2 * 8253/8254 interval timer emulation
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.
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:
17 * The above copyright notice and this permission notice shall be included in
18 * all copies or substantial portions of the Software.
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
29 * Sheng Yang <sheng.yang@intel.com>
30 * Based on QEMU and Xen.
33 #define pr_fmt(fmt) "pit: " fmt
35 #include <linux/kvm_host.h>
36 #include <linux/slab.h>
44 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
46 #define mod_64(x, y) ((x) % (y))
49 #define RW_STATE_LSB 1
50 #define RW_STATE_MSB 2
51 #define RW_STATE_WORD0 3
52 #define RW_STATE_WORD1 4
54 /* Compute with 96 bit intermediate result: (a*b)/c */
55 static u64 muldiv64(u64 a, u32 b, u32 c)
66 rl = (u64)u.l.low * (u64)b;
67 rh = (u64)u.l.high * (u64)b;
69 res.l.high = div64_u64(rh, c);
70 res.l.low = div64_u64(((mod_64(rh, c) << 32) + (rl & 0xffffffff)), c);
74 static void pit_set_gate(struct kvm_pit *pit, int channel, u32 val)
76 struct kvm_kpit_channel_state *c = &pit->pit_state.channels[channel];
82 /* XXX: just disable/enable counting */
88 /* Restart counting on rising edge. */
90 c->count_load_time = ktime_get();
97 static int pit_get_gate(struct kvm_pit *pit, int channel)
99 return pit->pit_state.channels[channel].gate;
102 static s64 __kpit_elapsed(struct kvm_pit *pit)
106 struct kvm_kpit_state *ps = &pit->pit_state;
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
120 remaining = hrtimer_get_remaining(&ps->timer);
121 elapsed = ps->period - ktime_to_ns(remaining);
126 static s64 kpit_elapsed(struct kvm_pit *pit, struct kvm_kpit_channel_state *c,
130 return __kpit_elapsed(pit);
132 return ktime_to_ns(ktime_sub(ktime_get(), c->count_load_time));
135 static int pit_get_count(struct kvm_pit *pit, int channel)
137 struct kvm_kpit_channel_state *c = &pit->pit_state.channels[channel];
141 t = kpit_elapsed(pit, c, channel);
142 d = muldiv64(t, KVM_PIT_FREQ, NSEC_PER_SEC);
149 counter = (c->count - d) & 0xffff;
152 /* XXX: may be incorrect for odd counts */
153 counter = c->count - (mod_64((2 * d), c->count));
156 counter = c->count - mod_64(d, c->count);
162 static int pit_get_out(struct kvm_pit *pit, int channel)
164 struct kvm_kpit_channel_state *c = &pit->pit_state.channels[channel];
168 t = kpit_elapsed(pit, c, channel);
169 d = muldiv64(t, KVM_PIT_FREQ, NSEC_PER_SEC);
174 out = (d >= c->count);
177 out = (d < c->count);
180 out = ((mod_64(d, c->count) == 0) && (d != 0));
183 out = (mod_64(d, c->count) < ((c->count + 1) >> 1));
187 out = (d == c->count);
194 static void pit_latch_count(struct kvm_pit *pit, int channel)
196 struct kvm_kpit_channel_state *c = &pit->pit_state.channels[channel];
198 if (!c->count_latched) {
199 c->latched_count = pit_get_count(pit, channel);
200 c->count_latched = c->rw_mode;
204 static void pit_latch_status(struct kvm_pit *pit, int channel)
206 struct kvm_kpit_channel_state *c = &pit->pit_state.channels[channel];
208 if (!c->status_latched) {
209 /* TODO: Return NULL COUNT (bit 6). */
210 c->status = ((pit_get_out(pit, channel) << 7) |
214 c->status_latched = 1;
218 static inline struct kvm_pit *pit_state_to_pit(struct kvm_kpit_state *ps)
220 return container_of(ps, struct kvm_pit, pit_state);
223 static void kvm_pit_ack_irq(struct kvm_irq_ack_notifier *kian)
225 struct kvm_kpit_state *ps = container_of(kian, struct kvm_kpit_state,
227 struct kvm_pit *pit = pit_state_to_pit(ps);
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.
