xref: /linux/arch/s390/kernel/vtime.c (revision c98be0c96db00e9b6b02d31e0fa7590c54cdaaac)
1 /*
2  *    Virtual cpu timer based timer functions.
3  *
4  *    Copyright IBM Corp. 2004, 2012
5  *    Author(s): Jan Glauber <jan.glauber@de.ibm.com>
6  */
7 
8 #include <linux/kernel_stat.h>
9 #include <linux/notifier.h>
10 #include <linux/kprobes.h>
11 #include <linux/export.h>
12 #include <linux/kernel.h>
13 #include <linux/timex.h>
14 #include <linux/types.h>
15 #include <linux/time.h>
16 #include <linux/cpu.h>
17 #include <linux/smp.h>
18 
19 #include <asm/irq_regs.h>
20 #include <asm/cputime.h>
21 #include <asm/vtimer.h>
22 #include <asm/vtime.h>
23 #include <asm/irq.h>
24 #include "entry.h"
25 
26 static void virt_timer_expire(void);
27 
28 DEFINE_PER_CPU(struct s390_idle_data, s390_idle);
29 
30 static LIST_HEAD(virt_timer_list);
31 static DEFINE_SPINLOCK(virt_timer_lock);
32 static atomic64_t virt_timer_current;
33 static atomic64_t virt_timer_elapsed;
34 
35 static inline u64 get_vtimer(void)
36 {
37 	u64 timer;
38 
39 	asm volatile("stpt %0" : "=m" (timer));
40 	return timer;
41 }
42 
43 static inline void set_vtimer(u64 expires)
44 {
45 	u64 timer;
46 
47 	asm volatile(
48 		"	stpt	%0\n"	/* Store current cpu timer value */
49 		"	spt	%1"	/* Set new value imm. afterwards */
50 		: "=m" (timer) : "m" (expires));
51 	S390_lowcore.system_timer += S390_lowcore.last_update_timer - timer;
52 	S390_lowcore.last_update_timer = expires;
53 }
54 
55 static inline int virt_timer_forward(u64 elapsed)
56 {
57 	BUG_ON(!irqs_disabled());
58 
59 	if (list_empty(&virt_timer_list))
60 		return 0;
61 	elapsed = atomic64_add_return(elapsed, &virt_timer_elapsed);
62 	return elapsed >= atomic64_read(&virt_timer_current);
63 }
64 
65 /*
66  * Update process times based on virtual cpu times stored by entry.S
67  * to the lowcore fields user_timer, system_timer & steal_clock.
68  */
69 static int do_account_vtime(struct task_struct *tsk, int hardirq_offset)
70 {
71 	struct thread_info *ti = task_thread_info(tsk);
72 	u64 timer, clock, user, system, steal;
73 
74 	timer = S390_lowcore.last_update_timer;
75 	clock = S390_lowcore.last_update_clock;
76 	asm volatile(
77 		"	stpt	%0\n"	/* Store current cpu timer value */
78 		"	stck	%1"	/* Store current tod clock value */
79 		: "=m" (S390_lowcore.last_update_timer),
80 		  "=m" (S390_lowcore.last_update_clock));
81 	S390_lowcore.system_timer += timer - S390_lowcore.last_update_timer;
82 	S390_lowcore.steal_timer += S390_lowcore.last_update_clock - clock;
83 
84 	user = S390_lowcore.user_timer - ti->user_timer;
85 	S390_lowcore.steal_timer -= user;
86 	ti->user_timer = S390_lowcore.user_timer;
87 	account_user_time(tsk, user, user);
88 
89 	system = S390_lowcore.system_timer - ti->system_timer;
90 	S390_lowcore.steal_timer -= system;
91 	ti->system_timer = S390_lowcore.system_timer;
92 	account_system_time(tsk, hardirq_offset, system, system);
93 
94 	steal = S390_lowcore.steal_timer;
95 	if ((s64) steal > 0) {
96 		S390_lowcore.steal_timer = 0;
97 		account_steal_time(steal);
98 	}
99 
100 	return virt_timer_forward(user + system);
101 }
102 
103 void vtime_task_switch(struct task_struct *prev)
104 {
105 	struct thread_info *ti;
106 
107 	do_account_vtime(prev, 0);
108 	ti = task_thread_info(prev);
109 	ti->user_timer = S390_lowcore.user_timer;
110 	ti->system_timer = S390_lowcore.system_timer;
111 	ti = task_thread_info(current);
112 	S390_lowcore.user_timer = ti->user_timer;
113 	S390_lowcore.system_timer = ti->system_timer;
114 }
115 
116 /*
117  * In s390, accounting pending user time also implies
118  * accounting system time in order to correctly compute
119  * the stolen time accounting.
