1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
5 * This module enables machines with Intel VT-x extensions to run virtual
6 * machines without emulation or binary translation.
8 * Copyright (C) 2006 Qumranet, Inc.
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
16 #include <linux/frame.h>
17 #include <linux/highmem.h>
18 #include <linux/hrtimer.h>
19 #include <linux/kernel.h>
20 #include <linux/kvm_host.h>
21 #include <linux/module.h>
22 #include <linux/moduleparam.h>
23 #include <linux/mod_devicetable.h>
25 #include <linux/sched.h>
26 #include <linux/sched/smt.h>
27 #include <linux/slab.h>
28 #include <linux/tboot.h>
29 #include <linux/trace_events.h>
34 #include <asm/cpu_device_id.h>
35 #include <asm/debugreg.h>
37 #include <asm/fpu/internal.h>
39 #include <asm/irq_remapping.h>
40 #include <asm/kexec.h>
41 #include <asm/perf_event.h>
43 #include <asm/mmu_context.h>
44 #include <asm/mshyperv.h>
45 #include <asm/mwait.h>
46 #include <asm/spec-ctrl.h>
47 #include <asm/virtext.h>
50 #include "capabilities.h"
54 #include "kvm_cache_regs.h"
66 MODULE_AUTHOR("Qumranet");
67 MODULE_LICENSE("GPL");
70 static const struct x86_cpu_id vmx_cpu_id[] = {
71 X86_MATCH_FEATURE(X86_FEATURE_VMX, NULL),
74 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
77 bool __read_mostly enable_vpid = 1;
78 module_param_named(vpid, enable_vpid, bool, 0444);
80 static bool __read_mostly enable_vnmi = 1;
81 module_param_named(vnmi, enable_vnmi, bool, S_IRUGO);
83 bool __read_mostly flexpriority_enabled = 1;
84 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
86 bool __read_mostly enable_ept = 1;
87 module_param_named(ept, enable_ept, bool, S_IRUGO);
89 bool __read_mostly enable_unrestricted_guest = 1;
90 module_param_named(unrestricted_guest,
91 enable_unrestricted_guest, bool, S_IRUGO);
93 bool __read_mostly enable_ept_ad_bits = 1;
94 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
96 static bool __read_mostly emulate_invalid_guest_state = true;
97 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
99 static bool __read_mostly fasteoi = 1;
100 module_param(fasteoi, bool, S_IRUGO);
102 bool __read_mostly enable_apicv = 1;
103 module_param(enable_apicv, bool, S_IRUGO);
106 * If nested=1, nested virtualization is supported, i.e., guests may use
107 * VMX and be a hypervisor for its own guests. If nested=0, guests may not
108 * use VMX instructions.
110 static bool __read_mostly nested = 1;
111 module_param(nested, bool, S_IRUGO);
113 bool __read_mostly enable_pml = 1;
114 module_param_named(pml, enable_pml, bool, S_IRUGO);
116 static bool __read_mostly dump_invalid_vmcs = 0;
117 module_param(dump_invalid_vmcs, bool, 0644);
119 #define MSR_BITMAP_MODE_X2APIC 1
120 #define MSR_BITMAP_MODE_X2APIC_APICV 2
122 #define KVM_VMX_TSC_MULTIPLIER_MAX 0xffffffffffffffffULL
124 /* Guest_tsc -> host_tsc conversion requires 64-bit division. */
125 static int __read_mostly cpu_preemption_timer_multi;
126 static bool __read_mostly enable_preemption_timer = 1;
128 module_param_named(preemption_timer, enable_preemption_timer, bool, S_IRUGO);
131 #define KVM_VM_CR0_ALWAYS_OFF (X86_CR0_NW | X86_CR0_CD)
132 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR0_NE
133 #define KVM_VM_CR0_ALWAYS_ON \
134 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | \
135 X86_CR0_WP | X86_CR0_PG | X86_CR0_PE)
136 #define KVM_CR4_GUEST_OWNED_BITS \
137 (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \
138 | X86_CR4_OSXMMEXCPT | X86_CR4_LA57 | X86_CR4_TSD)
140 #define KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR4_VMXE
141 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
142 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
144 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
146 #define MSR_IA32_RTIT_STATUS_MASK (~(RTIT_STATUS_FILTEREN | \
147 RTIT_STATUS_CONTEXTEN | RTIT_STATUS_TRIGGEREN | \
148 RTIT_STATUS_ERROR | RTIT_STATUS_STOPPED | \
149 RTIT_STATUS_BYTECNT))
151 #define MSR_IA32_RTIT_OUTPUT_BASE_MASK \
152 (~((1UL << cpuid_query_maxphyaddr(vcpu)) - 1) | 0x7f)
155 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
156 * ple_gap: upper bound on the amount of time between two successive
157 * executions of PAUSE in a loop. Also indicate if ple enabled.
158 * According to test, this time is usually smaller than 128 cycles.
159 * ple_window: upper bound on the amount of time a guest is allowed to execute
160 * in a PAUSE loop. Tests indicate that most spinlocks are held for
161 * less than 2^12 cycles
162 * Time is measured based on a counter that runs at the same rate as the TSC,
163 * refer SDM volume 3b section 21.6.13 & 22.1.3.
165 static unsigned int ple_gap = KVM_DEFAULT_PLE_GAP;
166 module_param(ple_gap, uint, 0444);
168 static unsigned int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
169 module_param(ple_window, uint, 0444);
171 /* Default doubles per-vcpu window every exit. */
172 static unsigned int ple_window_grow = KVM_DEFAULT_PLE_WINDOW_GROW;
173 module_param(ple_window_grow, uint, 0444);
175 /* Default resets per-vcpu window every exit to ple_window. */
176 static unsigned int ple_window_shrink = KVM_DEFAULT_PLE_WINDOW_SHRINK;
177 module_param(ple_window_shrink, uint, 0444);
179 /* Default is to compute the maximum so we can never overflow. */
180 static unsigned int ple_window_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
181 module_param(ple_window_max, uint, 0444);
183 /* Default is SYSTEM mode, 1 for host-guest mode */
184 int __read_mostly pt_mode = PT_MODE_SYSTEM;
185 module_param(pt_mode, int, S_IRUGO);
187 static DEFINE_STATIC_KEY_FALSE(vmx_l1d_should_flush);
188 static DEFINE_STATIC_KEY_FALSE(vmx_l1d_flush_cond);
189 static DEFINE_MUTEX(vmx_l1d_flush_mutex);
191 /* Storage for pre module init parameter parsing */
192 static enum vmx_l1d_flush_state __read_mostly vmentry_l1d_flush_param = VMENTER_L1D_FLUSH_AUTO;
194 static const struct {
197 } vmentry_l1d_param[] = {
198 [VMENTER_L1D_FLUSH_AUTO] = {"auto", true},
199 [VMENTER_L1D_FLUSH_NEVER] = {"never", true},
200 [VMENTER_L1D_FLUSH_COND] = {"cond", true},
201 [VMENTER_L1D_FLUSH_ALWAYS] = {"always", true},
202 [VMENTER_L1D_FLUSH_EPT_DISABLED] = {"EPT disabled", false},
203 [VMENTER_L1D_FLUSH_NOT_REQUIRED] = {"not required", false},
206 #define L1D_CACHE_ORDER 4
207 static void *vmx_l1d_flush_pages;
209 static int vmx_setup_l1d_flush(enum vmx_l1d_flush_state l1tf)
214 if (!boot_cpu_has_bug(X86_BUG_L1TF)) {
215 l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED;
220 l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_EPT_DISABLED;
224 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES)) {
227 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, msr);
228 if (msr & ARCH_CAP_SKIP_VMENTRY_L1DFLUSH) {
229 l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED;
234 /* If set to auto use the default l1tf mitigation method */
235 if (l1tf == VMENTER_L1D_FLUSH_AUTO) {
236 switch (l1tf_mitigation) {
237 case L1TF_MITIGATION_OFF:
238 l1tf = VMENTER_L1D_FLUSH_NEVER;
240 case L1TF_MITIGATION_FLUSH_NOWARN:
241 case L1TF_MITIGATION_FLUSH:
242 case L1TF_MITIGATION_FLUSH_NOSMT:
243 l1tf = VMENTER_L1D_FLUSH_COND;
245 case L1TF_MITIGATION_FULL:
246 case L1TF_MITIGATION_FULL_FORCE:
247 l1tf = VMENTER_L1D_FLUSH_ALWAYS;
250 } else if (l1tf_mitigation == L1TF_MITIGATION_FULL_FORCE) {
251 l1tf = VMENTER_L1D_FLUSH_ALWAYS;
254 if (l1tf != VMENTER_L1D_FLUSH_NEVER && !vmx_l1d_flush_pages &&
255 !boot_cpu_has(X86_FEATURE_FLUSH_L1D)) {
257 * This allocation for vmx_l1d_flush_pages is not tied to a VM
258 * lifetime and so should not be charged to a memcg.
260 page = alloc_pages(GFP_KERNEL, L1D_CACHE_ORDER);
263 vmx_l1d_flush_pages = page_address(page);
266 * Initialize each page with a different pattern in
267 * order to protect against KSM in the nested
268 * virtualization case.
270 for (i = 0; i < 1u << L1D_CACHE_ORDER; ++i) {
271 memset(vmx_l1d_flush_pages + i * PAGE_SIZE, i + 1,
276 l1tf_vmx_mitigation = l1tf;
278 if (l1tf != VMENTER_L1D_FLUSH_NEVER)
279 static_branch_enable(&vmx_l1d_should_flush);
281 static_branch_disable(&vmx_l1d_should_flush);
283 if (l1tf == VMENTER_L1D_FLUSH_COND)
284 static_branch_enable(&vmx_l1d_flush_cond);
286 static_branch_disable(&vmx_l1d_flush_cond);
290 static int vmentry_l1d_flush_parse(const char *s)
295 for (i = 0; i < ARRAY_SIZE(vmentry_l1d_param); i++) {
296 if (vmentry_l1d_param[i].for_parse &&
297 sysfs_streq(s, vmentry_l1d_param[i].option))
304 static int vmentry_l1d_flush_set(const char *s, const struct kernel_param *kp)
308 l1tf = vmentry_l1d_flush_parse(s);
312 if (!boot_cpu_has(X86_BUG_L1TF))
316 * Has vmx_init() run already? If not then this is the pre init
317 * parameter parsing. In that case just store the value and let
318 * vmx_init() do the proper setup after enable_ept has been
321 if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_AUTO) {
322 vmentry_l1d_flush_param = l1tf;
326 mutex_lock(&vmx_l1d_flush_mutex);
327 ret = vmx_setup_l1d_flush(l1tf);
328 mutex_unlock(&vmx_l1d_flush_mutex);
332 static int vmentry_l1d_flush_get(char *s, const struct kernel_param *kp)
334 if (WARN_ON_ONCE(l1tf_vmx_mitigation >= ARRAY_SIZE(vmentry_l1d_param)))
335 return sprintf(s, "???\n");
337 return sprintf(s, "%s\n", vmentry_l1d_param[l1tf_vmx_mitigation].option);
340 static const struct kernel_param_ops vmentry_l1d_flush_ops = {
341 .set = vmentry_l1d_flush_set,
342 .get = vmentry_l1d_flush_get,
344 module_param_cb(vmentry_l1d_flush, &vmentry_l1d_flush_ops, NULL, 0644);
346 static bool guest_state_valid(struct kvm_vcpu *vcpu);
347 static u32 vmx_segment_access_rights(struct kvm_segment *var);
348 static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
351 void vmx_vmexit(void);
353 #define vmx_insn_failed(fmt...) \
356 pr_warn_ratelimited(fmt); \
359 asmlinkage void vmread_error(unsigned long field, bool fault)
362 kvm_spurious_fault();
364 vmx_insn_failed("kvm: vmread failed: field=%lx\n", field);
367 noinline void vmwrite_error(unsigned long field, unsigned long value)
369 vmx_insn_failed("kvm: vmwrite failed: field=%lx val=%lx err=%d\n",
370 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
373 noinline void vmclear_error(struct vmcs *vmcs, u64 phys_addr)
375 vmx_insn_failed("kvm: vmclear failed: %p/%llx\n", vmcs, phys_addr);
378 noinline void vmptrld_error(struct vmcs *vmcs, u64 phys_addr)
380 vmx_insn_failed("kvm: vmptrld failed: %p/%llx\n", vmcs, phys_addr);
383 noinline void invvpid_error(unsigned long ext, u16 vpid, gva_t gva)
385 vmx_insn_failed("kvm: invvpid failed: ext=0x%lx vpid=%u gva=0x%lx\n",
389 noinline void invept_error(unsigned long ext, u64 eptp, gpa_t gpa)
391 vmx_insn_failed("kvm: invept failed: ext=0x%lx eptp=%llx gpa=0x%llx\n",
395 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
396 DEFINE_PER_CPU(struct vmcs *, current_vmcs);
398 * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
399 * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
401 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
404 * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
405 * can find which vCPU should be waken up.
407 static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu);
408 static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock);
410 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
411 static DEFINE_SPINLOCK(vmx_vpid_lock);
413 struct vmcs_config vmcs_config;
414 struct vmx_capability vmx_capability;
416 #define VMX_SEGMENT_FIELD(seg) \
417 [VCPU_SREG_##seg] = { \
418 .selector = GUEST_##seg##_SELECTOR, \
419 .base = GUEST_##seg##_BASE, \
420 .limit = GUEST_##seg##_LIMIT, \
421 .ar_bytes = GUEST_##seg##_AR_BYTES, \
424 static const struct kvm_vmx_segment_field {
429 } kvm_vmx_segment_fields[] = {
430 VMX_SEGMENT_FIELD(CS),
431 VMX_SEGMENT_FIELD(DS),
432 VMX_SEGMENT_FIELD(ES),
433 VMX_SEGMENT_FIELD(FS),
434 VMX_SEGMENT_FIELD(GS),
435 VMX_SEGMENT_FIELD(SS),
436 VMX_SEGMENT_FIELD(TR),
437 VMX_SEGMENT_FIELD(LDTR),
440 static inline void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
442 vmx->segment_cache.bitmask = 0;
445 static unsigned long host_idt_base;
448 * Though SYSCALL is only supported in 64-bit mode on Intel CPUs, kvm
449 * will emulate SYSCALL in legacy mode if the vendor string in guest
450 * CPUID.0:{EBX,ECX,EDX} is "AuthenticAMD" or "AMDisbetter!" To
451 * support this emulation, IA32_STAR must always be included in
452 * vmx_msr_index[], even in i386 builds.
454 const u32 vmx_msr_index[] = {
456 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
458 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
462 #if IS_ENABLED(CONFIG_HYPERV)
463 static bool __read_mostly enlightened_vmcs = true;
464 module_param(enlightened_vmcs, bool, 0444);
466 /* check_ept_pointer() should be under protection of ept_pointer_lock. */
467 static void check_ept_pointer_match(struct kvm *kvm)
469 struct kvm_vcpu *vcpu;
470 u64 tmp_eptp = INVALID_PAGE;
473 kvm_for_each_vcpu(i, vcpu, kvm) {
474 if (!VALID_PAGE(tmp_eptp)) {
475 tmp_eptp = to_vmx(vcpu)->ept_pointer;
476 } else if (tmp_eptp != to_vmx(vcpu)->ept_pointer) {
477 to_kvm_vmx(kvm)->ept_pointers_match
478 = EPT_POINTERS_MISMATCH;
483 to_kvm_vmx(kvm)->ept_pointers_match = EPT_POINTERS_MATCH;
486 static int kvm_fill_hv_flush_list_func(struct hv_guest_mapping_flush_list *flush,
489 struct kvm_tlb_range *range = data;
491 return hyperv_fill_flush_guest_mapping_list(flush, range->start_gfn,
495 static inline int __hv_remote_flush_tlb_with_range(struct kvm *kvm,
496 struct kvm_vcpu *vcpu, struct kvm_tlb_range *range)
498 u64 ept_pointer = to_vmx(vcpu)->ept_pointer;
501 * FLUSH_GUEST_PHYSICAL_ADDRESS_SPACE hypercall needs address
502 * of the base of EPT PML4 table, strip off EPT configuration
506 return hyperv_flush_guest_mapping_range(ept_pointer & PAGE_MASK,
507 kvm_fill_hv_flush_list_func, (void *)range);
509 return hyperv_flush_guest_mapping(ept_pointer & PAGE_MASK);
512 static int hv_remote_flush_tlb_with_range(struct kvm *kvm,
513 struct kvm_tlb_range *range)
515 struct kvm_vcpu *vcpu;
518 spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
520 if (to_kvm_vmx(kvm)->ept_pointers_match == EPT_POINTERS_CHECK)
521 check_ept_pointer_match(kvm);
523 if (to_kvm_vmx(kvm)->ept_pointers_match != EPT_POINTERS_MATCH) {
524 kvm_for_each_vcpu(i, vcpu, kvm) {
525 /* If ept_pointer is invalid pointer, bypass flush request. */
526 if (VALID_PAGE(to_vmx(vcpu)->ept_pointer))
527 ret |= __hv_remote_flush_tlb_with_range(
531 ret = __hv_remote_flush_tlb_with_range(kvm,
532 kvm_get_vcpu(kvm, 0), range);
535 spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
538 static int hv_remote_flush_tlb(struct kvm *kvm)
540 return hv_remote_flush_tlb_with_range(kvm, NULL);
543 static int hv_enable_direct_tlbflush(struct kvm_vcpu *vcpu)
545 struct hv_enlightened_vmcs *evmcs;
546 struct hv_partition_assist_pg **p_hv_pa_pg =
547 &vcpu->kvm->arch.hyperv.hv_pa_pg;
549 * Synthetic VM-Exit is not enabled in current code and so All
550 * evmcs in singe VM shares same assist page.
553 *p_hv_pa_pg = kzalloc(PAGE_SIZE, GFP_KERNEL);
558 evmcs = (struct hv_enlightened_vmcs *)to_vmx(vcpu)->loaded_vmcs->vmcs;
560 evmcs->partition_assist_page =
562 evmcs->hv_vm_id = (unsigned long)vcpu->kvm;
563 evmcs->hv_enlightenments_control.nested_flush_hypercall = 1;
568 #endif /* IS_ENABLED(CONFIG_HYPERV) */
571 * Comment's format: document - errata name - stepping - processor name.
573 * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
575 static u32 vmx_preemption_cpu_tfms[] = {
576 /* 323344.pdf - BA86 - D0 - Xeon 7500 Series */
578 /* 323056.pdf - AAX65 - C2 - Xeon L3406 */
579 /* 322814.pdf - AAT59 - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
580 /* 322911.pdf - AAU65 - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
582 /* 322911.pdf - AAU65 - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
584 /* 322373.pdf - AAO95 - B1 - Xeon 3400 Series */
585 /* 322166.pdf - AAN92 - B1 - i7-800 and i5-700 Desktop */
587 * 320767.pdf - AAP86 - B1 -
588 * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
591 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
593 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
595 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
597 /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
598 /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
599 /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
601 /* Xeon E3-1220 V2 */
605 static inline bool cpu_has_broken_vmx_preemption_timer(void)
607 u32 eax = cpuid_eax(0x00000001), i;
609 /* Clear the reserved bits */
610 eax &= ~(0x3U << 14 | 0xfU << 28);
611 for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
612 if (eax == vmx_preemption_cpu_tfms[i])
618 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
620 return flexpriority_enabled && lapic_in_kernel(vcpu);
623 static inline bool report_flexpriority(void)
625 return flexpriority_enabled;
628 static inline int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
632 for (i = 0; i < vmx->nmsrs; ++i)
633 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
638 struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
642 i = __find_msr_index(vmx, msr);
644 return &vmx->guest_msrs[i];
648 static int vmx_set_guest_msr(struct vcpu_vmx *vmx, struct shared_msr_entry *msr, u64 data)
652 u64 old_msr_data = msr->data;
654 if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
656 ret = kvm_set_shared_msr(msr->index, msr->data,
660 msr->data = old_msr_data;
665 #ifdef CONFIG_KEXEC_CORE
666 static void crash_vmclear_local_loaded_vmcss(void)
668 int cpu = raw_smp_processor_id();
669 struct loaded_vmcs *v;
671 list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
672 loaded_vmcss_on_cpu_link)
675 #endif /* CONFIG_KEXEC_CORE */
677 static void __loaded_vmcs_clear(void *arg)
679 struct loaded_vmcs *loaded_vmcs = arg;
680 int cpu = raw_smp_processor_id();
682 if (loaded_vmcs->cpu != cpu)
683 return; /* vcpu migration can race with cpu offline */
684 if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
685 per_cpu(current_vmcs, cpu) = NULL;
687 vmcs_clear(loaded_vmcs->vmcs);
688 if (loaded_vmcs->shadow_vmcs && loaded_vmcs->launched)
689 vmcs_clear(loaded_vmcs->shadow_vmcs);
691 list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
694 * Ensure all writes to loaded_vmcs, including deleting it from its
695 * current percpu list, complete before setting loaded_vmcs->vcpu to
696 * -1, otherwise a different cpu can see vcpu == -1 first and add
697 * loaded_vmcs to its percpu list before it's deleted from this cpu's
698 * list. Pairs with the smp_rmb() in vmx_vcpu_load_vmcs().
702 loaded_vmcs->cpu = -1;
703 loaded_vmcs->launched = 0;
706 void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
708 int cpu = loaded_vmcs->cpu;
711 smp_call_function_single(cpu,
712 __loaded_vmcs_clear, loaded_vmcs, 1);
715 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
719 u32 mask = 1 << (seg * SEG_FIELD_NR + field);
721 if (!kvm_register_is_available(&vmx->vcpu, VCPU_EXREG_SEGMENTS)) {
722 kvm_register_mark_available(&vmx->vcpu, VCPU_EXREG_SEGMENTS);
723 vmx->segment_cache.bitmask = 0;
725 ret = vmx->segment_cache.bitmask & mask;
726 vmx->segment_cache.bitmask |= mask;
730 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
732 u16 *p = &vmx->segment_cache.seg[seg].selector;
734 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
735 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
739 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
741 ulong *p = &vmx->segment_cache.seg[seg].base;
743 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
744 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
748 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
750 u32 *p = &vmx->segment_cache.seg[seg].limit;
752 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
753 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
757 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
759 u32 *p = &vmx->segment_cache.seg[seg].ar;
761 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
762 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
766 void update_exception_bitmap(struct kvm_vcpu *vcpu)
770 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
771 (1u << DB_VECTOR) | (1u << AC_VECTOR);
773 * Guest access to VMware backdoor ports could legitimately
774 * trigger #GP because of TSS I/O permission bitmap.
775 * We intercept those #GP and allow access to them anyway
778 if (enable_vmware_backdoor)
779 eb |= (1u << GP_VECTOR);
780 if ((vcpu->guest_debug &
781 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
782 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
783 eb |= 1u << BP_VECTOR;
784 if (to_vmx(vcpu)->rmode.vm86_active)
787 eb &= ~(1u << PF_VECTOR);
789 /* When we are running a nested L2 guest and L1 specified for it a
790 * certain exception bitmap, we must trap the same exceptions and pass
791 * them to L1. When running L2, we will only handle the exceptions
792 * specified above if L1 did not want them.
794 if (is_guest_mode(vcpu))
795 eb |= get_vmcs12(vcpu)->exception_bitmap;
797 vmcs_write32(EXCEPTION_BITMAP, eb);
801 * Check if MSR is intercepted for currently loaded MSR bitmap.
803 static bool msr_write_intercepted(struct kvm_vcpu *vcpu, u32 msr)
805 unsigned long *msr_bitmap;
806 int f = sizeof(unsigned long);
808 if (!cpu_has_vmx_msr_bitmap())
811 msr_bitmap = to_vmx(vcpu)->loaded_vmcs->msr_bitmap;
814 return !!test_bit(msr, msr_bitmap + 0x800 / f);
815 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
817 return !!test_bit(msr, msr_bitmap + 0xc00 / f);
823 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
824 unsigned long entry, unsigned long exit)
826 vm_entry_controls_clearbit(vmx, entry);
827 vm_exit_controls_clearbit(vmx, exit);
830 int vmx_find_msr_index(struct vmx_msrs *m, u32 msr)
834 for (i = 0; i < m->nr; ++i) {
835 if (m->val[i].index == msr)
841 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
844 struct msr_autoload *m = &vmx->msr_autoload;
848 if (cpu_has_load_ia32_efer()) {
849 clear_atomic_switch_msr_special(vmx,
850 VM_ENTRY_LOAD_IA32_EFER,
851 VM_EXIT_LOAD_IA32_EFER);
855 case MSR_CORE_PERF_GLOBAL_CTRL:
856 if (cpu_has_load_perf_global_ctrl()) {
857 clear_atomic_switch_msr_special(vmx,
858 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
859 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
864 i = vmx_find_msr_index(&m->guest, msr);
868 m->guest.val[i] = m->guest.val[m->guest.nr];
869 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
872 i = vmx_find_msr_index(&m->host, msr);
877 m->host.val[i] = m->host.val[m->host.nr];
878 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
881 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
882 unsigned long entry, unsigned long exit,
883 unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
884 u64 guest_val, u64 host_val)
886 vmcs_write64(guest_val_vmcs, guest_val);
887 if (host_val_vmcs != HOST_IA32_EFER)
888 vmcs_write64(host_val_vmcs, host_val);
889 vm_entry_controls_setbit(vmx, entry);
890 vm_exit_controls_setbit(vmx, exit);
893 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
894 u64 guest_val, u64 host_val, bool entry_only)
897 struct msr_autoload *m = &vmx->msr_autoload;
901 if (cpu_has_load_ia32_efer()) {
902 add_atomic_switch_msr_special(vmx,
903 VM_ENTRY_LOAD_IA32_EFER,
904 VM_EXIT_LOAD_IA32_EFER,
907 guest_val, host_val);
911 case MSR_CORE_PERF_GLOBAL_CTRL:
912 if (cpu_has_load_perf_global_ctrl()) {
913 add_atomic_switch_msr_special(vmx,
914 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
915 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
916 GUEST_IA32_PERF_GLOBAL_CTRL,
917 HOST_IA32_PERF_GLOBAL_CTRL,
918 guest_val, host_val);
922 case MSR_IA32_PEBS_ENABLE:
923 /* PEBS needs a quiescent period after being disabled (to write
924 * a record). Disabling PEBS through VMX MSR swapping doesn't
925 * provide that period, so a CPU could write host's record into
928 wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
931 i = vmx_find_msr_index(&m->guest, msr);
933 j = vmx_find_msr_index(&m->host, msr);
935 if ((i < 0 && m->guest.nr == NR_LOADSTORE_MSRS) ||
936 (j < 0 && m->host.nr == NR_LOADSTORE_MSRS)) {
937 printk_once(KERN_WARNING "Not enough msr switch entries. "
938 "Can't add msr %x\n", msr);
943 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
945 m->guest.val[i].index = msr;
946 m->guest.val[i].value = guest_val;
953 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
955 m->host.val[j].index = msr;
956 m->host.val[j].value = host_val;
959 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
961 u64 guest_efer = vmx->vcpu.arch.efer;
964 /* Shadow paging assumes NX to be available. */
966 guest_efer |= EFER_NX;
969 * LMA and LME handled by hardware; SCE meaningless outside long mode.
971 ignore_bits |= EFER_SCE;
973 ignore_bits |= EFER_LMA | EFER_LME;
974 /* SCE is meaningful only in long mode on Intel */
975 if (guest_efer & EFER_LMA)
976 ignore_bits &= ~(u64)EFER_SCE;
980 * On EPT, we can't emulate NX, so we must switch EFER atomically.
981 * On CPUs that support "load IA32_EFER", always switch EFER
982 * atomically, since it's faster than switching it manually.
984 if (cpu_has_load_ia32_efer() ||
985 (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
986 if (!(guest_efer & EFER_LMA))
987 guest_efer &= ~EFER_LME;
988 if (guest_efer != host_efer)
989 add_atomic_switch_msr(vmx, MSR_EFER,
990 guest_efer, host_efer, false);
992 clear_atomic_switch_msr(vmx, MSR_EFER);
995 clear_atomic_switch_msr(vmx, MSR_EFER);
997 guest_efer &= ~ignore_bits;
998 guest_efer |= host_efer & ignore_bits;
1000 vmx->guest_msrs[efer_offset].data = guest_efer;
1001 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1007 #ifdef CONFIG_X86_32
1009 * On 32-bit kernels, VM exits still load the FS and GS bases from the
1010 * VMCS rather than the segment table. KVM uses this helper to figure
1011 * out the current bases to poke them into the VMCS before entry.
