xref: /linux/drivers/s390/cio/cmf.c (revision 7fc2cd2e4b398c57c9cf961cfea05eadbf34c05c)
1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  * Linux on zSeries Channel Measurement Facility support
4  *
5  * Copyright IBM Corp. 2000, 2006
6  *
7  * Authors: Arnd Bergmann <arndb@de.ibm.com>
8  *	    Cornelia Huck <cornelia.huck@de.ibm.com>
9  *
10  * original idea from Natarajan Krishnaswami <nkrishna@us.ibm.com>
11  */
12 
13 #define pr_fmt(fmt) "cio: " fmt
14 
15 #include <linux/memblock.h>
16 #include <linux/device.h>
17 #include <linux/init.h>
18 #include <linux/list.h>
19 #include <linux/export.h>
20 #include <linux/moduleparam.h>
21 #include <linux/slab.h>
22 #include <linux/timex.h>	/* get_tod_clock() */
23 
24 #include <asm/ccwdev.h>
25 #include <asm/cio.h>
26 #include <asm/cmb.h>
27 #include <asm/div64.h>
28 
29 #include "cio.h"
30 #include "css.h"
31 #include "device.h"
32 #include "ioasm.h"
33 #include "chsc.h"
34 
35 /*
36  * parameter to enable cmf during boot, possible uses are:
37  *  "s390cmf" -- enable cmf and allocate 2 MB of ram so measuring can be
38  *               used on any subchannel
39  *  "s390cmf=<num>" -- enable cmf and allocate enough memory to measure
40  *                     <num> subchannel, where <num> is an integer
41  *                     between 1 and 65535, default is 1024
42  */
43 #define ARGSTRING "s390cmf"
44 
45 /* indices for READCMB */
46 enum cmb_index {
47 	avg_utilization = -1,
48  /* basic and extended format: */
49 	cmb_ssch_rsch_count = 0,
50 	cmb_sample_count,
51 	cmb_device_connect_time,
52 	cmb_function_pending_time,
53 	cmb_device_disconnect_time,
54 	cmb_control_unit_queuing_time,
55 	cmb_device_active_only_time,
56  /* extended format only: */
57 	cmb_device_busy_time,
58 	cmb_initial_command_response_time,
59 };
60 
61 /**
62  * enum cmb_format - types of supported measurement block formats
63  *
64  * @CMF_BASIC:      traditional channel measurement blocks supported
65  *		    by all machines that we run on
66  * @CMF_EXTENDED:   improved format that was introduced with the z990
67  *		    machine
68  * @CMF_AUTODETECT: default: use extended format when running on a machine
69  *		    supporting extended format, otherwise fall back to
70  *		    basic format
71  */
72 enum cmb_format {
73 	CMF_BASIC,
74 	CMF_EXTENDED,
75 	CMF_AUTODETECT = -1,
76 };
77 
78 /*
79  * format - actual format for all measurement blocks
80  *
81  * The format module parameter can be set to a value of 0 (zero)
82  * or 1, indicating basic or extended format as described for
83  * enum cmb_format.
84  */
85 static int format = CMF_AUTODETECT;
86 module_param(format, bint, 0444);
87 
88 /**
89  * struct cmb_operations - functions to use depending on cmb_format
90  *
91  * Most of these functions operate on a struct ccw_device. There is only
92  * one instance of struct cmb_operations because the format of the measurement
93  * data is guaranteed to be the same for every ccw_device.
94  *
95  * @alloc:	allocate memory for a channel measurement block,
96  *		either with the help of a special pool or with kmalloc
97  * @free:	free memory allocated with @alloc
98  * @set:	enable or disable measurement
99  * @read:	read a measurement entry at an index
100  * @readall:	read a measurement block in a common format
101  * @reset:	clear the data in the associated measurement block and
102  *		reset its time stamp
103  */
104 struct cmb_operations {
105 	int  (*alloc)  (struct ccw_device *);
106 	void (*free)   (struct ccw_device *);
107 	int  (*set)    (struct ccw_device *, u32);
108 	u64  (*read)   (struct ccw_device *, int);
109 	int  (*readall)(struct ccw_device *, struct cmbdata *);
110 	void (*reset)  (struct ccw_device *);
111 /* private: */
112 	struct attribute_group *attr_group;
113 };
114 static struct cmb_operations *cmbops;
115 
116 struct cmb_data {
117 	void *hw_block;   /* Pointer to block updated by hardware */
118 	void *last_block; /* Last changed block copied from hardware block */
119 	int size;	  /* Size of hw_block and last_block */
120 	unsigned long long last_update;  /* when last_block was updated */
121 };
122 
123 /*
124  * Our user interface is designed in terms of nanoseconds,
125  * while the hardware measures total times in its own
126  * unit.
127  */
128 static inline u64 time_to_nsec(u32 value)
129 {
130 	return ((u64)value) * 128000ull;
131 }
132 
133 /*
134  * Users are usually interested in average times,
135  * not accumulated time.
136  * This also helps us with atomicity problems
137  * when reading single values.
138  */
139 static inline u64 time_to_avg_nsec(u32 value, u32 count)
140 {
141 	u64 ret;
142 
143 	/* no samples yet, avoid division by 0 */
144 	if (count == 0)
145 		return 0;
146 
147 	/* value comes in units of 128 µsec */
148 	ret = time_to_nsec(value);
149 	do_div(ret, count);
150 
151 	return ret;
152 }
153 
154 #define CMF_OFF 0
155 #define CMF_ON	2
156 
157 /*
158  * Activate or deactivate the channel monitor. When area is NULL,
159  * the monitor is deactivated. The channel monitor needs to
160  * be active in order to measure subchannels, which also need
161  * to be enabled.
