xref: /openbmc/linux/drivers/scsi/scsi.c (revision 606d099c)
1 /*
2  *  scsi.c Copyright (C) 1992 Drew Eckhardt
3  *         Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
4  *         Copyright (C) 2002, 2003 Christoph Hellwig
5  *
6  *  generic mid-level SCSI driver
7  *      Initial versions: Drew Eckhardt
8  *      Subsequent revisions: Eric Youngdale
9  *
10  *  <drew@colorado.edu>
11  *
12  *  Bug correction thanks go to :
13  *      Rik Faith <faith@cs.unc.edu>
14  *      Tommy Thorn <tthorn>
15  *      Thomas Wuensche <tw@fgb1.fgb.mw.tu-muenchen.de>
16  *
17  *  Modified by Eric Youngdale eric@andante.org or ericy@gnu.ai.mit.edu to
18  *  add scatter-gather, multiple outstanding request, and other
19  *  enhancements.
20  *
21  *  Native multichannel, wide scsi, /proc/scsi and hot plugging
22  *  support added by Michael Neuffer <mike@i-connect.net>
23  *
24  *  Added request_module("scsi_hostadapter") for kerneld:
25  *  (Put an "alias scsi_hostadapter your_hostadapter" in /etc/modprobe.conf)
26  *  Bjorn Ekwall  <bj0rn@blox.se>
27  *  (changed to kmod)
28  *
29  *  Major improvements to the timeout, abort, and reset processing,
30  *  as well as performance modifications for large queue depths by
31  *  Leonard N. Zubkoff <lnz@dandelion.com>
32  *
33  *  Converted cli() code to spinlocks, Ingo Molnar
34  *
35  *  Jiffies wrap fixes (host->resetting), 3 Dec 1998 Andrea Arcangeli
36  *
37  *  out_of_space hacks, D. Gilbert (dpg) 990608
38  */
39 
40 #include <linux/module.h>
41 #include <linux/moduleparam.h>
42 #include <linux/kernel.h>
43 #include <linux/sched.h>
44 #include <linux/timer.h>
45 #include <linux/string.h>
46 #include <linux/slab.h>
47 #include <linux/blkdev.h>
48 #include <linux/delay.h>
49 #include <linux/init.h>
50 #include <linux/completion.h>
51 #include <linux/unistd.h>
52 #include <linux/spinlock.h>
53 #include <linux/kmod.h>
54 #include <linux/interrupt.h>
55 #include <linux/notifier.h>
56 #include <linux/cpu.h>
57 #include <linux/mutex.h>
58 
59 #include <scsi/scsi.h>
60 #include <scsi/scsi_cmnd.h>
61 #include <scsi/scsi_dbg.h>
62 #include <scsi/scsi_device.h>
63 #include <scsi/scsi_eh.h>
64 #include <scsi/scsi_host.h>
65 #include <scsi/scsi_tcq.h>
66 
67 #include "scsi_priv.h"
68 #include "scsi_logging.h"
69 
70 static void scsi_done(struct scsi_cmnd *cmd);
71 
72 /*
73  * Definitions and constants.
74  */
75 
76 #define MIN_RESET_DELAY (2*HZ)
77 
78 /* Do not call reset on error if we just did a reset within 15 sec. */
79 #define MIN_RESET_PERIOD (15*HZ)
80 
81 /*
82  * Macro to determine the size of SCSI command. This macro takes vendor
83  * unique commands into account. SCSI commands in groups 6 and 7 are
84  * vendor unique and we will depend upon the command length being
85  * supplied correctly in cmd_len.
86  */
87 #define CDB_SIZE(cmd)	(((((cmd)->cmnd[0] >> 5) & 7) < 6) ? \
88 				COMMAND_SIZE((cmd)->cmnd[0]) : (cmd)->cmd_len)
89 
90 /*
91  * Note - the initial logging level can be set here to log events at boot time.
92  * After the system is up, you may enable logging via the /proc interface.
93  */
94 unsigned int scsi_logging_level;
95 #if defined(CONFIG_SCSI_LOGGING)
96 EXPORT_SYMBOL(scsi_logging_level);
97 #endif
98 
99 /* NB: These are exposed through /proc/scsi/scsi and form part of the ABI.
