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>
34 #include <asm/debugreg.h>
36 #include <asm/fpu/internal.h>
38 #include <asm/irq_remapping.h>
39 #include <asm/kexec.h>
40 #include <asm/perf_event.h>
42 #include <asm/mmu_context.h>
43 #include <asm/mshyperv.h>
44 #include <asm/spec-ctrl.h>
45 #include <asm/virtext.h>
48 #include "capabilities.h"
52 #include "kvm_cache_regs.h"
64 MODULE_AUTHOR("Qumranet");
65 MODULE_LICENSE("GPL");
67 static const struct x86_cpu_id vmx_cpu_id[] = {
68 X86_FEATURE_MATCH(X86_FEATURE_VMX),
71 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
73 bool __read_mostly enable_vpid = 1;
74 module_param_named(vpid, enable_vpid, bool, 0444);
76 static bool __read_mostly enable_vnmi = 1;
77 module_param_named(vnmi, enable_vnmi, bool, S_IRUGO);
79 bool __read_mostly flexpriority_enabled = 1;
80 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
82 bool __read_mostly enable_ept = 1;
83 module_param_named(ept, enable_ept, bool, S_IRUGO);
85 bool __read_mostly enable_unrestricted_guest = 1;
86 module_param_named(unrestricted_guest,
87 enable_unrestricted_guest, bool, S_IRUGO);
89 bool __read_mostly enable_ept_ad_bits = 1;
90 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
92 static bool __read_mostly emulate_invalid_guest_state = true;
93 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
95 static bool __read_mostly fasteoi = 1;
96 module_param(fasteoi, bool, S_IRUGO);
98 bool __read_mostly enable_apicv = 1;
99 module_param(enable_apicv, bool, S_IRUGO);
102 * If nested=1, nested virtualization is supported, i.e., guests may use
103 * VMX and be a hypervisor for its own guests. If nested=0, guests may not
104 * use VMX instructions.
106 static bool __read_mostly nested = 1;
107 module_param(nested, bool, S_IRUGO);
109 static u64 __read_mostly host_xss;
111 bool __read_mostly enable_pml = 1;
112 module_param_named(pml, enable_pml, bool, S_IRUGO);
114 static bool __read_mostly dump_invalid_vmcs = 0;
115 module_param(dump_invalid_vmcs, bool, 0644);
117 #define MSR_BITMAP_MODE_X2APIC 1
118 #define MSR_BITMAP_MODE_X2APIC_APICV 2
120 #define KVM_VMX_TSC_MULTIPLIER_MAX 0xffffffffffffffffULL
122 /* Guest_tsc -> host_tsc conversion requires 64-bit division. */
123 static int __read_mostly cpu_preemption_timer_multi;
124 static bool __read_mostly enable_preemption_timer = 1;
126 module_param_named(preemption_timer, enable_preemption_timer, bool, S_IRUGO);
129 #define KVM_VM_CR0_ALWAYS_OFF (X86_CR0_NW | X86_CR0_CD)
130 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR0_NE
131 #define KVM_VM_CR0_ALWAYS_ON \
132 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | \
133 X86_CR0_WP | X86_CR0_PG | X86_CR0_PE)
134 #define KVM_CR4_GUEST_OWNED_BITS \
135 (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \
136 | X86_CR4_OSXMMEXCPT | X86_CR4_LA57 | X86_CR4_TSD)
138 #define KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR4_VMXE
139 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
140 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
142 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
144 #define MSR_IA32_RTIT_STATUS_MASK (~(RTIT_STATUS_FILTEREN | \
145 RTIT_STATUS_CONTEXTEN | RTIT_STATUS_TRIGGEREN | \
146 RTIT_STATUS_ERROR | RTIT_STATUS_STOPPED | \
147 RTIT_STATUS_BYTECNT))
149 #define MSR_IA32_RTIT_OUTPUT_BASE_MASK \
150 (~((1UL << cpuid_query_maxphyaddr(vcpu)) - 1) | 0x7f)
153 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
154 * ple_gap: upper bound on the amount of time between two successive
155 * executions of PAUSE in a loop. Also indicate if ple enabled.
156 * According to test, this time is usually smaller than 128 cycles.
157 * ple_window: upper bound on the amount of time a guest is allowed to execute
158 * in a PAUSE loop. Tests indicate that most spinlocks are held for
159 * less than 2^12 cycles
160 * Time is measured based on a counter that runs at the same rate as the TSC,
161 * refer SDM volume 3b section 21.6.13 & 22.1.3.
163 static unsigned int ple_gap = KVM_DEFAULT_PLE_GAP;
164 module_param(ple_gap, uint, 0444);
166 static unsigned int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
167 module_param(ple_window, uint, 0444);
169 /* Default doubles per-vcpu window every exit. */
170 static unsigned int ple_window_grow = KVM_DEFAULT_PLE_WINDOW_GROW;
171 module_param(ple_window_grow, uint, 0444);
173 /* Default resets per-vcpu window every exit to ple_window. */
174 static unsigned int ple_window_shrink = KVM_DEFAULT_PLE_WINDOW_SHRINK;
175 module_param(ple_window_shrink, uint, 0444);
177 /* Default is to compute the maximum so we can never overflow. */
178 static unsigned int ple_window_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
179 module_param(ple_window_max, uint, 0444);
181 /* Default is SYSTEM mode, 1 for host-guest mode */
182 int __read_mostly pt_mode = PT_MODE_SYSTEM;
183 module_param(pt_mode, int, S_IRUGO);
185 static DEFINE_STATIC_KEY_FALSE(vmx_l1d_should_flush);
186 static DEFINE_STATIC_KEY_FALSE(vmx_l1d_flush_cond);
187 static DEFINE_MUTEX(vmx_l1d_flush_mutex);
189 /* Storage for pre module init parameter parsing */
190 static enum vmx_l1d_flush_state __read_mostly vmentry_l1d_flush_param = VMENTER_L1D_FLUSH_AUTO;
192 static const struct {
195 } vmentry_l1d_param[] = {
196 [VMENTER_L1D_FLUSH_AUTO] = {"auto", true},
197 [VMENTER_L1D_FLUSH_NEVER] = {"never", true},
198 [VMENTER_L1D_FLUSH_COND] = {"cond", true},
199 [VMENTER_L1D_FLUSH_ALWAYS] = {"always", true},
200 [VMENTER_L1D_FLUSH_EPT_DISABLED] = {"EPT disabled", false},
201 [VMENTER_L1D_FLUSH_NOT_REQUIRED] = {"not required", false},
204 #define L1D_CACHE_ORDER 4
205 static void *vmx_l1d_flush_pages;
207 static int vmx_setup_l1d_flush(enum vmx_l1d_flush_state l1tf)
212 if (!boot_cpu_has_bug(X86_BUG_L1TF)) {
213 l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED;
218 l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_EPT_DISABLED;
222 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES)) {
225 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, msr);
226 if (msr & ARCH_CAP_SKIP_VMENTRY_L1DFLUSH) {
227 l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED;
232 /* If set to auto use the default l1tf mitigation method */
233 if (l1tf == VMENTER_L1D_FLUSH_AUTO) {
234 switch (l1tf_mitigation) {
235 case L1TF_MITIGATION_OFF:
236 l1tf = VMENTER_L1D_FLUSH_NEVER;
238 case L1TF_MITIGATION_FLUSH_NOWARN:
239 case L1TF_MITIGATION_FLUSH:
240 case L1TF_MITIGATION_FLUSH_NOSMT:
241 l1tf = VMENTER_L1D_FLUSH_COND;
243 case L1TF_MITIGATION_FULL:
244 case L1TF_MITIGATION_FULL_FORCE:
245 l1tf = VMENTER_L1D_FLUSH_ALWAYS;
248 } else if (l1tf_mitigation == L1TF_MITIGATION_FULL_FORCE) {
249 l1tf = VMENTER_L1D_FLUSH_ALWAYS;
252 if (l1tf != VMENTER_L1D_FLUSH_NEVER && !vmx_l1d_flush_pages &&
253 !boot_cpu_has(X86_FEATURE_FLUSH_L1D)) {
255 * This allocation for vmx_l1d_flush_pages is not tied to a VM
256 * lifetime and so should not be charged to a memcg.
258 page = alloc_pages(GFP_KERNEL, L1D_CACHE_ORDER);
261 vmx_l1d_flush_pages = page_address(page);
264 * Initialize each page with a different pattern in
265 * order to protect against KSM in the nested
266 * virtualization case.
268 for (i = 0; i < 1u << L1D_CACHE_ORDER; ++i) {
269 memset(vmx_l1d_flush_pages + i * PAGE_SIZE, i + 1,
274 l1tf_vmx_mitigation = l1tf;
276 if (l1tf != VMENTER_L1D_FLUSH_NEVER)
277 static_branch_enable(&vmx_l1d_should_flush);
279 static_branch_disable(&vmx_l1d_should_flush);
281 if (l1tf == VMENTER_L1D_FLUSH_COND)
282 static_branch_enable(&vmx_l1d_flush_cond);
284 static_branch_disable(&vmx_l1d_flush_cond);
288 static int vmentry_l1d_flush_parse(const char *s)
293 for (i = 0; i < ARRAY_SIZE(vmentry_l1d_param); i++) {
294 if (vmentry_l1d_param[i].for_parse &&
295 sysfs_streq(s, vmentry_l1d_param[i].option))
302 static int vmentry_l1d_flush_set(const char *s, const struct kernel_param *kp)
306 l1tf = vmentry_l1d_flush_parse(s);
310 if (!boot_cpu_has(X86_BUG_L1TF))
314 * Has vmx_init() run already? If not then this is the pre init
315 * parameter parsing. In that case just store the value and let
316 * vmx_init() do the proper setup after enable_ept has been
319 if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_AUTO) {
320 vmentry_l1d_flush_param = l1tf;
324 mutex_lock(&vmx_l1d_flush_mutex);
325 ret = vmx_setup_l1d_flush(l1tf);
326 mutex_unlock(&vmx_l1d_flush_mutex);
330 static int vmentry_l1d_flush_get(char *s, const struct kernel_param *kp)
332 if (WARN_ON_ONCE(l1tf_vmx_mitigation >= ARRAY_SIZE(vmentry_l1d_param)))
333 return sprintf(s, "???\n");
335 return sprintf(s, "%s\n", vmentry_l1d_param[l1tf_vmx_mitigation].option);
338 static const struct kernel_param_ops vmentry_l1d_flush_ops = {
339 .set = vmentry_l1d_flush_set,
340 .get = vmentry_l1d_flush_get,
342 module_param_cb(vmentry_l1d_flush, &vmentry_l1d_flush_ops, NULL, 0644);
344 static bool guest_state_valid(struct kvm_vcpu *vcpu);
345 static u32 vmx_segment_access_rights(struct kvm_segment *var);
346 static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
349 void vmx_vmexit(void);
351 #define vmx_insn_failed(fmt...) \
354 pr_warn_ratelimited(fmt); \
357 asmlinkage void vmread_error(unsigned long field, bool fault)
360 kvm_spurious_fault();
362 vmx_insn_failed("kvm: vmread failed: field=%lx\n", field);
365 noinline void vmwrite_error(unsigned long field, unsigned long value)
367 vmx_insn_failed("kvm: vmwrite failed: field=%lx val=%lx err=%d\n",
368 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
371 noinline void vmclear_error(struct vmcs *vmcs, u64 phys_addr)
373 vmx_insn_failed("kvm: vmclear failed: %p/%llx\n", vmcs, phys_addr);
376 noinline void vmptrld_error(struct vmcs *vmcs, u64 phys_addr)
378 vmx_insn_failed("kvm: vmptrld failed: %p/%llx\n", vmcs, phys_addr);
381 noinline void invvpid_error(unsigned long ext, u16 vpid, gva_t gva)
383 vmx_insn_failed("kvm: invvpid failed: ext=0x%lx vpid=%u gva=0x%lx\n",
387 noinline void invept_error(unsigned long ext, u64 eptp, gpa_t gpa)
389 vmx_insn_failed("kvm: invept failed: ext=0x%lx eptp=%llx gpa=0x%llx\n",
393 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
394 DEFINE_PER_CPU(struct vmcs *, current_vmcs);
396 * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
397 * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
399 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
402 * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
403 * can find which vCPU should be waken up.
405 static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu);
406 static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock);
408 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
409 static DEFINE_SPINLOCK(vmx_vpid_lock);
411 struct vmcs_config vmcs_config;
412 struct vmx_capability vmx_capability;
414 #define VMX_SEGMENT_FIELD(seg) \
415 [VCPU_SREG_##seg] = { \
416 .selector = GUEST_##seg##_SELECTOR, \
417 .base = GUEST_##seg##_BASE, \
418 .limit = GUEST_##seg##_LIMIT, \
419 .ar_bytes = GUEST_##seg##_AR_BYTES, \
422 static const struct kvm_vmx_segment_field {
427 } kvm_vmx_segment_fields[] = {
428 VMX_SEGMENT_FIELD(CS),
429 VMX_SEGMENT_FIELD(DS),
430 VMX_SEGMENT_FIELD(ES),
431 VMX_SEGMENT_FIELD(FS),
432 VMX_SEGMENT_FIELD(GS),
433 VMX_SEGMENT_FIELD(SS),
434 VMX_SEGMENT_FIELD(TR),
435 VMX_SEGMENT_FIELD(LDTR),
439 static unsigned long host_idt_base;
442 * Though SYSCALL is only supported in 64-bit mode on Intel CPUs, kvm
443 * will emulate SYSCALL in legacy mode if the vendor string in guest
444 * CPUID.0:{EBX,ECX,EDX} is "AuthenticAMD" or "AMDisbetter!" To
445 * support this emulation, IA32_STAR must always be included in
446 * vmx_msr_index[], even in i386 builds.
448 const u32 vmx_msr_index[] = {
450 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
452 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
455 #if IS_ENABLED(CONFIG_HYPERV)
456 static bool __read_mostly enlightened_vmcs = true;
457 module_param(enlightened_vmcs, bool, 0444);
459 /* check_ept_pointer() should be under protection of ept_pointer_lock. */
460 static void check_ept_pointer_match(struct kvm *kvm)
462 struct kvm_vcpu *vcpu;
463 u64 tmp_eptp = INVALID_PAGE;
466 kvm_for_each_vcpu(i, vcpu, kvm) {
467 if (!VALID_PAGE(tmp_eptp)) {
468 tmp_eptp = to_vmx(vcpu)->ept_pointer;
469 } else if (tmp_eptp != to_vmx(vcpu)->ept_pointer) {
470 to_kvm_vmx(kvm)->ept_pointers_match
471 = EPT_POINTERS_MISMATCH;
476 to_kvm_vmx(kvm)->ept_pointers_match = EPT_POINTERS_MATCH;
479 static int kvm_fill_hv_flush_list_func(struct hv_guest_mapping_flush_list *flush,
482 struct kvm_tlb_range *range = data;
484 return hyperv_fill_flush_guest_mapping_list(flush, range->start_gfn,
488 static inline int __hv_remote_flush_tlb_with_range(struct kvm *kvm,
489 struct kvm_vcpu *vcpu, struct kvm_tlb_range *range)
491 u64 ept_pointer = to_vmx(vcpu)->ept_pointer;
494 * FLUSH_GUEST_PHYSICAL_ADDRESS_SPACE hypercall needs address
495 * of the base of EPT PML4 table, strip off EPT configuration
499 return hyperv_flush_guest_mapping_range(ept_pointer & PAGE_MASK,
500 kvm_fill_hv_flush_list_func, (void *)range);
502 return hyperv_flush_guest_mapping(ept_pointer & PAGE_MASK);
505 static int hv_remote_flush_tlb_with_range(struct kvm *kvm,
506 struct kvm_tlb_range *range)
508 struct kvm_vcpu *vcpu;
511 spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
513 if (to_kvm_vmx(kvm)->ept_pointers_match == EPT_POINTERS_CHECK)
514 check_ept_pointer_match(kvm);
516 if (to_kvm_vmx(kvm)->ept_pointers_match != EPT_POINTERS_MATCH) {
517 kvm_for_each_vcpu(i, vcpu, kvm) {
518 /* If ept_pointer is invalid pointer, bypass flush request. */
519 if (VALID_PAGE(to_vmx(vcpu)->ept_pointer))
520 ret |= __hv_remote_flush_tlb_with_range(
524 ret = __hv_remote_flush_tlb_with_range(kvm,
525 kvm_get_vcpu(kvm, 0), range);
528 spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
531 static int hv_remote_flush_tlb(struct kvm *kvm)
533 return hv_remote_flush_tlb_with_range(kvm, NULL);
536 static int hv_enable_direct_tlbflush(struct kvm_vcpu *vcpu)
538 struct hv_enlightened_vmcs *evmcs;
539 struct hv_partition_assist_pg **p_hv_pa_pg =
540 &vcpu->kvm->arch.hyperv.hv_pa_pg;
542 * Synthetic VM-Exit is not enabled in current code and so All
543 * evmcs in singe VM shares same assist page.
546 *p_hv_pa_pg = kzalloc(PAGE_SIZE, GFP_KERNEL);
551 evmcs = (struct hv_enlightened_vmcs *)to_vmx(vcpu)->loaded_vmcs->vmcs;
553 evmcs->partition_assist_page =
555 evmcs->hv_vm_id = (unsigned long)vcpu->kvm;
556 evmcs->hv_enlightenments_control.nested_flush_hypercall = 1;
561 #endif /* IS_ENABLED(CONFIG_HYPERV) */
564 * Comment's format: document - errata name - stepping - processor name.
566 * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
568 static u32 vmx_preemption_cpu_tfms[] = {
569 /* 323344.pdf - BA86 - D0 - Xeon 7500 Series */
571 /* 323056.pdf - AAX65 - C2 - Xeon L3406 */
572 /* 322814.pdf - AAT59 - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
573 /* 322911.pdf - AAU65 - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
575 /* 322911.pdf - AAU65 - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
577 /* 322373.pdf - AAO95 - B1 - Xeon 3400 Series */
578 /* 322166.pdf - AAN92 - B1 - i7-800 and i5-700 Desktop */
580 * 320767.pdf - AAP86 - B1 -
581 * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
584 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
586 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
588 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
590 /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
591 /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
592 /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
594 /* Xeon E3-1220 V2 */
598 static inline bool cpu_has_broken_vmx_preemption_timer(void)
600 u32 eax = cpuid_eax(0x00000001), i;
602 /* Clear the reserved bits */
603 eax &= ~(0x3U << 14 | 0xfU << 28);
604 for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
605 if (eax == vmx_preemption_cpu_tfms[i])
611 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
613 return flexpriority_enabled && lapic_in_kernel(vcpu);
616 static inline bool report_flexpriority(void)
618 return flexpriority_enabled;
621 static inline int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
625 for (i = 0; i < vmx->nmsrs; ++i)
626 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
631 struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
635 i = __find_msr_index(vmx, msr);
637 return &vmx->guest_msrs[i];
641 void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
643 vmcs_clear(loaded_vmcs->vmcs);
644 if (loaded_vmcs->shadow_vmcs && loaded_vmcs->launched)
645 vmcs_clear(loaded_vmcs->shadow_vmcs);
646 loaded_vmcs->cpu = -1;
647 loaded_vmcs->launched = 0;
650 #ifdef CONFIG_KEXEC_CORE
652 * This bitmap is used to indicate whether the vmclear
653 * operation is enabled on all cpus. All disabled by
656 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
658 static inline void crash_enable_local_vmclear(int cpu)
660 cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
663 static inline void crash_disable_local_vmclear(int cpu)
665 cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
668 static inline int crash_local_vmclear_enabled(int cpu)
670 return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
673 static void crash_vmclear_local_loaded_vmcss(void)
675 int cpu = raw_smp_processor_id();
676 struct loaded_vmcs *v;
678 if (!crash_local_vmclear_enabled(cpu))
681 list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
682 loaded_vmcss_on_cpu_link)
686 static inline void crash_enable_local_vmclear(int cpu) { }
687 static inline void crash_disable_local_vmclear(int cpu) { }
688 #endif /* CONFIG_KEXEC_CORE */
690 static void __loaded_vmcs_clear(void *arg)
692 struct loaded_vmcs *loaded_vmcs = arg;
693 int cpu = raw_smp_processor_id();
695 if (loaded_vmcs->cpu != cpu)
696 return; /* vcpu migration can race with cpu offline */
697 if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
698 per_cpu(current_vmcs, cpu) = NULL;
699 crash_disable_local_vmclear(cpu);
700 list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
703 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
704 * is before setting loaded_vmcs->vcpu to -1 which is done in
705 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
706 * then adds the vmcs into percpu list before it is deleted.
710 loaded_vmcs_init(loaded_vmcs);
711 crash_enable_local_vmclear(cpu);
714 void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
716 int cpu = loaded_vmcs->cpu;
719 smp_call_function_single(cpu,
720 __loaded_vmcs_clear, loaded_vmcs, 1);
723 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
727 u32 mask = 1 << (seg * SEG_FIELD_NR + field);
729 if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
730 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
731 vmx->segment_cache.bitmask = 0;
733 ret = vmx->segment_cache.bitmask & mask;
734 vmx->segment_cache.bitmask |= mask;
738 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
740 u16 *p = &vmx->segment_cache.seg[seg].selector;
742 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
743 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
747 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
749 ulong *p = &vmx->segment_cache.seg[seg].base;
751 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
752 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
756 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
758 u32 *p = &vmx->segment_cache.seg[seg].limit;
760 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
761 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
765 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
767 u32 *p = &vmx->segment_cache.seg[seg].ar;
769 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
770 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
774 void update_exception_bitmap(struct kvm_vcpu *vcpu)
778 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
779 (1u << DB_VECTOR) | (1u << AC_VECTOR);
781 * Guest access to VMware backdoor ports could legitimately
782 * trigger #GP because of TSS I/O permission bitmap.
783 * We intercept those #GP and allow access to them anyway
786 if (enable_vmware_backdoor)
787 eb |= (1u << GP_VECTOR);
788 if ((vcpu->guest_debug &
789 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
790 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
791 eb |= 1u << BP_VECTOR;
792 if (to_vmx(vcpu)->rmode.vm86_active)
795 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
797 /* When we are running a nested L2 guest and L1 specified for it a
798 * certain exception bitmap, we must trap the same exceptions and pass
799 * them to L1. When running L2, we will only handle the exceptions
800 * specified above if L1 did not want them.
802 if (is_guest_mode(vcpu))
803 eb |= get_vmcs12(vcpu)->exception_bitmap;
805 vmcs_write32(EXCEPTION_BITMAP, eb);
809 * Check if MSR is intercepted for currently loaded MSR bitmap.
811 static bool msr_write_intercepted(struct kvm_vcpu *vcpu, u32 msr)
813 unsigned long *msr_bitmap;
814 int f = sizeof(unsigned long);
816 if (!cpu_has_vmx_msr_bitmap())
819 msr_bitmap = to_vmx(vcpu)->loaded_vmcs->msr_bitmap;
822 return !!test_bit(msr, msr_bitmap + 0x800 / f);
823 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
825 return !!test_bit(msr, msr_bitmap + 0xc00 / f);
831 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
832 unsigned long entry, unsigned long exit)
834 vm_entry_controls_clearbit(vmx, entry);
835 vm_exit_controls_clearbit(vmx, exit);
838 static int find_msr(struct vmx_msrs *m, unsigned int msr)
842 for (i = 0; i < m->nr; ++i) {
843 if (m->val[i].index == msr)
849 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
852 struct msr_autoload *m = &vmx->msr_autoload;
856 if (cpu_has_load_ia32_efer()) {
857 clear_atomic_switch_msr_special(vmx,
858 VM_ENTRY_LOAD_IA32_EFER,
859 VM_EXIT_LOAD_IA32_EFER);
863 case MSR_CORE_PERF_GLOBAL_CTRL:
864 if (cpu_has_load_perf_global_ctrl()) {
865 clear_atomic_switch_msr_special(vmx,
866 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
867 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
872 i = find_msr(&m->guest, msr);
876 m->guest.val[i] = m->guest.val[m->guest.nr];
877 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
880 i = find_msr(&m->host, msr);
885 m->host.val[i] = m->host.val[m->host.nr];
886 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
889 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
890 unsigned long entry, unsigned long exit,
891 unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
892 u64 guest_val, u64 host_val)
894 vmcs_write64(guest_val_vmcs, guest_val);
895 if (host_val_vmcs != HOST_IA32_EFER)
896 vmcs_write64(host_val_vmcs, host_val);
897 vm_entry_controls_setbit(vmx, entry);
898 vm_exit_controls_setbit(vmx, exit);
901 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
902 u64 guest_val, u64 host_val, bool entry_only)
905 struct msr_autoload *m = &vmx->msr_autoload;
909 if (cpu_has_load_ia32_efer()) {
910 add_atomic_switch_msr_special(vmx,
911 VM_ENTRY_LOAD_IA32_EFER,
912 VM_EXIT_LOAD_IA32_EFER,
915 guest_val, host_val);
919 case MSR_CORE_PERF_GLOBAL_CTRL:
920 if (cpu_has_load_perf_global_ctrl()) {
921 add_atomic_switch_msr_special(vmx,
922 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
923 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
924 GUEST_IA32_PERF_GLOBAL_CTRL,
925 HOST_IA32_PERF_GLOBAL_CTRL,
926 guest_val, host_val);
930 case MSR_IA32_PEBS_ENABLE:
931 /* PEBS needs a quiescent period after being disabled (to write
932 * a record). Disabling PEBS through VMX MSR swapping doesn't
933 * provide that period, so a CPU could write host's record into
936 wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
939 i = find_msr(&m->guest, msr);
941 j = find_msr(&m->host, msr);
943 if ((i < 0 && m->guest.nr == NR_AUTOLOAD_MSRS) ||
944 (j < 0 && m->host.nr == NR_AUTOLOAD_MSRS)) {
945 printk_once(KERN_WARNING "Not enough msr switch entries. "
946 "Can't add msr %x\n", msr);
951 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
953 m->guest.val[i].index = msr;
954 m->guest.val[i].value = guest_val;
961 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
963 m->host.val[j].index = msr;
964 m->host.val[j].value = host_val;
967 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
969 u64 guest_efer = vmx->vcpu.arch.efer;
972 /* Shadow paging assumes NX to be available. */
974 guest_efer |= EFER_NX;
977 * LMA and LME handled by hardware; SCE meaningless outside long mode.
979 ignore_bits |= EFER_SCE;
981 ignore_bits |= EFER_LMA | EFER_LME;
982 /* SCE is meaningful only in long mode on Intel */
983 if (guest_efer & EFER_LMA)
984 ignore_bits &= ~(u64)EFER_SCE;
988 * On EPT, we can't emulate NX, so we must switch EFER atomically.
989 * On CPUs that support "load IA32_EFER", always switch EFER
990 * atomically, since it's faster than switching it manually.
992 if (cpu_has_load_ia32_efer() ||
993 (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
994 if (!(guest_efer & EFER_LMA))
995 guest_efer &= ~EFER_LME;
996 if (guest_efer != host_efer)
997 add_atomic_switch_msr(vmx, MSR_EFER,
998 guest_efer, host_efer, false);
1000 clear_atomic_switch_msr(vmx, MSR_EFER);
1003 clear_atomic_switch_msr(vmx, MSR_EFER);
1005 guest_efer &= ~ignore_bits;
1006 guest_efer |= host_efer & ignore_bits;
1008 vmx->guest_msrs[efer_offset].data = guest_efer;
1009 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1015 #ifdef CONFIG_X86_32
1017 * On 32-bit kernels, VM exits still load the FS and GS bases from the
1018 * VMCS rather than the segment table. KVM uses this helper to figure
1019 * out the current bases to poke them into the VMCS before entry.