234 if (atomic_dec_if_positive(&ps->pending) > 0)
235 queue_kthread_work(&pit->worker, &pit->expired);
238 void __kvm_migrate_pit_timer(struct kvm_vcpu *vcpu)
240 struct kvm_pit *pit = vcpu->kvm->arch.vpit;
241 struct hrtimer *timer;
243 if (!kvm_vcpu_is_bsp(vcpu) || !pit)
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);
253 static void destroy_pit_timer(struct kvm_pit *pit)
255 hrtimer_cancel(&pit->pit_state.timer);
256 flush_kthread_work(&pit->expired);
259 static void pit_do_work(struct kthread_work *work)
261 struct kvm_pit *pit = container_of(work, struct kvm_pit, expired);
262 struct kvm *kvm = pit->kvm;
263 struct kvm_vcpu *vcpu;
265 struct kvm_kpit_state *ps = &pit->pit_state;
267 if (ps->reinject && !atomic_xchg(&ps->irq_ack, 0))
270 kvm_set_irq(kvm, kvm->arch.vpit->irq_source_id, 0, 1, false);
271 kvm_set_irq(kvm, kvm->arch.vpit->irq_source_id, 0, 0, false);
274 * Provides NMI watchdog support via Virtual Wire mode.
275 * The route is: PIT -> LVT0 in NMI mode.
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.
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);
287 static enum hrtimer_restart pit_timer_fn(struct hrtimer *data)
289 struct kvm_kpit_state *ps = container_of(data, struct kvm_kpit_state, timer);
290 struct kvm_pit *pt = pit_state_to_pit(ps);
293 atomic_inc(&ps->pending);
295 queue_kthread_work(&pt->worker, &pt->expired);
297 if (ps->is_periodic) {
298 hrtimer_add_expires_ns(&ps->timer, ps->period);
299 return HRTIMER_RESTART;
301 return HRTIMER_NORESTART;
304 static inline void kvm_pit_reset_reinject(struct kvm_pit *pit)
306 atomic_set(&pit->pit_state.pending, 0);
307 atomic_set(&pit->pit_state.irq_ack, 1);
310 void kvm_pit_set_reinject(struct kvm_pit *pit, bool reinject)
312 struct kvm_kpit_state *ps = &pit->pit_state;
313 struct kvm *kvm = pit->kvm;
315 if (ps->reinject == 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);
324 kvm_unregister_irq_ack_notifier(kvm, &ps->irq_ack_notifier);
325 kvm_unregister_irq_mask_notifier(kvm, 0, &pit->mask_notifier);
328 ps->reinject = reinject;
331 static void create_pit_timer(struct kvm_pit *pit, u32 val, int is_period)
333 struct kvm_kpit_state *ps = &pit->pit_state;
334 struct kvm *kvm = pit->kvm;
337 if (!ioapic_in_kernel(kvm) ||
338 ps->flags & KVM_PIT_FLAGS_HPET_LEGACY)
341 interval = muldiv64(val, NSEC_PER_SEC, KVM_PIT_FREQ);
343 pr_debug("create pit timer, interval is %llu nsec\n", interval);
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;
351 ps->timer.function = pit_timer_fn;
353 kvm_pit_reset_reinject(pit);
356 * Do not allow the guest to program periodic timers with small
357 * interval, since the hrtimers are not throttled by the host
360 if (ps->is_periodic) {
361 s64 min_period = min_timer_period_us * 1000LL;
363 if (ps->period < min_period) {
365 "kvm: requested %lld ns "
366 "i8254 timer period limited to %lld ns\n",
367 ps->period, min_period);
368 ps->period = min_period;
372 hrtimer_start(&ps->timer, ktime_add_ns(ktime_get(), interval),
376 static void pit_load_count(struct kvm_pit *pit, int channel, u32 val)
378 struct kvm_kpit_state *ps = &pit->pit_state;
380 pr_debug("load_count val is %d, channel is %d\n", val, channel);
383 * The largest possible initial count is 0; this is equivalent
384 * to 216 for binary counting and 104 for BCD counting.
