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/highmem.h>
17 #include <linux/hrtimer.h>
18 #include <linux/kernel.h>
19 #include <linux/kvm_host.h>
20 #include <linux/module.h>
21 #include <linux/moduleparam.h>
22 #include <linux/mod_devicetable.h>
24 #include <linux/objtool.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>
30 #include <linux/entry-kvm.h>
35 #include <asm/cpu_device_id.h>
36 #include <asm/debugreg.h>
38 #include <asm/fpu/api.h>
39 #include <asm/fpu/xstate.h>
40 #include <asm/idtentry.h>
42 #include <asm/irq_remapping.h>
43 #include <asm/kexec.h>
44 #include <asm/perf_event.h>
45 #include <asm/mmu_context.h>
46 #include <asm/mshyperv.h>
47 #include <asm/mwait.h>
48 #include <asm/spec-ctrl.h>
49 #include <asm/virtext.h>
52 #include "capabilities.h"
56 #include "kvm_onhyperv.h"
58 #include "kvm_cache_regs.h"
70 MODULE_AUTHOR("Qumranet");
71 MODULE_LICENSE("GPL");
74 static const struct x86_cpu_id vmx_cpu_id[] = {
75 X86_MATCH_FEATURE(X86_FEATURE_VMX, NULL),
78 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
81 bool __read_mostly enable_vpid = 1;
82 module_param_named(vpid, enable_vpid, bool, 0444);
84 static bool __read_mostly enable_vnmi = 1;
85 module_param_named(vnmi, enable_vnmi, bool, S_IRUGO);
87 bool __read_mostly flexpriority_enabled = 1;
88 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
90 bool __read_mostly enable_ept = 1;
91 module_param_named(ept, enable_ept, bool, S_IRUGO);
93 bool __read_mostly enable_unrestricted_guest = 1;
94 module_param_named(unrestricted_guest,
95 enable_unrestricted_guest, bool, S_IRUGO);
97 bool __read_mostly enable_ept_ad_bits = 1;
98 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
100 static bool __read_mostly emulate_invalid_guest_state = true;
101 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
103 static bool __read_mostly fasteoi = 1;
104 module_param(fasteoi, bool, S_IRUGO);
106 module_param(enable_apicv, bool, S_IRUGO);
109 * If nested=1, nested virtualization is supported, i.e., guests may use
110 * VMX and be a hypervisor for its own guests. If nested=0, guests may not
111 * use VMX instructions.
113 static bool __read_mostly nested = 1;
114 module_param(nested, bool, S_IRUGO);
116 bool __read_mostly enable_pml = 1;
117 module_param_named(pml, enable_pml, bool, S_IRUGO);
119 static bool __read_mostly dump_invalid_vmcs = 0;
120 module_param(dump_invalid_vmcs, bool, 0644);
122 #define MSR_BITMAP_MODE_X2APIC 1
123 #define MSR_BITMAP_MODE_X2APIC_APICV 2
125 #define KVM_VMX_TSC_MULTIPLIER_MAX 0xffffffffffffffffULL
127 /* Guest_tsc -> host_tsc conversion requires 64-bit division. */
128 static int __read_mostly cpu_preemption_timer_multi;
129 static bool __read_mostly enable_preemption_timer = 1;
131 module_param_named(preemption_timer, enable_preemption_timer, bool, S_IRUGO);
134 extern bool __read_mostly allow_smaller_maxphyaddr;
135 module_param(allow_smaller_maxphyaddr, bool, S_IRUGO);
137 #define KVM_VM_CR0_ALWAYS_OFF (X86_CR0_NW | X86_CR0_CD)
138 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR0_NE
139 #define KVM_VM_CR0_ALWAYS_ON \
140 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
142 #define KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR4_VMXE
143 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
144 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
146 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
148 #define MSR_IA32_RTIT_STATUS_MASK (~(RTIT_STATUS_FILTEREN | \
149 RTIT_STATUS_CONTEXTEN | RTIT_STATUS_TRIGGEREN | \
150 RTIT_STATUS_ERROR | RTIT_STATUS_STOPPED | \
151 RTIT_STATUS_BYTECNT))
154 * List of MSRs that can be directly passed to the guest.
155 * In addition to these x2apic and PT MSRs are handled specially.
157 static u32 vmx_possible_passthrough_msrs[MAX_POSSIBLE_PASSTHROUGH_MSRS] = {
168 MSR_IA32_SYSENTER_CS,
169 MSR_IA32_SYSENTER_ESP,
170 MSR_IA32_SYSENTER_EIP,
172 MSR_CORE_C3_RESIDENCY,
173 MSR_CORE_C6_RESIDENCY,
174 MSR_CORE_C7_RESIDENCY,
178 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
179 * ple_gap: upper bound on the amount of time between two successive
180 * executions of PAUSE in a loop. Also indicate if ple enabled.
181 * According to test, this time is usually smaller than 128 cycles.
182 * ple_window: upper bound on the amount of time a guest is allowed to execute
183 * in a PAUSE loop. Tests indicate that most spinlocks are held for
184 * less than 2^12 cycles
185 * Time is measured based on a counter that runs at the same rate as the TSC,
186 * refer SDM volume 3b section 21.6.13 & 22.1.3.
188 static unsigned int ple_gap = KVM_DEFAULT_PLE_GAP;
189 module_param(ple_gap, uint, 0444);
191 static unsigned int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
192 module_param(ple_window, uint, 0444);
194 /* Default doubles per-vcpu window every exit. */
195 static unsigned int ple_window_grow = KVM_DEFAULT_PLE_WINDOW_GROW;
196 module_param(ple_window_grow, uint, 0444);
198 /* Default resets per-vcpu window every exit to ple_window. */
199 static unsigned int ple_window_shrink = KVM_DEFAULT_PLE_WINDOW_SHRINK;
200 module_param(ple_window_shrink, uint, 0444);
202 /* Default is to compute the maximum so we can never overflow. */
203 static unsigned int ple_window_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
204 module_param(ple_window_max, uint, 0444);
206 /* Default is SYSTEM mode, 1 for host-guest mode */
207 int __read_mostly pt_mode = PT_MODE_SYSTEM;
208 module_param(pt_mode, int, S_IRUGO);
210 static DEFINE_STATIC_KEY_FALSE(vmx_l1d_should_flush);
211 static DEFINE_STATIC_KEY_FALSE(vmx_l1d_flush_cond);
212 static DEFINE_MUTEX(vmx_l1d_flush_mutex);
214 /* Storage for pre module init parameter parsing */
215 static enum vmx_l1d_flush_state __read_mostly vmentry_l1d_flush_param = VMENTER_L1D_FLUSH_AUTO;
217 static const struct {
220 } vmentry_l1d_param[] = {
221 [VMENTER_L1D_FLUSH_AUTO] = {"auto", true},
222 [VMENTER_L1D_FLUSH_NEVER] = {"never", true},
223 [VMENTER_L1D_FLUSH_COND] = {"cond", true},
224 [VMENTER_L1D_FLUSH_ALWAYS] = {"always", true},
225 [VMENTER_L1D_FLUSH_EPT_DISABLED] = {"EPT disabled", false},
226 [VMENTER_L1D_FLUSH_NOT_REQUIRED] = {"not required", false},
229 #define L1D_CACHE_ORDER 4
230 static void *vmx_l1d_flush_pages;
232 static int vmx_setup_l1d_flush(enum vmx_l1d_flush_state l1tf)
237 if (!boot_cpu_has_bug(X86_BUG_L1TF)) {
238 l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED;
243 l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_EPT_DISABLED;
247 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES)) {
250 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, msr);
251 if (msr & ARCH_CAP_SKIP_VMENTRY_L1DFLUSH) {
252 l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED;
257 /* If set to auto use the default l1tf mitigation method */
258 if (l1tf == VMENTER_L1D_FLUSH_AUTO) {
259 switch (l1tf_mitigation) {
260 case L1TF_MITIGATION_OFF:
261 l1tf = VMENTER_L1D_FLUSH_NEVER;
263 case L1TF_MITIGATION_FLUSH_NOWARN:
264 case L1TF_MITIGATION_FLUSH:
265 case L1TF_MITIGATION_FLUSH_NOSMT:
266 l1tf = VMENTER_L1D_FLUSH_COND;
268 case L1TF_MITIGATION_FULL:
269 case L1TF_MITIGATION_FULL_FORCE:
270 l1tf = VMENTER_L1D_FLUSH_ALWAYS;
273 } else if (l1tf_mitigation == L1TF_MITIGATION_FULL_FORCE) {
274 l1tf = VMENTER_L1D_FLUSH_ALWAYS;
277 if (l1tf != VMENTER_L1D_FLUSH_NEVER && !vmx_l1d_flush_pages &&
278 !boot_cpu_has(X86_FEATURE_FLUSH_L1D)) {
280 * This allocation for vmx_l1d_flush_pages is not tied to a VM
281 * lifetime and so should not be charged to a memcg.
283 page = alloc_pages(GFP_KERNEL, L1D_CACHE_ORDER);
286 vmx_l1d_flush_pages = page_address(page);
289 * Initialize each page with a different pattern in
290 * order to protect against KSM in the nested
291 * virtualization case.
293 for (i = 0; i < 1u << L1D_CACHE_ORDER; ++i) {
294 memset(vmx_l1d_flush_pages + i * PAGE_SIZE, i + 1,
299 l1tf_vmx_mitigation = l1tf;
301 if (l1tf != VMENTER_L1D_FLUSH_NEVER)
302 static_branch_enable(&vmx_l1d_should_flush);
304 static_branch_disable(&vmx_l1d_should_flush);
306 if (l1tf == VMENTER_L1D_FLUSH_COND)
307 static_branch_enable(&vmx_l1d_flush_cond);
309 static_branch_disable(&vmx_l1d_flush_cond);
313 static int vmentry_l1d_flush_parse(const char *s)
318 for (i = 0; i < ARRAY_SIZE(vmentry_l1d_param); i++) {
319 if (vmentry_l1d_param[i].for_parse &&
320 sysfs_streq(s, vmentry_l1d_param[i].option))
327 static int vmentry_l1d_flush_set(const char *s, const struct kernel_param *kp)
331 l1tf = vmentry_l1d_flush_parse(s);
335 if (!boot_cpu_has(X86_BUG_L1TF))
339 * Has vmx_init() run already? If not then this is the pre init
340 * parameter parsing. In that case just store the value and let
341 * vmx_init() do the proper setup after enable_ept has been
344 if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_AUTO) {
345 vmentry_l1d_flush_param = l1tf;
349 mutex_lock(&vmx_l1d_flush_mutex);
350 ret = vmx_setup_l1d_flush(l1tf);
351 mutex_unlock(&vmx_l1d_flush_mutex);
355 static int vmentry_l1d_flush_get(char *s, const struct kernel_param *kp)
357 if (WARN_ON_ONCE(l1tf_vmx_mitigation >= ARRAY_SIZE(vmentry_l1d_param)))
358 return sprintf(s, "???\n");
360 return sprintf(s, "%s\n", vmentry_l1d_param[l1tf_vmx_mitigation].option);
363 static const struct kernel_param_ops vmentry_l1d_flush_ops = {
364 .set = vmentry_l1d_flush_set,
365 .get = vmentry_l1d_flush_get,
367 module_param_cb(vmentry_l1d_flush, &vmentry_l1d_flush_ops, NULL, 0644);
369 static u32 vmx_segment_access_rights(struct kvm_segment *var);
371 void vmx_vmexit(void);
373 #define vmx_insn_failed(fmt...) \
376 pr_warn_ratelimited(fmt); \
379 asmlinkage void vmread_error(unsigned long field, bool fault)
382 kvm_spurious_fault();
384 vmx_insn_failed("kvm: vmread failed: field=%lx\n", field);
387 noinline void vmwrite_error(unsigned long field, unsigned long value)
389 vmx_insn_failed("kvm: vmwrite failed: field=%lx val=%lx err=%d\n",
390 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
393 noinline void vmclear_error(struct vmcs *vmcs, u64 phys_addr)
395 vmx_insn_failed("kvm: vmclear failed: %p/%llx\n", vmcs, phys_addr);
398 noinline void vmptrld_error(struct vmcs *vmcs, u64 phys_addr)
400 vmx_insn_failed("kvm: vmptrld failed: %p/%llx\n", vmcs, phys_addr);
403 noinline void invvpid_error(unsigned long ext, u16 vpid, gva_t gva)
405 vmx_insn_failed("kvm: invvpid failed: ext=0x%lx vpid=%u gva=0x%lx\n",
409 noinline void invept_error(unsigned long ext, u64 eptp, gpa_t gpa)
411 vmx_insn_failed("kvm: invept failed: ext=0x%lx eptp=%llx gpa=0x%llx\n",
415 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
416 DEFINE_PER_CPU(struct vmcs *, current_vmcs);
418 * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
419 * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
421 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
423 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
424 static DEFINE_SPINLOCK(vmx_vpid_lock);
426 struct vmcs_config vmcs_config;
427 struct vmx_capability vmx_capability;
429 #define VMX_SEGMENT_FIELD(seg) \
430 [VCPU_SREG_##seg] = { \
431 .selector = GUEST_##seg##_SELECTOR, \
432 .base = GUEST_##seg##_BASE, \
433 .limit = GUEST_##seg##_LIMIT, \
434 .ar_bytes = GUEST_##seg##_AR_BYTES, \
437 static const struct kvm_vmx_segment_field {
442 } kvm_vmx_segment_fields[] = {
443 VMX_SEGMENT_FIELD(CS),
444 VMX_SEGMENT_FIELD(DS),
445 VMX_SEGMENT_FIELD(ES),
446 VMX_SEGMENT_FIELD(FS),
447 VMX_SEGMENT_FIELD(GS),
448 VMX_SEGMENT_FIELD(SS),
449 VMX_SEGMENT_FIELD(TR),
450 VMX_SEGMENT_FIELD(LDTR),
453 static inline void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
455 vmx->segment_cache.bitmask = 0;
458 static unsigned long host_idt_base;
460 #if IS_ENABLED(CONFIG_HYPERV)
461 static bool __read_mostly enlightened_vmcs = true;
462 module_param(enlightened_vmcs, bool, 0444);
464 static int hv_enable_direct_tlbflush(struct kvm_vcpu *vcpu)
466 struct hv_enlightened_vmcs *evmcs;
467 struct hv_partition_assist_pg **p_hv_pa_pg =
468 &to_kvm_hv(vcpu->kvm)->hv_pa_pg;
470 * Synthetic VM-Exit is not enabled in current code and so All
471 * evmcs in singe VM shares same assist page.
474 *p_hv_pa_pg = kzalloc(PAGE_SIZE, GFP_KERNEL_ACCOUNT);
479 evmcs = (struct hv_enlightened_vmcs *)to_vmx(vcpu)->loaded_vmcs->vmcs;
481 evmcs->partition_assist_page =
483 evmcs->hv_vm_id = (unsigned long)vcpu->kvm;
484 evmcs->hv_enlightenments_control.nested_flush_hypercall = 1;
489 #endif /* IS_ENABLED(CONFIG_HYPERV) */
492 * Comment's format: document - errata name - stepping - processor name.
494 * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
496 static u32 vmx_preemption_cpu_tfms[] = {
497 /* 323344.pdf - BA86 - D0 - Xeon 7500 Series */
499 /* 323056.pdf - AAX65 - C2 - Xeon L3406 */
500 /* 322814.pdf - AAT59 - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
501 /* 322911.pdf - AAU65 - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
503 /* 322911.pdf - AAU65 - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
505 /* 322373.pdf - AAO95 - B1 - Xeon 3400 Series */
506 /* 322166.pdf - AAN92 - B1 - i7-800 and i5-700 Desktop */
508 * 320767.pdf - AAP86 - B1 -
509 * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
512 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
514 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
516 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
518 /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
519 /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
520 /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
522 /* Xeon E3-1220 V2 */
526 static inline bool cpu_has_broken_vmx_preemption_timer(void)
528 u32 eax = cpuid_eax(0x00000001), i;
530 /* Clear the reserved bits */
531 eax &= ~(0x3U << 14 | 0xfU << 28);
532 for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
533 if (eax == vmx_preemption_cpu_tfms[i])
539 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
541 return flexpriority_enabled && lapic_in_kernel(vcpu);
544 static inline bool report_flexpriority(void)
546 return flexpriority_enabled;
549 static int possible_passthrough_msr_slot(u32 msr)
553 for (i = 0; i < ARRAY_SIZE(vmx_possible_passthrough_msrs); i++)
554 if (vmx_possible_passthrough_msrs[i] == msr)
560 static bool is_valid_passthrough_msr(u32 msr)
565 case 0x800 ... 0x8ff:
566 /* x2APIC MSRs. These are handled in vmx_update_msr_bitmap_x2apic() */
568 case MSR_IA32_RTIT_STATUS:
569 case MSR_IA32_RTIT_OUTPUT_BASE:
570 case MSR_IA32_RTIT_OUTPUT_MASK:
571 case MSR_IA32_RTIT_CR3_MATCH:
572 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
573 /* PT MSRs. These are handled in pt_update_intercept_for_msr() */
576 case MSR_LBR_INFO_0 ... MSR_LBR_INFO_0 + 31:
577 case MSR_LBR_NHM_FROM ... MSR_LBR_NHM_FROM + 31:
578 case MSR_LBR_NHM_TO ... MSR_LBR_NHM_TO + 31:
579 case MSR_LBR_CORE_FROM ... MSR_LBR_CORE_FROM + 8:
580 case MSR_LBR_CORE_TO ... MSR_LBR_CORE_TO + 8:
581 /* LBR MSRs. These are handled in vmx_update_intercept_for_lbr_msrs() */
585 r = possible_passthrough_msr_slot(msr) != -ENOENT;
587 WARN(!r, "Invalid MSR %x, please adapt vmx_possible_passthrough_msrs[]", msr);
592 struct vmx_uret_msr *vmx_find_uret_msr(struct vcpu_vmx *vmx, u32 msr)
596 i = kvm_find_user_return_msr(msr);
598 return &vmx->guest_uret_msrs[i];
602 static int vmx_set_guest_uret_msr(struct vcpu_vmx *vmx,
603 struct vmx_uret_msr *msr, u64 data)
605 unsigned int slot = msr - vmx->guest_uret_msrs;
608 if (msr->load_into_hardware) {
610 ret = kvm_set_user_return_msr(slot, data, msr->mask);
618 #ifdef CONFIG_KEXEC_CORE
619 static void crash_vmclear_local_loaded_vmcss(void)
621 int cpu = raw_smp_processor_id();
622 struct loaded_vmcs *v;
624 list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
625 loaded_vmcss_on_cpu_link)
628 #endif /* CONFIG_KEXEC_CORE */
630 static void __loaded_vmcs_clear(void *arg)
632 struct loaded_vmcs *loaded_vmcs = arg;
633 int cpu = raw_smp_processor_id();
635 if (loaded_vmcs->cpu != cpu)
636 return; /* vcpu migration can race with cpu offline */
637 if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
638 per_cpu(current_vmcs, cpu) = NULL;
640 vmcs_clear(loaded_vmcs->vmcs);
641 if (loaded_vmcs->shadow_vmcs && loaded_vmcs->launched)
642 vmcs_clear(loaded_vmcs->shadow_vmcs);
644 list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
647 * Ensure all writes to loaded_vmcs, including deleting it from its
648 * current percpu list, complete before setting loaded_vmcs->vcpu to
649 * -1, otherwise a different cpu can see vcpu == -1 first and add
650 * loaded_vmcs to its percpu list before it's deleted from this cpu's
651 * list. Pairs with the smp_rmb() in vmx_vcpu_load_vmcs().
655 loaded_vmcs->cpu = -1;
656 loaded_vmcs->launched = 0;
659 void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
661 int cpu = loaded_vmcs->cpu;
664 smp_call_function_single(cpu,
665 __loaded_vmcs_clear, loaded_vmcs, 1);
668 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
672 u32 mask = 1 << (seg * SEG_FIELD_NR + field);
674 if (!kvm_register_is_available(&vmx->vcpu, VCPU_EXREG_SEGMENTS)) {
675 kvm_register_mark_available(&vmx->vcpu, VCPU_EXREG_SEGMENTS);
676 vmx->segment_cache.bitmask = 0;
678 ret = vmx->segment_cache.bitmask & mask;
679 vmx->segment_cache.bitmask |= mask;
683 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
685 u16 *p = &vmx->segment_cache.seg[seg].selector;
687 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
688 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
692 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
694 ulong *p = &vmx->segment_cache.seg[seg].base;
696 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
697 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
701 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
703 u32 *p = &vmx->segment_cache.seg[seg].limit;
705 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
706 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
710 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
712 u32 *p = &vmx->segment_cache.seg[seg].ar;
714 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
715 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
719 void vmx_update_exception_bitmap(struct kvm_vcpu *vcpu)
723 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
724 (1u << DB_VECTOR) | (1u << AC_VECTOR);
726 * Guest access to VMware backdoor ports could legitimately
727 * trigger #GP because of TSS I/O permission bitmap.
728 * We intercept those #GP and allow access to them anyway
731 if (enable_vmware_backdoor)
732 eb |= (1u << GP_VECTOR);
733 if ((vcpu->guest_debug &
734 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
735 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
736 eb |= 1u << BP_VECTOR;
737 if (to_vmx(vcpu)->rmode.vm86_active)
739 if (!vmx_need_pf_intercept(vcpu))
740 eb &= ~(1u << PF_VECTOR);
742 /* When we are running a nested L2 guest and L1 specified for it a
743 * certain exception bitmap, we must trap the same exceptions and pass
744 * them to L1. When running L2, we will only handle the exceptions
745 * specified above if L1 did not want them.
747 if (is_guest_mode(vcpu))
748 eb |= get_vmcs12(vcpu)->exception_bitmap;
750 int mask = 0, match = 0;
752 if (enable_ept && (eb & (1u << PF_VECTOR))) {
754 * If EPT is enabled, #PF is currently only intercepted
755 * if MAXPHYADDR is smaller on the guest than on the
756 * host. In that case we only care about present,
757 * non-reserved faults. For vmcs02, however, PFEC_MASK
758 * and PFEC_MATCH are set in prepare_vmcs02_rare.
760 mask = PFERR_PRESENT_MASK | PFERR_RSVD_MASK;
761 match = PFERR_PRESENT_MASK;
763 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, mask);
764 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, match);
768 * Disabling xfd interception indicates that dynamic xfeatures
769 * might be used in the guest. Always trap #NM in this case
770 * to save guest xfd_err timely.
772 if (vcpu->arch.xfd_no_write_intercept)
773 eb |= (1u << NM_VECTOR);
775 vmcs_write32(EXCEPTION_BITMAP, eb);
779 * Check if MSR is intercepted for currently loaded MSR bitmap.
781 static bool msr_write_intercepted(struct vcpu_vmx *vmx, u32 msr)
783 if (!(exec_controls_get(vmx) & CPU_BASED_USE_MSR_BITMAPS))
786 return vmx_test_msr_bitmap_write(vmx->loaded_vmcs->msr_bitmap,
790 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
791 unsigned long entry, unsigned long exit)
793 vm_entry_controls_clearbit(vmx, entry);
794 vm_exit_controls_clearbit(vmx, exit);
797 int vmx_find_loadstore_msr_slot(struct vmx_msrs *m, u32 msr)
801 for (i = 0; i < m->nr; ++i) {
802 if (m->val[i].index == msr)
808 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
811 struct msr_autoload *m = &vmx->msr_autoload;
815 if (cpu_has_load_ia32_efer()) {
816 clear_atomic_switch_msr_special(vmx,
817 VM_ENTRY_LOAD_IA32_EFER,
818 VM_EXIT_LOAD_IA32_EFER);
822 case MSR_CORE_PERF_GLOBAL_CTRL:
823 if (cpu_has_load_perf_global_ctrl()) {
824 clear_atomic_switch_msr_special(vmx,
825 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
826 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
831 i = vmx_find_loadstore_msr_slot(&m->guest, msr);
835 m->guest.val[i] = m->guest.val[m->guest.nr];
836 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
839 i = vmx_find_loadstore_msr_slot(&m->host, msr);
844 m->host.val[i] = m->host.val[m->host.nr];
845 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
848 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
849 unsigned long entry, unsigned long exit,
850 unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
851 u64 guest_val, u64 host_val)
853 vmcs_write64(guest_val_vmcs, guest_val);
854 if (host_val_vmcs != HOST_IA32_EFER)
855 vmcs_write64(host_val_vmcs, host_val);
856 vm_entry_controls_setbit(vmx, entry);
857 vm_exit_controls_setbit(vmx, exit);
860 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
861 u64 guest_val, u64 host_val, bool entry_only)
864 struct msr_autoload *m = &vmx->msr_autoload;
868 if (cpu_has_load_ia32_efer()) {
869 add_atomic_switch_msr_special(vmx,
870 VM_ENTRY_LOAD_IA32_EFER,
871 VM_EXIT_LOAD_IA32_EFER,
874 guest_val, host_val);
878 case MSR_CORE_PERF_GLOBAL_CTRL:
879 if (cpu_has_load_perf_global_ctrl()) {
880 add_atomic_switch_msr_special(vmx,
881 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
882 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
883 GUEST_IA32_PERF_GLOBAL_CTRL,
884 HOST_IA32_PERF_GLOBAL_CTRL,
885 guest_val, host_val);
889 case MSR_IA32_PEBS_ENABLE:
890 /* PEBS needs a quiescent period after being disabled (to write
891 * a record). Disabling PEBS through VMX MSR swapping doesn't
892 * provide that period, so a CPU could write host's record into
895 wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
898 i = vmx_find_loadstore_msr_slot(&m->guest, msr);
900 j = vmx_find_loadstore_msr_slot(&m->host, msr);
902 if ((i < 0 && m->guest.nr == MAX_NR_LOADSTORE_MSRS) ||
903 (j < 0 && m->host.nr == MAX_NR_LOADSTORE_MSRS)) {
904 printk_once(KERN_WARNING "Not enough msr switch entries. "
905 "Can't add msr %x\n", msr);
910 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
912 m->guest.val[i].index = msr;
913 m->guest.val[i].value = guest_val;
920 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
922 m->host.val[j].index = msr;
923 m->host.val[j].value = host_val;
926 static bool update_transition_efer(struct vcpu_vmx *vmx)
928 u64 guest_efer = vmx->vcpu.arch.efer;
932 /* Shadow paging assumes NX to be available. */
934 guest_efer |= EFER_NX;
937 * LMA and LME handled by hardware; SCE meaningless outside long mode.
939 ignore_bits |= EFER_SCE;
941 ignore_bits |= EFER_LMA | EFER_LME;
942 /* SCE is meaningful only in long mode on Intel */
943 if (guest_efer & EFER_LMA)
944 ignore_bits &= ~(u64)EFER_SCE;
948 * On EPT, we can't emulate NX, so we must switch EFER atomically.
949 * On CPUs that support "load IA32_EFER", always switch EFER
950 * atomically, since it's faster than switching it manually.
952 if (cpu_has_load_ia32_efer() ||
953 (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
954 if (!(guest_efer & EFER_LMA))
955 guest_efer &= ~EFER_LME;
956 if (guest_efer != host_efer)
957 add_atomic_switch_msr(vmx, MSR_EFER,
958 guest_efer, host_efer, false);
960 clear_atomic_switch_msr(vmx, MSR_EFER);
964 i = kvm_find_user_return_msr(MSR_EFER);
968 clear_atomic_switch_msr(vmx, MSR_EFER);
970 guest_efer &= ~ignore_bits;
971 guest_efer |= host_efer & ignore_bits;
973 vmx->guest_uret_msrs[i].data = guest_efer;
974 vmx->guest_uret_msrs[i].mask = ~ignore_bits;
981 * On 32-bit kernels, VM exits still load the FS and GS bases from the
982 * VMCS rather than the segment table. KVM uses this helper to figure
983 * out the current bases to poke them into the VMCS before entry.
985 static unsigned long segment_base(u16 selector)
987 struct desc_struct *table;
990 if (!(selector & ~SEGMENT_RPL_MASK))
993 table = get_current_gdt_ro();
995 if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
996 u16 ldt_selector = kvm_read_ldt();
998 if (!(ldt_selector & ~SEGMENT_RPL_MASK))
1001 table = (struct desc_struct *)segment_base(ldt_selector);
1003 v = get_desc_base(&table[selector >> 3]);
1008 static inline bool pt_can_write_msr(struct vcpu_vmx *vmx)
1010 return vmx_pt_mode_is_host_guest() &&
1011 !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
1014 static inline bool pt_output_base_valid(struct kvm_vcpu *vcpu, u64 base)
1016 /* The base must be 128-byte aligned and a legal physical address. */
1017 return kvm_vcpu_is_legal_aligned_gpa(vcpu, base, 128);
1020 static inline void pt_load_msr(struct pt_ctx *ctx, u32 addr_range)
1024 wrmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1025 wrmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1026 wrmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1027 wrmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1028 for (i = 0; i < addr_range; i++) {
1029 wrmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1030 wrmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1034 static inline void pt_save_msr(struct pt_ctx *ctx, u32 addr_range)
1038 rdmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1039 rdmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1040 rdmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1041 rdmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1042 for (i = 0; i < addr_range; i++) {
1043 rdmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1044 rdmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1048 static void pt_guest_enter(struct vcpu_vmx *vmx)
1050 if (vmx_pt_mode_is_system())
1054 * GUEST_IA32_RTIT_CTL is already set in the VMCS.
1055 * Save host state before VM entry.
1057 rdmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1058 if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1059 wrmsrl(MSR_IA32_RTIT_CTL, 0);
1060 pt_save_msr(&vmx->pt_desc.host, vmx->pt_desc.num_address_ranges);
1061 pt_load_msr(&vmx->pt_desc.guest, vmx->pt_desc.num_address_ranges);
1065 static void pt_guest_exit(struct vcpu_vmx *vmx)
1067 if (vmx_pt_mode_is_system())
1070 if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1071 pt_save_msr(&vmx->pt_desc.guest, vmx->pt_desc.num_address_ranges);
1072 pt_load_msr(&vmx->pt_desc.host, vmx->pt_desc.num_address_ranges);
1076 * KVM requires VM_EXIT_CLEAR_IA32_RTIT_CTL to expose PT to the guest,
1077 * i.e. RTIT_CTL is always cleared on VM-Exit. Restore it if necessary.
1079 if (vmx->pt_desc.host.ctl)
1080 wrmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1083 void vmx_set_vmcs_host_state(struct vmcs_host_state *host, unsigned long cr3,
1084 u16 fs_sel, u16 gs_sel,
1085 unsigned long fs_base, unsigned long gs_base)
1087 if (unlikely(cr3 != host->cr3)) {
1088 vmcs_writel(HOST_CR3, cr3);
1091 if (unlikely(fs_sel != host->fs_sel)) {
1093 vmcs_write16(HOST_FS_SELECTOR, fs_sel);
1095 vmcs_write16(HOST_FS_SELECTOR, 0);
1096 host->fs_sel = fs_sel;
1098 if (unlikely(gs_sel != host->gs_sel)) {
1100 vmcs_write16(HOST_GS_SELECTOR, gs_sel);
1102 vmcs_write16(HOST_GS_SELECTOR, 0);
1103 host->gs_sel = gs_sel;
1105 if (unlikely(fs_base != host->fs_base)) {
1106 vmcs_writel(HOST_FS_BASE, fs_base);
1107 host->fs_base = fs_base;
1109 if (unlikely(gs_base != host->gs_base)) {
1110 vmcs_writel(HOST_GS_BASE, gs_base);
1111 host->gs_base = gs_base;
1115 void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu)
1117 struct vcpu_vmx *vmx = to_vmx(vcpu);
1118 struct vmcs_host_state *host_state;
1119 #ifdef CONFIG_X86_64
1120 int cpu = raw_smp_processor_id();
1122 unsigned long fs_base, gs_base;
1126 vmx->req_immediate_exit = false;
1129 * Note that guest MSRs to be saved/restored can also be changed
1130 * when guest state is loaded. This happens when guest transitions
1131 * to/from long-mode by setting MSR_EFER.LMA.
