Linux Audio

Check our new training course

Loading...
v6.8
  1// SPDX-License-Identifier: GPL-2.0
  2/*
  3 *    Virtual cpu timer based timer functions.
  4 *
  5 *    Copyright IBM Corp. 2004, 2012
  6 *    Author(s): Jan Glauber <jan.glauber@de.ibm.com>
  7 */
  8
  9#include <linux/kernel_stat.h>
 
 
 10#include <linux/export.h>
 11#include <linux/kernel.h>
 12#include <linux/timex.h>
 13#include <linux/types.h>
 14#include <linux/time.h>
 15#include <asm/alternative.h>
 
 
 
 16#include <asm/cputime.h>
 17#include <asm/vtimer.h>
 18#include <asm/vtime.h>
 19#include <asm/cpu_mf.h>
 20#include <asm/smp.h>
 21
 22#include "entry.h"
 23
 24static void virt_timer_expire(void);
 25
 
 
 26static LIST_HEAD(virt_timer_list);
 27static DEFINE_SPINLOCK(virt_timer_lock);
 28static atomic64_t virt_timer_current;
 29static atomic64_t virt_timer_elapsed;
 30
 31DEFINE_PER_CPU(u64, mt_cycles[8]);
 32static DEFINE_PER_CPU(u64, mt_scaling_mult) = { 1 };
 33static DEFINE_PER_CPU(u64, mt_scaling_div) = { 1 };
 34static DEFINE_PER_CPU(u64, mt_scaling_jiffies);
 35
 36static inline u64 get_vtimer(void)
 37{
 38	u64 timer;
 39
 40	asm volatile("stpt %0" : "=Q" (timer));
 41	return timer;
 42}
 43
 44static inline void set_vtimer(u64 expires)
 45{
 46	u64 timer;
 47
 48	asm volatile(
 49		"	stpt	%0\n"	/* Store current cpu timer value */
 50		"	spt	%1"	/* Set new value imm. afterwards */
 51		: "=Q" (timer) : "Q" (expires));
 52	S390_lowcore.system_timer += S390_lowcore.last_update_timer - timer;
 53	S390_lowcore.last_update_timer = expires;
 54}
 55
 56static inline int virt_timer_forward(u64 elapsed)
 57{
 58	BUG_ON(!irqs_disabled());
 59
 60	if (list_empty(&virt_timer_list))
 61		return 0;
 62	elapsed = atomic64_add_return(elapsed, &virt_timer_elapsed);
 63	return elapsed >= atomic64_read(&virt_timer_current);
 64}
 65
 66static void update_mt_scaling(void)
 67{
 68	u64 cycles_new[8], *cycles_old;
 69	u64 delta, fac, mult, div;
 70	int i;
 71
 72	stcctm(MT_DIAG, smp_cpu_mtid + 1, cycles_new);
 73	cycles_old = this_cpu_ptr(mt_cycles);
 74	fac = 1;
 75	mult = div = 0;
 76	for (i = 0; i <= smp_cpu_mtid; i++) {
 77		delta = cycles_new[i] - cycles_old[i];
 78		div += delta;
 79		mult *= i + 1;
 80		mult += delta * fac;
 81		fac *= i + 1;
 82	}
 83	div *= fac;
 84	if (div > 0) {
 85		/* Update scaling factor */
 86		__this_cpu_write(mt_scaling_mult, mult);
 87		__this_cpu_write(mt_scaling_div, div);
 88		memcpy(cycles_old, cycles_new,
 89		       sizeof(u64) * (smp_cpu_mtid + 1));
 90	}
 91	__this_cpu_write(mt_scaling_jiffies, jiffies_64);
 92}
 93
 94static inline u64 update_tsk_timer(unsigned long *tsk_vtime, u64 new)
 95{
 96	u64 delta;
 97
 98	delta = new - *tsk_vtime;
 99	*tsk_vtime = new;
100	return delta;
101}
102
103
104static inline u64 scale_vtime(u64 vtime)
105{
106	u64 mult = __this_cpu_read(mt_scaling_mult);
107	u64 div = __this_cpu_read(mt_scaling_div);
108
109	if (smp_cpu_mtid)
110		return vtime * mult / div;
111	return vtime;
112}
113
114static void account_system_index_scaled(struct task_struct *p, u64 cputime,
115					enum cpu_usage_stat index)
116{
117	p->stimescaled += cputime_to_nsecs(scale_vtime(cputime));
118	account_system_index_time(p, cputime_to_nsecs(cputime), index);
119}
120
121/*
122 * Update process times based on virtual cpu times stored by entry.S
123 * to the lowcore fields user_timer, system_timer & steal_clock.
124 */
125static int do_account_vtime(struct task_struct *tsk)
126{
127	u64 timer, clock, user, guest, system, hardirq, softirq;
 
