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_dbg.h>
23 #include <scsi/scsi_device.h>
24 #include <scsi/scsi_driver.h>
25 #include <scsi/scsi_eh.h>
26 #include <scsi/scsi_host.h>
27 #include <scsi/scsi_request.h>
29 #include "scsi_priv.h"
30 #include "scsi_logging.h"
33 #define SG_MEMPOOL_NR (sizeof(scsi_sg_pools)/sizeof(struct scsi_host_sg_pool))
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
;
86 req
->special
= (req
->flags
& REQ_SPECIAL
) ? cmd
->sc_request
: NULL
;
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 * Function: scsi_do_req
167 * Purpose: Queue a SCSI request
169 * Arguments: sreq - command descriptor.
170 * cmnd - actual SCSI command to be performed.
171 * buffer - data buffer.
172 * bufflen - size of data buffer.
173 * done - completion function to be run.
174 * timeout - how long to let it run before timeout.
175 * retries - number of retries we allow.
177 * Lock status: No locks held upon entry.
181 * Notes: This function is only used for queueing requests for things
182 * like ioctls and character device requests - this is because
183 * we essentially just inject a request into the queue for the
186 * In order to support the scsi_device_quiesce function, we
187 * now inject requests on the *head* of the device queue
188 * rather than the tail.
190 void scsi_do_req(struct scsi_request
*sreq
, const void *cmnd
,
191 void *buffer
, unsigned bufflen
,
192 void (*done
)(struct scsi_cmnd
*),
193 int timeout
, int retries
)
196 * If the upper level driver is reusing these things, then
197 * we should release the low-level block now. Another one will
198 * be allocated later when this request is getting queued.
200 __scsi_release_request(sreq
);
203 * Our own function scsi_done (which marks the host as not busy,
204 * disables the timeout counter, etc) will be called by us or by the
205 * scsi_hosts[host].queuecommand() function needs to also call
206 * the completion function for the high level driver.
208 memcpy(sreq
->sr_cmnd
, cmnd
, sizeof(sreq
->sr_cmnd
));
209 sreq
->sr_bufflen
= bufflen
;
210 sreq
->sr_buffer
= buffer
;
211 sreq
->sr_allowed
= retries
;
212 sreq
->sr_done
= done
;
213 sreq
->sr_timeout_per_command
= timeout
;
215 if (sreq
->sr_cmd_len
== 0)
216 sreq
->sr_cmd_len
= COMMAND_SIZE(sreq
->sr_cmnd
[0]);
219 * head injection *required* here otherwise quiesce won't work
221 * Because users of this function are apt to reuse requests with no
222 * modification, we have to sanitise the request flags here
224 sreq
->sr_request
->flags
&= ~REQ_DONTPREP
;
225 blk_insert_request(sreq
->sr_device
->request_queue
, sreq
->sr_request
,
228 EXPORT_SYMBOL(scsi_do_req
);
231 * scsi_execute - insert request and wait for the result
234 * @data_direction: data direction
235 * @buffer: data buffer
236 * @bufflen: len of buffer
237 * @sense: optional sense buffer
238 * @timeout: request timeout in seconds
239 * @retries: number of times to retry request
240 * @flags: or into request flags;
242 * returns the req->errors value which is the the scsi_cmnd result
245 int scsi_execute(struct scsi_device
*sdev
, const unsigned char *cmd
,
246 int data_direction
, void *buffer
, unsigned bufflen
,
247 unsigned char *sense
, int timeout
, int retries
, int flags
)
250 int write
= (data_direction
== DMA_TO_DEVICE
);
251 int ret
= DRIVER_ERROR
<< 24;
253 req
= blk_get_request(sdev
->request_queue
, write
, __GFP_WAIT
);
255 if (bufflen
&& blk_rq_map_kern(sdev
->request_queue
, req
,
256 buffer
, bufflen
, __GFP_WAIT
))
259 req
->cmd_len
= COMMAND_SIZE(cmd
[0]);
260 memcpy(req
->cmd
, cmd
, req
->cmd_len
);
263 req
->retries
= retries
;
264 req
->timeout
= timeout
;
265 req
->flags
|= flags
| REQ_BLOCK_PC
| REQ_SPECIAL
| REQ_QUIET
;
268 * head injection *required* here otherwise quiesce won't work
270 blk_execute_rq(req
->q
, NULL
, req
, 1);
274 blk_put_request(req
);
278 EXPORT_SYMBOL(scsi_execute
);
281 int scsi_execute_req(struct scsi_device
*sdev
, const unsigned char *cmd
,
282 int data_direction
, void *buffer
, unsigned bufflen
,
283 struct scsi_sense_hdr
*sshdr
, int timeout
, int retries
)
289 sense
= kmalloc(SCSI_SENSE_BUFFERSIZE
, GFP_NOIO
);
291 return DRIVER_ERROR
<< 24;
292 memset(sense
, 0, SCSI_SENSE_BUFFERSIZE
);
294 result
= scsi_execute(sdev
, cmd
, data_direction
, buffer
, bufflen
,
295 sense
, timeout
, retries
, 0);
297 scsi_normalize_sense(sense
, SCSI_SENSE_BUFFERSIZE
, sshdr
);
302 EXPORT_SYMBOL(scsi_execute_req
);
304 struct scsi_io_context
{
306 void (*done
)(void *data
, char *sense
, int result
, int resid
);
307 char sense
[SCSI_SENSE_BUFFERSIZE
];
310 static kmem_cache_t
*scsi_io_context_cache
;
312 static void scsi_end_async(struct request
*req
, int uptodate
)
314 struct scsi_io_context
*sioc
= req
->end_io_data
;
317 sioc
->done(sioc
->data
, sioc
->sense
, req
->errors
, req
->data_len
);
319 kmem_cache_free(scsi_io_context_cache
, sioc
);
320 __blk_put_request(req
->q
, req
);
323 static int scsi_merge_bio(struct request
*rq
, struct bio
*bio
)
325 struct request_queue
*q
= rq
->q
;
327 bio
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
328 if (rq_data_dir(rq
) == WRITE
)
329 bio
->bi_rw
|= (1 << BIO_RW
);
330 blk_queue_bounce(q
, &bio
);
333 blk_rq_bio_prep(q
, rq
, bio
);
334 else if (!q
->back_merge_fn(q
, rq
, bio
))
337 rq
->biotail
->bi_next
= bio
;
339 rq
->hard_nr_sectors
+= bio_sectors(bio
);
340 rq
->nr_sectors
= rq
->hard_nr_sectors
;
346 static int scsi_bi_endio(struct bio
*bio
, unsigned int bytes_done
, int error
)
356 * scsi_req_map_sg - map a scatterlist into a request
357 * @rq: request to fill
359 * @nsegs: number of elements
360 * @bufflen: len of buffer
361 * @gfp: memory allocation flags
363 * scsi_req_map_sg maps a scatterlist into a request so that the
364 * request can be sent to the block layer. We do not trust the scatterlist
365 * sent to use, as some ULDs use that struct to only organize the pages.
