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 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
832 if (is_long_mode(&vmx->vcpu))
833 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
835 for (i = 0; i < vmx->save_nmsrs; ++i)
836 kvm_set_shared_msr(vmx->guest_msrs[i].index,
837 vmx->guest_msrs[i].data,
838 vmx->guest_msrs[i].mask);
841 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
843 if (!vmx->host_state.loaded)
846 ++vmx->vcpu.stat.host_state_reload;
847 vmx->host_state.loaded = 0;
849 if (is_long_mode(&vmx->vcpu))
850 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
852 if (vmx->host_state.gs_ldt_reload_needed) {
853 kvm_load_ldt(vmx->host_state.ldt_sel);
855 load_gs_index(vmx->host_state.gs_sel);
857 loadsegment(gs, vmx->host_state.gs_sel);
860 if (vmx->host_state.fs_reload_needed)
861 loadsegment(fs, vmx->host_state.fs_sel);
864 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
866 if (current_thread_info()->status & TS_USEDFPU)
868 load_gdt(&__get_cpu_var(host_gdt));
871 static void vmx_load_host_state(struct vcpu_vmx *vmx)
874 __vmx_load_host_state(vmx);
879 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
880 * vcpu mutex is already taken.
882 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
884 struct vcpu_vmx *vmx = to_vmx(vcpu);
885 u64 tsc_this, delta, new_offset;
886 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
889 kvm_cpu_vmxon(phys_addr);
890 else if (vcpu->cpu != cpu)
893 if (per_cpu(current_vmcs, cpu) != vmx->vmcs) {
894 per_cpu(current_vmcs, cpu) = vmx->vmcs;
895 vmcs_load(vmx->vmcs);
898 if (vcpu->cpu != cpu) {
900 unsigned long sysenter_esp;
902 kvm_migrate_timers(vcpu);
903 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
905 list_add(&vmx->local_vcpus_link,
906 &per_cpu(vcpus_on_cpu, cpu));
911 * Linux uses per-cpu TSS and GDT, so set these when switching
914 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
915 native_store_gdt(&dt);
916 vmcs_writel(HOST_GDTR_BASE, dt.address); /* 22.2.4 */
918 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
919 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
922 * Make sure the time stamp counter is monotonous.
925 if (tsc_this < vcpu->arch.host_tsc) {
926 delta = vcpu->arch.host_tsc - tsc_this;
927 new_offset = vmcs_read64(TSC_OFFSET) + delta;
928 vmcs_write64(TSC_OFFSET, new_offset);
933 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
935 __vmx_load_host_state(to_vmx(vcpu));
936 if (!vmm_exclusive) {
937 __vcpu_clear(to_vmx(vcpu));
942 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
946 if (vcpu->fpu_active)
948 vcpu->fpu_active = 1;
949 cr0 = vmcs_readl(GUEST_CR0);
950 cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
951 cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
952 vmcs_writel(GUEST_CR0, cr0);
953 update_exception_bitmap(vcpu);
954 vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
955 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
958 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
960 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
962 vmx_decache_cr0_guest_bits(vcpu);
963 vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
964 update_exception_bitmap(vcpu);
965 vcpu->arch.cr0_guest_owned_bits = 0;
966 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
967 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
970 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
972 unsigned long rflags, save_rflags;
974 rflags = vmcs_readl(GUEST_RFLAGS);
975 if (to_vmx(vcpu)->rmode.vm86_active) {
976 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
977 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
978 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
983 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
985 if (to_vmx(vcpu)->rmode.vm86_active) {
986 to_vmx(vcpu)->rmode.save_rflags = rflags;
987 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
989 vmcs_writel(GUEST_RFLAGS, rflags);
992 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
994 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
997 if (interruptibility & GUEST_INTR_STATE_STI)
998 ret |= KVM_X86_SHADOW_INT_STI;
999 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1000 ret |= KVM_X86_SHADOW_INT_MOV_SS;
1005 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1007 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1008 u32 interruptibility = interruptibility_old;
1010 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1012 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1013 interruptibility |= GUEST_INTR_STATE_MOV_SS;
1014 else if (mask & KVM_X86_SHADOW_INT_STI)
1015 interruptibility |= GUEST_INTR_STATE_STI;
1017 if ((interruptibility != interruptibility_old))
1018 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1021 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
1025 rip = kvm_rip_read(vcpu);
1026 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1027 kvm_rip_write(vcpu, rip);
1029 /* skipping an emulated instruction also counts */
1030 vmx_set_interrupt_shadow(vcpu, 0);
1033 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
1034 bool has_error_code, u32 error_code,
1037 struct vcpu_vmx *vmx = to_vmx(vcpu);
1038 u32 intr_info = nr | INTR_INFO_VALID_MASK;
1040 if (has_error_code) {
1041 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1042 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1045 if (vmx->rmode.vm86_active) {
1046 vmx->rmode.irq.pending = true;
1047 vmx->rmode.irq.vector = nr;
1048 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
1049 if (kvm_exception_is_soft(nr))
1050 vmx->rmode.irq.rip +=
1051 vmx->vcpu.arch.event_exit_inst_len;
1052 intr_info |= INTR_TYPE_SOFT_INTR;
1053 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1054 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
1055 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
1059 if (kvm_exception_is_soft(nr)) {
1060 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1061 vmx->vcpu.arch.event_exit_inst_len);
1062 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1064 intr_info |= INTR_TYPE_HARD_EXCEPTION;
1066 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1069 static bool vmx_rdtscp_supported(void)
1071 return cpu_has_vmx_rdtscp();
1075 * Swap MSR entry in host/guest MSR entry array.
1077 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1079 struct shared_msr_entry tmp;
1081 tmp = vmx->guest_msrs[to];
1082 vmx->guest_msrs[to] = vmx->guest_msrs[from];
1083 vmx->guest_msrs[from] = tmp;
1087 * Set up the vmcs to automatically save and restore system
1088 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
1089 * mode, as fiddling with msrs is very expensive.
1091 static void setup_msrs(struct vcpu_vmx *vmx)
1093 int save_nmsrs, index;
1094 unsigned long *msr_bitmap;
1096 vmx_load_host_state(vmx);
1098 #ifdef CONFIG_X86_64
1099 if (is_long_mode(&vmx->vcpu)) {
1100 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1102 move_msr_up(vmx, index, save_nmsrs++);
1103 index = __find_msr_index(vmx, MSR_LSTAR);
1105 move_msr_up(vmx, index, save_nmsrs++);
1106 index = __find_msr_index(vmx, MSR_CSTAR);
1108 move_msr_up(vmx, index, save_nmsrs++);
1109 index = __find_msr_index(vmx, MSR_TSC_AUX);
1110 if (index >= 0 && vmx->rdtscp_enabled)
1111 move_msr_up(vmx, index, save_nmsrs++);
1113 * MSR_STAR is only needed on long mode guests, and only
1114 * if efer.sce is enabled.
1116 index = __find_msr_index(vmx, MSR_STAR);
1117 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
1118 move_msr_up(vmx, index, save_nmsrs++);
1121 index = __find_msr_index(vmx, MSR_EFER);
1122 if (index >= 0 && update_transition_efer(vmx, index))
1123 move_msr_up(vmx, index, save_nmsrs++);
1125 vmx->save_nmsrs = save_nmsrs;
1127 if (cpu_has_vmx_msr_bitmap()) {
1128 if (is_long_mode(&vmx->vcpu))
1129 msr_bitmap = vmx_msr_bitmap_longmode;
1131 msr_bitmap = vmx_msr_bitmap_legacy;
1133 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
1138 * reads and returns guest's timestamp counter "register"
1139 * guest_tsc = host_tsc + tsc_offset -- 21.3
1141 static u64 guest_read_tsc(void)
1143 u64 host_tsc, tsc_offset;
1146 tsc_offset = vmcs_read64(TSC_OFFSET);
1147 return host_tsc + tsc_offset;
1151 * writes 'guest_tsc' into guest's timestamp counter "register"
1152 * guest_tsc = host_tsc + tsc_offset ==> tsc_offset = guest_tsc - host_tsc
1154 static void guest_write_tsc(u64 guest_tsc, u64 host_tsc)
1156 vmcs_write64(TSC_OFFSET, guest_tsc - host_tsc);
1160 * Reads an msr value (of 'msr_index') into 'pdata'.
1161 * Returns 0 on success, non-0 otherwise.
1162 * Assumes vcpu_load() was already called.
1164 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1167 struct shared_msr_entry *msr;
1170 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
1174 switch (msr_index) {
1175 #ifdef CONFIG_X86_64
1177 data = vmcs_readl(GUEST_FS_BASE);
1180 data = vmcs_readl(GUEST_GS_BASE);
1182 case MSR_KERNEL_GS_BASE:
1183 vmx_load_host_state(to_vmx(vcpu));
1184 data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
1188 return kvm_get_msr_common(vcpu, msr_index, pdata);
1190 data = guest_read_tsc();
1192 case MSR_IA32_SYSENTER_CS:
1193 data = vmcs_read32(GUEST_SYSENTER_CS);
1195 case MSR_IA32_SYSENTER_EIP:
1196 data = vmcs_readl(GUEST_SYSENTER_EIP);
1198 case MSR_IA32_SYSENTER_ESP:
1199 data = vmcs_readl(GUEST_SYSENTER_ESP);
1202 if (!to_vmx(vcpu)->rdtscp_enabled)
1204 /* Otherwise falls through */
1206 vmx_load_host_state(to_vmx(vcpu));
1207 msr = find_msr_entry(to_vmx(vcpu), msr_index);
1209 vmx_load_host_state(to_vmx(vcpu));
1213 return kvm_get_msr_common(vcpu, msr_index, pdata);
1221 * Writes msr value into into the appropriate "register".
1222 * Returns 0 on success, non-0 otherwise.
1223 * Assumes vcpu_load() was already called.
