Loading...
1/*
2 * arch/s390/kernel/vtime.c
3 * Virtual cpu timer based timer functions.
4 *
5 * S390 version
6 * Copyright (C) 2004 IBM Deutschland Entwicklung GmbH, IBM Corporation
7 * Author(s): Jan Glauber <jan.glauber@de.ibm.com>
8 */
9
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>
24
25#include <asm/timer.h>
26#include <asm/irq_regs.h>
27#include <asm/cputime.h>
28#include <asm/irq.h>
29
30static DEFINE_PER_CPU(struct vtimer_queue, virt_cpu_timer);
31
32DEFINE_PER_CPU(struct s390_idle_data, s390_idle);
33
34static inline __u64 get_vtimer(void)
35{
36 __u64 timer;
37
38 asm volatile("STPT %0" : "=m" (timer));
39 return timer;
40}
41
42static inline void set_vtimer(__u64 expires)
43{
44 __u64 timer;
45
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;
51}
52
53/*
54 * Update process times based on virtual cpu times stored by entry.S
55 * to the lowcore fields user_timer, system_timer & steal_clock.
56 */
57static void do_account_vtime(struct task_struct *tsk, int hardirq_offset)
58{
59 struct thread_info *ti = task_thread_info(tsk);
60 __u64 timer, clock, user, system, steal;
61
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;
70
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);
75
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);
80
81 steal = S390_lowcore.steal_timer;
82 if ((s64) steal > 0) {
83 S390_lowcore.steal_timer = 0;
84 account_steal_time(steal);
85 }
86}
87
88void account_vtime(struct task_struct *prev, struct task_struct *next)
89{
90 struct thread_info *ti;
91
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;
99}
100
101void account_process_tick(struct task_struct *tsk, int user_tick)
102{
103 do_account_vtime(tsk, HARDIRQ_OFFSET);
104}
105
106/*
107 * Update process times based on virtual cpu times stored by entry.S
108 * to the lowcore fields user_timer, system_timer & steal_clock.
109 */
110void account_system_vtime(struct task_struct *tsk)
111{
112 struct thread_info *ti = task_thread_info(tsk);
113 __u64 timer, system;
114
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;
118
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);
123}
124EXPORT_SYMBOL_GPL(account_system_vtime);
125
126void __kprobes vtime_start_cpu(__u64 int_clock, __u64 enter_timer)
127{
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;
131
132 if (idle->idle_enter == 0ULL)
133 return;
134
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;
141
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;
145
146 /* Restart vtime CPU timer */
147 if (vq->do_spt) {
148 /* Program old expire value but first save progress. */
149 expires = vq->idle - enter_timer;
150 expires += get_vtimer();
151 set_vtimer(expires);
152 } else {
153 /* Don't account the CPU timer delta while the cpu was idle. */
154 vq->elapsed -= vq->idle - enter_timer;
155 }
156
157 idle->sequence++;
158 smp_wmb();
159 idle->idle_time += idle_time;
160 idle->idle_enter = 0ULL;
161 idle->idle_count++;
162 smp_wmb();
163 idle->sequence++;
164}
165
166void __kprobes vtime_stop_cpu(void)
167{
168 struct s390_idle_data *idle = &__get_cpu_var(s390_idle);
169 struct vtimer_queue *vq = &__get_cpu_var(virt_cpu_timer);
170 psw_t psw;
171
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;
174
175 idle->nohz_delay = 0;
176
177 /* Check if the CPU timer needs to be reprogrammed. */
178 if (vq->do_spt) {
179 __u64 vmax = VTIMER_MAX_SLICE;
180 /*
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.
190 */
191 asm volatile(
192#ifndef CONFIG_64BIT
193 " basr 1,0\n"
194 "0: ahi 1,1f-0b\n"
195 " st 1,4(%2)\n"
196#else /* CONFIG_64BIT */
197 " larl 1,1f\n"
198 " stg 1,8(%2)\n"
199#endif /* CONFIG_64BIT */
200 " stpt 0(%4)\n"
201 " spt 0(%5)\n"
202 " stck 0(%3)\n"
203#ifndef CONFIG_64BIT
204 " lpsw 0(%2)\n"
205#else /* CONFIG_64BIT */
206 " lpswe 0(%2)\n"
207#endif /* CONFIG_64BIT */
208 "1:"
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");
213 } else {
214 /*
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.
