s390/dasd: blk-mq conversion
[muen/linux.git] / drivers / s390 / block / dasd.c
1 /*
2  * Author(s)......: Holger Smolinski <Holger.Smolinski@de.ibm.com>
3  *                  Horst Hummel <Horst.Hummel@de.ibm.com>
4  *                  Carsten Otte <Cotte@de.ibm.com>
5  *                  Martin Schwidefsky <schwidefsky@de.ibm.com>
6  * Bugreports.to..: <Linux390@de.ibm.com>
7  * Copyright IBM Corp. 1999, 2009
8  */
9
10 #define KMSG_COMPONENT "dasd"
11 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
12
13 #include <linux/kmod.h>
14 #include <linux/init.h>
15 #include <linux/interrupt.h>
16 #include <linux/ctype.h>
17 #include <linux/major.h>
18 #include <linux/slab.h>
19 #include <linux/hdreg.h>
20 #include <linux/async.h>
21 #include <linux/mutex.h>
22 #include <linux/debugfs.h>
23 #include <linux/seq_file.h>
24 #include <linux/vmalloc.h>
25
26 #include <asm/ccwdev.h>
27 #include <asm/ebcdic.h>
28 #include <asm/idals.h>
29 #include <asm/itcw.h>
30 #include <asm/diag.h>
31
32 /* This is ugly... */
33 #define PRINTK_HEADER "dasd:"
34
35 #include "dasd_int.h"
36 /*
37  * SECTION: Constant definitions to be used within this file
38  */
39 #define DASD_CHANQ_MAX_SIZE 4
40
41 #define DASD_DIAG_MOD           "dasd_diag_mod"
42
43 /*
44  * SECTION: exported variables of dasd.c
45  */
46 debug_info_t *dasd_debug_area;
47 EXPORT_SYMBOL(dasd_debug_area);
48 static struct dentry *dasd_debugfs_root_entry;
49 struct dasd_discipline *dasd_diag_discipline_pointer;
50 EXPORT_SYMBOL(dasd_diag_discipline_pointer);
51 void dasd_int_handler(struct ccw_device *, unsigned long, struct irb *);
52
53 MODULE_AUTHOR("Holger Smolinski <Holger.Smolinski@de.ibm.com>");
54 MODULE_DESCRIPTION("Linux on S/390 DASD device driver,"
55                    " Copyright IBM Corp. 2000");
56 MODULE_SUPPORTED_DEVICE("dasd");
57 MODULE_LICENSE("GPL");
58
59 /*
60  * SECTION: prototypes for static functions of dasd.c
61  */
62 static int  dasd_alloc_queue(struct dasd_block *);
63 static void dasd_setup_queue(struct dasd_block *);
64 static void dasd_free_queue(struct dasd_block *);
65 static int dasd_flush_block_queue(struct dasd_block *);
66 static void dasd_device_tasklet(struct dasd_device *);
67 static void dasd_block_tasklet(struct dasd_block *);
68 static void do_kick_device(struct work_struct *);
69 static void do_restore_device(struct work_struct *);
70 static void do_reload_device(struct work_struct *);
71 static void do_requeue_requests(struct work_struct *);
72 static void dasd_return_cqr_cb(struct dasd_ccw_req *, void *);
73 static void dasd_device_timeout(unsigned long);
74 static void dasd_block_timeout(unsigned long);
75 static void __dasd_process_erp(struct dasd_device *, struct dasd_ccw_req *);
76 static void dasd_profile_init(struct dasd_profile *, struct dentry *);
77 static void dasd_profile_exit(struct dasd_profile *);
78 static void dasd_hosts_init(struct dentry *, struct dasd_device *);
79 static void dasd_hosts_exit(struct dasd_device *);
80
81 /*
82  * SECTION: Operations on the device structure.
83  */
84 static wait_queue_head_t dasd_init_waitq;
85 static wait_queue_head_t dasd_flush_wq;
86 static wait_queue_head_t generic_waitq;
87 static wait_queue_head_t shutdown_waitq;
88
89 /*
90  * Allocate memory for a new device structure.
91  */
92 struct dasd_device *dasd_alloc_device(void)
93 {
94         struct dasd_device *device;
95
96         device = kzalloc(sizeof(struct dasd_device), GFP_ATOMIC);
97         if (!device)
98                 return ERR_PTR(-ENOMEM);
99
100         /* Get two pages for normal block device operations. */
101         device->ccw_mem = (void *) __get_free_pages(GFP_ATOMIC | GFP_DMA, 1);
102         if (!device->ccw_mem) {
103                 kfree(device);
104                 return ERR_PTR(-ENOMEM);
105         }
106         /* Get one page for error recovery. */
107         device->erp_mem = (void *) get_zeroed_page(GFP_ATOMIC | GFP_DMA);
108         if (!device->erp_mem) {
109                 free_pages((unsigned long) device->ccw_mem, 1);
110                 kfree(device);
111                 return ERR_PTR(-ENOMEM);
112         }
113
114         dasd_init_chunklist(&device->ccw_chunks, device->ccw_mem, PAGE_SIZE*2);
115         dasd_init_chunklist(&device->erp_chunks, device->erp_mem, PAGE_SIZE);
116         spin_lock_init(&device->mem_lock);
117         atomic_set(&device->tasklet_scheduled, 0);
118         tasklet_init(&device->tasklet,
119                      (void (*)(unsigned long)) dasd_device_tasklet,
120                      (unsigned long) device);
121         INIT_LIST_HEAD(&device->ccw_queue);
122         init_timer(&device->timer);
123         device->timer.function = dasd_device_timeout;
124         device->timer.data = (unsigned long) device;
125         INIT_WORK(&device->kick_work, do_kick_device);
126         INIT_WORK(&device->restore_device, do_restore_device);
127         INIT_WORK(&device->reload_device, do_reload_device);
128         INIT_WORK(&device->requeue_requests, do_requeue_requests);
129         device->state = DASD_STATE_NEW;
130         device->target = DASD_STATE_NEW;
131         mutex_init(&device->state_mutex);
132         spin_lock_init(&device->profile.lock);
133         return device;
134 }
135
136 /*
137  * Free memory of a device structure.
138  */
139 void dasd_free_device(struct dasd_device *device)
140 {
141         kfree(device->private);
142         free_page((unsigned long) device->erp_mem);
143         free_pages((unsigned long) device->ccw_mem, 1);
144         kfree(device);
145 }
146
147 /*
148  * Allocate memory for a new device structure.
149  */
150 struct dasd_block *dasd_alloc_block(void)
151 {
152         struct dasd_block *block;
153
154         block = kzalloc(sizeof(*block), GFP_ATOMIC);
155         if (!block)
156                 return ERR_PTR(-ENOMEM);
157         /* open_count = 0 means device online but not in use */
158         atomic_set(&block->open_count, -1);
159
160         atomic_set(&block->tasklet_scheduled, 0);
161         tasklet_init(&block->tasklet,
162                      (void (*)(unsigned long)) dasd_block_tasklet,
163                      (unsigned long) block);
164         INIT_LIST_HEAD(&block->ccw_queue);
165         spin_lock_init(&block->queue_lock);
166         init_timer(&block->timer);
167         block->timer.function = dasd_block_timeout;
168         block->timer.data = (unsigned long) block;
169         spin_lock_init(&block->profile.lock);
170
171         return block;
172 }
173 EXPORT_SYMBOL_GPL(dasd_alloc_block);
174
175 /*
176  * Free memory of a device structure.
177  */
178 void dasd_free_block(struct dasd_block *block)
179 {
180         kfree(block);
181 }
182 EXPORT_SYMBOL_GPL(dasd_free_block);
183
184 /*
185  * Make a new device known to the system.
186  */
187 static int dasd_state_new_to_known(struct dasd_device *device)
188 {
189         int rc;
190
191         /*
192          * As long as the device is not in state DASD_STATE_NEW we want to
193          * keep the reference count > 0.
194          */
195         dasd_get_device(device);
196
197         if (device->block) {
198                 rc = dasd_alloc_queue(device->block);
199                 if (rc) {
200                         dasd_put_device(device);
201                         return rc;
202                 }
203         }
204         device->state = DASD_STATE_KNOWN;
205         return 0;
206 }
207
208 /*
209  * Let the system forget about a device.
210  */
211 static int dasd_state_known_to_new(struct dasd_device *device)
212 {
213         /* Disable extended error reporting for this device. */
214         dasd_eer_disable(device);
215         device->state = DASD_STATE_NEW;
216
217         if (device->block)
218                 dasd_free_queue(device->block);
219
220         /* Give up reference we took in dasd_state_new_to_known. */
221         dasd_put_device(device);
222         return 0;
223 }
224
225 static struct dentry *dasd_debugfs_setup(const char *name,
226                                          struct dentry *base_dentry)
227 {
228         struct dentry *pde;
229
230         if (!base_dentry)
231                 return NULL;
232         pde = debugfs_create_dir(name, base_dentry);
233         if (!pde || IS_ERR(pde))
234                 return NULL;
235         return pde;
236 }
237
238 /*
239  * Request the irq line for the device.
240  */
241 static int dasd_state_known_to_basic(struct dasd_device *device)
242 {
243         struct dasd_block *block = device->block;
244         int rc = 0;
245
246         /* Allocate and register gendisk structure. */
247         if (block) {
248                 rc = dasd_gendisk_alloc(block);
249                 if (rc)
250                         return rc;
251                 block->debugfs_dentry =
252                         dasd_debugfs_setup(block->gdp->disk_name,
253                                            dasd_debugfs_root_entry);
254                 dasd_profile_init(&block->profile, block->debugfs_dentry);
255                 if (dasd_global_profile_level == DASD_PROFILE_ON)
256                         dasd_profile_on(&device->block->profile);
257         }
258         device->debugfs_dentry =
259                 dasd_debugfs_setup(dev_name(&device->cdev->dev),
260                                    dasd_debugfs_root_entry);
261         dasd_profile_init(&device->profile, device->debugfs_dentry);
262         dasd_hosts_init(device->debugfs_dentry, device);
263
264         /* register 'device' debug area, used for all DBF_DEV_XXX calls */
265         device->debug_area = debug_register(dev_name(&device->cdev->dev), 4, 1,
266                                             8 * sizeof(long));
267         debug_register_view(device->debug_area, &debug_sprintf_view);
268         debug_set_level(device->debug_area, DBF_WARNING);
269         DBF_DEV_EVENT(DBF_EMERG, device, "%s", "debug area created");
270
271         device->state = DASD_STATE_BASIC;
272
273         return rc;
274 }
275
276 /*
277  * Release the irq line for the device. Terminate any running i/o.
278  */
279 static int dasd_state_basic_to_known(struct dasd_device *device)
280 {
281         int rc;
282
283         if (device->discipline->basic_to_known) {
284                 rc = device->discipline->basic_to_known(device);
285                 if (rc)
286                         return rc;
287         }
288
289         if (device->block) {
290                 dasd_profile_exit(&device->block->profile);
291                 debugfs_remove(device->block->debugfs_dentry);
292                 dasd_gendisk_free(device->block);
293                 dasd_block_clear_timer(device->block);
294         }
295         rc = dasd_flush_device_queue(device);
296         if (rc)
297                 return rc;
298         dasd_device_clear_timer(device);
299         dasd_profile_exit(&device->profile);
300         dasd_hosts_exit(device);
301         debugfs_remove(device->debugfs_dentry);
302         DBF_DEV_EVENT(DBF_EMERG, device, "%p debug area deleted", device);
303         if (device->debug_area != NULL) {
304                 debug_unregister(device->debug_area);
305                 device->debug_area = NULL;
306         }
307         device->state = DASD_STATE_KNOWN;
308         return 0;
309 }
310
311 /*
312  * Do the initial analysis. The do_analysis function may return
313  * -EAGAIN in which case the device keeps the state DASD_STATE_BASIC
314  * until the discipline decides to continue the startup sequence
315  * by calling the function dasd_change_state. The eckd disciplines
316  * uses this to start a ccw that detects the format. The completion
317  * interrupt for this detection ccw uses the kernel event daemon to
318  * trigger the call to dasd_change_state. All this is done in the
319  * discipline code, see dasd_eckd.c.
320  * After the analysis ccw is done (do_analysis returned 0) the block
321  * device is setup.
322  * In case the analysis returns an error, the device setup is stopped
323  * (a fake disk was already added to allow formatting).
324  */
325 static int dasd_state_basic_to_ready(struct dasd_device *device)
326 {
327         int rc;
328         struct dasd_block *block;
329         struct gendisk *disk;
330
331         rc = 0;
332         block = device->block;
333         /* make disk known with correct capacity */
334         if (block) {
335                 if (block->base->discipline->do_analysis != NULL)
336                         rc = block->base->discipline->do_analysis(block);
337                 if (rc) {
338                         if (rc != -EAGAIN) {
339                                 device->state = DASD_STATE_UNFMT;
340                                 disk = device->block->gdp;
341                                 kobject_uevent(&disk_to_dev(disk)->kobj,
342                                                KOBJ_CHANGE);
343                                 goto out;
344                         }
345                         return rc;
346                 }
347                 dasd_setup_queue(block);
348                 set_capacity(block->gdp,
349                              block->blocks << block->s2b_shift);
350                 device->state = DASD_STATE_READY;
351                 rc = dasd_scan_partitions(block);
352                 if (rc) {
353                         device->state = DASD_STATE_BASIC;
354                         return rc;
355                 }
356         } else {
357                 device->state = DASD_STATE_READY;
358         }
359 out:
360         if (device->discipline->basic_to_ready)
361                 rc = device->discipline->basic_to_ready(device);
362         return rc;
363 }
364
365 static inline
366 int _wait_for_empty_queues(struct dasd_device *device)
367 {
368         if (device->block)
369                 return list_empty(&device->ccw_queue) &&
370                         list_empty(&device->block->ccw_queue);
371         else
372                 return list_empty(&device->ccw_queue);
373 }
374
375 /*
376  * Remove device from block device layer. Destroy dirty buffers.
377  * Forget format information. Check if the target level is basic
378  * and if it is create fake disk for formatting.
379  */
380 static int dasd_state_ready_to_basic(struct dasd_device *device)
381 {
382         int rc;
383
384         device->state = DASD_STATE_BASIC;
385         if (device->block) {
386                 struct dasd_block *block = device->block;
387                 rc = dasd_flush_block_queue(block);
388                 if (rc) {
389                         device->state = DASD_STATE_READY;
390                         return rc;
391                 }
392                 dasd_destroy_partitions(block);
393                 block->blocks = 0;
394                 block->bp_block = 0;
395                 block->s2b_shift = 0;
396         }
397         return 0;
398 }
399
400 /*
401  * Back to basic.
402  */
403 static int dasd_state_unfmt_to_basic(struct dasd_device *device)
404 {
405         device->state = DASD_STATE_BASIC;
406         return 0;
407 }
408
409 /*
410  * Make the device online and schedule the bottom half to start
411  * the requeueing of requests from the linux request queue to the
412  * ccw queue.
413  */
414 static int
415 dasd_state_ready_to_online(struct dasd_device * device)
416 {
417         struct gendisk *disk;
418         struct disk_part_iter piter;
419         struct hd_struct *part;
420
421         device->state = DASD_STATE_ONLINE;
422         if (device->block) {
423                 dasd_schedule_block_bh(device->block);
424                 if ((device->features & DASD_FEATURE_USERAW)) {
425                         disk = device->block->gdp;
426                         kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
427                         return 0;
428                 }
429                 disk = device->block->bdev->bd_disk;
430                 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
431                 while ((part = disk_part_iter_next(&piter)))
432                         kobject_uevent(&part_to_dev(part)->kobj, KOBJ_CHANGE);
433                 disk_part_iter_exit(&piter);
434         }
435         return 0;
436 }
437
438 /*
439  * Stop the requeueing of requests again.
440  */
441 static int dasd_state_online_to_ready(struct dasd_device *device)
442 {
443         int rc;
444         struct gendisk *disk;
445         struct disk_part_iter piter;
446         struct hd_struct *part;
447
448         if (device->discipline->online_to_ready) {
449                 rc = device->discipline->online_to_ready(device);
450                 if (rc)
451                         return rc;
452         }
453
454         device->state = DASD_STATE_READY;
455         if (device->block && !(device->features & DASD_FEATURE_USERAW)) {
456                 disk = device->block->bdev->bd_disk;
457                 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
458                 while ((part = disk_part_iter_next(&piter)))
459                         kobject_uevent(&part_to_dev(part)->kobj, KOBJ_CHANGE);
460                 disk_part_iter_exit(&piter);
461         }
462         return 0;
463 }
464
465 /*
466  * Device startup state changes.
