arch/x86/kvm/i8254.c

   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 void kvm_pit_ack_irq(struct kvm_irq_ack_notifier *kian)
 219 {
 220         struct kvm_kpit_state *ps = container_of(kian, struct kvm_kpit_state,
 221                                                  irq_ack_notifier);
 222
 223         atomic_set(&ps->irq_ack, 1);
 224         /* irq_ack should be set before pending is read.  Order accesses with
 225          * inc(pending) in pit_timer_fn and xchg(irq_ack, 0) in pit_do_work.
 226          */
 227         smp_mb();
 228         if (atomic_dec_if_positive(&ps->pending) > 0)
 229                 queue_kthread_work(&ps->pit->worker, &ps->pit->expired);
 230 }
 231
 232 void __kvm_migrate_pit_timer(struct kvm_vcpu *vcpu)
 233 {
 234         struct kvm_pit *pit = vcpu->kvm->arch.vpit;
 235         struct hrtimer *timer;
 236
 237         if (!kvm_vcpu_is_bsp(vcpu) || !pit)
 238                 return;
 239
 240         timer = &pit->pit_state.timer;
 241         mutex_lock(&pit->pit_state.lock);
 242         if (hrtimer_cancel(timer))
 243                 hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
 244         mutex_unlock(&pit->pit_state.lock);
 245 }
 246
 247 static void destroy_pit_timer(struct kvm_pit *pit)
 248 {
 249         hrtimer_cancel(&pit->pit_state.timer);
 250         flush_kthread_work(&pit->expired);
 251 }
 252
 253 static void pit_do_work(struct kthread_work *work)
 254 {
 255         struct kvm_pit *pit = container_of(work, struct kvm_pit, expired);
 256         struct kvm *kvm = pit->kvm;
 257         struct kvm_vcpu *vcpu;
 258         int i;
 259         struct kvm_kpit_state *ps = &pit->pit_state;
 260
 261         if (ps->reinject && !atomic_xchg(&ps->irq_ack, 0))
 262                 return;
 263
 264         kvm_set_irq(kvm, kvm->arch.vpit->irq_source_id, 0, 1, false);
 265         kvm_set_irq(kvm, kvm->arch.vpit->irq_source_id, 0, 0, false);
 266
 267         /*
 268          * Provides NMI watchdog support via Virtual Wire mode.
 269          * The route is: PIT -> LVT0 in NMI mode.
 270          *
 271          * Note: Our Virtual Wire implementation does not follow
 272          * the MP specification.  We propagate a PIT interrupt to all
 273          * VCPUs and only when LVT0 is in NMI mode.  The interrupt can
 274          * also be simultaneously delivered through PIC and IOAPIC.
 275          */
 276         if (atomic_read(&kvm->arch.vapics_in_nmi_mode) > 0)
 277                 kvm_for_each_vcpu(i, vcpu, kvm)
 278                         kvm_apic_nmi_wd_deliver(vcpu);
 279 }
 280
 281 static enum hrtimer_restart pit_timer_fn(struct hrtimer *data)
 282 {
 283         struct kvm_kpit_state *ps = container_of(data, struct kvm_kpit_state, timer);
 284         struct kvm_pit *pt = ps->kvm->arch.vpit;
 285
 286         if (ps->reinject)
 287                 atomic_inc(&ps->pending);
 288
 289         queue_kthread_work(&pt->worker, &pt->expired);
 290
 291         if (ps->is_periodic) {
 292                 hrtimer_add_expires_ns(&ps->timer, ps->period);
 293                 return HRTIMER_RESTART;
 294         } else
 295                 return HRTIMER_NORESTART;
 296 }
 297
 298 static inline void kvm_pit_reset_reinject(struct kvm_pit *pit)
 299 {
 300         atomic_set(&pit->pit_state.pending, 0);
 301         atomic_set(&pit->pit_state.irq_ack, 1);
 302 }
 303
 304 void kvm_pit_set_reinject(struct kvm_pit *pit, bool reinject)
 305 {
 306         struct kvm_kpit_state *ps = &pit->pit_state;
 307         struct kvm *kvm = pit->kvm;
 308
 309         if (ps->reinject == reinject)
 310                 return;
 311
 312         if (reinject) {
 313                 /* The initial state is preserved while ps->reinject == 0. */
 314                 kvm_pit_reset_reinject(pit);
 315                 kvm_register_irq_ack_notifier(kvm, &ps->irq_ack_notifier);
 316                 kvm_register_irq_mask_notifier(kvm, 0, &pit->mask_notifier);
 317         } else {
 318                 kvm_unregister_irq_ack_notifier(kvm, &ps->irq_ack_notifier);
 319                 kvm_unregister_irq_mask_notifier(kvm, 0, &pit->mask_notifier);
 320         }
 321