367 static int scsi_req_map_sg(struct request
*rq
, struct scatterlist
*sgl
,
368 int nsegs
, unsigned bufflen
, gfp_t gfp
)
370 struct request_queue
*q
= rq
->q
;
371 int nr_pages
= (bufflen
+ sgl
[0].offset
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
372 unsigned int data_len
= 0, len
, bytes
, off
;
374 struct bio
*bio
= NULL
;
375 int i
, err
, nr_vecs
= 0;
377 for (i
= 0; i
< nsegs
; i
++) {
384 bytes
= min_t(unsigned int, len
, PAGE_SIZE
- off
);
387 nr_vecs
= min_t(int, BIO_MAX_PAGES
, nr_pages
);
390 bio
= bio_alloc(gfp
, nr_vecs
);
395 bio
->bi_end_io
= scsi_bi_endio
;
398 if (bio_add_pc_page(q
, bio
, page
, bytes
, off
) !=
405 if (bio
->bi_vcnt
>= nr_vecs
) {
406 err
= scsi_merge_bio(rq
, bio
);
408 bio_endio(bio
, bio
->bi_size
, 0);
420 rq
->buffer
= rq
->data
= NULL
;
421 rq
->data_len
= data_len
;
425 while ((bio
= rq
->bio
) != NULL
) {
426 rq
->bio
= bio
->bi_next
;
428 * call endio instead of bio_put incase it was bounced
430 bio_endio(bio
, bio
->bi_size
, 0);
437 * scsi_execute_async - insert request
440 * @cmd_len: length of scsi cdb
441 * @data_direction: data direction
442 * @buffer: data buffer (this can be a kernel buffer or scatterlist)
443 * @bufflen: len of buffer
444 * @use_sg: if buffer is a scatterlist this is the number of elements
445 * @timeout: request timeout in seconds
446 * @retries: number of times to retry request
447 * @flags: or into request flags
449 int scsi_execute_async(struct scsi_device
*sdev
, const unsigned char *cmd
,
450 int cmd_len
, int data_direction
, void *buffer
, unsigned bufflen
,
451 int use_sg
, int timeout
, int retries
, void *privdata
,
452 void (*done
)(void *, char *, int, int), gfp_t gfp
)
455 struct scsi_io_context
*sioc
;
457 int write
= (data_direction
== DMA_TO_DEVICE
);
459 sioc
= kmem_cache_alloc(scsi_io_context_cache
, gfp
);
461 return DRIVER_ERROR
<< 24;
462 memset(sioc
, 0, sizeof(*sioc
));
464 req
= blk_get_request(sdev
->request_queue
, write
, gfp
);
467 req
->flags
|= REQ_BLOCK_PC
| REQ_QUIET
;
470 err
= scsi_req_map_sg(req
, buffer
, use_sg
, bufflen
, gfp
);
472 err
= blk_rq_map_kern(req
->q
, req
, buffer
, bufflen
, gfp
);
477 req
->cmd_len
= cmd_len
;
478 memset(req
->cmd
, 0, BLK_MAX_CDB
); /* ATAPI hates garbage after CDB */
479 memcpy(req
->cmd
, cmd
, req
->cmd_len
);
480 req
->sense
= sioc
->sense
;
482 req
->timeout
= timeout
;
483 req
->retries
= retries
;
484 req
->end_io_data
= sioc
;
486 sioc
->data
= privdata
;
489 blk_execute_rq_nowait(req
->q
, NULL
, req
, 1, scsi_end_async
);
493 blk_put_request(req
);
496 return DRIVER_ERROR
<< 24;
498 EXPORT_SYMBOL_GPL(scsi_execute_async
);
501 * Function: scsi_init_cmd_errh()
503 * Purpose: Initialize cmd fields related to error handling.
505 * Arguments: cmd - command that is ready to be queued.
509 * Notes: This function has the job of initializing a number of
510 * fields related to error handling. Typically this will
511 * be called once for each command, as required.
513 static int scsi_init_cmd_errh(struct scsi_cmnd
*cmd
)
515 cmd
->serial_number
= 0;
517 memset(cmd
->sense_buffer
, 0, sizeof cmd
->sense_buffer
);
519 if (cmd
->cmd_len
== 0)
520 cmd
->cmd_len
= COMMAND_SIZE(cmd
->cmnd
[0]);
523 * We need saved copies of a number of fields - this is because
524 * error handling may need to overwrite these with different values
525 * to run different commands, and once error handling is complete,
526 * we will need to restore these values prior to running the actual
529 cmd
->old_use_sg
= cmd
->use_sg
;
530 cmd
->old_cmd_len
= cmd
->cmd_len
;
531 cmd
->sc_old_data_direction
= cmd
->sc_data_direction
;
532 cmd
->old_underflow
= cmd
->underflow
;
533 memcpy(cmd
->data_cmnd
, cmd
->cmnd
, sizeof(cmd
->cmnd
));
534 cmd
->buffer
= cmd
->request_buffer
;
535 cmd
->bufflen
= cmd
->request_bufflen
;
541 * Function: scsi_setup_cmd_retry()
543 * Purpose: Restore the command state for a retry
545 * Arguments: cmd - command to be restored
549 * Notes: Immediately prior to retrying a command, we need
550 * to restore certain fields that we saved above.
552 void scsi_setup_cmd_retry(struct scsi_cmnd
*cmd
)
554 memcpy(cmd
->cmnd
, cmd
->data_cmnd
, sizeof(cmd
->data_cmnd
));
555 cmd
->request_buffer
= cmd
->buffer
;
556 cmd
->request_bufflen
= cmd
->bufflen
;
557 cmd
->use_sg
= cmd
->old_use_sg
;
558 cmd
->cmd_len
= cmd
->old_cmd_len
;
559 cmd
->sc_data_direction
= cmd
->sc_old_data_direction
;
560 cmd
->underflow
= cmd
->old_underflow
;
563 void scsi_device_unbusy(struct scsi_device
*sdev
)
565 struct Scsi_Host
*shost
= sdev
->host
;
568 spin_lock_irqsave(shost
->host_lock
, flags
);
570 if (unlikely(scsi_host_in_recovery(shost
) &&
572 scsi_eh_wakeup(shost
);
573 spin_unlock(shost
->host_lock
);
574 spin_lock(sdev
->request_queue
->queue_lock
);
576 spin_unlock_irqrestore(sdev
->request_queue
->queue_lock
, flags
);
580 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
581 * and call blk_run_queue for all the scsi_devices on the target -
582 * including current_sdev first.
584 * Called with *no* scsi locks held.
586 static void scsi_single_lun_run(struct scsi_device
*current_sdev
)
588 struct Scsi_Host
*shost
= current_sdev
->host
;
589 struct scsi_device
*sdev
, *tmp
;
590 struct scsi_target
*starget
= scsi_target(current_sdev
);
593 spin_lock_irqsave(shost
->host_lock
, flags
);
594 starget
->starget_sdev_user
= NULL
;
595 spin_unlock_irqrestore(shost
->host_lock
, flags
);
598 * Call blk_run_queue for all LUNs on the target, starting with
599 * current_sdev. We race with others (to set starget_sdev_user),
600 * but in most cases, we will be first. Ideally, each LU on the
601 * target would get some limited time or requests on the target.
603 blk_run_queue(current_sdev
->request_queue
);
605 spin_lock_irqsave(shost
->host_lock
, flags
);
606 if (starget
->starget_sdev_user
)
608 list_for_each_entry_safe(sdev
, tmp
, &starget
->devices
,
609 same_target_siblings
) {
610 if (sdev
== current_sdev
)
612 if (scsi_device_get(sdev
))
615 spin_unlock_irqrestore(shost
->host_lock
, flags
);
616 blk_run_queue(sdev
->request_queue
);
617 spin_lock_irqsave(shost
->host_lock
, flags
);
619 scsi_device_put(sdev
);
622 spin_unlock_irqrestore(shost
->host_lock
, flags
);
626 * Function: scsi_run_queue()
628 * Purpose: Select a proper request queue to serve next
630 * Arguments: q - last request's queue
634 * Notes: The previous command was completely finished, start
635 * a new one if possible.
637 static void scsi_run_queue(struct request_queue
*q
)
639 struct scsi_device
*sdev
= q
->queuedata
;
640 struct Scsi_Host
*shost
= sdev
->host
;
643 if (sdev
->single_lun
)
644 scsi_single_lun_run(sdev
);
646 spin_lock_irqsave(shost
->host_lock
, flags
);
647 while (!list_empty(&shost
->starved_list
) &&
648 !shost
->host_blocked
&& !shost
->host_self_blocked
&&
649 !((shost
->can_queue
> 0) &&
650 (shost
->host_busy
>= shost
->can_queue
))) {
652 * As long as shost is accepting commands and we have
653 * starved queues, call blk_run_queue. scsi_request_fn
654 * drops the queue_lock and can add us back to the
657 * host_lock protects the starved_list and starved_entry.