467  */
468 static int dasd_increase_state(struct dasd_device *device)
469 {
470         int rc;
471
472         rc = 0;
473         if (device->state == DASD_STATE_NEW &&
474             device->target >= DASD_STATE_KNOWN)
475                 rc = dasd_state_new_to_known(device);
476
477         if (!rc &&
478             device->state == DASD_STATE_KNOWN &&
479             device->target >= DASD_STATE_BASIC)
480                 rc = dasd_state_known_to_basic(device);
481
482         if (!rc &&
483             device->state == DASD_STATE_BASIC &&
484             device->target >= DASD_STATE_READY)
485                 rc = dasd_state_basic_to_ready(device);
486
487         if (!rc &&
488             device->state == DASD_STATE_UNFMT &&
489             device->target > DASD_STATE_UNFMT)
490                 rc = -EPERM;
491
492         if (!rc &&
493             device->state == DASD_STATE_READY &&
494             device->target >= DASD_STATE_ONLINE)
495                 rc = dasd_state_ready_to_online(device);
496
497         return rc;
498 }
499
500 /*
501  * Device shutdown state changes.
502  */
503 static int dasd_decrease_state(struct dasd_device *device)
504 {
505         int rc;
506
507         rc = 0;
508         if (device->state == DASD_STATE_ONLINE &&
509             device->target <= DASD_STATE_READY)
510                 rc = dasd_state_online_to_ready(device);
511
512         if (!rc &&
513             device->state == DASD_STATE_READY &&
514             device->target <= DASD_STATE_BASIC)
515                 rc = dasd_state_ready_to_basic(device);
516
517         if (!rc &&
518             device->state == DASD_STATE_UNFMT &&
519             device->target <= DASD_STATE_BASIC)
520                 rc = dasd_state_unfmt_to_basic(device);
521
522         if (!rc &&
523             device->state == DASD_STATE_BASIC &&
524             device->target <= DASD_STATE_KNOWN)
525                 rc = dasd_state_basic_to_known(device);
526
527         if (!rc &&
528             device->state == DASD_STATE_KNOWN &&
529             device->target <= DASD_STATE_NEW)
530                 rc = dasd_state_known_to_new(device);
531
532         return rc;
533 }
534
535 /*
536  * This is the main startup/shutdown routine.
537  */
538 static void dasd_change_state(struct dasd_device *device)
539 {
540         int rc;
541
542         if (device->state == device->target)
543                 /* Already where we want to go today... */
544                 return;
545         if (device->state < device->target)
546                 rc = dasd_increase_state(device);
547         else
548                 rc = dasd_decrease_state(device);
549         if (rc == -EAGAIN)
550                 return;
551         if (rc)
552                 device->target = device->state;
553
554         /* let user-space know that the device status changed */
555         kobject_uevent(&device->cdev->dev.kobj, KOBJ_CHANGE);
556
557         if (device->state == device->target)
558                 wake_up(&dasd_init_waitq);
559 }
560
561 /*
562  * Kick starter for devices that did not complete the startup/shutdown
563  * procedure or were sleeping because of a pending state.
564  * dasd_kick_device will schedule a call do do_kick_device to the kernel
565  * event daemon.
566  */
567 static void do_kick_device(struct work_struct *work)
568 {
569         struct dasd_device *device = container_of(work, struct dasd_device, kick_work);
570         mutex_lock(&device->state_mutex);
571         dasd_change_state(device);
572         mutex_unlock(&device->state_mutex);
573         dasd_schedule_device_bh(device);
574         dasd_put_device(device);
575 }
576
577 void dasd_kick_device(struct dasd_device *device)
578 {
579         dasd_get_device(device);
580         /* queue call to dasd_kick_device to the kernel event daemon. */
581         if (!schedule_work(&device->kick_work))
582                 dasd_put_device(device);
583 }
584 EXPORT_SYMBOL(dasd_kick_device);
585
586 /*
587  * dasd_reload_device will schedule a call do do_reload_device to the kernel
588  * event daemon.
589  */
590 static void do_reload_device(struct work_struct *work)
591 {
592         struct dasd_device *device = container_of(work, struct dasd_device,
593                                                   reload_device);
594         device->discipline->reload(device);
595         dasd_put_device(device);
596 }
597
598 void dasd_reload_device(struct dasd_device *device)
599 {
600         dasd_get_device(device);
601         /* queue call to dasd_reload_device to the kernel event daemon. */
602         if (!schedule_work(&device->reload_device))
603                 dasd_put_device(device);
604 }
605 EXPORT_SYMBOL(dasd_reload_device);
606
607 /*
608  * dasd_restore_device will schedule a call do do_restore_device to the kernel
609  * event daemon.
610  */
611 static void do_restore_device(struct work_struct *work)
612 {
613         struct dasd_device *device = container_of(work, struct dasd_device,
614                                                   restore_device);
615         device->cdev->drv->restore(device->cdev);
616         dasd_put_device(device);
617 }
618
619 void dasd_restore_device(struct dasd_device *device)
620 {
621         dasd_get_device(device);
622         /* queue call to dasd_restore_device to the kernel event daemon. */
623         if (!schedule_work(&device->restore_device))
624                 dasd_put_device(device);
625 }
626
627 /*
628  * Set the target state for a device and starts the state change.
629  */
630 void dasd_set_target_state(struct dasd_device *device, int target)
631 {
632         dasd_get_device(device);
633         mutex_lock(&device->state_mutex);
634         /* If we are in probeonly mode stop at DASD_STATE_READY. */
635         if (dasd_probeonly && target > DASD_STATE_READY)
636                 target = DASD_STATE_READY;
637         if (device->target != target) {
638                 if (device->state == target)
639                         wake_up(&dasd_init_waitq);
640                 device->target = target;
641         }
642         if (device->state != device->target)
643                 dasd_change_state(device);
644         mutex_unlock(&device->state_mutex);
645         dasd_put_device(device);
646 }
647 EXPORT_SYMBOL(dasd_set_target_state);
648
649 /*
650  * Enable devices with device numbers in [from..to].
651  */
652 static inline int _wait_for_device(struct dasd_device *device)
653 {
654         return (device->state == device->target);
655 }
656
657 void dasd_enable_device(struct dasd_device *device)
658 {
659         dasd_set_target_state(device, DASD_STATE_ONLINE);
660         if (device->state <= DASD_STATE_KNOWN)
661                 /* No discipline for device found. */
662                 dasd_set_target_state(device, DASD_STATE_NEW);
663         /* Now wait for the devices to come up. */
664         wait_event(dasd_init_waitq, _wait_for_device(device));
665
666         dasd_reload_device(device);
667         if (device->discipline->kick_validate)
668                 device->discipline->kick_validate(device);
669 }
670 EXPORT_SYMBOL(dasd_enable_device);
671
672 /*
673  * SECTION: device operation (interrupt handler, start i/o, term i/o ...)
674  */
675
676 unsigned int dasd_global_profile_level = DASD_PROFILE_OFF;
677
678 #ifdef CONFIG_DASD_PROFILE
679 struct dasd_profile dasd_global_profile = {
680         .lock = __SPIN_LOCK_UNLOCKED(dasd_global_profile.lock),
681 };
682 static struct dentry *dasd_debugfs_global_entry;
683
684 /*
685  * Add profiling information for cqr before execution.
686  */
687 static void dasd_profile_start(struct dasd_block *block,
688                                struct dasd_ccw_req *cqr,
689                                struct request *req)
690 {
691         struct list_head *l;
692         unsigned int counter;
693         struct dasd_device *device;
694
695         /* count the length of the chanq for statistics */
696         counter = 0;
697         if (dasd_global_profile_level || block->profile.data)
698                 list_for_each(l, &block->ccw_queue)
699                         if (++counter >= 31)
700                                 break;
701
702         spin_lock(&dasd_global_profile.lock);
703         if (dasd_global_profile.data) {
704                 dasd_global_profile.data->dasd_io_nr_req[counter]++;
705                 if (rq_data_dir(req) == READ)
706                         dasd_global_profile.data->dasd_read_nr_req[counter]++;
707         }
708         spin_unlock(&dasd_global_profile.lock);
709
710         spin_lock(&block->profile.lock);
711         if (block->profile.data) {
712                 block->profile.data->dasd_io_nr_req[counter]++;
713                 if (rq_data_dir(req) == READ)
714                         block->profile.data->dasd_read_nr_req[counter]++;
715         }
716         spin_unlock(&block->profile.lock);
717
718         /*
719          * We count the request for the start device, even though it may run on
720          * some other device due to error recovery. This way we make sure that
721          * we count each request only once.
722          */
723         device = cqr->startdev;
724         if (device->profile.data) {
725                 counter = 1; /* request is not yet queued on the start device */
726                 list_for_each(l, &device->ccw_queue)
727                         if (++counter >= 31)
728                                 break;
729         }
730         spin_lock(&device->profile.lock);
731         if (device->profile.data) {
732                 device->profile.data->dasd_io_nr_req[counter]++;
733                 if (rq_data_dir(req) == READ)
734                         device->profile.data->dasd_read_nr_req[counter]++;
735         }
736         spin_unlock(&device->profile.lock);
737 }
738
739 /*
740  * Add profiling information for cqr after execution.
741  */
742
743 #define dasd_profile_counter(value, index)                         \
744 {                                                                  \
745         for (index = 0; index < 31 && value >> (2+index); index++) \
746                 ;                                                  \
747 }
748
749 static void dasd_profile_end_add_data(struct dasd_profile_info *data,
750                                       int is_alias,
751                                       int is_tpm,
752                                       int is_read,
753                                       long sectors,
754                                       int sectors_ind,
755                                       int tottime_ind,
756                                       int tottimeps_ind,
757                                       int strtime_ind,
758                                       int irqtime_ind,
759                                       int irqtimeps_ind,
760                                       int endtime_ind)
761 {
762         /* in case of an overflow, reset the whole profile */
763         if (data->dasd_io_reqs == UINT_MAX) {
764                         memset(data, 0, sizeof(*data));
765                         getnstimeofday(&data->starttod);
766         }
767         data->dasd_io_reqs++;
768         data->dasd_io_sects += sectors;
769         if (is_alias)
770                 data->dasd_io_alias++;
771         if (is_tpm)
772                 data->dasd_io_tpm++;
773
774         data->dasd_io_secs[sectors_ind]++;
775         data->dasd_io_times[tottime_ind]++;
776         data->dasd_io_timps[tottimeps_ind]++;
777         data->dasd_io_time1[strtime_ind]++;
778         data->dasd_io_time2[irqtime_ind]++;
779         data->dasd_io_time2ps[irqtimeps_ind]++;
780         data->dasd_io_time3[endtime_ind]++;
781
782         if (is_read) {
783                 data->dasd_read_reqs++;
784                 data->dasd_read_sects += sectors;
785                 if (is_alias)
786                         data->dasd_read_alias++;
787                 if (is_tpm)
788                         data->dasd_read_tpm++;
789                 data->dasd_read_secs[sectors_ind]++;
790                 data->dasd_read_times[tottime_ind]++;
791                 data->dasd_read_time1[strtime_ind]++;
792                 data->dasd_read_time2[irqtime_ind]++;
793                 data->dasd_read_time3[endtime_ind]++;
794         }
795 }
796
797 static void dasd_profile_end(struct dasd_block *block,
798                              struct dasd_ccw_req *cqr,
799                              struct request *req)
800 {
801         unsigned long strtime, irqtime, endtime, tottime;
802         unsigned long tottimeps, sectors;
803         struct dasd_device *device;
804         int sectors_ind, tottime_ind, tottimeps_ind, strtime_ind;
805         int irqtime_ind, irqtimeps_ind, endtime_ind;
806         struct dasd_profile_info *data;
807
808         device = cqr->startdev;
809         if (!(dasd_global_profile_level ||
810               block->profile.data ||
811               device->profile.data))
812                 return;
813
814         sectors = blk_rq_sectors(req);
815         if (!cqr->buildclk || !cqr->startclk ||
816             !cqr->stopclk || !cqr->endclk ||
817             !sectors)
818                 return;
819
820         strtime = ((cqr->startclk - cqr->buildclk) >> 12);
821         irqtime = ((cqr->stopclk - cqr->startclk) >> 12);
822         endtime = ((cqr->endclk - cqr->stopclk) >> 12);
823         tottime = ((cqr->endclk - cqr->buildclk) >> 12);
824         tottimeps = tottime / sectors;
825
826         dasd_profile_counter(sectors, sectors_ind);
827         dasd_profile_counter(tottime, tottime_ind);
828         dasd_profile_counter(tottimeps, tottimeps_ind);
829         dasd_profile_counter(strtime, strtime_ind);
830         dasd_profile_counter(irqtime, irqtime_ind);
831         dasd_profile_counter(irqtime / sectors, irqtimeps_ind);
832         dasd_profile_counter(endtime, endtime_ind);
833
834         spin_lock(&dasd_global_profile.lock);
835         if (dasd_global_profile.data) {
836                 data = dasd_global_profile.data;
837                 data->dasd_sum_times += tottime;
838                 data->dasd_sum_time_str += strtime;
839                 data->dasd_sum_time_irq += irqtime;
840                 data->dasd_sum_time_end += endtime;
841                 dasd_profile_end_add_data(dasd_global_profile.data,
842                                           cqr->startdev != block->base,
843                                           cqr->cpmode == 1,
844                                           rq_data_dir(req) == READ,
845                                           sectors, sectors_ind, tottime_ind,
846                                           tottimeps_ind, strtime_ind,
847                                           irqtime_ind, irqtimeps_ind,
848                                           endtime_ind);
849         }
850         spin_unlock(&dasd_global_profile.lock);
851
852         spin_lock(&block->profile.lock);
853         if (block->profile.data) {
854                 data = block->profile.data;
855                 data->dasd_sum_times += tottime;
856                 data->dasd_sum_time_str += strtime;
857                 data->dasd_sum_time_irq += irqtime;
858                 data->dasd_sum_time_end += endtime;
859                 dasd_profile_end_add_data(block->profile.data,
860                                           cqr->startdev != block->base,
861                                           cqr->cpmode == 1,
862                                           rq_data_dir(req) == READ,
863                                           sectors, sectors_ind, tottime_ind,
864                                           tottimeps_ind, strtime_ind,
865                                           irqtime_ind, irqtimeps_ind,
866                                           endtime_ind);
867         }
868         spin_unlock(&block->profile.lock);
869
870         spin_lock(&device->profile.lock);
871         if (device->profile.data) {
872                 data = device->profile.data;
873                 data->dasd_sum_times += tottime;
874                 data->dasd_sum_time_str += strtime;
875                 data->dasd_sum_time_irq += irqtime;
876                 data->dasd_sum_time_end += endtime;
877                 dasd_profile_end_add_data(device->profile.data,
878                                           cqr->startdev != block->base,
879                                           cqr->cpmode == 1,
880                                           rq_data_dir(req) == READ,
881                                           sectors, sectors_ind, tottime_ind,
882                                           tottimeps_ind, strtime_ind,
883                                           irqtime_ind, irqtimeps_ind,
884                                           endtime_ind);
885         }
886         spin_unlock(&device->profile.lock);
887 }
888
889 void dasd_profile_reset(struct dasd_profile *profile)
890 {
891         struct dasd_profile_info *data;
892
893         spin_lock_bh(&profile->lock);
894         data = profile->data;
895         if (!data) {
896                 spin_unlock_bh(&profile->lock);
897                 return;
898         }
899         memset(data, 0, sizeof(*data));
900         getnstimeofday(&data->starttod);
901         spin_unlock_bh(&profile->lock);
902 }
903
904 int dasd_profile_on(struct dasd_profile *profile)
905 {
906         struct dasd_profile_info *data;
907
908         data = kzalloc(sizeof(*data), GFP_KERNEL);
909         if (!data)
910                 return -ENOMEM;
911         spin_lock_bh(&profile->lock);
912         if (profile->data) {
913                 spin_unlock_bh(&profile->lock);
914                 kfree(data);
915                 return 0;
916         }
917         getnstimeofday(&data->starttod);
918         profile->data = data;
919         spin_unlock_bh(&profile->lock);
920         return 0;
921 }
922
923 void dasd_profile_off(struct dasd_profile *profile)
924 {
925         spin_lock_bh(&profile->lock);
926         kfree(profile->data);
927         profile->data = NULL;
928         spin_unlock_bh(&profile->lock);
929 }
930
931 char *dasd_get_user_string(const char __user *user_buf, size_t user_len)
932 {
933         char *buffer;
934
935         buffer = vmalloc(user_len + 1);
936         if (buffer == NULL)
937                 return ERR_PTR(-ENOMEM);
938         if (copy_from_user(buffer, user_buf, user_len) != 0) {
939                 vfree(buffer);
940                 return ERR_PTR(-EFAULT);
941         }
942         /* got the string, now strip linefeed. */
943         if (buffer[user_len - 1] == '\n')
944                 buffer[user_len - 1] = 0;
945         else
946                 buffer[user_len] = 0;
947         return buffer;
948 }
949
950 static ssize_t dasd_stats_write(struct file *file,
951                                 const char __user *user_buf,
952                                 size_t user_len, loff_t *pos)
953 {
954         char *buffer, *str;
955         int rc;
956         struct seq_file *m = (struct seq_file *)file->private_data;
957         struct dasd_profile *prof = m->private;
958
959         if (user_len > 65536)
960                 user_len = 65536;
961         buffer = dasd_get_user_string(user_buf, user_len);
962         if (IS_ERR(buffer))
963                 return PTR_ERR(buffer);
964
965         str = skip_spaces(buffer);
966         rc = user_len;
967         if (strncmp(str, "reset", 5) == 0) {
968                 dasd_profile_reset(prof);
969         } else if (strncmp(str, "on", 2) == 0) {
970                 rc = dasd_profile_on(prof);
971                 if (rc)
972                         goto out;
973                 rc = user_len;
974                 if (prof == &dasd_global_profile) {
975                         dasd_profile_reset(prof);
976                         dasd_global_profile_level = DASD_PROFILE_GLOBAL_ONLY;
977                 }
978         } else if (strncmp(str, "off", 3) == 0) {
979                 if (prof == &dasd_global_profile)
980                         dasd_global_profile_level = DASD_PROFILE_OFF;
981                 dasd_profile_off(prof);
982         } else
983                 rc = -EINVAL;
984 out:
985         vfree(buffer);
986         return rc;
987 }
988
989 static void dasd_stats_array(struct seq_file *m, unsigned int *array)
990 {
991         int i;
992
993         for (i = 0; i < 32; i++)
994                 seq_printf(m, "%u ", array[i]);
995         seq_putc(m, '\n');
996 }
997
998 static void dasd_stats_seq_print(struct seq_file *m,
999                                  struct dasd_profile_info *data)
1000 {
1001         seq_printf(m, "start_time %ld.%09ld\n",
1002                    data->starttod.tv_sec, data->starttod.tv_nsec);
1003         seq_printf(m, "total_requests %u\n", data->dasd_io_reqs);
1004         seq_printf(m, "total_sectors %u\n", data->dasd_io_sects);
1005         seq_printf(m, "total_pav %u\n", data->dasd_io_alias);
1006         seq_printf(m, "total_hpf %u\n", data->dasd_io_tpm);
1007         seq_printf(m, "avg_total %lu\n", data->dasd_io_reqs ?
