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 unsigned long host_idt_base;
443 * Though SYSCALL is only supported in 64-bit mode on Intel CPUs, kvm
444 * will emulate SYSCALL in legacy mode if the vendor string in guest
445 * CPUID.0:{EBX,ECX,EDX} is "AuthenticAMD" or "AMDisbetter!" To
446 * support this emulation, IA32_STAR must always be included in
447 * vmx_msr_index[], even in i386 builds.
449 const u32 vmx_msr_index[] = {
451 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
453 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
457 #if IS_ENABLED(CONFIG_HYPERV)
458 static bool __read_mostly enlightened_vmcs = true;
459 module_param(enlightened_vmcs, bool, 0444);
461 /* check_ept_pointer() should be under protection of ept_pointer_lock. */
462 static void check_ept_pointer_match(struct kvm *kvm)
464 struct kvm_vcpu *vcpu;
465 u64 tmp_eptp = INVALID_PAGE;
468 kvm_for_each_vcpu(i, vcpu, kvm) {
469 if (!VALID_PAGE(tmp_eptp)) {
470 tmp_eptp = to_vmx(vcpu)->ept_pointer;
471 } else if (tmp_eptp != to_vmx(vcpu)->ept_pointer) {
472 to_kvm_vmx(kvm)->ept_pointers_match
473 = EPT_POINTERS_MISMATCH;
478 to_kvm_vmx(kvm)->ept_pointers_match = EPT_POINTERS_MATCH;
481 static int kvm_fill_hv_flush_list_func(struct hv_guest_mapping_flush_list *flush,
484 struct kvm_tlb_range *range = data;
486 return hyperv_fill_flush_guest_mapping_list(flush, range->start_gfn,
490 static inline int __hv_remote_flush_tlb_with_range(struct kvm *kvm,
491 struct kvm_vcpu *vcpu, struct kvm_tlb_range *range)
493 u64 ept_pointer = to_vmx(vcpu)->ept_pointer;
496 * FLUSH_GUEST_PHYSICAL_ADDRESS_SPACE hypercall needs address
497 * of the base of EPT PML4 table, strip off EPT configuration
501 return hyperv_flush_guest_mapping_range(ept_pointer & PAGE_MASK,
502 kvm_fill_hv_flush_list_func, (void *)range);
504 return hyperv_flush_guest_mapping(ept_pointer & PAGE_MASK);
507 static int hv_remote_flush_tlb_with_range(struct kvm *kvm,
508 struct kvm_tlb_range *range)
510 struct kvm_vcpu *vcpu;
513 spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
515 if (to_kvm_vmx(kvm)->ept_pointers_match == EPT_POINTERS_CHECK)
516 check_ept_pointer_match(kvm);
518 if (to_kvm_vmx(kvm)->ept_pointers_match != EPT_POINTERS_MATCH) {
519 kvm_for_each_vcpu(i, vcpu, kvm) {
520 /* If ept_pointer is invalid pointer, bypass flush request. */
521 if (VALID_PAGE(to_vmx(vcpu)->ept_pointer))
522 ret |= __hv_remote_flush_tlb_with_range(
526 ret = __hv_remote_flush_tlb_with_range(kvm,
527 kvm_get_vcpu(kvm, 0), range);
530 spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
533 static int hv_remote_flush_tlb(struct kvm *kvm)
535 return hv_remote_flush_tlb_with_range(kvm, NULL);
538 static int hv_enable_direct_tlbflush(struct kvm_vcpu *vcpu)
540 struct hv_enlightened_vmcs *evmcs;
541 struct hv_partition_assist_pg **p_hv_pa_pg =
542 &vcpu->kvm->arch.hyperv.hv_pa_pg;
544 * Synthetic VM-Exit is not enabled in current code and so All
545 * evmcs in singe VM shares same assist page.
548 *p_hv_pa_pg = kzalloc(PAGE_SIZE, GFP_KERNEL);
553 evmcs = (struct hv_enlightened_vmcs *)to_vmx(vcpu)->loaded_vmcs->vmcs;
555 evmcs->partition_assist_page =
557 evmcs->hv_vm_id = (unsigned long)vcpu->kvm;
558 evmcs->hv_enlightenments_control.nested_flush_hypercall = 1;
563 #endif /* IS_ENABLED(CONFIG_HYPERV) */
566 * Comment's format: document - errata name - stepping - processor name.
568 * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
570 static u32 vmx_preemption_cpu_tfms[] = {
571 /* 323344.pdf - BA86 - D0 - Xeon 7500 Series */
573 /* 323056.pdf - AAX65 - C2 - Xeon L3406 */
574 /* 322814.pdf - AAT59 - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
575 /* 322911.pdf - AAU65 - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
577 /* 322911.pdf - AAU65 - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
579 /* 322373.pdf - AAO95 - B1 - Xeon 3400 Series */
580 /* 322166.pdf - AAN92 - B1 - i7-800 and i5-700 Desktop */
582 * 320767.pdf - AAP86 - B1 -
583 * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
586 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
588 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
590 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
592 /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
593 /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
594 /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
596 /* Xeon E3-1220 V2 */
600 static inline bool cpu_has_broken_vmx_preemption_timer(void)
602 u32 eax = cpuid_eax(0x00000001), i;
604 /* Clear the reserved bits */
605 eax &= ~(0x3U << 14 | 0xfU << 28);
606 for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
607 if (eax == vmx_preemption_cpu_tfms[i])
613 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
615 return flexpriority_enabled && lapic_in_kernel(vcpu);
618 static inline bool report_flexpriority(void)
620 return flexpriority_enabled;
623 static inline int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
627 for (i = 0; i < vmx->nmsrs; ++i)
628 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
633 struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
637 i = __find_msr_index(vmx, msr);
639 return &vmx->guest_msrs[i];
643 static int vmx_set_guest_msr(struct vcpu_vmx *vmx, struct shared_msr_entry *msr, u64 data)
647 u64 old_msr_data = msr->data;
649 if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
651 ret = kvm_set_shared_msr(msr->index, msr->data,
655 msr->data = old_msr_data;
660 #ifdef CONFIG_KEXEC_CORE
661 static void crash_vmclear_local_loaded_vmcss(void)
663 int cpu = raw_smp_processor_id();
664 struct loaded_vmcs *v;
666 list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
667 loaded_vmcss_on_cpu_link)
670 #endif /* CONFIG_KEXEC_CORE */
672 static void __loaded_vmcs_clear(void *arg)
674 struct loaded_vmcs *loaded_vmcs = arg;
675 int cpu = raw_smp_processor_id();
677 if (loaded_vmcs->cpu != cpu)
678 return; /* vcpu migration can race with cpu offline */
679 if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
680 per_cpu(current_vmcs, cpu) = NULL;
682 vmcs_clear(loaded_vmcs->vmcs);
683 if (loaded_vmcs->shadow_vmcs && loaded_vmcs->launched)
684 vmcs_clear(loaded_vmcs->shadow_vmcs);
686 list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
689 * Ensure all writes to loaded_vmcs, including deleting it from its
690 * current percpu list, complete before setting loaded_vmcs->vcpu to
691 * -1, otherwise a different cpu can see vcpu == -1 first and add
692 * loaded_vmcs to its percpu list before it's deleted from this cpu's
693 * list. Pairs with the smp_rmb() in vmx_vcpu_load_vmcs().
697 loaded_vmcs->cpu = -1;
698 loaded_vmcs->launched = 0;
701 void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
703 int cpu = loaded_vmcs->cpu;
706 smp_call_function_single(cpu,
707 __loaded_vmcs_clear, loaded_vmcs, 1);
710 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
714 u32 mask = 1 << (seg * SEG_FIELD_NR + field);
716 if (!kvm_register_is_available(&vmx->vcpu, VCPU_EXREG_SEGMENTS)) {
717 kvm_register_mark_available(&vmx->vcpu, VCPU_EXREG_SEGMENTS);
718 vmx->segment_cache.bitmask = 0;
720 ret = vmx->segment_cache.bitmask & mask;
721 vmx->segment_cache.bitmask |= mask;
725 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
727 u16 *p = &vmx->segment_cache.seg[seg].selector;
729 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
730 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
734 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
736 ulong *p = &vmx->segment_cache.seg[seg].base;
738 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
739 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
743 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
745 u32 *p = &vmx->segment_cache.seg[seg].limit;
747 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
748 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
752 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
754 u32 *p = &vmx->segment_cache.seg[seg].ar;
756 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
757 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
761 void update_exception_bitmap(struct kvm_vcpu *vcpu)
765 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
766 (1u << DB_VECTOR) | (1u << AC_VECTOR);
768 * Guest access to VMware backdoor ports could legitimately
769 * trigger #GP because of TSS I/O permission bitmap.
770 * We intercept those #GP and allow access to them anyway
773 if (enable_vmware_backdoor)
774 eb |= (1u << GP_VECTOR);
775 if ((vcpu->guest_debug &
776 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
777 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
778 eb |= 1u << BP_VECTOR;
779 if (to_vmx(vcpu)->rmode.vm86_active)
782 eb &= ~(1u << PF_VECTOR);
784 /* When we are running a nested L2 guest and L1 specified for it a
785 * certain exception bitmap, we must trap the same exceptions and pass
786 * them to L1. When running L2, we will only handle the exceptions
787 * specified above if L1 did not want them.
789 if (is_guest_mode(vcpu))
790 eb |= get_vmcs12(vcpu)->exception_bitmap;
792 vmcs_write32(EXCEPTION_BITMAP, eb);
796 * Check if MSR is intercepted for currently loaded MSR bitmap.
798 static bool msr_write_intercepted(struct kvm_vcpu *vcpu, u32 msr)
800 unsigned long *msr_bitmap;
801 int f = sizeof(unsigned long);
803 if (!cpu_has_vmx_msr_bitmap())
806 msr_bitmap = to_vmx(vcpu)->loaded_vmcs->msr_bitmap;
809 return !!test_bit(msr, msr_bitmap + 0x800 / f);
810 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
812 return !!test_bit(msr, msr_bitmap + 0xc00 / f);
818 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
819 unsigned long entry, unsigned long exit)
821 vm_entry_controls_clearbit(vmx, entry);
822 vm_exit_controls_clearbit(vmx, exit);
825 int vmx_find_msr_index(struct vmx_msrs *m, u32 msr)
829 for (i = 0; i < m->nr; ++i) {
830 if (m->val[i].index == msr)
836 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
839 struct msr_autoload *m = &vmx->msr_autoload;
843 if (cpu_has_load_ia32_efer()) {
844 clear_atomic_switch_msr_special(vmx,
845 VM_ENTRY_LOAD_IA32_EFER,
846 VM_EXIT_LOAD_IA32_EFER);
850 case MSR_CORE_PERF_GLOBAL_CTRL:
851 if (cpu_has_load_perf_global_ctrl()) {
852 clear_atomic_switch_msr_special(vmx,
853 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
854 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
859 i = vmx_find_msr_index(&m->guest, msr);
863 m->guest.val[i] = m->guest.val[m->guest.nr];
864 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
867 i = vmx_find_msr_index(&m->host, msr);
872 m->host.val[i] = m->host.val[m->host.nr];
873 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
876 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
877 unsigned long entry, unsigned long exit,
878 unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
879 u64 guest_val, u64 host_val)
881 vmcs_write64(guest_val_vmcs, guest_val);
882 if (host_val_vmcs != HOST_IA32_EFER)
883 vmcs_write64(host_val_vmcs, host_val);
884 vm_entry_controls_setbit(vmx, entry);
885 vm_exit_controls_setbit(vmx, exit);
888 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
889 u64 guest_val, u64 host_val, bool entry_only)
892 struct msr_autoload *m = &vmx->msr_autoload;
896 if (cpu_has_load_ia32_efer()) {
897 add_atomic_switch_msr_special(vmx,
898 VM_ENTRY_LOAD_IA32_EFER,
899 VM_EXIT_LOAD_IA32_EFER,
902 guest_val, host_val);
906 case MSR_CORE_PERF_GLOBAL_CTRL:
907 if (cpu_has_load_perf_global_ctrl()) {
908 add_atomic_switch_msr_special(vmx,
909 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
910 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
911 GUEST_IA32_PERF_GLOBAL_CTRL,
912 HOST_IA32_PERF_GLOBAL_CTRL,
913 guest_val, host_val);
917 case MSR_IA32_PEBS_ENABLE:
918 /* PEBS needs a quiescent period after being disabled (to write
919 * a record). Disabling PEBS through VMX MSR swapping doesn't
920 * provide that period, so a CPU could write host's record into
923 wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
926 i = vmx_find_msr_index(&m->guest, msr);
928 j = vmx_find_msr_index(&m->host, msr);
930 if ((i < 0 && m->guest.nr == NR_LOADSTORE_MSRS) ||
931 (j < 0 && m->host.nr == NR_LOADSTORE_MSRS)) {
932 printk_once(KERN_WARNING "Not enough msr switch entries. "
933 "Can't add msr %x\n", msr);
938 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
940 m->guest.val[i].index = msr;
941 m->guest.val[i].value = guest_val;
948 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
950 m->host.val[j].index = msr;
951 m->host.val[j].value = host_val;
954 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
956 u64 guest_efer = vmx->vcpu.arch.efer;
959 /* Shadow paging assumes NX to be available. */
961 guest_efer |= EFER_NX;
964 * LMA and LME handled by hardware; SCE meaningless outside long mode.
966 ignore_bits |= EFER_SCE;
968 ignore_bits |= EFER_LMA | EFER_LME;
969 /* SCE is meaningful only in long mode on Intel */
970 if (guest_efer & EFER_LMA)
971 ignore_bits &= ~(u64)EFER_SCE;
975 * On EPT, we can't emulate NX, so we must switch EFER atomically.
976 * On CPUs that support "load IA32_EFER", always switch EFER
977 * atomically, since it's faster than switching it manually.
979 if (cpu_has_load_ia32_efer() ||
980 (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
981 if (!(guest_efer & EFER_LMA))
982 guest_efer &= ~EFER_LME;
983 if (guest_efer != host_efer)
984 add_atomic_switch_msr(vmx, MSR_EFER,
985 guest_efer, host_efer, false);
987 clear_atomic_switch_msr(vmx, MSR_EFER);
990 clear_atomic_switch_msr(vmx, MSR_EFER);
992 guest_efer &= ~ignore_bits;
993 guest_efer |= host_efer & ignore_bits;
995 vmx->guest_msrs[efer_offset].data = guest_efer;
996 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1002 #ifdef CONFIG_X86_32
1004 * On 32-bit kernels, VM exits still load the FS and GS bases from the
1005 * VMCS rather than the segment table. KVM uses this helper to figure
1006 * out the current bases to poke them into the VMCS before entry.
1008 static unsigned long segment_base(u16 selector)
1010 struct desc_struct *table;
1013 if (!(selector & ~SEGMENT_RPL_MASK))
1016 table = get_current_gdt_ro();
1018 if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
1019 u16 ldt_selector = kvm_read_ldt();
1021 if (!(ldt_selector & ~SEGMENT_RPL_MASK))
1024 table = (struct desc_struct *)segment_base(ldt_selector);
1026 v = get_desc_base(&table[selector >> 3]);
1031 static inline bool pt_can_write_msr(struct vcpu_vmx *vmx)
1033 return vmx_pt_mode_is_host_guest() &&
1034 !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
1037 static inline void pt_load_msr(struct pt_ctx *ctx, u32 addr_range)
1041 wrmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1042 wrmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1043 wrmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1044 wrmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1045 for (i = 0; i < addr_range; i++) {
1046 wrmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1047 wrmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1051 static inline void pt_save_msr(struct pt_ctx *ctx, u32 addr_range)
1055 rdmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1056 rdmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1057 rdmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1058 rdmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1059 for (i = 0; i < addr_range; i++) {
1060 rdmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1061 rdmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1065 static void pt_guest_enter(struct vcpu_vmx *vmx)
1067 if (vmx_pt_mode_is_system())
1071 * GUEST_IA32_RTIT_CTL is already set in the VMCS.
1072 * Save host state before VM entry.
1074 rdmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1075 if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1076 wrmsrl(MSR_IA32_RTIT_CTL, 0);
1077 pt_save_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1078 pt_load_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1082 static void pt_guest_exit(struct vcpu_vmx *vmx)
1084 if (vmx_pt_mode_is_system())
1087 if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1088 pt_save_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1089 pt_load_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1092 /* Reload host state (IA32_RTIT_CTL will be cleared on VM exit). */
1093 wrmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1096 void vmx_set_host_fs_gs(struct vmcs_host_state *host, u16 fs_sel, u16 gs_sel,
1097 unsigned long fs_base, unsigned long gs_base)
1099 if (unlikely(fs_sel != host->fs_sel)) {
1101 vmcs_write16(HOST_FS_SELECTOR, fs_sel);
1103 vmcs_write16(HOST_FS_SELECTOR, 0);
1104 host->fs_sel = fs_sel;
1106 if (unlikely(gs_sel != host->gs_sel)) {
1108 vmcs_write16(HOST_GS_SELECTOR, gs_sel);
1110 vmcs_write16(HOST_GS_SELECTOR, 0);
1111 host->gs_sel = gs_sel;
1113 if (unlikely(fs_base != host->fs_base)) {
1114 vmcs_writel(HOST_FS_BASE, fs_base);
1115 host->fs_base = fs_base;
1117 if (unlikely(gs_base != host->gs_base)) {
1118 vmcs_writel(HOST_GS_BASE, gs_base);
1119 host->gs_base = gs_base;
1123 void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu)
1125 struct vcpu_vmx *vmx = to_vmx(vcpu);
1126 struct vmcs_host_state *host_state;
1127 #ifdef CONFIG_X86_64
1128 int cpu = raw_smp_processor_id();
1130 unsigned long fs_base, gs_base;
1134 vmx->req_immediate_exit = false;
1137 * Note that guest MSRs to be saved/restored can also be changed
1138 * when guest state is loaded. This happens when guest transitions
1139 * to/from long-mode by setting MSR_EFER.LMA.
1141 if (!vmx->guest_msrs_ready) {
1142 vmx->guest_msrs_ready = true;
1143 for (i = 0; i < vmx->save_nmsrs; ++i)
1144 kvm_set_shared_msr(vmx->guest_msrs[i].index,
1145 vmx->guest_msrs[i].data,
1146 vmx->guest_msrs[i].mask);
1150 if (vmx->nested.need_vmcs12_to_shadow_sync)
1151 nested_sync_vmcs12_to_shadow(vcpu);
1153 if (vmx->guest_state_loaded)
1156 host_state = &vmx->loaded_vmcs->host_state;
1159 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
1160 * allow segment selectors with cpl > 0 or ti == 1.
1162 host_state->ldt_sel = kvm_read_ldt();
1164 #ifdef CONFIG_X86_64
1165 savesegment(ds, host_state->ds_sel);
1166 savesegment(es, host_state->es_sel);
1168 gs_base = cpu_kernelmode_gs_base(cpu);
1169 if (likely(is_64bit_mm(current->mm))) {
1170 save_fsgs_for_kvm();
1171 fs_sel = current->thread.fsindex;
1172 gs_sel = current->thread.gsindex;
1173 fs_base = current->thread.fsbase;
1174 vmx->msr_host_kernel_gs_base = current->thread.gsbase;
1176 savesegment(fs, fs_sel);
1177 savesegment(gs, gs_sel);
1178 fs_base = read_msr(MSR_FS_BASE);
1179 vmx->msr_host_kernel_gs_base = read_msr(MSR_KERNEL_GS_BASE);
1182 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1184 savesegment(fs, fs_sel);
1185 savesegment(gs, gs_sel);
1186 fs_base = segment_base(fs_sel);
1187 gs_base = segment_base(gs_sel);
1190 vmx_set_host_fs_gs(host_state, fs_sel, gs_sel, fs_base, gs_base);
1191 vmx->guest_state_loaded = true;
1194 static void vmx_prepare_switch_to_host(struct vcpu_vmx *vmx)
1196 struct vmcs_host_state *host_state;
1198 if (!vmx->guest_state_loaded)
1201 host_state = &vmx->loaded_vmcs->host_state;
1203 ++vmx->vcpu.stat.host_state_reload;
1205 #ifdef CONFIG_X86_64
1206 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1208 if (host_state->ldt_sel || (host_state->gs_sel & 7)) {
1209 kvm_load_ldt(host_state->ldt_sel);
1210 #ifdef CONFIG_X86_64
1211 load_gs_index(host_state->gs_sel);
1213 loadsegment(gs, host_state->gs_sel);
1216 if (host_state->fs_sel & 7)
1217 loadsegment(fs, host_state->fs_sel);
1218 #ifdef CONFIG_X86_64
1219 if (unlikely(host_state->ds_sel | host_state->es_sel)) {
1220 loadsegment(ds, host_state->ds_sel);
1221 loadsegment(es, host_state->es_sel);
1224 invalidate_tss_limit();
1225 #ifdef CONFIG_X86_64
1226 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1228 load_fixmap_gdt(raw_smp_processor_id());
1229 vmx->guest_state_loaded = false;
1230 vmx->guest_msrs_ready = false;
1233 #ifdef CONFIG_X86_64
1234 static u64 vmx_read_guest_kernel_gs_base(struct vcpu_vmx *vmx)
1237 if (vmx->guest_state_loaded)
1238 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1240 return vmx->msr_guest_kernel_gs_base;
1243 static void vmx_write_guest_kernel_gs_base(struct vcpu_vmx *vmx, u64 data)
1246 if (vmx->guest_state_loaded)
1247 wrmsrl(MSR_KERNEL_GS_BASE, data);
1249 vmx->msr_guest_kernel_gs_base = data;
1253 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
1255 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1256 struct pi_desc old, new;
1260 * In case of hot-plug or hot-unplug, we may have to undo
1261 * vmx_vcpu_pi_put even if there is no assigned device. And we
1262 * always keep PI.NDST up to date for simplicity: it makes the
1263 * code easier, and CPU migration is not a fast path.
1265 if (!pi_test_sn(pi_desc) && vcpu->cpu == cpu)
1269 * If the 'nv' field is POSTED_INTR_WAKEUP_VECTOR, do not change
1270 * PI.NDST: pi_post_block is the one expected to change PID.NDST and the
1271 * wakeup handler expects the vCPU to be on the blocked_vcpu_list that
1272 * matches PI.NDST. Otherwise, a vcpu may not be able to be woken up
1275 if (pi_desc->nv == POSTED_INTR_WAKEUP_VECTOR || vcpu->cpu == cpu) {
1276 pi_clear_sn(pi_desc);
1277 goto after_clear_sn;
1280 /* The full case. */
1282 old.control = new.control = pi_desc->control;
1284 dest = cpu_physical_id(cpu);
1286 if (x2apic_enabled())
1289 new.ndst = (dest << 8) & 0xFF00;
1292 } while (cmpxchg64(&pi_desc->control, old.control,
1293 new.control) != old.control);
1298 * Clear SN before reading the bitmap. The VT-d firmware
1299 * writes the bitmap and reads SN atomically (5.2.3 in the
1300 * spec), so it doesn't really have a memory barrier that
1301 * pairs with this, but we cannot do that and we need one.
1303 smp_mb__after_atomic();
1305 if (!pi_is_pir_empty(pi_desc))
1309 void vmx_vcpu_load_vmcs(struct kvm_vcpu *vcpu, int cpu)
1311 struct vcpu_vmx *vmx = to_vmx(vcpu);
1312 bool already_loaded = vmx->loaded_vmcs->cpu == cpu;
1314 if (!already_loaded) {
1315 loaded_vmcs_clear(vmx->loaded_vmcs);
1316 local_irq_disable();
1319 * Ensure loaded_vmcs->cpu is read before adding loaded_vmcs to
1320 * this cpu's percpu list, otherwise it may not yet be deleted
1321 * from its previous cpu's percpu list. Pairs with the
1322 * smb_wmb() in __loaded_vmcs_clear().
1326 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1327 &per_cpu(loaded_vmcss_on_cpu, cpu));
1331 if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
1332 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1333 vmcs_load(vmx->loaded_vmcs->vmcs);
1334 indirect_branch_prediction_barrier();
1337 if (!already_loaded) {
1338 void *gdt = get_current_gdt_ro();
1339 unsigned long sysenter_esp;
1342 * Flush all EPTP/VPID contexts, the new pCPU may have stale
1343 * TLB entries from its previous association with the vCPU.
1345 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1348 * Linux uses per-cpu TSS and GDT, so set these when switching
1349 * processors. See 22.2.4.
1351 vmcs_writel(HOST_TR_BASE,
1352 (unsigned long)&get_cpu_entry_area(cpu)->tss.x86_tss);
1353 vmcs_writel(HOST_GDTR_BASE, (unsigned long)gdt); /* 22.2.4 */
1355 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1356 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1358 vmx->loaded_vmcs->cpu = cpu;
1361 /* Setup TSC multiplier */
1362 if (kvm_has_tsc_control &&
1363 vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio)
1364 decache_tsc_multiplier(vmx);
1368 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1369 * vcpu mutex is already taken.
1371 void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1373 struct vcpu_vmx *vmx = to_vmx(vcpu);
1375 vmx_vcpu_load_vmcs(vcpu, cpu);
1377 vmx_vcpu_pi_load(vcpu, cpu);
1379 vmx->host_pkru = read_pkru();
1380 vmx->host_debugctlmsr = get_debugctlmsr();
1383 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
1385 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1387 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
1388 !irq_remapping_cap(IRQ_POSTING_CAP) ||
1389 !kvm_vcpu_apicv_active(vcpu))
1392 /* Set SN when the vCPU is preempted */
1393 if (vcpu->preempted)
1397 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1399 vmx_vcpu_pi_put(vcpu);
1401 vmx_prepare_switch_to_host(to_vmx(vcpu));
1404 static bool emulation_required(struct kvm_vcpu *vcpu)
1406 return emulate_invalid_guest_state && !guest_state_valid(vcpu);
1409 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1411 struct vcpu_vmx *vmx = to_vmx(vcpu);
1412 unsigned long rflags, save_rflags;
1414 if (!kvm_register_is_available(vcpu, VCPU_EXREG_RFLAGS)) {
1415 kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1416 rflags = vmcs_readl(GUEST_RFLAGS);
1417 if (vmx->rmode.vm86_active) {
1418 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1419 save_rflags = vmx->rmode.save_rflags;
1420 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1422 vmx->rflags = rflags;
1427 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1429 struct vcpu_vmx *vmx = to_vmx(vcpu);
1430 unsigned long old_rflags;
1432 if (enable_unrestricted_guest) {
1433 kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1434 vmx->rflags = rflags;
1435 vmcs_writel(GUEST_RFLAGS, rflags);
1439 old_rflags = vmx_get_rflags(vcpu);
1440 vmx->rflags = rflags;
1441 if (vmx->rmode.vm86_active) {
1442 vmx->rmode.save_rflags = rflags;
1443 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1445 vmcs_writel(GUEST_RFLAGS, rflags);
1447 if ((old_rflags ^ vmx->rflags) & X86_EFLAGS_VM)
1448 vmx->emulation_required = emulation_required(vcpu);
1451 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
1453 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1456 if (interruptibility & GUEST_INTR_STATE_STI)
1457 ret |= KVM_X86_SHADOW_INT_STI;
1458 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1459 ret |= KVM_X86_SHADOW_INT_MOV_SS;
1464 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1466 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1467 u32 interruptibility = interruptibility_old;
1469 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1471 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1472 interruptibility |= GUEST_INTR_STATE_MOV_SS;
1473 else if (mask & KVM_X86_SHADOW_INT_STI)
1474 interruptibility |= GUEST_INTR_STATE_STI;
1476 if ((interruptibility != interruptibility_old))
1477 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1480 static int vmx_rtit_ctl_check(struct kvm_vcpu *vcpu, u64 data)
1482 struct vcpu_vmx *vmx = to_vmx(vcpu);
1483 unsigned long value;
1486 * Any MSR write that attempts to change bits marked reserved will
1489 if (data & vmx->pt_desc.ctl_bitmask)
1493 * Any attempt to modify IA32_RTIT_CTL while TraceEn is set will
1494 * result in a #GP unless the same write also clears TraceEn.
