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>
21 #include <scsi/scsi.h>
22 #include <scsi/scsi_cmnd.h>
23 #include <scsi/scsi_dbg.h>
24 #include <scsi/scsi_device.h>
25 #include <scsi/scsi_driver.h>
26 #include <scsi/scsi_eh.h>
27 #include <scsi/scsi_host.h>
29 #include "scsi_priv.h"
30 #include "scsi_logging.h"
33 #define SG_MEMPOOL_NR ARRAY_SIZE(scsi_sg_pools)
34 #define SG_MEMPOOL_SIZE 32
36 struct scsi_host_sg_pool
{
43 #if (SCSI_MAX_PHYS_SEGMENTS < 32)
44 #error SCSI_MAX_PHYS_SEGMENTS is too small
47 #define SP(x) { x, "sgpool-" #x }
48 static struct scsi_host_sg_pool scsi_sg_pools
[] = {
52 #if (SCSI_MAX_PHYS_SEGMENTS > 32)
54 #if (SCSI_MAX_PHYS_SEGMENTS > 64)
56 #if (SCSI_MAX_PHYS_SEGMENTS > 128)
58 #if (SCSI_MAX_PHYS_SEGMENTS > 256)
59 #error SCSI_MAX_PHYS_SEGMENTS is too large
67 static void scsi_run_queue(struct request_queue
*q
);
70 * Function: scsi_unprep_request()
72 * Purpose: Remove all preparation done for a request, including its
73 * associated scsi_cmnd, so that it can be requeued.
75 * Arguments: req - request to unprepare
77 * Lock status: Assumed that no locks are held upon entry.
81 static void scsi_unprep_request(struct request
*req
)
83 struct scsi_cmnd
*cmd
= req
->special
;
85 req
->flags
&= ~REQ_DONTPREP
;
88 scsi_put_command(cmd
);
92 * Function: scsi_queue_insert()
94 * Purpose: Insert a command in the midlevel queue.
96 * Arguments: cmd - command that we are adding to queue.
97 * reason - why we are inserting command to queue.
99 * Lock status: Assumed that lock is not held upon entry.
103 * Notes: We do this for one of two cases. Either the host is busy
104 * and it cannot accept any more commands for the time being,
105 * or the device returned QUEUE_FULL and can accept no more
107 * Notes: This could be called either from an interrupt context or a
108 * normal process context.
110 int scsi_queue_insert(struct scsi_cmnd
*cmd
, int reason
)
112 struct Scsi_Host
*host
= cmd
->device
->host
;
113 struct scsi_device
*device
= cmd
->device
;
114 struct request_queue
*q
= device
->request_queue
;
118 printk("Inserting command %p into mlqueue\n", cmd
));
121 * Set the appropriate busy bit for the device/host.
123 * If the host/device isn't busy, assume that something actually
124 * completed, and that we should be able to queue a command now.
126 * Note that the prior mid-layer assumption that any host could
127 * always queue at least one command is now broken. The mid-layer
128 * will implement a user specifiable stall (see
129 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
130 * if a command is requeued with no other commands outstanding
131 * either for the device or for the host.
133 if (reason
== SCSI_MLQUEUE_HOST_BUSY
)
134 host
->host_blocked
= host
->max_host_blocked
;
135 else if (reason
== SCSI_MLQUEUE_DEVICE_BUSY
)
136 device
->device_blocked
= device
->max_device_blocked
;
139 * Decrement the counters, since these commands are no longer
140 * active on the host/device.
142 scsi_device_unbusy(device
);
145 * Requeue this command. It will go before all other commands
146 * that are already in the queue.
148 * NOTE: there is magic here about the way the queue is plugged if
149 * we have no outstanding commands.
151 * Although we *don't* plug the queue, we call the request
152 * function. The SCSI request function detects the blocked condition
153 * and plugs the queue appropriately.
155 spin_lock_irqsave(q
->queue_lock
, flags
);
156 blk_requeue_request(q
, cmd
->request
);
157 spin_unlock_irqrestore(q
->queue_lock
, flags
);
165 * scsi_execute - insert request and wait for the result
168 * @data_direction: data direction
169 * @buffer: data buffer
170 * @bufflen: len of buffer
171 * @sense: optional sense buffer
172 * @timeout: request timeout in seconds
173 * @retries: number of times to retry request
174 * @flags: or into request flags;
176 * returns the req->errors value which is the the scsi_cmnd result
179 int scsi_execute(struct scsi_device
*sdev
, const unsigned char *cmd
,
180 int data_direction
, void *buffer
, unsigned bufflen
,
181 unsigned char *sense
, int timeout
, int retries
, int flags
)
184 int write
= (data_direction
== DMA_TO_DEVICE
);
185 int ret
= DRIVER_ERROR
<< 24;
187 req
= blk_get_request(sdev
->request_queue
, write
, __GFP_WAIT
);
189 if (bufflen
&& blk_rq_map_kern(sdev
->request_queue
, req
,
190 buffer
, bufflen
, __GFP_WAIT
))
193 req
->cmd_len
= COMMAND_SIZE(cmd
[0]);
194 memcpy(req
->cmd
, cmd
, req
->cmd_len
);
197 req
->retries
= retries
;
198 req
->timeout
= timeout
;
199 req
->flags
|= flags
| REQ_BLOCK_PC
| REQ_SPECIAL
| REQ_QUIET
;
202 * head injection *required* here otherwise quiesce won't work
204 blk_execute_rq(req
->q
, NULL
, req
, 1);
208 blk_put_request(req
);
212 EXPORT_SYMBOL(scsi_execute
);
215 int scsi_execute_req(struct scsi_device
*sdev
, const unsigned char *cmd
,
216 int data_direction
, void *buffer
, unsigned bufflen
,
217 struct scsi_sense_hdr
*sshdr
, int timeout
, int retries
)
223 sense
= kzalloc(SCSI_SENSE_BUFFERSIZE
, GFP_NOIO
);
225 return DRIVER_ERROR
<< 24;
227 result
= scsi_execute(sdev
, cmd
, data_direction
, buffer
, bufflen
,
228 sense
, timeout
, retries
, 0);
230 scsi_normalize_sense(sense
, SCSI_SENSE_BUFFERSIZE
, sshdr
);
235 EXPORT_SYMBOL(scsi_execute_req
);
237 struct scsi_io_context
{
239 void (*done
)(void *data
, char *sense
, int result
, int resid
);
240 char sense
[SCSI_SENSE_BUFFERSIZE
];
243 static kmem_cache_t
*scsi_io_context_cache
;
245 static void scsi_end_async(struct request
*req
, int uptodate
)
247 struct scsi_io_context
*sioc
= req
->end_io_data
;
250 sioc
->done(sioc
->data
, sioc
->sense
, req
->errors
, req
->data_len
);
252 kmem_cache_free(scsi_io_context_cache
, sioc
);
253 __blk_put_request(req
->q
, req
);
256 static int scsi_merge_bio(struct request
*rq
, struct bio
*bio
)
258 struct request_queue
*q
= rq
->q
;
260 bio
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
261 if (rq_data_dir(rq
) == WRITE
)
262 bio
->bi_rw
|= (1 << BIO_RW
);
263 blk_queue_bounce(q
, &bio
);
266 blk_rq_bio_prep(q
, rq
, bio
);
267 else if (!q
->back_merge_fn(q
, rq
, bio
))
270 rq
->biotail
->bi_next
= bio
;
272 rq
->hard_nr_sectors
+= bio_sectors(bio
);
273 rq
->nr_sectors
= rq
->hard_nr_sectors
;
279 static int scsi_bi_endio(struct bio
*bio
, unsigned int bytes_done
, int error
)
289 * scsi_req_map_sg - map a scatterlist into a request
290 * @rq: request to fill
292 * @nsegs: number of elements
293 * @bufflen: len of buffer
294 * @gfp: memory allocation flags
296 * scsi_req_map_sg maps a scatterlist into a request so that the
297 * request can be sent to the block layer. We do not trust the scatterlist
298 * sent to use, as some ULDs use that struct to only organize the pages.