100  * You may not alter any existing entry (although adding new ones is
101  * encouraged once assigned by ANSI/INCITS T10
102  */
103 static const char *const scsi_device_types[] = {
104 	"Direct-Access    ",
105 	"Sequential-Access",
106 	"Printer          ",
107 	"Processor        ",
108 	"WORM             ",
109 	"CD-ROM           ",
110 	"Scanner          ",
111 	"Optical Device   ",
112 	"Medium Changer   ",
113 	"Communications   ",
114 	"ASC IT8          ",
115 	"ASC IT8          ",
116 	"RAID             ",
117 	"Enclosure        ",
118 	"Direct-Access-RBC",
119 	"Optical card     ",
120 	"Bridge controller",
121 	"Object storage   ",
122 	"Automation/Drive ",
123 };
124 
125 const char * scsi_device_type(unsigned type)
126 {
127 	if (type == 0x1e)
128 		return "Well-known LUN   ";
129 	if (type == 0x1f)
130 		return "No Device        ";
131 	if (type >= ARRAY_SIZE(scsi_device_types))
132 		return "Unknown          ";
133 	return scsi_device_types[type];
134 }
135 
136 EXPORT_SYMBOL(scsi_device_type);
137 
138 struct scsi_host_cmd_pool {
139 	struct kmem_cache	*slab;
140 	unsigned int	users;
141 	char		*name;
142 	unsigned int	slab_flags;
143 	gfp_t		gfp_mask;
144 };
145 
146 static struct scsi_host_cmd_pool scsi_cmd_pool = {
147 	.name		= "scsi_cmd_cache",
148 	.slab_flags	= SLAB_HWCACHE_ALIGN,
149 };
150 
151 static struct scsi_host_cmd_pool scsi_cmd_dma_pool = {
152 	.name		= "scsi_cmd_cache(DMA)",
153 	.slab_flags	= SLAB_HWCACHE_ALIGN|SLAB_CACHE_DMA,
154 	.gfp_mask	= __GFP_DMA,
155 };
156 
157 static DEFINE_MUTEX(host_cmd_pool_mutex);
158 
159 struct scsi_cmnd *__scsi_get_command(struct Scsi_Host *shost, gfp_t gfp_mask)
160 {
161 	struct scsi_cmnd *cmd;
162 
163 	cmd = kmem_cache_alloc(shost->cmd_pool->slab,
164 			gfp_mask | shost->cmd_pool->gfp_mask);
165 
166 	if (unlikely(!cmd)) {
167 		unsigned long flags;
168 
169 		spin_lock_irqsave(&shost->free_list_lock, flags);
170 		if (likely(!list_empty(&shost->free_list))) {
171 			cmd = list_entry(shost->free_list.next,
172 					 struct scsi_cmnd, list);
173 			list_del_init(&cmd->list);
174 		}
175 		spin_unlock_irqrestore(&shost->free_list_lock, flags);
176 	}
177 
178 	return cmd;
179 }
180 EXPORT_SYMBOL_GPL(__scsi_get_command);
181 
182 /*
183  * Function:	scsi_get_command()
184  *
185  * Purpose:	Allocate and setup a scsi command block
186  *
187  * Arguments:	dev	- parent scsi device
188  *		gfp_mask- allocator flags
189  *
190  * Returns:	The allocated scsi command structure.
191  */
192 struct scsi_cmnd *scsi_get_command(struct scsi_device *dev, gfp_t gfp_mask)
193 {
194 	struct scsi_cmnd *cmd;
195 
196 	/* Bail if we can't get a reference to the device */
197 	if (!get_device(&dev->sdev_gendev))
198 		return NULL;
199 
200 	cmd = __scsi_get_command(dev->host, gfp_mask);
201 
202 	if (likely(cmd != NULL)) {
203 		unsigned long flags;
204 
205 		memset(cmd, 0, sizeof(*cmd));
206 		cmd->device = dev;
207 		init_timer(&cmd->eh_timeout);
208 		INIT_LIST_HEAD(&cmd->list);
209 		spin_lock_irqsave(&dev->list_lock, flags);
210 		list_add_tail(&cmd->list, &dev->cmd_list);
211 		spin_unlock_irqrestore(&dev->list_lock, flags);
212 		cmd->jiffies_at_alloc = jiffies;
213 	} else
214 		put_device(&dev->sdev_gendev);
215 
216 	return cmd;
217 }
218 EXPORT_SYMBOL(scsi_get_command);
219 
220 void __scsi_put_command(struct Scsi_Host *shost, struct scsi_cmnd *cmd,
221 			struct device *dev)
222 {
223 	unsigned long flags;
224 
225 	/* changing locks here, don't need to restore the irq state */
226 	spin_lock_irqsave(&shost->free_list_lock, flags);
227 	if (unlikely(list_empty(&shost->free_list))) {
228 		list_add(&cmd->list, &shost->free_list);
229 		cmd = NULL;
230 	}
231 	spin_unlock_irqrestore(&shost->free_list_lock, flags);
232 
233 	if (likely(cmd != NULL))
234 		kmem_cache_free(shost->cmd_pool->slab, cmd);
235 
236 	put_device(dev);
237 }
238 EXPORT_SYMBOL(__scsi_put_command);
239 
240 /*
241  * Function:	scsi_put_command()
242  *
243  * Purpose:	Free a scsi command block
244  *
245  * Arguments:	cmd	- command block to free
246  *
247  * Returns:	Nothing.
248  *
249  * Notes:	The command must not belong to any lists.
250  */
251 void scsi_put_command(struct scsi_cmnd *cmd)
252 {
253 	struct scsi_device *sdev = cmd->device;
254 	unsigned long flags;
255 
256 	/* serious error if the command hasn't come from a device list */
257 	spin_lock_irqsave(&cmd->device->list_lock, flags);
258 	BUG_ON(list_empty(&cmd->list));
259 	list_del_init(&cmd->list);
260 	spin_unlock_irqrestore(&cmd->device->list_lock, flags);
261 
262 	__scsi_put_command(cmd->device->host, cmd, &sdev->sdev_gendev);
263 }
264 EXPORT_SYMBOL(scsi_put_command);
265 
266 /*
267  * Function:	scsi_setup_command_freelist()
268  *
269  * Purpose:	Setup the command freelist for a scsi host.
270  *
271  * Arguments:	shost	- host to allocate the freelist for.
272  *
273  * Returns:	Nothing.
274  */
275 int scsi_setup_command_freelist(struct Scsi_Host *shost)
276 {
277 	struct scsi_host_cmd_pool *pool;
278 	struct scsi_cmnd *cmd;
279 
280 	spin_lock_init(&shost->free_list_lock);
281 	INIT_LIST_HEAD(&shost->free_list);
282 
283 	/*
284 	 * Select a command slab for this host and create it if not
285 	 * yet existant.
