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 <[email protected]>
  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         int 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_request_apicv_update(kvm, false,
 309                                          APICV_INHIBIT_REASON_PIT_REINJ);
 310                 /* The initial state is preserved while ps->reinject == 0. */
 311                 kvm_pit_reset_reinject(pit);
 312                 kvm_register_irq_ack_notifier(kvm, &ps->irq_ack_notifier);
 313                 kvm_register_irq_mask_notifier(kvm, 0, &pit->mask_notifier);
 314         } else {
 315                 kvm_request_apicv_update(kvm, true,
 316                                          APICV_INHIBIT_REASON_PIT_REINJ);
 317                 kvm_unregister_irq_ack_notifier(kvm, &ps->irq_ack_notifier);
 318                 kvm_unregister_irq_mask_notifier(kvm, 0, &pit->mask_notifier);
 319         }
 320
 321         atomic_set(&ps->reinject, reinject);
 322 }
 323
 324 static void create_pit_timer(struct kvm_pit *pit, u32 val, int is_period)
 325 {
 326         struct kvm_kpit_state *ps = &pit->pit_state;
 327         struct kvm *kvm = pit->kvm;
 328         s64 interval;
 329
 330         if (!ioapic_in_kernel(kvm) ||
 331             ps->flags & KVM_PIT_FLAGS_HPET_LEGACY)
 332                 return;
 333
 334         interval = mul_u64_u32_div(val, NSEC_PER_SEC, KVM_PIT_FREQ);
 335
 336         pr_debug("create pit timer, interval is %llu nsec\n", interval);
 337
 338         /* TODO The new value only affected after the retriggered */
 339         hrtimer_cancel(&ps->timer);
 340         kthread_flush_work(&pit->expired);
 341         ps->period = interval;
 342         ps->is_periodic = is_period;
 343
 344         kvm_pit_reset_reinject(pit);
 345
 346         /*
 347          * Do not allow the guest to program periodic timers with small
 348          * interval, since the hrtimers are not throttled by the host
 349          * scheduler.
 350          */
 351         if (ps->is_periodic) {
 352                 s64 min_period = min_timer_period_us * 1000LL;
 353
 354                 if (ps->period < min_period) {
 355                         pr_info_ratelimited(
 356                             "kvm: requested %lld ns "
 357                             "i8254 timer period limited to %lld ns\n",
 358                             ps->period, min_period);
 359                         ps->period = min_period;
 360                 }
 361         }
 362
 363         hrtimer_start(&ps->timer, ktime_add_ns(ktime_get(), interval),
 364                       HRTIMER_MODE_ABS);
 365 }
 366
 367 static void pit_load_count(struct kvm_pit *pit, int channel, u32 val)
 368 {
 369         struct kvm_kpit_state *ps = &pit->pit_state;
 370
 371         pr_debug("load_count val is %u, channel is %d\n", val, channel);
 372
 373         /*
 374          * The largest possible initial count is 0; this is equivalent
 375          * to 216 for binary counting and 104 for BCD counting.
 376          */
 377         if (val == 0)
 378                 val = 0x10000;
 379
 380         ps->channels[channel].count = val;
 381
 382         if (channel != 0) {
 383                 ps->channels[channel].count_load_time = ktime_get();
 384                 return;
 385         }
 386
 387         /* Two types of timer
 388          * mode 1 is one shot, mode 2 is period, otherwise del timer */
 389         switch (ps->channels[0].mode) {
 390         case 0:
 391         case 1:
 392         /* FIXME: enhance mode 4 precision */
 393         case 4:
 394                 create_pit_timer(pit, val, 0);
 395                 break;
 396         case 2:
 397         case 3:
 398                 create_pit_timer(pit, val, 1);
 399                 break;
 400         default:
 401                 destroy_pit_timer(pit);
 402         }
 403 }
 404
 405 void kvm_pit_load_count(struct kvm_pit *pit, int channel, u32 val,
 406                 int hpet_legacy_start)
 407 {
 408         u8 saved_mode;
 409
 410         WARN_ON_ONCE(!mutex_is_locked(&pit->pit_state.lock));
 411
 412         if (hpet_legacy_start) {
 413                 /* save existing mode for later reenablement */
 414                 WARN_ON(channel != 0);
 415                 saved_mode = pit->pit_state.channels[0].mode;
