1 /*
2  * arch/s390/appldata/appldata_base.c
3  *
4  * Base infrastructure for Linux-z/VM Monitor Stream, Stage 1.
5  * Exports appldata_register_ops() and appldata_unregister_ops() for the
6  * data gathering modules.
7  *
8  * Copyright IBM Corp. 2003, 2008
9  *
10  * Author: Gerald Schaefer <gerald.schaefer@de.ibm.com>
11  */
12 
13 #define KMSG_COMPONENT	"appldata"
14 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
15 
16 #include <linux/module.h>
17 #include <linux/init.h>
18 #include <linux/slab.h>
19 #include <linux/errno.h>
20 #include <linux/interrupt.h>
21 #include <linux/proc_fs.h>
22 #include <linux/mm.h>
23 #include <linux/swap.h>
24 #include <linux/pagemap.h>
25 #include <linux/sysctl.h>
26 #include <linux/notifier.h>
27 #include <linux/cpu.h>
28 #include <linux/workqueue.h>
29 #include <asm/appldata.h>
30 #include <asm/timer.h>
31 #include <asm/uaccess.h>
32 #include <asm/io.h>
33 #include <asm/smp.h>
34 
35 #include "appldata.h"
36 
37 
38 #define APPLDATA_CPU_INTERVAL	10000		/* default (CPU) time for
39 						   sampling interval in
40 						   milliseconds */
41 
42 #define TOD_MICRO	0x01000			/* nr. of TOD clock units
43 						   for 1 microsecond */
44 /*
45  * /proc entries (sysctl)
46  */
47 static const char appldata_proc_name[APPLDATA_PROC_NAME_LENGTH] = "appldata";
48 static int appldata_timer_handler(ctl_table *ctl, int write, struct file *filp,
49 				  void __user *buffer, size_t *lenp, loff_t *ppos);
50 static int appldata_interval_handler(ctl_table *ctl, int write,
51 					 struct file *filp,
52 					 void __user *buffer,
53 					 size_t *lenp, loff_t *ppos);
54 
55 static struct ctl_table_header *appldata_sysctl_header;
56 static struct ctl_table appldata_table[] = {
57 	{
58 		.procname	= "timer",
59 		.mode		= S_IRUGO | S_IWUSR,
60 		.proc_handler	= &appldata_timer_handler,
61 	},
62 	{
63 		.procname	= "interval",
64 		.mode		= S_IRUGO | S_IWUSR,
65 		.proc_handler	= &appldata_interval_handler,
66 	},
67 	{ },
68 };
69 
70 static struct ctl_table appldata_dir_table[] = {
71 	{
72 		.procname	= appldata_proc_name,
73 		.maxlen		= 0,
74 		.mode		= S_IRUGO | S_IXUGO,
75 		.child		= appldata_table,
76 	},
77 	{ },
78 };
79 
80 /*
81  * Timer
82  */
83 static DEFINE_PER_CPU(struct vtimer_list, appldata_timer);
84 static atomic_t appldata_expire_count = ATOMIC_INIT(0);
85 
86 static DEFINE_SPINLOCK(appldata_timer_lock);
87 static int appldata_interval = APPLDATA_CPU_INTERVAL;
88 static int appldata_timer_active;
89 
90 /*
91  * Work queue
92  */
93 static struct workqueue_struct *appldata_wq;
94 static void appldata_work_fn(struct work_struct *work);
95 static DECLARE_WORK(appldata_work, appldata_work_fn);
96 
97 
98 /*
99  * Ops list
100  */
101 static DEFINE_MUTEX(appldata_ops_mutex);
102 static LIST_HEAD(appldata_ops_list);
103 
104 
105 /*************************** timer, work, DIAG *******************************/
106 /*
107  * appldata_timer_function()
108  *
109  * schedule work and reschedule timer
110  */
111 static void appldata_timer_function(unsigned long data)
112 {
113 	if (atomic_dec_and_test(&appldata_expire_count)) {
114 		atomic_set(&appldata_expire_count, num_online_cpus());
