xref: /linux/arch/x86/kvm/i8254.c (revision b85d45947951d23cb22d90caecf4c1eb81342c96)
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 "irq.h"
39 #include "i8254.h"
40 #include "x86.h"
41 
42 #ifndef CONFIG_X86_64
43 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
44 #else
45 #define mod_64(x, y) ((x) % (y))
46 #endif
47 
48 #define RW_STATE_LSB 1
49 #define RW_STATE_MSB 2
50 #define RW_STATE_WORD0 3
51 #define RW_STATE_WORD1 4
52 
53 /* Compute with 96 bit intermediate result: (a*b)/c */
54 static u64 muldiv64(u64 a, u32 b, u32 c)
55 {
56 	union {
57 		u64 ll;
58 		struct {
59 			u32 low, high;
60 		} l;
61 	} u, res;
62 	u64 rl, rh;
63 
64 	u.ll = a;
65 	rl = (u64)u.l.low * (u64)b;
66 	rh = (u64)u.l.high * (u64)b;
67 	rh += (rl >> 32);
68 	res.l.high = div64_u64(rh, c);
69 	res.l.low = div64_u64(((mod_64(rh, c) << 32) + (rl & 0xffffffff)), c);
70 	return res.ll;
71 }
72 
73 static void pit_set_gate(struct kvm *kvm, int channel, u32 val)
74 {
75 	struct kvm_kpit_channel_state *c =
76 		&kvm->arch.vpit->pit_state.channels[channel];
77 
78 	WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
79 
80 	switch (c->mode) {
81 	default:
82 	case 0:
83 	case 4:
84 		/* XXX: just disable/enable counting */
85 		break;
86 	case 1:
87 	case 2:
88 	case 3:
89 	case 5:
90 		/* Restart counting on rising edge. */
91 		if (c->gate < val)
92 			c->count_load_time = ktime_get();
93 		break;
94 	}
95 
96 	c->gate = val;
97 }
98 
99 static int pit_get_gate(struct kvm *kvm, int channel)
100 {
101 	WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
102 
103 	return kvm->arch.vpit->pit_state.channels[channel].gate;
104 }
105 
106 static s64 __kpit_elapsed(struct kvm *kvm)
107 {
108 	s64 elapsed;
109 	ktime_t remaining;
110 	struct kvm_kpit_state *ps = &kvm->arch.vpit->pit_state;
111 
112 	if (!ps->period)
113 		return 0;
114 
115 	/*
116 	 * The Counter does not stop when it reaches zero. In
117 	 * Modes 0, 1, 4, and 5 the Counter ``wraps around'' to
118 	 * the highest count, either FFFF hex for binary counting
119 	 * or 9999 for BCD counting, and continues counting.
120 	 * Modes 2 and 3 are periodic; the Counter reloads
121 	 * itself with the initial count and continues counting
122 	 * from there.
123 	 */
124 	remaining = hrtimer_get_remaining(&ps->timer);
125 	elapsed = ps->period - ktime_to_ns(remaining);
126 
127 	return elapsed;
128 }
129 
130 static s64 kpit_elapsed(struct kvm *kvm, struct kvm_kpit_channel_state *c,
131 			int channel)
132 {
133 	if (channel == 0)
134 		return __kpit_elapsed(kvm);
135 
136 	return ktime_to_ns(ktime_sub(ktime_get(), c->count_load_time));
137 }
138 
139 static int pit_get_count(struct kvm *kvm, int channel)
140 {
141 	struct kvm_kpit_channel_state *c =
142 		&kvm->arch.vpit->pit_state.channels[channel];
143 	s64 d, t;
144 	int counter;
145 
146 	WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
147 
148 	t = kpit_elapsed(kvm, c, channel);
149 	d = muldiv64(t, KVM_PIT_FREQ, NSEC_PER_SEC);
150 
151 	switch (c->mode) {
152 	case 0:
153 	case 1:
154 	case 4:
155 	case 5:
156 		counter = (c->count - d) & 0xffff;
157 		break;
158 	case 3:
159 		/* XXX: may be incorrect for odd counts */
160 		counter = c->count - (mod_64((2 * d), c->count));
161 		break;
162 	default:
163 		counter = c->count - mod_64(d, c->count);
164 		break;
