2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affilates.
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
22 #include <linux/kvm_host.h>
23 #include <linux/module.h>
24 #include <linux/kernel.h>
26 #include <linux/highmem.h>
27 #include <linux/sched.h>
28 #include <linux/moduleparam.h>
29 #include <linux/ftrace_event.h>
30 #include <linux/slab.h>
31 #include <linux/tboot.h>
32 #include "kvm_cache_regs.h"
38 #include <asm/virtext.h>
45 #define __ex(x) __kvm_handle_fault_on_reboot(x)
47 MODULE_AUTHOR("Qumranet");
48 MODULE_LICENSE("GPL");
50 static int __read_mostly bypass_guest_pf = 1;
51 module_param(bypass_guest_pf, bool, S_IRUGO);
53 static int __read_mostly enable_vpid = 1;
54 module_param_named(vpid, enable_vpid, bool, 0444);
56 static int __read_mostly flexpriority_enabled = 1;
57 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
59 static int __read_mostly enable_ept = 1;
60 module_param_named(ept, enable_ept, bool, S_IRUGO);
62 static int __read_mostly enable_unrestricted_guest = 1;
63 module_param_named(unrestricted_guest,
64 enable_unrestricted_guest, bool, S_IRUGO);
66 static int __read_mostly emulate_invalid_guest_state = 0;
67 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
69 static int __read_mostly vmm_exclusive = 1;
70 module_param(vmm_exclusive, bool, S_IRUGO);
72 #define KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST \
73 (X86_CR0_WP | X86_CR0_NE | X86_CR0_NW | X86_CR0_CD)
74 #define KVM_GUEST_CR0_MASK \
75 (KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
76 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST \
77 (X86_CR0_WP | X86_CR0_NE)
78 #define KVM_VM_CR0_ALWAYS_ON \
79 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
80 #define KVM_CR4_GUEST_OWNED_BITS \
81 (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \
84 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
85 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
87 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
90 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
91 * ple_gap: upper bound on the amount of time between two successive
92 * executions of PAUSE in a loop. Also indicate if ple enabled.
93 * According to test, this time is usually small than 41 cycles.
94 * ple_window: upper bound on the amount of time a guest is allowed to execute
95 * in a PAUSE loop. Tests indicate that most spinlocks are held for
96 * less than 2^12 cycles
97 * Time is measured based on a counter that runs at the same rate as the TSC,
98 * refer SDM volume 3b section 21.6.13 & 22.1.3.
100 #define KVM_VMX_DEFAULT_PLE_GAP 41
101 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
102 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
103 module_param(ple_gap, int, S_IRUGO);
105 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
106 module_param(ple_window, int, S_IRUGO);
108 #define NR_AUTOLOAD_MSRS 1
116 struct shared_msr_entry {
123 struct kvm_vcpu vcpu;
124 struct list_head local_vcpus_link;
125 unsigned long host_rsp;
128 u32 idt_vectoring_info;
129 struct shared_msr_entry *guest_msrs;
133 u64 msr_host_kernel_gs_base;
134 u64 msr_guest_kernel_gs_base;
137 struct msr_autoload {
139 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
140 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
144 u16 fs_sel, gs_sel, ldt_sel;
145 int gs_ldt_reload_needed;
146 int fs_reload_needed;
151 struct kvm_save_segment {
156 } tr, es, ds, fs, gs;
164 bool emulation_required;
166 /* Support for vnmi-less CPUs */
167 int soft_vnmi_blocked;
169 s64 vnmi_blocked_time;
175 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
177 return container_of(vcpu, struct vcpu_vmx, vcpu);
180 static int init_rmode(struct kvm *kvm);
181 static u64 construct_eptp(unsigned long root_hpa);
182 static void kvm_cpu_vmxon(u64 addr);
183 static void kvm_cpu_vmxoff(void);
185 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
186 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
187 static DEFINE_PER_CPU(struct list_head, vcpus_on_cpu);
188 static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
190 static unsigned long *vmx_io_bitmap_a;
191 static unsigned long *vmx_io_bitmap_b;
192 static unsigned long *vmx_msr_bitmap_legacy;
193 static unsigned long *vmx_msr_bitmap_longmode;
195 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
196 static DEFINE_SPINLOCK(vmx_vpid_lock);
198 static struct vmcs_config {
202 u32 pin_based_exec_ctrl;
203 u32 cpu_based_exec_ctrl;
204 u32 cpu_based_2nd_exec_ctrl;
209 static struct vmx_capability {
214 #define VMX_SEGMENT_FIELD(seg) \
215 [VCPU_SREG_##seg] = { \
216 .selector = GUEST_##seg##_SELECTOR, \
217 .base = GUEST_##seg##_BASE, \
218 .limit = GUEST_##seg##_LIMIT, \
219 .ar_bytes = GUEST_##seg##_AR_BYTES, \
222 static struct kvm_vmx_segment_field {
227 } kvm_vmx_segment_fields[] = {
228 VMX_SEGMENT_FIELD(CS),
229 VMX_SEGMENT_FIELD(DS),
230 VMX_SEGMENT_FIELD(ES),
231 VMX_SEGMENT_FIELD(FS),
232 VMX_SEGMENT_FIELD(GS),
233 VMX_SEGMENT_FIELD(SS),
234 VMX_SEGMENT_FIELD(TR),
235 VMX_SEGMENT_FIELD(LDTR),
238 static u64 host_efer;
240 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
243 * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
244 * away by decrementing the array size.
246 static const u32 vmx_msr_index[] = {
248 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
250 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
252 #define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
254 static inline bool is_page_fault(u32 intr_info)
256 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
257 INTR_INFO_VALID_MASK)) ==
258 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
261 static inline bool is_no_device(u32 intr_info)
263 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
264 INTR_INFO_VALID_MASK)) ==
265 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
268 static inline bool is_invalid_opcode(u32 intr_info)
270 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
271 INTR_INFO_VALID_MASK)) ==
272 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
275 static inline bool is_external_interrupt(u32 intr_info)
277 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
278 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
281 static inline bool is_machine_check(u32 intr_info)
283 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
284 INTR_INFO_VALID_MASK)) ==
285 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
288 static inline bool cpu_has_vmx_msr_bitmap(void)
290 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
293 static inline bool cpu_has_vmx_tpr_shadow(void)
295 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
298 static inline bool vm_need_tpr_shadow(struct kvm *kvm)
300 return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm));
303 static inline bool cpu_has_secondary_exec_ctrls(void)
305 return vmcs_config.cpu_based_exec_ctrl &
306 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
309 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
311 return vmcs_config.cpu_based_2nd_exec_ctrl &
312 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
315 static inline bool cpu_has_vmx_flexpriority(void)
317 return cpu_has_vmx_tpr_shadow() &&
318 cpu_has_vmx_virtualize_apic_accesses();
321 static inline bool cpu_has_vmx_ept_execute_only(void)
323 return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
326 static inline bool cpu_has_vmx_eptp_uncacheable(void)
328 return vmx_capability.ept & VMX_EPTP_UC_BIT;
331 static inline bool cpu_has_vmx_eptp_writeback(void)
333 return vmx_capability.ept & VMX_EPTP_WB_BIT;
336 static inline bool cpu_has_vmx_ept_2m_page(void)
338 return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
341 static inline bool cpu_has_vmx_ept_1g_page(void)
343 return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
346 static inline bool cpu_has_vmx_ept_4levels(void)
348 return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
351 static inline bool cpu_has_vmx_invept_individual_addr(void)
353 return vmx_capability.ept & VMX_EPT_EXTENT_INDIVIDUAL_BIT;
356 static inline bool cpu_has_vmx_invept_context(void)
358 return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
361 static inline bool cpu_has_vmx_invept_global(void)
363 return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
366 static inline bool cpu_has_vmx_invvpid_single(void)
368 return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
371 static inline bool cpu_has_vmx_invvpid_global(void)
373 return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
376 static inline bool cpu_has_vmx_ept(void)
378 return vmcs_config.cpu_based_2nd_exec_ctrl &
379 SECONDARY_EXEC_ENABLE_EPT;
382 static inline bool cpu_has_vmx_unrestricted_guest(void)
384 return vmcs_config.cpu_based_2nd_exec_ctrl &
385 SECONDARY_EXEC_UNRESTRICTED_GUEST;
388 static inline bool cpu_has_vmx_ple(void)
390 return vmcs_config.cpu_based_2nd_exec_ctrl &
391 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
394 static inline bool vm_need_virtualize_apic_accesses(struct kvm *kvm)
396 return flexpriority_enabled && irqchip_in_kernel(kvm);
399 static inline bool cpu_has_vmx_vpid(void)
401 return vmcs_config.cpu_based_2nd_exec_ctrl &
402 SECONDARY_EXEC_ENABLE_VPID;
405 static inline bool cpu_has_vmx_rdtscp(void)
407 return vmcs_config.cpu_based_2nd_exec_ctrl &
408 SECONDARY_EXEC_RDTSCP;
411 static inline bool cpu_has_virtual_nmis(void)
413 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
416 static inline bool cpu_has_vmx_wbinvd_exit(void)
418 return vmcs_config.cpu_based_2nd_exec_ctrl &
419 SECONDARY_EXEC_WBINVD_EXITING;
422 static inline bool report_flexpriority(void)
424 return flexpriority_enabled;
427 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
431 for (i = 0; i < vmx->nmsrs; ++i)
432 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
437 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
443 } operand = { vpid, 0, gva };
445 asm volatile (__ex(ASM_VMX_INVVPID)
446 /* CF==1 or ZF==1 --> rc = -1 */
448 : : "a"(&operand), "c"(ext) : "cc", "memory");
451 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
455 } operand = {eptp, gpa};
457 asm volatile (__ex(ASM_VMX_INVEPT)
458 /* CF==1 or ZF==1 --> rc = -1 */
459 "; ja 1f ; ud2 ; 1:\n"
460 : : "a" (&operand), "c" (ext) : "cc", "memory");
463 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
467 i = __find_msr_index(vmx, msr);
469 return &vmx->guest_msrs[i];
473 static void vmcs_clear(struct vmcs *vmcs)
475 u64 phys_addr = __pa(vmcs);
478 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
479 : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
482 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
486 static void vmcs_load(struct vmcs *vmcs)
488 u64 phys_addr = __pa(vmcs);
491 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
492 : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
495 printk(KERN_ERR "kvm: vmptrld %p/%llx fail\n",
499 static void __vcpu_clear(void *arg)
501 struct vcpu_vmx *vmx = arg;
502 int cpu = raw_smp_processor_id();
504 if (vmx->vcpu.cpu == cpu)
505 vmcs_clear(vmx->vmcs);
506 if (per_cpu(current_vmcs, cpu) == vmx->vmcs)
507 per_cpu(current_vmcs, cpu) = NULL;
508 rdtscll(vmx->vcpu.arch.host_tsc);
509 list_del(&vmx->local_vcpus_link);
514 static void vcpu_clear(struct vcpu_vmx *vmx)
516 if (vmx->vcpu.cpu == -1)
518 smp_call_function_single(vmx->vcpu.cpu, __vcpu_clear, vmx, 1);
521 static inline void vpid_sync_vcpu_single(struct vcpu_vmx *vmx)
526 if (cpu_has_vmx_invvpid_single())
527 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
530 static inline void vpid_sync_vcpu_global(void)
532 if (cpu_has_vmx_invvpid_global())
533 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
536 static inline void vpid_sync_context(struct vcpu_vmx *vmx)
538 if (cpu_has_vmx_invvpid_single())
539 vpid_sync_vcpu_single(vmx);
541 vpid_sync_vcpu_global();
544 static inline void ept_sync_global(void)
546 if (cpu_has_vmx_invept_global())
547 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
550 static inline void ept_sync_context(u64 eptp)
553 if (cpu_has_vmx_invept_context())
554 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
560 static inline void ept_sync_individual_addr(u64 eptp, gpa_t gpa)
563 if (cpu_has_vmx_invept_individual_addr())
564 __invept(VMX_EPT_EXTENT_INDIVIDUAL_ADDR,
567 ept_sync_context(eptp);
571 static unsigned long vmcs_readl(unsigned long field)
575 asm volatile (__ex(ASM_VMX_VMREAD_RDX_RAX)
576 : "=a"(value) : "d"(field) : "cc");
580 static u16 vmcs_read16(unsigned long field)
582 return vmcs_readl(field);
585 static u32 vmcs_read32(unsigned long field)
587 return vmcs_readl(field);
590 static u64 vmcs_read64(unsigned long field)
593 return vmcs_readl(field);
595 return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
599 static noinline void vmwrite_error(unsigned long field, unsigned long value)
601 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
602 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
606 static void vmcs_writel(unsigned long field, unsigned long value)
610 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
611 : "=q"(error) : "a"(value), "d"(field) : "cc");
613 vmwrite_error(field, value);
616 static void vmcs_write16(unsigned long field, u16 value)
618 vmcs_writel(field, value);
621 static void vmcs_write32(unsigned long field, u32 value)
623 vmcs_writel(field, value);
626 static void vmcs_write64(unsigned long field, u64 value)
628 vmcs_writel(field, value);
629 #ifndef CONFIG_X86_64
631 vmcs_writel(field+1, value >> 32);
635 static void vmcs_clear_bits(unsigned long field, u32 mask)
637 vmcs_writel(field, vmcs_readl(field) & ~mask);
640 static void vmcs_set_bits(unsigned long field, u32 mask)
642 vmcs_writel(field, vmcs_readl(field) | mask);
645 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
649 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
650 (1u << NM_VECTOR) | (1u << DB_VECTOR);
651 if ((vcpu->guest_debug &
652 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
653 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
654 eb |= 1u << BP_VECTOR;
655 if (to_vmx(vcpu)->rmode.vm86_active)
658 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
659 if (vcpu->fpu_active)
660 eb &= ~(1u << NM_VECTOR);
661 vmcs_write32(EXCEPTION_BITMAP, eb);
664 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
667 struct msr_autoload *m = &vmx->msr_autoload;
669 for (i = 0; i < m->nr; ++i)
670 if (m->guest[i].index == msr)
676 m->guest[i] = m->guest[m->nr];
677 m->host[i] = m->host[m->nr];
678 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
679 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
682 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
683 u64 guest_val, u64 host_val)
686 struct msr_autoload *m = &vmx->msr_autoload;
688 for (i = 0; i < m->nr; ++i)
689 if (m->guest[i].index == msr)
694 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
695 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
698 m->guest[i].index = msr;
699 m->guest[i].value = guest_val;
700 m->host[i].index = msr;
701 m->host[i].value = host_val;
704 static void reload_tss(void)
707 * VT restores TR but not its size. Useless.