1496 if ((vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) &&
1497 ((vmx->pt_desc.guest.ctl ^ data) & ~RTIT_CTL_TRACEEN))
1501 * WRMSR to IA32_RTIT_CTL that sets TraceEn but clears this bit
1502 * and FabricEn would cause #GP, if
1503 * CPUID.(EAX=14H, ECX=0):ECX.SNGLRGNOUT[bit 2] = 0
1505 if ((data & RTIT_CTL_TRACEEN) && !(data & RTIT_CTL_TOPA) &&
1506 !(data & RTIT_CTL_FABRIC_EN) &&
1507 !intel_pt_validate_cap(vmx->pt_desc.caps,
1508 PT_CAP_single_range_output))
1512 * MTCFreq, CycThresh and PSBFreq encodings check, any MSR write that
1513 * utilize encodings marked reserved will casue a #GP fault.
1515 value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc_periods);
1516 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc) &&
1517 !test_bit((data & RTIT_CTL_MTC_RANGE) >>
1518 RTIT_CTL_MTC_RANGE_OFFSET, &value))
1520 value = intel_pt_validate_cap(vmx->pt_desc.caps,
1521 PT_CAP_cycle_thresholds);
1522 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1523 !test_bit((data & RTIT_CTL_CYC_THRESH) >>
1524 RTIT_CTL_CYC_THRESH_OFFSET, &value))
1526 value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_periods);
1527 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1528 !test_bit((data & RTIT_CTL_PSB_FREQ) >>
1529 RTIT_CTL_PSB_FREQ_OFFSET, &value))
1533 * If ADDRx_CFG is reserved or the encodings is >2 will
1534 * cause a #GP fault.
1536 value = (data & RTIT_CTL_ADDR0) >> RTIT_CTL_ADDR0_OFFSET;
1537 if ((value && (vmx->pt_desc.addr_range < 1)) || (value > 2))
1539 value = (data & RTIT_CTL_ADDR1) >> RTIT_CTL_ADDR1_OFFSET;
1540 if ((value && (vmx->pt_desc.addr_range < 2)) || (value > 2))
1542 value = (data & RTIT_CTL_ADDR2) >> RTIT_CTL_ADDR2_OFFSET;
1543 if ((value && (vmx->pt_desc.addr_range < 3)) || (value > 2))
1545 value = (data & RTIT_CTL_ADDR3) >> RTIT_CTL_ADDR3_OFFSET;
1546 if ((value && (vmx->pt_desc.addr_range < 4)) || (value > 2))
1552 static int skip_emulated_instruction(struct kvm_vcpu *vcpu)
1557 * Using VMCS.VM_EXIT_INSTRUCTION_LEN on EPT misconfig depends on
1558 * undefined behavior: Intel's SDM doesn't mandate the VMCS field be
1559 * set when EPT misconfig occurs. In practice, real hardware updates
1560 * VM_EXIT_INSTRUCTION_LEN on EPT misconfig, but other hypervisors
1561 * (namely Hyper-V) don't set it due to it being undefined behavior,
1562 * i.e. we end up advancing IP with some random value.
1564 if (!static_cpu_has(X86_FEATURE_HYPERVISOR) ||
1565 to_vmx(vcpu)->exit_reason != EXIT_REASON_EPT_MISCONFIG) {
1566 rip = kvm_rip_read(vcpu);
1567 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1568 kvm_rip_write(vcpu, rip);
1570 if (!kvm_emulate_instruction(vcpu, EMULTYPE_SKIP))
1574 /* skipping an emulated instruction also counts */
1575 vmx_set_interrupt_shadow(vcpu, 0);
1582 * Recognizes a pending MTF VM-exit and records the nested state for later
1585 static void vmx_update_emulated_instruction(struct kvm_vcpu *vcpu)
1587 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1588 struct vcpu_vmx *vmx = to_vmx(vcpu);
1590 if (!is_guest_mode(vcpu))
1594 * Per the SDM, MTF takes priority over debug-trap exceptions besides
1595 * T-bit traps. As instruction emulation is completed (i.e. at the
1596 * instruction boundary), any #DB exception pending delivery must be a
1597 * debug-trap. Record the pending MTF state to be delivered in
1598 * vmx_check_nested_events().
1600 if (nested_cpu_has_mtf(vmcs12) &&
1601 (!vcpu->arch.exception.pending ||
1602 vcpu->arch.exception.nr == DB_VECTOR))
1603 vmx->nested.mtf_pending = true;
1605 vmx->nested.mtf_pending = false;
1608 static int vmx_skip_emulated_instruction(struct kvm_vcpu *vcpu)
1610 vmx_update_emulated_instruction(vcpu);
1611 return skip_emulated_instruction(vcpu);
1614 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1617 * Ensure that we clear the HLT state in the VMCS. We don't need to
1618 * explicitly skip the instruction because if the HLT state is set,
1619 * then the instruction is already executing and RIP has already been
1622 if (kvm_hlt_in_guest(vcpu->kvm) &&
1623 vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1624 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1627 static void vmx_queue_exception(struct kvm_vcpu *vcpu)
1629 struct vcpu_vmx *vmx = to_vmx(vcpu);
1630 unsigned nr = vcpu->arch.exception.nr;
1631 bool has_error_code = vcpu->arch.exception.has_error_code;
1632 u32 error_code = vcpu->arch.exception.error_code;
1633 u32 intr_info = nr | INTR_INFO_VALID_MASK;
1635 kvm_deliver_exception_payload(vcpu);
1637 if (has_error_code) {
1638 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1639 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1642 if (vmx->rmode.vm86_active) {
1644 if (kvm_exception_is_soft(nr))
1645 inc_eip = vcpu->arch.event_exit_inst_len;
1646 kvm_inject_realmode_interrupt(vcpu, nr, inc_eip);
1650 WARN_ON_ONCE(vmx->emulation_required);
1652 if (kvm_exception_is_soft(nr)) {
1653 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1654 vmx->vcpu.arch.event_exit_inst_len);
1655 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1657 intr_info |= INTR_TYPE_HARD_EXCEPTION;
1659 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1661 vmx_clear_hlt(vcpu);
1665 * Swap MSR entry in host/guest MSR entry array.
1667 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1669 struct shared_msr_entry tmp;
1671 tmp = vmx->guest_msrs[to];
1672 vmx->guest_msrs[to] = vmx->guest_msrs[from];
1673 vmx->guest_msrs[from] = tmp;
1677 * Set up the vmcs to automatically save and restore system
1678 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
1679 * mode, as fiddling with msrs is very expensive.
1681 static void setup_msrs(struct vcpu_vmx *vmx)
1683 int save_nmsrs, index;
1686 #ifdef CONFIG_X86_64
1688 * The SYSCALL MSRs are only needed on long mode guests, and only
1689 * when EFER.SCE is set.
1691 if (is_long_mode(&vmx->vcpu) && (vmx->vcpu.arch.efer & EFER_SCE)) {
1692 index = __find_msr_index(vmx, MSR_STAR);
1694 move_msr_up(vmx, index, save_nmsrs++);
1695 index = __find_msr_index(vmx, MSR_LSTAR);
1697 move_msr_up(vmx, index, save_nmsrs++);
1698 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1700 move_msr_up(vmx, index, save_nmsrs++);
1703 index = __find_msr_index(vmx, MSR_EFER);
1704 if (index >= 0 && update_transition_efer(vmx, index))
1705 move_msr_up(vmx, index, save_nmsrs++);
1706 index = __find_msr_index(vmx, MSR_TSC_AUX);
1707 if (index >= 0 && guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDTSCP))
1708 move_msr_up(vmx, index, save_nmsrs++);
1709 index = __find_msr_index(vmx, MSR_IA32_TSX_CTRL);
1711 move_msr_up(vmx, index, save_nmsrs++);
1713 vmx->save_nmsrs = save_nmsrs;
1714 vmx->guest_msrs_ready = false;
1716 if (cpu_has_vmx_msr_bitmap())
1717 vmx_update_msr_bitmap(&vmx->vcpu);
1720 static u64 vmx_read_l1_tsc_offset(struct kvm_vcpu *vcpu)
1722 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1724 if (is_guest_mode(vcpu) &&
1725 (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING))
1726 return vcpu->arch.tsc_offset - vmcs12->tsc_offset;
1728 return vcpu->arch.tsc_offset;
1731 static u64 vmx_write_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1733 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1734 u64 g_tsc_offset = 0;
1737 * We're here if L1 chose not to trap WRMSR to TSC. According
1738 * to the spec, this should set L1's TSC; The offset that L1
1739 * set for L2 remains unchanged, and still needs to be added
1740 * to the newly set TSC to get L2's TSC.
1742 if (is_guest_mode(vcpu) &&
1743 (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING))
1744 g_tsc_offset = vmcs12->tsc_offset;
1746 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1747 vcpu->arch.tsc_offset - g_tsc_offset,
1749 vmcs_write64(TSC_OFFSET, offset + g_tsc_offset);
1750 return offset + g_tsc_offset;
1754 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
1755 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
1756 * all guests if the "nested" module option is off, and can also be disabled
1757 * for a single guest by disabling its VMX cpuid bit.
1759 bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
1761 return nested && guest_cpuid_has(vcpu, X86_FEATURE_VMX);
1764 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
1767 uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
1769 return !(val & ~valid_bits);
1772 static int vmx_get_msr_feature(struct kvm_msr_entry *msr)
1774 switch (msr->index) {
1775 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1778 return vmx_get_vmx_msr(&vmcs_config.nested, msr->index, &msr->data);
1785 * Reads an msr value (of 'msr_index') into 'pdata'.
1786 * Returns 0 on success, non-0 otherwise.
1787 * Assumes vcpu_load() was already called.
1789 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1791 struct vcpu_vmx *vmx = to_vmx(vcpu);
1792 struct shared_msr_entry *msr;
1795 switch (msr_info->index) {
1796 #ifdef CONFIG_X86_64
1798 msr_info->data = vmcs_readl(GUEST_FS_BASE);
1801 msr_info->data = vmcs_readl(GUEST_GS_BASE);
1803 case MSR_KERNEL_GS_BASE:
1804 msr_info->data = vmx_read_guest_kernel_gs_base(vmx);
1808 return kvm_get_msr_common(vcpu, msr_info);
1809 case MSR_IA32_TSX_CTRL:
1810 if (!msr_info->host_initiated &&
1811 !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
1813 goto find_shared_msr;
1814 case MSR_IA32_UMWAIT_CONTROL:
1815 if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
1818 msr_info->data = vmx->msr_ia32_umwait_control;
1820 case MSR_IA32_SPEC_CTRL:
1821 if (!msr_info->host_initiated &&
1822 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1825 msr_info->data = to_vmx(vcpu)->spec_ctrl;
1827 case MSR_IA32_SYSENTER_CS:
1828 msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
1830 case MSR_IA32_SYSENTER_EIP:
1831 msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
1833 case MSR_IA32_SYSENTER_ESP:
1834 msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
1836 case MSR_IA32_BNDCFGS:
1837 if (!kvm_mpx_supported() ||
1838 (!msr_info->host_initiated &&
1839 !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1841 msr_info->data = vmcs_read64(GUEST_BNDCFGS);
1843 case MSR_IA32_MCG_EXT_CTL:
1844 if (!msr_info->host_initiated &&
1845 !(vmx->msr_ia32_feature_control &
1846 FEAT_CTL_LMCE_ENABLED))
1848 msr_info->data = vcpu->arch.mcg_ext_ctl;
1850 case MSR_IA32_FEAT_CTL:
1851 msr_info->data = vmx->msr_ia32_feature_control;
1853 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1854 if (!nested_vmx_allowed(vcpu))
1856 if (vmx_get_vmx_msr(&vmx->nested.msrs, msr_info->index,
1860 * Enlightened VMCS v1 doesn't have certain fields, but buggy
1861 * Hyper-V versions are still trying to use corresponding
1862 * features when they are exposed. Filter out the essential
1865 if (!msr_info->host_initiated &&
1866 vmx->nested.enlightened_vmcs_enabled)
1867 nested_evmcs_filter_control_msr(msr_info->index,
1870 case MSR_IA32_RTIT_CTL:
1871 if (!vmx_pt_mode_is_host_guest())
1873 msr_info->data = vmx->pt_desc.guest.ctl;
1875 case MSR_IA32_RTIT_STATUS:
1876 if (!vmx_pt_mode_is_host_guest())
1878 msr_info->data = vmx->pt_desc.guest.status;
1880 case MSR_IA32_RTIT_CR3_MATCH:
1881 if (!vmx_pt_mode_is_host_guest() ||
1882 !intel_pt_validate_cap(vmx->pt_desc.caps,
1883 PT_CAP_cr3_filtering))
1885 msr_info->data = vmx->pt_desc.guest.cr3_match;
1887 case MSR_IA32_RTIT_OUTPUT_BASE:
1888 if (!vmx_pt_mode_is_host_guest() ||
1889 (!intel_pt_validate_cap(vmx->pt_desc.caps,
1890 PT_CAP_topa_output) &&
1891 !intel_pt_validate_cap(vmx->pt_desc.caps,
1892 PT_CAP_single_range_output)))
1894 msr_info->data = vmx->pt_desc.guest.output_base;
1896 case MSR_IA32_RTIT_OUTPUT_MASK:
1897 if (!vmx_pt_mode_is_host_guest() ||
1898 (!intel_pt_validate_cap(vmx->pt_desc.caps,
1899 PT_CAP_topa_output) &&
1900 !intel_pt_validate_cap(vmx->pt_desc.caps,
1901 PT_CAP_single_range_output)))
1903 msr_info->data = vmx->pt_desc.guest.output_mask;
1905 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
1906 index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
1907 if (!vmx_pt_mode_is_host_guest() ||
1908 (index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
1909 PT_CAP_num_address_ranges)))
1912 msr_info->data = vmx->pt_desc.guest.addr_b[index / 2];
1914 msr_info->data = vmx->pt_desc.guest.addr_a[index / 2];
1917 if (!msr_info->host_initiated &&
1918 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
1920 goto find_shared_msr;
1923 msr = find_msr_entry(vmx, msr_info->index);
1925 msr_info->data = msr->data;
1928 return kvm_get_msr_common(vcpu, msr_info);
1935 * Writes msr value into the appropriate "register".
1936 * Returns 0 on success, non-0 otherwise.
1937 * Assumes vcpu_load() was already called.
1939 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1941 struct vcpu_vmx *vmx = to_vmx(vcpu);
1942 struct shared_msr_entry *msr;
1944 u32 msr_index = msr_info->index;
1945 u64 data = msr_info->data;
1948 switch (msr_index) {
1950 ret = kvm_set_msr_common(vcpu, msr_info);
1952 #ifdef CONFIG_X86_64
1954 vmx_segment_cache_clear(vmx);
1955 vmcs_writel(GUEST_FS_BASE, data);
1958 vmx_segment_cache_clear(vmx);
1959 vmcs_writel(GUEST_GS_BASE, data);
1961 case MSR_KERNEL_GS_BASE:
1962 vmx_write_guest_kernel_gs_base(vmx, data);
1965 case MSR_IA32_SYSENTER_CS:
1966 if (is_guest_mode(vcpu))
1967 get_vmcs12(vcpu)->guest_sysenter_cs = data;
1968 vmcs_write32(GUEST_SYSENTER_CS, data);
1970 case MSR_IA32_SYSENTER_EIP:
1971 if (is_guest_mode(vcpu))
1972 get_vmcs12(vcpu)->guest_sysenter_eip = data;
1973 vmcs_writel(GUEST_SYSENTER_EIP, data);
1975 case MSR_IA32_SYSENTER_ESP:
1976 if (is_guest_mode(vcpu))
1977 get_vmcs12(vcpu)->guest_sysenter_esp = data;
1978 vmcs_writel(GUEST_SYSENTER_ESP, data);
1980 case MSR_IA32_DEBUGCTLMSR:
1981 if (is_guest_mode(vcpu) && get_vmcs12(vcpu)->vm_exit_controls &
1982 VM_EXIT_SAVE_DEBUG_CONTROLS)
1983 get_vmcs12(vcpu)->guest_ia32_debugctl = data;
1985 ret = kvm_set_msr_common(vcpu, msr_info);
1988 case MSR_IA32_BNDCFGS:
1989 if (!kvm_mpx_supported() ||
1990 (!msr_info->host_initiated &&
1991 !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1993 if (is_noncanonical_address(data & PAGE_MASK, vcpu) ||
1994 (data & MSR_IA32_BNDCFGS_RSVD))
1996 vmcs_write64(GUEST_BNDCFGS, data);
1998 case MSR_IA32_UMWAIT_CONTROL:
1999 if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
2002 /* The reserved bit 1 and non-32 bit [63:32] should be zero */
2003 if (data & (BIT_ULL(1) | GENMASK_ULL(63, 32)))
2006 vmx->msr_ia32_umwait_control = data;
2008 case MSR_IA32_SPEC_CTRL:
2009 if (!msr_info->host_initiated &&
2010 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
2013 if (data & ~kvm_spec_ctrl_valid_bits(vcpu))
2016 vmx->spec_ctrl = data;
2022 * When it's written (to non-zero) for the first time, pass
2026 * The handling of the MSR bitmap for L2 guests is done in
2027 * nested_vmx_prepare_msr_bitmap. We should not touch the
2028 * vmcs02.msr_bitmap here since it gets completely overwritten
2029 * in the merging. We update the vmcs01 here for L1 as well
2030 * since it will end up touching the MSR anyway now.
2032 vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap,
2036 case MSR_IA32_TSX_CTRL:
2037 if (!msr_info->host_initiated &&
2038 !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
2040 if (data & ~(TSX_CTRL_RTM_DISABLE | TSX_CTRL_CPUID_CLEAR))
2042 goto find_shared_msr;
2043 case MSR_IA32_PRED_CMD:
2044 if (!msr_info->host_initiated &&
2045 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
2048 if (data & ~PRED_CMD_IBPB)
2050 if (!boot_cpu_has(X86_FEATURE_SPEC_CTRL))
2055 wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
2059 * When it's written (to non-zero) for the first time, pass
2063 * The handling of the MSR bitmap for L2 guests is done in
2064 * nested_vmx_prepare_msr_bitmap. We should not touch the
2065 * vmcs02.msr_bitmap here since it gets completely overwritten
2068 vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap, MSR_IA32_PRED_CMD,
2071 case MSR_IA32_CR_PAT:
2072 if (!kvm_pat_valid(data))
2075 if (is_guest_mode(vcpu) &&
2076 get_vmcs12(vcpu)->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
2077 get_vmcs12(vcpu)->guest_ia32_pat = data;
2079 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2080 vmcs_write64(GUEST_IA32_PAT, data);
2081 vcpu->arch.pat = data;
2084 ret = kvm_set_msr_common(vcpu, msr_info);
2086 case MSR_IA32_TSC_ADJUST:
2087 ret = kvm_set_msr_common(vcpu, msr_info);
2089 case MSR_IA32_MCG_EXT_CTL:
2090 if ((!msr_info->host_initiated &&
2091 !(to_vmx(vcpu)->msr_ia32_feature_control &
2092 FEAT_CTL_LMCE_ENABLED)) ||
2093 (data & ~MCG_EXT_CTL_LMCE_EN))
2095 vcpu->arch.mcg_ext_ctl = data;
2097 case MSR_IA32_FEAT_CTL:
2098 if (!vmx_feature_control_msr_valid(vcpu, data) ||
2099 (to_vmx(vcpu)->msr_ia32_feature_control &
2100 FEAT_CTL_LOCKED && !msr_info->host_initiated))
2102 vmx->msr_ia32_feature_control = data;
2103 if (msr_info->host_initiated && data == 0)
2104 vmx_leave_nested(vcpu);
2106 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2107 if (!msr_info->host_initiated)
2108 return 1; /* they are read-only */
2109 if (!nested_vmx_allowed(vcpu))
2111 return vmx_set_vmx_msr(vcpu, msr_index, data);
2112 case MSR_IA32_RTIT_CTL:
2113 if (!vmx_pt_mode_is_host_guest() ||
2114 vmx_rtit_ctl_check(vcpu, data) ||
2117 vmcs_write64(GUEST_IA32_RTIT_CTL, data);
2118 vmx->pt_desc.guest.ctl = data;
2119 pt_update_intercept_for_msr(vmx);
2121 case MSR_IA32_RTIT_STATUS:
2122 if (!pt_can_write_msr(vmx))
2124 if (data & MSR_IA32_RTIT_STATUS_MASK)
2126 vmx->pt_desc.guest.status = data;
2128 case MSR_IA32_RTIT_CR3_MATCH:
2129 if (!pt_can_write_msr(vmx))
2131 if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2132 PT_CAP_cr3_filtering))
2134 vmx->pt_desc.guest.cr3_match = data;
2136 case MSR_IA32_RTIT_OUTPUT_BASE:
2137 if (!pt_can_write_msr(vmx))
2139 if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2140 PT_CAP_topa_output) &&
2141 !intel_pt_validate_cap(vmx->pt_desc.caps,
2142 PT_CAP_single_range_output))
2144 if (data & MSR_IA32_RTIT_OUTPUT_BASE_MASK)
2146 vmx->pt_desc.guest.output_base = data;
2148 case MSR_IA32_RTIT_OUTPUT_MASK:
2149 if (!pt_can_write_msr(vmx))
2151 if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2152 PT_CAP_topa_output) &&
2153 !intel_pt_validate_cap(vmx->pt_desc.caps,
2154 PT_CAP_single_range_output))
2156 vmx->pt_desc.guest.output_mask = data;
2158 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
2159 if (!pt_can_write_msr(vmx))
2161 index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
2162 if (index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
2163 PT_CAP_num_address_ranges))
2165 if (is_noncanonical_address(data, vcpu))
2168 vmx->pt_desc.guest.addr_b[index / 2] = data;
2170 vmx->pt_desc.guest.addr_a[index / 2] = data;
2173 if (!msr_info->host_initiated &&
2174 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
2176 /* Check reserved bit, higher 32 bits should be zero */
2177 if ((data >> 32) != 0)
2179 goto find_shared_msr;
2183 msr = find_msr_entry(vmx, msr_index);
2185 ret = vmx_set_guest_msr(vmx, msr, data);
2187 ret = kvm_set_msr_common(vcpu, msr_info);
2193 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2195 kvm_register_mark_available(vcpu, reg);
2199 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2202 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2204 case VCPU_EXREG_PDPTR:
2206 ept_save_pdptrs(vcpu);
2208 case VCPU_EXREG_CR3:
2209 if (enable_unrestricted_guest || (enable_ept && is_paging(vcpu)))
2210 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2218 static __init int cpu_has_kvm_support(void)
2220 return cpu_has_vmx();
2223 static __init int vmx_disabled_by_bios(void)
2225 return !boot_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
2226 !boot_cpu_has(X86_FEATURE_VMX);
2229 static int kvm_cpu_vmxon(u64 vmxon_pointer)
2233 cr4_set_bits(X86_CR4_VMXE);
2234 intel_pt_handle_vmx(1);
2236 asm_volatile_goto("1: vmxon %[vmxon_pointer]\n\t"
2237 _ASM_EXTABLE(1b, %l[fault])
2238 : : [vmxon_pointer] "m"(vmxon_pointer)
2243 WARN_ONCE(1, "VMXON faulted, MSR_IA32_FEAT_CTL (0x3a) = 0x%llx\n",
2244 rdmsrl_safe(MSR_IA32_FEAT_CTL, &msr) ? 0xdeadbeef : msr);
2245 intel_pt_handle_vmx(0);
2246 cr4_clear_bits(X86_CR4_VMXE);
2251 static int hardware_enable(void)
2253 int cpu = raw_smp_processor_id();
2254 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2257 if (cr4_read_shadow() & X86_CR4_VMXE)
2261 * This can happen if we hot-added a CPU but failed to allocate
2262 * VP assist page for it.
2264 if (static_branch_unlikely(&enable_evmcs) &&
2265 !hv_get_vp_assist_page(cpu))
2268 r = kvm_cpu_vmxon(phys_addr);
2278 static void vmclear_local_loaded_vmcss(void)
2280 int cpu = raw_smp_processor_id();
2281 struct loaded_vmcs *v, *n;
2283 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2284 loaded_vmcss_on_cpu_link)
2285 __loaded_vmcs_clear(v);
2289 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2292 static void kvm_cpu_vmxoff(void)
2294 asm volatile (__ex("vmxoff"));
2296 intel_pt_handle_vmx(0);
2297 cr4_clear_bits(X86_CR4_VMXE);
2300 static void hardware_disable(void)
2302 vmclear_local_loaded_vmcss();
2307 * There is no X86_FEATURE for SGX yet, but anyway we need to query CPUID
2308 * directly instead of going through cpu_has(), to ensure KVM is trapping
2309 * ENCLS whenever it's supported in hardware. It does not matter whether
2310 * the host OS supports or has enabled SGX.