658 * scsi_request_fn must get the host_lock before checking
659 * or modifying starved_list or starved_entry.
661 sdev
= list_entry(shost
->starved_list
.next
,
662 struct scsi_device
, starved_entry
);
663 list_del_init(&sdev
->starved_entry
);
664 spin_unlock_irqrestore(shost
->host_lock
, flags
);
666 blk_run_queue(sdev
->request_queue
);
668 spin_lock_irqsave(shost
->host_lock
, flags
);
669 if (unlikely(!list_empty(&sdev
->starved_entry
)))
671 * sdev lost a race, and was put back on the
672 * starved list. This is unlikely but without this
673 * in theory we could loop forever.
677 spin_unlock_irqrestore(shost
->host_lock
, flags
);
683 * Function: scsi_requeue_command()
685 * Purpose: Handle post-processing of completed commands.
687 * Arguments: q - queue to operate on
688 * cmd - command that may need to be requeued.
692 * Notes: After command completion, there may be blocks left
693 * over which weren't finished by the previous command
694 * this can be for a number of reasons - the main one is
695 * I/O errors in the middle of the request, in which case
696 * we need to request the blocks that come after the bad
698 * Notes: Upon return, cmd is a stale pointer.
700 static void scsi_requeue_command(struct request_queue
*q
, struct scsi_cmnd
*cmd
)
702 struct request
*req
= cmd
->request
;
705 scsi_unprep_request(req
);
706 spin_lock_irqsave(q
->queue_lock
, flags
);
707 blk_requeue_request(q
, req
);
708 spin_unlock_irqrestore(q
->queue_lock
, flags
);
713 void scsi_next_command(struct scsi_cmnd
*cmd
)
715 struct scsi_device
*sdev
= cmd
->device
;
716 struct request_queue
*q
= sdev
->request_queue
;
718 /* need to hold a reference on the device before we let go of the cmd */
719 get_device(&sdev
->sdev_gendev
);
721 scsi_put_command(cmd
);
724 /* ok to remove device now */
725 put_device(&sdev
->sdev_gendev
);
728 void scsi_run_host_queues(struct Scsi_Host
*shost
)
730 struct scsi_device
*sdev
;
732 shost_for_each_device(sdev
, shost
)
733 scsi_run_queue(sdev
->request_queue
);
737 * Function: scsi_end_request()
739 * Purpose: Post-processing of completed commands (usually invoked at end
740 * of upper level post-processing and scsi_io_completion).
742 * Arguments: cmd - command that is complete.
743 * uptodate - 1 if I/O indicates success, <= 0 for I/O error.
744 * bytes - number of bytes of completed I/O
745 * requeue - indicates whether we should requeue leftovers.
747 * Lock status: Assumed that lock is not held upon entry.
749 * Returns: cmd if requeue required, NULL otherwise.
751 * Notes: This is called for block device requests in order to
752 * mark some number of sectors as complete.
754 * We are guaranteeing that the request queue will be goosed
755 * at some point during this call.
756 * Notes: If cmd was requeued, upon return it will be a stale pointer.
758 static struct scsi_cmnd
*scsi_end_request(struct scsi_cmnd
*cmd
, int uptodate
,
759 int bytes
, int requeue
)
761 request_queue_t
*q
= cmd
->device
->request_queue
;
762 struct request
*req
= cmd
->request
;
766 * If there are blocks left over at the end, set up the command
767 * to queue the remainder of them.
769 if (end_that_request_chunk(req
, uptodate
, bytes
)) {
770 int leftover
= (req
->hard_nr_sectors
<< 9);
772 if (blk_pc_request(req
))
773 leftover
= req
->data_len
;
775 /* kill remainder if no retrys */
776 if (!uptodate
&& blk_noretry_request(req
))
777 end_that_request_chunk(req
, 0, leftover
);
781 * Bleah. Leftovers again. Stick the
782 * leftovers in the front of the
783 * queue, and goose the queue again.
785 scsi_requeue_command(q
, cmd
);
792 add_disk_randomness(req
->rq_disk
);
794 spin_lock_irqsave(q
->queue_lock
, flags
);
795 if (blk_rq_tagged(req
))
796 blk_queue_end_tag(q
, req
);
797 end_that_request_last(req
, uptodate
);
798 spin_unlock_irqrestore(q
->queue_lock
, flags
);
801 * This will goose the queue request function at the end, so we don't
802 * need to worry about launching another command.
804 scsi_next_command(cmd
);
808 static struct scatterlist
*scsi_alloc_sgtable(struct scsi_cmnd
*cmd
, gfp_t gfp_mask
)
810 struct scsi_host_sg_pool
*sgp
;
811 struct scatterlist
*sgl
;
813 BUG_ON(!cmd
->use_sg
);
815 switch (cmd
->use_sg
) {
825 #if (SCSI_MAX_PHYS_SEGMENTS > 32)
829 #if (SCSI_MAX_PHYS_SEGMENTS > 64)
833 #if (SCSI_MAX_PHYS_SEGMENTS > 128)
844 sgp
= scsi_sg_pools
+ cmd
->sglist_len
;
845 sgl
= mempool_alloc(sgp
->pool
, gfp_mask
);
849 static void scsi_free_sgtable(struct scatterlist
*sgl
, int index
)
851 struct scsi_host_sg_pool
*sgp
;
853 BUG_ON(index
>= SG_MEMPOOL_NR
);
855 sgp
= scsi_sg_pools
+ index
;
856 mempool_free(sgl
, sgp
->pool
);
860 * Function: scsi_release_buffers()
862 * Purpose: Completion processing for block device I/O requests.
864 * Arguments: cmd - command that we are bailing.
866 * Lock status: Assumed that no lock is held upon entry.
870 * Notes: In the event that an upper level driver rejects a
871 * command, we must release resources allocated during
872 * the __init_io() function. Primarily this would involve
873 * the scatter-gather table, and potentially any bounce
876 static void scsi_release_buffers(struct scsi_cmnd
*cmd
)
878 struct request
*req
= cmd
->request
;
881 * Free up any indirection buffers we allocated for DMA purposes.
884 scsi_free_sgtable(cmd
->request_buffer
, cmd
->sglist_len
);
885 else if (cmd
->request_buffer
!= req
->buffer
)
886 kfree(cmd
->request_buffer
);
889 * Zero these out. They now point to freed memory, and it is
890 * dangerous to hang onto the pointers.
894 cmd
->request_buffer
= NULL
;
895 cmd
->request_bufflen
= 0;
899 * Function: scsi_io_completion()
901 * Purpose: Completion processing for block device I/O requests.
903 * Arguments: cmd - command that is finished.
905 * Lock status: Assumed that no lock is held upon entry.
909 * Notes: This function is matched in terms of capabilities to
910 * the function that created the scatter-gather list.
911 * In other words, if there are no bounce buffers
912 * (the normal case for most drivers), we don't need
913 * the logic to deal with cleaning up afterwards.
915 * We must do one of several things here:
917 * a) Call scsi_end_request. This will finish off the
918 * specified number of sectors. If we are done, the
919 * command block will be released, and the queue
920 * function will be goosed. If we are not done, then
921 * scsi_end_request will directly goose the queue.
923 * b) We can just use scsi_requeue_command() here. This would
924 * be used if we just wanted to retry, for example.
