xref: /openbmc/linux/arch/s390/kernel/vtime.c (revision f42b3800)
1 /*
2  *  arch/s390/kernel/vtime.c
3  *    Virtual cpu timer based timer functions.
4  *
5  *  S390 version
6  *    Copyright (C) 2004 IBM Deutschland Entwicklung GmbH, IBM Corporation
7  *    Author(s): Jan Glauber <jan.glauber@de.ibm.com>
8  */
9 
10 #include <linux/module.h>
11 #include <linux/kernel.h>
12 #include <linux/time.h>
13 #include <linux/delay.h>
14 #include <linux/init.h>
15 #include <linux/smp.h>
16 #include <linux/types.h>
17 #include <linux/timex.h>
18 #include <linux/notifier.h>
19 #include <linux/kernel_stat.h>
20 #include <linux/rcupdate.h>
21 #include <linux/posix-timers.h>
22 
23 #include <asm/s390_ext.h>
24 #include <asm/timer.h>
25 #include <asm/irq_regs.h>
26 
27 static ext_int_info_t ext_int_info_timer;
28 static DEFINE_PER_CPU(struct vtimer_queue, virt_cpu_timer);
29 
30 #ifdef CONFIG_VIRT_CPU_ACCOUNTING
31 /*
32  * Update process times based on virtual cpu times stored by entry.S
33  * to the lowcore fields user_timer, system_timer & steal_clock.
34  */
35 void account_process_tick(struct task_struct *tsk, int user_tick)
36 {
37 	cputime_t cputime;
38 	__u64 timer, clock;
39 	int rcu_user_flag;
40 
41 	timer = S390_lowcore.last_update_timer;
42 	clock = S390_lowcore.last_update_clock;
43 	asm volatile ("  STPT %0\n"    /* Store current cpu timer value */
44 		      "  STCK %1"      /* Store current tod clock value */
45 		      : "=m" (S390_lowcore.last_update_timer),
46 		        "=m" (S390_lowcore.last_update_clock) );
47 	S390_lowcore.system_timer += timer - S390_lowcore.last_update_timer;
48 	S390_lowcore.steal_clock += S390_lowcore.last_update_clock - clock;
49 
50 	cputime = S390_lowcore.user_timer >> 12;
51 	rcu_user_flag = cputime != 0;
52 	S390_lowcore.user_timer -= cputime << 12;
53 	S390_lowcore.steal_clock -= cputime << 12;
54 	account_user_time(tsk, cputime);
55 
56 	cputime =  S390_lowcore.system_timer >> 12;
57 	S390_lowcore.system_timer -= cputime << 12;
58 	S390_lowcore.steal_clock -= cputime << 12;
59 	account_system_time(tsk, HARDIRQ_OFFSET, cputime);
60 
61 	cputime = S390_lowcore.steal_clock;
62 	if ((__s64) cputime > 0) {
63 		cputime >>= 12;
64 		S390_lowcore.steal_clock -= cputime << 12;
65 		account_steal_time(tsk, cputime);
66 	}
67 }
68 
69 /*
70  * Update process times based on virtual cpu times stored by entry.S
71  * to the lowcore fields user_timer, system_timer & steal_clock.
72  */
73 void account_vtime(struct task_struct *tsk)
74 {
75 	cputime_t cputime;
76 	__u64 timer;
77 
78 	timer = S390_lowcore.last_update_timer;
79 	asm volatile ("  STPT %0"    /* Store current cpu timer value */
80 		      : "=m" (S390_lowcore.last_update_timer) );
81 	S390_lowcore.system_timer += timer - S390_lowcore.last_update_timer;
82 
83 	cputime = S390_lowcore.user_timer >> 12;
84 	S390_lowcore.user_timer -= cputime << 12;
85 	S390_lowcore.steal_clock -= cputime << 12;
86 	account_user_time(tsk, cputime);
87 
88 	cputime =  S390_lowcore.system_timer >> 12;
89 	S390_lowcore.system_timer -= cputime << 12;
90 	S390_lowcore.steal_clock -= cputime << 12;
91 	account_system_time(tsk, 0, cputime);
92 }
93 
94 /*
95  * Update process times based on virtual cpu times stored by entry.S
96  * to the lowcore fields user_timer, system_timer & steal_clock.
