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;
1341 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1344 * Linux uses per-cpu TSS and GDT, so set these when switching
1345 * processors. See 22.2.4.
1347 vmcs_writel(HOST_TR_BASE,
1348 (unsigned long)&get_cpu_entry_area(cpu)->tss.x86_tss);
1349 vmcs_writel(HOST_GDTR_BASE, (unsigned long)gdt); /* 22.2.4 */
1351 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1352 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1354 vmx->loaded_vmcs->cpu = cpu;
1357 /* Setup TSC multiplier */
1358 if (kvm_has_tsc_control &&
1359 vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio)
1360 decache_tsc_multiplier(vmx);
1364 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1365 * vcpu mutex is already taken.
1367 void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1369 struct vcpu_vmx *vmx = to_vmx(vcpu);
1371 vmx_vcpu_load_vmcs(vcpu, cpu);
1373 vmx_vcpu_pi_load(vcpu, cpu);
1375 vmx->host_pkru = read_pkru();
1376 vmx->host_debugctlmsr = get_debugctlmsr();
1379 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
1381 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1383 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
1384 !irq_remapping_cap(IRQ_POSTING_CAP) ||
1385 !kvm_vcpu_apicv_active(vcpu))
1388 /* Set SN when the vCPU is preempted */
1389 if (vcpu->preempted)
1393 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1395 vmx_vcpu_pi_put(vcpu);
1397 vmx_prepare_switch_to_host(to_vmx(vcpu));
1400 static bool emulation_required(struct kvm_vcpu *vcpu)
1402 return emulate_invalid_guest_state && !guest_state_valid(vcpu);
1405 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1407 struct vcpu_vmx *vmx = to_vmx(vcpu);
1408 unsigned long rflags, save_rflags;
1410 if (!kvm_register_is_available(vcpu, VCPU_EXREG_RFLAGS)) {
1411 kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1412 rflags = vmcs_readl(GUEST_RFLAGS);
1413 if (vmx->rmode.vm86_active) {
1414 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1415 save_rflags = vmx->rmode.save_rflags;
1416 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1418 vmx->rflags = rflags;
1423 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1425 struct vcpu_vmx *vmx = to_vmx(vcpu);
1426 unsigned long old_rflags;
1428 if (enable_unrestricted_guest) {
1429 kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1430 vmx->rflags = rflags;
1431 vmcs_writel(GUEST_RFLAGS, rflags);
1435 old_rflags = vmx_get_rflags(vcpu);
1436 vmx->rflags = rflags;
1437 if (vmx->rmode.vm86_active) {
1438 vmx->rmode.save_rflags = rflags;
1439 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1441 vmcs_writel(GUEST_RFLAGS, rflags);
1443 if ((old_rflags ^ vmx->rflags) & X86_EFLAGS_VM)
1444 vmx->emulation_required = emulation_required(vcpu);
1447 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
1449 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1452 if (interruptibility & GUEST_INTR_STATE_STI)
1453 ret |= KVM_X86_SHADOW_INT_STI;
1454 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1455 ret |= KVM_X86_SHADOW_INT_MOV_SS;
1460 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1462 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1463 u32 interruptibility = interruptibility_old;
1465 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1467 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1468 interruptibility |= GUEST_INTR_STATE_MOV_SS;
1469 else if (mask & KVM_X86_SHADOW_INT_STI)
1470 interruptibility |= GUEST_INTR_STATE_STI;
1472 if ((interruptibility != interruptibility_old))
1473 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1476 static int vmx_rtit_ctl_check(struct kvm_vcpu *vcpu, u64 data)
1478 struct vcpu_vmx *vmx = to_vmx(vcpu);
1479 unsigned long value;
1482 * Any MSR write that attempts to change bits marked reserved will
1485 if (data & vmx->pt_desc.ctl_bitmask)
1489 * Any attempt to modify IA32_RTIT_CTL while TraceEn is set will
1490 * result in a #GP unless the same write also clears TraceEn.
1492 if ((vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) &&
1493 ((vmx->pt_desc.guest.ctl ^ data) & ~RTIT_CTL_TRACEEN))
1497 * WRMSR to IA32_RTIT_CTL that sets TraceEn but clears this bit
1498 * and FabricEn would cause #GP, if
1499 * CPUID.(EAX=14H, ECX=0):ECX.SNGLRGNOUT[bit 2] = 0
1501 if ((data & RTIT_CTL_TRACEEN) && !(data & RTIT_CTL_TOPA) &&
1502 !(data & RTIT_CTL_FABRIC_EN) &&
1503 !intel_pt_validate_cap(vmx->pt_desc.caps,
1504 PT_CAP_single_range_output))
1508 * MTCFreq, CycThresh and PSBFreq encodings check, any MSR write that
1509 * utilize encodings marked reserved will casue a #GP fault.
1511 value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc_periods);
1512 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc) &&
1513 !test_bit((data & RTIT_CTL_MTC_RANGE) >>
1514 RTIT_CTL_MTC_RANGE_OFFSET, &value))
1516 value = intel_pt_validate_cap(vmx->pt_desc.caps,
1517 PT_CAP_cycle_thresholds);
1518 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1519 !test_bit((data & RTIT_CTL_CYC_THRESH) >>
1520 RTIT_CTL_CYC_THRESH_OFFSET, &value))
1522 value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_periods);
1523 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1524 !test_bit((data & RTIT_CTL_PSB_FREQ) >>
1525 RTIT_CTL_PSB_FREQ_OFFSET, &value))
1529 * If ADDRx_CFG is reserved or the encodings is >2 will
1530 * cause a #GP fault.
1532 value = (data & RTIT_CTL_ADDR0) >> RTIT_CTL_ADDR0_OFFSET;
1533 if ((value && (vmx->pt_desc.addr_range < 1)) || (value > 2))
1535 value = (data & RTIT_CTL_ADDR1) >> RTIT_CTL_ADDR1_OFFSET;
1536 if ((value && (vmx->pt_desc.addr_range < 2)) || (value > 2))
1538 value = (data & RTIT_CTL_ADDR2) >> RTIT_CTL_ADDR2_OFFSET;
1539 if ((value && (vmx->pt_desc.addr_range < 3)) || (value > 2))
1541 value = (data & RTIT_CTL_ADDR3) >> RTIT_CTL_ADDR3_OFFSET;
1542 if ((value && (vmx->pt_desc.addr_range < 4)) || (value > 2))
1548 static int skip_emulated_instruction(struct kvm_vcpu *vcpu)
1553 * Using VMCS.VM_EXIT_INSTRUCTION_LEN on EPT misconfig depends on
1554 * undefined behavior: Intel's SDM doesn't mandate the VMCS field be
1555 * set when EPT misconfig occurs. In practice, real hardware updates
1556 * VM_EXIT_INSTRUCTION_LEN on EPT misconfig, but other hypervisors
1557 * (namely Hyper-V) don't set it due to it being undefined behavior,
1558 * i.e. we end up advancing IP with some random value.
1560 if (!static_cpu_has(X86_FEATURE_HYPERVISOR) ||
1561 to_vmx(vcpu)->exit_reason != EXIT_REASON_EPT_MISCONFIG) {
1562 rip = kvm_rip_read(vcpu);
1563 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1564 kvm_rip_write(vcpu, rip);
1566 if (!kvm_emulate_instruction(vcpu, EMULTYPE_SKIP))
1570 /* skipping an emulated instruction also counts */
1571 vmx_set_interrupt_shadow(vcpu, 0);
1578 * Recognizes a pending MTF VM-exit and records the nested state for later
1581 static void vmx_update_emulated_instruction(struct kvm_vcpu *vcpu)
1583 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1584 struct vcpu_vmx *vmx = to_vmx(vcpu);
1586 if (!is_guest_mode(vcpu))
1590 * Per the SDM, MTF takes priority over debug-trap exceptions besides
1591 * T-bit traps. As instruction emulation is completed (i.e. at the
1592 * instruction boundary), any #DB exception pending delivery must be a
1593 * debug-trap. Record the pending MTF state to be delivered in
1594 * vmx_check_nested_events().
1596 if (nested_cpu_has_mtf(vmcs12) &&
1597 (!vcpu->arch.exception.pending ||
1598 vcpu->arch.exception.nr == DB_VECTOR))
1599 vmx->nested.mtf_pending = true;
1601 vmx->nested.mtf_pending = false;
1604 static int vmx_skip_emulated_instruction(struct kvm_vcpu *vcpu)
1606 vmx_update_emulated_instruction(vcpu);
1607 return skip_emulated_instruction(vcpu);
1610 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1613 * Ensure that we clear the HLT state in the VMCS. We don't need to
1614 * explicitly skip the instruction because if the HLT state is set,
1615 * then the instruction is already executing and RIP has already been
1618 if (kvm_hlt_in_guest(vcpu->kvm) &&
1619 vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1620 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1623 static void vmx_queue_exception(struct kvm_vcpu *vcpu)
1625 struct vcpu_vmx *vmx = to_vmx(vcpu);
1626 unsigned nr = vcpu->arch.exception.nr;
1627 bool has_error_code = vcpu->arch.exception.has_error_code;
1628 u32 error_code = vcpu->arch.exception.error_code;
1629 u32 intr_info = nr | INTR_INFO_VALID_MASK;
1631 kvm_deliver_exception_payload(vcpu);
1633 if (has_error_code) {
1634 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1635 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1638 if (vmx->rmode.vm86_active) {
1640 if (kvm_exception_is_soft(nr))
1641 inc_eip = vcpu->arch.event_exit_inst_len;
1642 kvm_inject_realmode_interrupt(vcpu, nr, inc_eip);
1646 WARN_ON_ONCE(vmx->emulation_required);
1648 if (kvm_exception_is_soft(nr)) {
1649 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1650 vmx->vcpu.arch.event_exit_inst_len);
1651 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1653 intr_info |= INTR_TYPE_HARD_EXCEPTION;
1655 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1657 vmx_clear_hlt(vcpu);
1661 * Swap MSR entry in host/guest MSR entry array.
1663 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1665 struct shared_msr_entry tmp;
1667 tmp = vmx->guest_msrs[to];
1668 vmx->guest_msrs[to] = vmx->guest_msrs[from];
1669 vmx->guest_msrs[from] = tmp;
1673 * Set up the vmcs to automatically save and restore system
1674 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
1675 * mode, as fiddling with msrs is very expensive.
1677 static void setup_msrs(struct vcpu_vmx *vmx)
1679 int save_nmsrs, index;
1682 #ifdef CONFIG_X86_64
1684 * The SYSCALL MSRs are only needed on long mode guests, and only
1685 * when EFER.SCE is set.
1687 if (is_long_mode(&vmx->vcpu) && (vmx->vcpu.arch.efer & EFER_SCE)) {
1688 index = __find_msr_index(vmx, MSR_STAR);
1690 move_msr_up(vmx, index, save_nmsrs++);
1691 index = __find_msr_index(vmx, MSR_LSTAR);
1693 move_msr_up(vmx, index, save_nmsrs++);
1694 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1696 move_msr_up(vmx, index, save_nmsrs++);
1699 index = __find_msr_index(vmx, MSR_EFER);
1700 if (index >= 0 && update_transition_efer(vmx, index))
1701 move_msr_up(vmx, index, save_nmsrs++);
1702 index = __find_msr_index(vmx, MSR_TSC_AUX);
1703 if (index >= 0 && guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDTSCP))
1704 move_msr_up(vmx, index, save_nmsrs++);
1705 index = __find_msr_index(vmx, MSR_IA32_TSX_CTRL);
1707 move_msr_up(vmx, index, save_nmsrs++);
1709 vmx->save_nmsrs = save_nmsrs;
1710 vmx->guest_msrs_ready = false;
1712 if (cpu_has_vmx_msr_bitmap())
1713 vmx_update_msr_bitmap(&vmx->vcpu);
1716 static u64 vmx_read_l1_tsc_offset(struct kvm_vcpu *vcpu)
1718 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1720 if (is_guest_mode(vcpu) &&
1721 (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING))
1722 return vcpu->arch.tsc_offset - vmcs12->tsc_offset;
1724 return vcpu->arch.tsc_offset;
1727 static u64 vmx_write_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1729 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1730 u64 g_tsc_offset = 0;
1733 * We're here if L1 chose not to trap WRMSR to TSC. According
1734 * to the spec, this should set L1's TSC; The offset that L1
1735 * set for L2 remains unchanged, and still needs to be added
1736 * to the newly set TSC to get L2's TSC.
1738 if (is_guest_mode(vcpu) &&
1739 (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING))
1740 g_tsc_offset = vmcs12->tsc_offset;
1742 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1743 vcpu->arch.tsc_offset - g_tsc_offset,
1745 vmcs_write64(TSC_OFFSET, offset + g_tsc_offset);
1746 return offset + g_tsc_offset;
1750 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
1751 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
1752 * all guests if the "nested" module option is off, and can also be disabled
1753 * for a single guest by disabling its VMX cpuid bit.
1755 bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
1757 return nested && guest_cpuid_has(vcpu, X86_FEATURE_VMX);
1760 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
1763 uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
1765 return !(val & ~valid_bits);
1768 static int vmx_get_msr_feature(struct kvm_msr_entry *msr)
1770 switch (msr->index) {
1771 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1774 return vmx_get_vmx_msr(&vmcs_config.nested, msr->index, &msr->data);
1781 * Reads an msr value (of 'msr_index') into 'pdata'.
1782 * Returns 0 on success, non-0 otherwise.
1783 * Assumes vcpu_load() was already called.
1785 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1787 struct vcpu_vmx *vmx = to_vmx(vcpu);
1788 struct shared_msr_entry *msr;
1791 switch (msr_info->index) {
1792 #ifdef CONFIG_X86_64
1794 msr_info->data = vmcs_readl(GUEST_FS_BASE);
1797 msr_info->data = vmcs_readl(GUEST_GS_BASE);
1799 case MSR_KERNEL_GS_BASE:
1800 msr_info->data = vmx_read_guest_kernel_gs_base(vmx);
1804 return kvm_get_msr_common(vcpu, msr_info);
1805 case MSR_IA32_TSX_CTRL:
1806 if (!msr_info->host_initiated &&
1807 !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
1809 goto find_shared_msr;
1810 case MSR_IA32_UMWAIT_CONTROL:
1811 if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
1814 msr_info->data = vmx->msr_ia32_umwait_control;
1816 case MSR_IA32_SPEC_CTRL:
1817 if (!msr_info->host_initiated &&
1818 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1821 msr_info->data = to_vmx(vcpu)->spec_ctrl;
1823 case MSR_IA32_SYSENTER_CS:
1824 msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
1826 case MSR_IA32_SYSENTER_EIP:
1827 msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
1829 case MSR_IA32_SYSENTER_ESP:
1830 msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
1832 case MSR_IA32_BNDCFGS:
1833 if (!kvm_mpx_supported() ||
1834 (!msr_info->host_initiated &&
1835 !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1837 msr_info->data = vmcs_read64(GUEST_BNDCFGS);
1839 case MSR_IA32_MCG_EXT_CTL:
1840 if (!msr_info->host_initiated &&
1841 !(vmx->msr_ia32_feature_control &
1842 FEAT_CTL_LMCE_ENABLED))
1844 msr_info->data = vcpu->arch.mcg_ext_ctl;
1846 case MSR_IA32_FEAT_CTL:
1847 msr_info->data = vmx->msr_ia32_feature_control;
1849 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1850 if (!nested_vmx_allowed(vcpu))
1852 if (vmx_get_vmx_msr(&vmx->nested.msrs, msr_info->index,
1856 * Enlightened VMCS v1 doesn't have certain fields, but buggy
1857 * Hyper-V versions are still trying to use corresponding
1858 * features when they are exposed. Filter out the essential
1861 if (!msr_info->host_initiated &&
1862 vmx->nested.enlightened_vmcs_enabled)
1863 nested_evmcs_filter_control_msr(msr_info->index,
1866 case MSR_IA32_RTIT_CTL:
1867 if (!vmx_pt_mode_is_host_guest())
1869 msr_info->data = vmx->pt_desc.guest.ctl;
1871 case MSR_IA32_RTIT_STATUS:
1872 if (!vmx_pt_mode_is_host_guest())
1874 msr_info->data = vmx->pt_desc.guest.status;
1876 case MSR_IA32_RTIT_CR3_MATCH:
1877 if (!vmx_pt_mode_is_host_guest() ||
1878 !intel_pt_validate_cap(vmx->pt_desc.caps,
1879 PT_CAP_cr3_filtering))
1881 msr_info->data = vmx->pt_desc.guest.cr3_match;
1883 case MSR_IA32_RTIT_OUTPUT_BASE:
1884 if (!vmx_pt_mode_is_host_guest() ||
1885 (!intel_pt_validate_cap(vmx->pt_desc.caps,
1886 PT_CAP_topa_output) &&
1887 !intel_pt_validate_cap(vmx->pt_desc.caps,
1888 PT_CAP_single_range_output)))
1890 msr_info->data = vmx->pt_desc.guest.output_base;
1892 case MSR_IA32_RTIT_OUTPUT_MASK:
1893 if (!vmx_pt_mode_is_host_guest() ||
1894 (!intel_pt_validate_cap(vmx->pt_desc.caps,
1895 PT_CAP_topa_output) &&
1896 !intel_pt_validate_cap(vmx->pt_desc.caps,
1897 PT_CAP_single_range_output)))
1899 msr_info->data = vmx->pt_desc.guest.output_mask;
1901 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
1902 index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
1903 if (!vmx_pt_mode_is_host_guest() ||
1904 (index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
1905 PT_CAP_num_address_ranges)))
1908 msr_info->data = vmx->pt_desc.guest.addr_b[index / 2];
1910 msr_info->data = vmx->pt_desc.guest.addr_a[index / 2];
1913 if (!msr_info->host_initiated &&
1914 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
1916 goto find_shared_msr;
1919 msr = find_msr_entry(vmx, msr_info->index);
1921 msr_info->data = msr->data;
1924 return kvm_get_msr_common(vcpu, msr_info);
1931 * Writes msr value into the appropriate "register".
1932 * Returns 0 on success, non-0 otherwise.
1933 * Assumes vcpu_load() was already called.
1935 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1937 struct vcpu_vmx *vmx = to_vmx(vcpu);
1938 struct shared_msr_entry *msr;
1940 u32 msr_index = msr_info->index;
1941 u64 data = msr_info->data;
1944 switch (msr_index) {
1946 ret = kvm_set_msr_common(vcpu, msr_info);
1948 #ifdef CONFIG_X86_64
1950 vmx_segment_cache_clear(vmx);
1951 vmcs_writel(GUEST_FS_BASE, data);
1954 vmx_segment_cache_clear(vmx);
1955 vmcs_writel(GUEST_GS_BASE, data);
1957 case MSR_KERNEL_GS_BASE:
1958 vmx_write_guest_kernel_gs_base(vmx, data);
1961 case MSR_IA32_SYSENTER_CS:
1962 if (is_guest_mode(vcpu))
1963 get_vmcs12(vcpu)->guest_sysenter_cs = data;
1964 vmcs_write32(GUEST_SYSENTER_CS, data);
1966 case MSR_IA32_SYSENTER_EIP:
1967 if (is_guest_mode(vcpu))
1968 get_vmcs12(vcpu)->guest_sysenter_eip = data;
1969 vmcs_writel(GUEST_SYSENTER_EIP, data);
1971 case MSR_IA32_SYSENTER_ESP:
1972 if (is_guest_mode(vcpu))
1973 get_vmcs12(vcpu)->guest_sysenter_esp = data;
1974 vmcs_writel(GUEST_SYSENTER_ESP, data);
1976 case MSR_IA32_DEBUGCTLMSR:
1977 if (is_guest_mode(vcpu) && get_vmcs12(vcpu)->vm_exit_controls &
1978 VM_EXIT_SAVE_DEBUG_CONTROLS)
1979 get_vmcs12(vcpu)->guest_ia32_debugctl = data;
1981 ret = kvm_set_msr_common(vcpu, msr_info);
1984 case MSR_IA32_BNDCFGS:
1985 if (!kvm_mpx_supported() ||
1986 (!msr_info->host_initiated &&
1987 !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1989 if (is_noncanonical_address(data & PAGE_MASK, vcpu) ||
1990 (data & MSR_IA32_BNDCFGS_RSVD))
1992 vmcs_write64(GUEST_BNDCFGS, data);
1994 case MSR_IA32_UMWAIT_CONTROL:
1995 if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
1998 /* The reserved bit 1 and non-32 bit [63:32] should be zero */
1999 if (data & (BIT_ULL(1) | GENMASK_ULL(63, 32)))
2002 vmx->msr_ia32_umwait_control = data;
2004 case MSR_IA32_SPEC_CTRL:
2005 if (!msr_info->host_initiated &&
2006 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
2009 if (data & ~kvm_spec_ctrl_valid_bits(vcpu))
2012 vmx->spec_ctrl = data;
2018 * When it's written (to non-zero) for the first time, pass
2022 * The handling of the MSR bitmap for L2 guests is done in
2023 * nested_vmx_prepare_msr_bitmap. We should not touch the
2024 * vmcs02.msr_bitmap here since it gets completely overwritten
2025 * in the merging. We update the vmcs01 here for L1 as well
2026 * since it will end up touching the MSR anyway now.