 416                 pit->pit_state.channels[0].mode = 0xff; /* disable timer */
 417                 pit_load_count(pit, channel, val);
 418                 pit->pit_state.channels[0].mode = saved_mode;
 419         } else {
 420                 pit_load_count(pit, channel, val);
 421         }
 422 }
 423
 424 static inline struct kvm_pit *dev_to_pit(struct kvm_io_device *dev)
 425 {
 426         return container_of(dev, struct kvm_pit, dev);
 427 }
 428
 429 static inline struct kvm_pit *speaker_to_pit(struct kvm_io_device *dev)
 430 {
 431         return container_of(dev, struct kvm_pit, speaker_dev);
 432 }
 433
 434 static inline int pit_in_range(gpa_t addr)
 435 {
 436         return ((addr >= KVM_PIT_BASE_ADDRESS) &&
 437                 (addr < KVM_PIT_BASE_ADDRESS + KVM_PIT_MEM_LENGTH));
 438 }
 439
 440 static int pit_ioport_write(struct kvm_vcpu *vcpu,
 441                                 struct kvm_io_device *this,
 442                             gpa_t addr, int len, const void *data)
 443 {
 444         struct kvm_pit *pit = dev_to_pit(this);
 445         struct kvm_kpit_state *pit_state = &pit->pit_state;
 446         int channel, access;
 447         struct kvm_kpit_channel_state *s;
 448         u32 val = *(u32 *) data;
 449         if (!pit_in_range(addr))
 450                 return -EOPNOTSUPP;
 451
 452         val  &= 0xff;
 453         addr &= KVM_PIT_CHANNEL_MASK;
 454
 455         mutex_lock(&pit_state->lock);
 456
 457         if (val != 0)
 458                 pr_debug("write addr is 0x%x, len is %d, val is 0x%x\n",
 459                          (unsigned int)addr, len, val);
 460
 461         if (addr == 3) {
 462                 channel = val >> 6;
 463                 if (channel == 3) {
 464                         /* Read-Back Command. */
 465                         for (channel = 0; channel < 3; channel++) {
 466                                 if (val & (2 << channel)) {
 467                                         if (!(val & 0x20))
 468                                                 pit_latch_count(pit, channel);
 469                                         if (!(val & 0x10))
 470                                                 pit_latch_status(pit, channel);
 471                                 }
 472                         }
 473                 } else {
 474                         /* Select Counter <channel>. */
 475                         s = &pit_state->channels[channel];
 476                         access = (val >> 4) & KVM_PIT_CHANNEL_MASK;
 477                         if (access == 0) {
 478                                 pit_latch_count(pit, channel);
 479                         } else {
 480                                 s->rw_mode = access;
 481                                 s->read_state = access;
 482                                 s->write_state = access;
 483                                 s->mode = (val >> 1) & 7;
 484                                 if (s->mode > 5)
 485                                         s->mode -= 4;
 486                                 s->bcd = val & 1;
 487                         }
 488                 }
 489         } else {
 490                 /* Write Count. */
 491                 s = &pit_state->channels[addr];
 492                 switch (s->write_state) {
 493                 default:
 494                 case RW_STATE_LSB:
 495                         pit_load_count(pit, addr, val);
 496                         break;
 497                 case RW_STATE_MSB:
 498                         pit_load_count(pit, addr, val << 8);
 499                         break;
 500                 case RW_STATE_WORD0:
 501                         s->write_latch = val;
 502                         s->write_state = RW_STATE_WORD1;
 503                         break;
 504                 case RW_STATE_WORD1:
 505                         pit_load_count(pit, addr, s->write_latch | (val << 8));
 506                         s->write_state = RW_STATE_WORD0;
 507                         break;
 508                 }
 509         }
 510
 511         mutex_unlock(&pit_state->lock);
 512         return 0;
 513 }
 514
 515 static int pit_ioport_read(struct kvm_vcpu *vcpu,
 516                            struct kvm_io_device *this,
 517                            gpa_t addr, int len, void *data)
 518 {
 519         struct kvm_pit *pit = dev_to_pit(this);
