1 2 /* 3 * Local APIC virtualization 4 * 5 * Copyright (C) 2006 Qumranet, Inc. 6 * Copyright (C) 2007 Novell 7 * Copyright (C) 2007 Intel 8 * Copyright 2009 Red Hat, Inc. and/or its affiliates. 9 * 10 * Authors: 11 * Dor Laor <dor.laor@qumranet.com> 12 * Gregory Haskins <ghaskins@novell.com> 13 * Yaozu (Eddie) Dong <eddie.dong@intel.com> 14 * 15 * Based on Xen 3.1 code, Copyright (c) 2004, Intel Corporation. 16 * 17 * This work is licensed under the terms of the GNU GPL, version 2. See 18 * the COPYING file in the top-level directory. 19 */ 20 21 #include <linux/kvm_host.h> 22 #include <linux/kvm.h> 23 #include <linux/mm.h> 24 #include <linux/highmem.h> 25 #include <linux/smp.h> 26 #include <linux/hrtimer.h> 27 #include <linux/io.h> 28 #include <linux/module.h> 29 #include <linux/math64.h> 30 #include <linux/slab.h> 31 #include <asm/processor.h> 32 #include <asm/msr.h> 33 #include <asm/page.h> 34 #include <asm/current.h> 35 #include <asm/apicdef.h> 36 #include <asm/delay.h> 37 #include <linux/atomic.h> 38 #include <linux/jump_label.h> 39 #include "kvm_cache_regs.h" 40 #include "irq.h" 41 #include "trace.h" 42 #include "x86.h" 43 #include "cpuid.h" 44 #include "hyperv.h" 45 46 #ifndef CONFIG_X86_64 47 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y)) 48 #else 49 #define mod_64(x, y) ((x) % (y)) 50 #endif 51 52 #define PRId64 "d" 53 #define PRIx64 "llx" 54 #define PRIu64 "u" 55 #define PRIo64 "o" 56 57 #define APIC_BUS_CYCLE_NS 1 58 59 /* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */ 60 #define apic_debug(fmt, arg...) 61 62 #define APIC_LVT_NUM 6 63 /* 14 is the version for Xeon and Pentium 8.4.8*/ 64 #define APIC_VERSION (0x14UL | ((APIC_LVT_NUM - 1) << 16)) 65 #define LAPIC_MMIO_LENGTH (1 << 12) 66 /* followed define is not in apicdef.h */ 67 #define APIC_SHORT_MASK 0xc0000 68 #define APIC_DEST_NOSHORT 0x0 69 #define APIC_DEST_MASK 0x800 70 #define MAX_APIC_VECTOR 256 71 #define APIC_VECTORS_PER_REG 32 72 73 #define APIC_BROADCAST 0xFF 74 #define X2APIC_BROADCAST 0xFFFFFFFFul 75 76 #define VEC_POS(v) ((v) & (32 - 1)) 77 #define REG_POS(v) (((v) >> 5) << 4) 78 79 static inline void apic_set_reg(struct kvm_lapic *apic, int reg_off, u32 val) 80 { 81 *((u32 *) (apic->regs + reg_off)) = val; 82 } 83 84 static inline int apic_test_vector(int vec, void *bitmap) 85 { 86 return test_bit(VEC_POS(vec), (bitmap) + REG_POS(vec)); 87 } 88 89 bool kvm_apic_pending_eoi(struct kvm_vcpu *vcpu, int vector) 90 { 91 struct kvm_lapic *apic = vcpu->arch.apic; 92 93 return apic_test_vector(vector, apic->regs + APIC_ISR) || 94 apic_test_vector(vector, apic->regs + APIC_IRR); 95 } 96 97 static inline void apic_set_vector(int vec, void *bitmap) 98 { 99 set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec)); 100 } 101 102 static inline void apic_clear_vector(int vec, void *bitmap) 103 { 104 clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec)); 105 } 106 107 static inline int __apic_test_and_set_vector(int vec, void *bitmap) 108 { 109 return __test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec)); 110 } 111 112 static inline int __apic_test_and_clear_vector(int vec, void *bitmap) 113 { 114 return __test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec)); 115 } 116 117 struct static_key_deferred apic_hw_disabled __read_mostly; 118 struct static_key_deferred apic_sw_disabled __read_mostly; 119 120 static inline int apic_enabled(struct kvm_lapic *apic) 121 { 122 return kvm_apic_sw_enabled(apic) && kvm_apic_hw_enabled(apic); 123 } 124 125 #define LVT_MASK \ 126 (APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK) 127 128 #define LINT_MASK \ 129 (LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \ 130 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER) 131 132 /* The logical map is definitely wrong if we have multiple 133 * modes at the same time. (Physical map is always right.) 134 */ 135 static inline bool kvm_apic_logical_map_valid(struct kvm_apic_map *map) 136 { 137 return !(map->mode & (map->mode - 1)); 138 } 139 140 static inline void 141 apic_logical_id(struct kvm_apic_map *map, u32 dest_id, u16 *cid, u16 *lid) 142 { 143 unsigned lid_bits; 144 145 BUILD_BUG_ON(KVM_APIC_MODE_XAPIC_CLUSTER != 4); 146 BUILD_BUG_ON(KVM_APIC_MODE_XAPIC_FLAT != 8); 147 BUILD_BUG_ON(KVM_APIC_MODE_X2APIC != 16); 148 lid_bits = map->mode; 149 150 *cid = dest_id >> lid_bits; 151 *lid = dest_id & ((1 << lid_bits) - 1); 152 } 153 154 static void recalculate_apic_map(struct kvm *kvm) 155 { 156 struct kvm_apic_map *new, *old = NULL; 157 struct kvm_vcpu *vcpu; 158 int i; 159 160 new = kzalloc(sizeof(struct kvm_apic_map), GFP_KERNEL); 161 162 mutex_lock(&kvm->arch.apic_map_lock); 163 164 if (!new) 165 goto out; 166 167 kvm_for_each_vcpu(i, vcpu, kvm) { 168 struct kvm_lapic *apic = vcpu->arch.apic; 169 u16 cid, lid; 170 u32 ldr, aid; 171 172 if (!kvm_apic_present(vcpu)) 173 continue; 174 175 aid = kvm_apic_id(apic); 176 ldr = kvm_apic_get_reg(apic, APIC_LDR); 177 178 if (aid < ARRAY_SIZE(new->phys_map)) 179 new->phys_map[aid] = apic; 180 181 if (apic_x2apic_mode(apic)) { 182 new->mode |= KVM_APIC_MODE_X2APIC; 183 } else if (ldr) { 184 ldr = GET_APIC_LOGICAL_ID(ldr); 185 if (kvm_apic_get_reg(apic, APIC_DFR) == APIC_DFR_FLAT) 186 new->mode |= KVM_APIC_MODE_XAPIC_FLAT; 187 else 188 new->mode |= KVM_APIC_MODE_XAPIC_CLUSTER; 189 } 190 191 if (!kvm_apic_logical_map_valid(new)) 192 continue; 193 194 apic_logical_id(new, ldr, &cid, &lid); 195 196 if (lid && cid < ARRAY_SIZE(new->logical_map)) 197 new->logical_map[cid][ffs(lid) - 1] = apic; 198 } 199 out: 200 old = rcu_dereference_protected(kvm->arch.apic_map, 201 lockdep_is_held(&kvm->arch.apic_map_lock)); 202 rcu_assign_pointer(kvm->arch.apic_map, new); 203 mutex_unlock(&kvm->arch.apic_map_lock); 204 205 if (old) 206 kfree_rcu(old, rcu); 207 208 kvm_make_scan_ioapic_request(kvm); 209 } 210 211 static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val) 212 { 213 bool enabled = val & APIC_SPIV_APIC_ENABLED; 214 215 apic_set_reg(apic, APIC_SPIV, val); 216 217 if (enabled != apic->sw_enabled) { 218 apic->sw_enabled = enabled; 219 if (enabled) { 220 static_key_slow_dec_deferred(&apic_sw_disabled); 221 recalculate_apic_map(apic->vcpu->kvm); 222 } else 223 static_key_slow_inc(&apic_sw_disabled.key); 224 } 225 } 226 227 static inline void kvm_apic_set_id(struct kvm_lapic *apic, u8 id) 228 { 229 apic_set_reg(apic, APIC_ID, id << 24); 230 recalculate_apic_map(apic->vcpu->kvm); 231 } 232 233 static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id) 234 { 235 apic_set_reg(apic, APIC_LDR, id); 236 recalculate_apic_map(apic->vcpu->kvm); 237 } 238 239 static inline void kvm_apic_set_x2apic_id(struct kvm_lapic *apic, u8 id) 240 { 241 u32 ldr = ((id >> 4) << 16) | (1 << (id & 0xf)); 242 243 apic_set_reg(apic, APIC_ID, id << 24); 244 apic_set_reg(apic, APIC_LDR, ldr); 245 recalculate_apic_map(apic->vcpu->kvm); 246 } 247 248 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type) 249 { 250 return !