perf_counter: PERF_SAMPLE_ID and inherited counters
[linux-2.6/verdex.git] / arch / x86 / kvm / i8254.c
blob4d6f0d293ee29bb303dbe5e4b81b4cb3c9e2b6e2
1 /*
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
9 * Permission is hereby granted, free of charge, to any person obtaining a copy
10 * of this software and associated documentation files (the "Software"), to deal
11 * in the Software without restriction, including without limitation the rights
12 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
13 * copies of the Software, and to permit persons to whom the Software is
14 * furnished to do so, subject to the following conditions:
16 * The above copyright notice and this permission notice shall be included in
17 * all copies or substantial portions of the Software.
19 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
20 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
21 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
22 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
23 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
24 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
25 * THE SOFTWARE.
27 * Authors:
28 * Sheng Yang <sheng.yang@intel.com>
29 * Based on QEMU and Xen.
32 #include <linux/kvm_host.h>
34 #include "irq.h"
35 #include "i8254.h"
37 #ifndef CONFIG_X86_64
38 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
39 #else
40 #define mod_64(x, y) ((x) % (y))
41 #endif
43 #define RW_STATE_LSB 1
44 #define RW_STATE_MSB 2
45 #define RW_STATE_WORD0 3
46 #define RW_STATE_WORD1 4
48 /* Compute with 96 bit intermediate result: (a*b)/c */
49 static u64 muldiv64(u64 a, u32 b, u32 c)
51 union {
52 u64 ll;
53 struct {
54 u32 low, high;
55 } l;
56 } u, res;
57 u64 rl, rh;
59 u.ll = a;
60 rl = (u64)u.l.low * (u64)b;
61 rh = (u64)u.l.high * (u64)b;
62 rh += (rl >> 32);
63 res.l.high = div64_u64(rh, c);
64 res.l.low = div64_u64(((mod_64(rh, c) << 32) + (rl & 0xffffffff)), c);
65 return res.ll;
68 static void pit_set_gate(struct kvm *kvm, int channel, u32 val)
70 struct kvm_kpit_channel_state *c =
71 &kvm->arch.vpit->pit_state.channels[channel];
73 WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
75 switch (c->mode) {
76 default:
77 case 0:
78 case 4:
79 /* XXX: just disable/enable counting */
80 break;
81 case 1:
82 case 2:
83 case 3:
84 case 5:
85 /* Restart counting on rising edge. */
86 if (c->gate < val)
87 c->count_load_time = ktime_get();
88 break;
91 c->gate = val;
94 static int pit_get_gate(struct kvm *kvm, int channel)
96 WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
98 return kvm->arch.vpit->pit_state.channels[channel].gate;
101 static s64 __kpit_elapsed(struct kvm *kvm)
103 s64 elapsed;
104 ktime_t remaining;
105 struct kvm_kpit_state *ps = &kvm->arch.vpit->pit_state;
108 * The Counter does not stop when it reaches zero. In
109 * Modes 0, 1, 4, and 5 the Counter ``wraps around'' to
110 * the highest count, either FFFF hex for binary counting
111 * or 9999 for BCD counting, and continues counting.
112 * Modes 2 and 3 are periodic; the Counter reloads
113 * itself with the initial count and continues counting
114 * from there.
