xref: /linux/arch/x86/kvm/i8254.c (revision 0da908c291070d89482f6211dbe81d4d43c3f7cb)
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 static void pit_set_gate(struct kvm_pit *pit, int channel, u32 val)
55 {
56 	struct kvm_kpit_channel_state *c = &pit->pit_state.channels[channel];
57 
58 	switch (c->mode) {
59 	default:
60 	case 0:
61 	case 4:
62 		/* XXX: just disable/enable counting */
63 		break;
64 	case 1:
65 	case 2:
66 	case 3:
67 	case 5:
68 		/* Restart counting on rising edge. */
69 		if (c->gate < val)
70 			c->count_load_time = ktime_get();
71 		break;
72 	}
73 
74 	c->gate = val;
75 }
76 
77 static int pit_get_gate(struct kvm_pit *pit, int channel)
78 {
79 	return pit->pit_state.channels[channel].gate;
80 }
81 
82 static s64 __kpit_elapsed(struct kvm_pit *pit)
83 {
84 	s64 elapsed;
85 	ktime_t remaining;
86 	struct kvm_kpit_state *ps = &pit->pit_state;
87 
88 	if (!ps->period)
89 		return 0;
90 
91 	/*
92 	 * The Counter does not stop when it reaches zero. In
93 	 * Modes 0, 1, 4, and 5 the Counter ``wraps around'' to
94 	 * the highest count, either FFFF hex for binary counting
95 	 * or 9999 for BCD counting, and continues counting.
96 	 * Modes 2 and 3 are periodic; the Counter reloads
97 	 * itself with the initial count and continues counting
98 	 * from there.
99 	 */
100 	remaining = hrtimer_get_remaining(&ps->timer);
101 	elapsed = ps->period - ktime_to_ns(remaining);
102 
103 	return elapsed;
104 }
105 
106 static s64 kpit_elapsed(struct kvm_pit *pit, struct kvm_kpit_channel_state *c,
107 			int channel)
108 {
109 	if (channel == 0)
110 		return __kpit_elapsed(pit);
111 
112 	return ktime_to_ns(ktime_sub(ktime_get(), c->count_load_time));
113 }
114 
115 static int pit_get_count(struct kvm_pit *pit, int channel)
116 {
117 	struct kvm_kpit_channel_state *c = &pit->pit_state.channels[channel];
118 	s64 d, t;
119 	int counter;
120 
121 	t = kpit_elapsed(pit, c, channel);
122 	d = mul_u64_u32_div(t, KVM_PIT_FREQ, NSEC_PER_SEC);
123 
124 	switch (c->mode) {
125 	case 0:
126 	case 1:
127 	case 4:
128 	case 5:
129 		counter = (c->count - d) & 0xffff;
130 		break;
131 	case 3:
132 		/* XXX: may be incorrect for odd counts */
133 		counter = c->count - (mod_64((2 * d), c->count));
134 		break;
135 	default:
136 		counter = c->count - mod_64(d, c->count);
137 		break;
138 	}
139 	return counter;
140 }
141 
142 static int pit_get_out(struct kvm_pit *pit, int channel)
143 {
144 	struct kvm_kpit_channel_state *c = &pit->pit_state.channels[channel];
145 	s64 d, t;
146 	int out;
147 
148 	t = kpit_elapsed(pit, c, channel);
149 	d = mul_u64_u32_div(t, KVM_PIT_FREQ, NSEC_PER_SEC);
150 
151 	switch (c->mode) {
152 	default:
153 	case 0:
154 		out = (d >= c->count);
155 		break;
156 	case 1:
157 		out = (d < c->count);
158 		break;
159 	case 2:
160 		out = ((mod_64(d, c->count) == 0) && (d != 0));
161 		break;
162 	case 3:
163 		out = (mod_64(d, c->count) < ((c->count + 1) >> 1));
164 		break;
165 	case 4:
166 	case 5:
167 		out = (d == c->count);
168 		break;
169 	}
170 
171 	return out;
172 }
173 
174 static void pit_latch_count(struct kvm_pit *pit, int channel)
175 {
176 	struct kvm_kpit_channel_state *c = &pit->pit_state.channels[channel];
177 
178 	if (!c->count_latched) {
179 		c->latched_count = pit_get_count(pit, channel);
180 		c->count_latched = c->rw_mode;
181 	}
182 }
183 
184 static void pit_latch_status(struct kvm_pit *pit, int channel)
185 {
186 	struct kvm_kpit_channel_state *c = &pit->pit_state.channels[channel];
187 
188 	if (!c->status_latched) {
189 		/* TODO: Return NULL COUNT (bit 6). */
190 		c->status = ((pit_get_out(pit, channel) << 7) |
191 				(c->rw_mode << 4) |
192 				(c->mode << 1) |
193 				c->bcd);
194 		c->status_latched = 1;
195 	}
196 }
197 
198 static inline struct kvm_pit *pit_state_to_pit(struct kvm_kpit_state *ps)
199 {
200 	return container_of(ps, struct kvm_pit, pit_state);
201 }
202 
203 static void kvm_pit_ack_irq(struct kvm_irq_ack_notifier *kian)
204 {
205 	struct kvm_kpit_state *ps = container_of(kian, struct kvm_kpit_state,
206 						 irq_ack_notifier);
207 	struct kvm_pit *pit = pit_state_to_pit(ps);
208 
209 	atomic_set(&ps->irq_ack, 1);
210 	/* irq_ack should be set before pending is read.  Order accesses with
211 	 * inc(pending) in pit_timer_fn and xchg(irq_ack, 0) in pit_do_work.
212 	 */
213 	smp_mb();
214 	if (atomic_dec_if_positive(&ps->pending) > 0)
215 		kthread_queue_work(pit->worker, &pit->expired);
216 }
217 
218 void __kvm_migrate_pit_timer(struct kvm_vcpu *vcpu)
219 {
220 	struct kvm_pit *pit = vcpu->kvm->arch.vpit;
221 	struct hrtimer *timer;
222 
223 	/* Somewhat arbitrarily make vcpu0 the owner of the PIT. */
224 	if (vcpu->vcpu_id || !pit)
225 		return;
226 
227 	timer = &pit->pit_state.timer;
228 	mutex_lock(&pit->pit_state.lock);
229 	if (hrtimer_cancel(timer))
230 		hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
231 	mutex_unlock(&pit->pit_state.lock);
232 }
233 
234 static void destroy_pit_timer(struct kvm_pit *pit)
235 {
236 	hrtimer_cancel(&pit->pit_state.timer);
237 	kthread_flush_work(&pit->expired);
238 }
239 
240 static void pit_do_work(struct kthread_work *work)
241 {
242 	struct kvm_pit *pit = container_of(work, struct kvm_pit, expired);
243 	struct kvm *kvm = pit->kvm;
244 	struct kvm_vcpu *vcpu;
245 	unsigned long i;
246 	struct kvm_kpit_state *ps = &pit->pit_state;
247 
248 	if (atomic_read(&ps->reinject) && !atomic_xchg(&ps->irq_ack, 0))
249 		return;
250 
251 	kvm_set_irq(kvm, pit->irq_source_id, 0, 1, false);
252 	kvm_set_irq(kvm, pit->irq_source_id, 0, 0, false);
253 
254 	/*
255 	 * Provides NMI watchdog support via Virtual Wire mode.
256 	 * The route is: PIT -> LVT0 in NMI mode.
257 	 *
258 	 * Note: Our Virtual Wire implementation does not follow
259 	 * the MP specification.  We propagate a PIT interrupt to all
260 	 * VCPUs and only when LVT0 is in NMI mode.  The interrupt can
261 	 * also be simultaneously delivered through PIC and IOAPIC.
