aboutsummaryrefslogtreecommitdiffstats
path: root/arch/x86/kvm/x86.c
diff options
context:
space:
mode:
authorGlenn Elliott <gelliott@cs.unc.edu>2012-03-04 19:47:13 -0500
committerGlenn Elliott <gelliott@cs.unc.edu>2012-03-04 19:47:13 -0500
commitc71c03bda1e86c9d5198c5d83f712e695c4f2a1e (patch)
treeecb166cb3e2b7e2adb3b5e292245fefd23381ac8 /arch/x86/kvm/x86.c
parentea53c912f8a86a8567697115b6a0d8152beee5c8 (diff)
parent6a00f206debf8a5c8899055726ad127dbeeed098 (diff)
Merge branch 'mpi-master' into wip-k-fmlpwip-k-fmlp
Conflicts: litmus/sched_cedf.c
Diffstat (limited to 'arch/x86/kvm/x86.c')
-rw-r--r--arch/x86/kvm/x86.c1860
1 files changed, 1381 insertions, 479 deletions
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 3a09c625d526..77c9d8673dc4 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -6,7 +6,7 @@
6 * Copyright (C) 2006 Qumranet, Inc. 6 * Copyright (C) 2006 Qumranet, Inc.
7 * Copyright (C) 2008 Qumranet, Inc. 7 * Copyright (C) 2008 Qumranet, Inc.
8 * Copyright IBM Corporation, 2008 8 * Copyright IBM Corporation, 2008
9 * Copyright 2010 Red Hat, Inc. and/or its affilates. 9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
10 * 10 *
11 * Authors: 11 * Authors:
12 * Avi Kivity <avi@qumranet.com> 12 * Avi Kivity <avi@qumranet.com>
@@ -43,6 +43,7 @@
43#include <linux/slab.h> 43#include <linux/slab.h>
44#include <linux/perf_event.h> 44#include <linux/perf_event.h>
45#include <linux/uaccess.h> 45#include <linux/uaccess.h>
46#include <linux/hash.h>
46#include <trace/events/kvm.h> 47#include <trace/events/kvm.h>
47 48
48#define CREATE_TRACE_POINTS 49#define CREATE_TRACE_POINTS
@@ -55,32 +56,25 @@
55#include <asm/mce.h> 56#include <asm/mce.h>
56#include <asm/i387.h> 57#include <asm/i387.h>
57#include <asm/xcr.h> 58#include <asm/xcr.h>
59#include <asm/pvclock.h>
60#include <asm/div64.h>
58 61
59#define MAX_IO_MSRS 256 62#define MAX_IO_MSRS 256
60#define CR0_RESERVED_BITS \
61 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
62 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
63 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
64#define CR4_RESERVED_BITS \
65 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
66 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
67 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
68 | X86_CR4_OSXSAVE \
69 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
70
71#define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
72
73#define KVM_MAX_MCE_BANKS 32 63#define KVM_MAX_MCE_BANKS 32
74#define KVM_MCE_CAP_SUPPORTED MCG_CTL_P 64#define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
65
66#define emul_to_vcpu(ctxt) \
67 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
75 68
76/* EFER defaults: 69/* EFER defaults:
77 * - enable syscall per default because its emulated by KVM 70 * - enable syscall per default because its emulated by KVM
78 * - enable LME and LMA per default on 64 bit KVM 71 * - enable LME and LMA per default on 64 bit KVM
79 */ 72 */
80#ifdef CONFIG_X86_64 73#ifdef CONFIG_X86_64
81static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL; 74static
75u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
82#else 76#else
83static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL; 77static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
84#endif 78#endif
85 79
86#define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM 80#define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
@@ -96,6 +90,11 @@ EXPORT_SYMBOL_GPL(kvm_x86_ops);
96int ignore_msrs = 0; 90int ignore_msrs = 0;
97module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR); 91module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
98 92
93bool kvm_has_tsc_control;
94EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
95u32 kvm_max_guest_tsc_khz;
96EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
97
99#define KVM_NR_SHARED_MSRS 16 98#define KVM_NR_SHARED_MSRS 16
100 99
101struct kvm_shared_msrs_global { 100struct kvm_shared_msrs_global {
@@ -153,9 +152,13 @@ struct kvm_stats_debugfs_item debugfs_entries[] = {
153 152
154u64 __read_mostly host_xcr0; 153u64 __read_mostly host_xcr0;
155 154
156static inline u32 bit(int bitno) 155int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
156
157static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
157{ 158{
158 return 1 << (bitno & 31); 159 int i;
160 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
161 vcpu->arch.apf.gfns[i] = ~0;
159} 162}
160 163
161static void kvm_on_user_return(struct user_return_notifier *urn) 164static void kvm_on_user_return(struct user_return_notifier *urn)
@@ -282,6 +285,8 @@ static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
282 u32 prev_nr; 285 u32 prev_nr;
283 int class1, class2; 286 int class1, class2;
284 287
288 kvm_make_request(KVM_REQ_EVENT, vcpu);
289
285 if (!vcpu->arch.exception.pending) { 290 if (!vcpu->arch.exception.pending) {
286 queue: 291 queue:
287 vcpu->arch.exception.pending = true; 292 vcpu->arch.exception.pending = true;
@@ -327,16 +332,33 @@ void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
327} 332}
328EXPORT_SYMBOL_GPL(kvm_requeue_exception); 333EXPORT_SYMBOL_GPL(kvm_requeue_exception);
329 334
330void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr, 335void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
331 u32 error_code) 336{
337 if (err)
338 kvm_inject_gp(vcpu, 0);
339 else
340 kvm_x86_ops->skip_emulated_instruction(vcpu);
341}
342EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
343
344void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
332{ 345{
333 ++vcpu->stat.pf_guest; 346 ++vcpu->stat.pf_guest;
334 vcpu->arch.cr2 = addr; 347 vcpu->arch.cr2 = fault->address;
335 kvm_queue_exception_e(vcpu, PF_VECTOR, error_code); 348 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
349}
350
351void kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
352{
353 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
354 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
355 else
356 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
336} 357}
337 358
338void kvm_inject_nmi(struct kvm_vcpu *vcpu) 359void kvm_inject_nmi(struct kvm_vcpu *vcpu)
339{ 360{
361 kvm_make_request(KVM_REQ_EVENT, vcpu);
340 vcpu->arch.nmi_pending = 1; 362 vcpu->arch.nmi_pending = 1;
341} 363}
342EXPORT_SYMBOL_GPL(kvm_inject_nmi); 364EXPORT_SYMBOL_GPL(kvm_inject_nmi);
@@ -367,18 +389,49 @@ bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
367EXPORT_SYMBOL_GPL(kvm_require_cpl); 389EXPORT_SYMBOL_GPL(kvm_require_cpl);
368 390
369/* 391/*
392 * This function will be used to read from the physical memory of the currently
393 * running guest. The difference to kvm_read_guest_page is that this function
394 * can read from guest physical or from the guest's guest physical memory.
395 */
396int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
397 gfn_t ngfn, void *data, int offset, int len,
398 u32 access)
399{
400 gfn_t real_gfn;
401 gpa_t ngpa;
402
403 ngpa = gfn_to_gpa(ngfn);
404 real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
405 if (real_gfn == UNMAPPED_GVA)
406 return -EFAULT;
407
408 real_gfn = gpa_to_gfn(real_gfn);
409
410 return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
411}
412EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
413
414int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
415 void *data, int offset, int len, u32 access)
416{
417 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
418 data, offset, len, access);
419}
420
421/*
370 * Load the pae pdptrs. Return true is they are all valid. 422 * Load the pae pdptrs. Return true is they are all valid.
371 */ 423 */
372int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3) 424int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
373{ 425{
374 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT; 426 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
375 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2; 427 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
376 int i; 428 int i;
377 int ret; 429 int ret;
378 u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)]; 430 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
379 431
380 ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte, 432 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
381 offset * sizeof(u64), sizeof(pdpte)); 433 offset * sizeof(u64), sizeof(pdpte),
434 PFERR_USER_MASK|PFERR_WRITE_MASK);
382 if (ret < 0) { 435 if (ret < 0) {
383 ret = 0; 436 ret = 0;
384 goto out; 437 goto out;
@@ -392,7 +445,7 @@ int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
392 } 445 }
393 ret = 1; 446 ret = 1;
394 447
395 memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs)); 448 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
396 __set_bit(VCPU_EXREG_PDPTR, 449 __set_bit(VCPU_EXREG_PDPTR,
397 (unsigned long *)&vcpu->arch.regs_avail); 450 (unsigned long *)&vcpu->arch.regs_avail);
398 __set_bit(VCPU_EXREG_PDPTR, 451 __set_bit(VCPU_EXREG_PDPTR,
@@ -405,8 +458,10 @@ EXPORT_SYMBOL_GPL(load_pdptrs);
405 458
406static bool pdptrs_changed(struct kvm_vcpu *vcpu) 459static bool pdptrs_changed(struct kvm_vcpu *vcpu)
407{ 460{
408 u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)]; 461 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
409 bool changed = true; 462 bool changed = true;
463 int offset;
464 gfn_t gfn;
410 int r; 465 int r;
411 466
412 if (is_long_mode(vcpu) || !is_pae(vcpu)) 467 if (is_long_mode(vcpu) || !is_pae(vcpu))
@@ -416,10 +471,13 @@ static bool pdptrs_changed(struct kvm_vcpu *vcpu)
416 (unsigned long *)&vcpu->arch.regs_avail)) 471 (unsigned long *)&vcpu->arch.regs_avail))
417 return true; 472 return true;
418 473
419 r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte)); 474 gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
475 offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
476 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
477 PFERR_USER_MASK | PFERR_WRITE_MASK);
420 if (r < 0) 478 if (r < 0)
421 goto out; 479 goto out;
422 changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0; 480 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
423out: 481out:
424 482
425 return changed; 483 return changed;
@@ -458,12 +516,18 @@ int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
458 return 1; 516 return 1;
459 } else 517 } else
460#endif 518#endif
461 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) 519 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
520 kvm_read_cr3(vcpu)))
462 return 1; 521 return 1;
463 } 522 }
464 523
465 kvm_x86_ops->set_cr0(vcpu, cr0); 524 kvm_x86_ops->set_cr0(vcpu, cr0);
466 525
526 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
527 kvm_clear_async_pf_completion_queue(vcpu);
528 kvm_async_pf_hash_reset(vcpu);
529 }
530
467 if ((cr0 ^ old_cr0) & update_bits) 531 if ((cr0 ^ old_cr0) & update_bits)
468 kvm_mmu_reset_context(vcpu); 532 kvm_mmu_reset_context(vcpu);
469 return 0; 533 return 0;
@@ -547,7 +611,8 @@ int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
547 return 1; 611 return 1;
548 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE) 612 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
549 && ((cr4 ^ old_cr4) & pdptr_bits) 613 && ((cr4 ^ old_cr4) & pdptr_bits)
550 && !load_pdptrs(vcpu, vcpu->arch.cr3)) 614 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
615 kvm_read_cr3(vcpu)))
551 return 1; 616 return 1;
552 617
553 if (cr4 & X86_CR4_VMXE) 618 if (cr4 & X86_CR4_VMXE)
@@ -567,7 +632,7 @@ EXPORT_SYMBOL_GPL(kvm_set_cr4);
567 632
568int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3) 633int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
569{ 634{
570 if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) { 635 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
571 kvm_mmu_sync_roots(vcpu); 636 kvm_mmu_sync_roots(vcpu);
572 kvm_mmu_flush_tlb(vcpu); 637 kvm_mmu_flush_tlb(vcpu);
573 return 0; 638 return 0;
@@ -580,7 +645,8 @@ int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
580 if (is_pae(vcpu)) { 645 if (is_pae(vcpu)) {
581 if (cr3 & CR3_PAE_RESERVED_BITS) 646 if (cr3 & CR3_PAE_RESERVED_BITS)
582 return 1; 647 return 1;
583 if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) 648 if (is_paging(vcpu) &&
649 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
584 return 1; 650 return 1;
585 } 651 }
586 /* 652 /*
@@ -601,12 +667,13 @@ int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
601 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT))) 667 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
602 return 1; 668 return 1;
603 vcpu->arch.cr3 = cr3; 669 vcpu->arch.cr3 = cr3;
670 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
604 vcpu->arch.mmu.new_cr3(vcpu); 671 vcpu->arch.mmu.new_cr3(vcpu);
605 return 0; 672 return 0;
606} 673}
607EXPORT_SYMBOL_GPL(kvm_set_cr3); 674EXPORT_SYMBOL_GPL(kvm_set_cr3);
608 675
609int __kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8) 676int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
610{ 677{
611 if (cr8 & CR8_RESERVED_BITS) 678 if (cr8 & CR8_RESERVED_BITS)
612 return 1; 679 return 1;
@@ -616,12 +683,6 @@ int __kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
616 vcpu->arch.cr8 = cr8; 683 vcpu->arch.cr8 = cr8;
617 return 0; 684 return 0;
618} 685}
619
620void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
621{
622 if (__kvm_set_cr8(vcpu, cr8))
623 kvm_inject_gp(vcpu, 0);
624}
625EXPORT_SYMBOL_GPL(kvm_set_cr8); 686EXPORT_SYMBOL_GPL(kvm_set_cr8);
626 687
627unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu) 688unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
@@ -726,18 +787,18 @@ EXPORT_SYMBOL_GPL(kvm_get_dr);
726 * kvm-specific. Those are put in the beginning of the list. 787 * kvm-specific. Those are put in the beginning of the list.
727 */ 788 */
728 789
729#define KVM_SAVE_MSRS_BEGIN 7 790#define KVM_SAVE_MSRS_BEGIN 8
730static u32 msrs_to_save[] = { 791static u32 msrs_to_save[] = {
731 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK, 792 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
732 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW, 793 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
733 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL, 794 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
734 HV_X64_MSR_APIC_ASSIST_PAGE, 795 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN,
735 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP, 796 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
736 MSR_STAR, 797 MSR_STAR,
737#ifdef CONFIG_X86_64 798#ifdef CONFIG_X86_64
738 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR, 799 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
739#endif 800#endif
740 MSR_IA32_TSC, MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA 801 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
741}; 802};
742 803
743static unsigned num_msrs_to_save; 804static unsigned num_msrs_to_save;
@@ -781,7 +842,6 @@ static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
781 kvm_x86_ops->set_efer(vcpu, efer); 842 kvm_x86_ops->set_efer(vcpu, efer);
782 843
783 vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled; 844 vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
784 kvm_mmu_reset_context(vcpu);
785 845
786 /* Update reserved bits */ 846 /* Update reserved bits */
787 if ((efer ^ old_efer) & EFER_NX) 847 if ((efer ^ old_efer) & EFER_NX)
@@ -838,7 +898,7 @@ static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
838 898
839 /* 899 /*
840 * The guest calculates current wall clock time by adding 900 * The guest calculates current wall clock time by adding
841 * system time (updated by kvm_write_guest_time below) to the 901 * system time (updated by kvm_guest_time_update below) to the
842 * wall clock specified here. guest system time equals host 902 * wall clock specified here. guest system time equals host
843 * system time for us, thus we must fill in host boot time here. 903 * system time for us, thus we must fill in host boot time here.
844 */ 904 */
@@ -866,65 +926,235 @@ static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
866 return quotient; 926 return quotient;
867} 927}
868 928
869static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock) 929static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
930 s8 *pshift, u32 *pmultiplier)
870{ 931{
871 uint64_t nsecs = 1000000000LL; 932 uint64_t scaled64;
872 int32_t shift = 0; 933 int32_t shift = 0;
873 uint64_t tps64; 934 uint64_t tps64;
874 uint32_t tps32; 935 uint32_t tps32;
875 936
876 tps64 = tsc_khz * 1000LL; 937 tps64 = base_khz * 1000LL;
877 while (tps64 > nsecs*2) { 938 scaled64 = scaled_khz * 1000LL;
939 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
878 tps64 >>= 1; 940 tps64 >>= 1;
879 shift--; 941 shift--;
880 } 942 }
881 943
882 tps32 = (uint32_t)tps64; 944 tps32 = (uint32_t)tps64;
883 while (tps32 <= (uint32_t)nsecs) { 945 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
884 tps32 <<= 1; 946 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
947 scaled64 >>= 1;
948 else
949 tps32 <<= 1;
885 shift++; 950 shift++;
886 } 951 }
887 952
888 hv_clock->tsc_shift = shift; 953 *pshift = shift;
889 hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32); 954 *pmultiplier = div_frac(scaled64, tps32);
890 955
891 pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n", 956 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
892 __func__, tsc_khz, hv_clock->tsc_shift, 957 __func__, base_khz, scaled_khz, shift, *pmultiplier);
893 hv_clock->tsc_to_system_mul); 958}
959
960static inline u64 get_kernel_ns(void)
961{
962 struct timespec ts;
963
964 WARN_ON(preemptible());
965 ktime_get_ts(&ts);
966 monotonic_to_bootbased(&ts);
967 return timespec_to_ns(&ts);
894} 968}
895 969
896static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz); 970static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
971unsigned long max_tsc_khz;
897 972
898static void kvm_write_guest_time(struct kvm_vcpu *v) 973static inline int kvm_tsc_changes_freq(void)
974{
975 int cpu = get_cpu();
976 int ret = !boot_cpu_has(X86_FEATURE_CONSTANT_TSC) &&
977 cpufreq_quick_get(cpu) != 0;
978 put_cpu();
979 return ret;
980}
981
982static u64 vcpu_tsc_khz(struct kvm_vcpu *vcpu)
983{
984 if (vcpu->arch.virtual_tsc_khz)
985 return vcpu->arch.virtual_tsc_khz;
986 else
987 return __this_cpu_read(cpu_tsc_khz);
988}
989
990static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
991{
992 u64 ret;
993
994 WARN_ON(preemptible());
995 if (kvm_tsc_changes_freq())
996 printk_once(KERN_WARNING
997 "kvm: unreliable cycle conversion on adjustable rate TSC\n");
998 ret = nsec * vcpu_tsc_khz(vcpu);
999 do_div(ret, USEC_PER_SEC);
1000 return ret;
1001}
1002
1003static void kvm_init_tsc_catchup(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1004{
1005 /* Compute a scale to convert nanoseconds in TSC cycles */
1006 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1007 &vcpu->arch.tsc_catchup_shift,
1008 &vcpu->arch.tsc_catchup_mult);
1009}
1010
1011static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1012{
1013 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.last_tsc_nsec,
1014 vcpu->arch.tsc_catchup_mult,
1015 vcpu->arch.tsc_catchup_shift);
1016 tsc += vcpu->arch.last_tsc_write;
1017 return tsc;
1018}
1019
1020void kvm_write_tsc(struct kvm_vcpu *vcpu, u64 data)
1021{
1022 struct kvm *kvm = vcpu->kvm;
1023 u64 offset, ns, elapsed;
1024 unsigned long flags;
1025 s64 sdiff;
1026
1027 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1028 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1029 ns = get_kernel_ns();
1030 elapsed = ns - kvm->arch.last_tsc_nsec;
1031 sdiff = data - kvm->arch.last_tsc_write;
1032 if (sdiff < 0)
1033 sdiff = -sdiff;
1034
1035 /*
1036 * Special case: close write to TSC within 5 seconds of
1037 * another CPU is interpreted as an attempt to synchronize
1038 * The 5 seconds is to accommodate host load / swapping as
1039 * well as any reset of TSC during the boot process.
