xref: /linux/arch/s390/kernel/vtime.c (revision 0d456bad36d42d16022be045c8a53ddbb59ee478)
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/irq.h>
23 #include "entry.h"
24 
25 static void virt_timer_expire(void);
26 
27 DEFINE_PER_CPU(struct s390_idle_data, s390_idle);
28 
29 static LIST_HEAD(virt_timer_list);
30 static DEFINE_SPINLOCK(virt_timer_lock);
31 static atomic64_t virt_timer_current;
32 static atomic64_t virt_timer_elapsed;
33 
34 static inline u64 get_vtimer(void)
35 {
36 	u64 timer;
37 
38 	asm volatile("stpt %0" : "=m" (timer));
39 	return timer;
40 }
41 
42 static inline void set_vtimer(u64 expires)
43 {
44 	u64 timer;
45 
46 	asm volatile(
47 		"	stpt	%0\n"	/* Store current cpu timer value */
48 		"	spt	%1"	/* Set new value imm. 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 static inline int virt_timer_forward(u64 elapsed)
55 {
56 	BUG_ON(!irqs_disabled());
57 
58 	if (list_empty(&virt_timer_list))
59 		return 0;
60 	elapsed = atomic64_add_return(elapsed, &virt_timer_elapsed);
61 	return elapsed >= atomic64_read(&virt_timer_current);
62 }
63 
64 /*
65  * Update process times based on virtual cpu times stored by entry.S
66  * to the lowcore fields user_timer, system_timer & steal_clock.
67  */
68 static int do_account_vtime(struct task_struct *tsk, int hardirq_offset)
69 {
70 	struct thread_info *ti = task_thread_info(tsk);
71 	u64 timer, clock, user, system, steal;
72 
73 	timer = S390_lowcore.last_update_timer;
74 	clock = S390_lowcore.last_update_clock;
75 	asm volatile(
76 		"	stpt	%0\n"	/* Store current cpu timer value */
77 		"	stck	%1"	/* Store current tod clock value */
78 		: "=m" (S390_lowcore.last_update_timer),
79 		  "=m" (S390_lowcore.last_update_clock));
80 	S390_lowcore.system_timer += timer - S390_lowcore.last_update_timer;
81 	S390_lowcore.steal_timer += S390_lowcore.last_update_clock - clock;
82 
83 	user = S390_lowcore.user_timer - ti->user_timer;
84 	S390_lowcore.steal_timer -= user;
85 	ti->user_timer = S390_lowcore.user_timer;
86 	account_user_time(tsk, user, user);
87 
88 	system = S390_lowcore.system_timer - ti->system_timer;
89 	S390_lowcore.steal_timer -= system;
90 	ti->system_timer = S390_lowcore.system_timer;
91 	account_system_time(tsk, hardirq_offset, system, system);
92 
93 	steal = S390_lowcore.steal_timer;
94 	if ((s64) steal > 0) {
95 		S390_lowcore.steal_timer = 0;
96 		account_steal_time(steal);
97 	}
98 
99 	return virt_timer_forward(user + system);
100 }
101 
102 void vtime_task_switch(struct task_struct *prev)
103 {
104 	struct thread_info *ti;
105 
106 	do_account_vtime(prev, 0);
107 	ti = task_thread_info(prev);
108 	ti->user_timer = S390_lowcore.user_timer;
109 	ti->system_timer = S390_lowcore.system_timer;
110 	ti = task_thread_info(current);
111 	S390_lowcore.user_timer = ti->user_timer;
112 	S390_lowcore.system_timer = ti->system_timer;
113 }
114 
115 /*
116  * In s390, accounting pending user time also implies
117  * accounting system time in order to correctly compute
118  * the stolen time accounting.
119  */
120 void vtime_account_user(struct task_struct *tsk)
121 {
122 	if (do_account_vtime(tsk, HARDIRQ_OFFSET))
123 		virt_timer_expire();
124 }
125 
126 /*
127  * Update process times based on virtual cpu times stored by entry.S
128  * to the lowcore fields user_timer, system_timer & steal_clock.