128
129	timer = S390_lowcore.last_update_timer;
130	clock = S390_lowcore.last_update_clock;
131	asm volatile(
132		"	stpt	%0\n"	/* Store current cpu timer value */
133		"	stckf	%1"	/* Store current tod clock value */
134		: "=Q" (S390_lowcore.last_update_timer),
135		  "=Q" (S390_lowcore.last_update_clock)
136		: : "cc");
137	clock = S390_lowcore.last_update_clock - clock;
138	timer -= S390_lowcore.last_update_timer;
139
140	if (hardirq_count())
141		S390_lowcore.hardirq_timer += timer;
142	else
143		S390_lowcore.system_timer += timer;
144
145	/* Update MT utilization calculation */
146	if (smp_cpu_mtid &&
147	    time_after64(jiffies_64, this_cpu_read(mt_scaling_jiffies)))
148		update_mt_scaling();
149
150	/* Calculate cputime delta */
151	user = update_tsk_timer(&tsk->thread.user_timer,
152				READ_ONCE(S390_lowcore.user_timer));
153	guest = update_tsk_timer(&tsk->thread.guest_timer,
154				 READ_ONCE(S390_lowcore.guest_timer));
155	system = update_tsk_timer(&tsk->thread.system_timer,
156				  READ_ONCE(S390_lowcore.system_timer));
157	hardirq = update_tsk_timer(&tsk->thread.hardirq_timer,
158				   READ_ONCE(S390_lowcore.hardirq_timer));
159	softirq = update_tsk_timer(&tsk->thread.softirq_timer,
160				   READ_ONCE(S390_lowcore.softirq_timer));
161	S390_lowcore.steal_timer +=
162		clock - user - guest - system - hardirq - softirq;
163
164	/* Push account value */
165	if (user) {
166		account_user_time(tsk, cputime_to_nsecs(user));
167		tsk->utimescaled += cputime_to_nsecs(scale_vtime(user));
168	}
169
170	if (guest) {
171		account_guest_time(tsk, cputime_to_nsecs(guest));
172		tsk->utimescaled += cputime_to_nsecs(scale_vtime(guest));
 