116 remaining = hrtimer_expires_remaining(&ps->pit_timer.timer);
117 elapsed = ps->pit_timer.period - ktime_to_ns(remaining);
118 elapsed = mod_64(elapsed, ps->pit_timer.period);
120 return elapsed;
123 static s64 kpit_elapsed(struct kvm *kvm, struct kvm_kpit_channel_state *c,
124 int channel)
126 if (channel == 0)
127 return __kpit_elapsed(kvm);
129 return ktime_to_ns(ktime_sub(ktime_get(), c->count_load_time));
132 static int pit_get_count(struct kvm *kvm, int channel)
134 struct kvm_kpit_channel_state *c =
135 &kvm->arch.vpit->pit_state.channels[channel];
136 s64 d, t;
137 int counter;
139 WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
141 t = kpit_elapsed(kvm, c, channel);
142 d = muldiv64(t, KVM_PIT_FREQ, NSEC_PER_SEC);
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;
159 return counter;
162 static int pit_get_out(struct kvm *kvm, int channel)
164 struct kvm_kpit_channel_state *c =
165 &kvm->arch.vpit->pit_state.channels[channel];
166 s64 d, t;
167 int out;
169 WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
171 t = kpit_elapsed(kvm, c, channel);
172 d = muldiv64(t, KVM_PIT_FREQ, NSEC_PER_SEC);
174 switch (c->mode) {
175 default:
176 case 0:
177 out = (d >= c->count);
178 break;
179 case 1:
180 out = (d < c->count);
181 break;
182 case 2:
183 out = ((mod_64(d, c->count) == 0) && (d != 0));
184 break;
185 case 3:
186 out = (mod_64(d, c->count) < ((c->count + 1) >> 1));
187 break;
188 case 4:
189 case 5:
190 out = (d == c->count);
191 break;
194 return out;
197 static void pit_latch_count(struct kvm *kvm, int channel)
199 struct kvm_kpit_channel_state *c =
200 &kvm->arch.vpit->pit_state.channels[channel];
202 WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
204 if (!c->count_latched) {
205 c->latched_count = pit_get_count(kvm, channel);
206 c->count_latched = c->rw_mode;
210 static void pit_latch_status(struct kvm *kvm, int channel)
212 struct kvm_kpit_channel_state *c =
213 &kvm->arch.vpit->pit_state.channels[channel];
215 WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
217 if (!c->status_latched) {
218 /* TODO: Return NULL COUNT (bit 6). */
219 c->status = ((pit_get_out(kvm, channel) << 7) |
220 (c->rw_mode << 4) |
221 (c->mode << 1) |
222 c->bcd);
223 c->status_latched = 1;
227 int pit_has_pending_timer(struct kvm_vcpu *vcpu)
229 struct kvm_pit *pit = vcpu->kvm->arch.vpit;
231 if (pit && vcpu->vcpu_id == 0 && pit->pit_state.irq_ack)
232 return atomic_read(&pit->pit_state.pit_timer.pending);
233 return 0;
236 static void kvm_pit_ack_irq(struct kvm_irq_ack_notifier *kian)
238 struct kvm_kpit_state *ps = container_of(kian, struct kvm_kpit_state,
239 irq_ack_notifier);
240 spin_lock(&ps->inject_lock);
241 if (atomic_dec_return(&ps->pit_timer.pending) < 0)
242 atomic_inc(&ps->pit_timer.pending);
243 ps->irq_ack = 1;
244 spin_unlock(&ps->inject_lock);
247 void __kvm_migrate_pit_timer(struct kvm_vcpu *vcpu)
249 struct kvm_pit *pit = vcpu->kvm->arch.vpit;
250 struct hrtimer *timer;
252 if (vcpu->vcpu_id != 0 || !pit)
253 return;
255 timer = &pit->pit_state.pit_timer.timer;
256 if (hrtimer_cancel(timer))
257 hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
260 static void destroy_pit_timer(struct kvm_timer *pt)
262 pr_debug("pit: execute del timer!\n");
263 hrtimer_cancel(&pt->timer);
266 static bool kpit_is_periodic(struct kvm_timer *ktimer)
268 struct kvm_kpit_state *ps = container_of(ktimer, struct kvm_kpit_state,
269 pit_timer);
270 return ps->is_periodic;
273 static struct kvm_timer_ops kpit_ops = {
274 .is_periodic = kpit_is_periodic,
277 static void create_pit_timer(struct kvm_kpit_state *ps, u32 val, int is_period)
279 struct kvm_timer *pt = &ps->pit_timer;
280 s64 interval;
282 interval = muldiv64(val, NSEC_PER_SEC, KVM_PIT_FREQ);
284 pr_debug("pit: create pit timer, interval is %llu nsec\n", interval);
286 /* TODO The new value only affected after the retriggered */
287 hrtimer_cancel(&pt->timer);
288 pt->period = interval;
289 ps->is_periodic = is_period;
291 pt->timer.function = kvm_timer_fn;
292 pt->t_ops = &kpit_ops;
293 pt->kvm = ps->pit->kvm;
294 pt->vcpu_id = 0;
296 atomic_set(&pt->pending, 0);
297 ps->irq_ack = 1;
299 hrtimer_start(&pt->timer, ktime_add_ns(ktime_get(), interval),
300 HRTIMER_MODE_ABS);
303 static void pit_load_count(struct kvm *kvm, int channel, u32 val)
305 struct kvm_kpit_state *ps = &kvm->arch.vpit->pit_state;
307 WARN_ON(!mutex_is_locked(&ps->lock));
309 pr_debug("pit: load_count val is %d, channel is %d\n", val, channel);
312 * The largest possible initial count is 0; this is equivalent
313 * to 216 for binary counting and 104 for BCD counting.