165 	}
166 	return counter;
167 }
168 
169 static int pit_get_out(struct kvm *kvm, int channel)
170 {
171 	struct kvm_kpit_channel_state *c =
172 		&kvm->arch.vpit->pit_state.channels[channel];
173 	s64 d, t;
174 	int out;
175 
176 	WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
177 
178 	t = kpit_elapsed(kvm, c, channel);
179 	d = muldiv64(t, KVM_PIT_FREQ, NSEC_PER_SEC);
180 
181 	switch (c->mode) {
182 	default:
183 	case 0:
184 		out = (d >= c->count);
185 		break;
186 	case 1:
187 		out = (d < c->count);
188 		break;
189 	case 2:
190 		out = ((mod_64(d, c->count) == 0) && (d != 0));
191 		break;
192 	case 3:
193 		out = (mod_64(d, c->count) < ((c->count + 1) >> 1));
194 		break;
195 	case 4:
196 	case 5:
197 		out = (d == c->count);
198 		break;
199 	}
200 
201 	return out;
202 }
203 
204 static void pit_latch_count(struct kvm *kvm, int channel)
205 {
206 	struct kvm_kpit_channel_state *c =
207 		&kvm->arch.vpit->pit_state.channels[channel];
208 
209 	WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
210 
211 	if (!c->count_latched) {
212 		c->latched_count = pit_get_count(kvm, channel);
213 		c->count_latched = c->rw_mode;
214 	}
215 }
216 
217 static void pit_latch_status(struct kvm *kvm, int channel)
218 {
219 	struct kvm_kpit_channel_state *c =
220 		&kvm->arch.vpit->pit_state.channels[channel];
221 
222 	WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
223 
224 	if (!c->status_latched) {
225 		/* TODO: Return NULL COUNT (bit 6). */
226 		c->status = ((pit_get_out(kvm, channel) << 7) |
227 				(c->rw_mode << 4) |
228 				(c->mode << 1) |
229 				c->bcd);
230 		c->status_latched = 1;
231 	}
232 }
233 
234 static void kvm_pit_ack_irq(struct kvm_irq_ack_notifier *kian)
235 {
236 	struct kvm_kpit_state *ps = container_of(kian, struct kvm_kpit_state,
237 						 irq_ack_notifier);
238 	int value;
239 
240 	spin_lock(&ps->inject_lock);
241 	value = atomic_dec_return(&ps->pending);
242 	if (value < 0)
243 		/* spurious acks can be generated if, for example, the
244 		 * PIC is being reset.  Handle it gracefully here
245 		 */
246 		atomic_inc(&ps->pending);
247 	else if (value > 0)
248 		/* in this case, we had multiple outstanding pit interrupts
249 		 * that we needed to inject.  Reinject
250 		 */
251 		queue_kthread_work(&ps->pit->worker, &ps->pit->expired);
252 	ps->irq_ack = 1;
253 	spin_unlock(&ps->inject_lock);
254 }
255 
256 void __kvm_migrate_pit_timer(struct kvm_vcpu *vcpu)
257 {
258 	struct kvm_pit *pit = vcpu->kvm->arch.vpit;
259 	struct hrtimer *timer;
260 
261 	if (!kvm_vcpu_is_bsp(vcpu) || !pit)
262 		return;
263 
264 	timer = &pit->pit_state.timer;
265 	mutex_lock(&pit->pit_state.lock);
266 	if (hrtimer_cancel(timer))
267 		hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
268 	mutex_unlock(&pit->pit_state.lock);
269 }
270 
271 static void destroy_pit_timer(struct kvm_pit *pit)
272 {
273 	hrtimer_cancel(&pit->pit_state.timer);
274 	flush_kthread_work(&pit->expired);
275 }
276 
277 static void pit_do_work(struct kthread_work *work)
278 {
279 	struct kvm_pit *pit = container_of(work, struct kvm_pit, expired);
280 	struct kvm *kvm = pit->kvm;
281 	struct kvm_vcpu *vcpu;
282 	int i;
283 	struct kvm_kpit_state *ps = &pit->pit_state;
284 	int inject = 0;
285 
286 	/* Try to inject pending interrupts when
287 	 * last one has been acked.
