Loading...
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 void set_vtimer(u64 expires)
37{
38 struct lowcore *lc = get_lowcore();
39 u64 timer;
40
41 asm volatile(
42 " stpt %0\n" /* Store current cpu timer value */
43 " spt %1" /* Set new value imm. afterwards */
44 : "=Q" (timer) : "Q" (expires));
45 lc->system_timer += lc->last_update_timer - timer;
46 lc->last_update_timer = expires;
47}
48
49static inline int virt_timer_forward(u64 elapsed)
50{
51 BUG_ON(!irqs_disabled());
52
53 if (list_empty(&virt_timer_list))
54 return 0;
55 elapsed = atomic64_add_return(elapsed, &virt_timer_elapsed);
56 return elapsed >= atomic64_read(&virt_timer_current);
57}
58
59static void update_mt_scaling(void)
60{
61 u64 cycles_new[8], *cycles_old;
62 u64 delta, fac, mult, div;
63 int i;
64
65 stcctm(MT_DIAG, smp_cpu_mtid + 1, cycles_new);
66 cycles_old = this_cpu_ptr(mt_cycles);
67 fac = 1;
68 mult = div = 0;
69 for (i = 0; i <= smp_cpu_mtid; i++) {
70 delta = cycles_new[i] - cycles_old[i];
71 div += delta;
72 mult *= i + 1;
73 mult += delta * fac;
74 fac *= i + 1;
75 }
76 div *= fac;
77 if (div > 0) {
78 /* Update scaling factor */
79 __this_cpu_write(mt_scaling_mult, mult);
80 __this_cpu_write(mt_scaling_div, div);
81 memcpy(cycles_old, cycles_new,
82 sizeof(u64) * (smp_cpu_mtid + 1));
83 }
84 __this_cpu_write(mt_scaling_jiffies, jiffies_64);
85}
86
87static inline u64 update_tsk_timer(unsigned long *tsk_vtime, u64 new)
88{
89 u64 delta;
90
91 delta = new - *tsk_vtime;
92 *tsk_vtime = new;
93 return delta;
94}
95
96
97static inline u64 scale_vtime(u64 vtime)
98{
99 u64 mult = __this_cpu_read(mt_scaling_mult);
100 u64 div = __this_cpu_read(mt_scaling_div);
101
102 if (smp_cpu_mtid)
103 return vtime * mult / div;
104 return vtime;
105}
106
107static void account_system_index_scaled(struct task_struct *p, u64 cputime,
108 enum cpu_usage_stat index)
109{
110 p->stimescaled += cputime_to_nsecs(scale_vtime(cputime));
111 account_system_index_time(p, cputime_to_nsecs(cputime), index);
112}
113
114/*
115 * Update process times based on virtual cpu times stored by entry.S
116 * to the lowcore fields user_timer, system_timer & steal_clock.
117 */
118static int do_account_vtime(struct task_struct *tsk)
119{
120 u64 timer, clock, user, guest, system, hardirq, softirq;
121 struct lowcore *lc = get_lowcore();
122
123 timer = lc->last_update_timer;
124 clock = lc->last_update_clock;
125 asm volatile(
126 " stpt %0\n" /* Store current cpu timer value */
127 " stckf %1" /* Store current tod clock value */
128 : "=Q" (lc->last_update_timer),
129 "=Q" (lc->last_update_clock)
130 : : "cc");
131 clock = lc->last_update_clock - clock;
132 timer -= lc->last_update_timer;
133
134 if (hardirq_count())
135 lc->hardirq_timer += timer;
136 else
137 lc->system_timer += timer;
138
139 /* Update MT utilization calculation */
140 if (smp_cpu_mtid &&
141 time_after64(jiffies_64, this_cpu_read(mt_scaling_jiffies)))
142 update_mt_scaling();
143
144 /* Calculate cputime delta */
145 user = update_tsk_timer(&tsk->thread.user_timer,
146 READ_ONCE(lc->user_timer));
147 guest = update_tsk_timer(&tsk->thread.guest_timer,
148 READ_ONCE(lc->guest_timer));
149 system = update_tsk_timer(&tsk->thread.system_timer,
150 READ_ONCE(lc->system_timer));
151 hardirq = update_tsk_timer(&tsk->thread.hardirq_timer,
152 READ_ONCE(lc->hardirq_timer));
153 softirq = update_tsk_timer(&tsk->thread.softirq_timer,
154 READ_ONCE(lc->softirq_timer));
155 lc->steal_timer +=
156 clock - user - guest - system - hardirq - softirq;
157
158 /* Push account value */
159 if (user) {
160 account_user_time(tsk, cputime_to_nsecs(user));
161 tsk->utimescaled += cputime_to_nsecs(scale_vtime(user));
162 }
163
164 if (guest) {
165 account_guest_time(tsk, cputime_to_nsecs(guest));
166 tsk->utimescaled += cputime_to_nsecs(scale_vtime(guest));
167 }
168
169 if (system)
170 account_system_index_scaled(tsk, system, CPUTIME_SYSTEM);
171 if (hardirq)
172 account_system_index_scaled(tsk, hardirq, CPUTIME_IRQ);
173 if (softirq)
174 account_system_index_scaled(tsk, softirq, CPUTIME_SOFTIRQ);
175
176 return virt_timer_forward(user + guest + system + hardirq + softirq);
177}
178
179void vtime_task_switch(struct task_struct *prev)
180{
181 struct lowcore *lc = get_lowcore();
182
183 do_account_vtime(prev);
184 prev->thread.user_timer = lc->user_timer;
185 prev->thread.guest_timer = lc->guest_timer;
186 prev->thread.system_timer = lc->system_timer;
187 prev->thread.hardirq_timer = lc->hardirq_timer;
188 prev->thread.softirq_timer = lc->softirq_timer;
189 lc->user_timer = current->thread.user_timer;
190 lc->guest_timer = current->thread.guest_timer;
191 lc->system_timer = current->thread.system_timer;
192 lc->hardirq_timer = current->thread.hardirq_timer;
193 lc->softirq_timer = current->thread.softirq_timer;
194}
195
196/*
197 * In s390, accounting pending user time also implies
198 * accounting system time in order to correctly compute
199 * the stolen time accounting.
