1 2 /* 3 * Local APIC virtualization 4 * 5 * Copyright (C) 2006 Qumranet, Inc. 6 * Copyright (C) 2007 Novell 7 * Copyright (C) 2007 Intel 8 * 9 * Authors: 10 * Dor Laor <dor.laor@qumranet.com> 11 * Gregory Haskins <ghaskins@novell.com> 12 * Yaozu (Eddie) Dong <eddie.dong@intel.com> 13 * 14 * Based on Xen 3.1 code, Copyright (c) 2004, Intel Corporation. 15 * 16 * This work is licensed under the terms of the GNU GPL, version 2. See 17 * the COPYING file in the top-level directory. 18 */ 19 20 #include <linux/kvm_host.h> 21 #include <linux/kvm.h> 22 #include <linux/mm.h> 23 #include <linux/highmem.h> 24 #include <linux/smp.h> 25 #include <linux/hrtimer.h> 26 #include <linux/io.h> 27 #include <linux/module.h> 28 #include <asm/processor.h> 29 #include <asm/msr.h> 30 #include <asm/page.h> 31 #include <asm/current.h> 32 #include <asm/apicdef.h> 33 #include <asm/atomic.h> 34 #include <asm/div64.h> 35 #include "irq.h" 36 37 #define PRId64 "d" 38 #define PRIx64 "llx" 39 #define PRIu64 "u" 40 #define PRIo64 "o" 41 42 #define APIC_BUS_CYCLE_NS 1 43 44 /* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */ 45 #define apic_debug(fmt, arg...) 46 47 #define APIC_LVT_NUM 6 48 /* 14 is the version for Xeon and Pentium 8.4.8*/ 49 #define APIC_VERSION (0x14UL | ((APIC_LVT_NUM - 1) << 16)) 50 #define LAPIC_MMIO_LENGTH (1 << 12) 51 /* followed define is not in apicdef.h */ 52 #define APIC_SHORT_MASK 0xc0000 53 #define APIC_DEST_NOSHORT 0x0 54 #define APIC_DEST_MASK 0x800 55 #define MAX_APIC_VECTOR 256 56 57 #define VEC_POS(v) ((v) & (32 - 1)) 58 #define REG_POS(v) (((v) >> 5) << 4) 59 60 static inline u32 apic_get_reg(struct kvm_lapic *apic, int reg_off) 61 { 62 return *((u32 *) (apic->regs + reg_off)); 63 } 64 65 static inline void apic_set_reg(struct kvm_lapic *apic, int reg_off, u32 val) 66 { 67 *((u32 *) (apic->regs + reg_off)) = val; 68 } 69 70 static inline int apic_test_and_set_vector(int vec, void *bitmap) 71 { 72 return test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec)); 73 } 74 75 static inline int apic_test_and_clear_vector(int vec, void *bitmap) 76 { 77 return test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec)); 78 } 79 80 static inline void apic_set_vector(int vec, void *bitmap) 81 { 82 set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec)); 83 } 84 85 static inline void apic_clear_vector(int vec, void *bitmap) 86 { 87 clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec)); 88 } 89 90 static inline int apic_hw_enabled(struct kvm_lapic *apic) 91 { 92 return (apic)->vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE; 93 } 94 95 static inline int apic_sw_enabled(struct kvm_lapic *apic) 96 { 97 return apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_APIC_ENABLED; 98 } 99 100 static inline int apic_enabled(struct kvm_lapic *apic) 101 { 102 return apic_sw_enabled(apic) && apic_hw_enabled(apic); 103 } 104 105 #define LVT_MASK \ 106 (APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK) 107 108 #define LINT_MASK \ 109 (LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \ 110 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER) 111 112 static inline int kvm_apic_id(struct kvm_lapic *apic) 113 { 114 return (apic_get_reg(apic, APIC_ID) >> 24) & 0xff; 115 } 116 117 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type) 118 { 119 return !