xref: /linux/arch/s390/kernel/vtime.c (revision fe6bce8d30a86c693bf7cfbf4759cbafd121289f)
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/export.h>
10 #include <linux/kernel.h>
11 #include <linux/timex.h>
12 #include <linux/types.h>
13 #include <linux/time.h>
14 
15 #include <asm/cputime.h>
16 #include <asm/vtimer.h>
17 #include <asm/vtime.h>
18 #include <asm/cpu_mf.h>
19 #include <asm/smp.h>
20 
21 #include "entry.h"
22 
23 static void virt_timer_expire(void);
24 
25 static LIST_HEAD(virt_timer_list);
26 static DEFINE_SPINLOCK(virt_timer_lock);
27 static atomic64_t virt_timer_current;
28 static atomic64_t virt_timer_elapsed;
29 
30 DEFINE_PER_CPU(u64, mt_cycles[8]);
31 static DEFINE_PER_CPU(u64, mt_scaling_mult) = { 1 };
32 static DEFINE_PER_CPU(u64, mt_scaling_div) = { 1 };
33 static DEFINE_PER_CPU(u64, mt_scaling_jiffies);
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 static void update_mt_scaling(void)
66 {
67 	u64 cycles_new[8], *cycles_old;
68 	u64 delta, fac, mult, div;
69 	int i;
70 
71 	stcctm5(smp_cpu_mtid + 1, cycles_new);
72 	cycles_old = this_cpu_ptr(mt_cycles);
73 	fac = 1;
74 	mult = div = 0;
75 	for (i = 0; i <= smp_cpu_mtid; i++) {
76 		delta = cycles_new[i] - cycles_old[i];
77 		div += delta;
78 		mult *= i + 1;
79 		mult += delta * fac;
80 		fac *= i + 1;
81 	}
82 	div *= fac;
83 	if (div > 0) {
84 		/* Update scaling factor */
85 		__this_cpu_write(mt_scaling_mult, mult);
86 		__this_cpu_write(mt_scaling_div, div);
87 		memcpy(cycles_old, cycles_new,
88 		       sizeof(u64) * (smp_cpu_mtid + 1));
89 	}
90 	__this_cpu_write(mt_scaling_jiffies, jiffies_64);
91 }
92 
93 /*
94  * Update process times based on virtual cpu times stored by entry.S
95  * to the lowcore fields user_timer, system_timer & steal_clock.
96  */
97 static int do_account_vtime(struct task_struct *tsk, int hardirq_offset)
98 {
99 	struct thread_info *ti = task_thread_info(tsk);
100 	u64 timer, clock, user, system, steal;
101 	u64 user_scaled, system_scaled;
102 
103 	timer = S390_lowcore.last_update_timer;
104 	clock = S390_lowcore.last_update_clock;
105 	asm volatile(
106 		"	stpt	%0\n"	/* Store current cpu timer value */
107 #ifdef CONFIG_HAVE_MARCH_Z9_109_FEATURES
108 		"	stckf	%1"	/* Store current tod clock value */
109 #else
110 		"	stck	%1"	/* Store current tod clock value */
111 #endif
112 		: "=m" (S390_lowcore.last_update_timer),
113 		  "=m" (S390_lowcore.last_update_clock));
114 	S390_lowcore.system_timer += timer - S390_lowcore.last_update_timer;
115 	S390_lowcore.steal_timer += S390_lowcore.last_update_clock - clock;
116 
117 	/* Update MT utilization calculation */
118 	if (smp_cpu_mtid &&
119 	    time_after64(jiffies_64, this_cpu_read(mt_scaling_jiffies)))
120 		update_mt_scaling();
121 
122 	user = S390_lowcore.user_timer - ti->user_timer;
123 	S390_lowcore.steal_timer -= user;
124 	ti->user_timer = S390_lowcore.user_timer;
125 
126 	system = S390_lowcore.system_timer - ti->system_timer;
127 	S390_lowcore.steal_timer -= system;
128 	ti->system_timer = S390_lowcore.system_timer;
129 
130 	user_scaled = user;
131 	system_scaled = system;
132 	/* Do MT utilization scaling */
133 	if (smp_cpu_mtid) {
134 		u64 mult = __this_cpu_read(mt_scaling_mult);
135 		u64 div = __this_cpu_read(mt_scaling_div);
136 
137 		user_scaled = (user_scaled * mult) / div;
138 		system_scaled = (system_scaled * mult) / div;
139 	}
140 	account_user_time(tsk, user);
141 	tsk->utimescaled += user_scaled;
142 	account_system_time(tsk, hardirq_offset, system);
143 	tsk->stimescaled += system_scaled;
144 
145 	steal = S390_lowcore.steal_timer;
146 	if ((s64) steal > 0) {
147 		S390_lowcore.steal_timer = 0;
148 		account_steal_time(steal);
149 	}
150 
151 	return virt_timer_forward(user + system);
152 }
153 
154 void vtime_task_switch(struct task_struct *prev)
155 {
156 	struct thread_info *ti;
157 
158 	do_account_vtime(prev, 0);
159 	ti = task_thread_info(prev);
160 	ti->user_timer = S390_lowcore.user_timer;
161 	ti->system_timer = S390_lowcore.system_timer;
162 	ti = task_thread_info(current);
163 	S390_lowcore.user_timer = ti->user_timer;
164 	S390_lowcore.system_timer = ti->system_timer;
165 }
166 
167 /*
168  * In s390, accounting pending user time also implies
169  * accounting system time in order to correctly compute
170  * the stolen time accounting.
