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}
v4.6
 
  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/export.h>
 10#include <linux/kernel.h>
 11#include <linux/timex.h>
 12#include <linux/types.h>
 13#include <linux/time.h>
 14
 15#include <asm/cputime.h>
 16#include <asm/vtimer.h>
 17#include <asm/vtime.h>
 18#include <asm/cpu_mf.h>
 19#include <asm/smp.h>
 20
 
 
 21static void virt_timer_expire(void);
 22
 23static LIST_HEAD(virt_timer_list);
 24static DEFINE_SPINLOCK(virt_timer_lock);
 25static atomic64_t virt_timer_current;
 26static atomic64_t virt_timer_elapsed;
 27
 28DEFINE_PER_CPU(u64, mt_cycles[8]);
 29static DEFINE_PER_CPU(u64, mt_scaling_mult) = { 1 };
 30static DEFINE_PER_CPU(u64, mt_scaling_div) = { 1 };
 31static DEFINE_PER_CPU(u64, mt_scaling_jiffies);
 32
 33static inline u64 get_vtimer(void)
 34{
 35	u64 timer;
 36
 37	asm volatile("stpt %0" : "=m" (timer));
 38	return timer;
 39}
 40
 41static inline void set_vtimer(u64 expires)
 42{
 43	u64 timer;
 44
 45	asm volatile(
 46		"	stpt	%0\n"	/* Store current cpu timer value */
 47		"	spt	%1"	/* Set new value imm. afterwards */
 48		: "=m" (timer) : "m" (expires));
 49	S390_lowcore.system_timer += S390_lowcore.last_update_timer - timer;
 50	S390_lowcore.last_update_timer = expires;
 51}
 52
 53static inline int virt_timer_forward(u64 elapsed)
 54{
 55	BUG_ON(!irqs_disabled());
 56
 57	if (list_empty(&virt_timer_list))
 58		return 0;
 59	elapsed = atomic64_add_return(elapsed, &virt_timer_elapsed);
 60	return elapsed >= atomic64_read(&virt_timer_current);
 61}
 62
 63static void update_mt_scaling(void)
 64{
 65	u64 cycles_new[8], *cycles_old;
 66	u64 delta, fac, mult, div;
 67	int i;
 68
 69	stcctm5(smp_cpu_mtid + 1, cycles_new);
 70	cycles_old = this_cpu_ptr(mt_cycles);
 71	fac = 1;
 72	mult = div = 0;
 73	for (i = 0; i <= smp_cpu_mtid; i++) {
 74		delta = cycles_new[i] - cycles_old[i];
 75		div += delta;
 76		mult *= i + 1;
 77		mult += delta * fac;
 78		fac *= i + 1;
 79	}
 80	div *= fac;
 81	if (div > 0) {
 82		/* Update scaling factor */
 83		__this_cpu_write(mt_scaling_mult, mult);
 84		__this_cpu_write(mt_scaling_div, div);
 85		memcpy(cycles_old, cycles_new,
 86		       sizeof(u64) * (smp_cpu_mtid + 1));
 87	}
 88	__this_cpu_write(mt_scaling_jiffies, jiffies_64);
 89}
 90
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 91/*
 92 * Update process times based on virtual cpu times stored by entry.S
 93 * to the lowcore fields user_timer, system_timer & steal_clock.
 94 */
 95static int do_account_vtime(struct task_struct *tsk, int hardirq_offset)
 96{
 97	struct thread_info *ti = task_thread_info(tsk);
 98	u64 timer, clock, user, system, steal;
 99	u64 user_scaled, system_scaled;
100
101	timer = S390_lowcore.last_update_timer;
102	clock = S390_lowcore.last_update_clock;
103	asm volatile(
104		"	stpt	%0\n"	/* Store current cpu timer value */
105#ifdef CONFIG_HAVE_MARCH_Z9_109_FEATURES
106		"	stckf	%1"	/* Store current tod clock value */
107#else
108		"	stck	%1"	/* Store current tod clock value */
109#endif
110		: "=m" (S390_lowcore.last_update_timer),
111		  "=m" (S390_lowcore.last_update_clock));
112	S390_lowcore.system_timer += timer - S390_lowcore.last_update_timer;
113	S390_lowcore.steal_timer += S390_lowcore.last_update_clock - clock;
 
 
 
114
115	/* Update MT utilization calculation */
116	if (smp_cpu_mtid &&
117	    time_after64(jiffies_64, this_cpu_read(mt_scaling_jiffies)))
118		update_mt_scaling();
119
120	user = S390_lowcore.user_timer - ti->user_timer;
121	S390_lowcore.steal_timer -= user;
122	ti->user_timer = S390_lowcore.user_timer;
123
124	system = S390_lowcore.system_timer - ti->system_timer;
125	S390_lowcore.steal_timer -= system;
126	ti->system_timer = S390_lowcore.system_timer;
127
128	user_scaled = user;
129	system_scaled = system;
130	/* Do MT utilization scaling */
131	if (smp_cpu_mtid) {
132		u64 mult = __this_cpu_read(mt_scaling_mult);
133		u64 div = __this_cpu_read(mt_scaling_div);
 