(apic_get_reg(apic, lvt_type) & APIC_LVT_MASKED); 120 } 121 122 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type) 123 { 124 return apic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK; 125 } 126 127 static inline int apic_lvtt_period(struct kvm_lapic *apic) 128 { 129 return apic_get_reg(apic, APIC_LVTT) & APIC_LVT_TIMER_PERIODIC; 130 } 131 132 static unsigned int apic_lvt_mask[APIC_LVT_NUM] = { 133 LVT_MASK | APIC_LVT_TIMER_PERIODIC, /* LVTT */ 134 LVT_MASK | APIC_MODE_MASK, /* LVTTHMR */ 135 LVT_MASK | APIC_MODE_MASK, /* LVTPC */ 136 LINT_MASK, LINT_MASK, /* LVT0-1 */ 137 LVT_MASK /* LVTERR */ 138 }; 139 140 static int find_highest_vector(void *bitmap) 141 { 142 u32 *word = bitmap; 143 int word_offset = MAX_APIC_VECTOR >> 5; 144 145 while ((word_offset != 0) && (word[(--word_offset) << 2] == 0)) 146 continue; 147 148 if (likely(!word_offset && !word[0])) 149 return -1; 150 else 151 return fls(word[word_offset << 2]) - 1 + (word_offset << 5); 152 } 153 154 static inline int apic_test_and_set_irr(int vec, struct kvm_lapic *apic) 155 { 156 return apic_test_and_set_vector(vec, apic->regs + APIC_IRR); 157 } 158 159 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic) 160 { 161 apic_clear_vector(vec, apic->regs + APIC_IRR); 162 } 163 164 static inline int apic_find_highest_irr(struct kvm_lapic *apic) 165 { 166 int result; 167 168 result = find_highest_vector(apic->regs + APIC_IRR); 169 ASSERT(result == -1 || result >= 16); 170 171 return result; 172 } 173 174 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu) 175 { 176 struct kvm_lapic *apic = vcpu->arch.apic; 177 int highest_irr; 178 179 if (!apic) 180 return 0; 181 highest_irr = apic_find_highest_irr(apic); 182 183 return highest_irr; 184 } 185 EXPORT_SYMBOL_GPL(kvm_lapic_find_highest_irr); 186 187 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, u8 vec, u8 trig) 188 { 189 struct kvm_lapic *apic = vcpu->arch.apic; 190 191 if (!apic_test_and_set_irr(vec, apic)) { 192 /* a new pending irq is set in IRR */ 193 if (trig) 194 apic_set_vector(vec, apic->regs + APIC_TMR); 195 else 196 apic_clear_vector(vec, apic->regs + APIC_TMR); 197 kvm_vcpu_kick(apic->vcpu); 198 return 1; 199 } 200 return 0; 201 } 202 203 static inline int apic_find_highest_isr(struct kvm_lapic *apic) 204 { 205 int result; 206 207 result = find_highest_vector(apic->regs + APIC_ISR); 208 ASSERT(result == -1 || result >= 16); 209 210 return result; 211 } 212 213 static void apic_update_ppr(struct kvm_lapic *apic) 214 { 215 u32 tpr, isrv, ppr; 216 int isr; 217 218 tpr = apic_get_reg(apic, APIC_TASKPRI); 219 isr = apic_find_highest_isr(apic); 220 isrv = (isr != -1) ? isr : 0; 221 222 if ((tpr & 0xf0) >= (isrv & 0xf0)) 223 ppr = tpr & 0xff; 224 else 225 ppr = isrv & 0xf0; 226 227 apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x", 228 apic, ppr, isr, isrv); 229 230 apic_set_reg(apic, APIC_PROCPRI, ppr); 231 } 232 233 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr) 234 { 235 apic_set_reg(apic, APIC_TASKPRI, tpr); 236 apic_update_ppr(apic); 237 } 238 239 int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u16 dest) 240 { 241 return kvm_apic_id(apic) == dest; 242 } 243 244 int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u8 mda) 245 { 246 int result = 0; 247 u8 logical_id; 248 249 logical_id = GET_APIC_LOGICAL_ID(apic_get_reg(apic, APIC_LDR)); 250 251 switch (apic_get_reg(apic, APIC_DFR)) { 252 case APIC_DFR_FLAT: 253 if (logical_id & mda) 254 result = 1; 255 break; 256 case APIC_DFR_CLUSTER: 257 if (((logical_id >> 4) == (mda >> 0x4)) 258 && (logical_id & mda & 0xf)) 259 result = 1; 260 break; 261 default: 262 printk(KERN_WARNING "Bad DFR vcpu %d: %08x\n", 263 apic->vcpu->vcpu_id, apic_get_reg(apic, APIC_DFR)); 264 break; 265 } 266 267 return result; 268 } 269 270 static int apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source, 271 int short_hand, int dest, int dest_mode) 272 { 273 int result = 0; 274 struct kvm_lapic *target = vcpu->arch.apic; 275 276 apic_debug("target %p, source %p, dest 0x%x, " 277 "dest_mode 0x%x, short_hand 0x%x", 278 target, source, dest, dest_mode, short_hand); 279 280 ASSERT(!target); 281 switch (short_hand) { 282 case APIC_DEST_NOSHORT: 283 if (dest_mode == 0) { 284 /* Physical mode. */ 285 if ((dest == 0xFF) || (dest == kvm_apic_id(target))) 286 result = 1; 287 } else 288 /* Logical mode. */ 289 result = kvm_apic_match_logical_addr(target, dest); 290 break; 291 case APIC_DEST_SELF: 292 if (target == source) 293 result = 1; 294 break; 295 case APIC_DEST_ALLINC: 296 result = 1; 297 break; 298 case APIC_DEST_ALLBUT: 299 if (target != source) 300 result = 1; 301 break; 302 default: 303 printk(KERN_WARNING "Bad dest shorthand value %x\n", 304 short_hand); 305 break; 306 } 307 308 return result; 309 } 310 311 /* 312 * Add a pending IRQ into lapic. 