2 * scsi_lib.c Copyright (C) 1999 Eric Youngdale
4 * SCSI queueing library.
5 * Initial versions: Eric Youngdale (eric@andante.org).
6 * Based upon conversations with large numbers
7 * of people at Linux Expo.
10 #include <linux/bio.h>
11 #include <linux/blkdev.h>
12 #include <linux/completion.h>
13 #include <linux/kernel.h>
14 #include <linux/mempool.h>
15 #include <linux/slab.h>
16 #include <linux/init.h>
17 #include <linux/pci.h>
18 #include <linux/delay.h>
19 #include <linux/hardirq.h>
20 #include <linux/scatterlist.h>
22 #include <scsi/scsi.h>
23 #include <scsi/scsi_cmnd.h>
24 #include <scsi/scsi_dbg.h>
25 #include <scsi/scsi_device.h>
26 #include <scsi/scsi_driver.h>
27 #include <scsi/scsi_eh.h>
28 #include <scsi/scsi_host.h>
30 #include "scsi_priv.h"
31 #include "scsi_logging.h"
34 #define SG_MEMPOOL_NR ARRAY_SIZE(scsi_sg_pools)
35 #define SG_MEMPOOL_SIZE 2
37 struct scsi_host_sg_pool
{
40 struct kmem_cache
*slab
;
44 #define SP(x) { x, "sgpool-" #x }
45 static struct scsi_host_sg_pool scsi_sg_pools
[] = {
54 static void scsi_run_queue(struct request_queue
*q
);
57 * Function: scsi_unprep_request()
59 * Purpose: Remove all preparation done for a request, including its
60 * associated scsi_cmnd, so that it can be requeued.
62 * Arguments: req - request to unprepare
64 * Lock status: Assumed that no locks are held upon entry.
68 static void scsi_unprep_request(struct request
*req
)
70 struct scsi_cmnd
*cmd
= req
->special
;
72 req
->cmd_flags
&= ~REQ_DONTPREP
;
75 scsi_put_command(cmd
);
79 * Function: scsi_queue_insert()
81 * Purpose: Insert a command in the midlevel queue.
83 * Arguments: cmd - command that we are adding to queue.
84 * reason - why we are inserting command to queue.
86 * Lock status: Assumed that lock is not held upon entry.
90 * Notes: We do this for one of two cases. Either the host is busy
91 * and it cannot accept any more commands for the time being,
92 * or the device returned QUEUE_FULL and can accept no more
94 * Notes: This could be called either from an interrupt context or a
95 * normal process context.
97 int scsi_queue_insert(struct scsi_cmnd
*cmd
, int reason
)
99 struct Scsi_Host
*host
= cmd
->device
->host
;
100 struct scsi_device
*device
= cmd
->device
;
101 struct request_queue
*q
= device
->request_queue
;
105 printk("Inserting command %p into mlqueue\n", cmd
));
108 * Set the appropriate busy bit for the device/host.
110 * If the host/device isn't busy, assume that something actually
111 * completed, and that we should be able to queue a command now.
113 * Note that the prior mid-layer assumption that any host could
114 * always queue at least one command is now broken. The mid-layer
115 * will implement a user specifiable stall (see
116 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
117 * if a command is requeued with no other commands outstanding
118 * either for the device or for the host.
120 if (reason
== SCSI_MLQUEUE_HOST_BUSY
)
121 host
->host_blocked
= host
->max_host_blocked
;
122 else if (reason
== SCSI_MLQUEUE_DEVICE_BUSY
)
123 device
->device_blocked
= device
->max_device_blocked
;
126 * Decrement the counters, since these commands are no longer
127 * active on the host/device.
129 scsi_device_unbusy(device
);
132 * Requeue this command. It will go before all other commands
133 * that are already in the queue.
135 * NOTE: there is magic here about the way the queue is plugged if
136 * we have no outstanding commands.
138 * Although we *don't* plug the queue, we call the request
139 * function. The SCSI request function detects the blocked condition
140 * and plugs the queue appropriately.
142 spin_lock_irqsave(q
->queue_lock
, flags
);
143 blk_requeue_request(q
, cmd
->request
);
144 spin_unlock_irqrestore(q
->queue_lock
, flags
);
152 * scsi_execute - insert request and wait for the result
155 * @data_direction: data direction
156 * @buffer: data buffer
157 * @bufflen: len of buffer
158 * @sense: optional sense buffer
159 * @timeout: request timeout in seconds
160 * @retries: number of times to retry request
161 * @flags: or into request flags;
163 * returns the req->errors value which is the scsi_cmnd result
166 int scsi_execute(struct scsi_device
*sdev
, const unsigned char *cmd
,
167 int data_direction
, void *buffer
, unsigned bufflen
,
168 unsigned char *sense
, int timeout
, int retries
, int flags
)
171 int write
= (data_direction
== DMA_TO_DEVICE
);
172 int ret
= DRIVER_ERROR
<< 24;
174 req
= blk_get_request(sdev
->request_queue
, write
, __GFP_WAIT
);
176 if (bufflen
&& blk_rq_map_kern(sdev
->request_queue
, req
,
177 buffer
, bufflen
, __GFP_WAIT
))
180 req
->cmd_len
= COMMAND_SIZE(cmd
[0]);
181 memcpy(req
->cmd
, cmd
, req
->cmd_len
);
184 req
->retries
= retries
;
185 req
->timeout
= timeout
;
186 req
->cmd_type
= REQ_TYPE_BLOCK_PC
;
187 req
->cmd_flags
|= flags
| REQ_QUIET
| REQ_PREEMPT
;
190 * head injection *required* here otherwise quiesce won't work
192 blk_execute_rq(req
->q
, NULL
, req
, 1);
196 blk_put_request(req
);
200 EXPORT_SYMBOL(scsi_execute
);
203 int scsi_execute_req(struct scsi_device
*sdev
, const unsigned char *cmd
,
204 int data_direction
, void *buffer
, unsigned bufflen
,
205 struct scsi_sense_hdr
*sshdr
, int timeout
, int retries
)
211 sense
= kzalloc(SCSI_SENSE_BUFFERSIZE
, GFP_NOIO
);
213 return DRIVER_ERROR
<< 24;
215 result
= scsi_execute(sdev
, cmd
, data_direction
, buffer
, bufflen
,
216 sense
, timeout
, retries
, 0);
218 scsi_normalize_sense(sense
, SCSI_SENSE_BUFFERSIZE
, sshdr
);
223 EXPORT_SYMBOL(scsi_execute_req
);
225 struct scsi_io_context
{
227 void (*done
)(void *data
, char *sense
, int result
, int resid
);
228 char sense
[SCSI_SENSE_BUFFERSIZE
];
231 static struct kmem_cache
*scsi_io_context_cache
;
233 static void scsi_end_async(struct request
*req
, int uptodate
)
235 struct scsi_io_context
*sioc
= req
->end_io_data
;
238 sioc
->done(sioc
->data
, sioc
->sense
, req
->errors
, req
->data_len
);
240 kmem_cache_free(scsi_io_context_cache
, sioc
);
241 __blk_put_request(req
->q
, req
);
244 static int scsi_merge_bio(struct request
*rq
, struct bio
*bio
)
246 struct request_queue
*q
= rq
->q
;
248 bio
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
249 if (rq_data_dir(rq
) == WRITE
)
250 bio
->bi_rw
|= (1 << BIO_RW
);
251 blk_queue_bounce(q
, &bio
);
253 return blk_rq_append_bio(q
, rq
, bio
);
256 static void scsi_bi_endio(struct bio
*bio
, int error
)
262 * scsi_req_map_sg - map a scatterlist into a request
263 * @rq: request to fill
265 * @nsegs: number of elements
266 * @bufflen: len of buffer
267 * @gfp: memory allocation flags
269 * scsi_req_map_sg maps a scatterlist into a request so that the
270 * request can be sent to the block layer. We do not trust the scatterlist
271 * sent to use, as some ULDs use that struct to only organize the pages.