162  */
163 static inline void cmf_activate(void *area, unsigned int onoff)
164 {
165 	/* activate channel measurement */
166 	asm volatile(
167 		"	lgr	1,%[r1]\n"
168 		"	lgr	2,%[mbo]\n"
169 		"	schm"
170 		:
171 		: [r1] "d" ((unsigned long)onoff),
172 		  [mbo] "d" (virt_to_phys(area))
173 		: "1", "2");
174 }
175 
176 static int set_schib(struct ccw_device *cdev, u32 mme, int mbfc,
177 		     unsigned long address)
178 {
179 	struct subchannel *sch = to_subchannel(cdev->dev.parent);
180 	int ret;
181 
182 	sch->config.mme = mme;
183 	sch->config.mbfc = mbfc;
184 	/* address can be either a block address or a block index */
185 	if (mbfc)
186 		sch->config.mba = address;
187 	else
188 		sch->config.mbi = address;
189 
190 	ret = cio_commit_config(sch);
191 	if (!mme && ret == -ENODEV) {
192 		/*
193 		 * The task was to disable measurement block updates but
194 		 * the subchannel is already gone. Report success.
195 		 */
196 		ret = 0;
197 	}
198 	return ret;
199 }
200 
201 struct set_schib_struct {
202 	u32 mme;
203 	int mbfc;
204 	unsigned long address;
205 	wait_queue_head_t wait;
206 	int ret;
207 };
208 
209 #define CMF_PENDING 1
210 #define SET_SCHIB_TIMEOUT (10 * HZ)
211 
212 static int set_schib_wait(struct ccw_device *cdev, u32 mme,
213 			  int mbfc, unsigned long address)
214 {
215 	struct set_schib_struct set_data;
216 	int ret = -ENODEV;
217 
218 	spin_lock_irq(cdev->ccwlock);
219 	if (!cdev->private->cmb)
220 		goto out;
221 
222 	ret = set_schib(cdev, mme, mbfc, address);
223 	if (ret != -EBUSY)
224 		goto out;
225 
226 	/* if the device is not online, don't even try again */
227 	if (cdev->private->state != DEV_STATE_ONLINE)
228 		goto out;
229 
230 	init_waitqueue_head(&set_data.wait);
231 	set_data.mme = mme;
232 	set_data.mbfc = mbfc;
233 	set_data.address = address;
234 	set_data.ret = CMF_PENDING;
235 
236 	cdev->private->state = DEV_STATE_CMFCHANGE;
237 	cdev->private->cmb_wait = &set_data;
238 	spin_unlock_irq(cdev->ccwlock);
239 
240 	ret = wait_event_interruptible_timeout(set_data.wait,
241 					       set_data.ret != CMF_PENDING,
242 					       SET_SCHIB_TIMEOUT);
243 	spin_lock_irq(cdev->ccwlock);
244 	if (ret <= 0) {
245 		if (set_data.ret == CMF_PENDING) {
246 			set_data.ret = (ret == 0) ? -ETIME : ret;
247 			if (cdev->private->state == DEV_STATE_CMFCHANGE)
248 				cdev->private->state = DEV_STATE_ONLINE;
249 		}
250 	}
251 	cdev->private->cmb_wait = NULL;
252 	ret = set_data.ret;
253 out:
254 	spin_unlock_irq(cdev->ccwlock);
255 	return ret;
256 }
257 
258 void retry_set_schib(struct ccw_device *cdev)
259 {
260 	struct set_schib_struct *set_data = cdev->private->cmb_wait;
261 
262 	if (!set_data)
263 		return;
264 
265 	set_data->ret = set_schib(cdev, set_data->mme, set_data->mbfc,
266 				  set_data->address);
267 	wake_up(&set_data->wait);
268 }
269 
270 static int cmf_copy_block(struct ccw_device *cdev)
271 {
272 	struct subchannel *sch = to_subchannel(cdev->dev.parent);
273 	struct cmb_data *cmb_data;
274 	void *hw_block;
275 
276 	if (cio_update_schib(sch))
277 		return -ENODEV;
278 
279 	if (scsw_fctl(&sch->schib.scsw) & SCSW_FCTL_START_FUNC) {
280 		/* Don't copy if a start function is in progress. */
281 		if ((!(scsw_actl(&sch->schib.scsw) & SCSW_ACTL_SUSPENDED)) &&
282 		    (scsw_actl(&sch->schib.scsw) &
283 		     (SCSW_ACTL_DEVACT | SCSW_ACTL_SCHACT)) &&
284 		    (!(scsw_stctl(&sch->schib.scsw) & SCSW_STCTL_SEC_STATUS)))
285 			return -EBUSY;
286 	}
287 	cmb_data = cdev->private->cmb;
288 	hw_block = cmb_data->hw_block;
289 	memcpy(cmb_data->last_block, hw_block, cmb_data->size);
290 	cmb_data->last_update = get_tod_clock();
291 	return 0;
292 }
293 
294 struct copy_block_struct {
295 	wait_queue_head_t wait;
296 	int ret;
297 };
298 
299 static int cmf_cmb_copy_wait(struct ccw_device *cdev)
300 {
301 	struct copy_block_struct copy_block;
302 	int ret = -ENODEV;
303 
304 	spin_lock_irq(cdev->ccwlock);
305 	if (!cdev->private->cmb)
306 		goto out;
307 
308 	ret = cmf_copy_block(cdev);
309 	if (ret != -EBUSY)
310 		goto out;
311 
312 	if (cdev->private->state != DEV_STATE_ONLINE)
313 		goto out;
314 
315 	init_waitqueue_head(&copy_block.wait);
316 	copy_block.ret = CMF_PENDING;
317 
318 	cdev->private->state = DEV_STATE_CMFUPDATE;
319 	cdev->private->cmb_wait = &copy_block;
320 	spin_unlock_irq(cdev->ccwlock);
321 
322 	ret = wait_event_interruptible(copy_block.wait,
323 				       copy_block.ret != CMF_PENDING);
324 	spin_lock_irq(cdev->ccwlock);
325 	if (ret) {
326 		if (copy_block.ret == CMF_PENDING) {
327 			copy_block.ret = -ERESTARTSYS;
328 			if (cdev->private->state == DEV_STATE_CMFUPDATE)
329 				cdev->private->state = DEV_STATE_ONLINE;
330 		}
331 	}
332 	cdev->private->cmb_wait = NULL;
333 	ret = copy_block.ret;
334 out:
335 	spin_unlock_irq(cdev->ccwlock);
336 	return ret;
337 }
338 
339 void cmf_retry_copy_block(struct ccw_device *cdev)
340 {
341 	struct copy_block_struct *copy_block = cdev->private->cmb_wait;
342 
343 	if (!copy_block)
344 		return;
345 
346 	copy_block->ret = cmf_copy_block(cdev);
347 	wake_up(&copy_block->wait);
348 }
349 
350 static void cmf_generic_reset(struct ccw_device *cdev)
351 {
352 	struct cmb_data *cmb_data;
353 
354 	spin_lock_irq(cdev->ccwlock);
355 	cmb_data = cdev->private->cmb;
356 	if (cmb_data) {
357 		memset(cmb_data->last_block, 0, cmb_data->size);
358 		/*
359 		 * Need to reset hw block as well to make the hardware start
360 		 * from 0 again.