313 * Return 1 if successfully added and 0 if discarded. 314 */ 315 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode, 316 int vector, int level, int trig_mode) 317 { 318 int orig_irr, result = 0; 319 struct kvm_vcpu *vcpu = apic->vcpu; 320 321 switch (delivery_mode) { 322 case APIC_DM_FIXED: 323 case APIC_DM_LOWEST: 324 /* FIXME add logic for vcpu on reset */ 325 if (unlikely(!apic_enabled(apic))) 326 break; 327 328 orig_irr = apic_test_and_set_irr(vector, apic); 329 if (orig_irr && trig_mode) { 330 apic_debug("level trig mode repeatedly for vector %d", 331 vector); 332 break; 333 } 334 335 if (trig_mode) { 336 apic_debug("level trig mode for vector %d", vector); 337 apic_set_vector(vector, apic->regs + APIC_TMR); 338 } else 339 apic_clear_vector(vector, apic->regs + APIC_TMR); 340 341 if (vcpu->arch.mp_state == VCPU_MP_STATE_RUNNABLE) 342 kvm_vcpu_kick(vcpu); 343 else if (vcpu->arch.mp_state == VCPU_MP_STATE_HALTED) { 344 vcpu->arch.mp_state = VCPU_MP_STATE_RUNNABLE; 345 if (waitqueue_active(&vcpu->wq)) 346 wake_up_interruptible(&vcpu->wq); 347 } 348 349 result = (orig_irr == 0); 350 break; 351 352 case APIC_DM_REMRD: 353 printk(KERN_DEBUG "Ignoring delivery mode 3\n"); 354 break; 355 356 case APIC_DM_SMI: 357 printk(KERN_DEBUG "Ignoring guest SMI\n"); 358 break; 359 case APIC_DM_NMI: 360 printk(KERN_DEBUG "Ignoring guest NMI\n"); 361 break; 362 363 case APIC_DM_INIT: 364 if (level) { 365 if (vcpu->arch.mp_state == VCPU_MP_STATE_RUNNABLE) 366 printk(KERN_DEBUG 367 "INIT on a runnable vcpu %d\n", 368 vcpu->vcpu_id); 369 vcpu->arch.mp_state = VCPU_MP_STATE_INIT_RECEIVED; 370 kvm_vcpu_kick(vcpu); 371 } else { 372 printk(KERN_DEBUG 373 "Ignoring de-assert INIT to vcpu %d\n", 374 vcpu->vcpu_id); 375 } 376 377 break; 378 379 case APIC_DM_STARTUP: 380 printk(KERN_DEBUG "SIPI to vcpu %d vector 0x%02x\n", 381 vcpu->vcpu_id, vector); 382 if (vcpu->arch.mp_state == VCPU_MP_STATE_INIT_RECEIVED) { 383 vcpu->arch.sipi_vector = vector; 384 vcpu->arch.mp_state = VCPU_MP_STATE_SIPI_RECEIVED; 385 if (waitqueue_active(&vcpu->wq)) 386 wake_up_interruptible(&vcpu->wq); 387 } 388 break; 389 390 default: 391 printk(KERN_ERR "TODO: unsupported delivery mode %x\n", 392 delivery_mode); 393 break; 394 } 395 return result; 396 } 397 398 static struct kvm_lapic *kvm_apic_round_robin(struct kvm *kvm, u8 vector, 399 unsigned long bitmap) 400 { 401 int last; 402 int next; 403 struct kvm_lapic *apic = NULL; 404 405 last = kvm->arch.round_robin_prev_vcpu; 406 next = last; 407 408 do { 409 if (++next == KVM_MAX_VCPUS) 410 next = 0; 411 if (kvm->vcpus[next] == NULL || !test_bit(next, &bitmap)) 412 continue; 413 apic = kvm->vcpus[next]->arch.apic; 414 if (apic && apic_enabled(apic)) 415 break; 416 apic = NULL; 417 } while (next != last); 418 kvm->arch.round_robin_prev_vcpu = next; 419 420 if (!apic) 421 printk(KERN_DEBUG "vcpu not ready for apic_round_robin\n"); 422 423 return apic; 424 } 425 426 struct kvm_vcpu *kvm_get_lowest_prio_vcpu(struct kvm *kvm, u8 vector, 427 unsigned long bitmap) 428 { 429 struct kvm_lapic *apic; 430 431 apic = kvm_apic_round_robin(kvm, vector, bitmap); 432 if (apic) 433 return apic->vcpu; 434 return NULL; 435 } 436 437 static void apic_set_eoi(struct kvm_lapic *apic) 438 { 439 int vector = apic_find_highest_isr(apic); 440 441 /* 442 * Not every write EOI will has corresponding ISR, 