273 static int scsi_req_map_sg(struct request
*rq
, struct scatterlist
*sgl
,
274 int nsegs
, unsigned bufflen
, gfp_t gfp
)
276 struct request_queue
*q
= rq
->q
;
277 int nr_pages
= (bufflen
+ sgl
[0].offset
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
278 unsigned int data_len
= bufflen
, len
, bytes
, off
;
279 struct scatterlist
*sg
;
281 struct bio
*bio
= NULL
;
282 int i
, err
, nr_vecs
= 0;
284 for_each_sg(sgl
, sg
, nsegs
, i
) {
290 while (len
> 0 && data_len
> 0) {
292 * sg sends a scatterlist that is larger than
293 * the data_len it wants transferred for certain
296 bytes
= min_t(unsigned int, len
, PAGE_SIZE
- off
);
297 bytes
= min(bytes
, data_len
);
300 nr_vecs
= min_t(int, BIO_MAX_PAGES
, nr_pages
);
303 bio
= bio_alloc(gfp
, nr_vecs
);
308 bio
->bi_end_io
= scsi_bi_endio
;
311 if (bio_add_pc_page(q
, bio
, page
, bytes
, off
) !=
318 if (bio
->bi_vcnt
>= nr_vecs
) {
319 err
= scsi_merge_bio(rq
, bio
);
334 rq
->buffer
= rq
->data
= NULL
;
335 rq
->data_len
= bufflen
;
339 while ((bio
= rq
->bio
) != NULL
) {
340 rq
->bio
= bio
->bi_next
;
342 * call endio instead of bio_put incase it was bounced
351 * scsi_execute_async - insert request
354 * @cmd_len: length of scsi cdb
355 * @data_direction: data direction
356 * @buffer: data buffer (this can be a kernel buffer or scatterlist)
357 * @bufflen: len of buffer
358 * @use_sg: if buffer is a scatterlist this is the number of elements
359 * @timeout: request timeout in seconds
360 * @retries: number of times to retry request
361 * @flags: or into request flags
363 int scsi_execute_async(struct scsi_device
*sdev
, const unsigned char *cmd
,
364 int cmd_len
, int data_direction
, void *buffer
, unsigned bufflen
,
365 int use_sg
, int timeout
, int retries
, void *privdata
,
366 void (*done
)(void *, char *, int, int), gfp_t gfp
)
369 struct scsi_io_context
*sioc
;
371 int write
= (data_direction
== DMA_TO_DEVICE
);
373 sioc
= kmem_cache_zalloc(scsi_io_context_cache
, gfp
);
375 return DRIVER_ERROR
<< 24;
377 req
= blk_get_request(sdev
->request_queue
, write
, gfp
);
380 req
->cmd_type
= REQ_TYPE_BLOCK_PC
;
381 req
->cmd_flags
|= REQ_QUIET
;
384 err
= scsi_req_map_sg(req
, buffer
, use_sg
, bufflen
, gfp
);
386 err
= blk_rq_map_kern(req
->q
, req
, buffer
, bufflen
, gfp
);
391 req
->cmd_len
= cmd_len
;
392 memset(req
->cmd
, 0, BLK_MAX_CDB
); /* ATAPI hates garbage after CDB */
393 memcpy(req
->cmd
, cmd
, req
->cmd_len
);
394 req
->sense
= sioc
->sense
;
396 req
->timeout
= timeout
;
397 req
->retries
= retries
;
398 req
->end_io_data
= sioc
;
400 sioc
->data
= privdata
;
403 blk_execute_rq_nowait(req
->q
, NULL
, req
, 1, scsi_end_async
);
407 blk_put_request(req
);
409 kmem_cache_free(scsi_io_context_cache
, sioc
);
410 return DRIVER_ERROR
<< 24;
412 EXPORT_SYMBOL_GPL(scsi_execute_async
);
415 * Function: scsi_init_cmd_errh()
417 * Purpose: Initialize cmd fields related to error handling.
419 * Arguments: cmd - command that is ready to be queued.
421 * Notes: This function has the job of initializing a number of
422 * fields related to error handling. Typically this will
423 * be called once for each command, as required.
425 static void scsi_init_cmd_errh(struct scsi_cmnd
*cmd
)
427 cmd
->serial_number
= 0;
429 memset(cmd
->sense_buffer
, 0, sizeof cmd
->sense_buffer
);
430 if (cmd
->cmd_len
== 0)
431 cmd
->cmd_len
= COMMAND_SIZE(cmd
->cmnd
[0]);
434 void scsi_device_unbusy(struct scsi_device
*sdev
)
436 struct Scsi_Host
*shost
= sdev
->host
;
439 spin_lock_irqsave(shost
->host_lock
, flags
);
441 if (unlikely(scsi_host_in_recovery(shost
) &&
442 (shost
->host_failed
|| shost
->host_eh_scheduled
)))
443 scsi_eh_wakeup(shost
);
444 spin_unlock(shost
->host_lock
);
445 spin_lock(sdev
->request_queue
->queue_lock
);
447 spin_unlock_irqrestore(sdev
->request_queue
->queue_lock
, flags
);
451 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
452 * and call blk_run_queue for all the scsi_devices on the target -
453 * including current_sdev first.
455 * Called with *no* scsi locks held.
457 static void scsi_single_lun_run(struct scsi_device
*current_sdev
)
459 struct Scsi_Host
*shost
= current_sdev
->host
;
460 struct scsi_device
*sdev
, *tmp
;
461 struct scsi_target
*starget
= scsi_target(current_sdev
);
464 spin_lock_irqsave(shost
->host_lock
, flags
);
465 starget
->starget_sdev_user
= NULL
;
466 spin_unlock_irqrestore(shost
->host_lock
, flags
);
469 * Call blk_run_queue for all LUNs on the target, starting with
470 * current_sdev. We race with others (to set starget_sdev_user),
471 * but in most cases, we will be first. Ideally, each LU on the
472 * target would get some limited time or requests on the target.
474 blk_run_queue(current_sdev
->request_queue
);
476 spin_lock_irqsave(shost
->host_lock
, flags
);
477 if (starget
->starget_sdev_user
)
479 list_for_each_entry_safe(sdev
, tmp
, &starget
->devices
,
480 same_target_siblings
) {
481 if (sdev
== current_sdev
)
483 if (scsi_device_get(sdev
))
486 spin_unlock_irqrestore(shost
->host_lock
, flags
);
487 blk_run_queue(sdev
->request_queue
);
488 spin_lock_irqsave(shost
->host_lock
, flags
);
490 scsi_device_put(sdev
);
493 spin_unlock_irqrestore(shost
->host_lock
, flags
);
497 * Function: scsi_run_queue()
499 * Purpose: Select a proper request queue to serve next
501 * Arguments: q - last request's queue
505 * Notes: The previous command was completely finished, start
506 * a new one if possible.
508 static void scsi_run_queue(struct request_queue
*q
)
510 struct scsi_device
*sdev
= q
->queuedata
;
511 struct Scsi_Host
*shost
= sdev
->host
;
514 if (sdev
->single_lun
)
515 scsi_single_lun_run(sdev
);
517 spin_lock_irqsave(shost
->host_lock
, flags
);
518 while (!list_empty(&shost
->starved_list
) &&
519 !shost
->host_blocked
&& !shost
->host_self_blocked
&&
520 !((shost
->can_queue
> 0) &&
521 (shost
->host_busy
>= shost
->can_queue
))) {
523 * As long as shost is accepting commands and we have
524 * starved queues, call blk_run_queue. scsi_request_fn
525 * drops the queue_lock and can add us back to the
528 * host_lock protects the starved_list and starved_entry.
529 * scsi_request_fn must get the host_lock before checking
530 * or modifying starved_list or starved_entry.
532 sdev
= list_entry(shost
->starved_list
.next
,
533 struct scsi_device
, starved_entry
);
534 list_del_init(&sdev
->starved_entry
);
535 spin_unlock_irqrestore(shost
->host_lock
, flags
);
538 if (test_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
) &&
539 !test_and_set_bit(QUEUE_FLAG_REENTER
,
540 &sdev
->request_queue
->queue_flags
)) {
541 blk_run_queue(sdev
->request_queue
);
542 clear_bit(QUEUE_FLAG_REENTER
,
543 &sdev
->request_queue
->queue_flags
);
545 blk_run_queue(sdev
->request_queue
);
547 spin_lock_irqsave(shost
->host_lock
, flags
);
548 if (unlikely(!list_empty(&sdev
->starved_entry
)))
550 * sdev lost a race, and was put back on the
551 * starved list. This is unlikely but without this
552 * in theory we could loop forever.
556 spin_unlock_irqrestore(shost
->host_lock
, flags
);
562 * Function: scsi_requeue_command()
564 * Purpose: Handle post-processing of completed commands.