2028 vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap,
2032 case MSR_IA32_TSX_CTRL:
2033 if (!msr_info->host_initiated &&
2034 !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
2036 if (data & ~(TSX_CTRL_RTM_DISABLE | TSX_CTRL_CPUID_CLEAR))
2038 goto find_shared_msr;
2039 case MSR_IA32_PRED_CMD:
2040 if (!msr_info->host_initiated &&
2041 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
2044 if (data & ~PRED_CMD_IBPB)
2046 if (!boot_cpu_has(X86_FEATURE_SPEC_CTRL))
2051 wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
2055 * When it's written (to non-zero) for the first time, pass
2059 * The handling of the MSR bitmap for L2 guests is done in
2060 * nested_vmx_prepare_msr_bitmap. We should not touch the
2061 * vmcs02.msr_bitmap here since it gets completely overwritten
2064 vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap, MSR_IA32_PRED_CMD,
2067 case MSR_IA32_CR_PAT:
2068 if (!kvm_pat_valid(data))
2071 if (is_guest_mode(vcpu) &&
2072 get_vmcs12(vcpu)->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
2073 get_vmcs12(vcpu)->guest_ia32_pat = data;
2075 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2076 vmcs_write64(GUEST_IA32_PAT, data);
2077 vcpu->arch.pat = data;
2080 ret = kvm_set_msr_common(vcpu, msr_info);
2082 case MSR_IA32_TSC_ADJUST:
2083 ret = kvm_set_msr_common(vcpu, msr_info);
2085 case MSR_IA32_MCG_EXT_CTL:
2086 if ((!msr_info->host_initiated &&
2087 !(to_vmx(vcpu)->msr_ia32_feature_control &
2088 FEAT_CTL_LMCE_ENABLED)) ||
2089 (data & ~MCG_EXT_CTL_LMCE_EN))
2091 vcpu->arch.mcg_ext_ctl = data;
2093 case MSR_IA32_FEAT_CTL:
2094 if (!vmx_feature_control_msr_valid(vcpu, data) ||
2095 (to_vmx(vcpu)->msr_ia32_feature_control &
2096 FEAT_CTL_LOCKED && !msr_info->host_initiated))
2098 vmx->msr_ia32_feature_control = data;
2099 if (msr_info->host_initiated && data == 0)
2100 vmx_leave_nested(vcpu);
2102 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2103 if (!msr_info->host_initiated)
2104 return 1; /* they are read-only */
2105 if (!nested_vmx_allowed(vcpu))
2107 return vmx_set_vmx_msr(vcpu, msr_index, data);
2108 case MSR_IA32_RTIT_CTL:
2109 if (!vmx_pt_mode_is_host_guest() ||
2110 vmx_rtit_ctl_check(vcpu, data) ||
2113 vmcs_write64(GUEST_IA32_RTIT_CTL, data);
2114 vmx->pt_desc.guest.ctl = data;
2115 pt_update_intercept_for_msr(vmx);
2117 case MSR_IA32_RTIT_STATUS:
2118 if (!pt_can_write_msr(vmx))
2120 if (data & MSR_IA32_RTIT_STATUS_MASK)
2122 vmx->pt_desc.guest.status = data;
2124 case MSR_IA32_RTIT_CR3_MATCH:
2125 if (!pt_can_write_msr(vmx))
2127 if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2128 PT_CAP_cr3_filtering))
2130 vmx->pt_desc.guest.cr3_match = data;
2132 case MSR_IA32_RTIT_OUTPUT_BASE:
2133 if (!pt_can_write_msr(vmx))
2135 if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2136 PT_CAP_topa_output) &&
2137 !intel_pt_validate_cap(vmx->pt_desc.caps,
2138 PT_CAP_single_range_output))
2140 if (data & MSR_IA32_RTIT_OUTPUT_BASE_MASK)
2142 vmx->pt_desc.guest.output_base = data;
2144 case MSR_IA32_RTIT_OUTPUT_MASK:
2145 if (!pt_can_write_msr(vmx))
2147 if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2148 PT_CAP_topa_output) &&
2149 !intel_pt_validate_cap(vmx->pt_desc.caps,
2150 PT_CAP_single_range_output))
2152 vmx->pt_desc.guest.output_mask = data;
2154 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
2155 if (!pt_can_write_msr(vmx))
2157 index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
2158 if (index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
2159 PT_CAP_num_address_ranges))
2161 if (is_noncanonical_address(data, vcpu))
2164 vmx->pt_desc.guest.addr_b[index / 2] = data;
2166 vmx->pt_desc.guest.addr_a[index / 2] = data;
2169 if (!msr_info->host_initiated &&
2170 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
2172 /* Check reserved bit, higher 32 bits should be zero */
2173 if ((data >> 32) != 0)
2175 goto find_shared_msr;
2179 msr = find_msr_entry(vmx, msr_index);
2181 ret = vmx_set_guest_msr(vmx, msr, data);
2183 ret = kvm_set_msr_common(vcpu, msr_info);
2189 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2191 kvm_register_mark_available(vcpu, reg);
2195 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2198 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2200 case VCPU_EXREG_PDPTR:
2202 ept_save_pdptrs(vcpu);
2204 case VCPU_EXREG_CR3:
2205 if (enable_unrestricted_guest || (enable_ept && is_paging(vcpu)))
2206 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2214 static __init int cpu_has_kvm_support(void)
2216 return cpu_has_vmx();
2219 static __init int vmx_disabled_by_bios(void)
2221 return !boot_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
2222 !boot_cpu_has(X86_FEATURE_VMX);
2225 static int kvm_cpu_vmxon(u64 vmxon_pointer)
2229 cr4_set_bits(X86_CR4_VMXE);
2230 intel_pt_handle_vmx(1);
2232 asm_volatile_goto("1: vmxon %[vmxon_pointer]\n\t"
2233 _ASM_EXTABLE(1b, %l[fault])
2234 : : [vmxon_pointer] "m"(vmxon_pointer)
2239 WARN_ONCE(1, "VMXON faulted, MSR_IA32_FEAT_CTL (0x3a) = 0x%llx\n",
2240 rdmsrl_safe(MSR_IA32_FEAT_CTL, &msr) ? 0xdeadbeef : msr);
2241 intel_pt_handle_vmx(0);
2242 cr4_clear_bits(X86_CR4_VMXE);
2247 static int hardware_enable(void)
2249 int cpu = raw_smp_processor_id();
2250 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2253 if (cr4_read_shadow() & X86_CR4_VMXE)
2257 * This can happen if we hot-added a CPU but failed to allocate
2258 * VP assist page for it.
2260 if (static_branch_unlikely(&enable_evmcs) &&
2261 !hv_get_vp_assist_page(cpu))
2264 r = kvm_cpu_vmxon(phys_addr);
2274 static void vmclear_local_loaded_vmcss(void)
2276 int cpu = raw_smp_processor_id();
2277 struct loaded_vmcs *v, *n;
2279 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2280 loaded_vmcss_on_cpu_link)
2281 __loaded_vmcs_clear(v);
2285 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2288 static void kvm_cpu_vmxoff(void)
2290 asm volatile (__ex("vmxoff"));
2292 intel_pt_handle_vmx(0);
2293 cr4_clear_bits(X86_CR4_VMXE);
2296 static void hardware_disable(void)
2298 vmclear_local_loaded_vmcss();
2303 * There is no X86_FEATURE for SGX yet, but anyway we need to query CPUID
2304 * directly instead of going through cpu_has(), to ensure KVM is trapping
2305 * ENCLS whenever it's supported in hardware. It does not matter whether
2306 * the host OS supports or has enabled SGX.
2308 static bool cpu_has_sgx(void)
2310 return cpuid_eax(0) >= 0x12 && (cpuid_eax(0x12) & BIT(0));
2313 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2314 u32 msr, u32 *result)
2316 u32 vmx_msr_low, vmx_msr_high;
2317 u32 ctl = ctl_min | ctl_opt;
2319 rdmsr(msr, vmx_msr_low, vmx_msr_high);
2321 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2322 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
2324 /* Ensure minimum (required) set of control bits are supported. */
2332 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf,
2333 struct vmx_capability *vmx_cap)
2335 u32 vmx_msr_low, vmx_msr_high;
2336 u32 min, opt, min2, opt2;
2337 u32 _pin_based_exec_control = 0;
2338 u32 _cpu_based_exec_control = 0;
2339 u32 _cpu_based_2nd_exec_control = 0;
2340 u32 _vmexit_control = 0;
2341 u32 _vmentry_control = 0;
2343 memset(vmcs_conf, 0, sizeof(*vmcs_conf));
2344 min = CPU_BASED_HLT_EXITING |
2345 #ifdef CONFIG_X86_64
2346 CPU_BASED_CR8_LOAD_EXITING |
2347 CPU_BASED_CR8_STORE_EXITING |
2349 CPU_BASED_CR3_LOAD_EXITING |
2350 CPU_BASED_CR3_STORE_EXITING |
2351 CPU_BASED_UNCOND_IO_EXITING |
2352 CPU_BASED_MOV_DR_EXITING |
2353 CPU_BASED_USE_TSC_OFFSETTING |
2354 CPU_BASED_MWAIT_EXITING |
2355 CPU_BASED_MONITOR_EXITING |
2356 CPU_BASED_INVLPG_EXITING |
2357 CPU_BASED_RDPMC_EXITING;
2359 opt = CPU_BASED_TPR_SHADOW |
2360 CPU_BASED_USE_MSR_BITMAPS |
2361 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2362 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2363 &_cpu_based_exec_control) < 0)
2365 #ifdef CONFIG_X86_64
2366 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2367 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2368 ~CPU_BASED_CR8_STORE_EXITING;
2370 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2372 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2373 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2374 SECONDARY_EXEC_WBINVD_EXITING |
2375 SECONDARY_EXEC_ENABLE_VPID |
2376 SECONDARY_EXEC_ENABLE_EPT |
2377 SECONDARY_EXEC_UNRESTRICTED_GUEST |
2378 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2379 SECONDARY_EXEC_DESC |
2380 SECONDARY_EXEC_RDTSCP |
2381 SECONDARY_EXEC_ENABLE_INVPCID |
2382 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2383 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2384 SECONDARY_EXEC_SHADOW_VMCS |
2385 SECONDARY_EXEC_XSAVES |
2386 SECONDARY_EXEC_RDSEED_EXITING |
2387 SECONDARY_EXEC_RDRAND_EXITING |
2388 SECONDARY_EXEC_ENABLE_PML |
2389 SECONDARY_EXEC_TSC_SCALING |
2390 SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE |
2391 SECONDARY_EXEC_PT_USE_GPA |
2392 SECONDARY_EXEC_PT_CONCEAL_VMX |
2393 SECONDARY_EXEC_ENABLE_VMFUNC;
2395 opt2 |= SECONDARY_EXEC_ENCLS_EXITING;
2396 if (adjust_vmx_controls(min2, opt2,
2397 MSR_IA32_VMX_PROCBASED_CTLS2,
2398 &_cpu_based_2nd_exec_control) < 0)
2401 #ifndef CONFIG_X86_64
2402 if (!(_cpu_based_2nd_exec_control &
2403 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2404 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2407 if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2408 _cpu_based_2nd_exec_control &= ~(
2409 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2410 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2411 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
2413 rdmsr_safe(MSR_IA32_VMX_EPT_VPID_CAP,
2414 &vmx_cap->ept, &vmx_cap->vpid);
2416 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2417 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2419 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2420 CPU_BASED_CR3_STORE_EXITING |
2421 CPU_BASED_INVLPG_EXITING);
2422 } else if (vmx_cap->ept) {
2424 pr_warn_once("EPT CAP should not exist if not support "
2425 "1-setting enable EPT VM-execution control\n");
2427 if (!(_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_VPID) &&
2430 pr_warn_once("VPID CAP should not exist if not support "
2431 "1-setting enable VPID VM-execution control\n");
2434 min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
2435 #ifdef CONFIG_X86_64
2436 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2438 opt = VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL |
2439 VM_EXIT_LOAD_IA32_PAT |
2440 VM_EXIT_LOAD_IA32_EFER |
2441 VM_EXIT_CLEAR_BNDCFGS |
2442 VM_EXIT_PT_CONCEAL_PIP |
2443 VM_EXIT_CLEAR_IA32_RTIT_CTL;
2444 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2445 &_vmexit_control) < 0)
2448 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2449 opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR |
2450 PIN_BASED_VMX_PREEMPTION_TIMER;
2451 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2452 &_pin_based_exec_control) < 0)
2455 if (cpu_has_broken_vmx_preemption_timer())
2456 _pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
2457 if (!(_cpu_based_2nd_exec_control &
2458 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
2459 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
2461 min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
2462 opt = VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL |
2463 VM_ENTRY_LOAD_IA32_PAT |
2464 VM_ENTRY_LOAD_IA32_EFER |
2465 VM_ENTRY_LOAD_BNDCFGS |
2466 VM_ENTRY_PT_CONCEAL_PIP |
2467 VM_ENTRY_LOAD_IA32_RTIT_CTL;
2468 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2469 &_vmentry_control) < 0)
2473 * Some cpus support VM_{ENTRY,EXIT}_IA32_PERF_GLOBAL_CTRL but they
2474 * can't be used due to an errata where VM Exit may incorrectly clear
2475 * IA32_PERF_GLOBAL_CTRL[34:32]. Workaround the errata by using the
2476 * MSR load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2478 if (boot_cpu_data.x86 == 0x6) {
2479 switch (boot_cpu_data.x86_model) {
2480 case 26: /* AAK155 */
2481 case 30: /* AAP115 */
2482 case 37: /* AAT100 */
2483 case 44: /* BC86,AAY89,BD102 */
2485 _vmentry_control &= ~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
2486 _vmexit_control &= ~VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
2487 pr_warn_once("kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
2488 "does not work properly. Using workaround\n");
2496 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2498 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2499 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2502 #ifdef CONFIG_X86_64
2503 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2504 if (vmx_msr_high & (1u<<16))
2508 /* Require Write-Back (WB) memory type for VMCS accesses. */
2509 if (((vmx_msr_high >> 18) & 15) != 6)
2512 vmcs_conf->size = vmx_msr_high & 0x1fff;
2513 vmcs_conf->order = get_order(vmcs_conf->size);
2514 vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
2516 vmcs_conf->revision_id = vmx_msr_low;
2518 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2519 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2520 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2521 vmcs_conf->vmexit_ctrl = _vmexit_control;
2522 vmcs_conf->vmentry_ctrl = _vmentry_control;
2524 if (static_branch_unlikely(&enable_evmcs))
2525 evmcs_sanitize_exec_ctrls(vmcs_conf);
2530 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags)
2532 int node = cpu_to_node(cpu);
2536 pages = __alloc_pages_node(node, flags, vmcs_config.order);
2539 vmcs = page_address(pages);
2540 memset(vmcs, 0, vmcs_config.size);
2542 /* KVM supports Enlightened VMCS v1 only */
2543 if (static_branch_unlikely(&enable_evmcs))
2544 vmcs->hdr.revision_id = KVM_EVMCS_VERSION;
2546 vmcs->hdr.revision_id = vmcs_config.revision_id;
2549 vmcs->hdr.shadow_vmcs = 1;
2553 void free_vmcs(struct vmcs *vmcs)
2555 free_pages((unsigned long)vmcs, vmcs_config.order);
2559 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2561 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2563 if (!loaded_vmcs->vmcs)
2565 loaded_vmcs_clear(loaded_vmcs);
2566 free_vmcs(loaded_vmcs->vmcs);
2567 loaded_vmcs->vmcs = NULL;
2568 if (loaded_vmcs->msr_bitmap)
2569 free_page((unsigned long)loaded_vmcs->msr_bitmap);
2570 WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
2573 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2575 loaded_vmcs->vmcs = alloc_vmcs(false);
2576 if (!loaded_vmcs->vmcs)
2579 vmcs_clear(loaded_vmcs->vmcs);
2581 loaded_vmcs->shadow_vmcs = NULL;
2582 loaded_vmcs->hv_timer_soft_disabled = false;
2583 loaded_vmcs->cpu = -1;
2584 loaded_vmcs->launched = 0;
2586 if (cpu_has_vmx_msr_bitmap()) {
2587 loaded_vmcs->msr_bitmap = (unsigned long *)
2588 __get_free_page(GFP_KERNEL_ACCOUNT);
2589 if (!loaded_vmcs->msr_bitmap)
2591 memset(loaded_vmcs->msr_bitmap, 0xff, PAGE_SIZE);
2593 if (IS_ENABLED(CONFIG_HYPERV) &&
2594 static_branch_unlikely(&enable_evmcs) &&
2595 (ms_hyperv.nested_features & HV_X64_NESTED_MSR_BITMAP)) {
2596 struct hv_enlightened_vmcs *evmcs =
2597 (struct hv_enlightened_vmcs *)loaded_vmcs->vmcs;
2599 evmcs->hv_enlightenments_control.msr_bitmap = 1;
2603 memset(&loaded_vmcs->host_state, 0, sizeof(struct vmcs_host_state));
2604 memset(&loaded_vmcs->controls_shadow, 0,
2605 sizeof(struct vmcs_controls_shadow));
2610 free_loaded_vmcs(loaded_vmcs);
2614 static void free_kvm_area(void)
2618 for_each_possible_cpu(cpu) {
2619 free_vmcs(per_cpu(vmxarea, cpu));
2620 per_cpu(vmxarea, cpu) = NULL;
2624 static __init int alloc_kvm_area(void)
2628 for_each_possible_cpu(cpu) {
2631 vmcs = alloc_vmcs_cpu(false, cpu, GFP_KERNEL);
2638 * When eVMCS is enabled, alloc_vmcs_cpu() sets
2639 * vmcs->revision_id to KVM_EVMCS_VERSION instead of
2640 * revision_id reported by MSR_IA32_VMX_BASIC.
2642 * However, even though not explicitly documented by
2643 * TLFS, VMXArea passed as VMXON argument should
2644 * still be marked with revision_id reported by
2647 if (static_branch_unlikely(&enable_evmcs))
2648 vmcs->hdr.revision_id = vmcs_config.revision_id;
2650 per_cpu(vmxarea, cpu) = vmcs;
2655 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
2656 struct kvm_segment *save)
2658 if (!emulate_invalid_guest_state) {
2660 * CS and SS RPL should be equal during guest entry according
2661 * to VMX spec, but in reality it is not always so. Since vcpu
2662 * is in the middle of the transition from real mode to
2663 * protected mode it is safe to assume that RPL 0 is a good
2666 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
2667 save->selector &= ~SEGMENT_RPL_MASK;
2668 save->dpl = save->selector & SEGMENT_RPL_MASK;
2671 vmx_set_segment(vcpu, save, seg);
2674 static void enter_pmode(struct kvm_vcpu *vcpu)
2676 unsigned long flags;
2677 struct vcpu_vmx *vmx = to_vmx(vcpu);
2680 * Update real mode segment cache. It may be not up-to-date if sement
2681 * register was written while vcpu was in a guest mode.
2683 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2684 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2685 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2686 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2687 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2688 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2690 vmx->rmode.vm86_active = 0;
2692 vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2694 flags = vmcs_readl(GUEST_RFLAGS);
2695 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2696 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2697 vmcs_writel(GUEST_RFLAGS, flags);
2699 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
2700 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
2702 update_exception_bitmap(vcpu);
2704 fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2705 fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2706 fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2707 fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2708 fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2709 fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2712 static void fix_rmode_seg(int seg, struct kvm_segment *save)
2714 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2715 struct kvm_segment var = *save;
2718 if (seg == VCPU_SREG_CS)
2721 if (!emulate_invalid_guest_state) {
2722 var.selector = var.base >> 4;
2723 var.base = var.base & 0xffff0;
2733 if (save->base & 0xf)
2734 printk_once(KERN_WARNING "kvm: segment base is not "
2735 "paragraph aligned when entering "
2736 "protected mode (seg=%d)", seg);
2739 vmcs_write16(sf->selector, var.selector);
2740 vmcs_writel(sf->base, var.base);
2741 vmcs_write32(sf->limit, var.limit);
2742 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
2745 static void enter_rmode(struct kvm_vcpu *vcpu)
2747 unsigned long flags;
2748 struct vcpu_vmx *vmx = to_vmx(vcpu);
2749 struct kvm_vmx *kvm_vmx = to_kvm_vmx(vcpu->kvm);
2751 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2752 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2753 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2754 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2755 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2756 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2757 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2759 vmx->rmode.vm86_active = 1;
2762 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
2763 * vcpu. Warn the user that an update is overdue.
2765 if (!kvm_vmx->tss_addr)
2766 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
2767 "called before entering vcpu\n");
2769 vmx_segment_cache_clear(vmx);
2771 vmcs_writel(GUEST_TR_BASE, kvm_vmx->tss_addr);
2772 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
2773 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2775 flags = vmcs_readl(GUEST_RFLAGS);
2776 vmx->rmode.save_rflags = flags;
2778 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2780 vmcs_writel(GUEST_RFLAGS, flags);
2781 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
2782 update_exception_bitmap(vcpu);
2784 fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2785 fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2786 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2787 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2788 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2789 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2791 kvm_mmu_reset_context(vcpu);
2794 void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
2796 struct vcpu_vmx *vmx = to_vmx(vcpu);
2797 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
2802 vcpu->arch.efer = efer;
2803 if (efer & EFER_LMA) {
2804 vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2807 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2809 msr->data = efer & ~EFER_LME;
2814 #ifdef CONFIG_X86_64
2816 static void enter_lmode(struct kvm_vcpu *vcpu)
2820 vmx_segment_cache_clear(to_vmx(vcpu));
2822 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
2823 if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
2824 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
2826 vmcs_write32(GUEST_TR_AR_BYTES,
2827 (guest_tr_ar & ~VMX_AR_TYPE_MASK)
2828 | VMX_AR_TYPE_BUSY_64_TSS);
2830 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
2833 static void exit_lmode(struct kvm_vcpu *vcpu)
2835 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2836 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
2841 static void vmx_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t addr)
2843 int vpid = to_vmx(vcpu)->vpid;
2845 if (!vpid_sync_vcpu_addr(vpid, addr))
2846 vpid_sync_context(vpid);
2849 * If VPIDs are not supported or enabled, then the above is a no-op.
2850 * But we don't really need a TLB flush in that case anyway, because
2851 * each VM entry/exit includes an implicit flush when VPID is 0.
2855 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
2857 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
2859 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
2860 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
2863 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
2865 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
2867 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
2868 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
2871 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
2873 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2875 if (!kvm_register_is_dirty(vcpu, VCPU_EXREG_PDPTR))
2878 if (is_pae_paging(vcpu)) {
2879 vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
2880 vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
2881 vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
2882 vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
2886 void ept_save_pdptrs(struct kvm_vcpu *vcpu)
2888 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2890 if (is_pae_paging(vcpu)) {
2891 mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
2892 mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
2893 mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
2894 mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
2897 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
2900 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
2902 struct kvm_vcpu *vcpu)
2904 struct vcpu_vmx *vmx = to_vmx(vcpu);
2906 if (!kvm_register_is_available(vcpu, VCPU_EXREG_CR3))
2907 vmx_cache_reg(vcpu, VCPU_EXREG_CR3);
2908 if (!(cr0 & X86_CR0_PG)) {
2909 /* From paging/starting to nonpaging */
2910 exec_controls_setbit(vmx, CPU_BASED_CR3_LOAD_EXITING |
2911 CPU_BASED_CR3_STORE_EXITING);
2912 vcpu->arch.cr0 = cr0;
2913 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2914 } else if (!is_paging(vcpu)) {
2915 /* From nonpaging to paging */
2916 exec_controls_clearbit(vmx, CPU_BASED_CR3_LOAD_EXITING |
2917 CPU_BASED_CR3_STORE_EXITING);
2918 vcpu->arch.cr0 = cr0;
2919 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2922 if (!(cr0 & X86_CR0_WP))
2923 *hw_cr0 &= ~X86_CR0_WP;
2926 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
2928 struct vcpu_vmx *vmx = to_vmx(vcpu);
2929 unsigned long hw_cr0;
2931 hw_cr0 = (cr0 & ~KVM_VM_CR0_ALWAYS_OFF);
2932 if (enable_unrestricted_guest)
2933 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
2935 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
2937 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
2940 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
2944 #ifdef CONFIG_X86_64
2945 if (vcpu->arch.efer & EFER_LME) {
2946 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
2948 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
2953 if (enable_ept && !enable_unrestricted_guest)
2954 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
2956 vmcs_writel(CR0_READ_SHADOW, cr0);
2957 vmcs_writel(GUEST_CR0, hw_cr0);
2958 vcpu->arch.cr0 = cr0;
2960 /* depends on vcpu->arch.cr0 to be set to a new value */
2961 vmx->emulation_required = emulation_required(vcpu);
2964 static int get_ept_level(struct kvm_vcpu *vcpu)
2966 if (is_guest_mode(vcpu) && nested_cpu_has_ept(get_vmcs12(vcpu)))
2967 return vmx_eptp_page_walk_level(nested_ept_get_eptp(vcpu));
2968 if (cpu_has_vmx_ept_5levels() && (cpuid_maxphyaddr(vcpu) > 48))
2973 u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa)
2975 u64 eptp = VMX_EPTP_MT_WB;
2977 eptp |= (get_ept_level(vcpu) == 5) ? VMX_EPTP_PWL_5 : VMX_EPTP_PWL_4;
2979 if (enable_ept_ad_bits &&
2980 (!is_guest_mode(vcpu) || nested_ept_ad_enabled(vcpu)))
2981 eptp |= VMX_EPTP_AD_ENABLE_BIT;
2982 eptp |= (root_hpa & PAGE_MASK);
2987 void vmx_load_mmu_pgd(struct kvm_vcpu *vcpu, unsigned long cr3)
2989 struct kvm *kvm = vcpu->kvm;
2990 bool update_guest_cr3 = true;
2991 unsigned long guest_cr3;
2996 eptp = construct_eptp(vcpu, cr3);
2997 vmcs_write64(EPT_POINTER, eptp);
2999 if (kvm_x86_ops.tlb_remote_flush) {
3000 spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
3001 to_vmx(vcpu)->ept_pointer = eptp;
3002 to_kvm_vmx(kvm)->ept_pointers_match
3003 = EPT_POINTERS_CHECK;
3004 spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
3007 /* Loading vmcs02.GUEST_CR3 is handled by nested VM-Enter. */
3008 if (is_guest_mode(vcpu))
3009 update_guest_cr3 = false;
3010 else if (!enable_unrestricted_guest && !is_paging(vcpu))
3011 guest_cr3 = to_kvm_vmx(kvm)->ept_identity_map_addr;
3012 else if (test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
3013 guest_cr3 = vcpu->arch.cr3;
3014 else /* vmcs01.GUEST_CR3 is already up-to-date. */
3015 update_guest_cr3 = false;
3016 ept_load_pdptrs(vcpu);
3019 if (update_guest_cr3)
3020 vmcs_writel(GUEST_CR3, guest_cr3);
3023 int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3025 struct vcpu_vmx *vmx = to_vmx(vcpu);
3027 * Pass through host's Machine Check Enable value to hw_cr4, which
3028 * is in force while we are in guest mode. Do not let guests control
3029 * this bit, even if host CR4.MCE == 0.