710 struct desc_struct *descs;
712 native_store_gdt(&gdt);
713 descs = (void *)gdt.address;
714 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
718 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
723 guest_efer = vmx->vcpu.arch.efer;
726 * NX is emulated; LMA and LME handled by hardware; SCE meaninless
729 ignore_bits = EFER_NX | EFER_SCE;
731 ignore_bits |= EFER_LMA | EFER_LME;
732 /* SCE is meaningful only in long mode on Intel */
733 if (guest_efer & EFER_LMA)
734 ignore_bits &= ~(u64)EFER_SCE;
736 guest_efer &= ~ignore_bits;
737 guest_efer |= host_efer & ignore_bits;
738 vmx->guest_msrs[efer_offset].data = guest_efer;
739 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
741 clear_atomic_switch_msr(vmx, MSR_EFER);
742 /* On ept, can't emulate nx, and must switch nx atomically */
743 if (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX)) {
744 guest_efer = vmx->vcpu.arch.efer;
745 if (!(guest_efer & EFER_LMA))
746 guest_efer &= ~EFER_LME;
747 add_atomic_switch_msr(vmx, MSR_EFER, guest_efer, host_efer);
754 static unsigned long segment_base(u16 selector)
757 struct desc_struct *d;
758 unsigned long table_base;
761 if (!(selector & ~3))
764 native_store_gdt(&gdt);
765 table_base = gdt.address;
767 if (selector & 4) { /* from ldt */
768 u16 ldt_selector = kvm_read_ldt();
770 if (!(ldt_selector & ~3))
773 table_base = segment_base(ldt_selector);
775 d = (struct desc_struct *)(table_base + (selector & ~7));
776 v = get_desc_base(d);
778 if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
779 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
784 static inline unsigned long kvm_read_tr_base(void)
787 asm("str %0" : "=g"(tr));
788 return segment_base(tr);
791 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
793 struct vcpu_vmx *vmx = to_vmx(vcpu);
796 if (vmx->host_state.loaded)
799 vmx->host_state.loaded = 1;
801 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
802 * allow segment selectors with cpl > 0 or ti == 1.
804 vmx->host_state.ldt_sel = kvm_read_ldt();
805 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
806 savesegment(fs, vmx->host_state.fs_sel);
807 if (!(vmx->host_state.fs_sel & 7)) {
808 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
809 vmx->host_state.fs_reload_needed = 0;
811 vmcs_write16(HOST_FS_SELECTOR, 0);
812 vmx->host_state.fs_reload_needed = 1;
814 savesegment(gs, vmx->host_state.gs_sel);
815 if (!(vmx->host_state.gs_sel & 7))
816 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
818 vmcs_write16(HOST_GS_SELECTOR, 0);
819 vmx->host_state.gs_ldt_reload_needed = 1;
823 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
824 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
826 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
827 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
831 if (is_long_mode(&vmx->vcpu)) {
832 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
833 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
836 for (i = 0; i < vmx->save_nmsrs; ++i)
837 kvm_set_shared_msr(vmx->guest_msrs[i].index,
838 vmx->guest_msrs[i].data,
839 vmx->guest_msrs[i].mask);
842 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
844 if (!vmx->host_state.loaded)
847 ++vmx->vcpu.stat.host_state_reload;
848 vmx->host_state.loaded = 0;
849 if (vmx->host_state.fs_reload_needed)
850 loadsegment(fs, vmx->host_state.fs_sel);
851 if (vmx->host_state.gs_ldt_reload_needed) {
852 kvm_load_ldt(vmx->host_state.ldt_sel);
854 load_gs_index(vmx->host_state.gs_sel);
855 wrmsrl(MSR_KERNEL_GS_BASE, current->thread.gs);
857 loadsegment(gs, vmx->host_state.gs_sel);
862 if (is_long_mode(&vmx->vcpu)) {
863 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
864 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
867 if (current_thread_info()->status & TS_USEDFPU)
869 load_gdt(&__get_cpu_var(host_gdt));
872 static void vmx_load_host_state(struct vcpu_vmx *vmx)
875 __vmx_load_host_state(vmx);
880 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
881 * vcpu mutex is already taken.
883 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
885 struct vcpu_vmx *vmx = to_vmx(vcpu);
886 u64 tsc_this, delta, new_offset;
887 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
890 kvm_cpu_vmxon(phys_addr);
891 else if (vcpu->cpu != cpu)
894 if (per_cpu(current_vmcs, cpu) != vmx->vmcs) {
895 per_cpu(current_vmcs, cpu) = vmx->vmcs;
896 vmcs_load(vmx->vmcs);
899 if (vcpu->cpu != cpu) {
901 unsigned long sysenter_esp;
903 kvm_migrate_timers(vcpu);
904 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
906 list_add(&vmx->local_vcpus_link,
907 &per_cpu(vcpus_on_cpu, cpu));
912 * Linux uses per-cpu TSS and GDT, so set these when switching
915 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
916 native_store_gdt(&dt);
917 vmcs_writel(HOST_GDTR_BASE, dt.address); /* 22.2.4 */
919 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
920 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
923 * Make sure the time stamp counter is monotonous.
926 if (tsc_this < vcpu->arch.host_tsc) {
927 delta = vcpu->arch.host_tsc - tsc_this;
928 new_offset = vmcs_read64(TSC_OFFSET) + delta;
929 vmcs_write64(TSC_OFFSET, new_offset);
934 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
936 __vmx_load_host_state(to_vmx(vcpu));
937 if (!vmm_exclusive) {
938 __vcpu_clear(to_vmx(vcpu));
943 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
947 if (vcpu->fpu_active)
949 vcpu->fpu_active = 1;
950 cr0 = vmcs_readl(GUEST_CR0);
951 cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
952 cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
953 vmcs_writel(GUEST_CR0, cr0);
954 update_exception_bitmap(vcpu);
955 vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
956 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
959 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
961 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
963 vmx_decache_cr0_guest_bits(vcpu);
964 vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
965 update_exception_bitmap(vcpu);
966 vcpu->arch.cr0_guest_owned_bits = 0;
967 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
968 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
971 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
973 unsigned long rflags, save_rflags;
975 rflags = vmcs_readl(GUEST_RFLAGS);
976 if (to_vmx(vcpu)->rmode.vm86_active) {
977 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
978 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
979 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
984 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
986 if (to_vmx(vcpu)->rmode.vm86_active) {
987 to_vmx(vcpu)->rmode.save_rflags = rflags;
988 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
990 vmcs_writel(GUEST_RFLAGS, rflags);
993 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
995 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
998 if (interruptibility & GUEST_INTR_STATE_STI)
999 ret |= KVM_X86_SHADOW_INT_STI;
1000 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1001 ret |= KVM_X86_SHADOW_INT_MOV_SS;
1006 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1008 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1009 u32 interruptibility = interruptibility_old;
1011 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1013 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1014 interruptibility |= GUEST_INTR_STATE_MOV_SS;
1015 else if (mask & KVM_X86_SHADOW_INT_STI)
1016 interruptibility |= GUEST_INTR_STATE_STI;
1018 if ((interruptibility != interruptibility_old))
1019 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1022 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
1026 rip = kvm_rip_read(vcpu);
1027 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1028 kvm_rip_write(vcpu, rip);
1030 /* skipping an emulated instruction also counts */
1031 vmx_set_interrupt_shadow(vcpu, 0);
1034 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
1035 bool has_error_code, u32 error_code,
1038 struct vcpu_vmx *vmx = to_vmx(vcpu);
1039 u32 intr_info = nr | INTR_INFO_VALID_MASK;
1041 if (has_error_code) {
1042 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1043 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1046 if (vmx->rmode.vm86_active) {
1047 vmx->rmode.irq.pending = true;
1048 vmx->rmode.irq.vector = nr;
1049 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
1050 if (kvm_exception_is_soft(nr))
1051 vmx->rmode.irq.rip +=
1052 vmx->vcpu.arch.event_exit_inst_len;
1053 intr_info |= INTR_TYPE_SOFT_INTR;
1054 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1055 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
1056 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
1060 if (kvm_exception_is_soft(nr)) {
1061 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1062 vmx->vcpu.arch.event_exit_inst_len);
1063 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1065 intr_info |= INTR_TYPE_HARD_EXCEPTION;
1067 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1070 static bool vmx_rdtscp_supported(void)
1072 return cpu_has_vmx_rdtscp();
1076 * Swap MSR entry in host/guest MSR entry array.
1078 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1080 struct shared_msr_entry tmp;
1082 tmp = vmx->guest_msrs[to];
1083 vmx->guest_msrs[to] = vmx->guest_msrs[from];
1084 vmx->guest_msrs[from] = tmp;
1088 * Set up the vmcs to automatically save and restore system
1089 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
1090 * mode, as fiddling with msrs is very expensive.
1092 static void setup_msrs(struct vcpu_vmx *vmx)
1094 int save_nmsrs, index;
1095 unsigned long *msr_bitmap;
1097 vmx_load_host_state(vmx);
1099 #ifdef CONFIG_X86_64
1100 if (is_long_mode(&vmx->vcpu)) {
1101 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1103 move_msr_up(vmx, index, save_nmsrs++);
1104 index = __find_msr_index(vmx, MSR_LSTAR);
1106 move_msr_up(vmx, index, save_nmsrs++);
1107 index = __find_msr_index(vmx, MSR_CSTAR);
1109 move_msr_up(vmx, index, save_nmsrs++);
1110 index = __find_msr_index(vmx, MSR_TSC_AUX);
1111 if (index >= 0 && vmx->rdtscp_enabled)
1112 move_msr_up(vmx, index, save_nmsrs++);
1114 * MSR_STAR is only needed on long mode guests, and only
1115 * if efer.sce is enabled.
1117 index = __find_msr_index(vmx, MSR_STAR);
1118 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
1119 move_msr_up(vmx, index, save_nmsrs++);
1122 index = __find_msr_index(vmx, MSR_EFER);
1123 if (index >= 0 && update_transition_efer(vmx, index))
1124 move_msr_up(vmx, index, save_nmsrs++);
1126 vmx->save_nmsrs = save_nmsrs;
1128 if (cpu_has_vmx_msr_bitmap()) {
1129 if (is_long_mode(&vmx->vcpu))
1130 msr_bitmap = vmx_msr_bitmap_longmode;
1132 msr_bitmap = vmx_msr_bitmap_legacy;
1134 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
1139 * reads and returns guest's timestamp counter "register"
1140 * guest_tsc = host_tsc + tsc_offset -- 21.3
1142 static u64 guest_read_tsc(void)
1144 u64 host_tsc, tsc_offset;
1147 tsc_offset = vmcs_read64(TSC_OFFSET);
1148 return host_tsc + tsc_offset;
1152 * writes 'guest_tsc' into guest's timestamp counter "register"
1153 * guest_tsc = host_tsc + tsc_offset ==> tsc_offset = guest_tsc - host_tsc
1155 static void guest_write_tsc(u64 guest_tsc, u64 host_tsc)
1157 vmcs_write64(TSC_OFFSET, guest_tsc - host_tsc);
1161 * Reads an msr value (of 'msr_index') into 'pdata'.
1162 * Returns 0 on success, non-0 otherwise.
1163 * Assumes vcpu_load() was already called.
1165 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1168 struct shared_msr_entry *msr;
1171 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
1175 switch (msr_index) {
1176 #ifdef CONFIG_X86_64
1178 data = vmcs_readl(GUEST_FS_BASE);
1181 data = vmcs_readl(GUEST_GS_BASE);
1183 case MSR_KERNEL_GS_BASE:
1184 vmx_load_host_state(to_vmx(vcpu));
1185 data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
1189 return kvm_get_msr_common(vcpu, msr_index, pdata);
1191 data = guest_read_tsc();
1193 case MSR_IA32_SYSENTER_CS:
1194 data = vmcs_read32(GUEST_SYSENTER_CS);
1196 case MSR_IA32_SYSENTER_EIP:
1197 data = vmcs_readl(GUEST_SYSENTER_EIP);
1199 case MSR_IA32_SYSENTER_ESP:
1200 data = vmcs_readl(GUEST_SYSENTER_ESP);
1203 if (!to_vmx(vcpu)->rdtscp_enabled)
1205 /* Otherwise falls through */
1207 vmx_load_host_state(to_vmx(vcpu));
1208 msr = find_msr_entry(to_vmx(vcpu), msr_index);
1210 vmx_load_host_state(to_vmx(vcpu));
1214 return kvm_get_msr_common(vcpu, msr_index, pdata);
1222 * Writes msr value into into the appropriate "register".
