Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/* ------------------------------------------------------------
3 * ibmvscsi.c
4 * (C) Copyright IBM Corporation 1994, 2004
5 * Authors: Colin DeVilbiss (devilbis@us.ibm.com)
6 * Santiago Leon (santil@us.ibm.com)
7 * Dave Boutcher (sleddog@us.ibm.com)
8 *
9 * ------------------------------------------------------------
10 * Emulation of a SCSI host adapter for Virtual I/O devices
11 *
12 * This driver supports the SCSI adapter implemented by the IBM
13 * Power5 firmware. That SCSI adapter is not a physical adapter,
14 * but allows Linux SCSI peripheral drivers to directly
15 * access devices in another logical partition on the physical system.
16 *
17 * The virtual adapter(s) are present in the open firmware device
18 * tree just like real adapters.
19 *
20 * One of the capabilities provided on these systems is the ability
21 * to DMA between partitions. The architecture states that for VSCSI,
22 * the server side is allowed to DMA to and from the client. The client
23 * is never trusted to DMA to or from the server directly.
24 *
25 * Messages are sent between partitions on a "Command/Response Queue"
26 * (CRQ), which is just a buffer of 16 byte entries in the receiver's
27 * Senders cannot access the buffer directly, but send messages by
28 * making a hypervisor call and passing in the 16 bytes. The hypervisor
29 * puts the message in the next 16 byte space in round-robin fashion,
30 * turns on the high order bit of the message (the valid bit), and
31 * generates an interrupt to the receiver (if interrupts are turned on.)
32 * The receiver just turns off the valid bit when they have copied out
33 * the message.
34 *
35 * The VSCSI client builds a SCSI Remote Protocol (SRP) Information Unit
36 * (IU) (as defined in the T10 standard available at www.t10.org), gets
37 * a DMA address for the message, and sends it to the server as the
38 * payload of a CRQ message. The server DMAs the SRP IU and processes it,
39 * including doing any additional data transfers. When it is done, it
40 * DMAs the SRP response back to the same address as the request came from,
41 * and sends a CRQ message back to inform the client that the request has
42 * completed.
43 *
44 * TODO: This is currently pretty tied to the IBM pSeries hypervisor
45 * interfaces. It would be really nice to abstract this above an RDMA
46 * layer.
47 */
48
49#include <linux/module.h>
50#include <linux/moduleparam.h>
51#include <linux/dma-mapping.h>
52#include <linux/delay.h>
53#include <linux/slab.h>
54#include <linux/of.h>
55#include <linux/pm.h>
56#include <linux/kthread.h>
57#include <asm/firmware.h>
58#include <asm/vio.h>
59#include <scsi/scsi.h>
60#include <scsi/scsi_cmnd.h>
61#include <scsi/scsi_host.h>
62#include <scsi/scsi_device.h>
63#include <scsi/scsi_transport_srp.h>
64#include "ibmvscsi.h"
65
66/* The values below are somewhat arbitrary default values, but
67 * OS/400 will use 3 busses (disks, CDs, tapes, I think.)
68 * Note that there are 3 bits of channel value, 6 bits of id, and
69 * 5 bits of LUN.
70 */
71static int max_id = 64;
72static int max_channel = 3;
73static int init_timeout = 300;
74static int login_timeout = 60;
75static int info_timeout = 30;
76static int abort_timeout = 60;
77static int reset_timeout = 60;
78static int max_requests = IBMVSCSI_MAX_REQUESTS_DEFAULT;
79static int max_events = IBMVSCSI_MAX_REQUESTS_DEFAULT + 2;
80static int fast_fail = 1;
81static int client_reserve = 1;
82static char partition_name[96] = "UNKNOWN";
83static unsigned int partition_number = -1;
84static LIST_HEAD(ibmvscsi_head);
85static DEFINE_SPINLOCK(ibmvscsi_driver_lock);
86
87static struct scsi_transport_template *ibmvscsi_transport_template;
88
89#define IBMVSCSI_VERSION "1.5.9"
90
91MODULE_DESCRIPTION("IBM Virtual SCSI");
92MODULE_AUTHOR("Dave Boutcher");
93MODULE_LICENSE("GPL");
94MODULE_VERSION(IBMVSCSI_VERSION);
95
96module_param_named(max_id, max_id, int, S_IRUGO | S_IWUSR);
97MODULE_PARM_DESC(max_id, "Largest ID value for each channel [Default=64]");
98module_param_named(max_channel, max_channel, int, S_IRUGO | S_IWUSR);
99MODULE_PARM_DESC(max_channel, "Largest channel value [Default=3]");
100module_param_named(init_timeout, init_timeout, int, S_IRUGO | S_IWUSR);
101MODULE_PARM_DESC(init_timeout, "Initialization timeout in seconds");
102module_param_named(max_requests, max_requests, int, S_IRUGO);
103MODULE_PARM_DESC(max_requests, "Maximum requests for this adapter");
104module_param_named(fast_fail, fast_fail, int, S_IRUGO | S_IWUSR);
105MODULE_PARM_DESC(fast_fail, "Enable fast fail. [Default=1]");
106module_param_named(client_reserve, client_reserve, int, S_IRUGO );
107MODULE_PARM_DESC(client_reserve, "Attempt client managed reserve/release");
108
109static void ibmvscsi_handle_crq(struct viosrp_crq *crq,
110 struct ibmvscsi_host_data *hostdata);
111
112/* ------------------------------------------------------------
113 * Routines for managing the command/response queue
114 */
115/**
116 * ibmvscsi_handle_event: - Interrupt handler for crq events
117 * @irq: number of irq to handle, not used
118 * @dev_instance: ibmvscsi_host_data of host that received interrupt
119 *
120 * Disables interrupts and schedules srp_task
121 * Always returns IRQ_HANDLED
122 */
123static irqreturn_t ibmvscsi_handle_event(int irq, void *dev_instance)
124{
125 struct ibmvscsi_host_data *hostdata =
126 (struct ibmvscsi_host_data *)dev_instance;
127 vio_disable_interrupts(to_vio_dev(hostdata->dev));
128 tasklet_schedule(&hostdata->srp_task);
129 return IRQ_HANDLED;
130}
131
132/**
133 * ibmvscsi_release_crq_queue() - Deallocates data and unregisters CRQ
134 * @queue: crq_queue to initialize and register
135 * @hostdata: ibmvscsi_host_data of host
136 * @max_requests: maximum requests (unused)
137 *
138 * Frees irq, deallocates a page for messages, unmaps dma, and unregisters
139 * the crq with the hypervisor.
140 */
141static void ibmvscsi_release_crq_queue(struct crq_queue *queue,
142 struct ibmvscsi_host_data *hostdata,
143 int max_requests)
144{
145 long rc = 0;
146 struct vio_dev *vdev = to_vio_dev(hostdata->dev);
147 free_irq(vdev->irq, (void *)hostdata);
148 tasklet_kill(&hostdata->srp_task);
149 do {
150 if (rc)
151 msleep(100);
152 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
153 } while ((rc == H_BUSY) || (H_IS_LONG_BUSY(rc)));
154 dma_unmap_single(hostdata->dev,
155 queue->msg_token,
156 queue->size * sizeof(*queue->msgs), DMA_BIDIRECTIONAL);
157 free_page((unsigned long)queue->msgs);
158}
159
160/**
161 * crq_queue_next_crq: - Returns the next entry in message queue
162 * @queue: crq_queue to use
163 *
164 * Returns pointer to next entry in queue, or NULL if there are no new
165 * entried in the CRQ.
166 */
167static struct viosrp_crq *crq_queue_next_crq(struct crq_queue *queue)
168{
169 struct viosrp_crq *crq;
170 unsigned long flags;
171
172 spin_lock_irqsave(&queue->lock, flags);
173 crq = &queue->msgs[queue->cur];
174 if (crq->valid != VIOSRP_CRQ_FREE) {
175 if (++queue->cur == queue->size)
176 queue->cur = 0;
177
178 /* Ensure the read of the valid bit occurs before reading any
179 * other bits of the CRQ entry
180 */
181 rmb();
182 } else
183 crq = NULL;
184 spin_unlock_irqrestore(&queue->lock, flags);
185
186 return crq;
187}
188
189/**
190 * ibmvscsi_send_crq: - Send a CRQ
191 * @hostdata: the adapter
192 * @word1: the first 64 bits of the data
193 * @word2: the second 64 bits of the data
194 */
195static int ibmvscsi_send_crq(struct ibmvscsi_host_data *hostdata,
196 u64 word1, u64 word2)
197{
198 struct vio_dev *vdev = to_vio_dev(hostdata->dev);
199
200 /*
201 * Ensure the command buffer is flushed to memory before handing it
202 * over to the VIOS to prevent it from fetching any stale data.
203 */
204 mb();
205 return plpar_hcall_norets(H_SEND_CRQ, vdev->unit_address, word1, word2);
206}
207
208/**
209 * ibmvscsi_task: - Process srps asynchronously
210 * @data: ibmvscsi_host_data of host
211 */
212static void ibmvscsi_task(void *data)
213{
214 struct ibmvscsi_host_data *hostdata = (struct ibmvscsi_host_data *)data;
215 struct vio_dev *vdev = to_vio_dev(hostdata->dev);
216 struct viosrp_crq *crq;
217 int done = 0;
218
219 while (!done) {
220 /* Pull all the valid messages off the CRQ */
221 while ((crq = crq_queue_next_crq(&hostdata->queue)) != NULL) {
222 ibmvscsi_handle_crq(crq, hostdata);
223 crq->valid = VIOSRP_CRQ_FREE;
224 wmb();
225 }
226
227 vio_enable_interrupts(vdev);
228 crq = crq_queue_next_crq(&hostdata->queue);
229 if (crq != NULL) {
230 vio_disable_interrupts(vdev);
231 ibmvscsi_handle_crq(crq, hostdata);
232 crq->valid = VIOSRP_CRQ_FREE;
233 wmb();
234 } else {
235 done = 1;
236 }
237 }
238}
239
240static void gather_partition_info(void)
241{
242 const char *ppartition_name;
243 const __be32 *p_number_ptr;
244
245 /* Retrieve information about this partition */
246 if (!of_root)
247 return;
248
249 of_node_get(of_root);
250
251 ppartition_name = of_get_property(of_root, "ibm,partition-name", NULL);
252 if (ppartition_name)
253 strscpy(partition_name, ppartition_name,
254 sizeof(partition_name));
255 p_number_ptr = of_get_property(of_root, "ibm,partition-no", NULL);
256 if (p_number_ptr)
257 partition_number = of_read_number(p_number_ptr, 1);
258 of_node_put(of_root);
259}
260
261static void set_adapter_info(struct ibmvscsi_host_data *hostdata)
262{
263 memset(&hostdata->madapter_info, 0x00,
264 sizeof(hostdata->madapter_info));
265
266 dev_info(hostdata->dev, "SRP_VERSION: %s\n", SRP_VERSION);
267 strcpy(hostdata->madapter_info.srp_version, SRP_VERSION);
268
269 strscpy(hostdata->madapter_info.partition_name, partition_name,
270 sizeof(hostdata->madapter_info.partition_name));
271
272 hostdata->madapter_info.partition_number =
273 cpu_to_be32(partition_number);
274
275 hostdata->madapter_info.mad_version = cpu_to_be32(SRP_MAD_VERSION_1);
276 hostdata->madapter_info.os_type = cpu_to_be32(SRP_MAD_OS_LINUX);
277}
278
279/**
280 * ibmvscsi_reset_crq_queue() - resets a crq after a failure
281 * @queue: crq_queue to initialize and register
282 * @hostdata: ibmvscsi_host_data of host
283 */
284static int ibmvscsi_reset_crq_queue(struct crq_queue *queue,
285 struct ibmvscsi_host_data *hostdata)
286{
287 int rc = 0;
288 struct vio_dev *vdev = to_vio_dev(hostdata->dev);
289
290 /* Close the CRQ */
291 do {
292 if (rc)
293 msleep(100);
294 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
295 } while ((rc == H_BUSY) || (H_IS_LONG_BUSY(rc)));
296
297 /* Clean out the queue */
298 memset(queue->msgs, 0x00, PAGE_SIZE);
299 queue->cur = 0;
300
301 set_adapter_info(hostdata);
302
303 /* And re-open it again */
304 rc = plpar_hcall_norets(H_REG_CRQ,
305 vdev->unit_address,
306 queue->msg_token, PAGE_SIZE);
307 if (rc == H_CLOSED) {
308 /* Adapter is good, but other end is not ready */
309 dev_warn(hostdata->dev, "Partner adapter not ready\n");
310 } else if (rc != 0) {
311 dev_warn(hostdata->dev, "couldn't register crq--rc 0x%x\n", rc);
312 }
313 return rc;
314}
315
316/**
317 * ibmvscsi_init_crq_queue() - Initializes and registers CRQ with hypervisor
318 * @queue: crq_queue to initialize and register
319 * @hostdata: ibmvscsi_host_data of host
320 * @max_requests: maximum requests (unused)
321 *
322 * Allocates a page for messages, maps it for dma, and registers
323 * the crq with the hypervisor.
324 * Returns zero on success.
325 */
326static int ibmvscsi_init_crq_queue(struct crq_queue *queue,
327 struct ibmvscsi_host_data *hostdata,
328 int max_requests)
329{
330 int rc;
331 int retrc;
332 struct vio_dev *vdev = to_vio_dev(hostdata->dev);
333
334 queue->msgs = (struct viosrp_crq *)get_zeroed_page(GFP_KERNEL);
335
336 if (!queue->msgs)
337 goto malloc_failed;
338 queue->size = PAGE_SIZE / sizeof(*queue->msgs);
339
340 queue->msg_token = dma_map_single(hostdata->dev, queue->msgs,
341 queue->size * sizeof(*queue->msgs),
342 DMA_BIDIRECTIONAL);
343
344 if (dma_mapping_error(hostdata->dev, queue->msg_token))
345 goto map_failed;
346
347 gather_partition_info();
348 set_adapter_info(hostdata);
349
350 retrc = rc = plpar_hcall_norets(H_REG_CRQ,
351 vdev->unit_address,
352 queue->msg_token, PAGE_SIZE);
353 if (rc == H_RESOURCE)
354 /* maybe kexecing and resource is busy. try a reset */
355 rc = ibmvscsi_reset_crq_queue(queue,
356 hostdata);
357
358 if (rc == H_CLOSED) {
359 /* Adapter is good, but other end is not ready */
360 dev_warn(hostdata->dev, "Partner adapter not ready\n");
361 retrc = 0;
362 } else if (rc != 0) {
363 dev_warn(hostdata->dev, "Error %d opening adapter\n", rc);
364 goto reg_crq_failed;
365 }
366
367 queue->cur = 0;
368 spin_lock_init(&queue->lock);
369
370 tasklet_init(&hostdata->srp_task, (void *)ibmvscsi_task,
371 (unsigned long)hostdata);
372
373 if (request_irq(vdev->irq,
374 ibmvscsi_handle_event,
375 0, "ibmvscsi", (void *)hostdata) != 0) {
376 dev_err(hostdata->dev, "couldn't register irq 0x%x\n",
377 vdev->irq);
378 goto req_irq_failed;
379 }
380
381 rc = vio_enable_interrupts(vdev);
382 if (rc != 0) {
383 dev_err(hostdata->dev, "Error %d enabling interrupts!!!\n", rc);
384 goto req_irq_failed;
385 }
386
387 return retrc;
388
389 req_irq_failed:
390 tasklet_kill(&hostdata->srp_task);
391 rc = 0;
392 do {
393 if (rc)
394 msleep(100);
395 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
396 } while ((rc == H_BUSY) || (H_IS_LONG_BUSY(rc)));
397 reg_crq_failed:
398 dma_unmap_single(hostdata->dev,
399 queue->msg_token,
400 queue->size * sizeof(*queue->msgs), DMA_BIDIRECTIONAL);
401 map_failed:
402 free_page((unsigned long)queue->msgs);
403 malloc_failed:
404 return -1;
405}
406
407/**
408 * ibmvscsi_reenable_crq_queue() - reenables a crq after
409 * @queue: crq_queue to initialize and register
410 * @hostdata: ibmvscsi_host_data of host
411 */
412static int ibmvscsi_reenable_crq_queue(struct crq_queue *queue,
413 struct ibmvscsi_host_data *hostdata)
414{
415 int rc = 0;
416 struct vio_dev *vdev = to_vio_dev(hostdata->dev);
417
418 set_adapter_info(hostdata);
419
420 /* Re-enable the CRQ */
421 do {
422 if (rc)
423 msleep(100);
424 rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address);
425 } while ((rc == H_IN_PROGRESS) || (rc == H_BUSY) || (H_IS_LONG_BUSY(rc)));
426
427 if (rc)
428 dev_err(hostdata->dev, "Error %d enabling adapter\n", rc);
429 return rc;
430}
431
432/* ------------------------------------------------------------
433 * Routines for the event pool and event structs
434 */
435/**
436 * initialize_event_pool: - Allocates and initializes the event pool for a host
437 * @pool: event_pool to be initialized
438 * @size: Number of events in pool
439 * @hostdata: ibmvscsi_host_data who owns the event pool
440 *
441 * Returns zero on success.
442 */
443static int initialize_event_pool(struct event_pool *pool,
444 int size, struct ibmvscsi_host_data *hostdata)
445{
446 int i;
447
448 pool->size = size;
449 pool->next = 0;
450 pool->events = kcalloc(pool->size, sizeof(*pool->events), GFP_KERNEL);
451 if (!pool->events)
452 return -ENOMEM;
453
454 pool->iu_storage =
455 dma_alloc_coherent(hostdata->dev,
456 pool->size * sizeof(*pool->iu_storage),
457 &pool->iu_token, GFP_KERNEL);
458 if (!pool->iu_storage) {
459 kfree(pool->events);
460 return -ENOMEM;
461 }
462
463 for (i = 0; i < pool->size; ++i) {
464 struct srp_event_struct *evt = &pool->events[i];
465 memset(&evt->crq, 0x00, sizeof(evt->crq));
466 atomic_set(&evt->free, 1);
467 evt->crq.valid = VIOSRP_CRQ_CMD_RSP;
468 evt->crq.IU_length = cpu_to_be16(sizeof(*evt->xfer_iu));
469 evt->crq.IU_data_ptr = cpu_to_be64(pool->iu_token +
470 sizeof(*evt->xfer_iu) * i);
471 evt->xfer_iu = pool->iu_storage + i;
472 evt->hostdata = hostdata;
473 evt->ext_list = NULL;
474 evt->ext_list_token = 0;
475 }
476
477 return 0;
478}
479
480/**
481 * release_event_pool() - Frees memory of an event pool of a host
482 * @pool: event_pool to be released
483 * @hostdata: ibmvscsi_host_data who owns the even pool
484 *
485 * Returns zero on success.
486 */
487static void release_event_pool(struct event_pool *pool,
488 struct ibmvscsi_host_data *hostdata)
489{
490 int i, in_use = 0;
491 for (i = 0; i < pool->size; ++i) {
492 if (atomic_read(&pool->events[i].free) != 1)
493 ++in_use;
494 if (pool->events[i].ext_list) {
495 dma_free_coherent(hostdata->dev,
496 SG_ALL * sizeof(struct srp_direct_buf),
497 pool->events[i].ext_list,
498 pool->events[i].ext_list_token);
499 }
500 }
501 if (in_use)
502 dev_warn(hostdata->dev, "releasing event pool with %d "
503 "events still in use?\n", in_use);
504 kfree(pool->events);
505 dma_free_coherent(hostdata->dev,
506 pool->size * sizeof(*pool->iu_storage),
507 pool->iu_storage, pool->iu_token);
508}
509
510/**
511 * valid_event_struct: - Determines if event is valid.
512 * @pool: event_pool that contains the event
513 * @evt: srp_event_struct to be checked for validity
514 *
515 * Returns zero if event is invalid, one otherwise.
516*/
517static int valid_event_struct(struct event_pool *pool,
518 struct srp_event_struct *evt)
519{
520 int index = evt - pool->events;
521 if (index < 0 || index >= pool->size) /* outside of bounds */
522 return 0;
523 if (evt != pool->events + index) /* unaligned */
524 return 0;
525 return 1;
526}
527
528/**
529 * free_event_struct() - Changes status of event to "free"
530 * @pool: event_pool that contains the event
531 * @evt: srp_event_struct to be modified
532 */
533static void free_event_struct(struct event_pool *pool,
534 struct srp_event_struct *evt)
535{
536 if (!valid_event_struct(pool, evt)) {
537 dev_err(evt->hostdata->dev, "Freeing invalid event_struct %p "
538 "(not in pool %p)\n", evt, pool->events);
539 return;
540 }
541 if (atomic_inc_return(&evt->free) != 1) {
542 dev_err(evt->hostdata->dev, "Freeing event_struct %p "
543 "which is not in use!\n", evt);
544 return;
545 }
546}
547
548/**
549 * get_event_struct() - Gets the next free event in pool
550 * @pool: event_pool that contains the events to be searched
551 *
552 * Returns the next event in "free" state, and NULL if none are free.
553 * Note that no synchronization is done here, we assume the host_lock
554 * will syncrhonze things.
555*/
556static struct srp_event_struct *get_event_struct(struct event_pool *pool)
557{
558 int i;
559 int poolsize = pool->size;
560 int offset = pool->next;
561
562 for (i = 0; i < poolsize; i++) {
563 offset = (offset + 1) % poolsize;
564 if (!atomic_dec_if_positive(&pool->events[offset].free)) {
565 pool->next = offset;
566 return &pool->events[offset];
567 }
568 }
569
570 printk(KERN_ERR "ibmvscsi: found no event struct in pool!\n");
571 return NULL;
572}
573
574/**
575 * init_event_struct: Initialize fields in an event struct that are always
576 * required.
577 * @evt_struct: The event
578 * @done: Routine to call when the event is responded to
579 * @format: SRP or MAD format
580 * @timeout: timeout value set in the CRQ
581 */
582static void init_event_struct(struct srp_event_struct *evt_struct,
583 void (*done) (struct srp_event_struct *),
584 u8 format,
585 int timeout)
586{
587 evt_struct->cmnd = NULL;
588 evt_struct->cmnd_done = NULL;
589 evt_struct->sync_srp = NULL;
590 evt_struct->crq.format = format;
591 evt_struct->crq.timeout = cpu_to_be16(timeout);
592 evt_struct->done = done;
593}
594
595/* ------------------------------------------------------------
596 * Routines for receiving SCSI responses from the hosting partition
597 */
598
599/*
600 * set_srp_direction: Set the fields in the srp related to data
601 * direction and number of buffers based on the direction in
602 * the scsi_cmnd and the number of buffers
603 */
604static void set_srp_direction(struct scsi_cmnd *cmd,
605 struct srp_cmd *srp_cmd,
606 int numbuf)
607{
608 u8 fmt;
609
610 if (numbuf == 0)
611 return;
612
613 if (numbuf == 1)
614 fmt = SRP_DATA_DESC_DIRECT;
615 else {
616 fmt = SRP_DATA_DESC_INDIRECT;
617 numbuf = min(numbuf, MAX_INDIRECT_BUFS);
618
619 if (cmd->sc_data_direction == DMA_TO_DEVICE)
620 srp_cmd->data_out_desc_cnt = numbuf;
621 else
622 srp_cmd->data_in_desc_cnt = numbuf;
623 }
624
625 if (cmd->sc_data_direction == DMA_TO_DEVICE)
626 srp_cmd->buf_fmt = fmt << 4;
627 else
628 srp_cmd->buf_fmt = fmt;
629}
630
631/**
632 * unmap_cmd_data: - Unmap data pointed in srp_cmd based on the format
633 * @cmd: srp_cmd whose additional_data member will be unmapped
634 * @evt_struct: the event
635 * @dev: device for which the memory is mapped
636 */
637static void unmap_cmd_data(struct srp_cmd *cmd,
638 struct srp_event_struct *evt_struct,
639 struct device *dev)
640{
641 u8 out_fmt, in_fmt;
642
643 out_fmt = cmd->buf_fmt >> 4;
644 in_fmt = cmd->buf_fmt & ((1U << 4) - 1);
645
646 if (out_fmt == SRP_NO_DATA_DESC && in_fmt == SRP_NO_DATA_DESC)
647 return;
648
649 if (evt_struct->cmnd)
650 scsi_dma_unmap(evt_struct->cmnd);
651}
652
653static int map_sg_list(struct scsi_cmnd *cmd, int nseg,
654 struct srp_direct_buf *md)
655{
656 int i;
657 struct scatterlist *sg;
658 u64 total_length = 0;
659
660 scsi_for_each_sg(cmd, sg, nseg, i) {
661 struct srp_direct_buf *descr = md + i;
662 descr->va = cpu_to_be64(sg_dma_address(sg));
663 descr->len = cpu_to_be32(sg_dma_len(sg));
664 descr->key = 0;
665 total_length += sg_dma_len(sg);
666 }
667 return total_length;
668}
669
670/**
671 * map_sg_data: - Maps dma for a scatterlist and initializes descriptor fields
672 * @cmd: struct scsi_cmnd with the scatterlist
673 * @evt_struct: struct srp_event_struct to map
674 * @srp_cmd: srp_cmd that contains the memory descriptor
675 * @dev: device for which to map dma memory
676 *
677 * Called by map_data_for_srp_cmd() when building srp cmd from scsi cmd.