3031 unsigned long hw_cr4;
3033 hw_cr4 = (cr4_read_shadow() & X86_CR4_MCE) | (cr4 & ~X86_CR4_MCE);
3034 if (enable_unrestricted_guest)
3035 hw_cr4 |= KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST;
3036 else if (vmx->rmode.vm86_active)
3037 hw_cr4 |= KVM_RMODE_VM_CR4_ALWAYS_ON;
3039 hw_cr4 |= KVM_PMODE_VM_CR4_ALWAYS_ON;
3041 if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated()) {
3042 if (cr4 & X86_CR4_UMIP) {
3043 secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_DESC);
3044 hw_cr4 &= ~X86_CR4_UMIP;
3045 } else if (!is_guest_mode(vcpu) ||
3046 !nested_cpu_has2(get_vmcs12(vcpu), SECONDARY_EXEC_DESC)) {
3047 secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_DESC);
3051 if (cr4 & X86_CR4_VMXE) {
3053 * To use VMXON (and later other VMX instructions), a guest
3054 * must first be able to turn on cr4.VMXE (see handle_vmon()).
3055 * So basically the check on whether to allow nested VMX
3056 * is here. We operate under the default treatment of SMM,
3057 * so VMX cannot be enabled under SMM.
3059 if (!nested_vmx_allowed(vcpu) || is_smm(vcpu))
3063 if (vmx->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
3066 vcpu->arch.cr4 = cr4;
3068 if (!enable_unrestricted_guest) {
3070 if (!is_paging(vcpu)) {
3071 hw_cr4 &= ~X86_CR4_PAE;
3072 hw_cr4 |= X86_CR4_PSE;
3073 } else if (!(cr4 & X86_CR4_PAE)) {
3074 hw_cr4 &= ~X86_CR4_PAE;
3079 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
3080 * hardware. To emulate this behavior, SMEP/SMAP/PKU needs
3081 * to be manually disabled when guest switches to non-paging
3084 * If !enable_unrestricted_guest, the CPU is always running
3085 * with CR0.PG=1 and CR4 needs to be modified.
3086 * If enable_unrestricted_guest, the CPU automatically
3087 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
3089 if (!is_paging(vcpu))
3090 hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
3093 vmcs_writel(CR4_READ_SHADOW, cr4);
3094 vmcs_writel(GUEST_CR4, hw_cr4);
3098 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3100 struct vcpu_vmx *vmx = to_vmx(vcpu);
3103 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3104 *var = vmx->rmode.segs[seg];
3105 if (seg == VCPU_SREG_TR
3106 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3108 var->base = vmx_read_guest_seg_base(vmx, seg);
3109 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3112 var->base = vmx_read_guest_seg_base(vmx, seg);
3113 var->limit = vmx_read_guest_seg_limit(vmx, seg);
3114 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3115 ar = vmx_read_guest_seg_ar(vmx, seg);
3116 var->unusable = (ar >> 16) & 1;
3117 var->type = ar & 15;
3118 var->s = (ar >> 4) & 1;
3119 var->dpl = (ar >> 5) & 3;
3121 * Some userspaces do not preserve unusable property. Since usable
3122 * segment has to be present according to VMX spec we can use present
3123 * property to amend userspace bug by making unusable segment always
3124 * nonpresent. vmx_segment_access_rights() already marks nonpresent
3125 * segment as unusable.
3127 var->present = !var->unusable;
3128 var->avl = (ar >> 12) & 1;
3129 var->l = (ar >> 13) & 1;
3130 var->db = (ar >> 14) & 1;
3131 var->g = (ar >> 15) & 1;
3134 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3136 struct kvm_segment s;
3138 if (to_vmx(vcpu)->rmode.vm86_active) {
3139 vmx_get_segment(vcpu, &s, seg);
3142 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3145 int vmx_get_cpl(struct kvm_vcpu *vcpu)
3147 struct vcpu_vmx *vmx = to_vmx(vcpu);
3149 if (unlikely(vmx->rmode.vm86_active))
3152 int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
3153 return VMX_AR_DPL(ar);
3157 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3161 if (var->unusable || !var->present)
3164 ar = var->type & 15;
3165 ar |= (var->s & 1) << 4;
3166 ar |= (var->dpl & 3) << 5;
3167 ar |= (var->present & 1) << 7;
3168 ar |= (var->avl & 1) << 12;
3169 ar |= (var->l & 1) << 13;
3170 ar |= (var->db & 1) << 14;
3171 ar |= (var->g & 1) << 15;
3177 void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3179 struct vcpu_vmx *vmx = to_vmx(vcpu);
3180 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3182 vmx_segment_cache_clear(vmx);
3184 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3185 vmx->rmode.segs[seg] = *var;
3186 if (seg == VCPU_SREG_TR)
3187 vmcs_write16(sf->selector, var->selector);
3189 fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3193 vmcs_writel(sf->base, var->base);
3194 vmcs_write32(sf->limit, var->limit);
3195 vmcs_write16(sf->selector, var->selector);
3198 * Fix the "Accessed" bit in AR field of segment registers for older
3200 * IA32 arch specifies that at the time of processor reset the
3201 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3202 * is setting it to 0 in the userland code. This causes invalid guest
3203 * state vmexit when "unrestricted guest" mode is turned on.
3204 * Fix for this setup issue in cpu_reset is being pushed in the qemu
3205 * tree. Newer qemu binaries with that qemu fix would not need this
3208 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3209 var->type |= 0x1; /* Accessed */
3211 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3214 vmx->emulation_required = emulation_required(vcpu);
3217 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3219 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3221 *db = (ar >> 14) & 1;
3222 *l = (ar >> 13) & 1;
3225 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3227 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3228 dt->address = vmcs_readl(GUEST_IDTR_BASE);
3231 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3233 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3234 vmcs_writel(GUEST_IDTR_BASE, dt->address);
3237 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3239 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3240 dt->address = vmcs_readl(GUEST_GDTR_BASE);
3243 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3245 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3246 vmcs_writel(GUEST_GDTR_BASE, dt->address);
3249 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3251 struct kvm_segment var;
3254 vmx_get_segment(vcpu, &var, seg);
3256 if (seg == VCPU_SREG_CS)
3258 ar = vmx_segment_access_rights(&var);
3260 if (var.base != (var.selector << 4))
3262 if (var.limit != 0xffff)
3270 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3272 struct kvm_segment cs;
3273 unsigned int cs_rpl;
3275 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3276 cs_rpl = cs.selector & SEGMENT_RPL_MASK;
3280 if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
3284 if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
3285 if (cs.dpl > cs_rpl)
3288 if (cs.dpl != cs_rpl)
3294 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3298 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3300 struct kvm_segment ss;
3301 unsigned int ss_rpl;
3303 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3304 ss_rpl = ss.selector & SEGMENT_RPL_MASK;
3308 if (ss.type != 3 && ss.type != 7)
3312 if (ss.dpl != ss_rpl) /* DPL != RPL */
3320 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3322 struct kvm_segment var;
3325 vmx_get_segment(vcpu, &var, seg);
3326 rpl = var.selector & SEGMENT_RPL_MASK;
3334 if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
3335 if (var.dpl < rpl) /* DPL < RPL */
3339 /* TODO: Add other members to kvm_segment_field to allow checking for other access
3345 static bool tr_valid(struct kvm_vcpu *vcpu)
3347 struct kvm_segment tr;
3349 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3353 if (tr.selector & SEGMENT_TI_MASK) /* TI = 1 */
3355 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3363 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3365 struct kvm_segment ldtr;
3367 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3371 if (ldtr.selector & SEGMENT_TI_MASK) /* TI = 1 */
3381 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3383 struct kvm_segment cs, ss;
3385 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3386 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3388 return ((cs.selector & SEGMENT_RPL_MASK) ==
3389 (ss.selector & SEGMENT_RPL_MASK));
3393 * Check if guest state is valid. Returns true if valid, false if
3395 * We assume that registers are always usable
3397 static bool guest_state_valid(struct kvm_vcpu *vcpu)
3399 if (enable_unrestricted_guest)
3402 /* real mode guest state checks */
3403 if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
3404 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3406 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3408 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3410 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3412 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3414 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3417 /* protected mode guest state checks */
3418 if (!cs_ss_rpl_check(vcpu))
3420 if (!code_segment_valid(vcpu))
3422 if (!stack_segment_valid(vcpu))
3424 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3426 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3428 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3430 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3432 if (!tr_valid(vcpu))
3434 if (!ldtr_valid(vcpu))
3438 * - Add checks on RIP
3439 * - Add checks on RFLAGS
3445 static int init_rmode_tss(struct kvm *kvm)
3451 idx = srcu_read_lock(&kvm->srcu);
3452 fn = to_kvm_vmx(kvm)->tss_addr >> PAGE_SHIFT;
3453 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3456 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3457 r = kvm_write_guest_page(kvm, fn++, &data,
3458 TSS_IOPB_BASE_OFFSET, sizeof(u16));
3461 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3464 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3468 r = kvm_write_guest_page(kvm, fn, &data,
3469 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3472 srcu_read_unlock(&kvm->srcu, idx);
3476 static int init_rmode_identity_map(struct kvm *kvm)
3478 struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm);
3480 kvm_pfn_t identity_map_pfn;
3483 /* Protect kvm_vmx->ept_identity_pagetable_done. */
3484 mutex_lock(&kvm->slots_lock);
3486 if (likely(kvm_vmx->ept_identity_pagetable_done))
3489 if (!kvm_vmx->ept_identity_map_addr)
3490 kvm_vmx->ept_identity_map_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
3491 identity_map_pfn = kvm_vmx->ept_identity_map_addr >> PAGE_SHIFT;
3493 r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
3494 kvm_vmx->ept_identity_map_addr, PAGE_SIZE);
3498 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3501 /* Set up identity-mapping pagetable for EPT in real mode */
3502 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3503 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3504 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3505 r = kvm_write_guest_page(kvm, identity_map_pfn,
3506 &tmp, i * sizeof(tmp), sizeof(tmp));
3510 kvm_vmx->ept_identity_pagetable_done = true;
3513 mutex_unlock(&kvm->slots_lock);
3517 static void seg_setup(int seg)
3519 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3522 vmcs_write16(sf->selector, 0);
3523 vmcs_writel(sf->base, 0);
3524 vmcs_write32(sf->limit, 0xffff);
3526 if (seg == VCPU_SREG_CS)
3527 ar |= 0x08; /* code segment */
3529 vmcs_write32(sf->ar_bytes, ar);
3532 static int alloc_apic_access_page(struct kvm *kvm)
3537 mutex_lock(&kvm->slots_lock);
3538 if (kvm->arch.apic_access_page_done)
3540 r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
3541 APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
3545 page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
3546 if (is_error_page(page)) {
3552 * Do not pin the page in memory, so that memory hot-unplug
3553 * is able to migrate it.
3556 kvm->arch.apic_access_page_done = true;
3558 mutex_unlock(&kvm->slots_lock);
3562 int allocate_vpid(void)
3568 spin_lock(&vmx_vpid_lock);
3569 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3570 if (vpid < VMX_NR_VPIDS)
3571 __set_bit(vpid, vmx_vpid_bitmap);
3574 spin_unlock(&vmx_vpid_lock);
3578 void free_vpid(int vpid)
3580 if (!enable_vpid || vpid == 0)
3582 spin_lock(&vmx_vpid_lock);
3583 __clear_bit(vpid, vmx_vpid_bitmap);
3584 spin_unlock(&vmx_vpid_lock);
3587 static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
3590 int f = sizeof(unsigned long);
3592 if (!cpu_has_vmx_msr_bitmap())
3595 if (static_branch_unlikely(&enable_evmcs))
3596 evmcs_touch_msr_bitmap();
3599 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3600 * have the write-low and read-high bitmap offsets the wrong way round.
3601 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3603 if (msr <= 0x1fff) {
3604 if (type & MSR_TYPE_R)
3606 __clear_bit(msr, msr_bitmap + 0x000 / f);
3608 if (type & MSR_TYPE_W)
3610 __clear_bit(msr, msr_bitmap + 0x800 / f);
3612 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3614 if (type & MSR_TYPE_R)
3616 __clear_bit(msr, msr_bitmap + 0x400 / f);
3618 if (type & MSR_TYPE_W)
3620 __clear_bit(msr, msr_bitmap + 0xc00 / f);
3625 static __always_inline void vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
3628 int f = sizeof(unsigned long);
3630 if (!cpu_has_vmx_msr_bitmap())
3633 if (static_branch_unlikely(&enable_evmcs))
3634 evmcs_touch_msr_bitmap();
3637 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3638 * have the write-low and read-high bitmap offsets the wrong way round.
3639 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3641 if (msr <= 0x1fff) {
3642 if (type & MSR_TYPE_R)
3644 __set_bit(msr, msr_bitmap + 0x000 / f);
3646 if (type & MSR_TYPE_W)
3648 __set_bit(msr, msr_bitmap + 0x800 / f);
3650 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3652 if (type & MSR_TYPE_R)
3654 __set_bit(msr, msr_bitmap + 0x400 / f);
3656 if (type & MSR_TYPE_W)
3658 __set_bit(msr, msr_bitmap + 0xc00 / f);
3663 static __always_inline void vmx_set_intercept_for_msr(unsigned long *msr_bitmap,
3664 u32 msr, int type, bool value)
3667 vmx_enable_intercept_for_msr(msr_bitmap, msr, type);
3669 vmx_disable_intercept_for_msr(msr_bitmap, msr, type);
3672 static u8 vmx_msr_bitmap_mode(struct kvm_vcpu *vcpu)
3676 if (cpu_has_secondary_exec_ctrls() &&
3677 (secondary_exec_controls_get(to_vmx(vcpu)) &
3678 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
3679 mode |= MSR_BITMAP_MODE_X2APIC;
3680 if (enable_apicv && kvm_vcpu_apicv_active(vcpu))
3681 mode |= MSR_BITMAP_MODE_X2APIC_APICV;
3687 static void vmx_update_msr_bitmap_x2apic(unsigned long *msr_bitmap,
3692 for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
3693 unsigned word = msr / BITS_PER_LONG;
3694 msr_bitmap[word] = (mode & MSR_BITMAP_MODE_X2APIC_APICV) ? 0 : ~0;
3695 msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
3698 if (mode & MSR_BITMAP_MODE_X2APIC) {
3700 * TPR reads and writes can be virtualized even if virtual interrupt
3701 * delivery is not in use.
3703 vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TASKPRI), MSR_TYPE_RW);
3704 if (mode & MSR_BITMAP_MODE_X2APIC_APICV) {
3705 vmx_enable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TMCCT), MSR_TYPE_R);
3706 vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_EOI), MSR_TYPE_W);
3707 vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_SELF_IPI), MSR_TYPE_W);
3712 void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu)
3714 struct vcpu_vmx *vmx = to_vmx(vcpu);
3715 unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3716 u8 mode = vmx_msr_bitmap_mode(vcpu);
3717 u8 changed = mode ^ vmx->msr_bitmap_mode;
3722 if (changed & (MSR_BITMAP_MODE_X2APIC | MSR_BITMAP_MODE_X2APIC_APICV))
3723 vmx_update_msr_bitmap_x2apic(msr_bitmap, mode);
3725 vmx->msr_bitmap_mode = mode;
3728 void pt_update_intercept_for_msr(struct vcpu_vmx *vmx)
3730 unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3731 bool flag = !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
3734 vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_STATUS,
3736 vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_BASE,
3738 vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_MASK,
3740 vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_CR3_MATCH,
3742 for (i = 0; i < vmx->pt_desc.addr_range; i++) {
3743 vmx_set_intercept_for_msr(msr_bitmap,
3744 MSR_IA32_RTIT_ADDR0_A + i * 2, MSR_TYPE_RW, flag);
3745 vmx_set_intercept_for_msr(msr_bitmap,
3746 MSR_IA32_RTIT_ADDR0_B + i * 2, MSR_TYPE_RW, flag);
3750 static bool vmx_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
3752 struct vcpu_vmx *vmx = to_vmx(vcpu);
3757 if (WARN_ON_ONCE(!is_guest_mode(vcpu)) ||
3758 !nested_cpu_has_vid(get_vmcs12(vcpu)) ||
3759 WARN_ON_ONCE(!vmx->nested.virtual_apic_map.gfn))
3762 rvi = vmx_get_rvi();
3764 vapic_page = vmx->nested.virtual_apic_map.hva;
3765 vppr = *((u32 *)(vapic_page + APIC_PROCPRI));
3767 return ((rvi & 0xf0) > (vppr & 0xf0));
3770 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu,
3774 int pi_vec = nested ? POSTED_INTR_NESTED_VECTOR : POSTED_INTR_VECTOR;
3776 if (vcpu->mode == IN_GUEST_MODE) {
3778 * The vector of interrupt to be delivered to vcpu had
3779 * been set in PIR before this function.
3781 * Following cases will be reached in this block, and
3782 * we always send a notification event in all cases as
3785 * Case 1: vcpu keeps in non-root mode. Sending a
3786 * notification event posts the interrupt to vcpu.
3788 * Case 2: vcpu exits to root mode and is still
3789 * runnable. PIR will be synced to vIRR before the
3790 * next vcpu entry. Sending a notification event in
3791 * this case has no effect, as vcpu is not in root
3794 * Case 3: vcpu exits to root mode and is blocked.
3795 * vcpu_block() has already synced PIR to vIRR and
3796 * never blocks vcpu if vIRR is not cleared. Therefore,
3797 * a blocked vcpu here does not wait for any requested
3798 * interrupts in PIR, and sending a notification event
3799 * which has no effect is safe here.
3802 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu), pi_vec);
3809 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
3812 struct vcpu_vmx *vmx = to_vmx(vcpu);
3814 if (is_guest_mode(vcpu) &&
3815 vector == vmx->nested.posted_intr_nv) {
3817 * If a posted intr is not recognized by hardware,
3818 * we will accomplish it in the next vmentry.
3820 vmx->nested.pi_pending = true;
3821 kvm_make_request(KVM_REQ_EVENT, vcpu);
3822 /* the PIR and ON have been set by L1. */
3823 if (!kvm_vcpu_trigger_posted_interrupt(vcpu, true))
3824 kvm_vcpu_kick(vcpu);
3830 * Send interrupt to vcpu via posted interrupt way.
3831 * 1. If target vcpu is running(non-root mode), send posted interrupt
3832 * notification to vcpu and hardware will sync PIR to vIRR atomically.
3833 * 2. If target vcpu isn't running(root mode), kick it to pick up the
3834 * interrupt from PIR in next vmentry.
3836 static int vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
3838 struct vcpu_vmx *vmx = to_vmx(vcpu);
3841 r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
3845 if (!vcpu->arch.apicv_active)
3848 if (pi_test_and_set_pir(vector, &vmx->pi_desc))
3851 /* If a previous notification has sent the IPI, nothing to do. */
3852 if (pi_test_and_set_on(&vmx->pi_desc))
3855 if (!kvm_vcpu_trigger_posted_interrupt(vcpu, false))
3856 kvm_vcpu_kick(vcpu);
3862 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
3863 * will not change in the lifetime of the guest.
3864 * Note that host-state that does change is set elsewhere. E.g., host-state
3865 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
3867 void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
3871 unsigned long cr0, cr3, cr4;
3874 WARN_ON(cr0 & X86_CR0_TS);
3875 vmcs_writel(HOST_CR0, cr0); /* 22.2.3 */
3878 * Save the most likely value for this task's CR3 in the VMCS.
3879 * We can't use __get_current_cr3_fast() because we're not atomic.
3882 vmcs_writel(HOST_CR3, cr3); /* 22.2.3 FIXME: shadow tables */
3883 vmx->loaded_vmcs->host_state.cr3 = cr3;
3885 /* Save the most likely value for this task's CR4 in the VMCS. */
3886 cr4 = cr4_read_shadow();
3887 vmcs_writel(HOST_CR4, cr4); /* 22.2.3, 22.2.5 */
3888 vmx->loaded_vmcs->host_state.cr4 = cr4;
3890 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
3891 #ifdef CONFIG_X86_64
3893 * Load null selectors, so we can avoid reloading them in
3894 * vmx_prepare_switch_to_host(), in case userspace uses
3895 * the null selectors too (the expected case).
3897 vmcs_write16(HOST_DS_SELECTOR, 0);
3898 vmcs_write16(HOST_ES_SELECTOR, 0);
3900 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
3901 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
3903 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
3904 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
3906 vmcs_writel(HOST_IDTR_BASE, host_idt_base); /* 22.2.4 */
3908 vmcs_writel(HOST_RIP, (unsigned long)vmx_vmexit); /* 22.2.5 */
3910 rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
3911 vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
3912 rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
3913 vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */
3915 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
3916 rdmsr(MSR_IA32_CR_PAT, low32, high32);
3917 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
3920 if (cpu_has_load_ia32_efer())
3921 vmcs_write64(HOST_IA32_EFER, host_efer);
3924 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
3926 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
3928 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
3929 if (is_guest_mode(&vmx->vcpu))
3930 vmx->vcpu.arch.cr4_guest_owned_bits &=
3931 ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
3932 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
3935 u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
3937 u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
3939 if (!kvm_vcpu_apicv_active(&vmx->vcpu))
3940 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
3943 pin_based_exec_ctrl &= ~PIN_BASED_VIRTUAL_NMIS;
3945 if (!enable_preemption_timer)
3946 pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
3948 return pin_based_exec_ctrl;
3951 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
3953 struct vcpu_vmx *vmx = to_vmx(vcpu);
3955 pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
3956 if (cpu_has_secondary_exec_ctrls()) {
3957 if (kvm_vcpu_apicv_active(vcpu))
3958 secondary_exec_controls_setbit(vmx,
3959 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3960 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3962 secondary_exec_controls_clearbit(vmx,
3963 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3964 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3967 if (cpu_has_vmx_msr_bitmap())
3968 vmx_update_msr_bitmap(vcpu);
3971 u32 vmx_exec_control(struct vcpu_vmx *vmx)
3973 u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
3975 if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
3976 exec_control &= ~CPU_BASED_MOV_DR_EXITING;
3978 if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
3979 exec_control &= ~CPU_BASED_TPR_SHADOW;
3980 #ifdef CONFIG_X86_64
3981 exec_control |= CPU_BASED_CR8_STORE_EXITING |
3982 CPU_BASED_CR8_LOAD_EXITING;
3986 exec_control |= CPU_BASED_CR3_STORE_EXITING |
3987 CPU_BASED_CR3_LOAD_EXITING |
3988 CPU_BASED_INVLPG_EXITING;
3989 if (kvm_mwait_in_guest(vmx->vcpu.kvm))
3990 exec_control &= ~(CPU_BASED_MWAIT_EXITING |
3991 CPU_BASED_MONITOR_EXITING);
3992 if (kvm_hlt_in_guest(vmx->vcpu.kvm))
3993 exec_control &= ~CPU_BASED_HLT_EXITING;
3994 return exec_control;
3998 static void vmx_compute_secondary_exec_control(struct vcpu_vmx *vmx)
4000 struct kvm_vcpu *vcpu = &vmx->vcpu;
4002 u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
4004 if (vmx_pt_mode_is_system())
4005 exec_control &= ~(SECONDARY_EXEC_PT_USE_GPA | SECONDARY_EXEC_PT_CONCEAL_VMX);
4006 if (!cpu_need_virtualize_apic_accesses(vcpu))
4007 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
4009 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
4011 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
4012 enable_unrestricted_guest = 0;
4014 if (!enable_unrestricted_guest)
4015 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4016 if (kvm_pause_in_guest(vmx->vcpu.kvm))
4017 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
4018 if (!kvm_vcpu_apicv_active(vcpu))
4019 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
4020 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4021 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
4023 /* SECONDARY_EXEC_DESC is enabled/disabled on writes to CR4.UMIP,
4024 * in vmx_set_cr4. */
4025 exec_control &= ~SECONDARY_EXEC_DESC;
4027 /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4029 We can NOT enable shadow_vmcs here because we don't have yet
4032 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4035 exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
4037 if (vmx_xsaves_supported()) {
4038 /* Exposing XSAVES only when XSAVE is exposed */
4039 bool xsaves_enabled =
4040 boot_cpu_has(X86_FEATURE_XSAVE) &&
4041 guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
4042 guest_cpuid_has(vcpu, X86_FEATURE_XSAVES);
4044 vcpu->arch.xsaves_enabled = xsaves_enabled;
4046 if (!xsaves_enabled)
4047 exec_control &= ~SECONDARY_EXEC_XSAVES;
4051 vmx->nested.msrs.secondary_ctls_high |=
4052 SECONDARY_EXEC_XSAVES;
4054 vmx->nested.msrs.secondary_ctls_high &=
4055 ~SECONDARY_EXEC_XSAVES;
4059 if (cpu_has_vmx_rdtscp()) {
4060 bool rdtscp_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP);
4061 if (!rdtscp_enabled)
4062 exec_control &= ~SECONDARY_EXEC_RDTSCP;
4066 vmx->nested.msrs.secondary_ctls_high |=
4067 SECONDARY_EXEC_RDTSCP;
4069 vmx->nested.msrs.secondary_ctls_high &=
4070 ~SECONDARY_EXEC_RDTSCP;
4074 if (cpu_has_vmx_invpcid()) {
4075 /* Exposing INVPCID only when PCID is exposed */
4076 bool invpcid_enabled =
4077 guest_cpuid_has(vcpu, X86_FEATURE_INVPCID) &&
4078 guest_cpuid_has(vcpu, X86_FEATURE_PCID);
4080 if (!invpcid_enabled) {
4081 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
4082 guest_cpuid_clear(vcpu, X86_FEATURE_INVPCID);
4086 if (invpcid_enabled)
4087 vmx->nested.msrs.secondary_ctls_high |=
4088 SECONDARY_EXEC_ENABLE_INVPCID;
4090 vmx->nested.msrs.secondary_ctls_high &=
4091 ~SECONDARY_EXEC_ENABLE_INVPCID;
4095 if (vmx_rdrand_supported()) {
4096 bool rdrand_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDRAND);
4098 exec_control &= ~SECONDARY_EXEC_RDRAND_EXITING;
4102 vmx->nested.msrs.secondary_ctls_high |=
4103 SECONDARY_EXEC_RDRAND_EXITING;
4105 vmx->nested.msrs.secondary_ctls_high &=
4106 ~SECONDARY_EXEC_RDRAND_EXITING;
4110 if (vmx_rdseed_supported()) {
4111 bool rdseed_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDSEED);
4113 exec_control &= ~SECONDARY_EXEC_RDSEED_EXITING;
4117 vmx->nested.msrs.secondary_ctls_high |=
4118 SECONDARY_EXEC_RDSEED_EXITING;
4120 vmx->nested.msrs.secondary_ctls_high &=
4121 ~SECONDARY_EXEC_RDSEED_EXITING;
4125 if (vmx_waitpkg_supported()) {
4126 bool waitpkg_enabled =
4127 guest_cpuid_has(vcpu, X86_FEATURE_WAITPKG);
4129 if (!waitpkg_enabled)
4130 exec_control &= ~SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4133 if (waitpkg_enabled)
4134 vmx->nested.msrs.secondary_ctls_high |=
4135 SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4137 vmx->nested.msrs.secondary_ctls_high &=
4138 ~SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4142 vmx->secondary_exec_control = exec_control;
4145 static void ept_set_mmio_spte_mask(void)
4148 * EPT Misconfigurations can be generated if the value of bits 2:0
4149 * of an EPT paging-structure entry is 110b (write/execute).