300 static int scsi_req_map_sg(struct request
*rq
, struct scatterlist
*sgl
,
301 int nsegs
, unsigned bufflen
, gfp_t gfp
)
303 struct request_queue
*q
= rq
->q
;
304 int nr_pages
= (bufflen
+ sgl
[0].offset
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
305 unsigned int data_len
= 0, len
, bytes
, off
;
307 struct bio
*bio
= NULL
;
308 int i
, err
, nr_vecs
= 0;
310 for (i
= 0; i
< nsegs
; i
++) {
317 bytes
= min_t(unsigned int, len
, PAGE_SIZE
- off
);
320 nr_vecs
= min_t(int, BIO_MAX_PAGES
, nr_pages
);
323 bio
= bio_alloc(gfp
, nr_vecs
);
328 bio
->bi_end_io
= scsi_bi_endio
;
331 if (bio_add_pc_page(q
, bio
, page
, bytes
, off
) !=
338 if (bio
->bi_vcnt
>= nr_vecs
) {
339 err
= scsi_merge_bio(rq
, bio
);
341 bio_endio(bio
, bio
->bi_size
, 0);
353 rq
->buffer
= rq
->data
= NULL
;
354 rq
->data_len
= data_len
;
358 while ((bio
= rq
->bio
) != NULL
) {
359 rq
->bio
= bio
->bi_next
;
361 * call endio instead of bio_put incase it was bounced
363 bio_endio(bio
, bio
->bi_size
, 0);
370 * scsi_execute_async - insert request
373 * @cmd_len: length of scsi cdb
374 * @data_direction: data direction
375 * @buffer: data buffer (this can be a kernel buffer or scatterlist)
376 * @bufflen: len of buffer
377 * @use_sg: if buffer is a scatterlist this is the number of elements
378 * @timeout: request timeout in seconds
379 * @retries: number of times to retry request
380 * @flags: or into request flags
382 int scsi_execute_async(struct scsi_device
*sdev
, const unsigned char *cmd
,
383 int cmd_len
, int data_direction
, void *buffer
, unsigned bufflen
,
384 int use_sg
, int timeout
, int retries
, void *privdata
,
385 void (*done
)(void *, char *, int, int), gfp_t gfp
)
388 struct scsi_io_context
*sioc
;
390 int write
= (data_direction
== DMA_TO_DEVICE
);
392 sioc
= kmem_cache_alloc(scsi_io_context_cache
, gfp
);
394 return DRIVER_ERROR
<< 24;
395 memset(sioc
, 0, sizeof(*sioc
));
397 req
= blk_get_request(sdev
->request_queue
, write
, gfp
);
400 req
->flags
|= REQ_BLOCK_PC
| REQ_QUIET
;
403 err
= scsi_req_map_sg(req
, buffer
, use_sg
, bufflen
, gfp
);
405 err
= blk_rq_map_kern(req
->q
, req
, buffer
, bufflen
, gfp
);
410 req
->cmd_len
= cmd_len
;
411 memcpy(req
->cmd
, cmd
, req
->cmd_len
);
412 req
->sense
= sioc
->sense
;
414 req
->timeout
= timeout
;
415 req
->retries
= retries
;
416 req
->end_io_data
= sioc
;
418 sioc
->data
= privdata
;
421 blk_execute_rq_nowait(req
->q
, NULL
, req
, 1, scsi_end_async
);
425 blk_put_request(req
);
428 return DRIVER_ERROR
<< 24;
430 EXPORT_SYMBOL_GPL(scsi_execute_async
);
433 * Function: scsi_init_cmd_errh()
435 * Purpose: Initialize cmd fields related to error handling.
437 * Arguments: cmd - command that is ready to be queued.
441 * Notes: This function has the job of initializing a number of
442 * fields related to error handling. Typically this will
443 * be called once for each command, as required.
445 static int scsi_init_cmd_errh(struct scsi_cmnd
*cmd
)
447 cmd
->serial_number
= 0;
449 memset(cmd
->sense_buffer
, 0, sizeof cmd
->sense_buffer
);
451 if (cmd
->cmd_len
== 0)
452 cmd
->cmd_len
= COMMAND_SIZE(cmd
->cmnd
[0]);
455 * We need saved copies of a number of fields - this is because
456 * error handling may need to overwrite these with different values
457 * to run different commands, and once error handling is complete,
458 * we will need to restore these values prior to running the actual
461 cmd
->old_use_sg
= cmd
->use_sg
;
462 cmd
->old_cmd_len
= cmd
->cmd_len
;
463 cmd
->sc_old_data_direction
= cmd
->sc_data_direction
;
464 cmd
->old_underflow
= cmd
->underflow
;
465 memcpy(cmd
->data_cmnd
, cmd
->cmnd
, sizeof(cmd
->cmnd
));
466 cmd
->buffer
= cmd
->request_buffer
;
467 cmd
->bufflen
= cmd
->request_bufflen
;
473 * Function: scsi_setup_cmd_retry()
475 * Purpose: Restore the command state for a retry
477 * Arguments: cmd - command to be restored
481 * Notes: Immediately prior to retrying a command, we need
482 * to restore certain fields that we saved above.
484 void scsi_setup_cmd_retry(struct scsi_cmnd
*cmd
)
486 memcpy(cmd
->cmnd
, cmd
->data_cmnd
, sizeof(cmd
->data_cmnd
));
487 cmd
->request_buffer
= cmd
->buffer
;
488 cmd
->request_bufflen
= cmd
->bufflen
;
489 cmd
->use_sg
= cmd
->old_use_sg
;
490 cmd
->cmd_len
= cmd
->old_cmd_len
;
491 cmd
->sc_data_direction
= cmd
->sc_old_data_direction
;
492 cmd
->underflow
= cmd
->old_underflow
;
495 void scsi_device_unbusy(struct scsi_device
*sdev
)
497 struct Scsi_Host
*shost
= sdev
->host
;
500 spin_lock_irqsave(shost
->host_lock
, flags
);
502 if (unlikely(scsi_host_in_recovery(shost
) &&
503 (shost
->host_failed
|| shost
->host_eh_scheduled
)))
504 scsi_eh_wakeup(shost
);
505 spin_unlock(shost
->host_lock
);
506 spin_lock(sdev
->request_queue
->queue_lock
);
508 spin_unlock_irqrestore(sdev
->request_queue
->queue_lock
, flags
);
512 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
513 * and call blk_run_queue for all the scsi_devices on the target -
514 * including current_sdev first.
516 * Called with *no* scsi locks held.
518 static void scsi_single_lun_run(struct scsi_device
*current_sdev
)
520 struct Scsi_Host
*shost
= current_sdev
->host
;
521 struct scsi_device
*sdev
, *tmp
;
522 struct scsi_target
*starget
= scsi_target(current_sdev
);
525 spin_lock_irqsave(shost
->host_lock
, flags
);
526 starget
->starget_sdev_user
= NULL
;
527 spin_unlock_irqrestore(shost
->host_lock
, flags
);
530 * Call blk_run_queue for all LUNs on the target, starting with
531 * current_sdev. We race with others (to set starget_sdev_user),
532 * but in most cases, we will be first. Ideally, each LU on the
533 * target would get some limited time or requests on the target.
535 blk_run_queue(current_sdev
->request_queue
);
537 spin_lock_irqsave(shost
->host_lock
, flags
);
538 if (starget
->starget_sdev_user
)
540 list_for_each_entry_safe(sdev
, tmp
, &starget
->devices
,
541 same_target_siblings
) {
542 if (sdev
== current_sdev
)
544 if (scsi_device_get(sdev
))
547 spin_unlock_irqrestore(shost
->host_lock
, flags
);
548 blk_run_queue(sdev
->request_queue
);
549 spin_lock_irqsave(shost
->host_lock
, flags
);
551 scsi_device_put(sdev
);
554 spin_unlock_irqrestore(shost
->host_lock
, flags
);
558 * Function: scsi_run_queue()
560 * Purpose: Select a proper request queue to serve next
562 * Arguments: q - last request's queue
566 * Notes: The previous command was completely finished, start
567 * a new one if possible.
569 static void scsi_run_queue(struct request_queue
*q
)
571 struct scsi_device
*sdev
= q
->queuedata
;
572 struct Scsi_Host
*shost
= sdev
->host
;
575 if (sdev
->single_lun
)
576 scsi_single_lun_run(sdev
);
578 spin_lock_irqsave(shost
->host_lock
, flags
);
579 while (!list_empty(&shost
->starved_list
) &&
580 !shost
->host_blocked
&& !shost
->host_self_blocked
&&
581 !((shost
->can_queue
> 0) &&
582 (shost
->host_busy
>= shost
->can_queue
))) {
584 * As long as shost is accepting commands and we have
585 * starved queues, call blk_run_queue. scsi_request_fn
586 * drops the queue_lock and can add us back to the
589 * host_lock protects the starved_list and starved_entry.
