1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
5 * This module enables machines with Intel VT-x extensions to run virtual
6 * machines without emulation or binary translation.
8 * Copyright (C) 2006 Qumranet, Inc.
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
16 #include <linux/frame.h>
17 #include <linux/highmem.h>
18 #include <linux/hrtimer.h>
19 #include <linux/kernel.h>
20 #include <linux/kvm_host.h>
21 #include <linux/module.h>
22 #include <linux/moduleparam.h>
23 #include <linux/mod_devicetable.h>
25 #include <linux/sched.h>
26 #include <linux/sched/smt.h>
27 #include <linux/slab.h>
28 #include <linux/tboot.h>
29 #include <linux/trace_events.h>
34 #include <asm/debugreg.h>
36 #include <asm/fpu/internal.h>
38 #include <asm/irq_remapping.h>
39 #include <asm/kexec.h>
40 #include <asm/perf_event.h>
42 #include <asm/mmu_context.h>
43 #include <asm/mshyperv.h>
44 #include <asm/spec-ctrl.h>
45 #include <asm/virtext.h>
48 #include "capabilities.h"
52 #include "kvm_cache_regs.h"
64 MODULE_AUTHOR("Qumranet");
65 MODULE_LICENSE("GPL");
68 static const struct x86_cpu_id vmx_cpu_id[] = {
69 X86_FEATURE_MATCH(X86_FEATURE_VMX),
72 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
75 bool __read_mostly enable_vpid = 1;
76 module_param_named(vpid, enable_vpid, bool, 0444);
78 static bool __read_mostly enable_vnmi = 1;
79 module_param_named(vnmi, enable_vnmi, bool, S_IRUGO);
81 bool __read_mostly flexpriority_enabled = 1;
82 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
84 bool __read_mostly enable_ept = 1;
85 module_param_named(ept, enable_ept, bool, S_IRUGO);
87 bool __read_mostly enable_unrestricted_guest = 1;
88 module_param_named(unrestricted_guest,
89 enable_unrestricted_guest, bool, S_IRUGO);
91 bool __read_mostly enable_ept_ad_bits = 1;
92 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
94 static bool __read_mostly emulate_invalid_guest_state = true;
95 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
97 static bool __read_mostly fasteoi = 1;
98 module_param(fasteoi, bool, S_IRUGO);
100 bool __read_mostly enable_apicv = 1;
101 module_param(enable_apicv, bool, S_IRUGO);
104 * If nested=1, nested virtualization is supported, i.e., guests may use
105 * VMX and be a hypervisor for its own guests. If nested=0, guests may not
106 * use VMX instructions.
108 static bool __read_mostly nested = 1;
109 module_param(nested, bool, S_IRUGO);
111 bool __read_mostly enable_pml = 1;
112 module_param_named(pml, enable_pml, bool, S_IRUGO);
114 static bool __read_mostly dump_invalid_vmcs = 0;
115 module_param(dump_invalid_vmcs, bool, 0644);
117 #define MSR_BITMAP_MODE_X2APIC 1
118 #define MSR_BITMAP_MODE_X2APIC_APICV 2
120 #define KVM_VMX_TSC_MULTIPLIER_MAX 0xffffffffffffffffULL
122 /* Guest_tsc -> host_tsc conversion requires 64-bit division. */
123 static int __read_mostly cpu_preemption_timer_multi;
124 static bool __read_mostly enable_preemption_timer = 1;
126 module_param_named(preemption_timer, enable_preemption_timer, bool, S_IRUGO);
129 #define KVM_VM_CR0_ALWAYS_OFF (X86_CR0_NW | X86_CR0_CD)
130 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR0_NE
131 #define KVM_VM_CR0_ALWAYS_ON \
132 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | \
133 X86_CR0_WP | X86_CR0_PG | X86_CR0_PE)
134 #define KVM_CR4_GUEST_OWNED_BITS \
135 (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \
136 | X86_CR4_OSXMMEXCPT | X86_CR4_LA57 | X86_CR4_TSD)
138 #define KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR4_VMXE
139 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
140 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
142 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
144 #define MSR_IA32_RTIT_STATUS_MASK (~(RTIT_STATUS_FILTEREN | \
145 RTIT_STATUS_CONTEXTEN | RTIT_STATUS_TRIGGEREN | \
146 RTIT_STATUS_ERROR | RTIT_STATUS_STOPPED | \
147 RTIT_STATUS_BYTECNT))
149 #define MSR_IA32_RTIT_OUTPUT_BASE_MASK \
150 (~((1UL << cpuid_query_maxphyaddr(vcpu)) - 1) | 0x7f)
153 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
154 * ple_gap: upper bound on the amount of time between two successive
155 * executions of PAUSE in a loop. Also indicate if ple enabled.
156 * According to test, this time is usually smaller than 128 cycles.
157 * ple_window: upper bound on the amount of time a guest is allowed to execute
158 * in a PAUSE loop. Tests indicate that most spinlocks are held for
159 * less than 2^12 cycles
160 * Time is measured based on a counter that runs at the same rate as the TSC,
161 * refer SDM volume 3b section 21.6.13 & 22.1.3.
163 static unsigned int ple_gap = KVM_DEFAULT_PLE_GAP;
164 module_param(ple_gap, uint, 0444);
166 static unsigned int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
167 module_param(ple_window, uint, 0444);
169 /* Default doubles per-vcpu window every exit. */
170 static unsigned int ple_window_grow = KVM_DEFAULT_PLE_WINDOW_GROW;
171 module_param(ple_window_grow, uint, 0444);
173 /* Default resets per-vcpu window every exit to ple_window. */
174 static unsigned int ple_window_shrink = KVM_DEFAULT_PLE_WINDOW_SHRINK;
175 module_param(ple_window_shrink, uint, 0444);
177 /* Default is to compute the maximum so we can never overflow. */
178 static unsigned int ple_window_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
179 module_param(ple_window_max, uint, 0444);
181 /* Default is SYSTEM mode, 1 for host-guest mode */
182 int __read_mostly pt_mode = PT_MODE_SYSTEM;
183 module_param(pt_mode, int, S_IRUGO);
185 static DEFINE_STATIC_KEY_FALSE(vmx_l1d_should_flush);
186 static DEFINE_STATIC_KEY_FALSE(vmx_l1d_flush_cond);
187 static DEFINE_MUTEX(vmx_l1d_flush_mutex);
189 /* Storage for pre module init parameter parsing */
190 static enum vmx_l1d_flush_state __read_mostly vmentry_l1d_flush_param = VMENTER_L1D_FLUSH_AUTO;
192 static const struct {
195 } vmentry_l1d_param[] = {
196 [VMENTER_L1D_FLUSH_AUTO] = {"auto", true},
197 [VMENTER_L1D_FLUSH_NEVER] = {"never", true},
198 [VMENTER_L1D_FLUSH_COND] = {"cond", true},
199 [VMENTER_L1D_FLUSH_ALWAYS] = {"always", true},
200 [VMENTER_L1D_FLUSH_EPT_DISABLED] = {"EPT disabled", false},
201 [VMENTER_L1D_FLUSH_NOT_REQUIRED] = {"not required", false},
204 #define L1D_CACHE_ORDER 4
205 static void *vmx_l1d_flush_pages;
207 static int vmx_setup_l1d_flush(enum vmx_l1d_flush_state l1tf)
212 if (!boot_cpu_has_bug(X86_BUG_L1TF)) {
213 l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED;
218 l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_EPT_DISABLED;
222 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES)) {
225 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, msr);
226 if (msr & ARCH_CAP_SKIP_VMENTRY_L1DFLUSH) {
227 l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED;
232 /* If set to auto use the default l1tf mitigation method */
233 if (l1tf == VMENTER_L1D_FLUSH_AUTO) {
234 switch (l1tf_mitigation) {
235 case L1TF_MITIGATION_OFF:
236 l1tf = VMENTER_L1D_FLUSH_NEVER;
238 case L1TF_MITIGATION_FLUSH_NOWARN:
239 case L1TF_MITIGATION_FLUSH:
240 case L1TF_MITIGATION_FLUSH_NOSMT:
241 l1tf = VMENTER_L1D_FLUSH_COND;
243 case L1TF_MITIGATION_FULL:
244 case L1TF_MITIGATION_FULL_FORCE:
245 l1tf = VMENTER_L1D_FLUSH_ALWAYS;
248 } else if (l1tf_mitigation == L1TF_MITIGATION_FULL_FORCE) {
249 l1tf = VMENTER_L1D_FLUSH_ALWAYS;
252 if (l1tf != VMENTER_L1D_FLUSH_NEVER && !vmx_l1d_flush_pages &&
253 !boot_cpu_has(X86_FEATURE_FLUSH_L1D)) {
255 * This allocation for vmx_l1d_flush_pages is not tied to a VM
256 * lifetime and so should not be charged to a memcg.
258 page = alloc_pages(GFP_KERNEL, L1D_CACHE_ORDER);
261 vmx_l1d_flush_pages = page_address(page);
264 * Initialize each page with a different pattern in
265 * order to protect against KSM in the nested
266 * virtualization case.
268 for (i = 0; i < 1u << L1D_CACHE_ORDER; ++i) {
269 memset(vmx_l1d_flush_pages + i * PAGE_SIZE, i + 1,
274 l1tf_vmx_mitigation = l1tf;
276 if (l1tf != VMENTER_L1D_FLUSH_NEVER)
277 static_branch_enable(&vmx_l1d_should_flush);
279 static_branch_disable(&vmx_l1d_should_flush);
281 if (l1tf == VMENTER_L1D_FLUSH_COND)
282 static_branch_enable(&vmx_l1d_flush_cond);
284 static_branch_disable(&vmx_l1d_flush_cond);
288 static int vmentry_l1d_flush_parse(const char *s)
293 for (i = 0; i < ARRAY_SIZE(vmentry_l1d_param); i++) {
294 if (vmentry_l1d_param[i].for_parse &&
295 sysfs_streq(s, vmentry_l1d_param[i].option))
302 static int vmentry_l1d_flush_set(const char *s, const struct kernel_param *kp)
306 l1tf = vmentry_l1d_flush_parse(s);
310 if (!boot_cpu_has(X86_BUG_L1TF))
314 * Has vmx_init() run already? If not then this is the pre init
315 * parameter parsing. In that case just store the value and let
316 * vmx_init() do the proper setup after enable_ept has been
319 if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_AUTO) {
320 vmentry_l1d_flush_param = l1tf;
324 mutex_lock(&vmx_l1d_flush_mutex);
325 ret = vmx_setup_l1d_flush(l1tf);
326 mutex_unlock(&vmx_l1d_flush_mutex);
330 static int vmentry_l1d_flush_get(char *s, const struct kernel_param *kp)
332 if (WARN_ON_ONCE(l1tf_vmx_mitigation >= ARRAY_SIZE(vmentry_l1d_param)))
333 return sprintf(s, "???\n");
335 return sprintf(s, "%s\n", vmentry_l1d_param[l1tf_vmx_mitigation].option);
338 static const struct kernel_param_ops vmentry_l1d_flush_ops = {
339 .set = vmentry_l1d_flush_set,
340 .get = vmentry_l1d_flush_get,
342 module_param_cb(vmentry_l1d_flush, &vmentry_l1d_flush_ops, NULL, 0644);
344 static bool guest_state_valid(struct kvm_vcpu *vcpu);
345 static u32 vmx_segment_access_rights(struct kvm_segment *var);
346 static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
349 void vmx_vmexit(void);
351 #define vmx_insn_failed(fmt...) \
354 pr_warn_ratelimited(fmt); \
357 asmlinkage void vmread_error(unsigned long field, bool fault)
360 kvm_spurious_fault();
362 vmx_insn_failed("kvm: vmread failed: field=%lx\n", field);
365 noinline void vmwrite_error(unsigned long field, unsigned long value)
367 vmx_insn_failed("kvm: vmwrite failed: field=%lx val=%lx err=%d\n",
368 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
371 noinline void vmclear_error(struct vmcs *vmcs, u64 phys_addr)
373 vmx_insn_failed("kvm: vmclear failed: %p/%llx\n", vmcs, phys_addr);
376 noinline void vmptrld_error(struct vmcs *vmcs, u64 phys_addr)
378 vmx_insn_failed("kvm: vmptrld failed: %p/%llx\n", vmcs, phys_addr);
381 noinline void invvpid_error(unsigned long ext, u16 vpid, gva_t gva)
383 vmx_insn_failed("kvm: invvpid failed: ext=0x%lx vpid=%u gva=0x%lx\n",
387 noinline void invept_error(unsigned long ext, u64 eptp, gpa_t gpa)
389 vmx_insn_failed("kvm: invept failed: ext=0x%lx eptp=%llx gpa=0x%llx\n",
393 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
394 DEFINE_PER_CPU(struct vmcs *, current_vmcs);
396 * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
397 * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
399 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
402 * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
403 * can find which vCPU should be waken up.
405 static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu);
406 static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock);
408 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
409 static DEFINE_SPINLOCK(vmx_vpid_lock);
411 struct vmcs_config vmcs_config;
412 struct vmx_capability vmx_capability;
414 #define VMX_SEGMENT_FIELD(seg) \
415 [VCPU_SREG_##seg] = { \
416 .selector = GUEST_##seg##_SELECTOR, \
417 .base = GUEST_##seg##_BASE, \
418 .limit = GUEST_##seg##_LIMIT, \
419 .ar_bytes = GUEST_##seg##_AR_BYTES, \
422 static const struct kvm_vmx_segment_field {
427 } kvm_vmx_segment_fields[] = {
428 VMX_SEGMENT_FIELD(CS),
429 VMX_SEGMENT_FIELD(DS),
430 VMX_SEGMENT_FIELD(ES),
431 VMX_SEGMENT_FIELD(FS),
432 VMX_SEGMENT_FIELD(GS),
433 VMX_SEGMENT_FIELD(SS),
434 VMX_SEGMENT_FIELD(TR),
435 VMX_SEGMENT_FIELD(LDTR),
438 static unsigned long host_idt_base;
441 * Though SYSCALL is only supported in 64-bit mode on Intel CPUs, kvm
442 * will emulate SYSCALL in legacy mode if the vendor string in guest
443 * CPUID.0:{EBX,ECX,EDX} is "AuthenticAMD" or "AMDisbetter!" To
444 * support this emulation, IA32_STAR must always be included in
445 * vmx_msr_index[], even in i386 builds.
447 const u32 vmx_msr_index[] = {
449 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
451 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
455 #if IS_ENABLED(CONFIG_HYPERV)
456 static bool __read_mostly enlightened_vmcs = true;
457 module_param(enlightened_vmcs, bool, 0444);
459 /* check_ept_pointer() should be under protection of ept_pointer_lock. */
460 static void check_ept_pointer_match(struct kvm *kvm)
462 struct kvm_vcpu *vcpu;
463 u64 tmp_eptp = INVALID_PAGE;
466 kvm_for_each_vcpu(i, vcpu, kvm) {
467 if (!VALID_PAGE(tmp_eptp)) {
468 tmp_eptp = to_vmx(vcpu)->ept_pointer;
469 } else if (tmp_eptp != to_vmx(vcpu)->ept_pointer) {
470 to_kvm_vmx(kvm)->ept_pointers_match
471 = EPT_POINTERS_MISMATCH;
476 to_kvm_vmx(kvm)->ept_pointers_match = EPT_POINTERS_MATCH;
479 static int kvm_fill_hv_flush_list_func(struct hv_guest_mapping_flush_list *flush,
482 struct kvm_tlb_range *range = data;
484 return hyperv_fill_flush_guest_mapping_list(flush, range->start_gfn,
488 static inline int __hv_remote_flush_tlb_with_range(struct kvm *kvm,
489 struct kvm_vcpu *vcpu, struct kvm_tlb_range *range)
491 u64 ept_pointer = to_vmx(vcpu)->ept_pointer;
494 * FLUSH_GUEST_PHYSICAL_ADDRESS_SPACE hypercall needs address
495 * of the base of EPT PML4 table, strip off EPT configuration
499 return hyperv_flush_guest_mapping_range(ept_pointer & PAGE_MASK,
500 kvm_fill_hv_flush_list_func, (void *)range);
502 return hyperv_flush_guest_mapping(ept_pointer & PAGE_MASK);
505 static int hv_remote_flush_tlb_with_range(struct kvm *kvm,
506 struct kvm_tlb_range *range)
508 struct kvm_vcpu *vcpu;
511 spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
513 if (to_kvm_vmx(kvm)->ept_pointers_match == EPT_POINTERS_CHECK)
514 check_ept_pointer_match(kvm);
516 if (to_kvm_vmx(kvm)->ept_pointers_match != EPT_POINTERS_MATCH) {
517 kvm_for_each_vcpu(i, vcpu, kvm) {
518 /* If ept_pointer is invalid pointer, bypass flush request. */
519 if (VALID_PAGE(to_vmx(vcpu)->ept_pointer))
520 ret |= __hv_remote_flush_tlb_with_range(
524 ret = __hv_remote_flush_tlb_with_range(kvm,
525 kvm_get_vcpu(kvm, 0), range);
528 spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
531 static int hv_remote_flush_tlb(struct kvm *kvm)
533 return hv_remote_flush_tlb_with_range(kvm, NULL);
536 static int hv_enable_direct_tlbflush(struct kvm_vcpu *vcpu)
538 struct hv_enlightened_vmcs *evmcs;
539 struct hv_partition_assist_pg **p_hv_pa_pg =
540 &vcpu->kvm->arch.hyperv.hv_pa_pg;
542 * Synthetic VM-Exit is not enabled in current code and so All
543 * evmcs in singe VM shares same assist page.
546 *p_hv_pa_pg = kzalloc(PAGE_SIZE, GFP_KERNEL);
551 evmcs = (struct hv_enlightened_vmcs *)to_vmx(vcpu)->loaded_vmcs->vmcs;
553 evmcs->partition_assist_page =
555 evmcs->hv_vm_id = (unsigned long)vcpu->kvm;
556 evmcs->hv_enlightenments_control.nested_flush_hypercall = 1;
561 #endif /* IS_ENABLED(CONFIG_HYPERV) */
564 * Comment's format: document - errata name - stepping - processor name.
566 * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
568 static u32 vmx_preemption_cpu_tfms[] = {
569 /* 323344.pdf - BA86 - D0 - Xeon 7500 Series */
571 /* 323056.pdf - AAX65 - C2 - Xeon L3406 */
572 /* 322814.pdf - AAT59 - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
573 /* 322911.pdf - AAU65 - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
575 /* 322911.pdf - AAU65 - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
577 /* 322373.pdf - AAO95 - B1 - Xeon 3400 Series */
578 /* 322166.pdf - AAN92 - B1 - i7-800 and i5-700 Desktop */
580 * 320767.pdf - AAP86 - B1 -
581 * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
584 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
586 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
588 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
590 /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
591 /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
592 /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
594 /* Xeon E3-1220 V2 */
598 static inline bool cpu_has_broken_vmx_preemption_timer(void)
600 u32 eax = cpuid_eax(0x00000001), i;
602 /* Clear the reserved bits */
603 eax &= ~(0x3U << 14 | 0xfU << 28);
604 for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
605 if (eax == vmx_preemption_cpu_tfms[i])
611 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
613 return flexpriority_enabled && lapic_in_kernel(vcpu);
616 static inline bool report_flexpriority(void)
618 return flexpriority_enabled;
621 static inline int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
625 for (i = 0; i < vmx->nmsrs; ++i)
626 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
631 struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
635 i = __find_msr_index(vmx, msr);
637 return &vmx->guest_msrs[i];
641 static int vmx_set_guest_msr(struct vcpu_vmx *vmx, struct shared_msr_entry *msr, u64 data)
645 u64 old_msr_data = msr->data;
647 if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
649 ret = kvm_set_shared_msr(msr->index, msr->data,
653 msr->data = old_msr_data;
658 #ifdef CONFIG_KEXEC_CORE
659 static void crash_vmclear_local_loaded_vmcss(void)
661 int cpu = raw_smp_processor_id();
662 struct loaded_vmcs *v;
664 list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
665 loaded_vmcss_on_cpu_link)
668 #endif /* CONFIG_KEXEC_CORE */
670 static void __loaded_vmcs_clear(void *arg)
672 struct loaded_vmcs *loaded_vmcs = arg;
673 int cpu = raw_smp_processor_id();
675 if (loaded_vmcs->cpu != cpu)
676 return; /* vcpu migration can race with cpu offline */
677 if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
678 per_cpu(current_vmcs, cpu) = NULL;
680 vmcs_clear(loaded_vmcs->vmcs);
681 if (loaded_vmcs->shadow_vmcs && loaded_vmcs->launched)
682 vmcs_clear(loaded_vmcs->shadow_vmcs);
684 list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
687 * Ensure all writes to loaded_vmcs, including deleting it from its
688 * current percpu list, complete before setting loaded_vmcs->vcpu to
689 * -1, otherwise a different cpu can see vcpu == -1 first and add
690 * loaded_vmcs to its percpu list before it's deleted from this cpu's
691 * list. Pairs with the smp_rmb() in vmx_vcpu_load_vmcs().
695 loaded_vmcs->cpu = -1;
696 loaded_vmcs->launched = 0;
699 void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
701 int cpu = loaded_vmcs->cpu;
704 smp_call_function_single(cpu,
705 __loaded_vmcs_clear, loaded_vmcs, 1);
708 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
712 u32 mask = 1 << (seg * SEG_FIELD_NR + field);
714 if (!kvm_register_is_available(&vmx->vcpu, VCPU_EXREG_SEGMENTS)) {
715 kvm_register_mark_available(&vmx->vcpu, VCPU_EXREG_SEGMENTS);
716 vmx->segment_cache.bitmask = 0;
718 ret = vmx->segment_cache.bitmask & mask;
719 vmx->segment_cache.bitmask |= mask;
723 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
725 u16 *p = &vmx->segment_cache.seg[seg].selector;
727 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
728 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
732 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
734 ulong *p = &vmx->segment_cache.seg[seg].base;
736 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
737 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
741 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
743 u32 *p = &vmx->segment_cache.seg[seg].limit;
745 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
746 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
750 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
752 u32 *p = &vmx->segment_cache.seg[seg].ar;
754 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
755 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
759 void update_exception_bitmap(struct kvm_vcpu *vcpu)
763 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
764 (1u << DB_VECTOR) | (1u << AC_VECTOR);
766 * Guest access to VMware backdoor ports could legitimately
767 * trigger #GP because of TSS I/O permission bitmap.
768 * We intercept those #GP and allow access to them anyway
771 if (enable_vmware_backdoor)
772 eb |= (1u << GP_VECTOR);
773 if ((vcpu->guest_debug &
774 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
775 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
776 eb |= 1u << BP_VECTOR;
777 if (to_vmx(vcpu)->rmode.vm86_active)
780 eb &= ~(1u << PF_VECTOR);
782 /* When we are running a nested L2 guest and L1 specified for it a
783 * certain exception bitmap, we must trap the same exceptions and pass
784 * them to L1. When running L2, we will only handle the exceptions
785 * specified above if L1 did not want them.
787 if (is_guest_mode(vcpu))
788 eb |= get_vmcs12(vcpu)->exception_bitmap;
790 vmcs_write32(EXCEPTION_BITMAP, eb);
794 * Check if MSR is intercepted for currently loaded MSR bitmap.
796 static bool msr_write_intercepted(struct kvm_vcpu *vcpu, u32 msr)
798 unsigned long *msr_bitmap;
799 int f = sizeof(unsigned long);
801 if (!cpu_has_vmx_msr_bitmap())
804 msr_bitmap = to_vmx(vcpu)->loaded_vmcs->msr_bitmap;
807 return !!test_bit(msr, msr_bitmap + 0x800 / f);
808 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
810 return !!test_bit(msr, msr_bitmap + 0xc00 / f);
816 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
817 unsigned long entry, unsigned long exit)
819 vm_entry_controls_clearbit(vmx, entry);
820 vm_exit_controls_clearbit(vmx, exit);
823 int vmx_find_msr_index(struct vmx_msrs *m, u32 msr)
827 for (i = 0; i < m->nr; ++i) {
828 if (m->val[i].index == msr)
834 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
837 struct msr_autoload *m = &vmx->msr_autoload;
841 if (cpu_has_load_ia32_efer()) {
842 clear_atomic_switch_msr_special(vmx,
843 VM_ENTRY_LOAD_IA32_EFER,
844 VM_EXIT_LOAD_IA32_EFER);
848 case MSR_CORE_PERF_GLOBAL_CTRL:
849 if (cpu_has_load_perf_global_ctrl()) {
850 clear_atomic_switch_msr_special(vmx,
851 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
852 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
857 i = vmx_find_msr_index(&m->guest, msr);
861 m->guest.val[i] = m->guest.val[m->guest.nr];
862 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
865 i = vmx_find_msr_index(&m->host, msr);
870 m->host.val[i] = m->host.val[m->host.nr];
871 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
874 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
875 unsigned long entry, unsigned long exit,
876 unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
877 u64 guest_val, u64 host_val)
879 vmcs_write64(guest_val_vmcs, guest_val);
880 if (host_val_vmcs != HOST_IA32_EFER)
881 vmcs_write64(host_val_vmcs, host_val);
882 vm_entry_controls_setbit(vmx, entry);
883 vm_exit_controls_setbit(vmx, exit);
886 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
887 u64 guest_val, u64 host_val, bool entry_only)
890 struct msr_autoload *m = &vmx->msr_autoload;
894 if (cpu_has_load_ia32_efer()) {
895 add_atomic_switch_msr_special(vmx,
896 VM_ENTRY_LOAD_IA32_EFER,
897 VM_EXIT_LOAD_IA32_EFER,
900 guest_val, host_val);
904 case MSR_CORE_PERF_GLOBAL_CTRL:
905 if (cpu_has_load_perf_global_ctrl()) {
906 add_atomic_switch_msr_special(vmx,
907 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
908 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
909 GUEST_IA32_PERF_GLOBAL_CTRL,
910 HOST_IA32_PERF_GLOBAL_CTRL,
911 guest_val, host_val);
915 case MSR_IA32_PEBS_ENABLE:
916 /* PEBS needs a quiescent period after being disabled (to write
917 * a record). Disabling PEBS through VMX MSR swapping doesn't
918 * provide that period, so a CPU could write host's record into
921 wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
924 i = vmx_find_msr_index(&m->guest, msr);
926 j = vmx_find_msr_index(&m->host, msr);
928 if ((i < 0 && m->guest.nr == NR_LOADSTORE_MSRS) ||
929 (j < 0 && m->host.nr == NR_LOADSTORE_MSRS)) {
930 printk_once(KERN_WARNING "Not enough msr switch entries. "
931 "Can't add msr %x\n", msr);
936 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
938 m->guest.val[i].index = msr;
939 m->guest.val[i].value = guest_val;
946 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
948 m->host.val[j].index = msr;
949 m->host.val[j].value = host_val;
952 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
954 u64 guest_efer = vmx->vcpu.arch.efer;
957 /* Shadow paging assumes NX to be available. */
959 guest_efer |= EFER_NX;
962 * LMA and LME handled by hardware; SCE meaningless outside long mode.
964 ignore_bits |= EFER_SCE;
966 ignore_bits |= EFER_LMA | EFER_LME;
967 /* SCE is meaningful only in long mode on Intel */
968 if (guest_efer & EFER_LMA)
969 ignore_bits &= ~(u64)EFER_SCE;
973 * On EPT, we can't emulate NX, so we must switch EFER atomically.
974 * On CPUs that support "load IA32_EFER", always switch EFER
975 * atomically, since it's faster than switching it manually.
977 if (cpu_has_load_ia32_efer() ||
978 (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
979 if (!(guest_efer & EFER_LMA))
980 guest_efer &= ~EFER_LME;
981 if (guest_efer != host_efer)
982 add_atomic_switch_msr(vmx, MSR_EFER,
983 guest_efer, host_efer, false);
985 clear_atomic_switch_msr(vmx, MSR_EFER);
988 clear_atomic_switch_msr(vmx, MSR_EFER);
990 guest_efer &= ~ignore_bits;
991 guest_efer |= host_efer & ignore_bits;
993 vmx->guest_msrs[efer_offset].data = guest_efer;
994 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1000 #ifdef CONFIG_X86_32
1002 * On 32-bit kernels, VM exits still load the FS and GS bases from the
1003 * VMCS rather than the segment table. KVM uses this helper to figure
1004 * out the current bases to poke them into the VMCS before entry.
