xref: /linux/drivers/s390/cio/cmf.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
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 KMSG_COMPONENT "cio"
14 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
15 
16 #include <linux/memblock.h>
17 #include <linux/device.h>
18 #include <linux/init.h>
19 #include <linux/list.h>
20 #include <linux/export.h>
21 #include <linux/moduleparam.h>
22 #include <linux/slab.h>
23 #include <linux/timex.h>	/* get_tod_clock() */
24 
25 #include <asm/ccwdev.h>
26 #include <asm/cio.h>
27 #include <asm/cmb.h>
28 #include <asm/div64.h>
29 
30 #include "cio.h"
31 #include "css.h"
32 #include "device.h"
33 #include "ioasm.h"
34 #include "chsc.h"
35 
36 /*
37  * parameter to enable cmf during boot, possible uses are:
38  *  "s390cmf" -- enable cmf and allocate 2 MB of ram so measuring can be
39  *               used on any subchannel
40  *  "s390cmf=<num>" -- enable cmf and allocate enough memory to measure
41  *                     <num> subchannel, where <num> is an integer
42  *                     between 1 and 65535, default is 1024
43  */
44 #define ARGSTRING "s390cmf"
45 
46 /* indices for READCMB */
47 enum cmb_index {
48 	avg_utilization = -1,
49  /* basic and exended format: */
50 	cmb_ssch_rsch_count = 0,
51 	cmb_sample_count,
52 	cmb_device_connect_time,
53 	cmb_function_pending_time,
54 	cmb_device_disconnect_time,
55 	cmb_control_unit_queuing_time,
56 	cmb_device_active_only_time,
57  /* extended format only: */
58 	cmb_device_busy_time,
59 	cmb_initial_command_response_time,
60 };
61 
62 /**
63  * enum cmb_format - types of supported measurement block formats
64  *
65  * @CMF_BASIC:      traditional channel measurement blocks supported
66  *		    by all machines that we run on
67  * @CMF_EXTENDED:   improved format that was introduced with the z990
68  *		    machine
69  * @CMF_AUTODETECT: default: use extended format when running on a machine
70  *		    supporting extended format, otherwise fall back to
71  *		    basic format
72  */
73 enum cmb_format {
74 	CMF_BASIC,
75 	CMF_EXTENDED,
76 	CMF_AUTODETECT = -1,
77 };
78 
79 /*
80  * format - actual format for all measurement blocks
81  *
82  * The format module parameter can be set to a value of 0 (zero)
83  * or 1, indicating basic or extended format as described for
84  * enum cmb_format.
85  */
86 static int format = CMF_AUTODETECT;
87 module_param(format, bint, 0444);
88 
89 /**
90  * struct cmb_operations - functions to use depending on cmb_format
91  *
92  * Most of these functions operate on a struct ccw_device. There is only
93  * one instance of struct cmb_operations because the format of the measurement
94  * data is guaranteed to be the same for every ccw_device.
95  *
96  * @alloc:	allocate memory for a channel measurement block,
97  *		either with the help of a special pool or with kmalloc
98  * @free:	free memory allocated with @alloc
99  * @set:	enable or disable measurement
100  * @read:	read a measurement entry at an index
101  * @readall:	read a measurement block in a common format
102  * @reset:	clear the data in the associated measurement block and
103  *		reset its time stamp
104  */
105 struct cmb_operations {
106 	int  (*alloc)  (struct ccw_device *);
107 	void (*free)   (struct ccw_device *);
108 	int  (*set)    (struct ccw_device *, u32);
109 	u64  (*read)   (struct ccw_device *, int);
110 	int  (*readall)(struct ccw_device *, struct cmbdata *);
111 	void (*reset)  (struct ccw_device *);
112 /* private: */
113 	struct attribute_group *attr_group;
114 };
115 static struct cmb_operations *cmbops;
116 
117 struct cmb_data {
118 	void *hw_block;   /* Pointer to block updated by hardware */
119 	void *last_block; /* Last changed block copied from hardware block */
120 	int size;	  /* Size of hw_block and last_block */
121 	unsigned long long last_update;  /* when last_block was updated */
122 };
123 
124 /*
125  * Our user interface is designed in terms of nanoseconds,
126  * while the hardware measures total times in its own
127  * unit.
128  */
129 static inline u64 time_to_nsec(u32 value)
130 {
131 	return ((u64)value) * 128000ull;
132 }
133 
134 /*
135  * Users are usually interested in average times,
136  * not accumulated time.
137  * This also helps us with atomicity problems
138  * when reading sinlge values.
139  */
140 static inline u64 time_to_avg_nsec(u32 value, u32 count)
141 {
142 	u64 ret;
143 
144 	/* no samples yet, avoid division by 0 */
145 	if (count == 0)
146 		return 0;
147 
148 	/* value comes in units of 128 µsec */
149 	ret = time_to_nsec(value);
150 	do_div(ret, count);
151 
152 	return ret;
153 }
154 
155 #define CMF_OFF 0
156 #define CMF_ON	2
157 
158 /*
159  * Activate or deactivate the channel monitor. When area is NULL,
160  * the monitor is deactivated. The channel monitor needs to
161  * be active in order to measure subchannels, which also need
162  * to be enabled.
163  */
164 static inline void cmf_activate(void *area, unsigned int onoff)
165 {
166 	/* activate channel measurement */
167 	asm volatile(
168 		"	lgr	1,%[r1]\n"
169 		"	lgr	2,%[mbo]\n"
170 		"	schm\n"
171 		:
172 		: [r1] "d" ((unsigned long)onoff),
173 		  [mbo] "d" (virt_to_phys(area))
174 		: "1", "2");
175 }
176 
177 static int set_schib(struct ccw_device *cdev, u32 mme, int mbfc,
178 		     unsigned long address)
179 {
180 	struct subchannel *sch = to_subchannel(cdev->dev.parent);
181 	int ret;
182 
183 	sch->config.mme = mme;
184 	sch->config.mbfc = mbfc;
185 	/* address can be either a block address or a block index */
186 	if (mbfc)
187 		sch->config.mba = address;
188 	else
189 		sch->config.mbi = address;
190 
191 	ret = cio_commit_config(sch);
192 	if (!mme && ret == -ENODEV) {
193 		/*
194 		 * The task was to disable measurement block updates but
195 		 * the subchannel is already gone. Report success.