 322         ps->reinject = reinject;
 323 }
 324
 325 static void create_pit_timer(struct kvm_pit *pit, u32 val, int is_period)
 326 {
 327         struct kvm_kpit_state *ps = &pit->pit_state;
 328         struct kvm *kvm = pit->kvm;
 329         s64 interval;
 330
 331         if (!ioapic_in_kernel(kvm) ||
 332             ps->flags & KVM_PIT_FLAGS_HPET_LEGACY)
 333                 return;
 334
 335         interval = muldiv64(val, NSEC_PER_SEC, KVM_PIT_FREQ);
 336
 337         pr_debug("create pit timer, interval is %llu nsec\n", interval);
 338
 339         /* TODO The new value only affected after the retriggered */
 340         hrtimer_cancel(&ps->timer);
 341         flush_kthread_work(&ps->pit->expired);
 342         ps->period = interval;
 343         ps->is_periodic = is_period;
 344
 345         ps->timer.function = pit_timer_fn;
 346         ps->kvm = pit->kvm;
 347
 348         kvm_pit_reset_reinject(pit);
 349
 350         /*
 351          * Do not allow the guest to program periodic timers with small
 352          * interval, since the hrtimers are not throttled by the host
 353          * scheduler.
 354          */
 355         if (ps->is_periodic) {
 356                 s64 min_period = min_timer_period_us * 1000LL;
 357
 358                 if (ps->period < min_period) {
 359                         pr_info_ratelimited(
 360                             "kvm: requested %lld ns "
 361                             "i8254 timer period limited to %lld ns\n",
 362                             ps->period, min_period);
 363                         ps->period = min_period;
 364                 }
 365         }
 366
 367         hrtimer_start(&ps->timer, ktime_add_ns(ktime_get(), interval),
 368                       HRTIMER_MODE_ABS);
 369 }
 370
 371 static void pit_load_count(struct kvm_pit *pit, int channel, u32 val)
 372 {
 373         struct kvm_kpit_state *ps = &pit->pit_state;
 374
 375         pr_debug("load_count val is %d, channel is %d\n", val, channel);
 376
 377         /*
 378          * The largest possible initial count is 0; this is equivalent
 379          * to 216 for binary counting and 104 for BCD counting.
 380          */
 381         if (val == 0)
 382                 val = 0x10000;
 383
 384         ps->channels[channel].count = val;
 385
 386         if (channel != 0) {
 387                 ps->channels[channel].count_load_time = ktime_get();
 388                 return;
 389         }
 390
 391         /* Two types of timer
 392          * mode 1 is one shot, mode 2 is period, otherwise del timer */
 393         switch (ps->channels[0].mode) {
 394         case 0:
 395         case 1:
 396         /* FIXME: enhance mode 4 precision */
 397         case 4:
 398                 create_pit_timer(pit, val, 0);
 399                 break;
 400         case 2:
 401         case 3:
 402                 create_pit_timer(pit, val, 1);
 403                 break;
 404         default:
 405                 destroy_pit_timer(pit);
 406         }
 407 }
 408
 409 void kvm_pit_load_count(struct kvm_pit *pit, int channel, u32 val,
 410                 int hpet_legacy_start)
 411 {
 412         u8 saved_mode;
 413
 414         WARN_ON_ONCE(!mutex_is_locked(&pit->pit_state.lock));
 415
 416         if (hpet_legacy_start) {
 417                 /* save existing mode for later reenablement */
 418                 WARN_ON(channel != 0);
 419                 saved_mode = pit->pit_state.channels[0].mode;
 420                 pit->pit_state.channels[0].mode = 0xff; /* disable timer */
 421                 pit_load_count(pit, channel, val);
 422                 pit->pit_state.channels[0].mode = saved_mode;
 423         } else {
 424                 pit_load_count(pit, channel, val);
 425         }
 426 }
 427
 428 static inline struct kvm_pit *dev_to_pit(struct kvm_io_device *dev)
 429 {
 430         return container_of(dev, struct kvm_pit, dev);
 431 }
 432
 433 static inline struct kvm_pit *speaker_to_pit(struct kvm_io_device *dev)
 434 {
 435         return container_of(dev, struct kvm_pit, speaker_dev);
 436 }
 437
 438 static inline int pit_in_range(gpa_t addr)
 439 {
 440         return ((addr >= KVM_PIT_BASE_ADDRESS) &&
