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/cpu_device_id.h>
35 #include <asm/debugreg.h>
37 #include <asm/fpu/internal.h>
39 #include <asm/irq_remapping.h>
40 #include <asm/kexec.h>
41 #include <asm/perf_event.h>
43 #include <asm/mmu_context.h>
44 #include <asm/mshyperv.h>
45 #include <asm/mwait.h>
46 #include <asm/spec-ctrl.h>
47 #include <asm/virtext.h>
50 #include "capabilities.h"
54 #include "kvm_cache_regs.h"
66 MODULE_AUTHOR("Qumranet");
67 MODULE_LICENSE("GPL");
70 static const struct x86_cpu_id vmx_cpu_id[] = {
71 X86_MATCH_FEATURE(X86_FEATURE_VMX, NULL),
74 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
77 bool __read_mostly enable_vpid = 1;
78 module_param_named(vpid, enable_vpid, bool, 0444);
80 static bool __read_mostly enable_vnmi = 1;
81 module_param_named(vnmi, enable_vnmi, bool, S_IRUGO);
83 bool __read_mostly flexpriority_enabled = 1;
84 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
86 bool __read_mostly enable_ept = 1;
87 module_param_named(ept, enable_ept, bool, S_IRUGO);
89 bool __read_mostly enable_unrestricted_guest = 1;
90 module_param_named(unrestricted_guest,
91 enable_unrestricted_guest, bool, S_IRUGO);
93 bool __read_mostly enable_ept_ad_bits = 1;
94 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
96 static bool __read_mostly emulate_invalid_guest_state = true;
97 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
99 static bool __read_mostly fasteoi = 1;
100 module_param(fasteoi, bool, S_IRUGO);
102 bool __read_mostly enable_apicv = 1;
103 module_param(enable_apicv, bool, S_IRUGO);
106 * If nested=1, nested virtualization is supported, i.e., guests may use
107 * VMX and be a hypervisor for its own guests. If nested=0, guests may not
108 * use VMX instructions.
110 static bool __read_mostly nested = 1;
111 module_param(nested, bool, S_IRUGO);
113 bool __read_mostly enable_pml = 1;
114 module_param_named(pml, enable_pml, bool, S_IRUGO);
116 static bool __read_mostly dump_invalid_vmcs = 0;
117 module_param(dump_invalid_vmcs, bool, 0644);
119 #define MSR_BITMAP_MODE_X2APIC 1
120 #define MSR_BITMAP_MODE_X2APIC_APICV 2
122 #define KVM_VMX_TSC_MULTIPLIER_MAX 0xffffffffffffffffULL
124 /* Guest_tsc -> host_tsc conversion requires 64-bit division. */
125 static int __read_mostly cpu_preemption_timer_multi;
126 static bool __read_mostly enable_preemption_timer = 1;
128 module_param_named(preemption_timer, enable_preemption_timer, bool, S_IRUGO);
131 #define KVM_VM_CR0_ALWAYS_OFF (X86_CR0_NW | X86_CR0_CD)
132 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR0_NE
133 #define KVM_VM_CR0_ALWAYS_ON \
134 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | \
135 X86_CR0_WP | X86_CR0_PG | X86_CR0_PE)
136 #define KVM_CR4_GUEST_OWNED_BITS \
137 (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \
138 | X86_CR4_OSXMMEXCPT | X86_CR4_LA57 | X86_CR4_TSD)
140 #define KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR4_VMXE
141 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
142 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
144 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
146 #define MSR_IA32_RTIT_STATUS_MASK (~(RTIT_STATUS_FILTEREN | \
147 RTIT_STATUS_CONTEXTEN | RTIT_STATUS_TRIGGEREN | \
148 RTIT_STATUS_ERROR | RTIT_STATUS_STOPPED | \
149 RTIT_STATUS_BYTECNT))
151 #define MSR_IA32_RTIT_OUTPUT_BASE_MASK \
152 (~((1UL << cpuid_query_maxphyaddr(vcpu)) - 1) | 0x7f)
155 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
156 * ple_gap: upper bound on the amount of time between two successive
157 * executions of PAUSE in a loop. Also indicate if ple enabled.
158 * According to test, this time is usually smaller than 128 cycles.
159 * ple_window: upper bound on the amount of time a guest is allowed to execute
160 * in a PAUSE loop. Tests indicate that most spinlocks are held for
161 * less than 2^12 cycles
162 * Time is measured based on a counter that runs at the same rate as the TSC,
163 * refer SDM volume 3b section 21.6.13 & 22.1.3.
165 static unsigned int ple_gap = KVM_DEFAULT_PLE_GAP;
166 module_param(ple_gap, uint, 0444);
168 static unsigned int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
169 module_param(ple_window, uint, 0444);
171 /* Default doubles per-vcpu window every exit. */
172 static unsigned int ple_window_grow = KVM_DEFAULT_PLE_WINDOW_GROW;
173 module_param(ple_window_grow, uint, 0444);
175 /* Default resets per-vcpu window every exit to ple_window. */
176 static unsigned int ple_window_shrink = KVM_DEFAULT_PLE_WINDOW_SHRINK;
177 module_param(ple_window_shrink, uint, 0444);
179 /* Default is to compute the maximum so we can never overflow. */
180 static unsigned int ple_window_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
181 module_param(ple_window_max, uint, 0444);
183 /* Default is SYSTEM mode, 1 for host-guest mode */
184 int __read_mostly pt_mode = PT_MODE_SYSTEM;
185 module_param(pt_mode, int, S_IRUGO);
187 static DEFINE_STATIC_KEY_FALSE(vmx_l1d_should_flush);
188 static DEFINE_STATIC_KEY_FALSE(vmx_l1d_flush_cond);
189 static DEFINE_MUTEX(vmx_l1d_flush_mutex);
191 /* Storage for pre module init parameter parsing */
192 static enum vmx_l1d_flush_state __read_mostly vmentry_l1d_flush_param = VMENTER_L1D_FLUSH_AUTO;
194 static const struct {
197 } vmentry_l1d_param[] = {
198 [VMENTER_L1D_FLUSH_AUTO] = {"auto", true},
199 [VMENTER_L1D_FLUSH_NEVER] = {"never", true},
200 [VMENTER_L1D_FLUSH_COND] = {"cond", true},
201 [VMENTER_L1D_FLUSH_ALWAYS] = {"always", true},
202 [VMENTER_L1D_FLUSH_EPT_DISABLED] = {"EPT disabled", false},
203 [VMENTER_L1D_FLUSH_NOT_REQUIRED] = {"not required", false},
206 #define L1D_CACHE_ORDER 4
207 static void *vmx_l1d_flush_pages;
209 static int vmx_setup_l1d_flush(enum vmx_l1d_flush_state l1tf)
214 if (!boot_cpu_has_bug(X86_BUG_L1TF)) {
215 l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED;
220 l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_EPT_DISABLED;
224 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES)) {
227 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, msr);
228 if (msr & ARCH_CAP_SKIP_VMENTRY_L1DFLUSH) {
229 l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED;
234 /* If set to auto use the default l1tf mitigation method */
235 if (l1tf == VMENTER_L1D_FLUSH_AUTO) {
236 switch (l1tf_mitigation) {
237 case L1TF_MITIGATION_OFF:
238 l1tf = VMENTER_L1D_FLUSH_NEVER;
240 case L1TF_MITIGATION_FLUSH_NOWARN:
241 case L1TF_MITIGATION_FLUSH:
242 case L1TF_MITIGATION_FLUSH_NOSMT:
243 l1tf = VMENTER_L1D_FLUSH_COND;
245 case L1TF_MITIGATION_FULL:
246 case L1TF_MITIGATION_FULL_FORCE:
247 l1tf = VMENTER_L1D_FLUSH_ALWAYS;
250 } else if (l1tf_mitigation == L1TF_MITIGATION_FULL_FORCE) {
251 l1tf = VMENTER_L1D_FLUSH_ALWAYS;
254 if (l1tf != VMENTER_L1D_FLUSH_NEVER && !vmx_l1d_flush_pages &&
255 !boot_cpu_has(X86_FEATURE_FLUSH_L1D)) {
257 * This allocation for vmx_l1d_flush_pages is not tied to a VM
258 * lifetime and so should not be charged to a memcg.
260 page = alloc_pages(GFP_KERNEL, L1D_CACHE_ORDER);
263 vmx_l1d_flush_pages = page_address(page);
266 * Initialize each page with a different pattern in
267 * order to protect against KSM in the nested
268 * virtualization case.
270 for (i = 0; i < 1u << L1D_CACHE_ORDER; ++i) {
271 memset(vmx_l1d_flush_pages + i * PAGE_SIZE, i + 1,
276 l1tf_vmx_mitigation = l1tf;
278 if (l1tf != VMENTER_L1D_FLUSH_NEVER)
279 static_branch_enable(&vmx_l1d_should_flush);
281 static_branch_disable(&vmx_l1d_should_flush);
283 if (l1tf == VMENTER_L1D_FLUSH_COND)
284 static_branch_enable(&vmx_l1d_flush_cond);
286 static_branch_disable(&vmx_l1d_flush_cond);
290 static int vmentry_l1d_flush_parse(const char *s)
295 for (i = 0; i < ARRAY_SIZE(vmentry_l1d_param); i++) {
296 if (vmentry_l1d_param[i].for_parse &&
297 sysfs_streq(s, vmentry_l1d_param[i].option))
304 static int vmentry_l1d_flush_set(const char *s, const struct kernel_param *kp)
308 l1tf = vmentry_l1d_flush_parse(s);
312 if (!boot_cpu_has(X86_BUG_L1TF))
316 * Has vmx_init() run already? If not then this is the pre init
317 * parameter parsing. In that case just store the value and let
318 * vmx_init() do the proper setup after enable_ept has been
321 if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_AUTO) {
322 vmentry_l1d_flush_param = l1tf;
326 mutex_lock(&vmx_l1d_flush_mutex);
327 ret = vmx_setup_l1d_flush(l1tf);
328 mutex_unlock(&vmx_l1d_flush_mutex);
332 static int vmentry_l1d_flush_get(char *s, const struct kernel_param *kp)
334 if (WARN_ON_ONCE(l1tf_vmx_mitigation >= ARRAY_SIZE(vmentry_l1d_param)))
335 return sprintf(s, "???\n");
337 return sprintf(s, "%s\n", vmentry_l1d_param[l1tf_vmx_mitigation].option);
340 static const struct kernel_param_ops vmentry_l1d_flush_ops = {
341 .set = vmentry_l1d_flush_set,
342 .get = vmentry_l1d_flush_get,
344 module_param_cb(vmentry_l1d_flush, &vmentry_l1d_flush_ops, NULL, 0644);
346 static bool guest_state_valid(struct kvm_vcpu *vcpu);
347 static u32 vmx_segment_access_rights(struct kvm_segment *var);
348 static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
351 void vmx_vmexit(void);
353 #define vmx_insn_failed(fmt...) \
356 pr_warn_ratelimited(fmt); \
359 asmlinkage void vmread_error(unsigned long field, bool fault)
362 kvm_spurious_fault();
364 vmx_insn_failed("kvm: vmread failed: field=%lx\n", field);
367 noinline void vmwrite_error(unsigned long field, unsigned long value)
369 vmx_insn_failed("kvm: vmwrite failed: field=%lx val=%lx err=%d\n",
370 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
373 noinline void vmclear_error(struct vmcs *vmcs, u64 phys_addr)
375 vmx_insn_failed("kvm: vmclear failed: %p/%llx\n", vmcs, phys_addr);
378 noinline void vmptrld_error(struct vmcs *vmcs, u64 phys_addr)
380 vmx_insn_failed("kvm: vmptrld failed: %p/%llx\n", vmcs, phys_addr);
383 noinline void invvpid_error(unsigned long ext, u16 vpid, gva_t gva)
385 vmx_insn_failed("kvm: invvpid failed: ext=0x%lx vpid=%u gva=0x%lx\n",
389 noinline void invept_error(unsigned long ext, u64 eptp, gpa_t gpa)
391 vmx_insn_failed("kvm: invept failed: ext=0x%lx eptp=%llx gpa=0x%llx\n",
395 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
396 DEFINE_PER_CPU(struct vmcs *, current_vmcs);
398 * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
399 * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
401 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
404 * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
405 * can find which vCPU should be waken up.
407 static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu);
408 static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock);
410 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
411 static DEFINE_SPINLOCK(vmx_vpid_lock);
413 struct vmcs_config vmcs_config;
414 struct vmx_capability vmx_capability;
416 #define VMX_SEGMENT_FIELD(seg) \
417 [VCPU_SREG_##seg] = { \
418 .selector = GUEST_##seg##_SELECTOR, \
419 .base = GUEST_##seg##_BASE, \
420 .limit = GUEST_##seg##_LIMIT, \
421 .ar_bytes = GUEST_##seg##_AR_BYTES, \
424 static const struct kvm_vmx_segment_field {
429 } kvm_vmx_segment_fields[] = {
430 VMX_SEGMENT_FIELD(CS),
431 VMX_SEGMENT_FIELD(DS),
432 VMX_SEGMENT_FIELD(ES),
433 VMX_SEGMENT_FIELD(FS),
434 VMX_SEGMENT_FIELD(GS),
435 VMX_SEGMENT_FIELD(SS),
436 VMX_SEGMENT_FIELD(TR),
437 VMX_SEGMENT_FIELD(LDTR),
440 static inline void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
442 vmx->segment_cache.bitmask = 0;
445 static unsigned long host_idt_base;
448 * Though SYSCALL is only supported in 64-bit mode on Intel CPUs, kvm
449 * will emulate SYSCALL in legacy mode if the vendor string in guest
450 * CPUID.0:{EBX,ECX,EDX} is "AuthenticAMD" or "AMDisbetter!" To
451 * support this emulation, IA32_STAR must always be included in
452 * vmx_msr_index[], even in i386 builds.
454 const u32 vmx_msr_index[] = {
456 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
458 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
462 #if IS_ENABLED(CONFIG_HYPERV)
463 static bool __read_mostly enlightened_vmcs = true;
464 module_param(enlightened_vmcs, bool, 0444);
466 /* check_ept_pointer() should be under protection of ept_pointer_lock. */
467 static void check_ept_pointer_match(struct kvm *kvm)
469 struct kvm_vcpu *vcpu;
470 u64 tmp_eptp = INVALID_PAGE;
473 kvm_for_each_vcpu(i, vcpu, kvm) {
474 if (!VALID_PAGE(tmp_eptp)) {
475 tmp_eptp = to_vmx(vcpu)->ept_pointer;
476 } else if (tmp_eptp != to_vmx(vcpu)->ept_pointer) {
477 to_kvm_vmx(kvm)->ept_pointers_match
478 = EPT_POINTERS_MISMATCH;
483 to_kvm_vmx(kvm)->ept_pointers_match = EPT_POINTERS_MATCH;
486 static int kvm_fill_hv_flush_list_func(struct hv_guest_mapping_flush_list *flush,
489 struct kvm_tlb_range *range = data;
491 return hyperv_fill_flush_guest_mapping_list(flush, range->start_gfn,
495 static inline int __hv_remote_flush_tlb_with_range(struct kvm *kvm,
496 struct kvm_vcpu *vcpu, struct kvm_tlb_range *range)
498 u64 ept_pointer = to_vmx(vcpu)->ept_pointer;
501 * FLUSH_GUEST_PHYSICAL_ADDRESS_SPACE hypercall needs address
502 * of the base of EPT PML4 table, strip off EPT configuration
506 return hyperv_flush_guest_mapping_range(ept_pointer & PAGE_MASK,
507 kvm_fill_hv_flush_list_func, (void *)range);
509 return hyperv_flush_guest_mapping(ept_pointer & PAGE_MASK);
512 static int hv_remote_flush_tlb_with_range(struct kvm *kvm,
513 struct kvm_tlb_range *range)
515 struct kvm_vcpu *vcpu;
518 spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
520 if (to_kvm_vmx(kvm)->ept_pointers_match == EPT_POINTERS_CHECK)
521 check_ept_pointer_match(kvm);
523 if (to_kvm_vmx(kvm)->ept_pointers_match != EPT_POINTERS_MATCH) {
524 kvm_for_each_vcpu(i, vcpu, kvm) {
525 /* If ept_pointer is invalid pointer, bypass flush request. */
526 if (VALID_PAGE(to_vmx(vcpu)->ept_pointer))
527 ret |= __hv_remote_flush_tlb_with_range(
531 ret = __hv_remote_flush_tlb_with_range(kvm,
532 kvm_get_vcpu(kvm, 0), range);
535 spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
538 static int hv_remote_flush_tlb(struct kvm *kvm)
540 return hv_remote_flush_tlb_with_range(kvm, NULL);
543 static int hv_enable_direct_tlbflush(struct kvm_vcpu *vcpu)
545 struct hv_enlightened_vmcs *evmcs;
546 struct hv_partition_assist_pg **p_hv_pa_pg =
547 &vcpu->kvm->arch.hyperv.hv_pa_pg;
549 * Synthetic VM-Exit is not enabled in current code and so All
550 * evmcs in singe VM shares same assist page.
553 *p_hv_pa_pg = kzalloc(PAGE_SIZE, GFP_KERNEL);
558 evmcs = (struct hv_enlightened_vmcs *)to_vmx(vcpu)->loaded_vmcs->vmcs;
560 evmcs->partition_assist_page =
562 evmcs->hv_vm_id = (unsigned long)vcpu->kvm;
563 evmcs->hv_enlightenments_control.nested_flush_hypercall = 1;
568 #endif /* IS_ENABLED(CONFIG_HYPERV) */
571 * Comment's format: document - errata name - stepping - processor name.
573 * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
575 static u32 vmx_preemption_cpu_tfms[] = {
576 /* 323344.pdf - BA86 - D0 - Xeon 7500 Series */
578 /* 323056.pdf - AAX65 - C2 - Xeon L3406 */
579 /* 322814.pdf - AAT59 - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
580 /* 322911.pdf - AAU65 - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
582 /* 322911.pdf - AAU65 - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
584 /* 322373.pdf - AAO95 - B1 - Xeon 3400 Series */
585 /* 322166.pdf - AAN92 - B1 - i7-800 and i5-700 Desktop */
587 * 320767.pdf - AAP86 - B1 -
588 * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
591 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
593 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
595 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
597 /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
598 /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
599 /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
601 /* Xeon E3-1220 V2 */
605 static inline bool cpu_has_broken_vmx_preemption_timer(void)
607 u32 eax = cpuid_eax(0x00000001), i;
609 /* Clear the reserved bits */
610 eax &= ~(0x3U << 14 | 0xfU << 28);
611 for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
612 if (eax == vmx_preemption_cpu_tfms[i])
618 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
620 return flexpriority_enabled && lapic_in_kernel(vcpu);
623 static inline bool report_flexpriority(void)
625 return flexpriority_enabled;
628 static inline int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
632 for (i = 0; i < vmx->nmsrs; ++i)
633 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
638 struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
642 i = __find_msr_index(vmx, msr);
644 return &vmx->guest_msrs[i];
648 static int vmx_set_guest_msr(struct vcpu_vmx *vmx, struct shared_msr_entry *msr, u64 data)
652 u64 old_msr_data = msr->data;
654 if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
656 ret = kvm_set_shared_msr(msr->index, msr->data,
660 msr->data = old_msr_data;
665 #ifdef CONFIG_KEXEC_CORE
666 static void crash_vmclear_local_loaded_vmcss(void)
668 int cpu = raw_smp_processor_id();
669 struct loaded_vmcs *v;
671 list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
672 loaded_vmcss_on_cpu_link)
675 #endif /* CONFIG_KEXEC_CORE */
677 static void __loaded_vmcs_clear(void *arg)
679 struct loaded_vmcs *loaded_vmcs = arg;
680 int cpu = raw_smp_processor_id();
682 if (loaded_vmcs->cpu != cpu)
683 return; /* vcpu migration can race with cpu offline */
684 if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
685 per_cpu(current_vmcs, cpu) = NULL;
687 vmcs_clear(loaded_vmcs->vmcs);
688 if (loaded_vmcs->shadow_vmcs && loaded_vmcs->launched)
689 vmcs_clear(loaded_vmcs->shadow_vmcs);
691 list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
694 * Ensure all writes to loaded_vmcs, including deleting it from its
695 * current percpu list, complete before setting loaded_vmcs->vcpu to
696 * -1, otherwise a different cpu can see vcpu == -1 first and add
697 * loaded_vmcs to its percpu list before it's deleted from this cpu's
698 * list. Pairs with the smp_rmb() in vmx_vcpu_load_vmcs().
702 loaded_vmcs->cpu = -1;
703 loaded_vmcs->launched = 0;
706 void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
708 int cpu = loaded_vmcs->cpu;
711 smp_call_function_single(cpu,
712 __loaded_vmcs_clear, loaded_vmcs, 1);
715 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
719 u32 mask = 1 << (seg * SEG_FIELD_NR + field);
721 if (!kvm_register_is_available(&vmx->vcpu, VCPU_EXREG_SEGMENTS)) {
722 kvm_register_mark_available(&vmx->vcpu, VCPU_EXREG_SEGMENTS);
723 vmx->segment_cache.bitmask = 0;
725 ret = vmx->segment_cache.bitmask & mask;
726 vmx->segment_cache.bitmask |= mask;
730 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
732 u16 *p = &vmx->segment_cache.seg[seg].selector;
734 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
735 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
739 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
741 ulong *p = &vmx->segment_cache.seg[seg].base;
743 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
744 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
748 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
750 u32 *p = &vmx->segment_cache.seg[seg].limit;
752 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
753 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
757 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
759 u32 *p = &vmx->segment_cache.seg[seg].ar;
761 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
762 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
766 void update_exception_bitmap(struct kvm_vcpu *vcpu)
770 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
771 (1u << DB_VECTOR) | (1u << AC_VECTOR);
773 * Guest access to VMware backdoor ports could legitimately
774 * trigger #GP because of TSS I/O permission bitmap.
775 * We intercept those #GP and allow access to them anyway
778 if (enable_vmware_backdoor)
779 eb |= (1u << GP_VECTOR);
780 if ((vcpu->guest_debug &
781 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
782 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
783 eb |= 1u << BP_VECTOR;
784 if (to_vmx(vcpu)->rmode.vm86_active)
787 eb &= ~(1u << PF_VECTOR);
789 /* When we are running a nested L2 guest and L1 specified for it a
790 * certain exception bitmap, we must trap the same exceptions and pass
791 * them to L1. When running L2, we will only handle the exceptions
792 * specified above if L1 did not want them.
794 if (is_guest_mode(vcpu))
795 eb |= get_vmcs12(vcpu)->exception_bitmap;
797 vmcs_write32(EXCEPTION_BITMAP, eb);
801 * Check if MSR is intercepted for currently loaded MSR bitmap.
803 static bool msr_write_intercepted(struct kvm_vcpu *vcpu, u32 msr)
805 unsigned long *msr_bitmap;
806 int f = sizeof(unsigned long);
808 if (!cpu_has_vmx_msr_bitmap())
811 msr_bitmap = to_vmx(vcpu)->loaded_vmcs->msr_bitmap;
814 return !!test_bit(msr, msr_bitmap + 0x800 / f);
815 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
817 return !!test_bit(msr, msr_bitmap + 0xc00 / f);
823 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
824 unsigned long entry, unsigned long exit)
826 vm_entry_controls_clearbit(vmx, entry);
827 vm_exit_controls_clearbit(vmx, exit);
830 int vmx_find_msr_index(struct vmx_msrs *m, u32 msr)
834 for (i = 0; i < m->nr; ++i) {
835 if (m->val[i].index == msr)
841 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
844 struct msr_autoload *m = &vmx->msr_autoload;
848 if (cpu_has_load_ia32_efer()) {
849 clear_atomic_switch_msr_special(vmx,
850 VM_ENTRY_LOAD_IA32_EFER,
851 VM_EXIT_LOAD_IA32_EFER);
855 case MSR_CORE_PERF_GLOBAL_CTRL:
856 if (cpu_has_load_perf_global_ctrl()) {
857 clear_atomic_switch_msr_special(vmx,
858 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
859 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
864 i = vmx_find_msr_index(&m->guest, msr);
868 m->guest.val[i] = m->guest.val[m->guest.nr];
869 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
872 i = vmx_find_msr_index(&m->host, msr);
877 m->host.val[i] = m->host.val[m->host.nr];
878 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
881 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
882 unsigned long entry, unsigned long exit,
883 unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
884 u64 guest_val, u64 host_val)
886 vmcs_write64(guest_val_vmcs, guest_val);
887 if (host_val_vmcs != HOST_IA32_EFER)
888 vmcs_write64(host_val_vmcs, host_val);
889 vm_entry_controls_setbit(vmx, entry);
890 vm_exit_controls_setbit(vmx, exit);
893 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
894 u64 guest_val, u64 host_val, bool entry_only)
897 struct msr_autoload *m = &vmx->msr_autoload;
901 if (cpu_has_load_ia32_efer()) {
902 add_atomic_switch_msr_special(vmx,
903 VM_ENTRY_LOAD_IA32_EFER,
904 VM_EXIT_LOAD_IA32_EFER,
907 guest_val, host_val);
911 case MSR_CORE_PERF_GLOBAL_CTRL:
912 if (cpu_has_load_perf_global_ctrl()) {
913 add_atomic_switch_msr_special(vmx,
914 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
915 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
916 GUEST_IA32_PERF_GLOBAL_CTRL,
917 HOST_IA32_PERF_GLOBAL_CTRL,
918 guest_val, host_val);
922 case MSR_IA32_PEBS_ENABLE:
923 /* PEBS needs a quiescent period after being disabled (to write
924 * a record). Disabling PEBS through VMX MSR swapping doesn't
925 * provide that period, so a CPU could write host's record into
928 wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
931 i = vmx_find_msr_index(&m->guest, msr);
933 j = vmx_find_msr_index(&m->host, msr);
935 if ((i < 0 && m->guest.nr == NR_LOADSTORE_MSRS) ||
936 (j < 0 && m->host.nr == NR_LOADSTORE_MSRS)) {
937 printk_once(KERN_WARNING "Not enough msr switch entries. "
938 "Can't add msr %x\n", msr);
943 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
945 m->guest.val[i].index = msr;
946 m->guest.val[i].value = guest_val;
953 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
955 m->host.val[j].index = msr;
956 m->host.val[j].value = host_val;
959 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
961 u64 guest_efer = vmx->vcpu.arch.efer;
964 /* Shadow paging assumes NX to be available. */
966 guest_efer |= EFER_NX;
969 * LMA and LME handled by hardware; SCE meaningless outside long mode.
971 ignore_bits |= EFER_SCE;
973 ignore_bits |= EFER_LMA | EFER_LME;
974 /* SCE is meaningful only in long mode on Intel */
975 if (guest_efer & EFER_LMA)
976 ignore_bits &= ~(u64)EFER_SCE;
980 * On EPT, we can't emulate NX, so we must switch EFER atomically.
981 * On CPUs that support "load IA32_EFER", always switch EFER
982 * atomically, since it's faster than switching it manually.
984 if (cpu_has_load_ia32_efer() ||
985 (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
986 if (!(guest_efer & EFER_LMA))
987 guest_efer &= ~EFER_LME;
988 if (guest_efer != host_efer)
989 add_atomic_switch_msr(vmx, MSR_EFER,
990 guest_efer, host_efer, false);
992 clear_atomic_switch_msr(vmx, MSR_EFER);
995 clear_atomic_switch_msr(vmx, MSR_EFER);
997 guest_efer &= ~ignore_bits;
998 guest_efer |= host_efer & ignore_bits;
1000 vmx->guest_msrs[efer_offset].data = guest_efer;
1001 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1007 #ifdef CONFIG_X86_32
1009 * On 32-bit kernels, VM exits still load the FS and GS bases from the
1010 * VMCS rather than the segment table. KVM uses this helper to figure
1011 * out the current bases to poke them into the VMCS before entry.
1013 static unsigned long segment_base(u16 selector)
1015 struct desc_struct *table;
1018 if (!(selector & ~SEGMENT_RPL_MASK))
1021 table = get_current_gdt_ro();
1023 if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
1024 u16 ldt_selector = kvm_read_ldt();
1026 if (!(ldt_selector & ~SEGMENT_RPL_MASK))
1029 table = (struct desc_struct *)segment_base(ldt_selector);
1031 v = get_desc_base(&table[selector >> 3]);
1036 static inline bool pt_can_write_msr(struct vcpu_vmx *vmx)
1038 return vmx_pt_mode_is_host_guest() &&
1039 !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
1042 static inline void pt_load_msr(struct pt_ctx *ctx, u32 addr_range)
1046 wrmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1047 wrmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1048 wrmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1049 wrmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1050 for (i = 0; i < addr_range; i++) {
1051 wrmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1052 wrmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1056 static inline void pt_save_msr(struct pt_ctx *ctx, u32 addr_range)
1060 rdmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1061 rdmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1062 rdmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1063 rdmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1064 for (i = 0; i < addr_range; i++) {
1065 rdmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1066 rdmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1070 static void pt_guest_enter(struct vcpu_vmx *vmx)
1072 if (vmx_pt_mode_is_system())
1076 * GUEST_IA32_RTIT_CTL is already set in the VMCS.
1077 * Save host state before VM entry.
