Merge branch 'for-linus' of git://oss.sgi.com:8090/xfs/xfs-2.6
[linux-2.6] / drivers / s390 / cio / cmf.c
1 /*
2  * linux/drivers/s390/cio/cmf.c
3  *
4  * Linux on zSeries Channel Measurement Facility support
5  *
6  * Copyright 2000,2006 IBM Corporation
7  *
8  * Authors: Arnd Bergmann <arndb@de.ibm.com>
9  *          Cornelia Huck <cornelia.huck@de.ibm.com>
10  *
11  * original idea from Natarajan Krishnaswami <nkrishna@us.ibm.com>
12  *
13  * This program is free software; you can redistribute it and/or modify
14  * it under the terms of the GNU General Public License as published by
15  * the Free Software Foundation; either version 2, or (at your option)
16  * any later version.
17  *
18  * This program is distributed in the hope that it will be useful,
19  * but WITHOUT ANY WARRANTY; without even the implied warranty of
20  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
21  * GNU General Public License for more details.
22  *
23  * You should have received a copy of the GNU General Public License
24  * along with this program; if not, write to the Free Software
25  * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
26  */
27
28 #include <linux/bootmem.h>
29 #include <linux/device.h>
30 #include <linux/init.h>
31 #include <linux/list.h>
32 #include <linux/module.h>
33 #include <linux/moduleparam.h>
34 #include <linux/slab.h>
35 #include <linux/timex.h>        /* get_clock() */
36
37 #include <asm/ccwdev.h>
38 #include <asm/cio.h>
39 #include <asm/cmb.h>
40 #include <asm/div64.h>
41
42 #include "cio.h"
43 #include "css.h"
44 #include "device.h"
45 #include "ioasm.h"
46 #include "chsc.h"
47
48 /*
49  * parameter to enable cmf during boot, possible uses are:
50  *  "s390cmf" -- enable cmf and allocate 2 MB of ram so measuring can be
51  *               used on any subchannel
52  *  "s390cmf=<num>" -- enable cmf and allocate enough memory to measure
53  *                     <num> subchannel, where <num> is an integer
54  *                     between 1 and 65535, default is 1024
55  */
56 #define ARGSTRING "s390cmf"
57
58 /* indices for READCMB */
59 enum cmb_index {
60  /* basic and exended format: */
61         cmb_ssch_rsch_count,
62         cmb_sample_count,
63         cmb_device_connect_time,
64         cmb_function_pending_time,
65         cmb_device_disconnect_time,
66         cmb_control_unit_queuing_time,
67         cmb_device_active_only_time,
68  /* extended format only: */
69         cmb_device_busy_time,
70         cmb_initial_command_response_time,
71 };
72
73 /**
74  * enum cmb_format - types of supported measurement block formats
75  *
76  * @CMF_BASIC:      traditional channel measurement blocks supported
77  *                  by all machines that we run on
78  * @CMF_EXTENDED:   improved format that was introduced with the z990
79  *                  machine
80  * @CMF_AUTODETECT: default: use extended format when running on a machine
81  *                  supporting extended format, otherwise fall back to
82  *                  basic format
83  */
84 enum cmb_format {
85         CMF_BASIC,
86         CMF_EXTENDED,
87         CMF_AUTODETECT = -1,
88 };
89
90 /*
91  * format - actual format for all measurement blocks
92  *
93  * The format module parameter can be set to a value of 0 (zero)
94  * or 1, indicating basic or extended format as described for
95  * enum cmb_format.
96  */
97 static int format = CMF_AUTODETECT;
98 module_param(format, bool, 0444);
99
100 /**
101  * struct cmb_operations - functions to use depending on cmb_format
102  *
103  * Most of these functions operate on a struct ccw_device. There is only
104  * one instance of struct cmb_operations because the format of the measurement
105  * data is guaranteed to be the same for every ccw_device.
106  *
107  * @alloc:      allocate memory for a channel measurement block,
108  *              either with the help of a special pool or with kmalloc
109  * @free:       free memory allocated with @alloc
110  * @set:        enable or disable measurement
111  * @read:       read a measurement entry at an index
112  * @readall:    read a measurement block in a common format
113  * @reset:      clear the data in the associated measurement block and
114  *              reset its time stamp
115  * @align:      align an allocated block so that the hardware can use it
116  */
117 struct cmb_operations {
118         int  (*alloc)  (struct ccw_device *);
119         void (*free)   (struct ccw_device *);
120         int  (*set)    (struct ccw_device *, u32);
121         u64  (*read)   (struct ccw_device *, int);
122         int  (*readall)(struct ccw_device *, struct cmbdata *);
123         void (*reset)  (struct ccw_device *);
124         void *(*align) (void *);
125 /* private: */
126         struct attribute_group *attr_group;
127 };
128 static struct cmb_operations *cmbops;
129
130 struct cmb_data {
131         void *hw_block;   /* Pointer to block updated by hardware */
132         void *last_block; /* Last changed block copied from hardware block */
133         int size;         /* Size of hw_block and last_block */
134         unsigned long long last_update;  /* when last_block was updated */
135 };
136
137 /*
138  * Our user interface is designed in terms of nanoseconds,
139  * while the hardware measures total times in its own
140  * unit.
141  */
142 static inline u64 time_to_nsec(u32 value)
143 {
144         return ((u64)value) * 128000ull;
145 }
146
147 /*
148  * Users are usually interested in average times,
149  * not accumulated time.
150  * This also helps us with atomicity problems
151  * when reading sinlge values.
152  */
153 static inline u64 time_to_avg_nsec(u32 value, u32 count)
154 {
155         u64 ret;
156
157         /* no samples yet, avoid division by 0 */
158         if (count == 0)
159                 return 0;
160
161         /* value comes in units of 128 µsec */
162         ret = time_to_nsec(value);
163         do_div(ret, count);
164
165         return ret;
166 }
167
168 /*
169  * Activate or deactivate the channel monitor. When area is NULL,
170  * the monitor is deactivated. The channel monitor needs to
171  * be active in order to measure subchannels, which also need
172  * to be enabled.
