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