1079 rdmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1080 if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1081 wrmsrl(MSR_IA32_RTIT_CTL, 0);
1082 pt_save_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1083 pt_load_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1087 static void pt_guest_exit(struct vcpu_vmx *vmx)
1089 if (vmx_pt_mode_is_system())
1092 if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1093 pt_save_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1094 pt_load_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1097 /* Reload host state (IA32_RTIT_CTL will be cleared on VM exit). */
1098 wrmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1101 void vmx_set_host_fs_gs(struct vmcs_host_state *host, u16 fs_sel, u16 gs_sel,
1102 unsigned long fs_base, unsigned long gs_base)
1104 if (unlikely(fs_sel != host->fs_sel)) {
1106 vmcs_write16(HOST_FS_SELECTOR, fs_sel);
1108 vmcs_write16(HOST_FS_SELECTOR, 0);
1109 host->fs_sel = fs_sel;
1111 if (unlikely(gs_sel != host->gs_sel)) {
1113 vmcs_write16(HOST_GS_SELECTOR, gs_sel);
1115 vmcs_write16(HOST_GS_SELECTOR, 0);
1116 host->gs_sel = gs_sel;
1118 if (unlikely(fs_base != host->fs_base)) {
1119 vmcs_writel(HOST_FS_BASE, fs_base);
1120 host->fs_base = fs_base;
1122 if (unlikely(gs_base != host->gs_base)) {
1123 vmcs_writel(HOST_GS_BASE, gs_base);
1124 host->gs_base = gs_base;
1128 void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu)
1130 struct vcpu_vmx *vmx = to_vmx(vcpu);
1131 struct vmcs_host_state *host_state;
1132 #ifdef CONFIG_X86_64
1133 int cpu = raw_smp_processor_id();
1135 unsigned long fs_base, gs_base;
1139 vmx->req_immediate_exit = false;
1142 * Note that guest MSRs to be saved/restored can also be changed
1143 * when guest state is loaded. This happens when guest transitions
1144 * to/from long-mode by setting MSR_EFER.LMA.
1146 if (!vmx->guest_msrs_ready) {
1147 vmx->guest_msrs_ready = true;
1148 for (i = 0; i < vmx->save_nmsrs; ++i)
1149 kvm_set_shared_msr(vmx->guest_msrs[i].index,
1150 vmx->guest_msrs[i].data,
1151 vmx->guest_msrs[i].mask);
1155 if (vmx->nested.need_vmcs12_to_shadow_sync)
1156 nested_sync_vmcs12_to_shadow(vcpu);
1158 if (vmx->guest_state_loaded)
1161 host_state = &vmx->loaded_vmcs->host_state;
1164 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
1165 * allow segment selectors with cpl > 0 or ti == 1.
1167 host_state->ldt_sel = kvm_read_ldt();
1169 #ifdef CONFIG_X86_64
1170 savesegment(ds, host_state->ds_sel);
1171 savesegment(es, host_state->es_sel);
1173 gs_base = cpu_kernelmode_gs_base(cpu);
1174 if (likely(is_64bit_mm(current->mm))) {
1175 save_fsgs_for_kvm();
1176 fs_sel = current->thread.fsindex;
1177 gs_sel = current->thread.gsindex;
1178 fs_base = current->thread.fsbase;
1179 vmx->msr_host_kernel_gs_base = current->thread.gsbase;
1181 savesegment(fs, fs_sel);
1182 savesegment(gs, gs_sel);
1183 fs_base = read_msr(MSR_FS_BASE);
1184 vmx->msr_host_kernel_gs_base = read_msr(MSR_KERNEL_GS_BASE);
1187 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1189 savesegment(fs, fs_sel);
1190 savesegment(gs, gs_sel);
1191 fs_base = segment_base(fs_sel);
1192 gs_base = segment_base(gs_sel);
1195 vmx_set_host_fs_gs(host_state, fs_sel, gs_sel, fs_base, gs_base);
1196 vmx->guest_state_loaded = true;
1199 static void vmx_prepare_switch_to_host(struct vcpu_vmx *vmx)
1201 struct vmcs_host_state *host_state;
1203 if (!vmx->guest_state_loaded)
1206 host_state = &vmx->loaded_vmcs->host_state;
1208 ++vmx->vcpu.stat.host_state_reload;
1210 #ifdef CONFIG_X86_64
1211 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1213 if (host_state->ldt_sel || (host_state->gs_sel & 7)) {
1214 kvm_load_ldt(host_state->ldt_sel);
1215 #ifdef CONFIG_X86_64
1216 load_gs_index(host_state->gs_sel);
1218 loadsegment(gs, host_state->gs_sel);
1221 if (host_state->fs_sel & 7)
1222 loadsegment(fs, host_state->fs_sel);
1223 #ifdef CONFIG_X86_64
1224 if (unlikely(host_state->ds_sel | host_state->es_sel)) {
1225 loadsegment(ds, host_state->ds_sel);
1226 loadsegment(es, host_state->es_sel);
1229 invalidate_tss_limit();
1230 #ifdef CONFIG_X86_64
1231 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1233 load_fixmap_gdt(raw_smp_processor_id());
1234 vmx->guest_state_loaded = false;
1235 vmx->guest_msrs_ready = false;
1238 #ifdef CONFIG_X86_64
1239 static u64 vmx_read_guest_kernel_gs_base(struct vcpu_vmx *vmx)
1242 if (vmx->guest_state_loaded)
1243 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1245 return vmx->msr_guest_kernel_gs_base;
1248 static void vmx_write_guest_kernel_gs_base(struct vcpu_vmx *vmx, u64 data)
1251 if (vmx->guest_state_loaded)
1252 wrmsrl(MSR_KERNEL_GS_BASE, data);
1254 vmx->msr_guest_kernel_gs_base = data;
1258 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
1260 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1261 struct pi_desc old, new;
1265 * In case of hot-plug or hot-unplug, we may have to undo
1266 * vmx_vcpu_pi_put even if there is no assigned device. And we
1267 * always keep PI.NDST up to date for simplicity: it makes the
1268 * code easier, and CPU migration is not a fast path.
1270 if (!pi_test_sn(pi_desc) && vcpu->cpu == cpu)
1274 * If the 'nv' field is POSTED_INTR_WAKEUP_VECTOR, do not change
1275 * PI.NDST: pi_post_block is the one expected to change PID.NDST and the
1276 * wakeup handler expects the vCPU to be on the blocked_vcpu_list that
1277 * matches PI.NDST. Otherwise, a vcpu may not be able to be woken up
1280 if (pi_desc->nv == POSTED_INTR_WAKEUP_VECTOR || vcpu->cpu == cpu) {
1281 pi_clear_sn(pi_desc);
1282 goto after_clear_sn;
1285 /* The full case. */
1287 old.control = new.control = pi_desc->control;
1289 dest = cpu_physical_id(cpu);
1291 if (x2apic_enabled())
1294 new.ndst = (dest << 8) & 0xFF00;
1297 } while (cmpxchg64(&pi_desc->control, old.control,
1298 new.control) != old.control);
1303 * Clear SN before reading the bitmap. The VT-d firmware
1304 * writes the bitmap and reads SN atomically (5.2.3 in the
1305 * spec), so it doesn't really have a memory barrier that
1306 * pairs with this, but we cannot do that and we need one.
1308 smp_mb__after_atomic();
1310 if (!pi_is_pir_empty(pi_desc))
1314 void vmx_vcpu_load_vmcs(struct kvm_vcpu *vcpu, int cpu,
1315 struct loaded_vmcs *buddy)
1317 struct vcpu_vmx *vmx = to_vmx(vcpu);
1318 bool already_loaded = vmx->loaded_vmcs->cpu == cpu;
1321 if (!already_loaded) {
1322 loaded_vmcs_clear(vmx->loaded_vmcs);
1323 local_irq_disable();
1326 * Ensure loaded_vmcs->cpu is read before adding loaded_vmcs to
1327 * this cpu's percpu list, otherwise it may not yet be deleted
1328 * from its previous cpu's percpu list. Pairs with the
1329 * smb_wmb() in __loaded_vmcs_clear().
1333 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1334 &per_cpu(loaded_vmcss_on_cpu, cpu));
1338 prev = per_cpu(current_vmcs, cpu);
1339 if (prev != vmx->loaded_vmcs->vmcs) {
1340 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1341 vmcs_load(vmx->loaded_vmcs->vmcs);
1344 * No indirect branch prediction barrier needed when switching
1345 * the active VMCS within a guest, e.g. on nested VM-Enter.
1346 * The L1 VMM can protect itself with retpolines, IBPB or IBRS.
1348 if (!buddy || WARN_ON_ONCE(buddy->vmcs != prev))
1349 indirect_branch_prediction_barrier();
1352 if (!already_loaded) {
1353 void *gdt = get_current_gdt_ro();
1354 unsigned long sysenter_esp;
1357 * Flush all EPTP/VPID contexts, the new pCPU may have stale
1358 * TLB entries from its previous association with the vCPU.
1360 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1363 * Linux uses per-cpu TSS and GDT, so set these when switching
1364 * processors. See 22.2.4.
1366 vmcs_writel(HOST_TR_BASE,
1367 (unsigned long)&get_cpu_entry_area(cpu)->tss.x86_tss);
1368 vmcs_writel(HOST_GDTR_BASE, (unsigned long)gdt); /* 22.2.4 */
1370 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1371 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1373 vmx->loaded_vmcs->cpu = cpu;
1376 /* Setup TSC multiplier */
1377 if (kvm_has_tsc_control &&
1378 vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio)
1379 decache_tsc_multiplier(vmx);
1383 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1384 * vcpu mutex is already taken.
1386 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1388 struct vcpu_vmx *vmx = to_vmx(vcpu);
1390 vmx_vcpu_load_vmcs(vcpu, cpu, NULL);
1392 vmx_vcpu_pi_load(vcpu, cpu);
1394 vmx->host_debugctlmsr = get_debugctlmsr();
1397 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
1399 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1401 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
1402 !irq_remapping_cap(IRQ_POSTING_CAP) ||
1403 !kvm_vcpu_apicv_active(vcpu))
1406 /* Set SN when the vCPU is preempted */
1407 if (vcpu->preempted)
1411 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1413 vmx_vcpu_pi_put(vcpu);
1415 vmx_prepare_switch_to_host(to_vmx(vcpu));
1418 static bool emulation_required(struct kvm_vcpu *vcpu)
1420 return emulate_invalid_guest_state && !guest_state_valid(vcpu);
1423 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1425 struct vcpu_vmx *vmx = to_vmx(vcpu);
1426 unsigned long rflags, save_rflags;
1428 if (!kvm_register_is_available(vcpu, VCPU_EXREG_RFLAGS)) {
1429 kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1430 rflags = vmcs_readl(GUEST_RFLAGS);
1431 if (vmx->rmode.vm86_active) {
1432 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1433 save_rflags = vmx->rmode.save_rflags;
1434 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1436 vmx->rflags = rflags;
1441 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1443 struct vcpu_vmx *vmx = to_vmx(vcpu);
1444 unsigned long old_rflags;
1446 if (enable_unrestricted_guest) {
1447 kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1448 vmx->rflags = rflags;
1449 vmcs_writel(GUEST_RFLAGS, rflags);
1453 old_rflags = vmx_get_rflags(vcpu);
1454 vmx->rflags = rflags;
1455 if (vmx->rmode.vm86_active) {
1456 vmx->rmode.save_rflags = rflags;
1457 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1459 vmcs_writel(GUEST_RFLAGS, rflags);
1461 if ((old_rflags ^ vmx->rflags) & X86_EFLAGS_VM)
1462 vmx->emulation_required = emulation_required(vcpu);
1465 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
1467 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1470 if (interruptibility & GUEST_INTR_STATE_STI)
1471 ret |= KVM_X86_SHADOW_INT_STI;
1472 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1473 ret |= KVM_X86_SHADOW_INT_MOV_SS;
1478 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1480 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1481 u32 interruptibility = interruptibility_old;
1483 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1485 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1486 interruptibility |= GUEST_INTR_STATE_MOV_SS;
1487 else if (mask & KVM_X86_SHADOW_INT_STI)
1488 interruptibility |= GUEST_INTR_STATE_STI;
1490 if ((interruptibility != interruptibility_old))
1491 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1494 static int vmx_rtit_ctl_check(struct kvm_vcpu *vcpu, u64 data)
1496 struct vcpu_vmx *vmx = to_vmx(vcpu);
1497 unsigned long value;
1500 * Any MSR write that attempts to change bits marked reserved will
1503 if (data & vmx->pt_desc.ctl_bitmask)
1507 * Any attempt to modify IA32_RTIT_CTL while TraceEn is set will
1508 * result in a #GP unless the same write also clears TraceEn.
1510 if ((vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) &&
1511 ((vmx->pt_desc.guest.ctl ^ data) & ~RTIT_CTL_TRACEEN))
1515 * WRMSR to IA32_RTIT_CTL that sets TraceEn but clears this bit
1516 * and FabricEn would cause #GP, if
1517 * CPUID.(EAX=14H, ECX=0):ECX.SNGLRGNOUT[bit 2] = 0
1519 if ((data & RTIT_CTL_TRACEEN) && !(data & RTIT_CTL_TOPA) &&
1520 !(data & RTIT_CTL_FABRIC_EN) &&
1521 !intel_pt_validate_cap(vmx->pt_desc.caps,
1522 PT_CAP_single_range_output))
1526 * MTCFreq, CycThresh and PSBFreq encodings check, any MSR write that
1527 * utilize encodings marked reserved will casue a #GP fault.
1529 value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc_periods);
1530 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc) &&
1531 !test_bit((data & RTIT_CTL_MTC_RANGE) >>
1532 RTIT_CTL_MTC_RANGE_OFFSET, &value))
1534 value = intel_pt_validate_cap(vmx->pt_desc.caps,
1535 PT_CAP_cycle_thresholds);
1536 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1537 !test_bit((data & RTIT_CTL_CYC_THRESH) >>
1538 RTIT_CTL_CYC_THRESH_OFFSET, &value))
1540 value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_periods);
1541 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1542 !test_bit((data & RTIT_CTL_PSB_FREQ) >>
1543 RTIT_CTL_PSB_FREQ_OFFSET, &value))
1547 * If ADDRx_CFG is reserved or the encodings is >2 will
1548 * cause a #GP fault.
1550 value = (data & RTIT_CTL_ADDR0) >> RTIT_CTL_ADDR0_OFFSET;
1551 if ((value && (vmx->pt_desc.addr_range < 1)) || (value > 2))
1553 value = (data & RTIT_CTL_ADDR1) >> RTIT_CTL_ADDR1_OFFSET;
1554 if ((value && (vmx->pt_desc.addr_range < 2)) || (value > 2))
1556 value = (data & RTIT_CTL_ADDR2) >> RTIT_CTL_ADDR2_OFFSET;
1557 if ((value && (vmx->pt_desc.addr_range < 3)) || (value > 2))
1559 value = (data & RTIT_CTL_ADDR3) >> RTIT_CTL_ADDR3_OFFSET;
1560 if ((value && (vmx->pt_desc.addr_range < 4)) || (value > 2))
1566 static int skip_emulated_instruction(struct kvm_vcpu *vcpu)
1568 unsigned long rip, orig_rip;
1571 * Using VMCS.VM_EXIT_INSTRUCTION_LEN on EPT misconfig depends on
1572 * undefined behavior: Intel's SDM doesn't mandate the VMCS field be
1573 * set when EPT misconfig occurs. In practice, real hardware updates
1574 * VM_EXIT_INSTRUCTION_LEN on EPT misconfig, but other hypervisors
1575 * (namely Hyper-V) don't set it due to it being undefined behavior,
1576 * i.e. we end up advancing IP with some random value.
1578 if (!static_cpu_has(X86_FEATURE_HYPERVISOR) ||
1579 to_vmx(vcpu)->exit_reason != EXIT_REASON_EPT_MISCONFIG) {
1580 orig_rip = kvm_rip_read(vcpu);
1581 rip = orig_rip + vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1582 #ifdef CONFIG_X86_64
1584 * We need to mask out the high 32 bits of RIP if not in 64-bit
1585 * mode, but just finding out that we are in 64-bit mode is
1586 * quite expensive. Only do it if there was a carry.
1588 if (unlikely(((rip ^ orig_rip) >> 31) == 3) && !is_64_bit_mode(vcpu))
1591 kvm_rip_write(vcpu, rip);
1593 if (!kvm_emulate_instruction(vcpu, EMULTYPE_SKIP))
1597 /* skipping an emulated instruction also counts */
1598 vmx_set_interrupt_shadow(vcpu, 0);
1605 * Recognizes a pending MTF VM-exit and records the nested state for later
1608 static void vmx_update_emulated_instruction(struct kvm_vcpu *vcpu)
1610 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1611 struct vcpu_vmx *vmx = to_vmx(vcpu);
1613 if (!is_guest_mode(vcpu))
1617 * Per the SDM, MTF takes priority over debug-trap exceptions besides
1618 * T-bit traps. As instruction emulation is completed (i.e. at the
1619 * instruction boundary), any #DB exception pending delivery must be a
1620 * debug-trap. Record the pending MTF state to be delivered in
1621 * vmx_check_nested_events().
1623 if (nested_cpu_has_mtf(vmcs12) &&
1624 (!vcpu->arch.exception.pending ||
1625 vcpu->arch.exception.nr == DB_VECTOR))
1626 vmx->nested.mtf_pending = true;
1628 vmx->nested.mtf_pending = false;
1631 static int vmx_skip_emulated_instruction(struct kvm_vcpu *vcpu)
1633 vmx_update_emulated_instruction(vcpu);
1634 return skip_emulated_instruction(vcpu);
1637 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1640 * Ensure that we clear the HLT state in the VMCS. We don't need to
1641 * explicitly skip the instruction because if the HLT state is set,
1642 * then the instruction is already executing and RIP has already been
1645 if (kvm_hlt_in_guest(vcpu->kvm) &&
1646 vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1647 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1650 static void vmx_queue_exception(struct kvm_vcpu *vcpu)
1652 struct vcpu_vmx *vmx = to_vmx(vcpu);
1653 unsigned nr = vcpu->arch.exception.nr;
1654 bool has_error_code = vcpu->arch.exception.has_error_code;
1655 u32 error_code = vcpu->arch.exception.error_code;
1656 u32 intr_info = nr | INTR_INFO_VALID_MASK;
1658 kvm_deliver_exception_payload(vcpu);
1660 if (has_error_code) {
1661 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1662 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1665 if (vmx->rmode.vm86_active) {
1667 if (kvm_exception_is_soft(nr))
1668 inc_eip = vcpu->arch.event_exit_inst_len;
1669 kvm_inject_realmode_interrupt(vcpu, nr, inc_eip);
1673 WARN_ON_ONCE(vmx->emulation_required);
1675 if (kvm_exception_is_soft(nr)) {
1676 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1677 vmx->vcpu.arch.event_exit_inst_len);
1678 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1680 intr_info |= INTR_TYPE_HARD_EXCEPTION;
1682 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1684 vmx_clear_hlt(vcpu);
1688 * Swap MSR entry in host/guest MSR entry array.
1690 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1692 struct shared_msr_entry tmp;
1694 tmp = vmx->guest_msrs[to];
1695 vmx->guest_msrs[to] = vmx->guest_msrs[from];
1696 vmx->guest_msrs[from] = tmp;
1700 * Set up the vmcs to automatically save and restore system
1701 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
1702 * mode, as fiddling with msrs is very expensive.
1704 static void setup_msrs(struct vcpu_vmx *vmx)
1706 int save_nmsrs, index;
1709 #ifdef CONFIG_X86_64
1711 * The SYSCALL MSRs are only needed on long mode guests, and only
1712 * when EFER.SCE is set.
1714 if (is_long_mode(&vmx->vcpu) && (vmx->vcpu.arch.efer & EFER_SCE)) {
1715 index = __find_msr_index(vmx, MSR_STAR);
1717 move_msr_up(vmx, index, save_nmsrs++);
1718 index = __find_msr_index(vmx, MSR_LSTAR);
1720 move_msr_up(vmx, index, save_nmsrs++);
1721 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1723 move_msr_up(vmx, index, save_nmsrs++);
1726 index = __find_msr_index(vmx, MSR_EFER);
1727 if (index >= 0 && update_transition_efer(vmx, index))
1728 move_msr_up(vmx, index, save_nmsrs++);
1729 index = __find_msr_index(vmx, MSR_TSC_AUX);
1730 if (index >= 0 && guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDTSCP))
1731 move_msr_up(vmx, index, save_nmsrs++);
1732 index = __find_msr_index(vmx, MSR_IA32_TSX_CTRL);
1734 move_msr_up(vmx, index, save_nmsrs++);
1736 vmx->save_nmsrs = save_nmsrs;
1737 vmx->guest_msrs_ready = false;
1739 if (cpu_has_vmx_msr_bitmap())
1740 vmx_update_msr_bitmap(&vmx->vcpu);
1743 static u64 vmx_write_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1745 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1746 u64 g_tsc_offset = 0;
1749 * We're here if L1 chose not to trap WRMSR to TSC. According
1750 * to the spec, this should set L1's TSC; The offset that L1
1751 * set for L2 remains unchanged, and still needs to be added
1752 * to the newly set TSC to get L2's TSC.
1754 if (is_guest_mode(vcpu) &&
1755 (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING))
1756 g_tsc_offset = vmcs12->tsc_offset;
1758 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1759 vcpu->arch.tsc_offset - g_tsc_offset,
1761 vmcs_write64(TSC_OFFSET, offset + g_tsc_offset);
1762 return offset + g_tsc_offset;
1766 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
1767 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
1768 * all guests if the "nested" module option is off, and can also be disabled
1769 * for a single guest by disabling its VMX cpuid bit.
1771 bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
1773 return nested && guest_cpuid_has(vcpu, X86_FEATURE_VMX);
1776 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
1779 uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
1781 return !(val & ~valid_bits);
1784 static int vmx_get_msr_feature(struct kvm_msr_entry *msr)
1786 switch (msr->index) {
1787 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1790 return vmx_get_vmx_msr(&vmcs_config.nested, msr->index, &msr->data);
1797 * Reads an msr value (of 'msr_index') into 'pdata'.
1798 * Returns 0 on success, non-0 otherwise.
1799 * Assumes vcpu_load() was already called.
1801 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1803 struct vcpu_vmx *vmx = to_vmx(vcpu);
1804 struct shared_msr_entry *msr;
1807 switch (msr_info->index) {
1808 #ifdef CONFIG_X86_64
1810 msr_info->data = vmcs_readl(GUEST_FS_BASE);
1813 msr_info->data = vmcs_readl(GUEST_GS_BASE);
1815 case MSR_KERNEL_GS_BASE:
1816 msr_info->data = vmx_read_guest_kernel_gs_base(vmx);
1820 return kvm_get_msr_common(vcpu, msr_info);
1821 case MSR_IA32_TSX_CTRL:
1822 if (!msr_info->host_initiated &&
1823 !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
1825 goto find_shared_msr;
1826 case MSR_IA32_UMWAIT_CONTROL:
1827 if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
1830 msr_info->data = vmx->msr_ia32_umwait_control;
1832 case MSR_IA32_SPEC_CTRL:
1833 if (!msr_info->host_initiated &&
1834 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1837 msr_info->data = to_vmx(vcpu)->spec_ctrl;
1839 case MSR_IA32_SYSENTER_CS:
1840 msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
1842 case MSR_IA32_SYSENTER_EIP:
1843 msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
1845 case MSR_IA32_SYSENTER_ESP:
1846 msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
1848 case MSR_IA32_BNDCFGS:
1849 if (!kvm_mpx_supported() ||
1850 (!msr_info->host_initiated &&
1851 !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1853 msr_info->data = vmcs_read64(GUEST_BNDCFGS);
1855 case MSR_IA32_MCG_EXT_CTL:
1856 if (!msr_info->host_initiated &&
1857 !(vmx->msr_ia32_feature_control &
1858 FEAT_CTL_LMCE_ENABLED))
1860 msr_info->data = vcpu->arch.mcg_ext_ctl;
1862 case MSR_IA32_FEAT_CTL:
1863 msr_info->data = vmx->msr_ia32_feature_control;
1865 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1866 if (!nested_vmx_allowed(vcpu))
1868 if (vmx_get_vmx_msr(&vmx->nested.msrs, msr_info->index,
1872 * Enlightened VMCS v1 doesn't have certain fields, but buggy
1873 * Hyper-V versions are still trying to use corresponding
1874 * features when they are exposed. Filter out the essential
1877 if (!msr_info->host_initiated &&
1878 vmx->nested.enlightened_vmcs_enabled)
1879 nested_evmcs_filter_control_msr(msr_info->index,
1882 case MSR_IA32_RTIT_CTL:
1883 if (!vmx_pt_mode_is_host_guest())
1885 msr_info->data = vmx->pt_desc.guest.ctl;
1887 case MSR_IA32_RTIT_STATUS:
1888 if (!vmx_pt_mode_is_host_guest())
1890 msr_info->data = vmx->pt_desc.guest.status;
1892 case MSR_IA32_RTIT_CR3_MATCH:
1893 if (!vmx_pt_mode_is_host_guest() ||
1894 !intel_pt_validate_cap(vmx->pt_desc.caps,
1895 PT_CAP_cr3_filtering))
1897 msr_info->data = vmx->pt_desc.guest.cr3_match;
1899 case MSR_IA32_RTIT_OUTPUT_BASE:
1900 if (!vmx_pt_mode_is_host_guest() ||
1901 (!intel_pt_validate_cap(vmx->pt_desc.caps,
1902 PT_CAP_topa_output) &&
1903 !intel_pt_validate_cap(vmx->pt_desc.caps,
1904 PT_CAP_single_range_output)))
1906 msr_info->data = vmx->pt_desc.guest.output_base;
1908 case MSR_IA32_RTIT_OUTPUT_MASK:
1909 if (!vmx_pt_mode_is_host_guest() ||
1910 (!intel_pt_validate_cap(vmx->pt_desc.caps,
1911 PT_CAP_topa_output) &&
1912 !intel_pt_validate_cap(vmx->pt_desc.caps,
1913 PT_CAP_single_range_output)))
1915 msr_info->data = vmx->pt_desc.guest.output_mask;
1917 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
1918 index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
1919 if (!vmx_pt_mode_is_host_guest() ||
1920 (index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
1921 PT_CAP_num_address_ranges)))
1924 msr_info->data = vmx->pt_desc.guest.addr_b[index / 2];
1926 msr_info->data = vmx->pt_desc.guest.addr_a[index / 2];
1929 if (!msr_info->host_initiated &&
1930 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
1932 goto find_shared_msr;
1935 msr = find_msr_entry(vmx, msr_info->index);
1937 msr_info->data = msr->data;
1940 return kvm_get_msr_common(vcpu, msr_info);
1946 static u64 nested_vmx_truncate_sysenter_addr(struct kvm_vcpu *vcpu,
1949 #ifdef CONFIG_X86_64
1950 if (!guest_cpuid_has(vcpu, X86_FEATURE_LM))
1953 return (unsigned long)data;
1957 * Writes msr value into the appropriate "register".
1958 * Returns 0 on success, non-0 otherwise.
1959 * Assumes vcpu_load() was already called.
1961 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1963 struct vcpu_vmx *vmx = to_vmx(vcpu);
1964 struct shared_msr_entry *msr;
1966 u32 msr_index = msr_info->index;
1967 u64 data = msr_info->data;
1970 switch (msr_index) {
1972 ret = kvm_set_msr_common(vcpu, msr_info);
1974 #ifdef CONFIG_X86_64
1976 vmx_segment_cache_clear(vmx);
1977 vmcs_writel(GUEST_FS_BASE, data);
1980 vmx_segment_cache_clear(vmx);
1981 vmcs_writel(GUEST_GS_BASE, data);
1983 case MSR_KERNEL_GS_BASE:
1984 vmx_write_guest_kernel_gs_base(vmx, data);
1987 case MSR_IA32_SYSENTER_CS:
1988 if (is_guest_mode(vcpu))
1989 get_vmcs12(vcpu)->guest_sysenter_cs = data;
1990 vmcs_write32(GUEST_SYSENTER_CS, data);
1992 case MSR_IA32_SYSENTER_EIP:
1993 if (is_guest_mode(vcpu)) {
1994 data = nested_vmx_truncate_sysenter_addr(vcpu, data);
1995 get_vmcs12(vcpu)->guest_sysenter_eip = data;
1997 vmcs_writel(GUEST_SYSENTER_EIP, data);
1999 case MSR_IA32_SYSENTER_ESP:
2000 if (is_guest_mode(vcpu)) {
2001 data = nested_vmx_truncate_sysenter_addr(vcpu, data);
2002 get_vmcs12(vcpu)->guest_sysenter_esp = data;
2004 vmcs_writel(GUEST_SYSENTER_ESP, data);
2006 case MSR_IA32_DEBUGCTLMSR:
2007 if (is_guest_mode(vcpu) && get_vmcs12(vcpu)->vm_exit_controls &
2008 VM_EXIT_SAVE_DEBUG_CONTROLS)
2009 get_vmcs12(vcpu)->guest_ia32_debugctl = data;
2011 ret = kvm_set_msr_common(vcpu, msr_info);
2014 case MSR_IA32_BNDCFGS:
2015 if (!kvm_mpx_supported() ||
2016 (!msr_info->host_initiated &&
2017 !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
2019 if (is_noncanonical_address(data & PAGE_MASK, vcpu) ||
2020 (data & MSR_IA32_BNDCFGS_RSVD))
2022 vmcs_write64(GUEST_BNDCFGS, data);
2024 case MSR_IA32_UMWAIT_CONTROL:
2025 if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
2028 /* The reserved bit 1 and non-32 bit [63:32] should be zero */
2029 if (data & (BIT_ULL(1) | GENMASK_ULL(63, 32)))
2032 vmx->msr_ia32_umwait_control = data;
2034 case MSR_IA32_SPEC_CTRL:
2035 if (!msr_info->host_initiated &&
2036 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
2039 if (data & ~kvm_spec_ctrl_valid_bits(vcpu))
2042 vmx->spec_ctrl = data;
2048 * When it's written (to non-zero) for the first time, pass
2052 * The handling of the MSR bitmap for L2 guests is done in
2053 * nested_vmx_prepare_msr_bitmap. We should not touch the
2054 * vmcs02.msr_bitmap here since it gets completely overwritten
2055 * in the merging. We update the vmcs01 here for L1 as well
2056 * since it will end up touching the MSR anyway now.