97  */
98 void account_system_vtime(struct task_struct *tsk)
99 {
100 	cputime_t cputime;
101 	__u64 timer;
102 
103 	timer = S390_lowcore.last_update_timer;
104 	asm volatile ("  STPT %0"    /* Store current cpu timer value */
105 		      : "=m" (S390_lowcore.last_update_timer) );
106 	S390_lowcore.system_timer += timer - S390_lowcore.last_update_timer;
107 
108 	cputime =  S390_lowcore.system_timer >> 12;
109 	S390_lowcore.system_timer -= cputime << 12;
110 	S390_lowcore.steal_clock -= cputime << 12;
111 	account_system_time(tsk, 0, cputime);
112 }
113 
114 static inline void set_vtimer(__u64 expires)
115 {
116 	__u64 timer;
117 
118 	asm volatile ("  STPT %0\n"  /* Store current cpu timer value */
119 		      "  SPT %1"     /* Set new value immediatly afterwards */
120 		      : "=m" (timer) : "m" (expires) );
121 	S390_lowcore.system_timer += S390_lowcore.last_update_timer - timer;
122 	S390_lowcore.last_update_timer = expires;
123 
124 	/* store expire time for this CPU timer */
125 	__get_cpu_var(virt_cpu_timer).to_expire = expires;
126 }
127 #else
128 static inline void set_vtimer(__u64 expires)
129 {
130 	S390_lowcore.last_update_timer = expires;
131 	asm volatile ("SPT %0" : : "m" (S390_lowcore.last_update_timer));
132 
133 	/* store expire time for this CPU timer */
134 	__get_cpu_var(virt_cpu_timer).to_expire = expires;
135 }
136 #endif
137 
138 static void start_cpu_timer(void)
139 {
140 	struct vtimer_queue *vt_list;
141 
142 	vt_list = &__get_cpu_var(virt_cpu_timer);
143 
144 	/* CPU timer interrupt is pending, don't reprogramm it */
145 	if (vt_list->idle & 1LL<<63)
146 		return;
147 
148 	if (!list_empty(&vt_list->list))
149 		set_vtimer(vt_list->idle);
150 }
151 
152 static void stop_cpu_timer(void)
153 {
154 	struct vtimer_queue *vt_list;
155 
156 	vt_list = &__get_cpu_var(virt_cpu_timer);
157 
158 	/* nothing to do */
159 	if (list_empty(&vt_list->list)) {
160 		vt_list->idle = VTIMER_MAX_SLICE;
161 		goto fire;
162 	}
163 
164 	/* store the actual expire value */
165 	asm volatile ("STPT %0" : "=m" (vt_list->idle));
166 
167 	/*
168 	 * If the CPU timer is negative we don't reprogramm
169 	 * it because we will get instantly an interrupt.
170 	 */
171 	if (vt_list->idle & 1LL<<63)
172 		return;
173 
174 	vt_list->offset += vt_list->to_expire - vt_list->idle;
175 
176 	/*
177 	 * We cannot halt the CPU timer, we just write a value that
178 	 * nearly never expires (only after 71 years) and re-write
179 	 * the stored expire value if we continue the timer
180 	 */
181  fire:
182 	set_vtimer(VTIMER_MAX_SLICE);
183 }
184 
185 /*
186  * Sorted add to a list. List is linear searched until first bigger
187  * element is found.
188  */
189 static void list_add_sorted(struct vtimer_list *timer, struct list_head *head)
190 {
191 	struct vtimer_list *event;
192 
193 	list_for_each_entry(event, head, entry) {
194 		if (event->expires > timer->expires) {
195 			list_add_tail(&timer->entry, &event->entry);
196 			return;
197 		}
198 	}
199 	list_add_tail(&timer->entry, head);
200 }
201 
202 /*
203  * Do the callback functions of expired vtimer events.
204  * Called from within the interrupt handler.
