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#include <linux/cpu.h>
23#include <linux/kprobes.h>
24
25#include <asm/timer.h>
26#include <asm/irq_regs.h>
27#include <asm/cputime.h>
28#include <asm/irq.h>
29#include "entry.h"
30
31static DEFINE_PER_CPU(struct vtimer_queue, virt_cpu_timer);
32
33DEFINE_PER_CPU(struct s390_idle_data, s390_idle);
34
35static inline __u64 get_vtimer(void)
36{
37	__u64 timer;
38
39	asm volatile("STPT %0" : "=m" (timer));
40	return timer;
41}
42
43static inline void set_vtimer(__u64 expires)
44{
45	__u64 timer;
46
47	asm volatile ("  STPT %0\n"  /* Store current cpu timer value */
48		      "  SPT %1"     /* Set new value immediately afterwards */
49		      : "=m" (timer) : "m" (expires) );
50	S390_lowcore.system_timer += S390_lowcore.last_update_timer - timer;
51	S390_lowcore.last_update_timer = expires;
52}
53
54/*
55 * Update process times based on virtual cpu times stored by entry.S
56 * to the lowcore fields user_timer, system_timer & steal_clock.
57 */
58static void do_account_vtime(struct task_struct *tsk, int hardirq_offset)
59{
60	struct thread_info *ti = task_thread_info(tsk);
61	__u64 timer, clock, user, system, steal;
62
63	timer = S390_lowcore.last_update_timer;
64	clock = S390_lowcore.last_update_clock;
65	asm volatile ("  STPT %0\n"    /* Store current cpu timer value */
66		      "  STCK %1"      /* Store current tod clock value */
67		      : "=m" (S390_lowcore.last_update_timer),
68		        "=m" (S390_lowcore.last_update_clock) );
69	S390_lowcore.system_timer += timer - S390_lowcore.last_update_timer;
70	S390_lowcore.steal_timer += S390_lowcore.last_update_clock - clock;
71
72	user = S390_lowcore.user_timer - ti->user_timer;
73	S390_lowcore.steal_timer -= user;
74	ti->user_timer = S390_lowcore.user_timer;
75	account_user_time(tsk, user, user);
76
77	system = S390_lowcore.system_timer - ti->system_timer;
78	S390_lowcore.steal_timer -= system;
79	ti->system_timer = S390_lowcore.system_timer;
80	account_system_time(tsk, hardirq_offset, system, system);
81
82	steal = S390_lowcore.steal_timer;
83	if ((s64) steal > 0) {
84		S390_lowcore.steal_timer = 0;
85		account_steal_time(steal);
86	}
87}
88
89void account_vtime(struct task_struct *prev, struct task_struct *next)
90{
91	struct thread_info *ti;
92
93	do_account_vtime(prev, 0);
94	ti = task_thread_info(prev);
95	ti->user_timer = S390_lowcore.user_timer;
96	ti->system_timer = S390_lowcore.system_timer;
97	ti = task_thread_info(next);
98	S390_lowcore.user_timer = ti->user_timer;
99	S390_lowcore.system_timer = ti->system_timer;
100}
101
102void account_process_tick(struct task_struct *tsk, int user_tick)
103{
104	do_account_vtime(tsk, HARDIRQ_OFFSET);
105}
106
107/*
108 * Update process times based on virtual cpu times stored by entry.S
109 * to the lowcore fields user_timer, system_timer & steal_clock.