1040 *
1041 * In that case, for a reliable TSC, we can match TSC offsets,
1042 * or make a best guest using elapsed value.
1043 */
1044 if (sdiff < nsec_to_cycles(vcpu, 5ULL * NSEC_PER_SEC) &&
1045 elapsed < 5ULL * NSEC_PER_SEC) {
1046 if (!check_tsc_unstable()) {
1047 offset = kvm->arch.last_tsc_offset;
1048 pr_debug("kvm: matched tsc offset for %llu\n", data);
1049 } else {
1050 u64 delta = nsec_to_cycles(vcpu, elapsed);
1051 offset += delta;
1052 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1053 }
1054 ns = kvm->arch.last_tsc_nsec;
1055 }
1056 kvm->arch.last_tsc_nsec = ns;
1057 kvm->arch.last_tsc_write = data;
1058 kvm->arch.last_tsc_offset = offset;
1059 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1060 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1061
1062 /* Reset of TSC must disable overshoot protection below */
1063 vcpu->arch.hv_clock.tsc_timestamp = 0;
1064 vcpu->arch.last_tsc_write = data;
1065 vcpu->arch.last_tsc_nsec = ns;
1066}
1067EXPORT_SYMBOL_GPL(kvm_write_tsc);
1068
1069static int kvm_guest_time_update(struct kvm_vcpu *v)
899{ 1070{
900 struct timespec ts;
901 unsigned long flags; 1071 unsigned long flags;
902 struct kvm_vcpu_arch *vcpu = &v->arch; 1072 struct kvm_vcpu_arch *vcpu = &v->arch;
903 void *shared_kaddr; 1073 void *shared_kaddr;
904 unsigned long this_tsc_khz; 1074 unsigned long this_tsc_khz;
1075 s64 kernel_ns, max_kernel_ns;
1076 u64 tsc_timestamp;
905 1077
906 if ((!vcpu->time_page)) 1078 /* Keep irq disabled to prevent changes to the clock */
907 return; 1079 local_irq_save(flags);
1080 kvm_get_msr(v, MSR_IA32_TSC, &tsc_timestamp);
1081 kernel_ns = get_kernel_ns();
1082 this_tsc_khz = vcpu_tsc_khz(v);
1083 if (unlikely(this_tsc_khz == 0)) {
1084 local_irq_restore(flags);
1085 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1086 return 1;
1087 }
908 1088
909 this_tsc_khz = get_cpu_var(cpu_tsc_khz); 1089 /*
910 if (unlikely(vcpu->hv_clock_tsc_khz != this_tsc_khz)) { 1090 * We may have to catch up the TSC to match elapsed wall clock
911 kvm_set_time_scale(this_tsc_khz, &vcpu->hv_clock); 1091 * time for two reasons, even if kvmclock is used.
912 vcpu->hv_clock_tsc_khz = this_tsc_khz; 1092 * 1) CPU could have been running below the maximum TSC rate
1093 * 2) Broken TSC compensation resets the base at each VCPU
1094 * entry to avoid unknown leaps of TSC even when running
1095 * again on the same CPU. This may cause apparent elapsed
1096 * time to disappear, and the guest to stand still or run
1097 * very slowly.
1098 */
1099 if (vcpu->tsc_catchup) {
1100 u64 tsc = compute_guest_tsc(v, kernel_ns);
1101 if (tsc > tsc_timestamp) {
1102 kvm_x86_ops->adjust_tsc_offset(v, tsc - tsc_timestamp);
1103 tsc_timestamp = tsc;
1104 }
913 } 1105 }
914 put_cpu_var(cpu_tsc_khz);
915 1106
916 /* Keep irq disabled to prevent changes to the clock */
917 local_irq_save(flags);
918 kvm_get_msr(v, MSR_IA32_TSC, &vcpu->hv_clock.tsc_timestamp);
919 ktime_get_ts(&ts);
920 monotonic_to_bootbased(&ts);
921 local_irq_restore(flags); 1107 local_irq_restore(flags);
922 1108
923 /* With all the info we got, fill in the values */ 1109 if (!vcpu->time_page)
1110 return 0;
924 1111
925 vcpu->hv_clock.system_time = ts.tv_nsec + 1112 /*
926 (NSEC_PER_SEC * (u64)ts.tv_sec) + v->kvm->arch.kvmclock_offset; 1113 * Time as measured by the TSC may go backwards when resetting the base
1114 * tsc_timestamp. The reason for this is that the TSC resolution is
1115 * higher than the resolution of the other clock scales. Thus, many
1116 * possible measurments of the TSC correspond to one measurement of any
1117 * other clock, and so a spread of values is possible. This is not a
1118 * problem for the computation of the nanosecond clock; with TSC rates
1119 * around 1GHZ, there can only be a few cycles which correspond to one
1120 * nanosecond value, and any path through this code will inevitably
1121 * take longer than that. However, with the kernel_ns value itself,
1122 * the precision may be much lower, down to HZ granularity. If the
1123 * first sampling of TSC against kernel_ns ends in the low part of the
1124 * range, and the second in the high end of the range, we can get:
1125 *
1126 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1127 *
1128 * As the sampling errors potentially range in the thousands of cycles,
1129 * it is possible such a time value has already been observed by the
1130 * guest. To protect against this, we must compute the system time as
1131 * observed by the guest and ensure the new system time is greater.
1132 */
1133 max_kernel_ns = 0;
1134 if (vcpu->hv_clock.tsc_timestamp && vcpu->last_guest_tsc) {
1135 max_kernel_ns = vcpu->last_guest_tsc -
1136 vcpu->hv_clock.tsc_timestamp;
1137 max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1138 vcpu->hv_clock.tsc_to_system_mul,
1139 vcpu->hv_clock.tsc_shift);
1140 max_kernel_ns += vcpu->last_kernel_ns;
1141 }
927 1142
1143 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1144 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1145 &vcpu->hv_clock.tsc_shift,
1146 &vcpu->hv_clock.tsc_to_system_mul);
1147 vcpu->hw_tsc_khz = this_tsc_khz;
1148 }
1149
1150 if (max_kernel_ns > kernel_ns)
1151 kernel_ns = max_kernel_ns;
1152
1153 /* With all the info we got, fill in the values */
1154 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1155 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1156 vcpu->last_kernel_ns = kernel_ns;
1157 vcpu->last_guest_tsc = tsc_timestamp;
928 vcpu->hv_clock.flags = 0; 1158 vcpu->hv_clock.flags = 0;
929 1159
930 /* 1160 /*
@@ -942,16 +1172,7 @@ static void kvm_write_guest_time(struct kvm_vcpu *v)
942 kunmap_atomic(shared_kaddr, KM_USER0); 1172 kunmap_atomic(shared_kaddr, KM_USER0);
943 1173
944 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT); 1174 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
945} 1175 return 0;
946
947static int kvm_request_guest_time_update(struct kvm_vcpu *v)
948{
949 struct kvm_vcpu_arch *vcpu = &v->arch;
950
951 if (!vcpu->time_page)
952 return 0;
953 kvm_make_request(KVM_REQ_KVMCLOCK_UPDATE, v);
954 return 1;
955} 1176}
956 1177
957static bool msr_mtrr_valid(unsigned msr) 1178static bool msr_mtrr_valid(unsigned msr)
@@ -1214,6 +1435,38 @@ static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1214 return 0; 1435 return 0;
1215} 1436}
1216 1437
1438static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1439{
1440 gpa_t gpa = data & ~0x3f;
1441
1442 /* Bits 2:5 are resrved, Should be zero */
1443 if (data & 0x3c)
1444 return 1;
1445
1446 vcpu->arch.apf.msr_val = data;
1447
1448 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1449 kvm_clear_async_pf_completion_queue(vcpu);
1450 kvm_async_pf_hash_reset(vcpu);
1451 return 0;
1452 }
1453
1454 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa))
1455 return 1;
1456
1457 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1458 kvm_async_pf_wakeup_all(vcpu);
1459 return 0;
1460}
1461
1462static void kvmclock_reset(struct kvm_vcpu *vcpu)
1463{
1464 if (vcpu->arch.time_page) {
1465 kvm_release_page_dirty(vcpu->arch.time_page);
1466 vcpu->arch.time_page = NULL;
1467 }
1468}
1469
1217int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data) 1470int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1218{ 1471{
1219 switch (msr) { 1472 switch (msr) {
@@ -1271,12 +1524,10 @@ int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1271 break; 1524 break;
1272 case MSR_KVM_SYSTEM_TIME_NEW: 1525 case MSR_KVM_SYSTEM_TIME_NEW:
1273 case MSR_KVM_SYSTEM_TIME: { 1526 case MSR_KVM_SYSTEM_TIME: {
1274 if (vcpu->arch.time_page) { 1527 kvmclock_reset(vcpu);
1275 kvm_release_page_dirty(vcpu->arch.time_page);
1276 vcpu->arch.time_page = NULL;
1277 }
1278 1528
1279 vcpu->arch.time = data; 1529 vcpu->arch.time = data;
1530 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1280 1531
1281 /* we verify if the enable bit is set... */ 1532 /* we verify if the enable bit is set... */
1282 if (!(data & 1)) 1533 if (!(data & 1))
@@ -1292,10 +1543,12 @@ int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1292 kvm_release_page_clean(vcpu->arch.time_page); 1543 kvm_release_page_clean(vcpu->arch.time_page);
1293 vcpu->arch.time_page = NULL; 1544 vcpu->arch.time_page = NULL;
1294 } 1545 }
1295
1296 kvm_request_guest_time_update(vcpu);
1297 break; 1546 break;
1298 } 1547 }
1548 case MSR_KVM_ASYNC_PF_EN:
1549 if (kvm_pv_enable_async_pf(vcpu, data))
1550 return 1;
1551 break;
1299 case MSR_IA32_MCG_CTL: 1552 case MSR_IA32_MCG_CTL:
1300 case MSR_IA32_MCG_STATUS: 1553 case MSR_IA32_MCG_STATUS:
1301 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1: 1554 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
@@ -1330,6 +1583,16 @@ int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1330 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: " 1583 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1331 "0x%x data 0x%llx\n", msr, data); 1584 "0x%x data 0x%llx\n", msr, data);
1332 break; 1585 break;
1586 case MSR_K7_CLK_CTL:
1587 /*
1588 * Ignore all writes to this no longer documented MSR.
1589 * Writes are only relevant for old K7 processors,
1590 * all pre-dating SVM, but a recommended workaround from
1591 * AMD for these chips. It is possible to speicify the
1592 * affected processor models on the command line, hence
1593 * the need to ignore the workaround.
1594 */
1595 break;
1333 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15: 1596 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1334 if (kvm_hv_msr_partition_wide(msr)) { 1597 if (kvm_hv_msr_partition_wide(msr)) {
1335 int r; 1598 int r;
@@ -1340,6 +1603,12 @@ int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1340 } else 1603 } else
1341 return set_msr_hyperv(vcpu, msr, data); 1604 return set_msr_hyperv(vcpu, msr, data);
1342 break; 1605 break;
1606 case MSR_IA32_BBL_CR_CTL3:
1607 /* Drop writes to this legacy MSR -- see rdmsr
1608 * counterpart for further detail.
1609 */
1610 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
1611 break;
1343 default: 1612 default:
1344 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr)) 1613 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1345 return xen_hvm_config(vcpu, data); 1614 return xen_hvm_config(vcpu, data);
@@ -1522,6 +1791,20 @@ int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1522 case 0xcd: /* fsb frequency */ 1791 case 0xcd: /* fsb frequency */
1523 data = 3; 1792 data = 3;
1524 break; 1793 break;
1794 /*
1795 * MSR_EBC_FREQUENCY_ID
1796 * Conservative value valid for even the basic CPU models.
1797 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
1798 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
1799 * and 266MHz for model 3, or 4. Set Core Clock
1800 * Frequency to System Bus Frequency Ratio to 1 (bits
1801 * 31:24) even though these are only valid for CPU
1802 * models > 2, however guests may end up dividing or
1803 * multiplying by zero otherwise.
1804 */
1805 case MSR_EBC_FREQUENCY_ID:
1806 data = 1 << 24;
1807 break;
1525 case MSR_IA32_APICBASE: 1808 case MSR_IA32_APICBASE:
1526 data = kvm_get_apic_base(vcpu); 1809 data = kvm_get_apic_base(vcpu);
1527 break; 1810 break;
@@ -1548,6 +1831,9 @@ int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1548 case MSR_KVM_SYSTEM_TIME_NEW: 1831 case MSR_KVM_SYSTEM_TIME_NEW:
1549 data = vcpu->arch.time; 1832 data = vcpu->arch.time;
1550 break; 1833 break;
1834 case MSR_KVM_ASYNC_PF_EN:
1835 data = vcpu->arch.apf.msr_val;
1836 break;
1551 case MSR_IA32_P5_MC_ADDR: 1837 case MSR_IA32_P5_MC_ADDR:
1552 case MSR_IA32_P5_MC_TYPE: 1838 case MSR_IA32_P5_MC_TYPE:
1553 case MSR_IA32_MCG_CAP: 1839 case MSR_IA32_MCG_CAP:
@@ -1555,6 +1841,18 @@ int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1555 case MSR_IA32_MCG_STATUS: 1841 case MSR_IA32_MCG_STATUS:
1556 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1: 1842 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1557 return get_msr_mce(vcpu, msr, pdata); 1843 return get_msr_mce(vcpu, msr, pdata);
1844 case MSR_K7_CLK_CTL:
1845 /*
1846 * Provide expected ramp-up count for K7. All other
1847 * are set to zero, indicating minimum divisors for
1848 * every field.
1849 *
1850 * This prevents guest kernels on AMD host with CPU
1851 * type 6, model 8 and higher from exploding due to
1852 * the rdmsr failing.