1008                    data->dasd_sum_times / data->dasd_io_reqs : 0UL);
1009         seq_printf(m, "avg_build_to_ssch %lu\n", data->dasd_io_reqs ?
1010                    data->dasd_sum_time_str / data->dasd_io_reqs : 0UL);
1011         seq_printf(m, "avg_ssch_to_irq %lu\n", data->dasd_io_reqs ?
1012                    data->dasd_sum_time_irq / data->dasd_io_reqs : 0UL);
1013         seq_printf(m, "avg_irq_to_end %lu\n", data->dasd_io_reqs ?
1014                    data->dasd_sum_time_end / data->dasd_io_reqs : 0UL);
1015         seq_puts(m, "histogram_sectors ");
1016         dasd_stats_array(m, data->dasd_io_secs);
1017         seq_puts(m, "histogram_io_times ");
1018         dasd_stats_array(m, data->dasd_io_times);
1019         seq_puts(m, "histogram_io_times_weighted ");
1020         dasd_stats_array(m, data->dasd_io_timps);
1021         seq_puts(m, "histogram_time_build_to_ssch ");
1022         dasd_stats_array(m, data->dasd_io_time1);
1023         seq_puts(m, "histogram_time_ssch_to_irq ");
1024         dasd_stats_array(m, data->dasd_io_time2);
1025         seq_puts(m, "histogram_time_ssch_to_irq_weighted ");
1026         dasd_stats_array(m, data->dasd_io_time2ps);
1027         seq_puts(m, "histogram_time_irq_to_end ");
1028         dasd_stats_array(m, data->dasd_io_time3);
1029         seq_puts(m, "histogram_ccw_queue_length ");
1030         dasd_stats_array(m, data->dasd_io_nr_req);
1031         seq_printf(m, "total_read_requests %u\n", data->dasd_read_reqs);
1032         seq_printf(m, "total_read_sectors %u\n", data->dasd_read_sects);
1033         seq_printf(m, "total_read_pav %u\n", data->dasd_read_alias);
1034         seq_printf(m, "total_read_hpf %u\n", data->dasd_read_tpm);
1035         seq_puts(m, "histogram_read_sectors ");
1036         dasd_stats_array(m, data->dasd_read_secs);
1037         seq_puts(m, "histogram_read_times ");
1038         dasd_stats_array(m, data->dasd_read_times);
1039         seq_puts(m, "histogram_read_time_build_to_ssch ");
1040         dasd_stats_array(m, data->dasd_read_time1);
1041         seq_puts(m, "histogram_read_time_ssch_to_irq ");
1042         dasd_stats_array(m, data->dasd_read_time2);
1043         seq_puts(m, "histogram_read_time_irq_to_end ");
1044         dasd_stats_array(m, data->dasd_read_time3);
1045         seq_puts(m, "histogram_read_ccw_queue_length ");
1046         dasd_stats_array(m, data->dasd_read_nr_req);
1047 }
1048
1049 static int dasd_stats_show(struct seq_file *m, void *v)
1050 {
1051         struct dasd_profile *profile;
1052         struct dasd_profile_info *data;
1053
1054         profile = m->private;
1055         spin_lock_bh(&profile->lock);
1056         data = profile->data;
1057         if (!data) {
1058                 spin_unlock_bh(&profile->lock);
1059                 seq_puts(m, "disabled\n");
1060                 return 0;
1061         }
1062         dasd_stats_seq_print(m, data);
1063         spin_unlock_bh(&profile->lock);
1064         return 0;
1065 }
1066
1067 static int dasd_stats_open(struct inode *inode, struct file *file)
1068 {
1069         struct dasd_profile *profile = inode->i_private;
1070         return single_open(file, dasd_stats_show, profile);
1071 }
1072
1073 static const struct file_operations dasd_stats_raw_fops = {
1074         .owner          = THIS_MODULE,
1075         .open           = dasd_stats_open,
1076         .read           = seq_read,
1077         .llseek         = seq_lseek,
1078         .release        = single_release,
1079         .write          = dasd_stats_write,
1080 };
1081
1082 static void dasd_profile_init(struct dasd_profile *profile,
1083                               struct dentry *base_dentry)
1084 {
1085         umode_t mode;
1086         struct dentry *pde;
1087
1088         if (!base_dentry)
1089                 return;
1090         profile->dentry = NULL;
1091         profile->data = NULL;
1092         mode = (S_IRUSR | S_IWUSR | S_IFREG);
1093         pde = debugfs_create_file("statistics", mode, base_dentry,
1094                                   profile, &dasd_stats_raw_fops);
1095         if (pde && !IS_ERR(pde))
1096                 profile->dentry = pde;
1097         return;
1098 }
1099
1100 static void dasd_profile_exit(struct dasd_profile *profile)
1101 {
1102         dasd_profile_off(profile);
1103         debugfs_remove(profile->dentry);
1104         profile->dentry = NULL;
1105 }
1106
1107 static void dasd_statistics_removeroot(void)
1108 {
1109         dasd_global_profile_level = DASD_PROFILE_OFF;
1110         dasd_profile_exit(&dasd_global_profile);
1111         debugfs_remove(dasd_debugfs_global_entry);
1112         debugfs_remove(dasd_debugfs_root_entry);
1113 }
1114
1115 static void dasd_statistics_createroot(void)
1116 {
1117         struct dentry *pde;
1118
1119         dasd_debugfs_root_entry = NULL;
1120         pde = debugfs_create_dir("dasd", NULL);
1121         if (!pde || IS_ERR(pde))
1122                 goto error;
1123         dasd_debugfs_root_entry = pde;
1124         pde = debugfs_create_dir("global", dasd_debugfs_root_entry);
1125         if (!pde || IS_ERR(pde))
1126                 goto error;
1127         dasd_debugfs_global_entry = pde;
1128         dasd_profile_init(&dasd_global_profile, dasd_debugfs_global_entry);
1129         return;
1130
1131 error:
1132         DBF_EVENT(DBF_ERR, "%s",
1133                   "Creation of the dasd debugfs interface failed");
1134         dasd_statistics_removeroot();
1135         return;
1136 }
1137
1138 #else
1139 #define dasd_profile_start(block, cqr, req) do {} while (0)
1140 #define dasd_profile_end(block, cqr, req) do {} while (0)
1141
1142 static void dasd_statistics_createroot(void)
1143 {
1144         return;
1145 }
1146
1147 static void dasd_statistics_removeroot(void)
1148 {
1149         return;
1150 }
1151
1152 int dasd_stats_generic_show(struct seq_file *m, void *v)
1153 {
1154         seq_puts(m, "Statistics are not activated in this kernel\n");
1155         return 0;
1156 }
1157
1158 static void dasd_profile_init(struct dasd_profile *profile,
1159                               struct dentry *base_dentry)
1160 {
1161         return;
1162 }
1163
1164 static void dasd_profile_exit(struct dasd_profile *profile)
1165 {
1166         return;
1167 }
1168
1169 int dasd_profile_on(struct dasd_profile *profile)
1170 {
1171         return 0;
1172 }
1173
1174 #endif                          /* CONFIG_DASD_PROFILE */
1175
1176 static int dasd_hosts_show(struct seq_file *m, void *v)
1177 {
1178         struct dasd_device *device;
1179         int rc = -EOPNOTSUPP;
1180
1181         device = m->private;
1182         dasd_get_device(device);
1183
1184         if (device->discipline->hosts_print)
1185                 rc = device->discipline->hosts_print(device, m);
1186
1187         dasd_put_device(device);
1188         return rc;
1189 }
1190
1191 static int dasd_hosts_open(struct inode *inode, struct file *file)
1192 {
1193         struct dasd_device *device = inode->i_private;
1194
1195         return single_open(file, dasd_hosts_show, device);
1196 }
1197
1198 static const struct file_operations dasd_hosts_fops = {
1199         .owner          = THIS_MODULE,
1200         .open           = dasd_hosts_open,
1201         .read           = seq_read,
1202         .llseek         = seq_lseek,
1203         .release        = single_release,
1204 };
1205
1206 static void dasd_hosts_exit(struct dasd_device *device)
1207 {
1208         debugfs_remove(device->hosts_dentry);
1209         device->hosts_dentry = NULL;
1210 }
1211
1212 static void dasd_hosts_init(struct dentry *base_dentry,
1213                             struct dasd_device *device)
1214 {
1215         struct dentry *pde;
1216         umode_t mode;
1217
1218         if (!base_dentry)
1219                 return;
1220
1221         mode = S_IRUSR | S_IFREG;
1222         pde = debugfs_create_file("host_access_list", mode, base_dentry,
1223                                   device, &dasd_hosts_fops);
1224         if (pde && !IS_ERR(pde))
1225                 device->hosts_dentry = pde;
1226 }
1227
1228 /*
1229  * Allocate memory for a channel program with 'cplength' channel
1230  * command words and 'datasize' additional space. There are two
1231  * variantes: 1) dasd_kmalloc_request uses kmalloc to get the needed
1232  * memory and 2) dasd_smalloc_request uses the static ccw memory
1233  * that gets allocated for each device.
1234  */
1235 struct dasd_ccw_req *dasd_kmalloc_request(int magic, int cplength,
1236                                           int datasize,
1237                                           struct dasd_device *device)
1238 {
1239         struct dasd_ccw_req *cqr;
1240
1241         /* Sanity checks */
1242         BUG_ON(datasize > PAGE_SIZE ||
1243              (cplength*sizeof(struct ccw1)) > PAGE_SIZE);
1244
1245         cqr = kzalloc(sizeof(struct dasd_ccw_req), GFP_ATOMIC);
1246         if (cqr == NULL)
1247                 return ERR_PTR(-ENOMEM);
1248         cqr->cpaddr = NULL;
1249         if (cplength > 0) {
1250                 cqr->cpaddr = kcalloc(cplength, sizeof(struct ccw1),
1251                                       GFP_ATOMIC | GFP_DMA);
1252                 if (cqr->cpaddr == NULL) {
1253                         kfree(cqr);
1254                         return ERR_PTR(-ENOMEM);
1255                 }
1256         }
1257         cqr->data = NULL;
1258         if (datasize > 0) {
1259                 cqr->data = kzalloc(datasize, GFP_ATOMIC | GFP_DMA);
1260                 if (cqr->data == NULL) {
1261                         kfree(cqr->cpaddr);
1262                         kfree(cqr);
1263                         return ERR_PTR(-ENOMEM);
1264                 }
1265         }
1266         cqr->magic =  magic;
1267         set_bit(DASD_CQR_FLAGS_USE_ERP, &cqr->flags);
1268         dasd_get_device(device);
1269         return cqr;
1270 }
1271 EXPORT_SYMBOL(dasd_kmalloc_request);
1272
1273 struct dasd_ccw_req *dasd_smalloc_request(int magic, int cplength,
1274                                           int datasize,
1275                                           struct dasd_device *device)
1276 {
1277         unsigned long flags;
1278         struct dasd_ccw_req *cqr;
1279         char *data;
1280         int size;
1281
1282         size = (sizeof(struct dasd_ccw_req) + 7L) & -8L;
1283         if (cplength > 0)
1284                 size += cplength * sizeof(struct ccw1);
1285         if (datasize > 0)
1286                 size += datasize;
1287         spin_lock_irqsave(&device->mem_lock, flags);
1288         cqr = (struct dasd_ccw_req *)
1289                 dasd_alloc_chunk(&device->ccw_chunks, size);
1290         spin_unlock_irqrestore(&device->mem_lock, flags);
1291         if (cqr == NULL)
1292                 return ERR_PTR(-ENOMEM);
1293         memset(cqr, 0, sizeof(struct dasd_ccw_req));
1294         data = (char *) cqr + ((sizeof(struct dasd_ccw_req) + 7L) & -8L);
1295         cqr->cpaddr = NULL;
1296         if (cplength > 0) {
1297                 cqr->cpaddr = (struct ccw1 *) data;
1298                 data += cplength*sizeof(struct ccw1);
1299                 memset(cqr->cpaddr, 0, cplength*sizeof(struct ccw1));
1300         }
1301         cqr->data = NULL;
1302         if (datasize > 0) {
1303                 cqr->data = data;
1304                 memset(cqr->data, 0, datasize);
1305         }
1306         cqr->magic = magic;
1307         set_bit(DASD_CQR_FLAGS_USE_ERP, &cqr->flags);
1308         dasd_get_device(device);
1309         return cqr;
1310 }
1311 EXPORT_SYMBOL(dasd_smalloc_request);
1312
1313 /*
1314  * Free memory of a channel program. This function needs to free all the
1315  * idal lists that might have been created by dasd_set_cda and the
1316  * struct dasd_ccw_req itself.