1006 static unsigned long segment_base(u16 selector)
1008 struct desc_struct *table;
1011 if (!(selector & ~SEGMENT_RPL_MASK))
1014 table = get_current_gdt_ro();
1016 if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
1017 u16 ldt_selector = kvm_read_ldt();
1019 if (!(ldt_selector & ~SEGMENT_RPL_MASK))
1022 table = (struct desc_struct *)segment_base(ldt_selector);
1024 v = get_desc_base(&table[selector >> 3]);
1029 static inline bool pt_can_write_msr(struct vcpu_vmx *vmx)
1031 return vmx_pt_mode_is_host_guest() &&
1032 !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
1035 static inline void pt_load_msr(struct pt_ctx *ctx, u32 addr_range)
1039 wrmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1040 wrmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1041 wrmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1042 wrmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1043 for (i = 0; i < addr_range; i++) {
1044 wrmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1045 wrmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1049 static inline void pt_save_msr(struct pt_ctx *ctx, u32 addr_range)
1053 rdmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1054 rdmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1055 rdmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1056 rdmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1057 for (i = 0; i < addr_range; i++) {
1058 rdmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1059 rdmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1063 static void pt_guest_enter(struct vcpu_vmx *vmx)
1065 if (vmx_pt_mode_is_system())
1069 * GUEST_IA32_RTIT_CTL is already set in the VMCS.
1070 * Save host state before VM entry.
1072 rdmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1073 if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1074 wrmsrl(MSR_IA32_RTIT_CTL, 0);
1075 pt_save_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1076 pt_load_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1080 static void pt_guest_exit(struct vcpu_vmx *vmx)
1082 if (vmx_pt_mode_is_system())
1085 if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1086 pt_save_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1087 pt_load_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1090 /* Reload host state (IA32_RTIT_CTL will be cleared on VM exit). */
1091 wrmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1094 void vmx_set_host_fs_gs(struct vmcs_host_state *host, u16 fs_sel, u16 gs_sel,
1095 unsigned long fs_base, unsigned long gs_base)
1097 if (unlikely(fs_sel != host->fs_sel)) {
1099 vmcs_write16(HOST_FS_SELECTOR, fs_sel);
1101 vmcs_write16(HOST_FS_SELECTOR, 0);
1102 host->fs_sel = fs_sel;
1104 if (unlikely(gs_sel != host->gs_sel)) {
1106 vmcs_write16(HOST_GS_SELECTOR, gs_sel);
1108 vmcs_write16(HOST_GS_SELECTOR, 0);
1109 host->gs_sel = gs_sel;
1111 if (unlikely(fs_base != host->fs_base)) {
1112 vmcs_writel(HOST_FS_BASE, fs_base);
1113 host->fs_base = fs_base;
1115 if (unlikely(gs_base != host->gs_base)) {
1116 vmcs_writel(HOST_GS_BASE, gs_base);
1117 host->gs_base = gs_base;
1121 void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu)
1123 struct vcpu_vmx *vmx = to_vmx(vcpu);
1124 struct vmcs_host_state *host_state;
1125 #ifdef CONFIG_X86_64
1126 int cpu = raw_smp_processor_id();
1128 unsigned long fs_base, gs_base;
1132 vmx->req_immediate_exit = false;
1135 * Note that guest MSRs to be saved/restored can also be changed
1136 * when guest state is loaded. This happens when guest transitions
1137 * to/from long-mode by setting MSR_EFER.LMA.
1139 if (!vmx->guest_msrs_ready) {
1140 vmx->guest_msrs_ready = true;
1141 for (i = 0; i < vmx->save_nmsrs; ++i)
1142 kvm_set_shared_msr(vmx->guest_msrs[i].index,
1143 vmx->guest_msrs[i].data,
1144 vmx->guest_msrs[i].mask);
1148 if (vmx->nested.need_vmcs12_to_shadow_sync)
1149 nested_sync_vmcs12_to_shadow(vcpu);
1151 if (vmx->guest_state_loaded)
1154 host_state = &vmx->loaded_vmcs->host_state;
1157 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
1158 * allow segment selectors with cpl > 0 or ti == 1.
1160 host_state->ldt_sel = kvm_read_ldt();
1162 #ifdef CONFIG_X86_64
1163 savesegment(ds, host_state->ds_sel);
1164 savesegment(es, host_state->es_sel);
1166 gs_base = cpu_kernelmode_gs_base(cpu);
1167 if (likely(is_64bit_mm(current->mm))) {
1168 save_fsgs_for_kvm();
1169 fs_sel = current->thread.fsindex;
1170 gs_sel = current->thread.gsindex;
1171 fs_base = current->thread.fsbase;
1172 vmx->msr_host_kernel_gs_base = current->thread.gsbase;
1174 savesegment(fs, fs_sel);
1175 savesegment(gs, gs_sel);
1176 fs_base = read_msr(MSR_FS_BASE);
1177 vmx->msr_host_kernel_gs_base = read_msr(MSR_KERNEL_GS_BASE);
1180 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1182 savesegment(fs, fs_sel);
1183 savesegment(gs, gs_sel);
1184 fs_base = segment_base(fs_sel);
1185 gs_base = segment_base(gs_sel);
1188 vmx_set_host_fs_gs(host_state, fs_sel, gs_sel, fs_base, gs_base);
1189 vmx->guest_state_loaded = true;
1192 static void vmx_prepare_switch_to_host(struct vcpu_vmx *vmx)
1194 struct vmcs_host_state *host_state;
1196 if (!vmx->guest_state_loaded)
1199 host_state = &vmx->loaded_vmcs->host_state;
1201 ++vmx->vcpu.stat.host_state_reload;
1203 #ifdef CONFIG_X86_64
1204 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1206 if (host_state->ldt_sel || (host_state->gs_sel & 7)) {
1207 kvm_load_ldt(host_state->ldt_sel);
1208 #ifdef CONFIG_X86_64
1209 load_gs_index(host_state->gs_sel);
1211 loadsegment(gs, host_state->gs_sel);
1214 if (host_state->fs_sel & 7)
1215 loadsegment(fs, host_state->fs_sel);
1216 #ifdef CONFIG_X86_64
1217 if (unlikely(host_state->ds_sel | host_state->es_sel)) {
1218 loadsegment(ds, host_state->ds_sel);
1219 loadsegment(es, host_state->es_sel);
1222 invalidate_tss_limit();
1223 #ifdef CONFIG_X86_64
1224 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1226 load_fixmap_gdt(raw_smp_processor_id());
1227 vmx->guest_state_loaded = false;
1228 vmx->guest_msrs_ready = false;
1231 #ifdef CONFIG_X86_64
1232 static u64 vmx_read_guest_kernel_gs_base(struct vcpu_vmx *vmx)
1235 if (vmx->guest_state_loaded)
1236 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1238 return vmx->msr_guest_kernel_gs_base;
1241 static void vmx_write_guest_kernel_gs_base(struct vcpu_vmx *vmx, u64 data)
1244 if (vmx->guest_state_loaded)
1245 wrmsrl(MSR_KERNEL_GS_BASE, data);
1247 vmx->msr_guest_kernel_gs_base = data;
1251 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
1253 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1254 struct pi_desc old, new;
1258 * In case of hot-plug or hot-unplug, we may have to undo
1259 * vmx_vcpu_pi_put even if there is no assigned device. And we
1260 * always keep PI.NDST up to date for simplicity: it makes the
1261 * code easier, and CPU migration is not a fast path.
1263 if (!pi_test_sn(pi_desc) && vcpu->cpu == cpu)
1267 * If the 'nv' field is POSTED_INTR_WAKEUP_VECTOR, do not change
1268 * PI.NDST: pi_post_block is the one expected to change PID.NDST and the
1269 * wakeup handler expects the vCPU to be on the blocked_vcpu_list that
1270 * matches PI.NDST. Otherwise, a vcpu may not be able to be woken up
1273 if (pi_desc->nv == POSTED_INTR_WAKEUP_VECTOR || vcpu->cpu == cpu) {
1274 pi_clear_sn(pi_desc);
1275 goto after_clear_sn;
1278 /* The full case. */
1280 old.control = new.control = pi_desc->control;
1282 dest = cpu_physical_id(cpu);
1284 if (x2apic_enabled())
1287 new.ndst = (dest << 8) & 0xFF00;
1290 } while (cmpxchg64(&pi_desc->control, old.control,
1291 new.control) != old.control);
1296 * Clear SN before reading the bitmap. The VT-d firmware
1297 * writes the bitmap and reads SN atomically (5.2.3 in the
1298 * spec), so it doesn't really have a memory barrier that
1299 * pairs with this, but we cannot do that and we need one.
1301 smp_mb__after_atomic();
1303 if (!pi_is_pir_empty(pi_desc))
1307 void vmx_vcpu_load_vmcs(struct kvm_vcpu *vcpu, int cpu)
1309 struct vcpu_vmx *vmx = to_vmx(vcpu);
1310 bool already_loaded = vmx->loaded_vmcs->cpu == cpu;
1312 if (!already_loaded) {
1313 loaded_vmcs_clear(vmx->loaded_vmcs);
1314 local_irq_disable();
1317 * Ensure loaded_vmcs->cpu is read before adding loaded_vmcs to
1318 * this cpu's percpu list, otherwise it may not yet be deleted
1319 * from its previous cpu's percpu list. Pairs with the
1320 * smb_wmb() in __loaded_vmcs_clear().
1324 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1325 &per_cpu(loaded_vmcss_on_cpu, cpu));
1329 if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
1330 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1331 vmcs_load(vmx->loaded_vmcs->vmcs);
1332 indirect_branch_prediction_barrier();
1335 if (!already_loaded) {
1336 void *gdt = get_current_gdt_ro();
1337 unsigned long sysenter_esp;
1339 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1342 * Linux uses per-cpu TSS and GDT, so set these when switching
1343 * processors. See 22.2.4.
1345 vmcs_writel(HOST_TR_BASE,
1346 (unsigned long)&get_cpu_entry_area(cpu)->tss.x86_tss);
1347 vmcs_writel(HOST_GDTR_BASE, (unsigned long)gdt); /* 22.2.4 */
1349 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1350 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1352 vmx->loaded_vmcs->cpu = cpu;
1355 /* Setup TSC multiplier */
1356 if (kvm_has_tsc_control &&
1357 vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio)
1358 decache_tsc_multiplier(vmx);
1362 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1363 * vcpu mutex is already taken.
1365 void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1367 struct vcpu_vmx *vmx = to_vmx(vcpu);
1369 vmx_vcpu_load_vmcs(vcpu, cpu);
1371 vmx_vcpu_pi_load(vcpu, cpu);
1373 vmx->host_pkru = read_pkru();
1374 vmx->host_debugctlmsr = get_debugctlmsr();
1377 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
1379 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1381 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
1382 !irq_remapping_cap(IRQ_POSTING_CAP) ||
1383 !kvm_vcpu_apicv_active(vcpu))
1386 /* Set SN when the vCPU is preempted */
1387 if (vcpu->preempted)
1391 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1393 vmx_vcpu_pi_put(vcpu);
1395 vmx_prepare_switch_to_host(to_vmx(vcpu));
1398 static bool emulation_required(struct kvm_vcpu *vcpu)
1400 return emulate_invalid_guest_state && !guest_state_valid(vcpu);
1403 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1405 struct vcpu_vmx *vmx = to_vmx(vcpu);
1406 unsigned long rflags, save_rflags;
1408 if (!kvm_register_is_available(vcpu, VCPU_EXREG_RFLAGS)) {
1409 kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1410 rflags = vmcs_readl(GUEST_RFLAGS);
1411 if (vmx->rmode.vm86_active) {
1412 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1413 save_rflags = vmx->rmode.save_rflags;
1414 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1416 vmx->rflags = rflags;
1421 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1423 struct vcpu_vmx *vmx = to_vmx(vcpu);
1424 unsigned long old_rflags;
1426 if (enable_unrestricted_guest) {
1427 kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1428 vmx->rflags = rflags;
1429 vmcs_writel(GUEST_RFLAGS, rflags);
1433 old_rflags = vmx_get_rflags(vcpu);
1434 vmx->rflags = rflags;
1435 if (vmx->rmode.vm86_active) {
1436 vmx->rmode.save_rflags = rflags;
1437 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1439 vmcs_writel(GUEST_RFLAGS, rflags);
1441 if ((old_rflags ^ vmx->rflags) & X86_EFLAGS_VM)
1442 vmx->emulation_required = emulation_required(vcpu);
1445 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
1447 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1450 if (interruptibility & GUEST_INTR_STATE_STI)
1451 ret |= KVM_X86_SHADOW_INT_STI;
1452 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1453 ret |= KVM_X86_SHADOW_INT_MOV_SS;
1458 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1460 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1461 u32 interruptibility = interruptibility_old;
1463 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1465 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1466 interruptibility |= GUEST_INTR_STATE_MOV_SS;
1467 else if (mask & KVM_X86_SHADOW_INT_STI)
1468 interruptibility |= GUEST_INTR_STATE_STI;
1470 if ((interruptibility != interruptibility_old))
1471 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1474 static int vmx_rtit_ctl_check(struct kvm_vcpu *vcpu, u64 data)
1476 struct vcpu_vmx *vmx = to_vmx(vcpu);
1477 unsigned long value;
1480 * Any MSR write that attempts to change bits marked reserved will
1483 if (data & vmx->pt_desc.ctl_bitmask)
1487 * Any attempt to modify IA32_RTIT_CTL while TraceEn is set will
1488 * result in a #GP unless the same write also clears TraceEn.
1490 if ((vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) &&
1491 ((vmx->pt_desc.guest.ctl ^ data) & ~RTIT_CTL_TRACEEN))
1495 * WRMSR to IA32_RTIT_CTL that sets TraceEn but clears this bit
1496 * and FabricEn would cause #GP, if
1497 * CPUID.(EAX=14H, ECX=0):ECX.SNGLRGNOUT[bit 2] = 0
1499 if ((data & RTIT_CTL_TRACEEN) && !(data & RTIT_CTL_TOPA) &&
1500 !(data & RTIT_CTL_FABRIC_EN) &&
1501 !intel_pt_validate_cap(vmx->pt_desc.caps,
1502 PT_CAP_single_range_output))
1506 * MTCFreq, CycThresh and PSBFreq encodings check, any MSR write that
1507 * utilize encodings marked reserved will casue a #GP fault.
1509 value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc_periods);
1510 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc) &&
1511 !test_bit((data & RTIT_CTL_MTC_RANGE) >>
1512 RTIT_CTL_MTC_RANGE_OFFSET, &value))
1514 value = intel_pt_validate_cap(vmx->pt_desc.caps,
1515 PT_CAP_cycle_thresholds);
1516 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1517 !test_bit((data & RTIT_CTL_CYC_THRESH) >>
1518 RTIT_CTL_CYC_THRESH_OFFSET, &value))
1520 value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_periods);
1521 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1522 !test_bit((data & RTIT_CTL_PSB_FREQ) >>
1523 RTIT_CTL_PSB_FREQ_OFFSET, &value))
1527 * If ADDRx_CFG is reserved or the encodings is >2 will
1528 * cause a #GP fault.
1530 value = (data & RTIT_CTL_ADDR0) >> RTIT_CTL_ADDR0_OFFSET;
1531 if ((value && (vmx->pt_desc.addr_range < 1)) || (value > 2))
1533 value = (data & RTIT_CTL_ADDR1) >> RTIT_CTL_ADDR1_OFFSET;
1534 if ((value && (vmx->pt_desc.addr_range < 2)) || (value > 2))
1536 value = (data & RTIT_CTL_ADDR2) >> RTIT_CTL_ADDR2_OFFSET;
1537 if ((value && (vmx->pt_desc.addr_range < 3)) || (value > 2))
1539 value = (data & RTIT_CTL_ADDR3) >> RTIT_CTL_ADDR3_OFFSET;
1540 if ((value && (vmx->pt_desc.addr_range < 4)) || (value > 2))
1546 static int skip_emulated_instruction(struct kvm_vcpu *vcpu)
1551 * Using VMCS.VM_EXIT_INSTRUCTION_LEN on EPT misconfig depends on
1552 * undefined behavior: Intel's SDM doesn't mandate the VMCS field be
1553 * set when EPT misconfig occurs. In practice, real hardware updates
1554 * VM_EXIT_INSTRUCTION_LEN on EPT misconfig, but other hypervisors
1555 * (namely Hyper-V) don't set it due to it being undefined behavior,
1556 * i.e. we end up advancing IP with some random value.
1558 if (!static_cpu_has(X86_FEATURE_HYPERVISOR) ||
1559 to_vmx(vcpu)->exit_reason != EXIT_REASON_EPT_MISCONFIG) {
1560 rip = kvm_rip_read(vcpu);
1561 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1562 kvm_rip_write(vcpu, rip);
1564 if (!kvm_emulate_instruction(vcpu, EMULTYPE_SKIP))
1568 /* skipping an emulated instruction also counts */
1569 vmx_set_interrupt_shadow(vcpu, 0);
1576 * Recognizes a pending MTF VM-exit and records the nested state for later
1579 static void vmx_update_emulated_instruction(struct kvm_vcpu *vcpu)
1581 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1582 struct vcpu_vmx *vmx = to_vmx(vcpu);
1584 if (!is_guest_mode(vcpu))
1588 * Per the SDM, MTF takes priority over debug-trap exceptions besides
1589 * T-bit traps. As instruction emulation is completed (i.e. at the
1590 * instruction boundary), any #DB exception pending delivery must be a
1591 * debug-trap. Record the pending MTF state to be delivered in
1592 * vmx_check_nested_events().
1594 if (nested_cpu_has_mtf(vmcs12) &&
1595 (!vcpu->arch.exception.pending ||
1596 vcpu->arch.exception.nr == DB_VECTOR))
1597 vmx->nested.mtf_pending = true;
1599 vmx->nested.mtf_pending = false;
1602 static int vmx_skip_emulated_instruction(struct kvm_vcpu *vcpu)
1604 vmx_update_emulated_instruction(vcpu);
1605 return skip_emulated_instruction(vcpu);
1608 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1611 * Ensure that we clear the HLT state in the VMCS. We don't need to
1612 * explicitly skip the instruction because if the HLT state is set,
1613 * then the instruction is already executing and RIP has already been
1616 if (kvm_hlt_in_guest(vcpu->kvm) &&
1617 vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1618 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1621 static void vmx_queue_exception(struct kvm_vcpu *vcpu)
1623 struct vcpu_vmx *vmx = to_vmx(vcpu);
1624 unsigned nr = vcpu->arch.exception.nr;
1625 bool has_error_code = vcpu->arch.exception.has_error_code;
1626 u32 error_code = vcpu->arch.exception.error_code;
1627 u32 intr_info = nr | INTR_INFO_VALID_MASK;
1629 kvm_deliver_exception_payload(vcpu);
1631 if (has_error_code) {
1632 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1633 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1636 if (vmx->rmode.vm86_active) {
1638 if (kvm_exception_is_soft(nr))
1639 inc_eip = vcpu->arch.event_exit_inst_len;
1640 kvm_inject_realmode_interrupt(vcpu, nr, inc_eip);
1644 WARN_ON_ONCE(vmx->emulation_required);
1646 if (kvm_exception_is_soft(nr)) {
1647 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1648 vmx->vcpu.arch.event_exit_inst_len);
1649 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1651 intr_info |= INTR_TYPE_HARD_EXCEPTION;
1653 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1655 vmx_clear_hlt(vcpu);
1659 * Swap MSR entry in host/guest MSR entry array.
1661 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1663 struct shared_msr_entry tmp;
1665 tmp = vmx->guest_msrs[to];
1666 vmx->guest_msrs[to] = vmx->guest_msrs[from];
1667 vmx->guest_msrs[from] = tmp;
1671 * Set up the vmcs to automatically save and restore system
1672 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
1673 * mode, as fiddling with msrs is very expensive.
1675 static void setup_msrs(struct vcpu_vmx *vmx)
1677 int save_nmsrs, index;
1680 #ifdef CONFIG_X86_64
1682 * The SYSCALL MSRs are only needed on long mode guests, and only
1683 * when EFER.SCE is set.
1685 if (is_long_mode(&vmx->vcpu) && (vmx->vcpu.arch.efer & EFER_SCE)) {
1686 index = __find_msr_index(vmx, MSR_STAR);
1688 move_msr_up(vmx, index, save_nmsrs++);
1689 index = __find_msr_index(vmx, MSR_LSTAR);
1691 move_msr_up(vmx, index, save_nmsrs++);
1692 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1694 move_msr_up(vmx, index, save_nmsrs++);
1697 index = __find_msr_index(vmx, MSR_EFER);
1698 if (index >= 0 && update_transition_efer(vmx, index))
1699 move_msr_up(vmx, index, save_nmsrs++);
1700 index = __find_msr_index(vmx, MSR_TSC_AUX);
1701 if (index >= 0 && guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDTSCP))
1702 move_msr_up(vmx, index, save_nmsrs++);
1703 index = __find_msr_index(vmx, MSR_IA32_TSX_CTRL);
1705 move_msr_up(vmx, index, save_nmsrs++);
1707 vmx->save_nmsrs = save_nmsrs;
1708 vmx->guest_msrs_ready = false;
1710 if (cpu_has_vmx_msr_bitmap())
1711 vmx_update_msr_bitmap(&vmx->vcpu);
1714 static u64 vmx_read_l1_tsc_offset(struct kvm_vcpu *vcpu)
1716 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1718 if (is_guest_mode(vcpu) &&
1719 (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING))
1720 return vcpu->arch.tsc_offset - vmcs12->tsc_offset;
1722 return vcpu->arch.tsc_offset;
1725 static u64 vmx_write_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1727 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1728 u64 g_tsc_offset = 0;
1731 * We're here if L1 chose not to trap WRMSR to TSC. According
1732 * to the spec, this should set L1's TSC; The offset that L1
1733 * set for L2 remains unchanged, and still needs to be added
1734 * to the newly set TSC to get L2's TSC.
1736 if (is_guest_mode(vcpu) &&
1737 (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING))
1738 g_tsc_offset = vmcs12->tsc_offset;
1740 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1741 vcpu->arch.tsc_offset - g_tsc_offset,
1743 vmcs_write64(TSC_OFFSET, offset + g_tsc_offset);
1744 return offset + g_tsc_offset;
1748 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
1749 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
1750 * all guests if the "nested" module option is off, and can also be disabled
1751 * for a single guest by disabling its VMX cpuid bit.
1753 bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
1755 return nested && guest_cpuid_has(vcpu, X86_FEATURE_VMX);
1758 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
1761 uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
1763 return !(val & ~valid_bits);
1766 static int vmx_get_msr_feature(struct kvm_msr_entry *msr)
1768 switch (msr->index) {
1769 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1772 return vmx_get_vmx_msr(&vmcs_config.nested, msr->index, &msr->data);
1779 * Reads an msr value (of 'msr_index') into 'pdata'.
1780 * Returns 0 on success, non-0 otherwise.
1781 * Assumes vcpu_load() was already called.
1783 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1785 struct vcpu_vmx *vmx = to_vmx(vcpu);
1786 struct shared_msr_entry *msr;
1789 switch (msr_info->index) {
1790 #ifdef CONFIG_X86_64
1792 msr_info->data = vmcs_readl(GUEST_FS_BASE);
1795 msr_info->data = vmcs_readl(GUEST_GS_BASE);
1797 case MSR_KERNEL_GS_BASE:
1798 msr_info->data = vmx_read_guest_kernel_gs_base(vmx);
1802 return kvm_get_msr_common(vcpu, msr_info);
1803 case MSR_IA32_TSX_CTRL:
1804 if (!msr_info->host_initiated &&
1805 !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
1807 goto find_shared_msr;
1808 case MSR_IA32_UMWAIT_CONTROL:
1809 if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
1812 msr_info->data = vmx->msr_ia32_umwait_control;
1814 case MSR_IA32_SPEC_CTRL:
1815 if (!msr_info->host_initiated &&
1816 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1819 msr_info->data = to_vmx(vcpu)->spec_ctrl;
1821 case MSR_IA32_SYSENTER_CS:
1822 msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
1824 case MSR_IA32_SYSENTER_EIP:
1825 msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
1827 case MSR_IA32_SYSENTER_ESP:
1828 msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
1830 case MSR_IA32_BNDCFGS:
1831 if (!kvm_mpx_supported() ||
1832 (!msr_info->host_initiated &&
1833 !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1835 msr_info->data = vmcs_read64(GUEST_BNDCFGS);
1837 case MSR_IA32_MCG_EXT_CTL:
1838 if (!msr_info->host_initiated &&
1839 !(vmx->msr_ia32_feature_control &
1840 FEAT_CTL_LMCE_ENABLED))
1842 msr_info->data = vcpu->arch.mcg_ext_ctl;
1844 case MSR_IA32_FEAT_CTL:
1845 msr_info->data = vmx->msr_ia32_feature_control;
1847 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1848 if (!nested_vmx_allowed(vcpu))
1850 if (vmx_get_vmx_msr(&vmx->nested.msrs, msr_info->index,
1854 * Enlightened VMCS v1 doesn't have certain fields, but buggy
1855 * Hyper-V versions are still trying to use corresponding
1856 * features when they are exposed. Filter out the essential
1859 if (!msr_info->host_initiated &&
1860 vmx->nested.enlightened_vmcs_enabled)
1861 nested_evmcs_filter_control_msr(msr_info->index,
1864 case MSR_IA32_RTIT_CTL:
1865 if (!vmx_pt_mode_is_host_guest())
1867 msr_info->data = vmx->pt_desc.guest.ctl;
1869 case MSR_IA32_RTIT_STATUS:
1870 if (!vmx_pt_mode_is_host_guest())
1872 msr_info->data = vmx->pt_desc.guest.status;
1874 case MSR_IA32_RTIT_CR3_MATCH:
1875 if (!vmx_pt_mode_is_host_guest() ||
1876 !intel_pt_validate_cap(vmx->pt_desc.caps,
1877 PT_CAP_cr3_filtering))
1879 msr_info->data = vmx->pt_desc.guest.cr3_match;
1881 case MSR_IA32_RTIT_OUTPUT_BASE:
1882 if (!vmx_pt_mode_is_host_guest() ||
1883 (!intel_pt_validate_cap(vmx->pt_desc.caps,
1884 PT_CAP_topa_output) &&
1885 !intel_pt_validate_cap(vmx->pt_desc.caps,
1886 PT_CAP_single_range_output)))
1888 msr_info->data = vmx->pt_desc.guest.output_base;
1890 case MSR_IA32_RTIT_OUTPUT_MASK:
1891 if (!vmx_pt_mode_is_host_guest() ||
1892 (!intel_pt_validate_cap(vmx->pt_desc.caps,
1893 PT_CAP_topa_output) &&
1894 !intel_pt_validate_cap(vmx->pt_desc.caps,
1895 PT_CAP_single_range_output)))
1897 msr_info->data = vmx->pt_desc.guest.output_mask;
1899 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
1900 index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
1901 if (!vmx_pt_mode_is_host_guest() ||
1902 (index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
1903 PT_CAP_num_address_ranges)))
1906 msr_info->data = vmx->pt_desc.guest.addr_b[index / 2];
1908 msr_info->data = vmx->pt_desc.guest.addr_a[index / 2];
1911 if (!msr_info->host_initiated &&
1912 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
1914 goto find_shared_msr;
1917 msr = find_msr_entry(vmx, msr_info->index);
1919 msr_info->data = msr->data;
1922 return kvm_get_msr_common(vcpu, msr_info);
1929 * Writes msr value into the appropriate "register".
1930 * Returns 0 on success, non-0 otherwise.
1931 * Assumes vcpu_load() was already called.