173	}
174
175	if (system)
176		account_system_index_scaled(tsk, system, CPUTIME_SYSTEM);
177	if (hardirq)
178		account_system_index_scaled(tsk, hardirq, CPUTIME_IRQ);
179	if (softirq)
180		account_system_index_scaled(tsk, softirq, CPUTIME_SOFTIRQ);
181
182	return virt_timer_forward(user + guest + system + hardirq + softirq);
183}
184
185void vtime_task_switch(struct task_struct *prev)
186{
187	do_account_vtime(prev);
188	prev->thread.user_timer = S390_lowcore.user_timer;
189	prev->thread.guest_timer = S390_lowcore.guest_timer;
190	prev->thread.system_timer = S390_lowcore.system_timer;
191	prev->thread.hardirq_timer = S390_lowcore.hardirq_timer;
192	prev->thread.softirq_timer = S390_lowcore.softirq_timer;
193	S390_lowcore.user_timer = current->thread.user_timer;
194	S390_lowcore.guest_timer = current->thread.guest_timer;
195	S390_lowcore.system_timer = current->thread.system_timer;
196	S390_lowcore.hardirq_timer = current->thread.hardirq_timer;
197	S390_lowcore.softirq_timer = current->thread.softirq_timer;
198}
199
200/*
201 * In s390, accounting pending user time also implies
202 * accounting system time in order to correctly compute
203 * the stolen time accounting.
204 */
205void vtime_flush(struct task_struct *tsk)
206{
207	u64 steal, avg_steal;
208
209	if (do_account_vtime(tsk))
210		virt_timer_expire();
211
212	steal = S390_lowcore.steal_timer;
213	avg_steal = S390_lowcore.avg_steal_timer / 2;
214	if ((s64) steal > 0) {
215		S390_lowcore.steal_timer = 0;
216		account_steal_time(cputime_to_nsecs(steal));
217		avg_steal += steal;
218	}
219	S390_lowcore.avg_steal_timer = avg_steal;
220}
221
222static u64 vtime_delta(void)
223{
224	u64 timer = S390_lowcore.last_update_timer;
225
226	S390_lowcore.last_update_timer = get_vtimer();
227
228	return timer - S390_lowcore.last_update_timer;
229}
230
231/*
232 * Update process times based on virtual cpu times stored by entry.S
233 * to the lowcore fields user_timer, system_timer & steal_clock.
234 */
235void vtime_account_kernel(struct task_struct *tsk)
236{
237	u64 delta = vtime_delta();
238
239	if (tsk->flags & PF_VCPU)
240		S390_lowcore.guest_timer += delta;
241	else
242		S390_lowcore.system_timer += delta;
243
244	virt_timer_forward(delta);
245}
246EXPORT_SYMBOL_GPL(vtime_account_kernel);
247
248void vtime_account_softirq(struct task_struct *tsk)
249{
250	u64 delta = vtime_delta();
251
252	S390_lowcore.softirq_timer += delta;
253
254	virt_timer_forward(delta);
255}
256
257void vtime_account_hardirq(struct task_struct *tsk)
258{
259	u64 delta = vtime_delta();
260
261	S390_lowcore.hardirq_timer += delta;
262
263	virt_timer_forward(delta);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
264}
265
266/*
267 * Sorted add to a list. List is linear searched until first bigger
268 * element is found.
269 */
270static void list_add_sorted(struct vtimer_list *timer, struct list_head *head)
271{
272	struct vtimer_list *tmp;
273
274	list_for_each_entry(tmp, head, entry) {
275		if (tmp->expires > timer->expires) {
276			list_add_tail(&timer->entry, &tmp->entry);
277			return;
278		}
279	}
280	list_add_tail(&timer->entry, head);
281}
282
283/*
284 * Handler for expired virtual CPU timer.
285 */
286static void virt_timer_expire(void)
287{
288	struct vtimer_list *timer, *tmp;
289	unsigned long elapsed;
290	LIST_HEAD(cb_list);
291
292	/* walk timer list, fire all expired timers */
293	spin_lock(&virt_timer_lock);
294	elapsed = atomic64_read(&virt_timer_elapsed);
295	list_for_each_entry_safe(timer, tmp, &virt_timer_list, entry) {
296		if (timer->expires < elapsed)
297			/* move expired timer to the callback queue */
298			list_move_tail(&timer->entry, &cb_list);
299		else
300			timer->expires -= elapsed;
301	}
302	if (!list_empty(&virt_timer_list)) {
303		timer = list_first_entry(&virt_timer_list,
304					 struct vtimer_list, entry);
305		atomic64_set(&virt_timer_current, timer->expires);
306	}
307	atomic64_sub(elapsed, &virt_timer_elapsed);
308	spin_unlock(&virt_timer_lock);
309
310	/* Do callbacks and recharge periodic timers */