223 */
224 asm volatile(
225#ifndef CONFIG_64BIT
226 " basr 1,0\n"
227 "0: ahi 1,1f-0b\n"
228 " st 1,4(%2)\n"
229#else /* CONFIG_64BIT */
230 " larl 1,1f\n"
231 " stg 1,8(%2)\n"
232#endif /* CONFIG_64BIT */
233 " stpt 0(%4)\n"
234 " stck 0(%3)\n"
235#ifndef CONFIG_64BIT
236 " lpsw 0(%2)\n"
237#else /* CONFIG_64BIT */
238 " lpswe 0(%2)\n"
239#endif /* CONFIG_64BIT */
240 "1:"
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");
245 }
246}
247
248cputime64_t s390_get_idle_time(int cpu)
249{
250 struct s390_idle_data *idle;
251 unsigned long long now, idle_time, idle_enter;
252 unsigned int sequence;
253
254 idle = &per_cpu(s390_idle, cpu);
255
256 now = get_clock();
257repeat:
258 sequence = idle->sequence;
259 smp_rmb();
260 if (sequence & 1)
261 goto repeat;
262 idle_time = 0;
263 idle_enter = idle->idle_enter;
264 if (idle_enter != 0ULL && idle_enter < now)
265 idle_time = now - idle_enter;
266 smp_rmb();
267 if (idle->sequence != sequence)
268 goto repeat;
269 return idle_time;
270}
271
272/*
273 * Sorted add to a list. List is linear searched until first bigger
274 * element is found.
275 */
276static void list_add_sorted(struct vtimer_list *timer, struct list_head *head)
277{
278 struct vtimer_list *event;
279
280 list_for_each_entry(event, head, entry) {
281 if (event->expires > timer->expires) {
282 list_add_tail(&timer->entry, &event->entry);
283 return;
284 }
285 }
286 list_add_tail(&timer->entry, head);
287}
288
289/*
290 * Do the callback functions of expired vtimer events.
291 * Called from within the interrupt handler.
292 */
293static void do_callbacks(struct list_head *cb_list)
294{
295 struct vtimer_queue *vq;
296 struct vtimer_list *event, *tmp;
297
298 if (list_empty(cb_list))
299 return;
300
301 vq = &__get_cpu_var(virt_cpu_timer);
302
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);
312 }
313 }
314}
315
316/*
317 * Handler for the virtual CPU timer.
318 */
319static void do_cpu_timer_interrupt(unsigned int ext_int_code,
320 unsigned int param32, unsigned long param64)
321{
322 struct vtimer_queue *vq;
323 struct vtimer_list *event, *tmp;
324 struct list_head cb_list; /* the callback queue */
325 __u64 elapsed, next;
326
327 kstat_cpu(smp_processor_id()).irqs[EXTINT_TMR]++;
328 INIT_LIST_HEAD(&cb_list);
329 vq = &__get_cpu_var(virt_cpu_timer);
330
331 /* walk timer list, fire all expired events */
332 spin_lock(&vq->lock);
333
334 elapsed = vq->elapsed + (vq->timer - S390_lowcore.async_enter_timer);
335 BUG_ON((s64) elapsed < 0);
336 vq->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);
341 else
342 event->expires -= elapsed;
343 }
344 spin_unlock(&vq->lock);
345
346 vq->do_spt = list_empty(&cb_list);
347 do_callbacks(&cb_list);
348
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;
355 } else
356 vq->do_spt = 0;
357 spin_unlock(&vq->lock);
358 /*
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
363 */
364 elapsed = S390_lowcore.async_enter_timer - get_vtimer();
365 set_vtimer(next - elapsed);
366 vq->timer = next - elapsed;
367 vq->elapsed = elapsed;
368}
369
370void init_virt_timer(struct vtimer_list *timer)
371{
372 timer->function = NULL;
373 INIT_LIST_HEAD(&timer->entry);
374}
375EXPORT_SYMBOL(init_virt_timer);
376
377static inline int vtimer_pending(struct vtimer_list *timer)
378{
379 return (!list_empty(&timer->entry));
380}
381
382/*
383 * this function should only run on the specified CPU
384 */
385static void internal_add_vtimer(struct vtimer_list *timer)
386{
387 struct vtimer_queue *vq;
388 unsigned long flags;
389 __u64 left, expires;
390
391 vq = &per_cpu(virt_cpu_timer, timer->cpu);
392 spin_lock_irqsave(&vq->lock, flags);
393
394 BUG_ON(timer->cpu != smp_processor_id());
395
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;
401 vq->elapsed = 0;
402 } else {
403 /* Check progress of old timers. */
404 expires = timer->expires;