1013 static unsigned long segment_base(u16 selector)
1015 struct desc_struct *table;
1018 if (!(selector & ~SEGMENT_RPL_MASK))
1021 table = get_current_gdt_ro();
1023 if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
1024 u16 ldt_selector = kvm_read_ldt();
1026 if (!(ldt_selector & ~SEGMENT_RPL_MASK))
1029 table = (struct desc_struct *)segment_base(ldt_selector);
1031 v = get_desc_base(&table[selector >> 3]);
1036 static inline bool pt_can_write_msr(struct vcpu_vmx *vmx)
1038 return vmx_pt_mode_is_host_guest() &&
1039 !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
1042 static inline void pt_load_msr(struct pt_ctx *ctx, u32 addr_range)
1046 wrmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1047 wrmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1048 wrmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1049 wrmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1050 for (i = 0; i < addr_range; i++) {
1051 wrmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1052 wrmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1056 static inline void pt_save_msr(struct pt_ctx *ctx, u32 addr_range)
1060 rdmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1061 rdmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1062 rdmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1063 rdmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1064 for (i = 0; i < addr_range; i++) {
1065 rdmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1066 rdmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1070 static void pt_guest_enter(struct vcpu_vmx *vmx)
1072 if (vmx_pt_mode_is_system())
1076 * GUEST_IA32_RTIT_CTL is already set in the VMCS.
1077 * Save host state before VM entry.
1079 rdmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1080 if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1081 wrmsrl(MSR_IA32_RTIT_CTL, 0);
1082 pt_save_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1083 pt_load_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1087 static void pt_guest_exit(struct vcpu_vmx *vmx)
1089 if (vmx_pt_mode_is_system())
1092 if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1093 pt_save_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1094 pt_load_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1097 /* Reload host state (IA32_RTIT_CTL will be cleared on VM exit). */
1098 wrmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1101 void vmx_set_host_fs_gs(struct vmcs_host_state *host, u16 fs_sel, u16 gs_sel,
1102 unsigned long fs_base, unsigned long gs_base)
1104 if (unlikely(fs_sel != host->fs_sel)) {
1106 vmcs_write16(HOST_FS_SELECTOR, fs_sel);
1108 vmcs_write16(HOST_FS_SELECTOR, 0);
1109 host->fs_sel = fs_sel;
1111 if (unlikely(gs_sel != host->gs_sel)) {
1113 vmcs_write16(HOST_GS_SELECTOR, gs_sel);
1115 vmcs_write16(HOST_GS_SELECTOR, 0);
1116 host->gs_sel = gs_sel;
1118 if (unlikely(fs_base != host->fs_base)) {
1119 vmcs_writel(HOST_FS_BASE, fs_base);
1120 host->fs_base = fs_base;
1122 if (unlikely(gs_base != host->gs_base)) {
1123 vmcs_writel(HOST_GS_BASE, gs_base);
1124 host->gs_base = gs_base;
1128 void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu)
1130 struct vcpu_vmx *vmx = to_vmx(vcpu);
1131 struct vmcs_host_state *host_state;
1132 #ifdef CONFIG_X86_64
1133 int cpu = raw_smp_processor_id();
1135 unsigned long fs_base, gs_base;
1139 vmx->req_immediate_exit = false;
1142 * Note that guest MSRs to be saved/restored can also be changed
1143 * when guest state is loaded. This happens when guest transitions
1144 * to/from long-mode by setting MSR_EFER.LMA.
1146 if (!vmx->guest_msrs_ready) {
1147 vmx->guest_msrs_ready = true;
1148 for (i = 0; i < vmx->save_nmsrs; ++i)
1149 kvm_set_shared_msr(vmx->guest_msrs[i].index,
1150 vmx->guest_msrs[i].data,
1151 vmx->guest_msrs[i].mask);
1155 if (vmx->nested.need_vmcs12_to_shadow_sync)
1156 nested_sync_vmcs12_to_shadow(vcpu);
1158 if (vmx->guest_state_loaded)
1161 host_state = &vmx->loaded_vmcs->host_state;
1164 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
1165 * allow segment selectors with cpl > 0 or ti == 1.
1167 host_state->ldt_sel = kvm_read_ldt();
1169 #ifdef CONFIG_X86_64
1170 savesegment(ds, host_state->ds_sel);
1171 savesegment(es, host_state->es_sel);
1173 gs_base = cpu_kernelmode_gs_base(cpu);
1174 if (likely(is_64bit_mm(current->mm))) {
1175 save_fsgs_for_kvm();
1176 fs_sel = current->thread.fsindex;
1177 gs_sel = current->thread.gsindex;
1178 fs_base = current->thread.fsbase;
1179 vmx->msr_host_kernel_gs_base = current->thread.gsbase;
1181 savesegment(fs, fs_sel);
1182 savesegment(gs, gs_sel);
1183 fs_base = read_msr(MSR_FS_BASE);
1184 vmx->msr_host_kernel_gs_base = read_msr(MSR_KERNEL_GS_BASE);
1187 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1189 savesegment(fs, fs_sel);
1190 savesegment(gs, gs_sel);
1191 fs_base = segment_base(fs_sel);
1192 gs_base = segment_base(gs_sel);
1195 vmx_set_host_fs_gs(host_state, fs_sel, gs_sel, fs_base, gs_base);
1196 vmx->guest_state_loaded = true;
1199 static void vmx_prepare_switch_to_host(struct vcpu_vmx *vmx)
1201 struct vmcs_host_state *host_state;
1203 if (!vmx->guest_state_loaded)
1206 host_state = &vmx->loaded_vmcs->host_state;
1208 ++vmx->vcpu.stat.host_state_reload;
1210 #ifdef CONFIG_X86_64
1211 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1213 if (host_state->ldt_sel || (host_state->gs_sel & 7)) {
1214 kvm_load_ldt(host_state->ldt_sel);
1215 #ifdef CONFIG_X86_64
1216 load_gs_index(host_state->gs_sel);
1218 loadsegment(gs, host_state->gs_sel);
1221 if (host_state->fs_sel & 7)
1222 loadsegment(fs, host_state->fs_sel);
1223 #ifdef CONFIG_X86_64
1224 if (unlikely(host_state->ds_sel | host_state->es_sel)) {
1225 loadsegment(ds, host_state->ds_sel);
1226 loadsegment(es, host_state->es_sel);
1229 invalidate_tss_limit();
1230 #ifdef CONFIG_X86_64
1231 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1233 load_fixmap_gdt(raw_smp_processor_id());
1234 vmx->guest_state_loaded = false;
1235 vmx->guest_msrs_ready = false;
1238 #ifdef CONFIG_X86_64
1239 static u64 vmx_read_guest_kernel_gs_base(struct vcpu_vmx *vmx)
1242 if (vmx->guest_state_loaded)
1243 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1245 return vmx->msr_guest_kernel_gs_base;
1248 static void vmx_write_guest_kernel_gs_base(struct vcpu_vmx *vmx, u64 data)
1251 if (vmx->guest_state_loaded)
1252 wrmsrl(MSR_KERNEL_GS_BASE, data);
1254 vmx->msr_guest_kernel_gs_base = data;
1258 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
1260 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1261 struct pi_desc old, new;
1265 * In case of hot-plug or hot-unplug, we may have to undo
1266 * vmx_vcpu_pi_put even if there is no assigned device. And we
1267 * always keep PI.NDST up to date for simplicity: it makes the
1268 * code easier, and CPU migration is not a fast path.
1270 if (!pi_test_sn(pi_desc) && vcpu->cpu == cpu)
1274 * If the 'nv' field is POSTED_INTR_WAKEUP_VECTOR, do not change
1275 * PI.NDST: pi_post_block is the one expected to change PID.NDST and the
1276 * wakeup handler expects the vCPU to be on the blocked_vcpu_list that
1277 * matches PI.NDST. Otherwise, a vcpu may not be able to be woken up
1280 if (pi_desc->nv == POSTED_INTR_WAKEUP_VECTOR || vcpu->cpu == cpu) {
1281 pi_clear_sn(pi_desc);
1282 goto after_clear_sn;
1285 /* The full case. */
1287 old.control = new.control = pi_desc->control;
1289 dest = cpu_physical_id(cpu);
1291 if (x2apic_enabled())
1294 new.ndst = (dest << 8) & 0xFF00;
1297 } while (cmpxchg64(&pi_desc->control, old.control,
1298 new.control) != old.control);
1303 * Clear SN before reading the bitmap. The VT-d firmware
1304 * writes the bitmap and reads SN atomically (5.2.3 in the
1305 * spec), so it doesn't really have a memory barrier that
1306 * pairs with this, but we cannot do that and we need one.
1308 smp_mb__after_atomic();
1310 if (!pi_is_pir_empty(pi_desc))
1314 void vmx_vcpu_load_vmcs(struct kvm_vcpu *vcpu, int cpu)
1316 struct vcpu_vmx *vmx = to_vmx(vcpu);
1317 bool already_loaded = vmx->loaded_vmcs->cpu == cpu;
1319 if (!already_loaded) {
1320 loaded_vmcs_clear(vmx->loaded_vmcs);
1321 local_irq_disable();
1324 * Ensure loaded_vmcs->cpu is read before adding loaded_vmcs to
1325 * this cpu's percpu list, otherwise it may not yet be deleted
1326 * from its previous cpu's percpu list. Pairs with the
1327 * smb_wmb() in __loaded_vmcs_clear().
1331 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1332 &per_cpu(loaded_vmcss_on_cpu, cpu));
1336 if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
1337 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1338 vmcs_load(vmx->loaded_vmcs->vmcs);
1339 indirect_branch_prediction_barrier();
1342 if (!already_loaded) {
1343 void *gdt = get_current_gdt_ro();
1344 unsigned long sysenter_esp;
1347 * Flush all EPTP/VPID contexts, the new pCPU may have stale
1348 * TLB entries from its previous association with the vCPU.
1350 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1353 * Linux uses per-cpu TSS and GDT, so set these when switching
1354 * processors. See 22.2.4.
1356 vmcs_writel(HOST_TR_BASE,
1357 (unsigned long)&get_cpu_entry_area(cpu)->tss.x86_tss);
1358 vmcs_writel(HOST_GDTR_BASE, (unsigned long)gdt); /* 22.2.4 */
1360 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1361 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1363 vmx->loaded_vmcs->cpu = cpu;
1366 /* Setup TSC multiplier */
1367 if (kvm_has_tsc_control &&
1368 vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio)
1369 decache_tsc_multiplier(vmx);
1373 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1374 * vcpu mutex is already taken.
1376 void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1378 struct vcpu_vmx *vmx = to_vmx(vcpu);
1380 vmx_vcpu_load_vmcs(vcpu, cpu);
1382 vmx_vcpu_pi_load(vcpu, cpu);
1384 vmx->host_pkru = read_pkru();
1385 vmx->host_debugctlmsr = get_debugctlmsr();
1388 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
1390 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1392 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
1393 !irq_remapping_cap(IRQ_POSTING_CAP) ||
1394 !kvm_vcpu_apicv_active(vcpu))
1397 /* Set SN when the vCPU is preempted */
1398 if (vcpu->preempted)
1402 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1404 vmx_vcpu_pi_put(vcpu);
1406 vmx_prepare_switch_to_host(to_vmx(vcpu));
1409 static bool emulation_required(struct kvm_vcpu *vcpu)
1411 return emulate_invalid_guest_state && !guest_state_valid(vcpu);
1414 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1416 struct vcpu_vmx *vmx = to_vmx(vcpu);
1417 unsigned long rflags, save_rflags;
1419 if (!kvm_register_is_available(vcpu, VCPU_EXREG_RFLAGS)) {
1420 kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1421 rflags = vmcs_readl(GUEST_RFLAGS);
1422 if (vmx->rmode.vm86_active) {
1423 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1424 save_rflags = vmx->rmode.save_rflags;
1425 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1427 vmx->rflags = rflags;
1432 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1434 struct vcpu_vmx *vmx = to_vmx(vcpu);
1435 unsigned long old_rflags;
1437 if (enable_unrestricted_guest) {
1438 kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1439 vmx->rflags = rflags;
1440 vmcs_writel(GUEST_RFLAGS, rflags);
1444 old_rflags = vmx_get_rflags(vcpu);
1445 vmx->rflags = rflags;
1446 if (vmx->rmode.vm86_active) {
1447 vmx->rmode.save_rflags = rflags;
1448 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1450 vmcs_writel(GUEST_RFLAGS, rflags);
1452 if ((old_rflags ^ vmx->rflags) & X86_EFLAGS_VM)
1453 vmx->emulation_required = emulation_required(vcpu);
1456 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
1458 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1461 if (interruptibility & GUEST_INTR_STATE_STI)
1462 ret |= KVM_X86_SHADOW_INT_STI;
1463 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1464 ret |= KVM_X86_SHADOW_INT_MOV_SS;
1469 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1471 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1472 u32 interruptibility = interruptibility_old;
1474 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1476 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1477 interruptibility |= GUEST_INTR_STATE_MOV_SS;
1478 else if (mask & KVM_X86_SHADOW_INT_STI)
1479 interruptibility |= GUEST_INTR_STATE_STI;
1481 if ((interruptibility != interruptibility_old))
1482 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1485 static int vmx_rtit_ctl_check(struct kvm_vcpu *vcpu, u64 data)
1487 struct vcpu_vmx *vmx = to_vmx(vcpu);
1488 unsigned long value;
1491 * Any MSR write that attempts to change bits marked reserved will
1494 if (data & vmx->pt_desc.ctl_bitmask)
1498 * Any attempt to modify IA32_RTIT_CTL while TraceEn is set will
1499 * result in a #GP unless the same write also clears TraceEn.
1501 if ((vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) &&
1502 ((vmx->pt_desc.guest.ctl ^ data) & ~RTIT_CTL_TRACEEN))
1506 * WRMSR to IA32_RTIT_CTL that sets TraceEn but clears this bit
1507 * and FabricEn would cause #GP, if
1508 * CPUID.(EAX=14H, ECX=0):ECX.SNGLRGNOUT[bit 2] = 0
1510 if ((data & RTIT_CTL_TRACEEN) && !(data & RTIT_CTL_TOPA) &&
1511 !(data & RTIT_CTL_FABRIC_EN) &&
1512 !intel_pt_validate_cap(vmx->pt_desc.caps,
1513 PT_CAP_single_range_output))
1517 * MTCFreq, CycThresh and PSBFreq encodings check, any MSR write that
1518 * utilize encodings marked reserved will casue a #GP fault.
1520 value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc_periods);
1521 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc) &&
1522 !test_bit((data & RTIT_CTL_MTC_RANGE) >>
1523 RTIT_CTL_MTC_RANGE_OFFSET, &value))
1525 value = intel_pt_validate_cap(vmx->pt_desc.caps,
1526 PT_CAP_cycle_thresholds);
1527 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1528 !test_bit((data & RTIT_CTL_CYC_THRESH) >>
1529 RTIT_CTL_CYC_THRESH_OFFSET, &value))
1531 value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_periods);
1532 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1533 !test_bit((data & RTIT_CTL_PSB_FREQ) >>
1534 RTIT_CTL_PSB_FREQ_OFFSET, &value))
1538 * If ADDRx_CFG is reserved or the encodings is >2 will
1539 * cause a #GP fault.
1541 value = (data & RTIT_CTL_ADDR0) >> RTIT_CTL_ADDR0_OFFSET;
1542 if ((value && (vmx->pt_desc.addr_range < 1)) || (value > 2))
1544 value = (data & RTIT_CTL_ADDR1) >> RTIT_CTL_ADDR1_OFFSET;
1545 if ((value && (vmx->pt_desc.addr_range < 2)) || (value > 2))
1547 value = (data & RTIT_CTL_ADDR2) >> RTIT_CTL_ADDR2_OFFSET;
1548 if ((value && (vmx->pt_desc.addr_range < 3)) || (value > 2))
1550 value = (data & RTIT_CTL_ADDR3) >> RTIT_CTL_ADDR3_OFFSET;
1551 if ((value && (vmx->pt_desc.addr_range < 4)) || (value > 2))
1557 static int skip_emulated_instruction(struct kvm_vcpu *vcpu)
1562 * Using VMCS.VM_EXIT_INSTRUCTION_LEN on EPT misconfig depends on
1563 * undefined behavior: Intel's SDM doesn't mandate the VMCS field be
1564 * set when EPT misconfig occurs. In practice, real hardware updates
1565 * VM_EXIT_INSTRUCTION_LEN on EPT misconfig, but other hypervisors
1566 * (namely Hyper-V) don't set it due to it being undefined behavior,
1567 * i.e. we end up advancing IP with some random value.
1569 if (!static_cpu_has(X86_FEATURE_HYPERVISOR) ||
1570 to_vmx(vcpu)->exit_reason != EXIT_REASON_EPT_MISCONFIG) {
1571 rip = kvm_rip_read(vcpu);
1572 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1573 kvm_rip_write(vcpu, rip);
1575 if (!kvm_emulate_instruction(vcpu, EMULTYPE_SKIP))
1579 /* skipping an emulated instruction also counts */
1580 vmx_set_interrupt_shadow(vcpu, 0);
1587 * Recognizes a pending MTF VM-exit and records the nested state for later
1590 static void vmx_update_emulated_instruction(struct kvm_vcpu *vcpu)
1592 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1593 struct vcpu_vmx *vmx = to_vmx(vcpu);
1595 if (!is_guest_mode(vcpu))
1599 * Per the SDM, MTF takes priority over debug-trap exceptions besides
1600 * T-bit traps. As instruction emulation is completed (i.e. at the
1601 * instruction boundary), any #DB exception pending delivery must be a
1602 * debug-trap. Record the pending MTF state to be delivered in
1603 * vmx_check_nested_events().
1605 if (nested_cpu_has_mtf(vmcs12) &&
1606 (!vcpu->arch.exception.pending ||
1607 vcpu->arch.exception.nr == DB_VECTOR))
1608 vmx->nested.mtf_pending = true;
1610 vmx->nested.mtf_pending = false;
1613 static int vmx_skip_emulated_instruction(struct kvm_vcpu *vcpu)
1615 vmx_update_emulated_instruction(vcpu);
1616 return skip_emulated_instruction(vcpu);
1619 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1622 * Ensure that we clear the HLT state in the VMCS. We don't need to
1623 * explicitly skip the instruction because if the HLT state is set,
1624 * then the instruction is already executing and RIP has already been
1627 if (kvm_hlt_in_guest(vcpu->kvm) &&
1628 vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1629 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1632 static void vmx_queue_exception(struct kvm_vcpu *vcpu)
1634 struct vcpu_vmx *vmx = to_vmx(vcpu);
1635 unsigned nr = vcpu->arch.exception.nr;
1636 bool has_error_code = vcpu->arch.exception.has_error_code;
1637 u32 error_code = vcpu->arch.exception.error_code;
1638 u32 intr_info = nr | INTR_INFO_VALID_MASK;
1640 kvm_deliver_exception_payload(vcpu);
1642 if (has_error_code) {
1643 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1644 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1647 if (vmx->rmode.vm86_active) {
1649 if (kvm_exception_is_soft(nr))
1650 inc_eip = vcpu->arch.event_exit_inst_len;
1651 kvm_inject_realmode_interrupt(vcpu, nr, inc_eip);
1655 WARN_ON_ONCE(vmx->emulation_required);
1657 if (kvm_exception_is_soft(nr)) {
1658 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1659 vmx->vcpu.arch.event_exit_inst_len);
1660 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1662 intr_info |= INTR_TYPE_HARD_EXCEPTION;
1664 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1666 vmx_clear_hlt(vcpu);
1670 * Swap MSR entry in host/guest MSR entry array.
1672 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1674 struct shared_msr_entry tmp;
1676 tmp = vmx->guest_msrs[to];
1677 vmx->guest_msrs[to] = vmx->guest_msrs[from];
1678 vmx->guest_msrs[from] = tmp;
1682 * Set up the vmcs to automatically save and restore system
1683 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
1684 * mode, as fiddling with msrs is very expensive.
1686 static void setup_msrs(struct vcpu_vmx *vmx)
1688 int save_nmsrs, index;
1691 #ifdef CONFIG_X86_64
1693 * The SYSCALL MSRs are only needed on long mode guests, and only
1694 * when EFER.SCE is set.
1696 if (is_long_mode(&vmx->vcpu) && (vmx->vcpu.arch.efer & EFER_SCE)) {
1697 index = __find_msr_index(vmx, MSR_STAR);
1699 move_msr_up(vmx, index, save_nmsrs++);
1700 index = __find_msr_index(vmx, MSR_LSTAR);
1702 move_msr_up(vmx, index, save_nmsrs++);
1703 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1705 move_msr_up(vmx, index, save_nmsrs++);
1708 index = __find_msr_index(vmx, MSR_EFER);
1709 if (index >= 0 && update_transition_efer(vmx, index))
1710 move_msr_up(vmx, index, save_nmsrs++);
1711 index = __find_msr_index(vmx, MSR_TSC_AUX);
1712 if (index >= 0 && guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDTSCP))
1713 move_msr_up(vmx, index, save_nmsrs++);
1714 index = __find_msr_index(vmx, MSR_IA32_TSX_CTRL);
1716 move_msr_up(vmx, index, save_nmsrs++);
1718 vmx->save_nmsrs = save_nmsrs;
1719 vmx->guest_msrs_ready = false;
1721 if (cpu_has_vmx_msr_bitmap())
1722 vmx_update_msr_bitmap(&vmx->vcpu);
1725 static u64 vmx_read_l1_tsc_offset(struct kvm_vcpu *vcpu)
1727 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1729 if (is_guest_mode(vcpu) &&
1730 (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING))
1731 return vcpu->arch.tsc_offset - vmcs12->tsc_offset;
1733 return vcpu->arch.tsc_offset;
1736 static u64 vmx_write_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1738 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1739 u64 g_tsc_offset = 0;
1742 * We're here if L1 chose not to trap WRMSR to TSC. According
1743 * to the spec, this should set L1's TSC; The offset that L1
1744 * set for L2 remains unchanged, and still needs to be added
1745 * to the newly set TSC to get L2's TSC.
1747 if (is_guest_mode(vcpu) &&
1748 (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING))
1749 g_tsc_offset = vmcs12->tsc_offset;
1751 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1752 vcpu->arch.tsc_offset - g_tsc_offset,
1754 vmcs_write64(TSC_OFFSET, offset + g_tsc_offset);
1755 return offset + g_tsc_offset;
1759 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
1760 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
1761 * all guests if the "nested" module option is off, and can also be disabled
1762 * for a single guest by disabling its VMX cpuid bit.
1764 bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
1766 return nested && guest_cpuid_has(vcpu, X86_FEATURE_VMX);
1769 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
1772 uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
1774 return !(val & ~valid_bits);
1777 static int vmx_get_msr_feature(struct kvm_msr_entry *msr)
1779 switch (msr->index) {
1780 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1783 return vmx_get_vmx_msr(&vmcs_config.nested, msr->index, &msr->data);
1790 * Reads an msr value (of 'msr_index') into 'pdata'.
1791 * Returns 0 on success, non-0 otherwise.
1792 * Assumes vcpu_load() was already called.
1794 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1796 struct vcpu_vmx *vmx = to_vmx(vcpu);
1797 struct shared_msr_entry *msr;
1800 switch (msr_info->index) {
1801 #ifdef CONFIG_X86_64
1803 msr_info->data = vmcs_readl(GUEST_FS_BASE);
1806 msr_info->data = vmcs_readl(GUEST_GS_BASE);
1808 case MSR_KERNEL_GS_BASE:
1809 msr_info->data = vmx_read_guest_kernel_gs_base(vmx);
1813 return kvm_get_msr_common(vcpu, msr_info);
1814 case MSR_IA32_TSX_CTRL:
1815 if (!msr_info->host_initiated &&
1816 !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
1818 goto find_shared_msr;
1819 case MSR_IA32_UMWAIT_CONTROL:
1820 if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
1823 msr_info->data = vmx->msr_ia32_umwait_control;
1825 case MSR_IA32_SPEC_CTRL:
1826 if (!msr_info->host_initiated &&
1827 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1830 msr_info->data = to_vmx(vcpu)->spec_ctrl;
1832 case MSR_IA32_SYSENTER_CS:
1833 msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
1835 case MSR_IA32_SYSENTER_EIP:
1836 msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
1838 case MSR_IA32_SYSENTER_ESP:
1839 msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
1841 case MSR_IA32_BNDCFGS:
1842 if (!kvm_mpx_supported() ||
1843 (!msr_info->host_initiated &&
1844 !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1846 msr_info->data = vmcs_read64(GUEST_BNDCFGS);
1848 case MSR_IA32_MCG_EXT_CTL:
1849 if (!msr_info->host_initiated &&
1850 !(vmx->msr_ia32_feature_control &
1851 FEAT_CTL_LMCE_ENABLED))
1853 msr_info->data = vcpu->arch.mcg_ext_ctl;
1855 case MSR_IA32_FEAT_CTL:
1856 msr_info->data = vmx->msr_ia32_feature_control;
1858 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1859 if (!nested_vmx_allowed(vcpu))
1861 if (vmx_get_vmx_msr(&vmx->nested.msrs, msr_info->index,
1865 * Enlightened VMCS v1 doesn't have certain fields, but buggy
1866 * Hyper-V versions are still trying to use corresponding
1867 * features when they are exposed. Filter out the essential
1870 if (!msr_info->host_initiated &&
1871 vmx->nested.enlightened_vmcs_enabled)
1872 nested_evmcs_filter_control_msr(msr_info->index,
1875 case MSR_IA32_RTIT_CTL:
1876 if (!vmx_pt_mode_is_host_guest())
1878 msr_info->data = vmx->pt_desc.guest.ctl;
1880 case MSR_IA32_RTIT_STATUS:
1881 if (!vmx_pt_mode_is_host_guest())
1883 msr_info->data = vmx->pt_desc.guest.status;
1885 case MSR_IA32_RTIT_CR3_MATCH:
1886 if (!vmx_pt_mode_is_host_guest() ||
1887 !intel_pt_validate_cap(vmx->pt_desc.caps,
1888 PT_CAP_cr3_filtering))
1890 msr_info->data = vmx->pt_desc.guest.cr3_match;
1892 case MSR_IA32_RTIT_OUTPUT_BASE:
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_base;
1901 case MSR_IA32_RTIT_OUTPUT_MASK:
1902 if (!vmx_pt_mode_is_host_guest() ||
1903 (!intel_pt_validate_cap(vmx->pt_desc.caps,
1904 PT_CAP_topa_output) &&
1905 !intel_pt_validate_cap(vmx->pt_desc.caps,
1906 PT_CAP_single_range_output)))
1908 msr_info->data = vmx->pt_desc.guest.output_mask;
1910 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
1911 index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
1912 if (!vmx_pt_mode_is_host_guest() ||
1913 (index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
1914 PT_CAP_num_address_ranges)))
1917 msr_info->data = vmx->pt_desc.guest.addr_b[index / 2];
1919 msr_info->data = vmx->pt_desc.guest.addr_a[index / 2];
1922 if (!msr_info->host_initiated &&
1923 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
1925 goto find_shared_msr;
1928 msr = find_msr_entry(vmx, msr_info->index);
1930 msr_info->data = msr->data;
1933 return kvm_get_msr_common(vcpu, msr_info);
1940 * Writes msr value into the appropriate "register".
1941 * Returns 0 on success, non-0 otherwise.
1942 * Assumes vcpu_load() was already called.
1944 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1946 struct vcpu_vmx *vmx = to_vmx(vcpu);
1947 struct shared_msr_entry *msr;
1949 u32 msr_index = msr_info->index;
1950 u64 data = msr_info->data;
1953 switch (msr_index) {
1955 ret = kvm_set_msr_common(vcpu, msr_info);
1957 #ifdef CONFIG_X86_64
1959 vmx_segment_cache_clear(vmx);
1960 vmcs_writel(GUEST_FS_BASE, data);
1963 vmx_segment_cache_clear(vmx);
1964 vmcs_writel(GUEST_GS_BASE, data);
1966 case MSR_KERNEL_GS_BASE:
1967 vmx_write_guest_kernel_gs_base(vmx, data);
1970 case MSR_IA32_SYSENTER_CS:
1971 if (is_guest_mode(vcpu))
1972 get_vmcs12(vcpu)->guest_sysenter_cs = data;
1973 vmcs_write32(GUEST_SYSENTER_CS, data);
1975 case MSR_IA32_SYSENTER_EIP:
1976 if (is_guest_mode(vcpu))
1977 get_vmcs12(vcpu)->guest_sysenter_eip = data;
1978 vmcs_writel(GUEST_SYSENTER_EIP, data);
1980 case MSR_IA32_SYSENTER_ESP:
1981 if (is_guest_mode(vcpu))
1982 get_vmcs12(vcpu)->guest_sysenter_esp = data;
1983 vmcs_writel(GUEST_SYSENTER_ESP, data);
1985 case MSR_IA32_DEBUGCTLMSR:
1986 if (is_guest_mode(vcpu) && get_vmcs12(vcpu)->vm_exit_controls &
1987 VM_EXIT_SAVE_DEBUG_CONTROLS)
1988 get_vmcs12(vcpu)->guest_ia32_debugctl = data;
1990 ret = kvm_set_msr_common(vcpu, msr_info);
1993 case MSR_IA32_BNDCFGS:
1994 if (!kvm_mpx_supported() ||
1995 (!msr_info->host_initiated &&
1996 !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1998 if (is_noncanonical_address(data & PAGE_MASK, vcpu) ||
1999 (data & MSR_IA32_BNDCFGS_RSVD))
2001 vmcs_write64(GUEST_BNDCFGS, data);
2003 case MSR_IA32_UMWAIT_CONTROL:
2004 if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
2007 /* The reserved bit 1 and non-32 bit [63:32] should be zero */
2008 if (data & (BIT_ULL(1) | GENMASK_ULL(63, 32)))
2011 vmx->msr_ia32_umwait_control = data;
2013 case MSR_IA32_SPEC_CTRL:
2014 if (!msr_info->host_initiated &&
2015 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
2018 if (data & ~kvm_spec_ctrl_valid_bits(vcpu))
2021 vmx->spec_ctrl = data;
2027 * When it's written (to non-zero) for the first time, pass
2031 * The handling of the MSR bitmap for L2 guests is done in
2032 * nested_vmx_prepare_msr_bitmap. We should not touch the
2033 * vmcs02.msr_bitmap here since it gets completely overwritten
2034 * in the merging. We update the vmcs01 here for L1 as well
2035 * since it will end up touching the MSR anyway now.
