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>
23 #include <asm/s390_ext.h>
24 #include <asm/timer.h>
25 #include <asm/irq_regs.h>
26 #include <asm/cputime.h>
28 static DEFINE_PER_CPU(struct vtimer_queue
, virt_cpu_timer
);
30 DEFINE_PER_CPU(struct s390_idle_data
, s390_idle
);
32 static inline __u64
get_vtimer(void)
36 asm volatile("STPT %0" : "=m" (timer
));
40 static inline void set_vtimer(__u64 expires
)
44 asm volatile (" STPT %0\n" /* Store current cpu timer value */
45 " SPT %1" /* Set new value immediatly afterwards */
46 : "=m" (timer
) : "m" (expires
) );
47 S390_lowcore
.system_timer
+= S390_lowcore
.last_update_timer
- timer
;
48 S390_lowcore
.last_update_timer
= expires
;
52 * Update process times based on virtual cpu times stored by entry.S
53 * to the lowcore fields user_timer, system_timer & steal_clock.
55 static void do_account_vtime(struct task_struct
*tsk
, int hardirq_offset
)
57 struct thread_info
*ti
= task_thread_info(tsk
);
58 __u64 timer
, clock
, user
, system
, steal
;
60 timer
= S390_lowcore
.last_update_timer
;
61 clock
= S390_lowcore
.last_update_clock
;
62 asm volatile (" STPT %0\n" /* Store current cpu timer value */
63 " STCK %1" /* Store current tod clock value */
64 : "=m" (S390_lowcore
.last_update_timer
),
65 "=m" (S390_lowcore
.last_update_clock
) );
66 S390_lowcore
.system_timer
+= timer
- S390_lowcore
.last_update_timer
;
67 S390_lowcore
.steal_timer
+= S390_lowcore
.last_update_clock
- clock
;
69 user
= S390_lowcore
.user_timer
- ti
->user_timer
;
70 S390_lowcore
.steal_timer
-= user
;
71 ti
->user_timer
= S390_lowcore
.user_timer
;
72 account_user_time(tsk
, user
, user
);
74 system
= S390_lowcore
.system_timer
- ti
->system_timer
;
75 S390_lowcore
.steal_timer
-= system
;
76 ti
->system_timer
= S390_lowcore
.system_timer
;
77 account_system_time(tsk
, hardirq_offset
, system
, system
);
79 steal
= S390_lowcore
.steal_timer
;
80 if ((s64
) steal
> 0) {
81 S390_lowcore
.steal_timer
= 0;
82 account_steal_time(steal
);
86 void account_vtime(struct task_struct
*prev
, struct task_struct
*next
)
88 struct thread_info
*ti
;
90 do_account_vtime(prev
, 0);
91 ti
= task_thread_info(prev
);
92 ti
->user_timer
= S390_lowcore
.user_timer
;
93 ti
->system_timer
= S390_lowcore
.system_timer
;
94 ti
= task_thread_info(next
);
95 S390_lowcore
.user_timer
= ti
->user_timer
;
96 S390_lowcore
.system_timer
= ti
->system_timer
;
99 void account_process_tick(struct task_struct
*tsk
, int user_tick
)
101 do_account_vtime(tsk
, HARDIRQ_OFFSET
);
105 * Update process times based on virtual cpu times stored by entry.S
106 * to the lowcore fields user_timer, system_timer & steal_clock.
