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>
23 #include <linux/kprobes.h>
25 #include <asm/timer.h>
26 #include <asm/irq_regs.h>
27 #include <asm/cputime.h>
30 static DEFINE_PER_CPU(struct vtimer_queue, virt_cpu_timer);
32 DEFINE_PER_CPU(struct s390_idle_data, s390_idle);
34 static inline __u64 get_vtimer(void)
38 asm volatile("STPT %0" : "=m" (timer));
42 static inline void set_vtimer(__u64 expires)
46 asm volatile (" STPT %0\n" /* Store current cpu timer value */
47 " SPT %1" /* Set new value immediately afterwards */
48 : "=m" (timer) : "m" (expires) );
49 S390_lowcore.system_timer += S390_lowcore.last_update_timer - timer;
50 S390_lowcore.last_update_timer = expires;
54 * Update process times based on virtual cpu times stored by entry.S
55 * to the lowcore fields user_timer, system_timer & steal_clock.
57 static void do_account_vtime(struct task_struct *tsk, int hardirq_offset)
59 struct thread_info *ti = task_thread_info(tsk);
60 __u64 timer, clock, user, system, steal;
62 timer = S390_lowcore.last_update_timer;
63 clock = S390_lowcore.last_update_clock;
64 asm volatile (" STPT %0\n" /* Store current cpu timer value */
65 " STCK %1" /* Store current tod clock value */
66 : "=m" (S390_lowcore.last_update_timer),
67 "=m" (S390_lowcore.last_update_clock) );
68 S390_lowcore.system_timer += timer - S390_lowcore.last_update_timer;
69 S390_lowcore.steal_timer += S390_lowcore.last_update_clock - clock;
71 user = S390_lowcore.user_timer - ti->user_timer;
72 S390_lowcore.steal_timer -= user;
73 ti->user_timer = S390_lowcore.user_timer;
74 account_user_time(tsk, user, user);
76 system = S390_lowcore.system_timer - ti->system_timer;
77 S390_lowcore.steal_timer -= system;
78 ti->system_timer = S390_lowcore.system_timer;
79 account_system_time(tsk, hardirq_offset, system, system);
81 steal = S390_lowcore.steal_timer;
82 if ((s64) steal > 0) {
83 S390_lowcore.steal_timer = 0;
84 account_steal_time(steal);
88 void account_vtime(struct task_struct *prev, struct task_struct *next)
90 struct thread_info *ti;
92 do_account_vtime(prev, 0);
93 ti = task_thread_info(prev);
94 ti->user_timer = S390_lowcore.user_timer;
95 ti->system_timer = S390_lowcore.system_timer;
96 ti = task_thread_info(next);
97 S390_lowcore.user_timer = ti->user_timer;
98 S390_lowcore.system_timer = ti->system_timer;
101 void account_process_tick(struct task_struct *tsk, int user_tick)
103 do_account_vtime(tsk, HARDIRQ_OFFSET);
107 * Update process times based on virtual cpu times stored by entry.S
108 * to the lowcore fields user_timer, system_timer & steal_clock.
