1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
5 * This module enables machines with Intel VT-x extensions to run virtual
6 * machines without emulation or binary translation.
8 * Copyright (C) 2006 Qumranet, Inc.
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
16 #include <linux/frame.h>
17 #include <linux/highmem.h>
18 #include <linux/hrtimer.h>
19 #include <linux/kernel.h>
20 #include <linux/kvm_host.h>
21 #include <linux/module.h>
22 #include <linux/moduleparam.h>
23 #include <linux/mod_devicetable.h>
25 #include <linux/sched.h>
26 #include <linux/sched/smt.h>
27 #include <linux/slab.h>
28 #include <linux/tboot.h>
29 #include <linux/trace_events.h>
30 #include <linux/entry-kvm.h>
35 #include <asm/cpu_device_id.h>
36 #include <asm/debugreg.h>
38 #include <asm/fpu/internal.h>
40 #include <asm/irq_remapping.h>
41 #include <asm/kexec.h>
42 #include <asm/perf_event.h>
44 #include <asm/mmu_context.h>
45 #include <asm/mshyperv.h>
46 #include <asm/mwait.h>
47 #include <asm/spec-ctrl.h>
48 #include <asm/virtext.h>
51 #include "capabilities.h"
55 #include "kvm_cache_regs.h"
66 MODULE_AUTHOR("Qumranet");
67 MODULE_LICENSE("GPL");
70 static const struct x86_cpu_id vmx_cpu_id[] = {
71 X86_MATCH_FEATURE(X86_FEATURE_VMX, NULL),
74 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
77 bool __read_mostly enable_vpid = 1;
78 module_param_named(vpid, enable_vpid, bool, 0444);
80 static bool __read_mostly enable_vnmi = 1;
81 module_param_named(vnmi, enable_vnmi, bool, S_IRUGO);
83 bool __read_mostly flexpriority_enabled = 1;
84 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
86 bool __read_mostly enable_ept = 1;
87 module_param_named(ept, enable_ept, bool, S_IRUGO);
89 bool __read_mostly enable_unrestricted_guest = 1;
90 module_param_named(unrestricted_guest,
91 enable_unrestricted_guest, bool, S_IRUGO);
93 bool __read_mostly enable_ept_ad_bits = 1;
94 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
96 static bool __read_mostly emulate_invalid_guest_state = true;
97 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
99 static bool __read_mostly fasteoi = 1;
100 module_param(fasteoi, bool, S_IRUGO);
102 bool __read_mostly enable_apicv = 1;
103 module_param(enable_apicv, bool, S_IRUGO);
106 * If nested=1, nested virtualization is supported, i.e., guests may use
107 * VMX and be a hypervisor for its own guests. If nested=0, guests may not
108 * use VMX instructions.
110 static bool __read_mostly nested = 1;
111 module_param(nested, bool, S_IRUGO);
113 bool __read_mostly enable_pml = 1;
114 module_param_named(pml, enable_pml, bool, S_IRUGO);
116 static bool __read_mostly dump_invalid_vmcs = 0;
117 module_param(dump_invalid_vmcs, bool, 0644);
119 #define MSR_BITMAP_MODE_X2APIC 1
120 #define MSR_BITMAP_MODE_X2APIC_APICV 2
122 #define KVM_VMX_TSC_MULTIPLIER_MAX 0xffffffffffffffffULL
124 /* Guest_tsc -> host_tsc conversion requires 64-bit division. */
125 static int __read_mostly cpu_preemption_timer_multi;
126 static bool __read_mostly enable_preemption_timer = 1;
128 module_param_named(preemption_timer, enable_preemption_timer, bool, S_IRUGO);
131 #define KVM_VM_CR0_ALWAYS_OFF (X86_CR0_NW | X86_CR0_CD)
132 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR0_NE
133 #define KVM_VM_CR0_ALWAYS_ON \
134 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | \
135 X86_CR0_WP | X86_CR0_PG | X86_CR0_PE)
137 #define KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR4_VMXE
138 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
139 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
141 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
143 #define MSR_IA32_RTIT_STATUS_MASK (~(RTIT_STATUS_FILTEREN | \
144 RTIT_STATUS_CONTEXTEN | RTIT_STATUS_TRIGGEREN | \
145 RTIT_STATUS_ERROR | RTIT_STATUS_STOPPED | \
146 RTIT_STATUS_BYTECNT))
149 * List of MSRs that can be directly passed to the guest.
150 * In addition to these x2apic and PT MSRs are handled specially.
152 static u32 vmx_possible_passthrough_msrs[MAX_POSSIBLE_PASSTHROUGH_MSRS] = {
159 MSR_IA32_SYSENTER_CS,
160 MSR_IA32_SYSENTER_ESP,
161 MSR_IA32_SYSENTER_EIP,
163 MSR_CORE_C3_RESIDENCY,
164 MSR_CORE_C6_RESIDENCY,
165 MSR_CORE_C7_RESIDENCY,
169 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
170 * ple_gap: upper bound on the amount of time between two successive
171 * executions of PAUSE in a loop. Also indicate if ple enabled.
172 * According to test, this time is usually smaller than 128 cycles.
173 * ple_window: upper bound on the amount of time a guest is allowed to execute
174 * in a PAUSE loop. Tests indicate that most spinlocks are held for
175 * less than 2^12 cycles
176 * Time is measured based on a counter that runs at the same rate as the TSC,
177 * refer SDM volume 3b section 21.6.13 & 22.1.3.
179 static unsigned int ple_gap = KVM_DEFAULT_PLE_GAP;
180 module_param(ple_gap, uint, 0444);
182 static unsigned int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
183 module_param(ple_window, uint, 0444);
185 /* Default doubles per-vcpu window every exit. */
186 static unsigned int ple_window_grow = KVM_DEFAULT_PLE_WINDOW_GROW;
187 module_param(ple_window_grow, uint, 0444);
189 /* Default resets per-vcpu window every exit to ple_window. */
190 static unsigned int ple_window_shrink = KVM_DEFAULT_PLE_WINDOW_SHRINK;
191 module_param(ple_window_shrink, uint, 0444);
193 /* Default is to compute the maximum so we can never overflow. */
194 static unsigned int ple_window_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
195 module_param(ple_window_max, uint, 0444);
197 /* Default is SYSTEM mode, 1 for host-guest mode */
198 int __read_mostly pt_mode = PT_MODE_SYSTEM;
199 module_param(pt_mode, int, S_IRUGO);
201 static DEFINE_STATIC_KEY_FALSE(vmx_l1d_should_flush);
202 static DEFINE_STATIC_KEY_FALSE(vmx_l1d_flush_cond);
203 static DEFINE_MUTEX(vmx_l1d_flush_mutex);
205 /* Storage for pre module init parameter parsing */
206 static enum vmx_l1d_flush_state __read_mostly vmentry_l1d_flush_param = VMENTER_L1D_FLUSH_AUTO;
208 static const struct {
211 } vmentry_l1d_param[] = {
212 [VMENTER_L1D_FLUSH_AUTO] = {"auto", true},
213 [VMENTER_L1D_FLUSH_NEVER] = {"never", true},
214 [VMENTER_L1D_FLUSH_COND] = {"cond", true},
215 [VMENTER_L1D_FLUSH_ALWAYS] = {"always", true},
216 [VMENTER_L1D_FLUSH_EPT_DISABLED] = {"EPT disabled", false},
217 [VMENTER_L1D_FLUSH_NOT_REQUIRED] = {"not required", false},
220 #define L1D_CACHE_ORDER 4
221 static void *vmx_l1d_flush_pages;
223 static int vmx_setup_l1d_flush(enum vmx_l1d_flush_state l1tf)
228 if (!boot_cpu_has_bug(X86_BUG_L1TF)) {
229 l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED;
234 l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_EPT_DISABLED;
238 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES)) {
241 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, msr);
242 if (msr & ARCH_CAP_SKIP_VMENTRY_L1DFLUSH) {
243 l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED;
248 /* If set to auto use the default l1tf mitigation method */
249 if (l1tf == VMENTER_L1D_FLUSH_AUTO) {
250 switch (l1tf_mitigation) {
251 case L1TF_MITIGATION_OFF:
252 l1tf = VMENTER_L1D_FLUSH_NEVER;
254 case L1TF_MITIGATION_FLUSH_NOWARN:
255 case L1TF_MITIGATION_FLUSH:
256 case L1TF_MITIGATION_FLUSH_NOSMT:
257 l1tf = VMENTER_L1D_FLUSH_COND;
259 case L1TF_MITIGATION_FULL:
260 case L1TF_MITIGATION_FULL_FORCE:
261 l1tf = VMENTER_L1D_FLUSH_ALWAYS;
264 } else if (l1tf_mitigation == L1TF_MITIGATION_FULL_FORCE) {
265 l1tf = VMENTER_L1D_FLUSH_ALWAYS;
268 if (l1tf != VMENTER_L1D_FLUSH_NEVER && !vmx_l1d_flush_pages &&
269 !boot_cpu_has(X86_FEATURE_FLUSH_L1D)) {
271 * This allocation for vmx_l1d_flush_pages is not tied to a VM
272 * lifetime and so should not be charged to a memcg.
274 page = alloc_pages(GFP_KERNEL, L1D_CACHE_ORDER);
277 vmx_l1d_flush_pages = page_address(page);
280 * Initialize each page with a different pattern in
281 * order to protect against KSM in the nested
282 * virtualization case.
284 for (i = 0; i < 1u << L1D_CACHE_ORDER; ++i) {
285 memset(vmx_l1d_flush_pages + i * PAGE_SIZE, i + 1,
290 l1tf_vmx_mitigation = l1tf;
292 if (l1tf != VMENTER_L1D_FLUSH_NEVER)
293 static_branch_enable(&vmx_l1d_should_flush);
295 static_branch_disable(&vmx_l1d_should_flush);
297 if (l1tf == VMENTER_L1D_FLUSH_COND)
298 static_branch_enable(&vmx_l1d_flush_cond);
300 static_branch_disable(&vmx_l1d_flush_cond);
304 static int vmentry_l1d_flush_parse(const char *s)
309 for (i = 0; i < ARRAY_SIZE(vmentry_l1d_param); i++) {
310 if (vmentry_l1d_param[i].for_parse &&
311 sysfs_streq(s, vmentry_l1d_param[i].option))
318 static int vmentry_l1d_flush_set(const char *s, const struct kernel_param *kp)
322 l1tf = vmentry_l1d_flush_parse(s);
326 if (!boot_cpu_has(X86_BUG_L1TF))
330 * Has vmx_init() run already? If not then this is the pre init
331 * parameter parsing. In that case just store the value and let
332 * vmx_init() do the proper setup after enable_ept has been
335 if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_AUTO) {
336 vmentry_l1d_flush_param = l1tf;
340 mutex_lock(&vmx_l1d_flush_mutex);
341 ret = vmx_setup_l1d_flush(l1tf);
342 mutex_unlock(&vmx_l1d_flush_mutex);
346 static int vmentry_l1d_flush_get(char *s, const struct kernel_param *kp)
348 if (WARN_ON_ONCE(l1tf_vmx_mitigation >= ARRAY_SIZE(vmentry_l1d_param)))
349 return sprintf(s, "???\n");
351 return sprintf(s, "%s\n", vmentry_l1d_param[l1tf_vmx_mitigation].option);
354 static const struct kernel_param_ops vmentry_l1d_flush_ops = {
355 .set = vmentry_l1d_flush_set,
356 .get = vmentry_l1d_flush_get,
358 module_param_cb(vmentry_l1d_flush, &vmentry_l1d_flush_ops, NULL, 0644);
360 static u32 vmx_segment_access_rights(struct kvm_segment *var);
361 static __always_inline void vmx_disable_intercept_for_msr(struct kvm_vcpu *vcpu,
364 void vmx_vmexit(void);
366 #define vmx_insn_failed(fmt...) \
369 pr_warn_ratelimited(fmt); \
372 asmlinkage void vmread_error(unsigned long field, bool fault)
375 kvm_spurious_fault();
377 vmx_insn_failed("kvm: vmread failed: field=%lx\n", field);
380 noinline void vmwrite_error(unsigned long field, unsigned long value)
382 vmx_insn_failed("kvm: vmwrite failed: field=%lx val=%lx err=%d\n",
383 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
386 noinline void vmclear_error(struct vmcs *vmcs, u64 phys_addr)
388 vmx_insn_failed("kvm: vmclear failed: %p/%llx\n", vmcs, phys_addr);
391 noinline void vmptrld_error(struct vmcs *vmcs, u64 phys_addr)
393 vmx_insn_failed("kvm: vmptrld failed: %p/%llx\n", vmcs, phys_addr);
396 noinline void invvpid_error(unsigned long ext, u16 vpid, gva_t gva)
398 vmx_insn_failed("kvm: invvpid failed: ext=0x%lx vpid=%u gva=0x%lx\n",
402 noinline void invept_error(unsigned long ext, u64 eptp, gpa_t gpa)
404 vmx_insn_failed("kvm: invept failed: ext=0x%lx eptp=%llx gpa=0x%llx\n",
408 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
409 DEFINE_PER_CPU(struct vmcs *, current_vmcs);
411 * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
412 * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
414 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
416 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
417 static DEFINE_SPINLOCK(vmx_vpid_lock);
419 struct vmcs_config vmcs_config;
420 struct vmx_capability vmx_capability;
422 #define VMX_SEGMENT_FIELD(seg) \
423 [VCPU_SREG_##seg] = { \
424 .selector = GUEST_##seg##_SELECTOR, \
425 .base = GUEST_##seg##_BASE, \
426 .limit = GUEST_##seg##_LIMIT, \
427 .ar_bytes = GUEST_##seg##_AR_BYTES, \
430 static const struct kvm_vmx_segment_field {
435 } kvm_vmx_segment_fields[] = {
436 VMX_SEGMENT_FIELD(CS),
437 VMX_SEGMENT_FIELD(DS),
438 VMX_SEGMENT_FIELD(ES),
439 VMX_SEGMENT_FIELD(FS),
440 VMX_SEGMENT_FIELD(GS),
441 VMX_SEGMENT_FIELD(SS),
442 VMX_SEGMENT_FIELD(TR),
443 VMX_SEGMENT_FIELD(LDTR),
446 static inline void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
448 vmx->segment_cache.bitmask = 0;
451 static unsigned long host_idt_base;
454 * Though SYSCALL is only supported in 64-bit mode on Intel CPUs, kvm
455 * will emulate SYSCALL in legacy mode if the vendor string in guest
456 * CPUID.0:{EBX,ECX,EDX} is "AuthenticAMD" or "AMDisbetter!" To
457 * support this emulation, IA32_STAR must always be included in
458 * vmx_uret_msrs_list[], even in i386 builds.
460 static const u32 vmx_uret_msrs_list[] = {
462 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
464 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
468 #if IS_ENABLED(CONFIG_HYPERV)
469 static bool __read_mostly enlightened_vmcs = true;
470 module_param(enlightened_vmcs, bool, 0444);
472 /* check_ept_pointer() should be under protection of ept_pointer_lock. */
473 static void check_ept_pointer_match(struct kvm *kvm)
475 struct kvm_vcpu *vcpu;
476 u64 tmp_eptp = INVALID_PAGE;
479 kvm_for_each_vcpu(i, vcpu, kvm) {
480 if (!VALID_PAGE(tmp_eptp)) {
481 tmp_eptp = to_vmx(vcpu)->ept_pointer;
482 } else if (tmp_eptp != to_vmx(vcpu)->ept_pointer) {
483 to_kvm_vmx(kvm)->ept_pointers_match
484 = EPT_POINTERS_MISMATCH;
489 to_kvm_vmx(kvm)->ept_pointers_match = EPT_POINTERS_MATCH;
492 static int kvm_fill_hv_flush_list_func(struct hv_guest_mapping_flush_list *flush,
495 struct kvm_tlb_range *range = data;
497 return hyperv_fill_flush_guest_mapping_list(flush, range->start_gfn,
501 static inline int __hv_remote_flush_tlb_with_range(struct kvm *kvm,
502 struct kvm_vcpu *vcpu, struct kvm_tlb_range *range)
504 u64 ept_pointer = to_vmx(vcpu)->ept_pointer;
507 * FLUSH_GUEST_PHYSICAL_ADDRESS_SPACE hypercall needs address
508 * of the base of EPT PML4 table, strip off EPT configuration
512 return hyperv_flush_guest_mapping_range(ept_pointer & PAGE_MASK,
513 kvm_fill_hv_flush_list_func, (void *)range);
515 return hyperv_flush_guest_mapping(ept_pointer & PAGE_MASK);
518 static int hv_remote_flush_tlb_with_range(struct kvm *kvm,
519 struct kvm_tlb_range *range)
521 struct kvm_vcpu *vcpu;
524 spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
526 if (to_kvm_vmx(kvm)->ept_pointers_match == EPT_POINTERS_CHECK)
527 check_ept_pointer_match(kvm);
529 if (to_kvm_vmx(kvm)->ept_pointers_match != EPT_POINTERS_MATCH) {
530 kvm_for_each_vcpu(i, vcpu, kvm) {
531 /* If ept_pointer is invalid pointer, bypass flush request. */
532 if (VALID_PAGE(to_vmx(vcpu)->ept_pointer))
533 ret |= __hv_remote_flush_tlb_with_range(
537 ret = __hv_remote_flush_tlb_with_range(kvm,
538 kvm_get_vcpu(kvm, 0), range);
541 spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
544 static int hv_remote_flush_tlb(struct kvm *kvm)
546 return hv_remote_flush_tlb_with_range(kvm, NULL);
549 static int hv_enable_direct_tlbflush(struct kvm_vcpu *vcpu)
551 struct hv_enlightened_vmcs *evmcs;
552 struct hv_partition_assist_pg **p_hv_pa_pg =
553 &vcpu->kvm->arch.hyperv.hv_pa_pg;
555 * Synthetic VM-Exit is not enabled in current code and so All
556 * evmcs in singe VM shares same assist page.
559 *p_hv_pa_pg = kzalloc(PAGE_SIZE, GFP_KERNEL);
564 evmcs = (struct hv_enlightened_vmcs *)to_vmx(vcpu)->loaded_vmcs->vmcs;
566 evmcs->partition_assist_page =
568 evmcs->hv_vm_id = (unsigned long)vcpu->kvm;
569 evmcs->hv_enlightenments_control.nested_flush_hypercall = 1;
574 #endif /* IS_ENABLED(CONFIG_HYPERV) */
577 * Comment's format: document - errata name - stepping - processor name.
579 * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
581 static u32 vmx_preemption_cpu_tfms[] = {
582 /* 323344.pdf - BA86 - D0 - Xeon 7500 Series */
584 /* 323056.pdf - AAX65 - C2 - Xeon L3406 */
585 /* 322814.pdf - AAT59 - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
586 /* 322911.pdf - AAU65 - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
588 /* 322911.pdf - AAU65 - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
590 /* 322373.pdf - AAO95 - B1 - Xeon 3400 Series */
591 /* 322166.pdf - AAN92 - B1 - i7-800 and i5-700 Desktop */
593 * 320767.pdf - AAP86 - B1 -
594 * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
597 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
599 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
601 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
603 /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
604 /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
605 /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
607 /* Xeon E3-1220 V2 */
611 static inline bool cpu_has_broken_vmx_preemption_timer(void)
613 u32 eax = cpuid_eax(0x00000001), i;
615 /* Clear the reserved bits */
616 eax &= ~(0x3U << 14 | 0xfU << 28);
617 for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
618 if (eax == vmx_preemption_cpu_tfms[i])
624 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
626 return flexpriority_enabled && lapic_in_kernel(vcpu);
629 static inline bool report_flexpriority(void)
631 return flexpriority_enabled;
634 static int possible_passthrough_msr_slot(u32 msr)
638 for (i = 0; i < ARRAY_SIZE(vmx_possible_passthrough_msrs); i++)
639 if (vmx_possible_passthrough_msrs[i] == msr)
645 static bool is_valid_passthrough_msr(u32 msr)
650 case 0x800 ... 0x8ff:
651 /* x2APIC MSRs. These are handled in vmx_update_msr_bitmap_x2apic() */
653 case MSR_IA32_RTIT_STATUS:
654 case MSR_IA32_RTIT_OUTPUT_BASE:
655 case MSR_IA32_RTIT_OUTPUT_MASK:
656 case MSR_IA32_RTIT_CR3_MATCH:
657 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
658 /* PT MSRs. These are handled in pt_update_intercept_for_msr() */
662 r = possible_passthrough_msr_slot(msr) != -ENOENT;
664 WARN(!r, "Invalid MSR %x, please adapt vmx_possible_passthrough_msrs[]", msr);
669 static inline int __vmx_find_uret_msr(struct vcpu_vmx *vmx, u32 msr)
673 for (i = 0; i < vmx->nr_uret_msrs; ++i)
674 if (vmx_uret_msrs_list[vmx->guest_uret_msrs[i].slot] == msr)
679 struct vmx_uret_msr *vmx_find_uret_msr(struct vcpu_vmx *vmx, u32 msr)
683 i = __vmx_find_uret_msr(vmx, msr);
685 return &vmx->guest_uret_msrs[i];
689 static int vmx_set_guest_uret_msr(struct vcpu_vmx *vmx,
690 struct vmx_uret_msr *msr, u64 data)
694 u64 old_msr_data = msr->data;
696 if (msr - vmx->guest_uret_msrs < vmx->nr_active_uret_msrs) {
698 ret = kvm_set_user_return_msr(msr->slot, msr->data, msr->mask);
701 msr->data = old_msr_data;
706 #ifdef CONFIG_KEXEC_CORE
707 static void crash_vmclear_local_loaded_vmcss(void)
709 int cpu = raw_smp_processor_id();
710 struct loaded_vmcs *v;
712 list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
713 loaded_vmcss_on_cpu_link)
716 #endif /* CONFIG_KEXEC_CORE */
718 static void __loaded_vmcs_clear(void *arg)
720 struct loaded_vmcs *loaded_vmcs = arg;
721 int cpu = raw_smp_processor_id();
723 if (loaded_vmcs->cpu != cpu)
724 return; /* vcpu migration can race with cpu offline */
725 if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
726 per_cpu(current_vmcs, cpu) = NULL;
728 vmcs_clear(loaded_vmcs->vmcs);
729 if (loaded_vmcs->shadow_vmcs && loaded_vmcs->launched)
730 vmcs_clear(loaded_vmcs->shadow_vmcs);
732 list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
735 * Ensure all writes to loaded_vmcs, including deleting it from its
736 * current percpu list, complete before setting loaded_vmcs->vcpu to
737 * -1, otherwise a different cpu can see vcpu == -1 first and add
738 * loaded_vmcs to its percpu list before it's deleted from this cpu's
739 * list. Pairs with the smp_rmb() in vmx_vcpu_load_vmcs().
743 loaded_vmcs->cpu = -1;
744 loaded_vmcs->launched = 0;
747 void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
749 int cpu = loaded_vmcs->cpu;
752 smp_call_function_single(cpu,
753 __loaded_vmcs_clear, loaded_vmcs, 1);
756 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
760 u32 mask = 1 << (seg * SEG_FIELD_NR + field);
762 if (!kvm_register_is_available(&vmx->vcpu, VCPU_EXREG_SEGMENTS)) {
763 kvm_register_mark_available(&vmx->vcpu, VCPU_EXREG_SEGMENTS);
764 vmx->segment_cache.bitmask = 0;
766 ret = vmx->segment_cache.bitmask & mask;
767 vmx->segment_cache.bitmask |= mask;
771 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
773 u16 *p = &vmx->segment_cache.seg[seg].selector;
775 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
776 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
780 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
782 ulong *p = &vmx->segment_cache.seg[seg].base;
784 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
785 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
789 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
791 u32 *p = &vmx->segment_cache.seg[seg].limit;
793 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
794 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
798 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
800 u32 *p = &vmx->segment_cache.seg[seg].ar;
802 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
803 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
807 void update_exception_bitmap(struct kvm_vcpu *vcpu)
811 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
812 (1u << DB_VECTOR) | (1u << AC_VECTOR);
814 * Guest access to VMware backdoor ports could legitimately
815 * trigger #GP because of TSS I/O permission bitmap.
816 * We intercept those #GP and allow access to them anyway
819 if (enable_vmware_backdoor)
820 eb |= (1u << GP_VECTOR);
821 if ((vcpu->guest_debug &
822 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
823 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
824 eb |= 1u << BP_VECTOR;
825 if (to_vmx(vcpu)->rmode.vm86_active)
827 if (!vmx_need_pf_intercept(vcpu))
828 eb &= ~(1u << PF_VECTOR);
830 /* When we are running a nested L2 guest and L1 specified for it a
831 * certain exception bitmap, we must trap the same exceptions and pass
832 * them to L1. When running L2, we will only handle the exceptions
833 * specified above if L1 did not want them.
835 if (is_guest_mode(vcpu))
836 eb |= get_vmcs12(vcpu)->exception_bitmap;
838 vmcs_write32(EXCEPTION_BITMAP, eb);
842 * Check if MSR is intercepted for currently loaded MSR bitmap.
844 static bool msr_write_intercepted(struct kvm_vcpu *vcpu, u32 msr)
846 unsigned long *msr_bitmap;
847 int f = sizeof(unsigned long);
849 if (!cpu_has_vmx_msr_bitmap())
852 msr_bitmap = to_vmx(vcpu)->loaded_vmcs->msr_bitmap;
855 return !!test_bit(msr, msr_bitmap + 0x800 / f);
856 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
858 return !!test_bit(msr, msr_bitmap + 0xc00 / f);
864 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
865 unsigned long entry, unsigned long exit)
867 vm_entry_controls_clearbit(vmx, entry);
868 vm_exit_controls_clearbit(vmx, exit);
871 int vmx_find_loadstore_msr_slot(struct vmx_msrs *m, u32 msr)
875 for (i = 0; i < m->nr; ++i) {
876 if (m->val[i].index == msr)
882 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
885 struct msr_autoload *m = &vmx->msr_autoload;
889 if (cpu_has_load_ia32_efer()) {
890 clear_atomic_switch_msr_special(vmx,
891 VM_ENTRY_LOAD_IA32_EFER,
892 VM_EXIT_LOAD_IA32_EFER);
896 case MSR_CORE_PERF_GLOBAL_CTRL:
897 if (cpu_has_load_perf_global_ctrl()) {
898 clear_atomic_switch_msr_special(vmx,
899 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
900 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
905 i = vmx_find_loadstore_msr_slot(&m->guest, msr);
909 m->guest.val[i] = m->guest.val[m->guest.nr];
910 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
913 i = vmx_find_loadstore_msr_slot(&m->host, msr);
918 m->host.val[i] = m->host.val[m->host.nr];
919 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
922 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
923 unsigned long entry, unsigned long exit,
924 unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
925 u64 guest_val, u64 host_val)
927 vmcs_write64(guest_val_vmcs, guest_val);
928 if (host_val_vmcs != HOST_IA32_EFER)
929 vmcs_write64(host_val_vmcs, host_val);
930 vm_entry_controls_setbit(vmx, entry);
931 vm_exit_controls_setbit(vmx, exit);
934 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
935 u64 guest_val, u64 host_val, bool entry_only)
938 struct msr_autoload *m = &vmx->msr_autoload;
942 if (cpu_has_load_ia32_efer()) {
943 add_atomic_switch_msr_special(vmx,
944 VM_ENTRY_LOAD_IA32_EFER,
945 VM_EXIT_LOAD_IA32_EFER,
948 guest_val, host_val);
952 case MSR_CORE_PERF_GLOBAL_CTRL:
953 if (cpu_has_load_perf_global_ctrl()) {
954 add_atomic_switch_msr_special(vmx,
955 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
956 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
957 GUEST_IA32_PERF_GLOBAL_CTRL,
958 HOST_IA32_PERF_GLOBAL_CTRL,
959 guest_val, host_val);
963 case MSR_IA32_PEBS_ENABLE:
964 /* PEBS needs a quiescent period after being disabled (to write
965 * a record). Disabling PEBS through VMX MSR swapping doesn't
966 * provide that period, so a CPU could write host's record into
969 wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
972 i = vmx_find_loadstore_msr_slot(&m->guest, msr);
974 j = vmx_find_loadstore_msr_slot(&m->host, msr);
976 if ((i < 0 && m->guest.nr == MAX_NR_LOADSTORE_MSRS) ||
977 (j < 0 && m->host.nr == MAX_NR_LOADSTORE_MSRS)) {
978 printk_once(KERN_WARNING "Not enough msr switch entries. "
979 "Can't add msr %x\n", msr);
984 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
986 m->guest.val[i].index = msr;
987 m->guest.val[i].value = guest_val;
994 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
996 m->host.val[j].index = msr;
997 m->host.val[j].value = host_val;
1000 static bool update_transition_efer(struct vcpu_vmx *vmx)
1002 u64 guest_efer = vmx->vcpu.arch.efer;
1003 u64 ignore_bits = 0;
1006 /* Shadow paging assumes NX to be available. */
1008 guest_efer |= EFER_NX;
1011 * LMA and LME handled by hardware; SCE meaningless outside long mode.
1013 ignore_bits |= EFER_SCE;
1014 #ifdef CONFIG_X86_64
1015 ignore_bits |= EFER_LMA | EFER_LME;
1016 /* SCE is meaningful only in long mode on Intel */
1017 if (guest_efer & EFER_LMA)
1018 ignore_bits &= ~(u64)EFER_SCE;
1022 * On EPT, we can't emulate NX, so we must switch EFER atomically.
1023 * On CPUs that support "load IA32_EFER", always switch EFER
1024 * atomically, since it's faster than switching it manually.