205  */
206 static void do_callbacks(struct list_head *cb_list)
207 {
208 	struct vtimer_queue *vt_list;
209 	struct vtimer_list *event, *tmp;
210 	void (*fn)(unsigned long);
211 	unsigned long data;
212 
213 	if (list_empty(cb_list))
214 		return;
215 
216 	vt_list = &__get_cpu_var(virt_cpu_timer);
217 
218 	list_for_each_entry_safe(event, tmp, cb_list, entry) {
219 		fn = event->function;
220 		data = event->data;
221 		fn(data);
222 
223 		if (!event->interval)
224 			/* delete one shot timer */
225 			list_del_init(&event->entry);
226 		else {
227 			/* move interval timer back to list */
228 			spin_lock(&vt_list->lock);
229 			list_del_init(&event->entry);
230 			list_add_sorted(event, &vt_list->list);
231 			spin_unlock(&vt_list->lock);
232 		}
233 	}
234 }
235 
236 /*
237  * Handler for the virtual CPU timer.
238  */
239 static void do_cpu_timer_interrupt(__u16 error_code)
240 {
241 	__u64 next, delta;
242 	struct vtimer_queue *vt_list;
243 	struct vtimer_list *event, *tmp;
244 	struct list_head *ptr;
245 	/* the callback queue */
246 	struct list_head cb_list;
247 
248 	INIT_LIST_HEAD(&cb_list);
249 	vt_list = &__get_cpu_var(virt_cpu_timer);
250 
251 	/* walk timer list, fire all expired events */
252 	spin_lock(&vt_list->lock);
253 
254 	if (vt_list->to_expire < VTIMER_MAX_SLICE)
255 		vt_list->offset += vt_list->to_expire;
256 
257 	list_for_each_entry_safe(event, tmp, &vt_list->list, entry) {
258 		if (event->expires > vt_list->offset)
259 			/* found first unexpired event, leave */
260 			break;
261 
262 		/* re-charge interval timer, we have to add the offset */
263 		if (event->interval)
264 			event->expires = event->interval + vt_list->offset;
265 
266 		/* move expired timer to the callback queue */
267 		list_move_tail(&event->entry, &cb_list);
268 	}
269 	spin_unlock(&vt_list->lock);
270 	do_callbacks(&cb_list);
271 
272 	/* next event is first in list */
273 	spin_lock(&vt_list->lock);
274 	if (!list_empty(&vt_list->list)) {
275 		ptr = vt_list->list.next;
276 		event = list_entry(ptr, struct vtimer_list, entry);
277 		next = event->expires - vt_list->offset;
278 
279 		/* add the expired time from this interrupt handler
280 		 * and the callback functions
281 		 */
282 		asm volatile ("STPT %0" : "=m" (delta));
283 		delta = 0xffffffffffffffffLL - delta + 1;
284 		vt_list->offset += delta;
285 		next -= delta;
286 	} else {
287 		vt_list->offset = 0;
288 		next = VTIMER_MAX_SLICE;
289 	}
290 	spin_unlock(&vt_list->lock);
291 	set_vtimer(next);
292 }
293 
294 void init_virt_timer(struct vtimer_list *timer)
295 {
296 	timer->function = NULL;
297 	INIT_LIST_HEAD(&timer->entry);
298 	spin_lock_init(&timer->lock);
299 }
300 EXPORT_SYMBOL(init_virt_timer);
301 
302 static inline int vtimer_pending(struct vtimer_list *timer)
303 {
304 	return (!list_empty(&timer->entry));
305 }
306 
307 /*
308  * this function should only run on the specified CPU
309  */
310 static void internal_add_vtimer(struct vtimer_list *timer)
311 {
312 	unsigned long flags;
313 	__u64 done;
314 	struct vtimer_list *event;
315 	struct vtimer_queue *vt_list;
316 
317 	vt_list = &per_cpu(virt_cpu_timer, timer->cpu);
318 	spin_lock_irqsave(&vt_list->lock, flags);
319 
320 	if (timer->cpu != smp_processor_id())
321 		printk("internal_add_vtimer: BUG, running on wrong CPU");
322 
323 	/* if list is empty we only have to set the timer */
324 	if (list_empty(&vt_list->list)) {