(kvm_apic_get_reg(apic, lvt_type) & APIC_LVT_MASKED); 251 } 252 253 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type) 254 { 255 return kvm_apic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK; 256 } 257 258 static inline int apic_lvtt_oneshot(struct kvm_lapic *apic) 259 { 260 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_ONESHOT; 261 } 262 263 static inline int apic_lvtt_period(struct kvm_lapic *apic) 264 { 265 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_PERIODIC; 266 } 267 268 static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic) 269 { 270 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_TSCDEADLINE; 271 } 272 273 static inline int apic_lvt_nmi_mode(u32 lvt_val) 274 { 275 return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI; 276 } 277 278 void kvm_apic_set_version(struct kvm_vcpu *vcpu) 279 { 280 struct kvm_lapic *apic = vcpu->arch.apic; 281 struct kvm_cpuid_entry2 *feat; 282 u32 v = APIC_VERSION; 283 284 if (!kvm_vcpu_has_lapic(vcpu)) 285 return; 286 287 feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0); 288 if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31)))) 289 v |= APIC_LVR_DIRECTED_EOI; 290 apic_set_reg(apic, APIC_LVR, v); 291 } 292 293 static const unsigned int apic_lvt_mask[APIC_LVT_NUM] = { 294 LVT_MASK , /* part LVTT mask, timer mode mask added at runtime */ 295 LVT_MASK | APIC_MODE_MASK, /* LVTTHMR */ 296 LVT_MASK | APIC_MODE_MASK, /* LVTPC */ 297 LINT_MASK, LINT_MASK, /* LVT0-1 */ 298 LVT_MASK /* LVTERR */ 299 }; 300 301 static int find_highest_vector(void *bitmap) 302 { 303 int vec; 304 u32 *reg; 305 306 for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG; 307 vec >= 0; vec -= APIC_VECTORS_PER_REG) { 308 reg = bitmap + REG_POS(vec); 309 if (*reg) 310 return fls(*reg) - 1 + vec; 311 } 312 313 return -1; 314 } 315 316 static u8 count_vectors(void *bitmap) 317 { 318 int vec; 319 u32 *reg; 320 u8 count = 0; 321 322 for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) { 323 reg = bitmap + REG_POS(vec); 324 count += hweight32(*reg); 325 } 326 327 return count; 328 } 329 330 void __kvm_apic_update_irr(u32 *pir, void *regs) 331 { 332 u32 i, pir_val; 333 334 for (i = 0; i <= 7; i++) { 335 pir_val = xchg(&pir[i], 0); 336 if (pir_val) 337 *((u32 *)(regs + APIC_IRR + i * 0x10)) |= pir_val; 338 } 339 } 340 EXPORT_SYMBOL_GPL(__kvm_apic_update_irr); 341 342 void kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir) 343 { 344 struct kvm_lapic *apic = vcpu->arch.apic; 345 346 __kvm_apic_update_irr(pir, apic->regs); 347 348 kvm_make_request(KVM_REQ_EVENT, vcpu); 349 } 350 EXPORT_SYMBOL_GPL(kvm_apic_update_irr); 351 352 static inline void apic_set_irr(int vec, struct kvm_lapic *apic) 353 { 354 apic_set_vector(vec, apic->regs + APIC_IRR); 355 /* 356 * irr_pending must be true if any interrupt is pending; set it after 357 * APIC_IRR to avoid race with apic_clear_irr 358 */ 359 apic->irr_pending = true; 360 } 361 362 static inline int apic_search_irr(struct kvm_lapic *apic) 363 { 364 return find_highest_vector(apic->regs + APIC_IRR); 365 } 366 367 static inline int apic_find_highest_irr(struct kvm_lapic *apic) 368 { 369 int result; 370 371 /* 372 * Note that irr_pending is just a hint. It will be always 373 * true with virtual interrupt delivery enabled. 374 */ 375 if (!apic->irr_pending) 376 return -1; 377 378 if (apic->vcpu->arch.apicv_active) 379 kvm_x86_ops->sync_pir_to_irr(apic->vcpu); 380 result = apic_search_irr(apic); 381 ASSERT(result == -1 || result >= 16); 382 383 return result; 384 } 385 386 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic) 387 { 388 struct kvm_vcpu *vcpu; 389 390 vcpu = apic->vcpu; 391 392 if (unlikely(vcpu->arch.apicv_active)) { 393 /* try to update RVI */ 394 apic_clear_vector(vec, apic->regs + APIC_IRR); 395 kvm_make_request(KVM_REQ_EVENT, vcpu); 396 } else { 397 apic->irr_pending = false; 398 apic_clear_vector(vec, apic->regs + APIC_IRR); 399 if (apic_search_irr(apic) != -1) 400 apic->irr_pending = true; 401 } 402 } 403 404 static inline void apic_set_isr(int vec, struct kvm_lapic *apic) 405 { 406 struct kvm_vcpu *vcpu; 407 408 if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR)) 409 return; 410 411 vcpu = apic->vcpu; 412 413 /* 414 * With APIC virtualization enabled, all caching is disabled 415 * because the processor can modify ISR under the hood. Instead 416 * just set SVI. 417 */ 418 if (unlikely(vcpu->arch.apicv_active)) 419 kvm_x86_ops->hwapic_isr_update(vcpu->kvm, vec); 420 else { 421 ++apic->isr_count; 422 BUG_ON(apic->isr_count > MAX_APIC_VECTOR); 423 /* 424 * ISR (in service register) bit is set when injecting an interrupt. 425 * The highest vector is injected. Thus the latest bit set matches 426 * the highest bit in ISR. 427 */ 428 apic->highest_isr_cache = vec; 429 } 430 } 431 432 static inline int apic_find_highest_isr(struct kvm_lapic *apic) 433 { 434 int result; 435 436 /* 437 * Note that isr_count is always 1, and highest_isr_cache 438 * is always -1, with APIC virtualization enabled. 439 */ 440 if (!apic->isr_count) 441 return -1; 442 if (likely(apic->highest_isr_cache != -1)) 443 return apic->highest_isr_cache; 444 445 result = find_highest_vector(apic->regs + APIC_ISR); 446 ASSERT(result == -1 || result >= 16); 447 448 return result; 449 } 450 451 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic) 452 { 453 struct kvm_vcpu *vcpu; 454 if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR)) 455 return; 456 457 vcpu = apic->vcpu; 458 459 /* 460 * We do get here for APIC virtualization enabled if the guest 461 * uses the Hyper-V APIC enlightenment. In this case we may need 462 * to trigger a new interrupt delivery by writing the SVI field; 463 * on the other hand isr_count and highest_isr_cache are unused 464 * and must be left alone. 465 */ 466 if (unlikely(vcpu->arch.apicv_active)) 467 kvm_x86_ops->hwapic_isr_update(vcpu->kvm, 468 apic_find_highest_isr(apic)); 469 else { 470 --apic->isr_count; 471 BUG_ON(apic->isr_count < 0); 472 apic->highest_isr_cache = -1; 473 } 474 } 475 476 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu) 477 { 478 int highest_irr; 479 480 /* This may race with setting of irr in __apic_accept_irq() and 481 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq 482 * will cause vmexit immediately and the value will be recalculated 483 * on the next vmentry. 484 */ 485 if (!kvm_vcpu_has_lapic(vcpu)) 486 return 0; 487 highest_irr = apic_find_highest_irr(vcpu->arch.apic); 488 489 return highest_irr; 490 } 491 492 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode, 493 int vector, int level, int trig_mode, 494 unsigned long *dest_map); 495 496 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq, 497 unsigned long *dest_map) 498 { 499 struct kvm_lapic *apic = vcpu->arch.apic; 500 501 return __apic_accept_irq(apic, irq->delivery_mode, irq->vector, 502 irq->level, irq->trig_mode, dest_map); 503 } 504 505 static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val) 506 { 507 508 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val, 509 sizeof(val)); 510 } 511 512 static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val) 513 { 514 515 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val, 516 sizeof(*val)); 517 } 518 519 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu) 520 { 521 return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED; 522 } 523 524 static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu) 525 { 526 u8 val; 527 if (pv_eoi_get_user(vcpu, &val) < 0) 528 apic_debug("Can't read EOI MSR value: 0x%llx\n", 529 (unsigned long long)vcpu->arch.pv_eoi.msr_val); 530 return val & 0x1; 531 } 532 533 static void pv_eoi_set_pending(struct kvm_vcpu *vcpu) 534 { 535 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) { 536 apic_debug("Can't set EOI MSR value: 0x%llx\n", 537 (unsigned long long)vcpu->arch.pv_eoi.msr_val); 538 return; 539 } 540 __set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention); 541 } 542 543 static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu) 544 { 545 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) { 546 apic_debug("Can't clear EOI MSR value: 0x%llx\n", 547 (unsigned long long)vcpu->arch.pv_eoi.msr_val); 548 return; 549 } 550 __clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention); 551 } 552 553 static void apic_update_ppr(struct kvm_lapic *apic) 554 { 555 u32 tpr, isrv, ppr, old_ppr; 556 int isr; 557 558 old_ppr = kvm_apic_get_reg(apic, APIC_PROCPRI); 559 tpr = kvm_apic_get_reg(apic, APIC_TASKPRI); 560 isr = apic_find_highest_isr(apic); 561 isrv = (isr != -1) ? isr : 0; 562 563 if ((tpr & 0xf0) >= (isrv & 0xf0)) 564 ppr = tpr & 0xff; 565 else 566 ppr = isrv & 0xf0; 567 568 apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x", 569 apic, ppr, isr, isrv); 570 571 if (old_ppr != ppr) { 572 apic_set_reg(apic, APIC_PROCPRI, ppr); 573 if (ppr < old_ppr) 574 kvm_make_request(KVM_REQ_EVENT, apic->vcpu); 575 } 576 } 577 578 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr) 579 { 580 apic_set_reg(apic, APIC_TASKPRI, tpr); 581 apic_update_ppr(apic); 582 } 583 584 static bool kvm_apic_broadcast(struct kvm_lapic *apic, u32 mda) 585 { 586 if (apic_x2apic_mode(apic)) 587 return mda == X2APIC_BROADCAST; 588 589 return GET_APIC_DEST_FIELD(mda) == APIC_BROADCAST; 590 } 591 592 static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 mda) 593 { 594 if (kvm_apic_broadcast(apic, mda)) 595 return true; 596 597 if (apic_x2apic_mode(apic)) 598 return mda == kvm_apic_id(apic); 599 600 return mda == SET_APIC_DEST_FIELD(kvm_apic_id(apic)); 601 } 602 603 static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda) 604 { 605 u32 logical_id; 606 607 if (kvm_apic_broadcast(apic, mda)) 608 return true; 609 610 logical_id = kvm_apic_get_reg(apic, APIC_LDR); 611 612 if (apic_x2apic_mode(apic)) 613 return ((logical_id >> 16) == (mda >> 16)) 614 && (logical_id & mda & 0xffff) != 0; 615 616 logical_id = GET_APIC_LOGICAL_ID(logical_id); 617 mda = GET_APIC_DEST_FIELD(mda); 618 619 switch (kvm_apic_get_reg(apic, APIC_DFR)) { 620 case APIC_DFR_FLAT: 621 return (logical_id & mda) != 0; 622 case APIC_DFR_CLUSTER: 623 return ((logical_id >> 4) == (mda >> 4)) 624 && (logical_id & mda & 0xf) != 0; 625 default: 626 apic_debug("Bad DFR vcpu %d: %08x\n", 627 apic->vcpu->vcpu_id, kvm_apic_get_reg(apic, APIC_DFR)); 628 return false; 629 } 630 } 631 632 /* KVM APIC implementation has two quirks 633 * - dest always begins at 0 while xAPIC MDA has offset 24, 634 * - IOxAPIC messages have to be delivered (directly) to x2APIC. 635 */ 636 static u32 kvm_apic_mda(unsigned int dest_id, struct kvm_lapic *source, 637 struct kvm_lapic *target) 638 { 639 bool ipi = source != NULL; 640 bool x2apic_mda = apic_x2apic_mode(ipi ? source : target); 641 642 if (!ipi && dest_id == APIC_BROADCAST && x2apic_mda) 643 return X2APIC_BROADCAST; 644 645 return x2apic_mda ? dest_id : SET_APIC_DEST_FIELD(dest_id); 646 } 647 648 bool kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source, 649 int short_hand, unsigned int dest, int dest_mode) 650 { 651 struct kvm_lapic *target = vcpu->arch.apic; 652 u32 mda = kvm_apic_mda(dest, source, target); 653 654 apic_debug("target %p, source %p, dest 0x%x, " 655 "dest_mode 0x%x, short_hand 0x%x\n", 656 target, source, dest, dest_mode, short_hand); 657 658 ASSERT(target); 659 switch (short_hand) { 660 case APIC_DEST_NOSHORT: 661 if (dest_mode == APIC_DEST_PHYSICAL) 662 return kvm_apic_match_physical_addr(target, mda); 663 else 664 return kvm_apic_match_logical_addr(target, mda); 665 case APIC_DEST_SELF: 666 return target == source; 667 case APIC_DEST_ALLINC: 668 return true; 669 case APIC_DEST_ALLBUT: 670 return target != source; 671 default: 672 apic_debug("kvm: apic: Bad dest shorthand value %x\n", 673 short_hand); 674 return false; 675 } 676 } 677 678 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src, 679 struct kvm_lapic_irq *irq, int *r, unsigned long *dest_map) 680 { 681 struct kvm_apic_map *map; 682 unsigned long bitmap = 1; 683 struct kvm_lapic **dst; 684 int i; 685 bool ret, x2apic_ipi; 686 687 *r = -1; 688 689 if (irq->shorthand == APIC_DEST_SELF) { 690 *r = kvm_apic_set_irq(src->vcpu, irq, dest_map); 691 return true; 692 } 693 694 if (irq->shorthand) 695 return false; 696 697 x2apic_ipi = src && apic_x2apic_mode(src); 698 if (irq->dest_id == (x2apic_ipi ? X2APIC_BROADCAST : APIC_BROADCAST)) 699 return false; 700 701 ret = true; 702 rcu_read_lock(); 703 map = rcu_dereference(kvm->arch.apic_map); 704 705 if (!map) { 706 ret = false; 707 goto out; 708 } 709 710 if (irq->dest_mode == APIC_DEST_PHYSICAL) { 711 if (irq->dest_id >= ARRAY_SIZE(map->phys_map)) 712 goto out; 713 714 dst = &map->phys_map[irq->dest_id]; 715 } else { 716 u16 cid; 717 718 if (!kvm_apic_logical_map_valid(map)) { 719 ret = false; 720 goto out; 721 } 722 723 apic_logical_id(map, irq->dest_id, &cid, (u16 *)&bitmap); 724 725 if (cid >= ARRAY_SIZE(map->logical_map)) 726 goto out; 727 728 dst = map->logical_map[cid]; 729 730 if (kvm_lowest_prio_delivery(irq)) { 731 int l = -1; 732 for_each_set_bit(i, &bitmap, 16) { 733 if (!dst[i]) 734 continue; 735 if (l < 0) 736 l = i; 737 else if (kvm_apic_compare_prio(dst[i]->vcpu, dst[l]->vcpu) < 0) 738 l = i; 739 } 740 741 bitmap = (l >= 0) ? 1 << l : 0; 742 } 743 } 744 745 for_each_set_bit(i, &bitmap, 16) { 746 if (!dst[i]) 747 continue; 748 if (*r < 0) 749 *r = 0; 750 *r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map); 751 } 752 out: 753 rcu_read_unlock(); 754 return ret; 755 } 756 757 bool kvm_intr_is_single_vcpu_fast(struct kvm *kvm, struct kvm_lapic_irq *irq, 758 struct kvm_vcpu **dest_vcpu) 759 { 760 struct kvm_apic_map *map; 761 bool ret = false; 762 struct kvm_lapic *dst = NULL; 763 764 if (irq->shorthand) 765 return false; 766 767 rcu_read_lock(); 768 map = rcu_dereference(kvm->arch.apic_map); 769 770 if (!map) 771 goto out; 772 773 if (irq->dest_mode == APIC_DEST_PHYSICAL) { 774 if (irq->dest_id == 0xFF) 775 goto out; 776 777 if (irq->dest_id >= ARRAY_SIZE(map->phys_map)) 778 goto out; 779 780 dst = map->phys_map[irq->dest_id]; 781 if (dst && kvm_apic_present(dst->vcpu)) 782 *dest_vcpu = dst->vcpu; 783 else 784 goto out; 785 } else { 786 u16 cid; 787 unsigned long bitmap = 1; 788 int i, r = 0; 789 790 if (!kvm_apic_logical_map_valid(map)) 791 goto out; 792 793 apic_logical_id(map, irq->dest_id, &cid, (u16 *)&bitmap); 794 795 if (cid >= ARRAY_SIZE(map->logical_map)) 796 goto out; 797 798 for_each_set_bit(i, &bitmap, 16) { 799 dst = map->logical_map[cid][i]; 800 if (++r == 2) 801 goto out; 802 } 803 804 if (dst && kvm_apic_present(dst->vcpu)) 805 *dest_vcpu = dst->vcpu; 806 else 807 goto out; 808 } 809 810 ret = true; 811 out: 812 rcu_read_unlock(); 813 return ret; 814 } 815 816 /* 817 * Add a pending IRQ into lapic. 818 * Return 1 if successfully added and 0 if discarded. 819 */ 820 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode, 821 int vector, int level, int trig_mode, 822 unsigned long *dest_map) 823 { 824 int result = 0; 825 struct kvm_vcpu *vcpu = apic->vcpu; 826 827 trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode, 828 trig_mode, vector); 829 switch (delivery_mode) { 830 case APIC_DM_LOWEST: 831 vcpu->arch.apic_arb_prio++; 832 case APIC_DM_FIXED: 833 if (unlikely(trig_mode && !level)) 834 break; 835 836 /* FIXME add logic for vcpu on reset */ 837 if (unlikely(!apic_enabled(apic))) 838 break; 839 840 result = 1; 841 842 if (dest_map) 843 __set_bit(vcpu->vcpu_id, dest_map); 844 845 if (apic_test_vector(vector, apic->regs + APIC_TMR) != !!trig_mode) { 846 if (trig_mode) 847 apic_set_vector(vector, apic->regs + APIC_TMR); 848 else 849 apic_clear_vector(vector, apic->regs + APIC_TMR); 850 } 851 852 if (vcpu->arch.apicv_active) 853 kvm_x86_ops->deliver_posted_interrupt(vcpu, vector); 854 else { 855 apic_set_irr(vector, apic); 856 857 kvm_make_request(KVM_REQ_EVENT, vcpu); 858 kvm_vcpu_kick(vcpu); 859 } 860 break; 861 862 case APIC_DM_REMRD: 863 result = 1; 864 vcpu->arch.pv.pv_unhalted = 1; 865 kvm_make_request(KVM_REQ_EVENT, vcpu); 866 kvm_vcpu_kick(vcpu); 867 break; 868 869 case APIC_DM_SMI: 870 result = 1; 871 