1026 if (cpu_has_load_ia32_efer() ||
1027 (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
1028 if (!(guest_efer & EFER_LMA))
1029 guest_efer &= ~EFER_LME;
1030 if (guest_efer != host_efer)
1031 add_atomic_switch_msr(vmx, MSR_EFER,
1032 guest_efer, host_efer, false);
1034 clear_atomic_switch_msr(vmx, MSR_EFER);
1038 i = __vmx_find_uret_msr(vmx, MSR_EFER);
1042 clear_atomic_switch_msr(vmx, MSR_EFER);
1044 guest_efer &= ~ignore_bits;
1045 guest_efer |= host_efer & ignore_bits;
1047 vmx->guest_uret_msrs[i].data = guest_efer;
1048 vmx->guest_uret_msrs[i].mask = ~ignore_bits;
1053 #ifdef CONFIG_X86_32
1055 * On 32-bit kernels, VM exits still load the FS and GS bases from the
1056 * VMCS rather than the segment table. KVM uses this helper to figure
1057 * out the current bases to poke them into the VMCS before entry.
1059 static unsigned long segment_base(u16 selector)
1061 struct desc_struct *table;
1064 if (!(selector & ~SEGMENT_RPL_MASK))
1067 table = get_current_gdt_ro();
1069 if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
1070 u16 ldt_selector = kvm_read_ldt();
1072 if (!(ldt_selector & ~SEGMENT_RPL_MASK))
1075 table = (struct desc_struct *)segment_base(ldt_selector);
1077 v = get_desc_base(&table[selector >> 3]);
1082 static inline bool pt_can_write_msr(struct vcpu_vmx *vmx)
1084 return vmx_pt_mode_is_host_guest() &&
1085 !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
1088 static inline bool pt_output_base_valid(struct kvm_vcpu *vcpu, u64 base)
1090 /* The base must be 128-byte aligned and a legal physical address. */
1091 return !kvm_vcpu_is_illegal_gpa(vcpu, base) && !(base & 0x7f);
1094 static inline void pt_load_msr(struct pt_ctx *ctx, u32 addr_range)
1098 wrmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1099 wrmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1100 wrmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1101 wrmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1102 for (i = 0; i < addr_range; i++) {
1103 wrmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1104 wrmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1108 static inline void pt_save_msr(struct pt_ctx *ctx, u32 addr_range)
1112 rdmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1113 rdmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1114 rdmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1115 rdmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1116 for (i = 0; i < addr_range; i++) {
1117 rdmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1118 rdmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1122 static void pt_guest_enter(struct vcpu_vmx *vmx)
1124 if (vmx_pt_mode_is_system())
1128 * GUEST_IA32_RTIT_CTL is already set in the VMCS.
1129 * Save host state before VM entry.
1131 rdmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1132 if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1133 wrmsrl(MSR_IA32_RTIT_CTL, 0);
1134 pt_save_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1135 pt_load_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1139 static void pt_guest_exit(struct vcpu_vmx *vmx)
1141 if (vmx_pt_mode_is_system())
1144 if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1145 pt_save_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1146 pt_load_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1149 /* Reload host state (IA32_RTIT_CTL will be cleared on VM exit). */
1150 wrmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1153 void vmx_set_host_fs_gs(struct vmcs_host_state *host, u16 fs_sel, u16 gs_sel,
1154 unsigned long fs_base, unsigned long gs_base)
1156 if (unlikely(fs_sel != host->fs_sel)) {
1158 vmcs_write16(HOST_FS_SELECTOR, fs_sel);
1160 vmcs_write16(HOST_FS_SELECTOR, 0);
1161 host->fs_sel = fs_sel;
1163 if (unlikely(gs_sel != host->gs_sel)) {
1165 vmcs_write16(HOST_GS_SELECTOR, gs_sel);
1167 vmcs_write16(HOST_GS_SELECTOR, 0);
1168 host->gs_sel = gs_sel;
1170 if (unlikely(fs_base != host->fs_base)) {
1171 vmcs_writel(HOST_FS_BASE, fs_base);
1172 host->fs_base = fs_base;
1174 if (unlikely(gs_base != host->gs_base)) {
1175 vmcs_writel(HOST_GS_BASE, gs_base);
1176 host->gs_base = gs_base;
1180 void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu)
1182 struct vcpu_vmx *vmx = to_vmx(vcpu);
1183 struct vmcs_host_state *host_state;
1184 #ifdef CONFIG_X86_64
1185 int cpu = raw_smp_processor_id();
1187 unsigned long fs_base, gs_base;
1191 vmx->req_immediate_exit = false;
1194 * Note that guest MSRs to be saved/restored can also be changed
1195 * when guest state is loaded. This happens when guest transitions
1196 * to/from long-mode by setting MSR_EFER.LMA.
1198 if (!vmx->guest_uret_msrs_loaded) {
1199 vmx->guest_uret_msrs_loaded = true;
1200 for (i = 0; i < vmx->nr_active_uret_msrs; ++i)
1201 kvm_set_user_return_msr(vmx->guest_uret_msrs[i].slot,
1202 vmx->guest_uret_msrs[i].data,
1203 vmx->guest_uret_msrs[i].mask);
1207 if (vmx->nested.need_vmcs12_to_shadow_sync)
1208 nested_sync_vmcs12_to_shadow(vcpu);
1210 if (vmx->guest_state_loaded)
1213 host_state = &vmx->loaded_vmcs->host_state;
1216 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
1217 * allow segment selectors with cpl > 0 or ti == 1.
1219 host_state->ldt_sel = kvm_read_ldt();
1221 #ifdef CONFIG_X86_64
1222 savesegment(ds, host_state->ds_sel);
1223 savesegment(es, host_state->es_sel);
1225 gs_base = cpu_kernelmode_gs_base(cpu);
1226 if (likely(is_64bit_mm(current->mm))) {
1227 current_save_fsgs();
1228 fs_sel = current->thread.fsindex;
1229 gs_sel = current->thread.gsindex;
1230 fs_base = current->thread.fsbase;
1231 vmx->msr_host_kernel_gs_base = current->thread.gsbase;
1233 savesegment(fs, fs_sel);
1234 savesegment(gs, gs_sel);
1235 fs_base = read_msr(MSR_FS_BASE);
1236 vmx->msr_host_kernel_gs_base = read_msr(MSR_KERNEL_GS_BASE);
1239 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1241 savesegment(fs, fs_sel);
1242 savesegment(gs, gs_sel);
1243 fs_base = segment_base(fs_sel);
1244 gs_base = segment_base(gs_sel);
1247 vmx_set_host_fs_gs(host_state, fs_sel, gs_sel, fs_base, gs_base);
1248 vmx->guest_state_loaded = true;
1251 static void vmx_prepare_switch_to_host(struct vcpu_vmx *vmx)
1253 struct vmcs_host_state *host_state;
1255 if (!vmx->guest_state_loaded)
1258 host_state = &vmx->loaded_vmcs->host_state;
1260 ++vmx->vcpu.stat.host_state_reload;
1262 #ifdef CONFIG_X86_64
1263 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1265 if (host_state->ldt_sel || (host_state->gs_sel & 7)) {
1266 kvm_load_ldt(host_state->ldt_sel);
1267 #ifdef CONFIG_X86_64
1268 load_gs_index(host_state->gs_sel);
1270 loadsegment(gs, host_state->gs_sel);
1273 if (host_state->fs_sel & 7)
1274 loadsegment(fs, host_state->fs_sel);
1275 #ifdef CONFIG_X86_64
1276 if (unlikely(host_state->ds_sel | host_state->es_sel)) {
1277 loadsegment(ds, host_state->ds_sel);
1278 loadsegment(es, host_state->es_sel);
1281 invalidate_tss_limit();
1282 #ifdef CONFIG_X86_64
1283 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1285 load_fixmap_gdt(raw_smp_processor_id());
1286 vmx->guest_state_loaded = false;
1287 vmx->guest_uret_msrs_loaded = false;
1290 #ifdef CONFIG_X86_64
1291 static u64 vmx_read_guest_kernel_gs_base(struct vcpu_vmx *vmx)
1294 if (vmx->guest_state_loaded)
1295 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1297 return vmx->msr_guest_kernel_gs_base;
1300 static void vmx_write_guest_kernel_gs_base(struct vcpu_vmx *vmx, u64 data)
1303 if (vmx->guest_state_loaded)
1304 wrmsrl(MSR_KERNEL_GS_BASE, data);
1306 vmx->msr_guest_kernel_gs_base = data;
1310 void vmx_vcpu_load_vmcs(struct kvm_vcpu *vcpu, int cpu,
1311 struct loaded_vmcs *buddy)
1313 struct vcpu_vmx *vmx = to_vmx(vcpu);
1314 bool already_loaded = vmx->loaded_vmcs->cpu == cpu;
1317 if (!already_loaded) {
1318 loaded_vmcs_clear(vmx->loaded_vmcs);
1319 local_irq_disable();
1322 * Ensure loaded_vmcs->cpu is read before adding loaded_vmcs to
1323 * this cpu's percpu list, otherwise it may not yet be deleted
1324 * from its previous cpu's percpu list. Pairs with the
1325 * smb_wmb() in __loaded_vmcs_clear().
1329 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1330 &per_cpu(loaded_vmcss_on_cpu, cpu));
1334 prev = per_cpu(current_vmcs, cpu);
1335 if (prev != vmx->loaded_vmcs->vmcs) {
1336 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1337 vmcs_load(vmx->loaded_vmcs->vmcs);
1340 * No indirect branch prediction barrier needed when switching
1341 * the active VMCS within a guest, e.g. on nested VM-Enter.
1342 * The L1 VMM can protect itself with retpolines, IBPB or IBRS.
1344 if (!buddy || WARN_ON_ONCE(buddy->vmcs != prev))
1345 indirect_branch_prediction_barrier();
1348 if (!already_loaded) {
1349 void *gdt = get_current_gdt_ro();
1350 unsigned long sysenter_esp;
1353 * Flush all EPTP/VPID contexts, the new pCPU may have stale
1354 * TLB entries from its previous association with the vCPU.
1356 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1359 * Linux uses per-cpu TSS and GDT, so set these when switching
1360 * processors. See 22.2.4.
1362 vmcs_writel(HOST_TR_BASE,
1363 (unsigned long)&get_cpu_entry_area(cpu)->tss.x86_tss);
1364 vmcs_writel(HOST_GDTR_BASE, (unsigned long)gdt); /* 22.2.4 */
1366 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1367 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1369 vmx->loaded_vmcs->cpu = cpu;
1372 /* Setup TSC multiplier */
1373 if (kvm_has_tsc_control &&
1374 vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio)
1375 decache_tsc_multiplier(vmx);
1379 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1380 * vcpu mutex is already taken.
1382 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1384 struct vcpu_vmx *vmx = to_vmx(vcpu);
1386 vmx_vcpu_load_vmcs(vcpu, cpu, NULL);
1388 vmx_vcpu_pi_load(vcpu, cpu);
1390 vmx->host_debugctlmsr = get_debugctlmsr();
1393 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1395 vmx_vcpu_pi_put(vcpu);
1397 vmx_prepare_switch_to_host(to_vmx(vcpu));
1400 static bool emulation_required(struct kvm_vcpu *vcpu)
1402 return emulate_invalid_guest_state && !vmx_guest_state_valid(vcpu);
1405 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1407 struct vcpu_vmx *vmx = to_vmx(vcpu);
1408 unsigned long rflags, save_rflags;
1410 if (!kvm_register_is_available(vcpu, VCPU_EXREG_RFLAGS)) {
1411 kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1412 rflags = vmcs_readl(GUEST_RFLAGS);
1413 if (vmx->rmode.vm86_active) {
1414 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1415 save_rflags = vmx->rmode.save_rflags;
1416 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1418 vmx->rflags = rflags;
1423 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1425 struct vcpu_vmx *vmx = to_vmx(vcpu);
1426 unsigned long old_rflags;
1428 if (is_unrestricted_guest(vcpu)) {
1429 kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1430 vmx->rflags = rflags;
1431 vmcs_writel(GUEST_RFLAGS, rflags);
1435 old_rflags = vmx_get_rflags(vcpu);
1436 vmx->rflags = rflags;
1437 if (vmx->rmode.vm86_active) {
1438 vmx->rmode.save_rflags = rflags;
1439 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1441 vmcs_writel(GUEST_RFLAGS, rflags);
1443 if ((old_rflags ^ vmx->rflags) & X86_EFLAGS_VM)
1444 vmx->emulation_required = emulation_required(vcpu);
1447 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
1449 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1452 if (interruptibility & GUEST_INTR_STATE_STI)
1453 ret |= KVM_X86_SHADOW_INT_STI;
1454 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1455 ret |= KVM_X86_SHADOW_INT_MOV_SS;
1460 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1462 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1463 u32 interruptibility = interruptibility_old;
1465 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1467 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1468 interruptibility |= GUEST_INTR_STATE_MOV_SS;
1469 else if (mask & KVM_X86_SHADOW_INT_STI)
1470 interruptibility |= GUEST_INTR_STATE_STI;
1472 if ((interruptibility != interruptibility_old))
1473 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1476 static int vmx_rtit_ctl_check(struct kvm_vcpu *vcpu, u64 data)
1478 struct vcpu_vmx *vmx = to_vmx(vcpu);
1479 unsigned long value;
1482 * Any MSR write that attempts to change bits marked reserved will
1485 if (data & vmx->pt_desc.ctl_bitmask)
1489 * Any attempt to modify IA32_RTIT_CTL while TraceEn is set will
1490 * result in a #GP unless the same write also clears TraceEn.
1492 if ((vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) &&
1493 ((vmx->pt_desc.guest.ctl ^ data) & ~RTIT_CTL_TRACEEN))
1497 * WRMSR to IA32_RTIT_CTL that sets TraceEn but clears this bit
1498 * and FabricEn would cause #GP, if
1499 * CPUID.(EAX=14H, ECX=0):ECX.SNGLRGNOUT[bit 2] = 0
1501 if ((data & RTIT_CTL_TRACEEN) && !(data & RTIT_CTL_TOPA) &&
1502 !(data & RTIT_CTL_FABRIC_EN) &&
1503 !intel_pt_validate_cap(vmx->pt_desc.caps,
1504 PT_CAP_single_range_output))
1508 * MTCFreq, CycThresh and PSBFreq encodings check, any MSR write that
1509 * utilize encodings marked reserved will casue a #GP fault.
1511 value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc_periods);
1512 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc) &&
1513 !test_bit((data & RTIT_CTL_MTC_RANGE) >>
1514 RTIT_CTL_MTC_RANGE_OFFSET, &value))
1516 value = intel_pt_validate_cap(vmx->pt_desc.caps,
1517 PT_CAP_cycle_thresholds);
1518 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1519 !test_bit((data & RTIT_CTL_CYC_THRESH) >>
1520 RTIT_CTL_CYC_THRESH_OFFSET, &value))
1522 value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_periods);
1523 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1524 !test_bit((data & RTIT_CTL_PSB_FREQ) >>
1525 RTIT_CTL_PSB_FREQ_OFFSET, &value))
1529 * If ADDRx_CFG is reserved or the encodings is >2 will
1530 * cause a #GP fault.
1532 value = (data & RTIT_CTL_ADDR0) >> RTIT_CTL_ADDR0_OFFSET;
1533 if ((value && (vmx->pt_desc.addr_range < 1)) || (value > 2))
1535 value = (data & RTIT_CTL_ADDR1) >> RTIT_CTL_ADDR1_OFFSET;
1536 if ((value && (vmx->pt_desc.addr_range < 2)) || (value > 2))
1538 value = (data & RTIT_CTL_ADDR2) >> RTIT_CTL_ADDR2_OFFSET;
1539 if ((value && (vmx->pt_desc.addr_range < 3)) || (value > 2))
1541 value = (data & RTIT_CTL_ADDR3) >> RTIT_CTL_ADDR3_OFFSET;
1542 if ((value && (vmx->pt_desc.addr_range < 4)) || (value > 2))
1548 static bool vmx_can_emulate_instruction(struct kvm_vcpu *vcpu, void *insn, int insn_len)
1553 static int skip_emulated_instruction(struct kvm_vcpu *vcpu)
1555 unsigned long rip, orig_rip;
1558 * Using VMCS.VM_EXIT_INSTRUCTION_LEN on EPT misconfig depends on
1559 * undefined behavior: Intel's SDM doesn't mandate the VMCS field be
1560 * set when EPT misconfig occurs. In practice, real hardware updates
1561 * VM_EXIT_INSTRUCTION_LEN on EPT misconfig, but other hypervisors
1562 * (namely Hyper-V) don't set it due to it being undefined behavior,
1563 * i.e. we end up advancing IP with some random value.
1565 if (!static_cpu_has(X86_FEATURE_HYPERVISOR) ||
1566 to_vmx(vcpu)->exit_reason != EXIT_REASON_EPT_MISCONFIG) {
1567 orig_rip = kvm_rip_read(vcpu);
1568 rip = orig_rip + vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1569 #ifdef CONFIG_X86_64
1571 * We need to mask out the high 32 bits of RIP if not in 64-bit
1572 * mode, but just finding out that we are in 64-bit mode is
1573 * quite expensive. Only do it if there was a carry.
1575 if (unlikely(((rip ^ orig_rip) >> 31) == 3) && !is_64_bit_mode(vcpu))
1578 kvm_rip_write(vcpu, rip);
1580 if (!kvm_emulate_instruction(vcpu, EMULTYPE_SKIP))
1584 /* skipping an emulated instruction also counts */
1585 vmx_set_interrupt_shadow(vcpu, 0);
1591 * Recognizes a pending MTF VM-exit and records the nested state for later
1594 static void vmx_update_emulated_instruction(struct kvm_vcpu *vcpu)
1596 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1597 struct vcpu_vmx *vmx = to_vmx(vcpu);
1599 if (!is_guest_mode(vcpu))
1603 * Per the SDM, MTF takes priority over debug-trap exceptions besides
1604 * T-bit traps. As instruction emulation is completed (i.e. at the
1605 * instruction boundary), any #DB exception pending delivery must be a
1606 * debug-trap. Record the pending MTF state to be delivered in
1607 * vmx_check_nested_events().
1609 if (nested_cpu_has_mtf(vmcs12) &&
1610 (!vcpu->arch.exception.pending ||
1611 vcpu->arch.exception.nr == DB_VECTOR))
1612 vmx->nested.mtf_pending = true;
1614 vmx->nested.mtf_pending = false;
1617 static int vmx_skip_emulated_instruction(struct kvm_vcpu *vcpu)
1619 vmx_update_emulated_instruction(vcpu);
1620 return skip_emulated_instruction(vcpu);
1623 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1626 * Ensure that we clear the HLT state in the VMCS. We don't need to
1627 * explicitly skip the instruction because if the HLT state is set,
1628 * then the instruction is already executing and RIP has already been
1631 if (kvm_hlt_in_guest(vcpu->kvm) &&
1632 vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1633 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1636 static void vmx_queue_exception(struct kvm_vcpu *vcpu)
1638 struct vcpu_vmx *vmx = to_vmx(vcpu);
1639 unsigned nr = vcpu->arch.exception.nr;
1640 bool has_error_code = vcpu->arch.exception.has_error_code;
1641 u32 error_code = vcpu->arch.exception.error_code;
1642 u32 intr_info = nr | INTR_INFO_VALID_MASK;
1644 kvm_deliver_exception_payload(vcpu);
1646 if (has_error_code) {
1647 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1648 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1651 if (vmx->rmode.vm86_active) {
1653 if (kvm_exception_is_soft(nr))
1654 inc_eip = vcpu->arch.event_exit_inst_len;
1655 kvm_inject_realmode_interrupt(vcpu, nr, inc_eip);
1659 WARN_ON_ONCE(vmx->emulation_required);
1661 if (kvm_exception_is_soft(nr)) {
1662 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1663 vmx->vcpu.arch.event_exit_inst_len);
1664 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1666 intr_info |= INTR_TYPE_HARD_EXCEPTION;
1668 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1670 vmx_clear_hlt(vcpu);
1673 static void vmx_setup_uret_msr(struct vcpu_vmx *vmx, unsigned int msr)
1675 struct vmx_uret_msr tmp;
1678 from = __vmx_find_uret_msr(vmx, msr);
1681 to = vmx->nr_active_uret_msrs++;
1683 tmp = vmx->guest_uret_msrs[to];
1684 vmx->guest_uret_msrs[to] = vmx->guest_uret_msrs[from];
1685 vmx->guest_uret_msrs[from] = tmp;
1689 * Set up the vmcs to automatically save and restore system
1690 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
1691 * mode, as fiddling with msrs is very expensive.
1693 static void setup_msrs(struct vcpu_vmx *vmx)
1695 vmx->guest_uret_msrs_loaded = false;
1696 vmx->nr_active_uret_msrs = 0;
1697 #ifdef CONFIG_X86_64
1699 * The SYSCALL MSRs are only needed on long mode guests, and only
1700 * when EFER.SCE is set.
1702 if (is_long_mode(&vmx->vcpu) && (vmx->vcpu.arch.efer & EFER_SCE)) {
1703 vmx_setup_uret_msr(vmx, MSR_STAR);
1704 vmx_setup_uret_msr(vmx, MSR_LSTAR);
1705 vmx_setup_uret_msr(vmx, MSR_SYSCALL_MASK);
1708 if (update_transition_efer(vmx))
1709 vmx_setup_uret_msr(vmx, MSR_EFER);
1711 if (guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDTSCP))
1712 vmx_setup_uret_msr(vmx, MSR_TSC_AUX);
1714 vmx_setup_uret_msr(vmx, MSR_IA32_TSX_CTRL);
1716 if (cpu_has_vmx_msr_bitmap())
1717 vmx_update_msr_bitmap(&vmx->vcpu);
1720 static u64 vmx_write_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1722 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1723 u64 g_tsc_offset = 0;
1726 * We're here if L1 chose not to trap WRMSR to TSC. According
1727 * to the spec, this should set L1's TSC; The offset that L1
1728 * set for L2 remains unchanged, and still needs to be added
1729 * to the newly set TSC to get L2's TSC.
1731 if (is_guest_mode(vcpu) &&
1732 (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING))
1733 g_tsc_offset = vmcs12->tsc_offset;
1735 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1736 vcpu->arch.tsc_offset - g_tsc_offset,
1738 vmcs_write64(TSC_OFFSET, offset + g_tsc_offset);
1739 return offset + g_tsc_offset;
1743 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
1744 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
1745 * all guests if the "nested" module option is off, and can also be disabled
1746 * for a single guest by disabling its VMX cpuid bit.
1748 bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
1750 return nested && guest_cpuid_has(vcpu, X86_FEATURE_VMX);
1753 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
1756 uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
1758 return !(val & ~valid_bits);
1761 static int vmx_get_msr_feature(struct kvm_msr_entry *msr)
1763 switch (msr->index) {
1764 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1767 return vmx_get_vmx_msr(&vmcs_config.nested, msr->index, &msr->data);
1768 case MSR_IA32_PERF_CAPABILITIES:
1769 msr->data = vmx_get_perf_capabilities();
1772 return KVM_MSR_RET_INVALID;
1777 * Reads an msr value (of 'msr_index') into 'pdata'.
1778 * Returns 0 on success, non-0 otherwise.
1779 * Assumes vcpu_load() was already called.
1781 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1783 struct vcpu_vmx *vmx = to_vmx(vcpu);
1784 struct vmx_uret_msr *msr;
1787 switch (msr_info->index) {
1788 #ifdef CONFIG_X86_64
1790 msr_info->data = vmcs_readl(GUEST_FS_BASE);
1793 msr_info->data = vmcs_readl(GUEST_GS_BASE);
1795 case MSR_KERNEL_GS_BASE:
1796 msr_info->data = vmx_read_guest_kernel_gs_base(vmx);
1800 return kvm_get_msr_common(vcpu, msr_info);
1801 case MSR_IA32_TSX_CTRL:
1802 if (!msr_info->host_initiated &&
1803 !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
1806 case MSR_IA32_UMWAIT_CONTROL:
1807 if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
1810 msr_info->data = vmx->msr_ia32_umwait_control;
1812 case MSR_IA32_SPEC_CTRL:
1813 if (!msr_info->host_initiated &&
1814 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1817 msr_info->data = to_vmx(vcpu)->spec_ctrl;
1819 case MSR_IA32_SYSENTER_CS:
1820 msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
1822 case MSR_IA32_SYSENTER_EIP:
1823 msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
1825 case MSR_IA32_SYSENTER_ESP:
1826 msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
1828 case MSR_IA32_BNDCFGS:
1829 if (!kvm_mpx_supported() ||
1830 (!msr_info->host_initiated &&
1831 !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1833 msr_info->data = vmcs_read64(GUEST_BNDCFGS);
1835 case MSR_IA32_MCG_EXT_CTL:
1836 if (!msr_info->host_initiated &&
1837 !(vmx->msr_ia32_feature_control &
1838 FEAT_CTL_LMCE_ENABLED))
1840 msr_info->data = vcpu->arch.mcg_ext_ctl;
1842 case MSR_IA32_FEAT_CTL:
1843 msr_info->data = vmx->msr_ia32_feature_control;
1845 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1846 if (!nested_vmx_allowed(vcpu))
1848 if (vmx_get_vmx_msr(&vmx->nested.msrs, msr_info->index,
1852 * Enlightened VMCS v1 doesn't have certain fields, but buggy
1853 * Hyper-V versions are still trying to use corresponding
1854 * features when they are exposed. Filter out the essential
1857 if (!msr_info->host_initiated &&
1858 vmx->nested.enlightened_vmcs_enabled)
1859 nested_evmcs_filter_control_msr(msr_info->index,
1862 case MSR_IA32_RTIT_CTL:
1863 if (!vmx_pt_mode_is_host_guest())
1865 msr_info->data = vmx->pt_desc.guest.ctl;
1867 case MSR_IA32_RTIT_STATUS:
1868 if (!vmx_pt_mode_is_host_guest())
1870 msr_info->data = vmx->pt_desc.guest.status;
1872 case MSR_IA32_RTIT_CR3_MATCH:
1873 if (!vmx_pt_mode_is_host_guest() ||
1874 !intel_pt_validate_cap(vmx->pt_desc.caps,
1875 PT_CAP_cr3_filtering))
1877 msr_info->data = vmx->pt_desc.guest.cr3_match;
1879 case MSR_IA32_RTIT_OUTPUT_BASE:
1880 if (!vmx_pt_mode_is_host_guest() ||
1881 (!intel_pt_validate_cap(vmx->pt_desc.caps,
1882 PT_CAP_topa_output) &&
1883 !intel_pt_validate_cap(vmx->pt_desc.caps,
1884 PT_CAP_single_range_output)))
1886 msr_info->data = vmx->pt_desc.guest.output_base;
1888 case MSR_IA32_RTIT_OUTPUT_MASK:
1889 if (!vmx_pt_mode_is_host_guest() ||
1890 (!intel_pt_validate_cap(vmx->pt_desc.caps,
1891 PT_CAP_topa_output) &&
1892 !intel_pt_validate_cap(vmx->pt_desc.caps,
1893 PT_CAP_single_range_output)))
1895 msr_info->data = vmx->pt_desc.guest.output_mask;
1897 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
1898 index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
1899 if (!vmx_pt_mode_is_host_guest() ||
1900 (index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
1901 PT_CAP_num_address_ranges)))
1904 msr_info->data = vmx->pt_desc.guest.addr_b[index / 2];
1906 msr_info->data = vmx->pt_desc.guest.addr_a[index / 2];
1909 if (!msr_info->host_initiated &&
1910 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
1915 msr = vmx_find_uret_msr(vmx, msr_info->index);
1917 msr_info->data = msr->data;
1920 return kvm_get_msr_common(vcpu, msr_info);
1926 static u64 nested_vmx_truncate_sysenter_addr(struct kvm_vcpu *vcpu,
1929 #ifdef CONFIG_X86_64
1930 if (!guest_cpuid_has(vcpu, X86_FEATURE_LM))
1933 return (unsigned long)data;
1937 * Writes msr value into the appropriate "register".
1938 * Returns 0 on success, non-0 otherwise.
1939 * Assumes vcpu_load() was already called.
1941 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1943 struct vcpu_vmx *vmx = to_vmx(vcpu);
1944 struct vmx_uret_msr *msr;
1946 u32 msr_index = msr_info->index;
1947 u64 data = msr_info->data;
1950 switch (msr_index) {
1952 ret = kvm_set_msr_common(vcpu, msr_info);
1954 #ifdef CONFIG_X86_64
1956 vmx_segment_cache_clear(vmx);
1957 vmcs_writel(GUEST_FS_BASE, data);
1960 vmx_segment_cache_clear(vmx);
1961 vmcs_writel(GUEST_GS_BASE, data);
1963 case MSR_KERNEL_GS_BASE:
1964 vmx_write_guest_kernel_gs_base(vmx, data);
1967 case MSR_IA32_SYSENTER_CS:
1968 if (is_guest_mode(vcpu))
1969 get_vmcs12(vcpu)->guest_sysenter_cs = data;
1970 vmcs_write32(GUEST_SYSENTER_CS, data);
1972 case MSR_IA32_SYSENTER_EIP:
1973 if (is_guest_mode(vcpu)) {
1974 data = nested_vmx_truncate_sysenter_addr(vcpu, data);
1975 get_vmcs12(vcpu)->guest_sysenter_eip = data;
1977 vmcs_writel(GUEST_SYSENTER_EIP, data);
1979 case MSR_IA32_SYSENTER_ESP:
1980 if (is_guest_mode(vcpu)) {
1981 data = nested_vmx_truncate_sysenter_addr(vcpu, data);
1982 get_vmcs12(vcpu)->guest_sysenter_esp = data;
1984 vmcs_writel(GUEST_SYSENTER_ESP, data);
1986 case MSR_IA32_DEBUGCTLMSR:
1987 if (is_guest_mode(vcpu) && get_vmcs12(vcpu)->vm_exit_controls &
1988 VM_EXIT_SAVE_DEBUG_CONTROLS)
1989 get_vmcs12(vcpu)->guest_ia32_debugctl = data;
1991 ret = kvm_set_msr_common(vcpu, msr_info);
1994 case MSR_IA32_BNDCFGS:
1995 if (!kvm_mpx_supported() ||
1996 (!msr_info->host_initiated &&
1997 !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1999 if (is_noncanonical_address(data & PAGE_MASK, vcpu) ||
2000 (data & MSR_IA32_BNDCFGS_RSVD))
2002 vmcs_write64(GUEST_BNDCFGS, data);
2004 case MSR_IA32_UMWAIT_CONTROL:
2005 if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
2008 /* The reserved bit 1 and non-32 bit [63:32] should be zero */
2009 if (data & (BIT_ULL(1) | GENMASK_ULL(63, 32)))
2012 vmx->msr_ia32_umwait_control = data;
2014 case MSR_IA32_SPEC_CTRL:
2015 if (!msr_info->host_initiated &&
2016 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
2019 if (kvm_spec_ctrl_test_value(data))
2022 vmx->spec_ctrl = data;
2028 * When it's written (to non-zero) for the first time, pass
2032 * The handling of the MSR bitmap for L2 guests is done in
2033 * nested_vmx_prepare_msr_bitmap. We should not touch the
2034 * vmcs02.msr_bitmap here since it gets completely overwritten
2035 * in the merging. We update the vmcs01 here for L1 as well
2036 * since it will end up touching the MSR anyway now.