 520         struct kvm_kpit_state *pit_state = &pit->pit_state;
 521         int ret, count;
 522         struct kvm_kpit_channel_state *s;
 523         if (!pit_in_range(addr))
 524                 return -EOPNOTSUPP;
 525
 526         addr &= KVM_PIT_CHANNEL_MASK;
 527         if (addr == 3)
 528                 return 0;
 529
 530         s = &pit_state->channels[addr];
 531
 532         mutex_lock(&pit_state->lock);
 533
 534         if (s->status_latched) {
 535                 s->status_latched = 0;
 536                 ret = s->status;
 537         } else if (s->count_latched) {
 538                 switch (s->count_latched) {
 539                 default:
 540                 case RW_STATE_LSB:
 541                         ret = s->latched_count & 0xff;
 542                         s->count_latched = 0;
 543                         break;
 544                 case RW_STATE_MSB:
 545                         ret = s->latched_count >> 8;
 546                         s->count_latched = 0;
 547                         break;
 548                 case RW_STATE_WORD0:
 549                         ret = s->latched_count & 0xff;
 550                         s->count_latched = RW_STATE_MSB;
 551                         break;
 552                 }
 553         } else {
 554                 switch (s->read_state) {
 555                 default:
 556                 case RW_STATE_LSB:
 557                         count = pit_get_count(pit, addr);
 558                         ret = count & 0xff;
 559                         break;
 560                 case RW_STATE_MSB:
 561                         count = pit_get_count(pit, addr);
 562                         ret = (count >> 8) & 0xff;
 563                         break;
 564                 case RW_STATE_WORD0:
 565                         count = pit_get_count(pit, addr);
 566                         ret = count & 0xff;
 567                         s->read_state = RW_STATE_WORD1;
 568                         break;
 569                 case RW_STATE_WORD1:
 570                         count = pit_get_count(pit, addr);
 571                         ret = (count >> 8) & 0xff;
 572                         s->read_state = RW_STATE_WORD0;
 573                         break;
 574                 }
 575         }
 576
 577         if (len > sizeof(ret))
 578                 len = sizeof(ret);
 579         memcpy(data, (char *)&ret, len);
 580
 581         mutex_unlock(&pit_state->lock);
 582         return 0;
 583 }
 584
 585 static int speaker_ioport_write(struct kvm_vcpu *vcpu,
 586                                 struct kvm_io_device *this,
 587                                 gpa_t addr, int len, const void *data)
 588 {
 589         struct kvm_pit *pit = speaker_to_pit(this);
 590         struct kvm_kpit_state *pit_state = &pit->pit_state;
 591         u32 val = *(u32 *) data;
 592         if (addr != KVM_SPEAKER_BASE_ADDRESS)
 593                 return -EOPNOTSUPP;
 594
 595         mutex_lock(&pit_state->lock);
 596         pit_state->speaker_data_on = (val >> 1) & 1;
 597         pit_set_gate(pit, 2, val & 1);
 598         mutex_unlock(&pit_state->lock);
 599         return 0;
 600 }
 601
 602 static int speaker_ioport_read(struct kvm_vcpu *vcpu,
 603                                    struct kvm_io_device *this,
 604                                    gpa_t addr, int len, void *data)
 605 {
 606         struct kvm_pit *pit = speaker_to_pit(this);
 607         struct kvm_kpit_state *pit_state = &pit->pit_state;
 608         unsigned int refresh_clock;
 609         int ret;
 610         if (addr != KVM_SPEAKER_BASE_ADDRESS)
 611                 return -EOPNOTSUPP;
 612
 613         /* Refresh clock toggles at about 15us. We approximate as 2^14ns. */
 614         refresh_clock = ((unsigned int)ktime_to_ns(ktime_get()) >> 14) & 1;
 615
 616         mutex_lock(&pit_state->lock);
 617         ret = ((pit_state->speaker_data_on << 1) | pit_get_gate(pit, 2) |
 618                 (pit_get_out(pit, 2) << 5) | (refresh_clock << 4));
 619         if (len > sizeof(ret))
 620                 len = sizeof(ret);
 621         memcpy(data, (char *)&ret, len);
 622         mutex_unlock(&pit_state->lock);
 623         return 0;
 624 }
 625
 626 static void kvm_pit_reset(struct kvm_pit *pit)
 627 {
 628         int i;
 629         struct kvm_kpit_channel_state *c;