288 	 */
289 	spin_lock(&ps->inject_lock);
290 	if (ps->irq_ack) {
291 		ps->irq_ack = 0;
292 		inject = 1;
293 	}
294 	spin_unlock(&ps->inject_lock);
295 	if (inject) {
296 		kvm_set_irq(kvm, kvm->arch.vpit->irq_source_id, 0, 1, false);
297 		kvm_set_irq(kvm, kvm->arch.vpit->irq_source_id, 0, 0, false);
298 
299 		/*
300 		 * Provides NMI watchdog support via Virtual Wire mode.
301 		 * The route is: PIT -> PIC -> LVT0 in NMI mode.
302 		 *
303 		 * Note: Our Virtual Wire implementation is simplified, only
304 		 * propagating PIT interrupts to all VCPUs when they have set
305 		 * LVT0 to NMI delivery. Other PIC interrupts are just sent to
306 		 * VCPU0, and only if its LVT0 is in EXTINT mode.
307 		 */
308 		if (atomic_read(&kvm->arch.vapics_in_nmi_mode) > 0)
309 			kvm_for_each_vcpu(i, vcpu, kvm)
310 				kvm_apic_nmi_wd_deliver(vcpu);
311 	}
312 }
313 
314 static enum hrtimer_restart pit_timer_fn(struct hrtimer *data)
315 {
316 	struct kvm_kpit_state *ps = container_of(data, struct kvm_kpit_state, timer);
317 	struct kvm_pit *pt = ps->kvm->arch.vpit;
318 
319 	if (ps->reinject || !atomic_read(&ps->pending)) {
320 		atomic_inc(&ps->pending);
321 		queue_kthread_work(&pt->worker, &pt->expired);
322 	}
323 
324 	if (ps->is_periodic) {
325 		hrtimer_add_expires_ns(&ps->timer, ps->period);
326 		return HRTIMER_RESTART;
327 	} else
328 		return HRTIMER_NORESTART;
329 }
330 
331 static void create_pit_timer(struct kvm *kvm, u32 val, int is_period)
332 {
333 	struct kvm_kpit_state *ps = &kvm->arch.vpit->pit_state;
334 	s64 interval;
335 
336 	if (!irqchip_in_kernel(kvm) || ps->flags & KVM_PIT_FLAGS_HPET_LEGACY)
337 		return;
338 
339 	interval = muldiv64(val, NSEC_PER_SEC, KVM_PIT_FREQ);
340 
341 	pr_debug("create pit timer, interval is %llu nsec\n", interval);
342 
343 	/* TODO The new value only affected after the retriggered */
344 	hrtimer_cancel(&ps->timer);
345 	flush_kthread_work(&ps->pit->expired);
346 	ps->period = interval;
347 	ps->is_periodic = is_period;
348 
349 	ps->timer.function = pit_timer_fn;
350 	ps->kvm = ps->pit->kvm;
351 
352 	atomic_set(&ps->pending, 0);
353 	ps->irq_ack = 1;
354 
355 	/*
356 	 * Do not allow the guest to program periodic timers with small
357 	 * interval, since the hrtimers are not throttled by the host
358 	 * scheduler.
359 	 */
360 	if (ps->is_periodic) {
361 		s64 min_period = min_timer_period_us * 1000LL;
362 
363 		if (ps->period < min_period) {
364 			pr_info_ratelimited(
365 			    "kvm: requested %lld ns "
366 			    "i8254 timer period limited to %lld ns\n",
367 			    ps->period, min_period);
368 			ps->period = min_period;
369 		}
370 	}
371 
372 	hrtimer_start(&ps->timer, ktime_add_ns(ktime_get(), interval),
373 		      HRTIMER_MODE_ABS);
374 }
375 
376 static void pit_load_count(struct kvm *kvm, int channel, u32 val)
377 {
378 	struct kvm_kpit_state *ps = &kvm->arch.vpit->pit_state;
379 
380 	WARN_ON(!mutex_is_locked(&ps->lock));
381 
382 	pr_debug("load_count val is %d, channel is %d\n", val, channel);
383 
384 	/*
385 	 * The largest possible initial count is 0; this is equivalent
386 	 * to 216 for binary counting and 104 for BCD counting.