315 if (val == 0)
316 val = 0x10000;
318 ps->channels[channel].count = val;
320 if (channel != 0) {
321 ps->channels[channel].count_load_time = ktime_get();
322 return;
325 /* Two types of timer
326 * mode 1 is one shot, mode 2 is period, otherwise del timer */
327 switch (ps->channels[0].mode) {
328 case 0:
329 case 1:
330 /* FIXME: enhance mode 4 precision */
331 case 4:
332 create_pit_timer(ps, val, 0);
333 break;
334 case 2:
335 case 3:
336 create_pit_timer(ps, val, 1);
337 break;
338 default:
339 destroy_pit_timer(&ps->pit_timer);
343 void kvm_pit_load_count(struct kvm *kvm, int channel, u32 val)
345 mutex_lock(&kvm->arch.vpit->pit_state.lock);
346 pit_load_count(kvm, channel, val);
347 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
350 static void pit_ioport_write(struct kvm_io_device *this,
351 gpa_t addr, int len, const void *data)
353 struct kvm_pit *pit = (struct kvm_pit *)this->private;
354 struct kvm_kpit_state *pit_state = &pit->pit_state;
355 struct kvm *kvm = pit->kvm;
356 int channel, access;
357 struct kvm_kpit_channel_state *s;
358 u32 val = *(u32 *) data;
360 val &= 0xff;
361 addr &= KVM_PIT_CHANNEL_MASK;
363 mutex_lock(&pit_state->lock);
365 if (val != 0)
366 pr_debug("pit: write addr is 0x%x, len is %d, val is 0x%x\n",
367 (unsigned int)addr, len, val);
369 if (addr == 3) {
370 channel = val >> 6;
371 if (channel == 3) {
372 /* Read-Back Command. */
373 for (channel = 0; channel < 3; channel++) {
374 s = &pit_state->channels[channel];
375 if (val & (2 << channel)) {
376 if (!(val & 0x20))
377 pit_latch_count(kvm, channel);
378 if (!(val & 0x10))
379 pit_latch_status(kvm, channel);
382 } else {
383 /* Select Counter <channel>. */
384 s = &pit_state->channels[channel];
385 access = (val >> 4) & KVM_PIT_CHANNEL_MASK;
386 if (access == 0) {
387 pit_latch_count(kvm, channel);
388 } else {
389 s->rw_mode = access;
390 s->read_state = access;
391 s->write_state = access;
392 s->mode = (val >> 1) & 7;
393 if (s->mode > 5)
394 s->mode -= 4;
395 s->bcd = val & 1;
398 } else {
399 /* Write Count. */
400 s = &pit_state->channels[addr];
401 switch (s->write_state) {
402 default:
403 case RW_STATE_LSB:
404 pit_load_count(kvm, addr, val);
405 break;
406 case RW_STATE_MSB:
407 pit_load_count(kvm, addr, val << 8);
408 break;
409 case RW_STATE_WORD0:
410 s->write_latch = val;
411 s->write_state = RW_STATE_WORD1;
412 break;
413 case RW_STATE_WORD1:
414 pit_load_count(kvm, addr, s->write_latch | (val << 8));
415 s->write_state = RW_STATE_WORD0;
416 break;
420 mutex_unlock(&pit_state->lock);
423 static void pit_ioport_read(struct kvm_io_device *this,
424 gpa_t addr, int len, void *data)
426 struct kvm_pit *pit = (struct kvm_pit *)this->private;
427 struct kvm_kpit_state *pit_state = &pit->pit_state;
428 struct kvm *kvm = pit->kvm;
429 int ret, count;
430 struct kvm_kpit_channel_state *s;
432 addr &= KVM_PIT_CHANNEL_MASK;
433 s = &pit_state->channels[addr];
435 mutex_lock(&pit_state->lock);
437 if (s->status_latched) {
438 s->status_latched = 0;
439 ret = s->status;
440 } else if (s->count_latched) {
441 switch (s->count_latched) {
442 default:
443 case RW_STATE_LSB:
444 ret = s->latched_count & 0xff;
445 s->count_latched = 0;