2312 static bool cpu_has_sgx(void)
2314 return cpuid_eax(0) >= 0x12 && (cpuid_eax(0x12) & BIT(0));
2317 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2318 u32 msr, u32 *result)
2320 u32 vmx_msr_low, vmx_msr_high;
2321 u32 ctl = ctl_min | ctl_opt;
2323 rdmsr(msr, vmx_msr_low, vmx_msr_high);
2325 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2326 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
2328 /* Ensure minimum (required) set of control bits are supported. */
2336 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf,
2337 struct vmx_capability *vmx_cap)
2339 u32 vmx_msr_low, vmx_msr_high;
2340 u32 min, opt, min2, opt2;
2341 u32 _pin_based_exec_control = 0;
2342 u32 _cpu_based_exec_control = 0;
2343 u32 _cpu_based_2nd_exec_control = 0;
2344 u32 _vmexit_control = 0;
2345 u32 _vmentry_control = 0;
2347 memset(vmcs_conf, 0, sizeof(*vmcs_conf));
2348 min = CPU_BASED_HLT_EXITING |
2349 #ifdef CONFIG_X86_64
2350 CPU_BASED_CR8_LOAD_EXITING |
2351 CPU_BASED_CR8_STORE_EXITING |
2353 CPU_BASED_CR3_LOAD_EXITING |
2354 CPU_BASED_CR3_STORE_EXITING |
2355 CPU_BASED_UNCOND_IO_EXITING |
2356 CPU_BASED_MOV_DR_EXITING |
2357 CPU_BASED_USE_TSC_OFFSETTING |
2358 CPU_BASED_MWAIT_EXITING |
2359 CPU_BASED_MONITOR_EXITING |
2360 CPU_BASED_INVLPG_EXITING |
2361 CPU_BASED_RDPMC_EXITING;
2363 opt = CPU_BASED_TPR_SHADOW |
2364 CPU_BASED_USE_MSR_BITMAPS |
2365 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2366 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2367 &_cpu_based_exec_control) < 0)
2369 #ifdef CONFIG_X86_64
2370 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2371 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2372 ~CPU_BASED_CR8_STORE_EXITING;
2374 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2376 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2377 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2378 SECONDARY_EXEC_WBINVD_EXITING |
2379 SECONDARY_EXEC_ENABLE_VPID |
2380 SECONDARY_EXEC_ENABLE_EPT |
2381 SECONDARY_EXEC_UNRESTRICTED_GUEST |
2382 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2383 SECONDARY_EXEC_DESC |
2384 SECONDARY_EXEC_RDTSCP |
2385 SECONDARY_EXEC_ENABLE_INVPCID |
2386 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2387 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2388 SECONDARY_EXEC_SHADOW_VMCS |
2389 SECONDARY_EXEC_XSAVES |
2390 SECONDARY_EXEC_RDSEED_EXITING |
2391 SECONDARY_EXEC_RDRAND_EXITING |
2392 SECONDARY_EXEC_ENABLE_PML |
2393 SECONDARY_EXEC_TSC_SCALING |
2394 SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE |
2395 SECONDARY_EXEC_PT_USE_GPA |
2396 SECONDARY_EXEC_PT_CONCEAL_VMX |
2397 SECONDARY_EXEC_ENABLE_VMFUNC;
2399 opt2 |= SECONDARY_EXEC_ENCLS_EXITING;
2400 if (adjust_vmx_controls(min2, opt2,
2401 MSR_IA32_VMX_PROCBASED_CTLS2,
2402 &_cpu_based_2nd_exec_control) < 0)
2405 #ifndef CONFIG_X86_64
2406 if (!(_cpu_based_2nd_exec_control &
2407 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2408 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2411 if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2412 _cpu_based_2nd_exec_control &= ~(
2413 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2414 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2415 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
2417 rdmsr_safe(MSR_IA32_VMX_EPT_VPID_CAP,
2418 &vmx_cap->ept, &vmx_cap->vpid);
2420 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2421 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2423 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2424 CPU_BASED_CR3_STORE_EXITING |
2425 CPU_BASED_INVLPG_EXITING);
2426 } else if (vmx_cap->ept) {
2428 pr_warn_once("EPT CAP should not exist if not support "
2429 "1-setting enable EPT VM-execution control\n");
2431 if (!(_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_VPID) &&
2434 pr_warn_once("VPID CAP should not exist if not support "
2435 "1-setting enable VPID VM-execution control\n");
2438 min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
2439 #ifdef CONFIG_X86_64
2440 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2442 opt = VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL |
2443 VM_EXIT_LOAD_IA32_PAT |
2444 VM_EXIT_LOAD_IA32_EFER |
2445 VM_EXIT_CLEAR_BNDCFGS |
2446 VM_EXIT_PT_CONCEAL_PIP |
2447 VM_EXIT_CLEAR_IA32_RTIT_CTL;
2448 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2449 &_vmexit_control) < 0)
2452 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2453 opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR |
2454 PIN_BASED_VMX_PREEMPTION_TIMER;
2455 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2456 &_pin_based_exec_control) < 0)
2459 if (cpu_has_broken_vmx_preemption_timer())
2460 _pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
2461 if (!(_cpu_based_2nd_exec_control &
2462 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
2463 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
2465 min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
2466 opt = VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL |
2467 VM_ENTRY_LOAD_IA32_PAT |
2468 VM_ENTRY_LOAD_IA32_EFER |
2469 VM_ENTRY_LOAD_BNDCFGS |
2470 VM_ENTRY_PT_CONCEAL_PIP |
2471 VM_ENTRY_LOAD_IA32_RTIT_CTL;
2472 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2473 &_vmentry_control) < 0)
2477 * Some cpus support VM_{ENTRY,EXIT}_IA32_PERF_GLOBAL_CTRL but they
2478 * can't be used due to an errata where VM Exit may incorrectly clear
2479 * IA32_PERF_GLOBAL_CTRL[34:32]. Workaround the errata by using the
2480 * MSR load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2482 if (boot_cpu_data.x86 == 0x6) {
2483 switch (boot_cpu_data.x86_model) {
2484 case 26: /* AAK155 */
2485 case 30: /* AAP115 */
2486 case 37: /* AAT100 */
2487 case 44: /* BC86,AAY89,BD102 */
2489 _vmentry_control &= ~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
2490 _vmexit_control &= ~VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
2491 pr_warn_once("kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
2492 "does not work properly. Using workaround\n");
2500 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2502 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2503 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2506 #ifdef CONFIG_X86_64
2507 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2508 if (vmx_msr_high & (1u<<16))
2512 /* Require Write-Back (WB) memory type for VMCS accesses. */
2513 if (((vmx_msr_high >> 18) & 15) != 6)
2516 vmcs_conf->size = vmx_msr_high & 0x1fff;
2517 vmcs_conf->order = get_order(vmcs_conf->size);
2518 vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
2520 vmcs_conf->revision_id = vmx_msr_low;
2522 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2523 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2524 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2525 vmcs_conf->vmexit_ctrl = _vmexit_control;
2526 vmcs_conf->vmentry_ctrl = _vmentry_control;
2528 if (static_branch_unlikely(&enable_evmcs))
2529 evmcs_sanitize_exec_ctrls(vmcs_conf);
2534 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags)
2536 int node = cpu_to_node(cpu);
2540 pages = __alloc_pages_node(node, flags, vmcs_config.order);
2543 vmcs = page_address(pages);
2544 memset(vmcs, 0, vmcs_config.size);
2546 /* KVM supports Enlightened VMCS v1 only */
2547 if (static_branch_unlikely(&enable_evmcs))
2548 vmcs->hdr.revision_id = KVM_EVMCS_VERSION;
2550 vmcs->hdr.revision_id = vmcs_config.revision_id;
2553 vmcs->hdr.shadow_vmcs = 1;
2557 void free_vmcs(struct vmcs *vmcs)
2559 free_pages((unsigned long)vmcs, vmcs_config.order);
2563 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2565 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2567 if (!loaded_vmcs->vmcs)
2569 loaded_vmcs_clear(loaded_vmcs);
2570 free_vmcs(loaded_vmcs->vmcs);
2571 loaded_vmcs->vmcs = NULL;
2572 if (loaded_vmcs->msr_bitmap)
2573 free_page((unsigned long)loaded_vmcs->msr_bitmap);
2574 WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
2577 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2579 loaded_vmcs->vmcs = alloc_vmcs(false);
2580 if (!loaded_vmcs->vmcs)
2583 vmcs_clear(loaded_vmcs->vmcs);
2585 loaded_vmcs->shadow_vmcs = NULL;
2586 loaded_vmcs->hv_timer_soft_disabled = false;
2587 loaded_vmcs->cpu = -1;
2588 loaded_vmcs->launched = 0;
2590 if (cpu_has_vmx_msr_bitmap()) {
2591 loaded_vmcs->msr_bitmap = (unsigned long *)
2592 __get_free_page(GFP_KERNEL_ACCOUNT);
2593 if (!loaded_vmcs->msr_bitmap)
2595 memset(loaded_vmcs->msr_bitmap, 0xff, PAGE_SIZE);
2597 if (IS_ENABLED(CONFIG_HYPERV) &&
2598 static_branch_unlikely(&enable_evmcs) &&
2599 (ms_hyperv.nested_features & HV_X64_NESTED_MSR_BITMAP)) {
2600 struct hv_enlightened_vmcs *evmcs =
2601 (struct hv_enlightened_vmcs *)loaded_vmcs->vmcs;
2603 evmcs->hv_enlightenments_control.msr_bitmap = 1;
2607 memset(&loaded_vmcs->host_state, 0, sizeof(struct vmcs_host_state));
2608 memset(&loaded_vmcs->controls_shadow, 0,
2609 sizeof(struct vmcs_controls_shadow));
2614 free_loaded_vmcs(loaded_vmcs);
2618 static void free_kvm_area(void)
2622 for_each_possible_cpu(cpu) {
2623 free_vmcs(per_cpu(vmxarea, cpu));
2624 per_cpu(vmxarea, cpu) = NULL;
2628 static __init int alloc_kvm_area(void)
2632 for_each_possible_cpu(cpu) {
2635 vmcs = alloc_vmcs_cpu(false, cpu, GFP_KERNEL);
2642 * When eVMCS is enabled, alloc_vmcs_cpu() sets
2643 * vmcs->revision_id to KVM_EVMCS_VERSION instead of
2644 * revision_id reported by MSR_IA32_VMX_BASIC.
2646 * However, even though not explicitly documented by
2647 * TLFS, VMXArea passed as VMXON argument should
2648 * still be marked with revision_id reported by
2651 if (static_branch_unlikely(&enable_evmcs))
2652 vmcs->hdr.revision_id = vmcs_config.revision_id;
2654 per_cpu(vmxarea, cpu) = vmcs;
2659 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
2660 struct kvm_segment *save)
2662 if (!emulate_invalid_guest_state) {
2664 * CS and SS RPL should be equal during guest entry according
2665 * to VMX spec, but in reality it is not always so. Since vcpu
2666 * is in the middle of the transition from real mode to
2667 * protected mode it is safe to assume that RPL 0 is a good
2670 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
2671 save->selector &= ~SEGMENT_RPL_MASK;
2672 save->dpl = save->selector & SEGMENT_RPL_MASK;
2675 vmx_set_segment(vcpu, save, seg);
2678 static void enter_pmode(struct kvm_vcpu *vcpu)
2680 unsigned long flags;
2681 struct vcpu_vmx *vmx = to_vmx(vcpu);
2684 * Update real mode segment cache. It may be not up-to-date if sement
2685 * register was written while vcpu was in a guest mode.
2687 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2688 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2689 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2690 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2691 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2692 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2694 vmx->rmode.vm86_active = 0;
2696 vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2698 flags = vmcs_readl(GUEST_RFLAGS);
2699 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2700 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2701 vmcs_writel(GUEST_RFLAGS, flags);
2703 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
2704 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
2706 update_exception_bitmap(vcpu);
2708 fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2709 fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2710 fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2711 fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2712 fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2713 fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2716 static void fix_rmode_seg(int seg, struct kvm_segment *save)
2718 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2719 struct kvm_segment var = *save;
2722 if (seg == VCPU_SREG_CS)
2725 if (!emulate_invalid_guest_state) {
2726 var.selector = var.base >> 4;
2727 var.base = var.base & 0xffff0;
2737 if (save->base & 0xf)
2738 printk_once(KERN_WARNING "kvm: segment base is not "
2739 "paragraph aligned when entering "
2740 "protected mode (seg=%d)", seg);
2743 vmcs_write16(sf->selector, var.selector);
2744 vmcs_writel(sf->base, var.base);
2745 vmcs_write32(sf->limit, var.limit);
2746 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
2749 static void enter_rmode(struct kvm_vcpu *vcpu)
2751 unsigned long flags;
2752 struct vcpu_vmx *vmx = to_vmx(vcpu);
2753 struct kvm_vmx *kvm_vmx = to_kvm_vmx(vcpu->kvm);
2755 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2756 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2757 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2758 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2759 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2760 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2761 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2763 vmx->rmode.vm86_active = 1;
2766 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
2767 * vcpu. Warn the user that an update is overdue.
2769 if (!kvm_vmx->tss_addr)
2770 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
2771 "called before entering vcpu\n");
2773 vmx_segment_cache_clear(vmx);
2775 vmcs_writel(GUEST_TR_BASE, kvm_vmx->tss_addr);
2776 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
2777 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2779 flags = vmcs_readl(GUEST_RFLAGS);
2780 vmx->rmode.save_rflags = flags;
2782 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2784 vmcs_writel(GUEST_RFLAGS, flags);
2785 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
2786 update_exception_bitmap(vcpu);
2788 fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2789 fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2790 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2791 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2792 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2793 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2795 kvm_mmu_reset_context(vcpu);
2798 void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
2800 struct vcpu_vmx *vmx = to_vmx(vcpu);
2801 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
2806 vcpu->arch.efer = efer;
2807 if (efer & EFER_LMA) {
2808 vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2811 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2813 msr->data = efer & ~EFER_LME;
2818 #ifdef CONFIG_X86_64
2820 static void enter_lmode(struct kvm_vcpu *vcpu)
2824 vmx_segment_cache_clear(to_vmx(vcpu));
2826 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
2827 if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
2828 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
2830 vmcs_write32(GUEST_TR_AR_BYTES,
2831 (guest_tr_ar & ~VMX_AR_TYPE_MASK)
2832 | VMX_AR_TYPE_BUSY_64_TSS);
2834 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
2837 static void exit_lmode(struct kvm_vcpu *vcpu)
2839 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2840 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
2845 static void vmx_flush_tlb_all(struct kvm_vcpu *vcpu)
2847 struct vcpu_vmx *vmx = to_vmx(vcpu);
2850 * INVEPT must be issued when EPT is enabled, irrespective of VPID, as
2851 * the CPU is not required to invalidate guest-physical mappings on
2852 * VM-Entry, even if VPID is disabled. Guest-physical mappings are
2853 * associated with the root EPT structure and not any particular VPID
2854 * (INVVPID also isn't required to invalidate guest-physical mappings).
2858 } else if (enable_vpid) {
2859 if (cpu_has_vmx_invvpid_global()) {
2860 vpid_sync_vcpu_global();
2862 vpid_sync_vcpu_single(vmx->vpid);
2863 vpid_sync_vcpu_single(vmx->nested.vpid02);
2868 static void vmx_flush_tlb_current(struct kvm_vcpu *vcpu)
2870 u64 root_hpa = vcpu->arch.mmu->root_hpa;
2872 /* No flush required if the current context is invalid. */
2873 if (!VALID_PAGE(root_hpa))
2877 ept_sync_context(construct_eptp(vcpu, root_hpa));
2878 else if (!is_guest_mode(vcpu))
2879 vpid_sync_context(to_vmx(vcpu)->vpid);
2881 vpid_sync_context(nested_get_vpid02(vcpu));
2884 static void vmx_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t addr)
2887 * vpid_sync_vcpu_addr() is a nop if vmx->vpid==0, see the comment in
2888 * vmx_flush_tlb_guest() for an explanation of why this is ok.
2890 vpid_sync_vcpu_addr(to_vmx(vcpu)->vpid, addr);
2893 static void vmx_flush_tlb_guest(struct kvm_vcpu *vcpu)
2896 * vpid_sync_context() is a nop if vmx->vpid==0, e.g. if enable_vpid==0
2897 * or a vpid couldn't be allocated for this vCPU. VM-Enter and VM-Exit
2898 * are required to flush GVA->{G,H}PA mappings from the TLB if vpid is
2899 * disabled (VM-Enter with vpid enabled and vpid==0 is disallowed),
2900 * i.e. no explicit INVVPID is necessary.
2902 vpid_sync_context(to_vmx(vcpu)->vpid);
2905 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
2907 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
2909 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
2910 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
2913 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
2915 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
2917 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
2918 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
2921 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
2923 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2925 if (!kvm_register_is_dirty(vcpu, VCPU_EXREG_PDPTR))
2928 if (is_pae_paging(vcpu)) {
2929 vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
2930 vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
2931 vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
2932 vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
2936 void ept_save_pdptrs(struct kvm_vcpu *vcpu)
2938 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2940 if (is_pae_paging(vcpu)) {
2941 mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
2942 mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
2943 mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
2944 mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
2947 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
2950 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
2952 struct kvm_vcpu *vcpu)
2954 struct vcpu_vmx *vmx = to_vmx(vcpu);
2956 if (!kvm_register_is_available(vcpu, VCPU_EXREG_CR3))
2957 vmx_cache_reg(vcpu, VCPU_EXREG_CR3);
2958 if (!(cr0 & X86_CR0_PG)) {
2959 /* From paging/starting to nonpaging */
2960 exec_controls_setbit(vmx, CPU_BASED_CR3_LOAD_EXITING |
2961 CPU_BASED_CR3_STORE_EXITING);
2962 vcpu->arch.cr0 = cr0;
2963 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2964 } else if (!is_paging(vcpu)) {
2965 /* From nonpaging to paging */
2966 exec_controls_clearbit(vmx, CPU_BASED_CR3_LOAD_EXITING |
2967 CPU_BASED_CR3_STORE_EXITING);
2968 vcpu->arch.cr0 = cr0;
2969 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2972 if (!(cr0 & X86_CR0_WP))
2973 *hw_cr0 &= ~X86_CR0_WP;
2976 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
2978 struct vcpu_vmx *vmx = to_vmx(vcpu);
2979 unsigned long hw_cr0;
2981 hw_cr0 = (cr0 & ~KVM_VM_CR0_ALWAYS_OFF);
2982 if (enable_unrestricted_guest)
2983 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
2985 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
2987 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
2990 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
2994 #ifdef CONFIG_X86_64
2995 if (vcpu->arch.efer & EFER_LME) {
2996 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
2998 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
3003 if (enable_ept && !enable_unrestricted_guest)
3004 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
3006 vmcs_writel(CR0_READ_SHADOW, cr0);
3007 vmcs_writel(GUEST_CR0, hw_cr0);
3008 vcpu->arch.cr0 = cr0;
3010 /* depends on vcpu->arch.cr0 to be set to a new value */
3011 vmx->emulation_required = emulation_required(vcpu);
3014 static int get_ept_level(struct kvm_vcpu *vcpu)
3016 if (is_guest_mode(vcpu) && nested_cpu_has_ept(get_vmcs12(vcpu)))
3017 return vmx_eptp_page_walk_level(nested_ept_get_eptp(vcpu));
3018 if (cpu_has_vmx_ept_5levels() && (cpuid_maxphyaddr(vcpu) > 48))
3023 u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa)
3025 u64 eptp = VMX_EPTP_MT_WB;
3027 eptp |= (get_ept_level(vcpu) == 5) ? VMX_EPTP_PWL_5 : VMX_EPTP_PWL_4;
3029 if (enable_ept_ad_bits &&
3030 (!is_guest_mode(vcpu) || nested_ept_ad_enabled(vcpu)))
3031 eptp |= VMX_EPTP_AD_ENABLE_BIT;
3032 eptp |= (root_hpa & PAGE_MASK);
3037 void vmx_load_mmu_pgd(struct kvm_vcpu *vcpu, unsigned long pgd)
3039 struct kvm *kvm = vcpu->kvm;
3040 bool update_guest_cr3 = true;
3041 unsigned long guest_cr3;
3045 eptp = construct_eptp(vcpu, pgd);
3046 vmcs_write64(EPT_POINTER, eptp);
3048 if (kvm_x86_ops.tlb_remote_flush) {
3049 spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
3050 to_vmx(vcpu)->ept_pointer = eptp;
3051 to_kvm_vmx(kvm)->ept_pointers_match
3052 = EPT_POINTERS_CHECK;
3053 spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
3056 /* Loading vmcs02.GUEST_CR3 is handled by nested VM-Enter. */
3057 if (is_guest_mode(vcpu))
3058 update_guest_cr3 = false;
3059 else if (!enable_unrestricted_guest && !is_paging(vcpu))
3060 guest_cr3 = to_kvm_vmx(kvm)->ept_identity_map_addr;
3061 else if (test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
3062 guest_cr3 = vcpu->arch.cr3;
3063 else /* vmcs01.GUEST_CR3 is already up-to-date. */
3064 update_guest_cr3 = false;
3065 ept_load_pdptrs(vcpu);
3070 if (update_guest_cr3)
3071 vmcs_writel(GUEST_CR3, guest_cr3);
3074 int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3076 struct vcpu_vmx *vmx = to_vmx(vcpu);
3078 * Pass through host's Machine Check Enable value to hw_cr4, which
3079 * is in force while we are in guest mode. Do not let guests control
3080 * this bit, even if host CR4.MCE == 0.
3082 unsigned long hw_cr4;
3084 hw_cr4 = (cr4_read_shadow() & X86_CR4_MCE) | (cr4 & ~X86_CR4_MCE);
3085 if (enable_unrestricted_guest)
3086 hw_cr4 |= KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST;
3087 else if (vmx->rmode.vm86_active)
3088 hw_cr4 |= KVM_RMODE_VM_CR4_ALWAYS_ON;
3090 hw_cr4 |= KVM_PMODE_VM_CR4_ALWAYS_ON;
3092 if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated()) {
3093 if (cr4 & X86_CR4_UMIP) {
3094 secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_DESC);
3095 hw_cr4 &= ~X86_CR4_UMIP;
3096 } else if (!is_guest_mode(vcpu) ||
3097 !nested_cpu_has2(get_vmcs12(vcpu), SECONDARY_EXEC_DESC)) {
3098 secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_DESC);
3102 if (cr4 & X86_CR4_VMXE) {
3104 * To use VMXON (and later other VMX instructions), a guest
3105 * must first be able to turn on cr4.VMXE (see handle_vmon()).
3106 * So basically the check on whether to allow nested VMX
3107 * is here. We operate under the default treatment of SMM,
3108 * so VMX cannot be enabled under SMM.
3110 if (!nested_vmx_allowed(vcpu) || is_smm(vcpu))
3114 if (vmx->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
3117 vcpu->arch.cr4 = cr4;
3119 if (!enable_unrestricted_guest) {
3121 if (!is_paging(vcpu)) {
3122 hw_cr4 &= ~X86_CR4_PAE;
3123 hw_cr4 |= X86_CR4_PSE;
3124 } else if (!(cr4 & X86_CR4_PAE)) {
3125 hw_cr4 &= ~X86_CR4_PAE;
3130 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
3131 * hardware. To emulate this behavior, SMEP/SMAP/PKU needs
3132 * to be manually disabled when guest switches to non-paging
3135 * If !enable_unrestricted_guest, the CPU is always running
3136 * with CR0.PG=1 and CR4 needs to be modified.
3137 * If enable_unrestricted_guest, the CPU automatically
3138 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
3140 if (!is_paging(vcpu))
3141 hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
3144 vmcs_writel(CR4_READ_SHADOW, cr4);
3145 vmcs_writel(GUEST_CR4, hw_cr4);
3149 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3151 struct vcpu_vmx *vmx = to_vmx(vcpu);
3154 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3155 *var = vmx->rmode.segs[seg];
3156 if (seg == VCPU_SREG_TR
3157 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3159 var->base = vmx_read_guest_seg_base(vmx, seg);
3160 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3163 var->base = vmx_read_guest_seg_base(vmx, seg);
3164 var->limit = vmx_read_guest_seg_limit(vmx, seg);
3165 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3166 ar = vmx_read_guest_seg_ar(vmx, seg);
3167 var->unusable = (ar >> 16) & 1;
3168 var->type = ar & 15;
3169 var->s = (ar >> 4) & 1;
3170 var->dpl = (ar >> 5) & 3;
3172 * Some userspaces do not preserve unusable property. Since usable
3173 * segment has to be present according to VMX spec we can use present
3174 * property to amend userspace bug by making unusable segment always
3175 * nonpresent. vmx_segment_access_rights() already marks nonpresent
3176 * segment as unusable.
3178 var->present = !var->unusable;
3179 var->avl = (ar >> 12) & 1;
3180 var->l = (ar >> 13) & 1;
3181 var->db = (ar >> 14) & 1;
3182 var->g = (ar >> 15) & 1;
3185 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3187 struct kvm_segment s;
3189 if (to_vmx(vcpu)->rmode.vm86_active) {
3190 vmx_get_segment(vcpu, &s, seg);
3193 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3196 int vmx_get_cpl(struct kvm_vcpu *vcpu)
3198 struct vcpu_vmx *vmx = to_vmx(vcpu);
3200 if (unlikely(vmx->rmode.vm86_active))
3203 int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
3204 return VMX_AR_DPL(ar);
3208 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3212 if (var->unusable || !var->present)
3215 ar = var->type & 15;
3216 ar |= (var->s & 1) << 4;
3217 ar |= (var->dpl & 3) << 5;
3218 ar |= (var->present & 1) << 7;
3219 ar |= (var->avl & 1) << 12;
3220 ar |= (var->l & 1) << 13;
3221 ar |= (var->db & 1) << 14;
3222 ar |= (var->g & 1) << 15;
3228 void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3230 struct vcpu_vmx *vmx = to_vmx(vcpu);
3231 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3233 vmx_segment_cache_clear(vmx);
3235 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3236 vmx->rmode.segs[seg] = *var;
3237 if (seg == VCPU_SREG_TR)
3238 vmcs_write16(sf->selector, var->selector);
3240 fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3244 vmcs_writel(sf->base, var->base);
3245 vmcs_write32(sf->limit, var->limit);
3246 vmcs_write16(sf->selector, var->selector);
3249 * Fix the "Accessed" bit in AR field of segment registers for older
3251 * IA32 arch specifies that at the time of processor reset the
3252 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3253 * is setting it to 0 in the userland code. This causes invalid guest
3254 * state vmexit when "unrestricted guest" mode is turned on.
3255 * Fix for this setup issue in cpu_reset is being pushed in the qemu
3256 * tree. Newer qemu binaries with that qemu fix would not need this
3259 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3260 var->type |= 0x1; /* Accessed */
3262 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3265 vmx->emulation_required = emulation_required(vcpu);
3268 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3270 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3272 *db = (ar >> 14) & 1;
3273 *l = (ar >> 13) & 1;
3276 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3278 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3279 dt->address = vmcs_readl(GUEST_IDTR_BASE);
3282 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3284 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3285 vmcs_writel(GUEST_IDTR_BASE, dt->address);
3288 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3290 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3291 dt->address = vmcs_readl(GUEST_GDTR_BASE);
3294 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3296 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3297 vmcs_writel(GUEST_GDTR_BASE, dt->address);
3300 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3302 struct kvm_segment var;
3305 vmx_get_segment(vcpu, &var, seg);
3307 if (seg == VCPU_SREG_CS)
3309 ar = vmx_segment_access_rights(&var);
3311 if (var.base != (var.selector << 4))
3313 if (var.limit != 0xffff)
3321 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3323 struct kvm_segment cs;
3324 unsigned int cs_rpl;
3326 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3327 cs_rpl = cs.selector & SEGMENT_RPL_MASK;
3331 if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
3335 if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
3336 if (cs.dpl > cs_rpl)
3339 if (cs.dpl != cs_rpl)
3345 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3349 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3351 struct kvm_segment ss;
3352 unsigned int ss_rpl;
3354 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3355 ss_rpl = ss.selector & SEGMENT_RPL_MASK;
3359 if (ss.type != 3 && ss.type != 7)
3363 if (ss.dpl != ss_rpl) /* DPL != RPL */
3371 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3373 struct kvm_segment var;
3376 vmx_get_segment(vcpu, &var, seg);
3377 rpl = var.selector & SEGMENT_RPL_MASK;
3385 if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
3386 if (var.dpl < rpl) /* DPL < RPL */
3390 /* TODO: Add other members to kvm_segment_field to allow checking for other access
3396 static bool tr_valid(struct kvm_vcpu *vcpu)
3398 struct kvm_segment tr;
3400 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3404 if (tr.selector & SEGMENT_TI_MASK) /* TI = 1 */
3406 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3414 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3416 struct kvm_segment ldtr;
3418 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3422 if (ldtr.selector & SEGMENT_TI_MASK) /* TI = 1 */
3432 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3434 struct kvm_segment cs, ss;
3436 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3437 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3439 return ((cs.selector & SEGMENT_RPL_MASK) ==
3440 (ss.selector & SEGMENT_RPL_MASK));
3444 * Check if guest state is valid. Returns true if valid, false if
3446 * We assume that registers are always usable
3448 static bool guest_state_valid(struct kvm_vcpu *vcpu)
3450 if (enable_unrestricted_guest)
3453 /* real mode guest state checks */
3454 if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
3455 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3457 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3459 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3461 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3463 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3465 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3468 /* protected mode guest state checks */
3469 if (!cs_ss_rpl_check(vcpu))
3471 if (!code_segment_valid(vcpu))
3473 if (!stack_segment_valid(vcpu))
3475 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3477 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3479 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3481 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3483 if (!tr_valid(vcpu))
3485 if (!ldtr_valid(vcpu))
3489 * - Add checks on RIP
3490 * - Add checks on RFLAGS
3496 static int init_rmode_tss(struct kvm *kvm)
3502 idx = srcu_read_lock(&kvm->srcu);
3503 fn = to_kvm_vmx(kvm)->tss_addr >> PAGE_SHIFT;
3504 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3507 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3508 r = kvm_write_guest_page(kvm, fn++, &data,
3509 TSS_IOPB_BASE_OFFSET, sizeof(u16));
3512 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3515 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3519 r = kvm_write_guest_page(kvm, fn, &data,
3520 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3523 srcu_read_unlock(&kvm->srcu, idx);
3527 static int init_rmode_identity_map(struct kvm *kvm)
3529 struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm);
3531 kvm_pfn_t identity_map_pfn;
3534 /* Protect kvm_vmx->ept_identity_pagetable_done. */
3535 mutex_lock(&kvm->slots_lock);
3537 if (likely(kvm_vmx->ept_identity_pagetable_done))
3540 if (!kvm_vmx->ept_identity_map_addr)
3541 kvm_vmx->ept_identity_map_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
3542 identity_map_pfn = kvm_vmx->ept_identity_map_addr >> PAGE_SHIFT;
3544 r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
3545 kvm_vmx->ept_identity_map_addr, PAGE_SIZE);
3549 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3552 /* Set up identity-mapping pagetable for EPT in real mode */
3553 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3554 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3555 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3556 r = kvm_write_guest_page(kvm, identity_map_pfn,
3557 &tmp, i * sizeof(tmp), sizeof(tmp));
3561 kvm_vmx->ept_identity_pagetable_done = true;
3564 mutex_unlock(&kvm->slots_lock);
3568 static void seg_setup(int seg)
3570 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3573 vmcs_write16(sf->selector, 0);
3574 vmcs_writel(sf->base, 0);
3575 vmcs_write32(sf->limit, 0xffff);
3577 if (seg == VCPU_SREG_CS)
3578 ar |= 0x08; /* code segment */
3580 vmcs_write32(sf->ar_bytes, ar);
3583 static int alloc_apic_access_page(struct kvm *kvm)
3588 mutex_lock(&kvm->slots_lock);
3589 if (kvm->arch.apic_access_page_done)
3591 r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
3592 APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
3596 page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
3597 if (is_error_page(page)) {
3603 * Do not pin the page in memory, so that memory hot-unplug
3604 * is able to migrate it.
