2 * arch/s390/kernel/vtime.c
3 * Virtual cpu timer based timer functions.
6 * Copyright (C) 2004 IBM Deutschland Entwicklung GmbH, IBM Corporation
7 * Author(s): Jan Glauber <jan.glauber@de.ibm.com>
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>
24 #include <asm/s390_ext.h>
25 #include <asm/timer.h>
26 #include <asm/irq_regs.h>
27 #include <asm/cputime.h>
29 static DEFINE_PER_CPU(struct vtimer_queue, virt_cpu_timer);
31 DEFINE_PER_CPU(struct s390_idle_data, s390_idle);
33 static inline __u64 get_vtimer(void)
37 asm volatile("STPT %0" : "=m" (timer));
41 static inline void set_vtimer(__u64 expires)
45 asm volatile (" STPT %0\n" /* Store current cpu timer value */
46 " SPT %1" /* Set new value immediatly afterwards */
47 : "=m" (timer) : "m" (expires) );
48 S390_lowcore.system_timer += S390_lowcore.last_update_timer - timer;
49 S390_lowcore.last_update_timer = expires;
53 * Update process times based on virtual cpu times stored by entry.S
54 * to the lowcore fields user_timer, system_timer & steal_clock.
56 static void do_account_vtime(struct task_struct *tsk, int hardirq_offset)
58 struct thread_info *ti = task_thread_info(tsk);
59 __u64 timer, clock, user, system, steal;
61 timer = S390_lowcore.last_update_timer;
62 clock = S390_lowcore.last_update_clock;
63 asm volatile (" STPT %0\n" /* Store current cpu timer value */
64 " STCK %1" /* Store current tod clock value */
65 : "=m" (S390_lowcore.last_update_timer),
66 "=m" (S390_lowcore.last_update_clock) );
67 S390_lowcore.system_timer += timer - S390_lowcore.last_update_timer;
68 S390_lowcore.steal_timer += S390_lowcore.last_update_clock - clock;
70 user = S390_lowcore.user_timer - ti->user_timer;
71 S390_lowcore.steal_timer -= user;
72 ti->user_timer = S390_lowcore.user_timer;
73 account_user_time(tsk, user, user);
75 system = S390_lowcore.system_timer - ti->system_timer;
76 S390_lowcore.steal_timer -= system;
77 ti->system_timer = S390_lowcore.system_timer;
78 account_system_time(tsk, hardirq_offset, system, system);
80 steal = S390_lowcore.steal_timer;
81 if ((s64) steal > 0) {
82 S390_lowcore.steal_timer = 0;
83 account_steal_time(steal);
87 void account_vtime(struct task_struct *prev, struct task_struct *next)
89 struct thread_info *ti;
91 do_account_vtime(prev, 0);
92 ti = task_thread_info(prev);
93 ti->user_timer = S390_lowcore.user_timer;
94 ti->system_timer = S390_lowcore.system_timer;
95 ti = task_thread_info(next);
96 S390_lowcore.user_timer = ti->user_timer;
97 S390_lowcore.system_timer = ti->system_timer;
100 void account_process_tick(struct task_struct *tsk, int user_tick)
102 do_account_vtime(tsk, HARDIRQ_OFFSET);
106 * Update process times based on virtual cpu times stored by entry.S
107 * to the lowcore fields user_timer, system_timer & steal_clock.