1853 */
1854 data = 0x20000000;
1855 break;
1558 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15: 1856 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1559 if (kvm_hv_msr_partition_wide(msr)) { 1857 if (kvm_hv_msr_partition_wide(msr)) {
1560 int r; 1858 int r;
@@ -1565,6 +1863,19 @@ int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1565 } else 1863 } else
1566 return get_msr_hyperv(vcpu, msr, pdata); 1864 return get_msr_hyperv(vcpu, msr, pdata);
1567 break; 1865 break;
1866 case MSR_IA32_BBL_CR_CTL3:
1867 /* This legacy MSR exists but isn't fully documented in current
1868 * silicon. It is however accessed by winxp in very narrow
1869 * scenarios where it sets bit #19, itself documented as
1870 * a "reserved" bit. Best effort attempt to source coherent
1871 * read data here should the balance of the register be
1872 * interpreted by the guest:
1873 *
1874 * L2 cache control register 3: 64GB range, 256KB size,
1875 * enabled, latency 0x1, configured
1876 */
1877 data = 0xbe702111;
1878 break;
1568 default: 1879 default:
1569 if (!ignore_msrs) { 1880 if (!ignore_msrs) {
1570 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr); 1881 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
@@ -1665,6 +1976,7 @@ int kvm_dev_ioctl_check_extension(long ext)
1665 case KVM_CAP_NOP_IO_DELAY: 1976 case KVM_CAP_NOP_IO_DELAY:
1666 case KVM_CAP_MP_STATE: 1977 case KVM_CAP_MP_STATE:
1667 case KVM_CAP_SYNC_MMU: 1978 case KVM_CAP_SYNC_MMU:
1979 case KVM_CAP_USER_NMI:
1668 case KVM_CAP_REINJECT_CONTROL: 1980 case KVM_CAP_REINJECT_CONTROL:
1669 case KVM_CAP_IRQ_INJECT_STATUS: 1981 case KVM_CAP_IRQ_INJECT_STATUS:
1670 case KVM_CAP_ASSIGN_DEV_IRQ: 1982 case KVM_CAP_ASSIGN_DEV_IRQ:
@@ -1683,6 +1995,8 @@ int kvm_dev_ioctl_check_extension(long ext)
1683 case KVM_CAP_DEBUGREGS: 1995 case KVM_CAP_DEBUGREGS:
1684 case KVM_CAP_X86_ROBUST_SINGLESTEP: 1996 case KVM_CAP_X86_ROBUST_SINGLESTEP:
1685 case KVM_CAP_XSAVE: 1997 case KVM_CAP_XSAVE:
1998 case KVM_CAP_ASYNC_PF:
1999 case KVM_CAP_GET_TSC_KHZ:
1686 r = 1; 2000 r = 1;
1687 break; 2001 break;
1688 case KVM_CAP_COALESCED_MMIO: 2002 case KVM_CAP_COALESCED_MMIO:
@@ -1709,6 +2023,9 @@ int kvm_dev_ioctl_check_extension(long ext)
1709 case KVM_CAP_XCRS: 2023 case KVM_CAP_XCRS:
1710 r = cpu_has_xsave; 2024 r = cpu_has_xsave;
1711 break; 2025 break;
2026 case KVM_CAP_TSC_CONTROL:
2027 r = kvm_has_tsc_control;
2028 break;
1712 default: 2029 default:
1713 r = 0; 2030 r = 0;
1714 break; 2031 break;
@@ -1808,19 +2125,33 @@ void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1808 } 2125 }
1809 2126
1810 kvm_x86_ops->vcpu_load(vcpu, cpu); 2127 kvm_x86_ops->vcpu_load(vcpu, cpu);
1811 if (unlikely(per_cpu(cpu_tsc_khz, cpu) == 0)) { 2128 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
1812 unsigned long khz = cpufreq_quick_get(cpu); 2129 /* Make sure TSC doesn't go backwards */
1813 if (!khz) 2130 s64 tsc_delta;
1814 khz = tsc_khz; 2131 u64 tsc;
1815 per_cpu(cpu_tsc_khz, cpu) = khz; 2132
2133 kvm_get_msr(vcpu, MSR_IA32_TSC, &tsc);
2134 tsc_delta = !vcpu->arch.last_guest_tsc ? 0 :
2135 tsc - vcpu->arch.last_guest_tsc;
2136
2137 if (tsc_delta < 0)
2138 mark_tsc_unstable("KVM discovered backwards TSC");
2139 if (check_tsc_unstable()) {
2140 kvm_x86_ops->adjust_tsc_offset(vcpu, -tsc_delta);
2141 vcpu->arch.tsc_catchup = 1;
2142 }
2143 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2144 if (vcpu->cpu != cpu)
2145 kvm_migrate_timers(vcpu);
2146 vcpu->cpu = cpu;
1816 } 2147 }
1817 kvm_request_guest_time_update(vcpu);
1818} 2148}
1819 2149
1820void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) 2150void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1821{ 2151{
1822 kvm_x86_ops->vcpu_put(vcpu); 2152 kvm_x86_ops->vcpu_put(vcpu);
1823 kvm_put_guest_fpu(vcpu); 2153 kvm_put_guest_fpu(vcpu);
2154 kvm_get_msr(vcpu, MSR_IA32_TSC, &vcpu->arch.last_guest_tsc);
1824} 2155}
1825 2156
1826static int is_efer_nx(void) 2157static int is_efer_nx(void)
@@ -1937,6 +2268,11 @@ out:
1937 return r; 2268 return r;
1938} 2269}
1939 2270
2271static void cpuid_mask(u32 *word, int wordnum)
2272{
2273 *word &= boot_cpu_data.x86_capability[wordnum];
2274}
2275
1940static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function, 2276static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1941 u32 index) 2277 u32 index)
1942{ 2278{
@@ -1991,13 +2327,20 @@ static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1991 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ | 2327 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
1992 0 /* Reserved, DCA */ | F(XMM4_1) | 2328 0 /* Reserved, DCA */ | F(XMM4_1) |
1993 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) | 2329 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
1994 0 /* Reserved, AES */ | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX); 2330 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
2331 F(F16C);
1995 /* cpuid 0x80000001.ecx */ 2332 /* cpuid 0x80000001.ecx */
1996 const u32 kvm_supported_word6_x86_features = 2333 const u32 kvm_supported_word6_x86_features =
1997 F(LAHF_LM) | F(CMP_LEGACY) | F(SVM) | 0 /* ExtApicSpace */ | 2334 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
1998 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) | 2335 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
1999 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5) | 2336 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(XOP) |
2000 0 /* SKINIT */ | 0 /* WDT */; 2337 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
2338
2339 /* cpuid 0xC0000001.edx */
2340 const u32 kvm_supported_word5_x86_features =
2341 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
2342 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
2343 F(PMM) | F(PMM_EN);
2001 2344
2002 /* all calls to cpuid_count() should be made on the same cpu */ 2345 /* all calls to cpuid_count() should be made on the same cpu */
2003 get_cpu(); 2346 get_cpu();
@@ -2010,7 +2353,9 @@ static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2010 break; 2353 break;
2011 case 1: 2354 case 1:
2012 entry->edx &= kvm_supported_word0_x86_features; 2355 entry->edx &= kvm_supported_word0_x86_features;
2356 cpuid_mask(&entry->edx, 0);
2013 entry->ecx &= kvm_supported_word4_x86_features; 2357 entry->ecx &= kvm_supported_word4_x86_features;
2358 cpuid_mask(&entry->ecx, 4);
2014 /* we support x2apic emulation even if host does not support 2359 /* we support x2apic emulation even if host does not support
2015 * it since we emulate x2apic in software */ 2360 * it since we emulate x2apic in software */
2016 entry->ecx |= F(X2APIC); 2361 entry->ecx |= F(X2APIC);
@@ -2068,9 +2413,9 @@ static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2068 int i; 2413 int i;
2069 2414
2070 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; 2415 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2071 for (i = 1; *nent < maxnent; ++i) { 2416 for (i = 1; *nent < maxnent && i < 64; ++i) {
2072 if (entry[i - 1].eax == 0 && i != 2) 2417 if (entry[i].eax == 0)
2073 break; 2418 continue;
2074 do_cpuid_1_ent(&entry[i], function, i); 2419 do_cpuid_1_ent(&entry[i], function, i);
2075 entry[i].flags |= 2420 entry[i].flags |=
2076 KVM_CPUID_FLAG_SIGNIFCANT_INDEX; 2421 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
@@ -2091,6 +2436,7 @@ static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2091 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) | 2436 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
2092 (1 << KVM_FEATURE_NOP_IO_DELAY) | 2437 (1 << KVM_FEATURE_NOP_IO_DELAY) |
2093 (1 << KVM_FEATURE_CLOCKSOURCE2) | 2438 (1 << KVM_FEATURE_CLOCKSOURCE2) |
2439 (1 << KVM_FEATURE_ASYNC_PF) |
2094 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT); 2440 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT);
2095 entry->ebx = 0; 2441 entry->ebx = 0;
2096 entry->ecx = 0; 2442 entry->ecx = 0;
@@ -2101,7 +2447,23 @@ static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2101 break; 2447 break;
2102 case 0x80000001: 2448 case 0x80000001:
2103 entry->edx &= kvm_supported_word1_x86_features; 2449 entry->edx &= kvm_supported_word1_x86_features;
2450 cpuid_mask(&entry->edx, 1);
2104 entry->ecx &= kvm_supported_word6_x86_features; 2451 entry->ecx &= kvm_supported_word6_x86_features;
2452 cpuid_mask(&entry->ecx, 6);
2453 break;
2454 /*Add support for Centaur's CPUID instruction*/
2455 case 0xC0000000:
2456 /*Just support up to 0xC0000004 now*/
2457 entry->eax = min(entry->eax, 0xC0000004);
2458 break;
2459 case 0xC0000001:
2460 entry->edx &= kvm_supported_word5_x86_features;
2461 cpuid_mask(&entry->edx, 5);
2462 break;
2463 case 0xC0000002:
2464 case 0xC0000003:
2465 case 0xC0000004:
2466 /*Now nothing to do, reserved for the future*/
2105 break; 2467 break;
2106 } 2468 }
2107 2469
@@ -2149,6 +2511,26 @@ static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
2149 if (nent >= cpuid->nent) 2511 if (nent >= cpuid->nent)
2150 goto out_free; 2512 goto out_free;
2151 2513
2514 /* Add support for Centaur's CPUID instruction. */
2515 if (boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR) {
2516 do_cpuid_ent(&cpuid_entries[nent], 0xC0000000, 0,
2517 &nent, cpuid->nent);
2518
2519 r = -E2BIG;
2520 if (nent >= cpuid->nent)
2521 goto out_free;
2522
2523 limit = cpuid_entries[nent - 1].eax;
2524 for (func = 0xC0000001;
2525 func <= limit && nent < cpuid->nent; ++func)
2526 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2527 &nent, cpuid->nent);
2528
2529 r = -E2BIG;
2530 if (nent >= cpuid->nent)
2531 goto out_free;
2532 }
2533
2152 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_SIGNATURE, 0, &nent, 2534 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_SIGNATURE, 0, &nent,
2153 cpuid->nent); 2535 cpuid->nent);
2154 2536
@@ -2203,6 +2585,7 @@ static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2203 return -ENXIO; 2585 return -ENXIO;
2204 2586
2205 kvm_queue_interrupt(vcpu, irq->irq, false); 2587 kvm_queue_interrupt(vcpu, irq->irq, false);
2588 kvm_make_request(KVM_REQ_EVENT, vcpu);
2206 2589
2207 return 0; 2590 return 0;
2208} 2591}
@@ -2272,9 +2655,6 @@ static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2272 if (mce->status & MCI_STATUS_UC) { 2655 if (mce->status & MCI_STATUS_UC) {
2273 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) || 2656 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2274 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) { 2657 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2275 printk(KERN_DEBUG "kvm: set_mce: "
2276 "injects mce exception while "
2277 "previous one is in progress!\n");
2278 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu); 2658 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2279 return 0; 2659 return 0;
2280 } 2660 }
@@ -2305,6 +2685,7 @@ static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2305 !kvm_exception_is_soft(vcpu->arch.exception.nr); 2685 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2306 events->exception.nr = vcpu->arch.exception.nr; 2686 events->exception.nr = vcpu->arch.exception.nr;
2307 events->exception.has_error_code = vcpu->arch.exception.has_error_code; 2687 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2688 events->exception.pad = 0;
2308 events->exception.error_code = vcpu->arch.exception.error_code; 2689 events->exception.error_code = vcpu->arch.exception.error_code;
2309 2690
2310 events->interrupt.injected = 2691 events->interrupt.injected =
@@ -2318,12 +2699,14 @@ static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2318 events->nmi.injected = vcpu->arch.nmi_injected; 2699 events->nmi.injected = vcpu->arch.nmi_injected;
2319 events->nmi.pending = vcpu->arch.nmi_pending; 2700 events->nmi.pending = vcpu->arch.nmi_pending;
2320 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu); 2701 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2702 events->nmi.pad = 0;
2321 2703
2322 events->sipi_vector = vcpu->arch.sipi_vector; 2704 events->sipi_vector = vcpu->arch.sipi_vector;
2323 2705
2324 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING 2706 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2325 | KVM_VCPUEVENT_VALID_SIPI_VECTOR 2707 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2326 | KVM_VCPUEVENT_VALID_SHADOW); 2708 | KVM_VCPUEVENT_VALID_SHADOW);
2709 memset(&events->reserved, 0, sizeof(events->reserved));
2327} 2710}
2328 2711
2329static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu, 2712static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
@@ -2342,8 +2725,6 @@ static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2342 vcpu->arch.interrupt.pending = events->interrupt.injected; 2725 vcpu->arch.interrupt.pending = events->interrupt.injected;
2343 vcpu->arch.interrupt.nr = events->interrupt.nr; 2726 vcpu->arch.interrupt.nr = events->interrupt.nr;
2344 vcpu->arch.interrupt.soft = events->interrupt.soft; 2727 vcpu->arch.interrupt.soft = events->interrupt.soft;
2345 if (vcpu->arch.interrupt.pending && irqchip_in_kernel(vcpu->kvm))
2346 kvm_pic_clear_isr_ack(vcpu->kvm);
2347 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW) 2728 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2348 kvm_x86_ops->set_interrupt_shadow(vcpu, 2729 kvm_x86_ops->set_interrupt_shadow(vcpu,
2349 events->interrupt.shadow); 2730 events->interrupt.shadow);
@@ -2356,6 +2737,8 @@ static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2356 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR) 2737 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2357 vcpu->arch.sipi_vector = events->sipi_vector; 2738 vcpu->arch.sipi_vector = events->sipi_vector;
2358 2739
2740 kvm_make_request(KVM_REQ_EVENT, vcpu);
2741
2359 return 0; 2742 return 0;
2360} 2743}
2361 2744
@@ -2366,6 +2749,7 @@ static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2366 dbgregs->dr6 = vcpu->arch.dr6; 2749 dbgregs->dr6 = vcpu->arch.dr6;
2367 dbgregs->dr7 = vcpu->arch.dr7; 2750 dbgregs->dr7 = vcpu->arch.dr7;
2368 dbgregs->flags = 0; 2751 dbgregs->flags = 0;
2752 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2369} 2753}
2370 2754
2371static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu, 2755static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
@@ -2715,6 +3099,32 @@ long kvm_arch_vcpu_ioctl(struct file *filp,
2715 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs); 3099 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
2716 break; 3100 break;
2717 } 3101 }
3102 case KVM_SET_TSC_KHZ: {
3103 u32 user_tsc_khz;
3104
3105 r = -EINVAL;
3106 if (!kvm_has_tsc_control)
3107 break;
3108
3109 user_tsc_khz = (u32)arg;
3110
3111 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3112 goto out;
3113
3114 kvm_x86_ops->set_tsc_khz(vcpu, user_tsc_khz);
3115
3116 r = 0;
3117 goto out;
3118 }
3119 case KVM_GET_TSC_KHZ: {
3120 r = -EIO;
3121 if (check_tsc_unstable())
3122 goto out;
3123
3124 r = vcpu_tsc_khz(vcpu);
3125
3126 goto out;
3127 }
2718 default: 3128 default:
2719 r = -EINVAL; 3129 r = -EINVAL;
2720 } 3130 }
@@ -2759,7 +3169,7 @@ static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
2759 3169
2760static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm) 3170static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
2761{ 3171{
2762 return kvm->arch.n_alloc_mmu_pages; 3172 return kvm->arch.n_max_mmu_pages;
2763} 3173}
2764 3174
2765static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip) 3175static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
@@ -2795,18 +3205,18 @@ static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2795 r = 0; 3205 r = 0;
2796 switch (chip->chip_id) { 3206 switch (chip->chip_id) {
2797 case KVM_IRQCHIP_PIC_MASTER: 3207 case KVM_IRQCHIP_PIC_MASTER:
2798 raw_spin_lock(&pic_irqchip(kvm)->lock); 3208 spin_lock(&pic_irqchip(kvm)->lock);
2799 memcpy(&pic_irqchip(kvm)->pics[0], 3209 memcpy(&pic_irqchip(kvm)->pics[0],
2800 &chip->chip.pic, 3210 &chip->chip.pic,
2801 sizeof(struct kvm_pic_state)); 3211 sizeof(struct kvm_pic_state));
2802 raw_spin_unlock(&pic_irqchip(kvm)->lock); 3212 spin_unlock(&pic_irqchip(kvm)->lock);
2803 break; 3213 break;
2804 case KVM_IRQCHIP_PIC_SLAVE: 3214 case KVM_IRQCHIP_PIC_SLAVE:
2805 raw_spin_lock(&pic_irqchip(kvm)->lock); 3215 spin_lock(&pic_irqchip(kvm)->lock);
2806 memcpy(&pic_irqchip(kvm)->pics[1], 3216 memcpy(&pic_irqchip(kvm)->pics[1],
2807 &chip->chip.pic, 3217 &chip->chip.pic,
2808 sizeof(struct kvm_pic_state)); 3218 sizeof(struct kvm_pic_state));
2809 raw_spin_unlock(&pic_irqchip(kvm)->lock); 3219 spin_unlock(&pic_irqchip(kvm)->lock);
2810 break; 3220 break;
2811 case KVM_IRQCHIP_IOAPIC: 3221 case KVM_IRQCHIP_IOAPIC:
2812 r = kvm_set_ioapic(kvm, &chip->chip.ioapic); 3222 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
@@ -2849,6 +3259,7 @@ static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2849 sizeof(ps->channels)); 3259 sizeof(ps->channels));
2850 ps->flags = kvm->arch.vpit->pit_state.flags; 3260 ps->flags = kvm->arch.vpit->pit_state.flags;
2851 mutex_unlock(&kvm->arch.vpit->pit_state.lock); 3261 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3262 memset(&ps->reserved, 0, sizeof(ps->reserved));
2852 return r; 3263 return r;
2853} 3264}
2854 3265
@@ -2912,24 +3323,18 @@ int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2912 struct kvm_memslots *slots, *old_slots; 3323 struct kvm_memslots *slots, *old_slots;
2913 unsigned long *dirty_bitmap; 3324 unsigned long *dirty_bitmap;
2914 3325