kvm_make_request(KVM_REQ_SMI, vcpu); 872 kvm_vcpu_kick(vcpu); 873 break; 874 875 case APIC_DM_NMI: 876 result = 1; 877 kvm_inject_nmi(vcpu); 878 kvm_vcpu_kick(vcpu); 879 break; 880 881 case APIC_DM_INIT: 882 if (!trig_mode || level) { 883 result = 1; 884 /* assumes that there are only KVM_APIC_INIT/SIPI */ 885 apic->pending_events = (1UL << KVM_APIC_INIT); 886 /* make sure pending_events is visible before sending 887 * the request */ 888 smp_wmb(); 889 kvm_make_request(KVM_REQ_EVENT, vcpu); 890 kvm_vcpu_kick(vcpu); 891 } else { 892 apic_debug("Ignoring de-assert INIT to vcpu %d\n", 893 vcpu->vcpu_id); 894 } 895 break; 896 897 case APIC_DM_STARTUP: 898 apic_debug("SIPI to vcpu %d vector 0x%02x\n", 899 vcpu->vcpu_id, vector); 900 result = 1; 901 apic->sipi_vector = vector; 902 /* make sure sipi_vector is visible for the receiver */ 903 smp_wmb(); 904 set_bit(KVM_APIC_SIPI, &apic->pending_events); 905 kvm_make_request(KVM_REQ_EVENT, vcpu); 906 kvm_vcpu_kick(vcpu); 907 break; 908 909 case APIC_DM_EXTINT: 910 /* 911 * Should only be called by kvm_apic_local_deliver() with LVT0, 912 * before NMI watchdog was enabled. Already handled by 913 * kvm_apic_accept_pic_intr(). 914 */ 915 break; 916 917 default: 918 printk(KERN_ERR "TODO: unsupported delivery mode %x\n", 919 delivery_mode); 920 break; 921 } 922 return result; 923 } 924 925 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2) 926 { 927 return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio; 928 } 929 930 static bool kvm_ioapic_handles_vector(struct kvm_lapic *apic, int vector) 931 { 932 return test_bit(vector, apic->vcpu->arch.ioapic_handled_vectors); 933 } 934 935 static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector) 936 { 937 int trigger_mode; 938 939 /* Eoi the ioapic only if the ioapic doesn't own the vector. */ 940 if (!kvm_ioapic_handles_vector(apic, vector)) 941 return; 942 943 /* Request a KVM exit to inform the userspace IOAPIC. */ 944 if (irqchip_split(apic->vcpu->kvm)) { 945 apic->vcpu->arch.pending_ioapic_eoi = vector; 946 kvm_make_request(KVM_REQ_IOAPIC_EOI_EXIT, apic->vcpu); 947 return; 948 } 949 950 if (apic_test_vector(vector, apic->regs + APIC_TMR)) 951 trigger_mode = IOAPIC_LEVEL_TRIG; 952 else 953 trigger_mode = IOAPIC_EDGE_TRIG; 954 955 kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode); 956 } 957 958 static int apic_set_eoi(struct kvm_lapic *apic) 959 { 960 int vector = apic_find_highest_isr(apic); 961 962 trace_kvm_eoi(apic, vector); 963 964 /* 965 * Not every write EOI will has corresponding ISR, 966 * one example is when Kernel check timer on setup_IO_APIC 967 */ 968 if (vector == -1) 969 return vector; 970 971 apic_clear_isr(vector, apic); 972 apic_update_ppr(apic); 973 974 if (test_bit(vector, vcpu_to_synic(apic->vcpu)->vec_bitmap)) 975 kvm_hv_synic_send_eoi(apic->vcpu, vector); 976 977 kvm_ioapic_send_eoi(apic, vector); 978 kvm_make_request(KVM_REQ_EVENT, apic->vcpu); 979 return vector; 980 } 981 982 /* 983 * this interface assumes a trap-like exit, which has already finished 984 * desired side effect including vISR and vPPR update. 985 */ 986 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector) 987 { 988 struct kvm_lapic *apic = vcpu->arch.apic; 989 990 trace_kvm_eoi(apic, vector); 991 992 kvm_ioapic_send_eoi(apic, vector); 993 kvm_make_request(KVM_REQ_EVENT, apic->vcpu); 994 } 995 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated); 996 997 static void apic_send_ipi(struct kvm_lapic *apic) 998 { 999 u32 icr_low = kvm_apic_get_reg(apic, APIC_ICR); 1000 u32 icr_high = kvm_apic_get_reg(apic, APIC_ICR2); 1001 struct kvm_lapic_irq irq; 1002 1003 irq.vector = icr_low & APIC_VECTOR_MASK; 1004 irq.delivery_mode = icr_low & APIC_MODE_MASK; 1005 irq.dest_mode = icr_low & APIC_DEST_MASK; 1006 irq.level = (icr_low & APIC_INT_ASSERT) != 0; 1007 irq.trig_mode = icr_low & APIC_INT_LEVELTRIG; 1008 irq.shorthand = icr_low & APIC_SHORT_MASK; 1009 irq.msi_redir_hint = false; 1010 if (apic_x2apic_mode(apic)) 1011 irq.dest_id = icr_high; 1012 else 1013 irq.dest_id = GET_APIC_DEST_FIELD(icr_high); 1014 1015 trace_kvm_apic_ipi(icr_low, irq.dest_id); 1016 1017 apic_debug("icr_high 0x%x, icr_low 0x%x, " 1018 "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, " 1019 "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x, " 1020 "msi_redir_hint 0x%x\n", 1021 icr_high, icr_low, irq.shorthand, irq.dest_id, 1022 irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode, 1023 irq.vector, irq.msi_redir_hint); 1024 1025 kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL); 1026 } 1027 1028 static u32 apic_get_tmcct(struct kvm_lapic *apic) 1029 { 1030 ktime_t remaining; 1031 s64 ns; 1032 u32 tmcct; 1033 1034 ASSERT(apic != NULL); 1035 1036 /* if initial count is 0, current count should also be 0 */ 1037 if (kvm_apic_get_reg(apic, APIC_TMICT) == 0 || 1038 apic->lapic_timer.period == 0) 1039 return 0; 1040 1041 remaining = hrtimer_get_remaining(&apic->lapic_timer.timer); 1042 if (ktime_to_ns(remaining) < 0) 1043 remaining = ktime_set(0, 0); 1044 1045 ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period); 1046 tmcct = div64_u64(ns, 1047 (APIC_BUS_CYCLE_NS * apic->divide_count)); 1048 1049 return tmcct; 1050 } 1051 1052 static void __report_tpr_access(struct kvm_lapic *apic, bool write) 1053 { 1054 struct kvm_vcpu *vcpu = apic->vcpu; 1055 struct kvm_run *run = vcpu->run; 1056 1057 kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu); 1058 run->tpr_access.rip = kvm_rip_read(vcpu); 1059 run->tpr_access.is_write = write; 1060 } 1061 1062 static inline void report_tpr_access(struct kvm_lapic *apic, bool write) 1063 { 1064 if (apic->vcpu->arch.tpr_access_reporting) 1065 __report_tpr_access(apic, write); 1066 } 1067 1068 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset) 1069 { 1070 u32 val = 0; 1071 1072 if (offset >= LAPIC_MMIO_LENGTH) 1073 return 0; 1074 1075 switch (offset) { 1076 case APIC_ID: 1077 if (apic_x2apic_mode(apic)) 1078 val = kvm_apic_id(apic); 1079 else 1080 val = kvm_apic_id(apic) << 24; 1081 break; 1082 case APIC_ARBPRI: 1083 apic_debug("Access APIC ARBPRI register which is for P6\n"); 1084 break; 1085 1086 case APIC_TMCCT: /* Timer CCR */ 1087 if (apic_lvtt_tscdeadline(apic)) 1088 return 0; 1089 1090 val = apic_get_tmcct(apic); 1091 break; 1092 case APIC_PROCPRI: 1093 apic_update_ppr(apic); 1094 val = kvm_apic_get_reg(apic, offset); 1095 break; 1096 case APIC_TASKPRI: 1097 report_tpr_access(apic, false); 1098 /* fall thru */ 1099 default: 1100 val = kvm_apic_get_reg(apic, offset); 1101 break; 1102 } 1103 1104 return val; 1105 } 1106 1107 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev) 1108 { 1109 return container_of(dev, struct kvm_lapic, dev); 1110 } 1111 1112 static int apic_reg_read(struct kvm_lapic *apic, u32 offset, int len, 1113 void *data) 1114 { 1115 unsigned char alignment = offset & 0xf; 1116 u32 result; 1117 /* this bitmask has a bit cleared for each reserved register */ 1118 static const u64 rmask = 0x43ff01ffffffe70cULL; 1119 1120 if ((alignment + len) > 4) { 1121 apic_debug("KVM_APIC_READ: alignment error %x %d\n", 1122 offset, len); 1123 return 1; 1124 } 1125 1126 if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) { 1127 apic_debug("KVM_APIC_READ: read reserved register %x\n", 1128 offset); 1129 return 1; 1130 } 1131 1132 result = __apic_read(apic, offset & ~0xf); 1133 1134 trace_kvm_apic_read(offset, result); 1135 1136 switch (len) { 1137 case 1: 1138 case 2: 1139 case 4: 1140 memcpy(data, (char *)&result + alignment, len); 1141 break; 1142 default: 1143 printk(KERN_ERR "Local APIC read with len = %x, " 1144 "should be 1,2, or 4 instead\n", len); 1145 break; 1146 } 1147 return 0; 1148 } 1149 1150 