3607 kvm->arch.apic_access_page_done = true;
3609 mutex_unlock(&kvm->slots_lock);
3613 int allocate_vpid(void)
3619 spin_lock(&vmx_vpid_lock);
3620 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3621 if (vpid < VMX_NR_VPIDS)
3622 __set_bit(vpid, vmx_vpid_bitmap);
3625 spin_unlock(&vmx_vpid_lock);
3629 void free_vpid(int vpid)
3631 if (!enable_vpid || vpid == 0)
3633 spin_lock(&vmx_vpid_lock);
3634 __clear_bit(vpid, vmx_vpid_bitmap);
3635 spin_unlock(&vmx_vpid_lock);
3638 static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
3641 int f = sizeof(unsigned long);
3643 if (!cpu_has_vmx_msr_bitmap())
3646 if (static_branch_unlikely(&enable_evmcs))
3647 evmcs_touch_msr_bitmap();
3650 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3651 * have the write-low and read-high bitmap offsets the wrong way round.
3652 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3654 if (msr <= 0x1fff) {
3655 if (type & MSR_TYPE_R)
3657 __clear_bit(msr, msr_bitmap + 0x000 / f);
3659 if (type & MSR_TYPE_W)
3661 __clear_bit(msr, msr_bitmap + 0x800 / f);
3663 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3665 if (type & MSR_TYPE_R)
3667 __clear_bit(msr, msr_bitmap + 0x400 / f);
3669 if (type & MSR_TYPE_W)
3671 __clear_bit(msr, msr_bitmap + 0xc00 / f);
3676 static __always_inline void vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
3679 int f = sizeof(unsigned long);
3681 if (!cpu_has_vmx_msr_bitmap())
3684 if (static_branch_unlikely(&enable_evmcs))
3685 evmcs_touch_msr_bitmap();
3688 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3689 * have the write-low and read-high bitmap offsets the wrong way round.
3690 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3692 if (msr <= 0x1fff) {
3693 if (type & MSR_TYPE_R)
3695 __set_bit(msr, msr_bitmap + 0x000 / f);
3697 if (type & MSR_TYPE_W)
3699 __set_bit(msr, msr_bitmap + 0x800 / f);
3701 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3703 if (type & MSR_TYPE_R)
3705 __set_bit(msr, msr_bitmap + 0x400 / f);
3707 if (type & MSR_TYPE_W)
3709 __set_bit(msr, msr_bitmap + 0xc00 / f);
3714 static __always_inline void vmx_set_intercept_for_msr(unsigned long *msr_bitmap,
3715 u32 msr, int type, bool value)
3718 vmx_enable_intercept_for_msr(msr_bitmap, msr, type);
3720 vmx_disable_intercept_for_msr(msr_bitmap, msr, type);
3723 static u8 vmx_msr_bitmap_mode(struct kvm_vcpu *vcpu)
3727 if (cpu_has_secondary_exec_ctrls() &&
3728 (secondary_exec_controls_get(to_vmx(vcpu)) &
3729 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
3730 mode |= MSR_BITMAP_MODE_X2APIC;
3731 if (enable_apicv && kvm_vcpu_apicv_active(vcpu))
3732 mode |= MSR_BITMAP_MODE_X2APIC_APICV;
3738 static void vmx_update_msr_bitmap_x2apic(unsigned long *msr_bitmap,
3743 for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
3744 unsigned word = msr / BITS_PER_LONG;
3745 msr_bitmap[word] = (mode & MSR_BITMAP_MODE_X2APIC_APICV) ? 0 : ~0;
3746 msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
3749 if (mode & MSR_BITMAP_MODE_X2APIC) {
3751 * TPR reads and writes can be virtualized even if virtual interrupt
3752 * delivery is not in use.
3754 vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TASKPRI), MSR_TYPE_RW);
3755 if (mode & MSR_BITMAP_MODE_X2APIC_APICV) {
3756 vmx_enable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TMCCT), MSR_TYPE_R);
3757 vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_EOI), MSR_TYPE_W);
3758 vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_SELF_IPI), MSR_TYPE_W);
3763 void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu)
3765 struct vcpu_vmx *vmx = to_vmx(vcpu);
3766 unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3767 u8 mode = vmx_msr_bitmap_mode(vcpu);
3768 u8 changed = mode ^ vmx->msr_bitmap_mode;
3773 if (changed & (MSR_BITMAP_MODE_X2APIC | MSR_BITMAP_MODE_X2APIC_APICV))
3774 vmx_update_msr_bitmap_x2apic(msr_bitmap, mode);
3776 vmx->msr_bitmap_mode = mode;
3779 void pt_update_intercept_for_msr(struct vcpu_vmx *vmx)
3781 unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3782 bool flag = !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
3785 vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_STATUS,
3787 vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_BASE,
3789 vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_MASK,
3791 vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_CR3_MATCH,
3793 for (i = 0; i < vmx->pt_desc.addr_range; i++) {
3794 vmx_set_intercept_for_msr(msr_bitmap,
3795 MSR_IA32_RTIT_ADDR0_A + i * 2, MSR_TYPE_RW, flag);
3796 vmx_set_intercept_for_msr(msr_bitmap,
3797 MSR_IA32_RTIT_ADDR0_B + i * 2, MSR_TYPE_RW, flag);
3801 static bool vmx_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
3803 struct vcpu_vmx *vmx = to_vmx(vcpu);
3808 if (WARN_ON_ONCE(!is_guest_mode(vcpu)) ||
3809 !nested_cpu_has_vid(get_vmcs12(vcpu)) ||
3810 WARN_ON_ONCE(!vmx->nested.virtual_apic_map.gfn))
3813 rvi = vmx_get_rvi();
3815 vapic_page = vmx->nested.virtual_apic_map.hva;
3816 vppr = *((u32 *)(vapic_page + APIC_PROCPRI));
3818 return ((rvi & 0xf0) > (vppr & 0xf0));
3821 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu,
3825 int pi_vec = nested ? POSTED_INTR_NESTED_VECTOR : POSTED_INTR_VECTOR;
3827 if (vcpu->mode == IN_GUEST_MODE) {
3829 * The vector of interrupt to be delivered to vcpu had
3830 * been set in PIR before this function.
3832 * Following cases will be reached in this block, and
3833 * we always send a notification event in all cases as
3836 * Case 1: vcpu keeps in non-root mode. Sending a
3837 * notification event posts the interrupt to vcpu.
3839 * Case 2: vcpu exits to root mode and is still
3840 * runnable. PIR will be synced to vIRR before the
3841 * next vcpu entry. Sending a notification event in
3842 * this case has no effect, as vcpu is not in root
3845 * Case 3: vcpu exits to root mode and is blocked.
3846 * vcpu_block() has already synced PIR to vIRR and
3847 * never blocks vcpu if vIRR is not cleared. Therefore,
3848 * a blocked vcpu here does not wait for any requested
3849 * interrupts in PIR, and sending a notification event
3850 * which has no effect is safe here.
3853 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu), pi_vec);
3860 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
3863 struct vcpu_vmx *vmx = to_vmx(vcpu);
3865 if (is_guest_mode(vcpu) &&
3866 vector == vmx->nested.posted_intr_nv) {
3868 * If a posted intr is not recognized by hardware,
3869 * we will accomplish it in the next vmentry.
3871 vmx->nested.pi_pending = true;
3872 kvm_make_request(KVM_REQ_EVENT, vcpu);
3873 /* the PIR and ON have been set by L1. */
3874 if (!kvm_vcpu_trigger_posted_interrupt(vcpu, true))
3875 kvm_vcpu_kick(vcpu);
3881 * Send interrupt to vcpu via posted interrupt way.
3882 * 1. If target vcpu is running(non-root mode), send posted interrupt
3883 * notification to vcpu and hardware will sync PIR to vIRR atomically.
3884 * 2. If target vcpu isn't running(root mode), kick it to pick up the
3885 * interrupt from PIR in next vmentry.
3887 static int vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
3889 struct vcpu_vmx *vmx = to_vmx(vcpu);
3892 r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
3896 if (!vcpu->arch.apicv_active)
3899 if (pi_test_and_set_pir(vector, &vmx->pi_desc))
3902 /* If a previous notification has sent the IPI, nothing to do. */
3903 if (pi_test_and_set_on(&vmx->pi_desc))
3906 if (!kvm_vcpu_trigger_posted_interrupt(vcpu, false))
3907 kvm_vcpu_kick(vcpu);
3913 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
3914 * will not change in the lifetime of the guest.
3915 * Note that host-state that does change is set elsewhere. E.g., host-state
3916 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
3918 void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
3922 unsigned long cr0, cr3, cr4;
3925 WARN_ON(cr0 & X86_CR0_TS);
3926 vmcs_writel(HOST_CR0, cr0); /* 22.2.3 */
3929 * Save the most likely value for this task's CR3 in the VMCS.
3930 * We can't use __get_current_cr3_fast() because we're not atomic.
3933 vmcs_writel(HOST_CR3, cr3); /* 22.2.3 FIXME: shadow tables */
3934 vmx->loaded_vmcs->host_state.cr3 = cr3;
3936 /* Save the most likely value for this task's CR4 in the VMCS. */
3937 cr4 = cr4_read_shadow();
3938 vmcs_writel(HOST_CR4, cr4); /* 22.2.3, 22.2.5 */
3939 vmx->loaded_vmcs->host_state.cr4 = cr4;
3941 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
3942 #ifdef CONFIG_X86_64
3944 * Load null selectors, so we can avoid reloading them in
3945 * vmx_prepare_switch_to_host(), in case userspace uses
3946 * the null selectors too (the expected case).
3948 vmcs_write16(HOST_DS_SELECTOR, 0);
3949 vmcs_write16(HOST_ES_SELECTOR, 0);
3951 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
3952 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
3954 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
3955 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
3957 vmcs_writel(HOST_IDTR_BASE, host_idt_base); /* 22.2.4 */
3959 vmcs_writel(HOST_RIP, (unsigned long)vmx_vmexit); /* 22.2.5 */
3961 rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
3962 vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
3963 rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
3964 vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */
3966 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
3967 rdmsr(MSR_IA32_CR_PAT, low32, high32);
3968 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
3971 if (cpu_has_load_ia32_efer())
3972 vmcs_write64(HOST_IA32_EFER, host_efer);
3975 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
3977 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
3979 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
3980 if (is_guest_mode(&vmx->vcpu))
3981 vmx->vcpu.arch.cr4_guest_owned_bits &=
3982 ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
3983 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
3986 u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
3988 u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
3990 if (!kvm_vcpu_apicv_active(&vmx->vcpu))
3991 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
3994 pin_based_exec_ctrl &= ~PIN_BASED_VIRTUAL_NMIS;
3996 if (!enable_preemption_timer)
3997 pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
3999 return pin_based_exec_ctrl;
4002 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
4004 struct vcpu_vmx *vmx = to_vmx(vcpu);
4006 pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
4007 if (cpu_has_secondary_exec_ctrls()) {
4008 if (kvm_vcpu_apicv_active(vcpu))
4009 secondary_exec_controls_setbit(vmx,
4010 SECONDARY_EXEC_APIC_REGISTER_VIRT |
4011 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4013 secondary_exec_controls_clearbit(vmx,
4014 SECONDARY_EXEC_APIC_REGISTER_VIRT |
4015 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4018 if (cpu_has_vmx_msr_bitmap())
4019 vmx_update_msr_bitmap(vcpu);
4022 u32 vmx_exec_control(struct vcpu_vmx *vmx)
4024 u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
4026 if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
4027 exec_control &= ~CPU_BASED_MOV_DR_EXITING;
4029 if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
4030 exec_control &= ~CPU_BASED_TPR_SHADOW;
4031 #ifdef CONFIG_X86_64
4032 exec_control |= CPU_BASED_CR8_STORE_EXITING |
4033 CPU_BASED_CR8_LOAD_EXITING;
4037 exec_control |= CPU_BASED_CR3_STORE_EXITING |
4038 CPU_BASED_CR3_LOAD_EXITING |
4039 CPU_BASED_INVLPG_EXITING;
4040 if (kvm_mwait_in_guest(vmx->vcpu.kvm))
4041 exec_control &= ~(CPU_BASED_MWAIT_EXITING |
4042 CPU_BASED_MONITOR_EXITING);
4043 if (kvm_hlt_in_guest(vmx->vcpu.kvm))
4044 exec_control &= ~CPU_BASED_HLT_EXITING;
4045 return exec_control;
4049 static void vmx_compute_secondary_exec_control(struct vcpu_vmx *vmx)
4051 struct kvm_vcpu *vcpu = &vmx->vcpu;
4053 u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
4055 if (vmx_pt_mode_is_system())
4056 exec_control &= ~(SECONDARY_EXEC_PT_USE_GPA | SECONDARY_EXEC_PT_CONCEAL_VMX);
4057 if (!cpu_need_virtualize_apic_accesses(vcpu))
4058 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
4060 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
4062 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
4063 enable_unrestricted_guest = 0;
4065 if (!enable_unrestricted_guest)
4066 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4067 if (kvm_pause_in_guest(vmx->vcpu.kvm))
4068 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
4069 if (!kvm_vcpu_apicv_active(vcpu))
4070 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
4071 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4072 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
4074 /* SECONDARY_EXEC_DESC is enabled/disabled on writes to CR4.UMIP,
4075 * in vmx_set_cr4. */
4076 exec_control &= ~SECONDARY_EXEC_DESC;
4078 /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4080 We can NOT enable shadow_vmcs here because we don't have yet
4083 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4086 exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
4088 if (vmx_xsaves_supported()) {
4089 /* Exposing XSAVES only when XSAVE is exposed */
4090 bool xsaves_enabled =
4091 boot_cpu_has(X86_FEATURE_XSAVE) &&
4092 guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
4093 guest_cpuid_has(vcpu, X86_FEATURE_XSAVES);
4095 vcpu->arch.xsaves_enabled = xsaves_enabled;
4097 if (!xsaves_enabled)
4098 exec_control &= ~SECONDARY_EXEC_XSAVES;
4102 vmx->nested.msrs.secondary_ctls_high |=
4103 SECONDARY_EXEC_XSAVES;
4105 vmx->nested.msrs.secondary_ctls_high &=
4106 ~SECONDARY_EXEC_XSAVES;
4110 if (cpu_has_vmx_rdtscp()) {
4111 bool rdtscp_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP);
4112 if (!rdtscp_enabled)
4113 exec_control &= ~SECONDARY_EXEC_RDTSCP;
4117 vmx->nested.msrs.secondary_ctls_high |=
4118 SECONDARY_EXEC_RDTSCP;
4120 vmx->nested.msrs.secondary_ctls_high &=
4121 ~SECONDARY_EXEC_RDTSCP;
4125 if (cpu_has_vmx_invpcid()) {
4126 /* Exposing INVPCID only when PCID is exposed */
4127 bool invpcid_enabled =
4128 guest_cpuid_has(vcpu, X86_FEATURE_INVPCID) &&
4129 guest_cpuid_has(vcpu, X86_FEATURE_PCID);
4131 if (!invpcid_enabled) {
4132 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
4133 guest_cpuid_clear(vcpu, X86_FEATURE_INVPCID);
4137 if (invpcid_enabled)
4138 vmx->nested.msrs.secondary_ctls_high |=
4139 SECONDARY_EXEC_ENABLE_INVPCID;
4141 vmx->nested.msrs.secondary_ctls_high &=
4142 ~SECONDARY_EXEC_ENABLE_INVPCID;
4146 if (vmx_rdrand_supported()) {
4147 bool rdrand_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDRAND);
4149 exec_control &= ~SECONDARY_EXEC_RDRAND_EXITING;
4153 vmx->nested.msrs.secondary_ctls_high |=
4154 SECONDARY_EXEC_RDRAND_EXITING;
4156 vmx->nested.msrs.secondary_ctls_high &=
4157 ~SECONDARY_EXEC_RDRAND_EXITING;
4161 if (vmx_rdseed_supported()) {
4162 bool rdseed_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDSEED);
4164 exec_control &= ~SECONDARY_EXEC_RDSEED_EXITING;
4168 vmx->nested.msrs.secondary_ctls_high |=
4169 SECONDARY_EXEC_RDSEED_EXITING;
4171 vmx->nested.msrs.secondary_ctls_high &=
4172 ~SECONDARY_EXEC_RDSEED_EXITING;
4176 if (vmx_waitpkg_supported()) {
4177 bool waitpkg_enabled =
4178 guest_cpuid_has(vcpu, X86_FEATURE_WAITPKG);
4180 if (!waitpkg_enabled)
4181 exec_control &= ~SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4184 if (waitpkg_enabled)
4185 vmx->nested.msrs.secondary_ctls_high |=
4186 SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4188 vmx->nested.msrs.secondary_ctls_high &=
4189 ~SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4193 vmx->secondary_exec_control = exec_control;
4196 static void ept_set_mmio_spte_mask(void)
4199 * EPT Misconfigurations can be generated if the value of bits 2:0
4200 * of an EPT paging-structure entry is 110b (write/execute).
4202 kvm_mmu_set_mmio_spte_mask(VMX_EPT_RWX_MASK,
4203 VMX_EPT_MISCONFIG_WX_VALUE, 0);
4206 #define VMX_XSS_EXIT_BITMAP 0
4209 * Noting that the initialization of Guest-state Area of VMCS is in
4212 static void init_vmcs(struct vcpu_vmx *vmx)
4215 nested_vmx_set_vmcs_shadowing_bitmap();
4217 if (cpu_has_vmx_msr_bitmap())
4218 vmcs_write64(MSR_BITMAP, __pa(vmx->vmcs01.msr_bitmap));
4220 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4223 pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
4225 exec_controls_set(vmx, vmx_exec_control(vmx));
4227 if (cpu_has_secondary_exec_ctrls()) {
4228 vmx_compute_secondary_exec_control(vmx);
4229 secondary_exec_controls_set(vmx, vmx->secondary_exec_control);
4232 if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
4233 vmcs_write64(EOI_EXIT_BITMAP0, 0);
4234 vmcs_write64(EOI_EXIT_BITMAP1, 0);
4235 vmcs_write64(EOI_EXIT_BITMAP2, 0);
4236 vmcs_write64(EOI_EXIT_BITMAP3, 0);
4238 vmcs_write16(GUEST_INTR_STATUS, 0);
4240 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4241 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4244 if (!kvm_pause_in_guest(vmx->vcpu.kvm)) {
4245 vmcs_write32(PLE_GAP, ple_gap);
4246 vmx->ple_window = ple_window;
4247 vmx->ple_window_dirty = true;
4250 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4251 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4252 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
4254 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
4255 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
4256 vmx_set_constant_host_state(vmx);
4257 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4258 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4260 if (cpu_has_vmx_vmfunc())
4261 vmcs_write64(VM_FUNCTION_CONTROL, 0);
4263 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4264 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4265 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
4266 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4267 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
4269 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
4270 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
4272 vm_exit_controls_set(vmx, vmx_vmexit_ctrl());
4274 /* 22.2.1, 20.8.1 */
4275 vm_entry_controls_set(vmx, vmx_vmentry_ctrl());
4277 vmx->vcpu.arch.cr0_guest_owned_bits = X86_CR0_TS;
4278 vmcs_writel(CR0_GUEST_HOST_MASK, ~X86_CR0_TS);
4280 set_cr4_guest_host_mask(vmx);
4283 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4285 if (vmx_xsaves_supported())
4286 vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
4289 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
4290 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
4293 if (cpu_has_vmx_encls_vmexit())
4294 vmcs_write64(ENCLS_EXITING_BITMAP, -1ull);
4296 if (vmx_pt_mode_is_host_guest()) {
4297 memset(&vmx->pt_desc, 0, sizeof(vmx->pt_desc));
4298 /* Bit[6~0] are forced to 1, writes are ignored. */
4299 vmx->pt_desc.guest.output_mask = 0x7F;
4300 vmcs_write64(GUEST_IA32_RTIT_CTL, 0);
4304 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
4306 struct vcpu_vmx *vmx = to_vmx(vcpu);
4307 struct msr_data apic_base_msr;
4310 vmx->rmode.vm86_active = 0;
4313 vmx->msr_ia32_umwait_control = 0;
4315 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
4316 vmx->hv_deadline_tsc = -1;
4317 kvm_set_cr8(vcpu, 0);
4320 apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
4321 MSR_IA32_APICBASE_ENABLE;
4322 if (kvm_vcpu_is_reset_bsp(vcpu))
4323 apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
4324 apic_base_msr.host_initiated = true;
4325 kvm_set_apic_base(vcpu, &apic_base_msr);
4328 vmx_segment_cache_clear(vmx);
4330 seg_setup(VCPU_SREG_CS);
4331 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4332 vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
4334 seg_setup(VCPU_SREG_DS);
4335 seg_setup(VCPU_SREG_ES);
4336 seg_setup(VCPU_SREG_FS);
4337 seg_setup(VCPU_SREG_GS);
4338 seg_setup(VCPU_SREG_SS);
4340 vmcs_write16(GUEST_TR_SELECTOR, 0);
4341 vmcs_writel(GUEST_TR_BASE, 0);
4342 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4343 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4345 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4346 vmcs_writel(GUEST_LDTR_BASE, 0);
4347 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4348 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4351 vmcs_write32(GUEST_SYSENTER_CS, 0);
4352 vmcs_writel(GUEST_SYSENTER_ESP, 0);
4353 vmcs_writel(GUEST_SYSENTER_EIP, 0);
4354 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4357 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
4358 kvm_rip_write(vcpu, 0xfff0);
4360 vmcs_writel(GUEST_GDTR_BASE, 0);
4361 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4363 vmcs_writel(GUEST_IDTR_BASE, 0);
4364 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4366 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4367 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4368 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4369 if (kvm_mpx_supported())
4370 vmcs_write64(GUEST_BNDCFGS, 0);
4374 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
4376 if (cpu_has_vmx_tpr_shadow() && !init_event) {
4377 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4378 if (cpu_need_tpr_shadow(vcpu))
4379 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4380 __pa(vcpu->arch.apic->regs));
4381 vmcs_write32(TPR_THRESHOLD, 0);
4384 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4386 cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4387 vmx->vcpu.arch.cr0 = cr0;
4388 vmx_set_cr0(vcpu, cr0); /* enter rmode */
4389 vmx_set_cr4(vcpu, 0);
4390 vmx_set_efer(vcpu, 0);
4392 update_exception_bitmap(vcpu);
4394 vpid_sync_context(vmx->vpid);
4396 vmx_clear_hlt(vcpu);
4399 static void enable_irq_window(struct kvm_vcpu *vcpu)
4401 exec_controls_setbit(to_vmx(vcpu), CPU_BASED_INTR_WINDOW_EXITING);
4404 static void enable_nmi_window(struct kvm_vcpu *vcpu)
4407 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4408 enable_irq_window(vcpu);
4412 exec_controls_setbit(to_vmx(vcpu), CPU_BASED_NMI_WINDOW_EXITING);
4415 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4417 struct vcpu_vmx *vmx = to_vmx(vcpu);
4419 int irq = vcpu->arch.interrupt.nr;
4421 trace_kvm_inj_virq(irq);
4423 ++vcpu->stat.irq_injections;
4424 if (vmx->rmode.vm86_active) {
4426 if (vcpu->arch.interrupt.soft)
4427 inc_eip = vcpu->arch.event_exit_inst_len;
4428 kvm_inject_realmode_interrupt(vcpu, irq, inc_eip);
4431 intr = irq | INTR_INFO_VALID_MASK;
4432 if (vcpu->arch.interrupt.soft) {
4433 intr |= INTR_TYPE_SOFT_INTR;
4434 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4435 vmx->vcpu.arch.event_exit_inst_len);
4437 intr |= INTR_TYPE_EXT_INTR;
4438 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4440 vmx_clear_hlt(vcpu);
4443 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4445 struct vcpu_vmx *vmx = to_vmx(vcpu);
4449 * Tracking the NMI-blocked state in software is built upon
4450 * finding the next open IRQ window. This, in turn, depends on
4451 * well-behaving guests: They have to keep IRQs disabled at
4452 * least as long as the NMI handler runs. Otherwise we may
4453 * cause NMI nesting, maybe breaking the guest. But as this is
4454 * highly unlikely, we can live with the residual risk.