678 * Returns 1 on success.
679*/
680static int map_sg_data(struct scsi_cmnd *cmd,
681 struct srp_event_struct *evt_struct,
682 struct srp_cmd *srp_cmd, struct device *dev)
683{
684
685 int sg_mapped;
686 u64 total_length = 0;
687 struct srp_direct_buf *data =
688 (struct srp_direct_buf *) srp_cmd->add_data;
689 struct srp_indirect_buf *indirect =
690 (struct srp_indirect_buf *) data;
691
692 sg_mapped = scsi_dma_map(cmd);
693 if (!sg_mapped)
694 return 1;
695 else if (sg_mapped < 0)
696 return 0;
697
698 set_srp_direction(cmd, srp_cmd, sg_mapped);
699
700 /* special case; we can use a single direct descriptor */
701 if (sg_mapped == 1) {
702 map_sg_list(cmd, sg_mapped, data);
703 return 1;
704 }
705
706 indirect->table_desc.va = 0;
707 indirect->table_desc.len = cpu_to_be32(sg_mapped *
708 sizeof(struct srp_direct_buf));
709 indirect->table_desc.key = 0;
710
711 if (sg_mapped <= MAX_INDIRECT_BUFS) {
712 total_length = map_sg_list(cmd, sg_mapped,
713 &indirect->desc_list[0]);
714 indirect->len = cpu_to_be32(total_length);
715 return 1;
716 }
717
718 /* get indirect table */
719 if (!evt_struct->ext_list) {
720 evt_struct->ext_list = dma_alloc_coherent(dev,
721 SG_ALL * sizeof(struct srp_direct_buf),
722 &evt_struct->ext_list_token, 0);
723 if (!evt_struct->ext_list) {
724 if (!firmware_has_feature(FW_FEATURE_CMO))
725 sdev_printk(KERN_ERR, cmd->device,
726 "Can't allocate memory "
727 "for indirect table\n");
728 scsi_dma_unmap(cmd);
729 return 0;
730 }
731 }
732
733 total_length = map_sg_list(cmd, sg_mapped, evt_struct->ext_list);
734
735 indirect->len = cpu_to_be32(total_length);
736 indirect->table_desc.va = cpu_to_be64(evt_struct->ext_list_token);
737 indirect->table_desc.len = cpu_to_be32(sg_mapped *
738 sizeof(indirect->desc_list[0]));
739 memcpy(indirect->desc_list, evt_struct->ext_list,
740 MAX_INDIRECT_BUFS * sizeof(struct srp_direct_buf));
741 return 1;
742}
743
744/**
745 * map_data_for_srp_cmd: - Calls functions to map data for srp cmds
746 * @cmd: struct scsi_cmnd with the memory to be mapped
747 * @evt_struct: struct srp_event_struct to map
748 * @srp_cmd: srp_cmd that contains the memory descriptor
749 * @dev: dma device for which to map dma memory
750 *
751 * Called by scsi_cmd_to_srp_cmd() when converting scsi cmds to srp cmds
752 * Returns 1 on success.
753*/
754static int map_data_for_srp_cmd(struct scsi_cmnd *cmd,
755 struct srp_event_struct *evt_struct,
756 struct srp_cmd *srp_cmd, struct device *dev)
757{
758 switch (cmd->sc_data_direction) {
759 case DMA_FROM_DEVICE:
760 case DMA_TO_DEVICE:
761 break;
762 case DMA_NONE:
763 return 1;
764 case DMA_BIDIRECTIONAL:
765 sdev_printk(KERN_ERR, cmd->device,
766 "Can't map DMA_BIDIRECTIONAL to read/write\n");
767 return 0;
768 default:
769 sdev_printk(KERN_ERR, cmd->device,
770 "Unknown data direction 0x%02x; can't map!\n",
771 cmd->sc_data_direction);
772 return 0;
773 }
774
775 return map_sg_data(cmd, evt_struct, srp_cmd, dev);
776}
777
778/**
779 * purge_requests: Our virtual adapter just shut down. purge any sent requests
780 * @hostdata: the adapter
781 * @error_code: error code to return as the 'result'
782 */
783static void purge_requests(struct ibmvscsi_host_data *hostdata, int error_code)
784{
785 struct srp_event_struct *evt;
786 unsigned long flags;
787
788 spin_lock_irqsave(hostdata->host->host_lock, flags);
789 while (!list_empty(&hostdata->sent)) {
790 evt = list_first_entry(&hostdata->sent, struct srp_event_struct, list);
791 list_del(&evt->list);
792 del_timer(&evt->timer);
793
794 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
795 if (evt->cmnd) {
796 evt->cmnd->result = (error_code << 16);
797 unmap_cmd_data(&evt->iu.srp.cmd, evt,
798 evt->hostdata->dev);
799 if (evt->cmnd_done)
800 evt->cmnd_done(evt->cmnd);
801 } else if (evt->done && evt->crq.format != VIOSRP_MAD_FORMAT &&
802 evt->iu.srp.login_req.opcode != SRP_LOGIN_REQ)
803 evt->done(evt);
804 free_event_struct(&evt->hostdata->pool, evt);
805 spin_lock_irqsave(hostdata->host->host_lock, flags);
806 }
807 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
808}
809
810/**
811 * ibmvscsi_set_request_limit - Set the adapter request_limit in response to
812 * an adapter failure, reset, or SRP Login. Done under host lock to prevent
813 * race with SCSI command submission.
814 * @hostdata: adapter to adjust
815 * @limit: new request limit
816 */
817static void ibmvscsi_set_request_limit(struct ibmvscsi_host_data *hostdata, int limit)
818{
819 unsigned long flags;
820
821 spin_lock_irqsave(hostdata->host->host_lock, flags);
822 atomic_set(&hostdata->request_limit, limit);
823 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
824}
825
826/**
827 * ibmvscsi_reset_host - Reset the connection to the server
828 * @hostdata: struct ibmvscsi_host_data to reset
829*/
830static void ibmvscsi_reset_host(struct ibmvscsi_host_data *hostdata)
831{
832 scsi_block_requests(hostdata->host);
833 ibmvscsi_set_request_limit(hostdata, 0);
834
835 purge_requests(hostdata, DID_ERROR);
836 hostdata->action = IBMVSCSI_HOST_ACTION_RESET;
837 wake_up(&hostdata->work_wait_q);
838}
839
840/**
841 * ibmvscsi_timeout - Internal command timeout handler
842 * @t: struct srp_event_struct that timed out
843 *
844 * Called when an internally generated command times out
845*/
846static void ibmvscsi_timeout(struct timer_list *t)
847{
848 struct srp_event_struct *evt_struct = from_timer(evt_struct, t, timer);
849 struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
850
851 dev_err(hostdata->dev, "Command timed out (%x). Resetting connection\n",
852 evt_struct->iu.srp.cmd.opcode);
853
854 ibmvscsi_reset_host(hostdata);
855}
856
857
858/* ------------------------------------------------------------
859 * Routines for sending and receiving SRPs
860 */
861/**
862 * ibmvscsi_send_srp_event: - Transforms event to u64 array and calls send_crq()
863 * @evt_struct: evt_struct to be sent
864 * @hostdata: ibmvscsi_host_data of host
865 * @timeout: timeout in seconds - 0 means do not time command
866 *
867 * Returns the value returned from ibmvscsi_send_crq(). (Zero for success)
868 * Note that this routine assumes that host_lock is held for synchronization
869*/
870static int ibmvscsi_send_srp_event(struct srp_event_struct *evt_struct,
871 struct ibmvscsi_host_data *hostdata,
872 unsigned long timeout)
873{
874 __be64 *crq_as_u64 = (__be64 *)&evt_struct->crq;
875 int request_status = 0;
876 int rc;
877 int srp_req = 0;
878
879 /* If we have exhausted our request limit, just fail this request,
880 * unless it is for a reset or abort.
881 * Note that there are rare cases involving driver generated requests
882 * (such as task management requests) that the mid layer may think we
883 * can handle more requests (can_queue) when we actually can't
884 */
885 if (evt_struct->crq.format == VIOSRP_SRP_FORMAT) {
886 srp_req = 1;
887 request_status =
888 atomic_dec_if_positive(&hostdata->request_limit);
889 /* If request limit was -1 when we started, it is now even
890 * less than that
891 */
892 if (request_status < -1)
893 goto send_error;
894 /* Otherwise, we may have run out of requests. */
895 /* If request limit was 0 when we started the adapter is in the
896 * process of performing a login with the server adapter, or
897 * we may have run out of requests.
898 */
899 else if (request_status == -1 &&
900 evt_struct->iu.srp.login_req.opcode != SRP_LOGIN_REQ)
901 goto send_busy;
902 /* Abort and reset calls should make it through.
903 * Nothing except abort and reset should use the last two
904 * slots unless we had two or less to begin with.
905 */
906 else if (request_status < 2 &&
907 evt_struct->iu.srp.cmd.opcode != SRP_TSK_MGMT) {
908 /* In the case that we have less than two requests
909 * available, check the server limit as a combination
910 * of the request limit and the number of requests
911 * in-flight (the size of the send list). If the
912 * server limit is greater than 2, return busy so
913 * that the last two are reserved for reset and abort.
914 */
915 int server_limit = request_status;
916 struct srp_event_struct *tmp_evt;
917
918 list_for_each_entry(tmp_evt, &hostdata->sent, list) {
919 server_limit++;
920 }
921
922 if (server_limit > 2)
923 goto send_busy;
924 }
925 }
926
927 /* Copy the IU into the transfer area */
928 *evt_struct->xfer_iu = evt_struct->iu;
929 evt_struct->xfer_iu->srp.rsp.tag = (u64)evt_struct;
930
931 /* Add this to the sent list. We need to do this
932 * before we actually send
933 * in case it comes back REALLY fast
934 */
935 list_add_tail(&evt_struct->list, &hostdata->sent);
936
937 timer_setup(&evt_struct->timer, ibmvscsi_timeout, 0);
938 if (timeout) {
939 evt_struct->timer.expires = jiffies + (timeout * HZ);
940 add_timer(&evt_struct->timer);
941 }
942
943 rc = ibmvscsi_send_crq(hostdata, be64_to_cpu(crq_as_u64[0]),
944 be64_to_cpu(crq_as_u64[1]));
945 if (rc != 0) {
946 list_del(&evt_struct->list);
947 del_timer(&evt_struct->timer);
948
949 /* If send_crq returns H_CLOSED, return SCSI_MLQUEUE_HOST_BUSY.
950 * Firmware will send a CRQ with a transport event (0xFF) to
951 * tell this client what has happened to the transport. This
952 * will be handled in ibmvscsi_handle_crq()
953 */
954 if (rc == H_CLOSED) {
955 dev_warn(hostdata->dev, "send warning. "
956 "Receive queue closed, will retry.\n");
957 goto send_busy;
958 }
959 dev_err(hostdata->dev, "send error %d\n", rc);
960 if (srp_req)
961 atomic_inc(&hostdata->request_limit);
962 goto send_error;
963 }
964
965 return 0;
966
967 send_busy:
968 unmap_cmd_data(&evt_struct->iu.srp.cmd, evt_struct, hostdata->dev);
969
970 free_event_struct(&hostdata->pool, evt_struct);
971 if (srp_req && request_status != -1)
972 atomic_inc(&hostdata->request_limit);
973 return SCSI_MLQUEUE_HOST_BUSY;
974
975 send_error:
976 unmap_cmd_data(&evt_struct->iu.srp.cmd, evt_struct, hostdata->dev);
977
978 if (evt_struct->cmnd != NULL) {
979 evt_struct->cmnd->result = DID_ERROR << 16;
980 evt_struct->cmnd_done(evt_struct->cmnd);
981 } else if (evt_struct->done)
982 evt_struct->done(evt_struct);
983
984 free_event_struct(&hostdata->pool, evt_struct);
985 return 0;
986}
987
988/**
989 * handle_cmd_rsp: - Handle responses from commands
990 * @evt_struct: srp_event_struct to be handled
991 *
992 * Used as a callback by when sending scsi cmds.
993 * Gets called by ibmvscsi_handle_crq()
994*/
995static void handle_cmd_rsp(struct srp_event_struct *evt_struct)
996{
997 struct srp_rsp *rsp = &evt_struct->xfer_iu->srp.rsp;
998 struct scsi_cmnd *cmnd = evt_struct->cmnd;
999
1000 if (unlikely(rsp->opcode != SRP_RSP)) {
1001 if (printk_ratelimit())
1002 dev_warn(evt_struct->hostdata->dev,
1003 "bad SRP RSP type %#02x\n", rsp->opcode);
1004 }
1005
1006 if (cmnd) {
1007 cmnd->result |= rsp->status;
1008 if (scsi_status_is_check_condition(cmnd->result))
1009 memcpy(cmnd->sense_buffer,
1010 rsp->data,
1011 be32_to_cpu(rsp->sense_data_len));
1012 unmap_cmd_data(&evt_struct->iu.srp.cmd,
1013 evt_struct,
1014 evt_struct->hostdata->dev);
1015
1016 if (rsp->flags & SRP_RSP_FLAG_DOOVER)
1017 scsi_set_resid(cmnd,
1018 be32_to_cpu(rsp->data_out_res_cnt));
1019 else if (rsp->flags & SRP_RSP_FLAG_DIOVER)
1020 scsi_set_resid(cmnd, be32_to_cpu(rsp->data_in_res_cnt));
1021 }
1022
1023 if (evt_struct->cmnd_done)
1024 evt_struct->cmnd_done(cmnd);
1025}
1026
1027/**
1028 * lun_from_dev: - Returns the lun of the scsi device
1029 * @dev: struct scsi_device
1030 *
1031*/
1032static inline u16 lun_from_dev(struct scsi_device *dev)
1033{
1034 return (0x2 << 14) | (dev->id << 8) | (dev->channel << 5) | dev->lun;
1035}
1036
1037/**
1038 * ibmvscsi_queuecommand_lck() - The queuecommand function of the scsi template
1039 * @cmnd: struct scsi_cmnd to be executed
1040 * @done: Callback function to be called when cmd is completed
1041*/
1042static int ibmvscsi_queuecommand_lck(struct scsi_cmnd *cmnd)
1043{
1044 void (*done)(struct scsi_cmnd *) = scsi_done;
1045 struct srp_cmd *srp_cmd;
1046 struct srp_event_struct *evt_struct;
1047 struct srp_indirect_buf *indirect;
1048 struct ibmvscsi_host_data *hostdata = shost_priv(cmnd->device->host);
1049 u16 lun = lun_from_dev(cmnd->device);
1050 u8 out_fmt, in_fmt;
1051
1052 cmnd->result = (DID_OK << 16);
1053 evt_struct = get_event_struct(&hostdata->pool);
1054 if (!evt_struct)
1055 return SCSI_MLQUEUE_HOST_BUSY;
1056
1057 /* Set up the actual SRP IU */
1058 BUILD_BUG_ON(sizeof(evt_struct->iu.srp) != SRP_MAX_IU_LEN);
1059 memset(&evt_struct->iu.srp, 0x00, sizeof(evt_struct->iu.srp));
1060 srp_cmd = &evt_struct->iu.srp.cmd;
1061 srp_cmd->opcode = SRP_CMD;
1062 memcpy(srp_cmd->cdb, cmnd->cmnd, sizeof(srp_cmd->cdb));
1063 int_to_scsilun(lun, &srp_cmd->lun);
1064
1065 if (!map_data_for_srp_cmd(cmnd, evt_struct, srp_cmd, hostdata->dev)) {
1066 if (!firmware_has_feature(FW_FEATURE_CMO))
1067 sdev_printk(KERN_ERR, cmnd->device,
1068 "couldn't convert cmd to srp_cmd\n");
1069 free_event_struct(&hostdata->pool, evt_struct);
1070 return SCSI_MLQUEUE_HOST_BUSY;
1071 }
1072
1073 init_event_struct(evt_struct,
1074 handle_cmd_rsp,
1075 VIOSRP_SRP_FORMAT,
1076 scsi_cmd_to_rq(cmnd)->timeout / HZ);
1077
1078 evt_struct->cmnd = cmnd;
1079 evt_struct->cmnd_done = done;
1080
1081 /* Fix up dma address of the buffer itself */
1082 indirect = (struct srp_indirect_buf *) srp_cmd->add_data;
1083 out_fmt = srp_cmd->buf_fmt >> 4;
1084 in_fmt = srp_cmd->buf_fmt & ((1U << 4) - 1);
1085 if ((in_fmt == SRP_DATA_DESC_INDIRECT ||
1086 out_fmt == SRP_DATA_DESC_INDIRECT) &&
1087 indirect->table_desc.va == 0) {
1088 indirect->table_desc.va =
1089 cpu_to_be64(be64_to_cpu(evt_struct->crq.IU_data_ptr) +
1090 offsetof(struct srp_cmd, add_data) +
1091 offsetof(struct srp_indirect_buf, desc_list));
1092 }
1093
1094 return ibmvscsi_send_srp_event(evt_struct, hostdata, 0);
1095}
1096
1097static DEF_SCSI_QCMD(ibmvscsi_queuecommand)
1098
1099/* ------------------------------------------------------------
1100 * Routines for driver initialization
1101 */
1102
1103/**
1104 * map_persist_bufs: - Pre-map persistent data for adapter logins
1105 * @hostdata: ibmvscsi_host_data of host
1106 *
1107 * Map the capabilities and adapter info DMA buffers to avoid runtime failures.
1108 * Return 1 on error, 0 on success.
1109 */
1110static int map_persist_bufs(struct ibmvscsi_host_data *hostdata)
1111{
1112
1113 hostdata->caps_addr = dma_map_single(hostdata->dev, &hostdata->caps,
1114 sizeof(hostdata->caps), DMA_BIDIRECTIONAL);
1115
1116 if (dma_mapping_error(hostdata->dev, hostdata->caps_addr)) {
1117 dev_err(hostdata->dev, "Unable to map capabilities buffer!\n");
1118 return 1;
1119 }
1120
1121 hostdata->adapter_info_addr = dma_map_single(hostdata->dev,
1122 &hostdata->madapter_info,
1123 sizeof(hostdata->madapter_info),
1124 DMA_BIDIRECTIONAL);
1125 if (dma_mapping_error(hostdata->dev, hostdata->adapter_info_addr)) {
1126 dev_err(hostdata->dev, "Unable to map adapter info buffer!\n");
1127 dma_unmap_single(hostdata->dev, hostdata->caps_addr,
1128 sizeof(hostdata->caps), DMA_BIDIRECTIONAL);
1129 return 1;
1130 }
1131
1132 return 0;
1133}
1134
1135/**
1136 * unmap_persist_bufs: - Unmap persistent data needed for adapter logins
1137 * @hostdata: ibmvscsi_host_data of host
1138 *
1139 * Unmap the capabilities and adapter info DMA buffers
1140 */
1141static void unmap_persist_bufs(struct ibmvscsi_host_data *hostdata)
1142{
1143 dma_unmap_single(hostdata->dev, hostdata->caps_addr,
1144 sizeof(hostdata->caps), DMA_BIDIRECTIONAL);
1145
1146 dma_unmap_single(hostdata->dev, hostdata->adapter_info_addr,
1147 sizeof(hostdata->madapter_info), DMA_BIDIRECTIONAL);
1148}
1149
1150/**
1151 * login_rsp: - Handle response to SRP login request
1152 * @evt_struct: srp_event_struct with the response
1153 *
1154 * Used as a "done" callback by when sending srp_login. Gets called
1155 * by ibmvscsi_handle_crq()
1156*/
1157static void login_rsp(struct srp_event_struct *evt_struct)
1158{
1159 struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
1160 switch (evt_struct->xfer_iu->srp.login_rsp.opcode) {
1161 case SRP_LOGIN_RSP: /* it worked! */
1162 break;
1163 case SRP_LOGIN_REJ: /* refused! */
1164 dev_info(hostdata->dev, "SRP_LOGIN_REJ reason %u\n",
1165 evt_struct->xfer_iu->srp.login_rej.reason);
1166 /* Login failed. */
1167 ibmvscsi_set_request_limit(hostdata, -1);
1168 return;
1169 default:
1170 dev_err(hostdata->dev, "Invalid login response typecode 0x%02x!\n",
1171 evt_struct->xfer_iu->srp.login_rsp.opcode);
1172 /* Login failed. */
1173 ibmvscsi_set_request_limit(hostdata, -1);
1174 return;
1175 }
1176
1177 dev_info(hostdata->dev, "SRP_LOGIN succeeded\n");
1178 hostdata->client_migrated = 0;
1179
1180 /* Now we know what the real request-limit is.
1181 * This value is set rather than added to request_limit because
1182 * request_limit could have been set to -1 by this client.
1183 */
1184 ibmvscsi_set_request_limit(hostdata,
1185 be32_to_cpu(evt_struct->xfer_iu->srp.login_rsp.req_lim_delta));
1186
1187 /* If we had any pending I/Os, kick them */
1188 hostdata->action = IBMVSCSI_HOST_ACTION_UNBLOCK;
1189 wake_up(&hostdata->work_wait_q);
1190}
1191
1192/**
1193 * send_srp_login: - Sends the srp login
1194 * @hostdata: ibmvscsi_host_data of host
1195 *
1196 * Returns zero if successful.
1197*/
1198static int send_srp_login(struct ibmvscsi_host_data *hostdata)
1199{
1200 int rc;
1201 unsigned long flags;
1202 struct srp_login_req *login;
1203 struct srp_event_struct *evt_struct = get_event_struct(&hostdata->pool);
1204
1205 BUG_ON(!evt_struct);
1206 init_event_struct(evt_struct, login_rsp,
1207 VIOSRP_SRP_FORMAT, login_timeout);
1208
1209 login = &evt_struct->iu.srp.login_req;
1210 memset(login, 0, sizeof(*login));
1211 login->opcode = SRP_LOGIN_REQ;
1212 login->req_it_iu_len = cpu_to_be32(sizeof(union srp_iu));
1213 login->req_buf_fmt = cpu_to_be16(SRP_BUF_FORMAT_DIRECT |
1214 SRP_BUF_FORMAT_INDIRECT);
1215
1216 /* Start out with a request limit of 0, since this is negotiated in
1217 * the login request we are just sending and login requests always
1218 * get sent by the driver regardless of request_limit.