1223 * Returns 0 on success, non-0 otherwise.
1224 * Assumes vcpu_load() was already called.
1226 static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1228 struct vcpu_vmx *vmx = to_vmx(vcpu);
1229 struct shared_msr_entry *msr;
1233 switch (msr_index) {
1235 vmx_load_host_state(vmx);
1236 ret = kvm_set_msr_common(vcpu, msr_index, data);
1238 #ifdef CONFIG_X86_64
1240 vmcs_writel(GUEST_FS_BASE, data);
1243 vmcs_writel(GUEST_GS_BASE, data);
1245 case MSR_KERNEL_GS_BASE:
1246 vmx_load_host_state(vmx);
1247 vmx->msr_guest_kernel_gs_base = data;
1250 case MSR_IA32_SYSENTER_CS:
1251 vmcs_write32(GUEST_SYSENTER_CS, data);
1253 case MSR_IA32_SYSENTER_EIP:
1254 vmcs_writel(GUEST_SYSENTER_EIP, data);
1256 case MSR_IA32_SYSENTER_ESP:
1257 vmcs_writel(GUEST_SYSENTER_ESP, data);
1261 guest_write_tsc(data, host_tsc);
1263 case MSR_IA32_CR_PAT:
1264 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
1265 vmcs_write64(GUEST_IA32_PAT, data);
1266 vcpu->arch.pat = data;
1269 ret = kvm_set_msr_common(vcpu, msr_index, data);
1272 if (!vmx->rdtscp_enabled)
1274 /* Check reserved bit, higher 32 bits should be zero */
1275 if ((data >> 32) != 0)
1277 /* Otherwise falls through */
1279 msr = find_msr_entry(vmx, msr_index);
1281 vmx_load_host_state(vmx);
1285 ret = kvm_set_msr_common(vcpu, msr_index, data);
1291 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1293 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
1296 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
1299 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
1301 case VCPU_EXREG_PDPTR:
1303 ept_save_pdptrs(vcpu);
1310 static void set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
1312 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1313 vmcs_writel(GUEST_DR7, dbg->arch.debugreg[7]);
1315 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
1317 update_exception_bitmap(vcpu);
1320 static __init int cpu_has_kvm_support(void)
1322 return cpu_has_vmx();
1325 static __init int vmx_disabled_by_bios(void)
1329 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
1330 if (msr & FEATURE_CONTROL_LOCKED) {
1331 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
1334 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
1335 && !tboot_enabled())
1340 /* locked but not enabled */
1343 static void kvm_cpu_vmxon(u64 addr)
1345 asm volatile (ASM_VMX_VMXON_RAX
1346 : : "a"(&addr), "m"(addr)
1350 static int hardware_enable(void *garbage)
1352 int cpu = raw_smp_processor_id();
1353 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1356 if (read_cr4() & X86_CR4_VMXE)
1359 INIT_LIST_HEAD(&per_cpu(vcpus_on_cpu, cpu));
1360 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
1362 test_bits = FEATURE_CONTROL_LOCKED;
1363 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
1364 if (tboot_enabled())
1365 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
1367 if ((old & test_bits) != test_bits) {
1368 /* enable and lock */
1369 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
1371 write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
1373 if (vmm_exclusive) {
1374 kvm_cpu_vmxon(phys_addr);
1378 store_gdt(&__get_cpu_var(host_gdt));
1383 static void vmclear_local_vcpus(void)
1385 int cpu = raw_smp_processor_id();
1386 struct vcpu_vmx *vmx, *n;
1388 list_for_each_entry_safe(vmx, n, &per_cpu(vcpus_on_cpu, cpu),
1394 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
1397 static void kvm_cpu_vmxoff(void)
1399 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
1402 static void hardware_disable(void *garbage)
1404 if (vmm_exclusive) {
1405 vmclear_local_vcpus();
1408 write_cr4(read_cr4() & ~X86_CR4_VMXE);
1411 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
1412 u32 msr, u32 *result)
1414 u32 vmx_msr_low, vmx_msr_high;
1415 u32 ctl = ctl_min | ctl_opt;
1417 rdmsr(msr, vmx_msr_low, vmx_msr_high);
1419 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
1420 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
1422 /* Ensure minimum (required) set of control bits are supported. */
1430 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
1432 u32 vmx_msr_low, vmx_msr_high;
1433 u32 min, opt, min2, opt2;
1434 u32 _pin_based_exec_control = 0;
1435 u32 _cpu_based_exec_control = 0;
1436 u32 _cpu_based_2nd_exec_control = 0;
1437 u32 _vmexit_control = 0;
1438 u32 _vmentry_control = 0;
1440 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
1441 opt = PIN_BASED_VIRTUAL_NMIS;
1442 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
1443 &_pin_based_exec_control) < 0)
1446 min = CPU_BASED_HLT_EXITING |
1447 #ifdef CONFIG_X86_64
1448 CPU_BASED_CR8_LOAD_EXITING |
1449 CPU_BASED_CR8_STORE_EXITING |
1451 CPU_BASED_CR3_LOAD_EXITING |
1452 CPU_BASED_CR3_STORE_EXITING |
1453 CPU_BASED_USE_IO_BITMAPS |
1454 CPU_BASED_MOV_DR_EXITING |
1455 CPU_BASED_USE_TSC_OFFSETING |
1456 CPU_BASED_MWAIT_EXITING |
1457 CPU_BASED_MONITOR_EXITING |
1458 CPU_BASED_INVLPG_EXITING;
1459 opt = CPU_BASED_TPR_SHADOW |
1460 CPU_BASED_USE_MSR_BITMAPS |
1461 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1462 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
1463 &_cpu_based_exec_control) < 0)
1465 #ifdef CONFIG_X86_64
1466 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
1467 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
1468 ~CPU_BASED_CR8_STORE_EXITING;
1470 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
1472 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
1473 SECONDARY_EXEC_WBINVD_EXITING |
1474 SECONDARY_EXEC_ENABLE_VPID |
1475 SECONDARY_EXEC_ENABLE_EPT |
1476 SECONDARY_EXEC_UNRESTRICTED_GUEST |
1477 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
1478 SECONDARY_EXEC_RDTSCP;
1479 if (adjust_vmx_controls(min2, opt2,
1480 MSR_IA32_VMX_PROCBASED_CTLS2,
1481 &_cpu_based_2nd_exec_control) < 0)
1484 #ifndef CONFIG_X86_64
1485 if (!(_cpu_based_2nd_exec_control &
1486 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
1487 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
1489 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
1490 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
1492 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
1493 CPU_BASED_CR3_STORE_EXITING |
1494 CPU_BASED_INVLPG_EXITING);
1495 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
1496 vmx_capability.ept, vmx_capability.vpid);
1500 #ifdef CONFIG_X86_64
1501 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
1503 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT;
1504 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
1505 &_vmexit_control) < 0)
1509 opt = VM_ENTRY_LOAD_IA32_PAT;
1510 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
1511 &_vmentry_control) < 0)
1514 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
1516 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
1517 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
1520 #ifdef CONFIG_X86_64
1521 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
1522 if (vmx_msr_high & (1u<<16))
1526 /* Require Write-Back (WB) memory type for VMCS accesses. */
1527 if (((vmx_msr_high >> 18) & 15) != 6)
1530 vmcs_conf->size = vmx_msr_high & 0x1fff;
1531 vmcs_conf->order = get_order(vmcs_config.size);
1532 vmcs_conf->revision_id = vmx_msr_low;
1534 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
1535 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
1536 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
1537 vmcs_conf->vmexit_ctrl = _vmexit_control;
1538 vmcs_conf->vmentry_ctrl = _vmentry_control;
1543 static struct vmcs *alloc_vmcs_cpu(int cpu)
1545 int node = cpu_to_node(cpu);
1549 pages = alloc_pages_exact_node(node, GFP_KERNEL, vmcs_config.order);
1552 vmcs = page_address(pages);
1553 memset(vmcs, 0, vmcs_config.size);
1554 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
1558 static struct vmcs *alloc_vmcs(void)
1560 return alloc_vmcs_cpu(raw_smp_processor_id());
1563 static void free_vmcs(struct vmcs *vmcs)
1565 free_pages((unsigned long)vmcs, vmcs_config.order);
1568 static void free_kvm_area(void)
1572 for_each_possible_cpu(cpu) {
1573 free_vmcs(per_cpu(vmxarea, cpu));
1574 per_cpu(vmxarea, cpu) = NULL;
1578 static __init int alloc_kvm_area(void)
1582 for_each_possible_cpu(cpu) {
1585 vmcs = alloc_vmcs_cpu(cpu);
1591 per_cpu(vmxarea, cpu) = vmcs;
1596 static __init int hardware_setup(void)
1598 if (setup_vmcs_config(&vmcs_config) < 0)
1601 if (boot_cpu_has(X86_FEATURE_NX))
1602 kvm_enable_efer_bits(EFER_NX);
1604 if (!cpu_has_vmx_vpid())
1607 if (!cpu_has_vmx_ept() ||
1608 !cpu_has_vmx_ept_4levels()) {
1610 enable_unrestricted_guest = 0;
1613 if (!cpu_has_vmx_unrestricted_guest())
1614 enable_unrestricted_guest = 0;
1616 if (!cpu_has_vmx_flexpriority())
1617 flexpriority_enabled = 0;
1619 if (!cpu_has_vmx_tpr_shadow())
1620 kvm_x86_ops->update_cr8_intercept = NULL;
1622 if (enable_ept && !cpu_has_vmx_ept_2m_page())
1623 kvm_disable_largepages();
1625 if (!cpu_has_vmx_ple())
1628 return alloc_kvm_area();
1631 static __exit void hardware_unsetup(void)
1636 static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
1638 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1640 if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
1641 vmcs_write16(sf->selector, save->selector);
1642 vmcs_writel(sf->base, save->base);
1643 vmcs_write32(sf->limit, save->limit);
1644 vmcs_write32(sf->ar_bytes, save->ar);
1646 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
1648 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
1652 static void enter_pmode(struct kvm_vcpu *vcpu)
1654 unsigned long flags;
1655 struct vcpu_vmx *vmx = to_vmx(vcpu);
1657 vmx->emulation_required = 1;
1658 vmx->rmode.vm86_active = 0;
1660 vmcs_writel(GUEST_TR_BASE, vmx->rmode.tr.base);
1661 vmcs_write32(GUEST_TR_LIMIT, vmx->rmode.tr.limit);
1662 vmcs_write32(GUEST_TR_AR_BYTES, vmx->rmode.tr.ar);
1664 flags = vmcs_readl(GUEST_RFLAGS);
1665 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1666 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1667 vmcs_writel(GUEST_RFLAGS, flags);
1669 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
1670 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
1672 update_exception_bitmap(vcpu);
1674 if (emulate_invalid_guest_state)
1677 fix_pmode_dataseg(VCPU_SREG_ES, &vmx->rmode.es);
1678 fix_pmode_dataseg(VCPU_SREG_DS, &vmx->rmode.ds);
1679 fix_pmode_dataseg(VCPU_SREG_GS, &vmx->rmode.gs);
1680 fix_pmode_dataseg(VCPU_SREG_FS, &vmx->rmode.fs);
1682 vmcs_write16(GUEST_SS_SELECTOR, 0);
1683 vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
1685 vmcs_write16(GUEST_CS_SELECTOR,
1686 vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
1687 vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1690 static gva_t rmode_tss_base(struct kvm *kvm)
1692 if (!kvm->arch.tss_addr) {
1693 struct kvm_memslots *slots;
1696 slots = kvm_memslots(kvm);
1697 base_gfn = slots->memslots[0].base_gfn +
1698 kvm->memslots->memslots[0].npages - 3;
1699 return base_gfn << PAGE_SHIFT;
1701 return kvm->arch.tss_addr;
1704 static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
1706 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1708 save->selector = vmcs_read16(sf->selector);
1709 save->base = vmcs_readl(sf->base);
1710 save->limit = vmcs_read32(sf->limit);
1711 save->ar = vmcs_read32(sf->ar_bytes);
1712 vmcs_write16(sf->selector, save->base >> 4);
1713 vmcs_write32(sf->base, save->base & 0xfffff);
1714 vmcs_write32(sf->limit, 0xffff);
1715 vmcs_write32(sf->ar_bytes, 0xf3);
1718 static void enter_rmode(struct kvm_vcpu *vcpu)
1720 unsigned long flags;
1721 struct vcpu_vmx *vmx = to_vmx(vcpu);
1723 if (enable_unrestricted_guest)
1726 vmx->emulation_required = 1;
1727 vmx->rmode.vm86_active = 1;
1729 vmx->rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
1730 vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
1732 vmx->rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
1733 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
1735 vmx->rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
1736 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1738 flags = vmcs_readl(GUEST_RFLAGS);
1739 vmx->rmode.save_rflags = flags;
1741 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1743 vmcs_writel(GUEST_RFLAGS, flags);
1744 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
1745 update_exception_bitmap(vcpu);
1747 if (emulate_invalid_guest_state)
1748 goto continue_rmode;
1750 vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
1751 vmcs_write32(GUEST_SS_LIMIT, 0xffff);
1752 vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
1754 vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
1755 vmcs_write32(GUEST_CS_LIMIT, 0xffff);
1756 if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
1757 vmcs_writel(GUEST_CS_BASE, 0xf0000);
1758 vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
1760 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.es);
1761 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.ds);
1762 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.gs);
1763 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.fs);
1766 kvm_mmu_reset_context(vcpu);
1767 init_rmode(vcpu->kvm);
1770 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
1772 struct vcpu_vmx *vmx = to_vmx(vcpu);
1773 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
1779 * Force kernel_gs_base reloading before EFER changes, as control
1780 * of this msr depends on is_long_mode().