1225 static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1227 struct vcpu_vmx *vmx = to_vmx(vcpu);
1228 struct shared_msr_entry *msr;
1232 switch (msr_index) {
1234 vmx_load_host_state(vmx);
1235 ret = kvm_set_msr_common(vcpu, msr_index, data);
1237 #ifdef CONFIG_X86_64
1239 vmcs_writel(GUEST_FS_BASE, data);
1242 vmcs_writel(GUEST_GS_BASE, data);
1244 case MSR_KERNEL_GS_BASE:
1245 vmx_load_host_state(vmx);
1246 vmx->msr_guest_kernel_gs_base = data;
1249 case MSR_IA32_SYSENTER_CS:
1250 vmcs_write32(GUEST_SYSENTER_CS, data);
1252 case MSR_IA32_SYSENTER_EIP:
1253 vmcs_writel(GUEST_SYSENTER_EIP, data);
1255 case MSR_IA32_SYSENTER_ESP:
1256 vmcs_writel(GUEST_SYSENTER_ESP, data);
1260 guest_write_tsc(data, host_tsc);
1262 case MSR_IA32_CR_PAT:
1263 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
1264 vmcs_write64(GUEST_IA32_PAT, data);
1265 vcpu->arch.pat = data;
1268 ret = kvm_set_msr_common(vcpu, msr_index, data);
1271 if (!vmx->rdtscp_enabled)
1273 /* Check reserved bit, higher 32 bits should be zero */
1274 if ((data >> 32) != 0)
1276 /* Otherwise falls through */
1278 msr = find_msr_entry(vmx, msr_index);
1280 vmx_load_host_state(vmx);
1284 ret = kvm_set_msr_common(vcpu, msr_index, data);
1290 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1292 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
1295 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
1298 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
1300 case VCPU_EXREG_PDPTR:
1302 ept_save_pdptrs(vcpu);
1309 static void set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
1311 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1312 vmcs_writel(GUEST_DR7, dbg->arch.debugreg[7]);
1314 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
1316 update_exception_bitmap(vcpu);
1319 static __init int cpu_has_kvm_support(void)
1321 return cpu_has_vmx();
1324 static __init int vmx_disabled_by_bios(void)
1328 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
1329 if (msr & FEATURE_CONTROL_LOCKED) {
1330 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
1333 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
1334 && !tboot_enabled())
1339 /* locked but not enabled */
1342 static void kvm_cpu_vmxon(u64 addr)
1344 asm volatile (ASM_VMX_VMXON_RAX
1345 : : "a"(&addr), "m"(addr)
1349 static int hardware_enable(void *garbage)
1351 int cpu = raw_smp_processor_id();
1352 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1355 if (read_cr4() & X86_CR4_VMXE)
1358 INIT_LIST_HEAD(&per_cpu(vcpus_on_cpu, cpu));
1359 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
1361 test_bits = FEATURE_CONTROL_LOCKED;
1362 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
1363 if (tboot_enabled())
1364 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
1366 if ((old & test_bits) != test_bits) {
1367 /* enable and lock */
1368 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
1370 write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
1372 if (vmm_exclusive) {
1373 kvm_cpu_vmxon(phys_addr);
1377 store_gdt(&__get_cpu_var(host_gdt));
1382 static void vmclear_local_vcpus(void)
1384 int cpu = raw_smp_processor_id();
1385 struct vcpu_vmx *vmx, *n;
1387 list_for_each_entry_safe(vmx, n, &per_cpu(vcpus_on_cpu, cpu),
1393 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
1396 static void kvm_cpu_vmxoff(void)
1398 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
1401 static void hardware_disable(void *garbage)
1403 if (vmm_exclusive) {
1404 vmclear_local_vcpus();
1407 write_cr4(read_cr4() & ~X86_CR4_VMXE);
1410 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
1411 u32 msr, u32 *result)
1413 u32 vmx_msr_low, vmx_msr_high;
1414 u32 ctl = ctl_min | ctl_opt;
1416 rdmsr(msr, vmx_msr_low, vmx_msr_high);
1418 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
1419 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
1421 /* Ensure minimum (required) set of control bits are supported. */
1429 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
1431 u32 vmx_msr_low, vmx_msr_high;
1432 u32 min, opt, min2, opt2;
1433 u32 _pin_based_exec_control = 0;
1434 u32 _cpu_based_exec_control = 0;
1435 u32 _cpu_based_2nd_exec_control = 0;
1436 u32 _vmexit_control = 0;
1437 u32 _vmentry_control = 0;
1439 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
1440 opt = PIN_BASED_VIRTUAL_NMIS;
1441 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
1442 &_pin_based_exec_control) < 0)
1445 min = CPU_BASED_HLT_EXITING |
1446 #ifdef CONFIG_X86_64
1447 CPU_BASED_CR8_LOAD_EXITING |
1448 CPU_BASED_CR8_STORE_EXITING |
1450 CPU_BASED_CR3_LOAD_EXITING |
1451 CPU_BASED_CR3_STORE_EXITING |
1452 CPU_BASED_USE_IO_BITMAPS |
1453 CPU_BASED_MOV_DR_EXITING |
1454 CPU_BASED_USE_TSC_OFFSETING |
1455 CPU_BASED_MWAIT_EXITING |
1456 CPU_BASED_MONITOR_EXITING |
1457 CPU_BASED_INVLPG_EXITING;
1458 opt = CPU_BASED_TPR_SHADOW |
1459 CPU_BASED_USE_MSR_BITMAPS |
1460 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1461 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
1462 &_cpu_based_exec_control) < 0)
1464 #ifdef CONFIG_X86_64
1465 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
1466 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
1467 ~CPU_BASED_CR8_STORE_EXITING;
1469 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
1471 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
1472 SECONDARY_EXEC_WBINVD_EXITING |
1473 SECONDARY_EXEC_ENABLE_VPID |
1474 SECONDARY_EXEC_ENABLE_EPT |
1475 SECONDARY_EXEC_UNRESTRICTED_GUEST |
1476 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
1477 SECONDARY_EXEC_RDTSCP;
1478 if (adjust_vmx_controls(min2, opt2,
1479 MSR_IA32_VMX_PROCBASED_CTLS2,
1480 &_cpu_based_2nd_exec_control) < 0)
1483 #ifndef CONFIG_X86_64
1484 if (!(_cpu_based_2nd_exec_control &
1485 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
1486 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
1488 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
1489 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
1491 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
1492 CPU_BASED_CR3_STORE_EXITING |
1493 CPU_BASED_INVLPG_EXITING);
1494 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
1495 vmx_capability.ept, vmx_capability.vpid);
1499 #ifdef CONFIG_X86_64
1500 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
1502 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT;
1503 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
1504 &_vmexit_control) < 0)
1508 opt = VM_ENTRY_LOAD_IA32_PAT;
1509 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
1510 &_vmentry_control) < 0)
1513 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
1515 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
1516 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
1519 #ifdef CONFIG_X86_64
1520 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
1521 if (vmx_msr_high & (1u<<16))
1525 /* Require Write-Back (WB) memory type for VMCS accesses. */
1526 if (((vmx_msr_high >> 18) & 15) != 6)
1529 vmcs_conf->size = vmx_msr_high & 0x1fff;
1530 vmcs_conf->order = get_order(vmcs_config.size);
1531 vmcs_conf->revision_id = vmx_msr_low;
1533 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
1534 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
1535 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
1536 vmcs_conf->vmexit_ctrl = _vmexit_control;
1537 vmcs_conf->vmentry_ctrl = _vmentry_control;
1542 static struct vmcs *alloc_vmcs_cpu(int cpu)
1544 int node = cpu_to_node(cpu);
1548 pages = alloc_pages_exact_node(node, GFP_KERNEL, vmcs_config.order);
1551 vmcs = page_address(pages);
1552 memset(vmcs, 0, vmcs_config.size);
1553 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
1557 static struct vmcs *alloc_vmcs(void)
1559 return alloc_vmcs_cpu(raw_smp_processor_id());
1562 static void free_vmcs(struct vmcs *vmcs)
1564 free_pages((unsigned long)vmcs, vmcs_config.order);
1567 static void free_kvm_area(void)
1571 for_each_possible_cpu(cpu) {
1572 free_vmcs(per_cpu(vmxarea, cpu));
1573 per_cpu(vmxarea, cpu) = NULL;
1577 static __init int alloc_kvm_area(void)
1581 for_each_possible_cpu(cpu) {
1584 vmcs = alloc_vmcs_cpu(cpu);
1590 per_cpu(vmxarea, cpu) = vmcs;
1595 static __init int hardware_setup(void)
1597 if (setup_vmcs_config(&vmcs_config) < 0)
1600 if (boot_cpu_has(X86_FEATURE_NX))
1601 kvm_enable_efer_bits(EFER_NX);
1603 if (!cpu_has_vmx_vpid())
1606 if (!cpu_has_vmx_ept() ||
1607 !cpu_has_vmx_ept_4levels()) {
1609 enable_unrestricted_guest = 0;
1612 if (!cpu_has_vmx_unrestricted_guest())
1613 enable_unrestricted_guest = 0;
1615 if (!cpu_has_vmx_flexpriority())
1616 flexpriority_enabled = 0;
1618 if (!cpu_has_vmx_tpr_shadow())
1619 kvm_x86_ops->update_cr8_intercept = NULL;
1621 if (enable_ept && !cpu_has_vmx_ept_2m_page())
1622 kvm_disable_largepages();
1624 if (!cpu_has_vmx_ple())
1627 return alloc_kvm_area();
1630 static __exit void hardware_unsetup(void)
1635 static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
1637 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1639 if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
1640 vmcs_write16(sf->selector, save->selector);
1641 vmcs_writel(sf->base, save->base);
1642 vmcs_write32(sf->limit, save->limit);
1643 vmcs_write32(sf->ar_bytes, save->ar);
1645 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
1647 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
1651 static void enter_pmode(struct kvm_vcpu *vcpu)
1653 unsigned long flags;
1654 struct vcpu_vmx *vmx = to_vmx(vcpu);
1656 vmx->emulation_required = 1;
1657 vmx->rmode.vm86_active = 0;
1659 vmcs_writel(GUEST_TR_BASE, vmx->rmode.tr.base);
1660 vmcs_write32(GUEST_TR_LIMIT, vmx->rmode.tr.limit);
1661 vmcs_write32(GUEST_TR_AR_BYTES, vmx->rmode.tr.ar);
1663 flags = vmcs_readl(GUEST_RFLAGS);
1664 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1665 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1666 vmcs_writel(GUEST_RFLAGS, flags);
1668 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
1669 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
1671 update_exception_bitmap(vcpu);
1673 if (emulate_invalid_guest_state)
1676 fix_pmode_dataseg(VCPU_SREG_ES, &vmx->rmode.es);
1677 fix_pmode_dataseg(VCPU_SREG_DS, &vmx->rmode.ds);
1678 fix_pmode_dataseg(VCPU_SREG_GS, &vmx->rmode.gs);
1679 fix_pmode_dataseg(VCPU_SREG_FS, &vmx->rmode.fs);
1681 vmcs_write16(GUEST_SS_SELECTOR, 0);
1682 vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
1684 vmcs_write16(GUEST_CS_SELECTOR,
1685 vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
1686 vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1689 static gva_t rmode_tss_base(struct kvm *kvm)
1691 if (!kvm->arch.tss_addr) {
1692 struct kvm_memslots *slots;
1695 slots = kvm_memslots(kvm);
1696 base_gfn = slots->memslots[0].base_gfn +
1697 kvm->memslots->memslots[0].npages - 3;
1698 return base_gfn << PAGE_SHIFT;
1700 return kvm->arch.tss_addr;
1703 static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
1705 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1707 save->selector = vmcs_read16(sf->selector);
1708 save->base = vmcs_readl(sf->base);
1709 save->limit = vmcs_read32(sf->limit);
1710 save->ar = vmcs_read32(sf->ar_bytes);
1711 vmcs_write16(sf->selector, save->base >> 4);
1712 vmcs_write32(sf->base, save->base & 0xfffff);
1713 vmcs_write32(sf->limit, 0xffff);
1714 vmcs_write32(sf->ar_bytes, 0xf3);
1717 static void enter_rmode(struct kvm_vcpu *vcpu)
1719 unsigned long flags;
1720 struct vcpu_vmx *vmx = to_vmx(vcpu);
1722 if (enable_unrestricted_guest)
1725 vmx->emulation_required = 1;
1726 vmx->rmode.vm86_active = 1;
1728 vmx->rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
1729 vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
1731 vmx->rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
1732 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
1734 vmx->rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
1735 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1737 flags = vmcs_readl(GUEST_RFLAGS);
1738 vmx->rmode.save_rflags = flags;
1740 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1742 vmcs_writel(GUEST_RFLAGS, flags);
1743 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
1744 update_exception_bitmap(vcpu);
1746 if (emulate_invalid_guest_state)
1747 goto continue_rmode;
1749 vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
1750 vmcs_write32(GUEST_SS_LIMIT, 0xffff);
1751 vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
1753 vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
1754 vmcs_write32(GUEST_CS_LIMIT, 0xffff);
1755 if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
1756 vmcs_writel(GUEST_CS_BASE, 0xf0000);
1757 vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
1759 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.es);
1760 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.ds);
1761 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.gs);
1762 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.fs);
1765 kvm_mmu_reset_context(vcpu);
1766 init_rmode(vcpu->kvm);
1769 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
1771 struct vcpu_vmx *vmx = to_vmx(vcpu);
1772 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
1778 * Force kernel_gs_base reloading before EFER changes, as control
1779 * of this msr depends on is_long_mode().