590 * scsi_request_fn must get the host_lock before checking
591 * or modifying starved_list or starved_entry.
593 sdev
= list_entry(shost
->starved_list
.next
,
594 struct scsi_device
, starved_entry
);
595 list_del_init(&sdev
->starved_entry
);
596 spin_unlock_irqrestore(shost
->host_lock
, flags
);
598 blk_run_queue(sdev
->request_queue
);
600 spin_lock_irqsave(shost
->host_lock
, flags
);
601 if (unlikely(!list_empty(&sdev
->starved_entry
)))
603 * sdev lost a race, and was put back on the
604 * starved list. This is unlikely but without this
605 * in theory we could loop forever.
609 spin_unlock_irqrestore(shost
->host_lock
, flags
);
615 * Function: scsi_requeue_command()
617 * Purpose: Handle post-processing of completed commands.
619 * Arguments: q - queue to operate on
620 * cmd - command that may need to be requeued.
624 * Notes: After command completion, there may be blocks left
625 * over which weren't finished by the previous command
626 * this can be for a number of reasons - the main one is
627 * I/O errors in the middle of the request, in which case
628 * we need to request the blocks that come after the bad
630 * Notes: Upon return, cmd is a stale pointer.
632 static void scsi_requeue_command(struct request_queue
*q
, struct scsi_cmnd
*cmd
)
634 struct request
*req
= cmd
->request
;
637 scsi_unprep_request(req
);
638 spin_lock_irqsave(q
->queue_lock
, flags
);
639 blk_requeue_request(q
, req
);
640 spin_unlock_irqrestore(q
->queue_lock
, flags
);
645 void scsi_next_command(struct scsi_cmnd
*cmd
)
647 struct scsi_device
*sdev
= cmd
->device
;
648 struct request_queue
*q
= sdev
->request_queue
;
650 /* need to hold a reference on the device before we let go of the cmd */
651 get_device(&sdev
->sdev_gendev
);
653 scsi_put_command(cmd
);
656 /* ok to remove device now */
657 put_device(&sdev
->sdev_gendev
);
660 void scsi_run_host_queues(struct Scsi_Host
*shost
)
662 struct scsi_device
*sdev
;
664 shost_for_each_device(sdev
, shost
)
665 scsi_run_queue(sdev
->request_queue
);
669 * Function: scsi_end_request()
671 * Purpose: Post-processing of completed commands (usually invoked at end
672 * of upper level post-processing and scsi_io_completion).
674 * Arguments: cmd - command that is complete.
675 * uptodate - 1 if I/O indicates success, <= 0 for I/O error.
676 * bytes - number of bytes of completed I/O
677 * requeue - indicates whether we should requeue leftovers.
679 * Lock status: Assumed that lock is not held upon entry.
681 * Returns: cmd if requeue required, NULL otherwise.
683 * Notes: This is called for block device requests in order to
684 * mark some number of sectors as complete.
686 * We are guaranteeing that the request queue will be goosed
687 * at some point during this call.
688 * Notes: If cmd was requeued, upon return it will be a stale pointer.
690 static struct scsi_cmnd
*scsi_end_request(struct scsi_cmnd
*cmd
, int uptodate
,
691 int bytes
, int requeue
)
693 request_queue_t
*q
= cmd
->device
->request_queue
;
694 struct request
*req
= cmd
->request
;
698 * If there are blocks left over at the end, set up the command
699 * to queue the remainder of them.
701 if (end_that_request_chunk(req
, uptodate
, bytes
)) {
702 int leftover
= (req
->hard_nr_sectors
<< 9);
704 if (blk_pc_request(req
))
705 leftover
= req
->data_len
;
707 /* kill remainder if no retrys */
708 if (!uptodate
&& blk_noretry_request(req
))
709 end_that_request_chunk(req
, 0, leftover
);
713 * Bleah. Leftovers again. Stick the
714 * leftovers in the front of the
715 * queue, and goose the queue again.
717 scsi_requeue_command(q
, cmd
);
724 add_disk_randomness(req
->rq_disk
);
726 spin_lock_irqsave(q
->queue_lock
, flags
);
727 if (blk_rq_tagged(req
))
728 blk_queue_end_tag(q
, req
);
729 end_that_request_last(req
, uptodate
);
730 spin_unlock_irqrestore(q
->queue_lock
, flags
);
733 * This will goose the queue request function at the end, so we don't
734 * need to worry about launching another command.
736 scsi_next_command(cmd
);
740 static struct scatterlist
*scsi_alloc_sgtable(struct scsi_cmnd
*cmd
, gfp_t gfp_mask
)
742 struct scsi_host_sg_pool
*sgp
;
743 struct scatterlist
*sgl
;
745 BUG_ON(!cmd
->use_sg
);
747 switch (cmd
->use_sg
) {
757 #if (SCSI_MAX_PHYS_SEGMENTS > 32)
761 #if (SCSI_MAX_PHYS_SEGMENTS > 64)
765 #if (SCSI_MAX_PHYS_SEGMENTS > 128)
776 sgp
= scsi_sg_pools
+ cmd
->sglist_len
;
777 sgl
= mempool_alloc(sgp
->pool
, gfp_mask
);
781 static void scsi_free_sgtable(struct scatterlist
*sgl
, int index
)
783 struct scsi_host_sg_pool
*sgp
;
785 BUG_ON(index
>= SG_MEMPOOL_NR
);
787 sgp
= scsi_sg_pools
+ index
;
788 mempool_free(sgl
, sgp
->pool
);
792 * Function: scsi_release_buffers()
794 * Purpose: Completion processing for block device I/O requests.
796 * Arguments: cmd - command that we are bailing.
798 * Lock status: Assumed that no lock is held upon entry.
802 * Notes: In the event that an upper level driver rejects a
803 * command, we must release resources allocated during
804 * the __init_io() function. Primarily this would involve
805 * the scatter-gather table, and potentially any bounce
808 static void scsi_release_buffers(struct scsi_cmnd
*cmd
)
810 struct request
*req
= cmd
->request
;
813 * Free up any indirection buffers we allocated for DMA purposes.
816 scsi_free_sgtable(cmd
->request_buffer
, cmd
->sglist_len
);
817 else if (cmd
->request_buffer
!= req
->buffer
)
818 kfree(cmd
->request_buffer
);
821 * Zero these out. They now point to freed memory, and it is
822 * dangerous to hang onto the pointers.
826 cmd
->request_buffer
= NULL
;
827 cmd
->request_bufflen
= 0;
831 * Function: scsi_io_completion()
833 * Purpose: Completion processing for block device I/O requests.
835 * Arguments: cmd - command that is finished.
837 * Lock status: Assumed that no lock is held upon entry.
841 * Notes: This function is matched in terms of capabilities to
842 * the function that created the scatter-gather list.
843 * In other words, if there are no bounce buffers
844 * (the normal case for most drivers), we don't need
845 * the logic to deal with cleaning up afterwards.
847 * We must do one of several things here:
849 * a) Call scsi_end_request. This will finish off the
850 * specified number of sectors. If we are done, the
851 * command block will be released, and the queue
852 * function will be goosed. If we are not done, then
853 * scsi_end_request will directly goose the queue.
855 * b) We can just use scsi_requeue_command() here. This would
856 * be used if we just wanted to retry, for example.
858 void scsi_io_completion(struct scsi_cmnd
*cmd
, unsigned int good_bytes
)
860 int result
= cmd
->result
;
861 int this_count
= cmd
->bufflen
;
862 request_queue_t
*q
= cmd
->device
->request_queue
;
863 struct request
*req
= cmd
->request
;
864 int clear_errors
= 1;
865 struct scsi_sense_hdr sshdr
;
867 int sense_deferred
= 0;
870 * Free up any indirection buffers we allocated for DMA purposes.
871 * For the case of a READ, we need to copy the data out of the
872 * bounce buffer and into the real buffer.