446 break;
447 case RW_STATE_MSB:
448 ret = s->latched_count >> 8;
449 s->count_latched = 0;
450 break;
451 case RW_STATE_WORD0:
452 ret = s->latched_count & 0xff;
453 s->count_latched = RW_STATE_MSB;
454 break;
456 } else {
457 switch (s->read_state) {
458 default:
459 case RW_STATE_LSB:
460 count = pit_get_count(kvm, addr);
461 ret = count & 0xff;
462 break;
463 case RW_STATE_MSB:
464 count = pit_get_count(kvm, addr);
465 ret = (count >> 8) & 0xff;
466 break;
467 case RW_STATE_WORD0:
468 count = pit_get_count(kvm, addr);
469 ret = count & 0xff;
470 s->read_state = RW_STATE_WORD1;
471 break;
472 case RW_STATE_WORD1:
473 count = pit_get_count(kvm, addr);
474 ret = (count >> 8) & 0xff;
475 s->read_state = RW_STATE_WORD0;
476 break;
480 if (len > sizeof(ret))
481 len = sizeof(ret);
482 memcpy(data, (char *)&ret, len);
484 mutex_unlock(&pit_state->lock);
487 static int pit_in_range(struct kvm_io_device *this, gpa_t addr,
488 int len, int is_write)
490 return ((addr >= KVM_PIT_BASE_ADDRESS) &&
491 (addr < KVM_PIT_BASE_ADDRESS + KVM_PIT_MEM_LENGTH));
494 static void speaker_ioport_write(struct kvm_io_device *this,
495 gpa_t addr, int len, const void *data)
497 struct kvm_pit *pit = (struct kvm_pit *)this->private;
498 struct kvm_kpit_state *pit_state = &pit->pit_state;
499 struct kvm *kvm = pit->kvm;
500 u32 val = *(u32 *) data;
502 mutex_lock(&pit_state->lock);
503 pit_state->speaker_data_on = (val >> 1) & 1;
504 pit_set_gate(kvm, 2, val & 1);
505 mutex_unlock(&pit_state->lock);
508 static void speaker_ioport_read(struct kvm_io_device *this,
509 gpa_t addr, int len, void *data)
511 struct kvm_pit *pit = (struct kvm_pit *)this->private;
512 struct kvm_kpit_state *pit_state = &pit->pit_state;
513 struct kvm *kvm = pit->kvm;
514 unsigned int refresh_clock;
515 int ret;
517 /* Refresh clock toggles at about 15us. We approximate as 2^14ns. */
518 refresh_clock = ((unsigned int)ktime_to_ns(ktime_get()) >> 14) & 1;
520 mutex_lock(&pit_state->lock);
521 ret = ((pit_state->speaker_data_on << 1) | pit_get_gate(kvm, 2) |
522 (pit_get_out(kvm, 2) << 5) | (refresh_clock << 4));
523 if (len > sizeof(ret))
524 len = sizeof(ret);
525 memcpy(data, (char *)&ret, len);
526 mutex_unlock(&pit_state->lock);
529 static int speaker_in_range(struct kvm_io_device *this, gpa_t addr,
530 int len, int is_write)
532 return (addr == KVM_SPEAKER_BASE_ADDRESS);
535 void kvm_pit_reset(struct kvm_pit *pit)
537 int i;
538 struct kvm_kpit_channel_state *c;
540 mutex_lock(&pit->pit_state.lock);
541 for (i = 0; i < 3; i++) {
542 c = &pit->pit_state.channels[i];
543 c->mode = 0xff;
544 c->gate = (i != 2);
545 pit_load_count(pit->kvm, i, 0);
547 mutex_unlock(&pit->pit_state.lock);
549 atomic_set(&pit->pit_state.pit_timer.pending, 0);
550 pit->pit_state.irq_ack = 1;
553 static void pit_mask_notifer(struct kvm_irq_mask_notifier *kimn, bool mask)
555 struct kvm_pit *pit = container_of(kimn, struct kvm_pit, mask_notifier);
557 if (!mask) {
558 atomic_set(&pit->pit_state.pit_timer.pending, 0);
559 pit->pit_state.irq_ack = 1;
563 struct kvm_pit *kvm_create_pit(struct kvm *kvm)
565 struct kvm_pit *pit;
566 struct kvm_kpit_state *pit_state;
568 pit = kzalloc(sizeof(struct kvm_pit), GFP_KERNEL);