2058 vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap,
2062 case MSR_IA32_TSX_CTRL:
2063 if (!msr_info->host_initiated &&
2064 !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
2066 if (data & ~(TSX_CTRL_RTM_DISABLE | TSX_CTRL_CPUID_CLEAR))
2068 goto find_shared_msr;
2069 case MSR_IA32_PRED_CMD:
2070 if (!msr_info->host_initiated &&
2071 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
2074 if (data & ~PRED_CMD_IBPB)
2076 if (!boot_cpu_has(X86_FEATURE_SPEC_CTRL))
2081 wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
2085 * When it's written (to non-zero) for the first time, pass
2089 * The handling of the MSR bitmap for L2 guests is done in
2090 * nested_vmx_prepare_msr_bitmap. We should not touch the
2091 * vmcs02.msr_bitmap here since it gets completely overwritten
2094 vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap, MSR_IA32_PRED_CMD,
2097 case MSR_IA32_CR_PAT:
2098 if (!kvm_pat_valid(data))
2101 if (is_guest_mode(vcpu) &&
2102 get_vmcs12(vcpu)->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
2103 get_vmcs12(vcpu)->guest_ia32_pat = data;
2105 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2106 vmcs_write64(GUEST_IA32_PAT, data);
2107 vcpu->arch.pat = data;
2110 ret = kvm_set_msr_common(vcpu, msr_info);
2112 case MSR_IA32_TSC_ADJUST:
2113 ret = kvm_set_msr_common(vcpu, msr_info);
2115 case MSR_IA32_MCG_EXT_CTL:
2116 if ((!msr_info->host_initiated &&
2117 !(to_vmx(vcpu)->msr_ia32_feature_control &
2118 FEAT_CTL_LMCE_ENABLED)) ||
2119 (data & ~MCG_EXT_CTL_LMCE_EN))
2121 vcpu->arch.mcg_ext_ctl = data;
2123 case MSR_IA32_FEAT_CTL:
2124 if (!vmx_feature_control_msr_valid(vcpu, data) ||
2125 (to_vmx(vcpu)->msr_ia32_feature_control &
2126 FEAT_CTL_LOCKED && !msr_info->host_initiated))
2128 vmx->msr_ia32_feature_control = data;
2129 if (msr_info->host_initiated && data == 0)
2130 vmx_leave_nested(vcpu);
2132 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2133 if (!msr_info->host_initiated)
2134 return 1; /* they are read-only */
2135 if (!nested_vmx_allowed(vcpu))
2137 return vmx_set_vmx_msr(vcpu, msr_index, data);
2138 case MSR_IA32_RTIT_CTL:
2139 if (!vmx_pt_mode_is_host_guest() ||
2140 vmx_rtit_ctl_check(vcpu, data) ||
2143 vmcs_write64(GUEST_IA32_RTIT_CTL, data);
2144 vmx->pt_desc.guest.ctl = data;
2145 pt_update_intercept_for_msr(vmx);
2147 case MSR_IA32_RTIT_STATUS:
2148 if (!pt_can_write_msr(vmx))
2150 if (data & MSR_IA32_RTIT_STATUS_MASK)
2152 vmx->pt_desc.guest.status = data;
2154 case MSR_IA32_RTIT_CR3_MATCH:
2155 if (!pt_can_write_msr(vmx))
2157 if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2158 PT_CAP_cr3_filtering))
2160 vmx->pt_desc.guest.cr3_match = data;
2162 case MSR_IA32_RTIT_OUTPUT_BASE:
2163 if (!pt_can_write_msr(vmx))
2165 if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2166 PT_CAP_topa_output) &&
2167 !intel_pt_validate_cap(vmx->pt_desc.caps,
2168 PT_CAP_single_range_output))
2170 if (data & MSR_IA32_RTIT_OUTPUT_BASE_MASK)
2172 vmx->pt_desc.guest.output_base = data;
2174 case MSR_IA32_RTIT_OUTPUT_MASK:
2175 if (!pt_can_write_msr(vmx))
2177 if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2178 PT_CAP_topa_output) &&
2179 !intel_pt_validate_cap(vmx->pt_desc.caps,
2180 PT_CAP_single_range_output))
2182 vmx->pt_desc.guest.output_mask = data;
2184 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
2185 if (!pt_can_write_msr(vmx))
2187 index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
2188 if (index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
2189 PT_CAP_num_address_ranges))
2191 if (is_noncanonical_address(data, vcpu))
2194 vmx->pt_desc.guest.addr_b[index / 2] = data;
2196 vmx->pt_desc.guest.addr_a[index / 2] = data;
2199 if (!msr_info->host_initiated &&
2200 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
2202 /* Check reserved bit, higher 32 bits should be zero */
2203 if ((data >> 32) != 0)
2205 goto find_shared_msr;
2209 msr = find_msr_entry(vmx, msr_index);
2211 ret = vmx_set_guest_msr(vmx, msr, data);
2213 ret = kvm_set_msr_common(vcpu, msr_info);
2219 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2221 unsigned long guest_owned_bits;
2223 kvm_register_mark_available(vcpu, reg);
2227 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2230 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2232 case VCPU_EXREG_PDPTR:
2234 ept_save_pdptrs(vcpu);
2236 case VCPU_EXREG_CR0:
2237 guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
2239 vcpu->arch.cr0 &= ~guest_owned_bits;
2240 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & guest_owned_bits;
2242 case VCPU_EXREG_CR3:
2243 if (enable_unrestricted_guest || (enable_ept && is_paging(vcpu)))
2244 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2246 case VCPU_EXREG_CR4:
2247 guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
2249 vcpu->arch.cr4 &= ~guest_owned_bits;
2250 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & guest_owned_bits;
2258 static __init int cpu_has_kvm_support(void)
2260 return cpu_has_vmx();
2263 static __init int vmx_disabled_by_bios(void)
2265 return !boot_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
2266 !boot_cpu_has(X86_FEATURE_VMX);
2269 static int kvm_cpu_vmxon(u64 vmxon_pointer)
2273 cr4_set_bits(X86_CR4_VMXE);
2274 intel_pt_handle_vmx(1);
2276 asm_volatile_goto("1: vmxon %[vmxon_pointer]\n\t"
2277 _ASM_EXTABLE(1b, %l[fault])
2278 : : [vmxon_pointer] "m"(vmxon_pointer)
2283 WARN_ONCE(1, "VMXON faulted, MSR_IA32_FEAT_CTL (0x3a) = 0x%llx\n",
2284 rdmsrl_safe(MSR_IA32_FEAT_CTL, &msr) ? 0xdeadbeef : msr);
2285 intel_pt_handle_vmx(0);
2286 cr4_clear_bits(X86_CR4_VMXE);
2291 static int hardware_enable(void)
2293 int cpu = raw_smp_processor_id();
2294 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2297 if (cr4_read_shadow() & X86_CR4_VMXE)
2301 * This can happen if we hot-added a CPU but failed to allocate
2302 * VP assist page for it.
2304 if (static_branch_unlikely(&enable_evmcs) &&
2305 !hv_get_vp_assist_page(cpu))
2308 r = kvm_cpu_vmxon(phys_addr);
2318 static void vmclear_local_loaded_vmcss(void)
2320 int cpu = raw_smp_processor_id();
2321 struct loaded_vmcs *v, *n;
2323 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2324 loaded_vmcss_on_cpu_link)
2325 __loaded_vmcs_clear(v);
2329 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2332 static void kvm_cpu_vmxoff(void)
2334 asm volatile (__ex("vmxoff"));
2336 intel_pt_handle_vmx(0);
2337 cr4_clear_bits(X86_CR4_VMXE);
2340 static void hardware_disable(void)
2342 vmclear_local_loaded_vmcss();
2347 * There is no X86_FEATURE for SGX yet, but anyway we need to query CPUID
2348 * directly instead of going through cpu_has(), to ensure KVM is trapping
2349 * ENCLS whenever it's supported in hardware. It does not matter whether
2350 * the host OS supports or has enabled SGX.
2352 static bool cpu_has_sgx(void)
2354 return cpuid_eax(0) >= 0x12 && (cpuid_eax(0x12) & BIT(0));
2357 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2358 u32 msr, u32 *result)
2360 u32 vmx_msr_low, vmx_msr_high;
2361 u32 ctl = ctl_min | ctl_opt;
2363 rdmsr(msr, vmx_msr_low, vmx_msr_high);
2365 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2366 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
2368 /* Ensure minimum (required) set of control bits are supported. */
2376 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf,
2377 struct vmx_capability *vmx_cap)
2379 u32 vmx_msr_low, vmx_msr_high;
2380 u32 min, opt, min2, opt2;
2381 u32 _pin_based_exec_control = 0;
2382 u32 _cpu_based_exec_control = 0;
2383 u32 _cpu_based_2nd_exec_control = 0;
2384 u32 _vmexit_control = 0;
2385 u32 _vmentry_control = 0;
2387 memset(vmcs_conf, 0, sizeof(*vmcs_conf));
2388 min = CPU_BASED_HLT_EXITING |
2389 #ifdef CONFIG_X86_64
2390 CPU_BASED_CR8_LOAD_EXITING |
2391 CPU_BASED_CR8_STORE_EXITING |
2393 CPU_BASED_CR3_LOAD_EXITING |
2394 CPU_BASED_CR3_STORE_EXITING |
2395 CPU_BASED_UNCOND_IO_EXITING |
2396 CPU_BASED_MOV_DR_EXITING |
2397 CPU_BASED_USE_TSC_OFFSETTING |
2398 CPU_BASED_MWAIT_EXITING |
2399 CPU_BASED_MONITOR_EXITING |
2400 CPU_BASED_INVLPG_EXITING |
2401 CPU_BASED_RDPMC_EXITING;
2403 opt = CPU_BASED_TPR_SHADOW |
2404 CPU_BASED_USE_MSR_BITMAPS |
2405 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2406 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2407 &_cpu_based_exec_control) < 0)
2409 #ifdef CONFIG_X86_64
2410 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2411 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2412 ~CPU_BASED_CR8_STORE_EXITING;
2414 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2416 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2417 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2418 SECONDARY_EXEC_WBINVD_EXITING |
2419 SECONDARY_EXEC_ENABLE_VPID |
2420 SECONDARY_EXEC_ENABLE_EPT |
2421 SECONDARY_EXEC_UNRESTRICTED_GUEST |
2422 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2423 SECONDARY_EXEC_DESC |
2424 SECONDARY_EXEC_RDTSCP |
2425 SECONDARY_EXEC_ENABLE_INVPCID |
2426 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2427 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2428 SECONDARY_EXEC_SHADOW_VMCS |
2429 SECONDARY_EXEC_XSAVES |
2430 SECONDARY_EXEC_RDSEED_EXITING |
2431 SECONDARY_EXEC_RDRAND_EXITING |
2432 SECONDARY_EXEC_ENABLE_PML |
2433 SECONDARY_EXEC_TSC_SCALING |
2434 SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE |
2435 SECONDARY_EXEC_PT_USE_GPA |
2436 SECONDARY_EXEC_PT_CONCEAL_VMX |
2437 SECONDARY_EXEC_ENABLE_VMFUNC;
2439 opt2 |= SECONDARY_EXEC_ENCLS_EXITING;
2440 if (adjust_vmx_controls(min2, opt2,
2441 MSR_IA32_VMX_PROCBASED_CTLS2,
2442 &_cpu_based_2nd_exec_control) < 0)
2445 #ifndef CONFIG_X86_64
2446 if (!(_cpu_based_2nd_exec_control &
2447 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2448 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2451 if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2452 _cpu_based_2nd_exec_control &= ~(
2453 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2454 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2455 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
2457 rdmsr_safe(MSR_IA32_VMX_EPT_VPID_CAP,
2458 &vmx_cap->ept, &vmx_cap->vpid);
2460 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2461 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2463 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2464 CPU_BASED_CR3_STORE_EXITING |
2465 CPU_BASED_INVLPG_EXITING);
2466 } else if (vmx_cap->ept) {
2468 pr_warn_once("EPT CAP should not exist if not support "
2469 "1-setting enable EPT VM-execution control\n");
2471 if (!(_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_VPID) &&
2474 pr_warn_once("VPID CAP should not exist if not support "
2475 "1-setting enable VPID VM-execution control\n");
2478 min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
2479 #ifdef CONFIG_X86_64
2480 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2482 opt = VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL |
2483 VM_EXIT_LOAD_IA32_PAT |
2484 VM_EXIT_LOAD_IA32_EFER |
2485 VM_EXIT_CLEAR_BNDCFGS |
2486 VM_EXIT_PT_CONCEAL_PIP |
2487 VM_EXIT_CLEAR_IA32_RTIT_CTL;
2488 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2489 &_vmexit_control) < 0)
2492 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2493 opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR |
2494 PIN_BASED_VMX_PREEMPTION_TIMER;
2495 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2496 &_pin_based_exec_control) < 0)
2499 if (cpu_has_broken_vmx_preemption_timer())
2500 _pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
2501 if (!(_cpu_based_2nd_exec_control &
2502 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
2503 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
2505 min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
2506 opt = VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL |
2507 VM_ENTRY_LOAD_IA32_PAT |
2508 VM_ENTRY_LOAD_IA32_EFER |
2509 VM_ENTRY_LOAD_BNDCFGS |
2510 VM_ENTRY_PT_CONCEAL_PIP |
2511 VM_ENTRY_LOAD_IA32_RTIT_CTL;
2512 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2513 &_vmentry_control) < 0)
2517 * Some cpus support VM_{ENTRY,EXIT}_IA32_PERF_GLOBAL_CTRL but they
2518 * can't be used due to an errata where VM Exit may incorrectly clear
2519 * IA32_PERF_GLOBAL_CTRL[34:32]. Workaround the errata by using the
2520 * MSR load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2522 if (boot_cpu_data.x86 == 0x6) {
2523 switch (boot_cpu_data.x86_model) {
2524 case 26: /* AAK155 */
2525 case 30: /* AAP115 */
2526 case 37: /* AAT100 */
2527 case 44: /* BC86,AAY89,BD102 */
2529 _vmentry_control &= ~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
2530 _vmexit_control &= ~VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
2531 pr_warn_once("kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
2532 "does not work properly. Using workaround\n");
2540 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2542 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2543 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2546 #ifdef CONFIG_X86_64
2547 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2548 if (vmx_msr_high & (1u<<16))
2552 /* Require Write-Back (WB) memory type for VMCS accesses. */
2553 if (((vmx_msr_high >> 18) & 15) != 6)
2556 vmcs_conf->size = vmx_msr_high & 0x1fff;
2557 vmcs_conf->order = get_order(vmcs_conf->size);
2558 vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
2560 vmcs_conf->revision_id = vmx_msr_low;
2562 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2563 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2564 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2565 vmcs_conf->vmexit_ctrl = _vmexit_control;
2566 vmcs_conf->vmentry_ctrl = _vmentry_control;
2568 if (static_branch_unlikely(&enable_evmcs))
2569 evmcs_sanitize_exec_ctrls(vmcs_conf);
2574 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags)
2576 int node = cpu_to_node(cpu);
2580 pages = __alloc_pages_node(node, flags, vmcs_config.order);
2583 vmcs = page_address(pages);
2584 memset(vmcs, 0, vmcs_config.size);
2586 /* KVM supports Enlightened VMCS v1 only */
2587 if (static_branch_unlikely(&enable_evmcs))
2588 vmcs->hdr.revision_id = KVM_EVMCS_VERSION;
2590 vmcs->hdr.revision_id = vmcs_config.revision_id;
2593 vmcs->hdr.shadow_vmcs = 1;
2597 void free_vmcs(struct vmcs *vmcs)
2599 free_pages((unsigned long)vmcs, vmcs_config.order);
2603 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2605 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2607 if (!loaded_vmcs->vmcs)
2609 loaded_vmcs_clear(loaded_vmcs);
2610 free_vmcs(loaded_vmcs->vmcs);
2611 loaded_vmcs->vmcs = NULL;
2612 if (loaded_vmcs->msr_bitmap)
2613 free_page((unsigned long)loaded_vmcs->msr_bitmap);
2614 WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
2617 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2619 loaded_vmcs->vmcs = alloc_vmcs(false);
2620 if (!loaded_vmcs->vmcs)
2623 vmcs_clear(loaded_vmcs->vmcs);
2625 loaded_vmcs->shadow_vmcs = NULL;
2626 loaded_vmcs->hv_timer_soft_disabled = false;
2627 loaded_vmcs->cpu = -1;
2628 loaded_vmcs->launched = 0;
2630 if (cpu_has_vmx_msr_bitmap()) {
2631 loaded_vmcs->msr_bitmap = (unsigned long *)
2632 __get_free_page(GFP_KERNEL_ACCOUNT);
2633 if (!loaded_vmcs->msr_bitmap)
2635 memset(loaded_vmcs->msr_bitmap, 0xff, PAGE_SIZE);
2637 if (IS_ENABLED(CONFIG_HYPERV) &&
2638 static_branch_unlikely(&enable_evmcs) &&
2639 (ms_hyperv.nested_features & HV_X64_NESTED_MSR_BITMAP)) {
2640 struct hv_enlightened_vmcs *evmcs =
2641 (struct hv_enlightened_vmcs *)loaded_vmcs->vmcs;
2643 evmcs->hv_enlightenments_control.msr_bitmap = 1;
2647 memset(&loaded_vmcs->host_state, 0, sizeof(struct vmcs_host_state));
2648 memset(&loaded_vmcs->controls_shadow, 0,
2649 sizeof(struct vmcs_controls_shadow));
2654 free_loaded_vmcs(loaded_vmcs);
2658 static void free_kvm_area(void)
2662 for_each_possible_cpu(cpu) {
2663 free_vmcs(per_cpu(vmxarea, cpu));
2664 per_cpu(vmxarea, cpu) = NULL;
2668 static __init int alloc_kvm_area(void)
2672 for_each_possible_cpu(cpu) {
2675 vmcs = alloc_vmcs_cpu(false, cpu, GFP_KERNEL);
2682 * When eVMCS is enabled, alloc_vmcs_cpu() sets
2683 * vmcs->revision_id to KVM_EVMCS_VERSION instead of
2684 * revision_id reported by MSR_IA32_VMX_BASIC.
2686 * However, even though not explicitly documented by
2687 * TLFS, VMXArea passed as VMXON argument should
2688 * still be marked with revision_id reported by
2691 if (static_branch_unlikely(&enable_evmcs))
2692 vmcs->hdr.revision_id = vmcs_config.revision_id;
2694 per_cpu(vmxarea, cpu) = vmcs;
2699 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
2700 struct kvm_segment *save)
2702 if (!emulate_invalid_guest_state) {
2704 * CS and SS RPL should be equal during guest entry according
2705 * to VMX spec, but in reality it is not always so. Since vcpu
2706 * is in the middle of the transition from real mode to
2707 * protected mode it is safe to assume that RPL 0 is a good
2710 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
2711 save->selector &= ~SEGMENT_RPL_MASK;
2712 save->dpl = save->selector & SEGMENT_RPL_MASK;
2715 vmx_set_segment(vcpu, save, seg);
2718 static void enter_pmode(struct kvm_vcpu *vcpu)
2720 unsigned long flags;
2721 struct vcpu_vmx *vmx = to_vmx(vcpu);
2724 * Update real mode segment cache. It may be not up-to-date if sement
2725 * register was written while vcpu was in a guest mode.
2727 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2728 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2729 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2730 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2731 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2732 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2734 vmx->rmode.vm86_active = 0;
2736 vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2738 flags = vmcs_readl(GUEST_RFLAGS);
2739 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2740 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2741 vmcs_writel(GUEST_RFLAGS, flags);
2743 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
2744 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
2746 update_exception_bitmap(vcpu);
2748 fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2749 fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2750 fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2751 fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2752 fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2753 fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2756 static void fix_rmode_seg(int seg, struct kvm_segment *save)
2758 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2759 struct kvm_segment var = *save;
2762 if (seg == VCPU_SREG_CS)
2765 if (!emulate_invalid_guest_state) {
2766 var.selector = var.base >> 4;
2767 var.base = var.base & 0xffff0;
2777 if (save->base & 0xf)
2778 printk_once(KERN_WARNING "kvm: segment base is not "
2779 "paragraph aligned when entering "
2780 "protected mode (seg=%d)", seg);
2783 vmcs_write16(sf->selector, var.selector);
2784 vmcs_writel(sf->base, var.base);
2785 vmcs_write32(sf->limit, var.limit);
2786 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
2789 static void enter_rmode(struct kvm_vcpu *vcpu)
2791 unsigned long flags;
2792 struct vcpu_vmx *vmx = to_vmx(vcpu);
2793 struct kvm_vmx *kvm_vmx = to_kvm_vmx(vcpu->kvm);
2795 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2796 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2797 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2798 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2799 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2800 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2801 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2803 vmx->rmode.vm86_active = 1;
2806 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
2807 * vcpu. Warn the user that an update is overdue.
2809 if (!kvm_vmx->tss_addr)
2810 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
2811 "called before entering vcpu\n");
2813 vmx_segment_cache_clear(vmx);
2815 vmcs_writel(GUEST_TR_BASE, kvm_vmx->tss_addr);
2816 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
2817 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2819 flags = vmcs_readl(GUEST_RFLAGS);
2820 vmx->rmode.save_rflags = flags;
2822 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2824 vmcs_writel(GUEST_RFLAGS, flags);
2825 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
2826 update_exception_bitmap(vcpu);
2828 fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2829 fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2830 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2831 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2832 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2833 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2835 kvm_mmu_reset_context(vcpu);
2838 void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
2840 struct vcpu_vmx *vmx = to_vmx(vcpu);
2841 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
2846 vcpu->arch.efer = efer;
2847 if (efer & EFER_LMA) {
2848 vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2851 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2853 msr->data = efer & ~EFER_LME;
2858 #ifdef CONFIG_X86_64
2860 static void enter_lmode(struct kvm_vcpu *vcpu)
2864 vmx_segment_cache_clear(to_vmx(vcpu));
2866 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
2867 if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
2868 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
2870 vmcs_write32(GUEST_TR_AR_BYTES,
2871 (guest_tr_ar & ~VMX_AR_TYPE_MASK)
2872 | VMX_AR_TYPE_BUSY_64_TSS);
2874 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
2877 static void exit_lmode(struct kvm_vcpu *vcpu)
2879 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2880 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
2885 static void vmx_flush_tlb_all(struct kvm_vcpu *vcpu)
2887 struct vcpu_vmx *vmx = to_vmx(vcpu);
2890 * INVEPT must be issued when EPT is enabled, irrespective of VPID, as
2891 * the CPU is not required to invalidate guest-physical mappings on
2892 * VM-Entry, even if VPID is disabled. Guest-physical mappings are
2893 * associated with the root EPT structure and not any particular VPID
2894 * (INVVPID also isn't required to invalidate guest-physical mappings).
2898 } else if (enable_vpid) {
2899 if (cpu_has_vmx_invvpid_global()) {
2900 vpid_sync_vcpu_global();
2902 vpid_sync_vcpu_single(vmx->vpid);
2903 vpid_sync_vcpu_single(vmx->nested.vpid02);
2908 static void vmx_flush_tlb_current(struct kvm_vcpu *vcpu)
2910 u64 root_hpa = vcpu->arch.mmu->root_hpa;
2912 /* No flush required if the current context is invalid. */
2913 if (!VALID_PAGE(root_hpa))
2917 ept_sync_context(construct_eptp(vcpu, root_hpa));
2918 else if (!is_guest_mode(vcpu))
2919 vpid_sync_context(to_vmx(vcpu)->vpid);
2921 vpid_sync_context(nested_get_vpid02(vcpu));
2924 static void vmx_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t addr)
2927 * vpid_sync_vcpu_addr() is a nop if vmx->vpid==0, see the comment in
2928 * vmx_flush_tlb_guest() for an explanation of why this is ok.
2930 vpid_sync_vcpu_addr(to_vmx(vcpu)->vpid, addr);
2933 static void vmx_flush_tlb_guest(struct kvm_vcpu *vcpu)
2936 * vpid_sync_context() is a nop if vmx->vpid==0, e.g. if enable_vpid==0
2937 * or a vpid couldn't be allocated for this vCPU. VM-Enter and VM-Exit
2938 * are required to flush GVA->{G,H}PA mappings from the TLB if vpid is
2939 * disabled (VM-Enter with vpid enabled and vpid==0 is disallowed),
2940 * i.e. no explicit INVVPID is necessary.
2942 vpid_sync_context(to_vmx(vcpu)->vpid);
2945 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
2947 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2949 if (!kvm_register_is_dirty(vcpu, VCPU_EXREG_PDPTR))
2952 if (is_pae_paging(vcpu)) {
2953 vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
2954 vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
2955 vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
2956 vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
2960 void ept_save_pdptrs(struct kvm_vcpu *vcpu)
2962 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2964 if (WARN_ON_ONCE(!is_pae_paging(vcpu)))
2967 mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
2968 mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
2969 mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
2970 mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
2972 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
2975 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
2977 struct kvm_vcpu *vcpu)
2979 struct vcpu_vmx *vmx = to_vmx(vcpu);
2981 if (!kvm_register_is_available(vcpu, VCPU_EXREG_CR3))
2982 vmx_cache_reg(vcpu, VCPU_EXREG_CR3);
2983 if (!(cr0 & X86_CR0_PG)) {
2984 /* From paging/starting to nonpaging */
2985 exec_controls_setbit(vmx, CPU_BASED_CR3_LOAD_EXITING |
2986 CPU_BASED_CR3_STORE_EXITING);
2987 vcpu->arch.cr0 = cr0;
2988 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2989 } else if (!is_paging(vcpu)) {
2990 /* From nonpaging to paging */
2991 exec_controls_clearbit(vmx, CPU_BASED_CR3_LOAD_EXITING |
2992 CPU_BASED_CR3_STORE_EXITING);
2993 vcpu->arch.cr0 = cr0;
2994 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2997 if (!(cr0 & X86_CR0_WP))
2998 *hw_cr0 &= ~X86_CR0_WP;
3001 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
3003 struct vcpu_vmx *vmx = to_vmx(vcpu);
3004 unsigned long hw_cr0;
3006 hw_cr0 = (cr0 & ~KVM_VM_CR0_ALWAYS_OFF);
3007 if (enable_unrestricted_guest)
3008 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
3010 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
3012 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
3015 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
3019 #ifdef CONFIG_X86_64
3020 if (vcpu->arch.efer & EFER_LME) {
3021 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
3023 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
3028 if (enable_ept && !enable_unrestricted_guest)
3029 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
3031 vmcs_writel(CR0_READ_SHADOW, cr0);
3032 vmcs_writel(GUEST_CR0, hw_cr0);
3033 vcpu->arch.cr0 = cr0;
3034 kvm_register_mark_available(vcpu, VCPU_EXREG_CR0);
3036 /* depends on vcpu->arch.cr0 to be set to a new value */
3037 vmx->emulation_required = emulation_required(vcpu);
3040 static int vmx_get_tdp_level(struct kvm_vcpu *vcpu)
3042 if (cpu_has_vmx_ept_5levels() && (cpuid_maxphyaddr(vcpu) > 48))
3047 static int get_ept_level(struct kvm_vcpu *vcpu)
3049 if (is_guest_mode(vcpu) && nested_cpu_has_ept(get_vmcs12(vcpu)))
3050 return vmx_eptp_page_walk_level(nested_ept_get_eptp(vcpu));
3052 return vmx_get_tdp_level(vcpu);
3055 u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa)
3057 u64 eptp = VMX_EPTP_MT_WB;
3059 eptp |= (get_ept_level(vcpu) == 5) ? VMX_EPTP_PWL_5 : VMX_EPTP_PWL_4;
3061 if (enable_ept_ad_bits &&
3062 (!is_guest_mode(vcpu) || nested_ept_ad_enabled(vcpu)))
3063 eptp |= VMX_EPTP_AD_ENABLE_BIT;
3064 eptp |= (root_hpa & PAGE_MASK);
3069 void vmx_load_mmu_pgd(struct kvm_vcpu *vcpu, unsigned long pgd)
3071 struct kvm *kvm = vcpu->kvm;
3072 bool update_guest_cr3 = true;
3073 unsigned long guest_cr3;
3077 eptp = construct_eptp(vcpu, pgd);
3078 vmcs_write64(EPT_POINTER, eptp);
3080 if (kvm_x86_ops.tlb_remote_flush) {
3081 spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
3082 to_vmx(vcpu)->ept_pointer = eptp;
3083 to_kvm_vmx(kvm)->ept_pointers_match
3084 = EPT_POINTERS_CHECK;
3085 spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
3088 /* Loading vmcs02.GUEST_CR3 is handled by nested VM-Enter. */
3089 if (is_guest_mode(vcpu))
3090 update_guest_cr3 = false;
3091 else if (!enable_unrestricted_guest && !is_paging(vcpu))
3092 guest_cr3 = to_kvm_vmx(kvm)->ept_identity_map_addr;
3093 else if (test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
3094 guest_cr3 = vcpu->arch.cr3;
3095 else /* vmcs01.GUEST_CR3 is already up-to-date. */
3096 update_guest_cr3 = false;
3097 ept_load_pdptrs(vcpu);
3102 if (update_guest_cr3)
3103 vmcs_writel(GUEST_CR3, guest_cr3);
3106 int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3108 struct vcpu_vmx *vmx = to_vmx(vcpu);
3110 * Pass through host's Machine Check Enable value to hw_cr4, which
3111 * is in force while we are in guest mode. Do not let guests control
3112 * this bit, even if host CR4.MCE == 0.