 630
 631         pit->pit_state.flags = 0;
 632         for (i = 0; i < 3; i++) {
 633                 c = &pit->pit_state.channels[i];
 634                 c->mode = 0xff;
 635                 c->gate = (i != 2);
 636                 pit_load_count(pit, i, 0);
 637         }
 638
 639         kvm_pit_reset_reinject(pit);
 640 }
 641
 642 static void pit_mask_notifer(struct kvm_irq_mask_notifier *kimn, bool mask)
 643 {
 644         struct kvm_pit *pit = container_of(kimn, struct kvm_pit, mask_notifier);
 645
 646         if (!mask)
 647                 kvm_pit_reset_reinject(pit);
 648 }
 649
 650 static const struct kvm_io_device_ops pit_dev_ops = {
 651         .read     = pit_ioport_read,
 652         .write    = pit_ioport_write,
 653 };
 654
 655 static const struct kvm_io_device_ops speaker_dev_ops = {
 656         .read     = speaker_ioport_read,
 657         .write    = speaker_ioport_write,
 658 };
 659
 660 struct kvm_pit *kvm_create_pit(struct kvm *kvm, u32 flags)
 661 {
 662         struct kvm_pit *pit;
 663         struct kvm_kpit_state *pit_state;
 664         struct pid *pid;
 665         pid_t pid_nr;
 666         int ret;
 667
 668         pit = kzalloc(sizeof(struct kvm_pit), GFP_KERNEL_ACCOUNT);
 669         if (!pit)
 670                 return NULL;
 671
 672         pit->irq_source_id = kvm_request_irq_source_id(kvm);
 673         if (pit->irq_source_id < 0)
 674                 goto fail_request;
 675
 676         mutex_init(&pit->pit_state.lock);
 677
 678         pid = get_pid(task_tgid(current));
 679         pid_nr = pid_vnr(pid);
 680         put_pid(pid);
 681
 682         pit->worker = kthread_create_worker(0, "kvm-pit/%d", pid_nr);
 683         if (IS_ERR(pit->worker))
 684                 goto fail_kthread;
 685
 686         kthread_init_work(&pit->expired, pit_do_work);
 687
 688         pit->kvm = kvm;
 689
 690         pit_state = &pit->pit_state;
 691         hrtimer_init(&pit_state->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
 692         pit_state->timer.function = pit_timer_fn;
 693
 694         pit_state->irq_ack_notifier.gsi = 0;
 695         pit_state->irq_ack_notifier.irq_acked = kvm_pit_ack_irq;
 696         pit->mask_notifier.func = pit_mask_notifer;
 697
 698         kvm_pit_reset(pit);
 699
 700         kvm_pit_set_reinject(pit, true);
 701
 702         mutex_lock(&kvm->slots_lock);
 703         kvm_iodevice_init(&pit->dev, &pit_dev_ops);
 704         ret = kvm_io_bus_register_dev(kvm, KVM_PIO_BUS, KVM_PIT_BASE_ADDRESS,
 705                                       KVM_PIT_MEM_LENGTH, &pit->dev);
 706         if (ret < 0)
 707                 goto fail_register_pit;
 708
 709         if (flags & KVM_PIT_SPEAKER_DUMMY) {
 710                 kvm_iodevice_init(&pit->speaker_dev, &speaker_dev_ops);
 711                 ret = kvm_io_bus_register_dev(kvm, KVM_PIO_BUS,
 712                                               KVM_SPEAKER_BASE_ADDRESS, 4,
 713                                               &pit->speaker_dev);
 714                 if (ret < 0)
 715                         goto fail_register_speaker;
 716         }
 717         mutex_unlock(&kvm->slots_lock);
 718
 719         return pit;
 720
 721 fail_register_speaker:
 722         kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &pit->dev);
 723 fail_register_pit:
 724         mutex_unlock(&kvm->slots_lock);
 725         kvm_pit_set_reinject(pit, false);
 726         kthread_destroy_worker(pit->worker);
 727 fail_kthread:
 728         kvm_free_irq_source_id(kvm, pit->irq_source_id);
 729 fail_request:
 730         kfree(pit);
 731         return NULL;
 732 }
 733
 734 void kvm_free_pit(struct kvm *kvm)
 735 {
 736         struct kvm_pit *pit = kvm->arch.vpit;
 737
 738         if (pit) {
 739                 mutex_lock(&kvm->slots_lock);
 740                 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &pit->dev);
 741                 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &pit->speaker_dev);
 742                 mutex_unlock(&kvm->slots_lock);
 743                 kvm_pit_set_reinject(pit, false);
 744                 hrtimer_cancel(&pit->pit_state.timer);
 745                 kthread_destroy_worker(pit->worker);
 746                 kvm_free_irq_source_id(kvm, pit->irq_source_id);
 747                 kfree(pit);
 748         }
 749 }