875 scsi_free_sgtable(cmd
->buffer
, cmd
->sglist_len
);
876 else if (cmd
->buffer
!= req
->buffer
) {
877 if (rq_data_dir(req
) == READ
) {
879 char *to
= bio_kmap_irq(req
->bio
, &flags
);
880 memcpy(to
, cmd
->buffer
, cmd
->bufflen
);
881 bio_kunmap_irq(to
, &flags
);
887 sense_valid
= scsi_command_normalize_sense(cmd
, &sshdr
);
889 sense_deferred
= scsi_sense_is_deferred(&sshdr
);
891 if (blk_pc_request(req
)) { /* SG_IO ioctl from block level */
892 req
->errors
= result
;
895 if (sense_valid
&& req
->sense
) {
897 * SG_IO wants current and deferred errors
899 int len
= 8 + cmd
->sense_buffer
[7];
901 if (len
> SCSI_SENSE_BUFFERSIZE
)
902 len
= SCSI_SENSE_BUFFERSIZE
;
903 memcpy(req
->sense
, cmd
->sense_buffer
, len
);
904 req
->sense_len
= len
;
907 req
->data_len
= cmd
->resid
;
911 * Zero these out. They now point to freed memory, and it is
912 * dangerous to hang onto the pointers.
916 cmd
->request_buffer
= NULL
;
917 cmd
->request_bufflen
= 0;
920 * Next deal with any sectors which we were able to correctly
923 if (good_bytes
> 0) {
924 SCSI_LOG_HLCOMPLETE(1, printk("%ld sectors total, "
926 req
->nr_sectors
, good_bytes
));
927 SCSI_LOG_HLCOMPLETE(1, printk("use_sg is %d\n", cmd
->use_sg
));
932 /* A number of bytes were successfully read. If there
933 * is leftovers and there is some kind of error
934 * (result != 0), retry the rest.
936 if (scsi_end_request(cmd
, 1, good_bytes
, !!result
) == NULL
)
940 /* good_bytes = 0, or (inclusive) there were leftovers and
941 * result = 0, so scsi_end_request couldn't retry.
943 if (sense_valid
&& !sense_deferred
) {
944 switch (sshdr
.sense_key
) {
946 if (cmd
->device
->removable
) {
947 /* Detected disc change. Set a bit
948 * and quietly refuse further access.
950 cmd
->device
->changed
= 1;
951 scsi_end_request(cmd
, 0, this_count
, 1);
954 /* Must have been a power glitch, or a
955 * bus reset. Could not have been a
956 * media change, so we just retry the
957 * request and see what happens.
959 scsi_requeue_command(q
, cmd
);
963 case ILLEGAL_REQUEST
:
964 /* If we had an ILLEGAL REQUEST returned, then
965 * we may have performed an unsupported
966 * command. The only thing this should be
967 * would be a ten byte read where only a six
968 * byte read was supported. Also, on a system
969 * where READ CAPACITY failed, we may have
970 * read past the end of the disk.
972 if ((cmd
->device
->use_10_for_rw
&&
973 sshdr
.asc
== 0x20 && sshdr
.ascq
== 0x00) &&
974 (cmd
->cmnd
[0] == READ_10
||
975 cmd
->cmnd
[0] == WRITE_10
)) {
976 cmd
->device
->use_10_for_rw
= 0;
977 /* This will cause a retry with a
980 scsi_requeue_command(q
, cmd
);
983 scsi_end_request(cmd
, 0, this_count
, 1);
988 /* If the device is in the process of becoming
989 * ready, or has a temporary blockage, retry.
991 if (sshdr
.asc
== 0x04) {
992 switch (sshdr
.ascq
) {
993 case 0x01: /* becoming ready */
994 case 0x04: /* format in progress */
995 case 0x05: /* rebuild in progress */
996 case 0x06: /* recalculation in progress */
997 case 0x07: /* operation in progress */
998 case 0x08: /* Long write in progress */
999 case 0x09: /* self test in progress */
1000 scsi_requeue_command(q
, cmd
);
1006 if (!(req
->flags
& REQ_QUIET
)) {
1007 scmd_printk(KERN_INFO
, cmd
,
1008 "Device not ready: ");
1009 scsi_print_sense_hdr("", &sshdr
);
1011 scsi_end_request(cmd
, 0, this_count
, 1);
1013 case VOLUME_OVERFLOW
:
1014 if (!(req
->flags
& REQ_QUIET
)) {
1015 scmd_printk(KERN_INFO
, cmd
,
1016 "Volume overflow, CDB: ");
1017 __scsi_print_command(cmd
->data_cmnd
);
1018 scsi_print_sense("", cmd
);
1020 /* See SSC3rXX or current. */
1021 scsi_end_request(cmd
, 0, this_count
, 1);
1027 if (host_byte(result
) == DID_RESET
) {
1028 /* Third party bus reset or reset for error recovery
1029 * reasons. Just retry the request and see what
1032 scsi_requeue_command(q
, cmd
);
1036 if (!(req
->flags
& REQ_QUIET
)) {
1037 scmd_printk(KERN_INFO
, cmd
,
1038 "SCSI error: return code = 0x%08x\n",
1040 if (driver_byte(result
) & DRIVER_SENSE
)
1041 scsi_print_sense("", cmd
);
1044 scsi_end_request(cmd
, 0, this_count
, !result
);
1046 EXPORT_SYMBOL(scsi_io_completion
);
1049 * Function: scsi_init_io()
1051 * Purpose: SCSI I/O initialize function.
1053 * Arguments: cmd - Command descriptor we wish to initialize
1055 * Returns: 0 on success
1056 * BLKPREP_DEFER if the failure is retryable
1057 * BLKPREP_KILL if the failure is fatal
1059 static int scsi_init_io(struct scsi_cmnd
*cmd
)
1061 struct request
*req
= cmd
->request
;
1062 struct scatterlist
*sgpnt
;
1066 * if this is a rq->data based REQ_BLOCK_PC, setup for a non-sg xfer
1068 if ((req
->flags
& REQ_BLOCK_PC
) && !req
->bio
) {
1069 cmd
->request_bufflen
= req
->data_len
;
1070 cmd
->request_buffer
= req
->data
;
1071 req
->buffer
= req
->data
;
1077 * we used to not use scatter-gather for single segment request,
1078 * but now we do (it makes highmem I/O easier to support without
1081 cmd
->use_sg
= req
->nr_phys_segments
;
1084 * if sg table allocation fails, requeue request later.