1133 if (!vmx->guest_uret_msrs_loaded) {
1134 vmx->guest_uret_msrs_loaded = true;
1135 for (i = 0; i < kvm_nr_uret_msrs; ++i) {
1136 if (!vmx->guest_uret_msrs[i].load_into_hardware)
1139 kvm_set_user_return_msr(i,
1140 vmx->guest_uret_msrs[i].data,
1141 vmx->guest_uret_msrs[i].mask);
1145 if (vmx->nested.need_vmcs12_to_shadow_sync)
1146 nested_sync_vmcs12_to_shadow(vcpu);
1148 if (vmx->guest_state_loaded)
1151 host_state = &vmx->loaded_vmcs->host_state;
1154 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
1155 * allow segment selectors with cpl > 0 or ti == 1.
1157 host_state->ldt_sel = kvm_read_ldt();
1159 #ifdef CONFIG_X86_64
1160 savesegment(ds, host_state->ds_sel);
1161 savesegment(es, host_state->es_sel);
1163 gs_base = cpu_kernelmode_gs_base(cpu);
1164 if (likely(is_64bit_mm(current->mm))) {
1165 current_save_fsgs();
1166 fs_sel = current->thread.fsindex;
1167 gs_sel = current->thread.gsindex;
1168 fs_base = current->thread.fsbase;
1169 vmx->msr_host_kernel_gs_base = current->thread.gsbase;
1171 savesegment(fs, fs_sel);
1172 savesegment(gs, gs_sel);
1173 fs_base = read_msr(MSR_FS_BASE);
1174 vmx->msr_host_kernel_gs_base = read_msr(MSR_KERNEL_GS_BASE);
1177 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1179 savesegment(fs, fs_sel);
1180 savesegment(gs, gs_sel);
1181 fs_base = segment_base(fs_sel);
1182 gs_base = segment_base(gs_sel);
1185 vmx_set_vmcs_host_state(host_state, __get_current_cr3_fast(),
1186 fs_sel, gs_sel, fs_base, gs_base);
1188 vmx->guest_state_loaded = true;
1191 static void vmx_prepare_switch_to_host(struct vcpu_vmx *vmx)
1193 struct vmcs_host_state *host_state;
1195 if (!vmx->guest_state_loaded)
1198 host_state = &vmx->loaded_vmcs->host_state;
1200 ++vmx->vcpu.stat.host_state_reload;
1202 #ifdef CONFIG_X86_64
1203 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1205 if (host_state->ldt_sel || (host_state->gs_sel & 7)) {
1206 kvm_load_ldt(host_state->ldt_sel);
1207 #ifdef CONFIG_X86_64
1208 load_gs_index(host_state->gs_sel);
1210 loadsegment(gs, host_state->gs_sel);
1213 if (host_state->fs_sel & 7)
1214 loadsegment(fs, host_state->fs_sel);
1215 #ifdef CONFIG_X86_64
1216 if (unlikely(host_state->ds_sel | host_state->es_sel)) {
1217 loadsegment(ds, host_state->ds_sel);
1218 loadsegment(es, host_state->es_sel);
1221 invalidate_tss_limit();
1222 #ifdef CONFIG_X86_64
1223 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1225 load_fixmap_gdt(raw_smp_processor_id());
1226 vmx->guest_state_loaded = false;
1227 vmx->guest_uret_msrs_loaded = false;
1230 #ifdef CONFIG_X86_64
1231 static u64 vmx_read_guest_kernel_gs_base(struct vcpu_vmx *vmx)
1234 if (vmx->guest_state_loaded)
1235 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1237 return vmx->msr_guest_kernel_gs_base;
1240 static void vmx_write_guest_kernel_gs_base(struct vcpu_vmx *vmx, u64 data)
1243 if (vmx->guest_state_loaded)
1244 wrmsrl(MSR_KERNEL_GS_BASE, data);
1246 vmx->msr_guest_kernel_gs_base = data;
1250 void vmx_vcpu_load_vmcs(struct kvm_vcpu *vcpu, int cpu,
1251 struct loaded_vmcs *buddy)
1253 struct vcpu_vmx *vmx = to_vmx(vcpu);
1254 bool already_loaded = vmx->loaded_vmcs->cpu == cpu;
1257 if (!already_loaded) {
1258 loaded_vmcs_clear(vmx->loaded_vmcs);
1259 local_irq_disable();
1262 * Ensure loaded_vmcs->cpu is read before adding loaded_vmcs to
1263 * this cpu's percpu list, otherwise it may not yet be deleted
1264 * from its previous cpu's percpu list. Pairs with the
1265 * smb_wmb() in __loaded_vmcs_clear().
1269 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1270 &per_cpu(loaded_vmcss_on_cpu, cpu));
1274 prev = per_cpu(current_vmcs, cpu);
1275 if (prev != vmx->loaded_vmcs->vmcs) {
1276 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1277 vmcs_load(vmx->loaded_vmcs->vmcs);
1280 * No indirect branch prediction barrier needed when switching
1281 * the active VMCS within a guest, e.g. on nested VM-Enter.
1282 * The L1 VMM can protect itself with retpolines, IBPB or IBRS.
1284 if (!buddy || WARN_ON_ONCE(buddy->vmcs != prev))
1285 indirect_branch_prediction_barrier();
1288 if (!already_loaded) {
1289 void *gdt = get_current_gdt_ro();
1292 * Flush all EPTP/VPID contexts, the new pCPU may have stale
1293 * TLB entries from its previous association with the vCPU.
1295 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1298 * Linux uses per-cpu TSS and GDT, so set these when switching
1299 * processors. See 22.2.4.
1301 vmcs_writel(HOST_TR_BASE,
1302 (unsigned long)&get_cpu_entry_area(cpu)->tss.x86_tss);
1303 vmcs_writel(HOST_GDTR_BASE, (unsigned long)gdt); /* 22.2.4 */
1305 if (IS_ENABLED(CONFIG_IA32_EMULATION) || IS_ENABLED(CONFIG_X86_32)) {
1307 vmcs_writel(HOST_IA32_SYSENTER_ESP,
1308 (unsigned long)(cpu_entry_stack(cpu) + 1));
1311 vmx->loaded_vmcs->cpu = cpu;
1316 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1317 * vcpu mutex is already taken.
1319 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1321 struct vcpu_vmx *vmx = to_vmx(vcpu);
1323 vmx_vcpu_load_vmcs(vcpu, cpu, NULL);
1325 vmx_vcpu_pi_load(vcpu, cpu);
1327 vmx->host_debugctlmsr = get_debugctlmsr();
1330 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1332 vmx_vcpu_pi_put(vcpu);
1334 vmx_prepare_switch_to_host(to_vmx(vcpu));
1337 bool vmx_emulation_required(struct kvm_vcpu *vcpu)
1339 return emulate_invalid_guest_state && !vmx_guest_state_valid(vcpu);
1342 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1344 struct vcpu_vmx *vmx = to_vmx(vcpu);
1345 unsigned long rflags, save_rflags;
1347 if (!kvm_register_is_available(vcpu, VCPU_EXREG_RFLAGS)) {
1348 kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1349 rflags = vmcs_readl(GUEST_RFLAGS);
1350 if (vmx->rmode.vm86_active) {
1351 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1352 save_rflags = vmx->rmode.save_rflags;
1353 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1355 vmx->rflags = rflags;
1360 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1362 struct vcpu_vmx *vmx = to_vmx(vcpu);
1363 unsigned long old_rflags;
1365 if (is_unrestricted_guest(vcpu)) {
1366 kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1367 vmx->rflags = rflags;
1368 vmcs_writel(GUEST_RFLAGS, rflags);
1372 old_rflags = vmx_get_rflags(vcpu);
1373 vmx->rflags = rflags;
1374 if (vmx->rmode.vm86_active) {
1375 vmx->rmode.save_rflags = rflags;
1376 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1378 vmcs_writel(GUEST_RFLAGS, rflags);
1380 if ((old_rflags ^ vmx->rflags) & X86_EFLAGS_VM)
1381 vmx->emulation_required = vmx_emulation_required(vcpu);
1384 static bool vmx_get_if_flag(struct kvm_vcpu *vcpu)
1386 return vmx_get_rflags(vcpu) & X86_EFLAGS_IF;
1389 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
1391 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1394 if (interruptibility & GUEST_INTR_STATE_STI)
1395 ret |= KVM_X86_SHADOW_INT_STI;
1396 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1397 ret |= KVM_X86_SHADOW_INT_MOV_SS;
1402 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1404 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1405 u32 interruptibility = interruptibility_old;
1407 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1409 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1410 interruptibility |= GUEST_INTR_STATE_MOV_SS;
1411 else if (mask & KVM_X86_SHADOW_INT_STI)
1412 interruptibility |= GUEST_INTR_STATE_STI;
1414 if ((interruptibility != interruptibility_old))
1415 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1418 static int vmx_rtit_ctl_check(struct kvm_vcpu *vcpu, u64 data)
1420 struct vcpu_vmx *vmx = to_vmx(vcpu);
1421 unsigned long value;
1424 * Any MSR write that attempts to change bits marked reserved will
1427 if (data & vmx->pt_desc.ctl_bitmask)
1431 * Any attempt to modify IA32_RTIT_CTL while TraceEn is set will
1432 * result in a #GP unless the same write also clears TraceEn.
1434 if ((vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) &&
1435 ((vmx->pt_desc.guest.ctl ^ data) & ~RTIT_CTL_TRACEEN))
1439 * WRMSR to IA32_RTIT_CTL that sets TraceEn but clears this bit
1440 * and FabricEn would cause #GP, if
1441 * CPUID.(EAX=14H, ECX=0):ECX.SNGLRGNOUT[bit 2] = 0
1443 if ((data & RTIT_CTL_TRACEEN) && !(data & RTIT_CTL_TOPA) &&
1444 !(data & RTIT_CTL_FABRIC_EN) &&
1445 !intel_pt_validate_cap(vmx->pt_desc.caps,
1446 PT_CAP_single_range_output))
1450 * MTCFreq, CycThresh and PSBFreq encodings check, any MSR write that
1451 * utilize encodings marked reserved will cause a #GP fault.
1453 value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc_periods);
1454 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc) &&
1455 !test_bit((data & RTIT_CTL_MTC_RANGE) >>
1456 RTIT_CTL_MTC_RANGE_OFFSET, &value))
1458 value = intel_pt_validate_cap(vmx->pt_desc.caps,
1459 PT_CAP_cycle_thresholds);
1460 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1461 !test_bit((data & RTIT_CTL_CYC_THRESH) >>
1462 RTIT_CTL_CYC_THRESH_OFFSET, &value))
1464 value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_periods);
1465 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1466 !test_bit((data & RTIT_CTL_PSB_FREQ) >>
1467 RTIT_CTL_PSB_FREQ_OFFSET, &value))
1471 * If ADDRx_CFG is reserved or the encodings is >2 will
1472 * cause a #GP fault.
1474 value = (data & RTIT_CTL_ADDR0) >> RTIT_CTL_ADDR0_OFFSET;
1475 if ((value && (vmx->pt_desc.num_address_ranges < 1)) || (value > 2))
1477 value = (data & RTIT_CTL_ADDR1) >> RTIT_CTL_ADDR1_OFFSET;
1478 if ((value && (vmx->pt_desc.num_address_ranges < 2)) || (value > 2))
1480 value = (data & RTIT_CTL_ADDR2) >> RTIT_CTL_ADDR2_OFFSET;
1481 if ((value && (vmx->pt_desc.num_address_ranges < 3)) || (value > 2))
1483 value = (data & RTIT_CTL_ADDR3) >> RTIT_CTL_ADDR3_OFFSET;
1484 if ((value && (vmx->pt_desc.num_address_ranges < 4)) || (value > 2))
1490 static bool vmx_can_emulate_instruction(struct kvm_vcpu *vcpu, int emul_type,
1491 void *insn, int insn_len)
1494 * Emulation of instructions in SGX enclaves is impossible as RIP does
1495 * not point at the failing instruction, and even if it did, the code
1496 * stream is inaccessible. Inject #UD instead of exiting to userspace
1497 * so that guest userspace can't DoS the guest simply by triggering
1498 * emulation (enclaves are CPL3 only).
1500 if (to_vmx(vcpu)->exit_reason.enclave_mode) {
1501 kvm_queue_exception(vcpu, UD_VECTOR);
1507 static int skip_emulated_instruction(struct kvm_vcpu *vcpu)
1509 union vmx_exit_reason exit_reason = to_vmx(vcpu)->exit_reason;
1510 unsigned long rip, orig_rip;
1514 * Using VMCS.VM_EXIT_INSTRUCTION_LEN on EPT misconfig depends on
1515 * undefined behavior: Intel's SDM doesn't mandate the VMCS field be
1516 * set when EPT misconfig occurs. In practice, real hardware updates
1517 * VM_EXIT_INSTRUCTION_LEN on EPT misconfig, but other hypervisors
1518 * (namely Hyper-V) don't set it due to it being undefined behavior,
1519 * i.e. we end up advancing IP with some random value.
1521 if (!static_cpu_has(X86_FEATURE_HYPERVISOR) ||
1522 exit_reason.basic != EXIT_REASON_EPT_MISCONFIG) {
1523 instr_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1526 * Emulating an enclave's instructions isn't supported as KVM
1527 * cannot access the enclave's memory or its true RIP, e.g. the
1528 * vmcs.GUEST_RIP points at the exit point of the enclave, not
1529 * the RIP that actually triggered the VM-Exit. But, because
1530 * most instructions that cause VM-Exit will #UD in an enclave,
1531 * most instruction-based VM-Exits simply do not occur.
1533 * There are a few exceptions, notably the debug instructions
1534 * INT1ICEBRK and INT3, as they are allowed in debug enclaves
1535 * and generate #DB/#BP as expected, which KVM might intercept.
1536 * But again, the CPU does the dirty work and saves an instr
1537 * length of zero so VMMs don't shoot themselves in the foot.
1538 * WARN if KVM tries to skip a non-zero length instruction on
1539 * a VM-Exit from an enclave.
1544 WARN(exit_reason.enclave_mode,
1545 "KVM: skipping instruction after SGX enclave VM-Exit");
1547 orig_rip = kvm_rip_read(vcpu);
1548 rip = orig_rip + instr_len;
1549 #ifdef CONFIG_X86_64
1551 * We need to mask out the high 32 bits of RIP if not in 64-bit
1552 * mode, but just finding out that we are in 64-bit mode is
1553 * quite expensive. Only do it if there was a carry.
1555 if (unlikely(((rip ^ orig_rip) >> 31) == 3) && !is_64_bit_mode(vcpu))
1558 kvm_rip_write(vcpu, rip);
1560 if (!kvm_emulate_instruction(vcpu, EMULTYPE_SKIP))
1565 /* skipping an emulated instruction also counts */
1566 vmx_set_interrupt_shadow(vcpu, 0);
1572 * Recognizes a pending MTF VM-exit and records the nested state for later
1575 static void vmx_update_emulated_instruction(struct kvm_vcpu *vcpu)
1577 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1578 struct vcpu_vmx *vmx = to_vmx(vcpu);
1580 if (!is_guest_mode(vcpu))
1584 * Per the SDM, MTF takes priority over debug-trap exceptions besides
1585 * T-bit traps. As instruction emulation is completed (i.e. at the
1586 * instruction boundary), any #DB exception pending delivery must be a
1587 * debug-trap. Record the pending MTF state to be delivered in
1588 * vmx_check_nested_events().
1590 if (nested_cpu_has_mtf(vmcs12) &&
1591 (!vcpu->arch.exception.pending ||
1592 vcpu->arch.exception.nr == DB_VECTOR))
1593 vmx->nested.mtf_pending = true;
1595 vmx->nested.mtf_pending = false;
1598 static int vmx_skip_emulated_instruction(struct kvm_vcpu *vcpu)
1600 vmx_update_emulated_instruction(vcpu);
1601 return skip_emulated_instruction(vcpu);
1604 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1607 * Ensure that we clear the HLT state in the VMCS. We don't need to
1608 * explicitly skip the instruction because if the HLT state is set,
1609 * then the instruction is already executing and RIP has already been
1612 if (kvm_hlt_in_guest(vcpu->kvm) &&
1613 vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1614 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1617 static void vmx_queue_exception(struct kvm_vcpu *vcpu)
1619 struct vcpu_vmx *vmx = to_vmx(vcpu);
1620 unsigned nr = vcpu->arch.exception.nr;
1621 bool has_error_code = vcpu->arch.exception.has_error_code;
1622 u32 error_code = vcpu->arch.exception.error_code;
1623 u32 intr_info = nr | INTR_INFO_VALID_MASK;
1625 kvm_deliver_exception_payload(vcpu);
1627 if (has_error_code) {
1628 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1629 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1632 if (vmx->rmode.vm86_active) {
1634 if (kvm_exception_is_soft(nr))
1635 inc_eip = vcpu->arch.event_exit_inst_len;
1636 kvm_inject_realmode_interrupt(vcpu, nr, inc_eip);
1640 WARN_ON_ONCE(vmx->emulation_required);
1642 if (kvm_exception_is_soft(nr)) {
1643 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1644 vmx->vcpu.arch.event_exit_inst_len);
1645 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1647 intr_info |= INTR_TYPE_HARD_EXCEPTION;
1649 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1651 vmx_clear_hlt(vcpu);
1654 static void vmx_setup_uret_msr(struct vcpu_vmx *vmx, unsigned int msr,
1655 bool load_into_hardware)
1657 struct vmx_uret_msr *uret_msr;
1659 uret_msr = vmx_find_uret_msr(vmx, msr);
1663 uret_msr->load_into_hardware = load_into_hardware;
1667 * Configuring user return MSRs to automatically save, load, and restore MSRs
1668 * that need to be shoved into hardware when running the guest. Note, omitting
1669 * an MSR here does _NOT_ mean it's not emulated, only that it will not be
1670 * loaded into hardware when running the guest.
1672 static void vmx_setup_uret_msrs(struct vcpu_vmx *vmx)
1674 #ifdef CONFIG_X86_64
1675 bool load_syscall_msrs;
1678 * The SYSCALL MSRs are only needed on long mode guests, and only
1679 * when EFER.SCE is set.
1681 load_syscall_msrs = is_long_mode(&vmx->vcpu) &&
1682 (vmx->vcpu.arch.efer & EFER_SCE);
1684 vmx_setup_uret_msr(vmx, MSR_STAR, load_syscall_msrs);
1685 vmx_setup_uret_msr(vmx, MSR_LSTAR, load_syscall_msrs);
1686 vmx_setup_uret_msr(vmx, MSR_SYSCALL_MASK, load_syscall_msrs);
1688 vmx_setup_uret_msr(vmx, MSR_EFER, update_transition_efer(vmx));
1690 vmx_setup_uret_msr(vmx, MSR_TSC_AUX,
1691 guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDTSCP) ||
1692 guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDPID));
1695 * hle=0, rtm=0, tsx_ctrl=1 can be found with some combinations of new
1696 * kernel and old userspace. If those guests run on a tsx=off host, do
1697 * allow guests to use TSX_CTRL, but don't change the value in hardware
1698 * so that TSX remains always disabled.
1700 vmx_setup_uret_msr(vmx, MSR_IA32_TSX_CTRL, boot_cpu_has(X86_FEATURE_RTM));
1703 * The set of MSRs to load may have changed, reload MSRs before the
1706 vmx->guest_uret_msrs_loaded = false;
1709 u64 vmx_get_l2_tsc_offset(struct kvm_vcpu *vcpu)
1711 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1713 if (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETTING))
1714 return vmcs12->tsc_offset;
1719 u64 vmx_get_l2_tsc_multiplier(struct kvm_vcpu *vcpu)
1721 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1723 if (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETTING) &&
1724 nested_cpu_has2(vmcs12, SECONDARY_EXEC_TSC_SCALING))
1725 return vmcs12->tsc_multiplier;
1727 return kvm_default_tsc_scaling_ratio;
1730 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1732 vmcs_write64(TSC_OFFSET, offset);
1735 static void vmx_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 multiplier)
1737 vmcs_write64(TSC_MULTIPLIER, multiplier);
1741 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
1742 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
1743 * all guests if the "nested" module option is off, and can also be disabled
1744 * for a single guest by disabling its VMX cpuid bit.
1746 bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
1748 return nested && guest_cpuid_has(vcpu, X86_FEATURE_VMX);
1751 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
1754 uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
1756 return !(val & ~valid_bits);
1759 static int vmx_get_msr_feature(struct kvm_msr_entry *msr)
1761 switch (msr->index) {
1762 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1765 return vmx_get_vmx_msr(&vmcs_config.nested, msr->index, &msr->data);
1766 case MSR_IA32_PERF_CAPABILITIES:
1767 msr->data = vmx_get_perf_capabilities();
1770 return KVM_MSR_RET_INVALID;
1775 * Reads an msr value (of 'msr_info->index') into 'msr_info->data'.
1776 * Returns 0 on success, non-0 otherwise.
1777 * Assumes vcpu_load() was already called.
1779 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1781 struct vcpu_vmx *vmx = to_vmx(vcpu);
1782 struct vmx_uret_msr *msr;
1785 switch (msr_info->index) {
1786 #ifdef CONFIG_X86_64
1788 msr_info->data = vmcs_readl(GUEST_FS_BASE);
1791 msr_info->data = vmcs_readl(GUEST_GS_BASE);
1793 case MSR_KERNEL_GS_BASE:
1794 msr_info->data = vmx_read_guest_kernel_gs_base(vmx);
1798 return kvm_get_msr_common(vcpu, msr_info);
1799 case MSR_IA32_TSX_CTRL:
1800 if (!msr_info->host_initiated &&
1801 !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
1804 case MSR_IA32_UMWAIT_CONTROL:
1805 if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
1808 msr_info->data = vmx->msr_ia32_umwait_control;
1810 case MSR_IA32_SPEC_CTRL:
1811 if (!msr_info->host_initiated &&
1812 !guest_has_spec_ctrl_msr(vcpu))
1815 msr_info->data = to_vmx(vcpu)->spec_ctrl;
1817 case MSR_IA32_SYSENTER_CS:
1818 msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
1820 case MSR_IA32_SYSENTER_EIP:
1821 msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
1823 case MSR_IA32_SYSENTER_ESP:
1824 msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
1826 case MSR_IA32_BNDCFGS:
1827 if (!kvm_mpx_supported() ||
1828 (!msr_info->host_initiated &&
1829 !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1831 msr_info->data = vmcs_read64(GUEST_BNDCFGS);
1833 case MSR_IA32_MCG_EXT_CTL:
1834 if (!msr_info->host_initiated &&
1835 !(vmx->msr_ia32_feature_control &
1836 FEAT_CTL_LMCE_ENABLED))
1838 msr_info->data = vcpu->arch.mcg_ext_ctl;
1840 case MSR_IA32_FEAT_CTL:
1841 msr_info->data = vmx->msr_ia32_feature_control;
1843 case MSR_IA32_SGXLEPUBKEYHASH0 ... MSR_IA32_SGXLEPUBKEYHASH3:
1844 if (!msr_info->host_initiated &&
1845 !guest_cpuid_has(vcpu, X86_FEATURE_SGX_LC))
1847 msr_info->data = to_vmx(vcpu)->msr_ia32_sgxlepubkeyhash
1848 [msr_info->index - MSR_IA32_SGXLEPUBKEYHASH0];
1850 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1851 if (!nested_vmx_allowed(vcpu))
1853 if (vmx_get_vmx_msr(&vmx->nested.msrs, msr_info->index,
1857 * Enlightened VMCS v1 doesn't have certain VMCS fields but
1858 * instead of just ignoring the features, different Hyper-V
1859 * versions are either trying to use them and fail or do some
1860 * sanity checking and refuse to boot. Filter all unsupported
1863 if (!msr_info->host_initiated &&
1864 vmx->nested.enlightened_vmcs_enabled)
1865 nested_evmcs_filter_control_msr(msr_info->index,
1868 case MSR_IA32_RTIT_CTL:
1869 if (!vmx_pt_mode_is_host_guest())
1871 msr_info->data = vmx->pt_desc.guest.ctl;
1873 case MSR_IA32_RTIT_STATUS:
1874 if (!vmx_pt_mode_is_host_guest())
1876 msr_info->data = vmx->pt_desc.guest.status;
1878 case MSR_IA32_RTIT_CR3_MATCH:
1879 if (!vmx_pt_mode_is_host_guest() ||
1880 !intel_pt_validate_cap(vmx->pt_desc.caps,
1881 PT_CAP_cr3_filtering))
1883 msr_info->data = vmx->pt_desc.guest.cr3_match;
1885 case MSR_IA32_RTIT_OUTPUT_BASE:
1886 if (!vmx_pt_mode_is_host_guest() ||
1887 (!intel_pt_validate_cap(vmx->pt_desc.caps,
1888 PT_CAP_topa_output) &&
1889 !intel_pt_validate_cap(vmx->pt_desc.caps,
1890 PT_CAP_single_range_output)))
1892 msr_info->data = vmx->pt_desc.guest.output_base;
1894 case MSR_IA32_RTIT_OUTPUT_MASK:
1895 if (!vmx_pt_mode_is_host_guest() ||
1896 (!intel_pt_validate_cap(vmx->pt_desc.caps,
1897 PT_CAP_topa_output) &&
1898 !intel_pt_validate_cap(vmx->pt_desc.caps,
1899 PT_CAP_single_range_output)))
1901 msr_info->data = vmx->pt_desc.guest.output_mask;
1903 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
1904 index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
1905 if (!vmx_pt_mode_is_host_guest() ||
1906 (index >= 2 * vmx->pt_desc.num_address_ranges))
1909 msr_info->data = vmx->pt_desc.guest.addr_b[index / 2];
1911 msr_info->data = vmx->pt_desc.guest.addr_a[index / 2];
1913 case MSR_IA32_DEBUGCTLMSR:
1914 msr_info->data = vmcs_read64(GUEST_IA32_DEBUGCTL);
1918 msr = vmx_find_uret_msr(vmx, msr_info->index);
1920 msr_info->data = msr->data;
1923 return kvm_get_msr_common(vcpu, msr_info);
1929 static u64 nested_vmx_truncate_sysenter_addr(struct kvm_vcpu *vcpu,
1932 #ifdef CONFIG_X86_64
1933 if (!guest_cpuid_has(vcpu, X86_FEATURE_LM))
1936 return (unsigned long)data;
1939 static u64 vcpu_supported_debugctl(struct kvm_vcpu *vcpu)
1941 u64 debugctl = vmx_supported_debugctl();
1943 if (!intel_pmu_lbr_is_enabled(vcpu))
1944 debugctl &= ~DEBUGCTLMSR_LBR_MASK;
1946 if (!guest_cpuid_has(vcpu, X86_FEATURE_BUS_LOCK_DETECT))
1947 debugctl &= ~DEBUGCTLMSR_BUS_LOCK_DETECT;
1953 * Writes msr value into the appropriate "register".
1954 * Returns 0 on success, non-0 otherwise.
1955 * Assumes vcpu_load() was already called.
1957 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1959 struct vcpu_vmx *vmx = to_vmx(vcpu);
1960 struct vmx_uret_msr *msr;
1962 u32 msr_index = msr_info->index;
1963 u64 data = msr_info->data;
1966 switch (msr_index) {
1968 ret = kvm_set_msr_common(vcpu, msr_info);
1970 #ifdef CONFIG_X86_64
1972 vmx_segment_cache_clear(vmx);
1973 vmcs_writel(GUEST_FS_BASE, data);
1976 vmx_segment_cache_clear(vmx);
1977 vmcs_writel(GUEST_GS_BASE, data);
1979 case MSR_KERNEL_GS_BASE:
1980 vmx_write_guest_kernel_gs_base(vmx, data);
1983 ret = kvm_set_msr_common(vcpu, msr_info);
1985 * Always intercepting WRMSR could incur non-negligible
1986 * overhead given xfd might be changed frequently in
1987 * guest context switch. Disable write interception
1988 * upon the first write with a non-zero value (indicating
1989 * potential usage on dynamic xfeatures). Also update
1990 * exception bitmap to trap #NM for proper virtualization
1994 vmx_disable_intercept_for_msr(vcpu, MSR_IA32_XFD,
1996 vcpu->arch.xfd_no_write_intercept = true;
1997 vmx_update_exception_bitmap(vcpu);
2001 case MSR_IA32_SYSENTER_CS:
2002 if (is_guest_mode(vcpu))
2003 get_vmcs12(vcpu)->guest_sysenter_cs = data;
2004 vmcs_write32(GUEST_SYSENTER_CS, data);
2006 case MSR_IA32_SYSENTER_EIP:
2007 if (is_guest_mode(vcpu)) {
2008 data = nested_vmx_truncate_sysenter_addr(vcpu, data);
2009 get_vmcs12(vcpu)->guest_sysenter_eip = data;
2011 vmcs_writel(GUEST_SYSENTER_EIP, data);
2013 case MSR_IA32_SYSENTER_ESP:
2014 if (is_guest_mode(vcpu)) {
2015 data = nested_vmx_truncate_sysenter_addr(vcpu, data);
2016 get_vmcs12(vcpu)->guest_sysenter_esp = data;
2018 vmcs_writel(GUEST_SYSENTER_ESP, data);
2020 case MSR_IA32_DEBUGCTLMSR: {
2021 u64 invalid = data & ~vcpu_supported_debugctl(vcpu);
2022 if (invalid & (DEBUGCTLMSR_BTF|DEBUGCTLMSR_LBR)) {
2023 if (report_ignored_msrs)
2024 vcpu_unimpl(vcpu, "%s: BTF|LBR in IA32_DEBUGCTLMSR 0x%llx, nop\n",
2026 data &= ~(DEBUGCTLMSR_BTF|DEBUGCTLMSR_LBR);
2027 invalid &= ~(DEBUGCTLMSR_BTF|DEBUGCTLMSR_LBR);
2033 if (is_guest_mode(vcpu) && get_vmcs12(vcpu)->vm_exit_controls &
2034 VM_EXIT_SAVE_DEBUG_CONTROLS)
2035 get_vmcs12(vcpu)->guest_ia32_debugctl = data;
2037 vmcs_write64(GUEST_IA32_DEBUGCTL, data);
2038 if (intel_pmu_lbr_is_enabled(vcpu) && !to_vmx(vcpu)->lbr_desc.event &&
2039 (data & DEBUGCTLMSR_LBR))
2040 intel_pmu_create_guest_lbr_event(vcpu);
2043 case MSR_IA32_BNDCFGS:
2044 if (!kvm_mpx_supported() ||
2045 (!msr_info->host_initiated &&
2046 !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
2048 if (is_noncanonical_address(data & PAGE_MASK, vcpu) ||
2049 (data & MSR_IA32_BNDCFGS_RSVD))
2051 vmcs_write64(GUEST_BNDCFGS, data);
2053 case MSR_IA32_UMWAIT_CONTROL:
2054 if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
2057 /* The reserved bit 1 and non-32 bit [63:32] should be zero */
2058 if (data & (BIT_ULL(1) | GENMASK_ULL(63, 32)))
2061 vmx->msr_ia32_umwait_control = data;
2063 case MSR_IA32_SPEC_CTRL:
2064 if (!msr_info->host_initiated &&
2065 !guest_has_spec_ctrl_msr(vcpu))
2068 if (kvm_spec_ctrl_test_value(data))
2071 vmx->spec_ctrl = data;
2077 * When it's written (to non-zero) for the first time, pass
2081 * The handling of the MSR bitmap for L2 guests is done in
2082 * nested_vmx_prepare_msr_bitmap. We should not touch the
2083 * vmcs02.msr_bitmap here since it gets completely overwritten
2084 * in the merging. We update the vmcs01 here for L1 as well
2085 * since it will end up touching the MSR anyway now.