262 	 */
263 	if (atomic_read(&kvm->arch.vapics_in_nmi_mode) > 0)
264 		kvm_for_each_vcpu(i, vcpu, kvm)
265 			kvm_apic_nmi_wd_deliver(vcpu);
266 }
267 
268 static enum hrtimer_restart pit_timer_fn(struct hrtimer *data)
269 {
270 	struct kvm_kpit_state *ps = container_of(data, struct kvm_kpit_state, timer);
271 	struct kvm_pit *pt = pit_state_to_pit(ps);
272 
273 	if (atomic_read(&ps->reinject))
274 		atomic_inc(&ps->pending);
275 
276 	kthread_queue_work(pt->worker, &pt->expired);
277 
278 	if (ps->is_periodic) {
279 		hrtimer_add_expires_ns(&ps->timer, ps->period);
280 		return HRTIMER_RESTART;
281 	} else
282 		return HRTIMER_NORESTART;
283 }
284 
285 static inline void kvm_pit_reset_reinject(struct kvm_pit *pit)
286 {
287 	atomic_set(&pit->pit_state.pending, 0);
288 	atomic_set(&pit->pit_state.irq_ack, 1);
289 }
290 
291 void kvm_pit_set_reinject(struct kvm_pit *pit, bool reinject)
292 {
293 	struct kvm_kpit_state *ps = &pit->pit_state;
294 	struct kvm *kvm = pit->kvm;
295 
296 	if (atomic_read(&ps->reinject) == reinject)
297 		return;
298 
299 	/*
300 	 * AMD SVM AVIC accelerates EOI write and does not trap.
301 	 * This cause in-kernel PIT re-inject mode to fail
302 	 * since it checks ps->irq_ack before kvm_set_irq()
303 	 * and relies on the ack notifier to timely queue
304 	 * the pt->worker work iterm and reinject the missed tick.
305 	 * So, deactivate APICv when PIT is in reinject mode.
306 	 */
307 	if (reinject) {
308 		kvm_set_apicv_inhibit(kvm, APICV_INHIBIT_REASON_PIT_REINJ);
309 		/* The initial state is preserved while ps->reinject == 0. */
310 		kvm_pit_reset_reinject(pit);
311 		kvm_register_irq_ack_notifier(kvm, &ps->irq_ack_notifier);
312 		kvm_register_irq_mask_notifier(kvm, 0, &pit->mask_notifier);
313 	} else {
314 		kvm_clear_apicv_inhibit(kvm, APICV_INHIBIT_REASON_PIT_REINJ);
315 		kvm_unregister_irq_ack_notifier(kvm, &ps->irq_ack_notifier);
316 		kvm_unregister_irq_mask_notifier(kvm, 0, &pit->mask_notifier);
317 	}
318 
319 	atomic_set(&ps->reinject, reinject);
320 }
321 
322 static void create_pit_timer(struct kvm_pit *pit, u32 val, int is_period)
323 {
324 	struct kvm_kpit_state *ps = &pit->pit_state;
325 	struct kvm *kvm = pit->kvm;
326 	s64 interval;
327 
328 	if (!ioapic_in_kernel(kvm) ||
329 	    ps->flags & KVM_PIT_FLAGS_HPET_LEGACY)
330 		return;
331 
332 	interval = mul_u64_u32_div(val, NSEC_PER_SEC, KVM_PIT_FREQ);
333 
334 	pr_debug("create pit timer, interval is %llu nsec\n", interval);
335 
336 	/* TODO The new value only affected after the retriggered */
337 	hrtimer_cancel(&ps->timer);
338 	kthread_flush_work(&pit->expired);
339 	ps->period = interval;
340 	ps->is_periodic = is_period;
341 
342 	kvm_pit_reset_reinject(pit);
343 
344 	/*
345 	 * Do not allow the guest to program periodic timers with small
346 	 * interval, since the hrtimers are not throttled by the host
347 	 * scheduler.