1933 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1935 struct vcpu_vmx *vmx = to_vmx(vcpu);
1936 struct shared_msr_entry *msr;
1938 u32 msr_index = msr_info->index;
1939 u64 data = msr_info->data;
1942 switch (msr_index) {
1944 ret = kvm_set_msr_common(vcpu, msr_info);
1946 #ifdef CONFIG_X86_64
1948 vmx_segment_cache_clear(vmx);
1949 vmcs_writel(GUEST_FS_BASE, data);
1952 vmx_segment_cache_clear(vmx);
1953 vmcs_writel(GUEST_GS_BASE, data);
1955 case MSR_KERNEL_GS_BASE:
1956 vmx_write_guest_kernel_gs_base(vmx, data);
1959 case MSR_IA32_SYSENTER_CS:
1960 if (is_guest_mode(vcpu))
1961 get_vmcs12(vcpu)->guest_sysenter_cs = data;
1962 vmcs_write32(GUEST_SYSENTER_CS, data);
1964 case MSR_IA32_SYSENTER_EIP:
1965 if (is_guest_mode(vcpu))
1966 get_vmcs12(vcpu)->guest_sysenter_eip = data;
1967 vmcs_writel(GUEST_SYSENTER_EIP, data);
1969 case MSR_IA32_SYSENTER_ESP:
1970 if (is_guest_mode(vcpu))
1971 get_vmcs12(vcpu)->guest_sysenter_esp = data;
1972 vmcs_writel(GUEST_SYSENTER_ESP, data);
1974 case MSR_IA32_DEBUGCTLMSR:
1975 if (is_guest_mode(vcpu) && get_vmcs12(vcpu)->vm_exit_controls &
1976 VM_EXIT_SAVE_DEBUG_CONTROLS)
1977 get_vmcs12(vcpu)->guest_ia32_debugctl = data;
1979 ret = kvm_set_msr_common(vcpu, msr_info);
1982 case MSR_IA32_BNDCFGS:
1983 if (!kvm_mpx_supported() ||
1984 (!msr_info->host_initiated &&
1985 !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1987 if (is_noncanonical_address(data & PAGE_MASK, vcpu) ||
1988 (data & MSR_IA32_BNDCFGS_RSVD))
1990 vmcs_write64(GUEST_BNDCFGS, data);
1992 case MSR_IA32_UMWAIT_CONTROL:
1993 if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
1996 /* The reserved bit 1 and non-32 bit [63:32] should be zero */
1997 if (data & (BIT_ULL(1) | GENMASK_ULL(63, 32)))
2000 vmx->msr_ia32_umwait_control = data;
2002 case MSR_IA32_SPEC_CTRL:
2003 if (!msr_info->host_initiated &&
2004 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
2007 if (data & ~kvm_spec_ctrl_valid_bits(vcpu))
2010 vmx->spec_ctrl = data;
2016 * When it's written (to non-zero) for the first time, pass
2020 * The handling of the MSR bitmap for L2 guests is done in
2021 * nested_vmx_prepare_msr_bitmap. We should not touch the
2022 * vmcs02.msr_bitmap here since it gets completely overwritten
2023 * in the merging. We update the vmcs01 here for L1 as well
2024 * since it will end up touching the MSR anyway now.
2026 vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap,
2030 case MSR_IA32_TSX_CTRL:
2031 if (!msr_info->host_initiated &&
2032 !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
2034 if (data & ~(TSX_CTRL_RTM_DISABLE | TSX_CTRL_CPUID_CLEAR))
2036 goto find_shared_msr;
2037 case MSR_IA32_PRED_CMD:
2038 if (!msr_info->host_initiated &&
2039 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
2042 if (data & ~PRED_CMD_IBPB)
2044 if (!boot_cpu_has(X86_FEATURE_SPEC_CTRL))
2049 wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
2053 * When it's written (to non-zero) for the first time, pass
2057 * The handling of the MSR bitmap for L2 guests is done in
2058 * nested_vmx_prepare_msr_bitmap. We should not touch the
2059 * vmcs02.msr_bitmap here since it gets completely overwritten
2062 vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap, MSR_IA32_PRED_CMD,
2065 case MSR_IA32_CR_PAT:
2066 if (!kvm_pat_valid(data))
2069 if (is_guest_mode(vcpu) &&
2070 get_vmcs12(vcpu)->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
2071 get_vmcs12(vcpu)->guest_ia32_pat = data;
2073 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2074 vmcs_write64(GUEST_IA32_PAT, data);
2075 vcpu->arch.pat = data;
2078 ret = kvm_set_msr_common(vcpu, msr_info);
2080 case MSR_IA32_TSC_ADJUST:
2081 ret = kvm_set_msr_common(vcpu, msr_info);
2083 case MSR_IA32_MCG_EXT_CTL:
2084 if ((!msr_info->host_initiated &&
2085 !(to_vmx(vcpu)->msr_ia32_feature_control &
2086 FEAT_CTL_LMCE_ENABLED)) ||
2087 (data & ~MCG_EXT_CTL_LMCE_EN))
2089 vcpu->arch.mcg_ext_ctl = data;
2091 case MSR_IA32_FEAT_CTL:
2092 if (!vmx_feature_control_msr_valid(vcpu, data) ||
2093 (to_vmx(vcpu)->msr_ia32_feature_control &
2094 FEAT_CTL_LOCKED && !msr_info->host_initiated))
2096 vmx->msr_ia32_feature_control = data;
2097 if (msr_info->host_initiated && data == 0)
2098 vmx_leave_nested(vcpu);
2100 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2101 if (!msr_info->host_initiated)
2102 return 1; /* they are read-only */
2103 if (!nested_vmx_allowed(vcpu))
2105 return vmx_set_vmx_msr(vcpu, msr_index, data);
2106 case MSR_IA32_RTIT_CTL:
2107 if (!vmx_pt_mode_is_host_guest() ||
2108 vmx_rtit_ctl_check(vcpu, data) ||
2111 vmcs_write64(GUEST_IA32_RTIT_CTL, data);
2112 vmx->pt_desc.guest.ctl = data;
2113 pt_update_intercept_for_msr(vmx);
2115 case MSR_IA32_RTIT_STATUS:
2116 if (!pt_can_write_msr(vmx))
2118 if (data & MSR_IA32_RTIT_STATUS_MASK)
2120 vmx->pt_desc.guest.status = data;
2122 case MSR_IA32_RTIT_CR3_MATCH:
2123 if (!pt_can_write_msr(vmx))
2125 if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2126 PT_CAP_cr3_filtering))
2128 vmx->pt_desc.guest.cr3_match = data;
2130 case MSR_IA32_RTIT_OUTPUT_BASE:
2131 if (!pt_can_write_msr(vmx))
2133 if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2134 PT_CAP_topa_output) &&
2135 !intel_pt_validate_cap(vmx->pt_desc.caps,
2136 PT_CAP_single_range_output))
2138 if (data & MSR_IA32_RTIT_OUTPUT_BASE_MASK)
2140 vmx->pt_desc.guest.output_base = data;
2142 case MSR_IA32_RTIT_OUTPUT_MASK:
2143 if (!pt_can_write_msr(vmx))
2145 if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2146 PT_CAP_topa_output) &&
2147 !intel_pt_validate_cap(vmx->pt_desc.caps,
2148 PT_CAP_single_range_output))
2150 vmx->pt_desc.guest.output_mask = data;
2152 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
2153 if (!pt_can_write_msr(vmx))
2155 index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
2156 if (index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
2157 PT_CAP_num_address_ranges))
2159 if (is_noncanonical_address(data, vcpu))
2162 vmx->pt_desc.guest.addr_b[index / 2] = data;
2164 vmx->pt_desc.guest.addr_a[index / 2] = data;
2167 if (!msr_info->host_initiated &&
2168 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
2170 /* Check reserved bit, higher 32 bits should be zero */
2171 if ((data >> 32) != 0)
2173 goto find_shared_msr;
2177 msr = find_msr_entry(vmx, msr_index);
2179 ret = vmx_set_guest_msr(vmx, msr, data);
2181 ret = kvm_set_msr_common(vcpu, msr_info);
2187 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2189 kvm_register_mark_available(vcpu, reg);
2193 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2196 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2198 case VCPU_EXREG_PDPTR:
2200 ept_save_pdptrs(vcpu);
2202 case VCPU_EXREG_CR3:
2203 if (enable_unrestricted_guest || (enable_ept && is_paging(vcpu)))
2204 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2212 static __init int cpu_has_kvm_support(void)
2214 return cpu_has_vmx();
2217 static __init int vmx_disabled_by_bios(void)
2219 return !boot_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
2220 !boot_cpu_has(X86_FEATURE_VMX);
2223 static int kvm_cpu_vmxon(u64 vmxon_pointer)
2227 cr4_set_bits(X86_CR4_VMXE);
2228 intel_pt_handle_vmx(1);
2230 asm_volatile_goto("1: vmxon %[vmxon_pointer]\n\t"
2231 _ASM_EXTABLE(1b, %l[fault])
2232 : : [vmxon_pointer] "m"(vmxon_pointer)
2237 WARN_ONCE(1, "VMXON faulted, MSR_IA32_FEAT_CTL (0x3a) = 0x%llx\n",
2238 rdmsrl_safe(MSR_IA32_FEAT_CTL, &msr) ? 0xdeadbeef : msr);
2239 intel_pt_handle_vmx(0);
2240 cr4_clear_bits(X86_CR4_VMXE);
2245 static int hardware_enable(void)
2247 int cpu = raw_smp_processor_id();
2248 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2251 if (cr4_read_shadow() & X86_CR4_VMXE)
2255 * This can happen if we hot-added a CPU but failed to allocate
2256 * VP assist page for it.
2258 if (static_branch_unlikely(&enable_evmcs) &&
2259 !hv_get_vp_assist_page(cpu))
2262 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
2263 INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
2264 spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
2266 r = kvm_cpu_vmxon(phys_addr);
2276 static void vmclear_local_loaded_vmcss(void)
2278 int cpu = raw_smp_processor_id();
2279 struct loaded_vmcs *v, *n;
2281 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2282 loaded_vmcss_on_cpu_link)
2283 __loaded_vmcs_clear(v);
2287 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2290 static void kvm_cpu_vmxoff(void)
2292 asm volatile (__ex("vmxoff"));
2294 intel_pt_handle_vmx(0);
2295 cr4_clear_bits(X86_CR4_VMXE);
2298 static void hardware_disable(void)
2300 vmclear_local_loaded_vmcss();
2304 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2305 u32 msr, u32 *result)
2307 u32 vmx_msr_low, vmx_msr_high;
2308 u32 ctl = ctl_min | ctl_opt;
2310 rdmsr(msr, vmx_msr_low, vmx_msr_high);
2312 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2313 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
2315 /* Ensure minimum (required) set of control bits are supported. */
2323 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf,
2324 struct vmx_capability *vmx_cap)
2326 u32 vmx_msr_low, vmx_msr_high;
2327 u32 min, opt, min2, opt2;
2328 u32 _pin_based_exec_control = 0;
2329 u32 _cpu_based_exec_control = 0;
2330 u32 _cpu_based_2nd_exec_control = 0;
2331 u32 _vmexit_control = 0;
2332 u32 _vmentry_control = 0;
2334 memset(vmcs_conf, 0, sizeof(*vmcs_conf));
2335 min = CPU_BASED_HLT_EXITING |
2336 #ifdef CONFIG_X86_64
2337 CPU_BASED_CR8_LOAD_EXITING |
2338 CPU_BASED_CR8_STORE_EXITING |
2340 CPU_BASED_CR3_LOAD_EXITING |
2341 CPU_BASED_CR3_STORE_EXITING |
2342 CPU_BASED_UNCOND_IO_EXITING |
2343 CPU_BASED_MOV_DR_EXITING |
2344 CPU_BASED_USE_TSC_OFFSETTING |
2345 CPU_BASED_MWAIT_EXITING |
2346 CPU_BASED_MONITOR_EXITING |
2347 CPU_BASED_INVLPG_EXITING |
2348 CPU_BASED_RDPMC_EXITING;
2350 opt = CPU_BASED_TPR_SHADOW |
2351 CPU_BASED_USE_MSR_BITMAPS |
2352 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2353 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2354 &_cpu_based_exec_control) < 0)
2356 #ifdef CONFIG_X86_64
2357 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2358 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2359 ~CPU_BASED_CR8_STORE_EXITING;
2361 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2363 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2364 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2365 SECONDARY_EXEC_WBINVD_EXITING |
2366 SECONDARY_EXEC_ENABLE_VPID |
2367 SECONDARY_EXEC_ENABLE_EPT |
2368 SECONDARY_EXEC_UNRESTRICTED_GUEST |
2369 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2370 SECONDARY_EXEC_DESC |
2371 SECONDARY_EXEC_RDTSCP |
2372 SECONDARY_EXEC_ENABLE_INVPCID |
2373 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2374 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2375 SECONDARY_EXEC_SHADOW_VMCS |
2376 SECONDARY_EXEC_XSAVES |
2377 SECONDARY_EXEC_RDSEED_EXITING |
2378 SECONDARY_EXEC_RDRAND_EXITING |
2379 SECONDARY_EXEC_ENABLE_PML |
2380 SECONDARY_EXEC_TSC_SCALING |
2381 SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE |
2382 SECONDARY_EXEC_PT_USE_GPA |
2383 SECONDARY_EXEC_PT_CONCEAL_VMX |
2384 SECONDARY_EXEC_ENABLE_VMFUNC |
2385 SECONDARY_EXEC_ENCLS_EXITING;
2386 if (adjust_vmx_controls(min2, opt2,
2387 MSR_IA32_VMX_PROCBASED_CTLS2,
2388 &_cpu_based_2nd_exec_control) < 0)
2391 #ifndef CONFIG_X86_64
2392 if (!(_cpu_based_2nd_exec_control &
2393 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2394 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2397 if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2398 _cpu_based_2nd_exec_control &= ~(
2399 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2400 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2401 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
2403 rdmsr_safe(MSR_IA32_VMX_EPT_VPID_CAP,
2404 &vmx_cap->ept, &vmx_cap->vpid);
2406 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2407 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2409 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2410 CPU_BASED_CR3_STORE_EXITING |
2411 CPU_BASED_INVLPG_EXITING);
2412 } else if (vmx_cap->ept) {
2414 pr_warn_once("EPT CAP should not exist if not support "
2415 "1-setting enable EPT VM-execution control\n");
2417 if (!(_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_VPID) &&
2420 pr_warn_once("VPID CAP should not exist if not support "
2421 "1-setting enable VPID VM-execution control\n");
2424 min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
2425 #ifdef CONFIG_X86_64
2426 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2428 opt = VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL |
2429 VM_EXIT_LOAD_IA32_PAT |
2430 VM_EXIT_LOAD_IA32_EFER |
2431 VM_EXIT_CLEAR_BNDCFGS |
2432 VM_EXIT_PT_CONCEAL_PIP |
2433 VM_EXIT_CLEAR_IA32_RTIT_CTL;
2434 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2435 &_vmexit_control) < 0)
2438 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2439 opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR |
2440 PIN_BASED_VMX_PREEMPTION_TIMER;
2441 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2442 &_pin_based_exec_control) < 0)
2445 if (cpu_has_broken_vmx_preemption_timer())
2446 _pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
2447 if (!(_cpu_based_2nd_exec_control &
2448 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
2449 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
2451 min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
2452 opt = VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL |
2453 VM_ENTRY_LOAD_IA32_PAT |
2454 VM_ENTRY_LOAD_IA32_EFER |
2455 VM_ENTRY_LOAD_BNDCFGS |
2456 VM_ENTRY_PT_CONCEAL_PIP |
2457 VM_ENTRY_LOAD_IA32_RTIT_CTL;
2458 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2459 &_vmentry_control) < 0)
2463 * Some cpus support VM_{ENTRY,EXIT}_IA32_PERF_GLOBAL_CTRL but they
2464 * can't be used due to an errata where VM Exit may incorrectly clear
2465 * IA32_PERF_GLOBAL_CTRL[34:32]. Workaround the errata by using the
2466 * MSR load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2468 if (boot_cpu_data.x86 == 0x6) {
2469 switch (boot_cpu_data.x86_model) {
2470 case 26: /* AAK155 */
2471 case 30: /* AAP115 */
2472 case 37: /* AAT100 */
2473 case 44: /* BC86,AAY89,BD102 */
2475 _vmentry_control &= ~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
2476 _vmexit_control &= ~VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
2477 pr_warn_once("kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
2478 "does not work properly. Using workaround\n");
2486 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2488 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2489 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2492 #ifdef CONFIG_X86_64
2493 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2494 if (vmx_msr_high & (1u<<16))
2498 /* Require Write-Back (WB) memory type for VMCS accesses. */
2499 if (((vmx_msr_high >> 18) & 15) != 6)
2502 vmcs_conf->size = vmx_msr_high & 0x1fff;
2503 vmcs_conf->order = get_order(vmcs_conf->size);
2504 vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
2506 vmcs_conf->revision_id = vmx_msr_low;
2508 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2509 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2510 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2511 vmcs_conf->vmexit_ctrl = _vmexit_control;
2512 vmcs_conf->vmentry_ctrl = _vmentry_control;
2514 if (static_branch_unlikely(&enable_evmcs))
2515 evmcs_sanitize_exec_ctrls(vmcs_conf);
2520 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags)
2522 int node = cpu_to_node(cpu);
2526 pages = __alloc_pages_node(node, flags, vmcs_config.order);
2529 vmcs = page_address(pages);
2530 memset(vmcs, 0, vmcs_config.size);
2532 /* KVM supports Enlightened VMCS v1 only */
2533 if (static_branch_unlikely(&enable_evmcs))
2534 vmcs->hdr.revision_id = KVM_EVMCS_VERSION;
2536 vmcs->hdr.revision_id = vmcs_config.revision_id;
2539 vmcs->hdr.shadow_vmcs = 1;
2543 void free_vmcs(struct vmcs *vmcs)
2545 free_pages((unsigned long)vmcs, vmcs_config.order);
2549 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2551 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2553 if (!loaded_vmcs->vmcs)
2555 loaded_vmcs_clear(loaded_vmcs);
2556 free_vmcs(loaded_vmcs->vmcs);
2557 loaded_vmcs->vmcs = NULL;
2558 if (loaded_vmcs->msr_bitmap)
2559 free_page((unsigned long)loaded_vmcs->msr_bitmap);
2560 WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
2563 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2565 loaded_vmcs->vmcs = alloc_vmcs(false);
2566 if (!loaded_vmcs->vmcs)
2569 vmcs_clear(loaded_vmcs->vmcs);
2571 loaded_vmcs->shadow_vmcs = NULL;
2572 loaded_vmcs->hv_timer_soft_disabled = false;
2573 loaded_vmcs->cpu = -1;
2574 loaded_vmcs->launched = 0;
2576 if (cpu_has_vmx_msr_bitmap()) {
2577 loaded_vmcs->msr_bitmap = (unsigned long *)
2578 __get_free_page(GFP_KERNEL_ACCOUNT);
2579 if (!loaded_vmcs->msr_bitmap)
2581 memset(loaded_vmcs->msr_bitmap, 0xff, PAGE_SIZE);
2583 if (IS_ENABLED(CONFIG_HYPERV) &&
2584 static_branch_unlikely(&enable_evmcs) &&
2585 (ms_hyperv.nested_features & HV_X64_NESTED_MSR_BITMAP)) {
2586 struct hv_enlightened_vmcs *evmcs =
2587 (struct hv_enlightened_vmcs *)loaded_vmcs->vmcs;
2589 evmcs->hv_enlightenments_control.msr_bitmap = 1;
2593 memset(&loaded_vmcs->host_state, 0, sizeof(struct vmcs_host_state));
2594 memset(&loaded_vmcs->controls_shadow, 0,
2595 sizeof(struct vmcs_controls_shadow));
2600 free_loaded_vmcs(loaded_vmcs);
2604 static void free_kvm_area(void)
2608 for_each_possible_cpu(cpu) {
2609 free_vmcs(per_cpu(vmxarea, cpu));
2610 per_cpu(vmxarea, cpu) = NULL;
2614 static __init int alloc_kvm_area(void)
2618 for_each_possible_cpu(cpu) {
2621 vmcs = alloc_vmcs_cpu(false, cpu, GFP_KERNEL);
2628 * When eVMCS is enabled, alloc_vmcs_cpu() sets
2629 * vmcs->revision_id to KVM_EVMCS_VERSION instead of
2630 * revision_id reported by MSR_IA32_VMX_BASIC.
2632 * However, even though not explicitly documented by
2633 * TLFS, VMXArea passed as VMXON argument should
2634 * still be marked with revision_id reported by
2637 if (static_branch_unlikely(&enable_evmcs))
2638 vmcs->hdr.revision_id = vmcs_config.revision_id;
2640 per_cpu(vmxarea, cpu) = vmcs;
2645 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
2646 struct kvm_segment *save)
2648 if (!emulate_invalid_guest_state) {
2650 * CS and SS RPL should be equal during guest entry according
2651 * to VMX spec, but in reality it is not always so. Since vcpu
2652 * is in the middle of the transition from real mode to
2653 * protected mode it is safe to assume that RPL 0 is a good
2656 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
2657 save->selector &= ~SEGMENT_RPL_MASK;
2658 save->dpl = save->selector & SEGMENT_RPL_MASK;
2661 vmx_set_segment(vcpu, save, seg);
2664 static void enter_pmode(struct kvm_vcpu *vcpu)
2666 unsigned long flags;
2667 struct vcpu_vmx *vmx = to_vmx(vcpu);
2670 * Update real mode segment cache. It may be not up-to-date if sement
2671 * register was written while vcpu was in a guest mode.
2673 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2674 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2675 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2676 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2677 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2678 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2680 vmx->rmode.vm86_active = 0;
2682 vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2684 flags = vmcs_readl(GUEST_RFLAGS);
2685 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2686 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2687 vmcs_writel(GUEST_RFLAGS, flags);
2689 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
2690 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
2692 update_exception_bitmap(vcpu);
2694 fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2695 fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2696 fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2697 fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2698 fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2699 fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2702 static void fix_rmode_seg(int seg, struct kvm_segment *save)
2704 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2705 struct kvm_segment var = *save;
2708 if (seg == VCPU_SREG_CS)
2711 if (!emulate_invalid_guest_state) {
2712 var.selector = var.base >> 4;
2713 var.base = var.base & 0xffff0;
2723 if (save->base & 0xf)
2724 printk_once(KERN_WARNING "kvm: segment base is not "
2725 "paragraph aligned when entering "
2726 "protected mode (seg=%d)", seg);
2729 vmcs_write16(sf->selector, var.selector);
2730 vmcs_writel(sf->base, var.base);
2731 vmcs_write32(sf->limit, var.limit);
2732 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
2735 static void enter_rmode(struct kvm_vcpu *vcpu)
2737 unsigned long flags;
2738 struct vcpu_vmx *vmx = to_vmx(vcpu);
2739 struct kvm_vmx *kvm_vmx = to_kvm_vmx(vcpu->kvm);
2741 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2742 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2743 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2744 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2745 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2746 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2747 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2749 vmx->rmode.vm86_active = 1;
2752 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
2753 * vcpu. Warn the user that an update is overdue.
2755 if (!kvm_vmx->tss_addr)
2756 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
2757 "called before entering vcpu\n");
2759 vmx_segment_cache_clear(vmx);
2761 vmcs_writel(GUEST_TR_BASE, kvm_vmx->tss_addr);
2762 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
2763 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2765 flags = vmcs_readl(GUEST_RFLAGS);
2766 vmx->rmode.save_rflags = flags;
2768 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2770 vmcs_writel(GUEST_RFLAGS, flags);
2771 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
2772 update_exception_bitmap(vcpu);
2774 fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2775 fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2776 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2777 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2778 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2779 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2781 kvm_mmu_reset_context(vcpu);
2784 void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
2786 struct vcpu_vmx *vmx = to_vmx(vcpu);
2787 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
2792 vcpu->arch.efer = efer;
2793 if (efer & EFER_LMA) {
2794 vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2797 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2799 msr->data = efer & ~EFER_LME;
2804 #ifdef CONFIG_X86_64
2806 static void enter_lmode(struct kvm_vcpu *vcpu)
2810 vmx_segment_cache_clear(to_vmx(vcpu));
2812 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
2813 if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
2814 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
2816 vmcs_write32(GUEST_TR_AR_BYTES,
2817 (guest_tr_ar & ~VMX_AR_TYPE_MASK)
2818 | VMX_AR_TYPE_BUSY_64_TSS);
2820 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
2823 static void exit_lmode(struct kvm_vcpu *vcpu)
2825 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2826 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
2831 static void vmx_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t addr)
2833 int vpid = to_vmx(vcpu)->vpid;
2835 if (!vpid_sync_vcpu_addr(vpid, addr))
2836 vpid_sync_context(vpid);
2839 * If VPIDs are not supported or enabled, then the above is a no-op.
2840 * But we don't really need a TLB flush in that case anyway, because
2841 * each VM entry/exit includes an implicit flush when VPID is 0.
2845 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
2847 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
2849 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
2850 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
2853 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
2855 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
2857 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
2858 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
2861 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
2863 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2865 if (!kvm_register_is_dirty(vcpu, VCPU_EXREG_PDPTR))
2868 if (is_pae_paging(vcpu)) {
2869 vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
2870 vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
2871 vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
2872 vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
2876 void ept_save_pdptrs(struct kvm_vcpu *vcpu)
2878 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2880 if (is_pae_paging(vcpu)) {
2881 mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
2882 mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
2883 mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
2884 mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
2887 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
2890 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
2892 struct kvm_vcpu *vcpu)
2894 struct vcpu_vmx *vmx = to_vmx(vcpu);
2896 if (!kvm_register_is_available(vcpu, VCPU_EXREG_CR3))
2897 vmx_cache_reg(vcpu, VCPU_EXREG_CR3);
2898 if (!(cr0 & X86_CR0_PG)) {
2899 /* From paging/starting to nonpaging */
2900 exec_controls_setbit(vmx, CPU_BASED_CR3_LOAD_EXITING |
2901 CPU_BASED_CR3_STORE_EXITING);
2902 vcpu->arch.cr0 = cr0;
2903 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2904 } else if (!is_paging(vcpu)) {
2905 /* From nonpaging to paging */
2906 exec_controls_clearbit(vmx, CPU_BASED_CR3_LOAD_EXITING |
2907 CPU_BASED_CR3_STORE_EXITING);
2908 vcpu->arch.cr0 = cr0;
2909 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2912 if (!(cr0 & X86_CR0_WP))
2913 *hw_cr0 &= ~X86_CR0_WP;
2916 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
2918 struct vcpu_vmx *vmx = to_vmx(vcpu);
2919 unsigned long hw_cr0;
2921 hw_cr0 = (cr0 & ~KVM_VM_CR0_ALWAYS_OFF);
2922 if (enable_unrestricted_guest)
2923 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
2925 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
2927 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
2930 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
2934 #ifdef CONFIG_X86_64
2935 if (vcpu->arch.efer & EFER_LME) {
2936 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
2938 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
2943 if (enable_ept && !enable_unrestricted_guest)
2944 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
2946 vmcs_writel(CR0_READ_SHADOW, cr0);
2947 vmcs_writel(GUEST_CR0, hw_cr0);
2948 vcpu->arch.cr0 = cr0;
2950 /* depends on vcpu->arch.cr0 to be set to a new value */
2951 vmx->emulation_required = emulation_required(vcpu);
2954 static int get_ept_level(struct kvm_vcpu *vcpu)
2956 if (is_guest_mode(vcpu) && nested_cpu_has_ept(get_vmcs12(vcpu)))
2957 return vmx_eptp_page_walk_level(nested_ept_get_eptp(vcpu));
2958 if (cpu_has_vmx_ept_5levels() && (cpuid_maxphyaddr(vcpu) > 48))
2963 u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa)
2965 u64 eptp = VMX_EPTP_MT_WB;
2967 eptp |= (get_ept_level(vcpu) == 5) ? VMX_EPTP_PWL_5 : VMX_EPTP_PWL_4;
2969 if (enable_ept_ad_bits &&
2970 (!is_guest_mode(vcpu) || nested_ept_ad_enabled(vcpu)))
2971 eptp |= VMX_EPTP_AD_ENABLE_BIT;
2972 eptp |= (root_hpa & PAGE_MASK);
2977 void vmx_load_mmu_pgd(struct kvm_vcpu *vcpu, unsigned long cr3)
2979 struct kvm *kvm = vcpu->kvm;
2980 bool update_guest_cr3 = true;
2981 unsigned long guest_cr3;
2986 eptp = construct_eptp(vcpu, cr3);
2987 vmcs_write64(EPT_POINTER, eptp);
2989 if (kvm_x86_ops->tlb_remote_flush) {
2990 spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
2991 to_vmx(vcpu)->ept_pointer = eptp;
2992 to_kvm_vmx(kvm)->ept_pointers_match
2993 = EPT_POINTERS_CHECK;
2994 spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
2997 /* Loading vmcs02.GUEST_CR3 is handled by nested VM-Enter. */
2998 if (is_guest_mode(vcpu))
2999 update_guest_cr3 = false;
3000 else if (!enable_unrestricted_guest && !is_paging(vcpu))
3001 guest_cr3 = to_kvm_vmx(kvm)->ept_identity_map_addr;
3002 else if (test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
3003 guest_cr3 = vcpu->arch.cr3;
3004 else /* vmcs01.GUEST_CR3 is already up-to-date. */
3005 update_guest_cr3 = false;
3006 ept_load_pdptrs(vcpu);
3009 if (update_guest_cr3)
3010 vmcs_writel(GUEST_CR3, guest_cr3);
3013 int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3015 struct vcpu_vmx *vmx = to_vmx(vcpu);
3017 * Pass through host's Machine Check Enable value to hw_cr4, which
3018 * is in force while we are in guest mode. Do not let guests control
3019 * this bit, even if host CR4.MCE == 0.