1219 */
1220 ibmvscsi_set_request_limit(hostdata, 0);
1221
1222 spin_lock_irqsave(hostdata->host->host_lock, flags);
1223 rc = ibmvscsi_send_srp_event(evt_struct, hostdata, login_timeout * 2);
1224 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1225 dev_info(hostdata->dev, "sent SRP login\n");
1226 return rc;
1227};
1228
1229/**
1230 * capabilities_rsp: - Handle response to MAD adapter capabilities request
1231 * @evt_struct: srp_event_struct with the response
1232 *
1233 * Used as a "done" callback by when sending adapter_info.
1234 */
1235static void capabilities_rsp(struct srp_event_struct *evt_struct)
1236{
1237 struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
1238
1239 if (evt_struct->xfer_iu->mad.capabilities.common.status) {
1240 dev_err(hostdata->dev, "error 0x%X getting capabilities info\n",
1241 evt_struct->xfer_iu->mad.capabilities.common.status);
1242 } else {
1243 if (hostdata->caps.migration.common.server_support !=
1244 cpu_to_be16(SERVER_SUPPORTS_CAP))
1245 dev_info(hostdata->dev, "Partition migration not supported\n");
1246
1247 if (client_reserve) {
1248 if (hostdata->caps.reserve.common.server_support ==
1249 cpu_to_be16(SERVER_SUPPORTS_CAP))
1250 dev_info(hostdata->dev, "Client reserve enabled\n");
1251 else
1252 dev_info(hostdata->dev, "Client reserve not supported\n");
1253 }
1254 }
1255
1256 send_srp_login(hostdata);
1257}
1258
1259/**
1260 * send_mad_capabilities: - Sends the mad capabilities request
1261 * and stores the result so it can be retrieved with
1262 * @hostdata: ibmvscsi_host_data of host
1263 */
1264static void send_mad_capabilities(struct ibmvscsi_host_data *hostdata)
1265{
1266 struct viosrp_capabilities *req;
1267 struct srp_event_struct *evt_struct;
1268 unsigned long flags;
1269 struct device_node *of_node = hostdata->dev->of_node;
1270 const char *location;
1271
1272 evt_struct = get_event_struct(&hostdata->pool);
1273 BUG_ON(!evt_struct);
1274
1275 init_event_struct(evt_struct, capabilities_rsp,
1276 VIOSRP_MAD_FORMAT, info_timeout);
1277
1278 req = &evt_struct->iu.mad.capabilities;
1279 memset(req, 0, sizeof(*req));
1280
1281 hostdata->caps.flags = cpu_to_be32(CAP_LIST_SUPPORTED);
1282 if (hostdata->client_migrated)
1283 hostdata->caps.flags |= cpu_to_be32(CLIENT_MIGRATED);
1284
1285 strscpy(hostdata->caps.name, dev_name(&hostdata->host->shost_gendev),
1286 sizeof(hostdata->caps.name));
1287
1288 location = of_get_property(of_node, "ibm,loc-code", NULL);
1289 location = location ? location : dev_name(hostdata->dev);
1290 strscpy(hostdata->caps.loc, location, sizeof(hostdata->caps.loc));
1291
1292 req->common.type = cpu_to_be32(VIOSRP_CAPABILITIES_TYPE);
1293 req->buffer = cpu_to_be64(hostdata->caps_addr);
1294
1295 hostdata->caps.migration.common.cap_type =
1296 cpu_to_be32(MIGRATION_CAPABILITIES);
1297 hostdata->caps.migration.common.length =
1298 cpu_to_be16(sizeof(hostdata->caps.migration));
1299 hostdata->caps.migration.common.server_support =
1300 cpu_to_be16(SERVER_SUPPORTS_CAP);
1301 hostdata->caps.migration.ecl = cpu_to_be32(1);
1302
1303 if (client_reserve) {
1304 hostdata->caps.reserve.common.cap_type =
1305 cpu_to_be32(RESERVATION_CAPABILITIES);
1306 hostdata->caps.reserve.common.length =
1307 cpu_to_be16(sizeof(hostdata->caps.reserve));
1308 hostdata->caps.reserve.common.server_support =
1309 cpu_to_be16(SERVER_SUPPORTS_CAP);
1310 hostdata->caps.reserve.type =
1311 cpu_to_be32(CLIENT_RESERVE_SCSI_2);
1312 req->common.length =
1313 cpu_to_be16(sizeof(hostdata->caps));
1314 } else
1315 req->common.length = cpu_to_be16(sizeof(hostdata->caps) -
1316 sizeof(hostdata->caps.reserve));
1317
1318 spin_lock_irqsave(hostdata->host->host_lock, flags);
1319 if (ibmvscsi_send_srp_event(evt_struct, hostdata, info_timeout * 2))
1320 dev_err(hostdata->dev, "couldn't send CAPABILITIES_REQ!\n");
1321 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1322};
1323
1324/**
1325 * fast_fail_rsp: - Handle response to MAD enable fast fail
1326 * @evt_struct: srp_event_struct with the response
1327 *
1328 * Used as a "done" callback by when sending enable fast fail. Gets called
1329 * by ibmvscsi_handle_crq()
1330 */
1331static void fast_fail_rsp(struct srp_event_struct *evt_struct)
1332{
1333 struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
1334 u16 status = be16_to_cpu(evt_struct->xfer_iu->mad.fast_fail.common.status);
1335
1336 if (status == VIOSRP_MAD_NOT_SUPPORTED)
1337 dev_err(hostdata->dev, "fast_fail not supported in server\n");
1338 else if (status == VIOSRP_MAD_FAILED)
1339 dev_err(hostdata->dev, "fast_fail request failed\n");
1340 else if (status != VIOSRP_MAD_SUCCESS)
1341 dev_err(hostdata->dev, "error 0x%X enabling fast_fail\n", status);
1342
1343 send_mad_capabilities(hostdata);
1344}
1345
1346/**
1347 * enable_fast_fail() - Start host initialization
1348 * @hostdata: ibmvscsi_host_data of host
1349 *
1350 * Returns zero if successful.
1351 */
1352static int enable_fast_fail(struct ibmvscsi_host_data *hostdata)
1353{
1354 int rc;
1355 unsigned long flags;
1356 struct viosrp_fast_fail *fast_fail_mad;
1357 struct srp_event_struct *evt_struct;
1358
1359 if (!fast_fail) {
1360 send_mad_capabilities(hostdata);
1361 return 0;
1362 }
1363
1364 evt_struct = get_event_struct(&hostdata->pool);
1365 BUG_ON(!evt_struct);
1366
1367 init_event_struct(evt_struct, fast_fail_rsp, VIOSRP_MAD_FORMAT, info_timeout);
1368
1369 fast_fail_mad = &evt_struct->iu.mad.fast_fail;
1370 memset(fast_fail_mad, 0, sizeof(*fast_fail_mad));
1371 fast_fail_mad->common.type = cpu_to_be32(VIOSRP_ENABLE_FAST_FAIL);
1372 fast_fail_mad->common.length = cpu_to_be16(sizeof(*fast_fail_mad));
1373
1374 spin_lock_irqsave(hostdata->host->host_lock, flags);
1375 rc = ibmvscsi_send_srp_event(evt_struct, hostdata, info_timeout * 2);
1376 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1377 return rc;
1378}
1379
1380/**
1381 * adapter_info_rsp: - Handle response to MAD adapter info request
1382 * @evt_struct: srp_event_struct with the response
1383 *
1384 * Used as a "done" callback by when sending adapter_info. Gets called
1385 * by ibmvscsi_handle_crq()
1386*/
1387static void adapter_info_rsp(struct srp_event_struct *evt_struct)
1388{
1389 struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
1390
1391 if (evt_struct->xfer_iu->mad.adapter_info.common.status) {
1392 dev_err(hostdata->dev, "error %d getting adapter info\n",
1393 evt_struct->xfer_iu->mad.adapter_info.common.status);
1394 } else {
1395 dev_info(hostdata->dev, "host srp version: %s, "
1396 "host partition %s (%d), OS %d, max io %u\n",
1397 hostdata->madapter_info.srp_version,
1398 hostdata->madapter_info.partition_name,
1399 be32_to_cpu(hostdata->madapter_info.partition_number),
1400 be32_to_cpu(hostdata->madapter_info.os_type),
1401 be32_to_cpu(hostdata->madapter_info.port_max_txu[0]));
1402
1403 if (hostdata->madapter_info.port_max_txu[0])
1404 hostdata->host->max_sectors =
1405 be32_to_cpu(hostdata->madapter_info.port_max_txu[0]) >> 9;
1406
1407 if (be32_to_cpu(hostdata->madapter_info.os_type) == SRP_MAD_OS_AIX &&
1408 strcmp(hostdata->madapter_info.srp_version, "1.6a") <= 0) {
1409 dev_err(hostdata->dev, "host (Ver. %s) doesn't support large transfers\n",
1410 hostdata->madapter_info.srp_version);
1411 dev_err(hostdata->dev, "limiting scatterlists to %d\n",
1412 MAX_INDIRECT_BUFS);
1413 hostdata->host->sg_tablesize = MAX_INDIRECT_BUFS;
1414 }
1415
1416 if (be32_to_cpu(hostdata->madapter_info.os_type) == SRP_MAD_OS_AIX) {
1417 enable_fast_fail(hostdata);
1418 return;
1419 }
1420 }
1421
1422 send_srp_login(hostdata);
1423}
1424
1425/**
1426 * send_mad_adapter_info: - Sends the mad adapter info request
1427 * and stores the result so it can be retrieved with
1428 * sysfs. We COULD consider causing a failure if the
1429 * returned SRP version doesn't match ours.
1430 * @hostdata: ibmvscsi_host_data of host
1431 *
1432 * Returns zero if successful.
1433*/
1434static void send_mad_adapter_info(struct ibmvscsi_host_data *hostdata)
1435{
1436 struct viosrp_adapter_info *req;
1437 struct srp_event_struct *evt_struct;
1438 unsigned long flags;
1439
1440 evt_struct = get_event_struct(&hostdata->pool);
1441 BUG_ON(!evt_struct);
1442
1443 init_event_struct(evt_struct,
1444 adapter_info_rsp,
1445 VIOSRP_MAD_FORMAT,
1446 info_timeout);
1447
1448 req = &evt_struct->iu.mad.adapter_info;
1449 memset(req, 0x00, sizeof(*req));
1450
1451 req->common.type = cpu_to_be32(VIOSRP_ADAPTER_INFO_TYPE);
1452 req->common.length = cpu_to_be16(sizeof(hostdata->madapter_info));
1453 req->buffer = cpu_to_be64(hostdata->adapter_info_addr);
1454
1455 spin_lock_irqsave(hostdata->host->host_lock, flags);
1456 if (ibmvscsi_send_srp_event(evt_struct, hostdata, info_timeout * 2))
1457 dev_err(hostdata->dev, "couldn't send ADAPTER_INFO_REQ!\n");
1458 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1459};
1460
1461/*
1462 * init_adapter() - Start virtual adapter initialization sequence
1463 */
1464static void init_adapter(struct ibmvscsi_host_data *hostdata)
1465{
1466 send_mad_adapter_info(hostdata);
1467}
1468
1469/*
1470 * sync_completion: Signal that a synchronous command has completed
1471 * Note that after returning from this call, the evt_struct is freed.
1472 * the caller waiting on this completion shouldn't touch the evt_struct
1473 * again.
1474 */
1475static void sync_completion(struct srp_event_struct *evt_struct)
1476{
1477 /* copy the response back */
1478 if (evt_struct->sync_srp)
1479 *evt_struct->sync_srp = *evt_struct->xfer_iu;
1480
1481 complete(&evt_struct->comp);
1482}
1483
1484/*
1485 * ibmvscsi_eh_abort_handler: Abort a command...from scsi host template
1486 * send this over to the server and wait synchronously for the response
1487 */
1488static int ibmvscsi_eh_abort_handler(struct scsi_cmnd *cmd)
1489{
1490 struct ibmvscsi_host_data *hostdata = shost_priv(cmd->device->host);
1491 struct srp_tsk_mgmt *tsk_mgmt;
1492 struct srp_event_struct *evt;
1493 struct srp_event_struct *tmp_evt, *found_evt;
1494 union viosrp_iu srp_rsp;
1495 int rsp_rc;
1496 unsigned long flags;
1497 u16 lun = lun_from_dev(cmd->device);
1498 unsigned long wait_switch = 0;
1499
1500 /* First, find this command in our sent list so we can figure
1501 * out the correct tag
1502 */
1503 spin_lock_irqsave(hostdata->host->host_lock, flags);
1504 wait_switch = jiffies + (init_timeout * HZ);
1505 do {
1506 found_evt = NULL;
1507 list_for_each_entry(tmp_evt, &hostdata->sent, list) {
1508 if (tmp_evt->cmnd == cmd) {
1509 found_evt = tmp_evt;
1510 break;
1511 }
1512 }
1513
1514 if (!found_evt) {
1515 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1516 return SUCCESS;
1517 }
1518
1519 evt = get_event_struct(&hostdata->pool);
1520 if (evt == NULL) {
1521 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1522 sdev_printk(KERN_ERR, cmd->device,
1523 "failed to allocate abort event\n");
1524 return FAILED;
1525 }
1526
1527 init_event_struct(evt,
1528 sync_completion,
1529 VIOSRP_SRP_FORMAT,
1530 abort_timeout);
1531
1532 tsk_mgmt = &evt->iu.srp.tsk_mgmt;
1533
1534 /* Set up an abort SRP command */
1535 memset(tsk_mgmt, 0x00, sizeof(*tsk_mgmt));
1536 tsk_mgmt->opcode = SRP_TSK_MGMT;
1537 int_to_scsilun(lun, &tsk_mgmt->lun);
1538 tsk_mgmt->tsk_mgmt_func = SRP_TSK_ABORT_TASK;
1539 tsk_mgmt->task_tag = (u64) found_evt;
1540
1541 evt->sync_srp = &srp_rsp;
1542
1543 init_completion(&evt->comp);
1544 rsp_rc = ibmvscsi_send_srp_event(evt, hostdata, abort_timeout * 2);
1545
1546 if (rsp_rc != SCSI_MLQUEUE_HOST_BUSY)
1547 break;
1548
1549 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1550 msleep(10);
1551 spin_lock_irqsave(hostdata->host->host_lock, flags);
1552 } while (time_before(jiffies, wait_switch));
1553
1554 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1555
1556 if (rsp_rc != 0) {
1557 sdev_printk(KERN_ERR, cmd->device,
1558 "failed to send abort() event. rc=%d\n", rsp_rc);
1559 return FAILED;
1560 }
1561
1562 sdev_printk(KERN_INFO, cmd->device,
1563 "aborting command. lun 0x%llx, tag 0x%llx\n",
1564 (((u64) lun) << 48), (u64) found_evt);
1565
1566 wait_for_completion(&evt->comp);
1567
1568 /* make sure we got a good response */
1569 if (unlikely(srp_rsp.srp.rsp.opcode != SRP_RSP)) {
1570 if (printk_ratelimit())
1571 sdev_printk(KERN_WARNING, cmd->device, "abort bad SRP RSP type %d\n",
1572 srp_rsp.srp.rsp.opcode);
1573 return FAILED;
1574 }
1575
1576 if (srp_rsp.srp.rsp.flags & SRP_RSP_FLAG_RSPVALID)
1577 rsp_rc = *((int *)srp_rsp.srp.rsp.data);
1578 else
1579 rsp_rc = srp_rsp.srp.rsp.status;
1580
1581 if (rsp_rc) {
1582 if (printk_ratelimit())
1583 sdev_printk(KERN_WARNING, cmd->device,
1584 "abort code %d for task tag 0x%llx\n",
1585 rsp_rc, tsk_mgmt->task_tag);
1586 return FAILED;
1587 }
1588
1589 /* Because we dropped the spinlock above, it's possible
1590 * The event is no longer in our list. Make sure it didn't
1591 * complete while we were aborting
1592 */
1593 spin_lock_irqsave(hostdata->host->host_lock, flags);
1594 found_evt = NULL;
1595 list_for_each_entry(tmp_evt, &hostdata->sent, list) {
1596 if (tmp_evt->cmnd == cmd) {
1597 found_evt = tmp_evt;
1598 break;
1599 }
1600 }
1601
1602 if (found_evt == NULL) {
1603 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1604 sdev_printk(KERN_INFO, cmd->device, "aborted task tag 0x%llx completed\n",
1605 tsk_mgmt->task_tag);
1606 return SUCCESS;
1607 }
1608
1609 sdev_printk(KERN_INFO, cmd->device, "successfully aborted task tag 0x%llx\n",
1610 tsk_mgmt->task_tag);
1611
1612 cmd->result = (DID_ABORT << 16);
1613 list_del(&found_evt->list);
1614 unmap_cmd_data(&found_evt->iu.srp.cmd, found_evt,
1615 found_evt->hostdata->dev);
1616 free_event_struct(&found_evt->hostdata->pool, found_evt);
1617 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1618 atomic_inc(&hostdata->request_limit);
1619 return SUCCESS;
1620}
1621
1622/*
1623 * ibmvscsi_eh_device_reset_handler: Reset a single LUN...from scsi host
1624 * template send this over to the server and wait synchronously for the
1625 * response
1626 */
1627static int ibmvscsi_eh_device_reset_handler(struct scsi_cmnd *cmd)
1628{
1629 struct ibmvscsi_host_data *hostdata = shost_priv(cmd->device->host);
1630 struct srp_tsk_mgmt *tsk_mgmt;
1631 struct srp_event_struct *evt;
1632 struct srp_event_struct *tmp_evt, *pos;
1633 union viosrp_iu srp_rsp;
1634 int rsp_rc;
1635 unsigned long flags;
1636 u16 lun = lun_from_dev(cmd->device);
1637 unsigned long wait_switch = 0;
1638
1639 spin_lock_irqsave(hostdata->host->host_lock, flags);
1640 wait_switch = jiffies + (init_timeout * HZ);
1641 do {
1642 evt = get_event_struct(&hostdata->pool);
1643 if (evt == NULL) {
1644 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1645 sdev_printk(KERN_ERR, cmd->device,
1646 "failed to allocate reset event\n");
1647 return FAILED;
1648 }
1649
1650 init_event_struct(evt,
1651 sync_completion,
1652 VIOSRP_SRP_FORMAT,
1653 reset_timeout);
1654
1655 tsk_mgmt = &evt->iu.srp.tsk_mgmt;
1656
1657 /* Set up a lun reset SRP command */
1658 memset(tsk_mgmt, 0x00, sizeof(*tsk_mgmt));
1659 tsk_mgmt->opcode = SRP_TSK_MGMT;
1660 int_to_scsilun(lun, &tsk_mgmt->lun);
1661 tsk_mgmt->tsk_mgmt_func = SRP_TSK_LUN_RESET;
1662
1663 evt->sync_srp = &srp_rsp;
1664
1665 init_completion(&evt->comp);
1666 rsp_rc = ibmvscsi_send_srp_event(evt, hostdata, reset_timeout * 2);
1667
1668 if (rsp_rc != SCSI_MLQUEUE_HOST_BUSY)
1669 break;
1670
1671 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1672 msleep(10);
1673 spin_lock_irqsave(hostdata->host->host_lock, flags);
1674 } while (time_before(jiffies, wait_switch));
1675
1676 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1677
1678 if (rsp_rc != 0) {
1679 sdev_printk(KERN_ERR, cmd->device,
1680 "failed to send reset event. rc=%d\n", rsp_rc);
1681 return FAILED;
1682 }
1683
1684 sdev_printk(KERN_INFO, cmd->device, "resetting device. lun 0x%llx\n",
1685 (((u64) lun) << 48));
1686
1687 wait_for_completion(&evt->comp);
1688
1689 /* make sure we got a good response */
1690 if (unlikely(srp_rsp.srp.rsp.opcode != SRP_RSP)) {
1691 if (printk_ratelimit())
1692 sdev_printk(KERN_WARNING, cmd->device, "reset bad SRP RSP type %d\n",
1693 srp_rsp.srp.rsp.opcode);
1694 return FAILED;
1695 }
1696
1697 if (srp_rsp.srp.rsp.flags & SRP_RSP_FLAG_RSPVALID)
1698 rsp_rc = *((int *)srp_rsp.srp.rsp.data);
1699 else
1700 rsp_rc = srp_rsp.srp.rsp.status;
1701
1702 if (rsp_rc) {
1703 if (printk_ratelimit())
1704 sdev_printk(KERN_WARNING, cmd->device,
1705 "reset code %d for task tag 0x%llx\n",
1706 rsp_rc, tsk_mgmt->task_tag);
1707 return FAILED;
1708 }
1709
1710 /* We need to find all commands for this LUN that have not yet been
1711 * responded to, and fail them with DID_RESET
1712 */
1713 spin_lock_irqsave(hostdata->host->host_lock, flags);
1714 list_for_each_entry_safe(tmp_evt, pos, &hostdata->sent, list) {
1715 if ((tmp_evt->cmnd) && (tmp_evt->cmnd->device == cmd->device)) {
1716 if (tmp_evt->cmnd)
1717 tmp_evt->cmnd->result = (DID_RESET << 16);
1718 list_del(&tmp_evt->list);
1719 unmap_cmd_data(&tmp_evt->iu.srp.cmd, tmp_evt,
1720 tmp_evt->hostdata->dev);
1721 free_event_struct(&tmp_evt->hostdata->pool,
1722 tmp_evt);
1723 atomic_inc(&hostdata->request_limit);
1724 if (tmp_evt->cmnd_done)
1725 tmp_evt->cmnd_done(tmp_evt->cmnd);
1726 else if (tmp_evt->done)
1727 tmp_evt->done(tmp_evt);
1728 }
1729 }
1730 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1731 return SUCCESS;
1732}
1733
1734/**
1735 * ibmvscsi_eh_host_reset_handler - Reset the connection to the server
1736 * @cmd: struct scsi_cmnd having problems
1737*/
1738static int ibmvscsi_eh_host_reset_handler(struct scsi_cmnd *cmd)
1739{
1740 unsigned long wait_switch = 0;
1741 struct ibmvscsi_host_data *hostdata = shost_priv(cmd->device->host);
1742
1743 dev_err(hostdata->dev, "Resetting connection due to error recovery\n");
1744
1745 ibmvscsi_reset_host(hostdata);
1746
1747 for (wait_switch = jiffies + (init_timeout * HZ);
1748 time_before(jiffies, wait_switch) &&
1749 atomic_read(&hostdata->request_limit) < 2;) {
1750
1751 msleep(10);
1752 }
1753
1754 if (atomic_read(&hostdata->request_limit) <= 0)
1755 return FAILED;
1756
1757 return SUCCESS;
1758}
1759
1760/**
1761 * ibmvscsi_handle_crq: - Handles and frees received events in the CRQ
1762 * @crq: Command/Response queue
1763 * @hostdata: ibmvscsi_host_data of host
1764 *
1765*/
1766static void ibmvscsi_handle_crq(struct viosrp_crq *crq,
1767 struct ibmvscsi_host_data *hostdata)
1768{
1769 long rc;
1770 unsigned long flags;
1771 /* The hypervisor copies our tag value here so no byteswapping */
1772 struct srp_event_struct *evt_struct =
1773 (__force struct srp_event_struct *)crq->IU_data_ptr;
1774 switch (crq->valid) {
1775 case VIOSRP_CRQ_INIT_RSP: /* initialization */
1776 switch (crq->format) {
1777 case VIOSRP_CRQ_INIT: /* Initialization message */
1778 dev_info(hostdata->dev, "partner initialized\n");
1779 /* Send back a response */
1780 rc = ibmvscsi_send_crq(hostdata, 0xC002000000000000LL, 0);
1781 if (rc == 0) {
1782 /* Now login */
1783 init_adapter(hostdata);
1784 } else {
1785 dev_err(hostdata->dev, "Unable to send init rsp. rc=%ld\n", rc);
1786 }
1787
1788 break;
1789 case VIOSRP_CRQ_INIT_COMPLETE: /* Initialization response */
1790 dev_info(hostdata->dev, "partner initialization complete\n");
1791
1792 /* Now login */
1793 init_adapter(hostdata);
1794 break;
1795 default:
1796 dev_err(hostdata->dev, "unknown crq message type: %d\n", crq->format);
1797 }
1798 return;
1799 case VIOSRP_CRQ_XPORT_EVENT: /* Hypervisor telling us the connection is closed */
1800 scsi_block_requests(hostdata->host);
1801 ibmvscsi_set_request_limit(hostdata, 0);
1802 if (crq->format == 0x06) {
1803 /* We need to re-setup the interpartition connection */
1804 dev_info(hostdata->dev, "Re-enabling adapter!\n");
1805 hostdata->client_migrated = 1;
1806 hostdata->action = IBMVSCSI_HOST_ACTION_REENABLE;
1807 purge_requests(hostdata, DID_REQUEUE);
1808 wake_up(&hostdata->work_wait_q);
1809 } else {
1810 dev_err(hostdata->dev, "Virtual adapter failed rc %d!\n",
1811 crq->format);
1812 ibmvscsi_reset_host(hostdata);
1813 }
1814 return;
1815 case VIOSRP_CRQ_CMD_RSP: /* real payload */
1816 break;
1817 default:
1818 dev_err(hostdata->dev, "got an invalid message type 0x%02x\n",
1819 crq->valid);
1820 return;
1821 }
1822
1823 /* The only kind of payload CRQs we should get are responses to
1824 * things we send. Make sure this response is to something we
1825 * actually sent
1826 */
1827 if (!valid_event_struct(&hostdata->pool, evt_struct)) {
1828 dev_err(hostdata->dev, "returned correlation_token 0x%p is invalid!\n",
1829 evt_struct);
1830 return;
1831 }
1832
1833 if (atomic_read(&evt_struct->free)) {
1834 dev_err(hostdata->dev, "received duplicate correlation_token 0x%p!\n",
1835 evt_struct);
1836 return;
1837 }
1838
1839 if (crq->format == VIOSRP_SRP_FORMAT)
1840 atomic_add(be32_to_cpu(evt_struct->xfer_iu->srp.rsp.req_lim_delta),
1841 &hostdata->request_limit);
1842
1843 del_timer(&evt_struct->timer);
1844
1845 if ((crq->status != VIOSRP_OK && crq->status != VIOSRP_OK2) && evt_struct->cmnd)
1846 evt_struct->cmnd->result = DID_ERROR << 16;
1847 if (evt_struct->done)
1848 evt_struct->done(evt_struct);
1849 else
1850 dev_err(hostdata->dev, "returned done() is NULL; not running it!\n");
1851
1852 /*
1853 * Lock the host_lock before messing with these structures, since we
1854 * are running in a task context
1855 */
1856 spin_lock_irqsave(evt_struct->hostdata->host->host_lock, flags);
1857 list_del(&evt_struct->list);
1858 free_event_struct(&evt_struct->hostdata->pool, evt_struct);
1859 spin_unlock_irqrestore(evt_struct->hostdata->host->host_lock, flags);
1860}
1861
1862/**
1863 * ibmvscsi_slave_configure: Set the "allow_restart" flag for each disk.