2037 vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap,
2041 case MSR_IA32_TSX_CTRL:
2042 if (!msr_info->host_initiated &&
2043 !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
2045 if (data & ~(TSX_CTRL_RTM_DISABLE | TSX_CTRL_CPUID_CLEAR))
2047 goto find_shared_msr;
2048 case MSR_IA32_PRED_CMD:
2049 if (!msr_info->host_initiated &&
2050 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
2053 if (data & ~PRED_CMD_IBPB)
2055 if (!boot_cpu_has(X86_FEATURE_SPEC_CTRL))
2060 wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
2064 * When it's written (to non-zero) for the first time, pass
2068 * The handling of the MSR bitmap for L2 guests is done in
2069 * nested_vmx_prepare_msr_bitmap. We should not touch the
2070 * vmcs02.msr_bitmap here since it gets completely overwritten
2073 vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap, MSR_IA32_PRED_CMD,
2076 case MSR_IA32_CR_PAT:
2077 if (!kvm_pat_valid(data))
2080 if (is_guest_mode(vcpu) &&
2081 get_vmcs12(vcpu)->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
2082 get_vmcs12(vcpu)->guest_ia32_pat = data;
2084 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2085 vmcs_write64(GUEST_IA32_PAT, data);
2086 vcpu->arch.pat = data;
2089 ret = kvm_set_msr_common(vcpu, msr_info);
2091 case MSR_IA32_TSC_ADJUST:
2092 ret = kvm_set_msr_common(vcpu, msr_info);
2094 case MSR_IA32_MCG_EXT_CTL:
2095 if ((!msr_info->host_initiated &&
2096 !(to_vmx(vcpu)->msr_ia32_feature_control &
2097 FEAT_CTL_LMCE_ENABLED)) ||
2098 (data & ~MCG_EXT_CTL_LMCE_EN))
2100 vcpu->arch.mcg_ext_ctl = data;
2102 case MSR_IA32_FEAT_CTL:
2103 if (!vmx_feature_control_msr_valid(vcpu, data) ||
2104 (to_vmx(vcpu)->msr_ia32_feature_control &
2105 FEAT_CTL_LOCKED && !msr_info->host_initiated))
2107 vmx->msr_ia32_feature_control = data;
2108 if (msr_info->host_initiated && data == 0)
2109 vmx_leave_nested(vcpu);
2111 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2112 if (!msr_info->host_initiated)
2113 return 1; /* they are read-only */
2114 if (!nested_vmx_allowed(vcpu))
2116 return vmx_set_vmx_msr(vcpu, msr_index, data);
2117 case MSR_IA32_RTIT_CTL:
2118 if (!vmx_pt_mode_is_host_guest() ||
2119 vmx_rtit_ctl_check(vcpu, data) ||
2122 vmcs_write64(GUEST_IA32_RTIT_CTL, data);
2123 vmx->pt_desc.guest.ctl = data;
2124 pt_update_intercept_for_msr(vmx);
2126 case MSR_IA32_RTIT_STATUS:
2127 if (!pt_can_write_msr(vmx))
2129 if (data & MSR_IA32_RTIT_STATUS_MASK)
2131 vmx->pt_desc.guest.status = data;
2133 case MSR_IA32_RTIT_CR3_MATCH:
2134 if (!pt_can_write_msr(vmx))
2136 if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2137 PT_CAP_cr3_filtering))
2139 vmx->pt_desc.guest.cr3_match = data;
2141 case MSR_IA32_RTIT_OUTPUT_BASE:
2142 if (!pt_can_write_msr(vmx))
2144 if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2145 PT_CAP_topa_output) &&
2146 !intel_pt_validate_cap(vmx->pt_desc.caps,
2147 PT_CAP_single_range_output))
2149 if (data & MSR_IA32_RTIT_OUTPUT_BASE_MASK)
2151 vmx->pt_desc.guest.output_base = data;
2153 case MSR_IA32_RTIT_OUTPUT_MASK:
2154 if (!pt_can_write_msr(vmx))
2156 if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2157 PT_CAP_topa_output) &&
2158 !intel_pt_validate_cap(vmx->pt_desc.caps,
2159 PT_CAP_single_range_output))
2161 vmx->pt_desc.guest.output_mask = data;
2163 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
2164 if (!pt_can_write_msr(vmx))
2166 index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
2167 if (index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
2168 PT_CAP_num_address_ranges))
2170 if (is_noncanonical_address(data, vcpu))
2173 vmx->pt_desc.guest.addr_b[index / 2] = data;
2175 vmx->pt_desc.guest.addr_a[index / 2] = data;
2178 if (!msr_info->host_initiated &&
2179 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
2181 /* Check reserved bit, higher 32 bits should be zero */
2182 if ((data >> 32) != 0)
2184 goto find_shared_msr;
2188 msr = find_msr_entry(vmx, msr_index);
2190 ret = vmx_set_guest_msr(vmx, msr, data);
2192 ret = kvm_set_msr_common(vcpu, msr_info);
2198 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2200 kvm_register_mark_available(vcpu, reg);
2204 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2207 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2209 case VCPU_EXREG_PDPTR:
2211 ept_save_pdptrs(vcpu);
2213 case VCPU_EXREG_CR3:
2214 if (enable_unrestricted_guest || (enable_ept && is_paging(vcpu)))
2215 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2223 static __init int cpu_has_kvm_support(void)
2225 return cpu_has_vmx();
2228 static __init int vmx_disabled_by_bios(void)
2230 return !boot_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
2231 !boot_cpu_has(X86_FEATURE_VMX);
2234 static int kvm_cpu_vmxon(u64 vmxon_pointer)
2238 cr4_set_bits(X86_CR4_VMXE);
2239 intel_pt_handle_vmx(1);
2241 asm_volatile_goto("1: vmxon %[vmxon_pointer]\n\t"
2242 _ASM_EXTABLE(1b, %l[fault])
2243 : : [vmxon_pointer] "m"(vmxon_pointer)
2248 WARN_ONCE(1, "VMXON faulted, MSR_IA32_FEAT_CTL (0x3a) = 0x%llx\n",
2249 rdmsrl_safe(MSR_IA32_FEAT_CTL, &msr) ? 0xdeadbeef : msr);
2250 intel_pt_handle_vmx(0);
2251 cr4_clear_bits(X86_CR4_VMXE);
2256 static int hardware_enable(void)
2258 int cpu = raw_smp_processor_id();
2259 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2262 if (cr4_read_shadow() & X86_CR4_VMXE)
2266 * This can happen if we hot-added a CPU but failed to allocate
2267 * VP assist page for it.
2269 if (static_branch_unlikely(&enable_evmcs) &&
2270 !hv_get_vp_assist_page(cpu))
2273 r = kvm_cpu_vmxon(phys_addr);
2283 static void vmclear_local_loaded_vmcss(void)
2285 int cpu = raw_smp_processor_id();
2286 struct loaded_vmcs *v, *n;
2288 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2289 loaded_vmcss_on_cpu_link)
2290 __loaded_vmcs_clear(v);
2294 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2297 static void kvm_cpu_vmxoff(void)
2299 asm volatile (__ex("vmxoff"));
2301 intel_pt_handle_vmx(0);
2302 cr4_clear_bits(X86_CR4_VMXE);
2305 static void hardware_disable(void)
2307 vmclear_local_loaded_vmcss();
2312 * There is no X86_FEATURE for SGX yet, but anyway we need to query CPUID
2313 * directly instead of going through cpu_has(), to ensure KVM is trapping
2314 * ENCLS whenever it's supported in hardware. It does not matter whether
2315 * the host OS supports or has enabled SGX.
2317 static bool cpu_has_sgx(void)
2319 return cpuid_eax(0) >= 0x12 && (cpuid_eax(0x12) & BIT(0));
2322 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2323 u32 msr, u32 *result)
2325 u32 vmx_msr_low, vmx_msr_high;
2326 u32 ctl = ctl_min | ctl_opt;
2328 rdmsr(msr, vmx_msr_low, vmx_msr_high);
2330 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2331 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
2333 /* Ensure minimum (required) set of control bits are supported. */
2341 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf,
2342 struct vmx_capability *vmx_cap)
2344 u32 vmx_msr_low, vmx_msr_high;
2345 u32 min, opt, min2, opt2;
2346 u32 _pin_based_exec_control = 0;
2347 u32 _cpu_based_exec_control = 0;
2348 u32 _cpu_based_2nd_exec_control = 0;
2349 u32 _vmexit_control = 0;
2350 u32 _vmentry_control = 0;
2352 memset(vmcs_conf, 0, sizeof(*vmcs_conf));
2353 min = CPU_BASED_HLT_EXITING |
2354 #ifdef CONFIG_X86_64
2355 CPU_BASED_CR8_LOAD_EXITING |
2356 CPU_BASED_CR8_STORE_EXITING |
2358 CPU_BASED_CR3_LOAD_EXITING |
2359 CPU_BASED_CR3_STORE_EXITING |
2360 CPU_BASED_UNCOND_IO_EXITING |
2361 CPU_BASED_MOV_DR_EXITING |
2362 CPU_BASED_USE_TSC_OFFSETTING |
2363 CPU_BASED_MWAIT_EXITING |
2364 CPU_BASED_MONITOR_EXITING |
2365 CPU_BASED_INVLPG_EXITING |
2366 CPU_BASED_RDPMC_EXITING;
2368 opt = CPU_BASED_TPR_SHADOW |
2369 CPU_BASED_USE_MSR_BITMAPS |
2370 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2371 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2372 &_cpu_based_exec_control) < 0)
2374 #ifdef CONFIG_X86_64
2375 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2376 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2377 ~CPU_BASED_CR8_STORE_EXITING;
2379 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2381 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2382 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2383 SECONDARY_EXEC_WBINVD_EXITING |
2384 SECONDARY_EXEC_ENABLE_VPID |
2385 SECONDARY_EXEC_ENABLE_EPT |
2386 SECONDARY_EXEC_UNRESTRICTED_GUEST |
2387 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2388 SECONDARY_EXEC_DESC |
2389 SECONDARY_EXEC_RDTSCP |
2390 SECONDARY_EXEC_ENABLE_INVPCID |
2391 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2392 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2393 SECONDARY_EXEC_SHADOW_VMCS |
2394 SECONDARY_EXEC_XSAVES |
2395 SECONDARY_EXEC_RDSEED_EXITING |
2396 SECONDARY_EXEC_RDRAND_EXITING |
2397 SECONDARY_EXEC_ENABLE_PML |
2398 SECONDARY_EXEC_TSC_SCALING |
2399 SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE |
2400 SECONDARY_EXEC_PT_USE_GPA |
2401 SECONDARY_EXEC_PT_CONCEAL_VMX |
2402 SECONDARY_EXEC_ENABLE_VMFUNC;
2404 opt2 |= SECONDARY_EXEC_ENCLS_EXITING;
2405 if (adjust_vmx_controls(min2, opt2,
2406 MSR_IA32_VMX_PROCBASED_CTLS2,
2407 &_cpu_based_2nd_exec_control) < 0)
2410 #ifndef CONFIG_X86_64
2411 if (!(_cpu_based_2nd_exec_control &
2412 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2413 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2416 if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2417 _cpu_based_2nd_exec_control &= ~(
2418 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2419 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2420 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
2422 rdmsr_safe(MSR_IA32_VMX_EPT_VPID_CAP,
2423 &vmx_cap->ept, &vmx_cap->vpid);
2425 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2426 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2428 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2429 CPU_BASED_CR3_STORE_EXITING |
2430 CPU_BASED_INVLPG_EXITING);
2431 } else if (vmx_cap->ept) {
2433 pr_warn_once("EPT CAP should not exist if not support "
2434 "1-setting enable EPT VM-execution control\n");
2436 if (!(_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_VPID) &&
2439 pr_warn_once("VPID CAP should not exist if not support "
2440 "1-setting enable VPID VM-execution control\n");
2443 min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
2444 #ifdef CONFIG_X86_64
2445 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2447 opt = VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL |
2448 VM_EXIT_LOAD_IA32_PAT |
2449 VM_EXIT_LOAD_IA32_EFER |
2450 VM_EXIT_CLEAR_BNDCFGS |
2451 VM_EXIT_PT_CONCEAL_PIP |
2452 VM_EXIT_CLEAR_IA32_RTIT_CTL;
2453 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2454 &_vmexit_control) < 0)
2457 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2458 opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR |
2459 PIN_BASED_VMX_PREEMPTION_TIMER;
2460 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2461 &_pin_based_exec_control) < 0)
2464 if (cpu_has_broken_vmx_preemption_timer())
2465 _pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
2466 if (!(_cpu_based_2nd_exec_control &
2467 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
2468 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
2470 min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
2471 opt = VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL |
2472 VM_ENTRY_LOAD_IA32_PAT |
2473 VM_ENTRY_LOAD_IA32_EFER |
2474 VM_ENTRY_LOAD_BNDCFGS |
2475 VM_ENTRY_PT_CONCEAL_PIP |
2476 VM_ENTRY_LOAD_IA32_RTIT_CTL;
2477 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2478 &_vmentry_control) < 0)
2482 * Some cpus support VM_{ENTRY,EXIT}_IA32_PERF_GLOBAL_CTRL but they
2483 * can't be used due to an errata where VM Exit may incorrectly clear
2484 * IA32_PERF_GLOBAL_CTRL[34:32]. Workaround the errata by using the
2485 * MSR load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2487 if (boot_cpu_data.x86 == 0x6) {
2488 switch (boot_cpu_data.x86_model) {
2489 case 26: /* AAK155 */
2490 case 30: /* AAP115 */
2491 case 37: /* AAT100 */
2492 case 44: /* BC86,AAY89,BD102 */
2494 _vmentry_control &= ~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
2495 _vmexit_control &= ~VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
2496 pr_warn_once("kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
2497 "does not work properly. Using workaround\n");
2505 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2507 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2508 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2511 #ifdef CONFIG_X86_64
2512 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2513 if (vmx_msr_high & (1u<<16))
2517 /* Require Write-Back (WB) memory type for VMCS accesses. */
2518 if (((vmx_msr_high >> 18) & 15) != 6)
2521 vmcs_conf->size = vmx_msr_high & 0x1fff;
2522 vmcs_conf->order = get_order(vmcs_conf->size);
2523 vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
2525 vmcs_conf->revision_id = vmx_msr_low;
2527 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2528 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2529 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2530 vmcs_conf->vmexit_ctrl = _vmexit_control;
2531 vmcs_conf->vmentry_ctrl = _vmentry_control;
2533 if (static_branch_unlikely(&enable_evmcs))
2534 evmcs_sanitize_exec_ctrls(vmcs_conf);
2539 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags)
2541 int node = cpu_to_node(cpu);
2545 pages = __alloc_pages_node(node, flags, vmcs_config.order);
2548 vmcs = page_address(pages);
2549 memset(vmcs, 0, vmcs_config.size);
2551 /* KVM supports Enlightened VMCS v1 only */
2552 if (static_branch_unlikely(&enable_evmcs))
2553 vmcs->hdr.revision_id = KVM_EVMCS_VERSION;
2555 vmcs->hdr.revision_id = vmcs_config.revision_id;
2558 vmcs->hdr.shadow_vmcs = 1;
2562 void free_vmcs(struct vmcs *vmcs)
2564 free_pages((unsigned long)vmcs, vmcs_config.order);
2568 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2570 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2572 if (!loaded_vmcs->vmcs)
2574 loaded_vmcs_clear(loaded_vmcs);
2575 free_vmcs(loaded_vmcs->vmcs);
2576 loaded_vmcs->vmcs = NULL;
2577 if (loaded_vmcs->msr_bitmap)
2578 free_page((unsigned long)loaded_vmcs->msr_bitmap);
2579 WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
2582 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2584 loaded_vmcs->vmcs = alloc_vmcs(false);
2585 if (!loaded_vmcs->vmcs)
2588 vmcs_clear(loaded_vmcs->vmcs);
2590 loaded_vmcs->shadow_vmcs = NULL;
2591 loaded_vmcs->hv_timer_soft_disabled = false;
2592 loaded_vmcs->cpu = -1;
2593 loaded_vmcs->launched = 0;
2595 if (cpu_has_vmx_msr_bitmap()) {
2596 loaded_vmcs->msr_bitmap = (unsigned long *)
2597 __get_free_page(GFP_KERNEL_ACCOUNT);
2598 if (!loaded_vmcs->msr_bitmap)
2600 memset(loaded_vmcs->msr_bitmap, 0xff, PAGE_SIZE);
2602 if (IS_ENABLED(CONFIG_HYPERV) &&
2603 static_branch_unlikely(&enable_evmcs) &&
2604 (ms_hyperv.nested_features & HV_X64_NESTED_MSR_BITMAP)) {
2605 struct hv_enlightened_vmcs *evmcs =
2606 (struct hv_enlightened_vmcs *)loaded_vmcs->vmcs;
2608 evmcs->hv_enlightenments_control.msr_bitmap = 1;
2612 memset(&loaded_vmcs->host_state, 0, sizeof(struct vmcs_host_state));
2613 memset(&loaded_vmcs->controls_shadow, 0,
2614 sizeof(struct vmcs_controls_shadow));
2619 free_loaded_vmcs(loaded_vmcs);
2623 static void free_kvm_area(void)
2627 for_each_possible_cpu(cpu) {
2628 free_vmcs(per_cpu(vmxarea, cpu));
2629 per_cpu(vmxarea, cpu) = NULL;
2633 static __init int alloc_kvm_area(void)
2637 for_each_possible_cpu(cpu) {
2640 vmcs = alloc_vmcs_cpu(false, cpu, GFP_KERNEL);
2647 * When eVMCS is enabled, alloc_vmcs_cpu() sets
2648 * vmcs->revision_id to KVM_EVMCS_VERSION instead of
2649 * revision_id reported by MSR_IA32_VMX_BASIC.
2651 * However, even though not explicitly documented by
2652 * TLFS, VMXArea passed as VMXON argument should
2653 * still be marked with revision_id reported by
2656 if (static_branch_unlikely(&enable_evmcs))
2657 vmcs->hdr.revision_id = vmcs_config.revision_id;
2659 per_cpu(vmxarea, cpu) = vmcs;
2664 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
2665 struct kvm_segment *save)
2667 if (!emulate_invalid_guest_state) {
2669 * CS and SS RPL should be equal during guest entry according
2670 * to VMX spec, but in reality it is not always so. Since vcpu
2671 * is in the middle of the transition from real mode to
2672 * protected mode it is safe to assume that RPL 0 is a good
2675 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
2676 save->selector &= ~SEGMENT_RPL_MASK;
2677 save->dpl = save->selector & SEGMENT_RPL_MASK;
2680 vmx_set_segment(vcpu, save, seg);
2683 static void enter_pmode(struct kvm_vcpu *vcpu)
2685 unsigned long flags;
2686 struct vcpu_vmx *vmx = to_vmx(vcpu);
2689 * Update real mode segment cache. It may be not up-to-date if sement
2690 * register was written while vcpu was in a guest mode.
2692 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2693 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2694 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2695 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2696 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2697 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2699 vmx->rmode.vm86_active = 0;
2701 vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2703 flags = vmcs_readl(GUEST_RFLAGS);
2704 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2705 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2706 vmcs_writel(GUEST_RFLAGS, flags);
2708 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
2709 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
2711 update_exception_bitmap(vcpu);
2713 fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2714 fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2715 fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2716 fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2717 fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2718 fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2721 static void fix_rmode_seg(int seg, struct kvm_segment *save)
2723 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2724 struct kvm_segment var = *save;
2727 if (seg == VCPU_SREG_CS)
2730 if (!emulate_invalid_guest_state) {
2731 var.selector = var.base >> 4;
2732 var.base = var.base & 0xffff0;
2742 if (save->base & 0xf)
2743 printk_once(KERN_WARNING "kvm: segment base is not "
2744 "paragraph aligned when entering "
2745 "protected mode (seg=%d)", seg);
2748 vmcs_write16(sf->selector, var.selector);
2749 vmcs_writel(sf->base, var.base);
2750 vmcs_write32(sf->limit, var.limit);
2751 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
2754 static void enter_rmode(struct kvm_vcpu *vcpu)
2756 unsigned long flags;
2757 struct vcpu_vmx *vmx = to_vmx(vcpu);
2758 struct kvm_vmx *kvm_vmx = to_kvm_vmx(vcpu->kvm);
2760 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2761 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2762 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2763 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2764 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2765 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2766 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2768 vmx->rmode.vm86_active = 1;
2771 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
2772 * vcpu. Warn the user that an update is overdue.
2774 if (!kvm_vmx->tss_addr)
2775 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
2776 "called before entering vcpu\n");
2778 vmx_segment_cache_clear(vmx);
2780 vmcs_writel(GUEST_TR_BASE, kvm_vmx->tss_addr);
2781 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
2782 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2784 flags = vmcs_readl(GUEST_RFLAGS);
2785 vmx->rmode.save_rflags = flags;
2787 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2789 vmcs_writel(GUEST_RFLAGS, flags);
2790 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
2791 update_exception_bitmap(vcpu);
2793 fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2794 fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2795 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2796 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2797 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2798 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2800 kvm_mmu_reset_context(vcpu);
2803 void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
2805 struct vcpu_vmx *vmx = to_vmx(vcpu);
2806 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
2811 vcpu->arch.efer = efer;
2812 if (efer & EFER_LMA) {
2813 vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2816 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2818 msr->data = efer & ~EFER_LME;
2823 #ifdef CONFIG_X86_64
2825 static void enter_lmode(struct kvm_vcpu *vcpu)
2829 vmx_segment_cache_clear(to_vmx(vcpu));
2831 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
2832 if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
2833 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
2835 vmcs_write32(GUEST_TR_AR_BYTES,
2836 (guest_tr_ar & ~VMX_AR_TYPE_MASK)
2837 | VMX_AR_TYPE_BUSY_64_TSS);
2839 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
2842 static void exit_lmode(struct kvm_vcpu *vcpu)
2844 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2845 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
2850 static void vmx_flush_tlb_all(struct kvm_vcpu *vcpu)
2852 struct vcpu_vmx *vmx = to_vmx(vcpu);
2855 * INVEPT must be issued when EPT is enabled, irrespective of VPID, as
2856 * the CPU is not required to invalidate guest-physical mappings on
2857 * VM-Entry, even if VPID is disabled. Guest-physical mappings are
2858 * associated with the root EPT structure and not any particular VPID
2859 * (INVVPID also isn't required to invalidate guest-physical mappings).
2863 } else if (enable_vpid) {
2864 if (cpu_has_vmx_invvpid_global()) {
2865 vpid_sync_vcpu_global();
2867 vpid_sync_vcpu_single(vmx->vpid);
2868 vpid_sync_vcpu_single(vmx->nested.vpid02);
2873 static void vmx_flush_tlb_current(struct kvm_vcpu *vcpu)
2875 u64 root_hpa = vcpu->arch.mmu->root_hpa;
2877 /* No flush required if the current context is invalid. */
2878 if (!VALID_PAGE(root_hpa))
2882 ept_sync_context(construct_eptp(vcpu, root_hpa));
2883 else if (!is_guest_mode(vcpu))
2884 vpid_sync_context(to_vmx(vcpu)->vpid);
2886 vpid_sync_context(nested_get_vpid02(vcpu));
2889 static void vmx_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t addr)
2892 * vpid_sync_vcpu_addr() is a nop if vmx->vpid==0, see the comment in
2893 * vmx_flush_tlb_guest() for an explanation of why this is ok.
2895 vpid_sync_vcpu_addr(to_vmx(vcpu)->vpid, addr);
2898 static void vmx_flush_tlb_guest(struct kvm_vcpu *vcpu)
2901 * vpid_sync_context() is a nop if vmx->vpid==0, e.g. if enable_vpid==0
2902 * or a vpid couldn't be allocated for this vCPU. VM-Enter and VM-Exit
2903 * are required to flush GVA->{G,H}PA mappings from the TLB if vpid is
2904 * disabled (VM-Enter with vpid enabled and vpid==0 is disallowed),
2905 * i.e. no explicit INVVPID is necessary.
2907 vpid_sync_context(to_vmx(vcpu)->vpid);
2910 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
2912 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
2914 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
2915 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
2918 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
2920 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
2922 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
2923 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
2926 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
2928 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2930 if (!kvm_register_is_dirty(vcpu, VCPU_EXREG_PDPTR))
2933 if (is_pae_paging(vcpu)) {
2934 vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
2935 vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
2936 vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
2937 vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
2941 void ept_save_pdptrs(struct kvm_vcpu *vcpu)
2943 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2945 if (WARN_ON_ONCE(!is_pae_paging(vcpu)))
2948 mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
2949 mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
2950 mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
2951 mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
2953 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
2956 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
2958 struct kvm_vcpu *vcpu)
2960 struct vcpu_vmx *vmx = to_vmx(vcpu);
2962 if (!kvm_register_is_available(vcpu, VCPU_EXREG_CR3))
2963 vmx_cache_reg(vcpu, VCPU_EXREG_CR3);
2964 if (!(cr0 & X86_CR0_PG)) {
2965 /* From paging/starting to nonpaging */
2966 exec_controls_setbit(vmx, CPU_BASED_CR3_LOAD_EXITING |
2967 CPU_BASED_CR3_STORE_EXITING);
2968 vcpu->arch.cr0 = cr0;
2969 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2970 } else if (!is_paging(vcpu)) {
2971 /* From nonpaging to paging */
2972 exec_controls_clearbit(vmx, CPU_BASED_CR3_LOAD_EXITING |
2973 CPU_BASED_CR3_STORE_EXITING);
2974 vcpu->arch.cr0 = cr0;
2975 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2978 if (!(cr0 & X86_CR0_WP))
2979 *hw_cr0 &= ~X86_CR0_WP;
2982 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
2984 struct vcpu_vmx *vmx = to_vmx(vcpu);
2985 unsigned long hw_cr0;
2987 hw_cr0 = (cr0 & ~KVM_VM_CR0_ALWAYS_OFF);
2988 if (enable_unrestricted_guest)
2989 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
2991 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
2993 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
2996 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
3000 #ifdef CONFIG_X86_64
3001 if (vcpu->arch.efer & EFER_LME) {
3002 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
3004 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
3009 if (enable_ept && !enable_unrestricted_guest)
3010 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
3012 vmcs_writel(CR0_READ_SHADOW, cr0);
3013 vmcs_writel(GUEST_CR0, hw_cr0);
3014 vcpu->arch.cr0 = cr0;
3016 /* depends on vcpu->arch.cr0 to be set to a new value */
3017 vmx->emulation_required = emulation_required(vcpu);
3020 static int get_ept_level(struct kvm_vcpu *vcpu)
3022 if (is_guest_mode(vcpu) && nested_cpu_has_ept(get_vmcs12(vcpu)))
3023 return vmx_eptp_page_walk_level(nested_ept_get_eptp(vcpu));
3024 if (cpu_has_vmx_ept_5levels() && (cpuid_maxphyaddr(vcpu) > 48))
3029 u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa)
3031 u64 eptp = VMX_EPTP_MT_WB;
3033 eptp |= (get_ept_level(vcpu) == 5) ? VMX_EPTP_PWL_5 : VMX_EPTP_PWL_4;
3035 if (enable_ept_ad_bits &&
3036 (!is_guest_mode(vcpu) || nested_ept_ad_enabled(vcpu)))
3037 eptp |= VMX_EPTP_AD_ENABLE_BIT;
3038 eptp |= (root_hpa & PAGE_MASK);
3043 void vmx_load_mmu_pgd(struct kvm_vcpu *vcpu, unsigned long pgd)
3045 struct kvm *kvm = vcpu->kvm;
3046 bool update_guest_cr3 = true;
3047 unsigned long guest_cr3;
3051 eptp = construct_eptp(vcpu, pgd);
3052 vmcs_write64(EPT_POINTER, eptp);
3054 if (kvm_x86_ops.tlb_remote_flush) {
3055 spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
3056 to_vmx(vcpu)->ept_pointer = eptp;
3057 to_kvm_vmx(kvm)->ept_pointers_match
3058 = EPT_POINTERS_CHECK;
3059 spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
3062 /* Loading vmcs02.GUEST_CR3 is handled by nested VM-Enter. */
3063 if (is_guest_mode(vcpu))
3064 update_guest_cr3 = false;
3065 else if (!enable_unrestricted_guest && !is_paging(vcpu))
3066 guest_cr3 = to_kvm_vmx(kvm)->ept_identity_map_addr;
3067 else if (test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
3068 guest_cr3 = vcpu->arch.cr3;
3069 else /* vmcs01.GUEST_CR3 is already up-to-date. */
3070 update_guest_cr3 = false;
3071 ept_load_pdptrs(vcpu);
3076 if (update_guest_cr3)
3077 vmcs_writel(GUEST_CR3, guest_cr3);
3080 int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3082 struct vcpu_vmx *vmx = to_vmx(vcpu);
3084 * Pass through host's Machine Check Enable value to hw_cr4, which
3085 * is in force while we are in guest mode. Do not let guests control
3086 * this bit, even if host CR4.MCE == 0.
