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");
67 static const struct x86_cpu_id vmx_cpu_id[] = {
68 X86_FEATURE_MATCH(X86_FEATURE_VMX),
71 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
73 bool __read_mostly enable_vpid = 1;
74 module_param_named(vpid, enable_vpid, bool, 0444);
76 static bool __read_mostly enable_vnmi = 1;
77 module_param_named(vnmi, enable_vnmi, bool, S_IRUGO);
79 bool __read_mostly flexpriority_enabled = 1;
80 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
82 bool __read_mostly enable_ept = 1;
83 module_param_named(ept, enable_ept, bool, S_IRUGO);
85 bool __read_mostly enable_unrestricted_guest = 1;
86 module_param_named(unrestricted_guest,
87 enable_unrestricted_guest, bool, S_IRUGO);
89 bool __read_mostly enable_ept_ad_bits = 1;
90 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
92 static bool __read_mostly emulate_invalid_guest_state = true;
93 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
95 static bool __read_mostly fasteoi = 1;
96 module_param(fasteoi, bool, S_IRUGO);
98 static bool __read_mostly enable_apicv = 1;
99 module_param(enable_apicv, bool, S_IRUGO);
102 * If nested=1, nested virtualization is supported, i.e., guests may use
103 * VMX and be a hypervisor for its own guests. If nested=0, guests may not
104 * use VMX instructions.
106 static bool __read_mostly nested = 1;
107 module_param(nested, bool, S_IRUGO);
109 static u64 __read_mostly host_xss;
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)
213 l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_EPT_DISABLED;
217 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES)) {
220 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, msr);
221 if (msr & ARCH_CAP_SKIP_VMENTRY_L1DFLUSH) {
222 l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED;
227 /* If set to auto use the default l1tf mitigation method */
228 if (l1tf == VMENTER_L1D_FLUSH_AUTO) {
229 switch (l1tf_mitigation) {
230 case L1TF_MITIGATION_OFF:
231 l1tf = VMENTER_L1D_FLUSH_NEVER;
233 case L1TF_MITIGATION_FLUSH_NOWARN:
234 case L1TF_MITIGATION_FLUSH:
235 case L1TF_MITIGATION_FLUSH_NOSMT:
236 l1tf = VMENTER_L1D_FLUSH_COND;
238 case L1TF_MITIGATION_FULL:
239 case L1TF_MITIGATION_FULL_FORCE:
240 l1tf = VMENTER_L1D_FLUSH_ALWAYS;
243 } else if (l1tf_mitigation == L1TF_MITIGATION_FULL_FORCE) {
244 l1tf = VMENTER_L1D_FLUSH_ALWAYS;
247 if (l1tf != VMENTER_L1D_FLUSH_NEVER && !vmx_l1d_flush_pages &&
248 !boot_cpu_has(X86_FEATURE_FLUSH_L1D)) {
250 * This allocation for vmx_l1d_flush_pages is not tied to a VM
251 * lifetime and so should not be charged to a memcg.
253 page = alloc_pages(GFP_KERNEL, L1D_CACHE_ORDER);
256 vmx_l1d_flush_pages = page_address(page);
259 * Initialize each page with a different pattern in
260 * order to protect against KSM in the nested
261 * virtualization case.
263 for (i = 0; i < 1u << L1D_CACHE_ORDER; ++i) {
264 memset(vmx_l1d_flush_pages + i * PAGE_SIZE, i + 1,
269 l1tf_vmx_mitigation = l1tf;
271 if (l1tf != VMENTER_L1D_FLUSH_NEVER)
272 static_branch_enable(&vmx_l1d_should_flush);
274 static_branch_disable(&vmx_l1d_should_flush);
276 if (l1tf == VMENTER_L1D_FLUSH_COND)
277 static_branch_enable(&vmx_l1d_flush_cond);
279 static_branch_disable(&vmx_l1d_flush_cond);
283 static int vmentry_l1d_flush_parse(const char *s)
288 for (i = 0; i < ARRAY_SIZE(vmentry_l1d_param); i++) {
289 if (vmentry_l1d_param[i].for_parse &&
290 sysfs_streq(s, vmentry_l1d_param[i].option))
297 static int vmentry_l1d_flush_set(const char *s, const struct kernel_param *kp)
301 l1tf = vmentry_l1d_flush_parse(s);
305 if (!boot_cpu_has(X86_BUG_L1TF))
309 * Has vmx_init() run already? If not then this is the pre init
310 * parameter parsing. In that case just store the value and let
311 * vmx_init() do the proper setup after enable_ept has been
314 if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_AUTO) {
315 vmentry_l1d_flush_param = l1tf;
319 mutex_lock(&vmx_l1d_flush_mutex);
320 ret = vmx_setup_l1d_flush(l1tf);
321 mutex_unlock(&vmx_l1d_flush_mutex);
325 static int vmentry_l1d_flush_get(char *s, const struct kernel_param *kp)
327 if (WARN_ON_ONCE(l1tf_vmx_mitigation >= ARRAY_SIZE(vmentry_l1d_param)))
328 return sprintf(s, "???\n");
330 return sprintf(s, "%s\n", vmentry_l1d_param[l1tf_vmx_mitigation].option);
333 static const struct kernel_param_ops vmentry_l1d_flush_ops = {
334 .set = vmentry_l1d_flush_set,
335 .get = vmentry_l1d_flush_get,
337 module_param_cb(vmentry_l1d_flush, &vmentry_l1d_flush_ops, NULL, 0644);
339 static bool guest_state_valid(struct kvm_vcpu *vcpu);
340 static u32 vmx_segment_access_rights(struct kvm_segment *var);
341 static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
344 void vmx_vmexit(void);
346 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
347 DEFINE_PER_CPU(struct vmcs *, current_vmcs);
349 * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
350 * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
352 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
355 * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
356 * can find which vCPU should be waken up.
358 static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu);
359 static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock);
361 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
362 static DEFINE_SPINLOCK(vmx_vpid_lock);
364 struct vmcs_config vmcs_config;
365 struct vmx_capability vmx_capability;
367 #define VMX_SEGMENT_FIELD(seg) \
368 [VCPU_SREG_##seg] = { \
369 .selector = GUEST_##seg##_SELECTOR, \
370 .base = GUEST_##seg##_BASE, \
371 .limit = GUEST_##seg##_LIMIT, \
372 .ar_bytes = GUEST_##seg##_AR_BYTES, \
375 static const struct kvm_vmx_segment_field {
380 } kvm_vmx_segment_fields[] = {
381 VMX_SEGMENT_FIELD(CS),
382 VMX_SEGMENT_FIELD(DS),
383 VMX_SEGMENT_FIELD(ES),
384 VMX_SEGMENT_FIELD(FS),
385 VMX_SEGMENT_FIELD(GS),
386 VMX_SEGMENT_FIELD(SS),
387 VMX_SEGMENT_FIELD(TR),
388 VMX_SEGMENT_FIELD(LDTR),
394 * Though SYSCALL is only supported in 64-bit mode on Intel CPUs, kvm
395 * will emulate SYSCALL in legacy mode if the vendor string in guest
396 * CPUID.0:{EBX,ECX,EDX} is "AuthenticAMD" or "AMDisbetter!" To
397 * support this emulation, IA32_STAR must always be included in
398 * vmx_msr_index[], even in i386 builds.
400 const u32 vmx_msr_index[] = {
402 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
404 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
407 #if IS_ENABLED(CONFIG_HYPERV)
408 static bool __read_mostly enlightened_vmcs = true;
409 module_param(enlightened_vmcs, bool, 0444);
411 /* check_ept_pointer() should be under protection of ept_pointer_lock. */
412 static void check_ept_pointer_match(struct kvm *kvm)
414 struct kvm_vcpu *vcpu;
415 u64 tmp_eptp = INVALID_PAGE;
418 kvm_for_each_vcpu(i, vcpu, kvm) {
419 if (!VALID_PAGE(tmp_eptp)) {
420 tmp_eptp = to_vmx(vcpu)->ept_pointer;
421 } else if (tmp_eptp != to_vmx(vcpu)->ept_pointer) {
422 to_kvm_vmx(kvm)->ept_pointers_match
423 = EPT_POINTERS_MISMATCH;
428 to_kvm_vmx(kvm)->ept_pointers_match = EPT_POINTERS_MATCH;
431 static int kvm_fill_hv_flush_list_func(struct hv_guest_mapping_flush_list *flush,
434 struct kvm_tlb_range *range = data;
436 return hyperv_fill_flush_guest_mapping_list(flush, range->start_gfn,
440 static inline int __hv_remote_flush_tlb_with_range(struct kvm *kvm,
441 struct kvm_vcpu *vcpu, struct kvm_tlb_range *range)
443 u64 ept_pointer = to_vmx(vcpu)->ept_pointer;
446 * FLUSH_GUEST_PHYSICAL_ADDRESS_SPACE hypercall needs address
447 * of the base of EPT PML4 table, strip off EPT configuration
451 return hyperv_flush_guest_mapping_range(ept_pointer & PAGE_MASK,
452 kvm_fill_hv_flush_list_func, (void *)range);
454 return hyperv_flush_guest_mapping(ept_pointer & PAGE_MASK);
457 static int hv_remote_flush_tlb_with_range(struct kvm *kvm,
458 struct kvm_tlb_range *range)
460 struct kvm_vcpu *vcpu;
463 spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
465 if (to_kvm_vmx(kvm)->ept_pointers_match == EPT_POINTERS_CHECK)
466 check_ept_pointer_match(kvm);
468 if (to_kvm_vmx(kvm)->ept_pointers_match != EPT_POINTERS_MATCH) {
469 kvm_for_each_vcpu(i, vcpu, kvm) {
470 /* If ept_pointer is invalid pointer, bypass flush request. */
471 if (VALID_PAGE(to_vmx(vcpu)->ept_pointer))
472 ret |= __hv_remote_flush_tlb_with_range(
476 ret = __hv_remote_flush_tlb_with_range(kvm,
477 kvm_get_vcpu(kvm, 0), range);
480 spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
483 static int hv_remote_flush_tlb(struct kvm *kvm)
485 return hv_remote_flush_tlb_with_range(kvm, NULL);
488 #endif /* IS_ENABLED(CONFIG_HYPERV) */
491 * Comment's format: document - errata name - stepping - processor name.
493 * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
495 static u32 vmx_preemption_cpu_tfms[] = {
496 /* 323344.pdf - BA86 - D0 - Xeon 7500 Series */
498 /* 323056.pdf - AAX65 - C2 - Xeon L3406 */
499 /* 322814.pdf - AAT59 - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
500 /* 322911.pdf - AAU65 - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
502 /* 322911.pdf - AAU65 - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
504 /* 322373.pdf - AAO95 - B1 - Xeon 3400 Series */
505 /* 322166.pdf - AAN92 - B1 - i7-800 and i5-700 Desktop */
507 * 320767.pdf - AAP86 - B1 -
508 * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
511 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
513 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
515 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
517 /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
518 /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
519 /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
521 /* Xeon E3-1220 V2 */
525 static inline bool cpu_has_broken_vmx_preemption_timer(void)
527 u32 eax = cpuid_eax(0x00000001), i;
529 /* Clear the reserved bits */
530 eax &= ~(0x3U << 14 | 0xfU << 28);
531 for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
532 if (eax == vmx_preemption_cpu_tfms[i])
538 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
540 return flexpriority_enabled && lapic_in_kernel(vcpu);
543 static inline bool report_flexpriority(void)
545 return flexpriority_enabled;
548 static inline int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
552 for (i = 0; i < vmx->nmsrs; ++i)
553 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
558 struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
562 i = __find_msr_index(vmx, msr);
564 return &vmx->guest_msrs[i];
568 void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
570 vmcs_clear(loaded_vmcs->vmcs);
571 if (loaded_vmcs->shadow_vmcs && loaded_vmcs->launched)
572 vmcs_clear(loaded_vmcs->shadow_vmcs);
573 loaded_vmcs->cpu = -1;
574 loaded_vmcs->launched = 0;
577 #ifdef CONFIG_KEXEC_CORE
579 * This bitmap is used to indicate whether the vmclear
580 * operation is enabled on all cpus. All disabled by
583 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
585 static inline void crash_enable_local_vmclear(int cpu)
587 cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
590 static inline void crash_disable_local_vmclear(int cpu)
592 cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
595 static inline int crash_local_vmclear_enabled(int cpu)
597 return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
600 static void crash_vmclear_local_loaded_vmcss(void)
602 int cpu = raw_smp_processor_id();
603 struct loaded_vmcs *v;
605 if (!crash_local_vmclear_enabled(cpu))
608 list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
609 loaded_vmcss_on_cpu_link)
613 static inline void crash_enable_local_vmclear(int cpu) { }
614 static inline void crash_disable_local_vmclear(int cpu) { }
615 #endif /* CONFIG_KEXEC_CORE */
617 static void __loaded_vmcs_clear(void *arg)
619 struct loaded_vmcs *loaded_vmcs = arg;
620 int cpu = raw_smp_processor_id();
622 if (loaded_vmcs->cpu != cpu)
623 return; /* vcpu migration can race with cpu offline */
624 if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
625 per_cpu(current_vmcs, cpu) = NULL;
626 crash_disable_local_vmclear(cpu);
627 list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
630 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
631 * is before setting loaded_vmcs->vcpu to -1 which is done in
632 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
633 * then adds the vmcs into percpu list before it is deleted.
637 loaded_vmcs_init(loaded_vmcs);
638 crash_enable_local_vmclear(cpu);
641 void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
643 int cpu = loaded_vmcs->cpu;
646 smp_call_function_single(cpu,
647 __loaded_vmcs_clear, loaded_vmcs, 1);
650 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
654 u32 mask = 1 << (seg * SEG_FIELD_NR + field);
656 if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
657 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
658 vmx->segment_cache.bitmask = 0;
660 ret = vmx->segment_cache.bitmask & mask;
661 vmx->segment_cache.bitmask |= mask;
665 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
667 u16 *p = &vmx->segment_cache.seg[seg].selector;
669 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
670 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
674 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
676 ulong *p = &vmx->segment_cache.seg[seg].base;
678 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
679 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
683 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
685 u32 *p = &vmx->segment_cache.seg[seg].limit;
687 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
688 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
692 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
694 u32 *p = &vmx->segment_cache.seg[seg].ar;
696 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
697 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
701 void update_exception_bitmap(struct kvm_vcpu *vcpu)
705 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
706 (1u << DB_VECTOR) | (1u << AC_VECTOR);
708 * Guest access to VMware backdoor ports could legitimately
709 * trigger #GP because of TSS I/O permission bitmap.
710 * We intercept those #GP and allow access to them anyway
713 if (enable_vmware_backdoor)
714 eb |= (1u << GP_VECTOR);
715 if ((vcpu->guest_debug &
716 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
717 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
718 eb |= 1u << BP_VECTOR;
719 if (to_vmx(vcpu)->rmode.vm86_active)
722 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
724 /* When we are running a nested L2 guest and L1 specified for it a
725 * certain exception bitmap, we must trap the same exceptions and pass
726 * them to L1. When running L2, we will only handle the exceptions
727 * specified above if L1 did not want them.
729 if (is_guest_mode(vcpu))
730 eb |= get_vmcs12(vcpu)->exception_bitmap;
732 vmcs_write32(EXCEPTION_BITMAP, eb);
736 * Check if MSR is intercepted for currently loaded MSR bitmap.
738 static bool msr_write_intercepted(struct kvm_vcpu *vcpu, u32 msr)
740 unsigned long *msr_bitmap;
741 int f = sizeof(unsigned long);
743 if (!cpu_has_vmx_msr_bitmap())
746 msr_bitmap = to_vmx(vcpu)->loaded_vmcs->msr_bitmap;
749 return !!test_bit(msr, msr_bitmap + 0x800 / f);
750 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
752 return !!test_bit(msr, msr_bitmap + 0xc00 / f);
758 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
759 unsigned long entry, unsigned long exit)
761 vm_entry_controls_clearbit(vmx, entry);
762 vm_exit_controls_clearbit(vmx, exit);
765 static int find_msr(struct vmx_msrs *m, unsigned int msr)
769 for (i = 0; i < m->nr; ++i) {
770 if (m->val[i].index == msr)
776 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
779 struct msr_autoload *m = &vmx->msr_autoload;
783 if (cpu_has_load_ia32_efer()) {
784 clear_atomic_switch_msr_special(vmx,
785 VM_ENTRY_LOAD_IA32_EFER,
786 VM_EXIT_LOAD_IA32_EFER);
790 case MSR_CORE_PERF_GLOBAL_CTRL:
791 if (cpu_has_load_perf_global_ctrl()) {
792 clear_atomic_switch_msr_special(vmx,
793 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
794 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
799 i = find_msr(&m->guest, msr);
803 m->guest.val[i] = m->guest.val[m->guest.nr];
804 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
807 i = find_msr(&m->host, msr);
812 m->host.val[i] = m->host.val[m->host.nr];
813 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
816 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
817 unsigned long entry, unsigned long exit,
818 unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
819 u64 guest_val, u64 host_val)
821 vmcs_write64(guest_val_vmcs, guest_val);
822 if (host_val_vmcs != HOST_IA32_EFER)
823 vmcs_write64(host_val_vmcs, host_val);
824 vm_entry_controls_setbit(vmx, entry);
825 vm_exit_controls_setbit(vmx, exit);
828 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
829 u64 guest_val, u64 host_val, bool entry_only)
832 struct msr_autoload *m = &vmx->msr_autoload;
836 if (cpu_has_load_ia32_efer()) {
837 add_atomic_switch_msr_special(vmx,
838 VM_ENTRY_LOAD_IA32_EFER,
839 VM_EXIT_LOAD_IA32_EFER,
842 guest_val, host_val);
846 case MSR_CORE_PERF_GLOBAL_CTRL:
847 if (cpu_has_load_perf_global_ctrl()) {
848 add_atomic_switch_msr_special(vmx,
849 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
850 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
851 GUEST_IA32_PERF_GLOBAL_CTRL,
852 HOST_IA32_PERF_GLOBAL_CTRL,
853 guest_val, host_val);
857 case MSR_IA32_PEBS_ENABLE:
858 /* PEBS needs a quiescent period after being disabled (to write
859 * a record). Disabling PEBS through VMX MSR swapping doesn't
860 * provide that period, so a CPU could write host's record into
863 wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
866 i = find_msr(&m->guest, msr);
868 j = find_msr(&m->host, msr);
870 if ((i < 0 && m->guest.nr == NR_AUTOLOAD_MSRS) ||
871 (j < 0 && m->host.nr == NR_AUTOLOAD_MSRS)) {
872 printk_once(KERN_WARNING "Not enough msr switch entries. "
873 "Can't add msr %x\n", msr);
878 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
880 m->guest.val[i].index = msr;
881 m->guest.val[i].value = guest_val;
888 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
890 m->host.val[j].index = msr;
891 m->host.val[j].value = host_val;
894 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
896 u64 guest_efer = vmx->vcpu.arch.efer;
901 * NX is needed to handle CR0.WP=1, CR4.SMEP=1. Testing
902 * host CPUID is more efficient than testing guest CPUID
903 * or CR4. Host SMEP is anyway a requirement for guest SMEP.
905 if (boot_cpu_has(X86_FEATURE_SMEP))
906 guest_efer |= EFER_NX;
907 else if (!(guest_efer & EFER_NX))
908 ignore_bits |= EFER_NX;
912 * LMA and LME handled by hardware; SCE meaningless outside long mode.
914 ignore_bits |= EFER_SCE;
916 ignore_bits |= EFER_LMA | EFER_LME;
917 /* SCE is meaningful only in long mode on Intel */
918 if (guest_efer & EFER_LMA)
919 ignore_bits &= ~(u64)EFER_SCE;
923 * On EPT, we can't emulate NX, so we must switch EFER atomically.
924 * On CPUs that support "load IA32_EFER", always switch EFER
925 * atomically, since it's faster than switching it manually.
927 if (cpu_has_load_ia32_efer() ||
928 (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
929 if (!(guest_efer & EFER_LMA))
930 guest_efer &= ~EFER_LME;
931 if (guest_efer != host_efer)
932 add_atomic_switch_msr(vmx, MSR_EFER,
933 guest_efer, host_efer, false);
935 clear_atomic_switch_msr(vmx, MSR_EFER);
938 clear_atomic_switch_msr(vmx, MSR_EFER);
940 guest_efer &= ~ignore_bits;
941 guest_efer |= host_efer & ignore_bits;
943 vmx->guest_msrs[efer_offset].data = guest_efer;
944 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
952 * On 32-bit kernels, VM exits still load the FS and GS bases from the
953 * VMCS rather than the segment table. KVM uses this helper to figure
954 * out the current bases to poke them into the VMCS before entry.
956 static unsigned long segment_base(u16 selector)
958 struct desc_struct *table;
961 if (!(selector & ~SEGMENT_RPL_MASK))
964 table = get_current_gdt_ro();
966 if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
967 u16 ldt_selector = kvm_read_ldt();
969 if (!(ldt_selector & ~SEGMENT_RPL_MASK))
972 table = (struct desc_struct *)segment_base(ldt_selector);
974 v = get_desc_base(&table[selector >> 3]);
979 static inline void pt_load_msr(struct pt_ctx *ctx, u32 addr_range)
983 wrmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
984 wrmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
985 wrmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
986 wrmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
987 for (i = 0; i < addr_range; i++) {
988 wrmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
989 wrmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
993 static inline void pt_save_msr(struct pt_ctx *ctx, u32 addr_range)
997 rdmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
998 rdmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
999 rdmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1000 rdmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1001 for (i = 0; i < addr_range; i++) {
1002 rdmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1003 rdmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1007 static void pt_guest_enter(struct vcpu_vmx *vmx)
1009 if (pt_mode == PT_MODE_SYSTEM)
1013 * GUEST_IA32_RTIT_CTL is already set in the VMCS.
1014 * Save host state before VM entry.
1016 rdmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1017 if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1018 wrmsrl(MSR_IA32_RTIT_CTL, 0);
1019 pt_save_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1020 pt_load_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1024 static void pt_guest_exit(struct vcpu_vmx *vmx)
1026 if (pt_mode == PT_MODE_SYSTEM)
1029 if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1030 pt_save_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1031 pt_load_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1034 /* Reload host state (IA32_RTIT_CTL will be cleared on VM exit). */
1035 wrmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1038 void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu)
1040 struct vcpu_vmx *vmx = to_vmx(vcpu);
1041 struct vmcs_host_state *host_state;
1042 #ifdef CONFIG_X86_64
1043 int cpu = raw_smp_processor_id();
1045 unsigned long fs_base, gs_base;
1049 vmx->req_immediate_exit = false;
1052 * Note that guest MSRs to be saved/restored can also be changed
1053 * when guest state is loaded. This happens when guest transitions
1054 * to/from long-mode by setting MSR_EFER.LMA.
1056 if (!vmx->loaded_cpu_state || vmx->guest_msrs_dirty) {
1057 vmx->guest_msrs_dirty = false;
1058 for (i = 0; i < vmx->save_nmsrs; ++i)
1059 kvm_set_shared_msr(vmx->guest_msrs[i].index,
1060 vmx->guest_msrs[i].data,
1061 vmx->guest_msrs[i].mask);
1065 if (vmx->loaded_cpu_state)
1068 vmx->loaded_cpu_state = vmx->loaded_vmcs;
1069 host_state = &vmx->loaded_cpu_state->host_state;
1072 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
1073 * allow segment selectors with cpl > 0 or ti == 1.
1075 host_state->ldt_sel = kvm_read_ldt();
1077 #ifdef CONFIG_X86_64
1078 savesegment(ds, host_state->ds_sel);
1079 savesegment(es, host_state->es_sel);
1081 gs_base = cpu_kernelmode_gs_base(cpu);
1082 if (likely(is_64bit_mm(current->mm))) {
1083 save_fsgs_for_kvm();
1084 fs_sel = current->thread.fsindex;
1085 gs_sel = current->thread.gsindex;
1086 fs_base = current->thread.fsbase;
1087 vmx->msr_host_kernel_gs_base = current->thread.gsbase;
1089 savesegment(fs, fs_sel);
1090 savesegment(gs, gs_sel);
1091 fs_base = read_msr(MSR_FS_BASE);
1092 vmx->msr_host_kernel_gs_base = read_msr(MSR_KERNEL_GS_BASE);
1095 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1097 savesegment(fs, fs_sel);
1098 savesegment(gs, gs_sel);
1099 fs_base = segment_base(fs_sel);
1100 gs_base = segment_base(gs_sel);
1103 if (unlikely(fs_sel != host_state->fs_sel)) {
1105 vmcs_write16(HOST_FS_SELECTOR, fs_sel);
1107 vmcs_write16(HOST_FS_SELECTOR, 0);
1108 host_state->fs_sel = fs_sel;
1110 if (unlikely(gs_sel != host_state->gs_sel)) {
1112 vmcs_write16(HOST_GS_SELECTOR, gs_sel);
1114 vmcs_write16(HOST_GS_SELECTOR, 0);
1115 host_state->gs_sel = gs_sel;
1117 if (unlikely(fs_base != host_state->fs_base)) {
1118 vmcs_writel(HOST_FS_BASE, fs_base);
1119 host_state->fs_base = fs_base;
1121 if (unlikely(gs_base != host_state->gs_base)) {
1122 vmcs_writel(HOST_GS_BASE, gs_base);
1123 host_state->gs_base = gs_base;
1127 static void vmx_prepare_switch_to_host(struct vcpu_vmx *vmx)
1129 struct vmcs_host_state *host_state;
1131 if (!vmx->loaded_cpu_state)
1134 WARN_ON_ONCE(vmx->loaded_cpu_state != vmx->loaded_vmcs);
1135 host_state = &vmx->loaded_cpu_state->host_state;
1137 ++vmx->vcpu.stat.host_state_reload;
1138 vmx->loaded_cpu_state = NULL;
1140 #ifdef CONFIG_X86_64
1141 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1143 if (host_state->ldt_sel || (host_state->gs_sel & 7)) {
1144 kvm_load_ldt(host_state->ldt_sel);
1145 #ifdef CONFIG_X86_64
1146 load_gs_index(host_state->gs_sel);
1148 loadsegment(gs, host_state->gs_sel);
1151 if (host_state->fs_sel & 7)
1152 loadsegment(fs, host_state->fs_sel);
1153 #ifdef CONFIG_X86_64
1154 if (unlikely(host_state->ds_sel | host_state->es_sel)) {
1155 loadsegment(ds, host_state->ds_sel);
1156 loadsegment(es, host_state->es_sel);
1159 invalidate_tss_limit();
1160 #ifdef CONFIG_X86_64
1161 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1163 load_fixmap_gdt(raw_smp_processor_id());
1166 #ifdef CONFIG_X86_64
1167 static u64 vmx_read_guest_kernel_gs_base(struct vcpu_vmx *vmx)
1170 if (vmx->loaded_cpu_state)
1171 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1173 return vmx->msr_guest_kernel_gs_base;
1176 static void vmx_write_guest_kernel_gs_base(struct vcpu_vmx *vmx, u64 data)
1179 if (vmx->loaded_cpu_state)
1180 wrmsrl(MSR_KERNEL_GS_BASE, data);
1182 vmx->msr_guest_kernel_gs_base = data;
1186 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
1188 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1189 struct pi_desc old, new;
1193 * In case of hot-plug or hot-unplug, we may have to undo
1194 * vmx_vcpu_pi_put even if there is no assigned device. And we
1195 * always keep PI.NDST up to date for simplicity: it makes the
1196 * code easier, and CPU migration is not a fast path.