1864 * @sdev: struct scsi_device device to configure
1865 *
1866 * Enable allow_restart for a device if it is a disk. Adjust the
1867 * queue_depth here also as is required by the documentation for
1868 * struct scsi_host_template.
1869 */
1870static int ibmvscsi_slave_configure(struct scsi_device *sdev)
1871{
1872 struct Scsi_Host *shost = sdev->host;
1873 unsigned long lock_flags = 0;
1874
1875 spin_lock_irqsave(shost->host_lock, lock_flags);
1876 if (sdev->type == TYPE_DISK) {
1877 sdev->allow_restart = 1;
1878 blk_queue_rq_timeout(sdev->request_queue, 120 * HZ);
1879 }
1880 spin_unlock_irqrestore(shost->host_lock, lock_flags);
1881 return 0;
1882}
1883
1884/**
1885 * ibmvscsi_change_queue_depth - Change the device's queue depth
1886 * @sdev: scsi device struct
1887 * @qdepth: depth to set
1888 *
1889 * Return value:
1890 * actual depth set
1891 **/
1892static int ibmvscsi_change_queue_depth(struct scsi_device *sdev, int qdepth)
1893{
1894 if (qdepth > IBMVSCSI_MAX_CMDS_PER_LUN)
1895 qdepth = IBMVSCSI_MAX_CMDS_PER_LUN;
1896 return scsi_change_queue_depth(sdev, qdepth);
1897}
1898
1899/* ------------------------------------------------------------
1900 * sysfs attributes
1901 */
1902static ssize_t show_host_vhost_loc(struct device *dev,
1903 struct device_attribute *attr, char *buf)
1904{
1905 struct Scsi_Host *shost = class_to_shost(dev);
1906 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1907 int len;
1908
1909 len = snprintf(buf, sizeof(hostdata->caps.loc), "%s\n",
1910 hostdata->caps.loc);
1911 return len;
1912}
1913
1914static struct device_attribute ibmvscsi_host_vhost_loc = {
1915 .attr = {
1916 .name = "vhost_loc",
1917 .mode = S_IRUGO,
1918 },
1919 .show = show_host_vhost_loc,
1920};
1921
1922static ssize_t show_host_vhost_name(struct device *dev,
1923 struct device_attribute *attr, char *buf)
1924{
1925 struct Scsi_Host *shost = class_to_shost(dev);
1926 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1927 int len;
1928
1929 len = snprintf(buf, sizeof(hostdata->caps.name), "%s\n",
1930 hostdata->caps.name);
1931 return len;
1932}
1933
1934static struct device_attribute ibmvscsi_host_vhost_name = {
1935 .attr = {
1936 .name = "vhost_name",
1937 .mode = S_IRUGO,
1938 },
1939 .show = show_host_vhost_name,
1940};
1941
1942static ssize_t show_host_srp_version(struct device *dev,
1943 struct device_attribute *attr, char *buf)
1944{
1945 struct Scsi_Host *shost = class_to_shost(dev);
1946 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1947 int len;
1948
1949 len = snprintf(buf, PAGE_SIZE, "%s\n",
1950 hostdata->madapter_info.srp_version);
1951 return len;
1952}
1953
1954static struct device_attribute ibmvscsi_host_srp_version = {
1955 .attr = {
1956 .name = "srp_version",
1957 .mode = S_IRUGO,
1958 },
1959 .show = show_host_srp_version,
1960};
1961
1962static ssize_t show_host_partition_name(struct device *dev,
1963 struct device_attribute *attr,
1964 char *buf)
1965{
1966 struct Scsi_Host *shost = class_to_shost(dev);
1967 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1968 int len;
1969
1970 len = snprintf(buf, PAGE_SIZE, "%s\n",
1971 hostdata->madapter_info.partition_name);
1972 return len;
1973}
1974
1975static struct device_attribute ibmvscsi_host_partition_name = {
1976 .attr = {
1977 .name = "partition_name",
1978 .mode = S_IRUGO,
1979 },
1980 .show = show_host_partition_name,
1981};
1982
1983static ssize_t show_host_partition_number(struct device *dev,
1984 struct device_attribute *attr,
1985 char *buf)
1986{
1987 struct Scsi_Host *shost = class_to_shost(dev);
1988 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1989 int len;
1990
1991 len = snprintf(buf, PAGE_SIZE, "%d\n",
1992 be32_to_cpu(hostdata->madapter_info.partition_number));
1993 return len;
1994}
1995
1996static struct device_attribute ibmvscsi_host_partition_number = {
1997 .attr = {
1998 .name = "partition_number",
1999 .mode = S_IRUGO,
2000 },
2001 .show = show_host_partition_number,
2002};
2003
2004static ssize_t show_host_mad_version(struct device *dev,
2005 struct device_attribute *attr, char *buf)
2006{
2007 struct Scsi_Host *shost = class_to_shost(dev);
2008 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
2009 int len;
2010
2011 len = snprintf(buf, PAGE_SIZE, "%d\n",
2012 be32_to_cpu(hostdata->madapter_info.mad_version));
2013 return len;
2014}
2015
2016static struct device_attribute ibmvscsi_host_mad_version = {
2017 .attr = {
2018 .name = "mad_version",
2019 .mode = S_IRUGO,
2020 },
2021 .show = show_host_mad_version,
2022};
2023
2024static ssize_t show_host_os_type(struct device *dev,
2025 struct device_attribute *attr, char *buf)
2026{
2027 struct Scsi_Host *shost = class_to_shost(dev);
2028 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
2029 int len;
2030
2031 len = snprintf(buf, PAGE_SIZE, "%d\n",
2032 be32_to_cpu(hostdata->madapter_info.os_type));
2033 return len;
2034}
2035
2036static struct device_attribute ibmvscsi_host_os_type = {
2037 .attr = {
2038 .name = "os_type",
2039 .mode = S_IRUGO,
2040 },
2041 .show = show_host_os_type,
2042};
2043
2044static ssize_t show_host_config(struct device *dev,
2045 struct device_attribute *attr, char *buf)
2046{
2047 return 0;
2048}
2049
2050static struct device_attribute ibmvscsi_host_config = {
2051 .attr = {
2052 .name = "config",
2053 .mode = S_IRUGO,
2054 },
2055 .show = show_host_config,
2056};
2057
2058static int ibmvscsi_host_reset(struct Scsi_Host *shost, int reset_type)
2059{
2060 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
2061
2062 dev_info(hostdata->dev, "Initiating adapter reset!\n");
2063 ibmvscsi_reset_host(hostdata);
2064
2065 return 0;
2066}
2067
2068static struct attribute *ibmvscsi_host_attrs[] = {
2069 &ibmvscsi_host_vhost_loc.attr,
2070 &ibmvscsi_host_vhost_name.attr,
2071 &ibmvscsi_host_srp_version.attr,
2072 &ibmvscsi_host_partition_name.attr,
2073 &ibmvscsi_host_partition_number.attr,
2074 &ibmvscsi_host_mad_version.attr,
2075 &ibmvscsi_host_os_type.attr,
2076 &ibmvscsi_host_config.attr,
2077 NULL
2078};
2079
2080ATTRIBUTE_GROUPS(ibmvscsi_host);
2081
2082/* ------------------------------------------------------------
2083 * SCSI driver registration
2084 */
2085static struct scsi_host_template driver_template = {
2086 .module = THIS_MODULE,
2087 .name = "IBM POWER Virtual SCSI Adapter " IBMVSCSI_VERSION,
2088 .proc_name = "ibmvscsi",
2089 .queuecommand = ibmvscsi_queuecommand,
2090 .eh_timed_out = srp_timed_out,
2091 .eh_abort_handler = ibmvscsi_eh_abort_handler,
2092 .eh_device_reset_handler = ibmvscsi_eh_device_reset_handler,
2093 .eh_host_reset_handler = ibmvscsi_eh_host_reset_handler,
2094 .slave_configure = ibmvscsi_slave_configure,
2095 .change_queue_depth = ibmvscsi_change_queue_depth,
2096 .host_reset = ibmvscsi_host_reset,
2097 .cmd_per_lun = IBMVSCSI_CMDS_PER_LUN_DEFAULT,
2098 .can_queue = IBMVSCSI_MAX_REQUESTS_DEFAULT,
2099 .this_id = -1,
2100 .sg_tablesize = SG_ALL,
2101 .shost_groups = ibmvscsi_host_groups,
2102};
2103
2104/**
2105 * ibmvscsi_get_desired_dma - Calculate IO memory desired by the driver
2106 *
2107 * @vdev: struct vio_dev for the device whose desired IO mem is to be returned
2108 *
2109 * Return value:
2110 * Number of bytes of IO data the driver will need to perform well.
2111 */
2112static unsigned long ibmvscsi_get_desired_dma(struct vio_dev *vdev)
2113{
2114 /* iu_storage data allocated in initialize_event_pool */
2115 unsigned long desired_io = max_events * sizeof(union viosrp_iu);
2116
2117 /* add io space for sg data */
2118 desired_io += (IBMVSCSI_MAX_SECTORS_DEFAULT * 512 *
2119 IBMVSCSI_CMDS_PER_LUN_DEFAULT);
2120
2121 return desired_io;
2122}
2123
2124static void ibmvscsi_do_work(struct ibmvscsi_host_data *hostdata)
2125{
2126 unsigned long flags;
2127 int rc;
2128 char *action = "reset";
2129
2130 spin_lock_irqsave(hostdata->host->host_lock, flags);
2131 switch (hostdata->action) {
2132 case IBMVSCSI_HOST_ACTION_UNBLOCK:
2133 rc = 0;
2134 break;
2135 case IBMVSCSI_HOST_ACTION_RESET:
2136 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
2137 rc = ibmvscsi_reset_crq_queue(&hostdata->queue, hostdata);
2138 spin_lock_irqsave(hostdata->host->host_lock, flags);
2139 if (!rc)
2140 rc = ibmvscsi_send_crq(hostdata, 0xC001000000000000LL, 0);
2141 vio_enable_interrupts(to_vio_dev(hostdata->dev));
2142 break;
2143 case IBMVSCSI_HOST_ACTION_REENABLE:
2144 action = "enable";
2145 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
2146 rc = ibmvscsi_reenable_crq_queue(&hostdata->queue, hostdata);
2147 spin_lock_irqsave(hostdata->host->host_lock, flags);
2148 if (!rc)
2149 rc = ibmvscsi_send_crq(hostdata, 0xC001000000000000LL, 0);
2150 break;
2151 case IBMVSCSI_HOST_ACTION_NONE:
2152 default:
2153 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
2154 return;
2155 }
2156
2157 hostdata->action = IBMVSCSI_HOST_ACTION_NONE;
2158 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
2159
2160 if (rc) {
2161 ibmvscsi_set_request_limit(hostdata, -1);
2162 dev_err(hostdata->dev, "error after %s\n", action);
2163 }
2164
2165 scsi_unblock_requests(hostdata->host);
2166}
2167
2168static int __ibmvscsi_work_to_do(struct ibmvscsi_host_data *hostdata)
2169{
2170 if (kthread_should_stop())
2171 return 1;
2172 switch (hostdata->action) {
2173 case IBMVSCSI_HOST_ACTION_NONE:
2174 return 0;
2175 case IBMVSCSI_HOST_ACTION_RESET:
2176 case IBMVSCSI_HOST_ACTION_REENABLE:
2177 case IBMVSCSI_HOST_ACTION_UNBLOCK:
2178 default:
2179 break;
2180 }
2181
2182 return 1;
2183}
2184
2185static int ibmvscsi_work_to_do(struct ibmvscsi_host_data *hostdata)
2186{
2187 unsigned long flags;
2188 int rc;
2189
2190 spin_lock_irqsave(hostdata->host->host_lock, flags);
2191 rc = __ibmvscsi_work_to_do(hostdata);
2192 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
2193
2194 return rc;
2195}
2196
2197static int ibmvscsi_work(void *data)
2198{
2199 struct ibmvscsi_host_data *hostdata = data;
2200 int rc;
2201
2202 set_user_nice(current, MIN_NICE);
2203
2204 while (1) {
2205 rc = wait_event_interruptible(hostdata->work_wait_q,
2206 ibmvscsi_work_to_do(hostdata));
2207
2208 BUG_ON(rc);
2209
2210 if (kthread_should_stop())
2211 break;
2212
2213 ibmvscsi_do_work(hostdata);
2214 }
2215
2216 return 0;
2217}
2218
2219/*
2220 * Called by bus code for each adapter
2221 */
2222static int ibmvscsi_probe(struct vio_dev *vdev, const struct vio_device_id *id)
2223{
2224 struct ibmvscsi_host_data *hostdata;
2225 struct Scsi_Host *host;
2226 struct device *dev = &vdev->dev;
2227 struct srp_rport_identifiers ids;
2228 struct srp_rport *rport;
2229 unsigned long wait_switch = 0;
2230 int rc;
2231
2232 dev_set_drvdata(&vdev->dev, NULL);
2233
2234 host = scsi_host_alloc(&driver_template, sizeof(*hostdata));
2235 if (!host) {
2236 dev_err(&vdev->dev, "couldn't allocate host data\n");
2237 goto scsi_host_alloc_failed;
2238 }
2239
2240 host->transportt = ibmvscsi_transport_template;
2241 hostdata = shost_priv(host);
2242 memset(hostdata, 0x00, sizeof(*hostdata));
2243 INIT_LIST_HEAD(&hostdata->sent);
2244 init_waitqueue_head(&hostdata->work_wait_q);
2245 hostdata->host = host;
2246 hostdata->dev = dev;
2247 ibmvscsi_set_request_limit(hostdata, -1);
2248 hostdata->host->max_sectors = IBMVSCSI_MAX_SECTORS_DEFAULT;
2249
2250 if (map_persist_bufs(hostdata)) {
2251 dev_err(&vdev->dev, "couldn't map persistent buffers\n");
2252 goto persist_bufs_failed;
2253 }
2254
2255 hostdata->work_thread = kthread_run(ibmvscsi_work, hostdata, "%s_%d",
2256 "ibmvscsi", host->host_no);
2257
2258 if (IS_ERR(hostdata->work_thread)) {
2259 dev_err(&vdev->dev, "couldn't initialize kthread. rc=%ld\n",
2260 PTR_ERR(hostdata->work_thread));
2261 goto init_crq_failed;
2262 }
2263
2264 rc = ibmvscsi_init_crq_queue(&hostdata->queue, hostdata, max_events);
2265 if (rc != 0 && rc != H_RESOURCE) {
2266 dev_err(&vdev->dev, "couldn't initialize crq. rc=%d\n", rc);
2267 goto kill_kthread;
2268 }
2269 if (initialize_event_pool(&hostdata->pool, max_events, hostdata) != 0) {
2270 dev_err(&vdev->dev, "couldn't initialize event pool\n");
2271 goto init_pool_failed;
2272 }
2273
2274 host->max_lun = IBMVSCSI_MAX_LUN;
2275 host->max_id = max_id;
2276 host->max_channel = max_channel;
2277 host->max_cmd_len = 16;
2278
2279 dev_info(dev,
2280 "Maximum ID: %d Maximum LUN: %llu Maximum Channel: %d\n",
2281 host->max_id, host->max_lun, host->max_channel);
2282
2283 if (scsi_add_host(hostdata->host, hostdata->dev))
2284 goto add_host_failed;
2285
2286 /* we don't have a proper target_port_id so let's use the fake one */
2287 memcpy(ids.port_id, hostdata->madapter_info.partition_name,
2288 sizeof(ids.port_id));
2289 ids.roles = SRP_RPORT_ROLE_TARGET;
2290 rport = srp_rport_add(host, &ids);
2291 if (IS_ERR(rport))
2292 goto add_srp_port_failed;
2293
2294 /* Try to send an initialization message. Note that this is allowed
2295 * to fail if the other end is not acive. In that case we don't
2296 * want to scan
2297 */
2298 if (ibmvscsi_send_crq(hostdata, 0xC001000000000000LL, 0) == 0
2299 || rc == H_RESOURCE) {
2300 /*
2301 * Wait around max init_timeout secs for the adapter to finish
2302 * initializing. When we are done initializing, we will have a
2303 * valid request_limit. We don't want Linux scanning before
2304 * we are ready.
2305 */
2306 for (wait_switch = jiffies + (init_timeout * HZ);
2307 time_before(jiffies, wait_switch) &&
2308 atomic_read(&hostdata->request_limit) < 2;) {
2309
2310 msleep(10);
2311 }
2312
2313 /* if we now have a valid request_limit, initiate a scan */
2314 if (atomic_read(&hostdata->request_limit) > 0)
2315 scsi_scan_host(host);
2316 }
2317
2318 dev_set_drvdata(&vdev->dev, hostdata);
2319 spin_lock(&ibmvscsi_driver_lock);
2320 list_add_tail(&hostdata->host_list, &ibmvscsi_head);
2321 spin_unlock(&ibmvscsi_driver_lock);
2322 return 0;
2323
2324 add_srp_port_failed:
2325 scsi_remove_host(hostdata->host);
2326 add_host_failed:
2327 release_event_pool(&hostdata->pool, hostdata);
2328 init_pool_failed:
2329 ibmvscsi_release_crq_queue(&hostdata->queue, hostdata, max_events);
2330 kill_kthread:
2331 kthread_stop(hostdata->work_thread);
2332 init_crq_failed:
2333 unmap_persist_bufs(hostdata);
2334 persist_bufs_failed:
2335 scsi_host_put(host);
2336 scsi_host_alloc_failed:
2337 return -1;
2338}
2339
2340static void ibmvscsi_remove(struct vio_dev *vdev)
2341{
2342 struct ibmvscsi_host_data *hostdata = dev_get_drvdata(&vdev->dev);
2343
2344 srp_remove_host(hostdata->host);
2345 scsi_remove_host(hostdata->host);
2346
2347 purge_requests(hostdata, DID_ERROR);
2348 release_event_pool(&hostdata->pool, hostdata);
2349
2350 ibmvscsi_release_crq_queue(&hostdata->queue, hostdata,
2351 max_events);
2352
2353 kthread_stop(hostdata->work_thread);
2354 unmap_persist_bufs(hostdata);
2355
2356 spin_lock(&ibmvscsi_driver_lock);
2357 list_del(&hostdata->host_list);
2358 spin_unlock(&ibmvscsi_driver_lock);
2359
2360 scsi_host_put(hostdata->host);
2361}
2362
2363/**
2364 * ibmvscsi_resume: Resume from suspend
2365 * @dev: device struct
2366 *
2367 * We may have lost an interrupt across suspend/resume, so kick the
2368 * interrupt handler
2369 */
2370static int ibmvscsi_resume(struct device *dev)
2371{
2372 struct ibmvscsi_host_data *hostdata = dev_get_drvdata(dev);
2373 vio_disable_interrupts(to_vio_dev(hostdata->dev));
2374 tasklet_schedule(&hostdata->srp_task);
2375
2376 return 0;
2377}
2378
2379/*
2380 * ibmvscsi_device_table: Used by vio.c to match devices in the device tree we
2381 * support.
2382 */
2383static const struct vio_device_id ibmvscsi_device_table[] = {
2384 {"vscsi", "IBM,v-scsi"},
2385 { "", "" }
2386};
2387MODULE_DEVICE_TABLE(vio, ibmvscsi_device_table);
2388
2389static const struct dev_pm_ops ibmvscsi_pm_ops = {
2390 .resume = ibmvscsi_resume
2391};
2392
2393static struct vio_driver ibmvscsi_driver = {
2394 .id_table = ibmvscsi_device_table,
2395 .probe = ibmvscsi_probe,
2396 .remove = ibmvscsi_remove,
2397 .get_desired_dma = ibmvscsi_get_desired_dma,
2398 .name = "ibmvscsi",
2399 .pm = &ibmvscsi_pm_ops,
2400};
2401
2402static struct srp_function_template ibmvscsi_transport_functions = {
2403};
2404
2405static int __init ibmvscsi_module_init(void)
2406{
2407 int ret;
2408
2409 /* Ensure we have two requests to do error recovery */
2410 driver_template.can_queue = max_requests;
2411 max_events = max_requests + 2;
2412
2413 if (!firmware_has_feature(FW_FEATURE_VIO))
2414 return -ENODEV;
2415
2416 ibmvscsi_transport_template =
2417 srp_attach_transport(&ibmvscsi_transport_functions);
2418 if (!ibmvscsi_transport_template)
2419 return -ENOMEM;
2420
2421 ret = vio_register_driver(&ibmvscsi_driver);
2422 if (ret)
2423 srp_release_transport(ibmvscsi_transport_template);
2424 return ret;
2425}
2426
2427static void __exit ibmvscsi_module_exit(void)
2428{
2429 vio_unregister_driver(&ibmvscsi_driver);
2430 srp_release_transport(ibmvscsi_transport_template);
2431}
2432
2433module_init(ibmvscsi_module_init);
2434module_exit(ibmvscsi_module_exit);
1// SPDX-License-Identifier: GPL-2.0-or-later
2/* ------------------------------------------------------------
3 * ibmvscsi.c
4 * (C) Copyright IBM Corporation 1994, 2004
5 * Authors: Colin DeVilbiss (devilbis@us.ibm.com)
6 * Santiago Leon (santil@us.ibm.com)
7 * Dave Boutcher (sleddog@us.ibm.com)
8 *
9 * ------------------------------------------------------------
10 * Emulation of a SCSI host adapter for Virtual I/O devices
11 *
12 * This driver supports the SCSI adapter implemented by the IBM
13 * Power5 firmware. That SCSI adapter is not a physical adapter,
14 * but allows Linux SCSI peripheral drivers to directly
15 * access devices in another logical partition on the physical system.