115 		queue_work(appldata_wq, (struct work_struct *) data);
116 	}
117 }
118 
119 /*
120  * appldata_work_fn()
121  *
122  * call data gathering function for each (active) module
123  */
124 static void appldata_work_fn(struct work_struct *work)
125 {
126 	struct list_head *lh;
127 	struct appldata_ops *ops;
128 	int i;
129 
130 	i = 0;
131 	get_online_cpus();
132 	mutex_lock(&appldata_ops_mutex);
133 	list_for_each(lh, &appldata_ops_list) {
134 		ops = list_entry(lh, struct appldata_ops, list);
135 		if (ops->active == 1) {
136 			ops->callback(ops->data);
137 		}
138 	}
139 	mutex_unlock(&appldata_ops_mutex);
140 	put_online_cpus();
141 }
142 
143 /*
144  * appldata_diag()
145  *
146  * prepare parameter list, issue DIAG 0xDC
147  */
148 int appldata_diag(char record_nr, u16 function, unsigned long buffer,
149 			u16 length, char *mod_lvl)
150 {
151 	struct appldata_product_id id = {
152 		.prod_nr    = {0xD3, 0xC9, 0xD5, 0xE4,
153 			       0xE7, 0xD2, 0xD9},	/* "LINUXKR" */
154 		.prod_fn    = 0xD5D3,			/* "NL" */
155 		.version_nr = 0xF2F6,			/* "26" */
156 		.release_nr = 0xF0F1,			/* "01" */
157 	};
158 
159 	id.record_nr = record_nr;
160 	id.mod_lvl = (mod_lvl[0]) << 8 | mod_lvl[1];
161 	return appldata_asm(&id, function, (void *) buffer, length);
162 }
163 /************************ timer, work, DIAG <END> ****************************/
164 
165 
166 /****************************** /proc stuff **********************************/
167 
168 /*
169  * appldata_mod_vtimer_wrap()
170  *
171  * wrapper function for mod_virt_timer(), because smp_call_function_single()
172  * accepts only one parameter.
173  */
174 static void __appldata_mod_vtimer_wrap(void *p) {
175 	struct {
176 		struct vtimer_list *timer;
177 		u64    expires;
178 	} *args = p;
179 	mod_virt_timer_periodic(args->timer, args->expires);
180 }
181 
182 #define APPLDATA_ADD_TIMER	0
183 #define APPLDATA_DEL_TIMER	1
184 #define APPLDATA_MOD_TIMER	2
185 
186 /*
187  * __appldata_vtimer_setup()
188  *
189  * Add, delete or modify virtual timers on all online cpus.
190  * The caller needs to get the appldata_timer_lock spinlock.
191  */
192 static void
193 __appldata_vtimer_setup(int cmd)
194 {
195 	u64 per_cpu_interval;
196 	int i;
197 
198 	switch (cmd) {
199 	case APPLDATA_ADD_TIMER:
200 		if (appldata_timer_active)
201 			break;
202 		per_cpu_interval = (u64) (appldata_interval*1000 /
203 					  num_online_cpus()) * TOD_MICRO;
204 		for_each_online_cpu(i) {
205 			per_cpu(appldata_timer, i).expires = per_cpu_interval;
206 			smp_call_function_single(i, add_virt_timer_periodic,
207 						 &per_cpu(appldata_timer, i),
208 						 1);
209 		}
210 		appldata_timer_active = 1;
211 		break;
212 	case APPLDATA_DEL_TIMER:
213 		for_each_online_cpu(i)
214 			del_virt_timer(&per_cpu(appldata_timer, i));
215 		if (!appldata_timer_active)
216 			break;
217 		appldata_timer_active = 0;
218 		atomic_set(&appldata_expire_count, num_online_cpus());
219 		break;
220 	case APPLDATA_MOD_TIMER:
221 		per_cpu_interval = (u64) (appldata_interval*1000 /
222 					  num_online_cpus()) * TOD_MICRO;
223 		if (!appldata_timer_active)
224 			break;
225 		for_each_online_cpu(i) {