1782 vmx_load_host_state(to_vmx(vcpu));
1783 vcpu->arch.efer = efer;
1784 if (efer & EFER_LMA) {
1785 vmcs_write32(VM_ENTRY_CONTROLS,
1786 vmcs_read32(VM_ENTRY_CONTROLS) |
1787 VM_ENTRY_IA32E_MODE);
1790 vmcs_write32(VM_ENTRY_CONTROLS,
1791 vmcs_read32(VM_ENTRY_CONTROLS) &
1792 ~VM_ENTRY_IA32E_MODE);
1794 msr->data = efer & ~EFER_LME;
1799 #ifdef CONFIG_X86_64
1801 static void enter_lmode(struct kvm_vcpu *vcpu)
1805 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
1806 if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
1807 printk(KERN_DEBUG "%s: tss fixup for long mode. \n",
1809 vmcs_write32(GUEST_TR_AR_BYTES,
1810 (guest_tr_ar & ~AR_TYPE_MASK)
1811 | AR_TYPE_BUSY_64_TSS);
1813 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
1816 static void exit_lmode(struct kvm_vcpu *vcpu)
1818 vmcs_write32(VM_ENTRY_CONTROLS,
1819 vmcs_read32(VM_ENTRY_CONTROLS)
1820 & ~VM_ENTRY_IA32E_MODE);
1821 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
1826 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
1828 vpid_sync_context(to_vmx(vcpu));
1830 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
1832 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
1836 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
1838 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
1840 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
1841 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
1844 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1846 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
1848 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
1849 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
1852 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
1854 if (!test_bit(VCPU_EXREG_PDPTR,
1855 (unsigned long *)&vcpu->arch.regs_dirty))
1858 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1859 vmcs_write64(GUEST_PDPTR0, vcpu->arch.pdptrs[0]);
1860 vmcs_write64(GUEST_PDPTR1, vcpu->arch.pdptrs[1]);
1861 vmcs_write64(GUEST_PDPTR2, vcpu->arch.pdptrs[2]);
1862 vmcs_write64(GUEST_PDPTR3, vcpu->arch.pdptrs[3]);
1866 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
1868 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1869 vcpu->arch.pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
1870 vcpu->arch.pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
1871 vcpu->arch.pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
1872 vcpu->arch.pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
1875 __set_bit(VCPU_EXREG_PDPTR,
1876 (unsigned long *)&vcpu->arch.regs_avail);
1877 __set_bit(VCPU_EXREG_PDPTR,
1878 (unsigned long *)&vcpu->arch.regs_dirty);
1881 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
1883 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
1885 struct kvm_vcpu *vcpu)
1887 if (!(cr0 & X86_CR0_PG)) {
1888 /* From paging/starting to nonpaging */
1889 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
1890 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
1891 (CPU_BASED_CR3_LOAD_EXITING |
1892 CPU_BASED_CR3_STORE_EXITING));
1893 vcpu->arch.cr0 = cr0;
1894 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
1895 } else if (!is_paging(vcpu)) {
1896 /* From nonpaging to paging */
1897 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
1898 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
1899 ~(CPU_BASED_CR3_LOAD_EXITING |
1900 CPU_BASED_CR3_STORE_EXITING));
1901 vcpu->arch.cr0 = cr0;
1902 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
1905 if (!(cr0 & X86_CR0_WP))
1906 *hw_cr0 &= ~X86_CR0_WP;
1909 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1911 struct vcpu_vmx *vmx = to_vmx(vcpu);
1912 unsigned long hw_cr0;
1914 if (enable_unrestricted_guest)
1915 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST)
1916 | KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
1918 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON;
1920 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
1923 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
1926 #ifdef CONFIG_X86_64
1927 if (vcpu->arch.efer & EFER_LME) {
1928 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
1930 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
1936 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
1938 if (!vcpu->fpu_active)
1939 hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
1941 vmcs_writel(CR0_READ_SHADOW, cr0);
1942 vmcs_writel(GUEST_CR0, hw_cr0);
1943 vcpu->arch.cr0 = cr0;
1946 static u64 construct_eptp(unsigned long root_hpa)
1950 /* TODO write the value reading from MSR */
1951 eptp = VMX_EPT_DEFAULT_MT |
1952 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
1953 eptp |= (root_hpa & PAGE_MASK);
1958 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1960 unsigned long guest_cr3;
1965 eptp = construct_eptp(cr3);
1966 vmcs_write64(EPT_POINTER, eptp);
1967 guest_cr3 = is_paging(vcpu) ? vcpu->arch.cr3 :
1968 vcpu->kvm->arch.ept_identity_map_addr;
1969 ept_load_pdptrs(vcpu);
1972 vmx_flush_tlb(vcpu);
1973 vmcs_writel(GUEST_CR3, guest_cr3);
1976 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1978 unsigned long hw_cr4 = cr4 | (to_vmx(vcpu)->rmode.vm86_active ?
1979 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
1981 vcpu->arch.cr4 = cr4;
1983 if (!is_paging(vcpu)) {
1984 hw_cr4 &= ~X86_CR4_PAE;
1985 hw_cr4 |= X86_CR4_PSE;
1986 } else if (!(cr4 & X86_CR4_PAE)) {
1987 hw_cr4 &= ~X86_CR4_PAE;
1991 vmcs_writel(CR4_READ_SHADOW, cr4);
1992 vmcs_writel(GUEST_CR4, hw_cr4);
1995 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1997 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1999 return vmcs_readl(sf->base);
2002 static void vmx_get_segment(struct kvm_vcpu *vcpu,
2003 struct kvm_segment *var, int seg)
2005 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2008 var->base = vmcs_readl(sf->base);
2009 var->limit = vmcs_read32(sf->limit);
2010 var->selector = vmcs_read16(sf->selector);
2011 ar = vmcs_read32(sf->ar_bytes);
2012 if ((ar & AR_UNUSABLE_MASK) && !emulate_invalid_guest_state)
2014 var->type = ar & 15;
2015 var->s = (ar >> 4) & 1;
2016 var->dpl = (ar >> 5) & 3;
2017 var->present = (ar >> 7) & 1;
2018 var->avl = (ar >> 12) & 1;
2019 var->l = (ar >> 13) & 1;
2020 var->db = (ar >> 14) & 1;
2021 var->g = (ar >> 15) & 1;
2022 var->unusable = (ar >> 16) & 1;
2025 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
2027 if (!is_protmode(vcpu))
2030 if (vmx_get_rflags(vcpu) & X86_EFLAGS_VM) /* if virtual 8086 */
2033 return vmcs_read16(GUEST_CS_SELECTOR) & 3;
2036 static u32 vmx_segment_access_rights(struct kvm_segment *var)
2043 ar = var->type & 15;
2044 ar |= (var->s & 1) << 4;
2045 ar |= (var->dpl & 3) << 5;
2046 ar |= (var->present & 1) << 7;
2047 ar |= (var->avl & 1) << 12;
2048 ar |= (var->l & 1) << 13;
2049 ar |= (var->db & 1) << 14;
2050 ar |= (var->g & 1) << 15;
2052 if (ar == 0) /* a 0 value means unusable */
2053 ar = AR_UNUSABLE_MASK;
2058 static void vmx_set_segment(struct kvm_vcpu *vcpu,
2059 struct kvm_segment *var, int seg)
2061 struct vcpu_vmx *vmx = to_vmx(vcpu);
2062 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2065 if (vmx->rmode.vm86_active && seg == VCPU_SREG_TR) {
2066 vmx->rmode.tr.selector = var->selector;
2067 vmx->rmode.tr.base = var->base;
2068 vmx->rmode.tr.limit = var->limit;
2069 vmx->rmode.tr.ar = vmx_segment_access_rights(var);
2072 vmcs_writel(sf->base, var->base);
2073 vmcs_write32(sf->limit, var->limit);
2074 vmcs_write16(sf->selector, var->selector);
2075 if (vmx->rmode.vm86_active && var->s) {
2077 * Hack real-mode segments into vm86 compatibility.
2079 if (var->base == 0xffff0000 && var->selector == 0xf000)
2080 vmcs_writel(sf->base, 0xf0000);
2083 ar = vmx_segment_access_rights(var);
2086 * Fix the "Accessed" bit in AR field of segment registers for older
2088 * IA32 arch specifies that at the time of processor reset the
2089 * "Accessed" bit in the AR field of segment registers is 1. And qemu
2090 * is setting it to 0 in the usedland code. This causes invalid guest
2091 * state vmexit when "unrestricted guest" mode is turned on.
2092 * Fix for this setup issue in cpu_reset is being pushed in the qemu
2093 * tree. Newer qemu binaries with that qemu fix would not need this
2096 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
2097 ar |= 0x1; /* Accessed */
2099 vmcs_write32(sf->ar_bytes, ar);
2102 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2104 u32 ar = vmcs_read32(GUEST_CS_AR_BYTES);
2106 *db = (ar >> 14) & 1;
2107 *l = (ar >> 13) & 1;
2110 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2112 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
2113 dt->address = vmcs_readl(GUEST_IDTR_BASE);
2116 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2118 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
2119 vmcs_writel(GUEST_IDTR_BASE, dt->address);
2122 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2124 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
2125 dt->address = vmcs_readl(GUEST_GDTR_BASE);
2128 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2130 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
2131 vmcs_writel(GUEST_GDTR_BASE, dt->address);
2134 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
2136 struct kvm_segment var;
2139 vmx_get_segment(vcpu, &var, seg);
2140 ar = vmx_segment_access_rights(&var);
2142 if (var.base != (var.selector << 4))
2144 if (var.limit != 0xffff)
2152 static bool code_segment_valid(struct kvm_vcpu *vcpu)
2154 struct kvm_segment cs;
2155 unsigned int cs_rpl;
2157 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
2158 cs_rpl = cs.selector & SELECTOR_RPL_MASK;
2162 if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
2166 if (cs.type & AR_TYPE_WRITEABLE_MASK) {
2167 if (cs.dpl > cs_rpl)
2170 if (cs.dpl != cs_rpl)
2176 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
2180 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
2182 struct kvm_segment ss;
2183 unsigned int ss_rpl;
2185 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
2186 ss_rpl = ss.selector & SELECTOR_RPL_MASK;
2190 if (ss.type != 3 && ss.type != 7)
2194 if (ss.dpl != ss_rpl) /* DPL != RPL */
2202 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
2204 struct kvm_segment var;
2207 vmx_get_segment(vcpu, &var, seg);
2208 rpl = var.selector & SELECTOR_RPL_MASK;
2216 if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
2217 if (var.dpl < rpl) /* DPL < RPL */
2221 /* TODO: Add other members to kvm_segment_field to allow checking for other access
2227 static bool tr_valid(struct kvm_vcpu *vcpu)
2229 struct kvm_segment tr;
2231 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
2235 if (tr.selector & SELECTOR_TI_MASK) /* TI = 1 */
2237 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
2245 static bool ldtr_valid(struct kvm_vcpu *vcpu)
2247 struct kvm_segment ldtr;
2249 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
2253 if (ldtr.selector & SELECTOR_TI_MASK) /* TI = 1 */
2263 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
2265 struct kvm_segment cs, ss;
2267 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
2268 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
2270 return ((cs.selector & SELECTOR_RPL_MASK) ==
2271 (ss.selector & SELECTOR_RPL_MASK));
2275 * Check if guest state is valid. Returns true if valid, false if
2277 * We assume that registers are always usable
2279 static bool guest_state_valid(struct kvm_vcpu *vcpu)
2281 /* real mode guest state checks */
2282 if (!is_protmode(vcpu)) {
2283 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
2285 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
2287 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
2289 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
2291 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
2293 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
2296 /* protected mode guest state checks */
2297 if (!cs_ss_rpl_check(vcpu))
2299 if (!code_segment_valid(vcpu))
2301 if (!stack_segment_valid(vcpu))
2303 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
2305 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
2307 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
2309 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
2311 if (!tr_valid(vcpu))
2313 if (!ldtr_valid(vcpu))
2317 * - Add checks on RIP
2318 * - Add checks on RFLAGS
2324 static int init_rmode_tss(struct kvm *kvm)
2326 gfn_t fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
2331 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
2334 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
2335 r = kvm_write_guest_page(kvm, fn++, &data,
2336 TSS_IOPB_BASE_OFFSET, sizeof(u16));
2339 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
2342 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
2346 r = kvm_write_guest_page(kvm, fn, &data,
2347 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
2357 static int init_rmode_identity_map(struct kvm *kvm)
2360 pfn_t identity_map_pfn;
2365 if (unlikely(!kvm->arch.ept_identity_pagetable)) {
2366 printk(KERN_ERR "EPT: identity-mapping pagetable "
2367 "haven't been allocated!\n");
2370 if (likely(kvm->arch.ept_identity_pagetable_done))
2373 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
2374 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
2377 /* Set up identity-mapping pagetable for EPT in real mode */
2378 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
2379 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
2380 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
2381 r = kvm_write_guest_page(kvm, identity_map_pfn,
2382 &tmp, i * sizeof(tmp), sizeof(tmp));
2386 kvm->arch.ept_identity_pagetable_done = true;
2392 static void seg_setup(int seg)
2394 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2397 vmcs_write16(sf->selector, 0);
2398 vmcs_writel(sf->base, 0);
2399 vmcs_write32(sf->limit, 0xffff);
2400 if (enable_unrestricted_guest) {
2402 if (seg == VCPU_SREG_CS)
2403 ar |= 0x08; /* code segment */
2407 vmcs_write32(sf->ar_bytes, ar);
2410 static int alloc_apic_access_page(struct kvm *kvm)
2412 struct kvm_userspace_memory_region kvm_userspace_mem;
2415 mutex_lock(&kvm->slots_lock);
2416 if (kvm->arch.apic_access_page)
2418 kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
2419 kvm_userspace_mem.flags = 0;
2420 kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
2421 kvm_userspace_mem.memory_size = PAGE_SIZE;
2422 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2426 kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
2428 mutex_unlock(&kvm->slots_lock);
2432 static int alloc_identity_pagetable(struct kvm *kvm)
2434 struct kvm_userspace_memory_region kvm_userspace_mem;
2437 mutex_lock(&kvm->slots_lock);
2438 if (kvm->arch.ept_identity_pagetable)
2440 kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
2441 kvm_userspace_mem.flags = 0;
2442 kvm_userspace_mem.guest_phys_addr =
2443 kvm->arch.ept_identity_map_addr;
2444 kvm_userspace_mem.memory_size = PAGE_SIZE;
2445 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2449 kvm->arch.ept_identity_pagetable = gfn_to_page(kvm,
2450 kvm->arch.ept_identity_map_addr >> PAGE_SHIFT);
2452 mutex_unlock(&kvm->slots_lock);
2456 static void allocate_vpid(struct vcpu_vmx *vmx)
2463 spin_lock(&vmx_vpid_lock);
2464 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
2465 if (vpid < VMX_NR_VPIDS) {
2467 __set_bit(vpid, vmx_vpid_bitmap);
2469 spin_unlock(&vmx_vpid_lock);
2472 static void free_vpid(struct vcpu_vmx *vmx)
2476 spin_lock(&vmx_vpid_lock);
2478 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
2479 spin_unlock(&vmx_vpid_lock);
2482 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap, u32 msr)
2484 int f = sizeof(unsigned long);
2486 if (!cpu_has_vmx_msr_bitmap())
2490 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
2491 * have the write-low and read-high bitmap offsets the wrong way round.