2038 vmx_disable_intercept_for_msr(vcpu,
2042 case MSR_IA32_TSX_CTRL:
2043 if (!msr_info->host_initiated &&
2044 !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
2046 if (data & ~(TSX_CTRL_RTM_DISABLE | TSX_CTRL_CPUID_CLEAR))
2049 case MSR_IA32_PRED_CMD:
2050 if (!msr_info->host_initiated &&
2051 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
2054 if (data & ~PRED_CMD_IBPB)
2056 if (!boot_cpu_has(X86_FEATURE_SPEC_CTRL))
2061 wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
2065 * When it's written (to non-zero) for the first time, pass
2069 * The handling of the MSR bitmap for L2 guests is done in
2070 * nested_vmx_prepare_msr_bitmap. We should not touch the
2071 * vmcs02.msr_bitmap here since it gets completely overwritten
2074 vmx_disable_intercept_for_msr(vcpu, MSR_IA32_PRED_CMD, MSR_TYPE_W);
2076 case MSR_IA32_CR_PAT:
2077 if (!kvm_pat_valid(data))
2080 if (is_guest_mode(vcpu) &&
2081 get_vmcs12(vcpu)->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
2082 get_vmcs12(vcpu)->guest_ia32_pat = data;
2084 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2085 vmcs_write64(GUEST_IA32_PAT, data);
2086 vcpu->arch.pat = data;
2089 ret = kvm_set_msr_common(vcpu, msr_info);
2091 case MSR_IA32_TSC_ADJUST:
2092 ret = kvm_set_msr_common(vcpu, msr_info);
2094 case MSR_IA32_MCG_EXT_CTL:
2095 if ((!msr_info->host_initiated &&
2096 !(to_vmx(vcpu)->msr_ia32_feature_control &
2097 FEAT_CTL_LMCE_ENABLED)) ||
2098 (data & ~MCG_EXT_CTL_LMCE_EN))
2100 vcpu->arch.mcg_ext_ctl = data;
2102 case MSR_IA32_FEAT_CTL:
2103 if (!vmx_feature_control_msr_valid(vcpu, data) ||
2104 (to_vmx(vcpu)->msr_ia32_feature_control &
2105 FEAT_CTL_LOCKED && !msr_info->host_initiated))
2107 vmx->msr_ia32_feature_control = data;
2108 if (msr_info->host_initiated && data == 0)
2109 vmx_leave_nested(vcpu);
2111 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2112 if (!msr_info->host_initiated)
2113 return 1; /* they are read-only */
2114 if (!nested_vmx_allowed(vcpu))
2116 return vmx_set_vmx_msr(vcpu, msr_index, data);
2117 case MSR_IA32_RTIT_CTL:
2118 if (!vmx_pt_mode_is_host_guest() ||
2119 vmx_rtit_ctl_check(vcpu, data) ||
2122 vmcs_write64(GUEST_IA32_RTIT_CTL, data);
2123 vmx->pt_desc.guest.ctl = data;
2124 pt_update_intercept_for_msr(vcpu);
2126 case MSR_IA32_RTIT_STATUS:
2127 if (!pt_can_write_msr(vmx))
2129 if (data & MSR_IA32_RTIT_STATUS_MASK)
2131 vmx->pt_desc.guest.status = data;
2133 case MSR_IA32_RTIT_CR3_MATCH:
2134 if (!pt_can_write_msr(vmx))
2136 if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2137 PT_CAP_cr3_filtering))
2139 vmx->pt_desc.guest.cr3_match = data;
2141 case MSR_IA32_RTIT_OUTPUT_BASE:
2142 if (!pt_can_write_msr(vmx))
2144 if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2145 PT_CAP_topa_output) &&
2146 !intel_pt_validate_cap(vmx->pt_desc.caps,
2147 PT_CAP_single_range_output))
2149 if (!pt_output_base_valid(vcpu, data))
2151 vmx->pt_desc.guest.output_base = data;
2153 case MSR_IA32_RTIT_OUTPUT_MASK:
2154 if (!pt_can_write_msr(vmx))
2156 if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2157 PT_CAP_topa_output) &&
2158 !intel_pt_validate_cap(vmx->pt_desc.caps,
2159 PT_CAP_single_range_output))
2161 vmx->pt_desc.guest.output_mask = data;
2163 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
2164 if (!pt_can_write_msr(vmx))
2166 index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
2167 if (index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
2168 PT_CAP_num_address_ranges))
2170 if (is_noncanonical_address(data, vcpu))
2173 vmx->pt_desc.guest.addr_b[index / 2] = data;
2175 vmx->pt_desc.guest.addr_a[index / 2] = data;
2178 if (!msr_info->host_initiated &&
2179 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
2181 /* Check reserved bit, higher 32 bits should be zero */
2182 if ((data >> 32) != 0)
2188 msr = vmx_find_uret_msr(vmx, msr_index);
2190 ret = vmx_set_guest_uret_msr(vmx, msr, data);
2192 ret = kvm_set_msr_common(vcpu, msr_info);
2198 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2200 unsigned long guest_owned_bits;
2202 kvm_register_mark_available(vcpu, reg);
2206 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2209 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2211 case VCPU_EXREG_PDPTR:
2213 ept_save_pdptrs(vcpu);
2215 case VCPU_EXREG_CR0:
2216 guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
2218 vcpu->arch.cr0 &= ~guest_owned_bits;
2219 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & guest_owned_bits;
2221 case VCPU_EXREG_CR3:
2222 if (is_unrestricted_guest(vcpu) ||
2223 (enable_ept && is_paging(vcpu)))
2224 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2226 case VCPU_EXREG_CR4:
2227 guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
2229 vcpu->arch.cr4 &= ~guest_owned_bits;
2230 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & guest_owned_bits;
2238 static __init int cpu_has_kvm_support(void)
2240 return cpu_has_vmx();
2243 static __init int vmx_disabled_by_bios(void)
2245 return !boot_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
2246 !boot_cpu_has(X86_FEATURE_VMX);
2249 static int kvm_cpu_vmxon(u64 vmxon_pointer)
2253 cr4_set_bits(X86_CR4_VMXE);
2254 intel_pt_handle_vmx(1);
2256 asm_volatile_goto("1: vmxon %[vmxon_pointer]\n\t"
2257 _ASM_EXTABLE(1b, %l[fault])
2258 : : [vmxon_pointer] "m"(vmxon_pointer)
2263 WARN_ONCE(1, "VMXON faulted, MSR_IA32_FEAT_CTL (0x3a) = 0x%llx\n",
2264 rdmsrl_safe(MSR_IA32_FEAT_CTL, &msr) ? 0xdeadbeef : msr);
2265 intel_pt_handle_vmx(0);
2266 cr4_clear_bits(X86_CR4_VMXE);
2271 static int hardware_enable(void)
2273 int cpu = raw_smp_processor_id();
2274 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2277 if (cr4_read_shadow() & X86_CR4_VMXE)
2281 * This can happen if we hot-added a CPU but failed to allocate
2282 * VP assist page for it.
2284 if (static_branch_unlikely(&enable_evmcs) &&
2285 !hv_get_vp_assist_page(cpu))
2288 r = kvm_cpu_vmxon(phys_addr);
2298 static void vmclear_local_loaded_vmcss(void)
2300 int cpu = raw_smp_processor_id();
2301 struct loaded_vmcs *v, *n;
2303 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2304 loaded_vmcss_on_cpu_link)
2305 __loaded_vmcs_clear(v);
2309 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2312 static void kvm_cpu_vmxoff(void)
2314 asm volatile (__ex("vmxoff"));
2316 intel_pt_handle_vmx(0);
2317 cr4_clear_bits(X86_CR4_VMXE);
2320 static void hardware_disable(void)
2322 vmclear_local_loaded_vmcss();
2327 * There is no X86_FEATURE for SGX yet, but anyway we need to query CPUID
2328 * directly instead of going through cpu_has(), to ensure KVM is trapping
2329 * ENCLS whenever it's supported in hardware. It does not matter whether
2330 * the host OS supports or has enabled SGX.
2332 static bool cpu_has_sgx(void)
2334 return cpuid_eax(0) >= 0x12 && (cpuid_eax(0x12) & BIT(0));
2337 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2338 u32 msr, u32 *result)
2340 u32 vmx_msr_low, vmx_msr_high;
2341 u32 ctl = ctl_min | ctl_opt;
2343 rdmsr(msr, vmx_msr_low, vmx_msr_high);
2345 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2346 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
2348 /* Ensure minimum (required) set of control bits are supported. */
2356 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf,
2357 struct vmx_capability *vmx_cap)
2359 u32 vmx_msr_low, vmx_msr_high;
2360 u32 min, opt, min2, opt2;
2361 u32 _pin_based_exec_control = 0;
2362 u32 _cpu_based_exec_control = 0;
2363 u32 _cpu_based_2nd_exec_control = 0;
2364 u32 _vmexit_control = 0;
2365 u32 _vmentry_control = 0;
2367 memset(vmcs_conf, 0, sizeof(*vmcs_conf));
2368 min = CPU_BASED_HLT_EXITING |
2369 #ifdef CONFIG_X86_64
2370 CPU_BASED_CR8_LOAD_EXITING |
2371 CPU_BASED_CR8_STORE_EXITING |
2373 CPU_BASED_CR3_LOAD_EXITING |
2374 CPU_BASED_CR3_STORE_EXITING |
2375 CPU_BASED_UNCOND_IO_EXITING |
2376 CPU_BASED_MOV_DR_EXITING |
2377 CPU_BASED_USE_TSC_OFFSETTING |
2378 CPU_BASED_MWAIT_EXITING |
2379 CPU_BASED_MONITOR_EXITING |
2380 CPU_BASED_INVLPG_EXITING |
2381 CPU_BASED_RDPMC_EXITING;
2383 opt = CPU_BASED_TPR_SHADOW |
2384 CPU_BASED_USE_MSR_BITMAPS |
2385 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2386 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2387 &_cpu_based_exec_control) < 0)
2389 #ifdef CONFIG_X86_64
2390 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2391 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2392 ~CPU_BASED_CR8_STORE_EXITING;
2394 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2396 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2397 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2398 SECONDARY_EXEC_WBINVD_EXITING |
2399 SECONDARY_EXEC_ENABLE_VPID |
2400 SECONDARY_EXEC_ENABLE_EPT |
2401 SECONDARY_EXEC_UNRESTRICTED_GUEST |
2402 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2403 SECONDARY_EXEC_DESC |
2404 SECONDARY_EXEC_ENABLE_RDTSCP |
2405 SECONDARY_EXEC_ENABLE_INVPCID |
2406 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2407 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2408 SECONDARY_EXEC_SHADOW_VMCS |
2409 SECONDARY_EXEC_XSAVES |
2410 SECONDARY_EXEC_RDSEED_EXITING |
2411 SECONDARY_EXEC_RDRAND_EXITING |
2412 SECONDARY_EXEC_ENABLE_PML |
2413 SECONDARY_EXEC_TSC_SCALING |
2414 SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE |
2415 SECONDARY_EXEC_PT_USE_GPA |
2416 SECONDARY_EXEC_PT_CONCEAL_VMX |
2417 SECONDARY_EXEC_ENABLE_VMFUNC;
2419 opt2 |= SECONDARY_EXEC_ENCLS_EXITING;
2420 if (adjust_vmx_controls(min2, opt2,
2421 MSR_IA32_VMX_PROCBASED_CTLS2,
2422 &_cpu_based_2nd_exec_control) < 0)
2425 #ifndef CONFIG_X86_64
2426 if (!(_cpu_based_2nd_exec_control &
2427 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2428 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2431 if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2432 _cpu_based_2nd_exec_control &= ~(
2433 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2434 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2435 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
2437 rdmsr_safe(MSR_IA32_VMX_EPT_VPID_CAP,
2438 &vmx_cap->ept, &vmx_cap->vpid);
2440 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2441 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2443 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2444 CPU_BASED_CR3_STORE_EXITING |
2445 CPU_BASED_INVLPG_EXITING);
2446 } else if (vmx_cap->ept) {
2448 pr_warn_once("EPT CAP should not exist if not support "
2449 "1-setting enable EPT VM-execution control\n");
2451 if (!(_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_VPID) &&
2454 pr_warn_once("VPID CAP should not exist if not support "
2455 "1-setting enable VPID VM-execution control\n");
2458 min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
2459 #ifdef CONFIG_X86_64
2460 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2462 opt = VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL |
2463 VM_EXIT_LOAD_IA32_PAT |
2464 VM_EXIT_LOAD_IA32_EFER |
2465 VM_EXIT_CLEAR_BNDCFGS |
2466 VM_EXIT_PT_CONCEAL_PIP |
2467 VM_EXIT_CLEAR_IA32_RTIT_CTL;
2468 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2469 &_vmexit_control) < 0)
2472 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2473 opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR |
2474 PIN_BASED_VMX_PREEMPTION_TIMER;
2475 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2476 &_pin_based_exec_control) < 0)
2479 if (cpu_has_broken_vmx_preemption_timer())
2480 _pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
2481 if (!(_cpu_based_2nd_exec_control &
2482 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
2483 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
2485 min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
2486 opt = VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL |
2487 VM_ENTRY_LOAD_IA32_PAT |
2488 VM_ENTRY_LOAD_IA32_EFER |
2489 VM_ENTRY_LOAD_BNDCFGS |
2490 VM_ENTRY_PT_CONCEAL_PIP |
2491 VM_ENTRY_LOAD_IA32_RTIT_CTL;
2492 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2493 &_vmentry_control) < 0)
2497 * Some cpus support VM_{ENTRY,EXIT}_IA32_PERF_GLOBAL_CTRL but they
2498 * can't be used due to an errata where VM Exit may incorrectly clear
2499 * IA32_PERF_GLOBAL_CTRL[34:32]. Workaround the errata by using the
2500 * MSR load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2502 if (boot_cpu_data.x86 == 0x6) {
2503 switch (boot_cpu_data.x86_model) {
2504 case 26: /* AAK155 */
2505 case 30: /* AAP115 */
2506 case 37: /* AAT100 */
2507 case 44: /* BC86,AAY89,BD102 */
2509 _vmentry_control &= ~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
2510 _vmexit_control &= ~VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
2511 pr_warn_once("kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
2512 "does not work properly. Using workaround\n");
2520 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2522 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2523 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2526 #ifdef CONFIG_X86_64
2527 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2528 if (vmx_msr_high & (1u<<16))
2532 /* Require Write-Back (WB) memory type for VMCS accesses. */
2533 if (((vmx_msr_high >> 18) & 15) != 6)
2536 vmcs_conf->size = vmx_msr_high & 0x1fff;
2537 vmcs_conf->order = get_order(vmcs_conf->size);
2538 vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
2540 vmcs_conf->revision_id = vmx_msr_low;
2542 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2543 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2544 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2545 vmcs_conf->vmexit_ctrl = _vmexit_control;
2546 vmcs_conf->vmentry_ctrl = _vmentry_control;
2548 if (static_branch_unlikely(&enable_evmcs))
2549 evmcs_sanitize_exec_ctrls(vmcs_conf);
2554 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags)
2556 int node = cpu_to_node(cpu);
2560 pages = __alloc_pages_node(node, flags, vmcs_config.order);
2563 vmcs = page_address(pages);
2564 memset(vmcs, 0, vmcs_config.size);
2566 /* KVM supports Enlightened VMCS v1 only */
2567 if (static_branch_unlikely(&enable_evmcs))
2568 vmcs->hdr.revision_id = KVM_EVMCS_VERSION;
2570 vmcs->hdr.revision_id = vmcs_config.revision_id;
2573 vmcs->hdr.shadow_vmcs = 1;
2577 void free_vmcs(struct vmcs *vmcs)
2579 free_pages((unsigned long)vmcs, vmcs_config.order);
2583 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2585 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2587 if (!loaded_vmcs->vmcs)
2589 loaded_vmcs_clear(loaded_vmcs);
2590 free_vmcs(loaded_vmcs->vmcs);
2591 loaded_vmcs->vmcs = NULL;
2592 if (loaded_vmcs->msr_bitmap)
2593 free_page((unsigned long)loaded_vmcs->msr_bitmap);
2594 WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
2597 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2599 loaded_vmcs->vmcs = alloc_vmcs(false);
2600 if (!loaded_vmcs->vmcs)
2603 vmcs_clear(loaded_vmcs->vmcs);
2605 loaded_vmcs->shadow_vmcs = NULL;
2606 loaded_vmcs->hv_timer_soft_disabled = false;
2607 loaded_vmcs->cpu = -1;
2608 loaded_vmcs->launched = 0;
2610 if (cpu_has_vmx_msr_bitmap()) {
2611 loaded_vmcs->msr_bitmap = (unsigned long *)
2612 __get_free_page(GFP_KERNEL_ACCOUNT);
2613 if (!loaded_vmcs->msr_bitmap)
2615 memset(loaded_vmcs->msr_bitmap, 0xff, PAGE_SIZE);
2617 if (IS_ENABLED(CONFIG_HYPERV) &&
2618 static_branch_unlikely(&enable_evmcs) &&
2619 (ms_hyperv.nested_features & HV_X64_NESTED_MSR_BITMAP)) {
2620 struct hv_enlightened_vmcs *evmcs =
2621 (struct hv_enlightened_vmcs *)loaded_vmcs->vmcs;
2623 evmcs->hv_enlightenments_control.msr_bitmap = 1;
2627 memset(&loaded_vmcs->host_state, 0, sizeof(struct vmcs_host_state));
2628 memset(&loaded_vmcs->controls_shadow, 0,
2629 sizeof(struct vmcs_controls_shadow));
2634 free_loaded_vmcs(loaded_vmcs);
2638 static void free_kvm_area(void)
2642 for_each_possible_cpu(cpu) {
2643 free_vmcs(per_cpu(vmxarea, cpu));
2644 per_cpu(vmxarea, cpu) = NULL;
2648 static __init int alloc_kvm_area(void)
2652 for_each_possible_cpu(cpu) {
2655 vmcs = alloc_vmcs_cpu(false, cpu, GFP_KERNEL);
2662 * When eVMCS is enabled, alloc_vmcs_cpu() sets
2663 * vmcs->revision_id to KVM_EVMCS_VERSION instead of
2664 * revision_id reported by MSR_IA32_VMX_BASIC.
2666 * However, even though not explicitly documented by
2667 * TLFS, VMXArea passed as VMXON argument should
2668 * still be marked with revision_id reported by
2671 if (static_branch_unlikely(&enable_evmcs))
2672 vmcs->hdr.revision_id = vmcs_config.revision_id;
2674 per_cpu(vmxarea, cpu) = vmcs;
2679 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
2680 struct kvm_segment *save)
2682 if (!emulate_invalid_guest_state) {
2684 * CS and SS RPL should be equal during guest entry according
2685 * to VMX spec, but in reality it is not always so. Since vcpu
2686 * is in the middle of the transition from real mode to
2687 * protected mode it is safe to assume that RPL 0 is a good
2690 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
2691 save->selector &= ~SEGMENT_RPL_MASK;
2692 save->dpl = save->selector & SEGMENT_RPL_MASK;
2695 vmx_set_segment(vcpu, save, seg);
2698 static void enter_pmode(struct kvm_vcpu *vcpu)
2700 unsigned long flags;
2701 struct vcpu_vmx *vmx = to_vmx(vcpu);
2704 * Update real mode segment cache. It may be not up-to-date if sement
2705 * register was written while vcpu was in a guest mode.
2707 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2708 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2709 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2710 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2711 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2712 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2714 vmx->rmode.vm86_active = 0;
2716 vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2718 flags = vmcs_readl(GUEST_RFLAGS);
2719 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2720 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2721 vmcs_writel(GUEST_RFLAGS, flags);
2723 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
2724 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
2726 update_exception_bitmap(vcpu);
2728 fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2729 fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2730 fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2731 fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2732 fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2733 fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2736 static void fix_rmode_seg(int seg, struct kvm_segment *save)
2738 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2739 struct kvm_segment var = *save;
2742 if (seg == VCPU_SREG_CS)
2745 if (!emulate_invalid_guest_state) {
2746 var.selector = var.base >> 4;
2747 var.base = var.base & 0xffff0;
2757 if (save->base & 0xf)
2758 printk_once(KERN_WARNING "kvm: segment base is not "
2759 "paragraph aligned when entering "
2760 "protected mode (seg=%d)", seg);
2763 vmcs_write16(sf->selector, var.selector);
2764 vmcs_writel(sf->base, var.base);
2765 vmcs_write32(sf->limit, var.limit);
2766 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
2769 static void enter_rmode(struct kvm_vcpu *vcpu)
2771 unsigned long flags;
2772 struct vcpu_vmx *vmx = to_vmx(vcpu);
2773 struct kvm_vmx *kvm_vmx = to_kvm_vmx(vcpu->kvm);
2775 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2776 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2777 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2778 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2779 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2780 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2781 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2783 vmx->rmode.vm86_active = 1;
2786 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
2787 * vcpu. Warn the user that an update is overdue.
2789 if (!kvm_vmx->tss_addr)
2790 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
2791 "called before entering vcpu\n");
2793 vmx_segment_cache_clear(vmx);
2795 vmcs_writel(GUEST_TR_BASE, kvm_vmx->tss_addr);
2796 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
2797 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2799 flags = vmcs_readl(GUEST_RFLAGS);
2800 vmx->rmode.save_rflags = flags;
2802 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2804 vmcs_writel(GUEST_RFLAGS, flags);
2805 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
2806 update_exception_bitmap(vcpu);
2808 fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2809 fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2810 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2811 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2812 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2813 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2815 kvm_mmu_reset_context(vcpu);
2818 void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
2820 struct vcpu_vmx *vmx = to_vmx(vcpu);
2821 struct vmx_uret_msr *msr = vmx_find_uret_msr(vmx, MSR_EFER);
2826 vcpu->arch.efer = efer;
2827 if (efer & EFER_LMA) {
2828 vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2831 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2833 msr->data = efer & ~EFER_LME;
2838 #ifdef CONFIG_X86_64
2840 static void enter_lmode(struct kvm_vcpu *vcpu)
2844 vmx_segment_cache_clear(to_vmx(vcpu));
2846 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
2847 if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
2848 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
2850 vmcs_write32(GUEST_TR_AR_BYTES,
2851 (guest_tr_ar & ~VMX_AR_TYPE_MASK)
2852 | VMX_AR_TYPE_BUSY_64_TSS);
2854 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
2857 static void exit_lmode(struct kvm_vcpu *vcpu)
2859 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2860 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
2865 static void vmx_flush_tlb_all(struct kvm_vcpu *vcpu)
2867 struct vcpu_vmx *vmx = to_vmx(vcpu);
2870 * INVEPT must be issued when EPT is enabled, irrespective of VPID, as
2871 * the CPU is not required to invalidate guest-physical mappings on
2872 * VM-Entry, even if VPID is disabled. Guest-physical mappings are
2873 * associated with the root EPT structure and not any particular VPID
2874 * (INVVPID also isn't required to invalidate guest-physical mappings).
2878 } else if (enable_vpid) {
2879 if (cpu_has_vmx_invvpid_global()) {
2880 vpid_sync_vcpu_global();
2882 vpid_sync_vcpu_single(vmx->vpid);
2883 vpid_sync_vcpu_single(vmx->nested.vpid02);
2888 static void vmx_flush_tlb_current(struct kvm_vcpu *vcpu)
2890 struct kvm_mmu *mmu = vcpu->arch.mmu;
2891 u64 root_hpa = mmu->root_hpa;
2893 /* No flush required if the current context is invalid. */
2894 if (!VALID_PAGE(root_hpa))
2898 ept_sync_context(construct_eptp(vcpu, root_hpa,
2899 mmu->shadow_root_level));
2900 else if (!is_guest_mode(vcpu))
2901 vpid_sync_context(to_vmx(vcpu)->vpid);
2903 vpid_sync_context(nested_get_vpid02(vcpu));
2906 static void vmx_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t addr)
2909 * vpid_sync_vcpu_addr() is a nop if vmx->vpid==0, see the comment in
2910 * vmx_flush_tlb_guest() for an explanation of why this is ok.
2912 vpid_sync_vcpu_addr(to_vmx(vcpu)->vpid, addr);
2915 static void vmx_flush_tlb_guest(struct kvm_vcpu *vcpu)
2918 * vpid_sync_context() is a nop if vmx->vpid==0, e.g. if enable_vpid==0
2919 * or a vpid couldn't be allocated for this vCPU. VM-Enter and VM-Exit
2920 * are required to flush GVA->{G,H}PA mappings from the TLB if vpid is
2921 * disabled (VM-Enter with vpid enabled and vpid==0 is disallowed),
2922 * i.e. no explicit INVVPID is necessary.
2924 vpid_sync_context(to_vmx(vcpu)->vpid);
2927 void vmx_ept_load_pdptrs(struct kvm_vcpu *vcpu)
2929 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2931 if (!kvm_register_is_dirty(vcpu, VCPU_EXREG_PDPTR))
2934 if (is_pae_paging(vcpu)) {
2935 vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
2936 vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
2937 vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
2938 vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
2942 void ept_save_pdptrs(struct kvm_vcpu *vcpu)
2944 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2946 if (WARN_ON_ONCE(!is_pae_paging(vcpu)))
2949 mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
2950 mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
2951 mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
2952 mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
2954 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
2957 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
2959 struct kvm_vcpu *vcpu)
2961 struct vcpu_vmx *vmx = to_vmx(vcpu);
2963 if (!kvm_register_is_available(vcpu, VCPU_EXREG_CR3))
2964 vmx_cache_reg(vcpu, VCPU_EXREG_CR3);
2965 if (!(cr0 & X86_CR0_PG)) {
2966 /* From paging/starting to nonpaging */
2967 exec_controls_setbit(vmx, CPU_BASED_CR3_LOAD_EXITING |
2968 CPU_BASED_CR3_STORE_EXITING);
2969 vcpu->arch.cr0 = cr0;
2970 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2971 } else if (!is_paging(vcpu)) {
2972 /* From nonpaging to paging */
2973 exec_controls_clearbit(vmx, CPU_BASED_CR3_LOAD_EXITING |
2974 CPU_BASED_CR3_STORE_EXITING);
2975 vcpu->arch.cr0 = cr0;
2976 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2979 if (!(cr0 & X86_CR0_WP))
2980 *hw_cr0 &= ~X86_CR0_WP;
2983 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
2985 struct vcpu_vmx *vmx = to_vmx(vcpu);
2986 unsigned long hw_cr0;
2988 hw_cr0 = (cr0 & ~KVM_VM_CR0_ALWAYS_OFF);
2989 if (is_unrestricted_guest(vcpu))
2990 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
2992 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
2994 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
2997 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
3001 #ifdef CONFIG_X86_64
3002 if (vcpu->arch.efer & EFER_LME) {
3003 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
3005 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
3010 if (enable_ept && !is_unrestricted_guest(vcpu))
3011 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
3013 vmcs_writel(CR0_READ_SHADOW, cr0);
3014 vmcs_writel(GUEST_CR0, hw_cr0);
3015 vcpu->arch.cr0 = cr0;
3016 kvm_register_mark_available(vcpu, VCPU_EXREG_CR0);
3018 /* depends on vcpu->arch.cr0 to be set to a new value */
3019 vmx->emulation_required = emulation_required(vcpu);
3022 static int vmx_get_max_tdp_level(void)
3024 if (cpu_has_vmx_ept_5levels())
3029 u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa,
3032 u64 eptp = VMX_EPTP_MT_WB;
3034 eptp |= (root_level == 5) ? VMX_EPTP_PWL_5 : VMX_EPTP_PWL_4;
3036 if (enable_ept_ad_bits &&
3037 (!is_guest_mode(vcpu) || nested_ept_ad_enabled(vcpu)))
3038 eptp |= VMX_EPTP_AD_ENABLE_BIT;
3039 eptp |= (root_hpa & PAGE_MASK);
3044 static void vmx_load_mmu_pgd(struct kvm_vcpu *vcpu, unsigned long pgd,
3047 struct kvm *kvm = vcpu->kvm;
3048 bool update_guest_cr3 = true;
3049 unsigned long guest_cr3;
3053 eptp = construct_eptp(vcpu, pgd, pgd_level);
3054 vmcs_write64(EPT_POINTER, eptp);
3056 if (kvm_x86_ops.tlb_remote_flush) {
3057 spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
3058 to_vmx(vcpu)->ept_pointer = eptp;
3059 to_kvm_vmx(kvm)->ept_pointers_match
3060 = EPT_POINTERS_CHECK;
3061 spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
3064 if (!enable_unrestricted_guest && !is_paging(vcpu))
3065 guest_cr3 = to_kvm_vmx(kvm)->ept_identity_map_addr;
3066 else if (test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
3067 guest_cr3 = vcpu->arch.cr3;
3068 else /* vmcs01.GUEST_CR3 is already up-to-date. */
3069 update_guest_cr3 = false;
3070 vmx_ept_load_pdptrs(vcpu);
3075 if (update_guest_cr3)
3076 vmcs_writel(GUEST_CR3, guest_cr3);
3079 int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3081 struct vcpu_vmx *vmx = to_vmx(vcpu);
3083 * Pass through host's Machine Check Enable value to hw_cr4, which
3084 * is in force while we are in guest mode. Do not let guests control
3085 * this bit, even if host CR4.MCE == 0.