173  */
174 static inline void cmf_activate(void *area, unsigned int onoff)
175 {
176         register void * __gpr2 asm("2");
177         register long __gpr1 asm("1");
178
179         __gpr2 = area;
180         __gpr1 = onoff ? 2 : 0;
181         /* activate channel measurement */
182         asm("schm" : : "d" (__gpr2), "d" (__gpr1) );
183 }
184
185 static int set_schib(struct ccw_device *cdev, u32 mme, int mbfc,
186                      unsigned long address)
187 {
188         int ret;
189         int retry;
190         struct subchannel *sch;
191         struct schib *schib;
192
193         sch = to_subchannel(cdev->dev.parent);
194         schib = &sch->schib;
195         /* msch can silently fail, so do it again if necessary */
196         for (retry = 0; retry < 3; retry++) {
197                 /* prepare schib */
198                 stsch(sch->schid, schib);
199                 schib->pmcw.mme  = mme;
200                 schib->pmcw.mbfc = mbfc;
201                 /* address can be either a block address or a block index */
202                 if (mbfc)
203                         schib->mba = address;
204                 else
205                         schib->pmcw.mbi = address;
206
207                 /* try to submit it */
208                 switch(ret = msch_err(sch->schid, schib)) {
209                         case 0:
210                                 break;
211                         case 1:
212                         case 2: /* in I/O or status pending */
213                                 ret = -EBUSY;
214                                 break;
215                         case 3: /* subchannel is no longer valid */
216                                 ret = -ENODEV;
217                                 break;
218                         default: /* msch caught an exception */
219                                 ret = -EINVAL;
220                                 break;
221                 }
222                 stsch(sch->schid, schib); /* restore the schib */
223
224                 if (ret)
225                         break;
226
227                 /* check if it worked */
228                 if (schib->pmcw.mme  == mme &&
229                     schib->pmcw.mbfc == mbfc &&
230                     (mbfc ? (schib->mba == address)
231                           : (schib->pmcw.mbi == address)))
232                         return 0;
233
234                 ret = -EINVAL;
235         }
236
237         return ret;
238 }
239
240 struct set_schib_struct {
241         u32 mme;
242         int mbfc;
243         unsigned long address;
244         wait_queue_head_t wait;
245         int ret;
246         struct kref kref;
247 };
248
249 static void cmf_set_schib_release(struct kref *kref)
250 {
251         struct set_schib_struct *set_data;
252
253         set_data = container_of(kref, struct set_schib_struct, kref);
254         kfree(set_data);
255 }
256
257 #define CMF_PENDING 1
258
259 static int set_schib_wait(struct ccw_device *cdev, u32 mme,
260                                 int mbfc, unsigned long address)
261 {
262         struct set_schib_struct *set_data;
263         int ret;
264
265         spin_lock_irq(cdev->ccwlock);
266         if (!cdev->private->cmb) {
267                 ret = -ENODEV;
268                 goto out;
269         }
270         set_data = kzalloc(sizeof(struct set_schib_struct), GFP_ATOMIC);
271         if (!set_data) {
272                 ret = -ENOMEM;
273                 goto out;
274         }
275         init_waitqueue_head(&set_data->wait);
276         kref_init(&set_data->kref);
277         set_data->mme = mme;
278         set_data->mbfc = mbfc;
279         set_data->address = address;
280
281         ret = set_schib(cdev, mme, mbfc, address);
282         if (ret != -EBUSY)
283                 goto out_put;
284
285         if (cdev->private->state != DEV_STATE_ONLINE) {
286                 /* if the device is not online, don't even try again */
287                 ret = -EBUSY;
288                 goto out_put;
289         }
290
291         cdev->private->state = DEV_STATE_CMFCHANGE;
292         set_data->ret = CMF_PENDING;
293         cdev->private->cmb_wait = set_data;
294
295         spin_unlock_irq(cdev->ccwlock);
296         if (wait_event_interruptible(set_data->wait,
297                                      set_data->ret != CMF_PENDING)) {
298                 spin_lock_irq(cdev->ccwlock);
299                 if (set_data->ret == CMF_PENDING) {
300                         set_data->ret = -ERESTARTSYS;
301                         if (cdev->private->state == DEV_STATE_CMFCHANGE)
302                                 cdev->private->state = DEV_STATE_ONLINE;
303                 }
304                 spin_unlock_irq(cdev->ccwlock);
305         }
306         spin_lock_irq(cdev->ccwlock);
307         cdev->private->cmb_wait = NULL;
308         ret = set_data->ret;
309 out_put:
310         kref_put(&set_data->kref, cmf_set_schib_release);
311 out:
312         spin_unlock_irq(cdev->ccwlock);
313         return ret;
314 }
315
316 void retry_set_schib(struct ccw_device *cdev)
317 {
318         struct set_schib_struct *set_data;
319
320         set_data = cdev->private->cmb_wait;
321         if (!set_data) {
322                 WARN_ON(1);
323                 return;
324         }
325         kref_get(&set_data->kref);
326         set_data->ret = set_schib(cdev, set_data->mme, set_data->mbfc,
327                                   set_data->address);
328         wake_up(&set_data->wait);
329         kref_put(&set_data->kref, cmf_set_schib_release);
330 }
331
332 static int cmf_copy_block(struct ccw_device *cdev)
333 {
334         struct subchannel *sch;
335         void *reference_buf;
336         void *hw_block;
337         struct cmb_data *cmb_data;
338
339         sch = to_subchannel(cdev->dev.parent);
340
341         if (stsch(sch->schid, &sch->schib))
342                 return -ENODEV;
343
344         if (sch->schib.scsw.fctl & SCSW_FCTL_START_FUNC) {
345                 /* Don't copy if a start function is in progress. */
346                 if ((!(sch->schib.scsw.actl & SCSW_ACTL_SUSPENDED)) &&
347                     (sch->schib.scsw.actl &
348                      (SCSW_ACTL_DEVACT | SCSW_ACTL_SCHACT)) &&
349                     (!(sch->schib.scsw.stctl & SCSW_STCTL_SEC_STATUS)))