109 void account_system_vtime(struct task_struct *tsk)
111 struct thread_info *ti = task_thread_info(tsk);
114 timer = S390_lowcore.last_update_timer;
115 S390_lowcore.last_update_timer = get_vtimer();
116 S390_lowcore.system_timer += timer - S390_lowcore.last_update_timer;
118 system = S390_lowcore.system_timer - ti->system_timer;
119 S390_lowcore.steal_timer -= system;
120 ti->system_timer = S390_lowcore.system_timer;
121 account_system_time(tsk, 0, system, system);
123 EXPORT_SYMBOL_GPL(account_system_vtime);
125 void vtime_start_cpu(__u64 int_clock, __u64 enter_timer)
127 struct s390_idle_data *idle = &__get_cpu_var(s390_idle);
128 struct vtimer_queue *vq = &__get_cpu_var(virt_cpu_timer);
129 __u64 idle_time, expires;
131 if (idle->idle_enter == 0ULL)
134 /* Account time spent with enabled wait psw loaded as idle time. */
135 idle_time = int_clock - idle->idle_enter;
136 account_idle_time(idle_time);
137 S390_lowcore.steal_timer +=
138 idle->idle_enter - S390_lowcore.last_update_clock;
139 S390_lowcore.last_update_clock = int_clock;
141 /* Account system time spent going idle. */
142 S390_lowcore.system_timer += S390_lowcore.last_update_timer - vq->idle;
143 S390_lowcore.last_update_timer = enter_timer;
145 /* Restart vtime CPU timer */
147 /* Program old expire value but first save progress. */
148 expires = vq->idle - enter_timer;
149 expires += get_vtimer();
152 /* Don't account the CPU timer delta while the cpu was idle. */
153 vq->elapsed -= vq->idle - enter_timer;
158 idle->idle_time += idle_time;
159 idle->idle_enter = 0ULL;
165 void vtime_stop_cpu(void)
167 struct s390_idle_data *idle = &__get_cpu_var(s390_idle);
168 struct vtimer_queue *vq = &__get_cpu_var(virt_cpu_timer);
171 /* Wait for external, I/O or machine check interrupt. */
172 psw.mask = psw_kernel_bits | PSW_MASK_WAIT | PSW_MASK_IO | PSW_MASK_EXT;
174 idle->nohz_delay = 0;
176 /* Check if the CPU timer needs to be reprogrammed. */
178 __u64 vmax = VTIMER_MAX_SLICE;
180 * The inline assembly is equivalent to
181 * vq->idle = get_cpu_timer();
182 * set_cpu_timer(VTIMER_MAX_SLICE);
183 * idle->idle_enter = get_clock();
184 * __load_psw_mask(psw_kernel_bits | PSW_MASK_WAIT |
185 * PSW_MASK_IO | PSW_MASK_EXT);
186 * The difference is that the inline assembly makes sure that
187 * the last three instruction are stpt, stck and lpsw in that
188 * order. This is done to increase the precision.
195 #else /* CONFIG_64BIT */
198 #endif /* CONFIG_64BIT */
204 #else /* CONFIG_64BIT */
206 #endif /* CONFIG_64BIT */
208 : "=m" (idle->idle_enter), "=m" (vq->idle)
209 : "a" (&psw), "a" (&idle->idle_enter),
210 "a" (&vq->idle), "a" (&vmax), "m" (vmax), "m" (psw)
211 : "memory", "cc", "1");
214 * The inline assembly is equivalent to
215 * vq->idle = get_cpu_timer();
216 * idle->idle_enter = get_clock();
217 * __load_psw_mask(psw_kernel_bits | PSW_MASK_WAIT |
218 * PSW_MASK_IO | PSW_MASK_EXT);
219 * The difference is that the inline assembly makes sure that
220 * the last three instruction are stpt, stck and lpsw in that
221 * order. This is done to increase the precision.
228 #else /* CONFIG_64BIT */
231 #endif /* CONFIG_64BIT */
236 #else /* CONFIG_64BIT */
238 #endif /* CONFIG_64BIT */
240 : "=m" (idle->idle_enter), "=m" (vq->idle)
241 : "a" (&psw), "a" (&idle->idle_enter),
242 "a" (&vq->idle), "m" (psw)
243 : "memory", "cc", "1");
247 cputime64_t s390_get_idle_time(int cpu)
249 struct s390_idle_data *idle;
250 unsigned long long now, idle_time, idle_enter;
251 unsigned int sequence;
253 idle = &per_cpu(s390_idle, cpu);
257 sequence = idle->sequence;
262 idle_enter = idle->idle_enter;
263 if (idle_enter != 0ULL && idle_enter < now)
264 idle_time = now - idle_enter;
266 if (idle->sequence != sequence)
272 * Sorted add to a list. List is linear searched until first bigger
275 static void list_add_sorted(struct vtimer_list *timer, struct list_head *head)
277 struct vtimer_list *event;
279 list_for_each_entry(event, head, entry) {
280 if (event->expires > timer->expires) {
281 list_add_tail(&timer->entry, &event->entry);
285 list_add_tail(&timer->entry, head);
289 * Do the callback functions of expired vtimer events.