3088 unsigned long hw_cr4;
3090 hw_cr4 = (cr4_read_shadow() & X86_CR4_MCE) | (cr4 & ~X86_CR4_MCE);
3091 if (enable_unrestricted_guest)
3092 hw_cr4 |= KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST;
3093 else if (vmx->rmode.vm86_active)
3094 hw_cr4 |= KVM_RMODE_VM_CR4_ALWAYS_ON;
3096 hw_cr4 |= KVM_PMODE_VM_CR4_ALWAYS_ON;
3098 if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated()) {
3099 if (cr4 & X86_CR4_UMIP) {
3100 secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_DESC);
3101 hw_cr4 &= ~X86_CR4_UMIP;
3102 } else if (!is_guest_mode(vcpu) ||
3103 !nested_cpu_has2(get_vmcs12(vcpu), SECONDARY_EXEC_DESC)) {
3104 secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_DESC);
3108 if (cr4 & X86_CR4_VMXE) {
3110 * To use VMXON (and later other VMX instructions), a guest
3111 * must first be able to turn on cr4.VMXE (see handle_vmon()).
3112 * So basically the check on whether to allow nested VMX
3113 * is here. We operate under the default treatment of SMM,
3114 * so VMX cannot be enabled under SMM.
3116 if (!nested_vmx_allowed(vcpu) || is_smm(vcpu))
3120 if (vmx->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
3123 vcpu->arch.cr4 = cr4;
3125 if (!enable_unrestricted_guest) {
3127 if (!is_paging(vcpu)) {
3128 hw_cr4 &= ~X86_CR4_PAE;
3129 hw_cr4 |= X86_CR4_PSE;
3130 } else if (!(cr4 & X86_CR4_PAE)) {
3131 hw_cr4 &= ~X86_CR4_PAE;
3136 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
3137 * hardware. To emulate this behavior, SMEP/SMAP/PKU needs
3138 * to be manually disabled when guest switches to non-paging
3141 * If !enable_unrestricted_guest, the CPU is always running
3142 * with CR0.PG=1 and CR4 needs to be modified.
3143 * If enable_unrestricted_guest, the CPU automatically
3144 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
3146 if (!is_paging(vcpu))
3147 hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
3150 vmcs_writel(CR4_READ_SHADOW, cr4);
3151 vmcs_writel(GUEST_CR4, hw_cr4);
3155 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3157 struct vcpu_vmx *vmx = to_vmx(vcpu);
3160 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3161 *var = vmx->rmode.segs[seg];
3162 if (seg == VCPU_SREG_TR
3163 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3165 var->base = vmx_read_guest_seg_base(vmx, seg);
3166 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3169 var->base = vmx_read_guest_seg_base(vmx, seg);
3170 var->limit = vmx_read_guest_seg_limit(vmx, seg);
3171 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3172 ar = vmx_read_guest_seg_ar(vmx, seg);
3173 var->unusable = (ar >> 16) & 1;
3174 var->type = ar & 15;
3175 var->s = (ar >> 4) & 1;
3176 var->dpl = (ar >> 5) & 3;
3178 * Some userspaces do not preserve unusable property. Since usable
3179 * segment has to be present according to VMX spec we can use present
3180 * property to amend userspace bug by making unusable segment always
3181 * nonpresent. vmx_segment_access_rights() already marks nonpresent
3182 * segment as unusable.
3184 var->present = !var->unusable;
3185 var->avl = (ar >> 12) & 1;
3186 var->l = (ar >> 13) & 1;
3187 var->db = (ar >> 14) & 1;
3188 var->g = (ar >> 15) & 1;
3191 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3193 struct kvm_segment s;
3195 if (to_vmx(vcpu)->rmode.vm86_active) {
3196 vmx_get_segment(vcpu, &s, seg);
3199 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3202 int vmx_get_cpl(struct kvm_vcpu *vcpu)
3204 struct vcpu_vmx *vmx = to_vmx(vcpu);
3206 if (unlikely(vmx->rmode.vm86_active))
3209 int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
3210 return VMX_AR_DPL(ar);
3214 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3218 if (var->unusable || !var->present)
3221 ar = var->type & 15;
3222 ar |= (var->s & 1) << 4;
3223 ar |= (var->dpl & 3) << 5;
3224 ar |= (var->present & 1) << 7;
3225 ar |= (var->avl & 1) << 12;
3226 ar |= (var->l & 1) << 13;
3227 ar |= (var->db & 1) << 14;
3228 ar |= (var->g & 1) << 15;
3234 void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3236 struct vcpu_vmx *vmx = to_vmx(vcpu);
3237 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3239 vmx_segment_cache_clear(vmx);
3241 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3242 vmx->rmode.segs[seg] = *var;
3243 if (seg == VCPU_SREG_TR)
3244 vmcs_write16(sf->selector, var->selector);
3246 fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3250 vmcs_writel(sf->base, var->base);
3251 vmcs_write32(sf->limit, var->limit);
3252 vmcs_write16(sf->selector, var->selector);
3255 * Fix the "Accessed" bit in AR field of segment registers for older
3257 * IA32 arch specifies that at the time of processor reset the
3258 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3259 * is setting it to 0 in the userland code. This causes invalid guest
3260 * state vmexit when "unrestricted guest" mode is turned on.
3261 * Fix for this setup issue in cpu_reset is being pushed in the qemu
3262 * tree. Newer qemu binaries with that qemu fix would not need this
3265 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3266 var->type |= 0x1; /* Accessed */
3268 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3271 vmx->emulation_required = emulation_required(vcpu);
3274 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3276 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3278 *db = (ar >> 14) & 1;
3279 *l = (ar >> 13) & 1;
3282 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3284 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3285 dt->address = vmcs_readl(GUEST_IDTR_BASE);
3288 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3290 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3291 vmcs_writel(GUEST_IDTR_BASE, dt->address);
3294 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3296 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3297 dt->address = vmcs_readl(GUEST_GDTR_BASE);
3300 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3302 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3303 vmcs_writel(GUEST_GDTR_BASE, dt->address);
3306 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3308 struct kvm_segment var;
3311 vmx_get_segment(vcpu, &var, seg);
3313 if (seg == VCPU_SREG_CS)
3315 ar = vmx_segment_access_rights(&var);
3317 if (var.base != (var.selector << 4))
3319 if (var.limit != 0xffff)
3327 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3329 struct kvm_segment cs;
3330 unsigned int cs_rpl;
3332 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3333 cs_rpl = cs.selector & SEGMENT_RPL_MASK;
3337 if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
3341 if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
3342 if (cs.dpl > cs_rpl)
3345 if (cs.dpl != cs_rpl)
3351 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3355 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3357 struct kvm_segment ss;
3358 unsigned int ss_rpl;
3360 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3361 ss_rpl = ss.selector & SEGMENT_RPL_MASK;
3365 if (ss.type != 3 && ss.type != 7)
3369 if (ss.dpl != ss_rpl) /* DPL != RPL */
3377 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3379 struct kvm_segment var;
3382 vmx_get_segment(vcpu, &var, seg);
3383 rpl = var.selector & SEGMENT_RPL_MASK;
3391 if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
3392 if (var.dpl < rpl) /* DPL < RPL */
3396 /* TODO: Add other members to kvm_segment_field to allow checking for other access
3402 static bool tr_valid(struct kvm_vcpu *vcpu)
3404 struct kvm_segment tr;
3406 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3410 if (tr.selector & SEGMENT_TI_MASK) /* TI = 1 */
3412 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3420 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3422 struct kvm_segment ldtr;
3424 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3428 if (ldtr.selector & SEGMENT_TI_MASK) /* TI = 1 */
3438 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3440 struct kvm_segment cs, ss;
3442 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3443 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3445 return ((cs.selector & SEGMENT_RPL_MASK) ==
3446 (ss.selector & SEGMENT_RPL_MASK));
3450 * Check if guest state is valid. Returns true if valid, false if
3452 * We assume that registers are always usable
3454 static bool guest_state_valid(struct kvm_vcpu *vcpu)
3456 if (enable_unrestricted_guest)
3459 /* real mode guest state checks */
3460 if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
3461 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3463 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3465 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3467 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3469 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3471 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3474 /* protected mode guest state checks */
3475 if (!cs_ss_rpl_check(vcpu))
3477 if (!code_segment_valid(vcpu))
3479 if (!stack_segment_valid(vcpu))
3481 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3483 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3485 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3487 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3489 if (!tr_valid(vcpu))
3491 if (!ldtr_valid(vcpu))
3495 * - Add checks on RIP
3496 * - Add checks on RFLAGS
3502 static int init_rmode_tss(struct kvm *kvm)
3508 idx = srcu_read_lock(&kvm->srcu);
3509 fn = to_kvm_vmx(kvm)->tss_addr >> PAGE_SHIFT;
3510 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3513 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3514 r = kvm_write_guest_page(kvm, fn++, &data,
3515 TSS_IOPB_BASE_OFFSET, sizeof(u16));
3518 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3521 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3525 r = kvm_write_guest_page(kvm, fn, &data,
3526 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3529 srcu_read_unlock(&kvm->srcu, idx);
3533 static int init_rmode_identity_map(struct kvm *kvm)
3535 struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm);
3537 kvm_pfn_t identity_map_pfn;
3540 /* Protect kvm_vmx->ept_identity_pagetable_done. */
3541 mutex_lock(&kvm->slots_lock);
3543 if (likely(kvm_vmx->ept_identity_pagetable_done))
3546 if (!kvm_vmx->ept_identity_map_addr)
3547 kvm_vmx->ept_identity_map_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
3548 identity_map_pfn = kvm_vmx->ept_identity_map_addr >> PAGE_SHIFT;
3550 r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
3551 kvm_vmx->ept_identity_map_addr, PAGE_SIZE);
3555 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3558 /* Set up identity-mapping pagetable for EPT in real mode */
3559 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3560 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3561 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3562 r = kvm_write_guest_page(kvm, identity_map_pfn,
3563 &tmp, i * sizeof(tmp), sizeof(tmp));
3567 kvm_vmx->ept_identity_pagetable_done = true;
3570 mutex_unlock(&kvm->slots_lock);
3574 static void seg_setup(int seg)
3576 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3579 vmcs_write16(sf->selector, 0);
3580 vmcs_writel(sf->base, 0);
3581 vmcs_write32(sf->limit, 0xffff);
3583 if (seg == VCPU_SREG_CS)
3584 ar |= 0x08; /* code segment */
3586 vmcs_write32(sf->ar_bytes, ar);
3589 static int alloc_apic_access_page(struct kvm *kvm)
3594 mutex_lock(&kvm->slots_lock);
3595 if (kvm->arch.apic_access_page_done)
3597 r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
3598 APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
3602 page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
3603 if (is_error_page(page)) {
3609 * Do not pin the page in memory, so that memory hot-unplug
3610 * is able to migrate it.
3613 kvm->arch.apic_access_page_done = true;
3615 mutex_unlock(&kvm->slots_lock);
3619 int allocate_vpid(void)
3625 spin_lock(&vmx_vpid_lock);
3626 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3627 if (vpid < VMX_NR_VPIDS)
3628 __set_bit(vpid, vmx_vpid_bitmap);
3631 spin_unlock(&vmx_vpid_lock);
3635 void free_vpid(int vpid)
3637 if (!enable_vpid || vpid == 0)
3639 spin_lock(&vmx_vpid_lock);
3640 __clear_bit(vpid, vmx_vpid_bitmap);
3641 spin_unlock(&vmx_vpid_lock);
3644 static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
3647 int f = sizeof(unsigned long);
3649 if (!cpu_has_vmx_msr_bitmap())
3652 if (static_branch_unlikely(&enable_evmcs))
3653 evmcs_touch_msr_bitmap();
3656 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3657 * have the write-low and read-high bitmap offsets the wrong way round.
3658 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3660 if (msr <= 0x1fff) {
3661 if (type & MSR_TYPE_R)
3663 __clear_bit(msr, msr_bitmap + 0x000 / f);
3665 if (type & MSR_TYPE_W)
3667 __clear_bit(msr, msr_bitmap + 0x800 / f);
3669 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3671 if (type & MSR_TYPE_R)
3673 __clear_bit(msr, msr_bitmap + 0x400 / f);
3675 if (type & MSR_TYPE_W)
3677 __clear_bit(msr, msr_bitmap + 0xc00 / f);
3682 static __always_inline void vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
3685 int f = sizeof(unsigned long);
3687 if (!cpu_has_vmx_msr_bitmap())
3690 if (static_branch_unlikely(&enable_evmcs))
3691 evmcs_touch_msr_bitmap();
3694 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3695 * have the write-low and read-high bitmap offsets the wrong way round.
3696 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3698 if (msr <= 0x1fff) {
3699 if (type & MSR_TYPE_R)
3701 __set_bit(msr, msr_bitmap + 0x000 / f);
3703 if (type & MSR_TYPE_W)
3705 __set_bit(msr, msr_bitmap + 0x800 / f);
3707 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3709 if (type & MSR_TYPE_R)
3711 __set_bit(msr, msr_bitmap + 0x400 / f);
3713 if (type & MSR_TYPE_W)
3715 __set_bit(msr, msr_bitmap + 0xc00 / f);
3720 static __always_inline void vmx_set_intercept_for_msr(unsigned long *msr_bitmap,
3721 u32 msr, int type, bool value)
3724 vmx_enable_intercept_for_msr(msr_bitmap, msr, type);
3726 vmx_disable_intercept_for_msr(msr_bitmap, msr, type);
3729 static u8 vmx_msr_bitmap_mode(struct kvm_vcpu *vcpu)
3733 if (cpu_has_secondary_exec_ctrls() &&
3734 (secondary_exec_controls_get(to_vmx(vcpu)) &
3735 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
3736 mode |= MSR_BITMAP_MODE_X2APIC;
3737 if (enable_apicv && kvm_vcpu_apicv_active(vcpu))
3738 mode |= MSR_BITMAP_MODE_X2APIC_APICV;
3744 static void vmx_update_msr_bitmap_x2apic(unsigned long *msr_bitmap,
3749 for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
3750 unsigned word = msr / BITS_PER_LONG;
3751 msr_bitmap[word] = (mode & MSR_BITMAP_MODE_X2APIC_APICV) ? 0 : ~0;
3752 msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
3755 if (mode & MSR_BITMAP_MODE_X2APIC) {
3757 * TPR reads and writes can be virtualized even if virtual interrupt
3758 * delivery is not in use.
3760 vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TASKPRI), MSR_TYPE_RW);
3761 if (mode & MSR_BITMAP_MODE_X2APIC_APICV) {
3762 vmx_enable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TMCCT), MSR_TYPE_R);
3763 vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_EOI), MSR_TYPE_W);
3764 vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_SELF_IPI), MSR_TYPE_W);
3769 void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu)
3771 struct vcpu_vmx *vmx = to_vmx(vcpu);
3772 unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3773 u8 mode = vmx_msr_bitmap_mode(vcpu);
3774 u8 changed = mode ^ vmx->msr_bitmap_mode;
3779 if (changed & (MSR_BITMAP_MODE_X2APIC | MSR_BITMAP_MODE_X2APIC_APICV))
3780 vmx_update_msr_bitmap_x2apic(msr_bitmap, mode);
3782 vmx->msr_bitmap_mode = mode;
3785 void pt_update_intercept_for_msr(struct vcpu_vmx *vmx)
3787 unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3788 bool flag = !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
3791 vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_STATUS,
3793 vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_BASE,
3795 vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_MASK,
3797 vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_CR3_MATCH,
3799 for (i = 0; i < vmx->pt_desc.addr_range; i++) {
3800 vmx_set_intercept_for_msr(msr_bitmap,
3801 MSR_IA32_RTIT_ADDR0_A + i * 2, MSR_TYPE_RW, flag);
3802 vmx_set_intercept_for_msr(msr_bitmap,
3803 MSR_IA32_RTIT_ADDR0_B + i * 2, MSR_TYPE_RW, flag);
3807 static bool vmx_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
3809 struct vcpu_vmx *vmx = to_vmx(vcpu);
3814 if (WARN_ON_ONCE(!is_guest_mode(vcpu)) ||
3815 !nested_cpu_has_vid(get_vmcs12(vcpu)) ||
3816 WARN_ON_ONCE(!vmx->nested.virtual_apic_map.gfn))
3819 rvi = vmx_get_rvi();
3821 vapic_page = vmx->nested.virtual_apic_map.hva;
3822 vppr = *((u32 *)(vapic_page + APIC_PROCPRI));
3824 return ((rvi & 0xf0) > (vppr & 0xf0));
3827 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu,
3831 int pi_vec = nested ? POSTED_INTR_NESTED_VECTOR : POSTED_INTR_VECTOR;
3833 if (vcpu->mode == IN_GUEST_MODE) {
3835 * The vector of interrupt to be delivered to vcpu had
3836 * been set in PIR before this function.
3838 * Following cases will be reached in this block, and
3839 * we always send a notification event in all cases as
3842 * Case 1: vcpu keeps in non-root mode. Sending a
3843 * notification event posts the interrupt to vcpu.
3845 * Case 2: vcpu exits to root mode and is still
3846 * runnable. PIR will be synced to vIRR before the
3847 * next vcpu entry. Sending a notification event in
3848 * this case has no effect, as vcpu is not in root
3851 * Case 3: vcpu exits to root mode and is blocked.
3852 * vcpu_block() has already synced PIR to vIRR and
3853 * never blocks vcpu if vIRR is not cleared. Therefore,
3854 * a blocked vcpu here does not wait for any requested
3855 * interrupts in PIR, and sending a notification event
3856 * which has no effect is safe here.
3859 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu), pi_vec);
3866 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
3869 struct vcpu_vmx *vmx = to_vmx(vcpu);
3871 if (is_guest_mode(vcpu) &&
3872 vector == vmx->nested.posted_intr_nv) {
3874 * If a posted intr is not recognized by hardware,
3875 * we will accomplish it in the next vmentry.
3877 vmx->nested.pi_pending = true;
3878 kvm_make_request(KVM_REQ_EVENT, vcpu);
3879 /* the PIR and ON have been set by L1. */
3880 if (!kvm_vcpu_trigger_posted_interrupt(vcpu, true))
3881 kvm_vcpu_kick(vcpu);
3887 * Send interrupt to vcpu via posted interrupt way.
3888 * 1. If target vcpu is running(non-root mode), send posted interrupt
3889 * notification to vcpu and hardware will sync PIR to vIRR atomically.
3890 * 2. If target vcpu isn't running(root mode), kick it to pick up the
3891 * interrupt from PIR in next vmentry.
3893 static int vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
3895 struct vcpu_vmx *vmx = to_vmx(vcpu);
3898 r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
3902 if (!vcpu->arch.apicv_active)
3905 if (pi_test_and_set_pir(vector, &vmx->pi_desc))
3908 /* If a previous notification has sent the IPI, nothing to do. */
3909 if (pi_test_and_set_on(&vmx->pi_desc))
3912 if (!kvm_vcpu_trigger_posted_interrupt(vcpu, false))
3913 kvm_vcpu_kick(vcpu);
3919 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
3920 * will not change in the lifetime of the guest.
3921 * Note that host-state that does change is set elsewhere. E.g., host-state
3922 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
3924 void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
3928 unsigned long cr0, cr3, cr4;
3931 WARN_ON(cr0 & X86_CR0_TS);
3932 vmcs_writel(HOST_CR0, cr0); /* 22.2.3 */
3935 * Save the most likely value for this task's CR3 in the VMCS.
3936 * We can't use __get_current_cr3_fast() because we're not atomic.
3939 vmcs_writel(HOST_CR3, cr3); /* 22.2.3 FIXME: shadow tables */
3940 vmx->loaded_vmcs->host_state.cr3 = cr3;
3942 /* Save the most likely value for this task's CR4 in the VMCS. */
3943 cr4 = cr4_read_shadow();
3944 vmcs_writel(HOST_CR4, cr4); /* 22.2.3, 22.2.5 */
3945 vmx->loaded_vmcs->host_state.cr4 = cr4;
3947 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
3948 #ifdef CONFIG_X86_64
3950 * Load null selectors, so we can avoid reloading them in
3951 * vmx_prepare_switch_to_host(), in case userspace uses
3952 * the null selectors too (the expected case).
3954 vmcs_write16(HOST_DS_SELECTOR, 0);
3955 vmcs_write16(HOST_ES_SELECTOR, 0);
3957 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
3958 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
3960 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
3961 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
3963 vmcs_writel(HOST_IDTR_BASE, host_idt_base); /* 22.2.4 */
3965 vmcs_writel(HOST_RIP, (unsigned long)vmx_vmexit); /* 22.2.5 */
3967 rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
3968 vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
3969 rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
3970 vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */
3972 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
3973 rdmsr(MSR_IA32_CR_PAT, low32, high32);
3974 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
3977 if (cpu_has_load_ia32_efer())
3978 vmcs_write64(HOST_IA32_EFER, host_efer);
3981 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
3983 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
3985 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
3986 if (is_guest_mode(&vmx->vcpu))
3987 vmx->vcpu.arch.cr4_guest_owned_bits &=
3988 ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
3989 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
3992 u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
3994 u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
3996 if (!kvm_vcpu_apicv_active(&vmx->vcpu))
3997 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
4000 pin_based_exec_ctrl &= ~PIN_BASED_VIRTUAL_NMIS;
4002 if (!enable_preemption_timer)
4003 pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
4005 return pin_based_exec_ctrl;
4008 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
4010 struct vcpu_vmx *vmx = to_vmx(vcpu);
4012 pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
4013 if (cpu_has_secondary_exec_ctrls()) {
4014 if (kvm_vcpu_apicv_active(vcpu))
4015 secondary_exec_controls_setbit(vmx,
4016 SECONDARY_EXEC_APIC_REGISTER_VIRT |
4017 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4019 secondary_exec_controls_clearbit(vmx,
4020 SECONDARY_EXEC_APIC_REGISTER_VIRT |
4021 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4024 if (cpu_has_vmx_msr_bitmap())
4025 vmx_update_msr_bitmap(vcpu);
4028 u32 vmx_exec_control(struct vcpu_vmx *vmx)
4030 u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
4032 if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
4033 exec_control &= ~CPU_BASED_MOV_DR_EXITING;
4035 if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
4036 exec_control &= ~CPU_BASED_TPR_SHADOW;
4037 #ifdef CONFIG_X86_64
4038 exec_control |= CPU_BASED_CR8_STORE_EXITING |
4039 CPU_BASED_CR8_LOAD_EXITING;
4043 exec_control |= CPU_BASED_CR3_STORE_EXITING |
4044 CPU_BASED_CR3_LOAD_EXITING |
4045 CPU_BASED_INVLPG_EXITING;
4046 if (kvm_mwait_in_guest(vmx->vcpu.kvm))
4047 exec_control &= ~(CPU_BASED_MWAIT_EXITING |
4048 CPU_BASED_MONITOR_EXITING);
4049 if (kvm_hlt_in_guest(vmx->vcpu.kvm))
4050 exec_control &= ~CPU_BASED_HLT_EXITING;
4051 return exec_control;
4055 static void vmx_compute_secondary_exec_control(struct vcpu_vmx *vmx)
4057 struct kvm_vcpu *vcpu = &vmx->vcpu;
4059 u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
4061 if (vmx_pt_mode_is_system())
4062 exec_control &= ~(SECONDARY_EXEC_PT_USE_GPA | SECONDARY_EXEC_PT_CONCEAL_VMX);
4063 if (!cpu_need_virtualize_apic_accesses(vcpu))
4064 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
4066 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
4068 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
4069 enable_unrestricted_guest = 0;
4071 if (!enable_unrestricted_guest)
4072 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4073 if (kvm_pause_in_guest(vmx->vcpu.kvm))
4074 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
4075 if (!kvm_vcpu_apicv_active(vcpu))
4076 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
4077 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4078 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
4080 /* SECONDARY_EXEC_DESC is enabled/disabled on writes to CR4.UMIP,
4081 * in vmx_set_cr4. */
4082 exec_control &= ~SECONDARY_EXEC_DESC;
4084 /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4086 We can NOT enable shadow_vmcs here because we don't have yet
4089 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4092 exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
4094 if (vmx_xsaves_supported()) {
4095 /* Exposing XSAVES only when XSAVE is exposed */
4096 bool xsaves_enabled =
4097 boot_cpu_has(X86_FEATURE_XSAVE) &&
4098 guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
4099 guest_cpuid_has(vcpu, X86_FEATURE_XSAVES);
4101 vcpu->arch.xsaves_enabled = xsaves_enabled;
4103 if (!xsaves_enabled)
4104 exec_control &= ~SECONDARY_EXEC_XSAVES;
4108 vmx->nested.msrs.secondary_ctls_high |=
4109 SECONDARY_EXEC_XSAVES;
4111 vmx->nested.msrs.secondary_ctls_high &=
4112 ~SECONDARY_EXEC_XSAVES;
4116 if (cpu_has_vmx_rdtscp()) {
4117 bool rdtscp_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP);
4118 if (!rdtscp_enabled)
4119 exec_control &= ~SECONDARY_EXEC_RDTSCP;
4123 vmx->nested.msrs.secondary_ctls_high |=
4124 SECONDARY_EXEC_RDTSCP;
4126 vmx->nested.msrs.secondary_ctls_high &=
4127 ~SECONDARY_EXEC_RDTSCP;
4131 if (cpu_has_vmx_invpcid()) {
4132 /* Exposing INVPCID only when PCID is exposed */
4133 bool invpcid_enabled =
4134 guest_cpuid_has(vcpu, X86_FEATURE_INVPCID) &&
4135 guest_cpuid_has(vcpu, X86_FEATURE_PCID);
4137 if (!invpcid_enabled) {
4138 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
4139 guest_cpuid_clear(vcpu, X86_FEATURE_INVPCID);
4143 if (invpcid_enabled)
4144 vmx->nested.msrs.secondary_ctls_high |=
4145 SECONDARY_EXEC_ENABLE_INVPCID;
4147 vmx->nested.msrs.secondary_ctls_high &=
4148 ~SECONDARY_EXEC_ENABLE_INVPCID;
4152 if (vmx_rdrand_supported()) {
4153 bool rdrand_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDRAND);
4155 exec_control &= ~SECONDARY_EXEC_RDRAND_EXITING;
4159 vmx->nested.msrs.secondary_ctls_high |=
4160 SECONDARY_EXEC_RDRAND_EXITING;
4162 vmx->nested.msrs.secondary_ctls_high &=
4163 ~SECONDARY_EXEC_RDRAND_EXITING;
4167 if (vmx_rdseed_supported()) {
4168 bool rdseed_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDSEED);
4170 exec_control &= ~SECONDARY_EXEC_RDSEED_EXITING;
4174 vmx->nested.msrs.secondary_ctls_high |=
4175 SECONDARY_EXEC_RDSEED_EXITING;
4177 vmx->nested.msrs.secondary_ctls_high &=
4178 ~SECONDARY_EXEC_RDSEED_EXITING;
4182 if (vmx_waitpkg_supported()) {
4183 bool waitpkg_enabled =
4184 guest_cpuid_has(vcpu, X86_FEATURE_WAITPKG);
4186 if (!waitpkg_enabled)
4187 exec_control &= ~SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4190 if (waitpkg_enabled)
4191 vmx->nested.msrs.secondary_ctls_high |=
4192 SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4194 vmx->nested.msrs.secondary_ctls_high &=
4195 ~SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4199 vmx->secondary_exec_control = exec_control;
4202 static void ept_set_mmio_spte_mask(void)
4205 * EPT Misconfigurations can be generated if the value of bits 2:0
4206 * of an EPT paging-structure entry is 110b (write/execute).
