xref: /linux/drivers/s390/block/dasd.c (revision b233b28eac0cc37d07c2d007ea08c86c778c5af4)
1 /*
2  * File...........: linux/drivers/s390/block/dasd.c
3  * Author(s)......: Holger Smolinski <Holger.Smolinski@de.ibm.com>
4  *		    Horst Hummel <Horst.Hummel@de.ibm.com>
5  *		    Carsten Otte <Cotte@de.ibm.com>
6  *		    Martin Schwidefsky <schwidefsky@de.ibm.com>
7  * Bugreports.to..: <Linux390@de.ibm.com>
8  * (C) IBM Corporation, IBM Deutschland Entwicklung GmbH, 1999-2001
9  *
10  */
11 
12 #include <linux/kmod.h>
13 #include <linux/init.h>
14 #include <linux/interrupt.h>
15 #include <linux/ctype.h>
16 #include <linux/major.h>
17 #include <linux/slab.h>
18 #include <linux/buffer_head.h>
19 #include <linux/hdreg.h>
20 
21 #include <asm/ccwdev.h>
22 #include <asm/ebcdic.h>
23 #include <asm/idals.h>
24 #include <asm/todclk.h>
25 
26 /* This is ugly... */
27 #define PRINTK_HEADER "dasd:"
28 
29 #include "dasd_int.h"
30 /*
31  * SECTION: Constant definitions to be used within this file
32  */
33 #define DASD_CHANQ_MAX_SIZE 4
34 
35 /*
36  * SECTION: exported variables of dasd.c
37  */
38 debug_info_t *dasd_debug_area;
39 struct dasd_discipline *dasd_diag_discipline_pointer;
40 void dasd_int_handler(struct ccw_device *, unsigned long, struct irb *);
41 
42 MODULE_AUTHOR("Holger Smolinski <Holger.Smolinski@de.ibm.com>");
43 MODULE_DESCRIPTION("Linux on S/390 DASD device driver,"
44 		   " Copyright 2000 IBM Corporation");
45 MODULE_SUPPORTED_DEVICE("dasd");
46 MODULE_LICENSE("GPL");
47 
48 /*
49  * SECTION: prototypes for static functions of dasd.c
50  */
51 static int  dasd_alloc_queue(struct dasd_block *);
52 static void dasd_setup_queue(struct dasd_block *);
53 static void dasd_free_queue(struct dasd_block *);
54 static void dasd_flush_request_queue(struct dasd_block *);
55 static int dasd_flush_block_queue(struct dasd_block *);
56 static void dasd_device_tasklet(struct dasd_device *);
57 static void dasd_block_tasklet(struct dasd_block *);
58 static void do_kick_device(struct work_struct *);
59 static void dasd_return_cqr_cb(struct dasd_ccw_req *, void *);
60 static void dasd_device_timeout(unsigned long);
61 static void dasd_block_timeout(unsigned long);
62 
63 /*
64  * SECTION: Operations on the device structure.
65  */
66 static wait_queue_head_t dasd_init_waitq;
67 static wait_queue_head_t dasd_flush_wq;
68 static wait_queue_head_t generic_waitq;
69 
70 /*
71  * Allocate memory for a new device structure.
72  */
73 struct dasd_device *dasd_alloc_device(void)
74 {
75 	struct dasd_device *device;
76 
77 	device = kzalloc(sizeof(struct dasd_device), GFP_ATOMIC);
78 	if (!device)
79 		return ERR_PTR(-ENOMEM);
80 
81 	/* Get two pages for normal block device operations. */
82 	device->ccw_mem = (void *) __get_free_pages(GFP_ATOMIC | GFP_DMA, 1);
83 	if (!device->ccw_mem) {
84 		kfree(device);
85 		return ERR_PTR(-ENOMEM);
86 	}
87 	/* Get one page for error recovery. */
88 	device->erp_mem = (void *) get_zeroed_page(GFP_ATOMIC | GFP_DMA);
89 	if (!device->erp_mem) {
90 		free_pages((unsigned long) device->ccw_mem, 1);
91 		kfree(device);
92 		return ERR_PTR(-ENOMEM);
93 	}
94 
95 	dasd_init_chunklist(&device->ccw_chunks, device->ccw_mem, PAGE_SIZE*2);
96 	dasd_init_chunklist(&device->erp_chunks, device->erp_mem, PAGE_SIZE);
97 	spin_lock_init(&device->mem_lock);
98 	atomic_set(&device->tasklet_scheduled, 0);
99 	tasklet_init(&device->tasklet,
100 		     (void (*)(unsigned long)) dasd_device_tasklet,
101 		     (unsigned long) device);
102 	INIT_LIST_HEAD(&device->ccw_queue);
103 	init_timer(&device->timer);
104 	device->timer.function = dasd_device_timeout;
105 	device->timer.data = (unsigned long) device;
106 	INIT_WORK(&device->kick_work, do_kick_device);
107 	device->state = DASD_STATE_NEW;
108 	device->target = DASD_STATE_NEW;
109 
110 	return device;
111 }
112 
113 /*
114  * Free memory of a device structure.
115  */
116 void dasd_free_device(struct dasd_device *device)
117 {
118 	kfree(device->private);
119 	free_page((unsigned long) device->erp_mem);
120 	free_pages((unsigned long) device->ccw_mem, 1);
121 	kfree(device);
122 }
123 
124 /*
125  * Allocate memory for a new device structure.
126  */
127 struct dasd_block *dasd_alloc_block(void)
128 {
129 	struct dasd_block *block;
130 
131 	block = kzalloc(sizeof(*block), GFP_ATOMIC);
132 	if (!block)
133 		return ERR_PTR(-ENOMEM);
134 	/* open_count = 0 means device online but not in use */
135 	atomic_set(&block->open_count, -1);
136 
137 	spin_lock_init(&block->request_queue_lock);
138 	atomic_set(&block->tasklet_scheduled, 0);
139 	tasklet_init(&block->tasklet,
140 		     (void (*)(unsigned long)) dasd_block_tasklet,
141 		     (unsigned long) block);
142 	INIT_LIST_HEAD(&block->ccw_queue);
143 	spin_lock_init(&block->queue_lock);
144 	init_timer(&block->timer);
145 	block->timer.function = dasd_block_timeout;
146 	block->timer.data = (unsigned long) block;
147 
148 	return block;
149 }
150 
151 /*
152  * Free memory of a device structure.
153  */
154 void dasd_free_block(struct dasd_block *block)
155 {
156 	kfree(block);
157 }
158 
159 /*
160  * Make a new device known to the system.
161  */
162 static int dasd_state_new_to_known(struct dasd_device *device)
163 {
164 	int rc;
165 
166 	/*
167 	 * As long as the device is not in state DASD_STATE_NEW we want to
168 	 * keep the reference count > 0.
169 	 */
170 	dasd_get_device(device);
171 
172 	if (device->block) {
173 		rc = dasd_alloc_queue(device->block);
174 		if (rc) {
175 			dasd_put_device(device);
176 			return rc;
177 		}
178 	}
179 	device->state = DASD_STATE_KNOWN;
180 	return 0;
181 }
182 
183 /*
184  * Let the system forget about a device.
185  */
186 static int dasd_state_known_to_new(struct dasd_device *device)
187 {
188 	/* Disable extended error reporting for this device. */
189 	dasd_eer_disable(device);
190 	/* Forget the discipline information. */
191 	if (device->discipline) {
192 		if (device->discipline->uncheck_device)
193 			device->discipline->uncheck_device(device);
194 		module_put(device->discipline->owner);
195 	}
196 	device->discipline = NULL;
197 	if (device->base_discipline)
198 		module_put(device->base_discipline->owner);
199 	device->base_discipline = NULL;
200 	device->state = DASD_STATE_NEW;
201 
202 	if (device->block)
203 		dasd_free_queue(device->block);
204 
205 	/* Give up reference we took in dasd_state_new_to_known. */
206 	dasd_put_device(device);
207 	return 0;
208 }
209 
210 /*
211  * Request the irq line for the device.
212  */
213 static int dasd_state_known_to_basic(struct dasd_device *device)
214 {
215 	int rc;
216 
217 	/* Allocate and register gendisk structure. */
218 	if (device->block) {
219 		rc = dasd_gendisk_alloc(device->block);
220 		if (rc)
221 			return rc;
222 	}
223 	/* register 'device' debug area, used for all DBF_DEV_XXX calls */
224 	device->debug_area = debug_register(dev_name(&device->cdev->dev), 1, 1,
225 					    8 * sizeof(long));
226 	debug_register_view(device->debug_area, &debug_sprintf_view);
227 	debug_set_level(device->debug_area, DBF_WARNING);
228 	DBF_DEV_EVENT(DBF_EMERG, device, "%s", "debug area created");
229 
230 	device->state = DASD_STATE_BASIC;
231 	return 0;
232 }
233 
234 /*
235  * Release the irq line for the device. Terminate any running i/o.
236  */
237 static int dasd_state_basic_to_known(struct dasd_device *device)
238 {
239 	int rc;
240 	if (device->block) {
241 		dasd_gendisk_free(device->block);
242 		dasd_block_clear_timer(device->block);
243 	}
244 	rc = dasd_flush_device_queue(device);
245 	if (rc)
246 		return rc;
247 	dasd_device_clear_timer(device);
248 
249 	DBF_DEV_EVENT(DBF_EMERG, device, "%p debug area deleted", device);
250 	if (device->debug_area != NULL) {
251 		debug_unregister(device->debug_area);
252 		device->debug_area = NULL;
253 	}
254 	device->state = DASD_STATE_KNOWN;
255 	return 0;
256 }
257 
258 /*
259  * Do the initial analysis. The do_analysis function may return
260  * -EAGAIN in which case the device keeps the state DASD_STATE_BASIC
261  * until the discipline decides to continue the startup sequence
262  * by calling the function dasd_change_state. The eckd disciplines
263  * uses this to start a ccw that detects the format. The completion
264  * interrupt for this detection ccw uses the kernel event daemon to
265  * trigger the call to dasd_change_state. All this is done in the
266  * discipline code, see dasd_eckd.c.
267  * After the analysis ccw is done (do_analysis returned 0) the block
268  * device is setup.
269  * In case the analysis returns an error, the device setup is stopped
270  * (a fake disk was already added to allow formatting).
271  */
272 static int dasd_state_basic_to_ready(struct dasd_device *device)
273 {
274 	int rc;
275 	struct dasd_block *block;
276 
277 	rc = 0;
278 	block = device->block;
279 	/* make disk known with correct capacity */
280 	if (block) {
281 		if (block->base->discipline->do_analysis != NULL)
282 			rc = block->base->discipline->do_analysis(block);
283 		if (rc) {
284 			if (rc != -EAGAIN)
285 				device->state = DASD_STATE_UNFMT;
286 			return rc;
287 		}
288 		dasd_setup_queue(block);
289 		set_capacity(block->gdp,
290 			     block->blocks << block->s2b_shift);
291 		device->state = DASD_STATE_READY;
292 		rc = dasd_scan_partitions(block);
293 		if (rc)
294 			device->state = DASD_STATE_BASIC;
295 	} else {
296 		device->state = DASD_STATE_READY;
297 	}
298 	return rc;
299 }
300 
301 /*
302  * Remove device from block device layer. Destroy dirty buffers.
303  * Forget format information. Check if the target level is basic
304  * and if it is create fake disk for formatting.
305  */
306 static int dasd_state_ready_to_basic(struct dasd_device *device)
307 {
308 	int rc;
309 
310 	device->state = DASD_STATE_BASIC;
311 	if (device->block) {
312 		struct dasd_block *block = device->block;
313 		rc = dasd_flush_block_queue(block);
314 		if (rc) {
315 			device->state = DASD_STATE_READY;
316 			return rc;
317 		}
318 		dasd_destroy_partitions(block);
319 		dasd_flush_request_queue(block);
320 		block->blocks = 0;
321 		block->bp_block = 0;
322 		block->s2b_shift = 0;
323 	}
324 	return 0;
325 }
326 
327 /*
328  * Back to basic.
329  */
330 static int dasd_state_unfmt_to_basic(struct dasd_device *device)
331 {
332 	device->state = DASD_STATE_BASIC;
333 	return 0;
334 }
335 
336 /*
337  * Make the device online and schedule the bottom half to start
338  * the requeueing of requests from the linux request queue to the
339  * ccw queue.