926 void scsi_io_completion(struct scsi_cmnd
*cmd
, unsigned int good_bytes
,
927 unsigned int block_bytes
)
929 int result
= cmd
->result
;
930 int this_count
= cmd
->bufflen
;
931 request_queue_t
*q
= cmd
->device
->request_queue
;
932 struct request
*req
= cmd
->request
;
933 int clear_errors
= 1;
934 struct scsi_sense_hdr sshdr
;
936 int sense_deferred
= 0;
939 * Free up any indirection buffers we allocated for DMA purposes.
940 * For the case of a READ, we need to copy the data out of the
941 * bounce buffer and into the real buffer.
944 scsi_free_sgtable(cmd
->buffer
, cmd
->sglist_len
);
945 else if (cmd
->buffer
!= req
->buffer
) {
946 if (rq_data_dir(req
) == READ
) {
948 char *to
= bio_kmap_irq(req
->bio
, &flags
);
949 memcpy(to
, cmd
->buffer
, cmd
->bufflen
);
950 bio_kunmap_irq(to
, &flags
);
956 sense_valid
= scsi_command_normalize_sense(cmd
, &sshdr
);
958 sense_deferred
= scsi_sense_is_deferred(&sshdr
);
960 if (blk_pc_request(req
)) { /* SG_IO ioctl from block level */
961 req
->errors
= result
;
964 if (sense_valid
&& req
->sense
) {
966 * SG_IO wants current and deferred errors
968 int len
= 8 + cmd
->sense_buffer
[7];
970 if (len
> SCSI_SENSE_BUFFERSIZE
)
971 len
= SCSI_SENSE_BUFFERSIZE
;
972 memcpy(req
->sense
, cmd
->sense_buffer
, len
);
973 req
->sense_len
= len
;
976 req
->data_len
= cmd
->resid
;
980 * Zero these out. They now point to freed memory, and it is
981 * dangerous to hang onto the pointers.
985 cmd
->request_buffer
= NULL
;
986 cmd
->request_bufflen
= 0;
989 * Next deal with any sectors which we were able to correctly
992 if (good_bytes
>= 0) {
993 SCSI_LOG_HLCOMPLETE(1, printk("%ld sectors total, %d bytes done.\n",
994 req
->nr_sectors
, good_bytes
));
995 SCSI_LOG_HLCOMPLETE(1, printk("use_sg is %d\n", cmd
->use_sg
));
1000 * If multiple sectors are requested in one buffer, then
1001 * they will have been finished off by the first command.
1002 * If not, then we have a multi-buffer command.
1004 * If block_bytes != 0, it means we had a medium error
1005 * of some sort, and that we want to mark some number of
1006 * sectors as not uptodate. Thus we want to inhibit
1007 * requeueing right here - we will requeue down below
1008 * when we handle the bad sectors.
1012 * If the command completed without error, then either
1013 * finish off the rest of the command, or start a new one.
1015 if (scsi_end_request(cmd
, 1, good_bytes
, result
== 0) == NULL
)
1019 * Now, if we were good little boys and girls, Santa left us a request
1020 * sense buffer. We can extract information from this, so we
1021 * can choose a block to remap, etc.
1023 if (sense_valid
&& !sense_deferred
) {
1024 switch (sshdr
.sense_key
) {
1025 case UNIT_ATTENTION
:
1026 if (cmd
->device
->removable
) {
1027 /* detected disc change. set a bit
1028 * and quietly refuse further access.
1030 cmd
->device
->changed
= 1;
1031 scsi_end_request(cmd
, 0,
1036 * Must have been a power glitch, or a
1037 * bus reset. Could not have been a
1038 * media change, so we just retry the
1039 * request and see what happens.
1041 scsi_requeue_command(q
, cmd
);
1045 case ILLEGAL_REQUEST
:
1047 * If we had an ILLEGAL REQUEST returned, then we may
1048 * have performed an unsupported command. The only
1049 * thing this should be would be a ten byte read where
1050 * only a six byte read was supported. Also, on a
1051 * system where READ CAPACITY failed, we may have read
1052 * past the end of the disk.
1054 if ((cmd
->device
->use_10_for_rw
&&
1055 sshdr
.asc
== 0x20 && sshdr
.ascq
== 0x00) &&
1056 (cmd
->cmnd
[0] == READ_10
||
1057 cmd
->cmnd
[0] == WRITE_10
)) {
1058 cmd
->device
->use_10_for_rw
= 0;
1060 * This will cause a retry with a 6-byte
1063 scsi_requeue_command(q
, cmd
);
1066 scsi_end_request(cmd
, 0, this_count
, 1);
1072 * If the device is in the process of becoming ready,
1075 if (sshdr
.asc
== 0x04 && sshdr
.ascq
== 0x01) {
1076 scsi_requeue_command(q
, cmd
);
1079 if (!(req
->flags
& REQ_QUIET
))
1080 scmd_printk(KERN_INFO
, cmd
,
1081 "Device not ready.\n");
1082 scsi_end_request(cmd
, 0, this_count
, 1);
1084 case VOLUME_OVERFLOW
:
1085 if (!(req
->flags
& REQ_QUIET
)) {
1086 scmd_printk(KERN_INFO
, cmd
,
1087 "Volume overflow, CDB: ");
1088 __scsi_print_command(cmd
->data_cmnd
);
1089 scsi_print_sense("", cmd
);
1091 scsi_end_request(cmd
, 0, block_bytes
, 1);
1096 } /* driver byte != 0 */
1097 if (host_byte(result
) == DID_RESET
) {
1099 * Third party bus reset or reset for error
1100 * recovery reasons. Just retry the request
1101 * and see what happens.
1103 scsi_requeue_command(q
, cmd
);
1107 if (!(req
->flags
& REQ_QUIET
)) {
1108 scmd_printk(KERN_INFO
, cmd
,
1109 "SCSI error: return code = 0x%x\n", result
);
1111 if (driver_byte(result
) & DRIVER_SENSE
)
1112 scsi_print_sense("", cmd
);
1115 * Mark a single buffer as not uptodate. Queue the remainder.
1116 * We sometimes get this cruft in the event that a medium error
1117 * isn't properly reported.
1119 block_bytes
= req
->hard_cur_sectors
<< 9;
1121 block_bytes
= req
->data_len
;
1122 scsi_end_request(cmd
, 0, block_bytes
, 1);
1125 EXPORT_SYMBOL(scsi_io_completion
);
1128 * Function: scsi_init_io()
1130 * Purpose: SCSI I/O initialize function.
1132 * Arguments: cmd - Command descriptor we wish to initialize
1134 * Returns: 0 on success
1135 * BLKPREP_DEFER if the failure is retryable
1136 * BLKPREP_KILL if the failure is fatal
1138 static int scsi_init_io(struct scsi_cmnd
*cmd
)
1140 struct request
*req
= cmd
->request
;
1141 struct scatterlist
*sgpnt
;
1145 * if this is a rq->data based REQ_BLOCK_PC, setup for a non-sg xfer
1147 if ((req
->flags
& REQ_BLOCK_PC
) && !req
->bio
) {
1148 cmd
->request_bufflen
= req
->data_len
;
1149 cmd
->request_buffer
= req
->data
;
1150 req
->buffer
= req
->data
;
1156 * we used to not use scatter-gather for single segment request,
1157 * but now we do (it makes highmem I/O easier to support without
1160 cmd
->use_sg
= req
->nr_phys_segments
;
1163 * if sg table allocation fails, requeue request later.