196 		 */
197 		ret = 0;
198 	}
199 	return ret;
200 }
201 
202 struct set_schib_struct {
203 	u32 mme;
204 	int mbfc;
205 	unsigned long address;
206 	wait_queue_head_t wait;
207 	int ret;
208 };
209 
210 #define CMF_PENDING 1
211 #define SET_SCHIB_TIMEOUT (10 * HZ)
212 
213 static int set_schib_wait(struct ccw_device *cdev, u32 mme,
214 			  int mbfc, unsigned long address)
215 {
216 	struct set_schib_struct set_data;
217 	int ret = -ENODEV;
218 
219 	spin_lock_irq(cdev->ccwlock);
220 	if (!cdev->private->cmb)
221 		goto out;
222 
223 	ret = set_schib(cdev, mme, mbfc, address);
224 	if (ret != -EBUSY)
225 		goto out;
226 
227 	/* if the device is not online, don't even try again */
228 	if (cdev->private->state != DEV_STATE_ONLINE)
229 		goto out;
230 
231 	init_waitqueue_head(&set_data.wait);
232 	set_data.mme = mme;
233 	set_data.mbfc = mbfc;
234 	set_data.address = address;
235 	set_data.ret = CMF_PENDING;
236 
237 	cdev->private->state = DEV_STATE_CMFCHANGE;
238 	cdev->private->cmb_wait = &set_data;
239 	spin_unlock_irq(cdev->ccwlock);
240 
241 	ret = wait_event_interruptible_timeout(set_data.wait,
242 					       set_data.ret != CMF_PENDING,
243 					       SET_SCHIB_TIMEOUT);
244 	spin_lock_irq(cdev->ccwlock);
245 	if (ret <= 0) {
246 		if (set_data.ret == CMF_PENDING) {
247 			set_data.ret = (ret == 0) ? -ETIME : ret;
248 			if (cdev->private->state == DEV_STATE_CMFCHANGE)
249 				cdev->private->state = DEV_STATE_ONLINE;
250 		}
251 	}
252 	cdev->private->cmb_wait = NULL;
253 	ret = set_data.ret;
254 out:
255 	spin_unlock_irq(cdev->ccwlock);
256 	return ret;
257 }
258 
259 void retry_set_schib(struct ccw_device *cdev)
260 {
261 	struct set_schib_struct *set_data = cdev->private->cmb_wait;
262 
263 	if (!set_data)
264 		return;
265 
266 	set_data->ret = set_schib(cdev, set_data->mme, set_data->mbfc,
267 				  set_data->address);
268 	wake_up(&set_data->wait);
269 }
270 
271 static int cmf_copy_block(struct ccw_device *cdev)
272 {
273 	struct subchannel *sch = to_subchannel(cdev->dev.parent);
274 	struct cmb_data *cmb_data;
275 	void *hw_block;
276 
277 	if (cio_update_schib(sch))
278 		return -ENODEV;
279 
280 	if (scsw_fctl(&sch->schib.scsw) & SCSW_FCTL_START_FUNC) {
281 		/* Don't copy if a start function is in progress. */
282 		if ((!(scsw_actl(&sch->schib.scsw) & SCSW_ACTL_SUSPENDED)) &&
283 		    (scsw_actl(&sch->schib.scsw) &
284 		     (SCSW_ACTL_DEVACT | SCSW_ACTL_SCHACT)) &&
285 		    (!(scsw_stctl(&sch->schib.scsw) & SCSW_STCTL_SEC_STATUS)))
286 			return -EBUSY;
287 	}
288 	cmb_data = cdev->private->cmb;
289 	hw_block = cmb_data->hw_block;
290 	memcpy(cmb_data->last_block, hw_block, cmb_data->size);
291 	cmb_data->last_update = get_tod_clock();
292 	return 0;
293 }
294 
295 struct copy_block_struct {
296 	wait_queue_head_t wait;
297 	int ret;
298 };
299 
300 static int cmf_cmb_copy_wait(struct ccw_device *cdev)
301 {
302 	struct copy_block_struct copy_block;
303 	int ret = -ENODEV;
304 
305 	spin_lock_irq(cdev->ccwlock);
306 	if (!cdev->private->cmb)
307 		goto out;
308 
309 	ret = cmf_copy_block(cdev);
310 	if (ret != -EBUSY)
311 		goto out;
312 
313 	if (cdev->private->state != DEV_STATE_ONLINE)
314 		goto out;
315 
316 	init_waitqueue_head(&copy_block.wait);
317 	copy_block.ret = CMF_PENDING;
318 
319 	cdev->private->state = DEV_STATE_CMFUPDATE;
320 	cdev->private->cmb_wait = &copy_block;
321 	spin_unlock_irq(cdev->ccwlock);
322 
323 	ret = wait_event_interruptible(copy_block.wait,
324 				       copy_block.ret != CMF_PENDING);
325 	spin_lock_irq(cdev->ccwlock);
326 	if (ret) {
327 		if (copy_block.ret == CMF_PENDING) {
328 			copy_block.ret = -ERESTARTSYS;
329 			if (cdev->private->state == DEV_STATE_CMFUPDATE)
330 				cdev->private->state = DEV_STATE_ONLINE;
331 		}
332 	}
333 	cdev->private->cmb_wait = NULL;
334 	ret = copy_block.ret;
335 out:
336 	spin_unlock_irq(cdev->ccwlock);
337 	return ret;
338 }
339 
340 void cmf_retry_copy_block(struct ccw_device *cdev)
341 {
342 	struct copy_block_struct *copy_block = cdev->private->cmb_wait;
343 
344 	if (!copy_block)
345 		return;
346 
347 	copy_block->ret = cmf_copy_block(cdev);
348 	wake_up(&copy_block->wait);
349 }
350 
351 static void cmf_generic_reset(struct ccw_device *cdev)
352 {
353 	struct cmb_data *cmb_data;
354 
355 	spin_lock_irq(cdev->ccwlock);
356 	cmb_data = cdev->private->cmb;
357 	if (cmb_data) {
358 		memset(cmb_data->last_block, 0, cmb_data->size);
359 		/*
360 		 * Need to reset hw block as well to make the hardware start
361 		 * from 0 again.