16 *
17 * The virtual adapter(s) are present in the open firmware device
18 * tree just like real adapters.
19 *
20 * One of the capabilities provided on these systems is the ability
21 * to DMA between partitions. The architecture states that for VSCSI,
22 * the server side is allowed to DMA to and from the client. The client
23 * is never trusted to DMA to or from the server directly.
24 *
25 * Messages are sent between partitions on a "Command/Response Queue"
26 * (CRQ), which is just a buffer of 16 byte entries in the receiver's
27 * Senders cannot access the buffer directly, but send messages by
28 * making a hypervisor call and passing in the 16 bytes. The hypervisor
29 * puts the message in the next 16 byte space in round-robin fashion,
30 * turns on the high order bit of the message (the valid bit), and
31 * generates an interrupt to the receiver (if interrupts are turned on.)
32 * The receiver just turns off the valid bit when they have copied out
33 * the message.
34 *
35 * The VSCSI client builds a SCSI Remote Protocol (SRP) Information Unit
36 * (IU) (as defined in the T10 standard available at www.t10.org), gets
37 * a DMA address for the message, and sends it to the server as the
38 * payload of a CRQ message. The server DMAs the SRP IU and processes it,
39 * including doing any additional data transfers. When it is done, it
40 * DMAs the SRP response back to the same address as the request came from,
41 * and sends a CRQ message back to inform the client that the request has
42 * completed.
43 *
44 * TODO: This is currently pretty tied to the IBM pSeries hypervisor
45 * interfaces. It would be really nice to abstract this above an RDMA
46 * layer.
47 */
48
49#include <linux/module.h>
50#include <linux/moduleparam.h>
51#include <linux/dma-mapping.h>
52#include <linux/delay.h>
53#include <linux/slab.h>
54#include <linux/of.h>
55#include <linux/pm.h>
56#include <linux/kthread.h>
57#include <asm/firmware.h>
58#include <asm/vio.h>
59#include <scsi/scsi.h>
60#include <scsi/scsi_cmnd.h>
61#include <scsi/scsi_host.h>
62#include <scsi/scsi_device.h>
63#include <scsi/scsi_transport_srp.h>
64#include "ibmvscsi.h"
65
66/* The values below are somewhat arbitrary default values, but
67 * OS/400 will use 3 busses (disks, CDs, tapes, I think.)
68 * Note that there are 3 bits of channel value, 6 bits of id, and
69 * 5 bits of LUN.
70 */
71static int max_id = 64;
72static int max_channel = 3;
73static int init_timeout = 300;
74static int login_timeout = 60;
75static int info_timeout = 30;
76static int abort_timeout = 60;
77static int reset_timeout = 60;
78static int max_requests = IBMVSCSI_MAX_REQUESTS_DEFAULT;
79static int max_events = IBMVSCSI_MAX_REQUESTS_DEFAULT + 2;
80static int fast_fail = 1;
81static int client_reserve = 1;
82static char partition_name[96] = "UNKNOWN";
83static unsigned int partition_number = -1;
84static LIST_HEAD(ibmvscsi_head);
85static DEFINE_SPINLOCK(ibmvscsi_driver_lock);
86
87static struct scsi_transport_template *ibmvscsi_transport_template;
88
89#define IBMVSCSI_VERSION "1.5.9"
90
91MODULE_DESCRIPTION("IBM Virtual SCSI");
92MODULE_AUTHOR("Dave Boutcher");
93MODULE_LICENSE("GPL");
94MODULE_VERSION(IBMVSCSI_VERSION);
95
96module_param_named(max_id, max_id, int, S_IRUGO | S_IWUSR);
97MODULE_PARM_DESC(max_id, "Largest ID value for each channel [Default=64]");
98module_param_named(max_channel, max_channel, int, S_IRUGO | S_IWUSR);
99MODULE_PARM_DESC(max_channel, "Largest channel value [Default=3]");
100module_param_named(init_timeout, init_timeout, int, S_IRUGO | S_IWUSR);
101MODULE_PARM_DESC(init_timeout, "Initialization timeout in seconds");
102module_param_named(max_requests, max_requests, int, S_IRUGO);
103MODULE_PARM_DESC(max_requests, "Maximum requests for this adapter");
104module_param_named(fast_fail, fast_fail, int, S_IRUGO | S_IWUSR);
105MODULE_PARM_DESC(fast_fail, "Enable fast fail. [Default=1]");
106module_param_named(client_reserve, client_reserve, int, S_IRUGO );
107MODULE_PARM_DESC(client_reserve, "Attempt client managed reserve/release");
108
109static void ibmvscsi_handle_crq(struct viosrp_crq *crq,
110 struct ibmvscsi_host_data *hostdata);
111
112/* ------------------------------------------------------------
113 * Routines for managing the command/response queue
114 */
115/**
116 * ibmvscsi_handle_event: - Interrupt handler for crq events
117 * @irq: number of irq to handle, not used
118 * @dev_instance: ibmvscsi_host_data of host that received interrupt
119 *
120 * Disables interrupts and schedules srp_task
121 * Always returns IRQ_HANDLED
122 */
123static irqreturn_t ibmvscsi_handle_event(int irq, void *dev_instance)
124{
125 struct ibmvscsi_host_data *hostdata =
126 (struct ibmvscsi_host_data *)dev_instance;
127 vio_disable_interrupts(to_vio_dev(hostdata->dev));
128 tasklet_schedule(&hostdata->srp_task);
129 return IRQ_HANDLED;
130}
131
132/**
133 * release_crq_queue: - Deallocates data and unregisters CRQ
134 * @queue: crq_queue to initialize and register
135 * @host_data: ibmvscsi_host_data of host
136 *
137 * Frees irq, deallocates a page for messages, unmaps dma, and unregisters
138 * the crq with the hypervisor.
139 */
140static void ibmvscsi_release_crq_queue(struct crq_queue *queue,
141 struct ibmvscsi_host_data *hostdata,
142 int max_requests)
143{
144 long rc = 0;
145 struct vio_dev *vdev = to_vio_dev(hostdata->dev);
146 free_irq(vdev->irq, (void *)hostdata);
147 tasklet_kill(&hostdata->srp_task);
148 do {
149 if (rc)
150 msleep(100);
151 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
152 } while ((rc == H_BUSY) || (H_IS_LONG_BUSY(rc)));
153 dma_unmap_single(hostdata->dev,
154 queue->msg_token,
155 queue->size * sizeof(*queue->msgs), DMA_BIDIRECTIONAL);
156 free_page((unsigned long)queue->msgs);
157}
158
159/**
160 * crq_queue_next_crq: - Returns the next entry in message queue
161 * @queue: crq_queue to use
162 *
163 * Returns pointer to next entry in queue, or NULL if there are no new
164 * entried in the CRQ.
165 */
166static struct viosrp_crq *crq_queue_next_crq(struct crq_queue *queue)
167{
168 struct viosrp_crq *crq;
169 unsigned long flags;
170
171 spin_lock_irqsave(&queue->lock, flags);
172 crq = &queue->msgs[queue->cur];
173 if (crq->valid != VIOSRP_CRQ_FREE) {
174 if (++queue->cur == queue->size)
175 queue->cur = 0;
176
177 /* Ensure the read of the valid bit occurs before reading any
178 * other bits of the CRQ entry
179 */
180 rmb();
181 } else
182 crq = NULL;
183 spin_unlock_irqrestore(&queue->lock, flags);
184
185 return crq;
186}
187
188/**
189 * ibmvscsi_send_crq: - Send a CRQ
190 * @hostdata: the adapter
191 * @word1: the first 64 bits of the data
192 * @word2: the second 64 bits of the data
193 */
194static int ibmvscsi_send_crq(struct ibmvscsi_host_data *hostdata,
195 u64 word1, u64 word2)
196{
197 struct vio_dev *vdev = to_vio_dev(hostdata->dev);
198
199 /*
200 * Ensure the command buffer is flushed to memory before handing it
201 * over to the VIOS to prevent it from fetching any stale data.
202 */
203 mb();
204 return plpar_hcall_norets(H_SEND_CRQ, vdev->unit_address, word1, word2);
205}
206
207/**
208 * ibmvscsi_task: - Process srps asynchronously
209 * @data: ibmvscsi_host_data of host
210 */
211static void ibmvscsi_task(void *data)
212{
213 struct ibmvscsi_host_data *hostdata = (struct ibmvscsi_host_data *)data;
214 struct vio_dev *vdev = to_vio_dev(hostdata->dev);
215 struct viosrp_crq *crq;
216 int done = 0;
217
218 while (!done) {
219 /* Pull all the valid messages off the CRQ */
220 while ((crq = crq_queue_next_crq(&hostdata->queue)) != NULL) {
221 ibmvscsi_handle_crq(crq, hostdata);
222 crq->valid = VIOSRP_CRQ_FREE;
223 wmb();
224 }
225
226 vio_enable_interrupts(vdev);
227 crq = crq_queue_next_crq(&hostdata->queue);
228 if (crq != NULL) {
229 vio_disable_interrupts(vdev);
230 ibmvscsi_handle_crq(crq, hostdata);
231 crq->valid = VIOSRP_CRQ_FREE;
232 wmb();
233 } else {
234 done = 1;
235 }
236 }
237}
238
239static void gather_partition_info(void)
240{
241 const char *ppartition_name;
242 const __be32 *p_number_ptr;
243
244 /* Retrieve information about this partition */
245 if (!of_root)
246 return;
247
248 of_node_get(of_root);
249
250 ppartition_name = of_get_property(of_root, "ibm,partition-name", NULL);
251 if (ppartition_name)
252 strlcpy(partition_name, ppartition_name,
253 sizeof(partition_name));
254 p_number_ptr = of_get_property(of_root, "ibm,partition-no", NULL);
255 if (p_number_ptr)
256 partition_number = of_read_number(p_number_ptr, 1);
257 of_node_put(of_root);
258}
259
260static void set_adapter_info(struct ibmvscsi_host_data *hostdata)
261{
262 memset(&hostdata->madapter_info, 0x00,
263 sizeof(hostdata->madapter_info));
264
265 dev_info(hostdata->dev, "SRP_VERSION: %s\n", SRP_VERSION);
266 strcpy(hostdata->madapter_info.srp_version, SRP_VERSION);
267
268 strncpy(hostdata->madapter_info.partition_name, partition_name,
269 sizeof(hostdata->madapter_info.partition_name));
270
271 hostdata->madapter_info.partition_number =
272 cpu_to_be32(partition_number);
273
274 hostdata->madapter_info.mad_version = cpu_to_be32(SRP_MAD_VERSION_1);
275 hostdata->madapter_info.os_type = cpu_to_be32(SRP_MAD_OS_LINUX);
276}
277
278/**
279 * reset_crq_queue: - resets a crq after a failure
280 * @queue: crq_queue to initialize and register
281 * @hostdata: ibmvscsi_host_data of host
282 *
283 */
284static int ibmvscsi_reset_crq_queue(struct crq_queue *queue,
285 struct ibmvscsi_host_data *hostdata)
286{
287 int rc = 0;
288 struct vio_dev *vdev = to_vio_dev(hostdata->dev);
289
290 /* Close the CRQ */
291 do {
292 if (rc)
293 msleep(100);
294 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
295 } while ((rc == H_BUSY) || (H_IS_LONG_BUSY(rc)));
296
297 /* Clean out the queue */
298 memset(queue->msgs, 0x00, PAGE_SIZE);
299 queue->cur = 0;
300
301 set_adapter_info(hostdata);
302
303 /* And re-open it again */
304 rc = plpar_hcall_norets(H_REG_CRQ,
305 vdev->unit_address,
306 queue->msg_token, PAGE_SIZE);
307 if (rc == H_CLOSED) {
308 /* Adapter is good, but other end is not ready */
309 dev_warn(hostdata->dev, "Partner adapter not ready\n");
310 } else if (rc != 0) {
311 dev_warn(hostdata->dev, "couldn't register crq--rc 0x%x\n", rc);
312 }
313 return rc;
314}
315
316/**
317 * initialize_crq_queue: - Initializes and registers CRQ with hypervisor
318 * @queue: crq_queue to initialize and register
319 * @hostdata: ibmvscsi_host_data of host
320 *
321 * Allocates a page for messages, maps it for dma, and registers
322 * the crq with the hypervisor.
323 * Returns zero on success.
324 */
325static int ibmvscsi_init_crq_queue(struct crq_queue *queue,
326 struct ibmvscsi_host_data *hostdata,
327 int max_requests)
328{
329 int rc;
330 int retrc;
331 struct vio_dev *vdev = to_vio_dev(hostdata->dev);
332
333 queue->msgs = (struct viosrp_crq *)get_zeroed_page(GFP_KERNEL);
334
335 if (!queue->msgs)
336 goto malloc_failed;
337 queue->size = PAGE_SIZE / sizeof(*queue->msgs);
338
339 queue->msg_token = dma_map_single(hostdata->dev, queue->msgs,
340 queue->size * sizeof(*queue->msgs),
341 DMA_BIDIRECTIONAL);
342
343 if (dma_mapping_error(hostdata->dev, queue->msg_token))
344 goto map_failed;
345
346 gather_partition_info();
347 set_adapter_info(hostdata);
348
349 retrc = rc = plpar_hcall_norets(H_REG_CRQ,
350 vdev->unit_address,
351 queue->msg_token, PAGE_SIZE);
352 if (rc == H_RESOURCE)
353 /* maybe kexecing and resource is busy. try a reset */
354 rc = ibmvscsi_reset_crq_queue(queue,
355 hostdata);
356
357 if (rc == H_CLOSED) {
358 /* Adapter is good, but other end is not ready */
359 dev_warn(hostdata->dev, "Partner adapter not ready\n");
360 retrc = 0;
361 } else if (rc != 0) {
362 dev_warn(hostdata->dev, "Error %d opening adapter\n", rc);
363 goto reg_crq_failed;
364 }
365
366 queue->cur = 0;
367 spin_lock_init(&queue->lock);
368
369 tasklet_init(&hostdata->srp_task, (void *)ibmvscsi_task,
370 (unsigned long)hostdata);
371
372 if (request_irq(vdev->irq,
373 ibmvscsi_handle_event,
374 0, "ibmvscsi", (void *)hostdata) != 0) {
375 dev_err(hostdata->dev, "couldn't register irq 0x%x\n",
376 vdev->irq);
377 goto req_irq_failed;
378 }
379
380 rc = vio_enable_interrupts(vdev);
381 if (rc != 0) {
382 dev_err(hostdata->dev, "Error %d enabling interrupts!!!\n", rc);
383 goto req_irq_failed;
384 }
385
386 return retrc;
387
388 req_irq_failed:
389 tasklet_kill(&hostdata->srp_task);
390 rc = 0;
391 do {
392 if (rc)
393 msleep(100);
394 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
395 } while ((rc == H_BUSY) || (H_IS_LONG_BUSY(rc)));
396 reg_crq_failed:
397 dma_unmap_single(hostdata->dev,
398 queue->msg_token,
399 queue->size * sizeof(*queue->msgs), DMA_BIDIRECTIONAL);
400 map_failed:
401 free_page((unsigned long)queue->msgs);
402 malloc_failed:
403 return -1;
404}
405
406/**
407 * reenable_crq_queue: - reenables a crq after
408 * @queue: crq_queue to initialize and register
409 * @hostdata: ibmvscsi_host_data of host
410 *
411 */
412static int ibmvscsi_reenable_crq_queue(struct crq_queue *queue,
413 struct ibmvscsi_host_data *hostdata)
414{
415 int rc = 0;
416 struct vio_dev *vdev = to_vio_dev(hostdata->dev);
417
418 set_adapter_info(hostdata);
419
420 /* Re-enable the CRQ */
421 do {
422 if (rc)
423 msleep(100);
424 rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address);
425 } while ((rc == H_IN_PROGRESS) || (rc == H_BUSY) || (H_IS_LONG_BUSY(rc)));
426
427 if (rc)
428 dev_err(hostdata->dev, "Error %d enabling adapter\n", rc);
429 return rc;
430}
431
432/* ------------------------------------------------------------
433 * Routines for the event pool and event structs
434 */
435/**
436 * initialize_event_pool: - Allocates and initializes the event pool for a host
437 * @pool: event_pool to be initialized
438 * @size: Number of events in pool
439 * @hostdata: ibmvscsi_host_data who owns the event pool
440 *
441 * Returns zero on success.
442*/
443static int initialize_event_pool(struct event_pool *pool,
444 int size, struct ibmvscsi_host_data *hostdata)
445{
446 int i;
447
448 pool->size = size;
449 pool->next = 0;
450 pool->events = kcalloc(pool->size, sizeof(*pool->events), GFP_KERNEL);
451 if (!pool->events)
452 return -ENOMEM;
453
454 pool->iu_storage =
455 dma_alloc_coherent(hostdata->dev,
456 pool->size * sizeof(*pool->iu_storage),
457 &pool->iu_token, 0);
458 if (!pool->iu_storage) {
459 kfree(pool->events);
460 return -ENOMEM;
461 }
462
463 for (i = 0; i < pool->size; ++i) {
464 struct srp_event_struct *evt = &pool->events[i];
465 memset(&evt->crq, 0x00, sizeof(evt->crq));
466 atomic_set(&evt->free, 1);
467 evt->crq.valid = VIOSRP_CRQ_CMD_RSP;
468 evt->crq.IU_length = cpu_to_be16(sizeof(*evt->xfer_iu));
469 evt->crq.IU_data_ptr = cpu_to_be64(pool->iu_token +
470 sizeof(*evt->xfer_iu) * i);
471 evt->xfer_iu = pool->iu_storage + i;
472 evt->hostdata = hostdata;
473 evt->ext_list = NULL;
474 evt->ext_list_token = 0;
475 }
476
477 return 0;
478}
479
480/**
481 * release_event_pool: - Frees memory of an event pool of a host
482 * @pool: event_pool to be released
483 * @hostdata: ibmvscsi_host_data who owns the even pool
484 *
485 * Returns zero on success.
486*/
487static void release_event_pool(struct event_pool *pool,
488 struct ibmvscsi_host_data *hostdata)
489{
490 int i, in_use = 0;
491 for (i = 0; i < pool->size; ++i) {
492 if (atomic_read(&pool->events[i].free) != 1)
493 ++in_use;
494 if (pool->events[i].ext_list) {
495 dma_free_coherent(hostdata->dev,
496 SG_ALL * sizeof(struct srp_direct_buf),
497 pool->events[i].ext_list,
498 pool->events[i].ext_list_token);
499 }
500 }
501 if (in_use)
502 dev_warn(hostdata->dev, "releasing event pool with %d "
503 "events still in use?\n", in_use);
504 kfree(pool->events);
505 dma_free_coherent(hostdata->dev,
506 pool->size * sizeof(*pool->iu_storage),
507 pool->iu_storage, pool->iu_token);
508}
509
510/**
511 * valid_event_struct: - Determines if event is valid.
512 * @pool: event_pool that contains the event
513 * @evt: srp_event_struct to be checked for validity
514 *
515 * Returns zero if event is invalid, one otherwise.
516*/
517static int valid_event_struct(struct event_pool *pool,
518 struct srp_event_struct *evt)
519{
520 int index = evt - pool->events;
521 if (index < 0 || index >= pool->size) /* outside of bounds */
522 return 0;
523 if (evt != pool->events + index) /* unaligned */
524 return 0;
525 return 1;
526}
527
528/**
529 * ibmvscsi_free-event_struct: - Changes status of event to "free"
530 * @pool: event_pool that contains the event
531 * @evt: srp_event_struct to be modified
532 *
533*/
534static void free_event_struct(struct event_pool *pool,
535 struct srp_event_struct *evt)
536{
537 if (!valid_event_struct(pool, evt)) {
538 dev_err(evt->hostdata->dev, "Freeing invalid event_struct %p "
539 "(not in pool %p)\n", evt, pool->events);
540 return;
541 }
542 if (atomic_inc_return(&evt->free) != 1) {
543 dev_err(evt->hostdata->dev, "Freeing event_struct %p "
544 "which is not in use!\n", evt);
545 return;
546 }
547}
548
549/**
550 * get_evt_struct: - Gets the next free event in pool
551 * @pool: event_pool that contains the events to be searched
552 *
553 * Returns the next event in "free" state, and NULL if none are free.
554 * Note that no synchronization is done here, we assume the host_lock
555 * will syncrhonze things.
556*/
557static struct srp_event_struct *get_event_struct(struct event_pool *pool)
558{
559 int i;
560 int poolsize = pool->size;
561 int offset = pool->next;
562
563 for (i = 0; i < poolsize; i++) {
564 offset = (offset + 1) % poolsize;
565 if (!atomic_dec_if_positive(&pool->events[offset].free)) {
566 pool->next = offset;
567 return &pool->events[offset];
568 }
569 }
570
571 printk(KERN_ERR "ibmvscsi: found no event struct in pool!\n");
572 return NULL;
573}
574
575/**
576 * init_event_struct: Initialize fields in an event struct that are always
577 * required.