1317  */
1318 void dasd_kfree_request(struct dasd_ccw_req *cqr, struct dasd_device *device)
1319 {
1320         struct ccw1 *ccw;
1321
1322         /* Clear any idals used for the request. */
1323         ccw = cqr->cpaddr;
1324         do {
1325                 clear_normalized_cda(ccw);
1326         } while (ccw++->flags & (CCW_FLAG_CC | CCW_FLAG_DC));
1327         kfree(cqr->cpaddr);
1328         kfree(cqr->data);
1329         kfree(cqr);
1330         dasd_put_device(device);
1331 }
1332 EXPORT_SYMBOL(dasd_kfree_request);
1333
1334 void dasd_sfree_request(struct dasd_ccw_req *cqr, struct dasd_device *device)
1335 {
1336         unsigned long flags;
1337
1338         spin_lock_irqsave(&device->mem_lock, flags);
1339         dasd_free_chunk(&device->ccw_chunks, cqr);
1340         spin_unlock_irqrestore(&device->mem_lock, flags);
1341         dasd_put_device(device);
1342 }
1343 EXPORT_SYMBOL(dasd_sfree_request);
1344
1345 /*
1346  * Check discipline magic in cqr.
1347  */
1348 static inline int dasd_check_cqr(struct dasd_ccw_req *cqr)
1349 {
1350         struct dasd_device *device;
1351
1352         if (cqr == NULL)
1353                 return -EINVAL;
1354         device = cqr->startdev;
1355         if (strncmp((char *) &cqr->magic, device->discipline->ebcname, 4)) {
1356                 DBF_DEV_EVENT(DBF_WARNING, device,
1357                             " dasd_ccw_req 0x%08x magic doesn't match"
1358                             " discipline 0x%08x",
1359                             cqr->magic,
1360                             *(unsigned int *) device->discipline->name);
1361                 return -EINVAL;
1362         }
1363         return 0;
1364 }
1365
1366 /*
1367  * Terminate the current i/o and set the request to clear_pending.
1368  * Timer keeps device runnig.
1369  * ccw_device_clear can fail if the i/o subsystem
1370  * is in a bad mood.
1371  */
1372 int dasd_term_IO(struct dasd_ccw_req *cqr)
1373 {
1374         struct dasd_device *device;
1375         int retries, rc;
1376         char errorstring[ERRORLENGTH];
1377
1378         /* Check the cqr */
1379         rc = dasd_check_cqr(cqr);
1380         if (rc)
1381                 return rc;
1382         retries = 0;
1383         device = (struct dasd_device *) cqr->startdev;
1384         while ((retries < 5) && (cqr->status == DASD_CQR_IN_IO)) {
1385                 rc = ccw_device_clear(device->cdev, (long) cqr);
1386                 switch (rc) {
1387                 case 0: /* termination successful */
1388                         cqr->status = DASD_CQR_CLEAR_PENDING;
1389                         cqr->stopclk = get_tod_clock();
1390                         cqr->starttime = 0;
1391                         DBF_DEV_EVENT(DBF_DEBUG, device,
1392                                       "terminate cqr %p successful",
1393                                       cqr);
1394                         break;
1395                 case -ENODEV:
1396                         DBF_DEV_EVENT(DBF_ERR, device, "%s",
1397                                       "device gone, retry");
1398                         break;
1399                 case -EIO:
1400                         DBF_DEV_EVENT(DBF_ERR, device, "%s",
1401                                       "I/O error, retry");
1402                         break;
1403                 case -EINVAL:
1404                         /*
1405                          * device not valid so no I/O could be running
1406                          * handle CQR as termination successful
1407                          */
1408                         cqr->status = DASD_CQR_CLEARED;
1409                         cqr->stopclk = get_tod_clock();
1410                         cqr->starttime = 0;
1411                         /* no retries for invalid devices */
1412                         cqr->retries = -1;
1413                         DBF_DEV_EVENT(DBF_ERR, device, "%s",
1414                                       "EINVAL, handle as terminated");
1415                         /* fake rc to success */
1416                         rc = 0;
1417                         break;
1418                 case -EBUSY:
1419                         DBF_DEV_EVENT(DBF_ERR, device, "%s",
1420                                       "device busy, retry later");
1421                         break;
1422                 default:
1423                         /* internal error 10 - unknown rc*/
1424                         snprintf(errorstring, ERRORLENGTH, "10 %d", rc);
1425                         dev_err(&device->cdev->dev, "An error occurred in the "
1426                                 "DASD device driver, reason=%s\n", errorstring);
1427                         BUG();
1428                         break;
1429                 }
1430                 retries++;
1431         }
1432         dasd_schedule_device_bh(device);
1433         return rc;
1434 }
1435 EXPORT_SYMBOL(dasd_term_IO);
1436
1437 /*
1438  * Start the i/o. This start_IO can fail if the channel is really busy.
1439  * In that case set up a timer to start the request later.
1440  */
1441 int dasd_start_IO(struct dasd_ccw_req *cqr)
1442 {
1443         struct dasd_device *device;
1444         int rc;
1445         char errorstring[ERRORLENGTH];
1446
1447         /* Check the cqr */
1448         rc = dasd_check_cqr(cqr);
1449         if (rc) {
1450                 cqr->intrc = rc;
1451                 return rc;
1452         }
1453         device = (struct dasd_device *) cqr->startdev;
1454         if (((cqr->block &&
1455               test_bit(DASD_FLAG_LOCK_STOLEN, &cqr->block->base->flags)) ||
1456              test_bit(DASD_FLAG_LOCK_STOLEN, &device->flags)) &&
1457             !test_bit(DASD_CQR_ALLOW_SLOCK, &cqr->flags)) {
1458                 DBF_DEV_EVENT(DBF_DEBUG, device, "start_IO: return request %p "
1459                               "because of stolen lock", cqr);
1460                 cqr->status = DASD_CQR_ERROR;
1461                 cqr->intrc = -EPERM;
1462                 return -EPERM;
1463         }
1464         if (cqr->retries < 0) {
1465                 /* internal error 14 - start_IO run out of retries */
1466                 sprintf(errorstring, "14 %p", cqr);
1467                 dev_err(&device->cdev->dev, "An error occurred in the DASD "
1468                         "device driver, reason=%s\n", errorstring);
1469                 cqr->status = DASD_CQR_ERROR;
1470                 return -EIO;
1471         }
1472         cqr->startclk = get_tod_clock();
1473         cqr->starttime = jiffies;
1474         cqr->retries--;
1475         if (!test_bit(DASD_CQR_VERIFY_PATH, &cqr->flags)) {
1476                 cqr->lpm &= dasd_path_get_opm(device);
1477                 if (!cqr->lpm)
1478                         cqr->lpm = dasd_path_get_opm(device);
1479         }
1480         if (cqr->cpmode == 1) {
1481                 rc = ccw_device_tm_start(device->cdev, cqr->cpaddr,
1482                                          (long) cqr, cqr->lpm);
1483         } else {
1484                 rc = ccw_device_start(device->cdev, cqr->cpaddr,
1485                                       (long) cqr, cqr->lpm, 0);
1486         }
1487         switch (rc) {
1488         case 0:
1489                 cqr->status = DASD_CQR_IN_IO;
1490                 break;
1491         case -EBUSY:
1492                 DBF_DEV_EVENT(DBF_WARNING, device, "%s",
1493                               "start_IO: device busy, retry later");
1494                 break;
1495         case -ETIMEDOUT:
1496                 DBF_DEV_EVENT(DBF_WARNING, device, "%s",
1497                               "start_IO: request timeout, retry later");
1498                 break;
1499         case -EACCES:
1500                 /* -EACCES indicates that the request used only a subset of the
1501                  * available paths and all these paths are gone. If the lpm of
1502                  * this request was only a subset of the opm (e.g. the ppm) then
1503                  * we just do a retry with all available paths.
1504                  * If we already use the full opm, something is amiss, and we
1505                  * need a full path verification.
1506                  */
1507                 if (test_bit(DASD_CQR_VERIFY_PATH, &cqr->flags)) {
1508                         DBF_DEV_EVENT(DBF_WARNING, device,
1509                                       "start_IO: selected paths gone (%x)",
1510                                       cqr->lpm);
1511                 } else if (cqr->lpm != dasd_path_get_opm(device)) {
1512                         cqr->lpm = dasd_path_get_opm(device);
1513                         DBF_DEV_EVENT(DBF_DEBUG, device, "%s",
1514                                       "start_IO: selected paths gone,"
1515                                       " retry on all paths");
1516                 } else {
1517                         DBF_DEV_EVENT(DBF_WARNING, device, "%s",
1518                                       "start_IO: all paths in opm gone,"
1519                                       " do path verification");
1520                         dasd_generic_last_path_gone(device);
1521                         dasd_path_no_path(device);
1522                         dasd_path_set_tbvpm(device,
1523                                           ccw_device_get_path_mask(
1524                                                   device->cdev));
1525                 }
1526                 break;
1527         case -ENODEV:
1528                 DBF_DEV_EVENT(DBF_WARNING, device, "%s",
1529                               "start_IO: -ENODEV device gone, retry");
1530                 break;
1531         case -EIO:
1532                 DBF_DEV_EVENT(DBF_WARNING, device, "%s",
1533                               "start_IO: -EIO device gone, retry");
1534                 break;
1535         case -EINVAL:
1536                 /* most likely caused in power management context */
1537                 DBF_DEV_EVENT(DBF_WARNING, device, "%s",
1538                               "start_IO: -EINVAL device currently "
1539                               "not accessible");
1540                 break;
1541         default:
1542                 /* internal error 11 - unknown rc */
1543                 snprintf(errorstring, ERRORLENGTH, "11 %d", rc);
1544                 dev_err(&device->cdev->dev,
1545                         "An error occurred in the DASD device driver, "
1546                         "reason=%s\n", errorstring);
1547                 BUG();
1548                 break;
1549         }
1550         cqr->intrc = rc;
1551         return rc;
1552 }
1553 EXPORT_SYMBOL(dasd_start_IO);
1554
1555 /*
1556  * Timeout function for dasd devices. This is used for different purposes
1557  *  1) missing interrupt handler for normal operation
1558  *  2) delayed start of request where start_IO failed with -EBUSY
1559  *  3) timeout for missing state change interrupts
1560  * The head of the ccw queue will have status DASD_CQR_IN_IO for 1),
1561  * DASD_CQR_QUEUED for 2) and 3).
1562  */
1563 static void dasd_device_timeout(unsigned long ptr)
1564 {
1565         unsigned long flags;
1566         struct dasd_device *device;
1567
1568         device = (struct dasd_device *) ptr;
1569         spin_lock_irqsave(get_ccwdev_lock(device->cdev), flags);
1570         /* re-activate request queue */
1571         dasd_device_remove_stop_bits(device, DASD_STOPPED_PENDING);
1572         spin_unlock_irqrestore(get_ccwdev_lock(device->cdev), flags);
1573         dasd_schedule_device_bh(device);
1574 }
1575
1576 /*
1577  * Setup timeout for a device in jiffies.
1578  */
1579 void dasd_device_set_timer(struct dasd_device *device, int expires)
1580 {
1581         if (expires == 0)
1582                 del_timer(&device->timer);
1583         else
1584                 mod_timer(&device->timer, jiffies + expires);
1585 }
1586 EXPORT_SYMBOL(dasd_device_set_timer);
1587
1588 /*
1589  * Clear timeout for a device.
1590  */
1591 void dasd_device_clear_timer(struct dasd_device *device)
1592 {
1593         del_timer(&device->timer);
1594 }
1595 EXPORT_SYMBOL(dasd_device_clear_timer);
1596
1597 static void dasd_handle_killed_request(struct ccw_device *cdev,
1598                                        unsigned long intparm)
1599 {
1600         struct dasd_ccw_req *cqr;
1601         struct dasd_device *device;
1602
1603         if (!intparm)
1604                 return;
1605         cqr = (struct dasd_ccw_req *) intparm;
1606         if (cqr->status != DASD_CQR_IN_IO) {
1607                 DBF_EVENT_DEVID(DBF_DEBUG, cdev,
1608                                 "invalid status in handle_killed_request: "
1609                                 "%02x", cqr->status);
1610                 return;
1611         }
1612
1613         device = dasd_device_from_cdev_locked(cdev);
1614         if (IS_ERR(device)) {
1615                 DBF_EVENT_DEVID(DBF_DEBUG, cdev, "%s",
1616                                 "unable to get device from cdev");
1617                 return;
1618         }
1619
1620         if (!cqr->startdev ||
1621             device != cqr->startdev ||
1622             strncmp(cqr->startdev->discipline->ebcname,
1623                     (char *) &cqr->magic, 4)) {
1624                 DBF_EVENT_DEVID(DBF_DEBUG, cdev, "%s",
1625                                 "invalid device in request");
1626                 dasd_put_device(device);
1627                 return;
1628         }
1629
1630         /* Schedule request to be retried. */
1631         cqr->status = DASD_CQR_QUEUED;
1632
1633         dasd_device_clear_timer(device);
1634         dasd_schedule_device_bh(device);
1635         dasd_put_device(device);
1636 }
1637
1638 void dasd_generic_handle_state_change(struct dasd_device *device)
1639 {
1640         /* First of all start sense subsystem status request. */
1641         dasd_eer_snss(device);
1642
1643         dasd_device_remove_stop_bits(device, DASD_STOPPED_PENDING);
1644         dasd_schedule_device_bh(device);
1645         if (device->block) {
1646                 dasd_schedule_block_bh(device->block);
1647                 blk_mq_run_hw_queues(device->block->request_queue, true);
1648         }
1649 }
1650 EXPORT_SYMBOL_GPL(dasd_generic_handle_state_change);
1651
1652 static int dasd_check_hpf_error(struct irb *irb)
1653 {
1654         return (scsw_tm_is_valid_schxs(&irb->scsw) &&
1655             (irb->scsw.tm.sesq == SCSW_SESQ_DEV_NOFCX ||
1656              irb->scsw.tm.sesq == SCSW_SESQ_PATH_NOFCX));
1657 }
1658
1659 /*
1660  * Interrupt handler for "normal" ssch-io based dasd devices.
1661  */
1662 void dasd_int_handler(struct ccw_device *cdev, unsigned long intparm,
1663                       struct irb *irb)
1664 {
1665         struct dasd_ccw_req *cqr, *next;
1666         struct dasd_device *device;
1667         unsigned long now;
1668         int nrf_suppressed = 0;
1669         int fp_suppressed = 0;
1670         u8 *sense = NULL;
1671         int expires;
1672
1673         cqr = (struct dasd_ccw_req *) intparm;
1674         if (IS_ERR(irb)) {
1675                 switch (PTR_ERR(irb)) {
1676                 case -EIO:
1677                         if (cqr && cqr->status == DASD_CQR_CLEAR_PENDING) {
1678                                 device = cqr->startdev;
1679                                 cqr->status = DASD_CQR_CLEARED;
1680                                 dasd_device_clear_timer(device);
1681                                 wake_up(&dasd_flush_wq);
1682                                 dasd_schedule_device_bh(device);
1683                                 return;
1684                         }
1685                         break;
1686                 case -ETIMEDOUT:
1687                         DBF_EVENT_DEVID(DBF_WARNING, cdev, "%s: "
1688                                         "request timed out\n", __func__);
1689                         break;
1690                 default:
1691                         DBF_EVENT_DEVID(DBF_WARNING, cdev, "%s: "
1692                                         "unknown error %ld\n", __func__,
1693                                         PTR_ERR(irb));
1694                 }
1695                 dasd_handle_killed_request(cdev, intparm);
1696                 return;
1697         }
1698
1699         now = get_tod_clock();
1700         /* check for conditions that should be handled immediately */
1701         if (!cqr ||
1702             !(scsw_dstat(&irb->scsw) == (DEV_STAT_CHN_END | DEV_STAT_DEV_END) &&
1703               scsw_cstat(&irb->scsw) == 0)) {
1704                 if (cqr)
1705                         memcpy(&cqr->irb, irb, sizeof(*irb));
1706                 device = dasd_device_from_cdev_locked(cdev);
1707                 if (IS_ERR(device))
1708                         return;
1709                 /* ignore unsolicited interrupts for DIAG discipline */
1710                 if (device->discipline == dasd_diag_discipline_pointer) {
1711                         dasd_put_device(device);
1712                         return;
1713                 }
1714
1715                 /*
1716                  * In some cases 'File Protected' or 'No Record Found' errors
1717                  * might be expected and debug log messages for the
1718                  * corresponding interrupts shouldn't be written then.
1719                  * Check if either of the according suppress bits is set.