362 		 */
363 		memset(cmb_data->hw_block, 0, cmb_data->size);
364 		cmb_data->last_update = 0;
365 	}
366 	cdev->private->cmb_start_time = get_tod_clock();
367 	spin_unlock_irq(cdev->ccwlock);
368 }
369 
370 /**
371  * struct cmb_area - container for global cmb data
372  *
373  * @mem:	pointer to CMBs (only in basic measurement mode)
374  * @list:	contains a linked list of all subchannels
375  * @num_channels: number of channels to be measured
376  * @lock:	protect concurrent access to @mem and @list
377  */
378 struct cmb_area {
379 	struct cmb *mem;
380 	struct list_head list;
381 	int num_channels;
382 	spinlock_t lock;
383 };
384 
385 static struct cmb_area cmb_area = {
386 	.lock = __SPIN_LOCK_UNLOCKED(cmb_area.lock),
387 	.list = LIST_HEAD_INIT(cmb_area.list),
388 	.num_channels  = 1024,
389 };
390 
391 /* ****** old style CMB handling ********/
392 
393 /*
394  * Basic channel measurement blocks are allocated in one contiguous
395  * block of memory, which can not be moved as long as any channel
396  * is active. Therefore, a maximum number of subchannels needs to
397  * be defined somewhere. This is a module parameter, defaulting to
398  * a reasonable value of 1024, or 32 kb of memory.
399  * Current kernels don't allow kmalloc with more than 128kb, so the
400  * maximum is 4096.
401  */
402 
403 module_param_named(maxchannels, cmb_area.num_channels, uint, 0444);
404 
405 /**
406  * struct cmb - basic channel measurement block
407  * @ssch_rsch_count: number of ssch and rsch
408  * @sample_count: number of samples
409  * @device_connect_time: time of device connect
410  * @function_pending_time: time of function pending
411  * @device_disconnect_time: time of device disconnect
412  * @control_unit_queuing_time: time of control unit queuing
413  * @device_active_only_time: time of device active only
414  * @reserved: unused in basic measurement mode
415  *
416  * The measurement block as used by the hardware. The fields are described
417  * further in z/Architecture Principles of Operation, chapter 17.
418  *
419  * The cmb area made up from these blocks must be a contiguous array and may
420  * not be reallocated or freed.
421  * Only one cmb area can be present in the system.
422  */
423 struct cmb {
424 	u16 ssch_rsch_count;
425 	u16 sample_count;
426 	u32 device_connect_time;
427 	u32 function_pending_time;
428 	u32 device_disconnect_time;
429 	u32 control_unit_queuing_time;
430 	u32 device_active_only_time;
431 	u32 reserved[2];
432 };
433 
434 /*
435  * Insert a single device into the cmb_area list.
436  * Called with cmb_area.lock held from alloc_cmb.
437  */
438 static int alloc_cmb_single(struct ccw_device *cdev,
439 			    struct cmb_data *cmb_data)
440 {
441 	struct cmb *cmb;
442 	struct ccw_device_private *node;
443 	int ret;
444 
445 	spin_lock_irq(cdev->ccwlock);
446 	if (!list_empty(&cdev->private->cmb_list)) {
447 		ret = -EBUSY;
448 		goto out;
449 	}
450 
451 	/*
452 	 * Find first unused cmb in cmb_area.mem.
453 	 * This is a little tricky: cmb_area.list
454 	 * remains sorted by ->cmb->hw_data pointers.
455 	 */
456 	cmb = cmb_area.mem;
457 	list_for_each_entry(node, &cmb_area.list, cmb_list) {
458 		struct cmb_data *data;
459 		data = node->cmb;
460 		if ((struct cmb*)data->hw_block > cmb)
461 			break;
462 		cmb++;
463 	}
464 	if (cmb - cmb_area.mem >= cmb_area.num_channels) {
465 		ret = -ENOMEM;
466 		goto out;
467 	}
468 
469 	/* insert new cmb */
470 	list_add_tail(&cdev->private->cmb_list, &node->cmb_list);
471 	cmb_data->hw_block = cmb;
472 	cdev->private->cmb = cmb_data;
473 	ret = 0;
474 out:
475 	spin_unlock_irq(cdev->ccwlock);
476 	return ret;
477 }
478 
479 static int alloc_cmb(struct ccw_device *cdev)
480 {
481 	int ret;
482 	struct cmb *mem;
483 	ssize_t size;
484 	struct cmb_data *cmb_data;
485 
486 	/* Allocate private cmb_data. */
487 	cmb_data = kzalloc(sizeof(struct cmb_data), GFP_KERNEL);
488 	if (!cmb_data)
489 		return -ENOMEM;
490 
491 	cmb_data->last_block = kzalloc(sizeof(struct cmb), GFP_KERNEL);
492 	if (!cmb_data->last_block) {
493 		kfree(cmb_data);
494 		return -ENOMEM;
495 	}
496 	cmb_data->size = sizeof(struct cmb);
497 	spin_lock(&cmb_area.lock);
498 
499 	if (!cmb_area.mem) {
500 		/* there is no user yet, so we need a new area */
501 		size = sizeof(struct cmb) * cmb_area.num_channels;
502 		WARN_ON(!list_empty(&cmb_area.list));
503 
504 		spin_unlock(&cmb_area.lock);
505 		mem = (void *)__get_free_pages(GFP_KERNEL, get_order(size));
506 		spin_lock(&cmb_area.lock);
507 