226 			struct {
227 				struct vtimer_list *timer;
228 				u64    expires;
229 			} args;
230 			args.timer = &per_cpu(appldata_timer, i);
231 			args.expires = per_cpu_interval;
232 			smp_call_function_single(i, __appldata_mod_vtimer_wrap,
233 						 &args, 1);
234 		}
235 	}
236 }
237 
238 /*
239  * appldata_timer_handler()
240  *
241  * Start/Stop timer, show status of timer (0 = not active, 1 = active)
242  */
243 static int
244 appldata_timer_handler(ctl_table *ctl, int write, struct file *filp,
245 			   void __user *buffer, size_t *lenp, loff_t *ppos)
246 {
247 	int len;
248 	char buf[2];
249 
250 	if (!*lenp || *ppos) {
251 		*lenp = 0;
252 		return 0;
253 	}
254 	if (!write) {
255 		len = sprintf(buf, appldata_timer_active ? "1\n" : "0\n");
256 		if (len > *lenp)
257 			len = *lenp;
258 		if (copy_to_user(buffer, buf, len))
259 			return -EFAULT;
260 		goto out;
261 	}
262 	len = *lenp;
263 	if (copy_from_user(buf, buffer, len > sizeof(buf) ? sizeof(buf) : len))
264 		return -EFAULT;
265 	get_online_cpus();
266 	spin_lock(&appldata_timer_lock);
267 	if (buf[0] == '1')
268 		__appldata_vtimer_setup(APPLDATA_ADD_TIMER);
269 	else if (buf[0] == '0')
270 		__appldata_vtimer_setup(APPLDATA_DEL_TIMER);
271 	spin_unlock(&appldata_timer_lock);
272 	put_online_cpus();
273 out:
274 	*lenp = len;
275 	*ppos += len;
276 	return 0;
277 }
278 
279 /*
280  * appldata_interval_handler()
281  *
282  * Set (CPU) timer interval for collection of data (in milliseconds), show
283  * current timer interval.
284  */
285 static int
286 appldata_interval_handler(ctl_table *ctl, int write, struct file *filp,
287 			   void __user *buffer, size_t *lenp, loff_t *ppos)
288 {
289 	int len, interval;
290 	char buf[16];
291 
292 	if (!*lenp || *ppos) {
293 		*lenp = 0;
294 		return 0;
295 	}
296 	if (!write) {
297 		len = sprintf(buf, "%i\n", appldata_interval);
298 		if (len > *lenp)
299 			len = *lenp;
300 		if (copy_to_user(buffer, buf, len))
301 			return -EFAULT;
302 		goto out;
303 	}
304 	len = *lenp;
305 	if (copy_from_user(buf, buffer, len > sizeof(buf) ? sizeof(buf) : len)) {
306 		return -EFAULT;
307 	}
308 	interval = 0;
309 	sscanf(buf, "%i", &interval);
310 	if (interval <= 0)
311 		return -EINVAL;
312 
313 	get_online_cpus();
314 	spin_lock(&appldata_timer_lock);
315 	appldata_interval = interval;
316 	__appldata_vtimer_setup(APPLDATA_MOD_TIMER);
317 	spin_unlock(&appldata_timer_lock);
318 	put_online_cpus();
319 out:
320 	*lenp = len;
321 	*ppos += len;
322 	return 0;
323 }
324 
325 /*
326  * appldata_generic_handler()
327  *
328  * Generic start/stop monitoring and DIAG, show status of
329  * monitoring (0 = not in process, 1 = in process)
330  */
331 static int
332 appldata_generic_handler(ctl_table *ctl, int write, struct file *filp,
333 			   void __user *buffer, size_t *lenp, loff_t *ppos)
334 {
335 	struct appldata_ops *ops = NULL, *tmp_ops;
336 	int rc, len, found;
337 	char buf[2];
338 	struct list_head *lh;
339 
340 	found = 0;
341 	mutex_lock(&appldata_ops_mutex);
342 	list_for_each(lh, &appldata_ops_list) {
343 		tmp_ops = list_entry(lh, struct appldata_ops, list);
344 		if (&tmp_ops->ctl_table[2] == ctl) {
345 			found = 1;