200 */
201void vtime_flush(struct task_struct *tsk)
202{
203 struct lowcore *lc = get_lowcore();
204 u64 steal, avg_steal;
205
206 if (do_account_vtime(tsk))
207 virt_timer_expire();
208
209 steal = lc->steal_timer;
210 avg_steal = lc->avg_steal_timer;
211 if ((s64) steal > 0) {
212 lc->steal_timer = 0;
213 account_steal_time(cputime_to_nsecs(steal));
214 avg_steal += steal;
215 }
216 lc->avg_steal_timer = avg_steal / 2;
217}
218
219static u64 vtime_delta(void)
220{
221 struct lowcore *lc = get_lowcore();
222 u64 timer = lc->last_update_timer;
223
224 lc->last_update_timer = get_cpu_timer();
225 return timer - lc->last_update_timer;
226}
227
228/*
229 * Update process times based on virtual cpu times stored by entry.S
230 * to the lowcore fields user_timer, system_timer & steal_clock.
231 */
232void vtime_account_kernel(struct task_struct *tsk)
233{
234 struct lowcore *lc = get_lowcore();
235 u64 delta = vtime_delta();
236
237 if (tsk->flags & PF_VCPU)
238 lc->guest_timer += delta;
239 else
240 lc->system_timer += delta;
241
242 virt_timer_forward(delta);
243}
244EXPORT_SYMBOL_GPL(vtime_account_kernel);
245
246void vtime_account_softirq(struct task_struct *tsk)
247{
248 u64 delta = vtime_delta();
249
250 get_lowcore()->softirq_timer += delta;
251
252 virt_timer_forward(delta);
253}
254
255void vtime_account_hardirq(struct task_struct *tsk)
256{
257 u64 delta = vtime_delta();
258
259 get_lowcore()->hardirq_timer += delta;
260
261 virt_timer_forward(delta);
262}
263
264/*
265 * Sorted add to a list. List is linear searched until first bigger
266 * element is found.
267 */
268static void list_add_sorted(struct vtimer_list *timer, struct list_head *head)
269{
270 struct vtimer_list *tmp;
271
272 list_for_each_entry(tmp, head, entry) {
273 if (tmp->expires > timer->expires) {
274 list_add_tail(&timer->entry, &tmp->entry);
275 return;
276 }
277 }
278 list_add_tail(&timer->entry, head);
279}
280
281/*
282 * Handler for expired virtual CPU timer.