1021 static unsigned long segment_base(u16 selector)
1023 struct desc_struct *table;
1026 if (!(selector & ~SEGMENT_RPL_MASK))
1029 table = get_current_gdt_ro();
1031 if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
1032 u16 ldt_selector = kvm_read_ldt();
1034 if (!(ldt_selector & ~SEGMENT_RPL_MASK))
1037 table = (struct desc_struct *)segment_base(ldt_selector);
1039 v = get_desc_base(&table[selector >> 3]);
1044 static inline void pt_load_msr(struct pt_ctx *ctx, u32 addr_range)
1048 wrmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1049 wrmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1050 wrmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1051 wrmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1052 for (i = 0; i < addr_range; i++) {
1053 wrmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1054 wrmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1058 static inline void pt_save_msr(struct pt_ctx *ctx, u32 addr_range)
1062 rdmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1063 rdmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1064 rdmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1065 rdmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1066 for (i = 0; i < addr_range; i++) {
1067 rdmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1068 rdmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1072 static void pt_guest_enter(struct vcpu_vmx *vmx)
1074 if (pt_mode == PT_MODE_SYSTEM)
1078 * GUEST_IA32_RTIT_CTL is already set in the VMCS.
1079 * Save host state before VM entry.
1081 rdmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1082 if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1083 wrmsrl(MSR_IA32_RTIT_CTL, 0);
1084 pt_save_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1085 pt_load_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1089 static void pt_guest_exit(struct vcpu_vmx *vmx)
1091 if (pt_mode == PT_MODE_SYSTEM)
1094 if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1095 pt_save_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1096 pt_load_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1099 /* Reload host state (IA32_RTIT_CTL will be cleared on VM exit). */
1100 wrmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1103 void vmx_set_host_fs_gs(struct vmcs_host_state *host, u16 fs_sel, u16 gs_sel,
1104 unsigned long fs_base, unsigned long gs_base)
1106 if (unlikely(fs_sel != host->fs_sel)) {
1108 vmcs_write16(HOST_FS_SELECTOR, fs_sel);
1110 vmcs_write16(HOST_FS_SELECTOR, 0);
1111 host->fs_sel = fs_sel;
1113 if (unlikely(gs_sel != host->gs_sel)) {
1115 vmcs_write16(HOST_GS_SELECTOR, gs_sel);
1117 vmcs_write16(HOST_GS_SELECTOR, 0);
1118 host->gs_sel = gs_sel;
1120 if (unlikely(fs_base != host->fs_base)) {
1121 vmcs_writel(HOST_FS_BASE, fs_base);
1122 host->fs_base = fs_base;
1124 if (unlikely(gs_base != host->gs_base)) {
1125 vmcs_writel(HOST_GS_BASE, gs_base);
1126 host->gs_base = gs_base;
1130 void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu)
1132 struct vcpu_vmx *vmx = to_vmx(vcpu);
1133 struct vmcs_host_state *host_state;
1134 #ifdef CONFIG_X86_64
1135 int cpu = raw_smp_processor_id();
1137 unsigned long fs_base, gs_base;
1141 vmx->req_immediate_exit = false;
1144 * Note that guest MSRs to be saved/restored can also be changed
1145 * when guest state is loaded. This happens when guest transitions
1146 * to/from long-mode by setting MSR_EFER.LMA.
1148 if (!vmx->guest_msrs_ready) {
1149 vmx->guest_msrs_ready = true;
1150 for (i = 0; i < vmx->save_nmsrs; ++i)
1151 kvm_set_shared_msr(vmx->guest_msrs[i].index,
1152 vmx->guest_msrs[i].data,
1153 vmx->guest_msrs[i].mask);
1156 if (vmx->guest_state_loaded)
1159 host_state = &vmx->loaded_vmcs->host_state;
1162 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
1163 * allow segment selectors with cpl > 0 or ti == 1.
1165 host_state->ldt_sel = kvm_read_ldt();
1167 #ifdef CONFIG_X86_64
1168 savesegment(ds, host_state->ds_sel);
1169 savesegment(es, host_state->es_sel);
1171 gs_base = cpu_kernelmode_gs_base(cpu);
1172 if (likely(is_64bit_mm(current->mm))) {
1173 save_fsgs_for_kvm();
1174 fs_sel = current->thread.fsindex;
1175 gs_sel = current->thread.gsindex;
1176 fs_base = current->thread.fsbase;
1177 vmx->msr_host_kernel_gs_base = current->thread.gsbase;
1179 savesegment(fs, fs_sel);
1180 savesegment(gs, gs_sel);
1181 fs_base = read_msr(MSR_FS_BASE);
1182 vmx->msr_host_kernel_gs_base = read_msr(MSR_KERNEL_GS_BASE);
1185 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1187 savesegment(fs, fs_sel);
1188 savesegment(gs, gs_sel);
1189 fs_base = segment_base(fs_sel);
1190 gs_base = segment_base(gs_sel);
1193 vmx_set_host_fs_gs(host_state, fs_sel, gs_sel, fs_base, gs_base);
1194 vmx->guest_state_loaded = true;
1197 static void vmx_prepare_switch_to_host(struct vcpu_vmx *vmx)
1199 struct vmcs_host_state *host_state;
1201 if (!vmx->guest_state_loaded)
1204 host_state = &vmx->loaded_vmcs->host_state;
1206 ++vmx->vcpu.stat.host_state_reload;
1208 #ifdef CONFIG_X86_64
1209 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1211 if (host_state->ldt_sel || (host_state->gs_sel & 7)) {
1212 kvm_load_ldt(host_state->ldt_sel);
1213 #ifdef CONFIG_X86_64
1214 load_gs_index(host_state->gs_sel);
1216 loadsegment(gs, host_state->gs_sel);
1219 if (host_state->fs_sel & 7)
1220 loadsegment(fs, host_state->fs_sel);
1221 #ifdef CONFIG_X86_64
1222 if (unlikely(host_state->ds_sel | host_state->es_sel)) {
1223 loadsegment(ds, host_state->ds_sel);
1224 loadsegment(es, host_state->es_sel);
1227 invalidate_tss_limit();
1228 #ifdef CONFIG_X86_64
1229 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1231 load_fixmap_gdt(raw_smp_processor_id());
1232 vmx->guest_state_loaded = false;
1233 vmx->guest_msrs_ready = false;
1236 #ifdef CONFIG_X86_64
1237 static u64 vmx_read_guest_kernel_gs_base(struct vcpu_vmx *vmx)
1240 if (vmx->guest_state_loaded)
1241 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1243 return vmx->msr_guest_kernel_gs_base;
1246 static void vmx_write_guest_kernel_gs_base(struct vcpu_vmx *vmx, u64 data)
1249 if (vmx->guest_state_loaded)
1250 wrmsrl(MSR_KERNEL_GS_BASE, data);
1252 vmx->msr_guest_kernel_gs_base = data;
1256 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
1258 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1259 struct pi_desc old, new;
1263 * In case of hot-plug or hot-unplug, we may have to undo
1264 * vmx_vcpu_pi_put even if there is no assigned device. And we
1265 * always keep PI.NDST up to date for simplicity: it makes the
1266 * code easier, and CPU migration is not a fast path.
1268 if (!pi_test_sn(pi_desc) && vcpu->cpu == cpu)
1272 * If the 'nv' field is POSTED_INTR_WAKEUP_VECTOR, do not change
1273 * PI.NDST: pi_post_block is the one expected to change PID.NDST and the
1274 * wakeup handler expects the vCPU to be on the blocked_vcpu_list that
1275 * matches PI.NDST. Otherwise, a vcpu may not be able to be woken up
1278 if (pi_desc->nv == POSTED_INTR_WAKEUP_VECTOR || vcpu->cpu == cpu) {
1279 pi_clear_sn(pi_desc);
1280 goto after_clear_sn;
1283 /* The full case. */
1285 old.control = new.control = pi_desc->control;
1287 dest = cpu_physical_id(cpu);
1289 if (x2apic_enabled())
1292 new.ndst = (dest << 8) & 0xFF00;
1295 } while (cmpxchg64(&pi_desc->control, old.control,
1296 new.control) != old.control);
1301 * Clear SN before reading the bitmap. The VT-d firmware
1302 * writes the bitmap and reads SN atomically (5.2.3 in the
1303 * spec), so it doesn't really have a memory barrier that
1304 * pairs with this, but we cannot do that and we need one.
1306 smp_mb__after_atomic();
1308 if (!pi_is_pir_empty(pi_desc))
1312 void vmx_vcpu_load_vmcs(struct kvm_vcpu *vcpu, int cpu)
1314 struct vcpu_vmx *vmx = to_vmx(vcpu);
1315 bool already_loaded = vmx->loaded_vmcs->cpu == cpu;
1317 if (!already_loaded) {
1318 loaded_vmcs_clear(vmx->loaded_vmcs);
1319 local_irq_disable();
1320 crash_disable_local_vmclear(cpu);
1323 * Read loaded_vmcs->cpu should be before fetching
1324 * loaded_vmcs->loaded_vmcss_on_cpu_link.
1325 * See the comments in __loaded_vmcs_clear().
1329 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1330 &per_cpu(loaded_vmcss_on_cpu, cpu));
1331 crash_enable_local_vmclear(cpu);
1335 if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
1336 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1337 vmcs_load(vmx->loaded_vmcs->vmcs);
1338 indirect_branch_prediction_barrier();
1341 if (!already_loaded) {
1342 void *gdt = get_current_gdt_ro();
1343 unsigned long sysenter_esp;
1345 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1348 * Linux uses per-cpu TSS and GDT, so set these when switching
1349 * processors. See 22.2.4.
1351 vmcs_writel(HOST_TR_BASE,
1352 (unsigned long)&get_cpu_entry_area(cpu)->tss.x86_tss);
1353 vmcs_writel(HOST_GDTR_BASE, (unsigned long)gdt); /* 22.2.4 */
1356 * VM exits change the host TR limit to 0x67 after a VM
1357 * exit. This is okay, since 0x67 covers everything except
1358 * the IO bitmap and have have code to handle the IO bitmap
1359 * being lost after a VM exit.
1361 BUILD_BUG_ON(IO_BITMAP_OFFSET - 1 != 0x67);
1363 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1364 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1366 vmx->loaded_vmcs->cpu = cpu;
1369 /* Setup TSC multiplier */
1370 if (kvm_has_tsc_control &&
1371 vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio)
1372 decache_tsc_multiplier(vmx);
1376 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1377 * vcpu mutex is already taken.
1379 void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1381 struct vcpu_vmx *vmx = to_vmx(vcpu);
1383 vmx_vcpu_load_vmcs(vcpu, cpu);
1385 vmx_vcpu_pi_load(vcpu, cpu);
1387 vmx->host_pkru = read_pkru();
1388 vmx->host_debugctlmsr = get_debugctlmsr();
1391 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
1393 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1395 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
1396 !irq_remapping_cap(IRQ_POSTING_CAP) ||
1397 !kvm_vcpu_apicv_active(vcpu))
1400 /* Set SN when the vCPU is preempted */
1401 if (vcpu->preempted)
1405 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1407 vmx_vcpu_pi_put(vcpu);
1409 vmx_prepare_switch_to_host(to_vmx(vcpu));
1412 static bool emulation_required(struct kvm_vcpu *vcpu)
1414 return emulate_invalid_guest_state && !guest_state_valid(vcpu);
1417 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
1419 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1421 unsigned long rflags, save_rflags;
1423 if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
1424 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1425 rflags = vmcs_readl(GUEST_RFLAGS);
1426 if (to_vmx(vcpu)->rmode.vm86_active) {
1427 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1428 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
1429 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1431 to_vmx(vcpu)->rflags = rflags;
1433 return to_vmx(vcpu)->rflags;
1436 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1438 unsigned long old_rflags = vmx_get_rflags(vcpu);
1440 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1441 to_vmx(vcpu)->rflags = rflags;
1442 if (to_vmx(vcpu)->rmode.vm86_active) {
1443 to_vmx(vcpu)->rmode.save_rflags = rflags;
1444 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1446 vmcs_writel(GUEST_RFLAGS, rflags);
1448 if ((old_rflags ^ to_vmx(vcpu)->rflags) & X86_EFLAGS_VM)
1449 to_vmx(vcpu)->emulation_required = emulation_required(vcpu);
1452 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
1454 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1457 if (interruptibility & GUEST_INTR_STATE_STI)
1458 ret |= KVM_X86_SHADOW_INT_STI;
1459 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1460 ret |= KVM_X86_SHADOW_INT_MOV_SS;
1465 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1467 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1468 u32 interruptibility = interruptibility_old;
1470 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1472 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1473 interruptibility |= GUEST_INTR_STATE_MOV_SS;
1474 else if (mask & KVM_X86_SHADOW_INT_STI)
1475 interruptibility |= GUEST_INTR_STATE_STI;
1477 if ((interruptibility != interruptibility_old))
1478 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1481 static int vmx_rtit_ctl_check(struct kvm_vcpu *vcpu, u64 data)
1483 struct vcpu_vmx *vmx = to_vmx(vcpu);
1484 unsigned long value;
1487 * Any MSR write that attempts to change bits marked reserved will
1490 if (data & vmx->pt_desc.ctl_bitmask)
1494 * Any attempt to modify IA32_RTIT_CTL while TraceEn is set will
1495 * result in a #GP unless the same write also clears TraceEn.
1497 if ((vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) &&
1498 ((vmx->pt_desc.guest.ctl ^ data) & ~RTIT_CTL_TRACEEN))
1502 * WRMSR to IA32_RTIT_CTL that sets TraceEn but clears this bit
1503 * and FabricEn would cause #GP, if
1504 * CPUID.(EAX=14H, ECX=0):ECX.SNGLRGNOUT[bit 2] = 0
1506 if ((data & RTIT_CTL_TRACEEN) && !(data & RTIT_CTL_TOPA) &&
1507 !(data & RTIT_CTL_FABRIC_EN) &&
1508 !intel_pt_validate_cap(vmx->pt_desc.caps,
1509 PT_CAP_single_range_output))
1513 * MTCFreq, CycThresh and PSBFreq encodings check, any MSR write that
1514 * utilize encodings marked reserved will casue a #GP fault.
1516 value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc_periods);
1517 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc) &&
1518 !test_bit((data & RTIT_CTL_MTC_RANGE) >>
1519 RTIT_CTL_MTC_RANGE_OFFSET, &value))
1521 value = intel_pt_validate_cap(vmx->pt_desc.caps,
1522 PT_CAP_cycle_thresholds);
1523 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1524 !test_bit((data & RTIT_CTL_CYC_THRESH) >>
1525 RTIT_CTL_CYC_THRESH_OFFSET, &value))
1527 value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_periods);
1528 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1529 !test_bit((data & RTIT_CTL_PSB_FREQ) >>
1530 RTIT_CTL_PSB_FREQ_OFFSET, &value))
1534 * If ADDRx_CFG is reserved or the encodings is >2 will
1535 * cause a #GP fault.
1537 value = (data & RTIT_CTL_ADDR0) >> RTIT_CTL_ADDR0_OFFSET;
1538 if ((value && (vmx->pt_desc.addr_range < 1)) || (value > 2))
1540 value = (data & RTIT_CTL_ADDR1) >> RTIT_CTL_ADDR1_OFFSET;
1541 if ((value && (vmx->pt_desc.addr_range < 2)) || (value > 2))
1543 value = (data & RTIT_CTL_ADDR2) >> RTIT_CTL_ADDR2_OFFSET;
1544 if ((value && (vmx->pt_desc.addr_range < 3)) || (value > 2))
1546 value = (data & RTIT_CTL_ADDR3) >> RTIT_CTL_ADDR3_OFFSET;
1547 if ((value && (vmx->pt_desc.addr_range < 4)) || (value > 2))
1553 static int skip_emulated_instruction(struct kvm_vcpu *vcpu)
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 rip = kvm_rip_read(vcpu);
1568 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1569 kvm_rip_write(vcpu, rip);
1571 if (!kvm_emulate_instruction(vcpu, EMULTYPE_SKIP))
1575 /* skipping an emulated instruction also counts */
1576 vmx_set_interrupt_shadow(vcpu, 0);
1581 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1584 * Ensure that we clear the HLT state in the VMCS. We don't need to
1585 * explicitly skip the instruction because if the HLT state is set,
1586 * then the instruction is already executing and RIP has already been
1589 if (kvm_hlt_in_guest(vcpu->kvm) &&
1590 vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1591 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1594 static void vmx_queue_exception(struct kvm_vcpu *vcpu)
1596 struct vcpu_vmx *vmx = to_vmx(vcpu);
1597 unsigned nr = vcpu->arch.exception.nr;
1598 bool has_error_code = vcpu->arch.exception.has_error_code;
1599 u32 error_code = vcpu->arch.exception.error_code;
1600 u32 intr_info = nr | INTR_INFO_VALID_MASK;
1602 kvm_deliver_exception_payload(vcpu);
1604 if (has_error_code) {
1605 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1606 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1609 if (vmx->rmode.vm86_active) {
1611 if (kvm_exception_is_soft(nr))
1612 inc_eip = vcpu->arch.event_exit_inst_len;
1613 kvm_inject_realmode_interrupt(vcpu, nr, inc_eip);
1617 WARN_ON_ONCE(vmx->emulation_required);
1619 if (kvm_exception_is_soft(nr)) {
1620 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1621 vmx->vcpu.arch.event_exit_inst_len);
1622 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1624 intr_info |= INTR_TYPE_HARD_EXCEPTION;
1626 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1628 vmx_clear_hlt(vcpu);
1631 static bool vmx_rdtscp_supported(void)
1633 return cpu_has_vmx_rdtscp();
1636 static bool vmx_invpcid_supported(void)
1638 return cpu_has_vmx_invpcid();
1642 * Swap MSR entry in host/guest MSR entry array.
1644 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1646 struct shared_msr_entry tmp;
1648 tmp = vmx->guest_msrs[to];
1649 vmx->guest_msrs[to] = vmx->guest_msrs[from];
1650 vmx->guest_msrs[from] = tmp;
1654 * Set up the vmcs to automatically save and restore system
1655 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
1656 * mode, as fiddling with msrs is very expensive.
1658 static void setup_msrs(struct vcpu_vmx *vmx)
1660 int save_nmsrs, index;
1663 #ifdef CONFIG_X86_64
1665 * The SYSCALL MSRs are only needed on long mode guests, and only
1666 * when EFER.SCE is set.
1668 if (is_long_mode(&vmx->vcpu) && (vmx->vcpu.arch.efer & EFER_SCE)) {
1669 index = __find_msr_index(vmx, MSR_STAR);
1671 move_msr_up(vmx, index, save_nmsrs++);
1672 index = __find_msr_index(vmx, MSR_LSTAR);
1674 move_msr_up(vmx, index, save_nmsrs++);
1675 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1677 move_msr_up(vmx, index, save_nmsrs++);
1680 index = __find_msr_index(vmx, MSR_EFER);
1681 if (index >= 0 && update_transition_efer(vmx, index))
1682 move_msr_up(vmx, index, save_nmsrs++);
1683 index = __find_msr_index(vmx, MSR_TSC_AUX);
1684 if (index >= 0 && guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDTSCP))
1685 move_msr_up(vmx, index, save_nmsrs++);
1687 vmx->save_nmsrs = save_nmsrs;
1688 vmx->guest_msrs_ready = false;
1690 if (cpu_has_vmx_msr_bitmap())
1691 vmx_update_msr_bitmap(&vmx->vcpu);
1694 static u64 vmx_read_l1_tsc_offset(struct kvm_vcpu *vcpu)
1696 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1698 if (is_guest_mode(vcpu) &&
1699 (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING))
1700 return vcpu->arch.tsc_offset - vmcs12->tsc_offset;
1702 return vcpu->arch.tsc_offset;
1705 static u64 vmx_write_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1707 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1708 u64 g_tsc_offset = 0;
1711 * We're here if L1 chose not to trap WRMSR to TSC. According
1712 * to the spec, this should set L1's TSC; The offset that L1
1713 * set for L2 remains unchanged, and still needs to be added
1714 * to the newly set TSC to get L2's TSC.
1716 if (is_guest_mode(vcpu) &&
1717 (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING))
1718 g_tsc_offset = vmcs12->tsc_offset;
1720 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1721 vcpu->arch.tsc_offset - g_tsc_offset,
1723 vmcs_write64(TSC_OFFSET, offset + g_tsc_offset);
1724 return offset + g_tsc_offset;
1728 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
1729 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
1730 * all guests if the "nested" module option is off, and can also be disabled
1731 * for a single guest by disabling its VMX cpuid bit.
1733 bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
1735 return nested && guest_cpuid_has(vcpu, X86_FEATURE_VMX);
1738 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
1741 uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
1743 return !(val & ~valid_bits);
1746 static int vmx_get_msr_feature(struct kvm_msr_entry *msr)
1748 switch (msr->index) {
1749 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1752 return vmx_get_vmx_msr(&vmcs_config.nested, msr->index, &msr->data);
1761 * Reads an msr value (of 'msr_index') into 'pdata'.
1762 * Returns 0 on success, non-0 otherwise.
1763 * Assumes vcpu_load() was already called.
1765 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1767 struct vcpu_vmx *vmx = to_vmx(vcpu);
1768 struct shared_msr_entry *msr;
1771 switch (msr_info->index) {
1772 #ifdef CONFIG_X86_64
1774 msr_info->data = vmcs_readl(GUEST_FS_BASE);
1777 msr_info->data = vmcs_readl(GUEST_GS_BASE);
1779 case MSR_KERNEL_GS_BASE:
1780 msr_info->data = vmx_read_guest_kernel_gs_base(vmx);
1784 return kvm_get_msr_common(vcpu, msr_info);
1785 case MSR_IA32_UMWAIT_CONTROL:
1786 if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
1789 msr_info->data = vmx->msr_ia32_umwait_control;
1791 case MSR_IA32_SPEC_CTRL:
1792 if (!msr_info->host_initiated &&
1793 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1796 msr_info->data = to_vmx(vcpu)->spec_ctrl;
1798 case MSR_IA32_SYSENTER_CS:
1799 msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
1801 case MSR_IA32_SYSENTER_EIP:
1802 msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
1804 case MSR_IA32_SYSENTER_ESP:
1805 msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
1807 case MSR_IA32_BNDCFGS:
1808 if (!kvm_mpx_supported() ||
1809 (!msr_info->host_initiated &&
1810 !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1812 msr_info->data = vmcs_read64(GUEST_BNDCFGS);
1814 case MSR_IA32_MCG_EXT_CTL:
1815 if (!msr_info->host_initiated &&
1816 !(vmx->msr_ia32_feature_control &
1817 FEATURE_CONTROL_LMCE))
1819 msr_info->data = vcpu->arch.mcg_ext_ctl;
1821 case MSR_IA32_FEATURE_CONTROL:
1822 msr_info->data = vmx->msr_ia32_feature_control;
1824 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1825 if (!nested_vmx_allowed(vcpu))
1827 return vmx_get_vmx_msr(&vmx->nested.msrs, msr_info->index,
1830 if (!vmx_xsaves_supported() ||
1831 (!msr_info->host_initiated &&
1832 !(guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
1833 guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))))
1835 msr_info->data = vcpu->arch.ia32_xss;
1837 case MSR_IA32_RTIT_CTL:
1838 if (pt_mode != PT_MODE_HOST_GUEST)
1840 msr_info->data = vmx->pt_desc.guest.ctl;
1842 case MSR_IA32_RTIT_STATUS:
1843 if (pt_mode != PT_MODE_HOST_GUEST)
1845 msr_info->data = vmx->pt_desc.guest.status;
1847 case MSR_IA32_RTIT_CR3_MATCH:
1848 if ((pt_mode != PT_MODE_HOST_GUEST) ||
1849 !intel_pt_validate_cap(vmx->pt_desc.caps,
1850 PT_CAP_cr3_filtering))
1852 msr_info->data = vmx->pt_desc.guest.cr3_match;
1854 case MSR_IA32_RTIT_OUTPUT_BASE:
1855 if ((pt_mode != PT_MODE_HOST_GUEST) ||
1856 (!intel_pt_validate_cap(vmx->pt_desc.caps,
1857 PT_CAP_topa_output) &&
1858 !intel_pt_validate_cap(vmx->pt_desc.caps,
1859 PT_CAP_single_range_output)))
1861 msr_info->data = vmx->pt_desc.guest.output_base;
1863 case MSR_IA32_RTIT_OUTPUT_MASK:
1864 if ((pt_mode != PT_MODE_HOST_GUEST) ||
1865 (!intel_pt_validate_cap(vmx->pt_desc.caps,
1866 PT_CAP_topa_output) &&
1867 !intel_pt_validate_cap(vmx->pt_desc.caps,
1868 PT_CAP_single_range_output)))
1870 msr_info->data = vmx->pt_desc.guest.output_mask;
1872 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
1873 index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
1874 if ((pt_mode != PT_MODE_HOST_GUEST) ||
1875 (index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
1876 PT_CAP_num_address_ranges)))
1879 msr_info->data = vmx->pt_desc.guest.addr_b[index / 2];
1881 msr_info->data = vmx->pt_desc.guest.addr_a[index / 2];
1884 if (!msr_info->host_initiated &&
1885 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
1887 /* Else, falls through */
1889 msr = find_msr_entry(vmx, msr_info->index);
1891 msr_info->data = msr->data;
1894 return kvm_get_msr_common(vcpu, msr_info);
1901 * Writes msr value into into the appropriate "register".
1902 * Returns 0 on success, non-0 otherwise.
1903 * Assumes vcpu_load() was already called.
1905 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1907 struct vcpu_vmx *vmx = to_vmx(vcpu);
1908 struct shared_msr_entry *msr;
1910 u32 msr_index = msr_info->index;
1911 u64 data = msr_info->data;
1914 switch (msr_index) {
1916 ret = kvm_set_msr_common(vcpu, msr_info);
1918 #ifdef CONFIG_X86_64
1920 vmx_segment_cache_clear(vmx);
1921 vmcs_writel(GUEST_FS_BASE, data);
1924 vmx_segment_cache_clear(vmx);
1925 vmcs_writel(GUEST_GS_BASE, data);
1927 case MSR_KERNEL_GS_BASE:
1928 vmx_write_guest_kernel_gs_base(vmx, data);
1931 case MSR_IA32_SYSENTER_CS:
1932 if (is_guest_mode(vcpu))
1933 get_vmcs12(vcpu)->guest_sysenter_cs = data;
1934 vmcs_write32(GUEST_SYSENTER_CS, data);
1936 case MSR_IA32_SYSENTER_EIP:
1937 if (is_guest_mode(vcpu))
1938 get_vmcs12(vcpu)->guest_sysenter_eip = data;
1939 vmcs_writel(GUEST_SYSENTER_EIP, data);
1941 case MSR_IA32_SYSENTER_ESP:
1942 if (is_guest_mode(vcpu))
1943 get_vmcs12(vcpu)->guest_sysenter_esp = data;
1944 vmcs_writel(GUEST_SYSENTER_ESP, data);
1946 case MSR_IA32_DEBUGCTLMSR:
1947 if (is_guest_mode(vcpu) && get_vmcs12(vcpu)->vm_exit_controls &
1948 VM_EXIT_SAVE_DEBUG_CONTROLS)
1949 get_vmcs12(vcpu)->guest_ia32_debugctl = data;
1951 ret = kvm_set_msr_common(vcpu, msr_info);
1954 case MSR_IA32_BNDCFGS:
1955 if (!kvm_mpx_supported() ||
1956 (!msr_info->host_initiated &&
1957 !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1959 if (is_noncanonical_address(data & PAGE_MASK, vcpu) ||
1960 (data & MSR_IA32_BNDCFGS_RSVD))
1962 vmcs_write64(GUEST_BNDCFGS, data);
1964 case MSR_IA32_UMWAIT_CONTROL:
1965 if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
1968 /* The reserved bit 1 and non-32 bit [63:32] should be zero */
1969 if (data & (BIT_ULL(1) | GENMASK_ULL(63, 32)))
1972 vmx->msr_ia32_umwait_control = data;
1974 case MSR_IA32_SPEC_CTRL:
1975 if (!msr_info->host_initiated &&
1976 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1979 /* The STIBP bit doesn't fault even if it's not advertised */
1980 if (data & ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP | SPEC_CTRL_SSBD))
1983 vmx->spec_ctrl = data;
1990 * When it's written (to non-zero) for the first time, pass
1994 * The handling of the MSR bitmap for L2 guests is done in
1995 * nested_vmx_merge_msr_bitmap. We should not touch the
1996 * vmcs02.msr_bitmap here since it gets completely overwritten
1997 * in the merging. We update the vmcs01 here for L1 as well
1998 * since it will end up touching the MSR anyway now.