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr) 1151 { 1152 return kvm_apic_hw_enabled(apic) && 1153 addr >= apic->base_address && 1154 addr < apic->base_address + LAPIC_MMIO_LENGTH; 1155 } 1156 1157 static int apic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this, 1158 gpa_t address, int len, void *data) 1159 { 1160 struct kvm_lapic *apic = to_lapic(this); 1161 u32 offset = address - apic->base_address; 1162 1163 if (!apic_mmio_in_range(apic, address)) 1164 return -EOPNOTSUPP; 1165 1166 apic_reg_read(apic, offset, len, data); 1167 1168 return 0; 1169 } 1170 1171 static void update_divide_count(struct kvm_lapic *apic) 1172 { 1173 u32 tmp1, tmp2, tdcr; 1174 1175 tdcr = kvm_apic_get_reg(apic, APIC_TDCR); 1176 tmp1 = tdcr & 0xf; 1177 tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1; 1178 apic->divide_count = 0x1 << (tmp2 & 0x7); 1179 1180 apic_debug("timer divide count is 0x%x\n", 1181 apic->divide_count); 1182 } 1183 1184 static void apic_update_lvtt(struct kvm_lapic *apic) 1185 { 1186 u32 timer_mode = kvm_apic_get_reg(apic, APIC_LVTT) & 1187 apic->lapic_timer.timer_mode_mask; 1188 1189 if (apic->lapic_timer.timer_mode != timer_mode) { 1190 apic->lapic_timer.timer_mode = timer_mode; 1191 hrtimer_cancel(&apic->lapic_timer.timer); 1192 } 1193 } 1194 1195 static void apic_timer_expired(struct kvm_lapic *apic) 1196 { 1197 struct kvm_vcpu *vcpu = apic->vcpu; 1198 wait_queue_head_t *q = &vcpu->wq; 1199 struct kvm_timer *ktimer = &apic->lapic_timer; 1200 1201 if (atomic_read(&apic->lapic_timer.pending)) 1202 return; 1203 1204 atomic_inc(&apic->lapic_timer.pending); 1205 kvm_set_pending_timer(vcpu); 1206 1207 if (waitqueue_active(q)) 1208 wake_up_interruptible(q); 1209 1210 if (apic_lvtt_tscdeadline(apic)) 1211 ktimer->expired_tscdeadline = ktimer->tscdeadline; 1212 } 1213 1214 /* 1215 * On APICv, this test will cause a busy wait 1216 * during a higher-priority task. 1217 */ 1218 1219 static bool lapic_timer_int_injected(struct kvm_vcpu *vcpu) 1220 { 1221 struct kvm_lapic *apic = vcpu->arch.apic; 1222 u32 reg = kvm_apic_get_reg(apic, APIC_LVTT); 1223 1224 if (kvm_apic_hw_enabled(apic)) { 1225 int vec = reg & APIC_VECTOR_MASK; 1226 void *bitmap = apic->regs + APIC_ISR; 1227 1228 if (vcpu->arch.apicv_active) 1229 bitmap = apic->regs + APIC_IRR; 1230 1231 if (apic_test_vector(vec, bitmap)) 1232 return true; 1233 } 1234 return false; 1235 } 1236 1237 void wait_lapic_expire(struct kvm_vcpu *vcpu) 1238 { 1239 struct kvm_lapic *apic = vcpu->arch.apic; 1240 u64 guest_tsc, tsc_deadline; 1241 1242 if (!kvm_vcpu_has_lapic(vcpu)) 1243 return; 1244 1245 if (apic->lapic_timer.expired_tscdeadline == 0) 1246 return; 1247 1248 if (!lapic_timer_int_injected(vcpu)) 1249 return; 1250 1251 tsc_deadline = apic->lapic_timer.expired_tscdeadline; 1252 apic->lapic_timer.expired_tscdeadline = 0; 1253 guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc()); 1254 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, guest_tsc - tsc_deadline); 1255 1256 /* __delay is delay_tsc whenever the hardware has TSC, thus always. */ 1257 if (guest_tsc < tsc_deadline) 1258 __delay(tsc_deadline - guest_tsc); 1259 } 1260 1261 static void start_apic_timer(struct kvm_lapic *apic) 1262 { 1263 ktime_t now; 1264 1265 atomic_set(&apic->lapic_timer.pending, 0); 1266 1267 if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) { 1268 /* lapic timer in oneshot or periodic mode */ 1269 now = apic->lapic_timer.timer.base->get_time(); 1270 apic->lapic_timer.period = (u64)kvm_apic_get_reg(apic, APIC_TMICT) 1271 * APIC_BUS_CYCLE_NS * apic->divide_count; 1272 1273 if (!apic->lapic_timer.period) 1274 return; 1275 /* 1276 * Do not allow the guest to program periodic timers with small 1277 * interval, since the hrtimers are not throttled by the host 1278 * scheduler. 1279 */ 1280 if (apic_lvtt_period(apic)) { 1281 s64 min_period = min_timer_period_us * 1000LL; 1282 1283 if (apic->lapic_timer.period < min_period) { 1284 pr_info_ratelimited( 1285 "kvm: vcpu %i: requested %lld ns " 1286 "lapic timer period limited to %lld ns\n", 1287 apic->vcpu->vcpu_id, 1288 apic->lapic_timer.period, min_period); 1289 apic->lapic_timer.period = min_period; 1290 } 1291 } 1292 1293 hrtimer_start(&apic->lapic_timer.timer, 1294 ktime_add_ns(now, apic->lapic_timer.period), 1295 HRTIMER_MODE_ABS); 1296 1297 apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016" 1298 PRIx64 ", " 1299 "timer initial count 0x%x, period %lldns, " 1300 "expire @ 0x%016" PRIx64 ".\n", __func__, 1301 APIC_BUS_CYCLE_NS, ktime_to_ns(now), 1302 kvm_apic_get_reg(apic, APIC_TMICT), 1303 apic->lapic_timer.period, 1304 ktime_to_ns(ktime_add_ns(now, 1305 apic->lapic_timer.period))); 1306 } else if (apic_lvtt_tscdeadline(apic)) { 1307 /* lapic timer in tsc deadline mode */ 1308 u64 guest_tsc, tscdeadline = apic->lapic_timer.tscdeadline; 1309 u64 ns = 0; 1310 ktime_t expire; 1311 struct kvm_vcpu *vcpu = apic->vcpu; 1312 unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz; 1313 unsigned long flags; 1314 1315 if (unlikely(!tscdeadline || !this_tsc_khz)) 1316 return; 1317 1318 local_irq_save(flags); 1319 1320 now = apic->lapic_timer.timer.base->get_time(); 1321 guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc()); 1322 if (likely(tscdeadline > guest_tsc)) { 1323 ns = (tscdeadline - guest_tsc) * 1000000ULL; 1324 do_div(ns, this_tsc_khz); 1325 expire = ktime_add_ns(now, ns); 1326 expire = ktime_sub_ns(expire, lapic_timer_advance_ns); 1327 hrtimer_start(&apic->lapic_timer.timer, 1328 expire, HRTIMER_MODE_ABS); 1329 } else 1330 apic_timer_expired(apic); 1331 1332 local_irq_restore(flags); 1333 } 1334 } 1335 1336 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val) 1337 { 1338 bool lvt0_in_nmi_mode = apic_lvt_nmi_mode(lvt0_val); 1339 1340 if (apic->lvt0_in_nmi_mode != lvt0_in_nmi_mode) { 1341 apic->lvt0_in_nmi_mode = lvt0_in_nmi_mode; 1342 if (lvt0_in_nmi_mode) { 1343 apic_debug("Receive NMI setting on APIC_LVT0 " 1344 "for cpu %d\n", apic->vcpu->vcpu_id); 1345 atomic_inc(&apic->vcpu->kvm->arch.vapics_in_nmi_mode); 1346 } else 1347 atomic_dec(&apic->vcpu->kvm->arch.vapics_in_nmi_mode); 1348 } 1349 } 1350 1351 static int apic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val) 1352 { 1353 int ret = 0; 1354 1355 trace_kvm_apic_write(reg, val); 1356 1357 switch (reg) { 1358 case APIC_ID: /* Local APIC ID */ 1359 if (!apic_x2apic_mode(apic)) 1360 kvm_apic_set_id(apic, val >> 24); 1361 else 1362 ret = 1; 1363 break; 1364 1365 case APIC_TASKPRI: 1366 report_tpr_access(apic, true); 1367 apic_set_tpr(apic, val & 0xff); 1368 break; 1369 1370 case APIC_EOI: 1371 apic_set_eoi(apic); 1372 break; 1373 1374 case APIC_LDR: 1375 if (!apic_x2apic_mode(apic)) 1376 kvm_apic_set_ldr(apic, val & APIC_LDR_MASK); 1377 else 1378 ret = 1; 1379 break; 1380 1381 case APIC_DFR: 1382 if (!apic_x2apic_mode(apic)) { 1383 apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF); 1384 recalculate_apic_map(apic->vcpu->kvm); 1385 } else 1386 ret = 1; 1387 break; 1388 1389 case APIC_SPIV: { 1390 u32 mask = 0x3ff; 1391 if (kvm_apic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI) 1392 mask |= APIC_SPIV_DIRECTED_EOI; 1393 apic_set_spiv(apic, val & mask); 1394 if (!