120  */
121 void vtime_account_user(struct task_struct *tsk)
122 {
123 	if (do_account_vtime(tsk, HARDIRQ_OFFSET))
124 		virt_timer_expire();
125 }
126 
127 /*
128  * Update process times based on virtual cpu times stored by entry.S
129  * to the lowcore fields user_timer, system_timer & steal_clock.
130  */
131 void vtime_account_irq_enter(struct task_struct *tsk)
132 {
133 	struct thread_info *ti = task_thread_info(tsk);
134 	u64 timer, system;
135 
136 	WARN_ON_ONCE(!irqs_disabled());
137 
138 	timer = S390_lowcore.last_update_timer;
139 	S390_lowcore.last_update_timer = get_vtimer();
140 	S390_lowcore.system_timer += timer - S390_lowcore.last_update_timer;
141 
142 	system = S390_lowcore.system_timer - ti->system_timer;
143 	S390_lowcore.steal_timer -= system;
144 	ti->system_timer = S390_lowcore.system_timer;
145 	account_system_time(tsk, 0, system, system);
146 
147 	virt_timer_forward(system);
148 }
149 EXPORT_SYMBOL_GPL(vtime_account_irq_enter);
150 
151 void vtime_account_system(struct task_struct *tsk)
152 __attribute__((alias("vtime_account_irq_enter")));
153 EXPORT_SYMBOL_GPL(vtime_account_system);
154 
155 void __kprobes vtime_stop_cpu(void)
156 {
157 	struct s390_idle_data *idle = &__get_cpu_var(s390_idle);
158 	unsigned long long idle_time;
159 	unsigned long psw_mask;
160 
161 	trace_hardirqs_on();
162 
163 	/* Wait for external, I/O or machine check interrupt. */
164 	psw_mask = PSW_KERNEL_BITS | PSW_MASK_WAIT | PSW_MASK_DAT |
165 		PSW_MASK_IO | PSW_MASK_EXT | PSW_MASK_MCHECK;
166 	idle->nohz_delay = 0;
167 
168 	/* Call the assembler magic in entry.S */
169 	psw_idle(idle, psw_mask);
170 
171 	/* Account time spent with enabled wait psw loaded as idle time. */
172 	idle->sequence++;
173 	smp_wmb();
174 	idle_time = idle->clock_idle_exit - idle->clock_idle_enter;
175 	idle->clock_idle_enter = idle->clock_idle_exit = 0ULL;
176 	idle->idle_time += idle_time;
177 	idle->idle_count++;
178 	account_idle_time(idle_time);
179 	smp_wmb();
180 	idle->sequence++;
181 }
182 
183 cputime64_t s390_get_idle_time(int cpu)
184 {
185 	struct s390_idle_data *idle = &per_cpu(s390_idle, cpu);
186 	unsigned long long now, idle_enter, idle_exit;
187 	unsigned int sequence;
188 
189 	do {
190 		now = get_tod_clock();
191 		sequence = ACCESS_ONCE(idle->sequence);
192 		idle_enter = ACCESS_ONCE(idle->clock_idle_enter);
193 		idle_exit = ACCESS_ONCE(idle->clock_idle_exit);
194 	} while ((sequence & 1) || (ACCESS_ONCE(idle->sequence) != sequence));
195 	return idle_enter ? ((idle_exit ?: now) - idle_enter) : 0;
196 }
197 
198 /*
199  * Sorted add to a list. List is linear searched until first bigger
200  * element is found.