1720                  */
1721                 sense = dasd_get_sense(irb);
1722                 if (sense) {
1723                         fp_suppressed = (sense[1] & SNS1_FILE_PROTECTED) &&
1724                                 test_bit(DASD_CQR_SUPPRESS_FP, &cqr->flags);
1725                         nrf_suppressed = (sense[1] & SNS1_NO_REC_FOUND) &&
1726                                 test_bit(DASD_CQR_SUPPRESS_NRF, &cqr->flags);
1727                 }
1728                 if (!(fp_suppressed || nrf_suppressed))
1729                         device->discipline->dump_sense_dbf(device, irb, "int");
1730
1731                 if (device->features & DASD_FEATURE_ERPLOG)
1732                         device->discipline->dump_sense(device, cqr, irb);
1733                 device->discipline->check_for_device_change(device, cqr, irb);
1734                 dasd_put_device(device);
1735         }
1736
1737         /* check for for attention message */
1738         if (scsw_dstat(&irb->scsw) & DEV_STAT_ATTENTION) {
1739                 device = dasd_device_from_cdev_locked(cdev);
1740                 if (!IS_ERR(device)) {
1741                         device->discipline->check_attention(device,
1742                                                             irb->esw.esw1.lpum);
1743                         dasd_put_device(device);
1744                 }
1745         }
1746
1747         if (!cqr)
1748                 return;
1749
1750         device = (struct dasd_device *) cqr->startdev;
1751         if (!device ||
1752             strncmp(device->discipline->ebcname, (char *) &cqr->magic, 4)) {
1753                 DBF_EVENT_DEVID(DBF_DEBUG, cdev, "%s",
1754                                 "invalid device in request");
1755                 return;
1756         }
1757
1758         /* Check for clear pending */
1759         if (cqr->status == DASD_CQR_CLEAR_PENDING &&
1760             scsw_fctl(&irb->scsw) & SCSW_FCTL_CLEAR_FUNC) {
1761                 cqr->status = DASD_CQR_CLEARED;
1762                 dasd_device_clear_timer(device);
1763                 wake_up(&dasd_flush_wq);
1764                 dasd_schedule_device_bh(device);
1765                 return;
1766         }
1767
1768         /* check status - the request might have been killed by dyn detach */
1769         if (cqr->status != DASD_CQR_IN_IO) {
1770                 DBF_DEV_EVENT(DBF_DEBUG, device, "invalid status: bus_id %s, "
1771                               "status %02x", dev_name(&cdev->dev), cqr->status);
1772                 return;
1773         }
1774
1775         next = NULL;
1776         expires = 0;
1777         if (scsw_dstat(&irb->scsw) == (DEV_STAT_CHN_END | DEV_STAT_DEV_END) &&
1778             scsw_cstat(&irb->scsw) == 0) {
1779                 /* request was completed successfully */
1780                 cqr->status = DASD_CQR_SUCCESS;
1781                 cqr->stopclk = now;
1782                 /* Start first request on queue if possible -> fast_io. */
1783                 if (cqr->devlist.next != &device->ccw_queue) {
1784                         next = list_entry(cqr->devlist.next,
1785                                           struct dasd_ccw_req, devlist);
1786                 }
1787         } else {  /* error */
1788                 /* check for HPF error
1789                  * call discipline function to requeue all requests
1790                  * and disable HPF accordingly
1791                  */
1792                 if (cqr->cpmode && dasd_check_hpf_error(irb) &&
1793                     device->discipline->handle_hpf_error)
1794                         device->discipline->handle_hpf_error(device, irb);
1795                 /*
1796                  * If we don't want complex ERP for this request, then just
1797                  * reset this and retry it in the fastpath
1798                  */
1799                 if (!test_bit(DASD_CQR_FLAGS_USE_ERP, &cqr->flags) &&
1800                     cqr->retries > 0) {
1801                         if (cqr->lpm == dasd_path_get_opm(device))
1802                                 DBF_DEV_EVENT(DBF_DEBUG, device,
1803                                               "default ERP in fastpath "
1804                                               "(%i retries left)",
1805                                               cqr->retries);
1806                         if (!test_bit(DASD_CQR_VERIFY_PATH, &cqr->flags))
1807                                 cqr->lpm = dasd_path_get_opm(device);
1808                         cqr->status = DASD_CQR_QUEUED;
1809                         next = cqr;
1810                 } else
1811                         cqr->status = DASD_CQR_ERROR;
1812         }
1813         if (next && (next->status == DASD_CQR_QUEUED) &&
1814             (!device->stopped)) {
1815                 if (device->discipline->start_IO(next) == 0)
1816                         expires = next->expires;
1817         }
1818         if (expires != 0)
1819                 dasd_device_set_timer(device, expires);
1820         else
1821                 dasd_device_clear_timer(device);
1822         dasd_schedule_device_bh(device);
1823 }
1824 EXPORT_SYMBOL(dasd_int_handler);
1825
1826 enum uc_todo dasd_generic_uc_handler(struct ccw_device *cdev, struct irb *irb)
1827 {
1828         struct dasd_device *device;
1829
1830         device = dasd_device_from_cdev_locked(cdev);
1831
1832         if (IS_ERR(device))
1833                 goto out;
1834         if (test_bit(DASD_FLAG_OFFLINE, &device->flags) ||
1835            device->state != device->target ||
1836            !device->discipline->check_for_device_change){
1837                 dasd_put_device(device);
1838                 goto out;
1839         }
1840         if (device->discipline->dump_sense_dbf)
1841                 device->discipline->dump_sense_dbf(device, irb, "uc");
1842         device->discipline->check_for_device_change(device, NULL, irb);
1843         dasd_put_device(device);
1844 out:
1845         return UC_TODO_RETRY;
1846 }
1847 EXPORT_SYMBOL_GPL(dasd_generic_uc_handler);
1848
1849 /*
1850  * If we have an error on a dasd_block layer request then we cancel
1851  * and return all further requests from the same dasd_block as well.
1852  */
1853 static void __dasd_device_recovery(struct dasd_device *device,
1854                                    struct dasd_ccw_req *ref_cqr)
1855 {
1856         struct list_head *l, *n;
1857         struct dasd_ccw_req *cqr;
1858
1859         /*
1860          * only requeue request that came from the dasd_block layer
1861          */
1862         if (!ref_cqr->block)
1863                 return;
1864
1865         list_for_each_safe(l, n, &device->ccw_queue) {
1866                 cqr = list_entry(l, struct dasd_ccw_req, devlist);
1867                 if (cqr->status == DASD_CQR_QUEUED &&
1868                     ref_cqr->block == cqr->block) {
1869                         cqr->status = DASD_CQR_CLEARED;
1870                 }
1871         }
1872 };
1873
1874 /*
1875  * Remove those ccw requests from the queue that need to be returned
1876  * to the upper layer.
1877  */
1878 static void __dasd_device_process_ccw_queue(struct dasd_device *device,
1879                                             struct list_head *final_queue)
1880 {
1881         struct list_head *l, *n;
1882         struct dasd_ccw_req *cqr;
1883
1884         /* Process request with final status. */
1885         list_for_each_safe(l, n, &device->ccw_queue) {
1886                 cqr = list_entry(l, struct dasd_ccw_req, devlist);
1887
1888                 /* Skip any non-final request. */
1889                 if (cqr->status == DASD_CQR_QUEUED ||
1890                     cqr->status == DASD_CQR_IN_IO ||
1891                     cqr->status == DASD_CQR_CLEAR_PENDING)
1892                         continue;
1893                 if (cqr->status == DASD_CQR_ERROR) {
1894                         __dasd_device_recovery(device, cqr);
1895                 }
1896                 /* Rechain finished requests to final queue */
1897                 list_move_tail(&cqr->devlist, final_queue);
1898         }
1899 }
1900
1901 /*
1902  * the cqrs from the final queue are returned to the upper layer
1903  * by setting a dasd_block state and calling the callback function
1904  */
1905 static void __dasd_device_process_final_queue(struct dasd_device *device,
1906                                               struct list_head *final_queue)
1907 {
1908         struct list_head *l, *n;
1909         struct dasd_ccw_req *cqr;
1910         struct dasd_block *block;
1911         void (*callback)(struct dasd_ccw_req *, void *data);
1912         void *callback_data;
1913         char errorstring[ERRORLENGTH];
1914
1915         list_for_each_safe(l, n, final_queue) {
1916                 cqr = list_entry(l, struct dasd_ccw_req, devlist);
1917                 list_del_init(&cqr->devlist);
1918                 block = cqr->block;
1919                 callback = cqr->callback;
1920                 callback_data = cqr->callback_data;
1921                 if (block)
1922                         spin_lock_bh(&block->queue_lock);
1923                 switch (cqr->status) {
1924                 case DASD_CQR_SUCCESS:
1925                         cqr->status = DASD_CQR_DONE;
1926                         break;
1927                 case DASD_CQR_ERROR:
1928                         cqr->status = DASD_CQR_NEED_ERP;
1929                         break;
1930                 case DASD_CQR_CLEARED:
1931                         cqr->status = DASD_CQR_TERMINATED;
1932                         break;
1933                 default:
1934                         /* internal error 12 - wrong cqr status*/
1935                         snprintf(errorstring, ERRORLENGTH, "12 %p %x02", cqr, cqr->status);
1936                         dev_err(&device->cdev->dev,
1937                                 "An error occurred in the DASD device driver, "
1938                                 "reason=%s\n", errorstring);
1939                         BUG();
1940                 }
1941                 if (cqr->callback != NULL)
1942                         (callback)(cqr, callback_data);
1943                 if (block)
1944                         spin_unlock_bh(&block->queue_lock);
1945         }
1946 }
1947
1948 /*
1949  * Take a look at the first request on the ccw queue and check
1950  * if it reached its expire time. If so, terminate the IO.
1951  */
1952 static void __dasd_device_check_expire(struct dasd_device *device)
1953 {
1954         struct dasd_ccw_req *cqr;
1955
1956         if (list_empty(&device->ccw_queue))
1957                 return;
1958         cqr = list_entry(device->ccw_queue.next, struct dasd_ccw_req, devlist);
1959         if ((cqr->status == DASD_CQR_IN_IO && cqr->expires != 0) &&
1960             (time_after_eq(jiffies, cqr->expires + cqr->starttime))) {
1961                 if (test_bit(DASD_FLAG_SAFE_OFFLINE_RUNNING, &device->flags)) {
1962                         /*
1963                          * IO in safe offline processing should not
1964                          * run out of retries
1965                          */
1966                         cqr->retries++;
1967                 }
1968                 if (device->discipline->term_IO(cqr) != 0) {
1969                         /* Hmpf, try again in 5 sec */
1970                         dev_err(&device->cdev->dev,
1971                                 "cqr %p timed out (%lus) but cannot be "
1972                                 "ended, retrying in 5 s\n",
1973                                 cqr, (cqr->expires/HZ));
1974                         cqr->expires += 5*HZ;
1975                         dasd_device_set_timer(device, 5*HZ);
1976                 } else {
1977                         dev_err(&device->cdev->dev,
1978                                 "cqr %p timed out (%lus), %i retries "
1979                                 "remaining\n", cqr, (cqr->expires/HZ),
1980                                 cqr->retries);
1981                 }
1982         }
1983 }
1984
1985 /*
1986  * return 1 when device is not eligible for IO
1987  */
1988 static int __dasd_device_is_unusable(struct dasd_device *device,
1989                                      struct dasd_ccw_req *cqr)
1990 {
1991         int mask = ~(DASD_STOPPED_DC_WAIT | DASD_UNRESUMED_PM);
1992
1993         if (test_bit(DASD_FLAG_OFFLINE, &device->flags) &&
1994             !test_bit(DASD_FLAG_SAFE_OFFLINE_RUNNING, &device->flags)) {
1995                 /*
1996                  * dasd is being set offline
1997                  * but it is no safe offline where we have to allow I/O
1998                  */
1999                 return 1;
2000         }
2001         if (device->stopped) {
2002                 if (device->stopped & mask) {
2003                         /* stopped and CQR will not change that. */
2004                         return 1;
2005                 }
2006                 if (!test_bit(DASD_CQR_VERIFY_PATH, &cqr->flags)) {
2007                         /* CQR is not able to change device to
2008                          * operational. */
2009                         return 1;
2010                 }
2011                 /* CQR required to get device operational. */
2012         }
2013         return 0;
2014 }
2015
2016 /*
2017  * Take a look at the first request on the ccw queue and check
2018  * if it needs to be started.
2019  */
2020 static void __dasd_device_start_head(struct dasd_device *device)
2021 {
2022         struct dasd_ccw_req *cqr;
2023         int rc;
2024
2025         if (list_empty(&device->ccw_queue))
2026                 return;
2027         cqr = list_entry(device->ccw_queue.next, struct dasd_ccw_req, devlist);
2028         if (cqr->status != DASD_CQR_QUEUED)
2029                 return;
2030         /* if device is not usable return request to upper layer */
2031         if (__dasd_device_is_unusable(device, cqr)) {
2032                 cqr->intrc = -EAGAIN;
2033                 cqr->status = DASD_CQR_CLEARED;
2034                 dasd_schedule_device_bh(device);
2035                 return;
2036         }
2037
2038         rc = device->discipline->start_IO(cqr);
2039         if (rc == 0)
2040                 dasd_device_set_timer(device, cqr->expires);
2041         else if (rc == -EACCES) {
2042                 dasd_schedule_device_bh(device);
2043         } else
2044                 /* Hmpf, try again in 1/2 sec */
2045                 dasd_device_set_timer(device, 50);
2046 }
2047
2048 static void __dasd_device_check_path_events(struct dasd_device *device)
2049 {
2050         int rc;
2051
2052         if (!dasd_path_get_tbvpm(device))
2053                 return;
2054
2055         if (device->stopped &
2056             ~(DASD_STOPPED_DC_WAIT | DASD_UNRESUMED_PM))
2057                 return;
2058         rc = device->discipline->verify_path(device,
2059                                              dasd_path_get_tbvpm(device));
2060         if (rc)
2061                 dasd_device_set_timer(device, 50);
2062         else
2063                 dasd_path_clear_all_verify(device);
2064 };
2065
2066 /*
2067  * Go through all request on the dasd_device request queue,
2068  * terminate them on the cdev if necessary, and return them to the
2069  * submitting layer via callback.
2070  * Note:
2071  * Make sure that all 'submitting layers' still exist when
2072  * this function is called!. In other words, when 'device' is a base
2073  * device then all block layer requests must have been removed before
2074  * via dasd_flush_block_queue.
2075  */
2076 int dasd_flush_device_queue(struct dasd_device *device)
2077 {
2078         struct dasd_ccw_req *cqr, *n;
2079         int rc;
2080         struct list_head flush_queue;
2081
2082         INIT_LIST_HEAD(&flush_queue);
2083         spin_lock_irq(get_ccwdev_lock(device->cdev));
2084         rc = 0;
2085         list_for_each_entry_safe(cqr, n, &device->ccw_queue, devlist) {
2086                 /* Check status and move request to flush_queue */
2087                 switch (cqr->status) {
2088                 case DASD_CQR_IN_IO:
2089                         rc = device->discipline->term_IO(cqr);
2090                         if (rc) {
2091                                 /* unable to terminate requeust */
2092                                 dev_err(&device->cdev->dev,
2093                                         "Flushing the DASD request queue "
2094                                         "failed for request %p\n", cqr);
2095                                 /* stop flush processing */
2096                                 goto finished;
2097                         }
2098                         break;
2099                 case DASD_CQR_QUEUED:
2100                         cqr->stopclk = get_tod_clock();
2101                         cqr->status = DASD_CQR_CLEARED;
2102                         break;
2103                 default: /* no need to modify the others */
2104                         break;
2105                 }
2106                 list_move_tail(&cqr->devlist, &flush_queue);
2107         }
2108 finished:
2109         spin_unlock_irq(get_ccwdev_lock(device->cdev));
2110         /*
2111          * After this point all requests must be in state CLEAR_PENDING,
2112          * CLEARED, SUCCESS or ERROR. Now wait for CLEAR_PENDING to become
2113          * one of the others.
2114          */
2115         list_for_each_entry_safe(cqr, n, &flush_queue, devlist)
2116                 wait_event(dasd_flush_wq,
2117                            (cqr->status != DASD_CQR_CLEAR_PENDING));
2118         /*
2119          * Now set each request back to TERMINATED, DONE or NEED_ERP
2120          * and call the callback function of flushed requests
2121          */
2122         __dasd_device_process_final_queue(device, &flush_queue);
2123         return rc;
2124 }
2125 EXPORT_SYMBOL_GPL(dasd_flush_device_queue);
2126
2127 /*
2128  * Acquire the device lock and process queues for the device.