361 		 */
362 		memset(cmb_data->hw_block, 0, cmb_data->size);
363 		cmb_data->last_update = 0;
364 	}
365 	cdev->private->cmb_start_time = get_tod_clock();
366 	spin_unlock_irq(cdev->ccwlock);
367 }
368 
369 /**
370  * struct cmb_area - container for global cmb data
371  *
372  * @mem:	pointer to CMBs (only in basic measurement mode)
373  * @list:	contains a linked list of all subchannels
374  * @num_channels: number of channels to be measured
375  * @lock:	protect concurrent access to @mem and @list
376  */
377 struct cmb_area {
378 	struct cmb *mem;
379 	struct list_head list;
380 	int num_channels;
381 	spinlock_t lock;
382 };
383 
384 static struct cmb_area cmb_area = {
385 	.lock = __SPIN_LOCK_UNLOCKED(cmb_area.lock),
386 	.list = LIST_HEAD_INIT(cmb_area.list),
387 	.num_channels  = 1024,
388 };
389 
390 /* ****** old style CMB handling ********/
391 
392 /*
393  * Basic channel measurement blocks are allocated in one contiguous
394  * block of memory, which can not be moved as long as any channel
395  * is active. Therefore, a maximum number of subchannels needs to
396  * be defined somewhere. This is a module parameter, defaulting to
397  * a reasonable value of 1024, or 32 kb of memory.
398  * Current kernels don't allow kmalloc with more than 128kb, so the
399  * maximum is 4096.
400  */
401 
402 module_param_named(maxchannels, cmb_area.num_channels, uint, 0444);
403 
404 /**
405  * struct cmb - basic channel measurement block
406  * @ssch_rsch_count: number of ssch and rsch
407  * @sample_count: number of samples
408  * @device_connect_time: time of device connect
409  * @function_pending_time: time of function pending
410  * @device_disconnect_time: time of device disconnect
411  * @control_unit_queuing_time: time of control unit queuing
412  * @device_active_only_time: time of device active only
413  * @reserved: unused in basic measurement mode
414  *
415  * The measurement block as used by the hardware. The fields are described
416  * further in z/Architecture Principles of Operation, chapter 17.
417  *
418  * The cmb area made up from these blocks must be a contiguous array and may
419  * not be reallocated or freed.
420  * Only one cmb area can be present in the system.
421  */
422 struct cmb {
423 	u16 ssch_rsch_count;
424 	u16 sample_count;
425 	u32 device_connect_time;
426 	u32 function_pending_time;
427 	u32 device_disconnect_time;
428 	u32 control_unit_queuing_time;
429 	u32 device_active_only_time;
430 	u32 reserved[2];
431 };
432 
433 /*
434  * Insert a single device into the cmb_area list.
435  * Called with cmb_area.lock held from alloc_cmb.
436  */
437 static int alloc_cmb_single(struct ccw_device *cdev,
438 			    struct cmb_data *cmb_data)
439 {
440 	struct cmb *cmb;
441 	struct ccw_device_private *node;
442 	int ret;
443 
444 	spin_lock_irq(cdev->ccwlock);
445 	if (!list_empty(&cdev->private->cmb_list)) {
446 		ret = -EBUSY;
447 		goto out;
448 	}
449 
450 	/*
451 	 * Find first unused cmb in cmb_area.mem.
452 	 * This is a little tricky: cmb_area.list
453 	 * remains sorted by ->cmb->hw_data pointers.