387 	 */
388 	if (val == 0)
389 		val = 0x10000;
390 
391 	ps->channels[channel].count = val;
392 
393 	if (channel != 0) {
394 		ps->channels[channel].count_load_time = ktime_get();
395 		return;
396 	}
397 
398 	/* Two types of timer
399 	 * mode 1 is one shot, mode 2 is period, otherwise del timer */
400 	switch (ps->channels[0].mode) {
401 	case 0:
402 	case 1:
403         /* FIXME: enhance mode 4 precision */
404 	case 4:
405 		create_pit_timer(kvm, val, 0);
406 		break;
407 	case 2:
408 	case 3:
409 		create_pit_timer(kvm, val, 1);
410 		break;
411 	default:
412 		destroy_pit_timer(kvm->arch.vpit);
413 	}
414 }
415 
416 void kvm_pit_load_count(struct kvm *kvm, int channel, u32 val, int hpet_legacy_start)
417 {
418 	u8 saved_mode;
419 	if (hpet_legacy_start) {
420 		/* save existing mode for later reenablement */
421 		saved_mode = kvm->arch.vpit->pit_state.channels[0].mode;
422 		kvm->arch.vpit->pit_state.channels[0].mode = 0xff; /* disable timer */
423 		pit_load_count(kvm, channel, val);
424 		kvm->arch.vpit->pit_state.channels[0].mode = saved_mode;
425 	} else {
426 		pit_load_count(kvm, channel, val);
427 	}
428 }
429 
430 static inline struct kvm_pit *dev_to_pit(struct kvm_io_device *dev)
431 {
432 	return container_of(dev, struct kvm_pit, dev);
433 }
434 
435 static inline struct kvm_pit *speaker_to_pit(struct kvm_io_device *dev)
436 {
437 	return container_of(dev, struct kvm_pit, speaker_dev);
438 }
439 
440 static inline int pit_in_range(gpa_t addr)
441 {
442 	return ((addr >= KVM_PIT_BASE_ADDRESS) &&
443 		(addr < KVM_PIT_BASE_ADDRESS + KVM_PIT_MEM_LENGTH));
444 }
445 
446 static int pit_ioport_write(struct kvm_vcpu *vcpu,
447 				struct kvm_io_device *this,
448 			    gpa_t addr, int len, const void *data)
449 {
450 	struct kvm_pit *pit = dev_to_pit(this);
451 	struct kvm_kpit_state *pit_state = &pit->pit_state;
452 	struct kvm *kvm = pit->kvm;
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(kvm, channel);
477 					if (!(val & 0x10))
478 						pit_latch_status(kvm, 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(kvm, 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(kvm, addr, val);
504 			break;
505 		case RW_STATE_MSB:
506 			pit_load_count(kvm, 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(kvm, 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 	struct kvm *kvm = pit->kvm;
530 	int ret, count;
531 	struct kvm_kpit_channel_state *s;
532 	if (!pit_in_range(addr))
533 		return -EOPNOTSUPP;
534 
535 	addr &= KVM_PIT_CHANNEL_MASK;
536 	if (addr == 3)
537 		return 0;
538 
539 	s = &pit_state->channels[addr];
540 
541 	mutex_lock(&pit_state->lock);
542 
543 	if (s->status_latched) {
544 		s->status_latched = 0;
545 		ret = s->status;
546 	} else if (s->count_latched) {
547 		switch (s->count_latched) {
548 		default:
549 		case RW_STATE_LSB:
550 			ret = s->latched_count & 0xff;
551 			s->count_latched = 0;
552 			break;
553 		case RW_STATE_MSB:
554 			ret = s->latched_count >> 8;
555 			s->count_latched = 0;
556 			break;
557 		case RW_STATE_WORD0:
558 			ret = s->latched_count & 0xff;
559 			s->count_latched = RW_STATE_MSB;
560 			break;
561 		}
562 	} else {
563 		switch (s->read_state) {
564 		default:
565 		case RW_STATE_LSB:
566 			count = pit_get_count(kvm, addr);
567 			ret = count & 0xff;