3021 unsigned long hw_cr4;
3023 hw_cr4 = (cr4_read_shadow() & X86_CR4_MCE) | (cr4 & ~X86_CR4_MCE);
3024 if (enable_unrestricted_guest)
3025 hw_cr4 |= KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST;
3026 else if (vmx->rmode.vm86_active)
3027 hw_cr4 |= KVM_RMODE_VM_CR4_ALWAYS_ON;
3029 hw_cr4 |= KVM_PMODE_VM_CR4_ALWAYS_ON;
3031 if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated()) {
3032 if (cr4 & X86_CR4_UMIP) {
3033 secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_DESC);
3034 hw_cr4 &= ~X86_CR4_UMIP;
3035 } else if (!is_guest_mode(vcpu) ||
3036 !nested_cpu_has2(get_vmcs12(vcpu), SECONDARY_EXEC_DESC)) {
3037 secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_DESC);
3041 if (cr4 & X86_CR4_VMXE) {
3043 * To use VMXON (and later other VMX instructions), a guest
3044 * must first be able to turn on cr4.VMXE (see handle_vmon()).
3045 * So basically the check on whether to allow nested VMX
3046 * is here. We operate under the default treatment of SMM,
3047 * so VMX cannot be enabled under SMM.
3049 if (!nested_vmx_allowed(vcpu) || is_smm(vcpu))
3053 if (vmx->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
3056 vcpu->arch.cr4 = cr4;
3058 if (!enable_unrestricted_guest) {
3060 if (!is_paging(vcpu)) {
3061 hw_cr4 &= ~X86_CR4_PAE;
3062 hw_cr4 |= X86_CR4_PSE;
3063 } else if (!(cr4 & X86_CR4_PAE)) {
3064 hw_cr4 &= ~X86_CR4_PAE;
3069 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
3070 * hardware. To emulate this behavior, SMEP/SMAP/PKU needs
3071 * to be manually disabled when guest switches to non-paging
3074 * If !enable_unrestricted_guest, the CPU is always running
3075 * with CR0.PG=1 and CR4 needs to be modified.
3076 * If enable_unrestricted_guest, the CPU automatically
3077 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
3079 if (!is_paging(vcpu))
3080 hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
3083 vmcs_writel(CR4_READ_SHADOW, cr4);
3084 vmcs_writel(GUEST_CR4, hw_cr4);
3088 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3090 struct vcpu_vmx *vmx = to_vmx(vcpu);
3093 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3094 *var = vmx->rmode.segs[seg];
3095 if (seg == VCPU_SREG_TR
3096 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3098 var->base = vmx_read_guest_seg_base(vmx, seg);
3099 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3102 var->base = vmx_read_guest_seg_base(vmx, seg);
3103 var->limit = vmx_read_guest_seg_limit(vmx, seg);
3104 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3105 ar = vmx_read_guest_seg_ar(vmx, seg);
3106 var->unusable = (ar >> 16) & 1;
3107 var->type = ar & 15;
3108 var->s = (ar >> 4) & 1;
3109 var->dpl = (ar >> 5) & 3;
3111 * Some userspaces do not preserve unusable property. Since usable
3112 * segment has to be present according to VMX spec we can use present
3113 * property to amend userspace bug by making unusable segment always
3114 * nonpresent. vmx_segment_access_rights() already marks nonpresent
3115 * segment as unusable.
3117 var->present = !var->unusable;
3118 var->avl = (ar >> 12) & 1;
3119 var->l = (ar >> 13) & 1;
3120 var->db = (ar >> 14) & 1;
3121 var->g = (ar >> 15) & 1;
3124 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3126 struct kvm_segment s;
3128 if (to_vmx(vcpu)->rmode.vm86_active) {
3129 vmx_get_segment(vcpu, &s, seg);
3132 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3135 int vmx_get_cpl(struct kvm_vcpu *vcpu)
3137 struct vcpu_vmx *vmx = to_vmx(vcpu);
3139 if (unlikely(vmx->rmode.vm86_active))
3142 int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
3143 return VMX_AR_DPL(ar);
3147 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3151 if (var->unusable || !var->present)
3154 ar = var->type & 15;
3155 ar |= (var->s & 1) << 4;
3156 ar |= (var->dpl & 3) << 5;
3157 ar |= (var->present & 1) << 7;
3158 ar |= (var->avl & 1) << 12;
3159 ar |= (var->l & 1) << 13;
3160 ar |= (var->db & 1) << 14;
3161 ar |= (var->g & 1) << 15;
3167 void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3169 struct vcpu_vmx *vmx = to_vmx(vcpu);
3170 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3172 vmx_segment_cache_clear(vmx);
3174 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3175 vmx->rmode.segs[seg] = *var;
3176 if (seg == VCPU_SREG_TR)
3177 vmcs_write16(sf->selector, var->selector);
3179 fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3183 vmcs_writel(sf->base, var->base);
3184 vmcs_write32(sf->limit, var->limit);
3185 vmcs_write16(sf->selector, var->selector);
3188 * Fix the "Accessed" bit in AR field of segment registers for older
3190 * IA32 arch specifies that at the time of processor reset the
3191 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3192 * is setting it to 0 in the userland code. This causes invalid guest
3193 * state vmexit when "unrestricted guest" mode is turned on.
3194 * Fix for this setup issue in cpu_reset is being pushed in the qemu
3195 * tree. Newer qemu binaries with that qemu fix would not need this
3198 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3199 var->type |= 0x1; /* Accessed */
3201 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3204 vmx->emulation_required = emulation_required(vcpu);
3207 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3209 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3211 *db = (ar >> 14) & 1;
3212 *l = (ar >> 13) & 1;
3215 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3217 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3218 dt->address = vmcs_readl(GUEST_IDTR_BASE);
3221 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3223 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3224 vmcs_writel(GUEST_IDTR_BASE, dt->address);
3227 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3229 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3230 dt->address = vmcs_readl(GUEST_GDTR_BASE);
3233 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3235 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3236 vmcs_writel(GUEST_GDTR_BASE, dt->address);
3239 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3241 struct kvm_segment var;
3244 vmx_get_segment(vcpu, &var, seg);
3246 if (seg == VCPU_SREG_CS)
3248 ar = vmx_segment_access_rights(&var);
3250 if (var.base != (var.selector << 4))
3252 if (var.limit != 0xffff)
3260 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3262 struct kvm_segment cs;
3263 unsigned int cs_rpl;
3265 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3266 cs_rpl = cs.selector & SEGMENT_RPL_MASK;
3270 if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
3274 if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
3275 if (cs.dpl > cs_rpl)
3278 if (cs.dpl != cs_rpl)
3284 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3288 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3290 struct kvm_segment ss;
3291 unsigned int ss_rpl;
3293 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3294 ss_rpl = ss.selector & SEGMENT_RPL_MASK;
3298 if (ss.type != 3 && ss.type != 7)
3302 if (ss.dpl != ss_rpl) /* DPL != RPL */
3310 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3312 struct kvm_segment var;
3315 vmx_get_segment(vcpu, &var, seg);
3316 rpl = var.selector & SEGMENT_RPL_MASK;
3324 if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
3325 if (var.dpl < rpl) /* DPL < RPL */
3329 /* TODO: Add other members to kvm_segment_field to allow checking for other access
3335 static bool tr_valid(struct kvm_vcpu *vcpu)
3337 struct kvm_segment tr;
3339 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3343 if (tr.selector & SEGMENT_TI_MASK) /* TI = 1 */
3345 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3353 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3355 struct kvm_segment ldtr;
3357 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3361 if (ldtr.selector & SEGMENT_TI_MASK) /* TI = 1 */
3371 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3373 struct kvm_segment cs, ss;
3375 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3376 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3378 return ((cs.selector & SEGMENT_RPL_MASK) ==
3379 (ss.selector & SEGMENT_RPL_MASK));
3383 * Check if guest state is valid. Returns true if valid, false if
3385 * We assume that registers are always usable
3387 static bool guest_state_valid(struct kvm_vcpu *vcpu)
3389 if (enable_unrestricted_guest)
3392 /* real mode guest state checks */
3393 if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
3394 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3396 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3398 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3400 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3402 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3404 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3407 /* protected mode guest state checks */
3408 if (!cs_ss_rpl_check(vcpu))
3410 if (!code_segment_valid(vcpu))
3412 if (!stack_segment_valid(vcpu))
3414 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3416 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3418 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3420 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3422 if (!tr_valid(vcpu))
3424 if (!ldtr_valid(vcpu))
3428 * - Add checks on RIP
3429 * - Add checks on RFLAGS
3435 static int init_rmode_tss(struct kvm *kvm)
3441 idx = srcu_read_lock(&kvm->srcu);
3442 fn = to_kvm_vmx(kvm)->tss_addr >> PAGE_SHIFT;
3443 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3446 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3447 r = kvm_write_guest_page(kvm, fn++, &data,
3448 TSS_IOPB_BASE_OFFSET, sizeof(u16));
3451 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3454 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3458 r = kvm_write_guest_page(kvm, fn, &data,
3459 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3462 srcu_read_unlock(&kvm->srcu, idx);
3466 static int init_rmode_identity_map(struct kvm *kvm)
3468 struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm);
3470 kvm_pfn_t identity_map_pfn;
3473 /* Protect kvm_vmx->ept_identity_pagetable_done. */
3474 mutex_lock(&kvm->slots_lock);
3476 if (likely(kvm_vmx->ept_identity_pagetable_done))
3479 if (!kvm_vmx->ept_identity_map_addr)
3480 kvm_vmx->ept_identity_map_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
3481 identity_map_pfn = kvm_vmx->ept_identity_map_addr >> PAGE_SHIFT;
3483 r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
3484 kvm_vmx->ept_identity_map_addr, PAGE_SIZE);
3488 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3491 /* Set up identity-mapping pagetable for EPT in real mode */
3492 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3493 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3494 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3495 r = kvm_write_guest_page(kvm, identity_map_pfn,
3496 &tmp, i * sizeof(tmp), sizeof(tmp));
3500 kvm_vmx->ept_identity_pagetable_done = true;
3503 mutex_unlock(&kvm->slots_lock);
3507 static void seg_setup(int seg)
3509 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3512 vmcs_write16(sf->selector, 0);
3513 vmcs_writel(sf->base, 0);
3514 vmcs_write32(sf->limit, 0xffff);
3516 if (seg == VCPU_SREG_CS)
3517 ar |= 0x08; /* code segment */
3519 vmcs_write32(sf->ar_bytes, ar);
3522 static int alloc_apic_access_page(struct kvm *kvm)
3527 mutex_lock(&kvm->slots_lock);
3528 if (kvm->arch.apic_access_page_done)
3530 r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
3531 APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
3535 page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
3536 if (is_error_page(page)) {
3542 * Do not pin the page in memory, so that memory hot-unplug
3543 * is able to migrate it.
3546 kvm->arch.apic_access_page_done = true;
3548 mutex_unlock(&kvm->slots_lock);
3552 int allocate_vpid(void)
3558 spin_lock(&vmx_vpid_lock);
3559 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3560 if (vpid < VMX_NR_VPIDS)
3561 __set_bit(vpid, vmx_vpid_bitmap);
3564 spin_unlock(&vmx_vpid_lock);
3568 void free_vpid(int vpid)
3570 if (!enable_vpid || vpid == 0)
3572 spin_lock(&vmx_vpid_lock);
3573 __clear_bit(vpid, vmx_vpid_bitmap);
3574 spin_unlock(&vmx_vpid_lock);
3577 static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
3580 int f = sizeof(unsigned long);
3582 if (!cpu_has_vmx_msr_bitmap())
3585 if (static_branch_unlikely(&enable_evmcs))
3586 evmcs_touch_msr_bitmap();
3589 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3590 * have the write-low and read-high bitmap offsets the wrong way round.
3591 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3593 if (msr <= 0x1fff) {
3594 if (type & MSR_TYPE_R)
3596 __clear_bit(msr, msr_bitmap + 0x000 / f);
3598 if (type & MSR_TYPE_W)
3600 __clear_bit(msr, msr_bitmap + 0x800 / f);
3602 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3604 if (type & MSR_TYPE_R)
3606 __clear_bit(msr, msr_bitmap + 0x400 / f);
3608 if (type & MSR_TYPE_W)
3610 __clear_bit(msr, msr_bitmap + 0xc00 / f);
3615 static __always_inline void vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
3618 int f = sizeof(unsigned long);
3620 if (!cpu_has_vmx_msr_bitmap())
3623 if (static_branch_unlikely(&enable_evmcs))
3624 evmcs_touch_msr_bitmap();
3627 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3628 * have the write-low and read-high bitmap offsets the wrong way round.
3629 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3631 if (msr <= 0x1fff) {
3632 if (type & MSR_TYPE_R)
3634 __set_bit(msr, msr_bitmap + 0x000 / f);
3636 if (type & MSR_TYPE_W)
3638 __set_bit(msr, msr_bitmap + 0x800 / f);
3640 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3642 if (type & MSR_TYPE_R)
3644 __set_bit(msr, msr_bitmap + 0x400 / f);
3646 if (type & MSR_TYPE_W)
3648 __set_bit(msr, msr_bitmap + 0xc00 / f);
3653 static __always_inline void vmx_set_intercept_for_msr(unsigned long *msr_bitmap,
3654 u32 msr, int type, bool value)
3657 vmx_enable_intercept_for_msr(msr_bitmap, msr, type);
3659 vmx_disable_intercept_for_msr(msr_bitmap, msr, type);
3662 static u8 vmx_msr_bitmap_mode(struct kvm_vcpu *vcpu)
3666 if (cpu_has_secondary_exec_ctrls() &&
3667 (secondary_exec_controls_get(to_vmx(vcpu)) &
3668 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
3669 mode |= MSR_BITMAP_MODE_X2APIC;
3670 if (enable_apicv && kvm_vcpu_apicv_active(vcpu))
3671 mode |= MSR_BITMAP_MODE_X2APIC_APICV;
3677 static void vmx_update_msr_bitmap_x2apic(unsigned long *msr_bitmap,
3682 for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
3683 unsigned word = msr / BITS_PER_LONG;
3684 msr_bitmap[word] = (mode & MSR_BITMAP_MODE_X2APIC_APICV) ? 0 : ~0;
3685 msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
3688 if (mode & MSR_BITMAP_MODE_X2APIC) {
3690 * TPR reads and writes can be virtualized even if virtual interrupt
3691 * delivery is not in use.
3693 vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TASKPRI), MSR_TYPE_RW);
3694 if (mode & MSR_BITMAP_MODE_X2APIC_APICV) {
3695 vmx_enable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TMCCT), MSR_TYPE_R);
3696 vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_EOI), MSR_TYPE_W);
3697 vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_SELF_IPI), MSR_TYPE_W);
3702 void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu)
3704 struct vcpu_vmx *vmx = to_vmx(vcpu);
3705 unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3706 u8 mode = vmx_msr_bitmap_mode(vcpu);
3707 u8 changed = mode ^ vmx->msr_bitmap_mode;
3712 if (changed & (MSR_BITMAP_MODE_X2APIC | MSR_BITMAP_MODE_X2APIC_APICV))
3713 vmx_update_msr_bitmap_x2apic(msr_bitmap, mode);
3715 vmx->msr_bitmap_mode = mode;
3718 void pt_update_intercept_for_msr(struct vcpu_vmx *vmx)
3720 unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3721 bool flag = !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
3724 vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_STATUS,
3726 vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_BASE,
3728 vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_MASK,
3730 vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_CR3_MATCH,
3732 for (i = 0; i < vmx->pt_desc.addr_range; i++) {
3733 vmx_set_intercept_for_msr(msr_bitmap,
3734 MSR_IA32_RTIT_ADDR0_A + i * 2, MSR_TYPE_RW, flag);
3735 vmx_set_intercept_for_msr(msr_bitmap,
3736 MSR_IA32_RTIT_ADDR0_B + i * 2, MSR_TYPE_RW, flag);
3740 static bool vmx_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
3742 struct vcpu_vmx *vmx = to_vmx(vcpu);
3747 if (WARN_ON_ONCE(!is_guest_mode(vcpu)) ||
3748 !nested_cpu_has_vid(get_vmcs12(vcpu)) ||
3749 WARN_ON_ONCE(!vmx->nested.virtual_apic_map.gfn))
3752 rvi = vmx_get_rvi();
3754 vapic_page = vmx->nested.virtual_apic_map.hva;
3755 vppr = *((u32 *)(vapic_page + APIC_PROCPRI));
3757 return ((rvi & 0xf0) > (vppr & 0xf0));
3760 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu,
3764 int pi_vec = nested ? POSTED_INTR_NESTED_VECTOR : POSTED_INTR_VECTOR;
3766 if (vcpu->mode == IN_GUEST_MODE) {
3768 * The vector of interrupt to be delivered to vcpu had
3769 * been set in PIR before this function.
3771 * Following cases will be reached in this block, and
3772 * we always send a notification event in all cases as
3775 * Case 1: vcpu keeps in non-root mode. Sending a
3776 * notification event posts the interrupt to vcpu.
3778 * Case 2: vcpu exits to root mode and is still
3779 * runnable. PIR will be synced to vIRR before the
3780 * next vcpu entry. Sending a notification event in
3781 * this case has no effect, as vcpu is not in root
3784 * Case 3: vcpu exits to root mode and is blocked.
3785 * vcpu_block() has already synced PIR to vIRR and
3786 * never blocks vcpu if vIRR is not cleared. Therefore,
3787 * a blocked vcpu here does not wait for any requested
3788 * interrupts in PIR, and sending a notification event
3789 * which has no effect is safe here.
3792 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu), pi_vec);
3799 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
3802 struct vcpu_vmx *vmx = to_vmx(vcpu);
3804 if (is_guest_mode(vcpu) &&
3805 vector == vmx->nested.posted_intr_nv) {
3807 * If a posted intr is not recognized by hardware,
3808 * we will accomplish it in the next vmentry.
3810 vmx->nested.pi_pending = true;
3811 kvm_make_request(KVM_REQ_EVENT, vcpu);
3812 /* the PIR and ON have been set by L1. */
3813 if (!kvm_vcpu_trigger_posted_interrupt(vcpu, true))
3814 kvm_vcpu_kick(vcpu);
3820 * Send interrupt to vcpu via posted interrupt way.
3821 * 1. If target vcpu is running(non-root mode), send posted interrupt
3822 * notification to vcpu and hardware will sync PIR to vIRR atomically.
3823 * 2. If target vcpu isn't running(root mode), kick it to pick up the
3824 * interrupt from PIR in next vmentry.
3826 static int vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
3828 struct vcpu_vmx *vmx = to_vmx(vcpu);
3831 r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
3835 if (!vcpu->arch.apicv_active)
3838 if (pi_test_and_set_pir(vector, &vmx->pi_desc))
3841 /* If a previous notification has sent the IPI, nothing to do. */
3842 if (pi_test_and_set_on(&vmx->pi_desc))
3845 if (!kvm_vcpu_trigger_posted_interrupt(vcpu, false))
3846 kvm_vcpu_kick(vcpu);
3852 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
3853 * will not change in the lifetime of the guest.
3854 * Note that host-state that does change is set elsewhere. E.g., host-state
3855 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
3857 void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
3861 unsigned long cr0, cr3, cr4;
3864 WARN_ON(cr0 & X86_CR0_TS);
3865 vmcs_writel(HOST_CR0, cr0); /* 22.2.3 */
3868 * Save the most likely value for this task's CR3 in the VMCS.
3869 * We can't use __get_current_cr3_fast() because we're not atomic.
3872 vmcs_writel(HOST_CR3, cr3); /* 22.2.3 FIXME: shadow tables */
3873 vmx->loaded_vmcs->host_state.cr3 = cr3;
3875 /* Save the most likely value for this task's CR4 in the VMCS. */
3876 cr4 = cr4_read_shadow();
3877 vmcs_writel(HOST_CR4, cr4); /* 22.2.3, 22.2.5 */
3878 vmx->loaded_vmcs->host_state.cr4 = cr4;
3880 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
3881 #ifdef CONFIG_X86_64
3883 * Load null selectors, so we can avoid reloading them in
3884 * vmx_prepare_switch_to_host(), in case userspace uses
3885 * the null selectors too (the expected case).
3887 vmcs_write16(HOST_DS_SELECTOR, 0);
3888 vmcs_write16(HOST_ES_SELECTOR, 0);
3890 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
3891 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
3893 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
3894 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
3896 vmcs_writel(HOST_IDTR_BASE, host_idt_base); /* 22.2.4 */
3898 vmcs_writel(HOST_RIP, (unsigned long)vmx_vmexit); /* 22.2.5 */
3900 rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
3901 vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
3902 rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
3903 vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */
3905 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
3906 rdmsr(MSR_IA32_CR_PAT, low32, high32);
3907 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
3910 if (cpu_has_load_ia32_efer())
3911 vmcs_write64(HOST_IA32_EFER, host_efer);
3914 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
3916 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
3918 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
3919 if (is_guest_mode(&vmx->vcpu))
3920 vmx->vcpu.arch.cr4_guest_owned_bits &=
3921 ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
3922 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
3925 u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
3927 u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
3929 if (!kvm_vcpu_apicv_active(&vmx->vcpu))
3930 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
3933 pin_based_exec_ctrl &= ~PIN_BASED_VIRTUAL_NMIS;
3935 if (!enable_preemption_timer)
3936 pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
3938 return pin_based_exec_ctrl;
3941 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
3943 struct vcpu_vmx *vmx = to_vmx(vcpu);
3945 pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
3946 if (cpu_has_secondary_exec_ctrls()) {
3947 if (kvm_vcpu_apicv_active(vcpu))
3948 secondary_exec_controls_setbit(vmx,
3949 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3950 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3952 secondary_exec_controls_clearbit(vmx,
3953 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3954 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3957 if (cpu_has_vmx_msr_bitmap())
3958 vmx_update_msr_bitmap(vcpu);
3961 u32 vmx_exec_control(struct vcpu_vmx *vmx)
3963 u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
3965 if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
3966 exec_control &= ~CPU_BASED_MOV_DR_EXITING;
3968 if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
3969 exec_control &= ~CPU_BASED_TPR_SHADOW;
3970 #ifdef CONFIG_X86_64
3971 exec_control |= CPU_BASED_CR8_STORE_EXITING |
3972 CPU_BASED_CR8_LOAD_EXITING;
3976 exec_control |= CPU_BASED_CR3_STORE_EXITING |
3977 CPU_BASED_CR3_LOAD_EXITING |
3978 CPU_BASED_INVLPG_EXITING;
3979 if (kvm_mwait_in_guest(vmx->vcpu.kvm))
3980 exec_control &= ~(CPU_BASED_MWAIT_EXITING |
3981 CPU_BASED_MONITOR_EXITING);
3982 if (kvm_hlt_in_guest(vmx->vcpu.kvm))
3983 exec_control &= ~CPU_BASED_HLT_EXITING;
3984 return exec_control;
3988 static void vmx_compute_secondary_exec_control(struct vcpu_vmx *vmx)
3990 struct kvm_vcpu *vcpu = &vmx->vcpu;
3992 u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
3994 if (vmx_pt_mode_is_system())
3995 exec_control &= ~(SECONDARY_EXEC_PT_USE_GPA | SECONDARY_EXEC_PT_CONCEAL_VMX);
3996 if (!cpu_need_virtualize_apic_accesses(vcpu))
3997 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
3999 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
4001 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
4002 enable_unrestricted_guest = 0;
4004 if (!enable_unrestricted_guest)
4005 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4006 if (kvm_pause_in_guest(vmx->vcpu.kvm))
4007 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
4008 if (!kvm_vcpu_apicv_active(vcpu))
4009 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
4010 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4011 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
4013 /* SECONDARY_EXEC_DESC is enabled/disabled on writes to CR4.UMIP,
4014 * in vmx_set_cr4. */
4015 exec_control &= ~SECONDARY_EXEC_DESC;
4017 /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4019 We can NOT enable shadow_vmcs here because we don't have yet
4022 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4025 exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
4027 if (vmx_xsaves_supported()) {
4028 /* Exposing XSAVES only when XSAVE is exposed */
4029 bool xsaves_enabled =
4030 boot_cpu_has(X86_FEATURE_XSAVE) &&
4031 guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
4032 guest_cpuid_has(vcpu, X86_FEATURE_XSAVES);
4034 vcpu->arch.xsaves_enabled = xsaves_enabled;
4036 if (!xsaves_enabled)
4037 exec_control &= ~SECONDARY_EXEC_XSAVES;
4041 vmx->nested.msrs.secondary_ctls_high |=
4042 SECONDARY_EXEC_XSAVES;
4044 vmx->nested.msrs.secondary_ctls_high &=
4045 ~SECONDARY_EXEC_XSAVES;
4049 if (cpu_has_vmx_rdtscp()) {
4050 bool rdtscp_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP);
4051 if (!rdtscp_enabled)
4052 exec_control &= ~SECONDARY_EXEC_RDTSCP;
4056 vmx->nested.msrs.secondary_ctls_high |=
4057 SECONDARY_EXEC_RDTSCP;
4059 vmx->nested.msrs.secondary_ctls_high &=
4060 ~SECONDARY_EXEC_RDTSCP;
4064 if (cpu_has_vmx_invpcid()) {
4065 /* Exposing INVPCID only when PCID is exposed */
4066 bool invpcid_enabled =
4067 guest_cpuid_has(vcpu, X86_FEATURE_INVPCID) &&
4068 guest_cpuid_has(vcpu, X86_FEATURE_PCID);
4070 if (!invpcid_enabled) {
4071 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
4072 guest_cpuid_clear(vcpu, X86_FEATURE_INVPCID);
4076 if (invpcid_enabled)
4077 vmx->nested.msrs.secondary_ctls_high |=
4078 SECONDARY_EXEC_ENABLE_INVPCID;
4080 vmx->nested.msrs.secondary_ctls_high &=
4081 ~SECONDARY_EXEC_ENABLE_INVPCID;
4085 if (vmx_rdrand_supported()) {
4086 bool rdrand_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDRAND);
4088 exec_control &= ~SECONDARY_EXEC_RDRAND_EXITING;
4092 vmx->nested.msrs.secondary_ctls_high |=
4093 SECONDARY_EXEC_RDRAND_EXITING;
4095 vmx->nested.msrs.secondary_ctls_high &=
4096 ~SECONDARY_EXEC_RDRAND_EXITING;
4100 if (vmx_rdseed_supported()) {
4101 bool rdseed_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDSEED);
4103 exec_control &= ~SECONDARY_EXEC_RDSEED_EXITING;
4107 vmx->nested.msrs.secondary_ctls_high |=
4108 SECONDARY_EXEC_RDSEED_EXITING;
4110 vmx->nested.msrs.secondary_ctls_high &=
4111 ~SECONDARY_EXEC_RDSEED_EXITING;
4115 if (vmx_waitpkg_supported()) {
4116 bool waitpkg_enabled =
4117 guest_cpuid_has(vcpu, X86_FEATURE_WAITPKG);
4119 if (!waitpkg_enabled)
4120 exec_control &= ~SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4123 if (waitpkg_enabled)
4124 vmx->nested.msrs.secondary_ctls_high |=
4125 SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4127 vmx->nested.msrs.secondary_ctls_high &=
4128 ~SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4132 vmx->secondary_exec_control = exec_control;
4135 static void ept_set_mmio_spte_mask(void)
4138 * EPT Misconfigurations can be generated if the value of bits 2:0
4139 * of an EPT paging-structure entry is 110b (write/execute).