2087 vmx_disable_intercept_for_msr(vcpu,
2091 case MSR_IA32_TSX_CTRL:
2092 if (!msr_info->host_initiated &&
2093 !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
2095 if (data & ~(TSX_CTRL_RTM_DISABLE | TSX_CTRL_CPUID_CLEAR))
2098 case MSR_IA32_PRED_CMD:
2099 if (!msr_info->host_initiated &&
2100 !guest_has_pred_cmd_msr(vcpu))
2103 if (data & ~PRED_CMD_IBPB)
2105 if (!boot_cpu_has(X86_FEATURE_IBPB))
2110 wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
2114 * When it's written (to non-zero) for the first time, pass
2118 * The handling of the MSR bitmap for L2 guests is done in
2119 * nested_vmx_prepare_msr_bitmap. We should not touch the
2120 * vmcs02.msr_bitmap here since it gets completely overwritten
2123 vmx_disable_intercept_for_msr(vcpu, MSR_IA32_PRED_CMD, MSR_TYPE_W);
2125 case MSR_IA32_CR_PAT:
2126 if (!kvm_pat_valid(data))
2129 if (is_guest_mode(vcpu) &&
2130 get_vmcs12(vcpu)->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
2131 get_vmcs12(vcpu)->guest_ia32_pat = data;
2133 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2134 vmcs_write64(GUEST_IA32_PAT, data);
2135 vcpu->arch.pat = data;
2138 ret = kvm_set_msr_common(vcpu, msr_info);
2140 case MSR_IA32_MCG_EXT_CTL:
2141 if ((!msr_info->host_initiated &&
2142 !(to_vmx(vcpu)->msr_ia32_feature_control &
2143 FEAT_CTL_LMCE_ENABLED)) ||
2144 (data & ~MCG_EXT_CTL_LMCE_EN))
2146 vcpu->arch.mcg_ext_ctl = data;
2148 case MSR_IA32_FEAT_CTL:
2149 if (!vmx_feature_control_msr_valid(vcpu, data) ||
2150 (to_vmx(vcpu)->msr_ia32_feature_control &
2151 FEAT_CTL_LOCKED && !msr_info->host_initiated))
2153 vmx->msr_ia32_feature_control = data;
2154 if (msr_info->host_initiated && data == 0)
2155 vmx_leave_nested(vcpu);
2157 /* SGX may be enabled/disabled by guest's firmware */
2158 vmx_write_encls_bitmap(vcpu, NULL);
2160 case MSR_IA32_SGXLEPUBKEYHASH0 ... MSR_IA32_SGXLEPUBKEYHASH3:
2162 * On real hardware, the LE hash MSRs are writable before
2163 * the firmware sets bit 0 in MSR 0x7a ("activating" SGX),
2164 * at which point SGX related bits in IA32_FEATURE_CONTROL
2167 * KVM does not emulate SGX activation for simplicity, so
2168 * allow writes to the LE hash MSRs if IA32_FEATURE_CONTROL
2169 * is unlocked. This is technically not architectural
2170 * behavior, but it's close enough.
2172 if (!msr_info->host_initiated &&
2173 (!guest_cpuid_has(vcpu, X86_FEATURE_SGX_LC) ||
2174 ((vmx->msr_ia32_feature_control & FEAT_CTL_LOCKED) &&
2175 !(vmx->msr_ia32_feature_control & FEAT_CTL_SGX_LC_ENABLED))))
2177 vmx->msr_ia32_sgxlepubkeyhash
2178 [msr_index - MSR_IA32_SGXLEPUBKEYHASH0] = data;
2180 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2181 if (!msr_info->host_initiated)
2182 return 1; /* they are read-only */
2183 if (!nested_vmx_allowed(vcpu))
2185 return vmx_set_vmx_msr(vcpu, msr_index, data);
2186 case MSR_IA32_RTIT_CTL:
2187 if (!vmx_pt_mode_is_host_guest() ||
2188 vmx_rtit_ctl_check(vcpu, data) ||
2191 vmcs_write64(GUEST_IA32_RTIT_CTL, data);
2192 vmx->pt_desc.guest.ctl = data;
2193 pt_update_intercept_for_msr(vcpu);
2195 case MSR_IA32_RTIT_STATUS:
2196 if (!pt_can_write_msr(vmx))
2198 if (data & MSR_IA32_RTIT_STATUS_MASK)
2200 vmx->pt_desc.guest.status = data;
2202 case MSR_IA32_RTIT_CR3_MATCH:
2203 if (!pt_can_write_msr(vmx))
2205 if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2206 PT_CAP_cr3_filtering))
2208 vmx->pt_desc.guest.cr3_match = data;
2210 case MSR_IA32_RTIT_OUTPUT_BASE:
2211 if (!pt_can_write_msr(vmx))
2213 if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2214 PT_CAP_topa_output) &&
2215 !intel_pt_validate_cap(vmx->pt_desc.caps,
2216 PT_CAP_single_range_output))
2218 if (!pt_output_base_valid(vcpu, data))
2220 vmx->pt_desc.guest.output_base = data;
2222 case MSR_IA32_RTIT_OUTPUT_MASK:
2223 if (!pt_can_write_msr(vmx))
2225 if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2226 PT_CAP_topa_output) &&
2227 !intel_pt_validate_cap(vmx->pt_desc.caps,
2228 PT_CAP_single_range_output))
2230 vmx->pt_desc.guest.output_mask = data;
2232 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
2233 if (!pt_can_write_msr(vmx))
2235 index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
2236 if (index >= 2 * vmx->pt_desc.num_address_ranges)
2238 if (is_noncanonical_address(data, vcpu))
2241 vmx->pt_desc.guest.addr_b[index / 2] = data;
2243 vmx->pt_desc.guest.addr_a[index / 2] = data;
2245 case MSR_IA32_PERF_CAPABILITIES:
2246 if (data && !vcpu_to_pmu(vcpu)->version)
2248 if (data & PMU_CAP_LBR_FMT) {
2249 if ((data & PMU_CAP_LBR_FMT) !=
2250 (vmx_get_perf_capabilities() & PMU_CAP_LBR_FMT))
2252 if (!intel_pmu_lbr_is_compatible(vcpu))
2255 ret = kvm_set_msr_common(vcpu, msr_info);
2260 msr = vmx_find_uret_msr(vmx, msr_index);
2262 ret = vmx_set_guest_uret_msr(vmx, msr, data);
2264 ret = kvm_set_msr_common(vcpu, msr_info);
2270 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2272 unsigned long guest_owned_bits;
2274 kvm_register_mark_available(vcpu, reg);
2278 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2281 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2283 case VCPU_EXREG_PDPTR:
2285 ept_save_pdptrs(vcpu);
2287 case VCPU_EXREG_CR0:
2288 guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
2290 vcpu->arch.cr0 &= ~guest_owned_bits;
2291 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & guest_owned_bits;
2293 case VCPU_EXREG_CR3:
2295 * When intercepting CR3 loads, e.g. for shadowing paging, KVM's
2296 * CR3 is loaded into hardware, not the guest's CR3.
2298 if (!(exec_controls_get(to_vmx(vcpu)) & CPU_BASED_CR3_LOAD_EXITING))
2299 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2301 case VCPU_EXREG_CR4:
2302 guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
2304 vcpu->arch.cr4 &= ~guest_owned_bits;
2305 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & guest_owned_bits;
2308 KVM_BUG_ON(1, vcpu->kvm);
2313 static __init int cpu_has_kvm_support(void)
2315 return cpu_has_vmx();
2318 static __init int vmx_disabled_by_bios(void)
2320 return !boot_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
2321 !boot_cpu_has(X86_FEATURE_VMX);
2324 static int kvm_cpu_vmxon(u64 vmxon_pointer)
2328 cr4_set_bits(X86_CR4_VMXE);
2330 asm_volatile_goto("1: vmxon %[vmxon_pointer]\n\t"
2331 _ASM_EXTABLE(1b, %l[fault])
2332 : : [vmxon_pointer] "m"(vmxon_pointer)
2337 WARN_ONCE(1, "VMXON faulted, MSR_IA32_FEAT_CTL (0x3a) = 0x%llx\n",
2338 rdmsrl_safe(MSR_IA32_FEAT_CTL, &msr) ? 0xdeadbeef : msr);
2339 cr4_clear_bits(X86_CR4_VMXE);
2344 static int hardware_enable(void)
2346 int cpu = raw_smp_processor_id();
2347 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2350 if (cr4_read_shadow() & X86_CR4_VMXE)
2354 * This can happen if we hot-added a CPU but failed to allocate
2355 * VP assist page for it.
2357 if (static_branch_unlikely(&enable_evmcs) &&
2358 !hv_get_vp_assist_page(cpu))
2361 intel_pt_handle_vmx(1);
2363 r = kvm_cpu_vmxon(phys_addr);
2365 intel_pt_handle_vmx(0);
2375 static void vmclear_local_loaded_vmcss(void)
2377 int cpu = raw_smp_processor_id();
2378 struct loaded_vmcs *v, *n;
2380 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2381 loaded_vmcss_on_cpu_link)
2382 __loaded_vmcs_clear(v);
2385 static void hardware_disable(void)
2387 vmclear_local_loaded_vmcss();
2390 kvm_spurious_fault();
2392 intel_pt_handle_vmx(0);
2396 * There is no X86_FEATURE for SGX yet, but anyway we need to query CPUID
2397 * directly instead of going through cpu_has(), to ensure KVM is trapping
2398 * ENCLS whenever it's supported in hardware. It does not matter whether
2399 * the host OS supports or has enabled SGX.
2401 static bool cpu_has_sgx(void)
2403 return cpuid_eax(0) >= 0x12 && (cpuid_eax(0x12) & BIT(0));
2406 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2407 u32 msr, u32 *result)
2409 u32 vmx_msr_low, vmx_msr_high;
2410 u32 ctl = ctl_min | ctl_opt;
2412 rdmsr(msr, vmx_msr_low, vmx_msr_high);
2414 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2415 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
2417 /* Ensure minimum (required) set of control bits are supported. */
2425 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf,
2426 struct vmx_capability *vmx_cap)
2428 u32 vmx_msr_low, vmx_msr_high;
2429 u32 min, opt, min2, opt2;
2430 u32 _pin_based_exec_control = 0;
2431 u32 _cpu_based_exec_control = 0;
2432 u32 _cpu_based_2nd_exec_control = 0;
2433 u32 _vmexit_control = 0;
2434 u32 _vmentry_control = 0;
2436 memset(vmcs_conf, 0, sizeof(*vmcs_conf));
2437 min = CPU_BASED_HLT_EXITING |
2438 #ifdef CONFIG_X86_64
2439 CPU_BASED_CR8_LOAD_EXITING |
2440 CPU_BASED_CR8_STORE_EXITING |
2442 CPU_BASED_CR3_LOAD_EXITING |
2443 CPU_BASED_CR3_STORE_EXITING |
2444 CPU_BASED_UNCOND_IO_EXITING |
2445 CPU_BASED_MOV_DR_EXITING |
2446 CPU_BASED_USE_TSC_OFFSETTING |
2447 CPU_BASED_MWAIT_EXITING |
2448 CPU_BASED_MONITOR_EXITING |
2449 CPU_BASED_INVLPG_EXITING |
2450 CPU_BASED_RDPMC_EXITING;
2452 opt = CPU_BASED_TPR_SHADOW |
2453 CPU_BASED_USE_MSR_BITMAPS |
2454 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2455 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2456 &_cpu_based_exec_control) < 0)
2458 #ifdef CONFIG_X86_64
2459 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2460 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2461 ~CPU_BASED_CR8_STORE_EXITING;
2463 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2465 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2466 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2467 SECONDARY_EXEC_WBINVD_EXITING |
2468 SECONDARY_EXEC_ENABLE_VPID |
2469 SECONDARY_EXEC_ENABLE_EPT |
2470 SECONDARY_EXEC_UNRESTRICTED_GUEST |
2471 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2472 SECONDARY_EXEC_DESC |
2473 SECONDARY_EXEC_ENABLE_RDTSCP |
2474 SECONDARY_EXEC_ENABLE_INVPCID |
2475 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2476 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2477 SECONDARY_EXEC_SHADOW_VMCS |
2478 SECONDARY_EXEC_XSAVES |
2479 SECONDARY_EXEC_RDSEED_EXITING |
2480 SECONDARY_EXEC_RDRAND_EXITING |
2481 SECONDARY_EXEC_ENABLE_PML |
2482 SECONDARY_EXEC_TSC_SCALING |
2483 SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE |
2484 SECONDARY_EXEC_PT_USE_GPA |
2485 SECONDARY_EXEC_PT_CONCEAL_VMX |
2486 SECONDARY_EXEC_ENABLE_VMFUNC |
2487 SECONDARY_EXEC_BUS_LOCK_DETECTION;
2489 opt2 |= SECONDARY_EXEC_ENCLS_EXITING;
2490 if (adjust_vmx_controls(min2, opt2,
2491 MSR_IA32_VMX_PROCBASED_CTLS2,
2492 &_cpu_based_2nd_exec_control) < 0)
2495 #ifndef CONFIG_X86_64
2496 if (!(_cpu_based_2nd_exec_control &
2497 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2498 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2501 if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2502 _cpu_based_2nd_exec_control &= ~(
2503 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2504 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2505 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
2507 rdmsr_safe(MSR_IA32_VMX_EPT_VPID_CAP,
2508 &vmx_cap->ept, &vmx_cap->vpid);
2510 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2511 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2513 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2514 CPU_BASED_CR3_STORE_EXITING |
2515 CPU_BASED_INVLPG_EXITING);
2516 } else if (vmx_cap->ept) {
2518 pr_warn_once("EPT CAP should not exist if not support "
2519 "1-setting enable EPT VM-execution control\n");
2521 if (!(_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_VPID) &&
2524 pr_warn_once("VPID CAP should not exist if not support "
2525 "1-setting enable VPID VM-execution control\n");
2528 min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
2529 #ifdef CONFIG_X86_64
2530 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2532 opt = VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL |
2533 VM_EXIT_LOAD_IA32_PAT |
2534 VM_EXIT_LOAD_IA32_EFER |
2535 VM_EXIT_CLEAR_BNDCFGS |
2536 VM_EXIT_PT_CONCEAL_PIP |
2537 VM_EXIT_CLEAR_IA32_RTIT_CTL;
2538 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2539 &_vmexit_control) < 0)
2542 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2543 opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR |
2544 PIN_BASED_VMX_PREEMPTION_TIMER;
2545 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2546 &_pin_based_exec_control) < 0)
2549 if (cpu_has_broken_vmx_preemption_timer())
2550 _pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
2551 if (!(_cpu_based_2nd_exec_control &
2552 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
2553 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
2555 min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
2556 opt = VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL |
2557 VM_ENTRY_LOAD_IA32_PAT |
2558 VM_ENTRY_LOAD_IA32_EFER |
2559 VM_ENTRY_LOAD_BNDCFGS |
2560 VM_ENTRY_PT_CONCEAL_PIP |
2561 VM_ENTRY_LOAD_IA32_RTIT_CTL;
2562 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2563 &_vmentry_control) < 0)
2567 * Some cpus support VM_{ENTRY,EXIT}_IA32_PERF_GLOBAL_CTRL but they
2568 * can't be used due to an errata where VM Exit may incorrectly clear
2569 * IA32_PERF_GLOBAL_CTRL[34:32]. Workaround the errata by using the
2570 * MSR load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2572 if (boot_cpu_data.x86 == 0x6) {
2573 switch (boot_cpu_data.x86_model) {
2574 case 26: /* AAK155 */
2575 case 30: /* AAP115 */
2576 case 37: /* AAT100 */
2577 case 44: /* BC86,AAY89,BD102 */
2579 _vmentry_control &= ~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
2580 _vmexit_control &= ~VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
2581 pr_warn_once("kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
2582 "does not work properly. Using workaround\n");
2590 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2592 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2593 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2596 #ifdef CONFIG_X86_64
2597 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2598 if (vmx_msr_high & (1u<<16))
2602 /* Require Write-Back (WB) memory type for VMCS accesses. */
2603 if (((vmx_msr_high >> 18) & 15) != 6)
2606 vmcs_conf->size = vmx_msr_high & 0x1fff;
2607 vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
2609 vmcs_conf->revision_id = vmx_msr_low;
2611 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2612 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2613 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2614 vmcs_conf->vmexit_ctrl = _vmexit_control;
2615 vmcs_conf->vmentry_ctrl = _vmentry_control;
2617 #if IS_ENABLED(CONFIG_HYPERV)
2618 if (enlightened_vmcs)
2619 evmcs_sanitize_exec_ctrls(vmcs_conf);
2625 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags)
2627 int node = cpu_to_node(cpu);
2631 pages = __alloc_pages_node(node, flags, 0);
2634 vmcs = page_address(pages);
2635 memset(vmcs, 0, vmcs_config.size);
2637 /* KVM supports Enlightened VMCS v1 only */
2638 if (static_branch_unlikely(&enable_evmcs))
2639 vmcs->hdr.revision_id = KVM_EVMCS_VERSION;
2641 vmcs->hdr.revision_id = vmcs_config.revision_id;
2644 vmcs->hdr.shadow_vmcs = 1;
2648 void free_vmcs(struct vmcs *vmcs)
2650 free_page((unsigned long)vmcs);
2654 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2656 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2658 if (!loaded_vmcs->vmcs)
2660 loaded_vmcs_clear(loaded_vmcs);
2661 free_vmcs(loaded_vmcs->vmcs);
2662 loaded_vmcs->vmcs = NULL;
2663 if (loaded_vmcs->msr_bitmap)
2664 free_page((unsigned long)loaded_vmcs->msr_bitmap);
2665 WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
2668 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2670 loaded_vmcs->vmcs = alloc_vmcs(false);
2671 if (!loaded_vmcs->vmcs)
2674 vmcs_clear(loaded_vmcs->vmcs);
2676 loaded_vmcs->shadow_vmcs = NULL;
2677 loaded_vmcs->hv_timer_soft_disabled = false;
2678 loaded_vmcs->cpu = -1;
2679 loaded_vmcs->launched = 0;
2681 if (cpu_has_vmx_msr_bitmap()) {
2682 loaded_vmcs->msr_bitmap = (unsigned long *)
2683 __get_free_page(GFP_KERNEL_ACCOUNT);
2684 if (!loaded_vmcs->msr_bitmap)
2686 memset(loaded_vmcs->msr_bitmap, 0xff, PAGE_SIZE);
2689 memset(&loaded_vmcs->host_state, 0, sizeof(struct vmcs_host_state));
2690 memset(&loaded_vmcs->controls_shadow, 0,
2691 sizeof(struct vmcs_controls_shadow));
2696 free_loaded_vmcs(loaded_vmcs);
2700 static void free_kvm_area(void)
2704 for_each_possible_cpu(cpu) {
2705 free_vmcs(per_cpu(vmxarea, cpu));
2706 per_cpu(vmxarea, cpu) = NULL;
2710 static __init int alloc_kvm_area(void)
2714 for_each_possible_cpu(cpu) {
2717 vmcs = alloc_vmcs_cpu(false, cpu, GFP_KERNEL);
2724 * When eVMCS is enabled, alloc_vmcs_cpu() sets
2725 * vmcs->revision_id to KVM_EVMCS_VERSION instead of
2726 * revision_id reported by MSR_IA32_VMX_BASIC.
2728 * However, even though not explicitly documented by
2729 * TLFS, VMXArea passed as VMXON argument should
2730 * still be marked with revision_id reported by
2733 if (static_branch_unlikely(&enable_evmcs))
2734 vmcs->hdr.revision_id = vmcs_config.revision_id;
2736 per_cpu(vmxarea, cpu) = vmcs;
2741 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
2742 struct kvm_segment *save)
2744 if (!emulate_invalid_guest_state) {
2746 * CS and SS RPL should be equal during guest entry according
2747 * to VMX spec, but in reality it is not always so. Since vcpu
2748 * is in the middle of the transition from real mode to
2749 * protected mode it is safe to assume that RPL 0 is a good
2752 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
2753 save->selector &= ~SEGMENT_RPL_MASK;
2754 save->dpl = save->selector & SEGMENT_RPL_MASK;
2757 __vmx_set_segment(vcpu, save, seg);
2760 static void enter_pmode(struct kvm_vcpu *vcpu)
2762 unsigned long flags;
2763 struct vcpu_vmx *vmx = to_vmx(vcpu);
2766 * Update real mode segment cache. It may be not up-to-date if segment
2767 * register was written while vcpu was in a guest mode.
2769 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2770 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2771 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2772 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2773 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2774 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2776 vmx->rmode.vm86_active = 0;
2778 __vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2780 flags = vmcs_readl(GUEST_RFLAGS);
2781 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2782 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2783 vmcs_writel(GUEST_RFLAGS, flags);
2785 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
2786 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
2788 vmx_update_exception_bitmap(vcpu);
2790 fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2791 fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2792 fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2793 fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2794 fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2795 fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2798 static void fix_rmode_seg(int seg, struct kvm_segment *save)
2800 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2801 struct kvm_segment var = *save;
2804 if (seg == VCPU_SREG_CS)
2807 if (!emulate_invalid_guest_state) {
2808 var.selector = var.base >> 4;
2809 var.base = var.base & 0xffff0;
2819 if (save->base & 0xf)
2820 printk_once(KERN_WARNING "kvm: segment base is not "
2821 "paragraph aligned when entering "
2822 "protected mode (seg=%d)", seg);
2825 vmcs_write16(sf->selector, var.selector);
2826 vmcs_writel(sf->base, var.base);
2827 vmcs_write32(sf->limit, var.limit);
2828 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
2831 static void enter_rmode(struct kvm_vcpu *vcpu)
2833 unsigned long flags;
2834 struct vcpu_vmx *vmx = to_vmx(vcpu);
2835 struct kvm_vmx *kvm_vmx = to_kvm_vmx(vcpu->kvm);
2837 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2838 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2839 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2840 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2841 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2842 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2843 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2845 vmx->rmode.vm86_active = 1;
2848 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
2849 * vcpu. Warn the user that an update is overdue.
2851 if (!kvm_vmx->tss_addr)
2852 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
2853 "called before entering vcpu\n");
2855 vmx_segment_cache_clear(vmx);
2857 vmcs_writel(GUEST_TR_BASE, kvm_vmx->tss_addr);
2858 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
2859 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2861 flags = vmcs_readl(GUEST_RFLAGS);
2862 vmx->rmode.save_rflags = flags;
2864 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2866 vmcs_writel(GUEST_RFLAGS, flags);
2867 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
2868 vmx_update_exception_bitmap(vcpu);
2870 fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2871 fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2872 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2873 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2874 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2875 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2878 int vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
2880 struct vcpu_vmx *vmx = to_vmx(vcpu);
2881 struct vmx_uret_msr *msr = vmx_find_uret_msr(vmx, MSR_EFER);
2883 /* Nothing to do if hardware doesn't support EFER. */
2887 vcpu->arch.efer = efer;
2888 if (efer & EFER_LMA) {
2889 vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2892 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2894 msr->data = efer & ~EFER_LME;
2896 vmx_setup_uret_msrs(vmx);
2900 #ifdef CONFIG_X86_64
2902 static void enter_lmode(struct kvm_vcpu *vcpu)
2906 vmx_segment_cache_clear(to_vmx(vcpu));
2908 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
2909 if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
2910 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
2912 vmcs_write32(GUEST_TR_AR_BYTES,
2913 (guest_tr_ar & ~VMX_AR_TYPE_MASK)
2914 | VMX_AR_TYPE_BUSY_64_TSS);
2916 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
2919 static void exit_lmode(struct kvm_vcpu *vcpu)
2921 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2922 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
2927 static void vmx_flush_tlb_all(struct kvm_vcpu *vcpu)
2929 struct vcpu_vmx *vmx = to_vmx(vcpu);
2932 * INVEPT must be issued when EPT is enabled, irrespective of VPID, as
2933 * the CPU is not required to invalidate guest-physical mappings on
2934 * VM-Entry, even if VPID is disabled. Guest-physical mappings are
2935 * associated with the root EPT structure and not any particular VPID
2936 * (INVVPID also isn't required to invalidate guest-physical mappings).
2940 } else if (enable_vpid) {
2941 if (cpu_has_vmx_invvpid_global()) {
2942 vpid_sync_vcpu_global();
2944 vpid_sync_vcpu_single(vmx->vpid);
2945 vpid_sync_vcpu_single(vmx->nested.vpid02);
2950 static inline int vmx_get_current_vpid(struct kvm_vcpu *vcpu)
2952 if (is_guest_mode(vcpu))
2953 return nested_get_vpid02(vcpu);
2954 return to_vmx(vcpu)->vpid;
2957 static void vmx_flush_tlb_current(struct kvm_vcpu *vcpu)
2959 struct kvm_mmu *mmu = vcpu->arch.mmu;
2960 u64 root_hpa = mmu->root_hpa;
2962 /* No flush required if the current context is invalid. */
2963 if (!VALID_PAGE(root_hpa))
2967 ept_sync_context(construct_eptp(vcpu, root_hpa,
2968 mmu->shadow_root_level));
2970 vpid_sync_context(vmx_get_current_vpid(vcpu));
2973 static void vmx_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t addr)
2976 * vpid_sync_vcpu_addr() is a nop if vpid==0, see the comment in
2977 * vmx_flush_tlb_guest() for an explanation of why this is ok.
2979 vpid_sync_vcpu_addr(vmx_get_current_vpid(vcpu), addr);
2982 static void vmx_flush_tlb_guest(struct kvm_vcpu *vcpu)
2985 * vpid_sync_context() is a nop if vpid==0, e.g. if enable_vpid==0 or a
2986 * vpid couldn't be allocated for this vCPU. VM-Enter and VM-Exit are
2987 * required to flush GVA->{G,H}PA mappings from the TLB if vpid is
2988 * disabled (VM-Enter with vpid enabled and vpid==0 is disallowed),
2989 * i.e. no explicit INVVPID is necessary.
2991 vpid_sync_context(vmx_get_current_vpid(vcpu));
2994 void vmx_ept_load_pdptrs(struct kvm_vcpu *vcpu)
2996 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2998 if (!kvm_register_is_dirty(vcpu, VCPU_EXREG_PDPTR))
3001 if (is_pae_paging(vcpu)) {
3002 vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
3003 vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
3004 vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
3005 vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
3009 void ept_save_pdptrs(struct kvm_vcpu *vcpu)
3011 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
3013 if (WARN_ON_ONCE(!is_pae_paging(vcpu)))
3016 mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
3017 mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
3018 mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
3019 mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
3021 kvm_register_mark_available(vcpu, VCPU_EXREG_PDPTR);
3024 #define CR3_EXITING_BITS (CPU_BASED_CR3_LOAD_EXITING | \
3025 CPU_BASED_CR3_STORE_EXITING)
3027 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
3029 struct vcpu_vmx *vmx = to_vmx(vcpu);
3030 unsigned long hw_cr0, old_cr0_pg;
3033 old_cr0_pg = kvm_read_cr0_bits(vcpu, X86_CR0_PG);
3035 hw_cr0 = (cr0 & ~KVM_VM_CR0_ALWAYS_OFF);
3036 if (is_unrestricted_guest(vcpu))
3037 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
3039 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
3041 hw_cr0 |= X86_CR0_WP;
3043 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
3046 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
3050 vmcs_writel(CR0_READ_SHADOW, cr0);
3051 vmcs_writel(GUEST_CR0, hw_cr0);
3052 vcpu->arch.cr0 = cr0;
3053 kvm_register_mark_available(vcpu, VCPU_EXREG_CR0);
3055 #ifdef CONFIG_X86_64
3056 if (vcpu->arch.efer & EFER_LME) {
3057 if (!old_cr0_pg && (cr0 & X86_CR0_PG))
3059 else if (old_cr0_pg && !(cr0 & X86_CR0_PG))
3064 if (enable_ept && !is_unrestricted_guest(vcpu)) {
3066 * Ensure KVM has an up-to-date snapshot of the guest's CR3. If
3067 * the below code _enables_ CR3 exiting, vmx_cache_reg() will
3068 * (correctly) stop reading vmcs.GUEST_CR3 because it thinks
3069 * KVM's CR3 is installed.
3071 if (!kvm_register_is_available(vcpu, VCPU_EXREG_CR3))
3072 vmx_cache_reg(vcpu, VCPU_EXREG_CR3);
3075 * When running with EPT but not unrestricted guest, KVM must
3076 * intercept CR3 accesses when paging is _disabled_. This is
3077 * necessary because restricted guests can't actually run with
3078 * paging disabled, and so KVM stuffs its own CR3 in order to
3079 * run the guest when identity mapped page tables.
3081 * Do _NOT_ check the old CR0.PG, e.g. to optimize away the
3082 * update, it may be stale with respect to CR3 interception,
3083 * e.g. after nested VM-Enter.
3085 * Lastly, honor L1's desires, i.e. intercept CR3 loads and/or
3086 * stores to forward them to L1, even if KVM does not need to
3087 * intercept them to preserve its identity mapped page tables.
3089 if (!(cr0 & X86_CR0_PG)) {
3090 exec_controls_setbit(vmx, CR3_EXITING_BITS);
3091 } else if (!is_guest_mode(vcpu)) {
3092 exec_controls_clearbit(vmx, CR3_EXITING_BITS);
3094 tmp = exec_controls_get(vmx);
3095 tmp &= ~CR3_EXITING_BITS;
3096 tmp |= get_vmcs12(vcpu)->cpu_based_vm_exec_control & CR3_EXITING_BITS;
3097 exec_controls_set(vmx, tmp);
3100 /* Note, vmx_set_cr4() consumes the new vcpu->arch.cr0. */
3101 if ((old_cr0_pg ^ cr0) & X86_CR0_PG)
3102 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3105 * When !CR0_PG -> CR0_PG, vcpu->arch.cr3 becomes active, but
3106 * GUEST_CR3 is still vmx->ept_identity_map_addr if EPT + !URG.
3108 if (!(old_cr0_pg & X86_CR0_PG) && (cr0 & X86_CR0_PG))
3109 kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
3112 /* depends on vcpu->arch.cr0 to be set to a new value */
3113 vmx->emulation_required = vmx_emulation_required(vcpu);
3116 static int vmx_get_max_tdp_level(void)
3118 if (cpu_has_vmx_ept_5levels())
3123 u64 construct_eptp(struct kvm_vcpu *vcpu, hpa_t root_hpa, int root_level)
3125 u64 eptp = VMX_EPTP_MT_WB;
3127 eptp |= (root_level == 5) ? VMX_EPTP_PWL_5 : VMX_EPTP_PWL_4;
3129 if (enable_ept_ad_bits &&
3130 (!is_guest_mode(vcpu) || nested_ept_ad_enabled(vcpu)))
3131 eptp |= VMX_EPTP_AD_ENABLE_BIT;
3137 static void vmx_load_mmu_pgd(struct kvm_vcpu *vcpu, hpa_t root_hpa,
3140 struct kvm *kvm = vcpu->kvm;
3141 bool update_guest_cr3 = true;
3142 unsigned long guest_cr3;
3146 eptp = construct_eptp(vcpu, root_hpa, root_level);
3147 vmcs_write64(EPT_POINTER, eptp);
3149 hv_track_root_tdp(vcpu, root_hpa);
3151 if (!enable_unrestricted_guest && !is_paging(vcpu))
3152 guest_cr3 = to_kvm_vmx(kvm)->ept_identity_map_addr;
3153 else if (kvm_register_is_dirty(vcpu, VCPU_EXREG_CR3))
3154 guest_cr3 = vcpu->arch.cr3;
3155 else /* vmcs.GUEST_CR3 is already up-to-date. */
3156 update_guest_cr3 = false;
3157 vmx_ept_load_pdptrs(vcpu);
3159 guest_cr3 = root_hpa | kvm_get_active_pcid(vcpu);
3162 if (update_guest_cr3)
3163 vmcs_writel(GUEST_CR3, guest_cr3);
3167 static bool vmx_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3170 * We operate under the default treatment of SMM, so VMX cannot be
3171 * enabled under SMM. Note, whether or not VMXE is allowed at all is
3172 * handled by kvm_is_valid_cr4().