325 		/* reset the offset, this may happen if the last timer was
326 		 * just deleted by mod_virt_timer and the interrupt
327 		 * didn't happen until here
328 		 */
329 		vt_list->offset = 0;
330 		goto fire;
331 	}
332 
333 	/* save progress */
334 	asm volatile ("STPT %0" : "=m" (done));
335 
336 	/* calculate completed work */
337 	done = vt_list->to_expire - done + vt_list->offset;
338 	vt_list->offset = 0;
339 
340 	list_for_each_entry(event, &vt_list->list, entry)
341 		event->expires -= done;
342 
343  fire:
344 	list_add_sorted(timer, &vt_list->list);
345 
346 	/* get first element, which is the next vtimer slice */
347 	event = list_entry(vt_list->list.next, struct vtimer_list, entry);
348 
349 	set_vtimer(event->expires);
350 	spin_unlock_irqrestore(&vt_list->lock, flags);
351 	/* release CPU acquired in prepare_vtimer or mod_virt_timer() */
352 	put_cpu();
353 }
354 
355 static inline int prepare_vtimer(struct vtimer_list *timer)
356 {
357 	if (!timer->function) {
358 		printk("add_virt_timer: uninitialized timer\n");
359 		return -EINVAL;
360 	}
361 
362 	if (!timer->expires || timer->expires > VTIMER_MAX_SLICE) {
363 		printk("add_virt_timer: invalid timer expire value!\n");
364 		return -EINVAL;
365 	}
366 
367 	if (vtimer_pending(timer)) {
368 		printk("add_virt_timer: timer pending\n");
369 		return -EBUSY;
370 	}
371 
372 	timer->cpu = get_cpu();
373 	return 0;
374 }
375 
376 /*
377  * add_virt_timer - add an oneshot virtual CPU timer
378  */
379 void add_virt_timer(void *new)
380 {
381 	struct vtimer_list *timer;
382 
383 	timer = (struct vtimer_list *)new;
384 
385 	if (prepare_vtimer(timer) < 0)
386 		return;
387 
388 	timer->interval = 0;
389 	internal_add_vtimer(timer);
390 }
391 EXPORT_SYMBOL(add_virt_timer);
392 
393 /*
394  * add_virt_timer_int - add an interval virtual CPU timer
395  */
396 void add_virt_timer_periodic(void *new)
397 {
398 	struct vtimer_list *timer;
399 
400 	timer = (struct vtimer_list *)new;
401 
402 	if (prepare_vtimer(timer) < 0)
403 		return;
404 
405 	timer->interval = timer->expires;
406 	internal_add_vtimer(timer);
407 }
408 EXPORT_SYMBOL(add_virt_timer_periodic);
409 
410 /*
411  * If we change a pending timer the function must be called on the CPU
412  * where the timer is running on, e.g. by smp_call_function_single()
413  *
414  * The original mod_timer adds the timer if it is not pending. For compatibility
415  * we do the same. The timer will be added on the current CPU as a oneshot timer.
416  *
417  * returns whether it has modified a pending timer (1) or not (0)
418  */
419 int mod_virt_timer(struct vtimer_list *timer, __u64 expires)
420 {
421 	struct vtimer_queue *vt_list;
422 	unsigned long flags;
423 	int cpu;
424 
425 	if (!timer->function) {
426 		printk("mod_virt_timer: uninitialized timer\n");
427 		return	-EINVAL;
428 	}
429 
430 	if (!expires || expires > VTIMER_MAX_SLICE) {
431 		printk("mod_virt_timer: invalid expire range\n");
432 		return -EINVAL;
433 	}
434 
435 	/*
436 	 * This is a common optimization triggered by the
437 	 * networking code - if the timer is re-modified
438 	 * to be the same thing then just return:
439 	 */
440 	if (timer->expires == expires && vtimer_pending(timer))
441 		return 1;
442 
443 	cpu = get_cpu();
444 	vt_list = &per_cpu(virt_cpu_timer, cpu);
445 
446 	/* disable interrupts before test if timer is pending */
447 	spin_lock_irqsave(&vt_list->lock, flags);
448 
