1562 rc = ibmvscsi_send_srp_event(evt_struct, hostdata, info_timeout * 2);
1563 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1565 wait_for_completion(&evt_struct->comp);
1566 dma_unmap_single(hostdata->dev, addr, length, DMA_BIDIRECTIONAL);
1572 * ibmvscsi_slave_configure: Set the "allow_restart" flag for each disk.
1573 * @sdev: struct scsi_device device to configure
1575 * Enable allow_restart for a device if it is a disk. Adjust the
1576 * queue_depth here also as is required by the documentation for
1577 * struct scsi_host_template.
1579 static int ibmvscsi_slave_configure(struct scsi_device *sdev)
1581 struct Scsi_Host *shost = sdev->host;
1582 unsigned long lock_flags = 0;
1584 spin_lock_irqsave(shost->host_lock, lock_flags);
1585 if (sdev->type == TYPE_DISK) {
1586 sdev->allow_restart = 1;
1587 blk_queue_rq_timeout(sdev->request_queue, 120 * HZ);
1589 scsi_adjust_queue_depth(sdev, 0, shost->cmd_per_lun);
1590 spin_unlock_irqrestore(shost->host_lock, lock_flags);
1595 * ibmvscsi_change_queue_depth - Change the device's queue depth
1596 * @sdev: scsi device struct
1597 * @qdepth: depth to set
1598 * @reason: calling context
1603 static int ibmvscsi_change_queue_depth(struct scsi_device *sdev, int qdepth,
1606 if (reason != SCSI_QDEPTH_DEFAULT)
1609 if (qdepth > IBMVSCSI_MAX_CMDS_PER_LUN)
1610 qdepth = IBMVSCSI_MAX_CMDS_PER_LUN;
1612 scsi_adjust_queue_depth(sdev, 0, qdepth);
1613 return sdev->queue_depth;
1616 /* ------------------------------------------------------------
1619 static ssize_t show_host_vhost_loc(struct device *dev,
1620 struct device_attribute *attr, char *buf)
1622 struct Scsi_Host *shost = class_to_shost(dev);
1623 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1626 len = snprintf(buf, sizeof(hostdata->caps.loc), "%s\n",
1627 hostdata->caps.loc);
1631 static struct device_attribute ibmvscsi_host_vhost_loc = {
1633 .name = "vhost_loc",
1636 .show = show_host_vhost_loc,
1639 static ssize_t show_host_vhost_name(struct device *dev,
1640 struct device_attribute *attr, char *buf)
1642 struct Scsi_Host *shost = class_to_shost(dev);
1643 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1646 len = snprintf(buf, sizeof(hostdata->caps.name), "%s\n",
1647 hostdata->caps.name);
1651 static struct device_attribute ibmvscsi_host_vhost_name = {
1653 .name = "vhost_name",
1656 .show = show_host_vhost_name,
1659 static ssize_t show_host_srp_version(struct device *dev,
1660 struct device_attribute *attr, char *buf)
1662 struct Scsi_Host *shost = class_to_shost(dev);
1663 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1666 len = snprintf(buf, PAGE_SIZE, "%s\n",
1667 hostdata->madapter_info.srp_version);
1671 static struct device_attribute ibmvscsi_host_srp_version = {
1673 .name = "srp_version",
1676 .show = show_host_srp_version,
1679 static ssize_t show_host_partition_name(struct device *dev,
1680 struct device_attribute *attr,
1683 struct Scsi_Host *shost = class_to_shost(dev);
1684 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1687 len = snprintf(buf, PAGE_SIZE, "%s\n",
1688 hostdata->madapter_info.partition_name);
1692 static struct device_attribute ibmvscsi_host_partition_name = {