283 */
284static void virt_timer_expire(void)
285{
286 struct vtimer_list *timer, *tmp;
287 unsigned long elapsed;
288 LIST_HEAD(cb_list);
289
290 /* walk timer list, fire all expired timers */
291 spin_lock(&virt_timer_lock);
292 elapsed = atomic64_read(&virt_timer_elapsed);
293 list_for_each_entry_safe(timer, tmp, &virt_timer_list, entry) {
294 if (timer->expires < elapsed)
295 /* move expired timer to the callback queue */
296 list_move_tail(&timer->entry, &cb_list);
297 else
298 timer->expires -= elapsed;
299 }
300 if (!list_empty(&virt_timer_list)) {
301 timer = list_first_entry(&virt_timer_list,
302 struct vtimer_list, entry);
303 atomic64_set(&virt_timer_current, timer->expires);
304 }
305 atomic64_sub(elapsed, &virt_timer_elapsed);
306 spin_unlock(&virt_timer_lock);
307
308 /* Do callbacks and recharge periodic timers */
309 list_for_each_entry_safe(timer, tmp, &cb_list, entry) {
310 list_del_init(&timer->entry);
311 timer->function(timer->data);
312 if (timer->interval) {
313 /* Recharge interval timer */
314 timer->expires = timer->interval +
315 atomic64_read(&virt_timer_elapsed);
316 spin_lock(&virt_timer_lock);
317 list_add_sorted(timer, &virt_timer_list);
318 spin_unlock(&virt_timer_lock);
319 }
320 }
321}
322
323void init_virt_timer(struct vtimer_list *timer)
324{
325 timer->function = NULL;
326 INIT_LIST_HEAD(&timer->entry);
327}
328EXPORT_SYMBOL(init_virt_timer);
329
330static inline int vtimer_pending(struct vtimer_list *timer)
331{
332 return !list_empty(&timer->entry);
333}
334
335static void internal_add_vtimer(struct vtimer_list *timer)
336{
337 if (list_empty(&virt_timer_list)) {
338 /* First timer, just program it. */
339 atomic64_set(&virt_timer_current, timer->expires);
340 atomic64_set(&virt_timer_elapsed, 0);
341 list_add(&timer->entry, &virt_timer_list);
342 } else {
343 /* Update timer against current base. */
344 timer->expires += atomic64_read(&virt_timer_elapsed);
345 if (likely((s64) timer->expires <
346 (s64) atomic64_read(&virt_timer_current)))
347 /* The new timer expires before the current timer. */
348 atomic64_set(&virt_timer_current, timer->expires);
349 /* Insert new timer into the list. */
350 list_add_sorted(timer, &virt_timer_list);
351 }
352}
353
354static void __add_vtimer(struct vtimer_list *timer, int periodic)
355{
356 unsigned long flags;
357
358 timer->interval = periodic ? timer->expires : 0;
359 spin_lock_irqsave(&virt_timer_lock, flags);
360 internal_add_vtimer(timer);
361 spin_unlock_irqrestore(&virt_timer_lock, flags);
362}
363
364/*
365 * add_virt_timer - add a oneshot virtual CPU timer
366 */
367void add_virt_timer(struct vtimer_list *timer)
368{
369 __add_vtimer(timer, 0);
370}
371EXPORT_SYMBOL(add_virt_timer);
372
373/*
374 * add_virt_timer_int - add an interval virtual CPU timer
375 */
376void add_virt_timer_periodic(struct vtimer_list *timer)
377{
378 __add_vtimer(timer, 1);
379}
380EXPORT_SYMBOL(add_virt_timer_periodic);
381
382static int __mod_vtimer(struct vtimer_list *timer, u64 expires, int periodic)
383{
384 unsigned long flags;
385 int rc;
386
387 BUG_ON(!timer->function);
388
389 if (timer->expires == expires && vtimer_pending(timer))
390 return 1;
391 spin_lock_irqsave(&virt_timer_lock, flags);
392 rc = vtimer_pending(timer);
393 if (rc)
394 list_del_init(&timer->entry);
395 timer->interval = periodic ? expires : 0;
396 timer->expires = expires;
397 internal_add_vtimer(timer);
398 spin_unlock_irqrestore(&virt_timer_lock, flags);
399 return rc;
400}
401
402/*
403 * returns whether it has modified a pending timer (1) or not (0)
404 */
405int mod_virt_timer(struct vtimer_list *timer, u64 expires)
406{
407 return __mod_vtimer(timer, expires, 0);
408}
409EXPORT_SYMBOL(mod_virt_timer);
410
411/*
412 * returns whether it has modified a pending timer (1) or not (0)
413 */
414int mod_virt_timer_periodic(struct vtimer_list *timer, u64 expires)
415{
416 return __mod_vtimer(timer, expires, 1);
417}
418EXPORT_SYMBOL(mod_virt_timer_periodic);
419
420/*
421 * Delete a virtual timer.
422 *
423 * returns whether the deleted timer was pending (1) or not (0)
424 */
425int del_virt_timer(struct vtimer_list *timer)
426{
427 unsigned long flags;
428
429 if (!vtimer_pending(timer))
430 return 0;
431 spin_lock_irqsave(&virt_timer_lock, flags);
432 list_del_init(&timer->entry);
433 spin_unlock_irqrestore(&virt_timer_lock, flags);
434 return 1;
435}
436EXPORT_SYMBOL(del_virt_timer);
437
438/*
439 * Start the virtual CPU timer on the current CPU.
440 */
441void vtime_init(void)
442{
443 /* set initial cpu timer */
444 set_vtimer(VTIMER_MAX_SLICE);
445 /* Setup initial MT scaling values */
446 if (smp_cpu_mtid) {
447 __this_cpu_write(mt_scaling_jiffies, jiffies);
448 __this_cpu_write(mt_scaling_mult, 1);
449 __this_cpu_write(mt_scaling_div, 1);
450 stcctm(MT_DIAG, smp_cpu_mtid + 1, this_cpu_ptr(mt_cycles));
451 }
452}
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