290 * Called from within the interrupt handler.
292 static void do_callbacks(struct list_head *cb_list)
294 struct vtimer_queue *vq;
295 struct vtimer_list *event, *tmp;
297 if (list_empty(cb_list))
300 vq = &__get_cpu_var(virt_cpu_timer);
302 list_for_each_entry_safe(event, tmp, cb_list, entry) {
303 list_del_init(&event->entry);
304 (event->function)(event->data);
305 if (event->interval) {
306 /* Recharge interval timer */
307 event->expires = event->interval + vq->elapsed;
308 spin_lock(&vq->lock);
309 list_add_sorted(event, &vq->list);
310 spin_unlock(&vq->lock);
316 * Handler for the virtual CPU timer.
318 static void do_cpu_timer_interrupt(unsigned int ext_int_code,
319 unsigned int param32, unsigned long param64)
321 struct vtimer_queue *vq;
322 struct vtimer_list *event, *tmp;
323 struct list_head cb_list; /* the callback queue */
326 INIT_LIST_HEAD(&cb_list);
327 vq = &__get_cpu_var(virt_cpu_timer);
329 /* walk timer list, fire all expired events */
330 spin_lock(&vq->lock);
332 elapsed = vq->elapsed + (vq->timer - S390_lowcore.async_enter_timer);
333 BUG_ON((s64) elapsed < 0);
335 list_for_each_entry_safe(event, tmp, &vq->list, entry) {
336 if (event->expires < elapsed)
337 /* move expired timer to the callback queue */
338 list_move_tail(&event->entry, &cb_list);
340 event->expires -= elapsed;
342 spin_unlock(&vq->lock);
344 vq->do_spt = list_empty(&cb_list);
345 do_callbacks(&cb_list);
347 /* next event is first in list */
348 next = VTIMER_MAX_SLICE;
349 spin_lock(&vq->lock);
350 if (!list_empty(&vq->list)) {
351 event = list_first_entry(&vq->list, struct vtimer_list, entry);
352 next = event->expires;
355 spin_unlock(&vq->lock);
357 * To improve precision add the time spent by the
358 * interrupt handler to the elapsed time.
359 * Note: CPU timer counts down and we got an interrupt,
360 * the current content is negative
362 elapsed = S390_lowcore.async_enter_timer - get_vtimer();
363 set_vtimer(next - elapsed);
364 vq->timer = next - elapsed;
365 vq->elapsed = elapsed;
368 void init_virt_timer(struct vtimer_list *timer)
370 timer->function = NULL;
371 INIT_LIST_HEAD(&timer->entry);
373 EXPORT_SYMBOL(init_virt_timer);
375 static inline int vtimer_pending(struct vtimer_list *timer)
377 return (!list_empty(&timer->entry));
381 * this function should only run on the specified CPU
383 static void internal_add_vtimer(struct vtimer_list *timer)
385 struct vtimer_queue *vq;
389 vq = &per_cpu(virt_cpu_timer, timer->cpu);
390 spin_lock_irqsave(&vq->lock, flags);
392 BUG_ON(timer->cpu != smp_processor_id());
394 if (list_empty(&vq->list)) {
395 /* First timer on this cpu, just program it. */
396 list_add(&timer->entry, &vq->list);
397 set_vtimer(timer->expires);
398 vq->timer = timer->expires;
401 /* Check progress of old timers. */
402 expires = timer->expires;
404 if (likely((s64) expires < (s64) left)) {
405 /* The new timer expires before the current timer. */
407 vq->elapsed += vq->timer - left;
410 vq->elapsed += vq->timer - left;
413 /* Insert new timer into per cpu list. */
414 timer->expires += vq->elapsed;
415 list_add_sorted(timer, &vq->list);
418 spin_unlock_irqrestore(&vq->lock, flags);
419 /* release CPU acquired in prepare_vtimer or mod_virt_timer() */
423 static inline void prepare_vtimer(struct vtimer_list *timer)
425 BUG_ON(!timer->function);
426 BUG_ON(!timer->expires || timer->expires > VTIMER_MAX_SLICE);
427 BUG_ON(vtimer_pending(timer));
428 timer->cpu = get_cpu();
432 * add_virt_timer - add an oneshot virtual CPU timer