508 		if (cmb_area.mem) {
509 			/* ok, another thread was faster */
510 			free_pages((unsigned long)mem, get_order(size));
511 		} else if (!mem) {
512 			/* no luck */
513 			ret = -ENOMEM;
514 			goto out;
515 		} else {
516 			/* everything ok */
517 			memset(mem, 0, size);
518 			cmb_area.mem = mem;
519 			cmf_activate(cmb_area.mem, CMF_ON);
520 		}
521 	}
522 
523 	/* do the actual allocation */
524 	ret = alloc_cmb_single(cdev, cmb_data);
525 out:
526 	spin_unlock(&cmb_area.lock);
527 	if (ret) {
528 		kfree(cmb_data->last_block);
529 		kfree(cmb_data);
530 	}
531 	return ret;
532 }
533 
534 static void free_cmb(struct ccw_device *cdev)
535 {
536 	struct ccw_device_private *priv;
537 	struct cmb_data *cmb_data;
538 
539 	spin_lock(&cmb_area.lock);
540 	spin_lock_irq(cdev->ccwlock);
541 
542 	priv = cdev->private;
543 	cmb_data = priv->cmb;
544 	priv->cmb = NULL;
545 	if (cmb_data)
546 		kfree(cmb_data->last_block);
547 	kfree(cmb_data);
548 	list_del_init(&priv->cmb_list);
549 
550 	if (list_empty(&cmb_area.list)) {
551 		ssize_t size;
552 		size = sizeof(struct cmb) * cmb_area.num_channels;
553 		cmf_activate(NULL, CMF_OFF);
554 		free_pages((unsigned long)cmb_area.mem, get_order(size));
555 		cmb_area.mem = NULL;
556 	}
557 	spin_unlock_irq(cdev->ccwlock);
558 	spin_unlock(&cmb_area.lock);
559 }
560 
561 static int set_cmb(struct ccw_device *cdev, u32 mme)
562 {
563 	u16 offset;
564 	struct cmb_data *cmb_data;
565 	unsigned long flags;
566 
567 	spin_lock_irqsave(cdev->ccwlock, flags);
568 	if (!cdev->private->cmb) {
569 		spin_unlock_irqrestore(cdev->ccwlock, flags);
570 		return -EINVAL;
571 	}
572 	cmb_data = cdev->private->cmb;
573 	offset = mme ? (struct cmb *)cmb_data->hw_block - cmb_area.mem : 0;
574 	spin_unlock_irqrestore(cdev->ccwlock, flags);
575 
576 	return set_schib_wait(cdev, mme, 0, offset);
577 }
578 
579 /* calculate utilization in 0.1 percent units */
580 static u64 __cmb_utilization(u64 device_connect_time, u64 function_pending_time,
581 			     u64 device_disconnect_time, u64 start_time)
582 {
583 	u64 utilization, elapsed_time;
584 
585 	utilization = time_to_nsec(device_connect_time +
586 				   function_pending_time +
587 				   device_disconnect_time);
588 
589 	elapsed_time = get_tod_clock() - start_time;
590 	elapsed_time = tod_to_ns(elapsed_time);
591 	elapsed_time /= 1000;
592 
593 	return elapsed_time ? (utilization / elapsed_time) : 0;
594 }
595 
596 static u64 read_cmb(struct ccw_device *cdev, int index)
597 {
598 	struct cmb_data *cmb_data;
599 	unsigned long flags;
600 	struct cmb *cmb;
601 	u64 ret = 0;
602 	u32 val;
603 
604 	spin_lock_irqsave(cdev->ccwlock, flags);
605 	cmb_data = cdev->private->cmb;
606 	if (!cmb_data)
607 		goto out;
608 
609 	cmb = cmb_data->hw_block;
610 	switch (index) {
611 	case avg_utilization:
612 		ret = __cmb_utilization(cmb->device_connect_time,
613 					cmb->function_pending_time,
614 					cmb->device_disconnect_time,
615 					cdev->private->cmb_start_time);
616 		goto out;
617 	case cmb_ssch_rsch_count:
618 		ret = cmb->ssch_rsch_count;
619 		goto out;
620 	case cmb_sample_count:
621 		ret = cmb->sample_count;
622 		goto out;
623 	case cmb_device_connect_time:
624 		val = cmb->device_connect_time;
625 		break;
626 	case cmb_function_pending_time:
627 		val = cmb->function_pending_time;
628 		break;
629 	case cmb_device_disconnect_time:
630 		val = cmb->device_disconnect_time;
631 		break;
632 	case cmb_control_unit_queuing_time:
633 		val = cmb->control_unit_queuing_time;
634 		break;
635 	case cmb_device_active_only_time:
636 		val = cmb->device_active_only_time;
637 		break;
638 	default:
639 		goto out;
640 	}
641 	ret = time_to_avg_nsec(val, cmb->sample_count);
642 out:
643 	spin_unlock_irqrestore(cdev->ccwlock, flags);
644 	return ret;
645 }
646 
647 static int readall_cmb(struct ccw_device *cdev, struct cmbdata *data)
648 {
649 	struct cmb *cmb;
650 	struct cmb_data *cmb_data;
651 	u64 time;
652 	unsigned long flags;
653 	int ret;
654 
655 	ret = cmf_cmb_copy_wait(cdev);
656 	if (ret < 0)
657 		return ret;
658 	spin_lock_irqsave(cdev->ccwlock, flags);
659 	cmb_data = cdev->private->cmb;
660 	if (!cmb_data) {
661 		ret = -ENODEV;
662 		goto out;
663 	}
664 	if (cmb_data->last_update == 0) {
665 		ret = -EAGAIN;
666 		goto out;
667 	}
668 	cmb = cmb_data->last_block;
669 	time = cmb_data->last_update - cdev->private->cmb_start_time;
670 
671 	memset(data, 0, sizeof(struct cmbdata));
672 
673 	/* we only know values before device_busy_time */