566 * Arguments: q - queue to operate on
567 * cmd - command that may need to be requeued.
571 * Notes: After command completion, there may be blocks left
572 * over which weren't finished by the previous command
573 * this can be for a number of reasons - the main one is
574 * I/O errors in the middle of the request, in which case
575 * we need to request the blocks that come after the bad
577 * Notes: Upon return, cmd is a stale pointer.
579 static void scsi_requeue_command(struct request_queue
*q
, struct scsi_cmnd
*cmd
)
581 struct request
*req
= cmd
->request
;
584 scsi_unprep_request(req
);
585 spin_lock_irqsave(q
->queue_lock
, flags
);
586 blk_requeue_request(q
, req
);
587 spin_unlock_irqrestore(q
->queue_lock
, flags
);
592 void scsi_next_command(struct scsi_cmnd
*cmd
)
594 struct scsi_device
*sdev
= cmd
->device
;
595 struct request_queue
*q
= sdev
->request_queue
;
597 /* need to hold a reference on the device before we let go of the cmd */
598 get_device(&sdev
->sdev_gendev
);
600 scsi_put_command(cmd
);
603 /* ok to remove device now */
604 put_device(&sdev
->sdev_gendev
);
607 void scsi_run_host_queues(struct Scsi_Host
*shost
)
609 struct scsi_device
*sdev
;
611 shost_for_each_device(sdev
, shost
)
612 scsi_run_queue(sdev
->request_queue
);
616 * Function: scsi_end_request()
618 * Purpose: Post-processing of completed commands (usually invoked at end
619 * of upper level post-processing and scsi_io_completion).
621 * Arguments: cmd - command that is complete.
622 * uptodate - 1 if I/O indicates success, <= 0 for I/O error.
623 * bytes - number of bytes of completed I/O
624 * requeue - indicates whether we should requeue leftovers.
626 * Lock status: Assumed that lock is not held upon entry.
628 * Returns: cmd if requeue required, NULL otherwise.
630 * Notes: This is called for block device requests in order to
631 * mark some number of sectors as complete.
633 * We are guaranteeing that the request queue will be goosed
634 * at some point during this call.
635 * Notes: If cmd was requeued, upon return it will be a stale pointer.
637 static struct scsi_cmnd
*scsi_end_request(struct scsi_cmnd
*cmd
, int uptodate
,
638 int bytes
, int requeue
)
640 struct request_queue
*q
= cmd
->device
->request_queue
;
641 struct request
*req
= cmd
->request
;
645 * If there are blocks left over at the end, set up the command
646 * to queue the remainder of them.
648 if (end_that_request_chunk(req
, uptodate
, bytes
)) {
649 int leftover
= (req
->hard_nr_sectors
<< 9);
651 if (blk_pc_request(req
))
652 leftover
= req
->data_len
;
654 /* kill remainder if no retrys */
655 if (!uptodate
&& blk_noretry_request(req
))
656 end_that_request_chunk(req
, 0, leftover
);
660 * Bleah. Leftovers again. Stick the
661 * leftovers in the front of the
662 * queue, and goose the queue again.
664 scsi_requeue_command(q
, cmd
);
671 add_disk_randomness(req
->rq_disk
);
673 spin_lock_irqsave(q
->queue_lock
, flags
);
674 if (blk_rq_tagged(req
))
675 blk_queue_end_tag(q
, req
);
676 end_that_request_last(req
, uptodate
);
677 spin_unlock_irqrestore(q
->queue_lock
, flags
);
680 * This will goose the queue request function at the end, so we don't
681 * need to worry about launching another command.
683 scsi_next_command(cmd
);
688 * The maximum number of SG segments that we will put inside a scatterlist
689 * (unless chaining is used). Should ideally fit inside a single page, to
690 * avoid a higher order allocation.
692 #define SCSI_MAX_SG_SEGMENTS 128
695 * Like SCSI_MAX_SG_SEGMENTS, but for archs that have sg chaining. This limit
696 * is totally arbitrary, a setting of 2048 will get you at least 8mb ios.
698 #define SCSI_MAX_SG_CHAIN_SEGMENTS 2048
700 static inline unsigned int scsi_sgtable_index(unsigned short nents
)
717 case 65 ... SCSI_MAX_SG_SEGMENTS
:
721 printk(KERN_ERR
"scsi: bad segment count=%d\n", nents
);
728 struct scatterlist
*scsi_alloc_sgtable(struct scsi_cmnd
*cmd
, gfp_t gfp_mask
)
730 struct scsi_host_sg_pool
*sgp
;
731 struct scatterlist
*sgl
, *prev
, *ret
;
735 BUG_ON(!cmd
->use_sg
);
741 if (this > SCSI_MAX_SG_SEGMENTS
) {
742 this = SCSI_MAX_SG_SEGMENTS
- 1;
743 index
= SG_MEMPOOL_NR
- 1;
745 index
= scsi_sgtable_index(this);
749 sgp
= scsi_sg_pools
+ index
;
751 sgl
= mempool_alloc(sgp
->pool
, gfp_mask
);
755 memset(sgl
, 0, sizeof(*sgl
) * sgp
->size
);
758 * first loop through, set initial index and return value
761 cmd
->sglist_len
= index
;
766 * chain previous sglist, if any. we know the previous
767 * sglist must be the biggest one, or we would not have
768 * ended up doing another loop.
771 sg_chain(prev
, SCSI_MAX_SG_SEGMENTS
, sgl
);
774 * don't allow subsequent mempool allocs to sleep, it would
775 * violate the mempool principle.
777 gfp_mask
&= ~__GFP_WAIT
;
778 gfp_mask
|= __GFP_HIGH
;
783 * ->use_sg may get modified after dma mapping has potentially
784 * shrunk the number of segments, so keep a copy of it for free.
786 cmd
->__use_sg
= cmd
->use_sg
;
791 * Free entries chained off ret. Since we were trying to
792 * allocate another sglist, we know that all entries are of
795 sgp
= scsi_sg_pools
+ SG_MEMPOOL_NR
- 1;
797 ret
= &ret
[SCSI_MAX_SG_SEGMENTS
- 1];
799 while ((sgl
= sg_chain_ptr(ret
)) != NULL
) {
800 ret
= &sgl
[SCSI_MAX_SG_SEGMENTS
- 1];
801 mempool_free(sgl
, sgp
->pool
);
804 mempool_free(prev
, sgp
->pool
);
809 EXPORT_SYMBOL(scsi_alloc_sgtable
);
811 void scsi_free_sgtable(struct scsi_cmnd
*cmd
)
813 struct scatterlist
*sgl
= cmd
->request_buffer
;
814 struct scsi_host_sg_pool
*sgp
;
816 BUG_ON(cmd
->sglist_len
>= SG_MEMPOOL_NR
);
819 * if this is the biggest size sglist, check if we have
820 * chained parts we need to free
822 if (cmd
->__use_sg
> SCSI_MAX_SG_SEGMENTS
) {
823 unsigned short this, left
;
824 struct scatterlist
*next
;
827 left
= cmd
->__use_sg
- (SCSI_MAX_SG_SEGMENTS
- 1);
828 next
= sg_chain_ptr(&sgl
[SCSI_MAX_SG_SEGMENTS
- 1]);
829 while (left
&& next
) {
832 if (this > SCSI_MAX_SG_SEGMENTS
) {
833 this = SCSI_MAX_SG_SEGMENTS
- 1;
834 index
= SG_MEMPOOL_NR
- 1;
836 index
= scsi_sgtable_index(this);
840 sgp
= scsi_sg_pools
+ index
;
843 next
= sg_chain_ptr(&sgl
[sgp
->size
- 1]);
845 mempool_free(sgl
, sgp
->pool
);
849 * Restore original, will be freed below
851 sgl
= cmd
->request_buffer
;
854 sgp
= scsi_sg_pools
+ cmd
->sglist_len
;
855 mempool_free(sgl
, sgp
->pool
);
858 EXPORT_SYMBOL(scsi_free_sgtable
);
861 * Function: scsi_release_buffers()
863 * Purpose: Completion processing for block device I/O requests.
865 * Arguments: cmd - command that we are bailing.
867 * Lock status: Assumed that no lock is held upon entry.