340  */
341 static int
342 dasd_state_ready_to_online(struct dasd_device * device)
343 {
344 	int rc;
345 	struct gendisk *disk;
346 	struct disk_part_iter piter;
347 	struct hd_struct *part;
348 
349 	if (device->discipline->ready_to_online) {
350 		rc = device->discipline->ready_to_online(device);
351 		if (rc)
352 			return rc;
353 	}
354 	device->state = DASD_STATE_ONLINE;
355 	if (device->block) {
356 		dasd_schedule_block_bh(device->block);
357 		disk = device->block->bdev->bd_disk;
358 		disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
359 		while ((part = disk_part_iter_next(&piter)))
360 			kobject_uevent(&part_to_dev(part)->kobj, KOBJ_CHANGE);
361 		disk_part_iter_exit(&piter);
362 	}
363 	return 0;
364 }
365 
366 /*
367  * Stop the requeueing of requests again.
368  */
369 static int dasd_state_online_to_ready(struct dasd_device *device)
370 {
371 	int rc;
372 	struct gendisk *disk;
373 	struct disk_part_iter piter;
374 	struct hd_struct *part;
375 
376 	if (device->discipline->online_to_ready) {
377 		rc = device->discipline->online_to_ready(device);
378 		if (rc)
379 			return rc;
380 	}
381 	device->state = DASD_STATE_READY;
382 	if (device->block) {
383 		disk = device->block->bdev->bd_disk;
384 		disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
385 		while ((part = disk_part_iter_next(&piter)))
386 			kobject_uevent(&part_to_dev(part)->kobj, KOBJ_CHANGE);
387 		disk_part_iter_exit(&piter);
388 	}
389 	return 0;
390 }
391 
392 /*
393  * Device startup state changes.
394  */
395 static int dasd_increase_state(struct dasd_device *device)
396 {
397 	int rc;
398 
399 	rc = 0;
400 	if (device->state == DASD_STATE_NEW &&
401 	    device->target >= DASD_STATE_KNOWN)
402 		rc = dasd_state_new_to_known(device);
403 
404 	if (!rc &&
405 	    device->state == DASD_STATE_KNOWN &&
406 	    device->target >= DASD_STATE_BASIC)
407 		rc = dasd_state_known_to_basic(device);
408 
409 	if (!rc &&
410 	    device->state == DASD_STATE_BASIC &&
411 	    device->target >= DASD_STATE_READY)
412 		rc = dasd_state_basic_to_ready(device);
413 
414 	if (!rc &&
415 	    device->state == DASD_STATE_UNFMT &&
416 	    device->target > DASD_STATE_UNFMT)
417 		rc = -EPERM;
418 
419 	if (!rc &&
420 	    device->state == DASD_STATE_READY &&
421 	    device->target >= DASD_STATE_ONLINE)
422 		rc = dasd_state_ready_to_online(device);
423 
424 	return rc;
425 }
426 
427 /*
428  * Device shutdown state changes.
429  */
430 static int dasd_decrease_state(struct dasd_device *device)
431 {
432 	int rc;
433 
434 	rc = 0;
435 	if (device->state == DASD_STATE_ONLINE &&
436 	    device->target <= DASD_STATE_READY)
437 		rc = dasd_state_online_to_ready(device);
438 
439 	if (!rc &&
440 	    device->state == DASD_STATE_READY &&
441 	    device->target <= DASD_STATE_BASIC)
442 		rc = dasd_state_ready_to_basic(device);
443 
444 	if (!rc &&
445 	    device->state == DASD_STATE_UNFMT &&
446 	    device->target <= DASD_STATE_BASIC)
447 		rc = dasd_state_unfmt_to_basic(device);
448 
449 	if (!rc &&
450 	    device->state == DASD_STATE_BASIC &&
451 	    device->target <= DASD_STATE_KNOWN)
452 		rc = dasd_state_basic_to_known(device);
453 
454 	if (!rc &&
455 	    device->state == DASD_STATE_KNOWN &&
456 	    device->target <= DASD_STATE_NEW)
457 		rc = dasd_state_known_to_new(device);
458 
459 	return rc;
460 }
461 
462 /*
463  * This is the main startup/shutdown routine.
464  */
465 static void dasd_change_state(struct dasd_device *device)
466 {
467         int rc;
468 
469 	if (device->state == device->target)
470 		/* Already where we want to go today... */
471 		return;
472 	if (device->state < device->target)
473 		rc = dasd_increase_state(device);
474 	else
475 		rc = dasd_decrease_state(device);
476         if (rc && rc != -EAGAIN)
477                 device->target = device->state;
478 
479 	if (device->state == device->target)
480 		wake_up(&dasd_init_waitq);
481 
482 	/* let user-space know that the device status changed */
483 	kobject_uevent(&device->cdev->dev.kobj, KOBJ_CHANGE);
484 }
485 
486 /*
487  * Kick starter for devices that did not complete the startup/shutdown
488  * procedure or were sleeping because of a pending state.
489  * dasd_kick_device will schedule a call do do_kick_device to the kernel
490  * event daemon.
491  */
492 static void do_kick_device(struct work_struct *work)
493 {
494 	struct dasd_device *device = container_of(work, struct dasd_device, kick_work);
495 	dasd_change_state(device);
496 	dasd_schedule_device_bh(device);
497 	dasd_put_device(device);
498 }
499 
500 void dasd_kick_device(struct dasd_device *device)
501 {
502 	dasd_get_device(device);
503 	/* queue call to dasd_kick_device to the kernel event daemon. */
504 	schedule_work(&device->kick_work);
505 }
506 
507 /*
508  * Set the target state for a device and starts the state change.
509  */
510 void dasd_set_target_state(struct dasd_device *device, int target)
511 {
512 	/* If we are in probeonly mode stop at DASD_STATE_READY. */
513 	if (dasd_probeonly && target > DASD_STATE_READY)
514 		target = DASD_STATE_READY;
515 	if (device->target != target) {
516                 if (device->state == target)
517 			wake_up(&dasd_init_waitq);
518 		device->target = target;
519 	}
520 	if (device->state != device->target)
521 		dasd_change_state(device);
522 }
523 
524 /*
525  * Enable devices with device numbers in [from..to].
526  */
527 static inline int _wait_for_device(struct dasd_device *device)
528 {
529 	return (device->state == device->target);
530 }
531 
532 void dasd_enable_device(struct dasd_device *device)
533 {
534 	dasd_set_target_state(device, DASD_STATE_ONLINE);
535 	if (device->state <= DASD_STATE_KNOWN)
536 		/* No discipline for device found. */
537 		dasd_set_target_state(device, DASD_STATE_NEW);
538 	/* Now wait for the devices to come up. */
539 	wait_event(dasd_init_waitq, _wait_for_device(device));
540 }
541 
542 /*
543  * SECTION: device operation (interrupt handler, start i/o, term i/o ...)
544  */
545 #ifdef CONFIG_DASD_PROFILE
546 
547 struct dasd_profile_info_t dasd_global_profile;
548 unsigned int dasd_profile_level = DASD_PROFILE_OFF;
549 
550 /*
551  * Increments counter in global and local profiling structures.
552  */
553 #define dasd_profile_counter(value, counter, block) \
554 { \
555 	int index; \
556 	for (index = 0; index < 31 && value >> (2+index); index++); \
557 	dasd_global_profile.counter[index]++; \
558 	block->profile.counter[index]++; \
559 }
560 
561 /*
562  * Add profiling information for cqr before execution.
563  */
564 static void dasd_profile_start(struct dasd_block *block,
565 			       struct dasd_ccw_req *cqr,
566 			       struct request *req)
567 {
568 	struct list_head *l;
569 	unsigned int counter;
570 
571 	if (dasd_profile_level != DASD_PROFILE_ON)
572 		return;
573 
574 	/* count the length of the chanq for statistics */
575 	counter = 0;
576 	list_for_each(l, &block->ccw_queue)
577 		if (++counter >= 31)
578 			break;
579 	dasd_global_profile.dasd_io_nr_req[counter]++;
580 	block->profile.dasd_io_nr_req[counter]++;
581 }
582 
583 /*
584  * Add profiling information for cqr after execution.
585  */
586 static void dasd_profile_end(struct dasd_block *block,
587 			     struct dasd_ccw_req *cqr,
588 			     struct request *req)
589 {
590 	long strtime, irqtime, endtime, tottime;	/* in microseconds */
591 	long tottimeps, sectors;
592 
593 	if (dasd_profile_level != DASD_PROFILE_ON)
594 		return;
595 
596 	sectors = req->nr_sectors;
597 	if (!cqr->buildclk || !cqr->startclk ||
598 	    !cqr->stopclk || !cqr->endclk ||
599 	    !sectors)
600 		return;
601 
602 	strtime = ((cqr->startclk - cqr->buildclk) >> 12);
603 	irqtime = ((cqr->stopclk - cqr->startclk) >> 12);
604 	endtime = ((cqr->endclk - cqr->stopclk) >> 12);
605 	tottime = ((cqr->endclk - cqr->buildclk) >> 12);
606 	tottimeps = tottime / sectors;
607 
608 	if (!dasd_global_profile.dasd_io_reqs)
609 		memset(&dasd_global_profile, 0,
610 		       sizeof(struct dasd_profile_info_t));
611 	dasd_global_profile.dasd_io_reqs++;
612 	dasd_global_profile.dasd_io_sects += sectors;
613 
614 	if (!block->profile.dasd_io_reqs)
615 		memset(&block->profile, 0,
616 		       sizeof(struct dasd_profile_info_t));
617 	block->profile.dasd_io_reqs++;
618 	block->profile.dasd_io_sects += sectors;
619 
620 	dasd_profile_counter(sectors, dasd_io_secs, block);
621 	dasd_profile_counter(tottime, dasd_io_times, block);
622 	dasd_profile_counter(tottimeps, dasd_io_timps, block);
623 	dasd_profile_counter(strtime, dasd_io_time1, block);
624 	dasd_profile_counter(irqtime, dasd_io_time2, block);
625 	dasd_profile_counter(irqtime / sectors, dasd_io_time2ps, block);
626 	dasd_profile_counter(endtime, dasd_io_time3, block);
627 }
628 #else
629 #define dasd_profile_start(block, cqr, req) do {} while (0)
630 #define dasd_profile_end(block, cqr, req) do {} while (0)
631 #endif				/* CONFIG_DASD_PROFILE */
632 
633 /*
634  * Allocate memory for a channel program with 'cplength' channel
635  * command words and 'datasize' additional space. There are two
636  * variantes: 1) dasd_kmalloc_request uses kmalloc to get the needed
637  * memory and 2) dasd_smalloc_request uses the static ccw memory
638  * that gets allocated for each device.