110 void account_system_vtime(struct task_struct *tsk)
112 struct thread_info *ti = task_thread_info(tsk);
115 timer = S390_lowcore.last_update_timer;
116 S390_lowcore.last_update_timer = get_vtimer();
117 S390_lowcore.system_timer += timer - S390_lowcore.last_update_timer;
119 system = S390_lowcore.system_timer - ti->system_timer;
120 S390_lowcore.steal_timer -= system;
121 ti->system_timer = S390_lowcore.system_timer;
122 account_system_time(tsk, 0, system, system);
124 EXPORT_SYMBOL_GPL(account_system_vtime);
126 void __kprobes vtime_start_cpu(__u64 int_clock, __u64 enter_timer)
128 struct s390_idle_data *idle = &__get_cpu_var(s390_idle);
129 struct vtimer_queue *vq = &__get_cpu_var(virt_cpu_timer);
130 __u64 idle_time, expires;
132 if (idle->idle_enter == 0ULL)
135 /* Account time spent with enabled wait psw loaded as idle time. */
136 idle_time = int_clock - idle->idle_enter;
137 account_idle_time(idle_time);
138 S390_lowcore.steal_timer +=
139 idle->idle_enter - S390_lowcore.last_update_clock;
140 S390_lowcore.last_update_clock = int_clock;
142 /* Account system time spent going idle. */
143 S390_lowcore.system_timer += S390_lowcore.last_update_timer - vq->idle;
144 S390_lowcore.last_update_timer = enter_timer;
146 /* Restart vtime CPU timer */
148 /* Program old expire value but first save progress. */
149 expires = vq->idle - enter_timer;
150 expires += get_vtimer();
153 /* Don't account the CPU timer delta while the cpu was idle. */
154 vq->elapsed -= vq->idle - enter_timer;
159 idle->idle_time += idle_time;
160 idle->idle_enter = 0ULL;
166 void __kprobes vtime_stop_cpu(void)
168 struct s390_idle_data *idle = &__get_cpu_var(s390_idle);
169 struct vtimer_queue *vq = &__get_cpu_var(virt_cpu_timer);
172 /* Wait for external, I/O or machine check interrupt. */
173 psw.mask = psw_kernel_bits | PSW_MASK_WAIT | PSW_MASK_IO | PSW_MASK_EXT;
175 idle->nohz_delay = 0;
177 /* Check if the CPU timer needs to be reprogrammed. */
179 __u64 vmax = VTIMER_MAX_SLICE;
181 * The inline assembly is equivalent to
182 * vq->idle = get_cpu_timer();
183 * set_cpu_timer(VTIMER_MAX_SLICE);
184 * idle->idle_enter = get_clock();
185 * __load_psw_mask(psw_kernel_bits | PSW_MASK_WAIT |
186 * PSW_MASK_IO | PSW_MASK_EXT);
187 * The difference is that the inline assembly makes sure that
188 * the last three instruction are stpt, stck and lpsw in that
189 * order. This is done to increase the precision.
196 #else /* CONFIG_64BIT */
199 #endif /* CONFIG_64BIT */
205 #else /* CONFIG_64BIT */
207 #endif /* CONFIG_64BIT */
209 : "=m" (idle->idle_enter), "=m" (vq->idle)
210 : "a" (&psw), "a" (&idle->idle_enter),
211 "a" (&vq->idle), "a" (&vmax), "m" (vmax), "m" (psw)
212 : "memory", "cc", "1");
215 * The inline assembly is equivalent to
216 * vq->idle = get_cpu_timer();
217 * idle->idle_enter = get_clock();
218 * __load_psw_mask(psw_kernel_bits | PSW_MASK_WAIT |
219 * PSW_MASK_IO | PSW_MASK_EXT);
220 * The difference is that the inline assembly makes sure that
221 * the last three instruction are stpt, stck and lpsw in that
222 * order. This is done to increase the precision.
229 #else /* CONFIG_64BIT */
232 #endif /* CONFIG_64BIT */
237 #else /* CONFIG_64BIT */
239 #endif /* CONFIG_64BIT */
241 : "=m" (idle->idle_enter), "=m" (vq->idle)
242 : "a" (&psw), "a" (&idle->idle_enter),
243 "a" (&vq->idle), "m" (psw)
244 : "memory", "cc", "1");
248 cputime64_t s390_get_idle_time(int cpu)
250 struct s390_idle_data *idle;
251 unsigned long long now, idle_time, idle_enter;
252 unsigned int sequence;
254 idle = &per_cpu(s390_idle, cpu);
258 sequence = idle->sequence;
263 idle_enter = idle->idle_enter;
264 if (idle_enter != 0ULL && idle_enter < now)
265 idle_time = now - idle_enter;
267 if (idle->sequence != sequence)
273 * Sorted add to a list. List is linear searched until first bigger
276 static void list_add_sorted(struct vtimer_list *timer, struct list_head *head)
278 struct vtimer_list *event;
280 list_for_each_entry(event, head, entry) {
281 if (event->expires > timer->expires) {
282 list_add_tail(&timer->entry, &event->entry);
286 list_add_tail(&timer->entry, head);
290 * Do the callback functions of expired vtimer events.
291 * Called from within the interrupt handler.