871 * Notes: In the event that an upper level driver rejects a
872 * command, we must release resources allocated during
873 * the __init_io() function. Primarily this would involve
874 * the scatter-gather table, and potentially any bounce
877 static void scsi_release_buffers(struct scsi_cmnd
*cmd
)
880 scsi_free_sgtable(cmd
);
883 * Zero these out. They now point to freed memory, and it is
884 * dangerous to hang onto the pointers.
886 cmd
->request_buffer
= NULL
;
887 cmd
->request_bufflen
= 0;
891 * Function: scsi_io_completion()
893 * Purpose: Completion processing for block device I/O requests.
895 * Arguments: cmd - command that is finished.
897 * Lock status: Assumed that no lock is held upon entry.
901 * Notes: This function is matched in terms of capabilities to
902 * the function that created the scatter-gather list.
903 * In other words, if there are no bounce buffers
904 * (the normal case for most drivers), we don't need
905 * the logic to deal with cleaning up afterwards.
907 * We must do one of several things here:
909 * a) Call scsi_end_request. This will finish off the
910 * specified number of sectors. If we are done, the
911 * command block will be released, and the queue
912 * function will be goosed. If we are not done, then
913 * scsi_end_request will directly goose the queue.
915 * b) We can just use scsi_requeue_command() here. This would
916 * be used if we just wanted to retry, for example.
918 void scsi_io_completion(struct scsi_cmnd
*cmd
, unsigned int good_bytes
)
920 int result
= cmd
->result
;
921 int this_count
= cmd
->request_bufflen
;
922 struct request_queue
*q
= cmd
->device
->request_queue
;
923 struct request
*req
= cmd
->request
;
924 int clear_errors
= 1;
925 struct scsi_sense_hdr sshdr
;
927 int sense_deferred
= 0;
929 scsi_release_buffers(cmd
);
932 sense_valid
= scsi_command_normalize_sense(cmd
, &sshdr
);
934 sense_deferred
= scsi_sense_is_deferred(&sshdr
);
937 if (blk_pc_request(req
)) { /* SG_IO ioctl from block level */
938 req
->errors
= result
;
941 if (sense_valid
&& req
->sense
) {
943 * SG_IO wants current and deferred errors
945 int len
= 8 + cmd
->sense_buffer
[7];
947 if (len
> SCSI_SENSE_BUFFERSIZE
)
948 len
= SCSI_SENSE_BUFFERSIZE
;
949 memcpy(req
->sense
, cmd
->sense_buffer
, len
);
950 req
->sense_len
= len
;
953 req
->data_len
= cmd
->resid
;
957 * Next deal with any sectors which we were able to correctly
960 SCSI_LOG_HLCOMPLETE(1, printk("%ld sectors total, "
962 req
->nr_sectors
, good_bytes
));
963 SCSI_LOG_HLCOMPLETE(1, printk("use_sg is %d\n", cmd
->use_sg
));
968 /* A number of bytes were successfully read. If there
969 * are leftovers and there is some kind of error
970 * (result != 0), retry the rest.
972 if (scsi_end_request(cmd
, 1, good_bytes
, result
== 0) == NULL
)
975 /* good_bytes = 0, or (inclusive) there were leftovers and
976 * result = 0, so scsi_end_request couldn't retry.
978 if (sense_valid
&& !sense_deferred
) {
979 switch (sshdr
.sense_key
) {
981 if (cmd
->device
->removable
) {
982 /* Detected disc change. Set a bit
983 * and quietly refuse further access.
985 cmd
->device
->changed
= 1;
986 scsi_end_request(cmd
, 0, this_count
, 1);
989 /* Must have been a power glitch, or a
990 * bus reset. Could not have been a
991 * media change, so we just retry the
992 * request and see what happens.
994 scsi_requeue_command(q
, cmd
);
998 case ILLEGAL_REQUEST
:
999 /* If we had an ILLEGAL REQUEST returned, then
1000 * we may have performed an unsupported
1001 * command. The only thing this should be
1002 * would be a ten byte read where only a six
1003 * byte read was supported. Also, on a system
1004 * where READ CAPACITY failed, we may have
1005 * read past the end of the disk.
1007 if ((cmd
->device
->use_10_for_rw
&&
1008 sshdr
.asc
== 0x20 && sshdr
.ascq
== 0x00) &&
1009 (cmd
->cmnd
[0] == READ_10
||
1010 cmd
->cmnd
[0] == WRITE_10
)) {
1011 cmd
->device
->use_10_for_rw
= 0;
1012 /* This will cause a retry with a
1015 scsi_requeue_command(q
, cmd
);
1018 scsi_end_request(cmd
, 0, this_count
, 1);
1023 /* If the device is in the process of becoming
1024 * ready, or has a temporary blockage, retry.
1026 if (sshdr
.asc
== 0x04) {
1027 switch (sshdr
.ascq
) {
1028 case 0x01: /* becoming ready */
1029 case 0x04: /* format in progress */
1030 case 0x05: /* rebuild in progress */
1031 case 0x06: /* recalculation in progress */
1032 case 0x07: /* operation in progress */
1033 case 0x08: /* Long write in progress */
1034 case 0x09: /* self test in progress */
1035 scsi_requeue_command(q
, cmd
);
1041 if (!(req
->cmd_flags
& REQ_QUIET
))
1042 scsi_cmd_print_sense_hdr(cmd
,
1046 scsi_end_request(cmd
, 0, this_count
, 1);
1048 case VOLUME_OVERFLOW
:
1049 if (!(req
->cmd_flags
& REQ_QUIET
)) {
1050 scmd_printk(KERN_INFO
, cmd
,
1051 "Volume overflow, CDB: ");
1052 __scsi_print_command(cmd
->cmnd
);
1053 scsi_print_sense("", cmd
);
1055 /* See SSC3rXX or current. */
1056 scsi_end_request(cmd
, 0, this_count
, 1);
1062 if (host_byte(result
) == DID_RESET
) {
1063 /* Third party bus reset or reset for error recovery
1064 * reasons. Just retry the request and see what
1067 scsi_requeue_command(q
, cmd
);
1071 if (!(req
->cmd_flags
& REQ_QUIET
)) {
1072 scsi_print_result(cmd
);
1073 if (driver_byte(result
) & DRIVER_SENSE
)
1074 scsi_print_sense("", cmd
);
1077 scsi_end_request(cmd
, 0, this_count
, !result
);
1081 * Function: scsi_init_io()
1083 * Purpose: SCSI I/O initialize function.
1085 * Arguments: cmd - Command descriptor we wish to initialize
1087 * Returns: 0 on success
1088 * BLKPREP_DEFER if the failure is retryable
1089 * BLKPREP_KILL if the failure is fatal
1091 static int scsi_init_io(struct scsi_cmnd
*cmd
)
1093 struct request
*req
= cmd
->request
;
1097 * We used to not use scatter-gather for single segment request,
1098 * but now we do (it makes highmem I/O easier to support without
1101 cmd
->use_sg
= req
->nr_phys_segments
;
1104 * If sg table allocation fails, requeue request later.
1106 cmd
->request_buffer
= scsi_alloc_sgtable(cmd
, GFP_ATOMIC
);
1107 if (unlikely(!cmd
->request_buffer
)) {
1108 scsi_unprep_request(req
);
1109 return BLKPREP_DEFER
;
1113 if (blk_pc_request(req
))
1114 cmd
->request_bufflen
= req
->data_len
;
1116 cmd
->request_bufflen
= req
->nr_sectors
<< 9;
1119 * Next, walk the list, and fill in the addresses and sizes of
1122 count
= blk_rq_map_sg(req
->q
, req
, cmd
->request_buffer
);
1123 if (likely(count
<= cmd
->use_sg
)) {
1124 cmd
->use_sg
= count
;
1128 printk(KERN_ERR
"Incorrect number of segments after building list\n");
1129 printk(KERN_ERR
"counted %d, received %d\n", count
, cmd
->use_sg
);
1130 printk(KERN_ERR
"req nr_sec %lu, cur_nr_sec %u\n", req
->nr_sectors
,
1131 req
->current_nr_sectors
);
1133 return BLKPREP_KILL
;
1136 static struct scsi_cmnd
*scsi_get_cmd_from_req(struct scsi_device
*sdev
,
1137 struct request
*req
)
1139 struct scsi_cmnd
*cmd
;
1141 if (!req
->special
) {
1142 cmd
= scsi_get_command(sdev
, GFP_ATOMIC
);
1150 /* pull a tag out of the request if we have one */
1151 cmd
->tag
= req
->tag
;
1157 int scsi_setup_blk_pc_cmnd(struct scsi_device
*sdev
, struct request
*req
)
1159 struct scsi_cmnd
*cmd
;
1160 int ret
= scsi_prep_state_check(sdev
, req
);
1162 if (ret
!= BLKPREP_OK
)
1165 cmd
= scsi_get_cmd_from_req(sdev
, req
);
1167 return BLKPREP_DEFER
;
1170 * BLOCK_PC requests may transfer data, in which case they must
1171 * a bio attached to them. Or they might contain a SCSI command
1172 * that does not transfer data, in which case they may optionally
1173 * submit a request without an attached bio.