2000 vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap,
2004 case MSR_IA32_PRED_CMD:
2005 if (!msr_info->host_initiated &&
2006 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
2009 if (data & ~PRED_CMD_IBPB)
2015 wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
2019 * When it's written (to non-zero) for the first time, pass
2023 * The handling of the MSR bitmap for L2 guests is done in
2024 * nested_vmx_merge_msr_bitmap. We should not touch the
2025 * vmcs02.msr_bitmap here since it gets completely overwritten
2028 vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap, MSR_IA32_PRED_CMD,
2031 case MSR_IA32_CR_PAT:
2032 if (!kvm_pat_valid(data))
2035 if (is_guest_mode(vcpu) &&
2036 get_vmcs12(vcpu)->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
2037 get_vmcs12(vcpu)->guest_ia32_pat = data;
2039 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2040 vmcs_write64(GUEST_IA32_PAT, data);
2041 vcpu->arch.pat = data;
2044 ret = kvm_set_msr_common(vcpu, msr_info);
2046 case MSR_IA32_TSC_ADJUST:
2047 ret = kvm_set_msr_common(vcpu, msr_info);
2049 case MSR_IA32_MCG_EXT_CTL:
2050 if ((!msr_info->host_initiated &&
2051 !(to_vmx(vcpu)->msr_ia32_feature_control &
2052 FEATURE_CONTROL_LMCE)) ||
2053 (data & ~MCG_EXT_CTL_LMCE_EN))
2055 vcpu->arch.mcg_ext_ctl = data;
2057 case MSR_IA32_FEATURE_CONTROL:
2058 if (!vmx_feature_control_msr_valid(vcpu, data) ||
2059 (to_vmx(vcpu)->msr_ia32_feature_control &
2060 FEATURE_CONTROL_LOCKED && !msr_info->host_initiated))
2062 vmx->msr_ia32_feature_control = data;
2063 if (msr_info->host_initiated && data == 0)
2064 vmx_leave_nested(vcpu);
2066 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2067 if (!msr_info->host_initiated)
2068 return 1; /* they are read-only */
2069 if (!nested_vmx_allowed(vcpu))
2071 return vmx_set_vmx_msr(vcpu, msr_index, data);
2073 if (!vmx_xsaves_supported() ||
2074 (!msr_info->host_initiated &&
2075 !(guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
2076 guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))))
2079 * The only supported bit as of Skylake is bit 8, but
2080 * it is not supported on KVM.
2084 vcpu->arch.ia32_xss = data;
2085 if (vcpu->arch.ia32_xss != host_xss)
2086 add_atomic_switch_msr(vmx, MSR_IA32_XSS,
2087 vcpu->arch.ia32_xss, host_xss, false);
2089 clear_atomic_switch_msr(vmx, MSR_IA32_XSS);
2091 case MSR_IA32_RTIT_CTL:
2092 if ((pt_mode != PT_MODE_HOST_GUEST) ||
2093 vmx_rtit_ctl_check(vcpu, data) ||
2096 vmcs_write64(GUEST_IA32_RTIT_CTL, data);
2097 vmx->pt_desc.guest.ctl = data;
2098 pt_update_intercept_for_msr(vmx);
2100 case MSR_IA32_RTIT_STATUS:
2101 if ((pt_mode != PT_MODE_HOST_GUEST) ||
2102 (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
2103 (data & MSR_IA32_RTIT_STATUS_MASK))
2105 vmx->pt_desc.guest.status = data;
2107 case MSR_IA32_RTIT_CR3_MATCH:
2108 if ((pt_mode != PT_MODE_HOST_GUEST) ||
2109 (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
2110 !intel_pt_validate_cap(vmx->pt_desc.caps,
2111 PT_CAP_cr3_filtering))
2113 vmx->pt_desc.guest.cr3_match = data;
2115 case MSR_IA32_RTIT_OUTPUT_BASE:
2116 if ((pt_mode != PT_MODE_HOST_GUEST) ||
2117 (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
2118 (!intel_pt_validate_cap(vmx->pt_desc.caps,
2119 PT_CAP_topa_output) &&
2120 !intel_pt_validate_cap(vmx->pt_desc.caps,
2121 PT_CAP_single_range_output)) ||
2122 (data & MSR_IA32_RTIT_OUTPUT_BASE_MASK))
2124 vmx->pt_desc.guest.output_base = data;
2126 case MSR_IA32_RTIT_OUTPUT_MASK:
2127 if ((pt_mode != PT_MODE_HOST_GUEST) ||
2128 (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
2129 (!intel_pt_validate_cap(vmx->pt_desc.caps,
2130 PT_CAP_topa_output) &&
2131 !intel_pt_validate_cap(vmx->pt_desc.caps,
2132 PT_CAP_single_range_output)))
2134 vmx->pt_desc.guest.output_mask = data;
2136 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
2137 index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
2138 if ((pt_mode != PT_MODE_HOST_GUEST) ||
2139 (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
2140 (index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
2141 PT_CAP_num_address_ranges)))
2143 if (is_noncanonical_address(data, vcpu))
2146 vmx->pt_desc.guest.addr_b[index / 2] = data;
2148 vmx->pt_desc.guest.addr_a[index / 2] = data;
2151 if (!msr_info->host_initiated &&
2152 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
2154 /* Check reserved bit, higher 32 bits should be zero */
2155 if ((data >> 32) != 0)
2157 /* Else, falls through */
2159 msr = find_msr_entry(vmx, msr_index);
2161 u64 old_msr_data = msr->data;
2163 if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
2165 ret = kvm_set_shared_msr(msr->index, msr->data,
2169 msr->data = old_msr_data;
2173 ret = kvm_set_msr_common(vcpu, msr_info);
2179 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2181 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
2184 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2187 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2189 case VCPU_EXREG_PDPTR:
2191 ept_save_pdptrs(vcpu);
2198 static __init int cpu_has_kvm_support(void)
2200 return cpu_has_vmx();
2203 static __init int vmx_disabled_by_bios(void)
2207 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
2208 if (msr & FEATURE_CONTROL_LOCKED) {
2209 /* launched w/ TXT and VMX disabled */
2210 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2213 /* launched w/o TXT and VMX only enabled w/ TXT */
2214 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2215 && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2216 && !tboot_enabled()) {
2217 printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
2218 "activate TXT before enabling KVM\n");
2221 /* launched w/o TXT and VMX disabled */
2222 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2223 && !tboot_enabled())
2230 static void kvm_cpu_vmxon(u64 addr)
2232 cr4_set_bits(X86_CR4_VMXE);
2233 intel_pt_handle_vmx(1);
2235 asm volatile ("vmxon %0" : : "m"(addr));
2238 static int hardware_enable(void)
2240 int cpu = raw_smp_processor_id();
2241 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2244 if (cr4_read_shadow() & X86_CR4_VMXE)
2248 * This can happen if we hot-added a CPU but failed to allocate
2249 * VP assist page for it.
2251 if (static_branch_unlikely(&enable_evmcs) &&
2252 !hv_get_vp_assist_page(cpu))
2255 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
2256 INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
2257 spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
2260 * Now we can enable the vmclear operation in kdump
2261 * since the loaded_vmcss_on_cpu list on this cpu
2262 * has been initialized.
2264 * Though the cpu is not in VMX operation now, there
2265 * is no problem to enable the vmclear operation
2266 * for the loaded_vmcss_on_cpu list is empty!
2268 crash_enable_local_vmclear(cpu);
2270 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
2272 test_bits = FEATURE_CONTROL_LOCKED;
2273 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
2274 if (tboot_enabled())
2275 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
2277 if ((old & test_bits) != test_bits) {
2278 /* enable and lock */
2279 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
2281 kvm_cpu_vmxon(phys_addr);
2288 static void vmclear_local_loaded_vmcss(void)
2290 int cpu = raw_smp_processor_id();
2291 struct loaded_vmcs *v, *n;
2293 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2294 loaded_vmcss_on_cpu_link)
2295 __loaded_vmcs_clear(v);
2299 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2302 static void kvm_cpu_vmxoff(void)
2304 asm volatile (__ex("vmxoff"));
2306 intel_pt_handle_vmx(0);
2307 cr4_clear_bits(X86_CR4_VMXE);
2310 static void hardware_disable(void)
2312 vmclear_local_loaded_vmcss();
2316 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2317 u32 msr, u32 *result)
2319 u32 vmx_msr_low, vmx_msr_high;
2320 u32 ctl = ctl_min | ctl_opt;
2322 rdmsr(msr, vmx_msr_low, vmx_msr_high);
2324 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2325 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
2327 /* Ensure minimum (required) set of control bits are supported. */
2335 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf,
2336 struct vmx_capability *vmx_cap)
2338 u32 vmx_msr_low, vmx_msr_high;
2339 u32 min, opt, min2, opt2;
2340 u32 _pin_based_exec_control = 0;
2341 u32 _cpu_based_exec_control = 0;
2342 u32 _cpu_based_2nd_exec_control = 0;
2343 u32 _vmexit_control = 0;
2344 u32 _vmentry_control = 0;
2346 memset(vmcs_conf, 0, sizeof(*vmcs_conf));
2347 min = CPU_BASED_HLT_EXITING |
2348 #ifdef CONFIG_X86_64
2349 CPU_BASED_CR8_LOAD_EXITING |
2350 CPU_BASED_CR8_STORE_EXITING |
2352 CPU_BASED_CR3_LOAD_EXITING |
2353 CPU_BASED_CR3_STORE_EXITING |
2354 CPU_BASED_UNCOND_IO_EXITING |
2355 CPU_BASED_MOV_DR_EXITING |
2356 CPU_BASED_USE_TSC_OFFSETING |
2357 CPU_BASED_MWAIT_EXITING |
2358 CPU_BASED_MONITOR_EXITING |
2359 CPU_BASED_INVLPG_EXITING |
2360 CPU_BASED_RDPMC_EXITING;
2362 opt = CPU_BASED_TPR_SHADOW |
2363 CPU_BASED_USE_MSR_BITMAPS |
2364 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2365 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2366 &_cpu_based_exec_control) < 0)
2368 #ifdef CONFIG_X86_64
2369 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2370 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2371 ~CPU_BASED_CR8_STORE_EXITING;
2373 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2375 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2376 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2377 SECONDARY_EXEC_WBINVD_EXITING |
2378 SECONDARY_EXEC_ENABLE_VPID |
2379 SECONDARY_EXEC_ENABLE_EPT |
2380 SECONDARY_EXEC_UNRESTRICTED_GUEST |
2381 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2382 SECONDARY_EXEC_DESC |
2383 SECONDARY_EXEC_RDTSCP |
2384 SECONDARY_EXEC_ENABLE_INVPCID |
2385 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2386 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2387 SECONDARY_EXEC_SHADOW_VMCS |
2388 SECONDARY_EXEC_XSAVES |
2389 SECONDARY_EXEC_RDSEED_EXITING |
2390 SECONDARY_EXEC_RDRAND_EXITING |
2391 SECONDARY_EXEC_ENABLE_PML |
2392 SECONDARY_EXEC_TSC_SCALING |
2393 SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE |
2394 SECONDARY_EXEC_PT_USE_GPA |
2395 SECONDARY_EXEC_PT_CONCEAL_VMX |
2396 SECONDARY_EXEC_ENABLE_VMFUNC |
2397 SECONDARY_EXEC_ENCLS_EXITING;
2398 if (adjust_vmx_controls(min2, opt2,
2399 MSR_IA32_VMX_PROCBASED_CTLS2,
2400 &_cpu_based_2nd_exec_control) < 0)
2403 #ifndef CONFIG_X86_64
2404 if (!(_cpu_based_2nd_exec_control &
2405 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2406 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2409 if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2410 _cpu_based_2nd_exec_control &= ~(
2411 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2412 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2413 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
2415 rdmsr_safe(MSR_IA32_VMX_EPT_VPID_CAP,
2416 &vmx_cap->ept, &vmx_cap->vpid);
2418 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2419 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2421 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2422 CPU_BASED_CR3_STORE_EXITING |
2423 CPU_BASED_INVLPG_EXITING);
2424 } else if (vmx_cap->ept) {
2426 pr_warn_once("EPT CAP should not exist if not support "
2427 "1-setting enable EPT VM-execution control\n");
2429 if (!(_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_VPID) &&
2432 pr_warn_once("VPID CAP should not exist if not support "
2433 "1-setting enable VPID VM-execution control\n");
2436 min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
2437 #ifdef CONFIG_X86_64
2438 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2440 opt = VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL |
2441 VM_EXIT_LOAD_IA32_PAT |
2442 VM_EXIT_LOAD_IA32_EFER |
2443 VM_EXIT_CLEAR_BNDCFGS |
2444 VM_EXIT_PT_CONCEAL_PIP |
2445 VM_EXIT_CLEAR_IA32_RTIT_CTL;
2446 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2447 &_vmexit_control) < 0)
2450 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2451 opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR |
2452 PIN_BASED_VMX_PREEMPTION_TIMER;
2453 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2454 &_pin_based_exec_control) < 0)
2457 if (cpu_has_broken_vmx_preemption_timer())
2458 _pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
2459 if (!(_cpu_based_2nd_exec_control &
2460 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
2461 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
2463 min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
2464 opt = VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL |
2465 VM_ENTRY_LOAD_IA32_PAT |
2466 VM_ENTRY_LOAD_IA32_EFER |
2467 VM_ENTRY_LOAD_BNDCFGS |
2468 VM_ENTRY_PT_CONCEAL_PIP |
2469 VM_ENTRY_LOAD_IA32_RTIT_CTL;
2470 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2471 &_vmentry_control) < 0)
2475 * Some cpus support VM_{ENTRY,EXIT}_IA32_PERF_GLOBAL_CTRL but they
2476 * can't be used due to an errata where VM Exit may incorrectly clear
2477 * IA32_PERF_GLOBAL_CTRL[34:32]. Workaround the errata by using the
2478 * MSR load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2480 if (boot_cpu_data.x86 == 0x6) {
2481 switch (boot_cpu_data.x86_model) {
2482 case 26: /* AAK155 */
2483 case 30: /* AAP115 */
2484 case 37: /* AAT100 */
2485 case 44: /* BC86,AAY89,BD102 */
2487 _vmentry_control &= ~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
2488 _vmexit_control &= ~VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
2489 pr_warn_once("kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
2490 "does not work properly. Using workaround\n");
2498 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2500 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2501 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2504 #ifdef CONFIG_X86_64
2505 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2506 if (vmx_msr_high & (1u<<16))
2510 /* Require Write-Back (WB) memory type for VMCS accesses. */
2511 if (((vmx_msr_high >> 18) & 15) != 6)
2514 vmcs_conf->size = vmx_msr_high & 0x1fff;
2515 vmcs_conf->order = get_order(vmcs_conf->size);
2516 vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
2518 vmcs_conf->revision_id = vmx_msr_low;
2520 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2521 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2522 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2523 vmcs_conf->vmexit_ctrl = _vmexit_control;
2524 vmcs_conf->vmentry_ctrl = _vmentry_control;
2526 if (static_branch_unlikely(&enable_evmcs))
2527 evmcs_sanitize_exec_ctrls(vmcs_conf);
2532 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags)
2534 int node = cpu_to_node(cpu);
2538 pages = __alloc_pages_node(node, flags, vmcs_config.order);
2541 vmcs = page_address(pages);
2542 memset(vmcs, 0, vmcs_config.size);
2544 /* KVM supports Enlightened VMCS v1 only */
2545 if (static_branch_unlikely(&enable_evmcs))
2546 vmcs->hdr.revision_id = KVM_EVMCS_VERSION;
2548 vmcs->hdr.revision_id = vmcs_config.revision_id;
2551 vmcs->hdr.shadow_vmcs = 1;
2555 void free_vmcs(struct vmcs *vmcs)
2557 free_pages((unsigned long)vmcs, vmcs_config.order);
2561 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2563 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2565 if (!loaded_vmcs->vmcs)
2567 loaded_vmcs_clear(loaded_vmcs);
2568 free_vmcs(loaded_vmcs->vmcs);
2569 loaded_vmcs->vmcs = NULL;
2570 if (loaded_vmcs->msr_bitmap)
2571 free_page((unsigned long)loaded_vmcs->msr_bitmap);
2572 WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
2575 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2577 loaded_vmcs->vmcs = alloc_vmcs(false);
2578 if (!loaded_vmcs->vmcs)
2581 loaded_vmcs->shadow_vmcs = NULL;
2582 loaded_vmcs->hv_timer_soft_disabled = false;
2583 loaded_vmcs_init(loaded_vmcs);
2585 if (cpu_has_vmx_msr_bitmap()) {
2586 loaded_vmcs->msr_bitmap = (unsigned long *)
2587 __get_free_page(GFP_KERNEL_ACCOUNT);
2588 if (!loaded_vmcs->msr_bitmap)
2590 memset(loaded_vmcs->msr_bitmap, 0xff, PAGE_SIZE);
2592 if (IS_ENABLED(CONFIG_HYPERV) &&
2593 static_branch_unlikely(&enable_evmcs) &&
2594 (ms_hyperv.nested_features & HV_X64_NESTED_MSR_BITMAP)) {
2595 struct hv_enlightened_vmcs *evmcs =
2596 (struct hv_enlightened_vmcs *)loaded_vmcs->vmcs;
2598 evmcs->hv_enlightenments_control.msr_bitmap = 1;
2602 memset(&loaded_vmcs->host_state, 0, sizeof(struct vmcs_host_state));
2603 memset(&loaded_vmcs->controls_shadow, 0,
2604 sizeof(struct vmcs_controls_shadow));
2609 free_loaded_vmcs(loaded_vmcs);
2613 static void free_kvm_area(void)
2617 for_each_possible_cpu(cpu) {
2618 free_vmcs(per_cpu(vmxarea, cpu));
2619 per_cpu(vmxarea, cpu) = NULL;
2623 static __init int alloc_kvm_area(void)
2627 for_each_possible_cpu(cpu) {
2630 vmcs = alloc_vmcs_cpu(false, cpu, GFP_KERNEL);
2637 * When eVMCS is enabled, alloc_vmcs_cpu() sets
2638 * vmcs->revision_id to KVM_EVMCS_VERSION instead of
2639 * revision_id reported by MSR_IA32_VMX_BASIC.
2641 * However, even though not explicitly documented by
2642 * TLFS, VMXArea passed as VMXON argument should
2643 * still be marked with revision_id reported by
2646 if (static_branch_unlikely(&enable_evmcs))
2647 vmcs->hdr.revision_id = vmcs_config.revision_id;
2649 per_cpu(vmxarea, cpu) = vmcs;
2654 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
2655 struct kvm_segment *save)
2657 if (!emulate_invalid_guest_state) {
2659 * CS and SS RPL should be equal during guest entry according
2660 * to VMX spec, but in reality it is not always so. Since vcpu
2661 * is in the middle of the transition from real mode to
2662 * protected mode it is safe to assume that RPL 0 is a good
2665 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
2666 save->selector &= ~SEGMENT_RPL_MASK;
2667 save->dpl = save->selector & SEGMENT_RPL_MASK;
2670 vmx_set_segment(vcpu, save, seg);
2673 static void enter_pmode(struct kvm_vcpu *vcpu)
2675 unsigned long flags;
2676 struct vcpu_vmx *vmx = to_vmx(vcpu);
2679 * Update real mode segment cache. It may be not up-to-date if sement
2680 * register was written while vcpu was in a guest mode.
2682 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2683 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2684 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2685 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2686 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2687 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2689 vmx->rmode.vm86_active = 0;
2691 vmx_segment_cache_clear(vmx);
2693 vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2695 flags = vmcs_readl(GUEST_RFLAGS);
2696 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2697 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2698 vmcs_writel(GUEST_RFLAGS, flags);
2700 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
2701 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
2703 update_exception_bitmap(vcpu);
2705 fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2706 fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2707 fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2708 fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2709 fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2710 fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2713 static void fix_rmode_seg(int seg, struct kvm_segment *save)
2715 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2716 struct kvm_segment var = *save;
2719 if (seg == VCPU_SREG_CS)
2722 if (!emulate_invalid_guest_state) {
2723 var.selector = var.base >> 4;
2724 var.base = var.base & 0xffff0;
2734 if (save->base & 0xf)
2735 printk_once(KERN_WARNING "kvm: segment base is not "
2736 "paragraph aligned when entering "
2737 "protected mode (seg=%d)", seg);
2740 vmcs_write16(sf->selector, var.selector);
2741 vmcs_writel(sf->base, var.base);
2742 vmcs_write32(sf->limit, var.limit);
2743 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
2746 static void enter_rmode(struct kvm_vcpu *vcpu)
2748 unsigned long flags;
2749 struct vcpu_vmx *vmx = to_vmx(vcpu);
2750 struct kvm_vmx *kvm_vmx = to_kvm_vmx(vcpu->kvm);
2752 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2753 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2754 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2755 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2756 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2757 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2758 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2760 vmx->rmode.vm86_active = 1;
2763 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
2764 * vcpu. Warn the user that an update is overdue.
2766 if (!kvm_vmx->tss_addr)
2767 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
2768 "called before entering vcpu\n");
2770 vmx_segment_cache_clear(vmx);
2772 vmcs_writel(GUEST_TR_BASE, kvm_vmx->tss_addr);
2773 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
2774 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2776 flags = vmcs_readl(GUEST_RFLAGS);
2777 vmx->rmode.save_rflags = flags;
2779 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2781 vmcs_writel(GUEST_RFLAGS, flags);
2782 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
2783 update_exception_bitmap(vcpu);
2785 fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2786 fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2787 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2788 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2789 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2790 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2792 kvm_mmu_reset_context(vcpu);
2795 void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
2797 struct vcpu_vmx *vmx = to_vmx(vcpu);
2798 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
2803 vcpu->arch.efer = efer;
2804 if (efer & EFER_LMA) {
2805 vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2808 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2810 msr->data = efer & ~EFER_LME;
2815 #ifdef CONFIG_X86_64
2817 static void enter_lmode(struct kvm_vcpu *vcpu)
2821 vmx_segment_cache_clear(to_vmx(vcpu));
2823 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
2824 if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
2825 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
2827 vmcs_write32(GUEST_TR_AR_BYTES,
2828 (guest_tr_ar & ~VMX_AR_TYPE_MASK)
2829 | VMX_AR_TYPE_BUSY_64_TSS);
2831 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
2834 static void exit_lmode(struct kvm_vcpu *vcpu)
2836 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2837 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
2842 static void vmx_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t addr)
2844 int vpid = to_vmx(vcpu)->vpid;
2846 if (!vpid_sync_vcpu_addr(vpid, addr))
2847 vpid_sync_context(vpid);
2850 * If VPIDs are not supported or enabled, then the above is a no-op.
2851 * But we don't really need a TLB flush in that case anyway, because
2852 * each VM entry/exit includes an implicit flush when VPID is 0.
2856 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
2858 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
2860 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
2861 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
2864 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
2866 if (enable_unrestricted_guest || (enable_ept && is_paging(vcpu)))
2867 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2868 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
2871 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
2873 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
2875 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
2876 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
2879 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
2881 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2883 if (!test_bit(VCPU_EXREG_PDPTR,
2884 (unsigned long *)&vcpu->arch.regs_dirty))
2887 if (is_pae_paging(vcpu)) {
2888 vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
2889 vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
2890 vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
2891 vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
2895 void ept_save_pdptrs(struct kvm_vcpu *vcpu)
2897 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2899 if (is_pae_paging(vcpu)) {
2900 mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
2901 mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
2902 mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
2903 mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
2906 __set_bit(VCPU_EXREG_PDPTR,
2907 (unsigned long *)&vcpu->arch.regs_avail);
2908 __set_bit(VCPU_EXREG_PDPTR,
2909 (unsigned long *)&vcpu->arch.regs_dirty);
2912 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
2914 struct kvm_vcpu *vcpu)
2916 struct vcpu_vmx *vmx = to_vmx(vcpu);
2918 if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
2919 vmx_decache_cr3(vcpu);
2920 if (!(cr0 & X86_CR0_PG)) {
2921 /* From paging/starting to nonpaging */
2922 exec_controls_setbit(vmx, CPU_BASED_CR3_LOAD_EXITING |
2923 CPU_BASED_CR3_STORE_EXITING);
2924 vcpu->arch.cr0 = cr0;
2925 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2926 } else if (!is_paging(vcpu)) {
2927 /* From nonpaging to paging */
2928 exec_controls_clearbit(vmx, CPU_BASED_CR3_LOAD_EXITING |
2929 CPU_BASED_CR3_STORE_EXITING);
2930 vcpu->arch.cr0 = cr0;
2931 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2934 if (!(cr0 & X86_CR0_WP))
2935 *hw_cr0 &= ~X86_CR0_WP;
2938 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
2940 struct vcpu_vmx *vmx = to_vmx(vcpu);
2941 unsigned long hw_cr0;
2943 hw_cr0 = (cr0 & ~KVM_VM_CR0_ALWAYS_OFF);
2944 if (enable_unrestricted_guest)
2945 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
2947 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
2949 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
2952 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
2956 #ifdef CONFIG_X86_64
2957 if (vcpu->arch.efer & EFER_LME) {
2958 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
2960 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
2965 if (enable_ept && !enable_unrestricted_guest)
2966 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
2968 vmcs_writel(CR0_READ_SHADOW, cr0);
2969 vmcs_writel(GUEST_CR0, hw_cr0);
2970 vcpu->arch.cr0 = cr0;
2972 /* depends on vcpu->arch.cr0 to be set to a new value */
2973 vmx->emulation_required = emulation_required(vcpu);
2976 static int get_ept_level(struct kvm_vcpu *vcpu)
2978 /* Nested EPT currently only supports 4-level walks. */
2979 if (is_guest_mode(vcpu) && nested_cpu_has_ept(get_vmcs12(vcpu)))
2981 if (cpu_has_vmx_ept_5levels() && (cpuid_maxphyaddr(vcpu) > 48))
2986 u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa)
2988 u64 eptp = VMX_EPTP_MT_WB;
2990 eptp |= (get_ept_level(vcpu) == 5) ? VMX_EPTP_PWL_5 : VMX_EPTP_PWL_4;
2992 if (enable_ept_ad_bits &&
2993 (!is_guest_mode(vcpu) || nested_ept_ad_enabled(vcpu)))
2994 eptp |= VMX_EPTP_AD_ENABLE_BIT;
2995 eptp |= (root_hpa & PAGE_MASK);
3000 void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
3002 struct kvm *kvm = vcpu->kvm;
3003 bool update_guest_cr3 = true;
3004 unsigned long guest_cr3;
3009 eptp = construct_eptp(vcpu, cr3);
3010 vmcs_write64(EPT_POINTER, eptp);
3012 if (kvm_x86_ops->tlb_remote_flush) {
3013 spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
3014 to_vmx(vcpu)->ept_pointer = eptp;
3015 to_kvm_vmx(kvm)->ept_pointers_match
3016 = EPT_POINTERS_CHECK;
3017 spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
3020 /* Loading vmcs02.GUEST_CR3 is handled by nested VM-Enter. */
3021 if (is_guest_mode(vcpu))
3022 update_guest_cr3 = false;
3023 else if (enable_unrestricted_guest || is_paging(vcpu))
3024 guest_cr3 = kvm_read_cr3(vcpu);
3026 guest_cr3 = to_kvm_vmx(kvm)->ept_identity_map_addr;
3027 ept_load_pdptrs(vcpu);
3030 if (update_guest_cr3)
3031 vmcs_writel(GUEST_CR3, guest_cr3);
3034 int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3036 struct vcpu_vmx *vmx = to_vmx(vcpu);
3038 * Pass through host's Machine Check Enable value to hw_cr4, which
3039 * is in force while we are in guest mode. Do not let guests control
3040 * this bit, even if host CR4.MCE == 0.
3042 unsigned long hw_cr4;
3044 hw_cr4 = (cr4_read_shadow() & X86_CR4_MCE) | (cr4 & ~X86_CR4_MCE);
3045 if (enable_unrestricted_guest)
3046 hw_cr4 |= KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST;
3047 else if (vmx->rmode.vm86_active)
3048 hw_cr4 |= KVM_RMODE_VM_CR4_ALWAYS_ON;
3050 hw_cr4 |= KVM_PMODE_VM_CR4_ALWAYS_ON;
3052 if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated()) {
3053 if (cr4 & X86_CR4_UMIP) {
3054 secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_DESC);
3055 hw_cr4 &= ~X86_CR4_UMIP;
3056 } else if (!is_guest_mode(vcpu) ||
3057 !nested_cpu_has2(get_vmcs12(vcpu), SECONDARY_EXEC_DESC)) {
3058 secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_DESC);
3062 if (cr4 & X86_CR4_VMXE) {
3064 * To use VMXON (and later other VMX instructions), a guest
3065 * must first be able to turn on cr4.VMXE (see handle_vmon()).