674 	data->size = offsetof(struct cmbdata, device_busy_time);
675 
676 	data->elapsed_time = tod_to_ns(time);
677 
678 	/* copy data to new structure */
679 	data->ssch_rsch_count = cmb->ssch_rsch_count;
680 	data->sample_count = cmb->sample_count;
681 
682 	/* time fields are converted to nanoseconds while copying */
683 	data->device_connect_time = time_to_nsec(cmb->device_connect_time);
684 	data->function_pending_time = time_to_nsec(cmb->function_pending_time);
685 	data->device_disconnect_time =
686 		time_to_nsec(cmb->device_disconnect_time);
687 	data->control_unit_queuing_time
688 		= time_to_nsec(cmb->control_unit_queuing_time);
689 	data->device_active_only_time
690 		= time_to_nsec(cmb->device_active_only_time);
691 	ret = 0;
692 out:
693 	spin_unlock_irqrestore(cdev->ccwlock, flags);
694 	return ret;
695 }
696 
697 static void reset_cmb(struct ccw_device *cdev)
698 {
699 	cmf_generic_reset(cdev);
700 }
701 
702 static int cmf_enabled(struct ccw_device *cdev)
703 {
704 	int enabled;
705 
706 	spin_lock_irq(cdev->ccwlock);
707 	enabled = !!cdev->private->cmb;
708 	spin_unlock_irq(cdev->ccwlock);
709 
710 	return enabled;
711 }
712 
713 static struct attribute_group cmf_attr_group;
714 
715 static struct cmb_operations cmbops_basic = {
716 	.alloc	= alloc_cmb,
717 	.free	= free_cmb,
718 	.set	= set_cmb,
719 	.read	= read_cmb,
720 	.readall    = readall_cmb,
721 	.reset	    = reset_cmb,
722 	.attr_group = &cmf_attr_group,
723 };
724 
725 /* ******** extended cmb handling ********/
726 
727 /**
728  * struct cmbe - extended channel measurement block
729  * @ssch_rsch_count: number of ssch and rsch
730  * @sample_count: number of samples
731  * @device_connect_time: time of device connect
732  * @function_pending_time: time of function pending
733  * @device_disconnect_time: time of device disconnect
734  * @control_unit_queuing_time: time of control unit queuing
735  * @device_active_only_time: time of device active only
736  * @device_busy_time: time of device busy
737  * @initial_command_response_time: initial command response time
738  * @reserved: unused
739  *
740  * The measurement block as used by the hardware. May be in any 64 bit physical
741  * location.
742  * The fields are described further in z/Architecture Principles of Operation,
743  * third edition, chapter 17.
744  */
745 struct cmbe {
746 	u32 ssch_rsch_count;
747 	u32 sample_count;
748 	u32 device_connect_time;
749 	u32 function_pending_time;
750 	u32 device_disconnect_time;
751 	u32 control_unit_queuing_time;
752 	u32 device_active_only_time;
753 	u32 device_busy_time;
754 	u32 initial_command_response_time;
755 	u32 reserved[7];
756 } __packed __aligned(64);
757 
758 static struct kmem_cache *cmbe_cache;
759 
760 static int alloc_cmbe(struct ccw_device *cdev)
761 {
762 	struct cmb_data *cmb_data;
763 	struct cmbe *cmbe;
764 	int ret = -ENOMEM;
765 
766 	cmbe = kmem_cache_zalloc(cmbe_cache, GFP_KERNEL);
767 	if (!cmbe)
768 		return ret;
769 
770 	cmb_data = kzalloc(sizeof(*cmb_data), GFP_KERNEL);
771 	if (!cmb_data)
772 		goto out_free;
773 
774 	cmb_data->last_block = kzalloc(sizeof(struct cmbe), GFP_KERNEL);
775 	if (!cmb_data->last_block)
776 		goto out_free;
777 
778 	cmb_data->size = sizeof(*cmbe);
779 	cmb_data->hw_block = cmbe;
780 
781 	spin_lock(&cmb_area.lock);
782 	spin_lock_irq(cdev->ccwlock);
783 	if (cdev->private->cmb)
784 		goto out_unlock;
785 
786 	cdev->private->cmb = cmb_data;
787 
788 	/* activate global measurement if this is the first channel */
789 	if (list_empty(&cmb_area.list))
790 		cmf_activate(NULL, CMF_ON);
791 	list_add_tail(&cdev->private->cmb_list, &cmb_area.list);
792 
793 	spin_unlock_irq(cdev->ccwlock);
794 	spin_unlock(&cmb_area.lock);
795 	return 0;
796 
797 out_unlock:
798 	spin_unlock_irq(cdev->ccwlock);
799 	spin_unlock(&cmb_area.lock);
800 	ret = -EBUSY;
801 out_free:
802 	if (cmb_data)
803 		kfree(cmb_data->last_block);
804 	kfree(cmb_data);
805 	kmem_cache_free(cmbe_cache, cmbe);
806 
807 	return ret;
808 }
809 
810 static void free_cmbe(struct ccw_device *cdev)
811 {
812 	struct cmb_data *cmb_data;
813 
814 	spin_lock(&cmb_area.lock);
815 	spin_lock_irq(cdev->ccwlock);
816 	cmb_data = cdev->private->cmb;
817 	cdev->private->cmb = NULL;
818 	if (cmb_data) {
819 		kfree(cmb_data->last_block);
820 		kmem_cache_free(cmbe_cache, cmb_data->hw_block);
821 	}
822 	kfree(cmb_data);
823 
824 	/* deactivate global measurement if this is the last channel */
825 	list_del_init(&cdev->private->cmb_list);
826 	if (list_empty(&cmb_area.list))