4151 kvm_mmu_set_mmio_spte_mask(VMX_EPT_RWX_MASK,
4152 VMX_EPT_MISCONFIG_WX_VALUE, 0);
4155 #define VMX_XSS_EXIT_BITMAP 0
4158 * Noting that the initialization of Guest-state Area of VMCS is in
4161 static void init_vmcs(struct vcpu_vmx *vmx)
4164 nested_vmx_set_vmcs_shadowing_bitmap();
4166 if (cpu_has_vmx_msr_bitmap())
4167 vmcs_write64(MSR_BITMAP, __pa(vmx->vmcs01.msr_bitmap));
4169 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4172 pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
4174 exec_controls_set(vmx, vmx_exec_control(vmx));
4176 if (cpu_has_secondary_exec_ctrls()) {
4177 vmx_compute_secondary_exec_control(vmx);
4178 secondary_exec_controls_set(vmx, vmx->secondary_exec_control);
4181 if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
4182 vmcs_write64(EOI_EXIT_BITMAP0, 0);
4183 vmcs_write64(EOI_EXIT_BITMAP1, 0);
4184 vmcs_write64(EOI_EXIT_BITMAP2, 0);
4185 vmcs_write64(EOI_EXIT_BITMAP3, 0);
4187 vmcs_write16(GUEST_INTR_STATUS, 0);
4189 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4190 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4193 if (!kvm_pause_in_guest(vmx->vcpu.kvm)) {
4194 vmcs_write32(PLE_GAP, ple_gap);
4195 vmx->ple_window = ple_window;
4196 vmx->ple_window_dirty = true;
4199 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4200 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4201 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
4203 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
4204 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
4205 vmx_set_constant_host_state(vmx);
4206 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4207 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4209 if (cpu_has_vmx_vmfunc())
4210 vmcs_write64(VM_FUNCTION_CONTROL, 0);
4212 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4213 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4214 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
4215 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4216 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
4218 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
4219 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
4221 vm_exit_controls_set(vmx, vmx_vmexit_ctrl());
4223 /* 22.2.1, 20.8.1 */
4224 vm_entry_controls_set(vmx, vmx_vmentry_ctrl());
4226 vmx->vcpu.arch.cr0_guest_owned_bits = X86_CR0_TS;
4227 vmcs_writel(CR0_GUEST_HOST_MASK, ~X86_CR0_TS);
4229 set_cr4_guest_host_mask(vmx);
4232 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4234 if (vmx_xsaves_supported())
4235 vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
4238 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
4239 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
4242 if (cpu_has_vmx_encls_vmexit())
4243 vmcs_write64(ENCLS_EXITING_BITMAP, -1ull);
4245 if (vmx_pt_mode_is_host_guest()) {
4246 memset(&vmx->pt_desc, 0, sizeof(vmx->pt_desc));
4247 /* Bit[6~0] are forced to 1, writes are ignored. */
4248 vmx->pt_desc.guest.output_mask = 0x7F;
4249 vmcs_write64(GUEST_IA32_RTIT_CTL, 0);
4253 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
4255 struct vcpu_vmx *vmx = to_vmx(vcpu);
4256 struct msr_data apic_base_msr;
4259 vmx->rmode.vm86_active = 0;
4262 vmx->msr_ia32_umwait_control = 0;
4264 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
4265 vmx->hv_deadline_tsc = -1;
4266 kvm_set_cr8(vcpu, 0);
4269 apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
4270 MSR_IA32_APICBASE_ENABLE;
4271 if (kvm_vcpu_is_reset_bsp(vcpu))
4272 apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
4273 apic_base_msr.host_initiated = true;
4274 kvm_set_apic_base(vcpu, &apic_base_msr);
4277 vmx_segment_cache_clear(vmx);
4279 seg_setup(VCPU_SREG_CS);
4280 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4281 vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
4283 seg_setup(VCPU_SREG_DS);
4284 seg_setup(VCPU_SREG_ES);
4285 seg_setup(VCPU_SREG_FS);
4286 seg_setup(VCPU_SREG_GS);
4287 seg_setup(VCPU_SREG_SS);
4289 vmcs_write16(GUEST_TR_SELECTOR, 0);
4290 vmcs_writel(GUEST_TR_BASE, 0);
4291 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4292 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4294 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4295 vmcs_writel(GUEST_LDTR_BASE, 0);
4296 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4297 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4300 vmcs_write32(GUEST_SYSENTER_CS, 0);
4301 vmcs_writel(GUEST_SYSENTER_ESP, 0);
4302 vmcs_writel(GUEST_SYSENTER_EIP, 0);
4303 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4306 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
4307 kvm_rip_write(vcpu, 0xfff0);
4309 vmcs_writel(GUEST_GDTR_BASE, 0);
4310 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4312 vmcs_writel(GUEST_IDTR_BASE, 0);
4313 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4315 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4316 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4317 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4318 if (kvm_mpx_supported())
4319 vmcs_write64(GUEST_BNDCFGS, 0);
4323 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
4325 if (cpu_has_vmx_tpr_shadow() && !init_event) {
4326 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4327 if (cpu_need_tpr_shadow(vcpu))
4328 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4329 __pa(vcpu->arch.apic->regs));
4330 vmcs_write32(TPR_THRESHOLD, 0);
4333 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4335 cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4336 vmx->vcpu.arch.cr0 = cr0;
4337 vmx_set_cr0(vcpu, cr0); /* enter rmode */
4338 vmx_set_cr4(vcpu, 0);
4339 vmx_set_efer(vcpu, 0);
4341 update_exception_bitmap(vcpu);
4343 vpid_sync_context(vmx->vpid);
4345 vmx_clear_hlt(vcpu);
4348 static void enable_irq_window(struct kvm_vcpu *vcpu)
4350 exec_controls_setbit(to_vmx(vcpu), CPU_BASED_INTR_WINDOW_EXITING);
4353 static void enable_nmi_window(struct kvm_vcpu *vcpu)
4356 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4357 enable_irq_window(vcpu);
4361 exec_controls_setbit(to_vmx(vcpu), CPU_BASED_NMI_WINDOW_EXITING);
4364 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4366 struct vcpu_vmx *vmx = to_vmx(vcpu);
4368 int irq = vcpu->arch.interrupt.nr;
4370 trace_kvm_inj_virq(irq);
4372 ++vcpu->stat.irq_injections;
4373 if (vmx->rmode.vm86_active) {
4375 if (vcpu->arch.interrupt.soft)
4376 inc_eip = vcpu->arch.event_exit_inst_len;
4377 kvm_inject_realmode_interrupt(vcpu, irq, inc_eip);
4380 intr = irq | INTR_INFO_VALID_MASK;
4381 if (vcpu->arch.interrupt.soft) {
4382 intr |= INTR_TYPE_SOFT_INTR;
4383 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4384 vmx->vcpu.arch.event_exit_inst_len);
4386 intr |= INTR_TYPE_EXT_INTR;
4387 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4389 vmx_clear_hlt(vcpu);
4392 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4394 struct vcpu_vmx *vmx = to_vmx(vcpu);
4398 * Tracking the NMI-blocked state in software is built upon
4399 * finding the next open IRQ window. This, in turn, depends on
4400 * well-behaving guests: They have to keep IRQs disabled at
4401 * least as long as the NMI handler runs. Otherwise we may
4402 * cause NMI nesting, maybe breaking the guest. But as this is
4403 * highly unlikely, we can live with the residual risk.
4405 vmx->loaded_vmcs->soft_vnmi_blocked = 1;
4406 vmx->loaded_vmcs->vnmi_blocked_time = 0;
4409 ++vcpu->stat.nmi_injections;
4410 vmx->loaded_vmcs->nmi_known_unmasked = false;
4412 if (vmx->rmode.vm86_active) {
4413 kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0);
4417 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4418 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4420 vmx_clear_hlt(vcpu);
4423 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4425 struct vcpu_vmx *vmx = to_vmx(vcpu);
4429 return vmx->loaded_vmcs->soft_vnmi_blocked;
4430 if (vmx->loaded_vmcs->nmi_known_unmasked)
4432 masked = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4433 vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4437 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4439 struct vcpu_vmx *vmx = to_vmx(vcpu);
4442 if (vmx->loaded_vmcs->soft_vnmi_blocked != masked) {
4443 vmx->loaded_vmcs->soft_vnmi_blocked = masked;
4444 vmx->loaded_vmcs->vnmi_blocked_time = 0;
4447 vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4449 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4450 GUEST_INTR_STATE_NMI);
4452 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4453 GUEST_INTR_STATE_NMI);
4457 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
4459 if (to_vmx(vcpu)->nested.nested_run_pending)
4463 to_vmx(vcpu)->loaded_vmcs->soft_vnmi_blocked)
4466 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4467 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
4468 | GUEST_INTR_STATE_NMI));
4471 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
4473 if (to_vmx(vcpu)->nested.nested_run_pending)
4476 if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu))
4479 return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
4480 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4481 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4484 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4488 if (enable_unrestricted_guest)
4491 mutex_lock(&kvm->slots_lock);
4492 ret = __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
4494 mutex_unlock(&kvm->slots_lock);
4498 to_kvm_vmx(kvm)->tss_addr = addr;
4499 return init_rmode_tss(kvm);
4502 static int vmx_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
4504 to_kvm_vmx(kvm)->ept_identity_map_addr = ident_addr;
4508 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
4513 * Update instruction length as we may reinject the exception
4514 * from user space while in guest debugging mode.
4516 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4517 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4518 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4522 if (vcpu->guest_debug &
4523 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
4539 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4540 int vec, u32 err_code)
4543 * Instruction with address size override prefix opcode 0x67
4544 * Cause the #SS fault with 0 error code in VM86 mode.
4546 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
4547 if (kvm_emulate_instruction(vcpu, 0)) {
4548 if (vcpu->arch.halt_request) {
4549 vcpu->arch.halt_request = 0;
4550 return kvm_vcpu_halt(vcpu);
4558 * Forward all other exceptions that are valid in real mode.
4559 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4560 * the required debugging infrastructure rework.
4562 kvm_queue_exception(vcpu, vec);
4567 * Trigger machine check on the host. We assume all the MSRs are already set up
4568 * by the CPU and that we still run on the same CPU as the MCE occurred on.
4569 * We pass a fake environment to the machine check handler because we want
4570 * the guest to be always treated like user space, no matter what context
4571 * it used internally.
4573 static void kvm_machine_check(void)
4575 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
4576 struct pt_regs regs = {
4577 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
4578 .flags = X86_EFLAGS_IF,
4581 do_machine_check(®s, 0);
4585 static int handle_machine_check(struct kvm_vcpu *vcpu)
4587 /* handled by vmx_vcpu_run() */
4591 static int handle_exception_nmi(struct kvm_vcpu *vcpu)
4593 struct vcpu_vmx *vmx = to_vmx(vcpu);
4594 struct kvm_run *kvm_run = vcpu->run;
4595 u32 intr_info, ex_no, error_code;
4596 unsigned long cr2, rip, dr6;
4599 vect_info = vmx->idt_vectoring_info;
4600 intr_info = vmx->exit_intr_info;
4602 if (is_machine_check(intr_info) || is_nmi(intr_info))
4603 return 1; /* handled by handle_exception_nmi_irqoff() */
4605 if (is_invalid_opcode(intr_info))
4606 return handle_ud(vcpu);
4609 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4610 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4612 if (!vmx->rmode.vm86_active && is_gp_fault(intr_info)) {
4613 WARN_ON_ONCE(!enable_vmware_backdoor);
4616 * VMware backdoor emulation on #GP interception only handles
4617 * IN{S}, OUT{S}, and RDPMC, none of which generate a non-zero
4618 * error code on #GP.
4621 kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
4624 return kvm_emulate_instruction(vcpu, EMULTYPE_VMWARE_GP);
4628 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
4629 * MMIO, it is better to report an internal error.
4630 * See the comments in vmx_handle_exit.
4632 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4633 !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
4634 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4635 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4636 vcpu->run->internal.ndata = 3;
4637 vcpu->run->internal.data[0] = vect_info;
4638 vcpu->run->internal.data[1] = intr_info;
4639 vcpu->run->internal.data[2] = error_code;
4643 if (is_page_fault(intr_info)) {
4644 cr2 = vmcs_readl(EXIT_QUALIFICATION);
4645 /* EPT won't cause page fault directly */
4646 WARN_ON_ONCE(!vcpu->arch.apf.host_apf_reason && enable_ept);
4647 return kvm_handle_page_fault(vcpu, error_code, cr2, NULL, 0);
4650 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4652 if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
4653 return handle_rmode_exception(vcpu, ex_no, error_code);
4657 kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
4660 dr6 = vmcs_readl(EXIT_QUALIFICATION);
4661 if (!(vcpu->guest_debug &
4662 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4663 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
4664 vcpu->arch.dr6 |= dr6 | DR6_RTM;
4665 if (is_icebp(intr_info))
4666 WARN_ON(!skip_emulated_instruction(vcpu));
4668 kvm_queue_exception(vcpu, DB_VECTOR);
4671 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
4672 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4676 * Update instruction length as we may reinject #BP from
4677 * user space while in guest debugging mode. Reading it for
4678 * #DB as well causes no harm, it is not used in that case.
4680 vmx->vcpu.arch.event_exit_inst_len =
4681 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4682 kvm_run->exit_reason = KVM_EXIT_DEBUG;
4683 rip = kvm_rip_read(vcpu);
4684 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4685 kvm_run->debug.arch.exception = ex_no;
4688 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4689 kvm_run->ex.exception = ex_no;
4690 kvm_run->ex.error_code = error_code;
4696 static __always_inline int handle_external_interrupt(struct kvm_vcpu *vcpu)
4698 ++vcpu->stat.irq_exits;
4702 static int handle_triple_fault(struct kvm_vcpu *vcpu)
4704 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4705 vcpu->mmio_needed = 0;
4709 static int handle_io(struct kvm_vcpu *vcpu)
4711 unsigned long exit_qualification;
4712 int size, in, string;
4715 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4716 string = (exit_qualification & 16) != 0;
4718 ++vcpu->stat.io_exits;
4721 return kvm_emulate_instruction(vcpu, 0);
4723 port = exit_qualification >> 16;
4724 size = (exit_qualification & 7) + 1;
4725 in = (exit_qualification & 8) != 0;
4727 return kvm_fast_pio(vcpu, size, port, in);
4731 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4734 * Patch in the VMCALL instruction:
4736 hypercall[0] = 0x0f;
4737 hypercall[1] = 0x01;
4738 hypercall[2] = 0xc1;
4741 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
4742 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
4744 if (is_guest_mode(vcpu)) {
4745 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4746 unsigned long orig_val = val;
4749 * We get here when L2 changed cr0 in a way that did not change
4750 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
4751 * but did change L0 shadowed bits. So we first calculate the
4752 * effective cr0 value that L1 would like to write into the
4753 * hardware. It consists of the L2-owned bits from the new
4754 * value combined with the L1-owned bits from L1's guest_cr0.
4756 val = (val & ~vmcs12->cr0_guest_host_mask) |
4757 (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
4759 if (!nested_guest_cr0_valid(vcpu, val))
4762 if (kvm_set_cr0(vcpu, val))
4764 vmcs_writel(CR0_READ_SHADOW, orig_val);
4767 if (to_vmx(vcpu)->nested.vmxon &&
4768 !nested_host_cr0_valid(vcpu, val))
4771 return kvm_set_cr0(vcpu, val);
4775 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
4777 if (is_guest_mode(vcpu)) {
4778 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4779 unsigned long orig_val = val;
4781 /* analogously to handle_set_cr0 */
4782 val = (val & ~vmcs12->cr4_guest_host_mask) |
4783 (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
4784 if (kvm_set_cr4(vcpu, val))
4786 vmcs_writel(CR4_READ_SHADOW, orig_val);
4789 return kvm_set_cr4(vcpu, val);
4792 static int handle_desc(struct kvm_vcpu *vcpu)
4794 WARN_ON(!(vcpu->arch.cr4 & X86_CR4_UMIP));
4795 return kvm_emulate_instruction(vcpu, 0);
4798 static int handle_cr(struct kvm_vcpu *vcpu)
4800 unsigned long exit_qualification, val;
4806 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4807 cr = exit_qualification & 15;
4808 reg = (exit_qualification >> 8) & 15;
4809 switch ((exit_qualification >> 4) & 3) {
4810 case 0: /* mov to cr */
4811 val = kvm_register_readl(vcpu, reg);
4812 trace_kvm_cr_write(cr, val);
4815 err = handle_set_cr0(vcpu, val);
4816 return kvm_complete_insn_gp(vcpu, err);
4818 WARN_ON_ONCE(enable_unrestricted_guest);
4819 err = kvm_set_cr3(vcpu, val);
4820 return kvm_complete_insn_gp(vcpu, err);
4822 err = handle_set_cr4(vcpu, val);
4823 return kvm_complete_insn_gp(vcpu, err);
4825 u8 cr8_prev = kvm_get_cr8(vcpu);
4827 err = kvm_set_cr8(vcpu, cr8);
4828 ret = kvm_complete_insn_gp(vcpu, err);
4829 if (lapic_in_kernel(vcpu))
4831 if (cr8_prev <= cr8)
4834 * TODO: we might be squashing a
4835 * KVM_GUESTDBG_SINGLESTEP-triggered
4836 * KVM_EXIT_DEBUG here.
4838 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
4844 WARN_ONCE(1, "Guest should always own CR0.TS");
4845 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4846 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
4847 return kvm_skip_emulated_instruction(vcpu);
4848 case 1: /*mov from cr*/
4851 WARN_ON_ONCE(enable_unrestricted_guest);
4852 val = kvm_read_cr3(vcpu);
4853 kvm_register_write(vcpu, reg, val);
4854 trace_kvm_cr_read(cr, val);
4855 return kvm_skip_emulated_instruction(vcpu);
4857 val = kvm_get_cr8(vcpu);
4858 kvm_register_write(vcpu, reg, val);
4859 trace_kvm_cr_read(cr, val);
4860 return kvm_skip_emulated_instruction(vcpu);
4864 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
4865 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
4866 kvm_lmsw(vcpu, val);
4868 return kvm_skip_emulated_instruction(vcpu);
4872 vcpu->run->exit_reason = 0;
4873 vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
4874 (int)(exit_qualification >> 4) & 3, cr);
4878 static int handle_dr(struct kvm_vcpu *vcpu)
4880 unsigned long exit_qualification;
4883 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4884 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
4886 /* First, if DR does not exist, trigger UD */
4887 if (!kvm_require_dr(vcpu, dr))
4890 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
4891 if (!kvm_require_cpl(vcpu, 0))
4893 dr7 = vmcs_readl(GUEST_DR7);
4896 * As the vm-exit takes precedence over the debug trap, we
4897 * need to emulate the latter, either for the host or the
4898 * guest debugging itself.
4900 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
4901 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
4902 vcpu->run->debug.arch.dr7 = dr7;
4903 vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
4904 vcpu->run->debug.arch.exception = DB_VECTOR;
4905 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
4908 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
4909 vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
4910 kvm_queue_exception(vcpu, DB_VECTOR);
4915 if (vcpu->guest_debug == 0) {
4916 exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
4919 * No more DR vmexits; force a reload of the debug registers
4920 * and reenter on this instruction. The next vmexit will
4921 * retrieve the full state of the debug registers.
4923 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
4927 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
4928 if (exit_qualification & TYPE_MOV_FROM_DR) {
4931 if (kvm_get_dr(vcpu, dr, &val))
4933 kvm_register_write(vcpu, reg, val);
4935 if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
4938 return kvm_skip_emulated_instruction(vcpu);
4941 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
4943 return vcpu->arch.dr6;
4946 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
4950 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
4952 get_debugreg(vcpu->arch.db[0], 0);
4953 get_debugreg(vcpu->arch.db[1], 1);
4954 get_debugreg(vcpu->arch.db[2], 2);
4955 get_debugreg(vcpu->arch.db[3], 3);
4956 get_debugreg(vcpu->arch.dr6, 6);
4957 vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
4959 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
4960 exec_controls_setbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
4963 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
4965 vmcs_writel(GUEST_DR7, val);
4968 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
4970 kvm_apic_update_ppr(vcpu);
4974 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
4976 exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_INTR_WINDOW_EXITING);
4978 kvm_make_request(KVM_REQ_EVENT, vcpu);
4980 ++vcpu->stat.irq_window_exits;
4984 static int handle_vmcall(struct kvm_vcpu *vcpu)
4986 return kvm_emulate_hypercall(vcpu);
4989 static int handle_invd(struct kvm_vcpu *vcpu)
4991 return kvm_emulate_instruction(vcpu, 0);
4994 static int handle_invlpg(struct kvm_vcpu *vcpu)
4996 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4998 kvm_mmu_invlpg(vcpu, exit_qualification);
4999 return kvm_skip_emulated_instruction(vcpu);
5002 static int handle_rdpmc(struct kvm_vcpu *vcpu)
5006 err = kvm_rdpmc(vcpu);
5007 return kvm_complete_insn_gp(vcpu, err);
5010 static int handle_wbinvd(struct kvm_vcpu *vcpu)
5012 return kvm_emulate_wbinvd(vcpu);
5015 static int handle_xsetbv(struct kvm_vcpu *vcpu)
5017 u64 new_bv = kvm_read_edx_eax(vcpu);
5018 u32 index = kvm_rcx_read(vcpu);
5020 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
5021 return kvm_skip_emulated_instruction(vcpu);
5025 static int handle_apic_access(struct kvm_vcpu *vcpu)
5027 if (likely(fasteoi)) {
5028 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5029 int access_type, offset;
5031 access_type = exit_qualification & APIC_ACCESS_TYPE;
5032 offset = exit_qualification & APIC_ACCESS_OFFSET;
5034 * Sane guest uses MOV to write EOI, with written value
5035 * not cared. So make a short-circuit here by avoiding
5036 * heavy instruction emulation.