108 void account_system_vtime(struct task_struct
*tsk
)
110 struct thread_info
*ti
= task_thread_info(tsk
);
113 timer
= S390_lowcore
.last_update_timer
;
114 S390_lowcore
.last_update_timer
= get_vtimer();
115 S390_lowcore
.system_timer
+= timer
- S390_lowcore
.last_update_timer
;
117 system
= S390_lowcore
.system_timer
- ti
->system_timer
;
118 S390_lowcore
.steal_timer
-= system
;
119 ti
->system_timer
= S390_lowcore
.system_timer
;
120 account_system_time(tsk
, 0, system
, system
);
122 EXPORT_SYMBOL_GPL(account_system_vtime
);
124 void vtime_start_cpu(void)
126 struct s390_idle_data
*idle
= &__get_cpu_var(s390_idle
);
127 struct vtimer_queue
*vq
= &__get_cpu_var(virt_cpu_timer
);
128 __u64 idle_time
, expires
;
130 /* Account time spent with enabled wait psw loaded as idle time. */
131 idle_time
= S390_lowcore
.int_clock
- idle
->idle_enter
;
132 account_idle_time(idle_time
);
133 S390_lowcore
.steal_timer
+=
134 idle
->idle_enter
- S390_lowcore
.last_update_clock
;
135 S390_lowcore
.last_update_clock
= S390_lowcore
.int_clock
;
137 /* Account system time spent going idle. */
138 S390_lowcore
.system_timer
+= S390_lowcore
.last_update_timer
- vq
->idle
;
139 S390_lowcore
.last_update_timer
= S390_lowcore
.async_enter_timer
;
141 /* Restart vtime CPU timer */
143 /* Program old expire value but first save progress. */
144 expires
= vq
->idle
- S390_lowcore
.async_enter_timer
;
145 expires
+= get_vtimer();
148 /* Don't account the CPU timer delta while the cpu was idle. */
149 vq
->elapsed
-= vq
->idle
- S390_lowcore
.async_enter_timer
;
154 idle
->idle_time
+= idle_time
;
155 idle
->idle_enter
= 0ULL;
161 void vtime_stop_cpu(void)
163 struct s390_idle_data
*idle
= &__get_cpu_var(s390_idle
);
164 struct vtimer_queue
*vq
= &__get_cpu_var(virt_cpu_timer
);
167 /* Wait for external, I/O or machine check interrupt. */
168 psw
.mask
= psw_kernel_bits
| PSW_MASK_WAIT
| PSW_MASK_IO
| PSW_MASK_EXT
;
170 idle
->nohz_delay
= 0;
172 /* Check if the CPU timer needs to be reprogrammed. */
174 __u64 vmax
= VTIMER_MAX_SLICE
;
176 * The inline assembly is equivalent to
177 * vq->idle = get_cpu_timer();
178 * set_cpu_timer(VTIMER_MAX_SLICE);
179 * idle->idle_enter = get_clock();
180 * __load_psw_mask(psw_kernel_bits | PSW_MASK_WAIT |
181 * PSW_MASK_IO | PSW_MASK_EXT);
182 * The difference is that the inline assembly makes sure that
183 * the last three instruction are stpt, stck and lpsw in that
184 * order. This is done to increase the precision.
191 #else /* CONFIG_64BIT */
194 #endif /* CONFIG_64BIT */
200 #else /* CONFIG_64BIT */
202 #endif /* CONFIG_64BIT */
204 : "=m" (idle
->idle_enter
), "=m" (vq
->idle
)
205 : "a" (&psw
), "a" (&idle
->idle_enter
),
206 "a" (&vq
->idle
), "a" (&vmax
), "m" (vmax
), "m" (psw
)
207 : "memory", "cc", "1");
210 * The inline assembly is equivalent to
211 * vq->idle = get_cpu_timer();
212 * idle->idle_enter = get_clock();
213 * __load_psw_mask(psw_kernel_bits | PSW_MASK_WAIT |
214 * PSW_MASK_IO | PSW_MASK_EXT);
215 * The difference is that the inline assembly makes sure that
216 * the last three instruction are stpt, stck and lpsw in that
217 * order. This is done to increase the precision.
224 #else /* CONFIG_64BIT */
227 #endif /* CONFIG_64BIT */
232 #else /* CONFIG_64BIT */
234 #endif /* CONFIG_64BIT */
236 : "=m" (idle
->idle_enter
), "=m" (vq
->idle
)
237 : "a" (&psw
), "a" (&idle
->idle_enter
),
238 "a" (&vq
->idle
), "m" (psw
)
239 : "memory", "cc", "1");
243 cputime64_t
s390_get_idle_time(int cpu
)
245 struct s390_idle_data
*idle
;
246 unsigned long long now
, idle_time
, idle_enter
;
247 unsigned int sequence
;
249 idle
= &per_cpu(s390_idle
, cpu
);
253 sequence
= idle
->sequence
;
258 idle_enter
= idle
->idle_enter
;
259 if (idle_enter
!= 0ULL && idle_enter
< now
)
260 idle_time
= now
- idle_enter
;
262 if (idle
->sequence
!= sequence
)
268 * Sorted add to a list. List is linear searched until first bigger
271 static void list_add_sorted(struct vtimer_list
*timer
, struct list_head
*head
)
273 struct vtimer_list
*event
;
275 list_for_each_entry(event
, head
, entry
) {
276 if (event
->expires
> timer
->expires
) {
277 list_add_tail(&timer
->entry
, &event
->entry
);
281 list_add_tail(&timer
->entry
, head
);
285 * Do the callback functions of expired vtimer events.
