xref: /openbmc/linux/arch/x86/kvm/i8254.c (revision a0aace5ac0efdb2bcb71e10d9c9ca6a851fa59f9)
1 /*
2  * 8253/8254 interval timer emulation
3  *
4  * Copyright (c) 2003-2004 Fabrice Bellard
5  * Copyright (c) 2006 Intel Corporation
6  * Copyright (c) 2007 Keir Fraser, XenSource Inc
7  * Copyright (c) 2008 Intel Corporation
8  * Copyright 2009 Red Hat, Inc. and/or its affiliates.
9  *
10  * Permission is hereby granted, free of charge, to any person obtaining a copy
11  * of this software and associated documentation files (the "Software"), to deal
12  * in the Software without restriction, including without limitation the rights
13  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
14  * copies of the Software, and to permit persons to whom the Software is
15  * furnished to do so, subject to the following conditions:
16  *
17  * The above copyright notice and this permission notice shall be included in
18  * all copies or substantial portions of the Software.
19  *
20  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
21  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
22  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
23  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
24  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
25  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
26  * THE SOFTWARE.
27  *
28  * Authors:
29  *   Sheng Yang <sheng.yang@intel.com>
30  *   Based on QEMU and Xen.
31  */
32 
33 #define pr_fmt(fmt) "pit: " fmt
34 
35 #include <linux/kvm_host.h>
36 #include <linux/slab.h>
37 
38 #include "ioapic.h"
39 #include "irq.h"
40 #include "i8254.h"
41 #include "x86.h"
42 
43 #ifndef CONFIG_X86_64
44 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
45 #else
46 #define mod_64(x, y) ((x) % (y))
47 #endif
48 
49 #define RW_STATE_LSB 1
50 #define RW_STATE_MSB 2
51 #define RW_STATE_WORD0 3
52 #define RW_STATE_WORD1 4
53 
54 /* Compute with 96 bit intermediate result: (a*b)/c */
55 static u64 muldiv64(u64 a, u32 b, u32 c)
56 {
57 	union {
58 		u64 ll;
59 		struct {
60 			u32 low, high;
61 		} l;
62 	} u, res;
63 	u64 rl, rh;
64 
65 	u.ll = a;
66 	rl = (u64)u.l.low * (u64)b;
67 	rh = (u64)u.l.high * (u64)b;
68 	rh += (rl >> 32);
69 	res.l.high = div64_u64(rh, c);
70 	res.l.low = div64_u64(((mod_64(rh, c) << 32) + (rl & 0xffffffff)), c);
71 	return res.ll;
72 }
73 
74 static void pit_set_gate(struct kvm_pit *pit, int channel, u32 val)
75 {
76 	struct kvm_kpit_channel_state *c = &pit->pit_state.channels[channel];
77 
78 	switch (c->mode) {
79 	default:
80 	case 0:
81 	case 4:
82 		/* XXX: just disable/enable counting */
83 		break;
84 	case 1:
85 	case 2:
86 	case 3:
87 	case 5:
88 		/* Restart counting on rising edge. */
89 		if (c->gate < val)
90 			c->count_load_time = ktime_get();
91 		break;
92 	}
93 
94 	c->gate = val;
95 }
96 
97 static int pit_get_gate(struct kvm_pit *pit, int channel)
98 {
99 	return pit->pit_state.channels[channel].gate;
100 }
101 
102 static s64 __kpit_elapsed(struct kvm_pit *pit)
103 {
104 	s64 elapsed;
105 	ktime_t remaining;
106 	struct kvm_kpit_state *ps = &pit->pit_state;
107 
108 	if (!ps->period)
109 		return 0;
110 
111 	/*
112 	 * The Counter does not stop when it reaches zero. In
113 	 * Modes 0, 1, 4, and 5 the Counter ``wraps around'' to
114 	 * the highest count, either FFFF hex for binary counting
115 	 * or 9999 for BCD counting, and continues counting.
116 	 * Modes 2 and 3 are periodic; the Counter reloads
117 	 * itself with the initial count and continues counting
118 	 * from there.