 
 
 
 
134
135		user_scaled = (user_scaled * mult) / div;
136		system_scaled = (system_scaled * mult) / div;
 
137	}
138	account_user_time(tsk, user, user_scaled);
139	account_system_time(tsk, hardirq_offset, system, system_scaled);
140
141	steal = S390_lowcore.steal_timer;
142	if ((s64) steal > 0) {
143		S390_lowcore.steal_timer = 0;
144		account_steal_time(steal);
145	}
 
146
147	return virt_timer_forward(user + system);
148}
149
150void vtime_task_switch(struct task_struct *prev)
151{
152	struct thread_info *ti;
153
154	do_account_vtime(prev, 0);
155	ti = task_thread_info(prev);
156	ti->user_timer = S390_lowcore.user_timer;
157	ti->system_timer = S390_lowcore.system_timer;
158	ti = task_thread_info(current);
159	S390_lowcore.user_timer = ti->user_timer;
160	S390_lowcore.system_timer = ti->system_timer;
 
 
161}
162
163/*
164 * In s390, accounting pending user time also implies
165 * accounting system time in order to correctly compute
166 * the stolen time accounting.
167 */
168void vtime_account_user(struct task_struct *tsk)
169{
170	if (do_account_vtime(tsk, HARDIRQ_OFFSET))
 
 
171		virt_timer_expire();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
172}
173
174/*
175 * Update process times based on virtual cpu times stored by entry.S
176 * to the lowcore fields user_timer, system_timer & steal_clock.
177 */
178void vtime_account_irq_enter(struct task_struct *tsk)
179{
180	struct thread_info *ti = task_thread_info(tsk);
181	u64 timer, system, system_scaled;
182
183	timer = S390_lowcore.last_update_timer;
184	S390_lowcore.last_update_timer = get_vtimer();
185	S390_lowcore.system_timer += timer - S390_lowcore.last_update_timer;
 
186
187	/* Update MT utilization calculation */
188	if (smp_cpu_mtid &&
189	    time_after64(jiffies_64, this_cpu_read(mt_scaling_jiffies)))
190		update_mt_scaling();
191
192	system = S390_lowcore.system_timer - ti->system_timer;
193	S390_lowcore.steal_timer -= system;
194	ti->system_timer = S390_lowcore.system_timer;
195	system_scaled = system;
196	/* Do MT utilization scaling */
197	if (smp_cpu_mtid) {
198		u64 mult = __this_cpu_read(mt_scaling_mult);
199		u64 div = __this_cpu_read(mt_scaling_div);
200
201		system_scaled = (system_scaled * mult) / div;
202	}
203	account_system_time(tsk, 0, system, system_scaled);
204
205	virt_timer_forward(system);
206}
207EXPORT_SYMBOL_GPL(vtime_account_irq_enter);
208
209void vtime_account_system(struct task_struct *tsk)
210__attribute__((alias("vtime_account_irq_enter")));
211EXPORT_SYMBOL_GPL(vtime_account_system);
 
 
 
 
 