639  */
640 struct dasd_ccw_req *dasd_kmalloc_request(char *magic, int cplength,
641 					  int datasize,
642 					  struct dasd_device *device)
643 {
644 	struct dasd_ccw_req *cqr;
645 
646 	/* Sanity checks */
647 	BUG_ON( magic == NULL || datasize > PAGE_SIZE ||
648 	     (cplength*sizeof(struct ccw1)) > PAGE_SIZE);
649 
650 	cqr = kzalloc(sizeof(struct dasd_ccw_req), GFP_ATOMIC);
651 	if (cqr == NULL)
652 		return ERR_PTR(-ENOMEM);
653 	cqr->cpaddr = NULL;
654 	if (cplength > 0) {
655 		cqr->cpaddr = kcalloc(cplength, sizeof(struct ccw1),
656 				      GFP_ATOMIC | GFP_DMA);
657 		if (cqr->cpaddr == NULL) {
658 			kfree(cqr);
659 			return ERR_PTR(-ENOMEM);
660 		}
661 	}
662 	cqr->data = NULL;
663 	if (datasize > 0) {
664 		cqr->data = kzalloc(datasize, GFP_ATOMIC | GFP_DMA);
665 		if (cqr->data == NULL) {
666 			kfree(cqr->cpaddr);
667 			kfree(cqr);
668 			return ERR_PTR(-ENOMEM);
669 		}
670 	}
671 	strncpy((char *) &cqr->magic, magic, 4);
672 	ASCEBC((char *) &cqr->magic, 4);
673 	set_bit(DASD_CQR_FLAGS_USE_ERP, &cqr->flags);
674 	dasd_get_device(device);
675 	return cqr;
676 }
677 
678 struct dasd_ccw_req *dasd_smalloc_request(char *magic, int cplength,
679 					  int datasize,
680 					  struct dasd_device *device)
681 {
682 	unsigned long flags;
683 	struct dasd_ccw_req *cqr;
684 	char *data;
685 	int size;
686 
687 	/* Sanity checks */
688 	BUG_ON( magic == NULL || datasize > PAGE_SIZE ||
689 	     (cplength*sizeof(struct ccw1)) > PAGE_SIZE);
690 
691 	size = (sizeof(struct dasd_ccw_req) + 7L) & -8L;
692 	if (cplength > 0)
693 		size += cplength * sizeof(struct ccw1);
694 	if (datasize > 0)
695 		size += datasize;
696 	spin_lock_irqsave(&device->mem_lock, flags);
697 	cqr = (struct dasd_ccw_req *)
698 		dasd_alloc_chunk(&device->ccw_chunks, size);
699 	spin_unlock_irqrestore(&device->mem_lock, flags);
700 	if (cqr == NULL)
701 		return ERR_PTR(-ENOMEM);
702 	memset(cqr, 0, sizeof(struct dasd_ccw_req));
703 	data = (char *) cqr + ((sizeof(struct dasd_ccw_req) + 7L) & -8L);
704 	cqr->cpaddr = NULL;
705 	if (cplength > 0) {
706 		cqr->cpaddr = (struct ccw1 *) data;
707 		data += cplength*sizeof(struct ccw1);
708 		memset(cqr->cpaddr, 0, cplength*sizeof(struct ccw1));
709 	}
710 	cqr->data = NULL;
711 	if (datasize > 0) {
712 		cqr->data = data;
713  		memset(cqr->data, 0, datasize);
714 	}
715 	strncpy((char *) &cqr->magic, magic, 4);
716 	ASCEBC((char *) &cqr->magic, 4);
717 	set_bit(DASD_CQR_FLAGS_USE_ERP, &cqr->flags);
718 	dasd_get_device(device);
719 	return cqr;
720 }
721 
722 /*
723  * Free memory of a channel program. This function needs to free all the
724  * idal lists that might have been created by dasd_set_cda and the
725  * struct dasd_ccw_req itself.
726  */
727 void dasd_kfree_request(struct dasd_ccw_req *cqr, struct dasd_device *device)
728 {
729 #ifdef CONFIG_64BIT
730 	struct ccw1 *ccw;
731 
732 	/* Clear any idals used for the request. */
733 	ccw = cqr->cpaddr;
734 	do {
735 		clear_normalized_cda(ccw);
736 	} while (ccw++->flags & (CCW_FLAG_CC | CCW_FLAG_DC));
737 #endif
738 	kfree(cqr->cpaddr);
739 	kfree(cqr->data);
740 	kfree(cqr);
741 	dasd_put_device(device);
742 }
743 
744 void dasd_sfree_request(struct dasd_ccw_req *cqr, struct dasd_device *device)
745 {
746 	unsigned long flags;
747 
748 	spin_lock_irqsave(&device->mem_lock, flags);
749 	dasd_free_chunk(&device->ccw_chunks, cqr);
750 	spin_unlock_irqrestore(&device->mem_lock, flags);
751 	dasd_put_device(device);
752 }
753 
754 /*
755  * Check discipline magic in cqr.
756  */
757 static inline int dasd_check_cqr(struct dasd_ccw_req *cqr)
758 {
759 	struct dasd_device *device;
760 
761 	if (cqr == NULL)
762 		return -EINVAL;
763 	device = cqr->startdev;
764 	if (strncmp((char *) &cqr->magic, device->discipline->ebcname, 4)) {
765 		DEV_MESSAGE(KERN_WARNING, device,
766 			    " dasd_ccw_req 0x%08x magic doesn't match"
767 			    " discipline 0x%08x",
768 			    cqr->magic,
769 			    *(unsigned int *) device->discipline->name);
770 		return -EINVAL;
771 	}
772 	return 0;
773 }
774 
775 /*
776  * Terminate the current i/o and set the request to clear_pending.
777  * Timer keeps device runnig.
778  * ccw_device_clear can fail if the i/o subsystem
779  * is in a bad mood.
780  */
781 int dasd_term_IO(struct dasd_ccw_req *cqr)
782 {
783 	struct dasd_device *device;
784 	int retries, rc;
785 
786 	/* Check the cqr */
787 	rc = dasd_check_cqr(cqr);
788 	if (rc)
789 		return rc;
790 	retries = 0;
791 	device = (struct dasd_device *) cqr->startdev;
792 	while ((retries < 5) && (cqr->status == DASD_CQR_IN_IO)) {
793 		rc = ccw_device_clear(device->cdev, (long) cqr);
794 		switch (rc) {
795 		case 0:	/* termination successful */
796 			cqr->retries--;
797 			cqr->status = DASD_CQR_CLEAR_PENDING;
798 			cqr->stopclk = get_clock();
799 			cqr->starttime = 0;
800 			DBF_DEV_EVENT(DBF_DEBUG, device,
801 				      "terminate cqr %p successful",
802 				      cqr);
803 			break;
804 		case -ENODEV:
805 			DBF_DEV_EVENT(DBF_ERR, device, "%s",
806 				      "device gone, retry");
807 			break;
808 		case -EIO:
809 			DBF_DEV_EVENT(DBF_ERR, device, "%s",
810 				      "I/O error, retry");
811 			break;
812 		case -EINVAL:
813 		case -EBUSY:
814 			DBF_DEV_EVENT(DBF_ERR, device, "%s",
815 				      "device busy, retry later");
816 			break;
817 		default:
818 			DEV_MESSAGE(KERN_ERR, device,
819 				    "line %d unknown RC=%d, please "
820 				    "report to linux390@de.ibm.com",
821 				    __LINE__, rc);
822 			BUG();
823 			break;
824 		}
825 		retries++;
826 	}
827 	dasd_schedule_device_bh(device);
828 	return rc;
829 }
830 
831 /*
832  * Start the i/o. This start_IO can fail if the channel is really busy.
833  * In that case set up a timer to start the request later.
834  */
835 int dasd_start_IO(struct dasd_ccw_req *cqr)
836 {
837 	struct dasd_device *device;
838 	int rc;
839 
840 	/* Check the cqr */
841 	rc = dasd_check_cqr(cqr);
842 	if (rc)
843 		return rc;
844 	device = (struct dasd_device *) cqr->startdev;
845 	if (cqr->retries < 0) {
846 		DEV_MESSAGE(KERN_DEBUG, device,
847 			    "start_IO: request %p (%02x/%i) - no retry left.",
848 			    cqr, cqr->status, cqr->retries);
849 		cqr->status = DASD_CQR_ERROR;
850 		return -EIO;
851 	}
852 	cqr->startclk = get_clock();
853 	cqr->starttime = jiffies;
854 	cqr->retries--;
855 	rc = ccw_device_start(device->cdev, cqr->cpaddr, (long) cqr,
856 			      cqr->lpm, 0);
857 	switch (rc) {
858 	case 0:
859 		cqr->status = DASD_CQR_IN_IO;
860 		DBF_DEV_EVENT(DBF_DEBUG, device,
861 			      "start_IO: request %p started successful",
862 			      cqr);
863 		break;
864 	case -EBUSY:
865 		DBF_DEV_EVENT(DBF_ERR, device, "%s",
866 			      "start_IO: device busy, retry later");
867 		break;
868 	case -ETIMEDOUT:
869 		DBF_DEV_EVENT(DBF_ERR, device, "%s",
870 			      "start_IO: request timeout, retry later");
871 		break;
872 	case -EACCES:
873 		/* -EACCES indicates that the request used only a
874 		 * subset of the available pathes and all these
875 		 * pathes are gone.
876 		 * Do a retry with all available pathes.
877 		 */
878 		cqr->lpm = LPM_ANYPATH;
879 		DBF_DEV_EVENT(DBF_ERR, device, "%s",
880 			      "start_IO: selected pathes gone,"
881 			      " retry on all pathes");
882 		break;
883 	case -ENODEV:
884 	case -EIO:
885 		DBF_DEV_EVENT(DBF_ERR, device, "%s",
886 			      "start_IO: device gone, retry");
887 		break;
888 	default:
889 		DEV_MESSAGE(KERN_ERR, device,
890 			    "line %d unknown RC=%d, please report"
891 			    " to linux390@de.ibm.com", __LINE__, rc);
892 		BUG();
893 		break;
894 	}
895 	return rc;
896 }
897 
898 /*
899  * Timeout function for dasd devices. This is used for different purposes
900  *  1) missing interrupt handler for normal operation
901  *  2) delayed start of request where start_IO failed with -EBUSY
902  *  3) timeout for missing state change interrupts
903  * The head of the ccw queue will have status DASD_CQR_IN_IO for 1),
904  * DASD_CQR_QUEUED for 2) and 3).
905  */
906 static void dasd_device_timeout(unsigned long ptr)
907 {
908 	unsigned long flags;
909 	struct dasd_device *device;
910 
911 	device = (struct dasd_device *) ptr;
912 	spin_lock_irqsave(get_ccwdev_lock(device->cdev), flags);
913 	/* re-activate request queue */
914         device->stopped &= ~DASD_STOPPED_PENDING;
915 	spin_unlock_irqrestore(get_ccwdev_lock(device->cdev), flags);
916 	dasd_schedule_device_bh(device);
917 }
918 
919 /*
920  * Setup timeout for a device in jiffies.
921  */
922 void dasd_device_set_timer(struct dasd_device *device, int expires)
923 {
924 	if (expires == 0)
925 		del_timer(&device->timer);
926 	else
927 		mod_timer(&device->timer, jiffies + expires);
928 }
929 
930 /*
931  * Clear timeout for a device.
932  */
933 void dasd_device_clear_timer(struct dasd_device *device)
934 {
935 	del_timer(&device->timer);
936 }
937 
938 static void dasd_handle_killed_request(struct ccw_device *cdev,
939 				       unsigned long intparm)
940 {
941 	struct dasd_ccw_req *cqr;
942 	struct dasd_device *device;
943 
944 	if (!intparm)
945 		return;
946 	cqr = (struct dasd_ccw_req *) intparm;
947 	if (cqr->status != DASD_CQR_IN_IO) {
948 		MESSAGE(KERN_DEBUG,
949 			"invalid status in handle_killed_request: "
950 			"bus_id %s, status %02x",
951 			dev_name(&cdev->dev), cqr->status);
952 		return;
953 	}
954 
955 	device = (struct dasd_device *) cqr->startdev;
956 	if (device == NULL ||
957 	    device != dasd_device_from_cdev_locked(cdev) ||
958 	    strncmp(device->discipline->ebcname, (char *) &cqr->magic, 4)) {
959 		MESSAGE(KERN_DEBUG, "invalid device in request: bus_id %s",
960 			dev_name(&cdev->dev));
961 		return;
962 	}
963 
964 	/* Schedule request to be retried. */
965 	cqr->status = DASD_CQR_QUEUED;
966 
967 	dasd_device_clear_timer(device);
968 	dasd_schedule_device_bh(device);
969 	dasd_put_device(device);
970 }
971 
972 void dasd_generic_handle_state_change(struct dasd_device *device)
973 {
974 	/* First of all start sense subsystem status request. */
975 	dasd_eer_snss(device);
976 
977 	device->stopped &= ~DASD_STOPPED_PENDING;
978 	dasd_schedule_device_bh(device);
979 	if (device->block)
980 		dasd_schedule_block_bh(device->block);
981 }
982 
983 /*
984  * Interrupt handler for "normal" ssch-io based dasd devices.