2492 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
2494 if (msr <= 0x1fff) {
2495 __clear_bit(msr, msr_bitmap + 0x000 / f); /* read-low */
2496 __clear_bit(msr, msr_bitmap + 0x800 / f); /* write-low */
2497 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
2499 __clear_bit(msr, msr_bitmap + 0x400 / f); /* read-high */
2500 __clear_bit(msr, msr_bitmap + 0xc00 / f); /* write-high */
2504 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
2507 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy, msr);
2508 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode, msr);
2512 * Sets up the vmcs for emulated real mode.
2514 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
2516 u32 host_sysenter_cs, msr_low, msr_high;
2518 u64 host_pat, tsc_this, tsc_base;
2522 unsigned long kvm_vmx_return;
2526 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
2527 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
2529 if (cpu_has_vmx_msr_bitmap())
2530 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
2532 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
2535 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
2536 vmcs_config.pin_based_exec_ctrl);
2538 exec_control = vmcs_config.cpu_based_exec_ctrl;
2539 if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
2540 exec_control &= ~CPU_BASED_TPR_SHADOW;
2541 #ifdef CONFIG_X86_64
2542 exec_control |= CPU_BASED_CR8_STORE_EXITING |
2543 CPU_BASED_CR8_LOAD_EXITING;
2547 exec_control |= CPU_BASED_CR3_STORE_EXITING |
2548 CPU_BASED_CR3_LOAD_EXITING |
2549 CPU_BASED_INVLPG_EXITING;
2550 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
2552 if (cpu_has_secondary_exec_ctrls()) {
2553 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
2554 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2556 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
2558 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
2560 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
2561 enable_unrestricted_guest = 0;
2563 if (!enable_unrestricted_guest)
2564 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
2566 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
2567 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
2571 vmcs_write32(PLE_GAP, ple_gap);
2572 vmcs_write32(PLE_WINDOW, ple_window);
2575 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, !!bypass_guest_pf);
2576 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, !!bypass_guest_pf);
2577 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
2579 vmcs_writel(HOST_CR0, read_cr0() | X86_CR0_TS); /* 22.2.3 */
2580 vmcs_writel(HOST_CR4, read_cr4()); /* 22.2.3, 22.2.5 */
2581 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
2583 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
2584 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
2585 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
2586 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
2587 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
2588 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
2589 #ifdef CONFIG_X86_64
2590 rdmsrl(MSR_FS_BASE, a);
2591 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
2592 rdmsrl(MSR_GS_BASE, a);
2593 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
2595 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
2596 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
2599 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
2601 native_store_idt(&dt);
2602 vmcs_writel(HOST_IDTR_BASE, dt.address); /* 22.2.4 */
2604 asm("mov $.Lkvm_vmx_return, %0" : "=r"(kvm_vmx_return));
2605 vmcs_writel(HOST_RIP, kvm_vmx_return); /* 22.2.5 */
2606 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
2607 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
2608 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
2609 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
2610 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
2612 rdmsr(MSR_IA32_SYSENTER_CS, host_sysenter_cs, junk);
2613 vmcs_write32(HOST_IA32_SYSENTER_CS, host_sysenter_cs);
2614 rdmsrl(MSR_IA32_SYSENTER_ESP, a);
2615 vmcs_writel(HOST_IA32_SYSENTER_ESP, a); /* 22.2.3 */
2616 rdmsrl(MSR_IA32_SYSENTER_EIP, a);
2617 vmcs_writel(HOST_IA32_SYSENTER_EIP, a); /* 22.2.3 */
2619 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
2620 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2621 host_pat = msr_low | ((u64) msr_high << 32);
2622 vmcs_write64(HOST_IA32_PAT, host_pat);
2624 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2625 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2626 host_pat = msr_low | ((u64) msr_high << 32);
2627 /* Write the default value follow host pat */
2628 vmcs_write64(GUEST_IA32_PAT, host_pat);
2629 /* Keep arch.pat sync with GUEST_IA32_PAT */
2630 vmx->vcpu.arch.pat = host_pat;
2633 for (i = 0; i < NR_VMX_MSR; ++i) {
2634 u32 index = vmx_msr_index[i];
2635 u32 data_low, data_high;
2638 if (rdmsr_safe(index, &data_low, &data_high) < 0)
2640 if (wrmsr_safe(index, data_low, data_high) < 0)
2642 vmx->guest_msrs[j].index = i;
2643 vmx->guest_msrs[j].data = 0;
2644 vmx->guest_msrs[j].mask = -1ull;
2648 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
2650 /* 22.2.1, 20.8.1 */
2651 vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
2653 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
2654 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
2656 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
2657 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
2659 tsc_base = vmx->vcpu.kvm->arch.vm_init_tsc;
2661 if (tsc_this < vmx->vcpu.kvm->arch.vm_init_tsc)
2662 tsc_base = tsc_this;
2664 guest_write_tsc(0, tsc_base);
2669 static int init_rmode(struct kvm *kvm)
2673 idx = srcu_read_lock(&kvm->srcu);
2674 if (!init_rmode_tss(kvm))
2676 if (!init_rmode_identity_map(kvm))
2681 srcu_read_unlock(&kvm->srcu, idx);
2685 static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
2687 struct vcpu_vmx *vmx = to_vmx(vcpu);
2691 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
2692 if (!init_rmode(vmx->vcpu.kvm)) {
2697 vmx->rmode.vm86_active = 0;
2699 vmx->soft_vnmi_blocked = 0;
2701 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
2702 kvm_set_cr8(&vmx->vcpu, 0);
2703 msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
2704 if (kvm_vcpu_is_bsp(&vmx->vcpu))
2705 msr |= MSR_IA32_APICBASE_BSP;
2706 kvm_set_apic_base(&vmx->vcpu, msr);
2708 ret = fx_init(&vmx->vcpu);
2712 seg_setup(VCPU_SREG_CS);
2714 * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
2715 * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4. Sigh.
2717 if (kvm_vcpu_is_bsp(&vmx->vcpu)) {
2718 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
2719 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
2721 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
2722 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
2725 seg_setup(VCPU_SREG_DS);
2726 seg_setup(VCPU_SREG_ES);
2727 seg_setup(VCPU_SREG_FS);
2728 seg_setup(VCPU_SREG_GS);
2729 seg_setup(VCPU_SREG_SS);
2731 vmcs_write16(GUEST_TR_SELECTOR, 0);
2732 vmcs_writel(GUEST_TR_BASE, 0);
2733 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
2734 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2736 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
2737 vmcs_writel(GUEST_LDTR_BASE, 0);
2738 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
2739 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
2741 vmcs_write32(GUEST_SYSENTER_CS, 0);
2742 vmcs_writel(GUEST_SYSENTER_ESP, 0);
2743 vmcs_writel(GUEST_SYSENTER_EIP, 0);
2745 vmcs_writel(GUEST_RFLAGS, 0x02);
2746 if (kvm_vcpu_is_bsp(&vmx->vcpu))
2747 kvm_rip_write(vcpu, 0xfff0);
2749 kvm_rip_write(vcpu, 0);
2750 kvm_register_write(vcpu, VCPU_REGS_RSP, 0);
2752 vmcs_writel(GUEST_DR7, 0x400);
2754 vmcs_writel(GUEST_GDTR_BASE, 0);
2755 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
2757 vmcs_writel(GUEST_IDTR_BASE, 0);
2758 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
2760 vmcs_write32(GUEST_ACTIVITY_STATE, 0);
2761 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
2762 vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
2764 /* Special registers */
2765 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
2769 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
2771 if (cpu_has_vmx_tpr_shadow()) {
2772 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
2773 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
2774 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
2775 page_to_phys(vmx->vcpu.arch.apic->regs_page));
2776 vmcs_write32(TPR_THRESHOLD, 0);
2779 if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2780 vmcs_write64(APIC_ACCESS_ADDR,
2781 page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
2784 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2786 vmx->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
2787 vmx_set_cr0(&vmx->vcpu, kvm_read_cr0(vcpu)); /* enter rmode */
2788 vmx_set_cr4(&vmx->vcpu, 0);
2789 vmx_set_efer(&vmx->vcpu, 0);
2790 vmx_fpu_activate(&vmx->vcpu);
2791 update_exception_bitmap(&vmx->vcpu);
2793 vpid_sync_context(vmx);
2797 /* HACK: Don't enable emulation on guest boot/reset */
2798 vmx->emulation_required = 0;
2804 static void enable_irq_window(struct kvm_vcpu *vcpu)
2806 u32 cpu_based_vm_exec_control;
2808 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2809 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
2810 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2813 static void enable_nmi_window(struct kvm_vcpu *vcpu)
2815 u32 cpu_based_vm_exec_control;
2817 if (!cpu_has_virtual_nmis()) {
2818 enable_irq_window(vcpu);
2822 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2823 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
2824 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2827 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
2829 struct vcpu_vmx *vmx = to_vmx(vcpu);
2831 int irq = vcpu->arch.interrupt.nr;
2833 trace_kvm_inj_virq(irq);
2835 ++vcpu->stat.irq_injections;
2836 if (vmx->rmode.vm86_active) {
2837 vmx->rmode.irq.pending = true;
2838 vmx->rmode.irq.vector = irq;
2839 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
2840 if (vcpu->arch.interrupt.soft)
2841 vmx->rmode.irq.rip +=
2842 vmx->vcpu.arch.event_exit_inst_len;
2843 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2844 irq | INTR_TYPE_SOFT_INTR | INTR_INFO_VALID_MASK);
2845 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
2846 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
2849 intr = irq | INTR_INFO_VALID_MASK;
2850 if (vcpu->arch.interrupt.soft) {
2851 intr |= INTR_TYPE_SOFT_INTR;
2852 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2853 vmx->vcpu.arch.event_exit_inst_len);
2855 intr |= INTR_TYPE_EXT_INTR;
2856 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
2859 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
2861 struct vcpu_vmx *vmx = to_vmx(vcpu);
2863 if (!cpu_has_virtual_nmis()) {
2865 * Tracking the NMI-blocked state in software is built upon
2866 * finding the next open IRQ window. This, in turn, depends on
2867 * well-behaving guests: They have to keep IRQs disabled at
2868 * least as long as the NMI handler runs. Otherwise we may
2869 * cause NMI nesting, maybe breaking the guest. But as this is
2870 * highly unlikely, we can live with the residual risk.
2872 vmx->soft_vnmi_blocked = 1;
2873 vmx->vnmi_blocked_time = 0;
2876 ++vcpu->stat.nmi_injections;
2877 if (vmx->rmode.vm86_active) {
2878 vmx->rmode.irq.pending = true;
2879 vmx->rmode.irq.vector = NMI_VECTOR;
2880 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
2881 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2882 NMI_VECTOR | INTR_TYPE_SOFT_INTR |
2883 INTR_INFO_VALID_MASK);
2884 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
2885 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
2888 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2889 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
2892 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
2894 if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
2897 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
2898 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_NMI));
2901 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
2903 if (!cpu_has_virtual_nmis())
2904 return to_vmx(vcpu)->soft_vnmi_blocked;
2905 return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
2908 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
2910 struct vcpu_vmx *vmx = to_vmx(vcpu);
2912 if (!cpu_has_virtual_nmis()) {
2913 if (vmx->soft_vnmi_blocked != masked) {
2914 vmx->soft_vnmi_blocked = masked;
2915 vmx->vnmi_blocked_time = 0;
2919 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
2920 GUEST_INTR_STATE_NMI);
2922 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
2923 GUEST_INTR_STATE_NMI);
2927 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
2929 return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
2930 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
2931 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
2934 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
2937 struct kvm_userspace_memory_region tss_mem = {
2938 .slot = TSS_PRIVATE_MEMSLOT,
2939 .guest_phys_addr = addr,
2940 .memory_size = PAGE_SIZE * 3,
2944 ret = kvm_set_memory_region(kvm, &tss_mem, 0);
2947 kvm->arch.tss_addr = addr;
2951 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
2952 int vec, u32 err_code)
2955 * Instruction with address size override prefix opcode 0x67
2956 * Cause the #SS fault with 0 error code in VM86 mode.