3087 unsigned long hw_cr4;
3089 hw_cr4 = (cr4_read_shadow() & X86_CR4_MCE) | (cr4 & ~X86_CR4_MCE);
3090 if (is_unrestricted_guest(vcpu))
3091 hw_cr4 |= KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST;
3092 else if (vmx->rmode.vm86_active)
3093 hw_cr4 |= KVM_RMODE_VM_CR4_ALWAYS_ON;
3095 hw_cr4 |= KVM_PMODE_VM_CR4_ALWAYS_ON;
3097 if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated()) {
3098 if (cr4 & X86_CR4_UMIP) {
3099 secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_DESC);
3100 hw_cr4 &= ~X86_CR4_UMIP;
3101 } else if (!is_guest_mode(vcpu) ||
3102 !nested_cpu_has2(get_vmcs12(vcpu), SECONDARY_EXEC_DESC)) {
3103 secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_DESC);
3107 if (cr4 & X86_CR4_VMXE) {
3109 * To use VMXON (and later other VMX instructions), a guest
3110 * must first be able to turn on cr4.VMXE (see handle_vmon()).
3111 * So basically the check on whether to allow nested VMX
3112 * is here. We operate under the default treatment of SMM,
3113 * so VMX cannot be enabled under SMM.
3115 if (!nested_vmx_allowed(vcpu) || is_smm(vcpu))
3119 if (vmx->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
3122 vcpu->arch.cr4 = cr4;
3123 kvm_register_mark_available(vcpu, VCPU_EXREG_CR4);
3125 if (!is_unrestricted_guest(vcpu)) {
3127 if (!is_paging(vcpu)) {
3128 hw_cr4 &= ~X86_CR4_PAE;
3129 hw_cr4 |= X86_CR4_PSE;
3130 } else if (!(cr4 & X86_CR4_PAE)) {
3131 hw_cr4 &= ~X86_CR4_PAE;
3136 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
3137 * hardware. To emulate this behavior, SMEP/SMAP/PKU needs
3138 * to be manually disabled when guest switches to non-paging
3141 * If !enable_unrestricted_guest, the CPU is always running
3142 * with CR0.PG=1 and CR4 needs to be modified.
3143 * If enable_unrestricted_guest, the CPU automatically
3144 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
3146 if (!is_paging(vcpu))
3147 hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
3150 vmcs_writel(CR4_READ_SHADOW, cr4);
3151 vmcs_writel(GUEST_CR4, hw_cr4);
3155 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3157 struct vcpu_vmx *vmx = to_vmx(vcpu);
3160 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3161 *var = vmx->rmode.segs[seg];
3162 if (seg == VCPU_SREG_TR
3163 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3165 var->base = vmx_read_guest_seg_base(vmx, seg);
3166 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3169 var->base = vmx_read_guest_seg_base(vmx, seg);
3170 var->limit = vmx_read_guest_seg_limit(vmx, seg);
3171 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3172 ar = vmx_read_guest_seg_ar(vmx, seg);
3173 var->unusable = (ar >> 16) & 1;
3174 var->type = ar & 15;
3175 var->s = (ar >> 4) & 1;
3176 var->dpl = (ar >> 5) & 3;
3178 * Some userspaces do not preserve unusable property. Since usable
3179 * segment has to be present according to VMX spec we can use present
3180 * property to amend userspace bug by making unusable segment always
3181 * nonpresent. vmx_segment_access_rights() already marks nonpresent
3182 * segment as unusable.
3184 var->present = !var->unusable;
3185 var->avl = (ar >> 12) & 1;
3186 var->l = (ar >> 13) & 1;
3187 var->db = (ar >> 14) & 1;
3188 var->g = (ar >> 15) & 1;
3191 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3193 struct kvm_segment s;
3195 if (to_vmx(vcpu)->rmode.vm86_active) {
3196 vmx_get_segment(vcpu, &s, seg);
3199 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3202 int vmx_get_cpl(struct kvm_vcpu *vcpu)
3204 struct vcpu_vmx *vmx = to_vmx(vcpu);
3206 if (unlikely(vmx->rmode.vm86_active))
3209 int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
3210 return VMX_AR_DPL(ar);
3214 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3218 if (var->unusable || !var->present)
3221 ar = var->type & 15;
3222 ar |= (var->s & 1) << 4;
3223 ar |= (var->dpl & 3) << 5;
3224 ar |= (var->present & 1) << 7;
3225 ar |= (var->avl & 1) << 12;
3226 ar |= (var->l & 1) << 13;
3227 ar |= (var->db & 1) << 14;
3228 ar |= (var->g & 1) << 15;
3234 void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3236 struct vcpu_vmx *vmx = to_vmx(vcpu);
3237 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3239 vmx_segment_cache_clear(vmx);
3241 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3242 vmx->rmode.segs[seg] = *var;
3243 if (seg == VCPU_SREG_TR)
3244 vmcs_write16(sf->selector, var->selector);
3246 fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3250 vmcs_writel(sf->base, var->base);
3251 vmcs_write32(sf->limit, var->limit);
3252 vmcs_write16(sf->selector, var->selector);
3255 * Fix the "Accessed" bit in AR field of segment registers for older
3257 * IA32 arch specifies that at the time of processor reset the
3258 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3259 * is setting it to 0 in the userland code. This causes invalid guest
3260 * state vmexit when "unrestricted guest" mode is turned on.
3261 * Fix for this setup issue in cpu_reset is being pushed in the qemu
3262 * tree. Newer qemu binaries with that qemu fix would not need this
3265 if (is_unrestricted_guest(vcpu) && (seg != VCPU_SREG_LDTR))
3266 var->type |= 0x1; /* Accessed */
3268 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3271 vmx->emulation_required = emulation_required(vcpu);
3274 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3276 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3278 *db = (ar >> 14) & 1;
3279 *l = (ar >> 13) & 1;
3282 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3284 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3285 dt->address = vmcs_readl(GUEST_IDTR_BASE);
3288 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3290 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3291 vmcs_writel(GUEST_IDTR_BASE, dt->address);
3294 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3296 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3297 dt->address = vmcs_readl(GUEST_GDTR_BASE);
3300 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3302 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3303 vmcs_writel(GUEST_GDTR_BASE, dt->address);
3306 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3308 struct kvm_segment var;
3311 vmx_get_segment(vcpu, &var, seg);
3313 if (seg == VCPU_SREG_CS)
3315 ar = vmx_segment_access_rights(&var);
3317 if (var.base != (var.selector << 4))
3319 if (var.limit != 0xffff)
3327 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3329 struct kvm_segment cs;
3330 unsigned int cs_rpl;
3332 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3333 cs_rpl = cs.selector & SEGMENT_RPL_MASK;
3337 if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
3341 if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
3342 if (cs.dpl > cs_rpl)
3345 if (cs.dpl != cs_rpl)
3351 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3355 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3357 struct kvm_segment ss;
3358 unsigned int ss_rpl;
3360 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3361 ss_rpl = ss.selector & SEGMENT_RPL_MASK;
3365 if (ss.type != 3 && ss.type != 7)
3369 if (ss.dpl != ss_rpl) /* DPL != RPL */
3377 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3379 struct kvm_segment var;
3382 vmx_get_segment(vcpu, &var, seg);
3383 rpl = var.selector & SEGMENT_RPL_MASK;
3391 if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
3392 if (var.dpl < rpl) /* DPL < RPL */
3396 /* TODO: Add other members to kvm_segment_field to allow checking for other access
3402 static bool tr_valid(struct kvm_vcpu *vcpu)
3404 struct kvm_segment tr;
3406 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3410 if (tr.selector & SEGMENT_TI_MASK) /* TI = 1 */
3412 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3420 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3422 struct kvm_segment ldtr;
3424 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3428 if (ldtr.selector & SEGMENT_TI_MASK) /* TI = 1 */
3438 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3440 struct kvm_segment cs, ss;
3442 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3443 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3445 return ((cs.selector & SEGMENT_RPL_MASK) ==
3446 (ss.selector & SEGMENT_RPL_MASK));
3450 * Check if guest state is valid. Returns true if valid, false if
3452 * We assume that registers are always usable
3454 bool __vmx_guest_state_valid(struct kvm_vcpu *vcpu)
3456 /* real mode guest state checks */
3457 if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
3458 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3460 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3462 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3464 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3466 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3468 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3471 /* protected mode guest state checks */
3472 if (!cs_ss_rpl_check(vcpu))
3474 if (!code_segment_valid(vcpu))
3476 if (!stack_segment_valid(vcpu))
3478 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3480 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3482 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3484 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3486 if (!tr_valid(vcpu))
3488 if (!ldtr_valid(vcpu))
3492 * - Add checks on RIP
3493 * - Add checks on RFLAGS
3499 static int init_rmode_tss(struct kvm *kvm)
3505 idx = srcu_read_lock(&kvm->srcu);
3506 fn = to_kvm_vmx(kvm)->tss_addr >> PAGE_SHIFT;
3507 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3510 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3511 r = kvm_write_guest_page(kvm, fn++, &data,
3512 TSS_IOPB_BASE_OFFSET, sizeof(u16));
3515 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3518 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3522 r = kvm_write_guest_page(kvm, fn, &data,
3523 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3526 srcu_read_unlock(&kvm->srcu, idx);
3530 static int init_rmode_identity_map(struct kvm *kvm)
3532 struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm);
3534 kvm_pfn_t identity_map_pfn;
3537 /* Protect kvm_vmx->ept_identity_pagetable_done. */
3538 mutex_lock(&kvm->slots_lock);
3540 if (likely(kvm_vmx->ept_identity_pagetable_done))
3543 if (!kvm_vmx->ept_identity_map_addr)
3544 kvm_vmx->ept_identity_map_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
3545 identity_map_pfn = kvm_vmx->ept_identity_map_addr >> PAGE_SHIFT;
3547 r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
3548 kvm_vmx->ept_identity_map_addr, PAGE_SIZE);
3552 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3555 /* Set up identity-mapping pagetable for EPT in real mode */
3556 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3557 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3558 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3559 r = kvm_write_guest_page(kvm, identity_map_pfn,
3560 &tmp, i * sizeof(tmp), sizeof(tmp));
3564 kvm_vmx->ept_identity_pagetable_done = true;
3567 mutex_unlock(&kvm->slots_lock);
3571 static void seg_setup(int seg)
3573 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3576 vmcs_write16(sf->selector, 0);
3577 vmcs_writel(sf->base, 0);
3578 vmcs_write32(sf->limit, 0xffff);
3580 if (seg == VCPU_SREG_CS)
3581 ar |= 0x08; /* code segment */
3583 vmcs_write32(sf->ar_bytes, ar);
3586 static int alloc_apic_access_page(struct kvm *kvm)
3591 mutex_lock(&kvm->slots_lock);
3592 if (kvm->arch.apic_access_page_done)
3594 r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
3595 APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
3599 page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
3600 if (is_error_page(page)) {
3606 * Do not pin the page in memory, so that memory hot-unplug
3607 * is able to migrate it.
3610 kvm->arch.apic_access_page_done = true;
3612 mutex_unlock(&kvm->slots_lock);
3616 int allocate_vpid(void)
3622 spin_lock(&vmx_vpid_lock);
3623 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3624 if (vpid < VMX_NR_VPIDS)
3625 __set_bit(vpid, vmx_vpid_bitmap);
3628 spin_unlock(&vmx_vpid_lock);
3632 void free_vpid(int vpid)
3634 if (!enable_vpid || vpid == 0)
3636 spin_lock(&vmx_vpid_lock);
3637 __clear_bit(vpid, vmx_vpid_bitmap);
3638 spin_unlock(&vmx_vpid_lock);
3641 static void vmx_clear_msr_bitmap_read(ulong *msr_bitmap, u32 msr)
3643 int f = sizeof(unsigned long);
3646 __clear_bit(msr, msr_bitmap + 0x000 / f);
3647 else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff))
3648 __clear_bit(msr & 0x1fff, msr_bitmap + 0x400 / f);
3651 static void vmx_clear_msr_bitmap_write(ulong *msr_bitmap, u32 msr)
3653 int f = sizeof(unsigned long);
3656 __clear_bit(msr, msr_bitmap + 0x800 / f);
3657 else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff))
3658 __clear_bit(msr & 0x1fff, msr_bitmap + 0xc00 / f);
3661 static void vmx_set_msr_bitmap_read(ulong *msr_bitmap, u32 msr)
3663 int f = sizeof(unsigned long);
3666 __set_bit(msr, msr_bitmap + 0x000 / f);
3667 else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff))
3668 __set_bit(msr & 0x1fff, msr_bitmap + 0x400 / f);
3671 static void vmx_set_msr_bitmap_write(ulong *msr_bitmap, u32 msr)
3673 int f = sizeof(unsigned long);
3676 __set_bit(msr, msr_bitmap + 0x800 / f);
3677 else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff))
3678 __set_bit(msr & 0x1fff, msr_bitmap + 0xc00 / f);
3681 static __always_inline void vmx_disable_intercept_for_msr(struct kvm_vcpu *vcpu,
3684 struct vcpu_vmx *vmx = to_vmx(vcpu);
3685 unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3687 if (!cpu_has_vmx_msr_bitmap())
3690 if (static_branch_unlikely(&enable_evmcs))
3691 evmcs_touch_msr_bitmap();
3694 * Mark the desired intercept state in shadow bitmap, this is needed
3695 * for resync when the MSR filters change.
3697 if (is_valid_passthrough_msr(msr)) {
3698 int idx = possible_passthrough_msr_slot(msr);
3700 if (idx != -ENOENT) {
3701 if (type & MSR_TYPE_R)
3702 clear_bit(idx, vmx->shadow_msr_intercept.read);
3703 if (type & MSR_TYPE_W)
3704 clear_bit(idx, vmx->shadow_msr_intercept.write);
3708 if ((type & MSR_TYPE_R) &&
3709 !kvm_msr_allowed(vcpu, msr, KVM_MSR_FILTER_READ)) {
3710 vmx_set_msr_bitmap_read(msr_bitmap, msr);
3711 type &= ~MSR_TYPE_R;
3714 if ((type & MSR_TYPE_W) &&
3715 !kvm_msr_allowed(vcpu, msr, KVM_MSR_FILTER_WRITE)) {
3716 vmx_set_msr_bitmap_write(msr_bitmap, msr);
3717 type &= ~MSR_TYPE_W;
3720 if (type & MSR_TYPE_R)
3721 vmx_clear_msr_bitmap_read(msr_bitmap, msr);
3723 if (type & MSR_TYPE_W)
3724 vmx_clear_msr_bitmap_write(msr_bitmap, msr);
3727 static __always_inline void vmx_enable_intercept_for_msr(struct kvm_vcpu *vcpu,
3730 struct vcpu_vmx *vmx = to_vmx(vcpu);
3731 unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3733 if (!cpu_has_vmx_msr_bitmap())
3736 if (static_branch_unlikely(&enable_evmcs))
3737 evmcs_touch_msr_bitmap();
3740 * Mark the desired intercept state in shadow bitmap, this is needed
3741 * for resync when the MSR filter changes.
3743 if (is_valid_passthrough_msr(msr)) {
3744 int idx = possible_passthrough_msr_slot(msr);
3746 if (idx != -ENOENT) {
3747 if (type & MSR_TYPE_R)
3748 set_bit(idx, vmx->shadow_msr_intercept.read);
3749 if (type & MSR_TYPE_W)
3750 set_bit(idx, vmx->shadow_msr_intercept.write);
3754 if (type & MSR_TYPE_R)
3755 vmx_set_msr_bitmap_read(msr_bitmap, msr);
3757 if (type & MSR_TYPE_W)
3758 vmx_set_msr_bitmap_write(msr_bitmap, msr);
3761 static __always_inline void vmx_set_intercept_for_msr(struct kvm_vcpu *vcpu,
3762 u32 msr, int type, bool value)
3765 vmx_enable_intercept_for_msr(vcpu, msr, type);
3767 vmx_disable_intercept_for_msr(vcpu, msr, type);
3770 static u8 vmx_msr_bitmap_mode(struct kvm_vcpu *vcpu)
3774 if (cpu_has_secondary_exec_ctrls() &&
3775 (secondary_exec_controls_get(to_vmx(vcpu)) &
3776 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
3777 mode |= MSR_BITMAP_MODE_X2APIC;
3778 if (enable_apicv && kvm_vcpu_apicv_active(vcpu))
3779 mode |= MSR_BITMAP_MODE_X2APIC_APICV;
3785 static void vmx_update_msr_bitmap_x2apic(struct kvm_vcpu *vcpu, u8 mode)
3789 for (msr = 0x800; msr <= 0x8ff; msr++) {
3790 bool intercepted = !!(mode & MSR_BITMAP_MODE_X2APIC_APICV);
3792 vmx_set_intercept_for_msr(vcpu, msr, MSR_TYPE_RW, intercepted);
3795 if (mode & MSR_BITMAP_MODE_X2APIC) {
3797 * TPR reads and writes can be virtualized even if virtual interrupt
3798 * delivery is not in use.
3800 vmx_disable_intercept_for_msr(vcpu, X2APIC_MSR(APIC_TASKPRI), MSR_TYPE_RW);
3801 if (mode & MSR_BITMAP_MODE_X2APIC_APICV) {
3802 vmx_enable_intercept_for_msr(vcpu, X2APIC_MSR(APIC_TMCCT), MSR_TYPE_RW);
3803 vmx_disable_intercept_for_msr(vcpu, X2APIC_MSR(APIC_EOI), MSR_TYPE_W);
3804 vmx_disable_intercept_for_msr(vcpu, X2APIC_MSR(APIC_SELF_IPI), MSR_TYPE_W);
3809 void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu)
3811 struct vcpu_vmx *vmx = to_vmx(vcpu);
3812 u8 mode = vmx_msr_bitmap_mode(vcpu);
3813 u8 changed = mode ^ vmx->msr_bitmap_mode;
3818 if (changed & (MSR_BITMAP_MODE_X2APIC | MSR_BITMAP_MODE_X2APIC_APICV))
3819 vmx_update_msr_bitmap_x2apic(vcpu, mode);
3821 vmx->msr_bitmap_mode = mode;
3824 void pt_update_intercept_for_msr(struct kvm_vcpu *vcpu)
3826 struct vcpu_vmx *vmx = to_vmx(vcpu);
3827 bool flag = !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
3830 vmx_set_intercept_for_msr(vcpu, MSR_IA32_RTIT_STATUS, MSR_TYPE_RW, flag);
3831 vmx_set_intercept_for_msr(vcpu, MSR_IA32_RTIT_OUTPUT_BASE, MSR_TYPE_RW, flag);
3832 vmx_set_intercept_for_msr(vcpu, MSR_IA32_RTIT_OUTPUT_MASK, MSR_TYPE_RW, flag);
3833 vmx_set_intercept_for_msr(vcpu, MSR_IA32_RTIT_CR3_MATCH, MSR_TYPE_RW, flag);
3834 for (i = 0; i < vmx->pt_desc.addr_range; i++) {
3835 vmx_set_intercept_for_msr(vcpu, MSR_IA32_RTIT_ADDR0_A + i * 2, MSR_TYPE_RW, flag);
3836 vmx_set_intercept_for_msr(vcpu, MSR_IA32_RTIT_ADDR0_B + i * 2, MSR_TYPE_RW, flag);
3840 static bool vmx_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
3842 struct vcpu_vmx *vmx = to_vmx(vcpu);
3847 if (WARN_ON_ONCE(!is_guest_mode(vcpu)) ||
3848 !nested_cpu_has_vid(get_vmcs12(vcpu)) ||
3849 WARN_ON_ONCE(!vmx->nested.virtual_apic_map.gfn))
3852 rvi = vmx_get_rvi();
3854 vapic_page = vmx->nested.virtual_apic_map.hva;
3855 vppr = *((u32 *)(vapic_page + APIC_PROCPRI));
3857 return ((rvi & 0xf0) > (vppr & 0xf0));
3860 static void vmx_msr_filter_changed(struct kvm_vcpu *vcpu)
3862 struct vcpu_vmx *vmx = to_vmx(vcpu);
3866 * Set intercept permissions for all potentially passed through MSRs
3867 * again. They will automatically get filtered through the MSR filter,
3868 * so we are back in sync after this.
3870 for (i = 0; i < ARRAY_SIZE(vmx_possible_passthrough_msrs); i++) {
3871 u32 msr = vmx_possible_passthrough_msrs[i];
3872 bool read = test_bit(i, vmx->shadow_msr_intercept.read);
3873 bool write = test_bit(i, vmx->shadow_msr_intercept.write);
3875 vmx_set_intercept_for_msr(vcpu, msr, MSR_TYPE_R, read);
3876 vmx_set_intercept_for_msr(vcpu, msr, MSR_TYPE_W, write);
3879 pt_update_intercept_for_msr(vcpu);
3880 vmx_update_msr_bitmap_x2apic(vcpu, vmx_msr_bitmap_mode(vcpu));
3883 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu,
3887 int pi_vec = nested ? POSTED_INTR_NESTED_VECTOR : POSTED_INTR_VECTOR;
3889 if (vcpu->mode == IN_GUEST_MODE) {
3891 * The vector of interrupt to be delivered to vcpu had
3892 * been set in PIR before this function.
3894 * Following cases will be reached in this block, and
3895 * we always send a notification event in all cases as
3898 * Case 1: vcpu keeps in non-root mode. Sending a
3899 * notification event posts the interrupt to vcpu.
3901 * Case 2: vcpu exits to root mode and is still
3902 * runnable. PIR will be synced to vIRR before the
3903 * next vcpu entry. Sending a notification event in
3904 * this case has no effect, as vcpu is not in root
3907 * Case 3: vcpu exits to root mode and is blocked.
3908 * vcpu_block() has already synced PIR to vIRR and
3909 * never blocks vcpu if vIRR is not cleared. Therefore,
3910 * a blocked vcpu here does not wait for any requested
3911 * interrupts in PIR, and sending a notification event
3912 * which has no effect is safe here.
3915 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu), pi_vec);
3922 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
3925 struct vcpu_vmx *vmx = to_vmx(vcpu);
3927 if (is_guest_mode(vcpu) &&
3928 vector == vmx->nested.posted_intr_nv) {
3930 * If a posted intr is not recognized by hardware,
3931 * we will accomplish it in the next vmentry.
3933 vmx->nested.pi_pending = true;
3934 kvm_make_request(KVM_REQ_EVENT, vcpu);
3935 /* the PIR and ON have been set by L1. */
3936 if (!kvm_vcpu_trigger_posted_interrupt(vcpu, true))
3937 kvm_vcpu_kick(vcpu);
3943 * Send interrupt to vcpu via posted interrupt way.
3944 * 1. If target vcpu is running(non-root mode), send posted interrupt
3945 * notification to vcpu and hardware will sync PIR to vIRR atomically.
3946 * 2. If target vcpu isn't running(root mode), kick it to pick up the
3947 * interrupt from PIR in next vmentry.
3949 static int vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
3951 struct vcpu_vmx *vmx = to_vmx(vcpu);
3954 r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
3958 if (!vcpu->arch.apicv_active)
3961 if (pi_test_and_set_pir(vector, &vmx->pi_desc))
3964 /* If a previous notification has sent the IPI, nothing to do. */
3965 if (pi_test_and_set_on(&vmx->pi_desc))
3968 if (vcpu != kvm_get_running_vcpu() &&
3969 !kvm_vcpu_trigger_posted_interrupt(vcpu, false))
3970 kvm_vcpu_kick(vcpu);
3976 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
3977 * will not change in the lifetime of the guest.
3978 * Note that host-state that does change is set elsewhere. E.g., host-state
3979 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
3981 void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
3985 unsigned long cr0, cr3, cr4;
3988 WARN_ON(cr0 & X86_CR0_TS);
3989 vmcs_writel(HOST_CR0, cr0); /* 22.2.3 */
3992 * Save the most likely value for this task's CR3 in the VMCS.
3993 * We can't use __get_current_cr3_fast() because we're not atomic.
3996 vmcs_writel(HOST_CR3, cr3); /* 22.2.3 FIXME: shadow tables */
3997 vmx->loaded_vmcs->host_state.cr3 = cr3;
3999 /* Save the most likely value for this task's CR4 in the VMCS. */
4000 cr4 = cr4_read_shadow();
4001 vmcs_writel(HOST_CR4, cr4); /* 22.2.3, 22.2.5 */
4002 vmx->loaded_vmcs->host_state.cr4 = cr4;
4004 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
4005 #ifdef CONFIG_X86_64
4007 * Load null selectors, so we can avoid reloading them in
4008 * vmx_prepare_switch_to_host(), in case userspace uses
4009 * the null selectors too (the expected case).
4011 vmcs_write16(HOST_DS_SELECTOR, 0);
4012 vmcs_write16(HOST_ES_SELECTOR, 0);
4014 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4015 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4017 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4018 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
4020 vmcs_writel(HOST_IDTR_BASE, host_idt_base); /* 22.2.4 */
4022 vmcs_writel(HOST_RIP, (unsigned long)vmx_vmexit); /* 22.2.5 */
4024 rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
4025 vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
4026 rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
4027 vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */
4029 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
4030 rdmsr(MSR_IA32_CR_PAT, low32, high32);
4031 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
4034 if (cpu_has_load_ia32_efer())
4035 vmcs_write64(HOST_IA32_EFER, host_efer);
4038 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
4040 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_POSSIBLE_CR4_GUEST_BITS;
4042 vmx->vcpu.arch.cr4_guest_owned_bits &= ~X86_CR4_PGE;
4043 if (is_guest_mode(&vmx->vcpu))
4044 vmx->vcpu.arch.cr4_guest_owned_bits &=
4045 ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
4046 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
4049 u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
4051 u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
4053 if (!kvm_vcpu_apicv_active(&vmx->vcpu))
4054 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
4057 pin_based_exec_ctrl &= ~PIN_BASED_VIRTUAL_NMIS;
4059 if (!enable_preemption_timer)
4060 pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
4062 return pin_based_exec_ctrl;
4065 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
4067 struct vcpu_vmx *vmx = to_vmx(vcpu);
4069 pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
4070 if (cpu_has_secondary_exec_ctrls()) {
4071 if (kvm_vcpu_apicv_active(vcpu))
4072 secondary_exec_controls_setbit(vmx,
4073 SECONDARY_EXEC_APIC_REGISTER_VIRT |
4074 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4076 secondary_exec_controls_clearbit(vmx,
4077 SECONDARY_EXEC_APIC_REGISTER_VIRT |
4078 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4081 if (cpu_has_vmx_msr_bitmap())
4082 vmx_update_msr_bitmap(vcpu);
4085 u32 vmx_exec_control(struct vcpu_vmx *vmx)
4087 u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
4089 if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
4090 exec_control &= ~CPU_BASED_MOV_DR_EXITING;
4092 if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
4093 exec_control &= ~CPU_BASED_TPR_SHADOW;
4094 #ifdef CONFIG_X86_64
4095 exec_control |= CPU_BASED_CR8_STORE_EXITING |
4096 CPU_BASED_CR8_LOAD_EXITING;
4100 exec_control |= CPU_BASED_CR3_STORE_EXITING |
4101 CPU_BASED_CR3_LOAD_EXITING |
4102 CPU_BASED_INVLPG_EXITING;
4103 if (kvm_mwait_in_guest(vmx->vcpu.kvm))
4104 exec_control &= ~(CPU_BASED_MWAIT_EXITING |
4105 CPU_BASED_MONITOR_EXITING);
4106 if (kvm_hlt_in_guest(vmx->vcpu.kvm))
4107 exec_control &= ~CPU_BASED_HLT_EXITING;
4108 return exec_control;
4112 * Adjust a single secondary execution control bit to intercept/allow an
4113 * instruction in the guest. This is usually done based on whether or not a
4114 * feature has been exposed to the guest in order to correctly emulate faults.
4117 vmx_adjust_secondary_exec_control(struct vcpu_vmx *vmx, u32 *exec_control,
4118 u32 control, bool enabled, bool exiting)
4121 * If the control is for an opt-in feature, clear the control if the
4122 * feature is not exposed to the guest, i.e. not enabled. If the
4123 * control is opt-out, i.e. an exiting control, clear the control if
4124 * the feature _is_ exposed to the guest, i.e. exiting/interception is
4125 * disabled for the associated instruction. Note, the caller is
4126 * responsible presetting exec_control to set all supported bits.
4128 if (enabled == exiting)
4129 *exec_control &= ~control;
4132 * Update the nested MSR settings so that a nested VMM can/can't set
4133 * controls for features that are/aren't exposed to the guest.
4137 vmx->nested.msrs.secondary_ctls_high |= control;
4139 vmx->nested.msrs.secondary_ctls_high &= ~control;
4144 * Wrapper macro for the common case of adjusting a secondary execution control
4145 * based on a single guest CPUID bit, with a dedicated feature bit. This also
4146 * verifies that the control is actually supported by KVM and hardware.