389 ps->channels[channel].count = val;
392 ps->channels[channel].count_load_time = ktime_get();
396 /* Two types of timer
397 * mode 1 is one shot, mode 2 is period, otherwise del timer */
398 switch (ps->channels[0].mode) {
401 /* FIXME: enhance mode 4 precision */
403 create_pit_timer(pit, val, 0);
407 create_pit_timer(pit, val, 1);
410 destroy_pit_timer(pit);
414 void kvm_pit_load_count(struct kvm_pit *pit, int channel, u32 val,
415 int hpet_legacy_start)
419 WARN_ON_ONCE(!mutex_is_locked(&pit->pit_state.lock));
421 if (hpet_legacy_start) {
422 /* save existing mode for later reenablement */
423 WARN_ON(channel != 0);
424 saved_mode = pit->pit_state.channels[0].mode;
425 pit->pit_state.channels[0].mode = 0xff; /* disable timer */
426 pit_load_count(pit, channel, val);
427 pit->pit_state.channels[0].mode = saved_mode;
429 pit_load_count(pit, channel, val);
433 static inline struct kvm_pit *dev_to_pit(struct kvm_io_device *dev)
435 return container_of(dev, struct kvm_pit, dev);
438 static inline struct kvm_pit *speaker_to_pit(struct kvm_io_device *dev)
440 return container_of(dev, struct kvm_pit, speaker_dev);
443 static inline int pit_in_range(gpa_t addr)
445 return ((addr >= KVM_PIT_BASE_ADDRESS) &&
446 (addr < KVM_PIT_BASE_ADDRESS + KVM_PIT_MEM_LENGTH));
449 static int pit_ioport_write(struct kvm_vcpu *vcpu,
450 struct kvm_io_device *this,
451 gpa_t addr, int len, const void *data)
453 struct kvm_pit *pit = dev_to_pit(this);
454 struct kvm_kpit_state *pit_state = &pit->pit_state;
456 struct kvm_kpit_channel_state *s;
457 u32 val = *(u32 *) data;
458 if (!pit_in_range(addr))
462 addr &= KVM_PIT_CHANNEL_MASK;
464 mutex_lock(&pit_state->lock);
467 pr_debug("write addr is 0x%x, len is %d, val is 0x%x\n",
468 (unsigned int)addr, len, val);
473 /* Read-Back Command. */
474 for (channel = 0; channel < 3; channel++) {
475 s = &pit_state->channels[channel];
476 if (val & (2 << channel)) {
478 pit_latch_count(pit, channel);
480 pit_latch_status(pit, channel);
484 /* Select Counter <channel>. */
485 s = &pit_state->channels[channel];
486 access = (val >> 4) & KVM_PIT_CHANNEL_MASK;
488 pit_latch_count(pit, channel);
491 s->read_state = access;
492 s->write_state = access;
493 s->mode = (val >> 1) & 7;
501 s = &pit_state->channels[addr];
502 switch (s->write_state) {
505 pit_load_count(pit, addr, val);
508 pit_load_count(pit, addr, val << 8);
511 s->write_latch = val;
512 s->write_state = RW_STATE_WORD1;
515 pit_load_count(pit, addr, s->write_latch | (val << 8));
516 s->write_state = RW_STATE_WORD0;
521 mutex_unlock(&pit_state->lock);
525 static int pit_ioport_read(struct kvm_vcpu *vcpu,
526 struct kvm_io_device *this,
527 gpa_t addr, int len, void *data)
529 struct kvm_pit *pit = dev_to_pit(this);
530 struct kvm_kpit_state *pit_state = &pit->pit_state;
532 struct kvm_kpit_channel_state *s;
533 if (!pit_in_range(addr))
536 addr &= KVM_PIT_CHANNEL_MASK;
540 s = &pit_state->channels[addr];
542 mutex_lock(&pit_state->lock);
544 if (s->status_latched) {
545 s->status_latched = 0;
547 } else if (s->count_latched) {
548 switch (s->count_latched) {
551 ret = s->latched_count & 0xff;
552 s->count_latched = 0;
555 ret = s->latched_count >> 8;
556 s->count_latched = 0;
559 ret = s->latched_count & 0xff;
560 s->count_latched = RW_STATE_MSB;
564 switch (s->read_state) {
567 count = pit_get_count(pit, addr);
571 count = pit_get_count(pit, addr);
572 ret = (count >> 8) & 0xff;
575 count = pit_get_count(pit, addr);
577 s->read_state = RW_STATE_WORD1;
580 count = pit_get_count(pit, addr);
581 ret = (count >> 8) & 0xff;
582 s->read_state = RW_STATE_WORD0;
587 if (len > sizeof(ret))
589 memcpy(data, (char *)&ret, len);
591 mutex_unlock(&pit_state->lock);
595 static int speaker_ioport_write(struct kvm_vcpu *vcpu,