5038 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5039 (offset == APIC_EOI)) {
5040 kvm_lapic_set_eoi(vcpu);
5041 return kvm_skip_emulated_instruction(vcpu);
5044 return kvm_emulate_instruction(vcpu, 0);
5047 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5049 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5050 int vector = exit_qualification & 0xff;
5052 /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5053 kvm_apic_set_eoi_accelerated(vcpu, vector);
5057 static int handle_apic_write(struct kvm_vcpu *vcpu)
5059 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5060 u32 offset = exit_qualification & 0xfff;
5062 /* APIC-write VM exit is trap-like and thus no need to adjust IP */
5063 kvm_apic_write_nodecode(vcpu, offset);
5067 static int handle_task_switch(struct kvm_vcpu *vcpu)
5069 struct vcpu_vmx *vmx = to_vmx(vcpu);
5070 unsigned long exit_qualification;
5071 bool has_error_code = false;
5074 int reason, type, idt_v, idt_index;
5076 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5077 idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5078 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5080 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5082 reason = (u32)exit_qualification >> 30;
5083 if (reason == TASK_SWITCH_GATE && idt_v) {
5085 case INTR_TYPE_NMI_INTR:
5086 vcpu->arch.nmi_injected = false;
5087 vmx_set_nmi_mask(vcpu, true);
5089 case INTR_TYPE_EXT_INTR:
5090 case INTR_TYPE_SOFT_INTR:
5091 kvm_clear_interrupt_queue(vcpu);
5093 case INTR_TYPE_HARD_EXCEPTION:
5094 if (vmx->idt_vectoring_info &
5095 VECTORING_INFO_DELIVER_CODE_MASK) {
5096 has_error_code = true;
5098 vmcs_read32(IDT_VECTORING_ERROR_CODE);
5101 case INTR_TYPE_SOFT_EXCEPTION:
5102 kvm_clear_exception_queue(vcpu);
5108 tss_selector = exit_qualification;
5110 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5111 type != INTR_TYPE_EXT_INTR &&
5112 type != INTR_TYPE_NMI_INTR))
5113 WARN_ON(!skip_emulated_instruction(vcpu));
5116 * TODO: What about debug traps on tss switch?
5117 * Are we supposed to inject them and update dr6?
5119 return kvm_task_switch(vcpu, tss_selector,
5120 type == INTR_TYPE_SOFT_INTR ? idt_index : -1,
5121 reason, has_error_code, error_code);
5124 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5126 unsigned long exit_qualification;
5130 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5133 * EPT violation happened while executing iret from NMI,
5134 * "blocked by NMI" bit has to be set before next VM entry.
5135 * There are errata that may cause this bit to not be set:
5138 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5140 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
5141 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5143 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5144 trace_kvm_page_fault(gpa, exit_qualification);
5146 /* Is it a read fault? */
5147 error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
5148 ? PFERR_USER_MASK : 0;
5149 /* Is it a write fault? */
5150 error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
5151 ? PFERR_WRITE_MASK : 0;
5152 /* Is it a fetch fault? */
5153 error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
5154 ? PFERR_FETCH_MASK : 0;
5155 /* ept page table entry is present? */
5156 error_code |= (exit_qualification &
5157 (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE |
5158 EPT_VIOLATION_EXECUTABLE))
5159 ? PFERR_PRESENT_MASK : 0;
5161 error_code |= (exit_qualification & 0x100) != 0 ?
5162 PFERR_GUEST_FINAL_MASK : PFERR_GUEST_PAGE_MASK;
5164 vcpu->arch.exit_qualification = exit_qualification;
5165 return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5168 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5173 * A nested guest cannot optimize MMIO vmexits, because we have an
5174 * nGPA here instead of the required GPA.
5176 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5177 if (!is_guest_mode(vcpu) &&
5178 !kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5179 trace_kvm_fast_mmio(gpa);
5180 return kvm_skip_emulated_instruction(vcpu);
5183 return kvm_mmu_page_fault(vcpu, gpa, PFERR_RSVD_MASK, NULL, 0);
5186 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5188 WARN_ON_ONCE(!enable_vnmi);
5189 exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_NMI_WINDOW_EXITING);
5190 ++vcpu->stat.nmi_window_exits;
5191 kvm_make_request(KVM_REQ_EVENT, vcpu);
5196 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5198 struct vcpu_vmx *vmx = to_vmx(vcpu);
5199 bool intr_window_requested;
5200 unsigned count = 130;
5203 * We should never reach the point where we are emulating L2
5204 * due to invalid guest state as that means we incorrectly
5205 * allowed a nested VMEntry with an invalid vmcs12.
5207 WARN_ON_ONCE(vmx->emulation_required && vmx->nested.nested_run_pending);
5209 intr_window_requested = exec_controls_get(vmx) &
5210 CPU_BASED_INTR_WINDOW_EXITING;
5212 while (vmx->emulation_required && count-- != 0) {
5213 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
5214 return handle_interrupt_window(&vmx->vcpu);
5216 if (kvm_test_request(KVM_REQ_EVENT, vcpu))
5219 if (!kvm_emulate_instruction(vcpu, 0))
5222 if (vmx->emulation_required && !vmx->rmode.vm86_active &&
5223 vcpu->arch.exception.pending) {
5224 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5225 vcpu->run->internal.suberror =
5226 KVM_INTERNAL_ERROR_EMULATION;
5227 vcpu->run->internal.ndata = 0;
5231 if (vcpu->arch.halt_request) {
5232 vcpu->arch.halt_request = 0;
5233 return kvm_vcpu_halt(vcpu);
5237 * Note, return 1 and not 0, vcpu_run() is responsible for
5238 * morphing the pending signal into the proper return code.
5240 if (signal_pending(current))
5250 static void grow_ple_window(struct kvm_vcpu *vcpu)
5252 struct vcpu_vmx *vmx = to_vmx(vcpu);
5253 unsigned int old = vmx->ple_window;
5255 vmx->ple_window = __grow_ple_window(old, ple_window,
5259 if (vmx->ple_window != old) {
5260 vmx->ple_window_dirty = true;
5261 trace_kvm_ple_window_update(vcpu->vcpu_id,
5262 vmx->ple_window, old);
5266 static void shrink_ple_window(struct kvm_vcpu *vcpu)
5268 struct vcpu_vmx *vmx = to_vmx(vcpu);
5269 unsigned int old = vmx->ple_window;
5271 vmx->ple_window = __shrink_ple_window(old, ple_window,
5275 if (vmx->ple_window != old) {
5276 vmx->ple_window_dirty = true;
5277 trace_kvm_ple_window_update(vcpu->vcpu_id,
5278 vmx->ple_window, old);
5283 * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
5285 static void wakeup_handler(void)
5287 struct kvm_vcpu *vcpu;
5288 int cpu = smp_processor_id();
5290 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5291 list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
5292 blocked_vcpu_list) {
5293 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
5295 if (pi_test_on(pi_desc) == 1)
5296 kvm_vcpu_kick(vcpu);
5298 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5301 static void vmx_enable_tdp(void)
5303 kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK,
5304 enable_ept_ad_bits ? VMX_EPT_ACCESS_BIT : 0ull,
5305 enable_ept_ad_bits ? VMX_EPT_DIRTY_BIT : 0ull,
5306 0ull, VMX_EPT_EXECUTABLE_MASK,
5307 cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK,
5308 VMX_EPT_RWX_MASK, 0ull);
5310 ept_set_mmio_spte_mask();
5314 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5315 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5317 static int handle_pause(struct kvm_vcpu *vcpu)
5319 if (!kvm_pause_in_guest(vcpu->kvm))
5320 grow_ple_window(vcpu);
5323 * Intel sdm vol3 ch-25.1.3 says: The "PAUSE-loop exiting"
5324 * VM-execution control is ignored if CPL > 0. OTOH, KVM
5325 * never set PAUSE_EXITING and just set PLE if supported,
5326 * so the vcpu must be CPL=0 if it gets a PAUSE exit.
5328 kvm_vcpu_on_spin(vcpu, true);
5329 return kvm_skip_emulated_instruction(vcpu);
5332 static int handle_nop(struct kvm_vcpu *vcpu)
5334 return kvm_skip_emulated_instruction(vcpu);
5337 static int handle_mwait(struct kvm_vcpu *vcpu)
5339 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
5340 return handle_nop(vcpu);
5343 static int handle_invalid_op(struct kvm_vcpu *vcpu)
5345 kvm_queue_exception(vcpu, UD_VECTOR);
5349 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
5354 static int handle_monitor(struct kvm_vcpu *vcpu)
5356 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
5357 return handle_nop(vcpu);
5360 static int handle_invpcid(struct kvm_vcpu *vcpu)
5362 u32 vmx_instruction_info;
5366 struct x86_exception e;
5368 unsigned long roots_to_free = 0;
5374 if (!guest_cpuid_has(vcpu, X86_FEATURE_INVPCID)) {
5375 kvm_queue_exception(vcpu, UD_VECTOR);
5379 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5380 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
5383 kvm_inject_gp(vcpu, 0);
5387 /* According to the Intel instruction reference, the memory operand
5388 * is read even if it isn't needed (e.g., for type==all)
5390 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5391 vmx_instruction_info, false,
5392 sizeof(operand), &gva))
5395 if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
5396 kvm_inject_page_fault(vcpu, &e);
5400 if (operand.pcid >> 12 != 0) {
5401 kvm_inject_gp(vcpu, 0);
5405 pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
5408 case INVPCID_TYPE_INDIV_ADDR:
5409 if ((!pcid_enabled && (operand.pcid != 0)) ||
5410 is_noncanonical_address(operand.gla, vcpu)) {
5411 kvm_inject_gp(vcpu, 0);
5414 kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
5415 return kvm_skip_emulated_instruction(vcpu);
5417 case INVPCID_TYPE_SINGLE_CTXT:
5418 if (!pcid_enabled && (operand.pcid != 0)) {
5419 kvm_inject_gp(vcpu, 0);
5423 if (kvm_get_active_pcid(vcpu) == operand.pcid) {
5424 kvm_mmu_sync_roots(vcpu);
5425 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
5428 for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
5429 if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].cr3)
5431 roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
5433 kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
5435 * If neither the current cr3 nor any of the prev_roots use the
5436 * given PCID, then nothing needs to be done here because a
5437 * resync will happen anyway before switching to any other CR3.
5440 return kvm_skip_emulated_instruction(vcpu);
5442 case INVPCID_TYPE_ALL_NON_GLOBAL:
5444 * Currently, KVM doesn't mark global entries in the shadow
5445 * page tables, so a non-global flush just degenerates to a
5446 * global flush. If needed, we could optimize this later by
5447 * keeping track of global entries in shadow page tables.
5451 case INVPCID_TYPE_ALL_INCL_GLOBAL:
5452 kvm_mmu_unload(vcpu);
5453 return kvm_skip_emulated_instruction(vcpu);
5456 BUG(); /* We have already checked above that type <= 3 */
5460 static int handle_pml_full(struct kvm_vcpu *vcpu)
5462 unsigned long exit_qualification;
5464 trace_kvm_pml_full(vcpu->vcpu_id);
5466 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5469 * PML buffer FULL happened while executing iret from NMI,
5470 * "blocked by NMI" bit has to be set before next VM entry.
5472 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5474 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
5475 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5476 GUEST_INTR_STATE_NMI);
5479 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
5480 * here.., and there's no userspace involvement needed for PML.
5485 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
5487 struct vcpu_vmx *vmx = to_vmx(vcpu);
5489 if (!vmx->req_immediate_exit &&
5490 !unlikely(vmx->loaded_vmcs->hv_timer_soft_disabled))
5491 kvm_lapic_expired_hv_timer(vcpu);
5497 * When nested=0, all VMX instruction VM Exits filter here. The handlers
5498 * are overwritten by nested_vmx_setup() when nested=1.
5500 static int handle_vmx_instruction(struct kvm_vcpu *vcpu)
5502 kvm_queue_exception(vcpu, UD_VECTOR);
5506 static int handle_encls(struct kvm_vcpu *vcpu)
5509 * SGX virtualization is not yet supported. There is no software
5510 * enable bit for SGX, so we have to trap ENCLS and inject a #UD
5511 * to prevent the guest from executing ENCLS.
5513 kvm_queue_exception(vcpu, UD_VECTOR);
5518 * The exit handlers return 1 if the exit was handled fully and guest execution
5519 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
5520 * to be done to userspace and return 0.
5522 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
5523 [EXIT_REASON_EXCEPTION_NMI] = handle_exception_nmi,
5524 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
5525 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
5526 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
5527 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
5528 [EXIT_REASON_CR_ACCESS] = handle_cr,
5529 [EXIT_REASON_DR_ACCESS] = handle_dr,
5530 [EXIT_REASON_CPUID] = kvm_emulate_cpuid,
5531 [EXIT_REASON_MSR_READ] = kvm_emulate_rdmsr,
5532 [EXIT_REASON_MSR_WRITE] = kvm_emulate_wrmsr,
5533 [EXIT_REASON_INTERRUPT_WINDOW] = handle_interrupt_window,
5534 [EXIT_REASON_HLT] = kvm_emulate_halt,
5535 [EXIT_REASON_INVD] = handle_invd,
5536 [EXIT_REASON_INVLPG] = handle_invlpg,
5537 [EXIT_REASON_RDPMC] = handle_rdpmc,
5538 [EXIT_REASON_VMCALL] = handle_vmcall,
5539 [EXIT_REASON_VMCLEAR] = handle_vmx_instruction,
5540 [EXIT_REASON_VMLAUNCH] = handle_vmx_instruction,
5541 [EXIT_REASON_VMPTRLD] = handle_vmx_instruction,
5542 [EXIT_REASON_VMPTRST] = handle_vmx_instruction,
5543 [EXIT_REASON_VMREAD] = handle_vmx_instruction,
5544 [EXIT_REASON_VMRESUME] = handle_vmx_instruction,
5545 [EXIT_REASON_VMWRITE] = handle_vmx_instruction,
5546 [EXIT_REASON_VMOFF] = handle_vmx_instruction,
5547 [EXIT_REASON_VMON] = handle_vmx_instruction,
5548 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
5549 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
5550 [EXIT_REASON_APIC_WRITE] = handle_apic_write,
5551 [EXIT_REASON_EOI_INDUCED] = handle_apic_eoi_induced,
5552 [EXIT_REASON_WBINVD] = handle_wbinvd,
5553 [EXIT_REASON_XSETBV] = handle_xsetbv,
5554 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
5555 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
5556 [EXIT_REASON_GDTR_IDTR] = handle_desc,
5557 [EXIT_REASON_LDTR_TR] = handle_desc,
5558 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
5559 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
5560 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
5561 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_mwait,
5562 [EXIT_REASON_MONITOR_TRAP_FLAG] = handle_monitor_trap,
5563 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_monitor,
5564 [EXIT_REASON_INVEPT] = handle_vmx_instruction,
5565 [EXIT_REASON_INVVPID] = handle_vmx_instruction,
5566 [EXIT_REASON_RDRAND] = handle_invalid_op,
5567 [EXIT_REASON_RDSEED] = handle_invalid_op,
5568 [EXIT_REASON_PML_FULL] = handle_pml_full,
5569 [EXIT_REASON_INVPCID] = handle_invpcid,
5570 [EXIT_REASON_VMFUNC] = handle_vmx_instruction,
5571 [EXIT_REASON_PREEMPTION_TIMER] = handle_preemption_timer,
5572 [EXIT_REASON_ENCLS] = handle_encls,
5575 static const int kvm_vmx_max_exit_handlers =
5576 ARRAY_SIZE(kvm_vmx_exit_handlers);
5578 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
5580 *info1 = vmcs_readl(EXIT_QUALIFICATION);
5581 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
5584 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
5587 __free_page(vmx->pml_pg);
5592 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
5594 struct vcpu_vmx *vmx = to_vmx(vcpu);
5598 pml_idx = vmcs_read16(GUEST_PML_INDEX);
5600 /* Do nothing if PML buffer is empty */
5601 if (pml_idx == (PML_ENTITY_NUM - 1))
5604 /* PML index always points to next available PML buffer entity */
5605 if (pml_idx >= PML_ENTITY_NUM)
5610 pml_buf = page_address(vmx->pml_pg);
5611 for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
5614 gpa = pml_buf[pml_idx];
5615 WARN_ON(gpa & (PAGE_SIZE - 1));
5616 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5619 /* reset PML index */
5620 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
5624 * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
5625 * Called before reporting dirty_bitmap to userspace.
5627 static void kvm_flush_pml_buffers(struct kvm *kvm)
5630 struct kvm_vcpu *vcpu;
5632 * We only need to kick vcpu out of guest mode here, as PML buffer
5633 * is flushed at beginning of all VMEXITs, and it's obvious that only
5634 * vcpus running in guest are possible to have unflushed GPAs in PML
5637 kvm_for_each_vcpu(i, vcpu, kvm)
5638 kvm_vcpu_kick(vcpu);
5641 static void vmx_dump_sel(char *name, uint32_t sel)
5643 pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
5644 name, vmcs_read16(sel),
5645 vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
5646 vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
5647 vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
5650 static void vmx_dump_dtsel(char *name, uint32_t limit)
5652 pr_err("%s limit=0x%08x, base=0x%016lx\n",
5653 name, vmcs_read32(limit),
5654 vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
5657 void dump_vmcs(void)
5659 u32 vmentry_ctl, vmexit_ctl;
5660 u32 cpu_based_exec_ctrl, pin_based_exec_ctrl, secondary_exec_control;
5665 if (!dump_invalid_vmcs) {
5666 pr_warn_ratelimited("set kvm_intel.dump_invalid_vmcs=1 to dump internal KVM state.\n");
5670 vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
5671 vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
5672 cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5673 pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
5674 cr4 = vmcs_readl(GUEST_CR4);
5675 efer = vmcs_read64(GUEST_IA32_EFER);
5676 secondary_exec_control = 0;
5677 if (cpu_has_secondary_exec_ctrls())
5678 secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
5680 pr_err("*** Guest State ***\n");
5681 pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5682 vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
5683 vmcs_readl(CR0_GUEST_HOST_MASK));
5684 pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5685 cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
5686 pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
5687 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
5688 (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
5690 pr_err("PDPTR0 = 0x%016llx PDPTR1 = 0x%016llx\n",
5691 vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
5692 pr_err("PDPTR2 = 0x%016llx PDPTR3 = 0x%016llx\n",
5693 vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
5695 pr_err("RSP = 0x%016lx RIP = 0x%016lx\n",
5696 vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
5697 pr_err("RFLAGS=0x%08lx DR7 = 0x%016lx\n",
5698 vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
5699 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5700 vmcs_readl(GUEST_SYSENTER_ESP),
5701 vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
5702 vmx_dump_sel("CS: ", GUEST_CS_SELECTOR);
5703 vmx_dump_sel("DS: ", GUEST_DS_SELECTOR);
5704 vmx_dump_sel("SS: ", GUEST_SS_SELECTOR);
5705 vmx_dump_sel("ES: ", GUEST_ES_SELECTOR);
5706 vmx_dump_sel("FS: ", GUEST_FS_SELECTOR);
5707 vmx_dump_sel("GS: ", GUEST_GS_SELECTOR);
5708 vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
5709 vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
5710 vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
5711 vmx_dump_sel("TR: ", GUEST_TR_SELECTOR);
5712 if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
5713 (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
5714 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
5715 efer, vmcs_read64(GUEST_IA32_PAT));
5716 pr_err("DebugCtl = 0x%016llx DebugExceptions = 0x%016lx\n",
5717 vmcs_read64(GUEST_IA32_DEBUGCTL),
5718 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
5719 if (cpu_has_load_perf_global_ctrl() &&
5720 vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
5721 pr_err("PerfGlobCtl = 0x%016llx\n",
5722 vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
5723 if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
5724 pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
5725 pr_err("Interruptibility = %08x ActivityState = %08x\n",
5726 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
5727 vmcs_read32(GUEST_ACTIVITY_STATE));
5728 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
5729 pr_err("InterruptStatus = %04x\n",
5730 vmcs_read16(GUEST_INTR_STATUS));
5732 pr_err("*** Host State ***\n");
5733 pr_err("RIP = 0x%016lx RSP = 0x%016lx\n",
5734 vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
5735 pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
5736 vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
5737 vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
5738 vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
5739 vmcs_read16(HOST_TR_SELECTOR));
5740 pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
5741 vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
5742 vmcs_readl(HOST_TR_BASE));
5743 pr_err("GDTBase=%016lx IDTBase=%016lx\n",
5744 vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
5745 pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
5746 vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
5747 vmcs_readl(HOST_CR4));
5748 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5749 vmcs_readl(HOST_IA32_SYSENTER_ESP),
5750 vmcs_read32(HOST_IA32_SYSENTER_CS),
5751 vmcs_readl(HOST_IA32_SYSENTER_EIP));
5752 if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
5753 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
5754 vmcs_read64(HOST_IA32_EFER),
5755 vmcs_read64(HOST_IA32_PAT));
5756 if (cpu_has_load_perf_global_ctrl() &&
5757 vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
5758 pr_err("PerfGlobCtl = 0x%016llx\n",
5759 vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
5761 pr_err("*** Control State ***\n");
5762 pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
5763 pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
5764 pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
5765 pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
5766 vmcs_read32(EXCEPTION_BITMAP),
5767 vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
5768 vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
5769 pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
5770 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
5771 vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
5772 vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
5773 pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
5774 vmcs_read32(VM_EXIT_INTR_INFO),
5775 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
5776 vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
5777 pr_err(" reason=%08x qualification=%016lx\n",
5778 vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
5779 pr_err("IDTVectoring: info=%08x errcode=%08x\n",
5780 vmcs_read32(IDT_VECTORING_INFO_FIELD),
5781 vmcs_read32(IDT_VECTORING_ERROR_CODE));
5782 pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
5783 if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
5784 pr_err("TSC Multiplier = 0x%016llx\n",
5785 vmcs_read64(TSC_MULTIPLIER));
5786 if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW) {
5787 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
5788 u16 status = vmcs_read16(GUEST_INTR_STATUS);
5789 pr_err("SVI|RVI = %02x|%02x ", status >> 8, status & 0xff);
5791 pr_cont("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
5792 if (secondary_exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
5793 pr_err("APIC-access addr = 0x%016llx ", vmcs_read64(APIC_ACCESS_ADDR));
5794 pr_cont("virt-APIC addr = 0x%016llx\n", vmcs_read64(VIRTUAL_APIC_PAGE_ADDR));
5796 if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
5797 pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
5798 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
5799 pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
5800 n = vmcs_read32(CR3_TARGET_COUNT);
5801 for (i = 0; i + 1 < n; i += 4)
5802 pr_err("CR3 target%u=%016lx target%u=%016lx\n",
5803 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
5804 i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
5806 pr_err("CR3 target%u=%016lx\n",
5807 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
5808 if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
5809 pr_err("PLE Gap=%08x Window=%08x\n",
5810 vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
5811 if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
5812 pr_err("Virtual processor ID = 0x%04x\n",
5813 vmcs_read16(VIRTUAL_PROCESSOR_ID));
5817 * The guest has exited. See if we can fix it or if we need userspace
5820 static int vmx_handle_exit(struct kvm_vcpu *vcpu,
5821 enum exit_fastpath_completion exit_fastpath)
5823 struct vcpu_vmx *vmx = to_vmx(vcpu);
5824 u32 exit_reason = vmx->exit_reason;
5825 u32 vectoring_info = vmx->idt_vectoring_info;
5827 trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
5830 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
5831 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
5832 * querying dirty_bitmap, we only need to kick all vcpus out of guest
5833 * mode as if vcpus is in root mode, the PML buffer must has been
5837 vmx_flush_pml_buffer(vcpu);
5839 /* If guest state is invalid, start emulating */
5840 if (vmx->emulation_required)
5841 return handle_invalid_guest_state(vcpu);
5843 if (is_guest_mode(vcpu)) {
5845 * The host physical addresses of some pages of guest memory
5846 * are loaded into the vmcs02 (e.g. vmcs12's Virtual APIC
5847 * Page). The CPU may write to these pages via their host
5848 * physical address while L2 is running, bypassing any
5849 * address-translation-based dirty tracking (e.g. EPT write
5852 * Mark them dirty on every exit from L2 to prevent them from
5853 * getting out of sync with dirty tracking.