4141 kvm_mmu_set_mmio_spte_mask(VMX_EPT_RWX_MASK,
4142 VMX_EPT_MISCONFIG_WX_VALUE, 0);
4145 #define VMX_XSS_EXIT_BITMAP 0
4148 * Noting that the initialization of Guest-state Area of VMCS is in
4151 static void init_vmcs(struct vcpu_vmx *vmx)
4154 nested_vmx_set_vmcs_shadowing_bitmap();
4156 if (cpu_has_vmx_msr_bitmap())
4157 vmcs_write64(MSR_BITMAP, __pa(vmx->vmcs01.msr_bitmap));
4159 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4162 pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
4164 exec_controls_set(vmx, vmx_exec_control(vmx));
4166 if (cpu_has_secondary_exec_ctrls()) {
4167 vmx_compute_secondary_exec_control(vmx);
4168 secondary_exec_controls_set(vmx, vmx->secondary_exec_control);
4171 if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
4172 vmcs_write64(EOI_EXIT_BITMAP0, 0);
4173 vmcs_write64(EOI_EXIT_BITMAP1, 0);
4174 vmcs_write64(EOI_EXIT_BITMAP2, 0);
4175 vmcs_write64(EOI_EXIT_BITMAP3, 0);
4177 vmcs_write16(GUEST_INTR_STATUS, 0);
4179 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4180 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4183 if (!kvm_pause_in_guest(vmx->vcpu.kvm)) {
4184 vmcs_write32(PLE_GAP, ple_gap);
4185 vmx->ple_window = ple_window;
4186 vmx->ple_window_dirty = true;
4189 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4190 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4191 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
4193 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
4194 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
4195 vmx_set_constant_host_state(vmx);
4196 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4197 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4199 if (cpu_has_vmx_vmfunc())
4200 vmcs_write64(VM_FUNCTION_CONTROL, 0);
4202 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4203 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4204 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
4205 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4206 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
4208 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
4209 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
4211 vm_exit_controls_set(vmx, vmx_vmexit_ctrl());
4213 /* 22.2.1, 20.8.1 */
4214 vm_entry_controls_set(vmx, vmx_vmentry_ctrl());
4216 vmx->vcpu.arch.cr0_guest_owned_bits = X86_CR0_TS;
4217 vmcs_writel(CR0_GUEST_HOST_MASK, ~X86_CR0_TS);
4219 set_cr4_guest_host_mask(vmx);
4222 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4224 if (vmx_xsaves_supported())
4225 vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
4228 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
4229 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
4232 if (cpu_has_vmx_encls_vmexit())
4233 vmcs_write64(ENCLS_EXITING_BITMAP, -1ull);
4235 if (vmx_pt_mode_is_host_guest()) {
4236 memset(&vmx->pt_desc, 0, sizeof(vmx->pt_desc));
4237 /* Bit[6~0] are forced to 1, writes are ignored. */
4238 vmx->pt_desc.guest.output_mask = 0x7F;
4239 vmcs_write64(GUEST_IA32_RTIT_CTL, 0);
4243 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
4245 struct vcpu_vmx *vmx = to_vmx(vcpu);
4246 struct msr_data apic_base_msr;
4249 vmx->rmode.vm86_active = 0;
4252 vmx->msr_ia32_umwait_control = 0;
4254 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
4255 vmx->hv_deadline_tsc = -1;
4256 kvm_set_cr8(vcpu, 0);
4259 apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
4260 MSR_IA32_APICBASE_ENABLE;
4261 if (kvm_vcpu_is_reset_bsp(vcpu))
4262 apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
4263 apic_base_msr.host_initiated = true;
4264 kvm_set_apic_base(vcpu, &apic_base_msr);
4267 vmx_segment_cache_clear(vmx);
4269 seg_setup(VCPU_SREG_CS);
4270 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4271 vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
4273 seg_setup(VCPU_SREG_DS);
4274 seg_setup(VCPU_SREG_ES);
4275 seg_setup(VCPU_SREG_FS);
4276 seg_setup(VCPU_SREG_GS);
4277 seg_setup(VCPU_SREG_SS);
4279 vmcs_write16(GUEST_TR_SELECTOR, 0);
4280 vmcs_writel(GUEST_TR_BASE, 0);
4281 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4282 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4284 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4285 vmcs_writel(GUEST_LDTR_BASE, 0);
4286 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4287 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4290 vmcs_write32(GUEST_SYSENTER_CS, 0);
4291 vmcs_writel(GUEST_SYSENTER_ESP, 0);
4292 vmcs_writel(GUEST_SYSENTER_EIP, 0);
4293 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4296 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
4297 kvm_rip_write(vcpu, 0xfff0);
4299 vmcs_writel(GUEST_GDTR_BASE, 0);
4300 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4302 vmcs_writel(GUEST_IDTR_BASE, 0);
4303 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4305 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4306 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4307 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4308 if (kvm_mpx_supported())
4309 vmcs_write64(GUEST_BNDCFGS, 0);
4313 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
4315 if (cpu_has_vmx_tpr_shadow() && !init_event) {
4316 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4317 if (cpu_need_tpr_shadow(vcpu))
4318 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4319 __pa(vcpu->arch.apic->regs));
4320 vmcs_write32(TPR_THRESHOLD, 0);
4323 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4325 cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4326 vmx->vcpu.arch.cr0 = cr0;
4327 vmx_set_cr0(vcpu, cr0); /* enter rmode */
4328 vmx_set_cr4(vcpu, 0);
4329 vmx_set_efer(vcpu, 0);
4331 update_exception_bitmap(vcpu);
4333 vpid_sync_context(vmx->vpid);
4335 vmx_clear_hlt(vcpu);
4338 static void enable_irq_window(struct kvm_vcpu *vcpu)
4340 exec_controls_setbit(to_vmx(vcpu), CPU_BASED_INTR_WINDOW_EXITING);
4343 static void enable_nmi_window(struct kvm_vcpu *vcpu)
4346 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4347 enable_irq_window(vcpu);
4351 exec_controls_setbit(to_vmx(vcpu), CPU_BASED_NMI_WINDOW_EXITING);
4354 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4356 struct vcpu_vmx *vmx = to_vmx(vcpu);
4358 int irq = vcpu->arch.interrupt.nr;
4360 trace_kvm_inj_virq(irq);
4362 ++vcpu->stat.irq_injections;
4363 if (vmx->rmode.vm86_active) {
4365 if (vcpu->arch.interrupt.soft)
4366 inc_eip = vcpu->arch.event_exit_inst_len;
4367 kvm_inject_realmode_interrupt(vcpu, irq, inc_eip);
4370 intr = irq | INTR_INFO_VALID_MASK;
4371 if (vcpu->arch.interrupt.soft) {
4372 intr |= INTR_TYPE_SOFT_INTR;
4373 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4374 vmx->vcpu.arch.event_exit_inst_len);
4376 intr |= INTR_TYPE_EXT_INTR;
4377 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4379 vmx_clear_hlt(vcpu);
4382 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4384 struct vcpu_vmx *vmx = to_vmx(vcpu);
4388 * Tracking the NMI-blocked state in software is built upon
4389 * finding the next open IRQ window. This, in turn, depends on
4390 * well-behaving guests: They have to keep IRQs disabled at
4391 * least as long as the NMI handler runs. Otherwise we may
4392 * cause NMI nesting, maybe breaking the guest. But as this is
4393 * highly unlikely, we can live with the residual risk.
4395 vmx->loaded_vmcs->soft_vnmi_blocked = 1;
4396 vmx->loaded_vmcs->vnmi_blocked_time = 0;
4399 ++vcpu->stat.nmi_injections;
4400 vmx->loaded_vmcs->nmi_known_unmasked = false;
4402 if (vmx->rmode.vm86_active) {
4403 kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0);
4407 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4408 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4410 vmx_clear_hlt(vcpu);
4413 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4415 struct vcpu_vmx *vmx = to_vmx(vcpu);
4419 return vmx->loaded_vmcs->soft_vnmi_blocked;
4420 if (vmx->loaded_vmcs->nmi_known_unmasked)
4422 masked = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4423 vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4427 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4429 struct vcpu_vmx *vmx = to_vmx(vcpu);
4432 if (vmx->loaded_vmcs->soft_vnmi_blocked != masked) {
4433 vmx->loaded_vmcs->soft_vnmi_blocked = masked;
4434 vmx->loaded_vmcs->vnmi_blocked_time = 0;
4437 vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4439 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4440 GUEST_INTR_STATE_NMI);
4442 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4443 GUEST_INTR_STATE_NMI);
4447 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
4449 if (to_vmx(vcpu)->nested.nested_run_pending)
4453 to_vmx(vcpu)->loaded_vmcs->soft_vnmi_blocked)
4456 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4457 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
4458 | GUEST_INTR_STATE_NMI));
4461 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
4463 if (to_vmx(vcpu)->nested.nested_run_pending)
4466 if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu))
4469 return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
4470 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4471 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4474 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4478 if (enable_unrestricted_guest)
4481 mutex_lock(&kvm->slots_lock);
4482 ret = __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
4484 mutex_unlock(&kvm->slots_lock);
4488 to_kvm_vmx(kvm)->tss_addr = addr;
4489 return init_rmode_tss(kvm);
4492 static int vmx_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
4494 to_kvm_vmx(kvm)->ept_identity_map_addr = ident_addr;
4498 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
4503 * Update instruction length as we may reinject the exception
4504 * from user space while in guest debugging mode.
4506 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4507 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4508 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4512 if (vcpu->guest_debug &
4513 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
4529 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4530 int vec, u32 err_code)
4533 * Instruction with address size override prefix opcode 0x67
4534 * Cause the #SS fault with 0 error code in VM86 mode.
4536 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
4537 if (kvm_emulate_instruction(vcpu, 0)) {
4538 if (vcpu->arch.halt_request) {
4539 vcpu->arch.halt_request = 0;
4540 return kvm_vcpu_halt(vcpu);
4548 * Forward all other exceptions that are valid in real mode.
4549 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4550 * the required debugging infrastructure rework.
4552 kvm_queue_exception(vcpu, vec);
4557 * Trigger machine check on the host. We assume all the MSRs are already set up
4558 * by the CPU and that we still run on the same CPU as the MCE occurred on.
4559 * We pass a fake environment to the machine check handler because we want
4560 * the guest to be always treated like user space, no matter what context
4561 * it used internally.
4563 static void kvm_machine_check(void)
4565 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
4566 struct pt_regs regs = {
4567 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
4568 .flags = X86_EFLAGS_IF,
4571 do_machine_check(®s, 0);
4575 static int handle_machine_check(struct kvm_vcpu *vcpu)
4577 /* handled by vmx_vcpu_run() */
4581 static int handle_exception_nmi(struct kvm_vcpu *vcpu)
4583 struct vcpu_vmx *vmx = to_vmx(vcpu);
4584 struct kvm_run *kvm_run = vcpu->run;
4585 u32 intr_info, ex_no, error_code;
4586 unsigned long cr2, rip, dr6;
4589 vect_info = vmx->idt_vectoring_info;
4590 intr_info = vmx->exit_intr_info;
4592 if (is_machine_check(intr_info) || is_nmi(intr_info))
4593 return 1; /* handled by handle_exception_nmi_irqoff() */
4595 if (is_invalid_opcode(intr_info))
4596 return handle_ud(vcpu);
4599 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4600 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4602 if (!vmx->rmode.vm86_active && is_gp_fault(intr_info)) {
4603 WARN_ON_ONCE(!enable_vmware_backdoor);
4606 * VMware backdoor emulation on #GP interception only handles
4607 * IN{S}, OUT{S}, and RDPMC, none of which generate a non-zero
4608 * error code on #GP.
4611 kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
4614 return kvm_emulate_instruction(vcpu, EMULTYPE_VMWARE_GP);
4618 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
4619 * MMIO, it is better to report an internal error.
4620 * See the comments in vmx_handle_exit.
4622 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4623 !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
4624 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4625 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4626 vcpu->run->internal.ndata = 3;
4627 vcpu->run->internal.data[0] = vect_info;
4628 vcpu->run->internal.data[1] = intr_info;
4629 vcpu->run->internal.data[2] = error_code;
4633 if (is_page_fault(intr_info)) {
4634 cr2 = vmcs_readl(EXIT_QUALIFICATION);
4635 /* EPT won't cause page fault directly */
4636 WARN_ON_ONCE(!vcpu->arch.apf.host_apf_reason && enable_ept);
4637 return kvm_handle_page_fault(vcpu, error_code, cr2, NULL, 0);
4640 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4642 if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
4643 return handle_rmode_exception(vcpu, ex_no, error_code);
4647 kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
4650 dr6 = vmcs_readl(EXIT_QUALIFICATION);
4651 if (!(vcpu->guest_debug &
4652 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4653 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
4654 vcpu->arch.dr6 |= dr6 | DR6_RTM;
4655 if (is_icebp(intr_info))
4656 WARN_ON(!skip_emulated_instruction(vcpu));
4658 kvm_queue_exception(vcpu, DB_VECTOR);
4661 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
4662 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4666 * Update instruction length as we may reinject #BP from
4667 * user space while in guest debugging mode. Reading it for
4668 * #DB as well causes no harm, it is not used in that case.
4670 vmx->vcpu.arch.event_exit_inst_len =
4671 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4672 kvm_run->exit_reason = KVM_EXIT_DEBUG;
4673 rip = kvm_rip_read(vcpu);
4674 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4675 kvm_run->debug.arch.exception = ex_no;
4678 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4679 kvm_run->ex.exception = ex_no;
4680 kvm_run->ex.error_code = error_code;
4686 static __always_inline int handle_external_interrupt(struct kvm_vcpu *vcpu)
4688 ++vcpu->stat.irq_exits;
4692 static int handle_triple_fault(struct kvm_vcpu *vcpu)
4694 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4695 vcpu->mmio_needed = 0;
4699 static int handle_io(struct kvm_vcpu *vcpu)
4701 unsigned long exit_qualification;
4702 int size, in, string;
4705 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4706 string = (exit_qualification & 16) != 0;
4708 ++vcpu->stat.io_exits;
4711 return kvm_emulate_instruction(vcpu, 0);
4713 port = exit_qualification >> 16;
4714 size = (exit_qualification & 7) + 1;
4715 in = (exit_qualification & 8) != 0;
4717 return kvm_fast_pio(vcpu, size, port, in);
4721 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4724 * Patch in the VMCALL instruction:
4726 hypercall[0] = 0x0f;
4727 hypercall[1] = 0x01;
4728 hypercall[2] = 0xc1;
4731 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
4732 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
4734 if (is_guest_mode(vcpu)) {
4735 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4736 unsigned long orig_val = val;
4739 * We get here when L2 changed cr0 in a way that did not change
4740 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
4741 * but did change L0 shadowed bits. So we first calculate the
4742 * effective cr0 value that L1 would like to write into the
4743 * hardware. It consists of the L2-owned bits from the new
4744 * value combined with the L1-owned bits from L1's guest_cr0.
4746 val = (val & ~vmcs12->cr0_guest_host_mask) |
4747 (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
4749 if (!nested_guest_cr0_valid(vcpu, val))
4752 if (kvm_set_cr0(vcpu, val))
4754 vmcs_writel(CR0_READ_SHADOW, orig_val);
4757 if (to_vmx(vcpu)->nested.vmxon &&
4758 !nested_host_cr0_valid(vcpu, val))
4761 return kvm_set_cr0(vcpu, val);
4765 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
4767 if (is_guest_mode(vcpu)) {
4768 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4769 unsigned long orig_val = val;
4771 /* analogously to handle_set_cr0 */
4772 val = (val & ~vmcs12->cr4_guest_host_mask) |
4773 (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
4774 if (kvm_set_cr4(vcpu, val))
4776 vmcs_writel(CR4_READ_SHADOW, orig_val);
4779 return kvm_set_cr4(vcpu, val);
4782 static int handle_desc(struct kvm_vcpu *vcpu)
4784 WARN_ON(!(vcpu->arch.cr4 & X86_CR4_UMIP));
4785 return kvm_emulate_instruction(vcpu, 0);
4788 static int handle_cr(struct kvm_vcpu *vcpu)
4790 unsigned long exit_qualification, val;
4796 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4797 cr = exit_qualification & 15;
4798 reg = (exit_qualification >> 8) & 15;
4799 switch ((exit_qualification >> 4) & 3) {
4800 case 0: /* mov to cr */
4801 val = kvm_register_readl(vcpu, reg);
4802 trace_kvm_cr_write(cr, val);
4805 err = handle_set_cr0(vcpu, val);
4806 return kvm_complete_insn_gp(vcpu, err);
4808 WARN_ON_ONCE(enable_unrestricted_guest);
4809 err = kvm_set_cr3(vcpu, val);
4810 return kvm_complete_insn_gp(vcpu, err);
4812 err = handle_set_cr4(vcpu, val);
4813 return kvm_complete_insn_gp(vcpu, err);
4815 u8 cr8_prev = kvm_get_cr8(vcpu);
4817 err = kvm_set_cr8(vcpu, cr8);
4818 ret = kvm_complete_insn_gp(vcpu, err);
4819 if (lapic_in_kernel(vcpu))
4821 if (cr8_prev <= cr8)
4824 * TODO: we might be squashing a
4825 * KVM_GUESTDBG_SINGLESTEP-triggered
4826 * KVM_EXIT_DEBUG here.
4828 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
4834 WARN_ONCE(1, "Guest should always own CR0.TS");
4835 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4836 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
4837 return kvm_skip_emulated_instruction(vcpu);
4838 case 1: /*mov from cr*/
4841 WARN_ON_ONCE(enable_unrestricted_guest);
4842 val = kvm_read_cr3(vcpu);
4843 kvm_register_write(vcpu, reg, val);
4844 trace_kvm_cr_read(cr, val);
4845 return kvm_skip_emulated_instruction(vcpu);
4847 val = kvm_get_cr8(vcpu);
4848 kvm_register_write(vcpu, reg, val);
4849 trace_kvm_cr_read(cr, val);
4850 return kvm_skip_emulated_instruction(vcpu);
4854 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
4855 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
4856 kvm_lmsw(vcpu, val);
4858 return kvm_skip_emulated_instruction(vcpu);
4862 vcpu->run->exit_reason = 0;
4863 vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
4864 (int)(exit_qualification >> 4) & 3, cr);
4868 static int handle_dr(struct kvm_vcpu *vcpu)
4870 unsigned long exit_qualification;
4873 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4874 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
4876 /* First, if DR does not exist, trigger UD */
4877 if (!kvm_require_dr(vcpu, dr))
4880 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
4881 if (!kvm_require_cpl(vcpu, 0))
4883 dr7 = vmcs_readl(GUEST_DR7);
4886 * As the vm-exit takes precedence over the debug trap, we
4887 * need to emulate the latter, either for the host or the
4888 * guest debugging itself.
4890 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
4891 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
4892 vcpu->run->debug.arch.dr7 = dr7;
4893 vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
4894 vcpu->run->debug.arch.exception = DB_VECTOR;
4895 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
4898 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
4899 vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
4900 kvm_queue_exception(vcpu, DB_VECTOR);
4905 if (vcpu->guest_debug == 0) {
4906 exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
4909 * No more DR vmexits; force a reload of the debug registers
4910 * and reenter on this instruction. The next vmexit will
4911 * retrieve the full state of the debug registers.
4913 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
4917 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
4918 if (exit_qualification & TYPE_MOV_FROM_DR) {
4921 if (kvm_get_dr(vcpu, dr, &val))
4923 kvm_register_write(vcpu, reg, val);
4925 if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
4928 return kvm_skip_emulated_instruction(vcpu);
4931 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
4933 return vcpu->arch.dr6;
4936 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
4940 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
4942 get_debugreg(vcpu->arch.db[0], 0);
4943 get_debugreg(vcpu->arch.db[1], 1);
4944 get_debugreg(vcpu->arch.db[2], 2);
4945 get_debugreg(vcpu->arch.db[3], 3);
4946 get_debugreg(vcpu->arch.dr6, 6);
4947 vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
4949 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
4950 exec_controls_setbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
4953 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
4955 vmcs_writel(GUEST_DR7, val);
4958 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
4960 kvm_apic_update_ppr(vcpu);
4964 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
4966 exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_INTR_WINDOW_EXITING);
4968 kvm_make_request(KVM_REQ_EVENT, vcpu);
4970 ++vcpu->stat.irq_window_exits;
4974 static int handle_vmcall(struct kvm_vcpu *vcpu)
4976 return kvm_emulate_hypercall(vcpu);
4979 static int handle_invd(struct kvm_vcpu *vcpu)
4981 return kvm_emulate_instruction(vcpu, 0);
4984 static int handle_invlpg(struct kvm_vcpu *vcpu)
4986 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4988 kvm_mmu_invlpg(vcpu, exit_qualification);
4989 return kvm_skip_emulated_instruction(vcpu);
4992 static int handle_rdpmc(struct kvm_vcpu *vcpu)
4996 err = kvm_rdpmc(vcpu);
4997 return kvm_complete_insn_gp(vcpu, err);
5000 static int handle_wbinvd(struct kvm_vcpu *vcpu)
5002 return kvm_emulate_wbinvd(vcpu);
5005 static int handle_xsetbv(struct kvm_vcpu *vcpu)
5007 u64 new_bv = kvm_read_edx_eax(vcpu);
5008 u32 index = kvm_rcx_read(vcpu);
5010 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
5011 return kvm_skip_emulated_instruction(vcpu);
5015 static int handle_apic_access(struct kvm_vcpu *vcpu)
5017 if (likely(fasteoi)) {
5018 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5019 int access_type, offset;
5021 access_type = exit_qualification & APIC_ACCESS_TYPE;
5022 offset = exit_qualification & APIC_ACCESS_OFFSET;
5024 * Sane guest uses MOV to write EOI, with written value
5025 * not cared. So make a short-circuit here by avoiding
5026 * heavy instruction emulation.
5028 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5029 (offset == APIC_EOI)) {
5030 kvm_lapic_set_eoi(vcpu);
5031 return kvm_skip_emulated_instruction(vcpu);
5034 return kvm_emulate_instruction(vcpu, 0);
5037 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5039 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5040 int vector = exit_qualification & 0xff;
5042 /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5043 kvm_apic_set_eoi_accelerated(vcpu, vector);
5047 static int handle_apic_write(struct kvm_vcpu *vcpu)
5049 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5050 u32 offset = exit_qualification & 0xfff;
5052 /* APIC-write VM exit is trap-like and thus no need to adjust IP */
5053 kvm_apic_write_nodecode(vcpu, offset);
5057 static int handle_task_switch(struct kvm_vcpu *vcpu)
5059 struct vcpu_vmx *vmx = to_vmx(vcpu);
5060 unsigned long exit_qualification;
5061 bool has_error_code = false;
5064 int reason, type, idt_v, idt_index;
5066 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5067 idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5068 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5070 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5072 reason = (u32)exit_qualification >> 30;
5073 if (reason == TASK_SWITCH_GATE && idt_v) {
5075 case INTR_TYPE_NMI_INTR:
5076 vcpu->arch.nmi_injected = false;
5077 vmx_set_nmi_mask(vcpu, true);
5079 case INTR_TYPE_EXT_INTR:
5080 case INTR_TYPE_SOFT_INTR:
5081 kvm_clear_interrupt_queue(vcpu);
5083 case INTR_TYPE_HARD_EXCEPTION:
5084 if (vmx->idt_vectoring_info &
5085 VECTORING_INFO_DELIVER_CODE_MASK) {
5086 has_error_code = true;
5088 vmcs_read32(IDT_VECTORING_ERROR_CODE);
5091 case INTR_TYPE_SOFT_EXCEPTION:
5092 kvm_clear_exception_queue(vcpu);
5098 tss_selector = exit_qualification;
5100 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5101 type != INTR_TYPE_EXT_INTR &&
5102 type != INTR_TYPE_NMI_INTR))
5103 WARN_ON(!skip_emulated_instruction(vcpu));
5106 * TODO: What about debug traps on tss switch?
5107 * Are we supposed to inject them and update dr6?
5109 return kvm_task_switch(vcpu, tss_selector,
5110 type == INTR_TYPE_SOFT_INTR ? idt_index : -1,
5111 reason, has_error_code, error_code);
5114 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5116 unsigned long exit_qualification;
5120 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5123 * EPT violation happened while executing iret from NMI,
5124 * "blocked by NMI" bit has to be set before next VM entry.
5125 * There are errata that may cause this bit to not be set:
5128 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5130 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
5131 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5133 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5134 trace_kvm_page_fault(gpa, exit_qualification);
5136 /* Is it a read fault? */
5137 error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
5138 ? PFERR_USER_MASK : 0;
5139 /* Is it a write fault? */
5140 error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
5141 ? PFERR_WRITE_MASK : 0;
5142 /* Is it a fetch fault? */
5143 error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
5144 ? PFERR_FETCH_MASK : 0;
5145 /* ept page table entry is present? */
5146 error_code |= (exit_qualification &
5147 (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE |
5148 EPT_VIOLATION_EXECUTABLE))
5149 ? PFERR_PRESENT_MASK : 0;
5151 error_code |= (exit_qualification & 0x100) != 0 ?
5152 PFERR_GUEST_FINAL_MASK : PFERR_GUEST_PAGE_MASK;
5154 vcpu->arch.exit_qualification = exit_qualification;
5155 return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5158 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5163 * A nested guest cannot optimize MMIO vmexits, because we have an
5164 * nGPA here instead of the required GPA.
5166 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5167 if (!is_guest_mode(vcpu) &&
5168 !kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5169 trace_kvm_fast_mmio(gpa);
5170 return kvm_skip_emulated_instruction(vcpu);
5173 return kvm_mmu_page_fault(vcpu, gpa, PFERR_RSVD_MASK, NULL, 0);
5176 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5178 WARN_ON_ONCE(!enable_vnmi);
5179 exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_NMI_WINDOW_EXITING);
5180 ++vcpu->stat.nmi_window_exits;
5181 kvm_make_request(KVM_REQ_EVENT, vcpu);
5186 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5188 struct vcpu_vmx *vmx = to_vmx(vcpu);
5189 bool intr_window_requested;
5190 unsigned count = 130;
5193 * We should never reach the point where we are emulating L2
5194 * due to invalid guest state as that means we incorrectly
5195 * allowed a nested VMEntry with an invalid vmcs12.
5197 WARN_ON_ONCE(vmx->emulation_required && vmx->nested.nested_run_pending);
5199 intr_window_requested = exec_controls_get(vmx) &
5200 CPU_BASED_INTR_WINDOW_EXITING;
5202 while (vmx->emulation_required && count-- != 0) {
5203 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
5204 return handle_interrupt_window(&vmx->vcpu);
5206 if (kvm_test_request(KVM_REQ_EVENT, vcpu))
5209 if (!kvm_emulate_instruction(vcpu, 0))
5212 if (vmx->emulation_required && !vmx->rmode.vm86_active &&
5213 vcpu->arch.exception.pending) {
5214 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5215 vcpu->run->internal.suberror =
5216 KVM_INTERNAL_ERROR_EMULATION;
5217 vcpu->run->internal.ndata = 0;
5221 if (vcpu->arch.halt_request) {
5222 vcpu->arch.halt_request = 0;
5223 return kvm_vcpu_halt(vcpu);
5227 * Note, return 1 and not 0, vcpu_run() is responsible for
5228 * morphing the pending signal into the proper return code.
5230 if (signal_pending(current))
5240 static void grow_ple_window(struct kvm_vcpu *vcpu)
5242 struct vcpu_vmx *vmx = to_vmx(vcpu);
5243 unsigned int old = vmx->ple_window;
5245 vmx->ple_window = __grow_ple_window(old, ple_window,
5249 if (vmx->ple_window != old) {
5250 vmx->ple_window_dirty = true;
5251 trace_kvm_ple_window_update(vcpu->vcpu_id,
5252 vmx->ple_window, old);
5256 static void shrink_ple_window(struct kvm_vcpu *vcpu)
5258 struct vcpu_vmx *vmx = to_vmx(vcpu);
5259 unsigned int old = vmx->ple_window;
5261 vmx->ple_window = __shrink_ple_window(old, ple_window,
5265 if (vmx->ple_window != old) {
5266 vmx->ple_window_dirty = true;
5267 trace_kvm_ple_window_update(vcpu->vcpu_id,
5268 vmx->ple_window, old);
5273 * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
5275 static void wakeup_handler(void)
5277 struct kvm_vcpu *vcpu;
5278 int cpu = smp_processor_id();
5280 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5281 list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
5282 blocked_vcpu_list) {
5283 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
5285 if (pi_test_on(pi_desc) == 1)
5286 kvm_vcpu_kick(vcpu);
5288 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5291 static void vmx_enable_tdp(void)
5293 kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK,
5294 enable_ept_ad_bits ? VMX_EPT_ACCESS_BIT : 0ull,
5295 enable_ept_ad_bits ? VMX_EPT_DIRTY_BIT : 0ull,
5296 0ull, VMX_EPT_EXECUTABLE_MASK,
5297 cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK,
5298 VMX_EPT_RWX_MASK, 0ull);
5300 ept_set_mmio_spte_mask();
5304 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5305 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5307 static int handle_pause(struct kvm_vcpu *vcpu)
5309 if (!kvm_pause_in_guest(vcpu->kvm))
5310 grow_ple_window(vcpu);
5313 * Intel sdm vol3 ch-25.1.3 says: The "PAUSE-loop exiting"
5314 * VM-execution control is ignored if CPL > 0. OTOH, KVM
5315 * never set PAUSE_EXITING and just set PLE if supported,
5316 * so the vcpu must be CPL=0 if it gets a PAUSE exit.