2915 spin_lock(&kvm->mmu_lock); 3326 dirty_bitmap = memslot->dirty_bitmap_head;
2916 kvm_mmu_slot_remove_write_access(kvm, log->slot); 3327 if (memslot->dirty_bitmap == dirty_bitmap)
2917 spin_unlock(&kvm->mmu_lock); 3328 dirty_bitmap += n / sizeof(long);
2918
2919 r = -ENOMEM;
2920 dirty_bitmap = vmalloc(n);
2921 if (!dirty_bitmap)
2922 goto out;
2923 memset(dirty_bitmap, 0, n); 3329 memset(dirty_bitmap, 0, n);
2924 3330
2925 r = -ENOMEM; 3331 r = -ENOMEM;
2926 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL); 3332 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
2927 if (!slots) { 3333 if (!slots)
2928 vfree(dirty_bitmap);
2929 goto out; 3334 goto out;
2930 }
2931 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots)); 3335 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
2932 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap; 3336 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
3337 slots->generation++;
2933 3338
2934 old_slots = kvm->memslots; 3339 old_slots = kvm->memslots;
2935 rcu_assign_pointer(kvm->memslots, slots); 3340 rcu_assign_pointer(kvm->memslots, slots);
@@ -2937,12 +3342,13 @@ int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2937 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap; 3342 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
2938 kfree(old_slots); 3343 kfree(old_slots);
2939 3344
3345 spin_lock(&kvm->mmu_lock);
3346 kvm_mmu_slot_remove_write_access(kvm, log->slot);
3347 spin_unlock(&kvm->mmu_lock);
3348
2940 r = -EFAULT; 3349 r = -EFAULT;
2941 if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n)) { 3350 if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
2942 vfree(dirty_bitmap);
2943 goto out; 3351 goto out;
2944 }
2945 vfree(dirty_bitmap);
2946 } else { 3352 } else {
2947 r = -EFAULT; 3353 r = -EFAULT;
2948 if (clear_user(log->dirty_bitmap, n)) 3354 if (clear_user(log->dirty_bitmap, n))
@@ -3009,8 +3415,10 @@ long kvm_arch_vm_ioctl(struct file *filp,
3009 if (vpic) { 3415 if (vpic) {
3010 r = kvm_ioapic_init(kvm); 3416 r = kvm_ioapic_init(kvm);
3011 if (r) { 3417 if (r) {
3418 mutex_lock(&kvm->slots_lock);
3012 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, 3419 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3013 &vpic->dev); 3420 &vpic->dev);
3421 mutex_unlock(&kvm->slots_lock);
3014 kfree(vpic); 3422 kfree(vpic);
3015 goto create_irqchip_unlock; 3423 goto create_irqchip_unlock;
3016 } 3424 }
@@ -3021,10 +3429,12 @@ long kvm_arch_vm_ioctl(struct file *filp,
3021 smp_wmb(); 3429 smp_wmb();
3022 r = kvm_setup_default_irq_routing(kvm); 3430 r = kvm_setup_default_irq_routing(kvm);
3023 if (r) { 3431 if (r) {
3432 mutex_lock(&kvm->slots_lock);
3024 mutex_lock(&kvm->irq_lock); 3433 mutex_lock(&kvm->irq_lock);
3025 kvm_ioapic_destroy(kvm); 3434 kvm_ioapic_destroy(kvm);
3026 kvm_destroy_pic(kvm); 3435 kvm_destroy_pic(kvm);
3027 mutex_unlock(&kvm->irq_lock); 3436 mutex_unlock(&kvm->irq_lock);
3437 mutex_unlock(&kvm->slots_lock);
3028 } 3438 }
3029 create_irqchip_unlock: 3439 create_irqchip_unlock:
3030 mutex_unlock(&kvm->lock); 3440 mutex_unlock(&kvm->lock);
@@ -3200,7 +3610,6 @@ long kvm_arch_vm_ioctl(struct file *filp,
3200 break; 3610 break;
3201 } 3611 }
3202 case KVM_SET_CLOCK: { 3612 case KVM_SET_CLOCK: {
3203 struct timespec now;
3204 struct kvm_clock_data user_ns; 3613 struct kvm_clock_data user_ns;
3205 u64 now_ns; 3614 u64 now_ns;
3206 s64 delta; 3615 s64 delta;
@@ -3214,21 +3623,23 @@ long kvm_arch_vm_ioctl(struct file *filp,
3214 goto out; 3623 goto out;
3215 3624
3216 r = 0; 3625 r = 0;
3217 ktime_get_ts(&now); 3626 local_irq_disable();
3218 now_ns = timespec_to_ns(&now); 3627 now_ns = get_kernel_ns();
3219 delta = user_ns.clock - now_ns; 3628 delta = user_ns.clock - now_ns;
3629 local_irq_enable();
3220 kvm->arch.kvmclock_offset = delta; 3630 kvm->arch.kvmclock_offset = delta;
3221 break; 3631 break;
3222 } 3632 }
3223 case KVM_GET_CLOCK: { 3633 case KVM_GET_CLOCK: {
3224 struct timespec now;
3225 struct kvm_clock_data user_ns; 3634 struct kvm_clock_data user_ns;
3226 u64 now_ns; 3635 u64 now_ns;
3227 3636
3228 ktime_get_ts(&now); 3637 local_irq_disable();
3229 now_ns = timespec_to_ns(&now); 3638 now_ns = get_kernel_ns();
3230 user_ns.clock = kvm->arch.kvmclock_offset + now_ns; 3639 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3640 local_irq_enable();
3231 user_ns.flags = 0; 3641 user_ns.flags = 0;
3642 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3232 3643
3233 r = -EFAULT; 3644 r = -EFAULT;
3234 if (copy_to_user(argp, &user_ns, sizeof(user_ns))) 3645 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
@@ -3263,20 +3674,43 @@ static void kvm_init_msr_list(void)
3263static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len, 3674static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3264 const void *v) 3675 const void *v)
3265{ 3676{
3266 if (vcpu->arch.apic && 3677 int handled = 0;
3267 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, len, v)) 3678 int n;
3268 return 0; 3679
3680 do {
3681 n = min(len, 8);
3682 if (!(vcpu->arch.apic &&
3683 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3684 && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3685 break;
3686 handled += n;
3687 addr += n;
3688 len -= n;
3689 v += n;
3690 } while (len);
3269 3691
3270 return kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, len, v); 3692 return handled;
3271} 3693}
3272 3694
3273static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v) 3695static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3274{ 3696{
3275 if (vcpu->arch.apic && 3697 int handled = 0;
3276 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, len, v)) 3698 int n;
3277 return 0; 3699
3700 do {
3701 n = min(len, 8);
3702 if (!(vcpu->arch.apic &&
3703 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
3704 && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3705 break;
3706 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3707 handled += n;
3708 addr += n;
3709 len -= n;
3710 v += n;
3711 } while (len);
3278 3712
3279 return kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, len, v); 3713 return handled;
3280} 3714}
3281 3715
3282static void kvm_set_segment(struct kvm_vcpu *vcpu, 3716static void kvm_set_segment(struct kvm_vcpu *vcpu,
@@ -3291,49 +3725,71 @@ void kvm_get_segment(struct kvm_vcpu *vcpu,
3291 kvm_x86_ops->get_segment(vcpu, var, seg); 3725 kvm_x86_ops->get_segment(vcpu, var, seg);
3292} 3726}
3293 3727
3294gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva, u32 *error) 3728static gpa_t translate_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3729{
3730 return gpa;
3731}
3732
3733static gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3734{
3735 gpa_t t_gpa;
3736 struct x86_exception exception;
3737
3738 BUG_ON(!mmu_is_nested(vcpu));
3739
3740 /* NPT walks are always user-walks */
3741 access |= PFERR_USER_MASK;
3742 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
3743
3744 return t_gpa;
3745}
3746
3747gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3748 struct x86_exception *exception)
3295{ 3749{
3296 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0; 3750 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3297 return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, access, error); 3751 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3298} 3752}
3299 3753
3300 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva, u32 *error) 3754 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3755 struct x86_exception *exception)
3301{ 3756{
3302 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0; 3757 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3303 access |= PFERR_FETCH_MASK; 3758 access |= PFERR_FETCH_MASK;
3304 return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, access, error); 3759 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3305} 3760}
3306 3761
3307gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva, u32 *error) 3762gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3763 struct x86_exception *exception)
3308{ 3764{
3309 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0; 3765 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3310 access |= PFERR_WRITE_MASK; 3766 access |= PFERR_WRITE_MASK;
3311 return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, access, error); 3767 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3312} 3768}
3313 3769
3314/* uses this to access any guest's mapped memory without checking CPL */ 3770/* uses this to access any guest's mapped memory without checking CPL */
3315gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva, u32 *error) 3771gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3772 struct x86_exception *exception)
3316{ 3773{
3317 return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, 0, error); 3774 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
3318} 3775}
3319 3776
3320static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes, 3777static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3321 struct kvm_vcpu *vcpu, u32 access, 3778 struct kvm_vcpu *vcpu, u32 access,
3322 u32 *error) 3779 struct x86_exception *exception)
3323{ 3780{
3324 void *data = val; 3781 void *data = val;
3325 int r = X86EMUL_CONTINUE; 3782 int r = X86EMUL_CONTINUE;
3326 3783
3327 while (bytes) { 3784 while (bytes) {
3328 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr, access, error); 3785 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3786 exception);
3329 unsigned offset = addr & (PAGE_SIZE-1); 3787 unsigned offset = addr & (PAGE_SIZE-1);
3330 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset); 3788 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3331 int ret; 3789 int ret;
3332 3790
3333 if (gpa == UNMAPPED_GVA) { 3791 if (gpa == UNMAPPED_GVA)
3334 r = X86EMUL_PROPAGATE_FAULT; 3792 return X86EMUL_PROPAGATE_FAULT;
3335 goto out;
3336 }
3337 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread); 3793 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3338 if (ret < 0) { 3794 if (ret < 0) {
3339 r = X86EMUL_IO_NEEDED; 3795 r = X86EMUL_IO_NEEDED;
@@ -3349,47 +3805,56 @@ out:
3349} 3805}
3350 3806
3351/* used for instruction fetching */ 3807/* used for instruction fetching */
3352static int kvm_fetch_guest_virt(gva_t addr, void *val, unsigned int bytes, 3808static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
3353 struct kvm_vcpu *vcpu, u32 *error) 3809 gva_t addr, void *val, unsigned int bytes,
3810 struct x86_exception *exception)
3354{ 3811{
3812 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3355 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0; 3813 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3814
3356 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 3815 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3357 access | PFERR_FETCH_MASK, error); 3816 access | PFERR_FETCH_MASK,
3817 exception);
3358} 3818}
3359 3819
3360static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes, 3820static int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
3361 struct kvm_vcpu *vcpu, u32 *error) 3821 gva_t addr, void *val, unsigned int bytes,
3822 struct x86_exception *exception)
3362{ 3823{
3824 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3363 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0; 3825 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3826
3364 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, 3827 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3365 error); 3828 exception);
3366} 3829}
3367 3830
3368static int kvm_read_guest_virt_system(gva_t addr, void *val, unsigned int bytes, 3831static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3369 struct kvm_vcpu *vcpu, u32 *error) 3832 gva_t addr, void *val, unsigned int bytes,
3833 struct x86_exception *exception)
3370{ 3834{
3371 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, error); 3835 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3836 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
3372} 3837}
3373 3838
3374static int kvm_write_guest_virt_system(gva_t addr, void *val, 3839static int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3840 gva_t addr, void *val,
3375 unsigned int bytes, 3841 unsigned int bytes,
3376 struct kvm_vcpu *vcpu, 3842 struct x86_exception *exception)
3377 u32 *error)
3378{ 3843{
3844 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3379 void *data = val; 3845 void *data = val;
3380 int r = X86EMUL_CONTINUE; 3846 int r = X86EMUL_CONTINUE;
3381 3847
3382 while (bytes) { 3848 while (bytes) {
3383 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr, 3849 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
3384 PFERR_WRITE_MASK, error); 3850 PFERR_WRITE_MASK,
3851 exception);
3385 unsigned offset = addr & (PAGE_SIZE-1); 3852 unsigned offset = addr & (PAGE_SIZE-1);
3386 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset); 3853 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3387 int ret; 3854 int ret;
3388 3855
3389 if (gpa == UNMAPPED_GVA) { 3856 if (gpa == UNMAPPED_GVA)
3390 r = X86EMUL_PROPAGATE_FAULT; 3857 return X86EMUL_PROPAGATE_FAULT;
3391 goto out;
3392 }
3393 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite); 3858 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3394 if (ret < 0) { 3859 if (ret < 0) {
3395 r = X86EMUL_IO_NEEDED; 3860 r = X86EMUL_IO_NEEDED;
@@ -3404,13 +3869,15 @@ out:
3404 return r; 3869 return r;
3405} 3870}
3406 3871
3407static int emulator_read_emulated(unsigned long addr, 3872static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
3873 unsigned long addr,
3408 void *val, 3874 void *val,
3409 unsigned int bytes, 3875 unsigned int bytes,
3410 unsigned int *error_code, 3876 struct x86_exception *exception)
3411 struct kvm_vcpu *vcpu)
3412{ 3877{
3878 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3413 gpa_t gpa; 3879 gpa_t gpa;
3880 int handled;
3414 3881
3415 if (vcpu->mmio_read_completed) { 3882 if (vcpu->mmio_read_completed) {
3416 memcpy(val, vcpu->mmio_data, bytes); 3883 memcpy(val, vcpu->mmio_data, bytes);
@@ -3420,7 +3887,7 @@ static int emulator_read_emulated(unsigned long addr,
3420 return X86EMUL_CONTINUE; 3887 return X86EMUL_CONTINUE;
3421 } 3888 }
3422 3889
3423 gpa = kvm_mmu_gva_to_gpa_read(vcpu, addr, error_code); 3890 gpa = kvm_mmu_gva_to_gpa_read(vcpu, addr, exception);
3424 3891
3425 if (gpa == UNMAPPED_GVA) 3892 if (gpa == UNMAPPED_GVA)
3426 return X86EMUL_PROPAGATE_FAULT; 3893 return X86EMUL_PROPAGATE_FAULT;
@@ -3429,32 +3896,38 @@ static int emulator_read_emulated(unsigned long addr,
3429 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE) 3896 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3430 goto mmio; 3897 goto mmio;
3431 3898
3432 if (kvm_read_guest_virt(addr, val, bytes, vcpu, NULL) 3899 if (kvm_read_guest_virt(ctxt, addr, val, bytes, exception)
3433 == X86EMUL_CONTINUE) 3900 == X86EMUL_CONTINUE)
3434 return X86EMUL_CONTINUE; 3901 return X86EMUL_CONTINUE;
3435 3902
3436mmio: 3903mmio:
3437 /* 3904 /*
3438 * Is this MMIO handled locally? 3905 * Is this MMIO handled locally?
3439 */ 3906 */
3440 if (!vcpu_mmio_read(vcpu, gpa, bytes, val)) { 3907 handled = vcpu_mmio_read(vcpu, gpa, bytes, val);
3441 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, gpa, *(u64 *)val); 3908
3909 if (handled == bytes)
3442 return X86EMUL_CONTINUE; 3910 return X86EMUL_CONTINUE;
3443 } 3911
3912 gpa += handled;
3913 bytes -= handled;
3914 val += handled;
3444 3915
3445 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0); 3916 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
3446 3917
3447 vcpu->mmio_needed = 1; 3918 vcpu->mmio_needed = 1;
3448 vcpu->run->exit_reason = KVM_EXIT_MMIO; 3919 vcpu->run->exit_reason = KVM_EXIT_MMIO;
3449 vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa; 3920 vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3450 vcpu->run->mmio.len = vcpu->mmio_size = bytes; 3921 vcpu->mmio_size = bytes;
3922 vcpu->run->mmio.len = min(vcpu->mmio_size, 8);
3451 vcpu->run->mmio.is_write = vcpu->mmio_is_write = 0; 3923 vcpu->run->mmio.is_write = vcpu->mmio_is_write = 0;
3924 vcpu->mmio_index = 0;
3452 3925
3453 return X86EMUL_IO_NEEDED; 3926 return X86EMUL_IO_NEEDED;
3454} 3927}
3455 3928
3456int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa, 3929int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
3457 const void *val, int bytes) 3930 const void *val, int bytes)
3458{ 3931{
3459 int ret; 3932 int ret;
3460 3933
@@ -3468,12 +3941,13 @@ int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
3468static int emulator_write_emulated_onepage(unsigned long addr, 3941static int emulator_write_emulated_onepage(unsigned long addr,
3469 const void *val, 3942 const void *val,
3470 unsigned int bytes, 3943 unsigned int bytes,
3471 unsigned int *error_code, 3944 struct x86_exception *exception,
3472 struct kvm_vcpu *vcpu) 3945 struct kvm_vcpu *vcpu)
3473{ 3946{
3474 gpa_t gpa; 3947 gpa_t gpa;
3948 int handled;
3475 3949
3476 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, error_code); 3950 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, exception);
3477 3951
3478 if (gpa == UNMAPPED_GVA) 3952 if (gpa == UNMAPPED_GVA)
3479 return X86EMUL_PROPAGATE_FAULT; 3953 return X86EMUL_PROPAGATE_FAULT;
@@ -3490,31 +3964,41 @@ mmio:
3490 /* 3964 /*
3491 * Is this MMIO handled locally? 3965 * Is this MMIO handled locally?