119 	 */
120 	remaining = hrtimer_get_remaining(&ps->timer);
121 	elapsed = ps->period - ktime_to_ns(remaining);
122 
123 	return elapsed;
124 }
125 
126 static s64 kpit_elapsed(struct kvm_pit *pit, struct kvm_kpit_channel_state *c,
127 			int channel)
128 {
129 	if (channel == 0)
130 		return __kpit_elapsed(pit);
131 
132 	return ktime_to_ns(ktime_sub(ktime_get(), c->count_load_time));
133 }
134 
135 static int pit_get_count(struct kvm_pit *pit, int channel)
136 {
137 	struct kvm_kpit_channel_state *c = &pit->pit_state.channels[channel];
138 	s64 d, t;
139 	int counter;
140 
141 	t = kpit_elapsed(pit, c, channel);
142 	d = muldiv64(t, KVM_PIT_FREQ, NSEC_PER_SEC);
143 
144 	switch (c->mode) {
145 	case 0:
146 	case 1:
147 	case 4:
148 	case 5:
149 		counter = (c->count - d) & 0xffff;
150 		break;
151 	case 3:
152 		/* XXX: may be incorrect for odd counts */
153 		counter = c->count - (mod_64((2 * d), c->count));
154 		break;
155 	default:
156 		counter = c->count - mod_64(d, c->count);
157 		break;
158 	}
159 	return counter;
160 }
161 
162 static int pit_get_out(struct kvm_pit *pit, int channel)
163 {
164 	struct kvm_kpit_channel_state *c = &pit->pit_state.channels[channel];
165 	s64 d, t;
166 	int out;
167 
168 	t = kpit_elapsed(pit, c, channel);
169 	d = muldiv64(t, KVM_PIT_FREQ, NSEC_PER_SEC);
170 
171 	switch (c->mode) {
172 	default:
173 	case 0:
174 		out = (d >= c->count);
175 		break;
176 	case 1:
177 		out = (d < c->count);
178 		break;
179 	case 2:
180 		out = ((mod_64(d, c->count) == 0) && (d != 0));
181 		break;
182 	case 3:
183 		out = (mod_64(d, c->count) < ((c->count + 1) >> 1));
184 		break;
185 	case 4:
186 	case 5:
187 		out = (d == c->count);
188 		break;
189 	}
190 
191 	return out;
192 }
193 
194 static void pit_latch_count(struct kvm_pit *pit, int channel)
195 {
196 	struct kvm_kpit_channel_state *c = &pit->pit_state.channels[channel];
197 
198 	if (!c->count_latched) {
199 		c->latched_count = pit_get_count(pit, channel);
200 		c->count_latched = c->rw_mode;
201 	}
202 }
203 
204 static void pit_latch_status(struct kvm_pit *pit, int channel)
205 {
206 	struct kvm_kpit_channel_state *c = &pit->pit_state.channels[channel];
207 
208 	if (!c->status_latched) {
209 		/* TODO: Return NULL COUNT (bit 6). */
210 		c->status = ((pit_get_out(pit, channel) << 7) |
211 				(c->rw_mode << 4) |
212 				(c->mode << 1) |
213 				c->bcd);
214 		c->status_latched = 1;
215 	}
216 }
217 
218 static inline struct kvm_pit *pit_state_to_pit(struct kvm_kpit_state *ps)
219 {
220 	return container_of(ps, struct kvm_pit, pit_state);
221 }
222 
223 static void kvm_pit_ack_irq(struct kvm_irq_ack_notifier *kian)
224 {
225 	struct kvm_kpit_state *ps = container_of(kian, struct kvm_kpit_state,
226 						 irq_ack_notifier);
227 	struct kvm_pit *pit = pit_state_to_pit(ps);
228 
229 	atomic_set(&ps->irq_ack, 1);
230 	/* irq_ack should be set before pending is read.  Order accesses with
231 	 * inc(pending) in pit_timer_fn and xchg(irq_ack, 0) in pit_do_work.