1178 BUG_ON(!req
->nr_phys_segments
);
1180 ret
= scsi_init_io(cmd
);
1184 BUG_ON(req
->data_len
);
1187 cmd
->request_bufflen
= 0;
1188 cmd
->request_buffer
= NULL
;
1193 BUILD_BUG_ON(sizeof(req
->cmd
) > sizeof(cmd
->cmnd
));
1194 memcpy(cmd
->cmnd
, req
->cmd
, sizeof(cmd
->cmnd
));
1195 cmd
->cmd_len
= req
->cmd_len
;
1197 cmd
->sc_data_direction
= DMA_NONE
;
1198 else if (rq_data_dir(req
) == WRITE
)
1199 cmd
->sc_data_direction
= DMA_TO_DEVICE
;
1201 cmd
->sc_data_direction
= DMA_FROM_DEVICE
;
1203 cmd
->transfersize
= req
->data_len
;
1204 cmd
->allowed
= req
->retries
;
1205 cmd
->timeout_per_command
= req
->timeout
;
1208 EXPORT_SYMBOL(scsi_setup_blk_pc_cmnd
);
1211 * Setup a REQ_TYPE_FS command. These are simple read/write request
1212 * from filesystems that still need to be translated to SCSI CDBs from
1215 int scsi_setup_fs_cmnd(struct scsi_device
*sdev
, struct request
*req
)
1217 struct scsi_cmnd
*cmd
;
1218 int ret
= scsi_prep_state_check(sdev
, req
);
1220 if (ret
!= BLKPREP_OK
)
1223 * Filesystem requests must transfer data.
1225 BUG_ON(!req
->nr_phys_segments
);
1227 cmd
= scsi_get_cmd_from_req(sdev
, req
);
1229 return BLKPREP_DEFER
;
1231 return scsi_init_io(cmd
);
1233 EXPORT_SYMBOL(scsi_setup_fs_cmnd
);
1235 int scsi_prep_state_check(struct scsi_device
*sdev
, struct request
*req
)
1237 int ret
= BLKPREP_OK
;
1240 * If the device is not in running state we will reject some
1243 if (unlikely(sdev
->sdev_state
!= SDEV_RUNNING
)) {
1244 switch (sdev
->sdev_state
) {
1247 * If the device is offline we refuse to process any
1248 * commands. The device must be brought online
1249 * before trying any recovery commands.
1251 sdev_printk(KERN_ERR
, sdev
,
1252 "rejecting I/O to offline device\n");
1257 * If the device is fully deleted, we refuse to
1258 * process any commands as well.
1260 sdev_printk(KERN_ERR
, sdev
,
1261 "rejecting I/O to dead device\n");
1267 * If the devices is blocked we defer normal commands.
1269 if (!(req
->cmd_flags
& REQ_PREEMPT
))
1270 ret
= BLKPREP_DEFER
;
1274 * For any other not fully online state we only allow
1275 * special commands. In particular any user initiated
1276 * command is not allowed.
1278 if (!(req
->cmd_flags
& REQ_PREEMPT
))
1285 EXPORT_SYMBOL(scsi_prep_state_check
);
1287 int scsi_prep_return(struct request_queue
*q
, struct request
*req
, int ret
)
1289 struct scsi_device
*sdev
= q
->queuedata
;
1293 req
->errors
= DID_NO_CONNECT
<< 16;
1294 /* release the command and kill it */
1296 struct scsi_cmnd
*cmd
= req
->special
;
1297 scsi_release_buffers(cmd
);
1298 scsi_put_command(cmd
);
1299 req
->special
= NULL
;
1304 * If we defer, the elv_next_request() returns NULL, but the
1305 * queue must be restarted, so we plug here if no returning
1306 * command will automatically do that.
1308 if (sdev
->device_busy
== 0)
1312 req
->cmd_flags
|= REQ_DONTPREP
;
1317 EXPORT_SYMBOL(scsi_prep_return
);
1319 static int scsi_prep_fn(struct request_queue
*q
, struct request
*req
)
1321 struct scsi_device
*sdev
= q
->queuedata
;
1322 int ret
= BLKPREP_KILL
;
1324 if (req
->cmd_type
== REQ_TYPE_BLOCK_PC
)
1325 ret
= scsi_setup_blk_pc_cmnd(sdev
, req
);
1326 return scsi_prep_return(q
, req
, ret
);
1330 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1333 * Called with the queue_lock held.
1335 static inline int scsi_dev_queue_ready(struct request_queue
*q
,
1336 struct scsi_device
*sdev
)
1338 if (sdev
->device_busy
>= sdev
->queue_depth
)
1340 if (sdev
->device_busy
== 0 && sdev
->device_blocked
) {
1342 * unblock after device_blocked iterates to zero
1344 if (--sdev
->device_blocked
== 0) {
1346 sdev_printk(KERN_INFO
, sdev
,
1347 "unblocking device at zero depth\n"));
1353 if (sdev
->device_blocked
)
1360 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1361 * return 0. We must end up running the queue again whenever 0 is
1362 * returned, else IO can hang.
1364 * Called with host_lock held.
1366 static inline int scsi_host_queue_ready(struct request_queue
*q
,
1367 struct Scsi_Host
*shost
,
1368 struct scsi_device
*sdev
)
1370 if (scsi_host_in_recovery(shost
))
1372 if (shost
->host_busy
== 0 && shost
->host_blocked
) {
1374 * unblock after host_blocked iterates to zero
1376 if (--shost
->host_blocked
== 0) {
1378 printk("scsi%d unblocking host at zero depth\n",
1385 if ((shost
->can_queue
> 0 && shost
->host_busy
>= shost
->can_queue
) ||
1386 shost
->host_blocked
|| shost
->host_self_blocked
) {
1387 if (list_empty(&sdev
->starved_entry
))
1388 list_add_tail(&sdev
->starved_entry
, &shost
->starved_list
);
1392 /* We're OK to process the command, so we can't be starved */
1393 if (!list_empty(&sdev
->starved_entry
))
1394 list_del_init(&sdev
->starved_entry
);
1400 * Kill a request for a dead device
1402 static void scsi_kill_request(struct request
*req
, struct request_queue
*q
)
1404 struct scsi_cmnd
*cmd
= req
->special
;
1405 struct scsi_device
*sdev
= cmd
->device
;
1406 struct Scsi_Host
*shost
= sdev
->host
;
1408 blkdev_dequeue_request(req
);
1410 if (unlikely(cmd
== NULL
)) {
1411 printk(KERN_CRIT
"impossible request in %s.\n",
1416 scsi_init_cmd_errh(cmd
);
1417 cmd
->result
= DID_NO_CONNECT
<< 16;
1418 atomic_inc(&cmd
->device
->iorequest_cnt
);
1421 * SCSI request completion path will do scsi_device_unbusy(),
1422 * bump busy counts. To bump the counters, we need to dance
1423 * with the locks as normal issue path does.
1425 sdev
->device_busy
++;
1426 spin_unlock(sdev
->request_queue
->queue_lock
);
1427 spin_lock(shost
->host_lock
);
1429 spin_unlock(shost
->host_lock
);
1430 spin_lock(sdev
->request_queue
->queue_lock
);
1435 static void scsi_softirq_done(struct request
*rq
)
1437 struct scsi_cmnd
*cmd
= rq
->completion_data
;
1438 unsigned long wait_for
= (cmd
->allowed
+ 1) * cmd
->timeout_per_command
;
1441 INIT_LIST_HEAD(&cmd
->eh_entry
);
1443 disposition
= scsi_decide_disposition(cmd
);
1444 if (disposition
!= SUCCESS
&&
1445 time_before(cmd
->jiffies_at_alloc
+ wait_for
, jiffies
)) {
1446 sdev_printk(KERN_ERR
, cmd
->device
,
1447 "timing out command, waited %lus\n",
1449 disposition
= SUCCESS
;
1452 scsi_log_completion(cmd
, disposition
);
1454 switch (disposition
) {
1456 scsi_finish_command(cmd
);
1459 scsi_queue_insert(cmd
, SCSI_MLQUEUE_EH_RETRY
);
1461 case ADD_TO_MLQUEUE
:
1462 scsi_queue_insert(cmd
, SCSI_MLQUEUE_DEVICE_BUSY
);
1465 if (!scsi_eh_scmd_add(cmd
, 0))
1466 scsi_finish_command(cmd
);
1471 * Function: scsi_request_fn()
1473 * Purpose: Main strategy routine for SCSI.