985  */
986 void dasd_int_handler(struct ccw_device *cdev, unsigned long intparm,
987 		      struct irb *irb)
988 {
989 	struct dasd_ccw_req *cqr, *next;
990 	struct dasd_device *device;
991 	unsigned long long now;
992 	int expires;
993 
994 	if (IS_ERR(irb)) {
995 		switch (PTR_ERR(irb)) {
996 		case -EIO:
997 			break;
998 		case -ETIMEDOUT:
999 			printk(KERN_WARNING"%s(%s): request timed out\n",
1000 			       __func__, dev_name(&cdev->dev));
1001 			break;
1002 		default:
1003 			printk(KERN_WARNING"%s(%s): unknown error %ld\n",
1004 			       __func__, dev_name(&cdev->dev), PTR_ERR(irb));
1005 		}
1006 		dasd_handle_killed_request(cdev, intparm);
1007 		return;
1008 	}
1009 
1010 	now = get_clock();
1011 
1012 	DBF_EVENT(DBF_ERR, "Interrupt: bus_id %s CS/DS %04x ip %08x",
1013 		  dev_name(&cdev->dev), ((irb->scsw.cmd.cstat << 8) |
1014 		  irb->scsw.cmd.dstat), (unsigned int) intparm);
1015 
1016 	/* check for unsolicited interrupts */
1017 	cqr = (struct dasd_ccw_req *) intparm;
1018 	if (!cqr || ((irb->scsw.cmd.cc == 1) &&
1019 		     (irb->scsw.cmd.fctl & SCSW_FCTL_START_FUNC) &&
1020 		     (irb->scsw.cmd.stctl & SCSW_STCTL_STATUS_PEND))) {
1021 		if (cqr && cqr->status == DASD_CQR_IN_IO)
1022 			cqr->status = DASD_CQR_QUEUED;
1023 		device = dasd_device_from_cdev_locked(cdev);
1024 		if (!IS_ERR(device)) {
1025 			dasd_device_clear_timer(device);
1026 			device->discipline->handle_unsolicited_interrupt(device,
1027 									 irb);
1028 			dasd_put_device(device);
1029 		}
1030 		return;
1031 	}
1032 
1033 	device = (struct dasd_device *) cqr->startdev;
1034 	if (!device ||
1035 	    strncmp(device->discipline->ebcname, (char *) &cqr->magic, 4)) {
1036 		MESSAGE(KERN_DEBUG, "invalid device in request: bus_id %s",
1037 			dev_name(&cdev->dev));
1038 		return;
1039 	}
1040 
1041 	/* Check for clear pending */
1042 	if (cqr->status == DASD_CQR_CLEAR_PENDING &&
1043 	    irb->scsw.cmd.fctl & SCSW_FCTL_CLEAR_FUNC) {
1044 		cqr->status = DASD_CQR_CLEARED;
1045 		dasd_device_clear_timer(device);
1046 		wake_up(&dasd_flush_wq);
1047 		dasd_schedule_device_bh(device);
1048 		return;
1049 	}
1050 
1051  	/* check status - the request might have been killed by dyn detach */
1052 	if (cqr->status != DASD_CQR_IN_IO) {
1053 		MESSAGE(KERN_DEBUG,
1054 			"invalid status: bus_id %s, status %02x",
1055 			dev_name(&cdev->dev), cqr->status);
1056 		return;
1057 	}
1058 	DBF_DEV_EVENT(DBF_DEBUG, device, "Int: CS/DS 0x%04x for cqr %p",
1059 		      ((irb->scsw.cmd.cstat << 8) | irb->scsw.cmd.dstat), cqr);
1060 	next = NULL;
1061 	expires = 0;
1062 	if (irb->scsw.cmd.dstat == (DEV_STAT_CHN_END | DEV_STAT_DEV_END) &&
1063 	    irb->scsw.cmd.cstat == 0 && !irb->esw.esw0.erw.cons) {
1064 		/* request was completed successfully */
1065 		cqr->status = DASD_CQR_SUCCESS;
1066 		cqr->stopclk = now;
1067 		/* Start first request on queue if possible -> fast_io. */
1068 		if (cqr->devlist.next != &device->ccw_queue) {
1069 			next = list_entry(cqr->devlist.next,
1070 					  struct dasd_ccw_req, devlist);
1071 		}
1072 	} else {  /* error */
1073 		memcpy(&cqr->irb, irb, sizeof(struct irb));
1074 		if (device->features & DASD_FEATURE_ERPLOG) {
1075 			dasd_log_sense(cqr, irb);
1076 		}
1077 		/*
1078 		 * If we don't want complex ERP for this request, then just
1079 		 * reset this and retry it in the fastpath
1080 		 */
1081 		if (!test_bit(DASD_CQR_FLAGS_USE_ERP, &cqr->flags) &&
1082 		    cqr->retries > 0) {
1083 			DEV_MESSAGE(KERN_DEBUG, device,
1084 				    "default ERP in fastpath (%i retries left)",
1085 				    cqr->retries);
1086 			cqr->lpm    = LPM_ANYPATH;
1087 			cqr->status = DASD_CQR_QUEUED;
1088 			next = cqr;
1089 		} else
1090 			cqr->status = DASD_CQR_ERROR;
1091 	}
1092 	if (next && (next->status == DASD_CQR_QUEUED) &&
1093 	    (!device->stopped)) {
1094 		if (device->discipline->start_IO(next) == 0)
1095 			expires = next->expires;
1096 		else
1097 			DEV_MESSAGE(KERN_DEBUG, device, "%s",
1098 				    "Interrupt fastpath "
1099 				    "failed!");
1100 	}
1101 	if (expires != 0)
1102 		dasd_device_set_timer(device, expires);
1103 	else
1104 		dasd_device_clear_timer(device);
1105 	dasd_schedule_device_bh(device);
1106 }
1107 
1108 /*
1109  * If we have an error on a dasd_block layer request then we cancel
1110  * and return all further requests from the same dasd_block as well.
1111  */
1112 static void __dasd_device_recovery(struct dasd_device *device,
1113 				   struct dasd_ccw_req *ref_cqr)
1114 {
1115 	struct list_head *l, *n;
1116 	struct dasd_ccw_req *cqr;
1117 
1118 	/*
1119 	 * only requeue request that came from the dasd_block layer
1120 	 */
1121 	if (!ref_cqr->block)
1122 		return;
1123 
1124 	list_for_each_safe(l, n, &device->ccw_queue) {
1125 		cqr = list_entry(l, struct dasd_ccw_req, devlist);
1126 		if (cqr->status == DASD_CQR_QUEUED &&
1127 		    ref_cqr->block == cqr->block) {
1128 			cqr->status = DASD_CQR_CLEARED;
1129 		}
1130 	}
1131 };
1132 
1133 /*
1134  * Remove those ccw requests from the queue that need to be returned
1135  * to the upper layer.
1136  */
1137 static void __dasd_device_process_ccw_queue(struct dasd_device *device,
1138 					    struct list_head *final_queue)
1139 {
1140 	struct list_head *l, *n;
1141 	struct dasd_ccw_req *cqr;
1142 
1143 	/* Process request with final status. */
1144 	list_for_each_safe(l, n, &device->ccw_queue) {
1145 		cqr = list_entry(l, struct dasd_ccw_req, devlist);
1146 
1147 		/* Stop list processing at the first non-final request. */
1148 		if (cqr->status == DASD_CQR_QUEUED ||
1149 		    cqr->status == DASD_CQR_IN_IO ||
1150 		    cqr->status == DASD_CQR_CLEAR_PENDING)
1151 			break;
1152 		if (cqr->status == DASD_CQR_ERROR) {
1153 			__dasd_device_recovery(device, cqr);
1154 		}
1155 		/* Rechain finished requests to final queue */
1156 		list_move_tail(&cqr->devlist, final_queue);
1157 	}
1158 }
1159 
1160 /*
1161  * the cqrs from the final queue are returned to the upper layer
1162  * by setting a dasd_block state and calling the callback function
1163  */
1164 static void __dasd_device_process_final_queue(struct dasd_device *device,
1165 					      struct list_head *final_queue)
1166 {
1167 	struct list_head *l, *n;
1168 	struct dasd_ccw_req *cqr;
1169 	struct dasd_block *block;
1170 	void (*callback)(struct dasd_ccw_req *, void *data);
1171 	void *callback_data;
1172 
1173 	list_for_each_safe(l, n, final_queue) {
1174 		cqr = list_entry(l, struct dasd_ccw_req, devlist);
1175 		list_del_init(&cqr->devlist);
1176 		block = cqr->block;
1177 		callback = cqr->callback;
1178 		callback_data = cqr->callback_data;
1179 		if (block)
1180 			spin_lock_bh(&block->queue_lock);
1181 		switch (cqr->status) {
1182 		case DASD_CQR_SUCCESS:
1183 			cqr->status = DASD_CQR_DONE;
1184 			break;
1185 		case DASD_CQR_ERROR:
1186 			cqr->status = DASD_CQR_NEED_ERP;
1187 			break;
1188 		case DASD_CQR_CLEARED:
1189 			cqr->status = DASD_CQR_TERMINATED;
1190 			break;
1191 		default:
1192 			DEV_MESSAGE(KERN_ERR, device,
1193 				    "wrong cqr status in __dasd_process_final_queue "
1194 				    "for cqr %p, status %x",
1195 				    cqr, cqr->status);
1196 			BUG();
1197 		}
1198 		if (cqr->callback != NULL)
1199 			(callback)(cqr, callback_data);
1200 		if (block)
1201 			spin_unlock_bh(&block->queue_lock);
1202 	}
1203 }
1204 
1205 /*
1206  * Take a look at the first request on the ccw queue and check
1207  * if it reached its expire time. If so, terminate the IO.
1208  */
1209 static void __dasd_device_check_expire(struct dasd_device *device)
1210 {
1211 	struct dasd_ccw_req *cqr;
1212 
1213 	if (list_empty(&device->ccw_queue))
1214 		return;
1215 	cqr = list_entry(device->ccw_queue.next, struct dasd_ccw_req, devlist);
1216 	if ((cqr->status == DASD_CQR_IN_IO && cqr->expires != 0) &&
1217 	    (time_after_eq(jiffies, cqr->expires + cqr->starttime))) {
1218 		if (device->discipline->term_IO(cqr) != 0) {
1219 			/* Hmpf, try again in 5 sec */
1220 			DEV_MESSAGE(KERN_ERR, device,
1221 				    "internal error - timeout (%is) expired "
1222 				    "for cqr %p, termination failed, "
1223 				    "retrying in 5s",
1224 				    (cqr->expires/HZ), cqr);
1225 			cqr->expires += 5*HZ;
1226 			dasd_device_set_timer(device, 5*HZ);
1227 		} else {
1228 			DEV_MESSAGE(KERN_ERR, device,
1229 				    "internal error - timeout (%is) expired "
1230 				    "for cqr %p (%i retries left)",
1231 				    (cqr->expires/HZ), cqr, cqr->retries);
1232 		}
1233 	}
1234 }
1235 
1236 /*
1237  * Take a look at the first request on the ccw queue and check
1238  * if it needs to be started.
1239  */
1240 static void __dasd_device_start_head(struct dasd_device *device)
1241 {
1242 	struct dasd_ccw_req *cqr;
1243 	int rc;
1244 
1245 	if (list_empty(&device->ccw_queue))
1246 		return;
1247 	cqr = list_entry(device->ccw_queue.next, struct dasd_ccw_req, devlist);
1248 	if (cqr->status != DASD_CQR_QUEUED)
1249 		return;
1250 	/* when device is stopped, return request to previous layer */
1251 	if (device->stopped) {
1252 		cqr->status = DASD_CQR_CLEARED;
1253 		dasd_schedule_device_bh(device);
1254 		return;
1255 	}
1256 
1257 	rc = device->discipline->start_IO(cqr);
1258 	if (rc == 0)
1259 		dasd_device_set_timer(device, cqr->expires);
1260 	else if (rc == -EACCES) {
1261 		dasd_schedule_device_bh(device);
1262 	} else
1263 		/* Hmpf, try again in 1/2 sec */
1264 		dasd_device_set_timer(device, 50);
1265 }
1266 
1267 /*
1268  * Go through all request on the dasd_device request queue,
1269  * terminate them on the cdev if necessary, and return them to the
1270  * submitting layer via callback.
1271  * Note:
1272  * Make sure that all 'submitting layers' still exist when
1273  * this function is called!. In other words, when 'device' is a base
1274  * device then all block layer requests must have been removed before
1275  * via dasd_flush_block_queue.