405 left = get_vtimer();
406 if (likely((s64) expires < (s64) left)) {
407 /* The new timer expires before the current timer. */
408 set_vtimer(expires);
409 vq->elapsed += vq->timer - left;
410 vq->timer = expires;
411 } else {
412 vq->elapsed += vq->timer - left;
413 vq->timer = left;
414 }
415 /* Insert new timer into per cpu list. */
416 timer->expires += vq->elapsed;
417 list_add_sorted(timer, &vq->list);
418 }
419
420 spin_unlock_irqrestore(&vq->lock, flags);
421 /* release CPU acquired in prepare_vtimer or mod_virt_timer() */
422 put_cpu();
423}
424
425static inline void prepare_vtimer(struct vtimer_list *timer)
426{
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();
431}
432
433/*
434 * add_virt_timer - add an oneshot virtual CPU timer
435 */
436void add_virt_timer(void *new)
437{
438 struct vtimer_list *timer;
439
440 timer = (struct vtimer_list *)new;
441 prepare_vtimer(timer);
442 timer->interval = 0;
443 internal_add_vtimer(timer);
444}
445EXPORT_SYMBOL(add_virt_timer);
446
447/*
448 * add_virt_timer_int - add an interval virtual CPU timer
449 */
450void add_virt_timer_periodic(void *new)
451{
452 struct vtimer_list *timer;
453
454 timer = (struct vtimer_list *)new;
455 prepare_vtimer(timer);
456 timer->interval = timer->expires;
457 internal_add_vtimer(timer);
458}
459EXPORT_SYMBOL(add_virt_timer_periodic);
460
461int __mod_vtimer(struct vtimer_list *timer, __u64 expires, int periodic)
462{
463 struct vtimer_queue *vq;
464 unsigned long flags;
465 int cpu;
466
467 BUG_ON(!timer->function);
468 BUG_ON(!expires || expires > VTIMER_MAX_SLICE);
469
470 if (timer->expires == expires && vtimer_pending(timer))
471 return 1;
472
473 cpu = get_cpu();
474 vq = &per_cpu(virt_cpu_timer, cpu);
475
476 /* disable interrupts before test if timer is pending */
477 spin_lock_irqsave(&vq->lock, flags);
478
479 /* if timer isn't pending add it on the current CPU */
480 if (!vtimer_pending(timer)) {
481 spin_unlock_irqrestore(&vq->lock, flags);
482
483 if (periodic)
484 timer->interval = expires;
485 else
486 timer->interval = 0;
487 timer->expires = expires;
488 timer->cpu = cpu;
489 internal_add_vtimer(timer);
490 return 0;
491 }
492
493 /* check if we run on the right CPU */
494 BUG_ON(timer->cpu != cpu);
495
496 list_del_init(&timer->entry);
497 timer->expires = expires;
498 if (periodic)
499 timer->interval = expires;
500
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);
504 return 1;
505}
506
507/*
508 * If we change a pending timer the function must be called on the CPU
509 * where the timer is running on.
510 *
511 * returns whether it has modified a pending timer (1) or not (0)
512 */
513int mod_virt_timer(struct vtimer_list *timer, __u64 expires)
514{
515 return __mod_vtimer(timer, expires, 0);
516}
517EXPORT_SYMBOL(mod_virt_timer);
518
519/*
520 * If we change a pending timer the function must be called on the CPU
521 * where the timer is running on.
522 *
523 * returns whether it has modified a pending timer (1) or not (0)
524 */
525int mod_virt_timer_periodic(struct vtimer_list *timer, __u64 expires)
526{
527 return __mod_vtimer(timer, expires, 1);
528}
529EXPORT_SYMBOL(mod_virt_timer_periodic);
530
531/*
532 * delete a virtual timer
533 *
534 * returns whether the deleted timer was pending (1) or not (0)
535 */
536int del_virt_timer(struct vtimer_list *timer)
537{
538 unsigned long flags;
539 struct vtimer_queue *vq;
540
541 /* check if timer is pending */
542 if (!vtimer_pending(timer))
543 return 0;
544
545 vq = &per_cpu(virt_cpu_timer, timer->cpu);
546 spin_lock_irqsave(&vq->lock, flags);
547
548 /* we don't interrupt a running timer, just let it expire! */
549 list_del_init(&timer->entry);
550
551 spin_unlock_irqrestore(&vq->lock, flags);
552 return 1;
553}
554EXPORT_SYMBOL(del_virt_timer);
555
556/*
557 * Start the virtual CPU timer on the current CPU.