3492 */ 3966 */
3493 if (!vcpu_mmio_write(vcpu, gpa, bytes, val)) 3967 handled = vcpu_mmio_write(vcpu, gpa, bytes, val);
3968 if (handled == bytes)
3494 return X86EMUL_CONTINUE; 3969 return X86EMUL_CONTINUE;
3495 3970
3971 gpa += handled;
3972 bytes -= handled;
3973 val += handled;
3974
3496 vcpu->mmio_needed = 1; 3975 vcpu->mmio_needed = 1;
3976 memcpy(vcpu->mmio_data, val, bytes);
3497 vcpu->run->exit_reason = KVM_EXIT_MMIO; 3977 vcpu->run->exit_reason = KVM_EXIT_MMIO;
3498 vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa; 3978 vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3499 vcpu->run->mmio.len = vcpu->mmio_size = bytes; 3979 vcpu->mmio_size = bytes;
3980 vcpu->run->mmio.len = min(vcpu->mmio_size, 8);
3500 vcpu->run->mmio.is_write = vcpu->mmio_is_write = 1; 3981 vcpu->run->mmio.is_write = vcpu->mmio_is_write = 1;
3501 memcpy(vcpu->run->mmio.data, val, bytes); 3982 memcpy(vcpu->run->mmio.data, vcpu->mmio_data, 8);
3983 vcpu->mmio_index = 0;
3502 3984
3503 return X86EMUL_CONTINUE; 3985 return X86EMUL_CONTINUE;
3504} 3986}
3505 3987
3506int emulator_write_emulated(unsigned long addr, 3988int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
3989 unsigned long addr,
3507 const void *val, 3990 const void *val,
3508 unsigned int bytes, 3991 unsigned int bytes,
3509 unsigned int *error_code, 3992 struct x86_exception *exception)
3510 struct kvm_vcpu *vcpu)
3511{ 3993{
3994 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3995
3512 /* Crossing a page boundary? */ 3996 /* Crossing a page boundary? */
3513 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) { 3997 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
3514 int rc, now; 3998 int rc, now;
3515 3999
3516 now = -addr & ~PAGE_MASK; 4000 now = -addr & ~PAGE_MASK;
3517 rc = emulator_write_emulated_onepage(addr, val, now, error_code, 4001 rc = emulator_write_emulated_onepage(addr, val, now, exception,
3518 vcpu); 4002 vcpu);
3519 if (rc != X86EMUL_CONTINUE) 4003 if (rc != X86EMUL_CONTINUE)
3520 return rc; 4004 return rc;
@@ -3522,7 +4006,7 @@ int emulator_write_emulated(unsigned long addr,
3522 val += now; 4006 val += now;
3523 bytes -= now; 4007 bytes -= now;
3524 } 4008 }
3525 return emulator_write_emulated_onepage(addr, val, bytes, error_code, 4009 return emulator_write_emulated_onepage(addr, val, bytes, exception,
3526 vcpu); 4010 vcpu);
3527} 4011}
3528 4012
@@ -3536,13 +4020,14 @@ int emulator_write_emulated(unsigned long addr,
3536 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old)) 4020 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
3537#endif 4021#endif
3538 4022
3539static int emulator_cmpxchg_emulated(unsigned long addr, 4023static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4024 unsigned long addr,
3540 const void *old, 4025 const void *old,
3541 const void *new, 4026 const void *new,
3542 unsigned int bytes, 4027 unsigned int bytes,
3543 unsigned int *error_code, 4028 struct x86_exception *exception)
3544 struct kvm_vcpu *vcpu)
3545{ 4029{
4030 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3546 gpa_t gpa; 4031 gpa_t gpa;
3547 struct page *page; 4032 struct page *page;
3548 char *kaddr; 4033 char *kaddr;
@@ -3598,7 +4083,7 @@ static int emulator_cmpxchg_emulated(unsigned long addr,
3598emul_write: 4083emul_write:
3599 printk_once(KERN_WARNING "kvm: emulating exchange as write\n"); 4084 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
3600 4085
3601 return emulator_write_emulated(addr, new, bytes, error_code, vcpu); 4086 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
3602} 4087}
3603 4088
3604static int kernel_pio(struct kvm_vcpu *vcpu, void *pd) 4089static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
@@ -3617,13 +4102,16 @@ static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
3617} 4102}
3618 4103
3619 4104
3620static int emulator_pio_in_emulated(int size, unsigned short port, void *val, 4105static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
3621 unsigned int count, struct kvm_vcpu *vcpu) 4106 int size, unsigned short port, void *val,
4107 unsigned int count)
3622{ 4108{
4109 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4110
3623 if (vcpu->arch.pio.count) 4111 if (vcpu->arch.pio.count)
3624 goto data_avail; 4112 goto data_avail;
3625 4113
3626 trace_kvm_pio(1, port, size, 1); 4114 trace_kvm_pio(0, port, size, count);
3627 4115
3628 vcpu->arch.pio.port = port; 4116 vcpu->arch.pio.port = port;
3629 vcpu->arch.pio.in = 1; 4117 vcpu->arch.pio.in = 1;
@@ -3647,11 +4135,13 @@ static int emulator_pio_in_emulated(int size, unsigned short port, void *val,
3647 return 0; 4135 return 0;
3648} 4136}
3649 4137
3650static int emulator_pio_out_emulated(int size, unsigned short port, 4138static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
3651 const void *val, unsigned int count, 4139 int size, unsigned short port,
3652 struct kvm_vcpu *vcpu) 4140 const void *val, unsigned int count)
3653{ 4141{
3654 trace_kvm_pio(0, port, size, 1); 4142 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4143
4144 trace_kvm_pio(1, port, size, count);
3655 4145
3656 vcpu->arch.pio.port = port; 4146 vcpu->arch.pio.port = port;
3657 vcpu->arch.pio.in = 0; 4147 vcpu->arch.pio.in = 0;
@@ -3680,10 +4170,9 @@ static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
3680 return kvm_x86_ops->get_segment_base(vcpu, seg); 4170 return kvm_x86_ops->get_segment_base(vcpu, seg);
3681} 4171}
3682 4172
3683int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address) 4173static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
3684{ 4174{
3685 kvm_mmu_invlpg(vcpu, address); 4175 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
3686 return X86EMUL_CONTINUE;
3687} 4176}
3688 4177
3689int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu) 4178int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
@@ -3692,31 +4181,33 @@ int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
3692 return X86EMUL_CONTINUE; 4181 return X86EMUL_CONTINUE;
3693 4182
3694 if (kvm_x86_ops->has_wbinvd_exit()) { 4183 if (kvm_x86_ops->has_wbinvd_exit()) {
4184 int cpu = get_cpu();
4185
4186 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3695 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask, 4187 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
3696 wbinvd_ipi, NULL, 1); 4188 wbinvd_ipi, NULL, 1);
4189 put_cpu();
3697 cpumask_clear(vcpu->arch.wbinvd_dirty_mask); 4190 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
3698 } 4191 } else
3699 wbinvd(); 4192 wbinvd();
3700 return X86EMUL_CONTINUE; 4193 return X86EMUL_CONTINUE;
3701} 4194}
3702EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd); 4195EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
3703 4196
3704int emulate_clts(struct kvm_vcpu *vcpu) 4197static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
3705{ 4198{
3706 kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS)); 4199 kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
3707 kvm_x86_ops->fpu_activate(vcpu);
3708 return X86EMUL_CONTINUE;
3709} 4200}
3710 4201
3711int emulator_get_dr(int dr, unsigned long *dest, struct kvm_vcpu *vcpu) 4202int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
3712{ 4203{
3713 return _kvm_get_dr(vcpu, dr, dest); 4204 return _kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
3714} 4205}
3715 4206
3716int emulator_set_dr(int dr, unsigned long value, struct kvm_vcpu *vcpu) 4207int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
3717{ 4208{
3718 4209
3719 return __kvm_set_dr(vcpu, dr, value); 4210 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
3720} 4211}
3721 4212
3722static u64 mk_cr_64(u64 curr_cr, u32 new_val) 4213static u64 mk_cr_64(u64 curr_cr, u32 new_val)
@@ -3724,8 +4215,9 @@ static u64 mk_cr_64(u64 curr_cr, u32 new_val)
3724 return (curr_cr & ~((1ULL << 32) - 1)) | new_val; 4215 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
3725} 4216}
3726 4217
3727static unsigned long emulator_get_cr(int cr, struct kvm_vcpu *vcpu) 4218static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
3728{ 4219{
4220 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3729 unsigned long value; 4221 unsigned long value;
3730 4222
3731 switch (cr) { 4223 switch (cr) {
@@ -3736,7 +4228,7 @@ static unsigned long emulator_get_cr(int cr, struct kvm_vcpu *vcpu)
3736 value = vcpu->arch.cr2; 4228 value = vcpu->arch.cr2;
3737 break; 4229 break;
3738 case 3: 4230 case 3:
3739 value = vcpu->arch.cr3; 4231 value = kvm_read_cr3(vcpu);
3740 break; 4232 break;
3741 case 4: 4233 case 4:
3742 value = kvm_read_cr4(vcpu); 4234 value = kvm_read_cr4(vcpu);
@@ -3752,8 +4244,9 @@ static unsigned long emulator_get_cr(int cr, struct kvm_vcpu *vcpu)
3752 return value; 4244 return value;
3753} 4245}
3754 4246
3755static int emulator_set_cr(int cr, unsigned long val, struct kvm_vcpu *vcpu) 4247static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
3756{ 4248{
4249 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3757 int res = 0; 4250 int res = 0;
3758 4251
3759 switch (cr) { 4252 switch (cr) {
@@ -3770,7 +4263,7 @@ static int emulator_set_cr(int cr, unsigned long val, struct kvm_vcpu *vcpu)
3770 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val)); 4263 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
3771 break; 4264 break;
3772 case 8: 4265 case 8:
3773 res = __kvm_set_cr8(vcpu, val & 0xfUL); 4266 res = kvm_set_cr8(vcpu, val);
3774 break; 4267 break;
3775 default: 4268 default:
3776 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr); 4269 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
@@ -3780,28 +4273,45 @@ static int emulator_set_cr(int cr, unsigned long val, struct kvm_vcpu *vcpu)
3780 return res; 4273 return res;
3781} 4274}
3782 4275
3783static int emulator_get_cpl(struct kvm_vcpu *vcpu) 4276static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
3784{ 4277{
3785 return kvm_x86_ops->get_cpl(vcpu); 4278 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
3786} 4279}
3787 4280
3788static void emulator_get_gdt(struct desc_ptr *dt, struct kvm_vcpu *vcpu) 4281static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
3789{ 4282{
3790 kvm_x86_ops->get_gdt(vcpu, dt); 4283 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
3791} 4284}
3792 4285
3793static unsigned long emulator_get_cached_segment_base(int seg, 4286static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
3794 struct kvm_vcpu *vcpu)
3795{ 4287{
3796 return get_segment_base(vcpu, seg); 4288 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
3797} 4289}
3798 4290
3799static bool emulator_get_cached_descriptor(struct desc_struct *desc, int seg, 4291static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
3800 struct kvm_vcpu *vcpu) 4292{
4293 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4294}
4295
4296static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4297{
4298 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4299}
4300
4301static unsigned long emulator_get_cached_segment_base(
4302 struct x86_emulate_ctxt *ctxt, int seg)
4303{
4304 return get_segment_base(emul_to_vcpu(ctxt), seg);
4305}
4306
4307static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4308 struct desc_struct *desc, u32 *base3,
4309 int seg)
3801{ 4310{
3802 struct kvm_segment var; 4311 struct kvm_segment var;
3803 4312
3804 kvm_get_segment(vcpu, &var, seg); 4313 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4314 *selector = var.selector;
3805 4315
3806 if (var.unusable) 4316 if (var.unusable)
3807 return false; 4317 return false;
@@ -3810,6 +4320,10 @@ static bool emulator_get_cached_descriptor(struct desc_struct *desc, int seg,
3810 var.limit >>= 12; 4320 var.limit >>= 12;
3811 set_desc_limit(desc, var.limit); 4321 set_desc_limit(desc, var.limit);
3812 set_desc_base(desc, (unsigned long)var.base); 4322 set_desc_base(desc, (unsigned long)var.base);
4323#ifdef CONFIG_X86_64
4324 if (base3)
4325 *base3 = var.base >> 32;
4326#endif
3813 desc->type = var.type; 4327 desc->type = var.type;
3814 desc->s = var.s; 4328 desc->s = var.s;
3815 desc->dpl = var.dpl; 4329 desc->dpl = var.dpl;
@@ -3822,15 +4336,18 @@ static bool emulator_get_cached_descriptor(struct desc_struct *desc, int seg,
3822 return true; 4336 return true;
3823} 4337}
3824 4338
3825static void emulator_set_cached_descriptor(struct desc_struct *desc, int seg, 4339static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
3826 struct kvm_vcpu *vcpu) 4340 struct desc_struct *desc, u32 base3,
4341 int seg)
3827{ 4342{
4343 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3828 struct kvm_segment var; 4344 struct kvm_segment var;
3829 4345
3830 /* needed to preserve selector */ 4346 var.selector = selector;
3831 kvm_get_segment(vcpu, &var, seg);
3832
3833 var.base = get_desc_base(desc); 4347 var.base = get_desc_base(desc);
4348#ifdef CONFIG_X86_64
4349 var.base |= ((u64)base3) << 32;
4350#endif
3834 var.limit = get_desc_limit(desc); 4351 var.limit = get_desc_limit(desc);
3835 if (desc->g) 4352 if (desc->g)
3836 var.limit = (var.limit << 12) | 0xfff; 4353 var.limit = (var.limit << 12) | 0xfff;
@@ -3850,22 +4367,44 @@ static void emulator_set_cached_descriptor(struct desc_struct *desc, int seg,
3850 return; 4367 return;
3851} 4368}
3852 4369
3853static u16 emulator_get_segment_selector(int seg, struct kvm_vcpu *vcpu) 4370static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4371 u32 msr_index, u64 *pdata)
3854{ 4372{
3855 struct kvm_segment kvm_seg; 4373 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4374}
3856 4375
3857 kvm_get_segment(vcpu, &kvm_seg, seg); 4376static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
3858 return kvm_seg.selector; 4377 u32 msr_index, u64 data)
4378{
4379 return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
4380}
4381
4382static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4383{
4384 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4385}
4386
4387static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4388{
4389 preempt_disable();
4390 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4391 /*
4392 * CR0.TS may reference the host fpu state, not the guest fpu state,
4393 * so it may be clear at this point.
4394 */
4395 clts();
3859} 4396}
3860 4397
3861static void emulator_set_segment_selector(u16 sel, int seg, 4398static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
3862 struct kvm_vcpu *vcpu)
3863{ 4399{
3864 struct kvm_segment kvm_seg; 4400 preempt_enable();
4401}
3865 4402
3866 kvm_get_segment(vcpu, &kvm_seg, seg); 4403static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
3867 kvm_seg.selector = sel; 4404 struct x86_instruction_info *info,
3868 kvm_set_segment(vcpu, &kvm_seg, seg); 4405 enum x86_intercept_stage stage)
4406{
4407 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
3869} 4408}
3870 4409
3871static struct x86_emulate_ops emulate_ops = { 4410static struct x86_emulate_ops emulate_ops = {
@@ -3875,21 +4414,29 @@ static struct x86_emulate_ops emulate_ops = {
3875 .read_emulated = emulator_read_emulated, 4414 .read_emulated = emulator_read_emulated,
3876 .write_emulated = emulator_write_emulated, 4415 .write_emulated = emulator_write_emulated,
3877 .cmpxchg_emulated = emulator_cmpxchg_emulated, 4416 .cmpxchg_emulated = emulator_cmpxchg_emulated,
4417 .invlpg = emulator_invlpg,
3878 .pio_in_emulated = emulator_pio_in_emulated, 4418 .pio_in_emulated = emulator_pio_in_emulated,
3879 .pio_out_emulated = emulator_pio_out_emulated, 4419 .pio_out_emulated = emulator_pio_out_emulated,
3880 .get_cached_descriptor = emulator_get_cached_descriptor, 4420 .get_segment = emulator_get_segment,
3881 .set_cached_descriptor = emulator_set_cached_descriptor, 4421 .set_segment = emulator_set_segment,
3882 .get_segment_selector = emulator_get_segment_selector,
3883 .set_segment_selector = emulator_set_segment_selector,
3884 .get_cached_segment_base = emulator_get_cached_segment_base, 4422 .get_cached_segment_base = emulator_get_cached_segment_base,
3885 .get_gdt = emulator_get_gdt, 4423 .get_gdt = emulator_get_gdt,
4424 .get_idt = emulator_get_idt,
4425 .set_gdt = emulator_set_gdt,
4426 .set_idt = emulator_set_idt,
3886 .get_cr = emulator_get_cr, 4427 .get_cr = emulator_get_cr,
3887 .set_cr = emulator_set_cr, 4428 .set_cr = emulator_set_cr,
3888 .cpl = emulator_get_cpl, 4429 .cpl = emulator_get_cpl,
3889 .get_dr = emulator_get_dr, 4430 .get_dr = emulator_get_dr,
3890 .set_dr = emulator_set_dr, 4431 .set_dr = emulator_set_dr,
3891 .set_msr = kvm_set_msr, 4432 .set_msr = emulator_set_msr,
3892 .get_msr = kvm_get_msr, 4433 .get_msr = emulator_get_msr,
4434 .halt = emulator_halt,
4435 .wbinvd = emulator_wbinvd,
4436 .fix_hypercall = emulator_fix_hypercall,
4437 .get_fpu = emulator_get_fpu,
4438 .put_fpu = emulator_put_fpu,
4439 .intercept = emulator_intercept,
3893}; 4440};
3894 4441
3895static void cache_all_regs(struct kvm_vcpu *vcpu) 4442static void cache_all_regs(struct kvm_vcpu *vcpu)
@@ -3917,23 +4464,89 @@ static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
3917static void inject_emulated_exception(struct kvm_vcpu *vcpu) 4464static void inject_emulated_exception(struct kvm_vcpu *vcpu)
3918{ 4465{
3919 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt; 4466 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
3920 if (ctxt->exception == PF_VECTOR) 4467 if (ctxt->exception.vector == PF_VECTOR)
3921 kvm_inject_page_fault(vcpu, ctxt->cr2, ctxt->error_code); 4468 kvm_propagate_fault(vcpu, &ctxt->exception);
3922 else if (ctxt->error_code_valid) 4469 else if (ctxt->exception.error_code_valid)
3923 kvm_queue_exception_e(vcpu, ctxt->exception, ctxt->error_code); 4470 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4471 ctxt->exception.error_code);
4472 else
4473 kvm_queue_exception(vcpu, ctxt->exception.vector);
4474}
4475
4476static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4477{
4478 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4479 int cs_db, cs_l;
4480
4481 /*
4482 * TODO: fix emulate.c to use guest_read/write_register
4483 * instead of direct ->regs accesses, can save hundred cycles
4484 * on Intel for instructions that don't read/change RSP, for
4485 * for example.