3066 * So basically the check on whether to allow nested VMX
3067 * is here. We operate under the default treatment of SMM,
3068 * so VMX cannot be enabled under SMM.
3070 if (!nested_vmx_allowed(vcpu) || is_smm(vcpu))
3074 if (vmx->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
3077 vcpu->arch.cr4 = cr4;
3079 if (!enable_unrestricted_guest) {
3081 if (!is_paging(vcpu)) {
3082 hw_cr4 &= ~X86_CR4_PAE;
3083 hw_cr4 |= X86_CR4_PSE;
3084 } else if (!(cr4 & X86_CR4_PAE)) {
3085 hw_cr4 &= ~X86_CR4_PAE;
3090 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
3091 * hardware. To emulate this behavior, SMEP/SMAP/PKU needs
3092 * to be manually disabled when guest switches to non-paging
3095 * If !enable_unrestricted_guest, the CPU is always running
3096 * with CR0.PG=1 and CR4 needs to be modified.
3097 * If enable_unrestricted_guest, the CPU automatically
3098 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
3100 if (!is_paging(vcpu))
3101 hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
3104 vmcs_writel(CR4_READ_SHADOW, cr4);
3105 vmcs_writel(GUEST_CR4, hw_cr4);
3109 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3111 struct vcpu_vmx *vmx = to_vmx(vcpu);
3114 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3115 *var = vmx->rmode.segs[seg];
3116 if (seg == VCPU_SREG_TR
3117 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3119 var->base = vmx_read_guest_seg_base(vmx, seg);
3120 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3123 var->base = vmx_read_guest_seg_base(vmx, seg);
3124 var->limit = vmx_read_guest_seg_limit(vmx, seg);
3125 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3126 ar = vmx_read_guest_seg_ar(vmx, seg);
3127 var->unusable = (ar >> 16) & 1;
3128 var->type = ar & 15;
3129 var->s = (ar >> 4) & 1;
3130 var->dpl = (ar >> 5) & 3;
3132 * Some userspaces do not preserve unusable property. Since usable
3133 * segment has to be present according to VMX spec we can use present
3134 * property to amend userspace bug by making unusable segment always
3135 * nonpresent. vmx_segment_access_rights() already marks nonpresent
3136 * segment as unusable.
3138 var->present = !var->unusable;
3139 var->avl = (ar >> 12) & 1;
3140 var->l = (ar >> 13) & 1;
3141 var->db = (ar >> 14) & 1;
3142 var->g = (ar >> 15) & 1;
3145 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3147 struct kvm_segment s;
3149 if (to_vmx(vcpu)->rmode.vm86_active) {
3150 vmx_get_segment(vcpu, &s, seg);
3153 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3156 int vmx_get_cpl(struct kvm_vcpu *vcpu)
3158 struct vcpu_vmx *vmx = to_vmx(vcpu);
3160 if (unlikely(vmx->rmode.vm86_active))
3163 int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
3164 return VMX_AR_DPL(ar);
3168 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3172 if (var->unusable || !var->present)
3175 ar = var->type & 15;
3176 ar |= (var->s & 1) << 4;
3177 ar |= (var->dpl & 3) << 5;
3178 ar |= (var->present & 1) << 7;
3179 ar |= (var->avl & 1) << 12;
3180 ar |= (var->l & 1) << 13;
3181 ar |= (var->db & 1) << 14;
3182 ar |= (var->g & 1) << 15;
3188 void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3190 struct vcpu_vmx *vmx = to_vmx(vcpu);
3191 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3193 vmx_segment_cache_clear(vmx);
3195 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3196 vmx->rmode.segs[seg] = *var;
3197 if (seg == VCPU_SREG_TR)
3198 vmcs_write16(sf->selector, var->selector);
3200 fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3204 vmcs_writel(sf->base, var->base);
3205 vmcs_write32(sf->limit, var->limit);
3206 vmcs_write16(sf->selector, var->selector);
3209 * Fix the "Accessed" bit in AR field of segment registers for older
3211 * IA32 arch specifies that at the time of processor reset the
3212 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3213 * is setting it to 0 in the userland code. This causes invalid guest
3214 * state vmexit when "unrestricted guest" mode is turned on.
3215 * Fix for this setup issue in cpu_reset is being pushed in the qemu
3216 * tree. Newer qemu binaries with that qemu fix would not need this
3219 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3220 var->type |= 0x1; /* Accessed */
3222 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3225 vmx->emulation_required = emulation_required(vcpu);
3228 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3230 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3232 *db = (ar >> 14) & 1;
3233 *l = (ar >> 13) & 1;
3236 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3238 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3239 dt->address = vmcs_readl(GUEST_IDTR_BASE);
3242 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3244 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3245 vmcs_writel(GUEST_IDTR_BASE, dt->address);
3248 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3250 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3251 dt->address = vmcs_readl(GUEST_GDTR_BASE);
3254 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3256 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3257 vmcs_writel(GUEST_GDTR_BASE, dt->address);
3260 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3262 struct kvm_segment var;
3265 vmx_get_segment(vcpu, &var, seg);
3267 if (seg == VCPU_SREG_CS)
3269 ar = vmx_segment_access_rights(&var);
3271 if (var.base != (var.selector << 4))
3273 if (var.limit != 0xffff)
3281 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3283 struct kvm_segment cs;
3284 unsigned int cs_rpl;
3286 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3287 cs_rpl = cs.selector & SEGMENT_RPL_MASK;
3291 if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
3295 if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
3296 if (cs.dpl > cs_rpl)
3299 if (cs.dpl != cs_rpl)
3305 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3309 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3311 struct kvm_segment ss;
3312 unsigned int ss_rpl;
3314 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3315 ss_rpl = ss.selector & SEGMENT_RPL_MASK;
3319 if (ss.type != 3 && ss.type != 7)
3323 if (ss.dpl != ss_rpl) /* DPL != RPL */
3331 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3333 struct kvm_segment var;
3336 vmx_get_segment(vcpu, &var, seg);
3337 rpl = var.selector & SEGMENT_RPL_MASK;
3345 if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
3346 if (var.dpl < rpl) /* DPL < RPL */
3350 /* TODO: Add other members to kvm_segment_field to allow checking for other access
3356 static bool tr_valid(struct kvm_vcpu *vcpu)
3358 struct kvm_segment tr;
3360 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3364 if (tr.selector & SEGMENT_TI_MASK) /* TI = 1 */
3366 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3374 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3376 struct kvm_segment ldtr;
3378 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3382 if (ldtr.selector & SEGMENT_TI_MASK) /* TI = 1 */
3392 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3394 struct kvm_segment cs, ss;
3396 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3397 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3399 return ((cs.selector & SEGMENT_RPL_MASK) ==
3400 (ss.selector & SEGMENT_RPL_MASK));
3404 * Check if guest state is valid. Returns true if valid, false if
3406 * We assume that registers are always usable
3408 static bool guest_state_valid(struct kvm_vcpu *vcpu)
3410 if (enable_unrestricted_guest)
3413 /* real mode guest state checks */
3414 if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
3415 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3417 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3419 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3421 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3423 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3425 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3428 /* protected mode guest state checks */
3429 if (!cs_ss_rpl_check(vcpu))
3431 if (!code_segment_valid(vcpu))
3433 if (!stack_segment_valid(vcpu))
3435 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3437 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3439 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3441 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3443 if (!tr_valid(vcpu))
3445 if (!ldtr_valid(vcpu))
3449 * - Add checks on RIP
3450 * - Add checks on RFLAGS
3456 static int init_rmode_tss(struct kvm *kvm)
3462 idx = srcu_read_lock(&kvm->srcu);
3463 fn = to_kvm_vmx(kvm)->tss_addr >> PAGE_SHIFT;
3464 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3467 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3468 r = kvm_write_guest_page(kvm, fn++, &data,
3469 TSS_IOPB_BASE_OFFSET, sizeof(u16));
3472 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3475 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3479 r = kvm_write_guest_page(kvm, fn, &data,
3480 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3483 srcu_read_unlock(&kvm->srcu, idx);
3487 static int init_rmode_identity_map(struct kvm *kvm)
3489 struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm);
3491 kvm_pfn_t identity_map_pfn;
3494 /* Protect kvm_vmx->ept_identity_pagetable_done. */
3495 mutex_lock(&kvm->slots_lock);
3497 if (likely(kvm_vmx->ept_identity_pagetable_done))
3500 if (!kvm_vmx->ept_identity_map_addr)
3501 kvm_vmx->ept_identity_map_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
3502 identity_map_pfn = kvm_vmx->ept_identity_map_addr >> PAGE_SHIFT;
3504 r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
3505 kvm_vmx->ept_identity_map_addr, PAGE_SIZE);
3509 idx = srcu_read_lock(&kvm->srcu);
3510 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3513 /* Set up identity-mapping pagetable for EPT in real mode */
3514 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3515 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3516 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3517 r = kvm_write_guest_page(kvm, identity_map_pfn,
3518 &tmp, i * sizeof(tmp), sizeof(tmp));
3522 kvm_vmx->ept_identity_pagetable_done = true;
3525 srcu_read_unlock(&kvm->srcu, idx);
3528 mutex_unlock(&kvm->slots_lock);
3532 static void seg_setup(int seg)
3534 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3537 vmcs_write16(sf->selector, 0);
3538 vmcs_writel(sf->base, 0);
3539 vmcs_write32(sf->limit, 0xffff);
3541 if (seg == VCPU_SREG_CS)
3542 ar |= 0x08; /* code segment */
3544 vmcs_write32(sf->ar_bytes, ar);
3547 static int alloc_apic_access_page(struct kvm *kvm)
3552 mutex_lock(&kvm->slots_lock);
3553 if (kvm->arch.apic_access_page_done)
3555 r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
3556 APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
3560 page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
3561 if (is_error_page(page)) {
3567 * Do not pin the page in memory, so that memory hot-unplug
3568 * is able to migrate it.
3571 kvm->arch.apic_access_page_done = true;
3573 mutex_unlock(&kvm->slots_lock);
3577 int allocate_vpid(void)
3583 spin_lock(&vmx_vpid_lock);
3584 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3585 if (vpid < VMX_NR_VPIDS)
3586 __set_bit(vpid, vmx_vpid_bitmap);
3589 spin_unlock(&vmx_vpid_lock);
3593 void free_vpid(int vpid)
3595 if (!enable_vpid || vpid == 0)
3597 spin_lock(&vmx_vpid_lock);
3598 __clear_bit(vpid, vmx_vpid_bitmap);
3599 spin_unlock(&vmx_vpid_lock);
3602 static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
3605 int f = sizeof(unsigned long);
3607 if (!cpu_has_vmx_msr_bitmap())
3610 if (static_branch_unlikely(&enable_evmcs))
3611 evmcs_touch_msr_bitmap();
3614 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3615 * have the write-low and read-high bitmap offsets the wrong way round.
3616 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3618 if (msr <= 0x1fff) {
3619 if (type & MSR_TYPE_R)
3621 __clear_bit(msr, msr_bitmap + 0x000 / f);
3623 if (type & MSR_TYPE_W)
3625 __clear_bit(msr, msr_bitmap + 0x800 / f);
3627 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3629 if (type & MSR_TYPE_R)
3631 __clear_bit(msr, msr_bitmap + 0x400 / f);
3633 if (type & MSR_TYPE_W)
3635 __clear_bit(msr, msr_bitmap + 0xc00 / f);
3640 static __always_inline void vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
3643 int f = sizeof(unsigned long);
3645 if (!cpu_has_vmx_msr_bitmap())
3648 if (static_branch_unlikely(&enable_evmcs))
3649 evmcs_touch_msr_bitmap();
3652 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3653 * have the write-low and read-high bitmap offsets the wrong way round.
3654 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3656 if (msr <= 0x1fff) {
3657 if (type & MSR_TYPE_R)
3659 __set_bit(msr, msr_bitmap + 0x000 / f);
3661 if (type & MSR_TYPE_W)
3663 __set_bit(msr, msr_bitmap + 0x800 / f);
3665 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3667 if (type & MSR_TYPE_R)
3669 __set_bit(msr, msr_bitmap + 0x400 / f);
3671 if (type & MSR_TYPE_W)
3673 __set_bit(msr, msr_bitmap + 0xc00 / f);
3678 static __always_inline void vmx_set_intercept_for_msr(unsigned long *msr_bitmap,
3679 u32 msr, int type, bool value)
3682 vmx_enable_intercept_for_msr(msr_bitmap, msr, type);
3684 vmx_disable_intercept_for_msr(msr_bitmap, msr, type);
3687 static u8 vmx_msr_bitmap_mode(struct kvm_vcpu *vcpu)
3691 if (cpu_has_secondary_exec_ctrls() &&
3692 (secondary_exec_controls_get(to_vmx(vcpu)) &
3693 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
3694 mode |= MSR_BITMAP_MODE_X2APIC;
3695 if (enable_apicv && kvm_vcpu_apicv_active(vcpu))
3696 mode |= MSR_BITMAP_MODE_X2APIC_APICV;
3702 static void vmx_update_msr_bitmap_x2apic(unsigned long *msr_bitmap,
3707 for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
3708 unsigned word = msr / BITS_PER_LONG;
3709 msr_bitmap[word] = (mode & MSR_BITMAP_MODE_X2APIC_APICV) ? 0 : ~0;
3710 msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
3713 if (mode & MSR_BITMAP_MODE_X2APIC) {
3715 * TPR reads and writes can be virtualized even if virtual interrupt
3716 * delivery is not in use.
3718 vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TASKPRI), MSR_TYPE_RW);
3719 if (mode & MSR_BITMAP_MODE_X2APIC_APICV) {
3720 vmx_enable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TMCCT), MSR_TYPE_R);
3721 vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_EOI), MSR_TYPE_W);
3722 vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_SELF_IPI), MSR_TYPE_W);
3727 void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu)
3729 struct vcpu_vmx *vmx = to_vmx(vcpu);
3730 unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3731 u8 mode = vmx_msr_bitmap_mode(vcpu);
3732 u8 changed = mode ^ vmx->msr_bitmap_mode;
3737 if (changed & (MSR_BITMAP_MODE_X2APIC | MSR_BITMAP_MODE_X2APIC_APICV))
3738 vmx_update_msr_bitmap_x2apic(msr_bitmap, mode);
3740 vmx->msr_bitmap_mode = mode;
3743 void pt_update_intercept_for_msr(struct vcpu_vmx *vmx)
3745 unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3746 bool flag = !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
3749 vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_STATUS,
3751 vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_BASE,
3753 vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_MASK,
3755 vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_CR3_MATCH,
3757 for (i = 0; i < vmx->pt_desc.addr_range; i++) {
3758 vmx_set_intercept_for_msr(msr_bitmap,
3759 MSR_IA32_RTIT_ADDR0_A + i * 2, MSR_TYPE_RW, flag);
3760 vmx_set_intercept_for_msr(msr_bitmap,
3761 MSR_IA32_RTIT_ADDR0_B + i * 2, MSR_TYPE_RW, flag);
3765 static bool vmx_get_enable_apicv(struct kvm_vcpu *vcpu)
3767 return enable_apicv;
3770 static bool vmx_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
3772 struct vcpu_vmx *vmx = to_vmx(vcpu);
3777 if (WARN_ON_ONCE(!is_guest_mode(vcpu)) ||
3778 !nested_cpu_has_vid(get_vmcs12(vcpu)) ||
3779 WARN_ON_ONCE(!vmx->nested.virtual_apic_map.gfn))
3782 rvi = vmx_get_rvi();
3784 vapic_page = vmx->nested.virtual_apic_map.hva;
3785 vppr = *((u32 *)(vapic_page + APIC_PROCPRI));
3787 return ((rvi & 0xf0) > (vppr & 0xf0));
3790 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu,
3794 int pi_vec = nested ? POSTED_INTR_NESTED_VECTOR : POSTED_INTR_VECTOR;
3796 if (vcpu->mode == IN_GUEST_MODE) {
3798 * The vector of interrupt to be delivered to vcpu had
3799 * been set in PIR before this function.
3801 * Following cases will be reached in this block, and
3802 * we always send a notification event in all cases as
3805 * Case 1: vcpu keeps in non-root mode. Sending a
3806 * notification event posts the interrupt to vcpu.
3808 * Case 2: vcpu exits to root mode and is still
3809 * runnable. PIR will be synced to vIRR before the
3810 * next vcpu entry. Sending a notification event in
3811 * this case has no effect, as vcpu is not in root
3814 * Case 3: vcpu exits to root mode and is blocked.
3815 * vcpu_block() has already synced PIR to vIRR and
3816 * never blocks vcpu if vIRR is not cleared. Therefore,
3817 * a blocked vcpu here does not wait for any requested
3818 * interrupts in PIR, and sending a notification event
3819 * which has no effect is safe here.
3822 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu), pi_vec);
3829 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
3832 struct vcpu_vmx *vmx = to_vmx(vcpu);
3834 if (is_guest_mode(vcpu) &&
3835 vector == vmx->nested.posted_intr_nv) {
3837 * If a posted intr is not recognized by hardware,
3838 * we will accomplish it in the next vmentry.
3840 vmx->nested.pi_pending = true;
3841 kvm_make_request(KVM_REQ_EVENT, vcpu);
3842 /* the PIR and ON have been set by L1. */
3843 if (!kvm_vcpu_trigger_posted_interrupt(vcpu, true))
3844 kvm_vcpu_kick(vcpu);
3850 * Send interrupt to vcpu via posted interrupt way.
3851 * 1. If target vcpu is running(non-root mode), send posted interrupt
3852 * notification to vcpu and hardware will sync PIR to vIRR atomically.
3853 * 2. If target vcpu isn't running(root mode), kick it to pick up the
3854 * interrupt from PIR in next vmentry.
3856 static int vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
3858 struct vcpu_vmx *vmx = to_vmx(vcpu);
3861 r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
3865 if (!vcpu->arch.apicv_active)
3868 if (pi_test_and_set_pir(vector, &vmx->pi_desc))
3871 /* If a previous notification has sent the IPI, nothing to do. */
3872 if (pi_test_and_set_on(&vmx->pi_desc))
3875 if (!kvm_vcpu_trigger_posted_interrupt(vcpu, false))
3876 kvm_vcpu_kick(vcpu);
3882 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
3883 * will not change in the lifetime of the guest.
3884 * Note that host-state that does change is set elsewhere. E.g., host-state
3885 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
3887 void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
3891 unsigned long cr0, cr3, cr4;
3894 WARN_ON(cr0 & X86_CR0_TS);
3895 vmcs_writel(HOST_CR0, cr0); /* 22.2.3 */
3898 * Save the most likely value for this task's CR3 in the VMCS.
3899 * We can't use __get_current_cr3_fast() because we're not atomic.
3902 vmcs_writel(HOST_CR3, cr3); /* 22.2.3 FIXME: shadow tables */
3903 vmx->loaded_vmcs->host_state.cr3 = cr3;
3905 /* Save the most likely value for this task's CR4 in the VMCS. */
3906 cr4 = cr4_read_shadow();
3907 vmcs_writel(HOST_CR4, cr4); /* 22.2.3, 22.2.5 */
3908 vmx->loaded_vmcs->host_state.cr4 = cr4;
3910 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
3911 #ifdef CONFIG_X86_64
3913 * Load null selectors, so we can avoid reloading them in
3914 * vmx_prepare_switch_to_host(), in case userspace uses
3915 * the null selectors too (the expected case).
3917 vmcs_write16(HOST_DS_SELECTOR, 0);
3918 vmcs_write16(HOST_ES_SELECTOR, 0);
3920 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
3921 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
3923 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
3924 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
3926 vmcs_writel(HOST_IDTR_BASE, host_idt_base); /* 22.2.4 */
3928 vmcs_writel(HOST_RIP, (unsigned long)vmx_vmexit); /* 22.2.5 */
3930 rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
3931 vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
3932 rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
3933 vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */
3935 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
3936 rdmsr(MSR_IA32_CR_PAT, low32, high32);
3937 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
3940 if (cpu_has_load_ia32_efer())
3941 vmcs_write64(HOST_IA32_EFER, host_efer);
3944 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
3946 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
3948 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
3949 if (is_guest_mode(&vmx->vcpu))
3950 vmx->vcpu.arch.cr4_guest_owned_bits &=
3951 ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
3952 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
3955 u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
3957 u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
3959 if (!kvm_vcpu_apicv_active(&vmx->vcpu))
3960 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
3963 pin_based_exec_ctrl &= ~PIN_BASED_VIRTUAL_NMIS;
3965 if (!enable_preemption_timer)
3966 pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
3968 return pin_based_exec_ctrl;
3971 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
3973 struct vcpu_vmx *vmx = to_vmx(vcpu);
3975 pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
3976 if (cpu_has_secondary_exec_ctrls()) {
3977 if (kvm_vcpu_apicv_active(vcpu))
3978 secondary_exec_controls_setbit(vmx,
3979 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3980 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3982 secondary_exec_controls_clearbit(vmx,
3983 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3984 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3987 if (cpu_has_vmx_msr_bitmap())
3988 vmx_update_msr_bitmap(vcpu);
3991 u32 vmx_exec_control(struct vcpu_vmx *vmx)
3993 u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
3995 if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
3996 exec_control &= ~CPU_BASED_MOV_DR_EXITING;
3998 if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
3999 exec_control &= ~CPU_BASED_TPR_SHADOW;
4000 #ifdef CONFIG_X86_64
4001 exec_control |= CPU_BASED_CR8_STORE_EXITING |
4002 CPU_BASED_CR8_LOAD_EXITING;
4006 exec_control |= CPU_BASED_CR3_STORE_EXITING |
4007 CPU_BASED_CR3_LOAD_EXITING |
4008 CPU_BASED_INVLPG_EXITING;
4009 if (kvm_mwait_in_guest(vmx->vcpu.kvm))
4010 exec_control &= ~(CPU_BASED_MWAIT_EXITING |
4011 CPU_BASED_MONITOR_EXITING);
4012 if (kvm_hlt_in_guest(vmx->vcpu.kvm))
4013 exec_control &= ~CPU_BASED_HLT_EXITING;
4014 return exec_control;
4018 static void vmx_compute_secondary_exec_control(struct vcpu_vmx *vmx)
4020 struct kvm_vcpu *vcpu = &vmx->vcpu;
4022 u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
4024 if (pt_mode == PT_MODE_SYSTEM)
4025 exec_control &= ~(SECONDARY_EXEC_PT_USE_GPA | SECONDARY_EXEC_PT_CONCEAL_VMX);
4026 if (!cpu_need_virtualize_apic_accesses(vcpu))
4027 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
4029 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
4031 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
4032 enable_unrestricted_guest = 0;
4034 if (!enable_unrestricted_guest)
4035 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4036 if (kvm_pause_in_guest(vmx->vcpu.kvm))
4037 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
4038 if (!kvm_vcpu_apicv_active(vcpu))
4039 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
4040 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4041 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
4043 /* SECONDARY_EXEC_DESC is enabled/disabled on writes to CR4.UMIP,
4044 * in vmx_set_cr4. */
4045 exec_control &= ~SECONDARY_EXEC_DESC;
4047 /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4049 We can NOT enable shadow_vmcs here because we don't have yet
4052 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4055 exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
4057 if (vmx_xsaves_supported()) {
4058 /* Exposing XSAVES only when XSAVE is exposed */
4059 bool xsaves_enabled =
4060 guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
4061 guest_cpuid_has(vcpu, X86_FEATURE_XSAVES);
4063 if (!xsaves_enabled)
4064 exec_control &= ~SECONDARY_EXEC_XSAVES;
4068 vmx->nested.msrs.secondary_ctls_high |=
4069 SECONDARY_EXEC_XSAVES;
4071 vmx->nested.msrs.secondary_ctls_high &=
4072 ~SECONDARY_EXEC_XSAVES;
4076 if (vmx_rdtscp_supported()) {
4077 bool rdtscp_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP);
4078 if (!rdtscp_enabled)
4079 exec_control &= ~SECONDARY_EXEC_RDTSCP;
4083 vmx->nested.msrs.secondary_ctls_high |=
4084 SECONDARY_EXEC_RDTSCP;
4086 vmx->nested.msrs.secondary_ctls_high &=
4087 ~SECONDARY_EXEC_RDTSCP;
4091 if (vmx_invpcid_supported()) {
4092 /* Exposing INVPCID only when PCID is exposed */
4093 bool invpcid_enabled =
4094 guest_cpuid_has(vcpu, X86_FEATURE_INVPCID) &&
4095 guest_cpuid_has(vcpu, X86_FEATURE_PCID);
4097 if (!invpcid_enabled) {
4098 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
4099 guest_cpuid_clear(vcpu, X86_FEATURE_INVPCID);
4103 if (invpcid_enabled)
4104 vmx->nested.msrs.secondary_ctls_high |=
4105 SECONDARY_EXEC_ENABLE_INVPCID;
4107 vmx->nested.msrs.secondary_ctls_high &=
4108 ~SECONDARY_EXEC_ENABLE_INVPCID;
4112 if (vmx_rdrand_supported()) {
4113 bool rdrand_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDRAND);
4115 exec_control &= ~SECONDARY_EXEC_RDRAND_EXITING;
4119 vmx->nested.msrs.secondary_ctls_high |=
4120 SECONDARY_EXEC_RDRAND_EXITING;
4122 vmx->nested.msrs.secondary_ctls_high &=
4123 ~SECONDARY_EXEC_RDRAND_EXITING;
4127 if (vmx_rdseed_supported()) {
4128 bool rdseed_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDSEED);
4130 exec_control &= ~SECONDARY_EXEC_RDSEED_EXITING;
4134 vmx->nested.msrs.secondary_ctls_high |=
4135 SECONDARY_EXEC_RDSEED_EXITING;
4137 vmx->nested.msrs.secondary_ctls_high &=
4138 ~SECONDARY_EXEC_RDSEED_EXITING;
4142 if (vmx_waitpkg_supported()) {
4143 bool waitpkg_enabled =
4144 guest_cpuid_has(vcpu, X86_FEATURE_WAITPKG);
4146 if (!waitpkg_enabled)
4147 exec_control &= ~SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4150 if (waitpkg_enabled)
4151 vmx->nested.msrs.secondary_ctls_high |=
4152 SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4154 vmx->nested.msrs.secondary_ctls_high &=
4155 ~SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4159 vmx->secondary_exec_control = exec_control;
4162 static void ept_set_mmio_spte_mask(void)
4165 * EPT Misconfigurations can be generated if the value of bits 2:0
4166 * of an EPT paging-structure entry is 110b (write/execute).
4168 kvm_mmu_set_mmio_spte_mask(VMX_EPT_RWX_MASK,
4169 VMX_EPT_MISCONFIG_WX_VALUE, 0);
4172 #define VMX_XSS_EXIT_BITMAP 0
4175 * Sets up the vmcs for emulated real mode.