454 	 */
455 	cmb = cmb_area.mem;
456 	list_for_each_entry(node, &cmb_area.list, cmb_list) {
457 		struct cmb_data *data;
458 		data = node->cmb;
459 		if ((struct cmb*)data->hw_block > cmb)
460 			break;
461 		cmb++;
462 	}
463 	if (cmb - cmb_area.mem >= cmb_area.num_channels) {
464 		ret = -ENOMEM;
465 		goto out;
466 	}
467 
468 	/* insert new cmb */
469 	list_add_tail(&cdev->private->cmb_list, &node->cmb_list);
470 	cmb_data->hw_block = cmb;
471 	cdev->private->cmb = cmb_data;
472 	ret = 0;
473 out:
474 	spin_unlock_irq(cdev->ccwlock);
475 	return ret;
476 }
477 
478 static int alloc_cmb(struct ccw_device *cdev)
479 {
480 	int ret;
481 	struct cmb *mem;
482 	ssize_t size;
483 	struct cmb_data *cmb_data;
484 
485 	/* Allocate private cmb_data. */
486 	cmb_data = kzalloc(sizeof(struct cmb_data), GFP_KERNEL);
487 	if (!cmb_data)
488 		return -ENOMEM;
489 
490 	cmb_data->last_block = kzalloc(sizeof(struct cmb), GFP_KERNEL);
491 	if (!cmb_data->last_block) {
492 		kfree(cmb_data);
493 		return -ENOMEM;
494 	}
495 	cmb_data->size = sizeof(struct cmb);
496 	spin_lock(&cmb_area.lock);
497 
498 	if (!cmb_area.mem) {
499 		/* there is no user yet, so we need a new area */
500 		size = sizeof(struct cmb) * cmb_area.num_channels;
501 		WARN_ON(!list_empty(&cmb_area.list));
502 
503 		spin_unlock(&cmb_area.lock);
504 		mem = (void *)__get_free_pages(GFP_KERNEL, get_order(size));
505 		spin_lock(&cmb_area.lock);
506 
507 		if (cmb_area.mem) {
508 			/* ok, another thread was faster */
509 			free_pages((unsigned long)mem, get_order(size));
510 		} else if (!mem) {
511 			/* no luck */
512 			ret = -ENOMEM;
513 			goto out;
514 		} else {
515 			/* everything ok */
516 			memset(mem, 0, size);
517 			cmb_area.mem = mem;
518 			cmf_activate(cmb_area.mem, CMF_ON);
519 		}
520 	}
521 
522 	/* do the actual allocation */
523 	ret = alloc_cmb_single(cdev, cmb_data);
524 out:
525 	spin_unlock(&cmb_area.lock);
526 	if (ret) {
527 		kfree(cmb_data->last_block);
528 		kfree(cmb_data);
529 	}
530 	return ret;
531 }
532 
533 static void free_cmb(struct ccw_device *cdev)
534 {
535 	struct ccw_device_private *priv;
536 	struct cmb_data *cmb_data;
537 
538 	spin_lock(&cmb_area.lock);
539 	spin_lock_irq(cdev->ccwlock);
540 
541 	priv = cdev->private;
542 	cmb_data = priv->cmb;
543 	priv->cmb = NULL;
544 	if (cmb_data)
545 		kfree(cmb_data->last_block);
546 	kfree(cmb_data);
547 	list_del_init(&priv->cmb_list);
548 
549 	if (list_empty(&cmb_area.list)) {
550 		ssize_t size;
551 		size = sizeof(struct cmb) * cmb_area.num_channels;
552 		cmf_activate(NULL, CMF_OFF);
553 		free_pages((unsigned long)cmb_area.mem, get_order(size));
554 		cmb_area.mem = NULL;
555 	}
556 	spin_unlock_irq(cdev->ccwlock);
557 	spin_unlock(&cmb_area.lock);
558 }
559 
560 static int set_cmb(struct ccw_device *cdev, u32 mme)
561 {
562 	u16 offset;
563 	struct cmb_data *cmb_data;
564 	unsigned long flags;
565 
566 	spin_lock_irqsave(cdev->ccwlock, flags);
567 	if (!cdev->private->cmb) {
568 		spin_unlock_irqrestore(cdev->ccwlock, flags);
569 		return -EINVAL;
570 	}
571 	cmb_data = cdev->private->cmb;
572 	offset = mme ? (struct cmb *)cmb_data->hw_block - cmb_area.mem : 0;
573 	spin_unlock_irqrestore(cdev->ccwlock, flags);
574 
575 	return set_schib_wait(cdev, mme, 0, offset);
576 }
577 
578 /* calculate utilization in 0.1 percent units */
579 static u64 __cmb_utilization(u64 device_connect_time, u64 function_pending_time,
580 			     u64 device_disconnect_time, u64 start_time)
581 {
582 	u64 utilization, elapsed_time;
583 
584 	utilization = time_to_nsec(device_connect_time +
585 				   function_pending_time +
586 				   device_disconnect_time);
587 
588 	elapsed_time = get_tod_clock() - start_time;
589 	elapsed_time = tod_to_ns(elapsed_time);
590 	elapsed_time /= 1000;
591 
592 	return elapsed_time ? (utilization / elapsed_time) : 0;
593 }
594 
595 static u64 read_cmb(struct ccw_device *cdev, int index)
596 {
597 	struct cmb_data *cmb_data;
598 	unsigned long flags;
599 	struct cmb *cmb;
600 	u64 ret = 0;
601 	u32 val;
602 
603 	spin_lock_irqsave(cdev->ccwlock, flags);
604 	cmb_data = cdev->private->cmb;
605 	if (!cmb_data)
606 		goto out;
607 
608 	cmb = cmb_data->hw_block;
609 	switch (index) {
610 	case avg_utilization:
611 		ret = __cmb_utilization(cmb->device_connect_time,
612 					cmb->function_pending_time,
613 					cmb->device_disconnect_time,
614 					cdev->private->cmb_start_time);
615 		goto out;
616 	case cmb_ssch_rsch_count:
617 		ret = cmb->ssch_rsch_count;
618 		goto out;
619 	case cmb_sample_count:
620 		ret = cmb->sample_count;
621 		goto out;
622 	case cmb_device_connect_time:
623 		val = cmb->device_connect_time;
624 		break;
625 	case cmb_function_pending_time:
626 		val = cmb->function_pending_time;
627 		break;
628 	case cmb_device_disconnect_time:
629 		val = cmb->device_disconnect_time;
630 		break;