346 		}
347 	}
348 	if (!found) {
349 		mutex_unlock(&appldata_ops_mutex);
350 		return -ENODEV;
351 	}
352 	ops = ctl->data;
353 	if (!try_module_get(ops->owner)) {	// protect this function
354 		mutex_unlock(&appldata_ops_mutex);
355 		return -ENODEV;
356 	}
357 	mutex_unlock(&appldata_ops_mutex);
358 
359 	if (!*lenp || *ppos) {
360 		*lenp = 0;
361 		module_put(ops->owner);
362 		return 0;
363 	}
364 	if (!write) {
365 		len = sprintf(buf, ops->active ? "1\n" : "0\n");
366 		if (len > *lenp)
367 			len = *lenp;
368 		if (copy_to_user(buffer, buf, len)) {
369 			module_put(ops->owner);
370 			return -EFAULT;
371 		}
372 		goto out;
373 	}
374 	len = *lenp;
375 	if (copy_from_user(buf, buffer,
376 			   len > sizeof(buf) ? sizeof(buf) : len)) {
377 		module_put(ops->owner);
378 		return -EFAULT;
379 	}
380 
381 	mutex_lock(&appldata_ops_mutex);
382 	if ((buf[0] == '1') && (ops->active == 0)) {
383 		// protect work queue callback
384 		if (!try_module_get(ops->owner)) {
385 			mutex_unlock(&appldata_ops_mutex);
386 			module_put(ops->owner);
387 			return -ENODEV;
388 		}
389 		ops->callback(ops->data);	// init record
390 		rc = appldata_diag(ops->record_nr,
391 					APPLDATA_START_INTERVAL_REC,
392 					(unsigned long) ops->data, ops->size,
393 					ops->mod_lvl);
394 		if (rc != 0) {
395 			pr_err("Starting the data collection for %s "
396 			       "failed with rc=%d\n", ops->name, rc);
397 			module_put(ops->owner);
398 		} else
399 			ops->active = 1;
400 	} else if ((buf[0] == '0') && (ops->active == 1)) {
401 		ops->active = 0;
402 		rc = appldata_diag(ops->record_nr, APPLDATA_STOP_REC,
403 				(unsigned long) ops->data, ops->size,
404 				ops->mod_lvl);
405 		if (rc != 0)
406 			pr_err("Stopping the data collection for %s "
407 			       "failed with rc=%d\n", ops->name, rc);
408 		module_put(ops->owner);
409 	}
410 	mutex_unlock(&appldata_ops_mutex);
411 out:
412 	*lenp = len;
413 	*ppos += len;
414 	module_put(ops->owner);
415 	return 0;
416 }
417 
418 /*************************** /proc stuff <END> *******************************/
419 
420 
421 /************************* module-ops management *****************************/
422 /*
423  * appldata_register_ops()
424  *
425  * update ops list, register /proc/sys entries
426  */
427 int appldata_register_ops(struct appldata_ops *ops)
428 {
429 	if (ops->size > APPLDATA_MAX_REC_SIZE)
430 		return -EINVAL;
431 
432 	ops->ctl_table = kzalloc(4 * sizeof(struct ctl_table), GFP_KERNEL);
433 	if (!ops->ctl_table)
434 		return -ENOMEM;
435 
436 	mutex_lock(&appldata_ops_mutex);
437 	list_add(&ops->list, &appldata_ops_list);
438 	mutex_unlock(&appldata_ops_mutex);
439 
440 	ops->ctl_table[0].procname = appldata_proc_name;
441 	ops->ctl_table[0].maxlen   = 0;
442 	ops->ctl_table[0].mode     = S_IRUGO | S_IXUGO;
443 	ops->ctl_table[0].child    = &ops->ctl_table[2];
444 
445 	ops->ctl_table[2].procname = ops->name;
446 	ops->ctl_table[2].mode     = S_IRUGO | S_IWUSR;
447 	ops->ctl_table[2].proc_handler = appldata_generic_handler;
448 	ops->ctl_table[2].data = ops;
449 
450 	ops->sysctl_header = register_sysctl_table(ops->ctl_table);
451 	if (!ops->sysctl_header)
452 		goto out;