4177 static void vmx_vcpu_setup(struct vcpu_vmx *vmx)
4182 nested_vmx_vcpu_setup();
4184 if (cpu_has_vmx_msr_bitmap())
4185 vmcs_write64(MSR_BITMAP, __pa(vmx->vmcs01.msr_bitmap));
4187 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4190 pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
4191 vmx->hv_deadline_tsc = -1;
4193 exec_controls_set(vmx, vmx_exec_control(vmx));
4195 if (cpu_has_secondary_exec_ctrls()) {
4196 vmx_compute_secondary_exec_control(vmx);
4197 secondary_exec_controls_set(vmx, vmx->secondary_exec_control);
4200 if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
4201 vmcs_write64(EOI_EXIT_BITMAP0, 0);
4202 vmcs_write64(EOI_EXIT_BITMAP1, 0);
4203 vmcs_write64(EOI_EXIT_BITMAP2, 0);
4204 vmcs_write64(EOI_EXIT_BITMAP3, 0);
4206 vmcs_write16(GUEST_INTR_STATUS, 0);
4208 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4209 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4212 if (!kvm_pause_in_guest(vmx->vcpu.kvm)) {
4213 vmcs_write32(PLE_GAP, ple_gap);
4214 vmx->ple_window = ple_window;
4215 vmx->ple_window_dirty = true;
4218 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4219 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4220 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
4222 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
4223 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
4224 vmx_set_constant_host_state(vmx);
4225 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4226 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4228 if (cpu_has_vmx_vmfunc())
4229 vmcs_write64(VM_FUNCTION_CONTROL, 0);
4231 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4232 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4233 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
4234 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4235 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
4237 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
4238 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
4240 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
4241 u32 index = vmx_msr_index[i];
4242 u32 data_low, data_high;
4245 if (rdmsr_safe(index, &data_low, &data_high) < 0)
4247 if (wrmsr_safe(index, data_low, data_high) < 0)
4249 vmx->guest_msrs[j].index = i;
4250 vmx->guest_msrs[j].data = 0;
4251 vmx->guest_msrs[j].mask = -1ull;
4255 vm_exit_controls_set(vmx, vmx_vmexit_ctrl());
4257 /* 22.2.1, 20.8.1 */
4258 vm_entry_controls_set(vmx, vmx_vmentry_ctrl());
4260 vmx->vcpu.arch.cr0_guest_owned_bits = X86_CR0_TS;
4261 vmcs_writel(CR0_GUEST_HOST_MASK, ~X86_CR0_TS);
4263 set_cr4_guest_host_mask(vmx);
4265 if (vmx_xsaves_supported())
4266 vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
4269 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
4270 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
4273 if (cpu_has_vmx_encls_vmexit())
4274 vmcs_write64(ENCLS_EXITING_BITMAP, -1ull);
4276 if (pt_mode == PT_MODE_HOST_GUEST) {
4277 memset(&vmx->pt_desc, 0, sizeof(vmx->pt_desc));
4278 /* Bit[6~0] are forced to 1, writes are ignored. */
4279 vmx->pt_desc.guest.output_mask = 0x7F;
4280 vmcs_write64(GUEST_IA32_RTIT_CTL, 0);
4284 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
4286 struct vcpu_vmx *vmx = to_vmx(vcpu);
4287 struct msr_data apic_base_msr;
4290 vmx->rmode.vm86_active = 0;
4293 vmx->msr_ia32_umwait_control = 0;
4295 vcpu->arch.microcode_version = 0x100000000ULL;
4296 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
4297 vmx->hv_deadline_tsc = -1;
4298 kvm_set_cr8(vcpu, 0);
4301 apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
4302 MSR_IA32_APICBASE_ENABLE;
4303 if (kvm_vcpu_is_reset_bsp(vcpu))
4304 apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
4305 apic_base_msr.host_initiated = true;
4306 kvm_set_apic_base(vcpu, &apic_base_msr);
4309 vmx_segment_cache_clear(vmx);
4311 seg_setup(VCPU_SREG_CS);
4312 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4313 vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
4315 seg_setup(VCPU_SREG_DS);
4316 seg_setup(VCPU_SREG_ES);
4317 seg_setup(VCPU_SREG_FS);
4318 seg_setup(VCPU_SREG_GS);
4319 seg_setup(VCPU_SREG_SS);
4321 vmcs_write16(GUEST_TR_SELECTOR, 0);
4322 vmcs_writel(GUEST_TR_BASE, 0);
4323 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4324 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4326 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4327 vmcs_writel(GUEST_LDTR_BASE, 0);
4328 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4329 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4332 vmcs_write32(GUEST_SYSENTER_CS, 0);
4333 vmcs_writel(GUEST_SYSENTER_ESP, 0);
4334 vmcs_writel(GUEST_SYSENTER_EIP, 0);
4335 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4338 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
4339 kvm_rip_write(vcpu, 0xfff0);
4341 vmcs_writel(GUEST_GDTR_BASE, 0);
4342 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4344 vmcs_writel(GUEST_IDTR_BASE, 0);
4345 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4347 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4348 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4349 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4350 if (kvm_mpx_supported())
4351 vmcs_write64(GUEST_BNDCFGS, 0);
4355 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
4357 if (cpu_has_vmx_tpr_shadow() && !init_event) {
4358 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4359 if (cpu_need_tpr_shadow(vcpu))
4360 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4361 __pa(vcpu->arch.apic->regs));
4362 vmcs_write32(TPR_THRESHOLD, 0);
4365 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4368 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4370 cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4371 vmx->vcpu.arch.cr0 = cr0;
4372 vmx_set_cr0(vcpu, cr0); /* enter rmode */
4373 vmx_set_cr4(vcpu, 0);
4374 vmx_set_efer(vcpu, 0);
4376 update_exception_bitmap(vcpu);
4378 vpid_sync_context(vmx->vpid);
4380 vmx_clear_hlt(vcpu);
4383 static void enable_irq_window(struct kvm_vcpu *vcpu)
4385 exec_controls_setbit(to_vmx(vcpu), CPU_BASED_VIRTUAL_INTR_PENDING);
4388 static void enable_nmi_window(struct kvm_vcpu *vcpu)
4391 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4392 enable_irq_window(vcpu);
4396 exec_controls_setbit(to_vmx(vcpu), CPU_BASED_VIRTUAL_NMI_PENDING);
4399 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4401 struct vcpu_vmx *vmx = to_vmx(vcpu);
4403 int irq = vcpu->arch.interrupt.nr;
4405 trace_kvm_inj_virq(irq);
4407 ++vcpu->stat.irq_injections;
4408 if (vmx->rmode.vm86_active) {
4410 if (vcpu->arch.interrupt.soft)
4411 inc_eip = vcpu->arch.event_exit_inst_len;
4412 kvm_inject_realmode_interrupt(vcpu, irq, inc_eip);
4415 intr = irq | INTR_INFO_VALID_MASK;
4416 if (vcpu->arch.interrupt.soft) {
4417 intr |= INTR_TYPE_SOFT_INTR;
4418 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4419 vmx->vcpu.arch.event_exit_inst_len);
4421 intr |= INTR_TYPE_EXT_INTR;
4422 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4424 vmx_clear_hlt(vcpu);
4427 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4429 struct vcpu_vmx *vmx = to_vmx(vcpu);
4433 * Tracking the NMI-blocked state in software is built upon
4434 * finding the next open IRQ window. This, in turn, depends on
4435 * well-behaving guests: They have to keep IRQs disabled at
4436 * least as long as the NMI handler runs. Otherwise we may
4437 * cause NMI nesting, maybe breaking the guest. But as this is
4438 * highly unlikely, we can live with the residual risk.
4440 vmx->loaded_vmcs->soft_vnmi_blocked = 1;
4441 vmx->loaded_vmcs->vnmi_blocked_time = 0;
4444 ++vcpu->stat.nmi_injections;
4445 vmx->loaded_vmcs->nmi_known_unmasked = false;
4447 if (vmx->rmode.vm86_active) {
4448 kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0);
4452 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4453 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4455 vmx_clear_hlt(vcpu);
4458 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4460 struct vcpu_vmx *vmx = to_vmx(vcpu);
4464 return vmx->loaded_vmcs->soft_vnmi_blocked;
4465 if (vmx->loaded_vmcs->nmi_known_unmasked)
4467 masked = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4468 vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4472 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4474 struct vcpu_vmx *vmx = to_vmx(vcpu);
4477 if (vmx->loaded_vmcs->soft_vnmi_blocked != masked) {
4478 vmx->loaded_vmcs->soft_vnmi_blocked = masked;
4479 vmx->loaded_vmcs->vnmi_blocked_time = 0;
4482 vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4484 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4485 GUEST_INTR_STATE_NMI);
4487 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4488 GUEST_INTR_STATE_NMI);
4492 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
4494 if (to_vmx(vcpu)->nested.nested_run_pending)
4498 to_vmx(vcpu)->loaded_vmcs->soft_vnmi_blocked)
4501 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4502 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
4503 | GUEST_INTR_STATE_NMI));
4506 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
4508 if (to_vmx(vcpu)->nested.nested_run_pending)
4511 if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu))
4514 return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
4515 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4516 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4519 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4523 if (enable_unrestricted_guest)
4526 ret = x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
4530 to_kvm_vmx(kvm)->tss_addr = addr;
4531 return init_rmode_tss(kvm);
4534 static int vmx_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
4536 to_kvm_vmx(kvm)->ept_identity_map_addr = ident_addr;
4540 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
4545 * Update instruction length as we may reinject the exception
4546 * from user space while in guest debugging mode.
4548 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4549 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4550 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4554 if (vcpu->guest_debug &
4555 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
4572 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4573 int vec, u32 err_code)
4576 * Instruction with address size override prefix opcode 0x67
4577 * Cause the #SS fault with 0 error code in VM86 mode.
4579 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
4580 if (kvm_emulate_instruction(vcpu, 0)) {
4581 if (vcpu->arch.halt_request) {
4582 vcpu->arch.halt_request = 0;
4583 return kvm_vcpu_halt(vcpu);
4591 * Forward all other exceptions that are valid in real mode.
4592 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4593 * the required debugging infrastructure rework.
4595 kvm_queue_exception(vcpu, vec);
4600 * Trigger machine check on the host. We assume all the MSRs are already set up
4601 * by the CPU and that we still run on the same CPU as the MCE occurred on.
4602 * We pass a fake environment to the machine check handler because we want
4603 * the guest to be always treated like user space, no matter what context
4604 * it used internally.
4606 static void kvm_machine_check(void)
4608 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
4609 struct pt_regs regs = {
4610 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
4611 .flags = X86_EFLAGS_IF,
4614 do_machine_check(®s, 0);
4618 static int handle_machine_check(struct kvm_vcpu *vcpu)
4620 /* handled by vmx_vcpu_run() */
4624 static int handle_exception_nmi(struct kvm_vcpu *vcpu)
4626 struct vcpu_vmx *vmx = to_vmx(vcpu);
4627 struct kvm_run *kvm_run = vcpu->run;
4628 u32 intr_info, ex_no, error_code;
4629 unsigned long cr2, rip, dr6;
4632 vect_info = vmx->idt_vectoring_info;
4633 intr_info = vmx->exit_intr_info;
4635 if (is_machine_check(intr_info) || is_nmi(intr_info))
4636 return 1; /* handled by handle_exception_nmi_irqoff() */
4638 if (is_invalid_opcode(intr_info))
4639 return handle_ud(vcpu);
4642 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4643 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4645 if (!vmx->rmode.vm86_active && is_gp_fault(intr_info)) {
4646 WARN_ON_ONCE(!enable_vmware_backdoor);
4649 * VMware backdoor emulation on #GP interception only handles
4650 * IN{S}, OUT{S}, and RDPMC, none of which generate a non-zero
4651 * error code on #GP.
4654 kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
4657 return kvm_emulate_instruction(vcpu, EMULTYPE_VMWARE_GP);
4661 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
4662 * MMIO, it is better to report an internal error.
4663 * See the comments in vmx_handle_exit.
4665 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4666 !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
4667 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4668 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4669 vcpu->run->internal.ndata = 3;
4670 vcpu->run->internal.data[0] = vect_info;
4671 vcpu->run->internal.data[1] = intr_info;
4672 vcpu->run->internal.data[2] = error_code;
4676 if (is_page_fault(intr_info)) {
4677 cr2 = vmcs_readl(EXIT_QUALIFICATION);
4678 /* EPT won't cause page fault directly */
4679 WARN_ON_ONCE(!vcpu->arch.apf.host_apf_reason && enable_ept);
4680 return kvm_handle_page_fault(vcpu, error_code, cr2, NULL, 0);
4683 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4685 if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
4686 return handle_rmode_exception(vcpu, ex_no, error_code);
4690 kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
4693 dr6 = vmcs_readl(EXIT_QUALIFICATION);
4694 if (!(vcpu->guest_debug &
4695 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4696 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
4697 vcpu->arch.dr6 |= dr6 | DR6_RTM;
4698 if (is_icebp(intr_info))
4699 WARN_ON(!skip_emulated_instruction(vcpu));
4701 kvm_queue_exception(vcpu, DB_VECTOR);
4704 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
4705 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4709 * Update instruction length as we may reinject #BP from
4710 * user space while in guest debugging mode. Reading it for
4711 * #DB as well causes no harm, it is not used in that case.
4713 vmx->vcpu.arch.event_exit_inst_len =
4714 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4715 kvm_run->exit_reason = KVM_EXIT_DEBUG;
4716 rip = kvm_rip_read(vcpu);
4717 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4718 kvm_run->debug.arch.exception = ex_no;
4721 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4722 kvm_run->ex.exception = ex_no;
4723 kvm_run->ex.error_code = error_code;
4729 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
4731 ++vcpu->stat.irq_exits;
4735 static int handle_triple_fault(struct kvm_vcpu *vcpu)
4737 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4738 vcpu->mmio_needed = 0;
4742 static int handle_io(struct kvm_vcpu *vcpu)
4744 unsigned long exit_qualification;
4745 int size, in, string;
4748 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4749 string = (exit_qualification & 16) != 0;
4751 ++vcpu->stat.io_exits;
4754 return kvm_emulate_instruction(vcpu, 0);
4756 port = exit_qualification >> 16;
4757 size = (exit_qualification & 7) + 1;
4758 in = (exit_qualification & 8) != 0;
4760 return kvm_fast_pio(vcpu, size, port, in);
4764 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4767 * Patch in the VMCALL instruction:
4769 hypercall[0] = 0x0f;
4770 hypercall[1] = 0x01;
4771 hypercall[2] = 0xc1;
4774 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
4775 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
4777 if (is_guest_mode(vcpu)) {
4778 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4779 unsigned long orig_val = val;
4782 * We get here when L2 changed cr0 in a way that did not change
4783 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
4784 * but did change L0 shadowed bits. So we first calculate the
4785 * effective cr0 value that L1 would like to write into the
4786 * hardware. It consists of the L2-owned bits from the new
4787 * value combined with the L1-owned bits from L1's guest_cr0.
4789 val = (val & ~vmcs12->cr0_guest_host_mask) |
4790 (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
4792 if (!nested_guest_cr0_valid(vcpu, val))
4795 if (kvm_set_cr0(vcpu, val))
4797 vmcs_writel(CR0_READ_SHADOW, orig_val);
4800 if (to_vmx(vcpu)->nested.vmxon &&
4801 !nested_host_cr0_valid(vcpu, val))
4804 return kvm_set_cr0(vcpu, val);
4808 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
4810 if (is_guest_mode(vcpu)) {
4811 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4812 unsigned long orig_val = val;
4814 /* analogously to handle_set_cr0 */
4815 val = (val & ~vmcs12->cr4_guest_host_mask) |
4816 (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
4817 if (kvm_set_cr4(vcpu, val))
4819 vmcs_writel(CR4_READ_SHADOW, orig_val);
4822 return kvm_set_cr4(vcpu, val);
4825 static int handle_desc(struct kvm_vcpu *vcpu)
4827 WARN_ON(!(vcpu->arch.cr4 & X86_CR4_UMIP));
4828 return kvm_emulate_instruction(vcpu, 0);
4831 static int handle_cr(struct kvm_vcpu *vcpu)
4833 unsigned long exit_qualification, val;
4839 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4840 cr = exit_qualification & 15;
4841 reg = (exit_qualification >> 8) & 15;
4842 switch ((exit_qualification >> 4) & 3) {
4843 case 0: /* mov to cr */
4844 val = kvm_register_readl(vcpu, reg);
4845 trace_kvm_cr_write(cr, val);
4848 err = handle_set_cr0(vcpu, val);
4849 return kvm_complete_insn_gp(vcpu, err);
4851 WARN_ON_ONCE(enable_unrestricted_guest);
4852 err = kvm_set_cr3(vcpu, val);
4853 return kvm_complete_insn_gp(vcpu, err);
4855 err = handle_set_cr4(vcpu, val);
4856 return kvm_complete_insn_gp(vcpu, err);
4858 u8 cr8_prev = kvm_get_cr8(vcpu);
4860 err = kvm_set_cr8(vcpu, cr8);
4861 ret = kvm_complete_insn_gp(vcpu, err);
4862 if (lapic_in_kernel(vcpu))
4864 if (cr8_prev <= cr8)
4867 * TODO: we might be squashing a
4868 * KVM_GUESTDBG_SINGLESTEP-triggered
4869 * KVM_EXIT_DEBUG here.
4871 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
4877 WARN_ONCE(1, "Guest should always own CR0.TS");
4878 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4879 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
4880 return kvm_skip_emulated_instruction(vcpu);
4881 case 1: /*mov from cr*/
4884 WARN_ON_ONCE(enable_unrestricted_guest);
4885 val = kvm_read_cr3(vcpu);
4886 kvm_register_write(vcpu, reg, val);
4887 trace_kvm_cr_read(cr, val);
4888 return kvm_skip_emulated_instruction(vcpu);
4890 val = kvm_get_cr8(vcpu);
4891 kvm_register_write(vcpu, reg, val);
4892 trace_kvm_cr_read(cr, val);
4893 return kvm_skip_emulated_instruction(vcpu);
4897 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
4898 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
4899 kvm_lmsw(vcpu, val);
4901 return kvm_skip_emulated_instruction(vcpu);
4905 vcpu->run->exit_reason = 0;
4906 vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
4907 (int)(exit_qualification >> 4) & 3, cr);
4911 static int handle_dr(struct kvm_vcpu *vcpu)
4913 unsigned long exit_qualification;
4916 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4917 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
4919 /* First, if DR does not exist, trigger UD */
4920 if (!kvm_require_dr(vcpu, dr))
4923 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
4924 if (!kvm_require_cpl(vcpu, 0))
4926 dr7 = vmcs_readl(GUEST_DR7);
4929 * As the vm-exit takes precedence over the debug trap, we
4930 * need to emulate the latter, either for the host or the
4931 * guest debugging itself.
4933 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
4934 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
4935 vcpu->run->debug.arch.dr7 = dr7;
4936 vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
4937 vcpu->run->debug.arch.exception = DB_VECTOR;
4938 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
4941 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
4942 vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
4943 kvm_queue_exception(vcpu, DB_VECTOR);
4948 if (vcpu->guest_debug == 0) {
4949 exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
4952 * No more DR vmexits; force a reload of the debug registers
4953 * and reenter on this instruction. The next vmexit will
4954 * retrieve the full state of the debug registers.
4956 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
4960 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
4961 if (exit_qualification & TYPE_MOV_FROM_DR) {
4964 if (kvm_get_dr(vcpu, dr, &val))
4966 kvm_register_write(vcpu, reg, val);
4968 if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
4971 return kvm_skip_emulated_instruction(vcpu);
4974 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
4976 return vcpu->arch.dr6;
4979 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
4983 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
4985 get_debugreg(vcpu->arch.db[0], 0);
4986 get_debugreg(vcpu->arch.db[1], 1);
4987 get_debugreg(vcpu->arch.db[2], 2);
4988 get_debugreg(vcpu->arch.db[3], 3);
4989 get_debugreg(vcpu->arch.dr6, 6);
4990 vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
4992 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
4993 exec_controls_setbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
4996 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
4998 vmcs_writel(GUEST_DR7, val);
5001 static int handle_cpuid(struct kvm_vcpu *vcpu)
5003 return kvm_emulate_cpuid(vcpu);
5006 static int handle_rdmsr(struct kvm_vcpu *vcpu)
5008 return kvm_emulate_rdmsr(vcpu);
5011 static int handle_wrmsr(struct kvm_vcpu *vcpu)
5013 return kvm_emulate_wrmsr(vcpu);
5016 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
5018 kvm_apic_update_ppr(vcpu);
5022 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
5024 exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_VIRTUAL_INTR_PENDING);
5026 kvm_make_request(KVM_REQ_EVENT, vcpu);
5028 ++vcpu->stat.irq_window_exits;
5032 static int handle_halt(struct kvm_vcpu *vcpu)
5034 return kvm_emulate_halt(vcpu);
5037 static int handle_vmcall(struct kvm_vcpu *vcpu)
5039 return kvm_emulate_hypercall(vcpu);
5042 static int handle_invd(struct kvm_vcpu *vcpu)
5044 return kvm_emulate_instruction(vcpu, 0);
5047 static int handle_invlpg(struct kvm_vcpu *vcpu)
5049 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5051 kvm_mmu_invlpg(vcpu, exit_qualification);
5052 return kvm_skip_emulated_instruction(vcpu);
5055 static int handle_rdpmc(struct kvm_vcpu *vcpu)
5059 err = kvm_rdpmc(vcpu);
5060 return kvm_complete_insn_gp(vcpu, err);
5063 static int handle_wbinvd(struct kvm_vcpu *vcpu)
5065 return kvm_emulate_wbinvd(vcpu);
5068 static int handle_xsetbv(struct kvm_vcpu *vcpu)
5070 u64 new_bv = kvm_read_edx_eax(vcpu);
5071 u32 index = kvm_rcx_read(vcpu);
5073 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
5074 return kvm_skip_emulated_instruction(vcpu);
5078 static int handle_apic_access(struct kvm_vcpu *vcpu)
5080 if (likely(fasteoi)) {
5081 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5082 int access_type, offset;
5084 access_type = exit_qualification & APIC_ACCESS_TYPE;
5085 offset = exit_qualification & APIC_ACCESS_OFFSET;
5087 * Sane guest uses MOV to write EOI, with written value
5088 * not cared. So make a short-circuit here by avoiding
5089 * heavy instruction emulation.
5091 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5092 (offset == APIC_EOI)) {
5093 kvm_lapic_set_eoi(vcpu);
5094 return kvm_skip_emulated_instruction(vcpu);
5097 return kvm_emulate_instruction(vcpu, 0);
5100 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5102 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5103 int vector = exit_qualification & 0xff;
5105 /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5106 kvm_apic_set_eoi_accelerated(vcpu, vector);
5110 static int handle_apic_write(struct kvm_vcpu *vcpu)
5112 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5113 u32 offset = exit_qualification & 0xfff;
5115 /* APIC-write VM exit is trap-like and thus no need to adjust IP */
5116 kvm_apic_write_nodecode(vcpu, offset);
5120 static int handle_task_switch(struct kvm_vcpu *vcpu)
5122 struct vcpu_vmx *vmx = to_vmx(vcpu);
5123 unsigned long exit_qualification;
5124 bool has_error_code = false;
5127 int reason, type, idt_v, idt_index;
5129 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5130 idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5131 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5133 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5135 reason = (u32)exit_qualification >> 30;
5136 if (reason == TASK_SWITCH_GATE && idt_v) {
5138 case INTR_TYPE_NMI_INTR:
5139 vcpu->arch.nmi_injected = false;
5140 vmx_set_nmi_mask(vcpu, true);
5142 case INTR_TYPE_EXT_INTR:
5143 case INTR_TYPE_SOFT_INTR:
5144 kvm_clear_interrupt_queue(vcpu);
5146 case INTR_TYPE_HARD_EXCEPTION:
5147 if (vmx->idt_vectoring_info &
5148 VECTORING_INFO_DELIVER_CODE_MASK) {
5149 has_error_code = true;
5151 vmcs_read32(IDT_VECTORING_ERROR_CODE);
5154 case INTR_TYPE_SOFT_EXCEPTION:
5155 kvm_clear_exception_queue(vcpu);
5161 tss_selector = exit_qualification;
5163 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5164 type != INTR_TYPE_EXT_INTR &&
5165 type != INTR_TYPE_NMI_INTR))
5166 WARN_ON(!skip_emulated_instruction(vcpu));
5169 * TODO: What about debug traps on tss switch?
5170 * Are we supposed to inject them and update dr6?
5172 return kvm_task_switch(vcpu, tss_selector,
5173 type == INTR_TYPE_SOFT_INTR ? idt_index : -1,
5174 reason, has_error_code, error_code);
5177 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5179 unsigned long exit_qualification;
5183 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5186 * EPT violation happened while executing iret from NMI,
5187 * "blocked by NMI" bit has to be set before next VM entry.
5188 * There are errata that may cause this bit to not be set:
5191 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5193 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
5194 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5196 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5197 trace_kvm_page_fault(gpa, exit_qualification);
5199 /* Is it a read fault? */
5200 error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
5201 ? PFERR_USER_MASK : 0;
5202 /* Is it a write fault? */
5203 error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
5204 ? PFERR_WRITE_MASK : 0;
5205 /* Is it a fetch fault? */
5206 error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
5207 ? PFERR_FETCH_MASK : 0;
5208 /* ept page table entry is present? */
5209 error_code |= (exit_qualification &
5210 (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE |
5211 EPT_VIOLATION_EXECUTABLE))
5212 ? PFERR_PRESENT_MASK : 0;
5214 error_code |= (exit_qualification & 0x100) != 0 ?
5215 PFERR_GUEST_FINAL_MASK : PFERR_GUEST_PAGE_MASK;
5217 vcpu->arch.exit_qualification = exit_qualification;
5218 return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5221 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5226 * A nested guest cannot optimize MMIO vmexits, because we have an
5227 * nGPA here instead of the required GPA.
5229 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5230 if (!is_guest_mode(vcpu) &&
5231 !kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5232 trace_kvm_fast_mmio(gpa);
5233 return kvm_skip_emulated_instruction(vcpu);
5236 return kvm_mmu_page_fault(vcpu, gpa, PFERR_RSVD_MASK, NULL, 0);
5239 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5241 WARN_ON_ONCE(!enable_vnmi);
5242 exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_VIRTUAL_NMI_PENDING);
5243 ++vcpu->stat.nmi_window_exits;
5244 kvm_make_request(KVM_REQ_EVENT, vcpu);
5249 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5251 struct vcpu_vmx *vmx = to_vmx(vcpu);
5252 bool intr_window_requested;
5253 unsigned count = 130;
5256 * We should never reach the point where we are emulating L2
5257 * due to invalid guest state as that means we incorrectly
5258 * allowed a nested VMEntry with an invalid vmcs12.
5260 WARN_ON_ONCE(vmx->emulation_required && vmx->nested.nested_run_pending);
5262 intr_window_requested = exec_controls_get(vmx) &
5263 CPU_BASED_VIRTUAL_INTR_PENDING;
5265 while (vmx->emulation_required && count-- != 0) {
5266 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
5267 return handle_interrupt_window(&vmx->vcpu);
5269 if (kvm_test_request(KVM_REQ_EVENT, vcpu))
5272 if (!kvm_emulate_instruction(vcpu, 0))
5275 if (vmx->emulation_required && !vmx->rmode.vm86_active &&
5276 vcpu->arch.exception.pending) {
5277 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5278 vcpu->run->internal.suberror =
5279 KVM_INTERNAL_ERROR_EMULATION;
5280 vcpu->run->internal.ndata = 0;
5284 if (vcpu->arch.halt_request) {
5285 vcpu->arch.halt_request = 0;
5286 return kvm_vcpu_halt(vcpu);
5290 * Note, return 1 and not 0, vcpu_run() is responsible for
5291 * morphing the pending signal into the proper return code.
5293 if (signal_pending(current))
5303 static void grow_ple_window(struct kvm_vcpu *vcpu)
5305 struct vcpu_vmx *vmx = to_vmx(vcpu);
5306 unsigned int old = vmx->ple_window;
5308 vmx->ple_window = __grow_ple_window(old, ple_window,
5312 if (vmx->ple_window != old) {
5313 vmx->ple_window_dirty = true;
5314 trace_kvm_ple_window_update(vcpu->vcpu_id,
5315 vmx->ple_window, old);
5319 static void shrink_ple_window(struct kvm_vcpu *vcpu)
5321 struct vcpu_vmx *vmx = to_vmx(vcpu);
5322 unsigned int old = vmx->ple_window;
5324 vmx->ple_window = __shrink_ple_window(old, ple_window,
5328 if (vmx->ple_window != old) {
5329 vmx->ple_window_dirty = true;
5330 trace_kvm_ple_window_update(vcpu->vcpu_id,
5331 vmx->ple_window, old);
5336 * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
5338 static void wakeup_handler(void)
5340 struct kvm_vcpu *vcpu;
5341 int cpu = smp_processor_id();
5343 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5344 list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
5345 blocked_vcpu_list) {
5346 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
5348 if (pi_test_on(pi_desc) == 1)
5349 kvm_vcpu_kick(vcpu);
5351 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5354 static void vmx_enable_tdp(void)
5356 kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK,
5357 enable_ept_ad_bits ? VMX_EPT_ACCESS_BIT : 0ull,
5358 enable_ept_ad_bits ? VMX_EPT_DIRTY_BIT : 0ull,
5359 0ull, VMX_EPT_EXECUTABLE_MASK,
5360 cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK,
5361 VMX_EPT_RWX_MASK, 0ull);
5363 ept_set_mmio_spte_mask();
5368 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5369 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5371 static int handle_pause(struct kvm_vcpu *vcpu)
5373 if (!kvm_pause_in_guest(vcpu->kvm))
5374 grow_ple_window(vcpu);
5377 * Intel sdm vol3 ch-25.1.3 says: The "PAUSE-loop exiting"
5378 * VM-execution control is ignored if CPL > 0. OTOH, KVM
5379 * never set PAUSE_EXITING and just set PLE if supported,
5380 * so the vcpu must be CPL=0 if it gets a PAUSE exit.