286 	 */
287 	mutex_lock(&host_cmd_pool_mutex);
288 	pool = (shost->unchecked_isa_dma ? &scsi_cmd_dma_pool : &scsi_cmd_pool);
289 	if (!pool->users) {
290 		pool->slab = kmem_cache_create(pool->name,
291 				sizeof(struct scsi_cmnd), 0,
292 				pool->slab_flags, NULL, NULL);
293 		if (!pool->slab)
294 			goto fail;
295 	}
296 
297 	pool->users++;
298 	shost->cmd_pool = pool;
299 	mutex_unlock(&host_cmd_pool_mutex);
300 
301 	/*
302 	 * Get one backup command for this host.
303 	 */
304 	cmd = kmem_cache_alloc(shost->cmd_pool->slab,
305 			GFP_KERNEL | shost->cmd_pool->gfp_mask);
306 	if (!cmd)
307 		goto fail2;
308 	list_add(&cmd->list, &shost->free_list);
309 	return 0;
310 
311  fail2:
312 	if (!--pool->users)
313 		kmem_cache_destroy(pool->slab);
314 	return -ENOMEM;
315  fail:
316 	mutex_unlock(&host_cmd_pool_mutex);
317 	return -ENOMEM;
318 
319 }
320 
321 /*
322  * Function:	scsi_destroy_command_freelist()
323  *
324  * Purpose:	Release the command freelist for a scsi host.
325  *
326  * Arguments:	shost	- host that's freelist is going to be destroyed
327  */
328 void scsi_destroy_command_freelist(struct Scsi_Host *shost)
329 {
330 	while (!list_empty(&shost->free_list)) {
331 		struct scsi_cmnd *cmd;
332 
333 		cmd = list_entry(shost->free_list.next, struct scsi_cmnd, list);
334 		list_del_init(&cmd->list);
335 		kmem_cache_free(shost->cmd_pool->slab, cmd);
336 	}
337 
338 	mutex_lock(&host_cmd_pool_mutex);
339 	if (!--shost->cmd_pool->users)
340 		kmem_cache_destroy(shost->cmd_pool->slab);
341 	mutex_unlock(&host_cmd_pool_mutex);
342 }
343 
344 #ifdef CONFIG_SCSI_LOGGING
345 void scsi_log_send(struct scsi_cmnd *cmd)
346 {
347 	unsigned int level;
348 	struct scsi_device *sdev;
349 
350 	/*
351 	 * If ML QUEUE log level is greater than or equal to:
352 	 *
353 	 * 1: nothing (match completion)
354 	 *
355 	 * 2: log opcode + command of all commands
356 	 *
357 	 * 3: same as 2 plus dump cmd address
358 	 *
359 	 * 4: same as 3 plus dump extra junk
360 	 */
361 	if (unlikely(scsi_logging_level)) {
362 		level = SCSI_LOG_LEVEL(SCSI_LOG_MLQUEUE_SHIFT,
363 				       SCSI_LOG_MLQUEUE_BITS);
364 		if (level > 1) {
365 			sdev = cmd->device;
366 			sdev_printk(KERN_INFO, sdev, "send ");
367 			if (level > 2)
368 				printk("0x%p ", cmd);
369 			/*
370 			 * spaces to match disposition and cmd->result
371 			 * output in scsi_log_completion.
372 			 */
373 			printk("                 ");
374 			scsi_print_command(cmd);
375 			if (level > 3) {
376 				printk(KERN_INFO "buffer = 0x%p, bufflen = %d,"
377 				       " done = 0x%p, queuecommand 0x%p\n",
378 					cmd->request_buffer, cmd->request_bufflen,
379 					cmd->done,
380 					sdev->host->hostt->queuecommand);
381 
382 			}
383 		}
384 	}
385 }
386 
387 void scsi_log_completion(struct scsi_cmnd *cmd, int disposition)
388 {
389 	unsigned int level;
390 	struct scsi_device *sdev;
391 
392 	/*
393 	 * If ML COMPLETE log level is greater than or equal to:
394 	 *
395 	 * 1: log disposition, result, opcode + command, and conditionally
396 	 * sense data for failures or non SUCCESS dispositions.
397 	 *
398 	 * 2: same as 1 but for all command completions.
399 	 *
400 	 * 3: same as 2 plus dump cmd address
401 	 *
402 	 * 4: same as 3 plus dump extra junk
403 	 */
404 	if (unlikely(scsi_logging_level)) {
405 		level = SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
406 				       SCSI_LOG_MLCOMPLETE_BITS);
407 		if (((level > 0) && (cmd->result || disposition != SUCCESS)) ||
408 		    (level > 1)) {
409 			sdev = cmd->device;
410 			sdev_printk(KERN_INFO, sdev, "done ");
411 			if (level > 2)
412 				printk("0x%p ", cmd);
413 			/*
414 			 * Dump truncated values, so we usually fit within
415 			 * 80 chars.
416 			 */
417 			switch (disposition) {
418 			case SUCCESS:
419 				printk("SUCCESS");
420 				break;
421 			case NEEDS_RETRY:
422 				printk("RETRY  ");
423 				break;
424 			case ADD_TO_MLQUEUE:
425 				printk("MLQUEUE");
426 				break;
427 			case FAILED:
428 				printk("FAILED ");
429 				break;
430 			case TIMEOUT_ERROR:
431 				/*
432 				 * If called via scsi_times_out.