1198 if (!pi_test_sn(pi_desc) && vcpu->cpu == cpu)
1201 /* The full case. */
1203 old.control = new.control = pi_desc->control;
1205 dest = cpu_physical_id(cpu);
1207 if (x2apic_enabled())
1210 new.ndst = (dest << 8) & 0xFF00;
1213 } while (cmpxchg64(&pi_desc->control, old.control,
1214 new.control) != old.control);
1217 * Clear SN before reading the bitmap. The VT-d firmware
1218 * writes the bitmap and reads SN atomically (5.2.3 in the
1219 * spec), so it doesn't really have a memory barrier that
1220 * pairs with this, but we cannot do that and we need one.
1222 smp_mb__after_atomic();
1224 if (!bitmap_empty((unsigned long *)pi_desc->pir, NR_VECTORS))
1229 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1230 * vcpu mutex is already taken.
1232 void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1234 struct vcpu_vmx *vmx = to_vmx(vcpu);
1235 bool already_loaded = vmx->loaded_vmcs->cpu == cpu;
1237 if (!already_loaded) {
1238 loaded_vmcs_clear(vmx->loaded_vmcs);
1239 local_irq_disable();
1240 crash_disable_local_vmclear(cpu);
1243 * Read loaded_vmcs->cpu should be before fetching
1244 * loaded_vmcs->loaded_vmcss_on_cpu_link.
1245 * See the comments in __loaded_vmcs_clear().
1249 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1250 &per_cpu(loaded_vmcss_on_cpu, cpu));
1251 crash_enable_local_vmclear(cpu);
1255 if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
1256 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1257 vmcs_load(vmx->loaded_vmcs->vmcs);
1258 indirect_branch_prediction_barrier();
1261 if (!already_loaded) {
1262 void *gdt = get_current_gdt_ro();
1263 unsigned long sysenter_esp;
1265 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1268 * Linux uses per-cpu TSS and GDT, so set these when switching
1269 * processors. See 22.2.4.
1271 vmcs_writel(HOST_TR_BASE,
1272 (unsigned long)&get_cpu_entry_area(cpu)->tss.x86_tss);
1273 vmcs_writel(HOST_GDTR_BASE, (unsigned long)gdt); /* 22.2.4 */
1276 * VM exits change the host TR limit to 0x67 after a VM
1277 * exit. This is okay, since 0x67 covers everything except
1278 * the IO bitmap and have have code to handle the IO bitmap
1279 * being lost after a VM exit.
1281 BUILD_BUG_ON(IO_BITMAP_OFFSET - 1 != 0x67);
1283 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1284 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1286 vmx->loaded_vmcs->cpu = cpu;
1289 /* Setup TSC multiplier */
1290 if (kvm_has_tsc_control &&
1291 vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio)
1292 decache_tsc_multiplier(vmx);
1294 vmx_vcpu_pi_load(vcpu, cpu);
1295 vmx->host_pkru = read_pkru();
1296 vmx->host_debugctlmsr = get_debugctlmsr();
1299 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
1301 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1303 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
1304 !irq_remapping_cap(IRQ_POSTING_CAP) ||
1305 !kvm_vcpu_apicv_active(vcpu))
1308 /* Set SN when the vCPU is preempted */
1309 if (vcpu->preempted)
1313 void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1315 vmx_vcpu_pi_put(vcpu);
1317 vmx_prepare_switch_to_host(to_vmx(vcpu));
1320 static bool emulation_required(struct kvm_vcpu *vcpu)
1322 return emulate_invalid_guest_state && !guest_state_valid(vcpu);
1325 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
1327 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1329 unsigned long rflags, save_rflags;
1331 if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
1332 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1333 rflags = vmcs_readl(GUEST_RFLAGS);
1334 if (to_vmx(vcpu)->rmode.vm86_active) {
1335 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1336 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
1337 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1339 to_vmx(vcpu)->rflags = rflags;
1341 return to_vmx(vcpu)->rflags;
1344 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1346 unsigned long old_rflags = vmx_get_rflags(vcpu);
1348 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1349 to_vmx(vcpu)->rflags = rflags;
1350 if (to_vmx(vcpu)->rmode.vm86_active) {
1351 to_vmx(vcpu)->rmode.save_rflags = rflags;
1352 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1354 vmcs_writel(GUEST_RFLAGS, rflags);
1356 if ((old_rflags ^ to_vmx(vcpu)->rflags) & X86_EFLAGS_VM)
1357 to_vmx(vcpu)->emulation_required = emulation_required(vcpu);
1360 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
1362 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1365 if (interruptibility & GUEST_INTR_STATE_STI)
1366 ret |= KVM_X86_SHADOW_INT_STI;
1367 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1368 ret |= KVM_X86_SHADOW_INT_MOV_SS;
1373 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1375 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1376 u32 interruptibility = interruptibility_old;
1378 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1380 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1381 interruptibility |= GUEST_INTR_STATE_MOV_SS;
1382 else if (mask & KVM_X86_SHADOW_INT_STI)
1383 interruptibility |= GUEST_INTR_STATE_STI;
1385 if ((interruptibility != interruptibility_old))
1386 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1389 static int vmx_rtit_ctl_check(struct kvm_vcpu *vcpu, u64 data)
1391 struct vcpu_vmx *vmx = to_vmx(vcpu);
1392 unsigned long value;
1395 * Any MSR write that attempts to change bits marked reserved will
1398 if (data & vmx->pt_desc.ctl_bitmask)
1402 * Any attempt to modify IA32_RTIT_CTL while TraceEn is set will
1403 * result in a #GP unless the same write also clears TraceEn.
1405 if ((vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) &&
1406 ((vmx->pt_desc.guest.ctl ^ data) & ~RTIT_CTL_TRACEEN))
1410 * WRMSR to IA32_RTIT_CTL that sets TraceEn but clears this bit
1411 * and FabricEn would cause #GP, if
1412 * CPUID.(EAX=14H, ECX=0):ECX.SNGLRGNOUT[bit 2] = 0
1414 if ((data & RTIT_CTL_TRACEEN) && !(data & RTIT_CTL_TOPA) &&
1415 !(data & RTIT_CTL_FABRIC_EN) &&
1416 !intel_pt_validate_cap(vmx->pt_desc.caps,
1417 PT_CAP_single_range_output))
1421 * MTCFreq, CycThresh and PSBFreq encodings check, any MSR write that
1422 * utilize encodings marked reserved will casue a #GP fault.
1424 value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc_periods);
1425 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc) &&
1426 !test_bit((data & RTIT_CTL_MTC_RANGE) >>
1427 RTIT_CTL_MTC_RANGE_OFFSET, &value))
1429 value = intel_pt_validate_cap(vmx->pt_desc.caps,
1430 PT_CAP_cycle_thresholds);
1431 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1432 !test_bit((data & RTIT_CTL_CYC_THRESH) >>
1433 RTIT_CTL_CYC_THRESH_OFFSET, &value))
1435 value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_periods);
1436 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1437 !test_bit((data & RTIT_CTL_PSB_FREQ) >>
1438 RTIT_CTL_PSB_FREQ_OFFSET, &value))
1442 * If ADDRx_CFG is reserved or the encodings is >2 will
1443 * cause a #GP fault.
1445 value = (data & RTIT_CTL_ADDR0) >> RTIT_CTL_ADDR0_OFFSET;
1446 if ((value && (vmx->pt_desc.addr_range < 1)) || (value > 2))
1448 value = (data & RTIT_CTL_ADDR1) >> RTIT_CTL_ADDR1_OFFSET;
1449 if ((value && (vmx->pt_desc.addr_range < 2)) || (value > 2))
1451 value = (data & RTIT_CTL_ADDR2) >> RTIT_CTL_ADDR2_OFFSET;
1452 if ((value && (vmx->pt_desc.addr_range < 3)) || (value > 2))
1454 value = (data & RTIT_CTL_ADDR3) >> RTIT_CTL_ADDR3_OFFSET;
1455 if ((value && (vmx->pt_desc.addr_range < 4)) || (value > 2))
1462 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
1466 rip = kvm_rip_read(vcpu);
1467 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1468 kvm_rip_write(vcpu, rip);
1470 /* skipping an emulated instruction also counts */
1471 vmx_set_interrupt_shadow(vcpu, 0);
1474 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1477 * Ensure that we clear the HLT state in the VMCS. We don't need to
1478 * explicitly skip the instruction because if the HLT state is set,
1479 * then the instruction is already executing and RIP has already been
1482 if (kvm_hlt_in_guest(vcpu->kvm) &&
1483 vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1484 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1487 static void vmx_queue_exception(struct kvm_vcpu *vcpu)
1489 struct vcpu_vmx *vmx = to_vmx(vcpu);
1490 unsigned nr = vcpu->arch.exception.nr;
1491 bool has_error_code = vcpu->arch.exception.has_error_code;
1492 u32 error_code = vcpu->arch.exception.error_code;
1493 u32 intr_info = nr | INTR_INFO_VALID_MASK;
1495 kvm_deliver_exception_payload(vcpu);
1497 if (has_error_code) {
1498 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1499 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1502 if (vmx->rmode.vm86_active) {
1504 if (kvm_exception_is_soft(nr))
1505 inc_eip = vcpu->arch.event_exit_inst_len;
1506 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
1507 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
1511 WARN_ON_ONCE(vmx->emulation_required);
1513 if (kvm_exception_is_soft(nr)) {
1514 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1515 vmx->vcpu.arch.event_exit_inst_len);
1516 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1518 intr_info |= INTR_TYPE_HARD_EXCEPTION;
1520 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1522 vmx_clear_hlt(vcpu);
1525 static bool vmx_rdtscp_supported(void)
1527 return cpu_has_vmx_rdtscp();
1530 static bool vmx_invpcid_supported(void)
1532 return cpu_has_vmx_invpcid();
1536 * Swap MSR entry in host/guest MSR entry array.
1538 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1540 struct shared_msr_entry tmp;
1542 tmp = vmx->guest_msrs[to];
1543 vmx->guest_msrs[to] = vmx->guest_msrs[from];
1544 vmx->guest_msrs[from] = tmp;
1548 * Set up the vmcs to automatically save and restore system
1549 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
1550 * mode, as fiddling with msrs is very expensive.
1552 static void setup_msrs(struct vcpu_vmx *vmx)
1554 int save_nmsrs, index;
1557 #ifdef CONFIG_X86_64
1559 * The SYSCALL MSRs are only needed on long mode guests, and only
1560 * when EFER.SCE is set.
1562 if (is_long_mode(&vmx->vcpu) && (vmx->vcpu.arch.efer & EFER_SCE)) {
1563 index = __find_msr_index(vmx, MSR_STAR);
1565 move_msr_up(vmx, index, save_nmsrs++);
1566 index = __find_msr_index(vmx, MSR_LSTAR);
1568 move_msr_up(vmx, index, save_nmsrs++);
1569 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1571 move_msr_up(vmx, index, save_nmsrs++);
1574 index = __find_msr_index(vmx, MSR_EFER);
1575 if (index >= 0 && update_transition_efer(vmx, index))
1576 move_msr_up(vmx, index, save_nmsrs++);
1577 index = __find_msr_index(vmx, MSR_TSC_AUX);
1578 if (index >= 0 && guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDTSCP))
1579 move_msr_up(vmx, index, save_nmsrs++);
1581 vmx->save_nmsrs = save_nmsrs;
1582 vmx->guest_msrs_dirty = true;
1584 if (cpu_has_vmx_msr_bitmap())
1585 vmx_update_msr_bitmap(&vmx->vcpu);
1588 static u64 vmx_read_l1_tsc_offset(struct kvm_vcpu *vcpu)
1590 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1592 if (is_guest_mode(vcpu) &&
1593 (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING))
1594 return vcpu->arch.tsc_offset - vmcs12->tsc_offset;
1596 return vcpu->arch.tsc_offset;
1599 static u64 vmx_write_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1601 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1602 u64 g_tsc_offset = 0;
1605 * We're here if L1 chose not to trap WRMSR to TSC. According
1606 * to the spec, this should set L1's TSC; The offset that L1
1607 * set for L2 remains unchanged, and still needs to be added
1608 * to the newly set TSC to get L2's TSC.
1610 if (is_guest_mode(vcpu) &&
1611 (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING))
1612 g_tsc_offset = vmcs12->tsc_offset;
1614 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1615 vcpu->arch.tsc_offset - g_tsc_offset,
1617 vmcs_write64(TSC_OFFSET, offset + g_tsc_offset);
1618 return offset + g_tsc_offset;
1622 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
1623 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
1624 * all guests if the "nested" module option is off, and can also be disabled
1625 * for a single guest by disabling its VMX cpuid bit.
1627 bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
1629 return nested && guest_cpuid_has(vcpu, X86_FEATURE_VMX);
1632 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
1635 uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
1637 return !(val & ~valid_bits);
1640 static int vmx_get_msr_feature(struct kvm_msr_entry *msr)
1642 switch (msr->index) {
1643 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1646 return vmx_get_vmx_msr(&vmcs_config.nested, msr->index, &msr->data);
1655 * Reads an msr value (of 'msr_index') into 'pdata'.
1656 * Returns 0 on success, non-0 otherwise.
1657 * Assumes vcpu_load() was already called.
1659 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1661 struct vcpu_vmx *vmx = to_vmx(vcpu);
1662 struct shared_msr_entry *msr;
1665 switch (msr_info->index) {
1666 #ifdef CONFIG_X86_64
1668 msr_info->data = vmcs_readl(GUEST_FS_BASE);
1671 msr_info->data = vmcs_readl(GUEST_GS_BASE);
1673 case MSR_KERNEL_GS_BASE:
1674 msr_info->data = vmx_read_guest_kernel_gs_base(vmx);
1678 return kvm_get_msr_common(vcpu, msr_info);
1679 case MSR_IA32_SPEC_CTRL:
1680 if (!msr_info->host_initiated &&
1681 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1684 msr_info->data = to_vmx(vcpu)->spec_ctrl;
1686 case MSR_IA32_SYSENTER_CS:
1687 msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
1689 case MSR_IA32_SYSENTER_EIP:
1690 msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
1692 case MSR_IA32_SYSENTER_ESP:
1693 msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
1695 case MSR_IA32_POWER_CTL:
1696 msr_info->data = vmx->msr_ia32_power_ctl;
1698 case MSR_IA32_BNDCFGS:
1699 if (!kvm_mpx_supported() ||
1700 (!msr_info->host_initiated &&
1701 !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1703 msr_info->data = vmcs_read64(GUEST_BNDCFGS);
1705 case MSR_IA32_MCG_EXT_CTL:
1706 if (!msr_info->host_initiated &&
1707 !(vmx->msr_ia32_feature_control &
1708 FEATURE_CONTROL_LMCE))
1710 msr_info->data = vcpu->arch.mcg_ext_ctl;
1712 case MSR_IA32_FEATURE_CONTROL:
1713 msr_info->data = vmx->msr_ia32_feature_control;
1715 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1716 if (!nested_vmx_allowed(vcpu))
1718 return vmx_get_vmx_msr(&vmx->nested.msrs, msr_info->index,
1721 if (!vmx_xsaves_supported())
1723 msr_info->data = vcpu->arch.ia32_xss;
1725 case MSR_IA32_RTIT_CTL:
1726 if (pt_mode != PT_MODE_HOST_GUEST)
1728 msr_info->data = vmx->pt_desc.guest.ctl;
1730 case MSR_IA32_RTIT_STATUS:
1731 if (pt_mode != PT_MODE_HOST_GUEST)
1733 msr_info->data = vmx->pt_desc.guest.status;
1735 case MSR_IA32_RTIT_CR3_MATCH:
1736 if ((pt_mode != PT_MODE_HOST_GUEST) ||
1737 !intel_pt_validate_cap(vmx->pt_desc.caps,
1738 PT_CAP_cr3_filtering))
1740 msr_info->data = vmx->pt_desc.guest.cr3_match;
1742 case MSR_IA32_RTIT_OUTPUT_BASE:
1743 if ((pt_mode != PT_MODE_HOST_GUEST) ||
1744 (!intel_pt_validate_cap(vmx->pt_desc.caps,
1745 PT_CAP_topa_output) &&
1746 !intel_pt_validate_cap(vmx->pt_desc.caps,
1747 PT_CAP_single_range_output)))
1749 msr_info->data = vmx->pt_desc.guest.output_base;
1751 case MSR_IA32_RTIT_OUTPUT_MASK:
1752 if ((pt_mode != PT_MODE_HOST_GUEST) ||
1753 (!intel_pt_validate_cap(vmx->pt_desc.caps,
1754 PT_CAP_topa_output) &&
1755 !intel_pt_validate_cap(vmx->pt_desc.caps,
1756 PT_CAP_single_range_output)))
1758 msr_info->data = vmx->pt_desc.guest.output_mask;
1760 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
1761 index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
1762 if ((pt_mode != PT_MODE_HOST_GUEST) ||
1763 (index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
1764 PT_CAP_num_address_ranges)))
1767 msr_info->data = vmx->pt_desc.guest.addr_b[index / 2];
1769 msr_info->data = vmx->pt_desc.guest.addr_a[index / 2];
1772 if (!msr_info->host_initiated &&
1773 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
1775 /* Else, falls through */
1777 msr = find_msr_entry(vmx, msr_info->index);
1779 msr_info->data = msr->data;
1782 return kvm_get_msr_common(vcpu, msr_info);
1789 * Writes msr value into into the appropriate "register".
1790 * Returns 0 on success, non-0 otherwise.
1791 * Assumes vcpu_load() was already called.
1793 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1795 struct vcpu_vmx *vmx = to_vmx(vcpu);
1796 struct shared_msr_entry *msr;
1798 u32 msr_index = msr_info->index;
1799 u64 data = msr_info->data;
1802 switch (msr_index) {
1804 ret = kvm_set_msr_common(vcpu, msr_info);
1806 #ifdef CONFIG_X86_64
1808 vmx_segment_cache_clear(vmx);
1809 vmcs_writel(GUEST_FS_BASE, data);
1812 vmx_segment_cache_clear(vmx);
1813 vmcs_writel(GUEST_GS_BASE, data);
1815 case MSR_KERNEL_GS_BASE:
1816 vmx_write_guest_kernel_gs_base(vmx, data);
1819 case MSR_IA32_SYSENTER_CS:
1820 vmcs_write32(GUEST_SYSENTER_CS, data);
1822 case MSR_IA32_SYSENTER_EIP:
1823 vmcs_writel(GUEST_SYSENTER_EIP, data);
1825 case MSR_IA32_SYSENTER_ESP:
1826 vmcs_writel(GUEST_SYSENTER_ESP, data);
1828 case MSR_IA32_POWER_CTL:
1829 vmx->msr_ia32_power_ctl = data;
1831 case MSR_IA32_BNDCFGS:
1832 if (!kvm_mpx_supported() ||
1833 (!msr_info->host_initiated &&
1834 !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1836 if (is_noncanonical_address(data & PAGE_MASK, vcpu) ||
1837 (data & MSR_IA32_BNDCFGS_RSVD))
1839 vmcs_write64(GUEST_BNDCFGS, data);
1841 case MSR_IA32_SPEC_CTRL:
1842 if (!msr_info->host_initiated &&
1843 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1846 /* The STIBP bit doesn't fault even if it's not advertised */
1847 if (data & ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP | SPEC_CTRL_SSBD))
1850 vmx->spec_ctrl = data;
1857 * When it's written (to non-zero) for the first time, pass
1861 * The handling of the MSR bitmap for L2 guests is done in
1862 * nested_vmx_merge_msr_bitmap. We should not touch the
1863 * vmcs02.msr_bitmap here since it gets completely overwritten
1864 * in the merging. We update the vmcs01 here for L1 as well
1865 * since it will end up touching the MSR anyway now.
1867 vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap,
1871 case MSR_IA32_PRED_CMD:
1872 if (!msr_info->host_initiated &&
1873 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1876 if (data & ~PRED_CMD_IBPB)
1882 wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
1886 * When it's written (to non-zero) for the first time, pass
1890 * The handling of the MSR bitmap for L2 guests is done in
1891 * nested_vmx_merge_msr_bitmap. We should not touch the
1892 * vmcs02.msr_bitmap here since it gets completely overwritten
1895 vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap, MSR_IA32_PRED_CMD,
1898 case MSR_IA32_CR_PAT:
1899 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
1900 if (!kvm_pat_valid(data))
1902 vmcs_write64(GUEST_IA32_PAT, data);
1903 vcpu->arch.pat = data;
1906 ret = kvm_set_msr_common(vcpu, msr_info);
1908 case MSR_IA32_TSC_ADJUST:
1909 ret = kvm_set_msr_common(vcpu, msr_info);
1911 case MSR_IA32_MCG_EXT_CTL:
1912 if ((!msr_info->host_initiated &&
1913 !(to_vmx(vcpu)->msr_ia32_feature_control &
1914 FEATURE_CONTROL_LMCE)) ||
1915 (data & ~MCG_EXT_CTL_LMCE_EN))
1917 vcpu->arch.mcg_ext_ctl = data;
1919 case MSR_IA32_FEATURE_CONTROL:
1920 if (!vmx_feature_control_msr_valid(vcpu, data) ||
1921 (to_vmx(vcpu)->msr_ia32_feature_control &
1922 FEATURE_CONTROL_LOCKED && !msr_info->host_initiated))
1924 vmx->msr_ia32_feature_control = data;
1925 if (msr_info->host_initiated && data == 0)
1926 vmx_leave_nested(vcpu);
1928 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1929 if (!msr_info->host_initiated)
1930 return 1; /* they are read-only */
1931 if (!nested_vmx_allowed(vcpu))
1933 return vmx_set_vmx_msr(vcpu, msr_index, data);
1935 if (!vmx_xsaves_supported())
1938 * The only supported bit as of Skylake is bit 8, but
1939 * it is not supported on KVM.
1943 vcpu->arch.ia32_xss = data;
1944 if (vcpu->arch.ia32_xss != host_xss)
1945 add_atomic_switch_msr(vmx, MSR_IA32_XSS,
1946 vcpu->arch.ia32_xss, host_xss, false);
1948 clear_atomic_switch_msr(vmx, MSR_IA32_XSS);
1950 case MSR_IA32_RTIT_CTL:
1951 if ((pt_mode != PT_MODE_HOST_GUEST) ||
1952 vmx_rtit_ctl_check(vcpu, data) ||
1955 vmcs_write64(GUEST_IA32_RTIT_CTL, data);
1956 vmx->pt_desc.guest.ctl = data;
1957 pt_update_intercept_for_msr(vmx);
1959 case MSR_IA32_RTIT_STATUS:
1960 if ((pt_mode != PT_MODE_HOST_GUEST) ||
1961 (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
1962 (data & MSR_IA32_RTIT_STATUS_MASK))
1964 vmx->pt_desc.guest.status = data;
1966 case MSR_IA32_RTIT_CR3_MATCH:
1967 if ((pt_mode != PT_MODE_HOST_GUEST) ||
1968 (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
1969 !intel_pt_validate_cap(vmx->pt_desc.caps,
1970 PT_CAP_cr3_filtering))
1972 vmx->pt_desc.guest.cr3_match = data;
1974 case MSR_IA32_RTIT_OUTPUT_BASE:
1975 if ((pt_mode != PT_MODE_HOST_GUEST) ||
1976 (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
1977 (!intel_pt_validate_cap(vmx->pt_desc.caps,
1978 PT_CAP_topa_output) &&
1979 !intel_pt_validate_cap(vmx->pt_desc.caps,
1980 PT_CAP_single_range_output)) ||
1981 (data & MSR_IA32_RTIT_OUTPUT_BASE_MASK))
1983 vmx->pt_desc.guest.output_base = data;
1985 case MSR_IA32_RTIT_OUTPUT_MASK:
1986 if ((pt_mode != PT_MODE_HOST_GUEST) ||
1987 (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
1988 (!intel_pt_validate_cap(vmx->pt_desc.caps,
1989 PT_CAP_topa_output) &&
1990 !intel_pt_validate_cap(vmx->pt_desc.caps,
1991 PT_CAP_single_range_output)))
1993 vmx->pt_desc.guest.output_mask = data;
1995 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
1996 index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
1997 if ((pt_mode != PT_MODE_HOST_GUEST) ||
1998 (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
1999 (index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
2000 PT_CAP_num_address_ranges)))
2003 vmx->pt_desc.guest.addr_b[index / 2] = data;
2005 vmx->pt_desc.guest.addr_a[index / 2] = data;
2008 if (!msr_info->host_initiated &&
2009 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
2011 /* Check reserved bit, higher 32 bits should be zero */
2012 if ((data >> 32) != 0)
2014 /* Else, falls through */
2016 msr = find_msr_entry(vmx, msr_index);
2018 u64 old_msr_data = msr->data;
2020 if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
2022 ret = kvm_set_shared_msr(msr->index, msr->data,
2026 msr->data = old_msr_data;
2030 ret = kvm_set_msr_common(vcpu, msr_info);
2036 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2038 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
2041 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2044 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2046 case VCPU_EXREG_PDPTR:
2048 ept_save_pdptrs(vcpu);
2055 static __init int cpu_has_kvm_support(void)
2057 return cpu_has_vmx();
2060 static __init int vmx_disabled_by_bios(void)
2064 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
2065 if (msr & FEATURE_CONTROL_LOCKED) {
2066 /* launched w/ TXT and VMX disabled */
2067 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2070 /* launched w/o TXT and VMX only enabled w/ TXT */
2071 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2072 && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2073 && !tboot_enabled()) {
2074 printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
2075 "activate TXT before enabling KVM\n");
2078 /* launched w/o TXT and VMX disabled */
2079 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2080 && !tboot_enabled())
2087 static void kvm_cpu_vmxon(u64 addr)
2089 cr4_set_bits(X86_CR4_VMXE);
2090 intel_pt_handle_vmx(1);
2092 asm volatile ("vmxon %0" : : "m"(addr));
2095 static int hardware_enable(void)
2097 int cpu = raw_smp_processor_id();
2098 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2101 if (cr4_read_shadow() & X86_CR4_VMXE)
2105 * This can happen if we hot-added a CPU but failed to allocate
2106 * VP assist page for it.