4208 kvm_mmu_set_mmio_spte_mask(VMX_EPT_RWX_MASK,
4209 VMX_EPT_MISCONFIG_WX_VALUE, 0);
4212 #define VMX_XSS_EXIT_BITMAP 0
4215 * Noting that the initialization of Guest-state Area of VMCS is in
4218 static void init_vmcs(struct vcpu_vmx *vmx)
4221 nested_vmx_set_vmcs_shadowing_bitmap();
4223 if (cpu_has_vmx_msr_bitmap())
4224 vmcs_write64(MSR_BITMAP, __pa(vmx->vmcs01.msr_bitmap));
4226 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4229 pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
4231 exec_controls_set(vmx, vmx_exec_control(vmx));
4233 if (cpu_has_secondary_exec_ctrls()) {
4234 vmx_compute_secondary_exec_control(vmx);
4235 secondary_exec_controls_set(vmx, vmx->secondary_exec_control);
4238 if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
4239 vmcs_write64(EOI_EXIT_BITMAP0, 0);
4240 vmcs_write64(EOI_EXIT_BITMAP1, 0);
4241 vmcs_write64(EOI_EXIT_BITMAP2, 0);
4242 vmcs_write64(EOI_EXIT_BITMAP3, 0);
4244 vmcs_write16(GUEST_INTR_STATUS, 0);
4246 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4247 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4250 if (!kvm_pause_in_guest(vmx->vcpu.kvm)) {
4251 vmcs_write32(PLE_GAP, ple_gap);
4252 vmx->ple_window = ple_window;
4253 vmx->ple_window_dirty = true;
4256 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4257 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4258 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
4260 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
4261 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
4262 vmx_set_constant_host_state(vmx);
4263 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4264 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4266 if (cpu_has_vmx_vmfunc())
4267 vmcs_write64(VM_FUNCTION_CONTROL, 0);
4269 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4270 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4271 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
4272 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4273 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
4275 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
4276 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
4278 vm_exit_controls_set(vmx, vmx_vmexit_ctrl());
4280 /* 22.2.1, 20.8.1 */
4281 vm_entry_controls_set(vmx, vmx_vmentry_ctrl());
4283 vmx->vcpu.arch.cr0_guest_owned_bits = X86_CR0_TS;
4284 vmcs_writel(CR0_GUEST_HOST_MASK, ~X86_CR0_TS);
4286 set_cr4_guest_host_mask(vmx);
4289 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4291 if (vmx_xsaves_supported())
4292 vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
4295 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
4296 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
4299 if (cpu_has_vmx_encls_vmexit())
4300 vmcs_write64(ENCLS_EXITING_BITMAP, -1ull);
4302 if (vmx_pt_mode_is_host_guest()) {
4303 memset(&vmx->pt_desc, 0, sizeof(vmx->pt_desc));
4304 /* Bit[6~0] are forced to 1, writes are ignored. */
4305 vmx->pt_desc.guest.output_mask = 0x7F;
4306 vmcs_write64(GUEST_IA32_RTIT_CTL, 0);
4310 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
4312 struct vcpu_vmx *vmx = to_vmx(vcpu);
4313 struct msr_data apic_base_msr;
4316 vmx->rmode.vm86_active = 0;
4319 vmx->msr_ia32_umwait_control = 0;
4321 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
4322 vmx->hv_deadline_tsc = -1;
4323 kvm_set_cr8(vcpu, 0);
4326 apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
4327 MSR_IA32_APICBASE_ENABLE;
4328 if (kvm_vcpu_is_reset_bsp(vcpu))
4329 apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
4330 apic_base_msr.host_initiated = true;
4331 kvm_set_apic_base(vcpu, &apic_base_msr);
4334 vmx_segment_cache_clear(vmx);
4336 seg_setup(VCPU_SREG_CS);
4337 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4338 vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
4340 seg_setup(VCPU_SREG_DS);
4341 seg_setup(VCPU_SREG_ES);
4342 seg_setup(VCPU_SREG_FS);
4343 seg_setup(VCPU_SREG_GS);
4344 seg_setup(VCPU_SREG_SS);
4346 vmcs_write16(GUEST_TR_SELECTOR, 0);
4347 vmcs_writel(GUEST_TR_BASE, 0);
4348 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4349 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4351 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4352 vmcs_writel(GUEST_LDTR_BASE, 0);
4353 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4354 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4357 vmcs_write32(GUEST_SYSENTER_CS, 0);
4358 vmcs_writel(GUEST_SYSENTER_ESP, 0);
4359 vmcs_writel(GUEST_SYSENTER_EIP, 0);
4360 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4363 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
4364 kvm_rip_write(vcpu, 0xfff0);
4366 vmcs_writel(GUEST_GDTR_BASE, 0);
4367 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4369 vmcs_writel(GUEST_IDTR_BASE, 0);
4370 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4372 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4373 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4374 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4375 if (kvm_mpx_supported())
4376 vmcs_write64(GUEST_BNDCFGS, 0);
4380 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
4382 if (cpu_has_vmx_tpr_shadow() && !init_event) {
4383 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4384 if (cpu_need_tpr_shadow(vcpu))
4385 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4386 __pa(vcpu->arch.apic->regs));
4387 vmcs_write32(TPR_THRESHOLD, 0);
4390 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4392 cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4393 vmx->vcpu.arch.cr0 = cr0;
4394 vmx_set_cr0(vcpu, cr0); /* enter rmode */
4395 vmx_set_cr4(vcpu, 0);
4396 vmx_set_efer(vcpu, 0);
4398 update_exception_bitmap(vcpu);
4400 vpid_sync_context(vmx->vpid);
4402 vmx_clear_hlt(vcpu);
4405 static void enable_irq_window(struct kvm_vcpu *vcpu)
4407 exec_controls_setbit(to_vmx(vcpu), CPU_BASED_INTR_WINDOW_EXITING);
4410 static void enable_nmi_window(struct kvm_vcpu *vcpu)
4413 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4414 enable_irq_window(vcpu);
4418 exec_controls_setbit(to_vmx(vcpu), CPU_BASED_NMI_WINDOW_EXITING);
4421 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4423 struct vcpu_vmx *vmx = to_vmx(vcpu);
4425 int irq = vcpu->arch.interrupt.nr;
4427 trace_kvm_inj_virq(irq);
4429 ++vcpu->stat.irq_injections;
4430 if (vmx->rmode.vm86_active) {
4432 if (vcpu->arch.interrupt.soft)
4433 inc_eip = vcpu->arch.event_exit_inst_len;
4434 kvm_inject_realmode_interrupt(vcpu, irq, inc_eip);
4437 intr = irq | INTR_INFO_VALID_MASK;
4438 if (vcpu->arch.interrupt.soft) {
4439 intr |= INTR_TYPE_SOFT_INTR;
4440 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4441 vmx->vcpu.arch.event_exit_inst_len);
4443 intr |= INTR_TYPE_EXT_INTR;
4444 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4446 vmx_clear_hlt(vcpu);
4449 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4451 struct vcpu_vmx *vmx = to_vmx(vcpu);
4455 * Tracking the NMI-blocked state in software is built upon
4456 * finding the next open IRQ window. This, in turn, depends on
4457 * well-behaving guests: They have to keep IRQs disabled at
4458 * least as long as the NMI handler runs. Otherwise we may
4459 * cause NMI nesting, maybe breaking the guest. But as this is
4460 * highly unlikely, we can live with the residual risk.
4462 vmx->loaded_vmcs->soft_vnmi_blocked = 1;
4463 vmx->loaded_vmcs->vnmi_blocked_time = 0;
4466 ++vcpu->stat.nmi_injections;
4467 vmx->loaded_vmcs->nmi_known_unmasked = false;
4469 if (vmx->rmode.vm86_active) {
4470 kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0);
4474 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4475 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4477 vmx_clear_hlt(vcpu);
4480 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4482 struct vcpu_vmx *vmx = to_vmx(vcpu);
4486 return vmx->loaded_vmcs->soft_vnmi_blocked;
4487 if (vmx->loaded_vmcs->nmi_known_unmasked)
4489 masked = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4490 vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4494 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4496 struct vcpu_vmx *vmx = to_vmx(vcpu);
4499 if (vmx->loaded_vmcs->soft_vnmi_blocked != masked) {
4500 vmx->loaded_vmcs->soft_vnmi_blocked = masked;
4501 vmx->loaded_vmcs->vnmi_blocked_time = 0;
4504 vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4506 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4507 GUEST_INTR_STATE_NMI);
4509 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4510 GUEST_INTR_STATE_NMI);
4514 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
4516 if (to_vmx(vcpu)->nested.nested_run_pending)
4520 to_vmx(vcpu)->loaded_vmcs->soft_vnmi_blocked)
4523 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4524 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
4525 | GUEST_INTR_STATE_NMI));
4528 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
4530 if (to_vmx(vcpu)->nested.nested_run_pending)
4533 if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu))
4536 return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
4537 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4538 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4541 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4545 if (enable_unrestricted_guest)
4548 mutex_lock(&kvm->slots_lock);
4549 ret = __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
4551 mutex_unlock(&kvm->slots_lock);
4555 to_kvm_vmx(kvm)->tss_addr = addr;
4556 return init_rmode_tss(kvm);
4559 static int vmx_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
4561 to_kvm_vmx(kvm)->ept_identity_map_addr = ident_addr;
4565 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
4570 * Update instruction length as we may reinject the exception
4571 * from user space while in guest debugging mode.
4573 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4574 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4575 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4579 if (vcpu->guest_debug &
4580 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
4596 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4597 int vec, u32 err_code)
4600 * Instruction with address size override prefix opcode 0x67
4601 * Cause the #SS fault with 0 error code in VM86 mode.
4603 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
4604 if (kvm_emulate_instruction(vcpu, 0)) {
4605 if (vcpu->arch.halt_request) {
4606 vcpu->arch.halt_request = 0;
4607 return kvm_vcpu_halt(vcpu);
4615 * Forward all other exceptions that are valid in real mode.
4616 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4617 * the required debugging infrastructure rework.
4619 kvm_queue_exception(vcpu, vec);
4624 * Trigger machine check on the host. We assume all the MSRs are already set up
4625 * by the CPU and that we still run on the same CPU as the MCE occurred on.
4626 * We pass a fake environment to the machine check handler because we want
4627 * the guest to be always treated like user space, no matter what context
4628 * it used internally.
4630 static void kvm_machine_check(void)
4632 #if defined(CONFIG_X86_MCE)
4633 struct pt_regs regs = {
4634 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
4635 .flags = X86_EFLAGS_IF,
4638 do_machine_check(®s, 0);
4642 static int handle_machine_check(struct kvm_vcpu *vcpu)
4644 /* handled by vmx_vcpu_run() */
4648 static int handle_exception_nmi(struct kvm_vcpu *vcpu)
4650 struct vcpu_vmx *vmx = to_vmx(vcpu);
4651 struct kvm_run *kvm_run = vcpu->run;
4652 u32 intr_info, ex_no, error_code;
4653 unsigned long cr2, rip, dr6;
4656 vect_info = vmx->idt_vectoring_info;
4657 intr_info = vmx->exit_intr_info;
4659 if (is_machine_check(intr_info) || is_nmi(intr_info))
4660 return 1; /* handled by handle_exception_nmi_irqoff() */
4662 if (is_invalid_opcode(intr_info))
4663 return handle_ud(vcpu);
4666 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4667 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4669 if (!vmx->rmode.vm86_active && is_gp_fault(intr_info)) {
4670 WARN_ON_ONCE(!enable_vmware_backdoor);
4673 * VMware backdoor emulation on #GP interception only handles
4674 * IN{S}, OUT{S}, and RDPMC, none of which generate a non-zero
4675 * error code on #GP.
4678 kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
4681 return kvm_emulate_instruction(vcpu, EMULTYPE_VMWARE_GP);
4685 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
4686 * MMIO, it is better to report an internal error.
4687 * See the comments in vmx_handle_exit.
4689 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4690 !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
4691 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4692 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4693 vcpu->run->internal.ndata = 3;
4694 vcpu->run->internal.data[0] = vect_info;
4695 vcpu->run->internal.data[1] = intr_info;
4696 vcpu->run->internal.data[2] = error_code;
4700 if (is_page_fault(intr_info)) {
4701 cr2 = vmx_get_exit_qual(vcpu);
4702 /* EPT won't cause page fault directly */
4703 WARN_ON_ONCE(!vcpu->arch.apf.host_apf_reason && enable_ept);
4704 return kvm_handle_page_fault(vcpu, error_code, cr2, NULL, 0);
4707 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4709 if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
4710 return handle_rmode_exception(vcpu, ex_no, error_code);
4714 kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
4717 dr6 = vmx_get_exit_qual(vcpu);
4718 if (!(vcpu->guest_debug &
4719 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4720 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
4721 vcpu->arch.dr6 |= dr6 | DR6_RTM;
4722 if (is_icebp(intr_info))
4723 WARN_ON(!skip_emulated_instruction(vcpu));
4725 kvm_queue_exception(vcpu, DB_VECTOR);
4728 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
4729 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4733 * Update instruction length as we may reinject #BP from
4734 * user space while in guest debugging mode. Reading it for
4735 * #DB as well causes no harm, it is not used in that case.
4737 vmx->vcpu.arch.event_exit_inst_len =
4738 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4739 kvm_run->exit_reason = KVM_EXIT_DEBUG;
4740 rip = kvm_rip_read(vcpu);
4741 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4742 kvm_run->debug.arch.exception = ex_no;
4745 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4746 kvm_run->ex.exception = ex_no;
4747 kvm_run->ex.error_code = error_code;
4753 static __always_inline int handle_external_interrupt(struct kvm_vcpu *vcpu)
4755 ++vcpu->stat.irq_exits;
4759 static int handle_triple_fault(struct kvm_vcpu *vcpu)
4761 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4762 vcpu->mmio_needed = 0;
4766 static int handle_io(struct kvm_vcpu *vcpu)
4768 unsigned long exit_qualification;
4769 int size, in, string;
4772 exit_qualification = vmx_get_exit_qual(vcpu);
4773 string = (exit_qualification & 16) != 0;
4775 ++vcpu->stat.io_exits;
4778 return kvm_emulate_instruction(vcpu, 0);
4780 port = exit_qualification >> 16;
4781 size = (exit_qualification & 7) + 1;
4782 in = (exit_qualification & 8) != 0;
4784 return kvm_fast_pio(vcpu, size, port, in);
4788 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4791 * Patch in the VMCALL instruction:
4793 hypercall[0] = 0x0f;
4794 hypercall[1] = 0x01;
4795 hypercall[2] = 0xc1;
4798 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
4799 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
4801 if (is_guest_mode(vcpu)) {
4802 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4803 unsigned long orig_val = val;
4806 * We get here when L2 changed cr0 in a way that did not change
4807 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
4808 * but did change L0 shadowed bits. So we first calculate the
4809 * effective cr0 value that L1 would like to write into the
4810 * hardware. It consists of the L2-owned bits from the new
4811 * value combined with the L1-owned bits from L1's guest_cr0.
4813 val = (val & ~vmcs12->cr0_guest_host_mask) |
4814 (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
4816 if (!nested_guest_cr0_valid(vcpu, val))
4819 if (kvm_set_cr0(vcpu, val))
4821 vmcs_writel(CR0_READ_SHADOW, orig_val);
4824 if (to_vmx(vcpu)->nested.vmxon &&
4825 !nested_host_cr0_valid(vcpu, val))
4828 return kvm_set_cr0(vcpu, val);
4832 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
4834 if (is_guest_mode(vcpu)) {
4835 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4836 unsigned long orig_val = val;
4838 /* analogously to handle_set_cr0 */
4839 val = (val & ~vmcs12->cr4_guest_host_mask) |
4840 (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
4841 if (kvm_set_cr4(vcpu, val))
4843 vmcs_writel(CR4_READ_SHADOW, orig_val);
4846 return kvm_set_cr4(vcpu, val);
4849 static int handle_desc(struct kvm_vcpu *vcpu)
4851 WARN_ON(!(vcpu->arch.cr4 & X86_CR4_UMIP));
4852 return kvm_emulate_instruction(vcpu, 0);
4855 static int handle_cr(struct kvm_vcpu *vcpu)
4857 unsigned long exit_qualification, val;
4863 exit_qualification = vmx_get_exit_qual(vcpu);
4864 cr = exit_qualification & 15;
4865 reg = (exit_qualification >> 8) & 15;
4866 switch ((exit_qualification >> 4) & 3) {
4867 case 0: /* mov to cr */
4868 val = kvm_register_readl(vcpu, reg);
4869 trace_kvm_cr_write(cr, val);
4872 err = handle_set_cr0(vcpu, val);
4873 return kvm_complete_insn_gp(vcpu, err);
4875 WARN_ON_ONCE(enable_unrestricted_guest);
4876 err = kvm_set_cr3(vcpu, val);
4877 return kvm_complete_insn_gp(vcpu, err);
4879 err = handle_set_cr4(vcpu, val);
4880 return kvm_complete_insn_gp(vcpu, err);
4882 u8 cr8_prev = kvm_get_cr8(vcpu);
4884 err = kvm_set_cr8(vcpu, cr8);
4885 ret = kvm_complete_insn_gp(vcpu, err);
4886 if (lapic_in_kernel(vcpu))
4888 if (cr8_prev <= cr8)
4891 * TODO: we might be squashing a
4892 * KVM_GUESTDBG_SINGLESTEP-triggered
4893 * KVM_EXIT_DEBUG here.
4895 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
4901 WARN_ONCE(1, "Guest should always own CR0.TS");
4902 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4903 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
4904 return kvm_skip_emulated_instruction(vcpu);
4905 case 1: /*mov from cr*/
4908 WARN_ON_ONCE(enable_unrestricted_guest);
4909 val = kvm_read_cr3(vcpu);
4910 kvm_register_write(vcpu, reg, val);
4911 trace_kvm_cr_read(cr, val);
4912 return kvm_skip_emulated_instruction(vcpu);
4914 val = kvm_get_cr8(vcpu);
4915 kvm_register_write(vcpu, reg, val);
4916 trace_kvm_cr_read(cr, val);
4917 return kvm_skip_emulated_instruction(vcpu);
4921 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
4922 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
4923 kvm_lmsw(vcpu, val);
4925 return kvm_skip_emulated_instruction(vcpu);
4929 vcpu->run->exit_reason = 0;
4930 vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
4931 (int)(exit_qualification >> 4) & 3, cr);
4935 static int handle_dr(struct kvm_vcpu *vcpu)
4937 unsigned long exit_qualification;
4940 exit_qualification = vmx_get_exit_qual(vcpu);
4941 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
4943 /* First, if DR does not exist, trigger UD */
4944 if (!kvm_require_dr(vcpu, dr))
4947 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
4948 if (!kvm_require_cpl(vcpu, 0))
4950 dr7 = vmcs_readl(GUEST_DR7);
4953 * As the vm-exit takes precedence over the debug trap, we
4954 * need to emulate the latter, either for the host or the
4955 * guest debugging itself.
4957 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
4958 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
4959 vcpu->run->debug.arch.dr7 = dr7;
4960 vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
4961 vcpu->run->debug.arch.exception = DB_VECTOR;
4962 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
4965 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
4966 vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
4967 kvm_queue_exception(vcpu, DB_VECTOR);
4972 if (vcpu->guest_debug == 0) {
4973 exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
4976 * No more DR vmexits; force a reload of the debug registers
4977 * and reenter on this instruction. The next vmexit will
4978 * retrieve the full state of the debug registers.
4980 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
4984 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
4985 if (exit_qualification & TYPE_MOV_FROM_DR) {
4988 if (kvm_get_dr(vcpu, dr, &val))
4990 kvm_register_write(vcpu, reg, val);
4992 if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
4995 return kvm_skip_emulated_instruction(vcpu);
4998 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
5000 return vcpu->arch.dr6;
5003 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
5007 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
5009 get_debugreg(vcpu->arch.db[0], 0);
5010 get_debugreg(vcpu->arch.db[1], 1);
5011 get_debugreg(vcpu->arch.db[2], 2);
5012 get_debugreg(vcpu->arch.db[3], 3);
5013 get_debugreg(vcpu->arch.dr6, 6);
5014 vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
5016 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
5017 exec_controls_setbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
5020 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
5022 vmcs_writel(GUEST_DR7, val);
5025 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
5027 kvm_apic_update_ppr(vcpu);
5031 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
5033 exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_INTR_WINDOW_EXITING);
5035 kvm_make_request(KVM_REQ_EVENT, vcpu);
5037 ++vcpu->stat.irq_window_exits;
5041 static int handle_vmcall(struct kvm_vcpu *vcpu)
5043 return kvm_emulate_hypercall(vcpu);
5046 static int handle_invd(struct kvm_vcpu *vcpu)
5048 return kvm_emulate_instruction(vcpu, 0);
5051 static int handle_invlpg(struct kvm_vcpu *vcpu)
5053 unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5055 kvm_mmu_invlpg(vcpu, exit_qualification);
5056 return kvm_skip_emulated_instruction(vcpu);
5059 static int handle_rdpmc(struct kvm_vcpu *vcpu)
5063 err = kvm_rdpmc(vcpu);
5064 return kvm_complete_insn_gp(vcpu, err);
5067 static int handle_wbinvd(struct kvm_vcpu *vcpu)
5069 return kvm_emulate_wbinvd(vcpu);
5072 static int handle_xsetbv(struct kvm_vcpu *vcpu)
5074 u64 new_bv = kvm_read_edx_eax(vcpu);
5075 u32 index = kvm_rcx_read(vcpu);
5077 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
5078 return kvm_skip_emulated_instruction(vcpu);
5082 static int handle_apic_access(struct kvm_vcpu *vcpu)
5084 if (likely(fasteoi)) {
5085 unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5086 int access_type, offset;
5088 access_type = exit_qualification & APIC_ACCESS_TYPE;
5089 offset = exit_qualification & APIC_ACCESS_OFFSET;
5091 * Sane guest uses MOV to write EOI, with written value
5092 * not cared. So make a short-circuit here by avoiding
5093 * heavy instruction emulation.
5095 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5096 (offset == APIC_EOI)) {
5097 kvm_lapic_set_eoi(vcpu);
5098 return kvm_skip_emulated_instruction(vcpu);
5101 return kvm_emulate_instruction(vcpu, 0);
5104 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5106 unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5107 int vector = exit_qualification & 0xff;
5109 /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5110 kvm_apic_set_eoi_accelerated(vcpu, vector);
5114 static int handle_apic_write(struct kvm_vcpu *vcpu)
5116 unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5117 u32 offset = exit_qualification & 0xfff;
5119 /* APIC-write VM exit is trap-like and thus no need to adjust IP */
5120 kvm_apic_write_nodecode(vcpu, offset);
5124 static int handle_task_switch(struct kvm_vcpu *vcpu)
5126 struct vcpu_vmx *vmx = to_vmx(vcpu);
5127 unsigned long exit_qualification;
5128 bool has_error_code = false;
5131 int reason, type, idt_v, idt_index;
5133 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5134 idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5135 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5137 exit_qualification = vmx_get_exit_qual(vcpu);
5139 reason = (u32)exit_qualification >> 30;
5140 if (reason == TASK_SWITCH_GATE && idt_v) {
5142 case INTR_TYPE_NMI_INTR:
5143 vcpu->arch.nmi_injected = false;
5144 vmx_set_nmi_mask(vcpu, true);
5146 case INTR_TYPE_EXT_INTR:
5147 case INTR_TYPE_SOFT_INTR:
5148 kvm_clear_interrupt_queue(vcpu);
5150 case INTR_TYPE_HARD_EXCEPTION:
5151 if (vmx->idt_vectoring_info &
5152 VECTORING_INFO_DELIVER_CODE_MASK) {
5153 has_error_code = true;
5155 vmcs_read32(IDT_VECTORING_ERROR_CODE);
5158 case INTR_TYPE_SOFT_EXCEPTION:
5159 kvm_clear_exception_queue(vcpu);
5165 tss_selector = exit_qualification;
5167 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5168 type != INTR_TYPE_EXT_INTR &&
5169 type != INTR_TYPE_NMI_INTR))
5170 WARN_ON(!skip_emulated_instruction(vcpu));
5173 * TODO: What about debug traps on tss switch?
5174 * Are we supposed to inject them and update dr6?
5176 return kvm_task_switch(vcpu, tss_selector,
5177 type == INTR_TYPE_SOFT_INTR ? idt_index : -1,
5178 reason, has_error_code, error_code);
5181 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5183 unsigned long exit_qualification;
5187 exit_qualification = vmx_get_exit_qual(vcpu);
5190 * EPT violation happened while executing iret from NMI,
5191 * "blocked by NMI" bit has to be set before next VM entry.
5192 * There are errata that may cause this bit to not be set:
5195 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5197 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
5198 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5200 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5201 trace_kvm_page_fault(gpa, exit_qualification);
5203 /* Is it a read fault? */
5204 error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
5205 ? PFERR_USER_MASK : 0;
5206 /* Is it a write fault? */
5207 error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
5208 ? PFERR_WRITE_MASK : 0;
5209 /* Is it a fetch fault? */
5210 error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
5211 ? PFERR_FETCH_MASK : 0;
5212 /* ept page table entry is present? */
5213 error_code |= (exit_qualification &
5214 (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE |
5215 EPT_VIOLATION_EXECUTABLE))
5216 ? PFERR_PRESENT_MASK : 0;
5218 error_code |= (exit_qualification & 0x100) != 0 ?
5219 PFERR_GUEST_FINAL_MASK : PFERR_GUEST_PAGE_MASK;
5221 vcpu->arch.exit_qualification = exit_qualification;
5222 return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5225 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5230 * A nested guest cannot optimize MMIO vmexits, because we have an
5231 * nGPA here instead of the required GPA.
5233 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5234 if (!is_guest_mode(vcpu) &&
5235 !kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5236 trace_kvm_fast_mmio(gpa);
5237 return kvm_skip_emulated_instruction(vcpu);
5240 return kvm_mmu_page_fault(vcpu, gpa, PFERR_RSVD_MASK, NULL, 0);
5243 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5245 WARN_ON_ONCE(!enable_vnmi);
5246 exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_NMI_WINDOW_EXITING);
5247 ++vcpu->stat.nmi_window_exits;
5248 kvm_make_request(KVM_REQ_EVENT, vcpu);
5253 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5255 struct vcpu_vmx *vmx = to_vmx(vcpu);
5256 bool intr_window_requested;
5257 unsigned count = 130;
5260 * We should never reach the point where we are emulating L2
5261 * due to invalid guest state as that means we incorrectly
5262 * allowed a nested VMEntry with an invalid vmcs12.
5264 WARN_ON_ONCE(vmx->emulation_required && vmx->nested.nested_run_pending);
5266 intr_window_requested = exec_controls_get(vmx) &
5267 CPU_BASED_INTR_WINDOW_EXITING;
5269 while (vmx->emulation_required && count-- != 0) {
5270 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
5271 return handle_interrupt_window(&vmx->vcpu);
5273 if (kvm_test_request(KVM_REQ_EVENT, vcpu))
5276 if (!kvm_emulate_instruction(vcpu, 0))
5279 if (vmx->emulation_required && !vmx->rmode.vm86_active &&
5280 vcpu->arch.exception.pending) {
5281 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5282 vcpu->run->internal.suberror =
5283 KVM_INTERNAL_ERROR_EMULATION;
5284 vcpu->run->internal.ndata = 0;
5288 if (vcpu->arch.halt_request) {
5289 vcpu->arch.halt_request = 0;
5290 return kvm_vcpu_halt(vcpu);
5294 * Note, return 1 and not 0, vcpu_run() is responsible for
5295 * morphing the pending signal into the proper return code.
5297 if (signal_pending(current))
5307 static void grow_ple_window(struct kvm_vcpu *vcpu)
5309 struct vcpu_vmx *vmx = to_vmx(vcpu);
5310 unsigned int old = vmx->ple_window;
5312 vmx->ple_window = __grow_ple_window(old, ple_window,
5316 if (vmx->ple_window != old) {
5317 vmx->ple_window_dirty = true;
5318 trace_kvm_ple_window_update(vcpu->vcpu_id,
5319 vmx->ple_window, old);
5323 static void shrink_ple_window(struct kvm_vcpu *vcpu)
5325 struct vcpu_vmx *vmx = to_vmx(vcpu);
5326 unsigned int old = vmx->ple_window;
5328 vmx->ple_window = __shrink_ple_window(old, ple_window,
5332 if (vmx->ple_window != old) {
5333 vmx->ple_window_dirty = true;
5334 trace_kvm_ple_window_update(vcpu->vcpu_id,
5335 vmx->ple_window, old);
5340 * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
5342 static void wakeup_handler(void)
5344 struct kvm_vcpu *vcpu;
5345 int cpu = smp_processor_id();
5347 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5348 list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
5349 blocked_vcpu_list) {
5350 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
5352 if (pi_test_on(pi_desc) == 1)
5353 kvm_vcpu_kick(vcpu);
5355 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5358 static void vmx_enable_tdp(void)
5360 kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK,
5361 enable_ept_ad_bits ? VMX_EPT_ACCESS_BIT : 0ull,
5362 enable_ept_ad_bits ? VMX_EPT_DIRTY_BIT : 0ull,
5363 0ull, VMX_EPT_EXECUTABLE_MASK,
5364 cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK,
5365 VMX_EPT_RWX_MASK, 0ull);
5367 ept_set_mmio_spte_mask();
5371 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5372 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5374 static int handle_pause(struct kvm_vcpu *vcpu)
5376 if (!kvm_pause_in_guest(vcpu->kvm))
5377 grow_ple_window(vcpu);
5380 * Intel sdm vol3 ch-25.1.3 says: The "PAUSE-loop exiting"
5381 * VM-execution control is ignored if CPL > 0. OTOH, KVM
5382 * never set PAUSE_EXITING and just set PLE if supported,
5383 * so the vcpu must be CPL=0 if it gets a PAUSE exit.