3114 unsigned long hw_cr4;
3116 hw_cr4 = (cr4_read_shadow() & X86_CR4_MCE) | (cr4 & ~X86_CR4_MCE);
3117 if (enable_unrestricted_guest)
3118 hw_cr4 |= KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST;
3119 else if (vmx->rmode.vm86_active)
3120 hw_cr4 |= KVM_RMODE_VM_CR4_ALWAYS_ON;
3122 hw_cr4 |= KVM_PMODE_VM_CR4_ALWAYS_ON;
3124 if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated()) {
3125 if (cr4 & X86_CR4_UMIP) {
3126 secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_DESC);
3127 hw_cr4 &= ~X86_CR4_UMIP;
3128 } else if (!is_guest_mode(vcpu) ||
3129 !nested_cpu_has2(get_vmcs12(vcpu), SECONDARY_EXEC_DESC)) {
3130 secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_DESC);
3134 if (cr4 & X86_CR4_VMXE) {
3136 * To use VMXON (and later other VMX instructions), a guest
3137 * must first be able to turn on cr4.VMXE (see handle_vmon()).
3138 * So basically the check on whether to allow nested VMX
3139 * is here. We operate under the default treatment of SMM,
3140 * so VMX cannot be enabled under SMM.
3142 if (!nested_vmx_allowed(vcpu) || is_smm(vcpu))
3146 if (vmx->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
3149 vcpu->arch.cr4 = cr4;
3150 kvm_register_mark_available(vcpu, VCPU_EXREG_CR4);
3152 if (!enable_unrestricted_guest) {
3154 if (!is_paging(vcpu)) {
3155 hw_cr4 &= ~X86_CR4_PAE;
3156 hw_cr4 |= X86_CR4_PSE;
3157 } else if (!(cr4 & X86_CR4_PAE)) {
3158 hw_cr4 &= ~X86_CR4_PAE;
3163 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
3164 * hardware. To emulate this behavior, SMEP/SMAP/PKU needs
3165 * to be manually disabled when guest switches to non-paging
3168 * If !enable_unrestricted_guest, the CPU is always running
3169 * with CR0.PG=1 and CR4 needs to be modified.
3170 * If enable_unrestricted_guest, the CPU automatically
3171 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
3173 if (!is_paging(vcpu))
3174 hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
3177 vmcs_writel(CR4_READ_SHADOW, cr4);
3178 vmcs_writel(GUEST_CR4, hw_cr4);
3182 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3184 struct vcpu_vmx *vmx = to_vmx(vcpu);
3187 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3188 *var = vmx->rmode.segs[seg];
3189 if (seg == VCPU_SREG_TR
3190 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3192 var->base = vmx_read_guest_seg_base(vmx, seg);
3193 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3196 var->base = vmx_read_guest_seg_base(vmx, seg);
3197 var->limit = vmx_read_guest_seg_limit(vmx, seg);
3198 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3199 ar = vmx_read_guest_seg_ar(vmx, seg);
3200 var->unusable = (ar >> 16) & 1;
3201 var->type = ar & 15;
3202 var->s = (ar >> 4) & 1;
3203 var->dpl = (ar >> 5) & 3;
3205 * Some userspaces do not preserve unusable property. Since usable
3206 * segment has to be present according to VMX spec we can use present
3207 * property to amend userspace bug by making unusable segment always
3208 * nonpresent. vmx_segment_access_rights() already marks nonpresent
3209 * segment as unusable.
3211 var->present = !var->unusable;
3212 var->avl = (ar >> 12) & 1;
3213 var->l = (ar >> 13) & 1;
3214 var->db = (ar >> 14) & 1;
3215 var->g = (ar >> 15) & 1;
3218 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3220 struct kvm_segment s;
3222 if (to_vmx(vcpu)->rmode.vm86_active) {
3223 vmx_get_segment(vcpu, &s, seg);
3226 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3229 int vmx_get_cpl(struct kvm_vcpu *vcpu)
3231 struct vcpu_vmx *vmx = to_vmx(vcpu);
3233 if (unlikely(vmx->rmode.vm86_active))
3236 int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
3237 return VMX_AR_DPL(ar);
3241 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3245 if (var->unusable || !var->present)
3248 ar = var->type & 15;
3249 ar |= (var->s & 1) << 4;
3250 ar |= (var->dpl & 3) << 5;
3251 ar |= (var->present & 1) << 7;
3252 ar |= (var->avl & 1) << 12;
3253 ar |= (var->l & 1) << 13;
3254 ar |= (var->db & 1) << 14;
3255 ar |= (var->g & 1) << 15;
3261 void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3263 struct vcpu_vmx *vmx = to_vmx(vcpu);
3264 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3266 vmx_segment_cache_clear(vmx);
3268 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3269 vmx->rmode.segs[seg] = *var;
3270 if (seg == VCPU_SREG_TR)
3271 vmcs_write16(sf->selector, var->selector);
3273 fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3277 vmcs_writel(sf->base, var->base);
3278 vmcs_write32(sf->limit, var->limit);
3279 vmcs_write16(sf->selector, var->selector);
3282 * Fix the "Accessed" bit in AR field of segment registers for older
3284 * IA32 arch specifies that at the time of processor reset the
3285 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3286 * is setting it to 0 in the userland code. This causes invalid guest
3287 * state vmexit when "unrestricted guest" mode is turned on.
3288 * Fix for this setup issue in cpu_reset is being pushed in the qemu
3289 * tree. Newer qemu binaries with that qemu fix would not need this
3292 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3293 var->type |= 0x1; /* Accessed */
3295 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3298 vmx->emulation_required = emulation_required(vcpu);
3301 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3303 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3305 *db = (ar >> 14) & 1;
3306 *l = (ar >> 13) & 1;
3309 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3311 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3312 dt->address = vmcs_readl(GUEST_IDTR_BASE);
3315 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3317 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3318 vmcs_writel(GUEST_IDTR_BASE, dt->address);
3321 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3323 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3324 dt->address = vmcs_readl(GUEST_GDTR_BASE);
3327 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3329 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3330 vmcs_writel(GUEST_GDTR_BASE, dt->address);
3333 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3335 struct kvm_segment var;
3338 vmx_get_segment(vcpu, &var, seg);
3340 if (seg == VCPU_SREG_CS)
3342 ar = vmx_segment_access_rights(&var);
3344 if (var.base != (var.selector << 4))
3346 if (var.limit != 0xffff)
3354 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3356 struct kvm_segment cs;
3357 unsigned int cs_rpl;
3359 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3360 cs_rpl = cs.selector & SEGMENT_RPL_MASK;
3364 if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
3368 if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
3369 if (cs.dpl > cs_rpl)
3372 if (cs.dpl != cs_rpl)
3378 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3382 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3384 struct kvm_segment ss;
3385 unsigned int ss_rpl;
3387 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3388 ss_rpl = ss.selector & SEGMENT_RPL_MASK;
3392 if (ss.type != 3 && ss.type != 7)
3396 if (ss.dpl != ss_rpl) /* DPL != RPL */
3404 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3406 struct kvm_segment var;
3409 vmx_get_segment(vcpu, &var, seg);
3410 rpl = var.selector & SEGMENT_RPL_MASK;
3418 if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
3419 if (var.dpl < rpl) /* DPL < RPL */
3423 /* TODO: Add other members to kvm_segment_field to allow checking for other access
3429 static bool tr_valid(struct kvm_vcpu *vcpu)
3431 struct kvm_segment tr;
3433 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3437 if (tr.selector & SEGMENT_TI_MASK) /* TI = 1 */
3439 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3447 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3449 struct kvm_segment ldtr;
3451 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3455 if (ldtr.selector & SEGMENT_TI_MASK) /* TI = 1 */
3465 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3467 struct kvm_segment cs, ss;
3469 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3470 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3472 return ((cs.selector & SEGMENT_RPL_MASK) ==
3473 (ss.selector & SEGMENT_RPL_MASK));
3477 * Check if guest state is valid. Returns true if valid, false if
3479 * We assume that registers are always usable
3481 static bool guest_state_valid(struct kvm_vcpu *vcpu)
3483 if (enable_unrestricted_guest)
3486 /* real mode guest state checks */
3487 if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
3488 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3490 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3492 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3494 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3496 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3498 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3501 /* protected mode guest state checks */
3502 if (!cs_ss_rpl_check(vcpu))
3504 if (!code_segment_valid(vcpu))
3506 if (!stack_segment_valid(vcpu))
3508 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3510 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3512 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3514 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3516 if (!tr_valid(vcpu))
3518 if (!ldtr_valid(vcpu))
3522 * - Add checks on RIP
3523 * - Add checks on RFLAGS
3529 static int init_rmode_tss(struct kvm *kvm)
3535 idx = srcu_read_lock(&kvm->srcu);
3536 fn = to_kvm_vmx(kvm)->tss_addr >> PAGE_SHIFT;
3537 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3540 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3541 r = kvm_write_guest_page(kvm, fn++, &data,
3542 TSS_IOPB_BASE_OFFSET, sizeof(u16));
3545 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3548 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3552 r = kvm_write_guest_page(kvm, fn, &data,
3553 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3556 srcu_read_unlock(&kvm->srcu, idx);
3560 static int init_rmode_identity_map(struct kvm *kvm)
3562 struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm);
3564 kvm_pfn_t identity_map_pfn;
3567 /* Protect kvm_vmx->ept_identity_pagetable_done. */
3568 mutex_lock(&kvm->slots_lock);
3570 if (likely(kvm_vmx->ept_identity_pagetable_done))
3573 if (!kvm_vmx->ept_identity_map_addr)
3574 kvm_vmx->ept_identity_map_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
3575 identity_map_pfn = kvm_vmx->ept_identity_map_addr >> PAGE_SHIFT;
3577 r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
3578 kvm_vmx->ept_identity_map_addr, PAGE_SIZE);
3582 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3585 /* Set up identity-mapping pagetable for EPT in real mode */
3586 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3587 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3588 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3589 r = kvm_write_guest_page(kvm, identity_map_pfn,
3590 &tmp, i * sizeof(tmp), sizeof(tmp));
3594 kvm_vmx->ept_identity_pagetable_done = true;
3597 mutex_unlock(&kvm->slots_lock);
3601 static void seg_setup(int seg)
3603 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3606 vmcs_write16(sf->selector, 0);
3607 vmcs_writel(sf->base, 0);
3608 vmcs_write32(sf->limit, 0xffff);
3610 if (seg == VCPU_SREG_CS)
3611 ar |= 0x08; /* code segment */
3613 vmcs_write32(sf->ar_bytes, ar);
3616 static int alloc_apic_access_page(struct kvm *kvm)
3621 mutex_lock(&kvm->slots_lock);
3622 if (kvm->arch.apic_access_page_done)
3624 r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
3625 APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
3629 page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
3630 if (is_error_page(page)) {
3636 * Do not pin the page in memory, so that memory hot-unplug
3637 * is able to migrate it.
3640 kvm->arch.apic_access_page_done = true;
3642 mutex_unlock(&kvm->slots_lock);
3646 int allocate_vpid(void)
3652 spin_lock(&vmx_vpid_lock);
3653 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3654 if (vpid < VMX_NR_VPIDS)
3655 __set_bit(vpid, vmx_vpid_bitmap);
3658 spin_unlock(&vmx_vpid_lock);
3662 void free_vpid(int vpid)
3664 if (!enable_vpid || vpid == 0)
3666 spin_lock(&vmx_vpid_lock);
3667 __clear_bit(vpid, vmx_vpid_bitmap);
3668 spin_unlock(&vmx_vpid_lock);
3671 static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
3674 int f = sizeof(unsigned long);
3676 if (!cpu_has_vmx_msr_bitmap())
3679 if (static_branch_unlikely(&enable_evmcs))
3680 evmcs_touch_msr_bitmap();
3683 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3684 * have the write-low and read-high bitmap offsets the wrong way round.
3685 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3687 if (msr <= 0x1fff) {
3688 if (type & MSR_TYPE_R)
3690 __clear_bit(msr, msr_bitmap + 0x000 / f);
3692 if (type & MSR_TYPE_W)
3694 __clear_bit(msr, msr_bitmap + 0x800 / f);
3696 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3698 if (type & MSR_TYPE_R)
3700 __clear_bit(msr, msr_bitmap + 0x400 / f);
3702 if (type & MSR_TYPE_W)
3704 __clear_bit(msr, msr_bitmap + 0xc00 / f);
3709 static __always_inline void vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
3712 int f = sizeof(unsigned long);
3714 if (!cpu_has_vmx_msr_bitmap())
3717 if (static_branch_unlikely(&enable_evmcs))
3718 evmcs_touch_msr_bitmap();
3721 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3722 * have the write-low and read-high bitmap offsets the wrong way round.
3723 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3725 if (msr <= 0x1fff) {
3726 if (type & MSR_TYPE_R)
3728 __set_bit(msr, msr_bitmap + 0x000 / f);
3730 if (type & MSR_TYPE_W)
3732 __set_bit(msr, msr_bitmap + 0x800 / f);
3734 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3736 if (type & MSR_TYPE_R)
3738 __set_bit(msr, msr_bitmap + 0x400 / f);
3740 if (type & MSR_TYPE_W)
3742 __set_bit(msr, msr_bitmap + 0xc00 / f);
3747 static __always_inline void vmx_set_intercept_for_msr(unsigned long *msr_bitmap,
3748 u32 msr, int type, bool value)
3751 vmx_enable_intercept_for_msr(msr_bitmap, msr, type);
3753 vmx_disable_intercept_for_msr(msr_bitmap, msr, type);
3756 static u8 vmx_msr_bitmap_mode(struct kvm_vcpu *vcpu)
3760 if (cpu_has_secondary_exec_ctrls() &&
3761 (secondary_exec_controls_get(to_vmx(vcpu)) &
3762 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
3763 mode |= MSR_BITMAP_MODE_X2APIC;
3764 if (enable_apicv && kvm_vcpu_apicv_active(vcpu))
3765 mode |= MSR_BITMAP_MODE_X2APIC_APICV;
3771 static void vmx_update_msr_bitmap_x2apic(unsigned long *msr_bitmap,
3776 for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
3777 unsigned word = msr / BITS_PER_LONG;
3778 msr_bitmap[word] = (mode & MSR_BITMAP_MODE_X2APIC_APICV) ? 0 : ~0;
3779 msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
3782 if (mode & MSR_BITMAP_MODE_X2APIC) {
3784 * TPR reads and writes can be virtualized even if virtual interrupt
3785 * delivery is not in use.
3787 vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TASKPRI), MSR_TYPE_RW);
3788 if (mode & MSR_BITMAP_MODE_X2APIC_APICV) {
3789 vmx_enable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TMCCT), MSR_TYPE_R);
3790 vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_EOI), MSR_TYPE_W);
3791 vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_SELF_IPI), MSR_TYPE_W);
3796 void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu)
3798 struct vcpu_vmx *vmx = to_vmx(vcpu);
3799 unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3800 u8 mode = vmx_msr_bitmap_mode(vcpu);
3801 u8 changed = mode ^ vmx->msr_bitmap_mode;
3806 if (changed & (MSR_BITMAP_MODE_X2APIC | MSR_BITMAP_MODE_X2APIC_APICV))
3807 vmx_update_msr_bitmap_x2apic(msr_bitmap, mode);
3809 vmx->msr_bitmap_mode = mode;
3812 void pt_update_intercept_for_msr(struct vcpu_vmx *vmx)
3814 unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3815 bool flag = !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
3818 vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_STATUS,
3820 vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_BASE,
3822 vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_MASK,
3824 vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_CR3_MATCH,
3826 for (i = 0; i < vmx->pt_desc.addr_range; i++) {
3827 vmx_set_intercept_for_msr(msr_bitmap,
3828 MSR_IA32_RTIT_ADDR0_A + i * 2, MSR_TYPE_RW, flag);
3829 vmx_set_intercept_for_msr(msr_bitmap,
3830 MSR_IA32_RTIT_ADDR0_B + i * 2, MSR_TYPE_RW, flag);
3834 static bool vmx_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
3836 struct vcpu_vmx *vmx = to_vmx(vcpu);
3841 if (WARN_ON_ONCE(!is_guest_mode(vcpu)) ||
3842 !nested_cpu_has_vid(get_vmcs12(vcpu)) ||
3843 WARN_ON_ONCE(!vmx->nested.virtual_apic_map.gfn))
3846 rvi = vmx_get_rvi();
3848 vapic_page = vmx->nested.virtual_apic_map.hva;
3849 vppr = *((u32 *)(vapic_page + APIC_PROCPRI));
3851 return ((rvi & 0xf0) > (vppr & 0xf0));
3854 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu,
3858 int pi_vec = nested ? POSTED_INTR_NESTED_VECTOR : POSTED_INTR_VECTOR;
3860 if (vcpu->mode == IN_GUEST_MODE) {
3862 * The vector of interrupt to be delivered to vcpu had
3863 * been set in PIR before this function.
3865 * Following cases will be reached in this block, and
3866 * we always send a notification event in all cases as
3869 * Case 1: vcpu keeps in non-root mode. Sending a
3870 * notification event posts the interrupt to vcpu.
3872 * Case 2: vcpu exits to root mode and is still
3873 * runnable. PIR will be synced to vIRR before the
3874 * next vcpu entry. Sending a notification event in
3875 * this case has no effect, as vcpu is not in root
3878 * Case 3: vcpu exits to root mode and is blocked.
3879 * vcpu_block() has already synced PIR to vIRR and
3880 * never blocks vcpu if vIRR is not cleared. Therefore,
3881 * a blocked vcpu here does not wait for any requested
3882 * interrupts in PIR, and sending a notification event
3883 * which has no effect is safe here.
3886 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu), pi_vec);
3893 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
3896 struct vcpu_vmx *vmx = to_vmx(vcpu);
3898 if (is_guest_mode(vcpu) &&
3899 vector == vmx->nested.posted_intr_nv) {
3901 * If a posted intr is not recognized by hardware,
3902 * we will accomplish it in the next vmentry.
3904 vmx->nested.pi_pending = true;
3905 kvm_make_request(KVM_REQ_EVENT, vcpu);
3906 /* the PIR and ON have been set by L1. */
3907 if (!kvm_vcpu_trigger_posted_interrupt(vcpu, true))
3908 kvm_vcpu_kick(vcpu);
3914 * Send interrupt to vcpu via posted interrupt way.
3915 * 1. If target vcpu is running(non-root mode), send posted interrupt
3916 * notification to vcpu and hardware will sync PIR to vIRR atomically.
3917 * 2. If target vcpu isn't running(root mode), kick it to pick up the
3918 * interrupt from PIR in next vmentry.
3920 static int vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
3922 struct vcpu_vmx *vmx = to_vmx(vcpu);
3925 r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
3929 if (!vcpu->arch.apicv_active)
3932 if (pi_test_and_set_pir(vector, &vmx->pi_desc))
3935 /* If a previous notification has sent the IPI, nothing to do. */
3936 if (pi_test_and_set_on(&vmx->pi_desc))
3939 if (vcpu != kvm_get_running_vcpu() &&
3940 !kvm_vcpu_trigger_posted_interrupt(vcpu, false))
3941 kvm_vcpu_kick(vcpu);
3947 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
3948 * will not change in the lifetime of the guest.
3949 * Note that host-state that does change is set elsewhere. E.g., host-state
3950 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
3952 void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
3956 unsigned long cr0, cr3, cr4;
3959 WARN_ON(cr0 & X86_CR0_TS);
3960 vmcs_writel(HOST_CR0, cr0); /* 22.2.3 */
3963 * Save the most likely value for this task's CR3 in the VMCS.
3964 * We can't use __get_current_cr3_fast() because we're not atomic.
3967 vmcs_writel(HOST_CR3, cr3); /* 22.2.3 FIXME: shadow tables */
3968 vmx->loaded_vmcs->host_state.cr3 = cr3;
3970 /* Save the most likely value for this task's CR4 in the VMCS. */
3971 cr4 = cr4_read_shadow();
3972 vmcs_writel(HOST_CR4, cr4); /* 22.2.3, 22.2.5 */
3973 vmx->loaded_vmcs->host_state.cr4 = cr4;
3975 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
3976 #ifdef CONFIG_X86_64
3978 * Load null selectors, so we can avoid reloading them in
3979 * vmx_prepare_switch_to_host(), in case userspace uses
3980 * the null selectors too (the expected case).
3982 vmcs_write16(HOST_DS_SELECTOR, 0);
3983 vmcs_write16(HOST_ES_SELECTOR, 0);
3985 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
3986 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
3988 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
3989 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
3991 vmcs_writel(HOST_IDTR_BASE, host_idt_base); /* 22.2.4 */
3993 vmcs_writel(HOST_RIP, (unsigned long)vmx_vmexit); /* 22.2.5 */
3995 rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
3996 vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
3997 rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
3998 vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */
4000 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
4001 rdmsr(MSR_IA32_CR_PAT, low32, high32);
4002 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
4005 if (cpu_has_load_ia32_efer())
4006 vmcs_write64(HOST_IA32_EFER, host_efer);
4009 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
4011 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
4013 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
4014 if (is_guest_mode(&vmx->vcpu))
4015 vmx->vcpu.arch.cr4_guest_owned_bits &=
4016 ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
4017 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
4020 u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
4022 u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
4024 if (!kvm_vcpu_apicv_active(&vmx->vcpu))
4025 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
4028 pin_based_exec_ctrl &= ~PIN_BASED_VIRTUAL_NMIS;
4030 if (!enable_preemption_timer)
4031 pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
4033 return pin_based_exec_ctrl;
4036 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
4038 struct vcpu_vmx *vmx = to_vmx(vcpu);
4040 pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
4041 if (cpu_has_secondary_exec_ctrls()) {
4042 if (kvm_vcpu_apicv_active(vcpu))
4043 secondary_exec_controls_setbit(vmx,
4044 SECONDARY_EXEC_APIC_REGISTER_VIRT |
4045 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4047 secondary_exec_controls_clearbit(vmx,
4048 SECONDARY_EXEC_APIC_REGISTER_VIRT |
4049 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4052 if (cpu_has_vmx_msr_bitmap())
4053 vmx_update_msr_bitmap(vcpu);
4056 u32 vmx_exec_control(struct vcpu_vmx *vmx)
4058 u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
4060 if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
4061 exec_control &= ~CPU_BASED_MOV_DR_EXITING;
4063 if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
4064 exec_control &= ~CPU_BASED_TPR_SHADOW;
4065 #ifdef CONFIG_X86_64
4066 exec_control |= CPU_BASED_CR8_STORE_EXITING |
4067 CPU_BASED_CR8_LOAD_EXITING;
4071 exec_control |= CPU_BASED_CR3_STORE_EXITING |
4072 CPU_BASED_CR3_LOAD_EXITING |
4073 CPU_BASED_INVLPG_EXITING;
4074 if (kvm_mwait_in_guest(vmx->vcpu.kvm))
4075 exec_control &= ~(CPU_BASED_MWAIT_EXITING |
4076 CPU_BASED_MONITOR_EXITING);
4077 if (kvm_hlt_in_guest(vmx->vcpu.kvm))
4078 exec_control &= ~CPU_BASED_HLT_EXITING;
4079 return exec_control;
4083 static void vmx_compute_secondary_exec_control(struct vcpu_vmx *vmx)
4085 struct kvm_vcpu *vcpu = &vmx->vcpu;
4087 u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
4089 if (vmx_pt_mode_is_system())
4090 exec_control &= ~(SECONDARY_EXEC_PT_USE_GPA | SECONDARY_EXEC_PT_CONCEAL_VMX);
4091 if (!cpu_need_virtualize_apic_accesses(vcpu))
4092 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
4094 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
4096 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
4097 enable_unrestricted_guest = 0;
4099 if (!enable_unrestricted_guest)
4100 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4101 if (kvm_pause_in_guest(vmx->vcpu.kvm))
4102 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
4103 if (!kvm_vcpu_apicv_active(vcpu))
4104 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
4105 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4106 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
4108 /* SECONDARY_EXEC_DESC is enabled/disabled on writes to CR4.UMIP,
4109 * in vmx_set_cr4. */
4110 exec_control &= ~SECONDARY_EXEC_DESC;
4112 /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4114 We can NOT enable shadow_vmcs here because we don't have yet
4117 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4120 exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
4122 if (vmx_xsaves_supported()) {
4123 /* Exposing XSAVES only when XSAVE is exposed */
4124 bool xsaves_enabled =
4125 boot_cpu_has(X86_FEATURE_XSAVE) &&
4126 guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
4127 guest_cpuid_has(vcpu, X86_FEATURE_XSAVES);
4129 vcpu->arch.xsaves_enabled = xsaves_enabled;
4131 if (!xsaves_enabled)
4132 exec_control &= ~SECONDARY_EXEC_XSAVES;
4136 vmx->nested.msrs.secondary_ctls_high |=
4137 SECONDARY_EXEC_XSAVES;
4139 vmx->nested.msrs.secondary_ctls_high &=
4140 ~SECONDARY_EXEC_XSAVES;
4144 if (cpu_has_vmx_rdtscp()) {
4145 bool rdtscp_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP);
4146 if (!rdtscp_enabled)
4147 exec_control &= ~SECONDARY_EXEC_RDTSCP;
4151 vmx->nested.msrs.secondary_ctls_high |=
4152 SECONDARY_EXEC_RDTSCP;
4154 vmx->nested.msrs.secondary_ctls_high &=
4155 ~SECONDARY_EXEC_RDTSCP;
4159 if (cpu_has_vmx_invpcid()) {
4160 /* Exposing INVPCID only when PCID is exposed */
4161 bool invpcid_enabled =
4162 guest_cpuid_has(vcpu, X86_FEATURE_INVPCID) &&
4163 guest_cpuid_has(vcpu, X86_FEATURE_PCID);
4165 if (!invpcid_enabled) {
4166 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
4167 guest_cpuid_clear(vcpu, X86_FEATURE_INVPCID);
4171 if (invpcid_enabled)
4172 vmx->nested.msrs.secondary_ctls_high |=
4173 SECONDARY_EXEC_ENABLE_INVPCID;
4175 vmx->nested.msrs.secondary_ctls_high &=
4176 ~SECONDARY_EXEC_ENABLE_INVPCID;
4180 if (vmx_rdrand_supported()) {
4181 bool rdrand_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDRAND);
4183 exec_control &= ~SECONDARY_EXEC_RDRAND_EXITING;
4187 vmx->nested.msrs.secondary_ctls_high |=
4188 SECONDARY_EXEC_RDRAND_EXITING;
4190 vmx->nested.msrs.secondary_ctls_high &=
4191 ~SECONDARY_EXEC_RDRAND_EXITING;
4195 if (vmx_rdseed_supported()) {
4196 bool rdseed_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDSEED);
4198 exec_control &= ~SECONDARY_EXEC_RDSEED_EXITING;
4202 vmx->nested.msrs.secondary_ctls_high |=
4203 SECONDARY_EXEC_RDSEED_EXITING;
4205 vmx->nested.msrs.secondary_ctls_high &=
4206 ~SECONDARY_EXEC_RDSEED_EXITING;
4210 if (vmx_waitpkg_supported()) {
4211 bool waitpkg_enabled =
4212 guest_cpuid_has(vcpu, X86_FEATURE_WAITPKG);
4214 if (!waitpkg_enabled)
4215 exec_control &= ~SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4218 if (waitpkg_enabled)
4219 vmx->nested.msrs.secondary_ctls_high |=
4220 SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4222 vmx->nested.msrs.secondary_ctls_high &=
4223 ~SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4227 vmx->secondary_exec_control = exec_control;
4230 static void ept_set_mmio_spte_mask(void)
4233 * EPT Misconfigurations can be generated if the value of bits 2:0
4234 * of an EPT paging-structure entry is 110b (write/execute).