171  */
172 void vtime_account_user(struct task_struct *tsk)
173 {
174 	if (do_account_vtime(tsk, HARDIRQ_OFFSET))
175 		virt_timer_expire();
176 }
177 
178 /*
179  * Update process times based on virtual cpu times stored by entry.S
180  * to the lowcore fields user_timer, system_timer & steal_clock.
181  */
182 void vtime_account_irq_enter(struct task_struct *tsk)
183 {
184 	struct thread_info *ti = task_thread_info(tsk);
185 	u64 timer, system, system_scaled;
186 
187 	timer = S390_lowcore.last_update_timer;
188 	S390_lowcore.last_update_timer = get_vtimer();
189 	S390_lowcore.system_timer += timer - S390_lowcore.last_update_timer;
190 
191 	/* Update MT utilization calculation */
192 	if (smp_cpu_mtid &&
193 	    time_after64(jiffies_64, this_cpu_read(mt_scaling_jiffies)))
194 		update_mt_scaling();
195 
196 	system = S390_lowcore.system_timer - ti->system_timer;
197 	S390_lowcore.steal_timer -= system;
198 	ti->system_timer = S390_lowcore.system_timer;
199 	system_scaled = system;
200 	/* Do MT utilization scaling */
201 	if (smp_cpu_mtid) {
202 		u64 mult = __this_cpu_read(mt_scaling_mult);
203 		u64 div = __this_cpu_read(mt_scaling_div);
204 
205 		system_scaled = (system_scaled * mult) / div;
206 	}
207 	account_system_time(tsk, 0, system);
208 	tsk->stimescaled += system_scaled;
209 
210 	virt_timer_forward(system);
211 }
212 EXPORT_SYMBOL_GPL(vtime_account_irq_enter);
213 
214 void vtime_account_system(struct task_struct *tsk)
215 __attribute__((alias("vtime_account_irq_enter")));
216 EXPORT_SYMBOL_GPL(vtime_account_system);
217 
218 /*
219  * Sorted add to a list. List is linear searched until first bigger
220  * element is found.
221  */
222 static void list_add_sorted(struct vtimer_list *timer, struct list_head *head)
223 {
224 	struct vtimer_list *tmp;
225 
226 	list_for_each_entry(tmp, head, entry) {
227 		if (tmp->expires > timer->expires) {
228 			list_add_tail(&timer->entry, &tmp->entry);
229 			return;
230 		}
231 	}
232 	list_add_tail(&timer->entry, head);
233 }
234 
235 /*
236  * Handler for expired virtual CPU timer.
237  */
238 static void virt_timer_expire(void)
239 {
240 	struct vtimer_list *timer, *tmp;
241 	unsigned long elapsed;
242 	LIST_HEAD(cb_list);
243 
244 	/* walk timer list, fire all expired timers */
245 	spin_lock(&virt_timer_lock);
246 	elapsed = atomic64_read(&virt_timer_elapsed);
247 	list_for_each_entry_safe(timer, tmp, &virt_timer_list, entry) {
248 		if (timer->expires < elapsed)
249 			/* move expired timer to the callback queue */
250 			list_move_tail(&timer->entry, &cb_list);
251 		else
252 			timer->expires -= elapsed;
253 	}
254 	if (!list_empty(&virt_timer_list)) {
255 		timer = list_first_entry(&virt_timer_list,
256 					 struct vtimer_list, entry);
257 		atomic64_set(&virt_timer_current, timer->expires);
258 	}
259 	atomic64_sub(elapsed, &virt_timer_elapsed);
260 	spin_unlock(&virt_timer_lock);
261 
262 	/* Do callbacks and recharge periodic timers */
263 	list_for_each_entry_safe(timer, tmp, &cb_list, entry) {
264 		list_del_init(&timer->entry);
265 		timer->function(timer->data);
266 		if (timer->interval) {
267 			/* Recharge interval timer */
268 			timer->expires = timer->interval +
269 				atomic64_read(&virt_timer_elapsed);
270 			spin_lock(&virt_timer_lock);
271 			list_add_sorted(timer, &virt_timer_list);
272 			spin_unlock(&virt_timer_lock);
273 		}
274 	}
275 }
276 
277 void init_virt_timer(struct vtimer_list *timer)
278 {
279 	timer->function = NULL;
280 	INIT_LIST_HEAD(&timer->entry);
281 }
282 EXPORT_SYMBOL(init_virt_timer);
283 