(val & APIC_SPIV_APIC_ENABLED)) { 1395 int i; 1396 u32 lvt_val; 1397 1398 for (i = 0; i < APIC_LVT_NUM; i++) { 1399 lvt_val = kvm_apic_get_reg(apic, 1400 APIC_LVTT + 0x10 * i); 1401 apic_set_reg(apic, APIC_LVTT + 0x10 * i, 1402 lvt_val | APIC_LVT_MASKED); 1403 } 1404 apic_update_lvtt(apic); 1405 atomic_set(&apic->lapic_timer.pending, 0); 1406 1407 } 1408 break; 1409 } 1410 case APIC_ICR: 1411 /* No delay here, so we always clear the pending bit */ 1412 apic_set_reg(apic, APIC_ICR, val & ~(1 << 12)); 1413 apic_send_ipi(apic); 1414 break; 1415 1416 case APIC_ICR2: 1417 if (!apic_x2apic_mode(apic)) 1418 val &= 0xff000000; 1419 apic_set_reg(apic, APIC_ICR2, val); 1420 break; 1421 1422 case APIC_LVT0: 1423 apic_manage_nmi_watchdog(apic, val); 1424 case APIC_LVTTHMR: 1425 case APIC_LVTPC: 1426 case APIC_LVT1: 1427 case APIC_LVTERR: 1428 /* TODO: Check vector */ 1429 if (!kvm_apic_sw_enabled(apic)) 1430 val |= APIC_LVT_MASKED; 1431 1432 val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4]; 1433 apic_set_reg(apic, reg, val); 1434 1435 break; 1436 1437 case APIC_LVTT: 1438 if (!kvm_apic_sw_enabled(apic)) 1439 val |= APIC_LVT_MASKED; 1440 val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask); 1441 apic_set_reg(apic, APIC_LVTT, val); 1442 apic_update_lvtt(apic); 1443 break; 1444 1445 case APIC_TMICT: 1446 if (apic_lvtt_tscdeadline(apic)) 1447 break; 1448 1449 hrtimer_cancel(&apic->lapic_timer.timer); 1450 apic_set_reg(apic, APIC_TMICT, val); 1451 start_apic_timer(apic); 1452 break; 1453 1454 case APIC_TDCR: 1455 if (val & 4) 1456 apic_debug("KVM_WRITE:TDCR %x\n", val); 1457 apic_set_reg(apic, APIC_TDCR, val); 1458 update_divide_count(apic); 1459 break; 1460 1461 case APIC_ESR: 1462 if (apic_x2apic_mode(apic) && val != 0) { 1463 apic_debug("KVM_WRITE:ESR not zero %x\n", val); 1464 ret = 1; 1465 } 1466 break; 1467 1468 case APIC_SELF_IPI: 1469 if (apic_x2apic_mode(apic)) { 1470 apic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff)); 1471 } else 1472 ret = 1; 1473 break; 1474 default: 1475 ret = 1; 1476 break; 1477 } 1478 if (ret) 1479 apic_debug("Local APIC Write to read-only register %x\n", reg); 1480 return ret; 1481 } 1482 1483 static int apic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this, 1484 gpa_t address, int len, const void *data) 1485 { 1486 struct kvm_lapic *apic = to_lapic(this); 1487 unsigned int offset = address - apic->base_address; 1488 u32 val; 1489 1490 if (!apic_mmio_in_range(apic, address)) 1491 return -EOPNOTSUPP; 1492 1493 /* 1494 * APIC register must be aligned on 128-bits boundary. 1495 * 32/64/128 bits registers must be accessed thru 32 bits. 1496 * Refer SDM 8.4.1 1497 */ 1498 if (len != 4 || (offset & 0xf)) { 1499 /* Don't shout loud, $infamous_os would cause only noise. */ 1500 apic_debug("apic write: bad size=%d %lx\n", len, (long)address); 1501 return 0; 1502 } 1503 1504 val = *(u32*)data; 1505 1506 /* too common printing */ 1507 if (offset != APIC_EOI) 1508 apic_debug("%s: offset 0x%x with length 0x%x, and value is " 1509 "0x%x\n", __func__, offset, len, val); 1510 1511 apic_reg_write(apic, offset & 0xff0, val); 1512 1513 return 0; 1514 } 1515 1516 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu) 1517 { 1518 if (kvm_vcpu_has_lapic(vcpu)) 1519 apic_reg_write(vcpu->arch.apic, APIC_EOI, 0); 1520 } 1521 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi); 1522 1523 /* emulate APIC access in a trap manner */ 1524 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset) 1525 { 1526 u32 val = 0; 1527 1528 /* hw has done the conditional check and inst decode */ 1529 offset &= 0xff0; 1530 1531 apic_reg_read(vcpu->arch.apic, offset, 4, &val); 1532 1533 /* TODO: optimize to just emulate side effect w/o one more write */ 1534 apic_reg_write(vcpu->arch.apic, offset, val); 1535 } 1536 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode); 1537 1538 void kvm_free_lapic(struct kvm_vcpu *vcpu) 1539 { 1540 struct kvm_lapic *apic = vcpu->arch.apic; 1541 1542 if (!vcpu->arch.apic) 1543 return; 1544 1545 hrtimer_cancel(&apic->lapic_timer.timer); 1546 1547 if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE)) 1548 static_key_slow_dec_deferred(&apic_hw_disabled); 1549 1550 if (!apic->sw_enabled) 1551 static_key_slow_dec_deferred(&apic_sw_disabled); 1552 1553 if (apic->regs) 1554 free_page((unsigned long)apic->regs); 1555 1556 kfree(apic); 1557 } 1558 1559 /* 1560 *---------------------------------------------------------------------- 1561 * LAPIC interface 1562 *---------------------------------------------------------------------- 1563 */ 1564 1565 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu) 1566 { 1567 struct kvm_lapic *apic = vcpu->arch.apic; 1568 1569 if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) || 1570 apic_lvtt_period(apic)) 1571 return 0; 1572 1573 return apic->lapic_timer.tscdeadline; 1574 } 1575 1576 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data) 1577 { 1578 struct kvm_lapic *apic = vcpu->arch.apic; 1579 1580 if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) || 1581 apic_lvtt_period(apic)) 1582 return; 1583 1584 hrtimer_cancel(&apic->lapic_timer.timer); 1585 apic->lapic_timer.tscdeadline = data; 1586 start_apic_timer(apic); 1587 } 1588 1589 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8) 1590 { 1591 struct kvm_lapic *apic = vcpu->arch.apic; 1592 1593 if (!kvm_vcpu_has_lapic(vcpu)) 1594 return; 1595 1596 apic_set_tpr(apic, ((cr8 & 0x0f) << 4) 1597 | (kvm_apic_get_reg(apic, APIC_TASKPRI) & 4)); 1598 } 1599 1600 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu) 1601 { 1602 u64 tpr; 1603 1604 if (!kvm_vcpu_has_lapic(vcpu)) 1605 return 0; 1606 1607 tpr = (u64) kvm_apic_get_reg(vcpu->arch.apic, APIC_TASKPRI); 1608 1609 return (tpr & 0xf0) >> 4; 1610 } 1611 1612 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value) 1613 { 1614 u64 old_value = vcpu->arch.apic_base; 1615 struct kvm_lapic *apic = vcpu->arch.apic; 1616 1617 if (!apic) { 1618 value |= MSR_IA32_APICBASE_BSP; 1619 vcpu->arch.apic_base = value; 1620 return; 1621 } 1622 1623 vcpu->arch.apic_base = value; 1624 1625 /* update jump label if enable bit changes */ 1626 if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) { 1627 if (value & MSR_IA32_APICBASE_ENABLE) 1628 static_key_slow_dec_deferred(&apic_hw_disabled); 1629 else 1630 static_key_slow_inc(&apic_hw_disabled.key); 1631 recalculate_apic_map(vcpu->kvm); 1632 } 1633 1634 if ((old_value ^ value) & X2APIC_ENABLE) { 1635 if (value & X2APIC_ENABLE) { 1636 kvm_apic_set_x2apic_id(apic, vcpu->vcpu_id); 1637 kvm_x86_ops->set_virtual_x2apic_mode(vcpu, true); 1638 } else 1639 kvm_x86_ops->set_virtual_x2apic_mode(vcpu, false); 1640 } 1641 1642 apic->base_address = apic->vcpu->arch.apic_base & 1643 MSR_IA32_APICBASE_BASE; 1644 1645 if ((value & MSR_IA32_APICBASE_ENABLE) && 1646 apic->base_address != APIC_DEFAULT_PHYS_BASE) 1647 pr_warn_once("APIC base relocation is unsupported by KVM"); 1648 1649 /* with FSB delivery interrupt, we can restart APIC functionality */ 1650 apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is " 1651 "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address); 1652 1653 } 1654 1655 void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event) 1656 { 1657 struct kvm_lapic *apic; 1658 int i; 1659 1660 apic_debug("%s\n", __func__); 1661 1662 ASSERT(vcpu); 1663 apic = vcpu->arch.apic; 1664 ASSERT(apic != NULL); 1665 1666 /* Stop the timer in case it's a reset to an active apic */ 1667 hrtimer_cancel(&apic->lapic_timer.timer); 1668 1669 if (!init_event) 1670 kvm_apic_set_id(apic, vcpu->vcpu_id); 1671 kvm_apic_set_version(apic->vcpu); 1672 1673 for (i = 0; i < APIC_LVT_NUM; i++) 1674 apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED); 1675 apic_update_lvtt(apic); 1676 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_LINT0_REENABLED)) 1677 apic_set_reg(apic, APIC_LVT0, 1678 SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT)); 1679 apic_manage_nmi_watchdog(apic, kvm_apic_get_reg(apic, APIC_LVT0)); 1680 1681 apic_set_reg(apic, APIC_DFR, 0xffffffffU); 1682 apic_set_spiv(apic, 0xff); 1683 apic_set_reg(apic, APIC_TASKPRI, 0); 1684 if (!apic_x2apic_mode(apic)) 1685 kvm_apic_set_ldr(apic, 0); 1686 apic_set_reg(apic, APIC_ESR, 0); 1687 apic_set_reg(apic, APIC_ICR, 0); 1688 apic_set_reg(apic, APIC_ICR2, 0); 1689 apic_set_reg(apic, APIC_TDCR, 0); 1690 apic_set_reg(apic, APIC_TMICT, 0); 1691 for (i = 0; i < 8; i++) { 1692 apic_set_reg(apic, APIC_IRR + 0x10 * i, 0); 1693 apic_set_reg(apic, APIC_ISR + 0x10 * i, 0); 1694 apic_set_reg(apic, APIC_TMR + 0x10 * i, 0); 1695 } 1696 apic->irr_pending = vcpu->arch.apicv_active; 1697 apic->isr_count = vcpu->arch.apicv_active ? 