1694 .name = "partition_name",
1697 .show = show_host_partition_name,
1700 static ssize_t show_host_partition_number(struct device *dev,
1701 struct device_attribute *attr,
1704 struct Scsi_Host *shost = class_to_shost(dev);
1705 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1708 len = snprintf(buf, PAGE_SIZE, "%d\n",
1709 hostdata->madapter_info.partition_number);
1713 static struct device_attribute ibmvscsi_host_partition_number = {
1715 .name = "partition_number",
1718 .show = show_host_partition_number,
1721 static ssize_t show_host_mad_version(struct device *dev,
1722 struct device_attribute *attr, char *buf)
1724 struct Scsi_Host *shost = class_to_shost(dev);
1725 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1728 len = snprintf(buf, PAGE_SIZE, "%d\n",
1729 hostdata->madapter_info.mad_version);
1733 static struct device_attribute ibmvscsi_host_mad_version = {
1735 .name = "mad_version",
1738 .show = show_host_mad_version,
1741 static ssize_t show_host_os_type(struct device *dev,
1742 struct device_attribute *attr, char *buf)
1744 struct Scsi_Host *shost = class_to_shost(dev);
1745 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1748 len = snprintf(buf, PAGE_SIZE, "%d\n", hostdata->madapter_info.os_type);
1752 static struct device_attribute ibmvscsi_host_os_type = {
1757 .show = show_host_os_type,
1760 static ssize_t show_host_config(struct device *dev,
1761 struct device_attribute *attr, char *buf)
1763 struct Scsi_Host *shost = class_to_shost(dev);
1764 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1766 /* returns null-terminated host config data */
1767 if (ibmvscsi_do_host_config(hostdata, buf, PAGE_SIZE) == 0)
1773 static struct device_attribute ibmvscsi_host_config = {
1778 .show = show_host_config,
1781 static struct device_attribute *ibmvscsi_attrs[] = {
1782 &ibmvscsi_host_vhost_loc,
1783 &ibmvscsi_host_vhost_name,
1784 &ibmvscsi_host_srp_version,
1785 &ibmvscsi_host_partition_name,
1786 &ibmvscsi_host_partition_number,
1787 &ibmvscsi_host_mad_version,
1788 &ibmvscsi_host_os_type,
1789 &ibmvscsi_host_config,
1793 /* ------------------------------------------------------------
1794 * SCSI driver registration
1796 static struct scsi_host_template driver_template = {
1797 .module = THIS_MODULE,
1798 .name = "IBM POWER Virtual SCSI Adapter " IBMVSCSI_VERSION,
1799 .proc_name = "ibmvscsi",
1800 .queuecommand = ibmvscsi_queuecommand,
1801 .eh_abort_handler = ibmvscsi_eh_abort_handler,
1802 .eh_device_reset_handler = ibmvscsi_eh_device_reset_handler,
1803 .eh_host_reset_handler = ibmvscsi_eh_host_reset_handler,
1804 .slave_configure = ibmvscsi_slave_configure,
1805 .change_queue_depth = ibmvscsi_change_queue_depth,
1806 .cmd_per_lun = IBMVSCSI_CMDS_PER_LUN_DEFAULT,
1807 .can_queue = IBMVSCSI_MAX_REQUESTS_DEFAULT,
1809 .sg_tablesize = SG_ALL,
1810 .use_clustering = ENABLE_CLUSTERING,
1811 .shost_attrs = ibmvscsi_attrs,
1815 * ibmvscsi_get_desired_dma - Calculate IO memory desired by the driver
1817 * @vdev: struct vio_dev for the device whose desired IO mem is to be returned
1820 * Number of bytes of IO data the driver will need to perform well.