286 * Called from within the interrupt handler.
288 static void do_callbacks(struct list_head
*cb_list
)
290 struct vtimer_queue
*vq
;
291 struct vtimer_list
*event
, *tmp
;
293 if (list_empty(cb_list
))
296 vq
= &__get_cpu_var(virt_cpu_timer
);
298 list_for_each_entry_safe(event
, tmp
, cb_list
, entry
) {
299 list_del_init(&event
->entry
);
300 (event
->function
)(event
->data
);
301 if (event
->interval
) {
302 /* Recharge interval timer */
303 event
->expires
= event
->interval
+ vq
->elapsed
;
304 spin_lock(&vq
->lock
);
305 list_add_sorted(event
, &vq
->list
);
306 spin_unlock(&vq
->lock
);
312 * Handler for the virtual CPU timer.
314 static void do_cpu_timer_interrupt(__u16 error_code
)
316 struct vtimer_queue
*vq
;
317 struct vtimer_list
*event
, *tmp
;
318 struct list_head cb_list
; /* the callback queue */
321 INIT_LIST_HEAD(&cb_list
);
322 vq
= &__get_cpu_var(virt_cpu_timer
);
324 /* walk timer list, fire all expired events */
325 spin_lock(&vq
->lock
);
327 elapsed
= vq
->elapsed
+ (vq
->timer
- S390_lowcore
.async_enter_timer
);
328 BUG_ON((s64
) elapsed
< 0);
330 list_for_each_entry_safe(event
, tmp
, &vq
->list
, entry
) {
331 if (event
->expires
< elapsed
)
332 /* move expired timer to the callback queue */
333 list_move_tail(&event
->entry
, &cb_list
);
335 event
->expires
-= elapsed
;
337 spin_unlock(&vq
->lock
);
339 vq
->do_spt
= list_empty(&cb_list
);
340 do_callbacks(&cb_list
);
342 /* next event is first in list */
343 next
= VTIMER_MAX_SLICE
;
344 spin_lock(&vq
->lock
);
345 if (!list_empty(&vq
->list
)) {
346 event
= list_first_entry(&vq
->list
, struct vtimer_list
, entry
);
347 next
= event
->expires
;
350 spin_unlock(&vq
->lock
);
352 * To improve precision add the time spent by the
353 * interrupt handler to the elapsed time.
354 * Note: CPU timer counts down and we got an interrupt,
355 * the current content is negative
357 elapsed
= S390_lowcore
.async_enter_timer
- get_vtimer();
358 set_vtimer(next
- elapsed
);
359 vq
->timer
= next
- elapsed
;
360 vq
->elapsed
= elapsed
;
363 void init_virt_timer(struct vtimer_list
*timer
)
365 timer
->function
= NULL
;
366 INIT_LIST_HEAD(&timer
->entry
);
368 EXPORT_SYMBOL(init_virt_timer
);
370 static inline int vtimer_pending(struct vtimer_list
*timer
)
372 return (!list_empty(&timer
->entry
));
376 * this function should only run on the specified CPU
378 static void internal_add_vtimer(struct vtimer_list
*timer
)
380 struct vtimer_queue
*vq
;
384 vq
= &per_cpu(virt_cpu_timer
, timer
->cpu
);
385 spin_lock_irqsave(&vq
->lock
, flags
);
387 BUG_ON(timer
->cpu
!= smp_processor_id());
389 if (list_empty(&vq
->list
)) {
390 /* First timer on this cpu, just program it. */
391 list_add(&timer
->entry
, &vq
->list
);
392 set_vtimer(timer
->expires
);
393 vq
->timer
= timer
->expires
;
396 /* Check progress of old timers. */
397 expires
= timer
->expires
;
399 if (likely((s64
) expires
< (s64
) left
)) {
400 /* The new timer expires before the current timer. */
402 vq
->elapsed
+= vq
->timer
- left
;
405 vq
->elapsed
+= vq
->timer
- left
;
408 /* Insert new timer into per cpu list. */
409 timer
->expires
+= vq
->elapsed
;
410 list_add_sorted(timer
, &vq
->list
);
413 spin_unlock_irqrestore(&vq
->lock
, flags
);
414 /* release CPU acquired in prepare_vtimer or mod_virt_timer() */
418 static inline void prepare_vtimer(struct vtimer_list
*timer
)
420 BUG_ON(!timer
->function
);
421 BUG_ON(!timer
->expires