2108 if (static_branch_unlikely(&enable_evmcs) &&
2109 !hv_get_vp_assist_page(cpu))
2112 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
2113 INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
2114 spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
2117 * Now we can enable the vmclear operation in kdump
2118 * since the loaded_vmcss_on_cpu list on this cpu
2119 * has been initialized.
2121 * Though the cpu is not in VMX operation now, there
2122 * is no problem to enable the vmclear operation
2123 * for the loaded_vmcss_on_cpu list is empty!
2125 crash_enable_local_vmclear(cpu);
2127 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
2129 test_bits = FEATURE_CONTROL_LOCKED;
2130 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
2131 if (tboot_enabled())
2132 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
2134 if ((old & test_bits) != test_bits) {
2135 /* enable and lock */
2136 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
2138 kvm_cpu_vmxon(phys_addr);
2145 static void vmclear_local_loaded_vmcss(void)
2147 int cpu = raw_smp_processor_id();
2148 struct loaded_vmcs *v, *n;
2150 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2151 loaded_vmcss_on_cpu_link)
2152 __loaded_vmcs_clear(v);
2156 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2159 static void kvm_cpu_vmxoff(void)
2161 asm volatile (__ex("vmxoff"));
2163 intel_pt_handle_vmx(0);
2164 cr4_clear_bits(X86_CR4_VMXE);
2167 static void hardware_disable(void)
2169 vmclear_local_loaded_vmcss();
2173 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2174 u32 msr, u32 *result)
2176 u32 vmx_msr_low, vmx_msr_high;
2177 u32 ctl = ctl_min | ctl_opt;
2179 rdmsr(msr, vmx_msr_low, vmx_msr_high);
2181 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2182 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
2184 /* Ensure minimum (required) set of control bits are supported. */
2192 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf,
2193 struct vmx_capability *vmx_cap)
2195 u32 vmx_msr_low, vmx_msr_high;
2196 u32 min, opt, min2, opt2;
2197 u32 _pin_based_exec_control = 0;
2198 u32 _cpu_based_exec_control = 0;
2199 u32 _cpu_based_2nd_exec_control = 0;
2200 u32 _vmexit_control = 0;
2201 u32 _vmentry_control = 0;
2203 memset(vmcs_conf, 0, sizeof(*vmcs_conf));
2204 min = CPU_BASED_HLT_EXITING |
2205 #ifdef CONFIG_X86_64
2206 CPU_BASED_CR8_LOAD_EXITING |
2207 CPU_BASED_CR8_STORE_EXITING |
2209 CPU_BASED_CR3_LOAD_EXITING |
2210 CPU_BASED_CR3_STORE_EXITING |
2211 CPU_BASED_UNCOND_IO_EXITING |
2212 CPU_BASED_MOV_DR_EXITING |
2213 CPU_BASED_USE_TSC_OFFSETING |
2214 CPU_BASED_MWAIT_EXITING |
2215 CPU_BASED_MONITOR_EXITING |
2216 CPU_BASED_INVLPG_EXITING |
2217 CPU_BASED_RDPMC_EXITING;
2219 opt = CPU_BASED_TPR_SHADOW |
2220 CPU_BASED_USE_MSR_BITMAPS |
2221 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2222 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2223 &_cpu_based_exec_control) < 0)
2225 #ifdef CONFIG_X86_64
2226 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2227 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2228 ~CPU_BASED_CR8_STORE_EXITING;
2230 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2232 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2233 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2234 SECONDARY_EXEC_WBINVD_EXITING |
2235 SECONDARY_EXEC_ENABLE_VPID |
2236 SECONDARY_EXEC_ENABLE_EPT |
2237 SECONDARY_EXEC_UNRESTRICTED_GUEST |
2238 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2239 SECONDARY_EXEC_DESC |
2240 SECONDARY_EXEC_RDTSCP |
2241 SECONDARY_EXEC_ENABLE_INVPCID |
2242 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2243 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2244 SECONDARY_EXEC_SHADOW_VMCS |
2245 SECONDARY_EXEC_XSAVES |
2246 SECONDARY_EXEC_RDSEED_EXITING |
2247 SECONDARY_EXEC_RDRAND_EXITING |
2248 SECONDARY_EXEC_ENABLE_PML |
2249 SECONDARY_EXEC_TSC_SCALING |
2250 SECONDARY_EXEC_PT_USE_GPA |
2251 SECONDARY_EXEC_PT_CONCEAL_VMX |
2252 SECONDARY_EXEC_ENABLE_VMFUNC |
2253 SECONDARY_EXEC_ENCLS_EXITING;
2254 if (adjust_vmx_controls(min2, opt2,
2255 MSR_IA32_VMX_PROCBASED_CTLS2,
2256 &_cpu_based_2nd_exec_control) < 0)
2259 #ifndef CONFIG_X86_64
2260 if (!(_cpu_based_2nd_exec_control &
2261 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2262 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2265 if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2266 _cpu_based_2nd_exec_control &= ~(
2267 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2268 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2269 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
2271 rdmsr_safe(MSR_IA32_VMX_EPT_VPID_CAP,
2272 &vmx_cap->ept, &vmx_cap->vpid);
2274 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2275 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2277 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2278 CPU_BASED_CR3_STORE_EXITING |
2279 CPU_BASED_INVLPG_EXITING);
2280 } else if (vmx_cap->ept) {
2282 pr_warn_once("EPT CAP should not exist if not support "
2283 "1-setting enable EPT VM-execution control\n");
2285 if (!(_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_VPID) &&
2288 pr_warn_once("VPID CAP should not exist if not support "
2289 "1-setting enable VPID VM-execution control\n");
2292 min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
2293 #ifdef CONFIG_X86_64
2294 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2296 opt = VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL |
2297 VM_EXIT_LOAD_IA32_PAT |
2298 VM_EXIT_LOAD_IA32_EFER |
2299 VM_EXIT_CLEAR_BNDCFGS |
2300 VM_EXIT_PT_CONCEAL_PIP |
2301 VM_EXIT_CLEAR_IA32_RTIT_CTL;
2302 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2303 &_vmexit_control) < 0)
2306 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2307 opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR |
2308 PIN_BASED_VMX_PREEMPTION_TIMER;
2309 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2310 &_pin_based_exec_control) < 0)
2313 if (cpu_has_broken_vmx_preemption_timer())
2314 _pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
2315 if (!(_cpu_based_2nd_exec_control &
2316 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
2317 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
2319 min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
2320 opt = VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL |
2321 VM_ENTRY_LOAD_IA32_PAT |
2322 VM_ENTRY_LOAD_IA32_EFER |
2323 VM_ENTRY_LOAD_BNDCFGS |
2324 VM_ENTRY_PT_CONCEAL_PIP |
2325 VM_ENTRY_LOAD_IA32_RTIT_CTL;
2326 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2327 &_vmentry_control) < 0)
2331 * Some cpus support VM_{ENTRY,EXIT}_IA32_PERF_GLOBAL_CTRL but they
2332 * can't be used due to an errata where VM Exit may incorrectly clear
2333 * IA32_PERF_GLOBAL_CTRL[34:32]. Workaround the errata by using the
2334 * MSR load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2336 if (boot_cpu_data.x86 == 0x6) {
2337 switch (boot_cpu_data.x86_model) {
2338 case 26: /* AAK155 */
2339 case 30: /* AAP115 */
2340 case 37: /* AAT100 */
2341 case 44: /* BC86,AAY89,BD102 */
2343 _vmentry_control &= ~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
2344 _vmexit_control &= ~VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
2345 pr_warn_once("kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
2346 "does not work properly. Using workaround\n");
2354 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2356 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2357 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2360 #ifdef CONFIG_X86_64
2361 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2362 if (vmx_msr_high & (1u<<16))
2366 /* Require Write-Back (WB) memory type for VMCS accesses. */
2367 if (((vmx_msr_high >> 18) & 15) != 6)
2370 vmcs_conf->size = vmx_msr_high & 0x1fff;
2371 vmcs_conf->order = get_order(vmcs_conf->size);
2372 vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
2374 vmcs_conf->revision_id = vmx_msr_low;
2376 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2377 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2378 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2379 vmcs_conf->vmexit_ctrl = _vmexit_control;
2380 vmcs_conf->vmentry_ctrl = _vmentry_control;
2382 if (static_branch_unlikely(&enable_evmcs))
2383 evmcs_sanitize_exec_ctrls(vmcs_conf);
2388 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags)
2390 int node = cpu_to_node(cpu);
2394 pages = __alloc_pages_node(node, flags, vmcs_config.order);
2397 vmcs = page_address(pages);
2398 memset(vmcs, 0, vmcs_config.size);
2400 /* KVM supports Enlightened VMCS v1 only */
2401 if (static_branch_unlikely(&enable_evmcs))
2402 vmcs->hdr.revision_id = KVM_EVMCS_VERSION;
2404 vmcs->hdr.revision_id = vmcs_config.revision_id;
2407 vmcs->hdr.shadow_vmcs = 1;
2411 void free_vmcs(struct vmcs *vmcs)
2413 free_pages((unsigned long)vmcs, vmcs_config.order);
2417 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2419 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2421 if (!loaded_vmcs->vmcs)
2423 loaded_vmcs_clear(loaded_vmcs);
2424 free_vmcs(loaded_vmcs->vmcs);
2425 loaded_vmcs->vmcs = NULL;
2426 if (loaded_vmcs->msr_bitmap)
2427 free_page((unsigned long)loaded_vmcs->msr_bitmap);
2428 WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
2431 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2433 loaded_vmcs->vmcs = alloc_vmcs(false);
2434 if (!loaded_vmcs->vmcs)
2437 loaded_vmcs->shadow_vmcs = NULL;
2438 loaded_vmcs_init(loaded_vmcs);
2440 if (cpu_has_vmx_msr_bitmap()) {
2441 loaded_vmcs->msr_bitmap = (unsigned long *)
2442 __get_free_page(GFP_KERNEL_ACCOUNT);
2443 if (!loaded_vmcs->msr_bitmap)
2445 memset(loaded_vmcs->msr_bitmap, 0xff, PAGE_SIZE);
2447 if (IS_ENABLED(CONFIG_HYPERV) &&
2448 static_branch_unlikely(&enable_evmcs) &&
2449 (ms_hyperv.nested_features & HV_X64_NESTED_MSR_BITMAP)) {
2450 struct hv_enlightened_vmcs *evmcs =
2451 (struct hv_enlightened_vmcs *)loaded_vmcs->vmcs;
2453 evmcs->hv_enlightenments_control.msr_bitmap = 1;
2457 memset(&loaded_vmcs->host_state, 0, sizeof(struct vmcs_host_state));
2462 free_loaded_vmcs(loaded_vmcs);
2466 static void free_kvm_area(void)
2470 for_each_possible_cpu(cpu) {
2471 free_vmcs(per_cpu(vmxarea, cpu));
2472 per_cpu(vmxarea, cpu) = NULL;
2476 static __init int alloc_kvm_area(void)
2480 for_each_possible_cpu(cpu) {
2483 vmcs = alloc_vmcs_cpu(false, cpu, GFP_KERNEL);
2490 * When eVMCS is enabled, alloc_vmcs_cpu() sets
2491 * vmcs->revision_id to KVM_EVMCS_VERSION instead of
2492 * revision_id reported by MSR_IA32_VMX_BASIC.
2494 * However, even though not explicitly documented by
2495 * TLFS, VMXArea passed as VMXON argument should
2496 * still be marked with revision_id reported by
2499 if (static_branch_unlikely(&enable_evmcs))
2500 vmcs->hdr.revision_id = vmcs_config.revision_id;
2502 per_cpu(vmxarea, cpu) = vmcs;
2507 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
2508 struct kvm_segment *save)
2510 if (!emulate_invalid_guest_state) {
2512 * CS and SS RPL should be equal during guest entry according
2513 * to VMX spec, but in reality it is not always so. Since vcpu
2514 * is in the middle of the transition from real mode to
2515 * protected mode it is safe to assume that RPL 0 is a good
2518 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
2519 save->selector &= ~SEGMENT_RPL_MASK;
2520 save->dpl = save->selector & SEGMENT_RPL_MASK;
2523 vmx_set_segment(vcpu, save, seg);
2526 static void enter_pmode(struct kvm_vcpu *vcpu)
2528 unsigned long flags;
2529 struct vcpu_vmx *vmx = to_vmx(vcpu);
2532 * Update real mode segment cache. It may be not up-to-date if sement
2533 * register was written while vcpu was in a guest mode.
2535 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2536 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2537 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2538 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2539 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2540 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2542 vmx->rmode.vm86_active = 0;
2544 vmx_segment_cache_clear(vmx);
2546 vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2548 flags = vmcs_readl(GUEST_RFLAGS);
2549 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2550 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2551 vmcs_writel(GUEST_RFLAGS, flags);
2553 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
2554 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
2556 update_exception_bitmap(vcpu);
2558 fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2559 fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2560 fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2561 fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2562 fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2563 fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2566 static void fix_rmode_seg(int seg, struct kvm_segment *save)
2568 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2569 struct kvm_segment var = *save;
2572 if (seg == VCPU_SREG_CS)
2575 if (!emulate_invalid_guest_state) {
2576 var.selector = var.base >> 4;
2577 var.base = var.base & 0xffff0;
2587 if (save->base & 0xf)
2588 printk_once(KERN_WARNING "kvm: segment base is not "
2589 "paragraph aligned when entering "
2590 "protected mode (seg=%d)", seg);
2593 vmcs_write16(sf->selector, var.selector);
2594 vmcs_writel(sf->base, var.base);
2595 vmcs_write32(sf->limit, var.limit);
2596 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
2599 static void enter_rmode(struct kvm_vcpu *vcpu)
2601 unsigned long flags;
2602 struct vcpu_vmx *vmx = to_vmx(vcpu);
2603 struct kvm_vmx *kvm_vmx = to_kvm_vmx(vcpu->kvm);
2605 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2606 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2607 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2608 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2609 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2610 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2611 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2613 vmx->rmode.vm86_active = 1;
2616 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
2617 * vcpu. Warn the user that an update is overdue.
2619 if (!kvm_vmx->tss_addr)
2620 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
2621 "called before entering vcpu\n");
2623 vmx_segment_cache_clear(vmx);
2625 vmcs_writel(GUEST_TR_BASE, kvm_vmx->tss_addr);
2626 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
2627 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2629 flags = vmcs_readl(GUEST_RFLAGS);
2630 vmx->rmode.save_rflags = flags;
2632 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2634 vmcs_writel(GUEST_RFLAGS, flags);
2635 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
2636 update_exception_bitmap(vcpu);
2638 fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2639 fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2640 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2641 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2642 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2643 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2645 kvm_mmu_reset_context(vcpu);
2648 void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
2650 struct vcpu_vmx *vmx = to_vmx(vcpu);
2651 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
2656 vcpu->arch.efer = efer;
2657 if (efer & EFER_LMA) {
2658 vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2661 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2663 msr->data = efer & ~EFER_LME;
2668 #ifdef CONFIG_X86_64
2670 static void enter_lmode(struct kvm_vcpu *vcpu)
2674 vmx_segment_cache_clear(to_vmx(vcpu));
2676 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
2677 if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
2678 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
2680 vmcs_write32(GUEST_TR_AR_BYTES,
2681 (guest_tr_ar & ~VMX_AR_TYPE_MASK)
2682 | VMX_AR_TYPE_BUSY_64_TSS);
2684 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
2687 static void exit_lmode(struct kvm_vcpu *vcpu)
2689 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2690 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
2695 static void vmx_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t addr)
2697 int vpid = to_vmx(vcpu)->vpid;
2699 if (!vpid_sync_vcpu_addr(vpid, addr))
2700 vpid_sync_context(vpid);
2703 * If VPIDs are not supported or enabled, then the above is a no-op.
2704 * But we don't really need a TLB flush in that case anyway, because
2705 * each VM entry/exit includes an implicit flush when VPID is 0.
2709 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
2711 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
2713 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
2714 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
2717 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
2719 if (enable_unrestricted_guest || (enable_ept && is_paging(vcpu)))
2720 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2721 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
2724 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
2726 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
2728 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
2729 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
2732 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
2734 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2736 if (!test_bit(VCPU_EXREG_PDPTR,
2737 (unsigned long *)&vcpu->arch.regs_dirty))
2740 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
2741 vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
2742 vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
2743 vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
2744 vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
2748 void ept_save_pdptrs(struct kvm_vcpu *vcpu)
2750 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2752 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
2753 mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
2754 mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
2755 mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
2756 mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
2759 __set_bit(VCPU_EXREG_PDPTR,
2760 (unsigned long *)&vcpu->arch.regs_avail);
2761 __set_bit(VCPU_EXREG_PDPTR,
2762 (unsigned long *)&vcpu->arch.regs_dirty);
2765 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
2767 struct kvm_vcpu *vcpu)
2769 if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
2770 vmx_decache_cr3(vcpu);
2771 if (!(cr0 & X86_CR0_PG)) {
2772 /* From paging/starting to nonpaging */
2773 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
2774 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
2775 (CPU_BASED_CR3_LOAD_EXITING |
2776 CPU_BASED_CR3_STORE_EXITING));
2777 vcpu->arch.cr0 = cr0;
2778 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2779 } else if (!is_paging(vcpu)) {
2780 /* From nonpaging to paging */
2781 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
2782 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
2783 ~(CPU_BASED_CR3_LOAD_EXITING |
2784 CPU_BASED_CR3_STORE_EXITING));
2785 vcpu->arch.cr0 = cr0;
2786 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2789 if (!(cr0 & X86_CR0_WP))
2790 *hw_cr0 &= ~X86_CR0_WP;
2793 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
2795 struct vcpu_vmx *vmx = to_vmx(vcpu);
2796 unsigned long hw_cr0;
2798 hw_cr0 = (cr0 & ~KVM_VM_CR0_ALWAYS_OFF);
2799 if (enable_unrestricted_guest)
2800 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
2802 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
2804 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
2807 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
2811 #ifdef CONFIG_X86_64
2812 if (vcpu->arch.efer & EFER_LME) {
2813 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
2815 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
2820 if (enable_ept && !enable_unrestricted_guest)
2821 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
2823 vmcs_writel(CR0_READ_SHADOW, cr0);
2824 vmcs_writel(GUEST_CR0, hw_cr0);
2825 vcpu->arch.cr0 = cr0;
2827 /* depends on vcpu->arch.cr0 to be set to a new value */
2828 vmx->emulation_required = emulation_required(vcpu);
2831 static int get_ept_level(struct kvm_vcpu *vcpu)
2833 if (cpu_has_vmx_ept_5levels() && (cpuid_maxphyaddr(vcpu) > 48))
2838 u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa)
2840 u64 eptp = VMX_EPTP_MT_WB;
2842 eptp |= (get_ept_level(vcpu) == 5) ? VMX_EPTP_PWL_5 : VMX_EPTP_PWL_4;
2844 if (enable_ept_ad_bits &&
2845 (!is_guest_mode(vcpu) || nested_ept_ad_enabled(vcpu)))
2846 eptp |= VMX_EPTP_AD_ENABLE_BIT;
2847 eptp |= (root_hpa & PAGE_MASK);
2852 void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
2854 struct kvm *kvm = vcpu->kvm;
2855 unsigned long guest_cr3;
2860 eptp = construct_eptp(vcpu, cr3);
2861 vmcs_write64(EPT_POINTER, eptp);
2863 if (kvm_x86_ops->tlb_remote_flush) {
2864 spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
2865 to_vmx(vcpu)->ept_pointer = eptp;
2866 to_kvm_vmx(kvm)->ept_pointers_match
2867 = EPT_POINTERS_CHECK;
2868 spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
2871 if (enable_unrestricted_guest || is_paging(vcpu) ||
2872 is_guest_mode(vcpu))
2873 guest_cr3 = kvm_read_cr3(vcpu);
2875 guest_cr3 = to_kvm_vmx(kvm)->ept_identity_map_addr;
2876 ept_load_pdptrs(vcpu);
2879 vmcs_writel(GUEST_CR3, guest_cr3);
2882 int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
2885 * Pass through host's Machine Check Enable value to hw_cr4, which
2886 * is in force while we are in guest mode. Do not let guests control
2887 * this bit, even if host CR4.MCE == 0.
2889 unsigned long hw_cr4;
2891 hw_cr4 = (cr4_read_shadow() & X86_CR4_MCE) | (cr4 & ~X86_CR4_MCE);
2892 if (enable_unrestricted_guest)
2893 hw_cr4 |= KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST;
2894 else if (to_vmx(vcpu)->rmode.vm86_active)
2895 hw_cr4 |= KVM_RMODE_VM_CR4_ALWAYS_ON;
2897 hw_cr4 |= KVM_PMODE_VM_CR4_ALWAYS_ON;
2899 if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated()) {
2900 if (cr4 & X86_CR4_UMIP) {
2901 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
2902 SECONDARY_EXEC_DESC);
2903 hw_cr4 &= ~X86_CR4_UMIP;
2904 } else if (!is_guest_mode(vcpu) ||
2905 !nested_cpu_has2(get_vmcs12(vcpu), SECONDARY_EXEC_DESC))
2906 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
2907 SECONDARY_EXEC_DESC);
2910 if (cr4 & X86_CR4_VMXE) {
2912 * To use VMXON (and later other VMX instructions), a guest
2913 * must first be able to turn on cr4.VMXE (see handle_vmon()).
2914 * So basically the check on whether to allow nested VMX
2915 * is here. We operate under the default treatment of SMM,
2916 * so VMX cannot be enabled under SMM.
2918 if (!nested_vmx_allowed(vcpu) || is_smm(vcpu))
2922 if (to_vmx(vcpu)->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
2925 vcpu->arch.cr4 = cr4;
2927 if (!enable_unrestricted_guest) {
2929 if (!is_paging(vcpu)) {
2930 hw_cr4 &= ~X86_CR4_PAE;
2931 hw_cr4 |= X86_CR4_PSE;
2932 } else if (!(cr4 & X86_CR4_PAE)) {
2933 hw_cr4 &= ~X86_CR4_PAE;
2938 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
2939 * hardware. To emulate this behavior, SMEP/SMAP/PKU needs
2940 * to be manually disabled when guest switches to non-paging
2943 * If !enable_unrestricted_guest, the CPU is always running
2944 * with CR0.PG=1 and CR4 needs to be modified.
2945 * If enable_unrestricted_guest, the CPU automatically
2946 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
2948 if (!is_paging(vcpu))
2949 hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
2952 vmcs_writel(CR4_READ_SHADOW, cr4);
2953 vmcs_writel(GUEST_CR4, hw_cr4);
2957 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
2959 struct vcpu_vmx *vmx = to_vmx(vcpu);
2962 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
2963 *var = vmx->rmode.segs[seg];
2964 if (seg == VCPU_SREG_TR
2965 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
2967 var->base = vmx_read_guest_seg_base(vmx, seg);
2968 var->selector = vmx_read_guest_seg_selector(vmx, seg);
2971 var->base = vmx_read_guest_seg_base(vmx, seg);
2972 var->limit = vmx_read_guest_seg_limit(vmx, seg);
2973 var->selector = vmx_read_guest_seg_selector(vmx, seg);
2974 ar = vmx_read_guest_seg_ar(vmx, seg);
2975 var->unusable = (ar >> 16) & 1;
2976 var->type = ar & 15;
2977 var->s = (ar >> 4) & 1;
2978 var->dpl = (ar >> 5) & 3;
2980 * Some userspaces do not preserve unusable property. Since usable
2981 * segment has to be present according to VMX spec we can use present
2982 * property to amend userspace bug by making unusable segment always
2983 * nonpresent. vmx_segment_access_rights() already marks nonpresent
2984 * segment as unusable.
2986 var->present = !var->unusable;
2987 var->avl = (ar >> 12) & 1;
2988 var->l = (ar >> 13) & 1;
2989 var->db = (ar >> 14) & 1;
2990 var->g = (ar >> 15) & 1;
2993 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
2995 struct kvm_segment s;
2997 if (to_vmx(vcpu)->rmode.vm86_active) {
2998 vmx_get_segment(vcpu, &s, seg);
3001 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3004 int vmx_get_cpl(struct kvm_vcpu *vcpu)
3006 struct vcpu_vmx *vmx = to_vmx(vcpu);
3008 if (unlikely(vmx->rmode.vm86_active))
3011 int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
3012 return VMX_AR_DPL(ar);
3016 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3020 if (var->unusable || !var->present)
3023 ar = var->type & 15;
3024 ar |= (var->s & 1) << 4;
3025 ar |= (var->dpl & 3) << 5;
3026 ar |= (var->present & 1) << 7;
3027 ar |= (var->avl & 1) << 12;
3028 ar |= (var->l & 1) << 13;
3029 ar |= (var->db & 1) << 14;
3030 ar |= (var->g & 1) << 15;
3036 void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3038 struct vcpu_vmx *vmx = to_vmx(vcpu);
3039 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3041 vmx_segment_cache_clear(vmx);
3043 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3044 vmx->rmode.segs[seg] = *var;
3045 if (seg == VCPU_SREG_TR)
3046 vmcs_write16(sf->selector, var->selector);
3048 fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3052 vmcs_writel(sf->base, var->base);
3053 vmcs_write32(sf->limit, var->limit);
3054 vmcs_write16(sf->selector, var->selector);
3057 * Fix the "Accessed" bit in AR field of segment registers for older
3059 * IA32 arch specifies that at the time of processor reset the
3060 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3061 * is setting it to 0 in the userland code. This causes invalid guest
3062 * state vmexit when "unrestricted guest" mode is turned on.