1475 * Arguments: q - Pointer to actual queue.
1479 * Lock status: IO request lock assumed to be held when called.
1481 static void scsi_request_fn(struct request_queue
*q
)
1483 struct scsi_device
*sdev
= q
->queuedata
;
1484 struct Scsi_Host
*shost
;
1485 struct scsi_cmnd
*cmd
;
1486 struct request
*req
;
1489 printk("scsi: killing requests for dead queue\n");
1490 while ((req
= elv_next_request(q
)) != NULL
)
1491 scsi_kill_request(req
, q
);
1495 if(!get_device(&sdev
->sdev_gendev
))
1496 /* We must be tearing the block queue down already */
1500 * To start with, we keep looping until the queue is empty, or until
1501 * the host is no longer able to accept any more requests.
1504 while (!blk_queue_plugged(q
)) {
1507 * get next queueable request. We do this early to make sure
1508 * that the request is fully prepared even if we cannot
1511 req
= elv_next_request(q
);
1512 if (!req
|| !scsi_dev_queue_ready(q
, sdev
))
1515 if (unlikely(!scsi_device_online(sdev
))) {
1516 sdev_printk(KERN_ERR
, sdev
,
1517 "rejecting I/O to offline device\n");
1518 scsi_kill_request(req
, q
);
1524 * Remove the request from the request list.
1526 if (!(blk_queue_tagged(q
) && !blk_queue_start_tag(q
, req
)))
1527 blkdev_dequeue_request(req
);
1528 sdev
->device_busy
++;
1530 spin_unlock(q
->queue_lock
);
1532 if (unlikely(cmd
== NULL
)) {
1533 printk(KERN_CRIT
"impossible request in %s.\n"
1534 "please mail a stack trace to "
1535 "linux-scsi@vger.kernel.org\n",
1537 blk_dump_rq_flags(req
, "foo");
1540 spin_lock(shost
->host_lock
);
1542 if (!scsi_host_queue_ready(q
, shost
, sdev
))
1544 if (sdev
->single_lun
) {
1545 if (scsi_target(sdev
)->starget_sdev_user
&&
1546 scsi_target(sdev
)->starget_sdev_user
!= sdev
)
1548 scsi_target(sdev
)->starget_sdev_user
= sdev
;
1553 * XXX(hch): This is rather suboptimal, scsi_dispatch_cmd will
1554 * take the lock again.
1556 spin_unlock_irq(shost
->host_lock
);
1559 * Finally, initialize any error handling parameters, and set up
1560 * the timers for timeouts.
1562 scsi_init_cmd_errh(cmd
);
1565 * Dispatch the command to the low-level driver.
1567 rtn
= scsi_dispatch_cmd(cmd
);
1568 spin_lock_irq(q
->queue_lock
);
1570 /* we're refusing the command; because of
1571 * the way locks get dropped, we need to
1572 * check here if plugging is required */
1573 if(sdev
->device_busy
== 0)
1583 spin_unlock_irq(shost
->host_lock
);
1586 * lock q, handle tag, requeue req, and decrement device_busy. We
1587 * must return with queue_lock held.
1589 * Decrementing device_busy without checking it is OK, as all such
1590 * cases (host limits or settings) should run the queue at some
1593 spin_lock_irq(q
->queue_lock
);
1594 blk_requeue_request(q
, req
);
1595 sdev
->device_busy
--;
1596 if(sdev
->device_busy
== 0)
1599 /* must be careful here...if we trigger the ->remove() function
1600 * we cannot be holding the q lock */
1601 spin_unlock_irq(q
->queue_lock
);
1602 put_device(&sdev
->sdev_gendev
);
1603 spin_lock_irq(q
->queue_lock
);
1606 u64
scsi_calculate_bounce_limit(struct Scsi_Host
*shost
)
1608 struct device
*host_dev
;
1609 u64 bounce_limit
= 0xffffffff;
1611 if (shost
->unchecked_isa_dma
)
1612 return BLK_BOUNCE_ISA
;
1614 * Platforms with virtual-DMA translation
1615 * hardware have no practical limit.
1617 if (!PCI_DMA_BUS_IS_PHYS
)
1618 return BLK_BOUNCE_ANY
;
1620 host_dev
= scsi_get_device(shost
);
1621 if (host_dev
&& host_dev
->dma_mask
)
1622 bounce_limit
= *host_dev
->dma_mask
;
1624 return bounce_limit
;
1626 EXPORT_SYMBOL(scsi_calculate_bounce_limit
);
1628 struct request_queue
*__scsi_alloc_queue(struct Scsi_Host
*shost
,
1629 request_fn_proc
*request_fn
)
1631 struct request_queue
*q
;
1633 q
= blk_init_queue(request_fn
, NULL
);
1638 * this limit is imposed by hardware restrictions
1640 blk_queue_max_hw_segments(q
, shost
->sg_tablesize
);
1643 * In the future, sg chaining support will be mandatory and this
1644 * ifdef can then go away. Right now we don't have all archs
1645 * converted, so better keep it safe.
1647 #ifdef ARCH_HAS_SG_CHAIN
1648 if (shost
->use_sg_chaining
)
1649 blk_queue_max_phys_segments(q
, SCSI_MAX_SG_CHAIN_SEGMENTS
);
1651 blk_queue_max_phys_segments(q
, SCSI_MAX_SG_SEGMENTS
);
1653 blk_queue_max_phys_segments(q
, SCSI_MAX_SG_SEGMENTS
);
1656 blk_queue_max_sectors(q
, shost
->max_sectors
);
1657 blk_queue_bounce_limit(q
, scsi_calculate_bounce_limit(shost
));
1658 blk_queue_segment_boundary(q
, shost
->dma_boundary
);
1660 if (!shost
->use_clustering
)
1661 clear_bit(QUEUE_FLAG_CLUSTER
, &q
->queue_flags
);
1664 EXPORT_SYMBOL(__scsi_alloc_queue
);
1666 struct request_queue
*scsi_alloc_queue(struct scsi_device
*sdev
)
1668 struct request_queue
*q
;
1670 q
= __scsi_alloc_queue(sdev
->host
, scsi_request_fn
);
1674 blk_queue_prep_rq(q
, scsi_prep_fn
);
1675 blk_queue_softirq_done(q
, scsi_softirq_done
);
1679 void scsi_free_queue(struct request_queue
*q
)
1681 blk_cleanup_queue(q
);
1685 * Function: scsi_block_requests()
1687 * Purpose: Utility function used by low-level drivers to prevent further
1688 * commands from being queued to the device.
1690 * Arguments: shost - Host in question
1694 * Lock status: No locks are assumed held.
1696 * Notes: There is no timer nor any other means by which the requests
1697 * get unblocked other than the low-level driver calling
1698 * scsi_unblock_requests().
1700 void scsi_block_requests(struct Scsi_Host
*shost
)
1702 shost
->host_self_blocked
= 1;
1704 EXPORT_SYMBOL(scsi_block_requests
);
1707 * Function: scsi_unblock_requests()
1709 * Purpose: Utility function used by low-level drivers to allow further
1710 * commands from being queued to the device.
1712 * Arguments: shost - Host in question
1716 * Lock status: No locks are assumed held.
1718 * Notes: There is no timer nor any other means by which the requests
1719 * get unblocked other than the low-level driver calling
1720 * scsi_unblock_requests().
1722 * This is done as an API function so that changes to the
1723 * internals of the scsi mid-layer won't require wholesale
1724 * changes to drivers that use this feature.