558 */
559void init_cpu_vtimer(void)
560{
561 struct vtimer_queue *vq;
562
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);
567
568 /* enable cpu timer interrupts */
569 __ctl_set_bit(0,10);
570}
571
572static int __cpuinit s390_nohz_notify(struct notifier_block *self,
573 unsigned long action, void *hcpu)
574{
575 struct s390_idle_data *idle;
576 long cpu = (long) hcpu;
577
578 idle = &per_cpu(s390_idle, cpu);
579 switch (action) {
580 case CPU_DYING:
581 case CPU_DYING_FROZEN:
582 idle->nohz_delay = 0;
583 default:
584 break;
585 }
586 return NOTIFY_OK;
587}
588
589void __init vtime_init(void)
590{
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");
594
595 /* Enable cpu timer interrupts on the boot cpu. */
596 init_cpu_vtimer();
597 cpu_notifier(s390_nohz_notify, 0);
598}
599
1/*
2 * arch/s390/kernel/vtime.c
3 * Virtual cpu timer based timer functions.
4 *
5 * S390 version
6 * Copyright (C) 2004 IBM Deutschland Entwicklung GmbH, IBM Corporation
7 * Author(s): Jan Glauber <jan.glauber@de.ibm.com>
8 */
9
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>
24
25#include <asm/timer.h>
26#include <asm/irq_regs.h>
27#include <asm/cputime.h>
28#include <asm/irq.h>
29#include "entry.h"
30
31static DEFINE_PER_CPU(struct vtimer_queue, virt_cpu_timer);
32
33DEFINE_PER_CPU(struct s390_idle_data, s390_idle);
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 (" STPT %0\n" /* Store current cpu timer value */
48 " SPT %1" /* Set new value immediately afterwards */
49 : "=m" (timer) : "m" (expires) );
50 S390_lowcore.system_timer += S390_lowcore.last_update_timer - timer;
51 S390_lowcore.last_update_timer = expires;
52}
53
54/*
55 * Update process times based on virtual cpu times stored by entry.S
56 * to the lowcore fields user_timer, system_timer & steal_clock.
57 */
58static void do_account_vtime(struct task_struct *tsk, int hardirq_offset)
59{
60 struct thread_info *ti = task_thread_info(tsk);
61 __u64 timer, clock, user, system, steal;
62
63 timer = S390_lowcore.last_update_timer;
64 clock = S390_lowcore.last_update_clock;
65 asm volatile (" STPT %0\n" /* Store current cpu timer value */
66 " STCK %1" /* Store current tod clock value */
67 : "=m" (S390_lowcore.last_update_timer),
68 "=m" (S390_lowcore.last_update_clock) );
69 S390_lowcore.system_timer += timer - S390_lowcore.last_update_timer;
70 S390_lowcore.steal_timer += S390_lowcore.last_update_clock - clock;
71
72 user = S390_lowcore.user_timer - ti->user_timer;
73 S390_lowcore.steal_timer -= user;
74 ti->user_timer = S390_lowcore.user_timer;
75 account_user_time(tsk, user, user);
76
77 system = S390_lowcore.system_timer - ti->system_timer;
78 S390_lowcore.steal_timer -= system;
79 ti->system_timer = S390_lowcore.system_timer;
80 account_system_time(tsk, hardirq_offset, system, system);
81
82 steal = S390_lowcore.steal_timer;
83 if ((s64) steal > 0) {
84 S390_lowcore.steal_timer = 0;
85 account_steal_time(steal);
86 }
87}
88
89void account_vtime(struct task_struct *prev, struct task_struct *next)
90{
91 struct thread_info *ti;
92
93 do_account_vtime(prev, 0);
94 ti = task_thread_info(prev);
95 ti->user_timer = S390_lowcore.user_timer;
96 ti->system_timer = S390_lowcore.system_timer;
97 ti = task_thread_info(next);
98 S390_lowcore.user_timer = ti->user_timer;
99 S390_lowcore.system_timer = ti->system_timer;
100}
101
102void account_process_tick(struct task_struct *tsk, int user_tick)
103{
104 do_account_vtime(tsk, HARDIRQ_OFFSET);
105}
106
107/*
108 * Update process times based on virtual cpu times stored by entry.S
109 * to the lowcore fields user_timer, system_timer & steal_clock.