4456 vmx->loaded_vmcs->soft_vnmi_blocked = 1;
4457 vmx->loaded_vmcs->vnmi_blocked_time = 0;
4460 ++vcpu->stat.nmi_injections;
4461 vmx->loaded_vmcs->nmi_known_unmasked = false;
4463 if (vmx->rmode.vm86_active) {
4464 kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0);
4468 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4469 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4471 vmx_clear_hlt(vcpu);
4474 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4476 struct vcpu_vmx *vmx = to_vmx(vcpu);
4480 return vmx->loaded_vmcs->soft_vnmi_blocked;
4481 if (vmx->loaded_vmcs->nmi_known_unmasked)
4483 masked = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4484 vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4488 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4490 struct vcpu_vmx *vmx = to_vmx(vcpu);
4493 if (vmx->loaded_vmcs->soft_vnmi_blocked != masked) {
4494 vmx->loaded_vmcs->soft_vnmi_blocked = masked;
4495 vmx->loaded_vmcs->vnmi_blocked_time = 0;
4498 vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4500 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4501 GUEST_INTR_STATE_NMI);
4503 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4504 GUEST_INTR_STATE_NMI);
4508 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
4510 if (to_vmx(vcpu)->nested.nested_run_pending)
4514 to_vmx(vcpu)->loaded_vmcs->soft_vnmi_blocked)
4517 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4518 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
4519 | GUEST_INTR_STATE_NMI));
4522 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
4524 if (to_vmx(vcpu)->nested.nested_run_pending)
4527 if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu))
4530 return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
4531 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4532 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4535 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4539 if (enable_unrestricted_guest)
4542 mutex_lock(&kvm->slots_lock);
4543 ret = __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
4545 mutex_unlock(&kvm->slots_lock);
4549 to_kvm_vmx(kvm)->tss_addr = addr;
4550 return init_rmode_tss(kvm);
4553 static int vmx_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
4555 to_kvm_vmx(kvm)->ept_identity_map_addr = ident_addr;
4559 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
4564 * Update instruction length as we may reinject the exception
4565 * from user space while in guest debugging mode.
4567 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4568 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4569 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4573 if (vcpu->guest_debug &
4574 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
4590 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4591 int vec, u32 err_code)
4594 * Instruction with address size override prefix opcode 0x67
4595 * Cause the #SS fault with 0 error code in VM86 mode.
4597 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
4598 if (kvm_emulate_instruction(vcpu, 0)) {
4599 if (vcpu->arch.halt_request) {
4600 vcpu->arch.halt_request = 0;
4601 return kvm_vcpu_halt(vcpu);
4609 * Forward all other exceptions that are valid in real mode.
4610 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4611 * the required debugging infrastructure rework.
4613 kvm_queue_exception(vcpu, vec);
4618 * Trigger machine check on the host. We assume all the MSRs are already set up
4619 * by the CPU and that we still run on the same CPU as the MCE occurred on.
4620 * We pass a fake environment to the machine check handler because we want
4621 * the guest to be always treated like user space, no matter what context
4622 * it used internally.
4624 static void kvm_machine_check(void)
4626 #if defined(CONFIG_X86_MCE)
4627 struct pt_regs regs = {
4628 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
4629 .flags = X86_EFLAGS_IF,
4632 do_machine_check(®s, 0);
4636 static int handle_machine_check(struct kvm_vcpu *vcpu)
4638 /* handled by vmx_vcpu_run() */
4642 static int handle_exception_nmi(struct kvm_vcpu *vcpu)
4644 struct vcpu_vmx *vmx = to_vmx(vcpu);
4645 struct kvm_run *kvm_run = vcpu->run;
4646 u32 intr_info, ex_no, error_code;
4647 unsigned long cr2, rip, dr6;
4650 vect_info = vmx->idt_vectoring_info;
4651 intr_info = vmx->exit_intr_info;
4653 if (is_machine_check(intr_info) || is_nmi(intr_info))
4654 return 1; /* handled by handle_exception_nmi_irqoff() */
4656 if (is_invalid_opcode(intr_info))
4657 return handle_ud(vcpu);
4660 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4661 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4663 if (!vmx->rmode.vm86_active && is_gp_fault(intr_info)) {
4664 WARN_ON_ONCE(!enable_vmware_backdoor);
4667 * VMware backdoor emulation on #GP interception only handles
4668 * IN{S}, OUT{S}, and RDPMC, none of which generate a non-zero
4669 * error code on #GP.
4672 kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
4675 return kvm_emulate_instruction(vcpu, EMULTYPE_VMWARE_GP);
4679 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
4680 * MMIO, it is better to report an internal error.
4681 * See the comments in vmx_handle_exit.
4683 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4684 !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
4685 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4686 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4687 vcpu->run->internal.ndata = 3;
4688 vcpu->run->internal.data[0] = vect_info;
4689 vcpu->run->internal.data[1] = intr_info;
4690 vcpu->run->internal.data[2] = error_code;
4694 if (is_page_fault(intr_info)) {
4695 cr2 = vmcs_readl(EXIT_QUALIFICATION);
4696 /* EPT won't cause page fault directly */
4697 WARN_ON_ONCE(!vcpu->arch.apf.host_apf_reason && enable_ept);
4698 return kvm_handle_page_fault(vcpu, error_code, cr2, NULL, 0);
4701 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4703 if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
4704 return handle_rmode_exception(vcpu, ex_no, error_code);
4708 kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
4711 dr6 = vmcs_readl(EXIT_QUALIFICATION);
4712 if (!(vcpu->guest_debug &
4713 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4714 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
4715 vcpu->arch.dr6 |= dr6 | DR6_RTM;
4716 if (is_icebp(intr_info))
4717 WARN_ON(!skip_emulated_instruction(vcpu));
4719 kvm_queue_exception(vcpu, DB_VECTOR);
4722 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
4723 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4727 * Update instruction length as we may reinject #BP from
4728 * user space while in guest debugging mode. Reading it for
4729 * #DB as well causes no harm, it is not used in that case.
4731 vmx->vcpu.arch.event_exit_inst_len =
4732 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4733 kvm_run->exit_reason = KVM_EXIT_DEBUG;
4734 rip = kvm_rip_read(vcpu);
4735 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4736 kvm_run->debug.arch.exception = ex_no;
4739 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4740 kvm_run->ex.exception = ex_no;
4741 kvm_run->ex.error_code = error_code;
4747 static __always_inline int handle_external_interrupt(struct kvm_vcpu *vcpu)
4749 ++vcpu->stat.irq_exits;
4753 static int handle_triple_fault(struct kvm_vcpu *vcpu)
4755 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4756 vcpu->mmio_needed = 0;
4760 static int handle_io(struct kvm_vcpu *vcpu)
4762 unsigned long exit_qualification;
4763 int size, in, string;
4766 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4767 string = (exit_qualification & 16) != 0;
4769 ++vcpu->stat.io_exits;
4772 return kvm_emulate_instruction(vcpu, 0);
4774 port = exit_qualification >> 16;
4775 size = (exit_qualification & 7) + 1;
4776 in = (exit_qualification & 8) != 0;
4778 return kvm_fast_pio(vcpu, size, port, in);
4782 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4785 * Patch in the VMCALL instruction:
4787 hypercall[0] = 0x0f;
4788 hypercall[1] = 0x01;
4789 hypercall[2] = 0xc1;
4792 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
4793 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
4795 if (is_guest_mode(vcpu)) {
4796 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4797 unsigned long orig_val = val;
4800 * We get here when L2 changed cr0 in a way that did not change
4801 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
4802 * but did change L0 shadowed bits. So we first calculate the
4803 * effective cr0 value that L1 would like to write into the
4804 * hardware. It consists of the L2-owned bits from the new
4805 * value combined with the L1-owned bits from L1's guest_cr0.
4807 val = (val & ~vmcs12->cr0_guest_host_mask) |
4808 (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
4810 if (!nested_guest_cr0_valid(vcpu, val))
4813 if (kvm_set_cr0(vcpu, val))
4815 vmcs_writel(CR0_READ_SHADOW, orig_val);
4818 if (to_vmx(vcpu)->nested.vmxon &&
4819 !nested_host_cr0_valid(vcpu, val))
4822 return kvm_set_cr0(vcpu, val);
4826 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
4828 if (is_guest_mode(vcpu)) {
4829 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4830 unsigned long orig_val = val;
4832 /* analogously to handle_set_cr0 */
4833 val = (val & ~vmcs12->cr4_guest_host_mask) |
4834 (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
4835 if (kvm_set_cr4(vcpu, val))
4837 vmcs_writel(CR4_READ_SHADOW, orig_val);
4840 return kvm_set_cr4(vcpu, val);
4843 static int handle_desc(struct kvm_vcpu *vcpu)
4845 WARN_ON(!(vcpu->arch.cr4 & X86_CR4_UMIP));
4846 return kvm_emulate_instruction(vcpu, 0);
4849 static int handle_cr(struct kvm_vcpu *vcpu)
4851 unsigned long exit_qualification, val;
4857 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4858 cr = exit_qualification & 15;
4859 reg = (exit_qualification >> 8) & 15;
4860 switch ((exit_qualification >> 4) & 3) {
4861 case 0: /* mov to cr */
4862 val = kvm_register_readl(vcpu, reg);
4863 trace_kvm_cr_write(cr, val);
4866 err = handle_set_cr0(vcpu, val);
4867 return kvm_complete_insn_gp(vcpu, err);
4869 WARN_ON_ONCE(enable_unrestricted_guest);
4870 err = kvm_set_cr3(vcpu, val);
4871 return kvm_complete_insn_gp(vcpu, err);
4873 err = handle_set_cr4(vcpu, val);
4874 return kvm_complete_insn_gp(vcpu, err);
4876 u8 cr8_prev = kvm_get_cr8(vcpu);
4878 err = kvm_set_cr8(vcpu, cr8);
4879 ret = kvm_complete_insn_gp(vcpu, err);
4880 if (lapic_in_kernel(vcpu))
4882 if (cr8_prev <= cr8)
4885 * TODO: we might be squashing a
4886 * KVM_GUESTDBG_SINGLESTEP-triggered
4887 * KVM_EXIT_DEBUG here.
4889 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
4895 WARN_ONCE(1, "Guest should always own CR0.TS");
4896 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4897 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
4898 return kvm_skip_emulated_instruction(vcpu);
4899 case 1: /*mov from cr*/
4902 WARN_ON_ONCE(enable_unrestricted_guest);
4903 val = kvm_read_cr3(vcpu);
4904 kvm_register_write(vcpu, reg, val);
4905 trace_kvm_cr_read(cr, val);
4906 return kvm_skip_emulated_instruction(vcpu);
4908 val = kvm_get_cr8(vcpu);
4909 kvm_register_write(vcpu, reg, val);
4910 trace_kvm_cr_read(cr, val);
4911 return kvm_skip_emulated_instruction(vcpu);
4915 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
4916 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
4917 kvm_lmsw(vcpu, val);
4919 return kvm_skip_emulated_instruction(vcpu);
4923 vcpu->run->exit_reason = 0;
4924 vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
4925 (int)(exit_qualification >> 4) & 3, cr);
4929 static int handle_dr(struct kvm_vcpu *vcpu)
4931 unsigned long exit_qualification;
4934 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4935 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
4937 /* First, if DR does not exist, trigger UD */
4938 if (!kvm_require_dr(vcpu, dr))
4941 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
4942 if (!kvm_require_cpl(vcpu, 0))
4944 dr7 = vmcs_readl(GUEST_DR7);
4947 * As the vm-exit takes precedence over the debug trap, we
4948 * need to emulate the latter, either for the host or the
4949 * guest debugging itself.
4951 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
4952 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
4953 vcpu->run->debug.arch.dr7 = dr7;
4954 vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
4955 vcpu->run->debug.arch.exception = DB_VECTOR;
4956 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
4959 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
4960 vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
4961 kvm_queue_exception(vcpu, DB_VECTOR);
4966 if (vcpu->guest_debug == 0) {
4967 exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
4970 * No more DR vmexits; force a reload of the debug registers
4971 * and reenter on this instruction. The next vmexit will
4972 * retrieve the full state of the debug registers.
4974 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
4978 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
4979 if (exit_qualification & TYPE_MOV_FROM_DR) {
4982 if (kvm_get_dr(vcpu, dr, &val))
4984 kvm_register_write(vcpu, reg, val);
4986 if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
4989 return kvm_skip_emulated_instruction(vcpu);
4992 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
4994 return vcpu->arch.dr6;
4997 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
5001 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
5003 get_debugreg(vcpu->arch.db[0], 0);
5004 get_debugreg(vcpu->arch.db[1], 1);
5005 get_debugreg(vcpu->arch.db[2], 2);
5006 get_debugreg(vcpu->arch.db[3], 3);
5007 get_debugreg(vcpu->arch.dr6, 6);
5008 vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
5010 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
5011 exec_controls_setbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
5014 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
5016 vmcs_writel(GUEST_DR7, val);
5019 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
5021 kvm_apic_update_ppr(vcpu);
5025 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
5027 exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_INTR_WINDOW_EXITING);
5029 kvm_make_request(KVM_REQ_EVENT, vcpu);
5031 ++vcpu->stat.irq_window_exits;
5035 static int handle_vmcall(struct kvm_vcpu *vcpu)
5037 return kvm_emulate_hypercall(vcpu);
5040 static int handle_invd(struct kvm_vcpu *vcpu)
5042 return kvm_emulate_instruction(vcpu, 0);
5045 static int handle_invlpg(struct kvm_vcpu *vcpu)
5047 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5049 kvm_mmu_invlpg(vcpu, exit_qualification);
5050 return kvm_skip_emulated_instruction(vcpu);
5053 static int handle_rdpmc(struct kvm_vcpu *vcpu)
5057 err = kvm_rdpmc(vcpu);
5058 return kvm_complete_insn_gp(vcpu, err);
5061 static int handle_wbinvd(struct kvm_vcpu *vcpu)
5063 return kvm_emulate_wbinvd(vcpu);
5066 static int handle_xsetbv(struct kvm_vcpu *vcpu)
5068 u64 new_bv = kvm_read_edx_eax(vcpu);
5069 u32 index = kvm_rcx_read(vcpu);
5071 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
5072 return kvm_skip_emulated_instruction(vcpu);
5076 static int handle_apic_access(struct kvm_vcpu *vcpu)
5078 if (likely(fasteoi)) {
5079 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5080 int access_type, offset;
5082 access_type = exit_qualification & APIC_ACCESS_TYPE;
5083 offset = exit_qualification & APIC_ACCESS_OFFSET;
5085 * Sane guest uses MOV to write EOI, with written value
5086 * not cared. So make a short-circuit here by avoiding
5087 * heavy instruction emulation.
5089 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5090 (offset == APIC_EOI)) {
5091 kvm_lapic_set_eoi(vcpu);
5092 return kvm_skip_emulated_instruction(vcpu);
5095 return kvm_emulate_instruction(vcpu, 0);
5098 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5100 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5101 int vector = exit_qualification & 0xff;
5103 /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5104 kvm_apic_set_eoi_accelerated(vcpu, vector);
5108 static int handle_apic_write(struct kvm_vcpu *vcpu)
5110 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5111 u32 offset = exit_qualification & 0xfff;
5113 /* APIC-write VM exit is trap-like and thus no need to adjust IP */
5114 kvm_apic_write_nodecode(vcpu, offset);
5118 static int handle_task_switch(struct kvm_vcpu *vcpu)
5120 struct vcpu_vmx *vmx = to_vmx(vcpu);
5121 unsigned long exit_qualification;
5122 bool has_error_code = false;
5125 int reason, type, idt_v, idt_index;
5127 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5128 idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5129 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5131 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5133 reason = (u32)exit_qualification >> 30;
5134 if (reason == TASK_SWITCH_GATE && idt_v) {
5136 case INTR_TYPE_NMI_INTR:
5137 vcpu->arch.nmi_injected = false;
5138 vmx_set_nmi_mask(vcpu, true);
5140 case INTR_TYPE_EXT_INTR:
5141 case INTR_TYPE_SOFT_INTR:
5142 kvm_clear_interrupt_queue(vcpu);
5144 case INTR_TYPE_HARD_EXCEPTION:
5145 if (vmx->idt_vectoring_info &
5146 VECTORING_INFO_DELIVER_CODE_MASK) {
5147 has_error_code = true;
5149 vmcs_read32(IDT_VECTORING_ERROR_CODE);
5152 case INTR_TYPE_SOFT_EXCEPTION:
5153 kvm_clear_exception_queue(vcpu);
5159 tss_selector = exit_qualification;
5161 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5162 type != INTR_TYPE_EXT_INTR &&
5163 type != INTR_TYPE_NMI_INTR))
5164 WARN_ON(!skip_emulated_instruction(vcpu));
5167 * TODO: What about debug traps on tss switch?
5168 * Are we supposed to inject them and update dr6?
5170 return kvm_task_switch(vcpu, tss_selector,
5171 type == INTR_TYPE_SOFT_INTR ? idt_index : -1,
5172 reason, has_error_code, error_code);
5175 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5177 unsigned long exit_qualification;
5181 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5184 * EPT violation happened while executing iret from NMI,
5185 * "blocked by NMI" bit has to be set before next VM entry.
5186 * There are errata that may cause this bit to not be set:
5189 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5191 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
5192 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5194 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5195 trace_kvm_page_fault(gpa, exit_qualification);
5197 /* Is it a read fault? */
5198 error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
5199 ? PFERR_USER_MASK : 0;
5200 /* Is it a write fault? */
5201 error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
5202 ? PFERR_WRITE_MASK : 0;
5203 /* Is it a fetch fault? */
5204 error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
5205 ? PFERR_FETCH_MASK : 0;
5206 /* ept page table entry is present? */
5207 error_code |= (exit_qualification &
5208 (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE |
5209 EPT_VIOLATION_EXECUTABLE))
5210 ? PFERR_PRESENT_MASK : 0;
5212 error_code |= (exit_qualification & 0x100) != 0 ?
5213 PFERR_GUEST_FINAL_MASK : PFERR_GUEST_PAGE_MASK;
5215 vcpu->arch.exit_qualification = exit_qualification;
5216 return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5219 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5224 * A nested guest cannot optimize MMIO vmexits, because we have an
5225 * nGPA here instead of the required GPA.
5227 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5228 if (!is_guest_mode(vcpu) &&
5229 !kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5230 trace_kvm_fast_mmio(gpa);
5231 return kvm_skip_emulated_instruction(vcpu);
5234 return kvm_mmu_page_fault(vcpu, gpa, PFERR_RSVD_MASK, NULL, 0);
5237 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5239 WARN_ON_ONCE(!enable_vnmi);
5240 exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_NMI_WINDOW_EXITING);
5241 ++vcpu->stat.nmi_window_exits;
5242 kvm_make_request(KVM_REQ_EVENT, vcpu);
5247 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5249 struct vcpu_vmx *vmx = to_vmx(vcpu);
5250 bool intr_window_requested;
5251 unsigned count = 130;
5254 * We should never reach the point where we are emulating L2
5255 * due to invalid guest state as that means we incorrectly
5256 * allowed a nested VMEntry with an invalid vmcs12.
5258 WARN_ON_ONCE(vmx->emulation_required && vmx->nested.nested_run_pending);
5260 intr_window_requested = exec_controls_get(vmx) &
5261 CPU_BASED_INTR_WINDOW_EXITING;
5263 while (vmx->emulation_required && count-- != 0) {
5264 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
5265 return handle_interrupt_window(&vmx->vcpu);
5267 if (kvm_test_request(KVM_REQ_EVENT, vcpu))
5270 if (!kvm_emulate_instruction(vcpu, 0))
5273 if (vmx->emulation_required && !vmx->rmode.vm86_active &&
5274 vcpu->arch.exception.pending) {
5275 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5276 vcpu->run->internal.suberror =
5277 KVM_INTERNAL_ERROR_EMULATION;
5278 vcpu->run->internal.ndata = 0;
5282 if (vcpu->arch.halt_request) {
5283 vcpu->arch.halt_request = 0;
5284 return kvm_vcpu_halt(vcpu);
5288 * Note, return 1 and not 0, vcpu_run() is responsible for
5289 * morphing the pending signal into the proper return code.
5291 if (signal_pending(current))
5301 static void grow_ple_window(struct kvm_vcpu *vcpu)
5303 struct vcpu_vmx *vmx = to_vmx(vcpu);
5304 unsigned int old = vmx->ple_window;
5306 vmx->ple_window = __grow_ple_window(old, ple_window,
5310 if (vmx->ple_window != old) {
5311 vmx->ple_window_dirty = true;
5312 trace_kvm_ple_window_update(vcpu->vcpu_id,
5313 vmx->ple_window, old);
5317 static void shrink_ple_window(struct kvm_vcpu *vcpu)
5319 struct vcpu_vmx *vmx = to_vmx(vcpu);
5320 unsigned int old = vmx->ple_window;
5322 vmx->ple_window = __shrink_ple_window(old, ple_window,
5326 if (vmx->ple_window != old) {
5327 vmx->ple_window_dirty = true;
5328 trace_kvm_ple_window_update(vcpu->vcpu_id,
5329 vmx->ple_window, old);
5334 * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
5336 static void wakeup_handler(void)
5338 struct kvm_vcpu *vcpu;
5339 int cpu = smp_processor_id();
5341 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5342 list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
5343 blocked_vcpu_list) {
5344 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
5346 if (pi_test_on(pi_desc) == 1)
5347 kvm_vcpu_kick(vcpu);
5349 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5352 static void vmx_enable_tdp(void)
5354 kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK,
5355 enable_ept_ad_bits ? VMX_EPT_ACCESS_BIT : 0ull,
5356 enable_ept_ad_bits ? VMX_EPT_DIRTY_BIT : 0ull,
5357 0ull, VMX_EPT_EXECUTABLE_MASK,
5358 cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK,
5359 VMX_EPT_RWX_MASK, 0ull);
5361 ept_set_mmio_spte_mask();
5365 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5366 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5368 static int handle_pause(struct kvm_vcpu *vcpu)
5370 if (!kvm_pause_in_guest(vcpu->kvm))
5371 grow_ple_window(vcpu);
5374 * Intel sdm vol3 ch-25.1.3 says: The "PAUSE-loop exiting"
5375 * VM-execution control is ignored if CPL > 0. OTOH, KVM
5376 * never set PAUSE_EXITING and just set PLE if supported,
5377 * so the vcpu must be CPL=0 if it gets a PAUSE exit.
5379 kvm_vcpu_on_spin(vcpu, true);
5380 return kvm_skip_emulated_instruction(vcpu);
5383 static int handle_nop(struct kvm_vcpu *vcpu)
5385 return kvm_skip_emulated_instruction(vcpu);
5388 static int handle_mwait(struct kvm_vcpu *vcpu)
5390 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
5391 return handle_nop(vcpu);
5394 static int handle_invalid_op(struct kvm_vcpu *vcpu)
5396 kvm_queue_exception(vcpu, UD_VECTOR);
5400 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
5405 static int handle_monitor(struct kvm_vcpu *vcpu)
5407 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
5408 return handle_nop(vcpu);
5411 static int handle_invpcid(struct kvm_vcpu *vcpu)
5413 u32 vmx_instruction_info;
5417 struct x86_exception e;
5419 unsigned long roots_to_free = 0;
5425 if (!guest_cpuid_has(vcpu, X86_FEATURE_INVPCID)) {
5426 kvm_queue_exception(vcpu, UD_VECTOR);
5430 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5431 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
5434 kvm_inject_gp(vcpu, 0);
5438 /* According to the Intel instruction reference, the memory operand
5439 * is read even if it isn't needed (e.g., for type==all)
5441 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5442 vmx_instruction_info, false,
5443 sizeof(operand), &gva))
5446 if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
5447 kvm_inject_emulated_page_fault(vcpu, &e);
5451 if (operand.pcid >> 12 != 0) {
5452 kvm_inject_gp(vcpu, 0);
5456 pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
5459 case INVPCID_TYPE_INDIV_ADDR:
5460 if ((!pcid_enabled && (operand.pcid != 0)) ||
5461 is_noncanonical_address(operand.gla, vcpu)) {
5462 kvm_inject_gp(vcpu, 0);
5465 kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
5466 return kvm_skip_emulated_instruction(vcpu);
5468 case INVPCID_TYPE_SINGLE_CTXT:
5469 if (!pcid_enabled && (operand.pcid != 0)) {
5470 kvm_inject_gp(vcpu, 0);
5474 if (kvm_get_active_pcid(vcpu) == operand.pcid) {
5475 kvm_mmu_sync_roots(vcpu);
5476 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
5479 for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
5480 if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].pgd)
5482 roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
5484 kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
5486 * If neither the current cr3 nor any of the prev_roots use the
5487 * given PCID, then nothing needs to be done here because a
5488 * resync will happen anyway before switching to any other CR3.
5491 return kvm_skip_emulated_instruction(vcpu);
5493 case INVPCID_TYPE_ALL_NON_GLOBAL:
5495 * Currently, KVM doesn't mark global entries in the shadow
5496 * page tables, so a non-global flush just degenerates to a
5497 * global flush. If needed, we could optimize this later by
5498 * keeping track of global entries in shadow page tables.
5502 case INVPCID_TYPE_ALL_INCL_GLOBAL:
5503 kvm_mmu_unload(vcpu);
5504 return kvm_skip_emulated_instruction(vcpu);
5507 BUG(); /* We have already checked above that type <= 3 */
5511 static int handle_pml_full(struct kvm_vcpu *vcpu)
5513 unsigned long exit_qualification;
5515 trace_kvm_pml_full(vcpu->vcpu_id);
5517 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5520 * PML buffer FULL happened while executing iret from NMI,
5521 * "blocked by NMI" bit has to be set before next VM entry.
5523 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5525 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
5526 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5527 GUEST_INTR_STATE_NMI);
5530 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
5531 * here.., and there's no userspace involvement needed for PML.
5536 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
5538 struct vcpu_vmx *vmx = to_vmx(vcpu);
5540 if (!vmx->req_immediate_exit &&
5541 !unlikely(vmx->loaded_vmcs->hv_timer_soft_disabled))
5542 kvm_lapic_expired_hv_timer(vcpu);
5548 * When nested=0, all VMX instruction VM Exits filter here. The handlers
5549 * are overwritten by nested_vmx_setup() when nested=1.
5551 static int handle_vmx_instruction(struct kvm_vcpu *vcpu)
5553 kvm_queue_exception(vcpu, UD_VECTOR);
5557 static int handle_encls(struct kvm_vcpu *vcpu)
5560 * SGX virtualization is not yet supported. There is no software
5561 * enable bit for SGX, so we have to trap ENCLS and inject a #UD
5562 * to prevent the guest from executing ENCLS.
5564 kvm_queue_exception(vcpu, UD_VECTOR);
5569 * The exit handlers return 1 if the exit was handled fully and guest execution
5570 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
5571 * to be done to userspace and return 0.