2958 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
2959 if (emulate_instruction(vcpu, 0, 0, 0) == EMULATE_DONE)
2962 * Forward all other exceptions that are valid in real mode.
2963 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
2964 * the required debugging infrastructure rework.
2968 if (vcpu->guest_debug &
2969 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
2971 kvm_queue_exception(vcpu, vec);
2975 * Update instruction length as we may reinject the exception
2976 * from user space while in guest debugging mode.
2978 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
2979 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
2980 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
2991 kvm_queue_exception(vcpu, vec);
2998 * Trigger machine check on the host. We assume all the MSRs are already set up
2999 * by the CPU and that we still run on the same CPU as the MCE occurred on.
3000 * We pass a fake environment to the machine check handler because we want
3001 * the guest to be always treated like user space, no matter what context
3002 * it used internally.
3004 static void kvm_machine_check(void)
3006 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
3007 struct pt_regs regs = {
3008 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
3009 .flags = X86_EFLAGS_IF,
3012 do_machine_check(®s, 0);
3016 static int handle_machine_check(struct kvm_vcpu *vcpu)
3018 /* already handled by vcpu_run */
3022 static int handle_exception(struct kvm_vcpu *vcpu)
3024 struct vcpu_vmx *vmx = to_vmx(vcpu);
3025 struct kvm_run *kvm_run = vcpu->run;
3026 u32 intr_info, ex_no, error_code;
3027 unsigned long cr2, rip, dr6;
3029 enum emulation_result er;
3031 vect_info = vmx->idt_vectoring_info;
3032 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
3034 if (is_machine_check(intr_info))
3035 return handle_machine_check(vcpu);
3037 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
3038 !is_page_fault(intr_info)) {
3039 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3040 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
3041 vcpu->run->internal.ndata = 2;
3042 vcpu->run->internal.data[0] = vect_info;
3043 vcpu->run->internal.data[1] = intr_info;
3047 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
3048 return 1; /* already handled by vmx_vcpu_run() */
3050 if (is_no_device(intr_info)) {
3051 vmx_fpu_activate(vcpu);
3055 if (is_invalid_opcode(intr_info)) {
3056 er = emulate_instruction(vcpu, 0, 0, EMULTYPE_TRAP_UD);
3057 if (er != EMULATE_DONE)
3058 kvm_queue_exception(vcpu, UD_VECTOR);
3063 rip = kvm_rip_read(vcpu);
3064 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
3065 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
3066 if (is_page_fault(intr_info)) {
3067 /* EPT won't cause page fault directly */
3070 cr2 = vmcs_readl(EXIT_QUALIFICATION);
3071 trace_kvm_page_fault(cr2, error_code);
3073 if (kvm_event_needs_reinjection(vcpu))
3074 kvm_mmu_unprotect_page_virt(vcpu, cr2);
3075 return kvm_mmu_page_fault(vcpu, cr2, error_code);
3078 if (vmx->rmode.vm86_active &&
3079 handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
3081 if (vcpu->arch.halt_request) {
3082 vcpu->arch.halt_request = 0;
3083 return kvm_emulate_halt(vcpu);
3088 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
3091 dr6 = vmcs_readl(EXIT_QUALIFICATION);
3092 if (!(vcpu->guest_debug &
3093 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
3094 vcpu->arch.dr6 = dr6 | DR6_FIXED_1;
3095 kvm_queue_exception(vcpu, DB_VECTOR);
3098 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
3099 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
3103 * Update instruction length as we may reinject #BP from
3104 * user space while in guest debugging mode. Reading it for
3105 * #DB as well causes no harm, it is not used in that case.
3107 vmx->vcpu.arch.event_exit_inst_len =
3108 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3109 kvm_run->exit_reason = KVM_EXIT_DEBUG;
3110 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
3111 kvm_run->debug.arch.exception = ex_no;
3114 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
3115 kvm_run->ex.exception = ex_no;
3116 kvm_run->ex.error_code = error_code;
3122 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
3124 ++vcpu->stat.irq_exits;
3128 static int handle_triple_fault(struct kvm_vcpu *vcpu)
3130 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
3134 static int handle_io(struct kvm_vcpu *vcpu)
3136 unsigned long exit_qualification;
3137 int size, in, string;
3140 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3141 string = (exit_qualification & 16) != 0;
3142 in = (exit_qualification & 8) != 0;
3144 ++vcpu->stat.io_exits;
3147 return emulate_instruction(vcpu, 0, 0, 0) == EMULATE_DONE;
3149 port = exit_qualification >> 16;
3150 size = (exit_qualification & 7) + 1;
3151 skip_emulated_instruction(vcpu);
3153 return kvm_fast_pio_out(vcpu, size, port);
3157 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
3160 * Patch in the VMCALL instruction:
3162 hypercall[0] = 0x0f;
3163 hypercall[1] = 0x01;
3164 hypercall[2] = 0xc1;
3167 static void complete_insn_gp(struct kvm_vcpu *vcpu, int err)
3170 kvm_inject_gp(vcpu, 0);
3172 skip_emulated_instruction(vcpu);
3175 static int handle_cr(struct kvm_vcpu *vcpu)
3177 unsigned long exit_qualification, val;
3182 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3183 cr = exit_qualification & 15;
3184 reg = (exit_qualification >> 8) & 15;
3185 switch ((exit_qualification >> 4) & 3) {
3186 case 0: /* mov to cr */
3187 val = kvm_register_read(vcpu, reg);
3188 trace_kvm_cr_write(cr, val);
3191 err = kvm_set_cr0(vcpu, val);
3192 complete_insn_gp(vcpu, err);
3195 err = kvm_set_cr3(vcpu, val);
3196 complete_insn_gp(vcpu, err);
3199 err = kvm_set_cr4(vcpu, val);
3200 complete_insn_gp(vcpu, err);
3203 u8 cr8_prev = kvm_get_cr8(vcpu);
3204 u8 cr8 = kvm_register_read(vcpu, reg);
3205 kvm_set_cr8(vcpu, cr8);
3206 skip_emulated_instruction(vcpu);
3207 if (irqchip_in_kernel(vcpu->kvm))
3209 if (cr8_prev <= cr8)
3211 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
3217 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
3218 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
3219 skip_emulated_instruction(vcpu);
3220 vmx_fpu_activate(vcpu);
3222 case 1: /*mov from cr*/
3225 kvm_register_write(vcpu, reg, vcpu->arch.cr3);
3226 trace_kvm_cr_read(cr, vcpu->arch.cr3);
3227 skip_emulated_instruction(vcpu);
3230 val = kvm_get_cr8(vcpu);
3231 kvm_register_write(vcpu, reg, val);
3232 trace_kvm_cr_read(cr, val);
3233 skip_emulated_instruction(vcpu);
3238 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
3239 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
3240 kvm_lmsw(vcpu, val);
3242 skip_emulated_instruction(vcpu);
3247 vcpu->run->exit_reason = 0;
3248 pr_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
3249 (int)(exit_qualification >> 4) & 3, cr);
3253 static int handle_dr(struct kvm_vcpu *vcpu)
3255 unsigned long exit_qualification;
3258 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
3259 if (!kvm_require_cpl(vcpu, 0))
3261 dr = vmcs_readl(GUEST_DR7);
3264 * As the vm-exit takes precedence over the debug trap, we
3265 * need to emulate the latter, either for the host or the
3266 * guest debugging itself.
3268 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
3269 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
3270 vcpu->run->debug.arch.dr7 = dr;
3271 vcpu->run->debug.arch.pc =
3272 vmcs_readl(GUEST_CS_BASE) +
3273 vmcs_readl(GUEST_RIP);
3274 vcpu->run->debug.arch.exception = DB_VECTOR;
3275 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
3278 vcpu->arch.dr7 &= ~DR7_GD;
3279 vcpu->arch.dr6 |= DR6_BD;
3280 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
3281 kvm_queue_exception(vcpu, DB_VECTOR);
3286 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3287 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
3288 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
3289 if (exit_qualification & TYPE_MOV_FROM_DR) {
3291 if (!kvm_get_dr(vcpu, dr, &val))
3292 kvm_register_write(vcpu, reg, val);
3294 kvm_set_dr(vcpu, dr, vcpu->arch.regs[reg]);
3295 skip_emulated_instruction(vcpu);
3299 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
3301 vmcs_writel(GUEST_DR7, val);
3304 static int handle_cpuid(struct kvm_vcpu *vcpu)
3306 kvm_emulate_cpuid(vcpu);
3310 static int handle_rdmsr(struct kvm_vcpu *vcpu)
3312 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3315 if (vmx_get_msr(vcpu, ecx, &data)) {
3316 trace_kvm_msr_read_ex(ecx);
3317 kvm_inject_gp(vcpu, 0);
3321 trace_kvm_msr_read(ecx, data);
3323 /* FIXME: handling of bits 32:63 of rax, rdx */
3324 vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
3325 vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
3326 skip_emulated_instruction(vcpu);
3330 static int handle_wrmsr(struct kvm_vcpu *vcpu)
3332 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3333 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
3334 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
3336 if (vmx_set_msr(vcpu, ecx, data) != 0) {
3337 trace_kvm_msr_write_ex(ecx, data);
3338 kvm_inject_gp(vcpu, 0);
3342 trace_kvm_msr_write(ecx, data);
3343 skip_emulated_instruction(vcpu);
3347 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
3352 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
3354 u32 cpu_based_vm_exec_control;
3356 /* clear pending irq */
3357 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3358 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
3359 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3361 ++vcpu->stat.irq_window_exits;
3364 * If the user space waits to inject interrupts, exit as soon as
3367 if (!irqchip_in_kernel(vcpu->kvm) &&
3368 vcpu->run->request_interrupt_window &&
3369 !kvm_cpu_has_interrupt(vcpu)) {
3370 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
3376 static int handle_halt(struct kvm_vcpu *vcpu)
3378 skip_emulated_instruction(vcpu);
3379 return kvm_emulate_halt(vcpu);
3382 static int handle_vmcall(struct kvm_vcpu *vcpu)
3384 skip_emulated_instruction(vcpu);
3385 kvm_emulate_hypercall(vcpu);
3389 static int handle_vmx_insn(struct kvm_vcpu *vcpu)
3391 kvm_queue_exception(vcpu, UD_VECTOR);
3395 static int handle_invlpg(struct kvm_vcpu *vcpu)
3397 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3399 kvm_mmu_invlpg(vcpu, exit_qualification);
3400 skip_emulated_instruction(vcpu);
3404 static int handle_wbinvd(struct kvm_vcpu *vcpu)
3406 skip_emulated_instruction(vcpu);
3407 kvm_emulate_wbinvd(vcpu);
3411 static int handle_xsetbv(struct kvm_vcpu *vcpu)
3413 u64 new_bv = kvm_read_edx_eax(vcpu);
3414 u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3416 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
3417 skip_emulated_instruction(vcpu);
3421 static int handle_apic_access(struct kvm_vcpu *vcpu)
3423 return emulate_instruction(vcpu, 0, 0, 0) == EMULATE_DONE;
3426 static int handle_task_switch(struct kvm_vcpu *vcpu)
3428 struct vcpu_vmx *vmx = to_vmx(vcpu);
3429 unsigned long exit_qualification;
3430 bool has_error_code = false;
3433 int reason, type, idt_v;
3435 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
3436 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
3438 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3440 reason = (u32)exit_qualification >> 30;
3441 if (reason == TASK_SWITCH_GATE && idt_v) {
3443 case INTR_TYPE_NMI_INTR:
3444 vcpu->arch.nmi_injected = false;
3445 if (cpu_has_virtual_nmis())
3446 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3447 GUEST_INTR_STATE_NMI);
3449 case INTR_TYPE_EXT_INTR:
3450 case INTR_TYPE_SOFT_INTR:
3451 kvm_clear_interrupt_queue(vcpu);
3453 case INTR_TYPE_HARD_EXCEPTION:
3454 if (vmx->idt_vectoring_info &
3455 VECTORING_INFO_DELIVER_CODE_MASK) {
3456 has_error_code = true;
3458 vmcs_read32(IDT_VECTORING_ERROR_CODE);
3461 case INTR_TYPE_SOFT_EXCEPTION:
3462 kvm_clear_exception_queue(vcpu);
3468 tss_selector = exit_qualification;
3470 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
3471 type != INTR_TYPE_EXT_INTR &&
3472 type != INTR_TYPE_NMI_INTR))
3473 skip_emulated_instruction(vcpu);
3475 if (kvm_task_switch(vcpu, tss_selector, reason,
3476 has_error_code, error_code) == EMULATE_FAIL) {
3477 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3478 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
3479 vcpu->run->internal.ndata = 0;
3483 /* clear all local breakpoint enable flags */
3484 vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~55);
3487 * TODO: What about debug traps on tss switch?
3488 * Are we supposed to inject them and update dr6?