1781 vmx_load_host_state(to_vmx(vcpu));
1782 vcpu->arch.efer = efer;
1783 if (efer & EFER_LMA) {
1784 vmcs_write32(VM_ENTRY_CONTROLS,
1785 vmcs_read32(VM_ENTRY_CONTROLS) |
1786 VM_ENTRY_IA32E_MODE);
1789 vmcs_write32(VM_ENTRY_CONTROLS,
1790 vmcs_read32(VM_ENTRY_CONTROLS) &
1791 ~VM_ENTRY_IA32E_MODE);
1793 msr->data = efer & ~EFER_LME;
1798 #ifdef CONFIG_X86_64
1800 static void enter_lmode(struct kvm_vcpu *vcpu)
1804 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
1805 if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
1806 printk(KERN_DEBUG "%s: tss fixup for long mode. \n",
1808 vmcs_write32(GUEST_TR_AR_BYTES,
1809 (guest_tr_ar & ~AR_TYPE_MASK)
1810 | AR_TYPE_BUSY_64_TSS);
1812 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
1815 static void exit_lmode(struct kvm_vcpu *vcpu)
1817 vmcs_write32(VM_ENTRY_CONTROLS,
1818 vmcs_read32(VM_ENTRY_CONTROLS)
1819 & ~VM_ENTRY_IA32E_MODE);
1820 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
1825 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
1827 vpid_sync_context(to_vmx(vcpu));
1829 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
1831 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
1835 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
1837 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
1839 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
1840 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
1843 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1845 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
1847 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
1848 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
1851 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
1853 if (!test_bit(VCPU_EXREG_PDPTR,
1854 (unsigned long *)&vcpu->arch.regs_dirty))
1857 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1858 vmcs_write64(GUEST_PDPTR0, vcpu->arch.pdptrs[0]);
1859 vmcs_write64(GUEST_PDPTR1, vcpu->arch.pdptrs[1]);
1860 vmcs_write64(GUEST_PDPTR2, vcpu->arch.pdptrs[2]);
1861 vmcs_write64(GUEST_PDPTR3, vcpu->arch.pdptrs[3]);
1865 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
1867 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1868 vcpu->arch.pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
1869 vcpu->arch.pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
1870 vcpu->arch.pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
1871 vcpu->arch.pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
1874 __set_bit(VCPU_EXREG_PDPTR,
1875 (unsigned long *)&vcpu->arch.regs_avail);
1876 __set_bit(VCPU_EXREG_PDPTR,
1877 (unsigned long *)&vcpu->arch.regs_dirty);
1880 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
1882 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
1884 struct kvm_vcpu *vcpu)
1886 if (!(cr0 & X86_CR0_PG)) {
1887 /* From paging/starting to nonpaging */
1888 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
1889 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
1890 (CPU_BASED_CR3_LOAD_EXITING |
1891 CPU_BASED_CR3_STORE_EXITING));
1892 vcpu->arch.cr0 = cr0;
1893 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
1894 } else if (!is_paging(vcpu)) {
1895 /* From nonpaging to paging */
1896 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
1897 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
1898 ~(CPU_BASED_CR3_LOAD_EXITING |
1899 CPU_BASED_CR3_STORE_EXITING));
1900 vcpu->arch.cr0 = cr0;
1901 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
1904 if (!(cr0 & X86_CR0_WP))
1905 *hw_cr0 &= ~X86_CR0_WP;
1908 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1910 struct vcpu_vmx *vmx = to_vmx(vcpu);
1911 unsigned long hw_cr0;
1913 if (enable_unrestricted_guest)
1914 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST)
1915 | KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
1917 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON;
1919 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
1922 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
1925 #ifdef CONFIG_X86_64
1926 if (vcpu->arch.efer & EFER_LME) {
1927 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
1929 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
1935 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
1937 if (!vcpu->fpu_active)
1938 hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
1940 vmcs_writel(CR0_READ_SHADOW, cr0);
1941 vmcs_writel(GUEST_CR0, hw_cr0);
1942 vcpu->arch.cr0 = cr0;
1945 static u64 construct_eptp(unsigned long root_hpa)
1949 /* TODO write the value reading from MSR */
1950 eptp = VMX_EPT_DEFAULT_MT |
1951 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
1952 eptp |= (root_hpa & PAGE_MASK);
1957 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1959 unsigned long guest_cr3;
1964 eptp = construct_eptp(cr3);
1965 vmcs_write64(EPT_POINTER, eptp);
1966 guest_cr3 = is_paging(vcpu) ? vcpu->arch.cr3 :
1967 vcpu->kvm->arch.ept_identity_map_addr;
1968 ept_load_pdptrs(vcpu);
1971 vmx_flush_tlb(vcpu);
1972 vmcs_writel(GUEST_CR3, guest_cr3);
1975 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1977 unsigned long hw_cr4 = cr4 | (to_vmx(vcpu)->rmode.vm86_active ?
1978 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
1980 vcpu->arch.cr4 = cr4;
1982 if (!is_paging(vcpu)) {
1983 hw_cr4 &= ~X86_CR4_PAE;
1984 hw_cr4 |= X86_CR4_PSE;
1985 } else if (!(cr4 & X86_CR4_PAE)) {
1986 hw_cr4 &= ~X86_CR4_PAE;
1990 vmcs_writel(CR4_READ_SHADOW, cr4);
1991 vmcs_writel(GUEST_CR4, hw_cr4);
1994 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1996 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1998 return vmcs_readl(sf->base);
2001 static void vmx_get_segment(struct kvm_vcpu *vcpu,
2002 struct kvm_segment *var, int seg)
2004 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2007 var->base = vmcs_readl(sf->base);
2008 var->limit = vmcs_read32(sf->limit);
2009 var->selector = vmcs_read16(sf->selector);
2010 ar = vmcs_read32(sf->ar_bytes);
2011 if ((ar & AR_UNUSABLE_MASK) && !emulate_invalid_guest_state)
2013 var->type = ar & 15;
2014 var->s = (ar >> 4) & 1;
2015 var->dpl = (ar >> 5) & 3;
2016 var->present = (ar >> 7) & 1;
2017 var->avl = (ar >> 12) & 1;
2018 var->l = (ar >> 13) & 1;
2019 var->db = (ar >> 14) & 1;
2020 var->g = (ar >> 15) & 1;
2021 var->unusable = (ar >> 16) & 1;
2024 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
2026 if (!is_protmode(vcpu))
2029 if (vmx_get_rflags(vcpu) & X86_EFLAGS_VM) /* if virtual 8086 */
2032 return vmcs_read16(GUEST_CS_SELECTOR) & 3;
2035 static u32 vmx_segment_access_rights(struct kvm_segment *var)
2042 ar = var->type & 15;
2043 ar |= (var->s & 1) << 4;
2044 ar |= (var->dpl & 3) << 5;
2045 ar |= (var->present & 1) << 7;
2046 ar |= (var->avl & 1) << 12;
2047 ar |= (var->l & 1) << 13;
2048 ar |= (var->db & 1) << 14;
2049 ar |= (var->g & 1) << 15;
2051 if (ar == 0) /* a 0 value means unusable */
2052 ar = AR_UNUSABLE_MASK;
2057 static void vmx_set_segment(struct kvm_vcpu *vcpu,
2058 struct kvm_segment *var, int seg)
2060 struct vcpu_vmx *vmx = to_vmx(vcpu);
2061 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2064 if (vmx->rmode.vm86_active && seg == VCPU_SREG_TR) {
2065 vmx->rmode.tr.selector = var->selector;
2066 vmx->rmode.tr.base = var->base;
2067 vmx->rmode.tr.limit = var->limit;
2068 vmx->rmode.tr.ar = vmx_segment_access_rights(var);
2071 vmcs_writel(sf->base, var->base);
2072 vmcs_write32(sf->limit, var->limit);
2073 vmcs_write16(sf->selector, var->selector);
2074 if (vmx->rmode.vm86_active && var->s) {
2076 * Hack real-mode segments into vm86 compatibility.
2078 if (var->base == 0xffff0000 && var->selector == 0xf000)
2079 vmcs_writel(sf->base, 0xf0000);
2082 ar = vmx_segment_access_rights(var);
2085 * Fix the "Accessed" bit in AR field of segment registers for older
2087 * IA32 arch specifies that at the time of processor reset the
2088 * "Accessed" bit in the AR field of segment registers is 1. And qemu
2089 * is setting it to 0 in the usedland code. This causes invalid guest
2090 * state vmexit when "unrestricted guest" mode is turned on.