2129  */
2130 static void dasd_device_tasklet(struct dasd_device *device)
2131 {
2132         struct list_head final_queue;
2133
2134         atomic_set (&device->tasklet_scheduled, 0);
2135         INIT_LIST_HEAD(&final_queue);
2136         spin_lock_irq(get_ccwdev_lock(device->cdev));
2137         /* Check expire time of first request on the ccw queue. */
2138         __dasd_device_check_expire(device);
2139         /* find final requests on ccw queue */
2140         __dasd_device_process_ccw_queue(device, &final_queue);
2141         __dasd_device_check_path_events(device);
2142         spin_unlock_irq(get_ccwdev_lock(device->cdev));
2143         /* Now call the callback function of requests with final status */
2144         __dasd_device_process_final_queue(device, &final_queue);
2145         spin_lock_irq(get_ccwdev_lock(device->cdev));
2146         /* Now check if the head of the ccw queue needs to be started. */
2147         __dasd_device_start_head(device);
2148         spin_unlock_irq(get_ccwdev_lock(device->cdev));
2149         if (waitqueue_active(&shutdown_waitq))
2150                 wake_up(&shutdown_waitq);
2151         dasd_put_device(device);
2152 }
2153
2154 /*
2155  * Schedules a call to dasd_tasklet over the device tasklet.
2156  */
2157 void dasd_schedule_device_bh(struct dasd_device *device)
2158 {
2159         /* Protect against rescheduling. */
2160         if (atomic_cmpxchg (&device->tasklet_scheduled, 0, 1) != 0)
2161                 return;
2162         dasd_get_device(device);
2163         tasklet_hi_schedule(&device->tasklet);
2164 }
2165 EXPORT_SYMBOL(dasd_schedule_device_bh);
2166
2167 void dasd_device_set_stop_bits(struct dasd_device *device, int bits)
2168 {
2169         device->stopped |= bits;
2170 }
2171 EXPORT_SYMBOL_GPL(dasd_device_set_stop_bits);
2172
2173 void dasd_device_remove_stop_bits(struct dasd_device *device, int bits)
2174 {
2175         device->stopped &= ~bits;
2176         if (!device->stopped)
2177                 wake_up(&generic_waitq);
2178 }
2179 EXPORT_SYMBOL_GPL(dasd_device_remove_stop_bits);
2180
2181 /*
2182  * Queue a request to the head of the device ccw_queue.
2183  * Start the I/O if possible.
2184  */
2185 void dasd_add_request_head(struct dasd_ccw_req *cqr)
2186 {
2187         struct dasd_device *device;
2188         unsigned long flags;
2189
2190         device = cqr->startdev;
2191         spin_lock_irqsave(get_ccwdev_lock(device->cdev), flags);
2192         cqr->status = DASD_CQR_QUEUED;
2193         list_add(&cqr->devlist, &device->ccw_queue);
2194         /* let the bh start the request to keep them in order */
2195         dasd_schedule_device_bh(device);
2196         spin_unlock_irqrestore(get_ccwdev_lock(device->cdev), flags);
2197 }
2198 EXPORT_SYMBOL(dasd_add_request_head);
2199
2200 /*
2201  * Queue a request to the tail of the device ccw_queue.
2202  * Start the I/O if possible.
2203  */
2204 void dasd_add_request_tail(struct dasd_ccw_req *cqr)
2205 {
2206         struct dasd_device *device;
2207         unsigned long flags;
2208
2209         device = cqr->startdev;
2210         spin_lock_irqsave(get_ccwdev_lock(device->cdev), flags);
2211         cqr->status = DASD_CQR_QUEUED;
2212         list_add_tail(&cqr->devlist, &device->ccw_queue);
2213         /* let the bh start the request to keep them in order */
2214         dasd_schedule_device_bh(device);
2215         spin_unlock_irqrestore(get_ccwdev_lock(device->cdev), flags);
2216 }
2217 EXPORT_SYMBOL(dasd_add_request_tail);
2218
2219 /*
2220  * Wakeup helper for the 'sleep_on' functions.
2221  */
2222 void dasd_wakeup_cb(struct dasd_ccw_req *cqr, void *data)
2223 {
2224         spin_lock_irq(get_ccwdev_lock(cqr->startdev->cdev));
2225         cqr->callback_data = DASD_SLEEPON_END_TAG;
2226         spin_unlock_irq(get_ccwdev_lock(cqr->startdev->cdev));
2227         wake_up(&generic_waitq);
2228 }
2229 EXPORT_SYMBOL_GPL(dasd_wakeup_cb);
2230
2231 static inline int _wait_for_wakeup(struct dasd_ccw_req *cqr)
2232 {
2233         struct dasd_device *device;
2234         int rc;
2235
2236         device = cqr->startdev;
2237         spin_lock_irq(get_ccwdev_lock(device->cdev));
2238         rc = (cqr->callback_data == DASD_SLEEPON_END_TAG);
2239         spin_unlock_irq(get_ccwdev_lock(device->cdev));
2240         return rc;
2241 }
2242
2243 /*
2244  * checks if error recovery is necessary, returns 1 if yes, 0 otherwise.
2245  */
2246 static int __dasd_sleep_on_erp(struct dasd_ccw_req *cqr)
2247 {
2248         struct dasd_device *device;
2249         dasd_erp_fn_t erp_fn;
2250
2251         if (cqr->status == DASD_CQR_FILLED)
2252                 return 0;
2253         device = cqr->startdev;
2254         if (test_bit(DASD_CQR_FLAGS_USE_ERP, &cqr->flags)) {
2255                 if (cqr->status == DASD_CQR_TERMINATED) {
2256                         device->discipline->handle_terminated_request(cqr);
2257                         return 1;
2258                 }
2259                 if (cqr->status == DASD_CQR_NEED_ERP) {
2260                         erp_fn = device->discipline->erp_action(cqr);
2261                         erp_fn(cqr);
2262                         return 1;
2263                 }
2264                 if (cqr->status == DASD_CQR_FAILED)
2265                         dasd_log_sense(cqr, &cqr->irb);
2266                 if (cqr->refers) {
2267                         __dasd_process_erp(device, cqr);
2268                         return 1;
2269                 }
2270         }
2271         return 0;
2272 }
2273
2274 static int __dasd_sleep_on_loop_condition(struct dasd_ccw_req *cqr)
2275 {
2276         if (test_bit(DASD_CQR_FLAGS_USE_ERP, &cqr->flags)) {
2277                 if (cqr->refers) /* erp is not done yet */
2278                         return 1;
2279                 return ((cqr->status != DASD_CQR_DONE) &&
2280                         (cqr->status != DASD_CQR_FAILED));
2281         } else
2282                 return (cqr->status == DASD_CQR_FILLED);
2283 }
2284
2285 static int _dasd_sleep_on(struct dasd_ccw_req *maincqr, int interruptible)
2286 {
2287         struct dasd_device *device;
2288         int rc;
2289         struct list_head ccw_queue;
2290         struct dasd_ccw_req *cqr;
2291
2292         INIT_LIST_HEAD(&ccw_queue);
2293         maincqr->status = DASD_CQR_FILLED;
2294         device = maincqr->startdev;
2295         list_add(&maincqr->blocklist, &ccw_queue);
2296         for (cqr = maincqr;  __dasd_sleep_on_loop_condition(cqr);
2297              cqr = list_first_entry(&ccw_queue,
2298                                     struct dasd_ccw_req, blocklist)) {
2299
2300                 if (__dasd_sleep_on_erp(cqr))
2301                         continue;
2302                 if (cqr->status != DASD_CQR_FILLED) /* could be failed */
2303                         continue;
2304                 if (test_bit(DASD_FLAG_LOCK_STOLEN, &device->flags) &&
2305                     !test_bit(DASD_CQR_ALLOW_SLOCK, &cqr->flags)) {
2306                         cqr->status = DASD_CQR_FAILED;
2307                         cqr->intrc = -EPERM;
2308                         continue;
2309                 }
2310                 /* Non-temporary stop condition will trigger fail fast */
2311                 if (device->stopped & ~DASD_STOPPED_PENDING &&
2312                     test_bit(DASD_CQR_FLAGS_FAILFAST, &cqr->flags) &&
2313                     (!dasd_eer_enabled(device))) {
2314                         cqr->status = DASD_CQR_FAILED;
2315                         cqr->intrc = -ENOLINK;
2316                         continue;
2317                 }
2318                 /*
2319                  * Don't try to start requests if device is in
2320                  * offline processing, it might wait forever
2321                  */
2322                 if (test_bit(DASD_FLAG_OFFLINE, &device->flags)) {
2323                         cqr->status = DASD_CQR_FAILED;
2324                         cqr->intrc = -ENODEV;
2325                         continue;
2326                 }
2327                 /*
2328                  * Don't try to start requests if device is stopped
2329                  * except path verification requests
2330                  */
2331                 if (!test_bit(DASD_CQR_VERIFY_PATH, &cqr->flags)) {
2332                         if (interruptible) {
2333                                 rc = wait_event_interruptible(
2334                                         generic_waitq, !(device->stopped));
2335                                 if (rc == -ERESTARTSYS) {
2336                                         cqr->status = DASD_CQR_FAILED;
2337                                         maincqr->intrc = rc;
2338                                         continue;
2339                                 }
2340                         } else
2341                                 wait_event(generic_waitq, !(device->stopped));
2342                 }
2343                 if (!cqr->callback)
2344                         cqr->callback = dasd_wakeup_cb;
2345
2346                 cqr->callback_data = DASD_SLEEPON_START_TAG;
2347                 dasd_add_request_tail(cqr);
2348                 if (interruptible) {
2349                         rc = wait_event_interruptible(
2350                                 generic_waitq, _wait_for_wakeup(cqr));
2351                         if (rc == -ERESTARTSYS) {
2352                                 dasd_cancel_req(cqr);
2353                                 /* wait (non-interruptible) for final status */
2354                                 wait_event(generic_waitq,
2355                                            _wait_for_wakeup(cqr));
2356                                 cqr->status = DASD_CQR_FAILED;
2357                                 maincqr->intrc = rc;
2358                                 continue;
2359                         }
2360                 } else
2361                         wait_event(generic_waitq, _wait_for_wakeup(cqr));
2362         }
2363
2364         maincqr->endclk = get_tod_clock();
2365         if ((maincqr->status != DASD_CQR_DONE) &&
2366             (maincqr->intrc != -ERESTARTSYS))
2367                 dasd_log_sense(maincqr, &maincqr->irb);
2368         if (maincqr->status == DASD_CQR_DONE)
2369                 rc = 0;
2370         else if (maincqr->intrc)
2371                 rc = maincqr->intrc;
2372         else
2373                 rc = -EIO;
2374         return rc;
2375 }
2376
2377 static inline int _wait_for_wakeup_queue(struct list_head *ccw_queue)
2378 {
2379         struct dasd_ccw_req *cqr;
2380
2381         list_for_each_entry(cqr, ccw_queue, blocklist) {
2382                 if (cqr->callback_data != DASD_SLEEPON_END_TAG)
2383                         return 0;
2384         }
2385
2386         return 1;
2387 }
2388
2389 static int _dasd_sleep_on_queue(struct list_head *ccw_queue, int interruptible)
2390 {
2391         struct dasd_device *device;
2392         struct dasd_ccw_req *cqr, *n;
2393         u8 *sense = NULL;
2394         int rc;
2395
2396 retry:
2397         list_for_each_entry_safe(cqr, n, ccw_queue, blocklist) {
2398                 device = cqr->startdev;
2399                 if (cqr->status != DASD_CQR_FILLED) /*could be failed*/
2400                         continue;
2401
2402                 if (test_bit(DASD_FLAG_LOCK_STOLEN, &device->flags) &&
2403                     !test_bit(DASD_CQR_ALLOW_SLOCK, &cqr->flags)) {
2404                         cqr->status = DASD_CQR_FAILED;
2405                         cqr->intrc = -EPERM;
2406                         continue;
2407                 }
2408                 /*Non-temporary stop condition will trigger fail fast*/
2409                 if (device->stopped & ~DASD_STOPPED_PENDING &&
2410                     test_bit(DASD_CQR_FLAGS_FAILFAST, &cqr->flags) &&
2411                     !dasd_eer_enabled(device)) {
2412                         cqr->status = DASD_CQR_FAILED;
2413                         cqr->intrc = -EAGAIN;
2414                         continue;
2415                 }
2416
2417                 /*Don't try to start requests if device is stopped*/
2418                 if (interruptible) {
2419                         rc = wait_event_interruptible(
2420                                 generic_waitq, !device->stopped);
2421                         if (rc == -ERESTARTSYS) {
2422                                 cqr->status = DASD_CQR_FAILED;
2423                                 cqr->intrc = rc;
2424                                 continue;
2425                         }
2426                 } else
2427                         wait_event(generic_waitq, !(device->stopped));
2428
2429                 if (!cqr->callback)
2430                         cqr->callback = dasd_wakeup_cb;
2431                 cqr->callback_data = DASD_SLEEPON_START_TAG;
2432                 dasd_add_request_tail(cqr);
2433         }
2434
2435         wait_event(generic_waitq, _wait_for_wakeup_queue(ccw_queue));
2436
2437         rc = 0;
2438         list_for_each_entry_safe(cqr, n, ccw_queue, blocklist) {
2439                 /*
2440                  * In some cases the 'File Protected' or 'Incorrect Length'
2441                  * error might be expected and error recovery would be
2442                  * unnecessary in these cases.  Check if the according suppress
2443                  * bit is set.
2444                  */
2445                 sense = dasd_get_sense(&cqr->irb);
2446                 if (sense && sense[1] & SNS1_FILE_PROTECTED &&
2447                     test_bit(DASD_CQR_SUPPRESS_FP, &cqr->flags))
2448                         continue;
2449                 if (scsw_cstat(&cqr->irb.scsw) == 0x40 &&
2450                     test_bit(DASD_CQR_SUPPRESS_IL, &cqr->flags))
2451                         continue;
2452
2453                 /*
2454                  * for alias devices simplify error recovery and
2455                  * return to upper layer
2456                  * do not skip ERP requests
2457                  */
2458                 if (cqr->startdev != cqr->basedev && !cqr->refers &&
2459                     (cqr->status == DASD_CQR_TERMINATED ||
2460                      cqr->status == DASD_CQR_NEED_ERP))
2461                         return -EAGAIN;
2462
2463                 /* normal recovery for basedev IO */
2464                 if (__dasd_sleep_on_erp(cqr))
2465                         /* handle erp first */
2466                         goto retry;
2467         }
2468
2469         return 0;
2470 }
2471
2472 /*
2473  * Queue a request to the tail of the device ccw_queue and wait for
2474  * it's completion.
2475  */
2476 int dasd_sleep_on(struct dasd_ccw_req *cqr)
2477 {
2478         return _dasd_sleep_on(cqr, 0);
2479 }
2480 EXPORT_SYMBOL(dasd_sleep_on);
2481
2482 /*
2483  * Start requests from a ccw_queue and wait for their completion.
2484  */
2485 int dasd_sleep_on_queue(struct list_head *ccw_queue)
2486 {
2487         return _dasd_sleep_on_queue(ccw_queue, 0);
2488 }
2489 EXPORT_SYMBOL(dasd_sleep_on_queue);
2490
2491 /*
2492  * Queue a request to the tail of the device ccw_queue and wait
2493  * interruptible for it's completion.
2494  */
2495 int dasd_sleep_on_interruptible(struct dasd_ccw_req *cqr)
2496 {
2497         return _dasd_sleep_on(cqr, 1);
2498 }
2499 EXPORT_SYMBOL(dasd_sleep_on_interruptible);
2500
2501 /*
2502  * Whoa nelly now it gets really hairy. For some functions (e.g. steal lock
2503  * for eckd devices) the currently running request has to be terminated
2504  * and be put back to status queued, before the special request is added
2505  * to the head of the queue. Then the special request is waited on normally.