110 */
111void account_system_vtime(struct task_struct *tsk)
112{
113	struct thread_info *ti = task_thread_info(tsk);
114	__u64 timer, system;
115
116	timer = S390_lowcore.last_update_timer;
117	S390_lowcore.last_update_timer = get_vtimer();
118	S390_lowcore.system_timer += timer - S390_lowcore.last_update_timer;
119
120	system = S390_lowcore.system_timer - ti->system_timer;
121	S390_lowcore.steal_timer -= system;
122	ti->system_timer = S390_lowcore.system_timer;
123	account_system_time(tsk, 0, system, system);
124}
125EXPORT_SYMBOL_GPL(account_system_vtime);
126
127void __kprobes vtime_stop_cpu(void)
128{
129	struct s390_idle_data *idle = &__get_cpu_var(s390_idle);
130	struct vtimer_queue *vq = &__get_cpu_var(virt_cpu_timer);
131	unsigned long long idle_time;
132	unsigned long psw_mask;
133
134	trace_hardirqs_on();
135	/* Don't trace preempt off for idle. */
136	stop_critical_timings();
137
138	/* Wait for external, I/O or machine check interrupt. */
139	psw_mask = psw_kernel_bits | PSW_MASK_WAIT | PSW_MASK_DAT |
140		PSW_MASK_IO | PSW_MASK_EXT | PSW_MASK_MCHECK;
141	idle->nohz_delay = 0;
142
143	/* Call the assembler magic in entry.S */
144	psw_idle(idle, vq, psw_mask, !list_empty(&vq->list));
145
146	/* Reenable preemption tracer. */
147	start_critical_timings();
148
149	/* Account time spent with enabled wait psw loaded as idle time. */
150	idle->sequence++;
151	smp_wmb();
152	idle_time = idle->idle_exit - idle->idle_enter;
153	idle->idle_time += idle_time;
154	idle->idle_enter = idle->idle_exit = 0ULL;
155	idle->idle_count++;
156	account_idle_time(idle_time);
157	smp_wmb();
158	idle->sequence++;
159}
160
161cputime64_t s390_get_idle_time(int cpu)
162{
163	struct s390_idle_data *idle = &per_cpu(s390_idle, cpu);
164	unsigned long long now, idle_enter, idle_exit;
165	unsigned int sequence;
166
167	do {
168		now = get_clock();
169		sequence = ACCESS_ONCE(idle->sequence);
170		idle_enter = ACCESS_ONCE(idle->idle_enter);
171		idle_exit = ACCESS_ONCE(idle->idle_exit);
172	} while ((sequence & 1) || (idle->sequence != sequence));
173	return idle_enter ? ((idle_exit ? : now) - idle_enter) : 0;
174}
175
176/*
177 * Sorted add to a list. List is linear searched until first bigger
178 * element is found.
179 */
180static void list_add_sorted(struct vtimer_list *timer, struct list_head *head)
181{
182	struct vtimer_list *event;
183
184	list_for_each_entry(event, head, entry) {
185		if (event->expires > timer->expires) {
186			list_add_tail(&timer->entry, &event->entry);
187			return;
188		}
189	}
190	list_add_tail(&timer->entry, head);
191}
192
193/*
194 * Do the callback functions of expired vtimer events.
195 * Called from within the interrupt handler.
196 */
197static void do_callbacks(struct list_head *cb_list)
198{
199	struct vtimer_queue *vq;
200	struct vtimer_list *event, *tmp;
201
202	if (list_empty(cb_list))
203		return;
204
205	vq = &__get_cpu_var(virt_cpu_timer);
206
207	list_for_each_entry_safe(event, tmp, cb_list, entry) {
208		list_del_init(&event->entry);
209		(event->function)(event->data);
210		if (event->interval) {
211			/* Recharge interval timer */
212			event->expires = event->interval + vq->elapsed;
213			spin_lock(&vq->lock);
214			list_add_sorted(event, &vq->list);
215			spin_unlock(&vq->lock);
216		}
217	}
218}
219
220/*
221 * Handler for the virtual CPU timer.
222 */
223static void do_cpu_timer_interrupt(struct ext_code ext_code,
224				   unsigned int param32, unsigned long param64)
225{
226	struct vtimer_queue *vq;
227	struct vtimer_list *event, *tmp;
228	struct list_head cb_list;	/* the callback queue */
229	__u64 elapsed, next;
230
231	kstat_cpu(smp_processor_id()).irqs[EXTINT_TMR]++;
232	INIT_LIST_HEAD(&cb_list);
233	vq = &__get_cpu_var(virt_cpu_timer);
234
235	/* walk timer list, fire all expired events */
236	spin_lock(&vq->lock);
237
238	elapsed = vq->elapsed + (vq->timer - S390_lowcore.async_enter_timer);
239	BUG_ON((s64) elapsed < 0);
240	vq->elapsed = 0;
241	list_for_each_entry_safe(event, tmp, &vq->list, entry) {
242		if (event->expires < elapsed)
243			/* move expired timer to the callback queue */
244			list_move_tail(&event->entry, &cb_list);
245		else
246			event->expires -= elapsed;
247	}
248	spin_unlock(&vq->lock);
249
250	do_callbacks(&cb_list);
251
252	/* next event is first in list */
253	next = VTIMER_MAX_SLICE;
254	spin_lock(&vq->lock);
255	if (!list_empty(&vq->list)) {
256		event = list_first_entry(&vq->list, struct vtimer_list, entry);
257		next = event->expires;
258	}
259	spin_unlock(&vq->lock);
260	/*
261	 * To improve precision add the time spent by the
262	 * interrupt handler to the elapsed time.