|| timer
->expires
> VTIMER_MAX_SLICE
);
422 BUG_ON(vtimer_pending(timer
));
423 timer
->cpu
= get_cpu();
427 * add_virt_timer - add an oneshot virtual CPU timer
429 void add_virt_timer(void *new)
431 struct vtimer_list
*timer
;
433 timer
= (struct vtimer_list
*)new;
434 prepare_vtimer(timer
);
436 internal_add_vtimer(timer
);
438 EXPORT_SYMBOL(add_virt_timer
);
441 * add_virt_timer_int - add an interval virtual CPU timer
443 void add_virt_timer_periodic(void *new)
445 struct vtimer_list
*timer
;
447 timer
= (struct vtimer_list
*)new;
448 prepare_vtimer(timer
);
449 timer
->interval
= timer
->expires
;
450 internal_add_vtimer(timer
);
452 EXPORT_SYMBOL(add_virt_timer_periodic
);
454 int __mod_vtimer(struct vtimer_list
*timer
, __u64 expires
, int periodic
)
456 struct vtimer_queue
*vq
;
460 BUG_ON(!timer
->function
);
461 BUG_ON(!expires
|| expires
> VTIMER_MAX_SLICE
);
463 if (timer
->expires
== expires
&& vtimer_pending(timer
))
467 vq
= &per_cpu(virt_cpu_timer
, cpu
);
469 /* disable interrupts before test if timer is pending */
470 spin_lock_irqsave(&vq
->lock
, flags
);
472 /* if timer isn't pending add it on the current CPU */
473 if (!vtimer_pending(timer
)) {
474 spin_unlock_irqrestore(&vq
->lock
, flags
);
477 timer
->interval
= expires
;
480 timer
->expires
= expires
;
482 internal_add_vtimer(timer
);
486 /* check if we run on the right CPU */
487 BUG_ON(timer
->cpu
!= cpu
);
489 list_del_init(&timer
->entry
);
490 timer
->expires
= expires
;
492 timer
->interval
= expires
;
494 /* the timer can't expire anymore so we can release the lock */
495 spin_unlock_irqrestore(&vq
->lock
, flags
);
496 internal_add_vtimer(timer
);
501 * If we change a pending timer the function must be called on the CPU
502 * where the timer is running on.
504 * returns whether it has modified a pending timer (1) or not (0)
506 int mod_virt_timer(struct vtimer_list
*timer
, __u64 expires
)
508 return __mod_vtimer(timer
, expires
, 0);
510 EXPORT_SYMBOL(mod_virt_timer
);
513 * If we change a pending timer the function must be called on the CPU
514 * where the timer is running on.
516 * returns whether it has modified a pending timer (1) or not (0)
518 int mod_virt_timer_periodic(struct vtimer_list
*timer
, __u64 expires
)
520 return __mod_vtimer(timer
, expires
, 1);
522 EXPORT_SYMBOL(mod_virt_timer_periodic
);
525 * delete a virtual timer
527 * returns whether the deleted timer was pending (1) or not (0)
529 int del_virt_timer(struct vtimer_list
*timer
)
532 struct vtimer_queue
*vq
;
534 /* check if timer is pending */
535 if (!vtimer_pending(timer
))
538 vq
= &per_cpu(virt_cpu_timer
, timer
->cpu
);
539 spin_lock_irqsave(&vq
->lock
, flags
);
541 /* we don't interrupt a running timer, just let it expire! */
542 list_del_init(&timer
->entry
);
544 spin_unlock_irqrestore(&vq
->lock
, flags
);
547 EXPORT_SYMBOL(del_virt_timer
);
550 * Start the virtual CPU timer on the current CPU.
552 void init_cpu_vtimer(void)
554 struct vtimer_queue
*vq
;
556 /* initialize per cpu vtimer structure */
557 vq
= &__get_cpu_var(virt_cpu_timer
);
558 INIT_LIST_HEAD(&vq
->list
);
559 spin_lock_init(&vq
->lock
);
561 /* enable cpu timer interrupts */
565 void __init
vtime_init(void)
567 /* request the cpu timer external interrupt */
568 if (register_external_interrupt(0x1005, do_cpu_timer_interrupt
))
569 panic("Couldn't request external interrupt 0x1005");
571 /* Enable cpu timer interrupts on the boot cpu. */