4236 kvm_mmu_set_mmio_spte_mask(VMX_EPT_RWX_MASK,
4237 VMX_EPT_MISCONFIG_WX_VALUE, 0);
4240 #define VMX_XSS_EXIT_BITMAP 0
4243 * Noting that the initialization of Guest-state Area of VMCS is in
4246 static void init_vmcs(struct vcpu_vmx *vmx)
4249 nested_vmx_set_vmcs_shadowing_bitmap();
4251 if (cpu_has_vmx_msr_bitmap())
4252 vmcs_write64(MSR_BITMAP, __pa(vmx->vmcs01.msr_bitmap));
4254 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4257 pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
4259 exec_controls_set(vmx, vmx_exec_control(vmx));
4261 if (cpu_has_secondary_exec_ctrls()) {
4262 vmx_compute_secondary_exec_control(vmx);
4263 secondary_exec_controls_set(vmx, vmx->secondary_exec_control);
4266 if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
4267 vmcs_write64(EOI_EXIT_BITMAP0, 0);
4268 vmcs_write64(EOI_EXIT_BITMAP1, 0);
4269 vmcs_write64(EOI_EXIT_BITMAP2, 0);
4270 vmcs_write64(EOI_EXIT_BITMAP3, 0);
4272 vmcs_write16(GUEST_INTR_STATUS, 0);
4274 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4275 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4278 if (!kvm_pause_in_guest(vmx->vcpu.kvm)) {
4279 vmcs_write32(PLE_GAP, ple_gap);
4280 vmx->ple_window = ple_window;
4281 vmx->ple_window_dirty = true;
4284 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4285 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4286 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
4288 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
4289 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
4290 vmx_set_constant_host_state(vmx);
4291 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4292 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4294 if (cpu_has_vmx_vmfunc())
4295 vmcs_write64(VM_FUNCTION_CONTROL, 0);
4297 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4298 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4299 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
4300 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4301 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
4303 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
4304 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
4306 vm_exit_controls_set(vmx, vmx_vmexit_ctrl());
4308 /* 22.2.1, 20.8.1 */
4309 vm_entry_controls_set(vmx, vmx_vmentry_ctrl());
4311 vmx->vcpu.arch.cr0_guest_owned_bits = X86_CR0_TS;
4312 vmcs_writel(CR0_GUEST_HOST_MASK, ~X86_CR0_TS);
4314 set_cr4_guest_host_mask(vmx);
4317 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4319 if (vmx_xsaves_supported())
4320 vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
4323 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
4324 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
4327 if (cpu_has_vmx_encls_vmexit())
4328 vmcs_write64(ENCLS_EXITING_BITMAP, -1ull);
4330 if (vmx_pt_mode_is_host_guest()) {
4331 memset(&vmx->pt_desc, 0, sizeof(vmx->pt_desc));
4332 /* Bit[6~0] are forced to 1, writes are ignored. */
4333 vmx->pt_desc.guest.output_mask = 0x7F;
4334 vmcs_write64(GUEST_IA32_RTIT_CTL, 0);
4338 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
4340 struct vcpu_vmx *vmx = to_vmx(vcpu);
4341 struct msr_data apic_base_msr;
4344 vmx->rmode.vm86_active = 0;
4347 vmx->msr_ia32_umwait_control = 0;
4349 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
4350 vmx->hv_deadline_tsc = -1;
4351 kvm_set_cr8(vcpu, 0);
4354 apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
4355 MSR_IA32_APICBASE_ENABLE;
4356 if (kvm_vcpu_is_reset_bsp(vcpu))
4357 apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
4358 apic_base_msr.host_initiated = true;
4359 kvm_set_apic_base(vcpu, &apic_base_msr);
4362 vmx_segment_cache_clear(vmx);
4364 seg_setup(VCPU_SREG_CS);
4365 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4366 vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
4368 seg_setup(VCPU_SREG_DS);
4369 seg_setup(VCPU_SREG_ES);
4370 seg_setup(VCPU_SREG_FS);
4371 seg_setup(VCPU_SREG_GS);
4372 seg_setup(VCPU_SREG_SS);
4374 vmcs_write16(GUEST_TR_SELECTOR, 0);
4375 vmcs_writel(GUEST_TR_BASE, 0);
4376 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4377 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4379 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4380 vmcs_writel(GUEST_LDTR_BASE, 0);
4381 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4382 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4385 vmcs_write32(GUEST_SYSENTER_CS, 0);
4386 vmcs_writel(GUEST_SYSENTER_ESP, 0);
4387 vmcs_writel(GUEST_SYSENTER_EIP, 0);
4388 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4391 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
4392 kvm_rip_write(vcpu, 0xfff0);
4394 vmcs_writel(GUEST_GDTR_BASE, 0);
4395 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4397 vmcs_writel(GUEST_IDTR_BASE, 0);
4398 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4400 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4401 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4402 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4403 if (kvm_mpx_supported())
4404 vmcs_write64(GUEST_BNDCFGS, 0);
4408 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
4410 if (cpu_has_vmx_tpr_shadow() && !init_event) {
4411 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4412 if (cpu_need_tpr_shadow(vcpu))
4413 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4414 __pa(vcpu->arch.apic->regs));
4415 vmcs_write32(TPR_THRESHOLD, 0);
4418 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4420 cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4421 vmx->vcpu.arch.cr0 = cr0;
4422 vmx_set_cr0(vcpu, cr0); /* enter rmode */
4423 vmx_set_cr4(vcpu, 0);
4424 vmx_set_efer(vcpu, 0);
4426 update_exception_bitmap(vcpu);
4428 vpid_sync_context(vmx->vpid);
4430 vmx_clear_hlt(vcpu);
4433 static void enable_irq_window(struct kvm_vcpu *vcpu)
4435 exec_controls_setbit(to_vmx(vcpu), CPU_BASED_INTR_WINDOW_EXITING);
4438 static void enable_nmi_window(struct kvm_vcpu *vcpu)
4441 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4442 enable_irq_window(vcpu);
4446 exec_controls_setbit(to_vmx(vcpu), CPU_BASED_NMI_WINDOW_EXITING);
4449 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4451 struct vcpu_vmx *vmx = to_vmx(vcpu);
4453 int irq = vcpu->arch.interrupt.nr;
4455 trace_kvm_inj_virq(irq);
4457 ++vcpu->stat.irq_injections;
4458 if (vmx->rmode.vm86_active) {
4460 if (vcpu->arch.interrupt.soft)
4461 inc_eip = vcpu->arch.event_exit_inst_len;
4462 kvm_inject_realmode_interrupt(vcpu, irq, inc_eip);
4465 intr = irq | INTR_INFO_VALID_MASK;
4466 if (vcpu->arch.interrupt.soft) {
4467 intr |= INTR_TYPE_SOFT_INTR;
4468 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4469 vmx->vcpu.arch.event_exit_inst_len);
4471 intr |= INTR_TYPE_EXT_INTR;
4472 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4474 vmx_clear_hlt(vcpu);
4477 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4479 struct vcpu_vmx *vmx = to_vmx(vcpu);
4483 * Tracking the NMI-blocked state in software is built upon
4484 * finding the next open IRQ window. This, in turn, depends on
4485 * well-behaving guests: They have to keep IRQs disabled at
4486 * least as long as the NMI handler runs. Otherwise we may
4487 * cause NMI nesting, maybe breaking the guest. But as this is
4488 * highly unlikely, we can live with the residual risk.
4490 vmx->loaded_vmcs->soft_vnmi_blocked = 1;
4491 vmx->loaded_vmcs->vnmi_blocked_time = 0;
4494 ++vcpu->stat.nmi_injections;
4495 vmx->loaded_vmcs->nmi_known_unmasked = false;
4497 if (vmx->rmode.vm86_active) {
4498 kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0);
4502 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4503 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4505 vmx_clear_hlt(vcpu);
4508 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4510 struct vcpu_vmx *vmx = to_vmx(vcpu);
4514 return vmx->loaded_vmcs->soft_vnmi_blocked;
4515 if (vmx->loaded_vmcs->nmi_known_unmasked)
4517 masked = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4518 vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4522 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4524 struct vcpu_vmx *vmx = to_vmx(vcpu);
4527 if (vmx->loaded_vmcs->soft_vnmi_blocked != masked) {
4528 vmx->loaded_vmcs->soft_vnmi_blocked = masked;
4529 vmx->loaded_vmcs->vnmi_blocked_time = 0;
4532 vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4534 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4535 GUEST_INTR_STATE_NMI);
4537 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4538 GUEST_INTR_STATE_NMI);
4542 bool vmx_nmi_blocked(struct kvm_vcpu *vcpu)
4544 if (is_guest_mode(vcpu) && nested_exit_on_nmi(vcpu))
4547 if (!enable_vnmi && to_vmx(vcpu)->loaded_vmcs->soft_vnmi_blocked)
4550 return (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4551 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI |
4552 GUEST_INTR_STATE_NMI));
4555 static bool vmx_nmi_allowed(struct kvm_vcpu *vcpu, bool for_injection)
4557 if (to_vmx(vcpu)->nested.nested_run_pending)
4560 /* An NMI must not be injected into L2 if it's supposed to VM-Exit. */
4561 if (for_injection && is_guest_mode(vcpu) && nested_exit_on_nmi(vcpu))
4564 return !vmx_nmi_blocked(vcpu);
4567 bool vmx_interrupt_blocked(struct kvm_vcpu *vcpu)
4569 if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu))
4572 return !(vmx_get_rflags(vcpu) & X86_EFLAGS_IF) ||
4573 (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4574 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4577 static bool vmx_interrupt_allowed(struct kvm_vcpu *vcpu, bool for_injection)
4579 if (to_vmx(vcpu)->nested.nested_run_pending)
4583 * An IRQ must not be injected into L2 if it's supposed to VM-Exit,
4584 * e.g. if the IRQ arrived asynchronously after checking nested events.
4586 if (for_injection && is_guest_mode(vcpu) && nested_exit_on_intr(vcpu))
4589 return !vmx_interrupt_blocked(vcpu);
4592 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4596 if (enable_unrestricted_guest)
4599 mutex_lock(&kvm->slots_lock);
4600 ret = __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
4602 mutex_unlock(&kvm->slots_lock);
4606 to_kvm_vmx(kvm)->tss_addr = addr;
4607 return init_rmode_tss(kvm);
4610 static int vmx_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
4612 to_kvm_vmx(kvm)->ept_identity_map_addr = ident_addr;
4616 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
4621 * Update instruction length as we may reinject the exception
4622 * from user space while in guest debugging mode.
4624 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4625 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4626 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4630 if (vcpu->guest_debug &
4631 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
4647 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4648 int vec, u32 err_code)
4651 * Instruction with address size override prefix opcode 0x67
4652 * Cause the #SS fault with 0 error code in VM86 mode.
4654 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
4655 if (kvm_emulate_instruction(vcpu, 0)) {
4656 if (vcpu->arch.halt_request) {
4657 vcpu->arch.halt_request = 0;
4658 return kvm_vcpu_halt(vcpu);
4666 * Forward all other exceptions that are valid in real mode.
4667 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4668 * the required debugging infrastructure rework.
4670 kvm_queue_exception(vcpu, vec);
4675 * Trigger machine check on the host. We assume all the MSRs are already set up
4676 * by the CPU and that we still run on the same CPU as the MCE occurred on.
4677 * We pass a fake environment to the machine check handler because we want
4678 * the guest to be always treated like user space, no matter what context
4679 * it used internally.
4681 static void kvm_machine_check(void)
4683 #if defined(CONFIG_X86_MCE)
4684 struct pt_regs regs = {
4685 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
4686 .flags = X86_EFLAGS_IF,
4689 do_machine_check(®s, 0);
4693 static int handle_machine_check(struct kvm_vcpu *vcpu)
4695 /* handled by vmx_vcpu_run() */
4700 * If the host has split lock detection disabled, then #AC is
4701 * unconditionally injected into the guest, which is the pre split lock
4702 * detection behaviour.
4704 * If the host has split lock detection enabled then #AC is
4705 * only injected into the guest when:
4706 * - Guest CPL == 3 (user mode)
4707 * - Guest has #AC detection enabled in CR0
4708 * - Guest EFLAGS has AC bit set
4710 static inline bool guest_inject_ac(struct kvm_vcpu *vcpu)
4712 if (!boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
4715 return vmx_get_cpl(vcpu) == 3 && kvm_read_cr0_bits(vcpu, X86_CR0_AM) &&
4716 (kvm_get_rflags(vcpu) & X86_EFLAGS_AC);
4719 static int handle_exception_nmi(struct kvm_vcpu *vcpu)
4721 struct vcpu_vmx *vmx = to_vmx(vcpu);
4722 struct kvm_run *kvm_run = vcpu->run;
4723 u32 intr_info, ex_no, error_code;
4724 unsigned long cr2, rip, dr6;
4727 vect_info = vmx->idt_vectoring_info;
4728 intr_info = vmx_get_intr_info(vcpu);
4730 if (is_machine_check(intr_info) || is_nmi(intr_info))
4731 return 1; /* handled by handle_exception_nmi_irqoff() */
4733 if (is_invalid_opcode(intr_info))
4734 return handle_ud(vcpu);
4737 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4738 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4740 if (!vmx->rmode.vm86_active && is_gp_fault(intr_info)) {
4741 WARN_ON_ONCE(!enable_vmware_backdoor);
4744 * VMware backdoor emulation on #GP interception only handles
4745 * IN{S}, OUT{S}, and RDPMC, none of which generate a non-zero
4746 * error code on #GP.
4749 kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
4752 return kvm_emulate_instruction(vcpu, EMULTYPE_VMWARE_GP);
4756 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
4757 * MMIO, it is better to report an internal error.
4758 * See the comments in vmx_handle_exit.
4760 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4761 !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
4762 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4763 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4764 vcpu->run->internal.ndata = 3;
4765 vcpu->run->internal.data[0] = vect_info;
4766 vcpu->run->internal.data[1] = intr_info;
4767 vcpu->run->internal.data[2] = error_code;
4771 if (is_page_fault(intr_info)) {
4772 cr2 = vmx_get_exit_qual(vcpu);
4773 /* EPT won't cause page fault directly */
4774 WARN_ON_ONCE(!vcpu->arch.apf.host_apf_reason && enable_ept);
4775 return kvm_handle_page_fault(vcpu, error_code, cr2, NULL, 0);
4778 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4780 if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
4781 return handle_rmode_exception(vcpu, ex_no, error_code);
4785 dr6 = vmx_get_exit_qual(vcpu);
4786 if (!(vcpu->guest_debug &
4787 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4788 if (is_icebp(intr_info))
4789 WARN_ON(!skip_emulated_instruction(vcpu));
4791 kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
4794 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
4795 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4799 * Update instruction length as we may reinject #BP from
4800 * user space while in guest debugging mode. Reading it for
4801 * #DB as well causes no harm, it is not used in that case.
4803 vmx->vcpu.arch.event_exit_inst_len =
4804 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4805 kvm_run->exit_reason = KVM_EXIT_DEBUG;
4806 rip = kvm_rip_read(vcpu);
4807 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4808 kvm_run->debug.arch.exception = ex_no;
4811 if (guest_inject_ac(vcpu)) {
4812 kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
4817 * Handle split lock. Depending on detection mode this will
4818 * either warn and disable split lock detection for this
4819 * task or force SIGBUS on it.
4821 if (handle_guest_split_lock(kvm_rip_read(vcpu)))
4825 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4826 kvm_run->ex.exception = ex_no;
4827 kvm_run->ex.error_code = error_code;
4833 static __always_inline int handle_external_interrupt(struct kvm_vcpu *vcpu)
4835 ++vcpu->stat.irq_exits;
4839 static int handle_triple_fault(struct kvm_vcpu *vcpu)
4841 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4842 vcpu->mmio_needed = 0;
4846 static int handle_io(struct kvm_vcpu *vcpu)
4848 unsigned long exit_qualification;
4849 int size, in, string;
4852 exit_qualification = vmx_get_exit_qual(vcpu);
4853 string = (exit_qualification & 16) != 0;
4855 ++vcpu->stat.io_exits;
4858 return kvm_emulate_instruction(vcpu, 0);
4860 port = exit_qualification >> 16;
4861 size = (exit_qualification & 7) + 1;
4862 in = (exit_qualification & 8) != 0;
4864 return kvm_fast_pio(vcpu, size, port, in);
4868 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4871 * Patch in the VMCALL instruction:
4873 hypercall[0] = 0x0f;
4874 hypercall[1] = 0x01;
4875 hypercall[2] = 0xc1;
4878 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
4879 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
4881 if (is_guest_mode(vcpu)) {
4882 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4883 unsigned long orig_val = val;
4886 * We get here when L2 changed cr0 in a way that did not change
4887 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
4888 * but did change L0 shadowed bits. So we first calculate the
4889 * effective cr0 value that L1 would like to write into the
4890 * hardware. It consists of the L2-owned bits from the new
4891 * value combined with the L1-owned bits from L1's guest_cr0.
4893 val = (val & ~vmcs12->cr0_guest_host_mask) |
4894 (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
4896 if (!nested_guest_cr0_valid(vcpu, val))
4899 if (kvm_set_cr0(vcpu, val))
4901 vmcs_writel(CR0_READ_SHADOW, orig_val);
4904 if (to_vmx(vcpu)->nested.vmxon &&
4905 !nested_host_cr0_valid(vcpu, val))
4908 return kvm_set_cr0(vcpu, val);
4912 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
4914 if (is_guest_mode(vcpu)) {
4915 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4916 unsigned long orig_val = val;
4918 /* analogously to handle_set_cr0 */
4919 val = (val & ~vmcs12->cr4_guest_host_mask) |
4920 (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
4921 if (kvm_set_cr4(vcpu, val))
4923 vmcs_writel(CR4_READ_SHADOW, orig_val);
4926 return kvm_set_cr4(vcpu, val);
4929 static int handle_desc(struct kvm_vcpu *vcpu)
4931 WARN_ON(!(vcpu->arch.cr4 & X86_CR4_UMIP));
4932 return kvm_emulate_instruction(vcpu, 0);
4935 static int handle_cr(struct kvm_vcpu *vcpu)
4937 unsigned long exit_qualification, val;
4943 exit_qualification = vmx_get_exit_qual(vcpu);
4944 cr = exit_qualification & 15;
4945 reg = (exit_qualification >> 8) & 15;
4946 switch ((exit_qualification >> 4) & 3) {
4947 case 0: /* mov to cr */
4948 val = kvm_register_readl(vcpu, reg);
4949 trace_kvm_cr_write(cr, val);
4952 err = handle_set_cr0(vcpu, val);
4953 return kvm_complete_insn_gp(vcpu, err);
4955 WARN_ON_ONCE(enable_unrestricted_guest);
4956 err = kvm_set_cr3(vcpu, val);
4957 return kvm_complete_insn_gp(vcpu, err);
4959 err = handle_set_cr4(vcpu, val);
4960 return kvm_complete_insn_gp(vcpu, err);
4962 u8 cr8_prev = kvm_get_cr8(vcpu);
4964 err = kvm_set_cr8(vcpu, cr8);
4965 ret = kvm_complete_insn_gp(vcpu, err);
4966 if (lapic_in_kernel(vcpu))
4968 if (cr8_prev <= cr8)
4971 * TODO: we might be squashing a
4972 * KVM_GUESTDBG_SINGLESTEP-triggered
4973 * KVM_EXIT_DEBUG here.
4975 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
4981 WARN_ONCE(1, "Guest should always own CR0.TS");
4982 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4983 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
4984 return kvm_skip_emulated_instruction(vcpu);
4985 case 1: /*mov from cr*/
4988 WARN_ON_ONCE(enable_unrestricted_guest);
4989 val = kvm_read_cr3(vcpu);
4990 kvm_register_write(vcpu, reg, val);
4991 trace_kvm_cr_read(cr, val);
4992 return kvm_skip_emulated_instruction(vcpu);
4994 val = kvm_get_cr8(vcpu);
4995 kvm_register_write(vcpu, reg, val);
4996 trace_kvm_cr_read(cr, val);
4997 return kvm_skip_emulated_instruction(vcpu);
5001 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5002 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
5003 kvm_lmsw(vcpu, val);
5005 return kvm_skip_emulated_instruction(vcpu);
5009 vcpu->run->exit_reason = 0;
5010 vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
5011 (int)(exit_qualification >> 4) & 3, cr);
5015 static int handle_dr(struct kvm_vcpu *vcpu)
5017 unsigned long exit_qualification;
5020 exit_qualification = vmx_get_exit_qual(vcpu);
5021 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
5023 /* First, if DR does not exist, trigger UD */
5024 if (!kvm_require_dr(vcpu, dr))
5027 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
5028 if (!kvm_require_cpl(vcpu, 0))
5030 dr7 = vmcs_readl(GUEST_DR7);
5033 * As the vm-exit takes precedence over the debug trap, we
5034 * need to emulate the latter, either for the host or the
5035 * guest debugging itself.
5037 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5038 vcpu->run->debug.arch.dr6 = DR6_BD | DR6_RTM | DR6_FIXED_1;
5039 vcpu->run->debug.arch.dr7 = dr7;
5040 vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
5041 vcpu->run->debug.arch.exception = DB_VECTOR;
5042 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
5045 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BD);
5050 if (vcpu->guest_debug == 0) {
5051 exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
5054 * No more DR vmexits; force a reload of the debug registers
5055 * and reenter on this instruction. The next vmexit will
5056 * retrieve the full state of the debug registers.
5058 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
5062 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
5063 if (exit_qualification & TYPE_MOV_FROM_DR) {
5066 if (kvm_get_dr(vcpu, dr, &val))
5068 kvm_register_write(vcpu, reg, val);
5070 if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
5073 return kvm_skip_emulated_instruction(vcpu);
5076 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
5078 get_debugreg(vcpu->arch.db[0], 0);
5079 get_debugreg(vcpu->arch.db[1], 1);
5080 get_debugreg(vcpu->arch.db[2], 2);
5081 get_debugreg(vcpu->arch.db[3], 3);
5082 get_debugreg(vcpu->arch.dr6, 6);
5083 vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
5085 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
5086 exec_controls_setbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
5089 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
5091 vmcs_writel(GUEST_DR7, val);
5094 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
5096 kvm_apic_update_ppr(vcpu);
5100 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
5102 exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_INTR_WINDOW_EXITING);
5104 kvm_make_request(KVM_REQ_EVENT, vcpu);
5106 ++vcpu->stat.irq_window_exits;
5110 static int handle_vmcall(struct kvm_vcpu *vcpu)
5112 return kvm_emulate_hypercall(vcpu);
5115 static int handle_invd(struct kvm_vcpu *vcpu)
5117 return kvm_emulate_instruction(vcpu, 0);
5120 static int handle_invlpg(struct kvm_vcpu *vcpu)
5122 unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5124 kvm_mmu_invlpg(vcpu, exit_qualification);
5125 return kvm_skip_emulated_instruction(vcpu);
5128 static int handle_rdpmc(struct kvm_vcpu *vcpu)
5132 err = kvm_rdpmc(vcpu);
5133 return kvm_complete_insn_gp(vcpu, err);
5136 static int handle_wbinvd(struct kvm_vcpu *vcpu)
5138 return kvm_emulate_wbinvd(vcpu);
5141 static int handle_xsetbv(struct kvm_vcpu *vcpu)
5143 u64 new_bv = kvm_read_edx_eax(vcpu);
5144 u32 index = kvm_rcx_read(vcpu);
5146 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
5147 return kvm_skip_emulated_instruction(vcpu);
5151 static int handle_apic_access(struct kvm_vcpu *vcpu)
5153 if (likely(fasteoi)) {
5154 unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5155 int access_type, offset;
5157 access_type = exit_qualification & APIC_ACCESS_TYPE;
5158 offset = exit_qualification & APIC_ACCESS_OFFSET;
5160 * Sane guest uses MOV to write EOI, with written value
5161 * not cared. So make a short-circuit here by avoiding
5162 * heavy instruction emulation.
5164 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5165 (offset == APIC_EOI)) {
5166 kvm_lapic_set_eoi(vcpu);
5167 return kvm_skip_emulated_instruction(vcpu);
5170 return kvm_emulate_instruction(vcpu, 0);
5173 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5175 unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5176 int vector = exit_qualification & 0xff;
5178 /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5179 kvm_apic_set_eoi_accelerated(vcpu, vector);
5183 static int handle_apic_write(struct kvm_vcpu *vcpu)
5185 unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5186 u32 offset = exit_qualification & 0xfff;
5188 /* APIC-write VM exit is trap-like and thus no need to adjust IP */
5189 kvm_apic_write_nodecode(vcpu, offset);
5193 static int handle_task_switch(struct kvm_vcpu *vcpu)
5195 struct vcpu_vmx *vmx = to_vmx(vcpu);
5196 unsigned long exit_qualification;
5197 bool has_error_code = false;
5200 int reason, type, idt_v, idt_index;
5202 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5203 idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5204 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5206 exit_qualification = vmx_get_exit_qual(vcpu);
5208 reason = (u32)exit_qualification >> 30;
5209 if (reason == TASK_SWITCH_GATE && idt_v) {
5211 case INTR_TYPE_NMI_INTR:
5212 vcpu->arch.nmi_injected = false;
5213 vmx_set_nmi_mask(vcpu, true);
5215 case INTR_TYPE_EXT_INTR:
5216 case INTR_TYPE_SOFT_INTR:
5217 kvm_clear_interrupt_queue(vcpu);
5219 case INTR_TYPE_HARD_EXCEPTION:
5220 if (vmx->idt_vectoring_info &
5221 VECTORING_INFO_DELIVER_CODE_MASK) {
5222 has_error_code = true;
5224 vmcs_read32(IDT_VECTORING_ERROR_CODE);
5227 case INTR_TYPE_SOFT_EXCEPTION:
5228 kvm_clear_exception_queue(vcpu);
5234 tss_selector = exit_qualification;
5236 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5237 type != INTR_TYPE_EXT_INTR &&
5238 type != INTR_TYPE_NMI_INTR))
5239 WARN_ON(!skip_emulated_instruction(vcpu));
5242 * TODO: What about debug traps on tss switch?
5243 * Are we supposed to inject them and update dr6?
5245 return kvm_task_switch(vcpu, tss_selector,
5246 type == INTR_TYPE_SOFT_INTR ? idt_index : -1,
5247 reason, has_error_code, error_code);
5250 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5252 unsigned long exit_qualification;
5256 exit_qualification = vmx_get_exit_qual(vcpu);
5259 * EPT violation happened while executing iret from NMI,
5260 * "blocked by NMI" bit has to be set before next VM entry.
5261 * There are errata that may cause this bit to not be set:
5264 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5266 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
5267 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5269 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5270 trace_kvm_page_fault(gpa, exit_qualification);
5272 /* Is it a read fault? */
5273 error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
5274 ? PFERR_USER_MASK : 0;
5275 /* Is it a write fault? */
5276 error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
5277 ? PFERR_WRITE_MASK : 0;
5278 /* Is it a fetch fault? */
5279 error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
5280 ? PFERR_FETCH_MASK : 0;
5281 /* ept page table entry is present? */
5282 error_code |= (exit_qualification &
5283 (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE |
5284 EPT_VIOLATION_EXECUTABLE))
5285 ? PFERR_PRESENT_MASK : 0;
5287 error_code |= (exit_qualification & 0x100) != 0 ?
5288 PFERR_GUEST_FINAL_MASK : PFERR_GUEST_PAGE_MASK;
5290 vcpu->arch.exit_qualification = exit_qualification;
5291 return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5294 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5299 * A nested guest cannot optimize MMIO vmexits, because we have an
5300 * nGPA here instead of the required GPA.
5302 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5303 if (!is_guest_mode(vcpu) &&
5304 !kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5305 trace_kvm_fast_mmio(gpa);
5306 return kvm_skip_emulated_instruction(vcpu);
5309 return kvm_mmu_page_fault(vcpu, gpa, PFERR_RSVD_MASK, NULL, 0);
5312 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5314 WARN_ON_ONCE(!enable_vnmi);
5315 exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_NMI_WINDOW_EXITING);
5316 ++vcpu->stat.nmi_window_exits;
5317 kvm_make_request(KVM_REQ_EVENT, vcpu);
5322 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5324 struct vcpu_vmx *vmx = to_vmx(vcpu);
5325 bool intr_window_requested;
5326 unsigned count = 130;
5328 intr_window_requested = exec_controls_get(vmx) &
5329 CPU_BASED_INTR_WINDOW_EXITING;
5331 while (vmx->emulation_required && count-- != 0) {
5332 if (intr_window_requested && !vmx_interrupt_blocked(vcpu))
5333 return handle_interrupt_window(&vmx->vcpu);
5335 if (kvm_test_request(KVM_REQ_EVENT, vcpu))
5338 if (!kvm_emulate_instruction(vcpu, 0))
5341 if (vmx->emulation_required && !vmx->rmode.vm86_active &&
5342 vcpu->arch.exception.pending) {
5343 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5344 vcpu->run->internal.suberror =
5345 KVM_INTERNAL_ERROR_EMULATION;
5346 vcpu->run->internal.ndata = 0;
5350 if (vcpu->arch.halt_request) {
5351 vcpu->arch.halt_request = 0;
5352 return kvm_vcpu_halt(vcpu);
5356 * Note, return 1 and not 0, vcpu_run() is responsible for
5357 * morphing the pending signal into the proper return code.
5359 if (signal_pending(current))
5369 static void grow_ple_window(struct kvm_vcpu *vcpu)
5371 struct vcpu_vmx *vmx = to_vmx(vcpu);
5372 unsigned int old = vmx->ple_window;
5374 vmx->ple_window = __grow_ple_window(old, ple_window,
5378 if (vmx->ple_window != old) {
5379 vmx->ple_window_dirty = true;
5380 trace_kvm_ple_window_update(vcpu->vcpu_id,
5381 vmx->ple_window, old);
5385 static void shrink_ple_window(struct kvm_vcpu *vcpu)
5387 struct vcpu_vmx *vmx = to_vmx(vcpu);
5388 unsigned int old = vmx->ple_window;
5390 vmx->ple_window = __shrink_ple_window(old, ple_window,
5394 if (vmx->ple_window != old) {
5395 vmx->ple_window_dirty = true;
5396 trace_kvm_ple_window_update(vcpu->vcpu_id,
5397 vmx->ple_window, old);
5402 * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
5404 static void wakeup_handler(void)
5406 struct kvm_vcpu *vcpu;
5407 int cpu = smp_processor_id();
5409 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5410 list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
5411 blocked_vcpu_list) {
5412 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
5414 if (pi_test_on(pi_desc) == 1)
5415 kvm_vcpu_kick(vcpu);
5417 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5420 static void vmx_enable_tdp(void)
5422 kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK,
5423 enable_ept_ad_bits ? VMX_EPT_ACCESS_BIT : 0ull,
5424 enable_ept_ad_bits ? VMX_EPT_DIRTY_BIT : 0ull,
5425 0ull, VMX_EPT_EXECUTABLE_MASK,
5426 cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK,
5427 VMX_EPT_RWX_MASK, 0ull);
5429 ept_set_mmio_spte_mask();
5433 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5434 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5436 static int handle_pause(struct kvm_vcpu *vcpu)
5438 if (!kvm_pause_in_guest(vcpu->kvm))
5439 grow_ple_window(vcpu);
5442 * Intel sdm vol3 ch-25.1.3 says: The "PAUSE-loop exiting"
5443 * VM-execution control is ignored if CPL > 0. OTOH, KVM
5444 * never set PAUSE_EXITING and just set PLE if supported,
5445 * so the vcpu must be CPL=0 if it gets a PAUSE exit.