232 	 */
233 	smp_mb();
234 	if (atomic_dec_if_positive(&ps->pending) > 0)
235 		queue_kthread_work(&pit->worker, &pit->expired);
236 }
237 
238 void __kvm_migrate_pit_timer(struct kvm_vcpu *vcpu)
239 {
240 	struct kvm_pit *pit = vcpu->kvm->arch.vpit;
241 	struct hrtimer *timer;
242 
243 	if (!kvm_vcpu_is_bsp(vcpu) || !pit)
244 		return;
245 
246 	timer = &pit->pit_state.timer;
247 	mutex_lock(&pit->pit_state.lock);
248 	if (hrtimer_cancel(timer))
249 		hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
250 	mutex_unlock(&pit->pit_state.lock);
251 }
252 
253 static void destroy_pit_timer(struct kvm_pit *pit)
254 {
255 	hrtimer_cancel(&pit->pit_state.timer);
256 	flush_kthread_work(&pit->expired);
257 }
258 
259 static void pit_do_work(struct kthread_work *work)
260 {
261 	struct kvm_pit *pit = container_of(work, struct kvm_pit, expired);
262 	struct kvm *kvm = pit->kvm;
263 	struct kvm_vcpu *vcpu;
264 	int i;
265 	struct kvm_kpit_state *ps = &pit->pit_state;
266 
267 	if (atomic_read(&ps->reinject) && !atomic_xchg(&ps->irq_ack, 0))
268 		return;
269 
270 	kvm_set_irq(kvm, pit->irq_source_id, 0, 1, false);
271 	kvm_set_irq(kvm, pit->irq_source_id, 0, 0, false);
272 
273 	/*
274 	 * Provides NMI watchdog support via Virtual Wire mode.
275 	 * The route is: PIT -> LVT0 in NMI mode.
276 	 *
277 	 * Note: Our Virtual Wire implementation does not follow
278 	 * the MP specification.  We propagate a PIT interrupt to all
279 	 * VCPUs and only when LVT0 is in NMI mode.  The interrupt can
280 	 * also be simultaneously delivered through PIC and IOAPIC.
281 	 */
282 	if (atomic_read(&kvm->arch.vapics_in_nmi_mode) > 0)
283 		kvm_for_each_vcpu(i, vcpu, kvm)
284 			kvm_apic_nmi_wd_deliver(vcpu);
285 }
286 
287 static enum hrtimer_restart pit_timer_fn(struct hrtimer *data)
288 {
289 	struct kvm_kpit_state *ps = container_of(data, struct kvm_kpit_state, timer);
290 	struct kvm_pit *pt = pit_state_to_pit(ps);
291 
292 	if (atomic_read(&ps->reinject))
293 		atomic_inc(&ps->pending);
294 
295 	queue_kthread_work(&pt->worker, &pt->expired);
296 
297 	if (ps->is_periodic) {
298 		hrtimer_add_expires_ns(&ps->timer, ps->period);
299 		return HRTIMER_RESTART;
300 	} else
301 		return HRTIMER_NORESTART;
302 }
303 
304 static inline void kvm_pit_reset_reinject(struct kvm_pit *pit)
305 {
306 	atomic_set(&pit->pit_state.pending, 0);
307 	atomic_set(&pit->pit_state.irq_ack, 1);
308 }
309 
310 void kvm_pit_set_reinject(struct kvm_pit *pit, bool reinject)
311 {
312 	struct kvm_kpit_state *ps = &pit->pit_state;
313 	struct kvm *kvm = pit->kvm;
314 
315 	if (atomic_read(&ps->reinject) == reinject)
316 		return;
317 
318 	if (reinject) {
319 		/* The initial state is preserved while ps->reinject == 0. */
320 		kvm_pit_reset_reinject(pit);
321 		kvm_register_irq_ack_notifier(kvm, &ps->irq_ack_notifier);
322 		kvm_register_irq_mask_notifier(kvm, 0, &pit->mask_notifier);
323 	} else {
324 		kvm_unregister_irq_ack_notifier(kvm, &ps->irq_ack_notifier);
325 		kvm_unregister_irq_mask_notifier(kvm, 0, &pit->mask_notifier);
326 	}
327 
328 	atomic_set(&ps->reinject, reinject);
329 }
330 
331 static void create_pit_timer(struct kvm_pit *pit, u32 val, int is_period)
332 {
333 	struct kvm_kpit_state *ps = &pit->pit_state;
334 	struct kvm *kvm = pit->kvm;
335 	s64 interval;
336 
337 	if (!ioapic_in_kernel(kvm) ||
338 	    ps->flags & KVM_PIT_FLAGS_HPET_LEGACY)
339 		return;
340 
341 	interval = muldiv64(val, NSEC_PER_SEC, KVM_PIT_FREQ);
342 
343 	pr_debug("create pit timer, interval is %llu nsec\n", interval);
344 
345 	/* TODO The new value only affected after the retriggered */
346 	hrtimer_cancel(&ps->timer);
347 	flush_kthread_work(&pit->expired);
348 	ps->period = interval;
349 	ps->is_periodic = is_period;
350 
351 	kvm_pit_reset_reinject(pit);
352 
353 	/*
354 	 * Do not allow the guest to program periodic timers with small
355 	 * interval, since the hrtimers are not throttled by the host
356 	 * scheduler.