453 	return 0;
454 out:
455 	mutex_lock(&appldata_ops_mutex);
456 	list_del(&ops->list);
457 	mutex_unlock(&appldata_ops_mutex);
458 	kfree(ops->ctl_table);
459 	return -ENOMEM;
460 }
461 
462 /*
463  * appldata_unregister_ops()
464  *
465  * update ops list, unregister /proc entries, stop DIAG if necessary
466  */
467 void appldata_unregister_ops(struct appldata_ops *ops)
468 {
469 	mutex_lock(&appldata_ops_mutex);
470 	list_del(&ops->list);
471 	mutex_unlock(&appldata_ops_mutex);
472 	unregister_sysctl_table(ops->sysctl_header);
473 	kfree(ops->ctl_table);
474 }
475 /********************** module-ops management <END> **************************/
476 
477 
478 /******************************* init / exit *********************************/
479 
480 static void __cpuinit appldata_online_cpu(int cpu)
481 {
482 	init_virt_timer(&per_cpu(appldata_timer, cpu));
483 	per_cpu(appldata_timer, cpu).function = appldata_timer_function;
484 	per_cpu(appldata_timer, cpu).data = (unsigned long)
485 		&appldata_work;
486 	atomic_inc(&appldata_expire_count);
487 	spin_lock(&appldata_timer_lock);
488 	__appldata_vtimer_setup(APPLDATA_MOD_TIMER);
489 	spin_unlock(&appldata_timer_lock);
490 }
491 
492 static void __cpuinit appldata_offline_cpu(int cpu)
493 {
494 	del_virt_timer(&per_cpu(appldata_timer, cpu));
495 	if (atomic_dec_and_test(&appldata_expire_count)) {
496 		atomic_set(&appldata_expire_count, num_online_cpus());
497 		queue_work(appldata_wq, &appldata_work);
498 	}
499 	spin_lock(&appldata_timer_lock);
500 	__appldata_vtimer_setup(APPLDATA_MOD_TIMER);
501 	spin_unlock(&appldata_timer_lock);
502 }
503 
504 static int __cpuinit appldata_cpu_notify(struct notifier_block *self,
505 					 unsigned long action,
506 					 void *hcpu)
507 {
508 	switch (action) {
509 	case CPU_ONLINE:
510 	case CPU_ONLINE_FROZEN:
511 		appldata_online_cpu((long) hcpu);
512 		break;
513 	case CPU_DEAD:
514 	case CPU_DEAD_FROZEN:
515 		appldata_offline_cpu((long) hcpu);
516 		break;
517 	default:
518 		break;
519 	}
520 	return NOTIFY_OK;
521 }
522 
523 static struct notifier_block __cpuinitdata appldata_nb = {
524 	.notifier_call = appldata_cpu_notify,
525 };
526 
527 /*
528  * appldata_init()
529  *
530  * init timer, register /proc entries
531  */
532 static int __init appldata_init(void)
533 {
534 	int i;
535 
536 	appldata_wq = create_singlethread_workqueue("appldata");
537 	if (!appldata_wq)
538 		return -ENOMEM;
539 
540 	get_online_cpus();
541 	for_each_online_cpu(i)
542 		appldata_online_cpu(i);
543 	put_online_cpus();
544 
545 	/* Register cpu hotplug notifier */
546 	register_hotcpu_notifier(&appldata_nb);
547 
548 	appldata_sysctl_header = register_sysctl_table(appldata_dir_table);
549 	return 0;
550 }
551 
552 __initcall(appldata_init);
553 
554 /**************************** init / exit <END> ******************************/
555 
556 EXPORT_SYMBOL_GPL(appldata_register_ops);
557 EXPORT_SYMBOL_GPL(appldata_unregister_ops);
558 EXPORT_SYMBOL_GPL(appldata_diag);
559 
560 #ifdef CONFIG_SWAP
561 EXPORT_SYMBOL_GPL(si_swapinfo);
562 #endif
563 EXPORT_SYMBOL_GPL(nr_threads);
564 EXPORT_SYMBOL_GPL(nr_running);
565 EXPORT_SYMBOL_GPL(nr_iowait);
566