311	list_for_each_entry_safe(timer, tmp, &cb_list, entry) {
312		list_del_init(&timer->entry);
313		timer->function(timer->data);
314		if (timer->interval) {
315			/* Recharge interval timer */
316			timer->expires = timer->interval +
317				atomic64_read(&virt_timer_elapsed);
318			spin_lock(&virt_timer_lock);
319			list_add_sorted(timer, &virt_timer_list);
320			spin_unlock(&virt_timer_lock);
321		}
322	}
323}
324
325void init_virt_timer(struct vtimer_list *timer)
326{
327	timer->function = NULL;
328	INIT_LIST_HEAD(&timer->entry);
329}
330EXPORT_SYMBOL(init_virt_timer);
331
332static inline int vtimer_pending(struct vtimer_list *timer)
333{
334	return !list_empty(&timer->entry);
335}
336
337static void internal_add_vtimer(struct vtimer_list *timer)
338{
339	if (list_empty(&virt_timer_list)) {
340		/* First timer, just program it. */
341		atomic64_set(&virt_timer_current, timer->expires);
342		atomic64_set(&virt_timer_elapsed, 0);
343		list_add(&timer->entry, &virt_timer_list);
344	} else {
345		/* Update timer against current base. */
346		timer->expires += atomic64_read(&virt_timer_elapsed);
347		if (likely((s64) timer->expires <
348			   (s64) atomic64_read(&virt_timer_current)))
349			/* The new timer expires before the current timer. */
350			atomic64_set(&virt_timer_current, timer->expires);
351		/* Insert new timer into the list. */
352		list_add_sorted(timer, &virt_timer_list);
353	}
354}
355
356static void __add_vtimer(struct vtimer_list *timer, int periodic)
357{
358	unsigned long flags;
359
360	timer->interval = periodic ? timer->expires : 0;
361	spin_lock_irqsave(&virt_timer_lock, flags);
362	internal_add_vtimer(timer);
363	spin_unlock_irqrestore(&virt_timer_lock, flags);
364}
365
366/*
367 * add_virt_timer - add a oneshot virtual CPU timer
368 */
369void add_virt_timer(struct vtimer_list *timer)
370{
371	__add_vtimer(timer, 0);
372}
373EXPORT_SYMBOL(add_virt_timer);
374
375/*
376 * add_virt_timer_int - add an interval virtual CPU timer
377 */
378void add_virt_timer_periodic(struct vtimer_list *timer)
379{
380	__add_vtimer(timer, 1);
381}
382EXPORT_SYMBOL(add_virt_timer_periodic);
383
384static int __mod_vtimer(struct vtimer_list *timer, u64 expires, int periodic)
385{
386	unsigned long flags;
387	int rc;
388
389	BUG_ON(!timer->function);
390
391	if (timer->expires == expires && vtimer_pending(timer))
392		return 1;
393	spin_lock_irqsave(&virt_timer_lock, flags);
394	rc = vtimer_pending(timer);
395	if (rc)
396		list_del_init(&timer->entry);
397	timer->interval = periodic ? expires : 0;
398	timer->expires = expires;
399	internal_add_vtimer(timer);
400	spin_unlock_irqrestore(&virt_timer_lock, flags);
401	return rc;
402}
403
404/*
405 * returns whether it has modified a pending timer (1) or not (0)
406 */
407int mod_virt_timer(struct vtimer_list *timer, u64 expires)
408{
409	return __mod_vtimer(timer, expires, 0);
410}
411EXPORT_SYMBOL(mod_virt_timer);
412
413/*
414 * returns whether it has modified a pending timer (1) or not (0)
415 */
416int mod_virt_timer_periodic(struct vtimer_list *timer, u64 expires)
417{
418	return __mod_vtimer(timer, expires, 1);
419}
420EXPORT_SYMBOL(mod_virt_timer_periodic);
421
422/*
423 * Delete a virtual timer.
424 *
425 * returns whether the deleted timer was pending (1) or not (0)
426 */
427int del_virt_timer(struct vtimer_list *timer)
428{
429	unsigned long flags;
430
431	if (!vtimer_pending(timer))
432		return 0;
433	spin_lock_irqsave(&virt_timer_lock, flags);
434	list_del_init(&timer->entry);
435	spin_unlock_irqrestore(&virt_timer_lock, flags);
436	return 1;
437}
438EXPORT_SYMBOL(del_virt_timer);
439
440/*
441 * Start the virtual CPU timer on the current CPU.
442 */
443void vtime_init(void)
444{
445	/* set initial cpu timer */
446	set_vtimer(VTIMER_MAX_SLICE);
447	/* Setup initial MT scaling values */
448	if (smp_cpu_mtid) {
449		__this_cpu_write(mt_scaling_jiffies, jiffies);
450		__this_cpu_write(mt_scaling_mult, 1);
451		__this_cpu_write(mt_scaling_div, 1);
452		stcctm(MT_DIAG, smp_cpu_mtid + 1, this_cpu_ptr(mt_cycles));
 