578 * @evt: The event
579 * @done: Routine to call when the event is responded to
580 * @format: SRP or MAD format
581 * @timeout: timeout value set in the CRQ
582 */
583static void init_event_struct(struct srp_event_struct *evt_struct,
584 void (*done) (struct srp_event_struct *),
585 u8 format,
586 int timeout)
587{
588 evt_struct->cmnd = NULL;
589 evt_struct->cmnd_done = NULL;
590 evt_struct->sync_srp = NULL;
591 evt_struct->crq.format = format;
592 evt_struct->crq.timeout = cpu_to_be16(timeout);
593 evt_struct->done = done;
594}
595
596/* ------------------------------------------------------------
597 * Routines for receiving SCSI responses from the hosting partition
598 */
599
600/**
601 * set_srp_direction: Set the fields in the srp related to data
602 * direction and number of buffers based on the direction in
603 * the scsi_cmnd and the number of buffers
604 */
605static void set_srp_direction(struct scsi_cmnd *cmd,
606 struct srp_cmd *srp_cmd,
607 int numbuf)
608{
609 u8 fmt;
610
611 if (numbuf == 0)
612 return;
613
614 if (numbuf == 1)
615 fmt = SRP_DATA_DESC_DIRECT;
616 else {
617 fmt = SRP_DATA_DESC_INDIRECT;
618 numbuf = min(numbuf, MAX_INDIRECT_BUFS);
619
620 if (cmd->sc_data_direction == DMA_TO_DEVICE)
621 srp_cmd->data_out_desc_cnt = numbuf;
622 else
623 srp_cmd->data_in_desc_cnt = numbuf;
624 }
625
626 if (cmd->sc_data_direction == DMA_TO_DEVICE)
627 srp_cmd->buf_fmt = fmt << 4;
628 else
629 srp_cmd->buf_fmt = fmt;
630}
631
632/**
633 * unmap_cmd_data: - Unmap data pointed in srp_cmd based on the format
634 * @cmd: srp_cmd whose additional_data member will be unmapped
635 * @dev: device for which the memory is mapped
636 *
637*/
638static void unmap_cmd_data(struct srp_cmd *cmd,
639 struct srp_event_struct *evt_struct,
640 struct device *dev)
641{
642 u8 out_fmt, in_fmt;
643
644 out_fmt = cmd->buf_fmt >> 4;
645 in_fmt = cmd->buf_fmt & ((1U << 4) - 1);
646
647 if (out_fmt == SRP_NO_DATA_DESC && in_fmt == SRP_NO_DATA_DESC)
648 return;
649
650 if (evt_struct->cmnd)
651 scsi_dma_unmap(evt_struct->cmnd);
652}
653
654static int map_sg_list(struct scsi_cmnd *cmd, int nseg,
655 struct srp_direct_buf *md)
656{
657 int i;
658 struct scatterlist *sg;
659 u64 total_length = 0;
660
661 scsi_for_each_sg(cmd, sg, nseg, i) {
662 struct srp_direct_buf *descr = md + i;
663 descr->va = cpu_to_be64(sg_dma_address(sg));
664 descr->len = cpu_to_be32(sg_dma_len(sg));
665 descr->key = 0;
666 total_length += sg_dma_len(sg);
667 }
668 return total_length;
669}
670
671/**
672 * map_sg_data: - Maps dma for a scatterlist and initializes descriptor fields
673 * @cmd: struct scsi_cmnd with the scatterlist
674 * @srp_cmd: srp_cmd that contains the memory descriptor
675 * @dev: device for which to map dma memory
676 *
677 * Called by map_data_for_srp_cmd() when building srp cmd from scsi cmd.
678 * Returns 1 on success.
679*/
680static int map_sg_data(struct scsi_cmnd *cmd,
681 struct srp_event_struct *evt_struct,
682 struct srp_cmd *srp_cmd, struct device *dev)
683{
684
685 int sg_mapped;
686 u64 total_length = 0;
687 struct srp_direct_buf *data =
688 (struct srp_direct_buf *) srp_cmd->add_data;
689 struct srp_indirect_buf *indirect =
690 (struct srp_indirect_buf *) data;
691
692 sg_mapped = scsi_dma_map(cmd);
693 if (!sg_mapped)
694 return 1;
695 else if (sg_mapped < 0)
696 return 0;
697
698 set_srp_direction(cmd, srp_cmd, sg_mapped);
699
700 /* special case; we can use a single direct descriptor */
701 if (sg_mapped == 1) {
702 map_sg_list(cmd, sg_mapped, data);
703 return 1;
704 }
705
706 indirect->table_desc.va = 0;
707 indirect->table_desc.len = cpu_to_be32(sg_mapped *
708 sizeof(struct srp_direct_buf));
709 indirect->table_desc.key = 0;
710
711 if (sg_mapped <= MAX_INDIRECT_BUFS) {
712 total_length = map_sg_list(cmd, sg_mapped,
713 &indirect->desc_list[0]);
714 indirect->len = cpu_to_be32(total_length);
715 return 1;
716 }
717
718 /* get indirect table */
719 if (!evt_struct->ext_list) {
720 evt_struct->ext_list = (struct srp_direct_buf *)
721 dma_alloc_coherent(dev,
722 SG_ALL * sizeof(struct srp_direct_buf),
723 &evt_struct->ext_list_token, 0);
724 if (!evt_struct->ext_list) {
725 if (!firmware_has_feature(FW_FEATURE_CMO))
726 sdev_printk(KERN_ERR, cmd->device,
727 "Can't allocate memory "
728 "for indirect table\n");
729 scsi_dma_unmap(cmd);
730 return 0;
731 }
732 }
733
734 total_length = map_sg_list(cmd, sg_mapped, evt_struct->ext_list);
735
736 indirect->len = cpu_to_be32(total_length);
737 indirect->table_desc.va = cpu_to_be64(evt_struct->ext_list_token);
738 indirect->table_desc.len = cpu_to_be32(sg_mapped *
739 sizeof(indirect->desc_list[0]));
740 memcpy(indirect->desc_list, evt_struct->ext_list,
741 MAX_INDIRECT_BUFS * sizeof(struct srp_direct_buf));
742 return 1;
743}
744
745/**
746 * map_data_for_srp_cmd: - Calls functions to map data for srp cmds
747 * @cmd: struct scsi_cmnd with the memory to be mapped
748 * @srp_cmd: srp_cmd that contains the memory descriptor
749 * @dev: dma device for which to map dma memory
750 *
751 * Called by scsi_cmd_to_srp_cmd() when converting scsi cmds to srp cmds
752 * Returns 1 on success.
753*/
754static int map_data_for_srp_cmd(struct scsi_cmnd *cmd,
755 struct srp_event_struct *evt_struct,
756 struct srp_cmd *srp_cmd, struct device *dev)
757{
758 switch (cmd->sc_data_direction) {
759 case DMA_FROM_DEVICE:
760 case DMA_TO_DEVICE:
761 break;
762 case DMA_NONE:
763 return 1;
764 case DMA_BIDIRECTIONAL:
765 sdev_printk(KERN_ERR, cmd->device,
766 "Can't map DMA_BIDIRECTIONAL to read/write\n");
767 return 0;
768 default:
769 sdev_printk(KERN_ERR, cmd->device,
770 "Unknown data direction 0x%02x; can't map!\n",
771 cmd->sc_data_direction);
772 return 0;
773 }
774
775 return map_sg_data(cmd, evt_struct, srp_cmd, dev);
776}
777
778/**
779 * purge_requests: Our virtual adapter just shut down. purge any sent requests
780 * @hostdata: the adapter
781 */
782static void purge_requests(struct ibmvscsi_host_data *hostdata, int error_code)
783{
784 struct srp_event_struct *evt;
785 unsigned long flags;
786
787 spin_lock_irqsave(hostdata->host->host_lock, flags);
788 while (!list_empty(&hostdata->sent)) {
789 evt = list_first_entry(&hostdata->sent, struct srp_event_struct, list);
790 list_del(&evt->list);
791 del_timer(&evt->timer);
792
793 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
794 if (evt->cmnd) {
795 evt->cmnd->result = (error_code << 16);
796 unmap_cmd_data(&evt->iu.srp.cmd, evt,
797 evt->hostdata->dev);
798 if (evt->cmnd_done)
799 evt->cmnd_done(evt->cmnd);
800 } else if (evt->done && evt->crq.format != VIOSRP_MAD_FORMAT &&
801 evt->iu.srp.login_req.opcode != SRP_LOGIN_REQ)
802 evt->done(evt);
803 free_event_struct(&evt->hostdata->pool, evt);
804 spin_lock_irqsave(hostdata->host->host_lock, flags);
805 }
806 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
807}
808
809/**
810 * ibmvscsi_reset_host - Reset the connection to the server
811 * @hostdata: struct ibmvscsi_host_data to reset
812*/
813static void ibmvscsi_reset_host(struct ibmvscsi_host_data *hostdata)
814{
815 scsi_block_requests(hostdata->host);
816 atomic_set(&hostdata->request_limit, 0);
817
818 purge_requests(hostdata, DID_ERROR);
819 hostdata->action = IBMVSCSI_HOST_ACTION_RESET;
820 wake_up(&hostdata->work_wait_q);
821}
822
823/**
824 * ibmvscsi_timeout - Internal command timeout handler
825 * @evt_struct: struct srp_event_struct that timed out
826 *
827 * Called when an internally generated command times out
828*/
829static void ibmvscsi_timeout(struct timer_list *t)
830{
831 struct srp_event_struct *evt_struct = from_timer(evt_struct, t, timer);
832 struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
833
834 dev_err(hostdata->dev, "Command timed out (%x). Resetting connection\n",
835 evt_struct->iu.srp.cmd.opcode);
836
837 ibmvscsi_reset_host(hostdata);
838}
839
840
841/* ------------------------------------------------------------
842 * Routines for sending and receiving SRPs
843 */
844/**
845 * ibmvscsi_send_srp_event: - Transforms event to u64 array and calls send_crq()
846 * @evt_struct: evt_struct to be sent
847 * @hostdata: ibmvscsi_host_data of host
848 * @timeout: timeout in seconds - 0 means do not time command
849 *
850 * Returns the value returned from ibmvscsi_send_crq(). (Zero for success)
851 * Note that this routine assumes that host_lock is held for synchronization
852*/
853static int ibmvscsi_send_srp_event(struct srp_event_struct *evt_struct,
854 struct ibmvscsi_host_data *hostdata,
855 unsigned long timeout)
856{
857 __be64 *crq_as_u64 = (__be64 *)&evt_struct->crq;
858 int request_status = 0;
859 int rc;
860 int srp_req = 0;
861
862 /* If we have exhausted our request limit, just fail this request,
863 * unless it is for a reset or abort.
864 * Note that there are rare cases involving driver generated requests
865 * (such as task management requests) that the mid layer may think we
866 * can handle more requests (can_queue) when we actually can't
867 */
868 if (evt_struct->crq.format == VIOSRP_SRP_FORMAT) {
869 srp_req = 1;
870 request_status =
871 atomic_dec_if_positive(&hostdata->request_limit);
872 /* If request limit was -1 when we started, it is now even
873 * less than that
874 */
875 if (request_status < -1)
876 goto send_error;
877 /* Otherwise, we may have run out of requests. */
878 /* If request limit was 0 when we started the adapter is in the
879 * process of performing a login with the server adapter, or
880 * we may have run out of requests.
881 */
882 else if (request_status == -1 &&
883 evt_struct->iu.srp.login_req.opcode != SRP_LOGIN_REQ)
884 goto send_busy;
885 /* Abort and reset calls should make it through.
886 * Nothing except abort and reset should use the last two
887 * slots unless we had two or less to begin with.
888 */
889 else if (request_status < 2 &&
890 evt_struct->iu.srp.cmd.opcode != SRP_TSK_MGMT) {
891 /* In the case that we have less than two requests
892 * available, check the server limit as a combination
893 * of the request limit and the number of requests
894 * in-flight (the size of the send list). If the
895 * server limit is greater than 2, return busy so
896 * that the last two are reserved for reset and abort.
897 */
898 int server_limit = request_status;
899 struct srp_event_struct *tmp_evt;
900
901 list_for_each_entry(tmp_evt, &hostdata->sent, list) {
902 server_limit++;
903 }
904
905 if (server_limit > 2)
906 goto send_busy;
907 }
908 }
909
910 /* Copy the IU into the transfer area */
911 *evt_struct->xfer_iu = evt_struct->iu;
912 evt_struct->xfer_iu->srp.rsp.tag = (u64)evt_struct;
913
914 /* Add this to the sent list. We need to do this
915 * before we actually send
916 * in case it comes back REALLY fast
917 */
918 list_add_tail(&evt_struct->list, &hostdata->sent);
919
920 timer_setup(&evt_struct->timer, ibmvscsi_timeout, 0);
921 if (timeout) {
922 evt_struct->timer.expires = jiffies + (timeout * HZ);
923 add_timer(&evt_struct->timer);
924 }
925
926 rc = ibmvscsi_send_crq(hostdata, be64_to_cpu(crq_as_u64[0]),
927 be64_to_cpu(crq_as_u64[1]));
928 if (rc != 0) {
929 list_del(&evt_struct->list);
930 del_timer(&evt_struct->timer);
931
932 /* If send_crq returns H_CLOSED, return SCSI_MLQUEUE_HOST_BUSY.
933 * Firmware will send a CRQ with a transport event (0xFF) to
934 * tell this client what has happened to the transport. This
935 * will be handled in ibmvscsi_handle_crq()
936 */
937 if (rc == H_CLOSED) {
938 dev_warn(hostdata->dev, "send warning. "
939 "Receive queue closed, will retry.\n");
940 goto send_busy;
941 }
942 dev_err(hostdata->dev, "send error %d\n", rc);
943 if (srp_req)
944 atomic_inc(&hostdata->request_limit);
945 goto send_error;
946 }
947
948 return 0;
949
950 send_busy:
951 unmap_cmd_data(&evt_struct->iu.srp.cmd, evt_struct, hostdata->dev);
952
953 free_event_struct(&hostdata->pool, evt_struct);
954 if (srp_req && request_status != -1)
955 atomic_inc(&hostdata->request_limit);
956 return SCSI_MLQUEUE_HOST_BUSY;
957
958 send_error:
959 unmap_cmd_data(&evt_struct->iu.srp.cmd, evt_struct, hostdata->dev);
960
961 if (evt_struct->cmnd != NULL) {
962 evt_struct->cmnd->result = DID_ERROR << 16;
963 evt_struct->cmnd_done(evt_struct->cmnd);
964 } else if (evt_struct->done)
965 evt_struct->done(evt_struct);
966
967 free_event_struct(&hostdata->pool, evt_struct);
968 return 0;
969}
970
971/**
972 * handle_cmd_rsp: - Handle responses from commands
973 * @evt_struct: srp_event_struct to be handled
974 *
975 * Used as a callback by when sending scsi cmds.
976 * Gets called by ibmvscsi_handle_crq()
977*/
978static void handle_cmd_rsp(struct srp_event_struct *evt_struct)
979{
980 struct srp_rsp *rsp = &evt_struct->xfer_iu->srp.rsp;
981 struct scsi_cmnd *cmnd = evt_struct->cmnd;
982
983 if (unlikely(rsp->opcode != SRP_RSP)) {
984 if (printk_ratelimit())
985 dev_warn(evt_struct->hostdata->dev,
986 "bad SRP RSP type %#02x\n", rsp->opcode);
987 }
988
989 if (cmnd) {
990 cmnd->result |= rsp->status;
991 if (((cmnd->result >> 1) & 0x1f) == CHECK_CONDITION)
992 memcpy(cmnd->sense_buffer,
993 rsp->data,
994 be32_to_cpu(rsp->sense_data_len));
995 unmap_cmd_data(&evt_struct->iu.srp.cmd,
996 evt_struct,
997 evt_struct->hostdata->dev);
998
999 if (rsp->flags & SRP_RSP_FLAG_DOOVER)
1000 scsi_set_resid(cmnd,
1001 be32_to_cpu(rsp->data_out_res_cnt));
1002 else if (rsp->flags & SRP_RSP_FLAG_DIOVER)
1003 scsi_set_resid(cmnd, be32_to_cpu(rsp->data_in_res_cnt));
1004 }
1005
1006 if (evt_struct->cmnd_done)
1007 evt_struct->cmnd_done(cmnd);
1008}
1009
1010/**
1011 * lun_from_dev: - Returns the lun of the scsi device
1012 * @dev: struct scsi_device
1013 *
1014*/
1015static inline u16 lun_from_dev(struct scsi_device *dev)
1016{
1017 return (0x2 << 14) | (dev->id << 8) | (dev->channel << 5) | dev->lun;
1018}
1019
1020/**
1021 * ibmvscsi_queue: - The queuecommand function of the scsi template
1022 * @cmd: struct scsi_cmnd to be executed
1023 * @done: Callback function to be called when cmd is completed
1024*/
1025static int ibmvscsi_queuecommand_lck(struct scsi_cmnd *cmnd,
1026 void (*done) (struct scsi_cmnd *))
1027{
1028 struct srp_cmd *srp_cmd;
1029 struct srp_event_struct *evt_struct;
1030 struct srp_indirect_buf *indirect;
1031 struct ibmvscsi_host_data *hostdata = shost_priv(cmnd->device->host);
1032 u16 lun = lun_from_dev(cmnd->device);
1033 u8 out_fmt, in_fmt;
1034
1035 cmnd->result = (DID_OK << 16);
1036 evt_struct = get_event_struct(&hostdata->pool);
1037 if (!evt_struct)
1038 return SCSI_MLQUEUE_HOST_BUSY;
1039
1040 /* Set up the actual SRP IU */
1041 srp_cmd = &evt_struct->iu.srp.cmd;
1042 memset(srp_cmd, 0x00, SRP_MAX_IU_LEN);
1043 srp_cmd->opcode = SRP_CMD;
1044 memcpy(srp_cmd->cdb, cmnd->cmnd, sizeof(srp_cmd->cdb));
1045 int_to_scsilun(lun, &srp_cmd->lun);
1046
1047 if (!map_data_for_srp_cmd(cmnd, evt_struct, srp_cmd, hostdata->dev)) {
1048 if (!firmware_has_feature(FW_FEATURE_CMO))
1049 sdev_printk(KERN_ERR, cmnd->device,
1050 "couldn't convert cmd to srp_cmd\n");
1051 free_event_struct(&hostdata->pool, evt_struct);
1052 return SCSI_MLQUEUE_HOST_BUSY;
1053 }
1054
1055 init_event_struct(evt_struct,
1056 handle_cmd_rsp,
1057 VIOSRP_SRP_FORMAT,
1058 cmnd->request->timeout/HZ);
1059
1060 evt_struct->cmnd = cmnd;
1061 evt_struct->cmnd_done = done;
1062
1063 /* Fix up dma address of the buffer itself */
1064 indirect = (struct srp_indirect_buf *) srp_cmd->add_data;
1065 out_fmt = srp_cmd->buf_fmt >> 4;
1066 in_fmt = srp_cmd->buf_fmt & ((1U << 4) - 1);
1067 if ((in_fmt == SRP_DATA_DESC_INDIRECT ||
1068 out_fmt == SRP_DATA_DESC_INDIRECT) &&
1069 indirect->table_desc.va == 0) {
1070 indirect->table_desc.va =
1071 cpu_to_be64(be64_to_cpu(evt_struct->crq.IU_data_ptr) +
1072 offsetof(struct srp_cmd, add_data) +
1073 offsetof(struct srp_indirect_buf, desc_list));
1074 }
1075
1076 return ibmvscsi_send_srp_event(evt_struct, hostdata, 0);
1077}
1078
1079static DEF_SCSI_QCMD(ibmvscsi_queuecommand)
1080
1081/* ------------------------------------------------------------
1082 * Routines for driver initialization
1083 */
1084
1085/**
1086 * map_persist_bufs: - Pre-map persistent data for adapter logins
1087 * @hostdata: ibmvscsi_host_data of host
1088 *
1089 * Map the capabilities and adapter info DMA buffers to avoid runtime failures.
1090 * Return 1 on error, 0 on success.
1091 */
1092static int map_persist_bufs(struct ibmvscsi_host_data *hostdata)
1093{
1094
1095 hostdata->caps_addr = dma_map_single(hostdata->dev, &hostdata->caps,
1096 sizeof(hostdata->caps), DMA_BIDIRECTIONAL);
1097
1098 if (dma_mapping_error(hostdata->dev, hostdata->caps_addr)) {
1099 dev_err(hostdata->dev, "Unable to map capabilities buffer!\n");
1100 return 1;
1101 }
1102
1103 hostdata->adapter_info_addr = dma_map_single(hostdata->dev,
1104 &hostdata->madapter_info,
1105 sizeof(hostdata->madapter_info),
1106 DMA_BIDIRECTIONAL);
1107 if (dma_mapping_error(hostdata->dev, hostdata->adapter_info_addr)) {
1108 dev_err(hostdata->dev, "Unable to map adapter info buffer!\n");
1109 dma_unmap_single(hostdata->dev, hostdata->caps_addr,
1110 sizeof(hostdata->caps), DMA_BIDIRECTIONAL);
1111 return 1;
1112 }
1113
1114 return 0;
1115}
1116
1117/**
1118 * unmap_persist_bufs: - Unmap persistent data needed for adapter logins
1119 * @hostdata: ibmvscsi_host_data of host
1120 *
1121 * Unmap the capabilities and adapter info DMA buffers
1122 */
1123static void unmap_persist_bufs(struct ibmvscsi_host_data *hostdata)
1124{
1125 dma_unmap_single(hostdata->dev, hostdata->caps_addr,
1126 sizeof(hostdata->caps), DMA_BIDIRECTIONAL);
1127
1128 dma_unmap_single(hostdata->dev, hostdata->adapter_info_addr,
1129 sizeof(hostdata->madapter_info), DMA_BIDIRECTIONAL);
1130}
1131
1132/**
1133 * login_rsp: - Handle response to SRP login request
1134 * @evt_struct: srp_event_struct with the response
1135 *
1136 * Used as a "done" callback by when sending srp_login. Gets called
1137 * by ibmvscsi_handle_crq()
1138*/
1139static void login_rsp(struct srp_event_struct *evt_struct)
1140{
1141 struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
1142 switch (evt_struct->xfer_iu->srp.login_rsp.opcode) {
1143 case SRP_LOGIN_RSP: /* it worked! */
1144 break;
1145 case SRP_LOGIN_REJ: /* refused! */
1146 dev_info(hostdata->dev, "SRP_LOGIN_REJ reason %u\n",
1147 evt_struct->xfer_iu->srp.login_rej.reason);
1148 /* Login failed. */
1149 atomic_set(&hostdata->request_limit, -1);
1150 return;
1151 default:
1152 dev_err(hostdata->dev, "Invalid login response typecode 0x%02x!\n",
1153 evt_struct->xfer_iu->srp.login_rsp.opcode);
1154 /* Login failed. */
1155 atomic_set(&hostdata->request_limit, -1);
1156 return;
1157 }
1158
1159 dev_info(hostdata->dev, "SRP_LOGIN succeeded\n");
1160 hostdata->client_migrated = 0;
1161
1162 /* Now we know what the real request-limit is.
1163 * This value is set rather than added to request_limit because
1164 * request_limit could have been set to -1 by this client.
1165 */
1166 atomic_set(&hostdata->request_limit,
1167 be32_to_cpu(evt_struct->xfer_iu->srp.login_rsp.req_lim_delta));
1168
1169 /* If we had any pending I/Os, kick them */
1170 hostdata->action = IBMVSCSI_HOST_ACTION_UNBLOCK;
1171 wake_up(&hostdata->work_wait_q);
1172}
1173
1174/**
1175 * send_srp_login: - Sends the srp login
1176 * @hostdata: ibmvscsi_host_data of host
1177 *
1178 * Returns zero if successful.
1179*/
1180static int send_srp_login(struct ibmvscsi_host_data *hostdata)
1181{
1182 int rc;
1183 unsigned long flags;
1184 struct srp_login_req *login;
1185 struct srp_event_struct *evt_struct = get_event_struct(&hostdata->pool);
1186
1187 BUG_ON(!evt_struct);
1188 init_event_struct(evt_struct, login_rsp,
1189 VIOSRP_SRP_FORMAT, login_timeout);
1190
1191 login = &evt_struct->iu.srp.login_req;
1192 memset(login, 0, sizeof(*login));
1193 login->opcode = SRP_LOGIN_REQ;
1194 login->req_it_iu_len = cpu_to_be32(sizeof(union srp_iu));
1195 login->req_buf_fmt = cpu_to_be16(SRP_BUF_FORMAT_DIRECT |
1196 SRP_BUF_FORMAT_INDIRECT);
1197
1198 spin_lock_irqsave(hostdata->host->host_lock, flags);
1199 /* Start out with a request limit of 0, since this is negotiated in
1200 * the login request we are just sending and login requests always
1201 * get sent by the driver regardless of request_limit.
1202 */
1203 atomic_set(&hostdata->request_limit, 0);
1204
1205 rc = ibmvscsi_send_srp_event(evt_struct, hostdata, login_timeout * 2);
1206 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1207 dev_info(hostdata->dev, "sent SRP login\n");
1208 return rc;
1209};
1210
1211/**
1212 * capabilities_rsp: - Handle response to MAD adapter capabilities request
1213 * @evt_struct: srp_event_struct with the response
1214 *
1215 * Used as a "done" callback by when sending adapter_info.