1276  */
1277 int dasd_flush_device_queue(struct dasd_device *device)
1278 {
1279 	struct dasd_ccw_req *cqr, *n;
1280 	int rc;
1281 	struct list_head flush_queue;
1282 
1283 	INIT_LIST_HEAD(&flush_queue);
1284 	spin_lock_irq(get_ccwdev_lock(device->cdev));
1285 	rc = 0;
1286 	list_for_each_entry_safe(cqr, n, &device->ccw_queue, devlist) {
1287 		/* Check status and move request to flush_queue */
1288 		switch (cqr->status) {
1289 		case DASD_CQR_IN_IO:
1290 			rc = device->discipline->term_IO(cqr);
1291 			if (rc) {
1292 				/* unable to terminate requeust */
1293 				DEV_MESSAGE(KERN_ERR, device,
1294 					    "dasd flush ccw_queue is unable "
1295 					    " to terminate request %p",
1296 					    cqr);
1297 				/* stop flush processing */
1298 				goto finished;
1299 			}
1300 			break;
1301 		case DASD_CQR_QUEUED:
1302 			cqr->stopclk = get_clock();
1303 			cqr->status = DASD_CQR_CLEARED;
1304 			break;
1305 		default: /* no need to modify the others */
1306 			break;
1307 		}
1308 		list_move_tail(&cqr->devlist, &flush_queue);
1309 	}
1310 finished:
1311 	spin_unlock_irq(get_ccwdev_lock(device->cdev));
1312 	/*
1313 	 * After this point all requests must be in state CLEAR_PENDING,
1314 	 * CLEARED, SUCCESS or ERROR. Now wait for CLEAR_PENDING to become
1315 	 * one of the others.
1316 	 */
1317 	list_for_each_entry_safe(cqr, n, &flush_queue, devlist)
1318 		wait_event(dasd_flush_wq,
1319 			   (cqr->status != DASD_CQR_CLEAR_PENDING));
1320 	/*
1321 	 * Now set each request back to TERMINATED, DONE or NEED_ERP
1322 	 * and call the callback function of flushed requests
1323 	 */
1324 	__dasd_device_process_final_queue(device, &flush_queue);
1325 	return rc;
1326 }
1327 
1328 /*
1329  * Acquire the device lock and process queues for the device.
1330  */
1331 static void dasd_device_tasklet(struct dasd_device *device)
1332 {
1333 	struct list_head final_queue;
1334 
1335 	atomic_set (&device->tasklet_scheduled, 0);
1336 	INIT_LIST_HEAD(&final_queue);
1337 	spin_lock_irq(get_ccwdev_lock(device->cdev));
1338 	/* Check expire time of first request on the ccw queue. */
1339 	__dasd_device_check_expire(device);
1340 	/* find final requests on ccw queue */
1341 	__dasd_device_process_ccw_queue(device, &final_queue);
1342 	spin_unlock_irq(get_ccwdev_lock(device->cdev));
1343 	/* Now call the callback function of requests with final status */
1344 	__dasd_device_process_final_queue(device, &final_queue);
1345 	spin_lock_irq(get_ccwdev_lock(device->cdev));
1346 	/* Now check if the head of the ccw queue needs to be started. */
1347 	__dasd_device_start_head(device);
1348 	spin_unlock_irq(get_ccwdev_lock(device->cdev));
1349 	dasd_put_device(device);
1350 }
1351 
1352 /*
1353  * Schedules a call to dasd_tasklet over the device tasklet.
1354  */
1355 void dasd_schedule_device_bh(struct dasd_device *device)
1356 {
1357 	/* Protect against rescheduling. */
1358 	if (atomic_cmpxchg (&device->tasklet_scheduled, 0, 1) != 0)
1359 		return;
1360 	dasd_get_device(device);
1361 	tasklet_hi_schedule(&device->tasklet);
1362 }
1363 
1364 /*
1365  * Queue a request to the head of the device ccw_queue.
1366  * Start the I/O if possible.
1367  */
1368 void dasd_add_request_head(struct dasd_ccw_req *cqr)
1369 {
1370 	struct dasd_device *device;
1371 	unsigned long flags;
1372 
1373 	device = cqr->startdev;
1374 	spin_lock_irqsave(get_ccwdev_lock(device->cdev), flags);
1375 	cqr->status = DASD_CQR_QUEUED;
1376 	list_add(&cqr->devlist, &device->ccw_queue);
1377 	/* let the bh start the request to keep them in order */
1378 	dasd_schedule_device_bh(device);
1379 	spin_unlock_irqrestore(get_ccwdev_lock(device->cdev), flags);
1380 }
1381 
1382 /*
1383  * Queue a request to the tail of the device ccw_queue.
1384  * Start the I/O if possible.
1385  */
1386 void dasd_add_request_tail(struct dasd_ccw_req *cqr)
1387 {
1388 	struct dasd_device *device;
1389 	unsigned long flags;
1390 
1391 	device = cqr->startdev;
1392 	spin_lock_irqsave(get_ccwdev_lock(device->cdev), flags);
1393 	cqr->status = DASD_CQR_QUEUED;
1394 	list_add_tail(&cqr->devlist, &device->ccw_queue);
1395 	/* let the bh start the request to keep them in order */
1396 	dasd_schedule_device_bh(device);
1397 	spin_unlock_irqrestore(get_ccwdev_lock(device->cdev), flags);
1398 }
1399 
1400 /*
1401  * Wakeup helper for the 'sleep_on' functions.
1402  */
1403 static void dasd_wakeup_cb(struct dasd_ccw_req *cqr, void *data)
1404 {
1405 	wake_up((wait_queue_head_t *) data);
1406 }
1407 
1408 static inline int _wait_for_wakeup(struct dasd_ccw_req *cqr)
1409 {
1410 	struct dasd_device *device;
1411 	int rc;
1412 
1413 	device = cqr->startdev;
1414 	spin_lock_irq(get_ccwdev_lock(device->cdev));
1415 	rc = ((cqr->status == DASD_CQR_DONE ||
1416 	       cqr->status == DASD_CQR_NEED_ERP ||
1417 	       cqr->status == DASD_CQR_TERMINATED) &&
1418 	      list_empty(&cqr->devlist));
1419 	spin_unlock_irq(get_ccwdev_lock(device->cdev));
1420 	return rc;
1421 }
1422 
1423 /*
1424  * Queue a request to the tail of the device ccw_queue and wait for
1425  * it's completion.
1426  */
1427 int dasd_sleep_on(struct dasd_ccw_req *cqr)
1428 {
1429 	struct dasd_device *device;
1430 	int rc;
1431 
1432 	device = cqr->startdev;
1433 
1434 	cqr->callback = dasd_wakeup_cb;
1435 	cqr->callback_data = (void *) &generic_waitq;
1436 	dasd_add_request_tail(cqr);
1437 	wait_event(generic_waitq, _wait_for_wakeup(cqr));
1438 
1439 	/* Request status is either done or failed. */
1440 	rc = (cqr->status == DASD_CQR_DONE) ? 0 : -EIO;
1441 	return rc;
1442 }
1443 
1444 /*
1445  * Queue a request to the tail of the device ccw_queue and wait
1446  * interruptible for it's completion.
1447  */
1448 int dasd_sleep_on_interruptible(struct dasd_ccw_req *cqr)
1449 {
1450 	struct dasd_device *device;
1451 	int rc;
1452 
1453 	device = cqr->startdev;
1454 	cqr->callback = dasd_wakeup_cb;
1455 	cqr->callback_data = (void *) &generic_waitq;
1456 	dasd_add_request_tail(cqr);
1457 	rc = wait_event_interruptible(generic_waitq, _wait_for_wakeup(cqr));
1458 	if (rc == -ERESTARTSYS) {
1459 		dasd_cancel_req(cqr);
1460 		/* wait (non-interruptible) for final status */
1461 		wait_event(generic_waitq, _wait_for_wakeup(cqr));
1462 	}
1463 	rc = (cqr->status == DASD_CQR_DONE) ? 0 : -EIO;
1464 	return rc;
1465 }
1466 
1467 /*
1468  * Whoa nelly now it gets really hairy. For some functions (e.g. steal lock
1469  * for eckd devices) the currently running request has to be terminated
1470  * and be put back to status queued, before the special request is added
1471  * to the head of the queue. Then the special request is waited on normally.
1472  */
1473 static inline int _dasd_term_running_cqr(struct dasd_device *device)
1474 {
1475 	struct dasd_ccw_req *cqr;
1476 
1477 	if (list_empty(&device->ccw_queue))
1478 		return 0;
1479 	cqr = list_entry(device->ccw_queue.next, struct dasd_ccw_req, devlist);
1480 	return device->discipline->term_IO(cqr);
1481 }
1482 
1483 int dasd_sleep_on_immediatly(struct dasd_ccw_req *cqr)
1484 {
1485 	struct dasd_device *device;
1486 	int rc;
1487 
1488 	device = cqr->startdev;
1489 	spin_lock_irq(get_ccwdev_lock(device->cdev));
1490 	rc = _dasd_term_running_cqr(device);
1491 	if (rc) {
1492 		spin_unlock_irq(get_ccwdev_lock(device->cdev));
1493 		return rc;
1494 	}
1495 
1496 	cqr->callback = dasd_wakeup_cb;
1497 	cqr->callback_data = (void *) &generic_waitq;
1498 	cqr->status = DASD_CQR_QUEUED;
1499 	list_add(&cqr->devlist, &device->ccw_queue);
1500 
1501 	/* let the bh start the request to keep them in order */
1502 	dasd_schedule_device_bh(device);
1503 
1504 	spin_unlock_irq(get_ccwdev_lock(device->cdev));
1505 
1506 	wait_event(generic_waitq, _wait_for_wakeup(cqr));
1507 
1508 	/* Request status is either done or failed. */
1509 	rc = (cqr->status == DASD_CQR_DONE) ? 0 : -EIO;
1510 	return rc;
1511 }
1512 
1513 /*
1514  * Cancels a request that was started with dasd_sleep_on_req.
1515  * This is useful to timeout requests. The request will be
1516  * terminated if it is currently in i/o.
1517  * Returns 1 if the request has been terminated.
1518  *	   0 if there was no need to terminate the request (not started yet)
1519  *	   negative error code if termination failed
1520  * Cancellation of a request is an asynchronous operation! The calling
1521  * function has to wait until the request is properly returned via callback.
1522  */
1523 int dasd_cancel_req(struct dasd_ccw_req *cqr)
1524 {
1525 	struct dasd_device *device = cqr->startdev;
1526 	unsigned long flags;
1527 	int rc;
1528 
1529 	rc = 0;
1530 	spin_lock_irqsave(get_ccwdev_lock(device->cdev), flags);
1531 	switch (cqr->status) {
1532 	case DASD_CQR_QUEUED:
1533 		/* request was not started - just set to cleared */
1534 		cqr->status = DASD_CQR_CLEARED;
1535 		break;
1536 	case DASD_CQR_IN_IO:
1537 		/* request in IO - terminate IO and release again */
1538 		rc = device->discipline->term_IO(cqr);
1539 		if (rc) {
1540 			DEV_MESSAGE(KERN_ERR, device,
1541 				    "dasd_cancel_req is unable "
1542 				    " to terminate request %p, rc = %d",
1543 				    cqr, rc);
1544 		} else {
1545 			cqr->stopclk = get_clock();
1546 			rc = 1;
1547 		}
1548 		break;
1549 	default: /* already finished or clear pending - do nothing */
1550 		break;
1551 	}
1552 	spin_unlock_irqrestore(get_ccwdev_lock(device->cdev), flags);
1553 	dasd_schedule_device_bh(device);
1554 	return rc;
1555 }
1556 
1557 
1558 /*
1559  * SECTION: Operations of the dasd_block layer.
1560  */
1561 
1562 /*
1563  * Timeout function for dasd_block. This is used when the block layer
1564  * is waiting for something that may not come reliably, (e.g. a state
1565  * change interrupt)
1566  */
1567 static void dasd_block_timeout(unsigned long ptr)
1568 {
1569 	unsigned long flags;
1570 	struct dasd_block *block;
1571 
1572 	block = (struct dasd_block *) ptr;
1573 	spin_lock_irqsave(get_ccwdev_lock(block->base->cdev), flags);
1574 	/* re-activate request queue */
1575 	block->base->stopped &= ~DASD_STOPPED_PENDING;
1576 	spin_unlock_irqrestore(get_ccwdev_lock(block->base->cdev), flags);
1577 	dasd_schedule_block_bh(block);
1578 }
1579 
1580 /*
1581  * Setup timeout for a dasd_block in jiffies.