1165 sgpnt
= scsi_alloc_sgtable(cmd
, GFP_ATOMIC
);
1166 if (unlikely(!sgpnt
)) {
1167 scsi_unprep_request(req
);
1168 return BLKPREP_DEFER
;
1171 cmd
->request_buffer
= (char *) sgpnt
;
1172 cmd
->request_bufflen
= req
->nr_sectors
<< 9;
1173 if (blk_pc_request(req
))
1174 cmd
->request_bufflen
= req
->data_len
;
1178 * Next, walk the list, and fill in the addresses and sizes of
1181 count
= blk_rq_map_sg(req
->q
, req
, cmd
->request_buffer
);
1184 * mapped well, send it off
1186 if (likely(count
<= cmd
->use_sg
)) {
1187 cmd
->use_sg
= count
;
1191 printk(KERN_ERR
"Incorrect number of segments after building list\n");
1192 printk(KERN_ERR
"counted %d, received %d\n", count
, cmd
->use_sg
);
1193 printk(KERN_ERR
"req nr_sec %lu, cur_nr_sec %u\n", req
->nr_sectors
,
1194 req
->current_nr_sectors
);
1196 /* release the command and kill it */
1197 scsi_release_buffers(cmd
);
1198 scsi_put_command(cmd
);
1199 return BLKPREP_KILL
;
1202 static int scsi_issue_flush_fn(request_queue_t
*q
, struct gendisk
*disk
,
1203 sector_t
*error_sector
)
1205 struct scsi_device
*sdev
= q
->queuedata
;
1206 struct scsi_driver
*drv
;
1208 if (sdev
->sdev_state
!= SDEV_RUNNING
)
1211 drv
= *(struct scsi_driver
**) disk
->private_data
;
1212 if (drv
->issue_flush
)
1213 return drv
->issue_flush(&sdev
->sdev_gendev
, error_sector
);
1218 static void scsi_blk_pc_done(struct scsi_cmnd
*cmd
)
1220 BUG_ON(!blk_pc_request(cmd
->request
));
1222 * This will complete the whole command with uptodate=1 so
1223 * as far as the block layer is concerned the command completed
1224 * successfully. Since this is a REQ_BLOCK_PC command the
1225 * caller should check the request's errors value
1227 scsi_io_completion(cmd
, cmd
->bufflen
, 0);
1230 static void scsi_setup_blk_pc_cmnd(struct scsi_cmnd
*cmd
)
1232 struct request
*req
= cmd
->request
;
1234 BUG_ON(sizeof(req
->cmd
) > sizeof(cmd
->cmnd
));
1235 memcpy(cmd
->cmnd
, req
->cmd
, sizeof(cmd
->cmnd
));
1236 cmd
->cmd_len
= req
->cmd_len
;
1238 cmd
->sc_data_direction
= DMA_NONE
;
1239 else if (rq_data_dir(req
) == WRITE
)
1240 cmd
->sc_data_direction
= DMA_TO_DEVICE
;
1242 cmd
->sc_data_direction
= DMA_FROM_DEVICE
;
1244 cmd
->transfersize
= req
->data_len
;
1245 cmd
->allowed
= req
->retries
;
1246 cmd
->timeout_per_command
= req
->timeout
;
1247 cmd
->done
= scsi_blk_pc_done
;
1250 static int scsi_prep_fn(struct request_queue
*q
, struct request
*req
)
1252 struct scsi_device
*sdev
= q
->queuedata
;
1253 struct scsi_cmnd
*cmd
;
1254 int specials_only
= 0;
1257 * Just check to see if the device is online. If it isn't, we
1258 * refuse to process any commands. The device must be brought
1259 * online before trying any recovery commands
1261 if (unlikely(!scsi_device_online(sdev
))) {
1262 sdev_printk(KERN_ERR
, sdev
,
1263 "rejecting I/O to offline device\n");
1266 if (unlikely(sdev
->sdev_state
!= SDEV_RUNNING
)) {
1267 /* OK, we're not in a running state don't prep
1269 if (sdev
->sdev_state
== SDEV_DEL
) {
1270 /* Device is fully deleted, no commands
1271 * at all allowed down */
1272 sdev_printk(KERN_ERR
, sdev
,
1273 "rejecting I/O to dead device\n");
1276 /* OK, we only allow special commands (i.e. not
1277 * user initiated ones */
1278 specials_only
= sdev
->sdev_state
;
1282 * Find the actual device driver associated with this command.
1283 * The SPECIAL requests are things like character device or
1284 * ioctls, which did not originate from ll_rw_blk. Note that
1285 * the special field is also used to indicate the cmd for
1286 * the remainder of a partially fulfilled request that can
1287 * come up when there is a medium error. We have to treat
1288 * these two cases differently. We differentiate by looking
1289 * at request->cmd, as this tells us the real story.
1291 if (req
->flags
& REQ_SPECIAL
&& req
->special
) {
1292 struct scsi_request
*sreq
= req
->special
;
1294 if (sreq
->sr_magic
== SCSI_REQ_MAGIC
) {
1295 cmd
= scsi_get_command(sreq
->sr_device
, GFP_ATOMIC
);
1298 scsi_init_cmd_from_req(cmd
, sreq
);
1301 } else if (req
->flags
& (REQ_CMD
| REQ_BLOCK_PC
)) {
1303 if(unlikely(specials_only
) && !(req
->flags
& REQ_SPECIAL
)) {
1304 if(specials_only
== SDEV_QUIESCE
||
1305 specials_only
== SDEV_BLOCK
)
1308 sdev_printk(KERN_ERR
, sdev
,
1309 "rejecting I/O to device being removed\n");
1315 * Now try and find a command block that we can use.
1317 if (!req
->special
) {
1318 cmd
= scsi_get_command(sdev
, GFP_ATOMIC
);
1324 /* pull a tag out of the request if we have one */
1325 cmd
->tag
= req
->tag
;
1327 blk_dump_rq_flags(req
, "SCSI bad req");
1331 /* note the overloading of req->special. When the tag
1332 * is active it always means cmd. If the tag goes
1333 * back for re-queueing, it may be reset */
1338 * FIXME: drop the lock here because the functions below
1339 * expect to be called without the queue lock held. Also,
1340 * previously, we dequeued the request before dropping the
1341 * lock. We hope REQ_STARTED prevents anything untoward from
1344 if (req
->flags
& (REQ_CMD
| REQ_BLOCK_PC
)) {
1348 * This will do a couple of things:
1349 * 1) Fill in the actual SCSI command.
1350 * 2) Fill in any other upper-level specific fields
1353 * If this returns 0, it means that the request failed
1354 * (reading past end of disk, reading offline device,
1355 * etc). This won't actually talk to the device, but
1356 * some kinds of consistency checking may cause the
1357 * request to be rejected immediately.
1361 * This sets up the scatter-gather table (allocating if
1364 ret
= scsi_init_io(cmd
);
1366 /* For BLKPREP_KILL/DEFER the cmd was released */
1374 * Initialize the actual SCSI command for this request.
1376 if (req
->flags
& REQ_BLOCK_PC
) {
1377 scsi_setup_blk_pc_cmnd(cmd
);
1378 } else if (req
->rq_disk
) {
1379 struct scsi_driver
*drv
;
1381 drv
= *(struct scsi_driver
**)req
->rq_disk
->private_data
;
1382 if (unlikely(!drv
->init_command(cmd
))) {
1383 scsi_release_buffers(cmd
);
1384 scsi_put_command(cmd
);
1391 * The request is now prepped, no need to come back here
1393 req
->flags
|= REQ_DONTPREP
;
1397 /* If we defer, the elv_next_request() returns NULL, but the
1398 * queue must be restarted, so we plug here if no returning
1399 * command will automatically do that. */
1400 if (sdev
->device_busy
== 0)
1402 return BLKPREP_DEFER
;
1404 req
->errors
= DID_NO_CONNECT
<< 16;
1405 return BLKPREP_KILL
;
1409 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1412 * Called with the queue_lock held.