1 : 0; 1698 apic->highest_isr_cache = -1; 1699 update_divide_count(apic); 1700 atomic_set(&apic->lapic_timer.pending, 0); 1701 if (kvm_vcpu_is_bsp(vcpu)) 1702 kvm_lapic_set_base(vcpu, 1703 vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP); 1704 vcpu->arch.pv_eoi.msr_val = 0; 1705 apic_update_ppr(apic); 1706 1707 vcpu->arch.apic_arb_prio = 0; 1708 vcpu->arch.apic_attention = 0; 1709 1710 apic_debug("%s: vcpu=%p, id=%d, base_msr=" 1711 "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__, 1712 vcpu, kvm_apic_id(apic), 1713 vcpu->arch.apic_base, apic->base_address); 1714 } 1715 1716 /* 1717 *---------------------------------------------------------------------- 1718 * timer interface 1719 *---------------------------------------------------------------------- 1720 */ 1721 1722 static bool lapic_is_periodic(struct kvm_lapic *apic) 1723 { 1724 return apic_lvtt_period(apic); 1725 } 1726 1727 int apic_has_pending_timer(struct kvm_vcpu *vcpu) 1728 { 1729 struct kvm_lapic *apic = vcpu->arch.apic; 1730 1731 if (kvm_vcpu_has_lapic(vcpu) && apic_enabled(apic) && 1732 apic_lvt_enabled(apic, APIC_LVTT)) 1733 return atomic_read(&apic->lapic_timer.pending); 1734 1735 return 0; 1736 } 1737 1738 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type) 1739 { 1740 u32 reg = kvm_apic_get_reg(apic, lvt_type); 1741 int vector, mode, trig_mode; 1742 1743 if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) { 1744 vector = reg & APIC_VECTOR_MASK; 1745 mode = reg & APIC_MODE_MASK; 1746 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER; 1747 return __apic_accept_irq(apic, mode, vector, 1, trig_mode, 1748 NULL); 1749 } 1750 return 0; 1751 } 1752 1753 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu) 1754 { 1755 struct kvm_lapic *apic = vcpu->arch.apic; 1756 1757 if (apic) 1758 kvm_apic_local_deliver(apic, APIC_LVT0); 1759 } 1760 1761 static const struct kvm_io_device_ops apic_mmio_ops = { 1762 .read = apic_mmio_read, 1763 .write = apic_mmio_write, 1764 }; 1765 1766 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data) 1767 { 1768 struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer); 1769 struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer); 1770 1771 apic_timer_expired(apic); 1772 1773 if (lapic_is_periodic(apic)) { 1774 hrtimer_add_expires_ns(&ktimer->timer, ktimer->period); 1775 return HRTIMER_RESTART; 1776 } else 1777 return HRTIMER_NORESTART; 1778 } 1779 1780 int kvm_create_lapic(struct kvm_vcpu *vcpu) 1781 { 1782 struct kvm_lapic *apic; 1783 1784 ASSERT(vcpu != NULL); 1785 apic_debug("apic_init %d\n", vcpu->vcpu_id); 1786 1787 apic = kzalloc(sizeof(*apic), GFP_KERNEL); 1788 if (!apic) 1789 goto nomem; 1790 1791 vcpu->arch.apic = apic; 1792 1793 apic->regs = (void *)get_zeroed_page(GFP_KERNEL); 1794 if (!apic->regs) { 1795 printk(KERN_ERR "malloc apic regs error for vcpu %x\n", 1796 vcpu->vcpu_id); 1797 goto nomem_free_apic; 1798 } 1799 apic->vcpu = vcpu; 1800 1801 hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC, 1802 HRTIMER_MODE_ABS); 1803 apic->lapic_timer.timer.function = apic_timer_fn; 1804 1805 /* 1806 * APIC is created enabled. This will prevent kvm_lapic_set_base from 1807 * thinking that APIC satet has changed. 1808 */ 1809 vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE; 1810 kvm_lapic_set_base(vcpu, 1811 APIC_DEFAULT_PHYS_BASE | MSR_IA32_APICBASE_ENABLE); 1812 1813 static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */ 1814 kvm_lapic_reset(vcpu, false); 1815 kvm_iodevice_init(&apic->dev, &apic_mmio_ops); 1816 1817 return 0; 1818 nomem_free_apic: 1819 kfree(apic); 1820 nomem: 1821 return -ENOMEM; 1822 } 1823 1824 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu) 1825 { 1826 struct kvm_lapic *apic = vcpu->arch.apic; 1827 int highest_irr; 1828 1829 if (!kvm_vcpu_has_lapic(vcpu) || !apic_enabled(apic)) 1830 return -1; 1831 1832 apic_update_ppr(apic); 1833 highest_irr = apic_find_highest_irr(apic); 1834 if ((highest_irr == -1) || 1835 ((highest_irr & 0xF0) <= kvm_apic_get_reg(apic, APIC_PROCPRI))) 1836 return -1; 1837 return highest_irr; 1838 } 1839 1840 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu) 1841 { 1842 u32 lvt0 = kvm_apic_get_reg(vcpu->arch.apic, APIC_LVT0); 1843 int r = 0; 1844 1845 if (!kvm_apic_hw_enabled(vcpu->arch.apic)) 1846 r = 1; 1847 if ((lvt0 & APIC_LVT_MASKED) == 0 && 1848 GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT) 1849 r = 1; 1850 return r; 1851 } 1852 1853 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu) 1854 { 1855 struct kvm_lapic *apic = vcpu->arch.apic; 1856 1857 if (!kvm_vcpu_has_lapic(vcpu)) 1858 return; 1859 1860 if (atomic_read(&apic->lapic_timer.pending) > 0) { 1861 kvm_apic_local_deliver(apic, APIC_LVTT); 1862 if (apic_lvtt_tscdeadline(apic)) 1863 apic->lapic_timer.tscdeadline = 0; 1864 atomic_set(&apic->lapic_timer.pending, 0); 1865 } 1866 } 1867 1868 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu) 1869 { 1870 int vector = kvm_apic_has_interrupt(vcpu); 1871 struct kvm_lapic *apic = vcpu->arch.apic; 1872 1873 if (vector == -1) 1874 return -1; 1875 1876 /* 1877 * We get here even with APIC virtualization enabled, if doing 1878 * nested virtualization and L1 runs with the "acknowledge interrupt 1879 * on exit" mode. Then we cannot inject the interrupt via RVI, 1880 * because the process would deliver it through the IDT. 1881 */ 1882 1883 apic_set_isr(vector, apic); 1884 apic_update_ppr(apic); 1885 apic_clear_irr(vector, apic); 1886 1887 if (test_bit(vector, vcpu_to_synic(vcpu)->auto_eoi_bitmap)) { 1888 apic_clear_isr(vector, apic); 1889 apic_update_ppr(apic); 1890 } 1891 1892 return vector; 1893 } 1894 1895 void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu, 1896 struct kvm_lapic_state *s) 1897 { 1898 struct kvm_lapic *apic = vcpu->arch.apic; 1899 1900 kvm_lapic_set_base(vcpu, vcpu->arch.apic_base); 1901 /* set SPIV separately to get count of SW disabled APICs right */ 1902 apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV))); 1903 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s); 1904 /* call kvm_apic_set_id() to put apic into apic_map */ 1905 kvm_apic_set_id(apic, kvm_apic_id(apic)); 1906 kvm_apic_set_version(vcpu); 1907 1908 apic_update_ppr(apic); 1909 hrtimer_cancel(&apic->lapic_timer.timer); 1910 apic_update_lvtt(apic); 1911 apic_manage_nmi_watchdog(apic, kvm_apic_get_reg(apic, APIC_LVT0)); 1912 update_divide_count(apic); 1913 start_apic_timer(apic); 1914 apic->irr_pending = true; 1915 apic->isr_count = vcpu->arch.apicv_active ? 1916 1 : count_vectors(apic->regs + APIC_ISR); 1917 apic->highest_isr_cache = -1; 1918 if (vcpu->arch.apicv_active) { 1919 kvm_x86_ops->hwapic_irr_update(vcpu, 1920 apic_find_highest_irr(apic)); 1921 kvm_x86_ops->hwapic_isr_update(vcpu->kvm, 1922 apic_find_highest_isr(apic)); 1923 } 1924 kvm_make_request(KVM_REQ_EVENT, vcpu); 1925 if (ioapic_in_kernel(vcpu->kvm)) 1926 kvm_rtc_eoi_tracking_restore_one(vcpu); 1927 1928 vcpu->arch.apic_arb_prio = 0; 1929 } 1930 1931 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu) 1932 { 1933 struct hrtimer *timer; 1934 1935 if (!kvm_vcpu_has_lapic(vcpu)) 1936 return; 1937 1938 timer = &vcpu->arch.apic->lapic_timer.timer; 1939 if (hrtimer_cancel(timer)) 1940 hrtimer_start_expires(timer, HRTIMER_MODE_ABS); 1941 } 1942 1943 /* 1944 * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt 1945 * 1946 * Detect whether guest triggered PV EOI since the 1947 * last entry. If yes, set EOI on guests's behalf. 