433 				 */
434 				printk("TIMEOUT");
435 				break;
436 			default:
437 				printk("UNKNOWN");
438 			}
439 			printk(" %8x ", cmd->result);
440 			scsi_print_command(cmd);
441 			if (status_byte(cmd->result) & CHECK_CONDITION) {
442 				/*
443 				 * XXX The scsi_print_sense formatting/prefix
444 				 * doesn't match this function.
445 				 */
446 				scsi_print_sense("", cmd);
447 			}
448 			if (level > 3) {
449 				printk(KERN_INFO "scsi host busy %d failed %d\n",
450 				       sdev->host->host_busy,
451 				       sdev->host->host_failed);
452 			}
453 		}
454 	}
455 }
456 #endif
457 
458 /*
459  * Assign a serial number and pid to the request for error recovery
460  * and debugging purposes.  Protected by the Host_Lock of host.
461  */
462 static inline void scsi_cmd_get_serial(struct Scsi_Host *host, struct scsi_cmnd *cmd)
463 {
464 	cmd->serial_number = host->cmd_serial_number++;
465 	if (cmd->serial_number == 0)
466 		cmd->serial_number = host->cmd_serial_number++;
467 
468 	cmd->pid = host->cmd_pid++;
469 	if (cmd->pid == 0)
470 		cmd->pid = host->cmd_pid++;
471 }
472 
473 /*
474  * Function:    scsi_dispatch_command
475  *
476  * Purpose:     Dispatch a command to the low-level driver.
477  *
478  * Arguments:   cmd - command block we are dispatching.
479  *
480  * Notes:
481  */
482 int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
483 {
484 	struct Scsi_Host *host = cmd->device->host;
485 	unsigned long flags = 0;
486 	unsigned long timeout;
487 	int rtn = 0;
488 
489 	/* check if the device is still usable */
490 	if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
491 		/* in SDEV_DEL we error all commands. DID_NO_CONNECT
492 		 * returns an immediate error upwards, and signals
493 		 * that the device is no longer present */
494 		cmd->result = DID_NO_CONNECT << 16;
495 		atomic_inc(&cmd->device->iorequest_cnt);
496 		__scsi_done(cmd);
497 		/* return 0 (because the command has been processed) */
498 		goto out;
499 	}
500 
501 	/* Check to see if the scsi lld put this device into state SDEV_BLOCK. */
502 	if (unlikely(cmd->device->sdev_state == SDEV_BLOCK)) {
503 		/*
504 		 * in SDEV_BLOCK, the command is just put back on the device
505 		 * queue.  The suspend state has already blocked the queue so
506 		 * future requests should not occur until the device
507 		 * transitions out of the suspend state.
508 		 */
509 		scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
510 
511 		SCSI_LOG_MLQUEUE(3, printk("queuecommand : device blocked \n"));
512 
513 		/*
514 		 * NOTE: rtn is still zero here because we don't need the
515 		 * queue to be plugged on return (it's already stopped)
516 		 */
517 		goto out;
518 	}
519 
520 	/*
521 	 * If SCSI-2 or lower, store the LUN value in cmnd.
522 	 */
523 	if (cmd->device->scsi_level <= SCSI_2 &&
524 	    cmd->device->scsi_level != SCSI_UNKNOWN) {
525 		cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
526 			       (cmd->device->lun << 5 & 0xe0);
527 	}
528 
529 	/*
530 	 * We will wait MIN_RESET_DELAY clock ticks after the last reset so
531 	 * we can avoid the drive not being ready.
532 	 */
533 	timeout = host->last_reset + MIN_RESET_DELAY;
534 
535 	if (host->resetting && time_before(jiffies, timeout)) {
536 		int ticks_remaining = timeout - jiffies;
537 		/*
538 		 * NOTE: This may be executed from within an interrupt
539 		 * handler!  This is bad, but for now, it'll do.  The irq
540 		 * level of the interrupt handler has been masked out by the
541 		 * platform dependent interrupt handling code already, so the
542 		 * sti() here will not cause another call to the SCSI host's
543 		 * interrupt handler (assuming there is one irq-level per
544 		 * host).
545 		 */
546 		while (--ticks_remaining >= 0)
547 			mdelay(1 + 999 / HZ);
548 		host->resetting = 0;
549 	}
550 
551 	/*
552 	 * AK: unlikely race here: for some reason the timer could
553 	 * expire before the serial number is set up below.
554 	 */
555 	scsi_add_timer(cmd, cmd->timeout_per_command, scsi_times_out);
556 
557 	scsi_log_send(cmd);
558 
559 	/*
560 	 * We will use a queued command if possible, otherwise we will
561 	 * emulate the queuing and calling of completion function ourselves.
562 	 */
563 	atomic_inc(&cmd->device->iorequest_cnt);
564 
565 	/*
566 	 * Before we queue this command, check if the command
567 	 * length exceeds what the host adapter can handle.
568 	 */
569 	if (CDB_SIZE(cmd) > cmd->device->host->max_cmd_len) {
570 		SCSI_LOG_MLQUEUE(3,
571 				printk("queuecommand : command too long.\n"));
572 		cmd->result = (DID_ABORT << 16);
573 
574 		scsi_done(cmd);
575 		goto out;
576 	}
577 
578 	spin_lock_irqsave(host->host_lock, flags);
579 	scsi_cmd_get_serial(host, cmd);
580 
581 	if (unlikely(host->shost_state == SHOST_DEL)) {
582 		cmd->result = (DID_NO_CONNECT << 16);
583 		scsi_done(cmd);
584 	} else {
585 		rtn = host->hostt->queuecommand(cmd, scsi_done);
586 	}
587 	spin_unlock_irqrestore(host->host_lock, flags);
588 	if (rtn) {
589 		if (scsi_delete_timer(cmd)) {
590 			atomic_inc(&cmd->device->iodone_cnt);
591 			scsi_queue_insert(cmd,
592 					  (rtn == SCSI_MLQUEUE_DEVICE_BUSY) ?