1414 static inline int scsi_dev_queue_ready(struct request_queue
*q
,
1415 struct scsi_device
*sdev
)
1417 if (sdev
->device_busy
>= sdev
->queue_depth
)
1419 if (sdev
->device_busy
== 0 && sdev
->device_blocked
) {
1421 * unblock after device_blocked iterates to zero
1423 if (--sdev
->device_blocked
== 0) {
1425 sdev_printk(KERN_INFO
, sdev
,
1426 "unblocking device at zero depth\n"));
1432 if (sdev
->device_blocked
)
1439 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1440 * return 0. We must end up running the queue again whenever 0 is
1441 * returned, else IO can hang.
1443 * Called with host_lock held.
1445 static inline int scsi_host_queue_ready(struct request_queue
*q
,
1446 struct Scsi_Host
*shost
,
1447 struct scsi_device
*sdev
)
1449 if (scsi_host_in_recovery(shost
))
1451 if (shost
->host_busy
== 0 && shost
->host_blocked
) {
1453 * unblock after host_blocked iterates to zero
1455 if (--shost
->host_blocked
== 0) {
1457 printk("scsi%d unblocking host at zero depth\n",
1464 if ((shost
->can_queue
> 0 && shost
->host_busy
>= shost
->can_queue
) ||
1465 shost
->host_blocked
|| shost
->host_self_blocked
) {
1466 if (list_empty(&sdev
->starved_entry
))
1467 list_add_tail(&sdev
->starved_entry
, &shost
->starved_list
);
1471 /* We're OK to process the command, so we can't be starved */
1472 if (!list_empty(&sdev
->starved_entry
))
1473 list_del_init(&sdev
->starved_entry
);
1479 * Kill a request for a dead device
1481 static void scsi_kill_request(struct request
*req
, request_queue_t
*q
)
1483 struct scsi_cmnd
*cmd
= req
->special
;
1485 blkdev_dequeue_request(req
);
1487 if (unlikely(cmd
== NULL
)) {
1488 printk(KERN_CRIT
"impossible request in %s.\n",
1493 scsi_init_cmd_errh(cmd
);
1494 cmd
->result
= DID_NO_CONNECT
<< 16;
1495 atomic_inc(&cmd
->device
->iorequest_cnt
);
1499 static void scsi_softirq_done(struct request
*rq
)
1501 struct scsi_cmnd
*cmd
= rq
->completion_data
;
1502 unsigned long wait_for
= (cmd
->allowed
+ 1) * cmd
->timeout_per_command
;
1505 INIT_LIST_HEAD(&cmd
->eh_entry
);
1507 disposition
= scsi_decide_disposition(cmd
);
1508 if (disposition
!= SUCCESS
&&
1509 time_before(cmd
->jiffies_at_alloc
+ wait_for
, jiffies
)) {
1510 sdev_printk(KERN_ERR
, cmd
->device
,
1511 "timing out command, waited %lus\n",
1513 disposition
= SUCCESS
;
1516 scsi_log_completion(cmd
, disposition
);
1518 switch (disposition
) {
1520 scsi_finish_command(cmd
);
1523 scsi_retry_command(cmd
);
1525 case ADD_TO_MLQUEUE
:
1526 scsi_queue_insert(cmd
, SCSI_MLQUEUE_DEVICE_BUSY
);
1529 if (!scsi_eh_scmd_add(cmd
, 0))
1530 scsi_finish_command(cmd
);
1535 * Function: scsi_request_fn()
1537 * Purpose: Main strategy routine for SCSI.
1539 * Arguments: q - Pointer to actual queue.
1543 * Lock status: IO request lock assumed to be held when called.
1545 static void scsi_request_fn(struct request_queue
*q
)
1547 struct scsi_device
*sdev
= q
->queuedata
;
1548 struct Scsi_Host
*shost
;
1549 struct scsi_cmnd
*cmd
;
1550 struct request
*req
;
1553 printk("scsi: killing requests for dead queue\n");
1554 while ((req
= elv_next_request(q
)) != NULL
)
1555 scsi_kill_request(req
, q
);
1559 if(!get_device(&sdev
->sdev_gendev
))
1560 /* We must be tearing the block queue down already */
1564 * To start with, we keep looping until the queue is empty, or until
1565 * the host is no longer able to accept any more requests.
1568 while (!blk_queue_plugged(q
)) {
1571 * get next queueable request. We do this early to make sure
1572 * that the request is fully prepared even if we cannot
1575 req
= elv_next_request(q
);
1576 if (!req
|| !scsi_dev_queue_ready(q
, sdev
))
1579 if (unlikely(!scsi_device_online(sdev
))) {
1580 sdev_printk(KERN_ERR
, sdev
,
1581 "rejecting I/O to offline device\n");
1582 scsi_kill_request(req
, q
);
1588 * Remove the request from the request list.
1590 if (!(blk_queue_tagged(q
) && !blk_queue_start_tag(q
, req
)))
1591 blkdev_dequeue_request(req
);
1592 sdev
->device_busy
++;
1594 spin_unlock(q
->queue_lock
);
1596 if (unlikely(cmd
== NULL
)) {
1597 printk(KERN_CRIT
"impossible request in %s.\n"
1598 "please mail a stack trace to "
1599 "linux-scsi@vger.kernel.org",
1603 spin_lock(shost
->host_lock
);
1605 if (!scsi_host_queue_ready(q
, shost
, sdev
))
1607 if (sdev
->single_lun
) {
1608 if (scsi_target(sdev
)->starget_sdev_user
&&
1609 scsi_target(sdev
)->starget_sdev_user
!= sdev
)
1611 scsi_target(sdev
)->starget_sdev_user
= sdev
;
1616 * XXX(hch): This is rather suboptimal, scsi_dispatch_cmd will
1617 * take the lock again.
1619 spin_unlock_irq(shost
->host_lock
);
1622 * Finally, initialize any error handling parameters, and set up
1623 * the timers for timeouts.
1625 scsi_init_cmd_errh(cmd
);
1628 * Dispatch the command to the low-level driver.
1630 rtn
= scsi_dispatch_cmd(cmd
);
1631 spin_lock_irq(q
->queue_lock
);
1633 /* we're refusing the command; because of
1634 * the way locks get dropped, we need to
1635 * check here if plugging is required */
1636 if(sdev
->device_busy
== 0)
1646 spin_unlock_irq(shost
->host_lock
);
1649 * lock q, handle tag, requeue req, and decrement device_busy. We
1650 * must return with queue_lock held.
1652 * Decrementing device_busy without checking it is OK, as all such
1653 * cases (host limits or settings) should run the queue at some
1656 spin_lock_irq(q
->queue_lock
);
1657 blk_requeue_request(q
, req
);
1658 sdev
->device_busy
--;
1659 if(sdev
->device_busy
== 0)
1662 /* must be careful here...if we trigger the ->remove() function
1663 * we cannot be holding the q lock */
1664 spin_unlock_irq(q
->queue_lock
);
1665 put_device(&sdev
->sdev_gendev
);
1666 spin_lock_irq(q
->queue_lock
);
1669 u64
scsi_calculate_bounce_limit(struct Scsi_Host
*shost
)
1671 struct device
*host_dev
;
1672 u64 bounce_limit
= 0xffffffff;
1674 if (shost
->unchecked_isa_dma
)
1675 return BLK_BOUNCE_ISA
;
1677 * Platforms with virtual-DMA translation
1678 * hardware have no practical limit.