568 			break;
569 		case RW_STATE_MSB:
570 			count = pit_get_count(kvm, addr);
571 			ret = (count >> 8) & 0xff;
572 			break;
573 		case RW_STATE_WORD0:
574 			count = pit_get_count(kvm, addr);
575 			ret = count & 0xff;
576 			s->read_state = RW_STATE_WORD1;
577 			break;
578 		case RW_STATE_WORD1:
579 			count = pit_get_count(kvm, addr);
580 			ret = (count >> 8) & 0xff;
581 			s->read_state = RW_STATE_WORD0;
582 			break;
583 		}
584 	}
585 
586 	if (len > sizeof(ret))
587 		len = sizeof(ret);
588 	memcpy(data, (char *)&ret, len);
589 
590 	mutex_unlock(&pit_state->lock);
591 	return 0;
592 }
593 
594 static int speaker_ioport_write(struct kvm_vcpu *vcpu,
595 				struct kvm_io_device *this,
596 				gpa_t addr, int len, const void *data)
597 {
598 	struct kvm_pit *pit = speaker_to_pit(this);
599 	struct kvm_kpit_state *pit_state = &pit->pit_state;
600 	struct kvm *kvm = pit->kvm;
601 	u32 val = *(u32 *) data;
602 	if (addr != KVM_SPEAKER_BASE_ADDRESS)
603 		return -EOPNOTSUPP;
604 
605 	mutex_lock(&pit_state->lock);
606 	pit_state->speaker_data_on = (val >> 1) & 1;
607 	pit_set_gate(kvm, 2, val & 1);
608 	mutex_unlock(&pit_state->lock);
609 	return 0;
610 }
611 
612 static int speaker_ioport_read(struct kvm_vcpu *vcpu,
613 				   struct kvm_io_device *this,
614 				   gpa_t addr, int len, void *data)
615 {
616 	struct kvm_pit *pit = speaker_to_pit(this);
617 	struct kvm_kpit_state *pit_state = &pit->pit_state;
618 	struct kvm *kvm = pit->kvm;
619 	unsigned int refresh_clock;
620 	int ret;
621 	if (addr != KVM_SPEAKER_BASE_ADDRESS)
622 		return -EOPNOTSUPP;
623 
624 	/* Refresh clock toggles at about 15us. We approximate as 2^14ns. */
625 	refresh_clock = ((unsigned int)ktime_to_ns(ktime_get()) >> 14) & 1;
626 
627 	mutex_lock(&pit_state->lock);
628 	ret = ((pit_state->speaker_data_on << 1) | pit_get_gate(kvm, 2) |
629 		(pit_get_out(kvm, 2) << 5) | (refresh_clock << 4));
630 	if (len > sizeof(ret))
631 		len = sizeof(ret);
632 	memcpy(data, (char *)&ret, len);
633 	mutex_unlock(&pit_state->lock);
634 	return 0;
635 }
636 
637 void kvm_pit_reset(struct kvm_pit *pit)
638 {
639 	int i;
640 	struct kvm_kpit_channel_state *c;
641 
642 	mutex_lock(&pit->pit_state.lock);
643 	pit->pit_state.flags = 0;
644 	for (i = 0; i < 3; i++) {
645 		c = &pit->pit_state.channels[i];
646 		c->mode = 0xff;
647 		c->gate = (i != 2);
648 		pit_load_count(pit->kvm, i, 0);
649 	}
650 	mutex_unlock(&pit->pit_state.lock);
651 
652 	atomic_set(&pit->pit_state.pending, 0);
653 	pit->pit_state.irq_ack = 1;
654 }
655 
656 static void pit_mask_notifer(struct kvm_irq_mask_notifier *kimn, bool mask)
657 {
658 	struct kvm_pit *pit = container_of(kimn, struct kvm_pit, mask_notifier);
659 
660 	if (!mask) {
661 		atomic_set(&pit->pit_state.pending, 0);
662 		pit->pit_state.irq_ack = 1;
663 	}
664 }
665 
666 static const struct kvm_io_device_ops pit_dev_ops = {
667 	.read     = pit_ioport_read,
668 	.write    = pit_ioport_write,
669 };
670 
671 static const struct kvm_io_device_ops speaker_dev_ops = {
672 	.read     = speaker_ioport_read,
673 	.write    = speaker_ioport_write,
674 };
675 
676 /* Caller must hold slots_lock */
677 struct kvm_pit *kvm_create_pit(struct kvm *kvm, u32 flags)
678 {
679 	struct kvm_pit *pit;
680 	struct kvm_kpit_state *pit_state;