2091 * Fix for this setup issue in cpu_reset is being pushed in the qemu
2092 * tree. Newer qemu binaries with that qemu fix would not need this
2095 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
2096 ar |= 0x1; /* Accessed */
2098 vmcs_write32(sf->ar_bytes, ar);
2101 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2103 u32 ar = vmcs_read32(GUEST_CS_AR_BYTES);
2105 *db = (ar >> 14) & 1;
2106 *l = (ar >> 13) & 1;
2109 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2111 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
2112 dt->address = vmcs_readl(GUEST_IDTR_BASE);
2115 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2117 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
2118 vmcs_writel(GUEST_IDTR_BASE, dt->address);
2121 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2123 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
2124 dt->address = vmcs_readl(GUEST_GDTR_BASE);
2127 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2129 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
2130 vmcs_writel(GUEST_GDTR_BASE, dt->address);
2133 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
2135 struct kvm_segment var;
2138 vmx_get_segment(vcpu, &var, seg);
2139 ar = vmx_segment_access_rights(&var);
2141 if (var.base != (var.selector << 4))
2143 if (var.limit != 0xffff)
2151 static bool code_segment_valid(struct kvm_vcpu *vcpu)
2153 struct kvm_segment cs;
2154 unsigned int cs_rpl;
2156 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
2157 cs_rpl = cs.selector & SELECTOR_RPL_MASK;
2161 if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
2165 if (cs.type & AR_TYPE_WRITEABLE_MASK) {
2166 if (cs.dpl > cs_rpl)
2169 if (cs.dpl != cs_rpl)
2175 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
2179 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
2181 struct kvm_segment ss;
2182 unsigned int ss_rpl;
2184 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
2185 ss_rpl = ss.selector & SELECTOR_RPL_MASK;
2189 if (ss.type != 3 && ss.type != 7)
2193 if (ss.dpl != ss_rpl) /* DPL != RPL */
2201 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
2203 struct kvm_segment var;
2206 vmx_get_segment(vcpu, &var, seg);
2207 rpl = var.selector & SELECTOR_RPL_MASK;
2215 if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
2216 if (var.dpl < rpl) /* DPL < RPL */
2220 /* TODO: Add other members to kvm_segment_field to allow checking for other access
2226 static bool tr_valid(struct kvm_vcpu *vcpu)
2228 struct kvm_segment tr;
2230 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
2234 if (tr.selector & SELECTOR_TI_MASK) /* TI = 1 */
2236 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
2244 static bool ldtr_valid(struct kvm_vcpu *vcpu)
2246 struct kvm_segment ldtr;
2248 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
2252 if (ldtr.selector & SELECTOR_TI_MASK) /* TI = 1 */
2262 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
2264 struct kvm_segment cs, ss;
2266 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
2267 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
2269 return ((cs.selector & SELECTOR_RPL_MASK) ==
2270 (ss.selector & SELECTOR_RPL_MASK));
2274 * Check if guest state is valid. Returns true if valid, false if
2276 * We assume that registers are always usable
2278 static bool guest_state_valid(struct kvm_vcpu *vcpu)
2280 /* real mode guest state checks */
2281 if (!is_protmode(vcpu)) {
2282 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
2284 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
2286 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
2288 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
2290 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
2292 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
2295 /* protected mode guest state checks */
2296 if (!cs_ss_rpl_check(vcpu))
2298 if (!code_segment_valid(vcpu))
2300 if (!stack_segment_valid(vcpu))
2302 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
2304 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
2306 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
2308 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
2310 if (!tr_valid(vcpu))
2312 if (!ldtr_valid(vcpu))
2316 * - Add checks on RIP
2317 * - Add checks on RFLAGS
2323 static int init_rmode_tss(struct kvm *kvm)
2325 gfn_t fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
2330 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
2333 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
2334 r = kvm_write_guest_page(kvm, fn++, &data,
2335 TSS_IOPB_BASE_OFFSET, sizeof(u16));
2338 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
2341 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
2345 r = kvm_write_guest_page(kvm, fn, &data,
2346 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
2356 static int init_rmode_identity_map(struct kvm *kvm)
2359 pfn_t identity_map_pfn;
2364 if (unlikely(!kvm->arch.ept_identity_pagetable)) {
2365 printk(KERN_ERR "EPT: identity-mapping pagetable "
2366 "haven't been allocated!\n");
2369 if (likely(kvm->arch.ept_identity_pagetable_done))
2372 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
2373 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
2376 /* Set up identity-mapping pagetable for EPT in real mode */
2377 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
2378 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
2379 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
2380 r = kvm_write_guest_page(kvm, identity_map_pfn,
2381 &tmp, i * sizeof(tmp), sizeof(tmp));
2385 kvm->arch.ept_identity_pagetable_done = true;
2391 static void seg_setup(int seg)
2393 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2396 vmcs_write16(sf->selector, 0);
2397 vmcs_writel(sf->base, 0);
2398 vmcs_write32(sf->limit, 0xffff);
2399 if (enable_unrestricted_guest) {
2401 if (seg == VCPU_SREG_CS)
2402 ar |= 0x08; /* code segment */
2406 vmcs_write32(sf->ar_bytes, ar);
2409 static int alloc_apic_access_page(struct kvm *kvm)
2411 struct kvm_userspace_memory_region kvm_userspace_mem;
2414 mutex_lock(&kvm->slots_lock);
2415 if (kvm->arch.apic_access_page)
2417 kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
2418 kvm_userspace_mem.flags = 0;
2419 kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
2420 kvm_userspace_mem.memory_size = PAGE_SIZE;
2421 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2425 kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
2427 mutex_unlock(&kvm->slots_lock);
2431 static int alloc_identity_pagetable(struct kvm *kvm)
2433 struct kvm_userspace_memory_region kvm_userspace_mem;
2436 mutex_lock(&kvm->slots_lock);
2437 if (kvm->arch.ept_identity_pagetable)
2439 kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
2440 kvm_userspace_mem.flags = 0;
2441 kvm_userspace_mem.guest_phys_addr =
2442 kvm->arch.ept_identity_map_addr;
2443 kvm_userspace_mem.memory_size = PAGE_SIZE;
2444 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2448 kvm->arch.ept_identity_pagetable = gfn_to_page(kvm,
2449 kvm->arch.ept_identity_map_addr >> PAGE_SHIFT);
2451 mutex_unlock(&kvm->slots_lock);
2455 static void allocate_vpid(struct vcpu_vmx *vmx)
2462 spin_lock(&vmx_vpid_lock);
2463 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
2464 if (vpid < VMX_NR_VPIDS) {
2466 __set_bit(vpid, vmx_vpid_bitmap);
2468 spin_unlock(&vmx_vpid_lock);
2471 static void free_vpid(struct vcpu_vmx *vmx)
2475 spin_lock(&vmx_vpid_lock);
2477 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
2478 spin_unlock(&vmx_vpid_lock);
2481 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap, u32 msr)
2483 int f = sizeof(unsigned long);
2485 if (!cpu_has_vmx_msr_bitmap())
2489 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
2490 * have the write-low and read-high bitmap offsets the wrong way round.
2491 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
2493 if (msr <= 0x1fff) {
2494 __clear_bit(msr, msr_bitmap + 0x000 / f); /* read-low */
2495 __clear_bit(msr, msr_bitmap + 0x800 / f); /* write-low */
2496 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
2498 __clear_bit(msr, msr_bitmap + 0x400 / f); /* read-high */
2499 __clear_bit(msr, msr_bitmap + 0xc00 / f); /* write-high */
2503 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
2506 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy, msr);
2507 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode, msr);
2511 * Sets up the vmcs for emulated real mode.
2513 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
2515 u32 host_sysenter_cs, msr_low, msr_high;
2517 u64 host_pat, tsc_this, tsc_base;
2521 unsigned long kvm_vmx_return;
2525 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
2526 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
2528 if (cpu_has_vmx_msr_bitmap())
2529 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
2531 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
2534 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
2535 vmcs_config.pin_based_exec_ctrl);
2537 exec_control = vmcs_config.cpu_based_exec_ctrl;
2538 if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
2539 exec_control &= ~CPU_BASED_TPR_SHADOW;
2540 #ifdef CONFIG_X86_64
2541 exec_control |= CPU_BASED_CR8_STORE_EXITING |
2542 CPU_BASED_CR8_LOAD_EXITING;
2546 exec_control |= CPU_BASED_CR3_STORE_EXITING |
2547 CPU_BASED_CR3_LOAD_EXITING |
2548 CPU_BASED_INVLPG_EXITING;
2549 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
2551 if (cpu_has_secondary_exec_ctrls()) {
2552 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
2553 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2555 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
2557 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
2559 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
2560 enable_unrestricted_guest = 0;
2562 if (!enable_unrestricted_guest)
2563 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
2565 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
2566 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
2570 vmcs_write32(PLE_GAP, ple_gap);
2571 vmcs_write32(PLE_WINDOW, ple_window);
2574 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, !!bypass_guest_pf);
2575 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, !!bypass_guest_pf);
2576 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
2578 vmcs_writel(HOST_CR0, read_cr0() | X86_CR0_TS); /* 22.2.3 */
2579 vmcs_writel(HOST_CR4, read_cr4()); /* 22.2.3, 22.2.5 */
2580 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
2582 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
2583 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
2584 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
2585 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
2586 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
2587 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
2588 #ifdef CONFIG_X86_64
2589 rdmsrl(MSR_FS_BASE, a);
2590 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
2591 rdmsrl(MSR_GS_BASE, a);
2592 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
2594 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
2595 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
2598 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
2600 native_store_idt(&dt);
2601 vmcs_writel(HOST_IDTR_BASE, dt.address); /* 22.2.4 */
2603 asm("mov $.Lkvm_vmx_return, %0" : "=r"(kvm_vmx_return));
2604 vmcs_writel(HOST_RIP, kvm_vmx_return); /* 22.2.5 */
2605 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
2606 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
2607 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
2608 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
2609 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
2611 rdmsr(MSR_IA32_SYSENTER_CS, host_sysenter_cs, junk);
2612 vmcs_write32(HOST_IA32_SYSENTER_CS, host_sysenter_cs);
2613 rdmsrl(MSR_IA32_SYSENTER_ESP, a);
2614 vmcs_writel(HOST_IA32_SYSENTER_ESP, a); /* 22.2.3 */
2615 rdmsrl(MSR_IA32_SYSENTER_EIP, a);
2616 vmcs_writel(HOST_IA32_SYSENTER_EIP, a); /* 22.2.3 */
2618 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
2619 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2620 host_pat = msr_low | ((u64) msr_high << 32);
2621 vmcs_write64(HOST_IA32_PAT, host_pat);
2623 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2624 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2625 host_pat = msr_low | ((u64) msr_high << 32);
2626 /* Write the default value follow host pat */
2627 vmcs_write64(GUEST_IA32_PAT, host_pat);
2628 /* Keep arch.pat sync with GUEST_IA32_PAT */
2629 vmx->vcpu.arch.pat = host_pat;
2632 for (i = 0; i < NR_VMX_MSR; ++i) {
2633 u32 index = vmx_msr_index[i];
2634 u32 data_low, data_high;
2637 if (rdmsr_safe(index, &data_low, &data_high) < 0)
2639 if (wrmsr_safe(index, data_low, data_high) < 0)
2641 vmx->guest_msrs[j].index = i;
2642 vmx->guest_msrs[j].data = 0;
2643 vmx->guest_msrs[j].mask = -1ull;
2647 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
2649 /* 22.2.1, 20.8.1 */
2650 vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
2652 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
2653 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
2655 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
2656 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
2658 tsc_base = vmx->vcpu.kvm->arch.vm_init_tsc;
2660 if (tsc_this < vmx->vcpu.kvm->arch.vm_init_tsc)
2661 tsc_base = tsc_this;
2663 guest_write_tsc(0, tsc_base);
2668 static int init_rmode(struct kvm *kvm)
2672 idx = srcu_read_lock(&kvm->srcu);
2673 if (!init_rmode_tss(kvm))
2675 if (!init_rmode_identity_map(kvm))
2680 srcu_read_unlock(&kvm->srcu, idx);
2684 static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
2686 struct vcpu_vmx *vmx = to_vmx(vcpu);
2690 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
2691 if (!init_rmode(vmx->vcpu.kvm)) {
2696 vmx->rmode.vm86_active = 0;
2698 vmx->soft_vnmi_blocked = 0;
2700 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
2701 kvm_set_cr8(&vmx->vcpu, 0);
2702 msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
2703 if (kvm_vcpu_is_bsp(&vmx->vcpu))
2704 msr |= MSR_IA32_APICBASE_BSP;
2705 kvm_set_apic_base(&vmx->vcpu, msr);
2707 ret = fx_init(&vmx->vcpu);
2711 seg_setup(VCPU_SREG_CS);
2713 * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
2714 * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4. Sigh.