348 	 */
349 	if (ps->is_periodic) {
350 		s64 min_period = min_timer_period_us * 1000LL;
351 
352 		if (ps->period < min_period) {
353 			pr_info_ratelimited(
354 			    "kvm: requested %lld ns "
355 			    "i8254 timer period limited to %lld ns\n",
356 			    ps->period, min_period);
357 			ps->period = min_period;
358 		}
359 	}
360 
361 	hrtimer_start(&ps->timer, ktime_add_ns(ktime_get(), interval),
362 		      HRTIMER_MODE_ABS);
363 }
364 
365 static void pit_load_count(struct kvm_pit *pit, int channel, u32 val)
366 {
367 	struct kvm_kpit_state *ps = &pit->pit_state;
368 
369 	pr_debug("load_count val is %u, channel is %d\n", val, channel);
370 
371 	/*
372 	 * The largest possible initial count is 0; this is equivalent
373 	 * to 216 for binary counting and 104 for BCD counting.
374 	 */
375 	if (val == 0)
376 		val = 0x10000;
377 
378 	ps->channels[channel].count = val;
379 
380 	if (channel != 0) {
381 		ps->channels[channel].count_load_time = ktime_get();
382 		return;
383 	}
384 
385 	/* Two types of timer
386 	 * mode 1 is one shot, mode 2 is period, otherwise del timer */
387 	switch (ps->channels[0].mode) {
388 	case 0:
389 	case 1:
390         /* FIXME: enhance mode 4 precision */
391 	case 4:
392 		create_pit_timer(pit, val, 0);
393 		break;
394 	case 2:
395 	case 3:
396 		create_pit_timer(pit, val, 1);
397 		break;
398 	default:
399 		destroy_pit_timer(pit);
400 	}
401 }
402 
403 void kvm_pit_load_count(struct kvm_pit *pit, int channel, u32 val,
404 		int hpet_legacy_start)
405 {
406 	u8 saved_mode;
407 
408 	WARN_ON_ONCE(!mutex_is_locked(&pit->pit_state.lock));
409 
410 	if (hpet_legacy_start) {
411 		/* save existing mode for later reenablement */
412 		WARN_ON(channel != 0);
413 		saved_mode = pit->pit_state.channels[0].mode;
414 		pit->pit_state.channels[0].mode = 0xff; /* disable timer */
415 		pit_load_count(pit, channel, val);
416 		pit->pit_state.channels[0].mode = saved_mode;
417 	} else {
418 		pit_load_count(pit, channel, val);
419 	}
420 }
421 
422 static inline struct kvm_pit *dev_to_pit(struct kvm_io_device *dev)
423 {
424 	return container_of(dev, struct kvm_pit, dev);
425 }
426 
427 static inline struct kvm_pit *speaker_to_pit(struct kvm_io_device *dev)
428 {
429 	return container_of(dev, struct kvm_pit, speaker_dev);
430 }
431 
432 static inline int pit_in_range(gpa_t addr)
433 {
434 	return ((addr >= KVM_PIT_BASE_ADDRESS) &&
435 		(addr < KVM_PIT_BASE_ADDRESS + KVM_PIT_MEM_LENGTH));
436 }
437 
438 static int pit_ioport_write(struct kvm_vcpu *vcpu,
439 				struct kvm_io_device *this,
440 			    gpa_t addr, int len, const void *data)
441 {
442 	struct kvm_pit *pit = dev_to_pit(this);
443 	struct kvm_kpit_state *pit_state = &pit->pit_state;
444 	int channel, access;
445 	struct kvm_kpit_channel_state *s;
446 	u32 val = *(u32 *) data;
447 	if (!pit_in_range(addr))