 
 
 
 
 
 
 
453	}
 
 
 
 
 
 
 
 
454}
v3.15
 
  1/*
  2 *    Virtual cpu timer based timer functions.
  3 *
  4 *    Copyright IBM Corp. 2004, 2012
  5 *    Author(s): Jan Glauber <jan.glauber@de.ibm.com>
  6 */
  7
  8#include <linux/kernel_stat.h>
  9#include <linux/notifier.h>
 10#include <linux/kprobes.h>
 11#include <linux/export.h>
 12#include <linux/kernel.h>
 13#include <linux/timex.h>
 14#include <linux/types.h>
 15#include <linux/time.h>
 16#include <linux/cpu.h>
 17#include <linux/smp.h>
 18
 19#include <asm/irq_regs.h>
 20#include <asm/cputime.h>
 21#include <asm/vtimer.h>
 22#include <asm/vtime.h>
 23#include <asm/irq.h>
 
 
 24#include "entry.h"
 25
 26static void virt_timer_expire(void);
 27
 28DEFINE_PER_CPU(struct s390_idle_data, s390_idle);
 29
 30static LIST_HEAD(virt_timer_list);
 31static DEFINE_SPINLOCK(virt_timer_lock);
 32static atomic64_t virt_timer_current;
 33static atomic64_t virt_timer_elapsed;
 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(
 48		"	stpt	%0\n"	/* Store current cpu timer value */
 49		"	spt	%1"	/* Set new value imm. afterwards */
 50		: "=m" (timer) : "m" (expires));
 51	S390_lowcore.system_timer += S390_lowcore.last_update_timer - timer;
 52	S390_lowcore.last_update_timer = expires;
 53}
 54
 55static inline int virt_timer_forward(u64 elapsed)
 56{
 57	BUG_ON(!irqs_disabled());
 58
 59	if (list_empty(&virt_timer_list))
 60		return 0;
 61	elapsed = atomic64_add_return(elapsed, &virt_timer_elapsed);
 62	return elapsed >= atomic64_read(&virt_timer_current);
 63}
 64
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 65/*
 66 * Update process times based on virtual cpu times stored by entry.S
 67 * to the lowcore fields user_timer, system_timer & steal_clock.
 68 */
 69static int do_account_vtime(struct task_struct *tsk, int hardirq_offset)
 70{
 71	struct thread_info *ti = task_thread_info(tsk);
 72	u64 timer, clock, user, system, steal;
 73
 74	timer = S390_lowcore.last_update_timer;
 75	clock = S390_lowcore.last_update_clock;
 76	asm volatile(
 77		"	stpt	%0\n"	/* Store current cpu timer value */
 78		"	stck	%1"	/* Store current tod clock value */
 79		: "=m" (S390_lowcore.last_update_timer),
 80		  "=m" (S390_lowcore.last_update_clock));
 81	S390_lowcore.system_timer += timer - S390_lowcore.last_update_timer;
 82	S390_lowcore.steal_timer += S390_lowcore.last_update_clock - clock;
 83
 84	user = S390_lowcore.user_timer - ti->user_timer;
 85	S390_lowcore.steal_timer -= user;
 86	ti->user_timer = S390_lowcore.user_timer;
 87	account_user_time(tsk, user, user);
 88
 89	system = S390_lowcore.system_timer - ti->system_timer;
 90	S390_lowcore.steal_timer -= system;
 91	ti->system_timer = S390_lowcore.system_timer;
 92	account_system_time(tsk, hardirq_offset, system, system);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 93
 94	steal = S390_lowcore.steal_timer;
 95	if ((s64) steal > 0) {
 96		S390_lowcore.steal_timer = 0;
 97		account_steal_time(steal);
 98	}
 99
100	return virt_timer_forward(user + system);
 