3174 if ((cr4 & X86_CR4_VMXE) && is_smm(vcpu))
3177 if (to_vmx(vcpu)->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
3183 void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3185 unsigned long old_cr4 = vcpu->arch.cr4;
3186 struct vcpu_vmx *vmx = to_vmx(vcpu);
3188 * Pass through host's Machine Check Enable value to hw_cr4, which
3189 * is in force while we are in guest mode. Do not let guests control
3190 * this bit, even if host CR4.MCE == 0.
3192 unsigned long hw_cr4;
3194 hw_cr4 = (cr4_read_shadow() & X86_CR4_MCE) | (cr4 & ~X86_CR4_MCE);
3195 if (is_unrestricted_guest(vcpu))
3196 hw_cr4 |= KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST;
3197 else if (vmx->rmode.vm86_active)
3198 hw_cr4 |= KVM_RMODE_VM_CR4_ALWAYS_ON;
3200 hw_cr4 |= KVM_PMODE_VM_CR4_ALWAYS_ON;
3202 if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated()) {
3203 if (cr4 & X86_CR4_UMIP) {
3204 secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_DESC);
3205 hw_cr4 &= ~X86_CR4_UMIP;
3206 } else if (!is_guest_mode(vcpu) ||
3207 !nested_cpu_has2(get_vmcs12(vcpu), SECONDARY_EXEC_DESC)) {
3208 secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_DESC);
3212 vcpu->arch.cr4 = cr4;
3213 kvm_register_mark_available(vcpu, VCPU_EXREG_CR4);
3215 if (!is_unrestricted_guest(vcpu)) {
3217 if (!is_paging(vcpu)) {
3218 hw_cr4 &= ~X86_CR4_PAE;
3219 hw_cr4 |= X86_CR4_PSE;
3220 } else if (!(cr4 & X86_CR4_PAE)) {
3221 hw_cr4 &= ~X86_CR4_PAE;
3226 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
3227 * hardware. To emulate this behavior, SMEP/SMAP/PKU needs
3228 * to be manually disabled when guest switches to non-paging
3231 * If !enable_unrestricted_guest, the CPU is always running
3232 * with CR0.PG=1 and CR4 needs to be modified.
3233 * If enable_unrestricted_guest, the CPU automatically
3234 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
3236 if (!is_paging(vcpu))
3237 hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
3240 vmcs_writel(CR4_READ_SHADOW, cr4);
3241 vmcs_writel(GUEST_CR4, hw_cr4);
3243 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
3244 kvm_update_cpuid_runtime(vcpu);
3247 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3249 struct vcpu_vmx *vmx = to_vmx(vcpu);
3252 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3253 *var = vmx->rmode.segs[seg];
3254 if (seg == VCPU_SREG_TR
3255 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3257 var->base = vmx_read_guest_seg_base(vmx, seg);
3258 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3261 var->base = vmx_read_guest_seg_base(vmx, seg);
3262 var->limit = vmx_read_guest_seg_limit(vmx, seg);
3263 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3264 ar = vmx_read_guest_seg_ar(vmx, seg);
3265 var->unusable = (ar >> 16) & 1;
3266 var->type = ar & 15;
3267 var->s = (ar >> 4) & 1;
3268 var->dpl = (ar >> 5) & 3;
3270 * Some userspaces do not preserve unusable property. Since usable
3271 * segment has to be present according to VMX spec we can use present
3272 * property to amend userspace bug by making unusable segment always
3273 * nonpresent. vmx_segment_access_rights() already marks nonpresent
3274 * segment as unusable.
3276 var->present = !var->unusable;
3277 var->avl = (ar >> 12) & 1;
3278 var->l = (ar >> 13) & 1;
3279 var->db = (ar >> 14) & 1;
3280 var->g = (ar >> 15) & 1;
3283 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3285 struct kvm_segment s;
3287 if (to_vmx(vcpu)->rmode.vm86_active) {
3288 vmx_get_segment(vcpu, &s, seg);
3291 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3294 int vmx_get_cpl(struct kvm_vcpu *vcpu)
3296 struct vcpu_vmx *vmx = to_vmx(vcpu);
3298 if (unlikely(vmx->rmode.vm86_active))
3301 int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
3302 return VMX_AR_DPL(ar);
3306 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3310 if (var->unusable || !var->present)
3313 ar = var->type & 15;
3314 ar |= (var->s & 1) << 4;
3315 ar |= (var->dpl & 3) << 5;
3316 ar |= (var->present & 1) << 7;
3317 ar |= (var->avl & 1) << 12;
3318 ar |= (var->l & 1) << 13;
3319 ar |= (var->db & 1) << 14;
3320 ar |= (var->g & 1) << 15;
3326 void __vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3328 struct vcpu_vmx *vmx = to_vmx(vcpu);
3329 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3331 vmx_segment_cache_clear(vmx);
3333 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3334 vmx->rmode.segs[seg] = *var;
3335 if (seg == VCPU_SREG_TR)
3336 vmcs_write16(sf->selector, var->selector);
3338 fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3342 vmcs_writel(sf->base, var->base);
3343 vmcs_write32(sf->limit, var->limit);
3344 vmcs_write16(sf->selector, var->selector);
3347 * Fix the "Accessed" bit in AR field of segment registers for older
3349 * IA32 arch specifies that at the time of processor reset the
3350 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3351 * is setting it to 0 in the userland code. This causes invalid guest
3352 * state vmexit when "unrestricted guest" mode is turned on.
3353 * Fix for this setup issue in cpu_reset is being pushed in the qemu
3354 * tree. Newer qemu binaries with that qemu fix would not need this
3357 if (is_unrestricted_guest(vcpu) && (seg != VCPU_SREG_LDTR))
3358 var->type |= 0x1; /* Accessed */
3360 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3363 static void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3365 __vmx_set_segment(vcpu, var, seg);
3367 to_vmx(vcpu)->emulation_required = vmx_emulation_required(vcpu);
3370 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3372 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3374 *db = (ar >> 14) & 1;
3375 *l = (ar >> 13) & 1;
3378 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3380 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3381 dt->address = vmcs_readl(GUEST_IDTR_BASE);
3384 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3386 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3387 vmcs_writel(GUEST_IDTR_BASE, dt->address);
3390 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3392 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3393 dt->address = vmcs_readl(GUEST_GDTR_BASE);
3396 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3398 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3399 vmcs_writel(GUEST_GDTR_BASE, dt->address);
3402 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3404 struct kvm_segment var;
3407 vmx_get_segment(vcpu, &var, seg);
3409 if (seg == VCPU_SREG_CS)
3411 ar = vmx_segment_access_rights(&var);
3413 if (var.base != (var.selector << 4))
3415 if (var.limit != 0xffff)
3423 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3425 struct kvm_segment cs;
3426 unsigned int cs_rpl;
3428 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3429 cs_rpl = cs.selector & SEGMENT_RPL_MASK;
3433 if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
3437 if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
3438 if (cs.dpl > cs_rpl)
3441 if (cs.dpl != cs_rpl)
3447 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3451 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3453 struct kvm_segment ss;
3454 unsigned int ss_rpl;
3456 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3457 ss_rpl = ss.selector & SEGMENT_RPL_MASK;
3461 if (ss.type != 3 && ss.type != 7)
3465 if (ss.dpl != ss_rpl) /* DPL != RPL */
3473 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3475 struct kvm_segment var;
3478 vmx_get_segment(vcpu, &var, seg);
3479 rpl = var.selector & SEGMENT_RPL_MASK;
3487 if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
3488 if (var.dpl < rpl) /* DPL < RPL */
3492 /* TODO: Add other members to kvm_segment_field to allow checking for other access
3498 static bool tr_valid(struct kvm_vcpu *vcpu)
3500 struct kvm_segment tr;
3502 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3506 if (tr.selector & SEGMENT_TI_MASK) /* TI = 1 */
3508 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3516 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3518 struct kvm_segment ldtr;
3520 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3524 if (ldtr.selector & SEGMENT_TI_MASK) /* TI = 1 */
3534 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3536 struct kvm_segment cs, ss;
3538 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3539 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3541 return ((cs.selector & SEGMENT_RPL_MASK) ==
3542 (ss.selector & SEGMENT_RPL_MASK));
3546 * Check if guest state is valid. Returns true if valid, false if
3548 * We assume that registers are always usable
3550 bool __vmx_guest_state_valid(struct kvm_vcpu *vcpu)
3552 /* real mode guest state checks */
3553 if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
3554 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3556 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3558 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3560 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3562 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3564 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3567 /* protected mode guest state checks */
3568 if (!cs_ss_rpl_check(vcpu))
3570 if (!code_segment_valid(vcpu))
3572 if (!stack_segment_valid(vcpu))
3574 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3576 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3578 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3580 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3582 if (!tr_valid(vcpu))
3584 if (!ldtr_valid(vcpu))
3588 * - Add checks on RIP
3589 * - Add checks on RFLAGS
3595 static int init_rmode_tss(struct kvm *kvm, void __user *ua)
3597 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
3601 for (i = 0; i < 3; i++) {
3602 if (__copy_to_user(ua + PAGE_SIZE * i, zero_page, PAGE_SIZE))
3606 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3607 if (__copy_to_user(ua + TSS_IOPB_BASE_OFFSET, &data, sizeof(u16)))
3611 if (__copy_to_user(ua + RMODE_TSS_SIZE - 1, &data, sizeof(u8)))
3617 static int init_rmode_identity_map(struct kvm *kvm)
3619 struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm);
3624 /* Protect kvm_vmx->ept_identity_pagetable_done. */
3625 mutex_lock(&kvm->slots_lock);
3627 if (likely(kvm_vmx->ept_identity_pagetable_done))
3630 if (!kvm_vmx->ept_identity_map_addr)
3631 kvm_vmx->ept_identity_map_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
3633 uaddr = __x86_set_memory_region(kvm,
3634 IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
3635 kvm_vmx->ept_identity_map_addr,
3637 if (IS_ERR(uaddr)) {
3642 /* Set up identity-mapping pagetable for EPT in real mode */
3643 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3644 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3645 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3646 if (__copy_to_user(uaddr + i * sizeof(tmp), &tmp, sizeof(tmp))) {
3651 kvm_vmx->ept_identity_pagetable_done = true;
3654 mutex_unlock(&kvm->slots_lock);
3658 static void seg_setup(int seg)
3660 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3663 vmcs_write16(sf->selector, 0);
3664 vmcs_writel(sf->base, 0);
3665 vmcs_write32(sf->limit, 0xffff);
3667 if (seg == VCPU_SREG_CS)
3668 ar |= 0x08; /* code segment */
3670 vmcs_write32(sf->ar_bytes, ar);
3673 static int alloc_apic_access_page(struct kvm *kvm)
3679 mutex_lock(&kvm->slots_lock);
3680 if (kvm->arch.apic_access_memslot_enabled)
3682 hva = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
3683 APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
3689 page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
3690 if (is_error_page(page)) {
3696 * Do not pin the page in memory, so that memory hot-unplug
3697 * is able to migrate it.
3700 kvm->arch.apic_access_memslot_enabled = true;
3702 mutex_unlock(&kvm->slots_lock);
3706 int allocate_vpid(void)
3712 spin_lock(&vmx_vpid_lock);
3713 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3714 if (vpid < VMX_NR_VPIDS)
3715 __set_bit(vpid, vmx_vpid_bitmap);
3718 spin_unlock(&vmx_vpid_lock);
3722 void free_vpid(int vpid)
3724 if (!enable_vpid || vpid == 0)
3726 spin_lock(&vmx_vpid_lock);
3727 __clear_bit(vpid, vmx_vpid_bitmap);
3728 spin_unlock(&vmx_vpid_lock);
3731 static void vmx_msr_bitmap_l01_changed(struct vcpu_vmx *vmx)
3734 * When KVM is a nested hypervisor on top of Hyper-V and uses
3735 * 'Enlightened MSR Bitmap' feature L0 needs to know that MSR
3736 * bitmap has changed.
3738 if (static_branch_unlikely(&enable_evmcs))
3739 evmcs_touch_msr_bitmap();
3741 vmx->nested.force_msr_bitmap_recalc = true;
3744 void vmx_disable_intercept_for_msr(struct kvm_vcpu *vcpu, u32 msr, int type)
3746 struct vcpu_vmx *vmx = to_vmx(vcpu);
3747 unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3749 if (!cpu_has_vmx_msr_bitmap())
3752 vmx_msr_bitmap_l01_changed(vmx);
3755 * Mark the desired intercept state in shadow bitmap, this is needed
3756 * for resync when the MSR filters change.
3758 if (is_valid_passthrough_msr(msr)) {
3759 int idx = possible_passthrough_msr_slot(msr);
3761 if (idx != -ENOENT) {
3762 if (type & MSR_TYPE_R)
3763 clear_bit(idx, vmx->shadow_msr_intercept.read);
3764 if (type & MSR_TYPE_W)
3765 clear_bit(idx, vmx->shadow_msr_intercept.write);
3769 if ((type & MSR_TYPE_R) &&
3770 !kvm_msr_allowed(vcpu, msr, KVM_MSR_FILTER_READ)) {
3771 vmx_set_msr_bitmap_read(msr_bitmap, msr);
3772 type &= ~MSR_TYPE_R;
3775 if ((type & MSR_TYPE_W) &&
3776 !kvm_msr_allowed(vcpu, msr, KVM_MSR_FILTER_WRITE)) {
3777 vmx_set_msr_bitmap_write(msr_bitmap, msr);
3778 type &= ~MSR_TYPE_W;
3781 if (type & MSR_TYPE_R)
3782 vmx_clear_msr_bitmap_read(msr_bitmap, msr);
3784 if (type & MSR_TYPE_W)
3785 vmx_clear_msr_bitmap_write(msr_bitmap, msr);
3788 void vmx_enable_intercept_for_msr(struct kvm_vcpu *vcpu, u32 msr, int type)
3790 struct vcpu_vmx *vmx = to_vmx(vcpu);
3791 unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3793 if (!cpu_has_vmx_msr_bitmap())
3796 vmx_msr_bitmap_l01_changed(vmx);
3799 * Mark the desired intercept state in shadow bitmap, this is needed
3800 * for resync when the MSR filter changes.
3802 if (is_valid_passthrough_msr(msr)) {
3803 int idx = possible_passthrough_msr_slot(msr);
3805 if (idx != -ENOENT) {
3806 if (type & MSR_TYPE_R)
3807 set_bit(idx, vmx->shadow_msr_intercept.read);
3808 if (type & MSR_TYPE_W)
3809 set_bit(idx, vmx->shadow_msr_intercept.write);
3813 if (type & MSR_TYPE_R)
3814 vmx_set_msr_bitmap_read(msr_bitmap, msr);
3816 if (type & MSR_TYPE_W)
3817 vmx_set_msr_bitmap_write(msr_bitmap, msr);
3820 static void vmx_reset_x2apic_msrs(struct kvm_vcpu *vcpu, u8 mode)
3822 unsigned long *msr_bitmap = to_vmx(vcpu)->vmcs01.msr_bitmap;
3823 unsigned long read_intercept;
3826 read_intercept = (mode & MSR_BITMAP_MODE_X2APIC_APICV) ? 0 : ~0;
3828 for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
3829 unsigned int read_idx = msr / BITS_PER_LONG;
3830 unsigned int write_idx = read_idx + (0x800 / sizeof(long));
3832 msr_bitmap[read_idx] = read_intercept;
3833 msr_bitmap[write_idx] = ~0ul;
3837 static void vmx_update_msr_bitmap_x2apic(struct kvm_vcpu *vcpu)
3839 struct vcpu_vmx *vmx = to_vmx(vcpu);
3842 if (!cpu_has_vmx_msr_bitmap())
3845 if (cpu_has_secondary_exec_ctrls() &&
3846 (secondary_exec_controls_get(vmx) &
3847 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
3848 mode = MSR_BITMAP_MODE_X2APIC;
3849 if (enable_apicv && kvm_vcpu_apicv_active(vcpu))
3850 mode |= MSR_BITMAP_MODE_X2APIC_APICV;
3855 if (mode == vmx->x2apic_msr_bitmap_mode)
3858 vmx->x2apic_msr_bitmap_mode = mode;
3860 vmx_reset_x2apic_msrs(vcpu, mode);
3863 * TPR reads and writes can be virtualized even if virtual interrupt
3864 * delivery is not in use.
3866 vmx_set_intercept_for_msr(vcpu, X2APIC_MSR(APIC_TASKPRI), MSR_TYPE_RW,
3867 !(mode & MSR_BITMAP_MODE_X2APIC));
3869 if (mode & MSR_BITMAP_MODE_X2APIC_APICV) {
3870 vmx_enable_intercept_for_msr(vcpu, X2APIC_MSR(APIC_TMCCT), MSR_TYPE_RW);
3871 vmx_disable_intercept_for_msr(vcpu, X2APIC_MSR(APIC_EOI), MSR_TYPE_W);
3872 vmx_disable_intercept_for_msr(vcpu, X2APIC_MSR(APIC_SELF_IPI), MSR_TYPE_W);
3876 void pt_update_intercept_for_msr(struct kvm_vcpu *vcpu)
3878 struct vcpu_vmx *vmx = to_vmx(vcpu);
3879 bool flag = !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
3882 vmx_set_intercept_for_msr(vcpu, MSR_IA32_RTIT_STATUS, MSR_TYPE_RW, flag);
3883 vmx_set_intercept_for_msr(vcpu, MSR_IA32_RTIT_OUTPUT_BASE, MSR_TYPE_RW, flag);
3884 vmx_set_intercept_for_msr(vcpu, MSR_IA32_RTIT_OUTPUT_MASK, MSR_TYPE_RW, flag);
3885 vmx_set_intercept_for_msr(vcpu, MSR_IA32_RTIT_CR3_MATCH, MSR_TYPE_RW, flag);
3886 for (i = 0; i < vmx->pt_desc.num_address_ranges; i++) {
3887 vmx_set_intercept_for_msr(vcpu, MSR_IA32_RTIT_ADDR0_A + i * 2, MSR_TYPE_RW, flag);
3888 vmx_set_intercept_for_msr(vcpu, MSR_IA32_RTIT_ADDR0_B + i * 2, MSR_TYPE_RW, flag);
3892 static bool vmx_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
3894 struct vcpu_vmx *vmx = to_vmx(vcpu);
3899 if (WARN_ON_ONCE(!is_guest_mode(vcpu)) ||
3900 !nested_cpu_has_vid(get_vmcs12(vcpu)) ||
3901 WARN_ON_ONCE(!vmx->nested.virtual_apic_map.gfn))
3904 rvi = vmx_get_rvi();
3906 vapic_page = vmx->nested.virtual_apic_map.hva;
3907 vppr = *((u32 *)(vapic_page + APIC_PROCPRI));
3909 return ((rvi & 0xf0) > (vppr & 0xf0));
3912 static void vmx_msr_filter_changed(struct kvm_vcpu *vcpu)
3914 struct vcpu_vmx *vmx = to_vmx(vcpu);
3918 * Set intercept permissions for all potentially passed through MSRs
3919 * again. They will automatically get filtered through the MSR filter,
3920 * so we are back in sync after this.
3922 for (i = 0; i < ARRAY_SIZE(vmx_possible_passthrough_msrs); i++) {
3923 u32 msr = vmx_possible_passthrough_msrs[i];
3924 bool read = test_bit(i, vmx->shadow_msr_intercept.read);
3925 bool write = test_bit(i, vmx->shadow_msr_intercept.write);
3927 vmx_set_intercept_for_msr(vcpu, msr, MSR_TYPE_R, read);
3928 vmx_set_intercept_for_msr(vcpu, msr, MSR_TYPE_W, write);
3931 pt_update_intercept_for_msr(vcpu);
3934 static inline void kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu,
3938 if (vcpu->mode == IN_GUEST_MODE) {
3940 * The vector of interrupt to be delivered to vcpu had
3941 * been set in PIR before this function.
3943 * Following cases will be reached in this block, and
3944 * we always send a notification event in all cases as
3947 * Case 1: vcpu keeps in non-root mode. Sending a
3948 * notification event posts the interrupt to vcpu.
3950 * Case 2: vcpu exits to root mode and is still
3951 * runnable. PIR will be synced to vIRR before the
3952 * next vcpu entry. Sending a notification event in
3953 * this case has no effect, as vcpu is not in root
3956 * Case 3: vcpu exits to root mode and is blocked.
3957 * vcpu_block() has already synced PIR to vIRR and
3958 * never blocks vcpu if vIRR is not cleared. Therefore,
3959 * a blocked vcpu here does not wait for any requested
3960 * interrupts in PIR, and sending a notification event
3961 * which has no effect is safe here.
3964 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu), pi_vec);
3969 * The vCPU isn't in the guest; wake the vCPU in case it is blocking,
3970 * otherwise do nothing as KVM will grab the highest priority pending
3971 * IRQ via ->sync_pir_to_irr() in vcpu_enter_guest().
3973 kvm_vcpu_wake_up(vcpu);
3976 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
3979 struct vcpu_vmx *vmx = to_vmx(vcpu);
3981 if (is_guest_mode(vcpu) &&
3982 vector == vmx->nested.posted_intr_nv) {
3984 * If a posted intr is not recognized by hardware,
3985 * we will accomplish it in the next vmentry.
3987 vmx->nested.pi_pending = true;
3988 kvm_make_request(KVM_REQ_EVENT, vcpu);
3991 * This pairs with the smp_mb_*() after setting vcpu->mode in
3992 * vcpu_enter_guest() to guarantee the vCPU sees the event
3993 * request if triggering a posted interrupt "fails" because
3994 * vcpu->mode != IN_GUEST_MODE. The extra barrier is needed as
3995 * the smb_wmb() in kvm_make_request() only ensures everything
3996 * done before making the request is visible when the request
3997 * is visible, it doesn't ensure ordering between the store to
3998 * vcpu->requests and the load from vcpu->mode.
4000 smp_mb__after_atomic();
4002 /* the PIR and ON have been set by L1. */
4003 kvm_vcpu_trigger_posted_interrupt(vcpu, POSTED_INTR_NESTED_VECTOR);
4009 * Send interrupt to vcpu via posted interrupt way.
4010 * 1. If target vcpu is running(non-root mode), send posted interrupt
4011 * notification to vcpu and hardware will sync PIR to vIRR atomically.
4012 * 2. If target vcpu isn't running(root mode), kick it to pick up the
4013 * interrupt from PIR in next vmentry.
4015 static int vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
4017 struct vcpu_vmx *vmx = to_vmx(vcpu);
4020 r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
4024 if (!vcpu->arch.apicv_active)
4027 if (pi_test_and_set_pir(vector, &vmx->pi_desc))
4030 /* If a previous notification has sent the IPI, nothing to do. */
4031 if (pi_test_and_set_on(&vmx->pi_desc))
4035 * The implied barrier in pi_test_and_set_on() pairs with the smp_mb_*()
4036 * after setting vcpu->mode in vcpu_enter_guest(), thus the vCPU is
4037 * guaranteed to see PID.ON=1 and sync the PIR to IRR if triggering a
4038 * posted interrupt "fails" because vcpu->mode != IN_GUEST_MODE.
4040 kvm_vcpu_trigger_posted_interrupt(vcpu, POSTED_INTR_VECTOR);
4045 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
4046 * will not change in the lifetime of the guest.
4047 * Note that host-state that does change is set elsewhere. E.g., host-state
4048 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
4050 void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
4054 unsigned long cr0, cr3, cr4;
4057 WARN_ON(cr0 & X86_CR0_TS);
4058 vmcs_writel(HOST_CR0, cr0); /* 22.2.3 */
4061 * Save the most likely value for this task's CR3 in the VMCS.
4062 * We can't use __get_current_cr3_fast() because we're not atomic.
4065 vmcs_writel(HOST_CR3, cr3); /* 22.2.3 FIXME: shadow tables */
4066 vmx->loaded_vmcs->host_state.cr3 = cr3;
4068 /* Save the most likely value for this task's CR4 in the VMCS. */
4069 cr4 = cr4_read_shadow();
4070 vmcs_writel(HOST_CR4, cr4); /* 22.2.3, 22.2.5 */
4071 vmx->loaded_vmcs->host_state.cr4 = cr4;
4073 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
4074 #ifdef CONFIG_X86_64
4076 * Load null selectors, so we can avoid reloading them in
4077 * vmx_prepare_switch_to_host(), in case userspace uses
4078 * the null selectors too (the expected case).
4080 vmcs_write16(HOST_DS_SELECTOR, 0);
4081 vmcs_write16(HOST_ES_SELECTOR, 0);
4083 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4084 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4086 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4087 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
4089 vmcs_writel(HOST_IDTR_BASE, host_idt_base); /* 22.2.4 */
4091 vmcs_writel(HOST_RIP, (unsigned long)vmx_vmexit); /* 22.2.5 */
4093 rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
4094 vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
4097 * SYSENTER is used for 32-bit system calls on either 32-bit or
4098 * 64-bit kernels. It is always zero If neither is allowed, otherwise
4099 * vmx_vcpu_load_vmcs loads it with the per-CPU entry stack (and may
4100 * have already done so!).
4102 if (!IS_ENABLED(CONFIG_IA32_EMULATION) && !IS_ENABLED(CONFIG_X86_32))
4103 vmcs_writel(HOST_IA32_SYSENTER_ESP, 0);
4105 rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
4106 vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */
4108 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
4109 rdmsr(MSR_IA32_CR_PAT, low32, high32);
4110 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
4113 if (cpu_has_load_ia32_efer())
4114 vmcs_write64(HOST_IA32_EFER, host_efer);
4117 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
4119 struct kvm_vcpu *vcpu = &vmx->vcpu;
4121 vcpu->arch.cr4_guest_owned_bits = KVM_POSSIBLE_CR4_GUEST_BITS &
4122 ~vcpu->arch.cr4_guest_rsvd_bits;
4124 vcpu->arch.cr4_guest_owned_bits &= ~X86_CR4_TLBFLUSH_BITS;
4125 vcpu->arch.cr4_guest_owned_bits &= ~X86_CR4_PDPTR_BITS;
4127 if (is_guest_mode(&vmx->vcpu))
4128 vcpu->arch.cr4_guest_owned_bits &=
4129 ~get_vmcs12(vcpu)->cr4_guest_host_mask;
4130 vmcs_writel(CR4_GUEST_HOST_MASK, ~vcpu->arch.cr4_guest_owned_bits);
4133 static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
4135 u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
4137 if (!kvm_vcpu_apicv_active(&vmx->vcpu))
4138 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
4141 pin_based_exec_ctrl &= ~PIN_BASED_VIRTUAL_NMIS;
4143 if (!enable_preemption_timer)
4144 pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
4146 return pin_based_exec_ctrl;
4149 static u32 vmx_vmentry_ctrl(void)
4151 u32 vmentry_ctrl = vmcs_config.vmentry_ctrl;
4153 if (vmx_pt_mode_is_system())
4154 vmentry_ctrl &= ~(VM_ENTRY_PT_CONCEAL_PIP |
4155 VM_ENTRY_LOAD_IA32_RTIT_CTL);
4156 /* Loading of EFER and PERF_GLOBAL_CTRL are toggled dynamically */
4157 return vmentry_ctrl &
4158 ~(VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL | VM_ENTRY_LOAD_IA32_EFER);
4161 static u32 vmx_vmexit_ctrl(void)
4163 u32 vmexit_ctrl = vmcs_config.vmexit_ctrl;
4165 if (vmx_pt_mode_is_system())
4166 vmexit_ctrl &= ~(VM_EXIT_PT_CONCEAL_PIP |
4167 VM_EXIT_CLEAR_IA32_RTIT_CTL);
4168 /* Loading of EFER and PERF_GLOBAL_CTRL are toggled dynamically */
4169 return vmexit_ctrl &
4170 ~(VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL | VM_EXIT_LOAD_IA32_EFER);
4173 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
4175 struct vcpu_vmx *vmx = to_vmx(vcpu);
4177 pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
4178 if (cpu_has_secondary_exec_ctrls()) {
4179 if (kvm_vcpu_apicv_active(vcpu))
4180 secondary_exec_controls_setbit(vmx,
4181 SECONDARY_EXEC_APIC_REGISTER_VIRT |
4182 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4184 secondary_exec_controls_clearbit(vmx,
4185 SECONDARY_EXEC_APIC_REGISTER_VIRT |
4186 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4189 vmx_update_msr_bitmap_x2apic(vcpu);
4192 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
4194 u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
4196 if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
4197 exec_control &= ~CPU_BASED_MOV_DR_EXITING;
4199 if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
4200 exec_control &= ~CPU_BASED_TPR_SHADOW;
4201 #ifdef CONFIG_X86_64
4202 exec_control |= CPU_BASED_CR8_STORE_EXITING |
4203 CPU_BASED_CR8_LOAD_EXITING;
4207 exec_control |= CPU_BASED_CR3_STORE_EXITING |
4208 CPU_BASED_CR3_LOAD_EXITING |
4209 CPU_BASED_INVLPG_EXITING;
4210 if (kvm_mwait_in_guest(vmx->vcpu.kvm))
4211 exec_control &= ~(CPU_BASED_MWAIT_EXITING |
4212 CPU_BASED_MONITOR_EXITING);
4213 if (kvm_hlt_in_guest(vmx->vcpu.kvm))
4214 exec_control &= ~CPU_BASED_HLT_EXITING;
4215 return exec_control;
4219 * Adjust a single secondary execution control bit to intercept/allow an
4220 * instruction in the guest. This is usually done based on whether or not a
4221 * feature has been exposed to the guest in order to correctly emulate faults.
4224 vmx_adjust_secondary_exec_control(struct vcpu_vmx *vmx, u32 *exec_control,
4225 u32 control, bool enabled, bool exiting)
4228 * If the control is for an opt-in feature, clear the control if the
4229 * feature is not exposed to the guest, i.e. not enabled. If the
4230 * control is opt-out, i.e. an exiting control, clear the control if
4231 * the feature _is_ exposed to the guest, i.e. exiting/interception is
4232 * disabled for the associated instruction. Note, the caller is
4233 * responsible presetting exec_control to set all supported bits.
4235 if (enabled == exiting)
4236 *exec_control &= ~control;
4239 * Update the nested MSR settings so that a nested VMM can/can't set
4240 * controls for features that are/aren't exposed to the guest.
4244 vmx->nested.msrs.secondary_ctls_high |= control;
4246 vmx->nested.msrs.secondary_ctls_high &= ~control;
4251 * Wrapper macro for the common case of adjusting a secondary execution control
4252 * based on a single guest CPUID bit, with a dedicated feature bit. This also
4253 * verifies that the control is actually supported by KVM and hardware.
4255 #define vmx_adjust_sec_exec_control(vmx, exec_control, name, feat_name, ctrl_name, exiting) \
4259 if (cpu_has_vmx_##name()) { \
4260 __enabled = guest_cpuid_has(&(vmx)->vcpu, \
4261 X86_FEATURE_##feat_name); \
4262 vmx_adjust_secondary_exec_control(vmx, exec_control, \
4263 SECONDARY_EXEC_##ctrl_name, __enabled, exiting); \
4267 /* More macro magic for ENABLE_/opt-in versus _EXITING/opt-out controls. */
4268 #define vmx_adjust_sec_exec_feature(vmx, exec_control, lname, uname) \
4269 vmx_adjust_sec_exec_control(vmx, exec_control, lname, uname, ENABLE_##uname, false)
4271 #define vmx_adjust_sec_exec_exiting(vmx, exec_control, lname, uname) \
4272 vmx_adjust_sec_exec_control(vmx, exec_control, lname, uname, uname##_EXITING, true)
4274 static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
4276 struct kvm_vcpu *vcpu = &vmx->vcpu;
4278 u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
4280 if (vmx_pt_mode_is_system())
4281 exec_control &= ~(SECONDARY_EXEC_PT_USE_GPA | SECONDARY_EXEC_PT_CONCEAL_VMX);
4282 if (!cpu_need_virtualize_apic_accesses(vcpu))
4283 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
4285 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
4287 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
4288 enable_unrestricted_guest = 0;
4290 if (!enable_unrestricted_guest)
4291 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4292 if (kvm_pause_in_guest(vmx->vcpu.kvm))
4293 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
4294 if (!kvm_vcpu_apicv_active(vcpu))
4295 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
4296 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4297 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
4299 /* SECONDARY_EXEC_DESC is enabled/disabled on writes to CR4.UMIP,
4300 * in vmx_set_cr4. */
4301 exec_control &= ~SECONDARY_EXEC_DESC;
4303 /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4305 We can NOT enable shadow_vmcs here because we don't have yet
4308 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4311 * PML is enabled/disabled when dirty logging of memsmlots changes, but
4312 * it needs to be set here when dirty logging is already active, e.g.