4486 */
4487 cache_all_regs(vcpu);
4488
4489 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4490
4491 vcpu->arch.emulate_ctxt.eflags = kvm_get_rflags(vcpu);
4492 vcpu->arch.emulate_ctxt.eip = kvm_rip_read(vcpu);
4493 vcpu->arch.emulate_ctxt.mode =
4494 (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4495 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
4496 ? X86EMUL_MODE_VM86 : cs_l
4497 ? X86EMUL_MODE_PROT64 : cs_db
4498 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
4499 vcpu->arch.emulate_ctxt.guest_mode = is_guest_mode(vcpu);
4500 memset(c, 0, sizeof(struct decode_cache));
4501 memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
4502 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4503}
4504
4505int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
4506{
4507 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4508 int ret;
4509
4510 init_emulate_ctxt(vcpu);
4511
4512 vcpu->arch.emulate_ctxt.decode.op_bytes = 2;
4513 vcpu->arch.emulate_ctxt.decode.ad_bytes = 2;
4514 vcpu->arch.emulate_ctxt.decode.eip = vcpu->arch.emulate_ctxt.eip +
4515 inc_eip;
4516 ret = emulate_int_real(&vcpu->arch.emulate_ctxt, &emulate_ops, irq);
4517
4518 if (ret != X86EMUL_CONTINUE)
4519 return EMULATE_FAIL;
4520
4521 vcpu->arch.emulate_ctxt.eip = c->eip;
4522 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
4523 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
4524 kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
4525
4526 if (irq == NMI_VECTOR)
4527 vcpu->arch.nmi_pending = false;
3924 else 4528 else
3925 kvm_queue_exception(vcpu, ctxt->exception); 4529 vcpu->arch.interrupt.pending = false;
4530
4531 return EMULATE_DONE;
3926} 4532}
4533EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
3927 4534
3928static int handle_emulation_failure(struct kvm_vcpu *vcpu) 4535static int handle_emulation_failure(struct kvm_vcpu *vcpu)
3929{ 4536{
4537 int r = EMULATE_DONE;
4538
3930 ++vcpu->stat.insn_emulation_fail; 4539 ++vcpu->stat.insn_emulation_fail;
3931 trace_kvm_emulate_insn_failed(vcpu); 4540 trace_kvm_emulate_insn_failed(vcpu);
3932 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR; 4541 if (!is_guest_mode(vcpu)) {
3933 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION; 4542 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3934 vcpu->run->internal.ndata = 0; 4543 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4544 vcpu->run->internal.ndata = 0;
4545 r = EMULATE_FAIL;
4546 }
3935 kvm_queue_exception(vcpu, UD_VECTOR); 4547 kvm_queue_exception(vcpu, UD_VECTOR);
3936 return EMULATE_FAIL; 4548
4549 return r;
3937} 4550}
3938 4551
3939static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva) 4552static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva)
@@ -3962,74 +4575,34 @@ static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva)
3962 return false; 4575 return false;
3963} 4576}
3964 4577
3965int emulate_instruction(struct kvm_vcpu *vcpu, 4578int x86_emulate_instruction(struct kvm_vcpu *vcpu,
3966 unsigned long cr2, 4579 unsigned long cr2,
3967 u16 error_code, 4580 int emulation_type,
3968 int emulation_type) 4581 void *insn,
4582 int insn_len)
3969{ 4583{
3970 int r; 4584 int r;
3971 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode; 4585 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4586 bool writeback = true;
3972 4587
3973 kvm_clear_exception_queue(vcpu); 4588 kvm_clear_exception_queue(vcpu);
3974 vcpu->arch.mmio_fault_cr2 = cr2;
3975 /*
3976 * TODO: fix emulate.c to use guest_read/write_register
3977 * instead of direct ->regs accesses, can save hundred cycles
3978 * on Intel for instructions that don't read/change RSP, for
3979 * for example.
3980 */
3981 cache_all_regs(vcpu);
3982 4589
3983 if (!(emulation_type & EMULTYPE_NO_DECODE)) { 4590 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
3984 int cs_db, cs_l; 4591 init_emulate_ctxt(vcpu);
3985 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
3986
3987 vcpu->arch.emulate_ctxt.vcpu = vcpu;
3988 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
3989 vcpu->arch.emulate_ctxt.eip = kvm_rip_read(vcpu);
3990 vcpu->arch.emulate_ctxt.mode =
3991 (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
3992 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
3993 ? X86EMUL_MODE_VM86 : cs_l
3994 ? X86EMUL_MODE_PROT64 : cs_db
3995 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
3996 memset(c, 0, sizeof(struct decode_cache));
3997 memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
3998 vcpu->arch.emulate_ctxt.interruptibility = 0; 4592 vcpu->arch.emulate_ctxt.interruptibility = 0;
3999 vcpu->arch.emulate_ctxt.exception = -1; 4593 vcpu->arch.emulate_ctxt.have_exception = false;
4000 4594 vcpu->arch.emulate_ctxt.perm_ok = false;
4001 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
4002 trace_kvm_emulate_insn_start(vcpu);
4003 4595
4004 /* Only allow emulation of specific instructions on #UD 4596 vcpu->arch.emulate_ctxt.only_vendor_specific_insn
4005 * (namely VMMCALL, sysenter, sysexit, syscall)*/ 4597 = emulation_type & EMULTYPE_TRAP_UD;
4006 if (emulation_type & EMULTYPE_TRAP_UD) {
4007 if (!c->twobyte)
4008 return EMULATE_FAIL;
4009 switch (c->b) {
4010 case 0x01: /* VMMCALL */
4011 if (c->modrm_mod != 3 || c->modrm_rm != 1)
4012 return EMULATE_FAIL;
4013 break;
4014 case 0x34: /* sysenter */
4015 case 0x35: /* sysexit */
4016 if (c->modrm_mod != 0 || c->modrm_rm != 0)
4017 return EMULATE_FAIL;
4018 break;
4019 case 0x05: /* syscall */
4020 if (c->modrm_mod != 0 || c->modrm_rm != 0)
4021 return EMULATE_FAIL;
4022 break;
4023 default:
4024 return EMULATE_FAIL;
4025 }
4026 4598
4027 if (!(c->modrm_reg == 0 || c->modrm_reg == 3)) 4599 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, insn, insn_len);
4028 return EMULATE_FAIL;
4029 }
4030 4600
4601 trace_kvm_emulate_insn_start(vcpu);
4031 ++vcpu->stat.insn_emulation; 4602 ++vcpu->stat.insn_emulation;
4032 if (r) { 4603 if (r) {
4604 if (emulation_type & EMULTYPE_TRAP_UD)
4605 return EMULATE_FAIL;
4033 if (reexecute_instruction(vcpu, cr2)) 4606 if (reexecute_instruction(vcpu, cr2))
4034 return EMULATE_DONE; 4607 return EMULATE_DONE;
4035 if (emulation_type & EMULTYPE_SKIP) 4608 if (emulation_type & EMULTYPE_SKIP)
@@ -4043,62 +4616,87 @@ int emulate_instruction(struct kvm_vcpu *vcpu,
4043 return EMULATE_DONE; 4616 return EMULATE_DONE;
4044 } 4617 }
4045 4618
4046 /* this is needed for vmware backdor interface to work since it 4619 /* this is needed for vmware backdoor interface to work since it
4047 changes registers values during IO operation */ 4620 changes registers values during IO operation */
4048 memcpy(c->regs, vcpu->arch.regs, sizeof c->regs); 4621 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
4622 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4623 memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
4624 }
4049 4625
4050restart: 4626restart:
4051 r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops); 4627 r = x86_emulate_insn(&vcpu->arch.emulate_ctxt);
4628
4629 if (r == EMULATION_INTERCEPTED)
4630 return EMULATE_DONE;
4052 4631
4053 if (r) { /* emulation failed */ 4632 if (r == EMULATION_FAILED) {
4054 if (reexecute_instruction(vcpu, cr2)) 4633 if (reexecute_instruction(vcpu, cr2))
4055 return EMULATE_DONE; 4634 return EMULATE_DONE;
4056 4635
4057 return handle_emulation_failure(vcpu); 4636 return handle_emulation_failure(vcpu);
4058 } 4637 }
4059 4638
4060 toggle_interruptibility(vcpu, vcpu->arch.emulate_ctxt.interruptibility); 4639 if (vcpu->arch.emulate_ctxt.have_exception) {
4061 kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
4062 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
4063 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
4064
4065 if (vcpu->arch.emulate_ctxt.exception >= 0) {
4066 inject_emulated_exception(vcpu); 4640 inject_emulated_exception(vcpu);
4067 return EMULATE_DONE; 4641 r = EMULATE_DONE;
4068 } 4642 } else if (vcpu->arch.pio.count) {
4069
4070 if (vcpu->arch.pio.count) {
4071 if (!vcpu->arch.pio.in) 4643 if (!vcpu->arch.pio.in)
4072 vcpu->arch.pio.count = 0; 4644 vcpu->arch.pio.count = 0;
4073 return EMULATE_DO_MMIO; 4645 else
4074 } 4646 writeback = false;
4075 4647 r = EMULATE_DO_MMIO;
4076 if (vcpu->mmio_needed) { 4648 } else if (vcpu->mmio_needed) {
4077 if (vcpu->mmio_is_write) 4649 if (!vcpu->mmio_is_write)
4078 vcpu->mmio_needed = 0; 4650 writeback = false;
4079 return EMULATE_DO_MMIO; 4651 r = EMULATE_DO_MMIO;
4080 } 4652 } else if (r == EMULATION_RESTART)
4081
4082 if (vcpu->arch.emulate_ctxt.restart)
4083 goto restart; 4653 goto restart;
4654 else
4655 r = EMULATE_DONE;
4656
4657 if (writeback) {
4658 toggle_interruptibility(vcpu,
4659 vcpu->arch.emulate_ctxt.interruptibility);
4660 kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
4661 kvm_make_request(KVM_REQ_EVENT, vcpu);
4662 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
4663 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
4664 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
4665 } else
4666 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
4084 4667
4085 return EMULATE_DONE; 4668 return r;
4086} 4669}
4087EXPORT_SYMBOL_GPL(emulate_instruction); 4670EXPORT_SYMBOL_GPL(x86_emulate_instruction);
4088 4671
4089int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port) 4672int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
4090{ 4673{
4091 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX); 4674 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
4092 int ret = emulator_pio_out_emulated(size, port, &val, 1, vcpu); 4675 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
4676 size, port, &val, 1);
4093 /* do not return to emulator after return from userspace */ 4677 /* do not return to emulator after return from userspace */
4094 vcpu->arch.pio.count = 0; 4678 vcpu->arch.pio.count = 0;
4095 return ret; 4679 return ret;
4096} 4680}
4097EXPORT_SYMBOL_GPL(kvm_fast_pio_out); 4681EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
4098 4682
4099static void bounce_off(void *info) 4683static void tsc_bad(void *info)
4100{ 4684{
4101 /* nothing */ 4685 __this_cpu_write(cpu_tsc_khz, 0);
4686}
4687
4688static void tsc_khz_changed(void *data)
4689{
4690 struct cpufreq_freqs *freq = data;
4691 unsigned long khz = 0;
4692
4693 if (data)
4694 khz = freq->new;
4695 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4696 khz = cpufreq_quick_get(raw_smp_processor_id());
4697 if (!khz)
4698 khz = tsc_khz;
4699 __this_cpu_write(cpu_tsc_khz, khz);
4102} 4700}
4103 4701
4104static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val, 4702static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
@@ -4109,24 +4707,63 @@ static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long va
4109 struct kvm_vcpu *vcpu; 4707 struct kvm_vcpu *vcpu;
4110 int i, send_ipi = 0; 4708 int i, send_ipi = 0;
4111 4709
4710 /*
4711 * We allow guests to temporarily run on slowing clocks,
4712 * provided we notify them after, or to run on accelerating
4713 * clocks, provided we notify them before. Thus time never
4714 * goes backwards.
4715 *
4716 * However, we have a problem. We can't atomically update
4717 * the frequency of a given CPU from this function; it is
4718 * merely a notifier, which can be called from any CPU.
4719 * Changing the TSC frequency at arbitrary points in time
4720 * requires a recomputation of local variables related to
4721 * the TSC for each VCPU. We must flag these local variables
4722 * to be updated and be sure the update takes place with the
4723 * new frequency before any guests proceed.
4724 *
4725 * Unfortunately, the combination of hotplug CPU and frequency
4726 * change creates an intractable locking scenario; the order
4727 * of when these callouts happen is undefined with respect to
4728 * CPU hotplug, and they can race with each other. As such,
4729 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
4730 * undefined; you can actually have a CPU frequency change take
4731 * place in between the computation of X and the setting of the
4732 * variable. To protect against this problem, all updates of
4733 * the per_cpu tsc_khz variable are done in an interrupt
4734 * protected IPI, and all callers wishing to update the value
4735 * must wait for a synchronous IPI to complete (which is trivial
4736 * if the caller is on the CPU already). This establishes the
4737 * necessary total order on variable updates.
4738 *
4739 * Note that because a guest time update may take place
4740 * anytime after the setting of the VCPU's request bit, the
4741 * correct TSC value must be set before the request. However,
4742 * to ensure the update actually makes it to any guest which
4743 * starts running in hardware virtualization between the set
4744 * and the acquisition of the spinlock, we must also ping the
4745 * CPU after setting the request bit.
4746 *
4747 */
4748
4112 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new) 4749 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
4113 return 0; 4750 return 0;
4114 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new) 4751 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
4115 return 0; 4752 return 0;
4116 per_cpu(cpu_tsc_khz, freq->cpu) = freq->new;
4117 4753
4118 spin_lock(&kvm_lock); 4754 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4755
4756 raw_spin_lock(&kvm_lock);
4119 list_for_each_entry(kvm, &vm_list, vm_list) { 4757 list_for_each_entry(kvm, &vm_list, vm_list) {
4120 kvm_for_each_vcpu(i, vcpu, kvm) { 4758 kvm_for_each_vcpu(i, vcpu, kvm) {
4121 if (vcpu->cpu != freq->cpu) 4759 if (vcpu->cpu != freq->cpu)
4122 continue; 4760 continue;
4123 if (!kvm_request_guest_time_update(vcpu)) 4761 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4124 continue;
4125 if (vcpu->cpu != smp_processor_id()) 4762 if (vcpu->cpu != smp_processor_id())
4126 send_ipi++; 4763 send_ipi = 1;
4127 } 4764 }
4128 } 4765 }
4129 spin_unlock(&kvm_lock); 4766 raw_spin_unlock(&kvm_lock);
4130 4767
4131 if (freq->old < freq->new && send_ipi) { 4768 if (freq->old < freq->new && send_ipi) {
4132 /* 4769 /*
@@ -4141,32 +4778,59 @@ static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long va
4141 * guest context is entered kvmclock will be updated, 4778 * guest context is entered kvmclock will be updated,
4142 * so the guest will not see stale values. 4779 * so the guest will not see stale values.
4143 */ 4780 */
4144 smp_call_function_single(freq->cpu, bounce_off, NULL, 1); 4781 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4145 } 4782 }
4146 return 0; 4783 return 0;
4147} 4784}
4148 4785
4149static struct notifier_block kvmclock_cpufreq_notifier_block = { 4786static struct notifier_block kvmclock_cpufreq_notifier_block = {
4150 .notifier_call = kvmclock_cpufreq_notifier 4787 .notifier_call = kvmclock_cpufreq_notifier
4788};
4789
4790static int kvmclock_cpu_notifier(struct notifier_block *nfb,
4791 unsigned long action, void *hcpu)
4792{
4793 unsigned int cpu = (unsigned long)hcpu;
4794
4795 switch (action) {
4796 case CPU_ONLINE:
4797 case CPU_DOWN_FAILED:
4798 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4799 break;
4800 case CPU_DOWN_PREPARE:
4801 smp_call_function_single(cpu, tsc_bad, NULL, 1);
4802 break;
4803 }
4804 return NOTIFY_OK;
4805}
4806
4807static struct notifier_block kvmclock_cpu_notifier_block = {
4808 .notifier_call = kvmclock_cpu_notifier,
4809 .priority = -INT_MAX
4151}; 4810};
4152 4811
4153static void kvm_timer_init(void) 4812static void kvm_timer_init(void)
4154{ 4813{
4155 int cpu; 4814 int cpu;
4156 4815
4816 max_tsc_khz = tsc_khz;
4817 register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4157 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) { 4818 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
4819#ifdef CONFIG_CPU_FREQ
4820 struct cpufreq_policy policy;
4821 memset(&policy, 0, sizeof(policy));
4822 cpu = get_cpu();
4823 cpufreq_get_policy(&policy, cpu);
4824 if (policy.cpuinfo.max_freq)
4825 max_tsc_khz = policy.cpuinfo.max_freq;
4826 put_cpu();
4827#endif
4158 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block, 4828 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
4159 CPUFREQ_TRANSITION_NOTIFIER); 4829 CPUFREQ_TRANSITION_NOTIFIER);
4160 for_each_online_cpu(cpu) {
4161 unsigned long khz = cpufreq_get(cpu);
4162 if (!khz)
4163 khz = tsc_khz;
4164 per_cpu(cpu_tsc_khz, cpu) = khz;
4165 }
4166 } else {
4167 for_each_possible_cpu(cpu)
4168 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
4169 } 4830 }
4831 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
4832 for_each_online_cpu(cpu)
4833 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4170} 4834}
4171 4835
4172static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu); 4836static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
@@ -4244,7 +4908,6 @@ int kvm_arch_init(void *opaque)
4244 4908
4245 kvm_x86_ops = ops; 4909 kvm_x86_ops = ops;
4246 kvm_mmu_set_nonpresent_ptes(0ull, 0ull); 4910 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
4247 kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
4248 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK, 4911 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
4249 PT_DIRTY_MASK, PT64_NX_MASK, 0); 4912 PT_DIRTY_MASK, PT64_NX_MASK, 0);
4250 4913
@@ -4268,6 +4931,7 @@ void kvm_arch_exit(void)
4268 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) 4931 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4269 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block, 4932 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
4270 CPUFREQ_TRANSITION_NOTIFIER); 4933 CPUFREQ_TRANSITION_NOTIFIER);
4934 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4271 kvm_x86_ops = NULL; 4935 kvm_x86_ops = NULL;
4272 kvm_mmu_module_exit(); 4936 kvm_mmu_module_exit();
4273} 4937}
@@ -4403,8 +5067,9 @@ out:
4403} 5067}
4404EXPORT_SYMBOL_GPL(kvm_emulate_hypercall); 5068EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
4405 5069
4406int kvm_fix_hypercall(struct kvm_vcpu *vcpu) 5070int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
4407{ 5071{
5072 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4408 char instruction[3]; 5073 char instruction[3];
4409 unsigned long rip = kvm_rip_read(vcpu); 5074 unsigned long rip = kvm_rip_read(vcpu);
4410 5075
@@ -4417,21 +5082,8 @@ int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
4417 5082
4418 kvm_x86_ops->patch_hypercall(vcpu, instruction); 5083 kvm_x86_ops->patch_hypercall(vcpu, instruction);
4419 5084
4420 return emulator_write_emulated(rip, instruction, 3, NULL, vcpu); 5085 return emulator_write_emulated(&vcpu->arch.emulate_ctxt,
4421} 5086 rip, instruction, 3, NULL);
4422
4423void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4424{
4425 struct desc_ptr dt = { limit, base };
4426
4427 kvm_x86_ops->set_gdt(vcpu, &dt);
4428}
4429
4430void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4431{
4432 struct desc_ptr dt = { limit, base };
4433
4434 kvm_x86_ops->set_idt(vcpu, &dt);
4435} 5087}
4436 5088
4437static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i) 5089static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
@@ -4482,12 +5134,6 @@ struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
4482 best = e; 5134 best = e;
4483 break; 5135 break;
4484 } 5136 }
4485 /*
4486 * Both basic or both extended?