110 */
111void account_system_vtime(struct task_struct *tsk)
112{
113 struct thread_info *ti = task_thread_info(tsk);
114 __u64 timer, system;
115
116 timer = S390_lowcore.last_update_timer;
117 S390_lowcore.last_update_timer = get_vtimer();
118 S390_lowcore.system_timer += timer - S390_lowcore.last_update_timer;
119
120 system = S390_lowcore.system_timer - ti->system_timer;
121 S390_lowcore.steal_timer -= system;
122 ti->system_timer = S390_lowcore.system_timer;
123 account_system_time(tsk, 0, system, system);
124}
125EXPORT_SYMBOL_GPL(account_system_vtime);
126
127void __kprobes vtime_stop_cpu(void)
128{
129 struct s390_idle_data *idle = &__get_cpu_var(s390_idle);
130 struct vtimer_queue *vq = &__get_cpu_var(virt_cpu_timer);
131 unsigned long long idle_time;
132 unsigned long psw_mask;
133
134 trace_hardirqs_on();
135 /* Don't trace preempt off for idle. */
136 stop_critical_timings();
137
138 /* Wait for external, I/O or machine check interrupt. */
139 psw_mask = psw_kernel_bits | PSW_MASK_WAIT | PSW_MASK_DAT |
140 PSW_MASK_IO | PSW_MASK_EXT | PSW_MASK_MCHECK;
141 idle->nohz_delay = 0;
142
143 /* Call the assembler magic in entry.S */
144 psw_idle(idle, vq, psw_mask, !list_empty(&vq->list));
145
146 /* Reenable preemption tracer. */
147 start_critical_timings();
148
149 /* Account time spent with enabled wait psw loaded as idle time. */
150 idle->sequence++;
151 smp_wmb();
152 idle_time = idle->idle_exit - idle->idle_enter;
153 idle->idle_time += idle_time;
154 idle->idle_enter = idle->idle_exit = 0ULL;
155 idle->idle_count++;
156 account_idle_time(idle_time);
157 smp_wmb();
158 idle->sequence++;
159}
160
161cputime64_t s390_get_idle_time(int cpu)
162{
163 struct s390_idle_data *idle = &per_cpu(s390_idle, cpu);
164 unsigned long long now, idle_enter, idle_exit;
165 unsigned int sequence;
166
167 do {
168 now = get_clock();
169 sequence = ACCESS_ONCE(idle->sequence);
170 idle_enter = ACCESS_ONCE(idle->idle_enter);
171 idle_exit = ACCESS_ONCE(idle->idle_exit);
172 } while ((sequence & 1) || (idle->sequence != sequence));
173 return idle_enter ? ((idle_exit ? : now) - idle_enter) : 0;
174}
175
176/*
177 * Sorted add to a list. List is linear searched until first bigger
178 * element is found.
179 */
180static void list_add_sorted(struct vtimer_list *timer, struct list_head *head)
181{
182 struct vtimer_list *event;
183
184 list_for_each_entry(event, head, entry) {
185 if (event->expires > timer->expires) {
186 list_add_tail(&timer->entry, &event->entry);
187 return;
188 }
189 }
190 list_add_tail(&timer->entry, head);
191}
192
193/*
194 * Do the callback functions of expired vtimer events.
195 * Called from within the interrupt handler.
196 */
197static void do_callbacks(struct list_head *cb_list)
198{
199 struct vtimer_queue *vq;
200 struct vtimer_list *event, *tmp;
201
202 if (list_empty(cb_list))
203 return;
204
205 vq = &__get_cpu_var(virt_cpu_timer);
206
207 list_for_each_entry_safe(event, tmp, cb_list, entry) {
208 list_del_init(&event->entry);
209 (event->function)(event->data);
210 if (event->interval) {
211 /* Recharge interval timer */
212 event->expires = event->interval + vq->elapsed;
213 spin_lock(&vq->lock);
214 list_add_sorted(event, &vq->list);
215 spin_unlock(&vq->lock);
216 }
217 }
218}
219
220/*
221 * Handler for the virtual CPU timer.