443 * one example is when Kernel check timer on setup_IO_APIC 444 */ 445 if (vector == -1) 446 return; 447 448 apic_clear_vector(vector, apic->regs + APIC_ISR); 449 apic_update_ppr(apic); 450 451 if (apic_test_and_clear_vector(vector, apic->regs + APIC_TMR)) 452 kvm_ioapic_update_eoi(apic->vcpu->kvm, vector); 453 } 454 455 static void apic_send_ipi(struct kvm_lapic *apic) 456 { 457 u32 icr_low = apic_get_reg(apic, APIC_ICR); 458 u32 icr_high = apic_get_reg(apic, APIC_ICR2); 459 460 unsigned int dest = GET_APIC_DEST_FIELD(icr_high); 461 unsigned int short_hand = icr_low & APIC_SHORT_MASK; 462 unsigned int trig_mode = icr_low & APIC_INT_LEVELTRIG; 463 unsigned int level = icr_low & APIC_INT_ASSERT; 464 unsigned int dest_mode = icr_low & APIC_DEST_MASK; 465 unsigned int delivery_mode = icr_low & APIC_MODE_MASK; 466 unsigned int vector = icr_low & APIC_VECTOR_MASK; 467 468 struct kvm_vcpu *target; 469 struct kvm_vcpu *vcpu; 470 unsigned long lpr_map = 0; 471 int i; 472 473 apic_debug("icr_high 0x%x, icr_low 0x%x, " 474 "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, " 475 "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n", 476 icr_high, icr_low, short_hand, dest, 477 trig_mode, level, dest_mode, delivery_mode, vector); 478 479 for (i = 0; i < KVM_MAX_VCPUS; i++) { 480 vcpu = apic->vcpu->kvm->vcpus[i]; 481 if (!vcpu) 482 continue; 483 484 if (vcpu->arch.apic && 485 apic_match_dest(vcpu, apic, short_hand, dest, dest_mode)) { 486 if (delivery_mode == APIC_DM_LOWEST) 487 set_bit(vcpu->vcpu_id, &lpr_map); 488 else 489 __apic_accept_irq(vcpu->arch.apic, delivery_mode, 490 vector, level, trig_mode); 491 } 492 } 493 494 if (delivery_mode == APIC_DM_LOWEST) { 495 target = kvm_get_lowest_prio_vcpu(vcpu->kvm, vector, lpr_map); 496 if (target != NULL) 497 __apic_accept_irq(target->arch.apic, delivery_mode, 498 vector, level, trig_mode); 499 } 500 } 501 502 static u32 apic_get_tmcct(struct kvm_lapic *apic) 503 { 504 u64 counter_passed; 505 ktime_t passed, now; 506 u32 tmcct; 507 508 ASSERT(apic != NULL); 509 510 now = apic->timer.dev.base->get_time(); 511 tmcct = apic_get_reg(apic, APIC_TMICT); 512 513 /* if initial count is 0, current count should also be 0 */ 514 if (tmcct == 0) 515 return 0; 516 517 if (unlikely(ktime_to_ns(now) <= 518 ktime_to_ns(apic->timer.last_update))) { 519 /* Wrap around */ 520 passed = ktime_add(( { 521 (ktime_t) { 522 .tv64 = KTIME_MAX - 523 (apic->timer.last_update).tv64}; } 524 ), now); 525 apic_debug("time elapsed\n"); 526 } else 527 passed = ktime_sub(now, apic->timer.last_update); 528 529 counter_passed = div64_64(ktime_to_ns(passed), 530 (APIC_BUS_CYCLE_NS * apic->timer.divide_count)); 531 532 if (counter_passed > tmcct) { 533 if (unlikely(!apic_lvtt_period(apic))) { 534 /* one-shot timers stick at 0 until reset */ 535 tmcct = 0; 536 } else { 537 /* 538 * periodic timers reset to APIC_TMICT when they 539 * hit 0. The while loop simulates this happening N 540 * times. (counter_passed %= tmcct) would also work, 541 * but might be slower or not work on 32-bit?? 542 */ 543 while (counter_passed > tmcct) 544 counter_passed -= tmcct; 545 tmcct -= counter_passed; 546 } 547 } else { 548 tmcct -= counter_passed; 549 } 550 551 return tmcct; 552 } 553 554 static void __report_tpr_access(struct kvm_lapic *apic, bool write) 555 { 556 struct kvm_vcpu *vcpu = apic->vcpu; 557 struct kvm_run *run = vcpu->run; 558 559 set_bit(KVM_REQ_REPORT_TPR_ACCESS, &vcpu->requests); 560 kvm_x86_ops->cache_regs(vcpu); 561 run->tpr_access.rip = vcpu->arch.rip; 562 run->tpr_access.is_write = write; 563 } 564 565 static inline void report_tpr_access(struct kvm_lapic *apic, bool write) 566 { 567 if (apic->vcpu->arch.tpr_access_reporting) 568 __report_tpr_access(apic, write); 569 } 570 571 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset) 572 { 573 u32 val = 0; 574 