2716 if (kvm_vcpu_is_bsp(&vmx->vcpu)) {
2717 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
2718 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
2720 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
2721 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
2724 seg_setup(VCPU_SREG_DS);
2725 seg_setup(VCPU_SREG_ES);
2726 seg_setup(VCPU_SREG_FS);
2727 seg_setup(VCPU_SREG_GS);
2728 seg_setup(VCPU_SREG_SS);
2730 vmcs_write16(GUEST_TR_SELECTOR, 0);
2731 vmcs_writel(GUEST_TR_BASE, 0);
2732 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
2733 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2735 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
2736 vmcs_writel(GUEST_LDTR_BASE, 0);
2737 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
2738 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
2740 vmcs_write32(GUEST_SYSENTER_CS, 0);
2741 vmcs_writel(GUEST_SYSENTER_ESP, 0);
2742 vmcs_writel(GUEST_SYSENTER_EIP, 0);
2744 vmcs_writel(GUEST_RFLAGS, 0x02);
2745 if (kvm_vcpu_is_bsp(&vmx->vcpu))
2746 kvm_rip_write(vcpu, 0xfff0);
2748 kvm_rip_write(vcpu, 0);
2749 kvm_register_write(vcpu, VCPU_REGS_RSP, 0);
2751 vmcs_writel(GUEST_DR7, 0x400);
2753 vmcs_writel(GUEST_GDTR_BASE, 0);
2754 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
2756 vmcs_writel(GUEST_IDTR_BASE, 0);
2757 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
2759 vmcs_write32(GUEST_ACTIVITY_STATE, 0);
2760 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
2761 vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
2763 /* Special registers */
2764 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
2768 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
2770 if (cpu_has_vmx_tpr_shadow()) {
2771 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
2772 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
2773 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
2774 page_to_phys(vmx->vcpu.arch.apic->regs_page));
2775 vmcs_write32(TPR_THRESHOLD, 0);
2778 if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2779 vmcs_write64(APIC_ACCESS_ADDR,
2780 page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
2783 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2785 vmx->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
2786 vmx_set_cr0(&vmx->vcpu, kvm_read_cr0(vcpu)); /* enter rmode */
2787 vmx_set_cr4(&vmx->vcpu, 0);
2788 vmx_set_efer(&vmx->vcpu, 0);
2789 vmx_fpu_activate(&vmx->vcpu);
2790 update_exception_bitmap(&vmx->vcpu);
2792 vpid_sync_context(vmx);
2796 /* HACK: Don't enable emulation on guest boot/reset */
2797 vmx->emulation_required = 0;
2803 static void enable_irq_window(struct kvm_vcpu *vcpu)
2805 u32 cpu_based_vm_exec_control;
2807 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2808 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
2809 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2812 static void enable_nmi_window(struct kvm_vcpu *vcpu)
2814 u32 cpu_based_vm_exec_control;
2816 if (!cpu_has_virtual_nmis()) {
2817 enable_irq_window(vcpu);
2821 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2822 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
2823 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2826 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
2828 struct vcpu_vmx *vmx = to_vmx(vcpu);
2830 int irq = vcpu->arch.interrupt.nr;
2832 trace_kvm_inj_virq(irq);
2834 ++vcpu->stat.irq_injections;
2835 if (vmx->rmode.vm86_active) {
2836 vmx->rmode.irq.pending = true;
2837 vmx->rmode.irq.vector = irq;
2838 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
2839 if (vcpu->arch.interrupt.soft)
2840 vmx->rmode.irq.rip +=
2841 vmx->vcpu.arch.event_exit_inst_len;
2842 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2843 irq | INTR_TYPE_SOFT_INTR | INTR_INFO_VALID_MASK);
2844 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
2845 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
2848 intr = irq | INTR_INFO_VALID_MASK;
2849 if (vcpu->arch.interrupt.soft) {
2850 intr |= INTR_TYPE_SOFT_INTR;
2851 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2852 vmx->vcpu.arch.event_exit_inst_len);
2854 intr |= INTR_TYPE_EXT_INTR;
2855 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
2858 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
2860 struct vcpu_vmx *vmx = to_vmx(vcpu);
2862 if (!cpu_has_virtual_nmis()) {
2864 * Tracking the NMI-blocked state in software is built upon
2865 * finding the next open IRQ window. This, in turn, depends on
2866 * well-behaving guests: They have to keep IRQs disabled at
2867 * least as long as the NMI handler runs. Otherwise we may
2868 * cause NMI nesting, maybe breaking the guest. But as this is
2869 * highly unlikely, we can live with the residual risk.
2871 vmx->soft_vnmi_blocked = 1;
2872 vmx->vnmi_blocked_time = 0;
2875 ++vcpu->stat.nmi_injections;
2876 if (vmx->rmode.vm86_active) {
2877 vmx->rmode.irq.pending = true;
2878 vmx->rmode.irq.vector = NMI_VECTOR;
2879 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
2880 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2881 NMI_VECTOR | INTR_TYPE_SOFT_INTR |
2882 INTR_INFO_VALID_MASK);
2883 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
2884 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
2887 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2888 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
2891 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
2893 if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
2896 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
2897 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_NMI));
2900 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
2902 if (!cpu_has_virtual_nmis())
2903 return to_vmx(vcpu)->soft_vnmi_blocked;
2904 return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
2907 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
2909 struct vcpu_vmx *vmx = to_vmx(vcpu);
2911 if (!cpu_has_virtual_nmis()) {
2912 if (vmx->soft_vnmi_blocked != masked) {
2913 vmx->soft_vnmi_blocked = masked;
2914 vmx->vnmi_blocked_time = 0;
2918 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
2919 GUEST_INTR_STATE_NMI);
2921 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
2922 GUEST_INTR_STATE_NMI);
2926 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
2928 return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
2929 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
2930 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
2933 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
2936 struct kvm_userspace_memory_region tss_mem = {
2937 .slot = TSS_PRIVATE_MEMSLOT,
2938 .guest_phys_addr = addr,
2939 .memory_size = PAGE_SIZE * 3,
2943 ret = kvm_set_memory_region(kvm, &tss_mem, 0);
2946 kvm->arch.tss_addr = addr;
2950 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
2951 int vec, u32 err_code)
2954 * Instruction with address size override prefix opcode 0x67
2955 * Cause the #SS fault with 0 error code in VM86 mode.
2957 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
2958 if (emulate_instruction(vcpu, 0, 0, 0) == EMULATE_DONE)
2961 * Forward all other exceptions that are valid in real mode.
2962 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
2963 * the required debugging infrastructure rework.
2967 if (vcpu->guest_debug &
2968 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
2970 kvm_queue_exception(vcpu, vec);
2974 * Update instruction length as we may reinject the exception
2975 * from user space while in guest debugging mode.
2977 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
2978 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
2979 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
2990 kvm_queue_exception(vcpu, vec);
2997 * Trigger machine check on the host. We assume all the MSRs are already set up
2998 * by the CPU and that we still run on the same CPU as the MCE occurred on.
2999 * We pass a fake environment to the machine check handler because we want
3000 * the guest to be always treated like user space, no matter what context
3001 * it used internally.
3003 static void kvm_machine_check(void)
3005 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
3006 struct pt_regs regs = {
3007 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
3008 .flags = X86_EFLAGS_IF,
3011 do_machine_check(®s, 0);
3015 static int handle_machine_check(struct kvm_vcpu *vcpu)
3017 /* already handled by vcpu_run */
3021 static int handle_exception(struct kvm_vcpu *vcpu)
3023 struct vcpu_vmx *vmx = to_vmx(vcpu);
3024 struct kvm_run *kvm_run = vcpu->run;
3025 u32 intr_info, ex_no, error_code;
3026 unsigned long cr2, rip, dr6;
3028 enum emulation_result er;
3030 vect_info = vmx->idt_vectoring_info;
3031 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
3033 if (is_machine_check(intr_info))
3034 return handle_machine_check(vcpu);
3036 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
3037 !is_page_fault(intr_info)) {
3038 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3039 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
3040 vcpu->run->internal.ndata = 2;
3041 vcpu->run->internal.data[0] = vect_info;
3042 vcpu->run->internal.data[1] = intr_info;
3046 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
3047 return 1; /* already handled by vmx_vcpu_run() */
3049 if (is_no_device(intr_info)) {
3050 vmx_fpu_activate(vcpu);
3054 if (is_invalid_opcode(intr_info)) {
3055 er = emulate_instruction(vcpu, 0, 0, EMULTYPE_TRAP_UD);
3056 if (er != EMULATE_DONE)
3057 kvm_queue_exception(vcpu, UD_VECTOR);
3062 rip = kvm_rip_read(vcpu);
3063 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
3064 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
3065 if (is_page_fault(intr_info)) {
3066 /* EPT won't cause page fault directly */
3069 cr2 = vmcs_readl(EXIT_QUALIFICATION);
3070 trace_kvm_page_fault(cr2, error_code);
3072 if (kvm_event_needs_reinjection(vcpu))
3073 kvm_mmu_unprotect_page_virt(vcpu, cr2);
3074 return kvm_mmu_page_fault(vcpu, cr2, error_code);
3077 if (vmx->rmode.vm86_active &&
3078 handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
3080 if (vcpu->arch.halt_request) {
3081 vcpu->arch.halt_request = 0;
3082 return kvm_emulate_halt(vcpu);
3087 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
3090 dr6 = vmcs_readl(EXIT_QUALIFICATION);
3091 if (!(vcpu->guest_debug &
3092 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
3093 vcpu->arch.dr6 = dr6 | DR6_FIXED_1;
3094 kvm_queue_exception(vcpu, DB_VECTOR);
3097 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
3098 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
3102 * Update instruction length as we may reinject #BP from
3103 * user space while in guest debugging mode. Reading it for
3104 * #DB as well causes no harm, it is not used in that case.
3106 vmx->vcpu.arch.event_exit_inst_len =
3107 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3108 kvm_run->exit_reason = KVM_EXIT_DEBUG;
3109 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
3110 kvm_run->debug.arch.exception = ex_no;
3113 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
3114 kvm_run->ex.exception = ex_no;
3115 kvm_run->ex.error_code = error_code;
3121 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
3123 ++vcpu->stat.irq_exits;
3127 static int handle_triple_fault(struct kvm_vcpu *vcpu)
3129 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
3133 static int handle_io(struct kvm_vcpu *vcpu)
3135 unsigned long exit_qualification;
3136 int size, in, string;
3139 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3140 string = (exit_qualification & 16) != 0;
3141 in = (exit_qualification & 8) != 0;
3143 ++vcpu->stat.io_exits;
3146 return emulate_instruction(vcpu, 0, 0, 0) == EMULATE_DONE;
3148 port = exit_qualification >> 16;
3149 size = (exit_qualification & 7) + 1;
3150 skip_emulated_instruction(vcpu);
3152 return kvm_fast_pio_out(vcpu, size, port);
3156 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
3159 * Patch in the VMCALL instruction:
3161 hypercall[0] = 0x0f;
3162 hypercall[1] = 0x01;
3163 hypercall[2] = 0xc1;
3166 static void complete_insn_gp(struct kvm_vcpu *vcpu, int err)
3169 kvm_inject_gp(vcpu, 0);
3171 skip_emulated_instruction(vcpu);
3174 static int handle_cr(struct kvm_vcpu *vcpu)
3176 unsigned long exit_qualification, val;
3181 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3182 cr = exit_qualification & 15;
3183 reg = (exit_qualification >> 8) & 15;
3184 switch ((exit_qualification >> 4) & 3) {
3185 case 0: /* mov to cr */
3186 val = kvm_register_read(vcpu, reg);
3187 trace_kvm_cr_write(cr, val);
3190 err = kvm_set_cr0(vcpu, val);
3191 complete_insn_gp(vcpu, err);
3194 err = kvm_set_cr3(vcpu, val);
3195 complete_insn_gp(vcpu, err);
3198 err = kvm_set_cr4(vcpu, val);
3199 complete_insn_gp(vcpu, err);
3202 u8 cr8_prev = kvm_get_cr8(vcpu);
3203 u8 cr8 = kvm_register_read(vcpu, reg);
3204 kvm_set_cr8(vcpu, cr8);
3205 skip_emulated_instruction(vcpu);
3206 if (irqchip_in_kernel(vcpu->kvm))
3208 if (cr8_prev <= cr8)
3210 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
3216 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
3217 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
3218 skip_emulated_instruction(vcpu);
3219 vmx_fpu_activate(vcpu);
3221 case 1: /*mov from cr*/
3224 kvm_register_write(vcpu, reg, vcpu->arch.cr3);
3225 trace_kvm_cr_read(cr, vcpu->arch.cr3);
3226 skip_emulated_instruction(vcpu);
3229 val = kvm_get_cr8(vcpu);
3230 kvm_register_write(vcpu, reg, val);
3231 trace_kvm_cr_read(cr, val);
3232 skip_emulated_instruction(vcpu);
3237 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
3238 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
3239 kvm_lmsw(vcpu, val);
3241 skip_emulated_instruction(vcpu);
3246 vcpu->run->exit_reason = 0;
3247 pr_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
3248 (int)(exit_qualification >> 4) & 3, cr);
3252 static int handle_dr(struct kvm_vcpu *vcpu)
3254 unsigned long exit_qualification;
3257 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
3258 if (!kvm_require_cpl(vcpu, 0))
3260 dr = vmcs_readl(GUEST_DR7);
3263 * As the vm-exit takes precedence over the debug trap, we
3264 * need to emulate the latter, either for the host or the
3265 * guest debugging itself.