2506  */
2507 static inline int _dasd_term_running_cqr(struct dasd_device *device)
2508 {
2509         struct dasd_ccw_req *cqr;
2510         int rc;
2511
2512         if (list_empty(&device->ccw_queue))
2513                 return 0;
2514         cqr = list_entry(device->ccw_queue.next, struct dasd_ccw_req, devlist);
2515         rc = device->discipline->term_IO(cqr);
2516         if (!rc)
2517                 /*
2518                  * CQR terminated because a more important request is pending.
2519                  * Undo decreasing of retry counter because this is
2520                  * not an error case.
2521                  */
2522                 cqr->retries++;
2523         return rc;
2524 }
2525
2526 int dasd_sleep_on_immediatly(struct dasd_ccw_req *cqr)
2527 {
2528         struct dasd_device *device;
2529         int rc;
2530
2531         device = cqr->startdev;
2532         if (test_bit(DASD_FLAG_LOCK_STOLEN, &device->flags) &&
2533             !test_bit(DASD_CQR_ALLOW_SLOCK, &cqr->flags)) {
2534                 cqr->status = DASD_CQR_FAILED;
2535                 cqr->intrc = -EPERM;
2536                 return -EIO;
2537         }
2538         spin_lock_irq(get_ccwdev_lock(device->cdev));
2539         rc = _dasd_term_running_cqr(device);
2540         if (rc) {
2541                 spin_unlock_irq(get_ccwdev_lock(device->cdev));
2542                 return rc;
2543         }
2544         cqr->callback = dasd_wakeup_cb;
2545         cqr->callback_data = DASD_SLEEPON_START_TAG;
2546         cqr->status = DASD_CQR_QUEUED;
2547         /*
2548          * add new request as second
2549          * first the terminated cqr needs to be finished
2550          */
2551         list_add(&cqr->devlist, device->ccw_queue.next);
2552
2553         /* let the bh start the request to keep them in order */
2554         dasd_schedule_device_bh(device);
2555
2556         spin_unlock_irq(get_ccwdev_lock(device->cdev));
2557
2558         wait_event(generic_waitq, _wait_for_wakeup(cqr));
2559
2560         if (cqr->status == DASD_CQR_DONE)
2561                 rc = 0;
2562         else if (cqr->intrc)
2563                 rc = cqr->intrc;
2564         else
2565                 rc = -EIO;
2566
2567         /* kick tasklets */
2568         dasd_schedule_device_bh(device);
2569         if (device->block)
2570                 dasd_schedule_block_bh(device->block);
2571
2572         return rc;
2573 }
2574 EXPORT_SYMBOL(dasd_sleep_on_immediatly);
2575
2576 /*
2577  * Cancels a request that was started with dasd_sleep_on_req.
2578  * This is useful to timeout requests. The request will be
2579  * terminated if it is currently in i/o.
2580  * Returns 0 if request termination was successful
2581  *         negative error code if termination failed
2582  * Cancellation of a request is an asynchronous operation! The calling
2583  * function has to wait until the request is properly returned via callback.
2584  */
2585 int dasd_cancel_req(struct dasd_ccw_req *cqr)
2586 {
2587         struct dasd_device *device = cqr->startdev;
2588         unsigned long flags;
2589         int rc;
2590
2591         rc = 0;
2592         spin_lock_irqsave(get_ccwdev_lock(device->cdev), flags);
2593         switch (cqr->status) {
2594         case DASD_CQR_QUEUED:
2595                 /* request was not started - just set to cleared */
2596                 cqr->status = DASD_CQR_CLEARED;
2597                 if (cqr->callback_data == DASD_SLEEPON_START_TAG)
2598                         cqr->callback_data = DASD_SLEEPON_END_TAG;
2599                 break;
2600         case DASD_CQR_IN_IO:
2601                 /* request in IO - terminate IO and release again */
2602                 rc = device->discipline->term_IO(cqr);
2603                 if (rc) {
2604                         dev_err(&device->cdev->dev,
2605                                 "Cancelling request %p failed with rc=%d\n",
2606                                 cqr, rc);
2607                 } else {
2608                         cqr->stopclk = get_tod_clock();
2609                 }
2610                 break;
2611         default: /* already finished or clear pending - do nothing */
2612                 break;
2613         }
2614         spin_unlock_irqrestore(get_ccwdev_lock(device->cdev), flags);
2615         dasd_schedule_device_bh(device);
2616         return rc;
2617 }
2618 EXPORT_SYMBOL(dasd_cancel_req);
2619
2620 /*
2621  * SECTION: Operations of the dasd_block layer.
2622  */
2623
2624 /*
2625  * Timeout function for dasd_block. This is used when the block layer
2626  * is waiting for something that may not come reliably, (e.g. a state
2627  * change interrupt)
2628  */
2629 static void dasd_block_timeout(unsigned long ptr)
2630 {
2631         unsigned long flags;
2632         struct dasd_block *block;
2633
2634         block = (struct dasd_block *) ptr;
2635         spin_lock_irqsave(get_ccwdev_lock(block->base->cdev), flags);
2636         /* re-activate request queue */
2637         dasd_device_remove_stop_bits(block->base, DASD_STOPPED_PENDING);
2638         spin_unlock_irqrestore(get_ccwdev_lock(block->base->cdev), flags);
2639         dasd_schedule_block_bh(block);
2640         blk_mq_run_hw_queues(block->request_queue, true);
2641 }
2642
2643 /*
2644  * Setup timeout for a dasd_block in jiffies.
2645  */
2646 void dasd_block_set_timer(struct dasd_block *block, int expires)
2647 {
2648         if (expires == 0)
2649                 del_timer(&block->timer);
2650         else
2651                 mod_timer(&block->timer, jiffies + expires);
2652 }
2653 EXPORT_SYMBOL(dasd_block_set_timer);
2654
2655 /*
2656  * Clear timeout for a dasd_block.
2657  */
2658 void dasd_block_clear_timer(struct dasd_block *block)
2659 {
2660         del_timer(&block->timer);
2661 }
2662 EXPORT_SYMBOL(dasd_block_clear_timer);
2663
2664 /*
2665  * Process finished error recovery ccw.
2666  */
2667 static void __dasd_process_erp(struct dasd_device *device,
2668                                struct dasd_ccw_req *cqr)
2669 {
2670         dasd_erp_fn_t erp_fn;
2671
2672         if (cqr->status == DASD_CQR_DONE)
2673                 DBF_DEV_EVENT(DBF_NOTICE, device, "%s", "ERP successful");
2674         else
2675                 dev_err(&device->cdev->dev, "ERP failed for the DASD\n");
2676         erp_fn = device->discipline->erp_postaction(cqr);
2677         erp_fn(cqr);
2678 }
2679
2680 static void __dasd_cleanup_cqr(struct dasd_ccw_req *cqr)
2681 {
2682         struct request *req;
2683         blk_status_t error = BLK_STS_OK;
2684         int status;
2685
2686         req = (struct request *) cqr->callback_data;
2687         dasd_profile_end(cqr->block, cqr, req);
2688
2689         status = cqr->block->base->discipline->free_cp(cqr, req);
2690         if (status < 0)
2691                 error = errno_to_blk_status(status);
2692         else if (status == 0) {
2693                 switch (cqr->intrc) {
2694                 case -EPERM:
2695                         error = BLK_STS_NEXUS;
2696                         break;
2697                 case -ENOLINK:
2698                         error = BLK_STS_TRANSPORT;
2699                         break;
2700                 case -ETIMEDOUT:
2701                         error = BLK_STS_TIMEOUT;
2702                         break;
2703                 default:
2704                         error = BLK_STS_IOERR;
2705                         break;
2706                 }
2707         }
2708
2709         /*
2710          * We need to take care for ETIMEDOUT errors here since the
2711          * complete callback does not get called in this case.
2712          * Take care of all errors here and avoid additional code to
2713          * transfer the error value to the complete callback.
2714          */
2715         if (error) {
2716                 blk_mq_end_request(req, error);
2717                 blk_mq_run_hw_queues(req->q, true);
2718         } else {
2719                 blk_mq_complete_request(req);
2720         }
2721 }
2722
2723 /*
2724  * Process ccw request queue.
2725  */
2726 static void __dasd_process_block_ccw_queue(struct dasd_block *block,
2727                                            struct list_head *final_queue)
2728 {
2729         struct list_head *l, *n;
2730         struct dasd_ccw_req *cqr;
2731         dasd_erp_fn_t erp_fn;
2732         unsigned long flags;
2733         struct dasd_device *base = block->base;
2734
2735 restart:
2736         /* Process request with final status. */
2737         list_for_each_safe(l, n, &block->ccw_queue) {
2738                 cqr = list_entry(l, struct dasd_ccw_req, blocklist);
2739                 if (cqr->status != DASD_CQR_DONE &&
2740                     cqr->status != DASD_CQR_FAILED &&
2741                     cqr->status != DASD_CQR_NEED_ERP &&
2742                     cqr->status != DASD_CQR_TERMINATED)
2743                         continue;
2744
2745                 if (cqr->status == DASD_CQR_TERMINATED) {
2746                         base->discipline->handle_terminated_request(cqr);
2747                         goto restart;
2748                 }
2749
2750                 /*  Process requests that may be recovered */
2751                 if (cqr->status == DASD_CQR_NEED_ERP) {
2752                         erp_fn = base->discipline->erp_action(cqr);
2753                         if (IS_ERR(erp_fn(cqr)))
2754                                 continue;
2755                         goto restart;
2756                 }
2757
2758                 /* log sense for fatal error */
2759                 if (cqr->status == DASD_CQR_FAILED) {
2760                         dasd_log_sense(cqr, &cqr->irb);
2761                 }
2762
2763                 /* First of all call extended error reporting. */
2764                 if (dasd_eer_enabled(base) &&
2765                     cqr->status == DASD_CQR_FAILED) {
2766                         dasd_eer_write(base, cqr, DASD_EER_FATALERROR);
2767
2768                         /* restart request  */
2769                         cqr->status = DASD_CQR_FILLED;
2770                         cqr->retries = 255;
2771                         spin_lock_irqsave(get_ccwdev_lock(base->cdev), flags);
2772                         dasd_device_set_stop_bits(base, DASD_STOPPED_QUIESCE);
2773                         spin_unlock_irqrestore(get_ccwdev_lock(base->cdev),
2774                                                flags);
2775                         goto restart;
2776                 }
2777
2778                 /* Process finished ERP request. */
2779                 if (cqr->refers) {
2780                         __dasd_process_erp(base, cqr);
2781                         goto restart;
2782                 }
2783
2784                 /* Rechain finished requests to final queue */
2785                 cqr->endclk = get_tod_clock();
2786                 list_move_tail(&cqr->blocklist, final_queue);
2787         }
2788 }
2789
2790 static void dasd_return_cqr_cb(struct dasd_ccw_req *cqr, void *data)
2791 {
2792         dasd_schedule_block_bh(cqr->block);
2793 }
2794
2795 static void __dasd_block_start_head(struct dasd_block *block)
2796 {
2797         struct dasd_ccw_req *cqr;
2798
2799         if (list_empty(&block->ccw_queue))
2800                 return;
2801         /* We allways begin with the first requests on the queue, as some
2802          * of previously started requests have to be enqueued on a
2803          * dasd_device again for error recovery.
2804          */
2805         list_for_each_entry(cqr, &block->ccw_queue, blocklist) {
2806                 if (cqr->status != DASD_CQR_FILLED)
2807                         continue;
2808                 if (test_bit(DASD_FLAG_LOCK_STOLEN, &block->base->flags) &&
2809                     !test_bit(DASD_CQR_ALLOW_SLOCK, &cqr->flags)) {
2810                         cqr->status = DASD_CQR_FAILED;
2811                         cqr->intrc = -EPERM;
2812                         dasd_schedule_block_bh(block);
2813                         continue;
2814                 }
2815                 /* Non-temporary stop condition will trigger fail fast */
2816                 if (block->base->stopped & ~DASD_STOPPED_PENDING &&
2817                     test_bit(DASD_CQR_FLAGS_FAILFAST, &cqr->flags) &&
2818                     (!dasd_eer_enabled(block->base))) {
2819                         cqr->status = DASD_CQR_FAILED;
2820                         cqr->intrc = -ENOLINK;
2821                         dasd_schedule_block_bh(block);
2822                         continue;
2823                 }
2824                 /* Don't try to start requests if device is stopped */
2825                 if (block->base->stopped)
2826                         return;
2827
2828                 /* just a fail safe check, should not happen */
2829                 if (!cqr->startdev)
2830                         cqr->startdev = block->base;
2831
2832                 /* make sure that the requests we submit find their way back */
2833                 cqr->callback = dasd_return_cqr_cb;
2834
2835                 dasd_add_request_tail(cqr);
2836         }
2837 }
2838
2839 /*
2840  * Central dasd_block layer routine. Takes requests from the generic
2841  * block layer request queue, creates ccw requests, enqueues them on
2842  * a dasd_device and processes ccw requests that have been returned.
2843  */
2844 static void dasd_block_tasklet(struct dasd_block *block)
2845 {
2846         struct list_head final_queue;
2847         struct list_head *l, *n;
2848         struct dasd_ccw_req *cqr;
2849         struct dasd_queue *dq;
2850
2851         atomic_set(&block->tasklet_scheduled, 0);
2852         INIT_LIST_HEAD(&final_queue);
2853         spin_lock_irq(&block->queue_lock);
2854         /* Finish off requests on ccw queue */
2855         __dasd_process_block_ccw_queue(block, &final_queue);
2856         spin_unlock_irq(&block->queue_lock);
2857
2858         /* Now call the callback function of requests with final status */
2859         list_for_each_safe(l, n, &final_queue) {
2860                 cqr = list_entry(l, struct dasd_ccw_req, blocklist);
2861                 dq = cqr->dq;
2862                 spin_lock_irq(&dq->lock);
2863                 list_del_init(&cqr->blocklist);
2864                 __dasd_cleanup_cqr(cqr);
2865                 spin_unlock_irq(&dq->lock);
2866         }
2867
2868         spin_lock_irq(&block->queue_lock);
2869         /* Now check if the head of the ccw queue needs to be started. */
2870         __dasd_block_start_head(block);
2871         spin_unlock_irq(&block->queue_lock);
2872
2873         if (waitqueue_active(&shutdown_waitq))
2874                 wake_up(&shutdown_waitq);
2875         dasd_put_device(block->base);
2876 }
2877
2878 static void _dasd_wake_block_flush_cb(struct dasd_ccw_req *cqr, void *data)
2879 {
2880         wake_up(&dasd_flush_wq);
2881 }
2882
2883 /*
2884  * Requeue a request back to the block request queue
2885  * only works for block requests
2886  */
2887 static int _dasd_requeue_request(struct dasd_ccw_req *cqr)
2888 {
2889         struct dasd_block *block = cqr->block;
2890         struct request *req;
2891
2892         if (!block)
2893                 return -EINVAL;
2894         spin_lock_irq(&cqr->dq->lock);
2895         req = (struct request *) cqr->callback_data;
2896         blk_mq_requeue_request(req, false);
2897         spin_unlock_irq(&cqr->dq->lock);
2898
2899         return 0;
2900 }
2901
2902 /*
2903  * Go through all request on the dasd_block request queue, cancel them
2904  * on the respective dasd_device, and return them to the generic
2905  * block layer.
2906  */
2907 static int dasd_flush_block_queue(struct dasd_block *block)
2908 {
2909         struct dasd_ccw_req *cqr, *n;
2910         int rc, i;
2911         struct list_head flush_queue;
2912         unsigned long flags;
2913
2914         INIT_LIST_HEAD(&flush_queue);
2915         spin_lock_bh(&block->queue_lock);
2916         rc = 0;
2917 restart:
2918         list_for_each_entry_safe(cqr, n, &block->ccw_queue, blocklist) {
2919                 /* if this request currently owned by a dasd_device cancel it */
2920                 if (cqr->status >= DASD_CQR_QUEUED)
2921                         rc = dasd_cancel_req(cqr);
2922                 if (rc < 0)
2923                         break;
2924                 /* Rechain request (including erp chain) so it won't be
2925                  * touched by the dasd_block_tasklet anymore.
2926                  * Replace the callback so we notice when the request
2927                  * is returned from the dasd_device layer.