4148 #define vmx_adjust_sec_exec_control(vmx, exec_control, name, feat_name, ctrl_name, exiting) \
4152 if (cpu_has_vmx_##name()) { \
4153 __enabled = guest_cpuid_has(&(vmx)->vcpu, \
4154 X86_FEATURE_##feat_name); \
4155 vmx_adjust_secondary_exec_control(vmx, exec_control, \
4156 SECONDARY_EXEC_##ctrl_name, __enabled, exiting); \
4160 /* More macro magic for ENABLE_/opt-in versus _EXITING/opt-out controls. */
4161 #define vmx_adjust_sec_exec_feature(vmx, exec_control, lname, uname) \
4162 vmx_adjust_sec_exec_control(vmx, exec_control, lname, uname, ENABLE_##uname, false)
4164 #define vmx_adjust_sec_exec_exiting(vmx, exec_control, lname, uname) \
4165 vmx_adjust_sec_exec_control(vmx, exec_control, lname, uname, uname##_EXITING, true)
4167 static void vmx_compute_secondary_exec_control(struct vcpu_vmx *vmx)
4169 struct kvm_vcpu *vcpu = &vmx->vcpu;
4171 u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
4173 if (vmx_pt_mode_is_system())
4174 exec_control &= ~(SECONDARY_EXEC_PT_USE_GPA | SECONDARY_EXEC_PT_CONCEAL_VMX);
4175 if (!cpu_need_virtualize_apic_accesses(vcpu))
4176 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
4178 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
4180 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
4181 enable_unrestricted_guest = 0;
4183 if (!enable_unrestricted_guest)
4184 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4185 if (kvm_pause_in_guest(vmx->vcpu.kvm))
4186 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
4187 if (!kvm_vcpu_apicv_active(vcpu))
4188 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
4189 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4190 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
4192 /* SECONDARY_EXEC_DESC is enabled/disabled on writes to CR4.UMIP,
4193 * in vmx_set_cr4. */
4194 exec_control &= ~SECONDARY_EXEC_DESC;
4196 /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4198 We can NOT enable shadow_vmcs here because we don't have yet
4201 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4204 exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
4206 if (cpu_has_vmx_xsaves()) {
4207 /* Exposing XSAVES only when XSAVE is exposed */
4208 bool xsaves_enabled =
4209 boot_cpu_has(X86_FEATURE_XSAVE) &&
4210 guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
4211 guest_cpuid_has(vcpu, X86_FEATURE_XSAVES);
4213 vcpu->arch.xsaves_enabled = xsaves_enabled;
4215 vmx_adjust_secondary_exec_control(vmx, &exec_control,
4216 SECONDARY_EXEC_XSAVES,
4217 xsaves_enabled, false);
4220 vmx_adjust_sec_exec_feature(vmx, &exec_control, rdtscp, RDTSCP);
4223 * Expose INVPCID if and only if PCID is also exposed to the guest.
4224 * INVPCID takes a #UD when it's disabled in the VMCS, but a #GP or #PF
4225 * if CR4.PCIDE=0. Enumerating CPUID.INVPCID=1 would lead to incorrect
4226 * behavior from the guest perspective (it would expect #GP or #PF).
4228 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
4229 guest_cpuid_clear(vcpu, X86_FEATURE_INVPCID);
4230 vmx_adjust_sec_exec_feature(vmx, &exec_control, invpcid, INVPCID);
4233 vmx_adjust_sec_exec_exiting(vmx, &exec_control, rdrand, RDRAND);
4234 vmx_adjust_sec_exec_exiting(vmx, &exec_control, rdseed, RDSEED);
4236 vmx_adjust_sec_exec_control(vmx, &exec_control, waitpkg, WAITPKG,
4237 ENABLE_USR_WAIT_PAUSE, false);
4239 vmx->secondary_exec_control = exec_control;
4242 static void ept_set_mmio_spte_mask(void)
4245 * EPT Misconfigurations can be generated if the value of bits 2:0
4246 * of an EPT paging-structure entry is 110b (write/execute).
4248 kvm_mmu_set_mmio_spte_mask(VMX_EPT_MISCONFIG_WX_VALUE, 0);
4251 #define VMX_XSS_EXIT_BITMAP 0
4254 * Noting that the initialization of Guest-state Area of VMCS is in
4257 static void init_vmcs(struct vcpu_vmx *vmx)
4260 nested_vmx_set_vmcs_shadowing_bitmap();
4262 if (cpu_has_vmx_msr_bitmap())
4263 vmcs_write64(MSR_BITMAP, __pa(vmx->vmcs01.msr_bitmap));
4265 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4268 pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
4270 exec_controls_set(vmx, vmx_exec_control(vmx));
4272 if (cpu_has_secondary_exec_ctrls()) {
4273 vmx_compute_secondary_exec_control(vmx);
4274 secondary_exec_controls_set(vmx, vmx->secondary_exec_control);
4277 if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
4278 vmcs_write64(EOI_EXIT_BITMAP0, 0);
4279 vmcs_write64(EOI_EXIT_BITMAP1, 0);
4280 vmcs_write64(EOI_EXIT_BITMAP2, 0);
4281 vmcs_write64(EOI_EXIT_BITMAP3, 0);
4283 vmcs_write16(GUEST_INTR_STATUS, 0);
4285 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4286 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4289 if (!kvm_pause_in_guest(vmx->vcpu.kvm)) {
4290 vmcs_write32(PLE_GAP, ple_gap);
4291 vmx->ple_window = ple_window;
4292 vmx->ple_window_dirty = true;
4295 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4296 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4297 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
4299 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
4300 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
4301 vmx_set_constant_host_state(vmx);
4302 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4303 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4305 if (cpu_has_vmx_vmfunc())
4306 vmcs_write64(VM_FUNCTION_CONTROL, 0);
4308 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4309 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4310 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
4311 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4312 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
4314 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
4315 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
4317 vm_exit_controls_set(vmx, vmx_vmexit_ctrl());
4319 /* 22.2.1, 20.8.1 */
4320 vm_entry_controls_set(vmx, vmx_vmentry_ctrl());
4322 vmx->vcpu.arch.cr0_guest_owned_bits = KVM_POSSIBLE_CR0_GUEST_BITS;
4323 vmcs_writel(CR0_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr0_guest_owned_bits);
4325 set_cr4_guest_host_mask(vmx);
4328 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4330 if (cpu_has_vmx_xsaves())
4331 vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
4334 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
4335 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
4338 if (cpu_has_vmx_encls_vmexit())
4339 vmcs_write64(ENCLS_EXITING_BITMAP, -1ull);
4341 if (vmx_pt_mode_is_host_guest()) {
4342 memset(&vmx->pt_desc, 0, sizeof(vmx->pt_desc));
4343 /* Bit[6~0] are forced to 1, writes are ignored. */
4344 vmx->pt_desc.guest.output_mask = 0x7F;
4345 vmcs_write64(GUEST_IA32_RTIT_CTL, 0);
4349 * If EPT is enabled, #PF is only trapped if MAXPHYADDR is mismatched
4350 * between guest and host. In that case we only care about present
4354 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, PFERR_PRESENT_MASK);
4355 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, PFERR_PRESENT_MASK);
4359 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
4361 struct vcpu_vmx *vmx = to_vmx(vcpu);
4362 struct msr_data apic_base_msr;
4365 vmx->rmode.vm86_active = 0;
4368 vmx->msr_ia32_umwait_control = 0;
4370 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
4371 vmx->hv_deadline_tsc = -1;
4372 kvm_set_cr8(vcpu, 0);
4375 apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
4376 MSR_IA32_APICBASE_ENABLE;
4377 if (kvm_vcpu_is_reset_bsp(vcpu))
4378 apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
4379 apic_base_msr.host_initiated = true;
4380 kvm_set_apic_base(vcpu, &apic_base_msr);
4383 vmx_segment_cache_clear(vmx);
4385 seg_setup(VCPU_SREG_CS);
4386 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4387 vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
4389 seg_setup(VCPU_SREG_DS);
4390 seg_setup(VCPU_SREG_ES);
4391 seg_setup(VCPU_SREG_FS);
4392 seg_setup(VCPU_SREG_GS);
4393 seg_setup(VCPU_SREG_SS);
4395 vmcs_write16(GUEST_TR_SELECTOR, 0);
4396 vmcs_writel(GUEST_TR_BASE, 0);
4397 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4398 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4400 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4401 vmcs_writel(GUEST_LDTR_BASE, 0);
4402 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4403 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4406 vmcs_write32(GUEST_SYSENTER_CS, 0);
4407 vmcs_writel(GUEST_SYSENTER_ESP, 0);
4408 vmcs_writel(GUEST_SYSENTER_EIP, 0);
4409 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4412 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
4413 kvm_rip_write(vcpu, 0xfff0);
4415 vmcs_writel(GUEST_GDTR_BASE, 0);
4416 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4418 vmcs_writel(GUEST_IDTR_BASE, 0);
4419 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4421 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4422 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4423 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4424 if (kvm_mpx_supported())
4425 vmcs_write64(GUEST_BNDCFGS, 0);
4429 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
4431 if (cpu_has_vmx_tpr_shadow() && !init_event) {
4432 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4433 if (cpu_need_tpr_shadow(vcpu))
4434 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4435 __pa(vcpu->arch.apic->regs));
4436 vmcs_write32(TPR_THRESHOLD, 0);
4439 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4441 cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4442 vmx->vcpu.arch.cr0 = cr0;
4443 vmx_set_cr0(vcpu, cr0); /* enter rmode */
4444 vmx_set_cr4(vcpu, 0);
4445 vmx_set_efer(vcpu, 0);
4447 update_exception_bitmap(vcpu);
4449 vpid_sync_context(vmx->vpid);
4451 vmx_clear_hlt(vcpu);
4454 static void enable_irq_window(struct kvm_vcpu *vcpu)
4456 exec_controls_setbit(to_vmx(vcpu), CPU_BASED_INTR_WINDOW_EXITING);
4459 static void enable_nmi_window(struct kvm_vcpu *vcpu)
4462 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4463 enable_irq_window(vcpu);
4467 exec_controls_setbit(to_vmx(vcpu), CPU_BASED_NMI_WINDOW_EXITING);
4470 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4472 struct vcpu_vmx *vmx = to_vmx(vcpu);
4474 int irq = vcpu->arch.interrupt.nr;
4476 trace_kvm_inj_virq(irq);
4478 ++vcpu->stat.irq_injections;
4479 if (vmx->rmode.vm86_active) {
4481 if (vcpu->arch.interrupt.soft)
4482 inc_eip = vcpu->arch.event_exit_inst_len;
4483 kvm_inject_realmode_interrupt(vcpu, irq, inc_eip);
4486 intr = irq | INTR_INFO_VALID_MASK;
4487 if (vcpu->arch.interrupt.soft) {
4488 intr |= INTR_TYPE_SOFT_INTR;
4489 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4490 vmx->vcpu.arch.event_exit_inst_len);
4492 intr |= INTR_TYPE_EXT_INTR;
4493 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4495 vmx_clear_hlt(vcpu);
4498 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4500 struct vcpu_vmx *vmx = to_vmx(vcpu);
4504 * Tracking the NMI-blocked state in software is built upon
4505 * finding the next open IRQ window. This, in turn, depends on
4506 * well-behaving guests: They have to keep IRQs disabled at
4507 * least as long as the NMI handler runs. Otherwise we may
4508 * cause NMI nesting, maybe breaking the guest. But as this is
4509 * highly unlikely, we can live with the residual risk.
4511 vmx->loaded_vmcs->soft_vnmi_blocked = 1;
4512 vmx->loaded_vmcs->vnmi_blocked_time = 0;
4515 ++vcpu->stat.nmi_injections;
4516 vmx->loaded_vmcs->nmi_known_unmasked = false;
4518 if (vmx->rmode.vm86_active) {
4519 kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0);
4523 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4524 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4526 vmx_clear_hlt(vcpu);
4529 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4531 struct vcpu_vmx *vmx = to_vmx(vcpu);
4535 return vmx->loaded_vmcs->soft_vnmi_blocked;
4536 if (vmx->loaded_vmcs->nmi_known_unmasked)
4538 masked = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4539 vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4543 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4545 struct vcpu_vmx *vmx = to_vmx(vcpu);
4548 if (vmx->loaded_vmcs->soft_vnmi_blocked != masked) {
4549 vmx->loaded_vmcs->soft_vnmi_blocked = masked;
4550 vmx->loaded_vmcs->vnmi_blocked_time = 0;
4553 vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4555 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4556 GUEST_INTR_STATE_NMI);
4558 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4559 GUEST_INTR_STATE_NMI);
4563 bool vmx_nmi_blocked(struct kvm_vcpu *vcpu)
4565 if (is_guest_mode(vcpu) && nested_exit_on_nmi(vcpu))
4568 if (!enable_vnmi && to_vmx(vcpu)->loaded_vmcs->soft_vnmi_blocked)
4571 return (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4572 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI |
4573 GUEST_INTR_STATE_NMI));
4576 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu, bool for_injection)
4578 if (to_vmx(vcpu)->nested.nested_run_pending)
4581 /* An NMI must not be injected into L2 if it's supposed to VM-Exit. */
4582 if (for_injection && is_guest_mode(vcpu) && nested_exit_on_nmi(vcpu))
4585 return !vmx_nmi_blocked(vcpu);
4588 bool vmx_interrupt_blocked(struct kvm_vcpu *vcpu)
4590 if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu))
4593 return !(vmx_get_rflags(vcpu) & X86_EFLAGS_IF) ||
4594 (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4595 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4598 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu, bool for_injection)
4600 if (to_vmx(vcpu)->nested.nested_run_pending)
4604 * An IRQ must not be injected into L2 if it's supposed to VM-Exit,
4605 * e.g. if the IRQ arrived asynchronously after checking nested events.
4607 if (for_injection && is_guest_mode(vcpu) && nested_exit_on_intr(vcpu))
4610 return !vmx_interrupt_blocked(vcpu);
4613 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4617 if (enable_unrestricted_guest)
4620 mutex_lock(&kvm->slots_lock);
4621 ret = __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
4623 mutex_unlock(&kvm->slots_lock);
4627 to_kvm_vmx(kvm)->tss_addr = addr;
4628 return init_rmode_tss(kvm);
4631 static int vmx_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
4633 to_kvm_vmx(kvm)->ept_identity_map_addr = ident_addr;
4637 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
4642 * Update instruction length as we may reinject the exception
4643 * from user space while in guest debugging mode.
4645 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4646 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4647 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4651 return !(vcpu->guest_debug &
4652 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP));
4666 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4667 int vec, u32 err_code)
4670 * Instruction with address size override prefix opcode 0x67
4671 * Cause the #SS fault with 0 error code in VM86 mode.
4673 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
4674 if (kvm_emulate_instruction(vcpu, 0)) {
4675 if (vcpu->arch.halt_request) {
4676 vcpu->arch.halt_request = 0;
4677 return kvm_vcpu_halt(vcpu);
4685 * Forward all other exceptions that are valid in real mode.
4686 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4687 * the required debugging infrastructure rework.
4689 kvm_queue_exception(vcpu, vec);
4694 * Trigger machine check on the host. We assume all the MSRs are already set up
4695 * by the CPU and that we still run on the same CPU as the MCE occurred on.
4696 * We pass a fake environment to the machine check handler because we want
4697 * the guest to be always treated like user space, no matter what context
4698 * it used internally.
4700 static void kvm_machine_check(void)
4702 #if defined(CONFIG_X86_MCE)
4703 struct pt_regs regs = {
4704 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
4705 .flags = X86_EFLAGS_IF,
4708 do_machine_check(®s);
4712 static int handle_machine_check(struct kvm_vcpu *vcpu)
4714 /* handled by vmx_vcpu_run() */
4719 * If the host has split lock detection disabled, then #AC is
4720 * unconditionally injected into the guest, which is the pre split lock
4721 * detection behaviour.
4723 * If the host has split lock detection enabled then #AC is
4724 * only injected into the guest when:
4725 * - Guest CPL == 3 (user mode)
4726 * - Guest has #AC detection enabled in CR0
4727 * - Guest EFLAGS has AC bit set
4729 static inline bool guest_inject_ac(struct kvm_vcpu *vcpu)
4731 if (!boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
4734 return vmx_get_cpl(vcpu) == 3 && kvm_read_cr0_bits(vcpu, X86_CR0_AM) &&
4735 (kvm_get_rflags(vcpu) & X86_EFLAGS_AC);
4738 static int handle_exception_nmi(struct kvm_vcpu *vcpu)
4740 struct vcpu_vmx *vmx = to_vmx(vcpu);
4741 struct kvm_run *kvm_run = vcpu->run;
4742 u32 intr_info, ex_no, error_code;
4743 unsigned long cr2, rip, dr6;
4746 vect_info = vmx->idt_vectoring_info;
4747 intr_info = vmx_get_intr_info(vcpu);
4749 if (is_machine_check(intr_info) || is_nmi(intr_info))
4750 return 1; /* handled by handle_exception_nmi_irqoff() */
4752 if (is_invalid_opcode(intr_info))
4753 return handle_ud(vcpu);
4756 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4757 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4759 if (!vmx->rmode.vm86_active && is_gp_fault(intr_info)) {
4760 WARN_ON_ONCE(!enable_vmware_backdoor);
4763 * VMware backdoor emulation on #GP interception only handles
4764 * IN{S}, OUT{S}, and RDPMC, none of which generate a non-zero
4765 * error code on #GP.
4768 kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
4771 return kvm_emulate_instruction(vcpu, EMULTYPE_VMWARE_GP);
4775 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
4776 * MMIO, it is better to report an internal error.
4777 * See the comments in vmx_handle_exit.
4779 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4780 !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
4781 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4782 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4783 vcpu->run->internal.ndata = 4;
4784 vcpu->run->internal.data[0] = vect_info;
4785 vcpu->run->internal.data[1] = intr_info;
4786 vcpu->run->internal.data[2] = error_code;
4787 vcpu->run->internal.data[3] = vcpu->arch.last_vmentry_cpu;
4791 if (is_page_fault(intr_info)) {
4792 cr2 = vmx_get_exit_qual(vcpu);
4793 if (enable_ept && !vcpu->arch.apf.host_apf_flags) {
4795 * EPT will cause page fault only if we need to
4796 * detect illegal GPAs.
4798 kvm_fixup_and_inject_pf_error(vcpu, cr2, error_code);
4801 return kvm_handle_page_fault(vcpu, error_code, cr2, NULL, 0);
4804 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4806 if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
4807 return handle_rmode_exception(vcpu, ex_no, error_code);
4811 dr6 = vmx_get_exit_qual(vcpu);
4812 if (!(vcpu->guest_debug &
4813 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4814 if (is_icebp(intr_info))
4815 WARN_ON(!skip_emulated_instruction(vcpu));
4817 kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
4820 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
4821 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4825 * Update instruction length as we may reinject #BP from
4826 * user space while in guest debugging mode. Reading it for
4827 * #DB as well causes no harm, it is not used in that case.
4829 vmx->vcpu.arch.event_exit_inst_len =
4830 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4831 kvm_run->exit_reason = KVM_EXIT_DEBUG;
4832 rip = kvm_rip_read(vcpu);
4833 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4834 kvm_run->debug.arch.exception = ex_no;
4837 if (guest_inject_ac(vcpu)) {
4838 kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
4843 * Handle split lock. Depending on detection mode this will
4844 * either warn and disable split lock detection for this
4845 * task or force SIGBUS on it.
4847 if (handle_guest_split_lock(kvm_rip_read(vcpu)))
4851 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4852 kvm_run->ex.exception = ex_no;
4853 kvm_run->ex.error_code = error_code;
4859 static __always_inline int handle_external_interrupt(struct kvm_vcpu *vcpu)
4861 ++vcpu->stat.irq_exits;
4865 static int handle_triple_fault(struct kvm_vcpu *vcpu)
4867 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4868 vcpu->mmio_needed = 0;
4872 static int handle_io(struct kvm_vcpu *vcpu)
4874 unsigned long exit_qualification;
4875 int size, in, string;
4878 exit_qualification = vmx_get_exit_qual(vcpu);
4879 string = (exit_qualification & 16) != 0;
4881 ++vcpu->stat.io_exits;
4884 return kvm_emulate_instruction(vcpu, 0);
4886 port = exit_qualification >> 16;
4887 size = (exit_qualification & 7) + 1;
4888 in = (exit_qualification & 8) != 0;
4890 return kvm_fast_pio(vcpu, size, port, in);
4894 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4897 * Patch in the VMCALL instruction:
4899 hypercall[0] = 0x0f;
4900 hypercall[1] = 0x01;
4901 hypercall[2] = 0xc1;
4904 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
4905 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
4907 if (is_guest_mode(vcpu)) {
4908 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4909 unsigned long orig_val = val;
4912 * We get here when L2 changed cr0 in a way that did not change
4913 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
4914 * but did change L0 shadowed bits. So we first calculate the
4915 * effective cr0 value that L1 would like to write into the
4916 * hardware. It consists of the L2-owned bits from the new
4917 * value combined with the L1-owned bits from L1's guest_cr0.
4919 val = (val & ~vmcs12->cr0_guest_host_mask) |
4920 (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
4922 if (!nested_guest_cr0_valid(vcpu, val))
4925 if (kvm_set_cr0(vcpu, val))
4927 vmcs_writel(CR0_READ_SHADOW, orig_val);
4930 if (to_vmx(vcpu)->nested.vmxon &&
4931 !nested_host_cr0_valid(vcpu, val))
4934 return kvm_set_cr0(vcpu, val);
4938 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
4940 if (is_guest_mode(vcpu)) {
4941 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4942 unsigned long orig_val = val;
4944 /* analogously to handle_set_cr0 */
4945 val = (val & ~vmcs12->cr4_guest_host_mask) |
4946 (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
4947 if (kvm_set_cr4(vcpu, val))
4949 vmcs_writel(CR4_READ_SHADOW, orig_val);
4952 return kvm_set_cr4(vcpu, val);
4955 static int handle_desc(struct kvm_vcpu *vcpu)
4957 WARN_ON(!(vcpu->arch.cr4 & X86_CR4_UMIP));
4958 return kvm_emulate_instruction(vcpu, 0);
4961 static int handle_cr(struct kvm_vcpu *vcpu)
4963 unsigned long exit_qualification, val;
4969 exit_qualification = vmx_get_exit_qual(vcpu);
4970 cr = exit_qualification & 15;
4971 reg = (exit_qualification >> 8) & 15;
4972 switch ((exit_qualification >> 4) & 3) {
4973 case 0: /* mov to cr */
4974 val = kvm_register_readl(vcpu, reg);
4975 trace_kvm_cr_write(cr, val);
4978 err = handle_set_cr0(vcpu, val);
4979 return kvm_complete_insn_gp(vcpu, err);
4981 WARN_ON_ONCE(enable_unrestricted_guest);
4982 err = kvm_set_cr3(vcpu, val);
4983 return kvm_complete_insn_gp(vcpu, err);
4985 err = handle_set_cr4(vcpu, val);
4986 return kvm_complete_insn_gp(vcpu, err);
4988 u8 cr8_prev = kvm_get_cr8(vcpu);
4990 err = kvm_set_cr8(vcpu, cr8);
4991 ret = kvm_complete_insn_gp(vcpu, err);
4992 if (lapic_in_kernel(vcpu))
4994 if (cr8_prev <= cr8)
4997 * TODO: we might be squashing a
4998 * KVM_GUESTDBG_SINGLESTEP-triggered
4999 * KVM_EXIT_DEBUG here.
5001 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
5007 WARN_ONCE(1, "Guest should always own CR0.TS");
5008 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
5009 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
5010 return kvm_skip_emulated_instruction(vcpu);
5011 case 1: /*mov from cr*/
5014 WARN_ON_ONCE(enable_unrestricted_guest);
5015 val = kvm_read_cr3(vcpu);
5016 kvm_register_write(vcpu, reg, val);
5017 trace_kvm_cr_read(cr, val);
5018 return kvm_skip_emulated_instruction(vcpu);
5020 val = kvm_get_cr8(vcpu);
5021 kvm_register_write(vcpu, reg, val);
5022 trace_kvm_cr_read(cr, val);
5023 return kvm_skip_emulated_instruction(vcpu);
5027 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5028 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
5029 kvm_lmsw(vcpu, val);
5031 return kvm_skip_emulated_instruction(vcpu);
5035 vcpu->run->exit_reason = 0;
5036 vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
5037 (int)(exit_qualification >> 4) & 3, cr);
5041 static int handle_dr(struct kvm_vcpu *vcpu)
5043 unsigned long exit_qualification;
5046 exit_qualification = vmx_get_exit_qual(vcpu);
5047 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
5049 /* First, if DR does not exist, trigger UD */
5050 if (!kvm_require_dr(vcpu, dr))
5053 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
5054 if (!kvm_require_cpl(vcpu, 0))
5056 dr7 = vmcs_readl(GUEST_DR7);
5059 * As the vm-exit takes precedence over the debug trap, we
5060 * need to emulate the latter, either for the host or the
5061 * guest debugging itself.
5063 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5064 vcpu->run->debug.arch.dr6 = DR6_BD | DR6_RTM | DR6_FIXED_1;
5065 vcpu->run->debug.arch.dr7 = dr7;
5066 vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
5067 vcpu->run->debug.arch.exception = DB_VECTOR;
5068 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
5071 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BD);
5076 if (vcpu->guest_debug == 0) {
5077 exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
5080 * No more DR vmexits; force a reload of the debug registers
5081 * and reenter on this instruction. The next vmexit will
5082 * retrieve the full state of the debug registers.
5084 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
5088 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
5089 if (exit_qualification & TYPE_MOV_FROM_DR) {
5092 if (kvm_get_dr(vcpu, dr, &val))
5094 kvm_register_write(vcpu, reg, val);
5096 if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
5099 return kvm_skip_emulated_instruction(vcpu);
5102 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
5104 get_debugreg(vcpu->arch.db[0], 0);
5105 get_debugreg(vcpu->arch.db[1], 1);
5106 get_debugreg(vcpu->arch.db[2], 2);
5107 get_debugreg(vcpu->arch.db[3], 3);
5108 get_debugreg(vcpu->arch.dr6, 6);
5109 vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
5111 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
5112 exec_controls_setbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
5115 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
5117 vmcs_writel(GUEST_DR7, val);
5120 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
5122 kvm_apic_update_ppr(vcpu);
5126 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
5128 exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_INTR_WINDOW_EXITING);
5130 kvm_make_request(KVM_REQ_EVENT, vcpu);
5132 ++vcpu->stat.irq_window_exits;
5136 static int handle_vmcall(struct kvm_vcpu *vcpu)
5138 return kvm_emulate_hypercall(vcpu);
5141 static int handle_invd(struct kvm_vcpu *vcpu)
5143 /* Treat an INVD instruction as a NOP and just skip it. */
5144 return kvm_skip_emulated_instruction(vcpu);
5147 static int handle_invlpg(struct kvm_vcpu *vcpu)
5149 unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5151 kvm_mmu_invlpg(vcpu, exit_qualification);
5152 return kvm_skip_emulated_instruction(vcpu);
5155 static int handle_rdpmc(struct kvm_vcpu *vcpu)
5159 err = kvm_rdpmc(vcpu);
5160 return kvm_complete_insn_gp(vcpu, err);
5163 static int handle_wbinvd(struct kvm_vcpu *vcpu)
5165 return kvm_emulate_wbinvd(vcpu);
5168 static int handle_xsetbv(struct kvm_vcpu *vcpu)
5170 u64 new_bv = kvm_read_edx_eax(vcpu);
5171 u32 index = kvm_rcx_read(vcpu);
5173 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
5174 return kvm_skip_emulated_instruction(vcpu);
5178 static int handle_apic_access(struct kvm_vcpu *vcpu)
5180 if (likely(fasteoi)) {
5181 unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5182 int access_type, offset;
5184 access_type = exit_qualification & APIC_ACCESS_TYPE;
5185 offset = exit_qualification & APIC_ACCESS_OFFSET;
5187 * Sane guest uses MOV to write EOI, with written value
5188 * not cared. So make a short-circuit here by avoiding
5189 * heavy instruction emulation.
5191 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5192 (offset == APIC_EOI)) {
5193 kvm_lapic_set_eoi(vcpu);
5194 return kvm_skip_emulated_instruction(vcpu);
5197 return kvm_emulate_instruction(vcpu, 0);
5200 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5202 unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5203 int vector = exit_qualification & 0xff;
5205 /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5206 kvm_apic_set_eoi_accelerated(vcpu, vector);
5210 static int handle_apic_write(struct kvm_vcpu *vcpu)
5212 unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5213 u32 offset = exit_qualification & 0xfff;
5215 /* APIC-write VM exit is trap-like and thus no need to adjust IP */
5216 kvm_apic_write_nodecode(vcpu, offset);
5220 static int handle_task_switch(struct kvm_vcpu *vcpu)
5222 struct vcpu_vmx *vmx = to_vmx(vcpu);
5223 unsigned long exit_qualification;
5224 bool has_error_code = false;
5227 int reason, type, idt_v, idt_index;
5229 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5230 idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5231 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5233 exit_qualification = vmx_get_exit_qual(vcpu);
5235 reason = (u32)exit_qualification >> 30;
5236 if (reason == TASK_SWITCH_GATE && idt_v) {
5238 case INTR_TYPE_NMI_INTR:
5239 vcpu->arch.nmi_injected = false;
5240 vmx_set_nmi_mask(vcpu, true);
5242 case INTR_TYPE_EXT_INTR:
5243 case INTR_TYPE_SOFT_INTR:
5244 kvm_clear_interrupt_queue(vcpu);
5246 case INTR_TYPE_HARD_EXCEPTION:
5247 if (vmx->idt_vectoring_info &
5248 VECTORING_INFO_DELIVER_CODE_MASK) {
5249 has_error_code = true;
5251 vmcs_read32(IDT_VECTORING_ERROR_CODE);
5254 case INTR_TYPE_SOFT_EXCEPTION:
5255 kvm_clear_exception_queue(vcpu);
5261 tss_selector = exit_qualification;
5263 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5264 type != INTR_TYPE_EXT_INTR &&
5265 type != INTR_TYPE_NMI_INTR))
5266 WARN_ON(!skip_emulated_instruction(vcpu));
5269 * TODO: What about debug traps on tss switch?