3267 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
3268 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
3269 vcpu->run->debug.arch.dr7 = dr;
3270 vcpu->run->debug.arch.pc =
3271 vmcs_readl(GUEST_CS_BASE) +
3272 vmcs_readl(GUEST_RIP);
3273 vcpu->run->debug.arch.exception = DB_VECTOR;
3274 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
3277 vcpu->arch.dr7 &= ~DR7_GD;
3278 vcpu->arch.dr6 |= DR6_BD;
3279 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
3280 kvm_queue_exception(vcpu, DB_VECTOR);
3285 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3286 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
3287 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
3288 if (exit_qualification & TYPE_MOV_FROM_DR) {
3290 if (!kvm_get_dr(vcpu, dr, &val))
3291 kvm_register_write(vcpu, reg, val);
3293 kvm_set_dr(vcpu, dr, vcpu->arch.regs[reg]);
3294 skip_emulated_instruction(vcpu);
3298 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
3300 vmcs_writel(GUEST_DR7, val);
3303 static int handle_cpuid(struct kvm_vcpu *vcpu)
3305 kvm_emulate_cpuid(vcpu);
3309 static int handle_rdmsr(struct kvm_vcpu *vcpu)
3311 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3314 if (vmx_get_msr(vcpu, ecx, &data)) {
3315 trace_kvm_msr_read_ex(ecx);
3316 kvm_inject_gp(vcpu, 0);
3320 trace_kvm_msr_read(ecx, data);
3322 /* FIXME: handling of bits 32:63 of rax, rdx */
3323 vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
3324 vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
3325 skip_emulated_instruction(vcpu);
3329 static int handle_wrmsr(struct kvm_vcpu *vcpu)
3331 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3332 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
3333 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
3335 if (vmx_set_msr(vcpu, ecx, data) != 0) {
3336 trace_kvm_msr_write_ex(ecx, data);
3337 kvm_inject_gp(vcpu, 0);
3341 trace_kvm_msr_write(ecx, data);
3342 skip_emulated_instruction(vcpu);
3346 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
3351 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
3353 u32 cpu_based_vm_exec_control;
3355 /* clear pending irq */
3356 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3357 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
3358 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3360 ++vcpu->stat.irq_window_exits;
3363 * If the user space waits to inject interrupts, exit as soon as
3366 if (!irqchip_in_kernel(vcpu->kvm) &&
3367 vcpu->run->request_interrupt_window &&
3368 !kvm_cpu_has_interrupt(vcpu)) {
3369 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
3375 static int handle_halt(struct kvm_vcpu *vcpu)
3377 skip_emulated_instruction(vcpu);
3378 return kvm_emulate_halt(vcpu);
3381 static int handle_vmcall(struct kvm_vcpu *vcpu)
3383 skip_emulated_instruction(vcpu);
3384 kvm_emulate_hypercall(vcpu);
3388 static int handle_vmx_insn(struct kvm_vcpu *vcpu)
3390 kvm_queue_exception(vcpu, UD_VECTOR);
3394 static int handle_invlpg(struct kvm_vcpu *vcpu)
3396 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3398 kvm_mmu_invlpg(vcpu, exit_qualification);
3399 skip_emulated_instruction(vcpu);
3403 static int handle_wbinvd(struct kvm_vcpu *vcpu)
3405 skip_emulated_instruction(vcpu);
3406 kvm_emulate_wbinvd(vcpu);
3410 static int handle_xsetbv(struct kvm_vcpu *vcpu)
3412 u64 new_bv = kvm_read_edx_eax(vcpu);
3413 u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3415 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
3416 skip_emulated_instruction(vcpu);
3420 static int handle_apic_access(struct kvm_vcpu *vcpu)
3422 return emulate_instruction(vcpu, 0, 0, 0) == EMULATE_DONE;
3425 static int handle_task_switch(struct kvm_vcpu *vcpu)
3427 struct vcpu_vmx *vmx = to_vmx(vcpu);
3428 unsigned long exit_qualification;
3429 bool has_error_code = false;
3432 int reason, type, idt_v;
3434 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
3435 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
3437 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3439 reason = (u32)exit_qualification >> 30;
3440 if (reason == TASK_SWITCH_GATE && idt_v) {
3442 case INTR_TYPE_NMI_INTR:
3443 vcpu->arch.nmi_injected = false;
3444 if (cpu_has_virtual_nmis())
3445 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3446 GUEST_INTR_STATE_NMI);
3448 case INTR_TYPE_EXT_INTR:
3449 case INTR_TYPE_SOFT_INTR:
3450 kvm_clear_interrupt_queue(vcpu);
3452 case INTR_TYPE_HARD_EXCEPTION:
3453 if (vmx->idt_vectoring_info &
3454 VECTORING_INFO_DELIVER_CODE_MASK) {
3455 has_error_code = true;
3457 vmcs_read32(IDT_VECTORING_ERROR_CODE);
3460 case INTR_TYPE_SOFT_EXCEPTION:
3461 kvm_clear_exception_queue(vcpu);
3467 tss_selector = exit_qualification;
3469 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
3470 type != INTR_TYPE_EXT_INTR &&
3471 type != INTR_TYPE_NMI_INTR))
3472 skip_emulated_instruction(vcpu);
3474 if (kvm_task_switch(vcpu, tss_selector, reason,
3475 has_error_code, error_code) == EMULATE_FAIL) {
3476 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3477 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
3478 vcpu->run->internal.ndata = 0;
3482 /* clear all local breakpoint enable flags */
3483 vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~55);
3486 * TODO: What about debug traps on tss switch?
3487 * Are we supposed to inject them and update dr6?
3493 static int handle_ept_violation(struct kvm_vcpu *vcpu)
3495 unsigned long exit_qualification;
3499 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3501 if (exit_qualification & (1 << 6)) {
3502 printk(KERN_ERR "EPT: GPA exceeds GAW!\n");
3506 gla_validity = (exit_qualification >> 7) & 0x3;
3507 if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
3508 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
3509 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
3510 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
3511 vmcs_readl(GUEST_LINEAR_ADDRESS));
3512 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
3513 (long unsigned int)exit_qualification);
3514 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3515 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
3519 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
3520 trace_kvm_page_fault(gpa, exit_qualification);
3521 return kvm_mmu_page_fault(vcpu, gpa & PAGE_MASK, 0);
3524 static u64 ept_rsvd_mask(u64 spte, int level)
3529 for (i = 51; i > boot_cpu_data.x86_phys_bits; i--)
3530 mask |= (1ULL << i);
3533 /* bits 7:3 reserved */
3535 else if (level == 2) {
3536 if (spte & (1ULL << 7))
3537 /* 2MB ref, bits 20:12 reserved */
3540 /* bits 6:3 reserved */
3547 static void ept_misconfig_inspect_spte(struct kvm_vcpu *vcpu, u64 spte,
3550 printk(KERN_ERR "%s: spte 0x%llx level %d\n", __func__, spte, level);
3552 /* 010b (write-only) */
3553 WARN_ON((spte & 0x7) == 0x2);
3555 /* 110b (write/execute) */
3556 WARN_ON((spte & 0x7) == 0x6);
3558 /* 100b (execute-only) and value not supported by logical processor */
3559 if (!cpu_has_vmx_ept_execute_only())
3560 WARN_ON((spte & 0x7) == 0x4);
3564 u64 rsvd_bits = spte & ept_rsvd_mask(spte, level);
3566 if (rsvd_bits != 0) {
3567 printk(KERN_ERR "%s: rsvd_bits = 0x%llx\n",
3568 __func__, rsvd_bits);
3572 if (level == 1 || (level == 2 && (spte & (1ULL << 7)))) {
3573 u64 ept_mem_type = (spte & 0x38) >> 3;
3575 if (ept_mem_type == 2 || ept_mem_type == 3 ||
3576 ept_mem_type == 7) {
3577 printk(KERN_ERR "%s: ept_mem_type=0x%llx\n",
3578 __func__, ept_mem_type);
3585 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
3591 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
3593 printk(KERN_ERR "EPT: Misconfiguration.\n");
3594 printk(KERN_ERR "EPT: GPA: 0x%llx\n", gpa);
3596 nr_sptes = kvm_mmu_get_spte_hierarchy(vcpu, gpa, sptes);
3598 for (i = PT64_ROOT_LEVEL; i > PT64_ROOT_LEVEL - nr_sptes; --i)
3599 ept_misconfig_inspect_spte(vcpu, sptes[i-1], i);
3601 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3602 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
3607 static int handle_nmi_window(struct kvm_vcpu *vcpu)
3609 u32 cpu_based_vm_exec_control;
3611 /* clear pending NMI */
3612 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3613 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
3614 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3615 ++vcpu->stat.nmi_window_exits;
3620 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
3622 struct vcpu_vmx *vmx = to_vmx(vcpu);
3623 enum emulation_result err = EMULATE_DONE;
3626 while (!guest_state_valid(vcpu)) {
3627 err = emulate_instruction(vcpu, 0, 0, 0);
3629 if (err == EMULATE_DO_MMIO) {
3634 if (err != EMULATE_DONE)
3637 if (signal_pending(current))
3643 vmx->emulation_required = 0;
3649 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
3650 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
3652 static int handle_pause(struct kvm_vcpu *vcpu)
3654 skip_emulated_instruction(vcpu);
3655 kvm_vcpu_on_spin(vcpu);
3660 static int handle_invalid_op(struct kvm_vcpu *vcpu)
3662 kvm_queue_exception(vcpu, UD_VECTOR);
3667 * The exit handlers return 1 if the exit was handled fully and guest execution
3668 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
3669 * to be done to userspace and return 0.