201  */
202 static void list_add_sorted(struct vtimer_list *timer, struct list_head *head)
203 {
204 	struct vtimer_list *tmp;
205 
206 	list_for_each_entry(tmp, head, entry) {
207 		if (tmp->expires > timer->expires) {
208 			list_add_tail(&timer->entry, &tmp->entry);
209 			return;
210 		}
211 	}
212 	list_add_tail(&timer->entry, head);
213 }
214 
215 /*
216  * Handler for expired virtual CPU timer.
217  */
218 static void virt_timer_expire(void)
219 {
220 	struct vtimer_list *timer, *tmp;
221 	unsigned long elapsed;
222 	LIST_HEAD(cb_list);
223 
224 	/* walk timer list, fire all expired timers */
225 	spin_lock(&virt_timer_lock);
226 	elapsed = atomic64_read(&virt_timer_elapsed);
227 	list_for_each_entry_safe(timer, tmp, &virt_timer_list, entry) {
228 		if (timer->expires < elapsed)
229 			/* move expired timer to the callback queue */
230 			list_move_tail(&timer->entry, &cb_list);
231 		else
232 			timer->expires -= elapsed;
233 	}
234 	if (!list_empty(&virt_timer_list)) {
235 		timer = list_first_entry(&virt_timer_list,
236 					 struct vtimer_list, entry);
237 		atomic64_set(&virt_timer_current, timer->expires);
238 	}
239 	atomic64_sub(elapsed, &virt_timer_elapsed);
240 	spin_unlock(&virt_timer_lock);
241 
242 	/* Do callbacks and recharge periodic timers */
243 	list_for_each_entry_safe(timer, tmp, &cb_list, entry) {
244 		list_del_init(&timer->entry);
245 		timer->function(timer->data);
246 		if (timer->interval) {
247 			/* Recharge interval timer */
248 			timer->expires = timer->interval +
249 				atomic64_read(&virt_timer_elapsed);
250 			spin_lock(&virt_timer_lock);
251 			list_add_sorted(timer, &virt_timer_list);
252 			spin_unlock(&virt_timer_lock);
253 		}
254 	}
255 }
256 
257 void init_virt_timer(struct vtimer_list *timer)
258 {
259 	timer->function = NULL;
260 	INIT_LIST_HEAD(&timer->entry);
261 }
262 EXPORT_SYMBOL(init_virt_timer);
263 
264 static inline int vtimer_pending(struct vtimer_list *timer)
265 {
266 	return !list_empty(&timer->entry);
267 }
268 
269 static void internal_add_vtimer(struct vtimer_list *timer)
270 {
271 	if (list_empty(&virt_timer_list)) {
272 		/* First timer, just program it. */
273 		atomic64_set(&virt_timer_current, timer->expires);
274 		atomic64_set(&virt_timer_elapsed, 0);
275 		list_add(&timer->entry, &virt_timer_list);
276 	} else {
277 		/* Update timer against current base. */
278 		timer->expires += atomic64_read(&virt_timer_elapsed);
279 		if (likely((s64) timer->expires <
280 			   (s64) atomic64_read(&virt_timer_current)))
281 			/* The new timer expires before the current timer. */
282 			atomic64_set(&virt_timer_current, timer->expires);
283 		/* Insert new timer into the list. */
284 		list_add_sorted(timer, &virt_timer_list);
285 	}
286 }
287 
288 static void __add_vtimer(struct vtimer_list *timer, int periodic)
289 {
290 	unsigned long flags;
291 