5318 kvm_vcpu_on_spin(vcpu, true);
5319 return kvm_skip_emulated_instruction(vcpu);
5322 static int handle_nop(struct kvm_vcpu *vcpu)
5324 return kvm_skip_emulated_instruction(vcpu);
5327 static int handle_mwait(struct kvm_vcpu *vcpu)
5329 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
5330 return handle_nop(vcpu);
5333 static int handle_invalid_op(struct kvm_vcpu *vcpu)
5335 kvm_queue_exception(vcpu, UD_VECTOR);
5339 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
5344 static int handle_monitor(struct kvm_vcpu *vcpu)
5346 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
5347 return handle_nop(vcpu);
5350 static int handle_invpcid(struct kvm_vcpu *vcpu)
5352 u32 vmx_instruction_info;
5356 struct x86_exception e;
5358 unsigned long roots_to_free = 0;
5364 if (!guest_cpuid_has(vcpu, X86_FEATURE_INVPCID)) {
5365 kvm_queue_exception(vcpu, UD_VECTOR);
5369 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5370 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
5373 kvm_inject_gp(vcpu, 0);
5377 /* According to the Intel instruction reference, the memory operand
5378 * is read even if it isn't needed (e.g., for type==all)
5380 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5381 vmx_instruction_info, false,
5382 sizeof(operand), &gva))
5385 if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
5386 kvm_inject_page_fault(vcpu, &e);
5390 if (operand.pcid >> 12 != 0) {
5391 kvm_inject_gp(vcpu, 0);
5395 pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
5398 case INVPCID_TYPE_INDIV_ADDR:
5399 if ((!pcid_enabled && (operand.pcid != 0)) ||
5400 is_noncanonical_address(operand.gla, vcpu)) {
5401 kvm_inject_gp(vcpu, 0);
5404 kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
5405 return kvm_skip_emulated_instruction(vcpu);
5407 case INVPCID_TYPE_SINGLE_CTXT:
5408 if (!pcid_enabled && (operand.pcid != 0)) {
5409 kvm_inject_gp(vcpu, 0);
5413 if (kvm_get_active_pcid(vcpu) == operand.pcid) {
5414 kvm_mmu_sync_roots(vcpu);
5415 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
5418 for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
5419 if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].cr3)
5421 roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
5423 kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
5425 * If neither the current cr3 nor any of the prev_roots use the
5426 * given PCID, then nothing needs to be done here because a
5427 * resync will happen anyway before switching to any other CR3.
5430 return kvm_skip_emulated_instruction(vcpu);
5432 case INVPCID_TYPE_ALL_NON_GLOBAL:
5434 * Currently, KVM doesn't mark global entries in the shadow
5435 * page tables, so a non-global flush just degenerates to a
5436 * global flush. If needed, we could optimize this later by
5437 * keeping track of global entries in shadow page tables.
5441 case INVPCID_TYPE_ALL_INCL_GLOBAL:
5442 kvm_mmu_unload(vcpu);
5443 return kvm_skip_emulated_instruction(vcpu);
5446 BUG(); /* We have already checked above that type <= 3 */
5450 static int handle_pml_full(struct kvm_vcpu *vcpu)
5452 unsigned long exit_qualification;
5454 trace_kvm_pml_full(vcpu->vcpu_id);
5456 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5459 * PML buffer FULL happened while executing iret from NMI,
5460 * "blocked by NMI" bit has to be set before next VM entry.
5462 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5464 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
5465 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5466 GUEST_INTR_STATE_NMI);
5469 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
5470 * here.., and there's no userspace involvement needed for PML.
5475 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
5477 struct vcpu_vmx *vmx = to_vmx(vcpu);
5479 if (!vmx->req_immediate_exit &&
5480 !unlikely(vmx->loaded_vmcs->hv_timer_soft_disabled))
5481 kvm_lapic_expired_hv_timer(vcpu);
5487 * When nested=0, all VMX instruction VM Exits filter here. The handlers
5488 * are overwritten by nested_vmx_setup() when nested=1.
5490 static int handle_vmx_instruction(struct kvm_vcpu *vcpu)
5492 kvm_queue_exception(vcpu, UD_VECTOR);
5496 static int handle_encls(struct kvm_vcpu *vcpu)
5499 * SGX virtualization is not yet supported. There is no software
5500 * enable bit for SGX, so we have to trap ENCLS and inject a #UD
5501 * to prevent the guest from executing ENCLS.
5503 kvm_queue_exception(vcpu, UD_VECTOR);
5508 * The exit handlers return 1 if the exit was handled fully and guest execution
5509 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
5510 * to be done to userspace and return 0.
5512 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
5513 [EXIT_REASON_EXCEPTION_NMI] = handle_exception_nmi,
5514 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
5515 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
5516 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
5517 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
5518 [EXIT_REASON_CR_ACCESS] = handle_cr,
5519 [EXIT_REASON_DR_ACCESS] = handle_dr,
5520 [EXIT_REASON_CPUID] = kvm_emulate_cpuid,
5521 [EXIT_REASON_MSR_READ] = kvm_emulate_rdmsr,
5522 [EXIT_REASON_MSR_WRITE] = kvm_emulate_wrmsr,
5523 [EXIT_REASON_INTERRUPT_WINDOW] = handle_interrupt_window,
5524 [EXIT_REASON_HLT] = kvm_emulate_halt,
5525 [EXIT_REASON_INVD] = handle_invd,
5526 [EXIT_REASON_INVLPG] = handle_invlpg,
5527 [EXIT_REASON_RDPMC] = handle_rdpmc,
5528 [EXIT_REASON_VMCALL] = handle_vmcall,
5529 [EXIT_REASON_VMCLEAR] = handle_vmx_instruction,
5530 [EXIT_REASON_VMLAUNCH] = handle_vmx_instruction,
5531 [EXIT_REASON_VMPTRLD] = handle_vmx_instruction,
5532 [EXIT_REASON_VMPTRST] = handle_vmx_instruction,
5533 [EXIT_REASON_VMREAD] = handle_vmx_instruction,
5534 [EXIT_REASON_VMRESUME] = handle_vmx_instruction,
5535 [EXIT_REASON_VMWRITE] = handle_vmx_instruction,
5536 [EXIT_REASON_VMOFF] = handle_vmx_instruction,
5537 [EXIT_REASON_VMON] = handle_vmx_instruction,
5538 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
5539 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
5540 [EXIT_REASON_APIC_WRITE] = handle_apic_write,
5541 [EXIT_REASON_EOI_INDUCED] = handle_apic_eoi_induced,
5542 [EXIT_REASON_WBINVD] = handle_wbinvd,
5543 [EXIT_REASON_XSETBV] = handle_xsetbv,
5544 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
5545 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
5546 [EXIT_REASON_GDTR_IDTR] = handle_desc,
5547 [EXIT_REASON_LDTR_TR] = handle_desc,
5548 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
5549 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
5550 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
5551 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_mwait,
5552 [EXIT_REASON_MONITOR_TRAP_FLAG] = handle_monitor_trap,
5553 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_monitor,
5554 [EXIT_REASON_INVEPT] = handle_vmx_instruction,
5555 [EXIT_REASON_INVVPID] = handle_vmx_instruction,
5556 [EXIT_REASON_RDRAND] = handle_invalid_op,
5557 [EXIT_REASON_RDSEED] = handle_invalid_op,
5558 [EXIT_REASON_PML_FULL] = handle_pml_full,
5559 [EXIT_REASON_INVPCID] = handle_invpcid,
5560 [EXIT_REASON_VMFUNC] = handle_vmx_instruction,
5561 [EXIT_REASON_PREEMPTION_TIMER] = handle_preemption_timer,
5562 [EXIT_REASON_ENCLS] = handle_encls,
5565 static const int kvm_vmx_max_exit_handlers =
5566 ARRAY_SIZE(kvm_vmx_exit_handlers);
5568 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
5570 *info1 = vmcs_readl(EXIT_QUALIFICATION);
5571 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
5574 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
5577 __free_page(vmx->pml_pg);
5582 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
5584 struct vcpu_vmx *vmx = to_vmx(vcpu);
5588 pml_idx = vmcs_read16(GUEST_PML_INDEX);
5590 /* Do nothing if PML buffer is empty */
5591 if (pml_idx == (PML_ENTITY_NUM - 1))
5594 /* PML index always points to next available PML buffer entity */
5595 if (pml_idx >= PML_ENTITY_NUM)
5600 pml_buf = page_address(vmx->pml_pg);
5601 for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
5604 gpa = pml_buf[pml_idx];
5605 WARN_ON(gpa & (PAGE_SIZE - 1));
5606 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5609 /* reset PML index */
5610 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
5614 * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
5615 * Called before reporting dirty_bitmap to userspace.
5617 static void kvm_flush_pml_buffers(struct kvm *kvm)
5620 struct kvm_vcpu *vcpu;
5622 * We only need to kick vcpu out of guest mode here, as PML buffer
5623 * is flushed at beginning of all VMEXITs, and it's obvious that only
5624 * vcpus running in guest are possible to have unflushed GPAs in PML
5627 kvm_for_each_vcpu(i, vcpu, kvm)
5628 kvm_vcpu_kick(vcpu);
5631 static void vmx_dump_sel(char *name, uint32_t sel)
5633 pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
5634 name, vmcs_read16(sel),
5635 vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
5636 vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
5637 vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
5640 static void vmx_dump_dtsel(char *name, uint32_t limit)
5642 pr_err("%s limit=0x%08x, base=0x%016lx\n",
5643 name, vmcs_read32(limit),
5644 vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
5647 void dump_vmcs(void)
5649 u32 vmentry_ctl, vmexit_ctl;
5650 u32 cpu_based_exec_ctrl, pin_based_exec_ctrl, secondary_exec_control;
5655 if (!dump_invalid_vmcs) {
5656 pr_warn_ratelimited("set kvm_intel.dump_invalid_vmcs=1 to dump internal KVM state.\n");
5660 vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
5661 vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
5662 cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5663 pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
5664 cr4 = vmcs_readl(GUEST_CR4);
5665 efer = vmcs_read64(GUEST_IA32_EFER);
5666 secondary_exec_control = 0;
5667 if (cpu_has_secondary_exec_ctrls())
5668 secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
5670 pr_err("*** Guest State ***\n");
5671 pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5672 vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
5673 vmcs_readl(CR0_GUEST_HOST_MASK));
5674 pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5675 cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
5676 pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
5677 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
5678 (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
5680 pr_err("PDPTR0 = 0x%016llx PDPTR1 = 0x%016llx\n",
5681 vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
5682 pr_err("PDPTR2 = 0x%016llx PDPTR3 = 0x%016llx\n",
5683 vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
5685 pr_err("RSP = 0x%016lx RIP = 0x%016lx\n",
5686 vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
5687 pr_err("RFLAGS=0x%08lx DR7 = 0x%016lx\n",
5688 vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
5689 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5690 vmcs_readl(GUEST_SYSENTER_ESP),
5691 vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
5692 vmx_dump_sel("CS: ", GUEST_CS_SELECTOR);
5693 vmx_dump_sel("DS: ", GUEST_DS_SELECTOR);
5694 vmx_dump_sel("SS: ", GUEST_SS_SELECTOR);
5695 vmx_dump_sel("ES: ", GUEST_ES_SELECTOR);
5696 vmx_dump_sel("FS: ", GUEST_FS_SELECTOR);
5697 vmx_dump_sel("GS: ", GUEST_GS_SELECTOR);
5698 vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
5699 vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
5700 vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
5701 vmx_dump_sel("TR: ", GUEST_TR_SELECTOR);
5702 if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
5703 (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
5704 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
5705 efer, vmcs_read64(GUEST_IA32_PAT));
5706 pr_err("DebugCtl = 0x%016llx DebugExceptions = 0x%016lx\n",
5707 vmcs_read64(GUEST_IA32_DEBUGCTL),
5708 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
5709 if (cpu_has_load_perf_global_ctrl() &&
5710 vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
5711 pr_err("PerfGlobCtl = 0x%016llx\n",
5712 vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
5713 if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
5714 pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
5715 pr_err("Interruptibility = %08x ActivityState = %08x\n",
5716 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
5717 vmcs_read32(GUEST_ACTIVITY_STATE));
5718 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
5719 pr_err("InterruptStatus = %04x\n",
5720 vmcs_read16(GUEST_INTR_STATUS));
5722 pr_err("*** Host State ***\n");
5723 pr_err("RIP = 0x%016lx RSP = 0x%016lx\n",
5724 vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
5725 pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
5726 vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
5727 vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
5728 vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
5729 vmcs_read16(HOST_TR_SELECTOR));
5730 pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
5731 vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
5732 vmcs_readl(HOST_TR_BASE));
5733 pr_err("GDTBase=%016lx IDTBase=%016lx\n",
5734 vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
5735 pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
5736 vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
5737 vmcs_readl(HOST_CR4));
5738 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5739 vmcs_readl(HOST_IA32_SYSENTER_ESP),
5740 vmcs_read32(HOST_IA32_SYSENTER_CS),
5741 vmcs_readl(HOST_IA32_SYSENTER_EIP));
5742 if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
5743 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
5744 vmcs_read64(HOST_IA32_EFER),
5745 vmcs_read64(HOST_IA32_PAT));
5746 if (cpu_has_load_perf_global_ctrl() &&
5747 vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
5748 pr_err("PerfGlobCtl = 0x%016llx\n",
5749 vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
5751 pr_err("*** Control State ***\n");
5752 pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
5753 pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
5754 pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
5755 pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
5756 vmcs_read32(EXCEPTION_BITMAP),
5757 vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
5758 vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
5759 pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
5760 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
5761 vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
5762 vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
5763 pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
5764 vmcs_read32(VM_EXIT_INTR_INFO),
5765 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
5766 vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
5767 pr_err(" reason=%08x qualification=%016lx\n",
5768 vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
5769 pr_err("IDTVectoring: info=%08x errcode=%08x\n",
5770 vmcs_read32(IDT_VECTORING_INFO_FIELD),
5771 vmcs_read32(IDT_VECTORING_ERROR_CODE));
5772 pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
5773 if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
5774 pr_err("TSC Multiplier = 0x%016llx\n",
5775 vmcs_read64(TSC_MULTIPLIER));
5776 if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW) {
5777 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
5778 u16 status = vmcs_read16(GUEST_INTR_STATUS);
5779 pr_err("SVI|RVI = %02x|%02x ", status >> 8, status & 0xff);
5781 pr_cont("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
5782 if (secondary_exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
5783 pr_err("APIC-access addr = 0x%016llx ", vmcs_read64(APIC_ACCESS_ADDR));
5784 pr_cont("virt-APIC addr = 0x%016llx\n", vmcs_read64(VIRTUAL_APIC_PAGE_ADDR));
5786 if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
5787 pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
5788 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
5789 pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
5790 n = vmcs_read32(CR3_TARGET_COUNT);
5791 for (i = 0; i + 1 < n; i += 4)
5792 pr_err("CR3 target%u=%016lx target%u=%016lx\n",
5793 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
5794 i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
5796 pr_err("CR3 target%u=%016lx\n",
5797 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
5798 if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
5799 pr_err("PLE Gap=%08x Window=%08x\n",
5800 vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
5801 if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
5802 pr_err("Virtual processor ID = 0x%04x\n",
5803 vmcs_read16(VIRTUAL_PROCESSOR_ID));
5807 * The guest has exited. See if we can fix it or if we need userspace
5810 static int vmx_handle_exit(struct kvm_vcpu *vcpu,
5811 enum exit_fastpath_completion exit_fastpath)
5813 struct vcpu_vmx *vmx = to_vmx(vcpu);
5814 u32 exit_reason = vmx->exit_reason;
5815 u32 vectoring_info = vmx->idt_vectoring_info;
5817 trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
5820 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
5821 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
5822 * querying dirty_bitmap, we only need to kick all vcpus out of guest
5823 * mode as if vcpus is in root mode, the PML buffer must has been
5827 vmx_flush_pml_buffer(vcpu);
5829 /* If guest state is invalid, start emulating */
5830 if (vmx->emulation_required)
5831 return handle_invalid_guest_state(vcpu);
5833 if (is_guest_mode(vcpu)) {
5835 * The host physical addresses of some pages of guest memory
5836 * are loaded into the vmcs02 (e.g. vmcs12's Virtual APIC
5837 * Page). The CPU may write to these pages via their host
5838 * physical address while L2 is running, bypassing any
5839 * address-translation-based dirty tracking (e.g. EPT write
5842 * Mark them dirty on every exit from L2 to prevent them from
5843 * getting out of sync with dirty tracking.
5845 nested_mark_vmcs12_pages_dirty(vcpu);
5847 if (nested_vmx_exit_reflected(vcpu, exit_reason))
5848 return nested_vmx_reflect_vmexit(vcpu, exit_reason);
5851 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
5853 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5854 vcpu->run->fail_entry.hardware_entry_failure_reason
5859 if (unlikely(vmx->fail)) {
5861 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5862 vcpu->run->fail_entry.hardware_entry_failure_reason
5863 = vmcs_read32(VM_INSTRUCTION_ERROR);
5869 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
5870 * delivery event since it indicates guest is accessing MMIO.
5871 * The vm-exit can be triggered again after return to guest that
5872 * will cause infinite loop.
5874 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
5875 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
5876 exit_reason != EXIT_REASON_EPT_VIOLATION &&
5877 exit_reason != EXIT_REASON_PML_FULL &&
5878 exit_reason != EXIT_REASON_TASK_SWITCH)) {
5879 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5880 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
5881 vcpu->run->internal.ndata = 3;
5882 vcpu->run->internal.data[0] = vectoring_info;
5883 vcpu->run->internal.data[1] = exit_reason;
5884 vcpu->run->internal.data[2] = vcpu->arch.exit_qualification;
5885 if (exit_reason == EXIT_REASON_EPT_MISCONFIG) {
5886 vcpu->run->internal.ndata++;
5887 vcpu->run->internal.data[3] =
5888 vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5893 if (unlikely(!enable_vnmi &&
5894 vmx->loaded_vmcs->soft_vnmi_blocked)) {
5895 if (vmx_interrupt_allowed(vcpu)) {
5896 vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5897 } else if (vmx->loaded_vmcs->vnmi_blocked_time > 1000000000LL &&
5898 vcpu->arch.nmi_pending) {
5900 * This CPU don't support us in finding the end of an
5901 * NMI-blocked window if the guest runs with IRQs
5902 * disabled. So we pull the trigger after 1 s of
5903 * futile waiting, but inform the user about this.
5905 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
5906 "state on VCPU %d after 1 s timeout\n",
5907 __func__, vcpu->vcpu_id);
5908 vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5912 if (exit_fastpath == EXIT_FASTPATH_SKIP_EMUL_INS) {
5913 kvm_skip_emulated_instruction(vcpu);
5917 if (exit_reason >= kvm_vmx_max_exit_handlers)
5918 goto unexpected_vmexit;
5919 #ifdef CONFIG_RETPOLINE
5920 if (exit_reason == EXIT_REASON_MSR_WRITE)
5921 return kvm_emulate_wrmsr(vcpu);
5922 else if (exit_reason == EXIT_REASON_PREEMPTION_TIMER)
5923 return handle_preemption_timer(vcpu);
5924 else if (exit_reason == EXIT_REASON_INTERRUPT_WINDOW)
5925 return handle_interrupt_window(vcpu);
5926 else if (exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
5927 return handle_external_interrupt(vcpu);
5928 else if (exit_reason == EXIT_REASON_HLT)
5929 return kvm_emulate_halt(vcpu);
5930 else if (exit_reason == EXIT_REASON_EPT_MISCONFIG)
5931 return handle_ept_misconfig(vcpu);
5934 exit_reason = array_index_nospec(exit_reason,
5935 kvm_vmx_max_exit_handlers);
5936 if (!kvm_vmx_exit_handlers[exit_reason])
5937 goto unexpected_vmexit;
5939 return kvm_vmx_exit_handlers[exit_reason](vcpu);
5942 vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n", exit_reason);
5944 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5945 vcpu->run->internal.suberror =
5946 KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON;
5947 vcpu->run->internal.ndata = 1;
5948 vcpu->run->internal.data[0] = exit_reason;
5953 * Software based L1D cache flush which is used when microcode providing
5954 * the cache control MSR is not loaded.
5956 * The L1D cache is 32 KiB on Nehalem and later microarchitectures, but to
5957 * flush it is required to read in 64 KiB because the replacement algorithm
5958 * is not exactly LRU. This could be sized at runtime via topology
5959 * information but as all relevant affected CPUs have 32KiB L1D cache size
5960 * there is no point in doing so.
5962 static void vmx_l1d_flush(struct kvm_vcpu *vcpu)
5964 int size = PAGE_SIZE << L1D_CACHE_ORDER;
5967 * This code is only executed when the the flush mode is 'cond' or
5970 if (static_branch_likely(&vmx_l1d_flush_cond)) {
5974 * Clear the per-vcpu flush bit, it gets set again
5975 * either from vcpu_run() or from one of the unsafe
5978 flush_l1d = vcpu->arch.l1tf_flush_l1d;
5979 vcpu->arch.l1tf_flush_l1d = false;
5982 * Clear the per-cpu flush bit, it gets set again from
5983 * the interrupt handlers.
5985 flush_l1d |= kvm_get_cpu_l1tf_flush_l1d();
5986 kvm_clear_cpu_l1tf_flush_l1d();
5992 vcpu->stat.l1d_flush++;
5994 if (static_cpu_has(X86_FEATURE_FLUSH_L1D)) {
5995 wrmsrl(MSR_IA32_FLUSH_CMD, L1D_FLUSH);
6000 /* First ensure the pages are in the TLB */
6001 "xorl %%eax, %%eax\n"
6002 ".Lpopulate_tlb:\n\t"
6003 "movzbl (%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
6004 "addl $4096, %%eax\n\t"
6005 "cmpl %%eax, %[size]\n\t"
6006 "jne .Lpopulate_tlb\n\t"
6007 "xorl %%eax, %%eax\n\t"
6009 /* Now fill the cache */
6010 "xorl %%eax, %%eax\n"
6012 "movzbl (%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
6013 "addl $64, %%eax\n\t"
6014 "cmpl %%eax, %[size]\n\t"
6015 "jne .Lfill_cache\n\t"
6017 :: [flush_pages] "r" (vmx_l1d_flush_pages),
6019 : "eax", "ebx", "ecx", "edx");
6022 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
6024 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6027 if (is_guest_mode(vcpu) &&
6028 nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
6031 tpr_threshold = (irr == -1 || tpr < irr) ? 0 : irr;
6032 if (is_guest_mode(vcpu))
6033 to_vmx(vcpu)->nested.l1_tpr_threshold = tpr_threshold;
6035 vmcs_write32(TPR_THRESHOLD, tpr_threshold);
6038 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu)
6040 struct vcpu_vmx *vmx = to_vmx(vcpu);
6041 u32 sec_exec_control;
6043 if (!lapic_in_kernel(vcpu))
6046 if (!flexpriority_enabled &&
6047 !cpu_has_vmx_virtualize_x2apic_mode())
6050 /* Postpone execution until vmcs01 is the current VMCS. */
6051 if (is_guest_mode(vcpu)) {
6052 vmx->nested.change_vmcs01_virtual_apic_mode = true;
6056 sec_exec_control = secondary_exec_controls_get(vmx);
6057 sec_exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6058 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
6060 switch (kvm_get_apic_mode(vcpu)) {
6061 case LAPIC_MODE_INVALID:
6062 WARN_ONCE(true, "Invalid local APIC state");
6063 case LAPIC_MODE_DISABLED:
6065 case LAPIC_MODE_XAPIC:
6066 if (flexpriority_enabled) {
6068 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6069 vmx_flush_tlb(vcpu, true);
6072 case LAPIC_MODE_X2APIC:
6073 if (cpu_has_vmx_virtualize_x2apic_mode())
6075 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
6078 secondary_exec_controls_set(vmx, sec_exec_control);
6080 vmx_update_msr_bitmap(vcpu);
6083 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
6085 if (!is_guest_mode(vcpu)) {
6086 vmcs_write64(APIC_ACCESS_ADDR, hpa);
6087 vmx_flush_tlb(vcpu, true);
6091 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
6099 status = vmcs_read16(GUEST_INTR_STATUS);
6101 if (max_isr != old) {
6103 status |= max_isr << 8;
6104 vmcs_write16(GUEST_INTR_STATUS, status);
6108 static void vmx_set_rvi(int vector)
6116 status = vmcs_read16(GUEST_INTR_STATUS);
6117 old = (u8)status & 0xff;
6118 if ((u8)vector != old) {
6120 status |= (u8)vector;
6121 vmcs_write16(GUEST_INTR_STATUS, status);
6125 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
6128 * When running L2, updating RVI is only relevant when
6129 * vmcs12 virtual-interrupt-delivery enabled.
6130 * However, it can be enabled only when L1 also
6131 * intercepts external-interrupts and in that case
6132 * we should not update vmcs02 RVI but instead intercept
6133 * interrupt. Therefore, do nothing when running L2.
6135 if (!is_guest_mode(vcpu))
6136 vmx_set_rvi(max_irr);
6139 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
6141 struct vcpu_vmx *vmx = to_vmx(vcpu);
6143 bool max_irr_updated;
6145 WARN_ON(!vcpu->arch.apicv_active);
6146 if (pi_test_on(&vmx->pi_desc)) {
6147 pi_clear_on(&vmx->pi_desc);
6149 * IOMMU can write to PID.ON, so the barrier matters even on UP.
6150 * But on x86 this is just a compiler barrier anyway.
6152 smp_mb__after_atomic();
6154 kvm_apic_update_irr(vcpu, vmx->pi_desc.pir, &max_irr);
6157 * If we are running L2 and L1 has a new pending interrupt
6158 * which can be injected, we should re-evaluate
6159 * what should be done with this new L1 interrupt.
6160 * If L1 intercepts external-interrupts, we should
6161 * exit from L2 to L1. Otherwise, interrupt should be
6162 * delivered directly to L2.
6164 if (is_guest_mode(vcpu) && max_irr_updated) {
6165 if (nested_exit_on_intr(vcpu))
6166 kvm_vcpu_exiting_guest_mode(vcpu);
6168 kvm_make_request(KVM_REQ_EVENT, vcpu);
6171 max_irr = kvm_lapic_find_highest_irr(vcpu);
6173 vmx_hwapic_irr_update(vcpu, max_irr);
6177 static bool vmx_dy_apicv_has_pending_interrupt(struct kvm_vcpu *vcpu)
6179 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
6181 return pi_test_on(pi_desc) ||
6182 (pi_test_sn(pi_desc) && !pi_is_pir_empty(pi_desc));
6185 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
6187 if (!kvm_vcpu_apicv_active(vcpu))
6190 vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
6191 vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
6192 vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
6193 vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
6196 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
6198 struct vcpu_vmx *vmx = to_vmx(vcpu);
6200 pi_clear_on(&vmx->pi_desc);
6201 memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
6204 static void handle_exception_nmi_irqoff(struct vcpu_vmx *vmx)
6206 vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6208 /* if exit due to PF check for async PF */
6209 if (is_page_fault(vmx->exit_intr_info)) {
6210 vmx->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason();
6211 /* Handle machine checks before interrupts are enabled */
6212 } else if (is_machine_check(vmx->exit_intr_info)) {
6213 kvm_machine_check();
6214 /* We need to handle NMIs before interrupts are enabled */
6215 } else if (is_nmi(vmx->exit_intr_info)) {
6216 kvm_before_interrupt(&vmx->vcpu);
6218 kvm_after_interrupt(&vmx->vcpu);
6222 static void handle_external_interrupt_irqoff(struct kvm_vcpu *vcpu)
6224 unsigned int vector;
6225 unsigned long entry;
6226 #ifdef CONFIG_X86_64
6232 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6233 if (WARN_ONCE(!is_external_intr(intr_info),
6234 "KVM: unexpected VM-Exit interrupt info: 0x%x", intr_info))
6237 vector = intr_info & INTR_INFO_VECTOR_MASK;
6238 desc = (gate_desc *)host_idt_base + vector;
6239 entry = gate_offset(desc);
6241 kvm_before_interrupt(vcpu);
6244 #ifdef CONFIG_X86_64
6245 "mov %%" _ASM_SP ", %[sp]\n\t"
6246 "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
6251 __ASM_SIZE(push) " $%c[cs]\n\t"
6254 #ifdef CONFIG_X86_64
6259 THUNK_TARGET(entry),
6260 [ss]"i"(__KERNEL_DS),
6261 [cs]"i"(__KERNEL_CS)
6264 kvm_after_interrupt(vcpu);
6266 STACK_FRAME_NON_STANDARD(handle_external_interrupt_irqoff);
6268 static void vmx_handle_exit_irqoff(struct kvm_vcpu *vcpu,
6269 enum exit_fastpath_completion *exit_fastpath)
6271 struct vcpu_vmx *vmx = to_vmx(vcpu);
6273 if (vmx->exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
6274 handle_external_interrupt_irqoff(vcpu);
6275 else if (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI)
6276 handle_exception_nmi_irqoff(vmx);
6277 else if (!is_guest_mode(vcpu) &&
6278 vmx->exit_reason == EXIT_REASON_MSR_WRITE)
6279 *exit_fastpath = handle_fastpath_set_msr_irqoff(vcpu);
6282 static bool vmx_has_emulated_msr(int index)
6285 case MSR_IA32_SMBASE:
6287 * We cannot do SMM unless we can run the guest in big
6290 return enable_unrestricted_guest || emulate_invalid_guest_state;
6291 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
6293 case MSR_AMD64_VIRT_SPEC_CTRL:
6294 /* This is AMD only. */
6301 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
6306 bool idtv_info_valid;
6308 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6311 if (vmx->loaded_vmcs->nmi_known_unmasked)
6314 * Can't use vmx->exit_intr_info since we're not sure what
6315 * the exit reason is.