5447 kvm_vcpu_on_spin(vcpu, true);
5448 return kvm_skip_emulated_instruction(vcpu);
5451 static int handle_nop(struct kvm_vcpu *vcpu)
5453 return kvm_skip_emulated_instruction(vcpu);
5456 static int handle_mwait(struct kvm_vcpu *vcpu)
5458 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
5459 return handle_nop(vcpu);
5462 static int handle_invalid_op(struct kvm_vcpu *vcpu)
5464 kvm_queue_exception(vcpu, UD_VECTOR);
5468 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
5473 static int handle_monitor(struct kvm_vcpu *vcpu)
5475 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
5476 return handle_nop(vcpu);
5479 static int handle_invpcid(struct kvm_vcpu *vcpu)
5481 u32 vmx_instruction_info;
5485 struct x86_exception e;
5487 unsigned long roots_to_free = 0;
5493 if (!guest_cpuid_has(vcpu, X86_FEATURE_INVPCID)) {
5494 kvm_queue_exception(vcpu, UD_VECTOR);
5498 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5499 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
5502 kvm_inject_gp(vcpu, 0);
5506 /* According to the Intel instruction reference, the memory operand
5507 * is read even if it isn't needed (e.g., for type==all)
5509 if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5510 vmx_instruction_info, false,
5511 sizeof(operand), &gva))
5514 if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
5515 kvm_inject_emulated_page_fault(vcpu, &e);
5519 if (operand.pcid >> 12 != 0) {
5520 kvm_inject_gp(vcpu, 0);
5524 pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
5527 case INVPCID_TYPE_INDIV_ADDR:
5528 if ((!pcid_enabled && (operand.pcid != 0)) ||
5529 is_noncanonical_address(operand.gla, vcpu)) {
5530 kvm_inject_gp(vcpu, 0);
5533 kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
5534 return kvm_skip_emulated_instruction(vcpu);
5536 case INVPCID_TYPE_SINGLE_CTXT:
5537 if (!pcid_enabled && (operand.pcid != 0)) {
5538 kvm_inject_gp(vcpu, 0);
5542 if (kvm_get_active_pcid(vcpu) == operand.pcid) {
5543 kvm_mmu_sync_roots(vcpu);
5544 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
5547 for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
5548 if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].pgd)
5550 roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
5552 kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
5554 * If neither the current cr3 nor any of the prev_roots use the
5555 * given PCID, then nothing needs to be done here because a
5556 * resync will happen anyway before switching to any other CR3.
5559 return kvm_skip_emulated_instruction(vcpu);
5561 case INVPCID_TYPE_ALL_NON_GLOBAL:
5563 * Currently, KVM doesn't mark global entries in the shadow
5564 * page tables, so a non-global flush just degenerates to a
5565 * global flush. If needed, we could optimize this later by
5566 * keeping track of global entries in shadow page tables.
5570 case INVPCID_TYPE_ALL_INCL_GLOBAL:
5571 kvm_mmu_unload(vcpu);
5572 return kvm_skip_emulated_instruction(vcpu);
5575 BUG(); /* We have already checked above that type <= 3 */
5579 static int handle_pml_full(struct kvm_vcpu *vcpu)
5581 unsigned long exit_qualification;
5583 trace_kvm_pml_full(vcpu->vcpu_id);
5585 exit_qualification = vmx_get_exit_qual(vcpu);
5588 * PML buffer FULL happened while executing iret from NMI,
5589 * "blocked by NMI" bit has to be set before next VM entry.
5591 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5593 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
5594 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5595 GUEST_INTR_STATE_NMI);
5598 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
5599 * here.., and there's no userspace involvement needed for PML.
5604 static fastpath_t handle_fastpath_preemption_timer(struct kvm_vcpu *vcpu)
5606 struct vcpu_vmx *vmx = to_vmx(vcpu);
5608 if (!vmx->req_immediate_exit &&
5609 !unlikely(vmx->loaded_vmcs->hv_timer_soft_disabled)) {
5610 kvm_lapic_expired_hv_timer(vcpu);
5611 return EXIT_FASTPATH_REENTER_GUEST;
5614 return EXIT_FASTPATH_NONE;
5617 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
5619 handle_fastpath_preemption_timer(vcpu);
5624 * When nested=0, all VMX instruction VM Exits filter here. The handlers
5625 * are overwritten by nested_vmx_setup() when nested=1.
5627 static int handle_vmx_instruction(struct kvm_vcpu *vcpu)
5629 kvm_queue_exception(vcpu, UD_VECTOR);
5633 static int handle_encls(struct kvm_vcpu *vcpu)
5636 * SGX virtualization is not yet supported. There is no software
5637 * enable bit for SGX, so we have to trap ENCLS and inject a #UD
5638 * to prevent the guest from executing ENCLS.
5640 kvm_queue_exception(vcpu, UD_VECTOR);
5645 * The exit handlers return 1 if the exit was handled fully and guest execution
5646 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
5647 * to be done to userspace and return 0.
5649 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
5650 [EXIT_REASON_EXCEPTION_NMI] = handle_exception_nmi,
5651 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
5652 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
5653 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
5654 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
5655 [EXIT_REASON_CR_ACCESS] = handle_cr,
5656 [EXIT_REASON_DR_ACCESS] = handle_dr,
5657 [EXIT_REASON_CPUID] = kvm_emulate_cpuid,
5658 [EXIT_REASON_MSR_READ] = kvm_emulate_rdmsr,
5659 [EXIT_REASON_MSR_WRITE] = kvm_emulate_wrmsr,
5660 [EXIT_REASON_INTERRUPT_WINDOW] = handle_interrupt_window,
5661 [EXIT_REASON_HLT] = kvm_emulate_halt,
5662 [EXIT_REASON_INVD] = handle_invd,
5663 [EXIT_REASON_INVLPG] = handle_invlpg,
5664 [EXIT_REASON_RDPMC] = handle_rdpmc,
5665 [EXIT_REASON_VMCALL] = handle_vmcall,
5666 [EXIT_REASON_VMCLEAR] = handle_vmx_instruction,
5667 [EXIT_REASON_VMLAUNCH] = handle_vmx_instruction,
5668 [EXIT_REASON_VMPTRLD] = handle_vmx_instruction,
5669 [EXIT_REASON_VMPTRST] = handle_vmx_instruction,
5670 [EXIT_REASON_VMREAD] = handle_vmx_instruction,
5671 [EXIT_REASON_VMRESUME] = handle_vmx_instruction,
5672 [EXIT_REASON_VMWRITE] = handle_vmx_instruction,
5673 [EXIT_REASON_VMOFF] = handle_vmx_instruction,
5674 [EXIT_REASON_VMON] = handle_vmx_instruction,
5675 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
5676 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
5677 [EXIT_REASON_APIC_WRITE] = handle_apic_write,
5678 [EXIT_REASON_EOI_INDUCED] = handle_apic_eoi_induced,
5679 [EXIT_REASON_WBINVD] = handle_wbinvd,
5680 [EXIT_REASON_XSETBV] = handle_xsetbv,
5681 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
5682 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
5683 [EXIT_REASON_GDTR_IDTR] = handle_desc,
5684 [EXIT_REASON_LDTR_TR] = handle_desc,
5685 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
5686 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
5687 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
5688 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_mwait,
5689 [EXIT_REASON_MONITOR_TRAP_FLAG] = handle_monitor_trap,
5690 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_monitor,
5691 [EXIT_REASON_INVEPT] = handle_vmx_instruction,
5692 [EXIT_REASON_INVVPID] = handle_vmx_instruction,
5693 [EXIT_REASON_RDRAND] = handle_invalid_op,
5694 [EXIT_REASON_RDSEED] = handle_invalid_op,
5695 [EXIT_REASON_PML_FULL] = handle_pml_full,
5696 [EXIT_REASON_INVPCID] = handle_invpcid,
5697 [EXIT_REASON_VMFUNC] = handle_vmx_instruction,
5698 [EXIT_REASON_PREEMPTION_TIMER] = handle_preemption_timer,
5699 [EXIT_REASON_ENCLS] = handle_encls,
5702 static const int kvm_vmx_max_exit_handlers =
5703 ARRAY_SIZE(kvm_vmx_exit_handlers);
5705 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
5707 *info1 = vmx_get_exit_qual(vcpu);
5708 *info2 = vmx_get_intr_info(vcpu);
5711 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
5714 __free_page(vmx->pml_pg);
5719 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
5721 struct vcpu_vmx *vmx = to_vmx(vcpu);
5725 pml_idx = vmcs_read16(GUEST_PML_INDEX);
5727 /* Do nothing if PML buffer is empty */
5728 if (pml_idx == (PML_ENTITY_NUM - 1))
5731 /* PML index always points to next available PML buffer entity */
5732 if (pml_idx >= PML_ENTITY_NUM)
5737 pml_buf = page_address(vmx->pml_pg);
5738 for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
5741 gpa = pml_buf[pml_idx];
5742 WARN_ON(gpa & (PAGE_SIZE - 1));
5743 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5746 /* reset PML index */
5747 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
5751 * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
5752 * Called before reporting dirty_bitmap to userspace.
5754 static void kvm_flush_pml_buffers(struct kvm *kvm)
5757 struct kvm_vcpu *vcpu;
5759 * We only need to kick vcpu out of guest mode here, as PML buffer
5760 * is flushed at beginning of all VMEXITs, and it's obvious that only
5761 * vcpus running in guest are possible to have unflushed GPAs in PML
5764 kvm_for_each_vcpu(i, vcpu, kvm)
5765 kvm_vcpu_kick(vcpu);
5768 static void vmx_dump_sel(char *name, uint32_t sel)
5770 pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
5771 name, vmcs_read16(sel),
5772 vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
5773 vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
5774 vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
5777 static void vmx_dump_dtsel(char *name, uint32_t limit)
5779 pr_err("%s limit=0x%08x, base=0x%016lx\n",
5780 name, vmcs_read32(limit),
5781 vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
5784 void dump_vmcs(void)
5786 u32 vmentry_ctl, vmexit_ctl;
5787 u32 cpu_based_exec_ctrl, pin_based_exec_ctrl, secondary_exec_control;
5791 if (!dump_invalid_vmcs) {
5792 pr_warn_ratelimited("set kvm_intel.dump_invalid_vmcs=1 to dump internal KVM state.\n");
5796 vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
5797 vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
5798 cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5799 pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
5800 cr4 = vmcs_readl(GUEST_CR4);
5801 efer = vmcs_read64(GUEST_IA32_EFER);
5802 secondary_exec_control = 0;
5803 if (cpu_has_secondary_exec_ctrls())
5804 secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
5806 pr_err("*** Guest State ***\n");
5807 pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5808 vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
5809 vmcs_readl(CR0_GUEST_HOST_MASK));
5810 pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5811 cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
5812 pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
5813 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
5814 (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
5816 pr_err("PDPTR0 = 0x%016llx PDPTR1 = 0x%016llx\n",
5817 vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
5818 pr_err("PDPTR2 = 0x%016llx PDPTR3 = 0x%016llx\n",
5819 vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
5821 pr_err("RSP = 0x%016lx RIP = 0x%016lx\n",
5822 vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
5823 pr_err("RFLAGS=0x%08lx DR7 = 0x%016lx\n",
5824 vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
5825 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5826 vmcs_readl(GUEST_SYSENTER_ESP),
5827 vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
5828 vmx_dump_sel("CS: ", GUEST_CS_SELECTOR);
5829 vmx_dump_sel("DS: ", GUEST_DS_SELECTOR);
5830 vmx_dump_sel("SS: ", GUEST_SS_SELECTOR);
5831 vmx_dump_sel("ES: ", GUEST_ES_SELECTOR);
5832 vmx_dump_sel("FS: ", GUEST_FS_SELECTOR);
5833 vmx_dump_sel("GS: ", GUEST_GS_SELECTOR);
5834 vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
5835 vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
5836 vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
5837 vmx_dump_sel("TR: ", GUEST_TR_SELECTOR);
5838 if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
5839 (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
5840 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
5841 efer, vmcs_read64(GUEST_IA32_PAT));
5842 pr_err("DebugCtl = 0x%016llx DebugExceptions = 0x%016lx\n",
5843 vmcs_read64(GUEST_IA32_DEBUGCTL),
5844 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
5845 if (cpu_has_load_perf_global_ctrl() &&
5846 vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
5847 pr_err("PerfGlobCtl = 0x%016llx\n",
5848 vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
5849 if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
5850 pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
5851 pr_err("Interruptibility = %08x ActivityState = %08x\n",
5852 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
5853 vmcs_read32(GUEST_ACTIVITY_STATE));
5854 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
5855 pr_err("InterruptStatus = %04x\n",
5856 vmcs_read16(GUEST_INTR_STATUS));
5858 pr_err("*** Host State ***\n");
5859 pr_err("RIP = 0x%016lx RSP = 0x%016lx\n",
5860 vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
5861 pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
5862 vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
5863 vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
5864 vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
5865 vmcs_read16(HOST_TR_SELECTOR));
5866 pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
5867 vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
5868 vmcs_readl(HOST_TR_BASE));
5869 pr_err("GDTBase=%016lx IDTBase=%016lx\n",
5870 vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
5871 pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
5872 vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
5873 vmcs_readl(HOST_CR4));
5874 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5875 vmcs_readl(HOST_IA32_SYSENTER_ESP),
5876 vmcs_read32(HOST_IA32_SYSENTER_CS),
5877 vmcs_readl(HOST_IA32_SYSENTER_EIP));
5878 if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
5879 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
5880 vmcs_read64(HOST_IA32_EFER),
5881 vmcs_read64(HOST_IA32_PAT));
5882 if (cpu_has_load_perf_global_ctrl() &&
5883 vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
5884 pr_err("PerfGlobCtl = 0x%016llx\n",
5885 vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
5887 pr_err("*** Control State ***\n");
5888 pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
5889 pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
5890 pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
5891 pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
5892 vmcs_read32(EXCEPTION_BITMAP),
5893 vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
5894 vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
5895 pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
5896 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
5897 vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
5898 vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
5899 pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
5900 vmcs_read32(VM_EXIT_INTR_INFO),
5901 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
5902 vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
5903 pr_err(" reason=%08x qualification=%016lx\n",
5904 vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
5905 pr_err("IDTVectoring: info=%08x errcode=%08x\n",
5906 vmcs_read32(IDT_VECTORING_INFO_FIELD),
5907 vmcs_read32(IDT_VECTORING_ERROR_CODE));
5908 pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
5909 if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
5910 pr_err("TSC Multiplier = 0x%016llx\n",
5911 vmcs_read64(TSC_MULTIPLIER));
5912 if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW) {
5913 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
5914 u16 status = vmcs_read16(GUEST_INTR_STATUS);
5915 pr_err("SVI|RVI = %02x|%02x ", status >> 8, status & 0xff);
5917 pr_cont("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
5918 if (secondary_exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
5919 pr_err("APIC-access addr = 0x%016llx ", vmcs_read64(APIC_ACCESS_ADDR));
5920 pr_cont("virt-APIC addr = 0x%016llx\n", vmcs_read64(VIRTUAL_APIC_PAGE_ADDR));
5922 if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
5923 pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
5924 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
5925 pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
5926 if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
5927 pr_err("PLE Gap=%08x Window=%08x\n",
5928 vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
5929 if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
5930 pr_err("Virtual processor ID = 0x%04x\n",
5931 vmcs_read16(VIRTUAL_PROCESSOR_ID));
5935 * The guest has exited. See if we can fix it or if we need userspace
5938 static int vmx_handle_exit(struct kvm_vcpu *vcpu, fastpath_t exit_fastpath)
5940 struct vcpu_vmx *vmx = to_vmx(vcpu);
5941 u32 exit_reason = vmx->exit_reason;
5942 u32 vectoring_info = vmx->idt_vectoring_info;
5945 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
5946 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
5947 * querying dirty_bitmap, we only need to kick all vcpus out of guest
5948 * mode as if vcpus is in root mode, the PML buffer must has been
5952 vmx_flush_pml_buffer(vcpu);
5955 * We should never reach this point with a pending nested VM-Enter, and
5956 * more specifically emulation of L2 due to invalid guest state (see
5957 * below) should never happen as that means we incorrectly allowed a
5958 * nested VM-Enter with an invalid vmcs12.
5960 WARN_ON_ONCE(vmx->nested.nested_run_pending);
5962 /* If guest state is invalid, start emulating */
5963 if (vmx->emulation_required)
5964 return handle_invalid_guest_state(vcpu);
5966 if (is_guest_mode(vcpu)) {
5968 * The host physical addresses of some pages of guest memory
5969 * are loaded into the vmcs02 (e.g. vmcs12's Virtual APIC
5970 * Page). The CPU may write to these pages via their host
5971 * physical address while L2 is running, bypassing any
5972 * address-translation-based dirty tracking (e.g. EPT write
5975 * Mark them dirty on every exit from L2 to prevent them from
5976 * getting out of sync with dirty tracking.
5978 nested_mark_vmcs12_pages_dirty(vcpu);
5980 if (nested_vmx_reflect_vmexit(vcpu))
5984 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
5986 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5987 vcpu->run->fail_entry.hardware_entry_failure_reason
5992 if (unlikely(vmx->fail)) {
5994 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5995 vcpu->run->fail_entry.hardware_entry_failure_reason
5996 = vmcs_read32(VM_INSTRUCTION_ERROR);
6002 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
6003 * delivery event since it indicates guest is accessing MMIO.
6004 * The vm-exit can be triggered again after return to guest that
6005 * will cause infinite loop.
6007 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
6008 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
6009 exit_reason != EXIT_REASON_EPT_VIOLATION &&
6010 exit_reason != EXIT_REASON_PML_FULL &&
6011 exit_reason != EXIT_REASON_TASK_SWITCH)) {
6012 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6013 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
6014 vcpu->run->internal.ndata = 3;
6015 vcpu->run->internal.data[0] = vectoring_info;
6016 vcpu->run->internal.data[1] = exit_reason;
6017 vcpu->run->internal.data[2] = vcpu->arch.exit_qualification;
6018 if (exit_reason == EXIT_REASON_EPT_MISCONFIG) {
6019 vcpu->run->internal.ndata++;
6020 vcpu->run->internal.data[3] =
6021 vmcs_read64(GUEST_PHYSICAL_ADDRESS);
6026 if (unlikely(!enable_vnmi &&
6027 vmx->loaded_vmcs->soft_vnmi_blocked)) {
6028 if (!vmx_interrupt_blocked(vcpu)) {
6029 vmx->loaded_vmcs->soft_vnmi_blocked = 0;
6030 } else if (vmx->loaded_vmcs->vnmi_blocked_time > 1000000000LL &&
6031 vcpu->arch.nmi_pending) {
6033 * This CPU don't support us in finding the end of an
6034 * NMI-blocked window if the guest runs with IRQs
6035 * disabled. So we pull the trigger after 1 s of
6036 * futile waiting, but inform the user about this.
6038 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
6039 "state on VCPU %d after 1 s timeout\n",
6040 __func__, vcpu->vcpu_id);
6041 vmx->loaded_vmcs->soft_vnmi_blocked = 0;
6045 if (exit_fastpath != EXIT_FASTPATH_NONE)
6048 if (exit_reason >= kvm_vmx_max_exit_handlers)
6049 goto unexpected_vmexit;
6050 #ifdef CONFIG_RETPOLINE
6051 if (exit_reason == EXIT_REASON_MSR_WRITE)
6052 return kvm_emulate_wrmsr(vcpu);
6053 else if (exit_reason == EXIT_REASON_PREEMPTION_TIMER)
6054 return handle_preemption_timer(vcpu);
6055 else if (exit_reason == EXIT_REASON_INTERRUPT_WINDOW)
6056 return handle_interrupt_window(vcpu);
6057 else if (exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
6058 return handle_external_interrupt(vcpu);
6059 else if (exit_reason == EXIT_REASON_HLT)
6060 return kvm_emulate_halt(vcpu);
6061 else if (exit_reason == EXIT_REASON_EPT_MISCONFIG)
6062 return handle_ept_misconfig(vcpu);
6065 exit_reason = array_index_nospec(exit_reason,
6066 kvm_vmx_max_exit_handlers);
6067 if (!kvm_vmx_exit_handlers[exit_reason])
6068 goto unexpected_vmexit;
6070 return kvm_vmx_exit_handlers[exit_reason](vcpu);
6073 vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n", exit_reason);
6075 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6076 vcpu->run->internal.suberror =
6077 KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON;
6078 vcpu->run->internal.ndata = 1;
6079 vcpu->run->internal.data[0] = exit_reason;
6084 * Software based L1D cache flush which is used when microcode providing
6085 * the cache control MSR is not loaded.
6087 * The L1D cache is 32 KiB on Nehalem and later microarchitectures, but to
6088 * flush it is required to read in 64 KiB because the replacement algorithm
6089 * is not exactly LRU. This could be sized at runtime via topology
6090 * information but as all relevant affected CPUs have 32KiB L1D cache size
6091 * there is no point in doing so.
6093 static void vmx_l1d_flush(struct kvm_vcpu *vcpu)
6095 int size = PAGE_SIZE << L1D_CACHE_ORDER;
6098 * This code is only executed when the the flush mode is 'cond' or
6101 if (static_branch_likely(&vmx_l1d_flush_cond)) {
6105 * Clear the per-vcpu flush bit, it gets set again
6106 * either from vcpu_run() or from one of the unsafe
6109 flush_l1d = vcpu->arch.l1tf_flush_l1d;
6110 vcpu->arch.l1tf_flush_l1d = false;
6113 * Clear the per-cpu flush bit, it gets set again from
6114 * the interrupt handlers.
6116 flush_l1d |= kvm_get_cpu_l1tf_flush_l1d();
6117 kvm_clear_cpu_l1tf_flush_l1d();
6123 vcpu->stat.l1d_flush++;
6125 if (static_cpu_has(X86_FEATURE_FLUSH_L1D)) {
6126 wrmsrl(MSR_IA32_FLUSH_CMD, L1D_FLUSH);
6131 /* First ensure the pages are in the TLB */
6132 "xorl %%eax, %%eax\n"
6133 ".Lpopulate_tlb:\n\t"
6134 "movzbl (%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
6135 "addl $4096, %%eax\n\t"
6136 "cmpl %%eax, %[size]\n\t"
6137 "jne .Lpopulate_tlb\n\t"
6138 "xorl %%eax, %%eax\n\t"
6140 /* Now fill the cache */
6141 "xorl %%eax, %%eax\n"
6143 "movzbl (%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
6144 "addl $64, %%eax\n\t"
6145 "cmpl %%eax, %[size]\n\t"
6146 "jne .Lfill_cache\n\t"
6148 :: [flush_pages] "r" (vmx_l1d_flush_pages),
6150 : "eax", "ebx", "ecx", "edx");
6153 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
6155 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6158 if (is_guest_mode(vcpu) &&
6159 nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
6162 tpr_threshold = (irr == -1 || tpr < irr) ? 0 : irr;
6163 if (is_guest_mode(vcpu))
6164 to_vmx(vcpu)->nested.l1_tpr_threshold = tpr_threshold;
6166 vmcs_write32(TPR_THRESHOLD, tpr_threshold);
6169 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu)
6171 struct vcpu_vmx *vmx = to_vmx(vcpu);
6172 u32 sec_exec_control;
6174 if (!lapic_in_kernel(vcpu))
6177 if (!flexpriority_enabled &&
6178 !cpu_has_vmx_virtualize_x2apic_mode())
6181 /* Postpone execution until vmcs01 is the current VMCS. */
6182 if (is_guest_mode(vcpu)) {
6183 vmx->nested.change_vmcs01_virtual_apic_mode = true;
6187 sec_exec_control = secondary_exec_controls_get(vmx);
6188 sec_exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6189 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
6191 switch (kvm_get_apic_mode(vcpu)) {
6192 case LAPIC_MODE_INVALID:
6193 WARN_ONCE(true, "Invalid local APIC state");
6194 case LAPIC_MODE_DISABLED:
6196 case LAPIC_MODE_XAPIC:
6197 if (flexpriority_enabled) {
6199 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6200 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
6203 * Flush the TLB, reloading the APIC access page will
6204 * only do so if its physical address has changed, but
6205 * the guest may have inserted a non-APIC mapping into
6206 * the TLB while the APIC access page was disabled.
6208 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
6211 case LAPIC_MODE_X2APIC:
6212 if (cpu_has_vmx_virtualize_x2apic_mode())
6214 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
6217 secondary_exec_controls_set(vmx, sec_exec_control);
6219 vmx_update_msr_bitmap(vcpu);
6222 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu)
6226 /* Defer reload until vmcs01 is the current VMCS. */
6227 if (is_guest_mode(vcpu)) {
6228 to_vmx(vcpu)->nested.reload_vmcs01_apic_access_page = true;
6232 if (!(secondary_exec_controls_get(to_vmx(vcpu)) &
6233 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
6236 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6237 if (is_error_page(page))
6240 vmcs_write64(APIC_ACCESS_ADDR, page_to_phys(page));
6241 vmx_flush_tlb_current(vcpu);
6244 * Do not pin apic access page in memory, the MMU notifier
6245 * will call us again if it is migrated or swapped out.
6250 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
6258 status = vmcs_read16(GUEST_INTR_STATUS);
6260 if (max_isr != old) {
6262 status |= max_isr << 8;
6263 vmcs_write16(GUEST_INTR_STATUS, status);
6267 static void vmx_set_rvi(int vector)
6275 status = vmcs_read16(GUEST_INTR_STATUS);
6276 old = (u8)status & 0xff;
6277 if ((u8)vector != old) {
6279 status |= (u8)vector;
6280 vmcs_write16(GUEST_INTR_STATUS, status);
6284 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
6287 * When running L2, updating RVI is only relevant when
6288 * vmcs12 virtual-interrupt-delivery enabled.
6289 * However, it can be enabled only when L1 also
6290 * intercepts external-interrupts and in that case
6291 * we should not update vmcs02 RVI but instead intercept
6292 * interrupt. Therefore, do nothing when running L2.
6294 if (!is_guest_mode(vcpu))
6295 vmx_set_rvi(max_irr);
6298 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
6300 struct vcpu_vmx *vmx = to_vmx(vcpu);
6302 bool max_irr_updated;
6304 WARN_ON(!vcpu->arch.apicv_active);
6305 if (pi_test_on(&vmx->pi_desc)) {
6306 pi_clear_on(&vmx->pi_desc);
6308 * IOMMU can write to PID.ON, so the barrier matters even on UP.
6309 * But on x86 this is just a compiler barrier anyway.
6311 smp_mb__after_atomic();
6313 kvm_apic_update_irr(vcpu, vmx->pi_desc.pir, &max_irr);
6316 * If we are running L2 and L1 has a new pending interrupt
6317 * which can be injected, we should re-evaluate
6318 * what should be done with this new L1 interrupt.
6319 * If L1 intercepts external-interrupts, we should
6320 * exit from L2 to L1. Otherwise, interrupt should be
6321 * delivered directly to L2.
6323 if (is_guest_mode(vcpu) && max_irr_updated) {
6324 if (nested_exit_on_intr(vcpu))
6325 kvm_vcpu_exiting_guest_mode(vcpu);
6327 kvm_make_request(KVM_REQ_EVENT, vcpu);
6330 max_irr = kvm_lapic_find_highest_irr(vcpu);
6332 vmx_hwapic_irr_update(vcpu, max_irr);
6336 static bool vmx_dy_apicv_has_pending_interrupt(struct kvm_vcpu *vcpu)
6338 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
6340 return pi_test_on(pi_desc) ||
6341 (pi_test_sn(pi_desc) && !pi_is_pir_empty(pi_desc));
6344 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
6346 if (!kvm_vcpu_apicv_active(vcpu))
6349 vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
6350 vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
6351 vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
6352 vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
6355 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
6357 struct vcpu_vmx *vmx = to_vmx(vcpu);
6359 pi_clear_on(&vmx->pi_desc);
6360 memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
6363 static void handle_exception_nmi_irqoff(struct vcpu_vmx *vmx)
6365 u32 intr_info = vmx_get_intr_info(&vmx->vcpu);
6367 /* if exit due to PF check for async PF */
6368 if (is_page_fault(intr_info)) {
6369 vmx->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason();
6370 /* Handle machine checks before interrupts are enabled */
6371 } else if (is_machine_check(intr_info)) {
6372 kvm_machine_check();
6373 /* We need to handle NMIs before interrupts are enabled */
6374 } else if (is_nmi(intr_info)) {
6375 kvm_before_interrupt(&vmx->vcpu);
6377 kvm_after_interrupt(&vmx->vcpu);
6381 static void handle_external_interrupt_irqoff(struct kvm_vcpu *vcpu)
6383 unsigned int vector;
6384 unsigned long entry;
6385 #ifdef CONFIG_X86_64
6389 u32 intr_info = vmx_get_intr_info(vcpu);
6391 if (WARN_ONCE(!is_external_intr(intr_info),
6392 "KVM: unexpected VM-Exit interrupt info: 0x%x", intr_info))
6395 vector = intr_info & INTR_INFO_VECTOR_MASK;
6396 desc = (gate_desc *)host_idt_base + vector;
6397 entry = gate_offset(desc);
6399 kvm_before_interrupt(vcpu);
6402 #ifdef CONFIG_X86_64
6403 "mov %%rsp, %[sp]\n\t"
6404 "and $-16, %%rsp\n\t"
6412 #ifdef CONFIG_X86_64
6417 [thunk_target]"r"(entry),
6418 #ifdef CONFIG_X86_64
6419 [ss]"i"(__KERNEL_DS),
6421 [cs]"i"(__KERNEL_CS)
6424 kvm_after_interrupt(vcpu);
6426 STACK_FRAME_NON_STANDARD(handle_external_interrupt_irqoff);
6428 static void vmx_handle_exit_irqoff(struct kvm_vcpu *vcpu)
6430 struct vcpu_vmx *vmx = to_vmx(vcpu);
6432 if (vmx->exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
6433 handle_external_interrupt_irqoff(vcpu);
6434 else if (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI)
6435 handle_exception_nmi_irqoff(vmx);
6438 static bool vmx_has_emulated_msr(int index)
6441 case MSR_IA32_SMBASE:
6443 * We cannot do SMM unless we can run the guest in big
6446 return enable_unrestricted_guest || emulate_invalid_guest_state;
6447 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
6449 case MSR_AMD64_VIRT_SPEC_CTRL:
6450 /* This is AMD only. */
6457 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
6462 bool idtv_info_valid;
6464 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6467 if (vmx->loaded_vmcs->nmi_known_unmasked)
6470 exit_intr_info = vmx_get_intr_info(&vmx->vcpu);
6471 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
6472 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
6474 * SDM 3: 27.7.1.2 (September 2008)
6475 * Re-set bit "block by NMI" before VM entry if vmexit caused by
6476 * a guest IRET fault.