3494 static int handle_ept_violation(struct kvm_vcpu *vcpu)
3496 unsigned long exit_qualification;
3500 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3502 if (exit_qualification & (1 << 6)) {
3503 printk(KERN_ERR "EPT: GPA exceeds GAW!\n");
3507 gla_validity = (exit_qualification >> 7) & 0x3;
3508 if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
3509 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
3510 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
3511 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
3512 vmcs_readl(GUEST_LINEAR_ADDRESS));
3513 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
3514 (long unsigned int)exit_qualification);
3515 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3516 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
3520 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
3521 trace_kvm_page_fault(gpa, exit_qualification);
3522 return kvm_mmu_page_fault(vcpu, gpa & PAGE_MASK, 0);
3525 static u64 ept_rsvd_mask(u64 spte, int level)
3530 for (i = 51; i > boot_cpu_data.x86_phys_bits; i--)
3531 mask |= (1ULL << i);
3534 /* bits 7:3 reserved */
3536 else if (level == 2) {
3537 if (spte & (1ULL << 7))
3538 /* 2MB ref, bits 20:12 reserved */
3541 /* bits 6:3 reserved */
3548 static void ept_misconfig_inspect_spte(struct kvm_vcpu *vcpu, u64 spte,
3551 printk(KERN_ERR "%s: spte 0x%llx level %d\n", __func__, spte, level);
3553 /* 010b (write-only) */
3554 WARN_ON((spte & 0x7) == 0x2);
3556 /* 110b (write/execute) */
3557 WARN_ON((spte & 0x7) == 0x6);
3559 /* 100b (execute-only) and value not supported by logical processor */
3560 if (!cpu_has_vmx_ept_execute_only())
3561 WARN_ON((spte & 0x7) == 0x4);
3565 u64 rsvd_bits = spte & ept_rsvd_mask(spte, level);
3567 if (rsvd_bits != 0) {
3568 printk(KERN_ERR "%s: rsvd_bits = 0x%llx\n",
3569 __func__, rsvd_bits);
3573 if (level == 1 || (level == 2 && (spte & (1ULL << 7)))) {
3574 u64 ept_mem_type = (spte & 0x38) >> 3;
3576 if (ept_mem_type == 2 || ept_mem_type == 3 ||
3577 ept_mem_type == 7) {
3578 printk(KERN_ERR "%s: ept_mem_type=0x%llx\n",
3579 __func__, ept_mem_type);
3586 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
3592 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
3594 printk(KERN_ERR "EPT: Misconfiguration.\n");
3595 printk(KERN_ERR "EPT: GPA: 0x%llx\n", gpa);
3597 nr_sptes = kvm_mmu_get_spte_hierarchy(vcpu, gpa, sptes);
3599 for (i = PT64_ROOT_LEVEL; i > PT64_ROOT_LEVEL - nr_sptes; --i)
3600 ept_misconfig_inspect_spte(vcpu, sptes[i-1], i);
3602 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3603 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
3608 static int handle_nmi_window(struct kvm_vcpu *vcpu)
3610 u32 cpu_based_vm_exec_control;
3612 /* clear pending NMI */
3613 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3614 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
3615 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3616 ++vcpu->stat.nmi_window_exits;
3621 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
3623 struct vcpu_vmx *vmx = to_vmx(vcpu);
3624 enum emulation_result err = EMULATE_DONE;
3627 while (!guest_state_valid(vcpu)) {
3628 err = emulate_instruction(vcpu, 0, 0, 0);
3630 if (err == EMULATE_DO_MMIO) {
3635 if (err != EMULATE_DONE)
3638 if (signal_pending(current))
3644 vmx->emulation_required = 0;
3650 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
3651 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
3653 static int handle_pause(struct kvm_vcpu *vcpu)
3655 skip_emulated_instruction(vcpu);
3656 kvm_vcpu_on_spin(vcpu);
3661 static int handle_invalid_op(struct kvm_vcpu *vcpu)
3663 kvm_queue_exception(vcpu, UD_VECTOR);
3668 * The exit handlers return 1 if the exit was handled fully and guest execution
3669 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
3670 * to be done to userspace and return 0.
3672 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
3673 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
3674 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
3675 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
3676 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
3677 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
3678 [EXIT_REASON_CR_ACCESS] = handle_cr,
3679 [EXIT_REASON_DR_ACCESS] = handle_dr,
3680 [EXIT_REASON_CPUID] = handle_cpuid,
3681 [EXIT_REASON_MSR_READ] = handle_rdmsr,
3682 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
3683 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
3684 [EXIT_REASON_HLT] = handle_halt,
3685 [EXIT_REASON_INVLPG] = handle_invlpg,
3686 [EXIT_REASON_VMCALL] = handle_vmcall,
3687 [EXIT_REASON_VMCLEAR] = handle_vmx_insn,
3688 [EXIT_REASON_VMLAUNCH] = handle_vmx_insn,
3689 [EXIT_REASON_VMPTRLD] = handle_vmx_insn,
3690 [EXIT_REASON_VMPTRST] = handle_vmx_insn,
3691 [EXIT_REASON_VMREAD] = handle_vmx_insn,
3692 [EXIT_REASON_VMRESUME] = handle_vmx_insn,
3693 [EXIT_REASON_VMWRITE] = handle_vmx_insn,
3694 [EXIT_REASON_VMOFF] = handle_vmx_insn,
3695 [EXIT_REASON_VMON] = handle_vmx_insn,
3696 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
3697 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
3698 [EXIT_REASON_WBINVD] = handle_wbinvd,
3699 [EXIT_REASON_XSETBV] = handle_xsetbv,
3700 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
3701 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
3702 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
3703 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
3704 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
3705 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_invalid_op,
3706 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_invalid_op,
3709 static const int kvm_vmx_max_exit_handlers =
3710 ARRAY_SIZE(kvm_vmx_exit_handlers);
3713 * The guest has exited. See if we can fix it or if we need userspace
3716 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
3718 struct vcpu_vmx *vmx = to_vmx(vcpu);
3719 u32 exit_reason = vmx->exit_reason;
3720 u32 vectoring_info = vmx->idt_vectoring_info;
3722 trace_kvm_exit(exit_reason, vcpu);
3724 /* If guest state is invalid, start emulating */
3725 if (vmx->emulation_required && emulate_invalid_guest_state)
3726 return handle_invalid_guest_state(vcpu);
3728 /* Access CR3 don't cause VMExit in paging mode, so we need
3729 * to sync with guest real CR3. */
3730 if (enable_ept && is_paging(vcpu))
3731 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3733 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
3734 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3735 vcpu->run->fail_entry.hardware_entry_failure_reason
3740 if (unlikely(vmx->fail)) {
3741 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3742 vcpu->run->fail_entry.hardware_entry_failure_reason
3743 = vmcs_read32(VM_INSTRUCTION_ERROR);
3747 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
3748 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
3749 exit_reason != EXIT_REASON_EPT_VIOLATION &&
3750 exit_reason != EXIT_REASON_TASK_SWITCH))
3751 printk(KERN_WARNING "%s: unexpected, valid vectoring info "
3752 "(0x%x) and exit reason is 0x%x\n",
3753 __func__, vectoring_info, exit_reason);
3755 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked)) {
3756 if (vmx_interrupt_allowed(vcpu)) {
3757 vmx->soft_vnmi_blocked = 0;
3758 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
3759 vcpu->arch.nmi_pending) {
3761 * This CPU don't support us in finding the end of an
3762 * NMI-blocked window if the guest runs with IRQs
3763 * disabled. So we pull the trigger after 1 s of
3764 * futile waiting, but inform the user about this.
3766 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
3767 "state on VCPU %d after 1 s timeout\n",
3768 __func__, vcpu->vcpu_id);
3769 vmx->soft_vnmi_blocked = 0;
3773 if (exit_reason < kvm_vmx_max_exit_handlers
3774 && kvm_vmx_exit_handlers[exit_reason])
3775 return kvm_vmx_exit_handlers[exit_reason](vcpu);
3777 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3778 vcpu->run->hw.hardware_exit_reason = exit_reason;
3783 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
3785 if (irr == -1 || tpr < irr) {
3786 vmcs_write32(TPR_THRESHOLD, 0);
3790 vmcs_write32(TPR_THRESHOLD, irr);
3793 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
3796 u32 idt_vectoring_info = vmx->idt_vectoring_info;
3800 bool idtv_info_valid;
3802 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
3804 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
3806 /* Handle machine checks before interrupts are enabled */
3807 if ((vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY)
3808 || (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI
3809 && is_machine_check(exit_intr_info)))
3810 kvm_machine_check();
3812 /* We need to handle NMIs before interrupts are enabled */
3813 if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
3814 (exit_intr_info & INTR_INFO_VALID_MASK)) {
3815 kvm_before_handle_nmi(&vmx->vcpu);
3817 kvm_after_handle_nmi(&vmx->vcpu);
3820 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
3822 if (cpu_has_virtual_nmis()) {
3823 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
3824 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
3826 * SDM 3: 27.7.1.2 (September 2008)
3827 * Re-set bit "block by NMI" before VM entry if vmexit caused by
3828 * a guest IRET fault.
3829 * SDM 3: 23.2.2 (September 2008)
3830 * Bit 12 is undefined in any of the following cases:
3831 * If the VM exit sets the valid bit in the IDT-vectoring
3832 * information field.
3833 * If the VM exit is due to a double fault.
3835 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
3836 vector != DF_VECTOR && !idtv_info_valid)
3837 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3838 GUEST_INTR_STATE_NMI);
3839 } else if (unlikely(vmx->soft_vnmi_blocked))
3840 vmx->vnmi_blocked_time +=
3841 ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
3843 vmx->vcpu.arch.nmi_injected = false;
3844 kvm_clear_exception_queue(&vmx->vcpu);
3845 kvm_clear_interrupt_queue(&vmx->vcpu);
3847 if (!idtv_info_valid)
3850 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
3851 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
3854 case INTR_TYPE_NMI_INTR:
3855 vmx->vcpu.arch.nmi_injected = true;
3857 * SDM 3: 27.7.1.2 (September 2008)
3858 * Clear bit "block by NMI" before VM entry if a NMI
3861 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
3862 GUEST_INTR_STATE_NMI);
3864 case INTR_TYPE_SOFT_EXCEPTION:
3865 vmx->vcpu.arch.event_exit_inst_len =
3866 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3868 case INTR_TYPE_HARD_EXCEPTION:
3869 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
3870 u32 err = vmcs_read32(IDT_VECTORING_ERROR_CODE);
3871 kvm_queue_exception_e(&vmx->vcpu, vector, err);
3873 kvm_queue_exception(&vmx->vcpu, vector);
3875 case INTR_TYPE_SOFT_INTR:
3876 vmx->vcpu.arch.event_exit_inst_len =
3877 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3879 case INTR_TYPE_EXT_INTR:
3880 kvm_queue_interrupt(&vmx->vcpu, vector,
3881 type == INTR_TYPE_SOFT_INTR);
3889 * Failure to inject an interrupt should give us the information
3890 * in IDT_VECTORING_INFO_FIELD. However, if the failure occurs
3891 * when fetching the interrupt redirection bitmap in the real-mode
3892 * tss, this doesn't happen. So we do it ourselves.