350                         return -EBUSY;
351         }
352         cmb_data = cdev->private->cmb;
353         hw_block = cmbops->align(cmb_data->hw_block);
354         if (!memcmp(cmb_data->last_block, hw_block, cmb_data->size))
355                 /* No need to copy. */
356                 return 0;
357         reference_buf = kzalloc(cmb_data->size, GFP_ATOMIC);
358         if (!reference_buf)
359                 return -ENOMEM;
360         /* Ensure consistency of block copied from hardware. */
361         do {
362                 memcpy(cmb_data->last_block, hw_block, cmb_data->size);
363                 memcpy(reference_buf, hw_block, cmb_data->size);
364         } while (memcmp(cmb_data->last_block, reference_buf, cmb_data->size));
365         cmb_data->last_update = get_clock();
366         kfree(reference_buf);
367         return 0;
368 }
369
370 struct copy_block_struct {
371         wait_queue_head_t wait;
372         int ret;
373         struct kref kref;
374 };
375
376 static void cmf_copy_block_release(struct kref *kref)
377 {
378         struct copy_block_struct *copy_block;
379
380         copy_block = container_of(kref, struct copy_block_struct, kref);
381         kfree(copy_block);
382 }
383
384 static int cmf_cmb_copy_wait(struct ccw_device *cdev)
385 {
386         struct copy_block_struct *copy_block;
387         int ret;
388         unsigned long flags;
389
390         spin_lock_irqsave(cdev->ccwlock, flags);
391         if (!cdev->private->cmb) {
392                 ret = -ENODEV;
393                 goto out;
394         }
395         copy_block = kzalloc(sizeof(struct copy_block_struct), GFP_ATOMIC);
396         if (!copy_block) {
397                 ret = -ENOMEM;
398                 goto out;
399         }
400         init_waitqueue_head(&copy_block->wait);
401         kref_init(&copy_block->kref);
402
403         ret = cmf_copy_block(cdev);
404         if (ret != -EBUSY)
405                 goto out_put;
406
407         if (cdev->private->state != DEV_STATE_ONLINE) {
408                 ret = -EBUSY;
409                 goto out_put;
410         }
411
412         cdev->private->state = DEV_STATE_CMFUPDATE;
413         copy_block->ret = CMF_PENDING;
414         cdev->private->cmb_wait = copy_block;
415
416         spin_unlock_irqrestore(cdev->ccwlock, flags);
417         if (wait_event_interruptible(copy_block->wait,
418                                      copy_block->ret != CMF_PENDING)) {
419                 spin_lock_irqsave(cdev->ccwlock, flags);
420                 if (copy_block->ret == CMF_PENDING) {
421                         copy_block->ret = -ERESTARTSYS;
422                         if (cdev->private->state == DEV_STATE_CMFUPDATE)
423                                 cdev->private->state = DEV_STATE_ONLINE;
424                 }
425                 spin_unlock_irqrestore(cdev->ccwlock, flags);
426         }
427         spin_lock_irqsave(cdev->ccwlock, flags);
428         cdev->private->cmb_wait = NULL;
429         ret = copy_block->ret;
430 out_put:
431         kref_put(&copy_block->kref, cmf_copy_block_release);
432 out:
433         spin_unlock_irqrestore(cdev->ccwlock, flags);
434         return ret;
435 }
436
437 void cmf_retry_copy_block(struct ccw_device *cdev)
438 {
439         struct copy_block_struct *copy_block;
440
441         copy_block = cdev->private->cmb_wait;
442         if (!copy_block) {
443                 WARN_ON(1);
444                 return;
445         }
446         kref_get(&copy_block->kref);
447         copy_block->ret = cmf_copy_block(cdev);
448         wake_up(&copy_block->wait);
449         kref_put(&copy_block->kref, cmf_copy_block_release);
450 }
451
452 static void cmf_generic_reset(struct ccw_device *cdev)
453 {
454         struct cmb_data *cmb_data;
455
456         spin_lock_irq(cdev->ccwlock);
457         cmb_data = cdev->private->cmb;
458         if (cmb_data) {
459                 memset(cmb_data->last_block, 0, cmb_data->size);
460                 /*
461                  * Need to reset hw block as well to make the hardware start
462                  * from 0 again.
463                  */
464                 memset(cmbops->align(cmb_data->hw_block), 0, cmb_data->size);
465                 cmb_data->last_update = 0;
466         }
467         cdev->private->cmb_start_time = get_clock();
468         spin_unlock_irq(cdev->ccwlock);
469 }
470
471 /**
472  * struct cmb_area - container for global cmb data
473  *
474  * @mem:        pointer to CMBs (only in basic measurement mode)
475  * @list:       contains a linked list of all subchannels
476  * @num_channels: number of channels to be measured
477  * @lock:       protect concurrent access to @mem and @list
478  */
479 struct cmb_area {
480         struct cmb *mem;
481         struct list_head list;
482         int num_channels;
483         spinlock_t lock;
484 };
485
486 static struct cmb_area cmb_area = {
487         .lock = __SPIN_LOCK_UNLOCKED(cmb_area.lock),
488         .list = LIST_HEAD_INIT(cmb_area.list),
489         .num_channels  = 1024,
490 };
491
492 /* ****** old style CMB handling ********/
493
494 /*
495  * Basic channel measurement blocks are allocated in one contiguous
496  * block of memory, which can not be moved as long as any channel
497  * is active. Therefore, a maximum number of subchannels needs to
498  * be defined somewhere. This is a module parameter, defaulting to
499  * a resonable value of 1024, or 32 kb of memory.
500  * Current kernels don't allow kmalloc with more than 128kb, so the
501  * maximum is 4096.
502  */
503
504 module_param_named(maxchannels, cmb_area.num_channels, uint, 0444);
505
506 /**
507  * struct cmb - basic channel measurement block
508  * @ssch_rsch_count: number of ssch and rsch
509  * @sample_count: number of samples
510  * @device_connect_time: time of device connect
511  * @function_pending_time: time of function pending
512  * @device_disconnect_time: time of device disconnect
513  * @control_unit_queuing_time: time of control unit queuing
514  * @device_active_only_time: time of device active only
515  * @reserved: unused in basic measurement mode
516  *
517  * The measurement block as used by the hardware. The fields are described
518  * further in z/Architecture Principles of Operation, chapter 17.