4313 * if this vCPU was created after dirty logging was enabled.
4315 if (!vcpu->kvm->arch.cpu_dirty_logging_count)
4316 exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
4318 if (cpu_has_vmx_xsaves()) {
4319 /* Exposing XSAVES only when XSAVE is exposed */
4320 bool xsaves_enabled =
4321 boot_cpu_has(X86_FEATURE_XSAVE) &&
4322 guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
4323 guest_cpuid_has(vcpu, X86_FEATURE_XSAVES);
4325 vcpu->arch.xsaves_enabled = xsaves_enabled;
4327 vmx_adjust_secondary_exec_control(vmx, &exec_control,
4328 SECONDARY_EXEC_XSAVES,
4329 xsaves_enabled, false);
4333 * RDPID is also gated by ENABLE_RDTSCP, turn on the control if either
4334 * feature is exposed to the guest. This creates a virtualization hole
4335 * if both are supported in hardware but only one is exposed to the
4336 * guest, but letting the guest execute RDTSCP or RDPID when either one
4337 * is advertised is preferable to emulating the advertised instruction
4338 * in KVM on #UD, and obviously better than incorrectly injecting #UD.
4340 if (cpu_has_vmx_rdtscp()) {
4341 bool rdpid_or_rdtscp_enabled =
4342 guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) ||
4343 guest_cpuid_has(vcpu, X86_FEATURE_RDPID);
4345 vmx_adjust_secondary_exec_control(vmx, &exec_control,
4346 SECONDARY_EXEC_ENABLE_RDTSCP,
4347 rdpid_or_rdtscp_enabled, false);
4349 vmx_adjust_sec_exec_feature(vmx, &exec_control, invpcid, INVPCID);
4351 vmx_adjust_sec_exec_exiting(vmx, &exec_control, rdrand, RDRAND);
4352 vmx_adjust_sec_exec_exiting(vmx, &exec_control, rdseed, RDSEED);
4354 vmx_adjust_sec_exec_control(vmx, &exec_control, waitpkg, WAITPKG,
4355 ENABLE_USR_WAIT_PAUSE, false);
4357 if (!vcpu->kvm->arch.bus_lock_detection_enabled)
4358 exec_control &= ~SECONDARY_EXEC_BUS_LOCK_DETECTION;
4360 return exec_control;
4363 #define VMX_XSS_EXIT_BITMAP 0
4365 static void init_vmcs(struct vcpu_vmx *vmx)
4368 nested_vmx_set_vmcs_shadowing_bitmap();
4370 if (cpu_has_vmx_msr_bitmap())
4371 vmcs_write64(MSR_BITMAP, __pa(vmx->vmcs01.msr_bitmap));
4373 vmcs_write64(VMCS_LINK_POINTER, INVALID_GPA); /* 22.3.1.5 */
4376 pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
4378 exec_controls_set(vmx, vmx_exec_control(vmx));
4380 if (cpu_has_secondary_exec_ctrls())
4381 secondary_exec_controls_set(vmx, vmx_secondary_exec_control(vmx));
4383 if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
4384 vmcs_write64(EOI_EXIT_BITMAP0, 0);
4385 vmcs_write64(EOI_EXIT_BITMAP1, 0);
4386 vmcs_write64(EOI_EXIT_BITMAP2, 0);
4387 vmcs_write64(EOI_EXIT_BITMAP3, 0);
4389 vmcs_write16(GUEST_INTR_STATUS, 0);
4391 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4392 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4395 if (!kvm_pause_in_guest(vmx->vcpu.kvm)) {
4396 vmcs_write32(PLE_GAP, ple_gap);
4397 vmx->ple_window = ple_window;
4398 vmx->ple_window_dirty = true;
4401 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4402 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4403 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
4405 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
4406 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
4407 vmx_set_constant_host_state(vmx);
4408 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4409 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4411 if (cpu_has_vmx_vmfunc())
4412 vmcs_write64(VM_FUNCTION_CONTROL, 0);
4414 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4415 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4416 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
4417 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4418 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
4420 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
4421 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
4423 vm_exit_controls_set(vmx, vmx_vmexit_ctrl());
4425 /* 22.2.1, 20.8.1 */
4426 vm_entry_controls_set(vmx, vmx_vmentry_ctrl());
4428 vmx->vcpu.arch.cr0_guest_owned_bits = KVM_POSSIBLE_CR0_GUEST_BITS;
4429 vmcs_writel(CR0_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr0_guest_owned_bits);
4431 set_cr4_guest_host_mask(vmx);
4434 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4436 if (cpu_has_vmx_xsaves())
4437 vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
4440 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
4441 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
4444 vmx_write_encls_bitmap(&vmx->vcpu, NULL);
4446 if (vmx_pt_mode_is_host_guest()) {
4447 memset(&vmx->pt_desc, 0, sizeof(vmx->pt_desc));
4448 /* Bit[6~0] are forced to 1, writes are ignored. */
4449 vmx->pt_desc.guest.output_mask = 0x7F;
4450 vmcs_write64(GUEST_IA32_RTIT_CTL, 0);
4453 vmcs_write32(GUEST_SYSENTER_CS, 0);
4454 vmcs_writel(GUEST_SYSENTER_ESP, 0);
4455 vmcs_writel(GUEST_SYSENTER_EIP, 0);
4456 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4458 if (cpu_has_vmx_tpr_shadow()) {
4459 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4460 if (cpu_need_tpr_shadow(&vmx->vcpu))
4461 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4462 __pa(vmx->vcpu.arch.apic->regs));
4463 vmcs_write32(TPR_THRESHOLD, 0);
4466 vmx_setup_uret_msrs(vmx);
4469 static void __vmx_vcpu_reset(struct kvm_vcpu *vcpu)
4471 struct vcpu_vmx *vmx = to_vmx(vcpu);
4476 memcpy(&vmx->nested.msrs, &vmcs_config.nested, sizeof(vmx->nested.msrs));
4478 vcpu_setup_sgx_lepubkeyhash(vcpu);
4480 vmx->nested.posted_intr_nv = -1;
4481 vmx->nested.vmxon_ptr = INVALID_GPA;
4482 vmx->nested.current_vmptr = INVALID_GPA;
4483 vmx->nested.hv_evmcs_vmptr = EVMPTR_INVALID;
4485 vcpu->arch.microcode_version = 0x100000000ULL;
4486 vmx->msr_ia32_feature_control_valid_bits = FEAT_CTL_LOCKED;
4489 * Enforce invariant: pi_desc.nv is always either POSTED_INTR_VECTOR
4490 * or POSTED_INTR_WAKEUP_VECTOR.
4492 vmx->pi_desc.nv = POSTED_INTR_VECTOR;
4493 vmx->pi_desc.sn = 1;
4496 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
4498 struct vcpu_vmx *vmx = to_vmx(vcpu);
4501 __vmx_vcpu_reset(vcpu);
4503 vmx->rmode.vm86_active = 0;
4506 vmx->msr_ia32_umwait_control = 0;
4508 vmx->hv_deadline_tsc = -1;
4509 kvm_set_cr8(vcpu, 0);
4511 vmx_segment_cache_clear(vmx);
4512 kvm_register_mark_available(vcpu, VCPU_EXREG_SEGMENTS);
4514 seg_setup(VCPU_SREG_CS);
4515 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4516 vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
4518 seg_setup(VCPU_SREG_DS);
4519 seg_setup(VCPU_SREG_ES);
4520 seg_setup(VCPU_SREG_FS);
4521 seg_setup(VCPU_SREG_GS);
4522 seg_setup(VCPU_SREG_SS);
4524 vmcs_write16(GUEST_TR_SELECTOR, 0);
4525 vmcs_writel(GUEST_TR_BASE, 0);
4526 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4527 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4529 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4530 vmcs_writel(GUEST_LDTR_BASE, 0);
4531 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4532 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4534 vmcs_writel(GUEST_GDTR_BASE, 0);
4535 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4537 vmcs_writel(GUEST_IDTR_BASE, 0);
4538 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4540 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4541 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4542 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4543 if (kvm_mpx_supported())
4544 vmcs_write64(GUEST_BNDCFGS, 0);
4546 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
4548 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4550 vpid_sync_context(vmx->vpid);
4553 static void vmx_enable_irq_window(struct kvm_vcpu *vcpu)
4555 exec_controls_setbit(to_vmx(vcpu), CPU_BASED_INTR_WINDOW_EXITING);
4558 static void vmx_enable_nmi_window(struct kvm_vcpu *vcpu)
4561 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4562 vmx_enable_irq_window(vcpu);
4566 exec_controls_setbit(to_vmx(vcpu), CPU_BASED_NMI_WINDOW_EXITING);
4569 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4571 struct vcpu_vmx *vmx = to_vmx(vcpu);
4573 int irq = vcpu->arch.interrupt.nr;
4575 trace_kvm_inj_virq(irq);
4577 ++vcpu->stat.irq_injections;
4578 if (vmx->rmode.vm86_active) {
4580 if (vcpu->arch.interrupt.soft)
4581 inc_eip = vcpu->arch.event_exit_inst_len;
4582 kvm_inject_realmode_interrupt(vcpu, irq, inc_eip);
4585 intr = irq | INTR_INFO_VALID_MASK;
4586 if (vcpu->arch.interrupt.soft) {
4587 intr |= INTR_TYPE_SOFT_INTR;
4588 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4589 vmx->vcpu.arch.event_exit_inst_len);
4591 intr |= INTR_TYPE_EXT_INTR;
4592 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4594 vmx_clear_hlt(vcpu);
4597 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4599 struct vcpu_vmx *vmx = to_vmx(vcpu);
4603 * Tracking the NMI-blocked state in software is built upon
4604 * finding the next open IRQ window. This, in turn, depends on
4605 * well-behaving guests: They have to keep IRQs disabled at
4606 * least as long as the NMI handler runs. Otherwise we may
4607 * cause NMI nesting, maybe breaking the guest. But as this is
4608 * highly unlikely, we can live with the residual risk.
4610 vmx->loaded_vmcs->soft_vnmi_blocked = 1;
4611 vmx->loaded_vmcs->vnmi_blocked_time = 0;
4614 ++vcpu->stat.nmi_injections;
4615 vmx->loaded_vmcs->nmi_known_unmasked = false;
4617 if (vmx->rmode.vm86_active) {
4618 kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0);
4622 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4623 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4625 vmx_clear_hlt(vcpu);
4628 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4630 struct vcpu_vmx *vmx = to_vmx(vcpu);
4634 return vmx->loaded_vmcs->soft_vnmi_blocked;
4635 if (vmx->loaded_vmcs->nmi_known_unmasked)
4637 masked = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4638 vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4642 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4644 struct vcpu_vmx *vmx = to_vmx(vcpu);
4647 if (vmx->loaded_vmcs->soft_vnmi_blocked != masked) {
4648 vmx->loaded_vmcs->soft_vnmi_blocked = masked;
4649 vmx->loaded_vmcs->vnmi_blocked_time = 0;
4652 vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4654 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4655 GUEST_INTR_STATE_NMI);
4657 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4658 GUEST_INTR_STATE_NMI);
4662 bool vmx_nmi_blocked(struct kvm_vcpu *vcpu)
4664 if (is_guest_mode(vcpu) && nested_exit_on_nmi(vcpu))
4667 if (!enable_vnmi && to_vmx(vcpu)->loaded_vmcs->soft_vnmi_blocked)
4670 return (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4671 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI |
4672 GUEST_INTR_STATE_NMI));
4675 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu, bool for_injection)
4677 if (to_vmx(vcpu)->nested.nested_run_pending)
4680 /* An NMI must not be injected into L2 if it's supposed to VM-Exit. */
4681 if (for_injection && is_guest_mode(vcpu) && nested_exit_on_nmi(vcpu))
4684 return !vmx_nmi_blocked(vcpu);
4687 bool vmx_interrupt_blocked(struct kvm_vcpu *vcpu)
4689 if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu))
4692 return !(vmx_get_rflags(vcpu) & X86_EFLAGS_IF) ||
4693 (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4694 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4697 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu, bool for_injection)
4699 if (to_vmx(vcpu)->nested.nested_run_pending)
4703 * An IRQ must not be injected into L2 if it's supposed to VM-Exit,
4704 * e.g. if the IRQ arrived asynchronously after checking nested events.
4706 if (for_injection && is_guest_mode(vcpu) && nested_exit_on_intr(vcpu))
4709 return !vmx_interrupt_blocked(vcpu);
4712 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4716 if (enable_unrestricted_guest)
4719 mutex_lock(&kvm->slots_lock);
4720 ret = __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
4722 mutex_unlock(&kvm->slots_lock);
4725 return PTR_ERR(ret);
4727 to_kvm_vmx(kvm)->tss_addr = addr;
4729 return init_rmode_tss(kvm, ret);
4732 static int vmx_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
4734 to_kvm_vmx(kvm)->ept_identity_map_addr = ident_addr;
4738 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
4743 * Update instruction length as we may reinject the exception
4744 * from user space while in guest debugging mode.
4746 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4747 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4748 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4752 return !(vcpu->guest_debug &
4753 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP));
4767 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4768 int vec, u32 err_code)
4771 * Instruction with address size override prefix opcode 0x67
4772 * Cause the #SS fault with 0 error code in VM86 mode.
4774 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
4775 if (kvm_emulate_instruction(vcpu, 0)) {
4776 if (vcpu->arch.halt_request) {
4777 vcpu->arch.halt_request = 0;
4778 return kvm_emulate_halt_noskip(vcpu);
4786 * Forward all other exceptions that are valid in real mode.
4787 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4788 * the required debugging infrastructure rework.
4790 kvm_queue_exception(vcpu, vec);
4794 static int handle_machine_check(struct kvm_vcpu *vcpu)
4796 /* handled by vmx_vcpu_run() */
4801 * If the host has split lock detection disabled, then #AC is
4802 * unconditionally injected into the guest, which is the pre split lock
4803 * detection behaviour.
4805 * If the host has split lock detection enabled then #AC is
4806 * only injected into the guest when:
4807 * - Guest CPL == 3 (user mode)
4808 * - Guest has #AC detection enabled in CR0
4809 * - Guest EFLAGS has AC bit set
4811 bool vmx_guest_inject_ac(struct kvm_vcpu *vcpu)
4813 if (!boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
4816 return vmx_get_cpl(vcpu) == 3 && kvm_read_cr0_bits(vcpu, X86_CR0_AM) &&
4817 (kvm_get_rflags(vcpu) & X86_EFLAGS_AC);
4820 static int handle_exception_nmi(struct kvm_vcpu *vcpu)
4822 struct vcpu_vmx *vmx = to_vmx(vcpu);
4823 struct kvm_run *kvm_run = vcpu->run;
4824 u32 intr_info, ex_no, error_code;
4825 unsigned long cr2, dr6;
4828 vect_info = vmx->idt_vectoring_info;
4829 intr_info = vmx_get_intr_info(vcpu);
4831 if (is_machine_check(intr_info) || is_nmi(intr_info))
4832 return 1; /* handled by handle_exception_nmi_irqoff() */
4835 * Queue the exception here instead of in handle_nm_fault_irqoff().
4836 * This ensures the nested_vmx check is not skipped so vmexit can
4837 * be reflected to L1 (when it intercepts #NM) before reaching this
4840 if (is_nm_fault(intr_info)) {
4841 kvm_queue_exception(vcpu, NM_VECTOR);
4845 if (is_invalid_opcode(intr_info))
4846 return handle_ud(vcpu);
4849 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4850 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4852 if (!vmx->rmode.vm86_active && is_gp_fault(intr_info)) {
4853 WARN_ON_ONCE(!enable_vmware_backdoor);
4856 * VMware backdoor emulation on #GP interception only handles
4857 * IN{S}, OUT{S}, and RDPMC, none of which generate a non-zero
4858 * error code on #GP.
4861 kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
4864 return kvm_emulate_instruction(vcpu, EMULTYPE_VMWARE_GP);
4868 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
4869 * MMIO, it is better to report an internal error.
4870 * See the comments in vmx_handle_exit.
4872 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4873 !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
4874 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4875 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4876 vcpu->run->internal.ndata = 4;
4877 vcpu->run->internal.data[0] = vect_info;
4878 vcpu->run->internal.data[1] = intr_info;
4879 vcpu->run->internal.data[2] = error_code;
4880 vcpu->run->internal.data[3] = vcpu->arch.last_vmentry_cpu;
4884 if (is_page_fault(intr_info)) {
4885 cr2 = vmx_get_exit_qual(vcpu);
4886 if (enable_ept && !vcpu->arch.apf.host_apf_flags) {
4888 * EPT will cause page fault only if we need to
4889 * detect illegal GPAs.
4891 WARN_ON_ONCE(!allow_smaller_maxphyaddr);
4892 kvm_fixup_and_inject_pf_error(vcpu, cr2, error_code);
4895 return kvm_handle_page_fault(vcpu, error_code, cr2, NULL, 0);
4898 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4900 if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
4901 return handle_rmode_exception(vcpu, ex_no, error_code);
4905 dr6 = vmx_get_exit_qual(vcpu);
4906 if (!(vcpu->guest_debug &
4907 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4909 * If the #DB was due to ICEBP, a.k.a. INT1, skip the
4910 * instruction. ICEBP generates a trap-like #DB, but
4911 * despite its interception control being tied to #DB,
4912 * is an instruction intercept, i.e. the VM-Exit occurs
4913 * on the ICEBP itself. Note, skipping ICEBP also
4914 * clears STI and MOVSS blocking.
4916 * For all other #DBs, set vmcs.PENDING_DBG_EXCEPTIONS.BS
4917 * if single-step is enabled in RFLAGS and STI or MOVSS
4918 * blocking is active, as the CPU doesn't set the bit
4919 * on VM-Exit due to #DB interception. VM-Entry has a
4920 * consistency check that a single-step #DB is pending
4921 * in this scenario as the previous instruction cannot
4922 * have toggled RFLAGS.TF 0=>1 (because STI and POP/MOV
4923 * don't modify RFLAGS), therefore the one instruction
4924 * delay when activating single-step breakpoints must
4925 * have already expired. Note, the CPU sets/clears BS
4926 * as appropriate for all other VM-Exits types.
4928 if (is_icebp(intr_info))
4929 WARN_ON(!skip_emulated_instruction(vcpu));
4930 else if ((vmx_get_rflags(vcpu) & X86_EFLAGS_TF) &&
4931 (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4932 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS)))
4933 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
4934 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS) | DR6_BS);
4936 kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
4939 kvm_run->debug.arch.dr6 = dr6 | DR6_ACTIVE_LOW;
4940 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4944 * Update instruction length as we may reinject #BP from
4945 * user space while in guest debugging mode. Reading it for
4946 * #DB as well causes no harm, it is not used in that case.
4948 vmx->vcpu.arch.event_exit_inst_len =
4949 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4950 kvm_run->exit_reason = KVM_EXIT_DEBUG;
4951 kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
4952 kvm_run->debug.arch.exception = ex_no;
4955 if (vmx_guest_inject_ac(vcpu)) {
4956 kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
4961 * Handle split lock. Depending on detection mode this will
4962 * either warn and disable split lock detection for this
4963 * task or force SIGBUS on it.
4965 if (handle_guest_split_lock(kvm_rip_read(vcpu)))
4969 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4970 kvm_run->ex.exception = ex_no;
4971 kvm_run->ex.error_code = error_code;
4977 static __always_inline int handle_external_interrupt(struct kvm_vcpu *vcpu)
4979 ++vcpu->stat.irq_exits;
4983 static int handle_triple_fault(struct kvm_vcpu *vcpu)
4985 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4986 vcpu->mmio_needed = 0;
4990 static int handle_io(struct kvm_vcpu *vcpu)
4992 unsigned long exit_qualification;
4993 int size, in, string;
4996 exit_qualification = vmx_get_exit_qual(vcpu);
4997 string = (exit_qualification & 16) != 0;
4999 ++vcpu->stat.io_exits;
5002 return kvm_emulate_instruction(vcpu, 0);
5004 port = exit_qualification >> 16;
5005 size = (exit_qualification & 7) + 1;
5006 in = (exit_qualification & 8) != 0;
5008 return kvm_fast_pio(vcpu, size, port, in);
5012 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
5015 * Patch in the VMCALL instruction:
5017 hypercall[0] = 0x0f;
5018 hypercall[1] = 0x01;
5019 hypercall[2] = 0xc1;
5022 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
5023 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
5025 if (is_guest_mode(vcpu)) {
5026 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5027 unsigned long orig_val = val;
5030 * We get here when L2 changed cr0 in a way that did not change
5031 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
5032 * but did change L0 shadowed bits. So we first calculate the
5033 * effective cr0 value that L1 would like to write into the
5034 * hardware. It consists of the L2-owned bits from the new
5035 * value combined with the L1-owned bits from L1's guest_cr0.
5037 val = (val & ~vmcs12->cr0_guest_host_mask) |
5038 (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
5040 if (!nested_guest_cr0_valid(vcpu, val))
5043 if (kvm_set_cr0(vcpu, val))
5045 vmcs_writel(CR0_READ_SHADOW, orig_val);
5048 if (to_vmx(vcpu)->nested.vmxon &&
5049 !nested_host_cr0_valid(vcpu, val))
5052 return kvm_set_cr0(vcpu, val);
5056 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
5058 if (is_guest_mode(vcpu)) {
5059 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5060 unsigned long orig_val = val;
5062 /* analogously to handle_set_cr0 */
5063 val = (val & ~vmcs12->cr4_guest_host_mask) |
5064 (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
5065 if (kvm_set_cr4(vcpu, val))
5067 vmcs_writel(CR4_READ_SHADOW, orig_val);
5070 return kvm_set_cr4(vcpu, val);
5073 static int handle_desc(struct kvm_vcpu *vcpu)
5075 WARN_ON(!(vcpu->arch.cr4 & X86_CR4_UMIP));
5076 return kvm_emulate_instruction(vcpu, 0);
5079 static int handle_cr(struct kvm_vcpu *vcpu)
5081 unsigned long exit_qualification, val;
5087 exit_qualification = vmx_get_exit_qual(vcpu);
5088 cr = exit_qualification & 15;
5089 reg = (exit_qualification >> 8) & 15;
5090 switch ((exit_qualification >> 4) & 3) {
5091 case 0: /* mov to cr */
5092 val = kvm_register_read(vcpu, reg);
5093 trace_kvm_cr_write(cr, val);
5096 err = handle_set_cr0(vcpu, val);
5097 return kvm_complete_insn_gp(vcpu, err);
5099 WARN_ON_ONCE(enable_unrestricted_guest);
5101 err = kvm_set_cr3(vcpu, val);
5102 return kvm_complete_insn_gp(vcpu, err);
5104 err = handle_set_cr4(vcpu, val);
5105 return kvm_complete_insn_gp(vcpu, err);
5107 u8 cr8_prev = kvm_get_cr8(vcpu);
5109 err = kvm_set_cr8(vcpu, cr8);
5110 ret = kvm_complete_insn_gp(vcpu, err);
5111 if (lapic_in_kernel(vcpu))
5113 if (cr8_prev <= cr8)
5116 * TODO: we might be squashing a
5117 * KVM_GUESTDBG_SINGLESTEP-triggered
5118 * KVM_EXIT_DEBUG here.
5120 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
5126 KVM_BUG(1, vcpu->kvm, "Guest always owns CR0.TS");
5128 case 1: /*mov from cr*/
5131 WARN_ON_ONCE(enable_unrestricted_guest);
5133 val = kvm_read_cr3(vcpu);
5134 kvm_register_write(vcpu, reg, val);
5135 trace_kvm_cr_read(cr, val);
5136 return kvm_skip_emulated_instruction(vcpu);
5138 val = kvm_get_cr8(vcpu);
5139 kvm_register_write(vcpu, reg, val);
5140 trace_kvm_cr_read(cr, val);
5141 return kvm_skip_emulated_instruction(vcpu);
5145 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5146 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
5147 kvm_lmsw(vcpu, val);
5149 return kvm_skip_emulated_instruction(vcpu);
5153 vcpu->run->exit_reason = 0;
5154 vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
5155 (int)(exit_qualification >> 4) & 3, cr);
5159 static int handle_dr(struct kvm_vcpu *vcpu)
5161 unsigned long exit_qualification;
5165 exit_qualification = vmx_get_exit_qual(vcpu);
5166 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
5168 /* First, if DR does not exist, trigger UD */
5169 if (!kvm_require_dr(vcpu, dr))
5172 if (kvm_x86_ops.get_cpl(vcpu) > 0)
5175 dr7 = vmcs_readl(GUEST_DR7);
5178 * As the vm-exit takes precedence over the debug trap, we
5179 * need to emulate the latter, either for the host or the
5180 * guest debugging itself.
5182 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5183 vcpu->run->debug.arch.dr6 = DR6_BD | DR6_ACTIVE_LOW;
5184 vcpu->run->debug.arch.dr7 = dr7;
5185 vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
5186 vcpu->run->debug.arch.exception = DB_VECTOR;
5187 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
5190 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BD);
5195 if (vcpu->guest_debug == 0) {
5196 exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
5199 * No more DR vmexits; force a reload of the debug registers
5200 * and reenter on this instruction. The next vmexit will
5201 * retrieve the full state of the debug registers.
5203 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
5207 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
5208 if (exit_qualification & TYPE_MOV_FROM_DR) {
5211 kvm_get_dr(vcpu, dr, &val);
5212 kvm_register_write(vcpu, reg, val);
5215 err = kvm_set_dr(vcpu, dr, kvm_register_read(vcpu, reg));
5219 return kvm_complete_insn_gp(vcpu, err);
5222 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
5224 get_debugreg(vcpu->arch.db[0], 0);
5225 get_debugreg(vcpu->arch.db[1], 1);
5226 get_debugreg(vcpu->arch.db[2], 2);
5227 get_debugreg(vcpu->arch.db[3], 3);
5228 get_debugreg(vcpu->arch.dr6, 6);
5229 vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
5231 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
5232 exec_controls_setbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
5235 * exc_debug expects dr6 to be cleared after it runs, avoid that it sees
5236 * a stale dr6 from the guest.
5238 set_debugreg(DR6_RESERVED, 6);
5241 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
5243 vmcs_writel(GUEST_DR7, val);
5246 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
5248 kvm_apic_update_ppr(vcpu);
5252 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
5254 exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_INTR_WINDOW_EXITING);
5256 kvm_make_request(KVM_REQ_EVENT, vcpu);
5258 ++vcpu->stat.irq_window_exits;
5262 static int handle_invlpg(struct kvm_vcpu *vcpu)
5264 unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5266 kvm_mmu_invlpg(vcpu, exit_qualification);
5267 return kvm_skip_emulated_instruction(vcpu);
5270 static int handle_apic_access(struct kvm_vcpu *vcpu)
5272 if (likely(fasteoi)) {
5273 unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5274 int access_type, offset;
5276 access_type = exit_qualification & APIC_ACCESS_TYPE;
5277 offset = exit_qualification & APIC_ACCESS_OFFSET;
5279 * Sane guest uses MOV to write EOI, with written value
5280 * not cared. So make a short-circuit here by avoiding
5281 * heavy instruction emulation.
5283 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5284 (offset == APIC_EOI)) {
5285 kvm_lapic_set_eoi(vcpu);
5286 return kvm_skip_emulated_instruction(vcpu);
5289 return kvm_emulate_instruction(vcpu, 0);
5292 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5294 unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5295 int vector = exit_qualification & 0xff;
5297 /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5298 kvm_apic_set_eoi_accelerated(vcpu, vector);
5302 static int handle_apic_write(struct kvm_vcpu *vcpu)
5304 unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5305 u32 offset = exit_qualification & 0xfff;
5307 /* APIC-write VM exit is trap-like and thus no need to adjust IP */
5308 kvm_apic_write_nodecode(vcpu, offset);
5312 static int handle_task_switch(struct kvm_vcpu *vcpu)
5314 struct vcpu_vmx *vmx = to_vmx(vcpu);
5315 unsigned long exit_qualification;
5316 bool has_error_code = false;
5319 int reason, type, idt_v, idt_index;
5321 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5322 idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5323 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5325 exit_qualification = vmx_get_exit_qual(vcpu);
5327 reason = (u32)exit_qualification >> 30;
5328 if (reason == TASK_SWITCH_GATE && idt_v) {
5330 case INTR_TYPE_NMI_INTR:
5331 vcpu->arch.nmi_injected = false;
5332 vmx_set_nmi_mask(vcpu, true);
5334 case INTR_TYPE_EXT_INTR:
5335 case INTR_TYPE_SOFT_INTR:
5336 kvm_clear_interrupt_queue(vcpu);
5338 case INTR_TYPE_HARD_EXCEPTION:
5339 if (vmx->idt_vectoring_info &
5340 VECTORING_INFO_DELIVER_CODE_MASK) {
5341 has_error_code = true;
5343 vmcs_read32(IDT_VECTORING_ERROR_CODE);
5346 case INTR_TYPE_SOFT_EXCEPTION:
5347 kvm_clear_exception_queue(vcpu);
5353 tss_selector = exit_qualification;
5355 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5356 type != INTR_TYPE_EXT_INTR &&
5357 type != INTR_TYPE_NMI_INTR))
5358 WARN_ON(!skip_emulated_instruction(vcpu));
5361 * TODO: What about debug traps on tss switch?
5362 * Are we supposed to inject them and update dr6?
5364 return kvm_task_switch(vcpu, tss_selector,
5365 type == INTR_TYPE_SOFT_INTR ? idt_index : -1,
5366 reason, has_error_code, error_code);
5369 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5371 unsigned long exit_qualification;
5375 exit_qualification = vmx_get_exit_qual(vcpu);
5378 * EPT violation happened while executing iret from NMI,
5379 * "blocked by NMI" bit has to be set before next VM entry.
5380 * There are errata that may cause this bit to not be set:
5383 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5385 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
5386 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5388 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5389 trace_kvm_page_fault(gpa, exit_qualification);
5391 /* Is it a read fault? */
5392 error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
5393 ? PFERR_USER_MASK : 0;
5394 /* Is it a write fault? */
5395 error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
5396 ? PFERR_WRITE_MASK : 0;
5397 /* Is it a fetch fault? */
5398 error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
5399 ? PFERR_FETCH_MASK : 0;
5400 /* ept page table entry is present? */
5401 error_code |= (exit_qualification &
5402 (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE |
5403 EPT_VIOLATION_EXECUTABLE))
5404 ? PFERR_PRESENT_MASK : 0;
5406 error_code |= (exit_qualification & EPT_VIOLATION_GVA_TRANSLATED) != 0 ?
5407 PFERR_GUEST_FINAL_MASK : PFERR_GUEST_PAGE_MASK;
5409 vcpu->arch.exit_qualification = exit_qualification;
5412 * Check that the GPA doesn't exceed physical memory limits, as that is
5413 * a guest page fault. We have to emulate the instruction here, because
5414 * if the illegal address is that of a paging structure, then
5415 * EPT_VIOLATION_ACC_WRITE bit is set. Alternatively, if supported we
5416 * would also use advanced VM-exit information for EPT violations to
5417 * reconstruct the page fault error code.