569 if (!pit)
570 return NULL;
572 pit->irq_source_id = kvm_request_irq_source_id(kvm);
573 if (pit->irq_source_id < 0) {
574 kfree(pit);
575 return NULL;
578 mutex_init(&pit->pit_state.lock);
579 mutex_lock(&pit->pit_state.lock);
580 spin_lock_init(&pit->pit_state.inject_lock);
582 /* Initialize PIO device */
583 pit->dev.read = pit_ioport_read;
584 pit->dev.write = pit_ioport_write;
585 pit->dev.in_range = pit_in_range;
586 pit->dev.private = pit;
587 kvm_io_bus_register_dev(&kvm->pio_bus, &pit->dev);
589 pit->speaker_dev.read = speaker_ioport_read;
590 pit->speaker_dev.write = speaker_ioport_write;
591 pit->speaker_dev.in_range = speaker_in_range;
592 pit->speaker_dev.private = pit;
593 kvm_io_bus_register_dev(&kvm->pio_bus, &pit->speaker_dev);
595 kvm->arch.vpit = pit;
596 pit->kvm = kvm;
598 pit_state = &pit->pit_state;
599 pit_state->pit = pit;
600 hrtimer_init(&pit_state->pit_timer.timer,
601 CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
602 pit_state->irq_ack_notifier.gsi = 0;
603 pit_state->irq_ack_notifier.irq_acked = kvm_pit_ack_irq;
604 kvm_register_irq_ack_notifier(kvm, &pit_state->irq_ack_notifier);
605 pit_state->pit_timer.reinject = true;
606 mutex_unlock(&pit->pit_state.lock);
608 kvm_pit_reset(pit);
610 pit->mask_notifier.func = pit_mask_notifer;
611 kvm_register_irq_mask_notifier(kvm, 0, &pit->mask_notifier);
613 return pit;
616 void kvm_free_pit(struct kvm *kvm)
618 struct hrtimer *timer;
620 if (kvm->arch.vpit) {
621 kvm_unregister_irq_mask_notifier(kvm, 0,
622 &kvm->arch.vpit->mask_notifier);
623 mutex_lock(&kvm->arch.vpit->pit_state.lock);
624 timer = &kvm->arch.vpit->pit_state.pit_timer.timer;
625 hrtimer_cancel(timer);
626 kvm_free_irq_source_id(kvm, kvm->arch.vpit->irq_source_id);
627 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
628 kfree(kvm->arch.vpit);
632 static void __inject_pit_timer_intr(struct kvm *kvm)
634 struct kvm_vcpu *vcpu;
635 int i;
637 mutex_lock(&kvm->lock);
638 kvm_set_irq(kvm, kvm->arch.vpit->irq_source_id, 0, 1);
639 kvm_set_irq(kvm, kvm->arch.vpit->irq_source_id, 0, 0);
640 mutex_unlock(&kvm->lock);
643 * Provides NMI watchdog support via Virtual Wire mode.
644 * The route is: PIT -> PIC -> LVT0 in NMI mode.
646 * Note: Our Virtual Wire implementation is simplified, only
647 * propagating PIT interrupts to all VCPUs when they have set
648 * LVT0 to NMI delivery. Other PIC interrupts are just sent to
649 * VCPU0, and only if its LVT0 is in EXTINT mode.
651 if (kvm->arch.vapics_in_nmi_mode > 0)
652 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
653 vcpu = kvm->vcpus[i];
654 if (vcpu)
655 kvm_apic_nmi_wd_deliver(vcpu);
659 void kvm_inject_pit_timer_irqs(struct kvm_vcpu *vcpu)
661 struct kvm_pit *pit = vcpu->kvm->arch.vpit;
662 struct kvm *kvm = vcpu->kvm;
663 struct kvm_kpit_state *ps;
665 if (vcpu && pit) {
666 int inject = 0;
667 ps = &pit->pit_state;
669 /* Try to inject pending interrupts when
670 * last one has been acked.
672 spin_lock(&ps->inject_lock);
673 if (atomic_read(&ps->pit_timer.pending) && ps->irq_ack) {
674 ps->irq_ack = 0;
675 inject = 1;
677 spin_unlock(&ps->inject_lock);
678 if (inject)
679 __inject_pit_timer_intr(kvm);