1726 void scsi_unblock_requests(struct Scsi_Host
*shost
)
1728 shost
->host_self_blocked
= 0;
1729 scsi_run_host_queues(shost
);
1731 EXPORT_SYMBOL(scsi_unblock_requests
);
1733 int __init
scsi_init_queue(void)
1737 scsi_io_context_cache
= kmem_cache_create("scsi_io_context",
1738 sizeof(struct scsi_io_context
),
1740 if (!scsi_io_context_cache
) {
1741 printk(KERN_ERR
"SCSI: can't init scsi io context cache\n");
1745 for (i
= 0; i
< SG_MEMPOOL_NR
; i
++) {
1746 struct scsi_host_sg_pool
*sgp
= scsi_sg_pools
+ i
;
1747 int size
= sgp
->size
* sizeof(struct scatterlist
);
1749 sgp
->slab
= kmem_cache_create(sgp
->name
, size
, 0,
1750 SLAB_HWCACHE_ALIGN
, NULL
);
1752 printk(KERN_ERR
"SCSI: can't init sg slab %s\n",
1756 sgp
->pool
= mempool_create_slab_pool(SG_MEMPOOL_SIZE
,
1759 printk(KERN_ERR
"SCSI: can't init sg mempool %s\n",
1767 void scsi_exit_queue(void)
1771 kmem_cache_destroy(scsi_io_context_cache
);
1773 for (i
= 0; i
< SG_MEMPOOL_NR
; i
++) {
1774 struct scsi_host_sg_pool
*sgp
= scsi_sg_pools
+ i
;
1775 mempool_destroy(sgp
->pool
);
1776 kmem_cache_destroy(sgp
->slab
);
1781 * scsi_mode_select - issue a mode select
1782 * @sdev: SCSI device to be queried
1783 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1784 * @sp: Save page bit (0 == don't save, 1 == save)
1785 * @modepage: mode page being requested
1786 * @buffer: request buffer (may not be smaller than eight bytes)
1787 * @len: length of request buffer.
1788 * @timeout: command timeout
1789 * @retries: number of retries before failing
1790 * @data: returns a structure abstracting the mode header data
1791 * @sense: place to put sense data (or NULL if no sense to be collected).
1792 * must be SCSI_SENSE_BUFFERSIZE big.
1794 * Returns zero if successful; negative error number or scsi
1799 scsi_mode_select(struct scsi_device
*sdev
, int pf
, int sp
, int modepage
,
1800 unsigned char *buffer
, int len
, int timeout
, int retries
,
1801 struct scsi_mode_data
*data
, struct scsi_sense_hdr
*sshdr
)
1803 unsigned char cmd
[10];
1804 unsigned char *real_buffer
;
1807 memset(cmd
, 0, sizeof(cmd
));
1808 cmd
[1] = (pf
? 0x10 : 0) | (sp
? 0x01 : 0);
1810 if (sdev
->use_10_for_ms
) {
1813 real_buffer
= kmalloc(8 + len
, GFP_KERNEL
);
1816 memcpy(real_buffer
+ 8, buffer
, len
);
1820 real_buffer
[2] = data
->medium_type
;
1821 real_buffer
[3] = data
->device_specific
;
1822 real_buffer
[4] = data
->longlba
? 0x01 : 0;
1824 real_buffer
[6] = data
->block_descriptor_length
>> 8;
1825 real_buffer
[7] = data
->block_descriptor_length
;
1827 cmd
[0] = MODE_SELECT_10
;
1831 if (len
> 255 || data
->block_descriptor_length
> 255 ||
1835 real_buffer
= kmalloc(4 + len
, GFP_KERNEL
);
1838 memcpy(real_buffer
+ 4, buffer
, len
);
1841 real_buffer
[1] = data
->medium_type
;
1842 real_buffer
[2] = data
->device_specific
;
1843 real_buffer
[3] = data
->block_descriptor_length
;
1846 cmd
[0] = MODE_SELECT
;
1850 ret
= scsi_execute_req(sdev
, cmd
, DMA_TO_DEVICE
, real_buffer
, len
,
1851 sshdr
, timeout
, retries
);
1855 EXPORT_SYMBOL_GPL(scsi_mode_select
);
1858 * scsi_mode_sense - issue a mode sense, falling back from 10 to
1859 * six bytes if necessary.
1860 * @sdev: SCSI device to be queried
1861 * @dbd: set if mode sense will allow block descriptors to be returned
1862 * @modepage: mode page being requested
1863 * @buffer: request buffer (may not be smaller than eight bytes)
1864 * @len: length of request buffer.
1865 * @timeout: command timeout
1866 * @retries: number of retries before failing
1867 * @data: returns a structure abstracting the mode header data
1868 * @sense: place to put sense data (or NULL if no sense to be collected).
1869 * must be SCSI_SENSE_BUFFERSIZE big.
1871 * Returns zero if unsuccessful, or the header offset (either 4
1872 * or 8 depending on whether a six or ten byte command was
1873 * issued) if successful.
1876 scsi_mode_sense(struct scsi_device
*sdev
, int dbd
, int modepage
,
1877 unsigned char *buffer
, int len
, int timeout
, int retries
,
1878 struct scsi_mode_data
*data
, struct scsi_sense_hdr
*sshdr
)
1880 unsigned char cmd
[12];
1884 struct scsi_sense_hdr my_sshdr
;
1886 memset(data
, 0, sizeof(*data
));
1887 memset(&cmd
[0], 0, 12);
1888 cmd
[1] = dbd
& 0x18; /* allows DBD and LLBA bits */
1891 /* caller might not be interested in sense, but we need it */
1896 use_10_for_ms
= sdev
->use_10_for_ms
;
1898 if (use_10_for_ms
) {
1902 cmd
[0] = MODE_SENSE_10
;
1909 cmd
[0] = MODE_SENSE
;
1914 memset(buffer
, 0, len
);
1916 result
= scsi_execute_req(sdev
, cmd
, DMA_FROM_DEVICE
, buffer
, len
,
1917 sshdr
, timeout
, retries
);
1919 /* This code looks awful: what it's doing is making sure an
1920 * ILLEGAL REQUEST sense return identifies the actual command
1921 * byte as the problem. MODE_SENSE commands can return
1922 * ILLEGAL REQUEST if the code page isn't supported */
1924 if (use_10_for_ms
&& !scsi_status_is_good(result
) &&
1925 (driver_byte(result
) & DRIVER_SENSE
)) {
1926 if (scsi_sense_valid(sshdr
)) {
1927 if ((sshdr
->sense_key
== ILLEGAL_REQUEST
) &&
1928 (sshdr
->asc
== 0x20) && (sshdr
->ascq
== 0)) {
1930 * Invalid command operation code
1932 sdev
->use_10_for_ms
= 0;
1938 if(scsi_status_is_good(result
)) {
1939 if (unlikely(buffer
[0] == 0x86 && buffer
[1] == 0x0b &&
1940 (modepage
== 6 || modepage
== 8))) {
1941 /* Initio breakage? */
1944 data
->medium_type
= 0;
1945 data
->device_specific
= 0;
1947 data
->block_descriptor_length
= 0;
1948 } else if(use_10_for_ms
) {
1949 data
->length
= buffer
[0]*256 + buffer
[1] + 2;
1950 data
->medium_type
= buffer
[2];
1951 data
->device_specific
= buffer
[3];
1952 data
->longlba
= buffer
[4] & 0x01;
1953 data
->block_descriptor_length
= buffer
[6]*256
1956 data
->length
= buffer
[0] + 1;
1957 data
->medium_type
= buffer
[1];
1958 data
->device_specific
= buffer
[2];
1959 data
->block_descriptor_length
= buffer
[3];
1961 data
->header_length
= header_length
;
1966 EXPORT_SYMBOL(scsi_mode_sense
);
1969 scsi_test_unit_ready(struct scsi_device
*sdev
, int timeout
, int retries
)
1972 TEST_UNIT_READY
, 0, 0, 0, 0, 0,
1974 struct scsi_sense_hdr sshdr
;
1977 result
= scsi_execute_req(sdev
, cmd
, DMA_NONE
, NULL
, 0, &sshdr
,
1980 if ((driver_byte(result
) & DRIVER_SENSE
) && sdev
->removable
) {
1982 if ((scsi_sense_valid(&sshdr
)) &&
1983 ((sshdr
.sense_key
== UNIT_ATTENTION
) ||
1984 (sshdr
.sense_key
== NOT_READY
))) {
1991 EXPORT_SYMBOL(scsi_test_unit_ready
);
1994 * scsi_device_set_state - Take the given device through the device
1996 * @sdev: scsi device to change the state of.