681 	struct pid *pid;
682 	pid_t pid_nr;
683 	int ret;
684 
685 	pit = kzalloc(sizeof(struct kvm_pit), GFP_KERNEL);
686 	if (!pit)
687 		return NULL;
688 
689 	pit->irq_source_id = kvm_request_irq_source_id(kvm);
690 	if (pit->irq_source_id < 0) {
691 		kfree(pit);
692 		return NULL;
693 	}
694 
695 	mutex_init(&pit->pit_state.lock);
696 	mutex_lock(&pit->pit_state.lock);
697 	spin_lock_init(&pit->pit_state.inject_lock);
698 
699 	pid = get_pid(task_tgid(current));
700 	pid_nr = pid_vnr(pid);
701 	put_pid(pid);
702 
703 	init_kthread_worker(&pit->worker);
704 	pit->worker_task = kthread_run(kthread_worker_fn, &pit->worker,
705 				       "kvm-pit/%d", pid_nr);
706 	if (IS_ERR(pit->worker_task)) {
707 		mutex_unlock(&pit->pit_state.lock);
708 		kvm_free_irq_source_id(kvm, pit->irq_source_id);
709 		kfree(pit);
710 		return NULL;
711 	}
712 	init_kthread_work(&pit->expired, pit_do_work);
713 
714 	kvm->arch.vpit = pit;
715 	pit->kvm = kvm;
716 
717 	pit_state = &pit->pit_state;
718 	pit_state->pit = pit;
719 	hrtimer_init(&pit_state->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
720 	pit_state->irq_ack_notifier.gsi = 0;
721 	pit_state->irq_ack_notifier.irq_acked = kvm_pit_ack_irq;
722 	kvm_register_irq_ack_notifier(kvm, &pit_state->irq_ack_notifier);
723 	pit_state->reinject = true;
724 	mutex_unlock(&pit->pit_state.lock);
725 
726 	kvm_pit_reset(pit);
727 
728 	pit->mask_notifier.func = pit_mask_notifer;
729 	kvm_register_irq_mask_notifier(kvm, 0, &pit->mask_notifier);
730 
731 	kvm_iodevice_init(&pit->dev, &pit_dev_ops);
732 	ret = kvm_io_bus_register_dev(kvm, KVM_PIO_BUS, KVM_PIT_BASE_ADDRESS,
733 				      KVM_PIT_MEM_LENGTH, &pit->dev);
734 	if (ret < 0)
735 		goto fail;
736 
737 	if (flags & KVM_PIT_SPEAKER_DUMMY) {
738 		kvm_iodevice_init(&pit->speaker_dev, &speaker_dev_ops);
739 		ret = kvm_io_bus_register_dev(kvm, KVM_PIO_BUS,
740 					      KVM_SPEAKER_BASE_ADDRESS, 4,
741 					      &pit->speaker_dev);
742 		if (ret < 0)
743 			goto fail_unregister;
744 	}
745 
746 	return pit;
747 
748 fail_unregister:
749 	kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &pit->dev);
750 
751 fail:
752 	kvm_unregister_irq_mask_notifier(kvm, 0, &pit->mask_notifier);
753 	kvm_unregister_irq_ack_notifier(kvm, &pit_state->irq_ack_notifier);
754 	kvm_free_irq_source_id(kvm, pit->irq_source_id);
755 	kthread_stop(pit->worker_task);
756 	kfree(pit);
757 	return NULL;
758 }
759 
760 void kvm_free_pit(struct kvm *kvm)
761 {
762 	struct hrtimer *timer;
763 
764 	if (kvm->arch.vpit) {
765 		kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &kvm->arch.vpit->dev);
766 		kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
767 					      &kvm->arch.vpit->speaker_dev);
768 		kvm_unregister_irq_mask_notifier(kvm, 0,
769 					       &kvm->arch.vpit->mask_notifier);
770 		kvm_unregister_irq_ack_notifier(kvm,
771 				&kvm->arch.vpit->pit_state.irq_ack_notifier);
772 		mutex_lock(&kvm->arch.vpit->pit_state.lock);
773 		timer = &kvm->arch.vpit->pit_state.timer;
774 		hrtimer_cancel(timer);
775 		flush_kthread_work(&kvm->arch.vpit->expired);
776 		kthread_stop(kvm->arch.vpit->worker_task);
777 		kvm_free_irq_source_id(kvm, kvm->arch.vpit->irq_source_id);
778 		mutex_unlock(&kvm->arch.vpit->pit_state.lock);
779 		kfree(kvm->arch.vpit);
780 	}
781 }
782