3671 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
3672 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
3673 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
3674 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
3675 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
3676 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
3677 [EXIT_REASON_CR_ACCESS] = handle_cr,
3678 [EXIT_REASON_DR_ACCESS] = handle_dr,
3679 [EXIT_REASON_CPUID] = handle_cpuid,
3680 [EXIT_REASON_MSR_READ] = handle_rdmsr,
3681 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
3682 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
3683 [EXIT_REASON_HLT] = handle_halt,
3684 [EXIT_REASON_INVLPG] = handle_invlpg,
3685 [EXIT_REASON_VMCALL] = handle_vmcall,
3686 [EXIT_REASON_VMCLEAR] = handle_vmx_insn,
3687 [EXIT_REASON_VMLAUNCH] = handle_vmx_insn,
3688 [EXIT_REASON_VMPTRLD] = handle_vmx_insn,
3689 [EXIT_REASON_VMPTRST] = handle_vmx_insn,
3690 [EXIT_REASON_VMREAD] = handle_vmx_insn,
3691 [EXIT_REASON_VMRESUME] = handle_vmx_insn,
3692 [EXIT_REASON_VMWRITE] = handle_vmx_insn,
3693 [EXIT_REASON_VMOFF] = handle_vmx_insn,
3694 [EXIT_REASON_VMON] = handle_vmx_insn,
3695 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
3696 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
3697 [EXIT_REASON_WBINVD] = handle_wbinvd,
3698 [EXIT_REASON_XSETBV] = handle_xsetbv,
3699 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
3700 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
3701 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
3702 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
3703 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
3704 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_invalid_op,
3705 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_invalid_op,
3708 static const int kvm_vmx_max_exit_handlers =
3709 ARRAY_SIZE(kvm_vmx_exit_handlers);
3712 * The guest has exited. See if we can fix it or if we need userspace
3715 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
3717 struct vcpu_vmx *vmx = to_vmx(vcpu);
3718 u32 exit_reason = vmx->exit_reason;
3719 u32 vectoring_info = vmx->idt_vectoring_info;
3721 trace_kvm_exit(exit_reason, vcpu);
3723 /* If guest state is invalid, start emulating */
3724 if (vmx->emulation_required && emulate_invalid_guest_state)
3725 return handle_invalid_guest_state(vcpu);
3727 /* Access CR3 don't cause VMExit in paging mode, so we need
3728 * to sync with guest real CR3. */
3729 if (enable_ept && is_paging(vcpu))
3730 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3732 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
3733 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3734 vcpu->run->fail_entry.hardware_entry_failure_reason
3739 if (unlikely(vmx->fail)) {
3740 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3741 vcpu->run->fail_entry.hardware_entry_failure_reason
3742 = vmcs_read32(VM_INSTRUCTION_ERROR);
3746 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
3747 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
3748 exit_reason != EXIT_REASON_EPT_VIOLATION &&
3749 exit_reason != EXIT_REASON_TASK_SWITCH))
3750 printk(KERN_WARNING "%s: unexpected, valid vectoring info "
3751 "(0x%x) and exit reason is 0x%x\n",
3752 __func__, vectoring_info, exit_reason);
3754 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked)) {
3755 if (vmx_interrupt_allowed(vcpu)) {
3756 vmx->soft_vnmi_blocked = 0;
3757 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
3758 vcpu->arch.nmi_pending) {
3760 * This CPU don't support us in finding the end of an
3761 * NMI-blocked window if the guest runs with IRQs
3762 * disabled. So we pull the trigger after 1 s of
3763 * futile waiting, but inform the user about this.
3765 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
3766 "state on VCPU %d after 1 s timeout\n",
3767 __func__, vcpu->vcpu_id);
3768 vmx->soft_vnmi_blocked = 0;
3772 if (exit_reason < kvm_vmx_max_exit_handlers
3773 && kvm_vmx_exit_handlers[exit_reason])
3774 return kvm_vmx_exit_handlers[exit_reason](vcpu);
3776 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3777 vcpu->run->hw.hardware_exit_reason = exit_reason;
3782 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
3784 if (irr == -1 || tpr < irr) {
3785 vmcs_write32(TPR_THRESHOLD, 0);
3789 vmcs_write32(TPR_THRESHOLD, irr);
3792 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
3795 u32 idt_vectoring_info = vmx->idt_vectoring_info;
3799 bool idtv_info_valid;
3801 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
3803 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
3805 /* Handle machine checks before interrupts are enabled */
3806 if ((vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY)
3807 || (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI
3808 && is_machine_check(exit_intr_info)))
3809 kvm_machine_check();
3811 /* We need to handle NMIs before interrupts are enabled */
3812 if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
3813 (exit_intr_info & INTR_INFO_VALID_MASK)) {
3814 kvm_before_handle_nmi(&vmx->vcpu);
3816 kvm_after_handle_nmi(&vmx->vcpu);
3819 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
3821 if (cpu_has_virtual_nmis()) {
3822 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
3823 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
3825 * SDM 3: 27.7.1.2 (September 2008)
3826 * Re-set bit "block by NMI" before VM entry if vmexit caused by
3827 * a guest IRET fault.
3828 * SDM 3: 23.2.2 (September 2008)
3829 * Bit 12 is undefined in any of the following cases:
3830 * If the VM exit sets the valid bit in the IDT-vectoring
3831 * information field.
3832 * If the VM exit is due to a double fault.
3834 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
3835 vector != DF_VECTOR && !idtv_info_valid)
3836 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3837 GUEST_INTR_STATE_NMI);
3838 } else if (unlikely(vmx->soft_vnmi_blocked))
3839 vmx->vnmi_blocked_time +=
3840 ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
3842 vmx->vcpu.arch.nmi_injected = false;
3843 kvm_clear_exception_queue(&vmx->vcpu);
3844 kvm_clear_interrupt_queue(&vmx->vcpu);
3846 if (!idtv_info_valid)
3849 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
3850 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
3853 case INTR_TYPE_NMI_INTR:
3854 vmx->vcpu.arch.nmi_injected = true;
3856 * SDM 3: 27.7.1.2 (September 2008)
3857 * Clear bit "block by NMI" before VM entry if a NMI
3860 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
3861 GUEST_INTR_STATE_NMI);
3863 case INTR_TYPE_SOFT_EXCEPTION:
3864 vmx->vcpu.arch.event_exit_inst_len =
3865 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3867 case INTR_TYPE_HARD_EXCEPTION:
3868 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
3869 u32 err = vmcs_read32(IDT_VECTORING_ERROR_CODE);
3870 kvm_queue_exception_e(&vmx->vcpu, vector, err);
3872 kvm_queue_exception(&vmx->vcpu, vector);
3874 case INTR_TYPE_SOFT_INTR:
3875 vmx->vcpu.arch.event_exit_inst_len =
3876 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3878 case INTR_TYPE_EXT_INTR:
3879 kvm_queue_interrupt(&vmx->vcpu, vector,
3880 type == INTR_TYPE_SOFT_INTR);
3888 * Failure to inject an interrupt should give us the information
3889 * in IDT_VECTORING_INFO_FIELD. However, if the failure occurs
3890 * when fetching the interrupt redirection bitmap in the real-mode
3891 * tss, this doesn't happen. So we do it ourselves.
3893 static void fixup_rmode_irq(struct vcpu_vmx *vmx)
3895 vmx->rmode.irq.pending = 0;
3896 if (kvm_rip_read(&vmx->vcpu) + 1 != vmx->rmode.irq.rip)
3898 kvm_rip_write(&vmx->vcpu, vmx->rmode.irq.rip);
3899 if (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) {
3900 vmx->idt_vectoring_info &= ~VECTORING_INFO_TYPE_MASK;
3901 vmx->idt_vectoring_info |= INTR_TYPE_EXT_INTR;
3904 vmx->idt_vectoring_info =
3905 VECTORING_INFO_VALID_MASK
3906 | INTR_TYPE_EXT_INTR
3907 | vmx->rmode.irq.vector;
3910 #ifdef CONFIG_X86_64
3918 static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
3920 struct vcpu_vmx *vmx = to_vmx(vcpu);
3922 /* Record the guest's net vcpu time for enforced NMI injections. */
3923 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
3924 vmx->entry_time = ktime_get();
3926 /* Don't enter VMX if guest state is invalid, let the exit handler
3927 start emulation until we arrive back to a valid state */
3928 if (vmx->emulation_required && emulate_invalid_guest_state)
3931 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
3932 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
3933 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
3934 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
3936 /* When single-stepping over STI and MOV SS, we must clear the
3937 * corresponding interruptibility bits in the guest state. Otherwise
3938 * vmentry fails as it then expects bit 14 (BS) in pending debug
3939 * exceptions being set, but that's not correct for the guest debugging
3941 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3942 vmx_set_interrupt_shadow(vcpu, 0);
3945 /* Store host registers */
3946 "push %%"R"dx; push %%"R"bp;"
3948 "cmp %%"R"sp, %c[host_rsp](%0) \n\t"
3950 "mov %%"R"sp, %c[host_rsp](%0) \n\t"
3951 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
3953 /* Reload cr2 if changed */
3954 "mov %c[cr2](%0), %%"R"ax \n\t"
3955 "mov %%cr2, %%"R"dx \n\t"
3956 "cmp %%"R"ax, %%"R"dx \n\t"
3958 "mov %%"R"ax, %%cr2 \n\t"
3960 /* Check if vmlaunch of vmresume is needed */
3961 "cmpl $0, %c[launched](%0) \n\t"
3962 /* Load guest registers. Don't clobber flags. */
3963 "mov %c[rax](%0), %%"R"ax \n\t"
3964 "mov %c[rbx](%0), %%"R"bx \n\t"
3965 "mov %c[rdx](%0), %%"R"dx \n\t"
3966 "mov %c[rsi](%0), %%"R"si \n\t"
3967 "mov %c[rdi](%0), %%"R"di \n\t"
3968 "mov %c[rbp](%0), %%"R"bp \n\t"
3969 #ifdef CONFIG_X86_64
3970 "mov %c[r8](%0), %%r8 \n\t"
3971 "mov %c[r9](%0), %%r9 \n\t"
3972 "mov %c[r10](%0), %%r10 \n\t"
3973 "mov %c[r11](%0), %%r11 \n\t"
3974 "mov %c[r12](%0), %%r12 \n\t"
3975 "mov %c[r13](%0), %%r13 \n\t"
3976 "mov %c[r14](%0), %%r14 \n\t"
3977 "mov %c[r15](%0), %%r15 \n\t"
3979 "mov %c[rcx](%0), %%"R"cx \n\t" /* kills %0 (ecx) */
3981 /* Enter guest mode */
3982 "jne .Llaunched \n\t"
3983 __ex(ASM_VMX_VMLAUNCH) "\n\t"
3984 "jmp .Lkvm_vmx_return \n\t"
3985 ".Llaunched: " __ex(ASM_VMX_VMRESUME) "\n\t"
3986 ".Lkvm_vmx_return: "
3987 /* Save guest registers, load host registers, keep flags */
3988 "xchg %0, (%%"R"sp) \n\t"
3989 "mov %%"R"ax, %c[rax](%0) \n\t"
3990 "mov %%"R"bx, %c[rbx](%0) \n\t"
3991 "push"Q" (%%"R"sp); pop"Q" %c[rcx](%0) \n\t"
3992 "mov %%"R"dx, %c[rdx](%0) \n\t"
3993 "mov %%"R"si, %c[rsi](%0) \n\t"
3994 "mov %%"R"di, %c[rdi](%0) \n\t"
3995 "mov %%"R"bp, %c[rbp](%0) \n\t"
3996 #ifdef CONFIG_X86_64