575 if (offset >= LAPIC_MMIO_LENGTH) 576 return 0; 577 578 switch (offset) { 579 case APIC_ARBPRI: 580 printk(KERN_WARNING "Access APIC ARBPRI register " 581 "which is for P6\n"); 582 break; 583 584 case APIC_TMCCT: /* Timer CCR */ 585 val = apic_get_tmcct(apic); 586 break; 587 588 case APIC_TASKPRI: 589 report_tpr_access(apic, false); 590 /* fall thru */ 591 default: 592 apic_update_ppr(apic); 593 val = apic_get_reg(apic, offset); 594 break; 595 } 596 597 return val; 598 } 599 600 static void apic_mmio_read(struct kvm_io_device *this, 601 gpa_t address, int len, void *data) 602 { 603 struct kvm_lapic *apic = (struct kvm_lapic *)this->private; 604 unsigned int offset = address - apic->base_address; 605 unsigned char alignment = offset & 0xf; 606 u32 result; 607 608 if ((alignment + len) > 4) { 609 printk(KERN_ERR "KVM_APIC_READ: alignment error %lx %d", 610 (unsigned long)address, len); 611 return; 612 } 613 result = __apic_read(apic, offset & ~0xf); 614 615 switch (len) { 616 case 1: 617 case 2: 618 case 4: 619 memcpy(data, (char *)&result + alignment, len); 620 break; 621 default: 622 printk(KERN_ERR "Local APIC read with len = %x, " 623 "should be 1,2, or 4 instead\n", len); 624 break; 625 } 626 } 627 628 static void update_divide_count(struct kvm_lapic *apic) 629 { 630 u32 tmp1, tmp2, tdcr; 631 632 tdcr = apic_get_reg(apic, APIC_TDCR); 633 tmp1 = tdcr & 0xf; 634 tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1; 635 apic->timer.divide_count = 0x1 << (tmp2 & 0x7); 636 637 apic_debug("timer divide count is 0x%x\n", 638 apic->timer.divide_count); 639 } 640 641 static void start_apic_timer(struct kvm_lapic *apic) 642 { 643 ktime_t now = apic->timer.dev.base->get_time(); 644 645 apic->timer.last_update = now; 646 647 apic->timer.period = apic_get_reg(apic, APIC_TMICT) * 648 APIC_BUS_CYCLE_NS * apic->timer.divide_count; 649 atomic_set(&apic->timer.pending, 0); 650 hrtimer_start(&apic->timer.dev, 651 ktime_add_ns(now, apic->timer.period), 652 HRTIMER_MODE_ABS); 653 654 apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016" 655 PRIx64 ", " 656 "timer initial count 0x%x, period %lldns, " 657 "expire @ 0x%016" PRIx64 ".\n", __FUNCTION__, 658 APIC_BUS_CYCLE_NS, ktime_to_ns(now), 659 apic_get_reg(apic, APIC_TMICT), 660 apic->timer.period, 661 ktime_to_ns(ktime_add_ns(now, 662 apic->timer.period))); 663 } 664 665 static void apic_mmio_write(struct kvm_io_device *this, 666 gpa_t address, int len, const void *data) 667 { 668 struct kvm_lapic *apic = (struct kvm_lapic *)this->private; 669 unsigned int offset = address - apic->base_address; 670 unsigned char alignment = offset & 0xf; 671 u32 val; 672 673 /* 674 * APIC register must be aligned on 128-bits boundary. 675 * 32/64/128 bits registers must be accessed thru 32 bits. 676 * Refer SDM 8.4.1 677 */ 678 if (len != 4 || alignment) { 679 if (printk_ratelimit()) 680 printk(KERN_ERR "apic write: bad size=%d %lx\n", 681 len, (long)address); 682 return; 683 } 684 685 val = *(u32 *) data; 686 687 /* too common printing */ 688 if (offset != APIC_EOI) 689 apic_debug("%s: offset 0x%x with length 0x%x, and value is " 690 "0x%x\n", __FUNCTION__, offset, len, val); 691 692 offset &= 0xff0; 693 694 switch (offset) { 695 case APIC_ID: /* Local APIC ID */ 696 apic_set_reg(apic, APIC_ID, val); 697 break; 698 699 case APIC_TASKPRI: 700 report_tpr_access(apic, true); 701 apic_set_tpr(apic, val & 0xff); 702 break; 703 704 case APIC_EOI: 705 apic_set_eoi(apic); 706 break; 707 708 case APIC_LDR: 709 apic_set_reg(apic, APIC_LDR, val & APIC_LDR_MASK); 710 break; 711 712 case APIC_DFR: 713 apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF); 714 break; 715 716 case APIC_SPIV: 717 apic_set_reg(apic, APIC_SPIV, val & 0x3ff); 718 if (!