 441                 (addr < KVM_PIT_BASE_ADDRESS + KVM_PIT_MEM_LENGTH));
 442 }
 443
 444 static int pit_ioport_write(struct kvm_vcpu *vcpu,
 445                                 struct kvm_io_device *this,
 446                             gpa_t addr, int len, const void *data)
 447 {
 448         struct kvm_pit *pit = dev_to_pit(this);
 449         struct kvm_kpit_state *pit_state = &pit->pit_state;
 450         int channel, access;
 451         struct kvm_kpit_channel_state *s;
 452         u32 val = *(u32 *) data;
 453         if (!pit_in_range(addr))
 454                 return -EOPNOTSUPP;
 455
 456         val  &= 0xff;
 457         addr &= KVM_PIT_CHANNEL_MASK;
 458
 459         mutex_lock(&pit_state->lock);
 460
 461         if (val != 0)
 462                 pr_debug("write addr is 0x%x, len is %d, val is 0x%x\n",
 463                          (unsigned int)addr, len, val);
 464
 465         if (addr == 3) {
 466                 channel = val >> 6;
 467                 if (channel == 3) {
 468                         /* Read-Back Command. */
 469                         for (channel = 0; channel < 3; channel++) {
 470                                 s = &pit_state->channels[channel];
 471                                 if (val & (2 << channel)) {
 472                                         if (!(val & 0x20))
 473                                                 pit_latch_count(pit, channel);
 474                                         if (!(val & 0x10))
 475                                                 pit_latch_status(pit, channel);
 476                                 }
 477                         }
 478                 } else {
 479                         /* Select Counter <channel>. */
 480                         s = &pit_state->channels[channel];
 481                         access = (val >> 4) & KVM_PIT_CHANNEL_MASK;
 482                         if (access == 0) {
 483                                 pit_latch_count(pit, channel);
 484                         } else {
 485                                 s->rw_mode = access;
 486                                 s->read_state = access;
 487                                 s->write_state = access;
 488                                 s->mode = (val >> 1) & 7;
 489                                 if (s->mode > 5)
 490                                         s->mode -= 4;
 491                                 s->bcd = val & 1;
 492                         }
 493                 }
 494         } else {
 495                 /* Write Count. */
 496                 s = &pit_state->channels[addr];
 497                 switch (s->write_state) {
 498                 default:
 499                 case RW_STATE_LSB:
 500                         pit_load_count(pit, addr, val);
 501                         break;
 502                 case RW_STATE_MSB:
 503                         pit_load_count(pit, addr, val << 8);
 504                         break;
 505                 case RW_STATE_WORD0:
 506                         s->write_latch = val;
 507                         s->write_state = RW_STATE_WORD1;
 508                         break;
 509                 case RW_STATE_WORD1:
 510                         pit_load_count(pit, addr, s->write_latch | (val << 8));
 511                         s->write_state = RW_STATE_WORD0;
 512                         break;
 513                 }
 514         }
 515
 516         mutex_unlock(&pit_state->lock);
 517         return 0;
 518 }
 519
 520 static int pit_ioport_read(struct kvm_vcpu *vcpu,
 521                            struct kvm_io_device *this,
 522                            gpa_t addr, int len, void *data)
 523 {
 524         struct kvm_pit *pit = dev_to_pit(this);
 525         struct kvm_kpit_state *pit_state = &pit->pit_state;
 526         int ret, count;
 527         struct kvm_kpit_channel_state *s;
 528         if (!pit_in_range(addr))
 529                 return -EOPNOTSUPP;
 530
 531         addr &= KVM_PIT_CHANNEL_MASK;
 532         if (addr == 3)
 533                 return 0;
 534
 535         s = &pit_state->channels[addr];
 536
 537         mutex_lock(&pit_state->lock);
 538
 539         if (s->status_latched) {