827 		cmf_activate(NULL, CMF_OFF);
828 	spin_unlock_irq(cdev->ccwlock);
829 	spin_unlock(&cmb_area.lock);
830 }
831 
832 static int set_cmbe(struct ccw_device *cdev, u32 mme)
833 {
834 	unsigned long mba;
835 	struct cmb_data *cmb_data;
836 	unsigned long flags;
837 
838 	spin_lock_irqsave(cdev->ccwlock, flags);
839 	if (!cdev->private->cmb) {
840 		spin_unlock_irqrestore(cdev->ccwlock, flags);
841 		return -EINVAL;
842 	}
843 	cmb_data = cdev->private->cmb;
844 	mba = mme ? (unsigned long) cmb_data->hw_block : 0;
845 	spin_unlock_irqrestore(cdev->ccwlock, flags);
846 
847 	return set_schib_wait(cdev, mme, 1, mba);
848 }
849 
850 static u64 read_cmbe(struct ccw_device *cdev, int index)
851 {
852 	struct cmb_data *cmb_data;
853 	unsigned long flags;
854 	struct cmbe *cmb;
855 	u64 ret = 0;
856 	u32 val;
857 
858 	spin_lock_irqsave(cdev->ccwlock, flags);
859 	cmb_data = cdev->private->cmb;
860 	if (!cmb_data)
861 		goto out;
862 
863 	cmb = cmb_data->hw_block;
864 	switch (index) {
865 	case avg_utilization:
866 		ret = __cmb_utilization(cmb->device_connect_time,
867 					cmb->function_pending_time,
868 					cmb->device_disconnect_time,
869 					cdev->private->cmb_start_time);
870 		goto out;
871 	case cmb_ssch_rsch_count:
872 		ret = cmb->ssch_rsch_count;
873 		goto out;
874 	case cmb_sample_count:
875 		ret = cmb->sample_count;
876 		goto out;
877 	case cmb_device_connect_time:
878 		val = cmb->device_connect_time;
879 		break;
880 	case cmb_function_pending_time:
881 		val = cmb->function_pending_time;
882 		break;
883 	case cmb_device_disconnect_time:
884 		val = cmb->device_disconnect_time;
885 		break;
886 	case cmb_control_unit_queuing_time:
887 		val = cmb->control_unit_queuing_time;
888 		break;
889 	case cmb_device_active_only_time:
890 		val = cmb->device_active_only_time;
891 		break;
892 	case cmb_device_busy_time:
893 		val = cmb->device_busy_time;
894 		break;
895 	case cmb_initial_command_response_time:
896 		val = cmb->initial_command_response_time;
897 		break;
898 	default:
899 		goto out;
900 	}
901 	ret = time_to_avg_nsec(val, cmb->sample_count);
902 out:
903 	spin_unlock_irqrestore(cdev->ccwlock, flags);
904 	return ret;
905 }
906 
907 static int readall_cmbe(struct ccw_device *cdev, struct cmbdata *data)
908 {
909 	struct cmbe *cmb;
910 	struct cmb_data *cmb_data;
911 	u64 time;
912 	unsigned long flags;
913 	int ret;
914 
915 	ret = cmf_cmb_copy_wait(cdev);
916 	if (ret < 0)
917 		return ret;
918 	spin_lock_irqsave(cdev->ccwlock, flags);
919 	cmb_data = cdev->private->cmb;
920 	if (!cmb_data) {
921 		ret = -ENODEV;
922 		goto out;
923 	}
924 	if (cmb_data->last_update == 0) {
925 		ret = -EAGAIN;
926 		goto out;
927 	}
928 	time = cmb_data->last_update - cdev->private->cmb_start_time;
929 
930 	memset (data, 0, sizeof(struct cmbdata));
931 
932 	/* we only know values before device_busy_time */
933 	data->size = offsetof(struct cmbdata, device_busy_time);
934 
935 	data->elapsed_time = tod_to_ns(time);
936 
937 	cmb = cmb_data->last_block;
938 	/* copy data to new structure */
939 	data->ssch_rsch_count = cmb->ssch_rsch_count;
940 	data->sample_count = cmb->sample_count;
941 
942 	/* time fields are converted to nanoseconds while copying */
943 	data->device_connect_time = time_to_nsec(cmb->device_connect_time);
944 	data->function_pending_time = time_to_nsec(cmb->function_pending_time);
945 	data->device_disconnect_time =
946 		time_to_nsec(cmb->device_disconnect_time);
947 	data->control_unit_queuing_time
948 		= time_to_nsec(cmb->control_unit_queuing_time);
949 	data->device_active_only_time
950 		= time_to_nsec(cmb->device_active_only_time);
951 	data->device_busy_time = time_to_nsec(cmb->device_busy_time);
952 	data->initial_command_response_time
953 		= time_to_nsec(cmb->initial_command_response_time);
954 
955 	ret = 0;
956 out:
957 	spin_unlock_irqrestore(cdev->ccwlock, flags);
958 	return ret;
959 }
960 
961 static void reset_cmbe(struct ccw_device *cdev)
962 {
963 	cmf_generic_reset(cdev);
964 }
965 
966 static struct attribute_group cmf_attr_group_ext;
967 
968 static struct cmb_operations cmbops_extended = {
969 	.alloc	    = alloc_cmbe,
970 	.free	    = free_cmbe,
971 	.set	    = set_cmbe,
972 	.read	    = read_cmbe,
973 	.readall    = readall_cmbe,
974 	.reset	    = reset_cmbe,
975 	.attr_group = &cmf_attr_group_ext,
976 };
977 
978 static ssize_t cmb_show_attr(struct device *dev, char *buf, enum cmb_index idx)
979 {
980 	return sprintf(buf, "%lld\n",
981 		(unsigned long long) cmf_read(to_ccwdev(dev), idx));