593 					  rtn : SCSI_MLQUEUE_HOST_BUSY);
594 		}
595 		SCSI_LOG_MLQUEUE(3,
596 		    printk("queuecommand : request rejected\n"));
597 	}
598 
599  out:
600 	SCSI_LOG_MLQUEUE(3, printk("leaving scsi_dispatch_cmnd()\n"));
601 	return rtn;
602 }
603 
604 /**
605  * scsi_req_abort_cmd -- Request command recovery for the specified command
606  * cmd: pointer to the SCSI command of interest
607  *
608  * This function requests that SCSI Core start recovery for the
609  * command by deleting the timer and adding the command to the eh
610  * queue.  It can be called by either LLDDs or SCSI Core.  LLDDs who
611  * implement their own error recovery MAY ignore the timeout event if
612  * they generated scsi_req_abort_cmd.
613  */
614 void scsi_req_abort_cmd(struct scsi_cmnd *cmd)
615 {
616 	if (!scsi_delete_timer(cmd))
617 		return;
618 	scsi_times_out(cmd);
619 }
620 EXPORT_SYMBOL(scsi_req_abort_cmd);
621 
622 /**
623  * scsi_done - Enqueue the finished SCSI command into the done queue.
624  * @cmd: The SCSI Command for which a low-level device driver (LLDD) gives
625  * ownership back to SCSI Core -- i.e. the LLDD has finished with it.
626  *
627  * This function is the mid-level's (SCSI Core) interrupt routine, which
628  * regains ownership of the SCSI command (de facto) from a LLDD, and enqueues
629  * the command to the done queue for further processing.
630  *
631  * This is the producer of the done queue who enqueues at the tail.
632  *
633  * This function is interrupt context safe.
634  */
635 static void scsi_done(struct scsi_cmnd *cmd)
636 {
637 	/*
638 	 * We don't have to worry about this one timing out any more.
639 	 * If we are unable to remove the timer, then the command
640 	 * has already timed out.  In which case, we have no choice but to
641 	 * let the timeout function run, as we have no idea where in fact
642 	 * that function could really be.  It might be on another processor,
643 	 * etc, etc.
644 	 */
645 	if (!scsi_delete_timer(cmd))
646 		return;
647 	__scsi_done(cmd);
648 }
649 
650 /* Private entry to scsi_done() to complete a command when the timer
651  * isn't running --- used by scsi_times_out */
652 void __scsi_done(struct scsi_cmnd *cmd)
653 {
654 	struct request *rq = cmd->request;
655 
656 	/*
657 	 * Set the serial numbers back to zero
658 	 */
659 	cmd->serial_number = 0;
660 
661 	atomic_inc(&cmd->device->iodone_cnt);
662 	if (cmd->result)
663 		atomic_inc(&cmd->device->ioerr_cnt);
664 
665 	BUG_ON(!rq);
666 
667 	/*
668 	 * The uptodate/nbytes values don't matter, as we allow partial
669 	 * completes and thus will check this in the softirq callback
670 	 */
671 	rq->completion_data = cmd;
672 	blk_complete_request(rq);
673 }
674 
675 /*
676  * Function:    scsi_retry_command
677  *
678  * Purpose:     Send a command back to the low level to be retried.
679  *
680  * Notes:       This command is always executed in the context of the
681  *              bottom half handler, or the error handler thread. Low
682  *              level drivers should not become re-entrant as a result of
683  *              this.
684  */
685 int scsi_retry_command(struct scsi_cmnd *cmd)
686 {
687         /*
688          * Zero the sense information from the last time we tried
689          * this command.
690          */
691 	memset(cmd->sense_buffer, 0, sizeof(cmd->sense_buffer));
692 
693 	return scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
694 }
695 
696 /*
697  * Function:    scsi_finish_command
698  *
699  * Purpose:     Pass command off to upper layer for finishing of I/O
700  *              request, waking processes that are waiting on results,
701  *              etc.
702  */
703 void scsi_finish_command(struct scsi_cmnd *cmd)
704 {
705 	struct scsi_device *sdev = cmd->device;
706 	struct Scsi_Host *shost = sdev->host;
707 
708 	scsi_device_unbusy(sdev);
709 
710         /*
711          * Clear the flags which say that the device/host is no longer
712          * capable of accepting new commands.  These are set in scsi_queue.c
713          * for both the queue full condition on a device, and for a
714          * host full condition on the host.
715 	 *
716 	 * XXX(hch): What about locking?
717          */
718         shost->host_blocked = 0;
719         sdev->device_blocked = 0;
720 
721 	/*
722 	 * If we have valid sense information, then some kind of recovery
723 	 * must have taken place.  Make a note of this.