6477 * SDM 3: 23.2.2 (September 2008)
6478 * Bit 12 is undefined in any of the following cases:
6479 * If the VM exit sets the valid bit in the IDT-vectoring
6480 * information field.
6481 * If the VM exit is due to a double fault.
6483 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
6484 vector != DF_VECTOR && !idtv_info_valid)
6485 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
6486 GUEST_INTR_STATE_NMI);
6488 vmx->loaded_vmcs->nmi_known_unmasked =
6489 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
6490 & GUEST_INTR_STATE_NMI);
6491 } else if (unlikely(vmx->loaded_vmcs->soft_vnmi_blocked))
6492 vmx->loaded_vmcs->vnmi_blocked_time +=
6493 ktime_to_ns(ktime_sub(ktime_get(),
6494 vmx->loaded_vmcs->entry_time));
6497 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
6498 u32 idt_vectoring_info,
6499 int instr_len_field,
6500 int error_code_field)
6504 bool idtv_info_valid;
6506 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6508 vcpu->arch.nmi_injected = false;
6509 kvm_clear_exception_queue(vcpu);
6510 kvm_clear_interrupt_queue(vcpu);
6512 if (!idtv_info_valid)
6515 kvm_make_request(KVM_REQ_EVENT, vcpu);
6517 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
6518 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
6521 case INTR_TYPE_NMI_INTR:
6522 vcpu->arch.nmi_injected = true;
6524 * SDM 3: 27.7.1.2 (September 2008)
6525 * Clear bit "block by NMI" before VM entry if a NMI
6528 vmx_set_nmi_mask(vcpu, false);
6530 case INTR_TYPE_SOFT_EXCEPTION:
6531 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6533 case INTR_TYPE_HARD_EXCEPTION:
6534 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
6535 u32 err = vmcs_read32(error_code_field);
6536 kvm_requeue_exception_e(vcpu, vector, err);
6538 kvm_requeue_exception(vcpu, vector);
6540 case INTR_TYPE_SOFT_INTR:
6541 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6543 case INTR_TYPE_EXT_INTR:
6544 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
6551 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
6553 __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
6554 VM_EXIT_INSTRUCTION_LEN,
6555 IDT_VECTORING_ERROR_CODE);
6558 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
6560 __vmx_complete_interrupts(vcpu,
6561 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
6562 VM_ENTRY_INSTRUCTION_LEN,
6563 VM_ENTRY_EXCEPTION_ERROR_CODE);
6565 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
6568 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
6571 struct perf_guest_switch_msr *msrs;
6573 msrs = perf_guest_get_msrs(&nr_msrs);
6578 for (i = 0; i < nr_msrs; i++)
6579 if (msrs[i].host == msrs[i].guest)
6580 clear_atomic_switch_msr(vmx, msrs[i].msr);
6582 add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
6583 msrs[i].host, false);
6586 static void atomic_switch_umwait_control_msr(struct vcpu_vmx *vmx)
6588 u32 host_umwait_control;
6590 if (!vmx_has_waitpkg(vmx))
6593 host_umwait_control = get_umwait_control_msr();
6595 if (vmx->msr_ia32_umwait_control != host_umwait_control)
6596 add_atomic_switch_msr(vmx, MSR_IA32_UMWAIT_CONTROL,
6597 vmx->msr_ia32_umwait_control,
6598 host_umwait_control, false);
6600 clear_atomic_switch_msr(vmx, MSR_IA32_UMWAIT_CONTROL);
6603 static void vmx_update_hv_timer(struct kvm_vcpu *vcpu)
6605 struct vcpu_vmx *vmx = to_vmx(vcpu);
6609 if (vmx->req_immediate_exit) {
6610 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, 0);
6611 vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6612 } else if (vmx->hv_deadline_tsc != -1) {
6614 if (vmx->hv_deadline_tsc > tscl)
6615 /* set_hv_timer ensures the delta fits in 32-bits */
6616 delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
6617 cpu_preemption_timer_multi);
6621 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
6622 vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6623 } else if (!vmx->loaded_vmcs->hv_timer_soft_disabled) {
6624 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, -1);
6625 vmx->loaded_vmcs->hv_timer_soft_disabled = true;
6629 void vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp)
6631 if (unlikely(host_rsp != vmx->loaded_vmcs->host_state.rsp)) {
6632 vmx->loaded_vmcs->host_state.rsp = host_rsp;
6633 vmcs_writel(HOST_RSP, host_rsp);
6637 static fastpath_t vmx_exit_handlers_fastpath(struct kvm_vcpu *vcpu)
6639 switch (to_vmx(vcpu)->exit_reason) {
6640 case EXIT_REASON_MSR_WRITE:
6641 return handle_fastpath_set_msr_irqoff(vcpu);
6642 case EXIT_REASON_PREEMPTION_TIMER:
6643 return handle_fastpath_preemption_timer(vcpu);
6645 return EXIT_FASTPATH_NONE;
6649 bool __vmx_vcpu_run(struct vcpu_vmx *vmx, unsigned long *regs, bool launched);
6651 static fastpath_t vmx_vcpu_run(struct kvm_vcpu *vcpu)
6653 fastpath_t exit_fastpath;
6654 struct vcpu_vmx *vmx = to_vmx(vcpu);
6655 unsigned long cr3, cr4;
6658 /* Record the guest's net vcpu time for enforced NMI injections. */
6659 if (unlikely(!enable_vnmi &&
6660 vmx->loaded_vmcs->soft_vnmi_blocked))
6661 vmx->loaded_vmcs->entry_time = ktime_get();
6663 /* Don't enter VMX if guest state is invalid, let the exit handler
6664 start emulation until we arrive back to a valid state */
6665 if (vmx->emulation_required)
6666 return EXIT_FASTPATH_NONE;
6668 if (vmx->ple_window_dirty) {
6669 vmx->ple_window_dirty = false;
6670 vmcs_write32(PLE_WINDOW, vmx->ple_window);
6674 * We did this in prepare_switch_to_guest, because it needs to
6675 * be within srcu_read_lock.
6677 WARN_ON_ONCE(vmx->nested.need_vmcs12_to_shadow_sync);
6679 if (kvm_register_is_dirty(vcpu, VCPU_REGS_RSP))
6680 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
6681 if (kvm_register_is_dirty(vcpu, VCPU_REGS_RIP))
6682 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
6684 cr3 = __get_current_cr3_fast();
6685 if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
6686 vmcs_writel(HOST_CR3, cr3);
6687 vmx->loaded_vmcs->host_state.cr3 = cr3;
6690 cr4 = cr4_read_shadow();
6691 if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
6692 vmcs_writel(HOST_CR4, cr4);
6693 vmx->loaded_vmcs->host_state.cr4 = cr4;
6696 /* When single-stepping over STI and MOV SS, we must clear the
6697 * corresponding interruptibility bits in the guest state. Otherwise
6698 * vmentry fails as it then expects bit 14 (BS) in pending debug
6699 * exceptions being set, but that's not correct for the guest debugging
6701 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6702 vmx_set_interrupt_shadow(vcpu, 0);
6704 kvm_load_guest_xsave_state(vcpu);
6706 pt_guest_enter(vmx);
6708 if (vcpu_to_pmu(vcpu)->version)
6709 atomic_switch_perf_msrs(vmx);
6710 atomic_switch_umwait_control_msr(vmx);
6712 if (enable_preemption_timer)
6713 vmx_update_hv_timer(vcpu);
6715 if (lapic_in_kernel(vcpu) &&
6716 vcpu->arch.apic->lapic_timer.timer_advance_ns)
6717 kvm_wait_lapic_expire(vcpu);
6720 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
6721 * it's non-zero. Since vmentry is serialising on affected CPUs, there
6722 * is no need to worry about the conditional branch over the wrmsr
6723 * being speculatively taken.
6725 x86_spec_ctrl_set_guest(vmx->spec_ctrl, 0);
6727 /* L1D Flush includes CPU buffer clear to mitigate MDS */
6728 if (static_branch_unlikely(&vmx_l1d_should_flush))
6729 vmx_l1d_flush(vcpu);
6730 else if (static_branch_unlikely(&mds_user_clear))
6731 mds_clear_cpu_buffers();
6733 if (vcpu->arch.cr2 != read_cr2())
6734 write_cr2(vcpu->arch.cr2);
6736 vmx->fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
6737 vmx->loaded_vmcs->launched);
6739 vcpu->arch.cr2 = read_cr2();
6742 * We do not use IBRS in the kernel. If this vCPU has used the
6743 * SPEC_CTRL MSR it may have left it on; save the value and
6744 * turn it off. This is much more efficient than blindly adding
6745 * it to the atomic save/restore list. Especially as the former
6746 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
6748 * For non-nested case:
6749 * If the L01 MSR bitmap does not intercept the MSR, then we need to
6753 * If the L02 MSR bitmap does not intercept the MSR, then we need to
6756 if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
6757 vmx->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
6759 x86_spec_ctrl_restore_host(vmx->spec_ctrl, 0);
6761 /* All fields are clean at this point */
6762 if (static_branch_unlikely(&enable_evmcs))
6763 current_evmcs->hv_clean_fields |=
6764 HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
6766 if (static_branch_unlikely(&enable_evmcs))
6767 current_evmcs->hv_vp_id = vcpu->arch.hyperv.vp_index;
6769 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
6770 if (vmx->host_debugctlmsr)
6771 update_debugctlmsr(vmx->host_debugctlmsr);
6773 #ifndef CONFIG_X86_64
6775 * The sysexit path does not restore ds/es, so we must set them to
6776 * a reasonable value ourselves.
6778 * We can't defer this to vmx_prepare_switch_to_host() since that
6779 * function may be executed in interrupt context, which saves and
6780 * restore segments around it, nullifying its effect.
6782 loadsegment(ds, __USER_DS);
6783 loadsegment(es, __USER_DS);
6786 vmx_register_cache_reset(vcpu);
6790 kvm_load_host_xsave_state(vcpu);
6792 vmx->nested.nested_run_pending = 0;
6793 vmx->idt_vectoring_info = 0;
6795 if (unlikely(vmx->fail)) {
6796 vmx->exit_reason = 0xdead;
6797 return EXIT_FASTPATH_NONE;
6800 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
6801 if (unlikely((u16)vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY))
6802 kvm_machine_check();
6804 trace_kvm_exit(vmx->exit_reason, vcpu, KVM_ISA_VMX);
6806 if (unlikely(vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
6807 return EXIT_FASTPATH_NONE;
6809 vmx->loaded_vmcs->launched = 1;
6810 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
6812 vmx_recover_nmi_blocking(vmx);
6813 vmx_complete_interrupts(vmx);
6815 if (is_guest_mode(vcpu))
6816 return EXIT_FASTPATH_NONE;
6818 exit_fastpath = vmx_exit_handlers_fastpath(vcpu);
6819 if (exit_fastpath == EXIT_FASTPATH_REENTER_GUEST) {
6820 if (!kvm_vcpu_exit_request(vcpu)) {
6822 * FIXME: this goto should be a loop in vcpu_enter_guest,
6823 * but it would incur the cost of a retpoline for now.
6824 * Revisit once static calls are available.
6826 if (vcpu->arch.apicv_active)
6827 vmx_sync_pir_to_irr(vcpu);
6830 exit_fastpath = EXIT_FASTPATH_EXIT_HANDLED;
6833 return exit_fastpath;
6836 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
6838 struct vcpu_vmx *vmx = to_vmx(vcpu);
6841 vmx_destroy_pml_buffer(vmx);
6842 free_vpid(vmx->vpid);
6843 nested_vmx_free_vcpu(vcpu);
6844 free_loaded_vmcs(vmx->loaded_vmcs);
6847 static int vmx_create_vcpu(struct kvm_vcpu *vcpu)
6849 struct vcpu_vmx *vmx;
6850 unsigned long *msr_bitmap;
6853 BUILD_BUG_ON(offsetof(struct vcpu_vmx, vcpu) != 0);
6858 vmx->vpid = allocate_vpid();
6861 * If PML is turned on, failure on enabling PML just results in failure
6862 * of creating the vcpu, therefore we can simplify PML logic (by
6863 * avoiding dealing with cases, such as enabling PML partially on vcpus
6864 * for the guest), etc.
6867 vmx->pml_pg = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
6872 BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) != NR_SHARED_MSRS);
6874 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
6875 u32 index = vmx_msr_index[i];
6876 u32 data_low, data_high;
6879 if (rdmsr_safe(index, &data_low, &data_high) < 0)
6881 if (wrmsr_safe(index, data_low, data_high) < 0)
6884 vmx->guest_msrs[j].index = i;
6885 vmx->guest_msrs[j].data = 0;
6887 case MSR_IA32_TSX_CTRL:
6889 * No need to pass TSX_CTRL_CPUID_CLEAR through, so
6890 * let's avoid changing CPUID bits under the host
6893 vmx->guest_msrs[j].mask = ~(u64)TSX_CTRL_CPUID_CLEAR;
6896 vmx->guest_msrs[j].mask = -1ull;
6902 err = alloc_loaded_vmcs(&vmx->vmcs01);
6906 msr_bitmap = vmx->vmcs01.msr_bitmap;
6907 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_TSC, MSR_TYPE_R);
6908 vmx_disable_intercept_for_msr(msr_bitmap, MSR_FS_BASE, MSR_TYPE_RW);
6909 vmx_disable_intercept_for_msr(msr_bitmap, MSR_GS_BASE, MSR_TYPE_RW);
6910 vmx_disable_intercept_for_msr(msr_bitmap, MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
6911 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_CS, MSR_TYPE_RW);
6912 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_ESP, MSR_TYPE_RW);
6913 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_EIP, MSR_TYPE_RW);
6914 if (kvm_cstate_in_guest(vcpu->kvm)) {
6915 vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C1_RES, MSR_TYPE_R);
6916 vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C3_RESIDENCY, MSR_TYPE_R);
6917 vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C6_RESIDENCY, MSR_TYPE_R);
6918 vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C7_RESIDENCY, MSR_TYPE_R);
6920 vmx->msr_bitmap_mode = 0;
6922 vmx->loaded_vmcs = &vmx->vmcs01;
6924 vmx_vcpu_load(vcpu, cpu);
6929 if (cpu_need_virtualize_apic_accesses(vcpu)) {
6930 err = alloc_apic_access_page(vcpu->kvm);
6935 if (enable_ept && !enable_unrestricted_guest) {
6936 err = init_rmode_identity_map(vcpu->kvm);
6942 nested_vmx_setup_ctls_msrs(&vmx->nested.msrs,
6943 vmx_capability.ept);
6945 memset(&vmx->nested.msrs, 0, sizeof(vmx->nested.msrs));
6947 vmx->nested.posted_intr_nv = -1;
6948 vmx->nested.current_vmptr = -1ull;
6950 vcpu->arch.microcode_version = 0x100000000ULL;
6951 vmx->msr_ia32_feature_control_valid_bits = FEAT_CTL_LOCKED;
6954 * Enforce invariant: pi_desc.nv is always either POSTED_INTR_VECTOR
6955 * or POSTED_INTR_WAKEUP_VECTOR.
6957 vmx->pi_desc.nv = POSTED_INTR_VECTOR;
6958 vmx->pi_desc.sn = 1;
6960 vmx->ept_pointer = INVALID_PAGE;
6965 free_loaded_vmcs(vmx->loaded_vmcs);
6967 vmx_destroy_pml_buffer(vmx);
6969 free_vpid(vmx->vpid);
6973 #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"
6974 #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"
6976 static int vmx_vm_init(struct kvm *kvm)
6978 spin_lock_init(&to_kvm_vmx(kvm)->ept_pointer_lock);
6981 kvm->arch.pause_in_guest = true;
6983 if (boot_cpu_has(X86_BUG_L1TF) && enable_ept) {
6984 switch (l1tf_mitigation) {
6985 case L1TF_MITIGATION_OFF:
6986 case L1TF_MITIGATION_FLUSH_NOWARN:
6987 /* 'I explicitly don't care' is set */
6989 case L1TF_MITIGATION_FLUSH:
6990 case L1TF_MITIGATION_FLUSH_NOSMT:
6991 case L1TF_MITIGATION_FULL:
6993 * Warn upon starting the first VM in a potentially
6994 * insecure environment.
6996 if (sched_smt_active())
6997 pr_warn_once(L1TF_MSG_SMT);
6998 if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER)
6999 pr_warn_once(L1TF_MSG_L1D);
7001 case L1TF_MITIGATION_FULL_FORCE:
7002 /* Flush is enforced */
7006 kvm_apicv_init(kvm, enable_apicv);
7010 static int __init vmx_check_processor_compat(void)
7012 struct vmcs_config vmcs_conf;
7013 struct vmx_capability vmx_cap;
7015 if (!this_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
7016 !this_cpu_has(X86_FEATURE_VMX)) {
7017 pr_err("kvm: VMX is disabled on CPU %d\n", smp_processor_id());
7021 if (setup_vmcs_config(&vmcs_conf, &vmx_cap) < 0)
7024 nested_vmx_setup_ctls_msrs(&vmcs_conf.nested, vmx_cap.ept);
7025 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
7026 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
7027 smp_processor_id());
7033 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
7038 /* We wanted to honor guest CD/MTRR/PAT, but doing so could result in
7039 * memory aliases with conflicting memory types and sometimes MCEs.
7040 * We have to be careful as to what are honored and when.
7042 * For MMIO, guest CD/MTRR are ignored. The EPT memory type is set to
7043 * UC. The effective memory type is UC or WC depending on guest PAT.
7044 * This was historically the source of MCEs and we want to be
7047 * When there is no need to deal with noncoherent DMA (e.g., no VT-d
7048 * or VT-d has snoop control), guest CD/MTRR/PAT are all ignored. The
7049 * EPT memory type is set to WB. The effective memory type is forced
7052 * Otherwise, we trust guest. Guest CD/MTRR/PAT are all honored. The
7053 * EPT memory type is used to emulate guest CD/MTRR.
7057 cache = MTRR_TYPE_UNCACHABLE;
7061 if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
7062 ipat = VMX_EPT_IPAT_BIT;
7063 cache = MTRR_TYPE_WRBACK;
7067 if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
7068 ipat = VMX_EPT_IPAT_BIT;
7069 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
7070 cache = MTRR_TYPE_WRBACK;
7072 cache = MTRR_TYPE_UNCACHABLE;
7076 cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
7079 return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
7082 static void vmcs_set_secondary_exec_control(struct vcpu_vmx *vmx)
7085 * These bits in the secondary execution controls field
7086 * are dynamic, the others are mostly based on the hypervisor
7087 * architecture and the guest's CPUID. Do not touch the
7091 SECONDARY_EXEC_SHADOW_VMCS |
7092 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
7093 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
7094 SECONDARY_EXEC_DESC;
7096 u32 new_ctl = vmx->secondary_exec_control;
7097 u32 cur_ctl = secondary_exec_controls_get(vmx);
7099 secondary_exec_controls_set(vmx, (new_ctl & ~mask) | (cur_ctl & mask));
7103 * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
7104 * (indicating "allowed-1") if they are supported in the guest's CPUID.
7106 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
7108 struct vcpu_vmx *vmx = to_vmx(vcpu);
7109 struct kvm_cpuid_entry2 *entry;
7111 vmx->nested.msrs.cr0_fixed1 = 0xffffffff;
7112 vmx->nested.msrs.cr4_fixed1 = X86_CR4_PCE;
7114 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do { \
7115 if (entry && (entry->_reg & (_cpuid_mask))) \
7116 vmx->nested.msrs.cr4_fixed1 |= (_cr4_mask); \
7119 entry = kvm_find_cpuid_entry(vcpu, 0x1, 0);
7120 cr4_fixed1_update(X86_CR4_VME, edx, feature_bit(VME));
7121 cr4_fixed1_update(X86_CR4_PVI, edx, feature_bit(VME));
7122 cr4_fixed1_update(X86_CR4_TSD, edx, feature_bit(TSC));
7123 cr4_fixed1_update(X86_CR4_DE, edx, feature_bit(DE));
7124 cr4_fixed1_update(X86_CR4_PSE, edx, feature_bit(PSE));
7125 cr4_fixed1_update(X86_CR4_PAE, edx, feature_bit(PAE));
7126 cr4_fixed1_update(X86_CR4_MCE, edx, feature_bit(MCE));
7127 cr4_fixed1_update(X86_CR4_PGE, edx, feature_bit(PGE));
7128 cr4_fixed1_update(X86_CR4_OSFXSR, edx, feature_bit(FXSR));
7129 cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, feature_bit(XMM));
7130 cr4_fixed1_update(X86_CR4_VMXE, ecx, feature_bit(VMX));
7131 cr4_fixed1_update(X86_CR4_SMXE, ecx, feature_bit(SMX));
7132 cr4_fixed1_update(X86_CR4_PCIDE, ecx, feature_bit(PCID));
7133 cr4_fixed1_update(X86_CR4_OSXSAVE, ecx, feature_bit(XSAVE));
7135 entry = kvm_find_cpuid_entry(vcpu, 0x7, 0);
7136 cr4_fixed1_update(X86_CR4_FSGSBASE, ebx, feature_bit(FSGSBASE));
7137 cr4_fixed1_update(X86_CR4_SMEP, ebx, feature_bit(SMEP));
7138 cr4_fixed1_update(X86_CR4_SMAP, ebx, feature_bit(SMAP));
7139 cr4_fixed1_update(X86_CR4_PKE, ecx, feature_bit(PKU));
7140 cr4_fixed1_update(X86_CR4_UMIP, ecx, feature_bit(UMIP));
7141 cr4_fixed1_update(X86_CR4_LA57, ecx, feature_bit(LA57));
7143 #undef cr4_fixed1_update
7146 static void nested_vmx_entry_exit_ctls_update(struct kvm_vcpu *vcpu)
7148 struct vcpu_vmx *vmx = to_vmx(vcpu);
7150 if (kvm_mpx_supported()) {
7151 bool mpx_enabled = guest_cpuid_has(vcpu, X86_FEATURE_MPX);
7154 vmx->nested.msrs.entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
7155 vmx->nested.msrs.exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
7157 vmx->nested.msrs.entry_ctls_high &= ~VM_ENTRY_LOAD_BNDCFGS;
7158 vmx->nested.msrs.exit_ctls_high &= ~VM_EXIT_CLEAR_BNDCFGS;
7163 static void update_intel_pt_cfg(struct kvm_vcpu *vcpu)
7165 struct vcpu_vmx *vmx = to_vmx(vcpu);
7166 struct kvm_cpuid_entry2 *best = NULL;
7169 for (i = 0; i < PT_CPUID_LEAVES; i++) {
7170 best = kvm_find_cpuid_entry(vcpu, 0x14, i);
7173 vmx->pt_desc.caps[CPUID_EAX + i*PT_CPUID_REGS_NUM] = best->eax;
7174 vmx->pt_desc.caps[CPUID_EBX + i*PT_CPUID_REGS_NUM] = best->ebx;
7175 vmx->pt_desc.caps[CPUID_ECX + i*PT_CPUID_REGS_NUM] = best->ecx;
7176 vmx->pt_desc.caps[CPUID_EDX + i*PT_CPUID_REGS_NUM] = best->edx;
7179 /* Get the number of configurable Address Ranges for filtering */
7180 vmx->pt_desc.addr_range = intel_pt_validate_cap(vmx->pt_desc.caps,
7181 PT_CAP_num_address_ranges);
7183 /* Initialize and clear the no dependency bits */
7184 vmx->pt_desc.ctl_bitmask = ~(RTIT_CTL_TRACEEN | RTIT_CTL_OS |
7185 RTIT_CTL_USR | RTIT_CTL_TSC_EN | RTIT_CTL_DISRETC);
7188 * If CPUID.(EAX=14H,ECX=0):EBX[0]=1 CR3Filter can be set otherwise
7189 * will inject an #GP
7191 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_cr3_filtering))
7192 vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_CR3EN;
7195 * If CPUID.(EAX=14H,ECX=0):EBX[1]=1 CYCEn, CycThresh and
7196 * PSBFreq can be set
7198 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc))
7199 vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_CYCLEACC |
7200 RTIT_CTL_CYC_THRESH | RTIT_CTL_PSB_FREQ);
7203 * If CPUID.(EAX=14H,ECX=0):EBX[3]=1 MTCEn BranchEn and
7204 * MTCFreq can be set
7206 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc))
7207 vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_MTC_EN |
7208 RTIT_CTL_BRANCH_EN | RTIT_CTL_MTC_RANGE);
7210 /* If CPUID.(EAX=14H,ECX=0):EBX[4]=1 FUPonPTW and PTWEn can be set */
7211 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_ptwrite))
7212 vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_FUP_ON_PTW |
7215 /* If CPUID.(EAX=14H,ECX=0):EBX[5]=1 PwrEvEn can be set */
7216 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_power_event_trace))
7217 vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_PWR_EVT_EN;
7219 /* If CPUID.(EAX=14H,ECX=0):ECX[0]=1 ToPA can be set */
7220 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_topa_output))
7221 vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_TOPA;
7223 /* If CPUID.(EAX=14H,ECX=0):ECX[3]=1 FabircEn can be set */
7224 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_output_subsys))
7225 vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_FABRIC_EN;
7227 /* unmask address range configure area */
7228 for (i = 0; i < vmx->pt_desc.addr_range; i++)
7229 vmx->pt_desc.ctl_bitmask &= ~(0xfULL << (32 + i * 4));
7232 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
7234 struct vcpu_vmx *vmx = to_vmx(vcpu);
7236 /* xsaves_enabled is recomputed in vmx_compute_secondary_exec_control(). */
7237 vcpu->arch.xsaves_enabled = false;
7239 if (cpu_has_secondary_exec_ctrls()) {
7240 vmx_compute_secondary_exec_control(vmx);
7241 vmcs_set_secondary_exec_control(vmx);
7244 if (nested_vmx_allowed(vcpu))
7245 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7246 FEAT_CTL_VMX_ENABLED_INSIDE_SMX |
7247 FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX;
7249 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7250 ~(FEAT_CTL_VMX_ENABLED_INSIDE_SMX |
7251 FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX);
7253 if (nested_vmx_allowed(vcpu)) {
7254 nested_vmx_cr_fixed1_bits_update(vcpu);
7255 nested_vmx_entry_exit_ctls_update(vcpu);
7258 if (boot_cpu_has(X86_FEATURE_INTEL_PT) &&
7259 guest_cpuid_has(vcpu, X86_FEATURE_INTEL_PT))
7260 update_intel_pt_cfg(vcpu);
7262 if (boot_cpu_has(X86_FEATURE_RTM)) {
7263 struct shared_msr_entry *msr;
7264 msr = find_msr_entry(vmx, MSR_IA32_TSX_CTRL);
7266 bool enabled = guest_cpuid_has(vcpu, X86_FEATURE_RTM);
7267 vmx_set_guest_msr(vmx, msr, enabled ? 0 : TSX_CTRL_RTM_DISABLE);
7272 static __init void vmx_set_cpu_caps(void)
7278 kvm_cpu_cap_set(X86_FEATURE_VMX);
7281 if (kvm_mpx_supported())
7282 kvm_cpu_cap_check_and_set(X86_FEATURE_MPX);
7283 if (cpu_has_vmx_invpcid())
7284 kvm_cpu_cap_check_and_set(X86_FEATURE_INVPCID);
7285 if (vmx_pt_mode_is_host_guest())
7286 kvm_cpu_cap_check_and_set(X86_FEATURE_INTEL_PT);
7288 /* PKU is not yet implemented for shadow paging. */
7289 if (enable_ept && boot_cpu_has(X86_FEATURE_OSPKE))
7290 kvm_cpu_cap_check_and_set(X86_FEATURE_PKU);
7292 if (vmx_umip_emulated())
7293 kvm_cpu_cap_set(X86_FEATURE_UMIP);
7297 if (!vmx_xsaves_supported())
7298 kvm_cpu_cap_clear(X86_FEATURE_XSAVES);
7300 /* CPUID 0x80000001 */
7301 if (!cpu_has_vmx_rdtscp())
7302 kvm_cpu_cap_clear(X86_FEATURE_RDTSCP);
7305 static void vmx_request_immediate_exit(struct kvm_vcpu *vcpu)
7307 to_vmx(vcpu)->req_immediate_exit = true;
7310 static int vmx_check_intercept_io(struct kvm_vcpu *vcpu,
7311 struct x86_instruction_info *info)
7313 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7314 unsigned short port;
7318 if (info->intercept == x86_intercept_in ||
7319 info->intercept == x86_intercept_ins) {
7320 port = info->src_val;
7321 size = info->dst_bytes;
7323 port = info->dst_val;
7324 size = info->src_bytes;
7328 * If the 'use IO bitmaps' VM-execution control is 0, IO instruction
7329 * VM-exits depend on the 'unconditional IO exiting' VM-execution
7332 * Otherwise, IO instruction VM-exits are controlled by the IO bitmaps.