519  *
520  * The cmb area made up from these blocks must be a contiguous array and may
521  * not be reallocated or freed.
522  * Only one cmb area can be present in the system.
523  */
524 struct cmb {
525         u16 ssch_rsch_count;
526         u16 sample_count;
527         u32 device_connect_time;
528         u32 function_pending_time;
529         u32 device_disconnect_time;
530         u32 control_unit_queuing_time;
531         u32 device_active_only_time;
532         u32 reserved[2];
533 };
534
535 /*
536  * Insert a single device into the cmb_area list.
537  * Called with cmb_area.lock held from alloc_cmb.
538  */
539 static int alloc_cmb_single(struct ccw_device *cdev,
540                             struct cmb_data *cmb_data)
541 {
542         struct cmb *cmb;
543         struct ccw_device_private *node;
544         int ret;
545
546         spin_lock_irq(cdev->ccwlock);
547         if (!list_empty(&cdev->private->cmb_list)) {
548                 ret = -EBUSY;
549                 goto out;
550         }
551
552         /*
553          * Find first unused cmb in cmb_area.mem.
554          * This is a little tricky: cmb_area.list
555          * remains sorted by ->cmb->hw_data pointers.
556          */
557         cmb = cmb_area.mem;
558         list_for_each_entry(node, &cmb_area.list, cmb_list) {
559                 struct cmb_data *data;
560                 data = node->cmb;
561                 if ((struct cmb*)data->hw_block > cmb)
562                         break;
563                 cmb++;
564         }
565         if (cmb - cmb_area.mem >= cmb_area.num_channels) {
566                 ret = -ENOMEM;
567                 goto out;
568         }
569
570         /* insert new cmb */
571         list_add_tail(&cdev->private->cmb_list, &node->cmb_list);
572         cmb_data->hw_block = cmb;
573         cdev->private->cmb = cmb_data;
574         ret = 0;
575 out:
576         spin_unlock_irq(cdev->ccwlock);
577         return ret;
578 }
579
580 static int alloc_cmb(struct ccw_device *cdev)
581 {
582         int ret;
583         struct cmb *mem;
584         ssize_t size;
585         struct cmb_data *cmb_data;
586
587         /* Allocate private cmb_data. */
588         cmb_data = kzalloc(sizeof(struct cmb_data), GFP_KERNEL);
589         if (!cmb_data)
590                 return -ENOMEM;
591
592         cmb_data->last_block = kzalloc(sizeof(struct cmb), GFP_KERNEL);
593         if (!cmb_data->last_block) {
594                 kfree(cmb_data);
595                 return -ENOMEM;
596         }
597         cmb_data->size = sizeof(struct cmb);
598         spin_lock(&cmb_area.lock);
599
600         if (!cmb_area.mem) {
601                 /* there is no user yet, so we need a new area */
602                 size = sizeof(struct cmb) * cmb_area.num_channels;
603                 WARN_ON(!list_empty(&cmb_area.list));
604
605                 spin_unlock(&cmb_area.lock);
606                 mem = (void*)__get_free_pages(GFP_KERNEL | GFP_DMA,
607                                  get_order(size));
608                 spin_lock(&cmb_area.lock);
609
610                 if (cmb_area.mem) {
611                         /* ok, another thread was faster */
612                         free_pages((unsigned long)mem, get_order(size));
613                 } else if (!mem) {
614                         /* no luck */
615                         printk(KERN_WARNING "cio: failed to allocate area "
616                                "for measuring %d subchannels\n",
617                                cmb_area.num_channels);
618                         ret = -ENOMEM;
619                         goto out;
620                 } else {
621                         /* everything ok */
622                         memset(mem, 0, size);
623                         cmb_area.mem = mem;
624                         cmf_activate(cmb_area.mem, 1);
625                 }
626         }
627
628         /* do the actual allocation */
629         ret = alloc_cmb_single(cdev, cmb_data);
630 out:
631         spin_unlock(&cmb_area.lock);
632         if (ret) {
633                 kfree(cmb_data->last_block);
634                 kfree(cmb_data);
635         }
636         return ret;
637 }
638
639 static void free_cmb(struct ccw_device *cdev)
640 {
641         struct ccw_device_private *priv;
642         struct cmb_data *cmb_data;
643
644         spin_lock(&cmb_area.lock);
645         spin_lock_irq(cdev->ccwlock);
646
647         priv = cdev->private;
648
649         if (list_empty(&priv->cmb_list)) {
650                 /* already freed */
651                 goto out;
652         }
653
654         cmb_data = priv->cmb;
655         priv->cmb = NULL;
656         if (cmb_data)
657                 kfree(cmb_data->last_block);
658         kfree(cmb_data);
659         list_del_init(&priv->cmb_list);
660
661         if (list_empty(&cmb_area.list)) {
662                 ssize_t size;
663                 size = sizeof(struct cmb) * cmb_area.num_channels;
664                 cmf_activate(NULL, 0);
665                 free_pages((unsigned long)cmb_area.mem, get_order(size));
666                 cmb_area.mem = NULL;
667         }
668 out:
669         spin_unlock_irq(cdev->ccwlock);
670         spin_unlock(&cmb_area.lock);
671 }
672
673 static int set_cmb(struct ccw_device *cdev, u32 mme)
674 {
675         u16 offset;
676         struct cmb_data *cmb_data;
677         unsigned long flags;
678
679         spin_lock_irqsave(cdev->ccwlock, flags);
680         if (!cdev->private->cmb) {
681                 spin_unlock_irqrestore(cdev->ccwlock, flags);
682                 return -EINVAL;
683         }
684         cmb_data = cdev->private->cmb;
685         offset = mme ? (struct cmb *)cmb_data->hw_block - cmb_area.mem : 0;