724 	 */
725 	if (SCSI_SENSE_VALID(cmd))
726 		cmd->result |= (DRIVER_SENSE << 24);
727 
728 	SCSI_LOG_MLCOMPLETE(4, sdev_printk(KERN_INFO, sdev,
729 				"Notifying upper driver of completion "
730 				"(result %x)\n", cmd->result));
731 
732 	cmd->done(cmd);
733 }
734 EXPORT_SYMBOL(scsi_finish_command);
735 
736 /*
737  * Function:	scsi_adjust_queue_depth()
738  *
739  * Purpose:	Allow low level drivers to tell us to change the queue depth
740  * 		on a specific SCSI device
741  *
742  * Arguments:	sdev	- SCSI Device in question
743  * 		tagged	- Do we use tagged queueing (non-0) or do we treat
744  * 			  this device as an untagged device (0)
745  * 		tags	- Number of tags allowed if tagged queueing enabled,
746  * 			  or number of commands the low level driver can
747  * 			  queue up in non-tagged mode (as per cmd_per_lun).
748  *
749  * Returns:	Nothing
750  *
751  * Lock Status:	None held on entry
752  *
753  * Notes:	Low level drivers may call this at any time and we will do
754  * 		the right thing depending on whether or not the device is
755  * 		currently active and whether or not it even has the
756  * 		command blocks built yet.
757  */
758 void scsi_adjust_queue_depth(struct scsi_device *sdev, int tagged, int tags)
759 {
760 	unsigned long flags;
761 
762 	/*
763 	 * refuse to set tagged depth to an unworkable size
764 	 */
765 	if (tags <= 0)
766 		return;
767 
768 	spin_lock_irqsave(sdev->request_queue->queue_lock, flags);
769 
770 	/* Check to see if the queue is managed by the block layer
771 	 * if it is, and we fail to adjust the depth, exit */
772 	if (blk_queue_tagged(sdev->request_queue) &&
773 	    blk_queue_resize_tags(sdev->request_queue, tags) != 0)
774 		goto out;
775 
776 	sdev->queue_depth = tags;
777 	switch (tagged) {
778 		case MSG_ORDERED_TAG:
779 			sdev->ordered_tags = 1;
780 			sdev->simple_tags = 1;
781 			break;
782 		case MSG_SIMPLE_TAG:
783 			sdev->ordered_tags = 0;
784 			sdev->simple_tags = 1;
785 			break;
786 		default:
787 			sdev_printk(KERN_WARNING, sdev,
788 				    "scsi_adjust_queue_depth, bad queue type, "
789 				    "disabled\n");
790 		case 0:
791 			sdev->ordered_tags = sdev->simple_tags = 0;
792 			sdev->queue_depth = tags;
793 			break;
794 	}
795  out:
796 	spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags);
797 }
798 EXPORT_SYMBOL(scsi_adjust_queue_depth);
799 
800 /*
801  * Function:	scsi_track_queue_full()
802  *
803  * Purpose:	This function will track successive QUEUE_FULL events on a
804  * 		specific SCSI device to determine if and when there is a
805  * 		need to adjust the queue depth on the device.
806  *
807  * Arguments:	sdev	- SCSI Device in question
808  * 		depth	- Current number of outstanding SCSI commands on
809  * 			  this device, not counting the one returned as
810  * 			  QUEUE_FULL.
811  *
812  * Returns:	0 - No change needed
813  * 		>0 - Adjust queue depth to this new depth
814  * 		-1 - Drop back to untagged operation using host->cmd_per_lun
815  * 			as the untagged command depth
816  *
817  * Lock Status:	None held on entry
818  *
819  * Notes:	Low level drivers may call this at any time and we will do
820  * 		"The Right Thing."  We are interrupt context safe.
821  */
822 int scsi_track_queue_full(struct scsi_device *sdev, int depth)
823 {
824 	if ((jiffies >> 4) == sdev->last_queue_full_time)
825 		return 0;
826 
827 	sdev->last_queue_full_time = (jiffies >> 4);
828 	if (sdev->last_queue_full_depth != depth) {
829 		sdev->last_queue_full_count = 1;
830 		sdev->last_queue_full_depth = depth;
831 	} else {
832 		sdev->last_queue_full_count++;
833 	}
834 
835 	if (sdev->last_queue_full_count <= 10)
836 		return 0;
837 	if (sdev->last_queue_full_depth < 8) {
838 		/* Drop back to untagged */
839 		scsi_adjust_queue_depth(sdev, 0, sdev->host->cmd_per_lun);
840 		return -1;
841 	}
842 
843 	if (sdev->ordered_tags)
844 		scsi_adjust_queue_depth(sdev, MSG_ORDERED_TAG, depth);
845 	else
846 		scsi_adjust_queue_depth(sdev, MSG_SIMPLE_TAG, depth);
847 	return depth;
848 }
849 EXPORT_SYMBOL(scsi_track_queue_full);
850 
851 /**
852  * scsi_device_get  -  get an addition reference to a scsi_device
853  * @sdev:	device to get a reference to
854  *
855  * Gets a reference to the scsi_device and increments the use count
856  * of the underlying LLDD module.  You must hold host_lock of the
857  * parent Scsi_Host or already have a reference when calling this.
858  */
859 int scsi_device_get(struct scsi_device *sdev)
860 {
861 	if (sdev->sdev_state == SDEV_DEL)
862 		return -ENXIO;
863 	if (!get_device(&sdev->sdev_gendev))
864 		return -ENXIO;
865 	/* We can fail this if we're doing SCSI operations
866 	 * from module exit (like cache flush) */
867 	try_module_get(sdev->host->hostt->module);
868 
869 	return 0;
870 }
871 EXPORT_SYMBOL(scsi_device_get);
872 
873 /**
874  * scsi_device_put  -  release a reference to a scsi_device
875  * @sdev:	device to release a reference on.