7334 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
7335 intercept = nested_cpu_has(vmcs12,
7336 CPU_BASED_UNCOND_IO_EXITING);
7338 intercept = nested_vmx_check_io_bitmaps(vcpu, port, size);
7340 /* FIXME: produce nested vmexit and return X86EMUL_INTERCEPTED. */
7341 return intercept ? X86EMUL_UNHANDLEABLE : X86EMUL_CONTINUE;
7344 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
7345 struct x86_instruction_info *info,
7346 enum x86_intercept_stage stage,
7347 struct x86_exception *exception)
7349 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7351 switch (info->intercept) {
7353 * RDPID causes #UD if disabled through secondary execution controls.
7354 * Because it is marked as EmulateOnUD, we need to intercept it here.
7356 case x86_intercept_rdtscp:
7357 if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDTSCP)) {
7358 exception->vector = UD_VECTOR;
7359 exception->error_code_valid = false;
7360 return X86EMUL_PROPAGATE_FAULT;
7364 case x86_intercept_in:
7365 case x86_intercept_ins:
7366 case x86_intercept_out:
7367 case x86_intercept_outs:
7368 return vmx_check_intercept_io(vcpu, info);
7370 case x86_intercept_lgdt:
7371 case x86_intercept_lidt:
7372 case x86_intercept_lldt:
7373 case x86_intercept_ltr:
7374 case x86_intercept_sgdt:
7375 case x86_intercept_sidt:
7376 case x86_intercept_sldt:
7377 case x86_intercept_str:
7378 if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC))
7379 return X86EMUL_CONTINUE;
7381 /* FIXME: produce nested vmexit and return X86EMUL_INTERCEPTED. */
7384 /* TODO: check more intercepts... */
7389 return X86EMUL_UNHANDLEABLE;
7392 #ifdef CONFIG_X86_64
7393 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
7394 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
7395 u64 divisor, u64 *result)
7397 u64 low = a << shift, high = a >> (64 - shift);
7399 /* To avoid the overflow on divq */
7400 if (high >= divisor)
7403 /* Low hold the result, high hold rem which is discarded */
7404 asm("divq %2\n\t" : "=a" (low), "=d" (high) :
7405 "rm" (divisor), "0" (low), "1" (high));
7411 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc,
7414 struct vcpu_vmx *vmx;
7415 u64 tscl, guest_tscl, delta_tsc, lapic_timer_advance_cycles;
7416 struct kvm_timer *ktimer = &vcpu->arch.apic->lapic_timer;
7420 guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
7421 delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
7422 lapic_timer_advance_cycles = nsec_to_cycles(vcpu,
7423 ktimer->timer_advance_ns);
7425 if (delta_tsc > lapic_timer_advance_cycles)
7426 delta_tsc -= lapic_timer_advance_cycles;
7430 /* Convert to host delta tsc if tsc scaling is enabled */
7431 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
7432 delta_tsc && u64_shl_div_u64(delta_tsc,
7433 kvm_tsc_scaling_ratio_frac_bits,
7434 vcpu->arch.tsc_scaling_ratio, &delta_tsc))
7438 * If the delta tsc can't fit in the 32 bit after the multi shift,
7439 * we can't use the preemption timer.
7440 * It's possible that it fits on later vmentries, but checking
7441 * on every vmentry is costly so we just use an hrtimer.
7443 if (delta_tsc >> (cpu_preemption_timer_multi + 32))
7446 vmx->hv_deadline_tsc = tscl + delta_tsc;
7447 *expired = !delta_tsc;
7451 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
7453 to_vmx(vcpu)->hv_deadline_tsc = -1;
7457 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
7459 if (!kvm_pause_in_guest(vcpu->kvm))
7460 shrink_ple_window(vcpu);
7463 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
7464 struct kvm_memory_slot *slot)
7466 if (!kvm_dirty_log_manual_protect_and_init_set(kvm))
7467 kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
7468 kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
7471 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
7472 struct kvm_memory_slot *slot)
7474 kvm_mmu_slot_set_dirty(kvm, slot);
7477 static void vmx_flush_log_dirty(struct kvm *kvm)
7479 kvm_flush_pml_buffers(kvm);
7482 static int vmx_write_pml_buffer(struct kvm_vcpu *vcpu)
7484 struct vmcs12 *vmcs12;
7485 struct vcpu_vmx *vmx = to_vmx(vcpu);
7488 if (is_guest_mode(vcpu)) {
7489 WARN_ON_ONCE(vmx->nested.pml_full);
7492 * Check if PML is enabled for the nested guest.
7493 * Whether eptp bit 6 is set is already checked
7494 * as part of A/D emulation.
7496 vmcs12 = get_vmcs12(vcpu);
7497 if (!nested_cpu_has_pml(vmcs12))
7500 if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
7501 vmx->nested.pml_full = true;
7505 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS) & ~0xFFFull;
7506 dst = vmcs12->pml_address + sizeof(u64) * vmcs12->guest_pml_index;
7508 if (kvm_write_guest_page(vcpu->kvm, gpa_to_gfn(dst), &gpa,
7509 offset_in_page(dst), sizeof(gpa)))
7512 vmcs12->guest_pml_index--;
7518 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
7519 struct kvm_memory_slot *memslot,
7520 gfn_t offset, unsigned long mask)
7522 kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
7525 static void __pi_post_block(struct kvm_vcpu *vcpu)
7527 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7528 struct pi_desc old, new;
7532 old.control = new.control = pi_desc->control;
7533 WARN(old.nv != POSTED_INTR_WAKEUP_VECTOR,
7534 "Wakeup handler not enabled while the VCPU is blocked\n");
7536 dest = cpu_physical_id(vcpu->cpu);
7538 if (x2apic_enabled())
7541 new.ndst = (dest << 8) & 0xFF00;
7543 /* set 'NV' to 'notification vector' */
7544 new.nv = POSTED_INTR_VECTOR;
7545 } while (cmpxchg64(&pi_desc->control, old.control,
7546 new.control) != old.control);
7548 if (!WARN_ON_ONCE(vcpu->pre_pcpu == -1)) {
7549 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7550 list_del(&vcpu->blocked_vcpu_list);
7551 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7552 vcpu->pre_pcpu = -1;
7557 * This routine does the following things for vCPU which is going
7558 * to be blocked if VT-d PI is enabled.
7559 * - Store the vCPU to the wakeup list, so when interrupts happen
7560 * we can find the right vCPU to wake up.
7561 * - Change the Posted-interrupt descriptor as below:
7562 * 'NDST' <-- vcpu->pre_pcpu
7563 * 'NV' <-- POSTED_INTR_WAKEUP_VECTOR
7564 * - If 'ON' is set during this process, which means at least one
7565 * interrupt is posted for this vCPU, we cannot block it, in
7566 * this case, return 1, otherwise, return 0.
7569 static int pi_pre_block(struct kvm_vcpu *vcpu)
7572 struct pi_desc old, new;
7573 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7575 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
7576 !irq_remapping_cap(IRQ_POSTING_CAP) ||
7577 !kvm_vcpu_apicv_active(vcpu))
7580 WARN_ON(irqs_disabled());
7581 local_irq_disable();
7582 if (!WARN_ON_ONCE(vcpu->pre_pcpu != -1)) {
7583 vcpu->pre_pcpu = vcpu->cpu;
7584 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7585 list_add_tail(&vcpu->blocked_vcpu_list,
7586 &per_cpu(blocked_vcpu_on_cpu,
7588 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7592 old.control = new.control = pi_desc->control;
7594 WARN((pi_desc->sn == 1),
7595 "Warning: SN field of posted-interrupts "
7596 "is set before blocking\n");
7599 * Since vCPU can be preempted during this process,
7600 * vcpu->cpu could be different with pre_pcpu, we
7601 * need to set pre_pcpu as the destination of wakeup
7602 * notification event, then we can find the right vCPU
7603 * to wakeup in wakeup handler if interrupts happen
7604 * when the vCPU is in blocked state.
7606 dest = cpu_physical_id(vcpu->pre_pcpu);
7608 if (x2apic_enabled())
7611 new.ndst = (dest << 8) & 0xFF00;
7613 /* set 'NV' to 'wakeup vector' */
7614 new.nv = POSTED_INTR_WAKEUP_VECTOR;
7615 } while (cmpxchg64(&pi_desc->control, old.control,
7616 new.control) != old.control);
7618 /* We should not block the vCPU if an interrupt is posted for it. */
7619 if (pi_test_on(pi_desc) == 1)
7620 __pi_post_block(vcpu);
7623 return (vcpu->pre_pcpu == -1);
7626 static int vmx_pre_block(struct kvm_vcpu *vcpu)
7628 if (pi_pre_block(vcpu))
7631 if (kvm_lapic_hv_timer_in_use(vcpu))
7632 kvm_lapic_switch_to_sw_timer(vcpu);
7637 static void pi_post_block(struct kvm_vcpu *vcpu)
7639 if (vcpu->pre_pcpu == -1)
7642 WARN_ON(irqs_disabled());
7643 local_irq_disable();
7644 __pi_post_block(vcpu);
7648 static void vmx_post_block(struct kvm_vcpu *vcpu)
7650 if (kvm_x86_ops.set_hv_timer)
7651 kvm_lapic_switch_to_hv_timer(vcpu);
7653 pi_post_block(vcpu);
7657 * vmx_update_pi_irte - set IRTE for Posted-Interrupts
7660 * @host_irq: host irq of the interrupt
7661 * @guest_irq: gsi of the interrupt
7662 * @set: set or unset PI
7663 * returns 0 on success, < 0 on failure
7665 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
7666 uint32_t guest_irq, bool set)
7668 struct kvm_kernel_irq_routing_entry *e;
7669 struct kvm_irq_routing_table *irq_rt;
7670 struct kvm_lapic_irq irq;
7671 struct kvm_vcpu *vcpu;
7672 struct vcpu_data vcpu_info;
7675 if (!kvm_arch_has_assigned_device(kvm) ||
7676 !irq_remapping_cap(IRQ_POSTING_CAP) ||
7677 !kvm_vcpu_apicv_active(kvm->vcpus[0]))
7680 idx = srcu_read_lock(&kvm->irq_srcu);
7681 irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
7682 if (guest_irq >= irq_rt->nr_rt_entries ||
7683 hlist_empty(&irq_rt->map[guest_irq])) {
7684 pr_warn_once("no route for guest_irq %u/%u (broken user space?)\n",
7685 guest_irq, irq_rt->nr_rt_entries);
7689 hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
7690 if (e->type != KVM_IRQ_ROUTING_MSI)
7693 * VT-d PI cannot support posting multicast/broadcast
7694 * interrupts to a vCPU, we still use interrupt remapping
7695 * for these kind of interrupts.
7697 * For lowest-priority interrupts, we only support
7698 * those with single CPU as the destination, e.g. user
7699 * configures the interrupts via /proc/irq or uses
7700 * irqbalance to make the interrupts single-CPU.
7702 * We will support full lowest-priority interrupt later.
7704 * In addition, we can only inject generic interrupts using
7705 * the PI mechanism, refuse to route others through it.
7708 kvm_set_msi_irq(kvm, e, &irq);
7709 if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu) ||
7710 !kvm_irq_is_postable(&irq)) {
7712 * Make sure the IRTE is in remapped mode if
7713 * we don't handle it in posted mode.
7715 ret = irq_set_vcpu_affinity(host_irq, NULL);
7718 "failed to back to remapped mode, irq: %u\n",
7726 vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
7727 vcpu_info.vector = irq.vector;
7729 trace_kvm_pi_irte_update(host_irq, vcpu->vcpu_id, e->gsi,
7730 vcpu_info.vector, vcpu_info.pi_desc_addr, set);
7733 ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
7735 ret = irq_set_vcpu_affinity(host_irq, NULL);
7738 printk(KERN_INFO "%s: failed to update PI IRTE\n",
7746 srcu_read_unlock(&kvm->irq_srcu, idx);
7750 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
7752 if (vcpu->arch.mcg_cap & MCG_LMCE_P)
7753 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7754 FEAT_CTL_LMCE_ENABLED;
7756 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7757 ~FEAT_CTL_LMCE_ENABLED;
7760 static bool vmx_smi_allowed(struct kvm_vcpu *vcpu, bool for_injection)
7762 /* we need a nested vmexit to enter SMM, postpone if run is pending */
7763 if (to_vmx(vcpu)->nested.nested_run_pending)
7765 return !is_smm(vcpu);
7768 static int vmx_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
7770 struct vcpu_vmx *vmx = to_vmx(vcpu);
7772 vmx->nested.smm.guest_mode = is_guest_mode(vcpu);
7773 if (vmx->nested.smm.guest_mode)
7774 nested_vmx_vmexit(vcpu, -1, 0, 0);
7776 vmx->nested.smm.vmxon = vmx->nested.vmxon;
7777 vmx->nested.vmxon = false;
7778 vmx_clear_hlt(vcpu);
7782 static int vmx_pre_leave_smm(struct kvm_vcpu *vcpu, const char *smstate)
7784 struct vcpu_vmx *vmx = to_vmx(vcpu);
7787 if (vmx->nested.smm.vmxon) {
7788 vmx->nested.vmxon = true;
7789 vmx->nested.smm.vmxon = false;
7792 if (vmx->nested.smm.guest_mode) {
7793 ret = nested_vmx_enter_non_root_mode(vcpu, false);
7797 vmx->nested.smm.guest_mode = false;
7802 static int enable_smi_window(struct kvm_vcpu *vcpu)
7807 static bool vmx_need_emulation_on_page_fault(struct kvm_vcpu *vcpu)
7812 static bool vmx_apic_init_signal_blocked(struct kvm_vcpu *vcpu)
7814 return to_vmx(vcpu)->nested.vmxon;
7817 static void hardware_unsetup(void)
7820 nested_vmx_hardware_unsetup();
7825 static bool vmx_check_apicv_inhibit_reasons(ulong bit)
7827 ulong supported = BIT(APICV_INHIBIT_REASON_DISABLE) |
7828 BIT(APICV_INHIBIT_REASON_HYPERV);
7830 return supported & BIT(bit);
7833 static struct kvm_x86_ops vmx_x86_ops __initdata = {
7834 .hardware_unsetup = hardware_unsetup,
7836 .hardware_enable = hardware_enable,
7837 .hardware_disable = hardware_disable,
7838 .cpu_has_accelerated_tpr = report_flexpriority,
7839 .has_emulated_msr = vmx_has_emulated_msr,
7841 .vm_size = sizeof(struct kvm_vmx),
7842 .vm_init = vmx_vm_init,
7844 .vcpu_create = vmx_create_vcpu,
7845 .vcpu_free = vmx_free_vcpu,
7846 .vcpu_reset = vmx_vcpu_reset,
7848 .prepare_guest_switch = vmx_prepare_switch_to_guest,
7849 .vcpu_load = vmx_vcpu_load,
7850 .vcpu_put = vmx_vcpu_put,
7852 .update_bp_intercept = update_exception_bitmap,
7853 .get_msr_feature = vmx_get_msr_feature,
7854 .get_msr = vmx_get_msr,
7855 .set_msr = vmx_set_msr,
7856 .get_segment_base = vmx_get_segment_base,
7857 .get_segment = vmx_get_segment,
7858 .set_segment = vmx_set_segment,
7859 .get_cpl = vmx_get_cpl,
7860 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
7861 .set_cr0 = vmx_set_cr0,
7862 .set_cr4 = vmx_set_cr4,
7863 .set_efer = vmx_set_efer,
7864 .get_idt = vmx_get_idt,
7865 .set_idt = vmx_set_idt,
7866 .get_gdt = vmx_get_gdt,
7867 .set_gdt = vmx_set_gdt,
7868 .set_dr7 = vmx_set_dr7,
7869 .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
7870 .cache_reg = vmx_cache_reg,
7871 .get_rflags = vmx_get_rflags,
7872 .set_rflags = vmx_set_rflags,
7874 .tlb_flush_all = vmx_flush_tlb_all,
7875 .tlb_flush_current = vmx_flush_tlb_current,
7876 .tlb_flush_gva = vmx_flush_tlb_gva,
7877 .tlb_flush_guest = vmx_flush_tlb_guest,
7879 .run = vmx_vcpu_run,
7880 .handle_exit = vmx_handle_exit,
7881 .skip_emulated_instruction = vmx_skip_emulated_instruction,
7882 .update_emulated_instruction = vmx_update_emulated_instruction,
7883 .set_interrupt_shadow = vmx_set_interrupt_shadow,
7884 .get_interrupt_shadow = vmx_get_interrupt_shadow,
7885 .patch_hypercall = vmx_patch_hypercall,
7886 .set_irq = vmx_inject_irq,
7887 .set_nmi = vmx_inject_nmi,
7888 .queue_exception = vmx_queue_exception,
7889 .cancel_injection = vmx_cancel_injection,
7890 .interrupt_allowed = vmx_interrupt_allowed,
7891 .nmi_allowed = vmx_nmi_allowed,
7892 .get_nmi_mask = vmx_get_nmi_mask,
7893 .set_nmi_mask = vmx_set_nmi_mask,
7894 .enable_nmi_window = enable_nmi_window,
7895 .enable_irq_window = enable_irq_window,
7896 .update_cr8_intercept = update_cr8_intercept,
7897 .set_virtual_apic_mode = vmx_set_virtual_apic_mode,
7898 .set_apic_access_page_addr = vmx_set_apic_access_page_addr,
7899 .refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
7900 .load_eoi_exitmap = vmx_load_eoi_exitmap,
7901 .apicv_post_state_restore = vmx_apicv_post_state_restore,
7902 .check_apicv_inhibit_reasons = vmx_check_apicv_inhibit_reasons,
7903 .hwapic_irr_update = vmx_hwapic_irr_update,
7904 .hwapic_isr_update = vmx_hwapic_isr_update,
7905 .guest_apic_has_interrupt = vmx_guest_apic_has_interrupt,
7906 .sync_pir_to_irr = vmx_sync_pir_to_irr,
7907 .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
7908 .dy_apicv_has_pending_interrupt = vmx_dy_apicv_has_pending_interrupt,
7910 .set_tss_addr = vmx_set_tss_addr,
7911 .set_identity_map_addr = vmx_set_identity_map_addr,
7912 .get_tdp_level = vmx_get_tdp_level,
7913 .get_mt_mask = vmx_get_mt_mask,
7915 .get_exit_info = vmx_get_exit_info,
7917 .cpuid_update = vmx_cpuid_update,
7919 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
7921 .write_l1_tsc_offset = vmx_write_l1_tsc_offset,
7923 .load_mmu_pgd = vmx_load_mmu_pgd,
7925 .check_intercept = vmx_check_intercept,
7926 .handle_exit_irqoff = vmx_handle_exit_irqoff,
7928 .request_immediate_exit = vmx_request_immediate_exit,
7930 .sched_in = vmx_sched_in,
7932 .slot_enable_log_dirty = vmx_slot_enable_log_dirty,
7933 .slot_disable_log_dirty = vmx_slot_disable_log_dirty,
7934 .flush_log_dirty = vmx_flush_log_dirty,
7935 .enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
7936 .write_log_dirty = vmx_write_pml_buffer,
7938 .pre_block = vmx_pre_block,
7939 .post_block = vmx_post_block,
7941 .pmu_ops = &intel_pmu_ops,
7942 .nested_ops = &vmx_nested_ops,
7944 .update_pi_irte = vmx_update_pi_irte,
7946 #ifdef CONFIG_X86_64
7947 .set_hv_timer = vmx_set_hv_timer,
7948 .cancel_hv_timer = vmx_cancel_hv_timer,
7951 .setup_mce = vmx_setup_mce,
7953 .smi_allowed = vmx_smi_allowed,
7954 .pre_enter_smm = vmx_pre_enter_smm,
7955 .pre_leave_smm = vmx_pre_leave_smm,
7956 .enable_smi_window = enable_smi_window,
7958 .need_emulation_on_page_fault = vmx_need_emulation_on_page_fault,
7959 .apic_init_signal_blocked = vmx_apic_init_signal_blocked,
7962 static __init int hardware_setup(void)
7964 unsigned long host_bndcfgs;
7966 int r, i, ept_lpage_level;
7969 host_idt_base = dt.address;
7971 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
7972 kvm_define_shared_msr(i, vmx_msr_index[i]);
7974 if (setup_vmcs_config(&vmcs_config, &vmx_capability) < 0)
7977 if (boot_cpu_has(X86_FEATURE_NX))
7978 kvm_enable_efer_bits(EFER_NX);
7980 if (boot_cpu_has(X86_FEATURE_MPX)) {
7981 rdmsrl(MSR_IA32_BNDCFGS, host_bndcfgs);
7982 WARN_ONCE(host_bndcfgs, "KVM: BNDCFGS in host will be lost");
7985 if (!cpu_has_vmx_mpx())
7986 supported_xcr0 &= ~(XFEATURE_MASK_BNDREGS |
7987 XFEATURE_MASK_BNDCSR);
7989 if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
7990 !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
7993 if (!cpu_has_vmx_ept() ||
7994 !cpu_has_vmx_ept_4levels() ||
7995 !cpu_has_vmx_ept_mt_wb() ||
7996 !cpu_has_vmx_invept_global())
7999 if (!cpu_has_vmx_ept_ad_bits() || !enable_ept)
8000 enable_ept_ad_bits = 0;
8002 if (!cpu_has_vmx_unrestricted_guest() || !enable_ept)
8003 enable_unrestricted_guest = 0;
8005 if (!cpu_has_vmx_flexpriority())
8006 flexpriority_enabled = 0;
8008 if (!cpu_has_virtual_nmis())
8012 * set_apic_access_page_addr() is used to reload apic access
8013 * page upon invalidation. No need to do anything if not
8014 * using the APIC_ACCESS_ADDR VMCS field.
8016 if (!flexpriority_enabled)
8017 vmx_x86_ops.set_apic_access_page_addr = NULL;
8019 if (!cpu_has_vmx_tpr_shadow())
8020 vmx_x86_ops.update_cr8_intercept = NULL;
8022 #if IS_ENABLED(CONFIG_HYPERV)
8023 if (ms_hyperv.nested_features & HV_X64_NESTED_GUEST_MAPPING_FLUSH
8025 vmx_x86_ops.tlb_remote_flush = hv_remote_flush_tlb;
8026 vmx_x86_ops.tlb_remote_flush_with_range =
8027 hv_remote_flush_tlb_with_range;
8031 if (!cpu_has_vmx_ple()) {
8034 ple_window_grow = 0;
8036 ple_window_shrink = 0;
8039 if (!cpu_has_vmx_apicv()) {
8041 vmx_x86_ops.sync_pir_to_irr = NULL;
8044 if (cpu_has_vmx_tsc_scaling()) {
8045 kvm_has_tsc_control = true;
8046 kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
8047 kvm_tsc_scaling_ratio_frac_bits = 48;
8050 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
8056 ept_lpage_level = 0;
8057 else if (cpu_has_vmx_ept_1g_page())
8058 ept_lpage_level = PG_LEVEL_1G;
8059 else if (cpu_has_vmx_ept_2m_page())
8060 ept_lpage_level = PG_LEVEL_2M;
8062 ept_lpage_level = PG_LEVEL_4K;
8063 kvm_configure_mmu(enable_ept, ept_lpage_level);
8066 * Only enable PML when hardware supports PML feature, and both EPT
8067 * and EPT A/D bit features are enabled -- PML depends on them to work.
8069 if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
8073 vmx_x86_ops.slot_enable_log_dirty = NULL;
8074 vmx_x86_ops.slot_disable_log_dirty = NULL;
8075 vmx_x86_ops.flush_log_dirty = NULL;
8076 vmx_x86_ops.enable_log_dirty_pt_masked = NULL;
8079 if (!cpu_has_vmx_preemption_timer())
8080 enable_preemption_timer = false;
8082 if (enable_preemption_timer) {
8083 u64 use_timer_freq = 5000ULL * 1000 * 1000;
8086 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
8087 cpu_preemption_timer_multi =
8088 vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
8091 use_timer_freq = (u64)tsc_khz * 1000;
8092 use_timer_freq >>= cpu_preemption_timer_multi;
8095 * KVM "disables" the preemption timer by setting it to its max
8096 * value. Don't use the timer if it might cause spurious exits
8097 * at a rate faster than 0.1 Hz (of uninterrupted guest time).
8099 if (use_timer_freq > 0xffffffffu / 10)
8100 enable_preemption_timer = false;
8103 if (!enable_preemption_timer) {
8104 vmx_x86_ops.set_hv_timer = NULL;
8105 vmx_x86_ops.cancel_hv_timer = NULL;
8106 vmx_x86_ops.request_immediate_exit = __kvm_request_immediate_exit;
8109 kvm_set_posted_intr_wakeup_handler(wakeup_handler);
8111 kvm_mce_cap_supported |= MCG_LMCE_P;
8113 if (pt_mode != PT_MODE_SYSTEM && pt_mode != PT_MODE_HOST_GUEST)
8115 if (!enable_ept || !cpu_has_vmx_intel_pt())
8116 pt_mode = PT_MODE_SYSTEM;
8119 nested_vmx_setup_ctls_msrs(&vmcs_config.nested,
8120 vmx_capability.ept);
8122 r = nested_vmx_hardware_setup(&vmx_x86_ops,
8123 kvm_vmx_exit_handlers);
8130 r = alloc_kvm_area();
8132 nested_vmx_hardware_unsetup();
8136 static struct kvm_x86_init_ops vmx_init_ops __initdata = {
8137 .cpu_has_kvm_support = cpu_has_kvm_support,
8138 .disabled_by_bios = vmx_disabled_by_bios,
8139 .check_processor_compatibility = vmx_check_processor_compat,
8140 .hardware_setup = hardware_setup,
8142 .runtime_ops = &vmx_x86_ops,
8145 static void vmx_cleanup_l1d_flush(void)
8147 if (vmx_l1d_flush_pages) {
8148 free_pages((unsigned long)vmx_l1d_flush_pages, L1D_CACHE_ORDER);
8149 vmx_l1d_flush_pages = NULL;
8151 /* Restore state so sysfs ignores VMX */
8152 l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
8155 static void vmx_exit(void)
8157 #ifdef CONFIG_KEXEC_CORE
8158 RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
8164 #if IS_ENABLED(CONFIG_HYPERV)
8165 if (static_branch_unlikely(&enable_evmcs)) {
8167 struct hv_vp_assist_page *vp_ap;
8169 * Reset everything to support using non-enlightened VMCS
8170 * access later (e.g. when we reload the module with
8171 * enlightened_vmcs=0)
8173 for_each_online_cpu(cpu) {
8174 vp_ap = hv_get_vp_assist_page(cpu);
8179 vp_ap->nested_control.features.directhypercall = 0;
8180 vp_ap->current_nested_vmcs = 0;
8181 vp_ap->enlighten_vmentry = 0;
8184 static_branch_disable(&enable_evmcs);
8187 vmx_cleanup_l1d_flush();
8189 module_exit(vmx_exit);
8191 static int __init vmx_init(void)
8195 #if IS_ENABLED(CONFIG_HYPERV)
8197 * Enlightened VMCS usage should be recommended and the host needs
8198 * to support eVMCS v1 or above. We can also disable eVMCS support
8199 * with module parameter.
8201 if (enlightened_vmcs &&
8202 ms_hyperv.hints & HV_X64_ENLIGHTENED_VMCS_RECOMMENDED &&
8203 (ms_hyperv.nested_features & HV_X64_ENLIGHTENED_VMCS_VERSION) >=
8204 KVM_EVMCS_VERSION) {
8207 /* Check that we have assist pages on all online CPUs */
8208 for_each_online_cpu(cpu) {
8209 if (!hv_get_vp_assist_page(cpu)) {
8210 enlightened_vmcs = false;
8215 if (enlightened_vmcs) {
8216 pr_info("KVM: vmx: using Hyper-V Enlightened VMCS\n");
8217 static_branch_enable(&enable_evmcs);
8220 if (ms_hyperv.nested_features & HV_X64_NESTED_DIRECT_FLUSH)
8221 vmx_x86_ops.enable_direct_tlbflush
8222 = hv_enable_direct_tlbflush;
8225 enlightened_vmcs = false;
8229 r = kvm_init(&vmx_init_ops, sizeof(struct vcpu_vmx),
8230 __alignof__(struct vcpu_vmx), THIS_MODULE);
8235 * Must be called after kvm_init() so enable_ept is properly set
8236 * up. Hand the parameter mitigation value in which was stored in
8237 * the pre module init parser. If no parameter was given, it will
8238 * contain 'auto' which will be turned into the default 'cond'
8241 r = vmx_setup_l1d_flush(vmentry_l1d_flush_param);
8247 for_each_possible_cpu(cpu) {
8248 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
8249 INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
8250 spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
8253 #ifdef CONFIG_KEXEC_CORE
8254 rcu_assign_pointer(crash_vmclear_loaded_vmcss,
8255 crash_vmclear_local_loaded_vmcss);
8257 vmx_check_vmcs12_offsets();
8261 module_init(vmx_init);