3063 * Fix for this setup issue in cpu_reset is being pushed in the qemu
3064 * tree. Newer qemu binaries with that qemu fix would not need this
3067 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3068 var->type |= 0x1; /* Accessed */
3070 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3073 vmx->emulation_required = emulation_required(vcpu);
3076 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3078 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3080 *db = (ar >> 14) & 1;
3081 *l = (ar >> 13) & 1;
3084 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3086 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3087 dt->address = vmcs_readl(GUEST_IDTR_BASE);
3090 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3092 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3093 vmcs_writel(GUEST_IDTR_BASE, dt->address);
3096 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3098 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3099 dt->address = vmcs_readl(GUEST_GDTR_BASE);
3102 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3104 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3105 vmcs_writel(GUEST_GDTR_BASE, dt->address);
3108 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3110 struct kvm_segment var;
3113 vmx_get_segment(vcpu, &var, seg);
3115 if (seg == VCPU_SREG_CS)
3117 ar = vmx_segment_access_rights(&var);
3119 if (var.base != (var.selector << 4))
3121 if (var.limit != 0xffff)
3129 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3131 struct kvm_segment cs;
3132 unsigned int cs_rpl;
3134 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3135 cs_rpl = cs.selector & SEGMENT_RPL_MASK;
3139 if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
3143 if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
3144 if (cs.dpl > cs_rpl)
3147 if (cs.dpl != cs_rpl)
3153 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3157 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3159 struct kvm_segment ss;
3160 unsigned int ss_rpl;
3162 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3163 ss_rpl = ss.selector & SEGMENT_RPL_MASK;
3167 if (ss.type != 3 && ss.type != 7)
3171 if (ss.dpl != ss_rpl) /* DPL != RPL */
3179 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3181 struct kvm_segment var;
3184 vmx_get_segment(vcpu, &var, seg);
3185 rpl = var.selector & SEGMENT_RPL_MASK;
3193 if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
3194 if (var.dpl < rpl) /* DPL < RPL */
3198 /* TODO: Add other members to kvm_segment_field to allow checking for other access
3204 static bool tr_valid(struct kvm_vcpu *vcpu)
3206 struct kvm_segment tr;
3208 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3212 if (tr.selector & SEGMENT_TI_MASK) /* TI = 1 */
3214 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3222 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3224 struct kvm_segment ldtr;
3226 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3230 if (ldtr.selector & SEGMENT_TI_MASK) /* TI = 1 */
3240 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3242 struct kvm_segment cs, ss;
3244 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3245 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3247 return ((cs.selector & SEGMENT_RPL_MASK) ==
3248 (ss.selector & SEGMENT_RPL_MASK));
3252 * Check if guest state is valid. Returns true if valid, false if
3254 * We assume that registers are always usable
3256 static bool guest_state_valid(struct kvm_vcpu *vcpu)
3258 if (enable_unrestricted_guest)
3261 /* real mode guest state checks */
3262 if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
3263 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3265 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3267 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3269 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3271 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3273 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3276 /* protected mode guest state checks */
3277 if (!cs_ss_rpl_check(vcpu))
3279 if (!code_segment_valid(vcpu))
3281 if (!stack_segment_valid(vcpu))
3283 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3285 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3287 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3289 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3291 if (!tr_valid(vcpu))
3293 if (!ldtr_valid(vcpu))
3297 * - Add checks on RIP
3298 * - Add checks on RFLAGS
3304 static int init_rmode_tss(struct kvm *kvm)
3310 idx = srcu_read_lock(&kvm->srcu);
3311 fn = to_kvm_vmx(kvm)->tss_addr >> PAGE_SHIFT;
3312 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3315 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3316 r = kvm_write_guest_page(kvm, fn++, &data,
3317 TSS_IOPB_BASE_OFFSET, sizeof(u16));
3320 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3323 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3327 r = kvm_write_guest_page(kvm, fn, &data,
3328 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3331 srcu_read_unlock(&kvm->srcu, idx);
3335 static int init_rmode_identity_map(struct kvm *kvm)
3337 struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm);
3339 kvm_pfn_t identity_map_pfn;
3342 /* Protect kvm_vmx->ept_identity_pagetable_done. */
3343 mutex_lock(&kvm->slots_lock);
3345 if (likely(kvm_vmx->ept_identity_pagetable_done))
3348 if (!kvm_vmx->ept_identity_map_addr)
3349 kvm_vmx->ept_identity_map_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
3350 identity_map_pfn = kvm_vmx->ept_identity_map_addr >> PAGE_SHIFT;
3352 r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
3353 kvm_vmx->ept_identity_map_addr, PAGE_SIZE);
3357 idx = srcu_read_lock(&kvm->srcu);
3358 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3361 /* Set up identity-mapping pagetable for EPT in real mode */
3362 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3363 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3364 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3365 r = kvm_write_guest_page(kvm, identity_map_pfn,
3366 &tmp, i * sizeof(tmp), sizeof(tmp));
3370 kvm_vmx->ept_identity_pagetable_done = true;
3373 srcu_read_unlock(&kvm->srcu, idx);
3376 mutex_unlock(&kvm->slots_lock);
3380 static void seg_setup(int seg)
3382 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3385 vmcs_write16(sf->selector, 0);
3386 vmcs_writel(sf->base, 0);
3387 vmcs_write32(sf->limit, 0xffff);
3389 if (seg == VCPU_SREG_CS)
3390 ar |= 0x08; /* code segment */
3392 vmcs_write32(sf->ar_bytes, ar);
3395 static int alloc_apic_access_page(struct kvm *kvm)
3400 mutex_lock(&kvm->slots_lock);
3401 if (kvm->arch.apic_access_page_done)
3403 r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
3404 APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
3408 page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
3409 if (is_error_page(page)) {
3415 * Do not pin the page in memory, so that memory hot-unplug
3416 * is able to migrate it.
3419 kvm->arch.apic_access_page_done = true;
3421 mutex_unlock(&kvm->slots_lock);
3425 int allocate_vpid(void)
3431 spin_lock(&vmx_vpid_lock);
3432 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3433 if (vpid < VMX_NR_VPIDS)
3434 __set_bit(vpid, vmx_vpid_bitmap);
3437 spin_unlock(&vmx_vpid_lock);
3441 void free_vpid(int vpid)
3443 if (!enable_vpid || vpid == 0)
3445 spin_lock(&vmx_vpid_lock);
3446 __clear_bit(vpid, vmx_vpid_bitmap);
3447 spin_unlock(&vmx_vpid_lock);
3450 static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
3453 int f = sizeof(unsigned long);
3455 if (!cpu_has_vmx_msr_bitmap())
3458 if (static_branch_unlikely(&enable_evmcs))
3459 evmcs_touch_msr_bitmap();
3462 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3463 * have the write-low and read-high bitmap offsets the wrong way round.
3464 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3466 if (msr <= 0x1fff) {
3467 if (type & MSR_TYPE_R)
3469 __clear_bit(msr, msr_bitmap + 0x000 / f);
3471 if (type & MSR_TYPE_W)
3473 __clear_bit(msr, msr_bitmap + 0x800 / f);
3475 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3477 if (type & MSR_TYPE_R)
3479 __clear_bit(msr, msr_bitmap + 0x400 / f);
3481 if (type & MSR_TYPE_W)
3483 __clear_bit(msr, msr_bitmap + 0xc00 / f);
3488 static __always_inline void vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
3491 int f = sizeof(unsigned long);
3493 if (!cpu_has_vmx_msr_bitmap())
3496 if (static_branch_unlikely(&enable_evmcs))
3497 evmcs_touch_msr_bitmap();
3500 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3501 * have the write-low and read-high bitmap offsets the wrong way round.
3502 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3504 if (msr <= 0x1fff) {
3505 if (type & MSR_TYPE_R)
3507 __set_bit(msr, msr_bitmap + 0x000 / f);
3509 if (type & MSR_TYPE_W)
3511 __set_bit(msr, msr_bitmap + 0x800 / f);
3513 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3515 if (type & MSR_TYPE_R)
3517 __set_bit(msr, msr_bitmap + 0x400 / f);
3519 if (type & MSR_TYPE_W)
3521 __set_bit(msr, msr_bitmap + 0xc00 / f);
3526 static __always_inline void vmx_set_intercept_for_msr(unsigned long *msr_bitmap,
3527 u32 msr, int type, bool value)
3530 vmx_enable_intercept_for_msr(msr_bitmap, msr, type);
3532 vmx_disable_intercept_for_msr(msr_bitmap, msr, type);
3535 static u8 vmx_msr_bitmap_mode(struct kvm_vcpu *vcpu)
3539 if (cpu_has_secondary_exec_ctrls() &&
3540 (vmcs_read32(SECONDARY_VM_EXEC_CONTROL) &
3541 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
3542 mode |= MSR_BITMAP_MODE_X2APIC;
3543 if (enable_apicv && kvm_vcpu_apicv_active(vcpu))
3544 mode |= MSR_BITMAP_MODE_X2APIC_APICV;
3550 static void vmx_update_msr_bitmap_x2apic(unsigned long *msr_bitmap,
3555 for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
3556 unsigned word = msr / BITS_PER_LONG;
3557 msr_bitmap[word] = (mode & MSR_BITMAP_MODE_X2APIC_APICV) ? 0 : ~0;
3558 msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
3561 if (mode & MSR_BITMAP_MODE_X2APIC) {
3563 * TPR reads and writes can be virtualized even if virtual interrupt
3564 * delivery is not in use.
3566 vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TASKPRI), MSR_TYPE_RW);
3567 if (mode & MSR_BITMAP_MODE_X2APIC_APICV) {
3568 vmx_enable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TMCCT), MSR_TYPE_R);
3569 vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_EOI), MSR_TYPE_W);
3570 vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_SELF_IPI), MSR_TYPE_W);
3575 void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu)
3577 struct vcpu_vmx *vmx = to_vmx(vcpu);
3578 unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3579 u8 mode = vmx_msr_bitmap_mode(vcpu);
3580 u8 changed = mode ^ vmx->msr_bitmap_mode;
3585 if (changed & (MSR_BITMAP_MODE_X2APIC | MSR_BITMAP_MODE_X2APIC_APICV))
3586 vmx_update_msr_bitmap_x2apic(msr_bitmap, mode);
3588 vmx->msr_bitmap_mode = mode;
3591 void pt_update_intercept_for_msr(struct vcpu_vmx *vmx)
3593 unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3594 bool flag = !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
3597 vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_STATUS,
3599 vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_BASE,
3601 vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_MASK,
3603 vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_CR3_MATCH,
3605 for (i = 0; i < vmx->pt_desc.addr_range; i++) {
3606 vmx_set_intercept_for_msr(msr_bitmap,
3607 MSR_IA32_RTIT_ADDR0_A + i * 2, MSR_TYPE_RW, flag);
3608 vmx_set_intercept_for_msr(msr_bitmap,
3609 MSR_IA32_RTIT_ADDR0_B + i * 2, MSR_TYPE_RW, flag);
3613 static bool vmx_get_enable_apicv(struct kvm_vcpu *vcpu)
3615 return enable_apicv;
3618 static bool vmx_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
3620 struct vcpu_vmx *vmx = to_vmx(vcpu);
3625 if (WARN_ON_ONCE(!is_guest_mode(vcpu)) ||
3626 !nested_cpu_has_vid(get_vmcs12(vcpu)) ||
3627 WARN_ON_ONCE(!vmx->nested.virtual_apic_map.gfn))
3630 rvi = vmx_get_rvi();
3632 vapic_page = vmx->nested.virtual_apic_map.hva;
3633 vppr = *((u32 *)(vapic_page + APIC_PROCPRI));
3635 return ((rvi & 0xf0) > (vppr & 0xf0));
3638 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu,
3642 int pi_vec = nested ? POSTED_INTR_NESTED_VECTOR : POSTED_INTR_VECTOR;
3644 if (vcpu->mode == IN_GUEST_MODE) {
3646 * The vector of interrupt to be delivered to vcpu had
3647 * been set in PIR before this function.
3649 * Following cases will be reached in this block, and
3650 * we always send a notification event in all cases as
3653 * Case 1: vcpu keeps in non-root mode. Sending a
3654 * notification event posts the interrupt to vcpu.
3656 * Case 2: vcpu exits to root mode and is still
3657 * runnable. PIR will be synced to vIRR before the
3658 * next vcpu entry. Sending a notification event in
3659 * this case has no effect, as vcpu is not in root
3662 * Case 3: vcpu exits to root mode and is blocked.
3663 * vcpu_block() has already synced PIR to vIRR and
3664 * never blocks vcpu if vIRR is not cleared. Therefore,
3665 * a blocked vcpu here does not wait for any requested
3666 * interrupts in PIR, and sending a notification event
3667 * which has no effect is safe here.
3670 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu), pi_vec);
3677 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
3680 struct vcpu_vmx *vmx = to_vmx(vcpu);
3682 if (is_guest_mode(vcpu) &&
3683 vector == vmx->nested.posted_intr_nv) {
3685 * If a posted intr is not recognized by hardware,
3686 * we will accomplish it in the next vmentry.
3688 vmx->nested.pi_pending = true;
3689 kvm_make_request(KVM_REQ_EVENT, vcpu);
3690 /* the PIR and ON have been set by L1. */
3691 if (!kvm_vcpu_trigger_posted_interrupt(vcpu, true))
3692 kvm_vcpu_kick(vcpu);
3698 * Send interrupt to vcpu via posted interrupt way.
3699 * 1. If target vcpu is running(non-root mode), send posted interrupt
3700 * notification to vcpu and hardware will sync PIR to vIRR atomically.
3701 * 2. If target vcpu isn't running(root mode), kick it to pick up the
3702 * interrupt from PIR in next vmentry.
3704 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
3706 struct vcpu_vmx *vmx = to_vmx(vcpu);
3709 r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
3713 if (pi_test_and_set_pir(vector, &vmx->pi_desc))
3716 /* If a previous notification has sent the IPI, nothing to do. */
3717 if (pi_test_and_set_on(&vmx->pi_desc))
3720 if (!kvm_vcpu_trigger_posted_interrupt(vcpu, false))
3721 kvm_vcpu_kick(vcpu);
3725 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
3726 * will not change in the lifetime of the guest.
3727 * Note that host-state that does change is set elsewhere. E.g., host-state
3728 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
3730 void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
3735 unsigned long cr0, cr3, cr4;
3738 WARN_ON(cr0 & X86_CR0_TS);
3739 vmcs_writel(HOST_CR0, cr0); /* 22.2.3 */
3742 * Save the most likely value for this task's CR3 in the VMCS.
3743 * We can't use __get_current_cr3_fast() because we're not atomic.
3746 vmcs_writel(HOST_CR3, cr3); /* 22.2.3 FIXME: shadow tables */
3747 vmx->loaded_vmcs->host_state.cr3 = cr3;
3749 /* Save the most likely value for this task's CR4 in the VMCS. */
3750 cr4 = cr4_read_shadow();
3751 vmcs_writel(HOST_CR4, cr4); /* 22.2.3, 22.2.5 */
3752 vmx->loaded_vmcs->host_state.cr4 = cr4;
3754 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
3755 #ifdef CONFIG_X86_64
3757 * Load null selectors, so we can avoid reloading them in
3758 * vmx_prepare_switch_to_host(), in case userspace uses
3759 * the null selectors too (the expected case).
3761 vmcs_write16(HOST_DS_SELECTOR, 0);
3762 vmcs_write16(HOST_ES_SELECTOR, 0);
3764 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
3765 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
3767 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
3768 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
3771 vmcs_writel(HOST_IDTR_BASE, dt.address); /* 22.2.4 */
3772 vmx->host_idt_base = dt.address;
3774 vmcs_writel(HOST_RIP, (unsigned long)vmx_vmexit); /* 22.2.5 */
3776 rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
3777 vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
3778 rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
3779 vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */
3781 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
3782 rdmsr(MSR_IA32_CR_PAT, low32, high32);
3783 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
3786 if (cpu_has_load_ia32_efer())
3787 vmcs_write64(HOST_IA32_EFER, host_efer);
3790 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
3792 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
3794 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
3795 if (is_guest_mode(&vmx->vcpu))
3796 vmx->vcpu.arch.cr4_guest_owned_bits &=
3797 ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
3798 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
3801 static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
3803 u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
3805 if (!kvm_vcpu_apicv_active(&vmx->vcpu))
3806 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
3809 pin_based_exec_ctrl &= ~PIN_BASED_VIRTUAL_NMIS;
3811 /* Enable the preemption timer dynamically */
3812 pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
3813 return pin_based_exec_ctrl;
3816 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
3818 struct vcpu_vmx *vmx = to_vmx(vcpu);
3820 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
3821 if (cpu_has_secondary_exec_ctrls()) {
3822 if (kvm_vcpu_apicv_active(vcpu))
3823 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
3824 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3825 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3827 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
3828 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3829 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3832 if (cpu_has_vmx_msr_bitmap())
3833 vmx_update_msr_bitmap(vcpu);
3836 u32 vmx_exec_control(struct vcpu_vmx *vmx)
3838 u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
3840 if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
3841 exec_control &= ~CPU_BASED_MOV_DR_EXITING;
3843 if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
3844 exec_control &= ~CPU_BASED_TPR_SHADOW;
3845 #ifdef CONFIG_X86_64
3846 exec_control |= CPU_BASED_CR8_STORE_EXITING |
3847 CPU_BASED_CR8_LOAD_EXITING;
3851 exec_control |= CPU_BASED_CR3_STORE_EXITING |
3852 CPU_BASED_CR3_LOAD_EXITING |
3853 CPU_BASED_INVLPG_EXITING;
3854 if (kvm_mwait_in_guest(vmx->vcpu.kvm))
3855 exec_control &= ~(CPU_BASED_MWAIT_EXITING |
3856 CPU_BASED_MONITOR_EXITING);
3857 if (kvm_hlt_in_guest(vmx->vcpu.kvm))
3858 exec_control &= ~CPU_BASED_HLT_EXITING;
3859 return exec_control;
3863 static void vmx_compute_secondary_exec_control(struct vcpu_vmx *vmx)
3865 struct kvm_vcpu *vcpu = &vmx->vcpu;
3867 u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
3869 if (pt_mode == PT_MODE_SYSTEM)
3870 exec_control &= ~(SECONDARY_EXEC_PT_USE_GPA | SECONDARY_EXEC_PT_CONCEAL_VMX);
3871 if (!cpu_need_virtualize_apic_accesses(vcpu))
3872 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
3874 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
3876 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
3877 enable_unrestricted_guest = 0;
3879 if (!enable_unrestricted_guest)
3880 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
3881 if (kvm_pause_in_guest(vmx->vcpu.kvm))
3882 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
3883 if (!kvm_vcpu_apicv_active(vcpu))
3884 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
3885 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3886 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
3888 /* SECONDARY_EXEC_DESC is enabled/disabled on writes to CR4.UMIP,
3889 * in vmx_set_cr4. */
3890 exec_control &= ~SECONDARY_EXEC_DESC;
3892 /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
3894 We can NOT enable shadow_vmcs here because we don't have yet
3897 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
3900 exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
3902 if (vmx_xsaves_supported()) {
3903 /* Exposing XSAVES only when XSAVE is exposed */
3904 bool xsaves_enabled =
3905 guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
3906 guest_cpuid_has(vcpu, X86_FEATURE_XSAVES);
3908 if (!xsaves_enabled)
3909 exec_control &= ~SECONDARY_EXEC_XSAVES;
3913 vmx->nested.msrs.secondary_ctls_high |=
3914 SECONDARY_EXEC_XSAVES;
3916 vmx->nested.msrs.secondary_ctls_high &=
3917 ~SECONDARY_EXEC_XSAVES;
3921 if (vmx_rdtscp_supported()) {
3922 bool rdtscp_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP);
3923 if (!rdtscp_enabled)
3924 exec_control &= ~SECONDARY_EXEC_RDTSCP;
3928 vmx->nested.msrs.secondary_ctls_high |=
3929 SECONDARY_EXEC_RDTSCP;
3931 vmx->nested.msrs.secondary_ctls_high &=
3932 ~SECONDARY_EXEC_RDTSCP;
3936 if (vmx_invpcid_supported()) {
3937 /* Exposing INVPCID only when PCID is exposed */
3938 bool invpcid_enabled =
3939 guest_cpuid_has(vcpu, X86_FEATURE_INVPCID) &&
3940 guest_cpuid_has(vcpu, X86_FEATURE_PCID);
3942 if (!invpcid_enabled) {
3943 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
3944 guest_cpuid_clear(vcpu, X86_FEATURE_INVPCID);
3948 if (invpcid_enabled)
3949 vmx->nested.msrs.secondary_ctls_high |=
3950 SECONDARY_EXEC_ENABLE_INVPCID;
3952 vmx->nested.msrs.secondary_ctls_high &=
3953 ~SECONDARY_EXEC_ENABLE_INVPCID;
3957 if (vmx_rdrand_supported()) {
3958 bool rdrand_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDRAND);
3960 exec_control &= ~SECONDARY_EXEC_RDRAND_EXITING;
3964 vmx->nested.msrs.secondary_ctls_high |=
3965 SECONDARY_EXEC_RDRAND_EXITING;
3967 vmx->nested.msrs.secondary_ctls_high &=
3968 ~SECONDARY_EXEC_RDRAND_EXITING;
3972 if (vmx_rdseed_supported()) {
3973 bool rdseed_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDSEED);
3975 exec_control &= ~SECONDARY_EXEC_RDSEED_EXITING;
3979 vmx->nested.msrs.secondary_ctls_high |=
3980 SECONDARY_EXEC_RDSEED_EXITING;
3982 vmx->nested.msrs.secondary_ctls_high &=
3983 ~SECONDARY_EXEC_RDSEED_EXITING;
3987 vmx->secondary_exec_control = exec_control;
3990 static void ept_set_mmio_spte_mask(void)
3993 * EPT Misconfigurations can be generated if the value of bits 2:0
3994 * of an EPT paging-structure entry is 110b (write/execute).
3996 kvm_mmu_set_mmio_spte_mask(VMX_EPT_RWX_MASK,
3997 VMX_EPT_MISCONFIG_WX_VALUE);
4000 #define VMX_XSS_EXIT_BITMAP 0
4003 * Sets up the vmcs for emulated real mode.
4005 static void vmx_vcpu_setup(struct vcpu_vmx *vmx)
4010 nested_vmx_vcpu_setup();
4012 if (cpu_has_vmx_msr_bitmap())
4013 vmcs_write64(MSR_BITMAP, __pa(vmx->vmcs01.msr_bitmap));
4015 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4018 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
4019 vmx->hv_deadline_tsc = -1;
4021 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
4023 if (cpu_has_secondary_exec_ctrls()) {
4024 vmx_compute_secondary_exec_control(vmx);
4025 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
4026 vmx->secondary_exec_control);
4029 if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
4030 vmcs_write64(EOI_EXIT_BITMAP0, 0);
4031 vmcs_write64(EOI_EXIT_BITMAP1, 0);
4032 vmcs_write64(EOI_EXIT_BITMAP2, 0);
4033 vmcs_write64(EOI_EXIT_BITMAP3, 0);
4035 vmcs_write16(GUEST_INTR_STATUS, 0);
4037 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4038 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4041 if (!kvm_pause_in_guest(vmx->vcpu.kvm)) {
4042 vmcs_write32(PLE_GAP, ple_gap);
4043 vmx->ple_window = ple_window;
4044 vmx->ple_window_dirty = true;
4047 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4048 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4049 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
4051 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
4052 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
4053 vmx_set_constant_host_state(vmx);
4054 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4055 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4057 if (cpu_has_vmx_vmfunc())
4058 vmcs_write64(VM_FUNCTION_CONTROL, 0);
4060 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4061 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4062 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
4063 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4064 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
4066 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
4067 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
4069 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
4070 u32 index = vmx_msr_index[i];
4071 u32 data_low, data_high;
4074 if (rdmsr_safe(index, &data_low, &data_high) < 0)
4076 if (wrmsr_safe(index, data_low, data_high) < 0)
4078 vmx->guest_msrs[j].index = i;
4079 vmx->guest_msrs[j].data = 0;
4080 vmx->guest_msrs[j].mask = -1ull;
4084 vm_exit_controls_init(vmx, vmx_vmexit_ctrl());
4086 /* 22.2.1, 20.8.1 */
4087 vm_entry_controls_init(vmx, vmx_vmentry_ctrl());
4089 vmx->vcpu.arch.cr0_guest_owned_bits = X86_CR0_TS;
4090 vmcs_writel(CR0_GUEST_HOST_MASK, ~X86_CR0_TS);
4092 set_cr4_guest_host_mask(vmx);
4094 if (vmx_xsaves_supported())
4095 vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
4098 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
4099 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
4102 if (cpu_has_vmx_encls_vmexit())
4103 vmcs_write64(ENCLS_EXITING_BITMAP, -1ull);
4105 if (pt_mode == PT_MODE_HOST_GUEST) {
4106 memset(&vmx->pt_desc, 0, sizeof(vmx->pt_desc));
4107 /* Bit[6~0] are forced to 1, writes are ignored. */
4108 vmx->pt_desc.guest.output_mask = 0x7F;
4109 vmcs_write64(GUEST_IA32_RTIT_CTL, 0);
4113 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
4115 struct vcpu_vmx *vmx = to_vmx(vcpu);
4116 struct msr_data apic_base_msr;
4119 vmx->rmode.vm86_active = 0;
4122 vcpu->arch.microcode_version = 0x100000000ULL;
4123 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
4124 kvm_set_cr8(vcpu, 0);
4127 apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
4128 MSR_IA32_APICBASE_ENABLE;
4129 if (kvm_vcpu_is_reset_bsp(vcpu))
4130 apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
4131 apic_base_msr.host_initiated = true;
4132 kvm_set_apic_base(vcpu, &apic_base_msr);
4135 vmx_segment_cache_clear(vmx);
4137 seg_setup(VCPU_SREG_CS);
4138 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4139 vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
4141 seg_setup(VCPU_SREG_DS);
4142 seg_setup(VCPU_SREG_ES);
4143 seg_setup(VCPU_SREG_FS);
4144 seg_setup(VCPU_SREG_GS);
4145 seg_setup(VCPU_SREG_SS);
4147 vmcs_write16(GUEST_TR_SELECTOR, 0);
4148 vmcs_writel(GUEST_TR_BASE, 0);
4149 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4150 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4152 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4153 vmcs_writel(GUEST_LDTR_BASE, 0);
4154 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4155 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4158 vmcs_write32(GUEST_SYSENTER_CS, 0);
4159 vmcs_writel(GUEST_SYSENTER_ESP, 0);
4160 vmcs_writel(GUEST_SYSENTER_EIP, 0);
4161 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4164 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
4165 kvm_rip_write(vcpu, 0xfff0);
4167 vmcs_writel(GUEST_GDTR_BASE, 0);
4168 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4170 vmcs_writel(GUEST_IDTR_BASE, 0);
4171 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4173 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4174 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4175 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4176 if (kvm_mpx_supported())
4177 vmcs_write64(GUEST_BNDCFGS, 0);
4181 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
4183 if (cpu_has_vmx_tpr_shadow() && !init_event) {
4184 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4185 if (cpu_need_tpr_shadow(vcpu))
4186 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4187 __pa(vcpu->arch.apic->regs));
4188 vmcs_write32(TPR_THRESHOLD, 0);
4191 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4194 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4196 cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4197 vmx->vcpu.arch.cr0 = cr0;
4198 vmx_set_cr0(vcpu, cr0); /* enter rmode */
4199 vmx_set_cr4(vcpu, 0);
4200 vmx_set_efer(vcpu, 0);
4202 update_exception_bitmap(vcpu);
4204 vpid_sync_context(vmx->vpid);
4206 vmx_clear_hlt(vcpu);
4209 static void enable_irq_window(struct kvm_vcpu *vcpu)
4211 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
4212 CPU_BASED_VIRTUAL_INTR_PENDING);
4215 static void enable_nmi_window(struct kvm_vcpu *vcpu)
4218 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4219 enable_irq_window(vcpu);
4223 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
4224 CPU_BASED_VIRTUAL_NMI_PENDING);
4227 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4229 struct vcpu_vmx *vmx = to_vmx(vcpu);
4231 int irq = vcpu->arch.interrupt.nr;
4233 trace_kvm_inj_virq(irq);
4235 ++vcpu->stat.irq_injections;
4236 if (vmx->rmode.vm86_active) {
4238 if (vcpu->arch.interrupt.soft)
4239 inc_eip = vcpu->arch.event_exit_inst_len;
4240 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
4241 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4244 intr = irq | INTR_INFO_VALID_MASK;
4245 if (vcpu->arch.interrupt.soft) {
4246 intr |= INTR_TYPE_SOFT_INTR;
4247 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4248 vmx->vcpu.arch.event_exit_inst_len);
4250 intr |= INTR_TYPE_EXT_INTR;
4251 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4253 vmx_clear_hlt(vcpu);
4256 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4258 struct vcpu_vmx *vmx = to_vmx(vcpu);
4262 * Tracking the NMI-blocked state in software is built upon
4263 * finding the next open IRQ window. This, in turn, depends on
4264 * well-behaving guests: They have to keep IRQs disabled at
4265 * least as long as the NMI handler runs. Otherwise we may
4266 * cause NMI nesting, maybe breaking the guest. But as this is
4267 * highly unlikely, we can live with the residual risk.