293 static void do_callbacks(struct list_head *cb_list)
295 struct vtimer_queue *vq;
296 struct vtimer_list *event, *tmp;
298 if (list_empty(cb_list))
301 vq = &__get_cpu_var(virt_cpu_timer);
303 list_for_each_entry_safe(event, tmp, cb_list, entry) {
304 list_del_init(&event->entry);
305 (event->function)(event->data);
306 if (event->interval) {
307 /* Recharge interval timer */
308 event->expires = event->interval + vq->elapsed;
309 spin_lock(&vq->lock);
310 list_add_sorted(event, &vq->list);
311 spin_unlock(&vq->lock);
317 * Handler for the virtual CPU timer.
319 static void do_cpu_timer_interrupt(unsigned int ext_int_code,
320 unsigned int param32, unsigned long param64)
322 struct vtimer_queue *vq;
323 struct vtimer_list *event, *tmp;
324 struct list_head cb_list; /* the callback queue */
327 kstat_cpu(smp_processor_id()).irqs[EXTINT_TMR]++;
328 INIT_LIST_HEAD(&cb_list);
329 vq = &__get_cpu_var(virt_cpu_timer);
331 /* walk timer list, fire all expired events */
332 spin_lock(&vq->lock);
334 elapsed = vq->elapsed + (vq->timer - S390_lowcore.async_enter_timer);
335 BUG_ON((s64) elapsed < 0);
337 list_for_each_entry_safe(event, tmp, &vq->list, entry) {
338 if (event->expires < elapsed)
339 /* move expired timer to the callback queue */
340 list_move_tail(&event->entry, &cb_list);
342 event->expires -= elapsed;
344 spin_unlock(&vq->lock);
346 vq->do_spt = list_empty(&cb_list);
347 do_callbacks(&cb_list);
349 /* next event is first in list */
350 next = VTIMER_MAX_SLICE;
351 spin_lock(&vq->lock);
352 if (!list_empty(&vq->list)) {
353 event = list_first_entry(&vq->list, struct vtimer_list, entry);
354 next = event->expires;
357 spin_unlock(&vq->lock);
359 * To improve precision add the time spent by the
360 * interrupt handler to the elapsed time.
361 * Note: CPU timer counts down and we got an interrupt,
362 * the current content is negative
364 elapsed = S390_lowcore.async_enter_timer - get_vtimer();
365 set_vtimer(next - elapsed);
366 vq->timer = next - elapsed;
367 vq->elapsed = elapsed;
370 void init_virt_timer(struct vtimer_list *timer)
372 timer->function = NULL;
373 INIT_LIST_HEAD(&timer->entry);
375 EXPORT_SYMBOL(init_virt_timer);
377 static inline int vtimer_pending(struct vtimer_list *timer)
379 return (!list_empty(&timer->entry));
383 * this function should only run on the specified CPU
385 static void internal_add_vtimer(struct vtimer_list *timer)
387 struct vtimer_queue *vq;
391 vq = &per_cpu(virt_cpu_timer, timer->cpu);
392 spin_lock_irqsave(&vq->lock, flags);
394 BUG_ON(timer->cpu != smp_processor_id());
396 if (list_empty(&vq->list)) {
397 /* First timer on this cpu, just program it. */
398 list_add(&timer->entry, &vq->list);
399 set_vtimer(timer->expires);
400 vq->timer = timer->expires;
403 /* Check progress of old timers. */
404 expires = timer->expires;
406 if (likely((s64) expires < (s64) left)) {
407 /* The new timer expires before the current timer. */
409 vq->elapsed += vq->timer - left;
412 vq->elapsed += vq->timer - left;
415 /* Insert new timer into per cpu list. */
416 timer->expires += vq->elapsed;
417 list_add_sorted(timer, &vq->list);
420 spin_unlock_irqrestore(&vq->lock, flags);
421 /* release CPU acquired in prepare_vtimer or mod_virt_timer() */
425 static inline void prepare_vtimer(struct vtimer_list *timer)
427 BUG_ON(!timer->function);
428 BUG_ON(!timer->expires || timer->expires > VTIMER_MAX_SLICE);
429 BUG_ON(vtimer_pending(timer));
430 timer->cpu = get_cpu();
434 * add_virt_timer - add an oneshot virtual CPU timer