686         spin_unlock_irqrestore(cdev->ccwlock, flags);
687
688         return set_schib_wait(cdev, mme, 0, offset);
689 }
690
691 static u64 read_cmb(struct ccw_device *cdev, int index)
692 {
693         struct cmb *cmb;
694         u32 val;
695         int ret;
696         unsigned long flags;
697
698         ret = cmf_cmb_copy_wait(cdev);
699         if (ret < 0)
700                 return 0;
701
702         spin_lock_irqsave(cdev->ccwlock, flags);
703         if (!cdev->private->cmb) {
704                 ret = 0;
705                 goto out;
706         }
707         cmb = ((struct cmb_data *)cdev->private->cmb)->last_block;
708
709         switch (index) {
710         case cmb_ssch_rsch_count:
711                 ret = cmb->ssch_rsch_count;
712                 goto out;
713         case cmb_sample_count:
714                 ret = cmb->sample_count;
715                 goto out;
716         case cmb_device_connect_time:
717                 val = cmb->device_connect_time;
718                 break;
719         case cmb_function_pending_time:
720                 val = cmb->function_pending_time;
721                 break;
722         case cmb_device_disconnect_time:
723                 val = cmb->device_disconnect_time;
724                 break;
725         case cmb_control_unit_queuing_time:
726                 val = cmb->control_unit_queuing_time;
727                 break;
728         case cmb_device_active_only_time:
729                 val = cmb->device_active_only_time;
730                 break;
731         default:
732                 ret = 0;
733                 goto out;
734         }
735         ret = time_to_avg_nsec(val, cmb->sample_count);
736 out:
737         spin_unlock_irqrestore(cdev->ccwlock, flags);
738         return ret;
739 }
740
741 static int readall_cmb(struct ccw_device *cdev, struct cmbdata *data)
742 {
743         struct cmb *cmb;
744         struct cmb_data *cmb_data;
745         u64 time;
746         unsigned long flags;
747         int ret;
748
749         ret = cmf_cmb_copy_wait(cdev);
750         if (ret < 0)
751                 return ret;
752         spin_lock_irqsave(cdev->ccwlock, flags);
753         cmb_data = cdev->private->cmb;
754         if (!cmb_data) {
755                 ret = -ENODEV;
756                 goto out;
757         }
758         if (cmb_data->last_update == 0) {
759                 ret = -EAGAIN;
760                 goto out;
761         }
762         cmb = cmb_data->last_block;
763         time = cmb_data->last_update - cdev->private->cmb_start_time;
764
765         memset(data, 0, sizeof(struct cmbdata));
766
767         /* we only know values before device_busy_time */
768         data->size = offsetof(struct cmbdata, device_busy_time);
769
770         /* convert to nanoseconds */
771         data->elapsed_time = (time * 1000) >> 12;
772
773         /* copy data to new structure */
774         data->ssch_rsch_count = cmb->ssch_rsch_count;
775         data->sample_count = cmb->sample_count;
776
777         /* time fields are converted to nanoseconds while copying */
778         data->device_connect_time = time_to_nsec(cmb->device_connect_time);
779         data->function_pending_time = time_to_nsec(cmb->function_pending_time);
780         data->device_disconnect_time =
781                 time_to_nsec(cmb->device_disconnect_time);
782         data->control_unit_queuing_time
783                 = time_to_nsec(cmb->control_unit_queuing_time);
784         data->device_active_only_time
785                 = time_to_nsec(cmb->device_active_only_time);
786         ret = 0;
787 out:
788         spin_unlock_irqrestore(cdev->ccwlock, flags);
789         return ret;
790 }
791
792 static void reset_cmb(struct ccw_device *cdev)
793 {
794         cmf_generic_reset(cdev);
795 }
796
797 static void * align_cmb(void *area)
798 {
799         return area;
800 }
801
802 static struct attribute_group cmf_attr_group;
803
804 static struct cmb_operations cmbops_basic = {
805         .alloc  = alloc_cmb,
806         .free   = free_cmb,
807         .set    = set_cmb,
808         .read   = read_cmb,
809         .readall    = readall_cmb,
810         .reset      = reset_cmb,
811         .align      = align_cmb,
812         .attr_group = &cmf_attr_group,
813 };
814
815 /* ******** extended cmb handling ********/
816
817 /**
818  * struct cmbe - extended channel measurement block
819  * @ssch_rsch_count: number of ssch and rsch
820  * @sample_count: number of samples
821  * @device_connect_time: time of device connect
822  * @function_pending_time: time of function pending
823  * @device_disconnect_time: time of device disconnect
824  * @control_unit_queuing_time: time of control unit queuing
825  * @device_active_only_time: time of device active only
826  * @device_busy_time: time of device busy
827  * @initial_command_response_time: initial command response time
828  * @reserved: unused
829  *
830  * The measurement block as used by the hardware. May be in any 64 bit physical
831  * location.
832  * The fields are described further in z/Architecture Principles of Operation,
833  * third edition, chapter 17.
834  */
835 struct cmbe {
836         u32 ssch_rsch_count;
837         u32 sample_count;
838         u32 device_connect_time;
839         u32 function_pending_time;
840         u32 device_disconnect_time;
841         u32 control_unit_queuing_time;
842         u32 device_active_only_time;
843         u32 device_busy_time;
844         u32 initial_command_response_time;
845         u32 reserved[7];
846 };
847
848 /*
849  * kmalloc only guarantees 8 byte alignment, but we need cmbe
850  * pointers to be naturally aligned. Make sure to allocate
851  * enough space for two cmbes.