1680 if (!PCI_DMA_BUS_IS_PHYS
)
1681 return BLK_BOUNCE_ANY
;
1683 host_dev
= scsi_get_device(shost
);
1684 if (host_dev
&& host_dev
->dma_mask
)
1685 bounce_limit
= *host_dev
->dma_mask
;
1687 return bounce_limit
;
1689 EXPORT_SYMBOL(scsi_calculate_bounce_limit
);
1691 struct request_queue
*scsi_alloc_queue(struct scsi_device
*sdev
)
1693 struct Scsi_Host
*shost
= sdev
->host
;
1694 struct request_queue
*q
;
1696 q
= blk_init_queue(scsi_request_fn
, NULL
);
1700 blk_queue_prep_rq(q
, scsi_prep_fn
);
1702 blk_queue_max_hw_segments(q
, shost
->sg_tablesize
);
1703 blk_queue_max_phys_segments(q
, SCSI_MAX_PHYS_SEGMENTS
);
1704 blk_queue_max_sectors(q
, shost
->max_sectors
);
1705 blk_queue_bounce_limit(q
, scsi_calculate_bounce_limit(shost
));
1706 blk_queue_segment_boundary(q
, shost
->dma_boundary
);
1707 blk_queue_issue_flush_fn(q
, scsi_issue_flush_fn
);
1708 blk_queue_softirq_done(q
, scsi_softirq_done
);
1710 if (!shost
->use_clustering
)
1711 clear_bit(QUEUE_FLAG_CLUSTER
, &q
->queue_flags
);
1715 void scsi_free_queue(struct request_queue
*q
)
1717 blk_cleanup_queue(q
);
1721 * Function: scsi_block_requests()
1723 * Purpose: Utility function used by low-level drivers to prevent further
1724 * commands from being queued to the device.
1726 * Arguments: shost - Host in question
1730 * Lock status: No locks are assumed held.
1732 * Notes: There is no timer nor any other means by which the requests
1733 * get unblocked other than the low-level driver calling
1734 * scsi_unblock_requests().
1736 void scsi_block_requests(struct Scsi_Host
*shost
)
1738 shost
->host_self_blocked
= 1;
1740 EXPORT_SYMBOL(scsi_block_requests
);
1743 * Function: scsi_unblock_requests()
1745 * Purpose: Utility function used by low-level drivers to allow further
1746 * commands from being queued to the device.
1748 * Arguments: shost - Host in question
1752 * Lock status: No locks are assumed held.
1754 * Notes: There is no timer nor any other means by which the requests
1755 * get unblocked other than the low-level driver calling
1756 * scsi_unblock_requests().
1758 * This is done as an API function so that changes to the
1759 * internals of the scsi mid-layer won't require wholesale
1760 * changes to drivers that use this feature.
1762 void scsi_unblock_requests(struct Scsi_Host
*shost
)
1764 shost
->host_self_blocked
= 0;
1765 scsi_run_host_queues(shost
);
1767 EXPORT_SYMBOL(scsi_unblock_requests
);
1769 int __init
scsi_init_queue(void)
1773 scsi_io_context_cache
= kmem_cache_create("scsi_io_context",
1774 sizeof(struct scsi_io_context
),
1776 if (!scsi_io_context_cache
) {
1777 printk(KERN_ERR
"SCSI: can't init scsi io context cache\n");
1781 for (i
= 0; i
< SG_MEMPOOL_NR
; i
++) {
1782 struct scsi_host_sg_pool
*sgp
= scsi_sg_pools
+ i
;
1783 int size
= sgp
->size
* sizeof(struct scatterlist
);
1785 sgp
->slab
= kmem_cache_create(sgp
->name
, size
, 0,
1786 SLAB_HWCACHE_ALIGN
, NULL
, NULL
);
1788 printk(KERN_ERR
"SCSI: can't init sg slab %s\n",
1792 sgp
->pool
= mempool_create(SG_MEMPOOL_SIZE
,
1793 mempool_alloc_slab
, mempool_free_slab
,
1796 printk(KERN_ERR
"SCSI: can't init sg mempool %s\n",
1804 void scsi_exit_queue(void)
1808 kmem_cache_destroy(scsi_io_context_cache
);
1810 for (i
= 0; i
< SG_MEMPOOL_NR
; i
++) {
1811 struct scsi_host_sg_pool
*sgp
= scsi_sg_pools
+ i
;
1812 mempool_destroy(sgp
->pool
);
1813 kmem_cache_destroy(sgp
->slab
);
1817 * scsi_mode_sense - issue a mode sense, falling back from 10 to
1818 * six bytes if necessary.
1819 * @sdev: SCSI device to be queried
1820 * @dbd: set if mode sense will allow block descriptors to be returned
1821 * @modepage: mode page being requested
1822 * @buffer: request buffer (may not be smaller than eight bytes)
1823 * @len: length of request buffer.
1824 * @timeout: command timeout
1825 * @retries: number of retries before failing
1826 * @data: returns a structure abstracting the mode header data
1827 * @sense: place to put sense data (or NULL if no sense to be collected).
1828 * must be SCSI_SENSE_BUFFERSIZE big.
1830 * Returns zero if unsuccessful, or the header offset (either 4
1831 * or 8 depending on whether a six or ten byte command was
1832 * issued) if successful.
1835 scsi_mode_sense(struct scsi_device
*sdev
, int dbd
, int modepage
,
1836 unsigned char *buffer
, int len
, int timeout
, int retries
,
1837 struct scsi_mode_data
*data
, struct scsi_sense_hdr
*sshdr
) {
1838 unsigned char cmd
[12];
1842 struct scsi_sense_hdr my_sshdr
;
1844 memset(data
, 0, sizeof(*data
));
1845 memset(&cmd
[0], 0, 12);
1846 cmd
[1] = dbd
& 0x18; /* allows DBD and LLBA bits */
1849 /* caller might not be interested in sense, but we need it */
1854 use_10_for_ms
= sdev
->use_10_for_ms
;
1856 if (use_10_for_ms
) {
1860 cmd
[0] = MODE_SENSE_10
;
1867 cmd
[0] = MODE_SENSE
;
1872 memset(buffer
, 0, len
);
1874 result
= scsi_execute_req(sdev
, cmd
, DMA_FROM_DEVICE
, buffer
, len
,
1875 sshdr
, timeout
, retries
);
1877 /* This code looks awful: what it's doing is making sure an
1878 * ILLEGAL REQUEST sense return identifies the actual command
1879 * byte as the problem. MODE_SENSE commands can return
1880 * ILLEGAL REQUEST if the code page isn't supported */
1882 if (use_10_for_ms
&& !scsi_status_is_good(result
) &&
1883 (driver_byte(result
) & DRIVER_SENSE
)) {
1884 if (scsi_sense_valid(sshdr
)) {
1885 if ((sshdr
->sense_key
== ILLEGAL_REQUEST
) &&
1886 (sshdr
->asc
== 0x20) && (sshdr
->ascq
== 0)) {
1888 * Invalid command operation code
1890 sdev
->use_10_for_ms
= 0;
1896 if(scsi_status_is_good(result
)) {
1897 data
->header_length
= header_length
;
1899 data
->length
= buffer
[0]*256 + buffer
[1] + 2;
1900 data
->medium_type
= buffer
[2];
1901 data
->device_specific
= buffer
[3];
1902 data
->longlba
= buffer
[4] & 0x01;
1903 data
->block_descriptor_length
= buffer
[6]*256
1906 data
->length
= buffer
[0] + 1;
1907 data
->medium_type
= buffer
[1];
1908 data
->device_specific
= buffer
[2];
1909 data
->block_descriptor_length
= buffer
[3];
1915 EXPORT_SYMBOL(scsi_mode_sense
);
1918 scsi_test_unit_ready(struct scsi_device
*sdev
, int timeout
, int retries
)
1921 TEST_UNIT_READY
, 0, 0, 0, 0, 0,
1923 struct scsi_sense_hdr sshdr
;
1926 result
= scsi_execute_req(sdev
, cmd
, DMA_NONE
, NULL
, 0, &sshdr
,
1929 if ((driver_byte(result
) & DRIVER_SENSE
) && sdev
->removable
) {
1931 if ((scsi_sense_valid(&sshdr
)) &&
1932 ((sshdr
.sense_key
== UNIT_ATTENTION
) ||
1933 (sshdr
.sense_key
== NOT_READY
))) {
1940 EXPORT_SYMBOL(scsi_test_unit_ready
);
1943 * scsi_device_set_state - Take the given device through the device
1945 * @sdev: scsi device to change the state of.