 540                 s->status_latched = 0;
 541                 ret = s->status;
 542         } else if (s->count_latched) {
 543                 switch (s->count_latched) {
 544                 default:
 545                 case RW_STATE_LSB:
 546                         ret = s->latched_count & 0xff;
 547                         s->count_latched = 0;
 548                         break;
 549                 case RW_STATE_MSB:
 550                         ret = s->latched_count >> 8;
 551                         s->count_latched = 0;
 552                         break;
 553                 case RW_STATE_WORD0:
 554                         ret = s->latched_count & 0xff;
 555                         s->count_latched = RW_STATE_MSB;
 556                         break;
 557                 }
 558         } else {
 559                 switch (s->read_state) {
 560                 default:
 561                 case RW_STATE_LSB:
 562                         count = pit_get_count(pit, addr);
 563                         ret = count & 0xff;
 564                         break;
 565                 case RW_STATE_MSB:
 566                         count = pit_get_count(pit, addr);
 567                         ret = (count >> 8) & 0xff;
 568                         break;
 569                 case RW_STATE_WORD0:
 570                         count = pit_get_count(pit, addr);
 571                         ret = count & 0xff;
 572                         s->read_state = RW_STATE_WORD1;
 573                         break;
 574                 case RW_STATE_WORD1:
 575                         count = pit_get_count(pit, addr);
 576                         ret = (count >> 8) & 0xff;
 577                         s->read_state = RW_STATE_WORD0;
 578                         break;
 579                 }
 580         }
 581
 582         if (len > sizeof(ret))
 583                 len = sizeof(ret);
 584         memcpy(data, (char *)&ret, len);
 585
 586         mutex_unlock(&pit_state->lock);
 587         return 0;
 588 }
 589
 590 static int speaker_ioport_write(struct kvm_vcpu *vcpu,
 591                                 struct kvm_io_device *this,
 592                                 gpa_t addr, int len, const void *data)
 593 {
 594         struct kvm_pit *pit = speaker_to_pit(this);
 595         struct kvm_kpit_state *pit_state = &pit->pit_state;
 596         u32 val = *(u32 *) data;
 597         if (addr != KVM_SPEAKER_BASE_ADDRESS)
 598                 return -EOPNOTSUPP;
 599
 600         mutex_lock(&pit_state->lock);
 601         pit_state->speaker_data_on = (val >> 1) & 1;
 602         pit_set_gate(pit, 2, val & 1);
 603         mutex_unlock(&pit_state->lock);
 604         return 0;
 605 }
 606
 607 static int speaker_ioport_read(struct kvm_vcpu *vcpu,
 608                                    struct kvm_io_device *this,
 609                                    gpa_t addr, int len, void *data)
 610 {
 611         struct kvm_pit *pit = speaker_to_pit(this);
 612         struct kvm_kpit_state *pit_state = &pit->pit_state;
 613         unsigned int refresh_clock;
 614         int ret;
 615         if (addr != KVM_SPEAKER_BASE_ADDRESS)
 616                 return -EOPNOTSUPP;
 617
 618         /* Refresh clock toggles at about 15us. We approximate as 2^14ns. */
 619         refresh_clock = ((unsigned int)ktime_to_ns(ktime_get()) >> 14) & 1;
 620
 621         mutex_lock(&pit_state->lock);
 622         ret = ((pit_state->speaker_data_on << 1) | pit_get_gate(pit, 2) |
 623                 (pit_get_out(pit, 2) << 5) | (refresh_clock << 4));
 624         if (len > sizeof(ret))
 625                 len = sizeof(ret);
 626         memcpy(data, (char *)&ret, len);
 627         mutex_unlock(&pit_state->lock);
 628         return 0;
 629 }
 630
 631 static void kvm_pit_reset(struct kvm_pit *pit)
 632 {
 633         int i;
 634         struct kvm_kpit_channel_state *c;
 635
 636         pit->pit_state.flags = 0;
 637         for (i = 0; i < 3; i++) {
 638                 c = &pit->pit_state.channels[i];
 639                 c->mode = 0xff;
 640                 c->gate = (i != 2);
 641                 pit_load_count(pit, i, 0);