1086 sgpnt
= scsi_alloc_sgtable(cmd
, GFP_ATOMIC
);
1087 if (unlikely(!sgpnt
)) {
1088 scsi_unprep_request(req
);
1089 return BLKPREP_DEFER
;
1092 cmd
->request_buffer
= (char *) sgpnt
;
1093 cmd
->request_bufflen
= req
->nr_sectors
<< 9;
1094 if (blk_pc_request(req
))
1095 cmd
->request_bufflen
= req
->data_len
;
1099 * Next, walk the list, and fill in the addresses and sizes of
1102 count
= blk_rq_map_sg(req
->q
, req
, cmd
->request_buffer
);
1105 * mapped well, send it off
1107 if (likely(count
<= cmd
->use_sg
)) {
1108 cmd
->use_sg
= count
;
1112 printk(KERN_ERR
"Incorrect number of segments after building list\n");
1113 printk(KERN_ERR
"counted %d, received %d\n", count
, cmd
->use_sg
);
1114 printk(KERN_ERR
"req nr_sec %lu, cur_nr_sec %u\n", req
->nr_sectors
,
1115 req
->current_nr_sectors
);
1117 /* release the command and kill it */
1118 scsi_release_buffers(cmd
);
1119 scsi_put_command(cmd
);
1120 return BLKPREP_KILL
;
1123 static int scsi_issue_flush_fn(request_queue_t
*q
, struct gendisk
*disk
,
1124 sector_t
*error_sector
)
1126 struct scsi_device
*sdev
= q
->queuedata
;
1127 struct scsi_driver
*drv
;
1129 if (sdev
->sdev_state
!= SDEV_RUNNING
)
1132 drv
= *(struct scsi_driver
**) disk
->private_data
;
1133 if (drv
->issue_flush
)
1134 return drv
->issue_flush(&sdev
->sdev_gendev
, error_sector
);
1139 static void scsi_blk_pc_done(struct scsi_cmnd
*cmd
)
1141 BUG_ON(!blk_pc_request(cmd
->request
));
1143 * This will complete the whole command with uptodate=1 so
1144 * as far as the block layer is concerned the command completed
1145 * successfully. Since this is a REQ_BLOCK_PC command the
1146 * caller should check the request's errors value
1148 scsi_io_completion(cmd
, cmd
->bufflen
);
1151 static void scsi_setup_blk_pc_cmnd(struct scsi_cmnd
*cmd
)
1153 struct request
*req
= cmd
->request
;
1155 BUG_ON(sizeof(req
->cmd
) > sizeof(cmd
->cmnd
));
1156 memcpy(cmd
->cmnd
, req
->cmd
, sizeof(cmd
->cmnd
));
1157 cmd
->cmd_len
= req
->cmd_len
;
1159 cmd
->sc_data_direction
= DMA_NONE
;
1160 else if (rq_data_dir(req
) == WRITE
)
1161 cmd
->sc_data_direction
= DMA_TO_DEVICE
;
1163 cmd
->sc_data_direction
= DMA_FROM_DEVICE
;
1165 cmd
->transfersize
= req
->data_len
;
1166 cmd
->allowed
= req
->retries
;
1167 cmd
->timeout_per_command
= req
->timeout
;
1168 cmd
->done
= scsi_blk_pc_done
;
1171 static int scsi_prep_fn(struct request_queue
*q
, struct request
*req
)
1173 struct scsi_device
*sdev
= q
->queuedata
;
1174 struct scsi_cmnd
*cmd
;
1175 int specials_only
= 0;
1178 * Just check to see if the device is online. If it isn't, we
1179 * refuse to process any commands. The device must be brought
1180 * online before trying any recovery commands
1182 if (unlikely(!scsi_device_online(sdev
))) {
1183 sdev_printk(KERN_ERR
, sdev
,
1184 "rejecting I/O to offline device\n");
1187 if (unlikely(sdev
->sdev_state
!= SDEV_RUNNING
)) {
1188 /* OK, we're not in a running state don't prep
1190 if (sdev
->sdev_state
== SDEV_DEL
) {
1191 /* Device is fully deleted, no commands
1192 * at all allowed down */
1193 sdev_printk(KERN_ERR
, sdev
,
1194 "rejecting I/O to dead device\n");
1197 /* OK, we only allow special commands (i.e. not
1198 * user initiated ones */
1199 specials_only
= sdev
->sdev_state
;
1203 * Find the actual device driver associated with this command.
1204 * The SPECIAL requests are things like character device or
1205 * ioctls, which did not originate from ll_rw_blk. Note that
1206 * the special field is also used to indicate the cmd for
1207 * the remainder of a partially fulfilled request that can
1208 * come up when there is a medium error. We have to treat
1209 * these two cases differently. We differentiate by looking
1210 * at request->cmd, as this tells us the real story.
1212 if (req
->flags
& REQ_SPECIAL
&& req
->special
) {
1214 } else if (req
->flags
& (REQ_CMD
| REQ_BLOCK_PC
)) {
1216 if(unlikely(specials_only
) && !(req
->flags
& REQ_SPECIAL
)) {
1217 if(specials_only
== SDEV_QUIESCE
||
1218 specials_only
== SDEV_BLOCK
)
1221 sdev_printk(KERN_ERR
, sdev
,
1222 "rejecting I/O to device being removed\n");
1228 * Now try and find a command block that we can use.
1230 if (!req
->special
) {
1231 cmd
= scsi_get_command(sdev
, GFP_ATOMIC
);
1237 /* pull a tag out of the request if we have one */
1238 cmd
->tag
= req
->tag
;
1240 blk_dump_rq_flags(req
, "SCSI bad req");
1244 /* note the overloading of req->special. When the tag
1245 * is active it always means cmd. If the tag goes
1246 * back for re-queueing, it may be reset */
1251 * FIXME: drop the lock here because the functions below
1252 * expect to be called without the queue lock held. Also,
1253 * previously, we dequeued the request before dropping the
1254 * lock. We hope REQ_STARTED prevents anything untoward from
1257 if (req
->flags
& (REQ_CMD
| REQ_BLOCK_PC
)) {
1261 * This will do a couple of things:
1262 * 1) Fill in the actual SCSI command.
1263 * 2) Fill in any other upper-level specific fields
1266 * If this returns 0, it means that the request failed
1267 * (reading past end of disk, reading offline device,
1268 * etc). This won't actually talk to the device, but
1269 * some kinds of consistency checking may cause the
1270 * request to be rejected immediately.
1274 * This sets up the scatter-gather table (allocating if
1277 ret
= scsi_init_io(cmd
);
1279 /* For BLKPREP_KILL/DEFER the cmd was released */
1287 * Initialize the actual SCSI command for this request.
1289 if (req
->flags
& REQ_BLOCK_PC
) {
1290 scsi_setup_blk_pc_cmnd(cmd
);
1291 } else if (req
->rq_disk
) {
1292 struct scsi_driver
*drv
;
1294 drv
= *(struct scsi_driver
**)req
->rq_disk
->private_data
;
1295 if (unlikely(!drv
->init_command(cmd
))) {
1296 scsi_release_buffers(cmd
);
1297 scsi_put_command(cmd
);
1304 * The request is now prepped, no need to come back here
1306 req
->flags
|= REQ_DONTPREP
;
1310 /* If we defer, the elv_next_request() returns NULL, but the
1311 * queue must be restarted, so we plug here if no returning
1312 * command will automatically do that. */
1313 if (sdev
->device_busy
== 0)
1315 return BLKPREP_DEFER
;
1317 req
->errors
= DID_NO_CONNECT
<< 16;
1318 return BLKPREP_KILL
;
1322 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1325 * Called with the queue_lock held.
1327 static inline int scsi_dev_queue_ready(struct request_queue
*q
,
1328 struct scsi_device
*sdev
)
1330 if (sdev
->device_busy
>= sdev
->queue_depth
)
1332 if (sdev
->device_busy
== 0 && sdev
->device_blocked
) {
1334 * unblock after device_blocked iterates to zero
1336 if (--sdev
->device_blocked
== 0) {
1338 sdev_printk(KERN_INFO
, sdev
,
1339 "unblocking device at zero depth\n"));
1345 if (sdev
->device_blocked
)
1352 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1353 * return 0. We must end up running the queue again whenever 0 is
1354 * returned, else IO can hang.
1356 * Called with host_lock held.
1358 static inline int scsi_host_queue_ready(struct request_queue
*q
,
1359 struct Scsi_Host
*shost
,
1360 struct scsi_device
*sdev
)
1362 if (scsi_host_in_recovery(shost
))
1364 if (shost
->host_busy
== 0 && shost
->host_blocked
) {
1366 * unblock after host_blocked iterates to zero
1368 if (--shost
->host_blocked
== 0) {
1370 printk("scsi%d unblocking host at zero depth\n",
1377 if ((shost
->can_queue
> 0 && shost
->host_busy
>= shost
->can_queue
) ||
1378 shost
->host_blocked
|| shost
->host_self_blocked
) {
1379 if (list_empty(&sdev
->starved_entry
))
1380 list_add_tail(&sdev
->starved_entry
, &shost
->starved_list
);
1384 /* We're OK to process the command, so we can't be starved */
1385 if (!list_empty(&sdev
->starved_entry
))
1386 list_del_init(&sdev
->starved_entry
);
1392 * Kill a request for a dead device
1394 static void scsi_kill_request(struct request
*req
, request_queue_t
*q
)
1396 struct scsi_cmnd
*cmd
= req
->special
;
1397 struct scsi_device
*sdev
= cmd
->device
;
1398 struct Scsi_Host
*shost
= sdev
->host
;
1400 blkdev_dequeue_request(req
);
1402 if (unlikely(cmd
== NULL
)) {
1403 printk(KERN_CRIT
"impossible request in %s.\n",
1408 scsi_init_cmd_errh(cmd
);
1409 cmd
->result
= DID_NO_CONNECT
<< 16;
1410 atomic_inc(&cmd
->device
->iorequest_cnt
);
1413 * SCSI request completion path will do scsi_device_unbusy(),
1414 * bump busy counts. To bump the counters, we need to dance
1415 * with the locks as normal issue path does.