4269 vmx->loaded_vmcs->soft_vnmi_blocked = 1;
4270 vmx->loaded_vmcs->vnmi_blocked_time = 0;
4273 ++vcpu->stat.nmi_injections;
4274 vmx->loaded_vmcs->nmi_known_unmasked = false;
4276 if (vmx->rmode.vm86_active) {
4277 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
4278 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4282 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4283 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4285 vmx_clear_hlt(vcpu);
4288 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4290 struct vcpu_vmx *vmx = to_vmx(vcpu);
4294 return vmx->loaded_vmcs->soft_vnmi_blocked;
4295 if (vmx->loaded_vmcs->nmi_known_unmasked)
4297 masked = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4298 vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4302 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4304 struct vcpu_vmx *vmx = to_vmx(vcpu);
4307 if (vmx->loaded_vmcs->soft_vnmi_blocked != masked) {
4308 vmx->loaded_vmcs->soft_vnmi_blocked = masked;
4309 vmx->loaded_vmcs->vnmi_blocked_time = 0;
4312 vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4314 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4315 GUEST_INTR_STATE_NMI);
4317 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4318 GUEST_INTR_STATE_NMI);
4322 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
4324 if (to_vmx(vcpu)->nested.nested_run_pending)
4328 to_vmx(vcpu)->loaded_vmcs->soft_vnmi_blocked)
4331 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4332 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
4333 | GUEST_INTR_STATE_NMI));
4336 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
4338 return (!to_vmx(vcpu)->nested.nested_run_pending &&
4339 vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
4340 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4341 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4344 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4348 if (enable_unrestricted_guest)
4351 ret = x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
4355 to_kvm_vmx(kvm)->tss_addr = addr;
4356 return init_rmode_tss(kvm);
4359 static int vmx_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
4361 to_kvm_vmx(kvm)->ept_identity_map_addr = ident_addr;
4365 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
4370 * Update instruction length as we may reinject the exception
4371 * from user space while in guest debugging mode.
4373 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4374 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4375 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4379 if (vcpu->guest_debug &
4380 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
4397 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4398 int vec, u32 err_code)
4401 * Instruction with address size override prefix opcode 0x67
4402 * Cause the #SS fault with 0 error code in VM86 mode.
4404 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
4405 if (kvm_emulate_instruction(vcpu, 0) == EMULATE_DONE) {
4406 if (vcpu->arch.halt_request) {
4407 vcpu->arch.halt_request = 0;
4408 return kvm_vcpu_halt(vcpu);
4416 * Forward all other exceptions that are valid in real mode.
4417 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4418 * the required debugging infrastructure rework.
4420 kvm_queue_exception(vcpu, vec);
4425 * Trigger machine check on the host. We assume all the MSRs are already set up
4426 * by the CPU and that we still run on the same CPU as the MCE occurred on.
4427 * We pass a fake environment to the machine check handler because we want
4428 * the guest to be always treated like user space, no matter what context
4429 * it used internally.
4431 static void kvm_machine_check(void)
4433 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
4434 struct pt_regs regs = {
4435 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
4436 .flags = X86_EFLAGS_IF,
4439 do_machine_check(®s, 0);
4443 static int handle_machine_check(struct kvm_vcpu *vcpu)
4445 /* already handled by vcpu_run */
4449 static int handle_exception(struct kvm_vcpu *vcpu)
4451 struct vcpu_vmx *vmx = to_vmx(vcpu);
4452 struct kvm_run *kvm_run = vcpu->run;
4453 u32 intr_info, ex_no, error_code;
4454 unsigned long cr2, rip, dr6;
4456 enum emulation_result er;
4458 vect_info = vmx->idt_vectoring_info;
4459 intr_info = vmx->exit_intr_info;
4461 if (is_machine_check(intr_info))
4462 return handle_machine_check(vcpu);
4464 if (is_nmi(intr_info))
4465 return 1; /* already handled by vmx_vcpu_run() */
4467 if (is_invalid_opcode(intr_info))
4468 return handle_ud(vcpu);
4471 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4472 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4474 if (!vmx->rmode.vm86_active && is_gp_fault(intr_info)) {
4475 WARN_ON_ONCE(!enable_vmware_backdoor);
4476 er = kvm_emulate_instruction(vcpu,
4477 EMULTYPE_VMWARE | EMULTYPE_NO_UD_ON_FAIL);
4478 if (er == EMULATE_USER_EXIT)
4480 else if (er != EMULATE_DONE)
4481 kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
4486 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
4487 * MMIO, it is better to report an internal error.
4488 * See the comments in vmx_handle_exit.
4490 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4491 !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
4492 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4493 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4494 vcpu->run->internal.ndata = 3;
4495 vcpu->run->internal.data[0] = vect_info;
4496 vcpu->run->internal.data[1] = intr_info;
4497 vcpu->run->internal.data[2] = error_code;
4501 if (is_page_fault(intr_info)) {
4502 cr2 = vmcs_readl(EXIT_QUALIFICATION);
4503 /* EPT won't cause page fault directly */
4504 WARN_ON_ONCE(!vcpu->arch.apf.host_apf_reason && enable_ept);
4505 return kvm_handle_page_fault(vcpu, error_code, cr2, NULL, 0);
4508 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4510 if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
4511 return handle_rmode_exception(vcpu, ex_no, error_code);
4515 kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
4518 dr6 = vmcs_readl(EXIT_QUALIFICATION);
4519 if (!(vcpu->guest_debug &
4520 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4521 vcpu->arch.dr6 &= ~15;
4522 vcpu->arch.dr6 |= dr6 | DR6_RTM;
4523 if (is_icebp(intr_info))
4524 skip_emulated_instruction(vcpu);
4526 kvm_queue_exception(vcpu, DB_VECTOR);
4529 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
4530 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4534 * Update instruction length as we may reinject #BP from
4535 * user space while in guest debugging mode. Reading it for
4536 * #DB as well causes no harm, it is not used in that case.
4538 vmx->vcpu.arch.event_exit_inst_len =
4539 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4540 kvm_run->exit_reason = KVM_EXIT_DEBUG;
4541 rip = kvm_rip_read(vcpu);
4542 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4543 kvm_run->debug.arch.exception = ex_no;
4546 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4547 kvm_run->ex.exception = ex_no;
4548 kvm_run->ex.error_code = error_code;
4554 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
4556 ++vcpu->stat.irq_exits;
4560 static int handle_triple_fault(struct kvm_vcpu *vcpu)
4562 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4563 vcpu->mmio_needed = 0;
4567 static int handle_io(struct kvm_vcpu *vcpu)
4569 unsigned long exit_qualification;
4570 int size, in, string;
4573 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4574 string = (exit_qualification & 16) != 0;
4576 ++vcpu->stat.io_exits;
4579 return kvm_emulate_instruction(vcpu, 0) == EMULATE_DONE;
4581 port = exit_qualification >> 16;
4582 size = (exit_qualification & 7) + 1;
4583 in = (exit_qualification & 8) != 0;
4585 return kvm_fast_pio(vcpu, size, port, in);
4589 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4592 * Patch in the VMCALL instruction:
4594 hypercall[0] = 0x0f;
4595 hypercall[1] = 0x01;
4596 hypercall[2] = 0xc1;
4599 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
4600 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
4602 if (is_guest_mode(vcpu)) {
4603 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4604 unsigned long orig_val = val;
4607 * We get here when L2 changed cr0 in a way that did not change
4608 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
4609 * but did change L0 shadowed bits. So we first calculate the
4610 * effective cr0 value that L1 would like to write into the
4611 * hardware. It consists of the L2-owned bits from the new
4612 * value combined with the L1-owned bits from L1's guest_cr0.
4614 val = (val & ~vmcs12->cr0_guest_host_mask) |
4615 (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
4617 if (!nested_guest_cr0_valid(vcpu, val))
4620 if (kvm_set_cr0(vcpu, val))
4622 vmcs_writel(CR0_READ_SHADOW, orig_val);
4625 if (to_vmx(vcpu)->nested.vmxon &&
4626 !nested_host_cr0_valid(vcpu, val))
4629 return kvm_set_cr0(vcpu, val);
4633 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
4635 if (is_guest_mode(vcpu)) {
4636 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4637 unsigned long orig_val = val;
4639 /* analogously to handle_set_cr0 */
4640 val = (val & ~vmcs12->cr4_guest_host_mask) |
4641 (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
4642 if (kvm_set_cr4(vcpu, val))
4644 vmcs_writel(CR4_READ_SHADOW, orig_val);
4647 return kvm_set_cr4(vcpu, val);
4650 static int handle_desc(struct kvm_vcpu *vcpu)
4652 WARN_ON(!(vcpu->arch.cr4 & X86_CR4_UMIP));
4653 return kvm_emulate_instruction(vcpu, 0) == EMULATE_DONE;
4656 static int handle_cr(struct kvm_vcpu *vcpu)
4658 unsigned long exit_qualification, val;
4664 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4665 cr = exit_qualification & 15;
4666 reg = (exit_qualification >> 8) & 15;
4667 switch ((exit_qualification >> 4) & 3) {
4668 case 0: /* mov to cr */
4669 val = kvm_register_readl(vcpu, reg);
4670 trace_kvm_cr_write(cr, val);
4673 err = handle_set_cr0(vcpu, val);
4674 return kvm_complete_insn_gp(vcpu, err);
4676 WARN_ON_ONCE(enable_unrestricted_guest);
4677 err = kvm_set_cr3(vcpu, val);
4678 return kvm_complete_insn_gp(vcpu, err);
4680 err = handle_set_cr4(vcpu, val);
4681 return kvm_complete_insn_gp(vcpu, err);
4683 u8 cr8_prev = kvm_get_cr8(vcpu);
4685 err = kvm_set_cr8(vcpu, cr8);
4686 ret = kvm_complete_insn_gp(vcpu, err);
4687 if (lapic_in_kernel(vcpu))
4689 if (cr8_prev <= cr8)
4692 * TODO: we might be squashing a
4693 * KVM_GUESTDBG_SINGLESTEP-triggered
4694 * KVM_EXIT_DEBUG here.
4696 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
4702 WARN_ONCE(1, "Guest should always own CR0.TS");
4703 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4704 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
4705 return kvm_skip_emulated_instruction(vcpu);
4706 case 1: /*mov from cr*/
4709 WARN_ON_ONCE(enable_unrestricted_guest);
4710 val = kvm_read_cr3(vcpu);
4711 kvm_register_write(vcpu, reg, val);
4712 trace_kvm_cr_read(cr, val);
4713 return kvm_skip_emulated_instruction(vcpu);
4715 val = kvm_get_cr8(vcpu);
4716 kvm_register_write(vcpu, reg, val);
4717 trace_kvm_cr_read(cr, val);
4718 return kvm_skip_emulated_instruction(vcpu);
4722 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
4723 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
4724 kvm_lmsw(vcpu, val);
4726 return kvm_skip_emulated_instruction(vcpu);
4730 vcpu->run->exit_reason = 0;
4731 vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
4732 (int)(exit_qualification >> 4) & 3, cr);
4736 static int handle_dr(struct kvm_vcpu *vcpu)
4738 unsigned long exit_qualification;
4741 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4742 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
4744 /* First, if DR does not exist, trigger UD */
4745 if (!kvm_require_dr(vcpu, dr))
4748 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
4749 if (!kvm_require_cpl(vcpu, 0))
4751 dr7 = vmcs_readl(GUEST_DR7);
4754 * As the vm-exit takes precedence over the debug trap, we
4755 * need to emulate the latter, either for the host or the
4756 * guest debugging itself.
4758 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
4759 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
4760 vcpu->run->debug.arch.dr7 = dr7;
4761 vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
4762 vcpu->run->debug.arch.exception = DB_VECTOR;
4763 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
4766 vcpu->arch.dr6 &= ~15;
4767 vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
4768 kvm_queue_exception(vcpu, DB_VECTOR);
4773 if (vcpu->guest_debug == 0) {
4774 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
4775 CPU_BASED_MOV_DR_EXITING);
4778 * No more DR vmexits; force a reload of the debug registers
4779 * and reenter on this instruction. The next vmexit will
4780 * retrieve the full state of the debug registers.
4782 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
4786 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
4787 if (exit_qualification & TYPE_MOV_FROM_DR) {
4790 if (kvm_get_dr(vcpu, dr, &val))
4792 kvm_register_write(vcpu, reg, val);
4794 if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
4797 return kvm_skip_emulated_instruction(vcpu);
4800 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
4802 return vcpu->arch.dr6;
4805 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
4809 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
4811 get_debugreg(vcpu->arch.db[0], 0);
4812 get_debugreg(vcpu->arch.db[1], 1);
4813 get_debugreg(vcpu->arch.db[2], 2);
4814 get_debugreg(vcpu->arch.db[3], 3);
4815 get_debugreg(vcpu->arch.dr6, 6);
4816 vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
4818 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
4819 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL, CPU_BASED_MOV_DR_EXITING);
4822 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
4824 vmcs_writel(GUEST_DR7, val);
4827 static int handle_cpuid(struct kvm_vcpu *vcpu)
4829 return kvm_emulate_cpuid(vcpu);
4832 static int handle_rdmsr(struct kvm_vcpu *vcpu)
4834 u32 ecx = kvm_rcx_read(vcpu);
4835 struct msr_data msr_info;
4837 msr_info.index = ecx;
4838 msr_info.host_initiated = false;
4839 if (vmx_get_msr(vcpu, &msr_info)) {
4840 trace_kvm_msr_read_ex(ecx);
4841 kvm_inject_gp(vcpu, 0);
4845 trace_kvm_msr_read(ecx, msr_info.data);
4847 kvm_rax_write(vcpu, msr_info.data & -1u);
4848 kvm_rdx_write(vcpu, (msr_info.data >> 32) & -1u);
4849 return kvm_skip_emulated_instruction(vcpu);
4852 static int handle_wrmsr(struct kvm_vcpu *vcpu)
4854 struct msr_data msr;
4855 u32 ecx = kvm_rcx_read(vcpu);
4856 u64 data = kvm_read_edx_eax(vcpu);
4860 msr.host_initiated = false;
4861 if (kvm_set_msr(vcpu, &msr) != 0) {
4862 trace_kvm_msr_write_ex(ecx, data);
4863 kvm_inject_gp(vcpu, 0);
4867 trace_kvm_msr_write(ecx, data);
4868 return kvm_skip_emulated_instruction(vcpu);
4871 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
4873 kvm_apic_update_ppr(vcpu);
4877 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
4879 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
4880 CPU_BASED_VIRTUAL_INTR_PENDING);
4882 kvm_make_request(KVM_REQ_EVENT, vcpu);
4884 ++vcpu->stat.irq_window_exits;
4888 static int handle_halt(struct kvm_vcpu *vcpu)
4890 return kvm_emulate_halt(vcpu);
4893 static int handle_vmcall(struct kvm_vcpu *vcpu)
4895 return kvm_emulate_hypercall(vcpu);
4898 static int handle_invd(struct kvm_vcpu *vcpu)
4900 return kvm_emulate_instruction(vcpu, 0) == EMULATE_DONE;
4903 static int handle_invlpg(struct kvm_vcpu *vcpu)
4905 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4907 kvm_mmu_invlpg(vcpu, exit_qualification);
4908 return kvm_skip_emulated_instruction(vcpu);
4911 static int handle_rdpmc(struct kvm_vcpu *vcpu)
4915 err = kvm_rdpmc(vcpu);
4916 return kvm_complete_insn_gp(vcpu, err);
4919 static int handle_wbinvd(struct kvm_vcpu *vcpu)
4921 return kvm_emulate_wbinvd(vcpu);
4924 static int handle_xsetbv(struct kvm_vcpu *vcpu)
4926 u64 new_bv = kvm_read_edx_eax(vcpu);
4927 u32 index = kvm_rcx_read(vcpu);
4929 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
4930 return kvm_skip_emulated_instruction(vcpu);
4934 static int handle_xsaves(struct kvm_vcpu *vcpu)
4936 kvm_skip_emulated_instruction(vcpu);
4937 WARN(1, "this should never happen\n");
4941 static int handle_xrstors(struct kvm_vcpu *vcpu)
4943 kvm_skip_emulated_instruction(vcpu);
4944 WARN(1, "this should never happen\n");
4948 static int handle_apic_access(struct kvm_vcpu *vcpu)
4950 if (likely(fasteoi)) {
4951 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4952 int access_type, offset;
4954 access_type = exit_qualification & APIC_ACCESS_TYPE;
4955 offset = exit_qualification & APIC_ACCESS_OFFSET;
4957 * Sane guest uses MOV to write EOI, with written value
4958 * not cared. So make a short-circuit here by avoiding
4959 * heavy instruction emulation.
4961 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
4962 (offset == APIC_EOI)) {
4963 kvm_lapic_set_eoi(vcpu);
4964 return kvm_skip_emulated_instruction(vcpu);
4967 return kvm_emulate_instruction(vcpu, 0) == EMULATE_DONE;
4970 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
4972 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4973 int vector = exit_qualification & 0xff;
4975 /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
4976 kvm_apic_set_eoi_accelerated(vcpu, vector);
4980 static int handle_apic_write(struct kvm_vcpu *vcpu)
4982 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4983 u32 offset = exit_qualification & 0xfff;
4985 /* APIC-write VM exit is trap-like and thus no need to adjust IP */
4986 kvm_apic_write_nodecode(vcpu, offset);
4990 static int handle_task_switch(struct kvm_vcpu *vcpu)
4992 struct vcpu_vmx *vmx = to_vmx(vcpu);
4993 unsigned long exit_qualification;
4994 bool has_error_code = false;
4997 int reason, type, idt_v, idt_index;
4999 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5000 idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5001 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5003 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5005 reason = (u32)exit_qualification >> 30;
5006 if (reason == TASK_SWITCH_GATE && idt_v) {
5008 case INTR_TYPE_NMI_INTR:
5009 vcpu->arch.nmi_injected = false;
5010 vmx_set_nmi_mask(vcpu, true);
5012 case INTR_TYPE_EXT_INTR:
5013 case INTR_TYPE_SOFT_INTR:
5014 kvm_clear_interrupt_queue(vcpu);
5016 case INTR_TYPE_HARD_EXCEPTION:
5017 if (vmx->idt_vectoring_info &
5018 VECTORING_INFO_DELIVER_CODE_MASK) {
5019 has_error_code = true;
5021 vmcs_read32(IDT_VECTORING_ERROR_CODE);
5024 case INTR_TYPE_SOFT_EXCEPTION:
5025 kvm_clear_exception_queue(vcpu);
5031 tss_selector = exit_qualification;
5033 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5034 type != INTR_TYPE_EXT_INTR &&
5035 type != INTR_TYPE_NMI_INTR))
5036 skip_emulated_instruction(vcpu);
5038 if (kvm_task_switch(vcpu, tss_selector,
5039 type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
5040 has_error_code, error_code) == EMULATE_FAIL) {
5041 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5042 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5043 vcpu->run->internal.ndata = 0;
5048 * TODO: What about debug traps on tss switch?
5049 * Are we supposed to inject them and update dr6?
5055 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5057 unsigned long exit_qualification;
5061 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5064 * EPT violation happened while executing iret from NMI,
5065 * "blocked by NMI" bit has to be set before next VM entry.
5066 * There are errata that may cause this bit to not be set:
5069 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5071 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
5072 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5074 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5075 trace_kvm_page_fault(gpa, exit_qualification);
5077 /* Is it a read fault? */
5078 error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
5079 ? PFERR_USER_MASK : 0;
5080 /* Is it a write fault? */
5081 error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
5082 ? PFERR_WRITE_MASK : 0;
5083 /* Is it a fetch fault? */
5084 error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
5085 ? PFERR_FETCH_MASK : 0;
5086 /* ept page table entry is present? */
5087 error_code |= (exit_qualification &
5088 (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE |
5089 EPT_VIOLATION_EXECUTABLE))
5090 ? PFERR_PRESENT_MASK : 0;
5092 error_code |= (exit_qualification & 0x100) != 0 ?
5093 PFERR_GUEST_FINAL_MASK : PFERR_GUEST_PAGE_MASK;
5095 vcpu->arch.exit_qualification = exit_qualification;
5096 return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5099 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5104 * A nested guest cannot optimize MMIO vmexits, because we have an
5105 * nGPA here instead of the required GPA.
5107 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5108 if (!is_guest_mode(vcpu) &&
5109 !kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5110 trace_kvm_fast_mmio(gpa);
5112 * Doing kvm_skip_emulated_instruction() depends on undefined
5113 * behavior: Intel's manual doesn't mandate
5114 * VM_EXIT_INSTRUCTION_LEN to be set in VMCS when EPT MISCONFIG
5115 * occurs and while on real hardware it was observed to be set,
5116 * other hypervisors (namely Hyper-V) don't set it, we end up
5117 * advancing IP with some random value. Disable fast mmio when
5118 * running nested and keep it for real hardware in hope that
5119 * VM_EXIT_INSTRUCTION_LEN will always be set correctly.
5121 if (!static_cpu_has(X86_FEATURE_HYPERVISOR))
5122 return kvm_skip_emulated_instruction(vcpu);
5124 return kvm_emulate_instruction(vcpu, EMULTYPE_SKIP) ==
5128 return kvm_mmu_page_fault(vcpu, gpa, PFERR_RSVD_MASK, NULL, 0);
5131 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5133 WARN_ON_ONCE(!enable_vnmi);
5134 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
5135 CPU_BASED_VIRTUAL_NMI_PENDING);
5136 ++vcpu->stat.nmi_window_exits;
5137 kvm_make_request(KVM_REQ_EVENT, vcpu);
5142 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5144 struct vcpu_vmx *vmx = to_vmx(vcpu);
5145 enum emulation_result err = EMULATE_DONE;
5148 bool intr_window_requested;
5149 unsigned count = 130;
5152 * We should never reach the point where we are emulating L2
5153 * due to invalid guest state as that means we incorrectly
5154 * allowed a nested VMEntry with an invalid vmcs12.
5156 WARN_ON_ONCE(vmx->emulation_required && vmx->nested.nested_run_pending);
5158 cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5159 intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
5161 while (vmx->emulation_required && count-- != 0) {
5162 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
5163 return handle_interrupt_window(&vmx->vcpu);
5165 if (kvm_test_request(KVM_REQ_EVENT, vcpu))
5168 err = kvm_emulate_instruction(vcpu, 0);
5170 if (err == EMULATE_USER_EXIT) {
5171 ++vcpu->stat.mmio_exits;
5176 if (err != EMULATE_DONE)
5177 goto emulation_error;
5179 if (vmx->emulation_required && !vmx->rmode.vm86_active &&
5180 vcpu->arch.exception.pending)
5181 goto emulation_error;
5183 if (vcpu->arch.halt_request) {
5184 vcpu->arch.halt_request = 0;
5185 ret = kvm_vcpu_halt(vcpu);
5189 if (signal_pending(current))
5199 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5200 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5201 vcpu->run->internal.ndata = 0;
5205 static void grow_ple_window(struct kvm_vcpu *vcpu)
5207 struct vcpu_vmx *vmx = to_vmx(vcpu);
5208 int old = vmx->ple_window;
5210 vmx->ple_window = __grow_ple_window(old, ple_window,
5214 if (vmx->ple_window != old)
5215 vmx->ple_window_dirty = true;
5217 trace_kvm_ple_window_grow(vcpu->vcpu_id, vmx->ple_window, old);
5220 static void shrink_ple_window(struct kvm_vcpu *vcpu)
5222 struct vcpu_vmx *vmx = to_vmx(vcpu);
5223 int old = vmx->ple_window;
5225 vmx->ple_window = __shrink_ple_window(old, ple_window,
5229 if (vmx->ple_window != old)
5230 vmx->ple_window_dirty = true;
5232 trace_kvm_ple_window_shrink(vcpu->vcpu_id, vmx->ple_window, old);
5236 * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
5238 static void wakeup_handler(void)
5240 struct kvm_vcpu *vcpu;
5241 int cpu = smp_processor_id();
5243 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5244 list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
5245 blocked_vcpu_list) {
5246 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
5248 if (pi_test_on(pi_desc) == 1)
5249 kvm_vcpu_kick(vcpu);
5251 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5254 static void vmx_enable_tdp(void)
5256 kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK,
5257 enable_ept_ad_bits ? VMX_EPT_ACCESS_BIT : 0ull,
5258 enable_ept_ad_bits ? VMX_EPT_DIRTY_BIT : 0ull,
5259 0ull, VMX_EPT_EXECUTABLE_MASK,
5260 cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK,
5261 VMX_EPT_RWX_MASK, 0ull);
5263 ept_set_mmio_spte_mask();
5268 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5269 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5271 static int handle_pause(struct kvm_vcpu *vcpu)
5273 if (!kvm_pause_in_guest(vcpu->kvm))
5274 grow_ple_window(vcpu);
5277 * Intel sdm vol3 ch-25.1.3 says: The "PAUSE-loop exiting"
5278 * VM-execution control is ignored if CPL > 0. OTOH, KVM
5279 * never set PAUSE_EXITING and just set PLE if supported,
5280 * so the vcpu must be CPL=0 if it gets a PAUSE exit.