2928                  */
2929                 cqr->callback = _dasd_wake_block_flush_cb;
2930                 for (i = 0; cqr != NULL; cqr = cqr->refers, i++)
2931                         list_move_tail(&cqr->blocklist, &flush_queue);
2932                 if (i > 1)
2933                         /* moved more than one request - need to restart */
2934                         goto restart;
2935         }
2936         spin_unlock_bh(&block->queue_lock);
2937         /* Now call the callback function of flushed requests */
2938 restart_cb:
2939         list_for_each_entry_safe(cqr, n, &flush_queue, blocklist) {
2940                 wait_event(dasd_flush_wq, (cqr->status < DASD_CQR_QUEUED));
2941                 /* Process finished ERP request. */
2942                 if (cqr->refers) {
2943                         spin_lock_bh(&block->queue_lock);
2944                         __dasd_process_erp(block->base, cqr);
2945                         spin_unlock_bh(&block->queue_lock);
2946                         /* restart list_for_xx loop since dasd_process_erp
2947                          * might remove multiple elements */
2948                         goto restart_cb;
2949                 }
2950                 /* call the callback function */
2951                 spin_lock_irqsave(&cqr->dq->lock, flags);
2952                 cqr->endclk = get_tod_clock();
2953                 list_del_init(&cqr->blocklist);
2954                 __dasd_cleanup_cqr(cqr);
2955                 spin_unlock_irqrestore(&cqr->dq->lock, flags);
2956         }
2957         return rc;
2958 }
2959
2960 /*
2961  * Schedules a call to dasd_tasklet over the device tasklet.
2962  */
2963 void dasd_schedule_block_bh(struct dasd_block *block)
2964 {
2965         /* Protect against rescheduling. */
2966         if (atomic_cmpxchg(&block->tasklet_scheduled, 0, 1) != 0)
2967                 return;
2968         /* life cycle of block is bound to it's base device */
2969         dasd_get_device(block->base);
2970         tasklet_hi_schedule(&block->tasklet);
2971 }
2972 EXPORT_SYMBOL(dasd_schedule_block_bh);
2973
2974
2975 /*
2976  * SECTION: external block device operations
2977  * (request queue handling, open, release, etc.)
2978  */
2979
2980 /*
2981  * Dasd request queue function. Called from ll_rw_blk.c
2982  */
2983 static blk_status_t do_dasd_request(struct blk_mq_hw_ctx *hctx,
2984                                     const struct blk_mq_queue_data *qd)
2985 {
2986         struct dasd_block *block = hctx->queue->queuedata;
2987         struct dasd_queue *dq = hctx->driver_data;
2988         struct request *req = qd->rq;
2989         struct dasd_device *basedev;
2990         struct dasd_ccw_req *cqr;
2991         blk_status_t rc = BLK_STS_OK;
2992
2993         basedev = block->base;
2994         spin_lock_irq(&dq->lock);
2995         if (basedev->state < DASD_STATE_READY) {
2996                 DBF_DEV_EVENT(DBF_ERR, basedev,
2997                               "device not ready for request %p", req);
2998                 rc = BLK_STS_IOERR;
2999                 goto out;
3000         }
3001
3002         /*
3003          * if device is stopped do not fetch new requests
3004          * except failfast is active which will let requests fail
3005          * immediately in __dasd_block_start_head()
3006          */
3007         if (basedev->stopped && !(basedev->features & DASD_FEATURE_FAILFAST)) {
3008                 DBF_DEV_EVENT(DBF_ERR, basedev,
3009                               "device stopped request %p", req);
3010                 rc = BLK_STS_RESOURCE;
3011                 goto out;
3012         }
3013
3014         if (basedev->features & DASD_FEATURE_READONLY &&
3015             rq_data_dir(req) == WRITE) {
3016                 DBF_DEV_EVENT(DBF_ERR, basedev,
3017                               "Rejecting write request %p", req);
3018                 rc = BLK_STS_IOERR;
3019                 goto out;
3020         }
3021
3022         if (test_bit(DASD_FLAG_ABORTALL, &basedev->flags) &&
3023             (basedev->features & DASD_FEATURE_FAILFAST ||
3024              blk_noretry_request(req))) {
3025                 DBF_DEV_EVENT(DBF_ERR, basedev,
3026                               "Rejecting failfast request %p", req);
3027                 rc = BLK_STS_IOERR;
3028                 goto out;
3029         }
3030
3031         cqr = basedev->discipline->build_cp(basedev, block, req);
3032         if (IS_ERR(cqr)) {
3033                 if (PTR_ERR(cqr) == -EBUSY ||
3034                     PTR_ERR(cqr) == -ENOMEM ||
3035                     PTR_ERR(cqr) == -EAGAIN) {
3036                         rc = BLK_STS_RESOURCE;
3037                         goto out;
3038                 }
3039                 DBF_DEV_EVENT(DBF_ERR, basedev,
3040                               "CCW creation failed (rc=%ld) on request %p",
3041                               PTR_ERR(cqr), req);
3042                 rc = BLK_STS_IOERR;
3043                 goto out;
3044         }
3045         /*
3046          *  Note: callback is set to dasd_return_cqr_cb in
3047          * __dasd_block_start_head to cover erp requests as well
3048          */
3049         cqr->callback_data = req;
3050         cqr->status = DASD_CQR_FILLED;
3051         cqr->dq = dq;
3052         req->completion_data = cqr;
3053         blk_mq_start_request(req);
3054         spin_lock(&block->queue_lock);
3055         list_add_tail(&cqr->blocklist, &block->ccw_queue);
3056         INIT_LIST_HEAD(&cqr->devlist);
3057         dasd_profile_start(block, cqr, req);
3058         dasd_schedule_block_bh(block);
3059         spin_unlock(&block->queue_lock);
3060
3061 out:
3062         spin_unlock_irq(&dq->lock);
3063         return rc;
3064 }
3065
3066 /*
3067  * Block timeout callback, called from the block layer
3068  *
3069  * Return values:
3070  * BLK_EH_RESET_TIMER if the request should be left running
3071  * BLK_EH_NOT_HANDLED if the request is handled or terminated
3072  *                    by the driver.
3073  */
3074 enum blk_eh_timer_return dasd_times_out(struct request *req, bool reserved)
3075 {
3076         struct dasd_ccw_req *cqr = req->completion_data;
3077         struct dasd_block *block = req->q->queuedata;
3078         struct dasd_device *device;
3079         unsigned long flags;
3080         int rc = 0;
3081
3082         if (!cqr)
3083                 return BLK_EH_NOT_HANDLED;
3084
3085         spin_lock_irqsave(&cqr->dq->lock, flags);
3086         device = cqr->startdev ? cqr->startdev : block->base;
3087         if (!device->blk_timeout) {
3088                 spin_unlock_irqrestore(&cqr->dq->lock, flags);
3089                 return BLK_EH_RESET_TIMER;
3090         }
3091         DBF_DEV_EVENT(DBF_WARNING, device,
3092                       " dasd_times_out cqr %p status %x",
3093                       cqr, cqr->status);
3094
3095         spin_lock(&block->queue_lock);
3096         spin_lock(get_ccwdev_lock(device->cdev));
3097         cqr->retries = -1;
3098         cqr->intrc = -ETIMEDOUT;
3099         if (cqr->status >= DASD_CQR_QUEUED) {
3100                 spin_unlock(get_ccwdev_lock(device->cdev));
3101                 rc = dasd_cancel_req(cqr);
3102         } else if (cqr->status == DASD_CQR_FILLED ||
3103                    cqr->status == DASD_CQR_NEED_ERP) {
3104                 cqr->status = DASD_CQR_TERMINATED;
3105                 spin_unlock(get_ccwdev_lock(device->cdev));
3106         } else if (cqr->status == DASD_CQR_IN_ERP) {
3107                 struct dasd_ccw_req *searchcqr, *nextcqr, *tmpcqr;
3108
3109                 list_for_each_entry_safe(searchcqr, nextcqr,
3110                                          &block->ccw_queue, blocklist) {
3111                         tmpcqr = searchcqr;
3112                         while (tmpcqr->refers)
3113                                 tmpcqr = tmpcqr->refers;
3114                         if (tmpcqr != cqr)
3115                                 continue;
3116                         /* searchcqr is an ERP request for cqr */
3117                         searchcqr->retries = -1;
3118                         searchcqr->intrc = -ETIMEDOUT;
3119                         if (searchcqr->status >= DASD_CQR_QUEUED) {
3120                                 spin_unlock(get_ccwdev_lock(device->cdev));
3121                                 rc = dasd_cancel_req(searchcqr);
3122                                 spin_lock(get_ccwdev_lock(device->cdev));
3123                         } else if ((searchcqr->status == DASD_CQR_FILLED) ||
3124                                    (searchcqr->status == DASD_CQR_NEED_ERP)) {
3125                                 searchcqr->status = DASD_CQR_TERMINATED;
3126                                 rc = 0;
3127                         } else if (searchcqr->status == DASD_CQR_IN_ERP) {
3128                                 /*
3129                                  * Shouldn't happen; most recent ERP
3130                                  * request is at the front of queue
3131                                  */
3132                                 continue;
3133                         }
3134                         break;
3135                 }
3136                 spin_unlock(get_ccwdev_lock(device->cdev));
3137         }
3138         dasd_schedule_block_bh(block);
3139         spin_unlock(&block->queue_lock);
3140         spin_unlock_irqrestore(&cqr->dq->lock, flags);
3141
3142         return rc ? BLK_EH_RESET_TIMER : BLK_EH_NOT_HANDLED;
3143 }
3144
3145 static int dasd_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
3146                           unsigned int idx)
3147 {
3148         struct dasd_queue *dq = kzalloc(sizeof(*dq), GFP_KERNEL);
3149
3150         if (!dq)
3151                 return -ENOMEM;
3152
3153         spin_lock_init(&dq->lock);
3154         hctx->driver_data = dq;
3155
3156         return 0;
3157 }
3158
3159 static void dasd_exit_hctx(struct blk_mq_hw_ctx *hctx, unsigned int idx)
3160 {
3161         kfree(hctx->driver_data);
3162         hctx->driver_data = NULL;
3163 }
3164
3165 static void dasd_request_done(struct request *req)
3166 {
3167         blk_mq_end_request(req, 0);
3168         blk_mq_run_hw_queues(req->q, true);
3169 }
3170
3171 static struct blk_mq_ops dasd_mq_ops = {
3172         .queue_rq = do_dasd_request,
3173         .complete = dasd_request_done,
3174         .timeout = dasd_times_out,
3175         .init_hctx = dasd_init_hctx,
3176         .exit_hctx = dasd_exit_hctx,
3177 };
3178
3179 /*
3180  * Allocate and initialize request queue and default I/O scheduler.
3181  */
3182 static int dasd_alloc_queue(struct dasd_block *block)
3183 {
3184         int rc;
3185
3186         block->tag_set.ops = &dasd_mq_ops;
3187         block->tag_set.nr_hw_queues = DASD_NR_HW_QUEUES;
3188         block->tag_set.queue_depth = DASD_MAX_LCU_DEV * DASD_REQ_PER_DEV;
3189         block->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
3190
3191         rc = blk_mq_alloc_tag_set(&block->tag_set);
3192         if (rc)
3193                 return rc;
3194
3195         block->request_queue = blk_mq_init_queue(&block->tag_set);
3196         if (IS_ERR(block->request_queue))
3197                 return PTR_ERR(block->request_queue);
3198
3199         block->request_queue->queuedata = block;
3200
3201         return 0;
3202 }
3203
3204 /*
3205  * Allocate and initialize request queue.
3206  */
3207 static void dasd_setup_queue(struct dasd_block *block)
3208 {
3209         unsigned int logical_block_size = block->bp_block;
3210         struct request_queue *q = block->request_queue;
3211         unsigned int max_bytes, max_discard_sectors;
3212         int max;
3213
3214         if (block->base->features & DASD_FEATURE_USERAW) {
3215                 /*
3216                  * the max_blocks value for raw_track access is 256
3217                  * it is higher than the native ECKD value because we
3218                  * only need one ccw per track
3219                  * so the max_hw_sectors are
3220                  * 2048 x 512B = 1024kB = 16 tracks
3221                  */
3222                 max = 2048;
3223         } else {
3224                 max = block->base->discipline->max_blocks << block->s2b_shift;
3225         }
3226         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
3227         q->limits.max_dev_sectors = max;
3228         blk_queue_logical_block_size(q, logical_block_size);
3229         blk_queue_max_hw_sectors(q, max);
3230         blk_queue_max_segments(q, USHRT_MAX);
3231         /* with page sized segments we can translate each segement into
3232          * one idaw/tidaw
3233          */
3234         blk_queue_max_segment_size(q, PAGE_SIZE);
3235         blk_queue_segment_boundary(q, PAGE_SIZE - 1);
3236
3237         /* Only activate blocklayer discard support for devices that support it */
3238         if (block->base->features & DASD_FEATURE_DISCARD) {
3239                 q->limits.discard_granularity = logical_block_size;
3240                 q->limits.discard_alignment = PAGE_SIZE;
3241
3242                 /* Calculate max_discard_sectors and make it PAGE aligned */
3243                 max_bytes = USHRT_MAX * logical_block_size;
3244                 max_bytes = ALIGN(max_bytes, PAGE_SIZE) - PAGE_SIZE;
3245                 max_discard_sectors = max_bytes / logical_block_size;
3246
3247                 blk_queue_max_discard_sectors(q, max_discard_sectors);
3248                 blk_queue_max_write_zeroes_sectors(q, max_discard_sectors);
3249                 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
3250         }
3251 }
3252
3253 /*
3254  * Deactivate and free request queue.
3255  */
3256 static void dasd_free_queue(struct dasd_block *block)
3257 {
3258         if (block->request_queue) {
3259                 blk_cleanup_queue(block->request_queue);
3260                 blk_mq_free_tag_set(&block->tag_set);
3261                 block->request_queue = NULL;
3262         }
3263 }
3264
3265 static int dasd_open(struct block_device *bdev, fmode_t mode)
3266 {
3267         struct dasd_device *base;
3268         int rc;
3269
3270         base = dasd_device_from_gendisk(bdev->bd_disk);
3271         if (!base)
3272                 return -ENODEV;
3273
3274         atomic_inc(&base->block->open_count);
3275         if (test_bit(DASD_FLAG_OFFLINE, &base->flags)) {
3276                 rc = -ENODEV;
3277                 goto unlock;
3278         }
3279
3280         if (!try_module_get(base->discipline->owner)) {
3281                 rc = -EINVAL;
3282                 goto unlock;
3283         }
3284
3285         if (dasd_probeonly) {
3286                 dev_info(&base->cdev->dev,
3287                          "Accessing the DASD failed because it is in "
3288                          "probeonly mode\n");
3289                 rc = -EPERM;
3290                 goto out;
3291         }
3292
3293         if (base->state <= DASD_STATE_BASIC) {
3294                 DBF_DEV_EVENT(DBF_ERR, base, " %s",
3295                               " Cannot open unrecognized device");
3296                 rc = -ENODEV;
3297                 goto out;
3298         }
3299
3300         if ((mode & FMODE_WRITE) &&
3301             (test_bit(DASD_FLAG_DEVICE_RO, &base->flags) ||
3302              (base->features & DASD_FEATURE_READONLY))) {
3303                 rc = -EROFS;
3304                 goto out;
3305         }
3306
3307         dasd_put_device(base);
3308         return 0;
3309
3310 out:
3311         module_put(base->discipline->owner);
3312 unlock:
3313         atomic_dec(&base->block->open_count);
3314         dasd_put_device(base);
3315         return rc;
3316 }
3317
3318 static void dasd_release(struct gendisk *disk, fmode_t mode)
3319 {
3320         struct dasd_device *base = dasd_device_from_gendisk(disk);
3321         if (base) {
3322                 atomic_dec(&base->block->open_count);
3323                 module_put(base->discipline->owner);
3324                 dasd_put_device(base);
3325         }
3326 }
3327
3328 /*
3329  * Return disk geometry.
3330  */
3331 static int dasd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
3332 {
3333         struct dasd_device *base;
3334
3335         base = dasd_device_from_gendisk(bdev->bd_disk);
3336         if (!base)
3337                 return -ENODEV;
3338
3339         if (!base->discipline ||
3340             !base->discipline->fill_geometry) {
3341                 dasd_put_device(base);
3342                 return -EINVAL;
3343         }
3344         base->discipline->fill_geometry(base->block, geo);
3345         geo->start = get_start_sect(bdev) >> base->block->s2b_shift;
3346         dasd_put_device(base);
3347         return 0;
3348 }
3349