596 struct kvm_io_device *this,
597 gpa_t addr, int len, const void *data)
599 struct kvm_pit *pit = speaker_to_pit(this);
600 struct kvm_kpit_state *pit_state = &pit->pit_state;
601 u32 val = *(u32 *) data;
602 if (addr != KVM_SPEAKER_BASE_ADDRESS)
605 mutex_lock(&pit_state->lock);
606 pit_state->speaker_data_on = (val >> 1) & 1;
607 pit_set_gate(pit, 2, val & 1);
608 mutex_unlock(&pit_state->lock);
612 static int speaker_ioport_read(struct kvm_vcpu *vcpu,
613 struct kvm_io_device *this,
614 gpa_t addr, int len, void *data)
616 struct kvm_pit *pit = speaker_to_pit(this);
617 struct kvm_kpit_state *pit_state = &pit->pit_state;
618 unsigned int refresh_clock;
620 if (addr != KVM_SPEAKER_BASE_ADDRESS)
623 /* Refresh clock toggles at about 15us. We approximate as 2^14ns. */
624 refresh_clock = ((unsigned int)ktime_to_ns(ktime_get()) >> 14) & 1;
626 mutex_lock(&pit_state->lock);
627 ret = ((pit_state->speaker_data_on << 1) | pit_get_gate(pit, 2) |
628 (pit_get_out(pit, 2) << 5) | (refresh_clock << 4));
629 if (len > sizeof(ret))
631 memcpy(data, (char *)&ret, len);
632 mutex_unlock(&pit_state->lock);
636 static void kvm_pit_reset(struct kvm_pit *pit)
639 struct kvm_kpit_channel_state *c;
641 pit->pit_state.flags = 0;
642 for (i = 0; i < 3; i++) {
643 c = &pit->pit_state.channels[i];
646 pit_load_count(pit, i, 0);
649 kvm_pit_reset_reinject(pit);
652 static void pit_mask_notifer(struct kvm_irq_mask_notifier *kimn, bool mask)
654 struct kvm_pit *pit = container_of(kimn, struct kvm_pit, mask_notifier);
657 kvm_pit_reset_reinject(pit);
660 static const struct kvm_io_device_ops pit_dev_ops = {
661 .read = pit_ioport_read,
662 .write = pit_ioport_write,
665 static const struct kvm_io_device_ops speaker_dev_ops = {
666 .read = speaker_ioport_read,
667 .write = speaker_ioport_write,
670 /* Caller must hold slots_lock */
671 struct kvm_pit *kvm_create_pit(struct kvm *kvm, u32 flags)
674 struct kvm_kpit_state *pit_state;
679 pit = kzalloc(sizeof(struct kvm_pit), GFP_KERNEL);
683 pit->irq_source_id = kvm_request_irq_source_id(kvm);
684 if (pit->irq_source_id < 0)
687 mutex_init(&pit->pit_state.lock);
689 pid = get_pid(task_tgid(current));
690 pid_nr = pid_vnr(pid);
693 init_kthread_worker(&pit->worker);
694 pit->worker_task = kthread_run(kthread_worker_fn, &pit->worker,
695 "kvm-pit/%d", pid_nr);
696 if (IS_ERR(pit->worker_task))
699 init_kthread_work(&pit->expired, pit_do_work);
703 pit_state = &pit->pit_state;
704 hrtimer_init(&pit_state->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
706 pit_state->irq_ack_notifier.gsi = 0;
707 pit_state->irq_ack_notifier.irq_acked = kvm_pit_ack_irq;
708 pit->mask_notifier.func = pit_mask_notifer;
712 kvm_pit_set_reinject(pit, true);
714 kvm_iodevice_init(&pit->dev, &pit_dev_ops);
715 ret = kvm_io_bus_register_dev(kvm, KVM_PIO_BUS, KVM_PIT_BASE_ADDRESS,
716 KVM_PIT_MEM_LENGTH, &pit->dev);
718 goto fail_register_pit;
720 if (flags & KVM_PIT_SPEAKER_DUMMY) {
721 kvm_iodevice_init(&pit->speaker_dev, &speaker_dev_ops);
722 ret = kvm_io_bus_register_dev(kvm, KVM_PIO_BUS,
723 KVM_SPEAKER_BASE_ADDRESS, 4,
726 goto fail_register_speaker;
731 fail_register_speaker:
732 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &pit->dev);
734 kvm_pit_set_reinject(pit, false);
735 kthread_stop(pit->worker_task);
737 kvm_free_irq_source_id(kvm, pit->irq_source_id);
743 void kvm_free_pit(struct kvm *kvm)
745 struct kvm_pit *pit = kvm->arch.vpit;
748 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &pit->dev);
749 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &pit->speaker_dev);
750 kvm_pit_set_reinject(pit, false);
751 hrtimer_cancel(&pit->pit_state.timer);
752 flush_kthread_work(&pit->expired);
753 kthread_stop(pit->worker_task);
754 kvm_free_irq_source_id(kvm, pit->irq_source_id);