449 	/* if timer isn't pending add it on the current CPU */
450 	if (!vtimer_pending(timer)) {
451 		spin_unlock_irqrestore(&vt_list->lock, flags);
452 		/* we do not activate an interval timer with mod_virt_timer */
453 		timer->interval = 0;
454 		timer->expires = expires;
455 		timer->cpu = cpu;
456 		internal_add_vtimer(timer);
457 		return 0;
458 	}
459 
460 	/* check if we run on the right CPU */
461 	if (timer->cpu != cpu) {
462 		printk("mod_virt_timer: running on wrong CPU, check your code\n");
463 		spin_unlock_irqrestore(&vt_list->lock, flags);
464 		put_cpu();
465 		return -EINVAL;
466 	}
467 
468 	list_del_init(&timer->entry);
469 	timer->expires = expires;
470 
471 	/* also change the interval if we have an interval timer */
472 	if (timer->interval)
473 		timer->interval = expires;
474 
475 	/* the timer can't expire anymore so we can release the lock */
476 	spin_unlock_irqrestore(&vt_list->lock, flags);
477 	internal_add_vtimer(timer);
478 	return 1;
479 }
480 EXPORT_SYMBOL(mod_virt_timer);
481 
482 /*
483  * delete a virtual timer
484  *
485  * returns whether the deleted timer was pending (1) or not (0)
486  */
487 int del_virt_timer(struct vtimer_list *timer)
488 {
489 	unsigned long flags;
490 	struct vtimer_queue *vt_list;
491 
492 	/* check if timer is pending */
493 	if (!vtimer_pending(timer))
494 		return 0;
495 
496 	vt_list = &per_cpu(virt_cpu_timer, timer->cpu);
497 	spin_lock_irqsave(&vt_list->lock, flags);
498 
499 	/* we don't interrupt a running timer, just let it expire! */
500 	list_del_init(&timer->entry);
501 
502 	/* last timer removed */
503 	if (list_empty(&vt_list->list)) {
504 		vt_list->to_expire = 0;
505 		vt_list->offset = 0;
506 	}
507 
508 	spin_unlock_irqrestore(&vt_list->lock, flags);
509 	return 1;
510 }
511 EXPORT_SYMBOL(del_virt_timer);
512 
513 /*
514  * Start the virtual CPU timer on the current CPU.
515  */
516 void init_cpu_vtimer(void)
517 {
518 	struct vtimer_queue *vt_list;
519 
520 	/* kick the virtual timer */
521 	S390_lowcore.exit_timer = VTIMER_MAX_SLICE;
522 	S390_lowcore.last_update_timer = VTIMER_MAX_SLICE;
523 	asm volatile ("SPT %0" : : "m" (S390_lowcore.last_update_timer));
524 	asm volatile ("STCK %0" : "=m" (S390_lowcore.last_update_clock));
525 
526 	/* enable cpu timer interrupts */
527 	__ctl_set_bit(0,10);
528 
529 	vt_list = &__get_cpu_var(virt_cpu_timer);
530 	INIT_LIST_HEAD(&vt_list->list);
531 	spin_lock_init(&vt_list->lock);
532 	vt_list->to_expire = 0;
533 	vt_list->offset = 0;
534 	vt_list->idle = 0;
535 
536 }
537 
538 static int vtimer_idle_notify(struct notifier_block *self,
539 			      unsigned long action, void *hcpu)
540 {
541 	switch (action) {
542 	case S390_CPU_IDLE:
543 		stop_cpu_timer();
544 		break;
545 	case S390_CPU_NOT_IDLE:
546 		start_cpu_timer();
547 		break;
548 	}
549 	return NOTIFY_OK;
550 }
551 
552 static struct notifier_block vtimer_idle_nb = {
553 	.notifier_call = vtimer_idle_notify,
554 };
555 
556 void __init vtime_init(void)
557 {
558 	/* request the cpu timer external interrupt */
559 	if (register_early_external_interrupt(0x1005, do_cpu_timer_interrupt,
560 					      &ext_int_info_timer) != 0)
561 		panic("Couldn't request external interrupt 0x1005");
562 
563 	if (register_idle_notifier(&vtimer_idle_nb))
564 		panic("Couldn't register idle notifier");
565 
566 	/* Enable cpu timer interrupts on the boot cpu. */
567 	init_cpu_vtimer();
568 }
569 
570