[go: nahoru, domu]

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