357 	 */
358 	if (ps->is_periodic) {
359 		s64 min_period = min_timer_period_us * 1000LL;
360 
361 		if (ps->period < min_period) {
362 			pr_info_ratelimited(
363 			    "kvm: requested %lld ns "
364 			    "i8254 timer period limited to %lld ns\n",
365 			    ps->period, min_period);
366 			ps->period = min_period;
367 		}
368 	}
369 
370 	hrtimer_start(&ps->timer, ktime_add_ns(ktime_get(), interval),
371 		      HRTIMER_MODE_ABS);
372 }
373 
374 static void pit_load_count(struct kvm_pit *pit, int channel, u32 val)
375 {
376 	struct kvm_kpit_state *ps = &pit->pit_state;
377 
378 	pr_debug("load_count val is %d, channel is %d\n", val, channel);
379 
380 	/*
381 	 * The largest possible initial count is 0; this is equivalent
382 	 * to 216 for binary counting and 104 for BCD counting.
383 	 */
384 	if (val == 0)
385 		val = 0x10000;
386 
387 	ps->channels[channel].count = val;
388 
389 	if (channel != 0) {
390 		ps->channels[channel].count_load_time = ktime_get();
391 		return;
392 	}
393 
394 	/* Two types of timer
395 	 * mode 1 is one shot, mode 2 is period, otherwise del timer */
396 	switch (ps->channels[0].mode) {
397 	case 0:
398 	case 1:
399         /* FIXME: enhance mode 4 precision */
400 	case 4:
401 		create_pit_timer(pit, val, 0);
402 		break;
403 	case 2:
404 	case 3:
405 		create_pit_timer(pit, val, 1);
406 		break;
407 	default:
408 		destroy_pit_timer(pit);
409 	}
410 }
411 
412 void kvm_pit_load_count(struct kvm_pit *pit, int channel, u32 val,
413 		int hpet_legacy_start)
414 {
415 	u8 saved_mode;
416 
417 	WARN_ON_ONCE(!mutex_is_locked(&pit->pit_state.lock));
418 
419 	if (hpet_legacy_start) {
420 		/* save existing mode for later reenablement */
421 		WARN_ON(channel != 0);
422 		saved_mode = pit->pit_state.channels[0].mode;
423 		pit->pit_state.channels[0].mode = 0xff; /* disable timer */
424 		pit_load_count(pit, channel, val);
425 		pit->pit_state.channels[0].mode = saved_mode;
426 	} else {
427 		pit_load_count(pit, channel, val);
428 	}
429 }
430 
431 static inline struct kvm_pit *dev_to_pit(struct kvm_io_device *dev)
432 {
433 	return container_of(dev, struct kvm_pit, dev);
434 }
435 
436 static inline struct kvm_pit *speaker_to_pit(struct kvm_io_device *dev)
437 {
438 	return container_of(dev, struct kvm_pit, speaker_dev);
439 }
440 
441 static inline int pit_in_range(gpa_t addr)
442 {
443 	return ((addr >= KVM_PIT_BASE_ADDRESS) &&
444 		(addr < KVM_PIT_BASE_ADDRESS + KVM_PIT_MEM_LENGTH));
445 }
446 
447 static int pit_ioport_write(struct kvm_vcpu *vcpu,
448 				struct kvm_io_device *this,
449 			    gpa_t addr, int len, const void *data)
450 {
451 	struct kvm_pit *pit = dev_to_pit(this);
452 	struct kvm_kpit_state *pit_state = &pit->pit_state;
453 	int channel, access;
454 	struct kvm_kpit_channel_state *s;
455 	u32 val = *(u32 *) data;
