xref: /openbmc/linux/arch/s390/kernel/vtime.c (revision bc5aa3a0)
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, user_scaled);
141 	account_system_time(tsk, hardirq_offset, system, system_scaled);
142 
143 	steal = S390_lowcore.steal_timer;
144 	if ((s64) steal > 0) {
145 		S390_lowcore.steal_timer = 0;
146 		account_steal_time(steal);
147 	}
148 
149 	return virt_timer_forward(user + system);
150 }
151 
152 void vtime_task_switch(struct task_struct *prev)
153 {
154 	struct thread_info *ti;
155 
156 	do_account_vtime(prev, 0);
157 	ti = task_thread_info(prev);
158 	ti->user_timer = S390_lowcore.user_timer;
159 	ti->system_timer = S390_lowcore.system_timer;
160 	ti = task_thread_info(current);
161 	S390_lowcore.user_timer = ti->user_timer;
162 	S390_lowcore.system_timer = ti->system_timer;
163 }
164 
165 /*
166  * In s390, accounting pending user time also implies
167  * accounting system time in order to correctly compute
168  * the stolen time accounting.
169  */
170 void vtime_account_user(struct task_struct *tsk)
171 {
172 	if (do_account_vtime(tsk, HARDIRQ_OFFSET))
173 		virt_timer_expire();
174 }
175 
176 /*
177  * Update process times based on virtual cpu times stored by entry.S
178  * to the lowcore fields user_timer, system_timer & steal_clock.
179  */
180 void vtime_account_irq_enter(struct task_struct *tsk)
181 {
182 	struct thread_info *ti = task_thread_info(tsk);
183 	u64 timer, system, system_scaled;
184 
185 	timer = S390_lowcore.last_update_timer;
186 	S390_lowcore.last_update_timer = get_vtimer();
187 	S390_lowcore.system_timer += timer - S390_lowcore.last_update_timer;
188 
189 	/* Update MT utilization calculation */
190 	if (smp_cpu_mtid &&
191 	    time_after64(jiffies_64, this_cpu_read(mt_scaling_jiffies)))
192 		update_mt_scaling();
193 
194 	system = S390_lowcore.system_timer - ti->system_timer;
195 	S390_lowcore.steal_timer -= system;
196 	ti->system_timer = S390_lowcore.system_timer;
197 	system_scaled = system;
198 	/* Do MT utilization scaling */
199 	if (smp_cpu_mtid) {
200 		u64 mult = __this_cpu_read(mt_scaling_mult);
201 		u64 div = __this_cpu_read(mt_scaling_div);
202 
203 		system_scaled = (system_scaled * mult) / div;
204 	}
205 	account_system_time(tsk, 0, system, system_scaled);
206 
207 	virt_timer_forward(system);
208 }
209 EXPORT_SYMBOL_GPL(vtime_account_irq_enter);
210 
211 void vtime_account_system(struct task_struct *tsk)
212 __attribute__((alias("vtime_account_irq_enter")));
213 EXPORT_SYMBOL_GPL(vtime_account_system);
214 
215 /*
216  * Sorted add to a list. List is linear searched until first bigger
217  * element is found.
218  */
219 static void list_add_sorted(struct vtimer_list *timer, struct list_head *head)
220 {
221 	struct vtimer_list *tmp;
222 
223 	list_for_each_entry(tmp, head, entry) {
224 		if (tmp->expires > timer->expires) {
225 			list_add_tail(&timer->entry, &tmp->entry);
226 			return;
227 		}
228 	}
229 	list_add_tail(&timer->entry, head);
230 }
231 
232 /*
233  * Handler for expired virtual CPU timer.