129  */
130 void vtime_account(struct task_struct *tsk)
131 {
132 	struct thread_info *ti = task_thread_info(tsk);
133 	u64 timer, system;
134 
135 	WARN_ON_ONCE(!irqs_disabled());
136 
137 	timer = S390_lowcore.last_update_timer;
138 	S390_lowcore.last_update_timer = get_vtimer();
139 	S390_lowcore.system_timer += timer - S390_lowcore.last_update_timer;
140 
141 	system = S390_lowcore.system_timer - ti->system_timer;
142 	S390_lowcore.steal_timer -= system;
143 	ti->system_timer = S390_lowcore.system_timer;
144 	account_system_time(tsk, 0, system, system);
145 
146 	virt_timer_forward(system);
147 }
148 EXPORT_SYMBOL_GPL(vtime_account);
149 
150 void vtime_account_system(struct task_struct *tsk)
151 __attribute__((alias("vtime_account")));
152 EXPORT_SYMBOL_GPL(vtime_account_system);
153 
154 void __kprobes vtime_stop_cpu(void)
155 {
156 	struct s390_idle_data *idle = &__get_cpu_var(s390_idle);
157 	unsigned long long idle_time;
158 	unsigned long psw_mask;
159 
160 	trace_hardirqs_on();
161 	/* Don't trace preempt off for idle. */
162 	stop_critical_timings();
163 
164 	/* Wait for external, I/O or machine check interrupt. */
165 	psw_mask = psw_kernel_bits | PSW_MASK_WAIT | PSW_MASK_DAT |
166 		PSW_MASK_IO | PSW_MASK_EXT | PSW_MASK_MCHECK;
167 	idle->nohz_delay = 0;
168 
169 	/* Call the assembler magic in entry.S */
170 	psw_idle(idle, psw_mask);
171 
172 	/* Reenable preemption tracer. */
173 	start_critical_timings();
174 
175 	/* Account time spent with enabled wait psw loaded as idle time. */
176 	idle->sequence++;
177 	smp_wmb();
178 	idle_time = idle->clock_idle_exit - idle->clock_idle_enter;
179 	idle->clock_idle_enter = idle->clock_idle_exit = 0ULL;
180 	idle->idle_time += idle_time;
181 	idle->idle_count++;
182 	account_idle_time(idle_time);
183 	smp_wmb();
184 	idle->sequence++;
185 }
186 
187 cputime64_t s390_get_idle_time(int cpu)
188 {
189 	struct s390_idle_data *idle = &per_cpu(s390_idle, cpu);
190 	unsigned long long now, idle_enter, idle_exit;
191 	unsigned int sequence;
192 
193 	do {
194 		now = get_clock();
195 		sequence = ACCESS_ONCE(idle->sequence);
196 		idle_enter = ACCESS_ONCE(idle->clock_idle_enter);
197 		idle_exit = ACCESS_ONCE(idle->clock_idle_exit);
198 	} while ((sequence & 1) || (idle->sequence != sequence));
199 	return idle_enter ? ((idle_exit ?: now) - idle_enter) : 0;
200 }
201 
202 /*
203  * Sorted add to a list. List is linear searched until first bigger
204  * element is found.
205  */
206 static void list_add_sorted(struct vtimer_list *timer, struct list_head *head)
207 {
208 	struct vtimer_list *tmp;
209 
210 	list_for_each_entry(tmp, head, entry) {
211 		if (tmp->expires > timer->expires) {
212 			list_add_tail(&timer->entry, &tmp->entry);
213 			return;
214 		}
215 	}
216 	list_add_tail(&timer->entry, head);
217 }
218 
219 /*
220  * Handler for expired virtual CPU timer.
221  */
222 static void virt_timer_expire(void)
223 {
224 	struct vtimer_list *timer, *tmp;
225 	unsigned long elapsed;
226 	LIST_HEAD(cb_list);
227 
228 	/* walk timer list, fire all expired timers */
229 	spin_lock(&virt_timer_lock);
230 	elapsed = atomic64_read(&virt_timer_elapsed);
231 	list_for_each_entry_safe(timer, tmp, &virt_timer_list, entry) {
232 		if (timer->expires < elapsed)
233 			/* move expired timer to the callback queue */
234 			list_move_tail(&timer->entry, &cb_list);
235 		else
236 			timer->expires -= elapsed;
237 	}
238 	if (!list_empty(&virt_timer_list)) {
239 		timer = list_first_entry(&virt_timer_list,
240 					 struct vtimer_list, entry);
241 		atomic64_set(&virt_timer_current, timer->expires);
242 	}
243 	atomic64_sub(elapsed, &virt_timer_elapsed);
244 	spin_unlock(&virt_timer_lock);
245 
246 	/* Do callbacks and recharge periodic timers */
247 	list_for_each_entry_safe(timer, tmp, &cb_list, entry) {
248 		list_del_init(&timer->entry);
249 		timer->function(timer->data);
250 		if (timer->interval) {
251 			/* Recharge interval timer */
252 			timer->expires = timer->interval +
253 				atomic64_read(&virt_timer_elapsed);
254 			spin_lock(&virt_timer_lock);
255 			list_add_sorted(timer, &virt_timer_list);