1822 static unsigned long ibmvscsi_get_desired_dma(struct vio_dev *vdev)
1824 /* iu_storage data allocated in initialize_event_pool */
1825 unsigned long desired_io = max_events * sizeof(union viosrp_iu);
1827 /* add io space for sg data */
1828 desired_io += (IBMVSCSI_MAX_SECTORS_DEFAULT * 512 *
1829 IBMVSCSI_CMDS_PER_LUN_DEFAULT);
1834 static void ibmvscsi_do_work(struct ibmvscsi_host_data *hostdata)
1837 char *action = "reset";
1839 if (hostdata->reset_crq) {
1841 hostdata->reset_crq = 0;
1843 rc = ibmvscsi_ops->reset_crq_queue(&hostdata->queue, hostdata);
1845 rc = ibmvscsi_ops->send_crq(hostdata, 0xC001000000000000LL, 0);
1846 vio_enable_interrupts(to_vio_dev(hostdata->dev));
1847 } else if (hostdata->reenable_crq) {
1850 rc = ibmvscsi_ops->reenable_crq_queue(&hostdata->queue, hostdata);
1851 hostdata->reenable_crq = 0;
1853 rc = ibmvscsi_ops->send_crq(hostdata, 0xC001000000000000LL, 0);
1858 atomic_set(&hostdata->request_limit, -1);
1859 dev_err(hostdata->dev, "error after %s\n", action);
1862 scsi_unblock_requests(hostdata->host);
1865 static int ibmvscsi_work_to_do(struct ibmvscsi_host_data *hostdata)
1867 if (kthread_should_stop())
1869 else if (hostdata->reset_crq) {
1872 } else if (hostdata->reenable_crq) {
1880 static int ibmvscsi_work(void *data)
1882 struct ibmvscsi_host_data *hostdata = data;
1885 set_user_nice(current, -20);
1888 rc = wait_event_interruptible(hostdata->work_wait_q,
1889 ibmvscsi_work_to_do(hostdata));
1893 if (kthread_should_stop())
1896 ibmvscsi_do_work(hostdata);
1903 * Called by bus code for each adapter
1905 static int ibmvscsi_probe(struct vio_dev *vdev, const struct vio_device_id *id)
1907 struct ibmvscsi_host_data *hostdata;
1908 struct Scsi_Host *host;
1909 struct device *dev = &vdev->dev;
1910 struct srp_rport_identifiers ids;
1911 struct srp_rport *rport;
1912 unsigned long wait_switch = 0;
1915 dev_set_drvdata(&vdev->dev, NULL);
1917 host = scsi_host_alloc(&driver_template, sizeof(*hostdata));
1919 dev_err(&vdev->dev, "couldn't allocate host data\n");
1920 goto scsi_host_alloc_failed;
1923 host->transportt = ibmvscsi_transport_template;
1924 hostdata = shost_priv(host);
1925 memset(hostdata, 0x00, sizeof(*hostdata));
1926 INIT_LIST_HEAD(&hostdata->sent);
1927 init_waitqueue_head(&hostdata->work_wait_q);
1928 hostdata->host = host;
1929 hostdata->dev = dev;
1930 atomic_set(&hostdata->request_limit, -1);
1931 hostdata->host->max_sectors = IBMVSCSI_MAX_SECTORS_DEFAULT;
1933 if (map_persist_bufs(hostdata)) {
1934 dev_err(&vdev->dev, "couldn't map persistent buffers\n");
1935 goto persist_bufs_failed;
1938 hostdata->work_thread = kthread_run(ibmvscsi_work, hostdata, "%s_%d",
1939 "ibmvscsi", host->host_no);
1941 if (IS_ERR(hostdata->work_thread)) {
1942 dev_err(&vdev->dev, "couldn't initialize kthread. rc=%ld\n",
1943 PTR_ERR(hostdata->work_thread));
1944 goto init_crq_failed;
1947 rc = ibmvscsi_ops->init_crq_queue(&hostdata->queue, hostdata, max_events);
1948 if (rc != 0 && rc != H_RESOURCE) {
1949 dev_err(&vdev->dev, "couldn't initialize crq. rc=%d\n", rc);
1952 if (initialize_event_pool(&hostdata->pool, max_events, hostdata) != 0) {
1953 dev_err(&vdev->dev, "couldn't initialize event pool\n");
1954 goto init_pool_failed;
1958 host->max_id = max_id;
1959 host->max_channel = max_channel;
1960 host->max_cmd_len = 16;
1962 if (scsi_add_host(hostdata->host, hostdata->dev))
1963 goto add_host_failed;
1965 /* we don't have a proper target_port_id so let's use the fake one */