5855 nested_mark_vmcs12_pages_dirty(vcpu);
5857 if (nested_vmx_exit_reflected(vcpu, exit_reason))
5858 return nested_vmx_reflect_vmexit(vcpu, exit_reason);
5861 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
5863 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5864 vcpu->run->fail_entry.hardware_entry_failure_reason
5869 if (unlikely(vmx->fail)) {
5871 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5872 vcpu->run->fail_entry.hardware_entry_failure_reason
5873 = vmcs_read32(VM_INSTRUCTION_ERROR);
5879 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
5880 * delivery event since it indicates guest is accessing MMIO.
5881 * The vm-exit can be triggered again after return to guest that
5882 * will cause infinite loop.
5884 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
5885 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
5886 exit_reason != EXIT_REASON_EPT_VIOLATION &&
5887 exit_reason != EXIT_REASON_PML_FULL &&
5888 exit_reason != EXIT_REASON_TASK_SWITCH)) {
5889 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5890 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
5891 vcpu->run->internal.ndata = 3;
5892 vcpu->run->internal.data[0] = vectoring_info;
5893 vcpu->run->internal.data[1] = exit_reason;
5894 vcpu->run->internal.data[2] = vcpu->arch.exit_qualification;
5895 if (exit_reason == EXIT_REASON_EPT_MISCONFIG) {
5896 vcpu->run->internal.ndata++;
5897 vcpu->run->internal.data[3] =
5898 vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5903 if (unlikely(!enable_vnmi &&
5904 vmx->loaded_vmcs->soft_vnmi_blocked)) {
5905 if (vmx_interrupt_allowed(vcpu)) {
5906 vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5907 } else if (vmx->loaded_vmcs->vnmi_blocked_time > 1000000000LL &&
5908 vcpu->arch.nmi_pending) {
5910 * This CPU don't support us in finding the end of an
5911 * NMI-blocked window if the guest runs with IRQs
5912 * disabled. So we pull the trigger after 1 s of
5913 * futile waiting, but inform the user about this.
5915 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
5916 "state on VCPU %d after 1 s timeout\n",
5917 __func__, vcpu->vcpu_id);
5918 vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5922 if (exit_fastpath == EXIT_FASTPATH_SKIP_EMUL_INS) {
5923 kvm_skip_emulated_instruction(vcpu);
5927 if (exit_reason >= kvm_vmx_max_exit_handlers)
5928 goto unexpected_vmexit;
5929 #ifdef CONFIG_RETPOLINE
5930 if (exit_reason == EXIT_REASON_MSR_WRITE)
5931 return kvm_emulate_wrmsr(vcpu);
5932 else if (exit_reason == EXIT_REASON_PREEMPTION_TIMER)
5933 return handle_preemption_timer(vcpu);
5934 else if (exit_reason == EXIT_REASON_INTERRUPT_WINDOW)
5935 return handle_interrupt_window(vcpu);
5936 else if (exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
5937 return handle_external_interrupt(vcpu);
5938 else if (exit_reason == EXIT_REASON_HLT)
5939 return kvm_emulate_halt(vcpu);
5940 else if (exit_reason == EXIT_REASON_EPT_MISCONFIG)
5941 return handle_ept_misconfig(vcpu);
5944 exit_reason = array_index_nospec(exit_reason,
5945 kvm_vmx_max_exit_handlers);
5946 if (!kvm_vmx_exit_handlers[exit_reason])
5947 goto unexpected_vmexit;
5949 return kvm_vmx_exit_handlers[exit_reason](vcpu);
5952 vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n", exit_reason);
5954 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5955 vcpu->run->internal.suberror =
5956 KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON;
5957 vcpu->run->internal.ndata = 1;
5958 vcpu->run->internal.data[0] = exit_reason;
5963 * Software based L1D cache flush which is used when microcode providing
5964 * the cache control MSR is not loaded.
5966 * The L1D cache is 32 KiB on Nehalem and later microarchitectures, but to
5967 * flush it is required to read in 64 KiB because the replacement algorithm
5968 * is not exactly LRU. This could be sized at runtime via topology
5969 * information but as all relevant affected CPUs have 32KiB L1D cache size
5970 * there is no point in doing so.
5972 static void vmx_l1d_flush(struct kvm_vcpu *vcpu)
5974 int size = PAGE_SIZE << L1D_CACHE_ORDER;
5977 * This code is only executed when the the flush mode is 'cond' or
5980 if (static_branch_likely(&vmx_l1d_flush_cond)) {
5984 * Clear the per-vcpu flush bit, it gets set again
5985 * either from vcpu_run() or from one of the unsafe
5988 flush_l1d = vcpu->arch.l1tf_flush_l1d;
5989 vcpu->arch.l1tf_flush_l1d = false;
5992 * Clear the per-cpu flush bit, it gets set again from
5993 * the interrupt handlers.
5995 flush_l1d |= kvm_get_cpu_l1tf_flush_l1d();
5996 kvm_clear_cpu_l1tf_flush_l1d();
6002 vcpu->stat.l1d_flush++;
6004 if (static_cpu_has(X86_FEATURE_FLUSH_L1D)) {
6005 wrmsrl(MSR_IA32_FLUSH_CMD, L1D_FLUSH);
6010 /* First ensure the pages are in the TLB */
6011 "xorl %%eax, %%eax\n"
6012 ".Lpopulate_tlb:\n\t"
6013 "movzbl (%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
6014 "addl $4096, %%eax\n\t"
6015 "cmpl %%eax, %[size]\n\t"
6016 "jne .Lpopulate_tlb\n\t"
6017 "xorl %%eax, %%eax\n\t"
6019 /* Now fill the cache */
6020 "xorl %%eax, %%eax\n"
6022 "movzbl (%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
6023 "addl $64, %%eax\n\t"
6024 "cmpl %%eax, %[size]\n\t"
6025 "jne .Lfill_cache\n\t"
6027 :: [flush_pages] "r" (vmx_l1d_flush_pages),
6029 : "eax", "ebx", "ecx", "edx");
6032 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
6034 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6037 if (is_guest_mode(vcpu) &&
6038 nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
6041 tpr_threshold = (irr == -1 || tpr < irr) ? 0 : irr;
6042 if (is_guest_mode(vcpu))
6043 to_vmx(vcpu)->nested.l1_tpr_threshold = tpr_threshold;
6045 vmcs_write32(TPR_THRESHOLD, tpr_threshold);
6048 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu)
6050 struct vcpu_vmx *vmx = to_vmx(vcpu);
6051 u32 sec_exec_control;
6053 if (!lapic_in_kernel(vcpu))
6056 if (!flexpriority_enabled &&
6057 !cpu_has_vmx_virtualize_x2apic_mode())
6060 /* Postpone execution until vmcs01 is the current VMCS. */
6061 if (is_guest_mode(vcpu)) {
6062 vmx->nested.change_vmcs01_virtual_apic_mode = true;
6066 sec_exec_control = secondary_exec_controls_get(vmx);
6067 sec_exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6068 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
6070 switch (kvm_get_apic_mode(vcpu)) {
6071 case LAPIC_MODE_INVALID:
6072 WARN_ONCE(true, "Invalid local APIC state");
6073 case LAPIC_MODE_DISABLED:
6075 case LAPIC_MODE_XAPIC:
6076 if (flexpriority_enabled) {
6078 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6079 vmx_flush_tlb(vcpu, true);
6082 case LAPIC_MODE_X2APIC:
6083 if (cpu_has_vmx_virtualize_x2apic_mode())
6085 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
6088 secondary_exec_controls_set(vmx, sec_exec_control);
6090 vmx_update_msr_bitmap(vcpu);
6093 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
6095 if (!is_guest_mode(vcpu)) {
6096 vmcs_write64(APIC_ACCESS_ADDR, hpa);
6097 vmx_flush_tlb(vcpu, true);
6101 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
6109 status = vmcs_read16(GUEST_INTR_STATUS);
6111 if (max_isr != old) {
6113 status |= max_isr << 8;
6114 vmcs_write16(GUEST_INTR_STATUS, status);
6118 static void vmx_set_rvi(int vector)
6126 status = vmcs_read16(GUEST_INTR_STATUS);
6127 old = (u8)status & 0xff;
6128 if ((u8)vector != old) {
6130 status |= (u8)vector;
6131 vmcs_write16(GUEST_INTR_STATUS, status);
6135 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
6138 * When running L2, updating RVI is only relevant when
6139 * vmcs12 virtual-interrupt-delivery enabled.
6140 * However, it can be enabled only when L1 also
6141 * intercepts external-interrupts and in that case
6142 * we should not update vmcs02 RVI but instead intercept
6143 * interrupt. Therefore, do nothing when running L2.
6145 if (!is_guest_mode(vcpu))
6146 vmx_set_rvi(max_irr);
6149 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
6151 struct vcpu_vmx *vmx = to_vmx(vcpu);
6153 bool max_irr_updated;
6155 WARN_ON(!vcpu->arch.apicv_active);
6156 if (pi_test_on(&vmx->pi_desc)) {
6157 pi_clear_on(&vmx->pi_desc);
6159 * IOMMU can write to PID.ON, so the barrier matters even on UP.
6160 * But on x86 this is just a compiler barrier anyway.
6162 smp_mb__after_atomic();
6164 kvm_apic_update_irr(vcpu, vmx->pi_desc.pir, &max_irr);
6167 * If we are running L2 and L1 has a new pending interrupt
6168 * which can be injected, we should re-evaluate
6169 * what should be done with this new L1 interrupt.
6170 * If L1 intercepts external-interrupts, we should
6171 * exit from L2 to L1. Otherwise, interrupt should be
6172 * delivered directly to L2.
6174 if (is_guest_mode(vcpu) && max_irr_updated) {
6175 if (nested_exit_on_intr(vcpu))
6176 kvm_vcpu_exiting_guest_mode(vcpu);
6178 kvm_make_request(KVM_REQ_EVENT, vcpu);
6181 max_irr = kvm_lapic_find_highest_irr(vcpu);
6183 vmx_hwapic_irr_update(vcpu, max_irr);
6187 static bool vmx_dy_apicv_has_pending_interrupt(struct kvm_vcpu *vcpu)
6189 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
6191 return pi_test_on(pi_desc) ||
6192 (pi_test_sn(pi_desc) && !pi_is_pir_empty(pi_desc));
6195 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
6197 if (!kvm_vcpu_apicv_active(vcpu))
6200 vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
6201 vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
6202 vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
6203 vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
6206 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
6208 struct vcpu_vmx *vmx = to_vmx(vcpu);
6210 pi_clear_on(&vmx->pi_desc);
6211 memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
6214 static void handle_exception_nmi_irqoff(struct vcpu_vmx *vmx)
6216 vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6218 /* if exit due to PF check for async PF */
6219 if (is_page_fault(vmx->exit_intr_info)) {
6220 vmx->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason();
6221 /* Handle machine checks before interrupts are enabled */
6222 } else if (is_machine_check(vmx->exit_intr_info)) {
6223 kvm_machine_check();
6224 /* We need to handle NMIs before interrupts are enabled */
6225 } else if (is_nmi(vmx->exit_intr_info)) {
6226 kvm_before_interrupt(&vmx->vcpu);
6228 kvm_after_interrupt(&vmx->vcpu);
6232 static void handle_external_interrupt_irqoff(struct kvm_vcpu *vcpu)
6234 unsigned int vector;
6235 unsigned long entry;
6236 #ifdef CONFIG_X86_64
6242 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6243 if (WARN_ONCE(!is_external_intr(intr_info),
6244 "KVM: unexpected VM-Exit interrupt info: 0x%x", intr_info))
6247 vector = intr_info & INTR_INFO_VECTOR_MASK;
6248 desc = (gate_desc *)host_idt_base + vector;
6249 entry = gate_offset(desc);
6251 kvm_before_interrupt(vcpu);
6254 #ifdef CONFIG_X86_64
6255 "mov %%" _ASM_SP ", %[sp]\n\t"
6256 "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
6261 __ASM_SIZE(push) " $%c[cs]\n\t"
6264 #ifdef CONFIG_X86_64
6269 [thunk_target]"r"(entry),
6270 [ss]"i"(__KERNEL_DS),
6271 [cs]"i"(__KERNEL_CS)
6274 kvm_after_interrupt(vcpu);
6276 STACK_FRAME_NON_STANDARD(handle_external_interrupt_irqoff);
6278 static void vmx_handle_exit_irqoff(struct kvm_vcpu *vcpu,
6279 enum exit_fastpath_completion *exit_fastpath)
6281 struct vcpu_vmx *vmx = to_vmx(vcpu);
6283 if (vmx->exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
6284 handle_external_interrupt_irqoff(vcpu);
6285 else if (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI)
6286 handle_exception_nmi_irqoff(vmx);
6287 else if (!is_guest_mode(vcpu) &&
6288 vmx->exit_reason == EXIT_REASON_MSR_WRITE)
6289 *exit_fastpath = handle_fastpath_set_msr_irqoff(vcpu);
6292 static bool vmx_has_emulated_msr(int index)
6295 case MSR_IA32_SMBASE:
6297 * We cannot do SMM unless we can run the guest in big
6300 return enable_unrestricted_guest || emulate_invalid_guest_state;
6301 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
6303 case MSR_AMD64_VIRT_SPEC_CTRL:
6304 /* This is AMD only. */
6311 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
6316 bool idtv_info_valid;
6318 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6321 if (vmx->loaded_vmcs->nmi_known_unmasked)
6324 * Can't use vmx->exit_intr_info since we're not sure what
6325 * the exit reason is.
6327 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6328 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
6329 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
6331 * SDM 3: 27.7.1.2 (September 2008)
6332 * Re-set bit "block by NMI" before VM entry if vmexit caused by
6333 * a guest IRET fault.
6334 * SDM 3: 23.2.2 (September 2008)
6335 * Bit 12 is undefined in any of the following cases:
6336 * If the VM exit sets the valid bit in the IDT-vectoring
6337 * information field.
6338 * If the VM exit is due to a double fault.
6340 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
6341 vector != DF_VECTOR && !idtv_info_valid)
6342 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
6343 GUEST_INTR_STATE_NMI);
6345 vmx->loaded_vmcs->nmi_known_unmasked =
6346 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
6347 & GUEST_INTR_STATE_NMI);
6348 } else if (unlikely(vmx->loaded_vmcs->soft_vnmi_blocked))
6349 vmx->loaded_vmcs->vnmi_blocked_time +=
6350 ktime_to_ns(ktime_sub(ktime_get(),
6351 vmx->loaded_vmcs->entry_time));
6354 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
6355 u32 idt_vectoring_info,
6356 int instr_len_field,
6357 int error_code_field)
6361 bool idtv_info_valid;
6363 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6365 vcpu->arch.nmi_injected = false;
6366 kvm_clear_exception_queue(vcpu);
6367 kvm_clear_interrupt_queue(vcpu);
6369 if (!idtv_info_valid)
6372 kvm_make_request(KVM_REQ_EVENT, vcpu);
6374 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
6375 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
6378 case INTR_TYPE_NMI_INTR:
6379 vcpu->arch.nmi_injected = true;
6381 * SDM 3: 27.7.1.2 (September 2008)
6382 * Clear bit "block by NMI" before VM entry if a NMI
6385 vmx_set_nmi_mask(vcpu, false);
6387 case INTR_TYPE_SOFT_EXCEPTION:
6388 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6390 case INTR_TYPE_HARD_EXCEPTION:
6391 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
6392 u32 err = vmcs_read32(error_code_field);
6393 kvm_requeue_exception_e(vcpu, vector, err);
6395 kvm_requeue_exception(vcpu, vector);
6397 case INTR_TYPE_SOFT_INTR:
6398 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6400 case INTR_TYPE_EXT_INTR:
6401 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
6408 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
6410 __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
6411 VM_EXIT_INSTRUCTION_LEN,
6412 IDT_VECTORING_ERROR_CODE);
6415 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
6417 __vmx_complete_interrupts(vcpu,
6418 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
6419 VM_ENTRY_INSTRUCTION_LEN,
6420 VM_ENTRY_EXCEPTION_ERROR_CODE);
6422 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
6425 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
6428 struct perf_guest_switch_msr *msrs;
6430 msrs = perf_guest_get_msrs(&nr_msrs);
6435 for (i = 0; i < nr_msrs; i++)
6436 if (msrs[i].host == msrs[i].guest)
6437 clear_atomic_switch_msr(vmx, msrs[i].msr);
6439 add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
6440 msrs[i].host, false);
6443 static void atomic_switch_umwait_control_msr(struct vcpu_vmx *vmx)
6445 u32 host_umwait_control;
6447 if (!vmx_has_waitpkg(vmx))
6450 host_umwait_control = get_umwait_control_msr();
6452 if (vmx->msr_ia32_umwait_control != host_umwait_control)
6453 add_atomic_switch_msr(vmx, MSR_IA32_UMWAIT_CONTROL,
6454 vmx->msr_ia32_umwait_control,
6455 host_umwait_control, false);
6457 clear_atomic_switch_msr(vmx, MSR_IA32_UMWAIT_CONTROL);
6460 static void vmx_update_hv_timer(struct kvm_vcpu *vcpu)
6462 struct vcpu_vmx *vmx = to_vmx(vcpu);
6466 if (vmx->req_immediate_exit) {
6467 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, 0);
6468 vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6469 } else if (vmx->hv_deadline_tsc != -1) {
6471 if (vmx->hv_deadline_tsc > tscl)
6472 /* set_hv_timer ensures the delta fits in 32-bits */
6473 delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
6474 cpu_preemption_timer_multi);
6478 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
6479 vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6480 } else if (!vmx->loaded_vmcs->hv_timer_soft_disabled) {
6481 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, -1);
6482 vmx->loaded_vmcs->hv_timer_soft_disabled = true;
6486 void vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp)
6488 if (unlikely(host_rsp != vmx->loaded_vmcs->host_state.rsp)) {
6489 vmx->loaded_vmcs->host_state.rsp = host_rsp;
6490 vmcs_writel(HOST_RSP, host_rsp);
6494 bool __vmx_vcpu_run(struct vcpu_vmx *vmx, unsigned long *regs, bool launched);
6496 static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
6498 struct vcpu_vmx *vmx = to_vmx(vcpu);
6499 unsigned long cr3, cr4;
6501 /* Record the guest's net vcpu time for enforced NMI injections. */
6502 if (unlikely(!enable_vnmi &&
6503 vmx->loaded_vmcs->soft_vnmi_blocked))
6504 vmx->loaded_vmcs->entry_time = ktime_get();
6506 /* Don't enter VMX if guest state is invalid, let the exit handler
6507 start emulation until we arrive back to a valid state */
6508 if (vmx->emulation_required)
6511 if (vmx->ple_window_dirty) {
6512 vmx->ple_window_dirty = false;
6513 vmcs_write32(PLE_WINDOW, vmx->ple_window);
6517 * We did this in prepare_switch_to_guest, because it needs to
6518 * be within srcu_read_lock.
6520 WARN_ON_ONCE(vmx->nested.need_vmcs12_to_shadow_sync);
6522 if (kvm_register_is_dirty(vcpu, VCPU_REGS_RSP))
6523 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
6524 if (kvm_register_is_dirty(vcpu, VCPU_REGS_RIP))
6525 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
6527 cr3 = __get_current_cr3_fast();
6528 if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
6529 vmcs_writel(HOST_CR3, cr3);
6530 vmx->loaded_vmcs->host_state.cr3 = cr3;
6533 cr4 = cr4_read_shadow();
6534 if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
6535 vmcs_writel(HOST_CR4, cr4);
6536 vmx->loaded_vmcs->host_state.cr4 = cr4;
6539 /* When single-stepping over STI and MOV SS, we must clear the
6540 * corresponding interruptibility bits in the guest state. Otherwise
6541 * vmentry fails as it then expects bit 14 (BS) in pending debug
6542 * exceptions being set, but that's not correct for the guest debugging
6544 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6545 vmx_set_interrupt_shadow(vcpu, 0);
6547 kvm_load_guest_xsave_state(vcpu);
6549 if (static_cpu_has(X86_FEATURE_PKU) &&
6550 kvm_read_cr4_bits(vcpu, X86_CR4_PKE) &&
6551 vcpu->arch.pkru != vmx->host_pkru)
6552 __write_pkru(vcpu->arch.pkru);
6554 pt_guest_enter(vmx);
6556 if (vcpu_to_pmu(vcpu)->version)
6557 atomic_switch_perf_msrs(vmx);
6558 atomic_switch_umwait_control_msr(vmx);
6560 if (enable_preemption_timer)
6561 vmx_update_hv_timer(vcpu);
6563 if (lapic_in_kernel(vcpu) &&
6564 vcpu->arch.apic->lapic_timer.timer_advance_ns)
6565 kvm_wait_lapic_expire(vcpu);
6568 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
6569 * it's non-zero. Since vmentry is serialising on affected CPUs, there
6570 * is no need to worry about the conditional branch over the wrmsr
6571 * being speculatively taken.
6573 x86_spec_ctrl_set_guest(vmx->spec_ctrl, 0);
6575 /* L1D Flush includes CPU buffer clear to mitigate MDS */
6576 if (static_branch_unlikely(&vmx_l1d_should_flush))
6577 vmx_l1d_flush(vcpu);
6578 else if (static_branch_unlikely(&mds_user_clear))
6579 mds_clear_cpu_buffers();
6581 if (vcpu->arch.cr2 != read_cr2())
6582 write_cr2(vcpu->arch.cr2);
6584 vmx->fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
6585 vmx->loaded_vmcs->launched);
6587 vcpu->arch.cr2 = read_cr2();
6590 * We do not use IBRS in the kernel. If this vCPU has used the
6591 * SPEC_CTRL MSR it may have left it on; save the value and
6592 * turn it off. This is much more efficient than blindly adding
6593 * it to the atomic save/restore list. Especially as the former
6594 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
6596 * For non-nested case:
6597 * If the L01 MSR bitmap does not intercept the MSR, then we need to
6601 * If the L02 MSR bitmap does not intercept the MSR, then we need to
6604 if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
6605 vmx->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
6607 x86_spec_ctrl_restore_host(vmx->spec_ctrl, 0);
6609 /* All fields are clean at this point */
6610 if (static_branch_unlikely(&enable_evmcs))
6611 current_evmcs->hv_clean_fields |=
6612 HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
6614 if (static_branch_unlikely(&enable_evmcs))
6615 current_evmcs->hv_vp_id = vcpu->arch.hyperv.vp_index;
6617 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
6618 if (vmx->host_debugctlmsr)
6619 update_debugctlmsr(vmx->host_debugctlmsr);
6621 #ifndef CONFIG_X86_64
6623 * The sysexit path does not restore ds/es, so we must set them to
6624 * a reasonable value ourselves.
6626 * We can't defer this to vmx_prepare_switch_to_host() since that
6627 * function may be executed in interrupt context, which saves and
6628 * restore segments around it, nullifying its effect.