1582  */
1583 void dasd_block_set_timer(struct dasd_block *block, int expires)
1584 {
1585 	if (expires == 0)
1586 		del_timer(&block->timer);
1587 	else
1588 		mod_timer(&block->timer, jiffies + expires);
1589 }
1590 
1591 /*
1592  * Clear timeout for a dasd_block.
1593  */
1594 void dasd_block_clear_timer(struct dasd_block *block)
1595 {
1596 	del_timer(&block->timer);
1597 }
1598 
1599 /*
1600  * posts the buffer_cache about a finalized request
1601  */
1602 static inline void dasd_end_request(struct request *req, int error)
1603 {
1604 	if (__blk_end_request(req, error, blk_rq_bytes(req)))
1605 		BUG();
1606 }
1607 
1608 /*
1609  * Process finished error recovery ccw.
1610  */
1611 static inline void __dasd_block_process_erp(struct dasd_block *block,
1612 					    struct dasd_ccw_req *cqr)
1613 {
1614 	dasd_erp_fn_t erp_fn;
1615 	struct dasd_device *device = block->base;
1616 
1617 	if (cqr->status == DASD_CQR_DONE)
1618 		DBF_DEV_EVENT(DBF_NOTICE, device, "%s", "ERP successful");
1619 	else
1620 		DEV_MESSAGE(KERN_ERR, device, "%s", "ERP unsuccessful");
1621 	erp_fn = device->discipline->erp_postaction(cqr);
1622 	erp_fn(cqr);
1623 }
1624 
1625 /*
1626  * Fetch requests from the block device queue.
1627  */
1628 static void __dasd_process_request_queue(struct dasd_block *block)
1629 {
1630 	struct request_queue *queue;
1631 	struct request *req;
1632 	struct dasd_ccw_req *cqr;
1633 	struct dasd_device *basedev;
1634 	unsigned long flags;
1635 	queue = block->request_queue;
1636 	basedev = block->base;
1637 	/* No queue ? Then there is nothing to do. */
1638 	if (queue == NULL)
1639 		return;
1640 
1641 	/*
1642 	 * We requeue request from the block device queue to the ccw
1643 	 * queue only in two states. In state DASD_STATE_READY the
1644 	 * partition detection is done and we need to requeue requests
1645 	 * for that. State DASD_STATE_ONLINE is normal block device
1646 	 * operation.
1647 	 */
1648 	if (basedev->state < DASD_STATE_READY)
1649 		return;
1650 	/* Now we try to fetch requests from the request queue */
1651 	while (!blk_queue_plugged(queue) &&
1652 	       elv_next_request(queue)) {
1653 
1654 		req = elv_next_request(queue);
1655 
1656 		if (basedev->features & DASD_FEATURE_READONLY &&
1657 		    rq_data_dir(req) == WRITE) {
1658 			DBF_DEV_EVENT(DBF_ERR, basedev,
1659 				      "Rejecting write request %p",
1660 				      req);
1661 			blkdev_dequeue_request(req);
1662 			dasd_end_request(req, -EIO);
1663 			continue;
1664 		}
1665 		cqr = basedev->discipline->build_cp(basedev, block, req);
1666 		if (IS_ERR(cqr)) {
1667 			if (PTR_ERR(cqr) == -EBUSY)
1668 				break;	/* normal end condition */
1669 			if (PTR_ERR(cqr) == -ENOMEM)
1670 				break;	/* terminate request queue loop */
1671 			if (PTR_ERR(cqr) == -EAGAIN) {
1672 				/*
1673 				 * The current request cannot be build right
1674 				 * now, we have to try later. If this request
1675 				 * is the head-of-queue we stop the device
1676 				 * for 1/2 second.
1677 				 */
1678 				if (!list_empty(&block->ccw_queue))
1679 					break;
1680 				spin_lock_irqsave(get_ccwdev_lock(basedev->cdev), flags);
1681 				basedev->stopped |= DASD_STOPPED_PENDING;
1682 				spin_unlock_irqrestore(get_ccwdev_lock(basedev->cdev), flags);
1683 				dasd_block_set_timer(block, HZ/2);
1684 				break;
1685 			}
1686 			DBF_DEV_EVENT(DBF_ERR, basedev,
1687 				      "CCW creation failed (rc=%ld) "
1688 				      "on request %p",
1689 				      PTR_ERR(cqr), req);
1690 			blkdev_dequeue_request(req);
1691 			dasd_end_request(req, -EIO);
1692 			continue;
1693 		}
1694 		/*
1695 		 *  Note: callback is set to dasd_return_cqr_cb in
1696 		 * __dasd_block_start_head to cover erp requests as well
1697 		 */
1698 		cqr->callback_data = (void *) req;
1699 		cqr->status = DASD_CQR_FILLED;
1700 		blkdev_dequeue_request(req);
1701 		list_add_tail(&cqr->blocklist, &block->ccw_queue);
1702 		dasd_profile_start(block, cqr, req);
1703 	}
1704 }
1705 
1706 static void __dasd_cleanup_cqr(struct dasd_ccw_req *cqr)
1707 {
1708 	struct request *req;
1709 	int status;
1710 	int error = 0;
1711 
1712 	req = (struct request *) cqr->callback_data;
1713 	dasd_profile_end(cqr->block, cqr, req);
1714 	status = cqr->block->base->discipline->free_cp(cqr, req);
1715 	if (status <= 0)
1716 		error = status ? status : -EIO;
1717 	dasd_end_request(req, error);
1718 }
1719 
1720 /*
1721  * Process ccw request queue.
1722  */
1723 static void __dasd_process_block_ccw_queue(struct dasd_block *block,
1724 					   struct list_head *final_queue)
1725 {
1726 	struct list_head *l, *n;
1727 	struct dasd_ccw_req *cqr;
1728 	dasd_erp_fn_t erp_fn;
1729 	unsigned long flags;
1730 	struct dasd_device *base = block->base;
1731 
1732 restart:
1733 	/* Process request with final status. */
1734 	list_for_each_safe(l, n, &block->ccw_queue) {
1735 		cqr = list_entry(l, struct dasd_ccw_req, blocklist);
1736 		if (cqr->status != DASD_CQR_DONE &&
1737 		    cqr->status != DASD_CQR_FAILED &&
1738 		    cqr->status != DASD_CQR_NEED_ERP &&
1739 		    cqr->status != DASD_CQR_TERMINATED)
1740 			continue;
1741 
1742 		if (cqr->status == DASD_CQR_TERMINATED) {
1743 			base->discipline->handle_terminated_request(cqr);
1744 			goto restart;
1745 		}
1746 
1747 		/*  Process requests that may be recovered */
1748 		if (cqr->status == DASD_CQR_NEED_ERP) {
1749 			erp_fn = base->discipline->erp_action(cqr);
1750 			erp_fn(cqr);
1751 			goto restart;
1752 		}
1753 
1754 		/* log sense for fatal error */
1755 		if (cqr->status == DASD_CQR_FAILED) {
1756 			dasd_log_sense(cqr, &cqr->irb);
1757 		}
1758 
1759 		/* First of all call extended error reporting. */
1760 		if (dasd_eer_enabled(base) &&
1761 		    cqr->status == DASD_CQR_FAILED) {
1762 			dasd_eer_write(base, cqr, DASD_EER_FATALERROR);
1763 
1764 			/* restart request  */
1765 			cqr->status = DASD_CQR_FILLED;
1766 			cqr->retries = 255;
1767 			spin_lock_irqsave(get_ccwdev_lock(base->cdev), flags);
1768 			base->stopped |= DASD_STOPPED_QUIESCE;
1769 			spin_unlock_irqrestore(get_ccwdev_lock(base->cdev),
1770 					       flags);
1771 			goto restart;
1772 		}
1773 
1774 		/* Process finished ERP request. */
1775 		if (cqr->refers) {
1776 			__dasd_block_process_erp(block, cqr);
1777 			goto restart;
1778 		}
1779 
1780 		/* Rechain finished requests to final queue */
1781 		cqr->endclk = get_clock();
1782 		list_move_tail(&cqr->blocklist, final_queue);
1783 	}
1784 }
1785 
1786 static void dasd_return_cqr_cb(struct dasd_ccw_req *cqr, void *data)
1787 {
1788 	dasd_schedule_block_bh(cqr->block);
1789 }
1790 
1791 static void __dasd_block_start_head(struct dasd_block *block)
1792 {
1793 	struct dasd_ccw_req *cqr;
1794 
1795 	if (list_empty(&block->ccw_queue))
1796 		return;
1797 	/* We allways begin with the first requests on the queue, as some
1798 	 * of previously started requests have to be enqueued on a
1799 	 * dasd_device again for error recovery.
1800 	 */
1801 	list_for_each_entry(cqr, &block->ccw_queue, blocklist) {
1802 		if (cqr->status != DASD_CQR_FILLED)
1803 			continue;
1804 		/* Non-temporary stop condition will trigger fail fast */
1805 		if (block->base->stopped & ~DASD_STOPPED_PENDING &&
1806 		    test_bit(DASD_CQR_FLAGS_FAILFAST, &cqr->flags) &&
1807 		    (!dasd_eer_enabled(block->base))) {
1808 			cqr->status = DASD_CQR_FAILED;
1809 			dasd_schedule_block_bh(block);
1810 			continue;
1811 		}
1812 		/* Don't try to start requests if device is stopped */
1813 		if (block->base->stopped)
1814 			return;
1815 
1816 		/* just a fail safe check, should not happen */
1817 		if (!cqr->startdev)
1818 			cqr->startdev = block->base;
1819 
1820 		/* make sure that the requests we submit find their way back */
1821 		cqr->callback = dasd_return_cqr_cb;
1822 
1823 		dasd_add_request_tail(cqr);
1824 	}
1825 }
1826 
1827 /*
1828  * Central dasd_block layer routine. Takes requests from the generic
1829  * block layer request queue, creates ccw requests, enqueues them on
1830  * a dasd_device and processes ccw requests that have been returned.
1831  */
1832 static void dasd_block_tasklet(struct dasd_block *block)
1833 {
1834 	struct list_head final_queue;
1835 	struct list_head *l, *n;
1836 	struct dasd_ccw_req *cqr;
1837 
1838 	atomic_set(&block->tasklet_scheduled, 0);
1839 	INIT_LIST_HEAD(&final_queue);
1840 	spin_lock(&block->queue_lock);
1841 	/* Finish off requests on ccw queue */
1842 	__dasd_process_block_ccw_queue(block, &final_queue);
1843 	spin_unlock(&block->queue_lock);
1844 	/* Now call the callback function of requests with final status */
1845 	spin_lock_irq(&block->request_queue_lock);
1846 	list_for_each_safe(l, n, &final_queue) {
1847 		cqr = list_entry(l, struct dasd_ccw_req, blocklist);
1848 		list_del_init(&cqr->blocklist);
1849 		__dasd_cleanup_cqr(cqr);
1850 	}
1851 	spin_lock(&block->queue_lock);
1852 	/* Get new request from the block device request queue */
1853 	__dasd_process_request_queue(block);
1854 	/* Now check if the head of the ccw queue needs to be started. */
1855 	__dasd_block_start_head(block);
1856 	spin_unlock(&block->queue_lock);
1857 	spin_unlock_irq(&block->request_queue_lock);
1858 	dasd_put_device(block->base);
1859 }
1860 
1861 static void _dasd_wake_block_flush_cb(struct dasd_ccw_req *cqr, void *data)
1862 {
1863 	wake_up(&dasd_flush_wq);
1864 }
1865 
1866 /*
1867  * Go through all request on the dasd_block request queue, cancel them
1868  * on the respective dasd_device, and return them to the generic
1869  * block layer.
1870  */
1871 static int dasd_flush_block_queue(struct dasd_block *block)
1872 {
1873 	struct dasd_ccw_req *cqr, *n;
1874 	int rc, i;
1875 	struct list_head flush_queue;
1876 
1877 	INIT_LIST_HEAD(&flush_queue);
1878 	spin_lock_bh(&block->queue_lock);
1879 	rc = 0;
1880 restart:
1881 	list_for_each_entry_safe(cqr, n, &block->ccw_queue, blocklist) {
1882 		/* if this request currently owned by a dasd_device cancel it */
1883 		if (cqr->status >= DASD_CQR_QUEUED)
1884 			rc = dasd_cancel_req(cqr);
1885 		if (rc < 0)
1886 			break;
1887 		/* Rechain request (including erp chain) so it won't be
1888 		 * touched by the dasd_block_tasklet anymore.