284 static inline int vtimer_pending(struct vtimer_list *timer)
285 {
286 	return !list_empty(&timer->entry);
287 }
288 
289 static void internal_add_vtimer(struct vtimer_list *timer)
290 {
291 	if (list_empty(&virt_timer_list)) {
292 		/* First timer, just program it. */
293 		atomic64_set(&virt_timer_current, timer->expires);
294 		atomic64_set(&virt_timer_elapsed, 0);
295 		list_add(&timer->entry, &virt_timer_list);
296 	} else {
297 		/* Update timer against current base. */
298 		timer->expires += atomic64_read(&virt_timer_elapsed);
299 		if (likely((s64) timer->expires <
300 			   (s64) atomic64_read(&virt_timer_current)))
301 			/* The new timer expires before the current timer. */
302 			atomic64_set(&virt_timer_current, timer->expires);
303 		/* Insert new timer into the list. */
304 		list_add_sorted(timer, &virt_timer_list);
305 	}
306 }
307 
308 static void __add_vtimer(struct vtimer_list *timer, int periodic)
309 {
310 	unsigned long flags;
311 
312 	timer->interval = periodic ? timer->expires : 0;
313 	spin_lock_irqsave(&virt_timer_lock, flags);
314 	internal_add_vtimer(timer);
315 	spin_unlock_irqrestore(&virt_timer_lock, flags);
316 }
317 
318 /*
319  * add_virt_timer - add an oneshot virtual CPU timer
320  */
321 void add_virt_timer(struct vtimer_list *timer)
322 {
323 	__add_vtimer(timer, 0);
324 }
325 EXPORT_SYMBOL(add_virt_timer);
326 
327 /*
328  * add_virt_timer_int - add an interval virtual CPU timer
329  */
330 void add_virt_timer_periodic(struct vtimer_list *timer)
331 {
332 	__add_vtimer(timer, 1);
333 }
334 EXPORT_SYMBOL(add_virt_timer_periodic);
335 
336 static int __mod_vtimer(struct vtimer_list *timer, u64 expires, int periodic)
337 {
338 	unsigned long flags;
339 	int rc;
340 
341 	BUG_ON(!timer->function);
342 
343 	if (timer->expires == expires && vtimer_pending(timer))
344 		return 1;
345 	spin_lock_irqsave(&virt_timer_lock, flags);
346 	rc = vtimer_pending(timer);
347 	if (rc)
348 		list_del_init(&timer->entry);
349 	timer->interval = periodic ? expires : 0;
350 	timer->expires = expires;
351 	internal_add_vtimer(timer);
352 	spin_unlock_irqrestore(&virt_timer_lock, flags);
353 	return rc;
354 }
355 
356 /*
357  * returns whether it has modified a pending timer (1) or not (0)
358  */
359 int mod_virt_timer(struct vtimer_list *timer, u64 expires)
360 {
361 	return __mod_vtimer(timer, expires, 0);
362 }
363 EXPORT_SYMBOL(mod_virt_timer);
364 
365 /*
366  * returns whether it has modified a pending timer (1) or not (0)
367  */
368 int mod_virt_timer_periodic(struct vtimer_list *timer, u64 expires)
369 {
370 	return __mod_vtimer(timer, expires, 1);
371 }
372 EXPORT_SYMBOL(mod_virt_timer_periodic);
373 
374 /*
375  * Delete a virtual timer.
376  *
377  * returns whether the deleted timer was pending (1) or not (0)
378  */
379 int del_virt_timer(struct vtimer_list *timer)
380 {
381 	unsigned long flags;
382 
383 	if (!vtimer_pending(timer))
384 		return 0;
385 	spin_lock_irqsave(&virt_timer_lock, flags);
386 	list_del_init(&timer->entry);
387 	spin_unlock_irqrestore(&virt_timer_lock, flags);
388 	return 1;
389 }
390 EXPORT_SYMBOL(del_virt_timer);
391 
392 /*
393  * Start the virtual CPU timer on the current CPU.
394  */
395 void vtime_init(void)
396 {
397 	/* set initial cpu timer */
398 	set_vtimer(VTIMER_MAX_SLICE);
399 	/* Setup initial MT scaling values */
400 	if (smp_cpu_mtid) {
401 		__this_cpu_write(mt_scaling_jiffies, jiffies);
402 		__this_cpu_write(mt_scaling_mult, 1);
403 		__this_cpu_write(mt_scaling_div, 1);
404 		stcctm5(smp_cpu_mtid + 1, this_cpu_ptr(mt_cycles));
405 	}
406 }
407