5385 kvm_vcpu_on_spin(vcpu, true);
5386 return kvm_skip_emulated_instruction(vcpu);
5389 static int handle_nop(struct kvm_vcpu *vcpu)
5391 return kvm_skip_emulated_instruction(vcpu);
5394 static int handle_mwait(struct kvm_vcpu *vcpu)
5396 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
5397 return handle_nop(vcpu);
5400 static int handle_invalid_op(struct kvm_vcpu *vcpu)
5402 kvm_queue_exception(vcpu, UD_VECTOR);
5406 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
5411 static int handle_monitor(struct kvm_vcpu *vcpu)
5413 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
5414 return handle_nop(vcpu);
5417 static int handle_invpcid(struct kvm_vcpu *vcpu)
5419 u32 vmx_instruction_info;
5423 struct x86_exception e;
5425 unsigned long roots_to_free = 0;
5431 if (!guest_cpuid_has(vcpu, X86_FEATURE_INVPCID)) {
5432 kvm_queue_exception(vcpu, UD_VECTOR);
5436 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5437 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
5440 kvm_inject_gp(vcpu, 0);
5444 /* According to the Intel instruction reference, the memory operand
5445 * is read even if it isn't needed (e.g., for type==all)
5447 if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5448 vmx_instruction_info, false,
5449 sizeof(operand), &gva))
5452 if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
5453 kvm_inject_emulated_page_fault(vcpu, &e);
5457 if (operand.pcid >> 12 != 0) {
5458 kvm_inject_gp(vcpu, 0);
5462 pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
5465 case INVPCID_TYPE_INDIV_ADDR:
5466 if ((!pcid_enabled && (operand.pcid != 0)) ||
5467 is_noncanonical_address(operand.gla, vcpu)) {
5468 kvm_inject_gp(vcpu, 0);
5471 kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
5472 return kvm_skip_emulated_instruction(vcpu);
5474 case INVPCID_TYPE_SINGLE_CTXT:
5475 if (!pcid_enabled && (operand.pcid != 0)) {
5476 kvm_inject_gp(vcpu, 0);
5480 if (kvm_get_active_pcid(vcpu) == operand.pcid) {
5481 kvm_mmu_sync_roots(vcpu);
5482 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
5485 for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
5486 if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].pgd)
5488 roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
5490 kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
5492 * If neither the current cr3 nor any of the prev_roots use the
5493 * given PCID, then nothing needs to be done here because a
5494 * resync will happen anyway before switching to any other CR3.
5497 return kvm_skip_emulated_instruction(vcpu);
5499 case INVPCID_TYPE_ALL_NON_GLOBAL:
5501 * Currently, KVM doesn't mark global entries in the shadow
5502 * page tables, so a non-global flush just degenerates to a
5503 * global flush. If needed, we could optimize this later by
5504 * keeping track of global entries in shadow page tables.
5508 case INVPCID_TYPE_ALL_INCL_GLOBAL:
5509 kvm_mmu_unload(vcpu);
5510 return kvm_skip_emulated_instruction(vcpu);
5513 BUG(); /* We have already checked above that type <= 3 */
5517 static int handle_pml_full(struct kvm_vcpu *vcpu)
5519 unsigned long exit_qualification;
5521 trace_kvm_pml_full(vcpu->vcpu_id);
5523 exit_qualification = vmx_get_exit_qual(vcpu);
5526 * PML buffer FULL happened while executing iret from NMI,
5527 * "blocked by NMI" bit has to be set before next VM entry.
5529 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5531 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
5532 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5533 GUEST_INTR_STATE_NMI);
5536 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
5537 * here.., and there's no userspace involvement needed for PML.
5542 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
5544 struct vcpu_vmx *vmx = to_vmx(vcpu);
5546 if (!vmx->req_immediate_exit &&
5547 !unlikely(vmx->loaded_vmcs->hv_timer_soft_disabled))
5548 kvm_lapic_expired_hv_timer(vcpu);
5554 * When nested=0, all VMX instruction VM Exits filter here. The handlers
5555 * are overwritten by nested_vmx_setup() when nested=1.
5557 static int handle_vmx_instruction(struct kvm_vcpu *vcpu)
5559 kvm_queue_exception(vcpu, UD_VECTOR);
5563 static int handle_encls(struct kvm_vcpu *vcpu)
5566 * SGX virtualization is not yet supported. There is no software
5567 * enable bit for SGX, so we have to trap ENCLS and inject a #UD
5568 * to prevent the guest from executing ENCLS.
5570 kvm_queue_exception(vcpu, UD_VECTOR);
5575 * The exit handlers return 1 if the exit was handled fully and guest execution
5576 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
5577 * to be done to userspace and return 0.
5579 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
5580 [EXIT_REASON_EXCEPTION_NMI] = handle_exception_nmi,
5581 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
5582 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
5583 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
5584 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
5585 [EXIT_REASON_CR_ACCESS] = handle_cr,
5586 [EXIT_REASON_DR_ACCESS] = handle_dr,
5587 [EXIT_REASON_CPUID] = kvm_emulate_cpuid,
5588 [EXIT_REASON_MSR_READ] = kvm_emulate_rdmsr,
5589 [EXIT_REASON_MSR_WRITE] = kvm_emulate_wrmsr,
5590 [EXIT_REASON_INTERRUPT_WINDOW] = handle_interrupt_window,
5591 [EXIT_REASON_HLT] = kvm_emulate_halt,
5592 [EXIT_REASON_INVD] = handle_invd,
5593 [EXIT_REASON_INVLPG] = handle_invlpg,
5594 [EXIT_REASON_RDPMC] = handle_rdpmc,
5595 [EXIT_REASON_VMCALL] = handle_vmcall,
5596 [EXIT_REASON_VMCLEAR] = handle_vmx_instruction,
5597 [EXIT_REASON_VMLAUNCH] = handle_vmx_instruction,
5598 [EXIT_REASON_VMPTRLD] = handle_vmx_instruction,
5599 [EXIT_REASON_VMPTRST] = handle_vmx_instruction,
5600 [EXIT_REASON_VMREAD] = handle_vmx_instruction,
5601 [EXIT_REASON_VMRESUME] = handle_vmx_instruction,
5602 [EXIT_REASON_VMWRITE] = handle_vmx_instruction,
5603 [EXIT_REASON_VMOFF] = handle_vmx_instruction,
5604 [EXIT_REASON_VMON] = handle_vmx_instruction,
5605 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
5606 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
5607 [EXIT_REASON_APIC_WRITE] = handle_apic_write,
5608 [EXIT_REASON_EOI_INDUCED] = handle_apic_eoi_induced,
5609 [EXIT_REASON_WBINVD] = handle_wbinvd,
5610 [EXIT_REASON_XSETBV] = handle_xsetbv,
5611 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
5612 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
5613 [EXIT_REASON_GDTR_IDTR] = handle_desc,
5614 [EXIT_REASON_LDTR_TR] = handle_desc,
5615 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
5616 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
5617 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
5618 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_mwait,
5619 [EXIT_REASON_MONITOR_TRAP_FLAG] = handle_monitor_trap,
5620 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_monitor,
5621 [EXIT_REASON_INVEPT] = handle_vmx_instruction,
5622 [EXIT_REASON_INVVPID] = handle_vmx_instruction,
5623 [EXIT_REASON_RDRAND] = handle_invalid_op,
5624 [EXIT_REASON_RDSEED] = handle_invalid_op,
5625 [EXIT_REASON_PML_FULL] = handle_pml_full,
5626 [EXIT_REASON_INVPCID] = handle_invpcid,
5627 [EXIT_REASON_VMFUNC] = handle_vmx_instruction,
5628 [EXIT_REASON_PREEMPTION_TIMER] = handle_preemption_timer,
5629 [EXIT_REASON_ENCLS] = handle_encls,
5632 static const int kvm_vmx_max_exit_handlers =
5633 ARRAY_SIZE(kvm_vmx_exit_handlers);
5635 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
5637 *info1 = vmx_get_exit_qual(vcpu);
5638 *info2 = vmx_get_intr_info(vcpu);
5641 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
5644 __free_page(vmx->pml_pg);
5649 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
5651 struct vcpu_vmx *vmx = to_vmx(vcpu);
5655 pml_idx = vmcs_read16(GUEST_PML_INDEX);
5657 /* Do nothing if PML buffer is empty */
5658 if (pml_idx == (PML_ENTITY_NUM - 1))
5661 /* PML index always points to next available PML buffer entity */
5662 if (pml_idx >= PML_ENTITY_NUM)
5667 pml_buf = page_address(vmx->pml_pg);
5668 for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
5671 gpa = pml_buf[pml_idx];
5672 WARN_ON(gpa & (PAGE_SIZE - 1));
5673 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5676 /* reset PML index */
5677 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
5681 * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
5682 * Called before reporting dirty_bitmap to userspace.
5684 static void kvm_flush_pml_buffers(struct kvm *kvm)
5687 struct kvm_vcpu *vcpu;
5689 * We only need to kick vcpu out of guest mode here, as PML buffer
5690 * is flushed at beginning of all VMEXITs, and it's obvious that only
5691 * vcpus running in guest are possible to have unflushed GPAs in PML
5694 kvm_for_each_vcpu(i, vcpu, kvm)
5695 kvm_vcpu_kick(vcpu);
5698 static void vmx_dump_sel(char *name, uint32_t sel)
5700 pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
5701 name, vmcs_read16(sel),
5702 vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
5703 vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
5704 vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
5707 static void vmx_dump_dtsel(char *name, uint32_t limit)
5709 pr_err("%s limit=0x%08x, base=0x%016lx\n",
5710 name, vmcs_read32(limit),
5711 vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
5714 void dump_vmcs(void)
5716 u32 vmentry_ctl, vmexit_ctl;
5717 u32 cpu_based_exec_ctrl, pin_based_exec_ctrl, secondary_exec_control;
5722 if (!dump_invalid_vmcs) {
5723 pr_warn_ratelimited("set kvm_intel.dump_invalid_vmcs=1 to dump internal KVM state.\n");
5727 vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
5728 vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
5729 cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5730 pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
5731 cr4 = vmcs_readl(GUEST_CR4);
5732 efer = vmcs_read64(GUEST_IA32_EFER);
5733 secondary_exec_control = 0;
5734 if (cpu_has_secondary_exec_ctrls())
5735 secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
5737 pr_err("*** Guest State ***\n");
5738 pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5739 vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
5740 vmcs_readl(CR0_GUEST_HOST_MASK));
5741 pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5742 cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
5743 pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
5744 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
5745 (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
5747 pr_err("PDPTR0 = 0x%016llx PDPTR1 = 0x%016llx\n",
5748 vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
5749 pr_err("PDPTR2 = 0x%016llx PDPTR3 = 0x%016llx\n",
5750 vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
5752 pr_err("RSP = 0x%016lx RIP = 0x%016lx\n",
5753 vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
5754 pr_err("RFLAGS=0x%08lx DR7 = 0x%016lx\n",
5755 vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
5756 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5757 vmcs_readl(GUEST_SYSENTER_ESP),
5758 vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
5759 vmx_dump_sel("CS: ", GUEST_CS_SELECTOR);
5760 vmx_dump_sel("DS: ", GUEST_DS_SELECTOR);
5761 vmx_dump_sel("SS: ", GUEST_SS_SELECTOR);
5762 vmx_dump_sel("ES: ", GUEST_ES_SELECTOR);
5763 vmx_dump_sel("FS: ", GUEST_FS_SELECTOR);
5764 vmx_dump_sel("GS: ", GUEST_GS_SELECTOR);
5765 vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
5766 vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
5767 vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
5768 vmx_dump_sel("TR: ", GUEST_TR_SELECTOR);
5769 if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
5770 (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
5771 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
5772 efer, vmcs_read64(GUEST_IA32_PAT));
5773 pr_err("DebugCtl = 0x%016llx DebugExceptions = 0x%016lx\n",
5774 vmcs_read64(GUEST_IA32_DEBUGCTL),
5775 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
5776 if (cpu_has_load_perf_global_ctrl() &&
5777 vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
5778 pr_err("PerfGlobCtl = 0x%016llx\n",
5779 vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
5780 if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
5781 pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
5782 pr_err("Interruptibility = %08x ActivityState = %08x\n",
5783 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
5784 vmcs_read32(GUEST_ACTIVITY_STATE));
5785 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
5786 pr_err("InterruptStatus = %04x\n",
5787 vmcs_read16(GUEST_INTR_STATUS));
5789 pr_err("*** Host State ***\n");
5790 pr_err("RIP = 0x%016lx RSP = 0x%016lx\n",
5791 vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
5792 pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
5793 vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
5794 vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
5795 vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
5796 vmcs_read16(HOST_TR_SELECTOR));
5797 pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
5798 vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
5799 vmcs_readl(HOST_TR_BASE));
5800 pr_err("GDTBase=%016lx IDTBase=%016lx\n",
5801 vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
5802 pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
5803 vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
5804 vmcs_readl(HOST_CR4));
5805 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5806 vmcs_readl(HOST_IA32_SYSENTER_ESP),
5807 vmcs_read32(HOST_IA32_SYSENTER_CS),
5808 vmcs_readl(HOST_IA32_SYSENTER_EIP));
5809 if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
5810 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
5811 vmcs_read64(HOST_IA32_EFER),
5812 vmcs_read64(HOST_IA32_PAT));
5813 if (cpu_has_load_perf_global_ctrl() &&
5814 vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
5815 pr_err("PerfGlobCtl = 0x%016llx\n",
5816 vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
5818 pr_err("*** Control State ***\n");
5819 pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
5820 pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
5821 pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
5822 pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
5823 vmcs_read32(EXCEPTION_BITMAP),
5824 vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
5825 vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
5826 pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
5827 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
5828 vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
5829 vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
5830 pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
5831 vmcs_read32(VM_EXIT_INTR_INFO),
5832 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
5833 vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
5834 pr_err(" reason=%08x qualification=%016lx\n",
5835 vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
5836 pr_err("IDTVectoring: info=%08x errcode=%08x\n",
5837 vmcs_read32(IDT_VECTORING_INFO_FIELD),
5838 vmcs_read32(IDT_VECTORING_ERROR_CODE));
5839 pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
5840 if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
5841 pr_err("TSC Multiplier = 0x%016llx\n",
5842 vmcs_read64(TSC_MULTIPLIER));
5843 if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW) {
5844 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
5845 u16 status = vmcs_read16(GUEST_INTR_STATUS);
5846 pr_err("SVI|RVI = %02x|%02x ", status >> 8, status & 0xff);
5848 pr_cont("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
5849 if (secondary_exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
5850 pr_err("APIC-access addr = 0x%016llx ", vmcs_read64(APIC_ACCESS_ADDR));
5851 pr_cont("virt-APIC addr = 0x%016llx\n", vmcs_read64(VIRTUAL_APIC_PAGE_ADDR));
5853 if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
5854 pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
5855 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
5856 pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
5857 n = vmcs_read32(CR3_TARGET_COUNT);
5858 for (i = 0; i + 1 < n; i += 4)
5859 pr_err("CR3 target%u=%016lx target%u=%016lx\n",
5860 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
5861 i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
5863 pr_err("CR3 target%u=%016lx\n",
5864 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
5865 if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
5866 pr_err("PLE Gap=%08x Window=%08x\n",
5867 vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
5868 if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
5869 pr_err("Virtual processor ID = 0x%04x\n",
5870 vmcs_read16(VIRTUAL_PROCESSOR_ID));
5874 * The guest has exited. See if we can fix it or if we need userspace
5877 static int vmx_handle_exit(struct kvm_vcpu *vcpu,
5878 enum exit_fastpath_completion exit_fastpath)
5880 struct vcpu_vmx *vmx = to_vmx(vcpu);
5881 u32 exit_reason = vmx->exit_reason;
5882 u32 vectoring_info = vmx->idt_vectoring_info;
5884 trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
5887 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
5888 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
5889 * querying dirty_bitmap, we only need to kick all vcpus out of guest
5890 * mode as if vcpus is in root mode, the PML buffer must has been
5894 vmx_flush_pml_buffer(vcpu);
5896 /* If guest state is invalid, start emulating */
5897 if (vmx->emulation_required)
5898 return handle_invalid_guest_state(vcpu);
5900 if (is_guest_mode(vcpu)) {
5902 * The host physical addresses of some pages of guest memory
5903 * are loaded into the vmcs02 (e.g. vmcs12's Virtual APIC
5904 * Page). The CPU may write to these pages via their host
5905 * physical address while L2 is running, bypassing any
5906 * address-translation-based dirty tracking (e.g. EPT write
5909 * Mark them dirty on every exit from L2 to prevent them from
5910 * getting out of sync with dirty tracking.
5912 nested_mark_vmcs12_pages_dirty(vcpu);
5914 if (nested_vmx_reflect_vmexit(vcpu))
5918 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
5920 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5921 vcpu->run->fail_entry.hardware_entry_failure_reason
5926 if (unlikely(vmx->fail)) {
5928 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5929 vcpu->run->fail_entry.hardware_entry_failure_reason
5930 = vmcs_read32(VM_INSTRUCTION_ERROR);
5936 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
5937 * delivery event since it indicates guest is accessing MMIO.
5938 * The vm-exit can be triggered again after return to guest that
5939 * will cause infinite loop.
5941 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
5942 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
5943 exit_reason != EXIT_REASON_EPT_VIOLATION &&
5944 exit_reason != EXIT_REASON_PML_FULL &&
5945 exit_reason != EXIT_REASON_TASK_SWITCH)) {
5946 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5947 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
5948 vcpu->run->internal.ndata = 3;
5949 vcpu->run->internal.data[0] = vectoring_info;
5950 vcpu->run->internal.data[1] = exit_reason;
5951 vcpu->run->internal.data[2] = vcpu->arch.exit_qualification;
5952 if (exit_reason == EXIT_REASON_EPT_MISCONFIG) {
5953 vcpu->run->internal.ndata++;
5954 vcpu->run->internal.data[3] =
5955 vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5960 if (unlikely(!enable_vnmi &&
5961 vmx->loaded_vmcs->soft_vnmi_blocked)) {
5962 if (vmx_interrupt_allowed(vcpu)) {
5963 vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5964 } else if (vmx->loaded_vmcs->vnmi_blocked_time > 1000000000LL &&
5965 vcpu->arch.nmi_pending) {
5967 * This CPU don't support us in finding the end of an
5968 * NMI-blocked window if the guest runs with IRQs
5969 * disabled. So we pull the trigger after 1 s of
5970 * futile waiting, but inform the user about this.
5972 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
5973 "state on VCPU %d after 1 s timeout\n",
5974 __func__, vcpu->vcpu_id);
5975 vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5979 if (exit_fastpath == EXIT_FASTPATH_SKIP_EMUL_INS) {
5980 kvm_skip_emulated_instruction(vcpu);
5984 if (exit_reason >= kvm_vmx_max_exit_handlers)
5985 goto unexpected_vmexit;
5986 #ifdef CONFIG_RETPOLINE
5987 if (exit_reason == EXIT_REASON_MSR_WRITE)
5988 return kvm_emulate_wrmsr(vcpu);
5989 else if (exit_reason == EXIT_REASON_PREEMPTION_TIMER)
5990 return handle_preemption_timer(vcpu);
5991 else if (exit_reason == EXIT_REASON_INTERRUPT_WINDOW)
5992 return handle_interrupt_window(vcpu);
5993 else if (exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
5994 return handle_external_interrupt(vcpu);
5995 else if (exit_reason == EXIT_REASON_HLT)
5996 return kvm_emulate_halt(vcpu);
5997 else if (exit_reason == EXIT_REASON_EPT_MISCONFIG)
5998 return handle_ept_misconfig(vcpu);
6001 exit_reason = array_index_nospec(exit_reason,
6002 kvm_vmx_max_exit_handlers);
6003 if (!kvm_vmx_exit_handlers[exit_reason])
6004 goto unexpected_vmexit;
6006 return kvm_vmx_exit_handlers[exit_reason](vcpu);
6009 vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n", exit_reason);
6011 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6012 vcpu->run->internal.suberror =
6013 KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON;
6014 vcpu->run->internal.ndata = 1;
6015 vcpu->run->internal.data[0] = exit_reason;
6020 * Software based L1D cache flush which is used when microcode providing
6021 * the cache control MSR is not loaded.
6023 * The L1D cache is 32 KiB on Nehalem and later microarchitectures, but to
6024 * flush it is required to read in 64 KiB because the replacement algorithm
6025 * is not exactly LRU. This could be sized at runtime via topology
6026 * information but as all relevant affected CPUs have 32KiB L1D cache size
6027 * there is no point in doing so.
6029 static void vmx_l1d_flush(struct kvm_vcpu *vcpu)
6031 int size = PAGE_SIZE << L1D_CACHE_ORDER;
6034 * This code is only executed when the the flush mode is 'cond' or
6037 if (static_branch_likely(&vmx_l1d_flush_cond)) {
6041 * Clear the per-vcpu flush bit, it gets set again
6042 * either from vcpu_run() or from one of the unsafe
6045 flush_l1d = vcpu->arch.l1tf_flush_l1d;
6046 vcpu->arch.l1tf_flush_l1d = false;
6049 * Clear the per-cpu flush bit, it gets set again from
6050 * the interrupt handlers.
6052 flush_l1d |= kvm_get_cpu_l1tf_flush_l1d();
6053 kvm_clear_cpu_l1tf_flush_l1d();
6059 vcpu->stat.l1d_flush++;
6061 if (static_cpu_has(X86_FEATURE_FLUSH_L1D)) {
6062 wrmsrl(MSR_IA32_FLUSH_CMD, L1D_FLUSH);
6067 /* First ensure the pages are in the TLB */
6068 "xorl %%eax, %%eax\n"
6069 ".Lpopulate_tlb:\n\t"
6070 "movzbl (%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
6071 "addl $4096, %%eax\n\t"
6072 "cmpl %%eax, %[size]\n\t"
6073 "jne .Lpopulate_tlb\n\t"
6074 "xorl %%eax, %%eax\n\t"
6076 /* Now fill the cache */
6077 "xorl %%eax, %%eax\n"
6079 "movzbl (%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
6080 "addl $64, %%eax\n\t"
6081 "cmpl %%eax, %[size]\n\t"
6082 "jne .Lfill_cache\n\t"
6084 :: [flush_pages] "r" (vmx_l1d_flush_pages),
6086 : "eax", "ebx", "ecx", "edx");
6089 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
6091 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6094 if (is_guest_mode(vcpu) &&
6095 nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
6098 tpr_threshold = (irr == -1 || tpr < irr) ? 0 : irr;
6099 if (is_guest_mode(vcpu))
6100 to_vmx(vcpu)->nested.l1_tpr_threshold = tpr_threshold;
6102 vmcs_write32(TPR_THRESHOLD, tpr_threshold);
6105 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu)
6107 struct vcpu_vmx *vmx = to_vmx(vcpu);
6108 u32 sec_exec_control;
6110 if (!lapic_in_kernel(vcpu))
6113 if (!flexpriority_enabled &&
6114 !cpu_has_vmx_virtualize_x2apic_mode())
6117 /* Postpone execution until vmcs01 is the current VMCS. */
6118 if (is_guest_mode(vcpu)) {
6119 vmx->nested.change_vmcs01_virtual_apic_mode = true;
6123 sec_exec_control = secondary_exec_controls_get(vmx);
6124 sec_exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6125 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
6127 switch (kvm_get_apic_mode(vcpu)) {
6128 case LAPIC_MODE_INVALID:
6129 WARN_ONCE(true, "Invalid local APIC state");
6130 case LAPIC_MODE_DISABLED:
6132 case LAPIC_MODE_XAPIC:
6133 if (flexpriority_enabled) {
6135 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6136 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
6139 * Flush the TLB, reloading the APIC access page will
6140 * only do so if its physical address has changed, but
6141 * the guest may have inserted a non-APIC mapping into
6142 * the TLB while the APIC access page was disabled.
6144 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
6147 case LAPIC_MODE_X2APIC:
6148 if (cpu_has_vmx_virtualize_x2apic_mode())
6150 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
6153 secondary_exec_controls_set(vmx, sec_exec_control);
6155 vmx_update_msr_bitmap(vcpu);
6158 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu)
6162 /* Defer reload until vmcs01 is the current VMCS. */
6163 if (is_guest_mode(vcpu)) {
6164 to_vmx(vcpu)->nested.reload_vmcs01_apic_access_page = true;
6168 if (!(secondary_exec_controls_get(to_vmx(vcpu)) &
6169 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
6172 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6173 if (is_error_page(page))
6176 vmcs_write64(APIC_ACCESS_ADDR, page_to_phys(page));
6177 vmx_flush_tlb_current(vcpu);
6180 * Do not pin apic access page in memory, the MMU notifier
6181 * will call us again if it is migrated or swapped out.
6186 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
6194 status = vmcs_read16(GUEST_INTR_STATUS);
6196 if (max_isr != old) {
6198 status |= max_isr << 8;
6199 vmcs_write16(GUEST_INTR_STATUS, status);
6203 static void vmx_set_rvi(int vector)
6211 status = vmcs_read16(GUEST_INTR_STATUS);
6212 old = (u8)status & 0xff;
6213 if ((u8)vector != old) {
6215 status |= (u8)vector;
6216 vmcs_write16(GUEST_INTR_STATUS, status);
6220 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
6223 * When running L2, updating RVI is only relevant when
6224 * vmcs12 virtual-interrupt-delivery enabled.
6225 * However, it can be enabled only when L1 also
6226 * intercepts external-interrupts and in that case
6227 * we should not update vmcs02 RVI but instead intercept
6228 * interrupt. Therefore, do nothing when running L2.
6230 if (!is_guest_mode(vcpu))
6231 vmx_set_rvi(max_irr);
6234 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
6236 struct vcpu_vmx *vmx = to_vmx(vcpu);
6238 bool max_irr_updated;
6240 WARN_ON(!vcpu->arch.apicv_active);
6241 if (pi_test_on(&vmx->pi_desc)) {
6242 pi_clear_on(&vmx->pi_desc);
6244 * IOMMU can write to PID.ON, so the barrier matters even on UP.
6245 * But on x86 this is just a compiler barrier anyway.
6247 smp_mb__after_atomic();
6249 kvm_apic_update_irr(vcpu, vmx->pi_desc.pir, &max_irr);
6252 * If we are running L2 and L1 has a new pending interrupt
6253 * which can be injected, we should re-evaluate
6254 * what should be done with this new L1 interrupt.
6255 * If L1 intercepts external-interrupts, we should
6256 * exit from L2 to L1. Otherwise, interrupt should be
6257 * delivered directly to L2.
6259 if (is_guest_mode(vcpu) && max_irr_updated) {
6260 if (nested_exit_on_intr(vcpu))
6261 kvm_vcpu_exiting_guest_mode(vcpu);
6263 kvm_make_request(KVM_REQ_EVENT, vcpu);
6266 max_irr = kvm_lapic_find_highest_irr(vcpu);
6268 vmx_hwapic_irr_update(vcpu, max_irr);
6272 static bool vmx_dy_apicv_has_pending_interrupt(struct kvm_vcpu *vcpu)
6274 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
6276 return pi_test_on(pi_desc) ||
6277 (pi_test_sn(pi_desc) && !pi_is_pir_empty(pi_desc));
6280 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
6282 if (!kvm_vcpu_apicv_active(vcpu))
6285 vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
6286 vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
6287 vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
6288 vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
6291 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
6293 struct vcpu_vmx *vmx = to_vmx(vcpu);
6295 pi_clear_on(&vmx->pi_desc);
6296 memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
6299 static void handle_exception_nmi_irqoff(struct vcpu_vmx *vmx)
6301 u32 intr_info = vmx_get_intr_info(&vmx->vcpu);
6303 /* if exit due to PF check for async PF */
6304 if (is_page_fault(intr_info)) {
6305 vmx->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason();
6306 /* Handle machine checks before interrupts are enabled */
6307 } else if (is_machine_check(intr_info)) {
6308 kvm_machine_check();
6309 /* We need to handle NMIs before interrupts are enabled */
6310 } else if (is_nmi(intr_info)) {
6311 kvm_before_interrupt(&vmx->vcpu);
6313 kvm_after_interrupt(&vmx->vcpu);
6317 static void handle_external_interrupt_irqoff(struct kvm_vcpu *vcpu)
6319 unsigned int vector;
6320 unsigned long entry;
6321 #ifdef CONFIG_X86_64
6325 u32 intr_info = vmx_get_intr_info(vcpu);
6327 if (WARN_ONCE(!is_external_intr(intr_info),
6328 "KVM: unexpected VM-Exit interrupt info: 0x%x", intr_info))
6331 vector = intr_info & INTR_INFO_VECTOR_MASK;
6332 desc = (gate_desc *)host_idt_base + vector;
6333 entry = gate_offset(desc);
6335 kvm_before_interrupt(vcpu);
6338 #ifdef CONFIG_X86_64
6339 "mov %%" _ASM_SP ", %[sp]\n\t"
6340 "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
6345 __ASM_SIZE(push) " $%c[cs]\n\t"
6348 #ifdef CONFIG_X86_64
6353 [thunk_target]"r"(entry),
6354 [ss]"i"(__KERNEL_DS),
6355 [cs]"i"(__KERNEL_CS)
6358 kvm_after_interrupt(vcpu);
6360 STACK_FRAME_NON_STANDARD(handle_external_interrupt_irqoff);
6362 static void vmx_handle_exit_irqoff(struct kvm_vcpu *vcpu,
6363 enum exit_fastpath_completion *exit_fastpath)
6365 struct vcpu_vmx *vmx = to_vmx(vcpu);
6367 if (vmx->exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
6368 handle_external_interrupt_irqoff(vcpu);
6369 else if (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI)
6370 handle_exception_nmi_irqoff(vmx);
6371 else if (!is_guest_mode(vcpu) &&
6372 vmx->exit_reason == EXIT_REASON_MSR_WRITE)
6373 *exit_fastpath = handle_fastpath_set_msr_irqoff(vcpu);
6376 static bool vmx_has_emulated_msr(int index)
6379 case MSR_IA32_SMBASE:
6381 * We cannot do SMM unless we can run the guest in big
6384 return enable_unrestricted_guest || emulate_invalid_guest_state;
6385 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
6387 case MSR_AMD64_VIRT_SPEC_CTRL:
6388 /* This is AMD only. */
6395 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
6400 bool idtv_info_valid;
6402 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6405 if (vmx->loaded_vmcs->nmi_known_unmasked)
6408 exit_intr_info = vmx_get_intr_info(&vmx->vcpu);
6409 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
6410 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
6412 * SDM 3: 27.7.1.2 (September 2008)
6413 * Re-set bit "block by NMI" before VM entry if vmexit caused by
6414 * a guest IRET fault.