1889 		 * Replace the callback so we notice when the request
1890 		 * is returned from the dasd_device layer.
1891 		 */
1892 		cqr->callback = _dasd_wake_block_flush_cb;
1893 		for (i = 0; cqr != NULL; cqr = cqr->refers, i++)
1894 			list_move_tail(&cqr->blocklist, &flush_queue);
1895 		if (i > 1)
1896 			/* moved more than one request - need to restart */
1897 			goto restart;
1898 	}
1899 	spin_unlock_bh(&block->queue_lock);
1900 	/* Now call the callback function of flushed requests */
1901 restart_cb:
1902 	list_for_each_entry_safe(cqr, n, &flush_queue, blocklist) {
1903 		wait_event(dasd_flush_wq, (cqr->status < DASD_CQR_QUEUED));
1904 		/* Process finished ERP request. */
1905 		if (cqr->refers) {
1906 			spin_lock_bh(&block->queue_lock);
1907 			__dasd_block_process_erp(block, cqr);
1908 			spin_unlock_bh(&block->queue_lock);
1909 			/* restart list_for_xx loop since dasd_process_erp
1910 			 * might remove multiple elements */
1911 			goto restart_cb;
1912 		}
1913 		/* call the callback function */
1914 		spin_lock_irq(&block->request_queue_lock);
1915 		cqr->endclk = get_clock();
1916 		list_del_init(&cqr->blocklist);
1917 		__dasd_cleanup_cqr(cqr);
1918 		spin_unlock_irq(&block->request_queue_lock);
1919 	}
1920 	return rc;
1921 }
1922 
1923 /*
1924  * Schedules a call to dasd_tasklet over the device tasklet.
1925  */
1926 void dasd_schedule_block_bh(struct dasd_block *block)
1927 {
1928 	/* Protect against rescheduling. */
1929 	if (atomic_cmpxchg(&block->tasklet_scheduled, 0, 1) != 0)
1930 		return;
1931 	/* life cycle of block is bound to it's base device */
1932 	dasd_get_device(block->base);
1933 	tasklet_hi_schedule(&block->tasklet);
1934 }
1935 
1936 
1937 /*
1938  * SECTION: external block device operations
1939  * (request queue handling, open, release, etc.)
1940  */
1941 
1942 /*
1943  * Dasd request queue function. Called from ll_rw_blk.c
1944  */
1945 static void do_dasd_request(struct request_queue *queue)
1946 {
1947 	struct dasd_block *block;
1948 
1949 	block = queue->queuedata;
1950 	spin_lock(&block->queue_lock);
1951 	/* Get new request from the block device request queue */
1952 	__dasd_process_request_queue(block);
1953 	/* Now check if the head of the ccw queue needs to be started. */
1954 	__dasd_block_start_head(block);
1955 	spin_unlock(&block->queue_lock);
1956 }
1957 
1958 /*
1959  * Allocate and initialize request queue and default I/O scheduler.
1960  */
1961 static int dasd_alloc_queue(struct dasd_block *block)
1962 {
1963 	int rc;
1964 
1965 	block->request_queue = blk_init_queue(do_dasd_request,
1966 					       &block->request_queue_lock);
1967 	if (block->request_queue == NULL)
1968 		return -ENOMEM;
1969 
1970 	block->request_queue->queuedata = block;
1971 
1972 	elevator_exit(block->request_queue->elevator);
1973 	block->request_queue->elevator = NULL;
1974 	rc = elevator_init(block->request_queue, "deadline");
1975 	if (rc) {
1976 		blk_cleanup_queue(block->request_queue);
1977 		return rc;
1978 	}
1979 	return 0;
1980 }
1981 
1982 /*
1983  * Allocate and initialize request queue.
1984  */
1985 static void dasd_setup_queue(struct dasd_block *block)
1986 {
1987 	int max;
1988 
1989 	blk_queue_hardsect_size(block->request_queue, block->bp_block);
1990 	max = block->base->discipline->max_blocks << block->s2b_shift;
1991 	blk_queue_max_sectors(block->request_queue, max);
1992 	blk_queue_max_phys_segments(block->request_queue, -1L);
1993 	blk_queue_max_hw_segments(block->request_queue, -1L);
1994 	blk_queue_max_segment_size(block->request_queue, -1L);
1995 	blk_queue_segment_boundary(block->request_queue, -1L);
1996 	blk_queue_ordered(block->request_queue, QUEUE_ORDERED_DRAIN, NULL);
1997 }
1998 
1999 /*
2000  * Deactivate and free request queue.
2001  */
2002 static void dasd_free_queue(struct dasd_block *block)
2003 {
2004 	if (block->request_queue) {
2005 		blk_cleanup_queue(block->request_queue);
2006 		block->request_queue = NULL;
2007 	}
2008 }
2009 
2010 /*
2011  * Flush request on the request queue.
2012  */
2013 static void dasd_flush_request_queue(struct dasd_block *block)
2014 {
2015 	struct request *req;
2016 
2017 	if (!block->request_queue)
2018 		return;
2019 
2020 	spin_lock_irq(&block->request_queue_lock);
2021 	while ((req = elv_next_request(block->request_queue))) {
2022 		blkdev_dequeue_request(req);
2023 		dasd_end_request(req, -EIO);
2024 	}
2025 	spin_unlock_irq(&block->request_queue_lock);
2026 }
2027 
2028 static int dasd_open(struct block_device *bdev, fmode_t mode)
2029 {
2030 	struct dasd_block *block = bdev->bd_disk->private_data;
2031 	struct dasd_device *base = block->base;
2032 	int rc;
2033 
2034 	atomic_inc(&block->open_count);
2035 	if (test_bit(DASD_FLAG_OFFLINE, &base->flags)) {
2036 		rc = -ENODEV;
2037 		goto unlock;
2038 	}
2039 
2040 	if (!try_module_get(base->discipline->owner)) {
2041 		rc = -EINVAL;
2042 		goto unlock;
2043 	}
2044 
2045 	if (dasd_probeonly) {
2046 		DEV_MESSAGE(KERN_INFO, base, "%s",
2047 			    "No access to device due to probeonly mode");
2048 		rc = -EPERM;
2049 		goto out;
2050 	}
2051 
2052 	if (base->state <= DASD_STATE_BASIC) {
2053 		DBF_DEV_EVENT(DBF_ERR, base, " %s",
2054 			      " Cannot open unrecognized device");
2055 		rc = -ENODEV;
2056 		goto out;
2057 	}
2058 
2059 	return 0;
2060 
2061 out:
2062 	module_put(base->discipline->owner);
2063 unlock:
2064 	atomic_dec(&block->open_count);
2065 	return rc;
2066 }
2067 
2068 static int dasd_release(struct gendisk *disk, fmode_t mode)
2069 {
2070 	struct dasd_block *block = disk->private_data;
2071 
2072 	atomic_dec(&block->open_count);
2073 	module_put(block->base->discipline->owner);
2074 	return 0;
2075 }
2076 
2077 /*
2078  * Return disk geometry.
2079  */
2080 static int dasd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
2081 {
2082 	struct dasd_block *block;
2083 	struct dasd_device *base;
2084 
2085 	block = bdev->bd_disk->private_data;
2086 	base = block->base;
2087 	if (!block)
2088 		return -ENODEV;
2089 
2090 	if (!base->discipline ||
2091 	    !base->discipline->fill_geometry)
2092 		return -EINVAL;
2093 
2094 	base->discipline->fill_geometry(block, geo);
2095 	geo->start = get_start_sect(bdev) >> block->s2b_shift;
2096 	return 0;
2097 }
2098 
2099 struct block_device_operations
2100 dasd_device_operations = {
2101 	.owner		= THIS_MODULE,
2102 	.open		= dasd_open,
2103 	.release	= dasd_release,
2104 	.locked_ioctl	= dasd_ioctl,
2105 	.getgeo		= dasd_getgeo,
2106 };
2107 
2108 /*******************************************************************************
2109  * end of block device operations
2110  */
2111 
2112 static void
2113 dasd_exit(void)
2114 {
2115 #ifdef CONFIG_PROC_FS
2116 	dasd_proc_exit();
2117 #endif
2118 	dasd_eer_exit();
2119         if (dasd_page_cache != NULL) {
2120 		kmem_cache_destroy(dasd_page_cache);
2121 		dasd_page_cache = NULL;
2122 	}
2123 	dasd_gendisk_exit();
2124 	dasd_devmap_exit();
2125 	if (dasd_debug_area != NULL) {
2126 		debug_unregister(dasd_debug_area);
2127 		dasd_debug_area = NULL;
2128 	}
2129 }
2130 
2131 /*
2132  * SECTION: common functions for ccw_driver use
2133  */
2134 
2135 /*
2136  * Initial attempt at a probe function. this can be simplified once
2137  * the other detection code is gone.
2138  */
2139 int dasd_generic_probe(struct ccw_device *cdev,
2140 		       struct dasd_discipline *discipline)
2141 {
2142 	int ret;
2143 
2144 	ret = ccw_device_set_options(cdev, CCWDEV_DO_PATHGROUP);
2145 	if (ret) {
2146 		printk(KERN_WARNING
2147 		       "dasd_generic_probe: could not set ccw-device options "
2148 		       "for %s\n", dev_name(&cdev->dev));
2149 		return ret;
2150 	}
2151 	ret = dasd_add_sysfs_files(cdev);
2152 	if (ret) {
2153 		printk(KERN_WARNING
2154 		       "dasd_generic_probe: could not add sysfs entries "
2155 		       "for %s\n", dev_name(&cdev->dev));
2156 		return ret;
2157 	}
2158 	cdev->handler = &dasd_int_handler;
2159 
2160 	/*
2161 	 * Automatically online either all dasd devices (dasd_autodetect)
2162 	 * or all devices specified with dasd= parameters during
2163 	 * initial probe.
2164 	 */
2165 	if ((dasd_get_feature(cdev, DASD_FEATURE_INITIAL_ONLINE) > 0 ) ||
2166 	    (dasd_autodetect && dasd_busid_known(dev_name(&cdev->dev)) != 0))
2167 		ret = ccw_device_set_online(cdev);
2168 	if (ret)
2169 		printk(KERN_WARNING
2170 		       "dasd_generic_probe: could not initially "
2171 		       "online ccw-device %s; return code: %d\n",
2172 		       dev_name(&cdev->dev), ret);
2173 	return 0;
2174 }
2175 
2176 /*
2177  * This will one day be called from a global not_oper handler.
2178  * It is also used by driver_unregister during module unload.
2179  */
2180 void dasd_generic_remove(struct ccw_device *cdev)
2181 {
2182 	struct dasd_device *device;
2183 	struct dasd_block *block;
2184 
2185 	cdev->handler = NULL;
2186 
2187 	dasd_remove_sysfs_files(cdev);
2188 	device = dasd_device_from_cdev(cdev);
2189 	if (IS_ERR(device))
2190 		return;
2191 	if (test_and_set_bit(DASD_FLAG_OFFLINE, &device->flags)) {
2192 		/* Already doing offline processing */
2193 		dasd_put_device(device);
2194 		return;
2195 	}
2196 	/*
2197 	 * This device is removed unconditionally. Set offline
2198 	 * flag to prevent dasd_open from opening it while it is
2199 	 * no quite down yet.
2200 	 */
2201 	dasd_set_target_state(device, DASD_STATE_NEW);
2202 	/* dasd_delete_device destroys the device reference. */
2203 	block = device->block;
2204 	device->block = NULL;
2205 	dasd_delete_device(device);
2206 	/*
2207 	 * life cycle of block is bound to device, so delete it after
2208 	 * device was safely removed
2209 	 */
2210 	if (block)
2211 		dasd_free_block(block);
2212 }
2213 
2214 /*
2215  * Activate a device. This is called from dasd_{eckd,fba}_probe() when either
2216  * the device is detected for the first time and is supposed to be used
2217  * or the user has started activation through sysfs.