6630 loadsegment(ds, __USER_DS);
6631 loadsegment(es, __USER_DS);
6634 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
6635 | (1 << VCPU_EXREG_RFLAGS)
6636 | (1 << VCPU_EXREG_PDPTR)
6637 | (1 << VCPU_EXREG_SEGMENTS)
6638 | (1 << VCPU_EXREG_CR3));
6639 vcpu->arch.regs_dirty = 0;
6644 * eager fpu is enabled if PKEY is supported and CR4 is switched
6645 * back on host, so it is safe to read guest PKRU from current
6648 if (static_cpu_has(X86_FEATURE_PKU) &&
6649 kvm_read_cr4_bits(vcpu, X86_CR4_PKE)) {
6650 vcpu->arch.pkru = rdpkru();
6651 if (vcpu->arch.pkru != vmx->host_pkru)
6652 __write_pkru(vmx->host_pkru);
6655 kvm_load_host_xsave_state(vcpu);
6657 vmx->nested.nested_run_pending = 0;
6658 vmx->idt_vectoring_info = 0;
6660 vmx->exit_reason = vmx->fail ? 0xdead : vmcs_read32(VM_EXIT_REASON);
6661 if ((u16)vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY)
6662 kvm_machine_check();
6664 if (vmx->fail || (vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
6667 vmx->loaded_vmcs->launched = 1;
6668 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
6670 vmx_recover_nmi_blocking(vmx);
6671 vmx_complete_interrupts(vmx);
6674 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
6676 struct vcpu_vmx *vmx = to_vmx(vcpu);
6679 vmx_destroy_pml_buffer(vmx);
6680 free_vpid(vmx->vpid);
6681 nested_vmx_free_vcpu(vcpu);
6682 free_loaded_vmcs(vmx->loaded_vmcs);
6685 static int vmx_create_vcpu(struct kvm_vcpu *vcpu)
6687 struct vcpu_vmx *vmx;
6688 unsigned long *msr_bitmap;
6691 BUILD_BUG_ON(offsetof(struct vcpu_vmx, vcpu) != 0);
6696 vmx->vpid = allocate_vpid();
6699 * If PML is turned on, failure on enabling PML just results in failure
6700 * of creating the vcpu, therefore we can simplify PML logic (by
6701 * avoiding dealing with cases, such as enabling PML partially on vcpus
6702 * for the guest), etc.
6705 vmx->pml_pg = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
6710 BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) != NR_SHARED_MSRS);
6712 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
6713 u32 index = vmx_msr_index[i];
6714 u32 data_low, data_high;
6717 if (rdmsr_safe(index, &data_low, &data_high) < 0)
6719 if (wrmsr_safe(index, data_low, data_high) < 0)
6722 vmx->guest_msrs[j].index = i;
6723 vmx->guest_msrs[j].data = 0;
6725 case MSR_IA32_TSX_CTRL:
6727 * No need to pass TSX_CTRL_CPUID_CLEAR through, so
6728 * let's avoid changing CPUID bits under the host
6731 vmx->guest_msrs[j].mask = ~(u64)TSX_CTRL_CPUID_CLEAR;
6734 vmx->guest_msrs[j].mask = -1ull;
6740 err = alloc_loaded_vmcs(&vmx->vmcs01);
6744 msr_bitmap = vmx->vmcs01.msr_bitmap;
6745 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_TSC, MSR_TYPE_R);
6746 vmx_disable_intercept_for_msr(msr_bitmap, MSR_FS_BASE, MSR_TYPE_RW);
6747 vmx_disable_intercept_for_msr(msr_bitmap, MSR_GS_BASE, MSR_TYPE_RW);
6748 vmx_disable_intercept_for_msr(msr_bitmap, MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
6749 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_CS, MSR_TYPE_RW);
6750 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_ESP, MSR_TYPE_RW);
6751 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_EIP, MSR_TYPE_RW);
6752 if (kvm_cstate_in_guest(vcpu->kvm)) {
6753 vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C1_RES, MSR_TYPE_R);
6754 vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C3_RESIDENCY, MSR_TYPE_R);
6755 vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C6_RESIDENCY, MSR_TYPE_R);
6756 vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C7_RESIDENCY, MSR_TYPE_R);
6758 vmx->msr_bitmap_mode = 0;
6760 vmx->loaded_vmcs = &vmx->vmcs01;
6762 vmx_vcpu_load(vcpu, cpu);
6767 if (cpu_need_virtualize_apic_accesses(vcpu)) {
6768 err = alloc_apic_access_page(vcpu->kvm);
6773 if (enable_ept && !enable_unrestricted_guest) {
6774 err = init_rmode_identity_map(vcpu->kvm);
6780 nested_vmx_setup_ctls_msrs(&vmx->nested.msrs,
6781 vmx_capability.ept);
6783 memset(&vmx->nested.msrs, 0, sizeof(vmx->nested.msrs));
6785 vmx->nested.posted_intr_nv = -1;
6786 vmx->nested.current_vmptr = -1ull;
6788 vcpu->arch.microcode_version = 0x100000000ULL;
6789 vmx->msr_ia32_feature_control_valid_bits = FEAT_CTL_LOCKED;
6792 * Enforce invariant: pi_desc.nv is always either POSTED_INTR_VECTOR
6793 * or POSTED_INTR_WAKEUP_VECTOR.
6795 vmx->pi_desc.nv = POSTED_INTR_VECTOR;
6796 vmx->pi_desc.sn = 1;
6798 vmx->ept_pointer = INVALID_PAGE;
6803 free_loaded_vmcs(vmx->loaded_vmcs);
6805 vmx_destroy_pml_buffer(vmx);
6807 free_vpid(vmx->vpid);
6811 #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"
6812 #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"
6814 static int vmx_vm_init(struct kvm *kvm)
6816 spin_lock_init(&to_kvm_vmx(kvm)->ept_pointer_lock);
6819 kvm->arch.pause_in_guest = true;
6821 if (boot_cpu_has(X86_BUG_L1TF) && enable_ept) {
6822 switch (l1tf_mitigation) {
6823 case L1TF_MITIGATION_OFF:
6824 case L1TF_MITIGATION_FLUSH_NOWARN:
6825 /* 'I explicitly don't care' is set */
6827 case L1TF_MITIGATION_FLUSH:
6828 case L1TF_MITIGATION_FLUSH_NOSMT:
6829 case L1TF_MITIGATION_FULL:
6831 * Warn upon starting the first VM in a potentially
6832 * insecure environment.
6834 if (sched_smt_active())
6835 pr_warn_once(L1TF_MSG_SMT);
6836 if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER)
6837 pr_warn_once(L1TF_MSG_L1D);
6839 case L1TF_MITIGATION_FULL_FORCE:
6840 /* Flush is enforced */
6844 kvm_apicv_init(kvm, enable_apicv);
6848 static int __init vmx_check_processor_compat(void)
6850 struct vmcs_config vmcs_conf;
6851 struct vmx_capability vmx_cap;
6853 if (!this_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
6854 !this_cpu_has(X86_FEATURE_VMX)) {
6855 pr_err("kvm: VMX is disabled on CPU %d\n", smp_processor_id());
6859 if (setup_vmcs_config(&vmcs_conf, &vmx_cap) < 0)
6862 nested_vmx_setup_ctls_msrs(&vmcs_conf.nested, vmx_cap.ept);
6863 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
6864 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
6865 smp_processor_id());
6871 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
6876 /* We wanted to honor guest CD/MTRR/PAT, but doing so could result in
6877 * memory aliases with conflicting memory types and sometimes MCEs.
6878 * We have to be careful as to what are honored and when.
6880 * For MMIO, guest CD/MTRR are ignored. The EPT memory type is set to
6881 * UC. The effective memory type is UC or WC depending on guest PAT.
6882 * This was historically the source of MCEs and we want to be
6885 * When there is no need to deal with noncoherent DMA (e.g., no VT-d
6886 * or VT-d has snoop control), guest CD/MTRR/PAT are all ignored. The
6887 * EPT memory type is set to WB. The effective memory type is forced
6890 * Otherwise, we trust guest. Guest CD/MTRR/PAT are all honored. The
6891 * EPT memory type is used to emulate guest CD/MTRR.
6895 cache = MTRR_TYPE_UNCACHABLE;
6899 if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
6900 ipat = VMX_EPT_IPAT_BIT;
6901 cache = MTRR_TYPE_WRBACK;
6905 if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
6906 ipat = VMX_EPT_IPAT_BIT;
6907 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
6908 cache = MTRR_TYPE_WRBACK;
6910 cache = MTRR_TYPE_UNCACHABLE;
6914 cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
6917 return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
6920 static void vmcs_set_secondary_exec_control(struct vcpu_vmx *vmx)
6923 * These bits in the secondary execution controls field
6924 * are dynamic, the others are mostly based on the hypervisor
6925 * architecture and the guest's CPUID. Do not touch the
6929 SECONDARY_EXEC_SHADOW_VMCS |
6930 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
6931 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6932 SECONDARY_EXEC_DESC;
6934 u32 new_ctl = vmx->secondary_exec_control;
6935 u32 cur_ctl = secondary_exec_controls_get(vmx);
6937 secondary_exec_controls_set(vmx, (new_ctl & ~mask) | (cur_ctl & mask));
6941 * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
6942 * (indicating "allowed-1") if they are supported in the guest's CPUID.
6944 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
6946 struct vcpu_vmx *vmx = to_vmx(vcpu);
6947 struct kvm_cpuid_entry2 *entry;
6949 vmx->nested.msrs.cr0_fixed1 = 0xffffffff;
6950 vmx->nested.msrs.cr4_fixed1 = X86_CR4_PCE;
6952 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do { \
6953 if (entry && (entry->_reg & (_cpuid_mask))) \
6954 vmx->nested.msrs.cr4_fixed1 |= (_cr4_mask); \
6957 entry = kvm_find_cpuid_entry(vcpu, 0x1, 0);
6958 cr4_fixed1_update(X86_CR4_VME, edx, feature_bit(VME));
6959 cr4_fixed1_update(X86_CR4_PVI, edx, feature_bit(VME));
6960 cr4_fixed1_update(X86_CR4_TSD, edx, feature_bit(TSC));
6961 cr4_fixed1_update(X86_CR4_DE, edx, feature_bit(DE));
6962 cr4_fixed1_update(X86_CR4_PSE, edx, feature_bit(PSE));
6963 cr4_fixed1_update(X86_CR4_PAE, edx, feature_bit(PAE));
6964 cr4_fixed1_update(X86_CR4_MCE, edx, feature_bit(MCE));
6965 cr4_fixed1_update(X86_CR4_PGE, edx, feature_bit(PGE));
6966 cr4_fixed1_update(X86_CR4_OSFXSR, edx, feature_bit(FXSR));
6967 cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, feature_bit(XMM));
6968 cr4_fixed1_update(X86_CR4_VMXE, ecx, feature_bit(VMX));
6969 cr4_fixed1_update(X86_CR4_SMXE, ecx, feature_bit(SMX));
6970 cr4_fixed1_update(X86_CR4_PCIDE, ecx, feature_bit(PCID));
6971 cr4_fixed1_update(X86_CR4_OSXSAVE, ecx, feature_bit(XSAVE));
6973 entry = kvm_find_cpuid_entry(vcpu, 0x7, 0);
6974 cr4_fixed1_update(X86_CR4_FSGSBASE, ebx, feature_bit(FSGSBASE));
6975 cr4_fixed1_update(X86_CR4_SMEP, ebx, feature_bit(SMEP));
6976 cr4_fixed1_update(X86_CR4_SMAP, ebx, feature_bit(SMAP));
6977 cr4_fixed1_update(X86_CR4_PKE, ecx, feature_bit(PKU));
6978 cr4_fixed1_update(X86_CR4_UMIP, ecx, feature_bit(UMIP));
6979 cr4_fixed1_update(X86_CR4_LA57, ecx, feature_bit(LA57));
6981 #undef cr4_fixed1_update
6984 static void nested_vmx_entry_exit_ctls_update(struct kvm_vcpu *vcpu)
6986 struct vcpu_vmx *vmx = to_vmx(vcpu);
6988 if (kvm_mpx_supported()) {
6989 bool mpx_enabled = guest_cpuid_has(vcpu, X86_FEATURE_MPX);
6992 vmx->nested.msrs.entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
6993 vmx->nested.msrs.exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
6995 vmx->nested.msrs.entry_ctls_high &= ~VM_ENTRY_LOAD_BNDCFGS;
6996 vmx->nested.msrs.exit_ctls_high &= ~VM_EXIT_CLEAR_BNDCFGS;
7001 static void update_intel_pt_cfg(struct kvm_vcpu *vcpu)
7003 struct vcpu_vmx *vmx = to_vmx(vcpu);
7004 struct kvm_cpuid_entry2 *best = NULL;
7007 for (i = 0; i < PT_CPUID_LEAVES; i++) {
7008 best = kvm_find_cpuid_entry(vcpu, 0x14, i);
7011 vmx->pt_desc.caps[CPUID_EAX + i*PT_CPUID_REGS_NUM] = best->eax;
7012 vmx->pt_desc.caps[CPUID_EBX + i*PT_CPUID_REGS_NUM] = best->ebx;
7013 vmx->pt_desc.caps[CPUID_ECX + i*PT_CPUID_REGS_NUM] = best->ecx;
7014 vmx->pt_desc.caps[CPUID_EDX + i*PT_CPUID_REGS_NUM] = best->edx;
7017 /* Get the number of configurable Address Ranges for filtering */
7018 vmx->pt_desc.addr_range = intel_pt_validate_cap(vmx->pt_desc.caps,
7019 PT_CAP_num_address_ranges);
7021 /* Initialize and clear the no dependency bits */
7022 vmx->pt_desc.ctl_bitmask = ~(RTIT_CTL_TRACEEN | RTIT_CTL_OS |
7023 RTIT_CTL_USR | RTIT_CTL_TSC_EN | RTIT_CTL_DISRETC);
7026 * If CPUID.(EAX=14H,ECX=0):EBX[0]=1 CR3Filter can be set otherwise
7027 * will inject an #GP
7029 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_cr3_filtering))
7030 vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_CR3EN;
7033 * If CPUID.(EAX=14H,ECX=0):EBX[1]=1 CYCEn, CycThresh and
7034 * PSBFreq can be set
7036 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc))
7037 vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_CYCLEACC |
7038 RTIT_CTL_CYC_THRESH | RTIT_CTL_PSB_FREQ);
7041 * If CPUID.(EAX=14H,ECX=0):EBX[3]=1 MTCEn BranchEn and
7042 * MTCFreq can be set
7044 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc))
7045 vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_MTC_EN |
7046 RTIT_CTL_BRANCH_EN | RTIT_CTL_MTC_RANGE);
7048 /* If CPUID.(EAX=14H,ECX=0):EBX[4]=1 FUPonPTW and PTWEn can be set */
7049 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_ptwrite))
7050 vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_FUP_ON_PTW |
7053 /* If CPUID.(EAX=14H,ECX=0):EBX[5]=1 PwrEvEn can be set */
7054 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_power_event_trace))
7055 vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_PWR_EVT_EN;
7057 /* If CPUID.(EAX=14H,ECX=0):ECX[0]=1 ToPA can be set */
7058 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_topa_output))
7059 vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_TOPA;
7061 /* If CPUID.(EAX=14H,ECX=0):ECX[3]=1 FabircEn can be set */
7062 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_output_subsys))
7063 vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_FABRIC_EN;
7065 /* unmask address range configure area */
7066 for (i = 0; i < vmx->pt_desc.addr_range; i++)
7067 vmx->pt_desc.ctl_bitmask &= ~(0xfULL << (32 + i * 4));
7070 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
7072 struct vcpu_vmx *vmx = to_vmx(vcpu);
7074 /* xsaves_enabled is recomputed in vmx_compute_secondary_exec_control(). */
7075 vcpu->arch.xsaves_enabled = false;
7077 if (cpu_has_secondary_exec_ctrls()) {
7078 vmx_compute_secondary_exec_control(vmx);
7079 vmcs_set_secondary_exec_control(vmx);
7082 if (nested_vmx_allowed(vcpu))
7083 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7084 FEAT_CTL_VMX_ENABLED_INSIDE_SMX |
7085 FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX;
7087 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7088 ~(FEAT_CTL_VMX_ENABLED_INSIDE_SMX |
7089 FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX);
7091 if (nested_vmx_allowed(vcpu)) {
7092 nested_vmx_cr_fixed1_bits_update(vcpu);
7093 nested_vmx_entry_exit_ctls_update(vcpu);
7096 if (boot_cpu_has(X86_FEATURE_INTEL_PT) &&
7097 guest_cpuid_has(vcpu, X86_FEATURE_INTEL_PT))
7098 update_intel_pt_cfg(vcpu);
7100 if (boot_cpu_has(X86_FEATURE_RTM)) {
7101 struct shared_msr_entry *msr;
7102 msr = find_msr_entry(vmx, MSR_IA32_TSX_CTRL);
7104 bool enabled = guest_cpuid_has(vcpu, X86_FEATURE_RTM);
7105 vmx_set_guest_msr(vmx, msr, enabled ? 0 : TSX_CTRL_RTM_DISABLE);
7110 static __init void vmx_set_cpu_caps(void)
7116 kvm_cpu_cap_set(X86_FEATURE_VMX);
7119 if (kvm_mpx_supported())
7120 kvm_cpu_cap_check_and_set(X86_FEATURE_MPX);
7121 if (cpu_has_vmx_invpcid())
7122 kvm_cpu_cap_check_and_set(X86_FEATURE_INVPCID);
7123 if (vmx_pt_mode_is_host_guest())
7124 kvm_cpu_cap_check_and_set(X86_FEATURE_INTEL_PT);
7126 /* PKU is not yet implemented for shadow paging. */
7127 if (enable_ept && boot_cpu_has(X86_FEATURE_OSPKE))
7128 kvm_cpu_cap_check_and_set(X86_FEATURE_PKU);
7130 if (vmx_umip_emulated())
7131 kvm_cpu_cap_set(X86_FEATURE_UMIP);
7135 if (!vmx_xsaves_supported())
7136 kvm_cpu_cap_clear(X86_FEATURE_XSAVES);
7138 /* CPUID 0x80000001 */
7139 if (!cpu_has_vmx_rdtscp())
7140 kvm_cpu_cap_clear(X86_FEATURE_RDTSCP);
7143 static void vmx_request_immediate_exit(struct kvm_vcpu *vcpu)
7145 to_vmx(vcpu)->req_immediate_exit = true;
7148 static int vmx_check_intercept_io(struct kvm_vcpu *vcpu,
7149 struct x86_instruction_info *info)
7151 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7152 unsigned short port;
7156 if (info->intercept == x86_intercept_in ||
7157 info->intercept == x86_intercept_ins) {
7158 port = info->src_val;
7159 size = info->dst_bytes;
7161 port = info->dst_val;
7162 size = info->src_bytes;
7166 * If the 'use IO bitmaps' VM-execution control is 0, IO instruction
7167 * VM-exits depend on the 'unconditional IO exiting' VM-execution
7170 * Otherwise, IO instruction VM-exits are controlled by the IO bitmaps.
7172 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
7173 intercept = nested_cpu_has(vmcs12,
7174 CPU_BASED_UNCOND_IO_EXITING);
7176 intercept = nested_vmx_check_io_bitmaps(vcpu, port, size);
7178 /* FIXME: produce nested vmexit and return X86EMUL_INTERCEPTED. */
7179 return intercept ? X86EMUL_UNHANDLEABLE : X86EMUL_CONTINUE;
7182 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
7183 struct x86_instruction_info *info,
7184 enum x86_intercept_stage stage,
7185 struct x86_exception *exception)
7187 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7189 switch (info->intercept) {
7191 * RDPID causes #UD if disabled through secondary execution controls.
7192 * Because it is marked as EmulateOnUD, we need to intercept it here.
7194 case x86_intercept_rdtscp:
7195 if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDTSCP)) {
7196 exception->vector = UD_VECTOR;
7197 exception->error_code_valid = false;
7198 return X86EMUL_PROPAGATE_FAULT;
7202 case x86_intercept_in:
7203 case x86_intercept_ins:
7204 case x86_intercept_out:
7205 case x86_intercept_outs:
7206 return vmx_check_intercept_io(vcpu, info);
7208 case x86_intercept_lgdt:
7209 case x86_intercept_lidt:
7210 case x86_intercept_lldt:
7211 case x86_intercept_ltr:
7212 case x86_intercept_sgdt:
7213 case x86_intercept_sidt:
7214 case x86_intercept_sldt:
7215 case x86_intercept_str:
7216 if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC))
7217 return X86EMUL_CONTINUE;
7219 /* FIXME: produce nested vmexit and return X86EMUL_INTERCEPTED. */
7222 /* TODO: check more intercepts... */
7227 return X86EMUL_UNHANDLEABLE;
7230 #ifdef CONFIG_X86_64
7231 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
7232 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
7233 u64 divisor, u64 *result)
7235 u64 low = a << shift, high = a >> (64 - shift);
7237 /* To avoid the overflow on divq */
7238 if (high >= divisor)
7241 /* Low hold the result, high hold rem which is discarded */
7242 asm("divq %2\n\t" : "=a" (low), "=d" (high) :
7243 "rm" (divisor), "0" (low), "1" (high));
7249 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc,
7252 struct vcpu_vmx *vmx;
7253 u64 tscl, guest_tscl, delta_tsc, lapic_timer_advance_cycles;
7254 struct kvm_timer *ktimer = &vcpu->arch.apic->lapic_timer;
7256 if (kvm_mwait_in_guest(vcpu->kvm) ||
7257 kvm_can_post_timer_interrupt(vcpu))
7262 guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
7263 delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
7264 lapic_timer_advance_cycles = nsec_to_cycles(vcpu,
7265 ktimer->timer_advance_ns);
7267 if (delta_tsc > lapic_timer_advance_cycles)
7268 delta_tsc -= lapic_timer_advance_cycles;
7272 /* Convert to host delta tsc if tsc scaling is enabled */
7273 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
7274 delta_tsc && u64_shl_div_u64(delta_tsc,
7275 kvm_tsc_scaling_ratio_frac_bits,
7276 vcpu->arch.tsc_scaling_ratio, &delta_tsc))
7280 * If the delta tsc can't fit in the 32 bit after the multi shift,
7281 * we can't use the preemption timer.
7282 * It's possible that it fits on later vmentries, but checking
7283 * on every vmentry is costly so we just use an hrtimer.
7285 if (delta_tsc >> (cpu_preemption_timer_multi + 32))
7288 vmx->hv_deadline_tsc = tscl + delta_tsc;
7289 *expired = !delta_tsc;
7293 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
7295 to_vmx(vcpu)->hv_deadline_tsc = -1;
7299 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
7301 if (!kvm_pause_in_guest(vcpu->kvm))
7302 shrink_ple_window(vcpu);
7305 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
7306 struct kvm_memory_slot *slot)
7308 if (!kvm_dirty_log_manual_protect_and_init_set(kvm))
7309 kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
7310 kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
7313 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
7314 struct kvm_memory_slot *slot)
7316 kvm_mmu_slot_set_dirty(kvm, slot);
7319 static void vmx_flush_log_dirty(struct kvm *kvm)
7321 kvm_flush_pml_buffers(kvm);
7324 static int vmx_write_pml_buffer(struct kvm_vcpu *vcpu)
7326 struct vmcs12 *vmcs12;
7327 struct vcpu_vmx *vmx = to_vmx(vcpu);
7330 if (is_guest_mode(vcpu)) {
7331 WARN_ON_ONCE(vmx->nested.pml_full);
7334 * Check if PML is enabled for the nested guest.
7335 * Whether eptp bit 6 is set is already checked
7336 * as part of A/D emulation.