982 }
983 
984 static ssize_t cmb_show_avg_sample_interval(struct device *dev,
985 					    struct device_attribute *attr,
986 					    char *buf)
987 {
988 	struct ccw_device *cdev = to_ccwdev(dev);
989 	unsigned long count;
990 	long interval;
991 
992 	count = cmf_read(cdev, cmb_sample_count);
993 	spin_lock_irq(cdev->ccwlock);
994 	if (count) {
995 		interval = get_tod_clock() - cdev->private->cmb_start_time;
996 		interval = tod_to_ns(interval);
997 		interval /= count;
998 	} else
999 		interval = -1;
1000 	spin_unlock_irq(cdev->ccwlock);
1001 	return sprintf(buf, "%ld\n", interval);
1002 }
1003 
1004 static ssize_t cmb_show_avg_utilization(struct device *dev,
1005 					struct device_attribute *attr,
1006 					char *buf)
1007 {
1008 	unsigned long u = cmf_read(to_ccwdev(dev), avg_utilization);
1009 
1010 	return sprintf(buf, "%02lu.%01lu%%\n", u / 10, u % 10);
1011 }
1012 
1013 #define cmf_attr(name) \
1014 static ssize_t show_##name(struct device *dev, \
1015 			   struct device_attribute *attr, char *buf)	\
1016 { return cmb_show_attr((dev), buf, cmb_##name); } \
1017 static DEVICE_ATTR(name, 0444, show_##name, NULL);
1018 
1019 #define cmf_attr_avg(name) \
1020 static ssize_t show_avg_##name(struct device *dev, \
1021 			       struct device_attribute *attr, char *buf) \
1022 { return cmb_show_attr((dev), buf, cmb_##name); } \
1023 static DEVICE_ATTR(avg_##name, 0444, show_avg_##name, NULL);
1024 
1025 cmf_attr(ssch_rsch_count);
1026 cmf_attr(sample_count);
1027 cmf_attr_avg(device_connect_time);
1028 cmf_attr_avg(function_pending_time);
1029 cmf_attr_avg(device_disconnect_time);
1030 cmf_attr_avg(control_unit_queuing_time);
1031 cmf_attr_avg(device_active_only_time);
1032 cmf_attr_avg(device_busy_time);
1033 cmf_attr_avg(initial_command_response_time);
1034 
1035 static DEVICE_ATTR(avg_sample_interval, 0444, cmb_show_avg_sample_interval,
1036 		   NULL);
1037 static DEVICE_ATTR(avg_utilization, 0444, cmb_show_avg_utilization, NULL);
1038 
1039 static struct attribute *cmf_attributes[] = {
1040 	&dev_attr_avg_sample_interval.attr,
1041 	&dev_attr_avg_utilization.attr,
1042 	&dev_attr_ssch_rsch_count.attr,
1043 	&dev_attr_sample_count.attr,
1044 	&dev_attr_avg_device_connect_time.attr,
1045 	&dev_attr_avg_function_pending_time.attr,
1046 	&dev_attr_avg_device_disconnect_time.attr,
1047 	&dev_attr_avg_control_unit_queuing_time.attr,
1048 	&dev_attr_avg_device_active_only_time.attr,
1049 	NULL,
1050 };
1051 
1052 static struct attribute_group cmf_attr_group = {
1053 	.name  = "cmf",
1054 	.attrs = cmf_attributes,
1055 };
1056 
1057 static struct attribute *cmf_attributes_ext[] = {
1058 	&dev_attr_avg_sample_interval.attr,
1059 	&dev_attr_avg_utilization.attr,
1060 	&dev_attr_ssch_rsch_count.attr,
1061 	&dev_attr_sample_count.attr,
1062 	&dev_attr_avg_device_connect_time.attr,
1063 	&dev_attr_avg_function_pending_time.attr,
1064 	&dev_attr_avg_device_disconnect_time.attr,
1065 	&dev_attr_avg_control_unit_queuing_time.attr,
1066 	&dev_attr_avg_device_active_only_time.attr,
1067 	&dev_attr_avg_device_busy_time.attr,
1068 	&dev_attr_avg_initial_command_response_time.attr,
1069 	NULL,
1070 };
1071 
1072 static struct attribute_group cmf_attr_group_ext = {
1073 	.name  = "cmf",
1074 	.attrs = cmf_attributes_ext,
1075 };
1076 
1077 static ssize_t cmb_enable_show(struct device *dev,
1078 			       struct device_attribute *attr,
1079 			       char *buf)
1080 {
1081 	struct ccw_device *cdev = to_ccwdev(dev);
1082 
1083 	return sprintf(buf, "%d\n", cmf_enabled(cdev));
1084 }
1085 
1086 static ssize_t cmb_enable_store(struct device *dev,
1087 				struct device_attribute *attr, const char *buf,
1088 				size_t c)
1089 {
1090 	struct ccw_device *cdev = to_ccwdev(dev);
1091 	unsigned long val;
1092 	int ret;
1093 
1094 	ret = kstrtoul(buf, 16, &val);
1095 	if (ret)
1096 		return ret;
1097 
1098 	switch (val) {
1099 	case 0:
1100 		ret = disable_cmf(cdev);
1101 		break;
1102 	case 1:
1103 		ret = enable_cmf(cdev);
1104 		break;
1105 	default:
1106 		ret = -EINVAL;
1107 	}
1108 
1109 	return ret ? ret : c;
1110 }
1111 DEVICE_ATTR_RW(cmb_enable);
1112 
1113 /**
1114  * enable_cmf() - switch on the channel measurement for a specific device
1115  *  @cdev:	The ccw device to be enabled
1116  *
1117  *  Enable channel measurements for @cdev. If this is called on a device
1118  *  for which channel measurement is already enabled a reset of the
1119  *  measurement data is triggered.
1120  *  Returns: %0 for success or a negative error value.