1997 * @state: state to change to.
1999 * Returns zero if unsuccessful or an error if the requested
2000 * transition is illegal.
2003 scsi_device_set_state(struct scsi_device
*sdev
, enum scsi_device_state state
)
2005 enum scsi_device_state oldstate
= sdev
->sdev_state
;
2007 if (state
== oldstate
)
2012 /* There are no legal states that come back to
2013 * created. This is the manually initialised start
2087 sdev
->sdev_state
= state
;
2091 SCSI_LOG_ERROR_RECOVERY(1,
2092 sdev_printk(KERN_ERR
, sdev
,
2093 "Illegal state transition %s->%s\n",
2094 scsi_device_state_name(oldstate
),
2095 scsi_device_state_name(state
))
2099 EXPORT_SYMBOL(scsi_device_set_state
);
2102 * scsi_device_quiesce - Block user issued commands.
2103 * @sdev: scsi device to quiesce.
2105 * This works by trying to transition to the SDEV_QUIESCE state
2106 * (which must be a legal transition). When the device is in this
2107 * state, only special requests will be accepted, all others will
2108 * be deferred. Since special requests may also be requeued requests,
2109 * a successful return doesn't guarantee the device will be
2110 * totally quiescent.
2112 * Must be called with user context, may sleep.
2114 * Returns zero if unsuccessful or an error if not.
2117 scsi_device_quiesce(struct scsi_device
*sdev
)
2119 int err
= scsi_device_set_state(sdev
, SDEV_QUIESCE
);
2123 scsi_run_queue(sdev
->request_queue
);
2124 while (sdev
->device_busy
) {
2125 msleep_interruptible(200);
2126 scsi_run_queue(sdev
->request_queue
);
2130 EXPORT_SYMBOL(scsi_device_quiesce
);
2133 * scsi_device_resume - Restart user issued commands to a quiesced device.
2134 * @sdev: scsi device to resume.
2136 * Moves the device from quiesced back to running and restarts the
2139 * Must be called with user context, may sleep.
2142 scsi_device_resume(struct scsi_device
*sdev
)
2144 if(scsi_device_set_state(sdev
, SDEV_RUNNING
))
2146 scsi_run_queue(sdev
->request_queue
);
2148 EXPORT_SYMBOL(scsi_device_resume
);
2151 device_quiesce_fn(struct scsi_device
*sdev
, void *data
)
2153 scsi_device_quiesce(sdev
);
2157 scsi_target_quiesce(struct scsi_target
*starget
)
2159 starget_for_each_device(starget
, NULL
, device_quiesce_fn
);
2161 EXPORT_SYMBOL(scsi_target_quiesce
);
2164 device_resume_fn(struct scsi_device
*sdev
, void *data
)
2166 scsi_device_resume(sdev
);
2170 scsi_target_resume(struct scsi_target
*starget
)
2172 starget_for_each_device(starget
, NULL
, device_resume_fn
);
2174 EXPORT_SYMBOL(scsi_target_resume
);
2177 * scsi_internal_device_block - internal function to put a device
2178 * temporarily into the SDEV_BLOCK state
2179 * @sdev: device to block
2181 * Block request made by scsi lld's to temporarily stop all
2182 * scsi commands on the specified device. Called from interrupt
2183 * or normal process context.
2185 * Returns zero if successful or error if not
2188 * This routine transitions the device to the SDEV_BLOCK state
2189 * (which must be a legal transition). When the device is in this
2190 * state, all commands are deferred until the scsi lld reenables
2191 * the device with scsi_device_unblock or device_block_tmo fires.
2192 * This routine assumes the host_lock is held on entry.
2195 scsi_internal_device_block(struct scsi_device
*sdev
)
2197 struct request_queue
*q
= sdev
->request_queue
;
2198 unsigned long flags
;
2201 err
= scsi_device_set_state(sdev
, SDEV_BLOCK
);
2206 * The device has transitioned to SDEV_BLOCK. Stop the
2207 * block layer from calling the midlayer with this device's
2210 spin_lock_irqsave(q
->queue_lock
, flags
);
2212 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2216 EXPORT_SYMBOL_GPL(scsi_internal_device_block
);
2219 * scsi_internal_device_unblock - resume a device after a block request
2220 * @sdev: device to resume
2222 * Called by scsi lld's or the midlayer to restart the device queue
2223 * for the previously suspended scsi device. Called from interrupt or
2224 * normal process context.
2226 * Returns zero if successful or error if not.
2229 * This routine transitions the device to the SDEV_RUNNING state
2230 * (which must be a legal transition) allowing the midlayer to
2231 * goose the queue for this device. This routine assumes the
2232 * host_lock is held upon entry.
2235 scsi_internal_device_unblock(struct scsi_device
*sdev
)
2237 struct request_queue
*q
= sdev
->request_queue
;
2239 unsigned long flags
;
2242 * Try to transition the scsi device to SDEV_RUNNING
2243 * and goose the device queue if successful.
2245 err
= scsi_device_set_state(sdev
, SDEV_RUNNING
);
2249 spin_lock_irqsave(q
->queue_lock
, flags
);
2251 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2255 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock
);
2258 device_block(struct scsi_device
*sdev
, void *data
)
2260 scsi_internal_device_block(sdev
);
2264 target_block(struct device
*dev
, void *data
)
2266 if (scsi_is_target_device(dev
))
2267 starget_for_each_device(to_scsi_target(dev
), NULL
,
2273 scsi_target_block(struct device
*dev
)
2275 if (scsi_is_target_device(dev
))
2276 starget_for_each_device(to_scsi_target(dev
), NULL
,
2279 device_for_each_child(dev
, NULL
, target_block
);
2281 EXPORT_SYMBOL_GPL(scsi_target_block
);
2284 device_unblock(struct scsi_device
*sdev
, void *data
)
2286 scsi_internal_device_unblock(sdev
);
2290 target_unblock(struct device
*dev
, void *data
)
2292 if (scsi_is_target_device(dev
))
2293 starget_for_each_device(to_scsi_target(dev
), NULL
,
2299 scsi_target_unblock(struct device
*dev
)
2301 if (scsi_is_target_device(dev
))
2302 starget_for_each_device(to_scsi_target(dev
), NULL
,
2305 device_for_each_child(dev
, NULL
, target_unblock
);
2307 EXPORT_SYMBOL_GPL(scsi_target_unblock
);
2310 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2311 * @sg: scatter-gather list
2312 * @sg_count: number of segments in sg
2313 * @offset: offset in bytes into sg, on return offset into the mapped area
2314 * @len: bytes to map, on return number of bytes mapped
2316 * Returns virtual address of the start of the mapped page
2318 void *scsi_kmap_atomic_sg(struct scatterlist
*sgl
, int sg_count
,
2319 size_t *offset
, size_t *len
)
2322 size_t sg_len
= 0, len_complete
= 0;
2323 struct scatterlist
*sg
;
2326 WARN_ON(!irqs_disabled());
2328 for_each_sg(sgl
, sg
, sg_count
, i
) {
2329 len_complete
= sg_len
; /* Complete sg-entries */
2330 sg_len
+= sg
->length
;
2331 if (sg_len
> *offset
)
2335 if (unlikely(i
== sg_count
)) {
2336 printk(KERN_ERR
"%s: Bytes in sg: %zu, requested offset %zu, "
2338 __FUNCTION__
, sg_len
, *offset
, sg_count
);
2343 /* Offset starting from the beginning of first page in this sg-entry */
2344 *offset
= *offset
- len_complete
+ sg
->offset
;
2346 /* Assumption: contiguous pages can be accessed as "page + i" */
2347 page
= nth_page(sg
->page
, (*offset
>> PAGE_SHIFT
));
2348 *offset
&= ~PAGE_MASK
;
2350 /* Bytes in this sg-entry from *offset to the end of the page */
2351 sg_len
= PAGE_SIZE
- *offset
;
2355 return kmap_atomic(page
, KM_BIO_SRC_IRQ
);
2357 EXPORT_SYMBOL(scsi_kmap_atomic_sg
);
2360 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously
2361 * mapped with scsi_kmap_atomic_sg
2362 * @virt: virtual address to be unmapped
2364 void scsi_kunmap_atomic_sg(void *virt
)
2366 kunmap_atomic(virt
, KM_BIO_SRC_IRQ
);
2368 EXPORT_SYMBOL(scsi_kunmap_atomic_sg
);