1948 * Clear PV EOI in guest memory in any case. 1949 */ 1950 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu, 1951 struct kvm_lapic *apic) 1952 { 1953 bool pending; 1954 int vector; 1955 /* 1956 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host 1957 * and KVM_PV_EOI_ENABLED in guest memory as follows: 1958 * 1959 * KVM_APIC_PV_EOI_PENDING is unset: 1960 * -> host disabled PV EOI. 1961 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set: 1962 * -> host enabled PV EOI, guest did not execute EOI yet. 1963 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset: 1964 * -> host enabled PV EOI, guest executed EOI. 1965 */ 1966 BUG_ON(!pv_eoi_enabled(vcpu)); 1967 pending = pv_eoi_get_pending(vcpu); 1968 /* 1969 * Clear pending bit in any case: it will be set again on vmentry. 1970 * While this might not be ideal from performance point of view, 1971 * this makes sure pv eoi is only enabled when we know it's safe. 1972 */ 1973 pv_eoi_clr_pending(vcpu); 1974 if (pending) 1975 return; 1976 vector = apic_set_eoi(apic); 1977 trace_kvm_pv_eoi(apic, vector); 1978 } 1979 1980 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu) 1981 { 1982 u32 data; 1983 1984 if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention)) 1985 apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic); 1986 1987 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention)) 1988 return; 1989 1990 if (kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data, 1991 sizeof(u32))) 1992 return; 1993 1994 apic_set_tpr(vcpu->arch.apic, data & 0xff); 1995 } 1996 1997 /* 1998 * apic_sync_pv_eoi_to_guest - called before vmentry 1999 * 2000 * Detect whether it's safe to enable PV EOI and 2001 * if yes do so. 2002 */ 2003 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu, 2004 struct kvm_lapic *apic) 2005 { 2006 if (!pv_eoi_enabled(vcpu) || 2007 /* IRR set or many bits in ISR: could be nested. */ 2008 apic->irr_pending || 2009 /* Cache not set: could be safe but we don't bother. */ 2010 apic->highest_isr_cache == -1 || 2011 /* Need EOI to update ioapic. */ 2012 kvm_ioapic_handles_vector(apic, apic->highest_isr_cache)) { 2013 /* 2014 * PV EOI was disabled by apic_sync_pv_eoi_from_guest 2015 * so we need not do anything here. 2016 */ 2017 return; 2018 } 2019 2020 pv_eoi_set_pending(apic->vcpu); 2021 } 2022 2023 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu) 2024 { 2025 u32 data, tpr; 2026 int max_irr, max_isr; 2027 struct kvm_lapic *apic = vcpu->arch.apic; 2028 2029 apic_sync_pv_eoi_to_guest(vcpu, apic); 2030 2031 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention)) 2032 return; 2033 2034 tpr = kvm_apic_get_reg(apic, APIC_TASKPRI) & 0xff; 2035 max_irr = apic_find_highest_irr(apic); 2036 if (max_irr < 0) 2037 max_irr = 0; 2038 max_isr = apic_find_highest_isr(apic); 2039 if (max_isr < 0) 2040 max_isr = 0; 2041 data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24); 2042 2043 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data, 2044 sizeof(u32)); 2045 } 2046 2047 int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr) 2048 { 2049 if (vapic_addr) { 2050 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, 2051 &vcpu->arch.apic->vapic_cache, 2052 vapic_addr, sizeof(u32))) 2053 return -EINVAL; 2054 __set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention); 2055 } else { 2056 __clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention); 2057 } 2058 2059 vcpu->arch.apic->vapic_addr = vapic_addr; 2060 return 0; 2061 } 2062 2063 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data) 2064 { 2065 struct kvm_lapic *apic = vcpu->arch.apic; 2066 u32 reg = (msr - APIC_BASE_MSR) << 4; 2067 2068 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic)) 2069 return 1; 2070 2071 if (reg == APIC_ICR2) 2072 return 1; 2073 2074 /* if this is ICR write vector before command */ 2075 if (reg == APIC_ICR) 2076 apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32)); 2077 return apic_reg_write(apic, reg, (u32)data); 2078 } 2079 2080 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data) 2081 { 2082 struct kvm_lapic *apic = vcpu->arch.apic; 2083 u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0; 2084 2085 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic)) 2086 return 1; 2087 2088 if (reg == APIC_DFR || reg == APIC_ICR2) { 2089 apic_debug("KVM_APIC_READ: read x2apic reserved register %x\n", 2090 reg); 2091 return 1; 2092 } 2093 2094 if (apic_reg_read(apic, reg, 4, &low)) 2095 return 1; 2096 if (reg == APIC_ICR) 2097 apic_reg_read(apic, APIC_ICR2, 4, &high); 2098 2099 *data = (((u64)high) << 32) | low; 2100 2101 return 0; 2102 } 2103 2104 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data) 2105 { 2106 struct kvm_lapic *apic = vcpu->arch.apic; 2107 2108 if (!kvm_vcpu_has_lapic(vcpu)) 2109 return 1; 2110 2111 /* if this is ICR write vector before command */ 2112 if (reg == APIC_ICR) 2113 apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32)); 2114 return apic_reg_write(apic, reg, (u32)data); 2115 } 2116 2117 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data) 2118 { 2119 struct kvm_lapic *apic = vcpu->arch.apic; 2120 u32 low, high = 0; 2121 2122 if (!kvm_vcpu_has_lapic(vcpu)) 2123 return 1; 2124 2125 if (apic_reg_read(apic, reg, 4, &low)) 2126 return 1; 2127 if (reg == APIC_ICR) 2128 apic_reg_read(apic, APIC_ICR2, 4, &high); 2129 2130 *data = (((u64)high) << 32) | low; 2131 2132 return 0; 2133 } 2134 2135 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data) 2136 { 2137 u64 addr = data & ~KVM_MSR_ENABLED; 2138 if (!IS_ALIGNED(addr, 4)) 2139 return 1; 2140 2141 vcpu->arch.pv_eoi.msr_val = data; 2142 if (!pv_eoi_enabled(vcpu)) 2143 return 0; 2144 return kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.pv_eoi.data, 2145 addr, sizeof(u8)); 2146 } 2147 2148 void kvm_apic_accept_events(struct kvm_vcpu *vcpu) 2149 { 2150 struct kvm_lapic *apic = vcpu->arch.apic; 2151 u8 sipi_vector; 2152 unsigned long pe; 2153 2154 if (!kvm_vcpu_has_lapic(vcpu) || !apic->pending_events) 2155 return; 2156 2157 /* 2158 * INITs are latched while in SMM. Because an SMM CPU cannot 2159 * be in KVM_MP_STATE_INIT_RECEIVED state, just eat SIPIs 2160 * and delay processing of INIT until the next RSM. 2161 */ 2162 if (is_smm(vcpu)) { 2163 WARN_ON_ONCE(vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED); 2164 if (test_bit(KVM_APIC_SIPI, &apic->pending_events)) 2165 clear_bit(KVM_APIC_SIPI, &apic->pending_events); 2166 return; 2167 } 2168 2169 pe = xchg(&apic->pending_events, 0); 2170 if (test_bit(KVM_APIC_INIT, &pe)) { 2171 kvm_lapic_reset(vcpu, true); 2172 kvm_vcpu_reset(vcpu, true); 2173 if (kvm_vcpu_is_bsp(apic->vcpu)) 2174 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE; 2175 else 2176 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED; 2177 } 2178 if (test_bit(KVM_APIC_SIPI, &pe) && 2179 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) { 2180 /* evaluate pending_events before reading the vector */ 2181 smp_rmb(); 2182 sipi_vector = apic->sipi_vector; 2183 apic_debug("vcpu %d received sipi with vector # %x\n", 2184 vcpu->vcpu_id, sipi_vector); 2185 kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector); 2186 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE; 2187 } 2188 } 2189 2190 void kvm_lapic_init(void) 2191 { 2192 /* do not patch jump label more than once per second */ 2193 jump_label_rate_limit(&apic_hw_disabled, HZ); 2194 jump_label_rate_limit(&apic_sw_disabled, HZ); 2195 } 2196