263	 * Note: CPU timer counts down and we got an interrupt,
264	 *	 the current content is negative
265	 */
266	elapsed = S390_lowcore.async_enter_timer - get_vtimer();
267	set_vtimer(next - elapsed);
268	vq->timer = next - elapsed;
269	vq->elapsed = elapsed;
270}
271
272void init_virt_timer(struct vtimer_list *timer)
273{
274	timer->function = NULL;
275	INIT_LIST_HEAD(&timer->entry);
276}
277EXPORT_SYMBOL(init_virt_timer);
278
279static inline int vtimer_pending(struct vtimer_list *timer)
280{
281	return (!list_empty(&timer->entry));
282}
283
284/*
285 * this function should only run on the specified CPU
286 */
287static void internal_add_vtimer(struct vtimer_list *timer)
288{
289	struct vtimer_queue *vq;
290	unsigned long flags;
291	__u64 left, expires;
292
293	vq = &per_cpu(virt_cpu_timer, timer->cpu);
294	spin_lock_irqsave(&vq->lock, flags);
295
296	BUG_ON(timer->cpu != smp_processor_id());
297
298	if (list_empty(&vq->list)) {
299		/* First timer on this cpu, just program it. */
300		list_add(&timer->entry, &vq->list);
301		set_vtimer(timer->expires);
302		vq->timer = timer->expires;
303		vq->elapsed = 0;
304	} else {
305		/* Check progress of old timers. */
306		expires = timer->expires;
307		left = get_vtimer();
308		if (likely((s64) expires < (s64) left)) {
309			/* The new timer expires before the current timer. */
310			set_vtimer(expires);
311			vq->elapsed += vq->timer - left;
312			vq->timer = expires;
313		} else {
314			vq->elapsed += vq->timer - left;
315			vq->timer = left;
316		}
317		/* Insert new timer into per cpu list. */
318		timer->expires += vq->elapsed;
319		list_add_sorted(timer, &vq->list);
320	}
321
322	spin_unlock_irqrestore(&vq->lock, flags);
323	/* release CPU acquired in prepare_vtimer or mod_virt_timer() */
324	put_cpu();
325}
326
327static inline void prepare_vtimer(struct vtimer_list *timer)
328{
329	BUG_ON(!timer->function);
330	BUG_ON(!timer->expires || timer->expires > VTIMER_MAX_SLICE);
331	BUG_ON(vtimer_pending(timer));
332	timer->cpu = get_cpu();
333}
334
335/*
336 * add_virt_timer - add an oneshot virtual CPU timer
337 */
338void add_virt_timer(void *new)
339{
340	struct vtimer_list *timer;
341
342	timer = (struct vtimer_list *)new;
343	prepare_vtimer(timer);
344	timer->interval = 0;
345	internal_add_vtimer(timer);
346}
347EXPORT_SYMBOL(add_virt_timer);
348
349/*
350 * add_virt_timer_int - add an interval virtual CPU timer
351 */
352void add_virt_timer_periodic(void *new)
353{
354	struct vtimer_list *timer;
355
356	timer = (struct vtimer_list *)new;
357	prepare_vtimer(timer);
358	timer->interval = timer->expires;
359	internal_add_vtimer(timer);
360}
361EXPORT_SYMBOL(add_virt_timer_periodic);
362
363static int __mod_vtimer(struct vtimer_list *timer, __u64 expires, int periodic)
364{
365	struct vtimer_queue *vq;
366	unsigned long flags;
367	int cpu;
368
369	BUG_ON(!timer->function);
370	BUG_ON(!expires || expires > VTIMER_MAX_SLICE);
371
372	if (timer->expires == expires && vtimer_pending(timer))
373		return 1;
374
375	cpu = get_cpu();
376	vq = &per_cpu(virt_cpu_timer, cpu);
377
378	/* disable interrupts before test if timer is pending */
379	spin_lock_irqsave(&vq->lock, flags);
380
381	/* if timer isn't pending add it on the current CPU */
382	if (!vtimer_pending(timer)) {
383		spin_unlock_irqrestore(&vq->lock, flags);
384
385		if (periodic)
386			timer->interval = expires;
387		else
388			timer->interval = 0;
389		timer->expires = expires;
390		timer->cpu = cpu;
391		internal_add_vtimer(timer);
392		return 0;
393	}
394
395	/* check if we run on the right CPU */
396	BUG_ON(timer->cpu != cpu);
397
398	list_del_init(&timer->entry);
399	timer->expires = expires;
400	if (periodic)
401		timer->interval = expires;
402
403	/* the timer can't expire anymore so we can release the lock */
404	spin_unlock_irqrestore(&vq->lock, flags);
405	internal_add_vtimer(timer);
406	return 1;
407}
408
409/*
410 * If we change a pending timer the function must be called on the CPU
411 * where the timer is running on.