631 	case cmb_control_unit_queuing_time:
632 		val = cmb->control_unit_queuing_time;
633 		break;
634 	case cmb_device_active_only_time:
635 		val = cmb->device_active_only_time;
636 		break;
637 	default:
638 		goto out;
639 	}
640 	ret = time_to_avg_nsec(val, cmb->sample_count);
641 out:
642 	spin_unlock_irqrestore(cdev->ccwlock, flags);
643 	return ret;
644 }
645 
646 static int readall_cmb(struct ccw_device *cdev, struct cmbdata *data)
647 {
648 	struct cmb *cmb;
649 	struct cmb_data *cmb_data;
650 	u64 time;
651 	unsigned long flags;
652 	int ret;
653 
654 	ret = cmf_cmb_copy_wait(cdev);
655 	if (ret < 0)
656 		return ret;
657 	spin_lock_irqsave(cdev->ccwlock, flags);
658 	cmb_data = cdev->private->cmb;
659 	if (!cmb_data) {
660 		ret = -ENODEV;
661 		goto out;
662 	}
663 	if (cmb_data->last_update == 0) {
664 		ret = -EAGAIN;
665 		goto out;
666 	}
667 	cmb = cmb_data->last_block;
668 	time = cmb_data->last_update - cdev->private->cmb_start_time;
669 
670 	memset(data, 0, sizeof(struct cmbdata));
671 
672 	/* we only know values before device_busy_time */
673 	data->size = offsetof(struct cmbdata, device_busy_time);
674 
675 	data->elapsed_time = tod_to_ns(time);
676 
677 	/* copy data to new structure */
678 	data->ssch_rsch_count = cmb->ssch_rsch_count;
679 	data->sample_count = cmb->sample_count;
680 
681 	/* time fields are converted to nanoseconds while copying */
682 	data->device_connect_time = time_to_nsec(cmb->device_connect_time);
683 	data->function_pending_time = time_to_nsec(cmb->function_pending_time);
684 	data->device_disconnect_time =
685 		time_to_nsec(cmb->device_disconnect_time);
686 	data->control_unit_queuing_time
687 		= time_to_nsec(cmb->control_unit_queuing_time);
688 	data->device_active_only_time
689 		= time_to_nsec(cmb->device_active_only_time);
690 	ret = 0;
691 out:
692 	spin_unlock_irqrestore(cdev->ccwlock, flags);
693 	return ret;
694 }
695 
696 static void reset_cmb(struct ccw_device *cdev)
697 {
698 	cmf_generic_reset(cdev);
699 }
700 
701 static int cmf_enabled(struct ccw_device *cdev)
702 {
703 	int enabled;
704 
705 	spin_lock_irq(cdev->ccwlock);
706 	enabled = !!cdev->private->cmb;
707 	spin_unlock_irq(cdev->ccwlock);
708 
709 	return enabled;
710 }
711 
712 static struct attribute_group cmf_attr_group;
713 
714 static struct cmb_operations cmbops_basic = {
715 	.alloc	= alloc_cmb,
716 	.free	= free_cmb,
717 	.set	= set_cmb,
718 	.read	= read_cmb,
719 	.readall    = readall_cmb,
720 	.reset	    = reset_cmb,
721 	.attr_group = &cmf_attr_group,
722 };
723 
724 /* ******** extended cmb handling ********/
725 
726 /**
727  * struct cmbe - extended channel measurement block
728  * @ssch_rsch_count: number of ssch and rsch
729  * @sample_count: number of samples
730  * @device_connect_time: time of device connect
731  * @function_pending_time: time of function pending
732  * @device_disconnect_time: time of device disconnect
733  * @control_unit_queuing_time: time of control unit queuing
734  * @device_active_only_time: time of device active only
735  * @device_busy_time: time of device busy
736  * @initial_command_response_time: initial command response time
737  * @reserved: unused
738  *
739  * The measurement block as used by the hardware. May be in any 64 bit physical
740  * location.
741  * The fields are described further in z/Architecture Principles of Operation,
742  * third edition, chapter 17.
743  */
744 struct cmbe {
745 	u32 ssch_rsch_count;
746 	u32 sample_count;
747 	u32 device_connect_time;
748 	u32 function_pending_time;
749 	u32 device_disconnect_time;
750 	u32 control_unit_queuing_time;
751 	u32 device_active_only_time;
752 	u32 device_busy_time;
753 	u32 initial_command_response_time;
754 	u32 reserved[7];
755 } __packed __aligned(64);
756 
757 static struct kmem_cache *cmbe_cache;
758 
759 static int alloc_cmbe(struct ccw_device *cdev)
760 {
761 	struct cmb_data *cmb_data;
762 	struct cmbe *cmbe;
763 	int ret = -ENOMEM;
764 
765 	cmbe = kmem_cache_zalloc(cmbe_cache, GFP_KERNEL);
766 	if (!cmbe)
767 		return ret;
768 
769 	cmb_data = kzalloc(sizeof(*cmb_data), GFP_KERNEL);
770 	if (!cmb_data)
771 		goto out_free;
772 
773 	cmb_data->last_block = kzalloc(sizeof(struct cmbe), GFP_KERNEL);
774 	if (!cmb_data->last_block)
775 		goto out_free;
776 
777 	cmb_data->size = sizeof(*cmbe);
778 	cmb_data->hw_block = cmbe;
779 
780 	spin_lock(&cmb_area.lock);
781 	spin_lock_irq(cdev->ccwlock);
782 	if (cdev->private->cmb)
783 		goto out_unlock;
784 
785 	cdev->private->cmb = cmb_data;
786 
787 	/* activate global measurement if this is the first channel */
788 	if (list_empty(&cmb_area.list))
789 		cmf_activate(NULL, CMF_ON);
790 	list_add_tail(&cdev->private->cmb_list, &cmb_area.list);
791 
792 	spin_unlock_irq(cdev->ccwlock);
793 	spin_unlock(&cmb_area.lock);
794 	return 0;
795 
796 out_unlock:
797 	spin_unlock_irq(cdev->ccwlock);