5573 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
5574 [EXIT_REASON_EXCEPTION_NMI] = handle_exception_nmi,
5575 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
5576 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
5577 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
5578 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
5579 [EXIT_REASON_CR_ACCESS] = handle_cr,
5580 [EXIT_REASON_DR_ACCESS] = handle_dr,
5581 [EXIT_REASON_CPUID] = kvm_emulate_cpuid,
5582 [EXIT_REASON_MSR_READ] = kvm_emulate_rdmsr,
5583 [EXIT_REASON_MSR_WRITE] = kvm_emulate_wrmsr,
5584 [EXIT_REASON_INTERRUPT_WINDOW] = handle_interrupt_window,
5585 [EXIT_REASON_HLT] = kvm_emulate_halt,
5586 [EXIT_REASON_INVD] = handle_invd,
5587 [EXIT_REASON_INVLPG] = handle_invlpg,
5588 [EXIT_REASON_RDPMC] = handle_rdpmc,
5589 [EXIT_REASON_VMCALL] = handle_vmcall,
5590 [EXIT_REASON_VMCLEAR] = handle_vmx_instruction,
5591 [EXIT_REASON_VMLAUNCH] = handle_vmx_instruction,
5592 [EXIT_REASON_VMPTRLD] = handle_vmx_instruction,
5593 [EXIT_REASON_VMPTRST] = handle_vmx_instruction,
5594 [EXIT_REASON_VMREAD] = handle_vmx_instruction,
5595 [EXIT_REASON_VMRESUME] = handle_vmx_instruction,
5596 [EXIT_REASON_VMWRITE] = handle_vmx_instruction,
5597 [EXIT_REASON_VMOFF] = handle_vmx_instruction,
5598 [EXIT_REASON_VMON] = handle_vmx_instruction,
5599 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
5600 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
5601 [EXIT_REASON_APIC_WRITE] = handle_apic_write,
5602 [EXIT_REASON_EOI_INDUCED] = handle_apic_eoi_induced,
5603 [EXIT_REASON_WBINVD] = handle_wbinvd,
5604 [EXIT_REASON_XSETBV] = handle_xsetbv,
5605 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
5606 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
5607 [EXIT_REASON_GDTR_IDTR] = handle_desc,
5608 [EXIT_REASON_LDTR_TR] = handle_desc,
5609 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
5610 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
5611 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
5612 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_mwait,
5613 [EXIT_REASON_MONITOR_TRAP_FLAG] = handle_monitor_trap,
5614 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_monitor,
5615 [EXIT_REASON_INVEPT] = handle_vmx_instruction,
5616 [EXIT_REASON_INVVPID] = handle_vmx_instruction,
5617 [EXIT_REASON_RDRAND] = handle_invalid_op,
5618 [EXIT_REASON_RDSEED] = handle_invalid_op,
5619 [EXIT_REASON_PML_FULL] = handle_pml_full,
5620 [EXIT_REASON_INVPCID] = handle_invpcid,
5621 [EXIT_REASON_VMFUNC] = handle_vmx_instruction,
5622 [EXIT_REASON_PREEMPTION_TIMER] = handle_preemption_timer,
5623 [EXIT_REASON_ENCLS] = handle_encls,
5626 static const int kvm_vmx_max_exit_handlers =
5627 ARRAY_SIZE(kvm_vmx_exit_handlers);
5629 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
5631 *info1 = vmcs_readl(EXIT_QUALIFICATION);
5632 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
5635 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
5638 __free_page(vmx->pml_pg);
5643 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
5645 struct vcpu_vmx *vmx = to_vmx(vcpu);
5649 pml_idx = vmcs_read16(GUEST_PML_INDEX);
5651 /* Do nothing if PML buffer is empty */
5652 if (pml_idx == (PML_ENTITY_NUM - 1))
5655 /* PML index always points to next available PML buffer entity */
5656 if (pml_idx >= PML_ENTITY_NUM)
5661 pml_buf = page_address(vmx->pml_pg);
5662 for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
5665 gpa = pml_buf[pml_idx];
5666 WARN_ON(gpa & (PAGE_SIZE - 1));
5667 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5670 /* reset PML index */
5671 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
5675 * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
5676 * Called before reporting dirty_bitmap to userspace.
5678 static void kvm_flush_pml_buffers(struct kvm *kvm)
5681 struct kvm_vcpu *vcpu;
5683 * We only need to kick vcpu out of guest mode here, as PML buffer
5684 * is flushed at beginning of all VMEXITs, and it's obvious that only
5685 * vcpus running in guest are possible to have unflushed GPAs in PML
5688 kvm_for_each_vcpu(i, vcpu, kvm)
5689 kvm_vcpu_kick(vcpu);
5692 static void vmx_dump_sel(char *name, uint32_t sel)
5694 pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
5695 name, vmcs_read16(sel),
5696 vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
5697 vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
5698 vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
5701 static void vmx_dump_dtsel(char *name, uint32_t limit)
5703 pr_err("%s limit=0x%08x, base=0x%016lx\n",
5704 name, vmcs_read32(limit),
5705 vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
5708 void dump_vmcs(void)
5710 u32 vmentry_ctl, vmexit_ctl;
5711 u32 cpu_based_exec_ctrl, pin_based_exec_ctrl, secondary_exec_control;
5716 if (!dump_invalid_vmcs) {
5717 pr_warn_ratelimited("set kvm_intel.dump_invalid_vmcs=1 to dump internal KVM state.\n");
5721 vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
5722 vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
5723 cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5724 pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
5725 cr4 = vmcs_readl(GUEST_CR4);
5726 efer = vmcs_read64(GUEST_IA32_EFER);
5727 secondary_exec_control = 0;
5728 if (cpu_has_secondary_exec_ctrls())
5729 secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
5731 pr_err("*** Guest State ***\n");
5732 pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5733 vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
5734 vmcs_readl(CR0_GUEST_HOST_MASK));
5735 pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5736 cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
5737 pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
5738 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
5739 (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
5741 pr_err("PDPTR0 = 0x%016llx PDPTR1 = 0x%016llx\n",
5742 vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
5743 pr_err("PDPTR2 = 0x%016llx PDPTR3 = 0x%016llx\n",
5744 vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
5746 pr_err("RSP = 0x%016lx RIP = 0x%016lx\n",
5747 vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
5748 pr_err("RFLAGS=0x%08lx DR7 = 0x%016lx\n",
5749 vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
5750 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5751 vmcs_readl(GUEST_SYSENTER_ESP),
5752 vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
5753 vmx_dump_sel("CS: ", GUEST_CS_SELECTOR);
5754 vmx_dump_sel("DS: ", GUEST_DS_SELECTOR);
5755 vmx_dump_sel("SS: ", GUEST_SS_SELECTOR);
5756 vmx_dump_sel("ES: ", GUEST_ES_SELECTOR);
5757 vmx_dump_sel("FS: ", GUEST_FS_SELECTOR);
5758 vmx_dump_sel("GS: ", GUEST_GS_SELECTOR);
5759 vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
5760 vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
5761 vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
5762 vmx_dump_sel("TR: ", GUEST_TR_SELECTOR);
5763 if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
5764 (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
5765 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
5766 efer, vmcs_read64(GUEST_IA32_PAT));
5767 pr_err("DebugCtl = 0x%016llx DebugExceptions = 0x%016lx\n",
5768 vmcs_read64(GUEST_IA32_DEBUGCTL),
5769 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
5770 if (cpu_has_load_perf_global_ctrl() &&
5771 vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
5772 pr_err("PerfGlobCtl = 0x%016llx\n",
5773 vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
5774 if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
5775 pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
5776 pr_err("Interruptibility = %08x ActivityState = %08x\n",
5777 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
5778 vmcs_read32(GUEST_ACTIVITY_STATE));
5779 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
5780 pr_err("InterruptStatus = %04x\n",
5781 vmcs_read16(GUEST_INTR_STATUS));
5783 pr_err("*** Host State ***\n");
5784 pr_err("RIP = 0x%016lx RSP = 0x%016lx\n",
5785 vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
5786 pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
5787 vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
5788 vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
5789 vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
5790 vmcs_read16(HOST_TR_SELECTOR));
5791 pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
5792 vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
5793 vmcs_readl(HOST_TR_BASE));
5794 pr_err("GDTBase=%016lx IDTBase=%016lx\n",
5795 vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
5796 pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
5797 vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
5798 vmcs_readl(HOST_CR4));
5799 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5800 vmcs_readl(HOST_IA32_SYSENTER_ESP),
5801 vmcs_read32(HOST_IA32_SYSENTER_CS),
5802 vmcs_readl(HOST_IA32_SYSENTER_EIP));
5803 if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
5804 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
5805 vmcs_read64(HOST_IA32_EFER),
5806 vmcs_read64(HOST_IA32_PAT));
5807 if (cpu_has_load_perf_global_ctrl() &&
5808 vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
5809 pr_err("PerfGlobCtl = 0x%016llx\n",
5810 vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
5812 pr_err("*** Control State ***\n");
5813 pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
5814 pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
5815 pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
5816 pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
5817 vmcs_read32(EXCEPTION_BITMAP),
5818 vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
5819 vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
5820 pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
5821 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
5822 vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
5823 vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
5824 pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
5825 vmcs_read32(VM_EXIT_INTR_INFO),
5826 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
5827 vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
5828 pr_err(" reason=%08x qualification=%016lx\n",
5829 vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
5830 pr_err("IDTVectoring: info=%08x errcode=%08x\n",
5831 vmcs_read32(IDT_VECTORING_INFO_FIELD),
5832 vmcs_read32(IDT_VECTORING_ERROR_CODE));
5833 pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
5834 if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
5835 pr_err("TSC Multiplier = 0x%016llx\n",
5836 vmcs_read64(TSC_MULTIPLIER));
5837 if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW) {
5838 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
5839 u16 status = vmcs_read16(GUEST_INTR_STATUS);
5840 pr_err("SVI|RVI = %02x|%02x ", status >> 8, status & 0xff);
5842 pr_cont("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
5843 if (secondary_exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
5844 pr_err("APIC-access addr = 0x%016llx ", vmcs_read64(APIC_ACCESS_ADDR));
5845 pr_cont("virt-APIC addr = 0x%016llx\n", vmcs_read64(VIRTUAL_APIC_PAGE_ADDR));
5847 if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
5848 pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
5849 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
5850 pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
5851 n = vmcs_read32(CR3_TARGET_COUNT);
5852 for (i = 0; i + 1 < n; i += 4)
5853 pr_err("CR3 target%u=%016lx target%u=%016lx\n",
5854 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
5855 i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
5857 pr_err("CR3 target%u=%016lx\n",
5858 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
5859 if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
5860 pr_err("PLE Gap=%08x Window=%08x\n",
5861 vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
5862 if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
5863 pr_err("Virtual processor ID = 0x%04x\n",
5864 vmcs_read16(VIRTUAL_PROCESSOR_ID));
5868 * The guest has exited. See if we can fix it or if we need userspace
5871 static int vmx_handle_exit(struct kvm_vcpu *vcpu,
5872 enum exit_fastpath_completion exit_fastpath)
5874 struct vcpu_vmx *vmx = to_vmx(vcpu);
5875 u32 exit_reason = vmx->exit_reason;
5876 u32 vectoring_info = vmx->idt_vectoring_info;
5878 trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
5881 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
5882 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
5883 * querying dirty_bitmap, we only need to kick all vcpus out of guest
5884 * mode as if vcpus is in root mode, the PML buffer must has been
5888 vmx_flush_pml_buffer(vcpu);
5890 /* If guest state is invalid, start emulating */
5891 if (vmx->emulation_required)
5892 return handle_invalid_guest_state(vcpu);
5894 if (is_guest_mode(vcpu)) {
5896 * The host physical addresses of some pages of guest memory
5897 * are loaded into the vmcs02 (e.g. vmcs12's Virtual APIC
5898 * Page). The CPU may write to these pages via their host
5899 * physical address while L2 is running, bypassing any
5900 * address-translation-based dirty tracking (e.g. EPT write
5903 * Mark them dirty on every exit from L2 to prevent them from
5904 * getting out of sync with dirty tracking.
5906 nested_mark_vmcs12_pages_dirty(vcpu);
5908 if (nested_vmx_reflect_vmexit(vcpu))
5912 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
5914 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5915 vcpu->run->fail_entry.hardware_entry_failure_reason
5920 if (unlikely(vmx->fail)) {
5922 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5923 vcpu->run->fail_entry.hardware_entry_failure_reason
5924 = vmcs_read32(VM_INSTRUCTION_ERROR);
5930 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
5931 * delivery event since it indicates guest is accessing MMIO.
5932 * The vm-exit can be triggered again after return to guest that
5933 * will cause infinite loop.
5935 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
5936 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
5937 exit_reason != EXIT_REASON_EPT_VIOLATION &&
5938 exit_reason != EXIT_REASON_PML_FULL &&
5939 exit_reason != EXIT_REASON_TASK_SWITCH)) {
5940 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5941 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
5942 vcpu->run->internal.ndata = 3;
5943 vcpu->run->internal.data[0] = vectoring_info;
5944 vcpu->run->internal.data[1] = exit_reason;
5945 vcpu->run->internal.data[2] = vcpu->arch.exit_qualification;
5946 if (exit_reason == EXIT_REASON_EPT_MISCONFIG) {
5947 vcpu->run->internal.ndata++;
5948 vcpu->run->internal.data[3] =
5949 vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5954 if (unlikely(!enable_vnmi &&
5955 vmx->loaded_vmcs->soft_vnmi_blocked)) {
5956 if (vmx_interrupt_allowed(vcpu)) {
5957 vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5958 } else if (vmx->loaded_vmcs->vnmi_blocked_time > 1000000000LL &&
5959 vcpu->arch.nmi_pending) {
5961 * This CPU don't support us in finding the end of an
5962 * NMI-blocked window if the guest runs with IRQs
5963 * disabled. So we pull the trigger after 1 s of
5964 * futile waiting, but inform the user about this.
5966 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
5967 "state on VCPU %d after 1 s timeout\n",
5968 __func__, vcpu->vcpu_id);
5969 vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5973 if (exit_fastpath == EXIT_FASTPATH_SKIP_EMUL_INS) {
5974 kvm_skip_emulated_instruction(vcpu);
5978 if (exit_reason >= kvm_vmx_max_exit_handlers)
5979 goto unexpected_vmexit;
5980 #ifdef CONFIG_RETPOLINE
5981 if (exit_reason == EXIT_REASON_MSR_WRITE)
5982 return kvm_emulate_wrmsr(vcpu);
5983 else if (exit_reason == EXIT_REASON_PREEMPTION_TIMER)
5984 return handle_preemption_timer(vcpu);
5985 else if (exit_reason == EXIT_REASON_INTERRUPT_WINDOW)
5986 return handle_interrupt_window(vcpu);
5987 else if (exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
5988 return handle_external_interrupt(vcpu);
5989 else if (exit_reason == EXIT_REASON_HLT)
5990 return kvm_emulate_halt(vcpu);
5991 else if (exit_reason == EXIT_REASON_EPT_MISCONFIG)
5992 return handle_ept_misconfig(vcpu);
5995 exit_reason = array_index_nospec(exit_reason,
5996 kvm_vmx_max_exit_handlers);
5997 if (!kvm_vmx_exit_handlers[exit_reason])
5998 goto unexpected_vmexit;
6000 return kvm_vmx_exit_handlers[exit_reason](vcpu);
6003 vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n", exit_reason);
6005 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6006 vcpu->run->internal.suberror =
6007 KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON;
6008 vcpu->run->internal.ndata = 1;
6009 vcpu->run->internal.data[0] = exit_reason;
6014 * Software based L1D cache flush which is used when microcode providing
6015 * the cache control MSR is not loaded.
6017 * The L1D cache is 32 KiB on Nehalem and later microarchitectures, but to
6018 * flush it is required to read in 64 KiB because the replacement algorithm
6019 * is not exactly LRU. This could be sized at runtime via topology
6020 * information but as all relevant affected CPUs have 32KiB L1D cache size
6021 * there is no point in doing so.
6023 static void vmx_l1d_flush(struct kvm_vcpu *vcpu)
6025 int size = PAGE_SIZE << L1D_CACHE_ORDER;
6028 * This code is only executed when the the flush mode is 'cond' or
6031 if (static_branch_likely(&vmx_l1d_flush_cond)) {
6035 * Clear the per-vcpu flush bit, it gets set again
6036 * either from vcpu_run() or from one of the unsafe
6039 flush_l1d = vcpu->arch.l1tf_flush_l1d;
6040 vcpu->arch.l1tf_flush_l1d = false;
6043 * Clear the per-cpu flush bit, it gets set again from
6044 * the interrupt handlers.
6046 flush_l1d |= kvm_get_cpu_l1tf_flush_l1d();
6047 kvm_clear_cpu_l1tf_flush_l1d();
6053 vcpu->stat.l1d_flush++;
6055 if (static_cpu_has(X86_FEATURE_FLUSH_L1D)) {
6056 wrmsrl(MSR_IA32_FLUSH_CMD, L1D_FLUSH);
6061 /* First ensure the pages are in the TLB */
6062 "xorl %%eax, %%eax\n"
6063 ".Lpopulate_tlb:\n\t"
6064 "movzbl (%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
6065 "addl $4096, %%eax\n\t"
6066 "cmpl %%eax, %[size]\n\t"
6067 "jne .Lpopulate_tlb\n\t"
6068 "xorl %%eax, %%eax\n\t"
6070 /* Now fill the cache */
6071 "xorl %%eax, %%eax\n"
6073 "movzbl (%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
6074 "addl $64, %%eax\n\t"
6075 "cmpl %%eax, %[size]\n\t"
6076 "jne .Lfill_cache\n\t"
6078 :: [flush_pages] "r" (vmx_l1d_flush_pages),
6080 : "eax", "ebx", "ecx", "edx");
6083 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
6085 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6088 if (is_guest_mode(vcpu) &&
6089 nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
6092 tpr_threshold = (irr == -1 || tpr < irr) ? 0 : irr;
6093 if (is_guest_mode(vcpu))
6094 to_vmx(vcpu)->nested.l1_tpr_threshold = tpr_threshold;
6096 vmcs_write32(TPR_THRESHOLD, tpr_threshold);
6099 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu)
6101 struct vcpu_vmx *vmx = to_vmx(vcpu);
6102 u32 sec_exec_control;
6104 if (!lapic_in_kernel(vcpu))
6107 if (!flexpriority_enabled &&
6108 !cpu_has_vmx_virtualize_x2apic_mode())
6111 /* Postpone execution until vmcs01 is the current VMCS. */
6112 if (is_guest_mode(vcpu)) {
6113 vmx->nested.change_vmcs01_virtual_apic_mode = true;
6117 sec_exec_control = secondary_exec_controls_get(vmx);
6118 sec_exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6119 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
6121 switch (kvm_get_apic_mode(vcpu)) {
6122 case LAPIC_MODE_INVALID:
6123 WARN_ONCE(true, "Invalid local APIC state");
6124 case LAPIC_MODE_DISABLED:
6126 case LAPIC_MODE_XAPIC:
6127 if (flexpriority_enabled) {
6129 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6130 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
6133 * Flush the TLB, reloading the APIC access page will
6134 * only do so if its physical address has changed, but
6135 * the guest may have inserted a non-APIC mapping into
6136 * the TLB while the APIC access page was disabled.
6138 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
6141 case LAPIC_MODE_X2APIC:
6142 if (cpu_has_vmx_virtualize_x2apic_mode())
6144 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
6147 secondary_exec_controls_set(vmx, sec_exec_control);
6149 vmx_update_msr_bitmap(vcpu);
6152 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu)
6156 /* Defer reload until vmcs01 is the current VMCS. */
6157 if (is_guest_mode(vcpu)) {
6158 to_vmx(vcpu)->nested.reload_vmcs01_apic_access_page = true;
6162 if (!(secondary_exec_controls_get(to_vmx(vcpu)) &
6163 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
6166 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6167 if (is_error_page(page))
6170 vmcs_write64(APIC_ACCESS_ADDR, page_to_phys(page));
6171 vmx_flush_tlb_current(vcpu);
6174 * Do not pin apic access page in memory, the MMU notifier
6175 * will call us again if it is migrated or swapped out.
6180 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
6188 status = vmcs_read16(GUEST_INTR_STATUS);
6190 if (max_isr != old) {
6192 status |= max_isr << 8;
6193 vmcs_write16(GUEST_INTR_STATUS, status);
6197 static void vmx_set_rvi(int vector)
6205 status = vmcs_read16(GUEST_INTR_STATUS);
6206 old = (u8)status & 0xff;
6207 if ((u8)vector != old) {
6209 status |= (u8)vector;
6210 vmcs_write16(GUEST_INTR_STATUS, status);
6214 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
6217 * When running L2, updating RVI is only relevant when
6218 * vmcs12 virtual-interrupt-delivery enabled.
6219 * However, it can be enabled only when L1 also
6220 * intercepts external-interrupts and in that case
6221 * we should not update vmcs02 RVI but instead intercept
6222 * interrupt. Therefore, do nothing when running L2.
6224 if (!is_guest_mode(vcpu))
6225 vmx_set_rvi(max_irr);
6228 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
6230 struct vcpu_vmx *vmx = to_vmx(vcpu);
6232 bool max_irr_updated;
6234 WARN_ON(!vcpu->arch.apicv_active);
6235 if (pi_test_on(&vmx->pi_desc)) {
6236 pi_clear_on(&vmx->pi_desc);
6238 * IOMMU can write to PID.ON, so the barrier matters even on UP.
6239 * But on x86 this is just a compiler barrier anyway.
6241 smp_mb__after_atomic();
6243 kvm_apic_update_irr(vcpu, vmx->pi_desc.pir, &max_irr);
6246 * If we are running L2 and L1 has a new pending interrupt
6247 * which can be injected, we should re-evaluate
6248 * what should be done with this new L1 interrupt.
6249 * If L1 intercepts external-interrupts, we should
6250 * exit from L2 to L1. Otherwise, interrupt should be
6251 * delivered directly to L2.
6253 if (is_guest_mode(vcpu) && max_irr_updated) {
6254 if (nested_exit_on_intr(vcpu))
6255 kvm_vcpu_exiting_guest_mode(vcpu);
6257 kvm_make_request(KVM_REQ_EVENT, vcpu);
6260 max_irr = kvm_lapic_find_highest_irr(vcpu);
6262 vmx_hwapic_irr_update(vcpu, max_irr);
6266 static bool vmx_dy_apicv_has_pending_interrupt(struct kvm_vcpu *vcpu)
6268 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
6270 return pi_test_on(pi_desc) ||
6271 (pi_test_sn(pi_desc) && !pi_is_pir_empty(pi_desc));
6274 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
6276 if (!kvm_vcpu_apicv_active(vcpu))
6279 vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
6280 vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
6281 vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
6282 vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
6285 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
6287 struct vcpu_vmx *vmx = to_vmx(vcpu);
6289 pi_clear_on(&vmx->pi_desc);
6290 memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
6293 static void handle_exception_nmi_irqoff(struct vcpu_vmx *vmx)
6295 vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6297 /* if exit due to PF check for async PF */
6298 if (is_page_fault(vmx->exit_intr_info)) {
6299 vmx->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason();
6300 /* Handle machine checks before interrupts are enabled */
6301 } else if (is_machine_check(vmx->exit_intr_info)) {
6302 kvm_machine_check();
6303 /* We need to handle NMIs before interrupts are enabled */
6304 } else if (is_nmi(vmx->exit_intr_info)) {
6305 kvm_before_interrupt(&vmx->vcpu);
6307 kvm_after_interrupt(&vmx->vcpu);
6311 static void handle_external_interrupt_irqoff(struct kvm_vcpu *vcpu)
6313 unsigned int vector;
6314 unsigned long entry;
6315 #ifdef CONFIG_X86_64
6321 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6322 if (WARN_ONCE(!is_external_intr(intr_info),
6323 "KVM: unexpected VM-Exit interrupt info: 0x%x", intr_info))
6326 vector = intr_info & INTR_INFO_VECTOR_MASK;
6327 desc = (gate_desc *)host_idt_base + vector;
6328 entry = gate_offset(desc);
6330 kvm_before_interrupt(vcpu);
6333 #ifdef CONFIG_X86_64
6334 "mov %%" _ASM_SP ", %[sp]\n\t"
6335 "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
6340 __ASM_SIZE(push) " $%c[cs]\n\t"
6343 #ifdef CONFIG_X86_64
6348 [thunk_target]"r"(entry),
6349 [ss]"i"(__KERNEL_DS),
6350 [cs]"i"(__KERNEL_CS)
6353 kvm_after_interrupt(vcpu);
6355 STACK_FRAME_NON_STANDARD(handle_external_interrupt_irqoff);
6357 static void vmx_handle_exit_irqoff(struct kvm_vcpu *vcpu,
6358 enum exit_fastpath_completion *exit_fastpath)
6360 struct vcpu_vmx *vmx = to_vmx(vcpu);
6362 if (vmx->exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
6363 handle_external_interrupt_irqoff(vcpu);
6364 else if (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI)
6365 handle_exception_nmi_irqoff(vmx);
6366 else if (!is_guest_mode(vcpu) &&
6367 vmx->exit_reason == EXIT_REASON_MSR_WRITE)
6368 *exit_fastpath = handle_fastpath_set_msr_irqoff(vcpu);
6371 static bool vmx_has_emulated_msr(int index)
6374 case MSR_IA32_SMBASE:
6376 * We cannot do SMM unless we can run the guest in big
6379 return enable_unrestricted_guest || emulate_invalid_guest_state;
6380 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
6382 case MSR_AMD64_VIRT_SPEC_CTRL:
6383 /* This is AMD only. */
6390 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
6395 bool idtv_info_valid;
6397 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6400 if (vmx->loaded_vmcs->nmi_known_unmasked)
6403 * Can't use vmx->exit_intr_info since we're not sure what
6404 * the exit reason is.
6406 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6407 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
6408 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
6410 * SDM 3: 27.7.1.2 (September 2008)
6411 * Re-set bit "block by NMI" before VM entry if vmexit caused by
6412 * a guest IRET fault.
6413 * SDM 3: 23.2.2 (September 2008)
6414 * Bit 12 is undefined in any of the following cases:
6415 * If the VM exit sets the valid bit in the IDT-vectoring
6416 * information field.
6417 * If the VM exit is due to a double fault.