5382 kvm_vcpu_on_spin(vcpu, true);
5383 return kvm_skip_emulated_instruction(vcpu);
5386 static int handle_nop(struct kvm_vcpu *vcpu)
5388 return kvm_skip_emulated_instruction(vcpu);
5391 static int handle_mwait(struct kvm_vcpu *vcpu)
5393 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
5394 return handle_nop(vcpu);
5397 static int handle_invalid_op(struct kvm_vcpu *vcpu)
5399 kvm_queue_exception(vcpu, UD_VECTOR);
5403 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
5408 static int handle_monitor(struct kvm_vcpu *vcpu)
5410 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
5411 return handle_nop(vcpu);
5414 static int handle_invpcid(struct kvm_vcpu *vcpu)
5416 u32 vmx_instruction_info;
5420 struct x86_exception e;
5422 unsigned long roots_to_free = 0;
5428 if (!guest_cpuid_has(vcpu, X86_FEATURE_INVPCID)) {
5429 kvm_queue_exception(vcpu, UD_VECTOR);
5433 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5434 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
5437 kvm_inject_gp(vcpu, 0);
5441 /* According to the Intel instruction reference, the memory operand
5442 * is read even if it isn't needed (e.g., for type==all)
5444 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5445 vmx_instruction_info, false,
5446 sizeof(operand), &gva))
5449 if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
5450 kvm_inject_page_fault(vcpu, &e);
5454 if (operand.pcid >> 12 != 0) {
5455 kvm_inject_gp(vcpu, 0);
5459 pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
5462 case INVPCID_TYPE_INDIV_ADDR:
5463 if ((!pcid_enabled && (operand.pcid != 0)) ||
5464 is_noncanonical_address(operand.gla, vcpu)) {
5465 kvm_inject_gp(vcpu, 0);
5468 kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
5469 return kvm_skip_emulated_instruction(vcpu);
5471 case INVPCID_TYPE_SINGLE_CTXT:
5472 if (!pcid_enabled && (operand.pcid != 0)) {
5473 kvm_inject_gp(vcpu, 0);
5477 if (kvm_get_active_pcid(vcpu) == operand.pcid) {
5478 kvm_mmu_sync_roots(vcpu);
5479 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
5482 for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
5483 if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].cr3)
5485 roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
5487 kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
5489 * If neither the current cr3 nor any of the prev_roots use the
5490 * given PCID, then nothing needs to be done here because a
5491 * resync will happen anyway before switching to any other CR3.
5494 return kvm_skip_emulated_instruction(vcpu);
5496 case INVPCID_TYPE_ALL_NON_GLOBAL:
5498 * Currently, KVM doesn't mark global entries in the shadow
5499 * page tables, so a non-global flush just degenerates to a
5500 * global flush. If needed, we could optimize this later by
5501 * keeping track of global entries in shadow page tables.
5505 case INVPCID_TYPE_ALL_INCL_GLOBAL:
5506 kvm_mmu_unload(vcpu);
5507 return kvm_skip_emulated_instruction(vcpu);
5510 BUG(); /* We have already checked above that type <= 3 */
5514 static int handle_pml_full(struct kvm_vcpu *vcpu)
5516 unsigned long exit_qualification;
5518 trace_kvm_pml_full(vcpu->vcpu_id);
5520 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5523 * PML buffer FULL happened while executing iret from NMI,
5524 * "blocked by NMI" bit has to be set before next VM entry.
5526 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5528 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
5529 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5530 GUEST_INTR_STATE_NMI);
5533 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
5534 * here.., and there's no userspace involvement needed for PML.
5539 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
5541 struct vcpu_vmx *vmx = to_vmx(vcpu);
5543 if (!vmx->req_immediate_exit &&
5544 !unlikely(vmx->loaded_vmcs->hv_timer_soft_disabled))
5545 kvm_lapic_expired_hv_timer(vcpu);
5551 * When nested=0, all VMX instruction VM Exits filter here. The handlers
5552 * are overwritten by nested_vmx_setup() when nested=1.
5554 static int handle_vmx_instruction(struct kvm_vcpu *vcpu)
5556 kvm_queue_exception(vcpu, UD_VECTOR);
5560 static int handle_encls(struct kvm_vcpu *vcpu)
5563 * SGX virtualization is not yet supported. There is no software
5564 * enable bit for SGX, so we have to trap ENCLS and inject a #UD
5565 * to prevent the guest from executing ENCLS.
5567 kvm_queue_exception(vcpu, UD_VECTOR);
5572 * The exit handlers return 1 if the exit was handled fully and guest execution
5573 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
5574 * to be done to userspace and return 0.
5576 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
5577 [EXIT_REASON_EXCEPTION_NMI] = handle_exception_nmi,
5578 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
5579 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
5580 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
5581 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
5582 [EXIT_REASON_CR_ACCESS] = handle_cr,
5583 [EXIT_REASON_DR_ACCESS] = handle_dr,
5584 [EXIT_REASON_CPUID] = handle_cpuid,
5585 [EXIT_REASON_MSR_READ] = handle_rdmsr,
5586 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
5587 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
5588 [EXIT_REASON_HLT] = handle_halt,
5589 [EXIT_REASON_INVD] = handle_invd,
5590 [EXIT_REASON_INVLPG] = handle_invlpg,
5591 [EXIT_REASON_RDPMC] = handle_rdpmc,
5592 [EXIT_REASON_VMCALL] = handle_vmcall,
5593 [EXIT_REASON_VMCLEAR] = handle_vmx_instruction,
5594 [EXIT_REASON_VMLAUNCH] = handle_vmx_instruction,
5595 [EXIT_REASON_VMPTRLD] = handle_vmx_instruction,
5596 [EXIT_REASON_VMPTRST] = handle_vmx_instruction,
5597 [EXIT_REASON_VMREAD] = handle_vmx_instruction,
5598 [EXIT_REASON_VMRESUME] = handle_vmx_instruction,
5599 [EXIT_REASON_VMWRITE] = handle_vmx_instruction,
5600 [EXIT_REASON_VMOFF] = handle_vmx_instruction,
5601 [EXIT_REASON_VMON] = handle_vmx_instruction,
5602 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
5603 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
5604 [EXIT_REASON_APIC_WRITE] = handle_apic_write,
5605 [EXIT_REASON_EOI_INDUCED] = handle_apic_eoi_induced,
5606 [EXIT_REASON_WBINVD] = handle_wbinvd,
5607 [EXIT_REASON_XSETBV] = handle_xsetbv,
5608 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
5609 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
5610 [EXIT_REASON_GDTR_IDTR] = handle_desc,
5611 [EXIT_REASON_LDTR_TR] = handle_desc,
5612 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
5613 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
5614 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
5615 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_mwait,
5616 [EXIT_REASON_MONITOR_TRAP_FLAG] = handle_monitor_trap,
5617 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_monitor,
5618 [EXIT_REASON_INVEPT] = handle_vmx_instruction,
5619 [EXIT_REASON_INVVPID] = handle_vmx_instruction,
5620 [EXIT_REASON_RDRAND] = handle_invalid_op,
5621 [EXIT_REASON_RDSEED] = handle_invalid_op,
5622 [EXIT_REASON_PML_FULL] = handle_pml_full,
5623 [EXIT_REASON_INVPCID] = handle_invpcid,
5624 [EXIT_REASON_VMFUNC] = handle_vmx_instruction,
5625 [EXIT_REASON_PREEMPTION_TIMER] = handle_preemption_timer,
5626 [EXIT_REASON_ENCLS] = handle_encls,
5629 static const int kvm_vmx_max_exit_handlers =
5630 ARRAY_SIZE(kvm_vmx_exit_handlers);
5632 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
5634 *info1 = vmcs_readl(EXIT_QUALIFICATION);
5635 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
5638 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
5641 __free_page(vmx->pml_pg);
5646 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
5648 struct vcpu_vmx *vmx = to_vmx(vcpu);
5652 pml_idx = vmcs_read16(GUEST_PML_INDEX);
5654 /* Do nothing if PML buffer is empty */
5655 if (pml_idx == (PML_ENTITY_NUM - 1))
5658 /* PML index always points to next available PML buffer entity */
5659 if (pml_idx >= PML_ENTITY_NUM)
5664 pml_buf = page_address(vmx->pml_pg);
5665 for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
5668 gpa = pml_buf[pml_idx];
5669 WARN_ON(gpa & (PAGE_SIZE - 1));
5670 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5673 /* reset PML index */
5674 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
5678 * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
5679 * Called before reporting dirty_bitmap to userspace.
5681 static void kvm_flush_pml_buffers(struct kvm *kvm)
5684 struct kvm_vcpu *vcpu;
5686 * We only need to kick vcpu out of guest mode here, as PML buffer
5687 * is flushed at beginning of all VMEXITs, and it's obvious that only
5688 * vcpus running in guest are possible to have unflushed GPAs in PML
5691 kvm_for_each_vcpu(i, vcpu, kvm)
5692 kvm_vcpu_kick(vcpu);
5695 static void vmx_dump_sel(char *name, uint32_t sel)
5697 pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
5698 name, vmcs_read16(sel),
5699 vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
5700 vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
5701 vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
5704 static void vmx_dump_dtsel(char *name, uint32_t limit)
5706 pr_err("%s limit=0x%08x, base=0x%016lx\n",
5707 name, vmcs_read32(limit),
5708 vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
5711 void dump_vmcs(void)
5713 u32 vmentry_ctl, vmexit_ctl;
5714 u32 cpu_based_exec_ctrl, pin_based_exec_ctrl, secondary_exec_control;
5719 if (!dump_invalid_vmcs) {
5720 pr_warn_ratelimited("set kvm_intel.dump_invalid_vmcs=1 to dump internal KVM state.\n");
5724 vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
5725 vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
5726 cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5727 pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
5728 cr4 = vmcs_readl(GUEST_CR4);
5729 efer = vmcs_read64(GUEST_IA32_EFER);
5730 secondary_exec_control = 0;
5731 if (cpu_has_secondary_exec_ctrls())
5732 secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
5734 pr_err("*** Guest State ***\n");
5735 pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5736 vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
5737 vmcs_readl(CR0_GUEST_HOST_MASK));
5738 pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5739 cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
5740 pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
5741 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
5742 (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
5744 pr_err("PDPTR0 = 0x%016llx PDPTR1 = 0x%016llx\n",
5745 vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
5746 pr_err("PDPTR2 = 0x%016llx PDPTR3 = 0x%016llx\n",
5747 vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
5749 pr_err("RSP = 0x%016lx RIP = 0x%016lx\n",
5750 vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
5751 pr_err("RFLAGS=0x%08lx DR7 = 0x%016lx\n",
5752 vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
5753 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5754 vmcs_readl(GUEST_SYSENTER_ESP),
5755 vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
5756 vmx_dump_sel("CS: ", GUEST_CS_SELECTOR);
5757 vmx_dump_sel("DS: ", GUEST_DS_SELECTOR);
5758 vmx_dump_sel("SS: ", GUEST_SS_SELECTOR);
5759 vmx_dump_sel("ES: ", GUEST_ES_SELECTOR);
5760 vmx_dump_sel("FS: ", GUEST_FS_SELECTOR);
5761 vmx_dump_sel("GS: ", GUEST_GS_SELECTOR);
5762 vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
5763 vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
5764 vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
5765 vmx_dump_sel("TR: ", GUEST_TR_SELECTOR);
5766 if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
5767 (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
5768 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
5769 efer, vmcs_read64(GUEST_IA32_PAT));
5770 pr_err("DebugCtl = 0x%016llx DebugExceptions = 0x%016lx\n",
5771 vmcs_read64(GUEST_IA32_DEBUGCTL),
5772 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
5773 if (cpu_has_load_perf_global_ctrl() &&
5774 vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
5775 pr_err("PerfGlobCtl = 0x%016llx\n",
5776 vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
5777 if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
5778 pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
5779 pr_err("Interruptibility = %08x ActivityState = %08x\n",
5780 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
5781 vmcs_read32(GUEST_ACTIVITY_STATE));
5782 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
5783 pr_err("InterruptStatus = %04x\n",
5784 vmcs_read16(GUEST_INTR_STATUS));
5786 pr_err("*** Host State ***\n");
5787 pr_err("RIP = 0x%016lx RSP = 0x%016lx\n",
5788 vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
5789 pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
5790 vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
5791 vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
5792 vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
5793 vmcs_read16(HOST_TR_SELECTOR));
5794 pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
5795 vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
5796 vmcs_readl(HOST_TR_BASE));
5797 pr_err("GDTBase=%016lx IDTBase=%016lx\n",
5798 vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
5799 pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
5800 vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
5801 vmcs_readl(HOST_CR4));
5802 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5803 vmcs_readl(HOST_IA32_SYSENTER_ESP),
5804 vmcs_read32(HOST_IA32_SYSENTER_CS),
5805 vmcs_readl(HOST_IA32_SYSENTER_EIP));
5806 if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
5807 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
5808 vmcs_read64(HOST_IA32_EFER),
5809 vmcs_read64(HOST_IA32_PAT));
5810 if (cpu_has_load_perf_global_ctrl() &&
5811 vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
5812 pr_err("PerfGlobCtl = 0x%016llx\n",
5813 vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
5815 pr_err("*** Control State ***\n");
5816 pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
5817 pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
5818 pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
5819 pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
5820 vmcs_read32(EXCEPTION_BITMAP),
5821 vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
5822 vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
5823 pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
5824 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
5825 vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
5826 vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
5827 pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
5828 vmcs_read32(VM_EXIT_INTR_INFO),
5829 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
5830 vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
5831 pr_err(" reason=%08x qualification=%016lx\n",
5832 vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
5833 pr_err("IDTVectoring: info=%08x errcode=%08x\n",
5834 vmcs_read32(IDT_VECTORING_INFO_FIELD),
5835 vmcs_read32(IDT_VECTORING_ERROR_CODE));
5836 pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
5837 if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
5838 pr_err("TSC Multiplier = 0x%016llx\n",
5839 vmcs_read64(TSC_MULTIPLIER));
5840 if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW) {
5841 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
5842 u16 status = vmcs_read16(GUEST_INTR_STATUS);
5843 pr_err("SVI|RVI = %02x|%02x ", status >> 8, status & 0xff);
5845 pr_cont("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
5846 if (secondary_exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
5847 pr_err("APIC-access addr = 0x%016llx ", vmcs_read64(APIC_ACCESS_ADDR));
5848 pr_cont("virt-APIC addr = 0x%016llx\n", vmcs_read64(VIRTUAL_APIC_PAGE_ADDR));
5850 if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
5851 pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
5852 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
5853 pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
5854 n = vmcs_read32(CR3_TARGET_COUNT);
5855 for (i = 0; i + 1 < n; i += 4)
5856 pr_err("CR3 target%u=%016lx target%u=%016lx\n",
5857 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
5858 i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
5860 pr_err("CR3 target%u=%016lx\n",
5861 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
5862 if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
5863 pr_err("PLE Gap=%08x Window=%08x\n",
5864 vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
5865 if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
5866 pr_err("Virtual processor ID = 0x%04x\n",
5867 vmcs_read16(VIRTUAL_PROCESSOR_ID));
5871 * The guest has exited. See if we can fix it or if we need userspace
5874 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
5876 struct vcpu_vmx *vmx = to_vmx(vcpu);
5877 u32 exit_reason = vmx->exit_reason;
5878 u32 vectoring_info = vmx->idt_vectoring_info;
5880 trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
5883 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
5884 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
5885 * querying dirty_bitmap, we only need to kick all vcpus out of guest
5886 * mode as if vcpus is in root mode, the PML buffer must has been
5890 vmx_flush_pml_buffer(vcpu);
5892 /* If guest state is invalid, start emulating */
5893 if (vmx->emulation_required)
5894 return handle_invalid_guest_state(vcpu);
5896 if (is_guest_mode(vcpu) && nested_vmx_exit_reflected(vcpu, exit_reason))
5897 return nested_vmx_reflect_vmexit(vcpu, exit_reason);
5899 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
5901 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5902 vcpu->run->fail_entry.hardware_entry_failure_reason
5907 if (unlikely(vmx->fail)) {
5909 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5910 vcpu->run->fail_entry.hardware_entry_failure_reason
5911 = vmcs_read32(VM_INSTRUCTION_ERROR);
5917 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
5918 * delivery event since it indicates guest is accessing MMIO.
5919 * The vm-exit can be triggered again after return to guest that
5920 * will cause infinite loop.
5922 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
5923 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
5924 exit_reason != EXIT_REASON_EPT_VIOLATION &&
5925 exit_reason != EXIT_REASON_PML_FULL &&
5926 exit_reason != EXIT_REASON_TASK_SWITCH)) {
5927 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5928 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
5929 vcpu->run->internal.ndata = 3;
5930 vcpu->run->internal.data[0] = vectoring_info;
5931 vcpu->run->internal.data[1] = exit_reason;
5932 vcpu->run->internal.data[2] = vcpu->arch.exit_qualification;
5933 if (exit_reason == EXIT_REASON_EPT_MISCONFIG) {
5934 vcpu->run->internal.ndata++;
5935 vcpu->run->internal.data[3] =
5936 vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5941 if (unlikely(!enable_vnmi &&
5942 vmx->loaded_vmcs->soft_vnmi_blocked)) {
5943 if (vmx_interrupt_allowed(vcpu)) {
5944 vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5945 } else if (vmx->loaded_vmcs->vnmi_blocked_time > 1000000000LL &&
5946 vcpu->arch.nmi_pending) {
5948 * This CPU don't support us in finding the end of an
5949 * NMI-blocked window if the guest runs with IRQs
5950 * disabled. So we pull the trigger after 1 s of
5951 * futile waiting, but inform the user about this.
5953 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
5954 "state on VCPU %d after 1 s timeout\n",
5955 __func__, vcpu->vcpu_id);
5956 vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5960 if (exit_reason < kvm_vmx_max_exit_handlers
5961 && kvm_vmx_exit_handlers[exit_reason])
5962 return kvm_vmx_exit_handlers[exit_reason](vcpu);
5964 vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n",
5967 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5968 vcpu->run->internal.suberror =
5969 KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON;
5970 vcpu->run->internal.ndata = 1;
5971 vcpu->run->internal.data[0] = exit_reason;
5977 * Software based L1D cache flush which is used when microcode providing
5978 * the cache control MSR is not loaded.
5980 * The L1D cache is 32 KiB on Nehalem and later microarchitectures, but to
5981 * flush it is required to read in 64 KiB because the replacement algorithm
5982 * is not exactly LRU. This could be sized at runtime via topology
5983 * information but as all relevant affected CPUs have 32KiB L1D cache size
5984 * there is no point in doing so.
5986 static void vmx_l1d_flush(struct kvm_vcpu *vcpu)
5988 int size = PAGE_SIZE << L1D_CACHE_ORDER;
5991 * This code is only executed when the the flush mode is 'cond' or
5994 if (static_branch_likely(&vmx_l1d_flush_cond)) {
5998 * Clear the per-vcpu flush bit, it gets set again
5999 * either from vcpu_run() or from one of the unsafe
6002 flush_l1d = vcpu->arch.l1tf_flush_l1d;
6003 vcpu->arch.l1tf_flush_l1d = false;
6006 * Clear the per-cpu flush bit, it gets set again from
6007 * the interrupt handlers.
6009 flush_l1d |= kvm_get_cpu_l1tf_flush_l1d();
6010 kvm_clear_cpu_l1tf_flush_l1d();
6016 vcpu->stat.l1d_flush++;
6018 if (static_cpu_has(X86_FEATURE_FLUSH_L1D)) {
6019 wrmsrl(MSR_IA32_FLUSH_CMD, L1D_FLUSH);
6024 /* First ensure the pages are in the TLB */
6025 "xorl %%eax, %%eax\n"
6026 ".Lpopulate_tlb:\n\t"
6027 "movzbl (%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
6028 "addl $4096, %%eax\n\t"
6029 "cmpl %%eax, %[size]\n\t"
6030 "jne .Lpopulate_tlb\n\t"
6031 "xorl %%eax, %%eax\n\t"
6033 /* Now fill the cache */
6034 "xorl %%eax, %%eax\n"
6036 "movzbl (%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
6037 "addl $64, %%eax\n\t"
6038 "cmpl %%eax, %[size]\n\t"
6039 "jne .Lfill_cache\n\t"
6041 :: [flush_pages] "r" (vmx_l1d_flush_pages),
6043 : "eax", "ebx", "ecx", "edx");
6046 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
6048 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6050 if (is_guest_mode(vcpu) &&
6051 nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
6054 if (irr == -1 || tpr < irr) {
6055 vmcs_write32(TPR_THRESHOLD, 0);
6059 vmcs_write32(TPR_THRESHOLD, irr);
6062 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu)
6064 struct vcpu_vmx *vmx = to_vmx(vcpu);
6065 u32 sec_exec_control;
6067 if (!lapic_in_kernel(vcpu))
6070 if (!flexpriority_enabled &&
6071 !cpu_has_vmx_virtualize_x2apic_mode())
6074 /* Postpone execution until vmcs01 is the current VMCS. */
6075 if (is_guest_mode(vcpu)) {
6076 vmx->nested.change_vmcs01_virtual_apic_mode = true;
6080 sec_exec_control = secondary_exec_controls_get(vmx);
6081 sec_exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6082 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
6084 switch (kvm_get_apic_mode(vcpu)) {
6085 case LAPIC_MODE_INVALID:
6086 WARN_ONCE(true, "Invalid local APIC state");
6087 case LAPIC_MODE_DISABLED:
6089 case LAPIC_MODE_XAPIC:
6090 if (flexpriority_enabled) {
6092 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6093 vmx_flush_tlb(vcpu, true);
6096 case LAPIC_MODE_X2APIC:
6097 if (cpu_has_vmx_virtualize_x2apic_mode())
6099 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
6102 secondary_exec_controls_set(vmx, sec_exec_control);
6104 vmx_update_msr_bitmap(vcpu);
6107 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
6109 if (!is_guest_mode(vcpu)) {
6110 vmcs_write64(APIC_ACCESS_ADDR, hpa);
6111 vmx_flush_tlb(vcpu, true);
6115 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
6123 status = vmcs_read16(GUEST_INTR_STATUS);
6125 if (max_isr != old) {
6127 status |= max_isr << 8;
6128 vmcs_write16(GUEST_INTR_STATUS, status);
6132 static void vmx_set_rvi(int vector)
6140 status = vmcs_read16(GUEST_INTR_STATUS);
6141 old = (u8)status & 0xff;
6142 if ((u8)vector != old) {
6144 status |= (u8)vector;
6145 vmcs_write16(GUEST_INTR_STATUS, status);
6149 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
6152 * When running L2, updating RVI is only relevant when
6153 * vmcs12 virtual-interrupt-delivery enabled.
6154 * However, it can be enabled only when L1 also
6155 * intercepts external-interrupts and in that case
6156 * we should not update vmcs02 RVI but instead intercept
6157 * interrupt. Therefore, do nothing when running L2.
6159 if (!is_guest_mode(vcpu))
6160 vmx_set_rvi(max_irr);
6163 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
6165 struct vcpu_vmx *vmx = to_vmx(vcpu);
6167 bool max_irr_updated;
6169 WARN_ON(!vcpu->arch.apicv_active);
6170 if (pi_test_on(&vmx->pi_desc)) {
6171 pi_clear_on(&vmx->pi_desc);
6173 * IOMMU can write to PID.ON, so the barrier matters even on UP.
6174 * But on x86 this is just a compiler barrier anyway.
6176 smp_mb__after_atomic();
6178 kvm_apic_update_irr(vcpu, vmx->pi_desc.pir, &max_irr);
6181 * If we are running L2 and L1 has a new pending interrupt
6182 * which can be injected, we should re-evaluate
6183 * what should be done with this new L1 interrupt.
6184 * If L1 intercepts external-interrupts, we should
6185 * exit from L2 to L1. Otherwise, interrupt should be
6186 * delivered directly to L2.
6188 if (is_guest_mode(vcpu) && max_irr_updated) {
6189 if (nested_exit_on_intr(vcpu))
6190 kvm_vcpu_exiting_guest_mode(vcpu);
6192 kvm_make_request(KVM_REQ_EVENT, vcpu);
6195 max_irr = kvm_lapic_find_highest_irr(vcpu);
6197 vmx_hwapic_irr_update(vcpu, max_irr);
6201 static bool vmx_dy_apicv_has_pending_interrupt(struct kvm_vcpu *vcpu)
6203 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
6205 return pi_test_on(pi_desc) ||
6206 (pi_test_sn(pi_desc) && !pi_is_pir_empty(pi_desc));
6209 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
6211 if (!kvm_vcpu_apicv_active(vcpu))
6214 vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
6215 vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
6216 vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
6217 vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
6220 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
6222 struct vcpu_vmx *vmx = to_vmx(vcpu);
6224 pi_clear_on(&vmx->pi_desc);
6225 memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
6228 static void handle_exception_nmi_irqoff(struct vcpu_vmx *vmx)
6230 vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6232 /* if exit due to PF check for async PF */
6233 if (is_page_fault(vmx->exit_intr_info))
6234 vmx->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason();
6236 /* Handle machine checks before interrupts are enabled */
6237 if (is_machine_check(vmx->exit_intr_info))
6238 kvm_machine_check();
6240 /* We need to handle NMIs before interrupts are enabled */
6241 if (is_nmi(vmx->exit_intr_info)) {
6242 kvm_before_interrupt(&vmx->vcpu);
6244 kvm_after_interrupt(&vmx->vcpu);
6248 static void handle_external_interrupt_irqoff(struct kvm_vcpu *vcpu)
6250 unsigned int vector;
6251 unsigned long entry;
6252 #ifdef CONFIG_X86_64
6258 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6259 if (WARN_ONCE(!is_external_intr(intr_info),
6260 "KVM: unexpected VM-Exit interrupt info: 0x%x", intr_info))
6263 vector = intr_info & INTR_INFO_VECTOR_MASK;
6264 desc = (gate_desc *)host_idt_base + vector;
6265 entry = gate_offset(desc);
6267 kvm_before_interrupt(vcpu);
6270 #ifdef CONFIG_X86_64
6271 "mov %%" _ASM_SP ", %[sp]\n\t"
6272 "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
6277 __ASM_SIZE(push) " $%c[cs]\n\t"
6280 #ifdef CONFIG_X86_64
6285 THUNK_TARGET(entry),
6286 [ss]"i"(__KERNEL_DS),
6287 [cs]"i"(__KERNEL_CS)
6290 kvm_after_interrupt(vcpu);
6292 STACK_FRAME_NON_STANDARD(handle_external_interrupt_irqoff);
6294 static void vmx_handle_exit_irqoff(struct kvm_vcpu *vcpu)
6296 struct vcpu_vmx *vmx = to_vmx(vcpu);
6298 if (vmx->exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
6299 handle_external_interrupt_irqoff(vcpu);
6300 else if (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI)
6301 handle_exception_nmi_irqoff(vmx);
6304 static bool vmx_has_emulated_msr(int index)
6307 case MSR_IA32_SMBASE:
6309 * We cannot do SMM unless we can run the guest in big
6312 return enable_unrestricted_guest || emulate_invalid_guest_state;
6313 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
6315 case MSR_AMD64_VIRT_SPEC_CTRL:
6316 /* This is AMD only. */
6323 static bool vmx_pt_supported(void)
6325 return pt_mode == PT_MODE_HOST_GUEST;
6328 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
6333 bool idtv_info_valid;
6335 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6338 if (vmx->loaded_vmcs->nmi_known_unmasked)
6341 * Can't use vmx->exit_intr_info since we're not sure what
6342 * the exit reason is.