 642         }
 643
 644         kvm_pit_reset_reinject(pit);
 645 }
 646
 647 static void pit_mask_notifer(struct kvm_irq_mask_notifier *kimn, bool mask)
 648 {
 649         struct kvm_pit *pit = container_of(kimn, struct kvm_pit, mask_notifier);
 650
 651         if (!mask)
 652                 kvm_pit_reset_reinject(pit);
 653 }
 654
 655 static const struct kvm_io_device_ops pit_dev_ops = {
 656         .read     = pit_ioport_read,
 657         .write    = pit_ioport_write,
 658 };
 659
 660 static const struct kvm_io_device_ops speaker_dev_ops = {
 661         .read     = speaker_ioport_read,
 662         .write    = speaker_ioport_write,
 663 };
 664
 665 /* Caller must hold slots_lock */
 666 struct kvm_pit *kvm_create_pit(struct kvm *kvm, u32 flags)
 667 {
 668         struct kvm_pit *pit;
 669         struct kvm_kpit_state *pit_state;
 670         struct pid *pid;
 671         pid_t pid_nr;
 672         int ret;
 673
 674         pit = kzalloc(sizeof(struct kvm_pit), GFP_KERNEL);
 675         if (!pit)
 676                 return NULL;
 677
 678         pit->irq_source_id = kvm_request_irq_source_id(kvm);
 679         if (pit->irq_source_id < 0)
 680                 goto fail_request;
 681
 682         mutex_init(&pit->pit_state.lock);
 683
 684         pid = get_pid(task_tgid(current));
 685         pid_nr = pid_vnr(pid);
 686         put_pid(pid);
 687
 688         init_kthread_worker(&pit->worker);
 689         pit->worker_task = kthread_run(kthread_worker_fn, &pit->worker,
 690                                        "kvm-pit/%d", pid_nr);
 691         if (IS_ERR(pit->worker_task))
 692                 goto fail_kthread;
 693
 694         init_kthread_work(&pit->expired, pit_do_work);
 695
 696         pit->kvm = kvm;
 697
 698         pit_state = &pit->pit_state;
 699         pit_state->pit = pit;
 700         hrtimer_init(&pit_state->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
 701
 702         pit_state->irq_ack_notifier.gsi = 0;
 703         pit_state->irq_ack_notifier.irq_acked = kvm_pit_ack_irq;
 704         pit->mask_notifier.func = pit_mask_notifer;
 705
 706         kvm_pit_reset(pit);
 707
 708         kvm_pit_set_reinject(pit, true);
 709
 710         kvm_iodevice_init(&pit->dev, &pit_dev_ops);
 711         ret = kvm_io_bus_register_dev(kvm, KVM_PIO_BUS, KVM_PIT_BASE_ADDRESS,
 712                                       KVM_PIT_MEM_LENGTH, &pit->dev);
 713         if (ret < 0)
 714                 goto fail_register_pit;
 715
 716         if (flags & KVM_PIT_SPEAKER_DUMMY) {
 717                 kvm_iodevice_init(&pit->speaker_dev, &speaker_dev_ops);
 718                 ret = kvm_io_bus_register_dev(kvm, KVM_PIO_BUS,
 719                                               KVM_SPEAKER_BASE_ADDRESS, 4,
 720                                               &pit->speaker_dev);
 721                 if (ret < 0)
 722                         goto fail_register_speaker;
 723         }
 724
 725         return pit;
 726
 727 fail_register_speaker:
 728         kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &pit->dev);
 729 fail_register_pit:
 730         kvm_pit_set_reinject(pit, false);
 731         kthread_stop(pit->worker_task);
 732 fail_kthread:
 733         kvm_free_irq_source_id(kvm, pit->irq_source_id);
 734 fail_request:
 735         kfree(pit);
 736         return NULL;
 737 }
 738
 739 void kvm_free_pit(struct kvm *kvm)
 740 {
 741         struct hrtimer *timer;
 742
 743         if (kvm->arch.vpit) {
 744                 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &kvm->arch.vpit->dev);
 745                 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
 746                                               &kvm->arch.vpit->speaker_dev);
 747                 kvm_pit_set_reinject(kvm->arch.vpit, false);
 748                 timer = &kvm->arch.vpit->pit_state.timer;
 749                 hrtimer_cancel(timer);
 750                 flush_kthread_work(&kvm->arch.vpit->expired);
 751                 kthread_stop(kvm->arch.vpit->worker_task);
 752                 kvm_free_irq_source_id(kvm, kvm->arch.vpit->irq_source_id);
 753                 kfree(kvm->arch.vpit);
 754         }
 755 }