5419 if (unlikely(allow_smaller_maxphyaddr && kvm_vcpu_is_illegal_gpa(vcpu, gpa)))
5420 return kvm_emulate_instruction(vcpu, 0);
5422 return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5425 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5429 if (!vmx_can_emulate_instruction(vcpu, EMULTYPE_PF, NULL, 0))
5433 * A nested guest cannot optimize MMIO vmexits, because we have an
5434 * nGPA here instead of the required GPA.
5436 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5437 if (!is_guest_mode(vcpu) &&
5438 !kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5439 trace_kvm_fast_mmio(gpa);
5440 return kvm_skip_emulated_instruction(vcpu);
5443 return kvm_mmu_page_fault(vcpu, gpa, PFERR_RSVD_MASK, NULL, 0);
5446 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5448 if (KVM_BUG_ON(!enable_vnmi, vcpu->kvm))
5451 exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_NMI_WINDOW_EXITING);
5452 ++vcpu->stat.nmi_window_exits;
5453 kvm_make_request(KVM_REQ_EVENT, vcpu);
5458 static bool vmx_emulation_required_with_pending_exception(struct kvm_vcpu *vcpu)
5460 struct vcpu_vmx *vmx = to_vmx(vcpu);
5462 return vmx->emulation_required && !vmx->rmode.vm86_active &&
5463 vcpu->arch.exception.pending;
5466 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5468 struct vcpu_vmx *vmx = to_vmx(vcpu);
5469 bool intr_window_requested;
5470 unsigned count = 130;
5472 intr_window_requested = exec_controls_get(vmx) &
5473 CPU_BASED_INTR_WINDOW_EXITING;
5475 while (vmx->emulation_required && count-- != 0) {
5476 if (intr_window_requested && !vmx_interrupt_blocked(vcpu))
5477 return handle_interrupt_window(&vmx->vcpu);
5479 if (kvm_test_request(KVM_REQ_EVENT, vcpu))
5482 if (!kvm_emulate_instruction(vcpu, 0))
5485 if (vmx_emulation_required_with_pending_exception(vcpu)) {
5486 kvm_prepare_emulation_failure_exit(vcpu);
5490 if (vcpu->arch.halt_request) {
5491 vcpu->arch.halt_request = 0;
5492 return kvm_emulate_halt_noskip(vcpu);
5496 * Note, return 1 and not 0, vcpu_run() will invoke
5497 * xfer_to_guest_mode() which will create a proper return
5500 if (__xfer_to_guest_mode_work_pending())
5507 static int vmx_vcpu_pre_run(struct kvm_vcpu *vcpu)
5509 if (vmx_emulation_required_with_pending_exception(vcpu)) {
5510 kvm_prepare_emulation_failure_exit(vcpu);
5517 static void grow_ple_window(struct kvm_vcpu *vcpu)
5519 struct vcpu_vmx *vmx = to_vmx(vcpu);
5520 unsigned int old = vmx->ple_window;
5522 vmx->ple_window = __grow_ple_window(old, ple_window,
5526 if (vmx->ple_window != old) {
5527 vmx->ple_window_dirty = true;
5528 trace_kvm_ple_window_update(vcpu->vcpu_id,
5529 vmx->ple_window, old);
5533 static void shrink_ple_window(struct kvm_vcpu *vcpu)
5535 struct vcpu_vmx *vmx = to_vmx(vcpu);
5536 unsigned int old = vmx->ple_window;
5538 vmx->ple_window = __shrink_ple_window(old, ple_window,
5542 if (vmx->ple_window != old) {
5543 vmx->ple_window_dirty = true;
5544 trace_kvm_ple_window_update(vcpu->vcpu_id,
5545 vmx->ple_window, old);
5550 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5551 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5553 static int handle_pause(struct kvm_vcpu *vcpu)
5555 if (!kvm_pause_in_guest(vcpu->kvm))
5556 grow_ple_window(vcpu);
5559 * Intel sdm vol3 ch-25.1.3 says: The "PAUSE-loop exiting"
5560 * VM-execution control is ignored if CPL > 0. OTOH, KVM
5561 * never set PAUSE_EXITING and just set PLE if supported,
5562 * so the vcpu must be CPL=0 if it gets a PAUSE exit.
5564 kvm_vcpu_on_spin(vcpu, true);
5565 return kvm_skip_emulated_instruction(vcpu);
5568 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
5573 static int handle_invpcid(struct kvm_vcpu *vcpu)
5575 u32 vmx_instruction_info;
5584 if (!guest_cpuid_has(vcpu, X86_FEATURE_INVPCID)) {
5585 kvm_queue_exception(vcpu, UD_VECTOR);
5589 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5590 gpr_index = vmx_get_instr_info_reg2(vmx_instruction_info);
5591 type = kvm_register_read(vcpu, gpr_index);
5593 /* According to the Intel instruction reference, the memory operand
5594 * is read even if it isn't needed (e.g., for type==all)
5596 if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5597 vmx_instruction_info, false,
5598 sizeof(operand), &gva))
5601 return kvm_handle_invpcid(vcpu, type, gva);
5604 static int handle_pml_full(struct kvm_vcpu *vcpu)
5606 unsigned long exit_qualification;
5608 trace_kvm_pml_full(vcpu->vcpu_id);
5610 exit_qualification = vmx_get_exit_qual(vcpu);
5613 * PML buffer FULL happened while executing iret from NMI,
5614 * "blocked by NMI" bit has to be set before next VM entry.
5616 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5618 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
5619 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5620 GUEST_INTR_STATE_NMI);
5623 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
5624 * here.., and there's no userspace involvement needed for PML.
5629 static fastpath_t handle_fastpath_preemption_timer(struct kvm_vcpu *vcpu)
5631 struct vcpu_vmx *vmx = to_vmx(vcpu);
5633 if (!vmx->req_immediate_exit &&
5634 !unlikely(vmx->loaded_vmcs->hv_timer_soft_disabled)) {
5635 kvm_lapic_expired_hv_timer(vcpu);
5636 return EXIT_FASTPATH_REENTER_GUEST;
5639 return EXIT_FASTPATH_NONE;
5642 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
5644 handle_fastpath_preemption_timer(vcpu);
5649 * When nested=0, all VMX instruction VM Exits filter here. The handlers
5650 * are overwritten by nested_vmx_setup() when nested=1.
5652 static int handle_vmx_instruction(struct kvm_vcpu *vcpu)
5654 kvm_queue_exception(vcpu, UD_VECTOR);
5658 #ifndef CONFIG_X86_SGX_KVM
5659 static int handle_encls(struct kvm_vcpu *vcpu)
5662 * SGX virtualization is disabled. There is no software enable bit for
5663 * SGX, so KVM intercepts all ENCLS leafs and injects a #UD to prevent
5664 * the guest from executing ENCLS (when SGX is supported by hardware).
5666 kvm_queue_exception(vcpu, UD_VECTOR);
5669 #endif /* CONFIG_X86_SGX_KVM */
5671 static int handle_bus_lock_vmexit(struct kvm_vcpu *vcpu)
5674 * Hardware may or may not set the BUS_LOCK_DETECTED flag on BUS_LOCK
5675 * VM-Exits. Unconditionally set the flag here and leave the handling to
5676 * vmx_handle_exit().
5678 to_vmx(vcpu)->exit_reason.bus_lock_detected = true;
5683 * The exit handlers return 1 if the exit was handled fully and guest execution
5684 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
5685 * to be done to userspace and return 0.
5687 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
5688 [EXIT_REASON_EXCEPTION_NMI] = handle_exception_nmi,
5689 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
5690 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
5691 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
5692 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
5693 [EXIT_REASON_CR_ACCESS] = handle_cr,
5694 [EXIT_REASON_DR_ACCESS] = handle_dr,
5695 [EXIT_REASON_CPUID] = kvm_emulate_cpuid,
5696 [EXIT_REASON_MSR_READ] = kvm_emulate_rdmsr,
5697 [EXIT_REASON_MSR_WRITE] = kvm_emulate_wrmsr,
5698 [EXIT_REASON_INTERRUPT_WINDOW] = handle_interrupt_window,
5699 [EXIT_REASON_HLT] = kvm_emulate_halt,
5700 [EXIT_REASON_INVD] = kvm_emulate_invd,
5701 [EXIT_REASON_INVLPG] = handle_invlpg,
5702 [EXIT_REASON_RDPMC] = kvm_emulate_rdpmc,
5703 [EXIT_REASON_VMCALL] = kvm_emulate_hypercall,
5704 [EXIT_REASON_VMCLEAR] = handle_vmx_instruction,
5705 [EXIT_REASON_VMLAUNCH] = handle_vmx_instruction,
5706 [EXIT_REASON_VMPTRLD] = handle_vmx_instruction,
5707 [EXIT_REASON_VMPTRST] = handle_vmx_instruction,
5708 [EXIT_REASON_VMREAD] = handle_vmx_instruction,
5709 [EXIT_REASON_VMRESUME] = handle_vmx_instruction,
5710 [EXIT_REASON_VMWRITE] = handle_vmx_instruction,
5711 [EXIT_REASON_VMOFF] = handle_vmx_instruction,
5712 [EXIT_REASON_VMON] = handle_vmx_instruction,
5713 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
5714 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
5715 [EXIT_REASON_APIC_WRITE] = handle_apic_write,
5716 [EXIT_REASON_EOI_INDUCED] = handle_apic_eoi_induced,
5717 [EXIT_REASON_WBINVD] = kvm_emulate_wbinvd,
5718 [EXIT_REASON_XSETBV] = kvm_emulate_xsetbv,
5719 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
5720 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
5721 [EXIT_REASON_GDTR_IDTR] = handle_desc,
5722 [EXIT_REASON_LDTR_TR] = handle_desc,
5723 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
5724 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
5725 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
5726 [EXIT_REASON_MWAIT_INSTRUCTION] = kvm_emulate_mwait,
5727 [EXIT_REASON_MONITOR_TRAP_FLAG] = handle_monitor_trap,
5728 [EXIT_REASON_MONITOR_INSTRUCTION] = kvm_emulate_monitor,
5729 [EXIT_REASON_INVEPT] = handle_vmx_instruction,
5730 [EXIT_REASON_INVVPID] = handle_vmx_instruction,
5731 [EXIT_REASON_RDRAND] = kvm_handle_invalid_op,
5732 [EXIT_REASON_RDSEED] = kvm_handle_invalid_op,
5733 [EXIT_REASON_PML_FULL] = handle_pml_full,
5734 [EXIT_REASON_INVPCID] = handle_invpcid,
5735 [EXIT_REASON_VMFUNC] = handle_vmx_instruction,
5736 [EXIT_REASON_PREEMPTION_TIMER] = handle_preemption_timer,
5737 [EXIT_REASON_ENCLS] = handle_encls,
5738 [EXIT_REASON_BUS_LOCK] = handle_bus_lock_vmexit,
5741 static const int kvm_vmx_max_exit_handlers =
5742 ARRAY_SIZE(kvm_vmx_exit_handlers);
5744 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u32 *reason,
5745 u64 *info1, u64 *info2,
5746 u32 *intr_info, u32 *error_code)
5748 struct vcpu_vmx *vmx = to_vmx(vcpu);
5750 *reason = vmx->exit_reason.full;
5751 *info1 = vmx_get_exit_qual(vcpu);
5752 if (!(vmx->exit_reason.failed_vmentry)) {
5753 *info2 = vmx->idt_vectoring_info;
5754 *intr_info = vmx_get_intr_info(vcpu);
5755 if (is_exception_with_error_code(*intr_info))
5756 *error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
5766 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
5769 __free_page(vmx->pml_pg);
5774 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
5776 struct vcpu_vmx *vmx = to_vmx(vcpu);
5780 pml_idx = vmcs_read16(GUEST_PML_INDEX);
5782 /* Do nothing if PML buffer is empty */
5783 if (pml_idx == (PML_ENTITY_NUM - 1))
5786 /* PML index always points to next available PML buffer entity */
5787 if (pml_idx >= PML_ENTITY_NUM)
5792 pml_buf = page_address(vmx->pml_pg);
5793 for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
5796 gpa = pml_buf[pml_idx];
5797 WARN_ON(gpa & (PAGE_SIZE - 1));
5798 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5801 /* reset PML index */
5802 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
5805 static void vmx_dump_sel(char *name, uint32_t sel)
5807 pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
5808 name, vmcs_read16(sel),
5809 vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
5810 vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
5811 vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
5814 static void vmx_dump_dtsel(char *name, uint32_t limit)
5816 pr_err("%s limit=0x%08x, base=0x%016lx\n",
5817 name, vmcs_read32(limit),
5818 vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
5821 static void vmx_dump_msrs(char *name, struct vmx_msrs *m)
5824 struct vmx_msr_entry *e;
5826 pr_err("MSR %s:\n", name);
5827 for (i = 0, e = m->val; i < m->nr; ++i, ++e)
5828 pr_err(" %2d: msr=0x%08x value=0x%016llx\n", i, e->index, e->value);
5831 void dump_vmcs(struct kvm_vcpu *vcpu)
5833 struct vcpu_vmx *vmx = to_vmx(vcpu);
5834 u32 vmentry_ctl, vmexit_ctl;
5835 u32 cpu_based_exec_ctrl, pin_based_exec_ctrl, secondary_exec_control;
5839 if (!dump_invalid_vmcs) {
5840 pr_warn_ratelimited("set kvm_intel.dump_invalid_vmcs=1 to dump internal KVM state.\n");
5844 vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
5845 vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
5846 cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5847 pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
5848 cr4 = vmcs_readl(GUEST_CR4);
5849 secondary_exec_control = 0;
5850 if (cpu_has_secondary_exec_ctrls())
5851 secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
5853 pr_err("VMCS %p, last attempted VM-entry on CPU %d\n",
5854 vmx->loaded_vmcs->vmcs, vcpu->arch.last_vmentry_cpu);
5855 pr_err("*** Guest State ***\n");
5856 pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5857 vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
5858 vmcs_readl(CR0_GUEST_HOST_MASK));
5859 pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5860 cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
5861 pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
5862 if (cpu_has_vmx_ept()) {
5863 pr_err("PDPTR0 = 0x%016llx PDPTR1 = 0x%016llx\n",
5864 vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
5865 pr_err("PDPTR2 = 0x%016llx PDPTR3 = 0x%016llx\n",
5866 vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
5868 pr_err("RSP = 0x%016lx RIP = 0x%016lx\n",
5869 vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
5870 pr_err("RFLAGS=0x%08lx DR7 = 0x%016lx\n",
5871 vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
5872 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5873 vmcs_readl(GUEST_SYSENTER_ESP),
5874 vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
5875 vmx_dump_sel("CS: ", GUEST_CS_SELECTOR);
5876 vmx_dump_sel("DS: ", GUEST_DS_SELECTOR);
5877 vmx_dump_sel("SS: ", GUEST_SS_SELECTOR);
5878 vmx_dump_sel("ES: ", GUEST_ES_SELECTOR);
5879 vmx_dump_sel("FS: ", GUEST_FS_SELECTOR);
5880 vmx_dump_sel("GS: ", GUEST_GS_SELECTOR);
5881 vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
5882 vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
5883 vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
5884 vmx_dump_sel("TR: ", GUEST_TR_SELECTOR);
5885 efer_slot = vmx_find_loadstore_msr_slot(&vmx->msr_autoload.guest, MSR_EFER);
5886 if (vmentry_ctl & VM_ENTRY_LOAD_IA32_EFER)
5887 pr_err("EFER= 0x%016llx\n", vmcs_read64(GUEST_IA32_EFER));
5888 else if (efer_slot >= 0)
5889 pr_err("EFER= 0x%016llx (autoload)\n",
5890 vmx->msr_autoload.guest.val[efer_slot].value);
5891 else if (vmentry_ctl & VM_ENTRY_IA32E_MODE)
5892 pr_err("EFER= 0x%016llx (effective)\n",
5893 vcpu->arch.efer | (EFER_LMA | EFER_LME));
5895 pr_err("EFER= 0x%016llx (effective)\n",
5896 vcpu->arch.efer & ~(EFER_LMA | EFER_LME));
5897 if (vmentry_ctl & VM_ENTRY_LOAD_IA32_PAT)
5898 pr_err("PAT = 0x%016llx\n", vmcs_read64(GUEST_IA32_PAT));
5899 pr_err("DebugCtl = 0x%016llx DebugExceptions = 0x%016lx\n",
5900 vmcs_read64(GUEST_IA32_DEBUGCTL),
5901 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
5902 if (cpu_has_load_perf_global_ctrl() &&
5903 vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
5904 pr_err("PerfGlobCtl = 0x%016llx\n",
5905 vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
5906 if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
5907 pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
5908 pr_err("Interruptibility = %08x ActivityState = %08x\n",
5909 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
5910 vmcs_read32(GUEST_ACTIVITY_STATE));
5911 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
5912 pr_err("InterruptStatus = %04x\n",
5913 vmcs_read16(GUEST_INTR_STATUS));
5914 if (vmcs_read32(VM_ENTRY_MSR_LOAD_COUNT) > 0)
5915 vmx_dump_msrs("guest autoload", &vmx->msr_autoload.guest);
5916 if (vmcs_read32(VM_EXIT_MSR_STORE_COUNT) > 0)
5917 vmx_dump_msrs("guest autostore", &vmx->msr_autostore.guest);
5919 pr_err("*** Host State ***\n");
5920 pr_err("RIP = 0x%016lx RSP = 0x%016lx\n",
5921 vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
5922 pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
5923 vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
5924 vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
5925 vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
5926 vmcs_read16(HOST_TR_SELECTOR));
5927 pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
5928 vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
5929 vmcs_readl(HOST_TR_BASE));
5930 pr_err("GDTBase=%016lx IDTBase=%016lx\n",
5931 vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
5932 pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
5933 vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
5934 vmcs_readl(HOST_CR4));
5935 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5936 vmcs_readl(HOST_IA32_SYSENTER_ESP),
5937 vmcs_read32(HOST_IA32_SYSENTER_CS),
5938 vmcs_readl(HOST_IA32_SYSENTER_EIP));
5939 if (vmexit_ctl & VM_EXIT_LOAD_IA32_EFER)
5940 pr_err("EFER= 0x%016llx\n", vmcs_read64(HOST_IA32_EFER));
5941 if (vmexit_ctl & VM_EXIT_LOAD_IA32_PAT)
5942 pr_err("PAT = 0x%016llx\n", vmcs_read64(HOST_IA32_PAT));
5943 if (cpu_has_load_perf_global_ctrl() &&
5944 vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
5945 pr_err("PerfGlobCtl = 0x%016llx\n",
5946 vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
5947 if (vmcs_read32(VM_EXIT_MSR_LOAD_COUNT) > 0)
5948 vmx_dump_msrs("host autoload", &vmx->msr_autoload.host);
5950 pr_err("*** Control State ***\n");
5951 pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
5952 pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
5953 pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
5954 pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
5955 vmcs_read32(EXCEPTION_BITMAP),
5956 vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
5957 vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
5958 pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
5959 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
5960 vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
5961 vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
5962 pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
5963 vmcs_read32(VM_EXIT_INTR_INFO),
5964 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
5965 vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
5966 pr_err(" reason=%08x qualification=%016lx\n",
5967 vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
5968 pr_err("IDTVectoring: info=%08x errcode=%08x\n",
5969 vmcs_read32(IDT_VECTORING_INFO_FIELD),
5970 vmcs_read32(IDT_VECTORING_ERROR_CODE));
5971 pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
5972 if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
5973 pr_err("TSC Multiplier = 0x%016llx\n",
5974 vmcs_read64(TSC_MULTIPLIER));
5975 if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW) {
5976 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
5977 u16 status = vmcs_read16(GUEST_INTR_STATUS);
5978 pr_err("SVI|RVI = %02x|%02x ", status >> 8, status & 0xff);
5980 pr_cont("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
5981 if (secondary_exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
5982 pr_err("APIC-access addr = 0x%016llx ", vmcs_read64(APIC_ACCESS_ADDR));
5983 pr_cont("virt-APIC addr = 0x%016llx\n", vmcs_read64(VIRTUAL_APIC_PAGE_ADDR));
5985 if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
5986 pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
5987 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
5988 pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
5989 if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
5990 pr_err("PLE Gap=%08x Window=%08x\n",
5991 vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
5992 if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
5993 pr_err("Virtual processor ID = 0x%04x\n",
5994 vmcs_read16(VIRTUAL_PROCESSOR_ID));
5998 * The guest has exited. See if we can fix it or if we need userspace
6001 static int __vmx_handle_exit(struct kvm_vcpu *vcpu, fastpath_t exit_fastpath)
6003 struct vcpu_vmx *vmx = to_vmx(vcpu);
6004 union vmx_exit_reason exit_reason = vmx->exit_reason;
6005 u32 vectoring_info = vmx->idt_vectoring_info;
6006 u16 exit_handler_index;
6009 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
6010 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
6011 * querying dirty_bitmap, we only need to kick all vcpus out of guest
6012 * mode as if vcpus is in root mode, the PML buffer must has been
6013 * flushed already. Note, PML is never enabled in hardware while
6016 if (enable_pml && !is_guest_mode(vcpu))
6017 vmx_flush_pml_buffer(vcpu);
6020 * KVM should never reach this point with a pending nested VM-Enter.
6021 * More specifically, short-circuiting VM-Entry to emulate L2 due to
6022 * invalid guest state should never happen as that means KVM knowingly
6023 * allowed a nested VM-Enter with an invalid vmcs12. More below.
6025 if (KVM_BUG_ON(vmx->nested.nested_run_pending, vcpu->kvm))
6028 if (is_guest_mode(vcpu)) {
6030 * PML is never enabled when running L2, bail immediately if a
6031 * PML full exit occurs as something is horribly wrong.
6033 if (exit_reason.basic == EXIT_REASON_PML_FULL)
6034 goto unexpected_vmexit;
6037 * The host physical addresses of some pages of guest memory
6038 * are loaded into the vmcs02 (e.g. vmcs12's Virtual APIC
6039 * Page). The CPU may write to these pages via their host
6040 * physical address while L2 is running, bypassing any
6041 * address-translation-based dirty tracking (e.g. EPT write
6044 * Mark them dirty on every exit from L2 to prevent them from
6045 * getting out of sync with dirty tracking.
6047 nested_mark_vmcs12_pages_dirty(vcpu);
6050 * Synthesize a triple fault if L2 state is invalid. In normal
6051 * operation, nested VM-Enter rejects any attempt to enter L2
6052 * with invalid state. However, those checks are skipped if
6053 * state is being stuffed via RSM or KVM_SET_NESTED_STATE. If
6054 * L2 state is invalid, it means either L1 modified SMRAM state
6055 * or userspace provided bad state. Synthesize TRIPLE_FAULT as
6056 * doing so is architecturally allowed in the RSM case, and is
6057 * the least awful solution for the userspace case without
6058 * risking false positives.
6060 if (vmx->emulation_required) {
6061 nested_vmx_vmexit(vcpu, EXIT_REASON_TRIPLE_FAULT, 0, 0);
6065 if (nested_vmx_reflect_vmexit(vcpu))
6069 /* If guest state is invalid, start emulating. L2 is handled above. */
6070 if (vmx->emulation_required)
6071 return handle_invalid_guest_state(vcpu);
6073 if (exit_reason.failed_vmentry) {
6075 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
6076 vcpu->run->fail_entry.hardware_entry_failure_reason
6078 vcpu->run->fail_entry.cpu = vcpu->arch.last_vmentry_cpu;
6082 if (unlikely(vmx->fail)) {
6084 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
6085 vcpu->run->fail_entry.hardware_entry_failure_reason
6086 = vmcs_read32(VM_INSTRUCTION_ERROR);
6087 vcpu->run->fail_entry.cpu = vcpu->arch.last_vmentry_cpu;
6093 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
6094 * delivery event since it indicates guest is accessing MMIO.
6095 * The vm-exit can be triggered again after return to guest that
6096 * will cause infinite loop.
6098 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
6099 (exit_reason.basic != EXIT_REASON_EXCEPTION_NMI &&
6100 exit_reason.basic != EXIT_REASON_EPT_VIOLATION &&
6101 exit_reason.basic != EXIT_REASON_PML_FULL &&
6102 exit_reason.basic != EXIT_REASON_APIC_ACCESS &&
6103 exit_reason.basic != EXIT_REASON_TASK_SWITCH)) {
6106 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6107 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
6108 vcpu->run->internal.data[0] = vectoring_info;
6109 vcpu->run->internal.data[1] = exit_reason.full;
6110 vcpu->run->internal.data[2] = vcpu->arch.exit_qualification;
6111 if (exit_reason.basic == EXIT_REASON_EPT_MISCONFIG) {
6112 vcpu->run->internal.data[ndata++] =
6113 vmcs_read64(GUEST_PHYSICAL_ADDRESS);
6115 vcpu->run->internal.data[ndata++] = vcpu->arch.last_vmentry_cpu;
6116 vcpu->run->internal.ndata = ndata;
6120 if (unlikely(!enable_vnmi &&
6121 vmx->loaded_vmcs->soft_vnmi_blocked)) {
6122 if (!vmx_interrupt_blocked(vcpu)) {
6123 vmx->loaded_vmcs->soft_vnmi_blocked = 0;
6124 } else if (vmx->loaded_vmcs->vnmi_blocked_time > 1000000000LL &&
6125 vcpu->arch.nmi_pending) {
6127 * This CPU don't support us in finding the end of an
6128 * NMI-blocked window if the guest runs with IRQs
6129 * disabled. So we pull the trigger after 1 s of
6130 * futile waiting, but inform the user about this.
6132 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
6133 "state on VCPU %d after 1 s timeout\n",
6134 __func__, vcpu->vcpu_id);
6135 vmx->loaded_vmcs->soft_vnmi_blocked = 0;
6139 if (exit_fastpath != EXIT_FASTPATH_NONE)
6142 if (exit_reason.basic >= kvm_vmx_max_exit_handlers)
6143 goto unexpected_vmexit;
6144 #ifdef CONFIG_RETPOLINE
6145 if (exit_reason.basic == EXIT_REASON_MSR_WRITE)
6146 return kvm_emulate_wrmsr(vcpu);
6147 else if (exit_reason.basic == EXIT_REASON_PREEMPTION_TIMER)
6148 return handle_preemption_timer(vcpu);
6149 else if (exit_reason.basic == EXIT_REASON_INTERRUPT_WINDOW)
6150 return handle_interrupt_window(vcpu);
6151 else if (exit_reason.basic == EXIT_REASON_EXTERNAL_INTERRUPT)
6152 return handle_external_interrupt(vcpu);
6153 else if (exit_reason.basic == EXIT_REASON_HLT)
6154 return kvm_emulate_halt(vcpu);
6155 else if (exit_reason.basic == EXIT_REASON_EPT_MISCONFIG)
6156 return handle_ept_misconfig(vcpu);
6159 exit_handler_index = array_index_nospec((u16)exit_reason.basic,
6160 kvm_vmx_max_exit_handlers);
6161 if (!kvm_vmx_exit_handlers[exit_handler_index])
6162 goto unexpected_vmexit;
6164 return kvm_vmx_exit_handlers[exit_handler_index](vcpu);
6167 vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n",
6170 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6171 vcpu->run->internal.suberror =
6172 KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON;
6173 vcpu->run->internal.ndata = 2;
6174 vcpu->run->internal.data[0] = exit_reason.full;
6175 vcpu->run->internal.data[1] = vcpu->arch.last_vmentry_cpu;
6179 static int vmx_handle_exit(struct kvm_vcpu *vcpu, fastpath_t exit_fastpath)
6181 int ret = __vmx_handle_exit(vcpu, exit_fastpath);
6184 * Exit to user space when bus lock detected to inform that there is
6185 * a bus lock in guest.
6187 if (to_vmx(vcpu)->exit_reason.bus_lock_detected) {
6189 vcpu->run->exit_reason = KVM_EXIT_X86_BUS_LOCK;
6191 vcpu->run->flags |= KVM_RUN_X86_BUS_LOCK;
6198 * Software based L1D cache flush which is used when microcode providing
6199 * the cache control MSR is not loaded.
6201 * The L1D cache is 32 KiB on Nehalem and later microarchitectures, but to
6202 * flush it is required to read in 64 KiB because the replacement algorithm
6203 * is not exactly LRU. This could be sized at runtime via topology
6204 * information but as all relevant affected CPUs have 32KiB L1D cache size
6205 * there is no point in doing so.
6207 static noinstr void vmx_l1d_flush(struct kvm_vcpu *vcpu)
6209 int size = PAGE_SIZE << L1D_CACHE_ORDER;
6212 * This code is only executed when the the flush mode is 'cond' or
6215 if (static_branch_likely(&vmx_l1d_flush_cond)) {
6219 * Clear the per-vcpu flush bit, it gets set again
6220 * either from vcpu_run() or from one of the unsafe
6223 flush_l1d = vcpu->arch.l1tf_flush_l1d;
6224 vcpu->arch.l1tf_flush_l1d = false;
6227 * Clear the per-cpu flush bit, it gets set again from
6228 * the interrupt handlers.
6230 flush_l1d |= kvm_get_cpu_l1tf_flush_l1d();
6231 kvm_clear_cpu_l1tf_flush_l1d();
6237 vcpu->stat.l1d_flush++;
6239 if (static_cpu_has(X86_FEATURE_FLUSH_L1D)) {
6240 native_wrmsrl(MSR_IA32_FLUSH_CMD, L1D_FLUSH);
6245 /* First ensure the pages are in the TLB */
6246 "xorl %%eax, %%eax\n"
6247 ".Lpopulate_tlb:\n\t"
6248 "movzbl (%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
6249 "addl $4096, %%eax\n\t"
6250 "cmpl %%eax, %[size]\n\t"
6251 "jne .Lpopulate_tlb\n\t"
6252 "xorl %%eax, %%eax\n\t"
6254 /* Now fill the cache */
6255 "xorl %%eax, %%eax\n"
6257 "movzbl (%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
6258 "addl $64, %%eax\n\t"
6259 "cmpl %%eax, %[size]\n\t"
6260 "jne .Lfill_cache\n\t"
6262 :: [flush_pages] "r" (vmx_l1d_flush_pages),
6264 : "eax", "ebx", "ecx", "edx");
6267 static void vmx_update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
6269 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6272 if (is_guest_mode(vcpu) &&
6273 nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
6276 tpr_threshold = (irr == -1 || tpr < irr) ? 0 : irr;
6277 if (is_guest_mode(vcpu))
6278 to_vmx(vcpu)->nested.l1_tpr_threshold = tpr_threshold;
6280 vmcs_write32(TPR_THRESHOLD, tpr_threshold);
6283 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu)
6285 struct vcpu_vmx *vmx = to_vmx(vcpu);
6286 u32 sec_exec_control;
6288 if (!lapic_in_kernel(vcpu))
6291 if (!flexpriority_enabled &&
6292 !cpu_has_vmx_virtualize_x2apic_mode())
6295 /* Postpone execution until vmcs01 is the current VMCS. */
6296 if (is_guest_mode(vcpu)) {
6297 vmx->nested.change_vmcs01_virtual_apic_mode = true;
6301 sec_exec_control = secondary_exec_controls_get(vmx);
6302 sec_exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6303 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
6305 switch (kvm_get_apic_mode(vcpu)) {
6306 case LAPIC_MODE_INVALID:
6307 WARN_ONCE(true, "Invalid local APIC state");
6309 case LAPIC_MODE_DISABLED:
6311 case LAPIC_MODE_XAPIC:
6312 if (flexpriority_enabled) {
6314 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6315 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
6318 * Flush the TLB, reloading the APIC access page will
6319 * only do so if its physical address has changed, but
6320 * the guest may have inserted a non-APIC mapping into
6321 * the TLB while the APIC access page was disabled.