5270 * Are we supposed to inject them and update dr6?
5272 return kvm_task_switch(vcpu, tss_selector,
5273 type == INTR_TYPE_SOFT_INTR ? idt_index : -1,
5274 reason, has_error_code, error_code);
5277 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5279 unsigned long exit_qualification;
5283 exit_qualification = vmx_get_exit_qual(vcpu);
5286 * EPT violation happened while executing iret from NMI,
5287 * "blocked by NMI" bit has to be set before next VM entry.
5288 * There are errata that may cause this bit to not be set:
5291 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5293 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
5294 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5296 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5297 trace_kvm_page_fault(gpa, exit_qualification);
5299 /* Is it a read fault? */
5300 error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
5301 ? PFERR_USER_MASK : 0;
5302 /* Is it a write fault? */
5303 error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
5304 ? PFERR_WRITE_MASK : 0;
5305 /* Is it a fetch fault? */
5306 error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
5307 ? PFERR_FETCH_MASK : 0;
5308 /* ept page table entry is present? */
5309 error_code |= (exit_qualification &
5310 (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE |
5311 EPT_VIOLATION_EXECUTABLE))
5312 ? PFERR_PRESENT_MASK : 0;
5314 error_code |= (exit_qualification & 0x100) != 0 ?
5315 PFERR_GUEST_FINAL_MASK : PFERR_GUEST_PAGE_MASK;
5317 vcpu->arch.exit_qualification = exit_qualification;
5320 * Check that the GPA doesn't exceed physical memory limits, as that is
5321 * a guest page fault. We have to emulate the instruction here, because
5322 * if the illegal address is that of a paging structure, then
5323 * EPT_VIOLATION_ACC_WRITE bit is set. Alternatively, if supported we
5324 * would also use advanced VM-exit information for EPT violations to
5325 * reconstruct the page fault error code.
5327 if (unlikely(kvm_vcpu_is_illegal_gpa(vcpu, gpa)))
5328 return kvm_emulate_instruction(vcpu, 0);
5330 return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5333 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5338 * A nested guest cannot optimize MMIO vmexits, because we have an
5339 * nGPA here instead of the required GPA.
5341 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5342 if (!is_guest_mode(vcpu) &&
5343 !kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5344 trace_kvm_fast_mmio(gpa);
5345 return kvm_skip_emulated_instruction(vcpu);
5348 return kvm_mmu_page_fault(vcpu, gpa, PFERR_RSVD_MASK, NULL, 0);
5351 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5353 WARN_ON_ONCE(!enable_vnmi);
5354 exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_NMI_WINDOW_EXITING);
5355 ++vcpu->stat.nmi_window_exits;
5356 kvm_make_request(KVM_REQ_EVENT, vcpu);
5361 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5363 struct vcpu_vmx *vmx = to_vmx(vcpu);
5364 bool intr_window_requested;
5365 unsigned count = 130;
5367 intr_window_requested = exec_controls_get(vmx) &
5368 CPU_BASED_INTR_WINDOW_EXITING;
5370 while (vmx->emulation_required && count-- != 0) {
5371 if (intr_window_requested && !vmx_interrupt_blocked(vcpu))
5372 return handle_interrupt_window(&vmx->vcpu);
5374 if (kvm_test_request(KVM_REQ_EVENT, vcpu))
5377 if (!kvm_emulate_instruction(vcpu, 0))
5380 if (vmx->emulation_required && !vmx->rmode.vm86_active &&
5381 vcpu->arch.exception.pending) {
5382 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5383 vcpu->run->internal.suberror =
5384 KVM_INTERNAL_ERROR_EMULATION;
5385 vcpu->run->internal.ndata = 0;
5389 if (vcpu->arch.halt_request) {
5390 vcpu->arch.halt_request = 0;
5391 return kvm_vcpu_halt(vcpu);
5395 * Note, return 1 and not 0, vcpu_run() will invoke
5396 * xfer_to_guest_mode() which will create a proper return
5399 if (__xfer_to_guest_mode_work_pending())
5406 static void grow_ple_window(struct kvm_vcpu *vcpu)
5408 struct vcpu_vmx *vmx = to_vmx(vcpu);
5409 unsigned int old = vmx->ple_window;
5411 vmx->ple_window = __grow_ple_window(old, ple_window,
5415 if (vmx->ple_window != old) {
5416 vmx->ple_window_dirty = true;
5417 trace_kvm_ple_window_update(vcpu->vcpu_id,
5418 vmx->ple_window, old);
5422 static void shrink_ple_window(struct kvm_vcpu *vcpu)
5424 struct vcpu_vmx *vmx = to_vmx(vcpu);
5425 unsigned int old = vmx->ple_window;
5427 vmx->ple_window = __shrink_ple_window(old, ple_window,
5431 if (vmx->ple_window != old) {
5432 vmx->ple_window_dirty = true;
5433 trace_kvm_ple_window_update(vcpu->vcpu_id,
5434 vmx->ple_window, old);
5438 static void vmx_enable_tdp(void)
5440 kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK,
5441 enable_ept_ad_bits ? VMX_EPT_ACCESS_BIT : 0ull,
5442 enable_ept_ad_bits ? VMX_EPT_DIRTY_BIT : 0ull,
5443 0ull, VMX_EPT_EXECUTABLE_MASK,
5444 cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK,
5445 VMX_EPT_RWX_MASK, 0ull);
5447 ept_set_mmio_spte_mask();
5451 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5452 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5454 static int handle_pause(struct kvm_vcpu *vcpu)
5456 if (!kvm_pause_in_guest(vcpu->kvm))
5457 grow_ple_window(vcpu);
5460 * Intel sdm vol3 ch-25.1.3 says: The "PAUSE-loop exiting"
5461 * VM-execution control is ignored if CPL > 0. OTOH, KVM
5462 * never set PAUSE_EXITING and just set PLE if supported,
5463 * so the vcpu must be CPL=0 if it gets a PAUSE exit.
5465 kvm_vcpu_on_spin(vcpu, true);
5466 return kvm_skip_emulated_instruction(vcpu);
5469 static int handle_nop(struct kvm_vcpu *vcpu)
5471 return kvm_skip_emulated_instruction(vcpu);
5474 static int handle_mwait(struct kvm_vcpu *vcpu)
5476 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
5477 return handle_nop(vcpu);
5480 static int handle_invalid_op(struct kvm_vcpu *vcpu)
5482 kvm_queue_exception(vcpu, UD_VECTOR);
5486 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
5491 static int handle_monitor(struct kvm_vcpu *vcpu)
5493 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
5494 return handle_nop(vcpu);
5497 static int handle_invpcid(struct kvm_vcpu *vcpu)
5499 u32 vmx_instruction_info;
5507 if (!guest_cpuid_has(vcpu, X86_FEATURE_INVPCID)) {
5508 kvm_queue_exception(vcpu, UD_VECTOR);
5512 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5513 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
5516 kvm_inject_gp(vcpu, 0);
5520 /* According to the Intel instruction reference, the memory operand
5521 * is read even if it isn't needed (e.g., for type==all)
5523 if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5524 vmx_instruction_info, false,
5525 sizeof(operand), &gva))
5528 return kvm_handle_invpcid(vcpu, type, gva);
5531 static int handle_pml_full(struct kvm_vcpu *vcpu)
5533 unsigned long exit_qualification;
5535 trace_kvm_pml_full(vcpu->vcpu_id);
5537 exit_qualification = vmx_get_exit_qual(vcpu);
5540 * PML buffer FULL happened while executing iret from NMI,
5541 * "blocked by NMI" bit has to be set before next VM entry.
5543 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5545 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
5546 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5547 GUEST_INTR_STATE_NMI);
5550 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
5551 * here.., and there's no userspace involvement needed for PML.
5556 static fastpath_t handle_fastpath_preemption_timer(struct kvm_vcpu *vcpu)
5558 struct vcpu_vmx *vmx = to_vmx(vcpu);
5560 if (!vmx->req_immediate_exit &&
5561 !unlikely(vmx->loaded_vmcs->hv_timer_soft_disabled)) {
5562 kvm_lapic_expired_hv_timer(vcpu);
5563 return EXIT_FASTPATH_REENTER_GUEST;
5566 return EXIT_FASTPATH_NONE;
5569 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
5571 handle_fastpath_preemption_timer(vcpu);
5576 * When nested=0, all VMX instruction VM Exits filter here. The handlers
5577 * are overwritten by nested_vmx_setup() when nested=1.
5579 static int handle_vmx_instruction(struct kvm_vcpu *vcpu)
5581 kvm_queue_exception(vcpu, UD_VECTOR);
5585 static int handle_encls(struct kvm_vcpu *vcpu)
5588 * SGX virtualization is not yet supported. There is no software
5589 * enable bit for SGX, so we have to trap ENCLS and inject a #UD
5590 * to prevent the guest from executing ENCLS.
5592 kvm_queue_exception(vcpu, UD_VECTOR);
5597 * The exit handlers return 1 if the exit was handled fully and guest execution
5598 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
5599 * to be done to userspace and return 0.
5601 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
5602 [EXIT_REASON_EXCEPTION_NMI] = handle_exception_nmi,
5603 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
5604 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
5605 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
5606 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
5607 [EXIT_REASON_CR_ACCESS] = handle_cr,
5608 [EXIT_REASON_DR_ACCESS] = handle_dr,
5609 [EXIT_REASON_CPUID] = kvm_emulate_cpuid,
5610 [EXIT_REASON_MSR_READ] = kvm_emulate_rdmsr,
5611 [EXIT_REASON_MSR_WRITE] = kvm_emulate_wrmsr,
5612 [EXIT_REASON_INTERRUPT_WINDOW] = handle_interrupt_window,
5613 [EXIT_REASON_HLT] = kvm_emulate_halt,
5614 [EXIT_REASON_INVD] = handle_invd,
5615 [EXIT_REASON_INVLPG] = handle_invlpg,
5616 [EXIT_REASON_RDPMC] = handle_rdpmc,
5617 [EXIT_REASON_VMCALL] = handle_vmcall,
5618 [EXIT_REASON_VMCLEAR] = handle_vmx_instruction,
5619 [EXIT_REASON_VMLAUNCH] = handle_vmx_instruction,
5620 [EXIT_REASON_VMPTRLD] = handle_vmx_instruction,
5621 [EXIT_REASON_VMPTRST] = handle_vmx_instruction,
5622 [EXIT_REASON_VMREAD] = handle_vmx_instruction,
5623 [EXIT_REASON_VMRESUME] = handle_vmx_instruction,
5624 [EXIT_REASON_VMWRITE] = handle_vmx_instruction,
5625 [EXIT_REASON_VMOFF] = handle_vmx_instruction,
5626 [EXIT_REASON_VMON] = handle_vmx_instruction,
5627 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
5628 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
5629 [EXIT_REASON_APIC_WRITE] = handle_apic_write,
5630 [EXIT_REASON_EOI_INDUCED] = handle_apic_eoi_induced,
5631 [EXIT_REASON_WBINVD] = handle_wbinvd,
5632 [EXIT_REASON_XSETBV] = handle_xsetbv,
5633 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
5634 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
5635 [EXIT_REASON_GDTR_IDTR] = handle_desc,
5636 [EXIT_REASON_LDTR_TR] = handle_desc,
5637 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
5638 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
5639 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
5640 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_mwait,
5641 [EXIT_REASON_MONITOR_TRAP_FLAG] = handle_monitor_trap,
5642 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_monitor,
5643 [EXIT_REASON_INVEPT] = handle_vmx_instruction,
5644 [EXIT_REASON_INVVPID] = handle_vmx_instruction,
5645 [EXIT_REASON_RDRAND] = handle_invalid_op,
5646 [EXIT_REASON_RDSEED] = handle_invalid_op,
5647 [EXIT_REASON_PML_FULL] = handle_pml_full,
5648 [EXIT_REASON_INVPCID] = handle_invpcid,
5649 [EXIT_REASON_VMFUNC] = handle_vmx_instruction,
5650 [EXIT_REASON_PREEMPTION_TIMER] = handle_preemption_timer,
5651 [EXIT_REASON_ENCLS] = handle_encls,
5654 static const int kvm_vmx_max_exit_handlers =
5655 ARRAY_SIZE(kvm_vmx_exit_handlers);
5657 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2,
5658 u32 *intr_info, u32 *error_code)
5660 struct vcpu_vmx *vmx = to_vmx(vcpu);
5662 *info1 = vmx_get_exit_qual(vcpu);
5663 if (!(vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
5664 *info2 = vmx->idt_vectoring_info;
5665 *intr_info = vmx_get_intr_info(vcpu);
5666 if (is_exception_with_error_code(*intr_info))
5667 *error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
5677 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
5680 __free_page(vmx->pml_pg);
5685 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
5687 struct vcpu_vmx *vmx = to_vmx(vcpu);
5691 pml_idx = vmcs_read16(GUEST_PML_INDEX);
5693 /* Do nothing if PML buffer is empty */
5694 if (pml_idx == (PML_ENTITY_NUM - 1))
5697 /* PML index always points to next available PML buffer entity */
5698 if (pml_idx >= PML_ENTITY_NUM)
5703 pml_buf = page_address(vmx->pml_pg);
5704 for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
5707 gpa = pml_buf[pml_idx];
5708 WARN_ON(gpa & (PAGE_SIZE - 1));
5709 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5712 /* reset PML index */
5713 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
5717 * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
5718 * Called before reporting dirty_bitmap to userspace.
5720 static void kvm_flush_pml_buffers(struct kvm *kvm)
5723 struct kvm_vcpu *vcpu;
5725 * We only need to kick vcpu out of guest mode here, as PML buffer
5726 * is flushed at beginning of all VMEXITs, and it's obvious that only
5727 * vcpus running in guest are possible to have unflushed GPAs in PML
5730 kvm_for_each_vcpu(i, vcpu, kvm)
5731 kvm_vcpu_kick(vcpu);
5734 static void vmx_dump_sel(char *name, uint32_t sel)
5736 pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
5737 name, vmcs_read16(sel),
5738 vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
5739 vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
5740 vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
5743 static void vmx_dump_dtsel(char *name, uint32_t limit)
5745 pr_err("%s limit=0x%08x, base=0x%016lx\n",
5746 name, vmcs_read32(limit),
5747 vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
5750 void dump_vmcs(void)
5752 u32 vmentry_ctl, vmexit_ctl;
5753 u32 cpu_based_exec_ctrl, pin_based_exec_ctrl, secondary_exec_control;
5757 if (!dump_invalid_vmcs) {
5758 pr_warn_ratelimited("set kvm_intel.dump_invalid_vmcs=1 to dump internal KVM state.\n");
5762 vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
5763 vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
5764 cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5765 pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
5766 cr4 = vmcs_readl(GUEST_CR4);
5767 efer = vmcs_read64(GUEST_IA32_EFER);
5768 secondary_exec_control = 0;
5769 if (cpu_has_secondary_exec_ctrls())
5770 secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
5772 pr_err("*** Guest State ***\n");
5773 pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5774 vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
5775 vmcs_readl(CR0_GUEST_HOST_MASK));
5776 pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5777 cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
5778 pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
5779 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
5780 (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
5782 pr_err("PDPTR0 = 0x%016llx PDPTR1 = 0x%016llx\n",
5783 vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
5784 pr_err("PDPTR2 = 0x%016llx PDPTR3 = 0x%016llx\n",
5785 vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
5787 pr_err("RSP = 0x%016lx RIP = 0x%016lx\n",
5788 vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
5789 pr_err("RFLAGS=0x%08lx DR7 = 0x%016lx\n",
5790 vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
5791 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5792 vmcs_readl(GUEST_SYSENTER_ESP),
5793 vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
5794 vmx_dump_sel("CS: ", GUEST_CS_SELECTOR);
5795 vmx_dump_sel("DS: ", GUEST_DS_SELECTOR);
5796 vmx_dump_sel("SS: ", GUEST_SS_SELECTOR);
5797 vmx_dump_sel("ES: ", GUEST_ES_SELECTOR);
5798 vmx_dump_sel("FS: ", GUEST_FS_SELECTOR);
5799 vmx_dump_sel("GS: ", GUEST_GS_SELECTOR);
5800 vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
5801 vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
5802 vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
5803 vmx_dump_sel("TR: ", GUEST_TR_SELECTOR);
5804 if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
5805 (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
5806 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
5807 efer, vmcs_read64(GUEST_IA32_PAT));
5808 pr_err("DebugCtl = 0x%016llx DebugExceptions = 0x%016lx\n",
5809 vmcs_read64(GUEST_IA32_DEBUGCTL),
5810 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
5811 if (cpu_has_load_perf_global_ctrl() &&
5812 vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
5813 pr_err("PerfGlobCtl = 0x%016llx\n",
5814 vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
5815 if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
5816 pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
5817 pr_err("Interruptibility = %08x ActivityState = %08x\n",
5818 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
5819 vmcs_read32(GUEST_ACTIVITY_STATE));
5820 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
5821 pr_err("InterruptStatus = %04x\n",
5822 vmcs_read16(GUEST_INTR_STATUS));
5824 pr_err("*** Host State ***\n");
5825 pr_err("RIP = 0x%016lx RSP = 0x%016lx\n",
5826 vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
5827 pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
5828 vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
5829 vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
5830 vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
5831 vmcs_read16(HOST_TR_SELECTOR));
5832 pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
5833 vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
5834 vmcs_readl(HOST_TR_BASE));
5835 pr_err("GDTBase=%016lx IDTBase=%016lx\n",
5836 vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
5837 pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
5838 vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
5839 vmcs_readl(HOST_CR4));
5840 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5841 vmcs_readl(HOST_IA32_SYSENTER_ESP),
5842 vmcs_read32(HOST_IA32_SYSENTER_CS),
5843 vmcs_readl(HOST_IA32_SYSENTER_EIP));
5844 if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
5845 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
5846 vmcs_read64(HOST_IA32_EFER),
5847 vmcs_read64(HOST_IA32_PAT));
5848 if (cpu_has_load_perf_global_ctrl() &&
5849 vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
5850 pr_err("PerfGlobCtl = 0x%016llx\n",
5851 vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
5853 pr_err("*** Control State ***\n");
5854 pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
5855 pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
5856 pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
5857 pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
5858 vmcs_read32(EXCEPTION_BITMAP),
5859 vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
5860 vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
5861 pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
5862 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
5863 vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
5864 vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
5865 pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
5866 vmcs_read32(VM_EXIT_INTR_INFO),
5867 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
5868 vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
5869 pr_err(" reason=%08x qualification=%016lx\n",
5870 vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
5871 pr_err("IDTVectoring: info=%08x errcode=%08x\n",
5872 vmcs_read32(IDT_VECTORING_INFO_FIELD),
5873 vmcs_read32(IDT_VECTORING_ERROR_CODE));
5874 pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
5875 if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
5876 pr_err("TSC Multiplier = 0x%016llx\n",
5877 vmcs_read64(TSC_MULTIPLIER));
5878 if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW) {
5879 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
5880 u16 status = vmcs_read16(GUEST_INTR_STATUS);
5881 pr_err("SVI|RVI = %02x|%02x ", status >> 8, status & 0xff);
5883 pr_cont("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
5884 if (secondary_exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
5885 pr_err("APIC-access addr = 0x%016llx ", vmcs_read64(APIC_ACCESS_ADDR));
5886 pr_cont("virt-APIC addr = 0x%016llx\n", vmcs_read64(VIRTUAL_APIC_PAGE_ADDR));
5888 if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
5889 pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
5890 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
5891 pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
5892 if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
5893 pr_err("PLE Gap=%08x Window=%08x\n",
5894 vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
5895 if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
5896 pr_err("Virtual processor ID = 0x%04x\n",
5897 vmcs_read16(VIRTUAL_PROCESSOR_ID));
5901 * The guest has exited. See if we can fix it or if we need userspace
5904 static int vmx_handle_exit(struct kvm_vcpu *vcpu, fastpath_t exit_fastpath)
5906 struct vcpu_vmx *vmx = to_vmx(vcpu);
5907 u32 exit_reason = vmx->exit_reason;
5908 u32 vectoring_info = vmx->idt_vectoring_info;
5911 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
5912 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
5913 * querying dirty_bitmap, we only need to kick all vcpus out of guest
5914 * mode as if vcpus is in root mode, the PML buffer must has been
5918 vmx_flush_pml_buffer(vcpu);
5921 * We should never reach this point with a pending nested VM-Enter, and
5922 * more specifically emulation of L2 due to invalid guest state (see
5923 * below) should never happen as that means we incorrectly allowed a
5924 * nested VM-Enter with an invalid vmcs12.
5926 WARN_ON_ONCE(vmx->nested.nested_run_pending);
5928 /* If guest state is invalid, start emulating */
5929 if (vmx->emulation_required)
5930 return handle_invalid_guest_state(vcpu);
5932 if (is_guest_mode(vcpu)) {
5934 * The host physical addresses of some pages of guest memory
5935 * are loaded into the vmcs02 (e.g. vmcs12's Virtual APIC
5936 * Page). The CPU may write to these pages via their host
5937 * physical address while L2 is running, bypassing any
5938 * address-translation-based dirty tracking (e.g. EPT write
5941 * Mark them dirty on every exit from L2 to prevent them from
5942 * getting out of sync with dirty tracking.
5944 nested_mark_vmcs12_pages_dirty(vcpu);
5946 if (nested_vmx_reflect_vmexit(vcpu))
5950 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
5952 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5953 vcpu->run->fail_entry.hardware_entry_failure_reason
5955 vcpu->run->fail_entry.cpu = vcpu->arch.last_vmentry_cpu;
5959 if (unlikely(vmx->fail)) {
5961 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5962 vcpu->run->fail_entry.hardware_entry_failure_reason
5963 = vmcs_read32(VM_INSTRUCTION_ERROR);
5964 vcpu->run->fail_entry.cpu = vcpu->arch.last_vmentry_cpu;
5970 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
5971 * delivery event since it indicates guest is accessing MMIO.
5972 * The vm-exit can be triggered again after return to guest that
5973 * will cause infinite loop.
5975 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
5976 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
5977 exit_reason != EXIT_REASON_EPT_VIOLATION &&
5978 exit_reason != EXIT_REASON_PML_FULL &&
5979 exit_reason != EXIT_REASON_APIC_ACCESS &&
5980 exit_reason != EXIT_REASON_TASK_SWITCH)) {
5981 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5982 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
5983 vcpu->run->internal.ndata = 3;
5984 vcpu->run->internal.data[0] = vectoring_info;
5985 vcpu->run->internal.data[1] = exit_reason;
5986 vcpu->run->internal.data[2] = vcpu->arch.exit_qualification;
5987 if (exit_reason == EXIT_REASON_EPT_MISCONFIG) {
5988 vcpu->run->internal.ndata++;
5989 vcpu->run->internal.data[3] =
5990 vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5992 vcpu->run->internal.data[vcpu->run->internal.ndata++] =
5993 vcpu->arch.last_vmentry_cpu;
5997 if (unlikely(!enable_vnmi &&
5998 vmx->loaded_vmcs->soft_vnmi_blocked)) {
5999 if (!vmx_interrupt_blocked(vcpu)) {
6000 vmx->loaded_vmcs->soft_vnmi_blocked = 0;
6001 } else if (vmx->loaded_vmcs->vnmi_blocked_time > 1000000000LL &&
6002 vcpu->arch.nmi_pending) {
6004 * This CPU don't support us in finding the end of an
6005 * NMI-blocked window if the guest runs with IRQs
6006 * disabled. So we pull the trigger after 1 s of
6007 * futile waiting, but inform the user about this.
6009 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
6010 "state on VCPU %d after 1 s timeout\n",
6011 __func__, vcpu->vcpu_id);
6012 vmx->loaded_vmcs->soft_vnmi_blocked = 0;
6016 if (exit_fastpath != EXIT_FASTPATH_NONE)
6019 if (exit_reason >= kvm_vmx_max_exit_handlers)
6020 goto unexpected_vmexit;
6021 #ifdef CONFIG_RETPOLINE
6022 if (exit_reason == EXIT_REASON_MSR_WRITE)
6023 return kvm_emulate_wrmsr(vcpu);
6024 else if (exit_reason == EXIT_REASON_PREEMPTION_TIMER)
6025 return handle_preemption_timer(vcpu);
6026 else if (exit_reason == EXIT_REASON_INTERRUPT_WINDOW)
6027 return handle_interrupt_window(vcpu);
6028 else if (exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
6029 return handle_external_interrupt(vcpu);
6030 else if (exit_reason == EXIT_REASON_HLT)
6031 return kvm_emulate_halt(vcpu);
6032 else if (exit_reason == EXIT_REASON_EPT_MISCONFIG)
6033 return handle_ept_misconfig(vcpu);
6036 exit_reason = array_index_nospec(exit_reason,
6037 kvm_vmx_max_exit_handlers);
6038 if (!kvm_vmx_exit_handlers[exit_reason])
6039 goto unexpected_vmexit;
6041 return kvm_vmx_exit_handlers[exit_reason](vcpu);
6044 vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n", exit_reason);
6046 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6047 vcpu->run->internal.suberror =
6048 KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON;
6049 vcpu->run->internal.ndata = 2;
6050 vcpu->run->internal.data[0] = exit_reason;
6051 vcpu->run->internal.data[1] = vcpu->arch.last_vmentry_cpu;
6056 * Software based L1D cache flush which is used when microcode providing
6057 * the cache control MSR is not loaded.
6059 * The L1D cache is 32 KiB on Nehalem and later microarchitectures, but to
6060 * flush it is required to read in 64 KiB because the replacement algorithm
6061 * is not exactly LRU. This could be sized at runtime via topology
6062 * information but as all relevant affected CPUs have 32KiB L1D cache size
6063 * there is no point in doing so.
6065 static noinstr void vmx_l1d_flush(struct kvm_vcpu *vcpu)
6067 int size = PAGE_SIZE << L1D_CACHE_ORDER;
6070 * This code is only executed when the the flush mode is 'cond' or
6073 if (static_branch_likely(&vmx_l1d_flush_cond)) {
6077 * Clear the per-vcpu flush bit, it gets set again
6078 * either from vcpu_run() or from one of the unsafe
6081 flush_l1d = vcpu->arch.l1tf_flush_l1d;
6082 vcpu->arch.l1tf_flush_l1d = false;
6085 * Clear the per-cpu flush bit, it gets set again from
6086 * the interrupt handlers.
6088 flush_l1d |= kvm_get_cpu_l1tf_flush_l1d();
6089 kvm_clear_cpu_l1tf_flush_l1d();
6095 vcpu->stat.l1d_flush++;
6097 if (static_cpu_has(X86_FEATURE_FLUSH_L1D)) {
6098 native_wrmsrl(MSR_IA32_FLUSH_CMD, L1D_FLUSH);
6103 /* First ensure the pages are in the TLB */
6104 "xorl %%eax, %%eax\n"
6105 ".Lpopulate_tlb:\n\t"
6106 "movzbl (%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
6107 "addl $4096, %%eax\n\t"
6108 "cmpl %%eax, %[size]\n\t"
6109 "jne .Lpopulate_tlb\n\t"
6110 "xorl %%eax, %%eax\n\t"
6112 /* Now fill the cache */
6113 "xorl %%eax, %%eax\n"
6115 "movzbl (%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
6116 "addl $64, %%eax\n\t"
6117 "cmpl %%eax, %[size]\n\t"
6118 "jne .Lfill_cache\n\t"
6120 :: [flush_pages] "r" (vmx_l1d_flush_pages),
6122 : "eax", "ebx", "ecx", "edx");
6125 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
6127 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6130 if (is_guest_mode(vcpu) &&
6131 nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
6134 tpr_threshold = (irr == -1 || tpr < irr) ? 0 : irr;
6135 if (is_guest_mode(vcpu))
6136 to_vmx(vcpu)->nested.l1_tpr_threshold = tpr_threshold;
6138 vmcs_write32(TPR_THRESHOLD, tpr_threshold);
6141 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu)
6143 struct vcpu_vmx *vmx = to_vmx(vcpu);
6144 u32 sec_exec_control;
6146 if (!lapic_in_kernel(vcpu))
6149 if (!flexpriority_enabled &&
6150 !cpu_has_vmx_virtualize_x2apic_mode())
6153 /* Postpone execution until vmcs01 is the current VMCS. */
6154 if (is_guest_mode(vcpu)) {
6155 vmx->nested.change_vmcs01_virtual_apic_mode = true;
6159 sec_exec_control = secondary_exec_controls_get(vmx);
6160 sec_exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6161 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
6163 switch (kvm_get_apic_mode(vcpu)) {
6164 case LAPIC_MODE_INVALID:
6165 WARN_ONCE(true, "Invalid local APIC state");
6166 case LAPIC_MODE_DISABLED:
6168 case LAPIC_MODE_XAPIC:
6169 if (flexpriority_enabled) {
6171 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6172 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
6175 * Flush the TLB, reloading the APIC access page will
6176 * only do so if its physical address has changed, but
6177 * the guest may have inserted a non-APIC mapping into
6178 * the TLB while the APIC access page was disabled.