448 		return -EOPNOTSUPP;
449 
450 	val  &= 0xff;
451 	addr &= KVM_PIT_CHANNEL_MASK;
452 
453 	mutex_lock(&pit_state->lock);
454 
455 	if (val != 0)
456 		pr_debug("write addr is 0x%x, len is %d, val is 0x%x\n",
457 			 (unsigned int)addr, len, val);
458 
459 	if (addr == 3) {
460 		channel = val >> 6;
461 		if (channel == 3) {
462 			/* Read-Back Command. */
463 			for (channel = 0; channel < 3; channel++) {
464 				if (val & (2 << channel)) {
465 					if (!(val & 0x20))
466 						pit_latch_count(pit, channel);
467 					if (!(val & 0x10))
468 						pit_latch_status(pit, channel);
469 				}
470 			}
471 		} else {
472 			/* Select Counter <channel>. */
473 			s = &pit_state->channels[channel];
474 			access = (val >> 4) & KVM_PIT_CHANNEL_MASK;
475 			if (access == 0) {
476 				pit_latch_count(pit, channel);
477 			} else {
478 				s->rw_mode = access;
479 				s->read_state = access;
480 				s->write_state = access;
481 				s->mode = (val >> 1) & 7;
482 				if (s->mode > 5)
483 					s->mode -= 4;
484 				s->bcd = val & 1;
485 			}
486 		}
487 	} else {
488 		/* Write Count. */
489 		s = &pit_state->channels[addr];
490 		switch (s->write_state) {
491 		default:
492 		case RW_STATE_LSB:
493 			pit_load_count(pit, addr, val);
494 			break;
495 		case RW_STATE_MSB:
496 			pit_load_count(pit, addr, val << 8);
497 			break;
498 		case RW_STATE_WORD0:
499 			s->write_latch = val;
500 			s->write_state = RW_STATE_WORD1;
501 			break;
502 		case RW_STATE_WORD1:
503 			pit_load_count(pit, addr, s->write_latch | (val << 8));
504 			s->write_state = RW_STATE_WORD0;
505 			break;
506 		}
507 	}
508 
509 	mutex_unlock(&pit_state->lock);
510 	return 0;
511 }
512 
513 static int pit_ioport_read(struct kvm_vcpu *vcpu,
514 			   struct kvm_io_device *this,
515 			   gpa_t addr, int len, void *data)
516 {
517 	struct kvm_pit *pit = dev_to_pit(this);
518 	struct kvm_kpit_state *pit_state = &pit->pit_state;
519 	int ret, count;
520 	struct kvm_kpit_channel_state *s;
521 	if (!pit_in_range(addr))
522 		return -EOPNOTSUPP;
523 
524 	addr &= KVM_PIT_CHANNEL_MASK;
525 	if (addr == 3)
526 		return 0;
527 
528 	s = &pit_state->channels[addr];
529 
530 	mutex_lock(&pit_state->lock);
531 
532 	if (s->status_latched) {
533 		s->status_latched = 0;
534 		ret = s->status;
535 	} else if (s->count_latched) {
536 		switch (s->count_latched) {
537 		default:
538 		case RW_STATE_LSB:
539 			ret = s->latched_count & 0xff;
540 			s->count_latched = 0;
541 			break;
542 		case RW_STATE_MSB:
543 			ret = s->latched_count >> 8;
544 			s->count_latched = 0;
545 			break;
546 		case RW_STATE_WORD0:
547 			ret = s->latched_count & 0xff;
548 			s->count_latched = RW_STATE_MSB;
549 			break;
550 		}
551 	} else {
552 		switch (s->read_state) {
553 		default:
554 		case RW_STATE_LSB:
555 			count = pit_get_count(pit, addr);
556 			ret = count & 0xff;