3997 "mov %%r8, %c[r8](%0) \n\t"
3998 "mov %%r9, %c[r9](%0) \n\t"
3999 "mov %%r10, %c[r10](%0) \n\t"
4000 "mov %%r11, %c[r11](%0) \n\t"
4001 "mov %%r12, %c[r12](%0) \n\t"
4002 "mov %%r13, %c[r13](%0) \n\t"
4003 "mov %%r14, %c[r14](%0) \n\t"
4004 "mov %%r15, %c[r15](%0) \n\t"
4006 "mov %%cr2, %%"R"ax \n\t"
4007 "mov %%"R"ax, %c[cr2](%0) \n\t"
4009 "pop %%"R"bp; pop %%"R"bp; pop %%"R"dx \n\t"
4010 "setbe %c[fail](%0) \n\t"
4011 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
4012 [launched]"i"(offsetof(struct vcpu_vmx, launched)),
4013 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
4014 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
4015 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
4016 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
4017 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
4018 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
4019 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
4020 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
4021 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
4022 #ifdef CONFIG_X86_64
4023 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
4024 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
4025 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
4026 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
4027 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
4028 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
4029 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
4030 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
4032 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2))
4034 , R"bx", R"di", R"si"
4035 #ifdef CONFIG_X86_64
4036 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
4040 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
4041 | (1 << VCPU_EXREG_PDPTR));
4042 vcpu->arch.regs_dirty = 0;
4044 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
4045 if (vmx->rmode.irq.pending)
4046 fixup_rmode_irq(vmx);
4048 asm("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
4051 vmx_complete_interrupts(vmx);
4057 static void vmx_free_vmcs(struct kvm_vcpu *vcpu)
4059 struct vcpu_vmx *vmx = to_vmx(vcpu);
4063 free_vmcs(vmx->vmcs);
4068 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
4070 struct vcpu_vmx *vmx = to_vmx(vcpu);
4073 vmx_free_vmcs(vcpu);
4074 kfree(vmx->guest_msrs);
4075 kvm_vcpu_uninit(vcpu);
4076 kmem_cache_free(kvm_vcpu_cache, vmx);
4079 static inline void vmcs_init(struct vmcs *vmcs)
4081 u64 phys_addr = __pa(per_cpu(vmxarea, raw_smp_processor_id()));
4084 kvm_cpu_vmxon(phys_addr);
4092 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
4095 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
4099 return ERR_PTR(-ENOMEM);
4103 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
4107 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
4108 if (!vmx->guest_msrs) {
4113 vmx->vmcs = alloc_vmcs();
4117 vmcs_init(vmx->vmcs);
4120 vmx_vcpu_load(&vmx->vcpu, cpu);
4121 err = vmx_vcpu_setup(vmx);
4122 vmx_vcpu_put(&vmx->vcpu);
4126 if (vm_need_virtualize_apic_accesses(kvm))
4127 if (alloc_apic_access_page(kvm) != 0)
4131 if (!kvm->arch.ept_identity_map_addr)
4132 kvm->arch.ept_identity_map_addr =
4133 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
4134 if (alloc_identity_pagetable(kvm) != 0)
4141 free_vmcs(vmx->vmcs);
4143 kfree(vmx->guest_msrs);
4145 kvm_vcpu_uninit(&vmx->vcpu);
4148 kmem_cache_free(kvm_vcpu_cache, vmx);
4149 return ERR_PTR(err);
4152 static void __init vmx_check_processor_compat(void *rtn)
4154 struct vmcs_config vmcs_conf;
4157 if (setup_vmcs_config(&vmcs_conf) < 0)
4159 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
4160 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
4161 smp_processor_id());
4166 static int get_ept_level(void)
4168 return VMX_EPT_DEFAULT_GAW + 1;
4171 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
4175 /* For VT-d and EPT combination
4176 * 1. MMIO: always map as UC
4178 * a. VT-d without snooping control feature: can't guarantee the
4179 * result, try to trust guest.
4180 * b. VT-d with snooping control feature: snooping control feature of
4181 * VT-d engine can guarantee the cache correctness. Just set it
4182 * to WB to keep consistent with host. So the same as item 3.
4183 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
4184 * consistent with host MTRR
4187 ret = MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
4188 else if (vcpu->kvm->arch.iommu_domain &&
4189 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY))
4190 ret = kvm_get_guest_memory_type(vcpu, gfn) <<
4191 VMX_EPT_MT_EPTE_SHIFT;
4193 ret = (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT)
4199 #define _ER(x) { EXIT_REASON_##x, #x }
4201 static const struct trace_print_flags vmx_exit_reasons_str[] = {
4203 _ER(EXTERNAL_INTERRUPT),
4205 _ER(PENDING_INTERRUPT),
4225 _ER(IO_INSTRUCTION),
4228 _ER(MWAIT_INSTRUCTION),
4229 _ER(MONITOR_INSTRUCTION),
4230 _ER(PAUSE_INSTRUCTION),
4231 _ER(MCE_DURING_VMENTRY),
4232 _ER(TPR_BELOW_THRESHOLD),
4242 static int vmx_get_lpage_level(void)
4244 if (enable_ept && !cpu_has_vmx_ept_1g_page())
4245 return PT_DIRECTORY_LEVEL;
4247 /* For shadow and EPT supported 1GB page */
4248 return PT_PDPE_LEVEL;
4251 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
4253 struct kvm_cpuid_entry2 *best;
4254 struct vcpu_vmx *vmx = to_vmx(vcpu);
4257 vmx->rdtscp_enabled = false;
4258 if (vmx_rdtscp_supported()) {
4259 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
4260 if (exec_control & SECONDARY_EXEC_RDTSCP) {
4261 best = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
4262 if (best && (best->edx & bit(X86_FEATURE_RDTSCP)))
4263 vmx->rdtscp_enabled = true;
4265 exec_control &= ~SECONDARY_EXEC_RDTSCP;
4266 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
4273 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
4277 static struct kvm_x86_ops vmx_x86_ops = {
4278 .cpu_has_kvm_support = cpu_has_kvm_support,
4279 .disabled_by_bios = vmx_disabled_by_bios,
4280 .hardware_setup = hardware_setup,
4281 .hardware_unsetup = hardware_unsetup,
4282 .check_processor_compatibility = vmx_check_processor_compat,
4283 .hardware_enable = hardware_enable,
4284 .hardware_disable = hardware_disable,
4285 .cpu_has_accelerated_tpr = report_flexpriority,
4287 .vcpu_create = vmx_create_vcpu,
4288 .vcpu_free = vmx_free_vcpu,
4289 .vcpu_reset = vmx_vcpu_reset,
4291 .prepare_guest_switch = vmx_save_host_state,
4292 .vcpu_load = vmx_vcpu_load,
4293 .vcpu_put = vmx_vcpu_put,
4295 .set_guest_debug = set_guest_debug,
4296 .get_msr = vmx_get_msr,
4297 .set_msr = vmx_set_msr,
4298 .get_segment_base = vmx_get_segment_base,
4299 .get_segment = vmx_get_segment,
4300 .set_segment = vmx_set_segment,
4301 .get_cpl = vmx_get_cpl,
4302 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
4303 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
4304 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
4305 .set_cr0 = vmx_set_cr0,
4306 .set_cr3 = vmx_set_cr3,
4307 .set_cr4 = vmx_set_cr4,
4308 .set_efer = vmx_set_efer,
4309 .get_idt = vmx_get_idt,
4310 .set_idt = vmx_set_idt,
4311 .get_gdt = vmx_get_gdt,
4312 .set_gdt = vmx_set_gdt,
4313 .set_dr7 = vmx_set_dr7,
4314 .cache_reg = vmx_cache_reg,
4315 .get_rflags = vmx_get_rflags,
4316 .set_rflags = vmx_set_rflags,
4317 .fpu_activate = vmx_fpu_activate,
4318 .fpu_deactivate = vmx_fpu_deactivate,
4320 .tlb_flush = vmx_flush_tlb,
4322 .run = vmx_vcpu_run,
4323 .handle_exit = vmx_handle_exit,
4324 .skip_emulated_instruction = skip_emulated_instruction,
4325 .set_interrupt_shadow = vmx_set_interrupt_shadow,
4326 .get_interrupt_shadow = vmx_get_interrupt_shadow,
4327 .patch_hypercall = vmx_patch_hypercall,
4328 .set_irq = vmx_inject_irq,
4329 .set_nmi = vmx_inject_nmi,
4330 .queue_exception = vmx_queue_exception,
4331 .interrupt_allowed = vmx_interrupt_allowed,
4332 .nmi_allowed = vmx_nmi_allowed,
4333 .get_nmi_mask = vmx_get_nmi_mask,
4334 .set_nmi_mask = vmx_set_nmi_mask,
4335 .enable_nmi_window = enable_nmi_window,
4336 .enable_irq_window = enable_irq_window,
4337 .update_cr8_intercept = update_cr8_intercept,
4339 .set_tss_addr = vmx_set_tss_addr,
4340 .get_tdp_level = get_ept_level,
4341 .get_mt_mask = vmx_get_mt_mask,
4343 .exit_reasons_str = vmx_exit_reasons_str,
4344 .get_lpage_level = vmx_get_lpage_level,
4346 .cpuid_update = vmx_cpuid_update,
4348 .rdtscp_supported = vmx_rdtscp_supported,
4350 .set_supported_cpuid = vmx_set_supported_cpuid,
4352 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
4355 static int __init vmx_init(void)
4359 rdmsrl_safe(MSR_EFER, &host_efer);
4361 for (i = 0; i < NR_VMX_MSR; ++i)
4362 kvm_define_shared_msr(i, vmx_msr_index[i]);
4364 vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
4365 if (!vmx_io_bitmap_a)
4368 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
4369 if (!vmx_io_bitmap_b) {
4374 vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
4375 if (!vmx_msr_bitmap_legacy) {
4380 vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
4381 if (!vmx_msr_bitmap_longmode) {
4387 * Allow direct access to the PC debug port (it is often used for I/O
4388 * delays, but the vmexits simply slow things down).
4390 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
4391 clear_bit(0x80, vmx_io_bitmap_a);
4393 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
4395 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
4396 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
4398 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
4400 r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
4401 __alignof__(struct vcpu_vmx), THIS_MODULE);
4405 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
4406 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
4407 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
4408 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
4409 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
4410 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
4413 bypass_guest_pf = 0;
4414 kvm_mmu_set_base_ptes(VMX_EPT_READABLE_MASK |
4415 VMX_EPT_WRITABLE_MASK);
4416 kvm_mmu_set_mask_ptes(0ull, 0ull, 0ull, 0ull,
4417 VMX_EPT_EXECUTABLE_MASK);
4422 if (bypass_guest_pf)
4423 kvm_mmu_set_nonpresent_ptes(~0xffeull, 0ull);
4428 free_page((unsigned long)vmx_msr_bitmap_longmode);
4430 free_page((unsigned long)vmx_msr_bitmap_legacy);
4432 free_page((unsigned long)vmx_io_bitmap_b);
4434 free_page((unsigned long)vmx_io_bitmap_a);
4438 static void __exit vmx_exit(void)
4440 free_page((unsigned long)vmx_msr_bitmap_legacy);
4441 free_page((unsigned long)vmx_msr_bitmap_longmode);
4442 free_page((unsigned long)vmx_io_bitmap_b);
4443 free_page((unsigned long)vmx_io_bitmap_a);
4448 module_init(vmx_init)
4449 module_exit(vmx_exit)