6180 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
6183 case LAPIC_MODE_X2APIC:
6184 if (cpu_has_vmx_virtualize_x2apic_mode())
6186 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
6189 secondary_exec_controls_set(vmx, sec_exec_control);
6191 vmx_update_msr_bitmap(vcpu);
6194 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu)
6198 /* Defer reload until vmcs01 is the current VMCS. */
6199 if (is_guest_mode(vcpu)) {
6200 to_vmx(vcpu)->nested.reload_vmcs01_apic_access_page = true;
6204 if (!(secondary_exec_controls_get(to_vmx(vcpu)) &
6205 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
6208 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6209 if (is_error_page(page))
6212 vmcs_write64(APIC_ACCESS_ADDR, page_to_phys(page));
6213 vmx_flush_tlb_current(vcpu);
6216 * Do not pin apic access page in memory, the MMU notifier
6217 * will call us again if it is migrated or swapped out.
6222 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
6230 status = vmcs_read16(GUEST_INTR_STATUS);
6232 if (max_isr != old) {
6234 status |= max_isr << 8;
6235 vmcs_write16(GUEST_INTR_STATUS, status);
6239 static void vmx_set_rvi(int vector)
6247 status = vmcs_read16(GUEST_INTR_STATUS);
6248 old = (u8)status & 0xff;
6249 if ((u8)vector != old) {
6251 status |= (u8)vector;
6252 vmcs_write16(GUEST_INTR_STATUS, status);
6256 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
6259 * When running L2, updating RVI is only relevant when
6260 * vmcs12 virtual-interrupt-delivery enabled.
6261 * However, it can be enabled only when L1 also
6262 * intercepts external-interrupts and in that case
6263 * we should not update vmcs02 RVI but instead intercept
6264 * interrupt. Therefore, do nothing when running L2.
6266 if (!is_guest_mode(vcpu))
6267 vmx_set_rvi(max_irr);
6270 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
6272 struct vcpu_vmx *vmx = to_vmx(vcpu);
6274 bool max_irr_updated;
6276 WARN_ON(!vcpu->arch.apicv_active);
6277 if (pi_test_on(&vmx->pi_desc)) {
6278 pi_clear_on(&vmx->pi_desc);
6280 * IOMMU can write to PID.ON, so the barrier matters even on UP.
6281 * But on x86 this is just a compiler barrier anyway.
6283 smp_mb__after_atomic();
6285 kvm_apic_update_irr(vcpu, vmx->pi_desc.pir, &max_irr);
6288 * If we are running L2 and L1 has a new pending interrupt
6289 * which can be injected, we should re-evaluate
6290 * what should be done with this new L1 interrupt.
6291 * If L1 intercepts external-interrupts, we should
6292 * exit from L2 to L1. Otherwise, interrupt should be
6293 * delivered directly to L2.
6295 if (is_guest_mode(vcpu) && max_irr_updated) {
6296 if (nested_exit_on_intr(vcpu))
6297 kvm_vcpu_exiting_guest_mode(vcpu);
6299 kvm_make_request(KVM_REQ_EVENT, vcpu);
6302 max_irr = kvm_lapic_find_highest_irr(vcpu);
6304 vmx_hwapic_irr_update(vcpu, max_irr);
6308 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
6310 if (!kvm_vcpu_apicv_active(vcpu))
6313 vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
6314 vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
6315 vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
6316 vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
6319 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
6321 struct vcpu_vmx *vmx = to_vmx(vcpu);
6323 pi_clear_on(&vmx->pi_desc);
6324 memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
6327 void vmx_do_interrupt_nmi_irqoff(unsigned long entry);
6329 static void handle_interrupt_nmi_irqoff(struct kvm_vcpu *vcpu, u32 intr_info)
6331 unsigned int vector = intr_info & INTR_INFO_VECTOR_MASK;
6332 gate_desc *desc = (gate_desc *)host_idt_base + vector;
6334 kvm_before_interrupt(vcpu);
6335 vmx_do_interrupt_nmi_irqoff(gate_offset(desc));
6336 kvm_after_interrupt(vcpu);
6339 static void handle_exception_nmi_irqoff(struct vcpu_vmx *vmx)
6341 u32 intr_info = vmx_get_intr_info(&vmx->vcpu);
6343 /* if exit due to PF check for async PF */
6344 if (is_page_fault(intr_info))
6345 vmx->vcpu.arch.apf.host_apf_flags = kvm_read_and_reset_apf_flags();
6346 /* Handle machine checks before interrupts are enabled */
6347 else if (is_machine_check(intr_info))
6348 kvm_machine_check();
6349 /* We need to handle NMIs before interrupts are enabled */
6350 else if (is_nmi(intr_info))
6351 handle_interrupt_nmi_irqoff(&vmx->vcpu, intr_info);
6354 static void handle_external_interrupt_irqoff(struct kvm_vcpu *vcpu)
6356 u32 intr_info = vmx_get_intr_info(vcpu);
6358 if (WARN_ONCE(!is_external_intr(intr_info),
6359 "KVM: unexpected VM-Exit interrupt info: 0x%x", intr_info))
6362 handle_interrupt_nmi_irqoff(vcpu, intr_info);
6365 static void vmx_handle_exit_irqoff(struct kvm_vcpu *vcpu)
6367 struct vcpu_vmx *vmx = to_vmx(vcpu);
6369 if (vmx->exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
6370 handle_external_interrupt_irqoff(vcpu);
6371 else if (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI)
6372 handle_exception_nmi_irqoff(vmx);
6375 static bool vmx_has_emulated_msr(u32 index)
6378 case MSR_IA32_SMBASE:
6380 * We cannot do SMM unless we can run the guest in big
6383 return enable_unrestricted_guest || emulate_invalid_guest_state;
6384 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
6386 case MSR_AMD64_VIRT_SPEC_CTRL:
6387 /* This is AMD only. */
6394 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
6399 bool idtv_info_valid;
6401 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6404 if (vmx->loaded_vmcs->nmi_known_unmasked)
6407 exit_intr_info = vmx_get_intr_info(&vmx->vcpu);
6408 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
6409 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
6411 * SDM 3: 27.7.1.2 (September 2008)
6412 * Re-set bit "block by NMI" before VM entry if vmexit caused by
6413 * a guest IRET fault.
6414 * SDM 3: 23.2.2 (September 2008)
6415 * Bit 12 is undefined in any of the following cases:
6416 * If the VM exit sets the valid bit in the IDT-vectoring
6417 * information field.
6418 * If the VM exit is due to a double fault.
6420 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
6421 vector != DF_VECTOR && !idtv_info_valid)
6422 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
6423 GUEST_INTR_STATE_NMI);
6425 vmx->loaded_vmcs->nmi_known_unmasked =
6426 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
6427 & GUEST_INTR_STATE_NMI);
6428 } else if (unlikely(vmx->loaded_vmcs->soft_vnmi_blocked))
6429 vmx->loaded_vmcs->vnmi_blocked_time +=
6430 ktime_to_ns(ktime_sub(ktime_get(),
6431 vmx->loaded_vmcs->entry_time));
6434 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
6435 u32 idt_vectoring_info,
6436 int instr_len_field,
6437 int error_code_field)
6441 bool idtv_info_valid;
6443 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6445 vcpu->arch.nmi_injected = false;
6446 kvm_clear_exception_queue(vcpu);
6447 kvm_clear_interrupt_queue(vcpu);
6449 if (!idtv_info_valid)
6452 kvm_make_request(KVM_REQ_EVENT, vcpu);
6454 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
6455 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
6458 case INTR_TYPE_NMI_INTR:
6459 vcpu->arch.nmi_injected = true;
6461 * SDM 3: 27.7.1.2 (September 2008)
6462 * Clear bit "block by NMI" before VM entry if a NMI
6465 vmx_set_nmi_mask(vcpu, false);
6467 case INTR_TYPE_SOFT_EXCEPTION:
6468 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6470 case INTR_TYPE_HARD_EXCEPTION:
6471 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
6472 u32 err = vmcs_read32(error_code_field);
6473 kvm_requeue_exception_e(vcpu, vector, err);
6475 kvm_requeue_exception(vcpu, vector);
6477 case INTR_TYPE_SOFT_INTR:
6478 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6480 case INTR_TYPE_EXT_INTR:
6481 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
6488 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
6490 __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
6491 VM_EXIT_INSTRUCTION_LEN,
6492 IDT_VECTORING_ERROR_CODE);
6495 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
6497 __vmx_complete_interrupts(vcpu,
6498 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
6499 VM_ENTRY_INSTRUCTION_LEN,
6500 VM_ENTRY_EXCEPTION_ERROR_CODE);
6502 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
6505 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
6508 struct perf_guest_switch_msr *msrs;
6510 msrs = perf_guest_get_msrs(&nr_msrs);
6515 for (i = 0; i < nr_msrs; i++)
6516 if (msrs[i].host == msrs[i].guest)
6517 clear_atomic_switch_msr(vmx, msrs[i].msr);
6519 add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
6520 msrs[i].host, false);
6523 static void vmx_update_hv_timer(struct kvm_vcpu *vcpu)
6525 struct vcpu_vmx *vmx = to_vmx(vcpu);
6529 if (vmx->req_immediate_exit) {
6530 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, 0);
6531 vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6532 } else if (vmx->hv_deadline_tsc != -1) {
6534 if (vmx->hv_deadline_tsc > tscl)
6535 /* set_hv_timer ensures the delta fits in 32-bits */
6536 delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
6537 cpu_preemption_timer_multi);
6541 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
6542 vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6543 } else if (!vmx->loaded_vmcs->hv_timer_soft_disabled) {
6544 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, -1);
6545 vmx->loaded_vmcs->hv_timer_soft_disabled = true;
6549 void noinstr vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp)
6551 if (unlikely(host_rsp != vmx->loaded_vmcs->host_state.rsp)) {
6552 vmx->loaded_vmcs->host_state.rsp = host_rsp;
6553 vmcs_writel(HOST_RSP, host_rsp);
6557 static fastpath_t vmx_exit_handlers_fastpath(struct kvm_vcpu *vcpu)
6559 switch (to_vmx(vcpu)->exit_reason) {
6560 case EXIT_REASON_MSR_WRITE:
6561 return handle_fastpath_set_msr_irqoff(vcpu);
6562 case EXIT_REASON_PREEMPTION_TIMER:
6563 return handle_fastpath_preemption_timer(vcpu);
6565 return EXIT_FASTPATH_NONE;
6569 bool __vmx_vcpu_run(struct vcpu_vmx *vmx, unsigned long *regs, bool launched);
6571 static noinstr void vmx_vcpu_enter_exit(struct kvm_vcpu *vcpu,
6572 struct vcpu_vmx *vmx)
6575 * VMENTER enables interrupts (host state), but the kernel state is
6576 * interrupts disabled when this is invoked. Also tell RCU about
6577 * it. This is the same logic as for exit_to_user_mode().
6579 * This ensures that e.g. latency analysis on the host observes
6580 * guest mode as interrupt enabled.
6582 * guest_enter_irqoff() informs context tracking about the
6583 * transition to guest mode and if enabled adjusts RCU state
6586 instrumentation_begin();
6587 trace_hardirqs_on_prepare();
6588 lockdep_hardirqs_on_prepare(CALLER_ADDR0);
6589 instrumentation_end();
6591 guest_enter_irqoff();
6592 lockdep_hardirqs_on(CALLER_ADDR0);
6594 /* L1D Flush includes CPU buffer clear to mitigate MDS */
6595 if (static_branch_unlikely(&vmx_l1d_should_flush))
6596 vmx_l1d_flush(vcpu);
6597 else if (static_branch_unlikely(&mds_user_clear))
6598 mds_clear_cpu_buffers();
6600 if (vcpu->arch.cr2 != native_read_cr2())
6601 native_write_cr2(vcpu->arch.cr2);
6603 vmx->fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
6604 vmx->loaded_vmcs->launched);
6606 vcpu->arch.cr2 = native_read_cr2();
6609 * VMEXIT disables interrupts (host state), but tracing and lockdep
6610 * have them in state 'on' as recorded before entering guest mode.
6611 * Same as enter_from_user_mode().
6613 * guest_exit_irqoff() restores host context and reinstates RCU if
6614 * enabled and required.
6616 * This needs to be done before the below as native_read_msr()
6617 * contains a tracepoint and x86_spec_ctrl_restore_host() calls
6618 * into world and some more.
6620 lockdep_hardirqs_off(CALLER_ADDR0);
6621 guest_exit_irqoff();
6623 instrumentation_begin();
6624 trace_hardirqs_off_finish();
6625 instrumentation_end();
6628 static fastpath_t vmx_vcpu_run(struct kvm_vcpu *vcpu)
6630 fastpath_t exit_fastpath;
6631 struct vcpu_vmx *vmx = to_vmx(vcpu);
6632 unsigned long cr3, cr4;
6635 /* Record the guest's net vcpu time for enforced NMI injections. */
6636 if (unlikely(!enable_vnmi &&
6637 vmx->loaded_vmcs->soft_vnmi_blocked))
6638 vmx->loaded_vmcs->entry_time = ktime_get();
6640 /* Don't enter VMX if guest state is invalid, let the exit handler
6641 start emulation until we arrive back to a valid state */
6642 if (vmx->emulation_required)
6643 return EXIT_FASTPATH_NONE;
6645 if (vmx->ple_window_dirty) {
6646 vmx->ple_window_dirty = false;
6647 vmcs_write32(PLE_WINDOW, vmx->ple_window);
6651 * We did this in prepare_switch_to_guest, because it needs to
6652 * be within srcu_read_lock.
6654 WARN_ON_ONCE(vmx->nested.need_vmcs12_to_shadow_sync);
6656 if (kvm_register_is_dirty(vcpu, VCPU_REGS_RSP))
6657 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
6658 if (kvm_register_is_dirty(vcpu, VCPU_REGS_RIP))
6659 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
6661 cr3 = __get_current_cr3_fast();
6662 if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
6663 vmcs_writel(HOST_CR3, cr3);
6664 vmx->loaded_vmcs->host_state.cr3 = cr3;
6667 cr4 = cr4_read_shadow();
6668 if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
6669 vmcs_writel(HOST_CR4, cr4);
6670 vmx->loaded_vmcs->host_state.cr4 = cr4;
6673 /* When single-stepping over STI and MOV SS, we must clear the
6674 * corresponding interruptibility bits in the guest state. Otherwise
6675 * vmentry fails as it then expects bit 14 (BS) in pending debug
6676 * exceptions being set, but that's not correct for the guest debugging
6678 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6679 vmx_set_interrupt_shadow(vcpu, 0);
6681 kvm_load_guest_xsave_state(vcpu);
6683 pt_guest_enter(vmx);
6685 atomic_switch_perf_msrs(vmx);
6687 if (enable_preemption_timer)
6688 vmx_update_hv_timer(vcpu);
6690 kvm_wait_lapic_expire(vcpu);
6693 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
6694 * it's non-zero. Since vmentry is serialising on affected CPUs, there
6695 * is no need to worry about the conditional branch over the wrmsr
6696 * being speculatively taken.
6698 x86_spec_ctrl_set_guest(vmx->spec_ctrl, 0);
6700 /* The actual VMENTER/EXIT is in the .noinstr.text section. */
6701 vmx_vcpu_enter_exit(vcpu, vmx);
6704 * We do not use IBRS in the kernel. If this vCPU has used the
6705 * SPEC_CTRL MSR it may have left it on; save the value and
6706 * turn it off. This is much more efficient than blindly adding
6707 * it to the atomic save/restore list. Especially as the former
6708 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
6710 * For non-nested case:
6711 * If the L01 MSR bitmap does not intercept the MSR, then we need to
6715 * If the L02 MSR bitmap does not intercept the MSR, then we need to
6718 if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
6719 vmx->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
6721 x86_spec_ctrl_restore_host(vmx->spec_ctrl, 0);
6723 /* All fields are clean at this point */
6724 if (static_branch_unlikely(&enable_evmcs))
6725 current_evmcs->hv_clean_fields |=
6726 HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
6728 if (static_branch_unlikely(&enable_evmcs))
6729 current_evmcs->hv_vp_id = vcpu->arch.hyperv.vp_index;
6731 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
6732 if (vmx->host_debugctlmsr)
6733 update_debugctlmsr(vmx->host_debugctlmsr);
6735 #ifndef CONFIG_X86_64
6737 * The sysexit path does not restore ds/es, so we must set them to
6738 * a reasonable value ourselves.
6740 * We can't defer this to vmx_prepare_switch_to_host() since that
6741 * function may be executed in interrupt context, which saves and
6742 * restore segments around it, nullifying its effect.
6744 loadsegment(ds, __USER_DS);
6745 loadsegment(es, __USER_DS);
6748 vmx_register_cache_reset(vcpu);
6752 kvm_load_host_xsave_state(vcpu);
6754 vmx->nested.nested_run_pending = 0;
6755 vmx->idt_vectoring_info = 0;
6757 if (unlikely(vmx->fail)) {
6758 vmx->exit_reason = 0xdead;
6759 return EXIT_FASTPATH_NONE;
6762 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
6763 if (unlikely((u16)vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY))
6764 kvm_machine_check();
6766 trace_kvm_exit(vmx->exit_reason, vcpu, KVM_ISA_VMX);
6768 if (unlikely(vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
6769 return EXIT_FASTPATH_NONE;
6771 vmx->loaded_vmcs->launched = 1;
6772 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
6774 vmx_recover_nmi_blocking(vmx);
6775 vmx_complete_interrupts(vmx);
6777 if (is_guest_mode(vcpu))
6778 return EXIT_FASTPATH_NONE;
6780 exit_fastpath = vmx_exit_handlers_fastpath(vcpu);
6781 if (exit_fastpath == EXIT_FASTPATH_REENTER_GUEST) {
6782 if (!kvm_vcpu_exit_request(vcpu)) {
6784 * FIXME: this goto should be a loop in vcpu_enter_guest,
6785 * but it would incur the cost of a retpoline for now.
6786 * Revisit once static calls are available.
6788 if (vcpu->arch.apicv_active)
6789 vmx_sync_pir_to_irr(vcpu);
6792 exit_fastpath = EXIT_FASTPATH_EXIT_HANDLED;
6795 return exit_fastpath;
6798 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
6800 struct vcpu_vmx *vmx = to_vmx(vcpu);
6803 vmx_destroy_pml_buffer(vmx);
6804 free_vpid(vmx->vpid);
6805 nested_vmx_free_vcpu(vcpu);
6806 free_loaded_vmcs(vmx->loaded_vmcs);
6809 static int vmx_create_vcpu(struct kvm_vcpu *vcpu)
6811 struct vcpu_vmx *vmx;
6812 unsigned long *msr_bitmap;
6815 BUILD_BUG_ON(offsetof(struct vcpu_vmx, vcpu) != 0);
6820 vmx->vpid = allocate_vpid();
6823 * If PML is turned on, failure on enabling PML just results in failure
6824 * of creating the vcpu, therefore we can simplify PML logic (by
6825 * avoiding dealing with cases, such as enabling PML partially on vcpus
6826 * for the guest), etc.
6829 vmx->pml_pg = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
6834 BUILD_BUG_ON(ARRAY_SIZE(vmx_uret_msrs_list) != MAX_NR_USER_RETURN_MSRS);
6836 for (i = 0; i < ARRAY_SIZE(vmx_uret_msrs_list); ++i) {
6837 u32 index = vmx_uret_msrs_list[i];
6838 u32 data_low, data_high;
6839 int j = vmx->nr_uret_msrs;
6841 if (rdmsr_safe(index, &data_low, &data_high) < 0)
6843 if (wrmsr_safe(index, data_low, data_high) < 0)
6846 vmx->guest_uret_msrs[j].slot = i;
6847 vmx->guest_uret_msrs[j].data = 0;
6849 case MSR_IA32_TSX_CTRL:
6851 * No need to pass TSX_CTRL_CPUID_CLEAR through, so
6852 * let's avoid changing CPUID bits under the host
6855 vmx->guest_uret_msrs[j].mask = ~(u64)TSX_CTRL_CPUID_CLEAR;
6858 vmx->guest_uret_msrs[j].mask = -1ull;
6861 ++vmx->nr_uret_msrs;
6864 err = alloc_loaded_vmcs(&vmx->vmcs01);
6868 /* The MSR bitmap starts with all ones */
6869 bitmap_fill(vmx->shadow_msr_intercept.read, MAX_POSSIBLE_PASSTHROUGH_MSRS);
6870 bitmap_fill(vmx->shadow_msr_intercept.write, MAX_POSSIBLE_PASSTHROUGH_MSRS);
6872 msr_bitmap = vmx->vmcs01.msr_bitmap;
6873 vmx_disable_intercept_for_msr(vcpu, MSR_IA32_TSC, MSR_TYPE_R);
6874 vmx_disable_intercept_for_msr(vcpu, MSR_FS_BASE, MSR_TYPE_RW);
6875 vmx_disable_intercept_for_msr(vcpu, MSR_GS_BASE, MSR_TYPE_RW);
6876 vmx_disable_intercept_for_msr(vcpu, MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
6877 vmx_disable_intercept_for_msr(vcpu, MSR_IA32_SYSENTER_CS, MSR_TYPE_RW);
6878 vmx_disable_intercept_for_msr(vcpu, MSR_IA32_SYSENTER_ESP, MSR_TYPE_RW);
6879 vmx_disable_intercept_for_msr(vcpu, MSR_IA32_SYSENTER_EIP, MSR_TYPE_RW);
6880 if (kvm_cstate_in_guest(vcpu->kvm)) {
6881 vmx_disable_intercept_for_msr(vcpu, MSR_CORE_C1_RES, MSR_TYPE_R);
6882 vmx_disable_intercept_for_msr(vcpu, MSR_CORE_C3_RESIDENCY, MSR_TYPE_R);
6883 vmx_disable_intercept_for_msr(vcpu, MSR_CORE_C6_RESIDENCY, MSR_TYPE_R);
6884 vmx_disable_intercept_for_msr(vcpu, MSR_CORE_C7_RESIDENCY, MSR_TYPE_R);
6886 vmx->msr_bitmap_mode = 0;
6888 vmx->loaded_vmcs = &vmx->vmcs01;
6890 vmx_vcpu_load(vcpu, cpu);
6895 if (cpu_need_virtualize_apic_accesses(vcpu)) {
6896 err = alloc_apic_access_page(vcpu->kvm);
6901 if (enable_ept && !enable_unrestricted_guest) {
6902 err = init_rmode_identity_map(vcpu->kvm);
6908 memcpy(&vmx->nested.msrs, &vmcs_config.nested, sizeof(vmx->nested.msrs));
6910 memset(&vmx->nested.msrs, 0, sizeof(vmx->nested.msrs));
6912 vmx->nested.posted_intr_nv = -1;
6913 vmx->nested.current_vmptr = -1ull;
6915 vcpu->arch.microcode_version = 0x100000000ULL;
6916 vmx->msr_ia32_feature_control_valid_bits = FEAT_CTL_LOCKED;
6919 * Enforce invariant: pi_desc.nv is always either POSTED_INTR_VECTOR
6920 * or POSTED_INTR_WAKEUP_VECTOR.
6922 vmx->pi_desc.nv = POSTED_INTR_VECTOR;
6923 vmx->pi_desc.sn = 1;
6925 vmx->ept_pointer = INVALID_PAGE;
6930 free_loaded_vmcs(vmx->loaded_vmcs);
6932 vmx_destroy_pml_buffer(vmx);
6934 free_vpid(vmx->vpid);
6938 #define L1TF_MSG_SMT "L1TF CPU bug present and SMT on, data leak possible. See CVE-2018-3646 and https://www.kernel.org/doc/html/latest/admin-guide/hw-vuln/l1tf.html for details.\n"
6939 #define L1TF_MSG_L1D "L1TF CPU bug present and virtualization mitigation disabled, data leak possible. See CVE-2018-3646 and https://www.kernel.org/doc/html/latest/admin-guide/hw-vuln/l1tf.html for details.\n"
6941 static int vmx_vm_init(struct kvm *kvm)
6943 spin_lock_init(&to_kvm_vmx(kvm)->ept_pointer_lock);
6946 kvm->arch.pause_in_guest = true;
6948 if (boot_cpu_has(X86_BUG_L1TF) && enable_ept) {
6949 switch (l1tf_mitigation) {
6950 case L1TF_MITIGATION_OFF:
6951 case L1TF_MITIGATION_FLUSH_NOWARN:
6952 /* 'I explicitly don't care' is set */
6954 case L1TF_MITIGATION_FLUSH:
6955 case L1TF_MITIGATION_FLUSH_NOSMT:
6956 case L1TF_MITIGATION_FULL:
6958 * Warn upon starting the first VM in a potentially
6959 * insecure environment.
6961 if (sched_smt_active())
6962 pr_warn_once(L1TF_MSG_SMT);
6963 if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER)
6964 pr_warn_once(L1TF_MSG_L1D);
6966 case L1TF_MITIGATION_FULL_FORCE:
6967 /* Flush is enforced */
6971 kvm_apicv_init(kvm, enable_apicv);
6975 static int __init vmx_check_processor_compat(void)
6977 struct vmcs_config vmcs_conf;
6978 struct vmx_capability vmx_cap;
6980 if (!this_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
6981 !this_cpu_has(X86_FEATURE_VMX)) {
6982 pr_err("kvm: VMX is disabled on CPU %d\n", smp_processor_id());
6986 if (setup_vmcs_config(&vmcs_conf, &vmx_cap) < 0)
6989 nested_vmx_setup_ctls_msrs(&vmcs_conf.nested, vmx_cap.ept);
6990 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
6991 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
6992 smp_processor_id());
6998 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
7003 /* We wanted to honor guest CD/MTRR/PAT, but doing so could result in
7004 * memory aliases with conflicting memory types and sometimes MCEs.
7005 * We have to be careful as to what are honored and when.
7007 * For MMIO, guest CD/MTRR are ignored. The EPT memory type is set to
7008 * UC. The effective memory type is UC or WC depending on guest PAT.
7009 * This was historically the source of MCEs and we want to be
7012 * When there is no need to deal with noncoherent DMA (e.g., no VT-d
7013 * or VT-d has snoop control), guest CD/MTRR/PAT are all ignored. The
7014 * EPT memory type is set to WB. The effective memory type is forced
7017 * Otherwise, we trust guest. Guest CD/MTRR/PAT are all honored. The
7018 * EPT memory type is used to emulate guest CD/MTRR.
7022 cache = MTRR_TYPE_UNCACHABLE;
7026 if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
7027 ipat = VMX_EPT_IPAT_BIT;
7028 cache = MTRR_TYPE_WRBACK;
7032 if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
7033 ipat = VMX_EPT_IPAT_BIT;
7034 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
7035 cache = MTRR_TYPE_WRBACK;
7037 cache = MTRR_TYPE_UNCACHABLE;
7041 cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
7044 return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
7047 static void vmcs_set_secondary_exec_control(struct vcpu_vmx *vmx)
7050 * These bits in the secondary execution controls field
7051 * are dynamic, the others are mostly based on the hypervisor
7052 * architecture and the guest's CPUID. Do not touch the
7056 SECONDARY_EXEC_SHADOW_VMCS |
7057 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
7058 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
7059 SECONDARY_EXEC_DESC;
7061 u32 new_ctl = vmx->secondary_exec_control;
7062 u32 cur_ctl = secondary_exec_controls_get(vmx);
7064 secondary_exec_controls_set(vmx, (new_ctl & ~mask) | (cur_ctl & mask));
7068 * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
7069 * (indicating "allowed-1") if they are supported in the guest's CPUID.