1946 * @state: state to change to.
1948 * Returns zero if unsuccessful or an error if the requested
1949 * transition is illegal.
1952 scsi_device_set_state(struct scsi_device
*sdev
, enum scsi_device_state state
)
1954 enum scsi_device_state oldstate
= sdev
->sdev_state
;
1956 if (state
== oldstate
)
1961 /* There are no legal states that come back to
1962 * created. This is the manually initialised start
2032 sdev
->sdev_state
= state
;
2036 SCSI_LOG_ERROR_RECOVERY(1,
2037 sdev_printk(KERN_ERR
, sdev
,
2038 "Illegal state transition %s->%s\n",
2039 scsi_device_state_name(oldstate
),
2040 scsi_device_state_name(state
))
2044 EXPORT_SYMBOL(scsi_device_set_state
);
2047 * scsi_device_quiesce - Block user issued commands.
2048 * @sdev: scsi device to quiesce.
2050 * This works by trying to transition to the SDEV_QUIESCE state
2051 * (which must be a legal transition). When the device is in this
2052 * state, only special requests will be accepted, all others will
2053 * be deferred. Since special requests may also be requeued requests,
2054 * a successful return doesn't guarantee the device will be
2055 * totally quiescent.
2057 * Must be called with user context, may sleep.
2059 * Returns zero if unsuccessful or an error if not.
2062 scsi_device_quiesce(struct scsi_device
*sdev
)
2064 int err
= scsi_device_set_state(sdev
, SDEV_QUIESCE
);
2068 scsi_run_queue(sdev
->request_queue
);
2069 while (sdev
->device_busy
) {
2070 msleep_interruptible(200);
2071 scsi_run_queue(sdev
->request_queue
);
2075 EXPORT_SYMBOL(scsi_device_quiesce
);
2078 * scsi_device_resume - Restart user issued commands to a quiesced device.
2079 * @sdev: scsi device to resume.
2081 * Moves the device from quiesced back to running and restarts the
2084 * Must be called with user context, may sleep.
2087 scsi_device_resume(struct scsi_device
*sdev
)
2089 if(scsi_device_set_state(sdev
, SDEV_RUNNING
))
2091 scsi_run_queue(sdev
->request_queue
);
2093 EXPORT_SYMBOL(scsi_device_resume
);
2096 device_quiesce_fn(struct scsi_device
*sdev
, void *data
)
2098 scsi_device_quiesce(sdev
);
2102 scsi_target_quiesce(struct scsi_target
*starget
)
2104 starget_for_each_device(starget
, NULL
, device_quiesce_fn
);
2106 EXPORT_SYMBOL(scsi_target_quiesce
);
2109 device_resume_fn(struct scsi_device
*sdev
, void *data
)
2111 scsi_device_resume(sdev
);
2115 scsi_target_resume(struct scsi_target
*starget
)
2117 starget_for_each_device(starget
, NULL
, device_resume_fn
);
2119 EXPORT_SYMBOL(scsi_target_resume
);
2122 * scsi_internal_device_block - internal function to put a device
2123 * temporarily into the SDEV_BLOCK state
2124 * @sdev: device to block
2126 * Block request made by scsi lld's to temporarily stop all
2127 * scsi commands on the specified device. Called from interrupt
2128 * or normal process context.
2130 * Returns zero if successful or error if not
2133 * This routine transitions the device to the SDEV_BLOCK state
2134 * (which must be a legal transition). When the device is in this
2135 * state, all commands are deferred until the scsi lld reenables
2136 * the device with scsi_device_unblock or device_block_tmo fires.
2137 * This routine assumes the host_lock is held on entry.
2140 scsi_internal_device_block(struct scsi_device
*sdev
)
2142 request_queue_t
*q
= sdev
->request_queue
;
2143 unsigned long flags
;
2146 err
= scsi_device_set_state(sdev
, SDEV_BLOCK
);
2151 * The device has transitioned to SDEV_BLOCK. Stop the
2152 * block layer from calling the midlayer with this device's
2155 spin_lock_irqsave(q
->queue_lock
, flags
);
2157 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2161 EXPORT_SYMBOL_GPL(scsi_internal_device_block
);
2164 * scsi_internal_device_unblock - resume a device after a block request
2165 * @sdev: device to resume
2167 * Called by scsi lld's or the midlayer to restart the device queue
2168 * for the previously suspended scsi device. Called from interrupt or
2169 * normal process context.
2171 * Returns zero if successful or error if not.
2174 * This routine transitions the device to the SDEV_RUNNING state
2175 * (which must be a legal transition) allowing the midlayer to
2176 * goose the queue for this device. This routine assumes the
2177 * host_lock is held upon entry.
2180 scsi_internal_device_unblock(struct scsi_device
*sdev
)
2182 request_queue_t
*q
= sdev
->request_queue
;
2184 unsigned long flags
;
2187 * Try to transition the scsi device to SDEV_RUNNING
2188 * and goose the device queue if successful.
2190 err
= scsi_device_set_state(sdev
, SDEV_RUNNING
);
2194 spin_lock_irqsave(q
->queue_lock
, flags
);
2196 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2200 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock
);
2203 device_block(struct scsi_device
*sdev
, void *data
)
2205 scsi_internal_device_block(sdev
);
2209 target_block(struct device
*dev
, void *data
)
2211 if (scsi_is_target_device(dev
))
2212 starget_for_each_device(to_scsi_target(dev
), NULL
,
2218 scsi_target_block(struct device
*dev
)
2220 if (scsi_is_target_device(dev
))
2221 starget_for_each_device(to_scsi_target(dev
), NULL
,
2224 device_for_each_child(dev
, NULL
, target_block
);
2226 EXPORT_SYMBOL_GPL(scsi_target_block
);
2229 device_unblock(struct scsi_device
*sdev
, void *data
)
2231 scsi_internal_device_unblock(sdev
);
2235 target_unblock(struct device
*dev
, void *data
)
2237 if (scsi_is_target_device(dev
))
2238 starget_for_each_device(to_scsi_target(dev
), NULL
,
2244 scsi_target_unblock(struct device
*dev
)
2246 if (scsi_is_target_device(dev
))
2247 starget_for_each_device(to_scsi_target(dev
), NULL
,
2250 device_for_each_child(dev
, NULL
, target_unblock
);
2252 EXPORT_SYMBOL_GPL(scsi_target_unblock
);
2255 struct work_queue_work
{
2256 struct work_struct work
;
2261 static void execute_in_process_context_work(void *data
)
2263 void (*fn
)(void *data
);
2264 struct work_queue_work
*wqw
= data
;
2275 * scsi_execute_in_process_context - reliably execute the routine with user context
2276 * @fn: the function to execute
2277 * @data: data to pass to the function
2279 * Executes the function immediately if process context is available,
2280 * otherwise schedules the function for delayed execution.
2282 * Returns: 0 - function was executed
2283 * 1 - function was scheduled for execution
2286 int scsi_execute_in_process_context(void (*fn
)(void *data
), void *data
)
2288 struct work_queue_work
*wqw
;
2290 if (!in_interrupt()) {
2295 wqw
= kmalloc(sizeof(struct work_queue_work
), GFP_ATOMIC
);
2297 if (unlikely(!wqw
)) {
2298 printk(KERN_ERR
"Failed to allocate memory\n");
2303 INIT_WORK(&wqw
->work
, execute_in_process_context_work
, wqw
);
2306 schedule_work(&wqw
->work
);
2310 EXPORT_SYMBOL_GPL(scsi_execute_in_process_context
);