6415 * SDM 3: 23.2.2 (September 2008)
6416 * Bit 12 is undefined in any of the following cases:
6417 * If the VM exit sets the valid bit in the IDT-vectoring
6418 * information field.
6419 * If the VM exit is due to a double fault.
6421 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
6422 vector != DF_VECTOR && !idtv_info_valid)
6423 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
6424 GUEST_INTR_STATE_NMI);
6426 vmx->loaded_vmcs->nmi_known_unmasked =
6427 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
6428 & GUEST_INTR_STATE_NMI);
6429 } else if (unlikely(vmx->loaded_vmcs->soft_vnmi_blocked))
6430 vmx->loaded_vmcs->vnmi_blocked_time +=
6431 ktime_to_ns(ktime_sub(ktime_get(),
6432 vmx->loaded_vmcs->entry_time));
6435 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
6436 u32 idt_vectoring_info,
6437 int instr_len_field,
6438 int error_code_field)
6442 bool idtv_info_valid;
6444 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6446 vcpu->arch.nmi_injected = false;
6447 kvm_clear_exception_queue(vcpu);
6448 kvm_clear_interrupt_queue(vcpu);
6450 if (!idtv_info_valid)
6453 kvm_make_request(KVM_REQ_EVENT, vcpu);
6455 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
6456 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
6459 case INTR_TYPE_NMI_INTR:
6460 vcpu->arch.nmi_injected = true;
6462 * SDM 3: 27.7.1.2 (September 2008)
6463 * Clear bit "block by NMI" before VM entry if a NMI
6466 vmx_set_nmi_mask(vcpu, false);
6468 case INTR_TYPE_SOFT_EXCEPTION:
6469 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6471 case INTR_TYPE_HARD_EXCEPTION:
6472 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
6473 u32 err = vmcs_read32(error_code_field);
6474 kvm_requeue_exception_e(vcpu, vector, err);
6476 kvm_requeue_exception(vcpu, vector);
6478 case INTR_TYPE_SOFT_INTR:
6479 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6481 case INTR_TYPE_EXT_INTR:
6482 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
6489 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
6491 __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
6492 VM_EXIT_INSTRUCTION_LEN,
6493 IDT_VECTORING_ERROR_CODE);
6496 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
6498 __vmx_complete_interrupts(vcpu,
6499 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
6500 VM_ENTRY_INSTRUCTION_LEN,
6501 VM_ENTRY_EXCEPTION_ERROR_CODE);
6503 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
6506 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
6509 struct perf_guest_switch_msr *msrs;
6511 msrs = perf_guest_get_msrs(&nr_msrs);
6516 for (i = 0; i < nr_msrs; i++)
6517 if (msrs[i].host == msrs[i].guest)
6518 clear_atomic_switch_msr(vmx, msrs[i].msr);
6520 add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
6521 msrs[i].host, false);
6524 static void atomic_switch_umwait_control_msr(struct vcpu_vmx *vmx)
6526 u32 host_umwait_control;
6528 if (!vmx_has_waitpkg(vmx))
6531 host_umwait_control = get_umwait_control_msr();
6533 if (vmx->msr_ia32_umwait_control != host_umwait_control)
6534 add_atomic_switch_msr(vmx, MSR_IA32_UMWAIT_CONTROL,
6535 vmx->msr_ia32_umwait_control,
6536 host_umwait_control, false);
6538 clear_atomic_switch_msr(vmx, MSR_IA32_UMWAIT_CONTROL);
6541 static void vmx_update_hv_timer(struct kvm_vcpu *vcpu)
6543 struct vcpu_vmx *vmx = to_vmx(vcpu);
6547 if (vmx->req_immediate_exit) {
6548 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, 0);
6549 vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6550 } else if (vmx->hv_deadline_tsc != -1) {
6552 if (vmx->hv_deadline_tsc > tscl)
6553 /* set_hv_timer ensures the delta fits in 32-bits */
6554 delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
6555 cpu_preemption_timer_multi);
6559 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
6560 vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6561 } else if (!vmx->loaded_vmcs->hv_timer_soft_disabled) {
6562 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, -1);
6563 vmx->loaded_vmcs->hv_timer_soft_disabled = true;
6567 void vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp)
6569 if (unlikely(host_rsp != vmx->loaded_vmcs->host_state.rsp)) {
6570 vmx->loaded_vmcs->host_state.rsp = host_rsp;
6571 vmcs_writel(HOST_RSP, host_rsp);
6575 bool __vmx_vcpu_run(struct vcpu_vmx *vmx, unsigned long *regs, bool launched);
6577 static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
6579 struct vcpu_vmx *vmx = to_vmx(vcpu);
6580 unsigned long cr3, cr4;
6582 /* Record the guest's net vcpu time for enforced NMI injections. */
6583 if (unlikely(!enable_vnmi &&
6584 vmx->loaded_vmcs->soft_vnmi_blocked))
6585 vmx->loaded_vmcs->entry_time = ktime_get();
6587 /* Don't enter VMX if guest state is invalid, let the exit handler
6588 start emulation until we arrive back to a valid state */
6589 if (vmx->emulation_required)
6592 if (vmx->ple_window_dirty) {
6593 vmx->ple_window_dirty = false;
6594 vmcs_write32(PLE_WINDOW, vmx->ple_window);
6598 * We did this in prepare_switch_to_guest, because it needs to
6599 * be within srcu_read_lock.
6601 WARN_ON_ONCE(vmx->nested.need_vmcs12_to_shadow_sync);
6603 if (kvm_register_is_dirty(vcpu, VCPU_REGS_RSP))
6604 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
6605 if (kvm_register_is_dirty(vcpu, VCPU_REGS_RIP))
6606 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
6608 cr3 = __get_current_cr3_fast();
6609 if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
6610 vmcs_writel(HOST_CR3, cr3);
6611 vmx->loaded_vmcs->host_state.cr3 = cr3;
6614 cr4 = cr4_read_shadow();
6615 if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
6616 vmcs_writel(HOST_CR4, cr4);
6617 vmx->loaded_vmcs->host_state.cr4 = cr4;
6620 /* When single-stepping over STI and MOV SS, we must clear the
6621 * corresponding interruptibility bits in the guest state. Otherwise
6622 * vmentry fails as it then expects bit 14 (BS) in pending debug
6623 * exceptions being set, but that's not correct for the guest debugging
6625 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6626 vmx_set_interrupt_shadow(vcpu, 0);
6628 kvm_load_guest_xsave_state(vcpu);
6630 if (static_cpu_has(X86_FEATURE_PKU) &&
6631 kvm_read_cr4_bits(vcpu, X86_CR4_PKE) &&
6632 vcpu->arch.pkru != vmx->host_pkru)
6633 __write_pkru(vcpu->arch.pkru);
6635 pt_guest_enter(vmx);
6637 if (vcpu_to_pmu(vcpu)->version)
6638 atomic_switch_perf_msrs(vmx);
6639 atomic_switch_umwait_control_msr(vmx);
6641 if (enable_preemption_timer)
6642 vmx_update_hv_timer(vcpu);
6644 if (lapic_in_kernel(vcpu) &&
6645 vcpu->arch.apic->lapic_timer.timer_advance_ns)
6646 kvm_wait_lapic_expire(vcpu);
6649 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
6650 * it's non-zero. Since vmentry is serialising on affected CPUs, there
6651 * is no need to worry about the conditional branch over the wrmsr
6652 * being speculatively taken.
6654 x86_spec_ctrl_set_guest(vmx->spec_ctrl, 0);
6656 /* L1D Flush includes CPU buffer clear to mitigate MDS */
6657 if (static_branch_unlikely(&vmx_l1d_should_flush))
6658 vmx_l1d_flush(vcpu);
6659 else if (static_branch_unlikely(&mds_user_clear))
6660 mds_clear_cpu_buffers();
6662 if (vcpu->arch.cr2 != read_cr2())
6663 write_cr2(vcpu->arch.cr2);
6665 vmx->fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
6666 vmx->loaded_vmcs->launched);
6668 vcpu->arch.cr2 = read_cr2();
6671 * We do not use IBRS in the kernel. If this vCPU has used the
6672 * SPEC_CTRL MSR it may have left it on; save the value and
6673 * turn it off. This is much more efficient than blindly adding
6674 * it to the atomic save/restore list. Especially as the former
6675 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
6677 * For non-nested case:
6678 * If the L01 MSR bitmap does not intercept the MSR, then we need to
6682 * If the L02 MSR bitmap does not intercept the MSR, then we need to
6685 if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
6686 vmx->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
6688 x86_spec_ctrl_restore_host(vmx->spec_ctrl, 0);
6690 /* All fields are clean at this point */
6691 if (static_branch_unlikely(&enable_evmcs))
6692 current_evmcs->hv_clean_fields |=
6693 HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
6695 if (static_branch_unlikely(&enable_evmcs))
6696 current_evmcs->hv_vp_id = vcpu->arch.hyperv.vp_index;
6698 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
6699 if (vmx->host_debugctlmsr)
6700 update_debugctlmsr(vmx->host_debugctlmsr);
6702 #ifndef CONFIG_X86_64
6704 * The sysexit path does not restore ds/es, so we must set them to
6705 * a reasonable value ourselves.
6707 * We can't defer this to vmx_prepare_switch_to_host() since that
6708 * function may be executed in interrupt context, which saves and
6709 * restore segments around it, nullifying its effect.
6711 loadsegment(ds, __USER_DS);
6712 loadsegment(es, __USER_DS);
6715 vmx_register_cache_reset(vcpu);
6720 * eager fpu is enabled if PKEY is supported and CR4 is switched
6721 * back on host, so it is safe to read guest PKRU from current
6724 if (static_cpu_has(X86_FEATURE_PKU) &&
6725 kvm_read_cr4_bits(vcpu, X86_CR4_PKE)) {
6726 vcpu->arch.pkru = rdpkru();
6727 if (vcpu->arch.pkru != vmx->host_pkru)
6728 __write_pkru(vmx->host_pkru);
6731 kvm_load_host_xsave_state(vcpu);
6733 vmx->nested.nested_run_pending = 0;
6734 vmx->idt_vectoring_info = 0;
6736 vmx->exit_reason = vmx->fail ? 0xdead : vmcs_read32(VM_EXIT_REASON);
6737 if ((u16)vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY)
6738 kvm_machine_check();
6740 if (vmx->fail || (vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
6743 vmx->loaded_vmcs->launched = 1;
6744 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
6746 vmx_recover_nmi_blocking(vmx);
6747 vmx_complete_interrupts(vmx);
6750 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
6752 struct vcpu_vmx *vmx = to_vmx(vcpu);
6755 vmx_destroy_pml_buffer(vmx);
6756 free_vpid(vmx->vpid);
6757 nested_vmx_free_vcpu(vcpu);
6758 free_loaded_vmcs(vmx->loaded_vmcs);
6761 static int vmx_create_vcpu(struct kvm_vcpu *vcpu)
6763 struct vcpu_vmx *vmx;
6764 unsigned long *msr_bitmap;
6767 BUILD_BUG_ON(offsetof(struct vcpu_vmx, vcpu) != 0);
6772 vmx->vpid = allocate_vpid();
6775 * If PML is turned on, failure on enabling PML just results in failure
6776 * of creating the vcpu, therefore we can simplify PML logic (by
6777 * avoiding dealing with cases, such as enabling PML partially on vcpus
6778 * for the guest), etc.
6781 vmx->pml_pg = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
6786 BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) != NR_SHARED_MSRS);
6788 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
6789 u32 index = vmx_msr_index[i];
6790 u32 data_low, data_high;
6793 if (rdmsr_safe(index, &data_low, &data_high) < 0)
6795 if (wrmsr_safe(index, data_low, data_high) < 0)
6798 vmx->guest_msrs[j].index = i;
6799 vmx->guest_msrs[j].data = 0;
6801 case MSR_IA32_TSX_CTRL:
6803 * No need to pass TSX_CTRL_CPUID_CLEAR through, so
6804 * let's avoid changing CPUID bits under the host
6807 vmx->guest_msrs[j].mask = ~(u64)TSX_CTRL_CPUID_CLEAR;
6810 vmx->guest_msrs[j].mask = -1ull;
6816 err = alloc_loaded_vmcs(&vmx->vmcs01);
6820 msr_bitmap = vmx->vmcs01.msr_bitmap;
6821 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_TSC, MSR_TYPE_R);
6822 vmx_disable_intercept_for_msr(msr_bitmap, MSR_FS_BASE, MSR_TYPE_RW);
6823 vmx_disable_intercept_for_msr(msr_bitmap, MSR_GS_BASE, MSR_TYPE_RW);
6824 vmx_disable_intercept_for_msr(msr_bitmap, MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
6825 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_CS, MSR_TYPE_RW);
6826 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_ESP, MSR_TYPE_RW);
6827 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_EIP, MSR_TYPE_RW);
6828 if (kvm_cstate_in_guest(vcpu->kvm)) {
6829 vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C1_RES, MSR_TYPE_R);
6830 vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C3_RESIDENCY, MSR_TYPE_R);
6831 vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C6_RESIDENCY, MSR_TYPE_R);
6832 vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C7_RESIDENCY, MSR_TYPE_R);
6834 vmx->msr_bitmap_mode = 0;
6836 vmx->loaded_vmcs = &vmx->vmcs01;
6838 vmx_vcpu_load(vcpu, cpu);
6843 if (cpu_need_virtualize_apic_accesses(vcpu)) {
6844 err = alloc_apic_access_page(vcpu->kvm);
6849 if (enable_ept && !enable_unrestricted_guest) {
6850 err = init_rmode_identity_map(vcpu->kvm);
6856 nested_vmx_setup_ctls_msrs(&vmx->nested.msrs,
6857 vmx_capability.ept);
6859 memset(&vmx->nested.msrs, 0, sizeof(vmx->nested.msrs));
6861 vmx->nested.posted_intr_nv = -1;
6862 vmx->nested.current_vmptr = -1ull;
6864 vcpu->arch.microcode_version = 0x100000000ULL;
6865 vmx->msr_ia32_feature_control_valid_bits = FEAT_CTL_LOCKED;
6868 * Enforce invariant: pi_desc.nv is always either POSTED_INTR_VECTOR
6869 * or POSTED_INTR_WAKEUP_VECTOR.
6871 vmx->pi_desc.nv = POSTED_INTR_VECTOR;
6872 vmx->pi_desc.sn = 1;
6874 vmx->ept_pointer = INVALID_PAGE;
6879 free_loaded_vmcs(vmx->loaded_vmcs);
6881 vmx_destroy_pml_buffer(vmx);
6883 free_vpid(vmx->vpid);
6887 #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"
6888 #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"
6890 static int vmx_vm_init(struct kvm *kvm)
6892 spin_lock_init(&to_kvm_vmx(kvm)->ept_pointer_lock);
6895 kvm->arch.pause_in_guest = true;
6897 if (boot_cpu_has(X86_BUG_L1TF) && enable_ept) {
6898 switch (l1tf_mitigation) {
6899 case L1TF_MITIGATION_OFF:
6900 case L1TF_MITIGATION_FLUSH_NOWARN:
6901 /* 'I explicitly don't care' is set */
6903 case L1TF_MITIGATION_FLUSH:
6904 case L1TF_MITIGATION_FLUSH_NOSMT:
6905 case L1TF_MITIGATION_FULL:
6907 * Warn upon starting the first VM in a potentially
6908 * insecure environment.
6910 if (sched_smt_active())
6911 pr_warn_once(L1TF_MSG_SMT);
6912 if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER)
6913 pr_warn_once(L1TF_MSG_L1D);
6915 case L1TF_MITIGATION_FULL_FORCE:
6916 /* Flush is enforced */
6920 kvm_apicv_init(kvm, enable_apicv);
6924 static int __init vmx_check_processor_compat(void)
6926 struct vmcs_config vmcs_conf;
6927 struct vmx_capability vmx_cap;
6929 if (!this_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
6930 !this_cpu_has(X86_FEATURE_VMX)) {
6931 pr_err("kvm: VMX is disabled on CPU %d\n", smp_processor_id());
6935 if (setup_vmcs_config(&vmcs_conf, &vmx_cap) < 0)
6938 nested_vmx_setup_ctls_msrs(&vmcs_conf.nested, vmx_cap.ept);
6939 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
6940 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
6941 smp_processor_id());
6947 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
6952 /* We wanted to honor guest CD/MTRR/PAT, but doing so could result in
6953 * memory aliases with conflicting memory types and sometimes MCEs.
6954 * We have to be careful as to what are honored and when.
6956 * For MMIO, guest CD/MTRR are ignored. The EPT memory type is set to
6957 * UC. The effective memory type is UC or WC depending on guest PAT.
6958 * This was historically the source of MCEs and we want to be
6961 * When there is no need to deal with noncoherent DMA (e.g., no VT-d
6962 * or VT-d has snoop control), guest CD/MTRR/PAT are all ignored. The
6963 * EPT memory type is set to WB. The effective memory type is forced
6966 * Otherwise, we trust guest. Guest CD/MTRR/PAT are all honored. The
6967 * EPT memory type is used to emulate guest CD/MTRR.
6971 cache = MTRR_TYPE_UNCACHABLE;
6975 if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
6976 ipat = VMX_EPT_IPAT_BIT;
6977 cache = MTRR_TYPE_WRBACK;
6981 if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
6982 ipat = VMX_EPT_IPAT_BIT;
6983 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
6984 cache = MTRR_TYPE_WRBACK;
6986 cache = MTRR_TYPE_UNCACHABLE;
6990 cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
6993 return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
6996 static void vmcs_set_secondary_exec_control(struct vcpu_vmx *vmx)
6999 * These bits in the secondary execution controls field
7000 * are dynamic, the others are mostly based on the hypervisor
7001 * architecture and the guest's CPUID. Do not touch the
7005 SECONDARY_EXEC_SHADOW_VMCS |
7006 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
7007 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
7008 SECONDARY_EXEC_DESC;
7010 u32 new_ctl = vmx->secondary_exec_control;
7011 u32 cur_ctl = secondary_exec_controls_get(vmx);
7013 secondary_exec_controls_set(vmx, (new_ctl & ~mask) | (cur_ctl & mask));
7017 * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
7018 * (indicating "allowed-1") if they are supported in the guest's CPUID.
7020 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
7022 struct vcpu_vmx *vmx = to_vmx(vcpu);
7023 struct kvm_cpuid_entry2 *entry;
7025 vmx->nested.msrs.cr0_fixed1 = 0xffffffff;
7026 vmx->nested.msrs.cr4_fixed1 = X86_CR4_PCE;
7028 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do { \
7029 if (entry && (entry->_reg & (_cpuid_mask))) \
7030 vmx->nested.msrs.cr4_fixed1 |= (_cr4_mask); \
7033 entry = kvm_find_cpuid_entry(vcpu, 0x1, 0);
7034 cr4_fixed1_update(X86_CR4_VME, edx, feature_bit(VME));
7035 cr4_fixed1_update(X86_CR4_PVI, edx, feature_bit(VME));
7036 cr4_fixed1_update(X86_CR4_TSD, edx, feature_bit(TSC));
7037 cr4_fixed1_update(X86_CR4_DE, edx, feature_bit(DE));
7038 cr4_fixed1_update(X86_CR4_PSE, edx, feature_bit(PSE));
7039 cr4_fixed1_update(X86_CR4_PAE, edx, feature_bit(PAE));
7040 cr4_fixed1_update(X86_CR4_MCE, edx, feature_bit(MCE));
7041 cr4_fixed1_update(X86_CR4_PGE, edx, feature_bit(PGE));
7042 cr4_fixed1_update(X86_CR4_OSFXSR, edx, feature_bit(FXSR));
7043 cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, feature_bit(XMM));
7044 cr4_fixed1_update(X86_CR4_VMXE, ecx, feature_bit(VMX));
7045 cr4_fixed1_update(X86_CR4_SMXE, ecx, feature_bit(SMX));
7046 cr4_fixed1_update(X86_CR4_PCIDE, ecx, feature_bit(PCID));
7047 cr4_fixed1_update(X86_CR4_OSXSAVE, ecx, feature_bit(XSAVE));
7049 entry = kvm_find_cpuid_entry(vcpu, 0x7, 0);
7050 cr4_fixed1_update(X86_CR4_FSGSBASE, ebx, feature_bit(FSGSBASE));
7051 cr4_fixed1_update(X86_CR4_SMEP, ebx, feature_bit(SMEP));
7052 cr4_fixed1_update(X86_CR4_SMAP, ebx, feature_bit(SMAP));
7053 cr4_fixed1_update(X86_CR4_PKE, ecx, feature_bit(PKU));
7054 cr4_fixed1_update(X86_CR4_UMIP, ecx, feature_bit(UMIP));
7055 cr4_fixed1_update(X86_CR4_LA57, ecx, feature_bit(LA57));
7057 #undef cr4_fixed1_update
7060 static void nested_vmx_entry_exit_ctls_update(struct kvm_vcpu *vcpu)
7062 struct vcpu_vmx *vmx = to_vmx(vcpu);
7064 if (kvm_mpx_supported()) {
7065 bool mpx_enabled = guest_cpuid_has(vcpu, X86_FEATURE_MPX);
7068 vmx->nested.msrs.entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
7069 vmx->nested.msrs.exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
7071 vmx->nested.msrs.entry_ctls_high &= ~VM_ENTRY_LOAD_BNDCFGS;
7072 vmx->nested.msrs.exit_ctls_high &= ~VM_EXIT_CLEAR_BNDCFGS;
7077 static void update_intel_pt_cfg(struct kvm_vcpu *vcpu)
7079 struct vcpu_vmx *vmx = to_vmx(vcpu);
7080 struct kvm_cpuid_entry2 *best = NULL;
7083 for (i = 0; i < PT_CPUID_LEAVES; i++) {
7084 best = kvm_find_cpuid_entry(vcpu, 0x14, i);
7087 vmx->pt_desc.caps[CPUID_EAX + i*PT_CPUID_REGS_NUM] = best->eax;
7088 vmx->pt_desc.caps[CPUID_EBX + i*PT_CPUID_REGS_NUM] = best->ebx;
7089 vmx->pt_desc.caps[CPUID_ECX + i*PT_CPUID_REGS_NUM] = best->ecx;
7090 vmx->pt_desc.caps[CPUID_EDX + i*PT_CPUID_REGS_NUM] = best->edx;
7093 /* Get the number of configurable Address Ranges for filtering */
7094 vmx->pt_desc.addr_range = intel_pt_validate_cap(vmx->pt_desc.caps,
7095 PT_CAP_num_address_ranges);
7097 /* Initialize and clear the no dependency bits */
7098 vmx->pt_desc.ctl_bitmask = ~(RTIT_CTL_TRACEEN | RTIT_CTL_OS |
7099 RTIT_CTL_USR | RTIT_CTL_TSC_EN | RTIT_CTL_DISRETC);
7102 * If CPUID.(EAX=14H,ECX=0):EBX[0]=1 CR3Filter can be set otherwise
7103 * will inject an #GP
7105 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_cr3_filtering))
7106 vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_CR3EN;
7109 * If CPUID.(EAX=14H,ECX=0):EBX[1]=1 CYCEn, CycThresh and
7110 * PSBFreq can be set
7112 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc))
7113 vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_CYCLEACC |
7114 RTIT_CTL_CYC_THRESH | RTIT_CTL_PSB_FREQ);
7117 * If CPUID.(EAX=14H,ECX=0):EBX[3]=1 MTCEn BranchEn and
7118 * MTCFreq can be set
7120 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc))
7121 vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_MTC_EN |
7122 RTIT_CTL_BRANCH_EN | RTIT_CTL_MTC_RANGE);
7124 /* If CPUID.(EAX=14H,ECX=0):EBX[4]=1 FUPonPTW and PTWEn can be set */
7125 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_ptwrite))
7126 vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_FUP_ON_PTW |
7129 /* If CPUID.(EAX=14H,ECX=0):EBX[5]=1 PwrEvEn can be set */
7130 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_power_event_trace))
7131 vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_PWR_EVT_EN;
7133 /* If CPUID.(EAX=14H,ECX=0):ECX[0]=1 ToPA can be set */
7134 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_topa_output))
7135 vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_TOPA;
7137 /* If CPUID.(EAX=14H,ECX=0):ECX[3]=1 FabircEn can be set */
7138 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_output_subsys))
7139 vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_FABRIC_EN;
7141 /* unmask address range configure area */
7142 for (i = 0; i < vmx->pt_desc.addr_range; i++)
7143 vmx->pt_desc.ctl_bitmask &= ~(0xfULL << (32 + i * 4));
7146 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
7148 struct vcpu_vmx *vmx = to_vmx(vcpu);
7150 /* xsaves_enabled is recomputed in vmx_compute_secondary_exec_control(). */
7151 vcpu->arch.xsaves_enabled = false;
7153 if (cpu_has_secondary_exec_ctrls()) {
7154 vmx_compute_secondary_exec_control(vmx);
7155 vmcs_set_secondary_exec_control(vmx);
7158 if (nested_vmx_allowed(vcpu))
7159 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7160 FEAT_CTL_VMX_ENABLED_INSIDE_SMX |
7161 FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX;
7163 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7164 ~(FEAT_CTL_VMX_ENABLED_INSIDE_SMX |
7165 FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX);
7167 if (nested_vmx_allowed(vcpu)) {
7168 nested_vmx_cr_fixed1_bits_update(vcpu);
7169 nested_vmx_entry_exit_ctls_update(vcpu);
7172 if (boot_cpu_has(X86_FEATURE_INTEL_PT) &&
7173 guest_cpuid_has(vcpu, X86_FEATURE_INTEL_PT))
7174 update_intel_pt_cfg(vcpu);
7176 if (boot_cpu_has(X86_FEATURE_RTM)) {
7177 struct shared_msr_entry *msr;
7178 msr = find_msr_entry(vmx, MSR_IA32_TSX_CTRL);
7180 bool enabled = guest_cpuid_has(vcpu, X86_FEATURE_RTM);
7181 vmx_set_guest_msr(vmx, msr, enabled ? 0 : TSX_CTRL_RTM_DISABLE);
7186 static __init void vmx_set_cpu_caps(void)
7192 kvm_cpu_cap_set(X86_FEATURE_VMX);
7195 if (kvm_mpx_supported())
7196 kvm_cpu_cap_check_and_set(X86_FEATURE_MPX);
7197 if (cpu_has_vmx_invpcid())
7198 kvm_cpu_cap_check_and_set(X86_FEATURE_INVPCID);
7199 if (vmx_pt_mode_is_host_guest())
7200 kvm_cpu_cap_check_and_set(X86_FEATURE_INTEL_PT);
7202 /* PKU is not yet implemented for shadow paging. */
7203 if (enable_ept && boot_cpu_has(X86_FEATURE_OSPKE))
7204 kvm_cpu_cap_check_and_set(X86_FEATURE_PKU);
7206 if (vmx_umip_emulated())
7207 kvm_cpu_cap_set(X86_FEATURE_UMIP);
7211 if (!vmx_xsaves_supported())
7212 kvm_cpu_cap_clear(X86_FEATURE_XSAVES);
7214 /* CPUID 0x80000001 */
7215 if (!cpu_has_vmx_rdtscp())
7216 kvm_cpu_cap_clear(X86_FEATURE_RDTSCP);
7219 static void vmx_request_immediate_exit(struct kvm_vcpu *vcpu)
7221 to_vmx(vcpu)->req_immediate_exit = true;
7224 static int vmx_check_intercept_io(struct kvm_vcpu *vcpu,
7225 struct x86_instruction_info *info)
7227 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7228 unsigned short port;
7232 if (info->intercept == x86_intercept_in ||
7233 info->intercept == x86_intercept_ins) {
7234 port = info->src_val;
7235 size = info->dst_bytes;
7237 port = info->dst_val;
7238 size = info->src_bytes;
7242 * If the 'use IO bitmaps' VM-execution control is 0, IO instruction
7243 * VM-exits depend on the 'unconditional IO exiting' VM-execution
7246 * Otherwise, IO instruction VM-exits are controlled by the IO bitmaps.