3894 static void fixup_rmode_irq(struct vcpu_vmx *vmx)
3896 vmx->rmode.irq.pending = 0;
3897 if (kvm_rip_read(&vmx->vcpu) + 1 != vmx->rmode.irq.rip)
3899 kvm_rip_write(&vmx->vcpu, vmx->rmode.irq.rip);
3900 if (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) {
3901 vmx->idt_vectoring_info &= ~VECTORING_INFO_TYPE_MASK;
3902 vmx->idt_vectoring_info |= INTR_TYPE_EXT_INTR;
3905 vmx->idt_vectoring_info =
3906 VECTORING_INFO_VALID_MASK
3907 | INTR_TYPE_EXT_INTR
3908 | vmx->rmode.irq.vector;
3911 #ifdef CONFIG_X86_64
3919 static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
3921 struct vcpu_vmx *vmx = to_vmx(vcpu);
3923 /* Record the guest's net vcpu time for enforced NMI injections. */
3924 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
3925 vmx->entry_time = ktime_get();
3927 /* Don't enter VMX if guest state is invalid, let the exit handler
3928 start emulation until we arrive back to a valid state */
3929 if (vmx->emulation_required && emulate_invalid_guest_state)
3932 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
3933 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
3934 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
3935 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
3937 /* When single-stepping over STI and MOV SS, we must clear the
3938 * corresponding interruptibility bits in the guest state. Otherwise
3939 * vmentry fails as it then expects bit 14 (BS) in pending debug
3940 * exceptions being set, but that's not correct for the guest debugging
3942 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3943 vmx_set_interrupt_shadow(vcpu, 0);
3946 /* Store host registers */
3947 "push %%"R"dx; push %%"R"bp;"
3949 "cmp %%"R"sp, %c[host_rsp](%0) \n\t"
3951 "mov %%"R"sp, %c[host_rsp](%0) \n\t"
3952 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
3954 /* Reload cr2 if changed */
3955 "mov %c[cr2](%0), %%"R"ax \n\t"
3956 "mov %%cr2, %%"R"dx \n\t"
3957 "cmp %%"R"ax, %%"R"dx \n\t"
3959 "mov %%"R"ax, %%cr2 \n\t"
3961 /* Check if vmlaunch of vmresume is needed */
3962 "cmpl $0, %c[launched](%0) \n\t"
3963 /* Load guest registers. Don't clobber flags. */
3964 "mov %c[rax](%0), %%"R"ax \n\t"
3965 "mov %c[rbx](%0), %%"R"bx \n\t"
3966 "mov %c[rdx](%0), %%"R"dx \n\t"
3967 "mov %c[rsi](%0), %%"R"si \n\t"
3968 "mov %c[rdi](%0), %%"R"di \n\t"
3969 "mov %c[rbp](%0), %%"R"bp \n\t"
3970 #ifdef CONFIG_X86_64
3971 "mov %c[r8](%0), %%r8 \n\t"
3972 "mov %c[r9](%0), %%r9 \n\t"
3973 "mov %c[r10](%0), %%r10 \n\t"
3974 "mov %c[r11](%0), %%r11 \n\t"
3975 "mov %c[r12](%0), %%r12 \n\t"
3976 "mov %c[r13](%0), %%r13 \n\t"
3977 "mov %c[r14](%0), %%r14 \n\t"
3978 "mov %c[r15](%0), %%r15 \n\t"
3980 "mov %c[rcx](%0), %%"R"cx \n\t" /* kills %0 (ecx) */
3982 /* Enter guest mode */
3983 "jne .Llaunched \n\t"
3984 __ex(ASM_VMX_VMLAUNCH) "\n\t"
3985 "jmp .Lkvm_vmx_return \n\t"
3986 ".Llaunched: " __ex(ASM_VMX_VMRESUME) "\n\t"
3987 ".Lkvm_vmx_return: "
3988 /* Save guest registers, load host registers, keep flags */
3989 "xchg %0, (%%"R"sp) \n\t"
3990 "mov %%"R"ax, %c[rax](%0) \n\t"
3991 "mov %%"R"bx, %c[rbx](%0) \n\t"
3992 "push"Q" (%%"R"sp); pop"Q" %c[rcx](%0) \n\t"
3993 "mov %%"R"dx, %c[rdx](%0) \n\t"
3994 "mov %%"R"si, %c[rsi](%0) \n\t"
3995 "mov %%"R"di, %c[rdi](%0) \n\t"
3996 "mov %%"R"bp, %c[rbp](%0) \n\t"
3997 #ifdef CONFIG_X86_64
3998 "mov %%r8, %c[r8](%0) \n\t"
3999 "mov %%r9, %c[r9](%0) \n\t"
4000 "mov %%r10, %c[r10](%0) \n\t"
4001 "mov %%r11, %c[r11](%0) \n\t"
4002 "mov %%r12, %c[r12](%0) \n\t"
4003 "mov %%r13, %c[r13](%0) \n\t"
4004 "mov %%r14, %c[r14](%0) \n\t"
4005 "mov %%r15, %c[r15](%0) \n\t"
4007 "mov %%cr2, %%"R"ax \n\t"
4008 "mov %%"R"ax, %c[cr2](%0) \n\t"
4010 "pop %%"R"bp; pop %%"R"bp; pop %%"R"dx \n\t"
4011 "setbe %c[fail](%0) \n\t"
4012 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
4013 [launched]"i"(offsetof(struct vcpu_vmx, launched)),
4014 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
4015 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
4016 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
4017 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
4018 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
4019 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
4020 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
4021 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
4022 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
4023 #ifdef CONFIG_X86_64
4024 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
4025 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
4026 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
4027 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
4028 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
4029 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
4030 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
4031 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
4033 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2))
4035 , R"bx", R"di", R"si"
4036 #ifdef CONFIG_X86_64
4037 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
4041 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
4042 | (1 << VCPU_EXREG_PDPTR));
4043 vcpu->arch.regs_dirty = 0;
4045 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
4046 if (vmx->rmode.irq.pending)
4047 fixup_rmode_irq(vmx);
4049 asm("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
4052 vmx_complete_interrupts(vmx);
4058 static void vmx_free_vmcs(struct kvm_vcpu *vcpu)
4060 struct vcpu_vmx *vmx = to_vmx(vcpu);
4064 free_vmcs(vmx->vmcs);
4069 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
4071 struct vcpu_vmx *vmx = to_vmx(vcpu);
4074 vmx_free_vmcs(vcpu);
4075 kfree(vmx->guest_msrs);
4076 kvm_vcpu_uninit(vcpu);
4077 kmem_cache_free(kvm_vcpu_cache, vmx);
4080 static inline void vmcs_init(struct vmcs *vmcs)
4082 u64 phys_addr = __pa(per_cpu(vmxarea, raw_smp_processor_id()));
4085 kvm_cpu_vmxon(phys_addr);
4093 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
4096 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
4100 return ERR_PTR(-ENOMEM);
4104 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
4108 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
4109 if (!vmx->guest_msrs) {
4114 vmx->vmcs = alloc_vmcs();
4118 vmcs_init(vmx->vmcs);
4121 vmx_vcpu_load(&vmx->vcpu, cpu);
4122 err = vmx_vcpu_setup(vmx);
4123 vmx_vcpu_put(&vmx->vcpu);
4127 if (vm_need_virtualize_apic_accesses(kvm))
4128 if (alloc_apic_access_page(kvm) != 0)
4132 if (!kvm->arch.ept_identity_map_addr)
4133 kvm->arch.ept_identity_map_addr =
4134 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
4135 if (alloc_identity_pagetable(kvm) != 0)
4142 free_vmcs(vmx->vmcs);
4144 kfree(vmx->guest_msrs);
4146 kvm_vcpu_uninit(&vmx->vcpu);
4149 kmem_cache_free(kvm_vcpu_cache, vmx);
4150 return ERR_PTR(err);
4153 static void __init vmx_check_processor_compat(void *rtn)
4155 struct vmcs_config vmcs_conf;
4158 if (setup_vmcs_config(&vmcs_conf) < 0)
4160 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
4161 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
4162 smp_processor_id());
4167 static int get_ept_level(void)
4169 return VMX_EPT_DEFAULT_GAW + 1;
4172 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
4176 /* For VT-d and EPT combination
4177 * 1. MMIO: always map as UC
4179 * a. VT-d without snooping control feature: can't guarantee the
4180 * result, try to trust guest.
4181 * b. VT-d with snooping control feature: snooping control feature of
4182 * VT-d engine can guarantee the cache correctness. Just set it
4183 * to WB to keep consistent with host. So the same as item 3.
4184 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
4185 * consistent with host MTRR
4188 ret = MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
4189 else if (vcpu->kvm->arch.iommu_domain &&
4190 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY))
4191 ret = kvm_get_guest_memory_type(vcpu, gfn) <<
4192 VMX_EPT_MT_EPTE_SHIFT;
4194 ret = (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT)
4200 #define _ER(x) { EXIT_REASON_##x, #x }
4202 static const struct trace_print_flags vmx_exit_reasons_str[] = {
4204 _ER(EXTERNAL_INTERRUPT),
4206 _ER(PENDING_INTERRUPT),
4226 _ER(IO_INSTRUCTION),
4229 _ER(MWAIT_INSTRUCTION),
4230 _ER(MONITOR_INSTRUCTION),
4231 _ER(PAUSE_INSTRUCTION),
4232 _ER(MCE_DURING_VMENTRY),
4233 _ER(TPR_BELOW_THRESHOLD),
4243 static int vmx_get_lpage_level(void)
4245 if (enable_ept && !cpu_has_vmx_ept_1g_page())
4246 return PT_DIRECTORY_LEVEL;
4248 /* For shadow and EPT supported 1GB page */
4249 return PT_PDPE_LEVEL;
4252 static inline u32 bit(int bitno)
4254 return 1 << (bitno & 31);
4257 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
4259 struct kvm_cpuid_entry2 *best;
4260 struct vcpu_vmx *vmx = to_vmx(vcpu);
4263 vmx->rdtscp_enabled = false;
4264 if (vmx_rdtscp_supported()) {
4265 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
4266 if (exec_control & SECONDARY_EXEC_RDTSCP) {
4267 best = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
4268 if (best && (best->edx & bit(X86_FEATURE_RDTSCP)))
4269 vmx->rdtscp_enabled = true;
4271 exec_control &= ~SECONDARY_EXEC_RDTSCP;
4272 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
4279 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
4283 static struct kvm_x86_ops vmx_x86_ops = {
4284 .cpu_has_kvm_support = cpu_has_kvm_support,
4285 .disabled_by_bios = vmx_disabled_by_bios,
4286 .hardware_setup = hardware_setup,
4287 .hardware_unsetup = hardware_unsetup,
4288 .check_processor_compatibility = vmx_check_processor_compat,
4289 .hardware_enable = hardware_enable,
4290 .hardware_disable = hardware_disable,
4291 .cpu_has_accelerated_tpr = report_flexpriority,
4293 .vcpu_create = vmx_create_vcpu,
4294 .vcpu_free = vmx_free_vcpu,
4295 .vcpu_reset = vmx_vcpu_reset,
4297 .prepare_guest_switch = vmx_save_host_state,
4298 .vcpu_load = vmx_vcpu_load,
4299 .vcpu_put = vmx_vcpu_put,
4301 .set_guest_debug = set_guest_debug,
4302 .get_msr = vmx_get_msr,
4303 .set_msr = vmx_set_msr,
4304 .get_segment_base = vmx_get_segment_base,
4305 .get_segment = vmx_get_segment,
4306 .set_segment = vmx_set_segment,
4307 .get_cpl = vmx_get_cpl,
4308 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
4309 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
4310 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
4311 .set_cr0 = vmx_set_cr0,
4312 .set_cr3 = vmx_set_cr3,
4313 .set_cr4 = vmx_set_cr4,
4314 .set_efer = vmx_set_efer,
4315 .get_idt = vmx_get_idt,
4316 .set_idt = vmx_set_idt,
4317 .get_gdt = vmx_get_gdt,
4318 .set_gdt = vmx_set_gdt,
4319 .set_dr7 = vmx_set_dr7,
4320 .cache_reg = vmx_cache_reg,
4321 .get_rflags = vmx_get_rflags,
4322 .set_rflags = vmx_set_rflags,
4323 .fpu_activate = vmx_fpu_activate,
4324 .fpu_deactivate = vmx_fpu_deactivate,
4326 .tlb_flush = vmx_flush_tlb,
4328 .run = vmx_vcpu_run,
4329 .handle_exit = vmx_handle_exit,
4330 .skip_emulated_instruction = skip_emulated_instruction,
4331 .set_interrupt_shadow = vmx_set_interrupt_shadow,
4332 .get_interrupt_shadow = vmx_get_interrupt_shadow,
4333 .patch_hypercall = vmx_patch_hypercall,
4334 .set_irq = vmx_inject_irq,
4335 .set_nmi = vmx_inject_nmi,
4336 .queue_exception = vmx_queue_exception,
4337 .interrupt_allowed = vmx_interrupt_allowed,
4338 .nmi_allowed = vmx_nmi_allowed,
4339 .get_nmi_mask = vmx_get_nmi_mask,
4340 .set_nmi_mask = vmx_set_nmi_mask,
4341 .enable_nmi_window = enable_nmi_window,
4342 .enable_irq_window = enable_irq_window,
4343 .update_cr8_intercept = update_cr8_intercept,
4345 .set_tss_addr = vmx_set_tss_addr,
4346 .get_tdp_level = get_ept_level,
4347 .get_mt_mask = vmx_get_mt_mask,
4349 .exit_reasons_str = vmx_exit_reasons_str,
4350 .get_lpage_level = vmx_get_lpage_level,
4352 .cpuid_update = vmx_cpuid_update,
4354 .rdtscp_supported = vmx_rdtscp_supported,
4356 .set_supported_cpuid = vmx_set_supported_cpuid,
4358 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
4361 static int __init vmx_init(void)
4365 rdmsrl_safe(MSR_EFER, &host_efer);
4367 for (i = 0; i < NR_VMX_MSR; ++i)
4368 kvm_define_shared_msr(i, vmx_msr_index[i]);
4370 vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
4371 if (!vmx_io_bitmap_a)
4374 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
4375 if (!vmx_io_bitmap_b) {
4380 vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
4381 if (!vmx_msr_bitmap_legacy) {
4386 vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
4387 if (!vmx_msr_bitmap_longmode) {
4393 * Allow direct access to the PC debug port (it is often used for I/O
4394 * delays, but the vmexits simply slow things down).
4396 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
4397 clear_bit(0x80, vmx_io_bitmap_a);
4399 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
4401 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
4402 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
4404 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
4406 r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
4407 __alignof__(struct vcpu_vmx), THIS_MODULE);
4411 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
4412 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
4413 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
4414 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
4415 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
4416 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
4419 bypass_guest_pf = 0;
4420 kvm_mmu_set_base_ptes(VMX_EPT_READABLE_MASK |
4421 VMX_EPT_WRITABLE_MASK);
4422 kvm_mmu_set_mask_ptes(0ull, 0ull, 0ull, 0ull,
4423 VMX_EPT_EXECUTABLE_MASK);
4428 if (bypass_guest_pf)
4429 kvm_mmu_set_nonpresent_ptes(~0xffeull, 0ull);
4434 free_page((unsigned long)vmx_msr_bitmap_longmode);
4436 free_page((unsigned long)vmx_msr_bitmap_legacy);
4438 free_page((unsigned long)vmx_io_bitmap_b);
4440 free_page((unsigned long)vmx_io_bitmap_a);
4444 static void __exit vmx_exit(void)
4446 free_page((unsigned long)vmx_msr_bitmap_legacy);
4447 free_page((unsigned long)vmx_msr_bitmap_longmode);
4448 free_page((unsigned long)vmx_io_bitmap_b);
4449 free_page((unsigned long)vmx_io_bitmap_a);
4454 module_init(vmx_init)
4455 module_exit(vmx_exit)