6344 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6345 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
6346 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
6348 * SDM 3: 27.7.1.2 (September 2008)
6349 * Re-set bit "block by NMI" before VM entry if vmexit caused by
6350 * a guest IRET fault.
6351 * SDM 3: 23.2.2 (September 2008)
6352 * Bit 12 is undefined in any of the following cases:
6353 * If the VM exit sets the valid bit in the IDT-vectoring
6354 * information field.
6355 * If the VM exit is due to a double fault.
6357 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
6358 vector != DF_VECTOR && !idtv_info_valid)
6359 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
6360 GUEST_INTR_STATE_NMI);
6362 vmx->loaded_vmcs->nmi_known_unmasked =
6363 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
6364 & GUEST_INTR_STATE_NMI);
6365 } else if (unlikely(vmx->loaded_vmcs->soft_vnmi_blocked))
6366 vmx->loaded_vmcs->vnmi_blocked_time +=
6367 ktime_to_ns(ktime_sub(ktime_get(),
6368 vmx->loaded_vmcs->entry_time));
6371 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
6372 u32 idt_vectoring_info,
6373 int instr_len_field,
6374 int error_code_field)
6378 bool idtv_info_valid;
6380 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6382 vcpu->arch.nmi_injected = false;
6383 kvm_clear_exception_queue(vcpu);
6384 kvm_clear_interrupt_queue(vcpu);
6386 if (!idtv_info_valid)
6389 kvm_make_request(KVM_REQ_EVENT, vcpu);
6391 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
6392 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
6395 case INTR_TYPE_NMI_INTR:
6396 vcpu->arch.nmi_injected = true;
6398 * SDM 3: 27.7.1.2 (September 2008)
6399 * Clear bit "block by NMI" before VM entry if a NMI
6402 vmx_set_nmi_mask(vcpu, false);
6404 case INTR_TYPE_SOFT_EXCEPTION:
6405 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6407 case INTR_TYPE_HARD_EXCEPTION:
6408 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
6409 u32 err = vmcs_read32(error_code_field);
6410 kvm_requeue_exception_e(vcpu, vector, err);
6412 kvm_requeue_exception(vcpu, vector);
6414 case INTR_TYPE_SOFT_INTR:
6415 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6417 case INTR_TYPE_EXT_INTR:
6418 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
6425 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
6427 __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
6428 VM_EXIT_INSTRUCTION_LEN,
6429 IDT_VECTORING_ERROR_CODE);
6432 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
6434 __vmx_complete_interrupts(vcpu,
6435 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
6436 VM_ENTRY_INSTRUCTION_LEN,
6437 VM_ENTRY_EXCEPTION_ERROR_CODE);
6439 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
6442 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
6445 struct perf_guest_switch_msr *msrs;
6447 msrs = perf_guest_get_msrs(&nr_msrs);
6452 for (i = 0; i < nr_msrs; i++)
6453 if (msrs[i].host == msrs[i].guest)
6454 clear_atomic_switch_msr(vmx, msrs[i].msr);
6456 add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
6457 msrs[i].host, false);
6460 static void atomic_switch_umwait_control_msr(struct vcpu_vmx *vmx)
6462 u32 host_umwait_control;
6464 if (!vmx_has_waitpkg(vmx))
6467 host_umwait_control = get_umwait_control_msr();
6469 if (vmx->msr_ia32_umwait_control != host_umwait_control)
6470 add_atomic_switch_msr(vmx, MSR_IA32_UMWAIT_CONTROL,
6471 vmx->msr_ia32_umwait_control,
6472 host_umwait_control, false);
6474 clear_atomic_switch_msr(vmx, MSR_IA32_UMWAIT_CONTROL);
6477 static void vmx_update_hv_timer(struct kvm_vcpu *vcpu)
6479 struct vcpu_vmx *vmx = to_vmx(vcpu);
6483 if (vmx->req_immediate_exit) {
6484 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, 0);
6485 vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6486 } else if (vmx->hv_deadline_tsc != -1) {
6488 if (vmx->hv_deadline_tsc > tscl)
6489 /* set_hv_timer ensures the delta fits in 32-bits */
6490 delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
6491 cpu_preemption_timer_multi);
6495 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
6496 vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6497 } else if (!vmx->loaded_vmcs->hv_timer_soft_disabled) {
6498 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, -1);
6499 vmx->loaded_vmcs->hv_timer_soft_disabled = true;
6503 void vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp)
6505 if (unlikely(host_rsp != vmx->loaded_vmcs->host_state.rsp)) {
6506 vmx->loaded_vmcs->host_state.rsp = host_rsp;
6507 vmcs_writel(HOST_RSP, host_rsp);
6511 bool __vmx_vcpu_run(struct vcpu_vmx *vmx, unsigned long *regs, bool launched);
6513 static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
6515 struct vcpu_vmx *vmx = to_vmx(vcpu);
6516 unsigned long cr3, cr4;
6518 /* Record the guest's net vcpu time for enforced NMI injections. */
6519 if (unlikely(!enable_vnmi &&
6520 vmx->loaded_vmcs->soft_vnmi_blocked))
6521 vmx->loaded_vmcs->entry_time = ktime_get();
6523 /* Don't enter VMX if guest state is invalid, let the exit handler
6524 start emulation until we arrive back to a valid state */
6525 if (vmx->emulation_required)
6528 if (vmx->ple_window_dirty) {
6529 vmx->ple_window_dirty = false;
6530 vmcs_write32(PLE_WINDOW, vmx->ple_window);
6533 if (vmx->nested.need_vmcs12_to_shadow_sync)
6534 nested_sync_vmcs12_to_shadow(vcpu);
6536 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
6537 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
6538 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
6539 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
6541 cr3 = __get_current_cr3_fast();
6542 if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
6543 vmcs_writel(HOST_CR3, cr3);
6544 vmx->loaded_vmcs->host_state.cr3 = cr3;
6547 cr4 = cr4_read_shadow();
6548 if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
6549 vmcs_writel(HOST_CR4, cr4);
6550 vmx->loaded_vmcs->host_state.cr4 = cr4;
6553 /* When single-stepping over STI and MOV SS, we must clear the
6554 * corresponding interruptibility bits in the guest state. Otherwise
6555 * vmentry fails as it then expects bit 14 (BS) in pending debug
6556 * exceptions being set, but that's not correct for the guest debugging
6558 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6559 vmx_set_interrupt_shadow(vcpu, 0);
6561 kvm_load_guest_xcr0(vcpu);
6563 if (static_cpu_has(X86_FEATURE_PKU) &&
6564 kvm_read_cr4_bits(vcpu, X86_CR4_PKE) &&
6565 vcpu->arch.pkru != vmx->host_pkru)
6566 __write_pkru(vcpu->arch.pkru);
6568 pt_guest_enter(vmx);
6570 atomic_switch_perf_msrs(vmx);
6571 atomic_switch_umwait_control_msr(vmx);
6573 if (enable_preemption_timer)
6574 vmx_update_hv_timer(vcpu);
6576 if (lapic_in_kernel(vcpu) &&
6577 vcpu->arch.apic->lapic_timer.timer_advance_ns)
6578 kvm_wait_lapic_expire(vcpu);
6581 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
6582 * it's non-zero. Since vmentry is serialising on affected CPUs, there
6583 * is no need to worry about the conditional branch over the wrmsr
6584 * being speculatively taken.
6586 x86_spec_ctrl_set_guest(vmx->spec_ctrl, 0);
6588 /* L1D Flush includes CPU buffer clear to mitigate MDS */
6589 if (static_branch_unlikely(&vmx_l1d_should_flush))
6590 vmx_l1d_flush(vcpu);
6591 else if (static_branch_unlikely(&mds_user_clear))
6592 mds_clear_cpu_buffers();
6594 if (vcpu->arch.cr2 != read_cr2())
6595 write_cr2(vcpu->arch.cr2);
6597 vmx->fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
6598 vmx->loaded_vmcs->launched);
6600 vcpu->arch.cr2 = read_cr2();
6603 * We do not use IBRS in the kernel. If this vCPU has used the
6604 * SPEC_CTRL MSR it may have left it on; save the value and
6605 * turn it off. This is much more efficient than blindly adding
6606 * it to the atomic save/restore list. Especially as the former
6607 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
6609 * For non-nested case:
6610 * If the L01 MSR bitmap does not intercept the MSR, then we need to
6614 * If the L02 MSR bitmap does not intercept the MSR, then we need to
6617 if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
6618 vmx->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
6620 x86_spec_ctrl_restore_host(vmx->spec_ctrl, 0);
6622 /* All fields are clean at this point */
6623 if (static_branch_unlikely(&enable_evmcs))
6624 current_evmcs->hv_clean_fields |=
6625 HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
6627 if (static_branch_unlikely(&enable_evmcs))
6628 current_evmcs->hv_vp_id = vcpu->arch.hyperv.vp_index;
6630 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
6631 if (vmx->host_debugctlmsr)
6632 update_debugctlmsr(vmx->host_debugctlmsr);
6634 #ifndef CONFIG_X86_64
6636 * The sysexit path does not restore ds/es, so we must set them to
6637 * a reasonable value ourselves.
6639 * We can't defer this to vmx_prepare_switch_to_host() since that
6640 * function may be executed in interrupt context, which saves and
6641 * restore segments around it, nullifying its effect.
6643 loadsegment(ds, __USER_DS);
6644 loadsegment(es, __USER_DS);
6647 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
6648 | (1 << VCPU_EXREG_RFLAGS)
6649 | (1 << VCPU_EXREG_PDPTR)
6650 | (1 << VCPU_EXREG_SEGMENTS)
6651 | (1 << VCPU_EXREG_CR3));
6652 vcpu->arch.regs_dirty = 0;
6657 * eager fpu is enabled if PKEY is supported and CR4 is switched
6658 * back on host, so it is safe to read guest PKRU from current
6661 if (static_cpu_has(X86_FEATURE_PKU) &&
6662 kvm_read_cr4_bits(vcpu, X86_CR4_PKE)) {
6663 vcpu->arch.pkru = rdpkru();
6664 if (vcpu->arch.pkru != vmx->host_pkru)
6665 __write_pkru(vmx->host_pkru);
6668 kvm_put_guest_xcr0(vcpu);
6670 vmx->nested.nested_run_pending = 0;
6671 vmx->idt_vectoring_info = 0;
6673 vmx->exit_reason = vmx->fail ? 0xdead : vmcs_read32(VM_EXIT_REASON);
6674 if ((u16)vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY)
6675 kvm_machine_check();
6677 if (vmx->fail || (vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
6680 vmx->loaded_vmcs->launched = 1;
6681 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
6683 vmx_recover_nmi_blocking(vmx);
6684 vmx_complete_interrupts(vmx);
6687 static struct kvm *vmx_vm_alloc(void)
6689 struct kvm_vmx *kvm_vmx = __vmalloc(sizeof(struct kvm_vmx),
6690 GFP_KERNEL_ACCOUNT | __GFP_ZERO,
6692 return &kvm_vmx->kvm;
6695 static void vmx_vm_free(struct kvm *kvm)
6697 kfree(kvm->arch.hyperv.hv_pa_pg);
6698 vfree(to_kvm_vmx(kvm));
6701 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
6703 struct vcpu_vmx *vmx = to_vmx(vcpu);
6706 vmx_destroy_pml_buffer(vmx);
6707 free_vpid(vmx->vpid);
6708 nested_vmx_free_vcpu(vcpu);
6709 free_loaded_vmcs(vmx->loaded_vmcs);
6710 kfree(vmx->guest_msrs);
6711 kvm_vcpu_uninit(vcpu);
6712 kmem_cache_free(x86_fpu_cache, vmx->vcpu.arch.user_fpu);
6713 kmem_cache_free(x86_fpu_cache, vmx->vcpu.arch.guest_fpu);
6714 kmem_cache_free(kvm_vcpu_cache, vmx);
6717 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
6720 struct vcpu_vmx *vmx;
6721 unsigned long *msr_bitmap;
6724 BUILD_BUG_ON_MSG(offsetof(struct vcpu_vmx, vcpu) != 0,
6725 "struct kvm_vcpu must be at offset 0 for arch usercopy region");
6727 vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT);
6729 return ERR_PTR(-ENOMEM);
6731 vmx->vcpu.arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
6732 GFP_KERNEL_ACCOUNT);
6733 if (!vmx->vcpu.arch.user_fpu) {
6734 printk(KERN_ERR "kvm: failed to allocate kvm userspace's fpu\n");
6736 goto free_partial_vcpu;
6739 vmx->vcpu.arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
6740 GFP_KERNEL_ACCOUNT);
6741 if (!vmx->vcpu.arch.guest_fpu) {
6742 printk(KERN_ERR "kvm: failed to allocate vcpu's fpu\n");
6747 vmx->vpid = allocate_vpid();
6749 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
6756 * If PML is turned on, failure on enabling PML just results in failure
6757 * of creating the vcpu, therefore we can simplify PML logic (by
6758 * avoiding dealing with cases, such as enabling PML partially on vcpus
6759 * for the guest, etc.
6762 vmx->pml_pg = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
6767 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL_ACCOUNT);
6768 BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) * sizeof(vmx->guest_msrs[0])
6771 if (!vmx->guest_msrs)
6774 err = alloc_loaded_vmcs(&vmx->vmcs01);
6778 msr_bitmap = vmx->vmcs01.msr_bitmap;
6779 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_TSC, MSR_TYPE_R);
6780 vmx_disable_intercept_for_msr(msr_bitmap, MSR_FS_BASE, MSR_TYPE_RW);
6781 vmx_disable_intercept_for_msr(msr_bitmap, MSR_GS_BASE, MSR_TYPE_RW);
6782 vmx_disable_intercept_for_msr(msr_bitmap, MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
6783 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_CS, MSR_TYPE_RW);
6784 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_ESP, MSR_TYPE_RW);
6785 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_EIP, MSR_TYPE_RW);
6786 if (kvm_cstate_in_guest(kvm)) {
6787 vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C1_RES, MSR_TYPE_R);
6788 vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C3_RESIDENCY, MSR_TYPE_R);
6789 vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C6_RESIDENCY, MSR_TYPE_R);
6790 vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C7_RESIDENCY, MSR_TYPE_R);
6792 vmx->msr_bitmap_mode = 0;
6794 vmx->loaded_vmcs = &vmx->vmcs01;
6796 vmx_vcpu_load(&vmx->vcpu, cpu);
6797 vmx->vcpu.cpu = cpu;
6798 vmx_vcpu_setup(vmx);
6799 vmx_vcpu_put(&vmx->vcpu);
6801 if (cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
6802 err = alloc_apic_access_page(kvm);
6807 if (enable_ept && !enable_unrestricted_guest) {
6808 err = init_rmode_identity_map(kvm);
6814 nested_vmx_setup_ctls_msrs(&vmx->nested.msrs,
6815 vmx_capability.ept);
6817 memset(&vmx->nested.msrs, 0, sizeof(vmx->nested.msrs));
6819 vmx->nested.posted_intr_nv = -1;
6820 vmx->nested.current_vmptr = -1ull;
6822 vmx->msr_ia32_feature_control_valid_bits = FEATURE_CONTROL_LOCKED;
6825 * Enforce invariant: pi_desc.nv is always either POSTED_INTR_VECTOR
6826 * or POSTED_INTR_WAKEUP_VECTOR.
6828 vmx->pi_desc.nv = POSTED_INTR_VECTOR;
6829 vmx->pi_desc.sn = 1;
6831 vmx->ept_pointer = INVALID_PAGE;
6836 free_loaded_vmcs(vmx->loaded_vmcs);
6838 kfree(vmx->guest_msrs);
6840 vmx_destroy_pml_buffer(vmx);
6842 kvm_vcpu_uninit(&vmx->vcpu);
6844 free_vpid(vmx->vpid);
6845 kmem_cache_free(x86_fpu_cache, vmx->vcpu.arch.guest_fpu);
6847 kmem_cache_free(x86_fpu_cache, vmx->vcpu.arch.user_fpu);
6849 kmem_cache_free(kvm_vcpu_cache, vmx);
6850 return ERR_PTR(err);
6853 #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"
6854 #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"
6856 static int vmx_vm_init(struct kvm *kvm)
6858 spin_lock_init(&to_kvm_vmx(kvm)->ept_pointer_lock);
6861 kvm->arch.pause_in_guest = true;
6863 if (boot_cpu_has(X86_BUG_L1TF) && enable_ept) {
6864 switch (l1tf_mitigation) {
6865 case L1TF_MITIGATION_OFF:
6866 case L1TF_MITIGATION_FLUSH_NOWARN:
6867 /* 'I explicitly don't care' is set */
6869 case L1TF_MITIGATION_FLUSH:
6870 case L1TF_MITIGATION_FLUSH_NOSMT:
6871 case L1TF_MITIGATION_FULL:
6873 * Warn upon starting the first VM in a potentially
6874 * insecure environment.
6876 if (sched_smt_active())
6877 pr_warn_once(L1TF_MSG_SMT);
6878 if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER)
6879 pr_warn_once(L1TF_MSG_L1D);
6881 case L1TF_MITIGATION_FULL_FORCE:
6882 /* Flush is enforced */
6889 static int __init vmx_check_processor_compat(void)
6891 struct vmcs_config vmcs_conf;
6892 struct vmx_capability vmx_cap;
6894 if (setup_vmcs_config(&vmcs_conf, &vmx_cap) < 0)
6897 nested_vmx_setup_ctls_msrs(&vmcs_conf.nested, vmx_cap.ept);
6898 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
6899 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
6900 smp_processor_id());
6906 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
6911 /* For VT-d and EPT combination
6912 * 1. MMIO: always map as UC
6914 * a. VT-d without snooping control feature: can't guarantee the
6915 * result, try to trust guest.
6916 * b. VT-d with snooping control feature: snooping control feature of
6917 * VT-d engine can guarantee the cache correctness. Just set it
6918 * to WB to keep consistent with host. So the same as item 3.
6919 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
6920 * consistent with host MTRR
6923 cache = MTRR_TYPE_UNCACHABLE;
6927 if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
6928 ipat = VMX_EPT_IPAT_BIT;
6929 cache = MTRR_TYPE_WRBACK;
6933 if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
6934 ipat = VMX_EPT_IPAT_BIT;
6935 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
6936 cache = MTRR_TYPE_WRBACK;
6938 cache = MTRR_TYPE_UNCACHABLE;
6942 cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
6945 return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
6948 static int vmx_get_lpage_level(void)
6950 if (enable_ept && !cpu_has_vmx_ept_1g_page())
6951 return PT_DIRECTORY_LEVEL;
6953 /* For shadow and EPT supported 1GB page */
6954 return PT_PDPE_LEVEL;
6957 static void vmcs_set_secondary_exec_control(struct vcpu_vmx *vmx)
6960 * These bits in the secondary execution controls field
6961 * are dynamic, the others are mostly based on the hypervisor
6962 * architecture and the guest's CPUID. Do not touch the
6966 SECONDARY_EXEC_SHADOW_VMCS |
6967 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
6968 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6969 SECONDARY_EXEC_DESC;
6971 u32 new_ctl = vmx->secondary_exec_control;
6972 u32 cur_ctl = secondary_exec_controls_get(vmx);
6974 secondary_exec_controls_set(vmx, (new_ctl & ~mask) | (cur_ctl & mask));
6978 * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
6979 * (indicating "allowed-1") if they are supported in the guest's CPUID.
6981 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
6983 struct vcpu_vmx *vmx = to_vmx(vcpu);
6984 struct kvm_cpuid_entry2 *entry;
6986 vmx->nested.msrs.cr0_fixed1 = 0xffffffff;
6987 vmx->nested.msrs.cr4_fixed1 = X86_CR4_PCE;
6989 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do { \
6990 if (entry && (entry->_reg & (_cpuid_mask))) \
6991 vmx->nested.msrs.cr4_fixed1 |= (_cr4_mask); \
6994 entry = kvm_find_cpuid_entry(vcpu, 0x1, 0);
6995 cr4_fixed1_update(X86_CR4_VME, edx, bit(X86_FEATURE_VME));
6996 cr4_fixed1_update(X86_CR4_PVI, edx, bit(X86_FEATURE_VME));
6997 cr4_fixed1_update(X86_CR4_TSD, edx, bit(X86_FEATURE_TSC));
6998 cr4_fixed1_update(X86_CR4_DE, edx, bit(X86_FEATURE_DE));
6999 cr4_fixed1_update(X86_CR4_PSE, edx, bit(X86_FEATURE_PSE));
7000 cr4_fixed1_update(X86_CR4_PAE, edx, bit(X86_FEATURE_PAE));
7001 cr4_fixed1_update(X86_CR4_MCE, edx, bit(X86_FEATURE_MCE));
7002 cr4_fixed1_update(X86_CR4_PGE, edx, bit(X86_FEATURE_PGE));
7003 cr4_fixed1_update(X86_CR4_OSFXSR, edx, bit(X86_FEATURE_FXSR));
7004 cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, bit(X86_FEATURE_XMM));
7005 cr4_fixed1_update(X86_CR4_VMXE, ecx, bit(X86_FEATURE_VMX));
7006 cr4_fixed1_update(X86_CR4_SMXE, ecx, bit(X86_FEATURE_SMX));
7007 cr4_fixed1_update(X86_CR4_PCIDE, ecx, bit(X86_FEATURE_PCID));
7008 cr4_fixed1_update(X86_CR4_OSXSAVE, ecx, bit(X86_FEATURE_XSAVE));
7010 entry = kvm_find_cpuid_entry(vcpu, 0x7, 0);
7011 cr4_fixed1_update(X86_CR4_FSGSBASE, ebx, bit(X86_FEATURE_FSGSBASE));
7012 cr4_fixed1_update(X86_CR4_SMEP, ebx, bit(X86_FEATURE_SMEP));
7013 cr4_fixed1_update(X86_CR4_SMAP, ebx, bit(X86_FEATURE_SMAP));
7014 cr4_fixed1_update(X86_CR4_PKE, ecx, bit(X86_FEATURE_PKU));
7015 cr4_fixed1_update(X86_CR4_UMIP, ecx, bit(X86_FEATURE_UMIP));
7017 #undef cr4_fixed1_update
7020 static void nested_vmx_entry_exit_ctls_update(struct kvm_vcpu *vcpu)
7022 struct vcpu_vmx *vmx = to_vmx(vcpu);
7024 if (kvm_mpx_supported()) {
7025 bool mpx_enabled = guest_cpuid_has(vcpu, X86_FEATURE_MPX);
7028 vmx->nested.msrs.entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
7029 vmx->nested.msrs.exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
7031 vmx->nested.msrs.entry_ctls_high &= ~VM_ENTRY_LOAD_BNDCFGS;
7032 vmx->nested.msrs.exit_ctls_high &= ~VM_EXIT_CLEAR_BNDCFGS;
7037 static void update_intel_pt_cfg(struct kvm_vcpu *vcpu)
7039 struct vcpu_vmx *vmx = to_vmx(vcpu);
7040 struct kvm_cpuid_entry2 *best = NULL;
7043 for (i = 0; i < PT_CPUID_LEAVES; i++) {
7044 best = kvm_find_cpuid_entry(vcpu, 0x14, i);
7047 vmx->pt_desc.caps[CPUID_EAX + i*PT_CPUID_REGS_NUM] = best->eax;
7048 vmx->pt_desc.caps[CPUID_EBX + i*PT_CPUID_REGS_NUM] = best->ebx;
7049 vmx->pt_desc.caps[CPUID_ECX + i*PT_CPUID_REGS_NUM] = best->ecx;
7050 vmx->pt_desc.caps[CPUID_EDX + i*PT_CPUID_REGS_NUM] = best->edx;
7053 /* Get the number of configurable Address Ranges for filtering */
7054 vmx->pt_desc.addr_range = intel_pt_validate_cap(vmx->pt_desc.caps,
7055 PT_CAP_num_address_ranges);
7057 /* Initialize and clear the no dependency bits */
7058 vmx->pt_desc.ctl_bitmask = ~(RTIT_CTL_TRACEEN | RTIT_CTL_OS |
7059 RTIT_CTL_USR | RTIT_CTL_TSC_EN | RTIT_CTL_DISRETC);
7062 * If CPUID.(EAX=14H,ECX=0):EBX[0]=1 CR3Filter can be set otherwise
7063 * will inject an #GP
7065 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_cr3_filtering))
7066 vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_CR3EN;
7069 * If CPUID.(EAX=14H,ECX=0):EBX[1]=1 CYCEn, CycThresh and
7070 * PSBFreq can be set
7072 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc))
7073 vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_CYCLEACC |
7074 RTIT_CTL_CYC_THRESH | RTIT_CTL_PSB_FREQ);
7077 * If CPUID.(EAX=14H,ECX=0):EBX[3]=1 MTCEn BranchEn and
7078 * MTCFreq can be set
7080 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc))
7081 vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_MTC_EN |
7082 RTIT_CTL_BRANCH_EN | RTIT_CTL_MTC_RANGE);
7084 /* If CPUID.(EAX=14H,ECX=0):EBX[4]=1 FUPonPTW and PTWEn can be set */
7085 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_ptwrite))
7086 vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_FUP_ON_PTW |
7089 /* If CPUID.(EAX=14H,ECX=0):EBX[5]=1 PwrEvEn can be set */
7090 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_power_event_trace))
7091 vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_PWR_EVT_EN;
7093 /* If CPUID.(EAX=14H,ECX=0):ECX[0]=1 ToPA can be set */
7094 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_topa_output))
7095 vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_TOPA;
7097 /* If CPUID.(EAX=14H,ECX=0):ECX[3]=1 FabircEn can be set */
7098 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_output_subsys))
7099 vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_FABRIC_EN;
7101 /* unmask address range configure area */
7102 for (i = 0; i < vmx->pt_desc.addr_range; i++)
7103 vmx->pt_desc.ctl_bitmask &= ~(0xfULL << (32 + i * 4));
7106 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
7108 struct vcpu_vmx *vmx = to_vmx(vcpu);
7110 if (cpu_has_secondary_exec_ctrls()) {
7111 vmx_compute_secondary_exec_control(vmx);
7112 vmcs_set_secondary_exec_control(vmx);
7115 if (nested_vmx_allowed(vcpu))
7116 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7117 FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
7119 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7120 ~FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
7122 if (nested_vmx_allowed(vcpu)) {
7123 nested_vmx_cr_fixed1_bits_update(vcpu);
7124 nested_vmx_entry_exit_ctls_update(vcpu);
7127 if (boot_cpu_has(X86_FEATURE_INTEL_PT) &&
7128 guest_cpuid_has(vcpu, X86_FEATURE_INTEL_PT))
7129 update_intel_pt_cfg(vcpu);
7132 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
7134 if (func == 1 && nested)
7135 entry->ecx |= bit(X86_FEATURE_VMX);
7138 static void vmx_request_immediate_exit(struct kvm_vcpu *vcpu)
7140 to_vmx(vcpu)->req_immediate_exit = true;
7143 static int vmx_check_intercept_io(struct kvm_vcpu *vcpu,
7144 struct x86_instruction_info *info)
7146 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7147 unsigned short port;
7151 if (info->intercept == x86_intercept_in ||
7152 info->intercept == x86_intercept_ins) {
7153 port = info->src_val;
7154 size = info->dst_bytes;
7156 port = info->dst_val;
7157 size = info->src_bytes;
7161 * If the 'use IO bitmaps' VM-execution control is 0, IO instruction
7162 * VM-exits depend on the 'unconditional IO exiting' VM-execution
7165 * Otherwise, IO instruction VM-exits are controlled by the IO bitmaps.
7167 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
7168 intercept = nested_cpu_has(vmcs12,
7169 CPU_BASED_UNCOND_IO_EXITING);
7171 intercept = nested_vmx_check_io_bitmaps(vcpu, port, size);
7173 /* FIXME: produce nested vmexit and return X86EMUL_INTERCEPTED. */
7174 return intercept ? X86EMUL_UNHANDLEABLE : X86EMUL_CONTINUE;
7177 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
7178 struct x86_instruction_info *info,
7179 enum x86_intercept_stage stage)
7181 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7182 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7184 switch (info->intercept) {
7186 * RDPID causes #UD if disabled through secondary execution controls.