456 	if (!pit_in_range(addr))
457 		return -EOPNOTSUPP;
458 
459 	val  &= 0xff;
460 	addr &= KVM_PIT_CHANNEL_MASK;
461 
462 	mutex_lock(&pit_state->lock);
463 
464 	if (val != 0)
465 		pr_debug("write addr is 0x%x, len is %d, val is 0x%x\n",
466 			 (unsigned int)addr, len, val);
467 
468 	if (addr == 3) {
469 		channel = val >> 6;
470 		if (channel == 3) {
471 			/* Read-Back Command. */
472 			for (channel = 0; channel < 3; channel++) {
473 				s = &pit_state->channels[channel];
474 				if (val & (2 << channel)) {
475 					if (!(val & 0x20))
476 						pit_latch_count(pit, channel);
477 					if (!(val & 0x10))
478 						pit_latch_status(pit, channel);
479 				}
480 			}
481 		} else {
482 			/* Select Counter <channel>. */
483 			s = &pit_state->channels[channel];
484 			access = (val >> 4) & KVM_PIT_CHANNEL_MASK;
485 			if (access == 0) {
486 				pit_latch_count(pit, channel);
487 			} else {
488 				s->rw_mode = access;
489 				s->read_state = access;
490 				s->write_state = access;
491 				s->mode = (val >> 1) & 7;
492 				if (s->mode > 5)
493 					s->mode -= 4;
494 				s->bcd = val & 1;
495 			}
496 		}
497 	} else {
498 		/* Write Count. */
499 		s = &pit_state->channels[addr];
500 		switch (s->write_state) {
501 		default:
502 		case RW_STATE_LSB:
503 			pit_load_count(pit, addr, val);
504 			break;
505 		case RW_STATE_MSB:
506 			pit_load_count(pit, addr, val << 8);
507 			break;
508 		case RW_STATE_WORD0:
509 			s->write_latch = val;
510 			s->write_state = RW_STATE_WORD1;
511 			break;
512 		case RW_STATE_WORD1:
513 			pit_load_count(pit, addr, s->write_latch | (val << 8));
514 			s->write_state = RW_STATE_WORD0;
515 			break;
516 		}
517 	}
518 
519 	mutex_unlock(&pit_state->lock);
520 	return 0;
521 }
522 
523 static int pit_ioport_read(struct kvm_vcpu *vcpu,
524 			   struct kvm_io_device *this,
525 			   gpa_t addr, int len, void *data)
526 {
527 	struct kvm_pit *pit = dev_to_pit(this);
528 	struct kvm_kpit_state *pit_state = &pit->pit_state;
529 	int ret, count;
530 	struct kvm_kpit_channel_state *s;
531 	if (!pit_in_range(addr))
532 		return -EOPNOTSUPP;
533 
534 	addr &= KVM_PIT_CHANNEL_MASK;
535 	if (addr == 3)
536 		return 0;
537 
538 	s = &pit_state->channels[addr];
539 
540 	mutex_lock(&pit_state->lock);
541 
542 	if (s->status_latched) {
543 		s->status_latched = 0;
544 		ret = s->status;
545 	} else if (s->count_latched) {
546 		switch (s->count_latched) {
547 		default:
548 		case RW_STATE_LSB:
549 			ret = s->latched_count & 0xff;
550 			s->count_latched = 0;
551 			break;
552 		case RW_STATE_MSB:
553 			ret = s->latched_count >> 8;
554 			s->count_latched = 0;
555 			break;
556 		case RW_STATE_WORD0:
557 			ret = s->latched_count & 0xff;
558 			s->count_latched = RW_STATE_MSB;
559 			break;
560 		}
561 	} else {
562 		switch (s->read_state) {
563 		default:
564 		case RW_STATE_LSB:
565 			count = pit_get_count(pit, addr);
566 			ret = count & 0xff;
567 			break;
568 		case RW_STATE_MSB:
569 			count = pit_get_count(pit, addr);
570 			ret = (count >> 8) & 0xff;
571 			break;
572 		case RW_STATE_WORD0:
573 			count = pit_get_count(pit, addr);
574 			ret = count & 0xff;
575 			s->read_state = RW_STATE_WORD1;
576 			break;
577 		case RW_STATE_WORD1:
578 			count = pit_get_count(pit, addr);
579 			ret = (count >> 8) & 0xff;
580 			s->read_state = RW_STATE_WORD0;
581 			break;
582 		}
583 	}
584 
585 	if (len > sizeof(ret))
586 		len = sizeof(ret);
587 	memcpy(data, (char *)&ret, len);
588 
589 	mutex_unlock(&pit_state->lock);
590 	return 0;
591 }
592 
593 static int speaker_ioport_write(struct kvm_vcpu *vcpu,
594 				struct kvm_io_device *this,
595 				gpa_t addr, int len, const void *data)
596 {
597 	struct kvm_pit *pit = speaker_to_pit(this);
598 	struct kvm_kpit_state *pit_state = &pit->pit_state;
599 	u32 val = *(u32 *) data;
600 	if (addr != KVM_SPEAKER_BASE_ADDRESS)
601 		return -EOPNOTSUPP;
602 
603 	mutex_lock(&pit_state->lock);
604 	pit_state->speaker_data_on = (val >> 1) & 1;
605 	pit_set_gate(pit, 2, val & 1);
606 	mutex_unlock(&pit_state->lock);
607 	return 0;
608 }
609 
610 static int speaker_ioport_read(struct kvm_vcpu *vcpu,
611 				   struct kvm_io_device *this,
612 				   gpa_t addr, int len, void *data)
613 {
614 	struct kvm_pit *pit = speaker_to_pit(this);
615 	struct kvm_kpit_state *pit_state = &pit->pit_state;
616 	unsigned int refresh_clock;
617 	int ret;
618 	if (addr != KVM_SPEAKER_BASE_ADDRESS)
619 		return -EOPNOTSUPP;
620 
621 	/* Refresh clock toggles at about 15us. We approximate as 2^14ns. */
622 	refresh_clock = ((unsigned int)ktime_to_ns(ktime_get()) >> 14) & 1;
623 
624 	mutex_lock(&pit_state->lock);
625 	ret = ((pit_state->speaker_data_on << 1) | pit_get_gate(pit, 2) |
626 		(pit_get_out(pit, 2) << 5) | (refresh_clock << 4));
627 	if (len > sizeof(ret))
628 		len = sizeof(ret);
629 	memcpy(data, (char *)&ret, len);
630 	mutex_unlock(&pit_state->lock);
631 	return 0;
632 }
633 
634 static void kvm_pit_reset(struct kvm_pit *pit)
635 {
636 	int i;
637 	struct kvm_kpit_channel_state *c;
638 
639 	pit->pit_state.flags = 0;
640 	for (i = 0; i < 3; i++) {
641 		c = &pit->pit_state.channels[i];
642 		c->mode = 0xff;
643 		c->gate = (i != 2);
644 		pit_load_count(pit, i, 0);
645 	}
646 
647 	kvm_pit_reset_reinject(pit);
648 }
649 
650 static void pit_mask_notifer(struct kvm_irq_mask_notifier *kimn, bool mask)
651 {
652 	struct kvm_pit *pit = container_of(kimn, struct kvm_pit, mask_notifier);
653 
654 	if (!mask)
655 		kvm_pit_reset_reinject(pit);
656 }
657 
658 static const struct kvm_io_device_ops pit_dev_ops = {
659 	.read     = pit_ioport_read,
660 	.write    = pit_ioport_write,
661 };
662 