852  */
853 static inline struct cmbe *cmbe_align(struct cmbe *c)
854 {
855         unsigned long addr;
856         addr = ((unsigned long)c + sizeof (struct cmbe) - sizeof(long)) &
857                                  ~(sizeof (struct cmbe) - sizeof(long));
858         return (struct cmbe*)addr;
859 }
860
861 static int alloc_cmbe(struct ccw_device *cdev)
862 {
863         struct cmbe *cmbe;
864         struct cmb_data *cmb_data;
865         int ret;
866
867         cmbe = kzalloc (sizeof (*cmbe) * 2, GFP_KERNEL);
868         if (!cmbe)
869                 return -ENOMEM;
870         cmb_data = kzalloc(sizeof(struct cmb_data), GFP_KERNEL);
871         if (!cmb_data) {
872                 ret = -ENOMEM;
873                 goto out_free;
874         }
875         cmb_data->last_block = kzalloc(sizeof(struct cmbe), GFP_KERNEL);
876         if (!cmb_data->last_block) {
877                 ret = -ENOMEM;
878                 goto out_free;
879         }
880         cmb_data->size = sizeof(struct cmbe);
881         spin_lock_irq(cdev->ccwlock);
882         if (cdev->private->cmb) {
883                 spin_unlock_irq(cdev->ccwlock);
884                 ret = -EBUSY;
885                 goto out_free;
886         }
887         cmb_data->hw_block = cmbe;
888         cdev->private->cmb = cmb_data;
889         spin_unlock_irq(cdev->ccwlock);
890
891         /* activate global measurement if this is the first channel */
892         spin_lock(&cmb_area.lock);
893         if (list_empty(&cmb_area.list))
894                 cmf_activate(NULL, 1);
895         list_add_tail(&cdev->private->cmb_list, &cmb_area.list);
896         spin_unlock(&cmb_area.lock);
897
898         return 0;
899 out_free:
900         if (cmb_data)
901                 kfree(cmb_data->last_block);
902         kfree(cmb_data);
903         kfree(cmbe);
904         return ret;
905 }
906
907 static void free_cmbe(struct ccw_device *cdev)
908 {
909         struct cmb_data *cmb_data;
910
911         spin_lock_irq(cdev->ccwlock);
912         cmb_data = cdev->private->cmb;
913         cdev->private->cmb = NULL;
914         if (cmb_data)
915                 kfree(cmb_data->last_block);
916         kfree(cmb_data);
917         spin_unlock_irq(cdev->ccwlock);
918
919         /* deactivate global measurement if this is the last channel */
920         spin_lock(&cmb_area.lock);
921         list_del_init(&cdev->private->cmb_list);
922         if (list_empty(&cmb_area.list))
923                 cmf_activate(NULL, 0);
924         spin_unlock(&cmb_area.lock);
925 }
926
927 static int set_cmbe(struct ccw_device *cdev, u32 mme)
928 {
929         unsigned long mba;
930         struct cmb_data *cmb_data;
931         unsigned long flags;
932
933         spin_lock_irqsave(cdev->ccwlock, flags);
934         if (!cdev->private->cmb) {
935                 spin_unlock_irqrestore(cdev->ccwlock, flags);
936                 return -EINVAL;
937         }
938         cmb_data = cdev->private->cmb;
939         mba = mme ? (unsigned long) cmbe_align(cmb_data->hw_block) : 0;
940         spin_unlock_irqrestore(cdev->ccwlock, flags);
941
942         return set_schib_wait(cdev, mme, 1, mba);
943 }
944
945
946 static u64 read_cmbe(struct ccw_device *cdev, int index)
947 {
948         struct cmbe *cmb;
949         struct cmb_data *cmb_data;
950         u32 val;
951         int ret;
952         unsigned long flags;
953
954         ret = cmf_cmb_copy_wait(cdev);
955         if (ret < 0)
956                 return 0;
957
958         spin_lock_irqsave(cdev->ccwlock, flags);
959         cmb_data = cdev->private->cmb;
960         if (!cmb_data) {
961                 ret = 0;
962                 goto out;
963         }
964         cmb = cmb_data->last_block;
965
966         switch (index) {
967         case cmb_ssch_rsch_count:
968                 ret = cmb->ssch_rsch_count;
969                 goto out;
970         case cmb_sample_count:
971                 ret = cmb->sample_count;
972                 goto out;
973         case cmb_device_connect_time:
974                 val = cmb->device_connect_time;
975                 break;
976         case cmb_function_pending_time:
977                 val = cmb->function_pending_time;
978                 break;
979         case cmb_device_disconnect_time:
980                 val = cmb->device_disconnect_time;
981                 break;
982         case cmb_control_unit_queuing_time:
983                 val = cmb->control_unit_queuing_time;
984                 break;
985         case cmb_device_active_only_time:
986                 val = cmb->device_active_only_time;
987                 break;
988         case cmb_device_busy_time:
989                 val = cmb->device_busy_time;
990                 break;
991         case cmb_initial_command_response_time:
992                 val = cmb->initial_command_response_time;
993                 break;
994         default:
995                 ret = 0;
996                 goto out;
997         }
998         ret = time_to_avg_nsec(val, cmb->sample_count);
999 out:
1000         spin_unlock_irqrestore(cdev->ccwlock, flags);
1001         return ret;
1002 }
1003
1004 static int readall_cmbe(struct ccw_device *cdev, struct cmbdata *data)
1005 {
1006         struct cmbe *cmb;
1007         struct cmb_data *cmb_data;
1008         u64 time;
1009         unsigned long flags;
1010         int ret;
1011
1012         ret = cmf_cmb_copy_wait(cdev);
1013         if (ret < 0)
1014                 return ret;
1015         spin_lock_irqsave(cdev->ccwlock, flags);
1016         cmb_data = cdev->private->cmb;
1017         if (!cmb_data) {
1018                 ret = -ENODEV;
1019                 goto out;
1020         }
1021         if (cmb_data->last_update == 0) {
1022                 ret = -EAGAIN;
1023                 goto out;
1024         }
1025         time = cmb_data->last_update - cdev->private->cmb_start_time;
1026
1027         memset (data, 0, sizeof(struct cmbdata));
1028
1029         /* we only know values before device_busy_time */
1030         data->size = offsetof(struct cmbdata, device_busy_time);
1031
1032         /* conver to nanoseconds */
1033         data->elapsed_time = (time * 1000) >> 12;
1034
1035         cmb = cmb_data->last_block;
1036         /* copy data to new structure */
1037         data->ssch_rsch_count = cmb->ssch_rsch_count;
1038         data->sample_count = cmb->sample_count;
1039
1040         /* time fields are converted to nanoseconds while copying */
1041         data->device_connect_time = time_to_nsec(cmb->device_connect_time);
1042         data->function_pending_time = time_to_nsec(cmb->function_pending_time);
1043         data->device_disconnect_time =
1044                 time_to_nsec(cmb->device_disconnect_time);
1045         data->control_unit_queuing_time
1046                 = time_to_nsec(cmb->control_unit_queuing_time);
1047         data->device_active_only_time
1048                 = time_to_nsec(cmb->device_active_only_time);
1049         data->device_busy_time = time_to_nsec(cmb->device_busy_time);
1050         data->initial_command_response_time
1051                 = time_to_nsec(cmb->initial_command_response_time);
1052
1053         ret = 0;
1054 out:
1055         spin_unlock_irqrestore(cdev->ccwlock, flags);
1056         return ret;
1057 }
1058
1059 static void reset_cmbe(struct ccw_device *cdev)
1060 {
1061         cmf_generic_reset(cdev);
1062 }
1063
1064 static void * align_cmbe(void *area)
1065 {
1066         return cmbe_align(area);
1067 }
1068
1069 static struct attribute_group cmf_attr_group_ext;
1070
1071 static struct cmb_operations cmbops_extended = {
1072         .alloc      = alloc_cmbe,
1073         .free       = free_cmbe,
1074         .set        = set_cmbe,
1075         .read       = read_cmbe,
1076         .readall    = readall_cmbe,
1077         .reset      = reset_cmbe,
1078         .align      = align_cmbe,
1079         .attr_group = &cmf_attr_group_ext,
1080 };
1081
1082 static ssize_t cmb_show_attr(struct device *dev, char *buf, enum cmb_index idx)
1083 {
1084         return sprintf(buf, "%lld\n",
1085                 (unsigned long long) cmf_read(to_ccwdev(dev), idx));
1086 }
1087
1088 static ssize_t cmb_show_avg_sample_interval(struct device *dev,
1089                                             struct device_attribute *attr,
1090                                             char *buf)
1091 {
1092         struct ccw_device *cdev;
1093         long interval;
1094         unsigned long count;
1095         struct cmb_data *cmb_data;
1096
1097         cdev = to_ccwdev(dev);
1098         count = cmf_read(cdev, cmb_sample_count);
1099         spin_lock_irq(cdev->ccwlock);
1100         cmb_data = cdev->private->cmb;
1101         if (count) {
1102                 interval = cmb_data->last_update -
1103                         cdev->private->cmb_start_time;
1104                 interval = (interval * 1000) >> 12;
1105                 interval /= count;
1106         } else
1107                 interval = -1;
1108         spin_unlock_irq(cdev->ccwlock);
1109         return sprintf(buf, "%ld\n", interval);
1110 }
1111
1112 static ssize_t cmb_show_avg_utilization(struct device *dev,
1113                                         struct device_attribute *attr,
1114                                         char *buf)
1115 {
1116         struct cmbdata data;
1117         u64 utilization;
1118         unsigned long t, u;
1119         int ret;
1120
1121         ret = cmf_readall(to_ccwdev(dev), &data);
1122         if (ret == -EAGAIN || ret == -ENODEV)
1123                 /* No data (yet/currently) available to use for calculation. */
1124                 return sprintf(buf, "n/a\n");
1125         else if (ret)
1126                 return ret;
1127
1128         utilization = data.device_connect_time +
1129                       data.function_pending_time +
1130                       data.device_disconnect_time;
1131
1132         /* shift to avoid long long division */
1133         while (-1ul < (data.elapsed_time | utilization)) {
1134                 utilization >>= 8;
1135                 data.elapsed_time >>= 8;
1136         }
1137
1138         /* calculate value in 0.1 percent units */
1139         t = (unsigned long) data.elapsed_time / 1000;
1140         u = (unsigned long) utilization / t;
1141
1142         return sprintf(buf, "%02ld.%01ld%%\n", u/ 10, u - (u/ 10) * 10);
1143 }
1144
1145 #define cmf_attr(name) \
1146 static ssize_t show_##name(struct device *dev, \
1147                            struct device_attribute *attr, char *buf)    \
1148 { return cmb_show_attr((dev), buf, cmb_##name); } \
1149 static DEVICE_ATTR(name, 0444, show_##name, NULL);
1150
1151 #define cmf_attr_avg(name) \
1152 static ssize_t show_avg_##name(struct device *dev, \
1153                                struct device_attribute *attr, char *buf) \
1154 { return cmb_show_attr((dev), buf, cmb_##name); } \
1155 static DEVICE_ATTR(avg_##name, 0444, show_avg_##name, NULL);
1156
1157 cmf_attr(ssch_rsch_count);
1158 cmf_attr(sample_count);
1159 cmf_attr_avg(device_connect_time);
1160 cmf_attr_avg(function_pending_time);
1161 cmf_attr_avg(device_disconnect_time);
1162 cmf_attr_avg(control_unit_queuing_time);
1163 cmf_attr_avg(device_active_only_time);
1164 cmf_attr_avg(device_busy_time);
1165 cmf_attr_avg(initial_command_response_time);
1166
1167 static DEVICE_ATTR(avg_sample_interval, 0444, cmb_show_avg_sample_interval,
1168                    NULL);
1169 static DEVICE_ATTR(avg_utilization, 0444, cmb_show_avg_utilization, NULL);
1170
1171 static struct attribute *cmf_attributes[] = {
1172         &dev_attr_avg_sample_interval.attr,
1173         &dev_attr_avg_utilization.attr,
1174         &dev_attr_ssch_rsch_count.attr,
1175         &dev_attr_sample_count.attr,
1176         &dev_attr_avg_device_connect_time.attr,
1177         &dev_attr_avg_function_pending_time.attr,
1178         &dev_attr_avg_device_disconnect_time.attr,
1179         &dev_attr_avg_control_unit_queuing_time.attr,
1180         &dev_attr_avg_device_active_only_time.attr,
1181         NULL,
1182 };
1183
1184 static struct attribute_group cmf_attr_group = {
1185         .name  = "cmf",
1186         .attrs = cmf_attributes,
1187 };
1188
1189 static struct attribute *cmf_attributes_ext[] = {
1190         &dev_attr_avg_sample_interval.attr,
1191         &dev_attr_avg_utilization.attr,
1192         &dev_attr_ssch_rsch_count.attr,
1193         &dev_attr_sample_count.attr,
1194         &dev_attr_avg_device_connect_time.attr,
1195         &dev_attr_avg_function_pending_time.attr,
1196         &dev_attr_avg_device_disconnect_time.attr,
1197         &dev_attr_avg_control_unit_queuing_time.attr,
1198         &dev_attr_avg_device_active_only_time.attr,
1199         &dev_attr_avg_device_busy_time.attr,
1200         &dev_attr_avg_initial_command_response_time.attr,
1201         NULL,
1202 };
1203
1204 static struct attribute_group cmf_attr_group_ext = {
1205         .name  = "cmf",
1206         .attrs = cmf_attributes_ext,
1207 };
1208
1209 static ssize_t cmb_enable_show(struct device *dev,
1210                                struct device_attribute *attr,
1211                                char *buf)
1212 {
1213         return sprintf(buf, "%d\n", to_ccwdev(dev)->private->cmb ? 1 : 0);
1214 }
1215
1216 static ssize_t cmb_enable_store(struct device *dev,
1217                                 struct device_attribute *attr, const char *buf,
1218                                 size_t c)
1219 {
1220         struct ccw_device *cdev;
1221         int ret;
1222
1223         cdev = to_ccwdev(dev);
1224
1225         switch (buf[0]) {
1226         case '0':
1227                 ret = disable_cmf(cdev);
1228                 if (ret)
1229                         dev_info(&cdev->dev, "disable_cmf failed (%d)\n", ret);
1230                 break;
1231         case '1':
1232                 ret = enable_cmf(cdev);
1233                 if (ret && ret != -EBUSY)
1234                         dev_info(&cdev->dev, "enable_cmf failed (%d)\n", ret);
1235                 break;
1236         }
1237
1238         return c;
1239 }
1240
1241 DEVICE_ATTR(cmb_enable, 0644, cmb_enable_show, cmb_enable_store);
1242
1243 /**
1244  * enable_cmf() - switch on the channel measurement for a specific device
1245  *  @cdev:      The ccw device to be enabled
1246  *
1247  *  Returns %0 for success or a negative error value.