6317 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6318 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
6319 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
6321 * SDM 3: 27.7.1.2 (September 2008)
6322 * Re-set bit "block by NMI" before VM entry if vmexit caused by
6323 * a guest IRET fault.
6324 * SDM 3: 23.2.2 (September 2008)
6325 * Bit 12 is undefined in any of the following cases:
6326 * If the VM exit sets the valid bit in the IDT-vectoring
6327 * information field.
6328 * If the VM exit is due to a double fault.
6330 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
6331 vector != DF_VECTOR && !idtv_info_valid)
6332 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
6333 GUEST_INTR_STATE_NMI);
6335 vmx->loaded_vmcs->nmi_known_unmasked =
6336 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
6337 & GUEST_INTR_STATE_NMI);
6338 } else if (unlikely(vmx->loaded_vmcs->soft_vnmi_blocked))
6339 vmx->loaded_vmcs->vnmi_blocked_time +=
6340 ktime_to_ns(ktime_sub(ktime_get(),
6341 vmx->loaded_vmcs->entry_time));
6344 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
6345 u32 idt_vectoring_info,
6346 int instr_len_field,
6347 int error_code_field)
6351 bool idtv_info_valid;
6353 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6355 vcpu->arch.nmi_injected = false;
6356 kvm_clear_exception_queue(vcpu);
6357 kvm_clear_interrupt_queue(vcpu);
6359 if (!idtv_info_valid)
6362 kvm_make_request(KVM_REQ_EVENT, vcpu);
6364 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
6365 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
6368 case INTR_TYPE_NMI_INTR:
6369 vcpu->arch.nmi_injected = true;
6371 * SDM 3: 27.7.1.2 (September 2008)
6372 * Clear bit "block by NMI" before VM entry if a NMI
6375 vmx_set_nmi_mask(vcpu, false);
6377 case INTR_TYPE_SOFT_EXCEPTION:
6378 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6380 case INTR_TYPE_HARD_EXCEPTION:
6381 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
6382 u32 err = vmcs_read32(error_code_field);
6383 kvm_requeue_exception_e(vcpu, vector, err);
6385 kvm_requeue_exception(vcpu, vector);
6387 case INTR_TYPE_SOFT_INTR:
6388 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6390 case INTR_TYPE_EXT_INTR:
6391 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
6398 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
6400 __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
6401 VM_EXIT_INSTRUCTION_LEN,
6402 IDT_VECTORING_ERROR_CODE);
6405 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
6407 __vmx_complete_interrupts(vcpu,
6408 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
6409 VM_ENTRY_INSTRUCTION_LEN,
6410 VM_ENTRY_EXCEPTION_ERROR_CODE);
6412 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
6415 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
6418 struct perf_guest_switch_msr *msrs;
6420 msrs = perf_guest_get_msrs(&nr_msrs);
6425 for (i = 0; i < nr_msrs; i++)
6426 if (msrs[i].host == msrs[i].guest)
6427 clear_atomic_switch_msr(vmx, msrs[i].msr);
6429 add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
6430 msrs[i].host, false);
6433 static void atomic_switch_umwait_control_msr(struct vcpu_vmx *vmx)
6435 u32 host_umwait_control;
6437 if (!vmx_has_waitpkg(vmx))
6440 host_umwait_control = get_umwait_control_msr();
6442 if (vmx->msr_ia32_umwait_control != host_umwait_control)
6443 add_atomic_switch_msr(vmx, MSR_IA32_UMWAIT_CONTROL,
6444 vmx->msr_ia32_umwait_control,
6445 host_umwait_control, false);
6447 clear_atomic_switch_msr(vmx, MSR_IA32_UMWAIT_CONTROL);
6450 static void vmx_update_hv_timer(struct kvm_vcpu *vcpu)
6452 struct vcpu_vmx *vmx = to_vmx(vcpu);
6456 if (vmx->req_immediate_exit) {
6457 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, 0);
6458 vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6459 } else if (vmx->hv_deadline_tsc != -1) {
6461 if (vmx->hv_deadline_tsc > tscl)
6462 /* set_hv_timer ensures the delta fits in 32-bits */
6463 delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
6464 cpu_preemption_timer_multi);
6468 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
6469 vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6470 } else if (!vmx->loaded_vmcs->hv_timer_soft_disabled) {
6471 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, -1);
6472 vmx->loaded_vmcs->hv_timer_soft_disabled = true;
6476 void vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp)
6478 if (unlikely(host_rsp != vmx->loaded_vmcs->host_state.rsp)) {
6479 vmx->loaded_vmcs->host_state.rsp = host_rsp;
6480 vmcs_writel(HOST_RSP, host_rsp);
6484 bool __vmx_vcpu_run(struct vcpu_vmx *vmx, unsigned long *regs, bool launched);
6486 static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
6488 struct vcpu_vmx *vmx = to_vmx(vcpu);
6489 unsigned long cr3, cr4;
6491 /* Record the guest's net vcpu time for enforced NMI injections. */
6492 if (unlikely(!enable_vnmi &&
6493 vmx->loaded_vmcs->soft_vnmi_blocked))
6494 vmx->loaded_vmcs->entry_time = ktime_get();
6496 /* Don't enter VMX if guest state is invalid, let the exit handler
6497 start emulation until we arrive back to a valid state */
6498 if (vmx->emulation_required)
6501 if (vmx->ple_window_dirty) {
6502 vmx->ple_window_dirty = false;
6503 vmcs_write32(PLE_WINDOW, vmx->ple_window);
6507 * We did this in prepare_switch_to_guest, because it needs to
6508 * be within srcu_read_lock.
6510 WARN_ON_ONCE(vmx->nested.need_vmcs12_to_shadow_sync);
6512 if (kvm_register_is_dirty(vcpu, VCPU_REGS_RSP))
6513 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
6514 if (kvm_register_is_dirty(vcpu, VCPU_REGS_RIP))
6515 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
6517 cr3 = __get_current_cr3_fast();
6518 if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
6519 vmcs_writel(HOST_CR3, cr3);
6520 vmx->loaded_vmcs->host_state.cr3 = cr3;
6523 cr4 = cr4_read_shadow();
6524 if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
6525 vmcs_writel(HOST_CR4, cr4);
6526 vmx->loaded_vmcs->host_state.cr4 = cr4;
6529 /* When single-stepping over STI and MOV SS, we must clear the
6530 * corresponding interruptibility bits in the guest state. Otherwise
6531 * vmentry fails as it then expects bit 14 (BS) in pending debug
6532 * exceptions being set, but that's not correct for the guest debugging
6534 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6535 vmx_set_interrupt_shadow(vcpu, 0);
6537 kvm_load_guest_xsave_state(vcpu);
6539 if (static_cpu_has(X86_FEATURE_PKU) &&
6540 kvm_read_cr4_bits(vcpu, X86_CR4_PKE) &&
6541 vcpu->arch.pkru != vmx->host_pkru)
6542 __write_pkru(vcpu->arch.pkru);
6544 pt_guest_enter(vmx);
6546 if (vcpu_to_pmu(vcpu)->version)
6547 atomic_switch_perf_msrs(vmx);
6548 atomic_switch_umwait_control_msr(vmx);
6550 if (enable_preemption_timer)
6551 vmx_update_hv_timer(vcpu);
6553 if (lapic_in_kernel(vcpu) &&
6554 vcpu->arch.apic->lapic_timer.timer_advance_ns)
6555 kvm_wait_lapic_expire(vcpu);
6558 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
6559 * it's non-zero. Since vmentry is serialising on affected CPUs, there
6560 * is no need to worry about the conditional branch over the wrmsr
6561 * being speculatively taken.
6563 x86_spec_ctrl_set_guest(vmx->spec_ctrl, 0);
6565 /* L1D Flush includes CPU buffer clear to mitigate MDS */
6566 if (static_branch_unlikely(&vmx_l1d_should_flush))
6567 vmx_l1d_flush(vcpu);
6568 else if (static_branch_unlikely(&mds_user_clear))
6569 mds_clear_cpu_buffers();
6571 if (vcpu->arch.cr2 != read_cr2())
6572 write_cr2(vcpu->arch.cr2);
6574 vmx->fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
6575 vmx->loaded_vmcs->launched);
6577 vcpu->arch.cr2 = read_cr2();
6580 * We do not use IBRS in the kernel. If this vCPU has used the
6581 * SPEC_CTRL MSR it may have left it on; save the value and
6582 * turn it off. This is much more efficient than blindly adding
6583 * it to the atomic save/restore list. Especially as the former
6584 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
6586 * For non-nested case:
6587 * If the L01 MSR bitmap does not intercept the MSR, then we need to
6591 * If the L02 MSR bitmap does not intercept the MSR, then we need to
6594 if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
6595 vmx->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
6597 x86_spec_ctrl_restore_host(vmx->spec_ctrl, 0);
6599 /* All fields are clean at this point */
6600 if (static_branch_unlikely(&enable_evmcs))
6601 current_evmcs->hv_clean_fields |=
6602 HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
6604 if (static_branch_unlikely(&enable_evmcs))
6605 current_evmcs->hv_vp_id = vcpu->arch.hyperv.vp_index;
6607 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
6608 if (vmx->host_debugctlmsr)
6609 update_debugctlmsr(vmx->host_debugctlmsr);
6611 #ifndef CONFIG_X86_64
6613 * The sysexit path does not restore ds/es, so we must set them to
6614 * a reasonable value ourselves.
6616 * We can't defer this to vmx_prepare_switch_to_host() since that
6617 * function may be executed in interrupt context, which saves and
6618 * restore segments around it, nullifying its effect.
6620 loadsegment(ds, __USER_DS);
6621 loadsegment(es, __USER_DS);
6624 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
6625 | (1 << VCPU_EXREG_RFLAGS)
6626 | (1 << VCPU_EXREG_PDPTR)
6627 | (1 << VCPU_EXREG_SEGMENTS)
6628 | (1 << VCPU_EXREG_CR3));
6629 vcpu->arch.regs_dirty = 0;
6634 * eager fpu is enabled if PKEY is supported and CR4 is switched
6635 * back on host, so it is safe to read guest PKRU from current
6638 if (static_cpu_has(X86_FEATURE_PKU) &&
6639 kvm_read_cr4_bits(vcpu, X86_CR4_PKE)) {
6640 vcpu->arch.pkru = rdpkru();
6641 if (vcpu->arch.pkru != vmx->host_pkru)
6642 __write_pkru(vmx->host_pkru);
6645 kvm_load_host_xsave_state(vcpu);
6647 vmx->nested.nested_run_pending = 0;
6648 vmx->idt_vectoring_info = 0;
6650 vmx->exit_reason = vmx->fail ? 0xdead : vmcs_read32(VM_EXIT_REASON);
6651 if ((u16)vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY)
6652 kvm_machine_check();
6654 if (vmx->fail || (vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
6657 vmx->loaded_vmcs->launched = 1;
6658 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
6660 vmx_recover_nmi_blocking(vmx);
6661 vmx_complete_interrupts(vmx);
6664 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
6666 struct vcpu_vmx *vmx = to_vmx(vcpu);
6669 vmx_destroy_pml_buffer(vmx);
6670 free_vpid(vmx->vpid);
6671 nested_vmx_free_vcpu(vcpu);
6672 free_loaded_vmcs(vmx->loaded_vmcs);
6675 static int vmx_create_vcpu(struct kvm_vcpu *vcpu)
6677 struct vcpu_vmx *vmx;
6678 unsigned long *msr_bitmap;
6681 BUILD_BUG_ON(offsetof(struct vcpu_vmx, vcpu) != 0);
6686 vmx->vpid = allocate_vpid();
6689 * If PML is turned on, failure on enabling PML just results in failure
6690 * of creating the vcpu, therefore we can simplify PML logic (by
6691 * avoiding dealing with cases, such as enabling PML partially on vcpus
6692 * for the guest), etc.
6695 vmx->pml_pg = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
6700 BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) != NR_SHARED_MSRS);
6702 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
6703 u32 index = vmx_msr_index[i];
6704 u32 data_low, data_high;
6707 if (rdmsr_safe(index, &data_low, &data_high) < 0)
6709 if (wrmsr_safe(index, data_low, data_high) < 0)
6712 vmx->guest_msrs[j].index = i;
6713 vmx->guest_msrs[j].data = 0;
6715 case MSR_IA32_TSX_CTRL:
6717 * No need to pass TSX_CTRL_CPUID_CLEAR through, so
6718 * let's avoid changing CPUID bits under the host
6721 vmx->guest_msrs[j].mask = ~(u64)TSX_CTRL_CPUID_CLEAR;
6724 vmx->guest_msrs[j].mask = -1ull;
6730 err = alloc_loaded_vmcs(&vmx->vmcs01);
6734 msr_bitmap = vmx->vmcs01.msr_bitmap;
6735 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_TSC, MSR_TYPE_R);
6736 vmx_disable_intercept_for_msr(msr_bitmap, MSR_FS_BASE, MSR_TYPE_RW);
6737 vmx_disable_intercept_for_msr(msr_bitmap, MSR_GS_BASE, MSR_TYPE_RW);
6738 vmx_disable_intercept_for_msr(msr_bitmap, MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
6739 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_CS, MSR_TYPE_RW);
6740 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_ESP, MSR_TYPE_RW);
6741 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_EIP, MSR_TYPE_RW);
6742 if (kvm_cstate_in_guest(vcpu->kvm)) {
6743 vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C1_RES, MSR_TYPE_R);
6744 vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C3_RESIDENCY, MSR_TYPE_R);
6745 vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C6_RESIDENCY, MSR_TYPE_R);
6746 vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C7_RESIDENCY, MSR_TYPE_R);
6748 vmx->msr_bitmap_mode = 0;
6750 vmx->loaded_vmcs = &vmx->vmcs01;
6752 vmx_vcpu_load(vcpu, cpu);
6757 if (cpu_need_virtualize_apic_accesses(vcpu)) {
6758 err = alloc_apic_access_page(vcpu->kvm);
6763 if (enable_ept && !enable_unrestricted_guest) {
6764 err = init_rmode_identity_map(vcpu->kvm);
6770 nested_vmx_setup_ctls_msrs(&vmx->nested.msrs,
6771 vmx_capability.ept);
6773 memset(&vmx->nested.msrs, 0, sizeof(vmx->nested.msrs));
6775 vmx->nested.posted_intr_nv = -1;
6776 vmx->nested.current_vmptr = -1ull;
6778 vcpu->arch.microcode_version = 0x100000000ULL;
6779 vmx->msr_ia32_feature_control_valid_bits = FEAT_CTL_LOCKED;
6782 * Enforce invariant: pi_desc.nv is always either POSTED_INTR_VECTOR
6783 * or POSTED_INTR_WAKEUP_VECTOR.
6785 vmx->pi_desc.nv = POSTED_INTR_VECTOR;
6786 vmx->pi_desc.sn = 1;
6788 vmx->ept_pointer = INVALID_PAGE;
6793 free_loaded_vmcs(vmx->loaded_vmcs);
6795 vmx_destroy_pml_buffer(vmx);
6797 free_vpid(vmx->vpid);
6801 #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"
6802 #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"
6804 static int vmx_vm_init(struct kvm *kvm)
6806 spin_lock_init(&to_kvm_vmx(kvm)->ept_pointer_lock);
6809 kvm->arch.pause_in_guest = true;
6811 if (boot_cpu_has(X86_BUG_L1TF) && enable_ept) {
6812 switch (l1tf_mitigation) {
6813 case L1TF_MITIGATION_OFF:
6814 case L1TF_MITIGATION_FLUSH_NOWARN:
6815 /* 'I explicitly don't care' is set */
6817 case L1TF_MITIGATION_FLUSH:
6818 case L1TF_MITIGATION_FLUSH_NOSMT:
6819 case L1TF_MITIGATION_FULL:
6821 * Warn upon starting the first VM in a potentially
6822 * insecure environment.
6824 if (sched_smt_active())
6825 pr_warn_once(L1TF_MSG_SMT);
6826 if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER)
6827 pr_warn_once(L1TF_MSG_L1D);
6829 case L1TF_MITIGATION_FULL_FORCE:
6830 /* Flush is enforced */
6834 kvm_apicv_init(kvm, enable_apicv);
6838 static int __init vmx_check_processor_compat(void)
6840 struct vmcs_config vmcs_conf;
6841 struct vmx_capability vmx_cap;
6843 if (!this_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
6844 !this_cpu_has(X86_FEATURE_VMX)) {
6845 pr_err("kvm: VMX is disabled on CPU %d\n", smp_processor_id());
6849 if (setup_vmcs_config(&vmcs_conf, &vmx_cap) < 0)
6852 nested_vmx_setup_ctls_msrs(&vmcs_conf.nested, vmx_cap.ept);
6853 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
6854 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
6855 smp_processor_id());
6861 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
6866 /* We wanted to honor guest CD/MTRR/PAT, but doing so could result in
6867 * memory aliases with conflicting memory types and sometimes MCEs.
6868 * We have to be careful as to what are honored and when.
6870 * For MMIO, guest CD/MTRR are ignored. The EPT memory type is set to
6871 * UC. The effective memory type is UC or WC depending on guest PAT.
6872 * This was historically the source of MCEs and we want to be
6875 * When there is no need to deal with noncoherent DMA (e.g., no VT-d
6876 * or VT-d has snoop control), guest CD/MTRR/PAT are all ignored. The
6877 * EPT memory type is set to WB. The effective memory type is forced
6880 * Otherwise, we trust guest. Guest CD/MTRR/PAT are all honored. The
6881 * EPT memory type is used to emulate guest CD/MTRR.
6885 cache = MTRR_TYPE_UNCACHABLE;
6889 if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
6890 ipat = VMX_EPT_IPAT_BIT;
6891 cache = MTRR_TYPE_WRBACK;
6895 if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
6896 ipat = VMX_EPT_IPAT_BIT;
6897 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
6898 cache = MTRR_TYPE_WRBACK;
6900 cache = MTRR_TYPE_UNCACHABLE;
6904 cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
6907 return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
6910 static void vmcs_set_secondary_exec_control(struct vcpu_vmx *vmx)
6913 * These bits in the secondary execution controls field
6914 * are dynamic, the others are mostly based on the hypervisor
6915 * architecture and the guest's CPUID. Do not touch the
6919 SECONDARY_EXEC_SHADOW_VMCS |
6920 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
6921 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6922 SECONDARY_EXEC_DESC;
6924 u32 new_ctl = vmx->secondary_exec_control;
6925 u32 cur_ctl = secondary_exec_controls_get(vmx);
6927 secondary_exec_controls_set(vmx, (new_ctl & ~mask) | (cur_ctl & mask));
6931 * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
6932 * (indicating "allowed-1") if they are supported in the guest's CPUID.
6934 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
6936 struct vcpu_vmx *vmx = to_vmx(vcpu);
6937 struct kvm_cpuid_entry2 *entry;
6939 vmx->nested.msrs.cr0_fixed1 = 0xffffffff;
6940 vmx->nested.msrs.cr4_fixed1 = X86_CR4_PCE;
6942 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do { \
6943 if (entry && (entry->_reg & (_cpuid_mask))) \
6944 vmx->nested.msrs.cr4_fixed1 |= (_cr4_mask); \
6947 entry = kvm_find_cpuid_entry(vcpu, 0x1, 0);
6948 cr4_fixed1_update(X86_CR4_VME, edx, feature_bit(VME));
6949 cr4_fixed1_update(X86_CR4_PVI, edx, feature_bit(VME));
6950 cr4_fixed1_update(X86_CR4_TSD, edx, feature_bit(TSC));
6951 cr4_fixed1_update(X86_CR4_DE, edx, feature_bit(DE));
6952 cr4_fixed1_update(X86_CR4_PSE, edx, feature_bit(PSE));
6953 cr4_fixed1_update(X86_CR4_PAE, edx, feature_bit(PAE));
6954 cr4_fixed1_update(X86_CR4_MCE, edx, feature_bit(MCE));
6955 cr4_fixed1_update(X86_CR4_PGE, edx, feature_bit(PGE));
6956 cr4_fixed1_update(X86_CR4_OSFXSR, edx, feature_bit(FXSR));
6957 cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, feature_bit(XMM));
6958 cr4_fixed1_update(X86_CR4_VMXE, ecx, feature_bit(VMX));
6959 cr4_fixed1_update(X86_CR4_SMXE, ecx, feature_bit(SMX));
6960 cr4_fixed1_update(X86_CR4_PCIDE, ecx, feature_bit(PCID));
6961 cr4_fixed1_update(X86_CR4_OSXSAVE, ecx, feature_bit(XSAVE));
6963 entry = kvm_find_cpuid_entry(vcpu, 0x7, 0);
6964 cr4_fixed1_update(X86_CR4_FSGSBASE, ebx, feature_bit(FSGSBASE));
6965 cr4_fixed1_update(X86_CR4_SMEP, ebx, feature_bit(SMEP));
6966 cr4_fixed1_update(X86_CR4_SMAP, ebx, feature_bit(SMAP));
6967 cr4_fixed1_update(X86_CR4_PKE, ecx, feature_bit(PKU));
6968 cr4_fixed1_update(X86_CR4_UMIP, ecx, feature_bit(UMIP));
6969 cr4_fixed1_update(X86_CR4_LA57, ecx, feature_bit(LA57));
6971 #undef cr4_fixed1_update
6974 static void nested_vmx_entry_exit_ctls_update(struct kvm_vcpu *vcpu)
6976 struct vcpu_vmx *vmx = to_vmx(vcpu);
6978 if (kvm_mpx_supported()) {
6979 bool mpx_enabled = guest_cpuid_has(vcpu, X86_FEATURE_MPX);
6982 vmx->nested.msrs.entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
6983 vmx->nested.msrs.exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
6985 vmx->nested.msrs.entry_ctls_high &= ~VM_ENTRY_LOAD_BNDCFGS;
6986 vmx->nested.msrs.exit_ctls_high &= ~VM_EXIT_CLEAR_BNDCFGS;
6991 static void update_intel_pt_cfg(struct kvm_vcpu *vcpu)
6993 struct vcpu_vmx *vmx = to_vmx(vcpu);
6994 struct kvm_cpuid_entry2 *best = NULL;
6997 for (i = 0; i < PT_CPUID_LEAVES; i++) {
6998 best = kvm_find_cpuid_entry(vcpu, 0x14, i);
7001 vmx->pt_desc.caps[CPUID_EAX + i*PT_CPUID_REGS_NUM] = best->eax;
7002 vmx->pt_desc.caps[CPUID_EBX + i*PT_CPUID_REGS_NUM] = best->ebx;
7003 vmx->pt_desc.caps[CPUID_ECX + i*PT_CPUID_REGS_NUM] = best->ecx;
7004 vmx->pt_desc.caps[CPUID_EDX + i*PT_CPUID_REGS_NUM] = best->edx;
7007 /* Get the number of configurable Address Ranges for filtering */
7008 vmx->pt_desc.addr_range = intel_pt_validate_cap(vmx->pt_desc.caps,
7009 PT_CAP_num_address_ranges);
7011 /* Initialize and clear the no dependency bits */
7012 vmx->pt_desc.ctl_bitmask = ~(RTIT_CTL_TRACEEN | RTIT_CTL_OS |
7013 RTIT_CTL_USR | RTIT_CTL_TSC_EN | RTIT_CTL_DISRETC);
7016 * If CPUID.(EAX=14H,ECX=0):EBX[0]=1 CR3Filter can be set otherwise
7017 * will inject an #GP
7019 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_cr3_filtering))
7020 vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_CR3EN;
7023 * If CPUID.(EAX=14H,ECX=0):EBX[1]=1 CYCEn, CycThresh and
7024 * PSBFreq can be set
7026 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc))
7027 vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_CYCLEACC |
7028 RTIT_CTL_CYC_THRESH | RTIT_CTL_PSB_FREQ);
7031 * If CPUID.(EAX=14H,ECX=0):EBX[3]=1 MTCEn BranchEn and
7032 * MTCFreq can be set
7034 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc))
7035 vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_MTC_EN |
7036 RTIT_CTL_BRANCH_EN | RTIT_CTL_MTC_RANGE);
7038 /* If CPUID.(EAX=14H,ECX=0):EBX[4]=1 FUPonPTW and PTWEn can be set */
7039 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_ptwrite))
7040 vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_FUP_ON_PTW |
7043 /* If CPUID.(EAX=14H,ECX=0):EBX[5]=1 PwrEvEn can be set */
7044 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_power_event_trace))
7045 vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_PWR_EVT_EN;
7047 /* If CPUID.(EAX=14H,ECX=0):ECX[0]=1 ToPA can be set */
7048 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_topa_output))
7049 vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_TOPA;
7051 /* If CPUID.(EAX=14H,ECX=0):ECX[3]=1 FabircEn can be set */
7052 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_output_subsys))
7053 vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_FABRIC_EN;
7055 /* unmask address range configure area */
7056 for (i = 0; i < vmx->pt_desc.addr_range; i++)
7057 vmx->pt_desc.ctl_bitmask &= ~(0xfULL << (32 + i * 4));
7060 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
7062 struct vcpu_vmx *vmx = to_vmx(vcpu);
7064 /* xsaves_enabled is recomputed in vmx_compute_secondary_exec_control(). */
7065 vcpu->arch.xsaves_enabled = false;
7067 if (cpu_has_secondary_exec_ctrls()) {
7068 vmx_compute_secondary_exec_control(vmx);
7069 vmcs_set_secondary_exec_control(vmx);
7072 if (nested_vmx_allowed(vcpu))
7073 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7074 FEAT_CTL_VMX_ENABLED_INSIDE_SMX |
7075 FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX;
7077 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7078 ~(FEAT_CTL_VMX_ENABLED_INSIDE_SMX |
7079 FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX);
7081 if (nested_vmx_allowed(vcpu)) {
7082 nested_vmx_cr_fixed1_bits_update(vcpu);
7083 nested_vmx_entry_exit_ctls_update(vcpu);
7086 if (boot_cpu_has(X86_FEATURE_INTEL_PT) &&
7087 guest_cpuid_has(vcpu, X86_FEATURE_INTEL_PT))
7088 update_intel_pt_cfg(vcpu);
7090 if (boot_cpu_has(X86_FEATURE_RTM)) {
7091 struct shared_msr_entry *msr;
7092 msr = find_msr_entry(vmx, MSR_IA32_TSX_CTRL);
7094 bool enabled = guest_cpuid_has(vcpu, X86_FEATURE_RTM);
7095 vmx_set_guest_msr(vmx, msr, enabled ? 0 : TSX_CTRL_RTM_DISABLE);
7100 static __init void vmx_set_cpu_caps(void)
7106 kvm_cpu_cap_set(X86_FEATURE_VMX);
7109 if (kvm_mpx_supported())
7110 kvm_cpu_cap_check_and_set(X86_FEATURE_MPX);
7111 if (cpu_has_vmx_invpcid())
7112 kvm_cpu_cap_check_and_set(X86_FEATURE_INVPCID);
7113 if (vmx_pt_mode_is_host_guest())
7114 kvm_cpu_cap_check_and_set(X86_FEATURE_INTEL_PT);
7116 /* PKU is not yet implemented for shadow paging. */
7117 if (enable_ept && boot_cpu_has(X86_FEATURE_OSPKE))
7118 kvm_cpu_cap_check_and_set(X86_FEATURE_PKU);
7120 if (vmx_umip_emulated())
7121 kvm_cpu_cap_set(X86_FEATURE_UMIP);
7125 if (!vmx_xsaves_supported())
7126 kvm_cpu_cap_clear(X86_FEATURE_XSAVES);
7128 /* CPUID 0x80000001 */
7129 if (!cpu_has_vmx_rdtscp())
7130 kvm_cpu_cap_clear(X86_FEATURE_RDTSCP);
7133 static void vmx_request_immediate_exit(struct kvm_vcpu *vcpu)
7135 to_vmx(vcpu)->req_immediate_exit = true;
7138 static int vmx_check_intercept_io(struct kvm_vcpu *vcpu,
7139 struct x86_instruction_info *info)
7141 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7142 unsigned short port;
7146 if (info->intercept == x86_intercept_in ||
7147 info->intercept == x86_intercept_ins) {
7148 port = info->src_val;
7149 size = info->dst_bytes;
7151 port = info->dst_val;
7152 size = info->src_bytes;
7156 * If the 'use IO bitmaps' VM-execution control is 0, IO instruction
7157 * VM-exits depend on the 'unconditional IO exiting' VM-execution
7160 * Otherwise, IO instruction VM-exits are controlled by the IO bitmaps.