7071 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
7073 struct vcpu_vmx *vmx = to_vmx(vcpu);
7074 struct kvm_cpuid_entry2 *entry;
7076 vmx->nested.msrs.cr0_fixed1 = 0xffffffff;
7077 vmx->nested.msrs.cr4_fixed1 = X86_CR4_PCE;
7079 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do { \
7080 if (entry && (entry->_reg & (_cpuid_mask))) \
7081 vmx->nested.msrs.cr4_fixed1 |= (_cr4_mask); \
7084 entry = kvm_find_cpuid_entry(vcpu, 0x1, 0);
7085 cr4_fixed1_update(X86_CR4_VME, edx, feature_bit(VME));
7086 cr4_fixed1_update(X86_CR4_PVI, edx, feature_bit(VME));
7087 cr4_fixed1_update(X86_CR4_TSD, edx, feature_bit(TSC));
7088 cr4_fixed1_update(X86_CR4_DE, edx, feature_bit(DE));
7089 cr4_fixed1_update(X86_CR4_PSE, edx, feature_bit(PSE));
7090 cr4_fixed1_update(X86_CR4_PAE, edx, feature_bit(PAE));
7091 cr4_fixed1_update(X86_CR4_MCE, edx, feature_bit(MCE));
7092 cr4_fixed1_update(X86_CR4_PGE, edx, feature_bit(PGE));
7093 cr4_fixed1_update(X86_CR4_OSFXSR, edx, feature_bit(FXSR));
7094 cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, feature_bit(XMM));
7095 cr4_fixed1_update(X86_CR4_VMXE, ecx, feature_bit(VMX));
7096 cr4_fixed1_update(X86_CR4_SMXE, ecx, feature_bit(SMX));
7097 cr4_fixed1_update(X86_CR4_PCIDE, ecx, feature_bit(PCID));
7098 cr4_fixed1_update(X86_CR4_OSXSAVE, ecx, feature_bit(XSAVE));
7100 entry = kvm_find_cpuid_entry(vcpu, 0x7, 0);
7101 cr4_fixed1_update(X86_CR4_FSGSBASE, ebx, feature_bit(FSGSBASE));
7102 cr4_fixed1_update(X86_CR4_SMEP, ebx, feature_bit(SMEP));
7103 cr4_fixed1_update(X86_CR4_SMAP, ebx, feature_bit(SMAP));
7104 cr4_fixed1_update(X86_CR4_PKE, ecx, feature_bit(PKU));
7105 cr4_fixed1_update(X86_CR4_UMIP, ecx, feature_bit(UMIP));
7106 cr4_fixed1_update(X86_CR4_LA57, ecx, feature_bit(LA57));
7108 #undef cr4_fixed1_update
7111 static void nested_vmx_entry_exit_ctls_update(struct kvm_vcpu *vcpu)
7113 struct vcpu_vmx *vmx = to_vmx(vcpu);
7115 if (kvm_mpx_supported()) {
7116 bool mpx_enabled = guest_cpuid_has(vcpu, X86_FEATURE_MPX);
7119 vmx->nested.msrs.entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
7120 vmx->nested.msrs.exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
7122 vmx->nested.msrs.entry_ctls_high &= ~VM_ENTRY_LOAD_BNDCFGS;
7123 vmx->nested.msrs.exit_ctls_high &= ~VM_EXIT_CLEAR_BNDCFGS;
7128 static void update_intel_pt_cfg(struct kvm_vcpu *vcpu)
7130 struct vcpu_vmx *vmx = to_vmx(vcpu);
7131 struct kvm_cpuid_entry2 *best = NULL;
7134 for (i = 0; i < PT_CPUID_LEAVES; i++) {
7135 best = kvm_find_cpuid_entry(vcpu, 0x14, i);
7138 vmx->pt_desc.caps[CPUID_EAX + i*PT_CPUID_REGS_NUM] = best->eax;
7139 vmx->pt_desc.caps[CPUID_EBX + i*PT_CPUID_REGS_NUM] = best->ebx;
7140 vmx->pt_desc.caps[CPUID_ECX + i*PT_CPUID_REGS_NUM] = best->ecx;
7141 vmx->pt_desc.caps[CPUID_EDX + i*PT_CPUID_REGS_NUM] = best->edx;
7144 /* Get the number of configurable Address Ranges for filtering */
7145 vmx->pt_desc.addr_range = intel_pt_validate_cap(vmx->pt_desc.caps,
7146 PT_CAP_num_address_ranges);
7148 /* Initialize and clear the no dependency bits */
7149 vmx->pt_desc.ctl_bitmask = ~(RTIT_CTL_TRACEEN | RTIT_CTL_OS |
7150 RTIT_CTL_USR | RTIT_CTL_TSC_EN | RTIT_CTL_DISRETC);
7153 * If CPUID.(EAX=14H,ECX=0):EBX[0]=1 CR3Filter can be set otherwise
7154 * will inject an #GP
7156 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_cr3_filtering))
7157 vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_CR3EN;
7160 * If CPUID.(EAX=14H,ECX=0):EBX[1]=1 CYCEn, CycThresh and
7161 * PSBFreq can be set
7163 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc))
7164 vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_CYCLEACC |
7165 RTIT_CTL_CYC_THRESH | RTIT_CTL_PSB_FREQ);
7168 * If CPUID.(EAX=14H,ECX=0):EBX[3]=1 MTCEn BranchEn and
7169 * MTCFreq can be set
7171 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc))
7172 vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_MTC_EN |
7173 RTIT_CTL_BRANCH_EN | RTIT_CTL_MTC_RANGE);
7175 /* If CPUID.(EAX=14H,ECX=0):EBX[4]=1 FUPonPTW and PTWEn can be set */
7176 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_ptwrite))
7177 vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_FUP_ON_PTW |
7180 /* If CPUID.(EAX=14H,ECX=0):EBX[5]=1 PwrEvEn can be set */
7181 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_power_event_trace))
7182 vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_PWR_EVT_EN;
7184 /* If CPUID.(EAX=14H,ECX=0):ECX[0]=1 ToPA can be set */
7185 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_topa_output))
7186 vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_TOPA;
7188 /* If CPUID.(EAX=14H,ECX=0):ECX[3]=1 FabircEn can be set */
7189 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_output_subsys))
7190 vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_FABRIC_EN;
7192 /* unmask address range configure area */
7193 for (i = 0; i < vmx->pt_desc.addr_range; i++)
7194 vmx->pt_desc.ctl_bitmask &= ~(0xfULL << (32 + i * 4));
7197 static void vmx_vcpu_after_set_cpuid(struct kvm_vcpu *vcpu)
7199 struct vcpu_vmx *vmx = to_vmx(vcpu);
7201 /* xsaves_enabled is recomputed in vmx_compute_secondary_exec_control(). */
7202 vcpu->arch.xsaves_enabled = false;
7204 if (cpu_has_secondary_exec_ctrls()) {
7205 vmx_compute_secondary_exec_control(vmx);
7206 vmcs_set_secondary_exec_control(vmx);
7209 if (nested_vmx_allowed(vcpu))
7210 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7211 FEAT_CTL_VMX_ENABLED_INSIDE_SMX |
7212 FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX;
7214 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7215 ~(FEAT_CTL_VMX_ENABLED_INSIDE_SMX |
7216 FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX);
7218 if (nested_vmx_allowed(vcpu)) {
7219 nested_vmx_cr_fixed1_bits_update(vcpu);
7220 nested_vmx_entry_exit_ctls_update(vcpu);
7223 if (boot_cpu_has(X86_FEATURE_INTEL_PT) &&
7224 guest_cpuid_has(vcpu, X86_FEATURE_INTEL_PT))
7225 update_intel_pt_cfg(vcpu);
7227 if (boot_cpu_has(X86_FEATURE_RTM)) {
7228 struct vmx_uret_msr *msr;
7229 msr = vmx_find_uret_msr(vmx, MSR_IA32_TSX_CTRL);
7231 bool enabled = guest_cpuid_has(vcpu, X86_FEATURE_RTM);
7232 vmx_set_guest_uret_msr(vmx, msr, enabled ? 0 : TSX_CTRL_RTM_DISABLE);
7237 static __init void vmx_set_cpu_caps(void)
7243 kvm_cpu_cap_set(X86_FEATURE_VMX);
7246 if (kvm_mpx_supported())
7247 kvm_cpu_cap_check_and_set(X86_FEATURE_MPX);
7248 if (cpu_has_vmx_invpcid())
7249 kvm_cpu_cap_check_and_set(X86_FEATURE_INVPCID);
7250 if (vmx_pt_mode_is_host_guest())
7251 kvm_cpu_cap_check_and_set(X86_FEATURE_INTEL_PT);
7253 if (vmx_umip_emulated())
7254 kvm_cpu_cap_set(X86_FEATURE_UMIP);
7258 if (!cpu_has_vmx_xsaves())
7259 kvm_cpu_cap_clear(X86_FEATURE_XSAVES);
7261 /* CPUID 0x80000001 */
7262 if (!cpu_has_vmx_rdtscp())
7263 kvm_cpu_cap_clear(X86_FEATURE_RDTSCP);
7265 if (cpu_has_vmx_waitpkg())
7266 kvm_cpu_cap_check_and_set(X86_FEATURE_WAITPKG);
7269 static void vmx_request_immediate_exit(struct kvm_vcpu *vcpu)
7271 to_vmx(vcpu)->req_immediate_exit = true;
7274 static int vmx_check_intercept_io(struct kvm_vcpu *vcpu,
7275 struct x86_instruction_info *info)
7277 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7278 unsigned short port;
7282 if (info->intercept == x86_intercept_in ||
7283 info->intercept == x86_intercept_ins) {
7284 port = info->src_val;
7285 size = info->dst_bytes;
7287 port = info->dst_val;
7288 size = info->src_bytes;
7292 * If the 'use IO bitmaps' VM-execution control is 0, IO instruction
7293 * VM-exits depend on the 'unconditional IO exiting' VM-execution
7296 * Otherwise, IO instruction VM-exits are controlled by the IO bitmaps.
7298 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
7299 intercept = nested_cpu_has(vmcs12,
7300 CPU_BASED_UNCOND_IO_EXITING);
7302 intercept = nested_vmx_check_io_bitmaps(vcpu, port, size);
7304 /* FIXME: produce nested vmexit and return X86EMUL_INTERCEPTED. */
7305 return intercept ? X86EMUL_UNHANDLEABLE : X86EMUL_CONTINUE;
7308 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
7309 struct x86_instruction_info *info,
7310 enum x86_intercept_stage stage,
7311 struct x86_exception *exception)
7313 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7315 switch (info->intercept) {
7317 * RDPID causes #UD if disabled through secondary execution controls.
7318 * Because it is marked as EmulateOnUD, we need to intercept it here.
7320 case x86_intercept_rdtscp:
7321 if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_RDTSCP)) {
7322 exception->vector = UD_VECTOR;
7323 exception->error_code_valid = false;
7324 return X86EMUL_PROPAGATE_FAULT;
7328 case x86_intercept_in:
7329 case x86_intercept_ins:
7330 case x86_intercept_out:
7331 case x86_intercept_outs:
7332 return vmx_check_intercept_io(vcpu, info);
7334 case x86_intercept_lgdt:
7335 case x86_intercept_lidt:
7336 case x86_intercept_lldt:
7337 case x86_intercept_ltr:
7338 case x86_intercept_sgdt:
7339 case x86_intercept_sidt:
7340 case x86_intercept_sldt:
7341 case x86_intercept_str:
7342 if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC))
7343 return X86EMUL_CONTINUE;
7345 /* FIXME: produce nested vmexit and return X86EMUL_INTERCEPTED. */
7348 /* TODO: check more intercepts... */
7353 return X86EMUL_UNHANDLEABLE;
7356 #ifdef CONFIG_X86_64
7357 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
7358 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
7359 u64 divisor, u64 *result)
7361 u64 low = a << shift, high = a >> (64 - shift);
7363 /* To avoid the overflow on divq */
7364 if (high >= divisor)
7367 /* Low hold the result, high hold rem which is discarded */
7368 asm("divq %2\n\t" : "=a" (low), "=d" (high) :
7369 "rm" (divisor), "0" (low), "1" (high));
7375 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc,
7378 struct vcpu_vmx *vmx;
7379 u64 tscl, guest_tscl, delta_tsc, lapic_timer_advance_cycles;
7380 struct kvm_timer *ktimer = &vcpu->arch.apic->lapic_timer;
7384 guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
7385 delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
7386 lapic_timer_advance_cycles = nsec_to_cycles(vcpu,
7387 ktimer->timer_advance_ns);
7389 if (delta_tsc > lapic_timer_advance_cycles)
7390 delta_tsc -= lapic_timer_advance_cycles;
7394 /* Convert to host delta tsc if tsc scaling is enabled */
7395 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
7396 delta_tsc && u64_shl_div_u64(delta_tsc,
7397 kvm_tsc_scaling_ratio_frac_bits,
7398 vcpu->arch.tsc_scaling_ratio, &delta_tsc))
7402 * If the delta tsc can't fit in the 32 bit after the multi shift,
7403 * we can't use the preemption timer.
7404 * It's possible that it fits on later vmentries, but checking
7405 * on every vmentry is costly so we just use an hrtimer.
7407 if (delta_tsc >> (cpu_preemption_timer_multi + 32))
7410 vmx->hv_deadline_tsc = tscl + delta_tsc;
7411 *expired = !delta_tsc;
7415 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
7417 to_vmx(vcpu)->hv_deadline_tsc = -1;
7421 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
7423 if (!kvm_pause_in_guest(vcpu->kvm))
7424 shrink_ple_window(vcpu);
7427 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
7428 struct kvm_memory_slot *slot)
7430 if (!kvm_dirty_log_manual_protect_and_init_set(kvm))
7431 kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
7432 kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
7435 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
7436 struct kvm_memory_slot *slot)
7438 kvm_mmu_slot_set_dirty(kvm, slot);
7441 static void vmx_flush_log_dirty(struct kvm *kvm)
7443 kvm_flush_pml_buffers(kvm);
7446 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
7447 struct kvm_memory_slot *memslot,
7448 gfn_t offset, unsigned long mask)
7450 kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
7453 static int vmx_pre_block(struct kvm_vcpu *vcpu)
7455 if (pi_pre_block(vcpu))
7458 if (kvm_lapic_hv_timer_in_use(vcpu))
7459 kvm_lapic_switch_to_sw_timer(vcpu);
7464 static void vmx_post_block(struct kvm_vcpu *vcpu)
7466 if (kvm_x86_ops.set_hv_timer)
7467 kvm_lapic_switch_to_hv_timer(vcpu);
7469 pi_post_block(vcpu);
7472 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
7474 if (vcpu->arch.mcg_cap & MCG_LMCE_P)
7475 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7476 FEAT_CTL_LMCE_ENABLED;
7478 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7479 ~FEAT_CTL_LMCE_ENABLED;
7482 static int vmx_smi_allowed(struct kvm_vcpu *vcpu, bool for_injection)
7484 /* we need a nested vmexit to enter SMM, postpone if run is pending */
7485 if (to_vmx(vcpu)->nested.nested_run_pending)
7487 return !is_smm(vcpu);
7490 static int vmx_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
7492 struct vcpu_vmx *vmx = to_vmx(vcpu);
7494 vmx->nested.smm.guest_mode = is_guest_mode(vcpu);
7495 if (vmx->nested.smm.guest_mode)
7496 nested_vmx_vmexit(vcpu, -1, 0, 0);
7498 vmx->nested.smm.vmxon = vmx->nested.vmxon;
7499 vmx->nested.vmxon = false;
7500 vmx_clear_hlt(vcpu);
7504 static int vmx_pre_leave_smm(struct kvm_vcpu *vcpu, const char *smstate)
7506 struct vcpu_vmx *vmx = to_vmx(vcpu);
7509 if (vmx->nested.smm.vmxon) {
7510 vmx->nested.vmxon = true;
7511 vmx->nested.smm.vmxon = false;
7514 if (vmx->nested.smm.guest_mode) {
7515 ret = nested_vmx_enter_non_root_mode(vcpu, false);
7519 vmx->nested.smm.guest_mode = false;
7524 static void enable_smi_window(struct kvm_vcpu *vcpu)
7526 /* RSM will cause a vmexit anyway. */
7529 static bool vmx_apic_init_signal_blocked(struct kvm_vcpu *vcpu)
7531 return to_vmx(vcpu)->nested.vmxon;
7534 static void vmx_migrate_timers(struct kvm_vcpu *vcpu)
7536 if (is_guest_mode(vcpu)) {
7537 struct hrtimer *timer = &to_vmx(vcpu)->nested.preemption_timer;
7539 if (hrtimer_try_to_cancel(timer) == 1)
7540 hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED);
7544 static void hardware_unsetup(void)
7547 nested_vmx_hardware_unsetup();
7552 static bool vmx_check_apicv_inhibit_reasons(ulong bit)
7554 ulong supported = BIT(APICV_INHIBIT_REASON_DISABLE) |
7555 BIT(APICV_INHIBIT_REASON_HYPERV);
7557 return supported & BIT(bit);
7560 static struct kvm_x86_ops vmx_x86_ops __initdata = {
7561 .hardware_unsetup = hardware_unsetup,
7563 .hardware_enable = hardware_enable,
7564 .hardware_disable = hardware_disable,
7565 .cpu_has_accelerated_tpr = report_flexpriority,
7566 .has_emulated_msr = vmx_has_emulated_msr,
7568 .vm_size = sizeof(struct kvm_vmx),
7569 .vm_init = vmx_vm_init,
7571 .vcpu_create = vmx_create_vcpu,
7572 .vcpu_free = vmx_free_vcpu,
7573 .vcpu_reset = vmx_vcpu_reset,
7575 .prepare_guest_switch = vmx_prepare_switch_to_guest,
7576 .vcpu_load = vmx_vcpu_load,
7577 .vcpu_put = vmx_vcpu_put,
7579 .update_exception_bitmap = update_exception_bitmap,
7580 .get_msr_feature = vmx_get_msr_feature,
7581 .get_msr = vmx_get_msr,
7582 .set_msr = vmx_set_msr,
7583 .get_segment_base = vmx_get_segment_base,
7584 .get_segment = vmx_get_segment,
7585 .set_segment = vmx_set_segment,
7586 .get_cpl = vmx_get_cpl,
7587 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
7588 .set_cr0 = vmx_set_cr0,
7589 .set_cr4 = vmx_set_cr4,
7590 .set_efer = vmx_set_efer,
7591 .get_idt = vmx_get_idt,
7592 .set_idt = vmx_set_idt,
7593 .get_gdt = vmx_get_gdt,
7594 .set_gdt = vmx_set_gdt,
7595 .set_dr7 = vmx_set_dr7,
7596 .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
7597 .cache_reg = vmx_cache_reg,
7598 .get_rflags = vmx_get_rflags,
7599 .set_rflags = vmx_set_rflags,
7601 .tlb_flush_all = vmx_flush_tlb_all,
7602 .tlb_flush_current = vmx_flush_tlb_current,
7603 .tlb_flush_gva = vmx_flush_tlb_gva,
7604 .tlb_flush_guest = vmx_flush_tlb_guest,
7606 .run = vmx_vcpu_run,
7607 .handle_exit = vmx_handle_exit,
7608 .skip_emulated_instruction = vmx_skip_emulated_instruction,
7609 .update_emulated_instruction = vmx_update_emulated_instruction,
7610 .set_interrupt_shadow = vmx_set_interrupt_shadow,
7611 .get_interrupt_shadow = vmx_get_interrupt_shadow,
7612 .patch_hypercall = vmx_patch_hypercall,
7613 .set_irq = vmx_inject_irq,
7614 .set_nmi = vmx_inject_nmi,
7615 .queue_exception = vmx_queue_exception,
7616 .cancel_injection = vmx_cancel_injection,
7617 .interrupt_allowed = vmx_interrupt_allowed,
7618 .nmi_allowed = vmx_nmi_allowed,
7619 .get_nmi_mask = vmx_get_nmi_mask,
7620 .set_nmi_mask = vmx_set_nmi_mask,
7621 .enable_nmi_window = enable_nmi_window,
7622 .enable_irq_window = enable_irq_window,
7623 .update_cr8_intercept = update_cr8_intercept,
7624 .set_virtual_apic_mode = vmx_set_virtual_apic_mode,
7625 .set_apic_access_page_addr = vmx_set_apic_access_page_addr,
7626 .refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
7627 .load_eoi_exitmap = vmx_load_eoi_exitmap,
7628 .apicv_post_state_restore = vmx_apicv_post_state_restore,
7629 .check_apicv_inhibit_reasons = vmx_check_apicv_inhibit_reasons,
7630 .hwapic_irr_update = vmx_hwapic_irr_update,
7631 .hwapic_isr_update = vmx_hwapic_isr_update,
7632 .guest_apic_has_interrupt = vmx_guest_apic_has_interrupt,
7633 .sync_pir_to_irr = vmx_sync_pir_to_irr,
7634 .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
7635 .dy_apicv_has_pending_interrupt = pi_has_pending_interrupt,
7637 .set_tss_addr = vmx_set_tss_addr,
7638 .set_identity_map_addr = vmx_set_identity_map_addr,
7639 .get_mt_mask = vmx_get_mt_mask,
7641 .get_exit_info = vmx_get_exit_info,
7643 .vcpu_after_set_cpuid = vmx_vcpu_after_set_cpuid,
7645 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
7647 .write_l1_tsc_offset = vmx_write_l1_tsc_offset,
7649 .load_mmu_pgd = vmx_load_mmu_pgd,
7651 .check_intercept = vmx_check_intercept,
7652 .handle_exit_irqoff = vmx_handle_exit_irqoff,
7654 .request_immediate_exit = vmx_request_immediate_exit,
7656 .sched_in = vmx_sched_in,
7658 .slot_enable_log_dirty = vmx_slot_enable_log_dirty,
7659 .slot_disable_log_dirty = vmx_slot_disable_log_dirty,
7660 .flush_log_dirty = vmx_flush_log_dirty,
7661 .enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
7663 .pre_block = vmx_pre_block,
7664 .post_block = vmx_post_block,
7666 .pmu_ops = &intel_pmu_ops,
7667 .nested_ops = &vmx_nested_ops,
7669 .update_pi_irte = pi_update_irte,
7671 #ifdef CONFIG_X86_64
7672 .set_hv_timer = vmx_set_hv_timer,
7673 .cancel_hv_timer = vmx_cancel_hv_timer,
7676 .setup_mce = vmx_setup_mce,
7678 .smi_allowed = vmx_smi_allowed,
7679 .pre_enter_smm = vmx_pre_enter_smm,
7680 .pre_leave_smm = vmx_pre_leave_smm,
7681 .enable_smi_window = enable_smi_window,
7683 .can_emulate_instruction = vmx_can_emulate_instruction,
7684 .apic_init_signal_blocked = vmx_apic_init_signal_blocked,
7685 .migrate_timers = vmx_migrate_timers,
7687 .msr_filter_changed = vmx_msr_filter_changed,
7690 static __init int hardware_setup(void)
7692 unsigned long host_bndcfgs;
7694 int r, i, ept_lpage_level;
7697 host_idt_base = dt.address;
7699 for (i = 0; i < ARRAY_SIZE(vmx_uret_msrs_list); ++i)
7700 kvm_define_user_return_msr(i, vmx_uret_msrs_list[i]);
7702 if (setup_vmcs_config(&vmcs_config, &vmx_capability) < 0)
7705 if (boot_cpu_has(X86_FEATURE_NX))
7706 kvm_enable_efer_bits(EFER_NX);
7708 if (boot_cpu_has(X86_FEATURE_MPX)) {
7709 rdmsrl(MSR_IA32_BNDCFGS, host_bndcfgs);
7710 WARN_ONCE(host_bndcfgs, "KVM: BNDCFGS in host will be lost");
7713 if (!cpu_has_vmx_mpx())
7714 supported_xcr0 &= ~(XFEATURE_MASK_BNDREGS |
7715 XFEATURE_MASK_BNDCSR);
7717 if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
7718 !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
7721 if (!cpu_has_vmx_ept() ||
7722 !cpu_has_vmx_ept_4levels() ||
7723 !cpu_has_vmx_ept_mt_wb() ||
7724 !cpu_has_vmx_invept_global())
7727 if (!cpu_has_vmx_ept_ad_bits() || !enable_ept)
7728 enable_ept_ad_bits = 0;
7730 if (!cpu_has_vmx_unrestricted_guest() || !enable_ept)
7731 enable_unrestricted_guest = 0;
7733 if (!cpu_has_vmx_flexpriority())
7734 flexpriority_enabled = 0;
7736 if (!cpu_has_virtual_nmis())
7740 * set_apic_access_page_addr() is used to reload apic access
7741 * page upon invalidation. No need to do anything if not
7742 * using the APIC_ACCESS_ADDR VMCS field.
7744 if (!flexpriority_enabled)
7745 vmx_x86_ops.set_apic_access_page_addr = NULL;
7747 if (!cpu_has_vmx_tpr_shadow())
7748 vmx_x86_ops.update_cr8_intercept = NULL;
7750 #if IS_ENABLED(CONFIG_HYPERV)
7751 if (ms_hyperv.nested_features & HV_X64_NESTED_GUEST_MAPPING_FLUSH
7753 vmx_x86_ops.tlb_remote_flush = hv_remote_flush_tlb;
7754 vmx_x86_ops.tlb_remote_flush_with_range =
7755 hv_remote_flush_tlb_with_range;
7759 if (!cpu_has_vmx_ple()) {
7762 ple_window_grow = 0;
7764 ple_window_shrink = 0;
7767 if (!cpu_has_vmx_apicv()) {
7769 vmx_x86_ops.sync_pir_to_irr = NULL;
7772 if (cpu_has_vmx_tsc_scaling()) {
7773 kvm_has_tsc_control = true;
7774 kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
7775 kvm_tsc_scaling_ratio_frac_bits = 48;
7778 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
7784 ept_lpage_level = 0;
7785 else if (cpu_has_vmx_ept_1g_page())
7786 ept_lpage_level = PG_LEVEL_1G;
7787 else if (cpu_has_vmx_ept_2m_page())
7788 ept_lpage_level = PG_LEVEL_2M;
7790 ept_lpage_level = PG_LEVEL_4K;
7791 kvm_configure_mmu(enable_ept, vmx_get_max_tdp_level(), ept_lpage_level);
7794 * Only enable PML when hardware supports PML feature, and both EPT
7795 * and EPT A/D bit features are enabled -- PML depends on them to work.
7797 if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
7801 vmx_x86_ops.slot_enable_log_dirty = NULL;
7802 vmx_x86_ops.slot_disable_log_dirty = NULL;
7803 vmx_x86_ops.flush_log_dirty = NULL;
7804 vmx_x86_ops.enable_log_dirty_pt_masked = NULL;
7807 if (!cpu_has_vmx_preemption_timer())
7808 enable_preemption_timer = false;
7810 if (enable_preemption_timer) {
7811 u64 use_timer_freq = 5000ULL * 1000 * 1000;
7814 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
7815 cpu_preemption_timer_multi =
7816 vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
7819 use_timer_freq = (u64)tsc_khz * 1000;
7820 use_timer_freq >>= cpu_preemption_timer_multi;
7823 * KVM "disables" the preemption timer by setting it to its max
7824 * value. Don't use the timer if it might cause spurious exits
7825 * at a rate faster than 0.1 Hz (of uninterrupted guest time).
7827 if (use_timer_freq > 0xffffffffu / 10)
7828 enable_preemption_timer = false;
7831 if (!enable_preemption_timer) {
7832 vmx_x86_ops.set_hv_timer = NULL;
7833 vmx_x86_ops.cancel_hv_timer = NULL;
7834 vmx_x86_ops.request_immediate_exit = __kvm_request_immediate_exit;
7837 kvm_set_posted_intr_wakeup_handler(pi_wakeup_handler);
7839 kvm_mce_cap_supported |= MCG_LMCE_P;
7841 if (pt_mode != PT_MODE_SYSTEM && pt_mode != PT_MODE_HOST_GUEST)
7843 if (!enable_ept || !cpu_has_vmx_intel_pt())
7844 pt_mode = PT_MODE_SYSTEM;
7847 nested_vmx_setup_ctls_msrs(&vmcs_config.nested,
7848 vmx_capability.ept);
7850 r = nested_vmx_hardware_setup(kvm_vmx_exit_handlers);
7857 r = alloc_kvm_area();
7859 nested_vmx_hardware_unsetup();
7863 static struct kvm_x86_init_ops vmx_init_ops __initdata = {
7864 .cpu_has_kvm_support = cpu_has_kvm_support,
7865 .disabled_by_bios = vmx_disabled_by_bios,
7866 .check_processor_compatibility = vmx_check_processor_compat,
7867 .hardware_setup = hardware_setup,
7869 .runtime_ops = &vmx_x86_ops,
7872 static void vmx_cleanup_l1d_flush(void)
7874 if (vmx_l1d_flush_pages) {
7875 free_pages((unsigned long)vmx_l1d_flush_pages, L1D_CACHE_ORDER);
7876 vmx_l1d_flush_pages = NULL;
7878 /* Restore state so sysfs ignores VMX */
7879 l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
7882 static void vmx_exit(void)
7884 #ifdef CONFIG_KEXEC_CORE
7885 RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
7891 #if IS_ENABLED(CONFIG_HYPERV)
7892 if (static_branch_unlikely(&enable_evmcs)) {
7894 struct hv_vp_assist_page *vp_ap;
7896 * Reset everything to support using non-enlightened VMCS
7897 * access later (e.g. when we reload the module with
7898 * enlightened_vmcs=0)
7900 for_each_online_cpu(cpu) {
7901 vp_ap = hv_get_vp_assist_page(cpu);
7906 vp_ap->nested_control.features.directhypercall = 0;
7907 vp_ap->current_nested_vmcs = 0;
7908 vp_ap->enlighten_vmentry = 0;
7911 static_branch_disable(&enable_evmcs);
7914 vmx_cleanup_l1d_flush();
7916 module_exit(vmx_exit);
7918 static int __init vmx_init(void)
7922 #if IS_ENABLED(CONFIG_HYPERV)
7924 * Enlightened VMCS usage should be recommended and the host needs
7925 * to support eVMCS v1 or above. We can also disable eVMCS support
7926 * with module parameter.
7928 if (enlightened_vmcs &&
7929 ms_hyperv.hints & HV_X64_ENLIGHTENED_VMCS_RECOMMENDED &&
7930 (ms_hyperv.nested_features & HV_X64_ENLIGHTENED_VMCS_VERSION) >=
7931 KVM_EVMCS_VERSION) {
7934 /* Check that we have assist pages on all online CPUs */
7935 for_each_online_cpu(cpu) {
7936 if (!hv_get_vp_assist_page(cpu)) {
7937 enlightened_vmcs = false;
7942 if (enlightened_vmcs) {
7943 pr_info("KVM: vmx: using Hyper-V Enlightened VMCS\n");
7944 static_branch_enable(&enable_evmcs);
7947 if (ms_hyperv.nested_features & HV_X64_NESTED_DIRECT_FLUSH)
7948 vmx_x86_ops.enable_direct_tlbflush
7949 = hv_enable_direct_tlbflush;
7952 enlightened_vmcs = false;
7956 r = kvm_init(&vmx_init_ops, sizeof(struct vcpu_vmx),
7957 __alignof__(struct vcpu_vmx), THIS_MODULE);
7962 * Must be called after kvm_init() so enable_ept is properly set
7963 * up. Hand the parameter mitigation value in which was stored in
7964 * the pre module init parser. If no parameter was given, it will
7965 * contain 'auto' which will be turned into the default 'cond'
7968 r = vmx_setup_l1d_flush(vmentry_l1d_flush_param);
7974 for_each_possible_cpu(cpu) {
7975 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
7980 #ifdef CONFIG_KEXEC_CORE
7981 rcu_assign_pointer(crash_vmclear_loaded_vmcss,
7982 crash_vmclear_local_loaded_vmcss);
7984 vmx_check_vmcs12_offsets();
7987 * Intel processors don't have problems with
7988 * GUEST_MAXPHYADDR < HOST_MAXPHYADDR so enable
7989 * it for VMX by default
7991 allow_smaller_maxphyaddr = true;
7995 module_init(vmx_init);