412 *
413 * returns whether it has modified a pending timer (1) or not (0)
414 */
415int mod_virt_timer(struct vtimer_list *timer, __u64 expires)
416{
417	return __mod_vtimer(timer, expires, 0);
418}
419EXPORT_SYMBOL(mod_virt_timer);
420
421/*
422 * If we change a pending timer the function must be called on the CPU
423 * where the timer is running on.
424 *
425 * returns whether it has modified a pending timer (1) or not (0)
426 */
427int mod_virt_timer_periodic(struct vtimer_list *timer, __u64 expires)
428{
429	return __mod_vtimer(timer, expires, 1);
430}
431EXPORT_SYMBOL(mod_virt_timer_periodic);
432
433/*
434 * delete a virtual timer
435 *
436 * returns whether the deleted timer was pending (1) or not (0)
437 */
438int del_virt_timer(struct vtimer_list *timer)
439{
440	unsigned long flags;
441	struct vtimer_queue *vq;
442
443	/* check if timer is pending */
444	if (!vtimer_pending(timer))
445		return 0;
446
447	vq = &per_cpu(virt_cpu_timer, timer->cpu);
448	spin_lock_irqsave(&vq->lock, flags);
449
450	/* we don't interrupt a running timer, just let it expire! */
451	list_del_init(&timer->entry);
452
453	spin_unlock_irqrestore(&vq->lock, flags);
454	return 1;
455}
456EXPORT_SYMBOL(del_virt_timer);
457
458/*
459 * Start the virtual CPU timer on the current CPU.
460 */
461void init_cpu_vtimer(void)
462{
463	struct vtimer_queue *vq;
464
465	/* initialize per cpu vtimer structure */
466	vq = &__get_cpu_var(virt_cpu_timer);
467	INIT_LIST_HEAD(&vq->list);
468	spin_lock_init(&vq->lock);
469
470	/* enable cpu timer interrupts */
471	__ctl_set_bit(0,10);
472
473	/* set initial cpu timer */
474	set_vtimer(0x7fffffffffffffffULL);
475}
476
477static int __cpuinit s390_nohz_notify(struct notifier_block *self,
478				      unsigned long action, void *hcpu)
479{
480	struct s390_idle_data *idle;
481	long cpu = (long) hcpu;
482
483	idle = &per_cpu(s390_idle, cpu);
484	switch (action) {
485	case CPU_DYING:
486	case CPU_DYING_FROZEN:
487		idle->nohz_delay = 0;
488	default:
489		break;
490	}
491	return NOTIFY_OK;
492}
493
494void __init vtime_init(void)
495{
496	/* request the cpu timer external interrupt */
497	if (register_external_interrupt(0x1005, do_cpu_timer_interrupt))
498		panic("Couldn't request external interrupt 0x1005");
499
500	/* Enable cpu timer interrupts on the boot cpu. */
501	init_cpu_vtimer();
502	cpu_notifier(s390_nohz_notify, 0);
503}
504
505