1248  *
1249  *  Context:
1250  *    non-atomic
1251  */
1252 int enable_cmf(struct ccw_device *cdev)
1253 {
1254         int ret;
1255
1256         ret = cmbops->alloc(cdev);
1257         cmbops->reset(cdev);
1258         if (ret)
1259                 return ret;
1260         ret = cmbops->set(cdev, 2);
1261         if (ret) {
1262                 cmbops->free(cdev);
1263                 return ret;
1264         }
1265         ret = sysfs_create_group(&cdev->dev.kobj, cmbops->attr_group);
1266         if (!ret)
1267                 return 0;
1268         cmbops->set(cdev, 0);  //FIXME: this can fail
1269         cmbops->free(cdev);
1270         return ret;
1271 }
1272
1273 /**
1274  * disable_cmf() - switch off the channel measurement for a specific device
1275  *  @cdev:      The ccw device to be disabled
1276  *
1277  *  Returns %0 for success or a negative error value.
1278  *
1279  *  Context:
1280  *    non-atomic
1281  */
1282 int disable_cmf(struct ccw_device *cdev)
1283 {
1284         int ret;
1285
1286         ret = cmbops->set(cdev, 0);
1287         if (ret)
1288                 return ret;
1289         cmbops->free(cdev);
1290         sysfs_remove_group(&cdev->dev.kobj, cmbops->attr_group);
1291         return ret;
1292 }
1293
1294 /**
1295  * cmf_read() - read one value from the current channel measurement block
1296  * @cdev:       the channel to be read
1297  * @index:      the index of the value to be read
1298  *
1299  * Returns the value read or %0 if the value cannot be read.
1300  *
1301  *  Context:
1302  *    any
1303  */
1304 u64 cmf_read(struct ccw_device *cdev, int index)
1305 {
1306         return cmbops->read(cdev, index);
1307 }
1308
1309 /**
1310  * cmf_readall() - read the current channel measurement block
1311  * @cdev:       the channel to be read
1312  * @data:       a pointer to a data block that will be filled
1313  *
1314  * Returns %0 on success, a negative error value otherwise.
1315  *
1316  *  Context:
1317  *    any
1318  */
1319 int cmf_readall(struct ccw_device *cdev, struct cmbdata *data)
1320 {
1321         return cmbops->readall(cdev, data);
1322 }
1323
1324 /* Reenable cmf when a disconnected device becomes available again. */
1325 int cmf_reenable(struct ccw_device *cdev)
1326 {
1327         cmbops->reset(cdev);
1328         return cmbops->set(cdev, 2);
1329 }
1330
1331 static int __init init_cmf(void)
1332 {
1333         char *format_string;
1334         char *detect_string = "parameter";
1335
1336         /*
1337          * If the user did not give a parameter, see if we are running on a
1338          * machine supporting extended measurement blocks, otherwise fall back
1339          * to basic mode.
1340          */
1341         if (format == CMF_AUTODETECT) {
1342                 if (!css_characteristics_avail ||
1343                     !css_general_characteristics.ext_mb) {
1344                         format = CMF_BASIC;
1345                 } else {
1346                         format = CMF_EXTENDED;
1347                 }
1348                 detect_string = "autodetected";
1349         } else {
1350                 detect_string = "parameter";
1351         }
1352
1353         switch (format) {
1354         case CMF_BASIC:
1355                 format_string = "basic";
1356                 cmbops = &cmbops_basic;
1357                 break;
1358         case CMF_EXTENDED:
1359                 format_string = "extended";
1360                 cmbops = &cmbops_extended;
1361                 break;
1362         default:
1363                 printk(KERN_ERR "cio: Invalid format %d for channel "
1364                         "measurement facility\n", format);
1365                 return 1;
1366         }
1367
1368         printk(KERN_INFO "cio: Channel measurement facility using %s "
1369                "format (%s)\n", format_string, detect_string);
1370         return 0;
1371 }
1372
1373 module_init(init_cmf);
1374
1375
1376 MODULE_AUTHOR("Arnd Bergmann <arndb@de.ibm.com>");
1377 MODULE_LICENSE("GPL");
1378 MODULE_DESCRIPTION("channel measurement facility base driver\n"
1379                    "Copyright 2003 IBM Corporation\n");
1380
1381 EXPORT_SYMBOL_GPL(enable_cmf);
1382 EXPORT_SYMBOL_GPL(disable_cmf);
1383 EXPORT_SYMBOL_GPL(cmf_read);
1384 EXPORT_SYMBOL_GPL(cmf_readall);