6419 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
6420 vector != DF_VECTOR && !idtv_info_valid)
6421 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
6422 GUEST_INTR_STATE_NMI);
6424 vmx->loaded_vmcs->nmi_known_unmasked =
6425 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
6426 & GUEST_INTR_STATE_NMI);
6427 } else if (unlikely(vmx->loaded_vmcs->soft_vnmi_blocked))
6428 vmx->loaded_vmcs->vnmi_blocked_time +=
6429 ktime_to_ns(ktime_sub(ktime_get(),
6430 vmx->loaded_vmcs->entry_time));
6433 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
6434 u32 idt_vectoring_info,
6435 int instr_len_field,
6436 int error_code_field)
6440 bool idtv_info_valid;
6442 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6444 vcpu->arch.nmi_injected = false;
6445 kvm_clear_exception_queue(vcpu);
6446 kvm_clear_interrupt_queue(vcpu);
6448 if (!idtv_info_valid)
6451 kvm_make_request(KVM_REQ_EVENT, vcpu);
6453 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
6454 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
6457 case INTR_TYPE_NMI_INTR:
6458 vcpu->arch.nmi_injected = true;
6460 * SDM 3: 27.7.1.2 (September 2008)
6461 * Clear bit "block by NMI" before VM entry if a NMI
6464 vmx_set_nmi_mask(vcpu, false);
6466 case INTR_TYPE_SOFT_EXCEPTION:
6467 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6469 case INTR_TYPE_HARD_EXCEPTION:
6470 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
6471 u32 err = vmcs_read32(error_code_field);
6472 kvm_requeue_exception_e(vcpu, vector, err);
6474 kvm_requeue_exception(vcpu, vector);
6476 case INTR_TYPE_SOFT_INTR:
6477 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6479 case INTR_TYPE_EXT_INTR:
6480 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
6487 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
6489 __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
6490 VM_EXIT_INSTRUCTION_LEN,
6491 IDT_VECTORING_ERROR_CODE);
6494 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
6496 __vmx_complete_interrupts(vcpu,
6497 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
6498 VM_ENTRY_INSTRUCTION_LEN,
6499 VM_ENTRY_EXCEPTION_ERROR_CODE);
6501 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
6504 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
6507 struct perf_guest_switch_msr *msrs;
6509 msrs = perf_guest_get_msrs(&nr_msrs);
6514 for (i = 0; i < nr_msrs; i++)
6515 if (msrs[i].host == msrs[i].guest)
6516 clear_atomic_switch_msr(vmx, msrs[i].msr);
6518 add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
6519 msrs[i].host, false);
6522 static void atomic_switch_umwait_control_msr(struct vcpu_vmx *vmx)
6524 u32 host_umwait_control;
6526 if (!vmx_has_waitpkg(vmx))
6529 host_umwait_control = get_umwait_control_msr();
6531 if (vmx->msr_ia32_umwait_control != host_umwait_control)
6532 add_atomic_switch_msr(vmx, MSR_IA32_UMWAIT_CONTROL,
6533 vmx->msr_ia32_umwait_control,
6534 host_umwait_control, false);
6536 clear_atomic_switch_msr(vmx, MSR_IA32_UMWAIT_CONTROL);
6539 static void vmx_update_hv_timer(struct kvm_vcpu *vcpu)
6541 struct vcpu_vmx *vmx = to_vmx(vcpu);
6545 if (vmx->req_immediate_exit) {
6546 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, 0);
6547 vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6548 } else if (vmx->hv_deadline_tsc != -1) {
6550 if (vmx->hv_deadline_tsc > tscl)
6551 /* set_hv_timer ensures the delta fits in 32-bits */
6552 delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
6553 cpu_preemption_timer_multi);
6557 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
6558 vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6559 } else if (!vmx->loaded_vmcs->hv_timer_soft_disabled) {
6560 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, -1);
6561 vmx->loaded_vmcs->hv_timer_soft_disabled = true;
6565 void vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp)
6567 if (unlikely(host_rsp != vmx->loaded_vmcs->host_state.rsp)) {
6568 vmx->loaded_vmcs->host_state.rsp = host_rsp;
6569 vmcs_writel(HOST_RSP, host_rsp);
6573 bool __vmx_vcpu_run(struct vcpu_vmx *vmx, unsigned long *regs, bool launched);
6575 static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
6577 struct vcpu_vmx *vmx = to_vmx(vcpu);
6578 unsigned long cr3, cr4;
6580 /* Record the guest's net vcpu time for enforced NMI injections. */
6581 if (unlikely(!enable_vnmi &&
6582 vmx->loaded_vmcs->soft_vnmi_blocked))
6583 vmx->loaded_vmcs->entry_time = ktime_get();
6585 /* Don't enter VMX if guest state is invalid, let the exit handler
6586 start emulation until we arrive back to a valid state */
6587 if (vmx->emulation_required)
6590 if (vmx->ple_window_dirty) {
6591 vmx->ple_window_dirty = false;
6592 vmcs_write32(PLE_WINDOW, vmx->ple_window);
6596 * We did this in prepare_switch_to_guest, because it needs to
6597 * be within srcu_read_lock.
6599 WARN_ON_ONCE(vmx->nested.need_vmcs12_to_shadow_sync);
6601 if (kvm_register_is_dirty(vcpu, VCPU_REGS_RSP))
6602 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
6603 if (kvm_register_is_dirty(vcpu, VCPU_REGS_RIP))
6604 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
6606 cr3 = __get_current_cr3_fast();
6607 if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
6608 vmcs_writel(HOST_CR3, cr3);
6609 vmx->loaded_vmcs->host_state.cr3 = cr3;
6612 cr4 = cr4_read_shadow();
6613 if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
6614 vmcs_writel(HOST_CR4, cr4);
6615 vmx->loaded_vmcs->host_state.cr4 = cr4;
6618 /* When single-stepping over STI and MOV SS, we must clear the
6619 * corresponding interruptibility bits in the guest state. Otherwise
6620 * vmentry fails as it then expects bit 14 (BS) in pending debug
6621 * exceptions being set, but that's not correct for the guest debugging
6623 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6624 vmx_set_interrupt_shadow(vcpu, 0);
6626 kvm_load_guest_xsave_state(vcpu);
6628 if (static_cpu_has(X86_FEATURE_PKU) &&
6629 kvm_read_cr4_bits(vcpu, X86_CR4_PKE) &&
6630 vcpu->arch.pkru != vmx->host_pkru)
6631 __write_pkru(vcpu->arch.pkru);
6633 pt_guest_enter(vmx);
6635 if (vcpu_to_pmu(vcpu)->version)
6636 atomic_switch_perf_msrs(vmx);
6637 atomic_switch_umwait_control_msr(vmx);
6639 if (enable_preemption_timer)
6640 vmx_update_hv_timer(vcpu);
6642 if (lapic_in_kernel(vcpu) &&
6643 vcpu->arch.apic->lapic_timer.timer_advance_ns)
6644 kvm_wait_lapic_expire(vcpu);
6647 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
6648 * it's non-zero. Since vmentry is serialising on affected CPUs, there
6649 * is no need to worry about the conditional branch over the wrmsr
6650 * being speculatively taken.
6652 x86_spec_ctrl_set_guest(vmx->spec_ctrl, 0);
6654 /* L1D Flush includes CPU buffer clear to mitigate MDS */
6655 if (static_branch_unlikely(&vmx_l1d_should_flush))
6656 vmx_l1d_flush(vcpu);
6657 else if (static_branch_unlikely(&mds_user_clear))
6658 mds_clear_cpu_buffers();
6660 if (vcpu->arch.cr2 != read_cr2())
6661 write_cr2(vcpu->arch.cr2);
6663 vmx->fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
6664 vmx->loaded_vmcs->launched);
6666 vcpu->arch.cr2 = read_cr2();
6669 * We do not use IBRS in the kernel. If this vCPU has used the
6670 * SPEC_CTRL MSR it may have left it on; save the value and
6671 * turn it off. This is much more efficient than blindly adding
6672 * it to the atomic save/restore list. Especially as the former
6673 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
6675 * For non-nested case:
6676 * If the L01 MSR bitmap does not intercept the MSR, then we need to
6680 * If the L02 MSR bitmap does not intercept the MSR, then we need to
6683 if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
6684 vmx->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
6686 x86_spec_ctrl_restore_host(vmx->spec_ctrl, 0);
6688 /* All fields are clean at this point */
6689 if (static_branch_unlikely(&enable_evmcs))
6690 current_evmcs->hv_clean_fields |=
6691 HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
6693 if (static_branch_unlikely(&enable_evmcs))
6694 current_evmcs->hv_vp_id = vcpu->arch.hyperv.vp_index;
6696 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
6697 if (vmx->host_debugctlmsr)
6698 update_debugctlmsr(vmx->host_debugctlmsr);
6700 #ifndef CONFIG_X86_64
6702 * The sysexit path does not restore ds/es, so we must set them to
6703 * a reasonable value ourselves.
6705 * We can't defer this to vmx_prepare_switch_to_host() since that
6706 * function may be executed in interrupt context, which saves and
6707 * restore segments around it, nullifying its effect.
6709 loadsegment(ds, __USER_DS);
6710 loadsegment(es, __USER_DS);
6713 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
6714 | (1 << VCPU_EXREG_RFLAGS)
6715 | (1 << VCPU_EXREG_PDPTR)
6716 | (1 << VCPU_EXREG_SEGMENTS)
6717 | (1 << VCPU_EXREG_CR3));
6718 vcpu->arch.regs_dirty = 0;
6723 * eager fpu is enabled if PKEY is supported and CR4 is switched
6724 * back on host, so it is safe to read guest PKRU from current
6727 if (static_cpu_has(X86_FEATURE_PKU) &&
6728 kvm_read_cr4_bits(vcpu, X86_CR4_PKE)) {
6729 vcpu->arch.pkru = rdpkru();
6730 if (vcpu->arch.pkru != vmx->host_pkru)
6731 __write_pkru(vmx->host_pkru);
6734 kvm_load_host_xsave_state(vcpu);
6736 vmx->nested.nested_run_pending = 0;
6737 vmx->idt_vectoring_info = 0;
6739 vmx->exit_reason = vmx->fail ? 0xdead : vmcs_read32(VM_EXIT_REASON);
6740 if ((u16)vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY)
6741 kvm_machine_check();
6743 if (vmx->fail || (vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
6746 vmx->loaded_vmcs->launched = 1;
6747 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
6749 vmx_recover_nmi_blocking(vmx);
6750 vmx_complete_interrupts(vmx);
6753 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
6755 struct vcpu_vmx *vmx = to_vmx(vcpu);
6758 vmx_destroy_pml_buffer(vmx);
6759 free_vpid(vmx->vpid);
6760 nested_vmx_free_vcpu(vcpu);
6761 free_loaded_vmcs(vmx->loaded_vmcs);
6764 static int vmx_create_vcpu(struct kvm_vcpu *vcpu)
6766 struct vcpu_vmx *vmx;
6767 unsigned long *msr_bitmap;
6770 BUILD_BUG_ON(offsetof(struct vcpu_vmx, vcpu) != 0);
6775 vmx->vpid = allocate_vpid();
6778 * If PML is turned on, failure on enabling PML just results in failure
6779 * of creating the vcpu, therefore we can simplify PML logic (by
6780 * avoiding dealing with cases, such as enabling PML partially on vcpus
6781 * for the guest), etc.
6784 vmx->pml_pg = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
6789 BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) != NR_SHARED_MSRS);
6791 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
6792 u32 index = vmx_msr_index[i];
6793 u32 data_low, data_high;
6796 if (rdmsr_safe(index, &data_low, &data_high) < 0)
6798 if (wrmsr_safe(index, data_low, data_high) < 0)
6801 vmx->guest_msrs[j].index = i;
6802 vmx->guest_msrs[j].data = 0;
6804 case MSR_IA32_TSX_CTRL:
6806 * No need to pass TSX_CTRL_CPUID_CLEAR through, so
6807 * let's avoid changing CPUID bits under the host
6810 vmx->guest_msrs[j].mask = ~(u64)TSX_CTRL_CPUID_CLEAR;
6813 vmx->guest_msrs[j].mask = -1ull;
6819 err = alloc_loaded_vmcs(&vmx->vmcs01);
6823 msr_bitmap = vmx->vmcs01.msr_bitmap;
6824 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_TSC, MSR_TYPE_R);
6825 vmx_disable_intercept_for_msr(msr_bitmap, MSR_FS_BASE, MSR_TYPE_RW);
6826 vmx_disable_intercept_for_msr(msr_bitmap, MSR_GS_BASE, MSR_TYPE_RW);
6827 vmx_disable_intercept_for_msr(msr_bitmap, MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
6828 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_CS, MSR_TYPE_RW);
6829 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_ESP, MSR_TYPE_RW);
6830 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_EIP, MSR_TYPE_RW);
6831 if (kvm_cstate_in_guest(vcpu->kvm)) {
6832 vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C1_RES, MSR_TYPE_R);
6833 vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C3_RESIDENCY, MSR_TYPE_R);
6834 vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C6_RESIDENCY, MSR_TYPE_R);
6835 vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C7_RESIDENCY, MSR_TYPE_R);
6837 vmx->msr_bitmap_mode = 0;
6839 vmx->loaded_vmcs = &vmx->vmcs01;
6841 vmx_vcpu_load(vcpu, cpu);
6846 if (cpu_need_virtualize_apic_accesses(vcpu)) {
6847 err = alloc_apic_access_page(vcpu->kvm);
6852 if (enable_ept && !enable_unrestricted_guest) {
6853 err = init_rmode_identity_map(vcpu->kvm);
6859 nested_vmx_setup_ctls_msrs(&vmx->nested.msrs,
6860 vmx_capability.ept);
6862 memset(&vmx->nested.msrs, 0, sizeof(vmx->nested.msrs));
6864 vmx->nested.posted_intr_nv = -1;
6865 vmx->nested.current_vmptr = -1ull;
6867 vcpu->arch.microcode_version = 0x100000000ULL;
6868 vmx->msr_ia32_feature_control_valid_bits = FEAT_CTL_LOCKED;
6871 * Enforce invariant: pi_desc.nv is always either POSTED_INTR_VECTOR
6872 * or POSTED_INTR_WAKEUP_VECTOR.
6874 vmx->pi_desc.nv = POSTED_INTR_VECTOR;
6875 vmx->pi_desc.sn = 1;
6877 vmx->ept_pointer = INVALID_PAGE;
6882 free_loaded_vmcs(vmx->loaded_vmcs);
6884 vmx_destroy_pml_buffer(vmx);
6886 free_vpid(vmx->vpid);
6890 #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"
6891 #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"
6893 static int vmx_vm_init(struct kvm *kvm)
6895 spin_lock_init(&to_kvm_vmx(kvm)->ept_pointer_lock);
6898 kvm->arch.pause_in_guest = true;
6900 if (boot_cpu_has(X86_BUG_L1TF) && enable_ept) {
6901 switch (l1tf_mitigation) {
6902 case L1TF_MITIGATION_OFF:
6903 case L1TF_MITIGATION_FLUSH_NOWARN:
6904 /* 'I explicitly don't care' is set */
6906 case L1TF_MITIGATION_FLUSH:
6907 case L1TF_MITIGATION_FLUSH_NOSMT:
6908 case L1TF_MITIGATION_FULL:
6910 * Warn upon starting the first VM in a potentially
6911 * insecure environment.
6913 if (sched_smt_active())
6914 pr_warn_once(L1TF_MSG_SMT);
6915 if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER)
6916 pr_warn_once(L1TF_MSG_L1D);
6918 case L1TF_MITIGATION_FULL_FORCE:
6919 /* Flush is enforced */
6923 kvm_apicv_init(kvm, enable_apicv);
6927 static int __init vmx_check_processor_compat(void)
6929 struct vmcs_config vmcs_conf;
6930 struct vmx_capability vmx_cap;
6932 if (!this_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
6933 !this_cpu_has(X86_FEATURE_VMX)) {
6934 pr_err("kvm: VMX is disabled on CPU %d\n", smp_processor_id());
6938 if (setup_vmcs_config(&vmcs_conf, &vmx_cap) < 0)
6941 nested_vmx_setup_ctls_msrs(&vmcs_conf.nested, vmx_cap.ept);
6942 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
6943 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
6944 smp_processor_id());
6950 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
6955 /* We wanted to honor guest CD/MTRR/PAT, but doing so could result in
6956 * memory aliases with conflicting memory types and sometimes MCEs.
6957 * We have to be careful as to what are honored and when.
6959 * For MMIO, guest CD/MTRR are ignored. The EPT memory type is set to
6960 * UC. The effective memory type is UC or WC depending on guest PAT.
6961 * This was historically the source of MCEs and we want to be
6964 * When there is no need to deal with noncoherent DMA (e.g., no VT-d
6965 * or VT-d has snoop control), guest CD/MTRR/PAT are all ignored. The
6966 * EPT memory type is set to WB. The effective memory type is forced
6969 * Otherwise, we trust guest. Guest CD/MTRR/PAT are all honored. The
6970 * EPT memory type is used to emulate guest CD/MTRR.
6974 cache = MTRR_TYPE_UNCACHABLE;
6978 if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
6979 ipat = VMX_EPT_IPAT_BIT;
6980 cache = MTRR_TYPE_WRBACK;
6984 if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
6985 ipat = VMX_EPT_IPAT_BIT;
6986 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
6987 cache = MTRR_TYPE_WRBACK;
6989 cache = MTRR_TYPE_UNCACHABLE;
6993 cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
6996 return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
6999 static void vmcs_set_secondary_exec_control(struct vcpu_vmx *vmx)
7002 * These bits in the secondary execution controls field
7003 * are dynamic, the others are mostly based on the hypervisor
7004 * architecture and the guest's CPUID. Do not touch the
7008 SECONDARY_EXEC_SHADOW_VMCS |
7009 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
7010 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
7011 SECONDARY_EXEC_DESC;
7013 u32 new_ctl = vmx->secondary_exec_control;
7014 u32 cur_ctl = secondary_exec_controls_get(vmx);
7016 secondary_exec_controls_set(vmx, (new_ctl & ~mask) | (cur_ctl & mask));
7020 * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
7021 * (indicating "allowed-1") if they are supported in the guest's CPUID.
7023 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
7025 struct vcpu_vmx *vmx = to_vmx(vcpu);
7026 struct kvm_cpuid_entry2 *entry;
7028 vmx->nested.msrs.cr0_fixed1 = 0xffffffff;
7029 vmx->nested.msrs.cr4_fixed1 = X86_CR4_PCE;
7031 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do { \
7032 if (entry && (entry->_reg & (_cpuid_mask))) \
7033 vmx->nested.msrs.cr4_fixed1 |= (_cr4_mask); \
7036 entry = kvm_find_cpuid_entry(vcpu, 0x1, 0);
7037 cr4_fixed1_update(X86_CR4_VME, edx, feature_bit(VME));
7038 cr4_fixed1_update(X86_CR4_PVI, edx, feature_bit(VME));
7039 cr4_fixed1_update(X86_CR4_TSD, edx, feature_bit(TSC));
7040 cr4_fixed1_update(X86_CR4_DE, edx, feature_bit(DE));
7041 cr4_fixed1_update(X86_CR4_PSE, edx, feature_bit(PSE));
7042 cr4_fixed1_update(X86_CR4_PAE, edx, feature_bit(PAE));
7043 cr4_fixed1_update(X86_CR4_MCE, edx, feature_bit(MCE));
7044 cr4_fixed1_update(X86_CR4_PGE, edx, feature_bit(PGE));
7045 cr4_fixed1_update(X86_CR4_OSFXSR, edx, feature_bit(FXSR));
7046 cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, feature_bit(XMM));
7047 cr4_fixed1_update(X86_CR4_VMXE, ecx, feature_bit(VMX));
7048 cr4_fixed1_update(X86_CR4_SMXE, ecx, feature_bit(SMX));
7049 cr4_fixed1_update(X86_CR4_PCIDE, ecx, feature_bit(PCID));
7050 cr4_fixed1_update(X86_CR4_OSXSAVE, ecx, feature_bit(XSAVE));
7052 entry = kvm_find_cpuid_entry(vcpu, 0x7, 0);
7053 cr4_fixed1_update(X86_CR4_FSGSBASE, ebx, feature_bit(FSGSBASE));
7054 cr4_fixed1_update(X86_CR4_SMEP, ebx, feature_bit(SMEP));
7055 cr4_fixed1_update(X86_CR4_SMAP, ebx, feature_bit(SMAP));
7056 cr4_fixed1_update(X86_CR4_PKE, ecx, feature_bit(PKU));
7057 cr4_fixed1_update(X86_CR4_UMIP, ecx, feature_bit(UMIP));
7058 cr4_fixed1_update(X86_CR4_LA57, ecx, feature_bit(LA57));
7060 #undef cr4_fixed1_update
7063 static void nested_vmx_entry_exit_ctls_update(struct kvm_vcpu *vcpu)
7065 struct vcpu_vmx *vmx = to_vmx(vcpu);
7067 if (kvm_mpx_supported()) {
7068 bool mpx_enabled = guest_cpuid_has(vcpu, X86_FEATURE_MPX);
7071 vmx->nested.msrs.entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
7072 vmx->nested.msrs.exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
7074 vmx->nested.msrs.entry_ctls_high &= ~VM_ENTRY_LOAD_BNDCFGS;
7075 vmx->nested.msrs.exit_ctls_high &= ~VM_EXIT_CLEAR_BNDCFGS;
7080 static void update_intel_pt_cfg(struct kvm_vcpu *vcpu)
7082 struct vcpu_vmx *vmx = to_vmx(vcpu);
7083 struct kvm_cpuid_entry2 *best = NULL;
7086 for (i = 0; i < PT_CPUID_LEAVES; i++) {
7087 best = kvm_find_cpuid_entry(vcpu, 0x14, i);
7090 vmx->pt_desc.caps[CPUID_EAX + i*PT_CPUID_REGS_NUM] = best->eax;
7091 vmx->pt_desc.caps[CPUID_EBX + i*PT_CPUID_REGS_NUM] = best->ebx;
7092 vmx->pt_desc.caps[CPUID_ECX + i*PT_CPUID_REGS_NUM] = best->ecx;
7093 vmx->pt_desc.caps[CPUID_EDX + i*PT_CPUID_REGS_NUM] = best->edx;
7096 /* Get the number of configurable Address Ranges for filtering */
7097 vmx->pt_desc.addr_range = intel_pt_validate_cap(vmx->pt_desc.caps,
7098 PT_CAP_num_address_ranges);
7100 /* Initialize and clear the no dependency bits */
7101 vmx->pt_desc.ctl_bitmask = ~(RTIT_CTL_TRACEEN | RTIT_CTL_OS |
7102 RTIT_CTL_USR | RTIT_CTL_TSC_EN | RTIT_CTL_DISRETC);
7105 * If CPUID.(EAX=14H,ECX=0):EBX[0]=1 CR3Filter can be set otherwise
7106 * will inject an #GP
7108 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_cr3_filtering))
7109 vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_CR3EN;
7112 * If CPUID.(EAX=14H,ECX=0):EBX[1]=1 CYCEn, CycThresh and
7113 * PSBFreq can be set
7115 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc))
7116 vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_CYCLEACC |
7117 RTIT_CTL_CYC_THRESH | RTIT_CTL_PSB_FREQ);
7120 * If CPUID.(EAX=14H,ECX=0):EBX[3]=1 MTCEn BranchEn and
7121 * MTCFreq can be set
7123 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc))
7124 vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_MTC_EN |
7125 RTIT_CTL_BRANCH_EN | RTIT_CTL_MTC_RANGE);
7127 /* If CPUID.(EAX=14H,ECX=0):EBX[4]=1 FUPonPTW and PTWEn can be set */
7128 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_ptwrite))
7129 vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_FUP_ON_PTW |
7132 /* If CPUID.(EAX=14H,ECX=0):EBX[5]=1 PwrEvEn can be set */
7133 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_power_event_trace))
7134 vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_PWR_EVT_EN;
7136 /* If CPUID.(EAX=14H,ECX=0):ECX[0]=1 ToPA can be set */
7137 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_topa_output))
7138 vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_TOPA;
7140 /* If CPUID.(EAX=14H,ECX=0):ECX[3]=1 FabircEn can be set */
7141 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_output_subsys))
7142 vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_FABRIC_EN;
7144 /* unmask address range configure area */
7145 for (i = 0; i < vmx->pt_desc.addr_range; i++)
7146 vmx->pt_desc.ctl_bitmask &= ~(0xfULL << (32 + i * 4));
7149 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
7151 struct vcpu_vmx *vmx = to_vmx(vcpu);
7153 /* xsaves_enabled is recomputed in vmx_compute_secondary_exec_control(). */
7154 vcpu->arch.xsaves_enabled = false;
7156 if (cpu_has_secondary_exec_ctrls()) {
7157 vmx_compute_secondary_exec_control(vmx);
7158 vmcs_set_secondary_exec_control(vmx);
7161 if (nested_vmx_allowed(vcpu))
7162 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7163 FEAT_CTL_VMX_ENABLED_INSIDE_SMX |
7164 FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX;
7166 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7167 ~(FEAT_CTL_VMX_ENABLED_INSIDE_SMX |
7168 FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX);
7170 if (nested_vmx_allowed(vcpu)) {
7171 nested_vmx_cr_fixed1_bits_update(vcpu);
7172 nested_vmx_entry_exit_ctls_update(vcpu);
7175 if (boot_cpu_has(X86_FEATURE_INTEL_PT) &&
7176 guest_cpuid_has(vcpu, X86_FEATURE_INTEL_PT))
7177 update_intel_pt_cfg(vcpu);
7179 if (boot_cpu_has(X86_FEATURE_RTM)) {
7180 struct shared_msr_entry *msr;
7181 msr = find_msr_entry(vmx, MSR_IA32_TSX_CTRL);
7183 bool enabled = guest_cpuid_has(vcpu, X86_FEATURE_RTM);
7184 vmx_set_guest_msr(vmx, msr, enabled ? 0 : TSX_CTRL_RTM_DISABLE);
7189 static __init void vmx_set_cpu_caps(void)
7195 kvm_cpu_cap_set(X86_FEATURE_VMX);
7198 if (kvm_mpx_supported())
7199 kvm_cpu_cap_check_and_set(X86_FEATURE_MPX);
7200 if (cpu_has_vmx_invpcid())
7201 kvm_cpu_cap_check_and_set(X86_FEATURE_INVPCID);
7202 if (vmx_pt_mode_is_host_guest())
7203 kvm_cpu_cap_check_and_set(X86_FEATURE_INTEL_PT);
7205 /* PKU is not yet implemented for shadow paging. */
7206 if (enable_ept && boot_cpu_has(X86_FEATURE_OSPKE))
7207 kvm_cpu_cap_check_and_set(X86_FEATURE_PKU);
7209 if (vmx_umip_emulated())
7210 kvm_cpu_cap_set(X86_FEATURE_UMIP);
7214 if (!vmx_xsaves_supported())
7215 kvm_cpu_cap_clear(X86_FEATURE_XSAVES);
7217 /* CPUID 0x80000001 */
7218 if (!cpu_has_vmx_rdtscp())
7219 kvm_cpu_cap_clear(X86_FEATURE_RDTSCP);
7222 static void vmx_request_immediate_exit(struct kvm_vcpu *vcpu)
7224 to_vmx(vcpu)->req_immediate_exit = true;
7227 static int vmx_check_intercept_io(struct kvm_vcpu *vcpu,
7228 struct x86_instruction_info *info)
7230 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7231 unsigned short port;
7235 if (info->intercept == x86_intercept_in ||
7236 info->intercept == x86_intercept_ins) {
7237 port = info->src_val;
7238 size = info->dst_bytes;
7240 port = info->dst_val;
7241 size = info->src_bytes;
7245 * If the 'use IO bitmaps' VM-execution control is 0, IO instruction
7246 * VM-exits depend on the 'unconditional IO exiting' VM-execution
7249 * Otherwise, IO instruction VM-exits are controlled by the IO bitmaps.
7251 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
7252 intercept = nested_cpu_has(vmcs12,
7253 CPU_BASED_UNCOND_IO_EXITING);
7255 intercept = nested_vmx_check_io_bitmaps(vcpu, port, size);
7257 /* FIXME: produce nested vmexit and return X86EMUL_INTERCEPTED. */
7258 return intercept ? X86EMUL_UNHANDLEABLE : X86EMUL_CONTINUE;
7261 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
7262 struct x86_instruction_info *info,
7263 enum x86_intercept_stage stage,
7264 struct x86_exception *exception)
7266 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7268 switch (info->intercept) {
7270 * RDPID causes #UD if disabled through secondary execution controls.
7271 * Because it is marked as EmulateOnUD, we need to intercept it here.
7273 case x86_intercept_rdtscp:
7274 if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDTSCP)) {
7275 exception->vector = UD_VECTOR;
7276 exception->error_code_valid = false;
7277 return X86EMUL_PROPAGATE_FAULT;
7281 case x86_intercept_in:
7282 case x86_intercept_ins:
7283 case x86_intercept_out:
7284 case x86_intercept_outs:
7285 return vmx_check_intercept_io(vcpu, info);
7287 case x86_intercept_lgdt:
7288 case x86_intercept_lidt:
7289 case x86_intercept_lldt:
7290 case x86_intercept_ltr:
7291 case x86_intercept_sgdt:
7292 case x86_intercept_sidt:
7293 case x86_intercept_sldt:
7294 case x86_intercept_str:
7295 if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC))
7296 return X86EMUL_CONTINUE;
7298 /* FIXME: produce nested vmexit and return X86EMUL_INTERCEPTED. */
7301 /* TODO: check more intercepts... */
7306 return X86EMUL_UNHANDLEABLE;
7309 #ifdef CONFIG_X86_64
7310 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
7311 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
7312 u64 divisor, u64 *result)
7314 u64 low = a << shift, high = a >> (64 - shift);
7316 /* To avoid the overflow on divq */
7317 if (high >= divisor)
7320 /* Low hold the result, high hold rem which is discarded */
7321 asm("divq %2\n\t" : "=a" (low), "=d" (high) :
7322 "rm" (divisor), "0" (low), "1" (high));
7328 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc,
7331 struct vcpu_vmx *vmx;
7332 u64 tscl, guest_tscl, delta_tsc, lapic_timer_advance_cycles;
7333 struct kvm_timer *ktimer = &vcpu->arch.apic->lapic_timer;
7335 if (kvm_mwait_in_guest(vcpu->kvm) ||
7336 kvm_can_post_timer_interrupt(vcpu))
7341 guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
7342 delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
7343 lapic_timer_advance_cycles = nsec_to_cycles(vcpu,
7344 ktimer->timer_advance_ns);
7346 if (delta_tsc > lapic_timer_advance_cycles)
7347 delta_tsc -= lapic_timer_advance_cycles;
7351 /* Convert to host delta tsc if tsc scaling is enabled */
7352 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
7353 delta_tsc && u64_shl_div_u64(delta_tsc,
7354 kvm_tsc_scaling_ratio_frac_bits,
7355 vcpu->arch.tsc_scaling_ratio, &delta_tsc))
7359 * If the delta tsc can't fit in the 32 bit after the multi shift,
7360 * we can't use the preemption timer.