(val & APIC_SPIV_APIC_ENABLED)) { 719 int i; 720 u32 lvt_val; 721 722 for (i = 0; i < APIC_LVT_NUM; i++) { 723 lvt_val = apic_get_reg(apic, 724 APIC_LVTT + 0x10 * i); 725 apic_set_reg(apic, APIC_LVTT + 0x10 * i, 726 lvt_val | APIC_LVT_MASKED); 727 } 728 atomic_set(&apic->timer.pending, 0); 729 730 } 731 break; 732 733 case APIC_ICR: 734 /* No delay here, so we always clear the pending bit */ 735 apic_set_reg(apic, APIC_ICR, val & ~(1 << 12)); 736 apic_send_ipi(apic); 737 break; 738 739 case APIC_ICR2: 740 apic_set_reg(apic, APIC_ICR2, val & 0xff000000); 741 break; 742 743 case APIC_LVTT: 744 case APIC_LVTTHMR: 745 case APIC_LVTPC: 746 case APIC_LVT0: 747 case APIC_LVT1: 748 case APIC_LVTERR: 749 /* TODO: Check vector */ 750 if (!apic_sw_enabled(apic)) 751 val |= APIC_LVT_MASKED; 752 753 val &= apic_lvt_mask[(offset - APIC_LVTT) >> 4]; 754 apic_set_reg(apic, offset, val); 755 756 break; 757 758 case APIC_TMICT: 759 hrtimer_cancel(&apic->timer.dev); 760 apic_set_reg(apic, APIC_TMICT, val); 761 start_apic_timer(apic); 762 return; 763 764 case APIC_TDCR: 765 if (val & 4) 766 printk(KERN_ERR "KVM_WRITE:TDCR %x\n", val); 767 apic_set_reg(apic, APIC_TDCR, val); 768 update_divide_count(apic); 769 break; 770 771 default: 772 apic_debug("Local APIC Write to read-only register %x\n", 773 offset); 774 break; 775 } 776 777 } 778 779 static int apic_mmio_range(struct kvm_io_device *this, gpa_t addr) 780 { 781 struct kvm_lapic *apic = (struct kvm_lapic *)this->private; 782 int ret = 0; 783 784 785 if (apic_hw_enabled(apic) && 786 (addr >= apic->base_address) && 787 (addr < (apic->base_address + LAPIC_MMIO_LENGTH))) 788 ret = 1; 789 790 return ret; 791 } 792 793 void kvm_free_lapic(struct kvm_vcpu *vcpu) 794 { 795 if (!vcpu->arch.apic) 796 return; 797 798 hrtimer_cancel(&vcpu->arch.apic->timer.dev); 799 800 if (vcpu->arch.apic->regs_page) 801 __free_page(vcpu->arch.apic->regs_page); 802 803 kfree(vcpu->arch.apic); 804 } 805 806 /* 807 *---------------------------------------------------------------------- 808 * LAPIC interface 809 *---------------------------------------------------------------------- 810 */ 811 812 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8) 813 { 814 struct kvm_lapic *apic = vcpu->arch.apic; 815 816 if (!apic) 817 return; 818 apic_set_tpr(apic, ((cr8 & 0x0f) << 4) 819 | (apic_get_reg(apic, APIC_TASKPRI) & 4)); 820 } 821 822 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu) 823 { 824 struct kvm_lapic *apic = vcpu->arch.apic; 825 u64 tpr; 826 827 if (!apic) 828 return 0; 829 tpr = (u64) apic_get_reg(apic, APIC_TASKPRI); 830 831 return (tpr & 0xf0) >> 4; 832 } 833 EXPORT_SYMBOL_GPL(kvm_lapic_get_cr8); 834 835 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value) 836 { 837 struct kvm_lapic *apic = vcpu->arch.apic; 838 839 if (!apic) { 840 value |= MSR_IA32_APICBASE_BSP; 841 vcpu->arch.apic_base = value; 842 return; 843 } 844 if (apic->vcpu->vcpu_id) 845 value &= ~MSR_IA32_APICBASE_BSP; 846 847 vcpu->arch.apic_base = value; 848 apic->base_address = apic->vcpu->arch.apic_base & 849 MSR_IA32_APICBASE_BASE; 850 851 /* with FSB delivery interrupt, we can restart APIC functionality */ 852 apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is " 853 "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address); 854 855 } 856 857 u64 kvm_lapic_get_base(struct kvm_vcpu *vcpu) 858 { 859 return vcpu->arch.apic_base; 860 } 861 EXPORT_SYMBOL_GPL(kvm_lapic_get_base); 862 863 void kvm_lapic_reset(struct kvm_vcpu *vcpu) 864 { 865 struct kvm_lapic *apic; 866 int i; 867 868 apic_debug("%s\n", __FUNCTION__); 869 870 ASSERT(vcpu); 871 apic = vcpu->arch.apic; 872 ASSERT(apic != NULL); 873 874 /* Stop the timer in case it's a reset to an active apic */ 875 hrtimer_cancel(&apic->timer.dev); 876 877 apic_set_reg(apic, APIC_ID, vcpu->vcpu_id << 24); 878 apic_set_reg(apic, APIC_LVR, APIC_VERSION); 879 880 for (i = 0; i < APIC_LVT_NUM; i++) 881 apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED); 882 apic_set_reg(apic, APIC_LVT0, 883 SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT)); 884 885 apic_set_reg(apic, APIC_DFR, 0xffffffffU); 886 apic_set_reg(apic, APIC_SPIV, 0xff); 887 apic_set_reg(apic, APIC_TASKPRI, 0); 888 apic_set_reg(apic, APIC_LDR, 0); 889 apic_set_reg(apic, APIC_ESR, 0); 890 apic_set_reg(apic, APIC_ICR, 0); 891 apic_set_reg(apic, APIC_ICR2, 0); 892 apic_set_reg(apic, APIC_TDCR, 0); 893 apic_set_reg(apic, APIC_TMICT, 0); 894 for (i = 0; i < 8; i++) { 895 apic_set_reg(apic, APIC_IRR + 0x10 * i, 0); 896 apic_set_reg(apic, APIC_ISR + 0x10 * i, 0); 897 apic_set_reg(apic, APIC_TMR + 0x10 * i, 0); 898 } 899 update_divide_count(apic); 900 atomic_set(&apic->timer.pending, 0); 901 if (vcpu->vcpu_id == 0) 902 vcpu->arch.apic_base |= MSR_IA32_APICBASE_BSP; 903 apic_update_ppr(apic); 904 905 apic_debug(KERN_INFO "%s: vcpu=%p, id=%d, base_msr=" 906 "0x%016" PRIx64 ", base_address=0x%0lx.\n", __FUNCTION__, 907 vcpu, kvm_apic_id(apic), 908 vcpu->arch.apic_base, apic->base_address); 909 } 910 EXPORT_SYMBOL_GPL(kvm_lapic_reset); 911 912 int kvm_lapic_enabled(struct kvm_vcpu *vcpu) 913 { 914 struct kvm_lapic *apic = vcpu->arch.apic; 915 int ret = 0; 916 917 if (!apic) 918 return 0; 919 ret = apic_enabled(apic); 920 921 return ret; 922 } 923 EXPORT_SYMBOL_GPL(kvm_lapic_enabled); 924 925 /* 926 *---------------------------------------------------------------------- 927 * timer interface 928 *---------------------------------------------------------------------- 929 */ 930 931 /* TODO: make sure __apic_timer_fn runs in current pCPU */ 932 static int __apic_timer_fn(struct kvm_lapic *apic) 933 { 934 int result = 0; 935 wait_queue_head_t *q = &apic->vcpu->wq; 936 937 atomic_inc(&apic->timer.pending); 938 if (waitqueue_active(q)) { 939 apic->vcpu->arch.mp_state = VCPU_MP_STATE_RUNNABLE; 940 wake_up_interruptible(q); 941 } 942 if (apic_lvtt_period(apic)) { 943 result = 1; 944 apic->timer.dev.expires = ktime_add_ns( 945 apic->timer.dev.expires, 946 apic->timer.period); 947 } 948 return result; 949 } 950 951 static int __inject_apic_timer_irq(struct kvm_lapic *apic) 952 { 953 int vector; 954 955 vector = apic_lvt_vector(apic, APIC_LVTT); 956 return __apic_accept_irq(apic, APIC_DM_FIXED, vector, 1, 0); 957 } 958 959 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data) 960 { 961 struct kvm_lapic *apic; 962 int restart_timer = 0; 963 964 apic = container_of(data, struct kvm_lapic, timer.dev); 965 966 restart_timer = __apic_timer_fn(apic); 967 968 if (restart_timer) 969 return HRTIMER_RESTART; 970 else 971 return HRTIMER_NORESTART; 972 } 973 974 int kvm_create_lapic(struct kvm_vcpu *vcpu) 975 { 976 struct kvm_lapic *apic; 977 978 ASSERT(vcpu != NULL); 979 apic_debug("apic_init %d\n", vcpu->vcpu_id); 980 981 apic = kzalloc(sizeof(*apic), GFP_KERNEL); 982 if (!apic) 983 goto nomem; 984 985 vcpu->arch.apic = apic; 986 987 apic->regs_page = alloc_page(GFP_KERNEL); 988 if (apic->regs_page == NULL) { 989 printk(KERN_ERR "malloc apic regs error for vcpu %x\n", 990 vcpu->vcpu_id); 991 goto nomem_free_apic; 992 } 993 apic->regs = page_address(apic->regs_page); 994 memset(apic->regs, 0, PAGE_SIZE); 995 apic->vcpu = vcpu; 996 997 hrtimer_init(&apic->timer.dev, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); 998 apic->timer.dev.function = apic_timer_fn; 999 apic->base_address = APIC_DEFAULT_PHYS_BASE; 1000 vcpu->arch.apic_base = APIC_DEFAULT_PHYS_BASE; 1001 1002 kvm_lapic_reset(vcpu); 1003 apic->dev.read = apic_mmio_read; 1004 apic->dev.write = apic_mmio_write; 1005 apic->dev.in_range = apic_mmio_range; 1006 apic->dev.private = apic; 1007 1008 return 0; 1009 nomem_free_apic: 1010 kfree(apic); 1011 nomem: 1012 return -ENOMEM; 1013 } 1014 EXPORT_SYMBOL_GPL(kvm_create_lapic); 1015 1016 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu) 