222 */
223static void do_cpu_timer_interrupt(struct ext_code ext_code,
224 unsigned int param32, unsigned long param64)
225{
226 struct vtimer_queue *vq;
227 struct vtimer_list *event, *tmp;
228 struct list_head cb_list; /* the callback queue */
229 __u64 elapsed, next;
230
231 kstat_cpu(smp_processor_id()).irqs[EXTINT_TMR]++;
232 INIT_LIST_HEAD(&cb_list);
233 vq = &__get_cpu_var(virt_cpu_timer);
234
235 /* walk timer list, fire all expired events */
236 spin_lock(&vq->lock);
237
238 elapsed = vq->elapsed + (vq->timer - S390_lowcore.async_enter_timer);
239 BUG_ON((s64) elapsed < 0);
240 vq->elapsed = 0;
241 list_for_each_entry_safe(event, tmp, &vq->list, entry) {
242 if (event->expires < elapsed)
243 /* move expired timer to the callback queue */
244 list_move_tail(&event->entry, &cb_list);
245 else
246 event->expires -= elapsed;
247 }
248 spin_unlock(&vq->lock);
249
250 do_callbacks(&cb_list);
251
252 /* next event is first in list */
253 next = VTIMER_MAX_SLICE;
254 spin_lock(&vq->lock);
255 if (!list_empty(&vq->list)) {
256 event = list_first_entry(&vq->list, struct vtimer_list, entry);
257 next = event->expires;
258 }
259 spin_unlock(&vq->lock);
260 /*
261 * To improve precision add the time spent by the
262 * interrupt handler to the elapsed time.
263 * Note: CPU timer counts down and we got an interrupt,
264 * the current content is negative
265 */
266 elapsed = S390_lowcore.async_enter_timer - get_vtimer();
267 set_vtimer(next - elapsed);
268 vq->timer = next - elapsed;
269 vq->elapsed = elapsed;
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
284/*
285 * this function should only run on the specified CPU
286 */
287static void internal_add_vtimer(struct vtimer_list *timer)
288{
289 struct vtimer_queue *vq;
290 unsigned long flags;
291 __u64 left, expires;
292
293 vq = &per_cpu(virt_cpu_timer, timer->cpu);
294 spin_lock_irqsave(&vq->lock, flags);
295
296 BUG_ON(timer->cpu != smp_processor_id());
297
298 if (list_empty(&vq->list)) {
299 /* First timer on this cpu, just program it. */
300 list_add(&timer->entry, &vq->list);
301 set_vtimer(timer->expires);
302 vq->timer = timer->expires;
303 vq->elapsed = 0;
304 } else {
305 /* Check progress of old timers. */
306 expires = timer->expires;
307 left = get_vtimer();
308 if (likely((s64) expires < (s64) left)) {
309 /* The new timer expires before the current timer. */
310 set_vtimer(expires);
311 vq->elapsed += vq->timer - left;
312 vq->timer = expires;
313 } else {
314 vq->elapsed += vq->timer - left;
315 vq->timer = left;
316 }
317 /* Insert new timer into per cpu list. */
318 timer->expires += vq->elapsed;
319 list_add_sorted(timer, &vq->list);
320 }
321
322 spin_unlock_irqrestore(&vq->lock, flags);
323 /* release CPU acquired in prepare_vtimer or mod_virt_timer() */
324 put_cpu();
325}
326
327static inline void prepare_vtimer(struct vtimer_list *timer)
328{
329 BUG_ON(!timer->function);
330 BUG_ON(!timer->expires || timer->expires > VTIMER_MAX_SLICE);
331 BUG_ON(vtimer_pending(timer));
332 timer->cpu = get_cpu();
333}
334
335/*
336 * add_virt_timer - add an oneshot virtual CPU timer
337 */
338void add_virt_timer(void *new)
339{
340 struct vtimer_list *timer;
341
342 timer = (struct vtimer_list *)new;
343 prepare_vtimer(timer);
344 timer->interval = 0;
345 internal_add_vtimer(timer);
346}
347EXPORT_SYMBOL(add_virt_timer);
348
349/*
350 * add_virt_timer_int - add an interval virtual CPU timer
351 */
352void add_virt_timer_periodic(void *new)
353{
354 struct vtimer_list *timer;
355
356 timer = (struct vtimer_list *)new;
357 prepare_vtimer(timer);
358 timer->interval = timer->expires;
359 internal_add_vtimer(timer);
360}
361EXPORT_SYMBOL(add_virt_timer_periodic);