7338 vmcs12 = get_vmcs12(vcpu);
7339 if (!nested_cpu_has_pml(vmcs12))
7342 if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
7343 vmx->nested.pml_full = true;
7347 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS) & ~0xFFFull;
7348 dst = vmcs12->pml_address + sizeof(u64) * vmcs12->guest_pml_index;
7350 if (kvm_write_guest_page(vcpu->kvm, gpa_to_gfn(dst), &gpa,
7351 offset_in_page(dst), sizeof(gpa)))
7354 vmcs12->guest_pml_index--;
7360 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
7361 struct kvm_memory_slot *memslot,
7362 gfn_t offset, unsigned long mask)
7364 kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
7367 static void __pi_post_block(struct kvm_vcpu *vcpu)
7369 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7370 struct pi_desc old, new;
7374 old.control = new.control = pi_desc->control;
7375 WARN(old.nv != POSTED_INTR_WAKEUP_VECTOR,
7376 "Wakeup handler not enabled while the VCPU is blocked\n");
7378 dest = cpu_physical_id(vcpu->cpu);
7380 if (x2apic_enabled())
7383 new.ndst = (dest << 8) & 0xFF00;
7385 /* set 'NV' to 'notification vector' */
7386 new.nv = POSTED_INTR_VECTOR;
7387 } while (cmpxchg64(&pi_desc->control, old.control,
7388 new.control) != old.control);
7390 if (!WARN_ON_ONCE(vcpu->pre_pcpu == -1)) {
7391 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7392 list_del(&vcpu->blocked_vcpu_list);
7393 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7394 vcpu->pre_pcpu = -1;
7399 * This routine does the following things for vCPU which is going
7400 * to be blocked if VT-d PI is enabled.
7401 * - Store the vCPU to the wakeup list, so when interrupts happen
7402 * we can find the right vCPU to wake up.
7403 * - Change the Posted-interrupt descriptor as below:
7404 * 'NDST' <-- vcpu->pre_pcpu
7405 * 'NV' <-- POSTED_INTR_WAKEUP_VECTOR
7406 * - If 'ON' is set during this process, which means at least one
7407 * interrupt is posted for this vCPU, we cannot block it, in
7408 * this case, return 1, otherwise, return 0.
7411 static int pi_pre_block(struct kvm_vcpu *vcpu)
7414 struct pi_desc old, new;
7415 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7417 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
7418 !irq_remapping_cap(IRQ_POSTING_CAP) ||
7419 !kvm_vcpu_apicv_active(vcpu))
7422 WARN_ON(irqs_disabled());
7423 local_irq_disable();
7424 if (!WARN_ON_ONCE(vcpu->pre_pcpu != -1)) {
7425 vcpu->pre_pcpu = vcpu->cpu;
7426 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7427 list_add_tail(&vcpu->blocked_vcpu_list,
7428 &per_cpu(blocked_vcpu_on_cpu,
7430 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7434 old.control = new.control = pi_desc->control;
7436 WARN((pi_desc->sn == 1),
7437 "Warning: SN field of posted-interrupts "
7438 "is set before blocking\n");
7441 * Since vCPU can be preempted during this process,
7442 * vcpu->cpu could be different with pre_pcpu, we
7443 * need to set pre_pcpu as the destination of wakeup
7444 * notification event, then we can find the right vCPU
7445 * to wakeup in wakeup handler if interrupts happen
7446 * when the vCPU is in blocked state.
7448 dest = cpu_physical_id(vcpu->pre_pcpu);
7450 if (x2apic_enabled())
7453 new.ndst = (dest << 8) & 0xFF00;
7455 /* set 'NV' to 'wakeup vector' */
7456 new.nv = POSTED_INTR_WAKEUP_VECTOR;
7457 } while (cmpxchg64(&pi_desc->control, old.control,
7458 new.control) != old.control);
7460 /* We should not block the vCPU if an interrupt is posted for it. */
7461 if (pi_test_on(pi_desc) == 1)
7462 __pi_post_block(vcpu);
7465 return (vcpu->pre_pcpu == -1);
7468 static int vmx_pre_block(struct kvm_vcpu *vcpu)
7470 if (pi_pre_block(vcpu))
7473 if (kvm_lapic_hv_timer_in_use(vcpu))
7474 kvm_lapic_switch_to_sw_timer(vcpu);
7479 static void pi_post_block(struct kvm_vcpu *vcpu)
7481 if (vcpu->pre_pcpu == -1)
7484 WARN_ON(irqs_disabled());
7485 local_irq_disable();
7486 __pi_post_block(vcpu);
7490 static void vmx_post_block(struct kvm_vcpu *vcpu)
7492 if (kvm_x86_ops.set_hv_timer)
7493 kvm_lapic_switch_to_hv_timer(vcpu);
7495 pi_post_block(vcpu);
7499 * vmx_update_pi_irte - set IRTE for Posted-Interrupts
7502 * @host_irq: host irq of the interrupt
7503 * @guest_irq: gsi of the interrupt
7504 * @set: set or unset PI
7505 * returns 0 on success, < 0 on failure
7507 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
7508 uint32_t guest_irq, bool set)
7510 struct kvm_kernel_irq_routing_entry *e;
7511 struct kvm_irq_routing_table *irq_rt;
7512 struct kvm_lapic_irq irq;
7513 struct kvm_vcpu *vcpu;
7514 struct vcpu_data vcpu_info;
7517 if (!kvm_arch_has_assigned_device(kvm) ||
7518 !irq_remapping_cap(IRQ_POSTING_CAP) ||
7519 !kvm_vcpu_apicv_active(kvm->vcpus[0]))
7522 idx = srcu_read_lock(&kvm->irq_srcu);
7523 irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
7524 if (guest_irq >= irq_rt->nr_rt_entries ||
7525 hlist_empty(&irq_rt->map[guest_irq])) {
7526 pr_warn_once("no route for guest_irq %u/%u (broken user space?)\n",
7527 guest_irq, irq_rt->nr_rt_entries);
7531 hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
7532 if (e->type != KVM_IRQ_ROUTING_MSI)
7535 * VT-d PI cannot support posting multicast/broadcast
7536 * interrupts to a vCPU, we still use interrupt remapping
7537 * for these kind of interrupts.
7539 * For lowest-priority interrupts, we only support
7540 * those with single CPU as the destination, e.g. user
7541 * configures the interrupts via /proc/irq or uses
7542 * irqbalance to make the interrupts single-CPU.
7544 * We will support full lowest-priority interrupt later.
7546 * In addition, we can only inject generic interrupts using
7547 * the PI mechanism, refuse to route others through it.
7550 kvm_set_msi_irq(kvm, e, &irq);
7551 if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu) ||
7552 !kvm_irq_is_postable(&irq)) {
7554 * Make sure the IRTE is in remapped mode if
7555 * we don't handle it in posted mode.
7557 ret = irq_set_vcpu_affinity(host_irq, NULL);
7560 "failed to back to remapped mode, irq: %u\n",
7568 vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
7569 vcpu_info.vector = irq.vector;
7571 trace_kvm_pi_irte_update(host_irq, vcpu->vcpu_id, e->gsi,
7572 vcpu_info.vector, vcpu_info.pi_desc_addr, set);
7575 ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
7577 ret = irq_set_vcpu_affinity(host_irq, NULL);
7580 printk(KERN_INFO "%s: failed to update PI IRTE\n",
7588 srcu_read_unlock(&kvm->irq_srcu, idx);
7592 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
7594 if (vcpu->arch.mcg_cap & MCG_LMCE_P)
7595 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7596 FEAT_CTL_LMCE_ENABLED;
7598 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7599 ~FEAT_CTL_LMCE_ENABLED;
7602 static int vmx_smi_allowed(struct kvm_vcpu *vcpu)
7604 /* we need a nested vmexit to enter SMM, postpone if run is pending */
7605 if (to_vmx(vcpu)->nested.nested_run_pending)
7610 static int vmx_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
7612 struct vcpu_vmx *vmx = to_vmx(vcpu);
7614 vmx->nested.smm.guest_mode = is_guest_mode(vcpu);
7615 if (vmx->nested.smm.guest_mode)
7616 nested_vmx_vmexit(vcpu, -1, 0, 0);
7618 vmx->nested.smm.vmxon = vmx->nested.vmxon;
7619 vmx->nested.vmxon = false;
7620 vmx_clear_hlt(vcpu);
7624 static int vmx_pre_leave_smm(struct kvm_vcpu *vcpu, const char *smstate)
7626 struct vcpu_vmx *vmx = to_vmx(vcpu);
7629 if (vmx->nested.smm.vmxon) {
7630 vmx->nested.vmxon = true;
7631 vmx->nested.smm.vmxon = false;
7634 if (vmx->nested.smm.guest_mode) {
7635 ret = nested_vmx_enter_non_root_mode(vcpu, false);
7639 vmx->nested.smm.guest_mode = false;
7644 static int enable_smi_window(struct kvm_vcpu *vcpu)
7649 static bool vmx_need_emulation_on_page_fault(struct kvm_vcpu *vcpu)
7654 static bool vmx_apic_init_signal_blocked(struct kvm_vcpu *vcpu)
7656 return to_vmx(vcpu)->nested.vmxon;
7659 static void hardware_unsetup(void)
7662 nested_vmx_hardware_unsetup();
7667 static bool vmx_check_apicv_inhibit_reasons(ulong bit)
7669 ulong supported = BIT(APICV_INHIBIT_REASON_DISABLE) |
7670 BIT(APICV_INHIBIT_REASON_HYPERV);
7672 return supported & BIT(bit);
7675 static struct kvm_x86_ops vmx_x86_ops __initdata = {
7676 .hardware_unsetup = hardware_unsetup,
7678 .hardware_enable = hardware_enable,
7679 .hardware_disable = hardware_disable,
7680 .cpu_has_accelerated_tpr = report_flexpriority,
7681 .has_emulated_msr = vmx_has_emulated_msr,
7683 .vm_size = sizeof(struct kvm_vmx),
7684 .vm_init = vmx_vm_init,
7686 .vcpu_create = vmx_create_vcpu,
7687 .vcpu_free = vmx_free_vcpu,
7688 .vcpu_reset = vmx_vcpu_reset,
7690 .prepare_guest_switch = vmx_prepare_switch_to_guest,
7691 .vcpu_load = vmx_vcpu_load,
7692 .vcpu_put = vmx_vcpu_put,
7694 .update_bp_intercept = update_exception_bitmap,
7695 .get_msr_feature = vmx_get_msr_feature,
7696 .get_msr = vmx_get_msr,
7697 .set_msr = vmx_set_msr,
7698 .get_segment_base = vmx_get_segment_base,
7699 .get_segment = vmx_get_segment,
7700 .set_segment = vmx_set_segment,
7701 .get_cpl = vmx_get_cpl,
7702 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
7703 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
7704 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
7705 .set_cr0 = vmx_set_cr0,
7706 .set_cr4 = vmx_set_cr4,
7707 .set_efer = vmx_set_efer,
7708 .get_idt = vmx_get_idt,
7709 .set_idt = vmx_set_idt,
7710 .get_gdt = vmx_get_gdt,
7711 .set_gdt = vmx_set_gdt,
7712 .get_dr6 = vmx_get_dr6,
7713 .set_dr6 = vmx_set_dr6,
7714 .set_dr7 = vmx_set_dr7,
7715 .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
7716 .cache_reg = vmx_cache_reg,
7717 .get_rflags = vmx_get_rflags,
7718 .set_rflags = vmx_set_rflags,
7720 .tlb_flush = vmx_flush_tlb,
7721 .tlb_flush_gva = vmx_flush_tlb_gva,
7723 .run = vmx_vcpu_run,
7724 .handle_exit = vmx_handle_exit,
7725 .skip_emulated_instruction = vmx_skip_emulated_instruction,
7726 .update_emulated_instruction = vmx_update_emulated_instruction,
7727 .set_interrupt_shadow = vmx_set_interrupt_shadow,
7728 .get_interrupt_shadow = vmx_get_interrupt_shadow,
7729 .patch_hypercall = vmx_patch_hypercall,
7730 .set_irq = vmx_inject_irq,
7731 .set_nmi = vmx_inject_nmi,
7732 .queue_exception = vmx_queue_exception,
7733 .cancel_injection = vmx_cancel_injection,
7734 .interrupt_allowed = vmx_interrupt_allowed,
7735 .nmi_allowed = vmx_nmi_allowed,
7736 .get_nmi_mask = vmx_get_nmi_mask,
7737 .set_nmi_mask = vmx_set_nmi_mask,
7738 .enable_nmi_window = enable_nmi_window,
7739 .enable_irq_window = enable_irq_window,
7740 .update_cr8_intercept = update_cr8_intercept,
7741 .set_virtual_apic_mode = vmx_set_virtual_apic_mode,
7742 .set_apic_access_page_addr = vmx_set_apic_access_page_addr,
7743 .refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
7744 .load_eoi_exitmap = vmx_load_eoi_exitmap,
7745 .apicv_post_state_restore = vmx_apicv_post_state_restore,
7746 .check_apicv_inhibit_reasons = vmx_check_apicv_inhibit_reasons,
7747 .hwapic_irr_update = vmx_hwapic_irr_update,
7748 .hwapic_isr_update = vmx_hwapic_isr_update,
7749 .guest_apic_has_interrupt = vmx_guest_apic_has_interrupt,
7750 .sync_pir_to_irr = vmx_sync_pir_to_irr,
7751 .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
7752 .dy_apicv_has_pending_interrupt = vmx_dy_apicv_has_pending_interrupt,
7754 .set_tss_addr = vmx_set_tss_addr,
7755 .set_identity_map_addr = vmx_set_identity_map_addr,
7756 .get_tdp_level = get_ept_level,
7757 .get_mt_mask = vmx_get_mt_mask,
7759 .get_exit_info = vmx_get_exit_info,
7761 .cpuid_update = vmx_cpuid_update,
7763 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
7765 .read_l1_tsc_offset = vmx_read_l1_tsc_offset,
7766 .write_l1_tsc_offset = vmx_write_l1_tsc_offset,
7768 .load_mmu_pgd = vmx_load_mmu_pgd,
7770 .check_intercept = vmx_check_intercept,
7771 .handle_exit_irqoff = vmx_handle_exit_irqoff,
7773 .request_immediate_exit = vmx_request_immediate_exit,
7775 .sched_in = vmx_sched_in,
7777 .slot_enable_log_dirty = vmx_slot_enable_log_dirty,
7778 .slot_disable_log_dirty = vmx_slot_disable_log_dirty,
7779 .flush_log_dirty = vmx_flush_log_dirty,
7780 .enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
7781 .write_log_dirty = vmx_write_pml_buffer,
7783 .pre_block = vmx_pre_block,
7784 .post_block = vmx_post_block,
7786 .pmu_ops = &intel_pmu_ops,
7788 .update_pi_irte = vmx_update_pi_irte,
7790 #ifdef CONFIG_X86_64
7791 .set_hv_timer = vmx_set_hv_timer,
7792 .cancel_hv_timer = vmx_cancel_hv_timer,
7795 .setup_mce = vmx_setup_mce,
7797 .smi_allowed = vmx_smi_allowed,
7798 .pre_enter_smm = vmx_pre_enter_smm,
7799 .pre_leave_smm = vmx_pre_leave_smm,
7800 .enable_smi_window = enable_smi_window,
7802 .check_nested_events = NULL,
7803 .get_nested_state = NULL,
7804 .set_nested_state = NULL,
7805 .get_vmcs12_pages = NULL,
7806 .nested_enable_evmcs = NULL,
7807 .nested_get_evmcs_version = NULL,
7808 .need_emulation_on_page_fault = vmx_need_emulation_on_page_fault,
7809 .apic_init_signal_blocked = vmx_apic_init_signal_blocked,
7812 static __init int hardware_setup(void)
7814 unsigned long host_bndcfgs;
7816 int r, i, ept_lpage_level;
7819 host_idt_base = dt.address;
7821 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
7822 kvm_define_shared_msr(i, vmx_msr_index[i]);
7824 if (setup_vmcs_config(&vmcs_config, &vmx_capability) < 0)
7827 if (boot_cpu_has(X86_FEATURE_NX))
7828 kvm_enable_efer_bits(EFER_NX);
7830 if (boot_cpu_has(X86_FEATURE_MPX)) {
7831 rdmsrl(MSR_IA32_BNDCFGS, host_bndcfgs);
7832 WARN_ONCE(host_bndcfgs, "KVM: BNDCFGS in host will be lost");
7835 if (!cpu_has_vmx_mpx())
7836 supported_xcr0 &= ~(XFEATURE_MASK_BNDREGS |
7837 XFEATURE_MASK_BNDCSR);
7839 if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
7840 !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
7843 if (!cpu_has_vmx_ept() ||
7844 !cpu_has_vmx_ept_4levels() ||
7845 !cpu_has_vmx_ept_mt_wb() ||
7846 !cpu_has_vmx_invept_global())
7849 if (!cpu_has_vmx_ept_ad_bits() || !enable_ept)
7850 enable_ept_ad_bits = 0;
7852 if (!cpu_has_vmx_unrestricted_guest() || !enable_ept)
7853 enable_unrestricted_guest = 0;
7855 if (!cpu_has_vmx_flexpriority())
7856 flexpriority_enabled = 0;
7858 if (!cpu_has_virtual_nmis())
7862 * set_apic_access_page_addr() is used to reload apic access
7863 * page upon invalidation. No need to do anything if not
7864 * using the APIC_ACCESS_ADDR VMCS field.
7866 if (!flexpriority_enabled)
7867 vmx_x86_ops.set_apic_access_page_addr = NULL;
7869 if (!cpu_has_vmx_tpr_shadow())
7870 vmx_x86_ops.update_cr8_intercept = NULL;
7872 #if IS_ENABLED(CONFIG_HYPERV)
7873 if (ms_hyperv.nested_features & HV_X64_NESTED_GUEST_MAPPING_FLUSH
7875 vmx_x86_ops.tlb_remote_flush = hv_remote_flush_tlb;
7876 vmx_x86_ops.tlb_remote_flush_with_range =
7877 hv_remote_flush_tlb_with_range;
7881 if (!cpu_has_vmx_ple()) {
7884 ple_window_grow = 0;
7886 ple_window_shrink = 0;
7889 if (!cpu_has_vmx_apicv()) {
7891 vmx_x86_ops.sync_pir_to_irr = NULL;
7894 if (cpu_has_vmx_tsc_scaling()) {
7895 kvm_has_tsc_control = true;
7896 kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
7897 kvm_tsc_scaling_ratio_frac_bits = 48;
7900 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
7906 ept_lpage_level = 0;
7907 else if (cpu_has_vmx_ept_1g_page())
7908 ept_lpage_level = PT_PDPE_LEVEL;
7909 else if (cpu_has_vmx_ept_2m_page())
7910 ept_lpage_level = PT_DIRECTORY_LEVEL;
7912 ept_lpage_level = PT_PAGE_TABLE_LEVEL;
7913 kvm_configure_mmu(enable_ept, ept_lpage_level);
7916 * Only enable PML when hardware supports PML feature, and both EPT
7917 * and EPT A/D bit features are enabled -- PML depends on them to work.
7919 if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
7923 vmx_x86_ops.slot_enable_log_dirty = NULL;
7924 vmx_x86_ops.slot_disable_log_dirty = NULL;
7925 vmx_x86_ops.flush_log_dirty = NULL;
7926 vmx_x86_ops.enable_log_dirty_pt_masked = NULL;
7929 if (!cpu_has_vmx_preemption_timer())
7930 enable_preemption_timer = false;
7932 if (enable_preemption_timer) {
7933 u64 use_timer_freq = 5000ULL * 1000 * 1000;
7936 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
7937 cpu_preemption_timer_multi =
7938 vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
7941 use_timer_freq = (u64)tsc_khz * 1000;
7942 use_timer_freq >>= cpu_preemption_timer_multi;
7945 * KVM "disables" the preemption timer by setting it to its max
7946 * value. Don't use the timer if it might cause spurious exits
7947 * at a rate faster than 0.1 Hz (of uninterrupted guest time).
7949 if (use_timer_freq > 0xffffffffu / 10)
7950 enable_preemption_timer = false;
7953 if (!enable_preemption_timer) {
7954 vmx_x86_ops.set_hv_timer = NULL;
7955 vmx_x86_ops.cancel_hv_timer = NULL;
7956 vmx_x86_ops.request_immediate_exit = __kvm_request_immediate_exit;
7959 kvm_set_posted_intr_wakeup_handler(wakeup_handler);
7961 kvm_mce_cap_supported |= MCG_LMCE_P;
7963 if (pt_mode != PT_MODE_SYSTEM && pt_mode != PT_MODE_HOST_GUEST)
7965 if (!enable_ept || !cpu_has_vmx_intel_pt())
7966 pt_mode = PT_MODE_SYSTEM;
7969 nested_vmx_setup_ctls_msrs(&vmcs_config.nested,
7970 vmx_capability.ept);
7972 r = nested_vmx_hardware_setup(&vmx_x86_ops,
7973 kvm_vmx_exit_handlers);
7980 r = alloc_kvm_area();
7982 nested_vmx_hardware_unsetup();
7986 static struct kvm_x86_init_ops vmx_init_ops __initdata = {
7987 .cpu_has_kvm_support = cpu_has_kvm_support,
7988 .disabled_by_bios = vmx_disabled_by_bios,
7989 .check_processor_compatibility = vmx_check_processor_compat,
7990 .hardware_setup = hardware_setup,
7992 .runtime_ops = &vmx_x86_ops,
7995 static void vmx_cleanup_l1d_flush(void)
7997 if (vmx_l1d_flush_pages) {
7998 free_pages((unsigned long)vmx_l1d_flush_pages, L1D_CACHE_ORDER);
7999 vmx_l1d_flush_pages = NULL;
8001 /* Restore state so sysfs ignores VMX */
8002 l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
8005 static void vmx_exit(void)
8007 #ifdef CONFIG_KEXEC_CORE
8008 RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
8014 #if IS_ENABLED(CONFIG_HYPERV)
8015 if (static_branch_unlikely(&enable_evmcs)) {
8017 struct hv_vp_assist_page *vp_ap;
8019 * Reset everything to support using non-enlightened VMCS
8020 * access later (e.g. when we reload the module with
8021 * enlightened_vmcs=0)
8023 for_each_online_cpu(cpu) {
8024 vp_ap = hv_get_vp_assist_page(cpu);
8029 vp_ap->nested_control.features.directhypercall = 0;
8030 vp_ap->current_nested_vmcs = 0;
8031 vp_ap->enlighten_vmentry = 0;
8034 static_branch_disable(&enable_evmcs);
8037 vmx_cleanup_l1d_flush();
8039 module_exit(vmx_exit);
8041 static int __init vmx_init(void)
8045 #if IS_ENABLED(CONFIG_HYPERV)
8047 * Enlightened VMCS usage should be recommended and the host needs
8048 * to support eVMCS v1 or above. We can also disable eVMCS support
8049 * with module parameter.
8051 if (enlightened_vmcs &&
8052 ms_hyperv.hints & HV_X64_ENLIGHTENED_VMCS_RECOMMENDED &&
8053 (ms_hyperv.nested_features & HV_X64_ENLIGHTENED_VMCS_VERSION) >=
8054 KVM_EVMCS_VERSION) {
8057 /* Check that we have assist pages on all online CPUs */
8058 for_each_online_cpu(cpu) {
8059 if (!hv_get_vp_assist_page(cpu)) {
8060 enlightened_vmcs = false;
8065 if (enlightened_vmcs) {
8066 pr_info("KVM: vmx: using Hyper-V Enlightened VMCS\n");
8067 static_branch_enable(&enable_evmcs);
8070 if (ms_hyperv.nested_features & HV_X64_NESTED_DIRECT_FLUSH)
8071 vmx_x86_ops.enable_direct_tlbflush
8072 = hv_enable_direct_tlbflush;
8075 enlightened_vmcs = false;
8079 r = kvm_init(&vmx_init_ops, sizeof(struct vcpu_vmx),
8080 __alignof__(struct vcpu_vmx), THIS_MODULE);
8085 * Must be called after kvm_init() so enable_ept is properly set
8086 * up. Hand the parameter mitigation value in which was stored in
8087 * the pre module init parser. If no parameter was given, it will
8088 * contain 'auto' which will be turned into the default 'cond'
8091 r = vmx_setup_l1d_flush(vmentry_l1d_flush_param);
8097 for_each_possible_cpu(cpu) {
8098 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
8099 INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
8100 spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
8103 #ifdef CONFIG_KEXEC_CORE
8104 rcu_assign_pointer(crash_vmclear_loaded_vmcss,
8105 crash_vmclear_local_loaded_vmcss);
8107 vmx_check_vmcs12_offsets();
8111 module_init(vmx_init);