256 			spin_unlock(&virt_timer_lock);
257 		}
258 	}
259 }
260 
261 void init_virt_timer(struct vtimer_list *timer)
262 {
263 	timer->function = NULL;
264 	INIT_LIST_HEAD(&timer->entry);
265 }
266 EXPORT_SYMBOL(init_virt_timer);
267 
268 static inline int vtimer_pending(struct vtimer_list *timer)
269 {
270 	return !list_empty(&timer->entry);
271 }
272 
273 static void internal_add_vtimer(struct vtimer_list *timer)
274 {
275 	if (list_empty(&virt_timer_list)) {
276 		/* First timer, just program it. */
277 		atomic64_set(&virt_timer_current, timer->expires);
278 		atomic64_set(&virt_timer_elapsed, 0);
279 		list_add(&timer->entry, &virt_timer_list);
280 	} else {
281 		/* Update timer against current base. */
282 		timer->expires += atomic64_read(&virt_timer_elapsed);
283 		if (likely((s64) timer->expires <
284 			   (s64) atomic64_read(&virt_timer_current)))
285 			/* The new timer expires before the current timer. */
286 			atomic64_set(&virt_timer_current, timer->expires);
287 		/* Insert new timer into the list. */
288 		list_add_sorted(timer, &virt_timer_list);
289 	}
290 }
291 
292 static void __add_vtimer(struct vtimer_list *timer, int periodic)
293 {
294 	unsigned long flags;
295 
296 	timer->interval = periodic ? timer->expires : 0;
297 	spin_lock_irqsave(&virt_timer_lock, flags);
298 	internal_add_vtimer(timer);
299 	spin_unlock_irqrestore(&virt_timer_lock, flags);
300 }
301 
302 /*
303  * add_virt_timer - add an oneshot virtual CPU timer
304  */
305 void add_virt_timer(struct vtimer_list *timer)
306 {
307 	__add_vtimer(timer, 0);
308 }
309 EXPORT_SYMBOL(add_virt_timer);
310 
311 /*
312  * add_virt_timer_int - add an interval virtual CPU timer
313  */
314 void add_virt_timer_periodic(struct vtimer_list *timer)
315 {
316 	__add_vtimer(timer, 1);
317 }
318 EXPORT_SYMBOL(add_virt_timer_periodic);
319 
320 static int __mod_vtimer(struct vtimer_list *timer, u64 expires, int periodic)
321 {
322 	unsigned long flags;
323 	int rc;
324 
325 	BUG_ON(!timer->function);
326 
327 	if (timer->expires == expires && vtimer_pending(timer))
328 		return 1;
329 	spin_lock_irqsave(&virt_timer_lock, flags);
330 	rc = vtimer_pending(timer);
331 	if (rc)
332 		list_del_init(&timer->entry);
333 	timer->interval = periodic ? expires : 0;
334 	timer->expires = expires;
335 	internal_add_vtimer(timer);
336 	spin_unlock_irqrestore(&virt_timer_lock, flags);
337 	return rc;
338 }
339 
340 /*
341  * returns whether it has modified a pending timer (1) or not (0)
342  */
343 int mod_virt_timer(struct vtimer_list *timer, u64 expires)
344 {
345 	return __mod_vtimer(timer, expires, 0);
346 }
347 EXPORT_SYMBOL(mod_virt_timer);
348 
349 /*
350  * returns whether it has modified a pending timer (1) or not (0)
351  */
352 int mod_virt_timer_periodic(struct vtimer_list *timer, u64 expires)
353 {
354 	return __mod_vtimer(timer, expires, 1);
355 }
356 EXPORT_SYMBOL(mod_virt_timer_periodic);
357 
358 /*
359  * Delete a virtual timer.
360  *
361  * returns whether the deleted timer was pending (1) or not (0)
362  */
363 int del_virt_timer(struct vtimer_list *timer)
364 {
365 	unsigned long flags;
366 
367 	if (!vtimer_pending(timer))
368 		return 0;
369 	spin_lock_irqsave(&virt_timer_lock, flags);
370 	list_del_init(&timer->entry);
371 	spin_unlock_irqrestore(&virt_timer_lock, flags);
372 	return 1;
373 }
374 EXPORT_SYMBOL(del_virt_timer);
375 
376 /*
377  * Start the virtual CPU timer on the current CPU.
378  */
379 void __cpuinit init_cpu_vtimer(void)
380 {
381 	/* set initial cpu timer */
382 	set_vtimer(VTIMER_MAX_SLICE);
383 }
384 
385 static int __cpuinit s390_nohz_notify(struct notifier_block *self,
386 				      unsigned long action, void *hcpu)
387 {
388 	struct s390_idle_data *idle;
389 	long cpu = (long) hcpu;
390 
391 	idle = &per_cpu(s390_idle, cpu);
392 	switch (action & ~CPU_TASKS_FROZEN) {
393 	case CPU_DYING:
394 		idle->nohz_delay = 0;
395 	default:
396 		break;
397 	}
398 	return NOTIFY_OK;
399 }
400 
401 void __init vtime_init(void)
402 {
403 	/* Enable cpu timer interrupts on the boot cpu. */
404 	init_cpu_vtimer();
405 	cpu_notifier(s390_nohz_notify, 0);
406 }
407