6323 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
6326 case LAPIC_MODE_X2APIC:
6327 if (cpu_has_vmx_virtualize_x2apic_mode())
6329 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
6332 secondary_exec_controls_set(vmx, sec_exec_control);
6334 vmx_update_msr_bitmap_x2apic(vcpu);
6337 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu)
6341 /* Defer reload until vmcs01 is the current VMCS. */
6342 if (is_guest_mode(vcpu)) {
6343 to_vmx(vcpu)->nested.reload_vmcs01_apic_access_page = true;
6347 if (!(secondary_exec_controls_get(to_vmx(vcpu)) &
6348 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
6351 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6352 if (is_error_page(page))
6355 vmcs_write64(APIC_ACCESS_ADDR, page_to_phys(page));
6356 vmx_flush_tlb_current(vcpu);
6359 * Do not pin apic access page in memory, the MMU notifier
6360 * will call us again if it is migrated or swapped out.
6365 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
6373 status = vmcs_read16(GUEST_INTR_STATUS);
6375 if (max_isr != old) {
6377 status |= max_isr << 8;
6378 vmcs_write16(GUEST_INTR_STATUS, status);
6382 static void vmx_set_rvi(int vector)
6390 status = vmcs_read16(GUEST_INTR_STATUS);
6391 old = (u8)status & 0xff;
6392 if ((u8)vector != old) {
6394 status |= (u8)vector;
6395 vmcs_write16(GUEST_INTR_STATUS, status);
6399 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
6402 * When running L2, updating RVI is only relevant when
6403 * vmcs12 virtual-interrupt-delivery enabled.
6404 * However, it can be enabled only when L1 also
6405 * intercepts external-interrupts and in that case
6406 * we should not update vmcs02 RVI but instead intercept
6407 * interrupt. Therefore, do nothing when running L2.
6409 if (!is_guest_mode(vcpu))
6410 vmx_set_rvi(max_irr);
6413 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
6415 struct vcpu_vmx *vmx = to_vmx(vcpu);
6417 bool got_posted_interrupt;
6419 if (KVM_BUG_ON(!enable_apicv, vcpu->kvm))
6422 if (pi_test_on(&vmx->pi_desc)) {
6423 pi_clear_on(&vmx->pi_desc);
6425 * IOMMU can write to PID.ON, so the barrier matters even on UP.
6426 * But on x86 this is just a compiler barrier anyway.
6428 smp_mb__after_atomic();
6429 got_posted_interrupt =
6430 kvm_apic_update_irr(vcpu, vmx->pi_desc.pir, &max_irr);
6432 max_irr = kvm_lapic_find_highest_irr(vcpu);
6433 got_posted_interrupt = false;
6437 * Newly recognized interrupts are injected via either virtual interrupt
6438 * delivery (RVI) or KVM_REQ_EVENT. Virtual interrupt delivery is
6439 * disabled in two cases:
6441 * 1) If L2 is running and the vCPU has a new pending interrupt. If L1
6442 * wants to exit on interrupts, KVM_REQ_EVENT is needed to synthesize a
6443 * VM-Exit to L1. If L1 doesn't want to exit, the interrupt is injected
6444 * into L2, but KVM doesn't use virtual interrupt delivery to inject
6445 * interrupts into L2, and so KVM_REQ_EVENT is again needed.
6447 * 2) If APICv is disabled for this vCPU, assigned devices may still
6448 * attempt to post interrupts. The posted interrupt vector will cause
6449 * a VM-Exit and the subsequent entry will call sync_pir_to_irr.
6451 if (!is_guest_mode(vcpu) && kvm_vcpu_apicv_active(vcpu))
6452 vmx_set_rvi(max_irr);
6453 else if (got_posted_interrupt)
6454 kvm_make_request(KVM_REQ_EVENT, vcpu);
6459 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
6461 if (!kvm_vcpu_apicv_active(vcpu))
6464 vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
6465 vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
6466 vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
6467 vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
6470 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
6472 struct vcpu_vmx *vmx = to_vmx(vcpu);
6474 pi_clear_on(&vmx->pi_desc);
6475 memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
6478 void vmx_do_interrupt_nmi_irqoff(unsigned long entry);
6480 static void handle_interrupt_nmi_irqoff(struct kvm_vcpu *vcpu,
6481 unsigned long entry)
6483 bool is_nmi = entry == (unsigned long)asm_exc_nmi_noist;
6485 kvm_before_interrupt(vcpu, is_nmi ? KVM_HANDLING_NMI : KVM_HANDLING_IRQ);
6486 vmx_do_interrupt_nmi_irqoff(entry);
6487 kvm_after_interrupt(vcpu);
6490 static void handle_nm_fault_irqoff(struct kvm_vcpu *vcpu)
6493 * Save xfd_err to guest_fpu before interrupt is enabled, so the
6494 * MSR value is not clobbered by the host activity before the guest
6495 * has chance to consume it.
6497 * Do not blindly read xfd_err here, since this exception might
6498 * be caused by L1 interception on a platform which doesn't
6499 * support xfd at all.
6501 * Do it conditionally upon guest_fpu::xfd. xfd_err matters
6502 * only when xfd contains a non-zero value.
6504 * Queuing exception is done in vmx_handle_exit. See comment there.
6506 if (vcpu->arch.guest_fpu.fpstate->xfd)
6507 rdmsrl(MSR_IA32_XFD_ERR, vcpu->arch.guest_fpu.xfd_err);
6510 static void handle_exception_nmi_irqoff(struct vcpu_vmx *vmx)
6512 const unsigned long nmi_entry = (unsigned long)asm_exc_nmi_noist;
6513 u32 intr_info = vmx_get_intr_info(&vmx->vcpu);
6515 /* if exit due to PF check for async PF */
6516 if (is_page_fault(intr_info))
6517 vmx->vcpu.arch.apf.host_apf_flags = kvm_read_and_reset_apf_flags();
6518 /* if exit due to NM, handle before interrupts are enabled */
6519 else if (is_nm_fault(intr_info))
6520 handle_nm_fault_irqoff(&vmx->vcpu);
6521 /* Handle machine checks before interrupts are enabled */
6522 else if (is_machine_check(intr_info))
6523 kvm_machine_check();
6524 /* We need to handle NMIs before interrupts are enabled */
6525 else if (is_nmi(intr_info))
6526 handle_interrupt_nmi_irqoff(&vmx->vcpu, nmi_entry);
6529 static void handle_external_interrupt_irqoff(struct kvm_vcpu *vcpu)
6531 u32 intr_info = vmx_get_intr_info(vcpu);
6532 unsigned int vector = intr_info & INTR_INFO_VECTOR_MASK;
6533 gate_desc *desc = (gate_desc *)host_idt_base + vector;
6535 if (KVM_BUG(!is_external_intr(intr_info), vcpu->kvm,
6536 "KVM: unexpected VM-Exit interrupt info: 0x%x", intr_info))
6539 handle_interrupt_nmi_irqoff(vcpu, gate_offset(desc));
6542 static void vmx_handle_exit_irqoff(struct kvm_vcpu *vcpu)
6544 struct vcpu_vmx *vmx = to_vmx(vcpu);
6546 if (vmx->emulation_required)
6549 if (vmx->exit_reason.basic == EXIT_REASON_EXTERNAL_INTERRUPT)
6550 handle_external_interrupt_irqoff(vcpu);
6551 else if (vmx->exit_reason.basic == EXIT_REASON_EXCEPTION_NMI)
6552 handle_exception_nmi_irqoff(vmx);
6556 * The kvm parameter can be NULL (module initialization, or invocation before
6557 * VM creation). Be sure to check the kvm parameter before using it.
6559 static bool vmx_has_emulated_msr(struct kvm *kvm, u32 index)
6562 case MSR_IA32_SMBASE:
6564 * We cannot do SMM unless we can run the guest in big
6567 return enable_unrestricted_guest || emulate_invalid_guest_state;
6568 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
6570 case MSR_AMD64_VIRT_SPEC_CTRL:
6571 case MSR_AMD64_TSC_RATIO:
6572 /* This is AMD only. */
6579 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
6584 bool idtv_info_valid;
6586 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6589 if (vmx->loaded_vmcs->nmi_known_unmasked)
6592 exit_intr_info = vmx_get_intr_info(&vmx->vcpu);
6593 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
6594 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
6596 * SDM 3: 27.7.1.2 (September 2008)
6597 * Re-set bit "block by NMI" before VM entry if vmexit caused by
6598 * a guest IRET fault.
6599 * SDM 3: 23.2.2 (September 2008)
6600 * Bit 12 is undefined in any of the following cases:
6601 * If the VM exit sets the valid bit in the IDT-vectoring
6602 * information field.
6603 * If the VM exit is due to a double fault.
6605 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
6606 vector != DF_VECTOR && !idtv_info_valid)
6607 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
6608 GUEST_INTR_STATE_NMI);
6610 vmx->loaded_vmcs->nmi_known_unmasked =
6611 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
6612 & GUEST_INTR_STATE_NMI);
6613 } else if (unlikely(vmx->loaded_vmcs->soft_vnmi_blocked))
6614 vmx->loaded_vmcs->vnmi_blocked_time +=
6615 ktime_to_ns(ktime_sub(ktime_get(),
6616 vmx->loaded_vmcs->entry_time));
6619 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
6620 u32 idt_vectoring_info,
6621 int instr_len_field,
6622 int error_code_field)
6626 bool idtv_info_valid;
6628 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6630 vcpu->arch.nmi_injected = false;
6631 kvm_clear_exception_queue(vcpu);
6632 kvm_clear_interrupt_queue(vcpu);
6634 if (!idtv_info_valid)
6637 kvm_make_request(KVM_REQ_EVENT, vcpu);
6639 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
6640 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
6643 case INTR_TYPE_NMI_INTR:
6644 vcpu->arch.nmi_injected = true;
6646 * SDM 3: 27.7.1.2 (September 2008)
6647 * Clear bit "block by NMI" before VM entry if a NMI
6650 vmx_set_nmi_mask(vcpu, false);
6652 case INTR_TYPE_SOFT_EXCEPTION:
6653 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6655 case INTR_TYPE_HARD_EXCEPTION:
6656 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
6657 u32 err = vmcs_read32(error_code_field);
6658 kvm_requeue_exception_e(vcpu, vector, err);
6660 kvm_requeue_exception(vcpu, vector);
6662 case INTR_TYPE_SOFT_INTR:
6663 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6665 case INTR_TYPE_EXT_INTR:
6666 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
6673 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
6675 __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
6676 VM_EXIT_INSTRUCTION_LEN,
6677 IDT_VECTORING_ERROR_CODE);
6680 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
6682 __vmx_complete_interrupts(vcpu,
6683 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
6684 VM_ENTRY_INSTRUCTION_LEN,
6685 VM_ENTRY_EXCEPTION_ERROR_CODE);
6687 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
6690 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
6693 struct perf_guest_switch_msr *msrs;
6695 /* Note, nr_msrs may be garbage if perf_guest_get_msrs() returns NULL. */
6696 msrs = perf_guest_get_msrs(&nr_msrs);
6700 for (i = 0; i < nr_msrs; i++)
6701 if (msrs[i].host == msrs[i].guest)
6702 clear_atomic_switch_msr(vmx, msrs[i].msr);
6704 add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
6705 msrs[i].host, false);
6708 static void vmx_update_hv_timer(struct kvm_vcpu *vcpu)
6710 struct vcpu_vmx *vmx = to_vmx(vcpu);
6714 if (vmx->req_immediate_exit) {
6715 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, 0);
6716 vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6717 } else if (vmx->hv_deadline_tsc != -1) {
6719 if (vmx->hv_deadline_tsc > tscl)
6720 /* set_hv_timer ensures the delta fits in 32-bits */
6721 delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
6722 cpu_preemption_timer_multi);
6726 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
6727 vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6728 } else if (!vmx->loaded_vmcs->hv_timer_soft_disabled) {
6729 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, -1);
6730 vmx->loaded_vmcs->hv_timer_soft_disabled = true;
6734 void noinstr vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp)
6736 if (unlikely(host_rsp != vmx->loaded_vmcs->host_state.rsp)) {
6737 vmx->loaded_vmcs->host_state.rsp = host_rsp;
6738 vmcs_writel(HOST_RSP, host_rsp);
6742 static fastpath_t vmx_exit_handlers_fastpath(struct kvm_vcpu *vcpu)
6744 switch (to_vmx(vcpu)->exit_reason.basic) {
6745 case EXIT_REASON_MSR_WRITE:
6746 return handle_fastpath_set_msr_irqoff(vcpu);
6747 case EXIT_REASON_PREEMPTION_TIMER:
6748 return handle_fastpath_preemption_timer(vcpu);
6750 return EXIT_FASTPATH_NONE;
6754 static noinstr void vmx_vcpu_enter_exit(struct kvm_vcpu *vcpu,
6755 struct vcpu_vmx *vmx)
6757 kvm_guest_enter_irqoff();
6759 /* L1D Flush includes CPU buffer clear to mitigate MDS */
6760 if (static_branch_unlikely(&vmx_l1d_should_flush))
6761 vmx_l1d_flush(vcpu);
6762 else if (static_branch_unlikely(&mds_user_clear))
6763 mds_clear_cpu_buffers();
6765 if (vcpu->arch.cr2 != native_read_cr2())
6766 native_write_cr2(vcpu->arch.cr2);
6768 vmx->fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
6769 vmx->loaded_vmcs->launched);
6771 vcpu->arch.cr2 = native_read_cr2();
6773 kvm_guest_exit_irqoff();
6776 static fastpath_t vmx_vcpu_run(struct kvm_vcpu *vcpu)
6778 struct vcpu_vmx *vmx = to_vmx(vcpu);
6781 /* Record the guest's net vcpu time for enforced NMI injections. */
6782 if (unlikely(!enable_vnmi &&
6783 vmx->loaded_vmcs->soft_vnmi_blocked))
6784 vmx->loaded_vmcs->entry_time = ktime_get();
6787 * Don't enter VMX if guest state is invalid, let the exit handler
6788 * start emulation until we arrive back to a valid state. Synthesize a
6789 * consistency check VM-Exit due to invalid guest state and bail.
6791 if (unlikely(vmx->emulation_required)) {
6794 vmx->exit_reason.full = EXIT_REASON_INVALID_STATE;
6795 vmx->exit_reason.failed_vmentry = 1;
6796 kvm_register_mark_available(vcpu, VCPU_EXREG_EXIT_INFO_1);
6797 vmx->exit_qualification = ENTRY_FAIL_DEFAULT;
6798 kvm_register_mark_available(vcpu, VCPU_EXREG_EXIT_INFO_2);
6799 vmx->exit_intr_info = 0;
6800 return EXIT_FASTPATH_NONE;
6803 trace_kvm_entry(vcpu);
6805 if (vmx->ple_window_dirty) {
6806 vmx->ple_window_dirty = false;
6807 vmcs_write32(PLE_WINDOW, vmx->ple_window);
6811 * We did this in prepare_switch_to_guest, because it needs to
6812 * be within srcu_read_lock.
6814 WARN_ON_ONCE(vmx->nested.need_vmcs12_to_shadow_sync);
6816 if (kvm_register_is_dirty(vcpu, VCPU_REGS_RSP))
6817 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
6818 if (kvm_register_is_dirty(vcpu, VCPU_REGS_RIP))
6819 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
6820 vcpu->arch.regs_dirty = 0;
6822 cr4 = cr4_read_shadow();
6823 if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
6824 vmcs_writel(HOST_CR4, cr4);
6825 vmx->loaded_vmcs->host_state.cr4 = cr4;
6828 /* When KVM_DEBUGREG_WONT_EXIT, dr6 is accessible in guest. */
6829 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT))
6830 set_debugreg(vcpu->arch.dr6, 6);
6832 /* When single-stepping over STI and MOV SS, we must clear the
6833 * corresponding interruptibility bits in the guest state. Otherwise
6834 * vmentry fails as it then expects bit 14 (BS) in pending debug
6835 * exceptions being set, but that's not correct for the guest debugging
6837 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6838 vmx_set_interrupt_shadow(vcpu, 0);
6840 kvm_load_guest_xsave_state(vcpu);
6842 pt_guest_enter(vmx);
6844 atomic_switch_perf_msrs(vmx);
6845 if (intel_pmu_lbr_is_enabled(vcpu))
6846 vmx_passthrough_lbr_msrs(vcpu);
6848 if (enable_preemption_timer)
6849 vmx_update_hv_timer(vcpu);
6851 kvm_wait_lapic_expire(vcpu);
6854 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
6855 * it's non-zero. Since vmentry is serialising on affected CPUs, there
6856 * is no need to worry about the conditional branch over the wrmsr
6857 * being speculatively taken.
6859 x86_spec_ctrl_set_guest(vmx->spec_ctrl, 0);
6861 /* The actual VMENTER/EXIT is in the .noinstr.text section. */
6862 vmx_vcpu_enter_exit(vcpu, vmx);
6865 * We do not use IBRS in the kernel. If this vCPU has used the
6866 * SPEC_CTRL MSR it may have left it on; save the value and
6867 * turn it off. This is much more efficient than blindly adding
6868 * it to the atomic save/restore list. Especially as the former
6869 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
6871 * For non-nested case:
6872 * If the L01 MSR bitmap does not intercept the MSR, then we need to
6876 * If the L02 MSR bitmap does not intercept the MSR, then we need to
6879 if (unlikely(!msr_write_intercepted(vmx, MSR_IA32_SPEC_CTRL)))
6880 vmx->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
6882 x86_spec_ctrl_restore_host(vmx->spec_ctrl, 0);
6884 /* All fields are clean at this point */
6885 if (static_branch_unlikely(&enable_evmcs)) {
6886 current_evmcs->hv_clean_fields |=
6887 HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
6889 current_evmcs->hv_vp_id = kvm_hv_get_vpindex(vcpu);
6892 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
6893 if (vmx->host_debugctlmsr)
6894 update_debugctlmsr(vmx->host_debugctlmsr);
6896 #ifndef CONFIG_X86_64
6898 * The sysexit path does not restore ds/es, so we must set them to
6899 * a reasonable value ourselves.
6901 * We can't defer this to vmx_prepare_switch_to_host() since that
6902 * function may be executed in interrupt context, which saves and
6903 * restore segments around it, nullifying its effect.
6905 loadsegment(ds, __USER_DS);
6906 loadsegment(es, __USER_DS);
6909 vcpu->arch.regs_avail &= ~VMX_REGS_LAZY_LOAD_SET;
6913 kvm_load_host_xsave_state(vcpu);
6915 if (is_guest_mode(vcpu)) {
6917 * Track VMLAUNCH/VMRESUME that have made past guest state
6920 if (vmx->nested.nested_run_pending &&
6921 !vmx->exit_reason.failed_vmentry)
6922 ++vcpu->stat.nested_run;
6924 vmx->nested.nested_run_pending = 0;
6927 vmx->idt_vectoring_info = 0;
6929 if (unlikely(vmx->fail)) {
6930 vmx->exit_reason.full = 0xdead;
6931 return EXIT_FASTPATH_NONE;
6934 vmx->exit_reason.full = vmcs_read32(VM_EXIT_REASON);
6935 if (unlikely((u16)vmx->exit_reason.basic == EXIT_REASON_MCE_DURING_VMENTRY))
6936 kvm_machine_check();
6938 if (likely(!vmx->exit_reason.failed_vmentry))
6939 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
6941 trace_kvm_exit(vcpu, KVM_ISA_VMX);
6943 if (unlikely(vmx->exit_reason.failed_vmentry))
6944 return EXIT_FASTPATH_NONE;
6946 vmx->loaded_vmcs->launched = 1;
6948 vmx_recover_nmi_blocking(vmx);
6949 vmx_complete_interrupts(vmx);
6951 if (is_guest_mode(vcpu))
6952 return EXIT_FASTPATH_NONE;
6954 return vmx_exit_handlers_fastpath(vcpu);
6957 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
6959 struct vcpu_vmx *vmx = to_vmx(vcpu);
6962 vmx_destroy_pml_buffer(vmx);
6963 free_vpid(vmx->vpid);
6964 nested_vmx_free_vcpu(vcpu);
6965 free_loaded_vmcs(vmx->loaded_vmcs);
6968 static int vmx_create_vcpu(struct kvm_vcpu *vcpu)
6970 struct vmx_uret_msr *tsx_ctrl;
6971 struct vcpu_vmx *vmx;
6974 BUILD_BUG_ON(offsetof(struct vcpu_vmx, vcpu) != 0);
6977 INIT_LIST_HEAD(&vmx->pi_wakeup_list);
6981 vmx->vpid = allocate_vpid();
6984 * If PML is turned on, failure on enabling PML just results in failure
6985 * of creating the vcpu, therefore we can simplify PML logic (by
6986 * avoiding dealing with cases, such as enabling PML partially on vcpus
6987 * for the guest), etc.
6990 vmx->pml_pg = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
6995 for (i = 0; i < kvm_nr_uret_msrs; ++i)
6996 vmx->guest_uret_msrs[i].mask = -1ull;
6997 if (boot_cpu_has(X86_FEATURE_RTM)) {
6999 * TSX_CTRL_CPUID_CLEAR is handled in the CPUID interception.
7000 * Keep the host value unchanged to avoid changing CPUID bits
7001 * under the host kernel's feet.
7003 tsx_ctrl = vmx_find_uret_msr(vmx, MSR_IA32_TSX_CTRL);
7005 tsx_ctrl->mask = ~(u64)TSX_CTRL_CPUID_CLEAR;
7008 err = alloc_loaded_vmcs(&vmx->vmcs01);
7013 * Use Hyper-V 'Enlightened MSR Bitmap' feature when KVM runs as a
7014 * nested (L1) hypervisor and Hyper-V in L0 supports it. Enable the
7015 * feature only for vmcs01, KVM currently isn't equipped to realize any
7016 * performance benefits from enabling it for vmcs02.
7018 if (IS_ENABLED(CONFIG_HYPERV) && static_branch_unlikely(&enable_evmcs) &&
7019 (ms_hyperv.nested_features & HV_X64_NESTED_MSR_BITMAP)) {
7020 struct hv_enlightened_vmcs *evmcs = (void *)vmx->vmcs01.vmcs;
7022 evmcs->hv_enlightenments_control.msr_bitmap = 1;
7025 /* The MSR bitmap starts with all ones */
7026 bitmap_fill(vmx->shadow_msr_intercept.read, MAX_POSSIBLE_PASSTHROUGH_MSRS);
7027 bitmap_fill(vmx->shadow_msr_intercept.write, MAX_POSSIBLE_PASSTHROUGH_MSRS);
7029 vmx_disable_intercept_for_msr(vcpu, MSR_IA32_TSC, MSR_TYPE_R);
7030 #ifdef CONFIG_X86_64
7031 vmx_disable_intercept_for_msr(vcpu, MSR_FS_BASE, MSR_TYPE_RW);
7032 vmx_disable_intercept_for_msr(vcpu, MSR_GS_BASE, MSR_TYPE_RW);
7033 vmx_disable_intercept_for_msr(vcpu, MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
7035 vmx_disable_intercept_for_msr(vcpu, MSR_IA32_SYSENTER_CS, MSR_TYPE_RW);
7036 vmx_disable_intercept_for_msr(vcpu, MSR_IA32_SYSENTER_ESP, MSR_TYPE_RW);
7037 vmx_disable_intercept_for_msr(vcpu, MSR_IA32_SYSENTER_EIP, MSR_TYPE_RW);
7038 if (kvm_cstate_in_guest(vcpu->kvm)) {
7039 vmx_disable_intercept_for_msr(vcpu, MSR_CORE_C1_RES, MSR_TYPE_R);
7040 vmx_disable_intercept_for_msr(vcpu, MSR_CORE_C3_RESIDENCY, MSR_TYPE_R);
7041 vmx_disable_intercept_for_msr(vcpu, MSR_CORE_C6_RESIDENCY, MSR_TYPE_R);
7042 vmx_disable_intercept_for_msr(vcpu, MSR_CORE_C7_RESIDENCY, MSR_TYPE_R);
7045 vmx->loaded_vmcs = &vmx->vmcs01;
7047 if (cpu_need_virtualize_apic_accesses(vcpu)) {
7048 err = alloc_apic_access_page(vcpu->kvm);
7053 if (enable_ept && !enable_unrestricted_guest) {
7054 err = init_rmode_identity_map(vcpu->kvm);
7062 free_loaded_vmcs(vmx->loaded_vmcs);
7064 vmx_destroy_pml_buffer(vmx);
7066 free_vpid(vmx->vpid);
7070 #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"
7071 #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"
7073 static int vmx_vm_init(struct kvm *kvm)
7076 kvm->arch.pause_in_guest = true;
7078 if (boot_cpu_has(X86_BUG_L1TF) && enable_ept) {
7079 switch (l1tf_mitigation) {
7080 case L1TF_MITIGATION_OFF:
7081 case L1TF_MITIGATION_FLUSH_NOWARN:
7082 /* 'I explicitly don't care' is set */
7084 case L1TF_MITIGATION_FLUSH:
7085 case L1TF_MITIGATION_FLUSH_NOSMT:
7086 case L1TF_MITIGATION_FULL:
7088 * Warn upon starting the first VM in a potentially
7089 * insecure environment.
7091 if (sched_smt_active())
7092 pr_warn_once(L1TF_MSG_SMT);
7093 if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER)
7094 pr_warn_once(L1TF_MSG_L1D);
7096 case L1TF_MITIGATION_FULL_FORCE:
7097 /* Flush is enforced */
7104 static int __init vmx_check_processor_compat(void)
7106 struct vmcs_config vmcs_conf;
7107 struct vmx_capability vmx_cap;
7109 if (!this_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
7110 !this_cpu_has(X86_FEATURE_VMX)) {
7111 pr_err("kvm: VMX is disabled on CPU %d\n", smp_processor_id());
7115 if (setup_vmcs_config(&vmcs_conf, &vmx_cap) < 0)
7118 nested_vmx_setup_ctls_msrs(&vmcs_conf.nested, vmx_cap.ept);
7119 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
7120 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
7121 smp_processor_id());
7127 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
7131 /* We wanted to honor guest CD/MTRR/PAT, but doing so could result in
7132 * memory aliases with conflicting memory types and sometimes MCEs.
7133 * We have to be careful as to what are honored and when.
7135 * For MMIO, guest CD/MTRR are ignored. The EPT memory type is set to
7136 * UC. The effective memory type is UC or WC depending on guest PAT.
7137 * This was historically the source of MCEs and we want to be
7140 * When there is no need to deal with noncoherent DMA (e.g., no VT-d
7141 * or VT-d has snoop control), guest CD/MTRR/PAT are all ignored. The
7142 * EPT memory type is set to WB. The effective memory type is forced
7145 * Otherwise, we trust guest. Guest CD/MTRR/PAT are all honored. The
7146 * EPT memory type is used to emulate guest CD/MTRR.
7150 return MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
7152 if (!kvm_arch_has_noncoherent_dma(vcpu->kvm))
7153 return (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT) | VMX_EPT_IPAT_BIT;
7155 if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
7156 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
7157 cache = MTRR_TYPE_WRBACK;
7159 cache = MTRR_TYPE_UNCACHABLE;
7161 return (cache << VMX_EPT_MT_EPTE_SHIFT) | VMX_EPT_IPAT_BIT;
7164 return kvm_mtrr_get_guest_memory_type(vcpu, gfn) << VMX_EPT_MT_EPTE_SHIFT;
7167 static void vmcs_set_secondary_exec_control(struct vcpu_vmx *vmx, u32 new_ctl)
7170 * These bits in the secondary execution controls field
7171 * are dynamic, the others are mostly based on the hypervisor
7172 * architecture and the guest's CPUID. Do not touch the
7176 SECONDARY_EXEC_SHADOW_VMCS |
7177 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
7178 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
7179 SECONDARY_EXEC_DESC;
7181 u32 cur_ctl = secondary_exec_controls_get(vmx);
7183 secondary_exec_controls_set(vmx, (new_ctl & ~mask) | (cur_ctl & mask));
7187 * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
7188 * (indicating "allowed-1") if they are supported in the guest's CPUID.