1121  *  Context:
1122  *    non-atomic
1123  */
1124 int enable_cmf(struct ccw_device *cdev)
1125 {
1126 	int ret = 0;
1127 
1128 	device_lock(&cdev->dev);
1129 	if (cmf_enabled(cdev)) {
1130 		cmbops->reset(cdev);
1131 		goto out_unlock;
1132 	}
1133 	get_device(&cdev->dev);
1134 	ret = cmbops->alloc(cdev);
1135 	if (ret)
1136 		goto out;
1137 	cmbops->reset(cdev);
1138 	ret = sysfs_create_group(&cdev->dev.kobj, cmbops->attr_group);
1139 	if (ret) {
1140 		cmbops->free(cdev);
1141 		goto out;
1142 	}
1143 	ret = cmbops->set(cdev, 2);
1144 	if (ret) {
1145 		sysfs_remove_group(&cdev->dev.kobj, cmbops->attr_group);
1146 		cmbops->free(cdev);
1147 	}
1148 out:
1149 	if (ret)
1150 		put_device(&cdev->dev);
1151 out_unlock:
1152 	device_unlock(&cdev->dev);
1153 	return ret;
1154 }
1155 
1156 /**
1157  * __disable_cmf() - switch off the channel measurement for a specific device
1158  *  @cdev:	The ccw device to be disabled
1159  *
1160  *  Returns: %0 for success or a negative error value.
1161  *
1162  *  Context:
1163  *    non-atomic, device_lock() held.
1164  */
1165 int __disable_cmf(struct ccw_device *cdev)
1166 {
1167 	int ret;
1168 
1169 	ret = cmbops->set(cdev, 0);
1170 	if (ret)
1171 		return ret;
1172 
1173 	sysfs_remove_group(&cdev->dev.kobj, cmbops->attr_group);
1174 	cmbops->free(cdev);
1175 	put_device(&cdev->dev);
1176 
1177 	return ret;
1178 }
1179 
1180 /**
1181  * disable_cmf() - switch off the channel measurement for a specific device
1182  *  @cdev:	The ccw device to be disabled
1183  *
1184  *  Returns: %0 for success or a negative error value.
1185  *
1186  *  Context:
1187  *    non-atomic
1188  */
1189 int disable_cmf(struct ccw_device *cdev)
1190 {
1191 	int ret;
1192 
1193 	device_lock(&cdev->dev);
1194 	ret = __disable_cmf(cdev);
1195 	device_unlock(&cdev->dev);
1196 
1197 	return ret;
1198 }
1199 
1200 /**
1201  * cmf_read() - read one value from the current channel measurement block
1202  * @cdev:	the channel to be read
1203  * @index:	the index of the value to be read
1204  *
1205  * Returns: The value read or %0 if the value cannot be read.
1206  *
1207  *  Context:
1208  *    any
1209  */
1210 u64 cmf_read(struct ccw_device *cdev, int index)
1211 {
1212 	return cmbops->read(cdev, index);
1213 }
1214 
1215 /**
1216  * cmf_readall() - read the current channel measurement block
1217  * @cdev:	the channel to be read
1218  * @data:	a pointer to a data block that will be filled
1219  *
1220  * Returns: %0 on success, a negative error value otherwise.
1221  *
1222  *  Context:
1223  *    any
1224  */
1225 int cmf_readall(struct ccw_device *cdev, struct cmbdata *data)
1226 {
1227 	return cmbops->readall(cdev, data);
1228 }
1229 
1230 /* Reenable cmf when a disconnected device becomes available again. */
1231 int cmf_reenable(struct ccw_device *cdev)
1232 {
1233 	cmbops->reset(cdev);
1234 	return cmbops->set(cdev, 2);
1235 }
1236 
1237 /**
1238  * cmf_reactivate() - reactivate measurement block updates
1239  *
1240  * Use this during resume from hibernate.
1241  */
1242 void cmf_reactivate(void)
1243 {
1244 	spin_lock(&cmb_area.lock);
1245 	if (!list_empty(&cmb_area.list))
1246 		cmf_activate(cmb_area.mem, CMF_ON);
1247 	spin_unlock(&cmb_area.lock);
1248 }
1249 
1250 static int __init init_cmbe(void)
1251 {
1252 	cmbe_cache = kmem_cache_create("cmbe_cache", sizeof(struct cmbe),
1253 				       __alignof__(struct cmbe), 0, NULL);
1254 
1255 	return cmbe_cache ? 0 : -ENOMEM;
1256 }
1257 
1258 static int __init init_cmf(void)
1259 {
1260 	char *format_string;
1261 	char *detect_string;
1262 	int ret;
1263 
1264 	/*
1265 	 * If the user did not give a parameter, see if we are running on a
1266 	 * machine supporting extended measurement blocks, otherwise fall back
1267 	 * to basic mode.
1268 	 */
1269 	if (format == CMF_AUTODETECT) {
1270 		if (!css_general_characteristics.ext_mb) {
1271 			format = CMF_BASIC;
1272 		} else {
1273 			format = CMF_EXTENDED;
1274 		}
1275 		detect_string = "autodetected";
1276 	} else {
1277 		detect_string = "parameter";
1278 	}
1279 
1280 	switch (format) {
1281 	case CMF_BASIC:
1282 		format_string = "basic";
1283 		cmbops = &cmbops_basic;
1284 		break;
1285 	case CMF_EXTENDED:
1286 		format_string = "extended";
1287 		cmbops = &cmbops_extended;
1288 
1289 		ret = init_cmbe();
1290 		if (ret)
1291 			return ret;
1292 		break;
1293 	default:
1294 		return -EINVAL;
1295 	}
1296 	pr_info("Channel measurement facility initialized using format "
1297 		"%s (mode %s)\n", format_string, detect_string);
1298 	return 0;
1299 }
1300 device_initcall(init_cmf);
1301 
1302 EXPORT_SYMBOL_GPL(enable_cmf);
1303 EXPORT_SYMBOL_GPL(disable_cmf);
1304 EXPORT_SYMBOL_GPL(cmf_read);
1305 EXPORT_SYMBOL_GPL(cmf_readall);
1306