557 			break;
558 		case RW_STATE_MSB:
559 			count = pit_get_count(pit, addr);
560 			ret = (count >> 8) & 0xff;
561 			break;
562 		case RW_STATE_WORD0:
563 			count = pit_get_count(pit, addr);
564 			ret = count & 0xff;
565 			s->read_state = RW_STATE_WORD1;
566 			break;
567 		case RW_STATE_WORD1:
568 			count = pit_get_count(pit, addr);
569 			ret = (count >> 8) & 0xff;
570 			s->read_state = RW_STATE_WORD0;
571 			break;
572 		}
573 	}
574 
575 	if (len > sizeof(ret))
576 		len = sizeof(ret);
577 	memcpy(data, (char *)&ret, len);
578 
579 	mutex_unlock(&pit_state->lock);
580 	return 0;
581 }
582 
583 static int speaker_ioport_write(struct kvm_vcpu *vcpu,
584 				struct kvm_io_device *this,
585 				gpa_t addr, int len, const void *data)
586 {
587 	struct kvm_pit *pit = speaker_to_pit(this);
588 	struct kvm_kpit_state *pit_state = &pit->pit_state;
589 	u32 val = *(u32 *) data;
590 	if (addr != KVM_SPEAKER_BASE_ADDRESS)
591 		return -EOPNOTSUPP;
592 
593 	mutex_lock(&pit_state->lock);
594 	if (val & (1 << 1))
595 		pit_state->flags |= KVM_PIT_FLAGS_SPEAKER_DATA_ON;
596 	else
597 		pit_state->flags &= ~KVM_PIT_FLAGS_SPEAKER_DATA_ON;
598 	pit_set_gate(pit, 2, val & 1);
599 	mutex_unlock(&pit_state->lock);
600 	return 0;
601 }
602 
603 static int speaker_ioport_read(struct kvm_vcpu *vcpu,
604 				   struct kvm_io_device *this,
605 				   gpa_t addr, int len, void *data)
606 {
607 	struct kvm_pit *pit = speaker_to_pit(this);
608 	struct kvm_kpit_state *pit_state = &pit->pit_state;
609 	unsigned int refresh_clock;
610 	int ret;
611 	if (addr != KVM_SPEAKER_BASE_ADDRESS)
612 		return -EOPNOTSUPP;
613 
614 	/* Refresh clock toggles at about 15us. We approximate as 2^14ns. */
615 	refresh_clock = ((unsigned int)ktime_to_ns(ktime_get()) >> 14) & 1;
616 
617 	mutex_lock(&pit_state->lock);
618 	ret = (!!(pit_state->flags & KVM_PIT_FLAGS_SPEAKER_DATA_ON) << 1) |
619 		pit_get_gate(pit, 2) | (pit_get_out(pit, 2) << 5) |
620 		(refresh_clock << 4);
621 	if (len > sizeof(ret))
622 		len = sizeof(ret);
623 	memcpy(data, (char *)&ret, len);
624 	mutex_unlock(&pit_state->lock);
625 	return 0;
626 }
627 
628 static void kvm_pit_reset(struct kvm_pit *pit)
629 {
630 	int i;
631 	struct kvm_kpit_channel_state *c;
632 
633 	pit->pit_state.flags = 0;
634 	for (i = 0; i < 3; i++) {
635 		c = &pit->pit_state.channels[i];
636 		c->mode = 0xff;
637 		c->gate = (i != 2);
638 		pit_load_count(pit, i, 0);
639 	}
640 
641 	kvm_pit_reset_reinject(pit);
642 }
643 
644 static void pit_mask_notifer(struct kvm_irq_mask_notifier *kimn, bool mask)
645 {
646 	struct kvm_pit *pit = container_of(kimn, struct kvm_pit, mask_notifier);
647 
648 	if (!mask)
649 		kvm_pit_reset_reinject(pit);
650 }
651 
652 static const struct kvm_io_device_ops pit_dev_ops = {
653 	.read     = pit_ioport_read,
654 	.write    = pit_ioport_write,
655 };
656 