1417 sdev
->device_busy
++;
1418 spin_unlock(sdev
->request_queue
->queue_lock
);
1419 spin_lock(shost
->host_lock
);
1421 spin_unlock(shost
->host_lock
);
1422 spin_lock(sdev
->request_queue
->queue_lock
);
1427 static void scsi_softirq_done(struct request
*rq
)
1429 struct scsi_cmnd
*cmd
= rq
->completion_data
;
1430 unsigned long wait_for
= (cmd
->allowed
+ 1) * cmd
->timeout_per_command
;
1433 INIT_LIST_HEAD(&cmd
->eh_entry
);
1435 disposition
= scsi_decide_disposition(cmd
);
1436 if (disposition
!= SUCCESS
&&
1437 time_before(cmd
->jiffies_at_alloc
+ wait_for
, jiffies
)) {
1438 sdev_printk(KERN_ERR
, cmd
->device
,
1439 "timing out command, waited %lus\n",
1441 disposition
= SUCCESS
;
1444 scsi_log_completion(cmd
, disposition
);
1446 switch (disposition
) {
1448 scsi_finish_command(cmd
);
1451 scsi_retry_command(cmd
);
1453 case ADD_TO_MLQUEUE
:
1454 scsi_queue_insert(cmd
, SCSI_MLQUEUE_DEVICE_BUSY
);
1457 if (!scsi_eh_scmd_add(cmd
, 0))
1458 scsi_finish_command(cmd
);
1463 * Function: scsi_request_fn()
1465 * Purpose: Main strategy routine for SCSI.
1467 * Arguments: q - Pointer to actual queue.
1471 * Lock status: IO request lock assumed to be held when called.
1473 static void scsi_request_fn(struct request_queue
*q
)
1475 struct scsi_device
*sdev
= q
->queuedata
;
1476 struct Scsi_Host
*shost
;
1477 struct scsi_cmnd
*cmd
;
1478 struct request
*req
;
1481 printk("scsi: killing requests for dead queue\n");
1482 while ((req
= elv_next_request(q
)) != NULL
)
1483 scsi_kill_request(req
, q
);
1487 if(!get_device(&sdev
->sdev_gendev
))
1488 /* We must be tearing the block queue down already */
1492 * To start with, we keep looping until the queue is empty, or until
1493 * the host is no longer able to accept any more requests.
1496 while (!blk_queue_plugged(q
)) {
1499 * get next queueable request. We do this early to make sure
1500 * that the request is fully prepared even if we cannot
1503 req
= elv_next_request(q
);
1504 if (!req
|| !scsi_dev_queue_ready(q
, sdev
))
1507 if (unlikely(!scsi_device_online(sdev
))) {
1508 sdev_printk(KERN_ERR
, sdev
,
1509 "rejecting I/O to offline device\n");
1510 scsi_kill_request(req
, q
);
1516 * Remove the request from the request list.
1518 if (!(blk_queue_tagged(q
) && !blk_queue_start_tag(q
, req
)))
1519 blkdev_dequeue_request(req
);
1520 sdev
->device_busy
++;
1522 spin_unlock(q
->queue_lock
);
1524 if (unlikely(cmd
== NULL
)) {
1525 printk(KERN_CRIT
"impossible request in %s.\n"
1526 "please mail a stack trace to "
1527 "linux-scsi@vger.kernel.org",
1531 spin_lock(shost
->host_lock
);
1533 if (!scsi_host_queue_ready(q
, shost
, sdev
))
1535 if (sdev
->single_lun
) {
1536 if (scsi_target(sdev
)->starget_sdev_user
&&
1537 scsi_target(sdev
)->starget_sdev_user
!= sdev
)
1539 scsi_target(sdev
)->starget_sdev_user
= sdev
;
1544 * XXX(hch): This is rather suboptimal, scsi_dispatch_cmd will
1545 * take the lock again.
1547 spin_unlock_irq(shost
->host_lock
);
1550 * Finally, initialize any error handling parameters, and set up
1551 * the timers for timeouts.
1553 scsi_init_cmd_errh(cmd
);
1556 * Dispatch the command to the low-level driver.
1558 rtn
= scsi_dispatch_cmd(cmd
);
1559 spin_lock_irq(q
->queue_lock
);
1561 /* we're refusing the command; because of
1562 * the way locks get dropped, we need to
1563 * check here if plugging is required */
1564 if(sdev
->device_busy
== 0)
1574 spin_unlock_irq(shost
->host_lock
);
1577 * lock q, handle tag, requeue req, and decrement device_busy. We
1578 * must return with queue_lock held.
1580 * Decrementing device_busy without checking it is OK, as all such
1581 * cases (host limits or settings) should run the queue at some
1584 spin_lock_irq(q
->queue_lock
);
1585 blk_requeue_request(q
, req
);
1586 sdev
->device_busy
--;
1587 if(sdev
->device_busy
== 0)
1590 /* must be careful here...if we trigger the ->remove() function
1591 * we cannot be holding the q lock */
1592 spin_unlock_irq(q
->queue_lock
);
1593 put_device(&sdev
->sdev_gendev
);
1594 spin_lock_irq(q
->queue_lock
);
1597 u64
scsi_calculate_bounce_limit(struct Scsi_Host
*shost
)
1599 struct device
*host_dev
;
1600 u64 bounce_limit
= 0xffffffff;
1602 if (shost
->unchecked_isa_dma
)
1603 return BLK_BOUNCE_ISA
;
1605 * Platforms with virtual-DMA translation
1606 * hardware have no practical limit.
1608 if (!PCI_DMA_BUS_IS_PHYS
)
1609 return BLK_BOUNCE_ANY
;
1611 host_dev
= scsi_get_device(shost
);
1612 if (host_dev
&& host_dev
->dma_mask
)
1613 bounce_limit
= *host_dev
->dma_mask
;
1615 return bounce_limit
;
1617 EXPORT_SYMBOL(scsi_calculate_bounce_limit
);
1619 struct request_queue
*scsi_alloc_queue(struct scsi_device
*sdev
)
1621 struct Scsi_Host
*shost
= sdev
->host
;
1622 struct request_queue
*q
;
1624 q
= blk_init_queue(scsi_request_fn
, NULL
);
1628 blk_queue_prep_rq(q
, scsi_prep_fn
);
1630 blk_queue_max_hw_segments(q
, shost
->sg_tablesize
);
1631 blk_queue_max_phys_segments(q
, SCSI_MAX_PHYS_SEGMENTS
);
1632 blk_queue_max_sectors(q
, shost
->max_sectors
);
1633 blk_queue_bounce_limit(q
, scsi_calculate_bounce_limit(shost
));
1634 blk_queue_segment_boundary(q
, shost
->dma_boundary
);
1635 blk_queue_issue_flush_fn(q
, scsi_issue_flush_fn
);
1636 blk_queue_softirq_done(q
, scsi_softirq_done
);
1638 if (!shost
->use_clustering
)
1639 clear_bit(QUEUE_FLAG_CLUSTER
, &q
->queue_flags
);
1643 void scsi_free_queue(struct request_queue
*q
)
1645 blk_cleanup_queue(q
);
1649 * Function: scsi_block_requests()
1651 * Purpose: Utility function used by low-level drivers to prevent further
1652 * commands from being queued to the device.
1654 * Arguments: shost - Host in question
1658 * Lock status: No locks are assumed held.
1660 * Notes: There is no timer nor any other means by which the requests
1661 * get unblocked other than the low-level driver calling
1662 * scsi_unblock_requests().
1664 void scsi_block_requests(struct Scsi_Host
*shost
)
1666 shost
->host_self_blocked
= 1;
1668 EXPORT_SYMBOL(scsi_block_requests
);
1671 * Function: scsi_unblock_requests()
1673 * Purpose: Utility function used by low-level drivers to allow further
1674 * commands from being queued to the device.
1676 * Arguments: shost - Host in question
1680 * Lock status: No locks are assumed held.
1682 * Notes: There is no timer nor any other means by which the requests
1683 * get unblocked other than the low-level driver calling
1684 * scsi_unblock_requests().
1686 * This is done as an API function so that changes to the
1687 * internals of the scsi mid-layer won't require wholesale
1688 * changes to drivers that use this feature.