292 	timer->interval = periodic ? timer->expires : 0;
293 	spin_lock_irqsave(&virt_timer_lock, flags);
294 	internal_add_vtimer(timer);
295 	spin_unlock_irqrestore(&virt_timer_lock, flags);
296 }
297 
298 /*
299  * add_virt_timer - add an oneshot virtual CPU timer
300  */
301 void add_virt_timer(struct vtimer_list *timer)
302 {
303 	__add_vtimer(timer, 0);
304 }
305 EXPORT_SYMBOL(add_virt_timer);
306 
307 /*
308  * add_virt_timer_int - add an interval virtual CPU timer
309  */
310 void add_virt_timer_periodic(struct vtimer_list *timer)
311 {
312 	__add_vtimer(timer, 1);
313 }
314 EXPORT_SYMBOL(add_virt_timer_periodic);
315 
316 static int __mod_vtimer(struct vtimer_list *timer, u64 expires, int periodic)
317 {
318 	unsigned long flags;
319 	int rc;
320 
321 	BUG_ON(!timer->function);
322 
323 	if (timer->expires == expires && vtimer_pending(timer))
324 		return 1;
325 	spin_lock_irqsave(&virt_timer_lock, flags);
326 	rc = vtimer_pending(timer);
327 	if (rc)
328 		list_del_init(&timer->entry);
329 	timer->interval = periodic ? expires : 0;
330 	timer->expires = expires;
331 	internal_add_vtimer(timer);
332 	spin_unlock_irqrestore(&virt_timer_lock, flags);
333 	return rc;
334 }
335 
336 /*
337  * returns whether it has modified a pending timer (1) or not (0)
338  */
339 int mod_virt_timer(struct vtimer_list *timer, u64 expires)
340 {
341 	return __mod_vtimer(timer, expires, 0);
342 }
343 EXPORT_SYMBOL(mod_virt_timer);
344 
345 /*
346  * returns whether it has modified a pending timer (1) or not (0)
347  */
348 int mod_virt_timer_periodic(struct vtimer_list *timer, u64 expires)
349 {
350 	return __mod_vtimer(timer, expires, 1);
351 }
352 EXPORT_SYMBOL(mod_virt_timer_periodic);
353 
354 /*
355  * Delete a virtual timer.
356  *
357  * returns whether the deleted timer was pending (1) or not (0)
358  */
359 int del_virt_timer(struct vtimer_list *timer)
360 {
361 	unsigned long flags;
362 
363 	if (!vtimer_pending(timer))
364 		return 0;
365 	spin_lock_irqsave(&virt_timer_lock, flags);
366 	list_del_init(&timer->entry);
367 	spin_unlock_irqrestore(&virt_timer_lock, flags);
368 	return 1;
369 }
370 EXPORT_SYMBOL(del_virt_timer);
371 
372 /*
373  * Start the virtual CPU timer on the current CPU.
374  */
375 void init_cpu_vtimer(void)
376 {
377 	/* set initial cpu timer */
378 	set_vtimer(VTIMER_MAX_SLICE);
379 }
380 
381 static int s390_nohz_notify(struct notifier_block *self, unsigned long action,
382 			    void *hcpu)
383 {
384 	struct s390_idle_data *idle;
385 	long cpu = (long) hcpu;
386 
387 	idle = &per_cpu(s390_idle, cpu);
388 	switch (action & ~CPU_TASKS_FROZEN) {
389 	case CPU_DYING:
390 		idle->nohz_delay = 0;
391 	default:
392 		break;
393 	}
394 	return NOTIFY_OK;
395 }
396 
397 void __init vtime_init(void)
398 {
399 	/* Enable cpu timer interrupts on the boot cpu. */
400 	init_cpu_vtimer();
401 	cpu_notifier(s390_nohz_notify, 0);
402 }
403