434 void add_virt_timer(void *new)
436 struct vtimer_list *timer;
438 timer = (struct vtimer_list *)new;
439 prepare_vtimer(timer);
441 internal_add_vtimer(timer);
443 EXPORT_SYMBOL(add_virt_timer);
446 * add_virt_timer_int - add an interval virtual CPU timer
448 void add_virt_timer_periodic(void *new)
450 struct vtimer_list *timer;
452 timer = (struct vtimer_list *)new;
453 prepare_vtimer(timer);
454 timer->interval = timer->expires;
455 internal_add_vtimer(timer);
457 EXPORT_SYMBOL(add_virt_timer_periodic);
459 int __mod_vtimer(struct vtimer_list *timer, __u64 expires, int periodic)
461 struct vtimer_queue *vq;
465 BUG_ON(!timer->function);
466 BUG_ON(!expires || expires > VTIMER_MAX_SLICE);
468 if (timer->expires == expires && vtimer_pending(timer))
472 vq = &per_cpu(virt_cpu_timer, cpu);
474 /* disable interrupts before test if timer is pending */
475 spin_lock_irqsave(&vq->lock, flags);
477 /* if timer isn't pending add it on the current CPU */
478 if (!vtimer_pending(timer)) {
479 spin_unlock_irqrestore(&vq->lock, flags);
482 timer->interval = expires;
485 timer->expires = expires;
487 internal_add_vtimer(timer);
491 /* check if we run on the right CPU */
492 BUG_ON(timer->cpu != cpu);
494 list_del_init(&timer->entry);
495 timer->expires = expires;
497 timer->interval = expires;
499 /* the timer can't expire anymore so we can release the lock */
500 spin_unlock_irqrestore(&vq->lock, flags);
501 internal_add_vtimer(timer);
506 * If we change a pending timer the function must be called on the CPU
507 * where the timer is running on.
509 * returns whether it has modified a pending timer (1) or not (0)
511 int mod_virt_timer(struct vtimer_list *timer, __u64 expires)
513 return __mod_vtimer(timer, expires, 0);
515 EXPORT_SYMBOL(mod_virt_timer);
518 * If we change a pending timer the function must be called on the CPU
519 * where the timer is running on.
521 * returns whether it has modified a pending timer (1) or not (0)
523 int mod_virt_timer_periodic(struct vtimer_list *timer, __u64 expires)
525 return __mod_vtimer(timer, expires, 1);
527 EXPORT_SYMBOL(mod_virt_timer_periodic);
530 * delete a virtual timer
532 * returns whether the deleted timer was pending (1) or not (0)
534 int del_virt_timer(struct vtimer_list *timer)
537 struct vtimer_queue *vq;
539 /* check if timer is pending */
540 if (!vtimer_pending(timer))
543 vq = &per_cpu(virt_cpu_timer, timer->cpu);
544 spin_lock_irqsave(&vq->lock, flags);
546 /* we don't interrupt a running timer, just let it expire! */
547 list_del_init(&timer->entry);
549 spin_unlock_irqrestore(&vq->lock, flags);
552 EXPORT_SYMBOL(del_virt_timer);
555 * Start the virtual CPU timer on the current CPU.
557 void init_cpu_vtimer(void)
559 struct vtimer_queue *vq;
561 /* initialize per cpu vtimer structure */
562 vq = &__get_cpu_var(virt_cpu_timer);
563 INIT_LIST_HEAD(&vq->list);
564 spin_lock_init(&vq->lock);
566 /* enable cpu timer interrupts */
570 static int __cpuinit s390_nohz_notify(struct notifier_block *self,
571 unsigned long action, void *hcpu)
573 struct s390_idle_data *idle;
574 long cpu = (long) hcpu;
576 idle = &per_cpu(s390_idle, cpu);
579 case CPU_DYING_FROZEN:
580 idle->nohz_delay = 0;
587 void __init vtime_init(void)
589 /* request the cpu timer external interrupt */
590 if (register_external_interrupt(0x1005, do_cpu_timer_interrupt))
591 panic("Couldn't request external interrupt 0x1005");
593 /* Enable cpu timer interrupts on the boot cpu. */
595 cpu_notifier(s390_nohz_notify, 0);