798 	spin_unlock(&cmb_area.lock);
799 	ret = -EBUSY;
800 out_free:
801 	if (cmb_data)
802 		kfree(cmb_data->last_block);
803 	kfree(cmb_data);
804 	kmem_cache_free(cmbe_cache, cmbe);
805 
806 	return ret;
807 }
808 
809 static void free_cmbe(struct ccw_device *cdev)
810 {
811 	struct cmb_data *cmb_data;
812 
813 	spin_lock(&cmb_area.lock);
814 	spin_lock_irq(cdev->ccwlock);
815 	cmb_data = cdev->private->cmb;
816 	cdev->private->cmb = NULL;
817 	if (cmb_data) {
818 		kfree(cmb_data->last_block);
819 		kmem_cache_free(cmbe_cache, cmb_data->hw_block);
820 	}
821 	kfree(cmb_data);
822 
823 	/* deactivate global measurement if this is the last channel */
824 	list_del_init(&cdev->private->cmb_list);
825 	if (list_empty(&cmb_area.list))
826 		cmf_activate(NULL, CMF_OFF);
827 	spin_unlock_irq(cdev->ccwlock);
828 	spin_unlock(&cmb_area.lock);
829 }
830 
831 static int set_cmbe(struct ccw_device *cdev, u32 mme)
832 {
833 	unsigned long mba;
834 	struct cmb_data *cmb_data;
835 	unsigned long flags;
836 
837 	spin_lock_irqsave(cdev->ccwlock, flags);
838 	if (!cdev->private->cmb) {
839 		spin_unlock_irqrestore(cdev->ccwlock, flags);
840 		return -EINVAL;
841 	}
842 	cmb_data = cdev->private->cmb;
843 	mba = mme ? (unsigned long) cmb_data->hw_block : 0;
844 	spin_unlock_irqrestore(cdev->ccwlock, flags);
845 
846 	return set_schib_wait(cdev, mme, 1, mba);
847 }
848 
849 static u64 read_cmbe(struct ccw_device *cdev, int index)
850 {
851 	struct cmb_data *cmb_data;
852 	unsigned long flags;
853 	struct cmbe *cmb;
854 	u64 ret = 0;
855 	u32 val;
856 
857 	spin_lock_irqsave(cdev->ccwlock, flags);
858 	cmb_data = cdev->private->cmb;
859 	if (!cmb_data)
860 		goto out;
861 
862 	cmb = cmb_data->hw_block;
863 	switch (index) {
864 	case avg_utilization:
865 		ret = __cmb_utilization(cmb->device_connect_time,
866 					cmb->function_pending_time,
867 					cmb->device_disconnect_time,
868 					cdev->private->cmb_start_time);
869 		goto out;
870 	case cmb_ssch_rsch_count:
871 		ret = cmb->ssch_rsch_count;
872 		goto out;
873 	case cmb_sample_count:
874 		ret = cmb->sample_count;
875 		goto out;
876 	case cmb_device_connect_time:
877 		val = cmb->device_connect_time;
878 		break;
879 	case cmb_function_pending_time:
880 		val = cmb->function_pending_time;
881 		break;
882 	case cmb_device_disconnect_time:
883 		val = cmb->device_disconnect_time;
884 		break;
885 	case cmb_control_unit_queuing_time:
886 		val = cmb->control_unit_queuing_time;
887 		break;
888 	case cmb_device_active_only_time:
889 		val = cmb->device_active_only_time;
890 		break;
891 	case cmb_device_busy_time:
892 		val = cmb->device_busy_time;
893 		break;
894 	case cmb_initial_command_response_time:
895 		val = cmb->initial_command_response_time;
896 		break;
897 	default:
898 		goto out;
899 	}
900 	ret = time_to_avg_nsec(val, cmb->sample_count);
901 out:
902 	spin_unlock_irqrestore(cdev->ccwlock, flags);
903 	return ret;
904 }
905 
906 static int readall_cmbe(struct ccw_device *cdev, struct cmbdata *data)
907 {
908 	struct cmbe *cmb;
909 	struct cmb_data *cmb_data;
910 	u64 time;
911 	unsigned long flags;
912 	int ret;
913 
914 	ret = cmf_cmb_copy_wait(cdev);
915 	if (ret < 0)
916 		return ret;
917 	spin_lock_irqsave(cdev->ccwlock, flags);
918 	cmb_data = cdev->private->cmb;
919 	if (!cmb_data) {
920 		ret = -ENODEV;
921 		goto out;
922 	}
923 	if (cmb_data->last_update == 0) {
924 		ret = -EAGAIN;
925 		goto out;
926 	}
927 	time = cmb_data->last_update - cdev->private->cmb_start_time;
928 
929 	memset (data, 0, sizeof(struct cmbdata));
930 
931 	/* we only know values before device_busy_time */
932 	data->size = offsetof(struct cmbdata, device_busy_time);
933 
934 	data->elapsed_time = tod_to_ns(time);
935 
936 	cmb = cmb_data->last_block;
937 	/* copy data to new structure */
938 	data->ssch_rsch_count = cmb->ssch_rsch_count;
939 	data->sample_count = cmb->sample_count;
940 
941 	/* time fields are converted to nanoseconds while copying */
942 	data->device_connect_time = time_to_nsec(cmb->device_connect_time);
943 	data->function_pending_time = time_to_nsec(cmb->function_pending_time);
944 	data->device_disconnect_time =
945 		time_to_nsec(cmb->device_disconnect_time);
946 	data->control_unit_queuing_time
947 		= time_to_nsec(cmb->control_unit_queuing_time);
948 	data->device_active_only_time
949 		= time_to_nsec(cmb->device_active_only_time);
950 	data->device_busy_time = time_to_nsec(cmb->device_busy_time);
951 	data->initial_command_response_time
952 		= time_to_nsec(cmb->initial_command_response_time);
953 
954 	ret = 0;
955 out:
956 	spin_unlock_irqrestore(cdev->ccwlock, flags);
957 	return ret;
958 }
959 
960 static void reset_cmbe(struct ccw_device *cdev)
961 {
962 	cmf_generic_reset(cdev);
963 }
964 
965 static struct attribute_group cmf_attr_group_ext;
966 
967 static struct cmb_operations cmbops_extended = {
968 	.alloc	    = alloc_cmbe,