5282 kvm_vcpu_on_spin(vcpu, true);
5283 return kvm_skip_emulated_instruction(vcpu);
5286 static int handle_nop(struct kvm_vcpu *vcpu)
5288 return kvm_skip_emulated_instruction(vcpu);
5291 static int handle_mwait(struct kvm_vcpu *vcpu)
5293 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
5294 return handle_nop(vcpu);
5297 static int handle_invalid_op(struct kvm_vcpu *vcpu)
5299 kvm_queue_exception(vcpu, UD_VECTOR);
5303 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
5308 static int handle_monitor(struct kvm_vcpu *vcpu)
5310 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
5311 return handle_nop(vcpu);
5314 static int handle_invpcid(struct kvm_vcpu *vcpu)
5316 u32 vmx_instruction_info;
5320 struct x86_exception e;
5322 unsigned long roots_to_free = 0;
5328 if (!guest_cpuid_has(vcpu, X86_FEATURE_INVPCID)) {
5329 kvm_queue_exception(vcpu, UD_VECTOR);
5333 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5334 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
5337 kvm_inject_gp(vcpu, 0);
5341 /* According to the Intel instruction reference, the memory operand
5342 * is read even if it isn't needed (e.g., for type==all)
5344 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5345 vmx_instruction_info, false, &gva))
5348 if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
5349 kvm_inject_page_fault(vcpu, &e);
5353 if (operand.pcid >> 12 != 0) {
5354 kvm_inject_gp(vcpu, 0);
5358 pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
5361 case INVPCID_TYPE_INDIV_ADDR:
5362 if ((!pcid_enabled && (operand.pcid != 0)) ||
5363 is_noncanonical_address(operand.gla, vcpu)) {
5364 kvm_inject_gp(vcpu, 0);
5367 kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
5368 return kvm_skip_emulated_instruction(vcpu);
5370 case INVPCID_TYPE_SINGLE_CTXT:
5371 if (!pcid_enabled && (operand.pcid != 0)) {
5372 kvm_inject_gp(vcpu, 0);
5376 if (kvm_get_active_pcid(vcpu) == operand.pcid) {
5377 kvm_mmu_sync_roots(vcpu);
5378 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
5381 for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
5382 if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].cr3)
5384 roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
5386 kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
5388 * If neither the current cr3 nor any of the prev_roots use the
5389 * given PCID, then nothing needs to be done here because a
5390 * resync will happen anyway before switching to any other CR3.
5393 return kvm_skip_emulated_instruction(vcpu);
5395 case INVPCID_TYPE_ALL_NON_GLOBAL:
5397 * Currently, KVM doesn't mark global entries in the shadow
5398 * page tables, so a non-global flush just degenerates to a
5399 * global flush. If needed, we could optimize this later by
5400 * keeping track of global entries in shadow page tables.
5404 case INVPCID_TYPE_ALL_INCL_GLOBAL:
5405 kvm_mmu_unload(vcpu);
5406 return kvm_skip_emulated_instruction(vcpu);
5409 BUG(); /* We have already checked above that type <= 3 */
5413 static int handle_pml_full(struct kvm_vcpu *vcpu)
5415 unsigned long exit_qualification;
5417 trace_kvm_pml_full(vcpu->vcpu_id);
5419 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5422 * PML buffer FULL happened while executing iret from NMI,
5423 * "blocked by NMI" bit has to be set before next VM entry.
5425 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5427 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
5428 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5429 GUEST_INTR_STATE_NMI);
5432 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
5433 * here.., and there's no userspace involvement needed for PML.
5438 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
5440 if (!to_vmx(vcpu)->req_immediate_exit)
5441 kvm_lapic_expired_hv_timer(vcpu);
5446 * When nested=0, all VMX instruction VM Exits filter here. The handlers
5447 * are overwritten by nested_vmx_setup() when nested=1.
5449 static int handle_vmx_instruction(struct kvm_vcpu *vcpu)
5451 kvm_queue_exception(vcpu, UD_VECTOR);
5455 static int handle_encls(struct kvm_vcpu *vcpu)
5458 * SGX virtualization is not yet supported. There is no software
5459 * enable bit for SGX, so we have to trap ENCLS and inject a #UD
5460 * to prevent the guest from executing ENCLS.
5462 kvm_queue_exception(vcpu, UD_VECTOR);
5467 * The exit handlers return 1 if the exit was handled fully and guest execution
5468 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
5469 * to be done to userspace and return 0.
5471 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
5472 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
5473 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
5474 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
5475 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
5476 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
5477 [EXIT_REASON_CR_ACCESS] = handle_cr,
5478 [EXIT_REASON_DR_ACCESS] = handle_dr,
5479 [EXIT_REASON_CPUID] = handle_cpuid,
5480 [EXIT_REASON_MSR_READ] = handle_rdmsr,
5481 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
5482 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
5483 [EXIT_REASON_HLT] = handle_halt,
5484 [EXIT_REASON_INVD] = handle_invd,
5485 [EXIT_REASON_INVLPG] = handle_invlpg,
5486 [EXIT_REASON_RDPMC] = handle_rdpmc,
5487 [EXIT_REASON_VMCALL] = handle_vmcall,
5488 [EXIT_REASON_VMCLEAR] = handle_vmx_instruction,
5489 [EXIT_REASON_VMLAUNCH] = handle_vmx_instruction,
5490 [EXIT_REASON_VMPTRLD] = handle_vmx_instruction,
5491 [EXIT_REASON_VMPTRST] = handle_vmx_instruction,
5492 [EXIT_REASON_VMREAD] = handle_vmx_instruction,
5493 [EXIT_REASON_VMRESUME] = handle_vmx_instruction,
5494 [EXIT_REASON_VMWRITE] = handle_vmx_instruction,
5495 [EXIT_REASON_VMOFF] = handle_vmx_instruction,
5496 [EXIT_REASON_VMON] = handle_vmx_instruction,
5497 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
5498 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
5499 [EXIT_REASON_APIC_WRITE] = handle_apic_write,
5500 [EXIT_REASON_EOI_INDUCED] = handle_apic_eoi_induced,
5501 [EXIT_REASON_WBINVD] = handle_wbinvd,
5502 [EXIT_REASON_XSETBV] = handle_xsetbv,
5503 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
5504 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
5505 [EXIT_REASON_GDTR_IDTR] = handle_desc,
5506 [EXIT_REASON_LDTR_TR] = handle_desc,
5507 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
5508 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
5509 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
5510 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_mwait,
5511 [EXIT_REASON_MONITOR_TRAP_FLAG] = handle_monitor_trap,
5512 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_monitor,
5513 [EXIT_REASON_INVEPT] = handle_vmx_instruction,
5514 [EXIT_REASON_INVVPID] = handle_vmx_instruction,
5515 [EXIT_REASON_RDRAND] = handle_invalid_op,
5516 [EXIT_REASON_RDSEED] = handle_invalid_op,
5517 [EXIT_REASON_XSAVES] = handle_xsaves,
5518 [EXIT_REASON_XRSTORS] = handle_xrstors,
5519 [EXIT_REASON_PML_FULL] = handle_pml_full,
5520 [EXIT_REASON_INVPCID] = handle_invpcid,
5521 [EXIT_REASON_VMFUNC] = handle_vmx_instruction,
5522 [EXIT_REASON_PREEMPTION_TIMER] = handle_preemption_timer,
5523 [EXIT_REASON_ENCLS] = handle_encls,
5526 static const int kvm_vmx_max_exit_handlers =
5527 ARRAY_SIZE(kvm_vmx_exit_handlers);
5529 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
5531 *info1 = vmcs_readl(EXIT_QUALIFICATION);
5532 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
5535 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
5538 __free_page(vmx->pml_pg);
5543 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
5545 struct vcpu_vmx *vmx = to_vmx(vcpu);
5549 pml_idx = vmcs_read16(GUEST_PML_INDEX);
5551 /* Do nothing if PML buffer is empty */
5552 if (pml_idx == (PML_ENTITY_NUM - 1))
5555 /* PML index always points to next available PML buffer entity */
5556 if (pml_idx >= PML_ENTITY_NUM)
5561 pml_buf = page_address(vmx->pml_pg);
5562 for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
5565 gpa = pml_buf[pml_idx];
5566 WARN_ON(gpa & (PAGE_SIZE - 1));
5567 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5570 /* reset PML index */
5571 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
5575 * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
5576 * Called before reporting dirty_bitmap to userspace.
5578 static void kvm_flush_pml_buffers(struct kvm *kvm)
5581 struct kvm_vcpu *vcpu;
5583 * We only need to kick vcpu out of guest mode here, as PML buffer
5584 * is flushed at beginning of all VMEXITs, and it's obvious that only
5585 * vcpus running in guest are possible to have unflushed GPAs in PML
5588 kvm_for_each_vcpu(i, vcpu, kvm)
5589 kvm_vcpu_kick(vcpu);
5592 static void vmx_dump_sel(char *name, uint32_t sel)
5594 pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
5595 name, vmcs_read16(sel),
5596 vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
5597 vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
5598 vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
5601 static void vmx_dump_dtsel(char *name, uint32_t limit)
5603 pr_err("%s limit=0x%08x, base=0x%016lx\n",
5604 name, vmcs_read32(limit),
5605 vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
5608 void dump_vmcs(void)
5610 u32 vmentry_ctl, vmexit_ctl;
5611 u32 cpu_based_exec_ctrl, pin_based_exec_ctrl, secondary_exec_control;
5616 if (!dump_invalid_vmcs) {
5617 pr_warn_ratelimited("set kvm_intel.dump_invalid_vmcs=1 to dump internal KVM state.\n");
5621 vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
5622 vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
5623 cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5624 pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
5625 cr4 = vmcs_readl(GUEST_CR4);
5626 efer = vmcs_read64(GUEST_IA32_EFER);
5627 secondary_exec_control = 0;
5628 if (cpu_has_secondary_exec_ctrls())
5629 secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
5631 pr_err("*** Guest State ***\n");
5632 pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5633 vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
5634 vmcs_readl(CR0_GUEST_HOST_MASK));
5635 pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5636 cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
5637 pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
5638 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
5639 (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
5641 pr_err("PDPTR0 = 0x%016llx PDPTR1 = 0x%016llx\n",
5642 vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
5643 pr_err("PDPTR2 = 0x%016llx PDPTR3 = 0x%016llx\n",
5644 vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
5646 pr_err("RSP = 0x%016lx RIP = 0x%016lx\n",
5647 vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
5648 pr_err("RFLAGS=0x%08lx DR7 = 0x%016lx\n",
5649 vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
5650 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5651 vmcs_readl(GUEST_SYSENTER_ESP),
5652 vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
5653 vmx_dump_sel("CS: ", GUEST_CS_SELECTOR);
5654 vmx_dump_sel("DS: ", GUEST_DS_SELECTOR);
5655 vmx_dump_sel("SS: ", GUEST_SS_SELECTOR);
5656 vmx_dump_sel("ES: ", GUEST_ES_SELECTOR);
5657 vmx_dump_sel("FS: ", GUEST_FS_SELECTOR);
5658 vmx_dump_sel("GS: ", GUEST_GS_SELECTOR);
5659 vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
5660 vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
5661 vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
5662 vmx_dump_sel("TR: ", GUEST_TR_SELECTOR);
5663 if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
5664 (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
5665 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
5666 efer, vmcs_read64(GUEST_IA32_PAT));
5667 pr_err("DebugCtl = 0x%016llx DebugExceptions = 0x%016lx\n",
5668 vmcs_read64(GUEST_IA32_DEBUGCTL),
5669 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
5670 if (cpu_has_load_perf_global_ctrl() &&
5671 vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
5672 pr_err("PerfGlobCtl = 0x%016llx\n",
5673 vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
5674 if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
5675 pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
5676 pr_err("Interruptibility = %08x ActivityState = %08x\n",
5677 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
5678 vmcs_read32(GUEST_ACTIVITY_STATE));
5679 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
5680 pr_err("InterruptStatus = %04x\n",
5681 vmcs_read16(GUEST_INTR_STATUS));
5683 pr_err("*** Host State ***\n");
5684 pr_err("RIP = 0x%016lx RSP = 0x%016lx\n",
5685 vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
5686 pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
5687 vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
5688 vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
5689 vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
5690 vmcs_read16(HOST_TR_SELECTOR));
5691 pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
5692 vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
5693 vmcs_readl(HOST_TR_BASE));
5694 pr_err("GDTBase=%016lx IDTBase=%016lx\n",
5695 vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
5696 pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
5697 vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
5698 vmcs_readl(HOST_CR4));
5699 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5700 vmcs_readl(HOST_IA32_SYSENTER_ESP),
5701 vmcs_read32(HOST_IA32_SYSENTER_CS),
5702 vmcs_readl(HOST_IA32_SYSENTER_EIP));
5703 if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
5704 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
5705 vmcs_read64(HOST_IA32_EFER),
5706 vmcs_read64(HOST_IA32_PAT));
5707 if (cpu_has_load_perf_global_ctrl() &&
5708 vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
5709 pr_err("PerfGlobCtl = 0x%016llx\n",
5710 vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
5712 pr_err("*** Control State ***\n");
5713 pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
5714 pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
5715 pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
5716 pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
5717 vmcs_read32(EXCEPTION_BITMAP),
5718 vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
5719 vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
5720 pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
5721 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
5722 vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
5723 vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
5724 pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
5725 vmcs_read32(VM_EXIT_INTR_INFO),
5726 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
5727 vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
5728 pr_err(" reason=%08x qualification=%016lx\n",
5729 vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
5730 pr_err("IDTVectoring: info=%08x errcode=%08x\n",
5731 vmcs_read32(IDT_VECTORING_INFO_FIELD),
5732 vmcs_read32(IDT_VECTORING_ERROR_CODE));
5733 pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
5734 if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
5735 pr_err("TSC Multiplier = 0x%016llx\n",
5736 vmcs_read64(TSC_MULTIPLIER));
5737 if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW) {
5738 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
5739 u16 status = vmcs_read16(GUEST_INTR_STATUS);
5740 pr_err("SVI|RVI = %02x|%02x ", status >> 8, status & 0xff);
5742 pr_cont("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
5743 if (secondary_exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
5744 pr_err("APIC-access addr = 0x%016llx ", vmcs_read64(APIC_ACCESS_ADDR));
5745 pr_cont("virt-APIC addr = 0x%016llx\n", vmcs_read64(VIRTUAL_APIC_PAGE_ADDR));
5747 if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
5748 pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
5749 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
5750 pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
5751 n = vmcs_read32(CR3_TARGET_COUNT);
5752 for (i = 0; i + 1 < n; i += 4)
5753 pr_err("CR3 target%u=%016lx target%u=%016lx\n",
5754 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
5755 i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
5757 pr_err("CR3 target%u=%016lx\n",
5758 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
5759 if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
5760 pr_err("PLE Gap=%08x Window=%08x\n",
5761 vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
5762 if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
5763 pr_err("Virtual processor ID = 0x%04x\n",
5764 vmcs_read16(VIRTUAL_PROCESSOR_ID));
5768 * The guest has exited. See if we can fix it or if we need userspace
5771 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
5773 struct vcpu_vmx *vmx = to_vmx(vcpu);
5774 u32 exit_reason = vmx->exit_reason;
5775 u32 vectoring_info = vmx->idt_vectoring_info;
5777 trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
5780 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
5781 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
5782 * querying dirty_bitmap, we only need to kick all vcpus out of guest
5783 * mode as if vcpus is in root mode, the PML buffer must has been
5787 vmx_flush_pml_buffer(vcpu);
5789 /* If guest state is invalid, start emulating */
5790 if (vmx->emulation_required)
5791 return handle_invalid_guest_state(vcpu);
5793 if (is_guest_mode(vcpu) && nested_vmx_exit_reflected(vcpu, exit_reason))
5794 return nested_vmx_reflect_vmexit(vcpu, exit_reason);
5796 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
5798 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5799 vcpu->run->fail_entry.hardware_entry_failure_reason
5804 if (unlikely(vmx->fail)) {
5805 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5806 vcpu->run->fail_entry.hardware_entry_failure_reason
5807 = vmcs_read32(VM_INSTRUCTION_ERROR);
5813 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
5814 * delivery event since it indicates guest is accessing MMIO.
5815 * The vm-exit can be triggered again after return to guest that
5816 * will cause infinite loop.
5818 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
5819 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
5820 exit_reason != EXIT_REASON_EPT_VIOLATION &&
5821 exit_reason != EXIT_REASON_PML_FULL &&
5822 exit_reason != EXIT_REASON_TASK_SWITCH)) {
5823 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5824 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
5825 vcpu->run->internal.ndata = 3;
5826 vcpu->run->internal.data[0] = vectoring_info;
5827 vcpu->run->internal.data[1] = exit_reason;
5828 vcpu->run->internal.data[2] = vcpu->arch.exit_qualification;
5829 if (exit_reason == EXIT_REASON_EPT_MISCONFIG) {
5830 vcpu->run->internal.ndata++;
5831 vcpu->run->internal.data[3] =
5832 vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5837 if (unlikely(!enable_vnmi &&
5838 vmx->loaded_vmcs->soft_vnmi_blocked)) {
5839 if (vmx_interrupt_allowed(vcpu)) {
5840 vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5841 } else if (vmx->loaded_vmcs->vnmi_blocked_time > 1000000000LL &&
5842 vcpu->arch.nmi_pending) {
5844 * This CPU don't support us in finding the end of an
5845 * NMI-blocked window if the guest runs with IRQs
5846 * disabled. So we pull the trigger after 1 s of
5847 * futile waiting, but inform the user about this.
5849 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
5850 "state on VCPU %d after 1 s timeout\n",
5851 __func__, vcpu->vcpu_id);
5852 vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5856 if (exit_reason < kvm_vmx_max_exit_handlers
5857 && kvm_vmx_exit_handlers[exit_reason])
5858 return kvm_vmx_exit_handlers[exit_reason](vcpu);
5860 vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n",
5862 kvm_queue_exception(vcpu, UD_VECTOR);
5868 * Software based L1D cache flush which is used when microcode providing
5869 * the cache control MSR is not loaded.
5871 * The L1D cache is 32 KiB on Nehalem and later microarchitectures, but to
5872 * flush it is required to read in 64 KiB because the replacement algorithm
5873 * is not exactly LRU. This could be sized at runtime via topology
5874 * information but as all relevant affected CPUs have 32KiB L1D cache size
5875 * there is no point in doing so.
5877 static void vmx_l1d_flush(struct kvm_vcpu *vcpu)
5879 int size = PAGE_SIZE << L1D_CACHE_ORDER;
5882 * This code is only executed when the the flush mode is 'cond' or
5885 if (static_branch_likely(&vmx_l1d_flush_cond)) {
5889 * Clear the per-vcpu flush bit, it gets set again
5890 * either from vcpu_run() or from one of the unsafe
5893 flush_l1d = vcpu->arch.l1tf_flush_l1d;
5894 vcpu->arch.l1tf_flush_l1d = false;
5897 * Clear the per-cpu flush bit, it gets set again from
5898 * the interrupt handlers.
5900 flush_l1d |= kvm_get_cpu_l1tf_flush_l1d();
5901 kvm_clear_cpu_l1tf_flush_l1d();
5907 vcpu->stat.l1d_flush++;
5909 if (static_cpu_has(X86_FEATURE_FLUSH_L1D)) {
5910 wrmsrl(MSR_IA32_FLUSH_CMD, L1D_FLUSH);
5915 /* First ensure the pages are in the TLB */
5916 "xorl %%eax, %%eax\n"
5917 ".Lpopulate_tlb:\n\t"
5918 "movzbl (%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
5919 "addl $4096, %%eax\n\t"
5920 "cmpl %%eax, %[size]\n\t"
5921 "jne .Lpopulate_tlb\n\t"
5922 "xorl %%eax, %%eax\n\t"
5924 /* Now fill the cache */
5925 "xorl %%eax, %%eax\n"
5927 "movzbl (%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
5928 "addl $64, %%eax\n\t"
5929 "cmpl %%eax, %[size]\n\t"
5930 "jne .Lfill_cache\n\t"
5932 :: [flush_pages] "r" (vmx_l1d_flush_pages),
5934 : "eax", "ebx", "ecx", "edx");
5937 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
5939 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5941 if (is_guest_mode(vcpu) &&
5942 nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
5945 if (irr == -1 || tpr < irr) {
5946 vmcs_write32(TPR_THRESHOLD, 0);
5950 vmcs_write32(TPR_THRESHOLD, irr);
5953 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu)
5955 u32 sec_exec_control;
5957 if (!lapic_in_kernel(vcpu))
5960 if (!flexpriority_enabled &&
5961 !cpu_has_vmx_virtualize_x2apic_mode())
5964 /* Postpone execution until vmcs01 is the current VMCS. */
5965 if (is_guest_mode(vcpu)) {
5966 to_vmx(vcpu)->nested.change_vmcs01_virtual_apic_mode = true;
5970 sec_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
5971 sec_exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
5972 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
5974 switch (kvm_get_apic_mode(vcpu)) {
5975 case LAPIC_MODE_INVALID:
5976 WARN_ONCE(true, "Invalid local APIC state");
5977 case LAPIC_MODE_DISABLED:
5979 case LAPIC_MODE_XAPIC:
5980 if (flexpriority_enabled) {
5982 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
5983 vmx_flush_tlb(vcpu, true);
5986 case LAPIC_MODE_X2APIC:
5987 if (cpu_has_vmx_virtualize_x2apic_mode())
5989 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
5992 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, sec_exec_control);
5994 vmx_update_msr_bitmap(vcpu);
5997 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
5999 if (!is_guest_mode(vcpu)) {
6000 vmcs_write64(APIC_ACCESS_ADDR, hpa);
6001 vmx_flush_tlb(vcpu, true);
6005 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
6013 status = vmcs_read16(GUEST_INTR_STATUS);
6015 if (max_isr != old) {
6017 status |= max_isr << 8;
6018 vmcs_write16(GUEST_INTR_STATUS, status);
6022 static void vmx_set_rvi(int vector)
6030 status = vmcs_read16(GUEST_INTR_STATUS);
6031 old = (u8)status & 0xff;
6032 if ((u8)vector != old) {
6034 status |= (u8)vector;
6035 vmcs_write16(GUEST_INTR_STATUS, status);
6039 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
6042 * When running L2, updating RVI is only relevant when
6043 * vmcs12 virtual-interrupt-delivery enabled.
6044 * However, it can be enabled only when L1 also
6045 * intercepts external-interrupts and in that case
6046 * we should not update vmcs02 RVI but instead intercept
6047 * interrupt. Therefore, do nothing when running L2.
6049 if (!is_guest_mode(vcpu))
6050 vmx_set_rvi(max_irr);
6053 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
6055 struct vcpu_vmx *vmx = to_vmx(vcpu);
6057 bool max_irr_updated;
6059 WARN_ON(!vcpu->arch.apicv_active);
6060 if (pi_test_on(&vmx->pi_desc)) {
6061 pi_clear_on(&vmx->pi_desc);
6063 * IOMMU can write to PIR.ON, so the barrier matters even on UP.
6064 * But on x86 this is just a compiler barrier anyway.
6066 smp_mb__after_atomic();
6068 kvm_apic_update_irr(vcpu, vmx->pi_desc.pir, &max_irr);
6071 * If we are running L2 and L1 has a new pending interrupt
6072 * which can be injected, we should re-evaluate
6073 * what should be done with this new L1 interrupt.
6074 * If L1 intercepts external-interrupts, we should
6075 * exit from L2 to L1. Otherwise, interrupt should be
6076 * delivered directly to L2.
6078 if (is_guest_mode(vcpu) && max_irr_updated) {
6079 if (nested_exit_on_intr(vcpu))
6080 kvm_vcpu_exiting_guest_mode(vcpu);
6082 kvm_make_request(KVM_REQ_EVENT, vcpu);
6085 max_irr = kvm_lapic_find_highest_irr(vcpu);
6087 vmx_hwapic_irr_update(vcpu, max_irr);
6091 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
6093 if (!kvm_vcpu_apicv_active(vcpu))
6096 vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
6097 vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
6098 vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
6099 vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
6102 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
6104 struct vcpu_vmx *vmx = to_vmx(vcpu);
6106 pi_clear_on(&vmx->pi_desc);
6107 memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
6110 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
6112 u32 exit_intr_info = 0;
6113 u16 basic_exit_reason = (u16)vmx->exit_reason;
6115 if (!(basic_exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
6116 || basic_exit_reason == EXIT_REASON_EXCEPTION_NMI))
6119 if (!(vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
6120 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6121 vmx->exit_intr_info = exit_intr_info;
6123 /* if exit due to PF check for async PF */
6124 if (is_page_fault(exit_intr_info))
6125 vmx->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason();
6127 /* Handle machine checks before interrupts are enabled */
6128 if (basic_exit_reason == EXIT_REASON_MCE_DURING_VMENTRY ||
6129 is_machine_check(exit_intr_info))
6130 kvm_machine_check();
6132 /* We need to handle NMIs before interrupts are enabled */
6133 if (is_nmi(exit_intr_info)) {
6134 kvm_before_interrupt(&vmx->vcpu);
6136 kvm_after_interrupt(&vmx->vcpu);
6140 static void vmx_handle_external_intr(struct kvm_vcpu *vcpu)
6142 u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6144 if ((exit_intr_info & (INTR_INFO_VALID_MASK | INTR_INFO_INTR_TYPE_MASK))
6145 == (INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR)) {
6146 unsigned int vector;
6147 unsigned long entry;
6149 struct vcpu_vmx *vmx = to_vmx(vcpu);
6150 #ifdef CONFIG_X86_64
6154 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
6155 desc = (gate_desc *)vmx->host_idt_base + vector;
6156 entry = gate_offset(desc);
6158 #ifdef CONFIG_X86_64
6159 "mov %%" _ASM_SP ", %[sp]\n\t"
6160 "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
6165 __ASM_SIZE(push) " $%c[cs]\n\t"
6168 #ifdef CONFIG_X86_64
6173 THUNK_TARGET(entry),
6174 [ss]"i"(__KERNEL_DS),
6175 [cs]"i"(__KERNEL_CS)
6179 STACK_FRAME_NON_STANDARD(vmx_handle_external_intr);
6181 static bool vmx_has_emulated_msr(int index)
6184 case MSR_IA32_SMBASE:
6186 * We cannot do SMM unless we can run the guest in big
6189 return enable_unrestricted_guest || emulate_invalid_guest_state;
6190 case MSR_AMD64_VIRT_SPEC_CTRL:
6191 /* This is AMD only. */
6198 static bool vmx_pt_supported(void)
6200 return pt_mode == PT_MODE_HOST_GUEST;
6203 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
6208 bool idtv_info_valid;
6210 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6213 if (vmx->loaded_vmcs->nmi_known_unmasked)
6216 * Can't use vmx->exit_intr_info since we're not sure what
6217 * the exit reason is.