7190 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
7192 struct vcpu_vmx *vmx = to_vmx(vcpu);
7193 struct kvm_cpuid_entry2 *entry;
7195 vmx->nested.msrs.cr0_fixed1 = 0xffffffff;
7196 vmx->nested.msrs.cr4_fixed1 = X86_CR4_PCE;
7198 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do { \
7199 if (entry && (entry->_reg & (_cpuid_mask))) \
7200 vmx->nested.msrs.cr4_fixed1 |= (_cr4_mask); \
7203 entry = kvm_find_cpuid_entry(vcpu, 0x1, 0);
7204 cr4_fixed1_update(X86_CR4_VME, edx, feature_bit(VME));
7205 cr4_fixed1_update(X86_CR4_PVI, edx, feature_bit(VME));
7206 cr4_fixed1_update(X86_CR4_TSD, edx, feature_bit(TSC));
7207 cr4_fixed1_update(X86_CR4_DE, edx, feature_bit(DE));
7208 cr4_fixed1_update(X86_CR4_PSE, edx, feature_bit(PSE));
7209 cr4_fixed1_update(X86_CR4_PAE, edx, feature_bit(PAE));
7210 cr4_fixed1_update(X86_CR4_MCE, edx, feature_bit(MCE));
7211 cr4_fixed1_update(X86_CR4_PGE, edx, feature_bit(PGE));
7212 cr4_fixed1_update(X86_CR4_OSFXSR, edx, feature_bit(FXSR));
7213 cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, feature_bit(XMM));
7214 cr4_fixed1_update(X86_CR4_VMXE, ecx, feature_bit(VMX));
7215 cr4_fixed1_update(X86_CR4_SMXE, ecx, feature_bit(SMX));
7216 cr4_fixed1_update(X86_CR4_PCIDE, ecx, feature_bit(PCID));
7217 cr4_fixed1_update(X86_CR4_OSXSAVE, ecx, feature_bit(XSAVE));
7219 entry = kvm_find_cpuid_entry(vcpu, 0x7, 0);
7220 cr4_fixed1_update(X86_CR4_FSGSBASE, ebx, feature_bit(FSGSBASE));
7221 cr4_fixed1_update(X86_CR4_SMEP, ebx, feature_bit(SMEP));
7222 cr4_fixed1_update(X86_CR4_SMAP, ebx, feature_bit(SMAP));
7223 cr4_fixed1_update(X86_CR4_PKE, ecx, feature_bit(PKU));
7224 cr4_fixed1_update(X86_CR4_UMIP, ecx, feature_bit(UMIP));
7225 cr4_fixed1_update(X86_CR4_LA57, ecx, feature_bit(LA57));
7227 #undef cr4_fixed1_update
7230 static void nested_vmx_entry_exit_ctls_update(struct kvm_vcpu *vcpu)
7232 struct vcpu_vmx *vmx = to_vmx(vcpu);
7234 if (kvm_mpx_supported()) {
7235 bool mpx_enabled = guest_cpuid_has(vcpu, X86_FEATURE_MPX);
7238 vmx->nested.msrs.entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
7239 vmx->nested.msrs.exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
7241 vmx->nested.msrs.entry_ctls_high &= ~VM_ENTRY_LOAD_BNDCFGS;
7242 vmx->nested.msrs.exit_ctls_high &= ~VM_EXIT_CLEAR_BNDCFGS;
7247 static void update_intel_pt_cfg(struct kvm_vcpu *vcpu)
7249 struct vcpu_vmx *vmx = to_vmx(vcpu);
7250 struct kvm_cpuid_entry2 *best = NULL;
7253 for (i = 0; i < PT_CPUID_LEAVES; i++) {
7254 best = kvm_find_cpuid_entry(vcpu, 0x14, i);
7257 vmx->pt_desc.caps[CPUID_EAX + i*PT_CPUID_REGS_NUM] = best->eax;
7258 vmx->pt_desc.caps[CPUID_EBX + i*PT_CPUID_REGS_NUM] = best->ebx;
7259 vmx->pt_desc.caps[CPUID_ECX + i*PT_CPUID_REGS_NUM] = best->ecx;
7260 vmx->pt_desc.caps[CPUID_EDX + i*PT_CPUID_REGS_NUM] = best->edx;
7263 /* Get the number of configurable Address Ranges for filtering */
7264 vmx->pt_desc.num_address_ranges = intel_pt_validate_cap(vmx->pt_desc.caps,
7265 PT_CAP_num_address_ranges);
7267 /* Initialize and clear the no dependency bits */
7268 vmx->pt_desc.ctl_bitmask = ~(RTIT_CTL_TRACEEN | RTIT_CTL_OS |
7269 RTIT_CTL_USR | RTIT_CTL_TSC_EN | RTIT_CTL_DISRETC |
7270 RTIT_CTL_BRANCH_EN);
7273 * If CPUID.(EAX=14H,ECX=0):EBX[0]=1 CR3Filter can be set otherwise
7274 * will inject an #GP
7276 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_cr3_filtering))
7277 vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_CR3EN;
7280 * If CPUID.(EAX=14H,ECX=0):EBX[1]=1 CYCEn, CycThresh and
7281 * PSBFreq can be set
7283 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc))
7284 vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_CYCLEACC |
7285 RTIT_CTL_CYC_THRESH | RTIT_CTL_PSB_FREQ);
7288 * If CPUID.(EAX=14H,ECX=0):EBX[3]=1 MTCEn and MTCFreq can be set
7290 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc))
7291 vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_MTC_EN |
7292 RTIT_CTL_MTC_RANGE);
7294 /* If CPUID.(EAX=14H,ECX=0):EBX[4]=1 FUPonPTW and PTWEn can be set */
7295 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_ptwrite))
7296 vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_FUP_ON_PTW |
7299 /* If CPUID.(EAX=14H,ECX=0):EBX[5]=1 PwrEvEn can be set */
7300 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_power_event_trace))
7301 vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_PWR_EVT_EN;
7303 /* If CPUID.(EAX=14H,ECX=0):ECX[0]=1 ToPA can be set */
7304 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_topa_output))
7305 vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_TOPA;
7307 /* If CPUID.(EAX=14H,ECX=0):ECX[3]=1 FabricEn can be set */
7308 if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_output_subsys))
7309 vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_FABRIC_EN;
7311 /* unmask address range configure area */
7312 for (i = 0; i < vmx->pt_desc.num_address_ranges; i++)
7313 vmx->pt_desc.ctl_bitmask &= ~(0xfULL << (32 + i * 4));
7316 static void vmx_vcpu_after_set_cpuid(struct kvm_vcpu *vcpu)
7318 struct vcpu_vmx *vmx = to_vmx(vcpu);
7320 /* xsaves_enabled is recomputed in vmx_compute_secondary_exec_control(). */
7321 vcpu->arch.xsaves_enabled = false;
7323 vmx_setup_uret_msrs(vmx);
7325 if (cpu_has_secondary_exec_ctrls())
7326 vmcs_set_secondary_exec_control(vmx,
7327 vmx_secondary_exec_control(vmx));
7329 if (nested_vmx_allowed(vcpu))
7330 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7331 FEAT_CTL_VMX_ENABLED_INSIDE_SMX |
7332 FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX;
7334 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7335 ~(FEAT_CTL_VMX_ENABLED_INSIDE_SMX |
7336 FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX);
7338 if (nested_vmx_allowed(vcpu)) {
7339 nested_vmx_cr_fixed1_bits_update(vcpu);
7340 nested_vmx_entry_exit_ctls_update(vcpu);
7343 if (boot_cpu_has(X86_FEATURE_INTEL_PT) &&
7344 guest_cpuid_has(vcpu, X86_FEATURE_INTEL_PT))
7345 update_intel_pt_cfg(vcpu);
7347 if (boot_cpu_has(X86_FEATURE_RTM)) {
7348 struct vmx_uret_msr *msr;
7349 msr = vmx_find_uret_msr(vmx, MSR_IA32_TSX_CTRL);
7351 bool enabled = guest_cpuid_has(vcpu, X86_FEATURE_RTM);
7352 vmx_set_guest_uret_msr(vmx, msr, enabled ? 0 : TSX_CTRL_RTM_DISABLE);
7356 if (kvm_cpu_cap_has(X86_FEATURE_XFD))
7357 vmx_set_intercept_for_msr(vcpu, MSR_IA32_XFD_ERR, MSR_TYPE_R,
7358 !guest_cpuid_has(vcpu, X86_FEATURE_XFD));
7361 set_cr4_guest_host_mask(vmx);
7363 vmx_write_encls_bitmap(vcpu, NULL);
7364 if (guest_cpuid_has(vcpu, X86_FEATURE_SGX))
7365 vmx->msr_ia32_feature_control_valid_bits |= FEAT_CTL_SGX_ENABLED;
7367 vmx->msr_ia32_feature_control_valid_bits &= ~FEAT_CTL_SGX_ENABLED;
7369 if (guest_cpuid_has(vcpu, X86_FEATURE_SGX_LC))
7370 vmx->msr_ia32_feature_control_valid_bits |=
7371 FEAT_CTL_SGX_LC_ENABLED;
7373 vmx->msr_ia32_feature_control_valid_bits &=
7374 ~FEAT_CTL_SGX_LC_ENABLED;
7376 /* Refresh #PF interception to account for MAXPHYADDR changes. */
7377 vmx_update_exception_bitmap(vcpu);
7380 static __init void vmx_set_cpu_caps(void)
7386 kvm_cpu_cap_set(X86_FEATURE_VMX);
7389 if (kvm_mpx_supported())
7390 kvm_cpu_cap_check_and_set(X86_FEATURE_MPX);
7391 if (!cpu_has_vmx_invpcid())
7392 kvm_cpu_cap_clear(X86_FEATURE_INVPCID);
7393 if (vmx_pt_mode_is_host_guest())
7394 kvm_cpu_cap_check_and_set(X86_FEATURE_INTEL_PT);
7397 kvm_cpu_cap_clear(X86_FEATURE_SGX);
7398 kvm_cpu_cap_clear(X86_FEATURE_SGX_LC);
7399 kvm_cpu_cap_clear(X86_FEATURE_SGX1);
7400 kvm_cpu_cap_clear(X86_FEATURE_SGX2);
7403 if (vmx_umip_emulated())
7404 kvm_cpu_cap_set(X86_FEATURE_UMIP);
7408 if (!cpu_has_vmx_xsaves())
7409 kvm_cpu_cap_clear(X86_FEATURE_XSAVES);
7411 /* CPUID 0x80000001 and 0x7 (RDPID) */
7412 if (!cpu_has_vmx_rdtscp()) {
7413 kvm_cpu_cap_clear(X86_FEATURE_RDTSCP);
7414 kvm_cpu_cap_clear(X86_FEATURE_RDPID);
7417 if (cpu_has_vmx_waitpkg())
7418 kvm_cpu_cap_check_and_set(X86_FEATURE_WAITPKG);
7421 static void vmx_request_immediate_exit(struct kvm_vcpu *vcpu)
7423 to_vmx(vcpu)->req_immediate_exit = true;
7426 static int vmx_check_intercept_io(struct kvm_vcpu *vcpu,
7427 struct x86_instruction_info *info)
7429 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7430 unsigned short port;
7434 if (info->intercept == x86_intercept_in ||
7435 info->intercept == x86_intercept_ins) {
7436 port = info->src_val;
7437 size = info->dst_bytes;
7439 port = info->dst_val;
7440 size = info->src_bytes;
7444 * If the 'use IO bitmaps' VM-execution control is 0, IO instruction
7445 * VM-exits depend on the 'unconditional IO exiting' VM-execution
7448 * Otherwise, IO instruction VM-exits are controlled by the IO bitmaps.
7450 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
7451 intercept = nested_cpu_has(vmcs12,
7452 CPU_BASED_UNCOND_IO_EXITING);
7454 intercept = nested_vmx_check_io_bitmaps(vcpu, port, size);
7456 /* FIXME: produce nested vmexit and return X86EMUL_INTERCEPTED. */
7457 return intercept ? X86EMUL_UNHANDLEABLE : X86EMUL_CONTINUE;
7460 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
7461 struct x86_instruction_info *info,
7462 enum x86_intercept_stage stage,
7463 struct x86_exception *exception)
7465 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7467 switch (info->intercept) {
7469 * RDPID causes #UD if disabled through secondary execution controls.
7470 * Because it is marked as EmulateOnUD, we need to intercept it here.
7471 * Note, RDPID is hidden behind ENABLE_RDTSCP.
7473 case x86_intercept_rdpid:
7474 if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_RDTSCP)) {
7475 exception->vector = UD_VECTOR;
7476 exception->error_code_valid = false;
7477 return X86EMUL_PROPAGATE_FAULT;
7481 case x86_intercept_in:
7482 case x86_intercept_ins:
7483 case x86_intercept_out:
7484 case x86_intercept_outs:
7485 return vmx_check_intercept_io(vcpu, info);
7487 case x86_intercept_lgdt:
7488 case x86_intercept_lidt:
7489 case x86_intercept_lldt:
7490 case x86_intercept_ltr:
7491 case x86_intercept_sgdt:
7492 case x86_intercept_sidt:
7493 case x86_intercept_sldt:
7494 case x86_intercept_str:
7495 if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC))
7496 return X86EMUL_CONTINUE;
7498 /* FIXME: produce nested vmexit and return X86EMUL_INTERCEPTED. */
7501 /* TODO: check more intercepts... */
7506 return X86EMUL_UNHANDLEABLE;
7509 #ifdef CONFIG_X86_64
7510 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
7511 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
7512 u64 divisor, u64 *result)
7514 u64 low = a << shift, high = a >> (64 - shift);
7516 /* To avoid the overflow on divq */
7517 if (high >= divisor)
7520 /* Low hold the result, high hold rem which is discarded */
7521 asm("divq %2\n\t" : "=a" (low), "=d" (high) :
7522 "rm" (divisor), "0" (low), "1" (high));
7528 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc,
7531 struct vcpu_vmx *vmx;
7532 u64 tscl, guest_tscl, delta_tsc, lapic_timer_advance_cycles;
7533 struct kvm_timer *ktimer = &vcpu->arch.apic->lapic_timer;
7537 guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
7538 delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
7539 lapic_timer_advance_cycles = nsec_to_cycles(vcpu,
7540 ktimer->timer_advance_ns);
7542 if (delta_tsc > lapic_timer_advance_cycles)
7543 delta_tsc -= lapic_timer_advance_cycles;
7547 /* Convert to host delta tsc if tsc scaling is enabled */
7548 if (vcpu->arch.l1_tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
7549 delta_tsc && u64_shl_div_u64(delta_tsc,
7550 kvm_tsc_scaling_ratio_frac_bits,
7551 vcpu->arch.l1_tsc_scaling_ratio, &delta_tsc))
7555 * If the delta tsc can't fit in the 32 bit after the multi shift,
7556 * we can't use the preemption timer.
7557 * It's possible that it fits on later vmentries, but checking
7558 * on every vmentry is costly so we just use an hrtimer.
7560 if (delta_tsc >> (cpu_preemption_timer_multi + 32))
7563 vmx->hv_deadline_tsc = tscl + delta_tsc;
7564 *expired = !delta_tsc;
7568 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
7570 to_vmx(vcpu)->hv_deadline_tsc = -1;
7574 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
7576 if (!kvm_pause_in_guest(vcpu->kvm))
7577 shrink_ple_window(vcpu);
7580 void vmx_update_cpu_dirty_logging(struct kvm_vcpu *vcpu)
7582 struct vcpu_vmx *vmx = to_vmx(vcpu);
7584 if (is_guest_mode(vcpu)) {
7585 vmx->nested.update_vmcs01_cpu_dirty_logging = true;
7590 * Note, cpu_dirty_logging_count can be changed concurrent with this
7591 * code, but in that case another update request will be made and so
7592 * the guest will never run with a stale PML value.
7594 if (vcpu->kvm->arch.cpu_dirty_logging_count)
7595 secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_ENABLE_PML);
7597 secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_ENABLE_PML);
7600 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
7602 if (vcpu->arch.mcg_cap & MCG_LMCE_P)
7603 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7604 FEAT_CTL_LMCE_ENABLED;
7606 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7607 ~FEAT_CTL_LMCE_ENABLED;
7610 static int vmx_smi_allowed(struct kvm_vcpu *vcpu, bool for_injection)
7612 /* we need a nested vmexit to enter SMM, postpone if run is pending */
7613 if (to_vmx(vcpu)->nested.nested_run_pending)
7615 return !is_smm(vcpu);
7618 static int vmx_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
7620 struct vcpu_vmx *vmx = to_vmx(vcpu);
7622 vmx->nested.smm.guest_mode = is_guest_mode(vcpu);
7623 if (vmx->nested.smm.guest_mode)
7624 nested_vmx_vmexit(vcpu, -1, 0, 0);
7626 vmx->nested.smm.vmxon = vmx->nested.vmxon;
7627 vmx->nested.vmxon = false;
7628 vmx_clear_hlt(vcpu);
7632 static int vmx_leave_smm(struct kvm_vcpu *vcpu, const char *smstate)
7634 struct vcpu_vmx *vmx = to_vmx(vcpu);
7637 if (vmx->nested.smm.vmxon) {
7638 vmx->nested.vmxon = true;
7639 vmx->nested.smm.vmxon = false;
7642 if (vmx->nested.smm.guest_mode) {
7643 ret = nested_vmx_enter_non_root_mode(vcpu, false);
7647 vmx->nested.smm.guest_mode = false;
7652 static void vmx_enable_smi_window(struct kvm_vcpu *vcpu)
7654 /* RSM will cause a vmexit anyway. */
7657 static bool vmx_apic_init_signal_blocked(struct kvm_vcpu *vcpu)
7659 return to_vmx(vcpu)->nested.vmxon && !is_guest_mode(vcpu);
7662 static void vmx_migrate_timers(struct kvm_vcpu *vcpu)
7664 if (is_guest_mode(vcpu)) {
7665 struct hrtimer *timer = &to_vmx(vcpu)->nested.preemption_timer;
7667 if (hrtimer_try_to_cancel(timer) == 1)
7668 hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED);
7672 static void hardware_unsetup(void)
7674 kvm_set_posted_intr_wakeup_handler(NULL);
7677 nested_vmx_hardware_unsetup();
7682 static bool vmx_check_apicv_inhibit_reasons(ulong bit)
7684 ulong supported = BIT(APICV_INHIBIT_REASON_DISABLE) |
7685 BIT(APICV_INHIBIT_REASON_ABSENT) |
7686 BIT(APICV_INHIBIT_REASON_HYPERV) |
7687 BIT(APICV_INHIBIT_REASON_BLOCKIRQ);
7689 return supported & BIT(bit);
7692 static struct kvm_x86_ops vmx_x86_ops __initdata = {
7693 .name = "kvm_intel",
7695 .hardware_unsetup = hardware_unsetup,
7697 .hardware_enable = hardware_enable,
7698 .hardware_disable = hardware_disable,
7699 .cpu_has_accelerated_tpr = report_flexpriority,
7700 .has_emulated_msr = vmx_has_emulated_msr,
7702 .vm_size = sizeof(struct kvm_vmx),
7703 .vm_init = vmx_vm_init,
7705 .vcpu_create = vmx_create_vcpu,
7706 .vcpu_free = vmx_free_vcpu,
7707 .vcpu_reset = vmx_vcpu_reset,
7709 .prepare_guest_switch = vmx_prepare_switch_to_guest,
7710 .vcpu_load = vmx_vcpu_load,
7711 .vcpu_put = vmx_vcpu_put,
7713 .update_exception_bitmap = vmx_update_exception_bitmap,
7714 .get_msr_feature = vmx_get_msr_feature,
7715 .get_msr = vmx_get_msr,
7716 .set_msr = vmx_set_msr,
7717 .get_segment_base = vmx_get_segment_base,
7718 .get_segment = vmx_get_segment,
7719 .set_segment = vmx_set_segment,
7720 .get_cpl = vmx_get_cpl,
7721 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
7722 .set_cr0 = vmx_set_cr0,
7723 .is_valid_cr4 = vmx_is_valid_cr4,
7724 .set_cr4 = vmx_set_cr4,
7725 .set_efer = vmx_set_efer,
7726 .get_idt = vmx_get_idt,
7727 .set_idt = vmx_set_idt,
7728 .get_gdt = vmx_get_gdt,
7729 .set_gdt = vmx_set_gdt,
7730 .set_dr7 = vmx_set_dr7,
7731 .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
7732 .cache_reg = vmx_cache_reg,
7733 .get_rflags = vmx_get_rflags,
7734 .set_rflags = vmx_set_rflags,
7735 .get_if_flag = vmx_get_if_flag,
7737 .tlb_flush_all = vmx_flush_tlb_all,
7738 .tlb_flush_current = vmx_flush_tlb_current,
7739 .tlb_flush_gva = vmx_flush_tlb_gva,
7740 .tlb_flush_guest = vmx_flush_tlb_guest,
7742 .vcpu_pre_run = vmx_vcpu_pre_run,
7743 .run = vmx_vcpu_run,
7744 .handle_exit = vmx_handle_exit,
7745 .skip_emulated_instruction = vmx_skip_emulated_instruction,
7746 .update_emulated_instruction = vmx_update_emulated_instruction,
7747 .set_interrupt_shadow = vmx_set_interrupt_shadow,
7748 .get_interrupt_shadow = vmx_get_interrupt_shadow,
7749 .patch_hypercall = vmx_patch_hypercall,
7750 .set_irq = vmx_inject_irq,
7751 .set_nmi = vmx_inject_nmi,
7752 .queue_exception = vmx_queue_exception,
7753 .cancel_injection = vmx_cancel_injection,
7754 .interrupt_allowed = vmx_interrupt_allowed,
7755 .nmi_allowed = vmx_nmi_allowed,
7756 .get_nmi_mask = vmx_get_nmi_mask,
7757 .set_nmi_mask = vmx_set_nmi_mask,
7758 .enable_nmi_window = vmx_enable_nmi_window,
7759 .enable_irq_window = vmx_enable_irq_window,
7760 .update_cr8_intercept = vmx_update_cr8_intercept,
7761 .set_virtual_apic_mode = vmx_set_virtual_apic_mode,
7762 .set_apic_access_page_addr = vmx_set_apic_access_page_addr,
7763 .refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
7764 .load_eoi_exitmap = vmx_load_eoi_exitmap,
7765 .apicv_post_state_restore = vmx_apicv_post_state_restore,
7766 .check_apicv_inhibit_reasons = vmx_check_apicv_inhibit_reasons,
7767 .hwapic_irr_update = vmx_hwapic_irr_update,
7768 .hwapic_isr_update = vmx_hwapic_isr_update,
7769 .guest_apic_has_interrupt = vmx_guest_apic_has_interrupt,
7770 .sync_pir_to_irr = vmx_sync_pir_to_irr,
7771 .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
7772 .dy_apicv_has_pending_interrupt = pi_has_pending_interrupt,
7774 .set_tss_addr = vmx_set_tss_addr,
7775 .set_identity_map_addr = vmx_set_identity_map_addr,
7776 .get_mt_mask = vmx_get_mt_mask,
7778 .get_exit_info = vmx_get_exit_info,
7780 .vcpu_after_set_cpuid = vmx_vcpu_after_set_cpuid,
7782 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
7784 .get_l2_tsc_offset = vmx_get_l2_tsc_offset,
7785 .get_l2_tsc_multiplier = vmx_get_l2_tsc_multiplier,
7786 .write_tsc_offset = vmx_write_tsc_offset,
7787 .write_tsc_multiplier = vmx_write_tsc_multiplier,
7789 .load_mmu_pgd = vmx_load_mmu_pgd,
7791 .check_intercept = vmx_check_intercept,
7792 .handle_exit_irqoff = vmx_handle_exit_irqoff,
7794 .request_immediate_exit = vmx_request_immediate_exit,
7796 .sched_in = vmx_sched_in,
7798 .cpu_dirty_log_size = PML_ENTITY_NUM,
7799 .update_cpu_dirty_logging = vmx_update_cpu_dirty_logging,
7801 .pmu_ops = &intel_pmu_ops,
7802 .nested_ops = &vmx_nested_ops,
7804 .update_pi_irte = pi_update_irte,
7805 .start_assignment = vmx_pi_start_assignment,
7807 #ifdef CONFIG_X86_64
7808 .set_hv_timer = vmx_set_hv_timer,
7809 .cancel_hv_timer = vmx_cancel_hv_timer,
7812 .setup_mce = vmx_setup_mce,
7814 .smi_allowed = vmx_smi_allowed,
7815 .enter_smm = vmx_enter_smm,
7816 .leave_smm = vmx_leave_smm,
7817 .enable_smi_window = vmx_enable_smi_window,
7819 .can_emulate_instruction = vmx_can_emulate_instruction,
7820 .apic_init_signal_blocked = vmx_apic_init_signal_blocked,
7821 .migrate_timers = vmx_migrate_timers,
7823 .msr_filter_changed = vmx_msr_filter_changed,
7824 .complete_emulated_msr = kvm_complete_insn_gp,
7826 .vcpu_deliver_sipi_vector = kvm_vcpu_deliver_sipi_vector,
7829 static unsigned int vmx_handle_intel_pt_intr(void)
7831 struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
7833 /* '0' on failure so that the !PT case can use a RET0 static call. */
7834 if (!kvm_arch_pmi_in_guest(vcpu))
7837 kvm_make_request(KVM_REQ_PMI, vcpu);
7838 __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
7839 (unsigned long *)&vcpu->arch.pmu.global_status);
7843 static __init void vmx_setup_user_return_msrs(void)
7847 * Though SYSCALL is only supported in 64-bit mode on Intel CPUs, kvm
7848 * will emulate SYSCALL in legacy mode if the vendor string in guest
7849 * CPUID.0:{EBX,ECX,EDX} is "AuthenticAMD" or "AMDisbetter!" To
7850 * support this emulation, MSR_STAR is included in the list for i386,
7851 * but is never loaded into hardware. MSR_CSTAR is also never loaded
7852 * into hardware and is here purely for emulation purposes.
7854 const u32 vmx_uret_msrs_list[] = {
7855 #ifdef CONFIG_X86_64
7856 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
7858 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
7863 BUILD_BUG_ON(ARRAY_SIZE(vmx_uret_msrs_list) != MAX_NR_USER_RETURN_MSRS);
7865 for (i = 0; i < ARRAY_SIZE(vmx_uret_msrs_list); ++i)
7866 kvm_add_user_return_msr(vmx_uret_msrs_list[i]);
7869 static struct kvm_x86_init_ops vmx_init_ops __initdata;
7871 static __init int hardware_setup(void)
7873 unsigned long host_bndcfgs;
7878 host_idt_base = dt.address;
7880 vmx_setup_user_return_msrs();
7882 if (setup_vmcs_config(&vmcs_config, &vmx_capability) < 0)
7885 if (boot_cpu_has(X86_FEATURE_NX))
7886 kvm_enable_efer_bits(EFER_NX);
7888 if (boot_cpu_has(X86_FEATURE_MPX)) {
7889 rdmsrl(MSR_IA32_BNDCFGS, host_bndcfgs);
7890 WARN_ONCE(host_bndcfgs, "KVM: BNDCFGS in host will be lost");
7893 if (!cpu_has_vmx_mpx())
7894 supported_xcr0 &= ~(XFEATURE_MASK_BNDREGS |
7895 XFEATURE_MASK_BNDCSR);
7897 if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
7898 !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
7901 if (!cpu_has_vmx_ept() ||
7902 !cpu_has_vmx_ept_4levels() ||
7903 !cpu_has_vmx_ept_mt_wb() ||
7904 !cpu_has_vmx_invept_global())
7907 /* NX support is required for shadow paging. */
7908 if (!enable_ept && !boot_cpu_has(X86_FEATURE_NX)) {
7909 pr_err_ratelimited("kvm: NX (Execute Disable) not supported\n");
7913 if (!cpu_has_vmx_ept_ad_bits() || !enable_ept)
7914 enable_ept_ad_bits = 0;
7916 if (!cpu_has_vmx_unrestricted_guest() || !enable_ept)
7917 enable_unrestricted_guest = 0;
7919 if (!cpu_has_vmx_flexpriority())
7920 flexpriority_enabled = 0;
7922 if (!cpu_has_virtual_nmis())
7926 * set_apic_access_page_addr() is used to reload apic access
7927 * page upon invalidation. No need to do anything if not
7928 * using the APIC_ACCESS_ADDR VMCS field.
7930 if (!flexpriority_enabled)
7931 vmx_x86_ops.set_apic_access_page_addr = NULL;
7933 if (!cpu_has_vmx_tpr_shadow())
7934 vmx_x86_ops.update_cr8_intercept = NULL;
7936 #if IS_ENABLED(CONFIG_HYPERV)
7937 if (ms_hyperv.nested_features & HV_X64_NESTED_GUEST_MAPPING_FLUSH
7939 vmx_x86_ops.tlb_remote_flush = hv_remote_flush_tlb;
7940 vmx_x86_ops.tlb_remote_flush_with_range =
7941 hv_remote_flush_tlb_with_range;
7945 if (!cpu_has_vmx_ple()) {
7948 ple_window_grow = 0;
7950 ple_window_shrink = 0;
7953 if (!cpu_has_vmx_apicv())
7956 vmx_x86_ops.sync_pir_to_irr = NULL;
7958 if (cpu_has_vmx_tsc_scaling()) {
7959 kvm_has_tsc_control = true;
7960 kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
7961 kvm_tsc_scaling_ratio_frac_bits = 48;
7964 kvm_has_bus_lock_exit = cpu_has_vmx_bus_lock_detection();
7966 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
7969 kvm_mmu_set_ept_masks(enable_ept_ad_bits,
7970 cpu_has_vmx_ept_execute_only());
7972 kvm_configure_mmu(enable_ept, 0, vmx_get_max_tdp_level(),
7973 ept_caps_to_lpage_level(vmx_capability.ept));
7976 * Only enable PML when hardware supports PML feature, and both EPT
7977 * and EPT A/D bit features are enabled -- PML depends on them to work.
7979 if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
7983 vmx_x86_ops.cpu_dirty_log_size = 0;
7985 if (!cpu_has_vmx_preemption_timer())
7986 enable_preemption_timer = false;
7988 if (enable_preemption_timer) {
7989 u64 use_timer_freq = 5000ULL * 1000 * 1000;
7992 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
7993 cpu_preemption_timer_multi =
7994 vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
7997 use_timer_freq = (u64)tsc_khz * 1000;
7998 use_timer_freq >>= cpu_preemption_timer_multi;
8001 * KVM "disables" the preemption timer by setting it to its max
8002 * value. Don't use the timer if it might cause spurious exits
8003 * at a rate faster than 0.1 Hz (of uninterrupted guest time).
8005 if (use_timer_freq > 0xffffffffu / 10)
8006 enable_preemption_timer = false;
8009 if (!enable_preemption_timer) {
8010 vmx_x86_ops.set_hv_timer = NULL;
8011 vmx_x86_ops.cancel_hv_timer = NULL;
8012 vmx_x86_ops.request_immediate_exit = __kvm_request_immediate_exit;
8015 kvm_mce_cap_supported |= MCG_LMCE_P;
8017 if (pt_mode != PT_MODE_SYSTEM && pt_mode != PT_MODE_HOST_GUEST)
8019 if (!enable_ept || !cpu_has_vmx_intel_pt())
8020 pt_mode = PT_MODE_SYSTEM;
8021 if (pt_mode == PT_MODE_HOST_GUEST)
8022 vmx_init_ops.handle_intel_pt_intr = vmx_handle_intel_pt_intr;
8024 vmx_init_ops.handle_intel_pt_intr = NULL;
8026 setup_default_sgx_lepubkeyhash();
8029 nested_vmx_setup_ctls_msrs(&vmcs_config.nested,
8030 vmx_capability.ept);
8032 r = nested_vmx_hardware_setup(kvm_vmx_exit_handlers);
8039 r = alloc_kvm_area();
8041 nested_vmx_hardware_unsetup();
8043 kvm_set_posted_intr_wakeup_handler(pi_wakeup_handler);
8048 static struct kvm_x86_init_ops vmx_init_ops __initdata = {
8049 .cpu_has_kvm_support = cpu_has_kvm_support,
8050 .disabled_by_bios = vmx_disabled_by_bios,
8051 .check_processor_compatibility = vmx_check_processor_compat,
8052 .hardware_setup = hardware_setup,
8053 .handle_intel_pt_intr = NULL,
8055 .runtime_ops = &vmx_x86_ops,
8058 static void vmx_cleanup_l1d_flush(void)
8060 if (vmx_l1d_flush_pages) {
8061 free_pages((unsigned long)vmx_l1d_flush_pages, L1D_CACHE_ORDER);
8062 vmx_l1d_flush_pages = NULL;
8064 /* Restore state so sysfs ignores VMX */
8065 l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
8068 static void vmx_exit(void)
8070 #ifdef CONFIG_KEXEC_CORE
8071 RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
8077 #if IS_ENABLED(CONFIG_HYPERV)
8078 if (static_branch_unlikely(&enable_evmcs)) {
8080 struct hv_vp_assist_page *vp_ap;
8082 * Reset everything to support using non-enlightened VMCS
8083 * access later (e.g. when we reload the module with
8084 * enlightened_vmcs=0)
8086 for_each_online_cpu(cpu) {
8087 vp_ap = hv_get_vp_assist_page(cpu);
8092 vp_ap->nested_control.features.directhypercall = 0;
8093 vp_ap->current_nested_vmcs = 0;
8094 vp_ap->enlighten_vmentry = 0;
8097 static_branch_disable(&enable_evmcs);
8100 vmx_cleanup_l1d_flush();
8102 allow_smaller_maxphyaddr = false;
8104 module_exit(vmx_exit);
8106 static int __init vmx_init(void)
8110 #if IS_ENABLED(CONFIG_HYPERV)
8112 * Enlightened VMCS usage should be recommended and the host needs
8113 * to support eVMCS v1 or above. We can also disable eVMCS support
8114 * with module parameter.
8116 if (enlightened_vmcs &&
8117 ms_hyperv.hints & HV_X64_ENLIGHTENED_VMCS_RECOMMENDED &&
8118 (ms_hyperv.nested_features & HV_X64_ENLIGHTENED_VMCS_VERSION) >=
8119 KVM_EVMCS_VERSION) {
8122 /* Check that we have assist pages on all online CPUs */
8123 for_each_online_cpu(cpu) {
8124 if (!hv_get_vp_assist_page(cpu)) {
8125 enlightened_vmcs = false;
8130 if (enlightened_vmcs) {
8131 pr_info("KVM: vmx: using Hyper-V Enlightened VMCS\n");
8132 static_branch_enable(&enable_evmcs);
8135 if (ms_hyperv.nested_features & HV_X64_NESTED_DIRECT_FLUSH)
8136 vmx_x86_ops.enable_direct_tlbflush
8137 = hv_enable_direct_tlbflush;
8140 enlightened_vmcs = false;
8144 r = kvm_init(&vmx_init_ops, sizeof(struct vcpu_vmx),
8145 __alignof__(struct vcpu_vmx), THIS_MODULE);
8150 * Must be called after kvm_init() so enable_ept is properly set
8151 * up. Hand the parameter mitigation value in which was stored in
8152 * the pre module init parser. If no parameter was given, it will
8153 * contain 'auto' which will be turned into the default 'cond'
8156 r = vmx_setup_l1d_flush(vmentry_l1d_flush_param);
8162 for_each_possible_cpu(cpu) {
8163 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
8168 #ifdef CONFIG_KEXEC_CORE
8169 rcu_assign_pointer(crash_vmclear_loaded_vmcss,
8170 crash_vmclear_local_loaded_vmcss);
8172 vmx_check_vmcs12_offsets();
8175 * Shadow paging doesn't have a (further) performance penalty
8176 * from GUEST_MAXPHYADDR < HOST_MAXPHYADDR so enable it
8180 allow_smaller_maxphyaddr = true;
8184 module_init(vmx_init);