1690 void scsi_unblock_requests(struct Scsi_Host
*shost
)
1692 shost
->host_self_blocked
= 0;
1693 scsi_run_host_queues(shost
);
1695 EXPORT_SYMBOL(scsi_unblock_requests
);
1697 int __init
scsi_init_queue(void)
1701 scsi_io_context_cache
= kmem_cache_create("scsi_io_context",
1702 sizeof(struct scsi_io_context
),
1704 if (!scsi_io_context_cache
) {
1705 printk(KERN_ERR
"SCSI: can't init scsi io context cache\n");
1709 for (i
= 0; i
< SG_MEMPOOL_NR
; i
++) {
1710 struct scsi_host_sg_pool
*sgp
= scsi_sg_pools
+ i
;
1711 int size
= sgp
->size
* sizeof(struct scatterlist
);
1713 sgp
->slab
= kmem_cache_create(sgp
->name
, size
, 0,
1714 SLAB_HWCACHE_ALIGN
, NULL
, NULL
);
1716 printk(KERN_ERR
"SCSI: can't init sg slab %s\n",
1720 sgp
->pool
= mempool_create_slab_pool(SG_MEMPOOL_SIZE
,
1723 printk(KERN_ERR
"SCSI: can't init sg mempool %s\n",
1731 void scsi_exit_queue(void)
1735 kmem_cache_destroy(scsi_io_context_cache
);
1737 for (i
= 0; i
< SG_MEMPOOL_NR
; i
++) {
1738 struct scsi_host_sg_pool
*sgp
= scsi_sg_pools
+ i
;
1739 mempool_destroy(sgp
->pool
);
1740 kmem_cache_destroy(sgp
->slab
);
1745 * scsi_mode_select - issue a mode select
1746 * @sdev: SCSI device to be queried
1747 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1748 * @sp: Save page bit (0 == don't save, 1 == save)
1749 * @modepage: mode page being requested
1750 * @buffer: request buffer (may not be smaller than eight bytes)
1751 * @len: length of request buffer.
1752 * @timeout: command timeout
1753 * @retries: number of retries before failing
1754 * @data: returns a structure abstracting the mode header data
1755 * @sense: place to put sense data (or NULL if no sense to be collected).
1756 * must be SCSI_SENSE_BUFFERSIZE big.
1758 * Returns zero if successful; negative error number or scsi
1763 scsi_mode_select(struct scsi_device
*sdev
, int pf
, int sp
, int modepage
,
1764 unsigned char *buffer
, int len
, int timeout
, int retries
,
1765 struct scsi_mode_data
*data
, struct scsi_sense_hdr
*sshdr
)
1767 unsigned char cmd
[10];
1768 unsigned char *real_buffer
;
1771 memset(cmd
, 0, sizeof(cmd
));
1772 cmd
[1] = (pf
? 0x10 : 0) | (sp
? 0x01 : 0);
1774 if (sdev
->use_10_for_ms
) {
1777 real_buffer
= kmalloc(8 + len
, GFP_KERNEL
);
1780 memcpy(real_buffer
+ 8, buffer
, len
);
1784 real_buffer
[2] = data
->medium_type
;
1785 real_buffer
[3] = data
->device_specific
;
1786 real_buffer
[4] = data
->longlba
? 0x01 : 0;
1788 real_buffer
[6] = data
->block_descriptor_length
>> 8;
1789 real_buffer
[7] = data
->block_descriptor_length
;
1791 cmd
[0] = MODE_SELECT_10
;
1795 if (len
> 255 || data
->block_descriptor_length
> 255 ||
1799 real_buffer
= kmalloc(4 + len
, GFP_KERNEL
);
1802 memcpy(real_buffer
+ 4, buffer
, len
);
1805 real_buffer
[1] = data
->medium_type
;
1806 real_buffer
[2] = data
->device_specific
;
1807 real_buffer
[3] = data
->block_descriptor_length
;
1810 cmd
[0] = MODE_SELECT
;
1814 ret
= scsi_execute_req(sdev
, cmd
, DMA_TO_DEVICE
, real_buffer
, len
,
1815 sshdr
, timeout
, retries
);
1819 EXPORT_SYMBOL_GPL(scsi_mode_select
);
1822 * scsi_mode_sense - issue a mode sense, falling back from 10 to
1823 * six bytes if necessary.
1824 * @sdev: SCSI device to be queried
1825 * @dbd: set if mode sense will allow block descriptors to be returned
1826 * @modepage: mode page being requested
1827 * @buffer: request buffer (may not be smaller than eight bytes)
1828 * @len: length of request buffer.
1829 * @timeout: command timeout
1830 * @retries: number of retries before failing
1831 * @data: returns a structure abstracting the mode header data
1832 * @sense: place to put sense data (or NULL if no sense to be collected).
1833 * must be SCSI_SENSE_BUFFERSIZE big.
1835 * Returns zero if unsuccessful, or the header offset (either 4
1836 * or 8 depending on whether a six or ten byte command was
1837 * issued) if successful.
1840 scsi_mode_sense(struct scsi_device
*sdev
, int dbd
, int modepage
,
1841 unsigned char *buffer
, int len
, int timeout
, int retries
,
1842 struct scsi_mode_data
*data
, struct scsi_sense_hdr
*sshdr
)
1844 unsigned char cmd
[12];
1848 struct scsi_sense_hdr my_sshdr
;
1850 memset(data
, 0, sizeof(*data
));
1851 memset(&cmd
[0], 0, 12);
1852 cmd
[1] = dbd
& 0x18; /* allows DBD and LLBA bits */
1855 /* caller might not be interested in sense, but we need it */
1860 use_10_for_ms
= sdev
->use_10_for_ms
;
1862 if (use_10_for_ms
) {
1866 cmd
[0] = MODE_SENSE_10
;
1873 cmd
[0] = MODE_SENSE
;
1878 memset(buffer
, 0, len
);
1880 result
= scsi_execute_req(sdev
, cmd
, DMA_FROM_DEVICE
, buffer
, len
,
1881 sshdr
, timeout
, retries
);
1883 /* This code looks awful: what it's doing is making sure an
1884 * ILLEGAL REQUEST sense return identifies the actual command
1885 * byte as the problem. MODE_SENSE commands can return
1886 * ILLEGAL REQUEST if the code page isn't supported */
1888 if (use_10_for_ms
&& !scsi_status_is_good(result
) &&
1889 (driver_byte(result
) & DRIVER_SENSE
)) {
1890 if (scsi_sense_valid(sshdr
)) {
1891 if ((sshdr
->sense_key
== ILLEGAL_REQUEST
) &&
1892 (sshdr
->asc
== 0x20) && (sshdr
->ascq
== 0)) {
1894 * Invalid command operation code
1896 sdev
->use_10_for_ms
= 0;
1902 if(scsi_status_is_good(result
)) {
1903 if (unlikely(buffer
[0] == 0x86 && buffer
[1] == 0x0b &&
1904 (modepage
== 6 || modepage
== 8))) {
1905 /* Initio breakage? */
1908 data
->medium_type
= 0;
1909 data
->device_specific
= 0;
1911 data
->block_descriptor_length
= 0;
1912 } else if(use_10_for_ms
) {
1913 data
->length
= buffer
[0]*256 + buffer
[1] + 2;
1914 data
->medium_type
= buffer
[2];
1915 data
->device_specific
= buffer
[3];
1916 data
->longlba
= buffer
[4] & 0x01;
1917 data
->block_descriptor_length
= buffer
[6]*256
1920 data
->length
= buffer
[0] + 1;
1921 data
->medium_type
= buffer
[1];
1922 data
->device_specific
= buffer
[2];
1923 data
->block_descriptor_length
= buffer
[3];
1925 data
->header_length
= header_length
;
1930 EXPORT_SYMBOL(scsi_mode_sense
);
1933 scsi_test_unit_ready(struct scsi_device
*sdev
, int timeout
, int retries
)
1936 TEST_UNIT_READY
, 0, 0, 0, 0, 0,
1938 struct scsi_sense_hdr sshdr
;
1941 result
= scsi_execute_req(sdev
, cmd
, DMA_NONE
, NULL
, 0, &sshdr
,
1944 if ((driver_byte(result
) & DRIVER_SENSE
) && sdev
->removable
) {
1946 if ((scsi_sense_valid(&sshdr
)) &&
1947 ((sshdr
.sense_key
== UNIT_ATTENTION
) ||
1948 (sshdr
.sense_key
== NOT_READY
))) {
1955 EXPORT_SYMBOL(scsi_test_unit_ready
);
1958 * scsi_device_set_state - Take the given device through the device
1960 * @sdev: scsi device to change the state of.