6219 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6220 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
6221 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
6223 * SDM 3: 27.7.1.2 (September 2008)
6224 * Re-set bit "block by NMI" before VM entry if vmexit caused by
6225 * a guest IRET fault.
6226 * SDM 3: 23.2.2 (September 2008)
6227 * Bit 12 is undefined in any of the following cases:
6228 * If the VM exit sets the valid bit in the IDT-vectoring
6229 * information field.
6230 * If the VM exit is due to a double fault.
6232 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
6233 vector != DF_VECTOR && !idtv_info_valid)
6234 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
6235 GUEST_INTR_STATE_NMI);
6237 vmx->loaded_vmcs->nmi_known_unmasked =
6238 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
6239 & GUEST_INTR_STATE_NMI);
6240 } else if (unlikely(vmx->loaded_vmcs->soft_vnmi_blocked))
6241 vmx->loaded_vmcs->vnmi_blocked_time +=
6242 ktime_to_ns(ktime_sub(ktime_get(),
6243 vmx->loaded_vmcs->entry_time));
6246 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
6247 u32 idt_vectoring_info,
6248 int instr_len_field,
6249 int error_code_field)
6253 bool idtv_info_valid;
6255 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6257 vcpu->arch.nmi_injected = false;
6258 kvm_clear_exception_queue(vcpu);
6259 kvm_clear_interrupt_queue(vcpu);
6261 if (!idtv_info_valid)
6264 kvm_make_request(KVM_REQ_EVENT, vcpu);
6266 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
6267 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
6270 case INTR_TYPE_NMI_INTR:
6271 vcpu->arch.nmi_injected = true;
6273 * SDM 3: 27.7.1.2 (September 2008)
6274 * Clear bit "block by NMI" before VM entry if a NMI
6277 vmx_set_nmi_mask(vcpu, false);
6279 case INTR_TYPE_SOFT_EXCEPTION:
6280 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6282 case INTR_TYPE_HARD_EXCEPTION:
6283 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
6284 u32 err = vmcs_read32(error_code_field);
6285 kvm_requeue_exception_e(vcpu, vector, err);
6287 kvm_requeue_exception(vcpu, vector);
6289 case INTR_TYPE_SOFT_INTR:
6290 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6292 case INTR_TYPE_EXT_INTR:
6293 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
6300 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
6302 __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
6303 VM_EXIT_INSTRUCTION_LEN,
6304 IDT_VECTORING_ERROR_CODE);
6307 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
6309 __vmx_complete_interrupts(vcpu,
6310 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
6311 VM_ENTRY_INSTRUCTION_LEN,
6312 VM_ENTRY_EXCEPTION_ERROR_CODE);
6314 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
6317 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
6320 struct perf_guest_switch_msr *msrs;
6322 msrs = perf_guest_get_msrs(&nr_msrs);
6327 for (i = 0; i < nr_msrs; i++)
6328 if (msrs[i].host == msrs[i].guest)
6329 clear_atomic_switch_msr(vmx, msrs[i].msr);
6331 add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
6332 msrs[i].host, false);
6335 static void vmx_arm_hv_timer(struct vcpu_vmx *vmx, u32 val)
6337 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, val);
6338 if (!vmx->loaded_vmcs->hv_timer_armed)
6339 vmcs_set_bits(PIN_BASED_VM_EXEC_CONTROL,
6340 PIN_BASED_VMX_PREEMPTION_TIMER);
6341 vmx->loaded_vmcs->hv_timer_armed = true;
6344 static void vmx_update_hv_timer(struct kvm_vcpu *vcpu)
6346 struct vcpu_vmx *vmx = to_vmx(vcpu);
6350 if (vmx->req_immediate_exit) {
6351 vmx_arm_hv_timer(vmx, 0);
6355 if (vmx->hv_deadline_tsc != -1) {
6357 if (vmx->hv_deadline_tsc > tscl)
6358 /* set_hv_timer ensures the delta fits in 32-bits */
6359 delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
6360 cpu_preemption_timer_multi);
6364 vmx_arm_hv_timer(vmx, delta_tsc);
6368 if (vmx->loaded_vmcs->hv_timer_armed)
6369 vmcs_clear_bits(PIN_BASED_VM_EXEC_CONTROL,
6370 PIN_BASED_VMX_PREEMPTION_TIMER);
6371 vmx->loaded_vmcs->hv_timer_armed = false;
6374 void vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp)
6376 if (unlikely(host_rsp != vmx->loaded_vmcs->host_state.rsp)) {
6377 vmx->loaded_vmcs->host_state.rsp = host_rsp;
6378 vmcs_writel(HOST_RSP, host_rsp);
6382 bool __vmx_vcpu_run(struct vcpu_vmx *vmx, unsigned long *regs, bool launched);
6384 static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
6386 struct vcpu_vmx *vmx = to_vmx(vcpu);
6387 unsigned long cr3, cr4;
6389 /* Record the guest's net vcpu time for enforced NMI injections. */
6390 if (unlikely(!enable_vnmi &&
6391 vmx->loaded_vmcs->soft_vnmi_blocked))
6392 vmx->loaded_vmcs->entry_time = ktime_get();
6394 /* Don't enter VMX if guest state is invalid, let the exit handler
6395 start emulation until we arrive back to a valid state */
6396 if (vmx->emulation_required)
6399 if (vmx->ple_window_dirty) {
6400 vmx->ple_window_dirty = false;
6401 vmcs_write32(PLE_WINDOW, vmx->ple_window);
6404 if (vmx->nested.need_vmcs12_sync)
6405 nested_sync_from_vmcs12(vcpu);
6407 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
6408 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
6409 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
6410 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
6412 cr3 = __get_current_cr3_fast();
6413 if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
6414 vmcs_writel(HOST_CR3, cr3);
6415 vmx->loaded_vmcs->host_state.cr3 = cr3;
6418 cr4 = cr4_read_shadow();
6419 if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
6420 vmcs_writel(HOST_CR4, cr4);
6421 vmx->loaded_vmcs->host_state.cr4 = cr4;
6424 /* When single-stepping over STI and MOV SS, we must clear the
6425 * corresponding interruptibility bits in the guest state. Otherwise
6426 * vmentry fails as it then expects bit 14 (BS) in pending debug
6427 * exceptions being set, but that's not correct for the guest debugging
6429 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6430 vmx_set_interrupt_shadow(vcpu, 0);
6432 kvm_load_guest_xcr0(vcpu);
6434 if (static_cpu_has(X86_FEATURE_PKU) &&
6435 kvm_read_cr4_bits(vcpu, X86_CR4_PKE) &&
6436 vcpu->arch.pkru != vmx->host_pkru)
6437 __write_pkru(vcpu->arch.pkru);
6439 pt_guest_enter(vmx);
6441 atomic_switch_perf_msrs(vmx);
6443 vmx_update_hv_timer(vcpu);
6446 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
6447 * it's non-zero. Since vmentry is serialising on affected CPUs, there
6448 * is no need to worry about the conditional branch over the wrmsr
6449 * being speculatively taken.
6451 x86_spec_ctrl_set_guest(vmx->spec_ctrl, 0);
6453 /* L1D Flush includes CPU buffer clear to mitigate MDS */
6454 if (static_branch_unlikely(&vmx_l1d_should_flush))
6455 vmx_l1d_flush(vcpu);
6456 else if (static_branch_unlikely(&mds_user_clear))
6457 mds_clear_cpu_buffers();
6459 if (vcpu->arch.cr2 != read_cr2())
6460 write_cr2(vcpu->arch.cr2);
6462 vmx->fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
6463 vmx->loaded_vmcs->launched);
6465 vcpu->arch.cr2 = read_cr2();
6468 * We do not use IBRS in the kernel. If this vCPU has used the
6469 * SPEC_CTRL MSR it may have left it on; save the value and
6470 * turn it off. This is much more efficient than blindly adding
6471 * it to the atomic save/restore list. Especially as the former
6472 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
6474 * For non-nested case:
6475 * If the L01 MSR bitmap does not intercept the MSR, then we need to
6479 * If the L02 MSR bitmap does not intercept the MSR, then we need to
6482 if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
6483 vmx->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
6485 x86_spec_ctrl_restore_host(vmx->spec_ctrl, 0);
6487 /* All fields are clean at this point */
6488 if (static_branch_unlikely(&enable_evmcs))
6489 current_evmcs->hv_clean_fields |=
6490 HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
6492 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
6493 if (vmx->host_debugctlmsr)
6494 update_debugctlmsr(vmx->host_debugctlmsr);
6496 #ifndef CONFIG_X86_64
6498 * The sysexit path does not restore ds/es, so we must set them to
6499 * a reasonable value ourselves.
6501 * We can't defer this to vmx_prepare_switch_to_host() since that
6502 * function may be executed in interrupt context, which saves and
6503 * restore segments around it, nullifying its effect.
6505 loadsegment(ds, __USER_DS);
6506 loadsegment(es, __USER_DS);
6509 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
6510 | (1 << VCPU_EXREG_RFLAGS)
6511 | (1 << VCPU_EXREG_PDPTR)
6512 | (1 << VCPU_EXREG_SEGMENTS)
6513 | (1 << VCPU_EXREG_CR3));
6514 vcpu->arch.regs_dirty = 0;
6519 * eager fpu is enabled if PKEY is supported and CR4 is switched
6520 * back on host, so it is safe to read guest PKRU from current
6523 if (static_cpu_has(X86_FEATURE_PKU) &&
6524 kvm_read_cr4_bits(vcpu, X86_CR4_PKE)) {
6525 vcpu->arch.pkru = rdpkru();
6526 if (vcpu->arch.pkru != vmx->host_pkru)
6527 __write_pkru(vmx->host_pkru);
6530 kvm_put_guest_xcr0(vcpu);
6532 vmx->nested.nested_run_pending = 0;
6533 vmx->idt_vectoring_info = 0;
6535 vmx->exit_reason = vmx->fail ? 0xdead : vmcs_read32(VM_EXIT_REASON);
6536 if (vmx->fail || (vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
6539 vmx->loaded_vmcs->launched = 1;
6540 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
6542 vmx_complete_atomic_exit(vmx);
6543 vmx_recover_nmi_blocking(vmx);
6544 vmx_complete_interrupts(vmx);
6547 static struct kvm *vmx_vm_alloc(void)
6549 struct kvm_vmx *kvm_vmx = __vmalloc(sizeof(struct kvm_vmx),
6550 GFP_KERNEL_ACCOUNT | __GFP_ZERO,
6552 return &kvm_vmx->kvm;
6555 static void vmx_vm_free(struct kvm *kvm)
6557 vfree(to_kvm_vmx(kvm));
6560 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
6562 struct vcpu_vmx *vmx = to_vmx(vcpu);
6565 vmx_destroy_pml_buffer(vmx);
6566 free_vpid(vmx->vpid);
6567 nested_vmx_free_vcpu(vcpu);
6568 free_loaded_vmcs(vmx->loaded_vmcs);
6569 kfree(vmx->guest_msrs);
6570 kvm_vcpu_uninit(vcpu);
6571 kmem_cache_free(x86_fpu_cache, vmx->vcpu.arch.guest_fpu);
6572 kmem_cache_free(kvm_vcpu_cache, vmx);
6575 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
6578 struct vcpu_vmx *vmx;
6579 unsigned long *msr_bitmap;
6582 vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT);
6584 return ERR_PTR(-ENOMEM);
6586 vmx->vcpu.arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
6587 GFP_KERNEL_ACCOUNT);
6588 if (!vmx->vcpu.arch.guest_fpu) {
6589 printk(KERN_ERR "kvm: failed to allocate vcpu's fpu\n");
6591 goto free_partial_vcpu;
6594 vmx->vpid = allocate_vpid();
6596 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
6603 * If PML is turned on, failure on enabling PML just results in failure
6604 * of creating the vcpu, therefore we can simplify PML logic (by
6605 * avoiding dealing with cases, such as enabling PML partially on vcpus
6606 * for the guest, etc.
6609 vmx->pml_pg = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
6614 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL_ACCOUNT);
6615 BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) * sizeof(vmx->guest_msrs[0])
6618 if (!vmx->guest_msrs)
6621 err = alloc_loaded_vmcs(&vmx->vmcs01);
6625 msr_bitmap = vmx->vmcs01.msr_bitmap;
6626 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_TSC, MSR_TYPE_R);
6627 vmx_disable_intercept_for_msr(msr_bitmap, MSR_FS_BASE, MSR_TYPE_RW);
6628 vmx_disable_intercept_for_msr(msr_bitmap, MSR_GS_BASE, MSR_TYPE_RW);
6629 vmx_disable_intercept_for_msr(msr_bitmap, MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
6630 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_CS, MSR_TYPE_RW);
6631 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_ESP, MSR_TYPE_RW);
6632 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_EIP, MSR_TYPE_RW);
6633 vmx->msr_bitmap_mode = 0;
6635 vmx->loaded_vmcs = &vmx->vmcs01;
6637 vmx_vcpu_load(&vmx->vcpu, cpu);
6638 vmx->vcpu.cpu = cpu;
6639 vmx_vcpu_setup(vmx);
6640 vmx_vcpu_put(&vmx->vcpu);
6642 if (cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
6643 err = alloc_apic_access_page(kvm);
6648 if (enable_ept && !enable_unrestricted_guest) {
6649 err = init_rmode_identity_map(kvm);
6655 nested_vmx_setup_ctls_msrs(&vmx->nested.msrs,
6657 kvm_vcpu_apicv_active(&vmx->vcpu));
6659 memset(&vmx->nested.msrs, 0, sizeof(vmx->nested.msrs));
6661 vmx->nested.posted_intr_nv = -1;
6662 vmx->nested.current_vmptr = -1ull;
6664 vmx->msr_ia32_feature_control_valid_bits = FEATURE_CONTROL_LOCKED;
6667 * Enforce invariant: pi_desc.nv is always either POSTED_INTR_VECTOR
6668 * or POSTED_INTR_WAKEUP_VECTOR.
6670 vmx->pi_desc.nv = POSTED_INTR_VECTOR;
6671 vmx->pi_desc.sn = 1;
6673 vmx->ept_pointer = INVALID_PAGE;
6678 free_loaded_vmcs(vmx->loaded_vmcs);
6680 kfree(vmx->guest_msrs);
6682 vmx_destroy_pml_buffer(vmx);
6684 kvm_vcpu_uninit(&vmx->vcpu);
6686 free_vpid(vmx->vpid);
6687 kmem_cache_free(x86_fpu_cache, vmx->vcpu.arch.guest_fpu);
6689 kmem_cache_free(kvm_vcpu_cache, vmx);
6690 return ERR_PTR(err);
6693 #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"
6694 #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"
6696 static int vmx_vm_init(struct kvm *kvm)
6698 spin_lock_init(&to_kvm_vmx(kvm)->ept_pointer_lock);
6701 kvm->arch.pause_in_guest = true;
6703 if (boot_cpu_has(X86_BUG_L1TF) && enable_ept) {
6704 switch (l1tf_mitigation) {
6705 case L1TF_MITIGATION_OFF:
6706 case L1TF_MITIGATION_FLUSH_NOWARN:
6707 /* 'I explicitly don't care' is set */
6709 case L1TF_MITIGATION_FLUSH:
6710 case L1TF_MITIGATION_FLUSH_NOSMT:
6711 case L1TF_MITIGATION_FULL:
6713 * Warn upon starting the first VM in a potentially
6714 * insecure environment.
6716 if (sched_smt_active())
6717 pr_warn_once(L1TF_MSG_SMT);
6718 if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER)
6719 pr_warn_once(L1TF_MSG_L1D);
6721 case L1TF_MITIGATION_FULL_FORCE:
6722 /* Flush is enforced */
6729 static void __init vmx_check_processor_compat(void *rtn)
6731 struct vmcs_config vmcs_conf;
6732 struct vmx_capability vmx_cap;
6735 if (setup_vmcs_config(&vmcs_conf, &vmx_cap) < 0)
6738 nested_vmx_setup_ctls_msrs(&vmcs_conf.nested, vmx_cap.ept,
6740 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
6741 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
6742 smp_processor_id());
6747 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
6752 /* For VT-d and EPT combination
6753 * 1. MMIO: always map as UC
6755 * a. VT-d without snooping control feature: can't guarantee the
6756 * result, try to trust guest.
6757 * b. VT-d with snooping control feature: snooping control feature of
6758 * VT-d engine can guarantee the cache correctness. Just set it
6759 * to WB to keep consistent with host. So the same as item 3.
6760 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
6761 * consistent with host MTRR
6764 cache = MTRR_TYPE_UNCACHABLE;
6768 if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
6769 ipat = VMX_EPT_IPAT_BIT;
6770 cache = MTRR_TYPE_WRBACK;
6774 if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
6775 ipat = VMX_EPT_IPAT_BIT;
6776 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
6777 cache = MTRR_TYPE_WRBACK;
6779 cache = MTRR_TYPE_UNCACHABLE;
6783 cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
6786 return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
6789 static int vmx_get_lpage_level(void)
6791 if (enable_ept && !cpu_has_vmx_ept_1g_page())
6792 return PT_DIRECTORY_LEVEL;
6794 /* For shadow and EPT supported 1GB page */
6795 return PT_PDPE_LEVEL;
6798 static void vmcs_set_secondary_exec_control(u32 new_ctl)
6801 * These bits in the secondary execution controls field
6802 * are dynamic, the others are mostly based on the hypervisor
6803 * architecture and the guest's CPUID. Do not touch the
6807 SECONDARY_EXEC_SHADOW_VMCS |
6808 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
6809 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6810 SECONDARY_EXEC_DESC;
6812 u32 cur_ctl = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
6814 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
6815 (new_ctl & ~mask) | (cur_ctl & mask));
6819 * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
6820 * (indicating "allowed-1") if they are supported in the guest's CPUID.
6822 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
6824 struct vcpu_vmx *vmx = to_vmx(vcpu);
6825 struct kvm_cpuid_entry2 *entry;
6827 vmx->nested.msrs.cr0_fixed1 = 0xffffffff;
6828 vmx->nested.msrs.cr4_fixed1 = X86_CR4_PCE;
6830 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do { \
6831 if (entry && (entry->_reg & (_cpuid_mask))) \
6832 vmx->nested.msrs.cr4_fixed1 |= (_cr4_mask); \
6835 entry = kvm_find_cpuid_entry(vcpu, 0x1, 0);
6836 cr4_fixed1_update(X86_CR4_VME, edx, bit(X86_FEATURE_VME));
6837 cr4_fixed1_update(X86_CR4_PVI, edx, bit(X86_FEATURE_VME));
6838 cr4_fixed1_update(X86_CR4_TSD, edx, bit(X86_FEATURE_TSC));
6839 cr4_fixed1_update(X86_CR4_DE, edx, bit(X86_FEATURE_DE));
6840 cr4_fixed1_update(X86_CR4_PSE, edx, bit(X86_FEATURE_PSE));
6841 cr4_fixed1_update(X86_CR4_PAE, edx, bit(X86_FEATURE_PAE));
6842 cr4_fixed1_update(X86_CR4_MCE, edx, bit(X86_FEATURE_MCE));
6843 cr4_fixed1_update(X86_CR4_PGE, edx, bit(X86_FEATURE_PGE));
6844 cr4_fixed1_update(X86_CR4_OSFXSR, edx, bit(X86_FEATURE_FXSR));
6845 cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, bit(X86_FEATURE_XMM));
6846 cr4_fixed1_update(X86_CR4_VMXE, ecx, bit(X86_FEATURE_VMX));
6847 cr4_fixed1_update(X86_CR4_SMXE, ecx, bit(X86_FEATURE_SMX));
6848 cr4_fixed1_update(X86_CR4_PCIDE, ecx, bit(X86_FEATURE_PCID));
6849 cr4_fixed1_update(X86_CR4_OSXSAVE, ecx, bit(X86_FEATURE_XSAVE));
6851 entry = kvm_find_cpuid_entry(vcpu, 0x7, 0);
6852 cr4_fixed1_update(X86_CR4_FSGSBASE, ebx, bit(X86_FEATURE_FSGSBASE));
6853 cr4_fixed1_update(X86_CR4_SMEP, ebx, bit(X86_FEATURE_SMEP));
6854 cr4_fixed1_update(X86_CR4_SMAP, ebx, bit(X86_FEATURE_SMAP));
6855 cr4_fixed1_update(X86_CR4_PKE, ecx, bit(X86_FEATURE_PKU));
6856 cr4_fixed1_update(X86_CR4_UMIP, ecx, bit(X86_FEATURE_UMIP));
6858 #undef cr4_fixed1_update
6861 static void nested_vmx_entry_exit_ctls_update(struct kvm_vcpu *vcpu)
6863 struct vcpu_vmx *vmx = to_vmx(vcpu);
6865 if (kvm_mpx_supported()) {
6866 bool mpx_enabled = guest_cpuid_has(vcpu, X86_FEATURE_MPX);
6869 vmx->nested.msrs.entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
6870 vmx->nested.msrs.exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
6872 vmx->nested.msrs.entry_ctls_high &= ~VM_ENTRY_LOAD_BNDCFGS;
6873 vmx->nested.msrs.exit_ctls_high &= ~VM_EXIT_CLEAR_BNDCFGS;
6878 static void update_intel_pt_cfg(struct kvm_vcpu *vcpu)
6880 struct vcpu_vmx *vmx = to_vmx(vcpu);
6881 struct kvm_cpuid_entry2 *best = NULL;
6884 for (i = 0; i < PT_CPUID_LEAVES; i++) {
6885 best = kvm_find_cpuid_entry(vcpu, 0x14, i);
6888 vmx->pt_desc.caps[CPUID_EAX + i*PT_CPUID_REGS_NUM] = best->eax;
6889 vmx->pt_desc.caps[CPUID_EBX + i*PT_CPUID_REGS_NUM] = best->ebx;
6890 vmx->pt_desc.caps[CPUID_ECX + i*PT_CPUID_REGS_NUM] = best->ecx;
6891 vmx->pt_desc.caps[CPUID_EDX + i*PT_CPUID_REGS_NUM] = best->edx;
6894 /* Get the number of configurable Address Ranges for filtering */
6895 vmx->pt_desc.addr_range = intel_pt_validate_cap(vmx->pt_desc.caps,
6896 PT_CAP_num_address_ranges);
6898 /* Initialize and clear the no dependency bits */
6899 vmx->pt_desc.ctl_bitmask = ~(RTIT_CTL_TRACEEN | RTIT_CTL_OS |
6900 RTIT_CTL_USR | RTIT_CTL_TSC_EN | RTIT_CTL_DISRETC);
6903 * If CPUID.(EAX=14H,ECX=0):EBX[0]=1 CR3Filter can be set otherwise
6904 * will inject an #GP
6906 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_cr3_filtering))
6907 vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_CR3EN;
6910 * If CPUID.(EAX=14H,ECX=0):EBX[1]=1 CYCEn, CycThresh and
6911 * PSBFreq can be set
6913 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc))
6914 vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_CYCLEACC |
6915 RTIT_CTL_CYC_THRESH | RTIT_CTL_PSB_FREQ);
6918 * If CPUID.(EAX=14H,ECX=0):EBX[3]=1 MTCEn BranchEn and
6919 * MTCFreq can be set
6921 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc))
6922 vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_MTC_EN |
6923 RTIT_CTL_BRANCH_EN | RTIT_CTL_MTC_RANGE);
6925 /* If CPUID.(EAX=14H,ECX=0):EBX[4]=1 FUPonPTW and PTWEn can be set */
6926 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_ptwrite))
6927 vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_FUP_ON_PTW |
6930 /* If CPUID.(EAX=14H,ECX=0):EBX[5]=1 PwrEvEn can be set */
6931 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_power_event_trace))
6932 vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_PWR_EVT_EN;
6934 /* If CPUID.(EAX=14H,ECX=0):ECX[0]=1 ToPA can be set */
6935 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_topa_output))
6936 vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_TOPA;
6938 /* If CPUID.(EAX=14H,ECX=0):ECX[3]=1 FabircEn can be set */
6939 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_output_subsys))
6940 vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_FABRIC_EN;
6942 /* unmask address range configure area */
6943 for (i = 0; i < vmx->pt_desc.addr_range; i++)
6944 vmx->pt_desc.ctl_bitmask &= ~(0xfULL << (32 + i * 4));
6947 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
6949 struct vcpu_vmx *vmx = to_vmx(vcpu);
6951 if (cpu_has_secondary_exec_ctrls()) {
6952 vmx_compute_secondary_exec_control(vmx);
6953 vmcs_set_secondary_exec_control(vmx->secondary_exec_control);
6956 if (nested_vmx_allowed(vcpu))
6957 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
6958 FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
6960 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
6961 ~FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
6963 if (nested_vmx_allowed(vcpu)) {
6964 nested_vmx_cr_fixed1_bits_update(vcpu);
6965 nested_vmx_entry_exit_ctls_update(vcpu);
6968 if (boot_cpu_has(X86_FEATURE_INTEL_PT) &&
6969 guest_cpuid_has(vcpu, X86_FEATURE_INTEL_PT))
6970 update_intel_pt_cfg(vcpu);
6973 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
6975 if (func == 1 && nested)
6976 entry->ecx |= bit(X86_FEATURE_VMX);
6979 static void vmx_request_immediate_exit(struct kvm_vcpu *vcpu)
6981 to_vmx(vcpu)->req_immediate_exit = true;
6984 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
6985 struct x86_instruction_info *info,
6986 enum x86_intercept_stage stage)
6988 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6989 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6992 * RDPID causes #UD if disabled through secondary execution controls.