4487 */
4488 if (((e->function ^ function) & 0x80000000) == 0)
4489 if (!best || e->function > best->function)
4490 best = e;
4491 } 5137 }
4492 return best; 5138 return best;
4493} 5139}
@@ -4507,6 +5153,27 @@ not_found:
4507 return 36; 5153 return 36;
4508} 5154}
4509 5155
5156/*
5157 * If no match is found, check whether we exceed the vCPU's limit
5158 * and return the content of the highest valid _standard_ leaf instead.
5159 * This is to satisfy the CPUID specification.
5160 */
5161static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
5162 u32 function, u32 index)
5163{
5164 struct kvm_cpuid_entry2 *maxlevel;
5165
5166 maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
5167 if (!maxlevel || maxlevel->eax >= function)
5168 return NULL;
5169 if (function & 0x80000000) {
5170 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
5171 if (!maxlevel)
5172 return NULL;
5173 }
5174 return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
5175}
5176
4510void kvm_emulate_cpuid(struct kvm_vcpu *vcpu) 5177void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
4511{ 5178{
4512 u32 function, index; 5179 u32 function, index;
@@ -4519,6 +5186,10 @@ void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
4519 kvm_register_write(vcpu, VCPU_REGS_RCX, 0); 5186 kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
4520 kvm_register_write(vcpu, VCPU_REGS_RDX, 0); 5187 kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
4521 best = kvm_find_cpuid_entry(vcpu, function, index); 5188 best = kvm_find_cpuid_entry(vcpu, function, index);
5189
5190 if (!best)
5191 best = check_cpuid_limit(vcpu, function, index);
5192
4522 if (best) { 5193 if (best) {
4523 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax); 5194 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
4524 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx); 5195 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
@@ -4675,6 +5346,7 @@ static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
4675static int vcpu_enter_guest(struct kvm_vcpu *vcpu) 5346static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
4676{ 5347{
4677 int r; 5348 int r;
5349 bool nmi_pending;
4678 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) && 5350 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
4679 vcpu->run->request_interrupt_window; 5351 vcpu->run->request_interrupt_window;
4680 5352
@@ -4683,8 +5355,11 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
4683 kvm_mmu_unload(vcpu); 5355 kvm_mmu_unload(vcpu);
4684 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu)) 5356 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
4685 __kvm_migrate_timers(vcpu); 5357 __kvm_migrate_timers(vcpu);
4686 if (kvm_check_request(KVM_REQ_KVMCLOCK_UPDATE, vcpu)) 5358 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
4687 kvm_write_guest_time(vcpu); 5359 r = kvm_guest_time_update(vcpu);
5360 if (unlikely(r))
5361 goto out;
5362 }
4688 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu)) 5363 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
4689 kvm_mmu_sync_roots(vcpu); 5364 kvm_mmu_sync_roots(vcpu);
4690 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) 5365 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
@@ -4703,12 +5378,41 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
4703 vcpu->fpu_active = 0; 5378 vcpu->fpu_active = 0;
4704 kvm_x86_ops->fpu_deactivate(vcpu); 5379 kvm_x86_ops->fpu_deactivate(vcpu);
4705 } 5380 }
5381 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5382 /* Page is swapped out. Do synthetic halt */
5383 vcpu->arch.apf.halted = true;
5384 r = 1;
5385 goto out;
5386 }
4706 } 5387 }
4707 5388
4708 r = kvm_mmu_reload(vcpu); 5389 r = kvm_mmu_reload(vcpu);
4709 if (unlikely(r)) 5390 if (unlikely(r))
4710 goto out; 5391 goto out;
4711 5392
5393 /*
5394 * An NMI can be injected between local nmi_pending read and
5395 * vcpu->arch.nmi_pending read inside inject_pending_event().
5396 * But in that case, KVM_REQ_EVENT will be set, which makes
5397 * the race described above benign.
5398 */
5399 nmi_pending = ACCESS_ONCE(vcpu->arch.nmi_pending);
5400
5401 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5402 inject_pending_event(vcpu);
5403
5404 /* enable NMI/IRQ window open exits if needed */
5405 if (nmi_pending)
5406 kvm_x86_ops->enable_nmi_window(vcpu);
5407 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
5408 kvm_x86_ops->enable_irq_window(vcpu);
5409
5410 if (kvm_lapic_enabled(vcpu)) {
5411 update_cr8_intercept(vcpu);
5412 kvm_lapic_sync_to_vapic(vcpu);
5413 }
5414 }
5415
4712 preempt_disable(); 5416 preempt_disable();
4713 5417
4714 kvm_x86_ops->prepare_guest_switch(vcpu); 5418 kvm_x86_ops->prepare_guest_switch(vcpu);
@@ -4716,34 +5420,26 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
4716 kvm_load_guest_fpu(vcpu); 5420 kvm_load_guest_fpu(vcpu);
4717 kvm_load_guest_xcr0(vcpu); 5421 kvm_load_guest_xcr0(vcpu);
4718 5422
4719 atomic_set(&vcpu->guest_mode, 1); 5423 vcpu->mode = IN_GUEST_MODE;
4720 smp_wmb(); 5424
5425 /* We should set ->mode before check ->requests,
5426 * see the comment in make_all_cpus_request.
5427 */
5428 smp_mb();
4721 5429
4722 local_irq_disable(); 5430 local_irq_disable();
4723 5431
4724 if (!atomic_read(&vcpu->guest_mode) || vcpu->requests 5432 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
4725 || need_resched() || signal_pending(current)) { 5433 || need_resched() || signal_pending(current)) {
4726 atomic_set(&vcpu->guest_mode, 0); 5434 vcpu->mode = OUTSIDE_GUEST_MODE;
4727 smp_wmb(); 5435 smp_wmb();
4728 local_irq_enable(); 5436 local_irq_enable();
4729 preempt_enable(); 5437 preempt_enable();
5438 kvm_x86_ops->cancel_injection(vcpu);
4730 r = 1; 5439 r = 1;
4731 goto out; 5440 goto out;
4732 } 5441 }
4733 5442
4734 inject_pending_event(vcpu);
4735
4736 /* enable NMI/IRQ window open exits if needed */
4737 if (vcpu->arch.nmi_pending)
4738 kvm_x86_ops->enable_nmi_window(vcpu);
4739 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
4740 kvm_x86_ops->enable_irq_window(vcpu);
4741
4742 if (kvm_lapic_enabled(vcpu)) {
4743 update_cr8_intercept(vcpu);
4744 kvm_lapic_sync_to_vapic(vcpu);
4745 }
4746
4747 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx); 5443 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4748 5444
4749 kvm_guest_enter(); 5445 kvm_guest_enter();
@@ -4769,7 +5465,9 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
4769 if (hw_breakpoint_active()) 5465 if (hw_breakpoint_active())
4770 hw_breakpoint_restore(); 5466 hw_breakpoint_restore();
4771 5467
4772 atomic_set(&vcpu->guest_mode, 0); 5468 kvm_get_msr(vcpu, MSR_IA32_TSC, &vcpu->arch.last_guest_tsc);
5469
5470 vcpu->mode = OUTSIDE_GUEST_MODE;
4773 smp_wmb(); 5471 smp_wmb();
4774 local_irq_enable(); 5472 local_irq_enable();
4775 5473
@@ -4826,7 +5524,8 @@ static int __vcpu_run(struct kvm_vcpu *vcpu)
4826 5524
4827 r = 1; 5525 r = 1;
4828 while (r > 0) { 5526 while (r > 0) {
4829 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE) 5527 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
5528 !vcpu->arch.apf.halted)
4830 r = vcpu_enter_guest(vcpu); 5529 r = vcpu_enter_guest(vcpu);
4831 else { 5530 else {
4832 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx); 5531 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
@@ -4839,6 +5538,7 @@ static int __vcpu_run(struct kvm_vcpu *vcpu)
4839 vcpu->arch.mp_state = 5538 vcpu->arch.mp_state =
4840 KVM_MP_STATE_RUNNABLE; 5539 KVM_MP_STATE_RUNNABLE;
4841 case KVM_MP_STATE_RUNNABLE: 5540 case KVM_MP_STATE_RUNNABLE:
5541 vcpu->arch.apf.halted = false;
4842 break; 5542 break;
4843 case KVM_MP_STATE_SIPI_RECEIVED: 5543 case KVM_MP_STATE_SIPI_RECEIVED:
4844 default: 5544 default:
@@ -4860,6 +5560,9 @@ static int __vcpu_run(struct kvm_vcpu *vcpu)
4860 vcpu->run->exit_reason = KVM_EXIT_INTR; 5560 vcpu->run->exit_reason = KVM_EXIT_INTR;
4861 ++vcpu->stat.request_irq_exits; 5561 ++vcpu->stat.request_irq_exits;
4862 } 5562 }
5563
5564 kvm_check_async_pf_completion(vcpu);
5565
4863 if (signal_pending(current)) { 5566 if (signal_pending(current)) {
4864 r = -EINTR; 5567 r = -EINTR;
4865 vcpu->run->exit_reason = KVM_EXIT_INTR; 5568 vcpu->run->exit_reason = KVM_EXIT_INTR;
@@ -4879,11 +5582,49 @@ static int __vcpu_run(struct kvm_vcpu *vcpu)
4879 return r; 5582 return r;
4880} 5583}
4881 5584
5585static int complete_mmio(struct kvm_vcpu *vcpu)
5586{
5587 struct kvm_run *run = vcpu->run;
5588 int r;
5589
5590 if (!(vcpu->arch.pio.count || vcpu->mmio_needed))
5591 return 1;
5592
5593 if (vcpu->mmio_needed) {
5594 vcpu->mmio_needed = 0;
5595 if (!vcpu->mmio_is_write)
5596 memcpy(vcpu->mmio_data + vcpu->mmio_index,
5597 run->mmio.data, 8);
5598 vcpu->mmio_index += 8;
5599 if (vcpu->mmio_index < vcpu->mmio_size) {
5600 run->exit_reason = KVM_EXIT_MMIO;
5601 run->mmio.phys_addr = vcpu->mmio_phys_addr + vcpu->mmio_index;
5602 memcpy(run->mmio.data, vcpu->mmio_data + vcpu->mmio_index, 8);
5603 run->mmio.len = min(vcpu->mmio_size - vcpu->mmio_index, 8);
5604 run->mmio.is_write = vcpu->mmio_is_write;
5605 vcpu->mmio_needed = 1;
5606 return 0;
5607 }
5608 if (vcpu->mmio_is_write)
5609 return 1;
5610 vcpu->mmio_read_completed = 1;
5611 }
5612 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5613 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
5614 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5615 if (r != EMULATE_DONE)
5616 return 0;
5617 return 1;
5618}
5619
4882int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run) 5620int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
4883{ 5621{
4884 int r; 5622 int r;
4885 sigset_t sigsaved; 5623 sigset_t sigsaved;
4886 5624
5625 if (!tsk_used_math(current) && init_fpu(current))
5626 return -ENOMEM;
5627
4887 if (vcpu->sigset_active) 5628 if (vcpu->sigset_active)
4888 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved); 5629 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
4889 5630
@@ -4895,24 +5636,17 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
4895 } 5636 }
4896 5637
4897 /* re-sync apic's tpr */ 5638 /* re-sync apic's tpr */
4898 if (!irqchip_in_kernel(vcpu->kvm)) 5639 if (!irqchip_in_kernel(vcpu->kvm)) {
4899 kvm_set_cr8(vcpu, kvm_run->cr8); 5640 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
4900 5641 r = -EINVAL;
4901 if (vcpu->arch.pio.count || vcpu->mmio_needed ||
4902 vcpu->arch.emulate_ctxt.restart) {
4903 if (vcpu->mmio_needed) {
4904 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
4905 vcpu->mmio_read_completed = 1;
4906 vcpu->mmio_needed = 0;
4907 }
4908 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4909 r = emulate_instruction(vcpu, 0, 0, EMULTYPE_NO_DECODE);
4910 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4911 if (r != EMULATE_DONE) {
4912 r = 0;
4913 goto out; 5642 goto out;
4914 } 5643 }
4915 } 5644 }
5645
5646 r = complete_mmio(vcpu);
5647 if (r <= 0)
5648 goto out;
5649
4916 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) 5650 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
4917 kvm_register_write(vcpu, VCPU_REGS_RAX, 5651 kvm_register_write(vcpu, VCPU_REGS_RAX,
4918 kvm_run->hypercall.ret); 5652 kvm_run->hypercall.ret);
@@ -4929,6 +5663,18 @@ out:
4929 5663
4930int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs) 5664int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4931{ 5665{
5666 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
5667 /*
5668 * We are here if userspace calls get_regs() in the middle of
5669 * instruction emulation. Registers state needs to be copied
5670 * back from emulation context to vcpu. Usrapace shouldn't do
5671 * that usually, but some bad designed PV devices (vmware
5672 * backdoor interface) need this to work
5673 */
5674 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
5675 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
5676 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5677 }
4932 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX); 5678 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4933 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX); 5679 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4934 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX); 5680 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
@@ -4956,6 +5702,9 @@ int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4956 5702
4957int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs) 5703int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4958{ 5704{
5705 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
5706 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5707
4959 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax); 5708 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
4960 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx); 5709 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
4961 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx); 5710 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
@@ -4980,6 +5729,8 @@ int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4980 5729
4981 vcpu->arch.exception.pending = false; 5730 vcpu->arch.exception.pending = false;
4982 5731
5732 kvm_make_request(KVM_REQ_EVENT, vcpu);
5733
4983 return 0; 5734 return 0;
4984} 5735}
4985 5736
@@ -5017,7 +5768,7 @@ int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
5017 5768
5018 sregs->cr0 = kvm_read_cr0(vcpu); 5769 sregs->cr0 = kvm_read_cr0(vcpu);
5019 sregs->cr2 = vcpu->arch.cr2; 5770 sregs->cr2 = vcpu->arch.cr2;
5020 sregs->cr3 = vcpu->arch.cr3; 5771 sregs->cr3 = kvm_read_cr3(vcpu);
5021 sregs->cr4 = kvm_read_cr4(vcpu); 5772 sregs->cr4 = kvm_read_cr4(vcpu);
5022 sregs->cr8 = kvm_get_cr8(vcpu); 5773 sregs->cr8 = kvm_get_cr8(vcpu);
5023 sregs->efer = vcpu->arch.efer; 5774 sregs->efer = vcpu->arch.efer;
@@ -5043,6 +5794,7 @@ int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
5043 struct kvm_mp_state *mp_state) 5794 struct kvm_mp_state *mp_state)
5044{ 5795{
5045 vcpu->arch.mp_state = mp_state->mp_state; 5796 vcpu->arch.mp_state = mp_state->mp_state;
5797 kvm_make_request(KVM_REQ_EVENT, vcpu);
5046 return 0; 5798 return 0;
5047} 5799}
5048 5800
@@ -5050,24 +5802,11 @@ int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason,
5050 bool has_error_code, u32 error_code) 5802 bool has_error_code, u32 error_code)
5051{ 5803{
5052 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode; 5804 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
5053 int cs_db, cs_l, ret; 5805 int ret;
5054 cache_all_regs(vcpu);
5055
5056 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5057 5806
5058 vcpu->arch.emulate_ctxt.vcpu = vcpu; 5807 init_emulate_ctxt(vcpu);
5059 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
5060 vcpu->arch.emulate_ctxt.eip = kvm_rip_read(vcpu);
5061 vcpu->arch.emulate_ctxt.mode =
5062 (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
5063 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
5064 ? X86EMUL_MODE_VM86 : cs_l
5065 ? X86EMUL_MODE_PROT64 : cs_db
5066 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
5067 memset(c, 0, sizeof(struct decode_cache));
5068 memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
5069 5808
5070 ret = emulator_task_switch(&vcpu->arch.emulate_ctxt, &emulate_ops, 5809 ret = emulator_task_switch(&vcpu->arch.emulate_ctxt,
5071 tss_selector, reason, has_error_code, 5810 tss_selector, reason, has_error_code,
5072 error_code); 5811 error_code);
5073 5812
@@ -5076,7 +5815,8 @@ int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason,
5076 5815
5077 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs); 5816 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
5078 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip); 5817 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
5079 kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags); 5818 kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
5819 kvm_make_request(KVM_REQ_EVENT, vcpu);
5080 return EMULATE_DONE; 5820 return EMULATE_DONE;
5081} 5821}
5082EXPORT_SYMBOL_GPL(kvm_task_switch); 5822EXPORT_SYMBOL_GPL(kvm_task_switch);
@@ -5085,7 +5825,7 @@ int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
5085 struct kvm_sregs *sregs) 5825 struct kvm_sregs *sregs)
5086{ 5826{
5087 int mmu_reset_needed = 0; 5827 int mmu_reset_needed = 0;
5088 int pending_vec, max_bits; 5828 int pending_vec, max_bits, idx;
5089 struct desc_ptr dt; 5829 struct desc_ptr dt;
5090 5830
5091 dt.size = sregs->idt.limit; 5831 dt.size = sregs->idt.limit;
@@ -5096,8 +5836,9 @@ int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
5096 kvm_x86_ops->set_gdt(vcpu, &dt); 5836 kvm_x86_ops->set_gdt(vcpu, &dt);