 
 
 
 
 
 
101}
102
103void vtime_task_switch(struct task_struct *prev)
104{
105	struct thread_info *ti;
106
107	do_account_vtime(prev, 0);
108	ti = task_thread_info(prev);
109	ti->user_timer = S390_lowcore.user_timer;
110	ti->system_timer = S390_lowcore.system_timer;
111	ti = task_thread_info(current);
112	S390_lowcore.user_timer = ti->user_timer;
113	S390_lowcore.system_timer = ti->system_timer;
 
 
114}
115
116/*
117 * In s390, accounting pending user time also implies
118 * accounting system time in order to correctly compute
119 * the stolen time accounting.
120 */
121void vtime_account_user(struct task_struct *tsk)
122{
123	if (do_account_vtime(tsk, HARDIRQ_OFFSET))
 
 
124		virt_timer_expire();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
125}
126
127/*
128 * Update process times based on virtual cpu times stored by entry.S
129 * to the lowcore fields user_timer, system_timer & steal_clock.
130 */
131void vtime_account_irq_enter(struct task_struct *tsk)
 
 
 
 
 
 
 
 
 
 
 
 
 
132{
133	struct thread_info *ti = task_thread_info(tsk);
134	u64 timer, system;
 
135
136	WARN_ON_ONCE(!irqs_disabled());
 
137
138	timer = S390_lowcore.last_update_timer;
139	S390_lowcore.last_update_timer = get_vtimer();
140	S390_lowcore.system_timer += timer - S390_lowcore.last_update_timer;
 