1216 */
1217static void capabilities_rsp(struct srp_event_struct *evt_struct)
1218{
1219 struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
1220
1221 if (evt_struct->xfer_iu->mad.capabilities.common.status) {
1222 dev_err(hostdata->dev, "error 0x%X getting capabilities info\n",
1223 evt_struct->xfer_iu->mad.capabilities.common.status);
1224 } else {
1225 if (hostdata->caps.migration.common.server_support !=
1226 cpu_to_be16(SERVER_SUPPORTS_CAP))
1227 dev_info(hostdata->dev, "Partition migration not supported\n");
1228
1229 if (client_reserve) {
1230 if (hostdata->caps.reserve.common.server_support ==
1231 cpu_to_be16(SERVER_SUPPORTS_CAP))
1232 dev_info(hostdata->dev, "Client reserve enabled\n");
1233 else
1234 dev_info(hostdata->dev, "Client reserve not supported\n");
1235 }
1236 }
1237
1238 send_srp_login(hostdata);
1239}
1240
1241/**
1242 * send_mad_capabilities: - Sends the mad capabilities request
1243 * and stores the result so it can be retrieved with
1244 * @hostdata: ibmvscsi_host_data of host
1245 */
1246static void send_mad_capabilities(struct ibmvscsi_host_data *hostdata)
1247{
1248 struct viosrp_capabilities *req;
1249 struct srp_event_struct *evt_struct;
1250 unsigned long flags;
1251 struct device_node *of_node = hostdata->dev->of_node;
1252 const char *location;
1253
1254 evt_struct = get_event_struct(&hostdata->pool);
1255 BUG_ON(!evt_struct);
1256
1257 init_event_struct(evt_struct, capabilities_rsp,
1258 VIOSRP_MAD_FORMAT, info_timeout);
1259
1260 req = &evt_struct->iu.mad.capabilities;
1261 memset(req, 0, sizeof(*req));
1262
1263 hostdata->caps.flags = cpu_to_be32(CAP_LIST_SUPPORTED);
1264 if (hostdata->client_migrated)
1265 hostdata->caps.flags |= cpu_to_be32(CLIENT_MIGRATED);
1266
1267 strlcpy(hostdata->caps.name, dev_name(&hostdata->host->shost_gendev),
1268 sizeof(hostdata->caps.name));
1269
1270 location = of_get_property(of_node, "ibm,loc-code", NULL);
1271 location = location ? location : dev_name(hostdata->dev);
1272 strlcpy(hostdata->caps.loc, location, sizeof(hostdata->caps.loc));
1273
1274 req->common.type = cpu_to_be32(VIOSRP_CAPABILITIES_TYPE);
1275 req->buffer = cpu_to_be64(hostdata->caps_addr);
1276
1277 hostdata->caps.migration.common.cap_type =
1278 cpu_to_be32(MIGRATION_CAPABILITIES);
1279 hostdata->caps.migration.common.length =
1280 cpu_to_be16(sizeof(hostdata->caps.migration));
1281 hostdata->caps.migration.common.server_support =
1282 cpu_to_be16(SERVER_SUPPORTS_CAP);
1283 hostdata->caps.migration.ecl = cpu_to_be32(1);
1284
1285 if (client_reserve) {
1286 hostdata->caps.reserve.common.cap_type =
1287 cpu_to_be32(RESERVATION_CAPABILITIES);
1288 hostdata->caps.reserve.common.length =
1289 cpu_to_be16(sizeof(hostdata->caps.reserve));
1290 hostdata->caps.reserve.common.server_support =
1291 cpu_to_be16(SERVER_SUPPORTS_CAP);
1292 hostdata->caps.reserve.type =
1293 cpu_to_be32(CLIENT_RESERVE_SCSI_2);
1294 req->common.length =
1295 cpu_to_be16(sizeof(hostdata->caps));
1296 } else
1297 req->common.length = cpu_to_be16(sizeof(hostdata->caps) -
1298 sizeof(hostdata->caps.reserve));
1299
1300 spin_lock_irqsave(hostdata->host->host_lock, flags);
1301 if (ibmvscsi_send_srp_event(evt_struct, hostdata, info_timeout * 2))
1302 dev_err(hostdata->dev, "couldn't send CAPABILITIES_REQ!\n");
1303 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1304};
1305
1306/**
1307 * fast_fail_rsp: - Handle response to MAD enable fast fail
1308 * @evt_struct: srp_event_struct with the response
1309 *
1310 * Used as a "done" callback by when sending enable fast fail. Gets called
1311 * by ibmvscsi_handle_crq()
1312 */
1313static void fast_fail_rsp(struct srp_event_struct *evt_struct)
1314{
1315 struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
1316 u16 status = be16_to_cpu(evt_struct->xfer_iu->mad.fast_fail.common.status);
1317
1318 if (status == VIOSRP_MAD_NOT_SUPPORTED)
1319 dev_err(hostdata->dev, "fast_fail not supported in server\n");
1320 else if (status == VIOSRP_MAD_FAILED)
1321 dev_err(hostdata->dev, "fast_fail request failed\n");
1322 else if (status != VIOSRP_MAD_SUCCESS)
1323 dev_err(hostdata->dev, "error 0x%X enabling fast_fail\n", status);
1324
1325 send_mad_capabilities(hostdata);
1326}
1327
1328/**
1329 * init_host - Start host initialization
1330 * @hostdata: ibmvscsi_host_data of host
1331 *
1332 * Returns zero if successful.
1333 */
1334static int enable_fast_fail(struct ibmvscsi_host_data *hostdata)
1335{
1336 int rc;
1337 unsigned long flags;
1338 struct viosrp_fast_fail *fast_fail_mad;
1339 struct srp_event_struct *evt_struct;
1340
1341 if (!fast_fail) {
1342 send_mad_capabilities(hostdata);
1343 return 0;
1344 }
1345
1346 evt_struct = get_event_struct(&hostdata->pool);
1347 BUG_ON(!evt_struct);
1348
1349 init_event_struct(evt_struct, fast_fail_rsp, VIOSRP_MAD_FORMAT, info_timeout);
1350
1351 fast_fail_mad = &evt_struct->iu.mad.fast_fail;
1352 memset(fast_fail_mad, 0, sizeof(*fast_fail_mad));
1353 fast_fail_mad->common.type = cpu_to_be32(VIOSRP_ENABLE_FAST_FAIL);
1354 fast_fail_mad->common.length = cpu_to_be16(sizeof(*fast_fail_mad));
1355
1356 spin_lock_irqsave(hostdata->host->host_lock, flags);
1357 rc = ibmvscsi_send_srp_event(evt_struct, hostdata, info_timeout * 2);
1358 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1359 return rc;
1360}
1361
1362/**
1363 * adapter_info_rsp: - Handle response to MAD adapter info request
1364 * @evt_struct: srp_event_struct with the response
1365 *
1366 * Used as a "done" callback by when sending adapter_info. Gets called
1367 * by ibmvscsi_handle_crq()
1368*/
1369static void adapter_info_rsp(struct srp_event_struct *evt_struct)
1370{
1371 struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
1372
1373 if (evt_struct->xfer_iu->mad.adapter_info.common.status) {
1374 dev_err(hostdata->dev, "error %d getting adapter info\n",
1375 evt_struct->xfer_iu->mad.adapter_info.common.status);
1376 } else {
1377 dev_info(hostdata->dev, "host srp version: %s, "
1378 "host partition %s (%d), OS %d, max io %u\n",
1379 hostdata->madapter_info.srp_version,
1380 hostdata->madapter_info.partition_name,
1381 be32_to_cpu(hostdata->madapter_info.partition_number),
1382 be32_to_cpu(hostdata->madapter_info.os_type),
1383 be32_to_cpu(hostdata->madapter_info.port_max_txu[0]));
1384
1385 if (hostdata->madapter_info.port_max_txu[0])
1386 hostdata->host->max_sectors =
1387 be32_to_cpu(hostdata->madapter_info.port_max_txu[0]) >> 9;
1388
1389 if (be32_to_cpu(hostdata->madapter_info.os_type) == SRP_MAD_OS_AIX &&
1390 strcmp(hostdata->madapter_info.srp_version, "1.6a") <= 0) {
1391 dev_err(hostdata->dev, "host (Ver. %s) doesn't support large transfers\n",
1392 hostdata->madapter_info.srp_version);
1393 dev_err(hostdata->dev, "limiting scatterlists to %d\n",
1394 MAX_INDIRECT_BUFS);
1395 hostdata->host->sg_tablesize = MAX_INDIRECT_BUFS;
1396 }
1397
1398 if (be32_to_cpu(hostdata->madapter_info.os_type) == SRP_MAD_OS_AIX) {
1399 enable_fast_fail(hostdata);
1400 return;
1401 }
1402 }
1403
1404 send_srp_login(hostdata);
1405}
1406
1407/**
1408 * send_mad_adapter_info: - Sends the mad adapter info request
1409 * and stores the result so it can be retrieved with
1410 * sysfs. We COULD consider causing a failure if the
1411 * returned SRP version doesn't match ours.
1412 * @hostdata: ibmvscsi_host_data of host
1413 *
1414 * Returns zero if successful.
1415*/
1416static void send_mad_adapter_info(struct ibmvscsi_host_data *hostdata)
1417{
1418 struct viosrp_adapter_info *req;
1419 struct srp_event_struct *evt_struct;
1420 unsigned long flags;
1421
1422 evt_struct = get_event_struct(&hostdata->pool);
1423 BUG_ON(!evt_struct);
1424
1425 init_event_struct(evt_struct,
1426 adapter_info_rsp,
1427 VIOSRP_MAD_FORMAT,
1428 info_timeout);
1429
1430 req = &evt_struct->iu.mad.adapter_info;
1431 memset(req, 0x00, sizeof(*req));
1432
1433 req->common.type = cpu_to_be32(VIOSRP_ADAPTER_INFO_TYPE);
1434 req->common.length = cpu_to_be16(sizeof(hostdata->madapter_info));
1435 req->buffer = cpu_to_be64(hostdata->adapter_info_addr);
1436
1437 spin_lock_irqsave(hostdata->host->host_lock, flags);
1438 if (ibmvscsi_send_srp_event(evt_struct, hostdata, info_timeout * 2))
1439 dev_err(hostdata->dev, "couldn't send ADAPTER_INFO_REQ!\n");
1440 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1441};
1442
1443/**
1444 * init_adapter: Start virtual adapter initialization sequence
1445 *
1446 */
1447static void init_adapter(struct ibmvscsi_host_data *hostdata)
1448{
1449 send_mad_adapter_info(hostdata);
1450}
1451
1452/**
1453 * sync_completion: Signal that a synchronous command has completed
1454 * Note that after returning from this call, the evt_struct is freed.
1455 * the caller waiting on this completion shouldn't touch the evt_struct
1456 * again.
1457 */
1458static void sync_completion(struct srp_event_struct *evt_struct)
1459{
1460 /* copy the response back */
1461 if (evt_struct->sync_srp)
1462 *evt_struct->sync_srp = *evt_struct->xfer_iu;
1463
1464 complete(&evt_struct->comp);
1465}
1466
1467/**
1468 * ibmvscsi_abort: Abort a command...from scsi host template
1469 * send this over to the server and wait synchronously for the response
1470 */
1471static int ibmvscsi_eh_abort_handler(struct scsi_cmnd *cmd)
1472{
1473 struct ibmvscsi_host_data *hostdata = shost_priv(cmd->device->host);
1474 struct srp_tsk_mgmt *tsk_mgmt;
1475 struct srp_event_struct *evt;
1476 struct srp_event_struct *tmp_evt, *found_evt;
1477 union viosrp_iu srp_rsp;
1478 int rsp_rc;
1479 unsigned long flags;
1480 u16 lun = lun_from_dev(cmd->device);
1481 unsigned long wait_switch = 0;
1482
1483 /* First, find this command in our sent list so we can figure
1484 * out the correct tag
1485 */
1486 spin_lock_irqsave(hostdata->host->host_lock, flags);
1487 wait_switch = jiffies + (init_timeout * HZ);
1488 do {
1489 found_evt = NULL;
1490 list_for_each_entry(tmp_evt, &hostdata->sent, list) {
1491 if (tmp_evt->cmnd == cmd) {
1492 found_evt = tmp_evt;
1493 break;
1494 }
1495 }
1496
1497 if (!found_evt) {
1498 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1499 return SUCCESS;
1500 }
1501
1502 evt = get_event_struct(&hostdata->pool);
1503 if (evt == NULL) {
1504 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1505 sdev_printk(KERN_ERR, cmd->device,
1506 "failed to allocate abort event\n");
1507 return FAILED;
1508 }
1509
1510 init_event_struct(evt,
1511 sync_completion,
1512 VIOSRP_SRP_FORMAT,
1513 abort_timeout);
1514
1515 tsk_mgmt = &evt->iu.srp.tsk_mgmt;
1516
1517 /* Set up an abort SRP command */
1518 memset(tsk_mgmt, 0x00, sizeof(*tsk_mgmt));
1519 tsk_mgmt->opcode = SRP_TSK_MGMT;
1520 int_to_scsilun(lun, &tsk_mgmt->lun);
1521 tsk_mgmt->tsk_mgmt_func = SRP_TSK_ABORT_TASK;
1522 tsk_mgmt->task_tag = (u64) found_evt;
1523
1524 evt->sync_srp = &srp_rsp;
1525
1526 init_completion(&evt->comp);
1527 rsp_rc = ibmvscsi_send_srp_event(evt, hostdata, abort_timeout * 2);
1528
1529 if (rsp_rc != SCSI_MLQUEUE_HOST_BUSY)
1530 break;
1531
1532 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1533 msleep(10);
1534 spin_lock_irqsave(hostdata->host->host_lock, flags);
1535 } while (time_before(jiffies, wait_switch));
1536
1537 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1538
1539 if (rsp_rc != 0) {
1540 sdev_printk(KERN_ERR, cmd->device,
1541 "failed to send abort() event. rc=%d\n", rsp_rc);
1542 return FAILED;
1543 }
1544
1545 sdev_printk(KERN_INFO, cmd->device,
1546 "aborting command. lun 0x%llx, tag 0x%llx\n",
1547 (((u64) lun) << 48), (u64) found_evt);
1548
1549 wait_for_completion(&evt->comp);
1550
1551 /* make sure we got a good response */
1552 if (unlikely(srp_rsp.srp.rsp.opcode != SRP_RSP)) {
1553 if (printk_ratelimit())
1554 sdev_printk(KERN_WARNING, cmd->device, "abort bad SRP RSP type %d\n",
1555 srp_rsp.srp.rsp.opcode);
1556 return FAILED;
1557 }
1558
1559 if (srp_rsp.srp.rsp.flags & SRP_RSP_FLAG_RSPVALID)
1560 rsp_rc = *((int *)srp_rsp.srp.rsp.data);
1561 else
1562 rsp_rc = srp_rsp.srp.rsp.status;
1563
1564 if (rsp_rc) {
1565 if (printk_ratelimit())
1566 sdev_printk(KERN_WARNING, cmd->device,
1567 "abort code %d for task tag 0x%llx\n",
1568 rsp_rc, tsk_mgmt->task_tag);
1569 return FAILED;
1570 }
1571
1572 /* Because we dropped the spinlock above, it's possible
1573 * The event is no longer in our list. Make sure it didn't
1574 * complete while we were aborting
1575 */
1576 spin_lock_irqsave(hostdata->host->host_lock, flags);
1577 found_evt = NULL;
1578 list_for_each_entry(tmp_evt, &hostdata->sent, list) {
1579 if (tmp_evt->cmnd == cmd) {
1580 found_evt = tmp_evt;
1581 break;
1582 }
1583 }
1584
1585 if (found_evt == NULL) {
1586 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1587 sdev_printk(KERN_INFO, cmd->device, "aborted task tag 0x%llx completed\n",
1588 tsk_mgmt->task_tag);
1589 return SUCCESS;
1590 }
1591
1592 sdev_printk(KERN_INFO, cmd->device, "successfully aborted task tag 0x%llx\n",
1593 tsk_mgmt->task_tag);
1594
1595 cmd->result = (DID_ABORT << 16);
1596 list_del(&found_evt->list);
1597 unmap_cmd_data(&found_evt->iu.srp.cmd, found_evt,
1598 found_evt->hostdata->dev);
1599 free_event_struct(&found_evt->hostdata->pool, found_evt);
1600 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1601 atomic_inc(&hostdata->request_limit);
1602 return SUCCESS;
1603}
1604
1605/**
1606 * ibmvscsi_eh_device_reset_handler: Reset a single LUN...from scsi host
1607 * template send this over to the server and wait synchronously for the
1608 * response
1609 */
1610static int ibmvscsi_eh_device_reset_handler(struct scsi_cmnd *cmd)
1611{
1612 struct ibmvscsi_host_data *hostdata = shost_priv(cmd->device->host);
1613 struct srp_tsk_mgmt *tsk_mgmt;
1614 struct srp_event_struct *evt;
1615 struct srp_event_struct *tmp_evt, *pos;
1616 union viosrp_iu srp_rsp;
1617 int rsp_rc;
1618 unsigned long flags;
1619 u16 lun = lun_from_dev(cmd->device);
1620 unsigned long wait_switch = 0;
1621
1622 spin_lock_irqsave(hostdata->host->host_lock, flags);
1623 wait_switch = jiffies + (init_timeout * HZ);
1624 do {
1625 evt = get_event_struct(&hostdata->pool);
1626 if (evt == NULL) {
1627 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1628 sdev_printk(KERN_ERR, cmd->device,
1629 "failed to allocate reset event\n");
1630 return FAILED;
1631 }
1632
1633 init_event_struct(evt,
1634 sync_completion,
1635 VIOSRP_SRP_FORMAT,
1636 reset_timeout);
1637
1638 tsk_mgmt = &evt->iu.srp.tsk_mgmt;
1639
1640 /* Set up a lun reset SRP command */
1641 memset(tsk_mgmt, 0x00, sizeof(*tsk_mgmt));
1642 tsk_mgmt->opcode = SRP_TSK_MGMT;
1643 int_to_scsilun(lun, &tsk_mgmt->lun);
1644 tsk_mgmt->tsk_mgmt_func = SRP_TSK_LUN_RESET;
1645
1646 evt->sync_srp = &srp_rsp;
1647
1648 init_completion(&evt->comp);
1649 rsp_rc = ibmvscsi_send_srp_event(evt, hostdata, reset_timeout * 2);
1650
1651 if (rsp_rc != SCSI_MLQUEUE_HOST_BUSY)
1652 break;
1653
1654 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1655 msleep(10);
1656 spin_lock_irqsave(hostdata->host->host_lock, flags);
1657 } while (time_before(jiffies, wait_switch));
1658
1659 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1660
1661 if (rsp_rc != 0) {
1662 sdev_printk(KERN_ERR, cmd->device,
1663 "failed to send reset event. rc=%d\n", rsp_rc);
1664 return FAILED;
1665 }
1666
1667 sdev_printk(KERN_INFO, cmd->device, "resetting device. lun 0x%llx\n",
1668 (((u64) lun) << 48));
1669
1670 wait_for_completion(&evt->comp);
1671
1672 /* make sure we got a good response */
1673 if (unlikely(srp_rsp.srp.rsp.opcode != SRP_RSP)) {
1674 if (printk_ratelimit())
1675 sdev_printk(KERN_WARNING, cmd->device, "reset bad SRP RSP type %d\n",
1676 srp_rsp.srp.rsp.opcode);
1677 return FAILED;
1678 }
1679
1680 if (srp_rsp.srp.rsp.flags & SRP_RSP_FLAG_RSPVALID)
1681 rsp_rc = *((int *)srp_rsp.srp.rsp.data);
1682 else
1683 rsp_rc = srp_rsp.srp.rsp.status;
1684
1685 if (rsp_rc) {
1686 if (printk_ratelimit())
1687 sdev_printk(KERN_WARNING, cmd->device,
1688 "reset code %d for task tag 0x%llx\n",
1689 rsp_rc, tsk_mgmt->task_tag);
1690 return FAILED;
1691 }
1692
1693 /* We need to find all commands for this LUN that have not yet been
1694 * responded to, and fail them with DID_RESET
1695 */
1696 spin_lock_irqsave(hostdata->host->host_lock, flags);
1697 list_for_each_entry_safe(tmp_evt, pos, &hostdata->sent, list) {
1698 if ((tmp_evt->cmnd) && (tmp_evt->cmnd->device == cmd->device)) {
1699 if (tmp_evt->cmnd)
1700 tmp_evt->cmnd->result = (DID_RESET << 16);
1701 list_del(&tmp_evt->list);
1702 unmap_cmd_data(&tmp_evt->iu.srp.cmd, tmp_evt,
1703 tmp_evt->hostdata->dev);
1704 free_event_struct(&tmp_evt->hostdata->pool,
1705 tmp_evt);
1706 atomic_inc(&hostdata->request_limit);
1707 if (tmp_evt->cmnd_done)
1708 tmp_evt->cmnd_done(tmp_evt->cmnd);
1709 else if (tmp_evt->done)
1710 tmp_evt->done(tmp_evt);
1711 }
1712 }
1713 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1714 return SUCCESS;
1715}
1716
1717/**
1718 * ibmvscsi_eh_host_reset_handler - Reset the connection to the server
1719 * @cmd: struct scsi_cmnd having problems
1720*/
1721static int ibmvscsi_eh_host_reset_handler(struct scsi_cmnd *cmd)
1722{
1723 unsigned long wait_switch = 0;
1724 struct ibmvscsi_host_data *hostdata = shost_priv(cmd->device->host);
1725
1726 dev_err(hostdata->dev, "Resetting connection due to error recovery\n");
1727
1728 ibmvscsi_reset_host(hostdata);
1729
1730 for (wait_switch = jiffies + (init_timeout * HZ);
1731 time_before(jiffies, wait_switch) &&
1732 atomic_read(&hostdata->request_limit) < 2;) {
1733
1734 msleep(10);
1735 }
1736
1737 if (atomic_read(&hostdata->request_limit) <= 0)
1738 return FAILED;
1739
1740 return SUCCESS;
1741}
1742
1743/**
1744 * ibmvscsi_handle_crq: - Handles and frees received events in the CRQ
1745 * @crq: Command/Response queue
1746 * @hostdata: ibmvscsi_host_data of host
1747 *
1748*/
1749static void ibmvscsi_handle_crq(struct viosrp_crq *crq,
1750 struct ibmvscsi_host_data *hostdata)
1751{
1752 long rc;
1753 unsigned long flags;
1754 /* The hypervisor copies our tag value here so no byteswapping */
1755 struct srp_event_struct *evt_struct =
1756 (__force struct srp_event_struct *)crq->IU_data_ptr;
1757 switch (crq->valid) {
1758 case VIOSRP_CRQ_INIT_RSP: /* initialization */
1759 switch (crq->format) {
1760 case VIOSRP_CRQ_INIT: /* Initialization message */
1761 dev_info(hostdata->dev, "partner initialized\n");
1762 /* Send back a response */
1763 rc = ibmvscsi_send_crq(hostdata, 0xC002000000000000LL, 0);
1764 if (rc == 0) {
1765 /* Now login */
1766 init_adapter(hostdata);
1767 } else {
1768 dev_err(hostdata->dev, "Unable to send init rsp. rc=%ld\n", rc);
1769 }
1770
1771 break;
1772 case VIOSRP_CRQ_INIT_COMPLETE: /* Initialization response */
1773 dev_info(hostdata->dev, "partner initialization complete\n");
1774
1775 /* Now login */
1776 init_adapter(hostdata);
1777 break;
1778 default:
1779 dev_err(hostdata->dev, "unknown crq message type: %d\n", crq->format);
1780 }
1781 return;
1782 case VIOSRP_CRQ_XPORT_EVENT: /* Hypervisor telling us the connection is closed */
1783 scsi_block_requests(hostdata->host);
1784 atomic_set(&hostdata->request_limit, 0);
1785 if (crq->format == 0x06) {
1786 /* We need to re-setup the interpartition connection */
1787 dev_info(hostdata->dev, "Re-enabling adapter!\n");
1788 hostdata->client_migrated = 1;
1789 hostdata->action = IBMVSCSI_HOST_ACTION_REENABLE;
1790 purge_requests(hostdata, DID_REQUEUE);
1791 wake_up(&hostdata->work_wait_q);
1792 } else {
1793 dev_err(hostdata->dev, "Virtual adapter failed rc %d!\n",
1794 crq->format);
1795 ibmvscsi_reset_host(hostdata);
1796 }
1797 return;
1798 case VIOSRP_CRQ_CMD_RSP: /* real payload */
1799 break;
1800 default:
1801 dev_err(hostdata->dev, "got an invalid message type 0x%02x\n",
1802 crq->valid);
1803 return;
1804 }
1805
1806 /* The only kind of payload CRQs we should get are responses to
1807 * things we send. Make sure this response is to something we
1808 * actually sent
1809 */
1810 if (!valid_event_struct(&hostdata->pool, evt_struct)) {
1811 dev_err(hostdata->dev, "returned correlation_token 0x%p is invalid!\n",
1812 evt_struct);
1813 return;
1814 }
1815
1816 if (atomic_read(&evt_struct->free)) {
1817 dev_err(hostdata->dev, "received duplicate correlation_token 0x%p!\n",
1818 evt_struct);
1819 return;
1820 }
1821
1822 if (crq->format == VIOSRP_SRP_FORMAT)
1823 atomic_add(be32_to_cpu(evt_struct->xfer_iu->srp.rsp.req_lim_delta),
1824 &hostdata->request_limit);
1825
1826 del_timer(&evt_struct->timer);
1827
1828 if ((crq->status != VIOSRP_OK && crq->status != VIOSRP_OK2) && evt_struct->cmnd)
1829 evt_struct->cmnd->result = DID_ERROR << 16;
1830 if (evt_struct->done)
1831 evt_struct->done(evt_struct);
1832 else
1833 dev_err(hostdata->dev, "returned done() is NULL; not running it!\n");
1834
1835 /*
1836 * Lock the host_lock before messing with these structures, since we
1837 * are running in a task context
1838 */
1839 spin_lock_irqsave(evt_struct->hostdata->host->host_lock, flags);
1840 list_del(&evt_struct->list);
1841 free_event_struct(&evt_struct->hostdata->pool, evt_struct);
1842 spin_unlock_irqrestore(evt_struct->hostdata->host->host_lock, flags);
1843}
1844
1845/**
1846 * ibmvscsi_slave_configure: Set the "allow_restart" flag for each disk.