969 	.free	    = free_cmbe,
970 	.set	    = set_cmbe,
971 	.read	    = read_cmbe,
972 	.readall    = readall_cmbe,
973 	.reset	    = reset_cmbe,
974 	.attr_group = &cmf_attr_group_ext,
975 };
976 
977 static ssize_t cmb_show_attr(struct device *dev, char *buf, enum cmb_index idx)
978 {
979 	return sysfs_emit(buf, "%lld\n", cmf_read(to_ccwdev(dev), idx));
980 }
981 
982 static ssize_t cmb_show_avg_sample_interval(struct device *dev,
983 					    struct device_attribute *attr,
984 					    char *buf)
985 {
986 	struct ccw_device *cdev = to_ccwdev(dev);
987 	unsigned long count;
988 	long interval;
989 
990 	count = cmf_read(cdev, cmb_sample_count);
991 	spin_lock_irq(cdev->ccwlock);
992 	if (count) {
993 		interval = get_tod_clock() - cdev->private->cmb_start_time;
994 		interval = tod_to_ns(interval);
995 		interval /= count;
996 	} else
997 		interval = -1;
998 	spin_unlock_irq(cdev->ccwlock);
999 	return sysfs_emit(buf, "%ld\n", interval);
1000 }
1001 
1002 static ssize_t cmb_show_avg_utilization(struct device *dev,
1003 					struct device_attribute *attr,
1004 					char *buf)
1005 {
1006 	unsigned long u = cmf_read(to_ccwdev(dev), avg_utilization);
1007 
1008 	return sysfs_emit(buf, "%02lu.%01lu%%\n", u / 10, u % 10);
1009 }
1010 
1011 #define cmf_attr(name) \
1012 static ssize_t show_##name(struct device *dev, \
1013 			   struct device_attribute *attr, char *buf)	\
1014 { return cmb_show_attr((dev), buf, cmb_##name); } \
1015 static DEVICE_ATTR(name, 0444, show_##name, NULL);
1016 
1017 #define cmf_attr_avg(name) \
1018 static ssize_t show_avg_##name(struct device *dev, \
1019 			       struct device_attribute *attr, char *buf) \
1020 { return cmb_show_attr((dev), buf, cmb_##name); } \
1021 static DEVICE_ATTR(avg_##name, 0444, show_avg_##name, NULL);
1022 
1023 cmf_attr(ssch_rsch_count);
1024 cmf_attr(sample_count);
1025 cmf_attr_avg(device_connect_time);
1026 cmf_attr_avg(function_pending_time);
1027 cmf_attr_avg(device_disconnect_time);
1028 cmf_attr_avg(control_unit_queuing_time);
1029 cmf_attr_avg(device_active_only_time);
1030 cmf_attr_avg(device_busy_time);
1031 cmf_attr_avg(initial_command_response_time);
1032 
1033 static DEVICE_ATTR(avg_sample_interval, 0444, cmb_show_avg_sample_interval,
1034 		   NULL);
1035 static DEVICE_ATTR(avg_utilization, 0444, cmb_show_avg_utilization, NULL);
1036 
1037 static struct attribute *cmf_attributes[] = {
1038 	&dev_attr_avg_sample_interval.attr,
1039 	&dev_attr_avg_utilization.attr,
1040 	&dev_attr_ssch_rsch_count.attr,
1041 	&dev_attr_sample_count.attr,
1042 	&dev_attr_avg_device_connect_time.attr,
1043 	&dev_attr_avg_function_pending_time.attr,
1044 	&dev_attr_avg_device_disconnect_time.attr,
1045 	&dev_attr_avg_control_unit_queuing_time.attr,
1046 	&dev_attr_avg_device_active_only_time.attr,
1047 	NULL,
1048 };
1049 
1050 static struct attribute_group cmf_attr_group = {
1051 	.name  = "cmf",
1052 	.attrs = cmf_attributes,
1053 };
1054 
1055 static struct attribute *cmf_attributes_ext[] = {
1056 	&dev_attr_avg_sample_interval.attr,
1057 	&dev_attr_avg_utilization.attr,
1058 	&dev_attr_ssch_rsch_count.attr,
1059 	&dev_attr_sample_count.attr,
1060 	&dev_attr_avg_device_connect_time.attr,
1061 	&dev_attr_avg_function_pending_time.attr,
1062 	&dev_attr_avg_device_disconnect_time.attr,
1063 	&dev_attr_avg_control_unit_queuing_time.attr,
1064 	&dev_attr_avg_device_active_only_time.attr,
1065 	&dev_attr_avg_device_busy_time.attr,
1066 	&dev_attr_avg_initial_command_response_time.attr,
1067 	NULL,
1068 };
1069 
1070 static struct attribute_group cmf_attr_group_ext = {
1071 	.name  = "cmf",
1072 	.attrs = cmf_attributes_ext,
1073 };
1074 
1075 static ssize_t cmb_enable_show(struct device *dev,
1076 			       struct device_attribute *attr,
1077 			       char *buf)
1078 {
1079 	struct ccw_device *cdev = to_ccwdev(dev);
1080 
1081 	return sysfs_emit(buf, "%d\n", cmf_enabled(cdev));
1082 }
1083 
1084 static ssize_t cmb_enable_store(struct device *dev,
1085 				struct device_attribute *attr, const char *buf,
1086 				size_t c)
1087 {
1088 	struct ccw_device *cdev = to_ccwdev(dev);
1089 	unsigned long val;
1090 	int ret;
1091 
1092 	ret = kstrtoul(buf, 16, &val);
1093 	if (ret)
1094 		return ret;
1095 
1096 	switch (val) {
1097 	case 0:
1098 		ret = disable_cmf(cdev);
1099 		break;
1100 	case 1:
1101 		ret = enable_cmf(cdev);
1102 		break;
1103 	default:
1104 		ret = -EINVAL;
1105 	}
1106 
1107 	return ret ? ret : c;
1108 }
1109 DEVICE_ATTR_RW(cmb_enable);
1110 
1111 /**
1112  * enable_cmf() - switch on the channel measurement for a specific device
1113  *  @cdev:	The ccw device to be enabled
1114  *
1115  *  Enable channel measurements for @cdev. If this is called on a device
1116  *  for which channel measurement is already enabled a reset of the
1117  *  measurement data is triggered.
1118  *  Returns: %0 for success or a negative error value.