 
141
142	system = S390_lowcore.system_timer - ti->system_timer;
143	S390_lowcore.steal_timer -= system;
144	ti->system_timer = S390_lowcore.system_timer;
145	account_system_time(tsk, 0, system, system);
146
147	virt_timer_forward(system);
148}
149EXPORT_SYMBOL_GPL(vtime_account_irq_enter);
150
151void vtime_account_system(struct task_struct *tsk)
152__attribute__((alias("vtime_account_irq_enter")));
153EXPORT_SYMBOL_GPL(vtime_account_system);
154
155void __kprobes vtime_stop_cpu(void)
156{
157	struct s390_idle_data *idle = &__get_cpu_var(s390_idle);
158	unsigned long long idle_time;
159	unsigned long psw_mask;
160
161	trace_hardirqs_on();
162
163	/* Wait for external, I/O or machine check interrupt. */
164	psw_mask = PSW_KERNEL_BITS | PSW_MASK_WAIT | PSW_MASK_DAT |
165		PSW_MASK_IO | PSW_MASK_EXT | PSW_MASK_MCHECK;
166	idle->nohz_delay = 0;
167
168	/* Call the assembler magic in entry.S */
169	psw_idle(idle, psw_mask);
170
171	/* Account time spent with enabled wait psw loaded as idle time. */
172	idle->sequence++;
173	smp_wmb();
174	idle_time = idle->clock_idle_exit - idle->clock_idle_enter;
175	idle->clock_idle_enter = idle->clock_idle_exit = 0ULL;
176	idle->idle_time += idle_time;
177	idle->idle_count++;
178	account_idle_time(idle_time);
179	smp_wmb();
180	idle->sequence++;
181}
182
183cputime64_t s390_get_idle_time(int cpu)
184{
185	struct s390_idle_data *idle = &per_cpu(s390_idle, cpu);
186	unsigned long long now, idle_enter, idle_exit;
187	unsigned int sequence;
188
189	do {
190		now = get_tod_clock();
191		sequence = ACCESS_ONCE(idle->sequence);
192		idle_enter = ACCESS_ONCE(idle->clock_idle_enter);
193		idle_exit = ACCESS_ONCE(idle->clock_idle_exit);
194	} while ((sequence & 1) || (ACCESS_ONCE(idle->sequence) != sequence));
195	return idle_enter ? ((idle_exit ?: now) - idle_enter) : 0;
196}
197
198/*
199 * Sorted add to a list. List is linear searched until first bigger
200 * element is found.
201 */
202static void list_add_sorted(struct vtimer_list *timer, struct list_head *head)
203{
204	struct vtimer_list *tmp;
205
206	list_for_each_entry(tmp, head, entry) {
207		if (tmp->expires > timer->expires) {
208			list_add_tail(&timer->entry, &tmp->entry);
209			return;
210		}
211	}
212	list_add_tail(&timer->entry, head);
213}
214
215/*
216 * Handler for expired virtual CPU timer.
217 */
218static void virt_timer_expire(void)
219{
220	struct vtimer_list *timer, *tmp;
221	unsigned long elapsed;
222	LIST_HEAD(cb_list);
223
224	/* walk timer list, fire all expired timers */
225	spin_lock(&virt_timer_lock);
226	elapsed = atomic64_read(&virt_timer_elapsed);
227	list_for_each_entry_safe(timer, tmp, &virt_timer_list, entry) {
228		if (timer->expires < elapsed)
229			/* move expired timer to the callback queue */
230			list_move_tail(&timer->entry, &cb_list);
231		else
232			timer->expires -= elapsed;
233	}
234	if (!list_empty(&virt_timer_list)) {
235		timer = list_first_entry(&virt_timer_list,
236					 struct vtimer_list, entry);
237		atomic64_set(&virt_timer_current, timer->expires);
238	}
239	atomic64_sub(elapsed, &virt_timer_elapsed);
240	spin_unlock(&virt_timer_lock);
241
242	/* Do callbacks and recharge periodic timers */
243	list_for_each_entry_safe(timer, tmp, &cb_list, entry) {
244		list_del_init(&timer->entry);
245		timer->function(timer->data);
246		if (timer->interval) {
247			/* Recharge interval timer */
248			timer->expires = timer->interval +
249				atomic64_read(&virt_timer_elapsed);
250			spin_lock(&virt_timer_lock);
251			list_add_sorted(timer, &virt_timer_list);
252			spin_unlock(&virt_timer_lock);
253		}
254	}
255}
256
257void init_virt_timer(struct vtimer_list *timer)
258{
259	timer->function = NULL;
260	INIT_LIST_HEAD(&timer->entry);
261}
262EXPORT_SYMBOL(init_virt_timer);