1961 * @state: state to change to.
1963 * Returns zero if unsuccessful or an error if the requested
1964 * transition is illegal.
1967 scsi_device_set_state(struct scsi_device
*sdev
, enum scsi_device_state state
)
1969 enum scsi_device_state oldstate
= sdev
->sdev_state
;
1971 if (state
== oldstate
)
1976 /* There are no legal states that come back to
1977 * created. This is the manually initialised start
2051 sdev
->sdev_state
= state
;
2055 SCSI_LOG_ERROR_RECOVERY(1,
2056 sdev_printk(KERN_ERR
, sdev
,
2057 "Illegal state transition %s->%s\n",
2058 scsi_device_state_name(oldstate
),
2059 scsi_device_state_name(state
))
2063 EXPORT_SYMBOL(scsi_device_set_state
);
2066 * scsi_device_quiesce - Block user issued commands.
2067 * @sdev: scsi device to quiesce.
2069 * This works by trying to transition to the SDEV_QUIESCE state
2070 * (which must be a legal transition). When the device is in this
2071 * state, only special requests will be accepted, all others will
2072 * be deferred. Since special requests may also be requeued requests,
2073 * a successful return doesn't guarantee the device will be
2074 * totally quiescent.
2076 * Must be called with user context, may sleep.
2078 * Returns zero if unsuccessful or an error if not.
2081 scsi_device_quiesce(struct scsi_device
*sdev
)
2083 int err
= scsi_device_set_state(sdev
, SDEV_QUIESCE
);
2087 scsi_run_queue(sdev
->request_queue
);
2088 while (sdev
->device_busy
) {
2089 msleep_interruptible(200);
2090 scsi_run_queue(sdev
->request_queue
);
2094 EXPORT_SYMBOL(scsi_device_quiesce
);
2097 * scsi_device_resume - Restart user issued commands to a quiesced device.
2098 * @sdev: scsi device to resume.
2100 * Moves the device from quiesced back to running and restarts the
2103 * Must be called with user context, may sleep.
2106 scsi_device_resume(struct scsi_device
*sdev
)
2108 if(scsi_device_set_state(sdev
, SDEV_RUNNING
))
2110 scsi_run_queue(sdev
->request_queue
);
2112 EXPORT_SYMBOL(scsi_device_resume
);
2115 device_quiesce_fn(struct scsi_device
*sdev
, void *data
)
2117 scsi_device_quiesce(sdev
);
2121 scsi_target_quiesce(struct scsi_target
*starget
)
2123 starget_for_each_device(starget
, NULL
, device_quiesce_fn
);
2125 EXPORT_SYMBOL(scsi_target_quiesce
);
2128 device_resume_fn(struct scsi_device
*sdev
, void *data
)
2130 scsi_device_resume(sdev
);
2134 scsi_target_resume(struct scsi_target
*starget
)
2136 starget_for_each_device(starget
, NULL
, device_resume_fn
);
2138 EXPORT_SYMBOL(scsi_target_resume
);
2141 * scsi_internal_device_block - internal function to put a device
2142 * temporarily into the SDEV_BLOCK state
2143 * @sdev: device to block
2145 * Block request made by scsi lld's to temporarily stop all
2146 * scsi commands on the specified device. Called from interrupt
2147 * or normal process context.
2149 * Returns zero if successful or error if not
2152 * This routine transitions the device to the SDEV_BLOCK state
2153 * (which must be a legal transition). When the device is in this
2154 * state, all commands are deferred until the scsi lld reenables
2155 * the device with scsi_device_unblock or device_block_tmo fires.
2156 * This routine assumes the host_lock is held on entry.
2159 scsi_internal_device_block(struct scsi_device
*sdev
)
2161 request_queue_t
*q
= sdev
->request_queue
;
2162 unsigned long flags
;
2165 err
= scsi_device_set_state(sdev
, SDEV_BLOCK
);
2170 * The device has transitioned to SDEV_BLOCK. Stop the
2171 * block layer from calling the midlayer with this device's
2174 spin_lock_irqsave(q
->queue_lock
, flags
);
2176 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2180 EXPORT_SYMBOL_GPL(scsi_internal_device_block
);
2183 * scsi_internal_device_unblock - resume a device after a block request
2184 * @sdev: device to resume
2186 * Called by scsi lld's or the midlayer to restart the device queue
2187 * for the previously suspended scsi device. Called from interrupt or
2188 * normal process context.
2190 * Returns zero if successful or error if not.
2193 * This routine transitions the device to the SDEV_RUNNING state
2194 * (which must be a legal transition) allowing the midlayer to
2195 * goose the queue for this device. This routine assumes the
2196 * host_lock is held upon entry.
2199 scsi_internal_device_unblock(struct scsi_device
*sdev
)
2201 request_queue_t
*q
= sdev
->request_queue
;
2203 unsigned long flags
;
2206 * Try to transition the scsi device to SDEV_RUNNING
2207 * and goose the device queue if successful.
2209 err
= scsi_device_set_state(sdev
, SDEV_RUNNING
);
2213 spin_lock_irqsave(q
->queue_lock
, flags
);
2215 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2219 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock
);
2222 device_block(struct scsi_device
*sdev
, void *data
)
2224 scsi_internal_device_block(sdev
);
2228 target_block(struct device
*dev
, void *data
)
2230 if (scsi_is_target_device(dev
))
2231 starget_for_each_device(to_scsi_target(dev
), NULL
,
2237 scsi_target_block(struct device
*dev
)
2239 if (scsi_is_target_device(dev
))
2240 starget_for_each_device(to_scsi_target(dev
), NULL
,
2243 device_for_each_child(dev
, NULL
, target_block
);
2245 EXPORT_SYMBOL_GPL(scsi_target_block
);
2248 device_unblock(struct scsi_device
*sdev
, void *data
)
2250 scsi_internal_device_unblock(sdev
);
2254 target_unblock(struct device
*dev
, void *data
)
2256 if (scsi_is_target_device(dev
))
2257 starget_for_each_device(to_scsi_target(dev
), NULL
,
2263 scsi_target_unblock(struct device
*dev
)
2265 if (scsi_is_target_device(dev
))
2266 starget_for_each_device(to_scsi_target(dev
), NULL
,
2269 device_for_each_child(dev
, NULL
, target_unblock
);
2271 EXPORT_SYMBOL_GPL(scsi_target_unblock
);
2274 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2275 * @sg: scatter-gather list
2276 * @sg_count: number of segments in sg
2277 * @offset: offset in bytes into sg, on return offset into the mapped area
2278 * @len: bytes to map, on return number of bytes mapped
2280 * Returns virtual address of the start of the mapped page
2282 void *scsi_kmap_atomic_sg(struct scatterlist
*sg
, int sg_count
,
2283 size_t *offset
, size_t *len
)
2286 size_t sg_len
= 0, len_complete
= 0;
2289 for (i
= 0; i
< sg_count
; i
++) {
2290 len_complete
= sg_len
; /* Complete sg-entries */
2291 sg_len
+= sg
[i
].length
;
2292 if (sg_len
> *offset
)
2296 if (unlikely(i
== sg_count
)) {
2297 printk(KERN_ERR
"%s: Bytes in sg: %zu, requested offset %zu, "
2299 __FUNCTION__
, sg_len
, *offset
, sg_count
);
2304 /* Offset starting from the beginning of first page in this sg-entry */
2305 *offset
= *offset
- len_complete
+ sg
[i
].offset
;
2307 /* Assumption: contiguous pages can be accessed as "page + i" */
2308 page
= nth_page(sg
[i
].page
, (*offset
>> PAGE_SHIFT
));
2309 *offset
&= ~PAGE_MASK
;
2311 /* Bytes in this sg-entry from *offset to the end of the page */
2312 sg_len
= PAGE_SIZE
- *offset
;
2316 return kmap_atomic(page
, KM_BIO_SRC_IRQ
);
2318 EXPORT_SYMBOL(scsi_kmap_atomic_sg
);
2321 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously
2322 * mapped with scsi_kmap_atomic_sg
2323 * @virt: virtual address to be unmapped
2325 void scsi_kunmap_atomic_sg(void *virt
)
2327 kunmap_atomic(virt
, KM_BIO_SRC_IRQ
);
2329 EXPORT_SYMBOL(scsi_kunmap_atomic_sg
);