1017 { 1018 struct kvm_lapic *apic = vcpu->arch.apic; 1019 int highest_irr; 1020 1021 if (!apic || !apic_enabled(apic)) 1022 return -1; 1023 1024 apic_update_ppr(apic); 1025 highest_irr = apic_find_highest_irr(apic); 1026 if ((highest_irr == -1) || 1027 ((highest_irr & 0xF0) <= apic_get_reg(apic, APIC_PROCPRI))) 1028 return -1; 1029 return highest_irr; 1030 } 1031 1032 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu) 1033 { 1034 u32 lvt0 = apic_get_reg(vcpu->arch.apic, APIC_LVT0); 1035 int r = 0; 1036 1037 if (vcpu->vcpu_id == 0) { 1038 if (!apic_hw_enabled(vcpu->arch.apic)) 1039 r = 1; 1040 if ((lvt0 & APIC_LVT_MASKED) == 0 && 1041 GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT) 1042 r = 1; 1043 } 1044 return r; 1045 } 1046 1047 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu) 1048 { 1049 struct kvm_lapic *apic = vcpu->arch.apic; 1050 1051 if (apic && apic_lvt_enabled(apic, APIC_LVTT) && 1052 atomic_read(&apic->timer.pending) > 0) { 1053 if (__inject_apic_timer_irq(apic)) 1054 atomic_dec(&apic->timer.pending); 1055 } 1056 } 1057 1058 void kvm_apic_timer_intr_post(struct kvm_vcpu *vcpu, int vec) 1059 { 1060 struct kvm_lapic *apic = vcpu->arch.apic; 1061 1062 if (apic && apic_lvt_vector(apic, APIC_LVTT) == vec) 1063 apic->timer.last_update = ktime_add_ns( 1064 apic->timer.last_update, 1065 apic->timer.period); 1066 } 1067 1068 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu) 1069 { 1070 int vector = kvm_apic_has_interrupt(vcpu); 1071 struct kvm_lapic *apic = vcpu->arch.apic; 1072 1073 if (vector == -1) 1074 return -1; 1075 1076 apic_set_vector(vector, apic->regs + APIC_ISR); 1077 apic_update_ppr(apic); 1078 apic_clear_irr(vector, apic); 1079 return vector; 1080 } 1081 1082 void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu) 1083 { 1084 struct kvm_lapic *apic = vcpu->arch.apic; 1085 1086 apic->base_address = vcpu->arch.apic_base & 1087 MSR_IA32_APICBASE_BASE; 1088 apic_set_reg(apic, APIC_LVR, APIC_VERSION); 1089 apic_update_ppr(apic); 1090 hrtimer_cancel(&apic->timer.dev); 1091 update_divide_count(apic); 1092 start_apic_timer(apic); 1093 } 1094 1095 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu) 1096 { 1097 struct kvm_lapic *apic = vcpu->arch.apic; 1098 struct hrtimer *timer; 1099 1100 if (!apic) 1101 return; 1102 1103 timer = &apic->timer.dev; 1104 if (hrtimer_cancel(timer)) 1105 hrtimer_start(timer, timer->expires, HRTIMER_MODE_ABS); 1106 } 1107 1108 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu) 1109 { 1110 u32 data; 1111 void *vapic; 1112 1113 if (!irqchip_in_kernel(vcpu->kvm) || !vcpu->arch.apic->vapic_addr) 1114 return; 1115 1116 vapic = kmap_atomic(vcpu->arch.apic->vapic_page, KM_USER0); 1117 data = *(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr)); 1118 kunmap_atomic(vapic, KM_USER0); 1119 1120 apic_set_tpr(vcpu->arch.apic, data & 0xff); 1121 } 1122 1123 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu) 1124 { 1125 u32 data, tpr; 1126 int max_irr, max_isr; 1127 struct kvm_lapic *apic; 1128 void *vapic; 1129 1130 if (!irqchip_in_kernel(vcpu->kvm) || !vcpu->arch.apic->vapic_addr) 1131 return; 1132 1133 apic = vcpu->arch.apic; 1134 tpr = apic_get_reg(apic, APIC_TASKPRI) & 0xff; 1135 max_irr = apic_find_highest_irr(apic); 1136 if (max_irr < 0) 1137 max_irr = 0; 1138 max_isr = apic_find_highest_isr(apic); 1139 if (max_isr < 0) 1140 max_isr = 0; 1141 data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24); 1142 1143 vapic = kmap_atomic(vcpu->arch.apic->vapic_page, KM_USER0); 1144 *(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr)) = data; 1145 kunmap_atomic(vapic, KM_USER0); 1146 } 1147 1148 void kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr) 1149 { 1150 if (!irqchip_in_kernel(vcpu->kvm)) 1151 return; 1152 1153 vcpu->arch.apic->vapic_addr = vapic_addr; 1154 } 1155