7162 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
7163 intercept = nested_cpu_has(vmcs12,
7164 CPU_BASED_UNCOND_IO_EXITING);
7166 intercept = nested_vmx_check_io_bitmaps(vcpu, port, size);
7168 /* FIXME: produce nested vmexit and return X86EMUL_INTERCEPTED. */
7169 return intercept ? X86EMUL_UNHANDLEABLE : X86EMUL_CONTINUE;
7172 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
7173 struct x86_instruction_info *info,
7174 enum x86_intercept_stage stage,
7175 struct x86_exception *exception)
7177 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7179 switch (info->intercept) {
7181 * RDPID causes #UD if disabled through secondary execution controls.
7182 * Because it is marked as EmulateOnUD, we need to intercept it here.
7184 case x86_intercept_rdtscp:
7185 if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDTSCP)) {
7186 exception->vector = UD_VECTOR;
7187 exception->error_code_valid = false;
7188 return X86EMUL_PROPAGATE_FAULT;
7192 case x86_intercept_in:
7193 case x86_intercept_ins:
7194 case x86_intercept_out:
7195 case x86_intercept_outs:
7196 return vmx_check_intercept_io(vcpu, info);
7198 case x86_intercept_lgdt:
7199 case x86_intercept_lidt:
7200 case x86_intercept_lldt:
7201 case x86_intercept_ltr:
7202 case x86_intercept_sgdt:
7203 case x86_intercept_sidt:
7204 case x86_intercept_sldt:
7205 case x86_intercept_str:
7206 if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC))
7207 return X86EMUL_CONTINUE;
7209 /* FIXME: produce nested vmexit and return X86EMUL_INTERCEPTED. */
7212 /* TODO: check more intercepts... */
7217 return X86EMUL_UNHANDLEABLE;
7220 #ifdef CONFIG_X86_64
7221 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
7222 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
7223 u64 divisor, u64 *result)
7225 u64 low = a << shift, high = a >> (64 - shift);
7227 /* To avoid the overflow on divq */
7228 if (high >= divisor)
7231 /* Low hold the result, high hold rem which is discarded */
7232 asm("divq %2\n\t" : "=a" (low), "=d" (high) :
7233 "rm" (divisor), "0" (low), "1" (high));
7239 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc,
7242 struct vcpu_vmx *vmx;
7243 u64 tscl, guest_tscl, delta_tsc, lapic_timer_advance_cycles;
7244 struct kvm_timer *ktimer = &vcpu->arch.apic->lapic_timer;
7246 if (kvm_mwait_in_guest(vcpu->kvm) ||
7247 kvm_can_post_timer_interrupt(vcpu))
7252 guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
7253 delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
7254 lapic_timer_advance_cycles = nsec_to_cycles(vcpu,
7255 ktimer->timer_advance_ns);
7257 if (delta_tsc > lapic_timer_advance_cycles)
7258 delta_tsc -= lapic_timer_advance_cycles;
7262 /* Convert to host delta tsc if tsc scaling is enabled */
7263 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
7264 delta_tsc && u64_shl_div_u64(delta_tsc,
7265 kvm_tsc_scaling_ratio_frac_bits,
7266 vcpu->arch.tsc_scaling_ratio, &delta_tsc))
7270 * If the delta tsc can't fit in the 32 bit after the multi shift,
7271 * we can't use the preemption timer.
7272 * It's possible that it fits on later vmentries, but checking
7273 * on every vmentry is costly so we just use an hrtimer.
7275 if (delta_tsc >> (cpu_preemption_timer_multi + 32))
7278 vmx->hv_deadline_tsc = tscl + delta_tsc;
7279 *expired = !delta_tsc;
7283 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
7285 to_vmx(vcpu)->hv_deadline_tsc = -1;
7289 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
7291 if (!kvm_pause_in_guest(vcpu->kvm))
7292 shrink_ple_window(vcpu);
7295 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
7296 struct kvm_memory_slot *slot)
7298 if (!kvm_dirty_log_manual_protect_and_init_set(kvm))
7299 kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
7300 kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
7303 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
7304 struct kvm_memory_slot *slot)
7306 kvm_mmu_slot_set_dirty(kvm, slot);
7309 static void vmx_flush_log_dirty(struct kvm *kvm)
7311 kvm_flush_pml_buffers(kvm);
7314 static int vmx_write_pml_buffer(struct kvm_vcpu *vcpu)
7316 struct vmcs12 *vmcs12;
7317 struct vcpu_vmx *vmx = to_vmx(vcpu);
7320 if (is_guest_mode(vcpu)) {
7321 WARN_ON_ONCE(vmx->nested.pml_full);
7324 * Check if PML is enabled for the nested guest.
7325 * Whether eptp bit 6 is set is already checked
7326 * as part of A/D emulation.
7328 vmcs12 = get_vmcs12(vcpu);
7329 if (!nested_cpu_has_pml(vmcs12))
7332 if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
7333 vmx->nested.pml_full = true;
7337 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS) & ~0xFFFull;
7338 dst = vmcs12->pml_address + sizeof(u64) * vmcs12->guest_pml_index;
7340 if (kvm_write_guest_page(vcpu->kvm, gpa_to_gfn(dst), &gpa,
7341 offset_in_page(dst), sizeof(gpa)))
7344 vmcs12->guest_pml_index--;
7350 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
7351 struct kvm_memory_slot *memslot,
7352 gfn_t offset, unsigned long mask)
7354 kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
7357 static void __pi_post_block(struct kvm_vcpu *vcpu)
7359 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7360 struct pi_desc old, new;
7364 old.control = new.control = pi_desc->control;
7365 WARN(old.nv != POSTED_INTR_WAKEUP_VECTOR,
7366 "Wakeup handler not enabled while the VCPU is blocked\n");
7368 dest = cpu_physical_id(vcpu->cpu);
7370 if (x2apic_enabled())
7373 new.ndst = (dest << 8) & 0xFF00;
7375 /* set 'NV' to 'notification vector' */
7376 new.nv = POSTED_INTR_VECTOR;
7377 } while (cmpxchg64(&pi_desc->control, old.control,
7378 new.control) != old.control);
7380 if (!WARN_ON_ONCE(vcpu->pre_pcpu == -1)) {
7381 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7382 list_del(&vcpu->blocked_vcpu_list);
7383 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7384 vcpu->pre_pcpu = -1;
7389 * This routine does the following things for vCPU which is going
7390 * to be blocked if VT-d PI is enabled.
7391 * - Store the vCPU to the wakeup list, so when interrupts happen
7392 * we can find the right vCPU to wake up.
7393 * - Change the Posted-interrupt descriptor as below:
7394 * 'NDST' <-- vcpu->pre_pcpu
7395 * 'NV' <-- POSTED_INTR_WAKEUP_VECTOR
7396 * - If 'ON' is set during this process, which means at least one
7397 * interrupt is posted for this vCPU, we cannot block it, in
7398 * this case, return 1, otherwise, return 0.
7401 static int pi_pre_block(struct kvm_vcpu *vcpu)
7404 struct pi_desc old, new;
7405 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7407 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
7408 !irq_remapping_cap(IRQ_POSTING_CAP) ||
7409 !kvm_vcpu_apicv_active(vcpu))
7412 WARN_ON(irqs_disabled());
7413 local_irq_disable();
7414 if (!WARN_ON_ONCE(vcpu->pre_pcpu != -1)) {
7415 vcpu->pre_pcpu = vcpu->cpu;
7416 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7417 list_add_tail(&vcpu->blocked_vcpu_list,
7418 &per_cpu(blocked_vcpu_on_cpu,
7420 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7424 old.control = new.control = pi_desc->control;
7426 WARN((pi_desc->sn == 1),
7427 "Warning: SN field of posted-interrupts "
7428 "is set before blocking\n");
7431 * Since vCPU can be preempted during this process,
7432 * vcpu->cpu could be different with pre_pcpu, we
7433 * need to set pre_pcpu as the destination of wakeup
7434 * notification event, then we can find the right vCPU
7435 * to wakeup in wakeup handler if interrupts happen
7436 * when the vCPU is in blocked state.
7438 dest = cpu_physical_id(vcpu->pre_pcpu);
7440 if (x2apic_enabled())
7443 new.ndst = (dest << 8) & 0xFF00;
7445 /* set 'NV' to 'wakeup vector' */
7446 new.nv = POSTED_INTR_WAKEUP_VECTOR;
7447 } while (cmpxchg64(&pi_desc->control, old.control,
7448 new.control) != old.control);
7450 /* We should not block the vCPU if an interrupt is posted for it. */
7451 if (pi_test_on(pi_desc) == 1)
7452 __pi_post_block(vcpu);
7455 return (vcpu->pre_pcpu == -1);
7458 static int vmx_pre_block(struct kvm_vcpu *vcpu)
7460 if (pi_pre_block(vcpu))
7463 if (kvm_lapic_hv_timer_in_use(vcpu))
7464 kvm_lapic_switch_to_sw_timer(vcpu);
7469 static void pi_post_block(struct kvm_vcpu *vcpu)
7471 if (vcpu->pre_pcpu == -1)
7474 WARN_ON(irqs_disabled());
7475 local_irq_disable();
7476 __pi_post_block(vcpu);
7480 static void vmx_post_block(struct kvm_vcpu *vcpu)
7482 if (kvm_x86_ops->set_hv_timer)
7483 kvm_lapic_switch_to_hv_timer(vcpu);
7485 pi_post_block(vcpu);
7489 * vmx_update_pi_irte - set IRTE for Posted-Interrupts
7492 * @host_irq: host irq of the interrupt
7493 * @guest_irq: gsi of the interrupt
7494 * @set: set or unset PI
7495 * returns 0 on success, < 0 on failure
7497 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
7498 uint32_t guest_irq, bool set)
7500 struct kvm_kernel_irq_routing_entry *e;
7501 struct kvm_irq_routing_table *irq_rt;
7502 struct kvm_lapic_irq irq;
7503 struct kvm_vcpu *vcpu;
7504 struct vcpu_data vcpu_info;
7507 if (!kvm_arch_has_assigned_device(kvm) ||
7508 !irq_remapping_cap(IRQ_POSTING_CAP) ||
7509 !kvm_vcpu_apicv_active(kvm->vcpus[0]))
7512 idx = srcu_read_lock(&kvm->irq_srcu);
7513 irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
7514 if (guest_irq >= irq_rt->nr_rt_entries ||
7515 hlist_empty(&irq_rt->map[guest_irq])) {
7516 pr_warn_once("no route for guest_irq %u/%u (broken user space?)\n",
7517 guest_irq, irq_rt->nr_rt_entries);
7521 hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
7522 if (e->type != KVM_IRQ_ROUTING_MSI)
7525 * VT-d PI cannot support posting multicast/broadcast
7526 * interrupts to a vCPU, we still use interrupt remapping
7527 * for these kind of interrupts.
7529 * For lowest-priority interrupts, we only support
7530 * those with single CPU as the destination, e.g. user
7531 * configures the interrupts via /proc/irq or uses
7532 * irqbalance to make the interrupts single-CPU.
7534 * We will support full lowest-priority interrupt later.
7536 * In addition, we can only inject generic interrupts using
7537 * the PI mechanism, refuse to route others through it.
7540 kvm_set_msi_irq(kvm, e, &irq);
7541 if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu) ||
7542 !kvm_irq_is_postable(&irq)) {
7544 * Make sure the IRTE is in remapped mode if
7545 * we don't handle it in posted mode.
7547 ret = irq_set_vcpu_affinity(host_irq, NULL);
7550 "failed to back to remapped mode, irq: %u\n",
7558 vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
7559 vcpu_info.vector = irq.vector;
7561 trace_kvm_pi_irte_update(host_irq, vcpu->vcpu_id, e->gsi,
7562 vcpu_info.vector, vcpu_info.pi_desc_addr, set);
7565 ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
7567 ret = irq_set_vcpu_affinity(host_irq, NULL);
7570 printk(KERN_INFO "%s: failed to update PI IRTE\n",
7578 srcu_read_unlock(&kvm->irq_srcu, idx);
7582 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
7584 if (vcpu->arch.mcg_cap & MCG_LMCE_P)
7585 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7586 FEAT_CTL_LMCE_ENABLED;
7588 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7589 ~FEAT_CTL_LMCE_ENABLED;
7592 static int vmx_smi_allowed(struct kvm_vcpu *vcpu)
7594 /* we need a nested vmexit to enter SMM, postpone if run is pending */
7595 if (to_vmx(vcpu)->nested.nested_run_pending)
7600 static int vmx_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
7602 struct vcpu_vmx *vmx = to_vmx(vcpu);
7604 vmx->nested.smm.guest_mode = is_guest_mode(vcpu);
7605 if (vmx->nested.smm.guest_mode)
7606 nested_vmx_vmexit(vcpu, -1, 0, 0);
7608 vmx->nested.smm.vmxon = vmx->nested.vmxon;
7609 vmx->nested.vmxon = false;
7610 vmx_clear_hlt(vcpu);
7614 static int vmx_pre_leave_smm(struct kvm_vcpu *vcpu, const char *smstate)
7616 struct vcpu_vmx *vmx = to_vmx(vcpu);
7619 if (vmx->nested.smm.vmxon) {
7620 vmx->nested.vmxon = true;
7621 vmx->nested.smm.vmxon = false;
7624 if (vmx->nested.smm.guest_mode) {
7625 ret = nested_vmx_enter_non_root_mode(vcpu, false);
7629 vmx->nested.smm.guest_mode = false;
7634 static int enable_smi_window(struct kvm_vcpu *vcpu)
7639 static bool vmx_need_emulation_on_page_fault(struct kvm_vcpu *vcpu)
7644 static bool vmx_apic_init_signal_blocked(struct kvm_vcpu *vcpu)
7646 return to_vmx(vcpu)->nested.vmxon;
7649 static __init int hardware_setup(void)
7651 unsigned long host_bndcfgs;
7653 int r, i, ept_lpage_level;
7656 host_idt_base = dt.address;
7658 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
7659 kvm_define_shared_msr(i, vmx_msr_index[i]);
7661 if (setup_vmcs_config(&vmcs_config, &vmx_capability) < 0)
7664 if (boot_cpu_has(X86_FEATURE_NX))
7665 kvm_enable_efer_bits(EFER_NX);
7667 if (boot_cpu_has(X86_FEATURE_MPX)) {
7668 rdmsrl(MSR_IA32_BNDCFGS, host_bndcfgs);
7669 WARN_ONCE(host_bndcfgs, "KVM: BNDCFGS in host will be lost");
7672 if (!cpu_has_vmx_mpx())
7673 supported_xcr0 &= ~(XFEATURE_MASK_BNDREGS |
7674 XFEATURE_MASK_BNDCSR);
7676 if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
7677 !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
7680 if (!cpu_has_vmx_ept() ||
7681 !cpu_has_vmx_ept_4levels() ||
7682 !cpu_has_vmx_ept_mt_wb() ||
7683 !cpu_has_vmx_invept_global())
7686 if (!cpu_has_vmx_ept_ad_bits() || !enable_ept)
7687 enable_ept_ad_bits = 0;
7689 if (!cpu_has_vmx_unrestricted_guest() || !enable_ept)
7690 enable_unrestricted_guest = 0;
7692 if (!cpu_has_vmx_flexpriority())
7693 flexpriority_enabled = 0;
7695 if (!cpu_has_virtual_nmis())
7699 * set_apic_access_page_addr() is used to reload apic access
7700 * page upon invalidation. No need to do anything if not
7701 * using the APIC_ACCESS_ADDR VMCS field.
7703 if (!flexpriority_enabled)
7704 kvm_x86_ops->set_apic_access_page_addr = NULL;
7706 if (!cpu_has_vmx_tpr_shadow())
7707 kvm_x86_ops->update_cr8_intercept = NULL;
7709 #if IS_ENABLED(CONFIG_HYPERV)
7710 if (ms_hyperv.nested_features & HV_X64_NESTED_GUEST_MAPPING_FLUSH
7712 kvm_x86_ops->tlb_remote_flush = hv_remote_flush_tlb;
7713 kvm_x86_ops->tlb_remote_flush_with_range =
7714 hv_remote_flush_tlb_with_range;
7718 if (!cpu_has_vmx_ple()) {
7721 ple_window_grow = 0;
7723 ple_window_shrink = 0;
7726 if (!cpu_has_vmx_apicv()) {
7728 kvm_x86_ops->sync_pir_to_irr = NULL;
7731 if (cpu_has_vmx_tsc_scaling()) {
7732 kvm_has_tsc_control = true;
7733 kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
7734 kvm_tsc_scaling_ratio_frac_bits = 48;
7737 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
7743 ept_lpage_level = 0;
7744 else if (cpu_has_vmx_ept_1g_page())
7745 ept_lpage_level = PT_PDPE_LEVEL;
7746 else if (cpu_has_vmx_ept_2m_page())
7747 ept_lpage_level = PT_DIRECTORY_LEVEL;
7749 ept_lpage_level = PT_PAGE_TABLE_LEVEL;
7750 kvm_configure_mmu(enable_ept, ept_lpage_level);
7753 * Only enable PML when hardware supports PML feature, and both EPT
7754 * and EPT A/D bit features are enabled -- PML depends on them to work.
7756 if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
7760 kvm_x86_ops->slot_enable_log_dirty = NULL;
7761 kvm_x86_ops->slot_disable_log_dirty = NULL;
7762 kvm_x86_ops->flush_log_dirty = NULL;
7763 kvm_x86_ops->enable_log_dirty_pt_masked = NULL;
7766 if (!cpu_has_vmx_preemption_timer())
7767 enable_preemption_timer = false;
7769 if (enable_preemption_timer) {
7770 u64 use_timer_freq = 5000ULL * 1000 * 1000;
7773 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
7774 cpu_preemption_timer_multi =
7775 vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
7778 use_timer_freq = (u64)tsc_khz * 1000;
7779 use_timer_freq >>= cpu_preemption_timer_multi;
7782 * KVM "disables" the preemption timer by setting it to its max
7783 * value. Don't use the timer if it might cause spurious exits
7784 * at a rate faster than 0.1 Hz (of uninterrupted guest time).
7786 if (use_timer_freq > 0xffffffffu / 10)
7787 enable_preemption_timer = false;
7790 if (!enable_preemption_timer) {
7791 kvm_x86_ops->set_hv_timer = NULL;
7792 kvm_x86_ops->cancel_hv_timer = NULL;
7793 kvm_x86_ops->request_immediate_exit = __kvm_request_immediate_exit;
7796 kvm_set_posted_intr_wakeup_handler(wakeup_handler);
7798 kvm_mce_cap_supported |= MCG_LMCE_P;
7800 if (pt_mode != PT_MODE_SYSTEM && pt_mode != PT_MODE_HOST_GUEST)
7802 if (!enable_ept || !cpu_has_vmx_intel_pt())
7803 pt_mode = PT_MODE_SYSTEM;
7806 nested_vmx_setup_ctls_msrs(&vmcs_config.nested,
7807 vmx_capability.ept);
7809 r = nested_vmx_hardware_setup(kvm_vmx_exit_handlers);
7816 r = alloc_kvm_area();
7818 nested_vmx_hardware_unsetup();
7822 static __exit void hardware_unsetup(void)
7825 nested_vmx_hardware_unsetup();
7830 static bool vmx_check_apicv_inhibit_reasons(ulong bit)
7832 ulong supported = BIT(APICV_INHIBIT_REASON_DISABLE) |
7833 BIT(APICV_INHIBIT_REASON_HYPERV);
7835 return supported & BIT(bit);
7838 static struct kvm_x86_ops vmx_x86_ops __ro_after_init = {
7839 .cpu_has_kvm_support = cpu_has_kvm_support,
7840 .disabled_by_bios = vmx_disabled_by_bios,
7841 .hardware_setup = hardware_setup,
7842 .hardware_unsetup = hardware_unsetup,
7843 .check_processor_compatibility = vmx_check_processor_compat,
7844 .hardware_enable = hardware_enable,
7845 .hardware_disable = hardware_disable,
7846 .cpu_has_accelerated_tpr = report_flexpriority,
7847 .has_emulated_msr = vmx_has_emulated_msr,
7849 .vm_size = sizeof(struct kvm_vmx),
7850 .vm_init = vmx_vm_init,
7852 .vcpu_create = vmx_create_vcpu,
7853 .vcpu_free = vmx_free_vcpu,
7854 .vcpu_reset = vmx_vcpu_reset,
7856 .prepare_guest_switch = vmx_prepare_switch_to_guest,
7857 .vcpu_load = vmx_vcpu_load,
7858 .vcpu_put = vmx_vcpu_put,
7860 .update_bp_intercept = update_exception_bitmap,
7861 .get_msr_feature = vmx_get_msr_feature,
7862 .get_msr = vmx_get_msr,
7863 .set_msr = vmx_set_msr,
7864 .get_segment_base = vmx_get_segment_base,
7865 .get_segment = vmx_get_segment,
7866 .set_segment = vmx_set_segment,
7867 .get_cpl = vmx_get_cpl,
7868 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
7869 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
7870 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
7871 .set_cr0 = vmx_set_cr0,
7872 .set_cr4 = vmx_set_cr4,
7873 .set_efer = vmx_set_efer,
7874 .get_idt = vmx_get_idt,
7875 .set_idt = vmx_set_idt,
7876 .get_gdt = vmx_get_gdt,
7877 .set_gdt = vmx_set_gdt,
7878 .get_dr6 = vmx_get_dr6,
7879 .set_dr6 = vmx_set_dr6,
7880 .set_dr7 = vmx_set_dr7,
7881 .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
7882 .cache_reg = vmx_cache_reg,
7883 .get_rflags = vmx_get_rflags,
7884 .set_rflags = vmx_set_rflags,
7886 .tlb_flush = vmx_flush_tlb,
7887 .tlb_flush_gva = vmx_flush_tlb_gva,
7889 .run = vmx_vcpu_run,
7890 .handle_exit = vmx_handle_exit,
7891 .skip_emulated_instruction = vmx_skip_emulated_instruction,
7892 .update_emulated_instruction = vmx_update_emulated_instruction,
7893 .set_interrupt_shadow = vmx_set_interrupt_shadow,
7894 .get_interrupt_shadow = vmx_get_interrupt_shadow,
7895 .patch_hypercall = vmx_patch_hypercall,
7896 .set_irq = vmx_inject_irq,
7897 .set_nmi = vmx_inject_nmi,
7898 .queue_exception = vmx_queue_exception,
7899 .cancel_injection = vmx_cancel_injection,
7900 .interrupt_allowed = vmx_interrupt_allowed,
7901 .nmi_allowed = vmx_nmi_allowed,
7902 .get_nmi_mask = vmx_get_nmi_mask,
7903 .set_nmi_mask = vmx_set_nmi_mask,
7904 .enable_nmi_window = enable_nmi_window,
7905 .enable_irq_window = enable_irq_window,
7906 .update_cr8_intercept = update_cr8_intercept,
7907 .set_virtual_apic_mode = vmx_set_virtual_apic_mode,
7908 .set_apic_access_page_addr = vmx_set_apic_access_page_addr,
7909 .refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
7910 .load_eoi_exitmap = vmx_load_eoi_exitmap,
7911 .apicv_post_state_restore = vmx_apicv_post_state_restore,
7912 .check_apicv_inhibit_reasons = vmx_check_apicv_inhibit_reasons,
7913 .hwapic_irr_update = vmx_hwapic_irr_update,
7914 .hwapic_isr_update = vmx_hwapic_isr_update,
7915 .guest_apic_has_interrupt = vmx_guest_apic_has_interrupt,
7916 .sync_pir_to_irr = vmx_sync_pir_to_irr,
7917 .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
7918 .dy_apicv_has_pending_interrupt = vmx_dy_apicv_has_pending_interrupt,
7920 .set_tss_addr = vmx_set_tss_addr,
7921 .set_identity_map_addr = vmx_set_identity_map_addr,
7922 .get_tdp_level = get_ept_level,
7923 .get_mt_mask = vmx_get_mt_mask,
7925 .get_exit_info = vmx_get_exit_info,
7927 .cpuid_update = vmx_cpuid_update,
7929 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
7931 .read_l1_tsc_offset = vmx_read_l1_tsc_offset,
7932 .write_l1_tsc_offset = vmx_write_l1_tsc_offset,
7934 .load_mmu_pgd = vmx_load_mmu_pgd,
7936 .check_intercept = vmx_check_intercept,
7937 .handle_exit_irqoff = vmx_handle_exit_irqoff,
7939 .request_immediate_exit = vmx_request_immediate_exit,
7941 .sched_in = vmx_sched_in,
7943 .slot_enable_log_dirty = vmx_slot_enable_log_dirty,
7944 .slot_disable_log_dirty = vmx_slot_disable_log_dirty,
7945 .flush_log_dirty = vmx_flush_log_dirty,
7946 .enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
7947 .write_log_dirty = vmx_write_pml_buffer,
7949 .pre_block = vmx_pre_block,
7950 .post_block = vmx_post_block,
7952 .pmu_ops = &intel_pmu_ops,
7954 .update_pi_irte = vmx_update_pi_irte,
7956 #ifdef CONFIG_X86_64
7957 .set_hv_timer = vmx_set_hv_timer,
7958 .cancel_hv_timer = vmx_cancel_hv_timer,
7961 .setup_mce = vmx_setup_mce,
7963 .smi_allowed = vmx_smi_allowed,
7964 .pre_enter_smm = vmx_pre_enter_smm,
7965 .pre_leave_smm = vmx_pre_leave_smm,
7966 .enable_smi_window = enable_smi_window,
7968 .check_nested_events = NULL,
7969 .get_nested_state = NULL,
7970 .set_nested_state = NULL,
7971 .get_vmcs12_pages = NULL,
7972 .nested_enable_evmcs = NULL,
7973 .nested_get_evmcs_version = NULL,
7974 .need_emulation_on_page_fault = vmx_need_emulation_on_page_fault,
7975 .apic_init_signal_blocked = vmx_apic_init_signal_blocked,
7978 static void vmx_cleanup_l1d_flush(void)
7980 if (vmx_l1d_flush_pages) {
7981 free_pages((unsigned long)vmx_l1d_flush_pages, L1D_CACHE_ORDER);
7982 vmx_l1d_flush_pages = NULL;
7984 /* Restore state so sysfs ignores VMX */
7985 l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
7988 static void vmx_exit(void)
7990 #ifdef CONFIG_KEXEC_CORE
7991 RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
7997 #if IS_ENABLED(CONFIG_HYPERV)
7998 if (static_branch_unlikely(&enable_evmcs)) {
8000 struct hv_vp_assist_page *vp_ap;
8002 * Reset everything to support using non-enlightened VMCS
8003 * access later (e.g. when we reload the module with
8004 * enlightened_vmcs=0)
8006 for_each_online_cpu(cpu) {
8007 vp_ap = hv_get_vp_assist_page(cpu);
8012 vp_ap->nested_control.features.directhypercall = 0;
8013 vp_ap->current_nested_vmcs = 0;
8014 vp_ap->enlighten_vmentry = 0;
8017 static_branch_disable(&enable_evmcs);
8020 vmx_cleanup_l1d_flush();
8022 module_exit(vmx_exit);
8024 static int __init vmx_init(void)
8028 #if IS_ENABLED(CONFIG_HYPERV)
8030 * Enlightened VMCS usage should be recommended and the host needs
8031 * to support eVMCS v1 or above. We can also disable eVMCS support
8032 * with module parameter.
8034 if (enlightened_vmcs &&
8035 ms_hyperv.hints & HV_X64_ENLIGHTENED_VMCS_RECOMMENDED &&
8036 (ms_hyperv.nested_features & HV_X64_ENLIGHTENED_VMCS_VERSION) >=
8037 KVM_EVMCS_VERSION) {
8040 /* Check that we have assist pages on all online CPUs */
8041 for_each_online_cpu(cpu) {
8042 if (!hv_get_vp_assist_page(cpu)) {
8043 enlightened_vmcs = false;
8048 if (enlightened_vmcs) {
8049 pr_info("KVM: vmx: using Hyper-V Enlightened VMCS\n");
8050 static_branch_enable(&enable_evmcs);
8053 if (ms_hyperv.nested_features & HV_X64_NESTED_DIRECT_FLUSH)
8054 vmx_x86_ops.enable_direct_tlbflush
8055 = hv_enable_direct_tlbflush;
8058 enlightened_vmcs = false;
8062 r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
8063 __alignof__(struct vcpu_vmx), THIS_MODULE);
8068 * Must be called after kvm_init() so enable_ept is properly set
8069 * up. Hand the parameter mitigation value in which was stored in
8070 * the pre module init parser. If no parameter was given, it will
8071 * contain 'auto' which will be turned into the default 'cond'
8074 r = vmx_setup_l1d_flush(vmentry_l1d_flush_param);
8080 #ifdef CONFIG_KEXEC_CORE
8081 rcu_assign_pointer(crash_vmclear_loaded_vmcss,
8082 crash_vmclear_local_loaded_vmcss);
8084 vmx_check_vmcs12_offsets();
8088 module_init(vmx_init);