663 static const struct kvm_io_device_ops speaker_dev_ops = {
664 	.read     = speaker_ioport_read,
665 	.write    = speaker_ioport_write,
666 };
667 
668 /* Caller must hold slots_lock */
669 struct kvm_pit *kvm_create_pit(struct kvm *kvm, u32 flags)
670 {
671 	struct kvm_pit *pit;
672 	struct kvm_kpit_state *pit_state;
673 	struct pid *pid;
674 	pid_t pid_nr;
675 	int ret;
676 
677 	pit = kzalloc(sizeof(struct kvm_pit), GFP_KERNEL);
678 	if (!pit)
679 		return NULL;
680 
681 	pit->irq_source_id = kvm_request_irq_source_id(kvm);
682 	if (pit->irq_source_id < 0)
683 		goto fail_request;
684 
685 	mutex_init(&pit->pit_state.lock);
686 
687 	pid = get_pid(task_tgid(current));
688 	pid_nr = pid_vnr(pid);
689 	put_pid(pid);
690 
691 	init_kthread_worker(&pit->worker);
692 	pit->worker_task = kthread_run(kthread_worker_fn, &pit->worker,
693 				       "kvm-pit/%d", pid_nr);
694 	if (IS_ERR(pit->worker_task))
695 		goto fail_kthread;
696 
697 	init_kthread_work(&pit->expired, pit_do_work);
698 
699 	pit->kvm = kvm;
700 
701 	pit_state = &pit->pit_state;
702 	hrtimer_init(&pit_state->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
703 	pit_state->timer.function = pit_timer_fn;
704 
705 	pit_state->irq_ack_notifier.gsi = 0;
706 	pit_state->irq_ack_notifier.irq_acked = kvm_pit_ack_irq;
707 	pit->mask_notifier.func = pit_mask_notifer;
708 
709 	kvm_pit_reset(pit);
710 
711 	kvm_pit_set_reinject(pit, true);
712 
713 	kvm_iodevice_init(&pit->dev, &pit_dev_ops);
714 	ret = kvm_io_bus_register_dev(kvm, KVM_PIO_BUS, KVM_PIT_BASE_ADDRESS,
715 				      KVM_PIT_MEM_LENGTH, &pit->dev);
716 	if (ret < 0)
717 		goto fail_register_pit;
718 
719 	if (flags & KVM_PIT_SPEAKER_DUMMY) {
720 		kvm_iodevice_init(&pit->speaker_dev, &speaker_dev_ops);
721 		ret = kvm_io_bus_register_dev(kvm, KVM_PIO_BUS,
722 					      KVM_SPEAKER_BASE_ADDRESS, 4,
723 					      &pit->speaker_dev);
724 		if (ret < 0)
725 			goto fail_register_speaker;
726 	}
727 
728 	return pit;
729 
730 fail_register_speaker:
731 	kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &pit->dev);
732 fail_register_pit:
733 	kvm_pit_set_reinject(pit, false);
734 	kthread_stop(pit->worker_task);
735 fail_kthread:
736 	kvm_free_irq_source_id(kvm, pit->irq_source_id);
737 fail_request:
738 	kfree(pit);
739 	return NULL;
740 }
741 
742 void kvm_free_pit(struct kvm *kvm)
743 {
744 	struct kvm_pit *pit = kvm->arch.vpit;
745 
746 	if (pit) {
747 		kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &pit->dev);
748 		kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &pit->speaker_dev);
749 		kvm_pit_set_reinject(pit, false);
750 		hrtimer_cancel(&pit->pit_state.timer);
751 		flush_kthread_work(&pit->expired);
752 		kthread_stop(pit->worker_task);
753 		kvm_free_irq_source_id(kvm, pit->irq_source_id);
754 		kfree(pit);
755 	}
756 }
757