6993 * Because it is marked as EmulateOnUD, we need to intercept it here.
6995 if (info->intercept == x86_intercept_rdtscp &&
6996 !nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDTSCP)) {
6997 ctxt->exception.vector = UD_VECTOR;
6998 ctxt->exception.error_code_valid = false;
6999 return X86EMUL_PROPAGATE_FAULT;
7002 /* TODO: check more intercepts... */
7003 return X86EMUL_CONTINUE;
7006 #ifdef CONFIG_X86_64
7007 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
7008 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
7009 u64 divisor, u64 *result)
7011 u64 low = a << shift, high = a >> (64 - shift);
7013 /* To avoid the overflow on divq */
7014 if (high >= divisor)
7017 /* Low hold the result, high hold rem which is discarded */
7018 asm("divq %2\n\t" : "=a" (low), "=d" (high) :
7019 "rm" (divisor), "0" (low), "1" (high));
7025 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc,
7028 struct vcpu_vmx *vmx;
7029 u64 tscl, guest_tscl, delta_tsc, lapic_timer_advance_cycles;
7030 struct kvm_timer *ktimer = &vcpu->arch.apic->lapic_timer;
7032 if (kvm_mwait_in_guest(vcpu->kvm))
7037 guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
7038 delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
7039 lapic_timer_advance_cycles = nsec_to_cycles(vcpu,
7040 ktimer->timer_advance_ns);
7042 if (delta_tsc > lapic_timer_advance_cycles)
7043 delta_tsc -= lapic_timer_advance_cycles;
7047 /* Convert to host delta tsc if tsc scaling is enabled */
7048 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
7049 delta_tsc && u64_shl_div_u64(delta_tsc,
7050 kvm_tsc_scaling_ratio_frac_bits,
7051 vcpu->arch.tsc_scaling_ratio, &delta_tsc))
7055 * If the delta tsc can't fit in the 32 bit after the multi shift,
7056 * we can't use the preemption timer.
7057 * It's possible that it fits on later vmentries, but checking
7058 * on every vmentry is costly so we just use an hrtimer.
7060 if (delta_tsc >> (cpu_preemption_timer_multi + 32))
7063 vmx->hv_deadline_tsc = tscl + delta_tsc;
7064 *expired = !delta_tsc;
7068 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
7070 to_vmx(vcpu)->hv_deadline_tsc = -1;
7074 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
7076 if (!kvm_pause_in_guest(vcpu->kvm))
7077 shrink_ple_window(vcpu);
7080 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
7081 struct kvm_memory_slot *slot)
7083 kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
7084 kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
7087 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
7088 struct kvm_memory_slot *slot)
7090 kvm_mmu_slot_set_dirty(kvm, slot);
7093 static void vmx_flush_log_dirty(struct kvm *kvm)
7095 kvm_flush_pml_buffers(kvm);
7098 static int vmx_write_pml_buffer(struct kvm_vcpu *vcpu)
7100 struct vmcs12 *vmcs12;
7101 struct vcpu_vmx *vmx = to_vmx(vcpu);
7104 if (is_guest_mode(vcpu)) {
7105 WARN_ON_ONCE(vmx->nested.pml_full);
7108 * Check if PML is enabled for the nested guest.
7109 * Whether eptp bit 6 is set is already checked
7110 * as part of A/D emulation.
7112 vmcs12 = get_vmcs12(vcpu);
7113 if (!nested_cpu_has_pml(vmcs12))
7116 if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
7117 vmx->nested.pml_full = true;
7121 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS) & ~0xFFFull;
7122 dst = vmcs12->pml_address + sizeof(u64) * vmcs12->guest_pml_index;
7124 if (kvm_write_guest_page(vcpu->kvm, gpa_to_gfn(dst), &gpa,
7125 offset_in_page(dst), sizeof(gpa)))
7128 vmcs12->guest_pml_index--;
7134 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
7135 struct kvm_memory_slot *memslot,
7136 gfn_t offset, unsigned long mask)
7138 kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
7141 static void __pi_post_block(struct kvm_vcpu *vcpu)
7143 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7144 struct pi_desc old, new;
7148 old.control = new.control = pi_desc->control;
7149 WARN(old.nv != POSTED_INTR_WAKEUP_VECTOR,
7150 "Wakeup handler not enabled while the VCPU is blocked\n");
7152 dest = cpu_physical_id(vcpu->cpu);
7154 if (x2apic_enabled())
7157 new.ndst = (dest << 8) & 0xFF00;
7159 /* set 'NV' to 'notification vector' */
7160 new.nv = POSTED_INTR_VECTOR;
7161 } while (cmpxchg64(&pi_desc->control, old.control,
7162 new.control) != old.control);
7164 if (!WARN_ON_ONCE(vcpu->pre_pcpu == -1)) {
7165 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7166 list_del(&vcpu->blocked_vcpu_list);
7167 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7168 vcpu->pre_pcpu = -1;
7173 * This routine does the following things for vCPU which is going
7174 * to be blocked if VT-d PI is enabled.
7175 * - Store the vCPU to the wakeup list, so when interrupts happen
7176 * we can find the right vCPU to wake up.
7177 * - Change the Posted-interrupt descriptor as below:
7178 * 'NDST' <-- vcpu->pre_pcpu
7179 * 'NV' <-- POSTED_INTR_WAKEUP_VECTOR
7180 * - If 'ON' is set during this process, which means at least one
7181 * interrupt is posted for this vCPU, we cannot block it, in
7182 * this case, return 1, otherwise, return 0.
7185 static int pi_pre_block(struct kvm_vcpu *vcpu)
7188 struct pi_desc old, new;
7189 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7191 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
7192 !irq_remapping_cap(IRQ_POSTING_CAP) ||
7193 !kvm_vcpu_apicv_active(vcpu))
7196 WARN_ON(irqs_disabled());
7197 local_irq_disable();
7198 if (!WARN_ON_ONCE(vcpu->pre_pcpu != -1)) {
7199 vcpu->pre_pcpu = vcpu->cpu;
7200 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7201 list_add_tail(&vcpu->blocked_vcpu_list,
7202 &per_cpu(blocked_vcpu_on_cpu,
7204 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7208 old.control = new.control = pi_desc->control;
7210 WARN((pi_desc->sn == 1),
7211 "Warning: SN field of posted-interrupts "
7212 "is set before blocking\n");
7215 * Since vCPU can be preempted during this process,
7216 * vcpu->cpu could be different with pre_pcpu, we
7217 * need to set pre_pcpu as the destination of wakeup
7218 * notification event, then we can find the right vCPU
7219 * to wakeup in wakeup handler if interrupts happen
7220 * when the vCPU is in blocked state.
7222 dest = cpu_physical_id(vcpu->pre_pcpu);
7224 if (x2apic_enabled())
7227 new.ndst = (dest << 8) & 0xFF00;
7229 /* set 'NV' to 'wakeup vector' */
7230 new.nv = POSTED_INTR_WAKEUP_VECTOR;
7231 } while (cmpxchg64(&pi_desc->control, old.control,
7232 new.control) != old.control);
7234 /* We should not block the vCPU if an interrupt is posted for it. */
7235 if (pi_test_on(pi_desc) == 1)
7236 __pi_post_block(vcpu);
7239 return (vcpu->pre_pcpu == -1);
7242 static int vmx_pre_block(struct kvm_vcpu *vcpu)
7244 if (pi_pre_block(vcpu))
7247 if (kvm_lapic_hv_timer_in_use(vcpu))
7248 kvm_lapic_switch_to_sw_timer(vcpu);
7253 static void pi_post_block(struct kvm_vcpu *vcpu)
7255 if (vcpu->pre_pcpu == -1)
7258 WARN_ON(irqs_disabled());
7259 local_irq_disable();
7260 __pi_post_block(vcpu);
7264 static void vmx_post_block(struct kvm_vcpu *vcpu)
7266 if (kvm_x86_ops->set_hv_timer)
7267 kvm_lapic_switch_to_hv_timer(vcpu);
7269 pi_post_block(vcpu);
7273 * vmx_update_pi_irte - set IRTE for Posted-Interrupts
7276 * @host_irq: host irq of the interrupt
7277 * @guest_irq: gsi of the interrupt
7278 * @set: set or unset PI
7279 * returns 0 on success, < 0 on failure
7281 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
7282 uint32_t guest_irq, bool set)
7284 struct kvm_kernel_irq_routing_entry *e;
7285 struct kvm_irq_routing_table *irq_rt;
7286 struct kvm_lapic_irq irq;
7287 struct kvm_vcpu *vcpu;
7288 struct vcpu_data vcpu_info;
7291 if (!kvm_arch_has_assigned_device(kvm) ||
7292 !irq_remapping_cap(IRQ_POSTING_CAP) ||
7293 !kvm_vcpu_apicv_active(kvm->vcpus[0]))
7296 idx = srcu_read_lock(&kvm->irq_srcu);
7297 irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
7298 if (guest_irq >= irq_rt->nr_rt_entries ||
7299 hlist_empty(&irq_rt->map[guest_irq])) {
7300 pr_warn_once("no route for guest_irq %u/%u (broken user space?)\n",
7301 guest_irq, irq_rt->nr_rt_entries);
7305 hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
7306 if (e->type != KVM_IRQ_ROUTING_MSI)
7309 * VT-d PI cannot support posting multicast/broadcast
7310 * interrupts to a vCPU, we still use interrupt remapping
7311 * for these kind of interrupts.
7313 * For lowest-priority interrupts, we only support
7314 * those with single CPU as the destination, e.g. user
7315 * configures the interrupts via /proc/irq or uses
7316 * irqbalance to make the interrupts single-CPU.
7318 * We will support full lowest-priority interrupt later.
7321 kvm_set_msi_irq(kvm, e, &irq);
7322 if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) {
7324 * Make sure the IRTE is in remapped mode if
7325 * we don't handle it in posted mode.
7327 ret = irq_set_vcpu_affinity(host_irq, NULL);
7330 "failed to back to remapped mode, irq: %u\n",
7338 vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
7339 vcpu_info.vector = irq.vector;
7341 trace_kvm_pi_irte_update(host_irq, vcpu->vcpu_id, e->gsi,
7342 vcpu_info.vector, vcpu_info.pi_desc_addr, set);
7345 ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
7347 ret = irq_set_vcpu_affinity(host_irq, NULL);
7350 printk(KERN_INFO "%s: failed to update PI IRTE\n",
7358 srcu_read_unlock(&kvm->irq_srcu, idx);
7362 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
7364 if (vcpu->arch.mcg_cap & MCG_LMCE_P)
7365 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7366 FEATURE_CONTROL_LMCE;
7368 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7369 ~FEATURE_CONTROL_LMCE;
7372 static int vmx_smi_allowed(struct kvm_vcpu *vcpu)
7374 /* we need a nested vmexit to enter SMM, postpone if run is pending */
7375 if (to_vmx(vcpu)->nested.nested_run_pending)
7380 static int vmx_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
7382 struct vcpu_vmx *vmx = to_vmx(vcpu);
7384 vmx->nested.smm.guest_mode = is_guest_mode(vcpu);
7385 if (vmx->nested.smm.guest_mode)
7386 nested_vmx_vmexit(vcpu, -1, 0, 0);
7388 vmx->nested.smm.vmxon = vmx->nested.vmxon;
7389 vmx->nested.vmxon = false;
7390 vmx_clear_hlt(vcpu);
7394 static int vmx_pre_leave_smm(struct kvm_vcpu *vcpu, const char *smstate)
7396 struct vcpu_vmx *vmx = to_vmx(vcpu);
7399 if (vmx->nested.smm.vmxon) {
7400 vmx->nested.vmxon = true;
7401 vmx->nested.smm.vmxon = false;
7404 if (vmx->nested.smm.guest_mode) {
7405 ret = nested_vmx_enter_non_root_mode(vcpu, false);
7409 vmx->nested.smm.guest_mode = false;
7414 static int enable_smi_window(struct kvm_vcpu *vcpu)
7419 static bool vmx_need_emulation_on_page_fault(struct kvm_vcpu *vcpu)
7424 static __init int hardware_setup(void)
7426 unsigned long host_bndcfgs;
7429 rdmsrl_safe(MSR_EFER, &host_efer);
7431 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
7432 kvm_define_shared_msr(i, vmx_msr_index[i]);
7434 if (setup_vmcs_config(&vmcs_config, &vmx_capability) < 0)
7437 if (boot_cpu_has(X86_FEATURE_NX))
7438 kvm_enable_efer_bits(EFER_NX);
7440 if (boot_cpu_has(X86_FEATURE_MPX)) {
7441 rdmsrl(MSR_IA32_BNDCFGS, host_bndcfgs);
7442 WARN_ONCE(host_bndcfgs, "KVM: BNDCFGS in host will be lost");
7445 if (boot_cpu_has(X86_FEATURE_XSAVES))
7446 rdmsrl(MSR_IA32_XSS, host_xss);
7448 if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
7449 !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
7452 if (!cpu_has_vmx_ept() ||
7453 !cpu_has_vmx_ept_4levels() ||
7454 !cpu_has_vmx_ept_mt_wb() ||
7455 !cpu_has_vmx_invept_global())
7458 if (!cpu_has_vmx_ept_ad_bits() || !enable_ept)
7459 enable_ept_ad_bits = 0;
7461 if (!cpu_has_vmx_unrestricted_guest() || !enable_ept)
7462 enable_unrestricted_guest = 0;
7464 if (!cpu_has_vmx_flexpriority())
7465 flexpriority_enabled = 0;
7467 if (!cpu_has_virtual_nmis())
7471 * set_apic_access_page_addr() is used to reload apic access
7472 * page upon invalidation. No need to do anything if not
7473 * using the APIC_ACCESS_ADDR VMCS field.
7475 if (!flexpriority_enabled)
7476 kvm_x86_ops->set_apic_access_page_addr = NULL;
7478 if (!cpu_has_vmx_tpr_shadow())
7479 kvm_x86_ops->update_cr8_intercept = NULL;
7481 if (enable_ept && !cpu_has_vmx_ept_2m_page())
7482 kvm_disable_largepages();
7484 #if IS_ENABLED(CONFIG_HYPERV)
7485 if (ms_hyperv.nested_features & HV_X64_NESTED_GUEST_MAPPING_FLUSH
7487 kvm_x86_ops->tlb_remote_flush = hv_remote_flush_tlb;
7488 kvm_x86_ops->tlb_remote_flush_with_range =
7489 hv_remote_flush_tlb_with_range;
7493 if (!cpu_has_vmx_ple()) {
7496 ple_window_grow = 0;
7498 ple_window_shrink = 0;
7501 if (!cpu_has_vmx_apicv()) {
7503 kvm_x86_ops->sync_pir_to_irr = NULL;
7506 if (cpu_has_vmx_tsc_scaling()) {
7507 kvm_has_tsc_control = true;
7508 kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
7509 kvm_tsc_scaling_ratio_frac_bits = 48;
7512 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
7520 * Only enable PML when hardware supports PML feature, and both EPT
7521 * and EPT A/D bit features are enabled -- PML depends on them to work.
7523 if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
7527 kvm_x86_ops->slot_enable_log_dirty = NULL;
7528 kvm_x86_ops->slot_disable_log_dirty = NULL;
7529 kvm_x86_ops->flush_log_dirty = NULL;
7530 kvm_x86_ops->enable_log_dirty_pt_masked = NULL;
7533 if (!cpu_has_vmx_preemption_timer())
7534 kvm_x86_ops->request_immediate_exit = __kvm_request_immediate_exit;
7536 if (cpu_has_vmx_preemption_timer() && enable_preemption_timer) {
7539 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
7540 cpu_preemption_timer_multi =
7541 vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
7543 kvm_x86_ops->set_hv_timer = NULL;
7544 kvm_x86_ops->cancel_hv_timer = NULL;
7547 kvm_set_posted_intr_wakeup_handler(wakeup_handler);
7549 kvm_mce_cap_supported |= MCG_LMCE_P;
7551 if (pt_mode != PT_MODE_SYSTEM && pt_mode != PT_MODE_HOST_GUEST)
7553 if (!enable_ept || !cpu_has_vmx_intel_pt())
7554 pt_mode = PT_MODE_SYSTEM;
7557 nested_vmx_setup_ctls_msrs(&vmcs_config.nested,
7558 vmx_capability.ept, enable_apicv);
7560 r = nested_vmx_hardware_setup(kvm_vmx_exit_handlers);
7565 r = alloc_kvm_area();
7567 nested_vmx_hardware_unsetup();
7571 static __exit void hardware_unsetup(void)
7574 nested_vmx_hardware_unsetup();
7579 static struct kvm_x86_ops vmx_x86_ops __ro_after_init = {
7580 .cpu_has_kvm_support = cpu_has_kvm_support,
7581 .disabled_by_bios = vmx_disabled_by_bios,
7582 .hardware_setup = hardware_setup,
7583 .hardware_unsetup = hardware_unsetup,
7584 .check_processor_compatibility = vmx_check_processor_compat,
7585 .hardware_enable = hardware_enable,
7586 .hardware_disable = hardware_disable,
7587 .cpu_has_accelerated_tpr = report_flexpriority,
7588 .has_emulated_msr = vmx_has_emulated_msr,
7590 .vm_init = vmx_vm_init,
7591 .vm_alloc = vmx_vm_alloc,
7592 .vm_free = vmx_vm_free,
7594 .vcpu_create = vmx_create_vcpu,
7595 .vcpu_free = vmx_free_vcpu,
7596 .vcpu_reset = vmx_vcpu_reset,
7598 .prepare_guest_switch = vmx_prepare_switch_to_guest,
7599 .vcpu_load = vmx_vcpu_load,
7600 .vcpu_put = vmx_vcpu_put,
7602 .update_bp_intercept = update_exception_bitmap,
7603 .get_msr_feature = vmx_get_msr_feature,
7604 .get_msr = vmx_get_msr,
7605 .set_msr = vmx_set_msr,
7606 .get_segment_base = vmx_get_segment_base,
7607 .get_segment = vmx_get_segment,
7608 .set_segment = vmx_set_segment,
7609 .get_cpl = vmx_get_cpl,
7610 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
7611 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
7612 .decache_cr3 = vmx_decache_cr3,
7613 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
7614 .set_cr0 = vmx_set_cr0,
7615 .set_cr3 = vmx_set_cr3,
7616 .set_cr4 = vmx_set_cr4,
7617 .set_efer = vmx_set_efer,
7618 .get_idt = vmx_get_idt,
7619 .set_idt = vmx_set_idt,
7620 .get_gdt = vmx_get_gdt,
7621 .set_gdt = vmx_set_gdt,
7622 .get_dr6 = vmx_get_dr6,
7623 .set_dr6 = vmx_set_dr6,
7624 .set_dr7 = vmx_set_dr7,
7625 .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
7626 .cache_reg = vmx_cache_reg,
7627 .get_rflags = vmx_get_rflags,
7628 .set_rflags = vmx_set_rflags,
7630 .tlb_flush = vmx_flush_tlb,
7631 .tlb_flush_gva = vmx_flush_tlb_gva,
7633 .run = vmx_vcpu_run,
7634 .handle_exit = vmx_handle_exit,
7635 .skip_emulated_instruction = skip_emulated_instruction,
7636 .set_interrupt_shadow = vmx_set_interrupt_shadow,
7637 .get_interrupt_shadow = vmx_get_interrupt_shadow,
7638 .patch_hypercall = vmx_patch_hypercall,
7639 .set_irq = vmx_inject_irq,
7640 .set_nmi = vmx_inject_nmi,
7641 .queue_exception = vmx_queue_exception,
7642 .cancel_injection = vmx_cancel_injection,
7643 .interrupt_allowed = vmx_interrupt_allowed,
7644 .nmi_allowed = vmx_nmi_allowed,
7645 .get_nmi_mask = vmx_get_nmi_mask,
7646 .set_nmi_mask = vmx_set_nmi_mask,
7647 .enable_nmi_window = enable_nmi_window,
7648 .enable_irq_window = enable_irq_window,
7649 .update_cr8_intercept = update_cr8_intercept,
7650 .set_virtual_apic_mode = vmx_set_virtual_apic_mode,
7651 .set_apic_access_page_addr = vmx_set_apic_access_page_addr,
7652 .get_enable_apicv = vmx_get_enable_apicv,
7653 .refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
7654 .load_eoi_exitmap = vmx_load_eoi_exitmap,
7655 .apicv_post_state_restore = vmx_apicv_post_state_restore,
7656 .hwapic_irr_update = vmx_hwapic_irr_update,
7657 .hwapic_isr_update = vmx_hwapic_isr_update,
7658 .guest_apic_has_interrupt = vmx_guest_apic_has_interrupt,
7659 .sync_pir_to_irr = vmx_sync_pir_to_irr,
7660 .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
7662 .set_tss_addr = vmx_set_tss_addr,
7663 .set_identity_map_addr = vmx_set_identity_map_addr,
7664 .get_tdp_level = get_ept_level,
7665 .get_mt_mask = vmx_get_mt_mask,
7667 .get_exit_info = vmx_get_exit_info,
7669 .get_lpage_level = vmx_get_lpage_level,
7671 .cpuid_update = vmx_cpuid_update,
7673 .rdtscp_supported = vmx_rdtscp_supported,
7674 .invpcid_supported = vmx_invpcid_supported,
7676 .set_supported_cpuid = vmx_set_supported_cpuid,
7678 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
7680 .read_l1_tsc_offset = vmx_read_l1_tsc_offset,
7681 .write_l1_tsc_offset = vmx_write_l1_tsc_offset,
7683 .set_tdp_cr3 = vmx_set_cr3,
7685 .check_intercept = vmx_check_intercept,
7686 .handle_external_intr = vmx_handle_external_intr,
7687 .mpx_supported = vmx_mpx_supported,
7688 .xsaves_supported = vmx_xsaves_supported,
7689 .umip_emulated = vmx_umip_emulated,
7690 .pt_supported = vmx_pt_supported,
7692 .request_immediate_exit = vmx_request_immediate_exit,
7694 .sched_in = vmx_sched_in,
7696 .slot_enable_log_dirty = vmx_slot_enable_log_dirty,
7697 .slot_disable_log_dirty = vmx_slot_disable_log_dirty,
7698 .flush_log_dirty = vmx_flush_log_dirty,
7699 .enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
7700 .write_log_dirty = vmx_write_pml_buffer,
7702 .pre_block = vmx_pre_block,
7703 .post_block = vmx_post_block,
7705 .pmu_ops = &intel_pmu_ops,
7707 .update_pi_irte = vmx_update_pi_irte,
7709 #ifdef CONFIG_X86_64
7710 .set_hv_timer = vmx_set_hv_timer,
7711 .cancel_hv_timer = vmx_cancel_hv_timer,
7714 .setup_mce = vmx_setup_mce,
7716 .smi_allowed = vmx_smi_allowed,
7717 .pre_enter_smm = vmx_pre_enter_smm,
7718 .pre_leave_smm = vmx_pre_leave_smm,
7719 .enable_smi_window = enable_smi_window,
7721 .check_nested_events = NULL,
7722 .get_nested_state = NULL,
7723 .set_nested_state = NULL,
7724 .get_vmcs12_pages = NULL,
7725 .nested_enable_evmcs = NULL,
7726 .need_emulation_on_page_fault = vmx_need_emulation_on_page_fault,
7729 static void vmx_cleanup_l1d_flush(void)
7731 if (vmx_l1d_flush_pages) {
7732 free_pages((unsigned long)vmx_l1d_flush_pages, L1D_CACHE_ORDER);
7733 vmx_l1d_flush_pages = NULL;
7735 /* Restore state so sysfs ignores VMX */
7736 l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
7739 static void vmx_exit(void)
7741 #ifdef CONFIG_KEXEC_CORE
7742 RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
7748 #if IS_ENABLED(CONFIG_HYPERV)
7749 if (static_branch_unlikely(&enable_evmcs)) {
7751 struct hv_vp_assist_page *vp_ap;
7753 * Reset everything to support using non-enlightened VMCS
7754 * access later (e.g. when we reload the module with
7755 * enlightened_vmcs=0)
7757 for_each_online_cpu(cpu) {
7758 vp_ap = hv_get_vp_assist_page(cpu);
7763 vp_ap->current_nested_vmcs = 0;
7764 vp_ap->enlighten_vmentry = 0;
7767 static_branch_disable(&enable_evmcs);
7770 vmx_cleanup_l1d_flush();
7772 module_exit(vmx_exit);
7774 static int __init vmx_init(void)
7778 #if IS_ENABLED(CONFIG_HYPERV)
7780 * Enlightened VMCS usage should be recommended and the host needs
7781 * to support eVMCS v1 or above. We can also disable eVMCS support
7782 * with module parameter.
7784 if (enlightened_vmcs &&
7785 ms_hyperv.hints & HV_X64_ENLIGHTENED_VMCS_RECOMMENDED &&
7786 (ms_hyperv.nested_features & HV_X64_ENLIGHTENED_VMCS_VERSION) >=
7787 KVM_EVMCS_VERSION) {
7790 /* Check that we have assist pages on all online CPUs */
7791 for_each_online_cpu(cpu) {
7792 if (!hv_get_vp_assist_page(cpu)) {
7793 enlightened_vmcs = false;
7798 if (enlightened_vmcs) {
7799 pr_info("KVM: vmx: using Hyper-V Enlightened VMCS\n");
7800 static_branch_enable(&enable_evmcs);
7803 enlightened_vmcs = false;
7807 r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
7808 __alignof__(struct vcpu_vmx), THIS_MODULE);
7813 * Must be called after kvm_init() so enable_ept is properly set
7814 * up. Hand the parameter mitigation value in which was stored in
7815 * the pre module init parser. If no parameter was given, it will
7816 * contain 'auto' which will be turned into the default 'cond'
7819 if (boot_cpu_has(X86_BUG_L1TF)) {
7820 r = vmx_setup_l1d_flush(vmentry_l1d_flush_param);
7827 #ifdef CONFIG_KEXEC_CORE
7828 rcu_assign_pointer(crash_vmclear_loaded_vmcss,
7829 crash_vmclear_local_loaded_vmcss);
7831 vmx_check_vmcs12_offsets();
7835 module_init(vmx_init);