5097 5837
5098 vcpu->arch.cr2 = sregs->cr2; 5838 vcpu->arch.cr2 = sregs->cr2;
5099 mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3; 5839 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
5100 vcpu->arch.cr3 = sregs->cr3; 5840 vcpu->arch.cr3 = sregs->cr3;
5841 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
5101 5842
5102 kvm_set_cr8(vcpu, sregs->cr8); 5843 kvm_set_cr8(vcpu, sregs->cr8);
5103 5844
@@ -5111,10 +5852,15 @@ int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
5111 5852
5112 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4; 5853 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
5113 kvm_x86_ops->set_cr4(vcpu, sregs->cr4); 5854 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
5855 if (sregs->cr4 & X86_CR4_OSXSAVE)
5856 update_cpuid(vcpu);
5857
5858 idx = srcu_read_lock(&vcpu->kvm->srcu);
5114 if (!is_long_mode(vcpu) && is_pae(vcpu)) { 5859 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
5115 load_pdptrs(vcpu, vcpu->arch.cr3); 5860 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
5116 mmu_reset_needed = 1; 5861 mmu_reset_needed = 1;
5117 } 5862 }
5863 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5118 5864
5119 if (mmu_reset_needed) 5865 if (mmu_reset_needed)
5120 kvm_mmu_reset_context(vcpu); 5866 kvm_mmu_reset_context(vcpu);
@@ -5125,8 +5871,6 @@ int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
5125 if (pending_vec < max_bits) { 5871 if (pending_vec < max_bits) {
5126 kvm_queue_interrupt(vcpu, pending_vec, false); 5872 kvm_queue_interrupt(vcpu, pending_vec, false);
5127 pr_debug("Set back pending irq %d\n", pending_vec); 5873 pr_debug("Set back pending irq %d\n", pending_vec);
5128 if (irqchip_in_kernel(vcpu->kvm))
5129 kvm_pic_clear_isr_ack(vcpu->kvm);
5130 } 5874 }
5131 5875
5132 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS); 5876 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
@@ -5147,6 +5891,8 @@ int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
5147 !is_protmode(vcpu)) 5891 !is_protmode(vcpu))
5148 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE; 5892 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5149 5893
5894 kvm_make_request(KVM_REQ_EVENT, vcpu);
5895
5150 return 0; 5896 return 0;
5151} 5897}
5152 5898
@@ -5320,10 +6066,7 @@ void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
5320 6066
5321void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu) 6067void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
5322{ 6068{
5323 if (vcpu->arch.time_page) { 6069 kvmclock_reset(vcpu);
5324 kvm_release_page_dirty(vcpu->arch.time_page);
5325 vcpu->arch.time_page = NULL;
5326 }
5327 6070
5328 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask); 6071 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
5329 fx_free(vcpu); 6072 fx_free(vcpu);
@@ -5333,6 +6076,10 @@ void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
5333struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, 6076struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
5334 unsigned int id) 6077 unsigned int id)
5335{ 6078{
6079 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
6080 printk_once(KERN_WARNING
6081 "kvm: SMP vm created on host with unstable TSC; "
6082 "guest TSC will not be reliable\n");
5336 return kvm_x86_ops->vcpu_create(kvm, id); 6083 return kvm_x86_ops->vcpu_create(kvm, id);
5337} 6084}
5338 6085
@@ -5357,6 +6104,8 @@ free_vcpu:
5357 6104
5358void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu) 6105void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
5359{ 6106{
6107 vcpu->arch.apf.msr_val = 0;
6108
5360 vcpu_load(vcpu); 6109 vcpu_load(vcpu);
5361 kvm_mmu_unload(vcpu); 6110 kvm_mmu_unload(vcpu);
5362 vcpu_put(vcpu); 6111 vcpu_put(vcpu);
@@ -5375,22 +6124,29 @@ int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
5375 vcpu->arch.dr6 = DR6_FIXED_1; 6124 vcpu->arch.dr6 = DR6_FIXED_1;
5376 vcpu->arch.dr7 = DR7_FIXED_1; 6125 vcpu->arch.dr7 = DR7_FIXED_1;
5377 6126
6127 kvm_make_request(KVM_REQ_EVENT, vcpu);
6128 vcpu->arch.apf.msr_val = 0;
6129
6130 kvmclock_reset(vcpu);
6131
6132 kvm_clear_async_pf_completion_queue(vcpu);
6133 kvm_async_pf_hash_reset(vcpu);
6134 vcpu->arch.apf.halted = false;
6135
5378 return kvm_x86_ops->vcpu_reset(vcpu); 6136 return kvm_x86_ops->vcpu_reset(vcpu);
5379} 6137}
5380 6138
5381int kvm_arch_hardware_enable(void *garbage) 6139int kvm_arch_hardware_enable(void *garbage)
5382{ 6140{
5383 /* 6141 struct kvm *kvm;
5384 * Since this may be called from a hotplug notifcation, 6142 struct kvm_vcpu *vcpu;
5385 * we can't get the CPU frequency directly. 6143 int i;
5386 */
5387 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5388 int cpu = raw_smp_processor_id();
5389 per_cpu(cpu_tsc_khz, cpu) = 0;
5390 }
5391 6144
5392 kvm_shared_msr_cpu_online(); 6145 kvm_shared_msr_cpu_online();
5393 6146 list_for_each_entry(kvm, &vm_list, vm_list)
6147 kvm_for_each_vcpu(i, vcpu, kvm)
6148 if (vcpu->cpu == smp_processor_id())
6149 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5394 return kvm_x86_ops->hardware_enable(garbage); 6150 return kvm_x86_ops->hardware_enable(garbage);
5395} 6151}
5396 6152
@@ -5424,7 +6180,11 @@ int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
5424 BUG_ON(vcpu->kvm == NULL); 6180 BUG_ON(vcpu->kvm == NULL);
5425 kvm = vcpu->kvm; 6181 kvm = vcpu->kvm;
5426 6182
6183 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
6184 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
5427 vcpu->arch.mmu.root_hpa = INVALID_PAGE; 6185 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
6186 vcpu->arch.mmu.translate_gpa = translate_gpa;
6187 vcpu->arch.nested_mmu.translate_gpa = translate_nested_gpa;
5428 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu)) 6188 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
5429 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE; 6189 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5430 else 6190 else
@@ -5437,6 +6197,8 @@ int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
5437 } 6197 }
5438 vcpu->arch.pio_data = page_address(page); 6198 vcpu->arch.pio_data = page_address(page);
5439 6199
6200 kvm_init_tsc_catchup(vcpu, max_tsc_khz);
6201
5440 r = kvm_mmu_create(vcpu); 6202 r = kvm_mmu_create(vcpu);
5441 if (r < 0) 6203 if (r < 0)
5442 goto fail_free_pio_data; 6204 goto fail_free_pio_data;
@@ -5458,6 +6220,8 @@ int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
5458 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) 6220 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL))
5459 goto fail_free_mce_banks; 6221 goto fail_free_mce_banks;
5460 6222
6223 kvm_async_pf_hash_reset(vcpu);
6224
5461 return 0; 6225 return 0;
5462fail_free_mce_banks: 6226fail_free_mce_banks:
5463 kfree(vcpu->arch.mce_banks); 6227 kfree(vcpu->arch.mce_banks);
@@ -5483,22 +6247,17 @@ void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
5483 free_page((unsigned long)vcpu->arch.pio_data); 6247 free_page((unsigned long)vcpu->arch.pio_data);
5484} 6248}
5485 6249
5486struct kvm *kvm_arch_create_vm(void) 6250int kvm_arch_init_vm(struct kvm *kvm)
5487{ 6251{
5488 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
5489
5490 if (!kvm)
5491 return ERR_PTR(-ENOMEM);
5492
5493 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages); 6252 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
5494 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head); 6253 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
5495 6254
5496 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */ 6255 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
5497 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap); 6256 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
5498 6257
5499 rdtscll(kvm->arch.vm_init_tsc); 6258 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
5500 6259
5501 return kvm; 6260 return 0;
5502} 6261}
5503 6262
5504static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu) 6263static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
@@ -5516,8 +6275,10 @@ static void kvm_free_vcpus(struct kvm *kvm)
5516 /* 6275 /*
5517 * Unpin any mmu pages first. 6276 * Unpin any mmu pages first.
5518 */ 6277 */
5519 kvm_for_each_vcpu(i, vcpu, kvm) 6278 kvm_for_each_vcpu(i, vcpu, kvm) {
6279 kvm_clear_async_pf_completion_queue(vcpu);
5520 kvm_unload_vcpu_mmu(vcpu); 6280 kvm_unload_vcpu_mmu(vcpu);
6281 }
5521 kvm_for_each_vcpu(i, vcpu, kvm) 6282 kvm_for_each_vcpu(i, vcpu, kvm)
5522 kvm_arch_vcpu_free(vcpu); 6283 kvm_arch_vcpu_free(vcpu);
5523 6284
@@ -5541,13 +6302,10 @@ void kvm_arch_destroy_vm(struct kvm *kvm)
5541 kfree(kvm->arch.vpic); 6302 kfree(kvm->arch.vpic);
5542 kfree(kvm->arch.vioapic); 6303 kfree(kvm->arch.vioapic);
5543 kvm_free_vcpus(kvm); 6304 kvm_free_vcpus(kvm);
5544 kvm_free_physmem(kvm);
5545 if (kvm->arch.apic_access_page) 6305 if (kvm->arch.apic_access_page)
5546 put_page(kvm->arch.apic_access_page); 6306 put_page(kvm->arch.apic_access_page);
5547 if (kvm->arch.ept_identity_pagetable) 6307 if (kvm->arch.ept_identity_pagetable)
5548 put_page(kvm->arch.ept_identity_pagetable); 6308 put_page(kvm->arch.ept_identity_pagetable);
5549 cleanup_srcu_struct(&kvm->srcu);
5550 kfree(kvm);
5551} 6309}
5552 6310
5553int kvm_arch_prepare_memory_region(struct kvm *kvm, 6311int kvm_arch_prepare_memory_region(struct kvm *kvm,
@@ -5595,7 +6353,7 @@ void kvm_arch_commit_memory_region(struct kvm *kvm,
5595 int user_alloc) 6353 int user_alloc)
5596{ 6354{
5597 6355
5598 int npages = mem->memory_size >> PAGE_SHIFT; 6356 int nr_mmu_pages = 0, npages = mem->memory_size >> PAGE_SHIFT;
5599 6357
5600 if (!user_alloc && !old.user_alloc && old.rmap && !npages) { 6358 if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
5601 int ret; 6359 int ret;
@@ -5610,12 +6368,12 @@ void kvm_arch_commit_memory_region(struct kvm *kvm,
5610 "failed to munmap memory\n"); 6368 "failed to munmap memory\n");
5611 } 6369 }
5612 6370
6371 if (!kvm->arch.n_requested_mmu_pages)
6372 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
6373
5613 spin_lock(&kvm->mmu_lock); 6374 spin_lock(&kvm->mmu_lock);
5614 if (!kvm->arch.n_requested_mmu_pages) { 6375 if (nr_mmu_pages)
5615 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
5616 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages); 6376 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
5617 }
5618
5619 kvm_mmu_slot_remove_write_access(kvm, mem->slot); 6377 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
5620 spin_unlock(&kvm->mmu_lock); 6378 spin_unlock(&kvm->mmu_lock);
5621} 6379}
@@ -5628,7 +6386,9 @@ void kvm_arch_flush_shadow(struct kvm *kvm)
5628 6386
5629int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu) 6387int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
5630{ 6388{
5631 return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE 6389 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6390 !vcpu->arch.apf.halted)
6391 || !list_empty_careful(&vcpu->async_pf.done)
5632 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED 6392 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
5633 || vcpu->arch.nmi_pending || 6393 || vcpu->arch.nmi_pending ||
5634 (kvm_arch_interrupt_allowed(vcpu) && 6394 (kvm_arch_interrupt_allowed(vcpu) &&
@@ -5647,7 +6407,7 @@ void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
5647 6407
5648 me = get_cpu(); 6408 me = get_cpu();
5649 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu)) 6409 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
5650 if (atomic_xchg(&vcpu->guest_mode, 0)) 6410 if (kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE)
5651 smp_send_reschedule(cpu); 6411 smp_send_reschedule(cpu);
5652 put_cpu(); 6412 put_cpu();
5653} 6413}
@@ -5683,9 +6443,151 @@ void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
5683 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip)) 6443 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
5684 rflags |= X86_EFLAGS_TF; 6444 rflags |= X86_EFLAGS_TF;
5685 kvm_x86_ops->set_rflags(vcpu, rflags); 6445 kvm_x86_ops->set_rflags(vcpu, rflags);
6446 kvm_make_request(KVM_REQ_EVENT, vcpu);
5686} 6447}
5687EXPORT_SYMBOL_GPL(kvm_set_rflags); 6448EXPORT_SYMBOL_GPL(kvm_set_rflags);
5688 6449
6450void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
6451{
6452 int r;
6453
6454 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
6455 is_error_page(work->page))
6456 return;
6457
6458 r = kvm_mmu_reload(vcpu);
6459 if (unlikely(r))
6460 return;
6461
6462 if (!vcpu->arch.mmu.direct_map &&
6463 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
6464 return;
6465
6466 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
6467}
6468
6469static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
6470{
6471 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
6472}
6473
6474static inline u32 kvm_async_pf_next_probe(u32 key)
6475{
6476 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
6477}
6478
6479static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6480{
6481 u32 key = kvm_async_pf_hash_fn(gfn);
6482
6483 while (vcpu->arch.apf.gfns[key] != ~0)
6484 key = kvm_async_pf_next_probe(key);
6485
6486 vcpu->arch.apf.gfns[key] = gfn;
6487}
6488
6489static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
6490{
6491 int i;
6492 u32 key = kvm_async_pf_hash_fn(gfn);
6493
6494 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
6495 (vcpu->arch.apf.gfns[key] != gfn &&
6496 vcpu->arch.apf.gfns[key] != ~0); i++)
6497 key = kvm_async_pf_next_probe(key);
6498
6499 return key;
6500}
6501
6502bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6503{
6504 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
6505}
6506
6507static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6508{
6509 u32 i, j, k;
6510
6511 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
6512 while (true) {
6513 vcpu->arch.apf.gfns[i] = ~0;
6514 do {
6515 j = kvm_async_pf_next_probe(j);
6516 if (vcpu->arch.apf.gfns[j] == ~0)
6517 return;
6518 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
6519 /*
6520 * k lies cyclically in ]i,j]
6521 * | i.k.j |
6522 * |....j i.k.| or |.k..j i...|
6523 */
6524 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
6525 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
6526 i = j;
6527 }
6528}
6529
6530static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
6531{
6532
6533 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
6534 sizeof(val));
6535}
6536
6537void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
6538 struct kvm_async_pf *work)
6539{
6540 struct x86_exception fault;
6541
6542 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
6543 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
6544
6545 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
6546 (vcpu->arch.apf.send_user_only &&
6547 kvm_x86_ops->get_cpl(vcpu) == 0))
6548 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
6549 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
6550 fault.vector = PF_VECTOR;
6551 fault.error_code_valid = true;
6552 fault.error_code = 0;
6553 fault.nested_page_fault = false;
6554 fault.address = work->arch.token;
6555 kvm_inject_page_fault(vcpu, &fault);
6556 }
6557}
6558
6559void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
6560 struct kvm_async_pf *work)
6561{
6562 struct x86_exception fault;
6563
6564 trace_kvm_async_pf_ready(work->arch.token, work->gva);
6565 if (is_error_page(work->page))
6566 work->arch.token = ~0; /* broadcast wakeup */
6567 else
6568 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
6569
6570 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
6571 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
6572 fault.vector = PF_VECTOR;
6573 fault.error_code_valid = true;
6574 fault.error_code = 0;
6575 fault.nested_page_fault = false;
6576 fault.address = work->arch.token;
6577 kvm_inject_page_fault(vcpu, &fault);
6578 }
6579 vcpu->arch.apf.halted = false;
6580}
6581
6582bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
6583{
6584 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
6585 return true;
6586 else
6587 return !kvm_event_needs_reinjection(vcpu) &&
6588 kvm_x86_ops->interrupt_allowed(vcpu);
6589}
6590
5689EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit); 6591EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
5690EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq); 6592EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
5691EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault); 6593EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
generated by cgit v1.2.2 (git 2.25.0) at 2025-03-14 15:42:00 -0400