362
363static int __mod_vtimer(struct vtimer_list *timer, __u64 expires, int periodic)
364{
365 struct vtimer_queue *vq;
366 unsigned long flags;
367 int cpu;
368
369 BUG_ON(!timer->function);
370 BUG_ON(!expires || expires > VTIMER_MAX_SLICE);
371
372 if (timer->expires == expires && vtimer_pending(timer))
373 return 1;
374
375 cpu = get_cpu();
376 vq = &per_cpu(virt_cpu_timer, cpu);
377
378 /* disable interrupts before test if timer is pending */
379 spin_lock_irqsave(&vq->lock, flags);
380
381 /* if timer isn't pending add it on the current CPU */
382 if (!vtimer_pending(timer)) {
383 spin_unlock_irqrestore(&vq->lock, flags);
384
385 if (periodic)
386 timer->interval = expires;
387 else
388 timer->interval = 0;
389 timer->expires = expires;
390 timer->cpu = cpu;
391 internal_add_vtimer(timer);
392 return 0;
393 }
394
395 /* check if we run on the right CPU */
396 BUG_ON(timer->cpu != cpu);
397
398 list_del_init(&timer->entry);
399 timer->expires = expires;
400 if (periodic)
401 timer->interval = expires;
402
403 /* the timer can't expire anymore so we can release the lock */
404 spin_unlock_irqrestore(&vq->lock, flags);
405 internal_add_vtimer(timer);
406 return 1;
407}
408
409/*
410 * If we change a pending timer the function must be called on the CPU
411 * where the timer is running on.
412 *
413 * returns whether it has modified a pending timer (1) or not (0)
414 */
415int mod_virt_timer(struct vtimer_list *timer, __u64 expires)
416{
417 return __mod_vtimer(timer, expires, 0);
418}
419EXPORT_SYMBOL(mod_virt_timer);
420
421/*
422 * If we change a pending timer the function must be called on the CPU
423 * where the timer is running on.
424 *
425 * returns whether it has modified a pending timer (1) or not (0)
426 */
427int mod_virt_timer_periodic(struct vtimer_list *timer, __u64 expires)
428{
429 return __mod_vtimer(timer, expires, 1);
430}
431EXPORT_SYMBOL(mod_virt_timer_periodic);
432
433/*
434 * delete a virtual timer
435 *
436 * returns whether the deleted timer was pending (1) or not (0)
437 */
438int del_virt_timer(struct vtimer_list *timer)
439{
440 unsigned long flags;
441 struct vtimer_queue *vq;
442
443 /* check if timer is pending */
444 if (!vtimer_pending(timer))
445 return 0;
446
447 vq = &per_cpu(virt_cpu_timer, timer->cpu);
448 spin_lock_irqsave(&vq->lock, flags);
449
450 /* we don't interrupt a running timer, just let it expire! */
451 list_del_init(&timer->entry);
452
453 spin_unlock_irqrestore(&vq->lock, flags);
454 return 1;
455}
456EXPORT_SYMBOL(del_virt_timer);
457
458/*
459 * Start the virtual CPU timer on the current CPU.
460 */
461void init_cpu_vtimer(void)
462{
463 struct vtimer_queue *vq;
464
465 /* initialize per cpu vtimer structure */
466 vq = &__get_cpu_var(virt_cpu_timer);
467 INIT_LIST_HEAD(&vq->list);
468 spin_lock_init(&vq->lock);
469
470 /* enable cpu timer interrupts */
471 __ctl_set_bit(0,10);
472
473 /* set initial cpu timer */
474 set_vtimer(0x7fffffffffffffffULL);
475}
476
477static int __cpuinit s390_nohz_notify(struct notifier_block *self,
478 unsigned long action, void *hcpu)
479{
480 struct s390_idle_data *idle;
481 long cpu = (long) hcpu;
482
483 idle = &per_cpu(s390_idle, cpu);
484 switch (action) {
485 case CPU_DYING:
486 case CPU_DYING_FROZEN:
487 idle->nohz_delay = 0;
488 default:
489 break;
490 }
491 return NOTIFY_OK;
492}
493
494void __init vtime_init(void)
495{
496 /* request the cpu timer external interrupt */
497 if (register_external_interrupt(0x1005, do_cpu_timer_interrupt))
498 panic("Couldn't request external interrupt 0x1005");
499
500 /* Enable cpu timer interrupts on the boot cpu. */
501 init_cpu_vtimer();
502 cpu_notifier(s390_nohz_notify, 0);
503}
504