1119  *  Context:
1120  *    non-atomic
1121  */
1122 int enable_cmf(struct ccw_device *cdev)
1123 {
1124 	int ret = 0;
1125 
1126 	device_lock(&cdev->dev);
1127 	if (cmf_enabled(cdev)) {
1128 		cmbops->reset(cdev);
1129 		goto out_unlock;
1130 	}
1131 	get_device(&cdev->dev);
1132 	ret = cmbops->alloc(cdev);
1133 	if (ret)
1134 		goto out;
1135 	cmbops->reset(cdev);
1136 	ret = sysfs_create_group(&cdev->dev.kobj, cmbops->attr_group);
1137 	if (ret) {
1138 		cmbops->free(cdev);
1139 		goto out;
1140 	}
1141 	ret = cmbops->set(cdev, 2);
1142 	if (ret) {
1143 		sysfs_remove_group(&cdev->dev.kobj, cmbops->attr_group);
1144 		cmbops->free(cdev);
1145 	}
1146 out:
1147 	if (ret)
1148 		put_device(&cdev->dev);
1149 out_unlock:
1150 	device_unlock(&cdev->dev);
1151 	return ret;
1152 }
1153 
1154 /**
1155  * __disable_cmf() - switch off the channel measurement for a specific device
1156  *  @cdev:	The ccw device to be disabled
1157  *
1158  *  Returns: %0 for success or a negative error value.
1159  *
1160  *  Context:
1161  *    non-atomic, device_lock() held.
1162  */
1163 int __disable_cmf(struct ccw_device *cdev)
1164 {
1165 	int ret;
1166 
1167 	ret = cmbops->set(cdev, 0);
1168 	if (ret)
1169 		return ret;
1170 
1171 	sysfs_remove_group(&cdev->dev.kobj, cmbops->attr_group);
1172 	cmbops->free(cdev);
1173 	put_device(&cdev->dev);
1174 
1175 	return ret;
1176 }
1177 
1178 /**
1179  * disable_cmf() - switch off the channel measurement for a specific device
1180  *  @cdev:	The ccw device to be disabled
1181  *
1182  *  Returns: %0 for success or a negative error value.
1183  *
1184  *  Context:
1185  *    non-atomic
1186  */
1187 int disable_cmf(struct ccw_device *cdev)
1188 {
1189 	int ret;
1190 
1191 	device_lock(&cdev->dev);
1192 	ret = __disable_cmf(cdev);
1193 	device_unlock(&cdev->dev);
1194 
1195 	return ret;
1196 }
1197 
1198 /**
1199  * cmf_read() - read one value from the current channel measurement block
1200  * @cdev:	the channel to be read
1201  * @index:	the index of the value to be read
1202  *
1203  * Returns: The value read or %0 if the value cannot be read.
1204  *
1205  *  Context:
1206  *    any
1207  */
1208 u64 cmf_read(struct ccw_device *cdev, int index)
1209 {
1210 	return cmbops->read(cdev, index);
1211 }
1212 
1213 /**
1214  * cmf_readall() - read the current channel measurement block
1215  * @cdev:	the channel to be read
1216  * @data:	a pointer to a data block that will be filled
1217  *
1218  * Returns: %0 on success, a negative error value otherwise.
1219  *
1220  *  Context:
1221  *    any
1222  */
1223 int cmf_readall(struct ccw_device *cdev, struct cmbdata *data)
1224 {
1225 	return cmbops->readall(cdev, data);
1226 }
1227 
1228 /* Re-enable cmf when a disconnected device becomes available again. */
1229 int cmf_reenable(struct ccw_device *cdev)
1230 {
1231 	cmbops->reset(cdev);
1232 	return cmbops->set(cdev, 2);
1233 }
1234 
1235 /**
1236  * cmf_reactivate() - reactivate measurement block updates
1237  *
1238  * Use this during resume from hibernate.
1239  */
1240 void cmf_reactivate(void)
1241 {
1242 	spin_lock(&cmb_area.lock);
1243 	if (!list_empty(&cmb_area.list))
1244 		cmf_activate(cmb_area.mem, CMF_ON);
1245 	spin_unlock(&cmb_area.lock);
1246 }
1247 
1248 static int __init init_cmbe(void)
1249 {
1250 	cmbe_cache = kmem_cache_create("cmbe_cache", sizeof(struct cmbe),
1251 				       __alignof__(struct cmbe), 0, NULL);
1252 
1253 	return cmbe_cache ? 0 : -ENOMEM;
1254 }
1255 
1256 static int __init init_cmf(void)
1257 {
1258 	char *format_string;
1259 	char *detect_string;
1260 	int ret;
1261 
1262 	/*
1263 	 * If the user did not give a parameter, see if we are running on a
1264 	 * machine supporting extended measurement blocks, otherwise fall back
1265 	 * to basic mode.
1266 	 */
1267 	if (format == CMF_AUTODETECT) {
1268 		if (!css_general_characteristics.ext_mb) {
1269 			format = CMF_BASIC;
1270 		} else {
1271 			format = CMF_EXTENDED;
1272 		}
1273 		detect_string = "autodetected";
1274 	} else {
1275 		detect_string = "parameter";
1276 	}
1277 
1278 	switch (format) {
1279 	case CMF_BASIC:
1280 		format_string = "basic";
1281 		cmbops = &cmbops_basic;
1282 		break;
1283 	case CMF_EXTENDED:
1284 		format_string = "extended";
1285 		cmbops = &cmbops_extended;
1286 
1287 		ret = init_cmbe();
1288 		if (ret)
1289 			return ret;
1290 		break;
1291 	default:
1292 		return -EINVAL;
1293 	}
1294 	pr_info("Channel measurement facility initialized using format "
1295 		"%s (mode %s)\n", format_string, detect_string);
1296 	return 0;
1297 }
1298 device_initcall(init_cmf);
1299 
1300 EXPORT_SYMBOL_GPL(enable_cmf);
1301 EXPORT_SYMBOL_GPL(disable_cmf);
1302 EXPORT_SYMBOL_GPL(cmf_read);
1303 EXPORT_SYMBOL_GPL(cmf_readall);
1304