2218  */
2219 int dasd_generic_set_online(struct ccw_device *cdev,
2220 			    struct dasd_discipline *base_discipline)
2221 {
2222 	struct dasd_discipline *discipline;
2223 	struct dasd_device *device;
2224 	int rc;
2225 
2226 	/* first online clears initial online feature flag */
2227 	dasd_set_feature(cdev, DASD_FEATURE_INITIAL_ONLINE, 0);
2228 	device = dasd_create_device(cdev);
2229 	if (IS_ERR(device))
2230 		return PTR_ERR(device);
2231 
2232 	discipline = base_discipline;
2233 	if (device->features & DASD_FEATURE_USEDIAG) {
2234 	  	if (!dasd_diag_discipline_pointer) {
2235 		        printk (KERN_WARNING
2236 				"dasd_generic couldn't online device %s "
2237 				"- discipline DIAG not available\n",
2238 				dev_name(&cdev->dev));
2239 			dasd_delete_device(device);
2240 			return -ENODEV;
2241 		}
2242 		discipline = dasd_diag_discipline_pointer;
2243 	}
2244 	if (!try_module_get(base_discipline->owner)) {
2245 		dasd_delete_device(device);
2246 		return -EINVAL;
2247 	}
2248 	if (!try_module_get(discipline->owner)) {
2249 		module_put(base_discipline->owner);
2250 		dasd_delete_device(device);
2251 		return -EINVAL;
2252 	}
2253 	device->base_discipline = base_discipline;
2254 	device->discipline = discipline;
2255 
2256 	/* check_device will allocate block device if necessary */
2257 	rc = discipline->check_device(device);
2258 	if (rc) {
2259 		printk (KERN_WARNING
2260 			"dasd_generic couldn't online device %s "
2261 			"with discipline %s rc=%i\n",
2262 			dev_name(&cdev->dev), discipline->name, rc);
2263 		module_put(discipline->owner);
2264 		module_put(base_discipline->owner);
2265 		dasd_delete_device(device);
2266 		return rc;
2267 	}
2268 
2269 	dasd_set_target_state(device, DASD_STATE_ONLINE);
2270 	if (device->state <= DASD_STATE_KNOWN) {
2271 		printk (KERN_WARNING
2272 			"dasd_generic discipline not found for %s\n",
2273 			dev_name(&cdev->dev));
2274 		rc = -ENODEV;
2275 		dasd_set_target_state(device, DASD_STATE_NEW);
2276 		if (device->block)
2277 			dasd_free_block(device->block);
2278 		dasd_delete_device(device);
2279 	} else
2280 		pr_debug("dasd_generic device %s found\n",
2281 				dev_name(&cdev->dev));
2282 
2283 	/* FIXME: we have to wait for the root device but we don't want
2284 	 * to wait for each single device but for all at once. */
2285 	wait_event(dasd_init_waitq, _wait_for_device(device));
2286 
2287 	dasd_put_device(device);
2288 
2289 	return rc;
2290 }
2291 
2292 int dasd_generic_set_offline(struct ccw_device *cdev)
2293 {
2294 	struct dasd_device *device;
2295 	struct dasd_block *block;
2296 	int max_count, open_count;
2297 
2298 	device = dasd_device_from_cdev(cdev);
2299 	if (IS_ERR(device))
2300 		return PTR_ERR(device);
2301 	if (test_and_set_bit(DASD_FLAG_OFFLINE, &device->flags)) {
2302 		/* Already doing offline processing */
2303 		dasd_put_device(device);
2304 		return 0;
2305 	}
2306 	/*
2307 	 * We must make sure that this device is currently not in use.
2308 	 * The open_count is increased for every opener, that includes
2309 	 * the blkdev_get in dasd_scan_partitions. We are only interested
2310 	 * in the other openers.
2311 	 */
2312 	if (device->block) {
2313 		max_count = device->block->bdev ? 0 : -1;
2314 		open_count = atomic_read(&device->block->open_count);
2315 		if (open_count > max_count) {
2316 			if (open_count > 0)
2317 				printk(KERN_WARNING "Can't offline dasd "
2318 				       "device with open count = %i.\n",
2319 				       open_count);
2320 			else
2321 				printk(KERN_WARNING "%s",
2322 				       "Can't offline dasd device due "
2323 				       "to internal use\n");
2324 			clear_bit(DASD_FLAG_OFFLINE, &device->flags);
2325 			dasd_put_device(device);
2326 			return -EBUSY;
2327 		}
2328 	}
2329 	dasd_set_target_state(device, DASD_STATE_NEW);
2330 	/* dasd_delete_device destroys the device reference. */
2331 	block = device->block;
2332 	device->block = NULL;
2333 	dasd_delete_device(device);
2334 	/*
2335 	 * life cycle of block is bound to device, so delete it after
2336 	 * device was safely removed
2337 	 */
2338 	if (block)
2339 		dasd_free_block(block);
2340 	return 0;
2341 }
2342 
2343 int dasd_generic_notify(struct ccw_device *cdev, int event)
2344 {
2345 	struct dasd_device *device;
2346 	struct dasd_ccw_req *cqr;
2347 	int ret;
2348 
2349 	device = dasd_device_from_cdev_locked(cdev);
2350 	if (IS_ERR(device))
2351 		return 0;
2352 	ret = 0;
2353 	switch (event) {
2354 	case CIO_GONE:
2355 	case CIO_NO_PATH:
2356 		/* First of all call extended error reporting. */
2357 		dasd_eer_write(device, NULL, DASD_EER_NOPATH);
2358 
2359 		if (device->state < DASD_STATE_BASIC)
2360 			break;
2361 		/* Device is active. We want to keep it. */
2362 		list_for_each_entry(cqr, &device->ccw_queue, devlist)
2363 			if (cqr->status == DASD_CQR_IN_IO) {
2364 				cqr->status = DASD_CQR_QUEUED;
2365 				cqr->retries++;
2366 			}
2367 		device->stopped |= DASD_STOPPED_DC_WAIT;
2368 		dasd_device_clear_timer(device);
2369 		dasd_schedule_device_bh(device);
2370 		ret = 1;
2371 		break;
2372 	case CIO_OPER:
2373 		/* FIXME: add a sanity check. */
2374 		device->stopped &= ~DASD_STOPPED_DC_WAIT;
2375 		dasd_schedule_device_bh(device);
2376 		if (device->block)
2377 			dasd_schedule_block_bh(device->block);
2378 		ret = 1;
2379 		break;
2380 	}
2381 	dasd_put_device(device);
2382 	return ret;
2383 }
2384 
2385 static struct dasd_ccw_req *dasd_generic_build_rdc(struct dasd_device *device,
2386 						   void *rdc_buffer,
2387 						   int rdc_buffer_size,
2388 						   char *magic)
2389 {
2390 	struct dasd_ccw_req *cqr;
2391 	struct ccw1 *ccw;
2392 
2393 	cqr = dasd_smalloc_request(magic, 1 /* RDC */, rdc_buffer_size, device);
2394 
2395 	if (IS_ERR(cqr)) {
2396 		DEV_MESSAGE(KERN_WARNING, device, "%s",
2397 			    "Could not allocate RDC request");
2398 		return cqr;
2399 	}
2400 
2401 	ccw = cqr->cpaddr;
2402 	ccw->cmd_code = CCW_CMD_RDC;
2403 	ccw->cda = (__u32)(addr_t)rdc_buffer;
2404 	ccw->count = rdc_buffer_size;
2405 
2406 	cqr->startdev = device;
2407 	cqr->memdev = device;
2408 	cqr->expires = 10*HZ;
2409 	clear_bit(DASD_CQR_FLAGS_USE_ERP, &cqr->flags);
2410 	cqr->retries = 2;
2411 	cqr->buildclk = get_clock();
2412 	cqr->status = DASD_CQR_FILLED;
2413 	return cqr;
2414 }
2415 
2416 
2417 int dasd_generic_read_dev_chars(struct dasd_device *device, char *magic,
2418 				void **rdc_buffer, int rdc_buffer_size)
2419 {
2420 	int ret;
2421 	struct dasd_ccw_req *cqr;
2422 
2423 	cqr = dasd_generic_build_rdc(device, *rdc_buffer, rdc_buffer_size,
2424 				     magic);
2425 	if (IS_ERR(cqr))
2426 		return PTR_ERR(cqr);
2427 
2428 	ret = dasd_sleep_on(cqr);
2429 	dasd_sfree_request(cqr, cqr->memdev);
2430 	return ret;
2431 }
2432 EXPORT_SYMBOL_GPL(dasd_generic_read_dev_chars);
2433 
2434 static int __init dasd_init(void)
2435 {
2436 	int rc;
2437 
2438 	init_waitqueue_head(&dasd_init_waitq);
2439 	init_waitqueue_head(&dasd_flush_wq);
2440 	init_waitqueue_head(&generic_waitq);
2441 
2442 	/* register 'common' DASD debug area, used for all DBF_XXX calls */
2443 	dasd_debug_area = debug_register("dasd", 1, 1, 8 * sizeof(long));
2444 	if (dasd_debug_area == NULL) {
2445 		rc = -ENOMEM;
2446 		goto failed;
2447 	}
2448 	debug_register_view(dasd_debug_area, &debug_sprintf_view);
2449 	debug_set_level(dasd_debug_area, DBF_WARNING);
2450 
2451 	DBF_EVENT(DBF_EMERG, "%s", "debug area created");
2452 
2453 	dasd_diag_discipline_pointer = NULL;
2454 
2455 	rc = dasd_devmap_init();
2456 	if (rc)
2457 		goto failed;
2458 	rc = dasd_gendisk_init();
2459 	if (rc)
2460 		goto failed;
2461 	rc = dasd_parse();
2462 	if (rc)
2463 		goto failed;
2464 	rc = dasd_eer_init();
2465 	if (rc)
2466 		goto failed;
2467 #ifdef CONFIG_PROC_FS
2468 	rc = dasd_proc_init();
2469 	if (rc)
2470 		goto failed;
2471 #endif
2472 
2473 	return 0;
2474 failed:
2475 	MESSAGE(KERN_INFO, "%s", "initialization not performed due to errors");
2476 	dasd_exit();
2477 	return rc;
2478 }
2479 
2480 module_init(dasd_init);
2481 module_exit(dasd_exit);
2482 
2483 EXPORT_SYMBOL(dasd_debug_area);
2484 EXPORT_SYMBOL(dasd_diag_discipline_pointer);
2485 
2486 EXPORT_SYMBOL(dasd_add_request_head);
2487 EXPORT_SYMBOL(dasd_add_request_tail);
2488 EXPORT_SYMBOL(dasd_cancel_req);
2489 EXPORT_SYMBOL(dasd_device_clear_timer);
2490 EXPORT_SYMBOL(dasd_block_clear_timer);
2491 EXPORT_SYMBOL(dasd_enable_device);
2492 EXPORT_SYMBOL(dasd_int_handler);
2493 EXPORT_SYMBOL(dasd_kfree_request);
2494 EXPORT_SYMBOL(dasd_kick_device);
2495 EXPORT_SYMBOL(dasd_kmalloc_request);
2496 EXPORT_SYMBOL(dasd_schedule_device_bh);
2497 EXPORT_SYMBOL(dasd_schedule_block_bh);
2498 EXPORT_SYMBOL(dasd_set_target_state);
2499 EXPORT_SYMBOL(dasd_device_set_timer);
2500 EXPORT_SYMBOL(dasd_block_set_timer);
2501 EXPORT_SYMBOL(dasd_sfree_request);
2502 EXPORT_SYMBOL(dasd_sleep_on);
2503 EXPORT_SYMBOL(dasd_sleep_on_immediatly);
2504 EXPORT_SYMBOL(dasd_sleep_on_interruptible);
2505 EXPORT_SYMBOL(dasd_smalloc_request);
2506 EXPORT_SYMBOL(dasd_start_IO);
2507 EXPORT_SYMBOL(dasd_term_IO);
2508 
2509 EXPORT_SYMBOL_GPL(dasd_generic_probe);
2510 EXPORT_SYMBOL_GPL(dasd_generic_remove);
2511 EXPORT_SYMBOL_GPL(dasd_generic_notify);
2512 EXPORT_SYMBOL_GPL(dasd_generic_set_online);
2513 EXPORT_SYMBOL_GPL(dasd_generic_set_offline);
2514 EXPORT_SYMBOL_GPL(dasd_generic_handle_state_change);
2515 EXPORT_SYMBOL_GPL(dasd_flush_device_queue);
2516 EXPORT_SYMBOL_GPL(dasd_alloc_block);
2517 EXPORT_SYMBOL_GPL(dasd_free_block);
2518