7248 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
7249 intercept = nested_cpu_has(vmcs12,
7250 CPU_BASED_UNCOND_IO_EXITING);
7252 intercept = nested_vmx_check_io_bitmaps(vcpu, port, size);
7254 /* FIXME: produce nested vmexit and return X86EMUL_INTERCEPTED. */
7255 return intercept ? X86EMUL_UNHANDLEABLE : X86EMUL_CONTINUE;
7258 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
7259 struct x86_instruction_info *info,
7260 enum x86_intercept_stage stage,
7261 struct x86_exception *exception)
7263 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7265 switch (info->intercept) {
7267 * RDPID causes #UD if disabled through secondary execution controls.
7268 * Because it is marked as EmulateOnUD, we need to intercept it here.
7270 case x86_intercept_rdtscp:
7271 if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDTSCP)) {
7272 exception->vector = UD_VECTOR;
7273 exception->error_code_valid = false;
7274 return X86EMUL_PROPAGATE_FAULT;
7278 case x86_intercept_in:
7279 case x86_intercept_ins:
7280 case x86_intercept_out:
7281 case x86_intercept_outs:
7282 return vmx_check_intercept_io(vcpu, info);
7284 case x86_intercept_lgdt:
7285 case x86_intercept_lidt:
7286 case x86_intercept_lldt:
7287 case x86_intercept_ltr:
7288 case x86_intercept_sgdt:
7289 case x86_intercept_sidt:
7290 case x86_intercept_sldt:
7291 case x86_intercept_str:
7292 if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC))
7293 return X86EMUL_CONTINUE;
7295 /* FIXME: produce nested vmexit and return X86EMUL_INTERCEPTED. */
7298 /* TODO: check more intercepts... */
7303 return X86EMUL_UNHANDLEABLE;
7306 #ifdef CONFIG_X86_64
7307 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
7308 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
7309 u64 divisor, u64 *result)
7311 u64 low = a << shift, high = a >> (64 - shift);
7313 /* To avoid the overflow on divq */
7314 if (high >= divisor)
7317 /* Low hold the result, high hold rem which is discarded */
7318 asm("divq %2\n\t" : "=a" (low), "=d" (high) :
7319 "rm" (divisor), "0" (low), "1" (high));
7325 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc,
7328 struct vcpu_vmx *vmx;
7329 u64 tscl, guest_tscl, delta_tsc, lapic_timer_advance_cycles;
7330 struct kvm_timer *ktimer = &vcpu->arch.apic->lapic_timer;
7332 if (kvm_mwait_in_guest(vcpu->kvm) ||
7333 kvm_can_post_timer_interrupt(vcpu))
7338 guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
7339 delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
7340 lapic_timer_advance_cycles = nsec_to_cycles(vcpu,
7341 ktimer->timer_advance_ns);
7343 if (delta_tsc > lapic_timer_advance_cycles)
7344 delta_tsc -= lapic_timer_advance_cycles;
7348 /* Convert to host delta tsc if tsc scaling is enabled */
7349 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
7350 delta_tsc && u64_shl_div_u64(delta_tsc,
7351 kvm_tsc_scaling_ratio_frac_bits,
7352 vcpu->arch.tsc_scaling_ratio, &delta_tsc))
7356 * If the delta tsc can't fit in the 32 bit after the multi shift,
7357 * we can't use the preemption timer.
7358 * It's possible that it fits on later vmentries, but checking
7359 * on every vmentry is costly so we just use an hrtimer.
7361 if (delta_tsc >> (cpu_preemption_timer_multi + 32))
7364 vmx->hv_deadline_tsc = tscl + delta_tsc;
7365 *expired = !delta_tsc;
7369 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
7371 to_vmx(vcpu)->hv_deadline_tsc = -1;
7375 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
7377 if (!kvm_pause_in_guest(vcpu->kvm))
7378 shrink_ple_window(vcpu);
7381 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
7382 struct kvm_memory_slot *slot)
7384 if (!kvm_dirty_log_manual_protect_and_init_set(kvm))
7385 kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
7386 kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
7389 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
7390 struct kvm_memory_slot *slot)
7392 kvm_mmu_slot_set_dirty(kvm, slot);
7395 static void vmx_flush_log_dirty(struct kvm *kvm)
7397 kvm_flush_pml_buffers(kvm);
7400 static int vmx_write_pml_buffer(struct kvm_vcpu *vcpu)
7402 struct vmcs12 *vmcs12;
7403 struct vcpu_vmx *vmx = to_vmx(vcpu);
7406 if (is_guest_mode(vcpu)) {
7407 WARN_ON_ONCE(vmx->nested.pml_full);
7410 * Check if PML is enabled for the nested guest.
7411 * Whether eptp bit 6 is set is already checked
7412 * as part of A/D emulation.
7414 vmcs12 = get_vmcs12(vcpu);
7415 if (!nested_cpu_has_pml(vmcs12))
7418 if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
7419 vmx->nested.pml_full = true;
7423 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS) & ~0xFFFull;
7424 dst = vmcs12->pml_address + sizeof(u64) * vmcs12->guest_pml_index;
7426 if (kvm_write_guest_page(vcpu->kvm, gpa_to_gfn(dst), &gpa,
7427 offset_in_page(dst), sizeof(gpa)))
7430 vmcs12->guest_pml_index--;
7436 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
7437 struct kvm_memory_slot *memslot,
7438 gfn_t offset, unsigned long mask)
7440 kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
7443 static void __pi_post_block(struct kvm_vcpu *vcpu)
7445 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7446 struct pi_desc old, new;
7450 old.control = new.control = pi_desc->control;
7451 WARN(old.nv != POSTED_INTR_WAKEUP_VECTOR,
7452 "Wakeup handler not enabled while the VCPU is blocked\n");
7454 dest = cpu_physical_id(vcpu->cpu);
7456 if (x2apic_enabled())
7459 new.ndst = (dest << 8) & 0xFF00;
7461 /* set 'NV' to 'notification vector' */
7462 new.nv = POSTED_INTR_VECTOR;
7463 } while (cmpxchg64(&pi_desc->control, old.control,
7464 new.control) != old.control);
7466 if (!WARN_ON_ONCE(vcpu->pre_pcpu == -1)) {
7467 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7468 list_del(&vcpu->blocked_vcpu_list);
7469 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7470 vcpu->pre_pcpu = -1;
7475 * This routine does the following things for vCPU which is going
7476 * to be blocked if VT-d PI is enabled.
7477 * - Store the vCPU to the wakeup list, so when interrupts happen
7478 * we can find the right vCPU to wake up.
7479 * - Change the Posted-interrupt descriptor as below:
7480 * 'NDST' <-- vcpu->pre_pcpu
7481 * 'NV' <-- POSTED_INTR_WAKEUP_VECTOR
7482 * - If 'ON' is set during this process, which means at least one
7483 * interrupt is posted for this vCPU, we cannot block it, in
7484 * this case, return 1, otherwise, return 0.
7487 static int pi_pre_block(struct kvm_vcpu *vcpu)
7490 struct pi_desc old, new;
7491 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7493 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
7494 !irq_remapping_cap(IRQ_POSTING_CAP) ||
7495 !kvm_vcpu_apicv_active(vcpu))
7498 WARN_ON(irqs_disabled());
7499 local_irq_disable();
7500 if (!WARN_ON_ONCE(vcpu->pre_pcpu != -1)) {
7501 vcpu->pre_pcpu = vcpu->cpu;
7502 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7503 list_add_tail(&vcpu->blocked_vcpu_list,
7504 &per_cpu(blocked_vcpu_on_cpu,
7506 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7510 old.control = new.control = pi_desc->control;
7512 WARN((pi_desc->sn == 1),
7513 "Warning: SN field of posted-interrupts "
7514 "is set before blocking\n");
7517 * Since vCPU can be preempted during this process,
7518 * vcpu->cpu could be different with pre_pcpu, we
7519 * need to set pre_pcpu as the destination of wakeup
7520 * notification event, then we can find the right vCPU
7521 * to wakeup in wakeup handler if interrupts happen
7522 * when the vCPU is in blocked state.
7524 dest = cpu_physical_id(vcpu->pre_pcpu);
7526 if (x2apic_enabled())
7529 new.ndst = (dest << 8) & 0xFF00;
7531 /* set 'NV' to 'wakeup vector' */
7532 new.nv = POSTED_INTR_WAKEUP_VECTOR;
7533 } while (cmpxchg64(&pi_desc->control, old.control,
7534 new.control) != old.control);
7536 /* We should not block the vCPU if an interrupt is posted for it. */
7537 if (pi_test_on(pi_desc) == 1)
7538 __pi_post_block(vcpu);
7541 return (vcpu->pre_pcpu == -1);
7544 static int vmx_pre_block(struct kvm_vcpu *vcpu)
7546 if (pi_pre_block(vcpu))
7549 if (kvm_lapic_hv_timer_in_use(vcpu))
7550 kvm_lapic_switch_to_sw_timer(vcpu);
7555 static void pi_post_block(struct kvm_vcpu *vcpu)
7557 if (vcpu->pre_pcpu == -1)
7560 WARN_ON(irqs_disabled());
7561 local_irq_disable();
7562 __pi_post_block(vcpu);
7566 static void vmx_post_block(struct kvm_vcpu *vcpu)
7568 if (kvm_x86_ops.set_hv_timer)
7569 kvm_lapic_switch_to_hv_timer(vcpu);
7571 pi_post_block(vcpu);
7575 * vmx_update_pi_irte - set IRTE for Posted-Interrupts
7578 * @host_irq: host irq of the interrupt
7579 * @guest_irq: gsi of the interrupt
7580 * @set: set or unset PI
7581 * returns 0 on success, < 0 on failure
7583 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
7584 uint32_t guest_irq, bool set)
7586 struct kvm_kernel_irq_routing_entry *e;
7587 struct kvm_irq_routing_table *irq_rt;
7588 struct kvm_lapic_irq irq;
7589 struct kvm_vcpu *vcpu;
7590 struct vcpu_data vcpu_info;
7593 if (!kvm_arch_has_assigned_device(kvm) ||
7594 !irq_remapping_cap(IRQ_POSTING_CAP) ||
7595 !kvm_vcpu_apicv_active(kvm->vcpus[0]))
7598 idx = srcu_read_lock(&kvm->irq_srcu);
7599 irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
7600 if (guest_irq >= irq_rt->nr_rt_entries ||
7601 hlist_empty(&irq_rt->map[guest_irq])) {
7602 pr_warn_once("no route for guest_irq %u/%u (broken user space?)\n",
7603 guest_irq, irq_rt->nr_rt_entries);
7607 hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
7608 if (e->type != KVM_IRQ_ROUTING_MSI)
7611 * VT-d PI cannot support posting multicast/broadcast
7612 * interrupts to a vCPU, we still use interrupt remapping
7613 * for these kind of interrupts.
7615 * For lowest-priority interrupts, we only support
7616 * those with single CPU as the destination, e.g. user
7617 * configures the interrupts via /proc/irq or uses
7618 * irqbalance to make the interrupts single-CPU.
7620 * We will support full lowest-priority interrupt later.
7622 * In addition, we can only inject generic interrupts using
7623 * the PI mechanism, refuse to route others through it.
7626 kvm_set_msi_irq(kvm, e, &irq);
7627 if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu) ||
7628 !kvm_irq_is_postable(&irq)) {
7630 * Make sure the IRTE is in remapped mode if
7631 * we don't handle it in posted mode.
7633 ret = irq_set_vcpu_affinity(host_irq, NULL);
7636 "failed to back to remapped mode, irq: %u\n",
7644 vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
7645 vcpu_info.vector = irq.vector;
7647 trace_kvm_pi_irte_update(host_irq, vcpu->vcpu_id, e->gsi,
7648 vcpu_info.vector, vcpu_info.pi_desc_addr, set);
7651 ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
7653 ret = irq_set_vcpu_affinity(host_irq, NULL);
7656 printk(KERN_INFO "%s: failed to update PI IRTE\n",
7664 srcu_read_unlock(&kvm->irq_srcu, idx);
7668 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
7670 if (vcpu->arch.mcg_cap & MCG_LMCE_P)
7671 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7672 FEAT_CTL_LMCE_ENABLED;
7674 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7675 ~FEAT_CTL_LMCE_ENABLED;
7678 static int vmx_smi_allowed(struct kvm_vcpu *vcpu)
7680 /* we need a nested vmexit to enter SMM, postpone if run is pending */
7681 if (to_vmx(vcpu)->nested.nested_run_pending)
7686 static int vmx_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
7688 struct vcpu_vmx *vmx = to_vmx(vcpu);
7690 vmx->nested.smm.guest_mode = is_guest_mode(vcpu);
7691 if (vmx->nested.smm.guest_mode)
7692 nested_vmx_vmexit(vcpu, -1, 0, 0);
7694 vmx->nested.smm.vmxon = vmx->nested.vmxon;
7695 vmx->nested.vmxon = false;
7696 vmx_clear_hlt(vcpu);
7700 static int vmx_pre_leave_smm(struct kvm_vcpu *vcpu, const char *smstate)
7702 struct vcpu_vmx *vmx = to_vmx(vcpu);
7705 if (vmx->nested.smm.vmxon) {
7706 vmx->nested.vmxon = true;
7707 vmx->nested.smm.vmxon = false;
7710 if (vmx->nested.smm.guest_mode) {
7711 ret = nested_vmx_enter_non_root_mode(vcpu, false);
7715 vmx->nested.smm.guest_mode = false;
7720 static int enable_smi_window(struct kvm_vcpu *vcpu)
7725 static bool vmx_need_emulation_on_page_fault(struct kvm_vcpu *vcpu)
7730 static bool vmx_apic_init_signal_blocked(struct kvm_vcpu *vcpu)
7732 return to_vmx(vcpu)->nested.vmxon;
7735 static void hardware_unsetup(void)
7738 nested_vmx_hardware_unsetup();
7743 static bool vmx_check_apicv_inhibit_reasons(ulong bit)
7745 ulong supported = BIT(APICV_INHIBIT_REASON_DISABLE) |
7746 BIT(APICV_INHIBIT_REASON_HYPERV);
7748 return supported & BIT(bit);
7751 static struct kvm_x86_ops vmx_x86_ops __initdata = {
7752 .hardware_unsetup = hardware_unsetup,
7754 .hardware_enable = hardware_enable,
7755 .hardware_disable = hardware_disable,
7756 .cpu_has_accelerated_tpr = report_flexpriority,
7757 .has_emulated_msr = vmx_has_emulated_msr,
7759 .vm_size = sizeof(struct kvm_vmx),
7760 .vm_init = vmx_vm_init,
7762 .vcpu_create = vmx_create_vcpu,
7763 .vcpu_free = vmx_free_vcpu,
7764 .vcpu_reset = vmx_vcpu_reset,
7766 .prepare_guest_switch = vmx_prepare_switch_to_guest,
7767 .vcpu_load = vmx_vcpu_load,
7768 .vcpu_put = vmx_vcpu_put,
7770 .update_bp_intercept = update_exception_bitmap,
7771 .get_msr_feature = vmx_get_msr_feature,
7772 .get_msr = vmx_get_msr,
7773 .set_msr = vmx_set_msr,
7774 .get_segment_base = vmx_get_segment_base,
7775 .get_segment = vmx_get_segment,
7776 .set_segment = vmx_set_segment,
7777 .get_cpl = vmx_get_cpl,
7778 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
7779 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
7780 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
7781 .set_cr0 = vmx_set_cr0,
7782 .set_cr4 = vmx_set_cr4,
7783 .set_efer = vmx_set_efer,
7784 .get_idt = vmx_get_idt,
7785 .set_idt = vmx_set_idt,
7786 .get_gdt = vmx_get_gdt,
7787 .set_gdt = vmx_set_gdt,
7788 .get_dr6 = vmx_get_dr6,
7789 .set_dr6 = vmx_set_dr6,
7790 .set_dr7 = vmx_set_dr7,
7791 .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
7792 .cache_reg = vmx_cache_reg,
7793 .get_rflags = vmx_get_rflags,
7794 .set_rflags = vmx_set_rflags,
7796 .tlb_flush_all = vmx_flush_tlb_all,
7797 .tlb_flush_current = vmx_flush_tlb_current,
7798 .tlb_flush_gva = vmx_flush_tlb_gva,
7799 .tlb_flush_guest = vmx_flush_tlb_guest,
7801 .run = vmx_vcpu_run,
7802 .handle_exit = vmx_handle_exit,
7803 .skip_emulated_instruction = vmx_skip_emulated_instruction,
7804 .update_emulated_instruction = vmx_update_emulated_instruction,
7805 .set_interrupt_shadow = vmx_set_interrupt_shadow,
7806 .get_interrupt_shadow = vmx_get_interrupt_shadow,
7807 .patch_hypercall = vmx_patch_hypercall,
7808 .set_irq = vmx_inject_irq,
7809 .set_nmi = vmx_inject_nmi,
7810 .queue_exception = vmx_queue_exception,
7811 .cancel_injection = vmx_cancel_injection,
7812 .interrupt_allowed = vmx_interrupt_allowed,
7813 .nmi_allowed = vmx_nmi_allowed,
7814 .get_nmi_mask = vmx_get_nmi_mask,
7815 .set_nmi_mask = vmx_set_nmi_mask,
7816 .enable_nmi_window = enable_nmi_window,
7817 .enable_irq_window = enable_irq_window,
7818 .update_cr8_intercept = update_cr8_intercept,
7819 .set_virtual_apic_mode = vmx_set_virtual_apic_mode,
7820 .set_apic_access_page_addr = vmx_set_apic_access_page_addr,
7821 .refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
7822 .load_eoi_exitmap = vmx_load_eoi_exitmap,
7823 .apicv_post_state_restore = vmx_apicv_post_state_restore,
7824 .check_apicv_inhibit_reasons = vmx_check_apicv_inhibit_reasons,
7825 .hwapic_irr_update = vmx_hwapic_irr_update,
7826 .hwapic_isr_update = vmx_hwapic_isr_update,
7827 .guest_apic_has_interrupt = vmx_guest_apic_has_interrupt,
7828 .sync_pir_to_irr = vmx_sync_pir_to_irr,
7829 .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
7830 .dy_apicv_has_pending_interrupt = vmx_dy_apicv_has_pending_interrupt,
7832 .set_tss_addr = vmx_set_tss_addr,
7833 .set_identity_map_addr = vmx_set_identity_map_addr,
7834 .get_tdp_level = get_ept_level,
7835 .get_mt_mask = vmx_get_mt_mask,
7837 .get_exit_info = vmx_get_exit_info,
7839 .cpuid_update = vmx_cpuid_update,
7841 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
7843 .read_l1_tsc_offset = vmx_read_l1_tsc_offset,
7844 .write_l1_tsc_offset = vmx_write_l1_tsc_offset,
7846 .load_mmu_pgd = vmx_load_mmu_pgd,
7848 .check_intercept = vmx_check_intercept,
7849 .handle_exit_irqoff = vmx_handle_exit_irqoff,
7851 .request_immediate_exit = vmx_request_immediate_exit,
7853 .sched_in = vmx_sched_in,
7855 .slot_enable_log_dirty = vmx_slot_enable_log_dirty,
7856 .slot_disable_log_dirty = vmx_slot_disable_log_dirty,
7857 .flush_log_dirty = vmx_flush_log_dirty,
7858 .enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
7859 .write_log_dirty = vmx_write_pml_buffer,
7861 .pre_block = vmx_pre_block,
7862 .post_block = vmx_post_block,
7864 .pmu_ops = &intel_pmu_ops,
7866 .update_pi_irte = vmx_update_pi_irte,
7868 #ifdef CONFIG_X86_64
7869 .set_hv_timer = vmx_set_hv_timer,
7870 .cancel_hv_timer = vmx_cancel_hv_timer,
7873 .setup_mce = vmx_setup_mce,
7875 .smi_allowed = vmx_smi_allowed,
7876 .pre_enter_smm = vmx_pre_enter_smm,
7877 .pre_leave_smm = vmx_pre_leave_smm,
7878 .enable_smi_window = enable_smi_window,
7880 .check_nested_events = NULL,
7881 .get_nested_state = NULL,
7882 .set_nested_state = NULL,
7883 .get_vmcs12_pages = NULL,
7884 .nested_enable_evmcs = NULL,
7885 .nested_get_evmcs_version = NULL,
7886 .need_emulation_on_page_fault = vmx_need_emulation_on_page_fault,
7887 .apic_init_signal_blocked = vmx_apic_init_signal_blocked,
7890 static __init int hardware_setup(void)
7892 unsigned long host_bndcfgs;
7894 int r, i, ept_lpage_level;
7897 host_idt_base = dt.address;
7899 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
7900 kvm_define_shared_msr(i, vmx_msr_index[i]);
7902 if (setup_vmcs_config(&vmcs_config, &vmx_capability) < 0)
7905 if (boot_cpu_has(X86_FEATURE_NX))
7906 kvm_enable_efer_bits(EFER_NX);
7908 if (boot_cpu_has(X86_FEATURE_MPX)) {
7909 rdmsrl(MSR_IA32_BNDCFGS, host_bndcfgs);
7910 WARN_ONCE(host_bndcfgs, "KVM: BNDCFGS in host will be lost");
7913 if (!cpu_has_vmx_mpx())
7914 supported_xcr0 &= ~(XFEATURE_MASK_BNDREGS |
7915 XFEATURE_MASK_BNDCSR);
7917 if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
7918 !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
7921 if (!cpu_has_vmx_ept() ||
7922 !cpu_has_vmx_ept_4levels() ||
7923 !cpu_has_vmx_ept_mt_wb() ||
7924 !cpu_has_vmx_invept_global())
7927 if (!cpu_has_vmx_ept_ad_bits() || !enable_ept)
7928 enable_ept_ad_bits = 0;
7930 if (!cpu_has_vmx_unrestricted_guest() || !enable_ept)
7931 enable_unrestricted_guest = 0;
7933 if (!cpu_has_vmx_flexpriority())
7934 flexpriority_enabled = 0;
7936 if (!cpu_has_virtual_nmis())
7940 * set_apic_access_page_addr() is used to reload apic access
7941 * page upon invalidation. No need to do anything if not
7942 * using the APIC_ACCESS_ADDR VMCS field.
7944 if (!flexpriority_enabled)
7945 vmx_x86_ops.set_apic_access_page_addr = NULL;
7947 if (!cpu_has_vmx_tpr_shadow())
7948 vmx_x86_ops.update_cr8_intercept = NULL;
7950 #if IS_ENABLED(CONFIG_HYPERV)
7951 if (ms_hyperv.nested_features & HV_X64_NESTED_GUEST_MAPPING_FLUSH
7953 vmx_x86_ops.tlb_remote_flush = hv_remote_flush_tlb;
7954 vmx_x86_ops.tlb_remote_flush_with_range =
7955 hv_remote_flush_tlb_with_range;
7959 if (!cpu_has_vmx_ple()) {
7962 ple_window_grow = 0;
7964 ple_window_shrink = 0;
7967 if (!cpu_has_vmx_apicv()) {
7969 vmx_x86_ops.sync_pir_to_irr = NULL;
7972 if (cpu_has_vmx_tsc_scaling()) {
7973 kvm_has_tsc_control = true;
7974 kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
7975 kvm_tsc_scaling_ratio_frac_bits = 48;
7978 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
7984 ept_lpage_level = 0;
7985 else if (cpu_has_vmx_ept_1g_page())
7986 ept_lpage_level = PT_PDPE_LEVEL;
7987 else if (cpu_has_vmx_ept_2m_page())
7988 ept_lpage_level = PT_DIRECTORY_LEVEL;
7990 ept_lpage_level = PT_PAGE_TABLE_LEVEL;
7991 kvm_configure_mmu(enable_ept, ept_lpage_level);
7994 * Only enable PML when hardware supports PML feature, and both EPT
7995 * and EPT A/D bit features are enabled -- PML depends on them to work.
7997 if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
8001 vmx_x86_ops.slot_enable_log_dirty = NULL;
8002 vmx_x86_ops.slot_disable_log_dirty = NULL;
8003 vmx_x86_ops.flush_log_dirty = NULL;
8004 vmx_x86_ops.enable_log_dirty_pt_masked = NULL;
8007 if (!cpu_has_vmx_preemption_timer())
8008 enable_preemption_timer = false;
8010 if (enable_preemption_timer) {
8011 u64 use_timer_freq = 5000ULL * 1000 * 1000;
8014 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
8015 cpu_preemption_timer_multi =
8016 vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
8019 use_timer_freq = (u64)tsc_khz * 1000;
8020 use_timer_freq >>= cpu_preemption_timer_multi;
8023 * KVM "disables" the preemption timer by setting it to its max
8024 * value. Don't use the timer if it might cause spurious exits
8025 * at a rate faster than 0.1 Hz (of uninterrupted guest time).
8027 if (use_timer_freq > 0xffffffffu / 10)
8028 enable_preemption_timer = false;
8031 if (!enable_preemption_timer) {
8032 vmx_x86_ops.set_hv_timer = NULL;
8033 vmx_x86_ops.cancel_hv_timer = NULL;
8034 vmx_x86_ops.request_immediate_exit = __kvm_request_immediate_exit;
8037 kvm_set_posted_intr_wakeup_handler(wakeup_handler);
8039 kvm_mce_cap_supported |= MCG_LMCE_P;
8041 if (pt_mode != PT_MODE_SYSTEM && pt_mode != PT_MODE_HOST_GUEST)
8043 if (!enable_ept || !cpu_has_vmx_intel_pt())
8044 pt_mode = PT_MODE_SYSTEM;
8047 nested_vmx_setup_ctls_msrs(&vmcs_config.nested,
8048 vmx_capability.ept);
8050 r = nested_vmx_hardware_setup(&vmx_x86_ops,
8051 kvm_vmx_exit_handlers);
8058 r = alloc_kvm_area();
8060 nested_vmx_hardware_unsetup();
8064 static struct kvm_x86_init_ops vmx_init_ops __initdata = {
8065 .cpu_has_kvm_support = cpu_has_kvm_support,
8066 .disabled_by_bios = vmx_disabled_by_bios,
8067 .check_processor_compatibility = vmx_check_processor_compat,
8068 .hardware_setup = hardware_setup,
8070 .runtime_ops = &vmx_x86_ops,
8073 static void vmx_cleanup_l1d_flush(void)
8075 if (vmx_l1d_flush_pages) {
8076 free_pages((unsigned long)vmx_l1d_flush_pages, L1D_CACHE_ORDER);
8077 vmx_l1d_flush_pages = NULL;
8079 /* Restore state so sysfs ignores VMX */
8080 l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
8083 static void vmx_exit(void)
8085 #ifdef CONFIG_KEXEC_CORE
8086 RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
8092 #if IS_ENABLED(CONFIG_HYPERV)
8093 if (static_branch_unlikely(&enable_evmcs)) {
8095 struct hv_vp_assist_page *vp_ap;
8097 * Reset everything to support using non-enlightened VMCS
8098 * access later (e.g. when we reload the module with
8099 * enlightened_vmcs=0)
8101 for_each_online_cpu(cpu) {
8102 vp_ap = hv_get_vp_assist_page(cpu);
8107 vp_ap->nested_control.features.directhypercall = 0;
8108 vp_ap->current_nested_vmcs = 0;
8109 vp_ap->enlighten_vmentry = 0;
8112 static_branch_disable(&enable_evmcs);
8115 vmx_cleanup_l1d_flush();
8117 module_exit(vmx_exit);
8119 static int __init vmx_init(void)
8123 #if IS_ENABLED(CONFIG_HYPERV)
8125 * Enlightened VMCS usage should be recommended and the host needs
8126 * to support eVMCS v1 or above. We can also disable eVMCS support
8127 * with module parameter.
8129 if (enlightened_vmcs &&
8130 ms_hyperv.hints & HV_X64_ENLIGHTENED_VMCS_RECOMMENDED &&
8131 (ms_hyperv.nested_features & HV_X64_ENLIGHTENED_VMCS_VERSION) >=
8132 KVM_EVMCS_VERSION) {
8135 /* Check that we have assist pages on all online CPUs */
8136 for_each_online_cpu(cpu) {
8137 if (!hv_get_vp_assist_page(cpu)) {
8138 enlightened_vmcs = false;
8143 if (enlightened_vmcs) {
8144 pr_info("KVM: vmx: using Hyper-V Enlightened VMCS\n");
8145 static_branch_enable(&enable_evmcs);
8148 if (ms_hyperv.nested_features & HV_X64_NESTED_DIRECT_FLUSH)
8149 vmx_x86_ops.enable_direct_tlbflush
8150 = hv_enable_direct_tlbflush;
8153 enlightened_vmcs = false;
8157 r = kvm_init(&vmx_init_ops, sizeof(struct vcpu_vmx),
8158 __alignof__(struct vcpu_vmx), THIS_MODULE);
8163 * Must be called after kvm_init() so enable_ept is properly set
8164 * up. Hand the parameter mitigation value in which was stored in
8165 * the pre module init parser. If no parameter was given, it will
8166 * contain 'auto' which will be turned into the default 'cond'
8169 r = vmx_setup_l1d_flush(vmentry_l1d_flush_param);
8175 for_each_possible_cpu(cpu) {
8176 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
8177 INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
8178 spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
8181 #ifdef CONFIG_KEXEC_CORE
8182 rcu_assign_pointer(crash_vmclear_loaded_vmcss,
8183 crash_vmclear_local_loaded_vmcss);
8185 vmx_check_vmcs12_offsets();
8189 module_init(vmx_init);