263
264static inline int vtimer_pending(struct vtimer_list *timer)
265{
266	return !list_empty(&timer->entry);
267}
268
269static void internal_add_vtimer(struct vtimer_list *timer)
270{
271	if (list_empty(&virt_timer_list)) {
272		/* First timer, just program it. */
273		atomic64_set(&virt_timer_current, timer->expires);
274		atomic64_set(&virt_timer_elapsed, 0);
275		list_add(&timer->entry, &virt_timer_list);
276	} else {
277		/* Update timer against current base. */
278		timer->expires += atomic64_read(&virt_timer_elapsed);
279		if (likely((s64) timer->expires <
280			   (s64) atomic64_read(&virt_timer_current)))
281			/* The new timer expires before the current timer. */
282			atomic64_set(&virt_timer_current, timer->expires);
283		/* Insert new timer into the list. */
284		list_add_sorted(timer, &virt_timer_list);
285	}
286}
287
288static void __add_vtimer(struct vtimer_list *timer, int periodic)
289{
290	unsigned long flags;
291
292	timer->interval = periodic ? timer->expires : 0;
293	spin_lock_irqsave(&virt_timer_lock, flags);
294	internal_add_vtimer(timer);
295	spin_unlock_irqrestore(&virt_timer_lock, flags);
296}
297
298/*
299 * add_virt_timer - add an oneshot virtual CPU timer
300 */
301void add_virt_timer(struct vtimer_list *timer)
302{
303	__add_vtimer(timer, 0);
304}
305EXPORT_SYMBOL(add_virt_timer);
306
307/*
308 * add_virt_timer_int - add an interval virtual CPU timer
309 */
310void add_virt_timer_periodic(struct vtimer_list *timer)
311{
312	__add_vtimer(timer, 1);
313}
314EXPORT_SYMBOL(add_virt_timer_periodic);
315
316static int __mod_vtimer(struct vtimer_list *timer, u64 expires, int periodic)
317{
318	unsigned long flags;
319	int rc;
320
321	BUG_ON(!timer->function);
322
323	if (timer->expires == expires && vtimer_pending(timer))
324		return 1;
325	spin_lock_irqsave(&virt_timer_lock, flags);
326	rc = vtimer_pending(timer);
327	if (rc)
328		list_del_init(&timer->entry);
329	timer->interval = periodic ? expires : 0;
330	timer->expires = expires;
331	internal_add_vtimer(timer);
332	spin_unlock_irqrestore(&virt_timer_lock, flags);
333	return rc;
334}
335
336/*
337 * returns whether it has modified a pending timer (1) or not (0)
338 */
339int mod_virt_timer(struct vtimer_list *timer, u64 expires)
340{
341	return __mod_vtimer(timer, expires, 0);
342}
343EXPORT_SYMBOL(mod_virt_timer);
344
345/*
346 * returns whether it has modified a pending timer (1) or not (0)
347 */
348int mod_virt_timer_periodic(struct vtimer_list *timer, u64 expires)
349{
350	return __mod_vtimer(timer, expires, 1);
351}
352EXPORT_SYMBOL(mod_virt_timer_periodic);
353
354/*
355 * Delete a virtual timer.
356 *
357 * returns whether the deleted timer was pending (1) or not (0)
358 */
359int del_virt_timer(struct vtimer_list *timer)
360{
361	unsigned long flags;
362
363	if (!vtimer_pending(timer))
364		return 0;
365	spin_lock_irqsave(&virt_timer_lock, flags);
366	list_del_init(&timer->entry);
367	spin_unlock_irqrestore(&virt_timer_lock, flags);
368	return 1;
369}
370EXPORT_SYMBOL(del_virt_timer);
371
372/*
373 * Start the virtual CPU timer on the current CPU.
374 */
375void init_cpu_vtimer(void)
376{
377	/* set initial cpu timer */
378	set_vtimer(VTIMER_MAX_SLICE);
379}
380
381static int s390_nohz_notify(struct notifier_block *self, unsigned long action,
382			    void *hcpu)
383{
384	struct s390_idle_data *idle;
385	long cpu = (long) hcpu;
386
387	idle = &per_cpu(s390_idle, cpu);
388	switch (action & ~CPU_TASKS_FROZEN) {
389	case CPU_DYING:
390		idle->nohz_delay = 0;
391	default:
392		break;
393	}
394	return NOTIFY_OK;
395}
396
397void __init vtime_init(void)
398{
399	/* Enable cpu timer interrupts on the boot cpu. */
400	init_cpu_vtimer();
401	cpu_notifier(s390_nohz_notify, 0);
402}