7187 * Because it is marked as EmulateOnUD, we need to intercept it here.
7189 case x86_intercept_rdtscp:
7190 if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDTSCP)) {
7191 ctxt->exception.vector = UD_VECTOR;
7192 ctxt->exception.error_code_valid = false;
7193 return X86EMUL_PROPAGATE_FAULT;
7197 case x86_intercept_in:
7198 case x86_intercept_ins:
7199 case x86_intercept_out:
7200 case x86_intercept_outs:
7201 return vmx_check_intercept_io(vcpu, info);
7203 case x86_intercept_lgdt:
7204 case x86_intercept_lidt:
7205 case x86_intercept_lldt:
7206 case x86_intercept_ltr:
7207 case x86_intercept_sgdt:
7208 case x86_intercept_sidt:
7209 case x86_intercept_sldt:
7210 case x86_intercept_str:
7211 if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC))
7212 return X86EMUL_CONTINUE;
7214 /* FIXME: produce nested vmexit and return X86EMUL_INTERCEPTED. */
7217 /* TODO: check more intercepts... */
7222 return X86EMUL_UNHANDLEABLE;
7225 #ifdef CONFIG_X86_64
7226 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
7227 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
7228 u64 divisor, u64 *result)
7230 u64 low = a << shift, high = a >> (64 - shift);
7232 /* To avoid the overflow on divq */
7233 if (high >= divisor)
7236 /* Low hold the result, high hold rem which is discarded */
7237 asm("divq %2\n\t" : "=a" (low), "=d" (high) :
7238 "rm" (divisor), "0" (low), "1" (high));
7244 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc,
7247 struct vcpu_vmx *vmx;
7248 u64 tscl, guest_tscl, delta_tsc, lapic_timer_advance_cycles;
7249 struct kvm_timer *ktimer = &vcpu->arch.apic->lapic_timer;
7251 if (kvm_mwait_in_guest(vcpu->kvm) ||
7252 kvm_can_post_timer_interrupt(vcpu))
7257 guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
7258 delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
7259 lapic_timer_advance_cycles = nsec_to_cycles(vcpu,
7260 ktimer->timer_advance_ns);
7262 if (delta_tsc > lapic_timer_advance_cycles)
7263 delta_tsc -= lapic_timer_advance_cycles;
7267 /* Convert to host delta tsc if tsc scaling is enabled */
7268 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
7269 delta_tsc && u64_shl_div_u64(delta_tsc,
7270 kvm_tsc_scaling_ratio_frac_bits,
7271 vcpu->arch.tsc_scaling_ratio, &delta_tsc))
7275 * If the delta tsc can't fit in the 32 bit after the multi shift,
7276 * we can't use the preemption timer.
7277 * It's possible that it fits on later vmentries, but checking
7278 * on every vmentry is costly so we just use an hrtimer.
7280 if (delta_tsc >> (cpu_preemption_timer_multi + 32))
7283 vmx->hv_deadline_tsc = tscl + delta_tsc;
7284 *expired = !delta_tsc;
7288 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
7290 to_vmx(vcpu)->hv_deadline_tsc = -1;
7294 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
7296 if (!kvm_pause_in_guest(vcpu->kvm))
7297 shrink_ple_window(vcpu);
7300 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
7301 struct kvm_memory_slot *slot)
7303 kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
7304 kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
7307 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
7308 struct kvm_memory_slot *slot)
7310 kvm_mmu_slot_set_dirty(kvm, slot);
7313 static void vmx_flush_log_dirty(struct kvm *kvm)
7315 kvm_flush_pml_buffers(kvm);
7318 static int vmx_write_pml_buffer(struct kvm_vcpu *vcpu)
7320 struct vmcs12 *vmcs12;
7321 struct vcpu_vmx *vmx = to_vmx(vcpu);
7324 if (is_guest_mode(vcpu)) {
7325 WARN_ON_ONCE(vmx->nested.pml_full);
7328 * Check if PML is enabled for the nested guest.
7329 * Whether eptp bit 6 is set is already checked
7330 * as part of A/D emulation.
7332 vmcs12 = get_vmcs12(vcpu);
7333 if (!nested_cpu_has_pml(vmcs12))
7336 if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
7337 vmx->nested.pml_full = true;
7341 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS) & ~0xFFFull;
7342 dst = vmcs12->pml_address + sizeof(u64) * vmcs12->guest_pml_index;
7344 if (kvm_write_guest_page(vcpu->kvm, gpa_to_gfn(dst), &gpa,
7345 offset_in_page(dst), sizeof(gpa)))
7348 vmcs12->guest_pml_index--;
7354 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
7355 struct kvm_memory_slot *memslot,
7356 gfn_t offset, unsigned long mask)
7358 kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
7361 static void __pi_post_block(struct kvm_vcpu *vcpu)
7363 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7364 struct pi_desc old, new;
7368 old.control = new.control = pi_desc->control;
7369 WARN(old.nv != POSTED_INTR_WAKEUP_VECTOR,
7370 "Wakeup handler not enabled while the VCPU is blocked\n");
7372 dest = cpu_physical_id(vcpu->cpu);
7374 if (x2apic_enabled())
7377 new.ndst = (dest << 8) & 0xFF00;
7379 /* set 'NV' to 'notification vector' */
7380 new.nv = POSTED_INTR_VECTOR;
7381 } while (cmpxchg64(&pi_desc->control, old.control,
7382 new.control) != old.control);
7384 if (!WARN_ON_ONCE(vcpu->pre_pcpu == -1)) {
7385 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7386 list_del(&vcpu->blocked_vcpu_list);
7387 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7388 vcpu->pre_pcpu = -1;
7393 * This routine does the following things for vCPU which is going
7394 * to be blocked if VT-d PI is enabled.
7395 * - Store the vCPU to the wakeup list, so when interrupts happen
7396 * we can find the right vCPU to wake up.
7397 * - Change the Posted-interrupt descriptor as below:
7398 * 'NDST' <-- vcpu->pre_pcpu
7399 * 'NV' <-- POSTED_INTR_WAKEUP_VECTOR
7400 * - If 'ON' is set during this process, which means at least one
7401 * interrupt is posted for this vCPU, we cannot block it, in
7402 * this case, return 1, otherwise, return 0.
7405 static int pi_pre_block(struct kvm_vcpu *vcpu)
7408 struct pi_desc old, new;
7409 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7411 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
7412 !irq_remapping_cap(IRQ_POSTING_CAP) ||
7413 !kvm_vcpu_apicv_active(vcpu))
7416 WARN_ON(irqs_disabled());
7417 local_irq_disable();
7418 if (!WARN_ON_ONCE(vcpu->pre_pcpu != -1)) {
7419 vcpu->pre_pcpu = vcpu->cpu;
7420 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7421 list_add_tail(&vcpu->blocked_vcpu_list,
7422 &per_cpu(blocked_vcpu_on_cpu,
7424 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7428 old.control = new.control = pi_desc->control;
7430 WARN((pi_desc->sn == 1),
7431 "Warning: SN field of posted-interrupts "
7432 "is set before blocking\n");
7435 * Since vCPU can be preempted during this process,
7436 * vcpu->cpu could be different with pre_pcpu, we
7437 * need to set pre_pcpu as the destination of wakeup
7438 * notification event, then we can find the right vCPU
7439 * to wakeup in wakeup handler if interrupts happen
7440 * when the vCPU is in blocked state.
7442 dest = cpu_physical_id(vcpu->pre_pcpu);
7444 if (x2apic_enabled())
7447 new.ndst = (dest << 8) & 0xFF00;
7449 /* set 'NV' to 'wakeup vector' */
7450 new.nv = POSTED_INTR_WAKEUP_VECTOR;
7451 } while (cmpxchg64(&pi_desc->control, old.control,
7452 new.control) != old.control);
7454 /* We should not block the vCPU if an interrupt is posted for it. */
7455 if (pi_test_on(pi_desc) == 1)
7456 __pi_post_block(vcpu);
7459 return (vcpu->pre_pcpu == -1);
7462 static int vmx_pre_block(struct kvm_vcpu *vcpu)
7464 if (pi_pre_block(vcpu))
7467 if (kvm_lapic_hv_timer_in_use(vcpu))
7468 kvm_lapic_switch_to_sw_timer(vcpu);
7473 static void pi_post_block(struct kvm_vcpu *vcpu)
7475 if (vcpu->pre_pcpu == -1)
7478 WARN_ON(irqs_disabled());
7479 local_irq_disable();
7480 __pi_post_block(vcpu);
7484 static void vmx_post_block(struct kvm_vcpu *vcpu)
7486 if (kvm_x86_ops->set_hv_timer)
7487 kvm_lapic_switch_to_hv_timer(vcpu);
7489 pi_post_block(vcpu);
7493 * vmx_update_pi_irte - set IRTE for Posted-Interrupts
7496 * @host_irq: host irq of the interrupt
7497 * @guest_irq: gsi of the interrupt
7498 * @set: set or unset PI
7499 * returns 0 on success, < 0 on failure
7501 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
7502 uint32_t guest_irq, bool set)
7504 struct kvm_kernel_irq_routing_entry *e;
7505 struct kvm_irq_routing_table *irq_rt;
7506 struct kvm_lapic_irq irq;
7507 struct kvm_vcpu *vcpu;
7508 struct vcpu_data vcpu_info;
7511 if (!kvm_arch_has_assigned_device(kvm) ||
7512 !irq_remapping_cap(IRQ_POSTING_CAP) ||
7513 !kvm_vcpu_apicv_active(kvm->vcpus[0]))
7516 idx = srcu_read_lock(&kvm->irq_srcu);
7517 irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
7518 if (guest_irq >= irq_rt->nr_rt_entries ||
7519 hlist_empty(&irq_rt->map[guest_irq])) {
7520 pr_warn_once("no route for guest_irq %u/%u (broken user space?)\n",
7521 guest_irq, irq_rt->nr_rt_entries);
7525 hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
7526 if (e->type != KVM_IRQ_ROUTING_MSI)
7529 * VT-d PI cannot support posting multicast/broadcast
7530 * interrupts to a vCPU, we still use interrupt remapping
7531 * for these kind of interrupts.
7533 * For lowest-priority interrupts, we only support
7534 * those with single CPU as the destination, e.g. user
7535 * configures the interrupts via /proc/irq or uses
7536 * irqbalance to make the interrupts single-CPU.
7538 * We will support full lowest-priority interrupt later.
7540 * In addition, we can only inject generic interrupts using
7541 * the PI mechanism, refuse to route others through it.
7544 kvm_set_msi_irq(kvm, e, &irq);
7545 if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu) ||
7546 !kvm_irq_is_postable(&irq)) {
7548 * Make sure the IRTE is in remapped mode if
7549 * we don't handle it in posted mode.
7551 ret = irq_set_vcpu_affinity(host_irq, NULL);
7554 "failed to back to remapped mode, irq: %u\n",
7562 vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
7563 vcpu_info.vector = irq.vector;
7565 trace_kvm_pi_irte_update(host_irq, vcpu->vcpu_id, e->gsi,
7566 vcpu_info.vector, vcpu_info.pi_desc_addr, set);
7569 ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
7571 ret = irq_set_vcpu_affinity(host_irq, NULL);
7574 printk(KERN_INFO "%s: failed to update PI IRTE\n",
7582 srcu_read_unlock(&kvm->irq_srcu, idx);
7586 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
7588 if (vcpu->arch.mcg_cap & MCG_LMCE_P)
7589 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7590 FEATURE_CONTROL_LMCE;
7592 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7593 ~FEATURE_CONTROL_LMCE;
7596 static int vmx_smi_allowed(struct kvm_vcpu *vcpu)
7598 /* we need a nested vmexit to enter SMM, postpone if run is pending */
7599 if (to_vmx(vcpu)->nested.nested_run_pending)
7604 static int vmx_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
7606 struct vcpu_vmx *vmx = to_vmx(vcpu);
7608 vmx->nested.smm.guest_mode = is_guest_mode(vcpu);
7609 if (vmx->nested.smm.guest_mode)
7610 nested_vmx_vmexit(vcpu, -1, 0, 0);
7612 vmx->nested.smm.vmxon = vmx->nested.vmxon;
7613 vmx->nested.vmxon = false;
7614 vmx_clear_hlt(vcpu);
7618 static int vmx_pre_leave_smm(struct kvm_vcpu *vcpu, const char *smstate)
7620 struct vcpu_vmx *vmx = to_vmx(vcpu);
7623 if (vmx->nested.smm.vmxon) {
7624 vmx->nested.vmxon = true;
7625 vmx->nested.smm.vmxon = false;
7628 if (vmx->nested.smm.guest_mode) {
7629 ret = nested_vmx_enter_non_root_mode(vcpu, false);
7633 vmx->nested.smm.guest_mode = false;
7638 static int enable_smi_window(struct kvm_vcpu *vcpu)
7643 static bool vmx_need_emulation_on_page_fault(struct kvm_vcpu *vcpu)
7648 static bool vmx_apic_init_signal_blocked(struct kvm_vcpu *vcpu)
7650 return to_vmx(vcpu)->nested.vmxon;
7653 static __init int hardware_setup(void)
7655 unsigned long host_bndcfgs;
7659 rdmsrl_safe(MSR_EFER, &host_efer);
7662 host_idt_base = dt.address;
7664 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
7665 kvm_define_shared_msr(i, vmx_msr_index[i]);
7667 if (setup_vmcs_config(&vmcs_config, &vmx_capability) < 0)
7670 if (boot_cpu_has(X86_FEATURE_NX))
7671 kvm_enable_efer_bits(EFER_NX);
7673 if (boot_cpu_has(X86_FEATURE_MPX)) {
7674 rdmsrl(MSR_IA32_BNDCFGS, host_bndcfgs);
7675 WARN_ONCE(host_bndcfgs, "KVM: BNDCFGS in host will be lost");
7678 if (boot_cpu_has(X86_FEATURE_XSAVES))
7679 rdmsrl(MSR_IA32_XSS, host_xss);
7681 if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
7682 !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
7685 if (!cpu_has_vmx_ept() ||
7686 !cpu_has_vmx_ept_4levels() ||
7687 !cpu_has_vmx_ept_mt_wb() ||
7688 !cpu_has_vmx_invept_global())
7691 if (!cpu_has_vmx_ept_ad_bits() || !enable_ept)
7692 enable_ept_ad_bits = 0;
7694 if (!cpu_has_vmx_unrestricted_guest() || !enable_ept)
7695 enable_unrestricted_guest = 0;
7697 if (!cpu_has_vmx_flexpriority())
7698 flexpriority_enabled = 0;
7700 if (!cpu_has_virtual_nmis())
7704 * set_apic_access_page_addr() is used to reload apic access
7705 * page upon invalidation. No need to do anything if not
7706 * using the APIC_ACCESS_ADDR VMCS field.
7708 if (!flexpriority_enabled)
7709 kvm_x86_ops->set_apic_access_page_addr = NULL;
7711 if (!cpu_has_vmx_tpr_shadow())
7712 kvm_x86_ops->update_cr8_intercept = NULL;
7714 if (enable_ept && !cpu_has_vmx_ept_2m_page())
7715 kvm_disable_largepages();
7717 #if IS_ENABLED(CONFIG_HYPERV)
7718 if (ms_hyperv.nested_features & HV_X64_NESTED_GUEST_MAPPING_FLUSH
7720 kvm_x86_ops->tlb_remote_flush = hv_remote_flush_tlb;
7721 kvm_x86_ops->tlb_remote_flush_with_range =
7722 hv_remote_flush_tlb_with_range;
7726 if (!cpu_has_vmx_ple()) {
7729 ple_window_grow = 0;
7731 ple_window_shrink = 0;
7734 if (!cpu_has_vmx_apicv()) {
7736 kvm_x86_ops->sync_pir_to_irr = NULL;
7739 if (cpu_has_vmx_tsc_scaling()) {
7740 kvm_has_tsc_control = true;
7741 kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
7742 kvm_tsc_scaling_ratio_frac_bits = 48;
7745 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
7753 * Only enable PML when hardware supports PML feature, and both EPT
7754 * and EPT A/D bit features are enabled -- PML depends on them to work.
7756 if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
7760 kvm_x86_ops->slot_enable_log_dirty = NULL;
7761 kvm_x86_ops->slot_disable_log_dirty = NULL;
7762 kvm_x86_ops->flush_log_dirty = NULL;
7763 kvm_x86_ops->enable_log_dirty_pt_masked = NULL;
7766 if (!cpu_has_vmx_preemption_timer())
7767 enable_preemption_timer = false;
7769 if (enable_preemption_timer) {
7770 u64 use_timer_freq = 5000ULL * 1000 * 1000;
7773 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
7774 cpu_preemption_timer_multi =
7775 vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
7778 use_timer_freq = (u64)tsc_khz * 1000;
7779 use_timer_freq >>= cpu_preemption_timer_multi;
7782 * KVM "disables" the preemption timer by setting it to its max
7783 * value. Don't use the timer if it might cause spurious exits
7784 * at a rate faster than 0.1 Hz (of uninterrupted guest time).
7786 if (use_timer_freq > 0xffffffffu / 10)
7787 enable_preemption_timer = false;
7790 if (!enable_preemption_timer) {
7791 kvm_x86_ops->set_hv_timer = NULL;
7792 kvm_x86_ops->cancel_hv_timer = NULL;
7793 kvm_x86_ops->request_immediate_exit = __kvm_request_immediate_exit;
7796 kvm_set_posted_intr_wakeup_handler(wakeup_handler);
7798 kvm_mce_cap_supported |= MCG_LMCE_P;
7800 if (pt_mode != PT_MODE_SYSTEM && pt_mode != PT_MODE_HOST_GUEST)
7802 if (!enable_ept || !cpu_has_vmx_intel_pt())
7803 pt_mode = PT_MODE_SYSTEM;
7806 nested_vmx_setup_ctls_msrs(&vmcs_config.nested,
7807 vmx_capability.ept);
7809 r = nested_vmx_hardware_setup(kvm_vmx_exit_handlers);
7814 r = alloc_kvm_area();
7816 nested_vmx_hardware_unsetup();
7820 static __exit void hardware_unsetup(void)
7823 nested_vmx_hardware_unsetup();
7828 static struct kvm_x86_ops vmx_x86_ops __ro_after_init = {
7829 .cpu_has_kvm_support = cpu_has_kvm_support,
7830 .disabled_by_bios = vmx_disabled_by_bios,
7831 .hardware_setup = hardware_setup,
7832 .hardware_unsetup = hardware_unsetup,
7833 .check_processor_compatibility = vmx_check_processor_compat,
7834 .hardware_enable = hardware_enable,
7835 .hardware_disable = hardware_disable,
7836 .cpu_has_accelerated_tpr = report_flexpriority,
7837 .has_emulated_msr = vmx_has_emulated_msr,
7839 .vm_init = vmx_vm_init,
7840 .vm_alloc = vmx_vm_alloc,
7841 .vm_free = vmx_vm_free,
7843 .vcpu_create = vmx_create_vcpu,
7844 .vcpu_free = vmx_free_vcpu,
7845 .vcpu_reset = vmx_vcpu_reset,
7847 .prepare_guest_switch = vmx_prepare_switch_to_guest,
7848 .vcpu_load = vmx_vcpu_load,
7849 .vcpu_put = vmx_vcpu_put,
7851 .update_bp_intercept = update_exception_bitmap,
7852 .get_msr_feature = vmx_get_msr_feature,
7853 .get_msr = vmx_get_msr,
7854 .set_msr = vmx_set_msr,
7855 .get_segment_base = vmx_get_segment_base,
7856 .get_segment = vmx_get_segment,
7857 .set_segment = vmx_set_segment,
7858 .get_cpl = vmx_get_cpl,
7859 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
7860 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
7861 .decache_cr3 = vmx_decache_cr3,
7862 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
7863 .set_cr0 = vmx_set_cr0,
7864 .set_cr3 = vmx_set_cr3,
7865 .set_cr4 = vmx_set_cr4,
7866 .set_efer = vmx_set_efer,
7867 .get_idt = vmx_get_idt,
7868 .set_idt = vmx_set_idt,
7869 .get_gdt = vmx_get_gdt,
7870 .set_gdt = vmx_set_gdt,
7871 .get_dr6 = vmx_get_dr6,
7872 .set_dr6 = vmx_set_dr6,
7873 .set_dr7 = vmx_set_dr7,
7874 .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
7875 .cache_reg = vmx_cache_reg,
7876 .get_rflags = vmx_get_rflags,
7877 .set_rflags = vmx_set_rflags,
7879 .tlb_flush = vmx_flush_tlb,
7880 .tlb_flush_gva = vmx_flush_tlb_gva,
7882 .run = vmx_vcpu_run,
7883 .handle_exit = vmx_handle_exit,
7884 .skip_emulated_instruction = skip_emulated_instruction,
7885 .set_interrupt_shadow = vmx_set_interrupt_shadow,
7886 .get_interrupt_shadow = vmx_get_interrupt_shadow,
7887 .patch_hypercall = vmx_patch_hypercall,
7888 .set_irq = vmx_inject_irq,
7889 .set_nmi = vmx_inject_nmi,
7890 .queue_exception = vmx_queue_exception,
7891 .cancel_injection = vmx_cancel_injection,
7892 .interrupt_allowed = vmx_interrupt_allowed,
7893 .nmi_allowed = vmx_nmi_allowed,
7894 .get_nmi_mask = vmx_get_nmi_mask,
7895 .set_nmi_mask = vmx_set_nmi_mask,
7896 .enable_nmi_window = enable_nmi_window,
7897 .enable_irq_window = enable_irq_window,
7898 .update_cr8_intercept = update_cr8_intercept,
7899 .set_virtual_apic_mode = vmx_set_virtual_apic_mode,
7900 .set_apic_access_page_addr = vmx_set_apic_access_page_addr,
7901 .get_enable_apicv = vmx_get_enable_apicv,
7902 .refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
7903 .load_eoi_exitmap = vmx_load_eoi_exitmap,
7904 .apicv_post_state_restore = vmx_apicv_post_state_restore,
7905 .hwapic_irr_update = vmx_hwapic_irr_update,
7906 .hwapic_isr_update = vmx_hwapic_isr_update,
7907 .guest_apic_has_interrupt = vmx_guest_apic_has_interrupt,
7908 .sync_pir_to_irr = vmx_sync_pir_to_irr,
7909 .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
7910 .dy_apicv_has_pending_interrupt = vmx_dy_apicv_has_pending_interrupt,
7912 .set_tss_addr = vmx_set_tss_addr,
7913 .set_identity_map_addr = vmx_set_identity_map_addr,
7914 .get_tdp_level = get_ept_level,
7915 .get_mt_mask = vmx_get_mt_mask,
7917 .get_exit_info = vmx_get_exit_info,
7919 .get_lpage_level = vmx_get_lpage_level,
7921 .cpuid_update = vmx_cpuid_update,
7923 .rdtscp_supported = vmx_rdtscp_supported,
7924 .invpcid_supported = vmx_invpcid_supported,
7926 .set_supported_cpuid = vmx_set_supported_cpuid,
7928 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
7930 .read_l1_tsc_offset = vmx_read_l1_tsc_offset,
7931 .write_l1_tsc_offset = vmx_write_l1_tsc_offset,
7933 .set_tdp_cr3 = vmx_set_cr3,
7935 .check_intercept = vmx_check_intercept,
7936 .handle_exit_irqoff = vmx_handle_exit_irqoff,
7937 .mpx_supported = vmx_mpx_supported,
7938 .xsaves_supported = vmx_xsaves_supported,
7939 .umip_emulated = vmx_umip_emulated,
7940 .pt_supported = vmx_pt_supported,
7941 .pku_supported = vmx_pku_supported,
7943 .request_immediate_exit = vmx_request_immediate_exit,
7945 .sched_in = vmx_sched_in,
7947 .slot_enable_log_dirty = vmx_slot_enable_log_dirty,
7948 .slot_disable_log_dirty = vmx_slot_disable_log_dirty,
7949 .flush_log_dirty = vmx_flush_log_dirty,
7950 .enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
7951 .write_log_dirty = vmx_write_pml_buffer,
7953 .pre_block = vmx_pre_block,
7954 .post_block = vmx_post_block,
7956 .pmu_ops = &intel_pmu_ops,
7958 .update_pi_irte = vmx_update_pi_irte,
7960 #ifdef CONFIG_X86_64
7961 .set_hv_timer = vmx_set_hv_timer,
7962 .cancel_hv_timer = vmx_cancel_hv_timer,
7965 .setup_mce = vmx_setup_mce,
7967 .smi_allowed = vmx_smi_allowed,
7968 .pre_enter_smm = vmx_pre_enter_smm,
7969 .pre_leave_smm = vmx_pre_leave_smm,
7970 .enable_smi_window = enable_smi_window,
7972 .check_nested_events = NULL,
7973 .get_nested_state = NULL,
7974 .set_nested_state = NULL,
7975 .get_vmcs12_pages = NULL,
7976 .nested_enable_evmcs = NULL,
7977 .nested_get_evmcs_version = NULL,
7978 .need_emulation_on_page_fault = vmx_need_emulation_on_page_fault,
7979 .apic_init_signal_blocked = vmx_apic_init_signal_blocked,
7982 static void vmx_cleanup_l1d_flush(void)
7984 if (vmx_l1d_flush_pages) {
7985 free_pages((unsigned long)vmx_l1d_flush_pages, L1D_CACHE_ORDER);
7986 vmx_l1d_flush_pages = NULL;
7988 /* Restore state so sysfs ignores VMX */
7989 l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
7992 static void vmx_exit(void)
7994 #ifdef CONFIG_KEXEC_CORE
7995 RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
8001 #if IS_ENABLED(CONFIG_HYPERV)
8002 if (static_branch_unlikely(&enable_evmcs)) {
8004 struct hv_vp_assist_page *vp_ap;
8006 * Reset everything to support using non-enlightened VMCS
8007 * access later (e.g. when we reload the module with
8008 * enlightened_vmcs=0)
8010 for_each_online_cpu(cpu) {
8011 vp_ap = hv_get_vp_assist_page(cpu);
8016 vp_ap->nested_control.features.directhypercall = 0;
8017 vp_ap->current_nested_vmcs = 0;
8018 vp_ap->enlighten_vmentry = 0;
8021 static_branch_disable(&enable_evmcs);
8024 vmx_cleanup_l1d_flush();
8026 module_exit(vmx_exit);
8028 static int __init vmx_init(void)
8032 #if IS_ENABLED(CONFIG_HYPERV)
8034 * Enlightened VMCS usage should be recommended and the host needs
8035 * to support eVMCS v1 or above. We can also disable eVMCS support
8036 * with module parameter.
8038 if (enlightened_vmcs &&
8039 ms_hyperv.hints & HV_X64_ENLIGHTENED_VMCS_RECOMMENDED &&
8040 (ms_hyperv.nested_features & HV_X64_ENLIGHTENED_VMCS_VERSION) >=
8041 KVM_EVMCS_VERSION) {
8044 /* Check that we have assist pages on all online CPUs */
8045 for_each_online_cpu(cpu) {
8046 if (!hv_get_vp_assist_page(cpu)) {
8047 enlightened_vmcs = false;
8052 if (enlightened_vmcs) {
8053 pr_info("KVM: vmx: using Hyper-V Enlightened VMCS\n");
8054 static_branch_enable(&enable_evmcs);
8057 if (ms_hyperv.nested_features & HV_X64_NESTED_DIRECT_FLUSH)
8058 vmx_x86_ops.enable_direct_tlbflush
8059 = hv_enable_direct_tlbflush;
8062 enlightened_vmcs = false;
8066 r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
8067 __alignof__(struct vcpu_vmx), THIS_MODULE);
8072 * Must be called after kvm_init() so enable_ept is properly set
8073 * up. Hand the parameter mitigation value in which was stored in
8074 * the pre module init parser. If no parameter was given, it will
8075 * contain 'auto' which will be turned into the default 'cond'
8078 r = vmx_setup_l1d_flush(vmentry_l1d_flush_param);
8084 #ifdef CONFIG_KEXEC_CORE
8085 rcu_assign_pointer(crash_vmclear_loaded_vmcss,
8086 crash_vmclear_local_loaded_vmcss);
8088 vmx_check_vmcs12_offsets();
8092 module_init(vmx_init);