1847 * @sdev: struct scsi_device device to configure
1848 *
1849 * Enable allow_restart for a device if it is a disk. Adjust the
1850 * queue_depth here also as is required by the documentation for
1851 * struct scsi_host_template.
1852 */
1853static int ibmvscsi_slave_configure(struct scsi_device *sdev)
1854{
1855 struct Scsi_Host *shost = sdev->host;
1856 unsigned long lock_flags = 0;
1857
1858 spin_lock_irqsave(shost->host_lock, lock_flags);
1859 if (sdev->type == TYPE_DISK) {
1860 sdev->allow_restart = 1;
1861 blk_queue_rq_timeout(sdev->request_queue, 120 * HZ);
1862 }
1863 spin_unlock_irqrestore(shost->host_lock, lock_flags);
1864 return 0;
1865}
1866
1867/**
1868 * ibmvscsi_change_queue_depth - Change the device's queue depth
1869 * @sdev: scsi device struct
1870 * @qdepth: depth to set
1871 * @reason: calling context
1872 *
1873 * Return value:
1874 * actual depth set
1875 **/
1876static int ibmvscsi_change_queue_depth(struct scsi_device *sdev, int qdepth)
1877{
1878 if (qdepth > IBMVSCSI_MAX_CMDS_PER_LUN)
1879 qdepth = IBMVSCSI_MAX_CMDS_PER_LUN;
1880 return scsi_change_queue_depth(sdev, qdepth);
1881}
1882
1883/* ------------------------------------------------------------
1884 * sysfs attributes
1885 */
1886static ssize_t show_host_vhost_loc(struct device *dev,
1887 struct device_attribute *attr, char *buf)
1888{
1889 struct Scsi_Host *shost = class_to_shost(dev);
1890 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1891 int len;
1892
1893 len = snprintf(buf, sizeof(hostdata->caps.loc), "%s\n",
1894 hostdata->caps.loc);
1895 return len;
1896}
1897
1898static struct device_attribute ibmvscsi_host_vhost_loc = {
1899 .attr = {
1900 .name = "vhost_loc",
1901 .mode = S_IRUGO,
1902 },
1903 .show = show_host_vhost_loc,
1904};
1905
1906static ssize_t show_host_vhost_name(struct device *dev,
1907 struct device_attribute *attr, char *buf)
1908{
1909 struct Scsi_Host *shost = class_to_shost(dev);
1910 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1911 int len;
1912
1913 len = snprintf(buf, sizeof(hostdata->caps.name), "%s\n",
1914 hostdata->caps.name);
1915 return len;
1916}
1917
1918static struct device_attribute ibmvscsi_host_vhost_name = {
1919 .attr = {
1920 .name = "vhost_name",
1921 .mode = S_IRUGO,
1922 },
1923 .show = show_host_vhost_name,
1924};
1925
1926static ssize_t show_host_srp_version(struct device *dev,
1927 struct device_attribute *attr, char *buf)
1928{
1929 struct Scsi_Host *shost = class_to_shost(dev);
1930 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1931 int len;
1932
1933 len = snprintf(buf, PAGE_SIZE, "%s\n",
1934 hostdata->madapter_info.srp_version);
1935 return len;
1936}
1937
1938static struct device_attribute ibmvscsi_host_srp_version = {
1939 .attr = {
1940 .name = "srp_version",
1941 .mode = S_IRUGO,
1942 },
1943 .show = show_host_srp_version,
1944};
1945
1946static ssize_t show_host_partition_name(struct device *dev,
1947 struct device_attribute *attr,
1948 char *buf)
1949{
1950 struct Scsi_Host *shost = class_to_shost(dev);
1951 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1952 int len;
1953
1954 len = snprintf(buf, PAGE_SIZE, "%s\n",
1955 hostdata->madapter_info.partition_name);
1956 return len;
1957}
1958
1959static struct device_attribute ibmvscsi_host_partition_name = {
1960 .attr = {
1961 .name = "partition_name",
1962 .mode = S_IRUGO,
1963 },
1964 .show = show_host_partition_name,
1965};
1966
1967static ssize_t show_host_partition_number(struct device *dev,
1968 struct device_attribute *attr,
1969 char *buf)
1970{
1971 struct Scsi_Host *shost = class_to_shost(dev);
1972 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1973 int len;
1974
1975 len = snprintf(buf, PAGE_SIZE, "%d\n",
1976 be32_to_cpu(hostdata->madapter_info.partition_number));
1977 return len;
1978}
1979
1980static struct device_attribute ibmvscsi_host_partition_number = {
1981 .attr = {
1982 .name = "partition_number",
1983 .mode = S_IRUGO,
1984 },
1985 .show = show_host_partition_number,
1986};
1987
1988static ssize_t show_host_mad_version(struct device *dev,
1989 struct device_attribute *attr, char *buf)
1990{
1991 struct Scsi_Host *shost = class_to_shost(dev);
1992 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1993 int len;
1994
1995 len = snprintf(buf, PAGE_SIZE, "%d\n",
1996 be32_to_cpu(hostdata->madapter_info.mad_version));
1997 return len;
1998}
1999
2000static struct device_attribute ibmvscsi_host_mad_version = {
2001 .attr = {
2002 .name = "mad_version",
2003 .mode = S_IRUGO,
2004 },
2005 .show = show_host_mad_version,
2006};
2007
2008static ssize_t show_host_os_type(struct device *dev,
2009 struct device_attribute *attr, char *buf)
2010{
2011 struct Scsi_Host *shost = class_to_shost(dev);
2012 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
2013 int len;
2014
2015 len = snprintf(buf, PAGE_SIZE, "%d\n",
2016 be32_to_cpu(hostdata->madapter_info.os_type));
2017 return len;
2018}
2019
2020static struct device_attribute ibmvscsi_host_os_type = {
2021 .attr = {
2022 .name = "os_type",
2023 .mode = S_IRUGO,
2024 },
2025 .show = show_host_os_type,
2026};
2027
2028static ssize_t show_host_config(struct device *dev,
2029 struct device_attribute *attr, char *buf)
2030{
2031 return 0;
2032}
2033
2034static struct device_attribute ibmvscsi_host_config = {
2035 .attr = {
2036 .name = "config",
2037 .mode = S_IRUGO,
2038 },
2039 .show = show_host_config,
2040};
2041
2042static int ibmvscsi_host_reset(struct Scsi_Host *shost, int reset_type)
2043{
2044 struct ibmvscsi_host_data *hostdata = shost_priv(shost);
2045
2046 dev_info(hostdata->dev, "Initiating adapter reset!\n");
2047 ibmvscsi_reset_host(hostdata);
2048
2049 return 0;
2050}
2051
2052static struct device_attribute *ibmvscsi_attrs[] = {
2053 &ibmvscsi_host_vhost_loc,
2054 &ibmvscsi_host_vhost_name,
2055 &ibmvscsi_host_srp_version,
2056 &ibmvscsi_host_partition_name,
2057 &ibmvscsi_host_partition_number,
2058 &ibmvscsi_host_mad_version,
2059 &ibmvscsi_host_os_type,
2060 &ibmvscsi_host_config,
2061 NULL
2062};
2063
2064/* ------------------------------------------------------------
2065 * SCSI driver registration
2066 */
2067static struct scsi_host_template driver_template = {
2068 .module = THIS_MODULE,
2069 .name = "IBM POWER Virtual SCSI Adapter " IBMVSCSI_VERSION,
2070 .proc_name = "ibmvscsi",
2071 .queuecommand = ibmvscsi_queuecommand,
2072 .eh_timed_out = srp_timed_out,
2073 .eh_abort_handler = ibmvscsi_eh_abort_handler,
2074 .eh_device_reset_handler = ibmvscsi_eh_device_reset_handler,
2075 .eh_host_reset_handler = ibmvscsi_eh_host_reset_handler,
2076 .slave_configure = ibmvscsi_slave_configure,
2077 .change_queue_depth = ibmvscsi_change_queue_depth,
2078 .host_reset = ibmvscsi_host_reset,
2079 .cmd_per_lun = IBMVSCSI_CMDS_PER_LUN_DEFAULT,
2080 .can_queue = IBMVSCSI_MAX_REQUESTS_DEFAULT,
2081 .this_id = -1,
2082 .sg_tablesize = SG_ALL,
2083 .shost_attrs = ibmvscsi_attrs,
2084};
2085
2086/**
2087 * ibmvscsi_get_desired_dma - Calculate IO memory desired by the driver
2088 *
2089 * @vdev: struct vio_dev for the device whose desired IO mem is to be returned
2090 *
2091 * Return value:
2092 * Number of bytes of IO data the driver will need to perform well.
2093 */
2094static unsigned long ibmvscsi_get_desired_dma(struct vio_dev *vdev)
2095{
2096 /* iu_storage data allocated in initialize_event_pool */
2097 unsigned long desired_io = max_events * sizeof(union viosrp_iu);
2098
2099 /* add io space for sg data */
2100 desired_io += (IBMVSCSI_MAX_SECTORS_DEFAULT * 512 *
2101 IBMVSCSI_CMDS_PER_LUN_DEFAULT);
2102
2103 return desired_io;
2104}
2105
2106static void ibmvscsi_do_work(struct ibmvscsi_host_data *hostdata)
2107{
2108 unsigned long flags;
2109 int rc;
2110 char *action = "reset";
2111
2112 spin_lock_irqsave(hostdata->host->host_lock, flags);
2113 switch (hostdata->action) {
2114 case IBMVSCSI_HOST_ACTION_UNBLOCK:
2115 rc = 0;
2116 break;
2117 case IBMVSCSI_HOST_ACTION_RESET:
2118 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
2119 rc = ibmvscsi_reset_crq_queue(&hostdata->queue, hostdata);
2120 spin_lock_irqsave(hostdata->host->host_lock, flags);
2121 if (!rc)
2122 rc = ibmvscsi_send_crq(hostdata, 0xC001000000000000LL, 0);
2123 vio_enable_interrupts(to_vio_dev(hostdata->dev));
2124 break;
2125 case IBMVSCSI_HOST_ACTION_REENABLE:
2126 action = "enable";
2127 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
2128 rc = ibmvscsi_reenable_crq_queue(&hostdata->queue, hostdata);
2129 spin_lock_irqsave(hostdata->host->host_lock, flags);
2130 if (!rc)
2131 rc = ibmvscsi_send_crq(hostdata, 0xC001000000000000LL, 0);
2132 break;
2133 case IBMVSCSI_HOST_ACTION_NONE:
2134 default:
2135 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
2136 return;
2137 }
2138
2139 hostdata->action = IBMVSCSI_HOST_ACTION_NONE;
2140
2141 if (rc) {
2142 atomic_set(&hostdata->request_limit, -1);
2143 dev_err(hostdata->dev, "error after %s\n", action);
2144 }
2145 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
2146
2147 scsi_unblock_requests(hostdata->host);
2148}
2149
2150static int __ibmvscsi_work_to_do(struct ibmvscsi_host_data *hostdata)
2151{
2152 if (kthread_should_stop())
2153 return 1;
2154 switch (hostdata->action) {
2155 case IBMVSCSI_HOST_ACTION_NONE:
2156 return 0;
2157 case IBMVSCSI_HOST_ACTION_RESET:
2158 case IBMVSCSI_HOST_ACTION_REENABLE:
2159 case IBMVSCSI_HOST_ACTION_UNBLOCK:
2160 default:
2161 break;
2162 }
2163
2164 return 1;
2165}
2166
2167static int ibmvscsi_work_to_do(struct ibmvscsi_host_data *hostdata)
2168{
2169 unsigned long flags;
2170 int rc;
2171
2172 spin_lock_irqsave(hostdata->host->host_lock, flags);
2173 rc = __ibmvscsi_work_to_do(hostdata);
2174 spin_unlock_irqrestore(hostdata->host->host_lock, flags);
2175
2176 return rc;
2177}
2178
2179static int ibmvscsi_work(void *data)
2180{
2181 struct ibmvscsi_host_data *hostdata = data;
2182 int rc;
2183
2184 set_user_nice(current, MIN_NICE);
2185
2186 while (1) {
2187 rc = wait_event_interruptible(hostdata->work_wait_q,
2188 ibmvscsi_work_to_do(hostdata));
2189
2190 BUG_ON(rc);
2191
2192 if (kthread_should_stop())
2193 break;
2194
2195 ibmvscsi_do_work(hostdata);
2196 }
2197
2198 return 0;
2199}
2200
2201/**
2202 * Called by bus code for each adapter
2203 */
2204static int ibmvscsi_probe(struct vio_dev *vdev, const struct vio_device_id *id)
2205{
2206 struct ibmvscsi_host_data *hostdata;
2207 struct Scsi_Host *host;
2208 struct device *dev = &vdev->dev;
2209 struct srp_rport_identifiers ids;
2210 struct srp_rport *rport;
2211 unsigned long wait_switch = 0;
2212 int rc;
2213
2214 dev_set_drvdata(&vdev->dev, NULL);
2215
2216 host = scsi_host_alloc(&driver_template, sizeof(*hostdata));
2217 if (!host) {
2218 dev_err(&vdev->dev, "couldn't allocate host data\n");
2219 goto scsi_host_alloc_failed;
2220 }
2221
2222 host->transportt = ibmvscsi_transport_template;
2223 hostdata = shost_priv(host);
2224 memset(hostdata, 0x00, sizeof(*hostdata));
2225 INIT_LIST_HEAD(&hostdata->sent);
2226 init_waitqueue_head(&hostdata->work_wait_q);
2227 hostdata->host = host;
2228 hostdata->dev = dev;
2229 atomic_set(&hostdata->request_limit, -1);
2230 hostdata->host->max_sectors = IBMVSCSI_MAX_SECTORS_DEFAULT;
2231
2232 if (map_persist_bufs(hostdata)) {
2233 dev_err(&vdev->dev, "couldn't map persistent buffers\n");
2234 goto persist_bufs_failed;
2235 }
2236
2237 hostdata->work_thread = kthread_run(ibmvscsi_work, hostdata, "%s_%d",
2238 "ibmvscsi", host->host_no);
2239
2240 if (IS_ERR(hostdata->work_thread)) {
2241 dev_err(&vdev->dev, "couldn't initialize kthread. rc=%ld\n",
2242 PTR_ERR(hostdata->work_thread));
2243 goto init_crq_failed;
2244 }
2245
2246 rc = ibmvscsi_init_crq_queue(&hostdata->queue, hostdata, max_events);
2247 if (rc != 0 && rc != H_RESOURCE) {
2248 dev_err(&vdev->dev, "couldn't initialize crq. rc=%d\n", rc);
2249 goto kill_kthread;
2250 }
2251 if (initialize_event_pool(&hostdata->pool, max_events, hostdata) != 0) {
2252 dev_err(&vdev->dev, "couldn't initialize event pool\n");
2253 goto init_pool_failed;
2254 }
2255
2256 host->max_lun = IBMVSCSI_MAX_LUN;
2257 host->max_id = max_id;
2258 host->max_channel = max_channel;
2259 host->max_cmd_len = 16;
2260
2261 dev_info(dev,
2262 "Maximum ID: %d Maximum LUN: %llu Maximum Channel: %d\n",
2263 host->max_id, host->max_lun, host->max_channel);
2264
2265 if (scsi_add_host(hostdata->host, hostdata->dev))
2266 goto add_host_failed;
2267
2268 /* we don't have a proper target_port_id so let's use the fake one */
2269 memcpy(ids.port_id, hostdata->madapter_info.partition_name,
2270 sizeof(ids.port_id));
2271 ids.roles = SRP_RPORT_ROLE_TARGET;
2272 rport = srp_rport_add(host, &ids);
2273 if (IS_ERR(rport))
2274 goto add_srp_port_failed;
2275
2276 /* Try to send an initialization message. Note that this is allowed
2277 * to fail if the other end is not acive. In that case we don't
2278 * want to scan
2279 */
2280 if (ibmvscsi_send_crq(hostdata, 0xC001000000000000LL, 0) == 0
2281 || rc == H_RESOURCE) {
2282 /*
2283 * Wait around max init_timeout secs for the adapter to finish
2284 * initializing. When we are done initializing, we will have a
2285 * valid request_limit. We don't want Linux scanning before
2286 * we are ready.
2287 */
2288 for (wait_switch = jiffies + (init_timeout * HZ);
2289 time_before(jiffies, wait_switch) &&
2290 atomic_read(&hostdata->request_limit) < 2;) {
2291
2292 msleep(10);
2293 }
2294
2295 /* if we now have a valid request_limit, initiate a scan */
2296 if (atomic_read(&hostdata->request_limit) > 0)
2297 scsi_scan_host(host);
2298 }
2299
2300 dev_set_drvdata(&vdev->dev, hostdata);
2301 spin_lock(&ibmvscsi_driver_lock);
2302 list_add_tail(&hostdata->host_list, &ibmvscsi_head);
2303 spin_unlock(&ibmvscsi_driver_lock);
2304 return 0;
2305
2306 add_srp_port_failed:
2307 scsi_remove_host(hostdata->host);
2308 add_host_failed:
2309 release_event_pool(&hostdata->pool, hostdata);
2310 init_pool_failed:
2311 ibmvscsi_release_crq_queue(&hostdata->queue, hostdata, max_events);
2312 kill_kthread:
2313 kthread_stop(hostdata->work_thread);
2314 init_crq_failed:
2315 unmap_persist_bufs(hostdata);
2316 persist_bufs_failed:
2317 scsi_host_put(host);
2318 scsi_host_alloc_failed:
2319 return -1;
2320}
2321
2322static int ibmvscsi_remove(struct vio_dev *vdev)
2323{
2324 struct ibmvscsi_host_data *hostdata = dev_get_drvdata(&vdev->dev);
2325
2326 srp_remove_host(hostdata->host);
2327 scsi_remove_host(hostdata->host);
2328
2329 purge_requests(hostdata, DID_ERROR);
2330 release_event_pool(&hostdata->pool, hostdata);
2331
2332 ibmvscsi_release_crq_queue(&hostdata->queue, hostdata,
2333 max_events);
2334
2335 kthread_stop(hostdata->work_thread);
2336 unmap_persist_bufs(hostdata);
2337
2338 spin_lock(&ibmvscsi_driver_lock);
2339 list_del(&hostdata->host_list);
2340 spin_unlock(&ibmvscsi_driver_lock);
2341
2342 scsi_host_put(hostdata->host);
2343
2344 return 0;
2345}
2346
2347/**
2348 * ibmvscsi_resume: Resume from suspend
2349 * @dev: device struct
2350 *
2351 * We may have lost an interrupt across suspend/resume, so kick the
2352 * interrupt handler
2353 */
2354static int ibmvscsi_resume(struct device *dev)
2355{
2356 struct ibmvscsi_host_data *hostdata = dev_get_drvdata(dev);
2357 vio_disable_interrupts(to_vio_dev(hostdata->dev));
2358 tasklet_schedule(&hostdata->srp_task);
2359
2360 return 0;
2361}
2362
2363/**
2364 * ibmvscsi_device_table: Used by vio.c to match devices in the device tree we
2365 * support.
2366 */
2367static const struct vio_device_id ibmvscsi_device_table[] = {
2368 {"vscsi", "IBM,v-scsi"},
2369 { "", "" }
2370};
2371MODULE_DEVICE_TABLE(vio, ibmvscsi_device_table);
2372
2373static const struct dev_pm_ops ibmvscsi_pm_ops = {
2374 .resume = ibmvscsi_resume
2375};
2376
2377static struct vio_driver ibmvscsi_driver = {
2378 .id_table = ibmvscsi_device_table,
2379 .probe = ibmvscsi_probe,
2380 .remove = ibmvscsi_remove,
2381 .get_desired_dma = ibmvscsi_get_desired_dma,
2382 .name = "ibmvscsi",
2383 .pm = &ibmvscsi_pm_ops,
2384};
2385
2386static struct srp_function_template ibmvscsi_transport_functions = {
2387};
2388
2389static int __init ibmvscsi_module_init(void)
2390{
2391 int ret;
2392
2393 /* Ensure we have two requests to do error recovery */
2394 driver_template.can_queue = max_requests;
2395 max_events = max_requests + 2;
2396
2397 if (!firmware_has_feature(FW_FEATURE_VIO))
2398 return -ENODEV;
2399
2400 ibmvscsi_transport_template =
2401 srp_attach_transport(&ibmvscsi_transport_functions);
2402 if (!ibmvscsi_transport_template)
2403 return -ENOMEM;
2404
2405 ret = vio_register_driver(&ibmvscsi_driver);
2406 if (ret)
2407 srp_release_transport(ibmvscsi_transport_template);
2408 return ret;
2409}
2410
2411static void __exit ibmvscsi_module_exit(void)
2412{
2413 vio_unregister_driver(&ibmvscsi_driver);
2414 srp_release_transport(ibmvscsi_transport_template);
2415}
2416
2417module_init(ibmvscsi_module_init);
2418module_exit(ibmvscsi_module_exit);