657 static const struct kvm_io_device_ops speaker_dev_ops = {
658 	.read     = speaker_ioport_read,
659 	.write    = speaker_ioport_write,
660 };
661 
662 struct kvm_pit *kvm_create_pit(struct kvm *kvm, u32 flags)
663 {
664 	struct kvm_pit *pit;
665 	struct kvm_kpit_state *pit_state;
666 	struct pid *pid;
667 	pid_t pid_nr;
668 	int ret;
669 
670 	pit = kzalloc(sizeof(struct kvm_pit), GFP_KERNEL_ACCOUNT);
671 	if (!pit)
672 		return NULL;
673 
674 	pit->irq_source_id = kvm_request_irq_source_id(kvm);
675 	if (pit->irq_source_id < 0)
676 		goto fail_request;
677 
678 	mutex_init(&pit->pit_state.lock);
679 
680 	pid = get_pid(task_tgid(current));
681 	pid_nr = pid_vnr(pid);
682 	put_pid(pid);
683 
684 	pit->worker = kthread_create_worker(0, "kvm-pit/%d", pid_nr);
685 	if (IS_ERR(pit->worker))
686 		goto fail_kthread;
687 
688 	kthread_init_work(&pit->expired, pit_do_work);
689 
690 	pit->kvm = kvm;
691 
692 	pit_state = &pit->pit_state;
693 	hrtimer_init(&pit_state->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
694 	pit_state->timer.function = pit_timer_fn;
695 
696 	pit_state->irq_ack_notifier.gsi = 0;
697 	pit_state->irq_ack_notifier.irq_acked = kvm_pit_ack_irq;
698 	pit->mask_notifier.func = pit_mask_notifer;
699 
700 	kvm_pit_reset(pit);
701 
702 	kvm_pit_set_reinject(pit, true);
703 
704 	mutex_lock(&kvm->slots_lock);
705 	kvm_iodevice_init(&pit->dev, &pit_dev_ops);
706 	ret = kvm_io_bus_register_dev(kvm, KVM_PIO_BUS, KVM_PIT_BASE_ADDRESS,
707 				      KVM_PIT_MEM_LENGTH, &pit->dev);
708 	if (ret < 0)
709 		goto fail_register_pit;
710 
711 	if (flags & KVM_PIT_SPEAKER_DUMMY) {
712 		kvm_iodevice_init(&pit->speaker_dev, &speaker_dev_ops);
713 		ret = kvm_io_bus_register_dev(kvm, KVM_PIO_BUS,
714 					      KVM_SPEAKER_BASE_ADDRESS, 4,
715 					      &pit->speaker_dev);
716 		if (ret < 0)
717 			goto fail_register_speaker;
718 	}
719 	mutex_unlock(&kvm->slots_lock);
720 
721 	return pit;
722 
723 fail_register_speaker:
724 	kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &pit->dev);
725 fail_register_pit:
726 	mutex_unlock(&kvm->slots_lock);
727 	kvm_pit_set_reinject(pit, false);
728 	kthread_destroy_worker(pit->worker);
729 fail_kthread:
730 	kvm_free_irq_source_id(kvm, pit->irq_source_id);
731 fail_request:
732 	kfree(pit);
733 	return NULL;
734 }
735 
736 void kvm_free_pit(struct kvm *kvm)
737 {
738 	struct kvm_pit *pit = kvm->arch.vpit;
739 
740 	if (pit) {
741 		mutex_lock(&kvm->slots_lock);
742 		kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &pit->dev);
743 		kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &pit->speaker_dev);
744 		mutex_unlock(&kvm->slots_lock);
745 		kvm_pit_set_reinject(pit, false);
746 		hrtimer_cancel(&pit->pit_state.timer);
747 		kthread_destroy_worker(pit->worker);
748 		kvm_free_irq_source_id(kvm, pit->irq_source_id);
749 		kfree(pit);
750 	}
751 }
752