212
213/*
214 * Sorted add to a list. List is linear searched until first bigger
215 * element is found.
216 */
217static void list_add_sorted(struct vtimer_list *timer, struct list_head *head)
218{
219	struct vtimer_list *tmp;
220
221	list_for_each_entry(tmp, head, entry) {
222		if (tmp->expires > timer->expires) {
223			list_add_tail(&timer->entry, &tmp->entry);
224			return;
225		}
226	}
227	list_add_tail(&timer->entry, head);
228}
229
230/*
231 * Handler for expired virtual CPU timer.
232 */
233static void virt_timer_expire(void)
234{
235	struct vtimer_list *timer, *tmp;
236	unsigned long elapsed;
237	LIST_HEAD(cb_list);
238
239	/* walk timer list, fire all expired timers */
240	spin_lock(&virt_timer_lock);
241	elapsed = atomic64_read(&virt_timer_elapsed);
242	list_for_each_entry_safe(timer, tmp, &virt_timer_list, entry) {
243		if (timer->expires < elapsed)
244			/* move expired timer to the callback queue */
245			list_move_tail(&timer->entry, &cb_list);
246		else
247			timer->expires -= elapsed;
248	}
249	if (!list_empty(&virt_timer_list)) {
250		timer = list_first_entry(&virt_timer_list,
251					 struct vtimer_list, entry);
252		atomic64_set(&virt_timer_current, timer->expires);
253	}
254	atomic64_sub(elapsed, &virt_timer_elapsed);
255	spin_unlock(&virt_timer_lock);
256
257	/* Do callbacks and recharge periodic timers */
258	list_for_each_entry_safe(timer, tmp, &cb_list, entry) {
259		list_del_init(&timer->entry);
260		timer->function(timer->data);
261		if (timer->interval) {
262			/* Recharge interval timer */
263			timer->expires = timer->interval +
264				atomic64_read(&virt_timer_elapsed);
265			spin_lock(&virt_timer_lock);
266			list_add_sorted(timer, &virt_timer_list);
267			spin_unlock(&virt_timer_lock);
268		}
269	}
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
284static void internal_add_vtimer(struct vtimer_list *timer)
285{
286	if (list_empty(&virt_timer_list)) {
287		/* First timer, just program it. */
288		atomic64_set(&virt_timer_current, timer->expires);
289		atomic64_set(&virt_timer_elapsed, 0);
290		list_add(&timer->entry, &virt_timer_list);
291	} else {
292		/* Update timer against current base. */
293		timer->expires += atomic64_read(&virt_timer_elapsed);
294		if (likely((s64) timer->expires <
295			   (s64) atomic64_read(&virt_timer_current)))
296			/* The new timer expires before the current timer. */
297			atomic64_set(&virt_timer_current, timer->expires);
298		/* Insert new timer into the list. */
299		list_add_sorted(timer, &virt_timer_list);
300	}
301}
302
303static void __add_vtimer(struct vtimer_list *timer, int periodic)
304{
305	unsigned long flags;
306
307	timer->interval = periodic ? timer->expires : 0;
308	spin_lock_irqsave(&virt_timer_lock, flags);
309	internal_add_vtimer(timer);
310	spin_unlock_irqrestore(&virt_timer_lock, flags);
311}
312
313/*
314 * add_virt_timer - add an oneshot virtual CPU timer
315 */
316void add_virt_timer(struct vtimer_list *timer)
317{
318	__add_vtimer(timer, 0);
319}
320EXPORT_SYMBOL(add_virt_timer);
321
322/*
323 * add_virt_timer_int - add an interval virtual CPU timer
324 */
325void add_virt_timer_periodic(struct vtimer_list *timer)
326{
327	__add_vtimer(timer, 1);
328}
329EXPORT_SYMBOL(add_virt_timer_periodic);
330
331static int __mod_vtimer(struct vtimer_list *timer, u64 expires, int periodic)
332{
333	unsigned long flags;
334	int rc;
335
336	BUG_ON(!timer->function);
337
338	if (timer->expires == expires && vtimer_pending(timer))
339		return 1;
340	spin_lock_irqsave(&virt_timer_lock, flags);
341	rc = vtimer_pending(timer);
342	if (rc)
343		list_del_init(&timer->entry);
344	timer->interval = periodic ? expires : 0;
345	timer->expires = expires;
346	internal_add_vtimer(timer);
347	spin_unlock_irqrestore(&virt_timer_lock, flags);
348	return rc;
349}
350
351/*
352 * returns whether it has modified a pending timer (1) or not (0)
353 */
354int mod_virt_timer(struct vtimer_list *timer, u64 expires)
355{
356	return __mod_vtimer(timer, expires, 0);
357}
358EXPORT_SYMBOL(mod_virt_timer);
359
360/*
361 * returns whether it has modified a pending timer (1) or not (0)
362 */
363int mod_virt_timer_periodic(struct vtimer_list *timer, u64 expires)
364{
365	return __mod_vtimer(timer, expires, 1);
366}
367EXPORT_SYMBOL(mod_virt_timer_periodic);
368
369/*
370 * Delete a virtual timer.
371 *
372 * returns whether the deleted timer was pending (1) or not (0)
373 */
374int del_virt_timer(struct vtimer_list *timer)
375{
376	unsigned long flags;
377
378	if (!vtimer_pending(timer))
379		return 0;
380	spin_lock_irqsave(&virt_timer_lock, flags);
381	list_del_init(&timer->entry);
382	spin_unlock_irqrestore(&virt_timer_lock, flags);
383	return 1;
384}
385EXPORT_SYMBOL(del_virt_timer);
386
387/*
388 * Start the virtual CPU timer on the current CPU.
389 */
390void vtime_init(void)
391{
392	/* set initial cpu timer */
393	set_vtimer(VTIMER_MAX_SLICE);
394	/* Setup initial MT scaling values */
395	if (smp_cpu_mtid) {
396		__this_cpu_write(mt_scaling_jiffies, jiffies);
397		__this_cpu_write(mt_scaling_mult, 1);
398		__this_cpu_write(mt_scaling_div, 1);
399		stcctm5(smp_cpu_mtid + 1, this_cpu_ptr(mt_cycles));
400	}
401}