234  */
235 static void virt_timer_expire(void)
236 {
237 	struct vtimer_list *timer, *tmp;
238 	unsigned long elapsed;
239 	LIST_HEAD(cb_list);
240 
241 	/* walk timer list, fire all expired timers */
242 	spin_lock(&virt_timer_lock);
243 	elapsed = atomic64_read(&virt_timer_elapsed);
244 	list_for_each_entry_safe(timer, tmp, &virt_timer_list, entry) {
245 		if (timer->expires < elapsed)
246 			/* move expired timer to the callback queue */
247 			list_move_tail(&timer->entry, &cb_list);
248 		else
249 			timer->expires -= elapsed;
250 	}
251 	if (!list_empty(&virt_timer_list)) {
252 		timer = list_first_entry(&virt_timer_list,
253 					 struct vtimer_list, entry);
254 		atomic64_set(&virt_timer_current, timer->expires);
255 	}
256 	atomic64_sub(elapsed, &virt_timer_elapsed);
257 	spin_unlock(&virt_timer_lock);
258 
259 	/* Do callbacks and recharge periodic timers */
260 	list_for_each_entry_safe(timer, tmp, &cb_list, entry) {
261 		list_del_init(&timer->entry);
262 		timer->function(timer->data);
263 		if (timer->interval) {
264 			/* Recharge interval timer */
265 			timer->expires = timer->interval +
266 				atomic64_read(&virt_timer_elapsed);
267 			spin_lock(&virt_timer_lock);
268 			list_add_sorted(timer, &virt_timer_list);
269 			spin_unlock(&virt_timer_lock);
270 		}
271 	}
272 }
273 
274 void init_virt_timer(struct vtimer_list *timer)
275 {
276 	timer->function = NULL;
277 	INIT_LIST_HEAD(&timer->entry);
278 }
279 EXPORT_SYMBOL(init_virt_timer);
280 
281 static inline int vtimer_pending(struct vtimer_list *timer)
282 {
283 	return !list_empty(&timer->entry);
284 }
285 
286 static void internal_add_vtimer(struct vtimer_list *timer)
287 {
288 	if (list_empty(&virt_timer_list)) {
289 		/* First timer, just program it. */
290 		atomic64_set(&virt_timer_current, timer->expires);
291 		atomic64_set(&virt_timer_elapsed, 0);
292 		list_add(&timer->entry, &virt_timer_list);
293 	} else {
294 		/* Update timer against current base. */
295 		timer->expires += atomic64_read(&virt_timer_elapsed);
296 		if (likely((s64) timer->expires <
297 			   (s64) atomic64_read(&virt_timer_current)))
298 			/* The new timer expires before the current timer. */
299 			atomic64_set(&virt_timer_current, timer->expires);
300 		/* Insert new timer into the list. */
301 		list_add_sorted(timer, &virt_timer_list);
302 	}
303 }
304 
305 static void __add_vtimer(struct vtimer_list *timer, int periodic)
306 {
307 	unsigned long flags;
308 
309 	timer->interval = periodic ? timer->expires : 0;
310 	spin_lock_irqsave(&virt_timer_lock, flags);
311 	internal_add_vtimer(timer);
312 	spin_unlock_irqrestore(&virt_timer_lock, flags);
313 }
314 
315 /*
316  * add_virt_timer - add an oneshot virtual CPU timer
317  */
318 void add_virt_timer(struct vtimer_list *timer)
319 {
320 	__add_vtimer(timer, 0);
321 }
322 EXPORT_SYMBOL(add_virt_timer);
323 
324 /*
325  * add_virt_timer_int - add an interval virtual CPU timer
326  */
327 void add_virt_timer_periodic(struct vtimer_list *timer)
328 {
329 	__add_vtimer(timer, 1);
330 }
331 EXPORT_SYMBOL(add_virt_timer_periodic);
332 
333 static int __mod_vtimer(struct vtimer_list *timer, u64 expires, int periodic)
334 {
335 	unsigned long flags;
336 	int rc;
337 
338 	BUG_ON(!timer->function);
339 
340 	if (timer->expires == expires && vtimer_pending(timer))
341 		return 1;
342 	spin_lock_irqsave(&virt_timer_lock, flags);
343 	rc = vtimer_pending(timer);
344 	if (rc)
345 		list_del_init(&timer->entry);
346 	timer->interval = periodic ? expires : 0;
347 	timer->expires = expires;
348 	internal_add_vtimer(timer);
349 	spin_unlock_irqrestore(&virt_timer_lock, flags);
350 	return rc;
351 }
352 
353 /*
354  * returns whether it has modified a pending timer (1) or not (0)
355  */
356 int mod_virt_timer(struct vtimer_list *timer, u64 expires)
357 {
358 	return __mod_vtimer(timer, expires, 0);
359 }
360 EXPORT_SYMBOL(mod_virt_timer);
361 
362 /*
363  * returns whether it has modified a pending timer (1) or not (0)
364  */
365 int mod_virt_timer_periodic(struct vtimer_list *timer, u64 expires)
366 {
367 	return __mod_vtimer(timer, expires, 1);
368 }
369 EXPORT_SYMBOL(mod_virt_timer_periodic);
370 
371 /*
372  * Delete a virtual timer.
373  *
374  * returns whether the deleted timer was pending (1) or not (0)
375  */
376 int del_virt_timer(struct vtimer_list *timer)
377 {
378 	unsigned long flags;
379 
380 	if (!vtimer_pending(timer))
381 		return 0;
382 	spin_lock_irqsave(&virt_timer_lock, flags);
383 	list_del_init(&timer->entry);
384 	spin_unlock_irqrestore(&virt_timer_lock, flags);
385 	return 1;
386 }
387 EXPORT_SYMBOL(del_virt_timer);
388 
389 /*
390  * Start the virtual CPU timer on the current CPU.
391  */
392 void vtime_init(void)
393 {
394 	/* set initial cpu timer */
395 	set_vtimer(VTIMER_MAX_SLICE);
396 	/* Setup initial MT scaling values */
397 	if (smp_cpu_mtid) {
398 		__this_cpu_write(mt_scaling_jiffies, jiffies);
399 		__this_cpu_write(mt_scaling_mult, 1);
400 		__this_cpu_write(mt_scaling_div, 1);
401 		stcctm5(smp_cpu_mtid + 1, this_cpu_ptr(mt_cycles));
402 	}
403 }
404