876  *
877  * Release a reference to the scsi_device and decrements the use count
878  * of the underlying LLDD module.  The device is freed once the last
879  * user vanishes.
880  */
881 void scsi_device_put(struct scsi_device *sdev)
882 {
883 #ifdef CONFIG_MODULE_UNLOAD
884 	struct module *module = sdev->host->hostt->module;
885 
886 	/* The module refcount will be zero if scsi_device_get()
887 	 * was called from a module removal routine */
888 	if (module && module_refcount(module) != 0)
889 		module_put(module);
890 #endif
891 	put_device(&sdev->sdev_gendev);
892 }
893 EXPORT_SYMBOL(scsi_device_put);
894 
895 /* helper for shost_for_each_device, thus not documented */
896 struct scsi_device *__scsi_iterate_devices(struct Scsi_Host *shost,
897 					   struct scsi_device *prev)
898 {
899 	struct list_head *list = (prev ? &prev->siblings : &shost->__devices);
900 	struct scsi_device *next = NULL;
901 	unsigned long flags;
902 
903 	spin_lock_irqsave(shost->host_lock, flags);
904 	while (list->next != &shost->__devices) {
905 		next = list_entry(list->next, struct scsi_device, siblings);
906 		/* skip devices that we can't get a reference to */
907 		if (!scsi_device_get(next))
908 			break;
909 		next = NULL;
910 		list = list->next;
911 	}
912 	spin_unlock_irqrestore(shost->host_lock, flags);
913 
914 	if (prev)
915 		scsi_device_put(prev);
916 	return next;
917 }
918 EXPORT_SYMBOL(__scsi_iterate_devices);
919 
920 /**
921  * starget_for_each_device  -  helper to walk all devices of a target
922  * @starget:	target whose devices we want to iterate over.
923  *
924  * This traverses over each devices of @shost.  The devices have
925  * a reference that must be released by scsi_host_put when breaking
926  * out of the loop.
927  */
928 void starget_for_each_device(struct scsi_target *starget, void * data,
929 		     void (*fn)(struct scsi_device *, void *))
930 {
931 	struct Scsi_Host *shost = dev_to_shost(starget->dev.parent);
932 	struct scsi_device *sdev;
933 
934 	shost_for_each_device(sdev, shost) {
935 		if ((sdev->channel == starget->channel) &&
936 		    (sdev->id == starget->id))
937 			fn(sdev, data);
938 	}
939 }
940 EXPORT_SYMBOL(starget_for_each_device);
941 
942 /**
943  * __scsi_device_lookup_by_target - find a device given the target (UNLOCKED)
944  * @starget:	SCSI target pointer
945  * @lun:	SCSI Logical Unit Number
946  *
947  * Looks up the scsi_device with the specified @lun for a give
948  * @starget. The returned scsi_device does not have an additional
949  * reference.  You must hold the host's host_lock over this call and
950  * any access to the returned scsi_device.
951  *
952  * Note:  The only reason why drivers would want to use this is because
953  * they're need to access the device list in irq context.  Otherwise you
954  * really want to use scsi_device_lookup_by_target instead.
955  **/
956 struct scsi_device *__scsi_device_lookup_by_target(struct scsi_target *starget,
957 						   uint lun)
958 {
959 	struct scsi_device *sdev;
960 
961 	list_for_each_entry(sdev, &starget->devices, same_target_siblings) {
962 		if (sdev->lun ==lun)
963 			return sdev;
964 	}
965 
966 	return NULL;
967 }
968 EXPORT_SYMBOL(__scsi_device_lookup_by_target);
969 
970 /**
971  * scsi_device_lookup_by_target - find a device given the target
972  * @starget:	SCSI target pointer
973  * @lun:	SCSI Logical Unit Number
974  *
975  * Looks up the scsi_device with the specified @channel, @id, @lun for a
976  * give host.  The returned scsi_device has an additional reference that
977  * needs to be release with scsi_host_put once you're done with it.
978  **/
979 struct scsi_device *scsi_device_lookup_by_target(struct scsi_target *starget,
980 						 uint lun)
981 {
982 	struct scsi_device *sdev;
983 	struct Scsi_Host *shost = dev_to_shost(starget->dev.parent);
984 	unsigned long flags;
985 
986 	spin_lock_irqsave(shost->host_lock, flags);
987 	sdev = __scsi_device_lookup_by_target(starget, lun);
988 	if (sdev && scsi_device_get(sdev))
989 		sdev = NULL;
990 	spin_unlock_irqrestore(shost->host_lock, flags);
991 
992 	return sdev;
993 }
994 EXPORT_SYMBOL(scsi_device_lookup_by_target);
995 
996 /**
997  * scsi_device_lookup - find a device given the host (UNLOCKED)
998  * @shost:	SCSI host pointer
999  * @channel:	SCSI channel (zero if only one channel)
1000  * @pun:	SCSI target number (physical unit number)
1001  * @lun:	SCSI Logical Unit Number
1002  *
1003  * Looks up the scsi_device with the specified @channel, @id, @lun for a
1004  * give host. The returned scsi_device does not have an additional reference.
1005  * You must hold the host's host_lock over this call and any access to the
1006  * returned scsi_device.
1007  *
1008  * Note:  The only reason why drivers would want to use this is because
1009  * they're need to access the device list in irq context.  Otherwise you
1010  * really want to use scsi_device_lookup instead.