7361 * It's possible that it fits on later vmentries, but checking
7362 * on every vmentry is costly so we just use an hrtimer.
7364 if (delta_tsc >> (cpu_preemption_timer_multi + 32))
7367 vmx->hv_deadline_tsc = tscl + delta_tsc;
7368 *expired = !delta_tsc;
7372 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
7374 to_vmx(vcpu)->hv_deadline_tsc = -1;
7378 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
7380 if (!kvm_pause_in_guest(vcpu->kvm))
7381 shrink_ple_window(vcpu);
7384 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
7385 struct kvm_memory_slot *slot)
7387 if (!kvm_dirty_log_manual_protect_and_init_set(kvm))
7388 kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
7389 kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
7392 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
7393 struct kvm_memory_slot *slot)
7395 kvm_mmu_slot_set_dirty(kvm, slot);
7398 static void vmx_flush_log_dirty(struct kvm *kvm)
7400 kvm_flush_pml_buffers(kvm);
7403 static int vmx_write_pml_buffer(struct kvm_vcpu *vcpu)
7405 struct vmcs12 *vmcs12;
7406 struct vcpu_vmx *vmx = to_vmx(vcpu);
7409 if (is_guest_mode(vcpu)) {
7410 WARN_ON_ONCE(vmx->nested.pml_full);
7413 * Check if PML is enabled for the nested guest.
7414 * Whether eptp bit 6 is set is already checked
7415 * as part of A/D emulation.
7417 vmcs12 = get_vmcs12(vcpu);
7418 if (!nested_cpu_has_pml(vmcs12))
7421 if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
7422 vmx->nested.pml_full = true;
7426 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS) & ~0xFFFull;
7427 dst = vmcs12->pml_address + sizeof(u64) * vmcs12->guest_pml_index;
7429 if (kvm_write_guest_page(vcpu->kvm, gpa_to_gfn(dst), &gpa,
7430 offset_in_page(dst), sizeof(gpa)))
7433 vmcs12->guest_pml_index--;
7439 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
7440 struct kvm_memory_slot *memslot,
7441 gfn_t offset, unsigned long mask)
7443 kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
7446 static void __pi_post_block(struct kvm_vcpu *vcpu)
7448 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7449 struct pi_desc old, new;
7453 old.control = new.control = pi_desc->control;
7454 WARN(old.nv != POSTED_INTR_WAKEUP_VECTOR,
7455 "Wakeup handler not enabled while the VCPU is blocked\n");
7457 dest = cpu_physical_id(vcpu->cpu);
7459 if (x2apic_enabled())
7462 new.ndst = (dest << 8) & 0xFF00;
7464 /* set 'NV' to 'notification vector' */
7465 new.nv = POSTED_INTR_VECTOR;
7466 } while (cmpxchg64(&pi_desc->control, old.control,
7467 new.control) != old.control);
7469 if (!WARN_ON_ONCE(vcpu->pre_pcpu == -1)) {
7470 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7471 list_del(&vcpu->blocked_vcpu_list);
7472 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7473 vcpu->pre_pcpu = -1;
7478 * This routine does the following things for vCPU which is going
7479 * to be blocked if VT-d PI is enabled.
7480 * - Store the vCPU to the wakeup list, so when interrupts happen
7481 * we can find the right vCPU to wake up.
7482 * - Change the Posted-interrupt descriptor as below:
7483 * 'NDST' <-- vcpu->pre_pcpu
7484 * 'NV' <-- POSTED_INTR_WAKEUP_VECTOR
7485 * - If 'ON' is set during this process, which means at least one
7486 * interrupt is posted for this vCPU, we cannot block it, in
7487 * this case, return 1, otherwise, return 0.
7490 static int pi_pre_block(struct kvm_vcpu *vcpu)
7493 struct pi_desc old, new;
7494 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7496 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
7497 !irq_remapping_cap(IRQ_POSTING_CAP) ||
7498 !kvm_vcpu_apicv_active(vcpu))
7501 WARN_ON(irqs_disabled());
7502 local_irq_disable();
7503 if (!WARN_ON_ONCE(vcpu->pre_pcpu != -1)) {
7504 vcpu->pre_pcpu = vcpu->cpu;
7505 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7506 list_add_tail(&vcpu->blocked_vcpu_list,
7507 &per_cpu(blocked_vcpu_on_cpu,
7509 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7513 old.control = new.control = pi_desc->control;
7515 WARN((pi_desc->sn == 1),
7516 "Warning: SN field of posted-interrupts "
7517 "is set before blocking\n");
7520 * Since vCPU can be preempted during this process,
7521 * vcpu->cpu could be different with pre_pcpu, we
7522 * need to set pre_pcpu as the destination of wakeup
7523 * notification event, then we can find the right vCPU
7524 * to wakeup in wakeup handler if interrupts happen
7525 * when the vCPU is in blocked state.
7527 dest = cpu_physical_id(vcpu->pre_pcpu);
7529 if (x2apic_enabled())
7532 new.ndst = (dest << 8) & 0xFF00;
7534 /* set 'NV' to 'wakeup vector' */
7535 new.nv = POSTED_INTR_WAKEUP_VECTOR;
7536 } while (cmpxchg64(&pi_desc->control, old.control,
7537 new.control) != old.control);
7539 /* We should not block the vCPU if an interrupt is posted for it. */
7540 if (pi_test_on(pi_desc) == 1)
7541 __pi_post_block(vcpu);
7544 return (vcpu->pre_pcpu == -1);
7547 static int vmx_pre_block(struct kvm_vcpu *vcpu)
7549 if (pi_pre_block(vcpu))
7552 if (kvm_lapic_hv_timer_in_use(vcpu))
7553 kvm_lapic_switch_to_sw_timer(vcpu);
7558 static void pi_post_block(struct kvm_vcpu *vcpu)
7560 if (vcpu->pre_pcpu == -1)
7563 WARN_ON(irqs_disabled());
7564 local_irq_disable();
7565 __pi_post_block(vcpu);
7569 static void vmx_post_block(struct kvm_vcpu *vcpu)
7571 if (kvm_x86_ops.set_hv_timer)
7572 kvm_lapic_switch_to_hv_timer(vcpu);
7574 pi_post_block(vcpu);
7578 * vmx_update_pi_irte - set IRTE for Posted-Interrupts
7581 * @host_irq: host irq of the interrupt
7582 * @guest_irq: gsi of the interrupt
7583 * @set: set or unset PI
7584 * returns 0 on success, < 0 on failure
7586 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
7587 uint32_t guest_irq, bool set)
7589 struct kvm_kernel_irq_routing_entry *e;
7590 struct kvm_irq_routing_table *irq_rt;
7591 struct kvm_lapic_irq irq;
7592 struct kvm_vcpu *vcpu;
7593 struct vcpu_data vcpu_info;
7596 if (!kvm_arch_has_assigned_device(kvm) ||
7597 !irq_remapping_cap(IRQ_POSTING_CAP) ||
7598 !kvm_vcpu_apicv_active(kvm->vcpus[0]))
7601 idx = srcu_read_lock(&kvm->irq_srcu);
7602 irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
7603 if (guest_irq >= irq_rt->nr_rt_entries ||
7604 hlist_empty(&irq_rt->map[guest_irq])) {
7605 pr_warn_once("no route for guest_irq %u/%u (broken user space?)\n",
7606 guest_irq, irq_rt->nr_rt_entries);
7610 hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
7611 if (e->type != KVM_IRQ_ROUTING_MSI)
7614 * VT-d PI cannot support posting multicast/broadcast
7615 * interrupts to a vCPU, we still use interrupt remapping
7616 * for these kind of interrupts.
7618 * For lowest-priority interrupts, we only support
7619 * those with single CPU as the destination, e.g. user
7620 * configures the interrupts via /proc/irq or uses
7621 * irqbalance to make the interrupts single-CPU.
7623 * We will support full lowest-priority interrupt later.
7625 * In addition, we can only inject generic interrupts using
7626 * the PI mechanism, refuse to route others through it.
7629 kvm_set_msi_irq(kvm, e, &irq);
7630 if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu) ||
7631 !kvm_irq_is_postable(&irq)) {
7633 * Make sure the IRTE is in remapped mode if
7634 * we don't handle it in posted mode.
7636 ret = irq_set_vcpu_affinity(host_irq, NULL);
7639 "failed to back to remapped mode, irq: %u\n",
7647 vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
7648 vcpu_info.vector = irq.vector;
7650 trace_kvm_pi_irte_update(host_irq, vcpu->vcpu_id, e->gsi,
7651 vcpu_info.vector, vcpu_info.pi_desc_addr, set);
7654 ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
7656 ret = irq_set_vcpu_affinity(host_irq, NULL);
7659 printk(KERN_INFO "%s: failed to update PI IRTE\n",
7667 srcu_read_unlock(&kvm->irq_srcu, idx);
7671 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
7673 if (vcpu->arch.mcg_cap & MCG_LMCE_P)
7674 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7675 FEAT_CTL_LMCE_ENABLED;
7677 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7678 ~FEAT_CTL_LMCE_ENABLED;
7681 static int vmx_smi_allowed(struct kvm_vcpu *vcpu)
7683 /* we need a nested vmexit to enter SMM, postpone if run is pending */
7684 if (to_vmx(vcpu)->nested.nested_run_pending)
7689 static int vmx_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
7691 struct vcpu_vmx *vmx = to_vmx(vcpu);
7693 vmx->nested.smm.guest_mode = is_guest_mode(vcpu);
7694 if (vmx->nested.smm.guest_mode)
7695 nested_vmx_vmexit(vcpu, -1, 0, 0);
7697 vmx->nested.smm.vmxon = vmx->nested.vmxon;
7698 vmx->nested.vmxon = false;
7699 vmx_clear_hlt(vcpu);
7703 static int vmx_pre_leave_smm(struct kvm_vcpu *vcpu, const char *smstate)
7705 struct vcpu_vmx *vmx = to_vmx(vcpu);
7708 if (vmx->nested.smm.vmxon) {
7709 vmx->nested.vmxon = true;
7710 vmx->nested.smm.vmxon = false;
7713 if (vmx->nested.smm.guest_mode) {
7714 ret = nested_vmx_enter_non_root_mode(vcpu, false);
7718 vmx->nested.smm.guest_mode = false;
7723 static int enable_smi_window(struct kvm_vcpu *vcpu)
7728 static bool vmx_need_emulation_on_page_fault(struct kvm_vcpu *vcpu)
7733 static bool vmx_apic_init_signal_blocked(struct kvm_vcpu *vcpu)
7735 return to_vmx(vcpu)->nested.vmxon;
7738 static void hardware_unsetup(void)
7741 nested_vmx_hardware_unsetup();
7746 static bool vmx_check_apicv_inhibit_reasons(ulong bit)
7748 ulong supported = BIT(APICV_INHIBIT_REASON_DISABLE) |
7749 BIT(APICV_INHIBIT_REASON_HYPERV);
7751 return supported & BIT(bit);
7754 static struct kvm_x86_ops vmx_x86_ops __initdata = {
7755 .hardware_unsetup = hardware_unsetup,
7757 .hardware_enable = hardware_enable,
7758 .hardware_disable = hardware_disable,
7759 .cpu_has_accelerated_tpr = report_flexpriority,
7760 .has_emulated_msr = vmx_has_emulated_msr,
7762 .vm_size = sizeof(struct kvm_vmx),
7763 .vm_init = vmx_vm_init,
7765 .vcpu_create = vmx_create_vcpu,
7766 .vcpu_free = vmx_free_vcpu,
7767 .vcpu_reset = vmx_vcpu_reset,
7769 .prepare_guest_switch = vmx_prepare_switch_to_guest,
7770 .vcpu_load = vmx_vcpu_load,
7771 .vcpu_put = vmx_vcpu_put,
7773 .update_bp_intercept = update_exception_bitmap,
7774 .get_msr_feature = vmx_get_msr_feature,
7775 .get_msr = vmx_get_msr,
7776 .set_msr = vmx_set_msr,
7777 .get_segment_base = vmx_get_segment_base,
7778 .get_segment = vmx_get_segment,
7779 .set_segment = vmx_set_segment,
7780 .get_cpl = vmx_get_cpl,
7781 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
7782 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
7783 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
7784 .set_cr0 = vmx_set_cr0,
7785 .set_cr4 = vmx_set_cr4,
7786 .set_efer = vmx_set_efer,
7787 .get_idt = vmx_get_idt,
7788 .set_idt = vmx_set_idt,
7789 .get_gdt = vmx_get_gdt,
7790 .set_gdt = vmx_set_gdt,
7791 .get_dr6 = vmx_get_dr6,
7792 .set_dr6 = vmx_set_dr6,
7793 .set_dr7 = vmx_set_dr7,
7794 .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
7795 .cache_reg = vmx_cache_reg,
7796 .get_rflags = vmx_get_rflags,
7797 .set_rflags = vmx_set_rflags,
7799 .tlb_flush_all = vmx_flush_tlb_all,
7800 .tlb_flush_current = vmx_flush_tlb_current,
7801 .tlb_flush_gva = vmx_flush_tlb_gva,
7802 .tlb_flush_guest = vmx_flush_tlb_guest,
7804 .run = vmx_vcpu_run,
7805 .handle_exit = vmx_handle_exit,
7806 .skip_emulated_instruction = vmx_skip_emulated_instruction,
7807 .update_emulated_instruction = vmx_update_emulated_instruction,
7808 .set_interrupt_shadow = vmx_set_interrupt_shadow,
7809 .get_interrupt_shadow = vmx_get_interrupt_shadow,
7810 .patch_hypercall = vmx_patch_hypercall,
7811 .set_irq = vmx_inject_irq,
7812 .set_nmi = vmx_inject_nmi,
7813 .queue_exception = vmx_queue_exception,
7814 .cancel_injection = vmx_cancel_injection,
7815 .interrupt_allowed = vmx_interrupt_allowed,
7816 .nmi_allowed = vmx_nmi_allowed,
7817 .get_nmi_mask = vmx_get_nmi_mask,
7818 .set_nmi_mask = vmx_set_nmi_mask,
7819 .enable_nmi_window = enable_nmi_window,
7820 .enable_irq_window = enable_irq_window,
7821 .update_cr8_intercept = update_cr8_intercept,
7822 .set_virtual_apic_mode = vmx_set_virtual_apic_mode,
7823 .set_apic_access_page_addr = vmx_set_apic_access_page_addr,
7824 .refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
7825 .load_eoi_exitmap = vmx_load_eoi_exitmap,
7826 .apicv_post_state_restore = vmx_apicv_post_state_restore,
7827 .check_apicv_inhibit_reasons = vmx_check_apicv_inhibit_reasons,
7828 .hwapic_irr_update = vmx_hwapic_irr_update,
7829 .hwapic_isr_update = vmx_hwapic_isr_update,
7830 .guest_apic_has_interrupt = vmx_guest_apic_has_interrupt,
7831 .sync_pir_to_irr = vmx_sync_pir_to_irr,
7832 .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
7833 .dy_apicv_has_pending_interrupt = vmx_dy_apicv_has_pending_interrupt,
7835 .set_tss_addr = vmx_set_tss_addr,
7836 .set_identity_map_addr = vmx_set_identity_map_addr,
7837 .get_tdp_level = get_ept_level,
7838 .get_mt_mask = vmx_get_mt_mask,
7840 .get_exit_info = vmx_get_exit_info,
7842 .cpuid_update = vmx_cpuid_update,
7844 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
7846 .read_l1_tsc_offset = vmx_read_l1_tsc_offset,
7847 .write_l1_tsc_offset = vmx_write_l1_tsc_offset,
7849 .load_mmu_pgd = vmx_load_mmu_pgd,
7851 .check_intercept = vmx_check_intercept,
7852 .handle_exit_irqoff = vmx_handle_exit_irqoff,
7854 .request_immediate_exit = vmx_request_immediate_exit,
7856 .sched_in = vmx_sched_in,
7858 .slot_enable_log_dirty = vmx_slot_enable_log_dirty,
7859 .slot_disable_log_dirty = vmx_slot_disable_log_dirty,
7860 .flush_log_dirty = vmx_flush_log_dirty,
7861 .enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
7862 .write_log_dirty = vmx_write_pml_buffer,
7864 .pre_block = vmx_pre_block,
7865 .post_block = vmx_post_block,
7867 .pmu_ops = &intel_pmu_ops,
7869 .update_pi_irte = vmx_update_pi_irte,
7871 #ifdef CONFIG_X86_64
7872 .set_hv_timer = vmx_set_hv_timer,
7873 .cancel_hv_timer = vmx_cancel_hv_timer,
7876 .setup_mce = vmx_setup_mce,
7878 .smi_allowed = vmx_smi_allowed,
7879 .pre_enter_smm = vmx_pre_enter_smm,
7880 .pre_leave_smm = vmx_pre_leave_smm,
7881 .enable_smi_window = enable_smi_window,
7883 .check_nested_events = NULL,
7884 .get_nested_state = NULL,
7885 .set_nested_state = NULL,
7886 .get_vmcs12_pages = NULL,
7887 .nested_enable_evmcs = NULL,
7888 .nested_get_evmcs_version = NULL,
7889 .need_emulation_on_page_fault = vmx_need_emulation_on_page_fault,
7890 .apic_init_signal_blocked = vmx_apic_init_signal_blocked,
7893 static __init int hardware_setup(void)
7895 unsigned long host_bndcfgs;
7897 int r, i, ept_lpage_level;
7900 host_idt_base = dt.address;
7902 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
7903 kvm_define_shared_msr(i, vmx_msr_index[i]);
7905 if (setup_vmcs_config(&vmcs_config, &vmx_capability) < 0)
7908 if (boot_cpu_has(X86_FEATURE_NX))
7909 kvm_enable_efer_bits(EFER_NX);
7911 if (boot_cpu_has(X86_FEATURE_MPX)) {
7912 rdmsrl(MSR_IA32_BNDCFGS, host_bndcfgs);
7913 WARN_ONCE(host_bndcfgs, "KVM: BNDCFGS in host will be lost");
7916 if (!cpu_has_vmx_mpx())
7917 supported_xcr0 &= ~(XFEATURE_MASK_BNDREGS |
7918 XFEATURE_MASK_BNDCSR);
7920 if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
7921 !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
7924 if (!cpu_has_vmx_ept() ||
7925 !cpu_has_vmx_ept_4levels() ||
7926 !cpu_has_vmx_ept_mt_wb() ||
7927 !cpu_has_vmx_invept_global())
7930 if (!cpu_has_vmx_ept_ad_bits() || !enable_ept)
7931 enable_ept_ad_bits = 0;
7933 if (!cpu_has_vmx_unrestricted_guest() || !enable_ept)
7934 enable_unrestricted_guest = 0;
7936 if (!cpu_has_vmx_flexpriority())
7937 flexpriority_enabled = 0;
7939 if (!cpu_has_virtual_nmis())
7943 * set_apic_access_page_addr() is used to reload apic access
7944 * page upon invalidation. No need to do anything if not
7945 * using the APIC_ACCESS_ADDR VMCS field.
7947 if (!flexpriority_enabled)
7948 vmx_x86_ops.set_apic_access_page_addr = NULL;
7950 if (!cpu_has_vmx_tpr_shadow())
7951 vmx_x86_ops.update_cr8_intercept = NULL;
7953 #if IS_ENABLED(CONFIG_HYPERV)
7954 if (ms_hyperv.nested_features & HV_X64_NESTED_GUEST_MAPPING_FLUSH
7956 vmx_x86_ops.tlb_remote_flush = hv_remote_flush_tlb;
7957 vmx_x86_ops.tlb_remote_flush_with_range =
7958 hv_remote_flush_tlb_with_range;
7962 if (!cpu_has_vmx_ple()) {
7965 ple_window_grow = 0;
7967 ple_window_shrink = 0;
7970 if (!cpu_has_vmx_apicv()) {
7972 vmx_x86_ops.sync_pir_to_irr = NULL;
7975 if (cpu_has_vmx_tsc_scaling()) {
7976 kvm_has_tsc_control = true;
7977 kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
7978 kvm_tsc_scaling_ratio_frac_bits = 48;
7981 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
7987 ept_lpage_level = 0;
7988 else if (cpu_has_vmx_ept_1g_page())
7989 ept_lpage_level = PT_PDPE_LEVEL;
7990 else if (cpu_has_vmx_ept_2m_page())
7991 ept_lpage_level = PT_DIRECTORY_LEVEL;
7993 ept_lpage_level = PT_PAGE_TABLE_LEVEL;
7994 kvm_configure_mmu(enable_ept, ept_lpage_level);
7997 * Only enable PML when hardware supports PML feature, and both EPT
7998 * and EPT A/D bit features are enabled -- PML depends on them to work.
8000 if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
8004 vmx_x86_ops.slot_enable_log_dirty = NULL;
8005 vmx_x86_ops.slot_disable_log_dirty = NULL;
8006 vmx_x86_ops.flush_log_dirty = NULL;
8007 vmx_x86_ops.enable_log_dirty_pt_masked = NULL;
8010 if (!cpu_has_vmx_preemption_timer())
8011 enable_preemption_timer = false;
8013 if (enable_preemption_timer) {
8014 u64 use_timer_freq = 5000ULL * 1000 * 1000;
8017 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
8018 cpu_preemption_timer_multi =
8019 vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
8022 use_timer_freq = (u64)tsc_khz * 1000;
8023 use_timer_freq >>= cpu_preemption_timer_multi;
8026 * KVM "disables" the preemption timer by setting it to its max
8027 * value. Don't use the timer if it might cause spurious exits
8028 * at a rate faster than 0.1 Hz (of uninterrupted guest time).
8030 if (use_timer_freq > 0xffffffffu / 10)
8031 enable_preemption_timer = false;
8034 if (!enable_preemption_timer) {
8035 vmx_x86_ops.set_hv_timer = NULL;
8036 vmx_x86_ops.cancel_hv_timer = NULL;
8037 vmx_x86_ops.request_immediate_exit = __kvm_request_immediate_exit;
8040 kvm_set_posted_intr_wakeup_handler(wakeup_handler);
8042 kvm_mce_cap_supported |= MCG_LMCE_P;
8044 if (pt_mode != PT_MODE_SYSTEM && pt_mode != PT_MODE_HOST_GUEST)
8046 if (!enable_ept || !cpu_has_vmx_intel_pt())
8047 pt_mode = PT_MODE_SYSTEM;
8050 nested_vmx_setup_ctls_msrs(&vmcs_config.nested,
8051 vmx_capability.ept);
8053 r = nested_vmx_hardware_setup(&vmx_x86_ops,
8054 kvm_vmx_exit_handlers);
8061 r = alloc_kvm_area();
8063 nested_vmx_hardware_unsetup();
8067 static struct kvm_x86_init_ops vmx_init_ops __initdata = {
8068 .cpu_has_kvm_support = cpu_has_kvm_support,
8069 .disabled_by_bios = vmx_disabled_by_bios,
8070 .check_processor_compatibility = vmx_check_processor_compat,
8071 .hardware_setup = hardware_setup,
8073 .runtime_ops = &vmx_x86_ops,
8076 static void vmx_cleanup_l1d_flush(void)
8078 if (vmx_l1d_flush_pages) {
8079 free_pages((unsigned long)vmx_l1d_flush_pages, L1D_CACHE_ORDER);
8080 vmx_l1d_flush_pages = NULL;
8082 /* Restore state so sysfs ignores VMX */
8083 l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
8086 static void vmx_exit(void)
8088 #ifdef CONFIG_KEXEC_CORE
8089 RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
8095 #if IS_ENABLED(CONFIG_HYPERV)
8096 if (static_branch_unlikely(&enable_evmcs)) {
8098 struct hv_vp_assist_page *vp_ap;
8100 * Reset everything to support using non-enlightened VMCS
8101 * access later (e.g. when we reload the module with
8102 * enlightened_vmcs=0)
8104 for_each_online_cpu(cpu) {
8105 vp_ap = hv_get_vp_assist_page(cpu);
8110 vp_ap->nested_control.features.directhypercall = 0;
8111 vp_ap->current_nested_vmcs = 0;
8112 vp_ap->enlighten_vmentry = 0;
8115 static_branch_disable(&enable_evmcs);
8118 vmx_cleanup_l1d_flush();
8120 module_exit(vmx_exit);
8122 static int __init vmx_init(void)
8126 #if IS_ENABLED(CONFIG_HYPERV)
8128 * Enlightened VMCS usage should be recommended and the host needs
8129 * to support eVMCS v1 or above. We can also disable eVMCS support
8130 * with module parameter.
8132 if (enlightened_vmcs &&
8133 ms_hyperv.hints & HV_X64_ENLIGHTENED_VMCS_RECOMMENDED &&
8134 (ms_hyperv.nested_features & HV_X64_ENLIGHTENED_VMCS_VERSION) >=
8135 KVM_EVMCS_VERSION) {
8138 /* Check that we have assist pages on all online CPUs */
8139 for_each_online_cpu(cpu) {
8140 if (!hv_get_vp_assist_page(cpu)) {
8141 enlightened_vmcs = false;
8146 if (enlightened_vmcs) {
8147 pr_info("KVM: vmx: using Hyper-V Enlightened VMCS\n");
8148 static_branch_enable(&enable_evmcs);
8151 if (ms_hyperv.nested_features & HV_X64_NESTED_DIRECT_FLUSH)
8152 vmx_x86_ops.enable_direct_tlbflush
8153 = hv_enable_direct_tlbflush;
8156 enlightened_vmcs = false;
8160 r = kvm_init(&vmx_init_ops, sizeof(struct vcpu_vmx),
8161 __alignof__(struct vcpu_vmx), THIS_MODULE);
8166 * Must be called after kvm_init() so enable_ept is properly set
8167 * up. Hand the parameter mitigation value in which was stored in
8168 * the pre module init parser. If no parameter was given, it will
8169 * contain 'auto' which will be turned into the default 'cond'
8172 r = vmx_setup_l1d_flush(vmentry_l1d_flush_param);
8178 for_each_possible_cpu(cpu) {
8179 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
8180 INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
8181 spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
8184 #ifdef CONFIG_KEXEC_CORE
8185 rcu_assign_pointer(crash_vmclear_loaded_vmcss,
8186 crash_vmclear_local_loaded_vmcss);
8188 vmx_check_vmcs12_offsets();
8192 module_init(vmx_init);