436 void add_virt_timer(void *new)
438 struct vtimer_list *timer;
440 timer = (struct vtimer_list *)new;
441 prepare_vtimer(timer);
443 internal_add_vtimer(timer);
445 EXPORT_SYMBOL(add_virt_timer);
448 * add_virt_timer_int - add an interval virtual CPU timer
450 void add_virt_timer_periodic(void *new)
452 struct vtimer_list *timer;
454 timer = (struct vtimer_list *)new;
455 prepare_vtimer(timer);
456 timer->interval = timer->expires;
457 internal_add_vtimer(timer);
459 EXPORT_SYMBOL(add_virt_timer_periodic);
461 int __mod_vtimer(struct vtimer_list *timer, __u64 expires, int periodic)
463 struct vtimer_queue *vq;
467 BUG_ON(!timer->function);
468 BUG_ON(!expires || expires > VTIMER_MAX_SLICE);
470 if (timer->expires == expires && vtimer_pending(timer))
474 vq = &per_cpu(virt_cpu_timer, cpu);
476 /* disable interrupts before test if timer is pending */
477 spin_lock_irqsave(&vq->lock, flags);
479 /* if timer isn't pending add it on the current CPU */
480 if (!vtimer_pending(timer)) {
481 spin_unlock_irqrestore(&vq->lock, flags);
484 timer->interval = expires;
487 timer->expires = expires;
489 internal_add_vtimer(timer);
493 /* check if we run on the right CPU */
494 BUG_ON(timer->cpu != cpu);
496 list_del_init(&timer->entry);
497 timer->expires = expires;
499 timer->interval = expires;
501 /* the timer can't expire anymore so we can release the lock */
502 spin_unlock_irqrestore(&vq->lock, flags);
503 internal_add_vtimer(timer);
508 * If we change a pending timer the function must be called on the CPU
509 * where the timer is running on.
511 * returns whether it has modified a pending timer (1) or not (0)
513 int mod_virt_timer(struct vtimer_list *timer, __u64 expires)
515 return __mod_vtimer(timer, expires, 0);
517 EXPORT_SYMBOL(mod_virt_timer);
520 * If we change a pending timer the function must be called on the CPU
521 * where the timer is running on.
523 * returns whether it has modified a pending timer (1) or not (0)
525 int mod_virt_timer_periodic(struct vtimer_list *timer, __u64 expires)
527 return __mod_vtimer(timer, expires, 1);
529 EXPORT_SYMBOL(mod_virt_timer_periodic);
532 * delete a virtual timer
534 * returns whether the deleted timer was pending (1) or not (0)
536 int del_virt_timer(struct vtimer_list *timer)
539 struct vtimer_queue *vq;
541 /* check if timer is pending */
542 if (!vtimer_pending(timer))
545 vq = &per_cpu(virt_cpu_timer, timer->cpu);
546 spin_lock_irqsave(&vq->lock, flags);
548 /* we don't interrupt a running timer, just let it expire! */
549 list_del_init(&timer->entry);
551 spin_unlock_irqrestore(&vq->lock, flags);
554 EXPORT_SYMBOL(del_virt_timer);
557 * Start the virtual CPU timer on the current CPU.
559 void init_cpu_vtimer(void)
561 struct vtimer_queue *vq;
563 /* initialize per cpu vtimer structure */
564 vq = &__get_cpu_var(virt_cpu_timer);
565 INIT_LIST_HEAD(&vq->list);
566 spin_lock_init(&vq->lock);
568 /* enable cpu timer interrupts */
572 static int __cpuinit s390_nohz_notify(struct notifier_block *self,
573 unsigned long action, void *hcpu)
575 struct s390_idle_data *idle;
576 long cpu = (long) hcpu;
578 idle = &per_cpu(s390_idle, cpu);
581 case CPU_DYING_FROZEN:
582 idle->nohz_delay = 0;
589 void __init vtime_init(void)
591 /* request the cpu timer external interrupt */
592 if (register_external_interrupt(0x1005, do_cpu_timer_interrupt))
593 panic("Couldn't request external interrupt 0x1005");
595 /* Enable cpu timer interrupts on the boot cpu. */
597 cpu_notifier(s390_nohz_notify, 0);