1966 memcpy(ids.port_id, hostdata->madapter_info.partition_name,
1967 sizeof(ids.port_id));
1968 ids.roles = SRP_RPORT_ROLE_TARGET;
1969 rport = srp_rport_add(host, &ids);
1971 goto add_srp_port_failed;
1973 /* Try to send an initialization message. Note that this is allowed
1974 * to fail if the other end is not acive. In that case we don't
1977 if (ibmvscsi_ops->send_crq(hostdata, 0xC001000000000000LL, 0) == 0
1978 || rc == H_RESOURCE) {
1980 * Wait around max init_timeout secs for the adapter to finish
1981 * initializing. When we are done initializing, we will have a
1982 * valid request_limit. We don't want Linux scanning before
1985 for (wait_switch = jiffies + (init_timeout * HZ);
1986 time_before(jiffies, wait_switch) &&
1987 atomic_read(&hostdata->request_limit) < 2;) {
1992 /* if we now have a valid request_limit, initiate a scan */
1993 if (atomic_read(&hostdata->request_limit) > 0)
1994 scsi_scan_host(host);
1997 dev_set_drvdata(&vdev->dev, hostdata);
2000 add_srp_port_failed:
2001 scsi_remove_host(hostdata->host);
2003 release_event_pool(&hostdata->pool, hostdata);
2005 ibmvscsi_ops->release_crq_queue(&hostdata->queue, hostdata, max_events);
2007 kthread_stop(hostdata->work_thread);
2009 unmap_persist_bufs(hostdata);
2010 persist_bufs_failed:
2011 scsi_host_put(host);
2012 scsi_host_alloc_failed:
2016 static int ibmvscsi_remove(struct vio_dev *vdev)
2018 struct ibmvscsi_host_data *hostdata = dev_get_drvdata(&vdev->dev);
2019 unmap_persist_bufs(hostdata);
2020 release_event_pool(&hostdata->pool, hostdata);
2021 ibmvscsi_ops->release_crq_queue(&hostdata->queue, hostdata,
2024 kthread_stop(hostdata->work_thread);
2025 srp_remove_host(hostdata->host);
2026 scsi_remove_host(hostdata->host);
2027 scsi_host_put(hostdata->host);
2033 * ibmvscsi_resume: Resume from suspend
2034 * @dev: device struct
2036 * We may have lost an interrupt across suspend/resume, so kick the
2039 static int ibmvscsi_resume(struct device *dev)
2041 struct ibmvscsi_host_data *hostdata = dev_get_drvdata(dev);
2042 return ibmvscsi_ops->resume(hostdata);
2046 * ibmvscsi_device_table: Used by vio.c to match devices in the device tree we
2049 static struct vio_device_id ibmvscsi_device_table[] __devinitdata = {
2050 {"vscsi", "IBM,v-scsi"},
2053 MODULE_DEVICE_TABLE(vio, ibmvscsi_device_table);
2055 static struct dev_pm_ops ibmvscsi_pm_ops = {
2056 .resume = ibmvscsi_resume
2059 static struct vio_driver ibmvscsi_driver = {
2060 .id_table = ibmvscsi_device_table,
2061 .probe = ibmvscsi_probe,
2062 .remove = ibmvscsi_remove,
2063 .get_desired_dma = ibmvscsi_get_desired_dma,
2066 .owner = THIS_MODULE,
2067 .pm = &ibmvscsi_pm_ops,
2071 static struct srp_function_template ibmvscsi_transport_functions = {
2074 int __init ibmvscsi_module_init(void)
2078 /* Ensure we have two requests to do error recovery */
2079 driver_template.can_queue = max_requests;
2080 max_events = max_requests + 2;
2082 if (firmware_has_feature(FW_FEATURE_VIO))
2083 ibmvscsi_ops = &rpavscsi_ops;
2087 ibmvscsi_transport_template =
2088 srp_attach_transport(&ibmvscsi_transport_functions);
2089 if (!ibmvscsi_transport_template)
2092 ret = vio_register_driver(&ibmvscsi_driver);
2094 srp_release_transport(ibmvscsi_transport_template);
2098 void __exit ibmvscsi_module_exit(void)
2100 vio_unregister_driver(&ibmvscsi_driver);
2101 srp_release_transport(ibmvscsi_transport_template);
2104 module_init(ibmvscsi_module_init);
2105 module_exit(ibmvscsi_module_exit);