1011  **/
1012 struct scsi_device *__scsi_device_lookup(struct Scsi_Host *shost,
1013 		uint channel, uint id, uint lun)
1014 {
1015 	struct scsi_device *sdev;
1016 
1017 	list_for_each_entry(sdev, &shost->__devices, siblings) {
1018 		if (sdev->channel == channel && sdev->id == id &&
1019 				sdev->lun ==lun)
1020 			return sdev;
1021 	}
1022 
1023 	return NULL;
1024 }
1025 EXPORT_SYMBOL(__scsi_device_lookup);
1026 
1027 /**
1028  * scsi_device_lookup - find a device given the host
1029  * @shost:	SCSI host pointer
1030  * @channel:	SCSI channel (zero if only one channel)
1031  * @id:		SCSI target number (physical unit number)
1032  * @lun:	SCSI Logical Unit Number
1033  *
1034  * Looks up the scsi_device with the specified @channel, @id, @lun for a
1035  * give host.  The returned scsi_device has an additional reference that
1036  * needs to be release with scsi_host_put once you're done with it.
1037  **/
1038 struct scsi_device *scsi_device_lookup(struct Scsi_Host *shost,
1039 		uint channel, uint id, uint lun)
1040 {
1041 	struct scsi_device *sdev;
1042 	unsigned long flags;
1043 
1044 	spin_lock_irqsave(shost->host_lock, flags);
1045 	sdev = __scsi_device_lookup(shost, channel, id, lun);
1046 	if (sdev && scsi_device_get(sdev))
1047 		sdev = NULL;
1048 	spin_unlock_irqrestore(shost->host_lock, flags);
1049 
1050 	return sdev;
1051 }
1052 EXPORT_SYMBOL(scsi_device_lookup);
1053 
1054 /**
1055  * scsi_device_cancel - cancel outstanding IO to this device
1056  * @sdev:	Pointer to struct scsi_device
1057  * @recovery:	Boolean instructing function to recover device or not.
1058  *
1059  **/
1060 int scsi_device_cancel(struct scsi_device *sdev, int recovery)
1061 {
1062 	struct scsi_cmnd *scmd;
1063 	LIST_HEAD(active_list);
1064 	struct list_head *lh, *lh_sf;
1065 	unsigned long flags;
1066 
1067 	scsi_device_set_state(sdev, SDEV_CANCEL);
1068 
1069 	spin_lock_irqsave(&sdev->list_lock, flags);
1070 	list_for_each_entry(scmd, &sdev->cmd_list, list) {
1071 		if (scmd->request) {
1072 			/*
1073 			 * If we are unable to remove the timer, it means
1074 			 * that the command has already timed out or
1075 			 * finished.
1076 			 */
1077 			if (!scsi_delete_timer(scmd))
1078 				continue;
1079 			list_add_tail(&scmd->eh_entry, &active_list);
1080 		}
1081 	}
1082 	spin_unlock_irqrestore(&sdev->list_lock, flags);
1083 
1084 	if (!list_empty(&active_list)) {
1085 		list_for_each_safe(lh, lh_sf, &active_list) {
1086 			scmd = list_entry(lh, struct scsi_cmnd, eh_entry);
1087 			list_del_init(lh);
1088 			if (recovery &&
1089 			    !scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD)) {
1090 				scmd->result = (DID_ABORT << 16);
1091 				scsi_finish_command(scmd);
1092 			}
1093 		}
1094 	}
1095 
1096 	return 0;
1097 }
1098 EXPORT_SYMBOL(scsi_device_cancel);
1099 
1100 MODULE_DESCRIPTION("SCSI core");
1101 MODULE_LICENSE("GPL");
1102 
1103 module_param(scsi_logging_level, int, S_IRUGO|S_IWUSR);
1104 MODULE_PARM_DESC(scsi_logging_level, "a bit mask of logging levels");
1105 
1106 static int __init init_scsi(void)
1107 {
1108 	int error;
1109 
1110 	error = scsi_init_queue();
1111 	if (error)
1112 		return error;
1113 	error = scsi_init_procfs();
1114 	if (error)
1115 		goto cleanup_queue;
1116 	error = scsi_init_devinfo();
1117 	if (error)
1118 		goto cleanup_procfs;
1119 	error = scsi_init_hosts();
1120 	if (error)
1121 		goto cleanup_devlist;
1122 	error = scsi_init_sysctl();
1123 	if (error)
1124 		goto cleanup_hosts;
1125 	error = scsi_sysfs_register();
1126 	if (error)
1127 		goto cleanup_sysctl;
1128 
1129 	scsi_netlink_init();
1130 
1131 	printk(KERN_NOTICE "SCSI subsystem initialized\n");
1132 	return 0;
1133 
1134 cleanup_sysctl:
1135 	scsi_exit_sysctl();
1136 cleanup_hosts:
1137 	scsi_exit_hosts();
1138 cleanup_devlist:
1139 	scsi_exit_devinfo();
1140 cleanup_procfs:
1141 	scsi_exit_procfs();
1142 cleanup_queue:
1143 	scsi_exit_queue();
1144 	printk(KERN_ERR "SCSI subsystem failed to initialize, error = %d\n",
1145 	       -error);
1146 	return error;
1147 }
1148 
1149 static void __exit exit_scsi(void)
1150 {
1151 	scsi_netlink_exit();
1152 	scsi_sysfs_unregister();
1153 	scsi_exit_sysctl();
1154 	scsi_exit_hosts();
1155 	scsi_exit_devinfo();
1156 	scsi_exit_procfs();
1157 	scsi_exit_queue();
1158 }
1159 
1160 subsys_initcall(init_scsi);
1161 module_exit(exit_scsi);
1162